This is the first book devoted to Bose-Einstein condensation (BEC) as

an interdisciplinary subject, covering atomic and molecular physics, laser

physics, low temperature physics, nuclear physics and astrophysics. It
contains 18 authoritative review articles on experimental and theoretical
research on BEC and associated phenomena.
Bose-Einstein condensation is a phase transition in which a macro-
scopic number of particles all go into the same quantum state. It has been
known for some time that this phenomenon gives rise to superfluidity
in liquid helium. However, recent research deals with BEC in a num-
ber of other systems, such as excitons in semiconductors, spin-polarized
hydrogen, laser-cooled atoms, high-temperature superconductors, and
subatomic matter. The search for BEC in these new physical systems
is an exciting field of research, and this unique book gives an in-depth
report on recent progress, as well as suggesting promising research for
the future.
This book will be of interest to graduate students and research workers
in all fields of physics, but especially those in condensed matter, low
temperature, atomic and laser physics.
Bose-Einstein Condensation
BOSE-EINSTEIN
CONDENSATION

Edited by

A. GRIFFIN
University of Toronto, Toronto, Canada

D. W. SNOKE
University of Pittsburgh, Pittsburgh, USA

S. STRINGARI
University of Trento, Povo, Italy

CAMBRIDGE
UNIVERSITY PRESS
 Published by the Press Syndicate of the University of Cambridge
The Pitt Building, Trumpington Street, Cambridge CB2 1RP
40 West 20th Street, New York, NY 10011-4211, USA
10 Stamford Road, Oakleigh, Melbourne 3166, Australia

© Cambridge University Press 1995

First published 1995

Reprinted 1996
First paperback edition 1996

A catalogue record of this book is available from the British Library

Library of Congress cataloguing in publication data

Bose-Einstein condensation / edited by A. Griffin, D.W. Snoke, S. Stringari.
p. cm.
Includes index.
ISBN 0-521-46473-0
1. Bose-Einstein condensation. I. Griffin, Allan. II. Snoke, D. W.
III. Stringari, S.
QC175.47.B65B67 1995
530.4'2-dc20 94-27795 CIP

ISBN 0 521 46473 0 hardback

ISBN 0 521 58990 8 paperback

Transferred to digital printing 2002

TAG
 Contents

Preface page xi
Preface to paperback edition xiii

1 Introduction: Unifying Themes of Bose-Einstein

Condensation D. W, Snoke and G. Baym 1

Part one: Review Papers 13

2 Some Comments on Bose-Einstein Condensation P. Nozieres 15
3 Bose-Einstein Condensation and Superfluidity K. Huang 31
4 Bose-Einstein Condensation in Liquid Helium P. Sokol 51
5 Sum Rules and Bose-Einstein Condensation S. Stringari 86
6 Dilute-Degenerate Gases F. Laloe 99
7 Prospects for Bose-Einstein Condensation in Magnetically
Trapped Atomic Hydrogen T. J. Greytak 131
8 Spin-Polarized Hydrogen: Prospects for Bose-Einstein
Condensation and Two-Dimensional Superfluidity
/. F Silvera 160
9 Laser Cooling and Trapping of Neutral Atoms Y Castin,
J. Dalibard and C. Cohen-Tannoudji 173
10 Kinetics of Bose-Einstein Condensate Formation in an
Interacting Bose Gas Yu. Kagan 202
11 Condensate Formation in a Bose Gas H. T. C. Stoof 226
12 Macroscopic Coherent States of Excitons in Semiconductors
L. V. Keldysh 246
13 Bose-Einstein Condensation of a Nearly Ideal Gas: Excitons
in Cu 2 O J. P. Wolfe, J. L. Lin and D. W. Snoke 281
14 Bose-Einstein Condensation of Excitonic Particles in
Semiconductors A. Mysyrowicz 330

VII
viii Contents
15 Crossover from BCS Theory to Bose-Einstein Condensation
M. Randeria 355
16 Bose-Einstein Condensation of Bipolarons in High-T c
Superconductors J. Ranninger 393
17 The Bosonization Method in Nuclear Physics F. Iachello 418
18 Kaon Condensation in Dense Matter G. E. Brown 438
19 Broken Gauge Symmetry in a Bose Condensate A. J. Leggett 452

Part two: Brief Reports 463

20 BEC in Ultra-cold Cesium: Collisional Constraints
E. Tiesinga, A. J. Moerdijk, B. J. Verhaar, and H. T. C. Stoof 465
21 BEC and the Relaxation Explosion in Magnetically Trapped
Atomic Hydrogen T. W. Hijmans, Yu. Kagan,
G. V. Shlyapnikov and J. T. M. Walraven All
22 Quest for Kosterlitz-Thouless Transition in Two-Dimensional
Atomic Hydrogen A. Matsubara, T. Aral, S. Hotta,
J. S. Korhonen, T. Mizusaki and A. Hirai 478
23 BEC of Biexcitons in CuCl M. Hasuo and N. Nagasawa 487
24 The Influence of Polariton Effects on BEC of Biexcitons
A. L. Ivanov and H. Haug 496
25 Light-Induced BEC of Excitons and Biexcitons
A. I. Bobrysheva and S. A. Moskalenko 507
26 Evolution of a Nonequilibrium Polariton Condensate
/. V. Beloussov and Yu. M. Shvera 513
27 Excitonic Superfluidity in CU2O E. Fortin, E. Benson and
A. Mysyrowicz 519
28 On the Bose-Einstein Condensation of Excitons:
Finite-lifetime Composite Bosons S. G. Tikhodeev 524
29 Charged Bosons in Quantum Heterostructures
L. D. Shvartsman and J. E. Golub 532
30 Evidence for Bipolaronic Bose-liquid and BEC in
High-T c Oxides A. S. Alexandrov 541
31 The Dynamic Structure Function of Bose Liquids in the Deep
Inelastic Regime A. Belie 550
32 Possibilities for BEC of Positronium P. M. Platzman
and A. P. Mills, Jr 558
33 Bose-Einstein Condensation and Spin Waves R. Friedberg,
T. D. Lee and H. C. Ren 565
34 Universal Behaviour within the Nozieres-Schmitt-Rink
Theory F. Pistolesi and G. C. Strinati 569
 Contents ix
35 Bound States and Superfluidity in Strongly Coupled Fermion
Systems G. Ropke 574
36 Onset of Superfluidity in Nuclear Matter A. Hellmich,
G. Ropke, A. Schnell, and H. Stein 584

Appendix. BEC 93 Participant List 595

Index 597
 Preface

In recent years, the phenomenon of Bose-Einstein condensation (BEC)

has become an increasingly active area of research, both experimentally
and theoretically. Long associated with the study of superfluid 4He
and 3He-4He mixtures, current research increasingly deals with other
condensed matter systems, including spin-polarized hydrogen, excitons,
laser-cooled atoms, high-temperature superconductors, and subatomic
matter.
The present volume contains a series of authoritative review articles
on current BEC research and related phenomena. The editors invited
the leading experts to review their research field in such a way that
their articles would introduce non-experts to current research and at the
same time highlight some of the most promising topics for study in the
next decade. These articles contain new material which is not available
elsewhere. This is the first book devoted to BEC as an inter-disciplinary
subject in its own right. It covers research in atomic and molecular
physics, laser physics, low temperature physics, subatomic physics and
astrophysics.
In the opening chapter, Snoke and Baym introduce the various review
articles by discussing them in the context of the dominant themes in
current BEC studies. These themes include broken gauge invariance,
phase coherence in equilibrium and non-equilibrium situations, time
scales for the formation of a Bose condensate, Bose particles with a
finite lifetime, and BEC vs. BCS superfluidity in fermionic systems with
attractive interactions.
The interdisciplinary character of modern BEC research has led to its
being largely ignored by the regular low temperature conferences. The
genesis of the present volume was the suggestion in 1992 by David Snoke
that an international workshop devoted to current BEC research was long

XI
xii Preface

overdue. The idea then developed that the main focus of this first BEC
workshop should be review-type talks by carefully chosen international
experts and that these talks would be the basis of an authoritative book
on BEC research, suitable for someone just getting interested in this topic.
The International Workshop on Bose-Einstein Condensation (BEC 93)
was duly organized and held in Levico Terme, a small resort town near
Trento in northern Italy, 31 May to 4 June, 1993. It was a resounding
success from every point of view. Over 70 participants from over 20
countries (listed in an appendix at the end of this book) represented all
the major research groups currently working on BEC, and included a
good mix of experimentalists and theorists. Everyone had a sense that
this first BEC workshop was an historic occasion, helped by the fact
that the participants included many of the "founding fathers" of BEC
studies, such as K. Huang, S.A. Moskalenko and L.V. Keldysh. The
unifying themes of BEC research were stressed by having talks on quite
different sub-fields given each day.
We decided that the invited review articles could be submitted well
after the workshop, so that they would reflect the excitement and the
interdisciplinary insights which arose from the workshop. The present
volume contains a few additional invited review articles covering topics
which were not emphasized at BEC 93. At the workshop, over thirty
poster contributions were also presented. The second part of this book
contains brief research reports and mini-reviews based on some of these
contributions. The refereeing was done by the editors and the authors
of the invited papers. Many of the contributed papers have since been
submitted to research journals and for this reason are not included here.
There is only one review article devoted to superfluid 4He. It was felt
that since a recent book has given an extensive review of superfluid 4He
(A. Griffin, Excitations in a Bose-Condensed Liquid, Cambridge University
Press, 1993), it was not necessary to emphasize this subject in the present
volume.
The organizing committee of BEC 93 at Levico Terme consisted of
the editors, with the assistance of F. Laloe and G. Baym. In addition,
A. Mysyrowicz, S. Tikhodeev and D. van der Marel were very supportive
at the very early stages of organization. The major source of external
funding of BEC 93 was from the European Centre for Theoretical Studies
in Nuclear Physics and Related Areas (ECF), which has recently been
set up at the University of Trento. In addition, BEC 93 was financially
assisted by grants from CNRS (France) and the Department of Physics
at the University of Toronto, as well as funds from NSERC of Canada.
 Preface xiii
A crucial role was played by the workshop secretary, Jennifer Tarn
(Toronto).
The local organization at the University of Trento and Levico Terme
was carried out by Sandro Stringari, and we thank especially F. Dalfovo
and S. Giorgini for their help. The Grand Hotel Bellavista in Levico
Terme played an important role in producing a successful workshop,
providing the right atmosphere as well as fine Italian cuisine. Late into
each evening, heated discussions on topics such as timescales and broken
symmetry could be heard coming from the hotel bar.
Preparation of the book in the Cambridge University Press L£F]gX
format was carried out by David Snoke and Karl Schroeder (Toronto).
Most of the contributors made this job much easier for us by sending
in their articles in TgX or L^TgX. We would also like to thank Simon
Capelin of Cambridge University Press for his enthusiastic support right
from the beginning.
We hope this unique volume will introduce many readers to the chal-
lenging problems involved in the quest for BEC and set the stage for
future progress.
Allan Griffin, Toronto
David Snoke, Pittsburgh
Sandro Stringari, Trento

Preface to paperback edition

With the achievement of Bose-Einstein condensation (BEC) in a di-
lute, ultracold gas of alkali atoms (M.H. Anderson, J.R. Ensher, M.R.
Matthews, C.E. Wieman and E.A. Cornell, Science, 269, 198 (1995)),
the study of BEC and its consequences has become a topic of in-
tense worldwide interest. For recent research, see the BEC homepage
(http://amo.phy.gasou.edu/bec.html).
The review articles in this book provide an excellent source of back-
ground information about the properties of Bose-condensed systems. For
the many new researchers who have become interested in studying BEC
in trapped atomic gases, this book also puts the new discoveries into the
wider context of the search for BEC in other physical systems. We hope
this paperback edition will be a useful reference for this exciting new
area of research.
A. Griffin, D. Snoke, S. Stringari
 Preface xiii
A crucial role was played by the workshop secretary, Jennifer Tarn
(Toronto).
The local organization at the University of Trento and Levico Terme
was carried out by Sandro Stringari, and we thank especially F. Dalfovo
and S. Giorgini for their help. The Grand Hotel Bellavista in Levico
Terme played an important role in producing a successful workshop,
providing the right atmosphere as well as fine Italian cuisine. Late into
each evening, heated discussions on topics such as timescales and broken
symmetry could be heard coming from the hotel bar.
Preparation of the book in the Cambridge University Press L£F]gX
format was carried out by David Snoke and Karl Schroeder (Toronto).
Most of the contributors made this job much easier for us by sending
in their articles in TgX or L^TgX. We would also like to thank Simon
Capelin of Cambridge University Press for his enthusiastic support right
from the beginning.
We hope this unique volume will introduce many readers to the chal-
lenging problems involved in the quest for BEC and set the stage for
future progress.
Allan Griffin, Toronto
David Snoke, Pittsburgh
Sandro Stringari, Trento

Preface to paperback edition

With the achievement of Bose-Einstein condensation (BEC) in a di-
lute, ultracold gas of alkali atoms (M.H. Anderson, J.R. Ensher, M.R.
Matthews, C.E. Wieman and E.A. Cornell, Science, 269, 198 (1995)),
the study of BEC and its consequences has become a topic of in-
tense worldwide interest. For recent research, see the BEC homepage
(http://amo.phy.gasou.edu/bec.html).
The review articles in this book provide an excellent source of back-
ground information about the properties of Bose-condensed systems. For
the many new researchers who have become interested in studying BEC
in trapped atomic gases, this book also puts the new discoveries into the
wider context of the search for BEC in other physical systems. We hope
this paperback edition will be a useful reference for this exciting new
area of research.
A. Griffin, D. Snoke, S. Stringari
 Introduction: Unifying Themes of
Bose-Einstein Condensation
D. W. Snoket
Mechanics and Materials Technology Center
The Aerospace Corporation
El Segundo, CA 90009-2957, USA

Gordon Baym
Department of Physics
University of Illinois at Urbana-Champaign
1110W. Green St, Urbana, IL 61801, USA

After many years as a phenomenon with only one experimental example,

superfluid 4He, the concept of Bose-Einstein condensation (BEC) has
in recent years emerged in an array of exciting new experimental and
theoretical systems; indeed, the study of BEC has become a field in its
own right, no longer primarily a subfield of liquid helium studies. As
the articles in this book make apparent, BEC is a common phenomenon
occuring in physics on all scales, from condensed matter to nuclear, ele-
mentary particle, and astrophysics, with ideas flowing across boundaries
between fields. The systems range from gases, liquids, and solids, includ-
ing semiconductors and metals, to atomic nuclei, elementary particles
and matter in neutron stars and supernova explosions. Generally, the
bosonic degrees of freedom are composite, originating from underlying
fermionic degrees of freedom. Table 1 lists bosonic systems presently
under study.
The articles in this book bring out several unifying themes as well as
common problems in the study of BEC which transcend specific systems.
In this introduction, we give an overview of some of these major themes.
(i) Broken gauge symmetry. What is Bose-Einstein condensation?
What is the essential underlying physics? In the past forty years the
concept of broken gauge symmetry (also described as off-diagonal long-
range order (ODLRO) or long-range phase coherence), which is one of

f Present address: Dept of Physics and Astronomy, University of Pittsburgh, Pittsburgh,

PA 15260, USA.
 D. W. Snoke and G. Baym

Table 1. Bosons under study

Particle Composed of In Coherence seen in

Cooper pair e~~, e~ metals superconductivity

Cooper pair h+, h+ copper oxides high-Tc
superconductivity
exciton e-,h+ semiconductors luminescence
and drag-free
transport in CU2O
biexciton 2(e", h+) semiconductors luminescence
and optical phase
coherence in CuCl
positronium e", e + crystal vacancies (proposed)
hydrogen e-,p+ magnetic traps (in progress)
4 4 2+
He He , 2e" He-II superfluidity
3 3 2+ 3
He pairs 2( He , 2e") He-A,B phases superfluidity
133 55+
cesium Cs , 55e" laser traps (in progress)
interacting bosons nn or pp nuclei excitations
nucleonic pairing nn or pp nuclei moments of inertia
neutron stars superfluidity and
pulsar glitches
chiral condensates (99) vacuum elementary particle
structure
meson condensates pion condensate neutron star neutron stars,
= (ud), etc. matter supernovae
kaon condensate (proposed)
= (us)
Higgs boson (It) condensate vacuum elementary particle
(proposed) masses
 Introduction: Unifying Themes of Bose-Einstein Condensation 3
the more remarkable implications of quantum mechanics, has come to
the fore as the fundamental principle for understanding BEC. Nozieres,
Huang, Leggett, Stringari, Stoof, and several other authors in this book
discuss the implications of the theory of broken gauge symmetry. In the
opening review, Chapter 2, Nozieres outlines the essential physics which
determines whether a Bose condensate will win over competing phases
in a wide variety of systems. Leggett, in the concluding review, Chapter
19, examines the validity of the concept of broken symmetry at a very
fundamental level.
The concept of broken gauge symmetry, with the accompanying macro-
scopic wavefunction describing the condensate, was first introduced in
explaining superconductivity and superfluidity. Since then it has also
been widely applied in nuclear and elementary particle physics. Bosonic
condensates, e.g., with non-vanishing expectation values of quark oper-
ators, (MM), (dd), (ss), are believed to be fundamental features of the
vacuum and the structure of elementary particles such as the nucleon,
underlying the spontaneous breaking of the chiral symmetry of the strong
interactions [1]. In fact, the Higgs boson, which gives rise to the breaking
of the SU(2)xU(l) symmetry of the weak interactions, and is invoked
to generate the masses of elementary particles, has been suggested as
arising from a condensation of the top and antitop quarks, with a non-
vanishing expectation value (it) in the vacuum [2]. Mesonic condensates,
e.g., pion and kaon condensates, which are nuclear analogs of exciton
condensates, may be an important feature in dense neutron star matter,
as Brown discusses in his article. In nuclear physics, BCS-type pairing is
used to explain the reduced moments of inertia of heavier nuclei. Both
the neutron and proton components of nuclear and neutron star matter,
more generally, undergo pairing to become superfluid.
In high energy physics, the normal state of the vacuum contains a
condensate, and thus one can ask the opposite of the question asked
in condensed matter physics, namely, can one produce non-condensed
states? A program to do so, in collisions of ultrarelativistic heavy-ions,
has been underway at CERN and Brookhaven. In the normal vacuum,
quark condensates, (qq), are non-zero; in a collision, on timescales of
about 10~23 s and distance scales of order 10~12 cm, it is believed possible
to generate regions with (qq) = 0, in which chiral symmetry becomes
unbroken and the quarks deconfined [3]. Somewhat later the quarks form
hadrons, mostly n mesons, and the state returns to (qq) =fc 0. The dense
pion gas might itself Bose condense, although most likely its entropy is
too high.
4 D.W. Snoke and G. Baym
(ii) The telltale signal for BEC. What is the proof that a system is Bose-
Einstein condensed? Superfluid helium has been studied for decades, and
while we know that it is Bose-Einstein condensed, what is the "smoking
gun?" Almost thirty years ago, Hohenberg and Platzman [4] proposed
that neutron scattering experiments should yield the classic signal, a delta
function for the occupation number of the particles with zero momentum
associated with long-range phase coherence. As Sokol discusses in his
article in this book, however, such a clean signal is not likely to appear
any time soon in the experimental liquid helium data. The reason is that
liquid helium is a strongly interacting system, so that the condensate is
strongly depleted; furthermore, the short lifetime of the final state of the
recoiling atom broadens the neutron scattering data considerably. The
neutron scattering data have been shown to be consistent with many-
body numerical calculations of helium which predict about 9% of the
atoms with zero momentum at T = 0 [5, 6]. Nevertheless, the strongest
proof of the existence of a condensate in liquid helium probably comes
from analysis of the critical exponents of the superfluid phase transition.
(They are characteristic of a two-component order parameter.) Obtaining
accurate data on critical exponents in other experimental systems may
turn out to be much more difficult. In a recent book, Griffin [6] has
given a detailed review of the dynamical properties of liquid 4He based
on the role of an underlying Bose condensate.
What is the general relation between BEC and superfluidity? Do
the superfluid properties of 4He prove that it involves a Bose-Einstein
condensate? The classic work of Hohenberg and Martin, as well as Bo-
goliubov, in the 1960's showed how the two-fluid description of superfluid
4
He was a direct consequence of the existence of Bose broken symmetry
(see chapter 6 of Ref. [6]). That a condensate is not necessary is seen
in superfluid two-dimensional Kosterlitz-Thouless systems [7], which,
rather than having an infinitely long-ranged phase coherence, have a
particle correlation function (xp\r)xp(O)) with a power-law dependence
on r. Lasers illustrate that a condensate is not sufficient. Huang presents
a model of condensation in the presence of a highly disordered substrate
to suggest that, more generally, Bose-Einstein condensation is neither
necessary nor sufficient for superfluidity of boson systems.
(iii) Bose-Einstein vs. Fermi-Dirac degrees of freedom. Bosonic de-
grees of freedom are in general composite; indeed the only fundamental
bosons in nature appear to be the photon, the three massive vector
bosons mediating the weak interaction, the gluon, which mediates the
strong interaction, and the graviton. We must thus face the problem of
 Introduction: Unifying Themes of Bose-Einstein Condensation 5
understanding how bosonic degrees of freedom emerge from underlying
fermionic degrees of freedom. Fermionic systems, for example liquid
3
He and electrons in metals, become superfluid; to what extent can
these systems be regarded as Bose-Einstein condensates? As discussed
by Randeria in this book, the same underlying theory based on the
formation of Cooper pairs can yield both BEC and BCS in different
limits, BEC in the limit of particle-particle correlation length very short
compared to the average interparticle spacing, and BCS in the opposite
limit. One cannot in general always make a clean distinction between
a superfluid system of Cooper pairs and a Bose-Einstein condensate.
Superfluidity and superconductivity both stem from the same underlying
physics. In either case, the bosonic nature of the pairs is critical for un-
derstanding the "super" behavior. Ranninger and others in this volume
suggest possible mechanisms for understanding the new cuprate high-Tc
superconductors as a BEC of small Cooper pairs.
In condensed-matter physics, the internal fermion degrees of freedom
of the nuclei only play the role of determining the statistics of the
nucleus, since nuclear energy scales are so vastly larger than atomic
scales. The nuclear domain itself provides important examples of the
transition between Bose and Fermi degrees of freedom. Examples include
the Interacting Boson Model of nuclei, reviewed here by Iachello, the
description of mesons in terms of their underlying quark structure by
Rho [8] and the deconfinement transition between bosonic mesons and
fermionic quarks, as expected in ultrarelativistic heavy-ion collisions and
in the early universe [3]. In this book Ropke and Hellmich et al. treat
the BEC-BCS crossover in nuclear matter, analogous to that discussed
by Randeria in superconductors.
(iv) BEC of a weakly interacting gas. In superfluid helium, many
features associated with BEC are masked by the strongly interacting
character of the liquid [6]. Many of the novel systems currently studied
in the laboratory are weakly interacting boson gases, whose momentum
spectra and other properties can be much more amenable to theoretical
analysis. The theory of a weakly interacting gas has been developed at
length over the years, starting with London and Bogoliubov, and aspects
of the theory are reviewed by Huang, Stoof and Laloe in this volume;
Greytak and Silvera review experimental attempts to observe BEC in
spin-polarized hydrogen; Castin, Dalibard and Cohen-Tannoudji review
the search for BEC of laser-trapped atoms, in particular cesium; and
Wolfe, Lin and Snoke as well as Mysyrowicz review experimental work
on BEC of excitons, in particular in CU2O. Platzman and Mills report on
6 D.W. Snoke and G. Baym
a proposal for observation of BEC of positronium. All these systems are
expected to remain weakly interacting gases at the BEC phase transition,
and one expects that their behaviors in the quantum degenerate regime
can be well understood [9].
The question of the "smoking gun" for BEC, discussed above, has
become more important because of the recent progress on excitons in
CU2O. Work over the last ten years has shown that excitons can exceed the
critical density for condensation, that they exhibit an extremely narrow
energy distribution with full width at half maximum much less than UB T,
and that at high densities they move without friction through the crystal
over macroscopic distances. This system is clearly a highly quantum-
degenerate, weakly interacting boson gas. Determining the extent to
which this system exhibits off-diagonal long-range order will require
ingenuious experimental tests. One has the advantage, compared with
4
He, that measurement of exciton recombination luminescence provides
fairly direct information about the particle momentum distribution.
An important property of a weakly interacting Bose gas is the phe-
nomenon of "Bose narrowing," i.e., the energy distribution of a weakly
interacting gas should decrease in average energy at a given temperature
as density increases, opposite to the increase with density in the average
energy of a Fermi gas at a given temperature. So far, this effect, which
occurs at densities greater than about one-tenth of the critical density for
BEC, has been demonstrated only for excitons. Laser Doppler measure-
ments of the energy distribution of cold atoms may see Bose narrowing
in the near future, giving the first indication of their quantum degeneracy.
Another prediction for the weakly interacting gas is spatial condensa-
tion at the center of a three-dimensional potential well. As discussed in
the articles by Greytak and by Wolfe et al, this kind of trap is feasible for
both atoms and excitons. If BEC occurs in such a trap, the appearance
of an extremely narrow peak in the spatial distribution of the gas would
provide a dramatic "smoking gun."
Finally, weakly interacting gases can allow study of Bose-Einstein
statistics far from equilibrium. The process of stimulated emission,
familiar for photons, occurs for all scattering processes for bosons, since
scattering rates contain a factor (1 + Nf) for each final scattered Bose
particle, where Nf is the occupation number of the final state. This
enhancement of the density of final states, or stimulated scattering, is
opposite to the suppression of scattering of fermions by the factor (1 —
Nf) from the Pauli exclusion principle; occupied states "attract" other
bosons just as fermions "repel" like fermions. This effect can lead to an
 Introduction: Unifying Themes of Bose-Einstein Condensation 1
increase by orders of magnitude of the total scattering rate even in the
normal state, which should be observable in time-resolved studies of the
momentum spectra or spatial distribution of weakly interacting systems
created at or above the critical density but far from thermal equilibrium.
The case of short-lived biexcitons in the semiconductor CuCl illustrates
this effect. Although the biexciton lifetime is too short for a spontaneous
appearance of BEC, when a condensate is placed "by hand" via a laser
tuned to the ground state energy or very near to it, the biexcitons at low
energy "attract" other biexcitons to the same region of momentum space
[10]. Mysyrowicz reviews this work in CuCl here and discusses evidence
for the same effect in transport measurements in CU2O.
(v) The timescale for formation of BEC. One important feature in all
the weakly interacting boson systems discussed here is that the parti-
cles have, in general, a finite lifetime to remain in the system. Spin-
aligned atomic hydrogen can flip its spin (resulting in the formation of a
molecule), and atoms can evaporate from a trap. Excitons and positro-
nium are composed of particle-antiparticle pairs and can thus decay;
since these are not conserved, what are the experimental consequences
of the spontaneously broken gauge symmetry being only approximate?
To what extent can one describe systems with finite lifetime as Bose
condensed? For example, as discussed in several articles in this book,
excitons in CU2O appear to move without friction through the crystal
when they reach supercritical densities consistent with BEC. Should ex-
citons ever exhibit superfluidity? Kohn and Sherrington, in an oft-cited
paper [11], argued that excitons cannot be superfluid in a rigorous sense,
because they are particle-antiparticle pairs. Hanamura and Haug [12]
responded, however, that Kohn and Sherrington's argument depends
on an equilibrium picture, while an excitonic condensate is in fact an
inherently non-equilibrium ("quasiequilibrium") state. Keldysh briefly
addresses this issue in this book. Generally, the problem of seeing evi-
dence of superfluidity on timescales comparable to the particle lifetime
is challenging for both theory and experiment. Can one see analogs
of persistent currents, or of the Meissner effect in superconductors, or
reduced moment of inertia in He-II, and if so, what is the appropriate
superfluid mass density?
The controversial question of the timescale for onset of Bose-Einstein
condensation is reviewed at length in articles by Kagan and Stoof. The
general issue is the growth and evolution in space and time of the
boson correlation function, (\p\r)xp(O)). Several timescales are involved:
Suppose that a boson gas has density above the critical density for
8 D.W. Snoke and G. Baym
condensation, but is in a metastable state with no condensate present.
How long does the quasiparticle distribution of the system take to develop
a peak at zero kinetic energy of width less than /CBT? HOW long does
it take to develop a quasicondensate, i.e., a condensate whose magnitude
is locally that of the equilibrium condensate, but with phase fluctuations
arising from a random macroscopic distribution of topological defects?
Then how long does it take for the condensate to achieve long-range
phase coherence, leading to the characteristic delta function at p = 0 in
the momentum distribution? Finally, when does superfluidity appear?
(vi) Two-dimensional degenerate boson systems. Formation of a quasi-
condensate is also the basis of the explanation of the Kosterlitz-Thouless
(KT) transition of a boson gas in two dimensions, which leads to super-
fluidity even though a true condensate cannot appear. The KT transition
has been seen for (two-dimensional) helium films [7]; Silvera reviews
work on observation of a KT transition of spin-polarized hydrogen ad-
sorbed on a surface, and Matsubara et al. briefly report on recent
experimental work toward this goal. Semiconductor quantum-well struc-
tures provide another natural way to examine physics in two dimensions.
One such system contains as excitations the "dumbbell exciton" in which
the electron and hole in a two-dimensional well are separated by a po-
tential barrier[13] or by an applied electric field [14]. Early indications
[14] of a KT transition in this type of structure were later realized to be
obscured by trapping in random potentials created by surface irregular-
ities (similar to the scenario considered by Huang in this book) which
actually produced a Fermi-Dirac distribution of excitons [15]. The possi-
bility nevertheless still remains for observation of the Kosterlitz-Thouless
transition with better-grown quantum-well structures.
Various kinds of two-dimensional semiconductor structures have been
proposed which contain charged bosonic excitations that could in prin-
ciple become superfluid, conceivably at room temperature, and which
would be superconductors. "Magnetoexcitons," which have received
much attention in recent years [16], move perpendicular to applied mag-
netic and electric fields in a two-dimensional well, and could undergo
Bose-Einstein condensation [17, 18]. In this volume Shvartsman and
Golub speculate that the possibility exists for engineering the band struc-
ture of GaAs to yield a "bihole," a boson in two dimensions with charge
+2e. Since band-structure engineering of semiconductor structures has
reached such an advanced state, one can imagine great advances in the
near future in the manufacture of new kinds of superconductors based
on BEC.
 Introduction: Unifying Themes of Bose-Einstein Condensation 9
(vii) BEC and lasing. While excitons in CU2O have lifetimes long
compared to interaction times, in most semiconductors the interaction
time is comparable to the lifetime of the excitons due to coupling to the
photon field. In the limit of very short exciton lifetime, the excitonic BEC
essentially becomes indistinguishable from a laser. The observation of a
narrow photon emission peak from a semiconductor does not therefore
immediately indicate the presence of an excitonic BEC, since in many
cases it may equally well be described as superradiant emission due
to lasing [12]. As discussed by Keldysh in this volume, lasing and
excitonic (or biexcitonic) BEC can be seen as two limits of the same
theory. Lasing occurs in the case of strong electron-photon coupling
(recombination rate large compared to the interparticle scattering rate)
while the excitonic condensate occurs in the case of weak electron-photon
coupling (recombination rate slow compared to interparticle scattering
rate). A laser, in fact, can be seen as a Bose-Einstein condensate in which
the long-range order involves the photon states, while in the excitonic
condensate the long-range order involves the electronic states [19]. Many
of the same issues of the timescale for onset of condensation, discussed
above, have already been discussed in the context of the onset of lasing
[20]. The polariton condensate, reviewed by Keldysh (see also Beloussov
and Shvera) in this volume, represents a middle case between lasing
and excitonic BEC, since the polariton has a mixed character with both
exciton-like and photon-like behavior.
The overlap between BEC and lasing has been illustrated recently
in the context of understanding the origin of dark matter in the early
universe, where it has been argued that decay of a heavy species of
neutrino into a light fermion and a boson could give rise either to a
Bose-Einstein condensate [21] or to a "neutrino laser" [22].
(viii) BEC of small clusters of particles. The usual theory of BEC
assumes the thermodynamic limit of an infinite system. The concept
of bosonization, including BEC and superfluidity, has become valuable
in understanding the properties of nuclei, however, in which pairs of
nucleons bind via the strong force to produce effective bosonic degrees
of freedom. Iachello reviews the theory of bosonization; Ropke as well
as Hellmich et al. discuss the crossover from BEC to BCS in the theory
of nuclear matter. These concepts work quite well in small nuclei, which
are systems far from the thermodynamic limit; this is perhaps not so
surprising since numerical models of helium using of order 64 atoms have
been shown to reproduce the bulk superfluid properties of helium [23].
The theory of BEC of small clusters of bosons [24] may find extension in
10 D. W. Snoke and G. Baym
the future to atomic and molecular systems, since the experimental study
of atomic clusters has progressed tremendously in recent years. What
is the minimum size of a system that we may still call a Bose-Einstein
condensate?
(ix) BEC in random potentials. The appearance of a Bose condensate
in disordered systems is a very active area of research in condensed-
matter physics (the so-called "dirty boson" problem.) The key issues are
highlighted in the review article by Nozieres, and some specific aspects are
treated in the articles by Huang and Stringari, but this important topic is
not reviewed in depth in this book. Experimental systems to which this
theory may apply include excitons in quantum wells with rough surfaces,
and helium in porous media such as Vycor, the superconductor-insulator
transition in disordered films, and vortices in type-II superconductors.
As one can see from this short survey, the abundant themes in BEC
go beyond specific system properties, and present a unifying basis to the
study of the phenomenon in different systems. We hope that this volume
will serve as a stimulus to deeper understanding of these issues.

Acknowledgements. We appreciate the useful comments of A. Griffin.

This work has been supported in part by the Aerospace Sponsored
Research Program (D.S.) and by National Science Foundation Grant
DMR91-22385 (G.B.)

References
[1] J. Gasser and H. Leutwyler, Phys. Reports 87, 77 (1982).
[2] C. T. Hill, Phys. Lett. B 266, 419 (1991) and references therein.
[3] See, e.g., International conferences on quark matter and ultra-
relativistic nucleus-nucleus collisions, Nucl. Phys. A 544 (1992); in
press (1994).
[4] P. C. Hohenberg and P. M. Platzman, Phys. Rev. 152, 198 (1966).
[5] D.M. Ceperley and E.L. Pollock, Phys. Rev. Lett. 56, 351 (1986).
[6] A. Griffin, Excitations in a Bose-Condensed Liquid (Cambridge Uni-
versity Press, Cambridge, 1993).
[7] For a review, see D.R. Nelson, in Phase Transitions and Critical
Phenomena 7, C. Domb and J.L. Lebowitz, eds. (Academic, New
York, 1983).
[8] M. Rho, "Cheshire Cat Hadrons," Phys. Reports, in press (1994).
[9] See, e.g., V.N. Popov, Functional Integrals and Collective Excitations,
(Cambridge University Press, Cambridge, 1987), ch. 6.
 Introduction: Unifying Themes of Bose-Einstein Condensation 11
[10] N. Peghambarian, L.L. Chase, and A. Mysyrowicz, Phys. Rev. B 27,
2325 (1983).
[11] W. Kohn and D. Sherrington, Rev. Mod. Phys. 42, 1 (1970).
[12] E. Hanamura and H. Haug, Sol. State Comm. 15, 1567 (1974); Phys.
Reports 33, 209 (1977).
[13] KM. Peeters and J.E. Golub, Phys. Rev. B 43, 5159 (1991).
[14] T. Fukuzawa et a/., Phys. Rev. Lett. 64, 3066 (1990).
[15] J.A. Kash et al, Phys. Rev. Lett. 66, 2247 (1991).
[16] D.S. Chemla J.B. Stark, and W.H. Knox, Ultrafast Phenomena VIII,
(Springer, Berlin, 1993), p. 21.
[17] I.V. Lerner and Yu. E. Lozovik, Sov. Phys. JETP 53, 763 (1981).
[18] D. Paquet et al, Phys. Rev. B 32, 5208 (1985).
[19] The analogy between a laser and superfluid 4 He is nicely reviewed
by PC. Martin, in Low Temperature Physics— LT9, J.E. Daunt,
D.O. Edwards, F.J. Milford, and M. Yaqub, eds. (Plenum, New
York, 1965), p. 9.
[20] See, e.g., H. Haken, Rev. Mod. Phys. 47, 67 (1975).
[21] J. Madsen, Phys. Rev. Lett. 69, 571 (1992).
[22] N. Kaiser, R. A. Maleney, and G. D. Starkman, Phys. Rev. Letters
71, 1128 (1993).
[23] Finite size effects are discussed by E.L. Pollock, Phys. Rev. B 46,
3535 (1992).
[24] Ph. Sindzingre, M. Klein and D. M. Ceperley, Phys. Rev. Lett. 63,
1601 (1989).
 Part one
Review Papers
 Some Comments on Bose-Einstein
Condensation
P. Nozieres
Institut Laue-Langevin
B.P. 156, 38042 Grenoble Cedex 9
France

Abstract
The paper reviews some important features of Bose-Einstein condensation
such as the nature of the condensate, exchange interactions, phase locking.
The competition with other physical effects (crystallization, dissociation,
localization) is briefly discussed.

Bose-Einstein condensation was first described for an ideal gas of free

bosons, with mass m, density N. The chemical potential \i is obtained
from the conditions
1

where 8k = h2k2/2m. In a dimension d > 2, [i reaches 0 at some

temperature Tc such that

where v(s) ~ e.i 1 is the one particle density of states. Below Tc a

macroscopic number of particles No accumulates in the lowest state, \i
being locked to 0 (Fig.l). When d < 2 the integral in (1) diverges:
Bose-Einstein condensation does not occur (put another way, Tc is 0).
These conclusions hold for a uniform infinite system. It was pointed
out by Bagnato and Kleppner [1] that the condensation can be restored
at low dimensions if the system is put in a confining potential

Then the region accessible to a particle with energy s has a radius L

15
16 P. Nozieres

(1)
o
r^ ^ T
p i

c N

Fig. 1. The chemical potential \L and condensate population no of an ideal Bose

gas as a function of temperature.

Fig. 2. A sketch of the critical temperature of a confined ideal Bose gas as a

function of the exponent rj in the confining potential for d — 1 and d = 2 (after
Ref. 1).

and the corresponding density of states behaves as

v(e) - Ld£d2~ £a, with a = - + •= - 1.

rj 2
Playing with the exponent rj one can make the integral in (1) convergent
and Tc finite. The result is sketched in Fig. 2.

1 Fragmentation of the Condensate

The above standard wisdom actually leaves one central question unan-
swered: do the condensate particles accumulate in a single state? Why
don't they share between several states that are degenerate, or at least so
close in energy that it makes no difference in the thermodynamic limit?
 Some Comments on Bose-Einstein Condensation 17
The answer is non-trivial: it is the exchange interaction energy that makes
condensate fragmentation costly. Genuine Bose-Einstein condensation is
not an ideal gas effect: it implies interacting particles!
Consider the simplest type of scalar interaction for structureless parti-
cles
V
= 2 Yl VqKK'bk'-qbk+q-

The fully condensed ground state has all particles in the lowest state
N
J5) |vac>. (2)
The corresponding interaction energy is
2
\v(b*b*bb)
Eo = \v ^VN(Nl)
o(b*ob*obobo) = ^VoN(N-l) ~ ±VN
±V0N
Note that Vo is necessarily > 0 (i.e., repulsive): otherwise the system
would collapse spontaneously to infinite densities JV.
We compare that state with one in which the condensate is shared
between states 1 and 2, with occupancies JVi and N2 = N — N\

IW = J^^^'^^lvac). (3)
The kinetic energies are both close to 0 and negligible. The interaction
energy involves all possible contractions of operators in the expectation
value of V. It contains a Hartree term (between all species) and a Fock
exchange term. The latter only occurs between different species, since
there is only one type of contraction in the expectation value of b\b\b\b\.
Hence
Hartree Fock

I 1
E = y [ATi(ATi - 1) + N2(N2 - 1) + 2iViJV2] + V^
(4)

Since Vo > 0 fragmentation of the condensate costs a macroscopic

extensive exchange energy. Genuine Bose-Einstein condensation is due to
exchange, a feature which is often overlooked.
It may be objected that the use of Hartree-Fock wave functions is
unwarranted. Indeed, short range correlations modify the wave function
at short distances between particles, thereby leading to a depletion of
the condensate (NQ < N at T = 0). The ground state is complicated,
18 P. Nozieres
leading to all the subtleties of Bose liquid theory. The argument against
fragmentation nevertheless remains true, as it relies on the comparison of
two situations with the same amount of depletion and virtual excitation,
with or without fragmentation. The only difference is in the Fock
"intracondensate" exchange, everything else remaining the same. This
exchange energy is reduced by depletion, but it remains extensive.
The crucial role of particle interactions in Bose liquids is well known.
It is responsible for the finite compressibility of the liquid, and therefore
for the existence of sound modes at long wave length instead of free
particle excitations fa ^ k instead of k2). It leads to a finite coherence
length

As a result the effect of confinement in one and two-dimensions dis-

appears [2], In two-dimensions a transition remains which is of the
Kosterlitz-Thouless type (unbinding of vortices) - yet another conse-
quence of interactions. The effect of interactions is equally important for
the kinetics of condensation in a thermally quenched normal gas. While
for a free gas the energy narrowing of the No condensed particles is very
slow (the corresponding width decays as l/t)9 exchange effects make the
relaxation time atomic once the condensate has been nucleated.

2 Internal Degrees of Freedom

Assume now that the bosons have an internal quantum number a. This
may be nuclear spin for polarized atomic hydrogen, or the electric dipole
moment for optically active excitons, or orbital and spin total momentum
for a triplet pair of 3He atoms. Is the condensate a pure state or a mixture?
We may describe it by a density matrix

which can be diagonalized in the appropriate basis

Is there one or several non-zero na?

In order to answer this question we write the interaction within the
condensate as
H xp b*Oab*^b^'bo^
 Some Comments on Bose-Einstein Condensation 19
from which we infer the condensate energy

(( ) (5)

a,a'

Here we neglect the depletion via virtual excitation which should also
depend on the internal state a. The selection argument relies only on
intracondensate interactions, which is certainly an approximation. We
note first that the energy £o must be positive for any choice of na > 0:
otherwise the density would collapse. The nature of the minimum then
depends on the matrix A. Consider, for instance, a two-fold situation

A= [ 1 => EQ = e[n\ + n2]2 + 2(rj — e)n\n 2.

\n s J
The ground state is a pure state if rj > e (n\ or n2 = 0). It is, on the
contrary, a mixture if rj < e (n\ = n2 = 1/2). Note that in the Hartree-
Fock approximation the selection occurs to first order in V. It may be
that the degeneracy is not lifted at that order: then one must go beyond
Hartree-Fock.
As an example consider polarized atomic hydrogen, H|. In a magnetic
field the atoms undergo a nuclear Zeeman splitting, ±fiB. The lowest
energy state is undoubtedly a pure state in which all particles are in the
lowest nuclear state. But one can work at given magnetization Mz =
Nj — N±. If the condensate atoms are in a pure spin 1/2 state, a transverse
magnetization appears, the same for every atom, as shown in Fig. 3.
Internal coherence of the condensate implies nuclear ferromagnetism (of
course this magnetization precesses at the Larmor frequency). Conversely,
if one allows for a mixture of I and f atoms the transverse M can be
killed.
In practice the most obvious interactions are

(i) A scalar interaction

yp\P2 -> H tf = V5apSatp'.

The corresponding interaction energy

does not depend on the condensate structure.

 20 P. Nozieres

Mixture Pure state

Fig. 3. The onset of transverse ferromagnetism for a pure state Hj condensate.

(ii) An exchange interaction JS\ • S2 which is equally ineffective in lifting

the degeneracy (orbital symmetry implies a triplet spin state for a pair
of bosons, such that Si • S2 = 1/4). An anisotropic exchange, Jz ^= Jj_,
would probably do the job, but where does it come from?

3 Condensate Fluctuations and Phase Locking

We return to structureless bosons. Ensuring a non-fragmented condensate
we gain the Fock exchange energy,firstorder in V. Actually one can do
better allowing for coherent fluctuations of the condensate population
No. Instead of state (2) let us try a superposition
IVo) =e # 0 |vac); (6)
0 is a complex order parameter, the phase 8 of which is locked to a
specific value. Locking 8 means fluctuations in N. Indeed, N and 8 are
canonically conjugate: state (2) is obtained by an integration over 8

d8e-iNe\xp0(8)).
Jo
It is easily verified that (N) = |0| 2 : the energy is the same in (2) and (6).
The situation is different when we consider interaction terms such as
b^bkb-k + c.c.
which allow virtual excitations of two particles out of the condensate.
In order to take such a hybridization into account, we try a variational
wave function
\xp0) = (7)
 Some Comments on Bose-Einstein Condensation 21
Afe is obtained minimizing the energy. Equation (7) is nothing but the well-
known Bogoliubov approximation for Bose liquids. When calculating the
expectation value of the energy, we recover the usual Hartree and Fock
exchange contributions, which here have corrections of order
V0\h\2 , Vk\Xk\2.
They are the same whether N or 9 is locked. The novelty is the so-called
Bogoliubov term, which appears only when the phase is locked

Such a linear coupling between <j> and kk always lowers the energy, by
an extensive amount of order V2: hybridization is bound to occur. It
follows that phase locking is implied by Bose-Einstein condensation. Note
that the Bogoliubov energy favours pure state condensates: it is enough
to lift the degeneracy.
Phase locking is a genuine symmetry breaking, distinct from Bose-
Einstein condensation. It is responsible for superfluidity (superfluid
currents are due to a gradient in the locked phase 6). Its physical
consequences are well known : quantization of circulation in multiples
of h/m, vortices, Josephson effect through a weak link, etc.
As an unconventional example let us consider Bose condensation of
excitons in CU2O, extensively discussed by Wolfe et al. and Mysyrow-
icz [3]. These excitons exist in two brands, para (lowest in energy) and
ortho (some 12 meV higher). They are seen through phonon assisted
fluorescence (momentum k is not conserved), from which one infers the
exciton energy distribution n(e). Most of the data are concerned with
orthoexcitons, less "forbidden" than the para. As the density N grows,
quantum deviations to n(e) are clearly visible, but somehow one never
reaches Bose-Einstein condensation. One "surfs" on the transition line
without being able to cross it - why?
A tentative explanation might go as follows. Being lower, the paraex-
citons should be more numerous and consequently they should Bose
condense first. But they are harder to see: their condensation may es-
cape observation. Assume now that the ortho density grows through the
transition. Then a Josephson process is possible
O + O-+P +P.
Direct conversion of a single exciton is forbidden by symmetry, but
there is no such problem for a pair. That Josephson current is a.c, at
a frequency co corresponding to the 12 meV splitting - hence a strong
22 P. Nozieres
dissipation that precludes further cooling. One thereby would explain
why any onset of ortho condensation would immediately stop the process.
The above example may be wrong, but it points to an essential fact:
phase locking is a crucial feature of condensed systems.

4 The Enemies of Bose-Einstein Condensation

Condensation of bosons, or of fermion pairs through the appropriate
generalization, is a fragile effect that may easily be superseded by other
instabilities. We now survey some of the main competitors.

4.1 Crystallization
The energy per particle of a liquid phase is roughly an interaction Vm
averaged over a unit cell (the distance between atoms is free to vary,
except for some short range correlations). This may be compared with a
solid phase in which the particles are localized at lattice sites. The latter
energy contains a kinetic part due to localization of the particles, and a
potential energy Va corresponding to the lattice spacing a

N ma2
Because the repulsive potential V(r) decreases with r, we have Va < Vm.
It follows that the crystalline state will be favoured for strong coupling (it
optimizes the potential energy), while the Bose condensed liquid will win
at weak coupling (it minimizes the kinetic energy). An obvious example
is 4He, which is a superfluid at low pressure and which freezes at higher
densities.
Note that the kinetic energy that is relevant to the above competition
is that associated to coherent hopping of the particle from place to
place, from which one may construct Bloch states (remember that phase
coherence is an essential ingredient). Such a coherence may be very hard
to achieve. Consider for instance bipolarons, i.e. two fermions trapped in
the same cloud of phonons that they have generated. These bipolarons
are candidates for high temperature superconductivity. What matters
here is elastic coherent hopping of the bipolaron, leaving the phonon
bath in its ground state. The corresponding amplitude W involves the
overlap of phonon ground states O, and O7 when the particle is at sites i
or;,
W =
 Some Comments on Bose-Einstein Condensation 23
This overlap is usually very small and the kinetic energy is negligible
as compared to potential energy. Charge density waves (crystallization)
win!
The above competition is nicely formulated in the old lattice gas model
of Matsubara and Matsuda [4]. Consider N hard core bosons on 2N
sites. Due to the single occupancy constraint, each site has two states,
empty or filled. The bosons have commutation rules on different sites,
which is just the algebra of spins. The hard core boson gas is therefore
isomorphous to a spin 1/2 system. The hopping amplitude between
nearest neighbours is equivalent to a transverse (xy) magnetic coupling:

b*b) -» SfSj -> energy J±.

A nearest neighbour repulsion is equivalent to an Ising coupling

ntnj -> SizSjz - • energy J z.

Hence there are two possibilities:

(i) Jz > J±: the equivalent magnetic system is an Ising antiferromagnet,

corresponding to a crystal in which one sublattice is empty and the
other occupied.
(ii) Jz < J±: the magnetic ground state is an (xy) transverse ferromagnet.
The transverse magnetization is the superfluid order parameter, its
azimuth being the phase 9. The competition between kinetic energy
and interactions is clear. One can enrich the model by adding further
couplings, and it is easy to construct a phase diagram using a mean-
field approximation [5].

An interesting issue is the possible coexistence of Bose-Einstein con-

densation and crystallization. Put another way do super solids exist? The
question is relevant for vacancies in solid 4He. It has been argued [6]
that zero point vacancies could persist at zero temperature (if the tun-
neling bandwidth is larger than the creation energy), and indeed there
is experimental evidence for an unusually large concentration of such
vacancies [7]. If that holds vacancies might become superfluid, leading
to unusual Herring-Nabarro plastic flow of solid 4He. In practice this is
unlikely since vacancy coherent hopping is subject to a strong polaronic
reduction.
 24 P. Nozieres
4.2 Dissociation of Composite Bosons
We now turn to the more frequent case of bosons made out of two
fermions, appropriately bound, as they occur in superconductors or
excitons. Let Cka be the fermion destruction operator. The condensing
entity is a bound pair characterized by

q is the total momentum of the pair, cpk its internal wave function. The
Hartree ground state for the pairs (equivalent to (6)) is just

in which we have set

(8) is nothing but the usual BCS wave function, viewed here as pair
Bose-Einstein condensation. Vk is a variational parameter which allows
an interpolation between low and high densities. (Note that (8) ignores
the effect of screening and its interplay with binding: the real problem of
Mott transitions is not tackled.)
The superfluid state (8) exists only if the particles are subject to an
attraction. Two limiting cases are simple:
(i) In the dilute "molecular" limit the attraction must be strong enough
to create a bound state of two fermions ("dilute" means a particle
separation large as compared to the bound state radius). The bound
pairs behave as "point" bosons that condense in the usual way. Vk is
<C 1 and it reflects the internal state of the underlying fermions. The
gap A is the molecular binding energy, while the critical temperature
Tc is controlled by centre of mass motion of the bound pairs. For free
bosons Tc ~ AT5, while for a Hubbard lattice gas Tc « 1/17 (one must
virtually break the pair in order to make it hop).
(ii) In the opposite dense limit, the pairs strongly overlap and the exclusion
principle becomes the dominant feature. \vk\2 cannot exceed 1: it
extends to higher k in order to accommodate the N particles - in
practice to the Fermi momentum kF. In the limit of high densities
we recover the usual BCS result, with an exponentially small gap A.
The fermions are normal except for a narrow region of width A near
 Some Comments on Bose-Einstein Condensation 25

Fig. 4. The evolution of the BCS parameter x>\ as a function of energy efc for
increasing densities. Note the saturation due to the exclusion principle.

kf. Binding is now a cooperative effect, which occurs only due to

Bose-Einstein condensation (in contrast to the dilute limit in which
the two effects were distinct). The critical temperature Tc is controlled
by pair breaking: it is of order A.

In between these two limits the evolution is smooth, as shown on

Fig. 4: contrary to what has been sometimes claimed, superconductivity
is indeed Bose-Einstein condensation of pairs. The transition occurs
when the pair binding energy £# is comparable to the Fermi energy sp.
The behaviour of the gap A and chemical potential \x as a function of N
is sketched in Fig. 5.
While the condensation of dilute "point" pairs is a robust phenomenon,
the BCS state is fragile when A <C ep. If the interaction near the
Fermi surface has repulsive parts, Bose condensation may well disappear,
leading to a standard normal Fermi liquid. In the case of dense excitons,
any anisotropy will kill condensation: the Fermi surfaces of electrons and
holes need not be the same and pairing cannot occur. We conclude that
overlap and resulting dissociation of composite bosons is antagonizing to
Bose-Einstein condensation.

43 Disorder and Localization

Superfluidity implies a long range phase coherence, while localization
means no coherent extended state: clearly they are antagonist. In practice,
two very different localization mechanisms must be considered:

(i) Mott localization is due to interaction between particles. When the

number of particles N is commensurate to the number of sites iVL,
 26 P. Nozieres

Fig. 5. T h e evolution of the gap A a n d chemical potential j u a s a function of

density for attracting bosons.

the ground state becomes insulating beyond a critical interaction U c.

Such a localization is the ultimate stage of crystallization, above and
beyond appearance of a periodic order (a real solid is both periodic
and localized). The hard core Bose lattice mentioned earlier is one such
example. Mott transition occurs equally well for a full lattice, N = NL,
past a critical Uc. (In that case it is pure, unspoiled by charge density
waves.)
(ii) Anderson localization is due to disorder. A random one-body potential
creates destructive interference on the wave function. At low enough
kinetic energy the long range phase coherence is suppressed. For
moderate disorder a mobility edge s* appears on either side of the
electron band, separating extended states in the middle and localized
states near the edges. If the disorder is strong enough the edges may
close in: no extended state subsists and localization is complete.

Strictly speaking, the effect does not depend on statistics - except that
something must prevent accumulation of particles in the lowest local-
ized states. For fermions the exclusion principle does it. For bosons
one must rely on repulsive interactions that forbid large local densi-
ties. At first sight one expects a transition from a localized state at
small coupling - the so-called Bose glass - to a regular superfluid at
large coupling, when interactions force the chemical potential \i into
the region of extended states. But one might also argue that strong
 Some Comments on Bose-Einstein Condensation 27

Fig. 6. A sketch of the phase diagram of a disordered two-dimensional Bose

lattice gas with N = NL (after Ref. 9).

local repulsion produces constraints that favour localization. Such a

contradiction gives a feeling of how primitive the present theoretical
understanding is. The subject is indeed "hot" and a number of at-
tempts have been put forward [8]. Generally speaking, they fall in two
categories:
(i) Either they put emphasis on localization of individual particles, referred
to the mobility edge,
(ii) Or they consider the rigidity of the superfluid phase. Breakdown of
superfluidity corresponds to a vanishing rigidity - a truly collective
point of view.
Efforts to characterize the critical behaviour are somewhat prema-
ture, since even the qualitative nature of the effect is not well under-
stood.
Extensive numerical work has been performed on lattices, where An-
derson localization is mixed with the Mott transition. They confirm the
above qualitative picture. Results of Trivedi [9] for a two-dimensional
lattice with N = NL are shown in Fig. 6: the Bose glass is found at strong
disorder - yet another enemy of Bose condensation.
The problem of disordered Bose liquids is not a theorist's amusement:
it has important experimental applications. Liquid 4He in a porous
medium (e.g. Vycor) is one example. A Cooper pair in disordered al-
loys is another candidate. A particularly fascinating problem is that of
disordered thin films in which disorder is monitored by film thickness
d. One can make d much smaller than the film superconducting co-
herence length £,, in which case the system is two-dimensional. When
28 P. Nozieres

Fig. 7. The resistance of Bismuth films as a function of temperature for varying

thickness d (after Ref. 10).

the square resistance is comparable to the quantum scale h/4e2 a sharp

transition occurs at T = 0 as a function of d, between a superconducting
and an insulating state. This is shown dramatically in Fig. 7, which
displays the resistance of Bismuth films as a function of temperature for
varying d [10]: the direct passage from 0 to oo at zero temperature is
spectacular.
The problem is important and, to a large extent, open. It is attacked
either via renormalization methods or through numerical simulation. But
we are far from a real understanding. Even simple questions such as
the relative importance of disorder scale (vs £) and disorder strength (vs
h2/m£2) are not answered.
 Some Comments on Bose-Einstein Condensation 29
5 Conclusion
In this brief review we have emphasized a few simple questions and
remarks:
(i) Bose-Einstein condensation is more subtle than is usually assumed.
For instance the nature of the condensate (fragmentation, pure state
or mixture) is a non-trivial issue.

(ii) Interactions between particles are absolutely crucial - one may say
that genuine condensation is an effect of exchange coupling.

(iii) Bose-Einstein condensation is fragile. It competes with other instabil-

ities and it may be destroyed by disorder, dissociation, etc.
One question remains: can a Bose liquid be normal, breaking no
symmetry whatsoever (translation, gauge, etc.)? We argued that a hard
core Bose liquid is isomorphic to a spin 1/2 system: a normal state
would correspond to a singlet ground state of that equivalent magnetic
system. Clearly this is what happens for composite bosons when they
turn into a normal Fermi liquid (the ground state of which is a singlet).
Is such a normal state possible for point bosons? It would correspond
to what present wisdom calls an "RVB" state. Rather than playing with
uncontrolled approximate calculations for fermions, it would be much
better to start with point bosons. It has been known for a long time
that interactions deplete the condensate. Can that depletion proceed to
100%? If it does, what is the physical nature of the resulting state?
These questions are among the outstanding puzzles of Bose-Einstein
condensation.

Acknowledgements. This lecture was not given at the Trento workshop,

but at a meeting held in Paris at the Ecole Normale Superieure on 9 June,
1993 in honour of Claude Cohen-Tannoudji on his sixtieth birthday. I
would like to take this occasion to renew my congratulations to Claude:
Bose-Einstein condensation is far from his usual worries, but I am acting
as "le jongleur de Notre Dame". I am also grateful to Franck Laloe for
many passionate discussions - and for his disagreements!

References
[1] V. Bagnato and D. Kleppner Phys. Rev. A 44, 7439 (1991).
[2] S.I. Sevchenko, Sov. Phys. JETP 73, 1009 (1991).
30 P. Nozieres
[3] J.R Wolfe et al. and A. Mysyrowicz, this volume.
[4] T. Matsubara and H. Matsuda, Prog. Theoret. Phys. 16, 569 (1956);
17, 19 (1957).
[5] K.S. Liu and M.E. Fisher, J. Low Temp. Phys. 10, 655 (1973).
[6] A.F. Andreev and I.M. Lifshitz, Sov. Phys. JETP 29, 1107 (1969).
[7] G.A. Lengua and J.M. Goodkind, J. Low Temp. Phys. 79, 251 (1990).
[8] R. Ma, B.I. Halperin and PA. Lee, Phys. Rev. B 34, 3136 (1986);
M.P.A. Fisher, RB. Weichman, G. Grinstein and D.S. Fisher, Phys.
Rev. B 40, 546 (1989). An excellent review by TV. Ramakrishnan is
in press.
[9] N. Trivedi, in Computer Simulation Studies in Condensed Matter
Physics V, (Springer, Berlin, 1993); W. Krauth, N. Trivedi and D.
M. Ceperley, Phys. Rev. Lett. 67, 2307 (1991).
[10] D.B. Haviland, Y. Liu and A.M. Goldman, Phys. Rev. Lett. 62, 2180
(1989).
 Bose-Einstein Condensation and
Superfluidity
Kerson Huang
Department of Physics
and
Center for Theoretical Physics, Laboratory for Nuclear Science
Massachusetts Institute of Technology
Cambridge, MA 02139
USA

Abstract
We review generally accepted definitions of Bose-Einstein condensation and
superfluidity, emphasizing that they are independent concepts. These ideas
are illustrated in a dilute hard-sphere Bose gas, which is relevant to ex-
periments on excitons and spin-aligned atomic hydrogen. We then discuss
superfluid 4He in porous media, as simulated by different models in differ-
ent regimes. At low coverage, we model it by a dilute hard-sphere Bose gas
in random potentials, and show that superfluidity is destroyed through the
pinning of the Bose condensate by the external potentials. At full coverage,
we model the random medium by an ohmic network of random resistors,
and argue that the superfluid transition is a percolation transition in d=3,
with critical exponent 1.7.

This book is devoted to the phenomenon of Bose-Einstein condensa-

tion [1, 2] and inevitably, its relevance to superfluidity [3]. To provide
some background for other articles in this volume, I would like to summa-
rize some commonly accepted views on these phenomena, and illustrate
them in the context of a dilute hard-sphere Bose gas, a model in which
we have some control over the approximations made. I will also describe
some recent work on the effect of randomness on the Bose condensate,
which shows that Bose-Einstein condensation does not automatically
give rise to superfluidity.

1 Bose-Einstein Condensation
In an ideal Bose gas in three spatial dimensions (d=3), Bose-Einstein
condensation occurs when there is a finite density of particles in the

31
32 K. Huang
zero-momentum (k=0) state:

(alao)>O (1)

where a^ is the annihilation operator for the single-particle state of

momentum ftk, Q is the total volume and ( ) denotes thermal averaging.
In an interacting system, a Bose condensate can be defined in the same
manner, with thermal averaging taken with respect to the interacting
Hamiltonian. The Bose-Einstein condensation is a first-order phase
transition in an ideal gas, but not in interacting systems generally.
The condensate density appears in a correlation function sometimes
called the "one-particle density matrix":

f ~ e^\a\ak) (2)

where xp(x) is the boson field operator. The relation

(3)

has been used to theoretically estimate [4], and experimentally deter-

mine [5] the condensate fraction in 4He, with the result no/n « 0.1 in the
limit T = 0.
It has taken nearly forty years from the first discussion of Bose-
Einstein condensation to the realization that its true essence lies in
the twin phenomena of broken gauge symmetry, and phase coherence.
These are embodied in what is now generally accepted as the criterion
for Bose-Einstein condensation:
(ipHx)ip(y)) —+ <v>(x)»(y)> (4)
|x-y|->oo

where (y>(x)) ^= 0 is called the "condensate wave function", which, being

a complex number, has a magnitude and a phase. Broken symmetry
refers to the fact that, when the condensate wave function is non-
zero, the vacuum state must depend on its phase, even though the
Hamiltonian is invariant under a constant phase change of xp(x) (global
gauge invariance). Phase coherence refers to the fact that, in order to
have non-vanishing thermal average, the phase of xp(x) must be spatially
correlated over the entire system. The criterion (4) was first proposed
by Penrose and Onsager [4], and its significance emerged gradually,
through the work of many people, among them Goldstone [6], who
advanced the idea of spontaneous symmetry breaking, Yang [7], who
 Bose-Einstein Condensation and Superfluidity 33
termed the phenomenon "off-diagonal long-range order (ODLRO)", and
Anderson [8], who emphasized the phase coherence.
It is instructive to repeat Anderson's argument [8,9] that Bose-Einstein
condensation implies phase coherence. Suppose, for simplicity, that all N
particles in the system are in the fc=0 state. Imagine that the system is
subdivided into K macroscopic boxes of the same size. Thermodynamic
extensiveness requires that each box contain N/K particles in the fc=0
state. Let At be the annihilation operator of fc=0 particles in the ith box.
We must have

= N (5)
(A}Ai) = N/K
which lead to the relation (ajao) = K"1 ^(AJAi)+K~l J2i+j(AlAj)> or

N
= £ + £ H(AUj) (6)
For large K, the second term on the right side must dominate, and
therefore the phases of the At in different boxes must be correlated.
The correct order parameter for Bose-Einstein condensation is thus
the condensate wave function. As defined, however, it is a micro-
scopic property of the system, and we can rarely make headway with
microscopic calculations. It is more practical to adopt the Ginsberg-
Landau approach, and introduce a local order parameter </>(r) that is a
coarse-grained version of (y>(r)):
</>(r) = jR{r)em (7)
whose phase is related to the superfluid velocity through

ys = - V a (8)
m
where m is the boson mass. Since 0 should be single-valued, the change
of a over any closed path must be a multiple of 2n. This leads to the
Onsager-Feynman quantization condition [10, 11]
0c = 0,±l,±2,---) (9)
m
The Landau free energy is taken to be [12]

= Jddx [|^|Vtf>| 2 + c2\(t>\2 + c4\4>\* + ••] (10)

34 K. Huang
As usual, in d=3 we put C2 = y(T — Tc), to model the Bose-Einstein
condensation as a second-order phase transition. The dots in (10) rep-
resent terms that can be neglected near the transition point, when <f>
is small. The partition function is given by the functional integral
Z = J(D(l))(D(l)*)e~E[(j>]/kT. Critical exponents associated with the tran-
sition can be calculated in the mean-field approximation, and, more
elaborately, in a power series expansion in e, with d = 4 — e [13]. By
regarding (10) as a Hamiltonian, we can derive an equation of motion:

-^nV2<t> + c2<i> + cA\4,\24, = ihd-^ (li)

which is referred to as "the non-linear Schrodinger equation" (NLSE), or
Gross-Pitaevsky equation [14, 15]. It admits static vortex solutions [12],
spawns vortices in flows past obstacles [16], and can be used to discuss
the kinetics of Bose-Einstein condensation [17]. The Ginsberg-Landau
theory needs retooling in d=2, because strong phase fluctuations of the
local order parameter destroy the Bose condensate. To see this in a
simple way, let us assume that \<t>\ > 0, due to the potential-energy terms
in (10). Ignoring amplitude fluctuations, we consider the correlation
function
C(r) = (eia(r)e-ia(0)) (12)
where the average is taken with Gaussian weight (i.e., with only the
kinetic term in the Landau free energy). The calculation is elementary,
and yields C(r) oc e~G^T\ where G(r) is the Green's function of the Laplace
equation: G(r) oc lnr for d=2, and G(r) oc 1/r for d=3. Thus, up to a
multiplicative constant,

(d=3)
where b is a constant. We see that phase coherence is maintained to
infinite distances in d=3, but not in d=2 at finite temperatures. A
rigorous proof of the absence of Bose-Einstein condensation in d=2 at
finite temperatures was given by Hohenberg [18]. Equivalent statements
are that long-ranged order does not exist in d=2 (Mermin-Wagner
theorem) [19], and that spontaneous symmetry breaking does not occur
in d=2 (Coleman's theorem) [20].
Even though the correlation function in d=2 goes to zero asymp-
totically, it does so slowly, with power-law behavior. This means that
phase coherence persists over a finite but large distance, and a local Bose
condensate exists. The local Bose condensate supports vortex-antivortex
 Bose-Einstein Condensation and Superfluidity 35
pairs, whose unbinding leads to the Kosterlitz-Thouless transition [21].
Thus, despite the fact that there is no Bose condensate, there is a phase
transition, associated with what Kosterlitz and Thouless termed "topo-
logical order."
In the Ginsberg-Landau picture, the Kosterlitz-Thouless transition in
d=2 should not be associated with a sign change of c^. It is caused
by vorticity associated with phase fluctuations, and to support vortices
we must must have ci < 0. It would be interesting to discover such
a transition within the Ginsberg-Landau theory; but this has not been
attempted. Instead, one takes the two-dimensional XY model [22] as
a more expedient starting point. As we shall discuss later, this model
exhibits superfluidity in the low-temperature phase, even though there is
no Bose condensate.
The local Bose condensate in d=2 is relevant to current thinking in
high-Tc superconductivity, where one speaks of Bose-Einstein condensa-
tion of "holons" in the two-dimensional copper oxide plane. What one
means is a "local condensation", which presumably becomes global when
a small amount of interplane coupling is turned on.

2 Superfluidity
The term superfluidity covers a group of experimental phenomena in liq-
uid 4He, including frictionless flow, persistent current, wave propagation
on liquid surfaces, and other kinetic effects. Our understanding on this
subject is considerably less than that for Bose-Einstein condensation.
These first attempts to define superfluidity were based on specific
pictures of liquid 4He. Landau [23] proposed a low-temperature model
in which the excitations are quasiparticles, which become phonons in
the long-wavelength limit. Feynman [24] justified the picture from a
microscopic point of view, and argued that Bose-Einstein statistics rule
out all excitations except density fluctuations (phonons) in the immediate
neighborhood of the ground state.
Landau argued that a system with only phonon excitations will flow
without friction, because it cannot absorb arbitrarily small amounts of
energy-momentum transfer. The argument is as follows. At absolute zero
the only way to interact with the system is to excite phonons. Suppose
an external object transfers energy AE and momentum AP by creating
np phonons of momentum p and energy cp, where c is the sound velocity.
36 K. Huang
Then
AE = Y^ cpnp
p

V (14)

Thus |AP| < Y,PPnP> o r AE > c|AP|. If the external object is moving
with velocity ye, we must have AE = ye • AP. Therefore ve > c.
Using the quasiparticle picture, Khalatnikov [25] derived an expression
for the normal fluid density at finite temperatures, based on Gallilean
invariance. Suppose the energy of a quasiparticle of momentum p is
(Op, and the average occupation number for quasiparticles is n(co) =
[exp(co//cT) — I]" 1 . In a two-fluid picture, with superfluid velocity vs and
normal fluid velocity vn, the quasiparticle energy in a reference frame
moving with the superfluid is
(o'p = (op-p'(yn-ys). (15)
The current density in the co-moving frame is given by

= pn(yn-ys) (16)

where the right side is a definition of the normal fluid mass density pn.
By expanding in powers of (vn — vs) and identifying coefficients on both
sides, one easily obtains

Pn
~ 3j (2nfP dop (17)

The superfluid mass density is ps = p — pn, where p is the total mass

density. In the limit T —> 0, the occupation number vanishes like
n(co) « kT/co, and thus pn -> 0. This shows that the ground state is pure
superfluid. This result is not valid for fermionic quasiparticles, because
n(co) does not vanish at absolute zero.
A more general definition of the superfluid density is proposed by
Hohenberg and Martin [26]. They emphasized that, unlike the condensate
density, the superfluid density is not an equilibrium quantity but a
transport coefficient, and should be defined in terms of linear-response
theory.
Suppose we impose on the system an infinitesimal velocity field <5v(x, t),
which should be turned on and off adiabatically:
(5v(x,t)—+0 (18)
 Bose-Einstein Condensation and Superfluidity 37

6v

5v

(a) (b)

Fig. 1. (a) Longitudinal response: Place liquid between parallel plates that are
moving at velocity S\. The response gives the total density, (b) Transverse
response: Place liquid in long pipe moving with velocity S\. The response gives
the normal fluid density.

The linear response, as measured by the changed in the momentum

density g of the system, can be expressed in terms of the appropriate
Fourier transforms in the form
3 (g,(k, co)) = xtj(K co)dvj(k9 co) (19)

which defines the susceptibility %i;(k,co). We are concerned only with the
static susceptibility &/(k) = #i 7(k,0). By rotational invariance, it can be
decomposed into a longitudinal and a transverse part:

Z y (k) = ,• - *&) B(k2) (20)

In the limit k —> 0, the longitudinal response is the total density (statement
of the f-sum-rule), while the transverse response is defined as the normal
fluid density. Thus, we have

p = lim A(k2)

pn = lim B(k2) (21)

k-+0
Ps = Um[A(k2)-B(k2)].

An elementary explanation for these definitions due to Baym [27] goes

as follows. Since A(k2) and B(k2) are rotationally invariant, it suffices
to examine any component of Xij, say Xn- That is, we consider the
momentum response in the x direction, (5gi(k), due to an imposed velocity
field in the x direction, Sv\(k).
38 K. Huang
Let the system be contained in a finite box, whose dimensions eventu-
ally tend to infinity. Suppose first that the system is sandwiched between
two infinite plates normal to the x axis, with the separation subsequently
tending to infinity, as illustrated in Fig.l(a). This arrangement corre-
sponds to the limiting procedure ki —> 0, k^ —• 0, followed by k\ —• 0. In
this case, a velocity field can be created by pushing on the plates, with the
consequence that the whole system is pushed along the x direction. The
response is associated with the total density. Carrying out this limiting
procedure explicitly yields p = lim^o^C^2)-
Next suppose the system is in the shape of an infinitely long pipe,
along the x axis, with the cross section of the pipe tending to infinity
subsequently, as illustrated in Fig.l(b). This arrangement corresponds
to the limiting procedure k\ —• 0, followed by ki -> 0, £3 —• 0. In this
case, a velocity field can be created by dragging the walls of the pipe,
and the part of the the system responding to the shear force is identified
as the normal fluid. Carrying out this limiting procedure explicitly gives
pn = lim^o B(k2).
These considerations apply equally well in d=2. In fact, using the
definitions (21), Nelson and Kosterlitz [28] calculated the superfluid
density in the two-dimensional XY model, and showed that it has a finite
jump Aps at the Kosterlitz-Thouless transition temperature Tc:
(h/m)2kTcAps = 2/TT (22)
where m is the mass of a particle in the fluid. This result demonstrates
that superfluidity can occur without Bose-Einstein condensation. The
experimental verification of this jump in 4He films [29] validates the
linear-response definition of the superfluid density.
It is well to re-emphasize that superfluidity is a kinetic property of a
system, and is generally more complicated than equilibrium properties.
The arguments of Landau and Khalatnikov were formulated in the
context of liquid helium in the quasiparticle picture, and one can find
fault with them in a more general setting [30]. The linear-response
definition, on the other hand, is more general; but it may not be unique.
Indeed, one can question whether there exists a unique definition of
superfluidity in general [31].

3 Hard-Sphere Bose Gas

A Bose gas in d=3 with hard-sphere interactions can be treated analyti-
cally in the limit of low temperatures and densities, the criteria for which
 Bose-Einstein Condensation and Superfluidity 39

Transition line

-• V

Fig. 2. Shaded region indicates schematically the low-density and low-

temperature region in which our calculations are valid. It includes the transition
line of the Bose-Einstein condensation for the ideal Bose gas.

are

a/k
na3 (23)

where a is the hard-sphere diameter, n the particle density, and

k= ^2nh2/mkT (24)

is the thermal wavelength. The region satisfying these criteria, shown

shaded in the P-V diagram in Fig. 2, includes the transition line of the
ideal Bose gas. Experimental efforts are underway to observe superfluidity
in such a dilute gas, in the form of spin-aligned atomic hydrogen [32]
or excitons [33].
In the low-temperature regime, the hard-sphere diameter a enters as
the s-wave scattering length of the interparticle potential, the detailed
shape of which is irrelevant. On this basis, Lenz [34] first showed that,
to lowest order, the system behaves like a medium with a refractive
index, and the ground state energy per particle is shifted by an amount
proportional to na. Fermi [35] showed that, for low-energy scattering, all
potentials with the same scattering length a can be effectively replaced
by a pseudopotential

F(r) = (25)
m
Using this, we can get Lenz's result in first-order perturbation theory.
40 K. Huang
With the pseudopotential, the many-body Hamiltonian reads

(r)xp(r)xp{r)

44 ' (26)
k,p,q J
where Q is the volume of the system. The ground state energy is divergent
in higher orders; but it can be made finite by a simple subtraction
procedure [36].
The Hamiltonian can be diagonalized using Bogoliubov's method [37],
in which ao is replaced by a c-number y/No. The parameter No, which
labels the state we are considering, is the number of fc=0 particles in the
unperturbed state. We write it in the form
No = £N, (27)
where N is the total number of particles. The diagonalized form of the
Hamiltonian to order a5/2 is given by [38, 39]
H =

(28)

where n = JV/Q is the particle density. The quasiparticle annihilation

operator bk is related to a^a^ through a Bogoliubov transformation:

0k = / (fok — ^k^k)

at = (bt — Xkh) (29)

where

. (30)

The sound velocity deduced from the Bogoliubov quasiparticle spectrum

is c = (h/m)y/4nan£9 which agrees with that calculated via the compress-
ibility by differentiating Eo.
 Bose-Einstein Condensation and Superfluidity 41
The condensate density is defined by no = n — Q~l Sk^
ground state, which corresponds to £ = 1, we have

This shows that the interactions deplete the Bose condensate; but the
superfluidity density at absolute zero is n, from Khalatnikov's formula
(17).
The partition sum should extend over all values of £. For thermal
properties near the transition point, the important states have £ « 0, and
in first approximation we keep £ only to lowest order. This means that
the quasiparticle spectrum is replaced by a shifted free-particle spectrum
(ft2/2m)(k2 + Snan£). The calculation of the free energy is elementary [39],
and gives

(( h ) (32)
m \ 2
where F^ is the free energy of the ideal Bose gas, and | , the thermody-
namic average of £, is the same as that in the ideal Bose gas:
j_jl-{T/Tcf'2 (T<TC)
1.0 (T>TC),

where kTc = 2nh2n/[m(2.6l2)3/2]. To this order, the transition tem-

perature is the same as that of the ideal Bose gas. In a higher-order
calculation, the transition temperature is lowered from that of the ideal
gas by O{yjna*) [39, 40].
The isotherms are given in this approximation by

P = P(°) + / (47lha/m)v (v > vc)

^\{2nh2a/m){v-2+v-2) (v < vc)9
where P is the pressure, P^ the pressure of the ideal Bose gas, v = 1/n,
and vc = 2nh2/[mkT(2.612)3/2]. They are sketched in Fig. 3. The Bose-
Einstein condensation now appears as a second-order phase transition. In
higher-order calculations, it appears to be a first-order phase transition
with a very narrow transition region, of order y/na?, which may not be
significant. According to (34), the isothermal compressibility just above vc
is Ko = mv2/Snh2a. When the gas is compressed just across the transition
point, it drops abruptly by a factor of 2. This is a signal for Bose-Einstein
condensation that one hopes to observe in spin-aligned hydrogen.
42 K. Huang

Hard-sphere Bose gas

Fig. 3. Qualitative sketch of isotherm of hard-sphere Bose gas to lowest order of

approximation, compared with that of ideal Bose gas at the same temperature.
The isothermal compressibility changes discontinuously by a factor of 2 at the
transition point.

A more realistic simulation of 4He consists of adding an attractive

tail to the hard-sphere potential [39, 41]. Such a model qualitatively
reproduces the phase diagram of 4He more faithfully. There is now a
first-order gas-liquid phase transition, and Bose-Einstein condensation
occurs within the liquid phase, dividing it into liquid I and liquid II, just
like the X transition in liquid 4He.

4 Superfluidity in Random Media

We have seen that Bose-Einstein condensation is not necessary for su-
perfluidity. We now show that it is not sufficient. The basic idea is very
simple. Imagine a Bose gas in a randomly chosen potential well, which
presumably has bound states. If there are no interactions, then at abso-
lute zero all particles will condense into the lowest bound state. As long
as the bound-state wave function has a non-vanishing spatial integral
(fc=0 projection), we have a Bose condensate; but it is pinned by the
potential, which may be regarded as a kind of external wall. Thus, the
condensate belongs to the normal fluid rather than the superfluid. This
model is of course unrealistic. We need repulsive interparticle interactions
to make the energy extensive, and the picture becomes more complicated.
But the underlying idea remains, namely, localization effects in random
media may destroy superfluidity. This idea has been developed further
in tight-binding models, in which bosons can hop from site to site in
 Bose-Einstein Condensation and Superfluidity 43

(a) (b)
Fig. 4. (a) Transition surface for superfluid transition of 4He in porous media.
Superfluidity disappears at T=0 below a critical coverage. After Physics Today,
July, 1989, p.22. (b) Transition surface for superfluid transition in a hard-sphere
Bose gas in random external potentials in d=3. The total density is n, and the
superfluid particle density is ns.

a lattice, simulating quantum tunneling between localized states. The

phase diagram for such a system has been discussed. In particular, a
non-superfluid phase is proposed, with very specific properties dubbed
the "Bose glass" [42, 43, 44].
The experimental impetus for these theoretical considerations come
from observations on superfluid 4He adsorbed in various porous media:
Vycor, xerogel and aerogel [45].
The superfluid density was measured as a function of temperature
and coverage, using a torsional pendulum. A qualitative sketch of the
superfluid transition surface, synthesized from different experiments, is
shown in Fig.4(a). We can see that, at absolute zero, the superfluid density
as a function of coverage vanishes below a certain critical coverage. A
system of bosons in random potentials is an attempt to simulate some
aspects of these experiments. So far, signals for the "Bose glass" have
not been detected experimentally [46].
I would like to describe a model somewhat different from the tight-
binding kind - a dilute hard-sphere Bose gas in random potentials at
low temperatures, in d=3 [47]. Since, unlike the much-studied case of
fermion localization, interparticle interactions are necessary for stability,
we cannot start with an ideal Bose gas. The next best thing is to start with
44 K. Huang
a soluble interacting model, the dilute hard-sphere Bose gas. This model
should be compared with experiments at low coverage near absolute
zero, i.e., in the neighborhood of the critical coverage in Fig. 4(a).
An arbitrary external potential U(r) is introduced by adding to the
Hamiltonian the term Jflrxp^Uxp,of which we only keep the part that ex-
cites one particle from the condensate, and vice versa. In the Bogoliubov
approximation, the new term in the Hamiltonian is

where U^ is the Fourier transform of U(r). The new Hamiltonian can be

diagonalized by supplementing the Bogoliubov transformation (29) with
a c-number, k-dependent term. In term of the field operator, this means
that we write xp(r) = v(r) + n(r), where v(r) is a c-number vacuum field,
and rj(r) an operator. We then calculate the free-energy and relevant
correlation functions, all to lowest order in the external potential, which
is arbitrary but fixed.
We make quenched averages of the free energy and correlation func-
tions over external potentials by putting

^(UkU-k>)av = ^Uo, (36)

where Ro is the single parameter that characterizes the random potentials.

It has dimension (energy)2(length)3. If we picture U(r) to consist of
scattering centers of various positions and strengths, Ro would represent
the average of (density)(strength)2 of the scatterers.
The validity of the approximations made are subject to the conditions
(23) plus the fact that Ro is treated only to first order. This means that
the only effect of the random potentials is to deplete or enhance the
condensate, while rescattering effects are neglected.
I will only quote the results at absolute zero. In the unperturbed ground
state, the total density n is equal to that of the condensate no = No/SI.
When perturbations are turned on, the condensate is depleted, due to
both the hard-sphere interactions and to the random potentials. The
total density is thus made up of three terms:

n = n 0 + m + n R, (37)
where n\ arises from the hard-sphere interactions, and can be read off
(31), while nR arises from the random potentials:
 Bose-Einstein Condensation and Superfluidity 45

ni = .
\y%t\

(38)

where c\ = 8/(3TT1/2), C2 = m2/(87c3/2) (in units with h = 1.) Substituting

(38) into (37), we obtain a transcendental equation for no. The singularity
HR ~ a~1//2 when a —• 0 signals the collapse of the system when the
interparticle interactions are removed.
To physically understand nR, note that it can be represented in the
form
x
? ^TF-> (39)
where the factor E = m~xn2^ supplies an energy scale natural for a Bose
gas, T is the collision time for the scattering of a particle out of the
condensate:
\ R(E) (40)

where p(E) is the density of final states, taken to be one-phonon states

of momentum k and energy cfe, with c oc ^Jna/m:

-I
p(E) = / d\ d(E - ck) oc E2/c\ (41)

For the approximations to be valid, both a and RQ must be small. Thus

n\ must be small; but nR oc Ro/yja can have arbitrary values. Solving for
no from (37) and (38), we can easily see that no never vanishes, however
large RQ might be. Thus, at absolute zero, there will always be a Bose
condensate.
The superfluid density is calculated by calculating the particle density
nn of the normal fluid using (21). At absolute zero we find nn = (4/3)n#,
and thus the superfluid particle density is given by
4
ns = n--nR. (42)

The remarkable factor 4/3 means that the random potentials generate
an amount of normal fluid 4/3 times what it took from the condensate.
This implies that part of the condensate belonging to the normal fluid,
being pinned by the random scatterers - a example of boson localization.
We can now see that superfluidity can be destroyed, even though there
46 K. Huang
is a Bose condensate. From (37) and (42), we can deduce the relation

n, = ^[4(no+ * ! ) - * ] , (43)

where n\ may be neglected. Thus, ns vanishes when the condensate

is depleted by random scattering to the value no = n/4. The finite-
temperature calculations yield a superfluid transition surface shown in
Fig. 4(b), which is to be compared with the experimental one in Fig.
4(a). They have similar qualitative features, but there is a re-entrant
effect in the theoretical graph which has not been seen experimentally.
The detailed nature of the superfluid transition, as well as properties of
the "Bose glass" phase, is beyond the capabilities of this calculation.
In quenched averaging, there is no interference between members of
the ensemble over which we average. Our results thus indicate that
localization happens in a good fraction of the potentials in the ensemble.
The calculation with any single given potential, of course, cannot be
carried out analytically.
The d=2 case cannot be treated in the same manner, because there
is no Bose condensate at any finite temperature, due to strong phase
fluctuations. Thus we cannot replace ao by a c-number offixedphase. The
way to proceed is to make a canonical transformation \p = /pexp(i0).
We can then replace ajao by a c-number no in the expansion p = ajao+- • \
A Bogoliubov-like transformation then diagonalizes the Hamiltonian. In
this manner, it is found that, just as in the d=3 case, superfluidity can be
destroyed by random potentials [48, 49].
Finally, let us turn to superfluid flow in a fully saturated porous
medium. This corresponds to a cut at maximum coverage in the phase
diagram of Fig. 4(a). Experimental interest centers on the critical expo-
nent C at the superfluid transition, which is reported to be, respectively,
0.67, 0.89 and 0.80, for Vycor, xerogel and aerogel [45], as compared
with C = 0.672 for the bulk liquid. It is thought, therefore, that each
medium represents a new universality class. I would like to suggest that
actually there is only one universality class with £ = 1.7, as appropriate
for percolation in d=3. The difference in the apparent values of £ can
be attributed to the fact that the experiments have not reached the true
critical point [50].
The physics here is different from that in the low-density regime, and
calls for a different model. We simulate the flow channels in the sponge-
like medium as bonds in a cubic lattice, with channel radii chosen at
random from a given distribution. The momentum density for superfluid
 Bose-Einstein Condensation and Superfluidity 47

I ' " ' '

aerogel: (2.17)
+ 85% D 95%
X 91% + 98%
O 94%

-5 H Vycor(2.00)
* Xerogel ( 2.09 )

, I I . . .
-5 -4 -3 -2 -1
Log(T c -T)
Fig. 5. Log-log plot of superfluid density vs. temperature for liquid 4He in fully
saturated media. The number in braces after the name of each medium gives
the value of Tc used in the plot. Percentages correspond to porosity of aerogel
samples [51].

flow is given by
j = PsVs + Pn (44)

where the two terms refer to the superfluid and normal fluid respectively.
Assuming that all channels are so small that the normal fluid is pinned,
we put \n = 0. Thus, writing vs = V®, where O is proportional to the
superfluid phase, we have
j = PsV<D (45)
This looks like Ohm's law with current density j , electrostatic potential
O and electrical conductivity ps. Thus, each bond can be replaced by
a resistor, and the superfluid density of the medium corresponds to the
conductivity of the network.
In a long channel filled with liquid 4He, the superfluid transition
temperature decreases with the radius according to a power law [52]. At
sufficiently high temperatures, all bonds in the lattice are non-conducting,
and the overall conductance is zero. As the temperature is lowered, some
bonds begins to conduct; but the overall conductance remains zero,
until there is a percolating cluster. Therefore, the true critical point
corresponds to percolation in d=3.
This picture is verified by solving Kirchoff's equations numerically to
obtain the conductivity of the network, for various distributions of the
48 K. Huang
resistances. The critical exponent is independent of the distribution used.
However, the true critical region is very narrow. At lower temperatures,
there is a wide "mean-field" region in which the numerically calculated
superfluid density can be fit with a power law, whose exponent depends
on the distribution. By choosing the distribution appropriately, we can
reproduce the apparent exponent observed for the different media. In
Fig. 5 we plot all available data on a log-log plot to show that they do
not contradict our picture.
This work is supported in part by funds provided by the US Depart-
ment of Energy under contract # DE-AC02-76ER03069.

References
[I] S.N. Bose , Z. Phys. 26, 178 (1924).
[2] A. Einstein, Sitzber. Kgl. Preuss. Akad. Wiss., (1924), p.261 ; (1925),
p.3 . The story of Bose-Einstein statistics is told in A. Pais, Subtle
is the Lord, The Science and the Life of Albert Einstein (Clarendon
Press, Oxford, 1982), Ch. 23.
[3] L. Tisza, Nature, 141, 913 (1938),
[4] O. Penrose and L. Onsager, Phys. Rev. 104, 576 (1956).
[5] RE. Sokol, this volume.
[6] J. Goldstone, N. Cim. 19, 154 (1961).
[7] C.N. Yang, Rev. Mod. Phys. 34, 4 (1962).
[8] RW. Anderson, Rev. Mod. Phys. 38, 298 (1966).
[9] RW. Anderson, Basic Notions of Condensed Matter Physics
(Benjamin-Cummings, Menlo Park,1984), pp.l5ff.
[10] L. Onsager, Suppl. N. Cim. 6, 249 (1949).
[II] R.P. Feynman, in Progress in Low Temperature Physics, vol.1, C.J.
Gorter, ed. (North-Holland, Amsterdam, 1955), p. 17.
[12] V.L. Ginsberg and L.R Pitaevsky, Sov. Phys. JETP 34, 858 (1958).
[13] K.G. Wilson and J. Kogut, Phys. Rep. 12C, 75 (1974).
[14] E.P. Gross, N. Cim. 20, 454 (1961); J. Math. Phys. 4, 195 (1963).
[15] L.R Pitaevsky, Sov. Phys. JETP 13, 451 (1961).
[16] T. Frisch, Y. Pomeau, and S. Rica, Phys. Rev. Lett. 69, 1644 (1992).
[17] See the contributions of Yu. Kagan and H.T.C. Stoof in this volume.
[18] RC. Hohenberg, Phys. Rev. 158, 383 (1967).
[19] N.D. Mermin and H. Wagner, Phys. Rev. Lett. 22, 1133 (1966).
[20] S. Coleman, Comm. Math. Phys. 31, 259 (1973).
[21] J.M. Kosterlitz and D.J. Thouless, J. Phys. C 6, 1181 (1973).
 Bose-Einstein Condensation and Superfluidity 49
[22] J.V. Jose, L.R Kadanoff, S. Kirkpatrick, and D.R. Nelson, Phys.
Rev. B 16, 1217 (1977).
[23] L.D. Landau, J. Phys. USSR 5, 71 (1941).
[24] R.P. Feynman, Phys. Rev., 94, 262 (1954).
[25] I.M. Khalatnikov, Introduction to the Theory of Superfluidity (Ben-
jamin, New York, 1965), p.13.
[26] PC. Hohenberg and PC. Martin, Ann. Physics (NY) 34, 291 (1965).
[27] G. Baym, in Mathematical Methods of Solid State and Super fluid The-
ory, ed. R.C. Clark (Oliver and Boyd, Edinburgh, 1969), p.121; the
argument is reproduced in D. Forster, Hydrodynamics Fluctuations,
Broken Symmetry, and Correlation Functions (Benjamin, Reading,
MA, 1975), pp.219 ff.
[28] D.R. Nelson and J.M. Kosterlitz, Phys. Rev. Lett. 39, 1201 (1977);
D.R. Nelson in Phase Transitions and Critical Phenomena, vol.7, C.
Domb and J.L. Lebowitz, eds. (Academic, New York, 1983).
[29] I. Rudnick, Phys. Rev. Lett. 40, 1454 (1978); D J . Bishop and J.
Reppy, Phys. Rev. Lett. 40, 1727 (1978); E. Webster, G. Webster,
and M. Chester, Phys. Rev. Lett. 42, 243 (1978); J.A. Roth, G.J.
Jelatis, and J.D. Maynard, Phys. Rev. Lett. 44, 333 (1978).
[30] See the article by L. V. Keldysh, this volume.
[31] See the article by A. Leggett, this volume.
[32] See the articles by T. Greytak and I. Silvera, this volume.
[33] See the article by J. P. Wolfe, et al, this volume.
[34] W. Lenz, Z. Phys. 56, 778 (1929).
[35] E. Fermi, Riverca Sci. 7, 13 (1936).
[36] K. Huang and C.N. Yang, Phys. Rev. 105, 767 (1956).
[37] N.N. Bogoliubov, J. Phys. USSR, 2, 23 (1947).
[38] T.D. Lee, K. Huang, and C.N. Yang, Phys. Rev. 106, 1135 (1957).
[39] K. Huang, in Studies in Statistical Mechanics, vol.2, J. De Boer and
G.E. Uhlenbeck, eds. (North-Holland, Amsterdam, 1964).
[40] K. Huang, C.N. Yang, and J.M. Luttinger, Phys. Rev. 105, 776
(1957).
[41] K. Huang, Phys. Rev. 115, 765 (1959); 119, 1129 (1960).
[42] J.A. Hertz, L. Fleishman, and PW. Anderson, Phys. Rev. Lett. 43,
942 (1979).
[43] D.S. Fisher and M.P.A. Fisher, Phys. Rev. 61, 1847 (1988).
[44] M.P.A. Fisher, P.B. Weichman, G. Grinstein, and D.S. Fisher, Phys.
Rev. B 40, 546 (1989).
[45] For a comprehensive review, see J.D. Reppy, J. Low Temp. Phys.
87, 205 (1992).
50 K. Huang
[46] J.D. Reppy, private communication.
[47] K. Huang and H.F. Meng, Phys. Rev. Lett. 69, 644 (1992).
[48] H.F. Meng, Superfluidity and Random Media, Ph.D. Thesis, Math-
ematics Department, MIT, 1993.
[49] H.F. Meng, Quantum Theory of Two-Dimensional Interacting Bo-
son Systems, (MIT Mathematics Department preprint, 1993).
[50] K. Huang and H.F. Meng, Phys. Rev. B 48, 6687 (1993).
[51] The data in Fig. 5 is from J. D. Reppy and H. M. W. Chan, private
communication.
[52] R. Donnelly, R. Hills, and P. Roberts, Phys. Rev. Lett. 42, 75 (1979).
 Bose-Einstein Condensation in Liquid
Helium
P. E. Sokol
Physics Department
Pennsylvania State University
University Park, PA 16802
USA

Abstract
Liquid helium is the prototypical example of a superfluid - a liquid that
flows without viscosity and transfers heat without a temperature gradient.
These properties are intimately related to the Bose condensation that oc-
curs in this strongly interacting liquid. Bose condensation is most directly
observed in the single particle atomic momentum distribution, where the
Bose condensate appears as a delta function singularity. In this article, we
discuss the experimental techniques used to observe the condensate and the
current status of measurments of the Bose condensate in liquid helium.

1 Introduction

Liquid helium (4He) has fascinated physicists ever since Kammerlingh-

Onnes liquified the last of the so called permanent gases in 1908. How-
ever, evidence of what is without doubt the most fascinating property of
this unique liquid, superfluidity, was not reported until almost 25 years
later [1]. The superfluid phase, where heat is transferred without a ther-
mal gradient and massflowswithout a driving pressure, is a macroscopic
manifestation of microscopic quantum effects governing the behavior of
atoms [2]. Bose-Einstein condensation was first proposed by London [3]
as the microscopic explanation for these fascinating phenomena.
Helium is unique among condensed atomic systems since the bulk
properties of the liquid are dominated by quantum effects. The most im-
portant of these are the statistical effects that arise for identical particles.
For bosons, such as 4He atoms, there is no limitation on the number of
particles that can occupy a single quantum state. Thus, in the absence of

51
52 P. Sokol
interactions, all the atoms in the liquid would occupy a single momentum
state at zero temperature.
Liquid helium is also unique among the Bose systems considered in this
volume. Other systems, such as spin polarized hydrogen and excitons,
are expected to exhibit Bose condensation. (See the review articles by
Greytak and Silvera, Castin et al, Mysyrowicz and Wolfe et al. in this
volume.) At present, however, liquid helium is the only system where the
existence of an experimentally attainable Bose condensed phase is almost
universally accepted.
Unlike many of the other systems considered in this volume, liquid
helium is a strongly interacting system. The interaction between the
atoms, which is dominated by the hard core repulsion, is strong enough
that the properties of the liquid cannot be treated as a simple perturbation
of the Ideal Bose Gas, as is the case for more weakly interacting systems.
As we will see, these strong interactions have greatly complicated both
the theoretical and experimental studies of the Bose-condensed phase of
superfluid 4He.
We begin this review with a discussion of the Ideal Bose Gas, a model
system consisting of non-interacting particles. This model, while not
applicable to the strongly interacting liquid phase of helium, is useful in
that it provides a system where the effects of Bose condensation are clearly
delineated. The microscopic properties of liquid helium, which we discuss
next, are quite different from the ideal gas due to the strong interactions
between atoms. Despite these differences, similarities between the ideal
gas and the liquid systems remain. In particular, a Bose-condensed phase
still appears in liquid 4 He at low temperatures, although the fraction of
particles in the condensate is considerably reduced. Therefore, in this
review, the analogies between an ideal Bose gas and superfluid 4He are
stressed, even though the former does not exhibit the long-range phase
coherence associated with Bose-broken symmetry. (For a discussion of
this, see especially the articles in this book by Nozieres, Huang, Stringari
and Stoof.)
The single-particle momentum distribution is one of the few quantities
which directly reflects the appearance of a Bose condensate. Therefore,
we next turn our attention to the experimental technique that provides
information on the momentum distribution: Deep Inelastic Neutron
Scattering (DINS). This technique, due to the information on n(p) that it
provides, provides one of the few opportunities to determine no directly [4,
5,6].
Early attempts to measure the momentum distribution, and to obtain
 BEC in Liquid Helium 53
direct information on the condensate using DINS, were limited by the
fluxes and energies of neutrons available. Recent advances in neutron
sources, which now have both higher fluxes and higher energies, have led
to a new generation of experiments. These new experiments are reviewed,
and the information that can be extracted, both directly and with the
help of models, are described.
Finally, we turn our attention to the direct evidence for the condensate
in liquid helium. The experimentally observed scattering shows distinct
changes consistent with the appearance of a condensate. However, the
sought after direct evidence, an observation of the (5-function singularity
associated with the condensate, is not observed. Nevertheless, comparison
to current theoretical predictions and to empirical expressions based on
realistic models give a strong case for a finite value of no and reasonable
results for its temperature and density dependence.

2 The Ideal Bose Gas

Before examining the properties of liquid helium, it is instructive to
first examine the Ideal Bose Gas (IBG), whose properties are entirely
controlled by the Bose-Einstein statistics obeyed by the particles. All
the complicating effects of interactions, which play a major role in
determining the properties of liquid helium, are absent. The effects of
Bose-Einstein condensation can, therefore, be studied in fairly direct
fashion.
The IBG, in its purest form, consists of a collection of non-interacting
particles in a box [7]. The eigenstates of a single particle in a box are
plane wave states with a well-defined momentum p. In the case of the
IBG, these s ates are also the eigenstates of the many particle system. The
statistical description of the particles in this box consists of specifying
the number of particles occupying each of the single-particle eigenstates.
This function is the single-particle momentum distribution, n(p), which
specifies the number of particles in a state with momentum p.
The particles in the gas can be characterized by their thermal DeBroglie
wavelength, which is given by

where m is the mass of the particle and T is the temperature. When

AD is much less than the interparticle spacing, quantum effects will be
negligible and the particles will behave classically. Conversely, when
54 P. Sokol

(a)
(b)
— Ideal Bose gas
- - - - Ideal classical gas
1.5
\
\ T = 50 K
(d)u

\ T= 3.14K

0.5

0 0.2 0.4 0.6 0.8 1 1.

p(A"')

(d ) 1 000

Ideal Bose gas

1600 Ideal Bose gas
Ideal classical gas

T=0K
(d)u

T=2.5K
T=3.14K

400

0
V
Fig. 1. Temperature dependence of the momentum distribution in the Ideal
Bose Gas (solid line) and the classical result (dashed line). The Bose-Einstein
condensation temperature is 3.14 K, corresponding to the density of liquid helium
at SVP. (a) T = 50 K, (b) T = 6 K, (c) T = 2.5 K (a condensate is present but
is not visible in this diagram due to the singular behavior of the uncondensed
component) and (d) T = 0.

XD is large compared to the interparticle spacing, there will be a large

overlap of the wavefunctions for the particles, and quantum effects will
be important.
The properties of the ideal Bose gas at high temperatures, where XD is
smaller than the atomic spacing, are dominated by the thermal motion of
the particles. The momenta of the particles will be distributed according
to the classical Maxwell-Boltzman distribution. This is simply a Gaussian
with a width that is proportional to the temperature, as illustrated in
Fig. l(a).
Upon lowering of the temperature, XD of the particles begins to ap-
proach the interparticle spacing. The effects of quantum statistics become
evident as n(p) deviates from its classical Gaussian shape. In the case
of bosons, where there is no restriction on the occupancy of a state,
the quantum behavior is reflected by an increase in the occupation of
states at small p. This reflects the particles taking advantage of the lack
 BEC in Liquid Helium 55
of restrictions on the occupancy of a state to reduce the energy of the
system. Thus, n(p) assumes a non-Gaussian shape and becomes more
peaked at small p, as shown in Fig. l(b). It is interesting to note that
this behavior becomes evident at relatively large temperatures for the
IBG. Even at twice the critical temperature, deviations from the classical
Gaussian behavior are quite pronounced.
As the temperature is lowered further , the quantum statistical effects
become the dominant factor in determining the properties of the IBG,
and we eventually have a transition to the Bose-Einstein condensed
phase. The Bose-Einstein transition temperature is

TBE = (2nh2/l.S97mkB)p2/\ (2)

where p is the density of the gas. This transition temperature is simply

determined by the condition that XD is equal to some appropriate measure
of the mean interparticle spacing.
At temperatures higher than TBE, the occupancy of any particular
one-body momentum state remains finite. Below this temperature, a
macroscopic (i.e. of order the total number of particles in the system, N)
occupation of the p = 0 momentum state develops. This is the famous
"Bose-Einstein condensation". It is reflected in n(p) by the appearance
of a term proportional to a Dirac delta function S(p), with a coefficient
which determines the (finite) fraction of particles residing in this Bose
condensate.
At finite temperatures lower than TBE, n(p) consists of a 6 function
component with a weight determined by the condensate and a component
resulting from atoms not in the condensate, as shown in Fig. l(c). As the
temperature is lowered, the magnitude of the condensate term increases
and the non-condensed component decreases. At T = 0, the condensate
component contains all of the intensity, representing the fact that every
particle in the system is in the ground state, as shown in Fig. l(d).
A common misconception about the IBG is that the non-condensed
part of n(p) is similar to a classical distribution, i.e. a broad Gaussian.
Associated with this is the idea that the uncondensed n(p) is similar to the
distribution above Tx. Neither of these ideas is correct. The momentum
distribution below Tx becomes singular, with the leading term going as
1/p2, as can be seen in Fig. l(c). Above Tx, n(p) becomes sharply peaked,
but not singular. Neither of these distributions is anything like the simple
Gaussian form often depicted.
56 P. Sokol
3 The Interacting Liquid

We now turn our attention to liquid 4He. Unlike the IBG, helium atoms
do interact with each other. Due to these interactions, the properties of
liquid helium are significantly different from those of the IBG. However,
one can still have a macroscopic number of atoms with zero momentum
(the Bose condensate). The interactions do, however, significantly reduce
the occupation of the ground state below its value in an ideal gas.
Liquid helium is, in one sense, a weakly interacting system. The
attractive part of the helium potential is quite weak [8], which, when
coupled with the light mass of 4 He, leads to a large zero-point motion
that prevents the atoms from being strongly localized even in the con-
densed phases. Thus, the liquid phase is not even stable until very low
temperatures and a solid phase, in which the atoms must be localized
about lattice sites, does not exist unless substantial external pressure is
applied [9]. This weak localization leads to a large overlap of the atomic
wavefunctions which make quantum statistics an important factor in
determining the properties of the condensed phases.
At the same time, liquid 4 He is also a very strongly interacting sys-
tem [10]. The hard core repulsion between atoms is very large due to the
closed shell electronic structure. The large zero-point motion would lead
to a large overlap of the core wavefunctions. The hard core repulsion
prevents this and leads to large repulsive interaction energies between
the atoms.
The effects of the interactions in liquid helium can be highlighted by
comparing the phase diagram of the IBG and liquid helium, as shown
in Fig. 2. The IBG exhibits a single-phase transition from the high
temperature gas phase to the Bose-condensed phase. This transition,
which is marked by the appearance of a finite fraction of particles in
the ground state, increases with increasing density. Liquid helium, on
the other hand, has a considerably more complex phase diagram [11].
There is no low density phase at temperatures where Bose condensation
might occur. At low temperatures, the low density gas is not stable and
condenses to form a high density liquid. Thus, there is no region where
Bose condensation of 4He atoms is possible where the interactions might
be considered weak.
Another difference between the liquid and the ideal gas is the depen-
dence of the transition temperature on density. In the IBG the transition
temperature increases with density, reflecting the fact that overlap of the
wavefunctions occurs at higher temperatures as the density increases. In
 BEC in Liquid Helium 57

0.280

0.240 -
Solid

0.200 - -

0.160 - Superfluid \ / Liquid

u
0.120

A
0.080 -

0.040 -
1 naccessible

Tc of ideal Bose gas

J Gas

0.000
0.0 1.0 2.0 3.0 4.0 5.0 6.0 7.0
T(K)

Fig. 2. The phase diagram of liquid helium in the temperature-density plane. The
regions marked inaccessible are not stable in the bulk liquid. The Bose-Einstein
transition temperature as a function of density is also shown.

liquid helium, on the other hand, the transition temperature decreases

with increasing density (as shown by pressure studies), indicating that the
strong interactions present in the liquid destroy the condensate. Thus,
the IBG is not an appropriate model for liquid helium.
The eigenstates of the particles in the non-interacting gas are free
particle states. Thus, p is a "good" quantum number and no is just the
density of particles in the ground state. The nature of Bose condensation
in the liquid is more complicated since we are no longer dealing with free
particle states and p is no longer a "good" quantum number. However,
one still finds that there can be a finite fraction of atoms with zero
momentum, and define this as the "condensate fraction".
A more general definition of Bose condensation that is applicable to
interacting systems is required. The concept of off-diagonal, long-range
order [12, 13, 14] (ODLRO) provides the necessary new definition. (See
also the article by Huang in this volume.) This new phase is described
in terms of the one-body density matrix which is defined as the average
of the product of the field creation operator at point r and the field
58 P. Sokol
destruction operator at the origin,
Pi(r) = < tp*(r,r 2 ,...,r N )^(r,r 2 ,...,r N ) >, (3)
where < > denotes an ensemble average and y;(r,r 2,...,rjv) is the many
body wave function for the system. This T = 0 definition can be easily
generalized to finite T.
In the classical limit, pi(r) is approximately Gaussian with a width
determined by XD. The general criteria for Bose condensation is that
pi(r) develops an eigenvalue of finite size in a macroscopic system. This
corresponds to pi(r) having a finite value as r —• oo. The magnitude of
this finite value is just the condensate density no-
The single-particle density matrix is related to the momentum distri-
bution through a Fourier transform:

n{p) = j dre-*TPl(r). (4)

This is the projection of the single-particle density matrix, in the position
representation, onto a set of single-particle states in momentum space.
For a Bose system, the momentum distribution can be separated into two
components, a sharp delta-function term representing the condensate and
a smooth component corresponding to the atoms with finite momentum,
(5)
where the condensate density is determined by
limpi(r) = wo (6)
r—>oo

and the non-condensate portion has the Fourier representation

n\p) = J (7)

Typical behavior for both n(p) and p\(r) for an interacting Bose fluid
shown in Fig. 3. The dynamical short-range correlation due to the hard
core repulsion governs the small-r behavior of p\{r) and the large-/? be-
havior of n{p). The effects of statistics play little role in this small-r range
and pi(r) would have a similar shape for Fermion or Boltzmann (class-
ical) particles. The effects of statistical correlations are most apparent in
Pi(r) at large r and in n(p) at small p. The condensate, which gives rise to
a finite value of pi(r) at infinity, manifests itself as a ^-function at p = 0
in nip).
A simple physical interpretation of pi(r) exists in the case of bosons
[15]. As can be seen from the definition of pi(r), it is the product of a
 BEC in Liquid Helium 59

^ 0.5 -

1.0 2.0 3.0 4.0

Fig. 3. Typical behavior of the momentum distribution (a) and single-particle

density matrix (b) for a Bose-condensed system.

particle annihilation operator at r = 0 and a particle creation operator

at r. Thus, p\(r) can be viewed as the overlap in the wavefunction of the
system when a particle is removed at the origin and replaced a distance
r away. In a classical system, this overlap would die away quickly,
with a distance scale on the order of XQ. However, when a condensate
exists, this overlap persists out to infinite separations. Thus, there is
a finite possibility of removing a particle at the origin and placing it
back in the same quantum state, no matter what the separation. This
interpretation emphasizes the fact that the Bose-condensed phase has
long-range macroscopic coherence.
The dual nature of liquid helium, being both strongly and weakly
interacting, has proved an extreme challenge for theoretical studies. Con-
sequently, it has been difficult to develop a comprehensive microscopic
theory of the dense liquid phase using perturbative techniques [16]. Some
specific results can be obtained using field-theoretical Green's function
techniques [17]. For example, this approach shows how the superfluid
properties are related to the broken Bose symmetry associated with a
condensate. One is able to predict that n(p) should become singular at
small p when a condensate is present, due to coupling of fluctuations
in the condensate with phonons [18, 19]. It is still difficult, however, to
60 P. Sokol
carry out realistic, quantitative calculations within the field-theoretical
approach.
In view of the difficulty associated with field-theoretical approaches,
the most detailed results for the Bose condensate have come from nu-
merical calculations. Variational [20], Green's Function Monte Carlo
(GFMC) [21], and most recently, shadow wave function [22] techniques
have been used to calculate the properties of the ground state (i.e., at
T = 0). Perturbative techniques, such as Correlated Basis Functions,
have also been extensively applied to 4He [23].
Recently, progress has also been made in extending these results to
finite temperatures by working with excited states corresponding to ele-
mentary excitations generated by applying appropriate operators to the
ground state wavefunction. This technique has been quite successful at
lower temperatures [20] where the elementary excitations are simple den-
sity fluctuations (phonons). Techniques for higher temperatures closer to
the transition, where rotons are the dominant elementary excitations, are
still being developed [24].
Another technique which is extremely powerful is the Path Integral
Monte Carlo method [25], in which the single-particle density matrix
at some low temperature is formulated in terms of path integrals over
density matrices at higher temperatures. Since the single-particle den-
sity matrix at high temperature can be calculated quite accurately, this
method allows calculation of the momentum distribution at any temper-
ature, limited only by the computational power available.
These numerical techniques can be considered as, in a sense, first-
principles calculations. They use as input only the mass of the 4He atom,
the measured inter-atomic interaction potential, and the density. There
are no adjustable parameters that can be varied to obtain agreement with
experiment. In view of this, the excellent agreement of these calculations
with several measured properties for the ground state of the liquid [26],
such as the total energy and static structure of the liquid, is quite
impressive.
The most prominent feature in the ground state n(p) obtained from
these numerical techniques is the appearance of a condensate, a d-
function singularity at p = 0. The intensity of this <5-function, which is
the condensate density no, is about 9% of the total intensity. All recent'
calculations of the ground state of helium have obtained essentially the
same result. It is interesting to note that this is quite similar to the original
estimates by Penrose and Onsager for a simple hard-sphere liquid [12].
 BEC in Liquid Helium 61

0.150

0.125 —

0.100 —

0.075 —

0.050 —

0.025 —

0.000

(A"1)
Fig. 4. Theoretical calculations of the momentum distribution. The ground
state momentum distributions have been calculated using GFMC (solid) and
variational (dotted) techniques. The momentum distribution in the normal liquid
(dashed) has been calculated using PIMC techniques.

Fig. 4 shows two recent calculations of n(p) in the ground state, based
on GFMC [21] and variational [20] techniques.
The remainder of the momentum distribution, excluding the conden-
sate d -function, is known as the uncondensed momentum distribution
and exhibits several interesting features. At small p, which corresponds
to long-range interactions among the atoms, the effects of statistics are
the dominant factor. For example, singular behavior [18, 19] due to
the coupling of the condensate to long-wavelength collective excitations
(phonons) appears. This singular behavior can be obtained exactly at
small p, where the phonons are well defined, and goes as 1/p2 in the
ballistic regime and 1/p in the hydrodynamic regime. The variational
n(p) in Fig. 4 shows this behavior explicitly at small p. This singular
behavior is not present in the GFMC results, presumably due to the
relatively small size of the samples used in the numerical calculations.
Alternately, the large p behavior of n(p) is determined primarily by the
62 P. Sokol
short-range repulsive interaction between atoms, and the Bose statistics
play little role.
The great majority of numerical studies of the liquid have concentrated
on the ground state due to the intrinsic limitations of the GFMC and
variational approaches. However, Path Integral Monte Carlo (PIMC) [25]
methods have recently been applied to study the liquid properties at
finite temperatures. These calculations yield results similar to the ground
state calculations at low temperatures, i.e. a condensate fraction of
approximately 9%. However, they have the distinct advantage that they
can provide results at finite temperature. The condensate fraction is found
to exhibit a rapid increase of no entering the superfluid as temperature is
lowered, with very little variation with temperature thereafter.
The momentum distribution may also be calculated in the normal
liquid using PIMC. For example, Fig. 4 shows n(p) at 3.33 K, well above
the superfluid transition [25]. The momentum distribution is broad and
featureless with a nearly Gaussian form, the familiar classical result. The
width of the momentum distribution is determined by the quantum zero-
point motion of the liquid and is much larger than the width expected for
classical particles. However, aside from this, the shape of the momentum
distribution in the normal liquid shows little effect due to quantum
statistics, in contrast to the IBG discussed in Section 2.

4 Deep Inelastic Neutron Scattering

Inelastic neutron scattering at large momentum transfer Q provides the
most direct means of obtaining information on n(p) and hence the con-
densate [10]. The scattering of neutrons by a condensed sample, such
as liquid helium, is described by the double differential scattering cross
section, which can be written as [10]

KB-££« ft " ) + £
where aT, G\ and oc are the total, incoherent and coherent scattering cross
sections and kt and kf are the initial and final neutron wave vectors. The
interaction of the neutron with the sample is described by the coherent
and incoherent dynamic structure factors Sc(Q,co) and Si(Q,co), where co
and Q are the energy and momentum transfer of the scattered neutron.
The dynamic structure factors contain information on spatial and
temporal correlations between atoms in the sample [27]. The coherent
dynamical structure factor involves correlations between different atoms
 BEC in Liquid Helium 63
and provides information on the structure and time-dependent behavior
related to phonons and rotons. The incoherent cross section, in contrast,
contains information only on the correlations of a single atom at two
different times.
The type of information provided by neutron scattering measurements
depends on the wavelength of the probing particle. Neutrons with wave-
lengths 2n/Q comparable to the interparticle separation provide infor-
mation on both the collective and single-particle correlations. However,
when the wavelength is much less than the average interparticle spacing,
neutron scattering measurements only probe single-particle correlations.
In this limit the wavelength of the neutron is too short to probe collec-
tive effects. The scattering is then determined totally by the incoherent
structure factor and provides information only on the single-particle mo-
tion of atoms. Even in this single-particle limit, however, the scattering
still probes the full many body properties of the sample through the
interactions of the scattering atom with the other atoms in the sample.
The scattering simplifies even further if the energy transferred to an
atom by a neutron is large compared with the interaction energies with
neighboring atoms. This limit [4], known as the Impulse Approximation
(IA), assumes that the neutron scatters from a single atom in a time
so short that the neighboring particles cannot react to the perturbation
caused by the neutron. The response of the He atom is then determined
entirely by n(p). It recoils from the collision in a free particle state of
high momentum and energy and the scattering directly probes the initial
many body wavefunction of the system.
The scattering in this high momentum and energy transfer limit is
described by the Compton profile
1 f+cc
JIA(Y) = -— 2 / pn(p)dp, (9)
4pnz J\Y\
where p is the density. The scattering in the IA does not depend on the
energy and momentum transfer separately, but only through the scaling
variable
Y=(M/Q)(co-corl (10)

where a> and Q are the energy and momentum transfer of the scattered
neutron, M is the mass and cor = Q2/2M is the recoil energy of the
scattering atom.
The scattering exhibits some characteristic features [28]; it is symmet-
ric, centered at Y = 0 and depends on a> and Q only through the scaling
64 P. Sokol
variable Y. These conditions have often been taken as indicative of
scattering in the IA limit. However, it is important to note that they are
necessary, but not sufficient, conditions for the validity of the IA. It is
possible for the scattering to exhibit these features even when the IA is
not valid, such as in the case of a gas of hard spheres [29].
The scattering in the IA limit is directly related to n(p) and, in the
case of a liquid, measurements of J(Y) can be used to determine n(p)
directly. For example, a Gaussian n(p) implies a Gaussian J(Y) with the
same second moment. Furthermore, the condensate, which appears as a
three-dimensional (5-function in n(p), is a one-dimensional (5-function in
J(Y). Unfortunately, the extraction of information is hampered by the
fact that prominent — and physically interesting — features in n(p) may
not be strongly reflected in J(Y) [6].
To see this, consider the two recent calculations of the ground state
n(p) discussed earlier and shown in Fig. 5(a). Both calculations predict
a condensate fraction, which appears as a delta function, with no=9.2
%. Both calculations also predict quite similar behavior at intermediate
and large p. However, they differ markedly at small p. The variational
calculation exhibits singular behavior due to coupling of long wave-
length density fluctuations (phonons) to the condensate which are not
present in the GFMC result, presumably due to finite size effects in the
calculation.
While n(p) for these two calculations is quite different, the correspond-
ing JIA(Y), shown in Fig. 5(b), is remarkably similar. The singular
behavior, which is the dominant feature in the variational n(p) at small
/?, is quite small in J(Y). When instrumental broadening is taken into
account the (now small) differences between the predicted scattering
for the two calculations all but disappear, as shown in Fig. 5(c). The
predictions of the two quite different n(p)s are now nearly indistinguish-
able ! In principle, the differences between the two different n(p)s are still
present in Fig. 5(c). In practice, a measurement of the scattering would
need fantastically good statistical accuracy to ever hope to observe these
differences.
This insensitivity to the details at small p is a direct consequence of
the IA. The Compton profile (9), which is proportional to the measured
scattering, is the momentum distribution in the direction of the momen-
tum transfer averaged over the longitudinal components. Since this is
the integral of pn(p) singular behavior at small p will be suppressed due
to the factor of p in the integrand. Alternately, features at large p will be
enhanced for exactly the same reason.
 BEC in Liquid Helium 65

""1"":
0.5 0.5
— HNO/$ : : (c) A —HNC/$ :
dFMd
0.4 — / \ - 0.4
'— f
0.3 0.3
\ ~
'- \ -
0.2 — \ ~ h / \ H 0.2
"1"

0.1 - I \ - 0.1

0.0 0.0

-4 -2 4 - 4 - 2

Fig. 5. (a) Two theoretical calculations of n(p) in liquid helium at T = 0. The solid
curve is the GFMC result, and the dashed curve is the HNC/S variational result,
(b) The longitudinal momentum distributions described by J(Y) corresponding
to the two ground state n(p) in (a), (c) The combined effects of the convolution
of J(Y) in (b) with the instrumental resolution function and Silver's final-state
broadening.

Directly extracting n(p) from J(Y) suffers similar problems. Statistical

noise, which is present in any measurement, will allow a variety of
different scattering functions to be consistent with the data. This problem
is particulaly severe at small p where the inversion procedure magnifies
the errors. Therefore, in view of the difficulties in extracting n(p) from the
experimental results, we find it more appropriate to work directly with
J(Y). Theoretical predictions can be compared to the experimental data
using the I A. Working with J(Y), as opposed to n(p), has the distinct
advantage that the statistical errors on the data provide a direct measure
of the 'goodness-of-fit' between theory and experiment.
66 P. Sokol
The discussion above has implicitly assumed that we can measure the
scattering in the IA limit, complicated only by the effects of instrumental
resolution. However, formally, the IA is only valid in the limit of infinite
co and Q. In real experiments, where co and Q are finite and collisions
between atoms are always present, deviations from the IA must be taken
into account [30]. The important questions are then the size and form of
the deviations from the IA and the extent to which they limit our ability
to infer n(p) from scattering data. Obviously, the IA becomes more
accurate with larger co and Q (i.e. the shorter the probe/struck-particle
interaction time), the weaker the interactions between the struck particle
and its neighbors.
At high co and Q, the most important deviations from the IA are
due to collisions of the recoiling particle with its neighbors, known as
final-state effects. Elementary arguments [4] lead to the prediction that
the IA result will experience a Lorentzian broadening with a full width
at half maximum (FWHM)

AYFWHM » pcjtot(G), (11)

where (Ttot(Q) is the atom-atom scattering cross section. Heuristically, in

terms of the uncertainty principle, this broadening is due to the finite
'lifetime' of the struck particle before it collides with its neighbors. This
simple result for the width of final-state broadening holds in several
modern theories for FSE. However, there is much debate over the details
of the lineshape [30]. For example, the Lorentzian broadening model
cannot be rigorously correct since it violates the second-moment co2 sum
rule obeyed by S(Q,co). However, the simple result (11) indicates the
property which is the most important in determining the FSE in real
systems: the strength of the interparticle interactions.
An important consequence of the final-state broadening given by (11)
is that the scattering will approach the IA very slowly with increasing
Q, because cjtot(Q) decreases logarithmically with Q for He-He scatter-
ing [31]. Therefore, helium is very similar to a hard-sphere system [29]
where approximate Y -scaling behavior is obtained without the IA be-
ing valid. The final-state broadening interferes with observation of the
features in n(p) which are the most interesting, i.e. the Bose-condensate
peak in 4He.
Any attempt to determine n(p) must somehow take FSE into account.
At present, we depend on theoretical calculations to provide the appro-
priate corrections. Unfortunately, this adds another layer of complication
and uncertainty to the interpretation of the experimental results. Fortu-
 BEC in Liquid Helium 67

0.1 —

0.0

-4

Y (A" 1 )

Fig. 6. The observed scattering converted to J(Y), at a Q of 23 A"1 and a

temperature of 4.2 K. The calculated instrumental resolution (solid line) and
final-state broadening (dashed line) are shown. The instrumental and final-state
broadening have similar widths and are both much less than the intrinsic width
of the observed scattering.

nately, recent theoretical work has provided an accurate description of

the FSE broadening that can be used in extracting information about
n(p).
The recent non-Lorentzian broadening theory by Silver [32] for 4He,
based on earlier work by Gersch and Rodrigue [33], yields the final-state
broadening function shown in Fig. 6 which is to be convoluted with
the impulse approximation prediction to obtain the predicted scattering.
Similar results have been obtained by Carraro and Koonin [34] based on
a variational approach. The instrumental broadening, calculated using a
Monte Carlo simulation [35], is also shown in Fig. 6 for reference. As
can be seen, the instrumental resolution and final-state broadening are of
comparable width, and both are much narrower than the intrinsic width
of the helium scattering.
The FWHM of the function in Fig. 6 is comparable to that from the
simple Lorentzian broadening theory. However, the modern theories [32,
68 P. Sokol
34] predict negative wings at large \Y\ so that the resulting S(Q,a>)
satisfies the co2 sum rule which requires that the second moment of the
broadening function be zero. The additional physics which this theory
takes into account is the pair-correlation function of the ground state
of the interacting system, which governs the collision rate as a function
of recoil distance. The overall effect of final-state broadening predicted
by Silver is much smaller than the simple Lorentzian treatment, and
produces excellent agreement with recent experiments [36].

5 Experimental Results
The suggestion that neutron scattering measurements at large momentum
transfers could directly observe the condensate set in motion a number
of studies by a variety of investigators [37]. Unfortunately, none of
these studies has succeeded in reaching the original goal: a direct and
unambiguous observation of the condensate. In fact, with our current
understanding of these measurements and FSE which complicate them,
we now realize that a direct observation of a 3 -function component in
n(p) is unlikely in the foreseeable future [38]. However, these studies
of liquid helium have provided a wealth of detailed information on
the momentum distribution in this strongly interacting quantum liquid.
With the use of some well-founded theoretical results, information on
the magnitude of the condensate fraction can be extracted, including its
temperature and density dependence.
The earliest attempts to measure n(p) and no were carried out at reactor
based sources [39]. These measurements provided general information
on n(p) and pointed the way toward techniques to extract the condensate.
However, they were limited by the large FSE at the relatively low Q («
5-15 A" 1 ) attainable at reactor based sources. Some measurements were
carried out at larger Q, but only at the expense of resolution making it
difficult to obtain accurate information on n(p).
More recently, measurements [40, 41, 42, 43, 44] have been carried
out using spallation neutron sources, such as the Intense Pulsed Neutron
Source (IPNS) at Argonne National Laboratory. These sources have a
high flux of epithermal (high energy) neutrons allowing high resolution
measurements to be carried out at much larger Q than are obtainable
with reactors. The higher Q have the advantage that the scattering is
consistent with the predictions of the IA, in terms of the location and
symmetry of the observed scattering. In addition, FSE are more amenable
 BEC in Liquid Helium 69

0.2 —

0.1 —

0.0

-4

Y (A" 1 )

Fig. 7. Observed scattering at T = 0.35 K. The line is a fit of the model scattering
function discussed in the text broadened by the instrumental resolution function
and the final-state broadening function of Silver [32].

to theoretical treatment [30] at these higher Q. Therefore, information

on the underlying n(p) can be more readily obtained.
To illustrate the results obtainable at spallation sources, we will discuss
recent measurements [40, 41, 42] using the high resolution PHOENIX
spectrometer at the IPNS at Argonne. Fig. 6 shows the measured
scattering at a Q of 23 A" 1 , converted to J(Y), at 4.2 K in the normal
liquid. The scattering shown in Fig. 6 is in qualitative agreement with the
predictions of the IA. It is centered at, and symmetric about, Y — 0. In
addition, measurements [28] at a variety of Q have shown that the shape
of J(Y) is independent of Q above approximately 15 A" 1 . Thus, at least
at this qualitative level, the scattering is well described by the IA.
The observed scattering in the normal liquid is broad and featureless.
J(Y) is nearly Gaussian, as in classical liquids [52]. Upon cooling into
the superfluid phase the scattering, shown in Fig. 7, becomes visibly more
peaked near 7 = 0 . However, no distinct condensate peak is observed.
The increase in scattering at small Y is consistent with the appearance of
a condensate peak broadened by the finite instrumental resolution and
70 P. Sokol

0.00

-0.25

-0.50

-0.75
t
•- 4I , -, ,2, I , , , 0
3
, ! • , . , 21 , , . , 4I
1
Y (A- )
Fig. 8. Observed J(Y) at a variety of temperatures in the normal and superfluid
phases at a density of 0.147 g cm"3.

FSE. However, as we discuss below, it is also consistent with many other

forms for n(p) as well [41], some of which do not have a finite no.
The temperature dependence of the neutron scattering is displayed in
Fig. 8. The scattering is nearly Gaussian and temperature independent
above the superfluid transition. Below the transition, the scattering
becomes non-Gaussian with an increase in intensity around 7 = 0 ,
consistent with the appearance of a condensate. There is a rapid change
in the scattering from just below the transition to 1.5 K, after which
the scattering changes little. This is consistent with Path Integral Monte
Carlo predictions [25] that the condensate fraction is largely temperature
independent below 1.5 K but decreases rapidly above that temperature.
The ultimate goal of these studies is to obtain information on n(p) in
general and the condensate in particular. Towards that end, we would like
 BEC in Liquid Helium 71

11
• • I" " I• • " I" "1
All
0.50

(A"1)
Fig. 9. JModei(^) with instrumental resolution and FSE deconvoluted obtained
from the observed scattering shown in Fig. 8.

to remove the effects of instrumental resolution and FSE and to invert the
transformation between n(p) and J(Y). However, as discussed previously,
these are ill-defined procedures [45] that are strongly affected by statistical
noise present in the experimental data. Therefore, rather than attempting
to deconvolute the instrumental resolution, it is more appropriate to fit
an expression for J(Y), broadened by instrumental resolution and FSE,
to the observed scattering data. Theoretical predictions for J(Y) are
discussed in the following section. Here we discuss a simple expression
for J(Y) which has sufficient flexibility to reflect the behavior of the
true scattering accurately, yet which is also sufficiently constrained so
that unphysical behavior is not introduced due to the finite statistical
accuracy of the data.
The scattering function that we have found most convenient [41] for
describing the observed scattering is a sum of two Gaussians

(12)

whose amplitudes, widths and common center may be varied. This

form is not unique and many other forms could be used to fit the data.
Nevertheless it has sufficient flexibility that it can represent the scattering
in both the normal liquid, where the scattering is nearly Gaussian, and
the superfluid.
Fig. 9 shows the resultant model J(Y), which represents the scattering
72 P. Sokol

0.6 1 1
, , , , 1, , , i I
" I data "
T=0.35K
- (upper) 2d-t n 0 =10%-

0.4 —
: (a)
A - (middle) 2(3 +n0 =4%J
(lower) 2d+ n Q =0% -

0.2 -
]

0.0 ,1 i i , • \7 —
1
LTIWMI
1 1
-4 -2

- (upper) 2 d + n Q =0%

- (middle) 2 d + n Q = 4 % ,

- ( l o w e r ) 2 d + n Q =10% -

0.0

Fig. 10. (a) Three different two-Gaussian fits to the data at 0.35 K. (b) The
corresponding model momentum distributions JModei(Y). One is a two-Gaussian
fit with a condensate fraction of 10% , another is a two-Gaussian fit with
a condensate fraction of 4% , and the third is a two-Gaussian fit with no
condensate fraction.

after the removal of instrumental resolution and FSE, over a broad

range of temperatures. As can be seen, the corrected scattering is nearly
Gaussian above Tx. Below Tx the scattering becomes non-Gaussian, with
a significant increase in the strength at small Y. However, in no cases
do the fits indicate that we need a 6 -function component to describe the
scattering adequately.
The above results indicate that a <5-function component is not necessary
to explain the scattering in the superfluid. This does not, however,
mean that the scattering is not consistent with a momentum distribution
containing a sharp peak at p = 0. As an illustration of the range of
model scattering functions which can fit the data equally well, consider
 BEC in Liquid Helium 73
the three different two-Gaussian fits to the scattering at T = 0.35 K
shown in Fig. 10. The dashed and dotted lines show the underlying
distributions which best fit the data if they are constrained to possess
a narrow component by fixing the width of one of the Gaussians to
be very narrow {a = 0.03 in this case). The dashed line has a narrow
component with 10% of the total area, and the dotted line has a narrow
component with 4% of the total area. Finally, the solid line is a two-
Gaussian fit in which both the amplitudes and widths are allowed to vary.
The x2 values of these fits are very close to each other. If we think of
the narrow component of the fits as representing the contribution from
the condensate and the broad component as representing the momentum
distribution of the uncondensed atoms, then this example shows that, with
an appropriate choice for the uncondensed component, the scattering
data is consistent with a condensate fraction ranging from zero (no
condensate) to values above 10 %.

6 Comparison to Theory
A direct model-independent determination of no is beyond current ex-
perimental capabilities for the reasons discussed in Section 5. However,
we may still obtain information on the condensate by comparing theo-
retical calculations of n(p) with the experimental data. Such comparisons
provide a direct test of the theoretical predictions and, indirectly, give
information about the magnitude of the condensate.
The dashed line in Fig. 11 (a) shows the theoretical prediction for JIA(Y)
in the normal liquid using the PIMC calculations [25] of n(p). The the-
oretical n(p) has been converted to J(Y) using the IA and broadened
by the instrumental resolution. The agreement between the theoretical
predication and the experiment is excellent. In this case the IA, calcu-
lated using the theoretical n(p)9 provides an excellent description of the
scattering in the normal liquid.
FSE have little effect on the observed scattering in the normal liquid
at these Q values. However, they may be included by convoluting the
theoretical predictions with the broadening function shown in Fig. 6. The
solid lines in Fig. 11 (a) and (b) are obtained when FSE are included.
Within the statistical accuracy of the measurements, there is no observable
change when FSE are included. Since, based on the co2 sum rule, FSE do
not change the second moment of the scattering, they have little effect
on the broad, nearly Gaussian J(Y) of the normal liquid.
74 P. Sokol

1 1 1 1 1
. " " I " I ' " I " : . ...1 . 0.5
0.5 - (a) I 3.5 K : I 0.35 K :

0.4 — jjjjfe
H
—
l — 0.4

J(Y) °-;
0.3

0.2
rF /A\ I
h /\
- / \
-

-
0.3

0.2

\
1 1 1 1 1
i i 1 i

0.1 0.1

0.0
0.0 ^fTMi.,,1 I I I -

-4 -2

Y (A"1)
Fig. 11. Comparison of the observed scattering in (a) the normal liquid at 3.5 K
and (b) the superfluid at 0.35 K, with the theoretical predictions from GFMC
and PIMC calculations. The dashed lines are the theoretical predictions with only
instrumental broadening included. The solid lines are the theoretical predictions
with instrumental resolution and final-state effects included.

The dashed line in Fig. ll(b) shows a similar comparison of the GFMC
calculations [21] to the scattering in the superfluid. The agreement
between the theoretical and experimental results is quite poor, particularly
in the region of the peak center, where the condensate would have the
largest contribution. Based on this comparison alone, which has neglected
FSE, we would conclude that no, if not identically zero, would have a
much smaller value than the theoretical predictions.
The inclusion of FSE in the comparison between the experimental and
theoretical results is essential in the superfluid phase. This is due to the
appearance of a sharp feature in n(p) for the superfluid: the condensate.
While FSE have little effect on the broad component of the scattering, as
observed in the normal liquid, they significantly broaden the contribution
from the condensate. Taking FSE into account, the agreement between
theory and experiment is now excellent! For the first time, ab initio
numerical calculations of n(p) in the superfluid are in good agreement
with experiment [40].
An important point regarding the final-state corrections is in order
here. The final-state broadening prediction of Silver, as shown in Fig. 6,
has a narrow central peak and negative tails at high Y. The negative tails
are essential if the broadening function is to satisfy the second-moment
 BEC in Liquid Helium 75
sum rule. Thus, final-state effects not only broaden the condensate peak,
they also shift intensity throughout the entire spectrum. For the particular
form of the FSE used here, the negative tails will cause a depletion of the
scattering at intermediate Y when a condensate is present.
In view of the discussion in Section 5 regarding the relationship be-
tween J(Y) and n(p\ it is appropriate to examine how sensitive the
observed scattering is to the theoretical n(p). Inevitably there is a finite
statistical accuracy attached to the experimental results, and a whole
range of different n(p)s may give equally good agreement with the data.
If the statistical accuracy of the results is high then only a limited range
of n(p)s, all with very similar shapes, will be consistent with the data.
Alternatively, if the statistical accuracy is poor then the experimental
results will only place very weak constraints on the underlying shape of
n(p).
Unfortunately, it is not possible to vary the condensate fraction in
numerical calculations to see what effect it has on the comparison with
the experimental data. The numerical calculations use as input only the
known interatomic potential. Based on this potential, the full many body
state of the liquid is calculated. There are no parameters that one can
adjust to make no larger or smaller. The result of the simulation, just
as in the real system, is determined solely by the interaction between
the atoms. Short of adjusting the interaction, which would undoubtedly
change the entire n(p), we cannot adjust no.
We can, however, attempt to make a limited test of the sensitivity of
the experimental results to the value of the condensate fraction. The
momentum distribution can be decomposed into a condensate contri-
bution and a non-condensed contribution. We may adjust the relative
weights of the condensate component and the non-condensed compo-
nent such that they still satisfy the normalization of n(p). However,
once we choose a value for the condensate different from the theoret-
ical value the resultant momentum distribution no longer corresponds
to any ab initio calculation for helium. Thus, while we are testing the
sensitivity of the scattering to the value of no, it is only in the lim-
ited range where n(p) has a very special shape for the non-condensed
component.
With the above caveats in mind, we may replace the condensate d-
function with a Gaussian of variable width and amplitude, keeping the
shape of the non-condensate distribution the same as the theoretical
prediction. The best agreement is obtained when the width of the
Gaussian is less than 0.05 A" 1 and no is 10%, in agreement with the
76 P. Sokol
best numerical predictions for the condensate. Significant deviations are
observed when the width is greater than 0.2 A" 1 and no is less than 8%
or greater than 12%. For this particular model for the non-condensate
part of n(p) provided by the GFMC calculations, we find that there is
indeed a condensate with no — 10 ± 2%. However, we note that changing
the shape of the uncondensed n(p) would also have an effect on the value
for the condensate fraction.
In a similar fashion, the sensitivity to the expected singular behavior in
n(p) at small p can be examined. The GFMC results, which give excellent
agreement with the observed scattering, do not contain the expected
singular contribution. (Because of finite-size effects, these results are
only valid for p > TC/L, where L is the system size.) The variational
n(p), discussed previously, explicitly includes this behavior. Both results
are in excellant agreement with experiment. This is not very surprising,
since the weak singular behavior at small p is suppressed when n(p) is
transformed to J(Y), as discussed earlier. Thus, the predicted small p
singular behavior makes little contribution to the observed scattering
and, with the experimental techniques now available, will be difficult, if
not impossible, to observe.
Thus, the experimental results in the superfluid provide a clear indi-
cation of a narrow component in n(p) containing approximately 9-10%
of the intensity, which is precisely that expected for the condensate.
Unfortunately, due to the finite statistical error inherent in any exper-
iment, they cannot definitely prove the existence of a condensate in
the form of a <5-function. Some other singular behavior not associated
with a condensate could be responsible for the increase in the scatter-
ing at small p observed in the superfluid. As seen in the comparison
with the variational n(p), however, this would have to be very singular
behavior, much more than the 1/p singularity, to agree with the exper-
imental results. Thus, while the experimental results cannot rule out
a ground state n(p) which does not contain a S-function condensate,
they do provide strong evidence for a very narrow feature containing
10+2% of the total area. The excellent agreement with the numerical
results suggests that this very narrow feature is indeed due to the Bose
condensate [3].
Similar comparisons have been carried out at a variety of temperatures
and several different densities in both the normal and superfluid phases.
GFMC and variational calculations are used for comparison with low
temperature measurements and PIMC results are used for comparison
to measurements above 1 K. The agreement is excellent over the entire
 BEC in Liquid Helium 77
temperature range! Theory and experiment appear to have converged
for n(p) in liquid 4 He at low densities (SVP).

7 Empirical Estimates of the Condensate Fraction

Due to its importance in understanding the superfluid phase, considerable

emphasis has been placed on determining no. Since the width of FSE
broadening follows the very slow decrease of the He-He cross section at
higher Q, it is very unlikely that a condensate peak can be resolved in
deep inelastic neutron scattering experiments [38].
The excellent agreement between theoretical calculations of n(p) and
the scattering data is strong (though not conclusive) evidence for the
presence of a condensate. It is important to realize, however, that there
are a number of physical systems, such as 3 He- 4 He mixtures and 4 He in
confined geometries, for which accurate theoretical calculations of n(p)
are difficult to carry out. For these systems, it may not be possible to
carry out direct comparisons to theory.
Therefore, it is important to develop alternative techniques to extract
information on the magnitude of no from experimental data. All of these,
in one form or another, involve modeling n(p) in the superfluid to extract
a value for no. Early models used for extracting the condensate usually
consisted of simple functions that could be fit easily to the experimental
data, such as a sum of Gaussians. Results for no ranging from 0 (no
condensate) to 20% were obtained, leading to considerable confusion
about the condensate [46].
A major improvement in analysis of experimental results was intro-
duced by the work of Sears and coworkers [47], which attempted to
incorporate known behavior of the momentum distribution as a con-
straint in extracting the value of no- The momentum distribution in the
superfluid phase was modeled as a £-function due to the condensate, a
singular term induced by the condensate which had been predicted the-
oretically, and an uncondensed component which was assumed to have
the same shape as n(p) in the normal liquid. Using this technique, values
for the condensate fraction of 10-13% (when extrapolated to T = 0)
were obtained [5]. These results are in reasonable agreement with current
theoretical predictions. However, when an improved form for the singular
term is used [48], the inferred values of no are 4-5%, considerably below
the theoretical estimates.
The Sears method suffers from some important limitations. For ex-
78 P. Sokol
ample, it assumes that the shape of the non-singular part of n(p) in the
superfluid is the same shape as the full n(p) in the normal liquid just
above the superfluid transition. At best this is probably only a reasonable
first approximation. In addition, the necessity of measuring n(p) in the
normal liquid phase (above Tx) introduces other uncertainties which will
affect the determination of the condensate from the measurements below

An alternative method [50] for determining the magnitude of no was

recently introduced by Snow, Wang and Sokol. The expression used
for n(p) explicitly incorporates the known analytic behavior of n(p) at
small p and places minimal restrictions on the shape of n(p) in other
regions. This expression for n(p), converted to J(Y) and broadened by
instrumental resolution and FSE, is compared directly to the scattering
and the model parameters adjusted to obtain the best fit.
The expression used by Snow et al. separated n(p) into two components
at an arbitrary cutoff momentum pc. Above pc a single Gaussian was
used to model n(p). Below this cutoff a S-function, the known singular
behavior and, to allow matching at p c , a Gaussian were used to model
n(p). When the normalization of n(p) and continuity above and below pc
are enforced, this model has four adjustable parameters: the widths of
the two Gaussians o\ and 02, the cutoff momentum pc and the condensate
fraction no-
The fitting parameters are all related to various unknown aspects of
the momentum distribution. The cutoff momentum pc is the momentum
at which the condensate-induced singularity becomes negligible in com-
parison to the non-singular part of the momentum distribution. However,
the results are not sensitive to the choice of pc as long as it is in the
range 0.1 < p < 0.7 A" 1 . This is just the phonon region where the
theoretical expression for the condensate-induced singularity is expected
to be valid [18, 19]. The value of 02 is determined primarily by the width
of the momentum distribution in the intermediate-to-large Y region. The
value of G\, on the other hand, is not tightly constrained by the scattering
data. It represents the unknown effect near Y = 0 of the condensate on
the non-singular part of the momentum distribution. Finally, the value
for no is determined primarily by the scattering at small p. Thus, in
practice, this fit has just two adjustable parameters, no and 02, which are
determined by different regions of the observed scattering.
Fig. 12 shows no as a function of temperature at low density as
determined by the above procedure, along with the results of numerical
simulations. The extracted value of the T = 0 condensate fraction of
 BEC in Liquid Helium 79

15.0

12.5 -

10.0 -

2.5 —

0.0

Fig. 12. The value of n0 as a function of temperature at a constant density of

0.147 g cm"3, obtained from fits using the method of Snow et al.[50] (crosses).
Also shown are the theoretical calculations of n0 at T = 0 from GFMC (square)
and at finite temperatures from PIMC calculations (diamonds). The solid line is
a plot of the renormalization group theory prediction.

10% is in excellent agreement with the GFMC result of 9.2%. The weak
temperature dependence of n0 seen in these fits at low temperatures is
also consistent with the PIMC results [25] and the finite-temperature
extensions of variational calculations.
The condensate wave function must vanish at the second-order
superfluid-normal fluid phase transition, since it is the microscopic order
parameter for the superfluid phase. According to the renormalization
group theory for second-order phase transitions, the temperature de-
pendence of rc0 near Tx is expected to take the form n0 = At2P, where
t = (1 - T/Tx). Unfortunately, the errors in the current experimental
estimates are too large to place any useful constraints on the critical
exponent which controls the behavior of the condensate close to the
superfluid transition. However, we may compare this prediction with our
experimental results using the accepted value of j8=0.35 obtained from
80 P. Sokol

12.5 -

Liquid Inaccessible Solid

10.0 -
•1 «•
75
- 7
:
5.0 - • 1

D
2.5 -

nn
0.14 0.16
1 . . . , 1 , ,
0.18
I.,..
0.2

P (g/cc)
Fig. 13. The value of n0 as a function of density at 0.75 K obtained from fits
using the method of Snow et al. [50] (crosses). The GFMC results for n0 (squares)
and the HNC/S results (diamonds) are also shown.

superfluid fraction [49] and heat capacity [51] measurements. Such a

comparison is shown in Fig. 12 and, as can be seen, the agreement is
excellent over a surprisingly large temperature range.
Fig. 13 shows the density-dependence results at T = 0.75 K. The
estimates of no are systematically somewhat higher than the theoretical
results using the GFMC method, but show the same density dependence
within the experimental error. The condensate fraction decreases with
increasing density. Notice that the estimate for the condensate remains
finite close to the superfluid-solid phase transition. Since the superfluid-
solid phase transition is first order, there is no reason for the condensate
to vanish continuously near the critical density. We see no evidence for
the presence of a condensate in the solid phase, a further indication of
the adequacy of our method for extracting no-
As a final note, it is important to keep in mind that all these different
methods have one feature in common - they implicitly assume the
 BEC in Liquid Helium 81
existence of a condensate. As discussed previously, the actual scattering
data is also consistent with an increase in scattering at small p without
the appearance of a condensate, but then one would have to come up
with a plausible theory of the origin of this anomalous scattering.

8 Conclusions

We have now reached a stage where there is excellent agreement between

experiment and theory for n(p) in both the normal and superfluid phases
of liquid 4He. Direct comparison of the experimental results to theory
and to various empirical expressions provides convincing evidence for a
Bose condensate containing 10% of the atoms at T = 0. In addition,
the temperature and density dependence of no have been measured and
are also in agreement with theoretical predictions obtained by numerical
simulations.
Unfortunately, the original goal [4] of a direct observation of the
condensate fraction has not come to pass. In view of our current
understanding of FSE in helium, it is unlikely that this goal will ever
be reached in deep inelastic neutron scattering experiments. Thus, while
the current experiments provide convincing evidence for a condensate,
they do not provide the sought-after direct observation of the condensate
5 -function. The values inferred for no are dependent on the current
theories and models that are used to interpret experimental results.
The difficulties of observing Bose condensation directly in liquid helium
are somewhat surprising considering the ease with which the effects of
the condensate can be observed. Many easily measured quantities, such
as the superfluid fraction, second sound and thermal conductivity exhibit
large changes when Bose condensation occurs. However, due to the lack
of a detailed microscopic theory, none of these changes can be used to
determine the magnitude of the condensate directly. We note, however,
that the superfluid fraction and the specific heat have been calculated
using the PIMC method in conjunction with the condensate fraction [25].
Effectively, these results give us ps(T) as a function of no(T).
The situation is similar for many microscopic quantities which change
when Bose-Einstein condensation occurs. For example, the static struc-
ture factor S(Q) changes slightly when the liquid enters the superfluid
phase. Phenomenological theories attempting to relate these changes to
no provide reasonable results [53]. However, the predictions of these
theories have been shown to be incorrect [25, 54]. Similarly, attempts to
82 P. Sokol
relate changes in the average kinetic energy per atom to the condensate
fraction [5] have also been shown to be incorrect.
Lacking the ability to observe the condensate directly and unam-
biguously, attention has focused on semi-empirical expressions for the
momentum distribution. As we have seen, fits using these expressions
are capable of providing information on the magnitude of the conden-
sate, and they have been employed to study the temperature and density
dependence of the condensate.
It is essential to remember, however, that if the underlying assumptions
of the expressions used in the fits are incorrect, then the results inferred
will not have much meaning. It is interesting to note that the observed
scattering in liquid helium can be fit extremely well simply by using the
momentum distribution of the ideal Bose gas [55]. This holds true in both
the normal liquid phase and in the superfluid phase. The best description
of the scattering in the superfluid is obtained by using an ideal Bose gas
n(p) with no condensate. Based on this naive model we might conclude
that while Bose statistics alter n(p) from the classical result, as indicated
by the large increase in intensity at small p, a condensate is not necessary
to explain the observed scattering. This illustrates that it is essential
that only expressions with a sound physical basis must be employed if
meaningful information on the condensate is to be obtained. Moreover,
it warns that even for weakly interacting Bose gases, e.g. excitons, fits to
the ideal Bose gas momentum distribution should not be overinterpreted,
since interactions may significantly alter and obscure the appearance of
a condensate peak in the spectral data.

Acknowledgements. I would like to acknowledge the contributions of

T. R. Sosnick, W. M. Snow, Y. Wang, K. Herwig, and R. Blasdell in
the experimental measurements described here. I would also like to
acknowledge many useful conversations with R. N. Silver, A. Griffin, H.
R. Glyde, and W. G. Stirling.
This work was supported by NSF grant DMR-9123469 and OBES/
DMS support of the Intense Pulsed Neutron Source at Argonne National
Laboratory under DOE grant W-31-109-Eng-38.

References
[1] The cessation of bubbling as 4 He passed through the superfluid
transition was first noted by J. C. McLennan, H. D. Smith, and J.
O. Wilhelm, Phil. Mag. 14, 161 (1932).
 BEC in Liquid Helium 83
[2] The first experiments showing superfluidity explicitly were reported
by J.F. Allen and A.D. Misener, Nature 141, 75 (1938), and P.
Kapitza, Nature 141, 74 (1938).
[3] F. London, Nature 141, 643 (1938).
[4] This method was first proposed by P.C. Hohenberg and P.M. Platz-
man, Phys. Rev. 152, 198 (1966).
[5] H.R. Glyde and E.C. Svensson, in Methods of Experimental Physics,
K. Skold and D. L. Price, eds. (Academic, N.Y., 1987), Vol.23, Part
B, p. 303.
[6] RE. Sokol, T. R. Sosnick, and W.M. Snow, in Momentum Distri-
butions, R. N. Silver and P. E. Sokol, eds. (Plenum, N.Y, 1989),
p.139.
[7] R.K. Pathria, Statistical Mechanics (Pergamon, Oxford, 1972).
[8] R.A. Aziz, V.P.S. Nain, J.S. Cerley, W.L. Taylor, and G.T. Mc-
Conville, J. Chem. Phys. 70, 4330 (1979).
[9] W.E. Keller, Helium-3 and Helium-4 (Plenum, N.Y., 1969).
[10] P. E. Sokol, R.N. Silver and J. W. Clark, in Momentum Distributions,
R.N. Silver and P. E. Sokol, eds. (Plenum, N.Y, 1989).
[11] el Hadi et al, Physica 41, 289-304 (1969); Phys. Chem. Ref. Data
2, 923 (1973); 6, 51 (1977).
[12] O. Penrose and L. Onsager, Phys. Rev. 104, 576 (1956).
[13] C.N. Yang, Rev. Mod. Phys. B 34, 694 (1962).
[14] P.W. Anderson, Rev. Mod. Phys. B 38, 298 (1966).
[15] G. Baym, private communication.
[16] J. W. Clark and M.L. Ristig, in Momentum Distributions, R. N. Silver
and P. E. Sokol, eds. (Plenum, N.Y, 1989), p. 39.
[17] A. Griffin, Can. J. Phys. 65, 1368 (1987).
[18] P.C. Hohenberg and P.C. Martin, Ann. Phys. (NY) 34, 291 (1965).
[19] J. Gavoret and P. Nozieres, Ann. Phys. (NY) 28, 349 (1964).
[20] E. Manousakis, V. R. Pandharipande, and Q. N. Usmani, Phys. Rev.
B 33, 7022 (1985); E. Manousakis and V. R. Pandharipande, Phys.
Rev. B 30, 5064 (1984); E. Manousakis, in Momentum Distributions,
R. N. Silver and P. E. Sokol, eds. (Plenum, N.Y, 1989), p. 81.
[21] P.A. Whitlock and R. Panoff, Can. J. Phys. 65, 1409 (1987); R.M.
Panoff and P.A. Whitlock, in Momentum Distributions, R. N. Silver
and P. E. Sokol, eds. (Plenum, N.Y, 1989), p. 59.
[22] S.A. Vitiello, K.J. Runge, G.V. Chester and M.H. Kalos, Phys. Rev.
B 42, 228 (1990).
84 P. Sokol
[23] E. Feenberg, Theory of Quantum Fluids (Academic, N.Y., 1969); C.E.
Campbell, Progress in Liquid Physics, C.A. Croxton, ed. (Wiley,
N.Y., 1968), ch. 6.
[24] C.E. Campbell, in Excitations in Two-Dimensional and Three-
Dimensional Quantum Fluids, A.G.F. Wyatt and H.J. Lauter, eds.
(Plenum, N.Y., 1991), p. 159.
[25] D.M.Ceperly and E.L.Pollock, Phys. Rev. Lett. 56, 351 (1986);
D.M. Ceperly, and E.L. Pollock, Can. J. Phys. 65, 1416 (1987);
E.L. Pollock and D.M. Ceperley, Phys. Rev. B 36, 8343 (1987);
D.M. Ceperley, in Momentum Distributions, R.N. Silver and
RE. Sokol, eds. (Plenum, N.Y, 1989), p. 71.
[26] A.D.B. Woods and V.F. Sears, Phys. Rev. Lett. 39, 415 (1977); H.A.
Mook, Phys. Rev. Lett. 32, 1167 (1974).
[27] D.L. Price and K. Skold in Methods in Experimental Physics, Vol.
23A, D.L. Price and K. Skold, eds. (Academic, N.Y, 1987), p. 1.
[28] RE. Sokol, Can. J. Phys. 65, 1393 (1987).
[29] J.J. Weinstein and J.W. Negele, Phys. Rev. Lett. 49, 1016 (1982).
[30] R.N. Silver, in Momentum Distributions, R.N. Silver and RE. Sokol,
eds. (Plenum, N.Y, 1989), p. 111.
[31] R. Feltgen, H. Kirst, K.A. Koehler, and F. Torello, J. Chem. Phys.
26, 2630 (1982).
[32] R.N. Silver, Phys. Rev. B 37, 3794 (1988); Phys. Rev. B 39, 4022
(1989).
[33] H. A. Gersch and L. J. Rodriguez, Phys. Rev. A 8, 905 (1973); L.
J. Rodriguez, H. A. Gersch and H. A. Mook, Phys. Rev. A 9, 2085
(1974).
[34] C. Carraro and S.E. Koonin, Phys. Rev. Lett. 65, 2792 (1990); Phys.
Rev. B 41, 6741 (1990).
[35] P.E. Sokol, to be published.
[36] P.E. Sokol, T.R. Sosnick, W.M. Snow, and R.N. Silver, Phys. Rev.
B 43, 216 (1991); K.W. Herwig, P.E. Sokol, W.M. Snow, and R.C.
Blasdell, Phys. Rev. B 44, 308 (1991).
[37] For a brief review, see E.C. Svensson, in 75th Jubilee Conference on
Helium-4, J.G.M. Armitage, ed. (World Scientific, Singapore, 1983),
p. 10.
[38] P.E. Sokol and W.M. Snow, in Excitations in Two-Dimensional and
Three-Dimensional Quantum Fluids, A.G.F. Wyatt and H.J. Lauter,
eds. (Plenum, N.Y, 1991) p. 49.
[39] R.A. Cowley and A.D.B. Woods, Phys. Rev. Lett. 21, 787 (1968).
 BEC in Liquid Helium 85
[40] T.R. Sosnick, W.M. Snow, RE. Sokol, and R.N. Silver, Europhys.
Lett. 9, 707 (1989).
[41] T.R. Sosnick, W.M. Snow, and RE. Sokol, Phys. Rev. B41, 11185
(1990).
[42] K.W. Herwig, RE. Sokol, T.R. Sosnick, W.M. Snow, and R. C.
Blasdell, Phys. Rev. B 41, 103 (1990).
[43] K.H. Anderson, W.G. Stirling, A.D. Taylor, S.M. Bennington, Z.A.
Bowden, I. Bailey, and H.R. Glyde, Physica B 180, 865 (1993).
[44] J. Mayers and A.C. Evans, Rutherford Appleton Laboratory Report
RAL-91-048.
[45] D.S. Sivia and R.N. Silver, in Momentum Distributions, R.N. Silver
and RE. Sokol, eds. (Plenum, N.Y., 1989), p. 377.
[46] H. W. Jackson, Phys. Rev. A 10, 278 (1974).
[47] V.F. Sears, E.C. Svensson, P. Martel and A.D.B. Woods, Phys. Rev.
Lett. 49, 279 (1982).
[48] A. Griffin, Phys. Rev. B 32, 3289 (1985).
[49] J.A. Lipa and TCP. Chui, Phys. Rev. Lett. 51, 2291 (1983).
[50] W. M. Snow, Y. Wang and P. E. Sokol, Europhys. Lett. 19, 403
(1992).
[51] D.S. Greywall and G. Ahlers, Phys. Rev. A 7, 2145 (1973).
[52] V.F. Sears, Can. J. Phys. 63, 68 (1985).
[53] G. J. Hyland, G. Rowlands, and F. W. Cummings, Phys. Lett 31A,
465 (1970); F. W. Cummings, G. J. Hyland, and G. Rowlands, Phys.
Lett 86A, 370 (1981).
[54] G.L. Masserini, L. Reatto, and S.A. Vitiello, J. Low Temp. Phys. 89,
335 (1992).
[55] H.R. Glyde and W.G. Stirling, in Momentum Distributions, R.N. Sil-
ver and RE. Sokol, eds. (Plenum, N.Y, 1989), p. 123.
 Sum Rules and Bose-Einstein
Condensation
S Stringari
Dipartimento di Fisica
Universitd di Trento
1-38050 Povo
Italy

Abstract
Various sum rules, accounting for the coupling between density and particle
excitations and emphasizing in an explicit way the role of Bose-Einstein
condensation, are discussed. Important consequences on the fluctuations of
the particle operator as well as on the structure of elementary excitations
are reviewed. These include a recent generalization of the Hohenberg-
Mermin-Wagner theorem holding at zero temperature.

1 Introduction
The sum rule approach has been employed extensively in the literature
in order to explore various dynamic features of quantum many body
systems from a microscopic point of view (see [1] and references therein).
An important merit of the method is its explicit emphasis on the role of
conservation laws and of the symmetries of the problem. Furthermore,
the explicit determination of sum rules is relatively easy and often requires
only a limited knowledge of the system. Usually the sum rule approach
is, however, employed without giving special emphasis to the possible
occurrence of (spontaneously) broken symmetries. For example, the
most famous /-sum rule [2] holding for a large class of systems is not
affected by the existence of an order parameter in the system.
The purpose of this paper is to discuss a different class of sum rules
which are directly affected by the presence of a broken symmetry. These
sum rules can be used to predict significant properties of the system which
are the consequence of the existence of an order parameter. In this work
we will make explicit reference to Bose systems and to the consequences
of Bose-Einstein condensation (BEC). Most of these results can however

86
 Sum Rules and Bose-Einstein Condensation 87
be generalized to discuss other systems exhibiting spontaneously broken
symmetries.
In Section 2 we use a sum rule due to Bogoliubov in order to derive
important constraints on the fluctuations of the particle operator as
well as to obtain Goldstone-type bounds for the energy of elementary
excitations. In Section 3 we explore the consequences of BEC on the long
range behavior of the half diagonal two-body density matrix and discuss
the coupling between the density and particle pictures of elementary
excitations in Bose superfluids.

2 The Bogoliubov Sum Rule

The Bogoliubov sum rule [3, 4]

/:
is the most remarkable among the sum rules depending explicitly on the
order parameter n0 = jj | (a0) |2. In (1) s^A^B{o}) is the spectral function

| ~ ^ " e^En^m I At I "><* I B I m >

xd(co-En-Em) (2)
relative to the operators A, B and we have made the choice A^ = aq — al q
and B — pq (pq, aq and al q are the usual density, particle annihilation and
creation operators, respectively, and we have taken (ao) = (aj) = • N/^no)-
In deriving the result (1), we have used the completeness relation and
the Bose commutation relations for a and a\ Despite its simplicity, (1)
is at the basis of very important results characterizing the macroscopic
behavior of Bose superfluids. Its richness is mainly due to the fact that
it exploits in a microscopic way the effects of the "phase" operator,
proportional to the difference aq — al q . Some important applications of
(1) are discussed below.

2.1 The Hohenberg-Mermin-Wagner (HMW) Theorem

This theorem, of fundamental importance in condensed matter physics,
states that no long range order can exist (no = 0 in the Bose case) in a
significant class of one- and two-dimensional systems at finite tempera-
ture [5, 6, 7]. The theorem is based on the Bogoliubov inequality [3, 4]
{{A\A}){[B\[H,B]])>kBT | {[A\B]) |2 (3)
88 S. Stringari
for the fluctuations of the operator A (we assume (A) = (B) = 0).
Inequality (3) can be derived starting from the inequality [3, 4]
XAU(IB\[H,B]])>\([A\B])\2 (4)

for the static response relative to the operator A

= / )-d(o (5)
J—o 0)

and the result ({A^9A}) > 2UBTXA^A following from the fluctuation
dissipation theorem.
Choosing in the Bose case A* = aq and B = p q we obtain the inequal-
ity [5]

n(q) = (ajflq) ^ nom-^j- - - (6)

for the momentum distribution n(q) of the system. In order to derive

result (6) we have used the /-sum rule [2]

J coS(q,co)dco = iflpj, [/f,pq]]) = N^ (7)

holding for Galilean invariant potentials. In (7), S(q,a>) = j/ pt,p(co)/(l —

exp(—jSco)) is the usual dynamic structure function. Inequality (6) points
out the occurrence of an important infrared 1/q2 divergence in the
momentum distribution which originates from the thermal fluctuations of
the phase of the condensate. This behavior permits proof of the absence
of BEC in one- and two-dimensional systems at finite temperature. In
fact, due to such a divergence, the normalization condition for the
momentum distribution Xlqn(^) = N cannot be satisfied in one- and
two-dimensions, unless no = 0 [5].

2.2 Extension of the HMW Theorem at T—0

As clearly revealed by Eqs. (3) and (6), the HMW theorem does not
apply at zero temperature. Actually there are important examples of
two-dimensional systems obeying the HMW theorem at T =£ 0 and
exhibiting long range order in the ground state. The physical reason is
that quantum fluctuations have in general a weaker effect with respect
to thermal fluctuations in destroying the order parameter.
An extension of the Hohenberg-Mermin-Wagner theorem holding
 Sum Rules and Bose-Einstein Condensation 89
at zero temperature has been recently formulated by Pitaevskii and
Stringari [8]. To this purpose one uses the uncertainty principle inequality
({A\A})({B\B}) > | <[^,B]> | 2 (8)
rather than the Bogoliubov inequality (3). Both inequalities provide rigor-
ous constraints on the fluctuations of the operator A. However, while the
Bogoliubov inequality is sensitive to thermal fluctuations and becomes
less and less useful as T —> 0, the uncertainty principle is particularly
powerful in the low temperature regime dominated by quantum fluctua-
tions. By making the choice A^ = aq, B — pq one obtains the following
non-trivial inequality for the momentum distribution [8]:

of a Bose system where S(q) = jj(pqpq) is the static structure function.

The low q behavior of S(q) is fixed, at zero temperature, by the further
bound

S(q) <Jf coS(q, (o)dco f j-S(q, co)do> (10)

that can be easily calculated at small q. In fact, in this limit, the inverse
energy weighted sum rule (compressibility sum rule) / ^S(q,a>) = ^x(#)
approaches the compressibility parameter of the system (limq-+ox(cl) =
1/mc2), while the energy weighted sum rule (/-sum rule) is given by
(7). As a consequence S(q) < q/2mc at small q and the momentum
distribution n(q) diverges at least as
n(q) = nor^. (11)

It is worth pointing out that the l/q law for the momentum distribution,
already known in the literature [9, 10], has been obtained here without
any assumption on the nature of the elementary excitations of the system
(phonons in a neutral Bose superfluid) and follows uniquely from the
existence of BEC, the validity of the /-sum rule and the finiteness
of the compressibility. By imposing the proper normalization on the
momentum distribution one can then rule out the existence of BEC in
one-dimensional systems at zero temperature. Note that the ideal Bose
Gas does not violate the theorem since in this case the compressibility
sum rule for x(q) diverges as l/q2.
Starting from the same inequality (8), it is also possible to rule out the
existence of long range order in other one-dimensional systems at T = 0
90 S. Stringari
(isotropic antiferromagnets, crystals) [8]. The theorem does not apply
to one-dimensional ferromagnets since in this case the inverse energy
weighted sum rule (magnetic susceptibility) diverges as 1/q2.
Inequality (9) has been used recently [11] to rule out the existence of
BEC in the bosonic representation of the electronic wave function

^ . , ^ ) (12)
Uj

proposed by Girvin and MacDonald [12] for the fractional quantum

Hall effect. In Eq. (12), *¥ F is the fermionic wave function of electrons,
v = l/(2fc +1), where k is an integer, v is the usual filling factor and a,j is
the angle between the vector connecting particles i and j and an arbitrary
fixed axis. Using Laughlin's expression [13] for the ground state wave
function ^FV, the authors of Ref. [12] were able to show that there is no
Bose-Einstein condensation in the bosonic wave function ^ B , but only
algebraic long range order. The same result was obtained in Ref. [14]
starting directly from the Chern-Simons-Landau-Ginzburg theory. In
Ref. [11] the absence of long range order was proven starting directly
from the uncertainty principle inequality (9). The result follows from the
fact that a charged liquid in an external magnetic field is characterized
by a suppression of density fluctuations resulting in a quadratic law
2
S(q) = ^— (13)
2m<x>
for the static structure function (coc is the cyclotronic frquency). This
behaviour reflects the incompressibility of the system and is consistent
with Kohn's theorem [15]. Inequality (9) implies a 1/q2 divergence for
the momentum distribution of the bosonic wave function of this two-
dimensional problem. This is incompatible with the normalization of
n{q), fixed by the total number of electrons, unless the Bose condensate
relative to *¥B identically vanishes.

2.3 BEC and Excitation Energies

The sum rule technique can be used to obtain useful constraints on the
energy of elementary excitations. In particular, use of the Bogoliubov sum
rule (1) makes it possible to emphasize the role of the order parameter.
Given a pair of operators A and B one can derive the following
rigorous inequality, holding at zero temperature, for the energy a>o of the
 Sum Rules and Bose-Einstein Condensation 91
lowest state excited by the operators A and B:

2 ([A\[H,A]])([B\[H,B]])
M
°- \(\A\B\)\2 ' ( }

Bounds of the form (14) were first considered by Wagner [4]. Result (14)
can be obtained using the inequality (holding at zero temperature)

where the rhs coincides with the ratio between the energy weighted and
the inverse energy weighted sum rules relative to the operator A. Use of
the Bogoliubov inequality (4) then yields (14).
Result (14) has the important merit of providing a rigorous upper
bound for a>o in terms of quantities involving commutators. This is an
advantage, for example, with respect to the so-called Feynman bound:

involving the anticommutator {{A\A}) in the denominator and hence

requiring direct knowledge of the fluctuations of the operator A. The
occurrence of the anticommutator makes it difficult to exploit the low
momentum regime of elementary excitations. Vice versa, (14) can be
directly employed for this purpose.
Result (14) is particularly interesting when the expectation value of
the commutator [A\ B] is proportional to the order parameter of the
problem. This happens in a Bose superfluid with the choice A^ = aq — aLq
and B = p q , already considered in this work. One then finds

w 2(
° ^-;ri^ [ i^ + N F W p
(17)

where we have carried out the commutators using the grand canonical
hamiltonian H' = H — fiN (ja is the chemical potential) and taken a
central potential with Fourier transform V(p). The quantitities n(p) and
n(p) are defined by n(p) = (ajflp) and n(p) = \{a\aLv +a p a_ p ).
Result (17) is a rigorous inequality holding for any Bose system inter-
acting with central potentials. It has the form of a Goldstone theorem.
In fact, since its rhs behaves as q2 when <gr —>• 0, it proves the existence
of gapless excitations, provided the order parameter no is different from
zero.
It is worth noting that in the limit of a dilute Bose gas (/i = NV(Q\
92 S. Stringari

n(p) = n(p) = iV<5p,o,no = 1)> this upper bound coincides with the Bogoli-
ubov dispersion law
~2 r »i
(18)

for any value of q. This result follows from the fact that the sum rules
entering inequality (14) are exhausted by a single excitation, multi-particle
states playing a negligeable role in a dilute gas. Of course in a strongly
interacting system, such as liquid 4 He, multiparticle excitations are much
more important and the bound (17) turns out to be significantly higher
than the lowest excitation energy coo(q).
It is useful to compare the Goldstone-type inequality (17) with another,
also rigorous, upper bound, which is derivable from (14) by choosing

A = q • j , = [H, p j = ^ J2^ + 2<»' k )4 +q «k (19)

k

and B = p q . In (19), j q is the usual current operator. The resulting bound

then coincides with the ratio between the cubic energy weighted and the
energy weighted sum rules relative to the density operator p q [16,17]
([[p_q, # ] , [ # , [ # , pq]]])

The explicit calculation of the triple commutator yields (we choose q

along the z-axis) [16, 17]
(21)

i2 ni. int. n

m m
where (EK) is the ground state kinetic energy and g(s) is the pair corre-
lation function. The bound (20) then becomes:

< (f^)2 + ^(EK) + ?- [ds(l cosqz)g(s)W

- 2
zV(s) (23)
2m m mJ
and exhibits a quadratic behaviour in q as q —> 0. It is worth noting that
in the limit of a dilute Bose gas ((EK) = 0,g(s) = 1) both bounds (23)
and (17) coincide with the Bogoliubov dispersion relation (18).
The fact that it is possible to prove, via (23), the existence of gapless ex-
citations without assuming the existence of a broken symmetry is directly
connected (see Eq. (20)) with the conservation of the total current j q= o,
 Sum Rules and Bose-Einstein Condensation 93
holding in translationally invariant systems. This discussion also suggests
that Goldstone-type inequalities of form (17) are particularly useful in
sytems where the current is not conserved and where consequently only
the existence of a spontaneously broken symmetry can ensure in a simple
way the occurrence of gapless excitations. This is the case, for example,
of spin excitations in magnetic systems (spin current is not conserved) or
Bose systems in random external potentials.
Let us discuss, for example, the effects of an external potential of the
form (we take [/_k = U±)

on the sum rules discussed above. A first important result is that the
Goldstone-type upper bound (17) is not directly affected by the external
field because of the exact commutation property

(25)

Vice versa, the cubic energy weighted sum rule for the density operator
pq obtains an extra contribution from the external force given by

(26)

At small q the new term provides the leading contribution to the triple
commutator (21). Since the quantity ([p_q, [H,pq]]) is not changed by
the external force, being still given by the /-sum rule (7), the upper
bound (20) no longer vanishes with q. This different behaviour is due
to the fact that the current jq=o is not conserved in the presence of
the external field (24). This result also reveals that the relationship
jq = qy/Nno/2m(aq — a_q) between the current and the gradient of the
phase operator, holding in a translationally invariant dilute Bose gas, is
no longer valid in the presence of an external potential and gives rise
to a normal (non-superfluid) component of the density of the system at
zero temperature. This is discussed, for example, in Huang's article in
this volume.
94 S. Stringari
3 The Half Diagonal Two-body Density Matrix
In this section we discuss another sum rule, also sensitive to the presence
of Bose-Einstein condensation, given by [18]
p +co i
a
/ 1 _ -B -q W> (27)
J—
where si is the spectral function already introduced in Section 1 with
the choice A* = aq + a!q and £ = pq.
In contrast to (1), which describes the commutation relation between
the density and the phase operators, this sum rule cannot be expressed in
terms of a commutator. Physically it accounts for the coupling between
the density of the system and the modulus of the condensate, proportional
to aq + ai q . An interesting feature of this sum rule is that it fixes the long
range behaviour of the half diagonal two-body density matrix
P(2)(ri,r2;ri,r2) = ( ^ ( r O ^ m ^ m ^ ) ) . (28)
The occurrence of BEC is in fact not only relevant for the long-range
behaviour of the one-body density matrix
(29)
fixed by the condition
\imp(1\r,r') = pn0, (30)
r'->oo
but also for the one of the two-body matrix (28). The long-range order
(LRO) in the two-body density matrix (28) is naturally defined by [19]
lim p (2) (ri,r 2 ;ri,r 2 ) = n0p2(l + Fx(\ n - r 2 |)) (31)

and is characterized by the condensate fraction no as well as by the func-

tion F\(r). The properties of this function have been recently investigated
in Refs. [18, 19]. The link between the sum rule (27) and the long-range
behaviour of p(2) is fixed by the relation
(pq(aq + al q )) = V ^ ( l + 2Fx(q))9 (32)
where F\(q) is the Fourier transform of F\(r). At low q the behaviour
of this function is fixed by the density dependence of the condensate
according to [18]
, - 1 + 2*1(3) 1 d(nop) m ,
hm = — . (33)
g^o q 2nomc dp
The occurrence of LRO in the two-body density matrix implies the
 Sum Rules and Bose-Einstein Condensation 95
existence of a non-trivial relation for the chemical potential that can
be obtained starting from the following expression holding in systems
exhibiting Bose-Einstein condensation:
(
f^ (34)
Result (34) for the chemical potential follows from the property that the
p = 0 state in a Bose superfluid plays the role of a reservoir and that
consequently, adding (or destroying) a particle in this state yields the
equilibrium state relative to the N + 1 (N — 1) system. Starting from (34)
one easily derives the equation:
,
(ajao) p
Using (1) and (32), one then finds the exact non-trivial relationship [18]

= pJ drV(r)(l+F (r))
1 (36)

relating the LRO function F\(r) to the chemical potential of the system.
The result in (36) holds for any system exhibiting Bose condensation and
interacting with central potentials.
The LRO function F\(q) plays a crucial role in the coupling between
the density

and particle
\P) - ^ - , 1 0 ) (38,

states that provide natural approximations to the elementary excitations

of a Bose system. The expression in (37) is the famous Bijl-Feynman
approximation [20]. Both the density and particle pictures coincide with
the exact eigenstates in the limit of a dilute Bose gas. For a strongly
correlated liquid, they provide only an approximate description. The
coupling between the two states is different from zero because of the
occurrence of the Bose-Einstein condensation and is given by [1]:

(39)

The coupling turns out to be complete when q -• 0 (in fact in this limit
one has S(q) = q/2mc,n(q) = nomc/2q and F\(q) = —^), showing in
96 S. Stringari
an explicit way that in the hydrodynamic limit the density and particle
picture of elementary excitations of a Bose condensed system coincide.
The coupling between the density and particle pictures is at the basis
of fundamental properties exhibited by Bose superfluids (for a recent
exhaustive discussion see Ref. [21]).
It is finally interesting to compare the average excitations energies of
the states (37) and (38). Both energies provide a rigorous upper bound
to the energy of the lowest excited state of the system. The energy of the
Feynman state (37) is given by the famous result [20]

and is expressed in terms of static structure function. It is well known

that in liquid 4 He, (40) provides the exact dispersion relation in the low q
phonon regime, while it gives only a poor description at higher momenta.
The energy of the particle state (38) takes instead the form [1]

where fi is the chemical potential and the quantity W(q) is given by

W(q) = J drtdr^^ruOir^e-^-^Vin), (42)

with p( 2) (ri,0;r;,0) defined in (28).

It is also useful to note that, due to relations (35)—(36) for the chemical
potential, the particle energy (41) has no gap at q = 0 and vanishes
linearly with q [1].
It is possible to obtain an explicit expression for the average value of
the particle energy (41) in momentum space:

_p = ^4 - — - - ( 4 3 )

This average is sensitive to the values of eP(q) in the interval of momenta

where the quantity q2n(q) has a significant weight. In superfluid 4 He,
this corresponds to the range q — \ — 3A" 1 including the maxon and
roton region. Using the operator identity — J]p a p[^> a p] = EK + 2 F , the
average in (43) takes the form

€P=ii-
—((E K) + 2{V))9 (44)
N
where (EK) and (V) are the kinetic energy and the potential energy
 Sum Rules and Bose-Einstein Condensation 97
relative to the ground state. At zero pressure, where \i = j^((EK) + (V)),
(44) yields eP = -(V)/N = 21 ~ 22K in superfluid 4 He. It is instructive
to compare the above value with the corresponding average of the
Feynman energy:

(45)

Using microscopic estimates for S(q) and n(q), we find eF = 24 ~ 25K, a

value rather close to e>.
The fact that the energies of the density and particle states (37)—(38)
turn out to be comparable in the most relevant region, q = 1 — 3A" 1
reveals the importance of a careful microscopic investigation of the cou-
pling (induced by BEC) between the two pictures in order to obtain a
better understanding of excitations in Bose superfluids.

Acknowledgement. It is a great pleasure to thank L. Pitaevskii for a

stimulating collaboration and discussion.

References
[I] S. Stringari, Phys. Rev. B 46, 2974 (1992).
[2] D. Pines and Ph. Nozieres, The Theory of Quantum Liquids (Ben-
jamin, New York, 1966), Vol.1; Ph. Nozieres and D. Pines, The
Theory of Quantum Liquids (Addison-Wesley, 1990), Vol.11.
[3] N.N. Bogoliubov, Phys. Abh. SU 6, 1 (1962).
[4] H. Wagner, Z. Physik 195, 273 (1966).
[5] PC. Hohenberg, Phys. Rev. 158, 383 (1967).
[6] N.D. Mermin and H. Wagner, Phys. Rev. Lett. 17, 1133 (1966).
[7] N.D. Mermin, Phys. Rev. 176, 250 (1968).
[8] L. Pitaevskii and S. Stringari, J. Low Temp. Phys. 85, 377 (1991).
[9] T. Gavoret and Ph. Nozieres, Ann. Phys. (NY) 28, 349 (1964).
[10] L. Reatto and G.V. Chester, Phys. Rev. 155, 88 (1967).
[II] L. Pitaevskii and S. Stringari, Phys. Rev. B 47, 10915, (1993).
[12] S. Girvin and A. MacDonald, Phys. Rev. Lett. 58, 1252 (1987).
[13] R.B. Laughlin, Phys.Rev.Lett. 50, 1395 (1983).
[14] S.C. Zhang, Int. J. Mod. Phys. 6, 25 (1992).
[15] W. Kohn, Phys.Rev. 123, 1242 (1961).
[16] R.D. Puff, Phys. Rev. A 137, 406 (1965).
[17] D. Pines and C.-W. Woo, Phys. Rev. Lett. 24, 1044 (1970).
[18] S. Stringari, J. Low Temp. Phys. 84, 279 (1991).
98 S. Stringari
[19] M.L. Ristig and J. Clark, Phys. Rev. B 40, 4355 (1989).
[20] A. Bijl, Physica 8, 655 (1940); R.P. Feynman, Phys. Rev. 94, 262
(1954).
[21] A. Griffin, Excitations in a Bose-Condensed Liquid (Cambridge Uni-
versity Press, 1993).
 Dilute-Degenerate Gases
F. Laloe
Laboratoire de Physique de I'ENS'f
24, rue Lhomond
F75231 Paris, Cedex 05
France

Abstract

We study the properties of quantum gases which are dilute in terms of

the interactions but degenerate; this may be the case if the thermal wave-
length of the particles is much larger than the interatomic potential range.
We assume that the potential contains a short range repulsive part, which
creates strong correlations between particles that cannot be treated by or-
dinary perturbation theory, even if the region of configuration space where
they are dominant is relatively small. Our method combines two basic in-
gredients: Ursell operators, which generalize the Ursell functions that are
usually used to generate quantum cluster expansions, but are not in them-
selves symmetrized (the operators can be related to an auxiliary system of
distinguishable particles); in a distinct step, an exact treatment of parti-
cle indistinguishability is given through the inclusion of exchange cycles of
arbitrary length. For non-condensed boson systems, as well as for fermion
systems, the method provides a compact expression of the partition func-
tion that generalizes the well-known Beth-Uhlenbeck formula to arbitrary
degeneracy. For the study of the Bose-Einstein condensation, our treatment
retains more information on the short range correlations than, for instance,
the Hartree-Fock method. It also introduces a natural and explicit distinc-
tion between particles and quasiparticles; the latter may condense into a
single one-particle state, which obeys a non-linear equation somewhat sim-
ilar to the Gross-Pitaevskii equation. Condensation of pairs (bosons or
fermions) appears naturally in this formalism as a competing transition
process, as well as condensation of more complex objects (triplets, etc.),

Laboratoire de Spectroscopie Hertzienne de TENS, UA associee au CNRS 18, associe

a l'Universite Pierre et Marie Curie.

99
100 F. Laloe
without requiring any new ingredient, such as some variational wave func-
tion adequate for each channel of condensation.

1 Introduction
Recent progress in experiments looking for Bose-Einstein condensation
(BEC) in dilute systems, in spin polarized hydrogen [1, 2, 3], laser-cooled
clouds of atoms [4] and excitons[5, 6, 7], provides a natural stimulus
towards a more detailed understanding of the properties of degenerate
quantum gases. In these gases at very low temperatures, unusually large
values of the thermal wavelength XT may be obtained, and one can
reach situations where nX\ is of order one while the condition na3 <C 1
is still satisfied (n is the number density of particles and a is the range
of the interatomic potential); these systems are degenerate, but they
remain dilute in terms of the interactions. A motivation for this work is
the hope that BEC of these "dilute-degenerate gases" will be observed
experimentally and studied in detail in the near future.
A second motivation arises from the fact that dilute systems, such
as those studied in atomic physics, often provide useful model systems,
allowing calculations that give general views on physical mechanisms and
processes can be extended to denser systems. But the spirit of atomic
physics also implies a detailed treatment of two-body correlations (as
opposed to mean-field approximations); it is well-known that binary
collisions can be treated exactly, and there is no special reason to believe
that the occurrence of degeneracy will destroy this possibility. Moreover,
in the study of the Bose-Einstein and superfluid transitions, or processes
occurring in the build up of degeneracy, a microscopic approach may
be particularly useful because, after all, several basic questions in this
domain are still not answered very clearly. For instance, even at the
level of the one-body density operator, one can ask precisely how the
momentum distribution function is affected when one adds repulsive
hard cores to an ideal Bose condensed gas. By a continuity argument,
it is clear that the distribution will resemble that of the ideal gas, which
contains a delta function of momentum centred at zero velocity. But
two different scenarios are then conceivable: either the delta function
will survive the effects of the interaction, just changing its weight, or it
will acquire a finite width and become a "non-monochromatic peak".
The first scenario is commonly accepted, and indeed at zero temperature
there is a general argument by Penrose and Onsager [8] that proves its
validity; the argument basically uses only the fact that the ground state
 Dilute-Degenerate Gases 101
wave function of a system of bosons has no node and, moreover, gives a
good order of magnitude estimate of the "condensate fraction" in liquid
4
He. Nevertheless, at non-zero temperature, the argument does not apply
anymore since the N-body wave functions of the excited states oscillate
around zero, and the existence of a momentum delta function becomes
less clear. Generally speaking, in most treatments of the Bose-Einstein
condensation in dilute systems, the zero width of the peak is not proved,
but, rather, it is assumed. This is in particular true of all first order
perturbation calculations, or in the various forms of mean-field theories
where, by construction, the wave functions remain unaffected under the
effect of the interactions, and certainly do not vanish on any of the
contact hypersurfaces mentioned in footnote.f The zero width of the
peak at the origin is more an assumption than a result.^
If one now goes to the level of two-particle density operators (correla-
tions), more questions arise, maybe more important conceptually because
they are related to the exact relation between the Bose-Einstein conden-
sation and superfluidity. We know that the former occurs in ideal gases
while the latter does not, but in interacting systems are they fundamen-
tally the same phenomenon, always taking place at exactly the same
temperature? Or is it conceivable that, for some class of systems, two
distinct transitions will occur? The very notion of superfluidity implies
that there is a mobile condensate that is robust against the effects of
dissipation, and one would like to understand in detail what kind of cor-
relations, occurring presumably at the same time in the space of momenta
and coordinates of the particles, physically create this robustness.
Of course these questions have been discussed in the literature, and
there is already an impressive body of work on the properties of the
Bose-Einstein and superfluid transitions in dilute systems. This includes
various contributions made after the important ideas introduced by Bo-
goliubov [10], the method of the pseudopotentials developed by Huang,
Lee, Yang and collaborators [11, 12, 13], the "binary collision approxima-
tion" of Lee and Yang [14, 15], just to refer to some of the best-known

t Actually, reasoning in the configuration space of the system, where hard cores make all
iV-body wave functions of the system vanish on a large number of contact hypersur-
faces [11], gives the impression that the peak centred at zero velocity might indeed have
a finite width: one could expect some kind of Heisenberg relation between the distance
over which the wave function can propagate in a 3N dimension space without vanishing
(of the order of the classical mean-free-path for hard cores in the system) and the width
of the momentum peak at the origin. If that were the case, the peak would not be really
infinitely narrow, except in the limit of vanishing densities.
J For a critical discussion of the experimental evidence of a condensate fraction in
superfluid helium 4, see [9].
102 F. Laloe
contributions. In most cases the present work will, of course, not do
more than recover the same physical results from another point of view,
where two-body correlations are emphasized in more detail. In par-
ticular, in our effort to treat short range correlations as precisely as
possible, we will avoid any perturbation series in the interatomic po-
tential V, even if resummations are possible, for instance by selecting
ladder diagrams in order to reconstruct the T collision matrix from V.
Rather, we shall use cluster expansions which, as is well-known (see,
for instance, [16]), are a tool of choice for obtaining density expansions
instead of V expansions; they contain exponential expressions of the
hamiltonians which do not create any divergence problem for infinite
potentials (hard cores). For this purpose, we will make use of generalized
quantum Ursell functions, actually Ursell operators, which intrinsically
do not contain statistics (as opposed to the usual Ursell functions, which
are already symmetrized [16, 17]): their action is defined, not only in the
space of states of the real system of indistinguishable particles, but also
in the larger space of an auxiliary system of distinguishable particles. If
the system is dilute in terms of the interactions but not of statistics, it
is possible to limit the calculation to low order Ursell operators! while,
with the usual (fully symmetrized) Ursell functions, the order would tend
to diverge with increasing degeneracy.
This method means that we have to give up the formalism of sec-
ond quantization, despite its well-known usefulness, but this is the price
to pay for the introduction of interactions and statistics in completely
independent steps. It also implies that, at some point, an explicit sym-
metrization of the wave functions becomes indispensable and, moreover,
that no approximation whatsoever can be made at this step: otherwise
the possibility of treating degenerate systems would be lost. Fortunately
it turns out that the symmetrization operation can indeed be performed
exactly without great difficulties; this is done by introducing simple prod-
ucts of operators, corresponding to exchange cycles, or, more generally,
simple functions of operators that correspond to summations over the
size of cycles up to infinity. In this way, interactions and degeneracy are
treated at completely different orders - actually all orders for the latter
- and relatively simple and concise expressions of the partition function
are obtained. When applied to dilute systems (with respect to the inter-
actions), the method can be seen as complementary to the method of

t For instance, when a Bose-Einstein condensation takes place, the Ursell operators in
themselves have no singular variation.
 Dilute-Degenerate Gases 103
Belyaev for bosons [18] and that of Galitskii for fermions [19], extended
to non-zero temperatures.
The study of the Bose-Einstein condensation process itself in its various
forms (condensation of particles, of pairs, etc.) is delicate, because the
phase transition introduces a sensitivity to terms that would otherwise
remain small corrections. We propose to include them within a self
consistent equation for the "dressed Ursell operator". This leads to
results that remain mathematically simple and physically satisfying; for
instance, the repulsive part of the interactions tends to increase the value
of the chemical potential at which the condensation takes place; more
generally one determines qualitative predictions on the effects of the
repulsive or attractive part of the potential on the critical density. The
modification of the Ursell operator is somewhat similar to a Hartree-
Fock theory, but it is free of divergences for hard core potentials since it
incorporates short range correlations; it also includes some of the physics
contained in the non-linear Schrodinger equation [20], for instance the
modification of the condensate wave function under the effects of the
interactions. Finally, we will also see how condensation of pairs, triplets,
etc., appears naturally in this point of view. The present text does not give
the details of the calculations, which will be published elsewhere [21]; it
just attempts to summarize the physical ideas and the main results.

2 Exchange Effects and Products of Operators

The purpose of this section is to show how the operation of symmetriza-
tion introduced by quantum statistics may be performed easily by the
introduction of products of operators or of simple functions of operators.
This mechanism will be basic for all calculations in the following sections,
and will allow us to obtain partition functions as well as reduced density
matrices. As an illustration, at the end of this section we treat the ideal
gas by this method and recover known results.

2.1 Two-Particle System

In the study of the effects of exchange in atomic collisions, it is often
found that particle indistinguishability effects introduce, in the density
operator of an atom after collision, the product of the density operators
initially associated with the internal states of the atoms; examples are
spin exchange collisions [22], metastability exchange collisions [23] or
identical spin-rotation effects [24] which are the origin of nuclear spin
104 F. Laloe
waves occurring in spin polarized gases. Here we will see how products of
one body operators also occur in the evaluation of the partition function
for identical atoms and, after summation over the size of exchange cycles,
simple functions of these density operators. Our first example deals with
the simplest possible case: a two-atom system which we will define by a
one-particle density operator pF, describing external as well as internal
(spin) variables; the notation " F " is for "free", for reasons that will be
explained shortly.
Assume for a moment that the two atoms are distinguishable and that
the system is described by:

p(l,2) = p F (l)®pF (2). (1)

If one is interested in the reduced density operator pi describing one

atom only, one takes the partial trace:

p/(l) = Tr{p(l,2)} (2)

and one trivially obtains:

Pi = PF- (3)
Now, if the two particles are identical, equation (1) must be replaced
by:
p(l,2) ex [1 + rjPex] p F (l) <g> pF(2) [1 + rjPex], (4)

where rj = ±1 (bosons or fermions) and Pex is the exchange operator (we

are not interested here in the normalization factor of p and do not write
it explicitly). A simple calculation shows that the one-body operator, still
defined by (2), now has a slightly more complicated expression than (3):

pi oc pF + r\pF. (5)

The reason is that the two terms that are linear in the exchange operator
turn out to take the simple form of squares of pp. If, for simplicity, we
assume that the particles do not have internal states, pF can be defined
through its Wigner transform pj?(r,p), and it easy to show that the
Wigner transform of relation (4) is

u Pi> r2, P2) = direct term + rj x exchange term, (6)

where the direct term is simply the product:

pJf(r1)Pl )xpjf(r2,p2), (7)

 Dilute-Degenerate Gases 105
while the exchange term has the more complicated form, which is the
integral over r and k of the expression:

4 J.^-»') (8)
The former has no limitation concerning the distance at which the two
particles can be found, while the integral over k in the latter puts a
limit on this distance (of the order of the Planck constant divided by
the spread over relative momenta of the two particles); in other words
the exchange term has a range for the relative position which is of the
order of the De Broglie wavelength. This shows that the interpretation
of pr is simple: while pj gives the "full" one particle density operator,
defined independently of the position of the second particle, the operator
pr characterizes a "free" particle when it is far from the other (too far
away for exchange effects to play a role).
The calculations which lead to equation (5) also apply when the
particles have any kind of internal state. For instance one can assume
that the particles have (21 + 1) spin states, and that they are unpolarized
(equal probability to be in any of these states). It is then easy to see that
the exchange term is reduced by a factor (2/ + 1), which corresponds to the
reduction of effects of exchange when internal states of the particles can
be used to "tag" them. As expected, the factor disappears if all particles
are in the same spin state: all particles are then fully indistinguishable.

2.2 Many Particles

The same physical ideas are true for a many particle system. One may
consider the expression:

P/(l) = [ZFN]~\ Tr ( S
pF(l) <g> pF(2) <g>... 0 pF(N)S ) (9)
2,3,...,iV ^ J± si J

where the symmetrization operator S applies for bosons, the antisym-

metrization operator A for fermions and where Z^ is defined by:

ZN = Tr < PF(1) ® PF(2) ® ...pF(N) > (10)

106 F. Laloe
The operator p\ is obtained from pF after two successive, opposite,
operations:! first one goes from one particle to N by a tensor product,
secondly one comes back to one particle by a partial trace; but in between
symmetrization (or antisymmetrization) has been applied so that pj is
not simply equal to pp. The operators S and A can be expressed as a
sum of all permutation operators of the numbered particles:

where N is the number of particles and ea is the parity ±1 of the permu-

tations P a . A calculation of expression (10) can be made by expressing
every permutation as a product of cycles Q offcparticles:
i) x C2{ij')C2{n q)C2{sJ\ xC3( )..., (12)
mi factors m2 factors

with m\ cycles of length one (identity operators acting on the particles

that do not change place), m2 cycles of length two (transpositions of two
particles), m^ cycles of three particles, etc. Here Ck(i,j,k,..) denotes the
cycle where particle i replaces particle j \ particle j replaces particlefc,etc.
One obviously has:

Y^kmk=N. (13)
k

The trace over the N particles in (9) is taken in a space that is simply the
tensor product of N single particle spaces of state, so that every cycle of
lengthfcin (12) corresponds to a trace taken in a different subspace which
introduces a separate factor FV Because the numbering of the particles
does not affect the value of F^ (it just changes the names of dummy
variables), we can for convenience renumber the relevant particles from
1 tofc.The effect of Q is then to move particle 1 into the place initially
occupied by particle 2, particle 2 into the place occupied by particle 3,
and so on, until one comes back to the place of particle 1. Introducing
a complete set of states {| 1 : cpn >} in the space of states^ of particle 1,
t In (10), one can suppress one of the operators S (or A) inside the braces by using
a circular permutation of the operators under the trace, combined with the relation
S2 = S. In (9) the trace is only a partial trace so that the argument does not apply
anymore; but, because all one-particle operators inside the brace are the same, their
product commutes with 5 (or A), so that the same simplification can be made.
J If the particles have internal states, the index n symbolizes at the same time the orbital
quantum numbers as well as those characterizing the internal state. For instance, if the
particles have spin /, a summation such as J^ n contains in fact two summations, one
over orbital quantum numbers, and a second over (2/ + 1) spin states.
 Dilute-Degenerate Gases 107

{| 2 : cpn>} for particle 2, etc., one can then write:

< 1 : (
P n i \ P F ( 1 ) I 1 : <Pn2 > < 2 : cpn2 \ pF(2) \ 2 : c p m >

....X < k : c p n k \ pF(k) \ k : q > n i > = Tr {[pF]k}

(14)
The contribution of the cycle is therefore merely the trace of the fcth
power of the operator pF. Consequently, a given permutation Pa which
corresponds to a given set of values of the m& introduces into the
summation the following term:

At this point, if we call Yl{mk\ a summation over all possible ways to

break the number N into m^ according to (13), and c\m^\ the number
of permutations P a which correspond to any particular decomposition
of this type, we obtain:

* {mk} k

This formula uses the fact that cycles of an even number of particles
have parity —1 so that the parity of a cycle of length k is (— l)l+k. For
instance, if m\ = N (all the other nik being equal to zero), there is only
one permutation (the identity) that fits into this decomposition, and c is
equal to one. The other extreme happens when mN = 1 (all the other
m/c being equal to zero): one is then dealing with all cycles of maximum
length N, and it is easy to see that their number is (N— 1). More generally,
if we choose any combination of m^ obeying (13), we can distribute in
AM different ways the N particles into the N places contained inside
cycles, but all distributions do not generate different permutations P a for
two reasons: first every group of cycles of length k can be found in m&!
different orders; second, inside each cycle, any of the k particles can be
put arbitrarily in the first position, generating k different ways to recover
the same permutation. Therefore:

The summation over the m^ is difficult when equation (13) correlates

the variations of these indices, but becomes easy if this condition is
removed. This can be done by going to the grand canonical ensemble,
and defining:
108 F. Laloe

(where z is the fugacity, which would be equal to e ^ at thermal equi-

librium) ; all operators of interest can be obtained from this generating
function. The introduction of zN amounts to multiplying in (16) pF by z,
according to (13). Then, summing over m^ introduces the product for all
values of k of the exponential of the quantities rjk~l Tr {{rjzpF)k}, that
is, the exponential of a series over k that can be summed by the function
—rj Log (1 — rjzpF). Finally one is led to the simple result for the partition
function:
z £ = exp [-rj Tr {Log [1 - rjzpF]}} . (19)
Now, if we wish to calculate operators, in the canonical ensemble we
can introduce infinitesimal variations of the operator pF defined by:
dpF=de\(p)((p\pF, (20)
where de is an arbitrarily small number. One then obtains for the
variation of Z # :

dZFN = de Tr i pF(l) ® pF(2) ® ...pF(N)J2 I » : <P)(i • 9 I \ (21)

or:f
j e LogZj^ = iVTr{| (p)((p | p,} = N(q> \ p, | q>) (22)

(all derivatives with respect to e are taken at the value e = 0). If the grand
canonical ensemble is used, the N particle density operator is defined by:

pgc = [ZgFc]"' J2zN [^Pf(l) ® PFM ® ...PF(AT)^] • (23)

A variation of PF according to (20) creates a variation of Z^. given by:

N
dZFN- (24)

If we insert (21) into this equation and compare with (23), we see that
this sum over N is nothing but the product of Z^c by the sum of the
contributions of every subspace N = 1,2,3,.... of the Fock space to the
t Here we take the convention where the operator pj is normalized to the number N of
particles (instead of one).
 Dilute-Degenerate Gases 109
average value of a "one-particle operator", as usually defined in second
quantization. But this average value can also be expressed as a function
of the one particle density operator in the grand canonical ensemble pf
(normalized to the average number of particles), and we obtain:

= Tr{pf c | q>)(q> |} = (q> \ pf \ q>). (25)

This formula shows that pf is the "operator derivative" of z £ (in (20)

we have only considered diagonal variations of pp, but a generalization
to off-diagonal variations is easy). By choosing in succession for | q>) all
eigenvectors of pF, and by taking the derivative of (19), one obtains:

1 - Y\Zpp
The two particle density operator may be obtained by a similar method
from a variation of pp of the form:

dpF = [dei \(p)((p\ +de2 | x)(x I] PF (27)

but, for brevity, we do not give the calculation here.
We can now assume that pp takes the simple form (thermal equilibrium
for an ideal gas) : |
PF = e-l*H; (28)

where Ho is some one particle hamiltonian, for instance P2/2m for a

particle in a box; by replacing in (19) the trace by a sum over d3k, and in
(26) the operator pF by its diagonal element e~^2/c2/2m, one immediately
recovers the usual formulas that are found in textbooks on statistical
mechanics. The method that we have used is more indirect than the
traditional method, but it gives a physical interpretation to the term in
[pp]k that is obtained by expanding the fraction in (26): it corresponds
to the contribution of all cyclic exchange of k particles in the system. The
validity of (26) is also more general: it is not limited to any particular
choice for pF or to thermal equilibrium ;£ moreover, exactly as in Section
2.1, it is easy to include the case where particles have internal states,
for instance spin states. The formalism gives a good idea of the general
properties of a system of identical particles when the correlations have
the minimum value that remains compatible with statistics. A way to
t The operator pf does not have to have the same trace as pi, and we do not need to
insert into (28) any additional constant to reconstruct the partition function through
(10).
t Actually the complete flexibility in the choice of pf is important for taking the operator
derivative that has provided us with the reduced density operator pf.
110 F.Laloe
describe zpF is as the operator generalization of the chemical potential
fi: while \i is a convenient (but indirect) way to characterize the number
of particles N, the operator zpF may be used to indirectly characterize
all properties of the full one particle operator pf. It is therefore not
surprising that, at thermal equilibrium, there should be a simple relation
between the chemical potential and the trace of zpF [25].

3 Equation of State of a Dilute-Degenerate Gas

3.1 General Considerations
We now assume that the gas is at thermal equilibrium. In the canonical
ensemble, the partition function ZN of the system is given by:

(29)

where S applies for bosons while A applies for fermions, and where the
operators KM are denned by:
Xi = exp-jBHo(l)
K2 =

KN =
in an obvious notation. We use cluster techniques to expand the operators
Kt into Ursell operators \J\ according to:
Kt = l/i
K2(l,2) =
K 3 (l,2,3) =
+C/2(3,l)C/i(2) +173(1,2,3).
(31)
In terms of the Ursell operators, the N particle operator KN can be
written in the form:
,.)... (32)
m D
\ i) \ / m\ factors m'2 factors

where the first summation is made on all possible ways to decompose

the number of particles as:

~~ ni (33)
 Dilute-Degenerate Gases 111
The second summation corresponds to all non-equivalent ways to dis-
tribute the N particles into the variables of the Ursell operators, symbol-
ized by dots in (32). The only difference between our definitions and those
of Section 4.2 of [16] is that we use operators instead of symmetrized
functions; these operators are defined, not only within the state space
that is appropriate for bosons or fermions, but also in the larger space
obtained by the tensor product that occurs for distinguishable particles.
Hence the need for an explicit inclusion of S or A in (29).
As in the preceding section we decompose these operators into a sum
of permutations P a that, in turn, we decompose into independent cycles
of particles C:

= ^T E E fiQQOdQ x>T C2(,-

(m*} W mi factors m2 factors
XC3( )... (34)
where the first summation is similar to that of (32), while the second
corresponds to all non-equivalent ways to distribute numbers ranging
from one to N into the variables of the various Cs.
We can now insert (32) and (34) into (29) and obtain, inside a four
index summation, numbers that are traces calculated in the space of
distinguishable particles, i.e. in the ordinary tensor product of N single
particle state spaces. Inside most of these traces, factorization into simpler
terms occur, f For instance, if the term in question contains particle
number i contained at the same time in a U\ operator as well as in a
Ci, that particle completely separates by introducing the simple number
Tr{C/i}. Or, if n particles are all in separate I/is but contained inside the
same cycle Cn (or conversely inside the same Un but all in separate C\s),
this group of particles contributes by a factor Tr{[J7i]n} (or Tri..„ {[/„}).
More generally, in each term of the four index summation, particles
group into clusters (U-C clusters), which associate together all particles
that belong either to the same cycle Ck (with k > 1), or the same Ui (with
/ > 1). The general term is therefore the product of the contributions of
all the clusters that it contains:

{mi} {D} {mJ }{D'} dusters

(the number of clusters into which each term is factorized depends, in

t Here we give only the outline of the reasoning; a more detailed article will be published
[21].
112 F. Laloe
general, on this term). It is also clear that clusters differing only by the
labelling of particles which they contain give the same contribution. It
is therefore useful to work in terms of diagrams (U-C diagrams) which
emphasize the way particles are connected through exchange cycles and
Ursell operators, rather than their labelling. For instance, thefirstdia-
gram will correspond to any particle in a U\ and in a C\, and contribute
the value Tr{U\} as mentioned above; the second to Tr{[L/i]2}, the
third to Tri,2 {^2}, etc. Now, inside each term of the multiple summation,
a given diagram Fdiag may occur several times, and we call maiag the
number of times it is repeated. If, moreover, we simplify the notation
S{m,-} S{D} ^ n t o a single sum J2fP\ over the permutations, and similarly
E{m;} E{D'} i n t o E{t/}> w e obtain:

^ ^ ] - . (36)
' {Pa} {U} diag

If we call Hdiag th e number of particles that a given diagram contains, we

obviously have:
N = mdia
E gx ndia
g- (37)
diag

The next step is to remark that identical diagrams appear not only in
the same term of the double summation but also in many different terms.
Therefore, if Yl{mdia} symbolizes a summation over all possible ways to
decompose N according to (37), we can also write:

Z C
" = Jfi E ( " W >< f t [r*agnag ' (38)
' {wdiag} diag

where c {mdiag} is the numbers of terms in the double summation of (36)

that correspond to this particular decomposition of N. This number
can be obtained by distributing the N particles inside the sites of the
diagrams, which can be done in N\ different ways, and counting how
many times the same term of the double summation is obtained. For
instance, since there are m^g! ways to interchange the order of all U-C
clusters arising from the same diagram, there is a redundancy factor
equal to fldiag {mdiag 0 that comes in. Moreover, there are also equivalent
ways to put ndiag particles into one given diagram which lead to the same
U-C cluster. The detail of this counting depends on the precise way in
which the diagrams are defined, which we do not specify in detail here
(see Ref. [21]). This introduces an extra factor /diag that depends only on
 Dilute-Degenerate Gases 113
the topology and the size of the "branches" of the given diagram. The
net result is:

At this point, it is convenient to remove condition (37), which limits

the choice on the values of the possible numbers mdiag; this can be done
by using the grand canonical ensemble and its partition function:

Then a great simplification occurs because the sums over mdiag» which
are now independent, contain factorials according to (39) which intro-
duce simple exponentials; moreover the factors Q^N can be included by
multiplying every number Fdiag by e^"diae, so that:

Z gc . = n e x p [exp (jg/indiag) x £ ^ 1 . (41)

L Jdia
diag 8J

Finally, we obtain the grand potential in the form:f

- I LogZg., = - I £<*«*. x £ * . (42)

P P^s /diag
This is an exact formula, which gives the value of the pressure of the
system multiplied by its volume (and divided by the temperature), which
is an extensive quantity (as opposed to Z gc> ). It is therefore well adapted
to approximations, for instance truncations of the summation, and this
is the subject of the rest of this article.

3.2 Non-condensed Dilute Gas

We first check our result by assuming that the gas is ideal, so that all
operators Uk are zero except U\. In this case, only one class of diagram
exists, corresponding to cycles Cu of arbitrary length k. We then just need
to calculate the numerical contribution 1^ of these cycles, as well as the
counting factors /*. But this calculation has already been done in Section
2, with the replacement of pF by U\\ we have seen that F& = Tr {[Ui]k}9
while fk is equal to l//c. We therefore obtain for the grand potential of

t The logarithm of Z contains only connected U-C diagrams, a general well-known

property of diagram expansions.
114 F. Laloe
the ideal gas, by a simple change in notation in (19):

(43)

This formula is has already been discussed in Section 2.2.

What happens now if we add interactions? In this section, we limit
ourselves to the terms that correspond to second virial corrections for
the interactions, but we treat statistical effects exactly. In (42) there are
two different classes of diagrams that contain only one JJi: those where
the two particles in the U2 operator belong to two different exchange
cycles, and those for which they belong to the same cycle.
We begin with the first class of diagrams. In the simplest of them, only
two particles are involved, and both of them are contained in the C\
functions (they remain unaffected by exchange); this simply introduces
the contribution:
r n =Tr{C/ 2 (l,2)}. (44)

The next diagram in this series corresponds to three clustered particles,

two contained in the same Ui and two in one permutation operator C2.
The numerical value of this second diagram is:

r 2>1 = Tr
(45)
fTr

or:
r2>i = < 1 : q>Hl \< 2 : q>ni | l/ 2 (l,2) | 1 : <pni > | 2 : (pn2 >

<pni
(46)
that is:

Tr (47)

Similarly, one would calculate a contribution I\2 arising from the ex-
change of particles 2 and 3, and obtained by replacing U\(l) by U\(2) in
(47). More generally, when V2 clusters together k\ particles, belonging
to the same permutation cycle of length k\, with /C2 particles belonging
to another cycle of length fc2, the calculation of the effect of each of
these cycles remains very similar to that of Section 2.2: now we have
two particles that exchange separately with others, but the algebra of
 Dilute-Degenerate Gases 115

operators remains the same for each of them. We therefore obtain the
contribution:
Tr (48)

For this class of diagrams, the value of the counting factor 1// is simply
1/2; this is because, while the numbering of the particles inside (72
can be used to distinguish the various clusters corresponding to the
same diagram, (72(m,n) is not distinct from l/2(n, m). Now the final step,
according to (42), is to make a summation over all possible values of k\
and fc2 after inserting an exponential of /? times the chemical potential
multiplied by the number of particles contained in the diagram. We
denote the sum by:

(49)

The summation can be done by introducing a simple function of the l/i's

operators (a fraction), which results in a total contribution that is equal
to:
AZ d i r e c t = 1 Tr , 2) (50)

For the second class of diagrams, which we will call exchange diagrams,
the two particles contained in (72 are intermixed inside the same circular
permutation, which slightly complicates the situation. The first exchange
diagram corresponds to the two particles in question contained in the
same transposition:
^f = Tr{L/ 2(l,2)C 2(l,2)}=Tr{L/ 2(l,2)Pex}. (51)

The second will contain three particles:

r f | = Tr (52)

which is equal to:

U2(l92)

x < cpn3 I U{ I 1 : q>ni >

(53)
Now, we can use the equality:
: <pni | < 2 : q>ni \ C/ 2 (l,2) | 1 : q>ni >\ 2 : <pm >
l ;
=<l:<pni\<2:<pn2\U2(l,2)Pex\l:<Pn3>\2:<pn2>,
116 F. Laloe
which allows us to make the same summation over indices as in equation
(46) to obtain:

J {2(,)e^i()} (55)

A similar term occurs if the circular permutation C3(l,2,3) of (52) is

replaced by C 3 (l,3,2); the calculation can easily be repeated to give:

T2ef = Tr {U2(l92)PexUi(2)} . (56)

We will not give further details here, but from the preceding equations
it is not difficult to see that the generic term of this second class of
diagrams will be obtained from (48) by a simple replacement of U2 by
the product UjPex- Finally, the value of the grand potential (multiplied
by — /?), to first order in U2, will be

Log Zgx. = Log Zg°c. + AZ, (57)

with

This result has been obtained within an approximation which is basi-

cally a second virial treatment of the interactions, while it contains all
statistical corrections (we have summed the exchange cycles to all orders).
The formula therefore remains valid if the degree of degeneracy of the
gas is significant. Nevertheless, as pointed out for instance in Section 2.1
of Ref. [26] and in Ref. [27], virial series (even summed to infinity) are
no longer appropriate beyond values where the density exceeds that of
a phase transition. This is because singularities in the thermodynamic
quantities occur at a transition, in the limit of infinite systems. Therefore,
for bosons, the validity of (58) is limited to non-condensed systems; the
discussion of what happens when a Bose-Einstein condensation takes
place is postponed until Section 4.
Equation (58) can be seen as a generalization of the Beth-Uhlenbeck
formula: it reduces to it if the two denominators containing U\ functions
are replaced by one, an operation which is valid in the limit of low
densities where the chemical potential is very negative and e ^ is small.
The details of this calculation are given in [21].
 Dilute-Degenerate Gases 117
3.3 A Fictitious Imperfect Bose Gas
In the same spirit as Huang, Yang and Luttinger, who discuss the
properties of a "fictitious imperfect Bose gas" in their classic article
on the effect of hard sphere interactions on the properties of a gas
of bosons [28], we now examine the properties of a physical system,
assuming that its grand potential is exactly given by (57) and (58). In
fact, we will see in Section 4 that there is no reason to believe that any
real system will obey this equation exactly but, for future reference, it is
nevertheless useful to discuss the implications of (58) in more detail.
For a given value of the chemical potential fx, the number of particles
N of the system is given by:

N = /T 1 — Log Zg.c. (59)

What happens in an ideal gas of bosons is well-known; when fi increases,

starting from large negative values where the gas is classical, the density
of particles also increases and degeneracy builds up progressively; when
the critical value \i = 0 (in the thermodynamic limit) is reached, the
density of excited particles saturates and all additional particles have to
accumulate into a single quantum ground state. More precisely, what
happens is the following: as long as \i remains much smaller than — e\ 9
the opposite of the energy associated with the first excited statef, the
summation over the discrete energy states which occurs in the trace giving
the logarithm of Z®c can be replaced by an integral over momenta with
a good approximation. The density of particles is then proportional to
the integral over d?k of the derivative (with respect to fi) of the function
Log [1 — exp/J (fi — h2k2/2m)]. When k -> 0, this function only has a
logarithmic divergence, its derivative diverges only as k~2 at the origin,
and thus the integral over d?k remains finite (in three dimensions); this
limits the density of particles that can be obtained within this integral
approximation. But if// increases more and becomes comparable to —e\ 9
or even crosses this value and approaches zero, the discrete sum can no
longer be approximated by an integral; actually the population of the
ground state increases without limit, while that of the first excited state
does not; in other words, the ground state absorbs a significant pro-
portion of the particles. The transition between the two regimes occurs
when | /i | is of the order of e\9 an infinitesimal energy in the limit where

t We choose the origin of the energies so that the energy of the ground state is equal to
zero.
118 F. Laloe
the size L of the box containing the gas becomes very large.f When L
increases, the crossover becomes sharper and sharper and, in the ther-
modynamic limit, a singularity occurs at \i = 0, which is the origin of the
Bose-Einstein transition. Clearly, its very existence is related to the slow
divergence of a logarithmic function at the origin.
Before Bose-Einstein condensation takes place, the second term in the
right hand side of (57) remains an arbitrarily small correction if the range
of the interaction potential is reduced. Nevertheless, however small the
new term is, it completely changes the nature of the transition. From the
beginning, it is obvious that it does not contain logarithms, but fractions
which diverge more rapidly at the origin. It remains finite when \i —• 0,
because the behaviour at the origin of the integral involved in the trace
is of the type:

d3k2 < ki = 0,k 2 = 0 | U2 I ki = 0,k 2 = 0 > [k{\~2 [k2]~2, (60)

which is finite. However, this integral becomes divergent when a deriva-

tive with respect to \i in the denominators of the fractions in (58) is
taken, since variations at the origin in [fci]"4'2 [fo]"2'4 are introduced. In
other words, the second term will always end up dominant over the first
term for some value of the chemical potential jUcroSsover- For very dilute
systems, or for systems where the interaction potential is either suffi-
ciently weak or with a sufficiently small range, | /^crossover I is also very
small but, because its value does not depend on the size of the container
but only on microscopic quantities, in the thermodynamic limit it will
still be much larger than e\. The divergence in density introduced by
the interactions will always occur before that of the population of the
ground state, and therefore quench it. No Bose-Einstein transition then
remains possible: it is replaced by a continuous change of regime that
simulates the transition to some extent since, at some point, the system
starts to accumulate particles into a narrow peak of momentum centred
at the origin. The smaller the potential range, the narrower this peak,
but its width will never completely vanish: the situation is exactly that
of a "non-monochromatic peak" discussed in the introduction.

4 Bose-Einstein Condensation in a Dilute Gas

Equation (58) is an approximation which does not contain corrections
including higher powers of 1/2, as well as higher order operators (73, t/4,
t The single particle energy e\ is proportional to L~2.
 Dilute-Degenerate Gases 119
etc. We begin by a study of the effects of the terms containing binary
interactions only, i.e. Ui functions only. We will see that quadratic, cubic,
etc., terms in 1/2 can be incorporated by a "dressing" of the operators
by the interactions; this actually allows a summation of an infinite series
and restores, but for quasiparticles instead of particles, the phenomenon
of a "monochromatic Bose-Einstein condensation" that had disappeared
in the lowest order calculation of Section 3.3.

4.1 Condensation in the Presence of Interactions

In this section we study bosons only. How can we add U2 functions
into the first order U-C diagrams that we have studied and summed in
Section 3.2? If, for instance, we start from the direct diagram, where a
first 1/2 incorporates two particles belonging to two different cycles, one
can choose one particle in one of the cycles, remove its U\ and replace
it by a second I/2, grouping this particle together with another particle.
There are three ways in which this can be done:

(i) The second particle in the added U2 may belong to a new, third
exchange cycle of length ki; this will correspond to a "direct-direct
diagram". To calculate its contribution, one can use reasoning that
is similar to that of Section 3.2, basically because the two U2 func-
tions play a symmetrical role in the diagram. One therefore expects
contributions where, in (48), U\ will be replaced as follows:!

£/i(n)=> y^Tr^^q)^^-1} (61)

(the exponential in front corresponds to the number of additional
particles; there is no rj factor since we are dealing with bosons),
(ii) The new U2 may connect two particles belonging to the same cycle,
which can be obtained from the original cycle by adding k^ extra
particles between those in the new U2 (the new cycle then has a total
length /c2 +/C3); this is a mixed "direct-exchange diagram" and one
can convince oneself that its contribution will be obtained by the
substitution:

I7i(n) = > ^ e » Tr { U2(n9 q)Pex(n, q) [U^q)]"3'1} • (62)

f All the present discussion is more intuitive than rigorous; in particular the presence of
additional factors 1/2 in (61) and (62), arising from the counting factors / of the new
diagrams, is not obvious and requires a detailed study [21].
 120 F. Laloe
(iii) The new Ui may also reconnect the two initial cycles, creating a "C/2
direct loop" and generating a term that can be expressed as a trace
over two particles only (and not three as in the two preceding cases).
For the moment we leave aside this class of diagrams.
Now, if we start from the exchange diagrams, the situation is similar,
and there are also three possibilities:
(i) Adding a new cycle again gives a "direct-exchange" diagram.
(ii) One can also lengthen the initial diagram and, as in (ii) above, introduce
a new U2 that contains two particles; if these particles are not entangled
with the two particles in the other U2 (running along the permutation
cycle, one obtains the two particles of one of the Ui functions and
then the two of the other), we obtain what we will call an "exchange-
exchange diagram".
(iii) Finally, one can do as in (ii) but entangle the two couples of particles
inside the JJi functions (along the cycle one gets alternatively particles
belonging to each 1/2; this provides a diagram with a "I/2 exchange
loop" which we will also leave aside for the moment.
We therefore see that substitutions (61) and (62) again apply. If now
we group these results and sum over all values of £3, we obtain:

i Tr{uiM [1 + PeX] 1 ; / ; [ / ( 2 ) } • (63)

It is clear that the above procedure is iterative, and that it allows us
to build higher and higher order terms in the series of U-C diagrams. In
particular, when applied successively to all U\ functions that belong to
the pure exchange cycles that occur for the ideal gas, it generates from
the zero order terms the first order 1/2 terms that are contained in (58).
The ideal gas cycles have a counting factor 1// = 1/fe but, because each
of the U\ operators is modified in turn according to the above equations,
giving the same numerical result, an extra factor k arises that cancels the
first; this is why no logarithm is obtained in the first order correction.
On the whole, when (63) is iterated from zero up to infinity, one is led to
the substitution (we now use the variable 1 instead of n):

, 2 ) [l+Pex] 1 ^ }
It is useful to notice that, in this equation, the "dressed operator" JJ\ has
an implicit definition, since it also appears in the denominator of (64);
in other words, U\ is defined by a self-consistent equation.
 Dilute-Degenerate Gases 121
Now, when this substitution is done, the effects of the interactions
are contained in U\ itself and the value of the grand potential can be
obtained by the same calculation as for an ideal gas, but just replacing
U\ by U\. This gives the result|, similar to (43):

LogZg., - -Tr {Log [l - e ^ l ) ] } . (65)

This equation contains a summation of diagrams to infinite order in U2S0
that it is clearly more general than (57) and (58). Nevertheless, since the
diagrams containing loops are not included, it is not exact either. The
approximation behind (65) bears some resemblance with that of a mean-
field theory, but not in terms of the potential: here, Ursell operators are
used instead as the basic objects, so that the two body correlations are
included more precisely than in usual Hartree-Fock-type theories. For
instance, as in the preceding section, infinitely repulsive hard cores do
not introduce any divergence, the physical reason being that our method
incorporates the fact that the particles are prevented from coming close
to each other.
How does the Bose-Einstein condensation takes place in this scheme?
The general scenario will be similar to that of the ideal gas, but here
condensation will take place when the function inside the logarithm of
(65) diverges, that is, when the product of e^ by the largest eigenvalue
of the operator U\ becomes equal to one. If the range of the interaction
potential is small, the two operators U\ and U\ are not very different,
so that this largest eigenvalue will be one plus some small correction.
What matters is the sign of this correction, which according to (64)
depends in turn on the sign of the matrix elements of Ui. Because this
operator contains the difference between two exponentials, one involving
the negative of the sum of the kinetic energy of two particles added to
their interaction energy, the other involving the negative of the kinetic
energy only, we see that repulsive interactions will tend to give negative
values to its matrix elements, while attractive interactions will favour
positive values. We conclude that, for repulsive interactions, the Bose-
Einstein condensation phenomenon will take place when e^ > 1, that is,
for positive values of the chemical potential; for attractive interactions,
for negative values of the chemical potential. These results are physically
satisfactory and correspond to those found in the method used by Huang
and coworkers [12, 26], but here we do not approximate the potential
t It can easily be checked, for instance, that, if a lowest order approximation is made by
replacing l/i(l) by t/i(l) in the right hand side of (64), relation (58) is recovered.
122 F. Laloe
by a scattering length; in particular, no special instability is predicted if
the potential becomes attractive, while keeping a small range (or a small
scattering length) .
Another feature of our theory is that condensation takes place when
a divergence occurs in a function which depends on the operator U\(l),
which is not directly related to the one-body density operator as for
the ideal gas, but contains density corrections. The eigenstates of l/i(l)
are the energy states of the quasiparticles, which include the effects of
the interactions averaged through C/2 operators, and not those of the
particles. We therefore find a scheme that incorporates the notion of
quasiparticles, and relates real particles to quasiparticles in a quantitative
way, since through (64) it is possible to express the one-particle density
operator as a function of U\(l).

4.2 Condensation of Pairs

We now come back to the general case where particles may be fermions
or bosons; we study the effect of another class of diagrams which we
left aside in the preceding section, where the emphasis was put on the
diagrams where the t/2 functions do not connect the various cycles in
too intricate a way. Here, on the other hand, we study what happens if
we introduce the maximum connectivity possible between two different
cycles, or inside a given cycle. If, for instance, we start from the first
"direct U2 loop" diagram that we ignored in our study of the Bose-
Einstein condensation of interacting particles, we have to consider two
U2 functions connecting two pairs of particles that are both placed inside
cycles of size two, that is, transpositions. It is easy to see that this term,
which now contains a summation over two independent indices, has a
contribution:
e 4 ^Tr{[(7 2 (l,2)] 2 }. (66)

If three U2 functions connect two triplet of particles that belong to two

three-particle cycles, one obtains in the same way the term:

{[2(,)]} (67)

or, more generally, with two cycles of equal length k where all particles
are linked two by two by U2 functions:

{[2(,)]} (68)
 Dilute-Degenerate Gases 123
None of the above terms contains an n factor, even for fermions, because
two cycles of the same parity are always involved.
Now, starting from the diagram that we called the "U2 exchange loop"
in the preceding section, we can proceed in a similar way as before, but
this time we do not consider two cycles of equal length, but only one
cycle of even length. The latter is saturated at all points by U2 functions,
each taking two particles from similar places in the two different halves
of the cycle. The first term which was ignored in the preceding section,
is obtained from a four particle cycle and two U2 functions, and simple
reasoning shows that its value is:

{[2{h)ex]} (69)

(the factor rj arises because a cycle of an even number of particles is

odd). More generally, one obtains from a cycle acting on 2/c particles the
term:
^ { ^ } (70)

Obviously both kinds of terms (68) and (70) can easily be grouped by
introducing the operator U2 [1 + .rjPex], but we also note that another
modification of l ^ is necessary as well. The reason is that, if we want
to avoid a limitation of the validity of the results to lowest order in the
degree of degeneracy of the system, when calculating the contribution of
any class of diagrams we must include all those that contain additional
arbitrarily long exchange cycles. In other words, as in Sections 3.2 and
4.1, we must incorporate an arbitrarily large number of intermediate U\
functions at all places where they can be inserted into this new class
of diagrams, without changing their topology. Moreover, after inserting
these chains of JJ\ functions at any place between the JJ2 functions, we
must take the sum over the lengths of the chains. But we already know
from Section 3.2 what the effect of this process is: the summation merely
introduces operators [l — nQ^V\\~ • More precisely, since in the present
case this operation has to be carried out twice, once for each particle
contained in the U2 functions, two fractions are introduced. Therefore,
to incorporate all diagrams that we have discussed, we have to replace
I/2 in (68) by the new operator U2 defined as:

Finally we have to sum over all values of k. The situation, then, becomes
reminiscent of that for the ideal gas: because all pairs of particles
124 F. Laloe
appearing in the U2 functions play the same role, it turns out that the
counting factor of the diagrams is simply given by 1// = 2k. This implies
that one again obtains a logarithmic function,! so that the term to be
added to the logarithm of z is:
-Tr{Log[l-e 2 ^l/ 2 (l,2)]}. (72)

The physical interpretation of this new contribution to the grand

potential is simple: we are now dealing with the condensation of pairs
of particles, which is indeed possible in principle for both fermions and
bosons [29]. The mechanism is similar to that studied in Section 4.1, and
what determines the possible existence of a phase transition is the largest
eigenvalue of the operator U2. If it is positive, which requires some
attraction between the particles, a Bose-Einstein condensation channel
becomes available for a critical value of the chemical potential which is
proportional to the logarithm of this eigenvalue; if it is negative, equation
(72) becomes similar to that describng the grand potential of an ideal gas
of fermions, and no new transition occurs. In what follows, we assume
that the eigenvalue is indeed positive in order to discuss the influence
of the new channel of condensation. If this is the case, there is still a
difference with the usual condensation of single particles, since U2 acts
at the same time on two kinds of variables, those associated with the
centre of mass of the two interacting particles, and those associated with
the relative motion. The effect on the centre of mass is almost the same
as the effect of U\ on one particle; the only difference is that the mass is
twice as large, decreasing the de Broglie wavelength. On the other hand,
the effect on the variables of relative motion is significantly distinct, since
the difference which appears in U2 gives a finite (microscopic) range to
the action of the operator; the interaction potential may or may not
create bound states.
The two fractions containing the operators (7i(l,2) in (71) play a role
which is similar those occurring in (58): if the gas is non-degenerate,
they introduce small corrections, but if this is not the case, they can
completely change the properties of 1/2. They also have a simple physical
interpretation: using (26), which gives the one particle density operator
Pi to zero order in the interaction (with the notation JJ\ in place of pF\
one easily sees that they are equal to [1 +r\pi]. The latter is nothing
t In fact, we have already incorporated the terms k = 0 in the summation leading to
(65) so that, to be precise, one should not merely add (71) to that value of the grand
potential, but also remove afterwards the term that has been double counted, which is
in fact simply given by (58).
 Dilute-Degenerate Gases 125
but the usual statistical factor that appears in the Uehling-Uhlenbeck
transport equation for instance, or in the Landau theory of Fermi liquids
(more precisely the factors are an operator generalization, because U\
and I/2 do not commute in general). For highly degenerate fermions,
for instance, these statistical factors remove all the contributions to 1/2
which correspond to matrix elements of Ui between one particle states
below the Fermi level, and only those close to the surface of the Fermi
sphere contribute in practice.
Will the pair Bose-Einstein condensation really take place, or will it
be quenched by other processes? We begin the discussion with bosons.
If the potential is only weakly attractive, or if it has a very small range,
the largest eigenvalue of 1/2 will be smaller than that of \J\ (which
is close to one). Then the Bose-Einstein condensation of particles (or,
more precisely, of quasiparticles) will take place before the condensation
of pairs, that is, at a smaller value of \i. In this case, the critical
value of \i at which pair condensation occurs will never be attained
(with repulsive interactions, the chemical potential continues to increase
after Bose-Einstein condensation has occurred, but here we assume
attractive interaction f); the presence of pairs will then just bring a
small correction to the calculations of the preceding section. But, if the
interaction potential becomes more and more attractive, able to sustain
one or more bound states with negative energies, large eigenvalues of l/ 2
appear, and then it becomes perfectly possible that the condensation of
pairs will correspond to a smaller value of the chemical potential than
that of single particles. In this case, the condensation of pairs occurs
first): and the condensation of single particles never takes place. Our
model gives a physical picture based on a competition between these two
channels of condensation, among which the channel which dominates is
that for which the critical density (of particles, or of molecules) is reached
first.
For fermions, no condensation of particles can take place, so that
no process will stop the chemical potential from rising to positive val-
ues; the limit is the Fermi energy at low temperatures. Now, what we
have to compare is the largest eigenvalue of 1/2 with the exponential
Q-2PH s i n c e a t j o w temperatures the chemical potential remains roughly

f Also, for attractive interactions, the critical values of ft are negative, so that we do not
have to worry about singularities in (71).
J Of course, it is perfectly possible that another condensation process involving more than
two particles will, in turn, take over and quench the condensation of pairs; this possibity
includes ordinary condensation into a liquid or even a solid phase. For a brief discussion
of other condensation processes see next section.
126 F. Laloe
constant and positive, when /? increases the exponential varies rapidly
and always becomes smaller, at some point, than the eigenvalue. A sin-
gularity will then necessarily occur: even weakly attractive potentials
will allow a Bose-Einstein condensation of pairs. We therefore recover
a well-known property of fermions: at sufficiently low temperature,
arbitrarily small attractions destabilize the Fermi sphere, an essential
ingredient of the BCS theory [30]. If the temperature is not very low,
and if there is no well-formed Fermi sphere with a sharp limit (the
usual assumption in the BCS theory), formulas (71) and (72) still apply
without any particular modification; the fractions in (71) automatically
take care of the cross-over between the degenerate and classical regime,
as noted above. We note in passing that, instead of the potential or
of the scattering length which often appears in the theory of pairing
in Fermi systems [31], here the relevant matrix element is that of an
operator which depends both directly on the temperature through the
definition of I/2 and indirectly through the factors [1 — p/] in the def-
inition of U2', moreover it contains all partial scattering waves and
not only the s-wave. In addition, one may use (71) and (72) to dis-
cuss the cross-over between the two regimes where condensation takes
place, either for strongly bound binary molecules, or weakly bound BCS
pairs [32, 33, 34].f
A final remark is that what we have discussed, for simplicity, is the
condensation of bare, non-interacting, pairs. But by introducing the
contribution of U3 operators (and by replacing U\ by U\), it would be
possible to extend the theory in the same spirit as that of Section 4.1 and
define a dressing of the operator £/2 under the effect of the interactions
of pairs with a third partner which would lead to the condensation of
"quasipairs of particles".

43 Other Phase Transitions

A generalization of the preceding considerations and of (71) leads to
condensation of triplets; the contribution to the partition function is
similar to (72), except that U3 replaces U2 in a trace over three particles,
and more complicated exchange operators are present. More generally,
channels corresponding to the condensation of groups of k particles can
be written formally in terms of operators acting in the space of k particles.

f For more specific discussions including films, see Refs. [35] and [36].
 Dilute-Degenerate Gases 127
In nuclei, four fermion condensate has been discussed in the context of
the condensation of a particles [37].
But even all these condensation channels do not necessarily exhaust all
the diagrams contained in the partition function. If all sorts of diagrams
can indeed be grouped, as above, in condensation channels where they
generate logarithms, then all the physics is contained in competing Bose-
Einstein condensation processes of composite particles. If some other
diagrams cannot be regrouped in this way, then there will be a channel
for formation of a "non-monochromatic momentum peak", as discussed
in Section 3.3, in competition with the ordinary ("monochromatic")
condensation channels. Of course there is no proof at this stage that such
processes can exist, for a given class of interaction potentials, so that this
remark remains speculative.
Finally, we note that the self-consistent equation (64) which defines the
U operator of the quasiparticles is non-linear, and may contain insta-
bilities if the coupling is strong enough and the temperature sufficiently
low. It may happen that a spontaneous symmetry breaking process, anal-
ogous to ferromagnetism, will allow one to obtain lower values of the
grand potential than the values obtained without this process; this would
correspond to a superfluid instability.

5 Conclusion
In this article, we have used another version of perturbation theory, not
in terms of the interaction potential, but in terms of Ursell operators
which give rise to various contributions in the form of U-C diagrams.
This leads to the exact expression (42) for the grand potential of the
system, expressing it as a sum of various terms arising from these dia-
grams. Each of these terms is obtained as an integral (a trace) over a finite
number of variables. The expression is valid for dilute or dense systems
such as liquids or solids, but for gases it becomes simpler because it can
be truncated conveniently. A feature of the formalism is that the size
of exchange cycles appears explicitly, which provides a natural connec-
tion with the numerical work of Ceperley and Pollock [38], emphasizing
the role of cyclic exchange of all orders in Bose-Einstein condensa-
tion and superfluidity. In our case, the summations over the lengths
of cycles are straightforward, introducing simple functions (fractions)
of the operators. Several mechanisms can lead to a divergence of the
size of the exchange cycles in the relevant diagrams; each of them in-
troduces a separate channel for Bose-Einstein condensation. In other
128 F. Laloe
words, the formalism provides information on various physical pro-
cesses which occur in parallel, such as the condensation of particles
(or quasiparticles), that of pairs, or the contribution of larger clus-
ters, without requiring any ad hoc additional assumption. Refining
the results on a given condensation channel by including interactions
through a renormalization of the operators is also possible, at least
formally, by the introduction of integrals of finite dimension. For in-
stance, if necessary, one could improve the theory of condensation of
pairs by replacing, in the statistical factors appearing in the definition
of I/2, the bare operators U\ by dressed operators U. The method
gives simple formulas that remain valid at both low and high tem-
peratures (degenerate or classical systems); in particular, it includes
automatically all s,p,d,... collision channels. It could be adapted rel-
atively easily to the inclusion of external potentials (inhomogeneous
BEC), since this additional interaction is entirely contained in U\.
We plan to use our method for calculating condensate fractions in
this kind of problem. A more systematic comparison with the re-
sults of the literature is also desirable: energy of the ground state
for bosons [39, 18] and fermions [40, 19], as well as pseudopotential
calculations of the magnetic susceptibility for fermions. Finally, the
study of superfluid gases, especially in the case considered by Siggia and
Ruckenstein where there are two internal states which may condense
simultaneously [41], is another natural domain of application of our
formalism.

Acknowledgment. The author has benefited from very helpful discus-

sions with Edouard Brezin, Roger Balian and Philippe Nozieres. More
recently he has also had stimulating discussions with A. Stringari, K.
Huang and W. MuUin and is very grateful to Peter Griiter and to W.
Mullin for a careful reading of the manuscript and suggesting many
improvements.

References
[1] T.J. Greytak, this volume.
[2] I. Silvera, this volume.
[3] T.J. Greytak and D. Kleppner, in New Trends in Atomic Physics
Vol. II, G. Grynberg and R. Stora, eds. (North-Holland, N.Y, 1984),
p.1127.
[4] Y. Castin, J. Dalibard, and C. Cohen-Tannoudji, this volume.
 Dilute-Degenerate Gases 129
[5] J.P. Wolfe, Jia Ling Lin, and D.W. Snoke, this volume.
[6] A. Mysyrowicz, this volume.
[7] L.V. Keldysh, this volume.
[8] O. Penrose and L. Onsager, Phys. Rev. 104, 576 (1956).
[9] RE. Sokol, this volume.
[10] N. Bogoliubov, Journal of Physics USSR 11, 23 (1947).
[11] Kerson Huang, Statistical Mechanics, (Wiley, N.Y., 1963) Section
13.3; sec. ed. (1987), Section 10.5.
[12] K. Huang and C.N. Yang, Phys. Rev. 105, 767 (1957); K. Huang,
C.N. Yang and J.M. Luttinger, Phys. Rev. 105, 776 (1956); T.D. Lee,
K. Huang and C.N. Yang, Phys. Rev. 106, 1135 (1957).
[13] T.D. Lee and C.N. Yang, Phys. Rev. 112, 1419 (1958); 113, 1406
(1959).
[14] T.D. Lee and C.N. Yang, Phys. Rev. 113, 1165 (1959); 116, 25
(1959); 117, 12, 22 and 897 (1960).
[15] R.K. Pathria, Statistical Mechanics, (Pergamon, Oxford, 1972), Sec-
tion 9.7.
[16] K. Huang, loc. cit, Chapter 14; second edition (1987), Chapter 10.
[17] R.K. Pathria, Ref.[15], Section 9.6.
[18] S.T. Belyaev, Soviet Phys. JETP 34, 289 (1958).
[19] V.M. Galitskii, Soviet Phys. JETP 34, 104 (1958).
[20] E.P. Gross, Journ. Math. Phys. 4, 195 (1963).
[21] F. Laloe, J. Physique, in preparation
[22] M. Pinard and F. Laloe, J. Physique 41, 769 (1980).
[23] M. Pinard and F. Laloe, J. Physique 41, 799 (1980).
[24] C. Lhuillier et F. Laloe, J. Physique 43, 197 and 225 (1982).
[25] PJ. Nacher, G. Tastevin and F. Laloe, J. Physique I 1, 181 (1991).
[26] K. Huang, in Studies in Statistical Mechanics, Vol. II, J. De Boer
and G.E. Uhlenbeck, eds. (North-Holland, Amsterdam, 1964).
[27] C.N. Yang and T.D. Lee, Phys. Rev. 87, 404 (1952).
[28] K. Huang, C.N. Yang and J.M. Luttinger, Phys. Rev. 105, 776
(1957), Section 6.
[29] C.N. Yang, Rev. Mod. Phys. 34, 694 (1962).
[30] J. Bardeen, L.N. Cooper and J.R. Schrieffer, Phys. Rev. 108, 1175
(1957).
[31] E.M. Lifshitz and L.R Pitaevskii, Statistical Physics, Vol. II (Perga-
mon, Oxford, 1980), Sections 39-40.
130 F. Laloe
[32] A.J. Leggett, in Modern Trends in Condensed Matter Physics, Pro-
ceedings 16th Karpacz Winter School, A. Pekalski and J. Przystawa,
eds. (Springer, Berlin, 1980); AJ. Leggett, J. Physique coll. C7, 19
(1980).
[33] See review by M. Randeria, this volume.
[34] P. Nozieres and S. Schmitt-Rink, J. Low Temp. Phys. 59, 195 (1985).
[35] D.M. Eagles, Phys. Rev. 186, 456 (1969).
[36] K. Miyake, Progr. Theor. Phys. 69, 1794 (1983).
[37] M. Apostol, Phys. Lett. 110A, 141 (1985).
[38] D.M. Ceperley and E.L. Pollock, Phys. Rev. Lett. 56, 351 (1986);
Phys. Rev. B36, 8343 (1987); E.L. Pollock and D.M. Ceperley, Phys.
Rev. B 39, 2084 (1989).
[39] Lifshitz and Pitaevskii, Ref.[31], Section 25.
[40] Lifshitz and Pitaevskii, Ref.[31], Section 6.
[41] E.D. Siggia and A.E. Ruckenstein, Phys.Rev.Lett. 44, 1423 (1980);
J. Physique 40, C7 15 (1980); Phys. Rev. B 23, 3580 (1981).
Prospects for Bose-Einstein Condensation
in Magnetically Trapped Atomic
Hydrogen
Thomas J. Greytak
Physics Department
Massachusetts Institute of Technology
Cambridge, MA 02139
USA

Abstract
Atomic hydrogen would be an ideal substance in which to observe Bose-
Einstein condensation. Recombination of the atoms into molecules can be
slowed dramatically by polarizing the electronic spins in a high magnetic
field at low temperature, but achieving the necessary density in samples
confined by material walls has not been possible. Evaporative cooling of
magnetically trapped spin-polarized hydrogen is a more promising approach.
It has already produced temperatures as low as 100 /iK (well below the
laser cooling limit) at a density of 8 x 1013 cm" 3 . At this density, the
transition temperature is 30 //K. Reaching the transition by evaporative
cooling appears to be possible. Laser techniques are being developed to
detect the transition and to study the properties of the resulting gas.

1 Introduction

The first discussion of atomic hydrogen as a candidate for Bose-Einstein

condensation was published by Hecht [1] in 1959. He used the quan-
tum theory of corresponding states to conclude that atomic hydrogen,
prevented from recombining into molecules by a strong magnetic field,
would remain a gas down to the absolute zero of temperature. He gave
the expression for the transition temperature in such a gas in the absence
of interactions,

h2 / n y/3
2nkBM V2.612/
/ n \2/3
2O (1)
U.965xlO J '

131
132 T. J. Greytak
and pointed out that the gas would be a mixture of two of the hyper-
fine states of hydrogen. The idea was raised again by Stwalley and
Nosanow [2] in 1976. By this time more complete ground state energy
calculations had been done confirming the absence of a liquid phase
for atomic hydrogen. They pointed out that the presence of the in-
teraction between the atoms could lead to the Bose condensed state
being a superfluid as well. The weakness of the interaction and the
ability to change the density would allow a detailed comparison be-
tween theory and experiment. Although Stwalley and Nosanow saw
BEC as the ultimate long term goal of spin-polarized atomic hydrogen
research, they also expected that "these systems should yield exciting
and fundamental new information about the properties of quantum
systems".
The ultimate goal of BEC in hydrogen has yet to be realized, though
we seem to be quite close at the current time. On the other hand,
since spin-polarized atomic hydrogen was first created in the labora-
tory by Silvera and Walraven [3], the prediction of exciting results
from the "new" quantum system has certainly been fulfilled. Once
the atomic hydrogen has been electron spin polarized, it spontaneously
develops nuclear polarization [4, 5], The doubly polarized gas was
found to support sharply defined nuclear spin waves [6-8], even though
the mean separation between the atoms is much larger than their De-
Broglie wave length. The doubly polarized gas, when compressed to
higher densities [9, 10], became a model system for the study of three-
body recombination processes [11]. Cryogenic atomic hydrogen masers
[12-14] offer the the possibility of significant increases in frequency
stability over similar room temperature devices [15]. Most recently,
the cold hydrogen atoms have been used to study atom-surface inter-
actions, in particular the sticking probability of very cold atoms on
well-understood surfaces [16-18]. These aspects of spin-polarized atomic
hydrogen have been the subject of two extensive reviews [19, 20]. In
this article I will concentrate on the possibilities for attaining BEC.
I will review some of the early prospects and explain why they did
not come to fruition. I will discuss magnetic trapping and evapora-
tive cooling of the hydrogen, which seems to be the most promising
avenue toward BEC at present. And finally, I will examine what the
condensed state might look like in the trapped gas, and how it could be
studied.
 Prospects for BEC in Magnetically Trapped Atomic Hydrogen 133

<D
C

-0.2 -

-0.4 -

0.1 0.2 0.3 0.4 0.5 0.6

Magnetic field [tesla]
Fig. 1. Hyperfine states of atomic hydrogen. j and -f indicate, respectively, the
spin of the electron and the proton. Simple fills with high field seekers are done
at fields of the order of 8 T. Magnetic trapping of low field seekers involves fields
from 1 T down to 1 m l

2 Early Approaches

Figure 1 shows the splitting of the ground state of atomic hydrogen due
to the hyperfine interaction. The two lowest energy states, labeled a and
b, are called 'high field seekers' since they are drawn into a region of high
magnetic field. Similarly, the two highest energy states, c and d, are 'low
field seekers'. All four states are created in roughly equal amounts when
molecular hydrogen is dissociated in an RF discharge. In the simplest
experiments the atomic hydrogen develops electron polarization when the
atoms impinge on a region of high magnetic field (typically the center
of a superconducting magnet at about 8 Tesla) and low temperature
(typically 300 mK obtained by a dilution refrigerator). Atoms in the c
and d states are excluded from the high field region; those in the a and b
134 T. J. Greytak

(A) Copper rod

(B) Reference capacitor
(C) Capacitive pressure transducer
(D) Liquid 4 He pool
(E) Sintered copper
(F) Level sensor
(G) Bolometer
(H) Thermometer

(a)

Fig. 2. Sample chambers for high field seekers used by the MIT group, (a) For
simple fills. The inner diameter is 1 cm.

states enter, lose their excess energy through wall collisions, and become
trapped.
It is unfortunate that Maxwell's equations do not allow a maximum in
the magnitude of the magnetic field in a source free region. Otherwise,
the high field seekers could be held in a pure magnetic trap. In practice
the atoms are usually confined radially and from below by physical walls,
as is shown in Fig. 2(a). One could say that the magnetic field acts as a
cork in an otherwise open bottle.
The electron-polarized gas of a and b atoms is not stable against
recombination. The finite admixture of the 'wrong' direction of elec-
tron spin in the a state due to hyperfine mixing allows the reactions
a + b —• ortho H2 and a + a —• para H2. The recombination of
two hydrogen atoms releases 4.476 eV of energy. Initially most of this
energy goes into rotation and vibration of the molecule, leaving less
than 1% for translational motion [11]. Since a two-body collision can-
not conserve momentum while simultaneously releasing kinetic energy,
these reactions require the presence of a third body. The third body
could be another hydrogen atom, but at the densities achieved in simple
 Prospects for BEC in Magnetically Trapped Atomic Hydrogen 135

-Glass plate

\ Z:
Pressure
transducer

Film Sample
thermometers ^/chamber
f !spacer
Indium O-ring
Silver sinter-

-Piston

(b)
Fig. 2. (b) For compression experiments. The hydrogen is compressed into the
region above the sintered silver, 90 /urn high and 0.8 cm in diameter.

fill experiments the reaction takes place while the atoms are adsorbed
on the walls of the cell with the wall acting as the third body. Re-
combination is slowed, but not eliminated, by coating the walls of the
cell with superfluid helium to minimize the van der Waals attraction
between the atoms and the surface. The result of these reactions is
to remove all of the a atoms while leaving a finite fraction of the b
atoms. The resulting gas of b atoms is nuclear as well as electron-spin-
polarized.
Although the gas of pure b atoms is stable against recombination,
nuclear relaxation can occur during collisions [4], causing a transition
to the a state. The nuclear relaxation is driven by the fluctuating mag-
136 T. J. Greytak

Table 1. Typical results for simple fills

hyperfine state b
number of atoms 10 17
density 3 x 10 16
cm~ 3
temperature 300 mK
lifetime 150 min
nc at this temperature 8 x 10 19 c m " 3
Tc at this density 1.5 m K

netic field experienced by a given atom due to the electronic magnetic

moments of other passing hydrogen atoms in the gas. The resulting a
atom promptly recombines on the surface during a collision with a b
atom. Therefore the rate at which atoms disappear from the doubly
polarized gas due to recombination is just twice the nuclear relaxation
rate. The relaxation rate is proportional to the collision rate, and thus to
the square of the density. As a consequence the characteristic decay time
of the sample is inversely proportional to the density. Typical parameters
for the doubly polarized gas which can be obtained by simple fills are
shown in Table 1.
The table shows that at these temperatures the critical density for BEC
is three orders of magnitude greater than can easily be achieved with the
current sources. Is it possible that improvements in source flux can make
up the difference? For a given flux, the density in the cell builds up until
a steady state is reached in which the recombination rate matches the
flux. Since the recombination rate is proportional to the square of the
density, the steady state density varies as the square root of the flux. To
increase the density by three orders of magnitude, the flux would have to
be increased by a factor of a million. The heat load on the refrigerator
would increase by a similar factor. This is not a feasible approach to
attaining the phase transition.
Several groups tried to reach the transition by compressing the dou-
bly polarized gas. The MIT group used a piston [9] (see Fig. 2(b)).
Other groups compressed the hydrogen inside a bubble [10, 21]. At
the critical density of 8 x 1019 cm"3 the decay time for the sam-
ple due to nuclear relaxation in the bulk gas would be about three
seconds. Careful studies, however, showed that the compressed gas
recombined more rapidly with increasing density than was expected
from relaxation. At high enough densities, the samples simply ex-
 Prospects for BEC in Magnetically Trapped Atomic Hydrogen 137

Table 2. Results obtainable by compression

hyperfine state b
number of atoms 2 x 1016
density 4.5 x 1018 cm"3
temperature 550 mK
lifetime 5 sec
nc at this temperature 2 x 1020 cm"3
Tc at this density 44 mK

ploded. It was determined that the recombination was caused by a

three-body process in the gas, predicted by Kagan, Vartanyantz and
Shlyapnikov [11], which takes place even when the three atoms are all
in the b state. Typical results that can be obtained by compression are
given in Table 2.
The Bose-Einstein transition occurs when the thermal DeBroglie wave-
length of the atoms becomes comparable to their mean separation. The
compression experiments were an attempt to approach this condition by
decreasing the separation at fixed temperature. Another approach would
be to increase the DeBroglie wavelength by cooling the atoms at fixed
density. This approach fails in the presence of surfaces due to three-body
recombination. As the temperature is lowered, the density of atoms
adsorbed on the surface increases to the point where the three-body rate
on the surface becomes prohibitive.
Two other possibilities have been suggested for approaching BEC with
high field seeking states in the presence of surfaces. Theories suggest
that very high magnetic fields will decrease the three-body recombi-
nation rate [11, 22]. However, experiments done at fields as high as
20 T showed that the three-body rate had not yet fallen below its value
at 8 T [23]. Kagan and Shlyapnikov [24] have suggested a novel ex-
periment in which a strongly inhomogeneous field is used to compress
the spin-polarized hydrogen against a small region of the surface in an
otherwise open geometry. The hope is that the heat of recombination
can be dissipated in a region of the cell well removed from the small,
high density cloud, while the lost atoms are replenished from the low
density gas filling the greater part of the cell. This experiment has
been tried [25], but the problem of sample heating has not yet been
resolved.
 138 T. J. Greytak

Source solenoid
Confined atoms Quadrupole magnet
Hyperfine resonator

,1,,,;;;;,,,;,;,,,,,,r?
Source

Lower pinch Bias Upper pinch

Transverse rf coil solenoid solenoid solenoid

Fig. 3. Magnetic trapping and cooling geometry proposed by Hess.

3 Magnetic Trapping and Evaporative Cooling

The obstacles to BEC caused by three-body recombination inspired

Harald Hess to propose an alternative approach [26]. He recognized
that magnetic field configurations are possible which possess a local
minimum in the magnitude of the field and could be used to trap the
low field seeking states of hydrogen, c and d. In addition, he pointed
out that it would be possible to cool the trapped atoms evaporatively
to temperatures well below that of the surrounding cell walls. The one
disadvantage of working with the low field seekers is that they are not
as stable against relaxation due to collisions as are the high field seekers.
The number of low field seekers per unit time converted to high field
seeking states by electronic relaxation is proportional to the square of
the gas density. The relaxation time at a density of 1014 cm" 3 is about 10
seconds. Thus experiments with magnetically trapped low field seekers
will typically be limited to modest densities and will require cooling below
50 microkelvin in order to observe BEC.
The configuration originally proposed by Hess is shown in Fig. 3. Ra-
dial magnetic confinement of the atoms is accomplished with a cylindrical
quadrupolar field. The magnitude of this field is zero on the axis and
increases linearly with increasing radius. Axial confinement is provided
by a solenoid at either end of the trap. If the field on the axis were
allowed to remain at zero, atoms passing through the zero-field region
could undergo non-adiabatic transitions to the lower hyperfine states.
To avoid this, a small uniform bias field is applied by a solenoid about
 Prospects for BEC in Magnetically Trapped Atomic Hydrogen 139

(a) Filling (b ) Spontaneous cooling

Loss by evaporation
E/atom » <E>
Causes cool
Hotest Gas cools,
atoms settles deeper
leave into trap

Loss by spin-relaxation
E / atom < <E> Density stays about constant
Causes heating

( c ) Steady state temperature (d) Forced evaporative cooling

Fig. 4. Stages in the trapping and cooling of atomic hydrogen.

the trap center. The atomic hydrogen source is located in a high field
solenoid.
Figure 4 illustrates the trapping and cooling process. In Fig. 4(a)
two important energies are introduced. The potential energy difference
between the center of the trap and the wall of the cell is represented by
£wau. The radial trapping fields are proportional to £waii- In practice the
lowest barrier to escape from the trap, ^threshold, occurs on axis at the
lower pinch solenoid, ^threshold governs the evaporative cooling process.
Atoms which manage to reach a helium covered surface, either by going
radially out to the wall or (much more likely) by escaping over the
threshold and entering the bottom part of the cell, can be adsorbed and
recombine into molecules.
Collisions between the atoms are essential to the loading of the trap.
Otherwise, atoms from the source would accelerate into the trapping
region, move across unimpeded, ride up the far side of the potential and
leave the trap over ^threshold- Radially, the atoms would bounce back
and forth between the walls. It is only through atom-atom collisions
that a fraction of the atoms lose enough energy to become trapped
near the center of the cell. Soon all atoms with energies greater than
140 T. J. Greytak

^threshold are lost and the trapped atoms come to a quasi-equilibrium

distribution. The atoms in the high energy tail of this distribution escape.
The remaining atoms redistribute their energies through collisions and
the temperature of the gas decreases. This spontaneous cooling of the
gas by evaporation is illustrated in Fig. 4(b). The temperature of the
gas becomes low enough to effectively isolate it from the surrounding
walls. If evaporation were the only way in which atoms could escape the
trap, the gas would continue to cool, although at an ever decreasing rate,
forever!
There is another way for the atoms to escape, however. Spin relaxation,
since it is a two-body process, takes place preferentially where the density
is highest, near the center of the trap where the atoms have the least
potential energy. Thus the atoms lost through spin relaxation tend to
have energies less than the average, and the process heats the gas. The
temperature of the gas reaches a steady state in which the heating due
to relaxation exactly balances the cooling due to evaporation. This
is indicated schematically in Fig. 4(c). The steady state temperature
depends on the geometry of the trap and the details of each of the loss
mechanisms. For an early MIT trap, the steady state temperature was
about one-twelfth of ^threshold over a wide range of operating conditions.
Since the steady state temperature is proportional to ^threshold, the
gas can be cooled further by slowly decreasing ^threshold- The rate at
which ^threshold is changed must be slow compared to the equilibration
rate of the atoms within the trap. Although the total number of atoms
decreases in this process, the density need not. In fact, if ^threshold (the
lower pinch field) were decreased without changing Ewaii (the quadrupole
field), the cooled atoms would fall deeper into the trap and the density
would actually increase! Since a higher density means increased electronic
relaxation, there is an advantage in keeping the density roughly constant
during the forced evaporative cooling. This can be done by decreasing
£waii simultaneously with ^threshold- This is the situation depicted in
Fig. 4(d). By controlling the two energies separately, one can change the
temperature and the density independently during the cooling process.
The trap shown in Fig. 3 was built at MIT and cooled atoms to less
than 3 mK [27, 28]. An apparatus similar in design was built in Ams-
terdam [29]. Experiments done there confined the atoms while keeping
them in thermal contact with the wall; they achieved gas temperatures
down to 80 mK. A second generation magnetic trap was designed at MIT
by John Doyle. The aims were to increase the trapping volume, obtain
lower temperatures and provide for optical access to the low temperature
 Prospects for BEC in Magnetically Trapped Atomic Hydrogen 141

Atom
source

Magnet
system

Bolometer

Retro mirror
O-k^m MicroChannel
plate
Fig. 5. The magnetic trap in current use at MIT.

gas. A scale drawing of the new trap is shown in Fig. 5. I will discuss its
operation and capabilities in some detail since this trap, or one like it,
will be used to try to achieve BEC in atomic hydrogen.
The loading of the trap and the subsequent development of electronic
and nuclear polarization in the gas can be understood with the help of
the simplified sketches in Fig. 6. Molecular hydrogen is condensed as a
solid on the walls of the source region during the initial cool down of
142 T. J. Greytak

J LHe H He
He He H
H H H
H2 H2 H He
He He
He He H H
He He H H
He H
He H2
He H H
H2 He H HeH H
HeH H
He He
H2 He H
H H
He H2 H He H He

n r
H2 He H He

T'r H

H
He
Hc

H He
H
He
H

H
He
He
J
H
He
He
H

Source before fill Source after discharge Trap after expansi

cycle begins of the gas

J L
He He
J L
He He
J L
He He

He He He He He He

He He He He He He
H
He He He He

i
He d He
dC
d
He cJ
He He He He d d
He
c1
dd
H
H d! d
He He He He He He
H
H d d
H c
He H He He He He He

He He He He He He

He He He He He He
He He He He He He He He He

Trap after helium Trap after high field seekers Trap after spin-exchang
recondenses return to the source eliminates c states

Fig. 6. Steps leading to a doubly polarized gas of low field seeking hydrogen
atoms in the MIT trap.

the cell. Enough 4 He is then admitted to form a puddle on the bottom

of the experimental cell. Because it is a superfluid, the 4 He provides a
self healing saturated film which covers all surfaces in the source and
trapping regions of the cell.
Loading the trap begins with an RF discharge in the source region
which evaporates the helium and vaporizes and dissociates some of the
hydrogen. The resulting gas cloud expands into the larger trapping
region below. The helium aids in the transport of the hydrogen and in
its subsequent equilibration to the wall temperature. The helium acts
as a buffer which slows the diffusion of the hydrogen to the walls. The
wall temperature at this point is kept high enough (about 250 mK) that
hydrogen atoms which happen to stick to the surface equilibrate and
desorb before they can recombine. Since the helium is unaffected by the
 Prospects for BEC in Magnetically Trapped Atomic Hydrogen 143
trapping potential, helium-hydrogen collisions bring the hydrogen gas in
the trap to the wall temperature, which is below the 800 mK trap depth.
The helium then quickly recondenses on the walls. The pressure in the
gas puff is sufficient to drive even the high field seekers out of the high
magnetic field present in the source region. After the helium recondenses,
electronic polarization develops as the a and b state atoms are drawn
back into the source. Less than a second has elapsed since the discharge
was turned off.
Nuclear polarization develops in the trapped gas due to spin-exchange
collisions: c + c —• b + d. All the c atoms are eliminated by this
process; the resulting b atoms are ejected from the trap. After about
three seconds the only atoms left in the trap are in the d state. The
wall temperature is then lowered to less than 80 mK. Now any atom
that escapes from the trap is permanently adsorbed on the surface and
cannot return to heat the gas. Thus the d atoms are thermally isolated.
Atoms promoted by collisions to energies high enough to reach the walls
will escape over ^threshold- The vapor pressure of the helium is too low for
helium-hydrogen collisions to be important. Similarly, thermal radiation
from the walls has a negligible effect. The d atoms are left to decay
slowly due to electronic spin relaxation, the dominant process being
d + d -» a + a.
In the above method of trap loading using a single short RF discharge,
the various physical processes occur sequentially. Other methods using
different patterns of discharges have been tried in order to maximize the
final d state density. In particular if the discharge is pulsed continuously,
all the above processes occur simultaneously and the ultimate density of
d atoms is determined by a balance between the flux of d atoms from
the source and spin relaxation taking place either in the trap or on
the surface. Table 3 shows typical conditions in the trapped gas soon
after the trap has been filled but before forced evaporative cooling has
commenced.
Determining the properties of the trapped gas poses a particular prob-
lem since the introduction of a solid probe, such as a thermometer
or a pressure transducer, would cause the atoms to be adsorbed and
recombine. Optical probes that avoid this problem are being devel-
oped and will be discussed below. However, almost all measurements
to date have required that the atoms be released from the trap. The
hyperfine states in the gas are determined by zero-field hyperfine reso-
nance in the gas after it flows into a field free region [27] (see Fig. 3).
The total number of atoms can be determined from the strength of
144 T. J. Greytak

Table 3. Magnetic trap after filling

hyperfine state d
number of atoms 2 x 1014
density on axis 2 x 1013 cm"3
temperature 50 mK
lifetime 20018sec 3
nc at this temperature 6 x 10 cm"
Tc at this density 12 ^K

the magnetic resonance signal [27, 28] or, more easily, from the energy
liberated when the released atoms recombine [30]. After nuclear po-
larization has been established, the time evolution of the total number
of atoms in the trap, N(t), is governed by spin relaxation and depends
on the trap shape, the electronic spin relaxation rate constant g, and
the temperature of the gas. Studies of N(t) for isolated atoms [27]
(where T was estimated) and for atoms in weak thermal contact with
the walls [29] (where T was known) confirmed the values of g calculated
theoretically [31, 32]. Once g was established, the temperature of the
gas and its density could be determined indirectly from an analysis of
N(t) [28].
A more direct method of measuring the gas temperature, developed
by Doyle [30], is illustrated in Fig. 7. A highly sensitive bolometer is
used to measure the flux of atoms from the trap as ^threshold is ramped
down to zero at a uniform rate. If the time necessary to drop the field
is short compared to the thermal equilibration time in the gas, the flux
versus time gives the number density versus energy, n(E). The energy
distribution n(E) is important in itself; for example it can be used to study
the approach to equilibrium during and after evaporative cooling. The
temperature of the gas is determined by comparing n(E) to the calculated
distribution expected for the known magnetic field profile. Figure 8 shows
n(E) measured after different degrees of evaporative cooling in the MIT
trap [17]. The lowest temperature achieved in that experiment, 100 /xK,
is well below the recoil limit for Lyman-a laser cooling of hydrogen,
T = (hv)2/2mc2kB « 650 JJK. Some of the features of this gas are given
in Table 4.
Other properties of this coldest hydrogen gas may be of interest. The
mean thermal speed, v = ^%kBT/nM, is 145 cm/s. The gravitational
height, kBT/Mg, is 8.4 cm and is beginning to play a role in the vertical
 Prospects for BEC in Magnetically Trapped Atomic Hydrogen 145

Fig. 7. The energy distribution in the gas, n(£), can be determined by measuring
the flux of atoms out of the trap as the threshold for escape is lowered quickly
to zero.

distribution of atoms in the trap. The thermal DeBroglie wavelength,

1/2
A(T) =
2nMkBTJ
= 1.739 x 10"7 (2)

is 174 nm; it is getting close to the mean separation between atoms,

n~1/3, which is equal to 232 nm. These properties depend only on
the temperature. The mean free path, X = ^n/8/n<j, depends on the
density as well. The total scattering cross section for this cold gas can
be expressed in terms of the s-wave scattering length, a: a = Ana2. The
s-wave scattering length for spin-polarized hydrogen is 0.72 A [33]. Thus
146 T. J. Greytak

12 16
Energy [mK]

250 500 750 1000

Energy |>K]

200 300
Energy

Fig. 8. Energy distributions for evaporatively cooled atomic hydrogen. The solid
curves are the calculated distributions for atoms at (a) 1.1 mK, (b) 190 //K, and
(c) 100 fiK. The dashed curves in (c) are for temperatures 30% higher and lower
than the best fit value. The finite atom signals from E < 0 are due to a small
axial field inhomogeneity ( < 1 G) that slightly deforms the bottom of the trap.
 Prospects for BEC in Magnetically Trapped Atomic Hydrogen 147

Table 4. Coldest gas attained by evaporation

hyperfine state d
number of atoms 1011
3 x 13
density on axis 8 x 10 cm"3
temperature 100 ^K
lifetime 50 sec
nc at this temperature 5 x 1014 cm"3
Tc at this density 30 ^K

for a uniform gas at a density of 8 x 1013 cm 3 the mean free path would
be 12 cm and the mean free time, X/v, would be 83 ms. It is clear, then,
that the trapped atoms traverse the confining potential hundreds of times
between collisions.
As Table 4 indicates, one would have to cool this gas by only a little
more than a factor of three in order to attain BEC. Is this feasible?
For evaporative cooling to be effective the thermalization rate for the
gas (proportional to the collision rate) must be greater than the spin
relaxation rate. Both of these rates are proportional to the density,
but the collision rate is also proportional to the square root of the
temperature whereas the relaxation rate is independent of temperature
for low temperatures and fields. Therefore the efficiency of evaporative
cooling decreases as the temperature is lowered [26]. Nevertheless, model
calculations [34] indicate that evaporative cooling using the current MIT
trap can cross the phase boundary for BEC at densities between about
5 x 1013 and 5 x 1014 cm" 3 . In order to observe the phase transition, it
would be very useful to have a non-destructive means of studying the gas.
The following section discusses the expected properties of the hydrogen
gas as it undergoes BEC in the trap and a possible unambiguous way of
observing the phase transition.

4 Bose Condensation in a Magnetic Trap

Bose-Einstein condensation of hydrogen in a magnetic trap will differ in
two fundamental ways from the simple case treated in textbooks. The
gas will not be homogeneous since it is confined by a spatially varying
potential. Also, the interactions between the atoms, although weak, have
a marked effect on the properties of the gas.
BEC in non-uniform potentials has been treated by Bagnato, Pritchard
and Kleppner [35]. They find that the density at the onset of BEC is
148 T. J. Greytak
given by the same expression that applies to the homogeneous case,

nc(T) = 2.612A(T)-3
= 4.965 x 1020T3/2, (3)

irrespective of the nature of the trapping potential. For a given trap

the condensate will begin to form at the location of the minimum in the
potential energy. As more atoms are added at fixed temperature, or as the
temperature is lowered at fixed N, the phase boundary expands outward,
coinciding with surfaces of constant potential energy. The density is
equal to the critical density nc(T) on the boundary and is greater than
the critical density inside the boundary.
The properties of a weakly interacting Bose gas have been covered in
detail by Huang [36]. The interaction between two particles is normally
described by an interaction potential Vflri—T2l)> but it can alternatively be
described by the wavevector dependent partial-wave phase shifts, di(k).
These phase shifts determine the form of the scattering cross section.
At sufficiently low energies, such as those associated with hydrogen in
magnetic traps, only s-wave scattering need be considered. This means
that only a single number, the s-wave scattering length a, is necessary to
parameterize the interactions between the hydrogen atoms. If, in addition,
the density of the gas is low enough that three-body collisions can be
neglected (an1/3 < 1, which is certainly true in the trap), then the entire
many-body behavior of the system depends solely on a single parameter,
a. The value a = 0.72 A calculated for spin-polarized hydrogen by Friend
and Etters [33] was based on the precise potential between hydrogen
atoms determined by Kolos and Wolniewicz [37]. It is interesting to
note, though, that exactly the same value would result from hypothetical
hard-sphere atoms of diameter a.
Although the fundamental quantity which determines the properties
of the weakly interacting Bose gas is the s-wave scattering length, there
are several derived quantities which are useful in developing a physical
understanding of the behavior of the gas. The healing length, Y\, is the
characteristic length associated with the many-body wave function and
indicates how rapidly the density can change in space [38]

i/ = (Snan)-1/2
= 2.35 x 10 3 ^p. (4)
 Prospects for BEC in Magnetically Trapped Atomic Hydrogen 149
The interaction strength Vo has the units of energy times volume

V0/kB = 4.36 x 1(T22 K cm3.

It sets the energy scale for various spatial configurations of the density.
In particular, the ground state energy per particle in a uniform gas is
given by E(0)/N = nVo/2.
The sound speed in a uniform weakly interacting Bose gas is given by

u = ^nan)1'2. (6)

This quantity by itself is of little use in understanding the trapped

hydrogen. However, it allows one to motivate the characteristic time for
changes in the wave-function, TC, as being related to the time necessary
for a sound wave to propagate over a distance equal to the healing
length:

nVo ~ V 2 u
1.75 x 1010

There has long been concern about the time necessary to nucleate the
Bose-Einstein phase transition in a low density gas once the critical
density or temperature has been reached. Recent theoretical calculations
by Stoof [39] and Kagan, Svistunov and Shlyapnikov [40] indicate that
the time is short compared to the expected lifetime of the trapped gas.
The details of the two theories differ (see the articles by Kagan and Stoof
in this book.) However, there seems to be agreement that an observable
peaking of the energy density n{E) in a group of states close to E = 0
will occur on a timescale of the order of TC. For a particle density of
n = 8 x 1013 cm"3, the characteristic time is TC = 0.2 ms.
Let us consider a specific example of a configuration in which BEC
might occur in the trapped hydrogen. By the time the magnetic fields
have been reduced to relatively low values during the evaporative cooling
process, the potential near the center can be approximated by a parabolic
minimum. There is a great deal of freedom in choosing the exact shape
of the potential in which the coldest atoms will reside. One possible
potential which is both attainable and convenient is
U(r,z)/kB = 1.4 x 10~2r2 + 3.0 x 10"5z2, (8)
150 T. J. Greytak

z=0

Single-particle Thermal distribution

ground state r = 460 microns
r = 6 microns
Laser beam
r = 40 microns

de Broglie wavelength Healing length

A(T) = 0.32 microns 1 mm r| = 2.6 microns

Fig. 9. Spatial profiles in a realistic trap potential. The temperature is 30 //K,

and the corresponding critical density is 8 x 1013 cm"3. The phase boundary for
BEC has been drawn for a condensate fraction of 2.5%.

where the radial (r) and axial (z) coordinates are expressed in centimeters
and U/ks is expressed in units of degrees Kelvin. In an actual experiment
the trap shape, temperature and number of atoms will all be changing
simultaneously. I will consider here a simpler hypothetical situation where
the trap shape and the temperature are held fixed and the total number
of atoms in the trap is slowly increased. I will take the temperature to be
30 /iK. At this temperature the critical density is nc(T) = 8 x 1013 cm" 3 .
When N is so small that the density at the center of the trap is much
less than nc(T)9 the density profile is proportional to e~u^kBT', as in the
case of a classical gas. The surface on which the density has fallen to e~ l
of its maximum value is ellipsoid of revolution extending to r = 460 /am
at z = 0 and z = ±10000 fim on axis (22 times longer than it is wide).
This is shown in Fig. 9. As N increases the density distribution develops
more weight at the low values of U than would be expected from the
classical form. Critical density is reached at the center of the trap when
N = 4.32 x 1011. Consider the various distance scales. The thermal
DeBroglie wavelength A(T) = 0.32 fim at this temperature. The healing
length Y\ = 2.6 /mi at this density. If there were no interactions between
the particles, the atoms would begin to condense into the single particle
ground state associated with this harmonic potential. That wave function
would have a radial extent of about 6 jum, too narrow to be resolved
in Fig. 9. Because of the repulsive interactions, the actual condensate is
 Prospects for BEC in Magnetically Trapped Atomic Hydrogen 151
spread over a wider region. For example, when N reaches 4.55 x 1011
the fraction of atoms in the condensate is 2.5% and the phase boundary
extends out to r = 63 irni.
Once a condensed state has formed in the gas, the density at the center
of the trap increases rapidly above nc(T) as more atoms are added. Since
the atom loss rate due to electronic relaxation is proportional to the
square of the density, the gas should begin to decay more rapidly once
BEC has been established. Hijmans et al. [41] have suggested that the
density at the trap center builds up so quickly with increasing condensate
fraction that the resulting rapid relaxation and subsequent recombination
can be likened to an 'explosion' (see the contribution by Hijmans et al. in
this book.) In order to study this possibility for the specific trap discussed
above it is necessary to compute the complete density distribution, n(r).
A simple thermodynamic approach to calculating n(r) has been intro-
duced by Oliva [42]. In this approach the chemical potential ja is taken
to be constant throughout the system. At each point \i is assumed to be
the sum of the 'internal chemical potential' of the gas, /xo(n(r)), and the
potential energy at that location:
/i = /io(n(r)) + l7(r). (9)
The internal chemical potential is simply the chemical potential of a
homogeneous gas of the same density in the absence of an applied
potential. For the weakly interacting Bose gas|
fio(n(r)) = V0(n + nc(T)), n>nc(T), (10)
= 2nVo+kBT\nz, n < nc(T). (11)
In the normal state the expression for /zo(n(r)) involves the fugacity,
z = ^ O / / C B T , which must be determined from the relation

g3/2(z)- 2.612^, (12)

where gs/2(z) is the familiar function from the theory of the ideal Bose
gas. To find n(r) one first chooses a value for \x. Then at each point in the
gas \x0 is found from Eq. (9) and either Eq. (11) or Eq. (12) is inverted to
find the density at that point. Finally, one integrates n(r) over the trap to
find the total number of atoms N corresponding to that value of \i. Since
N is a monotonically increasing function of /*, one can watch n(r) evolve
t Expressions for most of the thermodynamic variables in a weakly interacting Bose gas
are given in the review by Greytak and Kleppner [19]. The chemical potential is most
easily found from the fact that it is the Gibbs function per particle: \i0 = G/N =
(E-TS+PV)/N.
152 T. J. Greytak
as the trap is filled by systematically increasing the values of \x used in
the procedure. This 'local density approximation' approach to finding
n(r) should be accurate except in those regions where n(r) is changing on
a scale comparable to the healing length.
The expression for the internal chemical potential in the condensed
state, Eq. (11), can easily be inverted and used with Eq. (9) to give an
analytic expression for the density:

n(r) = n(0)-U(r)/Vo n>nc(T). (13)

This expression assumes l/(r) has its minimum value of 0 at r = 0. The

density at the origin, n(0), is related to the overall chemical potential
of the gas: \i — Vo(n(0) + nc(T)). This is an attractively simple result.
The density inside the phase boundary is directly proportional (with a
negative sign) to U(r); the density profile will be an inverted replica of
the trapping potential.
Figure 10 demonstrates how rapidly the density increases near the cen-
ter of the trap as additional atoms are added, once the critical density has
been exceeded. The isothermal compressibility increases discontinuously
by a factor of two upon entering the condensed state, causing a similar
increase in the gradient of the density at the phase boundary [19]. This
factor of two change in the slope has been observed in earlier density
profile calculations [43-45] and is reproduced in the local density approx-
imation approach [42]. However, at the low gas densities encountered
in traps, the slope in the normal state flattens so rapidly just above the
transition that the apparent slope change upon entering the transition
region is much larger than a factor of two, as can be seen in the Fig. 10.
Figure 11 shows how the condensate fraction, density at the origin and
decay time vary as more atoms are added to the trap. For a homogeneous
Bose gas the fraction of atoms in the condensate is 1 — nc(T)/n. To
determine the number of 'condensate atoms' in the trap this fraction
is multiplied by the density and integrated over the condensed phase.
The condensate fraction rises rapidly with increasing JV. The density
at the origin is a particularly steep function of the condensate fraction.
Fortunately very little of the volume of the trap is associated with the
region near the origin so the effect on the decay time is mitigated. The
decay rate of the gas was calculated simply by integrating gn2 over
the trap, neglecting any possible changes in the dipolar rate constant
g in the condensed state, or possible coherence effects due to long
range correlations. The resulting decrease in the lifetime of the gas
 Prospects for BEC in Magnetically Trapped Atomic Hydrogen 153

0.005 0.010 0.015 0.020

Radial extent [cm]
Fig. 10. Density profiles showing the formation of the condensed phase in the
trap of Fig. 9. The lowest curve is the density just before condensation begins.
The other curves correspond to condensate fractions of 2.6 x 10~5, 1.5 x 10~4,
1.5 x 10"3 and 2.5 x 10"2.

would certainly be observable, but may not prove to be an experimental

obstacle, at least up to condensate fractions of a few percent.
The detection of BEC in a magnetic trap will most likely require the
development of new experimental probes. Optical techniques offer the
most promise. The Amsterdam group has developed a pulsed Lyman-
<x source which can be used to measure and manipulate the trapped
hydrogen in situ through single-photon absorption. They have used
1S-2P Lyman-a transmission spectroscopy to obtain information on
the temperature and density of the gas [46]. The absorption lines are
broadened by the Zeeman effect which depends on the magnetic field at
the atom. Thus the lineshape depends on n(r) through the field profile
B(r). A model was made of the lineshape that includes T and the density
on axis, n0. These two parameters were then determined by fitting the
 154 T. J. Greytak

0.04

42 44 46 48 0 0.01 0.02 0.03 0.04 0 0.01 0.02 0.03 0.04

Number of atoms [x10 1 0 ] Condensate fraction Condensate fraction

Fig. 11. Behavior of critical parameters during Bose condensation in the trapping
potential of Fig. 9.

results of the model to the observed lineshapes. The same source has
been used to demonstrate Doppler cooling and light-induced evaporation
of trapped hydrogen [47]. In the latter case optical absorption is used
to remove high energy atoms from the trap by turning them into high
field seekers. Using this technique, they were able to cool the trapped
hydrogen to 3 mK.
Jon Sandberg at MIT developed a different optical probe, based on
two-photon absorption. The process is illustrated in Fig. 12. A continuous
source of radiation at one half the 1S-2S frequency is focused to a thin
line in the trapped gas (see also Fig. 9). The light is reflected back along
the same path by a mirror located at low temperature. Some atoms
undergo a two-photon transition from the IS to the 2S state. The 2S
state is metastable, so a small electric field is applied to mix the 2S
and the 2P states. The atoms then decay rapidly from the 2P state by
emitting a Lyman-a photon. The number of Lyman-a photons detected
by a micro-channel plate at low temperature is directly proportional
to the number of hydrogen atoms along the light beam path. The
wavelength dependence of the detector sensitivity strongly discriminates
against light from the source itself. To further decrease the sensitivity
to the illuminating radiation, the source can be chopped and the Stark
mixing of the 2S and 2P states delayed until the 'off' interval.
One way of using the two-photon absorption to study the gas is to
map out the radial density profile by translating the beam across the gas
parallel to the axis (in practice, it may be simpler to move the gas across
 Prospects for BEC in Magnetically Trapped Atomic Hydrogen 155

i»
Laser beam ' l*

Mirror
s
Trapped gas

Doppler shifted

"k + ~k
-~k + (-~k)

Doppler free

"k + (-~k)
Fig. 12. Two-photon absorption in a cold hydrogen gas.

the beam). If the trap potential is known, the radial extent of n(r) gives
a measure of the temperature. BEC could be identified by the sudden
appearance of a sharp peak in the density at the center of the trap, as in
Fig. 10.
A more direct signature of the phase transition can be obtained by
measuring the spectrum of the absorption as the frequency of the source
is scanned about the atomic frequency. Figure 12 shows that there are
two types of processes which contribute to the two-photon absorption.
If one photon is taken from each of the two counter propagating waves,
there will be no first order Doppler shift associated with the atom's
motion. This process gives rise to an extremely narrow absorption line
(of the order of 3 KHz wide due primarily to the time of flight of an atom
across the beam [48]), centered at half the atomic frequency v o /2, with
a strength proportional to the total number of atoms illuminated. On
the other hand, if the two absorbed photons were traveling in the same
direction, there would be a Doppler shift of the absorption frequency

hk2 kp cos 6 ....

v - v0 2 = —— + ; , (14)
2nM 2nM
where k is the wavevector of the photon and p cos 6 is the component of
156 T. J. Greytak

' «N
T>T C

RECOIL SHIFT

Fig. 13. The spectrum of two-photon absorption in trapped hydrogen can be

used to identify Bose-Einstein condensation. Av is the difference between the
laser frequency and half of the 1S-2S frequency for hydrogen. The recoil shift
is 6.7 MHz. The full width at half height of the low density (classical) Doppler
part of the spectrum is 5.7 MHz at T = 30 /zK. The ratio of the strength of the
discrete component in the Doppler part of the spectrum below the transition to
the strength of the Doppler free part is proportional to the condensate fraction.
Adapted from Sandberg [48].

the atom's momentum along the axis. The familiar p dependent Doppler
term (which mirrors the probability distribution for a given component
of the atom's momentum) is displaced to higher frequencies by a recoil
term. In most gases the recoil shift is negligible compared to the width
of the Doppler spectrum; but in trapped hydrogen, because of the light
mass and low temperature, the width and the shift are comparable. This
is shown schematically in Fig. 13.
By sweeping the frequency of the source one can measure the momen-
tum distribution and, of course, deduce the temperature. The signal will
be quite weak compared to the Doppler free case, however, since this
part of the absorption spectrum is much broader than the line width
of the source. The situation changes below the transition when a finite
fraction of the atoms come almost to rest. In this case a sharp feature will
appear in the absorption centered at the recoil shift with a strength pro-
portional to the condensate fraction (see Fig. 13). Kagan, Svistunov and
Shlyapnikov [40] estimate that the states which develop high occupation
below the transition are within a range A£ ~ nV0 of E = 0. This would
give rise to a frequency spread of the order of one KHz in the recoil
shifted condensate line, which is less than the time of flight broadening
expected for the Doppler free component. These considerations have
led Sandberg [48] to suggest that measuring the two-photon absorption
spectrum may be the simplest and least ambiguous way of detecting BEC
in trapped hydrogen.
 Prospects for BEC in Magnetically Trapped Atomic Hydrogen 157
The continuous wave laser source for the two-photon absorption has
been completed. It produces 20 mW of tunable light at X = 243 nm
with a spectral width of about 2 KHz. Work is now underway to see the
absorption signal in the cold hydrogen.

Acknowledgments. The research on spin-polarized hydrogen at MIT has

been, since its inception, a collaborative effort between the author (a
condensed matter physicist) and Daniel Kleppner (an atomic physicist).
Many talented graduate students and post doctoral fellows have con-
tributed to the success of the work, as can be seen by the multitude of
names on our papers. Two individuals who deserve special recognition
for the most recent advances are John Doyle, who was responsible for
the second generation magnetic trap and Jon Sandberg, who built up
the entire UV laser system starting with an empty room. Our current
support is provided by the National Science Foundation under Grant No.
DMR-91-19426 and the Air Force Office of Scientific Research under
Grant No. F49620-92-J-0356.

References
[I] C. E. Hecht, Physica 25, 1159 (1959).
[2] W. C. Stwalley and L. H. Nosanow, Phys. Rev. Lett. 36, 910 (1976).
[3] I. F. Silvera and J. T. M. Walraven, Phys. Rev. Lett. 44, 164 (1980).
[4] B. W. Statt and A. J. Berlinsky, Phys. Rev. Lett. 45, 2105, (1980).
[5] R. W. Cline, T. J. Greytak and D. Kleppner, Phys. Rev. Lett. 47,
1195 (1981).
[6] E. P. Bashkin, Pis'ma Zh. Eksp. Teor. Fiz. 33 , 11 (1981) [Soviet
Physics JETP Lett. 33, 8 (1981)].
[7] C. Lhuillier and F. Laloe, J. Phys. (Paris) 43, 197, 225 (1982).
[8] B. R. Johnson, J. S. Denker, N. Bigelow, L. P. Levy, J. H. Freed and
D. M. Lee, Phys. Rev. Lett. 52, 1508 (1984).
[9] D. A. Bell, H. F. Hess, G. P. Kochanski, S. Buchman, L. Pollack,
Y. M. Xiao, D. Kleppner and T. J. Greytak, Phys. Rev. B 34, 7670
(1986).
[10] R. Sprik, J. T. M. Walraven and I. F. Silvera, Phys. Rev. B 32, 5668
(1985).
[II] Yu. Kagan, I. A. Vartanyantz and G. V. Shlyapnikov, Sov. Phys.
JETP 54, 590 (1981).
[12] H. F. Hess, G. P. Kochanski, J. M. Doyle, T. J. Greytak and D.
Kleppner, Phys. Rev. A 34 ,1602 (1986).
158 T. J. Greytak
[13] M. D. Hiirlimann, W. N. Hardy, A. J. Berlinsky and R. W. dine,
Phys. Rev. A 34, 1605 (1986).
[14] R. L. Walsworth Jr, I. F. Silvera, H. P. Godfried, C. C. Agosta, R.
F. C. Vessot and E. M. Mattison, Phys. Rev. A 34, 2550, (1986).
[15] A. J. Berlinsky and W. N. Hardy, 13th Annual Precision Time and
Time Interval Applications and Planning Meeting, 1981, NASA
Conference Publication 220, p. 547.
[16] J. J. Berkhout, E. J. Wolters, R. van Roijen and J. T. M. Walraven,
Phys. Rev. Lett. 57, 2387 (1986).
[17] J. M. Doyle, J. C. Sandberg, I. A. Yu, C. L. Cesar, D. Kleppner and
T. J. Greytak, Phys. Rev. Lett. 67, 603 (1991).
[18] I. A. Yu, J. M. Doyle, J. C. Sandberg, C. L. Cesar, D. Kleppner and
T. J. Greytak, Phys. Rev. Lett. 71, 1589 (1993).
[19] T. J. Greytak and D. Kleppner, in New Trends in Atomic Physics, G.
Grynberg and R. Stora, eds. (North-Holland, Amsterdam, 1984).
[20] I. F. Silvera and J. T. M. Walraven, Spin polarized atomic hydrogen,
in Progress in Low Temperature Physics, D.F. Brewer, ed. (North-
Holland, Amsterdam, 1986), Vol. X, p. 139.
[21] T Tommila, E. Tjukanov, M. Krusius and S. Jaakkola, Phys. Rev.
B 36, 6837 (1987).
[22] H. T. C. Stoof, L. P. H. de Goey, B. J. Verhaar and W. Glockle,
Phys. Rev. B 38, 11221 (1988).
[23] J. D. Gillaspy, I. F. Silvera and J. S. Brooks, Phys. Rev. B 40, 210
(1989).
[24] Yu. Kagan and G. V. Shlyapnikov, Phys. Lett. A 130, 483, (1988).
[25] E. Tjukanov, A. Ya. Katunin, A. I. Safonov, P. Arvela, M. Karhunen,
B. G. Lazarev, G. V. Shlyapnikov, 1.1. Lukashevich and S. Jaakkola,
Physica B 178, 129 (1992).
[26] H. F. Hess, Phys. Rev. B 34, 3476 (1986).
[27] H. F. Hess, G. P. Kochanski, J. M. Doyle, N. Masuhara, D. Kleppner
and T. J. Greytak, Phys. Rev. Lett. 59, 672 (1987).
[28] N. Masuhara, J. M. Doyle, J. Sandberg, D. Kleppner, T. J. Greytak,
H. F. Hess and G. P. Kochanski, Phys. Rev. Lett. 61, 935 (1988).
[29] R. van Roijen, J. J. Berkhout, S. Jaakkola and J. T. M. Walraven,
Phys. Rev. Lett. 61, 931 (1988).
[30] J. M. Doyle, J. C. Sandberg, N. Masuhara, I. A. Yu, D. Kleppner
and T. J. Greytak, J. Opt. Soc. Am. B 6, 2244 (1989).
[31] A. Lagendijk, I. F. Silvera and B. J. Verhaar, Phys. Rev. B 33, 626
(1986).
 Prospects for BEC in Magnetically Trapped Atomic Hydrogen 159
[32] H. T. C. Stoof, J. M. V. A. Koelman and B. J. Verhaar, Phys. Rev.
B 38, 4688 (1988).
[33] D. G. Friend and R. D. Etters, J. Low Temp. Phys. 39, 409 (1980).
[34] J. M. Doyle, Ph.D. thesis, MIT (1991).
[35] V. Bagnato, D. E. Pritchard and D. Kleppner, Phys. Rev. A 35, 4354
(1987).
[36] K. Huang, Studies in Statistical Mechanics II, J. DeBoer and G. E.
Uhlenbeck, eds. (North-Holland, Amsterdam, 1964).
[37] W. Kolos and L. Wolniewicz, J. Chem. Phys. 43, 2429 (1965), and
Chem. Phys. Lett. 24, 457 (1974).
[38] E. P. Gross, J. Math. Phys. 4, 195 (1963).
[39] H. T. C. Stoof, Phys. Rev. Lett. 66, 3148 (1991).
[40] Yu. M. Kagan, B. V. Svistunov and G. V. Shlyapnikov, Sov. Phys.
JETP 75, 387 (1992).
[41] T. W. Hijmans, Yu. Kagan, G. V. Shlyapnikov and J. T M. Walraven,
contribution to this volume.
[42] J. Oliva, Phys. Rev. B 39, 4197 (1989).
[43] V. V. Goldman, I. F. Silvera and A. J. Leggett, Phys. Rev. B 24, 2870
(1981).
[44] C. A. Condat and R. A. Guyer, Phys. Rev. B 24, 2874 (1981).
[45] D. A. Huse and E. D. Siggia, J. Low Temp. Phys. 46, 137 (1982).
[46] O. J. Luiten, H. G. C. Werij, I. D. Setija, M. W. Reynolds, T. W.
Hijmans and J. T. M. Walraven, Phys. Rev. Lett. 70, 544 (1993).
[47] I. D. Setija, H. G. C. Werij, O. J. Luiten, M. W. Reynolds, T. W.
Hijmans and J. T. M. Walraven, Phys. Rev. Lett. 70, 2257 (1993).
[48] J. C. Sandberg, Ph.D. thesis, MIT (1993).
 8
Spin-Polarized Hydrogen: Prospects for
Bose-Einstein Condensation and
Two-Dimensional Superfluidity
Isaac F. Silvera
Lyman Laboratory of Physics
Harvard University
Cambridge MA 02138
USA

Abstract

Spin-polarized atomic hydrogen continues to be one of the most promising

candidates for Bose condensation of an atomic system. In contrast to
liquid helium, hydrogen is gaseous and therefore its density can be changed
to vary its behavior from the weakly to the strongly interacting Bose gas.
Until now, efforts to Bose condense hydrogen have been thwarted by re-
combination and relaxation phenomena. After a long introduction to the
subject, two promising approaches to observe quantum degenerate behavior
are discussed: the microwave trap and a two-dimensional gas of hydrogen.

1 Introduction

Since the stabilization of atomic hydrogen as a spin-polarized gas (H j),

reported in 1980 [1] , there has been a continuous effort to observe Bose-
Einstein condensation (BEC) or other effects of quantum degeneracy
in this Bose gas. This challenge has not yet been realized due to the
instability of Hj towards recombination to H2 or relaxation among
hyperfine states as the conditions for BEC are approached. Although
this difficulty is formidable, hydrogen presents a unique opportunity to
study BEC and the related superfluidity, as it remains a gas to T = 0 [2].
By comparison, 4He, the only experimentally observed boson fluid, is
a strongly interacting superfluid Bose liquid, with little flexibility for
varying its density. As a result of the gaseous nature of hydrogen,
its density, n, can be varied over several orders of magnitude with the
possibility of studying BEC and its relationship to superfluidity, ranging
from the weakly to the strongly interacting boson gas.

160
 Spin-Polarized Hydrogen: BEC and 2D Superfluidity 161
The critical temperature for BEC is given by
Tc = 331(h2/mkB)n2/3 (1)
When evaluated at Tc = 100 mK, n has a value of 1.57 x 1019/cm3.
Hydrogen has been cooled to such temperatures but not at this density.
The difficulty in achieving the conditions for BEC is the rapid intrinsic
recombination rate to molecular hydrogen. This has two effects: (i) it
limits the density, as the recombination rate is proportional to n3 in the
gas phase, and (ii) the 4.6 eV per recombining pair is usually dissipated
in the gas and can result in severe overheating of the sample, preventing
the attainment of Tc.
Hydrogen was initially stabilized in a cell with a thin film of superfluid
4
He covering the copper walls of the container. Hydrogen atoms reside
in states in the gas phase or states on the surface, and the ratio depends
exponentially on the adsorption energy. Helium presents a surface to the
hydrogen atoms with a very small adsorption energy, £a//c# « IK. In the
high temperature limit the surface coverage o of hydrogen is given by
(T = nkexp(sa/kBT), (2)
where k is the thermal de Broglie wavelength. From this we see that as n
increases or T decreases, the coverage builds up. The build up of surface
density can be very detrimental, as on the surface recombination can take
place when two hydrogen atoms collide in the presence of a helium atom.
The rate for this process is proportional to a2. At very high density the
surface recombination is dominated by three-body hydrogen collisions,
proportional to a3; ultimately two- and three-body recombination can
prevent the achievement of low temperatures in a cell or the build up of
gas phase density at low temperatures. Due to this problem some effort
has been expended to use magnetic traps to isolate hydrogen from the
container walls and completely eliminate wall recombination.
A more precise description of the recombination problem shows that
the rates depend on the hyperfine states of the colliding atoms. Due to
the spin 1/2 of the electron and spin 1/2 of the proton, there are four
hyperfine states, delineated a, b, c and d in order of increasing energy,
as shown in Fig. 1. In a high magnetic field the a and b states are
predominantly electron spin-down states, while the c and d states are
spin-up states. The recombination rate constants depend sensitively on
the hyperfine states of the atoms. For example, if two of the atoms in a
recombination collision are in one of the electron spin-up states and one
of the spin-down states, we have the situation for unpolarized hydrogen
162 /. F. Silvera

Fig. 1. The hyperfine energy levels of hydrogen as a function of magnetic field.

High field states are shown in terms of the projections of the electron and nuclear
spins; a is the hyperfine coupling constant.

and the recombination rate constant is very large. The rate constant is
reduced by several orders of magnitude when all atoms are in either the
up-spin states or down-spin states, i.e. spin-polarized. The a and c states
have admixtures of spin-up and spin-down. This admixture becomes very
small in a large magneticfield,which enhances the stability. However, the
b and d states are pure electron spin states and gases made of either one
of these states have still higher stability, as a recombination channel is
closed. Thus, the stability of the gas can be greatly increased by placing
all of the atoms in hyperfine states b or d.
Two-body collisions can also result in relaxation amongst the hyperfine
states. This can have beneficial or detrimental effects. For example,
in the static magnetic trap, the gas very rapidly relaxes to a single
hyperfine state, d, due to spin-exchange processes. This state is very
stable. Ultimately, however, as the conditions for BEC are approached,
magnetic dipolar relaxation dominates the behavior [3]. The d-state
atoms relax to the spin-down states which are ejected from the trap.
These processes have prevented the attainment of BEC in the static trap,
as will be discussed elsewhere in this volume [4].
Spin-polarized hydrogen also exists as an almost two-dimensional gas
when condensed on the surface of helium. The atoms are bound to
the 4He surface by an adsorption energy sa/kB « IK and are in two-
dimensional free particle momentum states; in thermodynamic equilib-
rium the coverage a is given by Eq. (2). At high densities this expression
 Spin-Polarized Hydrogen: BEC and 2D Superfluidity 163
is modified according to the Bose distribution; due to effective repulsive
interactions between hydrogen atoms on the surface, the coverage satu-
rates at crsat « l x 10 14/cm2. The surface gas of hydrogen is expected
to demonstrate the behavior of a degenerate Bose gas. Although it can
be shown rigorously that Bose- Einstein condensation does not exist in
two dimensions, it is believed that a quasi-condensate with short-range
correlations will exist and the correlation length rapidly becomes the
size of a laboratory sample as temperature is reduced below the critical
value. The critical temperature for the Kosterlitz-Thouless transition to
two-dimensional superfluidity is
T]D = nh2as/2kBm , (3)
where as ~ a is the superfluid fraction [5]. For GS = <7sat, the saturation
coverage, the calculated critical temperature is 756 mK. It is believed that
coverages of hydrogen in this range can be achieved. Thus, it is quite
realistic to seek the KT transition in hydrogen. Until recently most efforts
have concentrated on achieving BEC in three dimensions; however, it
now seems that it may be easier to deal with the recombination problems
in two dimensions, so new efforts are underway.
In the preceding we have given a brief introduction to spin-polarized
hydrogen and Bose condensation. In the remainder of this paper we shall
concentrate the discussion on two promising approaches: the microwave
trap and two-dimensional quasi-BEC. There are a number of other
interesting ideas for BEC which have been reviewed recently by Silvera [6]
and Silvera and Reynolds [7]. An earlier review by Silvera and Walraven
gives a detailed description of the general development of the field up
to about 1985, with descriptions of the experimental techniques and an
exhaustive treatment of recombination and relaxation processes. The
review by Greytak and Kleppner [8] discusses many of the properties of
the weakly interacting Bose gas in the region of BEC.

2 The Microwave Trap

In the introduction we discussed the static magnetic trap which has
advanced to within a factor of two or three of the critical density for
BEC. Temperatures as low as a few hundred microkelvin have been
achieved in this trap [9] by means of evaporative cooling of atoms
trapped in the d-state (lowering of the potential walls of the trap to allow
hot atoms to escape, with thermalization of the remaining atoms at a
lower temperature). At these low temperatures, as BEC is approached,
164 /. F. Silvera
the density peaks up in the center of the trap and the relaxation rate,
quadratic in density, increases [3]. As a result the further increase
of density ceases and the path to BEC is blocked. Nevertheless, the
MIT group feels that they can map out a trajectory in n-T space
which will allow them to Bose condense a sample. If this occurs, a
second problem recently studied by Hijmans et al. [10] arises. This is
a so-called relaxation explosion, not actually an explosion, but a large
enhancement of the relaxation rate. As soon as the BEC phase line is
crossed, the peak density rapidly increases with a concomitant increase
in the relaxation rate and loss of sample. As a result the sample is
confined to the BEC line and the order parameter cannot develop in
magnitude.
The microwave trap (MWT) does not suffer from these problems
as it is the ground a or b states which are trapped. The MWT was
proposed by Agosta and Silvera [11], with further considerations by
Agosta et al [12]. At high frequencies it is possible to produce a
magnetic field which has either a maximum or a minimum. A very
convenient geometry is a spherical microwave cavity in which one of
the fundamental modes has a field maximum. This can be tuned
to the region of the a-d transition or the b-c transition of hydro-
gen. An analysis in terms of dressed-atom states shows that in the
region of resonance a potential well is developed in the microwave
field gradient. The depth of this well for trapping of atoms is of the
order of several millikelvin for reasonable microwave powers and cav-
ity quality factor. Thus, to load this trap very cold atoms must be
used.
The eigenstates of the Hamiltonian (dressed states) near resonance
between the a and d states are
\\N) = cos0\d)®\N- 1) +sin0|a) ® |AT)(a),

where N = 0,1,2, ... , and \N) denotes the state of the magnetic mode
with N photons before coupling to the atom. The angle 6 is given
by tan 20 = cor/S, where cor is the Rabi frequency (cor = yeBrf for a
circularly polarized microwave field of amplitude Brf) and d = co — COQ is
the detuning of the microwave frequency from the resonance frequency,
co0 = Bo/ye (Bo is the applied static field). The energies are

l }
E2N= Nhco - fcO/2(b)
 Spin-Polarized Hydrogen: BEC and 2D Superfluidity 165

o.u
Static trap at B=5.0T

5.0

4.0 / \

J 3.0 I :
b
£- 2.0
O
Static trap at B=0.05T •
1.0

r\r\
-4.0 -3.0 -2.0 -1.0 0.0 1.0 2.0 3.0 4.0
8/(0 (dimensionless)

Fig. 2. The spin relaxation loss rate in a microwave trap for hydrogen. The
dashed lines show the loss rate in static traps for two different magnetic field
amplitudes. By using a negative detuning the loss rate can be made negligible in
the microwave trap; for large positive detuning the microwave rate approaches
the static rate.

where Q = (co2 + (52)1/2. The energy levels vary with position through
the variation of Q on position. From Eq. (5) we see that state |2N)
is a trapped state and |1N) is an antitrapped state. By varying S,
the admixture of states can be varied, and, with a reasonable negative
detuning, the character of the state \2N) becomes almost completely \a) 9
the ground state. Agosta et al. [12] have calculated the loss rate due to
relaxation and find that it becomes negligibly small, as shown in Fig. 2.
Thus, with the MWT, the principle loss mechanism of the static trap,
spin relaxation, can be suppressed to a very low level. Moreover, the
relaxation explosion is not a characteristic of this trap. After loading the
trap, by further detuning, the potential walls are lowered and evaporative
cooling can be used to achieve very low temperatures; alternatively atoms
can be selectively pumped from the trapped to the antitrapped state for
forced evaporative cooling, which does not change the shape of the trap.
If BEC can be achieved the principle loss mechanism will be due to
three-body recombination in the center of the trap which will become
important in the density region of 1017 to 10 18 /cm 3 . Since the initial
densities for BEC in this trap will be of order 10 14 /cm 3 , there should
be no problem with a recombination "explosion" when BEC is achieved.
166 /. F. Silvera

MOT/molasses
lasern

(AW input
1 |XW detect

/
\
Cs Oven
Slowing 1
laser • 0 O(
f/cco
Kl ^ Field
—-^ camera

Coil

^ I
Probe
laser
Mirror
Photo
d iode

Fig. 3. A schematic diagram of the cesium trap, showing the microwave cavity
and the several laser beams used for cooling the atoms to be loaded into the
trap. Atoms could be detected by releasing them from the trap and observing
theirfluorescenceas they passed through a probe beam. An alternative means of
observation is to laser illuminate trapped atoms and observe them in fluorescence
with a ccd camera.

Silvera et al. [13] have briefly discussed the problem of atomic losses
due to microwave noise; this appears to be a technically resolvable
problem.
The principle difficulty with the MWT is that it can only be loaded
with very cold atoms. Silvera and Reynolds [14] proposed a hybrid
trap in which H atoms would first be cooled in a static trap (in, say,
the d-state), evaporatively cooled to a few hundred microkelvin, and
then transferred to the MWT. Since this is a very demanding experi-
ment, the MWT was recently demonstrated experimentally [6, 15] with
cesium atoms in an experimental setup shown in Fig. 3. The atoms
are first laser cooled to a few microkelvin using a magneto-optical trap
and polarization laser cooling; the laser is then turned off and the
microwave power, up to 80 Watts in an x-band microwave cavity, is
turned on. A complication in this experiment was that the microwave
potential was insufficiently deep to hold atoms against the gravitational
field. This problem was solved by using a static gradient magnetic field
 Spin-Polarized Hydrogen: BEC and 2D Superfluidity 167

To Mixing Chamber

I Cell cooling
Discharge cooling

DISCHARGE

Race-tracks
Al olloy former

Top view of the quadrupole magnet

Fig. 4. The hybrid static microwave trap under construction at Harvard. The
static trap consists of 11 separately controlled superconducting magnets, used
for trapping, cooling and spatially manipulating the atoms. After being cooled,
atoms will be transferred to the microwave trap by varying the magnetic field
gradients.

to levitate the Cs atoms, with the microwave field gradient confining

them. The trapped atoms could be observed visually by inducing op-
tical fluorescence with a pulsed resonant laser beam. Zhao et al. [16]
are constructing a MWT for hydrogen, shown in Fig. 4. This trap
should pave the way to BEC in a three-dimensional gas of hydrogen.
168 /. F Silvera
3 The KT Transition and Quasi Bose-Einstein Condensation

The quest for Bose condensation in hydrogen has been focused strongly
on the three-dimensional gas. The attempts to Bose condense hydrogen
as a high density gas, to date, have been thwarted by recombination
heating. Although there are some proposals for experiments which can
handle the heating problem [6] , efforts have moved in the direction of
very low density traps. With regard to two-dimensional KT superflu-
idity, quasi-Bose condensation there seems to have been few efforts in
this direction [17]. Only recently have researchers become aware that it
is probably much easier to produce hydrogen in this state experimen-
tally than three-dimensional BEC. There have been a number of recent
theoretical and experimental results which indicate that this avenue of
research may be the more propitious approach.
The principle of the experiment is to create a two-dimensional sur-
face (of helium on a substrate); hydrogen will condense on the surface
and be stabilized by a magnetic field. The hydrogen behaves as a
two-dimensional gas, populating states acording to the Bose-Einstein
distribution function. However, it has been shown that in two dimen-
sions the Bose-Einstein transition does not take place [18]. Bagnato and
Kleppner [19] studied the ideal Bose gas in two dimensions and found
that, for example, in a non-uniform field with a harmonic potential a BEC
transition does take place. However, Shevchenko [20] studied the same
problem for the weakly interacting gas and found that in this case the
BEC transition does not take place, but, rather, there is a KT transition
and that the critical temperature is very close to the BEC temperature
in the absence of interactions. He then studied the quasi-condensate and
its development in a finite system [21]. The quasi-condensate had been
studied earlier by Kagan et al. [22]. Stoof and Bijlsma [23] have also
studied the thermodynamic properties of the weakly interacting Bose gas
and arrive at similar conclusions. Thus, on theoretical grounds we con-
clude that in two dimensions in an inhomogeneous magnetic field there
is a KT transition and a quasicondensate which becomes macroscopic at
a temperature close to TC2D.
Earlier, it was believed that it would be very difficult to study two-
dimensional BEC because of the very rapid recombination at the surface
coverages which are required and the resultant heating of the sample.
However, recent experiments have shown that when molecules recombine
on a surface there is a high probability that they will form in highly
excited states of Hi and fly off the surface. The fraction F of the
 Spin-Polarized Hydrogen: BEC and 2D Superfluidity 169
recombination energy deposited on the surface is found to have an upper
bound of 0.04 for two-body recombination, and there is probably a
similar bound for three-body surface recombination [17, 24]. Thus, only
a very small fraction of the recombination energy goes into the surface
to heat the sample and the heating problem of three-dimensional gases
can be strongly suppressed.
On the theoretical side there are a few additional advantages for the
experimentalist in studying properties of a quantum degenerate system
in two dimensions. Kagan et al. [22] had shown that the recombination
rate constant, K, is reduced by a factor of 6 for three-body recombina-
tion between atoms in the condensate; for a fixed surface coverage the
recombination rate should decrease by this amount as T/T?D decreases
and almost all of the atoms are in the quasi-condensate or superfluid
fraction. The origin for this decrease is in the very simple form of the
wavefunction for the zero-momentum or ground state particles, which
leads to a ^ 3 ! reduction in the matrix element for the process. For
the same reason, the two-particle interaction potential is reduced by 2!
As a result, if there is thermodynamic equilibrium between surface state
atoms and a thermal reservoir of gas atoms in three-dimensions, then at
the transition there is an increase in surface density by about a factor
of two [25]. Since the recombination rate is proportional to KG3, the
rate is expected to have a slight increase at TC2D rather than a decrease.
The most recent development is a calculation by Stoof and Bijlsma [26],
who go beyond the approach of Kagan et al. [22] and, in addition to
wavefunction symmetry effects, they consider the many-body states of
the gas in greater detail. They find an enormous reduction of the rate
constant, more than 400, and an increase in the density step to about
3 at the critical temperature. Together, this predicts that the rate will
decrease by about 33/400 « 15 at the transition. This is likely an over-
estimate since the calculation is probably not valid in the critical region.
Nevertheless, outside the critical region there is still a large predicted
reduction in the recombination rate in the superfluid, according to this
result. From an experimental point of view these predictions suggest
a very nice property of the sample which can be used to identify the
transition, the recombination rate or the recombination heating of the
sample. It will certainly be interesting if such enormous changes (400)
can arise due to Bose effects.
Experimental attempts to study the KT transition are so meager that it
is not meaningful to assess how close experiments have approached this
transition. An experiment was started at the University of Amsterdam to
170 /. F. Silvera
observe the KT transition by optical methods using a Lyman-alpha laser
source. The experimental plan was to introduce a gas of hydrogen to a
small, cold helium surface on the surface of a large reservoir of atoms
stored at a higher temperature, with higher temperature walls. In this way
the main recombination losses would be on the cold surface and could be
maintained at a manageable level. The density of atoms on the surface
would be determined using the Lyman-alpha radiation to optically excite
the system such that the surface coverage could be measured. However, it
was realized that the hydrogen interacts with the surface via the surface
ripplons of the helium and at the low temperature of the experiment
these are weakly coupled to the helium phonon bath [27]. As a result
the ripplons heat up and the hydrogen would not be at a sufficiently low
temperature required for the KT transition; the experiment was aborted.
An experiment at the University of Kyoto [17] using a cold surface has
been carried out at a measured cell temperature of 87 mK; however the
coverages were much too low for the KT transition. This experiment
also determined that the F factor described above is about 0.015. An
experiment is currently being developed at Harvard University [28] using
a two-dimensional magnetic field gradient with a maximum field at the
center of a disc-shaped surface. In this way very high surface coverages
can be achieved in a very small area so that the total sample loss is
very small. The atoms that cover the surface are fed from a large buffer
volume at a higher temperature in such a way that the total number
of atoms on the surface is maintained constant; this inhibits the density
step increase at the critical temperature and should amplify sensitivity
to changes in recombination energy. The means of detection of the
KT transition will be a measurement of the recombination heating on
the cell surface. Similar experiments are planned at the University of
Amsterdam [29] and the University of Turku in collaboration with the
Kurchatov Institute [30], using optical detection or thermal detection
respectively.

4 Conclusions
For more than 13 years, since the initial stabilization of atomic hydro-
gen, researchers have been devising schemes to observe Bose-Einstein
condensation in the weakly interacting Bose gas. In the first several years
a number of clever experiments were attempted, only to be thwarted by
the rapid loss of sample, and heating due to recombination and relax-
ation among hyperfine levels. In this period, there has been an enormous
 Spin-Polarized Hydrogen: BEC and 2D Superfluidity 171
growth in the basic theoretical and experimental understanding of this
seemingly simple system. Fortified with this knowledge, new experiments
and approaches have been developed, bringing us within reach of the
goal of BEC, which at times has seemed unattainable. The coming years
promise exciting new results in this simple Boson gas.

Acknowledgment. I thank the Department of Energy, Grant Number

DE-FGO2-85ER45190 for support.

References
[I] I.F. Silvera and J.T.M. Walraven, Phys. Rev. Lett. 44, 164 (1980).
[2] I.F. Silvera and J.T.M. Walraven, Progress in Low Temp. Physics,
Vol. X, D.F. Brewer, ed. (Elsevier, Amsterdam, 1986), p. 139.
[3] A. Lagendijk, I.F. Silvera, and B.J. Verhaar, Phys. Rev. B 33, 626
(1986).
[4] See T.J. Greytak, this volume.
[5] D.O. Edwards, Physica B 109 and 110, 1531 (1982).
[6] I.F. Silvera, J. Low Temp. Phys. 89, 287 (1992).
[7] I.F. Silvera and M. Reynolds, J. Low Temp. Phys. 87, 343 (1992).
[8] T.J. Greytak and D. Kleppner, New Trends in Atomic Physics, Vol.
II, G. Grynberg and R. Stora, eds. (North Holland, Amsterdam,
1984), p. 1127.
[9] J.M. Doyle, J.C. Sandberg, LA. Yu, C.L. Cesar, D. Kleppner, and
T.J. Greytak, Phys. Rev. Lett. 67, 603 (1991).
[10] T.W. Hijmans, Y. Kagan, G.V. Shlyapnikov, and J.T.M. Walraven,
preprint (1993).
[II] C.C. Agosta and I.F. Silvera, Spin Polarized Quantum Systems,
S. Stringari, ed. (World Scientific, Singapore, 1989), p. 254.
[12] C.C. Agosta, I.F. Silvera, H.T.C. Stoof, and B.J. Verhaar, Phys. Rev.
Lett. 62, 2361 (1989).
[13] I.F. Silvera, C. Gerz, L.S. Goldner, W.D. Phillips, M.W. Reynolds,
S.L. Rolston, R.J.C. Spreeuw, and C.I. Westbrook, LT20, Physica
B 194^196, in press (1994).
[14] I.F. Silvera and M. Reynolds, Physica B 169, 449 (1991).
[15] L. Goldner, R.J.C. Spreeuw, C. Gerz, W.D. Phillips, M.W. Reynolds,
S.L. Rolston, I.F. Silvera, and C.L Westbrook, LT20, Physica B 194-
196, in press (1994).
[16] Z. Zhao, I.F. Silvera, and M. Reynolds, J. Low Temp. Phys. 89, 703
(1992).
172 /. F. Silvera
[17] A. Matsubara, T. Arai, S. Hotta, J.S. Korhonen, T. Mizusaki, and
A. Hirai, this volume.
[18] RC. Hohenberg, Phys. Rev. 158, 383 (1967).
[19] V. Bagnato and D. Kleppner, Phys. Rev. A 44 , 7439 (1991).
[20] S.I. Shevchenko, Sov. Phys. JETP 73, 1009 (1991).
[21] S.I. Shevchenko, Sov. J. Low Temp. Phys. 18, 223 (1992).
[22] Y. Kagan, B.V. Svistunov, and G.V. Shlyapnikov, Sov. Phys. JETP
66, 314 (1987).
[23] H.T.C. Stoof and M. Bijlsma, Phys. Rev. B 47 (2), 939 (1993).
[24] Z. Zhao, E.S. Meyer, B. Freedman, J. Kim, J.C. Mester, and I.F. Sil-
vera, LT20, Physica B 194-196, in press (1994).
[25] B.V. Svistunov, T.W. Hijmans, B.V. Shlyapnikov, and J.T.M. Wal-
raven, this volume.
[26] H. Stoof and M. Bijlsmaa, preprint (1993).
[27] M.W. Reynolds, I.D. Setija, and G.V. Shlyapnikov, Phys. Rev. B 46,
575 (1992).
[28] I.F. Silvera, unpublished.
[29] M. Reynolds, private communication.
[30] S. Jaakkola, private communication.
 Laser Cooling and Trapping of
Neutral Atoms
Y. Castin, J. Dalibard and C. Cohen-Tannoudji
Laboratoire de Spectroscopie Hertzienne de VE.N.S.'f
et College de France
24, rue Lhomond
F-75231 Paris Cedex 5
France

Abstract
We present a simple review of the basic physical processes allowing one to
control, with laser beams, the velocity and the position of neutral atoms.
The control of the velocity corresponds to a cooling of atoms, that is, a re-
duction of the atomic velocity spread around a given value. The control of
the position means a trapping of atoms in real space. The best present per-
formances will be given, in terms of the lowest temperatures and the highest
densities. The corresponding highest quantum degeneracy will also be esti-
mated. It is imposed by fundamental limits, which will be described briefly .
We also give the general trends in this field of research and outline the new
directions which look promising for observing quantum statistical effects in
laser cooled atomic samples, but which are for the moment restricted by
unsolved problems.

1 Introducing the Simple Schemes

The radiative forces acting on atoms in a light field can be split into
two parts, a reactive one and a dissipative one. The dissipative force
(radiation pressure), which basically involves scattering processes, is ve-
locity dependent. We will see that this dependence leads to the Doppler
cooling scheme and to the concept of optical molasses, and we will give
the corresponding minimal achievable temperature. The dissipative force
can be made position dependent, through a gradient of the magnetic
field, so that the atoms are also trapped in the so-called magneto-optical
trap.

f Unite de recherche de l'Ecole Normale Superieure et de FUniversite Paris 6, associee au

CNRS

173
174 Y. Castin, J. Dalibard and C. Cohen-Tannoudji
1.1 Effect of Light on the Atomic Internal State
Since the atoms in laser cooling are illuminated by quasi-resonant light,
they can be considered as two-level atoms, with a metastable state
g, called "ground state" in what follows, and an excited state e. The
transition between g and e has a resonant frequency COA and an electric
dipole d. The excited state can decay by spontaneous emission with rate
F. In this section, we consider the simple case of atoms with no sublevels
in the ground state g. The influence of the presence of Zeeman sublevels
in g on laser cooling is the subject of Section 2.
The laser electric field is a superposition of travelling waves of fre-
quency coL close to COA and of complex amplitude SQ. An important
parameter is 5 = COL — COA, the detuning of the laser frequency CDL from
the atomic frequency COA. The dipolar coupling between atom and light
leads to a time dependent amplitude of transition from g to e, denoted by
(Q/2)e-i0JL\ and from e to g, (Q*/2) eia)Lt. The parameter Q = -dS0/h is
called the Rabi frequency. The amplitude of transition can be made time
independent by the following time dependent unitary transformation S(t)
changing the excited state:

S(t) = eiC0Ltle){el (1)

which gives the |g) — • \e) transition the effective Bohr frequency COA —
CDL = — S instead of <X>A> This leads to the time independent effective
hamiltonian ifeff for the evolution of the internal atomic state:
-d-iY/2 0/2 \
(2)
"72 0 )•
This matrix is given in the {|e),|g)} basis and the — iT/2 term accounts
for the instability of the excited state.
The effect of the coupling will be considered in the perturbative regime:

The dimensionless quantity s is called the saturation parameter and gives

the relative amount of time spent by the atoms in the excited state. In
this regime, one of the eigenvectors of //eff remains close to g, with the
eigenvalue h(d — iT/2)s/2. The corresponding real part hdf = hds/2 is
the signature of a reactive effect: it describes the shift of the energy of
the ground state by the incoming light (light shift). The imaginary part is
the signature of a dissipative effect: the ground state g becomes unstable
by contamination by the excited state through the Rabi coupling; the
 Laser Cooling and Trapping of Neutral Atoms 175
corresponding rate V = Ys/2 is the excitation rate of the ground state
by the laser light. Note that the following discussion can be generalized
to arbitrary laser intensities (see e.g. the so-called "dressed atom" picture
in [1]).

1.2 Effect of Light on the Atomic External State

In contrast to the internal atomic variables, which correspond to the
electronic motion relative to the nucleus, the external variables are related
to the motion of the atomic center of mass; the important ones for laser
cooling are the position r and the momentum p of the atom. These
external variables are also changed by the interaction with the laser light.
For example, the momentum p shifts to p ± hk after the absorption or
the emission of one photon of momentum hk. The corresponding atomic
recoil velocity hk/M, where M is the mass of the atom, is an important
parameter in laser cooling, as we shall see; it is of the order of 3 mm/s
for cesium. The effect of light on atomic motion can be described in
terms of radiative forces. These forces can be dissipative or reactive,
depending on the contribution or non-contribution of the momentum of
spontaneously emitted photons.
For the simple case of an atom in a plane wave, the dissipative force
corresponds to the usual radiation pressure; it is due to the absorption
of one laser photon of momentum hkL and the spontaneous emission of
one fluorescence photon of momentum hks. The corresponding average
change in atomic momentum is (<5p) = (hkL — hks} = hkL; the mean
contribution of hks is zero because spontaneous emission occurs with the
same probability in two opposite directions. The mean radiative force
is then simply hkL times the rate of absorption-spontaneous emission
cycles. This rate, of the order of Ts/2 at low saturation, cannot exceed
F/2 at high saturation. The dissipative force thus saturates at high laser
intensity to hkiJ'/2, a limit almost 104 higher than the gravitational force
Mg for cesium.
The reactive forces derive from an effective potential, which is the
position dependent light shift of the atomic level g; it is due to successive
absorption-stimulated emission cycles. Consider, for example, the light
shift U of the atomic ground state in the vicinity of a gaussian laser beam.
When the laser light is detuned to the red (5 < 0), U(r) is negative and
maximal in absolute value in the laser beam, and close to zero outside the
laser beam. The effect of U(r) leads to a trapping of the atoms around
the maxima of intensity of the laser light. This has been used to make a
176 Y. Castin, J. Dalibard and C. Cohen-Tannoudji
trap, at a detuning large enough for the dissipative effects (heating due to
spontaneous emission) to be negligible [2]. An improved version of the
optical trap, using stabilization by radiative pressure forces, is presented
in [3].

1.3 Doppler Cooling and the Doppler Limit

The Doppler cooling scheme has been proposed independently by Hansch
and Schawlow as well as Wineland and Dehmelt in 1975 [4, 5]. It relies
on the atomic velocity dependence of the radiation pressure due to the
Doppler effect. In this scheme, the cooling of the atomic velocity along a
given direction z is obtained by the superposition of two counterpropagat-
ing running waves along z. The two waves have the same weak amplitude
S'o and the same frequency coL, detuned to the red (5 = coL — coA < 0).
Consider an atom of velocity v, and call k the wave vector of the laser
wave copropagating with the atom (k • v > 0). In the atomic rest frame,
this wave has an apparent frequency coL — k • v < coL < O>A, SO it is moved
farther from resonance by the Doppler effect —k • v. The counterpropa-
gating wave, with wave vector —k, is on the contrary, is brought closer
to resonance by the Doppler effect (k • v > 0) (see Fig. 1). The atom will
then absorb photons preferentially in the counterpropagating wave, so
that it feels a mean radiative force opposed to its velocity along z. This
force vanishes for an atom at rest, and can be shown to behave as a
linear friction force:

Fz = -<xvZ9 (4)

where a is a friction coefficient, for sufficiently slow atoms. Cooling is

provided along the other directions x and y by use of two additional
standing waves. The corresponding light field acts as a viscous medium
on atomic motion, the so-called "optical molasses" [6].
Spontaneous emission plays an essential role in the cooling mechanism:
it allows the dissipation of energy by emission of fluorescence photons
with a frequency a>s > &>L- However, the random momentum recoil of
the atoms after spontaneous emission is responsible for a heating, i.e. an
increase of the mean atomic kinetic energy, which can be described, as in
standard brownian motion theory, by a momentum diffusion coefficient
D:
 Laser Cooling and Trapping of Neutral Atoms 177

k,

vww* CJ T

WWV* O *AAAAA
cj L -k L v co L +k L v

Fig. 1. The laser waves of the Doppler cooling configuration, in the laboratory
(a) and in the atomic rest frame (b).

This heating can counterbalance the effect of the damping force:

Jt (Pi)Icooiing = ~jy7 (Pi) > i = x,y,z (6)

so that equilibrium is reached. The corresponding stationary velocity
distribution is gaussian, with an effective temperature T given by Eintein's
law:
kB T = | , (7)

where ks is the Boltzmann constant. After an explicit derivation of a and

D for the two-level atom model, one finds that the minimal temperature
To, called the Doppler limit, is obtained for a detuning 5 = —T/2 and is
given by [7, 8]:

kBTD = —. (8)

It corresponds to To = 120 fiK for cesium atoms.

1.4 The Magneto-Optical Trap (MOT)

The effect of Doppler cooling is a compression in momentum space, but
it has no confining effect in real space, the atoms performing simply a
spatial diffusive random walk in the optical molasses. In order to trap
the atoms in real space, the radiation pressure is also made position
dependent via a gradient of magnetic field.
The principle of the MOT can be explained using the following model
(see Fig. 2). Consider an atomic transition occurring between a ground
178 Y. Castin, J. Dalibard and C. Cohen-Tannoudji

Fig. 2. The principle of the magneto-optical trap on a jg = 0 —• je = 1 transition.

state of angular momentum jg = 0 and an excited state of angular

momentum je = 1. Atoms are excited by two counterpropagating laser
waves which are, respectively , G+ and o- polarized along z. The fact that
the excited state has several Zeeman sublevels \e,m = — l) z , \e9m = 0) z
and \e,m = l) z of angular momentum mh along z is essential. First, these
sublevels undergo different Zeeman shifts due to the external magnetic
field, say — \iB, 0 and JXB, respectively, for a magnetic field B along the
quantization axis z, \i > 0 being proportional to the magnetic dipole
moment in the excited state. Second, the ground state is coupled to \e, 1)
only by absorption of photons in the <r+ laser wave, and to |e,— 1) only
by absorption of photons in the o- laser wave. Suppose now that B
varies linearly along z and vanishes in z = 0. An atom at rest in z = 0
sees no mean radiative force, the two laser beams being exactly equally
detuned far from resonance. In z > 0, however, because of the Zeeman
shifts, the resonance frequency for the |g,0) <—> \e,l) transition is larger
than coA, and is smaller than COA for the |g,0) <—• \e,—l) transition. The
a- laser wave is then closer to resonance than the a+ laser wave, and
the atom at rest feels a net radiative force pushing it towards z = 0.
The conclusion is reversed for negative values of z. In this configuration,
there is thus, in addition to the damping force, a trapping force, linear
with z around z = 0, so that the net mean radiative force along z reads:

Fz = —<xv z — KZ. (9)

The atomic motion is damped in an effective external harmonic potential,

 Laser Cooling and Trapping of Neutral Atoms 179
and the atomic cloud develops a finite spatial extension given in steady
state by:
2
K(z )=kBT = ^. (10)

The first experimental demonstration of the MOT has been reported

in [9].

2 Beating the Doppler Limit

Since 1988, precise experimental measurements of the temperatures of
optical molasses have been performed [10-16]. The lowest temperatures
are found to be well below the Doppler limit (8). For a heavy atom such
as cesium, the lowest measured temperatures are of the order of 3 /zK,
a factor of 40 smaller than the Doppler limit, and are observed at large
atom-laser detuning: \8\ > F, instead of 3 = — F/2 expected from the
Doppler cooling theory. All these results are experimental evidence of
very efficient new cooling mechanisms.
These mechanisms have been identified theoretically on simple one-
dimensional (ID) models [17, 18]. They rely strongly on the existence
of several degenerate sublevels in the atomic state g, a feature left out
in the two-level atom model used for Doppler cooling but present in
experimental optical molasses, because of the hyperfine and Zeeman
structures of the atomic ground state. A second important feature
is the spatial variation of the laser field polarization on the optical
wavelength scale, a condition which is automatically fulfilled in three-
dimensional (3D) laser configurations, and which is obtained in the
simple one-dimensional models by giving different polarizations to the
two counterpropagating waves.
In this section, we will analyze in some detail only one of the presently
known polarization gradient cooling mechanisms, relying on the so-called
"Sisyphus effect" [17]. The corresponding minimal kinetic energies no
longer scale as hT, as for Doppler cooling, but as the recoil energy
ER = h2k2/2M, i.e. the mean increase of the atomic kinetic energy after
the spontaneous emission of a single photon. Note that ER is well below
the Doppler limit hT/2 for the typical atomic transitions used in optical
molasses. In the case of cesium atoms, for example, hT/2 is one thousand
times larger than the recoil energy. In this very cold regime, it has been
predicted theoretically and checked experimentally that quantization of
atomic motion plays an important role in Sisyphus cooling. This has
180 Y. Castin, J. Dalibard and C. Cohen-Tannoudji

0
/ A/8 A/4 3A/8 A/2 z
y/
Fig. 3. Resulting electric field in the x-y laser configuration.

led to a new picture for the laser cooled atomic samples, the so-called
"optical lattices".

2.1 Optical Pumping and Light Shifts

As seen in Section 1.1 for the two-level atom case in the low saturation
regime the effect of light on the internal atomic state is to shift the energy
levels (reactive effect) and to give a finite lifetime to the ground state
(dissipative effect). When the atom has several Zeeman sublevels in the
ground state, the reactive effect can lead to different light shifts of the
sublevels, and the dissipation effect can lead to real transitions between
the various sublevels, through optical pumping cycles [19].
The last two features play an important role in laser cooling. They are
present in the simple one-dimensional scheme depicted in Fig. 3. The light
field is obtained by superposition of two coherent counterpropagating
running waves along z, with the same amplitude g$ and frequency coL, but
with orthogonal linear polarizations. For an appropriate phase choice,
the total electric field is given by:
g(z9t) = £+{z)e~i(OLt + c.c (11)
+ ikz ikz
<f (z) = £0[exe -ieye- ], (12)
where e x , e^ are unit vectors along x and y. One can check that the
polarization of &(z, t) depends on z, with a periodicity kov\/2, where /Lopt =
2n/k is the optical wavelength. For example, at z = 0, A opt/2,..., $(z,i) is
G- circularly polarized along z; it is a+ polarized at z = A opt /4, 3A opt /4,...;
it is linearly polarized along (e x — e^) jyjl at z = /l op t/8, 52 op t/8,... and
along (ex + ey) /y/2 at z = 3/l opt /8, 7A opt /8.
Consider an atom with an angular momentum jg = 1/2 in the ground
 Laser Cooling and Trapping of Neutral Atoms 181

J=3/2

(a)
Fig. 4. (a) Atomic-level scheme and intensity factors (square of the Clebsch-
Gordan coefficients) for a jg — 1/2 —• je = 3/2 atomic transition, (b) In the
Sisyphus cooling configuration, the position-dependent light shifts of the ground
state sublevels and a typical atomic trajectory in the position-energy space.

state and an angular momentum je = 3/2 in the excited state. This atom
is illuminated by the previous field configuration, with a laser frequency
a>L below the resonance frequency COA :

8 = COL — coA < 0 (13)

The transition amplitudes between the atomic energy levels of a given
angular momentum along z, {|g,m), m = ±1/2} and {|e,ra'),mr =
±3/2, ±1/2}, by absorption or stimulated emission of laser photons
are now weighted by the appropriate Clebsch-Gordan coefficients (see
Fig. 4(a)) and thus depend on the polarization of the electric field.
Where the light is purely o- polarized, the |g, —1/2) sublevel is coupled
to the electric field with an amplitude y/3 larger than the |g,+1/2)
sublevel. It therefore experiences a light shift three times larger, hds
instead of hds/3 for | g , + l / 2 ) , with a saturation parameter given by (3).|
In the presence of pure <r_ light, the sublevel |g,—1/2) not only has a
well defined light shift, but is also a trap for the internal atomic state.
It is coupled by laser light only to the \e,— 3/2) excited state sublevel,
which by spontaneous emission can only decay to |g,—1/2). If the atom
is initially in the |e, +3/2) excited sublevel, it is eventually decays to the
t At first sight, there seems to be a discrepancy of a factor of 2 between the light shift
(and also the excitation rate) derived here and the one given in Section 1.1. In fact, at
z = 0, the electric field amplitude is y[2$Q> instead of SQ in Section 1.1.
182 Y. Castin, J. Dalibard and C. Cohen-Tannoudji
\g, +1/2) ground sublevel by spontaneous emission. If the atom is initially
in the {|g,+l/2),|e,—1/2)} manifold, it is put back in |g,+l/2) after a
spontaneous emission with probability 1/3 (G- emission) and is trapped
into \g, —1/2) with probability 2/3 (linearly polarized emission along z).
The population in |g,+1/2) therefore tends exponentially to zero for an
increasing number of fluorescence cycles. The corresponding transition
rate from |g,+1/2) to |g, —1/2) by the optical pumping mechanism is
found to be Ts • 1/3 • 2/3 = 2Ts/9.
When the light is purely cr+ polarized, the situation is reversed between
the sublevels. The atoms are now excited by optical pumping to the
|g,+l/2) sublevel, which undergoes the largest light shift. The present
configuration therefore exhibits a strong correlation between the spatial
modulation of light shifts and the spatial modulation of optical pumping
rates. For a negative atom-laser detuning <5, the optical pumping rates are
the largest, from the highest energy ground state sublevel to the lowest
one.

2.2 Sisyphus Cooling Mechanism

We now explain how cooling occurs in the previously described atom-
laser configuration. The key point is that the position dependent light
shifts of the sublevels |g,+1/2) appear as effective potentials U±(z) for
a moving atom. It will thus be convenient to introduce the mechanical
energies E± = Mv2/2 + U+(z), where v is the atomic velocity along z.
A typical trajectory in (z,E+) space is shown in Fig. 4(b). It starts
in the sublevel |g, —1/2), in a minimum of U-(z\ with a kinetic energy
larger than the modulation depth:

Uo = -^hds (14)

of the potential. If the optical pumping time TP from |g, —1/2) to

|g, +1/2) is long enough, the atom can climb the potential hill and reach
the top of U-(z) before changing its internal state. During this part of
the motion, the mechanical energy £_ remains constant and there is a
conversion of kinetic energy into potential energy. In the vicinity of the
maxima of U-(z)9 the laser light is o+ polarized and the atom has the
maximal probability to be optically pumped into the |g,+1/2) sublevel
by absorption of a laser photon followed by a spontaneous emission. If
such a process takes place, the atom is goes back into a valley, but for
the C/+(z) potential this time. The resulting decrease of potential energy
 Laser Cooling and Trapping of Neutral Atoms 183
leads to a corresponding excess of energy for the spontaneously emitted
photon with respect to hcoL. Dissipation of energy thus originates in this
cooling scheme from the spontaneous emission of anti-Stokes Raman
photons.
Thus the atoms do not stop climbing potential hills, as Sisyphus did
in Greek mythology, until their kinetic energy becomes of the order of
the modulation depth UQ. This intuitive reasoning is confirmed by a
theoretical analysis, similar to the one used already for Doppler cooling,
based on introducing a friction coefficient a and a momentum diffusion
coefficient D, as in brownian motion theory. The derivation of the
coefficients D and a leads to the equilibrium temperature given in [17]:

kBT = %^oc U0~-h8s. (15)

M

23 Limit of Sisyphus Cooling

The equilibrium temperature given in the previous section is proportional
to the laser intensity, and scales as 1/|<5| for large detuning (|<5| > F). This
has been confirmed experimentally in 3D on a wide range of detunings
and laser intensities, and for several atomic species [15, 16]. The precise
value of the slope for the 3D experiments cannot of course be deduced
from the simple ID model that we described. Actually, such a scaling law
for Sisyphus cooling cannot be realistic for any values of the parameters.
For example, it leads to a vanishing equilibrium temperature in the limit
of vanishing laser intensity. In this regime of very small light shifts, the
experimental results do indeed deviate from (15).
In order to have a more complete understanding of Sisyphus cooling,
one has to take into careful account the heating of the atoms due to
random recoil after spontaneous emission. The Sisyphus effect, corre-
sponding to a decrease of the potential energy of the order of UQ after
an optical pumping cycle, is then counterbalanced by an increase of the
kinetic energy of the order of the recoil energy ER = h2k2/2M. Cooling
works only when UQ is above a threshold of a few ER, and the minimal
achievable temperature is expected from (15) to scale as the recoil energy
ER.
The precise value of the Sisyphus cooling limit has been derived from
a quantum treatment of both the internal atomic state and the atomic
motion in laser light, for the simple ID model presented in [20]. For a
given atom-laser detuning <5/F, the dependence of the steady state kinetic
energy on the modulation depth UQ of the potential is shown in Fig. 5.
184 Y. Castin, J. Dalibard and C. Cohen-Tannoudji

1000

Fig. 5. In a full quantum treatment of Sisyphus cooling, mean kinetic energy

(p2)/2M in units of recoil energy ER = h2k2/2M as a function of the modulation
depth l/0 of the optical potential wells in units of ER. The atom-laser detuning
is S = ST.

The scaling law kBT ~ Uo is recovered in the limit of large U0/ER. The
existence of a threshold on Uo corresponds to a rapid increase of the
average kinetic energy, when UO/ER becomes too small. The smallest
possible value of the root mean square atomic momentum is of the order
of 5.5ft/c,reached for Uo ~ 100ER in the limit of large detunings \S\ > T.

2.4 Quantization of Motion in the Optical Potential Wells

It has been found to be fruitful to try to obtain more insight into the
coldest regime of Sisyphus cooling. One can check that in this regime
the following inequality holds:
> 1, (16)
where QOsc ~ k^/Uo/M is the oscillation frequency of the atoms in the
bottom of the potential wells and TP ~ 1/Ts is the optical pumping time.
This leads to a level spacing ~ hClosc for the quantized atomic motion in
U±(z)9 much larger than the radiative width h/xp of these levels due to
optical pumping.
Therefore, a good basis for the analysis of the cooling is provided by the
stationary solutions of the Schrodinger equation for the atomic motion in
U±(z) [21]. Since the potentials U±(z) are periodic, of period /lopt/2, the
corresponding energy spectrum has a band structure. The lowest energy
bands, corresponding to quasi-harmonic motion in the bottom of the
potential wells (see Fig. 6 taken from [22]), have a negligible tunnelling
width. On the contrary, the energy gaps become very small for the quasi-
 Laser Cooling and Trapping of Neutral Atoms 185

Fig. 6. Band structure of the energy spectrum for the quantum motion in U+(z)
optical potential wells. The vibrational levels are labelled v = 0,1,.... The dark
circular areas are proportional to the steady state populations of the vibrational
levels.

free motion well above the potential hills. The effect of optical pumping
in this basis can be described simply by transition rates between the
various energy levels, when the inequality (16) holds. The corresponding
rate equations allow one to calculate the steady state populations of the
energy bands. The maximal population in the most populated band v = 0
is found to be 0.34 in [21], for Uo ^ 60ER.
The existence of well resolved external quantum levels in ID Sisyphus
cooling has been demonstrated experimentally. A probe laser beam, with
a very small intensity, is sent through the optical molasses along the
direction of the laser cooling beams. Two side-bands are clearly observed
on the probe absorption spectrum, for a probe frequency cop = COL ±^OSC-
They correspond to stimulated Raman transitions between two successive
vibrational levels in the potential wells. Since the lowest level is the most
populated one, Raman cycles starting with atoms in the state v = 0
are expected to be dominant. When cop is below coL (cop ~ coL — £losc\
the dominant process is therefore the absorption of one photon in the
cooling beams followed by stimulated emission of one photon in the
probe beam, and the probe beam is amplified. When cop is above COL
(cop ~ coL + Qosc), the Raman processes are mainly the absorption of
one photon in the probe beam followed by the stimulated emission of
one photon in the cooling beams, and the probe beam is absorbed (see
Fig. 7 taken from [22]). The so-called "overtones" (Raman transitions
186 Y. Castin, J. Dalibard and C. Cohen-Tannoudji

Fig. 7. Examples of Raman processes induced between the atomic vibrational

levels by a weak probe beam of frequency cop passed through the one-dimensional
optical molasses. The populations of the energy levels are represented by dark
areas. In (a), the Raman processes lead to an amplification of the probe beam
(cop ~ coL — Qosc)- In (b), the Raman processes lead to an absorption of the probe
beam (cop ~ coL + Oosc).

with Sv = 2, dv = 3,...) can also be identified. A stimulated Rayleigh

line is also observed at CDP = coL ; it can be shown to give evidence for a
spatial antiferromagnetic order of the atoms [23, 24]. These experiments
are the first observations of quantized external levels of a neutral atom in
an optical field. An important point is that about half of the atoms are
expected to be in minimum uncertainty states [22]. These observations
have been confirmed by studies of the fluorescence spectra of ID optical
molasses [25].

2.5 Optical Lattices in 2D and 3D

In the preceding sections, we have seen that ID Sisyphus cooling in a
laser field can lead to a strong accumulation of the atoms around the
points where the field is circularly polarized. Such a property can be
generalized to two-dimensions (2D) and three-dimensions (3D).
Theoretical results are available in 2D [26, 27]. The quantum approach
is similar to the one used in ID. In this paper, the main results concern
the population of the energy levels in steady state. They are found to be
strongly reduced: the maximum population of the deepest band shifts
from 0.34 in ID to 0.09 in 2D. This does not mean that cooling is less
 Laser Cooling and Trapping of Neutral Atoms 187
efficient in 2D, but it is simply a signature of degeneracy effects among
the vibrational levels. Let us indeed consider the atomic motion in the
bottom of the 2D potential wells. It is found to be almost harmonic, and
the energy bands are labeled by the two quantum numbers nx and ny of
the corresponding 2D isotropic harmonic oscillator. If the motions along
x and y are considered as decoupled, the probability of finding the atoms
in the ground state \nx = 0,ny = 0) of their motion is the product of the
probabilities of finding them in the ground states \nx = 0) and \ny = 0)
of their motion along x and y respectively. This simple reasoning leads
to the estimate:

n^ = o ^(njl£ o ) 2 , (17)

not far from the exact results.

To obtain experimental evidence for vibrational levels in o+ optical
potential wells, one has to eliminate phase fluctuations between the
orthogonal running waves in order to avoid distorsion of the electric
field polarization away from <r+ in the minima of the wells. This has
been realized in Munich, with two and three orthogonal standing waves
with well controlled phases [28, 29]. In Paris, an alternative to this
approach has been developed. The idea is to use in riD the minimal
number of running waves leading to cooling, which is n + 1. In this
case, as is evident for the particular case n = 1, the periodic pattern of
potential wells is only shifted by phase fluctuations. A hexagonal lattice
in 2D, and a cubic body centered lattice in 3D, have been obtained and
have been analyzed by a probe beam [30].
All these results lead to a completely different physical picture of the
laser cooled atomic samples. The concept of optical molasses is replaced
by the one of optical lattices, which consist of periodic arrays of micro-
traps, with vibrational levels for the atomic motion. This description is
well suited to the discussion of statistical effects, the quantum degeneracy
(i.e. the population of the most populated state of the one-atom density
matrix) being derived simply from the population of the lowest energy
band. However, these optical lattices are for the moment quite empty,
only a small percentage of the sites being occupied. In order to increase
the quantum degeneracy, one has to increase the atomic density, using
an extra trapping mechanism.
188 Y. Castin, J. Dalibard and C. Cohen-Tannoudji
2.6 Experimental Results from the Magneto-Optical Trap

The geometry of the magneto-optical trap has been described via a simple
ID model in Section 1.4. It can be extended to a 3D configuration in the
following way. The magnetic field gradient is provided by two coaxial
coils travelled by opposite electric currents. Close to r = 0, where the
resulting magnetic field is vanishing, one has the approximate spatial
dependence for B:

(18)

Three pairs of counterpropagating o+ and a_ circularly polarized laser

waves, along x, y and z, provide 3D trapping and cooling in the presence
of the magnetic gradient. Note that the laser beams propagating along z
and perpendicularly to z have opposite helicities because magnetic field
gradients are of opposite signs along z and x, y.
Experimental studies of the magneto-optical trap for atoms in a cell
have provided evidence for two different regimes (see, for example, [31]
for cesium atoms).
First, a low density regime, with a density n less than 1010 atoms/cm3,
is obtained when a small number N of atoms (up to 106) are present in
the trap. The important point is that the temperatures in the trap are the
same (down to 3 JJK) as the ones measured in the o+—o- optical molasses
in the absence of magnetic field. The sub-Doppler cooling mechanisms,
which rely strongly on the light shifts of the various Zeeman ground
sublevels, are thus not destroyed by the magnetic field in the trap [31-
34]. This is due to the fact that the trapping mechanism accumulates the
atoms in a small region around B = 0. The corresponding minimal radii
of the atomic cloud in the trap are found experimentally to be of the
order of 40 fim for a magnetic field gradient of 10 gauss/cm.
When a large number N of atoms are present in the trap (up to 3-1O8),
a high density regime is observed, with maximal densities n around
3 • 10 12 atoms/cm 3. However, the temperatures in the trap are then
found to be higher than the ones in the (7+ — <7_ optical molasses, the
corresponding excess of temperature AT scaling simply as iV1/3 for a
given laser detuning and intensity. The highest quantum degeneracies in
the trap are actually not obtained at the highest densities.
In order to estimate this quantum degeneracy, we suppose that the
mean number of atoms in the most populated quantum levels inside the
 Laser Cooling and Trapping of Neutral Atoms 189
trap is given by:
h
n m a x = nAj^ with A DB = ,. A/f/ . (19)

This formula, valid only for a gas of non-interacting particles in thermal

equilibrium in the low degeneracy regime, is found to be in reasonably
good agreement, when transposed to the 2D case, with the theoretical
results mentioned in Section 2.5. At moderate density n = 8 • 10 11
atoms/cm 3 and quite low temperature T = 10/iK, a situation which is a
compromise between the search for low temperatures and high densities
in the trap, equation (19) leads to a degeneracy n m a x of the order of
io-4.
We now give a qualitative account of the collective effects respon-
sible for the limits on the maximal densities n and the corresponding
temperatures T achievable in the magneto-optical trap.
At long range, the exchange of fluorescence photons between atoms
leads to a repulsive force, proportional to the flux of photons emitted
from one of the atoms through a fictitious surface located around the
other atom and of area crA, the resonant absorption cross section. It thus
scales as l/r\2, where r^ is the distance between the two atoms. This
force has a mean antitrapping effect, which limits the compression in real
space of the atomic cloud by the trap. It also fluctuates, which introduces
extra heating and increases the temperature [35, 36]. Fortunately, the
effect of this long range interaction should be reduced at high laser
detuning.
At short range (i.e. for ryi smaller than the optical wavelength /Lopt),
the dipolar interaction between an atom in excited state and an atom
in ground state has non-negligible effects. It shifts the energy levels by
an amount of the order of the natural width HT of the excited state, for
7*12 ~ ^opt/2ft. Such a shift is larger than the light shifts leading to the low
laser intensity Sisyphus effect and can destroy the cooling mechanism. At
high atomic densities, it can lead to the formation of molecules, with a
loss term of atoms from the trap scaling as n2 [37, 38].
The existence of such strong atomic interactions in the presence of laser
light thus leads to the following intuitive conclusion, that the density in
the magneto-optical trap cannot exceed a value such that the mean
distance between atoms becomes smaller than Aopt? which leads to a
maximal density n given by

<, t ~ 1. (20)
190 Y. Castin, J. Dalibard and C. Cohen-Tannoudji
This corresponds to n ~ 2 • 1012 atoms/cm 3 on cesium, of the order of
the experimental results.

3 Beating the Recoil Limit

One of the conclusions of the previous section was that it seemed diffi-
cult, in the presence of quasi-resonant laser light of wavelength lopU to
achieve atomic densities n so high that n2?opt > 1, while keeping the very
low temperatures realized at low densities. In the search for quantum
degeneracy, the regime where the atomic DeBroglie wavelength ADB is
much larger than the optical wavelength AOpt is thus particularly inter-
esting. In this case, the parameter nX^ can be on the order of unity,
with n2?opt still being very small, so that collisional effects between atoms
should remain small.
This interesting regime is called "subrecoil", because the condition
^DB = h/Ap > AOpt implies an equilibrium temperature T = A/? 2 /M/CB
much smaller than the so-called recoil temperature TR = h2k2/Mk^
where k — 2n/Xopt. The problem is that subrecoil temperatures cannot be
reached by the "usual" cooling scheme described in the previous sections.
In these schemes, indeed, spontaneous emission of fluorescence photons,
which is essential for carrying away the atomic mechanical energy, never
stops. The atoms thus continuously experience random recoils of size
hk in momentum space, so that the equilibrium temperature cannot be
much smaller than the recoil temperature.
In Section 3.1, we present a ID subrecoil cooling scheme called
"velocity-selective coherent population trapping" [39, 40]. It is not the
only one in the laser cooling field. Another scheme has been shown
experimentally to beat the recoil limit [41], and there are also several
theoretical proposals [42, 43]. The general idea of all these schemes is,
however, the same: it consists of accumulating the atoms into states that
are very weakly coupled to laser light and with a very narrow width in
momentum space. The fundamental limits of subrecoil cooling addressed
in Section 3.3 can thus be considered as being quite general.

3.1 ID Velocity-Selective Coherent Population Trapping

Consider the so-called "A type" atomic system illuminated by two mu-
tually coherent laser waves, of opposite wave vectors kez and —/cez along
z and of same frequency O>L, but with orthogonal circular polarizations,
respectively a+ and G- (see Fig. 8). We describe in a quantum manner
 Laser Cooling and Trapping of Neutral Atoms 191

Fig. 8. The A atomic system in the 0+—c_ laser configuration. In the absence
of spontaneous emission, note the existence of closed families of atomic states
labelled by p, the atomic momentum along z in the excited state.

the atomic motion along z by introducing the external states \p) corre-
sponding to the plane waves of momentum pez. The atomic dynamics
can then be analyzed as follows [40].
In the absence of spontaneous emission, an atom initially in the state
\eo,p), i.e. in the internal (electronic) state |eo) and with a center of
mass motion of momentum p along z, is coupled only to |g_,p — hk)
(by stimulated emission of a photon in the a+ wave) with an amplitude
a+, and to \g+,p + hk) (by stimulated emission of a photon in the
<j_ wave) with an amplitude a_. The space of atomic states can then
be split into a direct sum of an infinite number of closed families
^(p) = {\eo,p),\g->P — bk)>\g+>P + hk)} labelled by p. Since there is
only one internal excited state \eo) for two ground levels |g+), we can
form, inside each family ^(p), a particular linear combination \ipwc(p))
uncoupled to the laser:

[ l J f c ) | + ftfc)] (21)

Note that this combination is not a stationary state, because it has no

well defined kinetic energy:
(p-hk)2^(p + hk)\ (22)
except for the particular case p = 0, for which |T/;NC) is a perfectly
trapping state, called the "dark state".
Let us now take into account spontaneous emission. Starting from
|eo,p), the atom now eventually decays into \g±,p — pf), where pf is
the momentum of the fluorescence photon along z and is randomly
distributed between —hk and +hk. Spontaneous emission thus couples
family ^(p) to the "neighbouring" families #X//)> with pf = p+hk—p^ in
the range \p — 2hk,p + 2hk]. This random change of p after a fluorescence
cycle may transfer the atoms into the dark state |t/>Nc(0)) or into a quasi-
dark state for p very close to 0, where they remain trapped and where they
192 Y. Castin, J. Dalibard and C. Cohen-Tannoudji
pile up. The corresponding effect on the atomic momentum distribution
is to produce two peaks at p = ±hk, because |V?NC(0)) is a superposition
of states of momenta +ftk and — hk. The width of these peaks is not
limited by the recoil momentum hk, but scales as 1/y/i, where t is the
atom-laser interaction time, as has been checked numerically in [40]. For
an idea of the proof and an analysis of the height of the peaks as a
function of time, we refer to [44].

3.2 Experimental Results Below the Recoil Limit

The ID cooling scheme introduced in Section 3.1 has been demonstrated
experimentally on metastable helium [39], on a transition between a
ground state of angular momentum jg = 1 and an excited state of angu-
lar momentum je = 1. To see why the A structure of Section 3.1 is an
acceptable model in this geometry, let us introduce the internal atomic
states of angular momentum mh around z, {|g,m)z, |e,m)z,m = 0,±1}.
Since the laser field has no component along z, the V system formed by
{|e,-l) z ,|g,0) 2 ,|e,l)z}, and the A system {|g,-l) z ,|e,0) z ,|g,l) 2 }, are not
coherently coupled. The key point then is that the excited sublevel of
the A system \e,0)z cannot decay by spontaneous emission into |g,0)2,
because the corresponding Am = 0 Clebsch-Gordan coefficient is van-
ishing. Therefore, after a few spontaneous emissions from the excited
sublevels of the V system, \e,+l)Z9 the atoms are pumped into the A
system, from which they cannot escape.
Experimental results on the atomic momentum distribution after an
interaction time t = 350F"1, where hT is the natural width of the excited
state, are shown in Fig. 9. The two peaks in ±hk are visible above
the initial momentum distribution. Their half width at half maximum
dp allows one to define an effective temperaturef Teff of the order of
0.5 recoil temperature TR. Note that the velocity selective coherent
population trapping can be generalized on a jg = 1 —• je = 1 transition
to multidimensional cooling [40, 45, 46]. A 2D experiment on helium is
in progress in Paris.
As mentioned in the introductory remarks of this section, another
subrecoil cooling scheme has been demonstrated experimentally [41].
Stimulated Raman transitions, between two hyperfine ground states |1)
and 12), are produced by two counterpropagating synchronized laser
t More precisely, Sp is defined as the half width of the peaks at the relative height e~~1//2,
so that the equation Teff = Sp2/M defining Teff as a function of Sp is exact for a
Maxwell-Boltzmann distribution.
 Laser Cooling and Trapping of Neutral Atoms 193

N(p)

Fig. 9. Experimental results on the momentum distribution for ID subrecoil

cooling. The two peaks of the final distribution in ±hk (solid line) are above the
initial distribution (dashed line).

pulses and provide a high selectivity in atomic velocity. The atoms whose
velocity v meets the Raman resonance condition v = vo are transferred
into 12) by the pulses and are then "pushed" towards zero velocity back
into |1) through a pumping resonant traveling wave. Successive cooling
cycles of this type are then performed, with a scanning of vo from high
velocities towards low velocities, in order to collect most of the atoms.
Since vo is always different from 0, there is a small region around v = 0
where the atoms never meet the Raman resonance condition, and where
they pile up. Very sharp momentum distributions around p = 0 have thus
been obtained and observed in ID, with effective temperatures down to
TR/10.

3.3 The Limits of Subrecoil Cooling

In order to obtain observable effects depending on the quantum statistics
of the atoms, it is necessary to provide three-dimensional (3D) cool-
ing. For the moment, there is no experimental evidence of an efficient
velocity-selective coherent population trapping in 3D or even in 2D. In
the low density regime, new theoretical methods have been developed to
try to predict the evolution of the effective temperature as a function of
the interaction time [44]. They rely on Monte-Carlo wave function sim-
ulations, describing both the internal and external atomic state quantum
mechanically. This procedure has, first, the advantage of using much
smaller objects than the full atomic density matrix p in 2D or 3D (a
discretized wave function has Jf components, whereas p has Jf1 com-
ponents). Since it deals with stochastic objects, it has also allowed a
connection with efficient tools of statistical physics, the so-called Levy
194 Y. Castin, J. Dalibard and C. Cohen-Tannoudji
flights [47]. The maximal atomic density that one can achieve without de-
stroying the darkness of the trapping state by multiple photon scattering
between atoms is still an open question.
The subrecoil cooling schemes also have to face the severe problem of
gravity. Indeed, the atoms in the trapping state feel no radiative force,
so that the gravitational force can remove them from the dark state and
let them fall down! A possible solution to this problem is to try to
counterbalance gravity by a non-dissipative force by using, for example,
the non-resonant dipolar trap mentioned in Section 1.2. One can also
perform cooling in a free falling frame. This has been done already, for
ordinary optical molasses, in an airplane in parabolic flight [48].
A different and promising type of solution is adressed in the next
section. The idea is to use gravity as an ally for making a trap, instead
of trying to eliminate it.

4 What Next?
As explained in Section 3, the subrecoil cooling schemes known at present
cannot be applied efficiently in 3D because of gravity. In this section, we
explain how to make an atomic cavity by using gravity and a parabolic
atomic mirror (Section 4.1). A quantum picture for the atomic motion
inside this cavity is briefly given in Section 4.2: as light waves do in usual
laser cavities, matter waves have well defined modes in the gravitational
cavity, and one may hope to obtain quantum statistical effects if several
atoms occupy the same quantum modes.

4.1 A Gravitational Atomic Cavity

The realization of an atomic cavity requires non-dissipative bounding of
the atomic motion in all directions.
Bounding of vertical motion can be obtained through a horizontal
plane atomic mirror. The principle of such a device has been given
in [49]. By total internal reflection of a travelling laser wave inside a piece
of glass, an evanescent wave is produced in the vacuum above the glass,
with an intensity I(z) that is exponentially decreasing with increasing
z. When the laser frequency COL is larger than the atomic resonance
frequency COA, the atomic ground state experiences a positive light shift
in vacuum, which is a decreasing function of z and thus gives rise to a
repulsive potential for the atoms. The laser is detuned far from resonance
so that dissipative effects during reflection are small:fluorescencecycles
 Laser Cooling and Trapping of Neutral Atoms 195

Fig. 10. Principle and loading of the atomic gravitational cavity.

indeed have a rate scaling as I/(COL — &>A)2- However, COL has to be not
too far from resonance so that the laser has a sufficiently high intensity to
provide reactive effects (i.e. light shifts, proportional to //(COL — &>A)) able
to reflect the atoms. This last condition is very restrictive for thermal
atomic velocities, and in this case only reflection for grazing atomic
incidence is observed [50]. On the contrary, it is easily fulfilled using
laser cooling techniques, and reflection under normal atomic incidence
can then be obtained [51].
In order to bound the transverse atomic motion also, the plane mir-
ror of the previous scheme is replaced by a parabolic mirror, with an
evanescent wave still present at its surface. The atoms are released with
negligible initial velocities from a switched-off magneto-optical trap and
bounce off the evanescent wave (see Fig. 10). The corresponding clas-
sical orbits are paraxial and are shown to be stable [52]. Experimental
realizations are in progress in Paris (at the ENS) and in Gaithersburg
(at NIST). Recent observation of about ten bounces has been reported
in Paris for cesium atoms with the following experimental evidence. At
time t after the drop of the atoms, a probe beam is sent through a
small volume d V around the initial position of the atomic cloud, and the
fluorescence light is collected. This procedure gives a signal proportional
to the number N(t) of atoms in the volume dV. Such a measurement
is destructive because atoms are pushed away and heated by the probe
196 Y. Castin, J. Dalibard and C. Cohen-Tannoudji

200
Time in ms
Fig. 11. The number of atoms in the gravitational cavity in Paris is measured
as a function of time using thefluorescenceinduced by a probe beam (see text).
The points on the diagram give the number of atoms in the probe beam for
different times after their release. The curve is a fit calculated by a Monte-Carlo
simulation. The successive bounces are labelled 1,2... .

beam so that the whole process has to be restarted from the beginning if
one wishes to obtain N(tf) at a different time t''. The result of numerous
such cycles is shown in Fig. 11 (see [53]; less recent results, with only
four bounces, are given in [54]).

4.2 Quantum Degeneracy Effects in the Cavity

When the atomic motion is treated quantum mechanically, the dynamics
in the cavity is analogous to that of light in an optical cavity. The cor-
responding quantum modes for the matter waves have been investigated
theoretically [52]. They correspond mathematically to the stationary so-
lutions of the Schrodinger equation for the atomic wave function \p(r,t):
h2
ihdt\p = ~2M^V+ (23)

tp(T9t) = (j>(r)e-iEt/h (24)

 Laser Cooling and Trapping of Neutral Atoms 197

Fig. 12. The calculated probability distribution in real space of a mode in the
gravitational atomic cavity. The cross corresponds to the focus of the parabolic
surface of the mirror.

with appropriate boundary conditions, responsible for the discrete nature

of the allowed energies E. In an idealized model, the atomic mirror is
considered a perfect one and \p is required to vanish on the surface of the
mirror. The problem is then readily separated in parabolic coordinates,
so that one can solve separately for "transverse" and "vertical" motions.
As an illustrative example, a vertical cut of the probability distribution
in real space |</>(r)|2 is given in Fig. 12. The numerous horizontal node
lines indicate a highly excited vertical motion. On the other hand, the
transverse motion is in the ground state. Estimates of the density of
modes in the cavity for realistic experimental conditions are discussed
in [52].
The basic idea to obtain statistical effects in the gravitational cavity is
simply to put several (bosonic!) atoms in the same quantum mode. It
appears to be more promising to try to populate a highly excited mode
rather than the ground state. Such a mode has a much larger spatial
volume, so that the atomic density in real space is lower and collisional
effects are reduced for a given quantum degeneracy. The method of filling
one particular mode of the cavity is still a difficult open question. One has
to find efficient selective processes, and the subsequent atomic dynamics,
far from thermal equilibrium, should be close to that of photons in a
laser cavity, rather than to that of Bose-Einstein condensation.
198 Y. Castin, J. Dalibard and C. Cohen-Tannoudji
5 Conclusion
In this article, we have given a brief survey of what are currently quite
standard techniques in the laser cooling field. Ultracold atomic samples
are obtained in optical molasses, optical lattices and in the magneto-
optical traps, but quantum degeneracy remains small, for the following
fundamental reasons. Collective effects, such as the dipolar interaction
between atoms in ground and excited states, prevent us from reaching
densities n much larger than l/A?opV where Aopt is the optical wavelength.
On the other hand, the spontaneously emitted photons limit the maximal
atomic de Broglie wavelength ^B to a few 2 opt /10, so that nX^B should
not be much larger than 10~3.
We have also described some of the new potential offered by laser
manipulation of atoms in the search for quantum degeneracies.
First, subrecoil cooling could beat the previous simple reasoning. It
leads to de Broglie wavelengths larger than the optical wavelength, so
that condition nX^B ~ 1 could be satisfied even though nl3opt < 1.
However, one has to find a way to counterbalance gravity, and more
detailed investigations of the effect of multidimensionahty are required.
Second, the search for degeneracy effects among the highly excited
states of a gravitational atomic cavity is a challenging alternative to
Bose-Einstein condensation. It should have the advantage of reducing
the collisional constraints. But one first has to find an efficient way of
selecting such excited modes!

References
[1] J. Dalibard and C. Cohen-Tannoudji, J. Opt. Soc. Am. B 2, 1707
(1985).
[2] S. Chu, J. Bjorkholm, A. Ashkin and A. Cable, Phys. Rev. Lett. 57,
314 (1986).
[3] P. Gould, P. Lett, P. Julienne, W.D. Phillips, W. Thorsheim and J.
Weiner, Phys. Rev. Lett. 60, 788 (1988).
[4] T. Hansch and A. Schawlow, Opt. Commun. 13, 68 (1975).
[5] D. Wineland and H. Dehmelt, Bull. Am. Phys. Soc. 20, 637 (1975).
[6] S. Chu, L. Hollberg, J. Bjorkholm, A. Cable and A. Ashkin, Phys.
Rev. Lett. 55, 48 (1985).
[7] D J . Wineland and W.M. Itano, Phys. Rev. A 20, 1521 (1979).
[8] J.P. Gordon and A. Ashkin, Phys. Rev. A 21, 1606 (1980).
 Laser Cooling and Trapping of Neutral Atoms 199
[9] E.L. Raab, M. Prentiss, A. Cable, S. Chu and D.E. Pritchard, Phys.
Rev. Lett. 59, 2631 (1987).
[10] P. Lett, R. Watts, C. Westbrook, W.D. Phillips, P. Gould and H.
Metcalf, Phys. Rev. Lett. 61, 169 (1988).
[11] P. Lett, W.D. Phillips, S. Rolston, C. Tanner, R. Watts and C.
Westbrook, J. Opt. Soc. Am. B 6, 2084 (1989).
[12] J. Dalibard, C. Salomon, A. Aspect, E. Arimondo, R. Kaiser,
N. Vansteenkiste and C. Cohen-Tannoudji, Atomic Physics 11, S
Haroche, J.C Gay and G. Grynberg, eds. (World Scientific, Singa-
pore, 1989), p.199.
[13] Y. Shevy, D.S. Weiss, P.J. Ungar and S. Chu, Phys. Rev. Lett. 62,
1118 (1989).
[14] D.S. Weiss, E. Riis, Y. Shevy, P.J. Ungar and S. Chu, J. Opt. Soc.
Am. B 6, 2072 (1989).
[15] C. Salomon, J. Dalibard, W.D. Phillips, A. Clairon and S. Guellati,
Europhys. Lett. 12, 683 (1990).
[16] C. Gerz, T.W. Hodapp, P. Jessen, K.M. Jones, W.D. Phillips, C.J.
Westbrook and K. M0lmer, Europhys. Lett. 21, 661 (1993).
[17] J. Dalibard and C. Cohen-Tannoudji, J. Opt. Soc. Am. B 6, 2023
(1989).
[18] PJ. Ungar, D.S. Weiss, E. Riis and S. Chu, J. Opt. Soc. Am. B 6,
2058 (1989).
[19] C. Cohen-Tannoudji, in Fundamental systems in Quantum optics,
Proceedings of Session LIII of Les Houches Summer School, J.
Dalibard, J.-M. Raimond and J. Zinn-Justin, eds. (North-Holland,
Amsterdam, 1992).
[20] Y. Castin, J. Dalibard and C. Cohen-Tannoudji, Light Induced Ki-
netic Effects, L. Moi, S. Gozzini, C. Gabbanini, E. Arimondo and
F. Strumia, eds. (ETS Editrice, Pisa, 1991).
[21] Y. Castin and J. Dalibard, Europhys. Lett. 14, 761 (1991).
[22] J.Y. Courtois, these de doctorat de l'Ecole Polytechnique, Paris,
France, 1993.
[23] P. Verkerk, B. Lounis, C. Salomon, C. Cohen-Tannoudji, J.Y. Cour-
tois and G. Grynberg, Phys. Rev. Lett. 68, 3861 (1992).
[24] J.Y. Courtois and G. Grynberg, Phys. Rev. A 46, 7060 (1992).
[25] PS. Jessen, C. Gerz, P.D. Lett, W.D. Phillips, S.L. Rolston, R.J.C.
Spreeuw and C.I. Westbrook, Phys. Rev. Lett. 69, 49 (1992).
[26] K. Berg-Sorensen, Y. Castin, K. M0lmer and J. Dalibard, Europhys.
Lett. 22, 663 (1993).
200 Y. Castin, J. Dalibard and C. Cohen-Tannoudji
[27] Y. Castin, K. Berg-Sorensen, K. Molmer and J. Dalibard, to appear
in Fundamentals of Quantum Optics HI, Ehlotzky, ed. (Springer,
Berlin, 1993).
[28] A. Hemmerich, Zimmerman and T.W. Hansch, Europhys. Lett. 22,
89 (1993).
[29] A. Hemmerich and T.W. Hansch, Phys. Rev. Lett. 70, 410 (1993).
[30] G. Grynberg, B. Lounis, P. Verkerk, J.Y. Courtois and C. Salomon,
Phys. Rev. Lett. 70, 2249 (1993).
[31] A. Clairon, P. Laurent, A. Nadir, M. Drewsen, D. Grison, B. Lounis
and C. Salomon, in the Proceedings of the 6th European Frequency
and Time Forum, held at ESTEC, Noordwijk (ESA SP-340, June
1992).
[32] A.M. Steane and C.J. Foot, Europhys. Lett. 14, 231 (1991).
[33] A.M. Steane, M. Chowdhury and C.J. Foot, J. Opt. Soc. Am. B 9,
2142 (1992).
[34] C D . Wallace, T.P. Dinneen, K.Y.N. Tan, A. Kumarakrishnan, PL.
Gould and J. Javanainen, to be published (1993).
[35] T. Walker, D. Sesko and C. Wieman, Phys. Rev. Lett. 64,408 (1990).
[36] A.M. Smith and K. Burnett, J. Opt. Soc. Am. B 9, 1240 (1992).
[37] A. Gallagher and D.E. Pritchard, Phys. Rev. Lett. 63, 957 (1989).
[38] PS. Julienne and J. Vigue, Phys. Rev. A 44, 4464 (1991).
[39] A. Aspect, E. Arimondo, R. Kaiser, N. Vansteenkiste and C. Cohen-
Tannoudji, Phys. Rev. Lett. 61, 826 (1988).
[40] A. Aspect, E. Arimondo, R. Kaiser, N. Vansteenkiste and C. Cohen-
Tannoudji, J. Opt. Soc. Am. B 6, 2112 (1989).
[41] M. Kasevich and S. Chu, Phys. Rev. Lett. 69, 1741 (1992).
[42] H. Wallis and W. Ertmer, J. Opt. Soc. Am. B 6, 2211 (1989).
[43] K. Molmer, Phys. Rev. Lett. 66, 2301 (1991).
[44] F. Bardou, J.P Bouchaud, O. Emile, A. Aspect and C. Cohen-
Tannoudji, Phys. Rev. Lett. 72, 203 (1994).
[45] F. Mauri, F. Papoff and E. Arimondo, in Light Induced Kinetic
Effects, L. Moi, S. Gozzini, C. Gabbanini, E. Arimondo and F.
Strumia, eds. (ETS Editrice, Pisa, 1991).
[46] M.A. Ol'shanii and V.G. Minogin, in Light Induced Kinetic Effects,
L. Moi, S. Gozzini, C. Gabbanini, E. Arimondo and F. Strumia, eds.
(ETS Editrice, Pisa, 1991).
[47] J.P. Bouchaud and A. Georges, Phys. Rep. 195, 125 (1990).
[48] B. Lounis, J. Reichel and C. Salomon, C. R. Acad. Sci. Paris, Serie
II, 316, 739 (1993).
[49] R. Cook and R. Hill, Opt. Comm. 43, 258 (1982).
 Laser Cooling and Trapping of Neutral Atoms 201
[50] V. Balykin, V. Letokhov, Y. Ovchinikov and A. Sidorov, Phys. Rev.
Lett. 60, 2137 (1988).
[51] M. Kasevich, D. Weiss and S. Chu, Opt. Lett. 15, 607 (1990).
[52] H. Wallis, J. Dalibard and C. Cohen-Tannoudji, Appl. Phys. B 54,
407 (1992).
[53] C. Aminoff, A. Steane, P. Bouyer, P. Desbiolles, J. Dalibard and C.
Cohen-Tannoudji, to be published.
[54] C. Aminoff, P. Bouyer and P. Desbiolles, C. R. Acad. Sci. Paris
Serie II, 316, 1535 (1993).
 10
Kinetics of Bose-Einstein Condensate
Formation in an Interacting Bose Gas
Yu. Kagan
Department of Superconductivity and Solid State Physics
Kurchatov Institute
123182 Moscow
Russia

Abstract
The kinetics of the formation of coherent correlation properties associated
with Bose condensation is studied in detail. The evolution of a nonequilib-
rium state with no "condensate-seed" is related to a hierarchy of relaxation
times. At the first stage, a particle flux in energy space toward low energies
sets in. The evolution in this case is described by a nonlinear Boltzmann
equation, with a characteristic time given by interparticle collisions. When
the particles which will later form the condensate have a kinetic energy
which is less than the potential energy, a quasicondensate starts to form.
In this stage, fluctuations of the density (but not of the phase) are sup-
pressed and short-range coherent correlation properties are governed by
the equation of motion for a quasiclassical complex field. The next stage is
connected with the formation of the long-range order. The time for forming
topological order and therefore genuine superfluidity proves to be dependent
on the system size. The off-diagonal long-range order, arising after the at-
tenuation of long-wave phase fluctuations, has a size-dependent relaxation
time as well.

1 Introduction
The problem of Bose-Einstein Condensation (BEC) kinetics, being in-
teresting in itself, has acquired a special significance in connection with
experimental efforts to observe this condensation in a number of systems
with particles with a finite lifetime. Such systems include spin-polarized
atomic hydrogen [1], excitons [2] and biexcitons [3] in semiconductors
and, more recently, laser-cooled atomic systems [4]. In all cases, the
density is supposed to be low so that we always deal with a weakly inter-

202
 Kinetics of BEC Formation in an Interacting Bose Gas 203
acting Bose gas. This makes the investigation of these systems especially
attractive, because it allows the possibility of revealing the main features
which are characteristic for a Bose system in general, and, at the same
time, provides a chance for direct comparison with theory.
In all conceivable metastable systems whose evolution should lead
to BEC, the initial state contains no seed of condensate. If the final
equilibrium state is one with a sufficiently high condensate density, then
we are faced with the time evolution of an essentially nonequilibrium
system. Under these conditions the question of the time for attaining BEC,
or, more precisely, of the formation time of the appropriate correlation
properties, becomes very nontrivial. The answer to this question is
important not only for comparison with the lifetime of a system, but it
can also be crucial for the possibility of detecting the condensate.
In Ref. [5] it has been shown that the rate for inelastic processes
decreases substantially when a Bose condensate appears in the system.
This effect opens up an interesting opportunity for the observation of
BEC, particularly in an atomic hydrogen gas. Although there is no
true Bose condensate for T ^= 0 in the two-dimensional case, it was
shown in Ref. [6] that the change in the probability for the inelastic
processes persists below the Kosterlitz-Thouless transition, due to specific
properties of the two-dimensional quasicondensate (i.e. a condensate with
a fluctuating phase). However, in all cases this effect manifests itself only
after the development of specific quantum correlations in the process of
evolution of the system.
In a nonuniform space the opposite effect is realized: the appearance of
a condensate causes a sharp increase of the rate of any density-dependent
process (see, e.g. Ref. [7]). It allows BEC to be revealed even when the
condensate is localized in a very small volume. The latter is characteristic,
in particular, of a trap-type experiment (see, e.g. Ref. [8]). Again, we
need a certain time for the necessary quantum correlations to form.
One promising way of studying BEC in an excitonic gas is based on
the measurement of the transport motion of an expanding gas with the
aim of revealing the excitonic superfluidity [2,9]. The use of this method
assumes indirectly that the time of the formation of topological order is
short enough.
Several theoretical papers have addressed the question of the time evo-
lution of a nonequilibrium interacting Bose gas. Levich and Yakhot [10],
considering the evolution of an ideal Bose gas coupled to a heat bath,
came to the conclusion that the formation of a condensate peak requires
an infinite time. Nozieres [11] also obtained the same result. Recently,
204 Yu. Kagan
studying the kinetics of BEC in an ideal Bose gas, Tikhodeev [12] has
obtained the same answer, provided the size of the system goes to infinity.
Later, Levich and Yakhot [13] took into consideration the interparticle
interactions which should naturally play a dominant role in the temporal
evolution. Within the framework of the nonlinear Boltzmann kinetic
equation, they found a finite time for BEC formation. However, these
authors made a drastic modification of the collision integral in order
to simplify the calculation, which led to such a strong change of the
character of the evolution of the particle energy distribution that the
authors themselves doubted the validity of the results. Moreover, they
realized correctly that the kinetic equation is not applicable in the low
energy region where the kinetic energy becomes less than the potential
energy. The attempt to improve the kinetic equation was not adequate
because it still made use of the random-phase approximation, which fails
precisely in this region.
Snoke and Wolfe [14] undertook a direct numerical evaluation of the
Boltzmann kinetic equation for the weakly interacting Bose gas. The
calculations demonstrated the formation of the Bose distribution with
chemical potential \i close to zero on the time scale of the classical
scattering time of the particles. However, a condensate did not appear
even after a much longer period of time. Simultaneously, the conservation
of the particle number was violated. As we shall see below, both results
are not accidental and have a common origin.
Another extreme result was recently reported by Stoof [15], who
asserted that the time required for the formation of a Bose condensate
was ~ h/TC9 where Tc is the temperature of BEC. That result seems
surprising since it does not depend on the interparticle interaction, and
moreover this time is shorter than any characteristic interaction time in
a dilute Bose gas.
Actually, the question of the formation time of a Bose condensate does
not have an unambiguous answer. The real answer depends strongly on
the particular problem under consideration (see Ref. [16]). The ambiguity
arises at the stage of evolution when a substantial fraction of the particles
which should form the condensate in equilibrium is concentrated in the
energy interval e < noU, where U is the effective particle interaction
vertex, and no is the equilibrium density of the condensate. At this
stage the kinetic energy of the particles becomes smaller than their
average potential energy and a pronounced mixing of states with different
momenta k arises. The state of the system, taking into account that in
this case the occupation numbers obey the condition n^ > 1, can be
 Kinetics of BEC Formation in an Interacting Bose Gas 205
characterized by a quasiclassical complex-valued field ¥ (see e.g. [17])
whose dynamical evolution is described by the nonlinear Schrodinger-
type equation (h = 1)

i-^ = -^F+fr|¥| 2 ¥, (1.1)

where m is the mass of the particle. At this stage the random phase
approximation is not valid and (1.1) cannot be reduced to an equation for
occupation numbers. (There is a well-known analogy: the classical electric
field and the Maxwell equations as a limit of quantum electrodynamics
when the photon occupation numbers are much larger than unity.)
In the limit t —• oo, the solution of (1.1) approaches the value of the
complex order parameter with a macroscopically defined modulus and
phase which characterize the equilibrium state of an interacting Bose
gas at T < Tc. In the process of evolution, the modulus and the phase
of the function ¥ fluctuate strongly. Only with the relaxation of these
fluctuations do the proper coherent correlations start to set in. The
attenuation of the modulus or the density fluctuations begins first. This
leads to the formation of a short-range order with correlation properties
that are close to those of an equilibrium system with a true condensate
(see below). The characteristic time scale for this stage of the evolution
is given by

and, correspondingly, the length scale of the region where this short-range
order comes into play has the value of the correlation length

(L3)

Thus, even at this stage of evolution, the system should demonstrate a

number of physical properties identical to those of a Bose gas with a real
condensate. If we are interested in the processes governed by the local
correlation properties at distances r < rC9 we can conditionally regard
TC in (1.2) as a time scale for BEC (for the particles which are already
in the energy interval e ~ noU). Since, however, the relaxation of the
phase fluctuations requires a much longer time, a genuine condensate
with a S -function momentum distribution of condensate particles does
not appear even at t > TC. In a sense, we obtain a quasicondensate or
a condensate with a fluctuating phase. The organization of long-range
206 Yu. Kagan
order, which is necessary for the <5-function peak to build up, requires a
long time, depending on the size of a system.
The formation of the quasicondensate is inevitably accompanied by
the appearance of a huge number of topological defects. This is easy to
understand if one takes into account the absence of phase correlations
at a distance r > rc [18]. Hence, the quasicondensate does not imply
macroscopic superfluidity. Due to the short-range order, one can describe
the nonequilibrium state in terms of vortex lines and phonons, both being
characterized by essentially nonequilibrium distributions. The evolution
in this stage is related to the removal of the topological defects from the
system and the relaxation of anomalous phase fluctuations preventing
the formation of long-range order. The former occurs first, due to
a dissipative interaction of vortices with elementary excitations (see,
e.g. [19,20]). The decay of the vortex state with some characteristic time
TV should result in the formation of topological long-range order (TLRO)
and, thus, of macroscopic superfluidity. The appearance of TLRO does
not imply in general the appearance of the off-diagonal long-range order
(OLDRO) and, therefore, a 6-function peak in the particle distribution,
since phase fluctuations can survive independently of topological defects.
Then, one can expect that after the relaxation of topological defects,
the state of system will be similar to the state of a two-dimensional
superfluid at finite temperature (the existence of TLRO even though the
condensate is absent). Since the phase fluctuations are coupled to the
fluctuations of the phonon subsystem, the relaxation in this case and the
time T(f) for approaching OLDRO can be studied via the attenuation of
nonequilibrium phonon fluctuations [18]. The times TV and T^ are both
sensitive to the temperature and on the size of the system.
Returning to the formation of the quasicondensate, we note that the
time TC in (1.2) is not the only time scale and generally not the longest
one. Usually, the future condensate particles are initially located in an
energy interval far from the coherent region e < noli. As a result, the
initial evolution of the nonequilibrium subsystem can be described by
the Boltzmann kinetic equation. In the most interesting energy region,
where the nonlinearity of the collision integral plays a decisive role, the
occupation numbers obey the condition n^ > 1. This means that (1.1)
is correct in this region as well. But now all modes can be regarded as
independent and the use of the random-phase approximation is justified.
This leads to the relation < a^a^ >= /?k<5k,ks where a^ is a Fourier
component of the field *F. (This relation is not correct in the coherent
region.) Naturally, under these conditions the coherent correlations are
 Kinetics of EEC Formation in an Interacting Bose Gas 207
absent. It is important to emphasize that the region where (1.1) is correct
proves to be much wider than the coherent one, not to mention a narrow
region where (1.1) is used for the description of the superfluid state (the
Ginsburg-Gross-Pitaevskii equation - see, e.g. [21]). In the framework of
the random-phase approximation, (1.1) can be transformed into a kinetic
equation. In the homogeneous case, the latter assumes the form (see,
e.g. [22])
n]\ (1.4)

, [n]) = 1^5 f d3k2d3k3d3k4[nin2(n3 + n4) - («i + n2)n3n4]

x<5(ki + k2 — k 3 — k4)S(ei + e2 — e3 — e4),
where nt = n^et = kf/2m. Equation (1.4) differs from the general
Boltzmann equation for an interacting Bose gas only by the absence of
a spontaneous contribution to the scattering (% > 1!).
If one tries to use (1.4) and (1.5) in the entire energy interval including
the coherent region, one faces a serious difficulty. This becomes clear if
one compares (1.1) and (1.4). The kinetic equation (1.4) is independent
of the sign of U, whereas (1.1) is very sensitive to this sign. When U
is negative the evolution of *¥ in the coherent region leads to collapse
(see Ref. [22]) with the absorption of particles at e = 0, but when U is
positive, a condensate should form. The kinetic equation itself cannot
distinguish between these two cases. In numerical studies of (1.4) where
the additional special condition at e = 0 is not introduced, it becomes
impossible to prevent the solution from collapsing in the above sense (see
the discussion in [23]). Evidently, Ref. [14] has this problem.
The evolution in the energy region where the employment of (1.4) is
justified is characterized by a time scale (see Ref. [16] and below)

(!

which is the usual kinetic time of interparticle collision. Here a is the

scattering cross section, vj is the thermal velocity and n is the total
density of the gas. Comparing (1.6) and (1.2), we see that the condition

Tkin > ?c (1.7)

essentially always holds.
Thus, the kinetics of BEC are characterized by a hierarchy of time
scales and the overall scenario is as follows. Let us assume that after
fast cooling, the particles which will subsequently from the condensate
are initially concentrated at high energies, of the order of the equilibrium
208 Yu. Kagan
value Tc. Several collisions per particle in this case result in the formation
of a particle flux in energy space toward lower energies. The self-evolution
in the interval between Tc and n^tj (kinetic region) along the energy axis
is described by the Boltzmann kinetic equation. A certain energy e*
divides this interval in two parts. For e < e*, the occupation numbers n^
become greater than the equilibrium ones (as a result, n^ > 1) when the
particle flux approaches this region, and one can take advantage of (1.4).
At e > e*9 the general Boltzmann equation must be used. As the analysis
below shows, the time scale for the passage through both kinetic regions
is determined by (1.6).
According to (1.7), the initial stage of transformation in the coherent
region takes place much faster. Thus, the evolution in the kinetic region
turns out to be the limiting step for the formation time of a quasicon-
densate. Further steps in the BEC scenario are related to the relaxation
of the topological defects and the nonequilibrium phase fluctuation, with
times TV and TQ extremely long in comparison to the kinetic time ikin
given by (1.6).

2 Coherent Regime. Formation of the Quasicondensate

We now study the evolution in the coherent region e < noU, assuming
that the bulk of the particles which are about to form the condensate are
in this energy interval. As we have discussed, we have to use (1.1) for the
complex field *F in this case.
In the limit t —• oo, the function *F should reach its equilibrium value

¥ 0 = n0/2 exp(iO0 - W), fi0 = n0 U, (2.1)

where Oo is the equilibrium value of the condensate phase. We are

ignoring the particles outside the condensate in this energy region, since
their total number per unit volume is small in comparison with no.
Writing the function *¥ in the form

we can rewrite (1.1) as

i%
ot
= -^f
2m
+ noU(\f\2f-f).
In terms of dimensionless variables

t = t/ze, 7 = r/rc (2.3)

 Kinetics of BEC Formation in an Interacting Bose Gas 209
(see (1.2) and (1.3)), this equation becomes

i^ = -A F /+|/| 2 /-/- (2.4)

As t —• oo, we have / —• exp(z<Do)-

It follows from (2.4) that the evolution on length scales ~ rc occurs
over times ~ TC. With increasing t9 the gradients of the phase and the
density are smoothed out. This smoothing is equivalent to a decrease in
the scale of the important values of the wave vector k and frequencies
co in the Fourier representation of the function / . The excess energy
per particle ~ noU goes off into the kinetic region. It turns out that
the condition t > 1 is by itself a sufficient condition for substantial
suppression of the fluctuations in the absolute value of / , that is, the
density.
To demonstrate this point, we introduce a deviation 6n from the
mean spatial density no. Using (2.4), we write separate equations for the
absolute value and phase of function / . After linearization, the equations
for the Fourier components of the density and phase become

icokTc(5nk/n0) = -2r2k2Q>k,

icDkTc^k = (dnk/n0)(l + r2ck2/2). (2.5)

This system of equations yields a spectrum which essentially is the
Bogoliubov spectrum,
co2 = (2r2ck2/T2)(l+r2k2/2). (2.6)

From the first of the equations in (2.5), we find

[Snk/n0]2 = (4r4ck4/co2T2) | 0>, | 2 . (2.7)

For t > 1, the characteristic value k* of the momentum becomes smaller

than kc = r~l. For such values of k, the dispersion relation given by (2.6)
is acoustic,
co = ck9 c = 2l/2(rc/Tc) (2.8)

and we have

[Snk/n0]2 = 2r2k2 | % | 2 .

Hence we obtain

< (dn/no)2 > = X ) I ^ / " o l 2 =

k
210 Yu. Kagan
Taking account of the relation E ~ (k*/2m)noV for the kinetic energy
remaining in the system, and using

E * (rco/m) /
J
k
we find from (2.9)

< (dn/no)2 > ~ (fc./fco)2 < 1 • (2.10)

Consequently, over times determined by TC9 a state forms in which den-
sity fluctuations are essentially suppressed (as in a true condensate). It is
interesting to show that suppression of density fluctuations is a necessary
and sufficient condition for the same change in the probability of inelastic
processes to appear, as in the presence of a true condensate. With this
goal in mind, we consider the example of three-particle recombination,
which is a limiting process in the decay of high density spin-polarized
atomic hydrogen. For this process the probability is determined by the
correlation function

z = < £ + (r, tyv+(r9 t)* + (r, tyv(T, tyv(r9 tyv(r91) >, (2.11)
as was shown in Ref. [5] This is a typical short-range correlation function.
The coincidence of arguments reflects only the fact that the interaction
takes place at distances much less than rc and times much less than xc.
In the case at hand, this correlation function reduces to
Z=nl< f 3(r, t)/ 3 (r, t) > = < n3 > . (2.12)
If fluctuations in / are suppressed, then
Z=n\. (2.13)
If fluctuations are present, then
Z = n\ + 3n0 < (cSn0)2) > + < (<5n0)3) > . (2.14)

The expression for on can be written

dn/no = (| / | 2 -D (2.15)
Consequently,
2
< (Sn/n0) >• = «§[<l / I >-2
4
<\f 2> +1],
< («5n/n 0)3 > = «o[<l / |6> -3 <| /| 4 > +3< I / | 2 > - 1 ] - (2.16)
Near the boundary between the coherent and kinetic regions, where (2.1)
 Kinetics of BEC Formation in an Interacting Bose Gas 211
continues to be correct, all the field modes corresponding to different
wave vectors k can be regarded as independent, i.e.

< / k / k ' > = IM 2 (5 k , k .. (2.17)

Expanding / in a Fourier series, and using an analog of Wick's theorem,

we then find

<l / | 2> = E I h l 2 = !' <l / l 4 > = 2' <l / I6> = 6.

k

Using these in conjunction with (2.14)-(2.16), we obtain

Z = 6n30 . (2.18)

We conclude that when density fluctuations are substantially sup-

pressed, the probability for three-particle recombination is reduced from
(2.18) by a factor of six (see (2.13)). This result is precisely the same as
that found for the case of a true condensate [5,6]. From this point of
view, a state with narrow "precondensate" peak (fc* <Cfco)in the particle
distribution has all the properties of a true condensate for all processes
defined by short distance and short time interval correlations. Such a
state can be considered as a quasicondensate. These results demonstrate
that the formation of the quasicondensate starts only when the future
condensate particles turn out to be in a coherent region. After that, it
takes time of the order of TC given by (1.2) for a proper transformation
to evolve.

3 Kinetic Region. Linear Regime

We now consider the kinetic region, assuming self-evolution of a nonequi-
librium system. We suppose that the disturbance of the system after fast
cooling, as well as the location of extra particles which should later set up
the condensate, are related mainly to high energy degrees of freedom at
the initial time. Actually, this is rather typical for all conceivable systems.
To get away from the fluctuation region near Tc, we assume that
the condensate density no in the equilibrium will be restricted by the
inequality
no/n>(na3)1/3, (3.1)

in accordance with the Ginsburg criterion (see, e.g. Ref. [24]). Here n is
the total density of particles, and a is the s-wave scattering length. The
212 Yu. Kagan
latter is connected to U via the well-known relation

The small value of the gas parameter na3 allows (3.1) to be compatible
with the conditions
no/n < 1 (3.3)

and
T —T
-V-•* C
< 1, (3-4)
where T is the equilibrium temperature of the gas. We take advantage
of this possibility to make the discussion more transparent.
For the energy e* which defines the boundary of the nonlinear region,
we have in this case
e* < T. (3.5)
This means that in the first stage the excess particle flux in the energy
space covers a rather wide interval (T,e*), where the deviation of the
distribution function from its equilibrium value is relatively small.
The general kinetic equation for a Bose gas in a spatially uniform
case, for a distribution which is isotropic in terms of momentum, has the
well-known form (nt = ne.)

hi = -W f de2de3de4X(-, - , -)S(e{ +e2-e3- e4)

J ei ei €\

n2)n3n4l (3.6)
where
W = fr2m3/47r3fc3 = ^ (3.7)

and

X = \ J df2 sinC^sinMl)'/ 2 ] sinty(g)1/2] sinM^) 1 ' 2 ]. (3.8)

We have restored h in (3.7) to exhibit the dimensionality. The function
X for the entire set of e,- satisfying the energy conservation condition can
be reduced to the simple form (see Ref. [14])

. (3.9)
 Kinetics of BEC Formation in an Interacting Bose Gas 213
The temporal evolution in this region can be found by means of the solu-
tion of the linearized equation (3.6). The following simple considerations
enable one to reveal the main features of this stage.
We assume that a local quasiequilibrium is reached in the energy
interval where the bulk of excess particles is concentrated. Then the
particle distribution can be written as

ne » [exp[(e - fi(t))/T] - I]" 1 « T/(e - MO), (3.10)

with a time-dependent chemical potential fi(t) > 0. In general, the

temperature in (3.10) may also depend on the time. However, it is easy
to verify that this time dependence can be ignored by virtue of (3.1).
Relation (3.5) of course holds only for e > ii(t). Over most of the linear
interval, the condition \i <C e holds, as we will see below. We can thus
write
Ane = ne-ne « n(t)T/e 2. (3.11)

The e dependence corresponds to motion of the front of the distribution

60(0, with a width determined by the same parameter eo(O- The value
eo(O itself is found from the condition

a / e1/2Anede = n0, (3.12)

Jeo

where a = m3/2/21/27r2. The integral is determined by its lower limit and

has the value 2T/x(t)/€l0/2(t). Hence

pit) = no4 /2 (O/2aT. (3.13)

The boundary of the linear regime, 6*, is determined by the condition

Sn€/ne ~ 1. It can be found directly:

e. « T(n o/n)2 (3.14)

(here we have made use of the circumstance that T is close to Tc, and
we have used the relationship between Tc and n). In this case we have

»(t)=[e.e o(t)]1/2<eo(t). (3.15)

The parameter fi(t) lags behind 6o(O at all times and is comparable to this
quantity when the boundary of the linear regime, 6*, is reached. This cir-
cumstance demonstrates that it is legitimate to assume a quasiequilibrium
distribution of the form given in (3.10) and (3.11).
Let us find an expression for the particle flux in energy space. This
214 Yu. Kagan
flux is related to the motion of the front eo(t). Using (3.11) and (3.13) we
find
Q = -a4 /2 6 O n eo * -ino(W^o). (3.16)
The ratio eo/^o can be found from the kinetic equation (3.6). Analysis
of the collision integral shows that the main contribution is provided
by processes in which all particles have an energy of the same order of
magnitude. Besides, after linearization, the collision-integral operator is a
homogeneous functional of the energy to zeroth order. In other words, it
does not change under the replacement e,- —• >ie,. As a result, the collision
integral is characterized by an energy-independent relaxation time and
we have approximately
£oM> « -7 Akin • (3.17)
Explicit calculations using the kinetic equation (3.6) result in the es-
timate y « 4 (ikin is equal to (1.6) with o = 8na2). The particle flux in
energy space thus takes the simple form
2«yn o/2Tkin. (3.18)
It is interesting to note that the flux has a wave-shape with an increas-
ing front amplitude and a decreasing particle-distribution width during
the passage of the considered energy interval (see (3.11)). Using (3.17) and
(3.14), we can estimate the time which takes the front of the distribution
to reach the boundary of the linear region e* as

( (3.19)
y \nj
The time required to cross the linear region is thus determined by the
ordinary collision time scale Tkin? enhanced by a factor \n(n/no) at small
wo-

4 Nonlinear Kinetic Regime

Consider the nonlinear region n0U < e < e*, where we have
n€>n€>l. (4.1)
Under condition (3.1) the inequality
e, > n0U
is always satisfied, so the system necessarily passes through the nonlinear
kinetic regime. The evolution in this region is described by the kinetic
 Kinetics of BEC Formation in an Interacting Bose Gas 215
equation (1.4). After integrating over the directions of momenta, this
equation takes on the form (3.6) if the expression in square brackets is
replaced by
+ n4) - (n{ + n2)n3n4. (4.2)
It can be concluded from the structure of the collision integral that the
effective kinetic time is determined by

**~Wn¥. (4.3)
Analyzing a nonlinear kinetic equation of such a type, Zakharov [22]
found that the equation Jco\\ = 0 has two extra solutions, corresponding
to a steady state particle flux and to a steady state energy flux in the
energy space. In first case, the solution has the form

ne = A/e1/6 . (4.4)

Substituting this solution into (4.2), we find

T^(e)~WA2/eV\ (4.5)

Thus, the solution (4.4) leads to the enhancement of the collision processes
with decreasing energy in both the incoming and the outgoing terms.
Considering the particle flux moving toward 6 = 0, there are good rea-
sons to suppose that behind the front the particle distribution should be
close to (4.4). Direct numerical calculations performed by Svistunov [23]
have corroborated this conjecture.
Thus, the effective transformation from the distribution in (3.11) to
that in (4.4) starts at the energy e ~ e*. Once the distribution (4.4)
has been established behind the front ?o(O of a particle-flux wave, the
parameter A can be found from the condition

For ?o(O much lower than e. we have

Note that the restructuring of the distribution from (3.11) to (4.4) leaves a
substantial fraction of the particles in the tail of the distribution (e ~ e*)
Making use of (4.6), one can rewrite (4.5) in the form

3
- (4-7)
216 Yu. Kagan
Writing the kinetic equation as h€ = J+ — J_, we have J+ ~ n€/x^{e\
while for the left-hand side we have h€ ~ ne/Tkin. It means that in the
first approximation Jcoii « 0 at e < e* and, consequently, the solution
(4.4) is self-consistent.
The value of the flux associated with the motion of the front is
determined by a relation such as (3.16). Taking into account (4.4), we
obtain

( 4 .8)

Under the assumption of a constant flux, this relation leads to

£0(t)«e,(l-t/t 0)3, (4.9)
where
to = no/Q (4.10)

(to is reckoned from the beginning of the nonlinear regime). The quantity
to determines the time at which the front of the particle-flux wave arrives
in the coherent region. Using the value of the flux Q entering from the
linear region (see (3.18)), we immediately conclude to ~ ikin-
In the initial stage, the buildup of particles in the coherent region is
described by
nc*Q(t-t0). (4.11)

Considering the subsequent evolution, we can easily understand that the

particles in the kinetic region e > noli interact with the particles in
the coherent region just as in the case of a real condensate. Without
paying attention to the evolution in the coherent region by itself, one can
introduce as an approximation the following system of coupled equations
(see Ref. [23]),
he = -Jcoii( W,e) - ncfcoll([nle), (4.12)

he = ncJ'U[n\), (4.13)
where
W 1 f€l
J'con([n],ei) = r ^ { / de2[n1(n2 + ni_ 2 ) - n2n1-2]
a
6j JnoU

+2 de2[nin2-i-n2(ni+n2-i)]}, (4.14)
Jei
 Kinetics of BEC Formation in an Interacting Bose Gas 217

de2de3[n2n3 - n 2 + 3 (n 2 + n 3 )]. (4.15)

The first term on the right-hand side of (4.12) describes the interaction
of particles belonging to the kinetic region before and after a collision.
The distribution (4.4) causes the divergence of the integral in (4.15) at
low energies. If one defines the lower bound of the energy as eC9 one
obtains for the collision integral

•Cu ~ 4fr < 416)

This result helps one to understand that the transition to the coherent
region is evolving from intervals with successively increasing values of
energy. As we have already mentioned, the majority of the particles in
the nonlinear kinetic region are concentrated in the tail of the particle
distribution. This is why the transition will effectively be over when ec
increases up to energy of the order of e*. It is easy to estimate from
(4.13) and (4.15) that this time has the same order of magnitude as ikm-
Thus, the crossing of the nonlinear kinetic region and the arrival of
the majority of the excess particles into the coherent region occurs on the
time scale determined by ikm- Since ikin > ?c this means that the overall
time of the quasicondensate formation has the same time scale Tkin.
The results found above are not dependent on the inequality n0 < n.
As no increases, the boundary e* moves into the high energy region, and
the flux Q increases. However, the time scale stays the same. If no ~ n
holds, then e* ~ T and the nonlinear regime covers practically the entire
kinetic region while retaining Tkin as the time scale.

5 Formation of Topological Order

As we have discussed, the formation of the quasicondensate, related to
the suppression of density fluctuations, leaves out large long-range phase
fluctuations. The relaxation of these fluctuations on one hand and the
appearance, as a result, of true superfluidity and a genuine condensate,
require quite different time scales.
Suppose that the quasicondensate correlation properties have already
been set up in a region with size scale ro > rc. At distances greater
than r0 any phase correlations are absent. Let us mentally divide the
space into boxes with linear size / > ro(t). The absence of correlations
between the phase in the different boxes easily leads to the appearance
of a vortex line or a vortex ring of the size of the / scale. If we increase /
218 Yu. Kagan
but simultaneously remove a part of the phase field created by the rings
with size less than /, then the absence of the correlation between phases
in bigger boxes leads to the same conclusion. As a result, for the number
of rings W(R) with radius R in the volume unity we have
ww (51)
~h-
In the general case, the phase <I> of the function *F can be represented
as the sum
O = <D0 + (/>, (5.2)
where Oo is the phase field, determined purely by the configuration of
the vortex lines. To remove the ambiguity, the relation

= ±2n

for any contour enclosing a vortex line should be supplemented by the

equation A<X>o = 0 for the entire region out of the vortex lines.
If we make use of the same picture of uncorrelated boxes, one can
ascertain that the difference between the values of the regular part of
phase 0 in different boxes is of the order of magnitude n. Owing to the
arbitrariness of the box size this leads to the following relation for the
Fourier harmonics of the field (j):

I <t>k I2 k3 ~ 1, k< rfit). (5.3)

Making use of the well-known relation between | (/>& | 2 and the phonon
number occupations (see, e.g. Ref. [21]) allows us to proceed from the
relaxation of phase fluctuations to the evolution of the nonequilibrium
phonon distribution. Since in the process of the quasicondensate forma-
tion, the excitation spectrum becomes acoustic (see Section 2), we find,
with allowance for (5.3),
™» 1

k < rQX(t). (5.4)

Here c is the sound velocity (2.8). Let us first consider the relaxation of
topological defects. The main channel of the relaxation is self-annihilation
of the vortex rings caused by interaction with the elementary excitations
which provide the drag force and thus the dissipation of the energy.
The small-radius rings naturally disappear first. We define the minimum
radius in the distribution (5.1) at a moment t as R(t). Then it is easy to
 Kinetics of BEC Formation in an Interacting Bose Gas 219
make sure that the average radius of the rings is equal to R(t) and the
total length L of the vortex line in a unity volume equals

L(t) « J - . (5.5)
R(t)
This means that the relaxation at any moment t is related to the rings of
one size scale R(t).
In reality, the emerging vortex structure may resemble a tangle rather
than an ensemble of the rings. Usually, when studying the kinetics of
such a tangle, the average distance between the vortex lines is considered
as a basic parameter of a system. The same scale defines the characteristic
radius of the curvature. Since this parameter and L are linked by the
relation (5.5), the kinetics of the two models prove to be equivalent. For
simplicity we shall only refer to the first one.
The characteristic time of the self-annihilation of a ring of radius R is
(see, e.g. Ref. [19])
2nps R2
^R)™—]^RF C- (5 6)
"
Here y is the drag force coefficient per unit length of a vortex line. We
have taken into account that the core radius is of the order of rc and
R(t) > rc. It can be shown [18] that in an interacting Bose gas in the
temperature interval
n0U < T < TC9 (5.7)
the coefficient y is equal to
y = p s r o a, (5.8)
where

and To is the quantum of circulation.

It is a specific feature of the gas that a has a constant value in the
interval (5.7), where the elementary excitations practically coincide with
free particles. Any ring has a self-induced velocity related to the radius
value (the local radius of curvature for an element of the tangle structure).
One can estimate that in the time TV given by (5.6), the ring passes over
a distance
lR*^R. (5.10)
220 Yu. Kagan
Since a <C 1, this signifies that the ring experiences many crossing
encounters before disappearing. If these crossings are not accompanied
by reconnection, an estimate of the time for removal of topological
defects can be obtained from (5.6), (5.8) and (5.9). For a system of size
Do, we find

. (5.11)

Apparently, it is more realistic to think that the reconnection happens

efficiently (see, e.g. the discussion in Refs. [25] and [26]). Analysis of the
relaxation in this more complicated case [18] leads to the result

Note that this expression differs from (5.11) only in a substantial decrease
of the dependence upon the small interaction parameter a. (In the region
T <n0U one finds that a ~ T5 [27].)
Thus the time of the formation of topological long-range order, which
is determined by (5.12) (or (5.11)), turns out to be dependent on the size
of the system. In a time of the order of TV(DQ), the Bose gas becomes
truly superfluid.

6 Formation of the Off-diagonal Long-range Order

The complete relaxation of topological defects does not in the general
case imply the appearance of off-diagonal, long-range order together with
a strictly 5-function momentum distribution of particles. Accomplishing
the crossover from the quasicondensate to the genuine condensate also
requires the attenuation of the long-wave fluctuations of the phase 0
or the proper nonequilibrium distribution of phonons. It is interesting
to mention that the phonon distribution (5.4) leads to a logarithmic
divergence of the correlator < </>(r)</>(0) > at large distance, quite similar
to the two-dimensional case.
Considering the relaxation of the nonequilibrium distribution of phon-
ons (5.4), we deal with the problem of the attenuation of the sound
using the Bogoliubov dispersion law (2.8). At a moment t, when a
region free from vortices of size R(t) is formed, the dissipation of sound
with k > R(t)~{ proves to be much the same as in a true superfluid
system. This makes it possible to make use of some results obtained for
a superfluid liquid.
 Kinetics of BEC Formation in an Interacting Bose Gas 221
In a uniform system, in which the relaxation time is connected with k
stipulated by large distances, it is natural to assume the inequality
kl < 1. (6.1)
Here / is a free path length of particles. This means that we have to
analyze the attenuation of a sound in the hydrodynamic regime. As is
well known, condition (6.1) allows one to start with a nondissipative
system of equations and then to find the attenuation as a perturbation.
From the general system of the linear hydrodynamic equations for the
superfluid liquid, we have (see e.g. Ref. [28])

where pn is the density of normal liquid, and s is the entropy per unit
mass. Considering p and T as independent thermodynamic variables, we
can write for a small variation of parameters in a sound wave

Introducing (6.3) into (6.2), we find the system of equations for pr and
V . The equations (6.2)-(6.3) are correct in the general case. Applying
these equations to the interacting Bose gas, we find that the results differ
noticeably from the usual results for superfluid helium.
In the temperature interval (5.7) the normal excitations practically
have the dispersion law of free particles. Consequently, the normal
component has the properties of an ideal Bose gas. In this case, for the
temperature-dependent part of the pressure p T , we have

*0, (6.4)
V UP J T
and as a result
= c\ (6.5)

where c is the condensate sound velocity (2.8). On the other hand, we

have
c»,s,p n ~(^) 3/2 , ^ = f, (6.6)
Tc cv 3
where cv and cp are the specific heats at constant volume and pressure
222 Yu. Kagan
respectively. The solution of the system of equations (6.2) and (6.3), with
allowance for (6.5) and (6.6), leads to the following result:

The normal velocity vn can be found from the equation

Taking into account the fact that ps =£ f(p) we find, with allowance for
(6.6), that
3 V
(68)

Using (6.7) and (6.8), we can find the sound dissipation by employing
the standard approach (see, e.g. [28]). Since n o & < 71 and the specific
heat has temperature dependence given by (6.6), it is easy to show that
the role of the viscosity is negligible. As a result, the attenuation of
sound is determined by the thermal conductivity. Direct evaluation gives
for the phonon relaxation time

where K is the coefficient of the thermal conductivity. It is interesting to

compare TV(R) from (5.12) and Tph(k = R ):

(6.10)
•t ph(k=TCl) (na3)l'2Tc In R/rc'
For a realistic value of the system size, this ratio is always much greater
than unity. This assertion has an additional basis. As is well known,
sound reflection from a wall is accompanied by strong absorption (see,
e.g. Ref. [28]). At the same time, the phonon free path length approaches
the value Do much earlier than R(t) approaches this size.
Since the phase relaxation time T^ actually coincides with TPJ,, we can
conclude that %$ < xv. This signifies that the phase relaxation happens
faster than the vortex annihilation. Therefore, for the conditions un-
der consideration, complete long-range order should become established
when the topological order appears.
In a system of limited size, if the inequality / > Do is justified, the
hydrodynamic condition (6.1) is violated. In this case, the sound attenu-
ation is determined by the direct scattering of phonons with the normal
 Kinetics of BEC Formation in an Interacting Bose Gas 223
fluid excitations. The calculation of this process leads, in the temperature
interval (5.7), to the relation [18]

kT (6.11)

Comparing this result with (5.12), we find

TV(D0) Do T
(6.12)
Tph(k = D01) rc Tc '
Provided Do >> rc, (6.12) leads in many cases to the same conclusion as
before. However, we cannot exclude the case when the ratio Tv/Tph turns
out less than unity. In this case, T^ proves to be larger than xv and the
formation of the off-diagonal order is characterized by the time

(6 13)
< ~ Wo -
(we have again restored h in the explicit form).

7 Final Comments
In a number of experimental situations, nonuniform conditions are in-
volved and the condensate should form in a small volume. This takes
place, for example, in different kinds of magnetic traps or during the
creation of excitons by a laser pulse. We make some remarks in this
connection.
The formation of the quasicondensate in this case does not differ
noticeably from the uniform case characterized by the same scale of
relaxation time. The number of vortices is now limited, especially at low
densities. The time of their self-annihilation drastically shortens due to
the small size of the condensate location region. The free expansion of
the system, which is characteristic in particular of the excitonic gas in a
semiconductor, causes further decrease of the time TV. This is important
for the appearance of the superfluid properties of such a system.
The quasicondensate formation in a nonuniform case is accompanied
by a sharp increase of density. This leads inevitably to the strong
increase of the rate of any inelastic process which depends nonlinearly
on the density [29,7]. In a recent paper [30], it was shown that this fact
can prevent the nonequilibrium system from penetrating deeply into the
Bose condensation region. (It is useful to mention that all the results of
the preceding sections are correct at T ~ Tc.) One can conjecture that
the experimental observation [2] that the excitonic gas does not cross the
224 Yu. Kagan
BEC line while cooling is a consequence of just this effect. The motion
of the system along the BEC line during the evolution, which was also
revealed, can be explained by the adiabatic expansion of the Bose gas
which has the ratio cp/cv = 5/3.
It is clear that the general trends for nonuniform systems can be un-
derstood using the results obtained for the uniform case. However, the
precise values of the relaxation times are very sensitive to the details of
the experimental situation.

Acknowledgements. This study was supported by the Russian fund of fun-

damental research (#93-02-2540) and by the Netherlands Organization
for Scientific Research (NWO).

References
[I] TJ. Greytak and D. Kleppner, in: New Trends in Atomic Physics, G.
Grynberg and R. Stora, eds. (North-Holland, Amsterdam, 1984) v.2,
p. 1127; I.F. Silvera and J.T.M. Walraven, Progress in Low Temper-
ature Physics, D.F. Brewer, ed. (North-Holland, Amsterdam, 1986),
v.10, p.39; I.F. Silvera and M. Reynolds, Journ. Low Temp. Phys.
87, 343 (1992); see articles by Silvera and by Greytak, this volume.
[2] D.W. Snoke, J.P. Wolfe, and A. Mysyrowicz, Phys. Rev B 39, 11171
(1990); A.Mysyrowicz, D.W. Snoke, and J.P. Wolfe, Phys. Stat. Sol.
159, 387 (1990); see article by Wolfe et al, this volume.
[3] M. Hasuo, N. Nagasawa, T. Itoh, and A. Mysyrowicz, Phys. Rev.
Lett. 70, 1303 (1993).
[4] C. Monroe, W. Swann, H. Robinson, and C. Wieman, Phys. Rev.
Lett. 65, 1571 (1990); see article by Castin et al, this volume.
[5] Yu. Kagan, B.V. Svistunov, and G.V. Shlyapnikov, Pis'ma Zh. Eksp.
Teor. Fiz. 42, 169 (1985) [JETP Lett. 42, 209 (1985)].
[6] Yu. Kagan, B.V. Svistunov, and G.V. Shlyapnikov, Zh. Eksp. Teor.
Fiz. 93, 552 (1987) [Sov. Phys. JETP 66, 314 (1987)].
[7] Yu. Kagan and G.V. Shlyapnikov, Phys. Lett. A130, 483 (1988).
[8] H.E. Hess, G.P. Kochanski, J.M. Doyle, N. Masuhara, D. Kleppner
and TJ. Greytak. Phys. Rev. Lett. 59, 672 (1987); R. van Roijen,
J.J. Berkhout, S. Jakkola, and J.T.M. Walraven, Phys. Rev. Lett. 61,
931 (1988).
[9] E. Fortin and A. Mysyrowicz, Phys. Rev. Lett. 70, 3951 (1993).
[10] E. Levich and V Yakhot, Phys. Rev. B 15, 243 (1977).
[II] P. Nozieres (unpublished).
 Kinetics of BEC Formation in an Interacting Bose Gas 225
[12] S.G. Tikhodeev, Zh. Eksp. Teor. Fiz. 97, 681 (1990).
[13] E. Levich and V. Yakhot, J. Phys. A 11, 2237 (1978).
[14] D. Snoke and J.P. Wolfe, Phys. Rev. B 39, 4030 (1989).
[15] H.T.C. Stoof, Phys. Rev. Lett. 66, 3148 (1991).
[16] Yu. Kagan, B.V. Svistunov, and G.V. Shlyapnikov, Zh. Eksp. Teor.
Fiz. 101, 528 (1992)[ Sov. Phys. JETP 75, 387 (1992)].
[17] J.S. Langer, Phys. Rev. 167, 183 (1968).
[18] Yu. Kagan and B.V. Svistunov, Zh. Eksp. Teor. Fiz. 105, No. 2
(1994).
[19] C.P. Barenhi, R J . Donnelly, and W.F. Vinen, Journ. Low Temp.
Phys. 52, 189 (1983).
[20] J.T. Tough, in Progress in Low Temperature Physics, D.F. Brewer,
ed. (North-Holland, Amsterdam, 1982), v.8, p.133.
[21] E.M. Lifshitz and L.P. Pitaevskii, Statistical Physics, Part 2 (Perga-
mon, Oxford, 1980).
[22] V.E. Zakharov, in Basic Plasma Physics Vol. 2, A.A. Galeev and
R.N. Sudan, eds. (North-Holland, Amsterdam, 1984).
[23] B.V. Svistunov, J. Mosccow Phys. Soc. 1, 363 (1991).
[24] L.D. Landau and E.M. Lifshitz, Statistical Physics, Part 1 (Perga-
mon, Oxford, 1980), p. 476.
[25] K.W. Schwarz, Phys. Rev. B 18, 245 (1978).
[26] K.W. Schwarz, Phys. Rev. B 38, 2398 (1988).
[27] S.V. Iordanskii, Zh. Eksp. Teor. Fiz. 49, 225 (1966)[ Sov. Phys. JETP
22, 160, (1966)].
[28] L.D. Landau and E.M. Lifshitz, Fluid Mechanics (Pergamon, Oxford,
1959).
[29] V.V. Goldman, I.F. Silvera, and A.J. Leggett, Phys. Rev. B 24, 2870
(1981).
[30] T.W. Hijmans, Yu. Kagan, G.V. Shlyapnikov, and J.T.M. Walraven,
Phys. Rev. B 48, 12886 (1993); see also this volume.
 11
Condensate Formation in a Bose Gas
H. T. C. Stooff
Department of Physics, University of Illinois at Urbana-Champaign
Urbana, Illinois 61801, USA
and
Department of Theoretical Physics, Eindhoven University of Technology
5600 MB Eindhoven, The Netherlands

Abstract
Using magnetically trapped atomic hydrogen as an example, we investigate
the prospects of achieving Bose-Einstein condensation in a dilute Bose gas.
We show that, if the gas is quenched sufficiently far into the critical region
of the phase transition, the typical time scale for the nucleation of the
condensate density is short and of O(h/kBTc). As a result we find that
thermalizing elastic collisions act as a bottleneck for the formation of the
condensate. In the case of doubly polarized atomic hydrogen these occur
much more frequently than the inelastic collisions leading to decay and we
are led to the conclusion that Bose-Einstein condensation can indeed be
achieved within the lifetime of the gas.

1 Introduction
In the last few years it has been clearly demonstrated that not only
charged ions but also neutral atoms can be conveniently trapped and
cooled by means of electro-magnetic fields. Although the physics of
the various ingenious scenarios developed to accomplish this is already
interesting in itself [1], the opportunities offered by an atomic gas sample
at very low temperatures are exciting in their own right. Examples in this
respect are the performance of high-precision spectroscopy, the search
for a violation of CP invariance by measuring the electric dipole moment
of atomic cesium [2], the construction of an improved time standard
based on an atomic fountain [3] and the achievement of Bose-Einstein
condensation in a weakly interacting gas.
In particular the last objective has been pursued mainly with atomic
t Present address: Institute for Theoretical Physics, University of Utrecht, Utrecht,
The Netherlands.

226
 Condensate Formation in a Bose Gas 227
hydrogen [4, 5]. It has been recently proposed that the alkali-metal
vapors cesium [6] and lithium [7] are also suitable candidates for Bose-
Einstein condensation. We will nevertheless concentrate here on atomic
hydrogen, because it still seems to be the most promising system for
the observation of the phase transition in the near future. (See the
review articles by Greytak and Silvera in this volume.) Moreover, it has
the advantage that the atomic interaction potential is known to a high
degree of accuracy. As a result we can have confidence in the fact that
the scattering length is positive, which is required for the condensation to
take place in the gaseous phase [8], and small enough to rigorously justify
the approximations made in the following for the typical temperatures
(T ~ 10 JUK) and densities (n ~ 1014 cm"3) envisaged in the experiments.
Due to the spin of the electron and the proton, the ls-hyperfine
manifold of atomic hydrogen consists of four states which are in order of
increasing energy denoted by \a), \b), \c)9 and \d), respectively. Only the
\c) and \d) states can be trapped in a static magnetic trap, because in a
magnetic field they have predominantly an electron spin-up component
and are therefore low-field seeking [9]. Furthermore, if we load a trap with
atoms in these two hyperfine states, the \c) state is rapidly depopulated
as a result of the much larger probability for collisional relaxation to the
high-field seeking \a) and \b) states which are expelled from the trap. In
this manner the system polarizes spontaneously and we obtain a gas of
|d) -state atoms, known as doubly spin-polarized atomic hydrogen since
both the electron and the proton spin are directed along the magnetic
field. Unfortunately, such a doubly polarized hydrogen gas still decays
due to the dipole interaction between the magnetic moments of the
atoms. Although the time scale Tinei for this decay is much longer than
the time scale for the depopulation of the \c) state mentioned above, it
nevertheless limits the lifetime of the gas sample to the order of seconds
for the densities of interest [10].
Having filled the trap with doubly polarized atoms, we must subse-
quently lower the temperature of the gas to accomplish Bose-Einstein
condensation. At present it is believed that the most convenient way to
achieve this is by means of conventional [11, 12] or light-induced [13]
evaporative cooling. In both cases the idea is to remove, by lowering
the well-depth or by photon absorption in the perimeter, the most en-
ergetic particles from the trap and thus to create momentarily a highly
nonequilibrium energy distribution that will evolve into a new equilib-
rium distribution at a lower temperature. According to the quantum
Boltzmann equation describing this process, a typical time scale for the
228 H. T. C. Stoof
evolution is the average time between two elastic collisions rei = l/n(v(r),
with (vo) the thermal average of the relative velocity v of two colliding
atoms times their elastic cross section a. Clearly, rei must be small com-
pared to Tinei to ensure that thermal equilibrium is achieved within the
lifetime of the system. As a result, the minimum temperature that can
be reached by evaporative cooling is about 1 /xK and indeed below the
critical temperature of atomic hydrogen at a density of 1014 cm"3.
The previous discussion appears to indicate that a typical time scale
for the formation of the condensate is given by rei. However, this is
not correct because simple phase-space arguments show that a kinetic
equation cannot lead to a macroscopic occupation of the one-particle
ground state: considering a homogeneous system of N bosons in a
volume V, we find from the Boltzmann equation that the production
rate of the condensate fraction is
d No
(1)
dt N
where No is the number of particles in the zero-momentum state and C
is a constant of 0(1). Hence, in the thermodynamic limit (N, V —• oo in
such a way that their ratio n = N/V remains fixed) a nonzero production
rate is only possible if a condensate already exists [14] and we are forced
to conclude that Bose-Einstein condensation cannot be achieved by
evaporative cooling of the gas.

2 Nucleation
In the above argument we have only considered the effect of two-body
collisions. It is therefore legitimate to suggest that perhaps three- or
more body collisions are required for the formation of the condensate,
even though they are very improbable in a dilute gas [15]. However, we
can easily show that the same argument also applies to these processes:
in a m-body collision that produces one particle with zero momentum,
we have 2m —2 independent momentum summations, leading to a factor
of V2m~2. Moreover, the transition matrix element is proportional to
V - V~m due to the integration over the center-of-mass coordinate and the
normalization of the initial and final state wave functions, respectively. In
total, the production rate for the condensate fraction is thus proportional
to vlm-\Vx-m)2V-\\ +N0) or V-^l+No), which again vanishes in the
thermodynamic limit if there is no nucleus of the condensed phase. As
expected, the contributions from collisions that produce more than one
 Condensate Formation in a Bose Gas 229
zero-momentum particle have additional factors of V~l and vanish even
more rapidly if V —> oo.
Clearly, we have arrived at a nucleation problem for the achievement
of Bose-Einstein condensation which seriously endangers the success of
future experiments. Fortunately, we suspect that the line of reasoning
presented above is not completely rigorous because otherwise it implies
that liquid helium also cannot become superfluid, in evident disagreement
with experiment. Indeed, by using a kinetic equation to discuss the time
evolution of the gas we have in effect neglected the buildup of coherence
which is crucial for the formation of the condensate. Our previous
argument therefore only shows that by means of evaporative cooling
the gas is quenched into the critical region on a time scale xei, not
that Bose-Einstein condensation is impossible. To discuss that point we
need a different approach that accurately describes the time evolution of
the system after the quenching by taking the buildup of coherence into
account exactly. Such a nonequilibrium approach was recently developed
on the basis of the Keldysh formalism and can, in the case of a dilute
Bose gas, be seen as a generalization of the Landau theory of second-
order phase transitions [16, 17]. As a consequence it is useful to consider
the Landau theory first. This leads to a better understanding of the
more complicated nonequilibrium theory and ultimately of the physics
involved in the nucleation of Bose-Einstein condensation.

2.1 Landau Theory

As an introduction to the Landau theory of second-order phase transi-
tions we use the example of a ferromagnetic material [17]. To be more
specific we consider a cubic lattice with spins S, at the sites {i}. The
Hamiltonian is taken to be
» rSj, (2)
(Uj)

where J is the exchange energy and the sum is only over nearest neighbors.
For further convenience we also introduce the magnetization

Physically it is clear that this model has a phase transition at a critical

temperature Tc of O(J/&#). Above the critical temperature the thermal
fluctuations randomize the direction of the spins and the system is in
230 H. T. C. Stoof
a disordered (paramagnetic) state having a vanishing average magne-
tization (M) eq . However, below the critical temperature the thermal
fluctuations are not large enough to overcome the directional effect of
the Hamiltonian and the spins favor an ordered (ferromagnetic) state
with (M) 7^ 0. The different phases of the material are thus conve-
niently characterized by the average magnetization, which for this reason
is known as the order parameter of the ferromagnetic phase transition.
In the phenomenological approach put forward by Landau the above
mentioned temperature dependence of the equilibrium order parameter
(M) is reproduced by anticipating that the free-energy density of the
system at a fixed but not necessarily equilibrium value of the average
magnetization has the following expansion

/((M), T) ~ /(0, T) + a(T)(M) 2 + ^ ( M ) 4

(4)

for small values of (M), and that the coefficients of this expansion behave
near the critical temperature as

( ^ ) (5)

and
pm^Po, (6)
respectively, with ao and /?o positive constants.
Hence, above the critical temperature a(T) and fi(T) are both positive.
As a result the free energy is minimal for (M) = 0, which corresponds
exactly to the paramagnetic phase. Moreover, for temperatures below the
critical one a(T) is negative and the free energy is indeed minimized by
a nonzero average magnetization with magnitude y/—ot(T)/f}(T). Just
below the critical temperature the latter equals

(V)

which after substitution in Eq. (4) gives rise to an equilibrium free-energy

density of
V2

> T ) / ( 0 , T ) ^ ( l (8)

Therefore, the second derivative d2f/dT2 is discontinuous at the critical

temperature and the phase transition is of second order according to the
Ehrenfest nomenclature.
 Condensate Formation in a Bose Gas 231
Note that minimizing the free energy only fixes the magnitude and not
the direction of (M) . This degeneracy is caused by the fact that the
Hamiltonian in Eq. (2) is symmetric under an arbitrary rotation of all
the spins S*. Consequently, the free energy must be symmetric under a
rotation of the average magnetization and only even powers of (M) can
appear in its expansion (cf. Eq. (4)). Due to this behavior the ferromagnet
is a good example of a system with a spontaneously broken symmetry, i.e.
although the Hamiltonian is invariant under the operations of a group,
its ground state is not. In the case of a ferromagnet the symmetry group
is 50(3), which is broken spontaneously below the critical temperature
because the average magnetization points in a certain direction. Which
direction is chosen in practice depends on the surroundings of the system
and in particular on (arbitrary small) external magnetic fields that favor
a specific direction.
After this summary of the Landau theory we are now in a position
to introduce two time scales which turn out to be of great importance
for the nucleation of Bose-Einstein condensation. To do so we consider
the following experiment. Imagine that we have a piece of ferromagnetic
material at some temperature T\ above the critical temperature. Being
in thermal equilibrium the material is in the paramagnetic phase with
(M) = 0. We then quickly cool the material to a new temperature T^
below the critical temperature. If this is done sufficiently fast, the spins
will have no time to react and we obtain a nonequilibrium situation in
which the free energy has developed a 'double-well' structure but the
average magnetization is still zero. This is depicted in Fig. l(a). In such
a situation there is a typical time scale for the relaxation of the average
magnetization to its new equilibrium value y/—(x(T2)/P(T2)9 which we
denote Tcoh-
However, in the case of magnetically trapped atomic hydrogen, the
gas is isolated from its surroundings and it is not possible to perform the
cooling stage mentioned above. As a result the gas has to develop the
instability associated with the phase transition by itself. The time scale
corresponding to this process is called Tnuci and is schematically shown
in Fig. l(b). Combining the two processes we are led to the following
physical picture for the nucleation of Bose-Einstein condensation. After
the quench into the critical region the gas develops an instability on the
time scale Tnuci. On this time scale the actual nucleation takes place and a
small nucleus of the condensate is formed, which then grows on the time
scale TCoh as shown in Fig. 2. To solve the nucleation problem we are
thus left with the actual determination of these two time scales. Clearly,
232 H. T. C. Stoof

[a)

Fig. 1. Visualization of (a) the time scale rCOh for the relaxation of the order
parameter to its equilibrium value and (b) the time scale Tnuci associated with the
appearance of the instability.

before this can be done we need to know the correct order parameter of
the phase transition.

2.2 Order Parameter

Starting with the pioneering work of Bogoliubov [18], it is well known
that the order parameter for Bose-Einstein condensation in a weakly
interacting Bose gas is conveniently treated by using the method of
second quantization. In this method, all many-body observables are
expressed in terms of the creation and annihilation operators of a particle
at position x denoted by \p\\) and xp(x\ respectively [19]. For example,
for a gas of particles with mass m and a two-body interaction potential
 Condensate Formation in a Bose Gas 233

(b)

Fig. 2. Visualization of the time scales iCOh and Tnuci, using the time dependence
of (a) the order parameter and (b) the coefficient a of the quadratic term in the
free energy.

V(x — x'), the Hamiltonian equals

+ l- I dx [ dx! xp\x)\p\x!)V{x - x!)xp{x!)xp{x) (9)

and the total number of particles is given by

N = [ dx\p\x)\p{x). (10)

The method is also particularly useful for a Bose system because the per-
mutation symmetry of the many-body wave function is automatically ac-
counted for by assuming the Bose commutation relations [\p(x),\p(xf)] =
[xp^x^xp^x')] = 0 and [t/^xXt/^x')] = d(x — xf) between the creation
and annihilation operators.
234 H. T. C. Stoof
In the language of second quantization, the order parameter for the
dilute Bose gas is the expectation value (ip(x)). Analogous to the case
of the ferromagnetic phase transition, a nonzero value of this order pa-
rameter signals a spontaneously broken symmetry. Here the appropriate
symmetry group is (7(1), since the Hamiltonian of Eq. (9) is invariant
under the transformation xp(x) —• xp(x)eiS and y^(x) —• xp^(x)e~i9 of the
field operators, whereas their expectation values are clearly not. Notice
that the (7(1) symmetry of the Bose gas is closely related to the conser-
vation of particle number. This is most easily seen by observing that
the invariance of the Hamiltonian is due to the fact that each term in
the right-hand side of Eq. (9) contains an equal number of creation
and annihilation operators. The relationship can also be established in
a more formal way by noting that the (7(1) gauge transformations are
generated by the particle number operator. As we will see later on, it has
important consequences for the dynamics of the order parameter.
To understand why (ip(x)) is the order parameter associated with
Bose-Einstein condensation, it is convenient to use a momentum-space
description and to introduce the annihilation operator for a particle with
momentum hk

(11)

and the corresponding creation operator a^ by Hermitian conjugation.

The basis of states for the gas is then characterized by the occupation
numbers {N^}. If the gas is Bose-condensed, there is a macroscopic
occupation of the zero-momentum state and the relevant states are
|N0, {Nk}k=£o) ^ h on ly No proportional to N. Within this subspace of
states we have

(ao^o) = (No) * (No) + 1 = ( W > (12)

and we can neglect the fact that ao and a^ do not commute. As a result
we can treat these operators as complex numbers [18, 19] and say that
(OO^OQ) = (ao*)(ao) or equivalently that (ao) = y/No- In coordinate space,
the latter reads (ip(x)) = Jno, with no = No/V the condensate density.
The above argument essentially tells us that a sufficient condition for
a nonzero value of (xp(x)) is (No) > 1. Although this is intuitively
appealing, it is important to point out that it is not generally true.
Consider, for example, the ideal Bose gas [20]. In the grand canonical
 Condensate Formation in a Bose Gas 235
ensemble, the total number of particles in the gas is given by

where /? is l/fc^T, 6k is the kinetic energy h2k2/2m, £ is the fugacity e^

and fi is the chemical potential.
At high temperatures, the fugacity is small and we are allowed to take
the continuum limit of Eq. (13), which results in the equation of state

^ (14)

where we have introduced the de Broglie wavelength A = yj2nh2/mkBT

and the Bose functions gn(() defined by

Lowering the temperature while keeping the density fixed, the fugacity
increases until it ultimately reaches the value one at the critical temper-
ature
T =
2 (n ^ ( n \ 2 n 2*tf/nx2/3
° mkB Vg3/2(1)/ " mkB
At this point, Eq. (14) ceases to be valid because the occupation number
of the zero-momentum state, which is equal to £/(l — f), diverges and
must be taken out of the discrete sum in Eq. (13) before we take the
continuum limit. Moreover, we only need to treat the zero-momentum
term separately because in the thermodynamic limit the chemical po-
tential goes to zero as F" 1 , whereas the kinetic energy for the smallest
nonzero momentum decreases only as F~ 2/3 . Consequently, below the
critical temperature the equation of state becomes

and leads to a condensate density given by the well-known formula

3/2\
j • (18)
We thus find that the average occupation number (No) is at all temper-
atures given by C/(l—C)J t n a t is, its value in the grand canonical ensemble
with the density matrix e-P(R-^N), Since this density matrix commutes
with the particle number operator, we conclude that in the case of an
236 H. T. C. Stoof
ideal Bose gas there is a macroscopic occupation of the zero-momentum
state without a spontaneous breaking of the 17(1) symmetry. To show
more rigorously that (xp(x)) = 0 at all temperatures, we determine the
free-energy density of the gas as a function of the order parameter (\p(x)).
Dealing with a noninteracting system, it is not difficult to obtain
f((ip(x)),T) = -fi(T)\(xp(x))\2 (19)
for a homogeneous value of the order parameter. Because \i < 0 the
minimum is indeed always at (xp(x)) = 0 and it is necessary to identify
the condensate density no with the order parameter of the ideal Bose gas
(cf. Eq. (18)).
Notwithstanding the previous remarks, the order parameter for Bose-
Einstein condensation in a weakly interacting Bose gas is given by (xp(x)).
This was put on a firm theoretical basis by Hugenholtz and Pines [21],
who calculated the free energy as a function of the above order parameter
and showed that at sufficiently low temperatures the system develops an
instability that is removed by a nonzero value of (i/?(x)). In addition,
they derived an exact relationship between the chemical potential and the
condensate density, which turns out to be valid also in the nonequilibrium
problem of interest here and is important for an understanding of how
the 1/(1) symmetry is broken dynamically.

2.3 Condensation Time

We have argued that by means of evaporative cooling a doubly polarized
atomic hydrogen gas can be quenched into the critical region of the
phase transition and that this kinetic part of the condensation process is
described by a quantum Boltzmann equation. As a result, the gas acquires
on the time scale TC\ an equilibrium distribution with some temperature
T, which is slightly above the critical temperature To of the ideal Bose
gas because a condensate cannot be formed at this stage.
For the study of the subsequent coherent part of the condensation
process, it is therefore physically reasonable to assume that at a time to
the density matrix p(to) of the gas is well approximated by the density
matrix of an ideal Bose gas with temperature T. The evolution of
the order parameter (ip(x)) for times larger than to is then completely
determined by the Heisenberg equation of motion

ll = Mx,t),H] (20)
 Condensate Formation in a Bose Gas 237
for the field operator. Substituting the Hamiltonian in Eq. (9), and taking
the expectation value with respect to p(to\ we find

ihd{xpftJ)) = = ^ (V(x, r )> + J dx' V(x - x')(tpV, f)tp(x', t)xp(x, t)),

(21)
where the complicated part is the evaluation of (v^(x', t)\p(x',t)xp(x, t)).
In lowest order, we simply have
x,t)) ~ (ipHx',t))(xp(x',t))(xp(x,t))
t
,0Xv(x/,r)>, (22)
which, after substitution into Eq. (21), leads to
(v;(x 0) =
' f
+ / dx' V(x -
+ / dx' V(x - x'XvV, t))(xp(x\ t))(xp(x, t))9 (23)
and thus corresponds exactly to the Hartree-Fock approximation.
To proceed, we must restrict ourselves to the case of a dilute Bose
gas in the quantum regime. Introducing the scattering length a, which
is of the order of the range of the interaction, the quantum regime
is characterized by a/A <C 1. We therefore only need to consider s-
wave scattering and can neglect the momentum dependence of various
collisional quantities. In particular, we can replace the potential V(x — x')
by the contact interaction V05(x — x') with Vo = f dx V(x). Hence, in
the Hartree-Fock approximation we obtain

2
) V(*,t)>, (24)

having only the trivial solution (xp(\9t)) = 0 for a space and time-
independent order parameter. Within this lowest order approximation
we thus conclude that Tnuci = oo and that the formation of a condensate
will not take place.
Fortunately, it is well known that the Hartree-Fock approximation
is not sufficiently accurate for a dilute Bose gas because the diluteness
condition na3 <C 1 implies that we need to consider all two-body pro-
cesses, i.e., two particles must be allowed to interact more than once. The
appropriate approximation is therefore the ladder or T-matrix approx-
imation, which is displayed in terms of diagrams in Fig. 3. Moreover,
in the degenerate regime, where the temperature T is slightly larger
than To and the degeneracy parameter nA3 is of 0(1), the condition
238 H. T. C. Stoof

Fig. 3. Diagrammatic representation of the T-matrix equation. The wavy line

corresponds to the interaction and the straight line to-the noninteracting one-
particle Green's function.

a/A <C 1 implies that also naA2 < 1 or physically that the average ki-
netic energy of the gas is much larger than the typical interaction energy.
Consequently, an accurate discussion of the nucleation of Bose-Einstein
condensation in a weakly interacting Bose gas requires an evaluation of
(v?t(x', t)xp(x\ t)xp(x, t)) within the T-matrix approximation and to zeroth
order in the gas parameters a/A and naA2.
Although it is easy to formulate this objective, to actually perform the
calculation is considerably more difficult. It is most conveniently accom-
plished by making use of the Keldysh formalism [22] which has been
reviewed by Danielewicz [23] using operator methods. For a functional
formulation of this nonequilibrium theory and for the technical details of
the somewhat tedious mathematics, we refer to our previous papers [16].
Here we only present the final results and concentrate on the physics
involved.
Due to the fact that we are allowed to neglect the (relative) momentum
dependence of the T matrix, the equation of motion for the order
parameter (tp(x, t)) acquires the local form of a time-dependent Landau-
Ginzburg theory

^d h2V2
ih (xp(x,t)) = , (25)
Jt

which is recovered from a variational principle if we use the action

 Condensate Formation in a Bose Gas 239

s(+

Fig. 4. Time dependence of the coefficient for three different initial temper-
atures of the Bose gas.

S((xp(x,t)),T) = J dt J dx{ip(x,t)y

x in— . (26)

Here S^+\t)8(t — tf) is a simple approximation for the retarded self-energy

ftE(+)(0;£, tr) of a hydrogen atom with zero momentum and T (+) ~
4nh2a/m is the effective interaction between two such atoms. Clearly, the
action in Eq. (26) is the desired generalization of the Landau free energy
and corresponds precisely to the physical picture presented previously in
Figs. l(b) and 2(b).
Therefore, Tnuci is determined by the time dependence of the coefficient
S (+) which is shown in Fig. 4 for three different initial temperatures.
If the temperature T is much larger than To, S^+\t) is constant and
equal to $nh2an/m. In this region of the phase diagram, coherent
processes are negligible and the evolution of the gas is described by a
Boltzmann equation. Lowering the temperature, the occupation numbers
for momenta hk < O(h/A) increase and lead to an enhancement of the
coherent population of states with momenta hk < O(hjna) <C O(h/A).
This is signaled by the increasing correlation length £ = h/y/2mS^(co).
At the critical temperature Tc = T0(l + O(a/A0)), we have S(+)(oo) = 0
and this correlation length diverges. Below that temperature, but still
above To so as not to have a condensate already in the initial state,
()
actually changes sign and the gas develops the required instability
240 H. T. C. Stoof
for a Bose-Einstein condensation. The change of sign takes place at
H
) , (27)
kB(Tc- T)J
which shows that i nuc i is in general of O(h/kBTc) except for temperatures
very close to the critical temperature. Clearly, this time scale is due to the
fact that all states with momenta hk < O(h/A) cooperate in the coherent
population of the one-particle ground state.
After a small nucleus of O(n(a/Ac)2) has been formed, the subsequent
buildup of the condensate density is determined by the equation of
motion Eq. (25). Looking at the right-hand side, we immediately see
that the time scale icoh involved in this process is typically of order
h/n0T{+) or, equivalent^, of O((h/kBTc)(l/n0aA2c)). Therefore, we find
as
?coh > Trmcb anticipated in Fig. 2. The physical reason for this time
scale is that after the nucleation of the phase transition, the buildup of the
condensate density is accompanied by a depopulation of the momentum
states with hk < O(h^jnoa). As a result, it is not difficult to show that in
the limit t —• oo, the condensate density is of O(na/Ac) and thus we have
TCoh = O((Ac/a)2h/kBTc).
Finally, it is interesting to point out how the gas can conserve the total
number of particles and, apparently at the same time, break the 1/(1)
gauge symmetry that is responsible for this conservation law. To that
end, we write the field operator ip(x, t) as the sum of its expectation value
(xp(x9t)) and the fluctuation i//(x, t\ and introduce a time-dependent
chemical potential fi(t) by means of

(£ j* j (28)
Substituting the latter into the action of Eq. (26) and minimizing with
respect to jno(t) gives, for t > tC9

n
o(t) =

which determines the growth of the condensate density and is in effect a

nonequilibrium version of the Hugenholtz-Pines theorem [21]. Further-
more, by considering the fluctuations around y/no(t), we can show that
the chemical potential is determined by the constraint

n = no(t) + 1 J dx (xp'\x, t)tpr(x, t)), (30)

enforcing the conservation of particle number at all times. In the complex
 Condensate Formation in a Bose Gas 241

Fig. 5. Evolution of the complex order parameter (y?(x)), which is constrained

by the requirement of particle number conservation.

plane, (y>(x, t)) thus moves radially outward along a spiral, as shown in
Fig. 5. Consequently, the phase of the order parameter never has a fixed
value and the U(l) symmetry is not really broken dynamically. This is
of course expected since the system evolves according to a symmetric
Hamiltonian.

3 Conclusions and Discussion

We studied the evolution of a doubly polarized atomic hydrogen gas in
a magnetic trap and showed that by means of evaporative cooling the
gas can accomplish the Bose-Einstein phase transition within its lifetime
Tinei. The condensation process proceeds under these conditions in the
following three stages. In the first kinetic stage, the gas is quenched into
the critical region To < T < Tc. A typical time scale in this part of the
evolution is given by the time between elastic collisions i e i, which for a
degenerate gas is of O((A c/a)2ft//c#Tc). In the following coherent stage,
the actual nucleation takes place on the time scale TnUci = O(h/kBTc) by
means of a coherent population of the zero-momentum state. The small
nucleus formed in this manner then grows on the much longer time scale
= O((Ac/a)2h/kBTc) by a depopulation of the low-momentum states
242 H. T. C. Stoof
having hk < O(h/noa). In the third and last stage of the evolution,
the Bogoliubov quasiparticles produced in the previous stage have to
come into equilibrium with the condensate. This process can again be
treated by a kinetic equation and was studied by Eckern [24], who found
that the corresponding relaxation time irei is of order (Ac/a)3H/kBTc.
In the case of atomic hydrogen, this turns out to be comparable to
the lifetime of the system. Summarizing, we thus have the sequence
Tnuci <C ?coh — Tei ^ TTQ\ ~ Tinei for the various time scales involved in the
phase transition. The most important requirement for the achievement
of the phase transition is therefore rei <C Tinei, which is a relatively mild
requirement and should not pose an insurmountable problem for future
experiments aimed at the realization of Bose-Einstein condensation.
Having arrived at this conclusion, it is necessary to discuss recent
work by Kagan, Svistunov and Shlyapnikov [25] that also considers
the evolution of a weakly interacting Bose gas after the removal of the
most energetic atoms. Kagan and coworkers also conclude that the
evolution of the gas is divided into a kinetic and a subsequent coherent
stage. Moreover, their detailed study of the kinetic part of the evolution
confirms our conjecture that the gas is quenched into the critical region on
the time scale rei. The investigation of the coherent part, however, leads
to the extreme result that complete Bose-Einstein condensation cannot
occur in a finite amount of time. To understand why this conclusion is
reached we briefly summarize their line of thought. For more details, see
the review article by Kagan in this volume.
At the end of the kinetic stage, the gas has acquired large average
occupation numbers for the states with momenta hk < hko = O(hjnoa),
where no is the density of particles with these small momenta. Therefore,
Kagan, Svistunov and Shlyapnikov argue that for a study of the coherent
part of the evolution we must use the initial condition

(31)

together with the nonlinear Schrodinger equation

) v(x,t)), (32)

which has the equilibrium solution (tp(x, t)) = Jn$ exp(—i^oi) and /zo =
noT (+) . Consequently, all the particles that initially have momenta hk <
hko are in the limit t —• oo assumed to be in the condensate.
 Condensate Formation in a Bose Gas 243
Linearizing the Hamiltonian (32) around this equilibrium solution,
Kagan et a\. then observe that the energy involved with a magnitude
fluctuation of the order parameter is e^ + noT(+), whereas the energy
involved with a phase fluctuation is only e^. As a result, they assert
that on the time scale Tampi = Tcoh = 0(h/noT^), a state is formed in
which the amplitude of (xp(x,t)) is fixed, but the phase is still strongly
fluctuating because the corresponding time scale xph is much longer (and
even diverges as V2^ in the thermodynamic limit). Hence, at finite time,
the gas is in a state with only a quasicondensate [25, 26] and a complete
or true condensate is only formed in the limit t —• oo.
Clearly, this physical picture of two different time scales for the am-
plitude and phase fluctuations of the order parameter is only applicable
if these fluctuations exist independently of each other. Looking only at
the Hamiltonian this indeed seems to be the case. However, a correct
discussion of the fluctuations must be based on the equations of mo-
tion or, equivalently, the Lagrangian. The latter contains a first-order
time derivative which strongly couples the amplitude and phase fluctu-
ations. Therefore, a dilute Bose gas only has one collective mode (not
two) with a single dispersion relation, i.e. the well-known Bogoliubov
dispersion \A"k(ek + 2YIQT^\ and we are lead to TPH = Tampi- It is in-
teresting to note that in the case of a neutral BCS-type superfluid, we
do have two different time scales because the Lagrangian now contains
a second-order time derivative and the amplitude and phase fluctuations
are indeed independent in lowest order [27, 28].
An even more serious problem with the approach of Kagan, Svistunov
and Shlyapnikov is their claim that the use of the initial condition in Eq.
(31) is justified because (N^) > 1. As we have pointed out earlier, this is
not true in general. For (\p(x91)) to be nonzero, one must show that the
system has a corresponding instability. However, within the T -matrix
approximation it is not difficult to show that the instability associated
with a quasicondensate is always preceded by the instability correspond-
ing to the formation of a condensate. This implies that we always have to
take Bose-Einstein condensation into account first. After that has been
accomplished by means of the theory reviewed here, it is of course no
longer relevant to consider the appearance of a quasicondensate.

Acknowledgments. It is a great pleasure to thank Tony Leggett for

various helpful discussions and for giving me the opportunity to visit
the University of Illinois at Urbana-Champaign. I also benefited from
conversations with Steve Girvin, Daniel Loss, Kieran Mullen and Jook
244 H. T. C. Stoof
Walraven. This work was supported by the National Science Foundation
through Grant No. DMR-8822688.

References
[I] See the special issue of J. Opt. Soc. Am. B 6, No. 11 (1989), edited
by S. Chu and C. Wieman.
[2] W. Bernreuther and M. Suzuki, Rev. Mod. Phys. 63, 131 (1991).
[3] A. Clairon, C. Salomon, S. Guellati, and W.D. Phillips, Europhys.
Lett. 12, 683 (1990); K. Gibble and S. Chu, Phys. Rev. Lett. 70,
1771 (1993).
[4] H.F. Hess, G.P. Kochanski, J.M. Doyle, N. Masuhara, D. Kleppner,
and TJ. Greytak, Phys. Rev. Lett. 59, 672 (1987); N. Masuhara,
J.M. Doyle, J.C. Sandberg, D. Kleppner, T.J. Greytak, H.F. Hess,
and G.P. Kochanski, Phys. Rev. Lett. 61, 935 (1988).
[5] R. van Roijen, J.J. Berkhout, S. Jaakkola, and J.T.M. Walraven,
Phys. Rev. Lett. 61, 931 (1988).
[6] C. Monroe, W. Swann, H. Robinson, and C. Wieman, Phys. Rev.
Lett. 65, 1571 (1990).
[7] J.J. Tollett, C.C. Bradley, and R.G. Hulet, Bull. Am. Phys. Soc. 37,
1126(1992).
[8] H.T.C. Stoof, to be published.
[9] For a review we refer to T.J. Greytak and D. Kleppner, in New
Trends in Atomic Physics, edited by C. Grynberg and R. Stora
(North-Holland, Amsterdam, 1984), p. 1125 and I.F. Silvera and
J.T.M. Walraven, in Progress in Low Temperature Physics, edited by
D.F. Brewer (North-Holland, Amsterdam, 1986), Vol. 10, p. 139.
[10] A. Lagendijk, I.F. Silvera, and B.J. Verhaar, Phys. Rev. A 33, 626
(1986); H.T.C. Stoof, J.M.V.A. Koelman, and B.J. Verhaar, Phys.
Rev. B 38, 4688 (1988).
[II] J.M. Doyle, J.C. Sandberg, LA. Yu, C.L. Cesar, D. Kleppner, and
TJ. Greytak, Phys. Rev. Lett. 67, 603 (1991).
[12] O J . Luiten, H.G.C. Werij, I.D. Setija, T.W. Hijmans, and J.T.M.
Walraven, Phys. Rev. Lett. 70, 544 (1993).
[13] I.D. Setija, H.G.C. Werij, O J . Luiten, M.W. Reynolds, T.W. Hijmans,
and J.T.M. Walraven, Phys. Rev. Lett. 70, 2257 (1993).
[14] See also E. Levich and V. Yakhot, Phys. Rev. B 15, 243 (1977) and
S.G. Tikhodeev, Zh. Eksp. Teor. Fiz. 97, 681 (1990) [Sov. Phys.-JETP
70, 380 (1990)].
[15] D.W. Snoke and J.P. Wolfe, Phys. Rev. B 39, 4030 (1989).
 Condensate Formation in a Bose Gas 245
[16] H.T.C. Stoof, Phys. Rev. Lett. 66, 3148 (1991); Phys. Rev. A 45,
8398 (1992).
[17] For a more detailed discussion see, for instance, S. Ma, Modern
Theory of Critical Phenomena (Addison-Wesley, New York, 1976)
and J.W. Negele and H. Orland, Quantum Many-Particle Systems
(Addison-Wesley, New York, 1988).
[18] N.N. Bogoliubov, J. Phys. USSR (Moscow) 11, 23 (1947).
[19] See, for example, A.L. Fetter and J.D. Walecka, Quantum Theory of
Many-Particle Systems (McGraw-Hill, New York, 1971).
[20] K. Huang, Statistical Mechanics, second edn (Wiley, New York,
1987).
[21] N.M. Hugenholtz and D. Pines, Phys. Rev. 116, 489 (1958).
[22] L.V. Keldysh, Zh. Eksp. Teor. Fiz. 47, 1515 (1964) [Sov. Phys.-JETP
20, 235 (1965)].
[23] P. Danielewicz, Ann. Phys. (N.Y) 152, 239 (1984).
[24] U. Eckern, J. Low Temp. Phys. 54, 333 (1984).
[25] Yu. M. Kagan, B.V. Svistunov and G.V. Shlyapnikov, Zh. Eksp.
Teor. Fiz. 101, 528 (1992) [Sov. Phys.-JETP 74, 279 (1992)]; Sov.
Phys.-JETP 75, 387 (E) (1992); B.V. Svistunov, J. Moscow Phys. Soc.
1, 373 (1991); Yu. Kagan, this volume.
[26] V.N. Popov, Functional Integrals in Quantum Field Theory and Sta-
tistical Physics (Reidel, Dordrecht, 1983).
[27] N.N. Bogoliubov, V.V. Tolmachev, and D.N. Shirkov, New Methods
in the Theory of Superconductivity (Consultants Bureau Enterprises,
New York, 1959).
[28] RW. Anderson, Phys. Rev. 112, 1900 (1958).
 12
Macroscopic Coherent States of Excitons
in Semiconductors
L. V. Keldysh
P. N. Lebedev Physics Institute
Leninsky Prospect 53
117924 GSP, Moscow B-333
Russia

Abstract
Initially put forward by Moskalenko [1] and Blatt et al. [2], the idea of a
possible Bose-Einstein condensation (BEC) of excitons in semiconductors
has attracted the attention of both experimentalists and theoreticians for
more than three decades. At different stages of this long history, the results
of their efforts have been described and discussed in review articles [3-10].
A brief introduction and summary of the main qualitative conclusions of
this older work is presented here (Sections 1 and 2), followed by a more
detailed discussion of some more recent developments (Sections 3 and 4).

1 Electronic Excitations in Semiconductors

Schematically presented in Fig. 1 is the typical electronic spectrum of
a semiconductor: two bands (or two groups of bands) of continuous
spectrum - conduction (c) and valence (v) - separated by the energy gap
Eg = £c,min — EViJnax. In the ground state, all of the states in the valence
band(s) are occupied by valence electrons of the semiconductor, and all
states in the conduction band are empty.
The lowest single-particle electronic excitations are an additional elec-
tron (e) in the conduction band or a single empty state - a hole (h) -
in the valence band. Both of these excitation types are mobile fermions
(spin = 1/2) characterized by effective masses, me and mj, and effective
charges ee = e and eh = —e, respectively. Here e is the usual (negative)
elementary charge. Effective charges of electrons and holes are universal,
unlike the effective masses, which are different not only in different semi-
conductors but also in different bands of the same semiconductor and
for different directions in the same band (anisotropy).

246
 Macroscopic Coherent States of Excitons in Semiconductors 247

-
Conduction band

• e
'-' cmin

E g -Eo 2(E g -E o )-E D

Oh
Oh
Valence band

Fig. 1. Schematic picture of the main components of the nonequilibrium electron-

hole plasma - free electrons and holes, excitons and excitonic molecules - with
their characteristic energies. The top and bottom represent the continuous
spectrum of single-particle excitations - electrons and holes. The energy levels
of electrons and holes bound in excitons or excitonic molecules have no exact
meaning and are depicted only for the sake of illustration. Only the total energy
of a complex as a whole is well defined.

Being oppositely charged (quasi)free particles, the electron and hole

attract one another by Coulomb forces just as the electron and proton
do. And like an electron and proton which can be bound into a hydrogen
atom, the electron and hole can be bound via this interaction into the two-
particle excitation called the Wannier-Mott exciton [11, 12]. Being the
combination of two fermions, an exciton is expected to be a boson, and
so the problem of the possibility of BEC of excitons arises. But to discuss
this problem, we will need more information about the possible states
and transformations in the system of many excitons, or, more correctly,
in the many electron-hole system. The above-mentioned similarity to
the electron-proton system can be extended to the many-particle system
and so suggests the existence of the excitonic molecule (biexciton), the
bound state of two excitons similar to a hydrogen molecule [13, 14] and,
moreover, the gas-liquid-type phase transition of exciton and biexciton
"gas" into an electron-hole Fermi liquid (EHL) - the bound state of a
macroscopically large number of electrons and holes. [15, 6, 9, 10]
But this similarity is not complete. The electron-hole system possesses
248 L. V. Keldysh
some peculiar properties distinguishing it from any other. The most
important are
(1) Greatly reduced Coulomb interaction due to dielectric screening in
the host crystal. Typical values of dielectric constant 6 in semiconductors
are of the order of 10. Therefore, by the well-known Bohr formulas, the
binding energy of a hydrogen-like exciton

is a few orders of magnitude smaller than the atomic binding energy, and
the exciton effective radius
eh2
a0 = —- 2~ 10" 7 - l(T6cm (2)
me
is macroscopically large compared to interatomic distances in the host
crystal. Formulas (1) and (2) represent the natural units of quantum
scales of energy and length in the interacting electron-hole system, in
just the same sense as the Rydberg and Bohr radius are the natural scales
in atomic, molecular and solid state physics. Therefore in what follows
we will use these units. Instead of the temperature, a dimensionless
ratio kT/Eo, designated by T, will be used, and instead of particle
concentration the product na\, designated by n. In this notation, the
usual formula for the critical temperature of an excitonic gas BEC,
neglecting any interaction corrections, is given by

TC = 6.62^(V 3 . 0)
M g
Here M = me + m^ is the effective mass of the exciton, equal to the
sum of electron me and hole nih effective masses; m — m e- nih/M is their
reduced mass, entering (1) and (2); and g is the exciton ground state
statistical weight (degeneracy).
(2) The effective masses me and nth in any particular semiconductor,
though different, are usually of the same order of magnitude (or differ
at most by one order of magnitude). The absence in the electron-hole
system of really heavy particles such as nuclei results in a dominant role
of quantum effects at all temperatures T < 1. In particular as formula
(3) shows, Tc ^ 1 for n ~ 1, unlike, for example, in 4He where Tc ~ 10~5
in corresponding units. Other important manifestations of the absence of
heavy particles are very large zero vibrations of excitons in the excitonic
molecule. The amplitude of this vibration is of the order of, or even
larger than, the excitonic radius CIQ itself. So biexcitons appear to be
 Macroscopic Coherent States of Excitons in Semiconductors 249
very loosely bound complexes with binding energy ED not exeeding 0.1,
compared to 0.35 in the hydrogen molecule [16, 17].
For the same reason, nothing like an "excitonic crystal" can exist. The
weak van der Waals attraction dominating at large intermolecular dis-
tances is not able to confine light particles such as excitons or biexcitons.
According to Ref. [18], the s-wave scattering length of two excitonic
molecules is large (~ 7) and positive (repulsive). Therefore a condensed
phase of excitonic molecules - a molecular "liquid" similar to liquid
hydrogen - also cannot exist. But an electron-hole liquid (EHL) similar
to metallic hydrogen or alkali metals does exist [15, 6, 9, 10]. Unlike
common metals, in the EHL not only electrons but also holes ("nuclei")
are Fermi degenerate. If the effective masses of both electrons and holes
are more or less isotropic, the EHL at low temperatures transforms to
an "excitonic insulator" phase [19-21]. In this two-Fermi-liquid state the
collective pairing of electrons and holes in the vicinity of Fermi surfaces
arises, quite similar to BCS pairing in superconductors. This pairing
manifests itself in the appearance of the energy gaps A around the Fermi
surfaces, as shown in Fig. 2. These gaps may be considered a rem-
nant of the binding energy of a single exciton, transformed by collective
many-body interactions in the condensed phase. The gap diminishes fast
with increasing particle density or me-nth difference, and it does not exist
at all if the anisotropy is large. In that sense, the excitonic insulator
state in the nonequilibrium electron-hole (e-h) system is a coherent BEC
state of high density (n > 1) excitons, as the superconducting state is a
coherent BEC state of Cooper pairs. Like Cooper pairs, e-h pairs in an
excitonic insulator have very large radii ~ A"1/2, much larger than in-
terparticle distance. So they are true collective phenomenon. But, unlike
Cooper pairs in superconductors which are charged (—2e), excitons are
electrically neutral.
The above-described phenomena can be schematically represented by
the phase diagrams depicted in Figs. 3-4. They show possible states of
the e-h system in the plane (n, T). Here ft is the average particle density.
The phase diagram type of any particular semiconductor is defined
by the relationship of the binding energy per e-h pair in the biexciton
gas, given by (Eo + ED/2)9 and in the EHL, indicated by —/x/. Strictly
speaking, \i\ is the difference between the chemical potential per e-h pair
in the EHL and £ g , where Eg is the minimum energy of the free e-h
pair. Different phases presented in Figs. 3 and 4 are: G, the gas (plasma)
of weakly interacting electrons, holes, excitons and excitonic molecules;
BG, the degenerate gas of excitons and(or) excitonic molecules; ML, the
250 L. V. Keldysh

Fig. 2. Energy spectrum in the nonequilibrium "excitonic insulator" phase. The

striped regions are energy ranges occupied by electrons.

metallic liquid (EHL); and IL, the insulating liquid (excitonic insulator).
The striped area is the region of phase coexistance: droplets of liquid
phase surrounded by gas, degenerate or nondegenerate. Coherent states
BG or IL are present in the phase diagrams of the types shown in Figs.
3(b) and 4(a),(b). But up to now only the simplest type, shown in Fig.
3(a), is firmly established and studied experimentally, in germanium and
silicon.
Both phases in Fig. 3 correspond to the case | /x/ | > (EQ + ED/2);
that is, the EHL is more tightly bound than excitonic molecules. It
is theoretically proven and experimentally confirmed that the fulfilment
of this condition is strongly favored by multivalley band structure, i.e.
the presence in the electronic spectrum of a semiconductor of a few
equivalent groups of electrons or holes (or both). Just such a band
structure is characteristic of both germanium and silicon. No BEC exists
in the gas phase in this case, exactly as in the case of helium. It does exist
in the phase diagrams depicted in Fig. 4, corresponding to the fulfillment
 Macroscopic Coherent States of Excitons in Semiconductors 251

ML
ne(T)

log
(a)
Fig. 3. Possible types of nonequilibrium electron-hole system phase diagrams in
semiconductors with electron-hole liquid more tightly bound than the excitonic
molecule. The striped regions are the coexistence domains of the liquid and gas
phases.

o ne|
(a) (b)
Fig. 4. The same as in Fig. 3, but for semiconductors with excitonic molecules
more tightly bound than electron-hole liquid.

of the opposite condition,

0.5ED . (4)
Theoretically, such a case may be expected in semiconductors with
single-valley band structure for both electrons and holes, Fig. 4(a)
corresponding to semiconductors with a very large difference of the
electron and hole effective mass and moderate anisotropy, and Fig. 4(b)
to comparable effective masses and small anisotropy. Experimentally, the
fulfillment of (4) is not firmly established in any semiconductor. And it
should be noted also that the difference of binding energies per e-h pair
in the EHL and the biexciton in the case corresponding to Fig. 4(b) is
very small and comparable to the accuracy of theoretical computation.
So even the theoretical possibility of the phase diagram shown in Fig.
4(b) is, strictly speaking, not proven.
(3) Another important difference of the e-h system from any ordinary
252 L. V. Keldysh
matter is its essentially nonequilibrium nature. For sure, it also exists
under equilibrium conditions, but without any condensation phenom-
ena, because the number of e-h pairs itself is then determined by the
equilibrium conditions, i.e. by the balance of the thermal creation and
recombination processes. Therefore the chemical potential for pairs must
be equal to zero, while the above-described condensation phenomena can
take place only if it is equal to or larger than the minimal pair excitation
energy. It implies
min(£ g - (Eo + 0.5ED), Eg + /i,) < 0, (5)
which means that the binding energy per pair in the excitonic molecule
or in the EHL becomes larger than the minimal energy for a free e-h pair
creation, Eg. Such a possibility was also considered theoretically [19-22],
and it was recognized that it implies the reconstruction of the electronic
structure of the host semiconductor.
Indeed, the term " exciton" was introduced by Frenkel for the quasi-
particle which is essentially the quantum of excitation energy. The exciton
does not carry electric charge or real mass, although it possesses effec-
tive mass (exciton crystal momentum is conserved). It can also possess
angular momentum, dipole and magnetic moments, etc. The creation of
an exciton actually means the transition of an electron to some excited
level. The fulfillment of (5) in that sense signifies the existence of an-
other ground state electronic configuration than is supposed in Fig. 1,
with lower total energy. The equilibrium condensation of excitons with
zero chemical potential is, then, another formal language to describe
an electronic phase transition, i.e. reconstruction of electronic states in
the valence and conduction bands, accompanied by the formation of a
charge or spin density wave, magnetic ordering, or something like that.
Nothing like superfluidity can arise because the Hamiltonian does not
conserve the total number of excitons (e-h pairs). This inevitably results
in pinning of the "condensate" by interband scattering matrix elements
of the Coulomb interaction [23], which in this language appears as the
source of the spontaneous creation of two e-h pairs lifting the gauge
invariance of the Hamiltonian to yield a phase transformation, different
in different bands, of the kind
xpj -> xpj exp(f</>7). (6)
Here j = (c,v) is the band index. Also, interband scattering of electrons
by static lattice distortions - uniform or periodic - can serve as a single
pair creation source and thus a pinning mechanism.
 Macroscopic Coherent States of Excitons in Semiconductors 253
Quite different is the problem of the collective properties of the
nonequilibrium dense e-h system produced in a semiconductor by some
external action, usually illumination. Then the total number of e-h pairs
N (or as used in Figs. 3 and 4, their average density h = N/V, where V
is the volume of the specimen) really becomes an independent variable
with a value controlled by an external source. The conservation of N
is broken by recombination processes. But in some semiconductors the
direct radiative recombination is forbidden by a parity selection rule (as
in CU2O) or by quasimomentum conservation, as in germanium and sili-
con. The lifetime of the nonequilibrium e-h system in such a case may be
much longer than the thermalization time and the system appears to be
in quasiequilibrium, the only nonequilibrium parameter being the total
number of particles N itself. Under this condition, the phase diagrams of
Figs. 3 and 4 have an exact meaning. On the other hand, in semiconduc-
tors with a dipole-allowed excitonic radiative transition, the lifetimes are
of the order 10~9 — 10~10 s, and a dense enough e-h system is always far
from equilibrium, so that these phase diagrams can be considered only
as qualitative indications of trends in its evolution.
One more question of importance is whether excitons are really
bosons [24]. The reason for asking this is that the exciton is a com-
pound system composed of two fermions, and increasing the density of
excitons and their overlap also means increasing the local density of
fermions, perhaps leading to problems with the Pauli principle. Also,
from a formal point of view, the exciton annihilation operator A? is
composed of electron and hole annihilation operators aep and a^p (P, p
are momenta)
i p = ^<po(p)a e p +p a fc p_ p , (7)

where cpo is the wave function describing the relative e-h motion in the
exciton. The commutator of the two exciton operators, AF and A?>, is
then
"P' "P

<Po(<l - ^r)<P0(q - ^) S Iq-Q^q+Q

P . "I

(8)
where Q = (P' - P)/2. The righthand side of (8) differs from the usual
one for Bose operators <5pfp by operators that are of the order of n, the
exciton density. So for n ~ 1, the operator (7) has nothing to do with usual
254 L. V. Keldysh

Fig. 5. Momentum distribution function for electrons (or holes) in a system

of Bose-condensed excitons. Different curves correspond to different exciton
densities (n\ < ni < n3 < rc4).

Bose operators. What it really means and what happens to excitons at

high densities is illustrated by Fig. 5, in which the momentum distribution
function of electrons (or holes) for the system of Bose-condensed excitons
of different densities is presented. For a single exciton, it is | <po(p) I2> the
probability of finding an electron with momentum p in the exciton. For
low concentration it increases proportional to density, since all excitons
are identical: /(p) = n | <po(p) I2- But as /(p = 0) approaches 1 with
increasing n, it cannot continue to increase because of the Pauli principle
- /(p) is the fermion distribution function. So at larger densities, the
electrons and holes bound into excitons are forced to occupy states with
larger momenta until their /(p) also aproaches 1, and so the distribution
smoothly transforms to the Fermi distribution function. The binding
energy of the exciton transforms to the narrow gap around the Fermi
level which diminishes fast as the kinetic Fermi energy becomes large
compared to the Coulomb interaction. Thus Bose-condensed excitons
continuously transform to the "excitonic insulator" described above, and
finally to a degenerate EHL.
Such transformations are not unique for excitons. They must hold
also for 4 He or any other atomic or molecular system. But it is very
difficult experimentally to compress, for example, helium to densities
corresponding to n ~ 1. The peculiarities of excitons are their large radii
 Macroscopic Coherent States of Excitons in Semiconductors 255
and the possibility of creating them by external excitation, so that high
compression can be easily realized.

2 Polaritons
A very peculiar, but undoubtedly also the most important, case is pre-
sented by BEC of dipole-active excitons. Not only is it always essentially
a nonequilibrium problem, as explained above, but the nature of the
quasiparticles themselves is very special. They are the well-known polari-
tons - the quantum superposition of photons and electronic excitation
(excitons) [25]. In other words, a polariton is a real photon in a medium:
an exciton representing a polarization cloud accompanying propagation
of electromagnetic field through the medium [26]. These polarization ef-
fects are especially strong under resonant conditions, that is, for photon
energies hco close to the energy of exciton creation ha>o = (Eg — Eo).
Schematically depicted in the Fig. 6 is the well-known polariton disper-
sion law. Strong mixing arises close to the intersection point of the
undisturbed photon and exciton branches (ko,&>o), and the typical pic-
ture of level repulsion appears. So the problem of BEC of dipole-active
excitons is essentially that of photons, or, in other words, the problem of
a coherent electromagnetic wave of finite amplitude in the polariton fre-
quency range. It is a very familiar problem but is still far from trivial. It
includes all the many-body phenomena typical of BEC, but they acquire
the meaning of nonlinear optics phenomena. Two different problems are
usually considered in this context.
The accumulation of a macroscopic number of initially incoherent ex-
citation quanta in a single-photon mode is lasing. It was recognized long
ago [27-31] that the appearance of a coherent mode from multimode
intense noise in a pumped system is a very typical example of a nonequi-
librium phase transition. The approaches used for its general theoretical
treatment are based on mean-field approximations, the Langevin equa-
tion, the Fokker-Planck equation, etc. Their detailed review is beyond
the scope of this article. As applied to the particular case of dipole-active
excitons in semiconductors, the qualitative picture of the lasing process
should look like the following [32]. Initially excited electrons and holes
combine into excitons, which dissipate their kinetic energy by emission
of phonons - quanta of the crystal lattice vibrations - and so move
downwards in energy along the lower polariton branch (LPB) in Fig. 6.
Approaching the exciton-photon intersection point, which corresponds
to very small momenta ~ Eg/c (c the light velocity), they smoothly
256 L. V. Keldysh

1 1

/ /
UPB
.. — //^ LPB
////
//
//
////

i i

Fig. 6. Polariton dispersion law.

transform into polaritons, with the photonic component increasing at

the expense of excitonic. But this transformation is accompanied by a
decrease of energy dissipation rate, since just the excitonic component in-
teracts with lattice vibrations, and also because of the dramatic decrease
in the density of states due to increase of the polariton velocity below the
intersection point. Thus a "bottleneck" arises in this momentum range
where polaritons accumulate. Because of the small density of states in
the bottleneck region, polariton occupation numbers there easily reach
values of the order of unity even if the total exciton density is small,
n <C 1. Then the mechanism, common in laser physics, of line narrowing
by stimulated emission starts. As applied to the polariton kinetics, it
looks like stimulation of transitions to the bottleneck region. Under the
condition of sufficiently strong stationary pumping, this process results in
the imaginary part of the optical susceptibility becoming negative for one
or a few polariton modes. This denotes an instability of the stationary
incoherent radiation regime and the appearance of a finite-amplitude
 Macroscopic Coherent States of Excitons in Semiconductors 257
coherent polariton mode, i.e. BEC of polaritons. In the transient regime,
as a peak in polariton frequency distribution becomes large enough and
narrow, the exciton interaction synchronizes all of the modes inside this
peak. So in a finite time interval, the coherent state of a macroscopically
large number of polaritons arises. It should be noted that existing semi-
conductor lasers usually operate either at excitation densities n > 1 or
with participation of optical phonons, impurities or lattice defects, and
do not correspond to this idealized picture.
Another direction of research is the nonlinear optics proper, i.e. prop-
agation of the intense electromagnetic wave or pulse and its interaction
with the medium under conditions of polariton formation [31, 33]. In
such a case the coherence is induced by some external source of radiation.
In what follows, mainly this type of problem will be discussed. The main
difference in the formal description of these phenomena corresponding
to the above physical picture, compared to common "excitonic insulator"
theory, is the presence in the Hamiltonian of the source of e-h pairs -
the electromagnetic field /. The additional term is

Heh = J[xpe(r)r(T,t)xph(r) + xpl(r)f(r,t)ipl(r)]d\ (9)

where
p=- (10)
EE
is the matrix element of the electronic dipole transition between the
conduction and valence bands induced by the electric field S\ po is the
momentum of the photons; and p w is the interband momentum matrix
element. In its physical interpretation, / is the interband Rabi frequency
in terms of Eo/h.
It was indicated in Ref. [34] that for photon energies above the ab-
sorption threshold hco > Eg, a coherent field produces a collective excited
state of semiconductors quite similar to the excitonic insulator state, even
neglecting the Coulomb interaction. Creating electrons and holes coher-
ently in correlated pairs, the field itself acts as the pairing force. Also,
an energy gap arises in the electronic spectrum at the Fermi level with a
magnitude | / | (Fig. 2). The obvious condition for the existence of such
a state is / T > 1, where T is electron (hole) relaxation time. The complete
treatment of this problem needs inclusion of many-body effects. This was
done in Refs. [35-36] (see also the reviews and monographs [31, 33], [38-
40] and especially [37], where the problem has received its most complete
formulation, in terms of nonequilibrium Green functions [41]. (The de-
258 L. V. Keldysh
tails of this technique are presented in Refs. [42-51], [31] and many
other reviews and monographs.) Here we summarize only a few general
definitions and basic relations and also some features specific to the
nonequilibrium e-h system.
The nonequilibrium Green's function is a 2 x 2 matrix, which in the
so-called "triangular representation" looks as follows:

0 G \
) <»>

Q(a,r) a r e a(jvanced and retarded functions, respectively, defined in the

usual way

G(r)(x,x') = -i®(t ~ 0 < [ip(xW(x')]± >, (12)

G{a)(x,xf) = [G(r\x',x)]\ (13)

F(x,xf) = -i < [xp(xlxpHxf)]T > . (14)

Here x = (r, t)\ indices +(—) in (12) and —(+) in (14) refer to the case of
Fermi (Bose) statistics. In the electron-hole system all of these functions
are 2 x 2 matrices in band indices (c9v). In this notation, the interband
(electron-hole) Green's function

Gvc{x,x') = Ghe(x,xf) = Vr) y (15)

describes the exciton (electron-hole pair) condensate just as a Gor'kov

function in the theory of superconductivity describes a condensate of
Cooper pairs:

= - m - t') < [y>Ux),y)Ux')]+ >, (16)

= -f<[^(x),^V)]_>. (17)

Diagonal in both types of matrix indices, the functions Fee and FM at

coincident times t = tr are closely related to the single-particle den-
sity matrices and in the momentum representation to the distribution
functions (occupation numbers)

Fee(p;t,t) = -i[l -2nev(t)]. (18)

 Macroscopic Coherent States of Excitons in Semiconductors 259

The interband function F^(p;t, t) can be considered as an effective exci-

ton wave function. This interpretation follows from the comparison of
definitions (17) and (16) and is exact, at least for n <C 1. In a dense
many-body system it should not be treated too literally.
In general, Green's functions can be found from the Dyson equation

ijt - ^ - V 2 ) Gij(x,x') - f i*(x,x")Gkj{x\x')dAx"

= 5ijd(x-x'l (19)
where "self-energy" matrices %j also have "triangular" form
(2o)
*•• - (% f )
and can be expressed in terms of Green's functions and the interaction
potential in any approximation by a set of Feynman graphs. The specific
feature of the system under consideration is that, apart from the infinite
set of diagrams accounting for different order interaction processes, the
interband self-energy Ee/, contains, according to (9), the driving force
contribution
f<o)/Y Y/x = ( 0 f{x
ij( }
* \f(x) 0
The electric field $ entering into Eq. (10) has to be found self-consistently
from Maxwell's equation,

(22)

where the resonant excitonic contribution to the polarization [4] is

eh
0>ex = e< ye(x)riph(x) > = -^—=r [pCT ' Fhe(x,x)]* (23)

and e in (22) accounts for all the nonresonant contributions.

Equations (19)—(23) are the complete set describing formally all nonlin-
ear polarization problems. The interband matrix function Geh represents
the macroscopic coherent state of electron-hole pairs (excitons). In gen-
eral, the theory under consideration is that of a driven two-component
Fermi liquid with all its many-body aspects, with the additional compli-
cations due to being far from equilibrium. Therefore, the real problem in
any particular case is the choice of an appropriate approximation.
If the retardation effects in the self-energies may be neglected, i.e.
A

Sjy ~ d(t — tf), which is the case at least in the mean-field approximation,
260 L. V. Keldysh
(19) can be reduced to the equation for the single-time (t = tr) density
matrix [31, 33, 36, 37, 51]

/ l7 (r,r';t) = Fij(t9i';t = i!). (24)

Symbolically written, this equation is similar to the well-known Bloch

equation for two-level systems,

= (k + i ) f f ( h o + ± ) . (25)
Here the two-band Hamiltonian matrix is introduced by

M- ^vA • dtj. (26)

Matrices (I^) l7 and 0% are defined as (i^), 7 = 1$ , 0% = ZJ;a);

Eot and m, are the corresponding band-edge energy and the effective mass
(negative in the valence band).
In the simplest case of stationary and spatially homogeneous pumping,
the solution of (25) is

fuu - f(°) ^ ^ 1 ^h i
Uhh
~ feM £(P) + z«(p
£ S(r)
( (P)
P) + (P) ,f(Q)
(Q) , ,(0k
(0k ,28v
(28)

where

£(p) = Eg - Hco0 + p . (29)

Here, po and a>o are the momentum and frequency of the pumping wave,
respectively, and
{a)
f(0) = aeeM
CW)
> Je,h ~ 2y ' ^ ^
In the mean-field approximation, the self-energies themselves can be
expressed in terms of fiJ9

4W ^ f'5(r^) (31)
 Macroscopic Coherent States of Excitons in Semiconductors 261
and so the set of equations (25)-(30) is complete. Under stationary
conditions they reduce to two equations for the self-energies,

u (32)

_
(P)
-
The most important equation here is (32). It defines the gap in the
renormalized excitation spectrum,

+ T2 (—
\m e ~
For hco0 > \Eg+ Tir\0) | it is just equal to | I,eh |. So (32) is similar
to a BCS equation, but inhomogeneous and with a slightly different
integral kernel. The last difference is due to the nonequilibrium nature
of our problem. At moderate field | / | < 1, and not too close to
the resonance | Z^ | <C Eg — ha>o, Eq. (32) can be transformed to an
inhomogeneous Schrodinger equation for the exciton wave function. At
very large intensities | / | > 1, it gives the gap | Z^ |« /, as was obtained
in Ref. [34]. The functions / $ entering (32)-(33), defined by Eq. (30),
are connected to spontaneous fluctuations of occupation numbers. In
thermodynamic equilibrium, the ratios on the right-hand side of (31) are
found to be exactly

where nv(T) is the equilibrium Fermi distribution function, /i is the

chemical potential and ee>/,(p) are the energies of corresponding single-
particle excitations. Correct calculation of both the diagonal self-energy
matrix element Q(p) and the damping y(p) demands consideration of
relaxation processes at least to lowest order. But this is out of the scope
of our mean-field approximation. For our present considerations, it does
not seem to be crucial, since in an intrinsic semiconductor, nv(T) is very
small and it seems reasonable to suppose that the difference of | ffl \
from 1 may also be neglected in a pumped system, at least for fast
processes.
Nonlinearites in (31) account for the exchange interaction of electrons
262 L. V. Keldysh
and holes and so-called phase-space filling, which means reduction of
the interaction due to the blocking of some virtual scattering processes
because of occupation of corresponding final states.
The only shortcoming of this approach is the difficulty of including
the correlation effects, the most important of them being the formation
of excitonic molecules. Because of the polaritonic nature of excitons, the
existence of biexcitons implies the resonant mutual scattering of photons.
In the following sections, we will show that at | / | ~ 1 it can completely
dominate the optical properties in the vicinity of the exciton resonance.

3 Excitonic Molecules in a Coherent Cloud of Virtual Excitons

The existence of excitonic molecules - biexcitons (EM) - in semicon-
ductors is well-established theoretically [13, 14, 16, 17] and experimen-
tally [52-62]. It is also well recognized [62-67] that, as the excitons
themselves are the essential characteristic feature of semiconductor op-
tics, the biexciton contribution may be very important in the nonlinear
optical phenomena in the frequency range close to the intrinsic absorp-
tion threshold. As a rule, two different approaches have been used in the
theoretical description of EMs and their contribution to physical phe-
nomena: the direct microscopic treatment [68, 50], considering the EM
as a four-particle bound state - two electrons and two holes - and a phe-
nomenological one [63-66], introducing the EM as an independent boson
excitation capable of decaying into an exciton and a photon. Both of
these approaches, in many cases very successful, have essential disadvan-
tages for application to the problem under consideration. It seems that
the first approach is too complicated to be used as a starting point for a
true many-body problem. The second one is inadequate to account for
the changes of EM properties and parameters, similar to those described
above for excitons, due to interactions in a dense exciton-biexciton sys-
tem. Nevertheless, according to modern ideas [24, 37] about many-body
exciton physics, just these transformations, including an exciton's and
other complexes' continuous destruction, constitute the essence of the
high-excitation problem. Moreover, the assignment of definite statistics
(e.g. Bose-Einstein) to compound particles at such high densities becomes
inadequate, as explained in Section 2. At moderate exciton densities both
of these effects contribute corrections of the order n, where n is the exci-
ton dimensionless density. For the EM it will evidently be proportional
to na\x. Here, a\,x is the effective EM radius (in terms of the exciton
radius). As EM binding energies E\,x are typically an order of magnitude
 Macroscopic Coherent States of Excitons in Semiconductors 263
smaller than the corresponding exciton Rydberg EQ = 1, usually a^x must
be a few times larger than unity. So there is a relatively wide interval
of densities in which excitons are virtually unchanged, whereas the EM
may transform drastically. Nonlinear optical phenomena due to excitons
coherently driven by an intense electromagnetic pump, with the densities
in the interval described above, are the main subjects of this section.
For this purpose the following model seems to be adequate: excitons
will be considered as true boson particles, described in terms of the Bose
field operator xp(r, t) and the dispersion law ef^ = Eg — 1 +p2/2m. So
defined, \p(x) is essentially Feh(x,x) of Section 2. In contrast with that
section, an EM will be treated as composite particles - the bound state
of two excitons arising due to the interaction potential F(r,r'). Obviously
such an interaction has to be attractive. But then the well-known problem
in quantum liquid theory arises: at low temperatures, a system of Bose
particles with an attractive interaction is unstable against spontaneous
contraction. In order to avoid this difficulty, two equivalent types of
excitons xp^oc = 1,2) with the same parameters may be introduced. The
only distinction is that the excitons of different types attract each other
(U\2 = C/21 = — V < 0), whereas particles of the same type have a
repulsive interaction (U\\ = U22 = U > V > 0). Such an appoach
leads simultaneously to the existence of bound complexes of two excitons
of different types and to the stability of the many-particle system as a
whole, i.e. the stability of the "molecular gas" of excitons. Moreover, the
proposed model is not too artificial, since two types of exciton may be
considered as corresponding to singlet excitons with mutually opposite
directions of electron (and also hole) spins. The main difference of this
picture from reality is the absence of triplet excitons. Nevertheless, it
seems that such an assumption is not crucial because a triplet exciton in
the dipole approximation does not interact with photons (though it must
in a real system, due to electron exchange.) A similar model has been
used in Ref. [50]. Thus the exciton-exciton interaction is introduced as a
2 x 2 matrix:
U =UI - VTU (36)

where

0 1
T = l
1 0
 264 L. V. Keldysh

G g
a,x 0,x' a,x fiX

w
(p, c) (P, 0 (P, « ) (Po> o)

Fig. 7. Schematic illustration of elements in the Feynman graphs: Green's func-

tions and coherent exciton field amplitude.

Qualitatively this corresponds to the exchange interaction of two

hydrogen-like atoms (excitons).
From the theoretical point of view, the system under consideration
is a "driven Bose liquid". It arises from a macroscopically occupied
state (mode) tpa(r, t) of the excitons coherently created by an external
electromagnetic field <f (r, t). As the result of mutual scattering of excitons
of this mode, the other modes become populated and also the EM may
arise. A consistent description of this system may be given using the
nonequilibrium Green's functions technique [41], [46]-[49] and [31]. The
following functions are involved in such a treatment:
(1) Normal Green's functions ga^(x,xr), which describe the propagation of
a particle from space-time point x = (r, t) to x' including the possible
process of the change of the particle type (a —• /?).
(2) Anomalous Green's functions [46]-[49] ga^ (x,xf) which characterize
the macroscopically coherent states and describe the correlated ap-
pearance (or disappearance - %ap(x9x')) of two particles a and /? at
points x and x'9 respectively. Being the manifestation of the presence
of a macroscopically large number of coherently correlated pairs, these
functions contain complete information about the EM. In essence, gu
at equal times t = tr is the effective EM wave function. Throughout the
following discussion we will denote matrices (operators) by the 'hat'
(5) symbol and anomalous Green's functions and self energies by a
'tilde' (3).
Graphical representation of all these functions (in real space the same
as in momentum representation) is shown in the Fig. 7. Here, po and
coo are the momentum and frequency (energy) of the pump wave quanta,
respectively, p = 2p0 — p, e = 2co0 — €. It is easy to understand the origin
of the correlation of pairs of particles with momenta p and p. Such
a pair correlation arises as the result of the mutual scattering of two
initial excitons with the momenta po. Similar correlations are well-known
 Macroscopic Coherent States of Excitons in Semiconductors 265
in the phenomenological description of the nonlinear optical processes
such as four-wave mixing, self-phase modulation, etc. In what follows,
in order to consider the resonant phenomena in the vicinity of the
exciton and biexciton (two-photon) resonances, the "rotating coordinate
frame" method will be used. In this case, the frequency coo of the
pump wave becomes the origin of the frequency (energy) axis, so that
co —• co — co°,6p = Eg — 1 + p2/M — co° and co = —co. As the photon wave
vector po is very small in comparison to typical momenta of scattered
excitons, one can neglect it in the process of biexciton creation and put
p ~ —p. For example, e^ ~ e ^ = ef\ Only for an analysis of the
optical manifestations of the renormalization phenomena in the exciton-
EM system is it necessary to treat the difference between p and —p
explicitly. But in what follows we will assume the special arrangement
permitting realization of BEC of excitonic molecules with exactly zero
momentum: pumping by two counterpropagating, circularly polarized
waves [69, 70], which are in our model independent sources of excitons
of different species a. Then obviously p = —p.
As explained above, according to the nonequilibrium Green's function
formalism, each matrix element of g^ or §a£ matrices is itself a 2 x 2
matrix,

(38)

Here again, g^ a) denote retarded and advanced Green's functions, and

the diagonal element /^j is closely related to the distribution function.
But unlike the treatment of excitons in the preceding section, here we are
dealing with bosonic Green's functions. Therefore for t = t\

Q(V;t,t) = -i[l + 2np(t)}. (39)

The + sign in the above formula corresponds to Bose particles, and nv

are the occupation numbers of the excitons.
The Green's functions have to be found from the Dyson equations,
which are in this case similar to the corresponding equations for the Bose
liquid [46,51]:

J MP;*>0 -
J(p; t, fi)|y/Kp; tu tf)dh = S^S(t - 0 , (40)
266 L. V. Keldysh

at
\ J J

Here, the normal and anomalous self-energies 6^ and d^ include all

interaction processes in the system and are represented as usual by an
infinite series of Feynman graphs. The simplest class, which corresponds
to the self-consistent field approximation (SCF), is shown in Fig. 8. The
dashed lines in these graphs denote the interaction matrix U. Nonequi-
librium self-energies are also 2 x 2 matrices of the type

°g ) • (42)
Here, Q ^ is the noise correlator in the system, closely related to the
probability of incoherent creation and annihilation of excitons. The
macroscopic wave function tpa(Po>O of coherent excitons satisfies the
following field equation:
l d ,0A / x f irh ,. , /w/
T+ ~ 6PO V«(Po, t) - / (TjJ(po; t, t ) ^ ( p o ; t )dt
at j j

- / ff J(Po; *> * )\pp(po ;t)dt = fa(t), (43)

or, in a general spatially inhomogeneous case,

ih— + ^— V 2 — e0 ) xpoc(r,t)— I <r$(r,r f ;t 9 t f )\pp(r f ;t f )dr'dt'

t')drrdtf = / a ( r , t), (44)

where 6o = ^J»=o ^ s ^ e detuning of the pump frequency from the exciton

resonance, and the slowly varying electromagnetic field amplitude fl x(r, t)
is the source of the dipole-active coherent excitons. This function is
determined by the dipole matrix element m a of the a-exciton-photon
interaction:
ip
f?{t) = maim = —-{ Pcv)ai<t>(d)Si(t). (45)

Here, (pcv)<xi is the matrix element of the interband transition, mo is the

electron mass and </>(0) is the value of the exciton ground state wave
function for the coincident electron and hole coordinates.
 Macroscopic Coherent States of Excitons in Semiconductors 267

777/

Fig. 8. Diagonal (E) and anormalous (£) self-energy graphs in the mean-field
approximation.

The equations (40)-(41) are the matrix equations in the sense of the
definitions (38) and (42). In fact, the equations for g (r) and g (r) look
exactly like (40) and (41) with the substitution of the proper self-energy
functions. The same statement holds for the advanced Green's functions.
As to the function f^(p;t,tf), it can be expressed in terms of the self-
energies Q ^ , and the retarded and advanced functions [41]:

(46)

In what follows, the assumption that the self-energies Q ^ do not con-

siderably deviate from their quasiequihbrium values will again be used.
Also, all the retardation effects in the self-energies will be neglected.
Some results of such a treatment are presented in Section 4.
If the duration of the pumping pulse exceeds the relaxation time of
excitons, the quasi-stationary state arises. In this case, all self-energies
become time-independent functions and the following formulas can be
used:

(47)
268 L. V. Keldysh

(48)

0— = (Til + (T12 , <X~ = #11 ± #12 • (49)

Here, the symbols + and — correspond to the symmetric xp+ =
ip2) and antisymmetric tp_ = l/y/2(\pi — xp2) superpositions of different
types of excitons,
5±
gHp) = (50)
eL — I

4= |2 • (51)

The values ejj1 represent the energy difference between the excitonic levels
and the carrier frequency of the pump. In a sense, the last formula (52)
is an odd result. Instead of the usual resonance relation of the exciton
polarization \p to thefield/, the denominator in Eq. (52) is quite different
from 6jj=. This result is one of the manifestations of the presence of pair
coherence in the system under consideration.
The polarizability %(k, co) which describes the propagation of a weak
probe signal (k,co) in a semiconductor in the presence of a pump is given
by the following expression:
4TT I m + |2| g(k) |2
g(k) 2 (53)
co
CO — c 2 fe 2 — -
This formula directly manifests the interconnection of the polariton
branches k and k. The proper Green's function D(k) of the probe
electromagnetic field is given by
(co - e£)A+ + vp | m+ |2 6+
D(k) = (54)
A~A+ - 2vk | m +
with
A =(co — — e£)— \ m + (55)

A+ = (co - m+ \ (56)
and

Vn = - E — T 1 (57)
 Macroscopic Coherent States of Excitons in Semiconductors 269
The poles of D(k) describe the four new polariton-like branches. Each of
these branches reflects the joint propagation of the four particles - two
excitons and two photons with wave vectors k and k. Their dispersion
laws are

4 | m P [(e+ + 4 0 ) ) 2 + 2vpe+cop°>]\ , (58)

Here cffi = c|p| and np = — 0.5 [1 + i/*x(p; U t)] are the exciton occupation
numbers. It is easy to see from (54), and especially (58), that the coupling
of two polaritons p and p is determined by vp, which in its turn is
governed essentially by the | #+ |2? that is, by the EM density.

4 Results
In the framework of the model introduced, the considerations in the
previous section are quite general. In order to obtain a more detailed
description accounting for explicit dependences of self-energies on pump-
ing frequency and intensity, the mean-field approximation will be used
below. In this case a matrix equation similar to (25) can be used for the
density matrix /^(r, r'; t, i):

ihj-tfex = (h0 + £ (r) ) fex - fex (ho + £ (a) ) (59)

and retarded self-energies can be calculated in terms of the density matrix
l
<ja/Kr,r'; t) = -U^xr) [f/^(r,r'; t) + 2xpa(r

-dard(r-rf)] (60)
and
h = I - (Eg - Eo - hcoo - ^vA . (61)

It should be noted that, in its physical content, the mean-field approxima-

tion used here is essentially different from that used in Section 2. Having
introduced the exciton as a basic "elementary particle", we neglect the
possibility of treating its internal structure but include in the mean-field
treatment the exciton-exciton iteraction and even biexciton formation.
As explained above, the accuracy of this approach is of the order of n or
270 L. V. Keldysh
a^ <C 1. Within the same accuracy, we can neglect the effective radius of
the exciton-exciton iteraction potential, which is obviously of the order
of 1. Then considering Fourier components of matrix elements of U as
two different constants, we can adjust their values using the two most
important experimental data - the dissociation energy of the excitonic
molecule Ed and the low-energy mutual scattering length of two biexci-
tons. To be specific in what follows, the results for a two-dimensional
problem will be presented, i.e. excitons in a quantum well (QW) structure
and a pump wave propagating normal to the QW plane direction will
be treated. Many current experiments are done in such a geometry. Its
obvious advantages are that the exciting field amplitude is independent of
coordinates and undistorted by propagation phenomena, the excitons are
coherently created with zero momentum, etc. The well-known problem
of the destruction of the long-range order in two-dimensional systems by
fluctuations seems in our case to be unimportant because the coherence
is induced by an external source and so no phase degeneracy exists, and
also because the finite lifetime and excitation pulse duration essentially
restrict the logarithmic divergence of long-wave fluctuations.
Being no longer interested in the exciton internal structure, we will
now introduce new ("biexcitonic") scales of variables more convenient
for the following discussion, by putting the biexciton dissociation energy
Ed and its effective radius

abx = -j^= (62)

jME
equal to unity. As a result, time intervals in what follows will be measured
in terms of h/Ed and two-dimensional exciton densities in units of a^.
Dimensionless coupling constants will be introduced as
mil mV od
47m2 4nn2 a+ /
In order to avoid explicit treatment of relaxation processes only two
limiting regimes will be considered here: (1) stationary and (2) transient,
for time intervals short compared to any relaxation process.

4.1 Stationary Regime

This can be achieved if the excitation-pulse duration is large compared to
either the exciton phase-relaxation time or the coherence interval for the
pump field itself. Then formulas (47)-(51) within the above-described ap-
proximations give the possibility of expressing all the interesting variables
 Macroscopic Coherent States of Excitons in Semiconductors 271
in terms of two of them - the anomalous self-energies a±:
+ A 17 1 A »++W- 1. + , / 1±1 A 1 _
G± = 4TC( a - -A 1 - ^ -XrF + f a — AJ n 0 , (64)
where

n± = -?- • or± • Arsh ( ^ - ), (65)

•(£)•
\ c min/
and
r±) 2 - I 9± (66)
2
Here, no =| \p+ \ is the density of coherent excitons in the single-particle
condensate (polarization) and eo = e^l0 is the detuning of the pumping
wave from the unperturbed excitonic level. Two more equations define
the anomalous self-energies themselves:

- °~ 'ln^min = ~ 2 "0 ; (67)

o+ (1 - y l n e + J + o- (1 - ylne- i n ) = 2yn0. (68)

Together with formula (52), equations (64)-(68) are a complete set,
describing in the mean-field approximation the stationary states in the
model under consideration. They can be further simplified if X <C a. Then
it follows immediately from (68) that G~ « — 5 + and then from (64)-(66)
it follows that o~ « G + and e~in « 6^ in. The equations now reduce to
cr -hKrmin = —YIQ\ (69)

a = 4n(n + n0) = a faArshf-^) + n0] . (70)

Some results of the numerical solution of these equations are depicted
in Figs. 9-12. The first three of them show "resonant curves", i.e.
dependences of response - xp+ and a - on the detuning of pumping
frequency from the exciton resonance at a fixed value of a pumping field.
The zero of the frequency axis corresponds in these plots to the position
of the exciton level, and —1 to the biexciton resonance.
In these diagrams, the anomalous self-energy a is represented by
full lines and polarization \p+, designated as Co, by broken lines. Strictly
speaking, o cannot be considered as the "biexciton amplitude," in analogy
with the exciton amplitude \p+. But at low densities this interpretation
is exact, as can readily be seen from (65). It can be seen from these
plots that in the frequency range — 1 < coo < 1 and for fields > 0.2
the response is dominated by excitonic molecules. No resonant structure
272 L. V. Keldysh

2
/

o
o

\ "2-

Fig. 9. Coherent exciton amplitude Co (broken line) and anomalous self-energy Z

(full lines) dependences on the pumping field frequency co0 in a stationary regime
for pumping field strength / = 0.2 and effective repulsion coupling constant
a = 0.6.

Fig. 10. The same as in Fig. 9, but for / = 0.25 and a = 0.6.

exists around coo = 0. Only a loop in the interval 0.5 < a>o < 2 in
Fig. 9, also disappearing at higher field, is a small remnant of a "pure"
excitonic resonance observed at low fields. Instead of that, a typical
resonant structure for excitonic molecules is observed, accompanied by a
strongly nonlinear resonance of the excitonic polarization in the vicinity
of COQ = — 1. For frequencies above this stationary resonance the response
 Macroscopic Coherent States of Excitons in Semiconductors 273

Fig. 11. The same as in Fig. 9, but for / = 0.25 and a = 0.3.

becomes multistable. Two new states arise. The one with the smaller
value of excitonic amplitude corresponds to the one with larger biexciton
amplitude, and vice versa. Strictly speaking, solutions with coo > 0, c$ < 0
are unstable because of stimulated scattering into some states in the lower
polariton branch. Therefore, they will not be discussed in what follows.
Among the three stable (c0 > 0) solutions, one is "exciton dominated", i.e.
beyond the immediate vicinity of the resonance, the majority of excitons
are in the single-particle condensate-coherent polarization cloud. Two
other solutions are "biexciton dominated" - the majority of excitons
are bound into molecules at the expense of coherent polarization. But
nonlinear processes become greatly enhanced, such as four-wave mixing
(FWM) which increases with a, as can readily be seen from (54)-(57)
and the following formula, relating the exciton distribution function to
the anomalous self-energy
1
(71)

As or increases, this distribution transforms to a narrow peak with the

maximum value increasing as a2 and the halfwidth decreasing inversely
proportional to | o I1/2. This means an increase of the effective biexciton
radius proportional to | o |1//2.
As the pumping field strength increases, the resonance broadens, differ-
ent branches shifting in opposite directions from the resonant frequency.
So for any frequency co0 > — 1, at some critical field value dependent
274 L. V. Keldysh

?U-J

Fig. 12. Field dependence of the coherent exciton amplitude Co (full lines; note
the difference with Figs. 9-11!) and anomalous self-energy 2 (broken lines) for
pumping frequency COQ = —0.75.

on frequency, two solutions (both with o positive) - the "exciton domi-

nated" one and one "biexciton dominated"- converge and at still larger
fields disappear. Only the other "biexciton dominated" state- that with
a negative - persists. Such behavior is clearly seen in Fig. 12 where
field dependences of response y>+ (full lines) and a (broken lines) for
a>o = —0.75 and a = 0.75 are depicted. The common low-field "exciton
dominated" solution (the line coming from zero in Fig. 12) corresponding
to relatively small biexciton density (p ~ f2) disappears at the field value
/ ~ 1. The system must jump to the only remaining stationary state,
which will manifest itself in an abrupt increase of polarizability and large
increase of the FWM signal.
Both Figs. 9 and 10 correspond to a relatively large value of the
effective-repulsion coupling constant a = 0.6. At a < 0.5 the characteristic
appearance of the biexciton resonance is essentially different [51], as if
the effective nonlinearity changes its sign. It is shown in Fig. 11 for
a = 0.3. Now the multistable regime arises for frequencies below the
unperturbed biexciton resonance, starting at

coshz (72)
1-a
with a finite value of biexciton density | <r( | « sinhz. Here z is the root
 Macroscopic Coherent States of Excitons in Semiconductors 275

40 60 80 100 120 140

Time(ti/Ed)

Fig. 13. Time dependences of the coherent exciton amplitude c0 and normal
(Z) and anomalous (£) self-energies after smooth switching on of a pump field
(at t = 0) for the pumping field strength / = 0.05 and pumping frequency
COQ = —1.05. Time is in units ofh/Ed.

of the equation
tanh z = (73)
1 - a z.
In such a case, two more branches of stationary states exist, not shown
in Fig. 11, with very large biexciton densities | d \>\ <70 |. It seems that
if a is not too close to 0.5, these densities appear to be too large for
excitonic molecules to really survive.

4.2 Transient Regime

This regime is shown in Figs. 13-16. No relaxation processes are taken
into account. The pump field onset is supposed to have the form
f(t) = [1 — exp (—t/T)] with T = 4. This results in a partial smearing
of the Rabi oscillations for excitons. In these diagrams, one can see
even more dramatically than in the static regime that, in the vicinity
of the biexciton resonance, excitonic molecules completely dominate the
system response. Especially spectacular is the time dependence of the
normal self-energy, which, according to (70), contains the contributions
of both bound and free excitons proportional to their densities. The
striking similarity of the time evolution of both self-energies confirms
276 L. V. Keldysh

20 40 60 80 100 120 140

Time (ti/E d )

Fig. 14. The same as in Fig. 13, but / = 0.05, co0 = -1.025.

0 20 40 60 80 100 120 140

Time(h/Ed)

Fig. 15. The same as in Fig. 13, but / = 0.1, co0 = -1.05.

the overwhelming contribution of biexcitons. It is evident for both

frequencies - 0.05 detuning from the biexciton resonance in Figs. 13 and
15 and 0.025 in Figs. 14 and 16 - and for both (relatively small) pumping
fields - 0.05 in Figs. 13-14 and 0.1 in Figs. 15-16.
The Rabi oscillations for excitonic molecules are very slow and ex-
tremely nonlinear. At higher exitation and smaller detuning (Fig. 16)
 Macroscopic Coherent States of Excitons in Semiconductors 277

0 20 40 60 . 80 100 120 U0
Time
Fig. 16. The same as in Fig. 13, but / = 0.1, COQ = -1.025.

they even seem to become chaotic, which does not seem unexpected for
such a highly nonlinear system.
While small in reduced units, the field values supposed in Figs. 9-16
are really very large, corresponding to pumping intensities of the order
of tens and hundreds of GW/cm2. These phenomena can be observed
only in experiments with ultrashort pulses. Therefore, the details of the
relaxation mechanism become both important and interesting, as the
standard Boltzmann equation approach becomes inadequate. Different
theoretical approaches to this problem are presented in Refs. [71-73].
Being, in a sense, closely related to the problem of the coherence persis-
tance, it remains one of most important topics in the nonlinear optics of
semiconductors.

Acknowledgements. This work was partially supported by NATO Collab-

orative Research Grant CRG 930084. It is also a pleasure to acknowl-
edge here many illuminative and stimulating discussions with H. Haug,
A. Mysyrowicz, S. Moskalenko, N. Nagasawa, J. Wolfe, S. Tikhodeev
and especially A. Ivanov, whose assistance was very important. I am also
grateful to the Institute of Theoretical Physics of Frankfurt University
for hospitality while completing this work.
278 L. V. Keldysh
References
[I] S.A. Moskalenko, Fiz. Tverd. Tela 4, 276 (1962).
[2] LM. Blatt, K.W. Boer and W. Brandt, Phys. Rev. 126, 1691 (1962).
[3] S.A. Moskalenko, Bose-Einstein Condensation of Excitons and
Biexcitons, (Kishinev, RIO, 1970).
[4] L.V. Keldysh, in Problems of Theoretical Physics (Nauka, Moskow,
1972), p.433.
[5] E. Hanamura and H. Haug, Phys. Rep. 3, 209 (1977).
[6] T.M. Rice, Solid State Phys. 32, 1 (1977).
[7] A. Mysyrowicz, J. Phys. (Paris) 41 Suppl. 7, 281 (1980).
[8] C. Comte and P. Nozieres, J. Phys. (Paris) 43, 1069, 1083 (1982).
[9] L.V. Keldysh, Electron-hole liquid in semiconductors, in Modern
Problems of Condensed Matter Science 6, C D . Jeffries and L.V.
Keldysh, eds. (North-Holland, Amsterdam, 1987).
[10] L.V. Keldysh, Contemp. Phys. 27, 395 (1986).
[II] G.H. Wannier, Phys. Rev. 52, 191 (1937).
[12] N.F. Mott, Trans. Farad. Soc. 34, 500 (1938).
[13] S.A. Moskalenko, Opt. Spectrosk. 5, 147 (1958).
[14] M.A. Lampert. Phys. Rev. Lett. 1, 450 (1958).
[15] L.V. Keldysh, Proc. 9th Int. Conf on Physics of Semicond. (Moscow,
1968), p.1303.
[16] O. Akimoto and B. Hanamura, J. Phys. Soc. Jap. 33, 1537 (1972).
[17] W.F. Brinkman, T.M. Rice, and B. Bell, Phys. Rev. B 8, 1570 (1973).
[18] W.F. Brinkman and T.M. Rice, Phys. Rev. B 7, 1508 (1973).
[19] L.V. Keldysh and Yu.V Kopaev, Sov. Phys. Solid State 6, 2219
(1965).
[20] J. des Cloizeaux, J. Phys. Chem. Solids 26, 259 (1965).
[21] B.I. Halperin and T.M. Rice, Solid State Phys. 21, 115 (1968).
[22] E.A. Andryushin, L. V. Keldysh and A.R Silin, Sov. Phys. JETP
46, 616 (1977).
[23] R.R.Guseinov and L.VKeldysh, Sov. Phys. JETP 36, 1193, (1973).
[24] L.V. Keldysh and A.N. Kozlov, Sov. Phys. JETP 27, 521 (1968).
[25] J.J. Hopfield, Phys. Rev. 112, 1555 (1958)
[26] V.M. Agranovich and V.L. Ginzburg, Crystal Optics with Spatial
Dispersion and Excitons (Springer, Berlin, 1984).
[27] H.Haug, Z. Phys. 200, 57 (1967), Phys. Rev. 184, 338 (1969).
[28] R. Graham and H. Haken, Z. Phys. 237, 31 (1970).
[29] V Degiorgio and M.O. Scully, Phys. Rev. A 2, 1170 (1970).
[30] H. Haken, Rev. Mod. Phys. 47, 67 (1975).
 Macroscopic Coherent States of Excitons in Semiconductors 279
[31] H. Haug and S.W. Koch, Quantum Theory of Optical and Electronic
Properties of Semiconductors (World Scientific, London, 1993).
[32] Y. Toyozawa, Progr. Theor. Phys., Supplem.12, 11 (1959)
[33] A. Stahl and I. Balslev, Electrodynamics of the Semiconductor Band
Edge, Springer Tracts in Modern Physics 110 (Springer, Berlin,
1987).
[34] V.M. Galitskii et al, Zh. Eksp. Teor. Phys. 57, 207 (1969).
[35] V.F. Elesin and Yu.V. Kopaev, Zh. Eksp. Teor. Phys. 63, 1447
(1972).
[36] S. Schmitt-Rink and D.S. Chemla, Phys. Rev. Lett. 57, 2752 (1986).
[37] S. Schmitt-Rink, D.S. Chemla and H. Haug, Phys. Rev. B 37, 941
(1988).
[38] S. Schmitt-Rink, D.S. Chemla and D.A.B. Miller, Adv. Phys. 38,
89 (1989).
[39] H. Haug and S. Schmitt-Rink, Progress in Quant. Electron. 9,3
(1984).
[40] R. Zimmerman, Many Particle Theory of Highly Excited Semicon-
ductors, (Teubner, Leipzig, 1988).
[41] L.V. Keldysh, Sov. Phys. JETP 20, 1018 (1965).
[42] E.M. Lifshitz and L.P. Pitaevski, Statistical Physics, Part 2 (Perga-
mon, Oxford, 1980), chapters 10-11.
[43] J. Rammer and H. Smith, Rev. Mod. Phys. 58, 323 (1986).
[44] G.D. Mahan, Phys. Rev. 145, 251 (1987).
[45] K. Henneberger, Physica A150, 419 (1988).
[46] H. Haug, in Optical Nonlinearities and Instabilities in Semiconduc-
tors, H. Haug, ed. (Academic, New York, 1988) p. 53.
[47] A.L. Ivanov and L.V. Keldysh, Sov. Phys. JETP 57, 234 (1983).
[48] L.V. Keldysh and S.G. Tikhodeev, Sov. Phys. JETP 63, 1086 (1986).
[49] N.A. Gippius, L.V. Keldysh and S.G. Tikhodeev, Sov. Phys. JETP
64, 1344 (1986).
[50] A.L. Ivanov, L.V. Keldysh and V.V. Panashchenko, Sov. Phys. JETP
72, 359 (1991).
[51] L.V. Keldysh, Solid State Comm. 84, 37 (1992)
[52] A. Mysyrowicz, J.B. Grun, R. Levy, A. Bivas, and S. Nikitine, Phys.
Lett. A 26, 615 (1968).
[53] H. Suoma, T Goto, T. Ohta, and M. Ueta, J. Phys. Soc. Japan, 29,
697 (1970).
[54] N. Nagasawa, S. Koizumi, T. Mita, and M. Ueta, J. Lumin. 12/13,
587 (1976).
280 L. V. Keldysh
[55] F. Henneberger, K. Henneberger, and J. Voight, Phys. Stat. Sol.
(b) 83, 439 (1977).
[56] V.D. Phach, A. Bivas, B. Honerlage, and J.B. Grun, Phys. Status
Solidi (b) 84, 731 (1977).
[57] J.L. Oudar, A. Maruani, E. Batifol, and D.S. Chemla, J. Opt. Soc.
Am. 68, 1638 (1978).
[58] M. Ueta, T. Mito and T. Itoh, Solid State Comm. 32, 43 (1979).
[59] N. Nagasawa, T. Mito, and M. Ueta, J. Phys. Soc. Japan 41, 929
(1981).
[60] B. Honerlage, R. Levy, J.B. Grun, C. Klingshirn, and K. Bohnert,
Phys. Rep. 124, 161 (1985).
[61] V.D. Kulakovski, V.G. Lysenko and V.B. Timofeev, Sov. Phys. Usp.
28, 735 (1985).
[62] M. Ueta, H. Kanazaki, K. Kobajashi, Y. Tojozava, and E. Hana-
mura, Excitonic Processes in Solids (Springer, Berlin, 1986).
[63] F. Henneberger and J. Voigt, Phys. Status Solidi (b) 76, 313 (1976).
[64] V. May, K. Henneberger, and F. Henneberger, Phys. Status Solidi
(b) 94, 611 (1979).
[65] PL Khadzhi, S.A. Moskalenko, and S.N. Belkin, JETP Lett. 29,
200 (1979).
[66] H. Haug, R. Marz, and S. Schmitt-Rink, Phys. Rev. A 77, 287
(1980).
[67] R. Levy, B. Honerlage, and J. B. Grun, in Optical Nonlinearities and
Instabilities in Semiconductors, H. Haug, ed. (Academic, London,
1986), p. 181.
[68] M. Combescot and R. Combescot, Phys. Rev. Lett. 61, 117 (1988).
[69] M. Hasuo, N. Nagasawa, and A. Mysyrowicz, Phys. Stat. Sol. (b)
173, 255 (1992).
[70] M. Hasuo, N. Nagasawa, T. Itoh and A. Mysyrowicz, Phys. Rev.
Lett. 70, 1303 (1993).
[71] R. Zimmerman, Phys. Stat. Sol.(b) 173, 129 (1992).
[72] H. Haug, Phys. Stat. Sol.(b) 173, 139 (1992).
[73] M. Hartmann and W. Schafer, Phys. Stat. Sol.(b) 173, 165 (1992).
 13
Bose-Einstein Condensation of a Nearly
Ideal Gas: Excitons in C112O
J. P. Wolfe and J. L. Lin
Department of Physics
University of Illinois at Urbana-Champaign
Urbana, IL 61801, USA

D.W. Snokef
Mechanics and Materials Technology Center
The Aerospace Corporation
Los Angeles, CA 90009-2957, USA

Abstract
Systematic studies of the photoluminescence of free excitons in CU2O re-
veal the temporal and spatial properties of these particles under a vari-
ety of conditions. Their relaxation and diffusion are measured by time-
resolved spectroscopy and imaging. At low densities the thermalization
and diffusivity are well characterized by deformation-potential theory- i.e.,
electron-phonon coupling. At high densities, the excitonic gas establishes
a well-defined temperature as a result of frequent interparticle collisions
and displays quantum statistics. The orthoexciton gas tends to saturate at
the phase boundary for Bose-Einstein condensation, characterized by the
ideal-gas relation n — CT 3^2. Under appropriate conditions, the paraex-
citon energy distribution exhibits an extra component which is interpreted
as a Bose-Einstein condensate.

1 Excitons as a Nearly Ideal Gas

A gas within a solid. How can that be? That the elementary electronic
excitation of a semiconductor - the exciton - can be viewed as a nearly
free particle in a box takes quite a leap of the imagination. Yet the
basic wave nature of particles leads naturally to this picture. An electron
of mass mo in a region of volume V with constant potential has the
f Present address: Dept of Physics and Astronomy, University of Pittsburgh, Pittsburgh,
PA 15260, USA.

281
282 J. P. Wolfe, J. L. Lin and D. W. Snoke
well-known wave function,

w = 1 e i (kr- W ( ) ; (1)

corresponding to the kinetic energy,

E = h2k2/2m0, (2)
where k is the wave vector of the particle related to the classical mo-
mentum by hk = mov, and co is the wave frequency related to the kinetic
energy by hco = E. Similarly, an electron in a periodic potential (e.g.,
that of a crystalline lattice) is described by the famous Bloch wave
function,
xp = u k (r)e i(kT - a * ) , (3)
where Mk(r) is a periodic function whose period coincides with the lat-
tice structure. In its simplest form, Mk(r) is a sum of single atomic
wave functions residing on the nuclei in the crystal. Within some ar-
bitrary constant, the energy associated with a state near k = 0 (at
the band extremum, where Wk(r) depends only weakly on k) is given
by
E=h2k2/2me, (4)
where me is the effective mass of the electron reflecting the influence of
the atomic lattice. The miraculous implication here is that the electron
retains the same wave vector for all time; i.e., it moves without scattering
through this dense medium. This unique feature of the periodic potential
allows one to view the electron in a crystal nearly the same as one that
is freely moving in an empty box.
The extension to composite particles is straightforward. An electron
and a proton, by virtue of their Coulombic attraction, bind into a
hydrogen atom, whose center-of-mass wave vector, kcm = (mo+mp)vcm/ft,
is conserved. The hydrogenic wave function consists of an atomic-orbital
part multiplied by a plane-wave factor, exp[i(kcm • r — cot)]. The energy
spectrum for the bound pair is the Rydberg energy plus the kinetic energy
associated with the center of mass,
-e2 h2k2
cm
F — J_ (^\
2a0n2 2(me + mp)
where #o = h2/e2mr is the hydrogenic Bohr radius with reduced mass
mr = momp/(mo + mp), and n = 1,2,... is the principal quantum number.
The situation in the semiconductor is quite analogous: an electron in the
 BEC of a Nearly Ideal Gas: Excitons in CuiO 283

Table 1. Physical parameters for

Name Symbol Value Reference

Dielectric constant € 7 [2]
Electron mass me l.Omo [3,4]
Hole mass mh O.7mo [4]
Orthoexciton Mass m 3.Omo [5,6]
Rydberg (n >1) 97meV [1]
Binding energy (n=l) Ex 150 meV [7]
n=l Bohr radius oo = e 2 /2£ x e 7A
Band gap /7
^ gap
2.17 eV ( 4K) [1]
Ortho-Para 17
-C'ex
12meV [8]
exchange splitting

conduction band and a hole in the valence band bind together to form an
exciton with a center-of-mass wave vector kcm = ke + k/, and total mass
m = me + m^ The factor e2 in (5) and in ao is replaced by e2/e (where e
is the crystalline dielectric constant) and generally the mass of the hole is
comparable to that of the electron. In fact, the analogy between excitons
and positronium is even better than the analogy between excitons and
hydrogen, since the positron can be seen as a "hole" in the negative-
energy Dirac sea. The analogy between excitons and positronium is
better in CU2O than in most semiconductors, since the lowest conduction
band and the highest valence band are spherical and well separated
from other energy bands, and the effective electron and hole mass are
close to the free electron mass. This is illustrated by the fact that the
excitonic Rydberg in CU2O is almost exactly (13.6 eV)/2e2, the value
for positronium when e1 —• e2/e. Values of relevant parameters for the
crystal CU2O are summarized in Table 1. The Rydberg series of states
for excitons in CU2O has been clearly observed by optical absorption [1].
In the present review, we deal exclusively with the n = 1 ground state
because the thermal populations of the excited states are small, since the
binding energy of the exciton in CU2O is so large. The excitonic binding
energy Ex =150 meV is somewhat larger than the Rydberg, R^ = 91
meV, which describes the excited (n = 2,3,...) states, due to central-cell
corrections for this relatively small Wannier exciton.
Similar to the hydrogen atom, positronium and many atoms, the
exciton is a composite boson, having integer total spin. A collection
of excitons may be regarded as a gas of bosons provided the average
284 J. P. Wolfe, J. L. Lin and D. W. Snoke
distance between particles is much larger than the Bohr radius, or

na\ < 1, (6)

in terms of the gas density, n. For CU2O, ax equals 7 A(deduced from

the excitonic Rydberg Ex via the above equations), which requires that
n « 5 • 1021 cm"3, a condition which is well satisfied for the experiments
described in this paper. The condition for a gas of hard spheres of radius
ax to be weakly interacting is [9]

-, (7)
or
na\ < -?-, (8)

for the interaction vertex (7(0) = 4naxh2/m. This condition is also fulfilled
for the experiments discussed here.

2 Creation and Observation of Excitons

A pure semiconductor at zero temperature has no mobile charge carriers.
The valence band states are completely filled with electrons and the con-
duction band is completely empty. Photons with energy, hv, larger than
the semiconductor band gap, £ gap , create non-equilibrium electrons in the
conduction band and holes in the valence band which can bind pairwise
into excitons.f Once formed in a relatively pure crystal, the excitons
behave as mobile particles with finite lifetime. Unlike the single particle
in a box, however, the excitons find a variety of ways to scatter from
one wave vector to another. In particular the periodicity of the lattice
is easily broken by thermal vibrations, or phonons, so that excitons can
lose energy and cool down. The diffusivity and thermal relaxation of
excitons will be discussed in detail below. First we must consider how
excitons are detected.
Excitons make themselves visible by decaying. The radiative recombi-
nation of the electron and the hole composing the ground-state exciton
produces a photon with an energy,
f Under typical conditions the density of excitons, n, far exceeds the density of electrons
and holes, ne = nh. Statistical mechanics predicts nl/n « C(T)exp(Ex/kBT), where C{T)
is given approximately by HQ in (22) below, so that for Ex = 150 meV, kBT = kB(50 K)
= 4 meV and n = 1014 cm" 3 , we find ne/n » 10" 7 . See Ref. [10].
 BEC of a Nearly Ideal Gas: Excitons in CuiO 285

hv
(phonon assist)

Fig. 1. Recombination processes of excitons in C112O.

h2k2
hv = £ g a p — Ex + (direct transition) , (9)
2m
h2k2
hv = £ Ex + 2m (phonon-assisted) , (10)
gap
where E gap is the semiconductor energy gap, Ex = e2/2ax is the ex-
citonic binding energy, | and hQt is the energy of an optical phonon
which is emitted or absorbed simultaneously with the photon emission.
Momentum conservation simultaneously requires for the two types of
processes,
k = kphoton (direct transition) (11)

= kphoton + (phonon-assisted) (12)

These two recombination processes are illustrated in Fig. 1. The

direct process can only occur for an exciton with momentum matching
that of the emitted photon, yielding a sharp luminescence line at nearly
zero kinetic energy (E « E2ap/2mc2 « 10~ 6 eV). J The phonon-assisted
process, on the other hand, samples all of the occupied exciton states
because the exciton momentum is easily absorbed by the optical phonon.
Since the energy of the optical phonon (and the transition probability) is
t Ex = 150 meV is the binding energy of the lowest lying exciton, the paraexciton (see
below). For the orthoexciton, which lies higher by an exchange energy of 12 meV, Ex in
these expressions must be replaced by Ex — 12 meV. Applied stress modifies both £ g a p
and Ex and splits the orthoexciton levels.
t This one-photon decay process is allowed for excitons, unlike positronium, because
the energy gap of the semiconductor is far less than the positronium "gap" of 2mc2.
Two-photon decay is also allowed for excitons, like positronium, but the rate for this
process is much less than the single-photon emission processes.
286 J. P. Wolfe, J. L. Lin and D. W. Snoke

Table 2. Zone-center optical phonons in

Relative Radiative Efficiency

Designation Energy Ortho Para
3
rj 5 (IY) 11.0 meV 17 (Ref. [13]) 1 (Ref. [15])
2
rf 2 (iY) 13.8 meV 500 « 0
3
rr 5 (r 4 -) 18.7 meV
43meV
17 (Ref. [14])
3
« 0
3
rj(r^) 64meV 3
3
rr 5 (r 4 -) 79meV 8

only weakly dependent on the phonon wave vector,^ the phonon-assisted

process gives an accurate replica of the energy distribution of the excitons.
This situation is unlike that of liquid helium-4, in which the momentum
distribution can only be deduced indirectly from neutron scattering data.
Because there are several optical phonons with appropriate symmetry
to participate in this recombination process, each having a different
energyftCli9there are several "phonon replicas" of the excitonic kinetic
energy distribution observed in the recombination spectrum. Table 2 lists
the energies of zone-center optical phonons, from Ref. [12], denoted by
their irreducible representation in the cubic group, and gives the relative
radiative strengths of the corresponding phonon-assisted luminescence
processes.
The symmetries of the electronic and vibrational states are crucial to
the relaxation properties of the excitons - in particular, their lifetime
and scattering rates. A useful review of these symmetry properties is
given in Ref. [7]; also, Birman [16] gives a useful discussion of the
effect of symmetry on the phonon-assisted recombination rates. At this
point we only need to know the following facts: (1) CU2O is a forbidden
direct-gap semiconductor; that is, the parity of the conduction-band
minimum is the same as that of the valence-band maximum, implying
that dipole radiation is forbidden. This property leads to relatively
long excitonic lifetimes and correspondingly weak optical transitions. (2)
Because the exciton is composed of a spin-1/2 electron and spin-1/2
hole, its ground state is split into an "S = 0 like" paraexciton and
an "S = 1 like" orthoexciton by the electron-hole exchange interaction,
| For typical gas temperatures of 100 K or less, the exciton wave vectors are small
(k < 107 cm" 1 ) compared to the Brillouin-zone boundary (~ 108 cm" 1 ). The optical-
phonon dispersion, as measured by neutron scattering [11], is negligible for these wave
vectors.
 BEC of a Nearly Ideal Gas: Excitons in CujO 287

T=74K

1.99 2.01 2.03 2.05

Photon energy (eV)
Fig. 2. Luminescence from Cu20 at various temperatures. (From [18].)

also called the "virtual annihilation" interaction. Only the orthoexciton

feels this exchange energy - just as in positronium, the triplet state
has higher energy. | (3) The orthoexciton luminescence is quadrupole
allowed, whereas direct recombination of the paraexciton is forbidden
to all orders. Phonon-assisted recombination of paraexcitons is weakly
allowed, d) Formation of other complexes, such as biexcitons or electron-
hole liquid, is energetically unfavorable, due to the low band degeneracy
and near-unity electron-hole mass ratio [17].
Photoluminescence spectra of orthoexcitons in CU2O are shown in
Fig. 2. To obtain these, a high-quality natural crystal (from a copper
mine) is cooled in a cryostat and weakly excited with a continuous (cw)
Ar + laser beam (2.4 eV). The sharp spike is due to direct recombination
of the orthoexciton. The broad spectral lines flanking the direct emission
are phonon-assisted orthoexciton lines, corresponding to the emission
or absorption of a Tj"2 optical phonon with an energy of Ml = 14

I There exists some confusion about whether electron-hole exchange gives a positive shift
to the ortho state, or a negative shift to the para state. In general, the answer depends
on the symmetry of the conduction and valence bands. For CU2O, the ortho state is
shifted up; as noted in Ref [6], the mass renormalization of the ortho (M ort h o = 3mo
instead of Mortho =me-\-m\x = 1.1 mo) implies a k-dependent positive exchange-energy
shift of the ortho state, analogous to that of the positronium triplet state.
288 J. P. Wolfe, J. L. Lin and D. W. Snoke
meV. As the temperature is lowered, the thermal population of optical
phonons decreases and the phonon-emission line dominates. The overall
temperature shift of the spectrum is due to a blue shift of the band gap,
£gap, as the crystal thermally contracts.
The dashed lines in Fig. 2 show an analysis of the low temperature
orthoexciton spectra. The phonon-assisted spectra are fit to a Maxwell-
Boltzmann distribution of the form,

N(E) oc El/2e~E/kBT, (13)

where E is the kinetic energy of the exciton. The same fit temperature
T is assumed for the excitonic gas (giving the shape of the lines), and
for the crystal lattice (giving the relative intensity of the two lines,
/(emission)//(absorption) = ehQ/kBT). These data show unequivocally
that the orthoexcitons in weakly excited CU2O behave as a "classical
gas" in contact with a thermal reservoir, the lattice vibrations. The
lifetime of the orthoexcitons at low temperature is measured to be about
30 ns, so we conclude that the energy relaxation of these particles,
originating as hot electrons and holes, is sufficiently fast to reach thermal
equilibrium with the lattice within this time. As we shall see, the cooling
of the gas is accomplished by emission of phonons.
What is the cause of the 30 ns lifetime of the orthoexcitons? Recent
measurements indicate that the radiative lifetime of orthoexcitons is
much longer - several hundred nanoseconds [18]. It is most likely,
therefore, that the measured lifetime in this dilute-gas limit is due to
the decay of orthoexcitons to paraexcitons. Snoke et a\. [19] have
suggested that this interconversion process can take place by emission of
a single acoustic phonon. This mechanism appears to be consistent with
measurements of ortho-para interconversion times by Weiner et al. [20]
and with earlier measurements of luminescence intensities as a function
of temperature.
In contrast, the paraexcitons display lifetimes of many microseconds
in high quality naturally grown crystals [21], because their radiative
recombination is highly forbidden. | For a given experimental setup, the

The defect concentration in common synthetic crystals causes non-radiative lifetimes of

only nanoseconds or less, even for paraexcitons, which leads to non-Maxwellian exciton
distributions even at low density, since the excitons cannot fully thermalize by scattering
with phonons within their lifetime [22].
 BEC of a Nearly Ideal Gas: Excitons in Cu2O 289
luminescence intensities of the ortho and paraexcitons are given b y |

IO(E) = BNO(E)/Tro (ortho) , (14)

IP(E) = BNP(E)/Trp (para) , (15)

where the constant B contains the experimental collection efficiency

(which is the same for ortho and para), and N(E) are the respective
energy distributions. The most radiatively efficient phonon-replicas are
the rj~2 line for ortho and the I~Y5 line for para, and the ratio of the
radiative rates for these two processes is measured to be T rp /i ro « 500.
(See Table 2.) Consequently, the paraexcitons are difficult to observe in
pulsed experiments within time windows less than about 10 ns, and in
order to collect enough photon counts one must repeat the experiment
at least 105 times per second (a 100 kHz pulse-repetition rate).

3 Diffusion of Excitons
The long lifetimes of the paraexcitons have permitted Trauernicht et
al. [23] to observe their spatial diffusion by time-resolved luminescence
imaging. The two images of paraexciton luminescence in Fig. 3 show
the expansion of the paraexciton gas into the crystal. The diffusion
constant, D, can be determined by a straight-line fit to the square of the
cloud radius as a function of time [23]. A temperature dependence of
the measured diffusion constant is shown in Fig. 4. With each diffusion
measurement, the temperature of the paraexciton gas was obtained by
fitting its spectrum to the Maxwell-Boltzmann distribution (13).
The measured diffusion constants are huge, ranging up to 1000 cm 2 /s
for the lowest experimental temperature of 1.2 K. The strong temperature
dependence indicates that this process is mediated by thermal vibrations
of the lattice, i.e., exciton-phonon scattering. Taking into account the
Planckian distribution of phonons at temperature T, and making the
high-temperature approximation k%T > mv2L, where VL is the lattice

t Broadening of the spectrum associated with the lifetime of the excitons is negligible.
The Lorentzian broadening from the uncertainty principle implies a spectral full width
at half-maximum of 1.2 meV for a lifetime of 1 ps. Therefore, the broadening of the
orthoexciton spectrum due to a 30 ns lifetime is an unmeasureable 0.04 //eV. Also,
broadening due to exciton-phonon interactions, with a typical scattering time of 30 ps,
leads to a linewidth of only 0.04 meV at low temperature. On the other hand, interparticle
collisions can cause a measurable broadening at high gas densities [13].
290 J. P. Wolfe, J. L. Lin and D. W. Snoke

-crystal surface
Fig. 3. Time-resolved images of the paraexciton luminescence of Cu2O at T=2K,
obtained by x-y scanning of the crystal image across an entrance aperture of
a spectrometer, (a) £=0.2 fis after a 100-ns Ar + laser pulse is absorbed at the
crystal surface, (b) £=0.6 /is.(From [23].)

phonon velocity, deformation-potential theory gives the result

1 = S2mkBTh
(16)
T(k)
where p is the crystal mass density and S is the deformation potential
(basically the shift in £ gap with strain). If the excitons are distributed ac-
cording to a Maxwell-Boltzmann distribution at the lattice temperature,
then this leads to
D
^likBT =
^T~1'2' (17)

where R is a constant independent of T (see Ref. [23]). Figure 4 shows

that the measured diffusion constant for paraexcitons deviates markedly
from this predicted T~1/2 dependence for temperatures below about
30 K. Trauernicht et al. [24] have shown that this unusually sharp rise
in diffusivity at low temperatures can be understood in terms of the
breakdown of the high-temperature approximation given above. When
the exciton velocity is smaller than the lattice phonon velocity, energy
and momentum conservation forbid the exciton from emitting a phonon.
Therefore as the gas temperature is lowered, the diffusivity eventually
increases exponentially.!
t The extremely high diffusivities of the paraexcitons at low temperatures is a strong testa-
ment to the perfection of these natural-growth crystals. At T = 1.2 K, the corresponding
mean free path between scattering events is / = 3D/vay « 70/mi! The mean scattering
 BEC of a Nearly Ideal Gas: Excitons in 291

1000

CM

100

-r-l/2

10 10 30 100
T (K)

Fig. 4. Temperature dependence of the paraexciton diffusion constant in CU2O.

(From [23].)

Notice that in (16), the scattering rate varies inversely as the square
of the phonon velocity. For spherical electron and hole bands (as in
CU2O), only longitudinally polarized phonons can couple to the carriers.
Trauernicht et al. [24] show that if the crystal is stressed, the reduced
symmetry allows the excitons to couple also to transverse phonons, which
have a velocity about four times smaller than that of the longitudinal
phonons. Hence, an externally applied stress acts to reduce the diffusivity
of the excitons and may be an effective means of confining the excitons
to a smaller volume.

4 Quantum Statistics for the Nearly Ideal Bose Gas

We are all familiar with the profound effects of quantum statistics on
electrons in metals. As indistinguishable spin-1/2 particles, electrons are
fermions which are unable to share the occupancy of a single state. In
terms of scattering theory, this constraint implies that the probability of
an electron being scattered into a state with wavevector k is proportional

time is T = l/vav « 20 ns, which implies a mobility of x/m = 10 8 cm 2 /eV-s, perhaps the
highest measured for any electronic species in a non-superconducting bulk crystal.
292 J. P. Wolfe, J. L. Lin and D. W. Snoke
to 1 — /k, where /k is the occupation number of the state. This is the
Pauli exclusion principle at work.
The exciton, being composed of two spin-1/2 particles, is an integral-
spin particle, or composite boson. The quantum statistical rule for a
boson is that its probability of being scattered into the state of wavevector
k is proportional to 1 + /k- In other words, the likelihood that a boson
is scattered into a given state is enhanced if other bosons already occupy
that state. It can be shown [25] that this "stimulated" scattering leads to
the Bose-Einstein distribution function,

with E = h2k2/2m the kinetic energy of an exciton and \x the chemical

potential, which is measured with respect to the total energy of the
exciton at E = 0 and determined by the condition

where N is the total number of excitons in the system.

For a macroscopic volume, the number of plane-wave states per unit
energy for the ideal (non-interacting) gas has the form D(E) = C'EX^2.
Therefore, the number of excitons per unit energy is given by the product
of this density of states and the occupation number f(E):

If N excitons occupy a volume V, then C = (gV/4n2)(2m/h2)3/2 from

the usual counting of k states [26], where g is the spin multiplicity. In an
unstrained crystal, g = 3 for orthoexcitons and g = 1 for paraexcitons.
The constraint (19) may be restated as N = f N(E)dE, which in terms
of the gas density, n — N/V9 becomes
3/2

For a given density and temperature, (21) fixes the chemical potential,
fi, of the gas. By defining e = E/kBT and the "dimensionless chem-
ical potential" a = —fi/k BT, Eq. (21) can be written in terms of a
dimensionless integral as,
e l/2

-^-jde. ill)
 BEC of a Nearly Ideal Gas: Excitons in 293

Ideal Bose-Einstein distribution

1000

e = E/kT
Fig. 5. The Bose-Einstein distribution N(E) = f{E)D(E) for several values of a
for an ideal three-dimensional gas at constant potential.

where HQ = g(mkBT/2nh2)3/2 is the "quantum density", at which the

thermal DeBroglie wavelength X = 2n/kav of the particles is comparable
to the interparticle spacing. In the classical limit, a is much greater than
unity and (22) yields n = nQe~a, or \i = —k BT\n(riQ/n).
The shape of the kinetic energy distribution depends on the values of
fi (or a) and T. In Fig. 5 we plot (20) at a fixed gas temperature as
the density of the gas is hypothetically increased. At low densities, the
chemical potential is large and negative, which reduces the distribution in
(20) to the density-independent shape given by (13). As we increase n, the
constraint of Eq. (22) forces a closer to zero. When a approaches 1 (i.e.,
n —• Mg), the spectrum becomes more sharply peaked at low energies. In
this quantum regime, the (1 + / k ) final-state factors become important
and scattering into low-energy states is enhanced. Compared to a classical
distribution at temperature T, Bose statistics produce a narrowing of the
energy distribution, just the opposite of how Fermi statistics broaden the
distribution.^
As more particles are added to the gas at constant temperature, a
surprising thing happens. A density is reached where [i = 0 and the
integral in (21) is still finite, as indicated in Fig. 5. This fact can be
understood by expanding the integrand about the singularity at E = 0

f Fermi broadening of degenerate electron-hole plasmas has been clearly observed, for
example, in electron-hole droplets in Ge and Si [27].
294 J. P. Wolfe, J. L. Lin and D. W. Snoke
to obtain

which is integrable. The dilemma of where to put the next particles is

surmounted by realizing that the £ 1 / 2 density of states fails to completely
describe a system of discrete states. By bringing \i arbitrarily close to the
ground state of the system (corresponding to the longest wavelength that
matches the boundary conditions of the box) it is theoretically possible
to put an arbitrarily large number of particles into this "fc = 0" state. In
particular, in this idealized system, the number of particles in the ground
state is simply,

/<°>= ^ r r * ~g/a' (24)

which can become an arbitrarily large number as a approaches zero.
For a realistic volume, the first excited /c-state will also have a large
occupation number, but considerably fewer particles than the ground
state, due to the sharply peaked distribution function. This effect in which
thermodynamic equilibrium of the ideal Bose gas forces a macroscopic
number of particles into the lowest state is the idealized concept of
Bose-Einstein condensation.
Of course, the quantum states of real gases (atomic or excitonic) can be
significantly broadened by interparticle scattering, so the simplistic view
of well-defined plane-wave states at low energy is not perfectly valid.
Nevertheless, the non-interaction picture is a useful starting point for
considering the weakly interacting gas and, as we shall see, appears to do
a remarkably good job of describing the excitonic energy distributions
even at a « 0. For an excellent review of the thermodynamic properties
of a weakly interacting Bose gas, see Ref. [28].
The phase boundary for BEC of the ideal Bose gas is determined by
setting a = 0 in (21), yielding,

nc = 2.612nQ = CT3/\ (25)

with C = 2.612g(mkB/2KH2)3/2. This relation is plotted for orthoexcitons

(g = 3) and paraexcitons (g = 1) in Fig. 6, using the mass of the
orthoexciton in Table 1. We now see why excitons are a prime candidate
for observing this quantum effect: their effective mass is approximately
that of an electron, rather than that of an atom. At a temperature of 4 K,
a density of only about 1017 cm" 3 is required to reach the orthoexciton
phase boundary. Alternatively, an experimentally-attainable density of
 BEC of a Nearly Ideal Gas: Excitons in CU2O 295

100
BEC phase boundaries

//Vff,
m = 2.7m 0 / /

fc 10 - III!, / /
/ cond ensate /

10" 'ifttl/ll/,7/
IO IC
' / / / / , / /
I019
Density (cm 3 )
1020

Fig. 6. Phase diagram for excitons in Cu2O.

1019 cm 3 requires that the gas be cooled below 30 K. (In contrast, a

spin-aligned hydrogen gas at this density must be cooled to millikelvin
temperatures.)
Not surprisingly, nature often constructs unexpected and interesting
obstacles to lofty goals, as we shall see in the remainder of this article. An
interesting complication can be anticipated immediately: we are dealing
with a two-component gas, composed of the ("S = 0") paraexcitons and
("S = 1") orthoexcitons. We expect that the orthoexcitons and paraexci-
tons will come into thermal equilibrium, or quasiequilibrium (i.e., exhibit
the same gas temperature) within a few collision times, which is just a few
picoseconds at the relevant densities. However, as we discussed in Sec-
tion 2, the ortho-para interconversion rate is relatively slow - typically
30 nanoseconds. Therefore on a nanosecond time scale the two compo-
nents will not attain a chemical equilibrium. As they relax from a high
temperature and density, the ortho and para components should exhibit
distinct chemical potentials, determined by their respective densities.

5 Equilibration and Cooling of the Excitonic Gas

We are confronted with the task of producing an excitonic gas of suf-
ficiently high density and low temperature, as dictated by (25). In the
experiment of Fig. 3, orthoexcitons are produced by surface excitation of
the crystal. They diffuse away from the surface and convert to paraexci-
tons within a lifetime of % « 30 ns. With cw excitation, the steady-state
number of particles is given by N = Gz, where G is the pair generation
rate. For a typical laser power of 100 milliwatts, G = 3 • 1017 pairs/s,
which implies that N = 1010 particles are continuously present. The
steady-state volume of the orthoexciton gas is found to be in the range
296 J. P. Wolfe, J. L. Lin and D. W. Snoke
10~5 to 10~6 cm3 from an estimate of the excitonic diffusion length,
/ ~ y/Dx, and the area of the laser spot. The resulting density of 1015 to
1016 cm" 3 is well below the BEC phase boundary.
It turns out that higher densities are not so easily achieved by simply
raising the cw excitation power. The reason is that as the density is
raised the lifetimes of the excitons are shortened by particle-particle
interactions. In a collision, one exciton can recombine non-radiatively
by raising the kinetic energy of the other [29]. This process (Auger
recombination) leads to a sublinear increase in density with generation
rate, namely n oc y[G. Consequently, the most effective way of reaching
the quantum regime is by using pulsed surface excitation and observing the
excitonic gas before it diffuses far away from the surface. Instantaneous
densities of 1019 cm" 3 can be attained in this way, which may be viewed
as a sort of "inertial confinement." |
By "surface excitation" we mean that the incident light is absorbed
within a small distance from the crystal surface. A 2.4 eV (green) photon
from an Ar + laser has an absorption depth of about 1 fim in CU2O and
produces an "excess energy" of hv— E g2Lp «0.3 eV per pair, corresponding
to an initial carrier temperature of about 3000 K. The "hot" electrons
and holes quickly cool towards the lattice temperature by the emission of
phonons. As the kinetic energy falls below the excitonic binding energy
(150 meV), excitons are formed, which continue to cool by phonon
emission4
In discussing the approach to "equilibrium," we must consider three
separate questions: (1) Under what conditions is the excitonic gas de-
scribed by a temperature? (2) How quickly does the excitonic gas cool
to equilibrium with the crystal via phonon emission? (3) Under what
conditions is the exciton gas described by a single chemical potential?
The thermal equilibration time for the excitons to establish a gas
temperature has not been measured directly; some reasonable estimates
can be made, however. The exciton-exciton scattering rate is given
roughly by

— = nvavG, (26)

where t>av is the mean velocity and o is the exciton-exciton scattering

cross section. Although the exciton-exciton scattering cross section has
not been measured directly (Ref. [6] gives an upper bound) we can
t An alternative confinement method is considered in Section 11.
X There is evidence in some experiments that the excitons form almost immediately, cooling
as a unit towards the lattice temperature [30].
 BEC of a Nearly Ideal Gas: Excitons in 297

0 10 20 30 40
Kinetic energy (meV) \J 5 10
3 IU 15
ID
Kinetic energy (meV)

Fig. 7. (a) Theoretical particle distribution for a boson gas undergoing only
interparticle scattering, for n « nQ. The label of the curves is the number of
scattering events per particle since the initial creation. The last curve is within
1% of a Maxwell-Boltzmann distribution. (From Ref. [35].) (b) Theoretical
particle distribution for a boson gas undergoing only inlelastic scattering with
phonons, assuming an unchanging phonon bath temperature below the critical
temperature for condensation of the gas. The particle mass and particle-phonon
scattering matrix element are those of excitons in CU2O, but the model does not
take into account exciton-exciton interactions important in this density regime.
(From Ref. [6].)

assume that a is close to na2x.^ Then for a typical gas temperature

of 30 K and density of 1018 cm"3, the scattering time xe is about 30
picoseconds.
A simulation of hard sphere scattering, shown in Fig. 7(a), tells how
many scattering times, re, are required for a hard sphere gas to attain an
internal thermal equilibrium. The starting energy distribution is taken to
t For s-wave scattering of hard spheres with radius a, the cross section is Ana2. Exciton-
exciton scattering almost certainly is not well modeled by infinite-potential hard-sphere
scattering, however. The general form for exciton-exciton scattering has not been
calculated; Ref. [31] gives a trial form extrapolated from a calculation for hydrogen
which does not account for electron-hole exchange and reduced mass; Refs. [17, 32, 33]
give estimates of the average interaction including spin effects.
298 J. P. Wolfe, J. L. Lin and D. W. Snoke
be a narrow distribution at 20 meV. The details of this kinetic simulation
may be found in Refs. [34] and [35]. One finds that after only five scat-
tering times, independent of the shape of the initial energy distribution, the
gas attains a distribution indistinguishable from a Maxwell-Boltzmann
distribution. Therefore, the excitonic gas at relatively moderate densities
(n > 1017 cm~3) is capable of establishing an internal gas temperature, dif-
ferent from the lattice phonon temperature, in a very short period of time
- a few picoseconds.
This type of simulation of the particle dynamics has been extended to
higher densities where quantum statistics come into play. At high den-
sities, the factor 1 + /k enhances the scattering rate, but more scattering
events are required to establish the Bose-Einstein distribution. Moreover,
one is confronted with the important question of how long it takes for
a macroscopic condensate to form, a subject of some theoretical debate.
It turns out that density-of-states considerations indicate that the con-
densation process is greatly assisted by the interactions of particles with
phonons. As shown in Fig. 7(b), in an ideal gas interacting with phonons,
a macroscopic number of particles can enter near-zero energy states in
a sudden transition after a finite time. Of course, a large population at
near-zero energy is not the same thing as a condensate, and the articles
by Kagan and by Stoof in this volume give theoretical calculations of the
condensation time from a many-body approach. Nevertheless it seems
clear that the exciton gas can reach a very degenerate distribution within
its lifetime.
Now let us consider more directly the second question of how quickly
the kinetic energy of a hot gas is lost by phonon emission in a cold crystal.
Recently this cooling process has been measured by Snoke et al. [6] for
a low-density gas of excitons, and the results are shown in Fig. 8(a). The
crystal is excited by a 5 ps dye-laser pulse, and the photoluminescence
spectra are recorded by a streak camera with 10 ps time resolution.
(Readers who are curious about how one records a photon spectrum in
billionths to trillionths of a second will find a useful review in Ref. [36].)
The laser photons are tuned to 2.067 eV, which is 34 meV above the 2.033
eV band minimum of the orthoexcitons; in this phonon-assisted process,
14 meV goes towards creating a Fj~2 phonon and 20 meV is given as
kinetic energy to the exciton. The exciton luminescence is assisted by
emission of a Fj~2 phonon, so the zero in kinetic energy appears at about
2.018 eV in the data. At t = 0, the largest feature in the spectrum
corresponds to excitons with 20 meV kinetic energy.
As seen in Table 2, there are several types of optical phonons with
 BEC of a Nearly Ideal Gas: Excitons in CU2O 299

t(ps)

0.0

10.8 —
c
c 21.6
CD

32.4 E
U
N
43.2
W
E 10
54.0

64.8

2015 2030 2045 10 100

Photon energy (meV) T (K)

Fig. 8. (a) Time-resolved luminescence from CU2O following a very short (5 ps)
laser pulse at 2.033 eV, measured with a streak camera. The dashed lines are
a fit to the solution of the Boltzmann equation for exciton-phonon scattering
based on deformation-potential theory. (From Ref. [6].) (b) The energy loss rate
for excitons in a Maxwell-Boltzmann distribution and zero phonon temperature,
using the phonon emission rates deduced in Ref. [6].

hQt < 20 meV. The most favored process is emission of the 18.7-meV
TO phonon, apparent in the data of Fig. 8(a) in the peak near zero
kinetic energy. The other sharp peaks are longitudinal-acoustic (LA)
phonon emission peaks. Phonon emission depletes the 20-meV kinetic
energy states in about 30 ps, rather slow compared to semiconductors
such as GaAs, but still much less than the exciton lifetime. At 64.8 ps,
the distribution resembles a Maxwell-Boltzmann distribution with T «
50 K, which is still considerably higher than the 18 K lattice temperature.
The exciton gas does not reach the lattice temperature until hundreds of
picoseconds later (see Fig. 12). Experiments such as these give the matrix
elements for the optical and acoustic phonon emission processes. Fig. 8(b)
is a summary of the energy loss rate for an exciton gas at temperature
T, with the lattice temperature assumed to be zero. The dotted curve is
300 J. P. Wolfe, J. L. Lin and D. W. Snoke
the T3/2 behavior predicted in Section 3 for acoustic-phonon emission.
Deviation from this line at lower energies is due to the "freeze-out"
process indicated in Fig. 4. The enhancement at higher temperatures
is due to optical-phonon emission. The total of these processes gives a
temperature dependence of the rate of energy loss to phonons of roughly
T5'2 in the range 1-100 K.
Finally, we can ask under what conditions the ortho and para compo-
nents of the exciton gas reach chemical equilibrium. As mentioned above,
orthoexcitons can decay via a phonon-assisted process with a time scale
of 30 ns at 2 K and about 600 ps at 30 K [19, 20]. f Phonon-assisted
paraexciton conversion to the orthoexciton state lying 12 meV higher
requires phonon absorption, which is highly suppressed at low tempera-
ture. At high densities, however, the two components can "convert" into
each other via the Auger process discussed above, which destroys one
exciton and ionizes the other. The free carriers thus created can form
into either kind of exciton. This leads to a relatively efficient para-to-
ortho "up conversion" - orthoexcitons have been observed even at 2 K
at moderate densities [37]. Because this process is so fast, ortho and
para excitons will in general never be in chemical equilibrium, i.e. have
population ratio N0/Np = 3exp(—A/k BT), at densities above 1017 cm"3
and lattice temperatures below 30 K; instead, their relative number will
be determined by a rate balance between conversion processes. The two
components will have the same temperature but two different "effective
chemical potentials", oc0 and ocp.
If the two components were in chemical equilibrium, the orthoexcitons
would hardly ever have on0 less than A//CB T « 4 at 30 K, and they would
never deviate from a Maxwell-Boltzmann distribution (see Fig. 5) at low
temperature. As we shall see, in fact the orthoexcitons exhibit quantum
statistics over a wide range of density and temperature. The ortho-to-
para conversion process can be enhanced, however, by raising the lattice
temperature, pushing the gas toward chemical equilibrium.
Table 3 gives a summary of the various rates of relaxation processes
for excitons in CU2O. Relaxation and recombination processes of excitons
occur in CU2O on time scales which vary from picoseconds to millisec-
onds, and properly modeling this system requires understanding which
processes are important in a given density and temperature regime.

f Orthoexcitons are also expected to convert to paraexcitons by flipping spins in two-

body collisions, analogous to the process observed in hydrogen and positronium, but
definitive evidence for this process has not yet appeared. See Section 7.
 BEC of a Nearly Ideal Gas: Excitons in 301

Table 3. Relaxation processes of excitons in CuiO. (For excitons at

T=30 K and n=10 19 cm"3, and lattice at TL=0)
Process Timescale Proportional to Reference

exciton formation 250 fs [30]

from free carriers
exciton-exciton 3 ps nT1'2 [6]
scattering
exciton-phonon 30 ps ~ T5/2 [6]
scattering
exciton-exciton > 200 ps nT1'2 [38]
Auger recombination
ortho-to-para 600 ps ~ T 3/2 [19, 20]
phonon-assisted
conversion
para-to-ortho 30 ns ~ T3/2e~A/kBT [19, 20]
phonon-assisted
conversion
ortho radiative 300 ns constant [18]
recombination
para radiative >150/xs constant [21, 18]
recombination

6 The Exciton as a Boson: Spectroscopic Considerations

How do we analyze the luminescence spectra of excitons at high densi-
ties? Over a range of conditions, the ortho and paraexciton spectra are
expected to have the same temperature but different chemical potentials,
so we must consider each component separately. The basic procedure is
to extract a and T from a fit of a phonon-assisted spectrum to (20). In
the quantum regime, the chemical potential and temperature are quite
orthogonal fit parameters, as indicated in Fig. 5. This logarithmic plot
shows that the gas temperature manifests itself as a linear high-energy
tail, whereas the chemical potential (or a = —H/UBT) governs the sharp-
ness of the low-energy portion. Given the spectrally determined a and
T, we can then determine the density of this component of the gas by
calculating the integral in (22).
An important experimental example of the spectroscopic information
302 J. P. Wolfe, J. L. Lin and D. W. Snoke
gained from the two-component exciton gas is given in Fig. 9. Time-
resolved spectra following a 10-ns laser pulse capture the energy distri-
butions of the ortho and paraexcitons at "high" and "low" densities by
choosing sampling times of 16 ns and 44 ns, respectively, after the onset
of the laser pulse. The ortho and para components are superimposed by
shifting their spectral positions. At low densities (t > 44 ns), both the
para (solid line) and ortho (dashed line) spectra at all times exhibit a
Maxwell-Boltzmann distribution with T = 32 K, which is taken as the
lattice temperature.! The off-scale signal on the high-energy side of the
para spectrum is due to a strong ortho luminescence replica.
At high density (t = 16 ns), the energy distribution of the paraexcitons
is considerably narrower than that of the orthoexcitons, which fit a
Maxwellian distribution at a slightly elevated gas temperature of 38 K
(the gas is heated at early times). Assuming the same temperature for
the paraexcitons, the paraexciton distribution is described by a Bose-
Einstein distribution (circles) with a = 0.07. At these relatively high
lattice temperatures, the orthoexcitons convert efficiently to paraexcitons
via phonon emission, which are consequently the dominant species at
nearly all times. Part (c) of this figure plots the temporal evolution of the
spectral widths, clearly showing the narrowing of the paraexciton energy
distribution associated with Bose-Einstein statistics [40]. At the highest
densities, the full width at half maximum of the paraexciton spectrum
is about half that of a Maxwell-Boltzmann distribution at the lattice
temperature. No classical mechanism can explain this result, which does
not rely on any model-dependent fit. The exciton truly is a boson!
Now consider how a condensate would manifest itself in the lumines-
cence spectrum. If the gas density exceeds the critical density given by
(25), we might expect to see a sharp "condensate" peak near E = 0
superimposed on the broader "excited-state spectrum" given by (20).
Unfortunately, as a approaches zero the peak in the excited-state spec-
trum becomes arbitrarily intense and very close to E = 0. Taking the
derivative of N(E) in the quantum regime where a <C 1, one finds that
the peak in the excited-state spectrum occurs at E « |/z|. It appears that
for the ideal Bose gas, the energy distribution will continuously become
sharper and sharper without an abrupt change as the phase boundary is
crossed at a = 0.
Nevertheless even in this idealized case, the presence of condensate
should be readily detectable in the spectrum. Recall that the excited-
t Unlike other data presented below, the sample in this case is cooled by helium gas at
elevated temperature rather than immersed in liquid helium.
 BEC of a Nearly Ideal Gas: Excitons in CujO 303
Photon energy (meV) -ortho
2019 2021 2023 2025 2027

t=16ns $ W K ^ (a)
c
\

/
CD ~\

CD
N

"5
\a V

E
|V"T(OO

t=44 ns | h
r (b)
c
CD

CD

o
E T=32K ^
/
o

2010 2012 2014 2016 2018

Photon energy (meV) —para

-4 0 4 8 12 16 20 24 28 32 36 40 44 48
t(ns)
Fig. 9. (a) The orthoexciton (dashed line) and paraexciton (solid line) energy
distributions during laser-pulse excitation at photon energy of 2.4 eV. The two
spectra are normalized to the same height and shifted in energy for comparison.
The orthoexciton distribution fits a Maxwell-Boltzmann with T=38 K. The
paraexciton spectrum is obscured by orthoexciton luminescence at energies above
2.015 eV. The open circles are a 1.3 meV-broadened Bose-Einstein distribution
with a = 0.07 and T=38 K. (b) The distributions well after the laser pulse.
Both fit a Maxwell-Boltzmann distribution with T=32 K (open circles), (c) The
spectral FWHM of each energy distribution as a function of time (open circles:
orthoexcitons; solid circles: paraexcitons). The 10-ns laser pulse is centered at
£=10 ns. (From Ref. [39].)
304 J. P. Wolfe, J. L. Lin and D. W. Snoke

(c)

Fig. 10. (a) Excited-states Bose-Einstein distribution for a gas at constant po-
tential with a=0. (b) The distribution of (a) convolved with a triangular spectral-
resolution function, (c) The distribution of (a), plus a delta-function peak at
E = 0. (d) The distribution in (c), convolved with the same spectral resolution
function.

state spectrum is integrable, even at \i = 0, and that there is always

some experimental limit to the spectral resolution. This limitation can be
intrinsic to the gas- e.g., from collisional broadening - or simply from
the finite resolution of the spectrometer. These broadening mechanisms
will effectively integrate over the (hypothetical) infinity at E = 0, yielding
a finite excited-state spectrum even when JX = 0. To illustrate this
important point, Fig. 10(a) shows N(E) for a = 0, and Fig. 10(b) plots
the same spectrum convolved with a triangular "slit function", which is
characteristic of a grating spectrometer with equal input and output slits.
If we add to the excited-state spectrum a perfectly sharp delta-function
spike at E = 0 with, say, half of integrated intensity as the excited-
state spectrum, then the slit-convolved spectrum shown in Fig. 10(d) is
obtained. The contribution from the condensate is shaded. This spectrum
is quite distinguishable from the \i = 0 excited-state spectrum, and the
fraction of condensate (here 1/3) can be easily extracted.
 BEC of a Nearly Ideal Gas: Excitons in Cu^O 305
In what follows, we will examine the behavior of the excitonic gas
at high densities and low lattice temperature (Tiatt = 2 K). The system-
atic studies described below were motivated by early results of Hulin
et al. [41], who observed orthoexciton spectra - integrated over time
and space - that could be described by quantum-statistical distributions.
As we shall see, subsequent time- and space-resolved spectra provide
overwhelming evidence for quantum statistics of excitons in CU2O and
uncover an unusual "quantum saturation" effect, still not totally under-
stood. At the highest excitation levels, a rapid expansion of the gas is
observed together with interesting anomalies in the paraexciton spectra,
suggesting superfluidity of the excitonic gas. Most recently, compelling
spectroscopic evidence for BEC of paraexcitons has been obtained for
uniaxially stressed crystals.

7 Quantum Statistics of Excitons at High Gas Densities

Figure 11 shows phonon-assisted spectra of orthoexcitons following a
high-power, 100-ps Ar + laser pulse. The crystal is immersed in a su-
perfluid helium bath at 2 K. The luminescence is detected by photon
counting with 90 ps time resolution. The temperature of the gas at each
sampling time can be ascertained by fitting the high-energy tail of the
spectrum. The decay of the gas temperature determined in this way is
plotted as the open circles in Fig. 12. An interesting discovery is that the
decay rate of the temperature depends upon the initial density of the gas,
which is controlled by changing the spot size of the laser beam. In the
experiment of Fig. 11, the spot size is about 20 jt/m, corresponding to 108
W/cm2 for the 20 nJ pulse. If the spot size is increased to 2 mm, the
power density is (100)2 smaller and the resulting temperature evolution
is plotted as the solid dots. The cooling rate of about 100 ps is consistent
with the acoustic-phonon emission process discussed in Sections 3 and 5.
At high density, the gas retains a high temperature over a much longer
period. Possible sources of the inhibited cooling at high density will be
discussed below.
The spectra of Fig. 11 show clear evidence of quantum statistics. The
dashed curves are fits of Eq. (20) to these high-power spectra. Initially
the spectrum is nearly Maxwellian (compare to Fig. 5), but at later times
it takes on the more sharply peaked characteristic of a Bose gas in the
quantum regime. Indeed, at all times the gas is well described by Bose-
Einstein statistics, and over a broad time interval the chemical potential
assumes a highly quantum value of /z « —0.2/CBT. A value of a « 0.2
306 J. P. Wolfe, J. L. Lin and D. W. Snoke

t-i20ps (a)

unjts)

t«900 ps

•e ^ a—0.1
o

1
~ 10
;

2019 2021 2023 2025 2027 2029

Photon energy (meV)

Fig. 11. The orthoexciton luminescence at three densities plotted semilogarith-

mically, fit to the ideal Bose-Einstein distribution (20), with the values of a and
T shown. The fit distribution in each case is convolved with triangle function to
take into account the finite spectral resolution. (From Ref. [38].)

corresponds to a gas density which is a factor of about three lower than

the critical density for condensation.
Given fi and T, the density of the gas is determined by integrating
N(E)/V using (21). In Fig. 13 the spectroscopically determined [i and n
are plotted as a function of time after the pulse. Part (a) shows that the
rise and fall of n(t) closely follows that of the integrated intensity /tot(0>
implying that the effective volume of the gas is roughly constant over
the time span of the measurement. The rapid, non-exponential decay
of / to t(0 at subnanosecond times implies that some new relaxation or
recombination mechanism occurs at high density.
Most fascinating, during the cooling cycle, the degree of degeneracy
measured by a remains constant for a long period of time (Fig. 13(b)),
despite the large changes in density and temperature. This implies that
 BEC of a Nearly Ideal Gas: Excitons in 307

1000

fi
100-• o High density
°O
0

o O O
<» ° o o o o o
10-- Kl^ T
Low density -—.

T,ott
1 1 1
200 400 600 800 1000

t(ps)
Fig. 12. The effective temperature as a function of time for the case of high
density (n > 1019 cm" 3 ) and low density (n < 1017 cm" 3 ). The solid line in the
low density case is the prediction of the deformation-potential theory for phonon
emission discussed in the text. (From Ref. [38].)

the gas is evolving along a line parallel to the (/i = 0) phase boundary
for BEC, as shown by the open circles in Fig. 14. The effect of the gas
to remain close to the phase boundary over a wide range of densities
and temperatures has been termed "quantum saturation" by Snoke et
al. [42]. This remains one of the major puzzles of this remarkable
quantum gas.
Quantum saturation is even more pronounced in a "long-pulse" (10 ns)
experiment. In this case, the orthoexciton gas is observed with about 1 ns
resolution during and after the excitation pulse. Again, the gas tempera-
ture and chemical potential are found to be well defined throughout the
experiment. As the excitation pulse evolves, the temperature and density
of the gas increase along a \i « —O.l/c^T line to maximum values (at
about the peak of the pulse) and then decay along the same line, as shown
by the solid dots in Fig. 14. In these experiments, the signal levels are
high enough to observe the phonon-assisted spectra of the paraexcitons
simultaneously. At medium pulse intensities, the paraexcitons exhibit a
higher degree of degeneracy, as predicted in Fig. 6. At high pulse inten-
sities, interesting anomalies appear in the paraexciton spectrum which
were attributed to the formation of a condensate.
Clues to the origin of the quantum saturation effect may be extracted
from the observations of an increased orthoexciton decay rate and a
reduced cooling rate (Figs. 12 and 13). Figure 15 explicitly plots the
308 J. P. Wolfe, J. L. Lin and D. W. Snoke

I
•"fit (a)
1019 • s -
Eo

l0 18< <t • •
Q)
Q 1 •Laser intensify

•
m17 (

0.25

0.00
(b)
-0.25 II
-0.50

-0.75

-1.00
200 400 600 800 1000 1200
t(ps)
Fig. 13. The thermodynamic parameters of the orthoexciton gas in the case
shown in Fig. 11. At late times the uncertainty in the chemical potential \i grows
as the temperature lowers (Fig. 12) and the line narrows. The smallest possible
a is plotted as a solid dot, but, within the spectral resolution, the corresponding
spectra at late times are almost compatable with a classical distribution, as
indicated by the large error bars. (From Ref. [38].)

orthoexciton decay rate as a function of density. The linear dependence

indicates a two-body process for the destruction of orthoexcitons. Two
possible mechanisms are as follows. At high densities the probability
increases that the collision of two excitons can induce a non-radiative
decay of one of the excitons. In this "Auger recombination process",
the recombination energy (essentially £ gap — Ex) is given to the remain-
ing electron-hole pair as kinetic energy. The hot particles can always
transfer some of their kinetic energy to the remaining particles in the
gas, effectively heating the gas [43]. Another density-dependent heating
mechanism which might occur is collision-induced conversion of two
orthoexcitons into paraexcitons, which would raise the kinetic energy of
the gas by 24 meV (twice the exchange energy) per event. (See also
the discussion by Nozieres in this volume.) While this spin-flip process
 BEC of a Nearly Ideal Gas: Excitons in CujO 309

100 •

1017 1 0 18 10 19 1020
Density (cm" 3 )
Fig. 14. The quantum saturation of the orthoexciton gas, corresponding to a
temperature rise with increasing density, T oc n2/3. Solid circles are deduced
from fits to spectra during and after long (10-ns) laser pulses. (From Ref. [42].)
Open circles are deduced from fits to spectra following short (100-ps) laser pulses.
(From Ref. [38].) The dashed line gives the critical densities for BEC of the g = 3
orthoexcitons.

and/or Auger heating may be important in explaining the temperature

rise of the gas at high densities, it is difficult to see how they could
completely explain why the gas follows the quantum-statistical boundary
so closely.
Can the rise in temperature simply be due to heating of the lattice?
This explanation seems improbable due to the following argument: A
density of 2 • 1019 cm" 3 implies that the 3 • 1010 excitons (the number of
photons in a 0.01 / J pulse) occupy a volume of 1.5 • 10~9 cm3. If one
assumes a heat capacity of Cu 2 O given by 5.9 • 10" 3 J/kg-K (T/K)3 [44]
and that all of the excess photon energy (0.375 eV) imparted to the
excitons is converted to phonons, then the local temperature rise of
the crystal would be about 20 K - not enough to explain the 90 K
rise in the gas temperature. Also, there is no apparent band shift of
the exciton line (cf. Fig. 2), indicating that T^tt remains less than
15 K. A similar argument applies to the long-pulse case, where both the
number of particles and gas volume are about an order of magnitude
larger.
An interesting issue raised in earlier work [34] takes note of the fact
that at constant average energy, the temperature of an ideal boson gas
310 J. P. Wolfe, J. L. Lin and D. W. Snoke

1018 1019
Density (cm"

Fig. 15. The decay rate of orthoexcitons as a function of density, from the same
data as Fig. 12. (From Ref. [38].)

increases by a factor of two as the gas increases in density from the

classical to the quantum regime, simply due to the shape of the distri-
bution (20). Therefore if the rate of energy loss is constant per particle,
then increasing the gas density automatically increases the temperature.
Later work [45] showed, however, that increased degeneracy enhances
the rate of energy loss by acoustic phonon emission via the (1 + /k)
final-states factor, so that quantum statistics alone will lead to at most a
20% increase in the gas temperature and will not give the "saturation"
effect.
The question of the time scale for BEC, discussed in other articles
in this book, does not apply to the heating effect seen here. As shown
in earlier work based on solution of the Boltzmann equation for two-
body scattering [34, 6], a boson gas in the normal state can evolve
via elastic particle-particle or inelastic particle-phonon scattering to a
highly degenerate distribution within a few scattering times. Once the gas
has a large population in states near zero kinetic energy, the Boltzmann
equation breaks down and the question remains of how long the particles
need to organize into a true delta function. As long as the gas can be
considered weakly interacting, however, a slow process of long-range
organization would not yield a substantial increase in the average energy
of the gas.
 BEC of a Nearly Ideal Gas: Excitons in Cu2O 311
8 Effects of Inhomogeneities in the Exciton Gas

In all of the high-density experiments, the excitation light is absorbed

within about 1/im of the crystal surface. The resulting density and
temperature of the exciton gas are therefore inherently non-uniform. The
spectral fitting, on the other hand, has assumed a single temperature and
chemical potential. The fact that the spectra are so well characterized by
these two parameters implies that at a given time the energy distribution
of the gas is well described by an average temperature and chemical
potential.
To check this conclusion, Ref. [13] analyzed spectra resulting from
several spatial distributions of n and T. Since a full hydrodynamic
model of the density and temperature profile of the exciton gas in steady
state has not appeared [46], these calculations simply assumed "realistic"
distributions of the gas. In one model, the density was chosen to be a
Gaussian, n(z) = nmaxexp(—(z/z 0)2), decreasing with distance, z, from
the surface, and T was assumed to be a constant. For a given nmax and
T, Bose-Einstein spectra corresponding to n(zt) at a series of zt were
generated and summed to predict the overall energy distribution of the
inhomogeneous gas. The composite spectrum yielded a good fit to the
"saturated" experimental spectra when nmax = 0.8nc (or amin = 0.025). In
a second model, a Gaussian form for the density was assumed and the
local temperature was assumed to vary as the 2/3 power of the density, as
observed in the quantum saturation effect. Again a good fit was obtained
yielding amin =0.1 for a typical spectrum.
These calculations may have a bearing on why the orthoexciton chem-
ical potential appears to "saturate" at about -0.1 to -0.2 /c^T, rather
than \x = 0. The value a = 0.1 corresponds to an average density of the
gas which is about two times smaller than nc, which is consistent with a
spatial distribution varying from nmax « n c to zero. It may be that the
quantum saturation occurs when the orthoexciton gas is actually con-
densed in the region of highest density. If so, why would the condensate
fraction remain small, and what causes the gas to warm exactly along
the phase boundary? An interesting suggestion [47], is that when a
condensate is formed, it undergoes superfluid flow to a local minimum
in potential (perhaps associated with defects or the crystal surface). The
resulting increase in density produces an enhanced Auger recombina-
tion or ortho-para conversion, destroying the condensate and heating
the gas; i.e., keeping the gas precisely along the phase boundary! Any
312 J. P. Wolfe, J. L. Lin and D. W. Snoke
process which converts condensate particles into heat has this interesting
property.
It is interesting to note that the constant a implies that the gas follows
a line of constant entropy per particle, an "isentrope". A gas in free
expansion follows an isentrope, and the exciton gas created at the surface
approximates a free expansion when it moves away from the surface. We
note, however, that (1) the orthoexciton saturation is seen during times
when the effective volume of the gas is constant and density is rising or
falling due to particle generation (e.g. the data of Ref. [42], shown in
Fig. 14) or recombination (e.g. Fig. 13(a), following 100 ps pulses) and
(2) the excitons can lose energy at all times to the phonon field, so that
entropy is not strictly conserved.
Finally, we address the low-energy tail observed in the orthoexciton
spectra at high densities, e.g., Fig. 11. A possible source of this lumines-
cence at E < 0 is spectral broadening due to collisions. A convolution of
(20) with a Lorentzian of full width at half maximum of ~ 0.1 meV can
account for this spectral feature. This width corresponds to a collision
time of 10 ps, which is consistent in the classical approximation (26)
with a scattering length on the order of the Bohr radius of the order
of 7 A. Alternatively, a low-energy tail has been interpreted by Haug
and Kranz [31] as arising from interactions with the Bose condensate.
In that work, the experimental data of Hulin et al. [41] were fit to a
spectrum which included a low-energy tail arising from a 5% condensate
fraction, as well as substantial symmetric Lorentzian broadening which
was attributed to instrumental resolution. The fact that the broadening
seen in the low-energy tail is not proportional to density, as expected for
standard collision broadening, but instead increases disproportionately
at late times when the gas appears more degenerate [13], would seem to
support this interpretation. Although the instrumental spectral broaden-
ing in these recent experiments is much less than that assumed by Haug
and Kranz, one could argue that a small fraction of condensate leads to
the low-energy tail even though a sharp peak due to the condensate is
obscured by spatial inhomogeneities (see also Ref. [48]).

9 Enhanced Degeneracy of Orthoexcitons in Stressed Cu2O

How can one coerce the "saturated" orthoexciton gas into crossing the
phase boundary and condensing? One approach is to shift the phase
boundary to lower densities! Eq. (25) and Fig. 6 show that the critical
density for BEC is directly proportional to the ground-state multiplicity g.
 BEC of a Nearly Ideal Gas: Excitons in Cu2O 313

Zero stress [HO] stress

€2
g=3

2500
t = 6.4 ns
2000 cr =1.0 k b a r

T=40 K
(fit) -

-500
2020 2021 2022 2023 2024 2025
Photon energy (meV)-ortho
Fig. 16. Orthoexciton luminescence from Cu2O under [110] stress.

If the orthoexciton triplet can be split by applied stress or magnetic field

such that most of the particles relax to the lowest level, then the ground-
state multiplicity is effectively reduced from 3 to 1. Such a reduction in
the ground-state multiplicity displaces the orthoexciton phase boundary
a factor of three towards lower density, making it coincident with the
paraexciton phase boundary. If the experimental n(T) curve were to
remain the same as in Fig. 14, then the orthoexciton gas would indeed
cross the new phase boundary.
By uniformly stressing the crystal, Lin and Wolfe [49] have indeed been
able to effectively reduce g from 3 to 1. Uniaxial stress applied along
the crystalline [110] direction reduces the crystal symmetry from cubic
to orthorhombic. The energy splitting of the triplet has been measured
by Waters et al. [50] and fit theoretically by Trebin et al [51]; for stress
along the [110] axis the orthoexciton levels shift by approximately 0,
±1.5 meV/kbar, and the paraexciton level does not shift, to first order
[52]. The orthoexciton triplet is split into three levels, as shown in
Fig. 16 for a stress of o = 1.0 kBar = 108 Pa. Intense 10-ns Ar + laser
pulses are used. The upper solid curve in this figure is the orthoexciton
314 J. P. Wolfe, J. L. Lin and D. W. Snoke
spectrum obtained without selecting polarization of the luminescence. If
a linear polarizer is inserted into the detection path, it is possible to
nearly eliminate the luminescence from the upper two levels, obtaining
a spectrum close to that plotted by the open circles (upper trace). In
fact, these open circles are a good fit of (20) to the polarization-selected
spectrum, yielding the OL\ and T shown. By subtracting this theoretical
Bose-Einstein distribution from the upper trace, one obtains the contri-
butions from the upper stress-split levels, shown as the lower solid trace.
These excited states are then fit to (20) with the same T to obtain the a2
and 0C3 shown.
We see immediately that the excitons occupying the three levels are
not in chemical equilibrium with each other. That is, they do not share
the same chemical potential. In particular, the chemical potential for
the uppermost level is only about \i = — OAksT = —0.4 meV, relative to
its zero-kinetic-energy point at 2.0223 eV. Physically, this means that the
conversion time between the triplet levels is longer than the measurement
time (6.4 ns). Nevertheless, it appears from the spectrum that most of the
excitons are in the lowest level. For later sampling times, the luminescence
from the lowest level becomes even more dominant, as the excitons in
the split-off levels relax. Let us now concentrate on the distribution from
this ground level.
A quite remarkable spectrum appears when the gas density is the
highest, near the peak of the excitation pulse. A polarizer is used
to eliminate residual luminescence from the two excited states. The raw
spectrum is plotted both linearly and logarithmically as the solid curves in
Fig. 17 (a) and (b), respectively. The fit to the Bose-Einstein distribution
(20), convolved with the spectrometer slit resolution, is exceptionally
good, yielding a = 0.015 and T = 62 K. For comparison to classical
statistics, a Maxwell-Boltzmann distribution with T = 62 K is plotted
as a dashed line in both graphs.
Still, there is no evidence of a condensate in the orthoexciton spectrum.
The experimental n(T) has apparently been "pushed" to lower density
(or higher temperature) by the new phase boundary. The solid dots
in Fig. 18 are the results of analyzing the time-resolved spectra as a
function of sampling time. We see that the orthoexciton gas in the
stressed crystal displays a high degree of quantum saturation, closely
paralleling the g = 1 phase boundary plotted as the solid line in this
figure. For comparison, the data for the unstressed case is plotted as
the open circles, and the dashed line is the g = 3 phase boundary. In
view of these data, there is little doubt that the saturation effect is indeed
 BEC of a Nearly Ideal Gas: Excitons in CU2O 315
nit:

(a) t=9.6 ns
(b)
,6 2000 ft cr=1.0 kbar

jp
;cence intensity

1500
-

1000
/ - expt.
0 fit (T=62K, a ^0.015)

VI
<D
500

L MB V - MB (T=62K)

e 0 J 1 1 10" 1 ,

2020 2021 2022 2023 2024 2025 2020 2021 2022 2023 2024 2025
Photon energy (meV)
Fig. 17. Solid line: orthoexdton energy distribution in the lowest spin state in
stressed CU2O following intense laser exdtation. (Luminescence from higher spin
states has been subtracted via a polarizer.) Open circles: fit to the ideal-gas
Bose-Einstein distribution, (a) Linear plot, (b) Semilog plot.

100

# stressed
O unstressed

BEC boundary

10 10 •

Orthoexciton density (cm )

Fig. 18. Comparison of the orthoexciton "saturation" in the case of g = 1

(stressed Cu2O) and g = 3 (unstressed CU2O).

associated with the degenerate quantum nature of the gas and is not
solely due to an "accidental" coincidence of classical heating processes
(e.g., Auger or ortho-para conversion) with the BEC phase boundary.
316 J. P. Wolfe, J. L. Lin and D. W. Snoke

Photon energy (meV)-ortho

2021.2 2022.2 2023.2
t=15 ns
600 a=0.3 kbar
//'A.V ortho
para
•3 ;'/ X\
'2 400

200

3 A A K

2012 2013 2014

Photon energy (meV)-para

Fig. 19. The orthoexciton (dashed line) and paraexciton (solid line) energy dis-
tributions during laser-pulse excitation at photon energy of 2.4 eV in a [110]
stressed crystal. The two spectra are normalized to the same height and shifted
in energy for comparison.

10 BEC of Paraexcitons in Stressed Cu2O

Our attention has been fixed primarily on orthoexcitons because they are
so much easier to observe than paraexcitons. As discussed in Section 2
and listed in Table 2, the most intense paraexciton replica is assisted
by the Tj5 phonon and is 500 times less radiatively efficient than the
IY2 orthoexciton line that we have been studying. Nevertheless, when
the signal level is increased by widening the spectral and time resolution
somewhat and averaging over longer periods, the paraexciton lumines-
cence becomes observable.!
Despite the poor visibility of paraexcitons, we expect that their den-
sity will be comparable to that of the orthoexcitons at early times, t <
10 ns. As the orthoexcitons convert down to paraexcitons, the paraex-
citon population should greatly exceed that of the orthoexcitons at low
temperatures. Also, the paraexciton lifetime exceeds that of the orthoex-
t The paraexciton luminescence is readily observable in a ^sec-timescale experiment (e.g.
Section 3 and Ref. [23]) because one integrates the recombination photon flux over
times comparable to the lifetime of the paraexciton. In the pulsed experiments, the gas
density is only comparable to the quantum density for 20 ns or so following the peak of
the laser pulse (at t = 10 ns). The slow radiative rate of the paraexcitons (Table 3) does
not produce many photons in a few nanoseconds.
 BEC of a Nearly Ideal Gas: Excitons in Cu2O 317

a=0.3 kbar
• i
' • • • •
10 J

: / \ para (x500)
• c 0 ortho
O -
O
O
S7 oo(D O
_v v O
O
V V
laser V
V
o
V
o
7 . I . I .

10 20 30 40 50 60
Time (ns)

Fig. 20. The relative intensity of the ortho and para luminescence. Triangles give
the intensity of the laser vs. time, multiplied by an arbitrary constant.

citon by two to three orders of magnitude. Of course, the excitonic gas

will be expanding in the crystal as time progresses, which will increase
the volume occupied by the gas. As usual, we must resort to experiment
to find out if the paraexciton density can exceed the critical density for
BEC before diffusing away.
The simultaneous ortho and paraexciton spectra at a sampling time of
15 ns and an applied stress of a = 0.3 kBar are plotted in Fig. 19. The
two spectra are superimposed such that their E = 0 points coincide.!
The paraexciton spectrum is significantly narrower than the orthoexciton
spectrum and displays little evidence of the high-energy tail associated
with the excited-state spectrum. The orthoexciton spectrum is well fit to
the excited-state spectrum (20) with the parameters a = 0.1 and T =
19 K.
The relative number of ortho and paraexcitons is determined from their
spectrally integrated intensities, as plotted in Fig. 20 for an applied stress
of G = 0.3 kbar. The paraexciton intensity has been multiplied by a factor
of 500 to compensate for the different radiative efficiencies. The intensity
of the laser pulse is also shown. As expected, the ortho and paraexciton
populations are roughly comparable at early times, but near the peak
f For the orthoexciton, the photon energy for E = 0 is accurately determined in fitting the
data to the excited-state spectrum (20). For the paraexciton, we find E = 0 by fitting a
Maxwell-Boltzmann spectrum (13) to the late-time spectra, which can be averaged over
a long time window to enhance the signal-to-noise ratio.
 318 J. P. Wolfe, J. L. Lin and D. W. Stroke

(b)

time (ns) 3 J>^^ _/*

0
10 - (D
o
a, a, y
g
BEC boundary
o ortho
10 15 20 25 30 35 • para
Time (ns) 10
Density (cm )

Fig. 21. (a) Time evolution of the temperature and chemical potential determined
from the orthoexciton spectrum during and after an intense 10-ns laser pulse of
a Cu2O crystal stressed along [110] at 0.3 kbar. (b) Density and temperature for
the ortho and para gases during the same experiment. Labels for the para data
points indicate the elapsed nanoseconds.

of the pulse the number of paraexcitons begins to depart significantly

from that of the orthoexcitons. As the orthoexcitons continue to decay
and down-convert to paraexcitons, the paraexciton population grows and
then remains stable.
In Fig. 21 (a) we plot the time evolution of the temperature and chemical
potential determined from the orthoexciton spectrum. If we make the
reasonable assumption that, over the period of these measurements, the
orthoexciton and paraexciton components share the same temperature
and volume,! then we can obtain a rather accurate estimate of the
paraexciton density. As usual, the orthoexciton densities are determined
by integrating (20) using the parameters in Fig. 21 (a), and the results
are plotted as the open circles in Fig. 21(b). The concurrent paraexciton
density is then determined by multiplying the orthoexciton density by the
ratio of intensities plotted in Fig. 20. The solid dots in Fig. 21(b) show
that the paraexciton density determined in this way crosses the BEC
phase boundary at a time between 6 and 11 ns, which coincides with the
emergence of the narrow paraexciton spectrum.
In Fig. 22 we show an analysis of the paraexciton spectrum in terms

f That the orthoexciton and paraexciton gas occupy nearly the same volume during
the (10 ns) excitation pulse is quite reasonable considering that there are frequent
interparticle collisions to keep them together at high densities, and that the orthoexcitons
convert to form a good fraction of the paraexcitons. An imaging experiment with high
spatial resolution in an unstressed crystal has shown that the ortho and paraexciton
gases coexist spatially at least up to t = 25 ns (see Ref. [13].)
 BEC of a Nearly Ideal Gas: Excitons in CujO 319

Photon e n e r g y r (meV)-ortho
2021.2 2022.2 2023.2
1
i i

t=15 ns

h
600 CJ^O.S kbar

ortho
para
O fit
400 -
\ • (T=19K,a p =0)
S
^N - (A=0.5 meV)
A"

200 -

x
A k K C p ""

p
•

0(

2012 2013 2014

Photon energy (meV)-para

Fig. 22. The data of Fig. 19, fit to the BEC distribution discussed in the text.

of a Bose-Einstein condensate. The simple procedure of assuming a

delta-function at E = 0, convolved with the slit function as illustrated in
Fig. 10, does not yield a satisfactory fit. The extra luminescence at low
energy, however, is well fit by a Gaussian line of with 0.5 meV width,
blue shifted by 0.2 meV, convolved with the instrumental slit function
of 0.3 meV. This is plotted as the triangles in Fig. 22. The distribution
of thermally excited particles is blue shifted by the same amount in
this fit, although a fit with no blue shift of the excited particles is also
possible. The sum of the low-energy component and the a = 0 excited-
state component, with intensity determined by the ratio of ortho and
para intensities in Fig. 20(a), is plotted as the open circles. According
to Fig. 21(b), the condensate has a density which is about a factor of
3 above the phase boundary at 25-30 K and the density remains at
about 1019 cm" 3 over the measurement time, coinciding with a relatively
constant blue shift of the low-energy component.
We are compelled to conclude that this new component is the long-
sought excitonic condensate. The simplest explanation of the blue shift
of the condensate is that it represents the repulsive interactions of the
paraexcitons [48]. Indeed, the orthoexciton spectrum does not exhibit
this blue shift, and this is consistent with the predictions of Moskalenko
et al. [53] based on many-body theory that takes into account exciton-
320 J. P. Wolfe, J. L. Lin and D. W. Snoke
exciton interactions, that the para level should have much greater blue
shift at high density than the ortho level. Alternatively, a non-zero drift
velocity of the condensate would give a blue shift. The reason for the
finite width (0.5 meV) of the peak is unclear. A condensate with long-
range order is predicted to have a true delta-function distribution in the
ground state, even in the presence of interactions. The finite width may
be the result of spatial inhomogeneity which gives a distribution of blue
shifts of the gas, or it may indicate that a "quasicondensate" [47] has
appeared but long-range order of the condensate phase has not.
Why does the applied stress assist in the condensation of the
paraexcitons? Several factors come into play. First, by lowering the
orthoexciton multiplicity, stress causes the entire two-component gas to
have a higher level of degeneracy. Second, the applied stress can have
a beneficial effect on the confinement of the excitons. As mentioned
in Section 3, the diffusivity of paraexcitons is reduced as the symmetry
of the crystal is lowered. The gas volume can be estimated by simply
dividing the integrated intensity of the orthoexcitons (the open circles in
Fig. 20) by the spectroscopically determined density at each time. The
volume of the gas obtained this way increases slowly over the entire mea-
surement interval. In contrast, the volume obtained by Snoke et al. [13]
from spatial imaging in an unstressed crystal shows a considerable ex-
pansion during this same interval. Third, an increased exciton-phonon
coupling under crystal stress allows more efficient cooling of the gas. The
relative importance of these factors - or some not yet identified - may
be determined by using magnetic field, rather than stress, to reduce the
orthoexciton multiplicity.

11 Confinement of Excitons in a Parabolic Well

In all of the above, we have discussed BEC assuming a uniform potential,
that is, "particles in a box". An interesting alternative situation is an
ideal gas of bosons trapped in a parabolic potential of the form,
V(r) = Ar\ (27)
where r is the distance from the center of the well and A is the well pa-
rameter. As discussed previously [54] and by others in this volume, this
kind of potential allows the interesting possibility of a spatial condensa-
tion as a telltale of BEC of both atomic and excitonic gases. In other
words, BEC of particles trapped in a three-dimensional parabolic well is
predicted to appear as a very narrow spatial distribution of particles at
 BEC of a Nearly Ideal Gas: Excitons in CuiO 321
the exact energy minimum of the well, not only as a zero-kinetic-energy
peak in the energy distribution.
In the low density, classical case, the chemical potential is the sum of
the "internal" chemical potential (see the discussion of Eq. (22)),

flint = -kBTln(nQ/n), (28)

and an "external" chemical potential given by fiext = Ar2. In equilibrium,

Mint + Mext = constant over the well, leading to the spatial distribution of
the density,
n(r) = n(0)e-Ar2/kBT, (29)

a Gaussian profile. This expression is based on Maxwell-Boltzmann

statistics. It turns out that the spatial distribution of non-condensed
particles exhibiting Bose-Einstein statistics is not much different from
this form [54].
The single-particle states of the harmonic-oscillator potential given
above comprise an energy ladder, Es = (s + 3/2)hco9 with a> = (lA/m)1^1
and s = 1,2,... . Each level has a degeneracy of (s + l)(s 4- 2)/2, implying
that the density of states for s » 1 is proportional to E2. The full
expression in this limit is
(30)
i&
where g is the spin multiplicity of the particle. The chemical potential
is determined by the condition N(E) = f D(E)f(E)dE, where f(E) is
the distribution function (18). Setting \i = 0 in this equation yields the
critical condition for BEC,

Nc = 1202g(kBT/ha))\ (31)

For a given temperature, when the total number of particles in the well
exceeds iVc, the excess particles will theoretically occupy the ground state.
Herein lies the qualitative departure from the uniform-potential case:
since the ground state of the harmonic oscillator is a narrow Gaussian
wavefunction centered at the bottom of the well, BEC corresponds to a
spatial condensation rather than a fc-space condensation.
The spatial extent of the ground state for an ideal gas, calculated
simply as the wavefunction of a single particle in the ground state of
the well, is so small that the condition (6) for the excitons to remain
bosons, na3 <C 1, would not hold if a significant fraction of the gas
were condensed in this state. In this case, one cannot ignore the effect of
322 J. P. Wolfe, J. L. Lin and D. W. Snoke

Laser

(a) (b)
Fig. 23. (a) Equipotential curves for a Hertzian contact stress, (b) Drift of excitons
into a three-dimensional parabolic well created in a CU2O crystal with Hertzian
contact stress. This photoluminescence image of paraexcitons is obtained for
continuous excitation at the left surface with an Ar + laser. (From Ref. [54].)

interactions among the particles in the condensate. As shown by Hijmans

et al. [55], in the limit of weak interaction, assuming a condensate density
much greater than non-condensate density near the center of the well,
the spatial extent of the condensate is fixed by the relation
no(r)£/ + F(r) = ,u, (32)
where U is the interaction vertex (proportional to the scattering length),
V(r) is the external potential and // is the chemical potential determined
by ri(rc)U+V(rc) = \i, where ri is the excited-states density. The radius rc
of the region in which the condensate density is non-zero is determined
by the condition that the integral over the condensate density profile
over this region is equal to the number of condensate particles. The
volume of the condensate therefore depends on the condensate fraction.
Although the volume of the condensate taking into account interactions
is much larger than the single-particle (non-interacting) ground state, so
that the condition na3 < 1 holds, it should still be much smaller than
the excited-states spatial distribution and therefore should give a clear
telltale for the presence of a condensate, if its lifetime remains sufficiently
long.
A parabolic potential well can be created for excitons by non-uniformly
stressing the crystal. Figure 23(a) shows a schematic of the equipotential
 EEC of a Nearly Ideal Gas: Excitons in 323

o
15

<7>

Fig. 24. Spatial profiles of the photoluminescence from paraexcitons in the po-
tential well, during cw excitation with a dye laser. Here r is the distance from
the center of the well. (From Ref. [54].)

curves caused by a Hertzian-contact geometry. A shear-stress maximum

- or potential-energy minimum - is formed just below the contact area of
the rounded stress rod. Part (b) of this figure shows the drift of excitons
in CU2O towards the stress maximum from the surface where they are
created. The bright spot is a large collection of excitons in the parabolic
well. This "strain-confinement" technique has previously been exploited
for studying excitonic phases in Ge and Si [56].
Figure 24 shows spatial profiles of the photoluminescence from (long-
lived) paraexcitons in the potential well. The dots are fits to (29),
adjusted for stress dependence of the paraexciton radiative efficiency. The
fit temperatures are somewhat higher than the helium bath temperature,
indicating some heating in the well. According to (29), the widths of these
324 J. P. Wolfe, J. L. Lin and D. W. Snoke

i(D
Q

0.5 1.0 2.0 5.0 10 20 50 100

Absorbed laser power Pabs(mW)
Fig. 25. Estimated density of paraexcitons as a function of absorbed laser power,
when exciting the well directly with a dye laser. Here r is the distance from the
centre of well. (From Ref. [54])

profiles are proportional to T 1 / 2 and the effective volume is given by

n (2. \V1
(33)

Clearly, these excitons at relatively low density are behaving as a nearly

ideal classical gas. The effective volume of the gas is considerably larger
than in the "inertial-confinement" case discussed previously, so it is more
difficult to reach the condition for BEC.| However, in the strain well
the paraexcitons exhibit a much cooler temperature than in the inertial-
confinement case.
When the well is excited directly with a dye laser, by tuning the laser
wavelength exactly to the energy-gap minimum in the well, the gas density
in the strain well can be obtained simply by measuring the number of
absorbed photons. This estimate also requires the measured lifetime, T,
and effective volume, Ves. The dependence of the gas density, n, on cw
absorbed-power is plotted in Fig. 25. This P^ dependence is expected

The condensation condition for the two cases can be compared by noting that nc
 BEC of a Nearly Ideal Gas: Excitons in Cu2O 325
for Auger-limited lifetimes, where

^ = -Bn2 + (const)Pabs = 0 (34)

at
under steady-state excitation. For the gas temperature in this experiment,
the calculated critical density, nc « NC/VQQ, equals about 1017 cm"3,
which is one to two orders of magnitude above the highest experimental
density. Trauernicht et al. [37] discuss the difficulties in reaching the
phase boundary due to Auger recombination and heating of the gas.
One of the indicated solutions is to lower the bath temperature in order
to reduce the critical density. While quantum degeneracy has not yet been
reached, the strain-well approach still looks promising for the creation
and study of a quantum gas in a confining potential.

12 Conclusions
Despite their short lifetimes, free excitons in a high-purity crystal of
CU2O form a nearly ideal gas of particles. The excitonic gas obeys the
kinetic theory of gases, it moves via drift or diffusion through the crystal
long after the generating laser is off, it can be trapped in a potential
minimum, and on timescales longer than a nanosecond it exhibits a
Maxwell-Boltzmann or Bose-Einstein energy distribution characteristic
of an internal thermal equilibrium. Because the exciton lifetime is much
longer than the particle scattering time, the laws of equilibrium thermo-
dynamics apply. Indeed, the finite lifetime of the excitons has a distinct
benefit: due to the radiative decay of excitons, it is possible to observe
their energy distribution directly.
The search for BEC of excitons has progressed stepwise. First, it has
been proven that excitons behave statistically as bosons- namely, the
energy distribution narrows with increasing density, in contrast to the
broadening of the fermion distribution with increasing density. Second,
the BEC phase boundary has been mapped over more than an order
of magnitude of density variation, via the "quantum saturation" effect.
Third, it has been shown that the density of excitons can exceed the
critical density for condensation at a given temperature, under certain
conditions. Finally, an extremely narrow energy distribution (A£ <C fc^T)
of paraexcitons in the ground state has been observed under conditions
of n > nc.
A full understanding of these results awaits a complete theoretical
treatment of the weakly interacting exciton gas, including the hydro-
326 J. P. Wolfe, J. L. Lin and D. W. Snoke
dynamics of the expanding boson gas and the energy distribution of a
finite-lifetime, two-component gas with phonon and Auger interactions.
Perhaps the most significant lack in present theory is an exact calculation
or an accurate measurement of the exciton-exciton interaction potential:
the proper form can not be simply extrapolated from that of hydrogen,
due to the important effects of the light mass of the hole and the electron-
hole exchange. This interaction potential goes into the calculation of the
biexciton binding energy, the ortho-para splitting and interconversion,
and exciton-exciton scattering.
Nevertheless one of the beauties of this system is the fact that the laws
for an ideal Bose gas work so well over such a large range of density and
temperature, attested by the fact that the spatial diffusion data and the
spectra are well fit using simple Maxwell-Boltzmann or Bose-Einstein
distributions. All of this analysis supports the conclusion at this point
that paraexcitons do indeed undergo Bose-Einstein condensation.

Acknowledgments. We acknowledge the essential collaboration of A.

Mysysrowicz. The Illinois work is supported by National Foundation
Grant DMR 92-07458. One of us (D.W.S.) thanks the Aerospace Spon-
sored Research program for support for this work.

References
[I] S. Nikitine, J.B. Grun, and M. Sieskind, J. Phys. Chem. Solids 17,
292 (1961).
[2] M. O'Keefe, J. Chem. Phys. 39, 1789 (1963).
[3] A. Gotzene and C. Schwab, Solid State Comm. 18, 1565 (1976).
[4] J.W. Hodby et al, J. Phys. C 9, 1429 (1976).
[5] P.Y. Yu and Y.R. Shen, Phys. Rev. Lett. 32, 939 (1974); Phys. Rev.
B 12, 1377 (1975).
[6] D.W. Snoke, D. Braun, and M. Cardona, Phys. Rev. B 44, 2991
(1991).
[7] V.T. Agekyan, Phys. Stat. Sol. (a) 43, 11 (1977).
[8] RD. Bloch and C. Schwab, Phys. Rev. Lett. 41, 514 (1978).
[9] E.g. A.L. Fetter and J.D. Walecka, Quantum Theory of Many-Particle
Systems (McGraw-Hill, New York, 1971), p. 259.
[10] PL. Gourley and J.P. Wolfe, Phys. Rev. B 25, 6338 (1982).
[II] M.M. Beg and S.M. Shapiro, Phys. Rev. B 13, 1728 (1976).
[12] Y. Petroff, P.Y. Yu, Y.R. Shen, Phys. Rev. Lett. 29, 1558 (1972);
Phys. Rev. B 12, 2488 (1975).
 BEC of a Nearly Ideal Gas: Excitons in Cu2O 327
[13] D.W. Snoke, J.R Wolfe and A. Mysyrowicz, Phys. Rev. B 41 11171
(1990).
[14] K. Reimann and K. Syassen, Phys. Rev. B 39, 11113 (1989).
[15] The measured relative absorption ratio of 50 in Ref. [8] is reduced
by a factor of three to take into account the orthoexciton spin
multiplicity.
[16] J.L. Birman, Sol. State Comm. 13, 1189 (1978).
[17] F. Bassani and M. Rovere, Solid State Comm. 19, 887 (1976).
[18] D.W. Snoke, A J . Shields, and M. Cardona, Phys. Rev. B 45, 11693
(1992).
[19] D.W. Snoke, J.R Wolfe, and D.P. Trauernicht, Phys. Rev. B 41, 5266
(1990).
[20] J.S. Weiner et al, Solid State Comm. 46, 105 (1983).
[21] A. Mysyrowicz, D. Hulin, and A. Antonetti, Phys. Rev. Lett. 43,
1123 (1979).
[22] K. Reimann, J. Lum. 40/41, 475 (1988); A.V. Akimov and A.A.
Kaplyanskii, Proc. 5th Int. Conf. on Phonon Scattering, A.C. Ander-
son and J.R Wolfe, eds. (Springer, Berlin, 1986), p. 449.
[23] D.P. Trauernicht, J.R Wolfe, and A. Mysyrowicz, Phys. Rev. Lett.
52, 855 (1984).
[24] D.P. Trauernicht and J.P. Wolfe, Phys. Rev. B 33, 8506 (1986).
[25] E.g. D.I. Blokhintsev, Quantum Mechanics (Dordrecht-Holland, Am-
sterdam, 1964), p. 493ff.
[26] E.g. F.K. Richtmyer, E.H. Kennard, and John N. Cooper, Introduc-
tion to Modern Physics, 6th edition (McGraw-Hill, New York, 1969)
p. 571.
[27] For a review see J.C. Hensel, T.G. Phillips, and G.A. Thomas, Solid
State Physics 32, H. Ehrenreich, F. Seitz, and D. Turnbull, eds.
(Academic, New York, 1977).
[28] T.J. Greytak and D. Kleppner, in New Trends in Atomic Physics, G.
Grynberg and R. Stora, eds., (North-Holland, Amsterdam, 1984).
[29] R.C. Casella, J. Phys. Chem. Solids 24, 19 (1963).
[30] A. Mysyrowicz, D. Hulin, and E. Hanamura, in Ultrafast Phenomena
VII, (Springer, Berlin, 1990).
[31] H. Haug and H.H. Kranz, Z. Phys. B 53, 153 (1983).
[32] A. Quattropani and J.J. Forney, II Nuovo Cimento 39, 569 (1977).
[33] A.I. Bobrysheva and S.A. Mosklenko, Phys. Stat. Sol. (b) 119, 141
(1983); S.A. Moskalenko et al, J. Phys. C 18, 989 (1985); Phys. Stat.
Sol. (b) 129, 657 (1985); A.I. Bobtysheva et al, Phys. Stat. Sol. (b),
167, 625 (1991).
328 J. P. Wolfe, J. L. Lin and D. W. Snoke
[34] D.W. Snoke and J.R Wolfe, Phys. Rev. B 39, 4030 (1988).
[35] D.W. Snoke, W.W. Riihle, Y.-C. Lu, and E. Bauser, Phys. Rev. B 45,
10979 (1992).
[36] Semiconductors Probed by Ultrafast Laser Spectroscopy (Academic
Press, New York, 1984).
[37] D.R Trauernicht, A. Mysyrowicz, and J.R Wolfe, Phys. Rev. B 28,
3590 (1983); D.R Trauernicht, J.R Wolfe, and A. Mysyrowocz, Phys.
Rev. B 34, 2561 (1986).
[38] D.W. Snoke and J.R Wolfe, Phys. Rev. B 42, 7876 (1990).
[39] D.W. Snoke, J.-L. Lin, and J.R Wolfe, Phys. Rev. B 43, 1226 (1991).
[40] Bose narrowing was shown for excitons in stressed Ge by V.B. Tim-
ofeev, V.D. Kulakovskii, and I.V. Kukushkin, Physica B+C 117/118,
327 (1983).
[41] D. Hulin, A. Mysyrowicz, and C. Benoit a la Guillaume, Phys. Rev.
Lett. 45, 1970 (1980).
[42] D.W. Snoke, J.R Wolfe, and A. Mysyrowicz, Phys. Rev. Lett. 59, 827
(1987).
[43] A. Mysyrowicz, D. Hulin, and C. Benoit a la Guillaume, J. Lum.
24/25, 629 (1981).
[44] L.V. Gregor, J. Phys. Chem. 66, 1645 (1962).
[45] D.W. Snoke, University of Illinois at Urbana-Champaign Ph.D.
thesis, 1990.
[46] For a hydrodynamic model of excitons in CU2O without Auger
recombination and phonon emission, see B. Link and G. Baym,
Phys. Rev. Lett. 69, 2959 (1992).
[47] See the contribution of Yu. Kagan in this volume.
[48] H. Shi, G. Verechaka and A. Griffin, Phys. Rev. B 50, 1119 (1994)
have derived an expression for the decay luminescence spectrum
for an excitonic BEC in the weakly interacting gas approximation,
incorporating the low-energy tail and the blue shift of the spectrum.
[49] J.-L. Lin and J.R Wolfe, Phys. Rev. Lett. 71, 1223 (1993).
[50] R.G. Waters et a/, Phys. Rev. B 21, 1665 (1980).
[51] H.-R. Trebin, H.Z. Cummins, and J.L. Birman, Phys. Rev. B 23, 597
(1981).
[52] A. Mysyrowicz, D.R Trauernicht, J.R Wolfe, and H.-R. Trebin, Phys.
Rev. B 27, 2562 (1983).
[53] S.A. Moskalenko et a/., Phys. Stat. Sol. (b) 129, 657 (1985).
[54] See D.R Trauernicht, J.R Wolfe and A. Mysyrowicz, Phys. Rev. B
34, 2561 (1986).
 BEC of a Nearly Ideal Gas: Excitons in CU2O 329
[55] T.W. Hijmans, Yu. Kagan, G.V. Shlyapnikov, and J.T.M. Walraven,
to be published; see also D.A. Huse and E.D. Siggia, J. Low Temp.
Phys. 46, 137 (1982) and V.V. Goldman, I. Silvera, and AJ. Leggett,
Phys Rev. B 24, 2870 (1981).
[56] J.P. Wolfe and C D . Jeffries, in Electron-Hole Droplets in Semi-
conductors, C D . Jeffries and L.V. Keldysh, eds. (North-Holland,
Amsterdam, 1987).
 14
Bose-Einstein Condensation of Excitonic
Particles in Semiconductors
A. Mysyrowicz
LOA-ENSTA
Ecole Poly technique
F-91120 Palaiseau
France

Abstract
The case of excitons as candidates for Bose-Einstein condensation is dis-
cussed, and experimental results in CuCl and CU2O are presented. In CuCl,
spectral analysis of the luminescence from biexcitons as a function of their
density reveals a gradual evolution from classical statistics towards a quan-
tum degenerate regime. The appearance of a sharp emission line below a
critical temperature and above a critical density is attributed to the pres-
ence of a laser-induced Bose-Einstein condensate of excitonic molecules.
This interpretation is supported by pump-probe experiments which show
that additional particles injected in the presence of a biexciton condensate
are drawn into it.
In Cu^O, free exciton luminescence spectral analysis of ortho- and
paraexcitons reveals a gradual evolution from a classical to a Bose quantum
degenerate regime with increasing particle densities. Orthoexciton densities
close to the critical density for condensation are obtained at high inco-
herent excitation. Under similar pumping, paraexciton densities exceeding
the critical value are inferred from luminescence intensity ratios. Anoma-
lous transport properties of paraexcitons, such as ballistic propagation over
macroscopic distances and formation of soliton-like excitonic packets are
discussed as evidence for excitonic superfluidity.

1 Introduction
1.1 Excitons
The lowest electronically excited state of a non-metallic crystal corre-
sponds to the promotion of one electron from the top of the highest fully

330
 BEC of Excitonic Particles in Semiconductors 331
occupied valence band to the bottom of the next empty conduction band.
A correct evaluation of the required energy must include the Coulomb
correlation between the promoted electron and all other electrons left
behind in the valence band. A very fruitful conceptual approach (see the
article by Wolfe et al. in this volume) consists in treating this problem
by the equivalent point of view of the Coulomb interaction between a
negatively charged particle, the electron in the conduction band and a
positively charged particle, the electron vacancy or positively charged
hole left in the valence band. The problem reduces to the same two-
particle Schrodinger equation as for the hydrogen or positronium atom,

h2 2 h2 e2 1
Ve V
~2m~r ~ ^~ * ~ ~ ~ I °= E®' W
It is well-known that discrete solutions exist corresponding to bound
electron-hole pair states. These bound states are called excitons and
obey the hydrogenic relation

where EB = mre4/2h2£l = e2/2ax£o = h2/2mra2 is the exciton binding

energy, a\ = soh2/mre2 is the Bohr radius and mr = memh/(me + nth)
is the reduced effective mass, £o is the static dielectric constant of the
medium, np is the principal quantum number, me and m^ are the effective
electron and hold masses, and Mx = me + m^ is the translational mass
of the exciton [1]. Thus, in this point of view, supply of one quantum of
excitation energy to the crystal corresponds to the creation of one exciton,
a quasi-particle with the same internal energy structure as the hydrogen
atom or the positronium atom. For optically active excitons, this energy
supply can occur through absorption of a photon which occurs typically
over a penetration depth less than 1/im.
As a function of photon energy, the absorption leads to a hydrogenic
series of absorption lines below the energy band gap converging towards
an ionization continuum of free electron-hole pairs at the band gap en-
ergy. It is worth noting the difference between this absorption mechanism
and that occurring in atomic spectroscopy. In crystals the absorption of
light corresponds to the creation of excitons and not to electronic transi-
tions between terms of a series (this last process can occur with excitons
but the corresponding photon energy lies in the infrared region, in view
of the small exciton binding energy « meV) [1]. It is also important to
332 A. Mysyrowicz
keep in mind that optically inactive excitons exist in all semiconductors.
The number of possible excitonic states which can be formed with a
given set of valence and conduction bands is obtained from group-theory
arguments; there are at least two: a paraexciton with opposite electron
and hole spins which is optically inactive and an orthoexciton with par-
allel electron and hole spins which can be optically active. Even if they
do not contribute to the optical response of the material near the band
gap, paraexcitons are nevertheless formed when the crystal is excited in
the ionization continuum. Indeed, after creation of unbound electron-
hole pairs, fast relaxation of the excess energy of the free carriers occurs
on a subpicosecond time-scale through emission of phonons followed
by association of the free electrons and holes into the various varieties
of excitons. Of prime concern in the context of BEC is the fact that
optically active excitons are generally not the species lowest in energy
and therefore are not the best candidates for Bose condensation.
Besides an internal energy structure similar to that of hydrogen or
positronium atoms, excitons possess another attribute of real particles,
namely the ability to move freely inside the crystal. In this motion, the
center-of-mass of the electron-hole pair propagates through the crystal
carrying an electronic energy of the order of the band gap. Under normal
circumstances (low particle densities), the exciton transport is diffusive
with a diffusive coefficient being determined in high purity samples by
scattering with lattice phonons.
Local exciton densities reaching 1019 cm"3 can be achieved easily by
optical pumping of the crystal with modest pulsed lasers, since the excited
volume of the crystal can be as small as 10~10 cm3. For an exciton of
lifetime 10~9s, an optical pulse of the same duration with an energy
content of a few nanojoules should be sufficient in principle.

1.2 Biexcitons
It is possible to pursue the analogy between excitons and positronium or
hydrogen atoms one step further and consider the chemistry occurring at
higher densities when excitons start to interact. Like hydrogen atoms, ex-
citons can form molecules called biexcitons. The formation of a biexciton
due to the coupling of a pair of excitons and its analogy to a hydrogen
molecule has been experimentally well established. An excitonic molecule
in its ground state is similar to a hydrogen paramolecule with two holes
of opposite spins sharing two electrons, also with opposite spins. It can
 BEC of Excitonic Particles in Semiconductors 333
also propagate through the sample, carrying a total energy:

where 2E^ = 2(Eg — Bx) is the internal energy of two noninteracting exci-
tons at zero wave vectors, E®x is the biexciton energy at zero wave vector,
Dxx is the biexciton binding energy (the energy required to dissociate it
into two free excitons), and Kxx and Mxx are the center-of-mass wave
vector and mass of the biexciton, respectively.
The binding energy of the biexciton depends on the ratio of effective
masses, a = me/mh [2]. The limiting case a = 0 corresponds to the hydro-
gen molecule, with a well-established experimental binding energy. No
experimental verification of the other limit, a = 1 (molecule of positro-
nium) exists yet. However, variational calculations predict a biexciton
binding energy smaller by one order of magnitude.
Due to the small mass of excitons, typically of the order of the free
electron or less, quantum fluctuations are very important in determining
the internal structure of biexcitons. The biexciton zero-point energy is
comparable to the molecular binding energy itself. As a consequence,
vibrational levels are not stable and at most one excited level exists in
crystals with simple band structures (the orthobiexciton in the first rota-
tional level). This is in sharp contrast to the rich manifold of rotational
and vibrational levels of hydrogen molecules. This simplicity of the
molecular excitation spectrum is favorable for Bose condensation since
it prevents formation of a molecular liquid phase, as will be discussed
shortly.
Biexcitons cannot be created by optical means directly into their ground
state through a one-photon absorption process, since the required photon
energy is almost twice the band gap energy and therefore corresponds
to a region of strong absorption in the exciton ionization continuum.
It is possible to create biexcitons directly from the crystal ground state
by two-photon absorption. As shown by Hanamura [3], the two-photon
absorption cross section corresponding to the direct creation of a para-
biexciton has giant values, of the order /? « 1 cm- MW" 1 , or 5-6 orders
of magnitude more than for typical two-photon interband transitions in
crystals. Such huge values reflect the fact that a two-photon process
is well adapted to the creation of a biexciton, because two electrons
separated by a mean distance equal to the biexciton radius must be
simultaneously promoted from the valence to the conduction band. Con-
sequently, densities of biexcitons reaching 1019 cm"3 can be obtained
334 A. Mysyrowicz
with a laser of intensity less than 1 MW/cm2 [4]. In the context of
Bose condensation, it is important to note that the biexciton population
is nearly spatially uniform, since the two-photon attenuation of a laser
decreases as a function of crystal thickness z as

<4>
instead of the usual exponential attenuation from Beer-Lambert law
in a one-photon process. If two independent laser beams are used
to generate biexcitons directly, the wavevector of the created particles
can be controlled between 2Ko and 0 by changing the angle between
the incoming beam, where Ko is the magnitude of the incident photon
wavevector.
Biexcitons can also be created indirectly through free carrier (or ex-
citon) generation followed by biexciton formation [5]. In a steady state
regime, the ratio of biexciton to exciton populations Nxx/Nx is dictated
by the values of their respective chemical potentials. Biexciton densities
reaching 1020 cm"3 have been obtained through indirect band-to-band
pumping [6]. Note however that under indirect creation process, the
biexciton density is spatially nonuniform since it is proportional to the
local exciton density, itself spatially nonuniform. Also, the excess pump
photon energy leads to local heating of the exciton and biexciton gas.

2 Bose-Einstein Condensation of Excitonic Particles

Blatt et a\. [7] and, independently, Moskalenko [8] were the first to
point out that excitons consisting of an even number of fermions have
an integer spin value and as such are candidates for Bose condensation.
Theories of excitonic condensation (and superfluidity) have been given
by Keldysh and Kozlov [9], Haug and Hanamura [10] and Comte and
Nozieres [11]. In the theories of Keldysh and Nozieres and coworkers,
the starting point consists of a system of degenerate electrons and holes
of arbitrary density, the Bose model of ideal particles being just the
low density limit. At high carrier densities, the treatment is formally
similar to the BCS theory of superconductivity except that collective
pairing of electrons and holes occurs via Coulomb attraction (instead
of the phonon-mediated Cooper pairing of electrons in metals), making
the system unstable against the formation of a dielectric superfluid at
sufficiently low temperatures. Thus, it appears that excitonic superfluidity
is not restricted by the condition nxa3x <C 1 but should be possible over
 BEC of Excitonic Particles in Semiconductors 335
a wide range of electron-hole densities at low temperature. It must be
stressed, however, that occurrence of excitonic superfluidity in the high
density limit necessitates a direct band gap, because collective pairing
requires approximately equal but opposite electron and hole momenta.
Haug and Hanamura [10] have treated the problem in the spirit of the
Bogoliubov model, in the limit nxa3x <C 1, in which excitons are viewed
as approximate point bosons. Several assumptions are implicit in this
model. A first assumption is a predominantly repulsive exciton-exciton
potential. Otherwise, excitons aggregate at increasing densities into more
complex entities, the same problem which prevents Bose condensation
of a normal gas of hydrogen molecules. In crystals with indirect band
gap structure (with the extreme of valence and conduction band at
different points of the Brillouin zone), it is energetically favorable for
the system to separate into a low density exciton phase coexisting with
regions of a high density phase of a two-component electron-hole liquid
[12]. An increase of injected excitons in this case merely increases
the size of the electron-hole liquid drops, without changing the free
exciton density. Therefore, unless some artifice is used to prevent exciton
liquefaction, Bose condensation in the dilute limit is not possible in
indirect gap materials. In direct gap materials with simple band structure,
where exciton liquefaction into drops is not expected [12], the direct and
exchange Coulomb interactions leads to an overall attraction and in
principle to biexciton formation. However, a more detailed examination
is required when the ratio of electron to hole mass o « 1. In crystals with
large electron-hole exchange energy, the biexciton can become unstable
against dissociation in two paraexcitons [13].
On the other hand, in the other limit o <C 1, the formation of ex-
citonic molecules is unavoidable and is experimentally well established.
Biexcitons, with four fermi constituents, also fall into the category of
bosons and therefore can undergo Bose condensation themselves. It is
therefore necessary to examine the biexciton-biexciton potential in order
to determine whether the van der Waals attraction between molecules
can lead to a first order phase transition into a dielectric liquid before
BEC occurs in the gas phase. According to the quantum theory of corre-
sponding states [14], this should not be the case if the ratio of zero-point
kinetic energy to potential energy, expressed by the reduced quantum
parameter r\ = h2/Mef2, is larger than 3.5, assuming a Lennard-Jones
interaction potential U(r) = 4e[(//r) 1 2 -(//r) 6 ]. Here e is the depth of the
potential well and / the collision diameter. In actual excitonic systems n
is much larger than 3.5 [12], because of the small particle mass (leading
336 A. Mysyrowicz
to high zero kinetic energy) and therefore the system should remain a
gas even at T = 0. In that case, the results obtained for a weakly
interacting exciton gas can be applied to the gas of excitonic molecules
with little modification and Bose-Einstein condensation is expected for
the biexcitons.
Another assumption in the theory is to treat excitons as particles
of infinite lifetime. This condition is approximately fulfilled in most
semiconductors to the extent that the exciton decay time T > TC, where
TC is the mean collision time between particles. Typically TC is of the
order 10~ns or less for densities above 1016 cm"3. On the other hand,
the time zac required to establish true equilibrium with respect to the
lattice temperature is in general found to be very long, T <C rac, although
a quasiequilibrium is brought about in a time comparable with the
exciton-acoustic phonon relaxation time xac « 10~~9 — 10~10 s, even in
the absence of interparticle collisions. An additional problem is the
time required for an initially nonequilibrium system to undergo Bose
condensation. The understanding of the dynamics of Bose condensation
is the subject of current theoretical research (see the articles by Kagan
and Stoof in this book).
Finally, a last assumption underlying the nearly ideal Bose gas treat-
ment of excitons is a negligible interaction between excitons and the
radiation field. If the exciton-photon coupling is strong, then the true
elementary electronic excitations of the crystals are better described in
terms of polaritons [15]. By inspection of Fig. 1, it is immediately ap-
parent that no accumulation of polaritons near K = 0 is possible, and
therefore Bose condensation of dipole active excitons is impossible [16].
Excitonic molecules, on the other hand, do not suffer from this compli-
cation even if they are formed from dipole-active excitons, because they
have a quadrupole moment and consequently do not interact in first
order with the photon field.

3 Experimental Results in CuCl

This compound is a typical direct-band-gap semiconductor with dipole-
allowed interband transitions [17]. The crystal structure is zincblende
(cubic with no inversion symmetry) and belongs to the point group
Td. This type of band structure with no degeneracy of the valence
and conduction bands near K = 0 is the simplest possible. The value
of the forbidden gap is 3.43 eV at 4.2 K. The np = 1 exciton states
formed with the upper valence band of symmetry F7 and lower F6
 BEC of Excitonic Particles in Semiconductors 337
conduction band are split by electron-hole exchange interaction into a
dipole-allowed orthoexciton of symmetry Ts at 3.208 eV responsible for
a strong absorption line and a lower lying optically inactive paraexciton
F2 state at 3.202 eV. The binding energy of the exciton in CuCl, EB =
190 meV, is very large when compared to most semiconductors, and the
exciton radius is correspondingly small, a# = 0.7 nm. The electron-to-
hole mass ratio a is small, a = 0.2, favoring biexciton formation. The
internal energy of the biexciton is 6.372 eV, so that the energy required
to dissociate it into two free paraexcitons is Dxx = 29 meV. The biexciton
effective mass is known with high precision, Mxx = 5.29mo [18]. The
critical density for BEC estimated from the ideal gas model is 10 17 cm" 3
at 2 K. Because of its large binding energy, the biexciton is stable up to
at least T « 100 K. Also the fluid of biexcitons has been observed to be
stable up to densities n « 10 20 cm" 3 , at which point a sharp Mott-like
transition into a conducting plasma state is observed [6].

3.1 Biexciton Luminescence

An elementary step in the radiative decay process of a thermal biexciton
is as follows. A biexciton disintegrates into a free exciton and a photon.
The energy and momentum conservation conditions governing this decay
process are
Exx(Kxx) = hco + Ex(Kx), (5)

where

(6)
: 2M?
and
(7)
where q is the emitted photon wave vector.
The biexciton luminescence lineshape can be expressed as

I(hco) = C J d3Kxx J d3Kx\Wij(Kxx,Kx)\2f(Kxx)

-S(KXX -Kx- q)d[hco - (Exx - Ex)]9 (8)
where Wtj is the matrix element for the transition between the biexciton
and exciton branch, f(Kxx) is the biexciton statistical distribution, C is
a constant which includes the collection efficiency to the detector, and
the delta functions express the energy and momentum conservation. The
338 A. Mysyrowicz
dependence of Wy on the direction of polarization vectors of the photons
and excitons and the angle between emitted photon and restituted exciton
can be found in Ref. [19].
In a crystal with dipole-allowed excitons, such as CuCl, it is necessary
to take into account the coupling of the final exciton state with the
radiation field. This affects the luminescence spectrum in two ways [15].
First, it leads to a splitting between longitudinal and transverse excitons
which are the final states in the transition. The splitting is given by the
relation

fi
where the parameter ft characterizes the oscillator strength of the ex-
citonic transition. The longitudinal and uncoupled transverse exciton
energies are

ET(K)=ET(0)

Therefore, one expects two luminescence bands, terminating in the trans-

verse and longitudinal branches, called MT and ML as shown in Fig. 1.
Secondly, due to its coupling to the radiating field, the transverse ex-
citon branch acquires a strong dispersion around Ko, where Ko is the
wavevector of a photon of the same energy as shown in Fig. 1. In this
region of strong dispersion, the transverse exciton becomes a polariton
(a mixed exciton-photon mode). In a classical gas most biexcitons have
wavevectors Kxx > Ko so that the polariton effect can be neglected. In
this case, (8) reduces to the simple form

7 ( M oc (E°xx -E°x- Hco)^2e-^-E"-h^kBT. (11)

Thus, the emission from a classical biexciton gas takes the form of
two bands with inverted Maxwell-Boltzmann shape. Photons of lower
energies correspond to decaying biexcitons with larger kinetic energy,
and vice versa (see Fig. 1.) An example of a biexciton luminescence
spectrum obtained in a thin crystalline CuCl film of high quality under
weak pumping conditions such that the density of biexcitons nxx <C ncr,
together with a best fit assuming a classical distribution of biexcitons, is
shown in the upper curve of Fig. 2 [19].
In a quantum degenerate regime, the polariton effect becomes impor-
tant since most biexcitons have small wavevectors and the final state in
 BEC of Excitonic Particles in Semiconductors 339

B-E distribution E Maxwell distribution

-exciton

EM/2
K

Fig. 1. Excitation diagram of a crystal with dipole-active excitons. The upper

curve represents the dispersion of the biexciton. The lower two curves give the
dispersion of the longitudinal exciton and the transverse polariton.

the transition correspond therefore to the region of strong dispersion of

the polariton. The polariton dispersion is given by

h2c2K2 4TT p
= 1 + 1-E2/E2T(KY (12)

An example of a high density biexciton luminescence spectrum recorded

in a high quality 4.5 fim thin crystalline film of CuCl held at T = 25 K
is shown in the lower curve of Fig. 2. In order to reduce heating of
the film, the excess energy imparted to the biexciton gas is minimized
by using two-photon excitation tuned to the biexciton resonance. This
excitation scheme has the additional advantage of creating a spatially
uniform population of biexcitons in the sample, thereby avoiding difficul-
ties associated with particle migration. A large sample surface of radius
« 100 — 500 /im is excited and only the emission from the central part is
collected through a small pinhole of radius 10 /mi put in direct contact
340 A. Mysyrowicz

I i 1
CuCI Excitation for K = 0

TL = 25K

^t>^V y >o |I = - oo

J
tensi

.g
TL = 5K
O
<D
o IT H
</> XT ^a P A. T =25
umi ne

p d ««
/

I I i
3.170 3.165 3.160
Photon energy (eV)

Fig. 2. Luminescence spectrum of CuCI under weak (upper) and strong two-
photon excitation. The open circles are calculated lineshapes assuming a classical
and a Bose quantum distribution with n — 0.

with the sample surface, so that lateral spatial nonuniformities become

negligible [19]. A best fit to the measured spectrum is obtained by using
a value \i = 0 for the chemical potential in the Bose-Einstein function
1
/(£) = (13)
e(E-n)/kBT _ I

for the particle distribution, indicating that the critical density for BEC
is reached. Additional measurements for intermediate pump intensities
show a gradual evolution between these two limiting cases. Also, mea-
surements as a function of sample temperature show a disappearance of
quantum degeneracy in the biexciton population above T « 40 K, irre-
spective of the pump intensity, in good agreement with the predictions
from the ideal Bose gas [19].
 BEC of Excitonic Particles in Semiconductors 341

Photon energy (5-meV intervals)

Fig. 3. Biexciton luminescence with single beam excitation for several temper-
atures of the sample. The spectra are shifted so that the sharp line NT lines
coincide. The shift is due to the change of band gap energy with T.

Above a critical input intensity a sharp line, denoted NT, grows on the
high energy side of the Mr band if a single-beam, resonant two-photon
excitation is used. This line shows remarkable properties, such as a large
spatial anisotropy, even though the crystal has cubic symmetry. It is
intense in the backward detection geometry (with respect to the incident
beam), but very weak in the forward geometry [19]. Further, the strength
of the line shows a marked dependence upon sample temperature (see
Fig. 3), with a threshold temperature for appearance varying with input
excitation in the manner shown in Fig. 4. The appearance of iVj has been
interpreted as evidence for the occurrence of Bose-Einstein condensation
in the biexciton gas [19], with the condensation taking place at the
wavevector imposed by the incident excitation. Since the radiative decay
342 A. Mysyrowicz

CuCl
2.8 |im film
20K

i
%
8
£

0.025 In(x1500)

3.170 3.165 3.160

Proton energy (eV)

Fig. 4. Biexciton luminescence for a range of incident pump laser intensities.

of the condensate at Kxx = 2KQ leaves a final exciton with a wavevector

between KQ and 3Xo, depending on the direction of the emitted photon,
the polariton aspect of the final exciton is now crucial. Inspection of the
polariton dispersion curve of CuCl shows that in the backward detection
geometry, the remaining transverse polariton at 3X 0 is essentially exciton-
like, with the complementary emitted photon lying on the high energy
side of the Mr band. In the opposite limiting case (forward detection
geometry), the final polariton is photon-like; in fact, the whole decay
process is better described in this latter case as a two-photon decay, with
both emitted photon-like polaritons having the same energy hco = Exx/2.
This explains the spatial anisotropy of the condensate emission, directly
reflecting the macroscopic occupation of a biexciton state with finite
wavevector along a preferential direction.
 BEC of Excitonic Particles in Semiconductors 343

1 1 i i i i

Iol CuCI 2.9 um film

25K

k ^J
| \

\
\
X

probe
pump only
Xi

J
Uo/2

M V
\
>v
pump only
x2
^^(no pump)

I \ probe
\
^^^ ^ _ (witn pump )
j ^ \ probe
V (with pump)
^S. x32
i i
3.175 3.170 3.165 3.175 3.170 3.165
Photon energy (eV)

Fig. 5. Biexciton luminescence for excitation and detection conditions described

in the text. Traces labeled probe (no pump) and probe (with pump) are obtained
with synchronous differential detection.

3.2 Pump-Probe Experiments

The presence of a Bose condensate of biexcitons can be verified by pump
and probe experiments. Two weak, counterpropagating probe beams,
each of which is detuned by a small amount E = 2 meV to either side
of the biexciton resonance, are used to inject a small probe density of
biexcitons through the simultaneous absorption of one photon from each
of the probe beams. The resulting molecular luminescence of Fig. 5
(left-hand side, middle curve) shows the expected inverted Maxwell-
Boltzmann shape indicative of a low density gas. In a second step, a
strong pump beam induces a condensate at 2Kn- The corresponding
luminescence spectrum displays the appearance of the narrow, spatially
anisotropic line due to the luminescence from the condensate (Fig. 5, left-
hand side, upper curve). If the probe biexcitons are now injected in the
presence of the intense pump beam, the recombination radiation from
probe particles only (as detected with differential techniques) is substan-
tially modified (Fig. 5, left-hand side, lower curve). The observation of a
344 A. Mysyrowicz
sharp emission at the position of Nj and the simultaneous disappearance
of the broader Mj band is indicative of a large compression of the phase
volume occupied by the probe-excited biexcitons [19]. This attraction in
momentum space is expected if probe particles are added to the system in
the presence of a condensate of excitonic molecules. It has been verified
that this attraction effect only occurs if the crystal temperature is kept
below 40 K, in accordance with the evaluation of the critical density for
a density of biexcitons reaching at most 1020 cm"3.

33 Phase Conjugation Experiments

By definition, a conjugate mirror transforms an incoming optical wave

£(r, t) = 1 [eo(x, y)^'-Ko*+«*> + CBCJ (14)

into a reflected wave Er(r,t) with the temporal oscillation unchanged but
a reversed spatial phase

Er(r, t) = 1 [em0(x, yyM+Ko.-oo) + cc^ (15)

This amounts to an operation of time inversion performed on the incident

wave, leaving the spatial part unchanged, as can be seen immediately by
writing the complete expression for the waves explicitly.

Er(r, t) = 1 [em0{x, yyM+Ko.-«>o) + ^ y)e-*»t+K*-to)]

= E(r,-t) (16)
Therefore, the reflected wave Er(r, t) behaves as if it retraces the path
of the incoming wave back in time, as in a movie played backwards.
This intriguing property leads to automatic compensation, after reflec-
tion, of distortions of the phase front experienced by an incident wave
on its way to the conjugate mirror. It is possible to take advantage
of this property to perform spectroscopy in the space domain instead
of the frequency domain. With a small aperture one can discriminate
a coherent wave of well-defined K-vector reflected by conjugate mir-
ror from other radiations of the same frequency but different K -vectors
originating from the same mirror (see Fig. 6). A true phase conjugate
mirror is obtained by inducing a coherent, spatially uniform oscillation
at frequency co. The medium acts as a two-photon amplifier, ampli-
fying an incident beam (co,Ki). In nonlinear optics, the macroscopic
oscillation at 2co is induced by two counterpropagating beams of the
 BEC of Excitonic Particles in Semiconductors 345

50% mirror

to detector | /

conjugate mirror
pinhole "™ ™"

Fig. 6. Schematic diagram of a phase conjugate mirror allowing spatial discrim-

ination of the emission from the mirror.

same intensity and frequency co. However, a Bose condensate of biex-

citons at K = 0 can also provide a coherent oscillation at 2co and
consequently fulfil the same function. Therefore by detecting a phase
conjugate signal (co, — Kt) upon irradiation of a crystal with an incident
probe beam (co,K,), one obtains evidence for the presence of a biexciton
condensate.
This method has been recently implemented in CuCl [20]. A new
aspect of the experiment is the pumping scheme used to obtain a Bose-
Einstein condensation of biexcitons. In the experiment, described in
more detail by Hasuo et al. in this volume, an incoherent pump source is
used to create the biexciton population indirectly, via exciton generation
followed by molecular pairing. An increase of phase conjugate signal is
observed, indicative of an increased coherent oscillation at 2co. This is at
first sight surprising, since injection of random particles should normally
lead to an effective collision-induced dephasing of a prepared coherent
state. This opposite trend is attributed to condensation of part of the
biexciton population. However, analysis of the conjugate reflectance
indicates a small condensate fraction, of the order of a few percent at
best, despite the fact that the estimate local density of biexcitons should
reach values comparable to those obtained with resonant pumping in
the luminescence experiments described above. In the opinion of the
author, the origin of the observed saturation of the condensed fraction
at highest pump intensities results from two detrimental factors. First,
the indirect pumping scheme delivers excess energy per absorbed photon,
leading to local heating of the excitonic fluid mixture. Second, in contrast
to a resonant two-photon excitation scheme with homogeneous volume
excitation, the density of biexcitons is spatially nonuniform, being initially
confined to the proximity of the sample surface. Fluid expansion driven
346 A. Mysyrowicz
by the density gradient is expected to occur, reducing the number of
particles collapsing into a condensate at K = 0.
In summary, several experiments show indication that a Bose conden-
sation of biexcitons can be achieved in CuCl. The condensate is stable
even at finite K vectors, implying that a superfluid motion of a spa-
tially uniform biexciton fluid takes place at sufficiently low temperatures
T < 40 K.

4 Experimental Results in Cu2O

This compound is well known for the variety of its excitonic features
which provide a model case for the theory of Wannier excitons [21].
The crystal structure is cubic, with inversion symmetry. Consequently,
parity conservation rules apply in the optical transitions. The band
structure of Q12O consists of a non-degenerate upper valence band
(symmetry P*) and a lower conduction band (F^) with extrema located
at the center of the Brillouin zone at K = 0 and separated by an
energy gap of 2.175 eV at 2 K. The n = 1 exciton is split, because of
electron-hole exchange interaction, into a lower lying singly degenerate
Tj paraexciton (Xp = 2.022 eV at 2 K) and a triply degenerate Tj
orthoexciton (Xo = 2.034 eV at 2 K) [22]. As in CuCl, the exciton
binding is large, EB — 150 meV, and the corresponding exciton Bohr
radius is small, ax = 0.7 nm. The large electron-hole exchange energy
Exc = 12 meV and nearly equal electron-to-hole effective mass ratio
a = me/mh = 0.73 conspire to prevent biexciton formation [13], so that
free paraexcitons remain the particles with lowest energy per electron-
hole pair even at high exciton densities. Because of the positive parity
of both valence and conduction bands, ortho- and paraexcitons do
not interact, in the dipole approximation, with the radiation field. As
a consequence, the complication introduced by the polariton effect is
avoided and both species acquire a long radiative lifetime.
The orthoexciton decay time To is limited at low temperatures T <
10 K by ortho-paraexciton down conversion to a value TO « y~\ where
y = 1085 is the ortho-para conversion rate [23]. A much longer decay
time is measured at higher temperatures, TO = 1.1 us at 70 K [23]. At
this temperature the populations of orthoexcitons and paraexcitons are
in thermal equilibrium, so that the effective decay time for both species
is governed by the relation 1/T = I/TO + 1/TP. On the other hand,
it is known from absorption experiments that the ratio of transition
probabilities of ortho- and paraexcitons from the crystal ground state
 BEC of Excitonic Particles in Semiconductors 347
differs by a factor of 500 [24] so that the effective decay rate becomes
T « To and consequently TP ~ 500T. Therefore, one deduces a paraexciton
radiative lifetime TP of 0.85 ms at 70 K. This value, unusually large for
excitons in a direct gap semiconductor, is due to the combined effect of
parity and spin flip selection rules. The paraexciton lifetime measured at
2 K was found to be 13 jus [23]. Note that in these experiments xv was
underestimated, since the decay was registered from paraexcitons created
near the sample front surface and therefore did not take into account
migration of particles out of the observation region. Recently, a long
exciton decay time could be observed even at room temperature [25],
confirming the stability of excitons in CU2O.
The mass of the orthoexciton is 2.7 mo [26]; the paraexciton mass is
assumed to be the same. From the ideal Bose model, one estimates a
critical density for Bose condensation nc < 1017 cm"3 at T = 2 K. In
the absence of migration, this density should be attained locally with
a focussed cw optical pump source of mW power tuned to a region of
strong absorption in the crystal.

4.1 Exciton Luminescence

Guided by these considerations, initial attempts to observe quantum
statistics focussed on the spectral analysis of the paraexciton luminescence
using cw optical pumping of the crystal [23]. The only possible radiative
channel for paraexcitons is through simultaneous emission of a photon
and a momentum- and parity-conserving phonon r^ 5 of energy 10.8 meV,
hereafter denoted Xp — r^T5. Since the participating optical phonon has no
dispersion of its energy versus wave vector at zone center [26], and since
the transition matrix element for the decay process is also independent of
K, the spectral shape of this luminescence line directly yields the statistical
distribution of excitons present in the crystal. Indications of a deviation
from classical statistics could be observed in Ref. [23]. However, the
paraexciton density extracted from lineshape analysis was lower by one
order of magnitude than expected from an evaluation of the absorbed
photon density. This difference was attributed to exciton migration out
of the excited volume, confirmed by later transport measurements.
Study of exciton statistics through spectral analysis of the orthoexciton
luminescence Xo — F^2, where the participating phonon IY2 has an energy
of 14 meV, is much easier to perform since the radiative probability is
larger by a factor of 500. Gradual deviation from a Maxwell-Boltzmann
towards a Bose lineshape become apparent in the orthoexciton emission
348 A. Mysyrowicz

Fig. 7. Luminescence spectrum of the orthoexciton in CU2O under cw excitation

showing gradual evolution from classical to Bose distribution. The inset shows a
spectrum obtained with pulsed excitation (after Ref. [27]).

if the cw excitation rate is increased keeping all other experimental con-

ditions identical [27]. (See Fig. 7.) The increase of quantum degeneracy
is accompanied by an increase of effective temperature associated with
the gas of particles. Under very intense excitation using a pulsed dye
laser (peak intensity /o ^ 107 W/cm2) it is possible to reach situations
where the chemical potential of the gas is very close to zero (see inset of
Fig. 7) [27]. Time-resolved spectra [28] are discussed in this volume by
Wolfe, Lin and Snoke. Excellent fits to a Bose-Einstein distribution are
observed with values of the chemical potential close to zero, indicating
 BEC of Excitonic Particles in Semiconductors 349
that the critical density for orthoexcitons is reached. The rise of exciton
gas temperature with increasing particle density prevents the gas from
Bose condensing, although the critical density is reached asymptotically
at Teff = 10 K.
Time-resolved spectra of the paraexcitons recorded under similar con-
ditions show that the degree of degeneracy is higher than for orthoex-
citons, implying that the paraexciton density is higher. The ratio of
ortho- to paraexciton population can be deduced from the ratio of the
respective luminescence intensities. It could be shown in this way that
the paraexciton density in the sample is higher than nc [28].

4.2 Exciton Transport in

Perhaps the most interesting aspect of an excitonic Bose condensate is
the possibility of observing superfluidity, a dragfree motion of part of the
excitonic fluid through the sample. As pointed out by Gergel et ah [30],
this should translate into a change of the penetration distance inside the
crystal of an exciton fluid initially located close to the surface. Instead
of a diffusive motion, with a penetration depth U = y/Dr where D is the
diffusion coefficient and x the exciton lifetime, a ballistic propagation over
a distance reaching /s « vcrT, where vcr is the critical Landau velocity,
should be observed.
Information on excitonic transport properties at the early stages fol-
lowing exciton formation can be obtained from time- and space-resolved
luminescence studies [29]. A small part of the sample near the front
surface is imaged on the front slit of a monochromator. The band pass
of the monocromator is adjusted so as to select the emission line from
the species to be examined. Using pulsed excitation and time-resolved
detection of the projected sample image as a function of distance from
the excitation spot provides the sought information. Side view detec-
tion is convenient to detect the expansion of the excitonic cloud at the
early stages towards the interior of the sample, following front surface
excitation.
Results shown in Fig. 8 show a ballistic propagation at early times with
a speed dependent on initial density. At the highest densities supersonic
speeds are detected. Link and Baym [31] have performed a theoretical
analysis of the expansion rate and have concluded that these results
indicate a dragfree hydrodynamic flow, consistent with the idea that
excitonic superfluidity is involved.
Time- and space-resolved detection of excitons at much higher pen-
350 A. Mysyrowicz

i i

240 Full power /

D = 600 cm2/sec

0 20 40 60 80 100
t(ns)

Fig. 8. Expansion of the orthoexciton cloud towards the interior of the crystal,
as measured by time- and space-resolved luminescence, for three different initial
exciton densities, where A is the distance of half maximum of the luminescence
intensity from the surface of the crystal. The dashed line corresponds to the
expected behaviour of a diffusive motion with a diffusion constant as shown.

etration depths inside the crystal are obtained by the exciton-mediated

photovoltaic effect [32], in which an electric current is generated in the
absence of an external applied voltage at a semiconductor-metal contact
when excitons reach the interface. The external current results from
the dissociation of excitons into free electrons and holes at the inter-
face because of the built-in potential difference due to the presence of
a Schottky barrier. This method provides a local detector with excellent
spatial resolution since the active dissociation layer is less than 1 fim
thick. It also has good time resolution if the electrodes at the contact are
impedance-adapted for fast detection.
Application of this detection method has revealed a very unusual
excitonic transport regime in Cu2O [33]. At sufficiently high particle
densities and low temperatures, optically inactive paraexcitons initially
created with randomfe-vectorsnear the front surface of a single crystal
 BEC of Excitonic Particles in Semiconductors 351

0.3 0.7 1.1 1.5

Time (jusec)

Fig. 9. Photovoltaic signal obtained in a Cu2O sample of thickness d = 2.5 mm

held at 1.95 K. Delay between optical pump pulses is S t = 125 ns. Both pump
pulse intensities are « 100 KW/cm 2. The delayed traces correspond to (a) the
signal obtained at the sample back surface with the pump pulse alone, (b) the
probe pulse alone, (c) by sequential illumination of the sample. Trace (d) is the
algebraic sum of pump and probe induced signals [(a) + (b)]. (From Ref. [33].)

were observed to propagate ballistically through the sample in the form of

a packet of limited size over unusually large distances, approaching 1 cm.
This anomalous propagation at constant speed is highly temperature
sensitive and takes place under excitation conditions corresponding to
initial supercritical densities near the front surface. This anomalous
transport has been attributed to excitonic superfluidity [33], and is further
discussed by Fortin et al. in this volume.
Evidence for interaction between two ballistic packets, has been ob-
tained very recently [34]. As shown in Fig. 9, the exciton-mediated photo-
voltaic signals obtained in a crystal excited with two successive short
optical pulses of equal intensities (curve c) is larger than the algebraic
sum of the signals from the individual pulses (curve d). Results obtained
[34] with two optical pulses of unequal intensities and larger time sep-
aration are shown in Fig. 10. When both pulses illuminate the sample
sequentially, a second excitonic wavepacket appears instead of the weak
diffusive response observed in the presence of the second pulse alone.
352 A. Mysyrowicz

-0.1 0.7 1.1 1.5

Time (//sec)
Fig. 10. Photovoltaic signal of a C112O sample of thickness d = 2.5 mm held at
2K. Trace (a) is obtained with pump pulse alone. Trace (b) probe pulse alone.
Trace (c) pump and probe pulse in sequence with St — 278 ns. Trace (d) algebraic
difference between traces (c) and (a). The pump intensity is of the order 100
kW/cm2. The prompt signals at t = 0 and t = 0.278^s arefiducialsfor the pump
and probe optical pulses.

Formation of this second packet is attributed to quantum attraction of

thermal excitons left in the trail of the first packet towards the second
propagating packet. However, this interpretation is only suggestive at the
present time.
A theoretical model starting from first principles describing the motion
of a spatially nonuniform Bose condensate amid a gas of thermal particles
has been very recently developed by Hanamura [35]. The equation of
motion of the condensate is derived to second order in perturbation
theory by taking into account the interaction of condensed excitons with
normal excitons as well as lattice phonons. A stationary solution for
the envelope of the superfluid macroscopic wavefunction is given by the
symmetric solitary wave of intensity

\R-vst\
<t>2 = |0 o | 2 sech
Ro
 BEC of Excitonic Particles in Semiconductors 353
Comparison of the experimental results with the prediction of the model
reveals excellent agreement [34]. One interesting aspect of the prop-
agation of such a soliton pulse is its velocity, vs = 2.98 x 105 cm/s,
which is significantly smaller than the velocity of sound in the crystal,
v\ = 4.5 x 105 cm/s. Further investigations are needed to understand
more fully the formation and dynamics of the spatially non-uniform
condensate on which our interpretation rests.

5 Conclusions
There is growing evidence that excitonic particles in semiconductors can
exhibit features expected in a weakly interacting Bose gas. Luminescence
and pump/test experiments show features which are explained in a
natural way by the ideal, or nearly ideal, Bose gas model. Anomalous
transport of excitons initially confined close to the crystal surface is
observed. This ballistic transport over remarkably long distances suggests
that a superfluid phase of excitons is involved.

References
[I] Typical values of as and EB range from a few meV in Si or Ge to
a fraction of one eV in alkali or copper halides with corresponding
Bohr radius ranging from several tens of nm to less than one nm.
[2] O. Akimoto and E. Hanamura, Phys. Soc. Japan 33, 1537 (1972).
[3] E. Hanamura, Solid State Comm. 12, 951 (1973); A.L. Ivanov and
H. Haug Phys. Rev. B 48, 1490 (1993).
[4] G.M. Gale and A. Mysyrowicz, Phys. Rev. Lett. 32, 727 (1974).
[5] N. Nagasawa, T. Mita and M. Ueta, J. Phys. Soc. Japan 41, 929
(1976).
[6] D. Hulin, A. Mysyrowicz and A. Antonetti, J. Luminescence 30, 290
(1985).
[7] I.M. Blatt, K.W. Boer and W. Brandt, Phys. Rev. 126 1691 (1962).
[8] S.A. Moskalenko, Soviet Physics Solid State 4, 199, (1962).
[9] L.V. Keldysh and A.N. Kozlov, JETP 27 521, (1968). L.V. Keldysh,
Y.V. Kopaev, Soviet Physics Solid State 6, 2219 (1965).
[10] E. Hanamura and H. Haug, Physics Reports C33, 209 (1979).
[II] C. Comte and P. Nozieres, J. Physique (Paris) 43, 1069 (1982).
[12] L.V. Keldysh in Proc. 9th Int. Conf. on Physics of Semiconductors
(Moscow 1968) p. 1303; see also T.M. Rice, J. Haensel, T. Phillips,
G.A. Thomas, in Solid State Physics 32, 1 (1977).
354 A. Mysyrowicz
[13] F. Bassani and M. Rovere, Solid State Comm. 19, 887 (1976).
[14] L.H. Nosanow, J. de Physique 41 C7-1 (1980).
[15] J.J. Hopfield, Phys. Rev. 112, 1555 (1958).
[16] R.S. Knox, Theory of Excitons (Academic Press, New York, 1963).
[17] For a review see A. Goldmann, Phys. Stat. Sol. B 81 9 (1977).
[18] For a review of the literature on biexcitons in CuCl, see, for instance,
C. Klingshirn and H. Haug, Physics Reports 70 315 (1981).
[19] L.L. Chase, N. Peyghambarian, G. Grynberg and A. Mysyrowicz,
Phys. Rev. Lett. 42, 1231 (1979); N. Peyghambarian, L.L. Chase and
A. Mysyrowicz, Phys. Rev. B 27 2325 (1983).
[20] M. Hasuo, N. Nagasawa, T. Itoh, and A. Mysyrowicz, Phys. Rev.
Lett. 70, 1303 (1993).
[21] For a review of excitonic properties of CU2O see, for instance, S.
Nikitine in Optical Properties of Solids, S. Nudelman and S.S. Mitra,
eds. (Plenum, New York, 1969).
[22] LI. Kreingold and V.L. Makarov, Soviet Physics Solid State 15, 890
(1973).
[23] A. Mysyrowicz, D. Hulin and A. Antonetti, Phys. Rev. Lett. 43,
1123, 1275(E) (1979).
[24] P.D. Bloch and C. Schwab, Phys. Rev. Lett. 41, 514 (1978).
[25] D.W. Snoke, A.J. Shields and M. Cardona, Phys. Rev. B 45, 11693
(1992).
[26] P. Yu and Y.R. Shen, Phys. Rev. Lett. 32, 939 (1974).
[27] D. Hulin, A. Mysyrowicz and C. Benoit a la Guillaume, Phys. Rev.
Lett. 45, 1970 (1980).
[28] D. Snoke, J.P. Wolfe and A. Mysyrowicz, Phys. Rev. Lett. 59, 827
(1987).
[29] D. Snoke, J.P. Wolfe and A. Mysyrowicz, Phys. Rev. B 41, 11171
(1990). See also article by Wolfe et al. in this volume.
[30] V.A. Gergel, R.F. Kazarinov and R.A. Suris, JETP 27, 159 (1968).
[31] B. Link and G. Baym, Phys. Rev. Lett. 69, 2959 (1992).
[32] E. Tselepis, E. Fortin and A. Mysyrowicz, Phys. Rev. Lett. 59, 2107
(1987).
[33] E. Fortin, S. Fafard and A. Mysyrowicz, Phys. Rev. Lett. 70, 3951
(1993).
[34] A. Mysyrowicz, E. Fortin, E. Benson, S. Fafard and E. Hanamura,
to be published; M. Inoue, E. Hanamura, J. Phys. Soc. Japan 41,
771 (1976).
[35] E. Hanamura, Solid State Comm. 91, 889 (1994).
 15
Crossover from BCS Theory to
Bose-Einstein Condensation
Mohit Randeria
Argonne National Laboratory
MSD 223, 9700 S. Cass Avenue
Argonne IL 60439
USA

Abstract
This article gives a detailed review of our current theoretical understanding
of the crossover from cooperative Cooper pairing to independent bound-
state formation and Bose-Einstein condensation in Fermi systems with in-
creasing attractive interactions.

1 Introduction
There are two well-known paradigms for understanding the phenomena
of superconductivity and superfluidity:

(I) BCS theory [1], in which the normal state is a degenerate Fermi liquid
that undergoes a pairing instability at a temperature Tc < e^, the
degeneracy scale. The formation of Cooper pairs and their condensa-
tion (macroscopic occupation of a single quantum state) both occur
simultaneously at the transition temperature Tc.
(II) Bose-Einstein condensation (BEC) of bosons at a Tc of the order of
their degeneracy temperature. At a fundamental level these bosons
(for example, 4He or excitons) are invariably composite objects made
up of an even number of fermions. The composite particles form at
some very high temperature scale of the order of their dissociation
temperature TdiSSoc> and these "pre-formed" bosons then condense at
the BEC Tc < Tdissoc.
In most cases of experimental interest the system under consideration
clearly falls into one category or the other. For instance, 3 He is a
Fermi superfluid described by (I) whereas 4He is a Bose superfluid (II).
Essentially all cases of superconductivity in metals that we understand

355
356 M. Randeria
reasonably well are much closer to (I) than to (II), as evidenced by
the existence of fermionic quasiparticles above Tc. Nevertheless, as the
rest of this review will show, it is of great interest to consider models
which interpolate between these two extremes: the weak attraction, high
density limit is described by BCS theory while the strong coupling,
low density limit consists of tightly bound bosons. One outcome of
such a study would be a deeper understanding of the phenomena of
superconductivity, even for systems which are clearly closer to one of the
limits. In addition, there are experimental systems which may be in a
regime intermediate between these two extreme limiting cases, and then
one is forced to study the crossover from collective Cooper pairing to the
formation and condensation of independent bosons.
In the remainder of this section we will first give a very quick review
of the historical development of ideas in this subject, then discuss the
reasons for the recent resurgence of interest in this problem, and conclude
with an outline of the rest of the paper. Readers who would like a preview
of the main results of this article may wish to look at the summary in
the concluding section.

/./ History
The problem of BCS versus Bose-Einstein condensation is an old one.
Some of the earliest attempts in the pre-BCS era to theoretically under-
stand superconductivity in metals were in terms of Bose-Einstein con-
densation, an approach which was perhaps taken farthest by Schafroth,
Blatt and Butler [2], Then came the striking success of the Bardeen-
Cooper-Schreiffer (BCS) theory [1] and superconductivity in metals was
finally understood as a pairing instability of a Fermi liquid state. In
the immediate aftermath of BCS the differences with BEC were empha-
sized : that the Cooper pairs, which were highly overlapping in real space,
should not be thought of as composite bosons, and that the correlations
in the BCS condensate were best described in terms of "momentum space
pairing" rather than "real space pairing" in a highly degenerate Fermi
system.
There are, of course, many similarities in the behavior of Fermi and
Bose superfluids insofar as their macroscopic, coherent properties are
concerned. Off-diagonal long range order (ODLRO) [3] in an appropriate
density matrix is a unifying concept in the study of these two types of
superfluids: the macroscopic occupation of a single quantum state forms
the basis for understanding all bulk superfluid systems. Despite these
 Crossover from BCS Theory to BEC 357
similarities between the BCS condensate and the Bose condensate, there
are equally obvious differences at the microscopic level. In particular, the
normal states above Tc are completely different.
Perhaps the first discussion of the possibility of a crossover from
a BCS state to BEC as a function of some parameter, in this case
decreasing carrier density, was by Eagles [4] in the context of a theory
of superconductivity in low carrier concentration systems such as SrTiO3
doped with Zr. This remarkable paper has gone virtually unrecognized
and many results obtained in subsequent papers were, in fact, anticipated
by Eagles.
In a seminal paper, Leggett [5] studied a dilute gas of fermions at
T = 0 with an attractive interaction and showed within a variational
approach that there was a smooth crossover from a BCS ground state
with Cooper pairs overlapping in space to a condensate of tightly bound
diatomic molecules. One of his main motivations was to ask to what
extent one might be able to describe Cooper pairs in superfluid 3He as
giant diatomic molecules. (See also [6-8].)
Some years later this question was taken up by Nozieres and Schmitt-
Rink [9] motivated by the problem of exciton condensation [10] where
it might, in fact, be possible experimentally to go from one limit to
the other by varying the density of carriers. These authors extended
the previous analysis to lattice models and, more importantly, to finite
temperatures. Using a diagrammatic formulation they showed that within
their approximation, the transition temperature Tc evolves smoothly as
a function of the attractive coupling from the BCS to the Bose limit.
The papers of Leggett and of Nozieres and Schmitt-Rink have had
a major influence on all the subsequent work in this field, and we shall
return to a more detailed discussion of these later in this review.

1.2 Current Interest

There has been a resurgence of interest in superconductivity after the
discovery of the copper-oxide based materials with high transition tem-
peratures [11]. It is now clear that in the copper-oxide systems, the size
of the pairs, as estimated from the Ginzburg-Landau coherence length
(since these systems are believed to be in the clean limit), is only a few
times the lattice spacing [12]. This is in marked contrast to materials well
described by conventional BCS theory where the pair size greatly exceeds
the lattice spacing, or the average distance between carriers. In this sense,
the new superconductors are different from BCS systems, and the pairing
358 M. Randeria
in these systems is likely to be in a regime intermediate between Cooper
pairs and composite bosons [13].
The nature of the normal state just above Tc in the BCS-Bose crossover
raises a very interesting question [14]: how does one crossover from
a Fermi liquid at weak coupling to a normal Bose liquid of tightly
bound pairs in the strong coupling limit? Monte Carlo simulations of
attractive fermions in two dimensions have recently shown [14] that at
moderate couplings where one still has a degenerate Fermi system the
spin correlations are very different from that expected of a Fermi liquid,
showing characteristic "spin-gap" anomalies. These results, which will be
described at length below, are particularly interesting because many of
the underdoped cuprates show precisely these anomalies.
It must be emphasized that the theoretical problem of high Tc super-
conductivity in its full complexity goes beyond the simple ideas described
in this paper. The present approach does not address issues such as the
mechanism that gives rise to such high transition temperatures, the prox-
imity to a metal-insulator transition, the role of antiferromagnetic spin
fluctuations, and the symmetry of the order parameter. However, because
the ideas presented here are simple and general, the author believes that
they are relevant to the high Tc systems independent of microscopic
details. For a complete description, one must have a microscopic theory
that gives the correct symmetry of the order parameter and includes the
physics of short coherence length superconductors discussed here.
We must also mention several other topics of current interest where
questions closely related to the ones discussed in this paper arise. (1) The
physics of tightly-bound real space pairs has been explored in detail in the
context of bipolaron theories [15, 16]. These studies (see also the review
by Ranninger in this volume) are directly relevant to the Bose limit of the
crossover problem. (2) The problem of Bose condensation of excitons [10]
has already been alluded to. The possibility of biexciton formation and
phase separation due to formation of electron-hole droplets presumably
complicates a simple crossover scenario. (3) The crossover from itinerant
to local-moment magnetism has been studied intensively in the context
of metallic magnetism [17], and more recently for the insulating state
of the half-filled repulsive Hubbard model [18]. (4) A problem of great
current interest is the superconductor-insulator transition in disordered
systems [19]. The present considerations might have some bearing upon
such questions as: To what extent do purely bosonic models describe
this transition and the two phases on either side of it? What is the role
of the fermionic degrees of freedom? (5) Issues related to the BCS-Bose
 Crossover from BCS Theory to BEC 359
crossover arise in the study of the phases of hydrogen under high pressure
[20]. (6) Related questions also arise in nuclear physics, for example in
the study of the equation of state of extended nuclear matter [21].

1.3 Outline
The remainder of this article is organized as follows. In Section 2 we
describe two models for studying the BCS-Bose crossover.
In Section 3 we first introduce the functional integral formalism which
will be used in the first part of the paper. Using this we address the
question of pair formation versus condensation. The problems associated
with a finite temperature analysis in two dimensions are mentioned
briefly.
We discuss time-dependent Ginzburg-Landau theory in Section 4, and
describe how the dynamics of the pairs just above Tc evolves from the
BCS to the Bose regime.
We turn to the broken symmetry state in Section 5 focusing mainly
on the T = 0 case. The crossover in the ground state and single-particle
excitation spectrum is discussed using a variational formulation.
In Section 6 we describe the evolution of the collective modes using a
generalized RPA analysis.
In Section 7 we discuss numerical results obtained by the quantum
Monte Carlo method, focusing on recent results on the normal state
correlation functions in the crossover regime in two dimensions.
A summary and conclusions are presented in Section 8.
The main focus of this review will be the work done by the author and
his collaborators (in rough chronological order): J.M. Duan, L.Y. Shieh,
J.R. Engelbrecht, L. Belkhir, N. Trivedi, C. Sa de Melo, R. Scalettar and
A. Moreo. An attempt has been made to reference much of the recent
literature relevant to the models of Section 2 but the author apologizes
in advance to those whose work has been left out.

2 Preliminaries
In this section we introduce two models, one in the continuum and the
other on a lattice, of fermions with attractive two-body interactions. Our
aim is to see how, as a function of increasing attraction, these systems
evolve between two apparently very different regimes (see Fig. 1): BCS
theory in weak coupling with a Fermi liquid normal state and tightly
360 M. Randeria

1
Bose
liquid

Fermi /
liquid^A,
BCS Bose
Coupling

Fig. 1. Schematic phase diagram of the attractive fermion problem. The full
curve is the transition temperature Tc for a continuum model (see text), and the
dashed curve is Tc for the lattice (attractive Hubbard) model. The shaded region
represents a crossover below which pair correlations are important. In the Bose
limit, the crossover scale is the pair dissociation temperature.

bound pairs which Bose condense from a normal Bose liquid in strong
coupling.
The continuum model has the Hamiltonian density
V2 1

The chemical potential fi fixes the average density n. We choose units in

which h =fcfl= 1 and the system is in a unit volume.
Since we want to describe the Bose regime in terms of its constituent
fermions, we have to allow the magnitude of the attraction g to be
arbitrary. In addition, we cannot use a simple BCS-like cutoff coD <C eF
since we must allow for the possibility of all the fermions to be affected
by the interactions, not just a small fraction within a shell around the
Fermi energy. We will thus study a dilute Fermi gas in which the range of
the attractive interactions is much smaller than the average interparticle
spacing, and the attraction can be characterized by a shape-independent
parameter, the scattering length as in three dimensions. We will describe
exactly how the "renormalized" coupling as replaces the "bare" g in
Section 3.
While the continuum model described above is very useful for analyt-
ical approaches, the attractive Hubbard model
c
H = —t y^ iacjo — U y , ni\nii ~ f1 / . ( n i| + ni[) (2)
ija i i

is in many ways simpler. There is no need for the coupling constant

 Crossover from BCS Theory to BEC 361
renormalization: the on-site attraction U is cutoff in momentum space by
the bandwidth W = 2zt, where z is the coordination number. Another
appealing feature is that even in the large- U limit there is no phase
separation [9]; one only has pairs and not quartets or larger clusters.
For large U two fermions form an on-site singlet and other fermions are
prevented by the Pauli exclusion principle from coming on that site. (In
the continuum model it is difficult to prove that phase separation does
not occur. If it does, one can certainly imagine modifying the interactions
between fermions which prevent this from happening while not affecting
the BCS-Bose crossover physics that we are interested in).
In the weak coupling limit, the lattice and continuum models describe
essentially the same physics, since the Cooper pairs corresponding to
(2) turn out to be much larger than the lattice spacing a. In the
strong coupling limit significant differences arise, since once the pair size
becomes comparable to a, lattice effects naturally become very important.
For U/t > 1 the effective low energy description [16] of (2), in the
subspace of empty and doubly occupied sites, is in terms of a hard-core
Bose lattice gas with hopping £eff ~ t2/U. Since the bosons become
heavier with increasing coupling the superconducting Tc decreases as
t2/U in strong coupling (see Fig. 1). This is in contrast to the continuum
case where the boson mass is 2m in the strong coupling limit and, as we
shall show, Tc saturates at the BEC value.
While in the continuum model there is a single dimensionless parameter
1/kfds, the attractive Hubbard Hamiltonian is characterized by two:
the coupling U/t and the filling factor / , controlled by ja. Note that
/ = (n)/2, i.e., one-half the average site occupancy. We will only be
concerned here with systems away from half-filling, / ^ 1/2, where one
has a superconducting ground state; for / = 1/2 the ground state has
both superconducting and charge-density wave order [16].
Most, but not all, of the analytical results described below will be
for the continuum model (1), while all of the numerical progress on the
crossover question has been in the context of the lattice Hamiltonian (2).

3 Tc and Above: Pair Formation vs. Condensation

We will begin with a description of a functional integral formulation of
the continuum model (1) which permits us to make useful approximations
at all couplings and temperatures both above and below Tc. Within a
single formalism this allows us to recover all the previously known results
- which had been obtained using different techniques in different regimes
362 M. Randeria
- and to go beyond them. All of the results from the functional integral
approach described below were obtained in collaboration with Sa de
Melo and Engelbrecht, for detailed references, see [22, 23].
The partition function Z, at a temperature j?"1, is written as an
imaginary time functional integral [24] with action

S = I dx I dx(ipa(x)dxxpo(x) + JtT)9 (3)

where x = (x, T). Decoupling the fermion interaction with the Hubbard-
Stratonovich field A(X,T), which couples to xpxp, and integrating out the
fermions we obtain Z = jDADAexp— Seff[A,A]. The effective action

Seff [A(x)] = - fdxfdx\A{x)\2 - TrIn G"1 [A(x)]. (4)

g Jo J
is written in terms of the inverse Nambu propagator

We now analyze this functional integral in terms of static, uniform

saddle-points and fluctuations about these. The trivial saddle-point A = 0,
which is stable at sufficiently high T for all couplings, will become
unstable below a certain temperature which we denote, for now, by To
(rather than Tc for reasons which will become clear). The transition
temperature is then defined by the solution of SS^/dA [A = 0] = 0,
which can be written as 1/g = ^ktanh(£k/2Tb)/2£k.
Before proceeding further, we need to describe how we regulate the
ultraviolet divergence in the above equation. The idea is to replace the
bare g by the low energy limit of the the two-body T-matrix (in the
absence of a medium). In three dimensions (3D) we use [25]
2£k)-1, (6)
|k|<A

which defines the 5-wave scattering length as. Recall that as a function
of the bare interaction, l/as increases monotonically from —oo for a very
weak attraction to +oo for strongly attractive interaction. Beyond the
two-body bound state threshold in vacuum (l/as = 0), as is the "size"
of this bound state with binding energy E\> = 1 /ma]. The dimensionless
coupling constant in the dilute gas model is then l/kFaS9 which ranges
from —oo in the weak coupling BCS limit to +oo in the strong coupling
Bose limit.
 Crossover from BCS Theory to BEC 363
Using (6) we obtain the equation for the transition temperature in
terms of the renormalized coupling as:
tanh (£k/2T0) 1
m
=£ (7)

with £k = £k — I* and e^ = |k|2/2m. There are two unknowns in this

equation, To and /z, and thus we need another equation, N = — dCl/djj.,
which fixes the chemical potential for a given density. Note that unlike
the BCS analysis in which one may set \i = eF (the non-interacting Fermi
energy) at the outset, in the crossover problem \i will turn out to be a
very strong function of the coupling as one goes into the Bose regime
where all the particles are affected by the attractive interactions.

3.1 Saddle Point Analysis: Pair Formation

The saddle-point estimate To is obtained by using the approximation
Qo = 5eff [A = 0] //? for the thermodynamic potential. The number
equation is then given by

n = n o (//,r) = ^ [ l - t a n h l ^ l 1. (8)
k / /

In the weak coupling limit, l/kFas —• —oo and we find the BCS results
fi = eF and To = Se^yn^epQxp (—n/2k F\as\), where y ~ 1.781.
The equations can also be solved analytically in the strong coupling
limit, where we find that the roles of the gap and number equations are
reversed: the gap equation (7) determines \x, while the number equation
(8) determines To. In this limit l/kFas —• +oo and one finds tightly bound
pairs with binding energy Eb = \/ma2s. The non-degenerate Fermi system
has \i ~ -Eb/2 and one finds To ~ Eb/2ln (Eb/eF)y2.
This unbounded growth of the "transition temperature" is an artifact
of the approximation and there is, in fact, no sharp phase transition
at To (outside of weak coupling where To = Tc ignoring the small
effects of thermal fluctuations). The point is that the saddle-point ap-
proximation becomes progressively worse with increasing coupling: the
trivial saddle-point A = 0 can only describe a normal state consisting
of essentially non-interacting fermions. While this is adequate for weak
coupling, in the strong coupling limit unbound fermions are obtained
in the normal state only at very high temperatures. In this limit, where
the system is completely non-degenerate, a simple "chemical equilib-
rium" analysis (boson ^ 2 fermions) yields a dissociation temperature
364 M. Randeria
Tdissoc = Eb/\n {Eb/eF) . The logarithm in the denominator represents
an entropic contribution favoring broken pairs and leads to Tdissoc < Eb,
the binding energy. We thus see that in strong coupling To is related to
the pair dissociation scale rather than the Tc ( < To) at which coherence
is established.
A numerical solution of (7) and (8) yields To smoothly interpolating
between the two limiting cases analyzed above. This temperature repre-
sents a "pairing scale" below which pair correlations become important;
for sufficiently large attraction these correlations describe the formation
of real bound states. For temperatures above the pairing scale To, the
system may be described as essentially consisting of unbound fermions.

3.2 Beyond Saddle Point: Pair Condensation

Since the saddle-point estimate for Tc fails with increasing coupling, we
must include the effects of fluctuations about A(x, T) = 0. We show that
treating the fluctuations at the Gaussian level produces a sensible strong
coupling answer for Tc provided their frequency dependence is retained.
This is, in fact, simply another way to state Nozieres and Schmitt-Rink's
(NSR) results since the Gaussian theory for Tc is shown to be identical
to their diagrammatic calculation [9]. One of the nice things about
the functional integral approach is that it provides a systematic way to
check the adequacy of the Gaussian approximation by looking at the
fourth order corrections (see below) and, in principle, to systematically
go beyond it. (In addition, the functional integral approach also works
below Tc.)
Before plunging into the details of the calculation, it may be worth
commenting on static spatial ("classical") fluctuations versus dynamic
("quantum") fluctuations. The role of spatial fluctuations is well-known
from the study of critical phenomena. Dynamic fluctuations, however,
become crucial in determining Tc with increasing coupling because the
bosonic degrees of freedom present in the strong coupling normal state
can only be adequately described by retaining the full frequency depen-
dence of the fluctuations about the trivial saddle-point. In some crude
sense, the formation of tightly bound pairs is a quantum fluctuation
which must be included in the functional integral.
To second order in A the action is

Seff - Seff [0] + E IA <*

 Crossover from BCS Theory to BEC 365
where

Here nk is a Fermi function and zg/ = i2ln/(l are the Matsubara Bose
frequencies. The resulting expression for the thermodynamic potential
Q = Qo - jS"1 ^2qMl lnF(q,i<?/), is identical to that obtained by NSR [9]
by summing particle-particle ladders for the free energy. Following these
authors, it is convenient to rewrite Q in terms of a phase shift defined
by F(q,co ± iO) = |F(q,co) | exp(±i<5 (q,a>)). The new number equation
N = — SQ/d/z is then given by

n = no(fi, T) + ] T I — nB(co&q,co)9 (10)

q J-oo H Ol1

where the "free" no is defined in (8) and HB{CO) = l/(exp(/?co) — 1) is

the Bose function. The second term in (10) describes the effects of
pair correlations, and, for sufficiently large attraction, the formation of
two-body bound states.
We note in passing that in the non-degenerate limit, say at very high
temperatures, the Gaussian formalism reduces to the classic analysis of
Beth and Uhlenbeck on the second virial coefficien [26]. To see this, note
that, when nk <C 1, F coincides with the familiar T-matrix describing two-
particle scattering in the absence of a medium. The resulting expression
for Q then relates the thermodynamics to the two-body scattering phase
shift, as first described by Beth-Uhlenbeck. In this sense the Gaussian
theory is a generalization of their result to two-body scattering in a
medium [9].
The temperature Tc at which long range order is established is defined
by the solution of (7) together with (10). In weak coupling, the results
are essentially unaffected by the inclusion of Gaussian fluctuations in the
number equation and Tc is the same as the saddle-point To obtained
earlier. A numerical solution of the equations shows that with increasing
coupling, Tc deviates from To and goes to a finite strong coupling limit
which coincides with the BEC transition temperature (see Fig. 2). The
chemical potential /x(Tc) for the fermions also smoothly interpolates
between the weak and strong coupling limits as shown in Fig. 3. In fact,
the equations may be solved analytically in the Bose limit. From (7) we
find JX{TC) = —Et,/2, which is one-half the energy required to break a
pair. Further, (10) may be simplified (using the properties of F described
366 M. Randeria

Fig. 2. The transition temperature Tc (full line) and the crossover scale To (dashed
line) plotted as a function of the dimensionless coupling l/kFas for an attractive
Fermi gas in three-dimensions. The BCS limit corresponds to l/kFas —• — oo and
the Bose limit to l/kFas -> oo. All temperatures are in units of the noninteracting
Fermi energy eF. To is obtained from a solution of the saddle-point equations,
while Tc is obtained from the Gaussian approximation described in the text.
(From Ref. [22].)

p 1 • 1 111 11
1 ' ' ' '' ' ' -

\ :
0
-
V
-1 - \ -

T . . . . 1 . . . . .... i A.-
- 2 - 1 0 1
l/(k r a.)
Fig. 3. The chemical potential fi/eF, obtained from the Gaussian approximation,
plotted as a function of the dimensionless coupling l/kFas for an attractive Fermi
gas in three-dimensions. (From Ref. [22].)

below; see also Ref. [9]) to obtain Tc = [n/2C(3/2)]2/3 n/m = 0.218eF,

which is the BEC result for bosons of mass 2m and density n/2.
To understand how this strong coupling result emerges from our
formalism, note that in this limit F(q,z) has an isolated pole on the real
axis, for each q, representing a two-body bound state with center-of-
mass momentum q. Since the energy required to break a pair is very
 Crossover from BCS Theory to BEC 367
large, this pole is widely separated from the branch cut representing the
continuum of two-particle fermionic excitations. The low energy physics
for T <C Tdissoc is thus dominated by this pole, and one can approximate
T(q, iqm) ~ R(q)/ [iqm - cob(q) + 2/i], where cob(q) ~ -Eb + \q\2/4m is the
dispersion relation for a pair of mass 2m. The partition function, after
rescaling the Afield,may be rewritten as

-Z.J d(j)d(f) exp (f)q (iqi - (ob(q)

which is nothing but the partition function of a free Bose gas. Thus we
see how the BEC result is obtained in the strong coupling limit. In the
next section we will go beyond the Gaussian approximation and see that
the fourth order term describes a repulsive two-body interaction between
the bosons. As is well-known, this repulsion stablizes the low temperature
phase, but does not substantially affect the condensation temperature.
It may be worth commenting on the fact that the numerical result for
Tc (see Fig. 2) appears to be a non-monotonic function of l/kFas with
a maximum value at intermediate coupling which is slightly larger than
the BEC value. We note that this is where the Gaussian approximation
becomes the most questionable (see next section), and even within this
approximation the numerical accuracy is the least. Nevertheless, this
qualitative feature was also seen in Nozieres and Schmitt-Rink's results
for a somewhat different model with a separable potential. The very
large maximum obtained in some other calculations [27] is certainly an
artifact of the approximate form used for the number equation.
Finally, we must mention two other approaches to the finite temper-
ature crossover problem. Haussmann [28] has generalized the Nozieres-
Schmitt-Rink diagrammatic approach using a self-consistent Green's
function formulation which appears to be promising. Gyorffy and
coworkers [29] have used the coherent potential approximation (CPA) to
study the attractive Hubbard model. This method correctly emphasizes
the phase coherence aspect of the transition. However, while this theory
gives reasonable results in the strong coupling regime, it appears not to
work for sufficiently small attraction, at least in its present form.

3.3 Two Dimensions (2D)

The case of two dimensions has recently attracted much attention partly
because of its possible relevance to the layered high temperature super-
368 M. Randeria
conductors. We will discuss the T = 0 formulation of the crossover in the
two-dimensional (2D) ground state and collective modes in later sections.
Here we discuss the question of the Tc crossover. As is well-known, phase
fluctuations destroy off-diagonal long range order (ODLRO) in a strictly
2D system. Thus Tc must be interpreted as the Kosterlitz-Thouless (KT)
temperature below which there is algebraic order and a finite superfluid
density.
Most of the analyses [30] of the Tc crossover in 2D have been based
on a Nozieres-Schmitt-Rink approach. As shown above, this is simply
a Gaussian approximation to the functional integral which perhaps ex-
plains the difficulties faced in these calculations. While incorporating the
all-important temporal fluctuations, one is also including the spatial fluc-
tuations at the Gaussian level. On the one hand, Gaussian fluctuations
destroy ODLRO in 2D and if one searches for a "T c " at which LRO sets
in one should get zero, and, on the other, they are completely inadequate
to describe the KT transition. It is not an easy matter to develop a theory
of the KT transition starting with fermions with attractive interactions
even in the weak coupling limit. Therefore, it is perhaps not surprising
that an analytical treatment of the finite T crossover problem in 2D is
still an open problem. We will describe numerical results from quantum
Monte Carlo simulations in Section 7. For a completely different ap-
proach starting from the broken symmetry state and using the KT jump
in the superfluid density ps to determine Tc (see Ref. [31]).

4 Time-Dependent Ginzburg-Landau Theory near Tc -

In this section, we will discuss the effective low energy, long-wavelength
action describing the bosonic degree of freedom A(x, t), and the resulting
equation of motion: the time-dependent Ginzburg-Landau (TDGL)
equation. These results are based on Refs. [22, 23]; the results for
the static Ginzburg-Landau theory were first given by Dreschler and
Zwerger [27]. There are two distinct aims of such a study. (1) By
looking at the A4 terms in the action one can assess the validity of the
Gaussian theory described earlier. (2) It is of great interest to understand
the evolution of the low frequency dynamics of the bosonic degrees of
freedom above Tc, since the behavior in the two limits is very different:
diffusive in the BCS regime and propagating in the Bose limit.
We begin by expanding the action (4) to fourth order in A(q), where
 Crossover from BCS Theory to BEC 369

q = (q, iQm\ to obtain

l 2
(11)

The Gaussian piece was already studied above (see (9)), and the coefficient
of the nonlinear term is given below in terms of fermion propagators. It
suffices to set the external frequencies and momenta - the arguments of
b - to zero, so that

where ikn = i(2n + l)n/P is a Matsubara Fermi frequency.

To study the evolution of the TDGL, we consider SSeft/dA*(q) = 0
near Tc for A slowly varying in space and time. For the static part of the
problem, one needs to make the expansion p-^qjO) = a + c\q\2/2m + ....
It is straightforward to find general expressions for the coefficients a, b and
c. These are in the form of momentum integrals which can be evaluated
numerically once Tc and fi(Tc) are determined from the equations of the
previous section. We will not quote these general results here (which
can be found in Refs. [22, 23, 27]) and will only give the BCS and Bose
limiting values below, where the integrals are analytically tractable.
We next turn to the much more interesting question of the time-
dependence of the linearized TDGL equation. Given the questions about
the validity of various microscopic derivations [32] of TDGL in the weak
coupling BCS regime, it is important to make some clarifying remarks.
Our main aim is to study the dynamics of the pairs above Tc, which is in
some sense a simple regime; the complications arise below T c if one really
wants to study the nonequihbrium dynamics of the order parameter. For
our purposes it is valid to use the Matsubara technique. Further, by
studying fluctuations about the broken symmetry state we can argue [23]
that our results are also valid just below Tc provided
|A0| <co<min{T c ,| ) u c |}, (13)
so that Ao, the non-trivial solution of of the saddle-point equation (see
next section), is the smallest energy scale in the problem.
We need to make a low frequency expansion of Q(iqi) = F -1 (q =
0, iqi)—T~ l(0,0) after analytic continuation. (Note that in Ref. [27] an ex-
pansion in Matsubara frequencies was made before analytic continuation,
370 M. Randeria
thus yielding unphysical results.) We find that Q(co+iO+) = Q'((o)-iQ"(co)
with
y
= KCO & def -? ; (14)
Jo 2(e - /i)(co -2e + 2\x)
2(e /i)(co 2e + 2\x)
and
Q"(co) = K2-3/2TZ (2// + co)1/2 tanh (co/4Tc) 0 (2/i + co), (15)
where K = N(eF)/\feF. A low frequency (co <C Tc) expansion of Q is
possible both in the BCS and Bose limits, where the condition co <C \/x\ is
automatically satisfied. Note, however, that for that intermediate coupling
for which pt(Tc) = 0, F(q = 0,z) has a branch point at the origin, the
point about which we would have liked to expand. The vanishing energy
scale n(Tc) does not permit a low frequency expansion and a linearized
TDGL description fails. Strictly speaking, condition (13) implies that
a linearized TDGL analysis is not valid inside a small region \p\ < Ao
about fi(Tc) = 0.
We can use the expansion Q(co + i0+) = —dco + ..., provided (13) is
satisfied, where

Z ^ ^ (16)
Collecting the results of the small |q| and small co expansions, and
transforming to (x, t), we obtain the result

a + b\A (x, t) | 2 - ^ V 2
- idj\ A(x, 0 = 0. (17)

We now discuss the two limiting cases, and the intervening singular point,
in more detail.
In weak coupling, we obtain the well-known results a = N(eF)ln(T/
Tc), b = 7C(3)AT(eF)/87r2T2, c = U(?>)N(eF)eF/\2n2T? and d = (inN(eF)/
8TC) -[1 - i(2Tc/neF)]. By rescaling the order parameter *F = >/2cA, we
obtain the conventional TDGL equation:
(e + | ¥ | 2 - <£LV2 + TGLdt) ¥(x, t) = 0, (18)
where e = \T - Tc\/Tc < 1. The characteristic length scale is £(T) =
<^GL£~ 1/2 with £GL ~ vF/Tc, which is the size of the Cooper pair £pair
(defined in the following section). The width of the Ginzburg region,
defined in the usual way as that range of reduced temperatures e for
which the effects of the fourth order term become important, is very
small, being of order (Tc/eF)4.
 Crossover from BCS Theory to BEC 371
The dynamics of *F in the BCS limit is overdamped with characteristic
time scale T(T) = TGL£~\ where TGL = n/STc. This damping arises
from the continuum of fermionic excitations into which a pair can decay.
There is in addition an (9 (Tc/eF) propagating part coming from the real
part of d, since our model does not impose particle-hole symmetry [33].
As the coupling increases, we see from (16) that the coefficient of
the propagating piece grows while that of the damped part diminishes.
The singular point fi(Tc) = 0 separates the regime of damped pair
excitations from that of propagating pairs. For stronger coupling, the
fermionic excitations have a gap in the normal state, and an essentially
propagating mode is obtained at this level of approximation [34]. The
full significance of the singular point at intermediate coupling where
fi(Tc) = 0 is not clear at the present time. Nor is it known whether
the thermally excited modes, ignored in this description, smooth out the
effects of this singularity.
In the extreme strong coupling limit we find the TDGL coefficients to
be given by a = KU (\fi\^2 - (Eb/2)V2), b = KU/32\II\3/2, C = KTT/16|/I|1/2,
and d = Kn/lfi^2, where K = N(eF)/elJ2. Note that we can set \i ~ —E b/2
in b,c and d. Using the normalization ^o = s/dA we can rewrite (17) as
- jW0 + U\¥o\ 2Vo ~ (2M)"1V2^o - id Wo = 0. (19)
This is simply the Gross-Pitaevskii equation for a dilute gas of bosons
of mass M = 2m with a repulsive interaction U = Ana^/M character-
ized by a (boson) scattering length a\> = 2as > 0 with n ^ <C 1, where
nb = n/2 ~ kp (in terms of the kF of the constituent fermions). This re-
pulsive interaction between the composite bosons has also been obtained
independently by Haussmann using a self-consistent Green's function
method [28].
The chemical potential of the bosons controls the phase transition via
the change in sign of ]i = Eb — 2|/x|. The prefactors of the divergent length
and time scales at this transition are given by £GL — kjl/^fkFas > ^pair —
as and TQL — T~l/(kFas) respectively. These are both much longer than
the microscopic scales because of the diluteness condition kFas <C 1.
Consequently the Ginzburg region (s <C kFas) is again small in strong
coupling.
We will find in the next section that the size of the pairs ^pair defined in
terms of the T = 0 wavefunction is a monotonically decreasing function
of the attractive interaction, as one might guess on physical grounds. The
"Ginzburg-Landau" coherence length in the dilute gas model is, however,
a non-monotonic function of the coupling. A numerical evaluation of the
372 M. Randeria
TDGL coefficients shows that £GL decreases with increasing interactions
at first, reaches a minimum value of order of the interparticle spacing fe^ 1
at intermediate coupling, and then increases again attaining the strong
coupling form discussed above.
Closely related to the large Ginzburg-Landau lengths in the two
extreme limits is the small Ginzburg region found in both the BCS
and the Bose limits of the dilute gas model. Thus the neglect of the
fourth order contributions is justified in both limits. However, in the
intermediate coupling regime, where /CJ^GL ~ 1, one does not have a
small parameter which controls the calculation: the Ginzburg region is
of order e = (T—T c)/Tc ~ 0(1). The only reason to trust the intermediate
coupling results, at least qualitatively, is that they smoothly interpolate
between two non-trivial limiting cases. At T=0 the variational principle
provides further justification, as will be seen in the following section.

5 Broken Symmetry State - Mainly T=0

The functional integral approach described above works equally well [23]
in the broken symmetry state for all T < Tc; however, except for a few
brief remarks at the end, we shall mainly focus on a T = 0 variational
approach [5]. The two approaches are in fact completely equivalent at
T = 0.

5.7 Ground State Crossover

In the absence of a small parameter, it is very useful to have a variational
formulation. This approach works for both the continuum and the lattice
models of Section 2. We begin with the BCS variational wavefunction

where |wk|2 + W 2 = 1- It is guaranteed to give the standard results in

the BCS limit; the reason why one might expect it to work even in the
the Bose limit of tightly bound pairs is the following. We can rewrite

= const.
k

= const, x exp I ^ <Pk4|c-kj ) I0)'

V k /
 Crossover from BCS Theory to BEC 373
where q>k = ^k/wk- The iV-particle projection of the last expression yields

which looks formally indentical to a condensate of composite bosons

with an internal pair wavefunction cp. Thus in a limit in which the "size"
of the pairs described by cp is much less than the average spacing between
the fermions, |*F) in (21) describes a condensate of composite bosons.
The calculation then proceeds along the usual BCS route [1], with
one difference which we note below. We define, as usual, v£ — 1 — u\ =
j(l — £k/£k) where E^ = J£%, + A£ with £k = s^ — [i. The gap function
Ak is determined by the standard gap equation Ak = — ^k, V^A^/IE^,
where V^p represents the attractive interaction.
An important difference with the standard BCS method is that the
chemical potential \i for the fermions must be determined self-consistently
with the gap function [4, 5]. In the weak coupling limit \i = eF, where
eF is the Fermi energy of the non-interacting system to an excellent
approximation. However with increasing attraction the size of the Cooper
pairs decreases and consequently the momentum occupation probability
nk = 2v\ broadens out considerably, resulting in a significant shift in \i.
This is determined by the number equation Ysknk = ^> which can be
written as

5.2 Three Dimensions (3D)

To be concrete we first focus on the attractive Fermi gas (1) in 3D. The
gap parameter can be shown to be k independent in the low energy limit,
so that Ak = A. (The gap and number equations derived above work
equally well for the attractive Hubbard model (2); see Section 6 and
Refs. [9, 16, 35].) Using the regularization procedure (6) we obtain the
"renormalized" gap equation [25]

M2N(6F) (24)

The order parameter A and chemical potential \i at T = 0 are obtained

by solving (23) and (24). In the weak coupling BCS limit (l/kfas —• — oo)
we find \x = eF and A = Se~2€fQxp (—n/2kp\as\) so that A = nTc/y,
374 M. Randeria
(y ~ 1.781). In the opposite strong coupling limit (l/kFas —• +00),
we expect \i < 0 and \\i\ > Ao. Analyzing (24) and (23) under these
conditions, which we verify at the end, the integrals are tractable. We
obtain A = (16/3TT) €F/y/k^a~S9 and ILL = —E^/l + 4nasn/m, where
Eb = I/ma] is the pair binding energy. It is worth recalling that in strong
coupling fi(Tc) = —Eb/2, and the shift in JLL(T = 0) is simply a result of
the repulsive interaction U = 4nas/m between the bosons.
A numerical solution [23] of the gap and number equations shows a
smooth (but in this particular case, a fairly rapid) crossover from the
BCS to the Bose limit. Note that there is no singularity at either the
two-body bound state threshold (\/kpas = 0) or at the point at which
/z = 0.
The size of the pairs may be estimated using

^pair = <V>k|r2|V>k>/<V>klV>k>, (25)

where the "internal pair wave function" is [36] xp^ = Ak/2£k, which is
related to, but distinct from, the wave function cp^ in (22). Evaluating
the matrix element (using r 2 —• V£) we find that in the BCS limit
£pair ~ ^ F / A > kp1, which is the interparticle spacing. With increasing
coupling the pair size decreases monotonically, with a limiting form
£Pair ~ as < kjl in the Bose regime.
There is an amusing feature in the evolution of the gap to single-
particle (fermionic) excitations [4, 5]. This is given by the minimum of
the Bogoliubov quasiparticle energy E^ within the band, namely,

£ k _ / ^ + AY' . (26)

From this it immediately follows that

F / A for/i>0
£ g a p
~ \ (/i2 + A^)1/2 for^<0 (2?)

(for a simple way to see this, see Fig. 6 of Ref. [13]). As pointed out
by Leggett [5], this has an important consequence for the case of non-
5-wave pairs. Even if one has nodes in the gap function in the BCS
limit, there is a non-vanishing gap in the composite boson limit which
simply corresponds to the ionization energy of the diatomic molecule.
The reader can find a detailed discussion of the crossover in non-s-wave
superconductors in Ref. [7] (for the 3D case) and Ref. [13] (in two
dimensions).
 Crossover from BCS Theory to EEC 375
5.5 Two Dimensions (2D)
It was shown by Miyake [8] and by Duan, Shieh and the author
(RDS) [13] that, in a 2D dilute gas model, the existence of a two-
body bound state in vacuum is a necessary (and sufficient) condition
for a Cooper instability. For purely attractive two-body potentials V(r)
this is not surprising since an infinitesimal attraction leads to a bound
pair in vacuum. However, as emphasized by RDS, when J d2rV(r) > 0
(e.g., hard-core plus attraction, where this integral is infinite), one has
to cross a finite threshold in the attraction before a bound state forms
in vacuum (absence of a medium), and the non-trivial result is that one
has to cross the same threshold for a pairing instability in a degenerate
Fermi system. In addition RDS showed that this result was special to
the s-wave channel: an /-wave bound state in vacuum is not a necessary
condition for a pairing instability with / > 0 in 2D.
The above result is intimately related to the jump discontinuity in the
2D density of states (DOS) at the bottom of the band, or the related
logarithmic singularity in the 2D propagator and T -matrix. The low
energy s-wave T-matrix for two-particle scattering in vacuum (i.e., in the
absence of a medium), which we shall need below, has the form
To(co) = (4n/m) [- ln(co/Ea) + in] ~*, (28)
where Ea is the binding energy of the two-body bound state. The
continuum model in 2D has another feature - a constant DOS - which
greatly simplifies the crossover calculation [4, 8, 13]. For the dilute
gas model in 2D this allows an exact solution of the T = 0 gap and
number equations for all couplings leading to very simple and transparent
results [13].
The variational gap equation 1/g = — ^ k l/2£k, with Ak = A, has an
ultraviolet divergence which is regulated by using the T-matrix. Unlike
the 3D case (6), however, we cannot take the zero energy limit at the
outset since the two-dimensional T-matrix (28) vanishes logarithmically.
We find the "renormalized" gap equation
i
4TT _ r d r Ll
mT0(2co) " Jo £ k L £ k - c o - i O + ~ £ i ; j ' ( )

which is independent of co. The only way in which the attractive interac-
tion enters is via the energy of the bound state in vacuum, which must
exist as described above.
The dimensionless coupling constant in the 2D case - the analog of
1/kpds in 3D - is given by Ea/eF9 which varies from 0 (in the BCS limit)
376 M. Randeria
to oo (in the Bose limit). It is then easy to solve the above gap equation
together with (23) for arbitrary Ea/eF to obtain

A = \/2eFEa and // = ey - Ea/2. (30)

This simple analytical result clearly shows a smooth crossover from the
BCS to the Bose limit. Note that the essential singularity of the BCS
limit is buried inside the dependence of the binding energy Ea on the
attractive interaction. For a detailed discussion of the 2D crossover
problem, including an extension to higher angular momentum pairing,
the reader is referred to Ref. [13].

5.4 Functional Integral Derivation of T=0 Results

We briefly describe, following Ref. [23], how one arrives at (24) and (23)
from the functional integral of Section 3. Below Tc one has a stable
uniform, static solution A(X,T) = A ^ 0 of the saddle-point equation
dSeff[A]/8A = 0. Using (4) and after performing the Matsubara sum, this
can be rewritten as the familiar gap equation
)
2Ek
k

The number equation is obtained, as before, by using the saddle-point

approximation for the thermodynamic potential Qo = Seff [^o]/P in JV =
This leads to

In the T = 0 limit, the above equations clearly reduce to the gap and
number equations derived from the variational approach above.
It is important to emphasize that the saddle-point approximation is
much more reliable for T <C Tc than it was for T > Tc. Outside of the
weak coupling regime, the important physics of pairing correlations, and
eventually bound state formation, above Tc is in fluctuations beyond the
trivial saddle-point. In the broken symmetry state, however, the non-
trivial saddle-point of Seff includes the non-perturbative effects of bound
state formation and condensation. At least for T <C Tc, the fluctuations
about the non-trivial saddle-point make only a small correction even for
strong coupling.
 Crossover from BCS Theory to BEC 377
6 Collective Excitations
Gaussian fluctuations about the broken symmetry saddle-point represent
the collective excitations of the superconductor. The functional integral
approach is a very elegant way to study the crossover problem for
collective modes, see Ref. [23]. We present here a simpler derivation
using the old-fashioned RPA [37] based on the work of Belkhir and
the author [35]. We should also mention that related results on the
evolution of the collective modes have been obtained by many authors
for the BCS-Bose crossover [38, 39], for the crossover in the exciton
condensation problem [40], and for the crossover from spin-density wave
(SDW) to local moment antiferromagnetism [18, 41].
The RPA analysis of collective modes has also been extended to
charged systems. While long range Coulomb interactions have been
largely ignored up to this point in our discussion, we discuss them here
since they have a profound impact on the collective modes [37, 42]. At
the end of the section we will briefly mention some results on charged
systems based on Ref. [35]. For a comprehensive treatment of plasmons
in layered systems, especially in the context of the BCS-Bose crossover,
the reader is referred to the recent work of Cote and Griffin [39].
Our study of collective modes is motivated by the following reasons.
The important excitations in the BCS limit involve broken pairs. With
increasing attraction these are pushed to very high energies, and in
the Bose limit it is the collective modes which are the dominant low-
energy excitations. Thus one might expect that a proper description of
the intermediate coupling regime must include both broken pairs and
collective modes.
Our starting point is the attractive Hubbard Hamiltonian (2) written
in momentum space

H
= Y,(£k " ric&ck* ~ M 5 1 ck+qi4-qick'lCkh (33)
M
k,a kk'q
where Sk = —2t J^f=1 cos(fcja) and M is the total number of lattice sites.
The "reduced" Hamiltonian ifbcs includes only the fc' = — k part of the
interaction term in (33), and thus only describes the interaction between
pairs with zero center-of-mass momentum.
The variational solution described above can be recast in the form
of a linearization of the equations of motion for CkG and c£a with the
dynamics governed by ifbcs- This is the well-known Hartree-Fock-
Bogoliubov scheme and one arrives at the familiar number and gap
378 M. Randeria
equations. A and the chemical potential ju are given by the solution
of 1/(7 = (2M)- 1 £ k £ ^ and / = (2M)" 1 £ k (1 - £k/Ek). Here £k =
ek-Ji, Ek = (<^+A 2 ) 1/2 , and / = N/2M is the filling factor. The only real
difference with the continuum model treated in earlier sections is that
one must take into account the Hartree shift in the chemical potential:
/} = // + fU, which can be significant for large U.
To determine the collective mode spectrum we study the time evolution
of density fluctuations c^+q(TCka- Its equation of motion is coupled to
that of pairs with a finite center of mass momentum cjj"+q+c+k\ and
c-k-q[Ck\^ resulting from the particle-hole mixing due to the condensate:
(cfctcifci) ^ 0- The overall strategy of the RPA is simple: first, to linearize
the equations of motion [37] with respect to the mean field ground state,
and, second, to diagonalize the resulting Anderson-Rickayzen equations
by finding the appropriate eigenoperators for the collective coordinates.
At the mean field level one has diagonalized only the H\>cs part of the
full Hamiltonian H = H\>cs + Hmi. At the RPA level, we treat the small
fluctuations introduced by H{nt which describes the interaction between
the Bogoliubov quasiparticles.
The actual implementation of this idea is algebraically messy. We
omit all details, referring the reader to Ref. [35], which generalizes the
weak coupling calculation of Bardasis and Schrieffer [43] to the crossover
problem. The BCS limit is considerably simplified by an approximate
particle-hole symmetry arising from the fact that only fermions within a
thin shell symmetric about the Fermi energy are affected by the pairing.
For arbitrary U, one does not have such a p-h symmetry and the
calculations are more complicated.
To determine the collective excitation spectrum co(q) we take the matrix
elements of the equations of motion between the ground state (which is
the BCS state renormalized by the zero-point motion of the collective
modes), and a state containing exactly one quantum of excitation. The
resulting secular equation is

+ UIE,n,n(q)] Ul^jiq) 2UIE/vn(q)

UIE,n,m(<l) [1 + ] = 0, (34)
(U/2)ImjKj{q)
with
1 . . n(k n\h(h n\r(k n\
(35)

Here a,fc,c denote any one of the following quantities: the excitation
 Crossover from BCS Theory to BEC 379
energy co(q)9 or the quasiparticle energy EkA = Ek+q+Ek, or the coherence
factors l{k,q) = ukuk+q + vkvk+q, m(k9q) = ukvk+q + vkuk+q, and n(k,q) =
ukuk+q — vkvk+q. We now solve (34) for co{q\ restricting attention to the
long wavelength q —> 0 regime.
In weak coupling, the integrals Iw,n,i and IE,n,m vanish as a result of the
p-h symmetry discussed above, and we are left with a 2 x 2 determinant
to solve. The small q and co expansion of the remaining terms in
(34) is conveniently written in terms of four quantities: x = X^^/f 3 >
y = ZkWktf/El w = Zk^U/El and z = A2^k(Wki)2/E5k. The
momentum sums can be readily evaluated to obtain the q —• 0 dispersion
relation in d dimensions

<»(«) = [<«£>/<*] 1/2 (1 - UN(0))1?2 q. (36)

Here (vp) = Ny(0)/N(0) is the mean squared Fermi velocity, where the
density of states N(£) = (2n)-d J ddkd(£k - f), and ATV({) = (2n)-d J ddk
(Vfc^)2(5(^fe — £). Note that (vp)1/2 is ta times a dimensionless function
of the filling / (for plots of this in 2D and 3D see Ref. [35]). The lattice
result (36) is essentially the same as Anderson's continuum result [37]
(where y/(vp) = PF/M), since the size of the bound pairs is much larger
than the lattice spacing a in weak coupling.
In the strong coupling limit the gap and the number equations can
be solved, to leading order in a = 2t/U <C 1, to obtain A = U[f(l —
Z)] 1 / 2 [1 - da2] and Ji = 1/(2/ - 1) [l + 2doc2] /2. To make contact with
the Bogoliubov theory for the dilute Bose gas we focus on the low
density limit / < 1. The elements of the determinant in (34) can
be expanded as follows: 1 + UIE,n,n = 4 / [l +4da 2 ], 1 + UIE,m,m =
1 - 16/da 2 , UIEM = ~2fl/2 [1 + da2], UIm^ = -co [l + 2da2] /U, and
UIw,m,i = 2cafV2 [-1 + Ida2] /U. We also find, independent of the filling,
1 + UIEU = oc2q2a2/2 -co2/U2. We thus obtain, for / < 1 and U/t > 1,
the excitation dispersion relation
co(q) = y/4df(4t2a/U)q. (37)
To make contact with Bogoliubov theory, we note that in the strong
coupling limit the fermions bind into on-site singlet pairs, and the attrac-
tive Hubbard model can be mapped onto a system of hard core bosons
described by //b0Se = (2t2/U) Ylij(~K^j + ninj)- The pairs move only
via virtual dissociation which leads to a t2/U hopping amplitude, and
the hard core constraint comes from the Pauli principle for the con-
stituent fermions. In the long wavelength limit, the collective excitation
of a dilute {n\>a\ <C 1) 3D Bose gas is the sound mode with dispersion
380 M. Randeria

co = (Ann^a^/m^) q. Here the density of bosons n\, = f/a3, the effec-

tive mass m\y ~ l/t\>a2 = U/2t2a2, and the boson scattering length a^ is
of the order of the lattice spacing a. This leads to a speed of sound which
is essentially the same as that in (37), up to numerical factors of order
unity, which are not unexpected given the crude estimate of a^. Quite
remarkably, starting with interacting fermions and using RPA we were
able to reach the regime of a low density Bose gas in the strong coupling
limit.
For intermediate couplings we have solved the RPA equation (34)
numerically as a function of the coupling U/t and for various fillings / ;
see Ref. [35] for details. The numerical results show that the spectrum
smoothly interpolates between the Anderson mode in weak coupling and
the Bogoliubov sound mode in strong coupling.
The evolution of the eigenvector corresponding the collective mode is
also of some interest. For q = 0 the eigenvector is a pure phase mode, for
all couplings, as required by Goldstone's theorem. However, as shown
in Ref. [23], for small q there is a small admixture of the amplitude
mode. The magnitude of this mixing is controlled by, in addition to q,
the deviation from p-h symmetry, and therefore grows with the coupling.

6.1 Charged Systems

We briefly summarize the results of an analysis [35] which includes the
effects of the long ranged Coulomb interaction
2(d-l)ne2
Vc{q) = (38)
cflqd-i '
where ad is the unit cell volume in d-dimensions. The effective interaction
in the particle-particle channel is taken to be the attraction —U as before;
however the bare Vc is used in the particle-hole channel. This can be
shown to modify the equations of motion as follows: in the third row
of (34) each factor of U is replaced by Vc(q\ while the first two rows
remain unchanged.
For charged systems we work only in the high density regime, since at
sufficiently low density one would expect an instability to a Wigner crystal
of pairs (which is clearly outside the scope of this analysis). For the lattice
model under consideration we work exactly at half-filling / = 1/2 since
this simplifies the algebra considerably. We only quote the final results
for q —• 0. In the weak coupling limit in 3D one finds the well-known
result [37] that the Anderson mode of the neutral system is pushed up
 Crossover from BCS Theory to BEC 381
to the plasma frequency: co2 = Sne2(vj)N(0)/3. For U/t > 1 in 3D
we find co2 = (4ne2/a3)(4t2a2/U). Using the boson mass m^ = U/2t2a2,
charge e& = 2e, and density n^ = l/2a 3 , this result may be rewritten
as co2 = 4nnBe^/niB which is exactly the plasma frequency of a dense
charged bose gas [44].
In 2D the plasmon does not have a gap and we find a Jq dispersion.
The weak coupling result in 2D is co2 = 2ne2(vp)N(0)q9 while in the
Bose limit we find co2 = (Sne2t2/U) q which may be rewritten as co =
(2ne^nB/mB) Jq. The smooth evolution of the plasmon from the BCS
to the Bose limit is demonstrated by a numerical solution of the RPA
equations in Ref. [35].

7 Numerical Studies: Normal State Crossover

The absence of a small parameter in the intermediate coupling regime,
the importance of dynamic fluctuations, and the special problems of two
dimensions discussed in Section 3, all suggest that numerical simulations
will play a very important role in elucidating the BCS-Bose crossover
problem. Quantum Monte Carlo (QMC) methods for strongly correlated
systems have been intensively investigated in recent years. While much
of the work on lattice fermions has focussed on the repulsive Hubbard
model [45], there have been some studies of the 2D attractive model
relevant to the present subject [46, 47, 14].
Very briefly, the finite temperature QMC scheme is as follows. Using an
imaginary-time coherent state path integral formulation, and a Hubbard-
Stratanovich decoupling, the fermion problem is written as a functional
integral over the bosonic decoupling fields with a complicated non-local
action. The bosonic path integral is then simulated using Monte Carlo
techniques. The determinants obtained from integrating out the fermion
fields can usually take on both positive and negative values, thus leading
to large statistical errors - the "sign problem" that plagues fermion Monte
Carlo calculations. An important feature of the attractive Hubbard model
is the absence of the sign problem [48]: for a charge decoupling, in which
the auxiliary field couples to (n^ + n^), the determinants are always
positive definite.
Scalettar et a\. [46] first studied the phase diagram of the 2D attractive
Hubbard model using QMC, and presented preliminary evidence for
superconductivity away from half-filling, and the coexistence of SC and
CDW, together with a vanishing TC9 at half-filling (n) = 1. Moreo
382 M. Randeria
and Scalapino [47] demonstrated the existence of off-diagonal algebraic
order for (n) ^ 1 and, using finite size scaling, estimated the Kosterlitz-
Thouless Tc. As a function of U and (n), the maximum Tc was found to
be around O.lt.

Anomalous Normal State Properties: Spin Gaps

The normal state, just above Tc, in the U/t <C 1 limit is a Fermi liquid,
while that in the opposite U/t > 1 limit is a normal Bose liquid. We
now describe the results of a QMC study [14] of how the normal state
evolves from a Fermi liquid-like regime to a Bose regime as a function
of increasing attraction. Understanding when and how deviations from
Fermi liquid behavior arise in strongly correlated Fermi systems is a
major topic of current research in many-body physics. The deviations
from Fermi liquid behavior discussed below are, in a sense, generic to
the normal state of a short coherence length superconductor.
We show that for intermediate coupling strengths the normal state
clearly deviates from a canonical Fermi liquid. Our main results are: (1)
the uniform, static spin susceptibility %s is strongly temperature depen-
dent, with dxs/dT > 0, for Tc < T < Tp, where Tc is the superconducting
transition temperature, and Tp a "pairing" scale below which strong sin-
glet pairing correlations develop. (Tp, which will be defined below, is
closely related to To of Section 3.) (2) This is accompanied by a "spin
gap", i.e, a reduction in the low frequency spectral weight, leading to an
NMR relaxation rate l/TxT which tracks xs{T). This is very different
from the Korringa law in a Fermi liquid, namely \/T\T ~ %*, with
temperature independent Pauli susceptibility Xs>
Our results provide a natural explanation for the anomalous spin
gap behavior of the NMR Knight shifts and relaxation rates ob-
served [49, 50, 51, 52] above Tc in many of the underdoped high Tc
superconductors. We discuss in more detail below the implications of
our results for experiments on the reduced oxygen YBCO systems, espe-
cially why the spin-gap anomalies are observed only in the underdoped
materials and not in the optimally doped ones. We also note that normal
state pseudo-gaps have also been observed in the optical conductivity [53]
of underdoped cuprates.
We emphasize that our results, which are perhaps easiest to rationalize
in the non-degenerate, pre-formed boson limit at large-1/, persist well
into the intermediate- U regime where one has a degenerate Fermi sys-
tem. This is important to establish because the materials which exhibit
 Crossover from BCS Theory to BEC 383
these anomalies also have a Fermi surface as measured by angle-resolved
photoemission spectroscopy (ARPES) [54]. Thus the spin gaps cannot
come from "pre-formed pairs" but, rather, from strong singlet correla-
tions.
We now give the details of our results. The numerical simulations are
on systems of size up to 8 x 8 with some data obtained from 16 x 16
lattices. Most of the data is at quarter-filling (n) = 0.5 for temperatures
T > OAt; note that Tc ~ 0.051 for U/t = 4 at this filling [47]. The
chemical potential /n is adjusted as a function of the interaction and
temperature to obtain the desired (n); see Fig. 1 of Ref. [14]. We find
dfi/dT < 0, as one usually expects, for all T and [/, in contrast to
the result of summing ladder diagrams [30] in 2D (see end of Section
2 for a discussion of problems with finite temperature calculations in
2D). Further, at low T the results agree with the T = 0 mean field
treatment [35] described in previous sections. Comparing the chemical
potential with the temperature we can check whether or not one has a
degenerate Fermi system, which is important for the reasons given above.
The uniform, static spin susceptibility Xs = x(Q -» 0,co = 0) as a
function of U and T, is shown in Fig. 4. For U = 0, x° has the expected
Pauli behavior for low temperatures and crosses over to a Curie-like form
at higher temperatures (with a small bump at T ~ 0.5 due to the log
singularity in the DOS at the band center). With increasing U the overall
magnitude of Xs decreases, which may be understood (at high T) within
RPA with xRPA = Z°/(l + Ux°). However, the behavior dxs/dT > 0
observed in the data is the result of strong singlet pair fluctuations in
the normal state and is, of course, completely absent in the RPA. For
large U one obtains tightly bound singlet pairs which contribute to Xs
only when they are ionized. What is remarkable is that such correlations
persist down to U = 4 where one has a degenerate Fermi system.
The temperature at which the Monte Carlo data deviates from the
RPA can be used to define a "pairing" scale Tp which is clearly larger
(for all the values of U studied) than the transition temperature Tc at
which coherence is established. We expect that Tp (defined from Xs) is
essentially the same as the crossover scale To of Section 3. The anomalous
behavior in the normal state discussed here is in the regime Tc < T < Tp.
For small U, Tp is expected to be the same as the mean-field transition
temperature Tc°, and one can describe the regime Tc < T < Tc° in terms
of superconducting fluctuations [55]. For large U, the two scales are
widely separated with Tc° ~ 2t/U, while Tp ~ U gives an estimate of
the pair binding energy. The anomalous temperature dependences in
384 M. Randeria

Fig. 4. The normal state spin susceptibility for the 2D attractive Hubbard model
at quarter-filling (n) = 0.5 plotted as a function of temperature, for various U
values. The full curve is for the non-interacting system. The U/t = 4 and 8 data
are from Monte Carlo simulations on an 8 x 8 lattice, and the U/t = 12 data are
on a 4 x 4 lattice. The error bars are of the size of the symbols. The RPA curve
is fit to the U/t = 4 data (not shown in the diagram) in the high temperature
regime 2t < T <4t using a "renormalized" U — 3.25. (From Ref. [14].)

the normal state then clearly arise from the formation of singlet pairs
without any coherence, and not from superconducting fluctuations.
A static susceptibility with dxs/dT > 0, though suggestive, is by itself
not sufficient to establish a "spin-gap", namely a reduction in the low fre-
quency spectral weight Im%(q, co), which is probed by the NMR relaxation
rate \/T\T = lim^oYlq Im z(^co)/co. In general, it is very difficult to
obtain Im/(q, co) from imaginary-time Monte Carlo data, since it requires
analytic continuation from Matsubara to real frequencies. However, we
use a trick that allows us to extract \/T\T directly from the imaginary
time spin-spin correlation function measured at T = 1/2T, provided the
temperature is much smaller than the characteristic frequency of the
spectral weight Im#(q, co) (see Ref. [14] for details).
The relaxation rate \/T\T determined from this analysis is plotted
in Fig. 5, it is found to decrease as T is lowered. In Fig. 6 we show
 Crossover from BCS Theory to BEC 385

0.1 1
1 1 1 ' 1
A
(n>=0.5-
A
A U=4 -
A D U=8
A
-0.05 -
A

A •
D
i rh Di 1 1 , 1
0
0 0.25 0.5 0.75
T/t
Fig. 5. The normal state NMR relaxation rate \/T\T (in arbitrary units) for the
2D attractive Hubbard model, at quarter-filling (n) = 0.5, for U/t = 4 and 8 on
8 x 8 lattices. (From Ref. [14].)

that \/T\T indeed tracks the static susceptibility &, thus establishing
spin-gap behavior in the normal state. It is worth noting that, for the
two values of U studied, the data for l/T\T vs. Xs appear to lie on the
same curve.
Is the opening up of a "spin gap" accompanied by a charge gap, or
a pseudo-gap? In this model the answer appears to be yes, although
more detailed work is necessary. The single-particle density of states
obtained via analytic continuation in Ref. [57] gives some evidence for
a gap-like structure developing above Tc. We have also studied the
momentum distribution function (rik). While (rik) is clearly broader than
what would be expected for a Fermi gas (Z = 1) with thermal smearing,
it is difficult to conclude (from fitting the Monte carlo data to various
functional forms) whether the observed behavior necessarily implies a
gap. However the measured (rik) again clearly demonstrates that the
system is in a degenerate Fermi regime.
Finally, we discuss the applicability of our results to normal state
386 M. Randeria

0.1 1
I 1
_ A
A
A

A
-0.05-
A

A (n)=0.5
A U=4
R • U=8 _
D
0 n i 1 , 1
0 0.05 0.1 0.15

Fig. 6. A parametric plot of the normal state NMR relaxation rate 1/TiT (from
Fig. 5) versus the spin susceptibility (from Fig. 4) for the 2D attractive Hubbard
model. For a Fermi liquid all of the data, for a given [/, should collapse to a
single point; instead the two quantities track each other: 1/TiT ~ Xs- (From
Ref [14].)

NMR experiments [49, 50, 51] on the YBa2Cu3O6+x systems; while these
are the best studied systems, spin-gap behavior is not [52] restricted
to them. Two issues (at least) need to be discussed: (1) the role of
antiferromagnetism (AFM), and (2) why spin gaps are seen in some
materials, and not others.
In the Tc = 60K system (x ~ 0.65) the Knight shift, which probes #s,
increases by a factor of four as T increases from Tc to 300K, which we
would identify as Tp. For the O and Y sites (where the form-factors [56]
filter out the AFM contribution) 1/TiT indeed tracks Xs below 300K.
Our model, which has no AFM, is able to explain these spin gap features
naturally. For the Cu site, l/T\ is larger in magnitude and shows spin gap
behavior only below 150K. This would appear to require a combination
of AFM fluctuations and a spin gap opening up due to enhanced pairing
correlations. Clearly our simple model does not have any non-trivial
q-dependence of the spin gap.
 Crossover from BCS Theory to EEC 387
In the Tc = 90K (x ~ 1.0) material, spin gap behavior is not seen (or
is seen over a very narrow regime). To discuss the range Tc < T < Tp
over which these anomalies will be seen, one needs to know how Tc
and Tp change as a function of doping (x). Since the attractive Hubbard
model is not a microscopic model for YBCO (e.g., half-filling in this model
has nothing to do with the magnetic insulator at x = 0), we can only
make qualitative remarks. Let us assume that once YBCO is metallic,
increasing x is analogous to increasing the carrier concentration n in the
model. For fixed (7, Tc increases with n, till n ~ 0.85, and then drops to
zero at n = 1.0 in 2D, as shown in Ref. [47]. On the other hand, Tp is
expected to be relatively independent of n for smallrc,to the extent that
one has independent pair formation, and could even drop with increasing
n due to the Pauli principle blocking the formation of independent pairs.
This crude argument suggests why the system with the highest Tc, and
highest carrier concentration, might have the smallest window between
the pairing temperature and Tc.

8 Summary and Conclusions

We have reviewed the subject of the crossover from cooperative Cooper
pairing to the Bose condensation of tightly bound pairs as the strength
of the attractive interaction increases. Most of the discussion centered
around a continuum model of an attractive Fermi gas or the attractive
Hubbard model on a lattice. A variety of techniques - variational, dia-
grammatic, equations of motion, functional integrals and quantum Monte
Carlo - have been used to study this problem. We briefly summarize the
main physical ideas that emerge from these studies.
First and foremost: at any given coupling, there is only one finite
temperature phase transition in this problem at Tc, which is from a
normal state to a superfluid state with ODLRO or, in two-dimensions,
algebraic order. Even though the weak and strong coupling limits appear
to be very different at first sight, there are no phase transitions as a
function of coupling (see Fig. 1).
In the broken symmetry state, below Tc, as a function of increasing
interactions, the system evolves from a BCS state, which is a condensate
of large, overlapping Cooper pairs, to a Bose condensate of tightly bound
composite bosons. Variationally, all that happens is that the pair wave
function shrinks in the relative coordinate; the large kinetic energy cost
required to do this being more than compensated by the even larger gain
from the attractive interaction. The important low energy excitations in
388 M. Randeria
the BCS limit are broken pairs with an exponentially small gap. With
increasing coupling these get pushed to higher energies. The Anderson-
Bogoliubov collective mode for the neutral BCS system evolves smoothly
into the Bogoliubov-Beliaev sound mode for the Bose gas. Inclusion of
long range Coulomb interactions modifies these modes into plasmons,
which also evolve smoothly from the BCS to the Bose limit.
The nature of the transition in the two limiting cases is rather different.
In the BCS case, the formation of the Cooper pairs and their conden-
sation occurs at the same temperature, which is Tc. The first deviations
from weak coupling may be understood in terms of superconducting
fluctuations. With increasing attraction the crossover scale To at which
pair correlations become manifest and the transition temperature T c get
increasingly separated. In the strong coupling regime, To is roughly the
temperature below which real bound states begin to appear. For T > To
the bound states are thermally dissociated. In the Bose limit, the for-
mation of the bound pairs and their condensation are two independent
processes with widely separated energy scales; the only phase transition
is at the condensation Tc.
The normal state just above Tc, in the BCS limit, is a Fermi liquid,
while that in the opposite strong coupling limit is a normal Bose liquid
of tightly bound pairs. How does the system evolve from a Fermi
liquid to a Bose liquid as a function of the interaction strength? In two
dimensions, numerical simulations are beginning to provide an answer.
For moderate values of the attraction, the normal state, even though it
is still in a degenerate Fermi regime, already shows anomalous magnetic
correlations due to the opening up of pseudo-gap in the normal state.
These effects are strikingly similar to anomalies seen in normal state
NMR and optical conductivity experiments on underdoped high Tc
systems.
In conclusion, the problem of the crossover from BCS theory to Bose-
Einstein condensation is a very rich one, and there are possibly new
surprises in the intermediate coupling regime that we have yet to un-
cover.

Acknowledgements. I would like to thank Jan Engelbrecht, Carlos Sa

de Melo, Nandini Trivedi and Lotfi Belkhir for invaluable input; much
of the work presented here was done in collaboration with them. I am
grateful to Tony Leggett for first introducing me to this problem and for
many conversations over the years. Finally I would like to thank Alexei
Abrikosov and Allan Griffin for their interest and encouragement.
 Crossover from BCS Theory to BEC 389
This work was supported by the US Department of Energy, Office of
Basic Energy Sciences under contract W-31-109-ENG-38.

References
[I] J. R. Schreiffer Theory of Superconductivity (Benjamin/Cummings,
Reading, 1964).
[2] For a review, see J. M. Blatt Theory of Superconductivity (Academic,
New York, 1964) and references therein.
[3] C. N. Yang, Rev. Mod. Phys. 34, 694 (1962).
[4] D. M. Eagles, Phys. Rev. 186, 456 (1969).
[5] A. J. Leggett in Modern Trends in the Theory of Condensed Matter,
A. Pekalski and R. Przystawa, eds. (Springer-Verlag, Berlin, 1980);
and J. Phys. (Paris) Colloq. 41, C7-19 (1980).
[6] N. D. Mermin and P. Muzikar, Phys. Rev. 21, 980 (1980).
[7] M. G. McClure, D. Phil, thesis, University of Sussex (1981).
[8] K. Miyake, Prog. Theor. Phys. 69, 1794 (1983).
[9] P. Nozieres and S. Schmitt-Rink, J. Low Temp. Phys. 59, 195 (1985).
[10] See C. Comte and P. Nozieres, J. Phys. (Paris) 43, 1069, 1083 (1982);
H. Haug and S. Schmitt-Rink, Prog. Quant. Elect. 9, 3 (1984); and
references therein.
[II] J. Bednorz and K.A. Muller, Z. Phys. B 64, 189 (1986); M.K. Wu et
ah, Phys. Rev. Lett. 58, 908 (1987).
[12] See High Temperature Superconductivity, Proceedings of the Los
Alamos Conference, K. Bedell et al, eds. (Addison-Wesley, Redwood
City, Calif., 1990).
[13] M. Randeria, J. Duan, and L. Shieh, Phys. Rev. Lett. 62, 981 (1989);
Phys. Rev. 41, 327 (1990).
[14] M. Randeria, N. Trivedi, A. Moreo, and R. T. Scalettar, Phys. Rev.
Lett. 69, 2001 (1992) and in preparation.
[15] A. S. Alexandrov and J. Ranninger, Phys. Rev. 23, 1796 (1981); S.
Robaszkiewicz, R. Micnas, and K. A. Chao, Phys. Rev. 23, 1447
(1981).
[16] R. Micnas, J. Ranninger, and S. Robaszkiewicz, Rev. Mod. Phys.
62, 113 (1990) and references therein.
[17] See the reviews by H. Capellmann, and by T. Moriya in Metallic
Magnetism, H. Capellmann, ed. (Springer, Berlin, 1986); G. G.
Lonzarich and L. Taillefer, J. Phys. C 43, 4339, 1083 (1985).
[18] J. R. Schrieffer, X. G. Wen, and S. C. Zhang, Phys. Rev. B 39, 11663
(1988).
390 M. Randeria
[19] M. P. A. Fisher, G. Grinstein, and S. M. Girvin Phys. Rev. Lett.
64, 587 (1990); A. F. Hebard and M. Paalanen, Phys. Rev. Lett. 69,
1604 (1992).
[20] K. Moulopoulos and N. W. Ashcroft, Phys. Rev. Lett. 66, 2915
(1991).
[21] M. Schmidt, G. Ropke, and H. Schulz, Ann. Phys. 202, 57 (1990);
T. Aim, B. L. Friman, G. Ropke, and H. Schulz, Nucl. Phys. A 551,
45 (1993); see also G. Ropke, this volume.
[22] C. A. R. Sa de Melo, M. Randeria, and J. R. Engelbrecht, Phys.
Rev. Lett. 71, 3202 (1993).
[23] J. R. Engelbrecht, C. A. R. Sa de Melo, and M. Randeria (preprint,
1993).
[24] V. N. Popov, Functional Integrals and Collective Excitations (Cam-
bridge University Press, 1987); for a nice introduction and references,
see V. Ambegaokar in Percolation, Localization and Superconductiv-
ity, A. M. Goldman and S. A. Wolf, eds. (Plenum, New York,
1984).
[25] For a more careful derivation, see M. Randeria, J. Duan and L.
Shieh, Phys. Rev. 41, 327 (1990).
[26] See, e.g., E. M. Lifshitz and L. P. Pitaevskii, Statistical Physics, Part
I (Pergamon, Oxford, 1980), Section 77.
[27] M. Drechsler and W. Zwerger, Ann. Physik 1, 15 (1992); W. Zwerger,
(preprint 1993).
[28] R. Haussmann, Z. Phys. B 91, 291 (1993), and private communica-
tion.
[29] B. L. Gyorffy, J. B. Staunton, and G. M. Stocks Phys. Rev. B 44,
5190 (1991).
[30] S. Schmitt-Rink, C M . Varma and A. Ruckenstein, Phys. Rev. Lett.
63, 445 (1989); J. Serene, Phys. Rev. B 40, 10873 (1989); A. Toku-
mitu, K. Miyake, and K. Yamada, Prog. Theor. Phys. Suppl. 106,
63 (1992); Phys. Rev. B 47, 11988 (1993).
[31] P.J. H. Denteneer, G. An, and J. M. J. van Leeuwen, Europhys. Lett.
16, 5 (1991).
[32] E. Abrahams and T. Tsuneto, Phys. Rev. 152, 416 (1966); see also
M. Cyrot, Rep. Prog. Phys. 36, 103 (1973) and references therein.
[33] H. Ebisawa and H. Fukuyama, Prog. Theor. Phys. 46, 1042 (1971).
[34] Even in a purely bosonic system the dynamics of the order parameter
will be damped sufficiently close to Tc (see P. Hohenberg and B.
I. Halperin, Rev. Mod. Phys. 49, 435 (1977), section VI E). Our
 Crossover from BCS Theory to BEC 391
analysis, however, is restricted to be valid only outside the critical
region.
[35] L. Belkhir and M. Randeria, Phys. Rev. B 45, 5087 (1992); Phys.
Rev. B 49, 6829 (1994).
[36] See, e.g., A. J. Leggett in Quantum Liquids, J. Ruvalds and T. Regge,
eds. (North Holland, Amsterdam, 1978), p. 167.
[37] P. W. Anderson, Phys. Rev. 112, 1900 (1958).
[38] J. Sofo, C. A. Balseiro, and H. E. Castillo, Phys. Rev. B 45, 9860
(1992); T. Kostryko and R. Micnas, Phys. Rev. B 46, 11025 (1993);
A. S. Alexandrov and S. G. Rubin, Phys. Rev. B 47, 5141 (1993).
[39] R. Cote and A. Griffin, Phys. Rev. B 48, 10404 (1993).
[40] S. Schmitt-Rink, D. S. Chemla, and H. Haug, Phys. Rev. B 37, 941,
(1988); R. Cote and A. Griffin, Phys. Rev. B 37, 4539 (1988).
[41] H. Monien and K. Bedell, Phys. Rev. B 45, 3164 (1992); P.J. H.
Denteneer and J. M. J. van Leeuwen, Europhys. Lett. 22, 413 (1993).
[42] For a review and references, see P. C. Martin in Superconductivity,
R. D. Parks, ed. (M. Dekker, New York, 1969), Part 1, p. 371.
[43] A. Bardasis and J. R. Schrieffer, Phys. Rev. 121, 1050 (1961).
[44] L. Foldy, Phys. Rev. 125, 2208 (1962).
[45] See the review by D. J. Scalapino in High Temperature Supercon-
ductivity, Proceedings of the Los Alamos Conference, K. Bedell et al,
eds. (Addison-Wesley, Redwood City, Calif., 1990).
[46] R. T. Scalettar et al, Phys. Rev. Lett. 62, 1407 (1989).
[47] A. Moreo and D. J. Scalapino, Phys. Rev. Lett. 66, 946 (1991).
[48] J. E. Hirsch, Phys. Rev. B 28, 4059 (1983).
[49] W. W. Warren et al, Phys. Rev. Lett. 62, 1193 (1989); R. E. Walstedt
et al, Phys. Rev. B 41, 9574 (1990).
[50] M. Takigawa et al, Phys. Rev. B 43, 247 (1991).
[51] H. Alloul, T. Ohno, and P. Mendels, Phys. Rev. Lett. 63, 1700 (1989).
[52] H. Zimmerman et al, Physica C 159, 681 (1989); R. E. Walstedt, R.
F. Bell and D. B. Mitzi, Phys. Rev. B 44, 7760 (1991).
[53] J. Orenstein et al, Phys. Rev. B 42, 6342 (1990); Z. Schlesinger et
al, Phys. Rev. Lett. 67, 2741 (1991); C. C. Homes et al, Phys. Rev.
Lett. 71, 1645 (1993).
[54] R. Liu et al, Phys. Rev. B 46, 11056 (1992).
[55] In 2D where the Kosterlitz-Thouless Tc is much lower than Tc°, it
is interesting to ask if deviations from Fermi liquid behavior persist
even as U —> 0. Finite size errors, which are more pronounced at
small U, make it difficult to address this question via Monte Carlo
calculations.
392 M. Randeria
[56] B. S. Shastry, Phys. Rev. Lett. 63, 1288 (1989); A. J. Millis, H.
Monien, and D. Pines, Phys. Rev. B 42, 167 (1990); N. Bulut. D.
Hone, D. J. Scalapino and N. E. Bickers, Phys. Rev. B 41, 1797
(1990).
[57] A. Moreo, D. J. Scalapino, and S. R. White, Phys. Rev. B 45, 7544
(1992); A. Moreo, N. Trivedi, and M. Randeria (unpublished).
 16
Bose-Einstein Condensation of
Bipolarons in High-Tc Superconductors
J. Ranninger
Centre de Recherches sur les Tres Basses Temperatures,
CNRS, BP 166
38042 Grenoble-Cedex 9
France

Abstract

The short coherence lengths and the low carrier concentrations in high
temperature superconductors (HTCSC) quite naturally favor a scenario
where preformed bound electron pairs undergo a Bose-Einstein conden-
sation (BEC). There are numerous experimental indications in the normal
state properties (anomalous diamagnetic susceptibility, NMR spin-lattice
relaxation, Knight shift and crystal electric field transitions, as well as
entropy non-linear in T and strong local lattice deformations seen by EX-
AFS), all of which indicate a characteristic temperature TPB associated
with the breaking of such preformed electron pairs and the existence of
a pseudogap above Tc. As the number of charge carriers is increased by
chemical doping, TPB decreases and the critical temperature Tc, where su-
perconductivity sets in, in this so-called "underdoped regime", increases.
The maximum value for Tc is reached when Tc < Tp#. Upon further dop-
ing, going into the so-called "overdoped regime", Tc decreases. We take this
as an indication that the superconducting state depends on the existence of
preformed electron pairs. The normal state properties of the "underdoped
regime" seem to show significant differences from Fermi liquid behavior,
while in the "overdoped regime" they seem to be typical of ordinary metals.
In order to capture this empirically established scenario in HTCSC, we
discuss a phenomenological model based on a mixture of localized bosons
(bound electron pairs such as bipolarons) and itinerant fermions (elec-
trons), with a local exchange between bosons and fermion pairs. Super-
conductivity in such a system materializes when, due to the boson-fermion
exchange mechanism, a BCS-like gap opens up in the electron spectrum. At
the same time, the initially localized bosons acquire itinerancy and condense
inside the energy gap of the fermions. The dependence of Tc on the total

393
394 J. Ranninger
number of carriers is reminiscent of the BEC temperature of lattice bosons,
with an effective mass determined by the boson-fermion pair exchange term.
Above Tc, the bosons behave as resonant states inside the Fermi sea, which
is manifest by the appearance of a pseudogap and a blocking of the chem-
ical potential close to the bosonic level

1 Introduction
Since the experimental discovery of the first superconducting material
(Hg) by Kamerlingh Onnes in 1911 [1], progress in producing super-
conducting materials with critical temperatures higher than about 20 K
stagnated until the discovery of the high Tc superconductors by Bednorz
and Muller and Chu et al. [2].
As far as the classical (pre-high Tc superconductors) are concerned,
they could be described by the BCS theory [3] based on the idea of
pairing of electrons in k-space - called Cooper pairs. Within the weak
coupling limit of this theory, the critical temperature Tc is given by:
Tc ~ 1.13co0e~1Aeff, (1.1)
where Aeff = \i — K \£ denotes the effective Coulomb repulsion and
X is the phonon-mediated electron-electron attraction. The prefactor
in the expression for Tc being proportional to a>o - the characteristic
phonon frequency of the material - perfectly describes the isotope effect
Tc oc M" 1 / 2 (M being the ionic mass), thus confirming the phonon-
induced origin of electron pairing.
It was quickly understood [4] that, on the basis of the weak coupling
theory of superconductivity, Tc could not be expected to exceed 30-40 K.
The reason for this lies in the interdependence of the parameters entering
this theory. The effective attraction given by Aeff comes about via a
subtle decrease in efficiency of the Coulomb repulsion \i —> /i* due to
retardation effects. In fact, the quantity

li-X = 4ne2N(eF)/(q2e(0,q)) (1.2)

is generally positive; N(eF) denotes the density-of-states at the Fermi level
and e(0,q) is the dielectric constant for zero frequency and momentum
q. The average (...) over q is taken over the Fermi surface of the
material. Due to the retardation of the phonon-mediated electron-
electron interaction, \i is reduced to fi* = fi/(l + filn(eF/coo)), which
provides the possibility of ja* — X < 0 and hence makes electron-electron
pairing possible.
 BEC of Bipolarons in High-Tc Superconductors 395
At first sight Tc, given by expression (1.1), could be increased by
increasing the characteristic phonon frequency coo- Since, however,
X = \v\2N(eF), where the electron-phonon coupling \v\2 ~ CDQ2, this
would effectively lead to a decrease rather than an increase of Tc. Like-
wise an increase of the density of states N(eF) - at first sight favorable
for increasing Tc - renders the system unstable via global lattice transfor-
mations before Tc can be substantially increased. Increasing the effective
electron-phonon coupling \v\2 may indeed be the only possibility of in-
creasing Tc - if we remain within the electron-phonon-mechanism of
superconductivity.
These inherent limitations of achieving values of Tc much higher than
30-40 K within the picture of weak coupling electron-phonon mediated
superconductivity has led a large number of theorists to look for pairing
mechanisms outside the scope of electron-phonon interaction in order
to describe the superconductivity in HTCSC. There may be no need for
this, either on theoretical grounds or on the basis of experimental facts
which we have up to now.
In principle, strong electron-phonon coupling can give values of Tc
much higher than 30-40 K. This reasoning is based on extensions of the
weak-coupling BCS theory to its strong-coupling formulation in terms
of the Eliashberg theory [5]. This theory is a self-consistent theory, in
principle valid for arbitrarily high values of electron-phonon coupling
but assuming the validity of the adiabatic approximation (or the Migdal
hypothesis), implying ksTc/ep <C 1 and a>o/eF < 1, where eF denotes
the Fermi energy. As we first showed [6], this hypothesis is violated for
A > 1 when the lattice becomes locally (rather than globally) unstable
and electrons become heavily dressed particles - so-called small polarons.
Such small polarons tend to attract each other and form locally bound
pairs, called bipolarons, which are real-space pairs, in contrast to Cooper
pairs. Superconductivity then results from a BEC of such bipolarons
(which are effectively bosons) in the sense of Schafroth superconductivity
[7].
The short coherence length £ (a few tens of Angstroms) and the
low concentration of charge carriers in the HTCSC, which makes £ only
slightly larger than the interparticle distance, favors such a picture of local
pair superconductivity in the form of a BEC. The experimental evidence
that such pairs really exist in the semiconducting parent compounds
of HTCSC is strong support - although not a proof - for a BEC of
bipolarons in such materials.
In Section 2, we shall review the experimental evidence for local
396 J. Ranninger
electron pairing in the semiconducting parent compounds of HTCSC .
These experiments strongly suggest that in the metallic phase of these
compounds, a charge transfer mechanism between bipolarons and pairs
of highly correlated electrons is at play. This leads us naturally to a
generic model of an interacting boson-fermion mixture for HTCSC. The
qualitative feature of such a model will be discussed in Section 3, where
we shall show under what circumstances a BEC of bipolarons could
possibly be realized in HTCSC. The compatibility of such a picture with
a number of experiments in these materials is discussed in Section 4.

2 Experimental Evidence for Bipolarons in HTCSC

Tightly bound electron pairs on small molecular clusters making up
crystalline lattices have been experimentally verified and studied in great
detail over the last 20 years. The most striking examples are Ti4_xVxO7
[8], NaxV2O5 [9] and WO3_X [10]. In Ti4_xVxO7, diatomic molecular
units occur in the form of Ti 3+ -Ti 3+ and Ti 4+ -Ti 4+ , each Ti being
surrounded by an oxygen octahedral ligand environment. The extra two
electrons sitting on Ti 3+ -Ti 3+ units are tightly bound, which is manifest
from the much shortened (~ 10%) intramolecular distance for those units
as compared to the Ti 4+ -Ti 4+ molecular units. The electron pairs located
on the Ti 3+ -Ti 3+ units have been called bipolarons because of their
lattice-induced pairing mechanism. In Ti4O7, at low temperature, Ti 3+ -
Ti3+ pairs alternate with Ti4+-Ti4+ pairs, thus forming a semiconducting
charge ordered state of bipolarons. Upon increasing the temperature (to
about 140 K), these bipolarons become dynamically disordered, showing
a thermally activated resistivity and a diamagnetic susceptibility. At still
higher temperatures (at about 150 K), a first-order phase transition leads
to a metallic state of single (unbound) electrons characterized by a Pauli
susceptibility. Upon doping Ti4C>7 with V (Ti4_xVxO7), the temperature
region of dynamically disordered bipolarons can be sizeably enlarged to
the detriment of the charge ordered state. In NaxV2C>5 and WO3_X, the
situation is very similar, involving V 4+-V4+, V5+-V5+ and W5+-W5+,
W 6+ -W 6+ molecular units respectively.
These experimental results led us early on [6] to investigate the pos-
sibility of bipolaronic superconductivity as a concrete example for a
manifestation of a Schafroth superconductivity which could potentially
be realized in systems with strong electron-lattice coupling. In princi-
ple it is conceivable that the above-mentioned materials show bipolaronic
superconductivity at sufficiently low temperatures provided that the bipo-
 BEC of Bipolarons in High-Tc Superconductors 397
laronic motion occurs via tunneling rather than thermal activation. One
of the biggest hindering factors for BEC to occur in these materials is
related to doping the system away from its charge ordered state. This has
been done so far by ion substitution (Ti —• V) on sites where bipolaron
formation normally occurs. This evidently introduces strong impurity
effects in the bipolaronic system, hindering the onset of BEC of the
bipolarons.
Materials where bipolaronic superconductivity could occur should be
sought in systems where the concentration of bipolarons can be varied
over a wide regime without disturbing the subsystem in which the bipo-
larons move, i.e. by doping outside the bipolaronic subsystem. Such
a situation might be approached in the cuprate HTCSC and possibly
Bai_xKxBiC>3 and the fullerenes.
In the following, we shall focus on hole-doped cuprate HTCSC, which
consist of CuC>2 planes and adjacent layers containing so-called apical
oxygens. The electrons in the CuC>2 layers are highly correlated due
to strong on-site Coulomb repulsion on Cu atoms and, moreover, are
strongly hybridized between the Cu 3dx2_y2 and O 2px>>, orbitals. For
the low doped semiconducting parent compounds of HTCSC such as
YBa2Cu3C>6+x, Bi2_:cSr2Ca2Cu30io+>;, La2_xSrxCuO4 ..., this leads to a
charge transfer gap. Doping HTCSC by ion implantation occurs exclu-
sively outside the correlated electron system and sensitively affects the
apical oxygen network, on which it leads to the formation of bipolarons.
Experimental evidence for this comes from resonant Raman scattering,
photo-luminescence and photo-absorption, as well as photo-modulation
studies.
For undoped parent compounds, resonant Raman scattering is ob-
served involving vibrations of the apex oxygens polarized orthogonally
to the CuC>2 planes. This Raman active mode (475 cm"1 in YBa2Cu3O6)
changes significantly upon low doping of the parent compound [11]; it
hardens (507 cm"1 in YBa2Cu3O6.i5). The resonant Raman scattering
becomes more pronounced and a large number of clearly detectable over-
tones appear when the incident frequency of the laser is in the resonant
region, i.e. around 1.16 eV. This energy being less than the charge transfer
gap (1.7 eV in YBa2Cu3O6) indicates that one is probing states inside
this charge transfer gap. Site selective 18O isotope doping [11] reveals
that the doping-induced 507 cm"1 modes are related to vibrations of
two O(4)-Cu(l)-O(4) units on either side of a dopant oxygen O(l) which
leads to a change of Cu+(1) to Cu++(1) and ultimately causes a change in
bond length of the O(4)-Cu(l)-O(4) unit, as well as a significant change
398 J. Ranninger
in its vibrational frequency. The distribution of spectral weights of the
overtones is characteristically different from those observed in strongly
anharmonic crystals, an unmistakable signature of polarons [11,12].
Photo-luminescence spectra carried out on the same samples [11]
indicate the presence of some narrow electronic band inside the charge
transfer gap. The observed photo-luminescence can be understood in
terms of electrons which have been excited across the charge transfer
gap into the conduction band and then recombine radiatively into in-gap
states of the O(4)-Cu(l)-O(4) units. This necessarily involves a charge
transfer between the CuC>2 planes and the apex oxygen network. Evidence
that a direct link exists between dramatic doping-induced changes in the
vibrational frequency (involving the apex oxygens) and the associated
charge excitations comes from photo-induced infrared absorption and
photo-modulation studies [12]. There, additional charge carriers are
created in the semiconducting parent compounds by laser illumination.
Such photo-induced charge carriers lead to the suppression of modes
which characterize the semiconducting phase and to the appearance
of modes which normally exist only in the metallic compounds [13].
This is a strong indication of the existence of polaronic charge carriers
in the metallic compounds. Moreover a broad spectrum of electronic
origin appears in the frequency regime of the charge transfer gap in
the insulating compounds. The intensities of the infrared absorption of
the newly created phonon mode and of the near infrared absorption of
the electronic excitation scale in the same way with the strength of the
laser illumination [12]. Since this scaling varies as the square root of the
laser power, this implies recombination processes involving two charge
carriers and hence provides evidence for bipolarons as stable entities.
These optical experiments are a clear proof of strong coupling of charge
carriers to the displacement of the apex oxygens polarized orthogonally
to the CuC>2 planes and bipolaron formation involving adjacent apex
oxygen units.
Although all these optical experiments were carried out exclusively on
the semiconducting parent compounds, they are nevertheless strongly in-
dicative of the physics which we can expect for the metallic compounds.
Upon increasing the chemical doping, the semiconducting state breaks
down, resulting in a metallic state triggered by intrinsic charge fluctua-
tions between the CuC>2 planes and the apex oxygen network. The type
of optical experiments performed on the semiconducting parent com-
pounds are such that they test these intrinsic charge fluctuations in an
out-of-equilibrium situation. This is manifest in:
 BEC of Bipolarons in High-Tc Superconductors 399
(i) the strong resonant enhanced Raman scattering of the apex oxygen
mode orthogonal to the C11O2 plane,
(ii) the effect of photo-induced carriers leading to the creation of phonon
modes in the insulating parent compounds which are characteristic of
the metallic compounds,
(iii) the strong correlation between these photo-induced modes and the
appearance of electronic excitations in the charge transfer gap,
(iv) the possibility of inducing a metallic (superconducting) state upon in-
tense photo-induced doping of the semiconducting parent compounds
[14].
The broad spectrum of electronic origin appearing in the frequency
regime of the charge transfer gap upon photo-doping finds its counterpart
in the metallic compounds in the strong non-Drude behavior of the
optical conductivity showing a similarly broad mid-and near-infrared
absorption band [15].
Charge fluctuations between the CuO2 layers and the apex oxygen
network in the metallic compounds lead to dynamical fluctuations of
the interatomic distance between the Cu ions in the CuC>2 planes and
the adjacent layers containing the apex oxygen atoms. Such effects
have been seen by EXAFS in a great variety of high Tc materials in
the metallic compounds [16,17]. It should be noted that in metallic
YBa2Cu3O6+x> these distance fluctuations are linked to bond fluctuations
of the O(4)-Cu(l)-O(4) units varying between 1.80 and 1.94 A for the
Cu(l)-O(4) distance [16]. These two distances correspond to those of
undoped YBa2Cu3C>6 consisting of O(4)-Cu+(1)-O(4) units and those of
low doped insulating YBa2Cu3O6+x, with oxygens intercalated between
two such units and with Cu+(1) being replaced by Cu++(1). In the optical
experiments discussed above, we have pointed out the dramatic change
in vibrational frequency associated with such bond length changes.
From the detailed analysis of the experiments discussed above, we
arrive at the following picture. For low doped materials, pairs of local-
ized charge carriers are created which remain tied to small molecular
units containing the apex oxygens as the essential ingredients. It is an
experimental fact that upon exceeding a certain rate of chemical doping,
the sites on which bipolarons form rearrange themselves and order along
chains of very long length [18]. Such a rearrangement of a fixed number
of bipolaronic sites (or doping atoms) forces the system to activate a
charge transfer from the CuC>2 planes towards the dopant atoms in order
to bind those dopant atoms in their new positions [19]. This is directly
400 J. Ranninger

Fig. 1. Schematic plot of a molecular cluster on which bipolaron<->electron pair

charge transfer takes place in metallic YBa2Cu3O6+x. The bipolarons are formed
on the complex involving the two O(4)-Cu(l)-O(4) (ii, i2) units on either side of
a dopant O(l) atom. The electrons reside in the Cu(2)-2O(3)-2O(4) units (ju ji)-

evident from the photo-emission data [20] which measure abrupt changes
of the number of Cu + + (1) atoms as a function of chemical doping and
also confirm equally abrupt increases of the overlap of electronic wave
functions between the Cu(2) ions in the CuC>2 planes and the O(4) apex
oxygens [21]. The reason for the ordering of the chemically doped atoms
leading to an homogeneous distribution of bipolaronic sites and the onset
of a metallic state must lie in the energy gained by the bipolarons due
to delocalization in the normal state and the possibility of achieveing a
macroscopic occupation of the q = 0 state in the superconducting state
due to a BEC of bipolarons. Judging from the EXAFS measurements,
this charge transfer is of a dynamical nature [16,17], leading to a situation
where a bipolaron on any arbitrary site of the atomic clusters involving
the apex oxygens can be exchanged by a pair of electrons in the adjacent
CuC>2 planes. This charge transfer can be described by:

(2.1)

where i\^ in (2.1) represent two adjacent molecular units on the apex
oxygen network and 7*1,2 two adjacent Cu(2)-2O(2)-2O(3) clusters in
the O1O2 planes (see Fig. 1, for a likely realization of this situation in
metallic YBa2Cu3O6+x). The bipolarons are expected to exist in the form
 BEC of Bipolarons in High-Tc Superconductors 401
of intersite bipolarons bfii2 = c+ cJ consisting of two polaronic charge
carriers:
cl = ct\ct)(x-xo))i, (2.2)

made of the bare charge carriers surrounded by strong local deformations

of the molecular units on which they are located (the O(4)-Cu(l)-O(4)
unit in YBa2Cu3C>6+x). The state \(f)(x — xo))i denotes the displaced
oscillator function characterizing the deformed molecular unit, where xo
measures the degree of deformation. Provided this deformation is very
local and large - as it usually is for polaronic charge carriers - ciG and
ci(J can be considered as commuting operators.

3 BEC in a Hybridized Boson-Fermion Mixture

On the basis of the above analysis of the experimental situation of
HTCSC, we now construct a generic model which we expect contains the
essential physics in those materials [22]. The number of charge carriers
being generally small (10 21 -10 22 cm" 3 ), we can safely neglect the spatial
overlap of different bipolarons. To a first approximation, this amounts to
replacing the intersite bipolarons by simple bosons defined on effective
cluster sites *f, such as the one depicted in Fig. 1. We can then write
the effective hybridization term in the form: (b^c^^c^ + h.c), where b+
corresponds to c^c^, and define a generic model to describe HTCSC in
form of the following interacting boson-fermion mixture hamiltonian:

(3.1)

The charge transfer between bipolarons and electron pairs has strength v
and the common chemical potential ensures conservation of charge car-
riers; Q denotes the normalization volume. The energy of the bipolaronic
level is given by A, which should roughly correspond to TPB-
We note that, already, the simplest case of a non-interacting boson-
fermion mixture (v = 0) leads to deviations from the BEC of a pure Bose
gas. To show this, we consider free bosons with a finite dispersion Ek =
^ + A — 2/z and free fermions with a dispersion ek = £% — n (see Fig. 2)
with £f = h2k2/2mF and £jf = H2k2/2mB, where mF(B) denotes the fermion
402 J. Ranninger

Fig. 2. Schematic plot of the fermion (ek) and boson (Ek) dispersion for the
non-interacting boson-fermion model.

(boson) mass. Using the standard three-dimensional density of states for

free bosons and fermions given by gF^ = • sf2mF^B,€1/2Q/2n2h3 = ge 1 / 2 ,
we have the following condition for the conservation of total particle
number:

mB
mF )fjT—*dee^n (e + A —
B 2\i) (3.2)

where no denotes the number of condensed bosons. For low T (i.e.

T <C fi) one finds:

(kBT)i (^ ) h(cp) 2no(T\ (3.3)

with (p = exp(P(A — 2/i)), h/iity) is a Riemann C function and \i = A/2

for T < TBEC- From (3.3), we note that there is a critical value of the
density of charge carriers (nc = 4g/3 x (A/2)3//2) below which BEC is not
possible. The BEC temperature TBEC is then determined by:

n = nc 2g (kBTBEc)2 — • (3.4)

In Fig. 3(a,b), we plot TBEc as a function of n and no(T) as a function

of T for n> nc. Thus we see that in the non-interacting boson-fermion
mixture, the fraction of condensed bosons is always bigger than that for
a simple free Bose gas due to the extra thermally excited bosons near
TBEC-
 BEC of Bipolarons in High-Tc Superconductors 403

(a)
Fig. 3. (a) TBEC as a function of the total carrier concentration n in the non-
interacting boson-fermion model, (b) The condensed fraction no(T) of bosons as
a function of temperature in the non-interacting boson-fermion model (continu-
ous line) and the pure Bose gas (dashed line).

We now return to the case of an interacting boson-fermion mixture

based on the Hamiltonian (3.1), but assuming from the outset itiner-
ant rather than localized bosons. We do this in anticipation of a fully
self-consistent treatment [23] of the Hamiltonian (3.1), where, as a re-
sult of the boson-fermion exchange term, the initially totally localized
bosons are renormalized into propagating - although heavily damped -
quasiparticles.
In order to describe the normal state properties of such a system, as
well as its transition towards a superconducting state, we must consider
the diagrams for the self-energy I*F(k,co) for the fermions (electrons in
the CuC>2 planes) and Zs(q,co) for the bosons (bipolarons in the apex
oxygen network) given in Fig. 4 and solve the coupled equations for the
fermion and boson Green's functions in a self-consistent way [23].
For an interacting Bose system, the chemical potential lies below the
unrenormalized energy spectrum of the bosons, i.e. Eo = A — 2/J, > 0.
Considering the boson and fermion self-energies in their simplest RPA
approximation, i.e. in terms of free fermion and boson Green's functions
we obtain:

k-q "7

(3.5)

For £o > 0 and T —• 0, is negligible in this expression for

404 J. Ranninger
The imaginary part of the fermion self-energy is then given by

r>(k, co) ~ 2nv2gF(E0 + fi- co)nF(E0 - co\ (3.6)

where gF(co) denotes the density of states of the fermions. The width
r>(k, co) is different from zero for Eo < co < EQ+II. The real part RF(k, CO)
of ZF(k,co) is obtained in the usual way via a Hilbert transformation
of FV(k,co). The renormalized spectrum cok of the fermions is then
determined by cok — e^ — RF(k9co\i) = 0, given by the set of intersections
of co — ek with RF(k, CO). AS long as Eo > 0, there always exists a wave
vector kF such that for k < kF, we have cok < 0 and for k > kF, we
have cok > 0. This determines the renormalized Fermi vector kF. As
Eo approaches zero, kF increases and a situation is finally reached for
which no kF can be found within the Brillouin zone for which cok>kF > 0.
This signals the breakdown of the Fermi liquid behavior, manifest in the
absence of a discontinuity of ^(cok).
The limit Eo —• 0 is subtle in the sense that riB(E q) in the expression
for £jr(k,a>) diverges for q —• 0. Rewriting riB(Eq) as:

nB(Eq) = nB(Eq) + SqO(no - nB(Eq)\ (3.7)

the dominant contribution to E/r(k, co) in the limit T -» 0 is determined

by:

. (3.8)

This yields the poles

]l + nov2 (3.9)
in the fermion Green's function and a corresponding Fermi distribution
function of the form:

The expressions (3.9) and (3.10) have exactly the BCS form for the
superconducting ground state but with a gap proportional to the number
of condensed bosons no.
Let us now turn to the properties of boson Green's function in the
same limiting case, i.e. T —• 0 and Eo —• 0. From the discussion of the
instability of the fermionic system, we know that it is triggered by the
macroscopic occupation of the q = 0 mode. The imaginary part of the
 BEC of Bipolarons in High-Tc Superconductors 405

(a) (b)

Fig. 4. Fermionic (a) and bosonic (b) self-energy diagrams in terms of fully
dressed self-consistent fermion and boson Green's functions.

boson self-energy for q = 0 is given by:

TB(q = 0,co) = 2nV- ] T tanh ^pekd(co - 2ek), (3.11)

which for co = 0 is identically zero, signaling the onset of undamped

bosonic excitations. This imposes the following constraint on the renor-
malized boson spectrum:

Eo = JW« = 0,co = 0) = ^P £ 1 tanh ^ , (3.12)

where RB(q,co) denotes the real part of the boson self-energy

Rewriting the sum over k in (3.12) as an integral over the density of
states of the fermions, we obtain:

Eo = * f" degF(e)—
)€ tanh he - |i). (3.13)
2 J I1 A

In the limit T -> 0, the integrand on the rhs of (3.13) diverges for e = \i
unless gF(e = ix) = 0. This implies the opening up of a gap in the one-
particle electron spectrum which is consistent with the conclusions drawn
earlier from the properties of the fermion Green's function, (3.6)-(3.9).
The arguments developed above clearly demonstrate the mutually
induced superconducting state in the boson-fermion mixture being due
to a BEC of the bosons which occurs with the simultaneous opening of
a gap in the excitation spectrum of the fermions. A self-consistent mean
field theory of this problem has previously been given [24,25].
A fully self-consistent treatment of this problem in the present scheme -
taking into account, in addition to the self-energy diagrams for the normal
phase (Fig. 4), all those including anomalous fermion and boson Green's
functions - will be discussed elsewhere [23]. The critical temperature
406 J. Ranninger

o.io

Fig. 5. Transition temperature Tc as a function of the total number of charge

carriers and for several values of the boson-fermion pair coupling v = v/W.

TC9 signaling the onset of superconductivity, is obtained as the limiting

value for // for a given particle concentration. Up to this density, there
exist undamped solutions of the self-consistently treated problem in the
normal state (involving only the self-energies illustrated in Fig. 4).
In Fig. 5 we plot Tc as a function of the total number of particles
n = 2nF + 2HB for various values of the dimensionless boson-fermion
pair coupling constant v = v/W; W denoting the fermion band width.
We notice that the density dependence of Tc(n) can be divided into two
limiting regimes. For low carrier concentrations n < ncr(v\ one obtains
BCS-like behavior where the electron-electron attraction comes about
by an exchange of virtual bosons - just as in BCS for virtual phonons.
For n > ncr(v), Tc(n) is similar to that obtained for the non-interacting
boson-fermion mixture (see Fig. 3(a)) and hence is essentially determined
by the BEC of the bosons in the system. The changeover from one to
the other regime occurs rather abruptly, leading to a sizeable increase
of Tc upon increasing n only very little. This fact clearly demonstrates
the interest in achieving a BEC in conection with efforts to obtain large
values of Tc. In the remainder of this section, we shall sketch the main
results of this treatment.
The fully self-consistent treatment of this problem shows the develop-
ment of a pseudogap in the normal state as the superconducting state
is approached upon lowering the temperature. At the same time, the
initially flat (<?-independent) distribution function for bosons at high
temperatures gets redistributed, shifting its intensity more and more to
smaller q values as T approaches Tc (see Fig. 6(a)). This is a precursor
 BEC of Bipolarons in High-Tc Superconductors 407

(a)

Fig. 6. Schematic plot of (a) the boson occupation number n?(T) and (b)
the fermion occupation number n£(T), for the fully self-consistently treated
interacting boson-fermion model.

effect of a BEC, similar to that seen in liquid 4He upon approaching Tx

from above.
The appearance of a pseudogap in the fermionic density of states
above Tc leads to a renormalization of the Fermi distribution function.
This shows non-Fermi liquid properties with a noticeable redistribu-
tion of spectral weight involving the entire Brillouin zone, as illustrated
schematically in Fig. 6(b). The wavevector kF for the Fermi subsystem,
and hence the number of fermions, varies only a little as a function of the
total number of carriers. This is due to the pinning of the chemical po-
tential in the vicinity of £o- On the contrary, the number of bosons varies
practically linearly with an increase in the total number of particles.
As concerns the superconducting state, the fully self-consistent calcu-
lation involving normal as well as anomalous fermion and boson Green's
functions yields an acoustic dispersion relation (in the absence of a long-
range Coulomb interaction) for the bosonic modes near T ~ 0, thus
signaling a superfluid phase.

4 Experimental Suggestions for Verification of a BEC in HTCSC

Because of the short coherence lengths (a few tens of A) and the low
carrier concentrations (1020-1021 cm"3), fluctuations are expected to play
a major role and it is therefore highly unlikely that a mean field theory
could be adequate to describe the superconducting transition in HTCSC.
Both the critical regime defined by £G = \TG — Tc\/Tc, where mean
field exponents cease to be valid (the Ginzburg criterion [26]), as well
 408 J. Ranninger
as the wider temperature regime, defined by £B = \TB — Tc\/Tc, where
a quantitative breakdown of mean field theory gradually sets in as
one approaches Tc (the Brout criterion [27]), are expected to be large:
C ~ 0.01 — 0.1 and ( B ~ 0.1 — 1 [28]. For these reasons alone, a theory
built on the assumption of preformed electron pairs (as part of the charge
carriers) which condense upon lowering the temperature, seems to be an
appropriate starting point to describe the high Tc phenomenon, quite
independent of any specific mechanism of binding.
Experimental evidence for such preformed electron pairs and for the
physical origin of their binding below a characteristic temperature T/>#
comes from:

(i) The comparative study of bipolaronic properties in the insulating and

metallic state, as discussed in Section 2.
(ii) Measurements of the magnetic susceptibility [29] (using SQUID and
torque measurements) showing strong diamagnetic contributions com-
ing from the charge carriers in the CuC>2 planes. These diamagnetic
contributions start to set in at temperatures T^a which are several
times the critical temperatures and, upon approaching TC9 outweigh
the van Vleck contribution close to Tc. This temperature-dependent
diamagnetism can be expressed in terms of a universal Curie-Weiss
law Xdia = — Cdm/(T — #dia) with Tc/0dia ~ 1, for a wide variety of
HTCSC. It can be excluded that this diamagnetism is related to super-
conducting fluctuations since in certain samples they can be seen up
to 30 times Tc. By increasing the number of carriers, Tdia diminishes,
thus reducing more and more the temperature region of observable
diamagnetism.
(iii) The dependence of the normal state entropy [30] as a function of
temperature shows strong non-linear behavior with an inflexion point
at a characteristic temperature TPB which strongly depends on the
total number of charge carriers. As the number of charge carriers
is increased, entering the so-called "overdoped regime", TPB tends to
Tc. For sufficiently high temperatures, the entropy can be reconciled
with that of itinerant fermions having small band width of the order
of 1000 K. It would be highly desirable to see to what extent Tdia
and TPB are correlated as a function of doping. The present status
of the experiments unfortunately does not permit such a quantitative
comparison.
(iv) Indirect evidence for the existence of preformed electron pairs comes
from the experimental verification of a pseudogap in the Fermi sea. As
 BEC of Bipolarons in High-Tc Superconductors 409
discussed in the previous section, such a pseudogap has to be expected
whenever long-lived electron pair states are present. Experimental
evidences for such a pseudogap include the temperature dependence
of the NMR Knight shift Ks, which appears to be dominated by
pseudogap transitions above as well as below Tc [31]. Similar findings
are obtained by inelastic neutron scattering techniques, measuring the
temperature dependence of the linewidth of the crystal electric field
transitions [32].

These experimental results strongly support the picture of a hybridized

boson-fermion mixture with bosons (bipolarons) breaking up into indi-
vidual fermions at TPB (~ 7dia?). It is universal behavior of HTCSC that
Tc increases linearly with the number of charge carriers [33] and varies
inversely with their mass [34] up to some critical concentration, beyond
which one enters the "overdoped regime" and Tc starts to decrease,
sometimes rapidly going to zero.
The "underdoped regime" shows a number of non-Fermi liquid prop-
erties such as: a violation of the Korringa law for NMR, a scaling of
1/TiT with x rather than x2 [36] (where T{ denotes the NMR relaxation
rate and x the magnetic susceptibility), a non-Drude behavior of the
optical conductivity and the absence of a T2 law for the resistivity at low
temperatures [35].
In contrast to the above, the "overdoped regime" seems to have
normal Fermi liquid properties and an entropy which varies linearly with
temperature for T > Tc, compatible with ordinary itinerant fermions
[30].
The fact that TPB (and seemingly also Tdia) tends to Tc as one ap-
proaches the "overdoped" metallic normal state could be an indication
that the maximum of Tc which can be obtained in HTCSC is limited by
the existence of preformed local pairs below TPB (~ Tdia)-
The study of correlation functions within the Bose-Einstein model is
presently not sufficiently advanced to verify the above-mentioned charac-
teristics. We have, however, information from a somewhat related model-
the negative U Hubbard model [25] - in which the generic transition
from weak coupling \U\/W <C 1 BCS-like behavior to strong coupling
\U\/W > 1 local pair superconducting behavior [37] has been fairly
well studied. Here U represents an on-site attraction between electrons
with opposite spin (in other words, the pair breaking energy ~ TPB)
and W denotes their band width. The maximum of Tc in this model
is close to TPB ~ U for a certain \U\/W of the order of unity. Larger
 410 J. Ranninger

values for \U\/W describe the "underdoped regime" with a normal state
between Tc and 7># which shows many of the anomalous normal state
properties [38], listed above, and which in many ways are reminiscent of
a purely bosonic system [39].
Concerning the physical foundations of the boson-fermion model, we
propose the following decisive experiments which would either verify or
disprove the relevance of this model:
(1) A verification of the disappearance of polaronic effects upon entering
the overdoped regime for T > Tc using resonant Raman spectroscopy,
photo-luminescence, photo-induced Raman measurements, the appear-
ance of polaron-induced local lattice modes and EXAFS, measuring
polaron-tunneling induced local fluctuations of intramolecular dis-
tances.
(2) A verification of the disappearance of the diamagnetic contribution to
the magnetic susceptibility (not due to superconducting fluctuations)
upon entering the "overdoped" regime. Tendencies suggesting this
disappearance have been observed [29].
(3) A detailed analysis of the doping dependence of the chemical potential
fi by photo-emission spectroscopy. According to the boson-fermion
model, \x is expected to be pinned by the presence of the bosonic
resonant state inside the Fermi sea. Our preliminary calculations on
this issue suggest a rather weak doping dependence of //, provided
the pairing energy A does not vary with doping. Experimentally, this
question is unsettled. There are experiments in which \i is practically
unchanged with doping [40] and those in which n behaves more
like doped semiconductors [41], showing rather significant changes
with doping. High resolution photo-emission experiments testing the
dependence of fi on the carrier concentration and on the temperature,
as well as the existence of a pseudogap, would be strong tests for the
existence of bosonic states (pre-existing electron pairs) above Tc.
Let us now turn to the properties of the superconducting state. Exper-
imentally there are a number of features [35] which clearly deviate from
ordinary BCS-like behavior, such as:
(1) Unusually strong effects of the superconducting state on the lattice
properties. Apart from the classical shift of phonon frequencies with
energies above and below the gap energy [42], one observes strongly
correlated motion of the molecular vibrations (seen by ion channeling
experiments [43]) and a freezing of the vibrational energy of individual
 BEC of Bipolarons in High-Tc Superconductors 411
atoms (seen by resonant neutron scattering [44]) which are triggered
by the onset of the superconducting state [45].
(2) An unusually large gap (2A/ks Tc ~ 5 — 8) which in practice does not
vary with temperature and whose signal fades away as one approaches
Tc from below. This is seen from tunneling spectroscopy [46], EELS
[47] and reflectivity measurements [48]. These lattice effects may be
of importance in limiting the fluctuations of the purely electronic
condensation.
(3) An essentially frequency-independent optical conductivity (for frequen-
cies less than the superconducting gap) which shows the same temper-
ature behavior as 1/TiT [49].
(4) Specific heat of the purely electronic contribution which scales sur-
prisingly well with that of 4 He over a wide temperature regime
0 < \T - Tc\/Tc < 0.2 [50], as shown in Fig. 7. The scaling of
the specific heat of HT CSC to that of liquid 4 He permits one to deter-
mine the number of "effective bosons" n& in such systems. The number
derived in this way compares well with the number of charge carriers
determined independently by Hall measurements [51]. Using the for-
mula for the Bose condensation temperature for quasi two-dimensional
systems,
u T2D+e _ h2nnd
B BEC l j
mab ln(2kB Tf^^lK)
permits one to derive the effective mass of the "effective bosons" mab
in the CuC>2 planes. Here mc denotes the "effective boson" mass and d
the size of the molar volume along the c direction. Upon inserting into
(4.1) the experimentally determined transition temperature for T^Q
and for nB the "effective boson" density derived from the comparison
of the specific heat of several HT CSC with that of liquid 4 He, one
obtains mab ~ 3-5 free electron masses if we assume mc/mab ~ 100.
This value agrees reasonably well with the effective masses deduced
from optical experiments [54]. Finally, using the expression for the
penetration depth of charged bosons

= ymabc2/16nnBe2 (4.2)

for the magnetic field orthogonal to the CuC>2 planes, and inserting
the values for mab and nB derived above, one obtains values such as
^H,C ~ 2500-5000 A which are quite compatible with those derived by
independent measurements [51,52].
412 J. Ranninger

40

Bi 2 S r 2 C a 2 C u 3 0 x ,T c = 107K
32 Y B a 2 C u 3 0 7 . 5 , T c = 91.9K
Y Ba 2 Cu 3 O 6 9 7 ) Tc = 92.5 K

1 24

Liquid He , T c = 2 . 1 7 K

O 16

0.6 0.8 1.0 1.2 1.4

TfTc

Fig. 7. Comparison of the experimentally determined specific heats of liquid

4
He and of YBa2Cu3O6+x and Bi2Sr2Ca2Cu3030io+x. The experimental points
denoted by +, A, o and • correspond to Refs. [51], [52] and [53] and [30]
respectively.

It is remarkable that reasonable estimates of the concentration and

mass of the charge carriers in HTCSC can be determined by simply
comparing the specific heat to that of liquid 4He and using the BEC
temperature as the critical temperature for superconductivity. At present
we do not have any understanding of why the specific heat of HTCSC
should fit that of liquid 4He over such a wide regime in temperature,
well beyond the point where universality and scaling can be expected.
Although this comparison itself is not a proof for a BEC in HTCSC,
it nevertheless is one more indication for the existence of preformed
electron pairs in these materials.
Recent angle-resolved photo-emission studies [55] measuring the super-
conducting gap and low T measurements of the penetration depth which
varies linearly with T [56] both appear to be indicative of d-wave symme-
try of the superconducting order parameter. The fact that d-wave pairing
rather than any other pairing should occur in HTCSC is a new feature
which should help to clarify the origin of the large values of Tc. It is clear
 BEC of Bipolarons in High-Tc Superconductors 413
that for systems with a short coherence length, the superconducting order
parameter must adapt to the local underlying crystal structure. On the
basis of an appropriate phenomenological model such as the extended
Hubbard model with on-site repulsion and nearest neighbor attraction
[57], one indeed obtains a whole series of superconducting ground states
with d,p and extended s-wave pairing depending on the concentration
of charge carriers. The boson-fermion model presented above would
also lead to symmetries of the gap which are not necessarily s-wave
pairing. We simply have to take into account in this model the precise
position (ii, i*2 in (2.1)) of bipolarons, stretched over adjacent sites, and
the induced pairing fluctuations of the electrons involving square config-
urations of four CuC>2 units (J\9 J2 in (2.1)) in the basal planes of the
cuprate HTCSC.

5 Conclusions
The origin of high Tc superconductivity is presently still a matter of
great controversy. Certainly, in these materials we are faced with strong
correlations between the electrons, which most theorists consider as
crucial to explain high Tc superconductivity in terms of a non-lattice-
driven mechanism. As we have stated, there is extensive experimental
evidence for strong coupling of electrons to certain local lattice modes
and consequent polaron formation. On the basis of these experiments,
we are led to a picture in which localized bipolarons form one of the
two subsystems of HTCSC, whereas the strongly correlated electrons
are confined to the other subsystem. A charge exchange, involving
bipolarons and electron pairs, is triggered when one exceeds a certain
doping concentration. At that point, an insulator-metal transition sets
in, monitored by the binding of extra dopant atoms in those materi-
als. The metallic regime consists of two apparently physically different,
regions:

(1) an "underdoped regime", in which Tc increases monotonically as a

function of the charge carrier concentration, and
(2) an "overdoped regime", in which Tc decreases with increased carrier
concentration.
Specific heat, susceptibility, reflectivity, NMR Knight shift and crystal
electric field excitations indicate that, in the "low doped regime", there
exists a pseudogap, which we interpret as being due to a bosonic reso-
414 J. Ranninger

nant state of bipolarons inside the Fermi sea. With increased doping, the
binding energy TPB of such a bipolaronic state decreases and eventually
coincides with the superconducting transition temperature Tc. At that
point, the maximum value of Tc is reached. Upon further doping, Tc
starts to decrease. This whole scenario has been described in terms of
a model of a boson-fermion mixture with an exchange of boson <-•
fermion pairs. Preliminary results of our theoretical investigation of this
model have been presented here [23].
Most of our theoretical arguments for developing the boson-fermion
scenario for HTCSC were based on experimental evidence coming from
the hole-doped cuprates. There is recent experimental evidence that
polaronic charge carriers also exist in electron-doped cuprates [58] as
well as in fullerenes [59]. This strengthens our conviction that the whole
high Tc superconductivity phenomenon has a unique mechanism, such
as the one described here.
Of course, for a satisfactory theory of high Tc superconductivity, the
problem of strong electron-phonon coupling has to be solved in con-
junction with the strong electron correlation problem. One of the key
questions to be addressed in this connection would be a careful study
of the correlation between the pseudogap exhibited in the electron spec-
trum and the pseudo spin gap seen in the magnetic susceptibility of
the insulating and superconducting cuprate HTCSC. The latter has been
extensively studied by inelastic neutron scattering [60] and NMR [61].
The microscopic origin of this spin gap and its evolution in the highly
doped superconducting samples is presently not understood. In princi-
ple, any formation of singlet pairs would produce such a pseudo spin
gap, whatever its origin might be: magnetic, purely correlation driven,
negative U-centers or bipolarons, etc. A prime question to answer
would be to check if a pseudogap in the electron spectrum exists in
the non-magnetic HTCSC. If so, then it would be conceivable that it is
the existence of such a pseudogap which is at the origin of the pseudo
spin gap in systems with magnetic correlations. Such a verification
would be decisive for our understanding of the origin of pairing in
HTCSC.

Acknowledgements. I would like to thank C. Berthier, C. di Castro,

D. Mihailovic, M. Miljak, Ph. Nozieres, J.M. Robin and N. Schopohl for
many valuable discussions, and T. Schneider for acquainting me with the
experimental results on the crystal field excitations and Knight shift data
in HTCSC.
 BEC of Bipolarons in High-Tc Superconductors 415
References
[I] H. Kamerlingh Onnes, Leiden Commun. 1206, 1226 (1911).
[2] J.G. Bednorz and K.A. Muller, Z. Phys. B 64, 189 (1986); C.W. Chu
et al, Phys. Rev. Lett. 58, 405 (1987).
[3] J. Bardeen, L. Cooper and R. Schrieffer, Phys. Rev. 108, 1175 (1957).
[4] For a review on this subject, see V.L. Ginzburg, Physica C 209, 1
(1993).
[5] G.M. Eliashberg, Sov. Phys. JETP 11, 696 (1960); D. Rainer, Prog.
Low Temp. Phys. 10, 371 (1986).
[6] A.S. Alexandrov and J. Ranninger, Phys. Rev. B 23, 1796 (1981);
ibid. 24, 1164 (1981); A.S. Alexandrov, J. Ranninger and S. Robasz-
kiewicz, Phys. Rev. B 33, 4526 (1986).
[7] M.R. Schafroth, Phys. Rev. 100, 463 (1955).
[8] B.K. Chakraverty and C. Schlenker, J. Physique 37, C4-353 (1976);
S. Lakkis et al, Phys. Rev. B 14, 1429 (1976).
[9] B.K. Chakraverty, M.J. Sienko and J. Bonnerot, Phys. Rev. B 17,
3781 (1978).
[10] O.F. Schirmer and E. Salje, J. Phys. C 13 L1067 (1980); R. Gehlig
and E. Salje, Phil. Mag. B 47, 229 (1983).
[II] V.N. Denisov et al, Phys. Rev. B 48, 16714 (1993).
[12] C. Taliani et al, in Electronic Properties of High Tc Superconductors
and Related Compounds, Springer Series of Solid State Physics 99,
H. Kuzmany, M. Mehring and J. Fink, eds. (Springer, Berlin,1990),
p. 280.
[13] D. Mihailovic et al, Phys. Rev. B 42, 7989 (1990).
[14] V. Kudinov et al, Phys. Lett. A 151, 358 (1990); V.V. Eremenko et
al, Physica C 185, 961 (1991); G. Yu et al, Phys. Rev. B 45, 4964
(1992).
[15] J. Orenstein et al, in Electronic Properties of High Tc Supercon-
ductors and Related Compounds, Springer Series of Solid State
Physics 99, H. Kuzmany, M. Mehring and J. Fink, eds. (Springer,
Berlin, 1990), p. 254 ; K. Kamarao et al, ibid. p. 260.
[16] J. Mustre de Leon et al, in Lattice Effects in High Tc Superconduc-
tors, Y Bar Yam, T. Egami, J. Mustre de Leon and A.R. Bishop,
eds. (World Scientific, Singapore, 1992), p. 93.
[17] A. Bianconi et al, Lattice Effects in High Tc Superconductors, Y. Bar
Yam, T. Egami, J. Mustre de Leon and A.R. Bishop, eds. (World
Scientific, Singapore, 1992), p. 65.
416 J. Ranninger
[18] D. De Fontain et al , Nature 343, 544 (1991); J.D. Jorgensen, ibid.
349, 565 (1991).
[19] J. Ranninger, Physica Scripta T 39, 61 (1991).
[20] H. Tolentino et al, Physica C 192, 115 (1992); G. Uimin and J.
Rossat-Mignod, Physica C 199, 251 (1992).
[21] A. Bianconi, Proc. Int. Conf. on Superconductivity (Bangalore), S.K.
Joshi, C.N.R. Rao and S.V. Subramanian, eds. (World Scientific,
Singapore, 1990), p. 488.
[22] We do not discuss alternative mechanisms which have been proposed
for HTCSC.
[23] J.M. Robin, Ph.D. Thesis, University of Grenoble; J. Ranninger and
J.M. Robin, to be published.
[24] J. Ranninger and S. Robaszkiewicz, Physica B 135, 468 (1985); S.
Robaszkiewicz, R. Micnas and J. Ranninger, Phys. Rev. B 37, 180
(1987).
[25] R. Micnas, J. Ranninger and S. Robaszkiewicz, Rev. Mod. Phys. 62,
113 (1990).
[26] V.L. Ginzburg, Sov. Phys. Solid State 2, 18244 (1960).
[27] R. Brout, Phys. Rev. 118, 1009 (1960).
[28] A. Kapitulnik et al, Phys. Rev. B 37, 537 (1988).
[29] M. Miljak et al, Solid State Commun. 85, 519 (1993) and references
therein.
[30] J.M. Loram et al, Phys. Rev. Lett. 71, 1740 (1993).
[31] N. Winzek et al, Physica C 205, 45 (1993).
[32] R. Osborn and E.A. Goremychkin, Physica C 185, 1179 (1991).
[33] YJ. Uemura et al, Phys. Rev. Lett. 62, 2317 (1989).
[34] R. Tournier et al, J. Mag. Mag. Mat. 76, 552 (1988).
[35] P. Allen, Comments on Condensed Matter Physics 15, 327 (1992).
[36] R.E. Walstedt et al., Phys. Rev. B 41, 9574 (1990); ibid 44 7760
(1991); M. Horvatic et al, Physica C 185/9, 1139 (1991).
[37] Ph. Nozieres and S. Schmitt-Rink, J. Low Temp. Phys. 59, 195
(1985).
[38] M. Randeria et al, Phys. Rev. Lett. 69, 2001 (1992).
[39] A.S. Alexandrov and N. Mott, Supercond. Sci. Technol. 6, 215
(1993).
[40] J.W. Allen et al9 Phys. Rev. Lett. 64, 595 (1990); Rong Liu et al,
Phys. Rev. B 46, 11056(1992).
[41] M.A. van Veenendaal et al, Phys. Rev. B 47, 446 (1993).
[42] C. Thomsen et al, Sol. State Comm. 75, 219 (1990).
 BEC of Bipolarons in High-Tc Superconductors 411
[43] L.E. Rehn et al, in Lattice Effects in High Tc Superconductors, Y.
Bar Yam, T. Egami, J. Mustre de Leon and A.R. Bishop, eds. (World
Scientific, Singapore, 1992), p. 27 ; K. Yamaya et al, ibid p. 33.
[44] H.A. Mook et al, Phys. Rev. Lett. 65, 2712 (1990).
[45] For a review on anomalous lattice properties in HT CSC, see J.
Ranninger, Z. Phys. B 84, 167 (1991).
[46] J.R. Kirtley, Int. J. Mod. Phys. 4, 201 (1990).
[47] B.N. Persson and J.E. Demuth, Phys. Rev. B 42, 8057 (1990).
[48] L.D. Rotter et al, Phys. Rev. Lett. 67, 2741 (1991); C.C. Homes et
al, Phys. Rev. Lett. 71, 1645 (1993).
[49] R.T. Collins et al, Phys. Rev. B 43, 8701 (1991).
[50] A.S. Alexandrov and J. Ranninger, Sol. State Comm. 81, 403 (1992).
[51] W. Schnelle et al, Physica C 161, 123 (1989).
[52] W. Schnelle et al, Physica C 168, 465 (1990).
[53] M.B. Salamon et al, Physica C 168, 283 (1990).
[54] D. Mihailovic et al, Phys. Rev. B 42, 7989 (1990); Xiang-Xin Bi
and P. Ecklund, Phys. Rev. Lett. 70, 2625 (1993).
[55] Z.X. Shen et al, Phys. Rev. Lett. 70, 1553 (1993).
[56] W.N. Hardy et al, Phys. Rev. Lett. 70, 3999 (1993).
[57] R. Micnas et al, Phys. Rev. B 37, 9410 (1988).
[58] P. Calvani et al, to be published.
[59] M. Knupfer et al, Phys. Rev. B 47, 13944 (1993); V.N. Denisov et
al, Sov. Phys. JETP 75, 158 (1992); V.N. Denisov et al, Synthetic
Metals 55-57, 3050 (1993).
[60] J. Rossat-Mignod et al, Physica C 169, 158 (1991).
[61] C. Berthier et al, Physica C 185/9, 1141 (1991).
 17
The Bosonization Method in
Nuclear Physics
F. Iachello
Center for Theoretical Physics, Sloane Physics Laboratory
Yale University
New Haven, CT 06511-8167

Abstract

The bosonization method developed in nuclear physics in the last 20 years

is briefly reviewed.

1 Introduction

In the last 20 years, a considerable amount of work has gone into

the development of bosonization methods for highly correlated finite
fermion systems. This work has been stimulated by the phenomenological
successes of the interacting boson model introduced in 1974 in which
atomic nuclei composed of nucleons (fermions) are treated in terms of
an interacting system of bosons (fermion pairs). In this article the logic
scheme of the bosonization method in nuclear physics will be briefly
reviewed. The purpose here is to provide the basic references upon
which the review is built and sources where further references can be
found. Bosonization methods were introduced much earlier than 1974
in connection with infinite Fermi systems. The method developed in
nuclear physics has some similarities with those developed in other areas
of physics, but it also has major differences. In nuclear physics one
maps not only operators but also states and seeks a description not
only of the ground state but also of the entire excitation spectrum.
Furthermore, number projection plays a very important role contrary
to the case of infinite systems where number projection is not relevant.
These differences, and others, will be briefly remarked upon in this article.

418
 The Bosonization Method in Nuclear Physics 419

V(r)

/
, /
s
/

Fig. 1. The self-consistent potential V(r) in nuclei. The single-particle levels are
labelled by the spectroscopic notation / 7 .

2 Strongly Interacting Finite Fermion Systems

2.1 Fermions and the Shell Model
Quite often in physics one has to deal with a system of interacting
fermions. In some cases, this system is finite, i.e. with a number of
particles of the order of n ~ 100. The best examples of finite Fermi
systems are atomic nuclei, where n ~ 2 — 250. In first approximation, one
can imagine that the individual fermions (nucleons) move in the average
field generated by all others. The average (or mean) field can be obtained
by a Hartree-Fock (or similar) calculation. The mean field in nuclei
has the shape depicted in Fig. 1. Although the mean field provides a
description of some properties of the ground state of the system, it does
not help much in the study of the excitation spectrum, which is often
dominated by the correlations (residual interactions). This means that
one has to deal with a finite Fermi system whose Hamiltonian is

HF = E o (1)

where ep is the single particle energy and vpp> the residual interaction
between particles, p and pf. The Hamiltonian (1) describes a "shell-
model". Introducing a label i to characterize the single-particle states,
420 F. Iachello
one can write a second quantized Hamiltonian

HF = Eo + J ^ €ia[ai + ^2 vu'kk>a!4akak' • (2)

The label i for a finite system is i = v,/,s,7,ra, where v is the principal

quantum number in the self-consistent potential of Fig. 1, / the orbital
angular momentum, s = 1/2 the spin, and j and m the total angular
momentum and its z-component. The spectroscopic notation vlj is often
used, as in Fig. 1. The Hamiltonian (2) is obviously identical to that
of infinite Fermi systems except that in that case the label i = k, s, <r,
where k is the momentum, s = 1/2 the spin and o its z-component. The
excitation spectrum of the system is dictated by the residual interaction.
The behavior of the residual interaction in nuclei is shown schematically
in Fig. 2. Here the matrix elements

Wj = ((If7/2)2J I v | (If7/2)2J) (3)

for two identical particles occupying the single-particle level I/7/2 is

shown as a function of the total angular momentum J = 0,2,4,6. (In
contrast to atoms where the appropriate coupling scheme is L — S, in
nuclei the dominant coupling scheme is j — j.) As one can see from Fig.
2, the interaction is large and attractive when J = 0, is still attractive
(but not so large) when J = 2, and it is small or repulsive when J > 4.

2.2 Fermion Pairs

The properties of the residual interaction shown in Fig. 2 suggest an
approximation scheme to deal with the excitation spectra of nuclei. This
approximation scheme consists in introducing correlated fermion pairs.
The second quantized forms of the pair operators are

S]

(4)

if ' ^ '*
 The Bosonization Method in Nuclear Physics 421

l7
/2

(MeV)

-0.5

- 1.0 -

- 1.5 -

Fig. 2. Matrix elements Wj of the residual nucleon-nucleon interaction for two

identical particles occupying the single particle level I/7/2, as a function of the
total angular momentum, J, of the two particles.

Here the crosses denote angular momentum couplings,

m,m'

(5)
m,m'

m,m'

One may note that the angular momentum zero pairs, called 5 pairs, are
similar to the Cooper pairs of superconductivity

where rf is the creation operator for electrons. The others (D pairs, G

pairs,...) represent a generalization of the pairs of superconductivity to
larger angular momenta. The number of pairs is obviously
N = l, (7)
where, for the moment, n=even.
States of the interacting fermion system are constructed by acting with
pair operators on the vacuum. If only S -pairs are retained, there is only
422 F Iachello
one state with wave function

where Jf is a normalization factor. This state (called generalized seniority

zero state) corresponds to the ground state of BCS theory in which
the number of particles has been appropriately projected. It is the
generalization of the seniority states introduced by Racah [1], as discussed
by Talmi, and others [2]. The unprojected BCS ground state [3]

k
is, when written for the finite system,

j,m>0

Unfortunately, as shown by many authors, number projection plays a

crucial role in finite systems, since the fluctuations are of order ^Jn and
n ~ 100. BCS is thus not appropriate for finite systems.
The more and more pair states are retained, the more and more the
full Hilbert space of the original fermion system is recovered. If S and D
pairs are retained, the wave functions are of the type

(S)(D)|0),« d =0,...,N, (11)

which include (8) but are more general. If S,D and G pairs are retained,
one has
(S) (D) (G) | 0). (12)

It should be remarked that introduction of at least D pairs is crucial

in nuclear physics, in view of the attractive nature of the J = 2 pairing
(quadrupole pairing).
By introducing correlated pairs, one truncates the original fermionic
space F into a smaller space of finite dimensions, called B in Fig. 3.
The complete fermionic space contains, in addition to the correlated pair
states, states in which the correlated pairs are broken, for example, when
S and D pairs are retained, the complete space is
 The Bosonization Method in Nuclear Physics 423

Fig. 3. The truncation-mapping procedure. F is the fermionic space, B the pair

space and & the bosonic space.

The broken pair states are those containing explicitly the fermion op-
erators cftj. Of course, in constructing the space (13) one has to be
careful about overcompleteness and orthogonality (called the decoupling
problem) but this problem is important only for very small systems.

2.3 Mapping Fermions into Bosons

Once fermion pairs have been introduced, the next step is to map fermion
pair states into boson states [4]. Mapping is a one to one correspondence
in which to each fermion pair state there corresponds a boson state.
When S and D pairs are retained, the mapping is done through the
introduction of s and d boson operators, s,d^ (in = 0,±l,±2), satisfying
Bose commutation relations

=0. (14)

The fermion pair states (11) are mapped into boson states

_L (15)

Explicitly, the first few fermion pair states and their mappings are:
424 F. Iachello

Fermion pair space B Boson space B

n = 0, v = 0 |0) JV = O, nd = 0 | 0)

n = 2, v= 0 S f | 0) N = 1, nd = 0 s1" | 0)
v= 2 D t 10) Md = 1 dt | o)

n = 4, t> = 0 5 t 2 | 0) JV = 2, nd = 0 s*2 | 0)
i? = 2 S ^ | 0) nd = l s V | 0)
0) nd = 2 rft2 | o)

(16)
The mapping is such that the number of bosons N is half the number of
fermions
N -- n (17)
=
2
and the number of d-bosons is half the fermion generalized seniority
V
nd ==
• (18)
2
As mentioned above, some subtle problems arise in the construction of
the pair space, requiring the introduction of projection operators to take
into account the correct orthogonality properties of the states. These
projection operators are denoted by <PV in Eq. (16). The truncation and
subsequent mapping procedure is shown schematically in Fig. 3.

2.4 Hamiltonian (and Other) Operators

Mapping of states is not sufficient to determine the properties of the
system. The next step is to map operators. For example, in order to
compute the excitation spectrum, one needs to construct, starting from
the fermion Hamiltonian (2), the boson Hamiltonian

where b\ denotes generically a boson operator. Although in general the

mapping from fermions to bosons can be done for the entire system, in
practice only the valence shells are usually considered. The valence shells
of nuclei between particle numbers 50 and 82 are shown in Fig. 4.
In order to obtain the boson Hamiltonian (19) one performs two steps:
 The Bosonization Method in Nuclear Physics 425

Proton levels

(82 )

E (MeV) ȣf
0 - 3S
./2
2d
3/2

-1.67 - 2d
5/2

-348 -
^7/2

)
<»

Fig. 4. The valence shells of nuclei between nucleon numbers 50 and 82.

(i) Determine the pair structure constants a,-, /J//, which appear in the
correlated pairs, Eq. (4). Several methods have been suggested, the
simplest one being that of diagonalizing HF in the two-particle valence
space. This method gives the structure constants a,, fijf, and the
single-particle boson energies, eT.
(ii) Determine the boson-boson interaction uxx>m'. Again, several methods
have been suggested, the simplest one being that of equating matrix
elements of the fermion Hamiltonian, HF, in the fermion space of
the left-hand side of Eq. (16), with matrix elements of the boson
Hamiltonian, HB, in the boson space of the right-hand side of Eq.
(16), schematically written
(HF) = (HB) (20)

for four particles (two bosons). This method is referred to as OAI map-
ping [4]. The boson-boson interaction appears here as the contribution
of two terms, the direct and the exchange terms. The exchange term,
which is very important for finite systems, arises from the composite
nature of the bosons (Pauli principle).
At the end of this procedure one has replaced a strongly interacting finite
fermion system by a strongly interacting boson system. The bosonization
of a strongly interacting finite fermion system is also called the Interacting
Boson Approximation (IBA) and it leads to a model of nuclei called the
426 F. Iachello
Interacting Boson Model (IBM) [5]. Two remarks are in order here.
First, the method discussed here is a generalization of methods used in
superconductivity. Superconductivity, when written in the language of
Sections 2-5, corresponds to retaining only S pairs, to considering only
J=0 pairing interactions and to neglecting number projection, hereby
assuming that the structure constants of the S pairs are given by

In Eq. (21), 8j are the single fermion energies, X is the Fermi energy and
A the pairing gap.
The second remark is that the bosonization method can be applied
not only to particle-particle fermion operators, where the product eft eft
is replaced by a boson operator b\ but also to particle-hole fermion
operators [6], where the product eft a is replaced by a boson operator, tf.
The equivalent of the bosonization method then becomes, in the infinite
system, the theory of excitons.

3 Strongly Interacting Finite Boson Systems

3.1 Bosons and the Shell Model
The procedure of Section 2 replaces a strongly interacting fermion system
by a strongly interacting boson system with Hamiltonian

o O<G'

Here sa are the single-particle boson energies and the vaa< are the boson-
boson interactions. A second quantized version of (22) is given in (19)
where T labels the single particle boson states. It is convenient here to
use a spectroscopic notation T = s,d,g,.... The Hamiltonian (22) or (19)
is that of a "shell-model" for bosons. As mentioned in Section 2, in
nuclei, in first approximation, only s and d bosons are important. For
this reason, considerable effort has gone into the study and solution of a
system composed of s and d bosons over the last 20 years.
 The Bosonization Method in Nuclear Physics All
3.2 The s-d Boson System
The s-d boson system, usually called interacting boson model [5], is
characterized by the Hamiltonian

+ >^ C'T \<t x tf] x [2 x 21

J
Z—• L |_L J L Jo
L=0,2,4

x^l x \d x s] + [rf1" x 51"! x [2 x 2] I

J LJ L J L -• I A

{0) (0) (0)

+ v'o [ [ i t X i t ] X[SX 5] + [St X 5 t ] X [2 x

J L J
Jo

1
(0)
(23)
0
This Hamiltonian is the most general Hamiltonian invariant under rota-
tions up to two-body boson-boson interactions. Although in principle,
the bosonization procedure, starting from a fermion two-body Hamilto-
nian may produce a boson-boson Hamiltonian with three-body, four-
body, ... terms, in practice it has been found that the higher order terms
are small and therefore it is sufficient to consider the Hamiltonian (23).
One is then asked to find the eigenvalues of HB for a system of N bosons,
where N is the number of active pairs. Several methods have been used
to find the spectrum of (23):

(i) Matrix diagonalization

This is the direct numerical diagonalization of (23). At the present time

computer programs exist that diagonalize (23) up to a number of bosons
N ~ 30. For larger values, the matrices to diagonalize become very
large and a direct diagonalization is not possible. None-the-less, since
the number of active pairs in nuclei is usually less than 30, this method
has been used extensively and, in combination with the bosonization
method, has allowed one to compute spectra of nuclei, starting from the
effective nucleon-nucleon interaction, with accuracies better than 10%.
An example is shown in Fig. 5.
428 F. Iachello

E
(MeV)
56 B0 76
3 -

Exp Th

2 - 6 ^

\ ^j

1 -
°
2
-—I

2 2

0 - o o

Fig. 5. Ab initio calculation of the excitation spectrum of 132Ba using the bosoniza-
tion method. On the left the experimental spectrum, on the right the calculated
spectrum.

(ii) Lie algebraic methods

Bilinear products of boson creation and annihilation operators

(24)

generate a Lie algebra (S. If the index a goes from 1 to n, the Lie algebra
is u(n). For the s-d boson system in which the index a goes from 1
to 6, the Lie algebra is ^ = M(6). The six components are the single
component of the s-boson and the five components of the d-boson, d^
(fi = 0,±l,±2). The Hamiltonian HB can be written in terms of the
operators Ga/j, as

(25)

All the powerful techniques of group theory can then be used to find
the eigenvalues of HB. This is the celebrated "Algebraic Theory" which
has been used extensively in recent years to solve many-body problems.
Algebraic theory is presented in some detail in Ref. [5] and will not be
discussed further here.
 The Bosonization Method in Nuclear Physics 429

E
(MeV) 48Cd62 E«p. Th.

3- (n d ,0) <"d 1) (n d -2,0). 1) (n d -2,0).

2- "A-
4
0*-
2*—
e
4
~CL>
0*—
~2*- 0*— -a-_
'"
2*— 2*-

0- 0*— 0*— U(5) .

Fig. 6. An example of a spherical nucleus: 110Cd. On the left the experimental

spectrum, on the right the calculated spectrum. Both the experimental and calcu-
lated spectra are slightly anharmonic. These anharmonicities can be accounted
for by going beyond the simple approximation (26), as discussed in Ref. [5].

(in) The method of Bose condensates

This method is not used much in practice in nuclei. I will present it here
in view of its relation to the subject of Bose condensation discussed in
this volume, and in view of its interest for systems with N large. I will
begin with the trivial case in which there is no boson-boson interaction.
In this case, the ground state is the state in which all bosons occupy the
lowest level (s-level)(called spherical Bose condensate),

-F=(S^)N I 0) • (26)
y/N\
Excitations are obtained by removing one boson from the lowest level
and placing it into the d-level

0), (27)

where Jf, Jf' are renormalization constants. The corresponding spec-

trum is a harmonic spectrum with energies s, 2s,... where s = &d ~ £s-
Several nuclei have been found with a spectrum of this type, one of
which is shown in Fig. 6.
A more interesting case is that in which there is a boson-boson inter-
action. In this case one can introduce an operator
b\ = (*t + £ a^J), fi = 0, ±1, ±2, (28)
430 F. Iachello
which contains some variational parameters aM. The ground state of a
strongly interacting finite boson system is then given by
t
&f I °) ^ ( + E «A)N I 0>. (29)
called an intrinsic (or coherent) state [7]. Since the operator b\ contains
a mixture of all boson operators, it is called a deformed Bose condensate.
In the last few years, the theory of deformed Bose condensates has been
developed considerably, and used to study both the ground states of finite
Bose systems and their intrinsic excitations. In the particular case of s-d
systems, a frame of reference can be chosen such that the expectation
values of all operators do not depend on all five variables, a^, but only
on two of them, called intrinsic (or Bohr) variables, p,y. Starting from
the six spherical boson operators st,d+2>^+i'^o>^-i>^-2> o n e c a n ^ e n
introduce six other boson operators that depend continuously on /? and
y as [8]

p cos ydl + - U sin y{d\2 + / 2) +

b\ = ( cos ydl + -j= sin y{d\2 + / 2) - jSs1"I ,

b\ = —= cos y(d\y + d\) + sin ydl, (30)

The original basis can then be transformed into a deformed boson basis,
obtained by acting with the operators of (30) on a vacuum state

^)bl(p,y)---\O). (31)
The number operator, when written in terms of the new operators
becomes
N = b\bc + b\bfi + b]by + b\bx + b\by + b%. (32)
The ground state is the state of Eq. (29),
N
\0)=\N;P,y). (33)
 The Bosonization Method in Nuclear Physics 431

E
(MeV)

(24,0) (20,2) (16,4) (18,0) (24,0) (20,2) (16,4) (0,0)

8—

6*—

P
Fig. 7. An example of a deformed nucleus: 156Gd.

Excitations are obtained by removing one boson from the condensate

(34)

These excitations are called intrinsic excitations. The operators b\,fcj,b\

generate spurious modes instead, corresponding to rotations of the con-
densate. The variational parameters /?, y are determined by minimizing
the energy with respect to /? and y,
(N;P,y\H\N;f},y)
E(N;p,y) = (35)
(N;P,y\N;P,y) '
Several nuclei have been found whose excitation spectrum can be well
described by this method, one of which is shown in Fig. 7.
Some remarks are in order here. The method of Bose condensates
is different from the theory of superfluidity [9], although it shares some
similarities with it. When the depletion of the condensate is small, one has
NS~N, Nd~0. (36)
The Bogoliubov Hamiltonian for the s-d boson system is
HB = Ef0-esN + t^N(N-l) + ed(dt'd) + v'0N [df xrff+ 2 x . (37)
This Hamiltonian is obtained from (23) by replacing the operators s^
and s by c-numbers, y/N. This approximation is valid only for weakly
interacting Bose systems. In nuclei, the interactions are strong and the
approximation (37) is not appropriate. Another comment is that the
evaluation of the ground state energy (35), and of the excitation energies,
is good to leading order in N, the number of bosons. The method of
Bose condensates is thus particularly appropriate where N is large. In
432 F. Iachello
this respect it is often called l/N expansion, since corrections to the
leading order in N can be constructed in a systematic way.

4 Two (or More) Types of Particles

4.1 Fermions
In some cases, one has to deal with finite systems composed of two (or
more) types of fermions. Again the best example of such a system is
the atomic nucleus composed of protons and neutrons. To the extent
that protons and neutrons are assumed to have the same properties (a
good approximation from the point of view of the strong interactions)
one can consider the nucleus as if it is composed of nucleons (protons
and neutrons). This is the approximation used in Sections 2 and 3.
However, there are some properties of nuclei that depend explicitly
on the difference between protons and neutrons. For a more refined
treatment of nuclei, one has therefore to deal with a system composed
of two types of particles. The shell model for two types of particles is
no different from that with one type of particles. There are now single-
particle levels both for protons and neutrons, as shown in Fig. 8. The
bosonization method can be used without major modification for these
systems.

4.2 Fermion Pairs

One introduces pairs composed of two protons and two neutrons [10]:

j? Z> Jn -{a Jnxa Jn)o ,

where now the indices n and v denote protons and neutrons. One can
consider in principle many pair states, but in practice, in nuclei, only s
and d pairs are important.

4.3 Bosonization
The pair space is then mapped onto a boson space composed of proton
bosons and neutron bosons s\.,d\^ ; sJ,dJjAi . One then obtains the boson
 The Bosonization Method in Nuclear Physics 433

Fig. 8. The valence shells for protons and neutrons between particle numbers
50 and 82 (a), and the boson model that replaces the fermion problem after
truncation and mapping (b).

Hamiltonian corresponding to a given fermion Hamiltonian. The boson

Hamiltonian has the structure

Hv (39)

where Hn and Hv describe protons and neutrons, respectively, and Vnv

their interaction. The Hn and Hv have the structure of Eq. (23) with an
index n and v attached.

4.4 Methods of Solution

The eigenvalue problem for (39) can be solved using the same techniques
discussed in Section 3.

(i) Matrix diagonalization

This technique is straightforward except that the dimensions of the ma-

trices here are even larger. Computer programs exist for the s-d boson
system up to Nn ~ 15 and Nv ~ 15.

(ii) Lie algebraic methods

These techniques are particularly useful here in view of the complexity

of the problem. The bilinear products
434 F. Iachello
G$ = b{nb^, (40)
{
G :j = blbp,v,
generate a Lie algebra, which is the direct sum of u(6) algebras
<$ = un(6)@uv(6). (41)
The Hamiltonian, and other operators, can all be written in terms of
the operator in (41) and Lie algebraic methods used to find their matrix
elements and eigenvalues. (See Ref. [5] for details.)

(in) Bose condensates.

Methods of Bose condensation can also be used here. The ground state
of the (deformed) two-fluid system is written as
| g) = (Nn \NV l)-1/2(btjN-(bljN* | 0) (42)
and excitations are constructed.

5 Mixed Systems of Bosons and Fermions

Another important case that has been extensively investigated over the
last 15 years is that of mixed systems of bosons and fermions [11]. In
nuclei this situation appears when one wishes to treat systems with an
odd number of protons and/or neutrons. In these cases, at least one
particle must be unpaired. If the pairs are bosonized one ends up with a
mixed system of bosons and fermions. Also to this situation belong those
cases in which the correlated pairs are broken and one has explicitly
two or more fermions in addition to the bosons. The bosonization
method has been used extensively to deal with these situations. One
starts from the purely fermionic Hamiltonian and introduces correlated
pairs as in Section 2. The pair states are mapped into boson states and
a Hamiltonian is constructed of the type
H = HB+HF + VBF, (43)
where HB and HF describe the system of bosons and fermions, respec-
tively, and VBF their interaction. Particular care must be exerted here
in order to avoid overcounting. Also the boson-fermion interaction has
a rather complex form in order to take into account the composite na-
ture of the bosons. The boson-fermion interaction is illustrated by the
diagrams in Fig. 9.
 The Bosonization Method in Nuclear Physics 435

" 1 1' 1 1
s d d

r
> • •
> -

(a) (b) (c) (d) (e) (f)

Fig. 9. The boson-fermion interaction appearing in Eq. (43). This interaction

contains direct (d,e,f) and exchange (a,b,c) terms. The exchange terms arise
from the Pauli principle.

The eigenvalue problem for the Hamiltonian H of Eq. (43) is solved

by methods similar to those discussed previously.

(i) Matrix diagonalization

(ii) Lie algebraic methods

In this case one takes advantage of the fact that the bilinear products of
creation and annihilation operators for bosons and fermions generate a
graded Lie algebra composed of the set of operators of the type
G(B) = })\b

<%? = a]aj, (44)

Ft = a]b<,,

If n is the dimension of the bosonic space (6 for the s-d system) and m is
the dimension of the fermionic space, the (m + n)2 operators of Eq. (44)
generate the graded algebra u(n/m). The powerful machinery of group
theory is then used to help find solutions of the eigenvalue problem for
a mixed system of boson and fermions.

(in) Coherent state methods

The treatment of mixed systems of bosons and fermions using the co-
herent state method is more complex than the corresponding case of
bosons. The reason is that, while it is possible to introduce in the bosonic
436 F. Iachello
coherent state (29) variational parameters which are c-numbers, when
one deals with fermions such an introduction is not possible. The varia-
tional parameters become Grassman variables in the fermionic case. As
a result, the use of coherent state methods for fermions has not been
developed much.

6 Conclusions
In this article I have presented a brief survey of the bosonization method
in finite systems. The literature on this subject is very extensive and some
references have been indicated in Sections 2-5. Others can be found in
the reference list of Refs. [5] and [11] and in Ref. [12], from which most
of the material reviewed here has been taken. The applications of the
method have up to now been confined to the structure of atomic nuclei
where the method has prove to be very useful. The method can be used
for other finite Fermi systems. An area where, in principle, it could be
very useful is that of the newly discovered metallic clusters. These, too,
are finite Fermi systems, composed of valence electrons. The number of
valence electrons is of the same order of magnitude as the number of
nucleons in nuclei, ~ 100. The residual interaction in metallic clusters
may be different from that of nuclei and therefore the corresponding
boson system may not be an s-d system. None the less, the bosonization
method is sufficiently general that it can be applied to these situations as
well. Work in this direction is in progress [13].

Acknowledgments. This work was supported in part by DOE Grant No.

DE-FG02-91ER40608.

References
[1] C. Racah, Phys. Rev. 63, 367 (1943).
[2] Y.K. Gambhir, A. Rimini and T. Weber, Phys. Rev. 188, 1513
(1969); B. Lorazo, Nucl. Phys. A 153, 255 (1970); I. Talmi, Nucl.
Phys. A 172, 1 (1971). See also I. Talmi, Simple Models of Complex
Nuclei, (Harwood, Chur, 1993) and references therein.
[3] J. Bardeen, L.N. Cooper and J.R. Schrieffer, Phys. Rev. 108, 1175
(1957).
[4] T. Otsuka, A. Arima and F. Iachello, Nucl. Phys. A 309, 1 (1978).
[5] For a review see F. Iachello and A. Arima, The Interacting Boson
Model, (Cambridge University Press, 1987).
 The Bosonization Method in Nuclear Physics 437
[6] H. Feshbach and F. Iachello, Phys. Lett. 45B, 7 (1980).
[7] A. Bohr and B.R. Mottelson, Physica Scripta 22, 468 (1980); A.E.L.
Dieperink, O. Scholten and F. Iachello, Phys. Rev. Lett. 44, 1747
(1980); J. Ginocchio and M.W. Kirson, Phys. Rev. Lett. 44, 1744
(1980).
[8] A. Leviatan, Phys. Lett. 143B, 25 (1984).
[9] N.N. Bogoliubov, J. Phys. USSR 11, 23 (1947).
[10] A. Arima, T. Otsuka, F. Iachello and I. Talmi, Phys. Lett. 66B, 205
(1977); T. Otsuka, A. Arima, F. Iachello and I. Talmi, Phys. Lett.
76B, 139 (1978).
[11] For a review, see F. Iachello and P. van Isacker, The Interacting
Boson-Fermion Model, (Cambridge University Press, 1991).
[12] F. Iachello and I. Talmi, Rev. Mod. Phys. 59, 339 (1987).
[13] F. Iachello, E. Lipparini and A. Ventura, Lecture Notes in Physics
404, 318 (Springer, Berlin, 1992).
 18
Kaon Condensation in Dense Matter
Gerald E. Brown
Department of Physics
State University of New York at Stony Brook
Stony Brook, New York 11794
USA

Abstract
The K~-meson of mass m^ = 494 MeV is similar to an exciton, consisting
of a strange quark, as particle, and u antiquark, as hole, bound together to
make the kaon. In dense nuclear matter the kaon feels an attractive mean
field from the nucleons which lowers its energy appreciably. As soon as its
energy is brought down to ~ half of its rest mass, it becomes energetically
favorable in neutron stars to replace electrons by kaons, the neutron stars
becoming nuclear matter stars at higher density. The kaons form a zero
momentum Bose-Einstein condensate.
The new equation of state of dense matter, with the inclusion of kaon
condensation, is substantially softer than conventional ones, meaning the
maximum mass of compact objects formed in the collapse of large stars is
only ~ 1.5 M o . With this equation of state, black holes are easy to form
and it is estimated that there are ~ 109 of them in our galaxy. In this sense,
a large number of black holes is the "smoking gun" for kaon condensation.

1 Introduction

A Bose-Einstein condensation which is somewhat exotic, even from

the nuclear and particle physics point of view, is kaon condensation,
condensation of K~-mesons in a zero momentum state. Yet, aside from
the fact that we must think of energies and densities orders of magnitude
higher than in the case of the exciton gas, the K~ can be viewed as
analogous to an exciton and a system of X~-mesons can easily be viewed
as a gas of nearly free particles, because their interactions with each
other are known to be weak.

438
 Kaon Condensation in Dense Matter 439
The K~-meson has a mass of
mK = 494 MeV . (1)
The kaon is a boson, which will be important for our condensation
scenario. It is composed of a strange quark and a nonstrange u antiquark;
i.e., \K~) = |us). Whereas we have heard, throughout this conference,
about the breaking of symmetry, in order to establish a Bose-Einstein
condensate, we should realize that from the viewpoint of QCD we live
in a world of broken symmetry, broken chiral symmetry. One of the
consequences of this is that the negative energy sea of quarks can be
considered to be filled, at least up to some cutoff momentum at which
asymptotic freedom takes over. The filled negative energy sea of quarks
plays the role of the valence band in CU2O, which is much discussed in
this volume (see the articles by Wolfe et al. and Mysyrowicz.) The u is
then a hole in the valence band of up quarks, and the s is a particle in
the conduction band of strange quarks. The K~ can, then, be considered
to be an exciton, with the s-particle somewhat heavier than the u-hole.
We shall work at temperatures very small compared with m^, a
few MeV in magnitude, and will assume for our purposes here that
T = 0. Temperature dependent effects have not yet been calculated for
kaon condensation. Our Bose-Einstein condensate is then quite trivial, a
condensate at zero temperature of K~-mesons in the state of zero mo-
mentum. I will try to convince you that such a condensate is most likely
formed in the collapse of a large star, and that kaon condensates are
integral components of the compact objects, often referred to as neutron
stars, formed in such collapses. Obviously, the question to be answered
is how the interactions compensate the mass (1) of the kaon. It must be
energetically favorable to introduce kaons, before they will appear.
We will develop the theoretical framework in the next section, or at
least, discuss it. Before doing this, let us review what has been deduced
empirically about the K~-nucleus interaction, using this knowledge to
build the K~-nuclear matter interaction. The K~-nucleus interaction has
been obtained recently by Friedman, Gal and Batty [1], who described
the kaonic atoms; i.e., the atoms formed by a K~-meson and the nucleus,
for nuclei across the periodic table. In the case of thef X-nucleon
interaction, there is the well-known A(1405) resonance, which lies below
the KN continuum, and thus would be a bound state if its decay through
the (TC,A) or (TC,Z) channels would be turned off [2, 3]. As it is, the
t The notation is that the K, the antikaon, has two components, K~ and K°. Similarly
K = (K+,K°).
440 G. E. Brown
A(1405) is a bound state in the continuum. Because of this bound state
below the X~-nucleon scattering threshold, the low-energy K~-nucleon
scattering amplitudes are repulsive and the optical model potential (mean
field potential)
2mKVmf = -4nbp , (2)
where b is the scattering amplitude, is repulsive. (Kinematical factors are
suppressed here. They go out in the end.)
However, since the K and nucleon form a bound state, albeit in
the continuum, the basic interaction between K and nucleon must be
attractive.
Friedman et al. [1] argue that as the nuclear density is increased, the
K~-nucleus interaction must turn attractive. From fitting kaonic atoms
across the periodic table, they deduce that the K~ in the middle of the
nickel nucleus (i.e., roughly at nuclear matter density po) experiences an
attraction of
Vmf(po) * -200 MeV . (3)
The precise value of this number is not so certain, because the analysis
is difficult due to the fact that the kaon can disappear through reactions
such as
K~ + N -> 2T + n (4)
and others; i.e., there is a large imaginary part to the potential. Let
us none the less accept that there is a large attraction of the order of
that given in (3). The simplest possible generalization of (3) to higher
densities is
Vmf(p) = -200(p/p 0 ) MeV. (5)
The energy of a K~-meson in dense nuclear matter then will be
eK- = 494 MeV - 200(p/p0) MeV . (6)
In the collapse of large stars, due to the presence of electrons a large
negative chemical potential \i- is established in the core of the collapsing
star. An example of such a star was the progenitor of Supernova 1987A,
which was known from observations of its luminosity and type, to be of
mass
Mstar = 18 + 2 M O , (7)
measured in units of the mass of our sun M o . In collapse calculations, the
 Kaon Condensation in Dense Matter 441
core of the star which became the compact object, often called neutron
star, was estimated in many calculations to be

Mcore S 1.5 M o . (8)

More important for us, the density reached in the center of the core is
generally found to be several times nuclear matter density po, the precise
value depending upon the equation of state employed.
During the collapse of the star and the supernova explosion, the
neutrinos formed in the capture reaction

e~ + p <-> n + v (9)

are trapped, so that chemical equilibrium establishes the following equal-

ity of chemical potentials:

fie + Up = Hn + Mv • (10)

However, later, as neutrinos leave, the /xv can be left out of this equation
and we obtain
lie = lin-fip. (11)

The electron chemical potential, which initially is the chemical poten-

tial for negative charge \ie = //_, is then determined by this chemical
equilibrium. For densities of 2-4 po, it generally lies between

200 MeV < fie < 300 MeV , (12)

the precise value depending somewhat on the nucleon-nucleon interac-

tions. Not only is fie large, but it tends to increase with density.
Since eK- (see (6)) decreases with density, there will come a critical
density at which
eK-{pc) = fie. (13)
Above this density it will be energetically favorable for electrons to
change into K~-mesons, through the reaction

e ^ K ~ + v, (14)

the neutrinos leaving the star.f We will give results of detailed calcula-
tions later.
The energy is lowest, in a mean-field description, if the kaons so formed
t The formation of kaon condensation will happen after the era of neutrino trapping, so
that neutrinos can freely leave the star.
442 G. E. Brown
go into a state of zero momentum. The kaon condensate can then be
described by a classical wave function (see (15) below).
Calculations to date of kaon condensation have been carried out at
zero temperature. However, temperatures as high as 75 MeV are reached
in the collapse of stars [4]. Characteristic thermal energies are eth ~ TTT.
Thus, eth is of the same order as the electron chemical potential /ie ~
225 MeV at condensation. Consequently, the thermal energy may hinder
Bose-Einstein condensation, as it seems to hinder the condensations of
spin-polarized hydrogen or excitons discussed in this volume. In the
case of excitons in CU2O discussed by Wolfe et al, modification of the
crystal symmetry by applied stress seems to give the excitons a new way
to eliminate thermal energy and entropy. Such a way is present in a
star. Neutrinos will carry off energy quickly, reducing the entropy to that
needed for Bose-Einstein condensation. But the precise time scale for all
of this still needs to be worked out.

2 Calculation of Kaon Condensation

Calculations of kaon condensation have been carried out using chiral
Lagrangians, based on the Gell-Mann, Levy a-model. In fact, they have
been carried out in the nonlinear cr-model, rather than the GL-linear
model, but we shall ignore this (important) distinction. The salient point
is that the GL-Hamiltonian plays the same role as the Ginzberg-Landau
expression for the free energy; we don't even need to change the initials!
One introduces the kaon condensate as
(K^^e-^VK, (15)

where VK is the amplitude of the classical condensed kaon field, taken

here to be in the state of zero momentum, therefore uniform in r. The
GL equations will involve |t;xl2 as a parameter, and kaon condensation
will be possible when the energy can be lowered by choosing a nonzero

It is useful to introduce the chiral angle 0, which will serve as order

parameter for the kaon condensation, by
d = y/2\vK\/f, (16)
where / is the pion decay constant, / = 93 MeV. The angle 6 is the chiral
angle, and the significance of this angle can best be appreciated from the
paper by Brown, Kubodera and Rho [5] where strangeness condensation
 Kaon Condensation in Dense Matter 443

9
Fig. 1. Ginzberg-Landau energy (the temperature T is taken to be zero) for
p > pc, where pc is the critical density for the phase transition.

is interpreted as rotation in 8, in the region from 0 to 90°, the effects of

the bare strange quark mass being rotated out.
Let us now discuss the construction of the phase transition. The
condition that there be a phase transition is [6] that the propagator

D(co) = (co2 - m \ - Tr(co))-1 (17)

have a pole at co = /*_. Here n(co) is the kaon self energy. In our model
it includes only s-wave interactions. Physically, the reason that co = ja-
is that given in the last section; below this energy it becomes favorable
for electrons to change into kaons.
We have investigated only the second order phase transition, which
takes place at pc = 3.22po, where

po = 2.5 x 1014 g/cm 3 (18)

is nuclear matter density. The number density of nucleons is given by

multiplying by Avogadro's constant Ao = 6 x 1023, so that at pc = 3.22p0,
the critical number density is

nc = 4.8 x 1040 . (19)

Of course, such a large number is mindboggling.

The Ginzberg-Landau free energy has the form (shown in Fig. 1)

E = -a(p)82 + ±84 (20)

for p > pc. Here we have rescaled 8 so that the coefficient of the 0 4-term
is \. The coefficient a(p) has the form

fl(p) = a ( p - p c ) (21)
444 G. E. Brown
for a second order phase transition. The a comes out of our GL-
Lagrangian. Minimizing the energy with respect to 6 gives

— v — ±*\*\yjv ,- Zu , \^Z)

where 0 is the solution of this equation,

A I I

9 = ya(p) — v a(p — pc) . (23)

The energy E is found to be

E = ~oc2(p-pc)2 . (24)

From this one can see that the gain in energy is very rapid above p = pc.
In Table 1 we show the results of calculations [7] for the scenario
outlined in the previous section. For completeness, we list refs. [8, 9, 10,
11, 12] as a chronological development of the development of the input
used in the GL equations.
Table 1 describes the composition of neutron stars at densities p > pc.
One sees that the proton fraction x rises rapidly with density towards 0.5
and that the electron fraction xe decreases. The dense part of the core is
no longer a neutron star, but nuclear matter, or a "nucleon" star. Thus,
the kaon condensation scenario has replaced neutron by nucleon stars.
This should have many major implications for their structure, which are
presently being worked out.
The tendency to go from neutron matter towards nuclear matter is
easily understood. Once the electrons can be replaced by kaons, the
high energies of the former (fermions) need not be introduced in order
to neutralize the charge on the protons. Other things being equal, the
strong neutron-proton attraction favors equal numbers of protons and
neutrons. In this way the nuclear symmetry energy can be brought lowest.

3 Lots of Black Holes are the Smoking Gun for Kaon Condensation
We would now like to discuss the effect of kaon condensation on the
structure of compact objects formed following the collapse of large stars.
In the last section, we saw that kaon condensation substantially lowered
the energy of the dense matter. It also greatly softens the pressure, so
that the maximum mass of the compact objects is lower than is usually
found in the literature. Indeed, Brown [13] found Mmax = 15 M o , just
about the same as (8), the estimated mass of the core formed in the
 Kaon Condensation in Dense Matter 445

Table 1. Results of the calculation of the energy gain with

respect to neutron rich nuclear matter in beta equilibrium.
Here u = p/po, 0 is the chiral angle, Ae is the gain in energy,
in MeV, \i is the chemical potential for negative charge, in
MeV, x is the proton fraction; XK, the kaon fraction; xe, the
electron fraction and x^, the muon fraction.

u e Ae A* X xK xe
3.22 0.0 0.0 224.0 0.163 0.000 0.096 0.067
3.72 24.1 -5.2 165.6 0.317 0.267 0.034 0.016
4.22 27.1 -15.8 127.4 0.385 0.368 0.013 0.003
4.72 27.4 -28.4 100.8 0.421 0.415 0.006 0.000
5.22 26.8 -41.9 81.3 0.444 0.441 0.003 0.000
5.72 26.0 -55.9 66.7 0.458 0.457 0.001 0.000
6.22 25.0 -70.1 55.5 0.468 0.467 0.001 0.000
6.72 24.1 -84.7 46.7 0.475 0.475 0.000 0.000
7.22 23.1 -99.4 39.7 0.480 0.480 0.000 0.000
7.72 22.3 -114.3 34.0 0.484 0.484 0.000 0.000
8.22 21.4 -129.5 29.4 0.487 0.487 0.000 0.000
8.72 20.6 -144.9 25.6 0.489 0.489 0.000 0.000
9.22 19.9 -160.5 22.4 0.491 0.491 0.000 0.000

explosion of SN 1987A. It has been suggested [14] that this core went
into a black hole.
Arguments have been given [15] that the compression modulus of
nuclear matter is somewhat smaller than the conventional value [16]
Ko = 210 + 30 MeV. It was found [13], however, that Ko had to be raised
to 190 MeV, the lower region of conventional values, in order to stabilize
a 1.50 M o nucleon star. Figure 2 shows the known measured masses
of neutron stars. It is quite striking that all of the accurately measured
masses in binaries lie below 1.5 M o , the largest being the 1.44 M o pulsar
in PSR 1913+16. The mass of Vela X-l lies somewhat higher, but the
relatively large error bars reach down to ~ 1.55 M o .
In the formation of binary pulsars, after the first large star of a
binary explodes in order to form a compact object, the latter and the
remaining large star share a period of common envelope. During this
period, accretion can be rather large, ~ 10~3 M o per year [18]. Given
this large possible accretion, it would seem strange if compact objects
larger than 1.44 MG would not be seen in binaries, were they to exist.
Of course, 1.44 M o is just the Chandrasekhar mass for an N = Z star,
so that the collapse of the cores of large stars begins only when they
446 G. E. Brown

4U 1538-52
Ccn X 3 i •
HerX-1 •—•-

SMCX-1 i •
Vela X - l I » i

L-» PSR 2303+<16

1 i PSR 2303+ 16 companion
1 •- PSR 1855+ )9
PSR 1534+ 2 companion
• PSR 1534+ 2
• PSR 1913+ 6 companion
• PSR 1913+ 6
» 1 PSR 2127+ 1C companion
1 t PSR 2127+ 1C
1

1 2
Neutron star mass (Mo)

Fig. 2. Measured masses of neutron stars, from Ref. [17].

are unstable, with masses somewhat less than 1.44 M 0 because of some
electron capture. However, with accretion following the collapse, the
compact cores formed from large stars are estimated to be of the order
1.4 M o . Thus, it is true that compact objects tend to be formed with
such a mass. Our argument is that accretion from a companion star
should produce, in some cases, a larger mass. Since these are not seen,
this argues in favor of the maximum mass being not much larger than
1.44 M o .
We next argue that the collapse scenario of the core of a large star
is radically changed at high densities because of the kaon condensate.
Let us first review the conventional collapse scenario. As can be seen
from Fig. 3, the maximum masses of neutron stars at short times are
less than the longtime cold maximum mass, even though the trapped
neutrinos give considerable pressure. The point is that nuclear matter is
substantially softer than neutron matter, and the former is turned into
the latter as the trapped neutrinos leave, allowing electron capture. The
conversion of protons to neutrons increases the pressure more than it is
decreased by the loss of neutrinos. Thus, if the mass of the compact object
exceeds 1.5 M o at early times (or any time), it will immediately collapse
into a black hole. Inclusion of thermal pressures, which are presently
being calculated, may raise the solid, short-time curve, slightly above
the dashed, long-time one, so that there would be a small "window"
 Kaon Condensation in Dense Matter 447

Fig. 3. Maximum masses of neutron stars in the conventional scenario of neutron

matter composition [19]. The compression modulus Ko has been chosen to be
small, equal to 130 MeV, so that the late time cold maximum mass, shown by
the dashed line, is equal to Mmax = 1.5 M o . The solid line shows the maximum
mass during the first few seconds during which neutrinos are trapped. A lepton
fraction of T> = 0.4 was chosen for the early times.

of masses for a star to exist for some seconds, and then collapse into a
black hole as it cooled. We shall next discuss what happens in connection
with the much larger window, found in the case of the equation of state
including kaon condensation.
In the case of kaon condensation, the original nuclear matter does not
change into neutron matter. Indeed, at the higher densities, the fraction
of protons goes to one half. The early time maximum masses, as a
function of central density, are shown by the solid line in Fig. 4.
As can be seen from Fig. 4, the compact object can be stabilized,
during the time that the neutrinos are trapped, and then collapse into a
black hole. This means that the star can explode, returning matter to the
galaxy, and then go into a black hole [20]. Such black holes will be of
relatively light mass, not much larger than 1.5 M o . In general, observers
have looked for invisible centers of gravitational attraction, heavier than
448 G. E. Brown

2.0 1
1 • 1 1 ' 1 ' 1 ' 1

1.5 - 1
II
ll
if
1
II

1.0 - -

0.5 - -

0.0 1 1 . | . 1 . 1 . 1 . 1
0.0 1.0 2.0 3.0

Fig. 4. As in Fig. 3 the solid line shows the short-time maximum nucleon star
mass, as function of central density, with an assumed lepton fraction of Yf —
0.4, consisting of electrons and trapped neutrinos. The dashed line shows the
long-time cold mass, after the neutrinos have left.

~ 3 M o , as candidates for black holes, in order to make sure that the

mass exceeded that of the maximum mass neutron star. The black holes
of mass ^1.5 M o provide a new region of masses to look for.
Inclusion of late time accretion increases the above window for nucle-
osynthesis with later collapse into black holes by an estimated 0.04 M o
[20].
The possibility of nucleosynthesis with later collapse into black holes
was pointed out by Woosley et a\. in 1986 [21]. The mechanism for
stabilizing the compact object, which could later go into a black hole,
over the short time necessary for nucleosynthesis was the thermal pressure
which results from the conversion of gravitational energy into thermal
energy as the core goes through the Kelvin-Helmholtz contraction. As
noted above, this is presently being worked out with detailed equations
of state. There will be a relatively small "window" for nucleosynthesis
with later collapse into a black hole in the standard scenario where the
 Kaon Condensation in Dense Matter 449
compact core ends up as a neutron star. The larger "window" will be
further increased in our case of the kaon condensate. We estimate that
inclusion of all of the effects noted above will bring the upper limit
for nucleosynthesis with later collapse into a black hole to > 1.8 M o ,
corresponding to main sequence stars of mass > 30 M o .
That main sequence stars of mass > 25 — 30 M o must end up in black
holes without producing nucleosynthesis, i.e., without returning matter
to the galaxy, is required by the observed abundances of elements [22].
Maeder's argument is based on the measurement of AY/AZ, the ratio of
helium abundance to that of metals, in low-metallicity galaxies, especially
irregular dwarf galaxies [23]. Measurement of the ratio avoids the
problem of determining the normalization of the stellar mass function.
Most of the total galactic stellar mass is in light stars, of mass less than
1 M o . These contribute neither to A Y nor to AZ, so they do not affect
the ratio. The ratio can be measured with good accuracy [23]:

^=4±1.3. (25)
AZ
If all stable stars of mass up to ~ 100 M o were to explode, returning
matter to the galaxy, this ratio would lie between 1 and 2. Helium is
produced chiefly by relatively light stars, metals by heavy stars, so that
cutting off the production by the heavy stars going directly into black
holes without nucleosynthesis increases the AY/AZ. Maeder [22] found,
using the standard initial mass function for stars, that Pagel's measured
AY/AZ was best reproduced by a cutoff of nucleosynthesis at a main
sequence stellar mass of ~ 22.5 M o . There is considerable uncertainty
in the initial mass function, as noted by Maeder [22], so that this limit
could easily be ~ 30 M o or even higher.
The Brown-Bethe estimate [20] of 30 M o as the cutoff for stars to
drop directly into black holes without nucleosynthesis and the scenario
discussed above that a large range of stars below this mass can first
accomplish nucleosynthesis and then collapse into (light mass) black
holes, extends the estimated range of main sequence masses for which
stars go into black holes down to ~ 18 M o [20]. Interestingly, this
possibly includes Supernova 1987A, with progenitor mass of 18 + 2 M o .
Indeed, arguments have been put forward [14] that SN 1987A probably
went into a black hole.
A good candidate for a similar situation is the supernova remnant Cas
A, where the supernova explosion — unobserved at the time — took
place in the late seventeenth century. From element abundances in the
450 G. E. Brown
ejecta it has been argued [24] that the progenitor was an ~ 20 M o star, of
roughly equal mass to the progenitor of SN 1987A. Although observers
have looked with high accuracy in several different ways at the center of
the supernova remnant, they have been unable to find a compact object.
The Brown-Bethe scenario suggests that about as many stars end up
in black holes as nucleon stars. They estimate that there are ~ 500
million heavy mass black holes and a comparable number of light mass
black holes in the Galaxy. In fact, compact remnants are seen in only
~ 20 of the roughly 150 supernova remnants in the Galaxy [25]. It may
turn out that a black hole is the more likely fate than a nucleon star for
the compact core of a large star.
I would like to thank M. Prakash for pointing out the large "window"
for nucleosynthesis, with later collapse into a black hole, in the case of
strangeness condensates, and for providing me with Figs. 3 and 4. I am
grateful to Vesteinn Thorsson for providing me with Table 1 and both to
him and to Mannque Rho for their collaboration on the theory of kaon
condensation. I am grateful to Hans Bethe, with whom the scenario for
black hole formation of ref. [20] was worked out.
This work was supported by the US Department of Energy Grant No.
DE-FG02-88ER40388.

References
[1] E. Friedman, A. Gal and C. J. Batty, Phys. Lett. B 308, 6 (1993).
[2] P. B. Siegel and W. Weise, Phys. Rev. C 38, 221 (1988).
[3] A. Muller-Groeling, K. Holinde and J. Speth, Nucl. Phys. A 513,
557 (1990).
[4] J. Cooperstein, in First symposium on nuclear physics in the universe,
ORNL, TN, 2^26 Sept. (1992), M. Guidry, ed. (Adam Hilger), in
press.
[5] G. E. Brown, K. Kubodera and M. Rho, Phys. Lett. B 192, 273
(1987).
[6] G. Baym and D. K. Campbell, Mesons in Nuclei III, M. Rho and
D. H. Wilkinson, eds. (North-Holland, Amsterdam, 1979), p. 1031.
[7] V. Thorsson, private communication.
[8] D. B. Kaplan and A. E. Nelson, Phys. Lett. B 175, 57 (1986).
[9] H. D. Politzer and M. B. Wise, Phys. Lett. B 274, 156 (1991).
[10] G. E. Brown, K. Kubodera, M. Rho and V. Thorsson, Phys. Lett.
B 291, 355 (1992).
 Kaon Condensation in Dense Matter 451
[11] G. E. Brown, C.-H. Lee, M. Rho and V. Thorsson, Nucl. Phys.
A 567, 937 (1994).
[12] G. E. Brown, Nucl. Phys. A, to be published.
[13] G. E. Brown, in First symposium on nuclear physics in the universe,
ORNL, TN, 24^26 Sept. (1992), M. Guidry, ed. (Adam Hilger), in
press.
[14] G. E. Brown, S. W. Bruenn and J. C. Wheeler, Comm. Astrophys.
16, 153 (1992).
[15] G. E. Brown, Nature 336, 519 (1988).
[16] J. P. Blaizot, D. Gogny and B. Grammaticos, Nucl. Phys. A 265,
315 (1976).
[17] S. E. Thorsett, Z. Arzoumanian, M. M. McKinnon and J. H. Taylor,
Astrophys. Journ. Lett, 405, L29 (1993).
[18] R. A. Chevalier, Astrophys. Journ. 346, 847 (1989) ; R. A. Chevalier,
Astrophys. Journ. 411, L33 (1993).
[19] M. Prakash, private communication. See also V. Thorsson, M.
Prakash and J. M. Lattimer, Nucl. Phys. A 572, 693 (1994).
[20] G. E. Brown and H. A. Bethe, Astrophys. Journ. 423, 659 (1994).
[21] S. E. Woosley and T. A. Weaver, Ann. Rev. Astron. and Astrophys.
24, 205 (1986); J. R. Wilson, R. Mayle, S. E. Woosley and T. A.
Weaver, Proc. 12th Texas Rel. Astrophys. Symp., Jerusalem, Israel,
16-20 Dec, (1984).
[22] A. Maeder, Astron. Astrophys. 264, 105 (1992).
[23] B. E. J. Pagel, E. A. Simonson, E. A. Terlevich and M. G. Edmunds,
MNRAS 255, 325 (1992).
[24] R. A. Chevalier and R. P. Kirshner, Astrophys. Journ. 233, 154
(1979).
[25] S. Kulkarni, private communication.
 19
Broken Gauge Symmetry in a Bose
Condensate
A. J. Leggett
Department of Physics
University of Illinois at Urbana-Champaign
Urbana, IL 61801
USA

Abstract

This paper examines the meaning of the "phase" of a Bose condensate,

and in particular the degree of validity of the often used analogy with
the direction of magnetization of a ferromagnet. It focusses on two spe-
cific questions: (i) Under what circumstances is the relative phase of two
condensates well defined? (ii) Would it be possible in principle to set up
a "standard of phase"? In the most obvious sense, the answer to (ii) is
concluded to be no.

As is well known, it is very fashionable nowadays to treat Bose con-

densation as a special case of the more general idea of spontaneously
broken symmetry, which is ubiquitous in condensed-matter physics [1],
The standard account goes something like this: Just as in a magnetic
material described by an isotropic Heisenberg model, the Hamiltonian
is invariant under simultaneous rotation of all the spins, so in a Bose
system described by the standard creation and annihilation operators
\p\r),\p(r) it is invariant under the global (7(1) gauge transformation
ip(r) —• \p(r)el(p, y^(r) —• \p\r)e~iq>. Thus, at first sight, symmetry forbids
either the expectation value < S > of the magnetization of the mag-
netic material, or the corresponding quantity < \p > in the Bose system,
to take a finite value. Nevertheless, we know that if the material in
question is ferromagnetic, then when it is below its Curie temperature
and we take the thermodynamic limit in the presence of "infinitesimal"
symmetry-breaking fields in the usual sense, the quantity < S > indeed
takes a finite (in fact macroscopically large) value, and thus the direction
of the magnetization is a well defined quantity. Similarly, it is argued, in
the Bose case when one is below the Bose condensation temperature and

452
 Broken Gauge Symmetry in a Bose Condensate 453
takes the thermodynamic limit in a similar way, the quantity rj = < \p >
("order parameter") becomes finite (in this case proportional to N1^2)
and its phase q> thus becomes well defined. The Bose condensate is thus
said to possess "spontaneously broken gauge symmetry", just as the fer-
romagnet below its Curie temperature possesses "spontaneously broken
rotational symmetry".
The aim of the present paper is to examine in detail the concept of
spontaneously broken gauge symmetry and hence, by implication, the
degree of validity of the analogy between the Bose condensate and the
ferromagnet. To some extent, it is a continuation of the discussion
begun in Ref. [2], to which I shall sometimes refer for technical details
for which there is no space here. It should be said at once that, at
least as regards the vast majority of the experiments apparently feasible
today, nothing of any great importance hinges on the outcome of the
discussion, and indeed the points at issue might be regarded as in some
sense "theological"; my reason for pursuing them nevertheless is simply
that I find existing treatments of the concept of broken gauge symmetry
somewhat unsatisfactory. However, as a by-product we shall find that
some experiments are suggested which are not obviously beyond existing
capabilities.
I wish to make three remarks before I start on the main theme.
First, I shall feel free in the following to use examples, and thought-
experiments, which involve not only "Bose condensation" in the strict
sense but also the "pseudo-Bose condensation" of Cooper pairs in Fermi
systems. In the latter system, the quantity which plays the role of the
order parameter < xp > in a Bose system is related to the so-called
"two-particle anomalous average" defined by

M r y ) MV>«(OVVKO>. (1)
For example, in a simple BCS superconductor the natural choice for the
order parameter is the quantity F^(r9r), and all considerations regarding
the phase, etc., then go forward just as in the Bose case.t In the case of
Fermi superfluids such as 3He with more complicated pairing structures,
there may be more than one order parameter (as in the spin-1/2 Bose
system formed by spin-polarized atomic hydrogen, see Ref. [2]) and their
relative phase then becomes of interest: see Ref. [2], and below.
Secondly, for the purposes of the present paper the relevant order
parameter will always be assumed to be constant in space, except when
t Except for a factor of 2 in the commutation relations, which is irrelevant to the
considerations of this paper and which I will generally ignore.
454 A. J. Leggett
there is a discrete discontinuity such as that across a Josephson junc-
tion. To be sure, the question of how far the concept of spontaneously
broken symmetry remains valid in the presence of substantial spatial
inhomogeneity is an important one, and may well be connected with
some of the intriguing questions raised by Kagan [3], but to keep the
discussion focused I shall not attempt to attack it here. However, one
may hope that an improved understanding of the conditions for, and
implications of, broken gauge symmetry in the simpler cases discussed
here may eventually shed some light on these more practically important
questions.
Thirdly, let us try to tighten up the Bose-ferromagnet analogy a little
by focusing specifically on two variables of the ferromagnet, namely the
z-component of total spin Sz and the angle 9 = tan~1(5>,/5x) made by the
component of spin in the xy plane with some arbitrary reference axis. We
could, in fact, at this point drop the requirement that the Hamiltonian of
the ferromagnet has the complete 0(3) symmetry and demand only that
it be invariant under rotation around the z-axis. Under these conditions
the formal analogy with the Bose system is complete: to make it explicit,
we define the operator S+ = Sx + iSy so that we have the commutation
relation
S+, (2)
in complete analogy with the relation for the Bose case
[N,xp+]=ip+, (3)
where N is the total number operator. Thus, Sz is the analog of N and
(—)6 of the "phase" cp of xp; Sz generates "rotations" of 6 just as N
does of cp. We shall tentatively treat 6 and cp as operators satisfying
the commutation relations [5Z, 6] = j, [AT, cp] = —i. Although there are,
of course, well known difficulties [4] in doing so when the number
of particles involved is small, these difficulties appear to be irrelevant
for systems of typical "laboratory" size (as is shown, inter alia, by the
success of theories in which the phase is treated as a quantum-mechanical
operator in predicting the rate of "macroscopic quantum tunneling" in
Josephson systems, see e.g. Ref. [5]), and in any case do not appear to
affect the questions that I shall consider in this paper. The meaning of
the relative phase as an operator is clarified explicitly below. Of course,
it is conceivable that in the comparison of the role of cp with that of 0,
a crucial role is played by the possibility that the ferromagnet has the
full 0(3) symmetry and not just the 1/(1) one corresponding to rotation
 Broken Gauge Symmetry in a Bose Condensate 455
around the z-axis; we return briefly to the possible relevance of this
circumstance at the end of the paper.
Let us first review briefly some of the considerations discussed in more
detail in Ref. [2]. We first note that for an isolated system with a fixed
number JV of particles it is possible to define the "order parameter" rj(r)
in an alternative way, namely

rj(r) = ^N~oXo(r\ (4)

where xo(r) is the single-particle wave function into which condensation

has taken place and No the number of condensed particles; for the
spatially homogeneous case discussed here, this reduces to the rather
trivial expression
n = V^o exp(icp\ (5)

where cp is explicitly the phase of xo- Since the latter has no physical
meaning, it is clear that neither does the phase of rj, as we should of
course infer, for definite N, from (3). Thus it is clear that to make sense of
the concept of the "phase" of the order parameter, we must first discuss
the conditions for defining relative phase, and then ask whether we can
in some sense define an "absolute" phase by means, for example, of a
universal "phase standard". Note at this point that we do not normally
regard a similar problem as arising for the ferromagnet, because we are
used to the idea that Sz need not be well defined (thus allowing 9 to
have a definite value), whereas the notion that N is not definite is more
difficult to accept. Whether or not this difference of perception is really
well based is a question we shall have to return to later.
Let us then consider, as in Ref. [2] , a system which has available to
it two states xu li i n t o which Bose condensation may take place. An
obvious example is two bulk superconductors (or superfluids) joined by
a Josephson junction; in this case the states xi and xi would of course
simply correspond to localization on one side of the junction or the
other. However, it is more convenient to discuss first a case which is
closer to that envisioned in section 3 of Ref. [2], namely the Cooper
pairs in superfluid 3He-A. If we ignore the orbital degrees of freedom,
and moreover suppose that the so-called d-vector (for notation, see e.g.
Ref. [6]) lies in the xy-plane, then the two states x\ a n d Xi m question
correspond to pairs formed with total spin ±1, respectively, along the
z-axis: we denote these by ||f > and \[{> respectively. The most general
form of wave function in which all Cooper pairs are "Bose-condensed"
456 A. J. Leggett
into a single wave function may thus be written schematically (a, b real)f

*¥ ~ (aeiA(p/2 |TT> +be-**/2 IU>) N / 2 = *(Ap) (6)

(TV being the total number of 3He atoms). In the theory of magnetic
resonance in superfluid 3He [7], the ground state wave function is taken
to be of the form (6) with a = b and the relative phase Acp of the up-
and down-spin pairs chosen so as to minimize the dipole energy4
What is the variable canonically conjugate to Acpl It is nothing but the
difference AN of the total number of up and down spins in the system
(actually, of course, this is the total z-component of spin within a factor,
but I avoid the notation Sz so as to reserve this for the ferromagnetic
analogy). That AN and Acp are indeed canonically conjugate variables
(in the limit N -• oo) may be seen in a number of different ways, e.g.
by going back to the microscopic definition (1) of the order parameter
(see Ref. [7], section 2). For our purposes the most instructive derivation
probably consists in considering the class of many-body wave functions
(a subspace of the complete many-body space, of course) defined by

¥ = f c(AcpmAcp)dA(p, (7)

when ^(Acp) is defined by (6) and c(Acp) is an arbitrary complex coeffi-

cient. Note that in this expression, ^(Acp) plays the role of | x > in the
standard Dirac notation and c(Acp) that of the "wave function" (proba-
bility amplitude) y)(x): cf. Ref. [8], section 3.4. It is then straightforward
to show by direct calculation that within the subspace of states defined
by (7) the operator —id/d(Acp) has the same matrix elements as AN9 so
that Aq> and AN are indeed canonically conjugate variables within this
subspace. It is possible to extend this result to more general states, but
not worth doing in the present context.
We now return to a question raised, but not answered, in Ref. [2],
namely: Why should we assume that the wave function of a Bose-
condensed system which has available to it two different states (here
| t t > and | | | > ) will always be of the form (6), rather than (say) of the
form
¥ = (ITT>)*^(IU>)*^ = ¥(AN) (8)
t For simplicity of presentation I ignore here the notational complications introduced by
antisymmetrization, etc.
J Note that, contrary to the case discussed in section 3 of Ref. [2], Aq> does not represent
the angle of the Cooper-pair spin in the xy-plane, but, rather, the angle of the direction
d along which they have spin projection is 0.
 Broken Gauge Symmetry in a Bose Condensate 457
(which, as mentioned in Ref. [2], is a special case of (7) with c(Acp) =
exp(—iAN .Acp))l It is clear that if we put no conditions whatever on the
physical situation envisaged, there is no particular reason to prefer (6)
to (8) (or indeed to a myriad of other possible states). However, if we
specify that the system is in or close to thermodynamic equilibrium, there
is a simple reason why (6) is usually preferred. Consider for definiteness,
as above, the case of superfluid 3He-A. The terms in the Hamiltonian
which depend on AN and Acp may be written

(9)

Here Xs is the spin susceptibility of the system (which, as it turns out,

is essentially just the normal-state Pauli value), and gD is the coupling
constant due to the dipole force (which, since it does not conserve spin
angular momentum alone, can have nonzero matrix elements between
the states | | | > and ||i>). The crucial observation, now, is that both Xs
and gD are extensive quantities, that is, they are proportional to the total
number of particles N. We can then see immediately from (9) that in
the thermodynamic limit N —> oo, it is always advantageous to have Acp
well defined and allow AN to fluctuate, i.e. the wave function is indeed
approximately of the form (6). A more exact consideration shows that
in this limit the groundstate wave function (7) is of harmonic-oscillator
form around Acp = 0, with the rms fluctuation of Acp proportional to
AT1/2 and that of AN to JV1/2.
It is clear that a similar argument will go through for any case in which
the states xi a n d li i n t o which Bose condensation can take place are
"extensive" (in a sense obvious from the above), provided only that the
term in the Hamiltonian which connects them (the generalization of go)
is not identically zero. Thus, given this condition, the relative phase of
the two condensates is indeed always well defined in the thermodynamic
limit, so that the concept of broken (relative) gauge symmetry is justified.!
While this conclusion is, of course, in no way novel, the argument given
above has the advantage of making it quite explicit that the occurrence
of broken (relative) symmetry in the thermodynamic limit is in no sense
"magic" but is simply a result of energy considerations, and thus that the
point at which we may legitimately assume that the limit is a function of
the strength of the "symmetry-breaking" interaction. If we imagine that
t It is needless to remark that similar arguments concerning the absolute "phase" in the
presence of external "source" terms are commonplace in the literature. The advantage
of the present formulation is that it uses no such unphysical assumptions.
458 A. J. Leggett
we could somehow adjust the strength of the dipole coupling constant
gr> in 3He-A, then the above argument makes it clear that as go tends to
zero, we would need to go to larger and larger volumes of the liquid to
stabilize the broken relative symmetry of the up- and down-spin Cooper
pairs. Indeed, it is interesting to inquire what is the order of magnitude of
the "critical" volume for the actual value of the dipole coupling realized
in nature. The answer is ~ 10~18cm3; since this is (by a numerical
accident) on the order of the cube of the coherence length, it seems
doubtful that one could obtain volumes of superfluid A-phase much less
than this. If this should be possible (e.g. in the form of thin films), one
would expect their NMR properties to be spectacularly different from
those of the bulk liquid.
Now let us turn our attention to the case in which the two relevant
states xi and xi into which Bose condensation can take place are spatially
distinct; the most obvious realization of this situation is of course via
the Josephson effect, and some aspects of it were already discussed in
Ref. [2]. In the context of the present discussion, the salient point is that,
in contrast to the "extensive" case treated above, the Josephson coupling
constant Ej (the analog of go in the above argument) is in principle
subject to adjustment by the experimenter as a function of time. We may
in fact assume that when Josephson contact is made Ej is sufficiently
large that the relative phase Acp of the Cooper-pair (or the atom) wave
functions localized on the left or right is very well defined, while when it
is broken, Ej may be effectively taken to be zero.
We first briefly revisit the question first raised by RW. Anderson and
discussed in section 5 of Ref. [2]: Do two superfluids which have never
"seen" one another possess a definite relative phase? The arguments
given in section 4 of Ref. [2], which show that even when the superfluids
have been in contact (and have acquired a definite relative phase) in the
past, this is eventually washed out, point strongly to the answer "no".
However, as pointed out in that reference (section 5) any attempt to
answer the question experimentally, by asking whether or not a Josephson
current flows when two such "memory-free" superfluids are connected,
may be frustrated by the fact that any attempt to determine this is liable,
according to the precepts of the quantum theory of measurement, itself to
"create" a relative phase, irrespective of whether or not there "really was"
one before the measurement.! Nevertheless, there exists an alternative
way of giving meaning to the question (although once it is framed in this
t A very detailed analysis of this phenomenon has recently been given by Hegstrom and
Sols [9].
 Broken Gauge Symmetry in a Bose Condensate 459
way, the answer is perhaps obvious): We assume that the two superfluids
together constitute a closed system (i.e., cannot exchange particles with
the rest of the world). The "experiment" simply consists in weighing them
at separate times t\ and ti, which can be arbitrarily far separated, so as to
determine the number difference AN at these times, without ever making
Josephson contact between them. Evidently, if, given a sufficient time
T, the two superfluids can acquire a definite relative phase even without
making Josephson contact, then for t2 — t\ » x we should find that
Ayfa) is definite (to order N~i/2) and hence the second measurement of
AN should in general give a different result from the first, the fluctuation
being presumably of order iV1/2. On the other hand, if the relative
phase Acp is not definite, then there is no reason why AN should vary in
time in the absence of contact (and indeed every reason why it should
not!), and one would expect the measured values of AN(t\) and AN(t2)
to be identified. I can see no reason whatever to doubt that it is the
latter conclusion which would be found experimentally, so that in this
(operationally defined) sense, the statement that "two superfluids which
have never seen one another before nevertheless have a definite relative
phase" is, I believe, false.
I now turn to what I believe is the most intriguing question in this whole
business, namely the question of a "phase standard". To orient ourselves,
let us note that, according to the considerations of Ref. [2], section 4,
once two superfluids have been placed in Josephson contact, then even
after separation they may be able to maintain a definite relative phase
for a time which can in principle be quite long by laboratory standards;
we may assume that by suitable engineering (e.g. placing the samples
in a spaceship) this time may be made as long as we please. Then
the question arises: Could we in principle "standardize" phases of (say)
buckets of helium the world over by placing them successively at some
central system maintained by a standards laboratory?! If this should
indeed turn out to be possible, then for all practical purposes we could
define an "absolute" phase simply as the phase relative to the standard.
Let us try to make this question a bit more precise. Suppose that two
buckets of helium (1 and the "standard", S) are known to have a definite
relative phase Acpis, e.g. because they have been placed in Josephson
contact and allowed to come to equilibrium (so that in fact Acp is = 0).

f Needless to say, even if the answer were to turn out to be yes, this is unlikely in the
extreme to be a practical proposition, inter alia because of the extreme difficulty of
controlling the chemical potentials acting on the different buckets sufficiently accurately
to prevent an unknown drift of their relative phase.
460 A. J. Leggett
Suppose further that the same is true of S and 2 (for the precise meaning
of this statement, see below). Is it then true that 1 and 2 have a definite
relative phase? Note that this question is experimentally answerable via
an ensemble of experiments in which 1 and 2 are placed in contact and
the "instantaneous" Josephson current between them is measured. If the
relative phase is definite, the same current will flow on all occasions,!
while if it is not true, the results will be randomly distributed.
It is, however, crucial to distinguish two different versions of this
thought-experiment. In version (A), S is placed in contact with 1 and 2
simultaneously and the whole system is allowed to come to equilibrium;
all contacts are then broken, and 1 is subsequently placed in contact with
2 without the mediation of S. In version (B), 1 is first placed in contact
with S, with 2 absent, and sufficient time is allowed for equilibrium
to be established. The 1-S contact is then broken, and subsequently
contact between S and 2 is made and time allowed for the attainment of
equilibrium. Finally this contact is also broken, and eventually 1 and 2
are put in contact as in version (A).
In view of space limitations, I shall give here only the results of the
analysis of these two thought-experiments; the details of the calculations,
which are not entirely trivial, will be presented elsewhere [10]. Case (A) is
straightforward: at the end of the day, a definite phase relation is indeed
established (subject, of course, to all the necessary caveats concerning
time-scales, etc., see Ref. [2]) between the superfluids 1 and 2, so that
on joining them a predictable Josephson current is obtained. Case (B)
is a little more tricky: one might intuitively suspect that the phase
relation established at the first stage between 1 and S would be "upset"
in the subsequent attainment of Josephson equilibrium between S and
2, and this indeed turns out to be the case, although, surprisingly, a
proper demonstration of this result requires explicit consideration of
the "environment" of the system (e.g., for the superconducting case,
the radiation field). Indeed, given a certain very plausible, though not
rigorously proved, assumption [10] concerning the dependence on the
total particle number involved of the final state of the environment, one
can demonstrate that no definite phase relation is established at the end
of the day between systems 1 and 2. Hence a "standard of superfluid
phase" is possible only in sense (A), not in sense (B).
We must finally return to the ferromagnetic analogy. Let us try, as
t In order to obtain a finite Josephson current, one could apply a known chemical potential
difference between 1 and 2 for a known time (so as to make Acpn nonzero before placing
them in contact).
 Broken Gauge Symmetry in a Bose Condensate 461
a thought-experiment, repeating the arguments made in this paper with
the "phase of a superfluid" replaced by the "direction of spin (in the xy-
plane) of a ferromagnet". Just as we have implicitly assumed throughout
that the total number of the relevant particles (electrons, He atoms, etc.)
in the universe is concerned, let us for the sake of the argument make the
analogous assumption of conservation of the total z-component of spin.f
We would then have to conclude that, strictly speaking, the absolute
direction of magnetization is meaningless and that we should only talk
about relative directions of spin of different magnets. That most of us are
happy in everyday life with the idea that for all practical purposes we can
talk about "absolute" directions would then seem to imply a tacit belief in
the possibility of "standardization". For the case of the superfluid phase,
we have argued that such standardization is possible only in sense (A),
which requires prima facie very special and artificial conditions: what, if
any, is the salient difference in the case of magnetization?
I believe that the crucial difference between the two cases is to be
sought in the very different role of what we have vaguely called the
"environment". It turns out to be implicit in our arguments about phase
standardization for a superfluid that the "environment" (radiation field,
phonons, normal electrons or whatever) either cannot exchange particles
with the "system" at all, or, even if it can, cannot exist (or is not typically
found to exist) in the coherent superposition of states of different particle
number (in the sense, explored above, in which a superfluid system does).
That this is so is essentially just the statement that most of the world
is not superfluid (let alone a superfluid state of the relevant particles) !J
On the other hand, just about any "environment" can not only exchange
angular momentum with the system, but has no difficulty in existing
in a coherent superposition of states of different Sz (or rather Jz: here
the exchange of spin and orbital angular momentum does play a vital
role). More accurately, the universe can easily exist in a superposition
of states which differ by the relative angular momentum of "system"
and "environment". (That this is so is no doubt not unconnected with
the question of the more general SO(3) invariance.) In other words, the
typical situation as regards magnetization corresponds to "version A"

f Of course in real life spin and orbital angular momentum are not separately conserved,
but taking this into account only pushes the problem one step back.
$ To be sure, a superconductor may induce a degree of phase coherence in a normal metal
adjacent to it - the so-called proximity effect (see e.g. Ref. [11], section 7.3). However,
the effects die out exponentially with distance into the normal metal and thus cannot
play the necessary "global" role.
462 A. J. Leggett
of our thought-experiment rather than "version B", so the problem of
standardization is automatically overcome.
A final thought: Will "version A" ever be realized in a practical sense
for superfluid systems? This idea is not quite so preposterous as it
sounds. If the dream of room-temperature superconductivity is attained
(and perhaps even if it is not), it is not impossible that by the year 2100
there will be a network of superconducting transmission lines covering
the globe, with no normal "breaks". Then, indeed, it would be possible
in principle for groups, based say, in Tokyo and Washington, DC, to
"mate" two pieces of superconductor which had never been in direct
physical contact in such a way as to produce a predictable Josephson
current on first acquaintance! Although this state of affairs would of
course in no way alter any of the considerations present above, it might
perhaps modify our perception of their significance somewhat.
This work was supported by the National Science Foundation under
Grant No. DMR 92-14236.

References
[I] RW. Anderson, Basic Notions of Condensed Matter Physics (W.A.
Benjamin, Menlo Park, California, 1984).
[2] A.J. Leggett and F. Sols, Found. Phys. 21, 353 (1991).
[3] Yu. Kagan, this volume.
[4] P. Carruthers and M. M. Nieto. Phys. Rev. Lett. 14, 387 (1965).
[5] A. J. Leggett, J. Phys. Soc. Japan Supp. 26-3, 1986 (1987).
[6] A. J. Leggett, Rev. Mod. Phys. 47, 331(1975).
[7] A. J. Leggett, Ann. Phys. 85, 11(1974).
[8] A. J. Leggett, in Chance and Matter, J. Souletie, J. Vannimenus and
R. Stora, eds. (North-Holland, Amsterdam, 1987).
[9] R. Hegstrom and F. Sols, submitted to Phys. Rev. A.
[10] A. J. Leggett, Found. Phys., to be published.
[II] P. G. de Gennes, Superconductivity of Metals and Alloys (W.A.
Benjamin, N.Y, 1966).
 Part two
Brief Reports
 20
BEC in Ultra-cold Cesium: Collisional
Constraints
E. Tiesinga, A. J. Moerdijk, B. J. Verhaar, and H. T. C. Stoof
Department of Physics, Eindhoven University of Technology,
P.O. Box 513
5600 MB Eindhoven
The Netherlands

Abstract

We study necessary conditions for the observation of Bose-Einstein con-

densation in a magnetically trapped sample of atomic cesium gas. These
constraints are due to interatomic collisions in the sample. We show that
the prospects for observing Bose-Einstein condensation are favorable for a
gas of ground-state Cs atoms in the highest state of the lowest hyperfine
manifold. An interesting aspect of the calculations is that the scattering
length for this f = 3, mj = — 3 hyperfine state shows pronounced reso-
nance structures as a function of applied magnetic field leading to varia-
tions of two orders of magnitude. Most importantly, the scattering length
can change sign near the resonances. This suggests a controllable means
to change the behavior of the Bose condensate because for negative val-
ues a condensate is unstable and other (quantum-)collective effects might
be observed. The origin of the resonances is understood from the bound
singlet and triplet rovibrational Cs2 states which are perturbed due to the
hyperfine and Zeeman interactions.

It is a long standing goal to achieve quantum-collective effects in atomic

Fermi- or Bose gases. The prominent reasons are that for a relatively
simple system with low density, a microscopic theoretical treatment of
the phase transition is still feasible, and to have an experimental testing
ground for more complicated quantum-coherent effects such as superflu-
idity in 4He and superconductivity in metals. Here, we focus on atomic
species which behave as (composite) bosons. A gaseous system like
this can Bose condense below a critical temperature, i.e. macroscopi-
cally occupy the one-particle ground state. For this to occur the atomic
wavefunctions need to overlap, which can be expressed as nA3 = 0(1),

465
466 E. Tiesinga, A. J. Moerdijk, B. J. Verhaar, and H. T. C. Stoof
where n is the density and A is the temperature-dependent DeBroglie
wavelength. The first system considered was a gas of ultracold hydrogen
[1] stored either in a gas cell with superfluid helium-lined walls or in a
wall-free magnetic trap. Although enormous progress has been made,
the required density/temperature combination has not been achieved.
Since the advent of laser cooling [2] other species have become avail-
able as candidates for observing Bose-Einstein condensation. In a few
milliseconds the (mostly alkali-metal) atoms are cooled from room tem-
perature to temperatures in the micro kelvin regime and after that stored
magnetically. One of the first experiments to magnetically trap heavy
alkali-metal atoms used 133Cs [3]. In this case the attempts so far reached
a density/temperature combination of 2 x 1010 cm"3 and 1//K. All exper-
iments aimed at Bose-Einstein condensation are hampered by collisions
between atoms. This limits the density and the lifetime of the sample to
a large extent. Since storage in a magnetic trap is restricted to low-field-
seeking hyperfine states of the electronic ground state, particle loss occurs
either by exothermic collisions releasing sufficient kinetic energy to leave
the trap or by transitions to high-field-seeking states. For sufficiently
high densities recombination is an important loss mechanism as well.
Although inelastic collisions limit the observable densities, the elastic
collisions are essential for the formation of the condensate [4]. This is
due to the fact that these collisions comprise the only process available
to transfer the atoms in the one-particle ground state. Furthermore, the
validity of the usual discussion of the Bose condensate in terms of a
Bogoliubov transformation [5] crucially depends on a positive value of
the scattering length, which is equivalent to a repulsive elastic collision
at low kinetic energies. For negative values the Bogoliubov theory breaks
down. This paper briefly discusses the trap-loss effects for atomic cesium
and focusses mainly on the resonant behavior of the scattering length
as a function of magnetic field. The latter promises to be an excellent
property for the study of the Bose condensate. An extensive discussion
of the density decay can be found in Ref. [6].
The starting point of the our discussion is the hyperfine level diagram
(Fig. 1) of the cesium ground-state (2Si/2) atom. If we consider the
magnetic field along the z-axis, the effective single-atom Hamiltonian
comprises hyperfine and Zeeman terms

HM = *£Se.Sn + {yeSe_ynS« )B9 ( J )

where Se and S" are electron and nuclear spin, respectively. Magnetic
 BEC in Ultra-cold Cesium: Collisional Constraints 467

Fig. 1. Hyperfine level diagram of 133Cs as a function of magnetic field. The

states are labeled |/,m/) and £hf = 4flhf-

traps generally operate with fields well below the critical value B c =
4ahf/(yeh) « 0.33 T, so that the total spin vector f = S e + Sn is still
conserved. Since the electron (nuclear) spin is 1/2 (7/2), the 16 hyperfine
states of cesium are conveniently labeled with |/,m/). The collision of
two such atoms is described by the effective two-body Hamiltonian [6],[7]

H = Z- (2)

containing a kinetic energy term with ft the reduced mass and two-body
interactions Vc and Vd. The central or exchange interaction Vc represents
an effective description of all Coulomb interactions between the electrons
and nuclei and depends only on the magnitude of the total electron spin
S = Sf + S|. It can be written as a sum of singlet and triplet terms

Vc = (3)

with Ps the projection operators on the subspaces with definite total

electron spin quantum number S = 0,1. The magnetic dipole-dipole
interaction Vd is classically understood as the coupling between either
the electronic or the nuclear magnetic moment of one atom with those of
the other. The singlet Cs-Cs potential VQ is taken from Demtroder and
468 E. Tiesinga, A. J. Moerdijk, B. J. Verhaar, and H. T. C. Stoof
coworkers [8], where it is derived from measurements of CS2 rotation-
vibration energies. The triplet rovibrational levels have not been mea-
sured and a theoretical calculation [9] is used. Hence, the rates presented
below give an indication of the order of magnitude. As will be discussed
below, however, the presence of resonances is independent of the precise
shape of the potentials.
Symmetry considerations show that the so-called exchange transitions,
induced by the central interaction, conserve the orbital angular momen-
tum quantum numbers ( and m^, as well as the total spin projection
Mp = ntfi + m/2. For zero magnetic field no preferential direction exists
and exchange transitions will conserve the total molecular spin F as well.
The dipole-dipole interaction, which is much weaker than the exchange
interaction and therefore has little effect on exchange transitions, induces
dipole transitions conserving wv + MF and satisfying |A/| = 0,2 with
{ = 0 -> i = 0 forbidden.
The scattering matrix describing all hyperfine-changing transitions is
found by rigorously solving the coupled channels equations. We find that
for the ultra-cold temperatures considered, a zero-temperature calculation
is sufficient. This implies that the initial state of the collision has zero
angular momentum. Moreover, the inelastic transitions are divided in
two classes. Their rate constants differ by several orders of magnitude.
Exchange transitions give a rate constant of the order of GC(T = 0) =
10~12cm~3/s and lead to a decay time of 102 x (1010cm~3/n) seconds.
Dipolar decay rate constants are much smaller and are of the order
of Gd(T = 0) = 2 x 10~15 cm~3/s corresponding to a decay time of
5 x 104 x (1010cm~3/w) seconds. The second-order spin-orbit interaction
[10] also gives rise to density decay. It is very qualitatively known but
most likely of the same order of magnitude as the dipole interaction. In
view of the uncertainties in the central interaction already present we
have omitted this decay process.
The distinction between the two classes makes those gases which
decay via dipolar relaxation preferable. It turns out that a sample of
magnetically trapped atomic cesium with all atoms in either the |4,4) or
the |3,—3) hyperfine level solely decays via the dipole-dipole interaction,
where the latter is a low-field-seeking state as long as the trapping
field remains below Bc. A collision between two doubly polarized |4,4)
hyperfine states is described by pure 5 = 1 triplet scattering, implying
the absence of exchange transitions. Furthermore, a gas of |3,—3)
atoms, with collisions described by both singlet and triplet terms, is long
lived since other hyperfine states accessible via the central interaction
 BEC in Ultra-cold Cesium: Collisional Constraints 469

500
(0,6,-6)
(£,F,Mp)=(0,7,-6)

6 (0^8,-6)

i
t-l
J
O
CO

-500
0.00 0.05 0.10

Fig. 2. Scattering length for elastic |3,— 3) + |3,—3) scattering as a function of

magneticfield.Labels denote quantum numbers (£9F,MF).

are energetically forbidden as long as collision energies remain below

the hyperfine splitting 4ahf « 0.4 K. Elastic scattering, essential for
thermalization of a sample as well as for the formation of the Bose
condensate, is for both samples governed by the central interaction.
For the doubly-polarized gas the scattering length is solely given by
t = 0 scattering in the triplet potential and is therefore independent of
magnetic field. The uncertainty in the singlet/triplet curve leaves the sign
of the scattering length undetermined. In the case of a gas of |3,— 3)
atoms, however, resonant behavior of the scattering length (See fig. 2)
is observed. This is due to a magnetic-field-dependent coupling between
the singlet and triplet potential curves.
At this point it is convenient to express the collision in terms of
the \(SI)F,MF) basis, with total nuclear spin I = SJ + SJ. Note that
in this basis the central interaction is diagonal while the hyperfine-
and to a lesser extent the Zeeman interaction couple the S = 0 and
1 potential curves. Moreover, the total energy of the initial |3,—3) +
|3,—3) state, which for low field values is approximatively given by
—9/2ahf + 3/%hy eB+ kinetic energy, lies below the theshold of the central
470 E. Tiesinga, A. J. Moerdijk, B. J. Verhaar, and H. T. C. Stoof
interaction. Therefore a resonance will occur when the total energy
in the system is close to the energy of £ = 0 vibrational levels of
the singlet and triplet potential to first-order shifted and split by the
hyperfine and Zeeman interaction, i.e. ((SI)F9MF\ J2HM\(SI)F9MF).
At first sight this indicates that a resonance corresponds to a specific
value of 5. A more comprehensive discussion[6] including higher-order
effects, however, shows that the hyperfine interaction must be included
more thoroughly and that bound states in the subspaces characterized
by £,F,MF bring about the resonances. The underlying mechanism is
more generally known as a Feshbach resonance. In view of the potential
curves used, the position of the resonances is unknown. However, the
energy separation between bound states of the subspaces /, F, Mp near
the initial energy of —9/2ahf is of the order of the hyperfine splitting
and is not sensitive to the precise potential shape. This implies that the
resonances are always present. Furthermore, the width of the F = 6
resonance, about 5 x 10~~3 T, is independent of the potential shape and
large enough to be of experimental interest. Note that the scattering
length changes from positive to negative values.
In conclusion, magnetically trapped cesium in the |3, — 3) hyperfine
state is a very promising candidate for the observation of quantum-
collective effects.

References
[1] I.F. Silvera and J.T.M. Walraven, Phys. Rev. Lett. 44, 164 (1980);
J.M. Doyle, J.C. Sandberg, LA. Yu, C.L. Cesar, D. Kleppner, and TJ.
Greytak, Phys. Rev. Lett. 67, 603 (1991); O.J. Luiten, H.G.C. Werij,
I.D. Setija, M.W. Reynolds, T.W. Hijmans, and J.T.M. Walraven,
Phys. Rev. Lett. 70, 544 (1993).
[2] The special issues of J. Opt. Soc. Am. B devoted to laser cooling;
Vol. 2 (1985) and Vol. 6 (1989).
[3] C. Monroe, W. Swann, H. Robinson, and C. Wieman, Phys. Rev.Lett.
65, 1571 (1990); C. Monroe, E. Cornell, C. Sackett, C. Myatt,and C.
Wieman, Phys. Rev. Lett. 70, 414 (1993).
[4] H.T.C. Stoof, Phys. Rev. Lett. 66, 3148(1991); H.T.C. Stoof, Phys.
Rev. A 45, 8398 (1992); Yu. M. Kagan, and G.V. Shlyapnikov, Zh.
Eksp. Teor. Fiz. 101, 528 (1992); B.V.Svistunov, J. Moscow Phys.
Soc. 1, 373 (1991). Contributions of these authors on nucleation can
also be found in this volume.
 BEC in Ultra-cold Cesium: Collisional Constraints 471
[5] A.L. Fetter and J.D. Walecka, Quantum Theory of Many-Particle
Systems, (McGraw-Hill, New York, 1971).
[6] E. Tiesinga, A.J. Moerdijk, BJ. Verhaar, and H.T.C. Stoof, Phys. Rev.
A 46, R1167 (1992); E. Tiesinga, BJ. Verhaar, and H.T.C. Stoof,
Phys. Rev. A, May (1993).
[7] H.T.C. Stoof, J.M.V.A. Koelman, and BJ. Verhaar, Phys. Rev. B 38,
4688 (1988).
[8] H. Weickenmeier, U. Diemer, W. Demtroder, and M. Broyer, Chem.
Phys. Lett. 124, 470 (1992); H. Weickenmeier, U. Diemer, M. Wahl,
M. Raab, W. Demtroder, and W. Muller, J. Chem. Phys. 82, 5354
(1985).
[9] M. Krauss and WJ. Stevens, J. Chem. Phys. 93, 4236(1990).
[10] P.S. Julienne, J. Mol. Spectrosc. 56, 270 (1975); S.R. Langhoff, J.
Chem. Phys. 61, 1708 (1974).
 21
BEC and the Relaxation Explosion in
Magnetically Trapped Atomic Hydrogen
T. W. Hijmans, Yu. Kagan,* G. V. Shlyapnikov,* and J. T. M. Walraven
Van der Waals-Zeeman Laboratoriwn, Universiteit van Amsterdam,
Valckenierstraat 65/67
1018 XE Amsterdam
The Netherlands

Abstract
We predict and analyze non-trivial relaxational behavior of magnetically
trapped gases near the Bose condensation temperature Tc. Due to strong
compression of the condensate by the inhomogeneous trapping field, par-
ticularly at low densities, the relaxation rate shows a strong, almost jump
wise, increase below Tc. As a consequence the maximum fraction of con-
densate particles is limited to a few percent. This phenomenon can be called
a "relaxation explosion". We discuss its implications for the detectability
of BEC in atomic hydrogen.

Magnetostatic traps offer the possibility to study gases of Bose particles

in the truly dilute limit, and have proved particularly fruitful [1, 2, 3, 4, 5]
in the study of atomic hydrogen (H). In these traps, proposed for H by
Hess [6], the effective elimination of physical boundaries is accomplished
by creating a magnetic field minimum in free space. This minimum
forms a potential well for electron spin-up polarized atoms (Hf), called
low-field seekers. The occurrence of Bose-Einstein condensation (BEC)
in such systems introduces qualitatively different behavior from the case
of a homogeneous Bose gas. This is related to the explosive increase
of the dipolar relaxation rate associated with the strong compression of
the condensate in an external potential. A similar phenomenon occurs
in connection with three-body recombination in high density systems [7]
The large condensate density resulting from this compression gives rise
to an increase of the rate of inelastic (dipolar) pair collisions, in which
spins of the colliding particles are flipped, producing high-field seeking

t Permanent address: Russian Research Center, Kurchatov Institute, Moscow 123182,

Russia.

472
 BEC and the Relaxation Explosion in Trapped Atomic H 473
atoms (Hj) which are ejected from the trap. The approach to BEC
is often [6, 8] described in terms of trajectories in density-temperature
space. We shall see that in H, the increase in relaxation rate resulting from
the appearance of the condensate, the "relaxation explosion", prevents
one from penetrating deep into the BEC region and markedly alters these
trajectories.
For simplicity we consider an isotropic harmonic trapping potential of
the form
V(r) = fiBB0 + ™co 2r\ (1)
where r is the distance to the trap center, and co is the oscillation
frequency. The critical BEC temperature is expressed by the relation
Tc = 3.31ft2n2/3/m, where n is the density of the gas at the trap center
and m is the mass of the atom. The density profiles in external potentials
were analyzed by Goldman et al. [9], and by Huse and Siggia [10]. The
critical temperature can be expressed in terms of the total number of
particles N9 and the parameters of the trapping potential [11]. For the
potential given by Eq. (1) we have
Tc = (N/ g 3 (l)) 1 / 3 ^. (2)
Here g3(l) = 1.20, where g/ is a Bose integral given by g«K£) =
£^i(£7«0.
Well above Tc, the number of relaxation events per unit time is given
by
= 2a /drn 2 (r), (3)

where n(r) is the density distribution and a is the rate constant for
dipolar decay. For hydrogen a « 10~ 15 cm3/s [12, 13]. Below Tc we
should replace Eq. (3) by a more general expression [14, 15]:

vr = 2a f dr i < <p\x)xp\r)<p{r)xp{x) >, (4)

where xp is the field operator of the atoms and < xp^xp^xpxp > is the local
two-particle correlator.
A detailed calculation [16] shows that, below Tc, the relaxation rate
can be expressed in terms of its value at Tc:
T AT
vr(T) = vr(Tc)(l + 7.5(-^) 3 ' 5 (—) 7 / 5 ) ; T < Tc (5)
Here, AT = Tc — T, and ncU represents the mean field interaction
474 T. W. Hijmans et al.

Fig. 1. Number of relaxation events per unit time for a fixed number of atoms,
normalized to the value at Tc. The solid line corresponds to nc = 1015 cm" 3 and
the dashed line to nc = 1013 cm" 3 .

energy, where nc is the critical density. For H, at T = 100 JJK, we have

nc « 5 x 10 14 cm"3 and ncU « 0.2 //K. This clearly shows the extreme
weakness of the interaction in this system.
Qualitatively, Eq. (5) can be understood as follows. Above Tc we
have, according to Eq. (3) vr ~ ctN2/Ve, where the classical effective
volume Ve is defined trough n(r) = (N/Ve)exp(—V(r)/T). Below T c we
have an additional contribution associated with the condensate particles:
VO. Explicit calculation [16] shows

v0 =
Clearly, as a result of the weak interaction, we have Vo < Ve. Hence,
even if No < N (and accordingly AT < T) the condensate can dominate
the relaxation rate. In Fig. 1 the value of v r (T)/v r (r c ) is given for H
atoms for nc = 1013cm"3 (Tc = 7JIK) and nc = 1015cm"3 (Tc = 150/iK).
To investigate the effect of the relaxation explosion, we use a simple
model for the kinetics of the cooling proces, in which we assume that
the cooling proceeds sufficiently slowly to consider the system as being
in thermal equilibrium. To obtain the trajectories in N — T space, we
start from the expressions for the internal energy of the trapped gas at
 BEC and the Relaxation Explosion in Trapped Atomic H 475
temperatures above and below Tc (here we neglect ncll):

;T>TC (7)

where £ is the fugacity exp(—fi/T). From this we obtain relations between

the time derivatives of T, U and JV:

\~W)
t 1J7
T = 4U iT*T- (10)

If we parametrize the cooling process by an energy removal rate

Rv = —U/U, and an accompanying rate RN of particle removal, we can
write the rate of change of particle number and internal energy as

Here Ur is time rate of change of energy associated with relaxation. As

the relaxing particles have a finite energy, the internal energy decreases
with decreasing particle number. Hence, Ur is always negative. From
Eq. (10) one thus finds that when T < Tc, because the energy does
not explicitly depend on N, relaxation does not lead to heating. In
contrast, from Eq. (9) we see that for T > Tc removal of particles
only leads to cooling if the energy of these particles is sufficiently high.
This is the principle of evaporative cooling. In this sense, any loss of
(above-condensate) particles leads to evaporative cooling for T < Tc.
Note, moreover, that the ratio f/N changes discontinuously at Tc. This
behavior has the same origin as the discontinuity in the derivative of
specific heat at the BEC point. A straightforward calculation shows
that il(T) = -(9/4)Tv r for T > Tc and U(T) = -(9/4)T0.8vr(Tc) for
T < Tc.
Using the above results we can easily obtain the trajectories in the
N — T plane, if we make the somewhat simplifying assumption that
Ru is proportional to the elastic collision rate Re = \GVTN/Ve. The
resulting trajectories are shown in Fig. 2 for two values of Ru = Ru/Rc-
Below Tc the relaxation explosion manifests itself as an attraction of
the trajectories towards the BEC line. This is particularly striking when
comparing with the corresponding trajectories obtained by assuming the
system to obey Boltzmann statistics at any temperature and density.
476 T. W. Hijmans et al.

1016

1014
10-5 10-4 10-3

Temperature (K)

Fig. 2. Cooling trajectories for Rv =0.1 and 0.05 (solid lines) plotted as N/Ve
vs T. The long-dashed curve is the BEC line. The short-dashed curves represent
the trajectories corresponding to a system obeying Boltzmann statistics.

This research was supported by the Nederlandse Organisatie voor

Wetenschappelijk Onderzoek (NWO-PIONIER and NWO-07-30-002).

References
[1] H.F. Hess, G. Kochanski, J.M. Doyle, N. Masuhara, D. Kleppner,
and T.J. Greytak, Phys. Rev. Lett. 59, 672 (1987).
[2] R. van Roijen, J.J. Berkhout, S. Jaakkola, and J.T.M. Walraven,
Phys. Rev. Lett. 61, 931 (1988).
[3] J.M. Doyle, J.C. Sandberg, LA. Yu, C.L. Cesar, D. Kleppner, and
T.J. Greytak, Phys. Rev. Lett. 67, 603 (1991).
[4] OJ. Luiten, H.G.C. Werij, I.D. Setija, M.W. Reynolds, T.W. Hijmans,
and J.T.M. Walraven, Phys. Rev. Lett. 70, 544 (1993).
[5] I.D. Setija, H.G.C. Werij, O.J.Luiten, M.W. Reynolds, T.W. Hijmans,
and J.T.M. Walraven, Phys. Rev. Lett. 70, 2257 (1993).
[6] H.F. Hess, Phys. Rev. B 34, 3476 (1986).
[7] Yu. Kagan and G.V. Shlyapnikov, Phys. Lett. A 130, 483 (1988).
[8] T.J. Tommila, Europhys. Lett. 57, 314 (1986).
[9] V.V. Goldman, I.F. Silvera, and A.J. Leggett, Phys. Rev. B 24, 2870
(1981).
 BEC and the Relaxation Explosion in Trapped Atomic H All
[10] D.A. Huse and E.D. Siggia, J. Low. Temp. Phys. 46, 137 (1982).
[11] V. Bagnato, D.E. Pritchard, and D. Kleppner, Phys. Rev. A 35 4354
(1987).
[12] A. Lagendijk, I.F. Silvera, and B.J. Verhaar, Phys. Rev. B 33, 626
(1986).
[13] H.T.C. Stoof, J.M.V.A. Koelman, and B.J. Verhaar, Phys. Rev. B 38,
4688 (1988).
[14] Yu. Kagan, B.V. Svistunov, and G.V. Shlyapnikov, Pis'ma Zh. Eksp.
Teor. Fiz. 48, 54 (1988) [JETP. Lett. 48, 56 (1988).
[15] H.T.C. Stoof, A.M.L. Jansen, J.M.V.A. Koelman, and B.J. Verhaar,
Phys. Rev. A 39, 3157 (1989).
[16] T.W. Hijmans, Yu. Kagan, G.V. Shlyapnikov, and J.T.M. Walraven,
Phys. Rev. B 48, 12886 (1993).
 22
Quest for Kosterlitz-Thouless Transition
in Two-Dimensional Atomic Hydrogen
A. Matsubara,f T. Arai, S. Hotta, J. S. Korhonen, T. Mizusaki,
and A. Hirai $
Department of Physics
Kyoto University
Kyoto 606-01
Japan

Abstract
We have cooled atomic hydrogen adsorbed on liquid helium to very low
temperatures: a surface temperature of SI mK was obtained with a density
of 2 • 1012 atoms/cm 2 . The two-dimensional hydrogen gas was cooled on
a cold spot, which was inside a large sample cell. When the cold spot is
much colder than the rest of the sample cell the surface density on it is
controlled by the hydrogen flux into the sample cell. We found that at least
98% of the recombination heat was carried away from the cold spot by
the excited H2 molecules. There was a heat input to the cold spot which
was independent of the presence of atomic hydrogen and which prevented
cooling of the adsorbed hydrogen below 87 mK. We attribute this heat input
to thermal conduction by ripplons.

1 Introduction
The experiments with polarized atomic hydrogen [1, 2] are usually made
in sample cells whose walls are covered with a helium film. The atomic
hydrogen is adsorbed on the He surface and forms a two-dimensional
(2D) gas, which is expected to undergo a Kosterlitz-Thouless transition
[3] to a superfluid state. The critical temperature of the transition is
TKT=nh2a/2kBm , (1)
where m is the mass of the hydrogen atom and a is the surface density.
If the adsorbed hydrogen is in thermal equilibrium with the bulk gas, it
is difficult to control the surface density and temperature independently:
t Permanent address : Department of Physics, Osaka-City University, Sumiyoshi-ku,
Osaka, 558 Japan,
t Deceased on 31 December, 1992.

478
 Quest for K-T Transition in 2D Atomic Hydrogen 479
the adsorption energy ea is about 1 K and when temperature is lowered
much below I K the surface density tends to increase exponentially and
soon approaches the saturation density 2 • 1014 atoms/cm2. This leads to
rapid three-body recombination at the surface, which prevents cooling
below TKT-
Our method of cooling 2D-HJ, is the following: inside a large sample
cell we prepared a small cold spot. The cold spot was thermally isolated
from the rest of the sample cell, and its temperature, Tcs, was kept
much lower than that of the sample cell. At sufficiently low TQS the
recombination occurs mainly on the cold spot. Then the surface density
is nearly independent of temperature and is determined by the input flux
of hydrogen to the sample cell. We found that although the recombination
occurs mainly on the cold spot, most of the recombination heat is carried
away from the cold spot by the excited H2 molecules.

2 Experimental Apparatus
We prepared doubly polarized atomic hydrogen, HJ^r, as follows. The
atomic hydrogen was produced at low temperature by rf discharge. It
was precooled by a baffle before it was stored in a buffer volume, which
was connected to the sample cell through a flow impedance. The buffer
and the sample cell were in a magnetic field of 8 T. If theflowimpedance,
volume of the buffer and the flux of hydrogen to the buffer are selected
properly, the atomic hydrogen in the buffer volume is doubly polarized.
The sample cell was a cylinder with radius 17.5 mm and height 75 mm.
Its walls were covered with superfluid 4He film of saturation thickness.
The cold spot was a copper disk of radius 2.2 mm. It was located at the
upper part of the sample cell. The cold spot surface was partly covered
with a sintered silver piece of radius 1.4 mm and estimated surface area
of 0.1m2. The cold spot was thermally isolated from the rest of the
copper sample cell by a piece of Stycast with an outer radius of 8 mm.
The discharge cell and the baffle were anchored to a heat exchanger
of the dilution refrigerator at a temperature of 0.5 K. The common
temperature of the buffer volume and the sample cell was fixed to a
temperature T#, between 290 and 430 mK, by a temperature controller.
The cold spot was connected by a mesh of copper wires to the mixing
chamber of the dilution refrigerator and its temperature was controlled
at Tcs.
We measured two quantities: the bulk density n of the hydrogen in
the sample cell and the increase in the heat input to the cold spot which
480 A. Matsubara et al.
was due to the introduction of hydrogen. In order to measure changes in
the heat input to the cold spot we fixed the temperatures of the mixing
chamber and the cold spot separately by temperature controllers. Then
the change in the feedback power of the cold spot controller was equal
(but opposite) to the power Q due to the introduction of hydrogen.
The bulk density n was measured by making all the hydrogen in the
sample cell recombine on a bolometer. The density was calculated from
the total recombination heat, which was obtained from the increase in
the sample cell temperature. With this method we were able to measure
only densities larger than 1012cm~3 but the absolute value of n was
obtained reliably. For smaller densities n was calculated from the rise in
the bolometer temperature.

3 Rate Equations for Hydrogen Densities

We assume that the atomic hydrogen in the sample cell was doubly
polarized. Then the data analysis can be based on the following equations
for the time evolution of the bulk density n of H|^- in the sample cell
and the surface density a on the cold spot:

(2)

Here V is the sample cell volume' Acs is the area of the cold spot; and
</) is the flux of HJ^r to the sample cell; Ks and Ls are the two- and
three-body recombination rates on the cold spot; R!jf*9 KeJ* and If^
are the effective one-, two- and three-body recombination rates, which
include the contribution from the recombination on the hot part of the
sample cell wall and are defined through equations R^n = RS(A/V)GH,
Kebffn2 = Kbn2+KS(A/V)(J2H and Lebffn3 = Lbn3+LS(A/V)(T3H. The sample
cell area is A and GH = nAGxp(ea/kBTH) is the surface density on the hot
walls of the sample cell; X is the thermal de Broglie wavelength. In our ex-
periments, only the recombination constants Ls = 1.2 x 10~24 cm4 s"1 [4]
and Rs = 8.5 cm"1 are important. The determination of Rs is discussed in
section 4. For values of Ka and L&, see Ref. [4], and for Ks, see Ref. [5].
In Eq. (2), the thermal velocity of the atoms in the bulk gas is vb =
m, and VbtiAcs/4 is the number of atoms colliding with the cold
 Quest for K-T Transition in 2D Atomic Hydrogen 481

T s (mK)
jioo_

10 20
1
1/Ts (K- )
Fig. 1. The calculated saturation values of the bulk density n in the sample cell
and the surface density a on the cold spot as a function of the inverse temperature
Tg"1 of the adsorbed hydrogen. In region (i) the cold spot does not affect the
bulk density; in regions (ii) and (iii) the recombination occurs mainly at the cold
spot.

spot surface in unit time. The sticking coefficient Sb = 0.33 T#/K [5, 6] is
the probability that a surface collision leads to the adsorption of the atom.
The mean residence time on the surface is TS = (4As/vsSs)exp(ea/kBTs),
where A, v and S must now be evaluated at the temperature of the
adsorbed hydrogen, denoted by Ts. Due to the thermal resistance
between the adsorbed hydrogen and the cold spot, Ts may not be the
same as Tcs-
The saturation values of n and <r, given by Eq. (2), are plotted in Fig. 1
as a function of Ts, at constant TH = 310mK and $ = 1.1 • lO^s" 1 .
There are three distinct temperature regions: (i) the high temperature,
(ii) the desorption limited, and (iii) the recombination limited regions. In
the high temperature region the recombination on the cold spot can be
neglected. In regions (ii) and (iii) the recombination occurs mainly on
the cold spot. Then a is given by the equation Acs(Kso2 + Lscr3) = <f> and
is almost independent of the cold spot temperature. A desired value of a
can be obtained by selecting an appropriate hydrogen flux <j). In region
(ii) the probability that the adsorbed atom desorbs before recombination
is large, while in region (iii) nearly all the atoms adsorbed on the cold
spot recombine before being desorbed.
482 A. Matsubara et a\.

T(mK)
1 190 80 60
10
TH=310mK

10 15 20
1/r (K-1)
Fig. 2. The saturated bulk density of hydrogen vs. the inverse cold spot temper-
ature TQI (filled circles) and the inverse temperature of the adsorbed hydrogen
Tjf1 (open circles). Ts is calculated from the measured heat input to the cold
spot and the thermal resistance between the ripplon and the phonon systems.
The dashed line is a guide for the eyes, and the solid curve is calculated from
Eq. (2).

4 Experimental Results and Data Analysis

The measurements of n were started about two hours after switching on
the pulsed rf discharge. This time was needed to saturate the density of
hydrogen in the buffer volume. First we determined the values of the
flux 4> and the recombination rate constant Rs. This was done at high
Ts (region (i) in Fig. 1 by recording n(t) after all the atomic hydrogen
was removed from the cell at t = 0. A fit of the measured n(t) to Eq. (2)
gave <j> = 1.1 x 1012 s"1 and Rs = 8.5 s. It turned out that one-body
recombination was the dominant mechanism in the bulk, because n was
rather low. Next we measured the saturation value of the hydrogen
density nsat. The solid circles in Fig. 2 show nsat as a function of the cold
spot temperature TQS> The solid line, denoted by ncai(Ts\ is calculated
from Eq. (2). The difference between nsat(Tcs) and ncai(Ts) indicates that
Ts was higher than Tcs- The observed value of nsat serves as a sensitive
thermometer for Ts in region (ii) in Fig. 1. The saturation value of
the heat input to the cold spot, Qsat, is shown in Fig. 3 as a function
of Tcs.
 Quest for K-T Transition in 2D Atomic Hydrogen 483

T C S (mK)
100 80 60

1/Tcs
Fig. 3. The heat input to the cold spot caused by the atomic hydrogen as a
function of the inverse cold spot temperature. The circles show the measured
data; the solid line is the power carried as the kinetic energy of the hydrogen
atoms. Qrecomb is the total power released in the recombinations. Only a fraction
smaller than 1.4% of QreComb goes to the cold spot.

All these data were taken under conditions such that the recombina-
tion occurs mainly on the cold spot. The total recombination heat power
Qrecomb> which is calculated from the flux (j>9 is shown in Fig. 3 by the
upper dashed line. Comparison with Qsat reveals that less than 1.4% of
the total recombination heat goes to the cold spot. This implies that the
H2 molecule, which is left in a highly excited state in the recombination,
cannot penetrate into liquid helium, and the recombination heat is deliv-
ered evenly to the sample cell. A similar result was obtained by Vasilyev
et ah [8] They measured an upper limit of 10% for the recombination
energy which is deposited into helium at the recombination site. The
heat input in Fig. 3 seems to vary roughly proportional to the density
nsat in Fig. 2. A mechanism which would show such a dependence is
heat transfer from the hot bulk hydrogen gas to the cold surface. The
calculated power Qkm, which is released when the hot hydrogen atoms
thermalize after being adsorbed, is shown in Fig. 3 by the solid line. The
local heat flux
= 2kB(TH - T(r))nvcc/4, (3)
484 A. Matsubara et al.
is calculated by using the measured density n. Here a = (3/2)5 is
the thermal accommodation coefficient, measured by Helffrich et al.
[9] Outside the cold spot, the radial distribution of the temperature,
T(r), is determined by the thermal conductivity of Stycast and is not
changed much by a small q^m- We estimate Qkm by integrating qkm
over the cold spot and the Stycast area. Qkin follows Qsat down to the
lowest cold spot temperature. It thus appears that the fraction of the
recombination heat which goes to the cold spot was much smaller than
1.4%.
We have used Qkin in estimating the difference between the tempera-
tures of the cold spot and the adsorbed hydrogen. According to Reynolds
et al. [7], the bottleneck in the heat conduction from the adsorbed hy-
drogen to the cold spot is between the ripplon and the phonon systems
in the helium. We thus assume that the adsorbed hydrogen is at the same
temperature as the ripplons and that the temperature of the phonons in
helium equals the cold spot temperature. Using the result of Reynolds
et al. we calculate the temperature difference T$ — TQS which is caused
by Qkin being dumped to the ripplons on the cold spot. The open cir-
cles in Fig. 2 show the measured nsat as a function of the calculated
7$. Agreement with ncaic(Ts) is good at Ts > 90 mK but at lower
temperatures the measured Qsat is too small to explain the temperature
difference Ts — TCs- This difference should be attributed to some resid-
ual heat input which does not depend on the presence of the atomic
hydrogen.
The saturation of Ts can be explained by the ripplon heat conduction.
Using the conduction measured by Mantz et al. [10, 11], we obtain a heat
input of about 30 nW to the cold spot under the conditions of Figs. 2
and 3. At low temperatures the heat carried by the ripplons is thus
much larger than Qsat and determines the minimum 7$. The calculated
minimum Ts = 85 mK is in good agreement with the value of 87 mK
obtained from Fig. 2.
In order to clarify the effect of the ripplon heat conduction, we mea-
sured nsat as a function of TH with fixed TQS = 80 mK. Using Eq. (2) we
estimated Ts from nsat; TS(TH) is shown by the circles in Fig. 4. The lines
are calculated by using the results of Reynolds et al. The dashed line
is obtained by considering only the ripplon heat conduction, while the
lower solid line results if only Qkin is taken into account. Both ripplons
and Qkm contribute to the upper solid line. Clearly, the heat carried by
ripplons becomes important for low T#.
 Quest for K-T Transition in 2D Atomic Hydrogen 485

300 400
TH(mK)

Fig. 4. Temperature of the adsorbed hydrogen Ts as a function of the sample

cell temperature TH. The temperature of the cold spot is fixed to TCs = 80 mK.
The open circles show Ts estimated from the measured nsaU using Eq. (2). The
lines are calculated from the thermal resistance between the ripplon and the
phonon systems and the heat input to the surface. Dashed line: only ripplon heat
conduction considered. Lower solid line: only Qun taken into account. Upper
solid line: combined effect of Qkin and ripplon heat conduction.

5 Summary
Our experiments demonstrate that two-dimensional atomic hydrogen gas
can be cooled on a small cold area much below the temperature of the
bulk gas. In particular, we have shown that the recombination heat was
not the limiting factor of the minimum temperature of the adsorbed
hydrogen: the minimum T$ was determined by the residual heat input
to the ripplon system of the He-film. This residual heat flux was due to
ripplon heat conduction. In future experiments we will try to cool the
two-dimensional hydrogen further by using thinner 4He films and 3He
or 3He-4He films.
Finally, we note that although we do not at the moment have a direct
signal from the adsorbed hydrogen we believe that the KT transition can
be detected even in the present set-up. This is due to a large decrease
in the surface recombination rate when the KT transition takes place
[12,13]. If the measurement is done in the desorption limited temperature
region (ii) in Fig. 1, the transition should be seen as a sudden increase in
n when Ts is lowered below TKT>
486 A. Matsubara et al.
References
[I] I.F. Silvera and J.T.M. Walraven in Progress in Low temperature
Physics, Vol. 10, edited by D.F. Brewer (North Holland, Amsterdam,
1986), p. 139.
[2] I.F. Silvera and M. Reynolds, J. Low Temp. Phys. 87, 345 (1992).
[3] J.M. Kosterlitz and D J . Thouless, J. Phys. C 6, 1181 (1973).
[4] D.A. Bell, H.F. Hess, G.P. Kochanski, S. Buchman, L. Pollack,
Y.M. Xiao, D. Kleppner, and TJ. Greytak, Phys. Rev. B 34, 7670
(1986).
[5] L. Pollack, S. Buchman, and T.J. Greytak, Phys. Rev. B 45, 2993
(1992).
[6] JJ. Berkhout, E.J. Wolters, R. van Roijen, and J.T.M. Walraven,
Phys. Rev. Lett. 57, 2387 (1986).
[7] M.W. Reynolds, I.D. Setija, and G.V. Shlyapnikov, Phys. Rev. B 46,
575 (1992).
[8] S.A. Vasilyev, E. Tjukanov, M. Mertig, A. Ya. Katunin, and
S. Jaakkola, Europhys. Lett. 24, 223 (1993).
[9] J. Helffrich, M.P. Maley, M. Krusius, and J.C. Wheatley, Phys. Rev.
B 34, 6550 (1986).
[10] D.O. Edwards, I.B. Mantz, and V.U. Nayak, J. Phys. (Paris), Colloq.
39, C6-300 (1978).
[II] I.B. Mantz, D.O. Edwards, and V.U. Nayak, Phys. Rev. Lett. 44,
663 (1980).
[12] Yu. Kagan, B.V Svistunov, and G.V. Shlyapnikov, Sov. Phys. JETP,
66, 314 (1988).
[13] H.T.C. Stoof and M. Bijlsma, preprint (1993).
 23
BEC of Biexcitons in CuCl
Masahiro Hasuo and Nobukata Nagasawa
Department of Physics
The University of Tokyo
7-3-1 Hongo, Bunkyo-ku, Tokyo, 113
Japan

Abstract
A new approach to detecting coherence associated with Bose-Einstein con-
densation of biexcitons at K=0 in CuCl is described. Remarkable enhance-
ment of the phase-conjugate signal at the two-photon resonance of the biex-
citons at K=0 is observed when high-density incoherent biexcitons are in-
jected, suggesting macroscopic increase of the occupation number of the
K=0 biexciton state and enhancement of the coherence. A small blue shift
of the K=0 biexciton resonance is also observed, suggesting the change
of the chemical potential of the biexciton due to the interaction between
biexcitons.

1 Introduction
An excitonic particle such as an exciton or a biexciton (excitonic molecule)
in a semiconductor crystal has a boson-like nature. If the system of
excitonic particles is regarded as an ideal Bose gas, Bose-Einstein con-
densation (BEC) may occur at the critical density of
_ 2.612 2mkBT 3/2
c { ] U)
~ J4^ ~W~ '
where m and T are the effective mass of the particle and the temperature
of the system, respectively.
In most semiconductors, the biexciton, a bound complex of two exci-
tons, is the most stable entity, with the lowest energy per electron-hole
pair. The CuCl crystal is famous as such a material, the excitonic param-
eters of which have been well determined experimentally. For example,
the effective mass of the biexciton is 5.3 times as large as the electron rest

487
488 M. Hasuo and N. Nagasawa
mass [1]. Thus, the corresponding Nc is estimated to be 2.2 x 1017/cm3
at T=2 K.
A biexciton in CuCl decomposes into one photon and one exciton
through a radiative process that gives characteristic luminescence called
M-emission. The spectral shape of the M-emission is well known to be
determined by the density of states and the distribution function of the
biexciton system. When the density of biexcitons, N, is far less than
Nc, the spectral shape is reproduced approximately by the Maxwell-
Boltzmann distribution function [2]. With increase of JY, it is expected
that the spectral shape changes, reflecting the Bose statistics due to the
Bose nature of the biexciton.
In fact, it has been reported that the spectral shape of M-emission
is reproduced by the Bose-Einstein distribution function corresponding
to fx=O under special experimental conditions, where \i is the chemical
potential of biexcitons [3]. A demonstration showing the quantum at-
traction between biexcitons in phase space has also been presented by
the use of a pump-and-probe method on the M-emission associated with
the two-photon excitation of biexcitons [4].
When the BEC occurs, it is expected that a macroscopic number of
biexcitons collapses to a single quantum state of zero momentum, K=0,
and forms a coherent quantum state. However, the coherence inherent in
the BEC has never been demonstrated experimentally so far. Recently,
we have proposed a new pump-and-probe method that aims to detect
the coherence [5].
We measure the change in the intensity of the phase-conjugate signal
generated from the third-order macroscopic non-linear polarization asso-
ciated with the coherent biexcitons populated at K=0 under the presence
of the high-density biexcitons injected incoherently. Fig. 1 shows a
schematic illustration of the phase conjugation. The relevant non-linear
polarization is expressed by
P^^^(Ei'E2)Em39 (2)
where E\ and Ej are the electric fields of the counter-propagating laser
beams 1 and 2 used for exciting the coherent biexcitons at X=0 [6].
Adjusting these beams' intensities, the density of the coherent biexcitons
populated at K=0 is substantially lower than Nc. The electric field £3 of
the laser beam 3 is used to induce a parametric scattering process associ-
ated with the coherent biexcitons, as shown in Fig. 1. The signal light field
has phase which is conjugate to that of £3, as seen in Eq. (2). Since this
phase-conjugate signal is generated from the coherent assembly of the
 BEC ofBiexcitons in CuCl 489

(a)

Beam! Beam2

Beam3
' Signal

(b)
- lbiexciton>

Beam2 Beam3
EGTA EGTA + 6

Beami Signal
EGTA EGTA - 6

Fig. 1. Geometry (a) and energy diagram (b) for the phase conjugation.

relevant microscopic polarizations which enter into %(3), one can monitor
the occupation number of the biexcitons at K=0 and its coherence. If
BEC is realized, it is expected that the intensity of the phase-conjugate
signal should show a remarkable increase as an indication of the onset
of BEC and of the enhancement of the relevant coherence.

2 Experimental Method
Figure 2 shows a schematic diagram of the experimental set-up. Each
photon energy of the counterpropagating beams 1 and 2, provided by
Dye Laser A, was adjusted to be the two-photon resonance energy of the
biexcitons at K=0, £Gr^=3.1858 eV. The spectral width of the laser light,
7i,2> was adjusted to be 138 jueV for a reason mentioned below. These
beams were focused in a sample. The spot size on the sample surface was
about 100 /xm</>, resulting in a total intensity of 160 kW/cm2. The phase
 490 M. Hasuo and N. Nagasawa

Dye laser A PBS Beam 2 ND

Beam 1

ND--

Monochromator • «— ^

P
HM /
ND Q
Dye 1 aser B
Beam 3 \/

Dye 1 aser C
Beam 4
Fig. 2. Schematic diagram of the experimental set-up. PBS :polarized beam
splitter, ND:neutral density filter, Qiquarter wave plate, P:polarizer, VA:variable
attenuator, HM :half mirror. Inset shows a phase-conjugate spectrum expected
under the present excitation conditions. For details, see text.

coherent signal from the coherent biexcitons generated by beams 1 and 2

was induced by the radiation of beam 3 from Dye Laser B. The photon
energy was set at EGTA+S, where S is the detuning from the resonance
energy and was 4 meV. The spectral width, y3, was set to be 37/*eV.
An inset of Fig. 2 shows a schematic illustration of a phase-conjugate
signal in the spectral domain under these excitation conditions. The
phase-conjugate signal appears at EGTA - d, shown in the diagram. The
corresponding transition is shown in Fig. l(b). The spectral width of
the phase-conjugate spectrum is determined mainly by the level width of
the biexciton at K=0, y0, through the relevant / (3) under the condition
7i,2 > 7o > 73- The spectral shape is deformed by a phase-matching
factor, fPMi given by
sin2AKL
(3)

where AK and L are the relevant phase mismatch [7] and the sample
thickness, respectively. The dashed curve shows the phase-matching fac-
 BEC of Biexcitons in CuCl 491
tor as a function of photon energy estimated for the present experimental
conditions.
The polarization of these laser beams is made circular so as to excite
mainly the K=0 biexcitons via the relevant selection rule of the two-
photon transition. The external angle between beams 1 and 3 was 10°,
which corresponds to an internal angle of 3.6°. The phase-conjugate
signal was analyzed with a high resolution monochromator (Jobin-Yvon
THR1500, band pass 16/ieV).
The fourth laser beam, from Dye Laser C, was used to generate the
incoherent biexcitons through a bimolecular collision process of photo-
excited excitons. We shall refer to this light as the incoherent pump
light hereafter. The photon energy and the spectral width of this light
were 3.2029 eV and 1.36 meV, respectively. The beam was focused on
the sample surface to a spot of 200 fimcj) in order to cover the region
illuminated by beams 1, 2 and 3. The intensity of this light was adjusted
continuously with a variable intensity attenuator. The maximum intensity
was 2.7 MW/cm2.
All the dye lasers (BBQ in p-dioxane) were pumped by a common
excimer laser (Lambda Physik 53MSC). The pulse duration of each dye
laser light was about 10 ns.
The sample was a 15 /mi thick single-crystal platelet, grown from vapor
phase. All the measurements were done at 2 K.

3 Results and Discussion

Figure 3 shows an example of the experimental results on the phase-
conjugate spectrum. Curve (a) shows the spectrum without injection of
the incoherent biexcitons. With increase in the intensity of the incoherent
pump light, the intensity of the phase-conjugate signal shows an increase,
as shown in curves (b)-(d). In these cases, the increase of the signal
was mainly localized in the peak region of the spectrum. A small blue
shift of the spectrum was also observed correlated to the increase of the
signal intensity. However, this trend reversed beyond about 1 MW/cm 2.
In these cases, the decrease of the signal was seen in the whole spectral
region.
Figure 4 shows the peak intensity of the phase-conjugate spectrum,
normalized by the spectrum without injection of the incoherent biexci-
tons, as a function of the intensity of the incoherent pump light. Open
squares correspond to the cases shown in Fig. 3. The dotted curve
492 M. Hasuo and N. Nagasawa

"CO

a>

0 -

3.1850 3.1855 3.1860

Photon energy (eV)

Fig. 3. Spectra of the phase-conjugate signal at 2 K with the intensities of the

incoherent pump light of (a) 0 kW/cm2, (b) 120 kW/cm2, (c) 260 kW/cm2, (d)
680 kW/cm2, (e) 930 kW/cm2, (f) 1800 kW/cm2 and (g) 2700 kW/cm2.

is a similar one, obtained by changing the intensity of the incoher-

ent pump light continuously by a variable intensity attenuator. In this
case, the monochromator's wavelength was fixed at the peak energy of
the phase-conjugate spectrum without injection of incoherent biexcitons.
The general trend was well reproduced.
The scale shown in the top of this diagram shows the density N
of incoherent biexcitons that was calculated from the intensity of the
incoherent pump light. This was estimated by solving the following rate
equations under steady state conditions:

dN
n 2 (4)
ym
~dt 2
dn 9
_ = -rjn1 + ymo\N - yexn + J, (5)
at
 BEC ofBiexcitons in CuCl 493

Biexciton density (xlO 1 7 /cm 3 )

0 1 5 10 50 100 200

0.0
500 1000 1500 2000 2500 3000
2
Intensity (kW/cm )

Fig. 4. The peak intensity of the phase-conjugate spectrum as a function of the

intensity of the incoherent pump light. The calculated biexciton density is scaled
at the top of the diagram. The arrow shows the critical density for BEC of ideal
Bose gas at 2 K. For details of the estimation of the biexciton density, see text.

where n is the density of excitons. The parameters listed below were

adopted for the present calculation:
rj : biexciton formation rate, 6 x 10~8cm3/s [8],
7moi '. radiative decay rate of a biexciton, 1/50 (ps)"1 [9],
7ex : decay rate of an exciton, 1/5 (ns)"1 [10].
In this analysis, the effective excitation density, J, is given by

J = (6)
LeSE
Here, R,I and E are the reflectivity, intensity and photon energy of the
incoherent pump light, respectively; Leff is the effective excitation depth
and was assumed to be expressed in the form

Leff = (7)

where vg and ysc are the group velocity and the scattering rate of the
exciton polaritons which are converted from the incoherent pump light
at the sample surface. They were adopted to be 1.5 x 10 6cm/s and
1 x 109s~1, respectively [10]. The collision rate of excitons to create the
incoherent biexcitons is rjn. As seen in Fig. 4, the remarkable increase
494 M. Hasuo and N. Nagasawa

102
Biexciton density ( c m 3 )

Fig. 5. Peak shift of the phase-conjugate spectrum as a function of the biexciton

density. The solid line shows the change of the chemical potential of biexcitons
obtained theoretically.

of the signal intensity was observed in the density region of Nc at 2 K,

marked by the arrow.
Figure 5 shows the energy shift of the peak of the phase-conjugate
spectrum as a function of N on a log-log scale. The change of the
chemical potential due to the interaction between particles, A£, may be
taken as
(8)
m
where OQ is the scattering length of the particles, 2.9 nm [11, 12]. For the
present experimental conditions, we take A£=5.25 x 10~16iV in units of
//eV, shown by the solid line in the diagram. The observed shift was only
reproduced quantitatively in the lowest excitation case. The origin of the
deviation in the high-density region is not known.

4 Conclusion
We have presented a new approach for observing the coherence associ-
ated with the BEC of the biexciton system in CuCl. A phase-conjugate
signal was used as a probe to monitor the macroscopic occupation of
the biexcitons at K=0 and the relevant coherence under the high-density
injection of the biexcitons of incoherent nature. A remarkable increase of
 BEC ofBiexcitons in CuCl 495
the signal intensity at the critical density for BEC and a blue shift of the
K = 0 biexciton resonance were observed. These results are attributed to
the presence of the Bose-Einstein condensate of biexcitons.

Acknowledgements. We acknowledge the support of CNRS and JSPS

through the joint collaborative program. We would like to express our
sincere thanks to Dr A. Mysyrowicz for his kind support and valuable
discussions at every stage of this work. We also thank Profs. T. Itoh, M.
Inoue, E. Hanamura, and M. Kuwata-Gonokami and Drs A. Ivanov and
T. Hatano for fruitful discussions. This work was partially supported by
the Grant-in-Aid for Scientific Research from the Ministry of Education,
Science and Culture.

References
[I] T. Mita, K. Sotome and M. Ueta, J. Phys. Soc. Jpn. 48, 496 (1980).
[2] H. Haug, and E. Hanamura, Physics Reports 33, 209 (1977).
[3] L.L. Chase, N. Peyghambarian, G. Grynberg, and A. Mysyrowicz,
Phys. Rev. Lett. 42, 1231 (1974).
[4] N. Peyghambarian, L.L. Chase, and A. Mysyrowicz, Optics Comm.
41, 178 (1982).
[5] M. Hasuo, N. Nagasawa, and A. Mysyrowicz, Phys. Stat. Sol.(b) 173,
255 (1992); M. Hasuo, N. Nagasawa, T. Itoh, and A. Mysyrowicz,
Phys. Rev. Lett. 70, 1303 (1993).
[6] G. Mizutani, and N. Nagasawa, J. Phys. Soc. Jpn. 52, 2251 (1983);
L.L. Chase, M.L. Claude, D. Hulin, and A. Mysyrowicz, Phys. Rev.
A 28, 3969 (1983).
[7] Y. Masumoto, and S. Shionoya, J. Phys. Soc. Jpn. 49, 2236 (1980).
[8] T. Hatano, doctoral thesis (The University of Tokyo, 1993).
[9] H. Akiyama, T. Kuga, M. Matsuoka, and M. Kuwata-Gonokami,
Phys. Rev. B 42, 5621 (1990).
[10] T Ikehara, and T. Itoh, Phys. Rev. B 44, 9283 (1991).
[II] L.V. Keldysh, and A.N. Kozlov, Sov. Phys. JETP 27, 521 (1968).
[12] M. Inoue, and E. Hanamura, J. Phys. Soc. Jpn. 41, 1273 (1976).
 24
The Influence of Polariton Effects on BEC
of Biexcitons
A. L. Ivanov and H. Haug
Institut fur Theoretische Physik
J.W. Goethe-Universitat
Robert-Mayer-Str. 8, D-60054 Frankfurt/Main
Germany

Abstract
The influence of the radiative decay of excitonic molecules on a possible
quasiequilibrium Bose-Einstein condensation (BEC) of excitonic molecule's
is examined with respect to the radiative renormalization of the excitonic
molecule energy (excitonic molecule Lamb shift). For the excitonic molecule
wave function, a Schrodinger equation which contains polariton effects is
derived and analyzed. Both the inverse excitonic molecule radiative lifetime
ym and the biexciton Lamb shift Am depend strongly on the total excitonic
molecule momentum K. The energy renormalization Am(K) leads to the ex-
citonic molecule effective mass modification and can result in a camel-back
structure at K = 0, which opposes a BEC of excitonic molecules at K= 0.

1 Introduction
Observations of a quasiequilibrium Bose-Einstein condensation (BEC)
of excitonic molecules (EM) have been attempted [1, 2, 3] following
its theoretical prediction [4, 5] (for reviews see, e.g. [6, 7]). Recent
approaches [8, 9] with high-precision techniques renewed the interest in
this phenomenon. Traditionally, one tries to detect BEC of the EMs
in luminescence. In the new approach the appearance of coherence in
the thermal system of the EMs has been tested by means of four-wave
mixing and treated as a fundamental manifestation of BEC. (See the
paper by Hasuo et al. in this book.) Both optical methods for the
BEC detection imply that the optical transition to the corresponding
intermediate exciton (IE) state is dipole-active. In this case the EM state
is unstable against optical decay with a "giant" oscillator strength [6].
The short radiative lifetime rm of the EM is a very serious obstacle for

496
 The Influence of Polariton Effects on BEC of Biexcitons 497
BEC. The influence of this factor has been considered exclusively in the
kinetics of the Bose condensate formation from arbitrary initial non-
equilibrium distributions of EMs (see, e.g. [10]). The strong interaction
with electromagnetic field also modifies the dispersion of the elementary
excitations, however. A well-known example is the Lamb shift of the
resonance energy of a dipole-active two-level system. Another example
is the polariton spectrum of dipole-active IEs. Here we study the EM
Lamb shift and show that this effect strongly influences the BEC of EMs.

2 Self-consistent Biexciton Wave Equation with Polariton Effects

Neglecting the interaction with an electromagnetic field, the complete
set of EM eigenfunctions ^ ( p ) and corresponding energies Q™ K can
be obtained from the following Schrodinger equation in the momentum
representation (with h = 1):
£ [(Hx (p + f) + Hx (-p + f)) df<fl + Wl2 (p - pi)]
Pi

Here, p and K are the momenta of the relative and center-of-mass motion
of the EM, respectively. H*(k) = k2/2M is the kinetic energy of an IE,
and W\2(p — Pi) is the attractive potential between two singlet IEs with
opposite internal electron spin orientation. Due to the quadratic IE
dispersion, the center-of-mass motion splits off, and the wave function
and energies of the relative motion are independent of K. This equation
in the underlying e-h picture has often been used to calculate the EM
properties variationally [11, 12, 13, 14]. In order to investigate the
influence of the polariton effects on the interior structure of EM we
derived self-consistently [15] an EM wave equation from the complete
e-/i-photon (e — h — y) Hamiltonian :
£ [{HP (p + f) + H<> (-p + f)) <5p,Pl + Wn (p, Pl ,K)]
Pi

with Hp(k) = co~(k), where co±(k) is the dispersion of the upper (+) or

m-
lower (-) polariton branch, which are given by the roots of
2

n (3)
€Q(O2 CO2 + COtp2/M — CO2
Here, coy(p) is the photon frequency and eo is the background optical
dielectric constant. The polariton parameter Qc is defined in terms of
498 A. L. Ivanov and H. Haug
the longitudinal-transverse (t — t) splitting co/f and the transverse exciton
frequency cot:
Q2C = 2cott(ot. (4)

The effective interaction Wn is determined by the x components of the

two interacting polaritons (with k1^ = ±p + f):
-pi), (5)
where
/(p,K) = <tr (k + ,co- (k + )) (j>~ (k-,©- (k-)), (6)
and

These x components satisfy the following conditions:

c/>+(p,co+(p))>0;
O; (8)

In Eq. (2) only the lower polariton dispersion branch is taken into
account. In the following we will consider only the ground states of the
IE and EM (J=l). The linear equation, Eq. (2), contains the radiative
decay of an EM state, as well as the renormalization of the EM energy
(EM Lamb shift) due to the polariton effects. When an IE in the
EM acquires a small momentum within the optical range (sectors A1A2
and B1B2 in Fig. 1), the EM undergoes a radiative annihilation. In
other words, the linear part of the lower polariton dispersion branch
corresponds to the case without a bound EM state. The radiative decay
of an EM stems from the polariton dispersion of its IE rather than from
the usual tunneling process out of a metastable state. In our picture
the EM luminescence is a continuous evolution of the EM interior state
rather than a discrete act of the EM optical conversion to y and IE.

3 Biexciton Lamb-Shift
The non-stationary solution of Eq. (2), ^(p), represents an outgoing
spherical wave due to the radiative decay of a EM state with Q£ =
Q£ + Am(K) - iym(K). Here, Q£ = Q£ =o + K2/2Mm is the unperturbed
EM energy with the EM translational mass Mm = 2M. For small relative
 The Influence of Polariton Effects on BEC of Biexcitons 499

03
GO

Fig. 1. Illustration of the optical decay of an EM.

momenta which belong to the optical range of the polariton spectrum, the
EM wavefunction (EMWF) ^(p) is strongly modified. This modification
of the EMWF cannot be treated perturbationally. Here the EM Lamb
shift Am(K) and the inverse radiative lifetime ym(K) are of the same order,
while in atomic optics the radiative energy shifts are always substantially
smaller than corresponding inverse lifetimes. In order to treat Eq. (2) we
decompose the EMWF in

where *F(p) is the known solution of unperturbed Eq. (1), while *Fi(p,K)
describes the influence of the polariton effects and is large only in the
optical range, that is, for p « a~\ where a m is the EM radius. For these
small momenta one neglects the momentum dependence of the attractive
potential Wn{\> — Pi) = Wnity =const.
Substituting Eq. (9) in Eq. (2), one finds the following integral equation
for ^ ( p ) in a D-dimensional system:

+/(p,K) W12(0) j = 0, (10)

500 A. L. Ivanov and H. Haug
where

£(p, K) = A (p, K) - £f (p, K) + emf (p, K)

A(p,K) = co- (p + K/2) +Q>- (-p + K/2) - f i g . (11)
m 2
Here, e = Q£ - 2a>, - X /4M < 0 is the unperturbed EM binding
energy. Eq. (10) is a Fredholm integral equation with separable kernel.
The corresponding solution is given by

where C = TTO with

(13)

(14)

The functions *F(p) and ^(pjK), as well as the basic Eq. (2), are symmet-
ric with respect to the substitution p —• —p. According to Eq. (12), the
roots po = Po(K) of the equation A(p, K) = 0 give rise to singularities in
^ ( P J K ) . These singularities describe the two outgoing polariton waves
with momenta po 4- K/2 and — po + K/2 in the decay of the EM with
K. A(p,K) = 0 is the energy-conservation law of this decay. The graphic
solution of this equation is presented in Fig. 2.
According to Eq. (2) the total EM momentum K influences the relative
motion of the IEs in the EM. The polariton dispersion makes *Fi(p,K),
Am(K), and ym(K) K-dependent in the optical range K < 2po(K = 0).
Formally, the solution (12) can be found for an arbitrary EM energy
Q£. In order to find the true EM energy for a given K, we change
Q£ continuously around the unperturbed EM energy Q£ and find the
value which minimizes the outgoing part / \x¥i(p,K)\2dDp/(2n)D and
maximizes the EM radiative lifetime. Although the wave functions (12)
cannot be normalized, one can compare the relative contributions of the
outgoing parts. For simplicity we present here only numerical solutions
for one-dimensional EM systems with D = 1. The fraction of the phase
space in which polariton effects dominate over the Coulomb IE-IE
interaction is of the order of (ampo(K = 0))D. Therefore, the influence
of polariton effects on the EMWF is much stronger in one dimension
than in three dimensions. We use the parameters of GaAs/GaAlAs
quantum-well wires [16]. For the attractive IE-IE potential
 The Influence of Polariton Effects on BEC of Biexcitons 501

m
Fig. 2. Graphic solution of Eq. (12). The points 0 and O\ correspond to outgoing
polaritons.

we use the simple form of the deuteron model [17] in dimensionless units
(normalized with the EM Rydberg and the EM radius)

Wn(p-Pi) = (15)

Eq. (1) can be treated with the trial EMWF

C
(16)

For P = 4.0, the parameter a = 0.5 minimizes the energy and gives
the binding energy em — 0.75. In dimensional units these parameters
correspond to em = -5 meV and EM Bohr radius am = 73.9A. The
accuracy of the variational procedure is about 2.5% [17].
502 A. L. Ivanov and H. Haug

100

50
-1.0

Fig. 3. Dependencies of the relative contributions of the outgoing parts of the

EMWF (see Eq. (17)) on the energy Q£ for various K values: 0 (full line),
3 x 105 cm' 1 (long dashes), 15 x 105 cm" 1 (medium dashes), 20 x 105 cm" 1 (short
dashes). The following parameters for GaAs/GaAlAs quantum wires have been
used: cot = 1.515 eV, e0 = 12.5,co^ = 0.08 meV, M = O.52mo,em = - 5 meV,
polariton absorption yx = 10 cm" 1 .

The set of curves

F=F p,ft£) \2/ (17)

which represent the relative contributions of the outgoing parts for a

given K, is shown in Fig. 3, where the points Ot correspond to the true EM
energy. The dependences ym = ym(K) and AQm(K) = h2K2/2Mm + Am(K)
are plotted in Fig. 4.
The EM Lamb shift Am(K) gives rise to a camel-back structure with a
relative maximum at K = 0 and a minimum at Ko ^ 2po(K = 0), where
po(K = 0) = 2.76 x lO^m" 1 (points O in Fig. 2). Experimentally, the EM
luminescence can be observed only for EMs with K > po(K = 0) both
due to the strong polariton dispersion and due to the interference with the
pump [2,3,7,8,9]. These EMs are strongly coupled to the electromagnetic
field which results in a large EM Lamb shift, as seen in the camel-back
 The Influence of Polariton Effects on BEC of Biexcitons 503

On r3

CD

-10-

-15
30

K (xlO cm"1)

Fig. 4. Biexciton energy AO£ = X2/2Mm + Am(K) (full line) and inverse radiative
lifetime ym(K) (dashed line) versus total momentum K. All parameters are the
same as in Fig. 3, except for yx = 500 cm"1.

structure around K = 0. In our theory this behavior has the following

explanation. Increasing the total EM momentum K from 0, one reaches
the point at which the second lower branch —a>~(—p + K/2) (left branch
of —co"(—p+K/2) in Fig. 2) touches and crosses the first polariton branch
co~(p+K/2) (right branch in Fig. 2). Thus, one receives beyond the points
0i additional solutions of the energy-conservation law A(p, K) = 0. The
touching point corresponds to vs = dA(p,K)/dp = 0 and gives rise to a
van Hove singularity. This singularity at Ko ^ 2po(K = 0) is responsible
for a strong coupling of the EM with Ko to the electromagnetic field
and results in a "jump" of the EM energy to smaller values (see Fig. 4).
For two dimensions/three dimensions the van Hove singularities are a
continuous family of critical points in the form of a circle (sphere) with
radius Ko. So the character of the van Hove singularity is independent of
the dimensionality of EM. The large spatial dispersion of Am(K) in the
optical range K ^ po(K = 0) allows one to discriminate the EM Lamb
504 A. L. Ivanov and H. Haug

Fig. 5. Schematic representation of the "Mexican hat" structure in 2D.

shift from the other EM renormalizations due to LO-phonon, IE-EM

and EM-EM interactions.
In CuCl, recent experiments [18]-[20] measured a radiative EM lifetime
of 50 — 90 ps under relatively weak resonant excitation. This lifetime
is much shorter than any other values previously reported and has
been attributed to the giant oscillator strength of the IE-EM optical
transition. Moreover, a strong dependence of ym(K) on K has been
observed [19]. Our theory gives a qualitative explanation (see Fig. 4) of
these experimental results.

4 Discussion
We expect that in a three-dimensional system such as bulk CuCl a
camel-back or, rather, "Mexican hat" structure will also be obtained
for the renormalized energy, particularly because the heavy hole mass
of this material results in a heavy EM mass of Mm = 5.2mo, which
favors the minimum at Ko. The magnitude of the minimum Am(X0),
however, is expected to be considerably smaller than in one dimension.
The appearance of this minimum will eliminate the possibility of a BEC
of EMs at K = 0. But at the same time a condensation in the minimum
at KQ ~ 2po(K = 0) (see Fig. 5) is not possible, because the relation for
 The Influence of Polariton Effects on BEC of Biexcitons 505
the total number of EMs

yields for all finite temperatures /x(iVm, T) < 0, if we scale the EM min-
imum energy at Ko to be zero. Our results are consistent with the
experimental data of Refs. [2, 3]. In these experiments a strong accumu-
lation of the EMs at K ~ 2po(K = 0), rather than at K = 0, has been
detected.

Acknowledgments. We appreciate valuable discussions with L.V. Keldysh,

A. Mysyrowicz, N. Nagasawa, M. Hasuo and D. Snoke. This work
has been supported by the Volkswagen Stiftung and by the NATO
Collaborative Research Grant No. 930084.

References
[I] N. Nagasawa, T. Mita, and M. Ueta, J. Phys. Soc. Jpn. 41, 929
(1976).
[2] L.L. Chase, N. Peyghambarian, G. Gryndberg, and A. Mysyrowicz,
Phys. Rev. Lett. 42, 1231 (1979).
[3] N. Peyghambarian, L.L. Chase, and A. Mysyrowicz, Phys. Rev. B
27, 2325 (1983).
[4] E. Hanamura and M. Inoue, in Proc. 11th International Conference
on the Physics of Semiconductors, Warsaw, (1972) p. 711.
[5] S.A. Moskalenko, Fiz. Tverd. Tela 4, 276 (1962)
[6] E. Hanamura and H. Haug, Physics Reports 33, 209 (1977).
[7] A. Mysyrowicz, J. de Physique (Supplement) 41, C -281 (1980).
[8] M. Hasuo, N. Nagasawa, and A. Mysyrowicz, Phys. Stat. Sol. (b)
173, 255 (1992).
[9] M. Hasuo, N. Nagasawa, T. Itoh, and A. Mysyrowicz, Phys. Rev.
Lett. 70, 1303 (1993).
[10] M. Ueta, H. Kanzaki, K. Kobayashi, Y. Toyozawa, and E. Hana-
mura, Excitonic Processes in Solids (Springer-Verlag, Berlin, 1986),
Springer Series in Solid-State Sciences 60, p. 110.
[II] O. Akimoto and E. Hanamura, Solid State Comm. 10, 253 (1972).
[12] A.I. Bobrysheva, M.F. Miglei, and M.I. Shmiglyuk, Phys. Stat. Sol.
(b) 53, 71 (1972).
[13] W.F. Brinkman, T.M. Rice, and B. Bell, Phys. Rev. B 8, 1570 (1973).
[14] M.I. Sheboul and W. Ekardt, Phys. Stat. Sol. (b) 73, 165 (1976).
506 A. L. Ivanov and H. Haug
[15] A.L. Ivanov and H. Haug, Phys. Rev. B 48, 1490 (1993).
[16] L. Banyai, I. Galbraith, C. Ell, and H. Haug, Phys. Rev. B 36, 6099
(1987).
[17] S. Fliigge, Practical Quantum Mechanics (Springer, Berlin 1974), p.
196.
[18] H. Akiyama, T. Kuga, M. Matsuoka, and M. Kuwata-Gonokami,
Phys. Rev. B 42, 5621 (1990).
[19] M. Hasuo, N. Nagasawa, and T. Itoh, Opt. Commun. 85, 219 (1991).
[20] T. Ikehara and T. Itoh, Sol. Stat. Comm. 79, 755 (1991).
 25
Light-Induced BEC of Excitons and
Biexcitons
A. I. Bobrysheva and S. A. Moskalenko
Institute of Applied Physics
Academy of Sciences of Moldova
5 Academy Street, Kishinev
Moldova

Abstract
We review the theory of coherent pairing of excitons and biexcitons in two-
dimensional and three-dimensional semiconductor structures.

1 Introduction
The coherent pairing of bosons, formally analogous to Cooper pairing
of electrons in superconductors, and the coexistence of particle and pair
Bose-Einstein condensation (BEC) was first studied in Refs. [1-4]. BEC
of excitons in semiconductors was investigated using the approximation
of coherent pairing of the electrons and holes [5]. Later on, the idea of
coherent pairing of bosons was applied to excitons [6-8]. In Ref. [6],
it was assumed that the boson pairs consist of two electron-hole pairs
instead of two excitons, and it was suggested that the new collective
state is a BEC even though the two-pair bound state does not exist. It
was proved independently [7, 8] that in a system of bosons and excitons
with an attractive pair interaction sufficient for biexciton formation, the
coherent pairing of excitons with momenta k and —k coincides with the
BEC of biexcitons with zero translational momentum.
The possibility of particle and pair BEC in a system of four species
of excitons with different values of spin projections of the electron and
hole was studied in Ref. [9]. It is known [10, 11] that the interaction of
an exciton of either type with other exciton species is repulsive on the
average, but that pairs of excitons with antiparallel spins of electrons and
holes interact attractively, and biexcitons can be formed. In two limiting
cases, when the exciton-exciton interaction energy W is much greater or
smaller than ortho-paraexciton splitting A0_p, it was shown that the BEC

507
508 A. I. Bobrysheva and S. A. Moskalenko
of biexcitons persists without collapse due to the internal spin structure
of excitons. In the case where A0_p < W the biexciton is formed by
all four species of excitons. When A0_p > W in a system of para-
and orthoexcitons, BEC of paraexcitons and biexcitons formed by two
orthoexcitons with zero total spin can take place simultaneously. Taking
the CuCl crystal as an example, it was proved [12] that in the presence of
nonresonant polarized laser radiation, which induces the polarized exci-
tonic BEC, coherent pairing of excitons takes place. Two induced BECs
coexist: that of polarized excitons with wave vector ko, and the biexci-
tonic one with wave vector 2ko. It was shown that the biexciton becomes
anisotropic in the sense that four species of excitons T$x, Tsy, r$z and
F2 do not participate equally in biexciton formation, but have different
weights. That is, the two-photon biexciton absorption becomes polarized.
In the present paper, this phenomenon is studied in a wider class of
crystals with different dimensionalities d = 2,3, in particular in quasi-
two-dimensional semiconductor structures with quantum wells such as
GaAs/AlxGai_xAs and ZnSe/(Zn,Mn)Se and three-dimensional semi-
conductor crystals such as CuCl. Optical methods which allow the
observation of the new electronic state of the crystal are of interest.
Therefore biexciton hyper-Raman scattering and luminescence under the
above-mentioned conditions are also studied.

2 The Exciton-Exciton Interaction

We first construct the exciton-exciton interaction Hamiltonian Hex-ex for
crystals which have a band structure which consists of a spin-degenerate
conduction band and twofold-degenerate valence bands. This model
suits the above-mentioned quantum wells (QW) which have two valence
bands with a maximum at k = 0 (Tv6 symmetry for heavy holes, and Tv7
for light holes) and the bottom of the conduction band also at k = 0
(symmetry Tc6). The CuCl bulk crystal band structure also consists of a
valence band with Tv7 symmetry and conduction band with Tc6 symmetry
at k = 0. Hex-ex can be constructed as follows:
16
Hex.ex = ( 2 L V £ E vm(k1,k2,k3,k4)
ro=l ki,k2,k3,k4

t )<Mki + k2, k3 + k4) (1)

Here ¥+ is the two-exciton creation operator characterized by the total
momentum of four fermions Jt and its projection Mt; vm is the Fourier
 Light-Induced BEC of Excitons and Biexcitons 509
transform of exciton-exciton interaction energy; Ld is the surface area of
two-dimensional system or the volume of the three-dimensional one.
Sixteen two-exciton wave functions *FTO(S,SZ) for a model with two
spin-degenerate bands and A0_p <C W have been constructed by Culik
[13] in terms of four electron and hole Fermi operators. Here S9SZ
mean the total spin and its projection, respectively, of four fermions.
They can be easily generalized for our case by a simple interchange of
the spin projections Sh of the holes by their momentum projections J\.
After that, four species of exciton operators characterized by the spin
projection S% of the electron and by J\ are introduced and expressed in
terms of operators d^Y^ for excitons belonging to a certain row £ of the
irreducible representation F. The final step is to represent the two-exciton
operator as a linear combination of the products of two operators a^Y^.
In the effective-mass approximation, vm obtained for the case of two spin-
degenerate bands for three-dimensional [14] and two-dimensional [15]
structures remain valid in our twofold-degenerate valence band model.
Only one function xFm results in an attractive interaction, implying v =
va,a. The remaining functions give rise to repulsive interaction and v == v v .
The subscripts a, a(s9 s) (the first symbol refers to electrons and the second
to holes) describe the symmetry of xFm with respect to the permutation
of creation operators of two identical fermions. For small k values the
vs,s(ki, k2, k3, k 4 ) ^ -v a , a (ki, k2, k3, k 4 ) and vm(k, k, k, k) = vm(0). For three-
dimensional systems vS)S(0) = (26/?>)nRy wa\D, but for two-dimensional
ones v w (0) = %nRy2Daw[\ - (315/2 1 2 )TT 2 ] .
In the case of the above-mentioned quantum wells, the four types
i = 1,2,3,4 of HH-excitons are characterized by the rows of the irre-
ducible representation E: £(x), E(y) and onefold representations A2,A\
respectively. The types of LH-excitons are E(x\ E{y), Bi, B\, and for
bulk CuCl F 5 (x), F 5 (Y), F 5 (z), F 2 . In terms of the exciton creation
operators a j ^ , the Hex-ex one obtains has an identical form in these
three cases. It can be written as follows:

Hex_ex = 4
ki,k2,q tr

(2)
510 A. I. Bobrysheva and S. A. Moskalenko

3 Coherent Pairing of Excitons Induced by Polarized Laser Radiation

The full Hamiltonian is written as
H = Ho + Hex-ex + Hex-R, (3)
where
4
^ (4)

Hex-ex has the form given by (2). We assume that laser radiation is
polarized along the x axis and excites the exciton mode E(x) with
the wave vector ko; Hex-R describes this exciton-electromagnetic field
interaction.
The induced BEC of polarized excitons in the state ko is introduced
in the Hamiltonian by Bogoliubov's translation operation. The biexciton
creation operator has the form

*2+ko = E CiL d/2

~ E ^)<ko+q4o-q > (5)
i=l q

where the function of relative motion O(q) and the coefficients Q obey
normalization conditions. Therefore, BEC of biexcitons in the state 2ko is
introduced by the operation of coherent pairing of excitons (ko+q,ko—q)
of the same type i.
The coefficients of the (uuvt) transformations are given by

Mq) = cosh(2C I y^O(q)), (6)

and
i;I(q) = sinh(2C i y^a)(q)), (7)

where n^ = N^L~d is the condensate biexciton density. We note that

without laser radiation, the absolute values of C, are equal.
The stability conditions of the new ground state of the system lead to
the conclusion that Ci = C3 = —C4 and allow the possibility of obtaining
equations which, along with the normalization conditions, determine both
C\, Ci as well as the condensate exciton n j ^ and biexciton n^ densities.
In the linear v,• approximation, we obtain:

= (|< v >\ TA)(y/2^3(\< v >\ +A)2 + (|< v >\ -A) 2 )" 1 , (8)
 Light-Induced BEC of Excitons and Biexcitons 511
where 1, 2 refer to signs minus, plus, respectively; < v > is the mean
potential energy of the internal motion; A is the biexciton detuning. It
follows from Eq. (7) that for A ^ 0, the number of excitons of each of the
four types forming the biexciton are different. In this sense the biexciton
is polarized. At A = 0, fast mutual conversions of the excitons remove
this inequality.

4 The Biexciton Two-Photon Absorption, Hyper-Raman Scattering and

Luminescence
The probability P of a two-photon biexciton transition is obtained in
the second order of perturbation theory. The biexciton wave function is
given by (5). If two probe photons have the same polarization et, then
the transition probability depends on the orientation of et relative to the
x axis along which the laser radiation is polarized. The relative value
P/Po is equal to 8 | Cx | 2 if et || x, and 8 | C2 | 2 if et _L x. Here Po is the
transition probability without pumping.
Let us consider the process of absorption of two photons with wave
vectors ko and energies in the spectral region £ m (0)/2, as a result of
which a photon with wave vector q# and polarization e# is emitted and
either a longitudinal L or transverse T exciton is created. The probability
PR of this hyper-Raman scattering process is described by third order
perturbation theory. It is enough to take into account only the T -exciton
and biexciton as intermediate states. It is assumed that the incident
radiation propagates along the z axis and is polarized along x. Then for
a (001) quantum well, we obtain P£ = 0, P$ ~ | CXC2 | 2 (e*y0)2, qR || x°.
2
The x° and y° are unit vectors. In the case of CuCl, we find P£ ~\ C\C 2 | ,
and the intensity of scattered light reaches a maximum in the direction
perpendicular tok 0 . In turn, P* ~ | Cx | 2 (| Cx \2 (eRx°)2+ \ C2 | 2 (e*y 0 ) 2 ).
The intensity of scattered light reaches a maximum in the direction
parallel to ko.
We would expect that the biexciton polarization manifests itself in the
M luminescence band in the case of two simultaneously acting pump and
probe beams. The pump beam produces the biexciton polarization, but
the weak unpolarized probe beam with frequency com/2 populates the
biexciton states. In CuCl, one has Pf* ~\ C2 | 2 , P™ H Q | 2 (e/x°)2 + |
C2 I2 (e/y0)2. In a (001) quantum well, P^ = 0, P^ ~ | C2 | 2 (e/y0)2. Here
e/ is the luminescence polarization vector.
512 A. I. Bobrysheva and S. A. Moskalenko
References
[I] J.G. Valatin and D. Batler, Nuovo Cimento 10, 37 (1958).
[2] A. Coniglio and M. Marinaro, Nuovo Cimento 48B, 249 (1967).
[3] A. Coniglio and F. Manciniand M. Maturi, Nuovo Cimento 63B,
227 (1969).
[4] W.A.B. Evans and Y. Imry, Nuovo Cimento 63B, 155 (1969).
[5] L.V. Keldysh and A.N. Kozlov, Zh. Eksp. Teor. Fiz. 54, 978 (1968).
[6] C.A. Mavroyannis, Phys. Rev. 10, 1741 (1974).
[7] P. Nozieres and D. Saint James, J. Physique 43, 1133 (1982).
[8] A.I. Bobrysheva and S.A. Moskalenko, Fiz. Tverd. Tela 25, 3282
(1983).
[9] A.I. Bobrysheva, S.A. Moskalenko, and Yu.M. Shvera, Phys. Stat.
Sol. (b) 147, 711 (1988).
[10] A.I. Bobrysheva, S.A. Moskalenko, and V.I. Vybornov, Phys. Stat.
Sol. (b) 76, K51 (1976).
[II] A.I. Bobrysheva and S.A. Moskalenko, Phys. Stat. Sol. (b) 119, 141
(1983).
[12] A.I. Bobrysheva, S.A. Moskalenko, and Hoang Ngok Cam, Zh.
Eksp. Teor. Fiz. 103, 301 (1993).
[13] F. Culik, Czech. J. Phys. B 16, 194 (1966).
[14] A.I. Bobrysheva, M.F. Miglei, and M.I. Shmigliuk, Phys. Stat. Sol.
(b) 53, 71 (1972).
[15] A.I. Bobrysheva, V.T. Zyukov, and S.I. Beryl, Phys. Stat. Sol. (b) 101,
(1980).
 26
Evolution of a Nonequilibrium Polariton
Condensate
I. V. Beloussov and Yu. M. Shvera
Institute of Applied Physics
Academy of Sciences of Moldova
5 Academy Street, Kishinev
Moldova

Abstract

We review the theory of the polariton condensate taking into account

exciton-exciton interactions.

It is known [1] that coherent electromagnetic radiation resonant with an

isolated exciton energy level excites in the crystal a coherent polariton
wave with the wave vector ko i= 0 — the non-equilibrium polariton
condensate. Different scattering processes accompanying its propagation
lead to the loss of the initial coherence of the polariton wave, complete
or partial depletion of the condensate, excitation of polaritons with wave
vector k ^ 0, and other phenomena.
In the present paper the effect of exciton-exciton scattering processes
on the properties of a coherently excited polariton system are discussed.
This scattering mechanism is of considerable interest due to recent experi-
mental investigations [2] and many interesting results (see e.g. Refs. [3,4])
obtained in theoretical study of dynamic and kinetic processes in a system
of interacting polaritons.
According to Refs. [4, 5], exciton-exciton scattering is very important
when coherent polaritons are excited in a certain spectral region in which
energy and momentum conservation laws allow real processes of two-
quantum excitation of polaritons from the condensate. These processes
lead to the instability of the condensed state of the polariton system.
The existence of this spectral region situated around the isolated exciton
resonance is due to the peculiarities of the polariton dispersion relation.
In [5] the energy spectrum of non-condensate polaritons, arising as
the result of decay of the coherent polariton wave, is studied. According
to [5], in some regions of k-space the polariton modes will be damped.

513
514 I. V. Beloussov and Yu. M. Shvera
The investigation reported in [5] was based on a model formally analo-
gous to that used by N.N. Bogoliubov in Ref. [6] to study the equilibrium
system of a weakly non-ideal Bose gas. In the non-equilibrium situation
considered in [5], in which decay of the polariton condensate and exci-
tation of non-condensate polaritons take place, this model is adequate
for the experimental situation only at the initial stage of the condensate
decay, when the number of condensate polaritons is still much greater
than the total number of non-condensate polaritons. The system is es-
sentially non-stationary, while a real energy spectrum implies a steady
state of the system (see e.g. Ref. [7], p.46). Thus the results of Refs. [4, 5]
for the polariton energy spectrum based on the above-mentioned model
are only approximate.
Because of the essential non-stationarity of the processes in the system,
the methods of non-equilibrium statistical mechanics must be used to
describe it adequately. The derivation of equations that describe the
kinetics of the polariton condensate decay and the excitation of quantum
fluctuations have some specific features due to the degeneracy in the
system. As the energy and wavevector of two non-condensate polaritons
can be equal to the energy and wavevector of two condensate polaritons,
respectively, there is degeneracy of two-particle states. Moreover, the
presence of the condensate in the system also leads to degeneracy due to
its macroscopic amplitude [8].
The correct description of the system with degeneracy requires the
introduction of abnormal (or off-diagonal) distribution functions [9]
an<
l ^k = (^k^ko-k), together with the usual (normal) ones
). Here ®jj~(<l>k) are Bose operators of creation (annihilation)
of a polariton on the lower branch with wave vector k. The appearance of
abnormal averages and the coherent part of the polariton field indicates
a breaking of the selection rules connected with the gauge invariance
of the system [9-11]. It can take place as a result of the action of
external classical sources, spontaneously, or because of non-invariant
initial conditions due to the action of external sources.
An earlier attempt to obtain kinetic equations for polaritons excited in
semiconductors by an external classical field has been given in Refs. [12,
13]. However, in this work the degeneracy of two-particle states was
not taken into account, and the abnormal distribution functions Ft were
not introduced. As a result, one expects that the equations obtained in
Refs. [12, 13] will lead to unphysical singularities.
Kinetic equations describing the evolution of partially coherent polari-
tons which take into account the degeneracy were obtained in [14, 15]
 Evolution of a Nonequilibrium Polariton Condensate 515
using the Keldysh method [16], presented in terms of functionals. They
coincide with the equations obtained in [17] using the non-equilibrium
statistical operator method [J8] and do not possess unphysical singu-
larities. One can find extended versions of Refs. [14, 15, 17] in Refs.
[19-21].
According to Refs. [15, 17], the kinetics of partially coherent polaritons
are described in the Born approximation by a closed set of nonlinear
integro-differential equations for the coherent part of polariton field ^ ^
and the normal n^ = N^ — ^kj^kol 2 and abnormal /k = F^ — ^Fj^
distribution functions. In the absence of quantum fluctuations described
by the functions n^ and /k, the equations become identities, and the
equation for ^Fka (a = 1,2 is the label of the polariton branch) coincides
with that obtained in Ref. [22] for a interacting system of coherent
excitons and photons. In the case when *Fko = 0 and /k = 0, the
equations obtained in [15, 17] reduce to the usual kinetic equation for
the distribution function N^ (see e.g. Ref. [23]).
The right-hand sides of the equations obtained in [15, 17] include terms
linear in the exciton-exciton interaction strength v, and terms propor-
tional to v2. Terms proportional to v correspond to the self-consistent
field approximation (SCFA), which neglects the higher correlation func-
tions. This approximation is sufficient to describe the initial stage of
evolution. It is shown [15] that depletion of the coherent part ^F^ a n d
excitation of quantum fluctuations n^ and / ^ in the condensate mode
occurs at this stage. Thus it partially loses its coherence. There exist two
regions of wave vector k ^ ko. In the first one, describing the instability
region, the number of polaritons increases monotonically; in the second
one, it is a periodic function of time with an amplitude decreasing with
the distance from the instability region.
As usual, the terms proportional to v2 describe the difference between
the number of processes (per unit of time) of polariton creation and
annihilation in a state with wavevector k. They only arise if non-
condensate polaritons exist in the system. They describe the evolution
which is slower than that described by terms proportional to v, which
correspond to resonant scattering with participation of two condensate
polaritons.
It should be mentioned that the SCFA takes into account the influence
of the excited non-condensate polaritons on the condensate. On the
other hand, it describes the fastest processes in the system. So one should
expect the establishment of steady state in the isolated polariton system
after the time interval T ~ h/vn (n is the condensate initial density).
516 I. V. Beloussov and Yu. M. Shvera
Terms in the kinetic equations proportional to v2 are responsible for its
further evolution.
Using considerations similar to those used in [24], we have found the
exact steady-state solution of the evolution equations given by the SCFA.
Formally, it has the following nature. In the Heisenberg representation,
the SCFA corresponds to the Hamiltonian [15] which has operator
terms O£^io-k w ^^ time-dependent coefficients £ = ^F^ + A (here
A = J2kfa)- A t t h e initial stage of evolution, when I^FkJ2 > |A|,
these terms are responsible for the appearance and development of the
polariton condensate instability [4, 5]. In the steady state, the two terms
in £, cancel each other, so that { = 0. Thus these terms disappear from
the Hamiltonian.
We note that in [4, 5], the influence of non-condensate polaritons on the
condensate was not taken into account, and the term A in the expression
for £ did not appear. As a result, in [4, 5] the possibility of obtaining a
steady-state solution, which is the result of mutual compensation of ^Fj^
and A, did not exist.
From the physical point of view, the results obtained here mean
that while the condensate is being depleted, and polaritons are being
excited in the instability region (modified by the concentration-dependent
corrections), the backward processes (in which scattering of two non-
condensate polaritons creates two polaritons in the condensate) become
more important. It results in slowing down of the parametric instability,
and finally in the establishment of a steady state. This state corresponds
to a dynamic equilibrium between the condensate and non-condensate
polaritons.
The steady state is characterized by the renormalized frequency of the
polariton condensate hco^ = hQ,^ + yjlvno and renormalized energies of
non-condensate polaritons E^ = hQ^ + 2vno (here Q^ is the frequency
of non-interacting polaritons). This result differs crucially from that
obtained in [4, 5]. Distribution functions of non-condensate polaritons
are localized in the regions of k-space, as determined by the resonant
condition E^ + £2ko-k — 2fta>ko = 0-
Future investigations of this problem require numerical analysis of the
set of integro-differential equations corresponding to the SCFA. Another
important problem, in our opinion, is the study of polariton kinetics
under the action of an external field.
 Evolution of a Nonequilibrium Polariton Condensate 517
We wish to thank Dr. A. Ivanov and Dr. S. Tikhodeev for helpful
discussions.

References
[I] A.L. Ivanov and L.V. Keldysh, Zh. Eksp. Teor. Fiz. 84, 404 (1983).
[2] M.V. Lebedev, Zh. Eksp. Teor. Fiz. 101, 957 (1992).
[3] S.A. Moskalenko, Introduction to the Theory of High Density Exci-
tons (§tiinja, Kishinev, 1983).
[4] M.I. Shmigliuk and V.N. Pitei, Coherent Polaritons in Semiconductors
(§tiin{a, Kishinev, 1989).
[5] V.N. Pitei, M.I. Shmigliuk, V.T. Zyukov, and M.F. Miglei, in Bose
Condensation of Polaritons in Semiconductors, pp. 58-79 (§tiinja,
Kishinev, 1985).
[6] N.N. Bogoliubov, J. Phys. USSR 11, 23 (1947); see also N.N. Bogoli-
ubov, Selected Works. Vol.2, pp. 242-257 (Naukova Dumka, Kiev,
1971).
[7] L.D. Landau and E.M. Lifshitz, Quantum Mechanics (Pergamon,
Oxford, 1963).
[8] J.R. Schrieffer, Theory of Superconductivity (Nauka, Moscow, 1970).
[9] N.N. Bogoliubov, Quasi-Averages in the Problems of Statistical
Mechanics. Dubna: Preprint of JINR R-1415, 1963; see also N.N.
Bogoliubov, Lectures on Quantum Statistics (Gordon and Breach,
NY, 1970), Vol. 2.
[10] N.N. Bogoliubov (Jr) and B.I. Sadovnikov, Some Questions of Sta-
tistical Mechanics (Vysshaya Shkola, Moscow, 1975).
[II] V.N. Popov and VS. Yarunin, Collective Effects in Quantum Statistics
of Radiation and Matter (Izd. LGU, Leningrad, 1985).
[12] S.A. Moskalenko, A.H. Rotaru and Yu.M. Shvera, in Laser Optics of
Condensed Matter, J.L. Birman, H.Z. Cummins and A.A. Kaplyan-
skii, eds. (Plenum, NY, 1988), pp. 331-336.
[13] VR. Misko, S.A. Moskalenko, A.H. Rotaru and Yu.M. Shvera,
Phys. Stat. Sol. (b) 159, 477 (1990).
[14] I.V Beloussov and Yu.M. Shvera, Phys. Stat. Sol. (b) 139, 91 (1990).
[15] I.V Beloussov and Yu.M. Shvera, Z. Phys. B 90, 51 (1993).
[16] L.V. Keldysh, Zh. Eksp. Teor. Fiz. 47, 1515 (1964).
[17] I.V. Beloussov and Yu.M. Shvera, Teor. Mat. Fiz. 85, 237 (1990).
[18] D.N. Zubarev, Non-Equilibrium Statistical Thermodynamics (Plenum,
NY, 1974).
518 I. V. Beloussov and Yu. M. Shvera
[19] I.V. Beloussov and Yu.M. Shvera, in Excitons and Biexcitons in
Confined Systems (§tiinja, Kishinev, 1990), pp.133-176.
[20] I.V. Beloussov and Yu.M. Shvera, in Interaction of Excitons with
Laser Radiation (§tiinja, Kishinev, 1991), pp.57-74.
[21] I.V. Beloussov and Yu.M. Shvera, in Nonlinear Optical Properties
of Excitons in Semiconductors of Different Dimensionalities (§tiinja,
Kishinev, 1991), pp.53-64.
[22] L.V. Keldysh, in Problems of Theoretical Physics (Nauka, Moscow,
1972), pp.433-444.
[23] A.I. Akhiezer and S.V. Peletminskii, Methods of Statistical Physics
(Nauka, Moscow, 1977).
[24] V.L. Safonov, Physica A 188, 675 (1992).
 27
Excitonic Superfluidity in CU2O
E. Fortin and E. Benson
Department of Physics
University of Ottawa
Ottawa, ON KIN 6N5
Canada
A. Mysyrowicz
LOA, Ecole Poly technique
Palaiseau
France

Abstract

Using the exciton-mediated photovoltaic effect, we examine exciton trans-

port over large distances in CuiO as a function of temperature and particle
density. Evidence for a phase transition at low temperatures and high den-
sities is attributed to the onset of excitonic superfluidity.

We have performed exciton transport measurements over a range of

temperatures and exciton densities in ultrapure, oriented large CU2O
single crystals. A sketch of the experimental method is shown in Fig. 1.
The crystal is illuminated on the back surface by 10 ns pulses from a
frequency-doubled YAG laser (X = 532 nm). The initial exciton density
created over an absorption depth (about one micron at X = 532 nm) can
be varied by inserting calibrated neutral density filters in the laser beam,
reaching values of up to 1019 cm" 3 . The excitons which have migrated
to the opposite face of the crystal are dissociated into free carriers by
the high electric field near the Cu Schottky contact [1] deposited in
a comb configuration together with an ohmic Au electrode, resulting
in an external current. A time-resolved measurement of that current
will give the velocity distribution of the excitons migrating through the
crystal. This method of detection - as opposed to photoluminescence -
is particularly well suited to the study of optically inactive paraexcitons
in CU2O; moreover, since the migration time is of the order of one
microsecond as compared to the lifetime of 13 ^s [2] for paraexcitons,
recombination processes have little influence on the measurements.
Figure 2 shows the time-resolved photocurrent at a temperature of
1.85K in a 3.56-mm-thick crystal for several illumination intensities, that
is, initial exciton densities. The initial signal at t = 0 results from the

519
520 E. Fortin, E. Benson and A. Mysyrowicz

Au electrode
Photocurrent
^ ^ 0 . 2 5 4 mm

Cu electrode

1 mm

f1 1 • 1\1
E|
|

Exciton
formation E>
Cuprite crystal
>

Laser
beam

Fig. 1. Schematic of the experiment. The sample is oriented with the < 100 >
direction along the edges.

direct illumination of the electrodes, and serves as a convenient measure

of the exciton detector response time as well as an independent optical
trigger for the 2 GHz digital oscilloscope. The delayed signal, about one
JUS later, results from the dissociation of the excitons having travelled
across the crystal. As the illumination intensity (exciton density) is
increased (bottom curve to top curve), a remarkable change occurs in
the exciton transport: from a diffusive behaviour characteristic of a free
gas expanding through a viscous medium, the propagation evolves, at
high exciton densities, to a ballistic behaviour where the majority of
particles propagate in a narrow range of speeds. The speed is found
to approach asymptotically the longitudinal sound velocity in CU2O [3]
fa = 4.5 x 105 cm/s) [4]. Measurements on a second crystal confirm
the dragfree motion over distances of nearly 1 cm. If Bose condensation
is responsible for the evolution shown, the crystal temperature should
strongly influence the exciton transport as well. Fig. 3 shows data for
the thinner crystal at constant illumination intensity of 1.2 x 106 W/cm2
at several temperatures. Here again the transport evolves from diffusive
to ballistic as the temperature is lowered. Note the narrow range of
temperatures over which this change occurs.
 Excitonic Superfluidity in Cu^O 521

Thi( dcness = 3 56 mm

I
1 V

1-
1
1
Jft m f /

0.00

Time (psec)

Fig. 2. Time-resolved photocurrent of sample 1 measured across a 50 Q resistance

at different illumination intensities and a constant temperature of 1.85 K. The
intensity increases from bottom to top curve from 3.75 x 104 to 1.5 x 106 W/cm2.
The curve shown by the arrow is for an intensity of 6 x 105 W/cm2.

J
rj^r\Kj\/\r^KJ^\r^\^^ ^^

0.00 0.50 1.00 1.50 2.00 2.50 3.00

Time (psec)

Fig. 3. Time-resolved photocurrent for sample 1 at different temperatures and

constant illumination intensity of 1.2 x 106 W/cm2.

We believe that these results are strong evidence for the onset of a
superfluid excitonic phase above a critical density and below a critical
temperature. Let us first compare the density of excitons inside the
initial packet to the critical density for Bose-Einstein condensation as
predicted by the ideal Bose model. Since the measurements are taken
522 E. Fortin, E. Benson and A. Mysyrowicz

Fig. 4. Integrated photocurrent of sample 1 for afixedtime window as a function

of temperature for two illumination intensities: squares, 1.2 x 106 W/cm2; circles,
6 x 105 W/cm2. Arrows show transition temperatures.

at long delays after the excitation, the temperature of the excitonic

fluid is imposed by the crystal temperature, and the initial mixture of
ortho- and para-excitons has now converted into an energetically lower
population of long-lived paraexcitons. The density can be estimated
from the knowledge of the initial number of excitons created by the laser
pulse, and the volume of the packet at t = 1 fis as given by its thickness
d ~ Aftv, where At is the temporal width and the lateral dimensions
are assumed to be equal to the initial ones (diameter = 2 mm). For
an excitation of 600 kW/cm 2 (corresponding to the onset of a ballistic
excitonic packet as shown in Fig. 2, fifth graph from the top), a value
of n(t = lfis) ~ 1.4 x 1017 cm" 3 is found, where a factor of 2 has been
included to take into account reflection losses at cryostat windows and
sample surface. This value is in good agreement with the calculated
critical density of 6 x 1016 cm" 3 for paraexcitons of mass m = 2.7mo at
T = 1.85K [5].
The critical density for condensation is predicted to vary with temper-
ature as nc = CT3/2. To test our results against this relationship, we plot
in Fig. 4 the delayed signals of Fig. 3 integrated over a, fixed time window
 Excitonic Superfluidity in CU2O 523
(defined as the full width at half maximum at the lowest temperature
of 1.9 K), as a function of temperature. A sharp transition is observed
at T = 3.1 K, which we can interpret as the critical temperature for
condensation at this particular exciton density. When the experiment is
repeated at an excitation intensity that is lower by a factor of 2, a similar
graph displays a transition temperature of 2K. Over a density change of
a factor of 2 the theory predicts a critical temperature variation of:

Tl = 2 2 ' 3 = 1.58
T
This is in good agreement with the experimentally observed value of 1.55.
In conclusion, detection of exciton transport with good time and
space resolution reveals the onset of an anomalous ballistic transport
below a critical temperature and above a critical particle density. The
conditions required for the appearance of ballistic transport are in very
good agreement with those predicted for Bose-Einstein condensation in
an ideal Bose gas. We interpret these results as evidence for a superfluid
excitonic phase associated with Bose-Einstein condensation. For more
details, see Ref. [6].

References
[1] E. Tselepis, E. Fortin and A. Mysyrowicz, Phys. Rev. Lett. 59, 2107
(1987).
[2] A. Mysyrowicz, D. Hulin and A. Antonetti, Phys. Rev. Lett. 43,
1123 (1979); D. W. Snoke, A. J. Shields and M. Cardona, Phys. Rev.
B45, 11693 (1992).
[3] A. Mysyrowicz and E. Fortin, to be published.
[4] J. Berger, J. Castaing and M. Fisher, J. de Physique 40, 13 (1979).
[5] N. Caswell, J. S. Weiner and P. Y. Yu, Solid State Comm. 40, 843
(1981).
[6] E. Fortin, S. Fafard and A. Mysyrowicz, Phys. Rev. Lett. 70, 3951
(1993).
 28
On the Bose-Einstein Condensation of
Excitons: Finite-lifetime Composite
Bosons
Sergei G. Tikhodeev
Institute of General Physics
Moscow 117333
Russia

Abstract
A brief review of experimental and theoretical work on Bose-Einstein con-
densation (BEC) in nonequilibrium systems is presented, with emphasis on
excitons in a semiconductor. Conditions for obtaining BEC in such sys-
tems are discussed, with special attention being paid to the effects of finite
lifetime, the role of spatial inhomogeneity and phonon-driven transport of
excitons.

1 Introduction
After many years of study, Bose-Einstein condensation (BEC) in a dilute
Bose gas has not yet been unambiguously established experimentally in
any real physical system. A systematic search has been undertaken in
atomic systems (atomic spin-polarized in magnetic field hydrogen [1]
and more recently in cesium [2]) and in excitonic systems in solids (see
e.g. [3, 4]).
Observation of BEC in an excitonic system is made easier [5, 6] because
of the following reasons: (i) small particle masses (of the order of the
free electron mass mo); (ii) the possibility of reaching sufficiently high gas
densities with an increase of excitation intensity; and (iii) the presence
of luminescence, making it possible to determine the particle distribution
function. However, excitons are finite-lifetime particles, and a number
of problems arise due to this fact. The goal of the present paper is
to analyze the effect of the finite lifetime on the possibility of BEC of
excitons.
In the case of incoherent excitation [7], a Bose-Einstein (non-
Maxwellian) distribution has been experimentally demonstrated in a sys-

524
 On the BEC of Excitons: Finite-Lifetime Composite Bosons 525

Table 1. Densities, temperatures, particle masses, degeneracy factors and

Silvera-Reynolds numbers of dilute Bose gases

n T
(cm-3) (K)

Atomic H a 2- 10 19 0.65 1836 1 0.23

Atomic H b 7- 10 13 10" 4 1 0.33
Excitons in stressed G e c 2.8 10 15 3.1 0.5 1 0.44
Orthoexcitons in Cu2O d ' e 8- 10 18 20 2.7 3 0.85
Paraexcitons in Cu2O e 5- 10 18 * 20 1 1.4
a
E . Tjukanov et a/, Physica B 178, 129 (1992).
b
J. M . Doyle et a/., Phys. Rev. Lett. 67, 603 (1991).
c
Ref. [8].
d
Ref. [9].
e
Ref. [11].
* Corrected number, see Ref. [12].

tem of magnetically spin-polarized excitons in uniaxially stressed Ge [8]

and of ortho- and paraexcitons in CU2O [9]-[12]. The best results to
date are listed in Table 1, where, in addition to the data on dilute Bose-
gas densities and temperatures achieved, the values of Silvera-Reynolds
number [1] S are shown,

where

is the BEC temperature of an ideal Bose gas with density n, and nc(T) is
the critical density at temperature T. For comparison, the corresponding
numbers are shown for atomic hydrogen and excitonic systems. The
logarithm of 8 is the dimensionless distance from a point (n, T) on a
phase diagram to the BEC phase boundary given by (2); if S > 1, the
conditions of BEC are met. (The mysterious number 13 = 22 + 32 is the
sum of the squared numerator and denominator of the exponent 2/3 in
the BEC criterion (2) .) One can see from Table 1 that, up to now, the
search for BEC in excitonic systems has been in the lead, and, in the case
of paraexcitons in CU2O, the conditions for BEC in an ideal Bose gas
have been met [13].
526 S. G. Tikhodeev
2 Experiments under Stationary Excitation
Stationary excitation was used in earlier experiments in CU2O [9] and in
Ge [8]. The flow of thermalizing excitons should compensate the decay
of condensate particles. The latter introduces some deviations of BEC
conditions from the ideal Bose gas given by (2) [14, 15]. The physical
reason for these modifications of the phase diagram is that during its
lifetime, a newly generated exciton should have enough time to diminish
its energy via exciton-exciton and exciton-phonon collisions and to reach
the k = 0 state.
To illustrate the modifications of BEC conditions, a standard kinetic
equation for excitons interacting with a phonon thermostat at T — 0 was
solved in Refs. [14, 15]. In the statistical limit V —• 00, N/V —• n = const,
this equation takes the form

(3)

/
p<q

0<q<p
(4)

where T is the lifetime and g p is the generation rate of excitons, as-

sumed to be a smooth function of momentum (i.e. without resonant
excitation of the condensate at p = 0). Such equations for biological
systems were analyzed previously in Ref. [16, 17]. These equations can
be solved analytically for some specific forms of W(q,p). It can be
shown [14, 15] that even at T = 0, the condensate does not appear until
the gas density exceeds some threshold value wth ~ ^V^phonA- Here
VQ1 ~ (mS/h)3 is a characteristic volume of momentum space, where the
thermalization of excitons scattering with phonons slows down due to
the energy-momentum conservation law, S is the velocity of sound in the
semiconductor, and TPhOn is a large subsonic exciton-phonon scattering
time.
Close to the BEC, when the occupation numbers become large, one
should obviously take into account exciton-exciton collisions, which
cause an increase of the excitonic flow into the k = 0 state. However,
it can be easily shown that even taking these collisions into account, we
 On the BEC of Excitons: Finite-Lifetime Composite Bosons 527
obtain a threshold density for condensate nucleation
1 m
4TT3/4 h2a\T
where a is the exciton scattering length and e is the mean exciton energy.
Assuming a ~ aQX and e ~ kBT, one can obtain nth ~ 10 13 cm~ 3 for
Ge and 10 15 cm~ 3 for CU2O. One can see that, fortunately, nth < w
(see Table 1) in both cases, so this mechanism cannot impede BEC in
experiments [8, 9]. However, it can become significant in other excitonic
systems with shorter exciton lifetimes.

3 Experiments under Pulsed Excitation. Nucleation Time

In order to achieve BEC in the case of pulsed excitation, the condensate
nucleation time should be less than the lifetime of the particles. Recently
the problem of the condensate nucleation time was addressed in a number
of papers [18]-[20]. The situation had been dramatized by the fact
pointed out for the first time in Ref. [21]: within the kinetic approach,
the kinetic equation for condensate density takes the form ^ oc no. (This
can be seen, e.g., from (3).) Thus, if the condensate density at the initial
moment is zero, the condensate nucleation time is infinite.
It was shown in Refs. [19, 20] that actually the condensate nucleation
time is finite, of the order of the usual kinetic collision time ikin ~ (nva)~ l.
The physical reason for this is the following [19]. In the region of small
energies close to the condensate, the kinetic approach is not valid, and
the system has to be described by means of a complex order parameter
*F, obeying the Ginzburg-Landau equation:
fixy 1
AT + m'*. (6)
The characteristic evolution time of *F is the correlation time Tcor
On the time scale of Tcor, a quasicoherent state arises with suppressed
density fluctuations. However, this stage of evolution is preceded by
a kinetic one, which takes a time Tkin > TCOr- Thus, ikin becomes a
characteristic quasicondensate nucleation time.
Although the analysis of Refs. [19, 20] was devoted to atomic systems,
it is also valid for excitonic systems. A consistent approach to deriving (6)
for excitons was developed in Ref. [22]. One can estimate x ~ 4nh2aex/m.
Then for excitons in CU2O, we obtain Tcor ~ 10~12 s, Tkin ~ 10~ n s (at
n ~ 10 19 cm~ 3 and T ~ 20 K). One can see that ikm <C T, and the
excitonic system in CU2O has enough time to develop a quasicondensate.
528 S. G. Tikhodeev

5 10 15 20 25 30 35
t(ns)
Fig. 1. Supersonic transport of orthoexcitons in CU2O at high power. The
distance A is measured from the surface to the innermost half maximum of
the spatial distribution, for the orthoexcitons as a function of time for several
laser-power densities. The dashed line has a slope given by the velocity of sound
S (from [11]).

4 Impact of the Spatial Inhomogeneity and Phonon Wind

In order to make the initial exciton density as large as possible, surface
excitation was used in [10]—[12] with a photon energy well above the band
gap of CU2O, hat — Eg > ksT. As a result, the excitonic system in these
experiments was very inhomogeneous, and the processes of exciton ther-
malization and condensate nucleation were proceeding at the same time
as the spreading of the exciton cloud into the sample bulk. Anomalously
fast (supersonic) nondiffusive transport of excitons was observed [11] at
higher excitation intensities (see Fig. 1) and was attributed to the onset of
exciton superfluidity. However, this fast transport can also be explained
by the joint action of diffusion and the so-called phonon wind, that is,
by nonequilibrium phonon-driven transport of excitons [23].
Phonon wind, produced by ballistically propagating nonequilibrium
phonons born during the thermalization of carriers, plays a very signif-
icant role in solids at low temperatures (see [24]-[26] for a review). For
example, it causes fast carrier transport in experiments with electron-hole
liquid in Ge [27]. Phonon wind effects are unavoidable if hco > Eg. They
should be important in experiments with surface excitation in CU2O, and
a simple model of phonon-driven transport of excitons was developed in
Ref. [23].
Figure 2 illustrates the dependence (calculated within this model) of
 On the BEC of Excitons: Finite-Lifetime Composite Bosons 529

125-

100 :

-15 5 15 25
t(10" 9 s)
Fig. 2. The time dependence of the FWHM of the carrier concentration for
different values of dimensionless excitation energy a = E/Es (Eq. (7)) in the case
of surface excitation (from [23]).

the FWHM of the exciton density on time and excitation intensity, which
agrees well with the experimental data. One can see in Fig. 2 that if

mhcoS2Ti
E>ES = (7)

a region of supersonic drift appears. (Here x\ is the excitation pulse

duration, xr is the exciton relaxation time, <rph is the exciton-phonon
cross section and ePW is the energy (per excitation quantum) which goes
into the phonon wind.) Comparing Figs. 1 and 2, one can estimate that
Es ~ 1.5 • 104 erg/cm 2 . Then <7ph£PW ~ 2 • 10~ 15 meVcm 2 , which is in
reasonable agreement with a simple estimate of this parameter [23, 28]
and also with data on phonon wind in other semiconductors [25]. Thus,
the phonon wind model may explain fast supersonic transport of excitons
without assuming their superfluidity. However, as was shown recently
in [29], this fast transport at higher excitation intensities may also be
explained by drag-free hydrodynamic flow, which may indicate that the
excitonic superfluidity has indeed been observed in experiments [11]. In
530 S. G. Tikhodeev
view of this controversy, the occurrence of superfluidity (over and above
BEC) in CU2O needs additional study.

References
[I] LF. Silvera and M. Reynolds, J. Low Temp. Phys. 87, 343 (1992).
[2] E. Tiesinga, A.J. Moerdijk, BJ. Verhaar and H.T.C. Stoof, Phys.
Rev. A 46, 1167(1992).
[3] E. Hanamura and H. Haug, Phys. Rep. 33C, 210 (1977).
[4] M. Ueta, H. Kanzaki, K. Kobayashi, Y. Toyozawa, and E. Hana-
mura, Excitonic Processes in Solids (Springer, Berlin, 1986).
[5] S.A.Moskalenko, Fiz. Tv. Tela 4, 276 (1962).
[6] J.M. Blatt, K.W. Boer, and W. Brandt, Phys. Rev. 126, 1691 (1962).
[7] We do not consider here the experiments with direct excitation of
coherent excitonic and biexcitonic states by an external coherent
source (EM wave) (see e.g. Refs. [3], [4] and M. Hasuo, N. Naga-
sawa, T. Itoh, and A. Mysyrowicz, Phys. Rev. Lett. 70, 1303 (1993)).
[8] I.V. Kukushkin, V.D. Kulakovskii, and V.B. Timofeev, JETP Lett.
34, 36 (1981).
[9] D. Hulin, A. Mysyrowicz, and G. Benoit a la Guillaume, Phys. Rev.
Lett. 45, 1970 (1980).
[10] D.W. Snoke, J.P. Wolfe, and A. Mysyrowicz, Phys. Rev. Lett. 59,
827 (1987).
[II] D.W. Snoke, J.R Wolfe, and A. Mysyrowicz, Phys. Rev. B 41, 11171
(1990).
[12] D.W. Snoke, Jia Ling Lin, and J.R Wolfe, Phys. Rev. 43, 1226
(1991).
[13] Further results for CU2O are reported in this volume. J.L. Lin
and J.R Wolfe obtain a Silvera-Reynolds number S = 2.8 (for
paraexcitons in stressed CU2O), and E. Fortin, E. Benson, and
A. Mysyrowicz report d as large as 5.
[14] S.G. Tikhodeev, Solid State Commun. 72, 1075 (1989).
[15] S.G. Tikhodeev, Sov. Phys. - JETP 70, 380 (1990).
[16] H. Frohlich, Int. J. Quantum Chem. 2, 641 (1968).
[17] N.G. Duffield, J. Phys. A: Math. Gen. 21, 625 (1988).
[18] D. Snoke and J.R Wolfe, Phys. Rev. B 39, 4030 (1989).
[19] Yu. Kagan, B.V. Svistunov, and G.V. Shlyapnikov, Zh. Eksp. Teor.
Fiz. 101, 528 (1992).
[20] H.T.C. Stoof, Phys. Rev. A 45, 8398 (1992).
[21] E. Levich and V. Yakhot, Phys. Rev. B 15, 243 (1977).
 On the BEC of Excitons: Finite-Lifetime Composite Bosons 531
[22] L.V. Keldysh, in Problems of Theoretical Physics, A Memorial Volume
to Igor E. Tamm (Nauka, Moscow, 1972), p. 433.
[23] A.E. Bulatov and S.G. Tikhodeev, Phys. Rev. B 46, 15053 (1992).
[24] J.R Wolfe, J. Lumin. 30, 82 (1985).
[25] S.G. Tikhodeev, Sov. Phys. Usp. 28, 1 (1985).
[26] L.V. Keldysh and N.N. Sibeldin, in Nonequilibrium Phonons in Non-
metallic Crystals, Eisenmenger and Kaplyanskii, eds. (Elsevier, Am-
sterdam, 1986), p.455.
[27] N.N. Sibeldin, V.B. Stopachinskii, S.G. Tikhodeev, and V.A. Tsetkov,
JETP Lett. 38, 207 (1983).
[28] V.V. Konopatskii, G.A. Kopelevich, and S.G. Tikhodeev, Zh. Eksp.
Teor. Fiz., in press (1994).
[29] B. Link and G. Baym, Phys. Rev. Lett. 69, 2959 (1992).
 29
Charged Bosons in Quantum
Heterostructures
L. D. Shvartsman and J. E. Golub
Racah Institute of Physics
The Hebrew University of Jerusalem
Jerusalem 91904
Israel

Abstract

We show that heterostructures may give rise to charged bosons. Bosons

may be formed as a result of two quantum-well holes pairing in a repul-
sive potential In the case where at least one hole has a negative effective
mass, the repulsive electrostatic interaction may be converted to an effec-
tive attraction. The effect is a general one, possible in quantum layers of
most cubic semiconductors. We compute the x-dependent hole-hole bind-
ing energy for AlGaAs/InxGa\-xAs/GaAs quantum layers, taking account
of the screening by a degenerate background gas of positive mass holes.
We conclude that bound hole-hole pairs should be observable in infrared
absorption experiments.

The possibility of creating a controlled gas of bosons has stimulated years

of work on excitonic systems [1-4], spin-polarized hydrogen [5], and atom
trapping [6]. But the only opportunity of charged boson formation in
solids which is widely accepted is Cooper pairing. The possibility of any
other mechanism of charged boson formation in solids is a very exciting
topic for investigation. Needless to say, the superfluidity of a charged
boson gas in solids would be manifested as superconductivity.
In this work we discuss regular electrostatic repulsion, which, in the
case where at least one of the carriers has a negative effective mass,
may be converted to an effective attraction, as a mechanism of pairing.
A composite boson, carrying charge 2|e|, may be formed in such a
case.
It is known that the presence of negative mass states of carriers is a
general property of crystal band structure. The pairing mechanism of
interest is realized in the following situations. First of all, it may be an

532
 Charged Bosons in Quantum Heterostructures 533
electrostatic interaction of two carriers when one of them is near the
bottom of one band and another one near the top of the next band. In
such a case, the pairs are excitonic charged complexes. The other variant
corresponds to the case of two carriers near the top of the band, so that
both of them have a negative mass. In both situations, it is necessary
to stress that we are considering an excited state of the system. As will
be clear in the first case, a boson may be formed only when the reduced
mass of the pair is negative and so its total mass is positive. For two
negative-mass carriers, both the total mass and the reduced mass of the
pair will be negative.
In order to make clear the physical nature of the pairing mechanism,
let us write down a regular hydrogen-like Schrodinger equation :
[-(ft2/2m*)V2 + e<j>(p)]ip(p) = Exp(p) (1)
It does not matter if it is a three- or a two-dimensional equation. In
the usual case, when m* > 0 is the reduced mass of the pair, and <t>(p)
is an attractive electrostatic potential, (1) describes bound states with a
discrete spectrum. In the three-dimensional situation the potential well,
of course, is supposed to be deep enough, and in the two-dimensional
situation, as is well known, at least one confined level always exists.
It is obvious that if we put in Eq. (1) a negative reduced mass but
a repulsive potential $(p), we still have the same equation. This would
describe a bound state of the complex not in the well but in the barrier.
The kinetic energy of the internal motion has a negative value in this
case.
The idea of a charged, exciton-like complex appeared as early as
1971 in connection with the observation of a reversed hydrogen-like
absorption spectrum in bulk crystals of BH3 [7, 8]. The reversal of the
usual hydrogenic spectrum for excitons was explained as a result of the
existence of a bound complex of two electrons of different bands, with
negative exciton reduced mass. A similar effect has been reported in bulk
ZnP2 [9]. The complex of two electrons was called a bielectron, and the
complex of two holes was called a bihole.
In the present paper, we show that quantum heterostructures of cubic
semiconductors may exhibit biholes, and that this characteristic is a
general one, possible in nearly all quantum wells of cubic semiconductors.
Specifically, we show that two holes may form a bound pair, and compute
the binding energy including the effects of the two-dimensional screening.
Being a quasistationary state, the pairs have a finite lifetime. We conclude
that experimentally observable binding energies are realizable, and we
534 L. D. Shvartsman and J. E. Golub

8 -5

Fig. 1. Valence subband structure for a GaAs quantum layer in the approximation
of an infinite potential barrier, K = kL/n is the dimensionless wavevector, where
k is the usual wavevector and L is the well width. The energy e is measured in
units of h2n2/2moL2.

argue that the pairs are long-lived objects. A detailed calculation of the
lifetime is postponed for a later publication.
The possibility of charged bosons in heterostructures follows from
the form of the valence subband structure of quantum wells of most
cubic semiconductors. A generic band structure is shown in Fig. 1.
A general feature of the dispersion is the presence of negative mass
states. These states exist in the first excited subband of two-dimensional
films of almost all cubic semiconductors [10-12]. They appear as a
sequence of multicomponent anisotropic structure of the Kohn-Luttinger
hamiltonian. For example [10-12], in deep potential wells of some
materials (e.g., GaAs, Ge, GaSb, InSb, InP, InAs), this subband is the
n = 2 heavy hole band [10]. In others (e.g., AlSb, ZnTe, GaP), it is the
n = 1 light hole band [10]. In shallow wells, the band ordering may be
reversed. However, in all these systems, the sign of the effective mass is
negative [13].
Consider the electrostatic interaction between two holes when one
belongs to the ground subband (mass m\ > 0), and the other to the
excited subband (mass mi < 0). If the reduced mass (m*)"1 = mj"1 + m^1
is negative, the electrostatic repulsion, as we saw above, is converted to
a net attraction. It leads to the formation of a bound pair, that is, to
the formation of a bihole. A necessary condition for bihole formation
 Charged Bosons in Quantum Heterostructures 535

1 2 3 4
x (percent)

Fig. 2. Indium-mole-fraction-dependent masses of the ground and excited sub-

bands for L = 10 nm. The absolute value of the h2 band mass is shown. The
arrow indicates the point at which the bihole reduced mass diverges.

is m* < 0 or
m2 (2)

Eq. (2) is satisfied in infinite quantum wells of all the materials mentioned
above except InSb and InAs. Eq. (2) may also be satisfied in finite wells
of properly chosen width and material composition. So, the possibility
of biholes is a general feature of quantum wells.
Bihole formation in quantum wells was first considered in 1983 [14].
We discuss below several important differences between experiments in
bulk crystals and the present work.
We next calculate the bihole binding energy using a quantum well
of AlGaAs/ InxGai_xAs/AlGaAs as a model system starting from a
Kohn-Luttinger multicomponent formulation. In any heterostructure,
changes in the doping x affect both the well-depth and the uniaxial
strain. The present system has the important calculational advantage
that the uniaxial strain dominates [15]. The calculation thus pro-
ceeds from a subband structure obtained assuming an infinite well, but
takes into account the x-dependent uniaxial strain in the well layer.
In particular, we assume the dependence of the effective masses upon
the indium mole fraction x shown in Fig. 2 [15, 16]. This assump-
tion permits us to calculate an x-dependent binding energy and then
adjust x to maximize the effect within the applicability range of the
model.
536 L. D. Shvartsman and J. E. Golub
Now let us consider (1) in the two-dimensional case. The electrostatic
potential c/>(p) in the presence of screening by the positive mass holes of

&keiKP(t>(K) (3)
the ground subband is given by

where [17-19]
c/>(K) = 2ne/e1(K + 2/a). (4)

Here, e = €2/e\; and ei(e2) is the dielectric constant of the well (barrier)
layers. In the numerical solutions below, we assume e = 1. Other
cases, for example, that of a free standing film, will be treated in a later
publication, a = eih2/m\e2 is a Bohr radius defined for holes in the first
subband. For p > L, we find for <f)(p) the asymptotic behavior

<t>{p) = (e/ex) {1/p - (n/a)[H0(2p/a) - N0(p/a)]), (5)

where HO(NQ) denotes the Struve (Neumann) function of order zero.

Eq. (5) expresses the rather non-coulombic force law obtained in the
narrow-well limit.
The bihole binding energy now is determined from (1), where m* < 0 is
the reduced mass, and <j>(p) is determined by (3) and (4). To solve (1), we
use a direct variational method in fc-space. We choose the two-parameter
variational function

t/>(p) = [(21)2'r/2nr(2/v)] " 2 exp[- Xpv], (6)

where X and v are variational parameters. This function satisfies the

trivial boundary conditions ip(0) = const, and \p(p —* 00) = 0. The
prefactors are chosen for normalization. The bihole binding energy is
found by maximizing (because of the reversed signs of the mass and
interaction) the total energy < T + U >xp. The kinetic energy is easily
found to be:

< T >y)= (nh2v/4m*)(2X)2/v/2nT(2/v). (7)

Calculation of the potential term is less straightforward, and will be

detailed elsewhere. The result is

e P
/ '
Combining (7) and (8), we find for the x-dependent bihole binding energy
 Charged Bosons in Quantum Heterostructures 537

0.6 1.0 1.4

x (percent)

Fig. 3. Indium-mole-fraction-dependent bihole binding energy computed using

(a) a single-parameter trial function, and (b) the trial function in (6).

EB(x) = max(< T(x) + U(x) >) (9)

= max C(A, v) \2n2h2v/4rn*(x)

pdp kdk 2ne2/e(k + 2/a(x))exp[-2lpv]J0(kp)

• /

where C(A,v) = {2X)2/v/2nT(2/v). We have calculated this dependence

numerically with the result shown in Fig. 3, which also shows the result
of a single-parameter variational calculation using a hydrogenic trial
function. (The one-parameter function is obtained from (6) by taking
v = 1.) The diagram shows that the addition of as little as 1.3% indium
to the GaAs quantum well-layer has the effect of increasing the bihole
binding energy from a small fraction of one meV to nearly 5 meV. The
reason for this increase in binding energy may be understood on simple
physical grounds. According to Fig. 2, strain induced by the lattice
mismatch has the effect of tuning the two hole masses to be more nearly
equal and opposite. As a result, the reduced mass goes to infinity, and
the binding energy diverges.
There are other means to make the hole masses nearly equal and
opposite. We illustrate this point using a well of AlyGai-^As/GaAs. For
y = 1, the hole masses are dissimilar [16] with m\ > \ni2\. By contrast, for
y = 0.3, we have the opposite situation [17]: m\ < \rn2l Since the masses
538 L. D. Shvartsman and J. E. Golub
vary monotonically with y, there is a value of y for which m\ « — mi and
the bihole binding energy again diverges. A similar situation occurs in
the Si/Ge system [11]. In the pure Ge infinite quantum well, \m2\ is too
small to permit an observable bihole binding energy. With the addition
of Si, mi grows and the situation m\ « — mi becomes possible.
The existence of biholes must be manifested as the presence of addi-
tional peaks between the hi-h 2 and hi-h 3 absorption lines. The hi-h2 and
hi-h3 absorption lines themselves have nonzero, temperature-independent
widths determined by the complex (nonparallel) subband structure [11]
and the dipole selection rules which, for example, forbid hi-h2 transi-
tions at k = 0 for TE polarization. While biholes are broadened by
disorder-related inhomogeneities, these may be suppressed through the
use of interrupted growth, atomic layer epitaxy, or through other techni-
cal means. The only fundamental limitation to the observation of biholes
comes from temperature broadening. Since the binding energies calcu-
lated above can be as large as 5 meV (« 60 K), the bihole line should
be observable in low temperature, infrared absorption experiments. The
range of validity of the present model can be understood as follows. The
model is inadequate in the limit of large binding energy because of our
neglect of the third (and higher lying) subbands. Thus, the model is
most accurate when the bihole level is much closer to the second sub-
band than to the third. In the opposite limit, that of very small binding
energies, the model fails because of inaccuracies associated with the use
of an approximate trial function, a situation common to all variational
calculations.
Because dispersion curves like those displayed in Fig. 1 are non-
parabolic, and bihole states lie within a continuous background of free
hole-hole states, the bihole state cannot be a fully stationary state in
the strict sense. The binding energy calculated above is thus the real
part of a complex energy E = EB + ih/T, where x gives the lifetime with
respect to various decay processes. Let us divide the decay channels into
two types: those which proceed directly, and those which require the
participation of a third body. The second class of decays proceeds with
the characteristic time of interaction with the third body, e.g., a phonon
scattering time. These processes proceed slowly with the result that h/x
is small compared to the binding energies (meV's) calculated above. The
first class of (direct) decay channels is forbidden for the structure calcu-
lated here as for many structures [20, 21]. To establish this fact, we note
that the complex subband dispersion prevents the spontaneous decay of
a bihole to an isoenergetic hole-hole pair state. As we have checked
 Charged Bosons in Quantum Heterostructures 539
numerically, such a decay process can not conserve both momentum and
energy. As long as a third body (e.g., a phonon) is required, the bihole
must be metastable.
To conclude, we have shown that bound hole-hole pairs are a generic
feature of quantum wells made of most cubic semiconductors. We
have calculated the binding energy of the bihole in the presence of a
degenerate gas of positive mass holes and shown that, by careful choice
of the quantum-well parameters, it may be made quite large. We conclude
that it should be possible to realize charged bosons experimentally using
heterostructures.
We gratefully acknowledge the support of the Ministry of Science and
Technology (3845/1/99).

References
[I] P.L. Gourley and J.P. Wolfe, Phys. Rev. Lett. 40, 526 (1978);
J. P. Wolfe and C. D. Jeffries, in Electron-Hole Droplets in Semi-
conductors, CD. Jeffries and L.V. Keldysh, eds. (North Holland,
Amsterdam, 1983) p. 431.
[2] D. Snoke, J.L. Lin, and J.P. Wolfe, Phys. Rev. B 43, 1226 (1991).
[3] D. Snoke and J.P. Wolfe, Phys. Rev. B 42, 7876 (1990).
[4] D. Snoke, J.P. Wolfe, and A. Mysyrowicz, Phys. Rev. Lett. 64, 2563
(1990).
[5] I.F. Silvera, Physica 109+110B, 1499 (1982); T. J. Greytak and D.
Kleppner, in New Trends in Atomic Physics, G. Grynberg and R.
Stora, eds. (North Holland, Amsterdam, 1984), Vol. II.
[6] E.L. Raab, M. Prentiss, Alex Cable, Steven Chu, and D. E. Pritchard,
Phys. Rev. Lett. 59, 2631 (1987), and references therein.
[7] E.F. Gross, V.I. Perel, and R.I. Shechmamet'ev, JETP Lett. 13, 229
(1971).
[8] E.F. Gross, N.V. Starostin, M.P. Shepilov and R. I. Shechmamet'ev,
Proceedings of the USSR Academy of Sciences 37, 885 (1973).
[9] A.V. Selkin, I.G. Stamov, N.U. Syrbu, and A.G. Umenets, JETP
Lett. 35, 57 (1982).
[10] L.D. Shvartsman, Ph.D. thesis, USSR Academy of Science Siberian
Branch, Institute of Semiconductors, Novosibirsk (1984).
[II] A.V. Chaplik and L.D. Shvartsman, Poverkhnost (Surface), No. 2,
73 (1982).
[12] L.D. Shvartsman, Sol. State Commun. 46, 787 (1983).
[13] Silicon is the only exception to this rule. See Refs. [10] and [12].
540 L. D. Shvartsman and J. E. Golub
[14] A.V. Chaplik and L.D. Shvartsman, in All Union School on Surface
Physics, (USSR Academy of Sciences, Chernogolovka, 1983), pro-
ceedings of the Third All Union School on Surface Physics held in
Tashkent, October, 1983, p. 123.
[15] B. Laikhtman, R.A. Kiehl and DJ. Frank, J. Appl. Phys. 70, 1531
(1991).
[16] O.V. Kibis and L.D. Shvartsman, Poverkhnost (Surface), No. 7, 119
(1985), in Russian.
[17] N.S. Rytova, Vestnik Moskovskogo Universiteta (Moscow University
Proceedings), No. 3, 30 (1967).
[18] L.V. Keldysh, JETP Lett. 29, 658 (1979).
[19] A.V. Chaplik and M.V. Entin, JETP 34, 1335 (1972).
[20] G. Bastard, in Physics and Applications of Quantum Wells and Su-
perlattices, E.E. Mendez and K. von Klitzing, eds. (Plenum, New
York, 1987).
[21] Direct decay can always be made forbidden by slight changes in
material composition, or other quantum-well parameters.
 30
Evidence for Bipolaronic Bose-liquid and
BEC in High-Tc Oxides
A. S. Alexandrov
Interdisciplinary Research Centre in Superconductivity
University of Cambridge
Madingley Road
Cambridge, CBS OHE
UK

Abstract
Recent experiments on the near-infrared absorption, thermal conductivity
and the critical field HC2 in several high-Tc oxides are interpreted as a
manifestation of the Bose-Einstein condensation of small bipolarons.

1 Basic Model for High-Tc Oxides

To describe low-energy spin and charge excitations of metal oxides and
doped fullerenes with bipolarons, Alexandrov and Mott [1, 2, 3, 4] have
suggested that bipolarons are intersite in two possible spin states (S = 0
or 1), and a proportion of bipolarons are in Anderson localised states.
Our assumption is that all electrons are bound in small singlet or triplet
bipolarons and they are responsible for the spin excitations. Hole pairs,
which appear with doping, are responsible for the low-energy charge
excitations of the CuC>2 plane. Above Tc, a material such as YBCO
contains a non-degenerate gas of these hole bipolarons in a singlet or in
a triplet state, with a slightly lower mass due to the lower binding energy.
The low-energy band-structure includes two bosonic bands (singlets
and triplets), separated by the singlet-triplet exchange energy J, estimated
to be of the order of a few hundred meV. The half-bandwidth w is of
the same order. The bipolaron binding energy is assumed to be large
(A » T), and therefore single polarons are irrelevant in the temperature
region under consideration.
We argue that many features of spin and charge excitations in metal
oxides can be described within our simple model. In this mini-review, we
discuss the recent observation of the significant influence of the super-

541
542 A. S. Alexandrov
conducting transition on the near-infrared absorption [5], the thermal
conductivity enhancement below Tc [6] and the upper critical field, di-
vergent at low temperature [7]. These observations can be interpreted
as a manifestation of the Bose-Einstein condensation of bipolarons in
copper-based oxides.

2 Evidence for Bose-Einstein Condensation from Near-Infrared

Absorption
Evidence for the bipolaronic origin of the "spin" gap and for the Bose-
Einstein condensation comes from the temperature dependence of the
near-infrared absorption. It is well known that for frequencies much
higher than the superconducting energy gap, no change with temperature
in optical absorption is expected within the BCS theory. However Dewing
and Salje [8] observed the effect of the superconducting phase transition
on the near infrared absorption, with the characteristic frequency
v ~ 0.7 eV in YBa2Cu3O7_<5. The relative change of the integrated optical
conductivity at the superconducting transition turns out to be as high as
10%.
With the sum rule for bipolaronic conductivity, one can describe
the experimentally observed temperature dependence of the integrated
absorption intensity /(T), shown in Fig. 1 [5]:

where the non-coherent contribution, responsible mainly for the temper-

ature dependence of the normal-state absorption, is
2w J T
3AT /I - ve-( + )/ \
in(T) = l l n ln ((l ye
(2)
2vv \ 1— ye~ J'1
and the coherent contribution is:
rlw/T

The chemical potential fi = T i n y is determined by the bipolaron density

n, which is the sum of singlets and triplets :

and fi = 0 if T < Tc. Here n is the total number of pairs per cell and
the density of states is assumed to be energy independent within bands.
 Evidence for Bipolaronic Bose-liquid and BEC in High-Tc 543

1.2 I I I I

1.0 —

0.8 -

nfi I I I

T/Tc

Fig. 1. Integrated optical absorption of inter-site bipolarons (solid line) and

experiment [5][YBa2Cu3O7 (triangles) and YBaCuO/Fe (circles)], J/Tc = 4.4,
5C = 0.014, 5 = 1.6, w/Tc = 1.9.

The temperature-independent constant A in (2) is proportional to the

difference of the non-coherent absorption of singlet and triplet pairs.
In general, singlets absorb light in a frequency window different from
triplets due to an internal bonding-antibonding transition. The triplet
half-bandwidth w is assumed to be the same as the singlet one, and the
dimensionless constant dc determines the relative value of the coherent
contribution (i.e. without phonon and spin-wave "shakeoff").
A remarkable change of the slope of the optical absorption below Tc
(see Fig. 1) is due to the temperature-dependent coherent contribution
to the sum rule. This contribution vanishes at T = 0 due to the Bose-
Einstein condensation of singlet bipolarons. The condensed pairs on a
lattice cannot absorb light coherently because of the parity and time-
reversal symmetry. The pronounced decrease of the integrated absorption
intensity in the normal state is explained by the temperature dependence
of the singlet bipolaron density on the scale T ~ J. One can fit the
experimental curve, Fig. 1, with the reasonable value of the singlet-triplet
exchange energy J/Tc ^ 4.4. The latter is close to the characteristic value
of the spin pseudogap in "1-2-3" determined from neutron scattering [9],
Ecap/Tc ^3.5 — 5 (depending on the doping).
544 A. S. Alexandrov
3 Thermal Conductivity
Using the Bogoliubov spectrum for two-dimensional charged bosons
e(k) ~ y/k9 one can understand the pronounced maximum in the thermal
conductivity observed in two-dimensional copper-based oxides [6]. For
two-dimensional charged bosons, the gapless two-dimensional Bogo-
liubov mode has a divergent group velocity de(k)/dk ~ l/y/k in the
long-wave limit. This leads to strong suppression of the transport scat-
tering rate of excitations in the superconducting state. Both the acoustical
phonon and the charged impurity transport relaxation times show a sharp
enhancement in the superconducting state. This enhancement is due to
screening by the Bose-Einstein condensate, and to a large group velocity
of the two-dimensional Bogoliubov mode. The Boltzmann equation for
the excitations yields the thermal conductivity in the superconducting
state [6]:

r x dx
* <*
Jo
where K^o is temperature independent, t = T/Tc is the reduced temper-
ature and rj is proportional to the charged impurity concentration.
With Eq. (5) one can explain the experimentally observed enhancement
of the thermal conductivity below Tc, shown in Fig. 2. It should be
mentioned that the commonly accepted explanation of this enhancement
due to the lattice contribution to the heat transport has now been
rejected [11].
For carriers with a double elementary charge, the Lorentz number
should be at least four times smaller than for degenerate electrons. This
fact enables one to understand [2] the near equality of the in-plane
thermal conductivity above 100K in the insulating and 90K crystals of
YBa 2 Cu 3 0 7 -a [12].

4 Upper Critical Field

The interacting charged Bose gas is condensed in a field lower than a
certain critical value H* because the interaction with impurities [13], or
between bosons [14], broadens the Landau levels and thereby eliminates
the one-dimensional singularity of the density of states. The critical field
of Bose-Einstein condensation has an unusual positive curvature near
Tco, H*(T) ~ {TcQ-Tf1 and diverges at T -* 0, where T& ~ 3.31n2/3/m
is the critical temperature of Bose-Einstein condensation of an ideal gas
in zero field.
 Evidence for Bipolaronic Bose-liquid and BEC in High-Tc 545

30
r\ =0.01
a/ '••
1.6 - 2^ 25
E 7/'b\'-. pre - an
/ \^ • post - an

\ ^2°
\«\\
/ 0.02
2 1.2 -
\ •AN
o
1 / \ ^nr.-.r.:-.-.:.: ~

I / 0.05
\ ^ \ / ,40
, \ ;
"I 0.8 80 120 160
T(K)
N\
0.4 - ^ ^

0.0 0.2 0.4 0.6 0.8 1.0

Temperature (T / Tc )

Fig. 2. Theoretical temperature dependence of the thermal conductivity of near-

two-dimensional charged bosons for different charged impurity concentrations
rj [6]. Insert: enhancement of the in-plane thermal conductivity in the supercon-
ducting state of YBa2Cu307-5 [10].

It was shown recently [7] that localization drastically changes the

low-temperature behaviour of the critical field. H* saturates with the
temperature lowering at some value of the impurity concentration and at
higher concentrations the re-entry effect to the normal state takes place.
H* is determined as the field in which the first non-zero solution
of the linearized stationary Ginzburg-Pitaevskii equation [15] for the
macroscopic condensate wave function tpo(r) = (N|v?(r)|iV + 1) (N —• oo,
N/V = n = const) appears:

where 2e is the charge of a boson; A(r), l/imp(r) and ft are the vector,
random, and chemical potentials, respectively.
The definition of H* in Eq. (6) is identical to that of the upper
critical field H& of BCS superconductors of the second kind. Therefore
H* determines the upper critical field of bipolaronic or any "bosonic"
superconductor.
The first non-trivial delocalized solution of Eq. (6) appears at \i = Ec.
546 A. S. Alexandrov
Thus the critical curve H*(T) is determined from the conservation of the
number of particles n under the condition that the chemical potential
coincides with the mobility edge Ec (for more details see Ref. [7]):

H'(T) = Hd(Tc0/T)3/2 l-CT/Trf)) 372 -p(T/y)) , (7)

\ yn
with

/c=0

and temperature independent Hd = 0O/2TC^Q. The "coherence" length £o

is determined by both the mean free path / and the interparticle distance

while (j>Q = n/e is the flux quantum.

The last term in (7) is the number of localized bosons, which can be cal-
culated with a "single well-single boson" approximation [2, 7] (localized
charged bosons obey Fermi-Dirac statistics):

WlW
*-7. <io»

where the density of localized states NL(C) may be approximated in many

cases by the exponential tail:

jVL(6) = ^ e x p ( ^ — ^ ) , (11)

with y of the order of the binding energy in a single random potential

well and HL the total number of localized states per unit cell.
The localization does not change the positive "3/2" curvature of the
critical magnetic field near Tc [13], as shown in Fig. 3. I believe that
this curvature is a universal feature of a charged Bose-gas, which does
not depend on a particular scattering mechanism or the approximations
made. At low temperature (T « y), the temperature dependence of if*
turns out to be drastically different for different impurity concentration,
as seen in Fig 3. An upward curvature of HC2 near Tc has been observed
in practically all superconducting oxides, including cubic ones.
Recently, divergent HC2(T) has been measured in a wide temperature
range starting from the mK-level up to Tc [16]. Resistively determined
Hc2 values from T/Tc = 0.0025 to T/Tc = 1 in a Tc = 20K single
crystal of Tl2Ba2CuO6 follow a temperature dependence that is in good
qualitative agreement with the type of curve shown in Fig. 3 for n^jn ^ 1.
 Evidence for Bipolaronic Bose-liquid and EEC in High-Tc 547

1.0
CD

0.8
• \ \ \

Kv
\ \ \
\ n L/n = o
0.6
a4
V \
.1 0.4 . \ \08

I6 0.2
/—^x.
1.1
v° \
\.

0.0
0.0 0.2 0.4 0.6 0.8 1.0
Temperature T / T co

Fig. 3. Temperature dependence of the critical magnetic field of Bose-Einstein

condensation [7](in units of Hd(Tco\n2/y)3/2) for different relative number of
localized states nL/n and y/Tco = 0.2.

Osofsky et al. [17] also observed a divergent upward temperature

dependence of the upper critical field Hc2(T) for thin BSCOfilms,which
was five times longer than expected for a conventional superconductor
at the lowest temperature. They compared their data with Eq. (7).

5 Conclusion
Earlier evidence for small (bi)polarons in copper-based oxides comes
from the photoinduced infrared absorption, measured by Heeger and by
Taliany [18, 19], from the observation by Sugai [20] of both infrared and
Raman-active vibration modes, and from XAFS results on the radial
distribution function of apex oxygen ions [21, 22].
The experimental observations discussed above, and many other fea-
tures of high-Tc copper oxides [3, 4], lead us to the conclusion that a
charged Bose-liquid of small intersite bipolarons is a simple but far-
reaching model of high- Tc superconductors.

Acknowledgments. I am grateful to Sir Nevill Mott, Yao Liang, Alex

Bratkovskii, Ekhardt Salje and Andrew Mackenzie for helpful stimulating
discussions. I appreciate the financial support from the Leverhulme Trust.
548 A. S. Alexandrov
References
[I] A.S. Alexandrov, J. Low Temp. Phys. 87, 721 (1992); Physica C 182,
327 (1991).
[2] N.F. Mott, Physica C 205, 191 (1993).
[3] A.S. Alexandrov and N.F. Mott, Supercond. Sci. Technol. 6, 215
(1993).
[4] A.S. Alexandrov and N.F. Mott, Invited paper at the MOS-
Conference, Eugene, Oregon, USA, July (1993), to be published.
[5] A.S. Alexandrov, A.M. Bratkovsky, N.F. Mott and E.K.H. Salje,
Physica C 215, 359 (1993).
[6] A.S. Alexandrov and N.F. Mott, Phys. Rev. Lett. 71, 1075 (1993).
[7] A.S. Alexandrov, Phys. Rev. B 48, 10571 (1993).
[8] H.L. Dewing and E.K.H. Salje, Supercond. Sci. Technol. 5, 50 (1992).
[9] J. Rossat-Mignod, L.P. Regnault, P. Bourges, C. Vettier, P. Burlet
and J.Y. Henry, Physica Scripta 45, 74 (1992).
[10] J.L. Cohn et a/., Phys. Rev. B 45, 13144 (1992).
[II] R.C. Yu, M.B. Salamon, J.P. Lu, and W.C. Lee, Phys. Rev. Lett. 69,
1431 (1992).
[12] S.J. Hagen, Z.Z. Wang, and N.P. Ong, Phys. Rev. B 40, 9389 (1989).
[13] A.S. Alexandrov, doctoral thesis, Moscow Engineering Physics
Institute, Moscow (1984); A.S. Alexandrov, J. Ranninger, and
S. Robaszkiewicz, Phys. Rev. B 33, 4526 (1986).
[14] A.S. Alexandrov, D.A. Samarchenko, and S.V. Traven, Zh. Eksp.
Teor. Fiz. 93, 1007 (1987) [Sov. Phys. JETP 66, 567 (1987)].
[15] E.M. Lifshitz and L.P. Pitaevskii, Statistical Physics, Part 2 (Perga-
mon, Oxford, 1980), p.117.
[16] A.P. Mackenzie, S.R. Julian, G.G. Lonzarich, A. Carrington,
S.D. Hughes, R.S. Liu and D.C. Sinclair, Phys. Rev. Lett. 71, 1238
(1993).
[17] M.S. Osofsky et a/., Phys. Rev. Lett. 71, 2315 (1993).
[18] Y.H. Kim, C M . Foster, A.J. Heeger, S. Cos, and G. Stucky, Phys.
Rev. B 38, 6478 (1988).
[19] C. Taliani, A.J. Pal, G. Ruani, R. Zamboni, X. Wei, and Z.V. Var-
dini, Electronic Properties of High Tc Superconductors and Related
Compounds, H. Kuzmany, M. Mehring, and J. Fink, eds., Springer
Series of Solid State Science 99 (Springer-Verlag, Berlin, 1990), p.
280.
[20] S. Sugai, Physica C 185-189, 76 (1991).
 Evidence for Bipolaronic Bose-liquid and BEC in High-Tc 549
[21] J. Mustre de Leon, S.D. Conradson, I. Batistic, and A.R. Bishop,
Phys. Rev. Lett. 65, 1675 (1990).
[22] J. Mustre de Leon, I. Batistic, A.R. Bishop, S.D. Conradson, and
S.A. Trugman, Phys. Rev. Lett. 68, 3236 (1992).
 31
The Dynamic Structure Function of Bose
Liquids in the Deep Inelastic Regime
A. Belie
International Center for Theoretical Physics
P.O.Box 586
34100 Trieste
Italy

Abstract
We study the dynamic structure function S(k,co) of Bose liquids in the
asymptotic limit k,co —• oo at constant y = j(co — k2/2m), using the or-
thogonal correlated basis of Feynman phonon states.

1 Introduction
In the last few years there has been a growing interest in the possibilities
for experimental determination of single-particle momentum distributions
in non-relativistic many-body systems by means of inelastic neutron
scattering at large momentum transfers. This interest was generated by
the advent of pulsed-neutron sources that made possible measurements
with substantially larger momentum transfers then before. For example,
in the case of liquid 4 He a momentum transfer of 20-30 A" 1 was
achieved [1], which is much larger than the rms momentum of atoms in
the liquid of ~ 1.6 A" 1 . If it is assumed that at such large momentum
transfers, the potentials between the atoms in the liquid are negligible
compared to the large kinetic energy of the struck atom, the deep inelastic
response is completely determined by the initial single-particle momentum
distribution. This assumption is known as the impulse approximation
(IA) [2, 3]. At large momentum transfers fe, the dynamic structure
function (DSF) 5(k,co) exhibits the phenomenon of y-scaling, i.e. the
combination J(y) = ^S(k,co), called the scaling function, depends solely
upon the scaling variable [2] y = f(co — ^ ) and not separately upon k
and co. The scaling function in the IA is given by:
i r00

dqq
JiA(y) = ncS(y) + ^ 5 " I

550
 The Dynamic Structure Function of Bose Liquids 551
where p is the density of the system, n(q) is the single-particle momentum
distribution and nc the corresponding condensate fraction. However, the
^-scaling property of the deep inelastic response is more general than the
IA in the sense that it may hold even when the IA fails.
If the IA is valid, the momentum distribution can be extracted from
the deep inelastic response in a model-independent way. This gives rise
to the exciting possibility that the momentum distribution is a directly
observable quantity. If this were true, not only would we be able to check
the accuracy of our calculations and determine interparticle potentials
more reliably, but also we could test fundamental physical ideas such as
Bose condensation!
Unfortunately, in the most interesting cases, such as liquid 4He, there
is a strong repulsive short-ranged component in the interparticle po-
tential which is not negligible compared to the recoil kinetic energy
even for the large momentum transfers achieved in recent experiments.
Thus the IA must be corrected to account for interaction of the re-
coiling particle with the surrounding medium. These corrections are
called final state interaction (FSI) corrections. Clearly, it would be
desirable to develop an analytic, first-principles theory of the dynamic
structure function at large momentum transfers, in order to understand
the FSI.
There were many attempts that start from the IA but take into account
the interactions in the initial state of the target, and improve upon it
by incorporating the effects of FSI [4]. This procedure becomes more
and more difficult as k decreases and the FSI become more important.
In contrast the response at small k and co is easily treated by using
Feynman phonon states [5]. In particular the orthogonal correlated basis
(OCB) formalism [6] based on Feynman's ideas is quite successful [7]
in explaining the observed S(k,co) at k < 2 A'1. The small and large k
methods have different starting points; the momentum distribution n(q)
is the main input for the IA, while the OCB formalism uses the static
structure function S(q) as the main input.
Although in principle the OCB formalism can be used to study S(k,co)
at all k and co, in practice the calculations [7] become technically com-
plicated as k increases. Nevertheless we have shown [8] that, by using
field-theoretical techniques in OCB perturbation theory, it is possible
to calculate the S(k,co) in the scaling limit, and thus extend this low
k method to the k —> oo limit. In this paper, we present results of
this approach as applied to the case of a Bose liquid at zero tem-
perature.
552 A. Belie
2 OCB Calculation of S(k,co) in the Scaling Limit
The DSF S(k,co) can be written as
1
S(k,co) = —ImD(k, co), (2)
n
where
k
N

n=0

x(H' [co - Ho + Eo + iff]"1)" Pk 1°)' (3)

is the density-density response function, pk = J2ie~lkTi *s ^ e Fourier
component of the density, rj is a positive infinitesimal, and the Hamil-
tonian of the system is divided into diagonal and off-diagonal parts with
respect to the set of intermediate states | i):
H = Ho + H\ (4)

H04j = Sijiil H \i), Hrij = { \ - 5 i j ) { i \ H \ j ) . (5)

If the intermediate states are taken to be | i) =| p)i | W)JV-I, where
| p) is a plane wave state of the struck particle, and | n) are eigenstates
for N — 1 background particles, then the expansion (3) becomes the
so-called Gersh series [9]. When the interparticle potential is weak, it is
an asymptotic expansion in powers of m/k around the zeroth-order term
that is equal to IA. However, when the interaction contains a hard core,
i.e. when the interparticle potential v(r) is divergent enough at small r so
that its Fourier transform v(p) is not defined, the Gersh series fails. The
reason for this failure is that the non-perturbed Hamiltonian decouples
the recoiling particle from the rest of the system and does not take into
account strong short-range correlations that would normally prevent
the struck particle from entering the hard-core region of background
particles.
Our choice for intermediate states are Feynman phonons (FP):

where
 The Dynamic Structure Function of Bose Liquids 553
and | 0) is the ground state of the system. FP states do contain the
appropriate short-range correlations, and are reasonably close to the
exact eigenstates of the Hamiltonian. Thus we expect Hr to be small
and series (3) to be rapidly convergent. However, they are not mutually
orthogonal, which is the general property of correlated basis (CB) states.
Such states can be used in the perturbation expansions, but require a
special formalism that takes into the account their non-orthogonality, and
it is not clear whether the non-orthogonality effects that are introduced
when the expansion is truncated are negligible or not. In order to use
normal perturbation theory, we have to orthogonalize the CB states.
This can be achieved using the Lowdin transformation [10], but resulting
OCB states have higher energies than CB states. Perturbative corrections
move the energy down again, and both the increase in energy due to
the Lowdin transform and the decrease due to perturbative corrections
are larger than the net displacement [6]. In view of these difficulties, the
OCB perturbation theory (OCBPT) is not the preferred tool for studies
of quantum liquids. Recently, however, a new orthogonalization scheme,
free of the problems mentioned above, was proposed [11]. It is expected
that the convergence of OCBPT does not depend upon the specific nature
of the bare interaction, i.e. if it has a hard-core or not. For these reasons,
we use it to study the scaling properties of the response at large k and co.
When the orthogonalized FP in (6) are used as intermediate states in
(3), it can be shown [8] that the diagonal matrix elements HQJJ are of
order O((fe/m)2), while the off-diagonal matrix elements H[j are of order
0(k/m\ from which the scaling property

-D(k,co) = Ds(y) + O(m/k) (8)

m
immediately follows. The only assumption used to derive this result is
that the static structure function S(k) of a Bose liquid approaches unity
exponentially fast as k —• oo, which is true even when the interatomic
interactions are hard, as for example, the Lenard-Jones 6-12 potential.
Here we do not address the problem of a hard sphere gas where (S(/c)—1)
does not go exponentially to zero at large k due to the singularity in
the derivative of the pair distribution function g(r) at r — rc (the hard
core radius). The correction term of order 0(m/k) in (8) was not studied
using OCBPT. It is known that it is an antisymmetric function of y, and
can be estimated using sum-rule arguments [12].
An exact calculation of the scaling function J(y) is difficult because of
the large number of different types of vertices, representing the matrix
554 A. Belie
elements Hfij9 in the expansion (3). However, all the diagrams based on
the vertex at which the hard phonon (with momentum ~ k) emits or
absorbs a soft phonon (with momentum <C k) can be summed by the
standard field-theoretical methods without further approximations. The
resulting system of coupled integral equations can be solved in the closed
form to obtain [8]:

j(y) = C I d(y) + - T ds cos(ys)

n
Jo

-1 >, (9)

where we have introduced the constant

(10)

The function /(/) depends on the static structure function S(q) only, and
is given by:

(11)
_n=0
where the coefficients ocn = 1,-1/2,3/8,... are those that appear in the
expansion of (1 + *)~ 1/2 , and the functions /„(/) are defined recursively
as:
(12)

r d3h |
), n>\. (13)
Eqs. (9)—(13) represent the main results of this work.

3 Discussion and Numerical Results

In the limit y —• 0 the scaling function J(y) is completely determined
by the properties of the function /(/) for small /. More precisely, from
(11)-(13) it follows that

lim J~Y = lir (14)

where c is the sound velocity in the liquid, and we obtain [8]

1
(15)
 The Dynamic Structure Function of Bose Liquids 555
where the constants A and B are given by

A „ r , r / 1 sin(s/)
r° ,,sin(s -i ,
A = AnpC I ds exp —?- / dl—V"
me
B = Cy. (16)

The scaling function in IA (Eq. (1)) exhibits the same small y behavior
with the coefficients
f00 me
CIA = nc, AIA= / dqqn(q), BIA = limqn(q) = nc— 9 (17)
2
Jo «^°
and it follows that the ratio B/C in our theory is identical to that in IA:
£ =^. (18)
C CIA
Next, we want to discuss the origin of the S -function peak at y = 0. In
general, the presence of the ^-function in DSF means that probe couples
to a long-lived state with a given energy and momentum. The inverse
lifetime of this excitation is expected to be proportional to k on the basis
of the semi-classical value T" 1 ~ pva, where v = k/m is the velocity
and a the average two-body scattering cross section. However, at y = 0
this argument fails, as can be seen in the second-order contribution
to (3). In that order, an off-shell FP with momentum k and energy
co = k2/2m + ky/m can decay into two FP of momenta k — 1 and 1, thus
acquiring a lifetime

[KtffWiy i)). (19)

z
pmJ
When y ± 0 we obtain T ~ fe"1, as expected from semi-classical argu-
ments, but when y = 0 (on-shell FP) the energy-conserving <5-function
forces the cosine k • 1 to be of the order k~x which makes the integrand
small, and gives the on-shell FP a long lifetime T ~ k.
In our calculation of the scaling function J(y), we have taken into
account only the processes in which one soft phonon is emitted or
absorbed at a time. We can estimate the validity of this approximation
by checking if J(y) satisfies the sum-rules [6]. It is easy to show that the
y° and y sum-rules:

r dy J(y) = 1, r dy yJ(y) = 0, (20)

J— 00 J—CO
556 A. Belie

o< 0.3 -

1.5 2.0 2.5 3.0

Fig. 1. The scaling functions of liquid 4He at equilibrium density and zero
temperature. Solid line: J(y) of Eq. (9) obtained using S(q) of Ref. 13; dotted
line: JiA{y) generated from n(q) of Ref. 14. The ^-function peaks at y = 0, with
the respective strengths of 0.15 and 0.092, are not shown.

are satisfied by J(y) in (9). For the y2 (kinetic energy) sum-rule, we obtain
<} /»OO 1 /*OO

-^Ek= dyy2J(y) = —Y- dll2f(l). (21)

The degree to which this sum-rule is satisfied is a measure of importance

of neglected processes.
Finally, in the uniform limit (|g(r) — 1| < 1), the scaling function J(y)
should be equal to the JiA(y)- It can be shown [8] that in this limit, the
function /(/) becomes:

MD = (22)

and that the scaling function J(y) of (9)-(10) reduces to (1), as expected.
When (9)—(13) are applied to the case of liquid 4 He, using the mea-
sured [13] S(q), the value C = 0.15 and the J(y) shown in Fig. 1 are
obtained. For comparison, JiAiy) generated from the variational n(q)
(Ref. [14]) which has the condensate fraction nc = 0.092, is also shown.
Finally, using the value Ek = 14.8K from Ref. [14] and S(q) from
Ref. [13], we find that
3 r
* dll2f(l) = 0.91. (23)

As was argued above, the closeness of this value to 1 suggests that the
 The Dynamic Structure Function of Bose Liquids 557
neglected terms in expansion (3) are much smaller than those taken into
account.

References
[I] RE. Sokol, in Momentum Distributions, R.N. Silver and RE. Sokol,
eds. (Plenum, New York, 1989).
[2] G.B. West, Phys. Rep. 18C, 263 (1975).
[3] RC. Hohenberg and P.M. Platzman, Phys. Rev. 152, 198 (1966).
[4] R.N. Silver, in Momentum Distributions, R.N. Silver and RE. Sokol,
eds. (Plenum, New York, 1989); A.S. Rinat, Phys. Rev. B 40, 6625
(1989).
[5] R.R Feynman, Phys. Rev. 94, 262 (1954).
[6] E. Feenberg, Theory of Quantum Liquids (Academic Press, New
York, 1969).
[7] E. Manousakis and V.R. Pandharipande, Phys. Rev. B 33,150 (1986).
[8] A. Belie and V.R. Pandharipande, Phys. Rev. B 45, 839 (1992).
[9] H.A. Gersch and L.J. Rodriguez, Phys. Rev. A 8, 905 (1973).
[10] P.O. Lowdin, J. Chem. Phys. 18, 365 (1950).
[II] S. Fantoni and V.R. Pandharipande, Phys. Rev. C 37, 1697 (1988).
[12] A. Belie and V.R. Pandharipande, Phys. Rev. B 39, 2696 (1989).
[13] H.N. Robkoff and R.B. Hallock, Phys. Rev. B 25, 1572 (1982).
[14] E. Manousakis, V.R. Pandharipande, and Q.N. Usmani, Phys. Rev.
B 31, 7022 (1985); ibid. B 43, 13587 (1991).
 32
Possibilities for BEC of Positronium
P. M. Platzman and A. P. Mills, Jr
AT&T Bell Laboratories
600 Mountain Avenue
Murray Hill, New Jersey 07974
USA

Abstract
We review the proposal for Bose-Einstein condensation of positronium
atoms. All of the ingredients necessary to achieve BEC of Ps atoms are
currently available.

1 Introduction
In this volume several authors have discussed and described a variety
of weakly interacting systems which might display BEC. In this short
contribution we suggest that a dense gas of positronium (Ps) atoms in
vacuum is a rather ideal but somewhat more exotic system that might
be a very good candidate for observing a weakly interacting BEC.
Recent investigations of the interactions of positrons (e+) and (Ps)
with solids have led to extraordinary improvements in the kinds of low
energy experiments we can do with the positron [1, 2]. All of the
ingredients necessary to achieve BEC of Ps atoms are currently available.
We envision a scenario where roughly N = 105 Ps atoms are trapped in a
volume V ~ 10~13 cm3 and allowed to cool through the Bose transition
temperature of 20 — 30K in a time of the order of nanoseconds. In the
following we discuss the relevant interactions and describe how Ps BEC
can be achieved.

2 Single Positronium Physics

Ps is comprised of an e+ — e~ bound in a hydrogenic orbit. Its mass, 2me,
is extremely light compared to H, an important ingredient for achieving
reasonable Bose condensation temperatures. Its binding energy (6.8 eV)

558
 Possibilities for BEC of Positronium 559
is half that of H. The ground state of Ps is a spin singlet separated from
an excited triplet by an energy AEST — 10K. Ps annihilates itself, i.e., it
becomes high energy 7-rays [3],
The annihilation characteristics of the Ps atom are dependent on which
of the two ground states it is in. The ground state singlet (s) is short
lived with a lifetime TS = 1.25 x 10~10 s. For a Ps atom at rest, the decay
occurs with the emission of two 0.5 MeV y-rays which come out precisely
in opposite directions due to momentum conservation. If the Ps atom
is moving with momentum /?, the y -rays come out with a small angle
0 ^ p/mc relative to each other. The nearby triplet state (t) is prohibited
by selection rules from decay into the two y -ray channel. Instead it decays
by three y -rays with a spread of energies and a much longer lifetime of
Tt = 1.42 x 10~7 s. A magnetic field mixes the triplet with singlet, thus
rapidly quenching the triplet.
Like hydrogen, the PS2 molecule exists in an overall singlet state [4]. It
has a binding energy EM = 0.4 eV. Two low energy Ps atoms scatter from
one another with a cross section o = Ana1 determined by how close the
bound state is to the continuum [5]. Specifically, as = (2mEM)~{^2 = 3 A,
and GS = 10~14 cm2. On the other hand the non-singlet channel has
a scattering length more like a Bohr radius, i.e., at = 1 A and ot =
10~15 cm2. Of course, the experiment we are considering here would,
for the first time, be sensitive to such cross sections and a measurement
of them should be possible.

3 e+ and Ps at Solid Surfaces

In order to make the case for Bose condensation we need to understand,
at least qualitatively, how a mildly energetic (« 10 keV) beam of e +
interacts with a semi-infinite slab of some simple solid [1]. The interaction
process is sketched in Fig. 1. The incident beam enters the solid and
begins to make electronic excitations in the material, plasmons, inner shell
ionizations, etc. The energetic e+ reaches an energy of 10 eV or so in a
short time T < 10~13 s, and does so in a distance of roughly 1000 A for
a 5 keV beam [2]. The mechanism is primarily plasmon emission. At
this point in its history, the e+ begins to slow down to thermal energies
by electron-hole pair excitations and phonon emission. In this regime,
its motion in the solid may be described as diffusive. Calculations and
experiments convincingly demonstrate that diffusion constants Dp for our
superthermal e+ are roughly 1 cm2/s [1]. Since annihilation times in the
solid xA are, in most materials, a fraction of a nanosecond, a typical
560 P. M. Platzmanand A. P. Mills, Jr

Vacuum Solid

Polarized
e+(=10keV) Plasmon

•HOeV)

Diffusion

Thermal e+

Fig. 1. Schematic of the interaction of a 10 keV beam with a typical solid such
as Si.

e+ diffuses a distance • xA = 2 x 103A. This simple estimate

suggests that some large fraction (50%) of our initial 5 keV e+ beam
returns to the surface, where we must consider its energetics.
The work function (the energy difference between the lowest state
inside the solid and the vacuum) for an e+ consists of three parts [1, 2].
First is the correlation energy, which is positive, i.e., it keeps the e+ inside
the solid. It must be less then the 6.8 eV Ps binding energy. Second
is the negative dipole part which exists because the e~ in the solid spill
out into the vacuum. The value of this energy varies but is typically
between 5 and 10 eV. Finally there is the negative kinetic energy part
typically 2-4 eV. This part comes from the fact that the bottom of the
e+ band moves up when the e+ is squeezed into the space between the
ions. These general considerations lead one, a priori, to believe that the
e+ work function could be negative. In fact, simple materials [Ni, Al, W,
...] do have negative e+ wavefunctions [1, 2].
An e+ arriving at the surface of a negative work function material
simply slides down the hill out into the vacuum. When it gets into the
 Possibilities for BEC of Positronium 561
regime where the electron density is low enough it can, and does, pick up
an electron and form neutral Ps which can also be ejected from the solid.
Emission of e+ and Ps occurs with about 50% probability. Typically,
the energy of the e+ is non-thermal, i.e., directed perpendicular to the
surface and greater than 1 eV, while the Ps energy distribution is similarly
non-thermal but is most often greater than 1 eV [2]. This unique surface
interaction characteristic permits the generation of very bright pulsed e +
beams (see Canter in Ref. [2]) and is essential, as we shall see, to the
proposed BEC experiment.

4 Dense Positronium
The "initial" condition for the Ps condensation experiment is sketched
in Fig. 2. A bunched 1 nanosecond brightness-enhanced microbeam
(diameter ljum) consisting of N > 106e+ at an energy of 5 keV is incident
at time t = 0 (1 nanosecond uncertainty) on the surface of a Si target
which has, at the point of entry, a small cavity etched into it. We
choose Si rather than a metal for several reasons, to be discussed shortly.
The geometry we envision will typically be a cylindrical cavity of 1 fim
diameter with a height of 1000 A. These dimensions are rather easily
achieved with conventional lithography techniques. The e+9 as discussed,
stop in the Si at a depth of about 1000 A. About one-quarter of the
initial e+, because of diffusion and the negative work function for Ps, will
be reemitted as electron volt Ps into the cavity. The singlet Ps rapidly
decays leaving us with a hot gas of bosons (triplet Ps atoms), at a density
n = N/V ^ 1018 cm"3.
In Ref. [6] we consider in some detail the time evolution of this
confined hot gas of Ps atoms due to collisions with the wall and with
other Ps atoms in the cavity. We conclude that such a gas would indeed
cool to a Bose condensed state by phonon emission at the walls in a
time short compared to a triplet lifetime. However, the serious difficulty
which must be avoided is quenching of the triplet state by collisions of
Ps atoms with the same wall and with each other.
Scattering from the walls occurs at a very rapid rate, with Tw — 1012
1
s" for Ps atoms with a energy of 0.1 eV. If the solid wall of the cavity
is metallic, i.e., there are unpaired spins, then exchange of the e~ in
the triplet Ps with e~ in the metal occurs with high probability and the
triplet state is quenched very rapidly. If the wall is insulating, i.e., the
electron spins are paired, then exchange scattering is often energetically
562 P. M. Platzmanand A. P. Mills, Jr

1 st remoderator
10 6 e+
1cm
5nsec
1mm

30jim|[

2nd remoderator Sample

Fig. 2. Sketch of a pulsed brightness enhanced e+ microbeam incident on a Si

target with a void etched in it. The dimensions are typical dimensions for the
proposed experiments.

forbidden and the only allowed process is a weak relativistic effect, which
is negligible.
The scattering of triplet Ps atoms with one another is the mechanism
by which the system equilibrates internally. This type of scattering can
also lead to exchange scattering which will annihilate the gas before it
reaches BEC. In the s wave scattering approximation the bulk triplet-
triplet scattering rate characterized by an s-wave scattering length of
lA is roughly Tt = 1010 s"1. It is rapid enough to equilibrate the gas as
it cools.
In Ref. [6] we have shown that collisions between triplets with their
spin pointing in different directions will surely lead to conversion of
triplet to singlet at roughly the same rate. However, it is possible to
prevent this type of annihilation by using a fully polarized e+ beam.
Fully polarized e+ beams are commonly available since all beta-decay
sources generate e+s by a parity non-conserving process which leads to
e+ polarized with their spin in the direction of their momentum. Slowing
down inside a low Z solid such as Si leaves the initially fully polarized
e+s about 50% percent polarized (see Berko in Ref. [2]).
For such a polarized beam of e+ incident on an unpolarized e~ target,
it is easy to show, based on rather general arguments [6], that triplet-
 Possibilities for BEC of Positronium 563

Counts'.
Bose
condensate

Maxwellian
Ps

e
Fig. 3. Sketch of a hypothetical y-ray angular correlation spectrum for BEC Ps.
The double humped structure is characteristic of the Bose condensed fraction.

triplet collisions rapidly lead to a fully polarized Ps gas consisting of

electrons and positrons with spins in the same direction as the initial
polarized e+ beam. This fully polarized gas decays with a long triplet
lifetime and should quite happily cool through the BEC.

5 Observation of the Bose Condensate

Because of the unique annihilation characteristics of Ps, we will be able
to observe the condensate directly, by converting a small but significant
fraction of triplets to singlets. This can be accomplished by applying
a small transverse magnetic field ~ 0.1T to mix the singlet and triplet
states. The small amount of singlet will annihilate into two y-rays. The
angle between the two y-rays, as we have pointed out, directly mirrors
the momentum of the Ps atom. Thus an angular correlation profile
with enough resolution will indicate the presence of a zero momentum
condensate. If the sample is below Tc, the angular correlation profile
will resemble the angular correlation data shown in Fig. 3. The best
angular resolution currently attainable is a A 0 = 0.1 milliradian. This
implies that we can measure the velocity distribution function with a
resolution At; ~ cA& ~ 3 x 106 cm/s, or about two times better than the
velocity of a Ps atom at room temperature. With some improvement in
A®, there should be unambiguous evidence of a qualitative peaking of
the angular correlation profile as a function of time. If the system were
to Bose condense, it might also be possible to demonstrate the existence
of a superfluid by placing a small hole in the cavity containing the Ps
564 P. M. Platzmanand A. P. Mills, Jr
atoms. Below Tc the "superfluid" might rapidly leak out to the vacuum,
where it could be easily detected.
Observation of a Bose condensate, while interesting in itself, will not
be the only goal of such experiments. We would, for the first time, begin
to obtain direct experimental information about the dense Ps gas, that
is, Ps-Ps collisions and possibly Ps molecule formation. Secondly, all of
the problems connected with the time evolution of the condensate, for
example, how such a condensate responds to changes in temperature and
density, will be observable. It may also be possible to look carefully near
the transition for deviations from ideal-gas behavior. Such non ideal
behavior will of course become more apparent as we increase the density
one to two orders of magnitude. In any event it seems likely that this
exotic, but rather ideal, system will be a fascinating one to look at in the
near future.

Acknowledgements. The authors would like to thank S. Berko, K. G.

Lynn and K. F. Canter for numerous discussions.

References
[1] A.P. Mills, Jr, in Positron Solid State Physics, edited by W. Brandt
and A. Dupasquier (North Holland, Amsterdam, 1983) p. 421; P.J.
Schultz and K.G. Lynn, Rev. Mod. Phys. 60, 701 (1988).
[2] Positron Studies of Solids Surfaces and Atoms, edited by A.P. Mills,
Jr, W.S. Crane and K.F. Canter (World Scientific, Singapore, 1985).
[3] H. Bethe and E. Salpeter, Quantum Mechanics of One and Two Elec-
tron Atoms (Academic, New York, 1957).
[4] E.A. Hylleraas and A. Ore, Phys. Rev. 71, 493 (1947); Y. K. Ho,
Phys. Rev. A 33, 3584 (1986). For a review see M.A. Abdel-Raouf,
Fortschr. Phys. 36, 521 (1988).
[5] T.Y. Wu and T. Ohmura, Quantum Theory of Scattering (Prentice Hall,
New Jersey, 1962).
[6] P.M. Platzman and A.P. Mills, Jr, Phys. Rev. B 49, 454 (1994).
 33
Bose-Einstein Condensation and Spin
Waves
R. Friedberg and T. D. Lee
Department of Physics
Columbia University
New York, NY 10027
USA
H. C. Ren
The Rockefeller University
New York, NY 10021
USA

The Quantum Lattice Model [1] is a natural basis for computer simu-
lations of Bose liquids as well as for possibly realistic models of mobile
bosonic excitations in a solid. It depicts particles obeying Bose statistics
occupying sites of a lattice and possessing kinetic energy by virtue of a
hopping term —2 ^2,{b\bj + b^bt) in the Hamiltonian, where ^ counts

each nearest neighbor pair once; but it is understood that no more than
one particle may occupy the same site.
It is of interest to add an attractive two-body potential at nearest
neighbor separation, so that the Hamiltonian may be written
bt + 2 ]T(bJ - b])(bt - bj) - a ^ b]b)btbj. (1)

The problem is then to study (1) in a Hilbert space restricted to states

satisfying bibt\ > = 0 for all i. A convenient formal way to achieve this is
to study the Hamiltonian
HG = H + G^bjbjbibt (2)
i

and take the limit G -» +oo in the results [2].

In the case where a = 0, this is the natural lattice analog of the hard-
sphere boson problem, and the "infinite" potential G can be dealt with
in the same way as the "infinite" potential forbidding two spheres to
penetrate. One introduces the binary dressed vertex F(P,k,q,£) obtained
by summing over all ladder diagrams involving two particles with initial
momenta ^P±k, final momenta ^P±q and external energy E. All physics

565
566 R. Friedberg, T. D. Lee and H. C. Ren
can be calculated in powers of F, which itself remains finite as G —• +00.
Particularly important is the long-wavelength scattering amplitude

iy o = ^(0,0,0,0) = 4/g 0, (3)

where go is the solution at the origin of the Poisson equation with unit
source. The continuum analog of |yo is 4na/m, where a,m are the
diameter and mass of the spheres.
If a 7^ 0 the same treatment can be applied with
1 4
" a
go ~ ^ (4)

where g\ is the solution of the same equation at neighboring sites; thus

x(go — gi) = 1> where x *s the number of nearest neighbors. For low-
momentum phenomena one replaces the a term in H by \yoLb\b\bibi
and omits higher ladder terms already included in F. The quantity yo
then determines the physics of the dilute system; for example the ground
state energy agrees with that found by substituting ^70 for Ana/m and
setting m = 4 (in accordance with the hopping term of (1)) in Lee and
Yang's formula [3], but at higher powers of density, the lattice departs
from the continuum.
An insight into the numerator 4 — a in (4) is gained by considering the
anisotropic spin-| Heisenberg model

HHeis = £ ( - ' « • *j + Ao\o)) - KLicrj. (5)

If we define

Ci = \ ^ + iaJ), c]=l-{<j*-iG>) (6)

so that
[ci,c}]=a*diJ = (l-2c)ci)5ij, (7)

then (6) becomes

\-2{c}Cj + c)ct) - (1 - A){\ - 2cjc,)(l - 2c)cj)]

cjc). (8)

Now if we could only replace the c\ by boson operators b\ (ignoring

 Bose-Einstein Condensation and Spin Waves 567
the term 2c-c; in (7)) we see that Hnets would be the same, apart from a
constant term, as if in (1), with the identifications
a = 4(1-A), (9)

A). (10)

Thus the isotropic Heisenberg model (A = 0) corresponds to the case

a = 4 or yo = 0 in (4).
Indeed, Dyson [4] has shown that the isotropic Heisenberg model can
be made to correspond with a bosonic system provided that in H we
replace the a term with a certain nonhermitian two-body scattering term
whose form is such that it vanishes at zero momentum. In this way
Dyson could treat the dynamics of spin waves.
We have found it simpler to treat (5) by relating it to (2). That is, spin
excitations are just like bosons with same-site exclusion, and we back this
up with a theorem [2] showing that those eigenvalues of HQ that remain
finite as G —> +oo do tend toward eigenvalues of Hnets + const, and that
the respective eigenvectors of HQ tend toward permitted states (no two
particles on a site) corresponding to the respective eigenvectors of Hneis-
Armed with this theorem, we can treat the anisotropic Heisenberg
model by converting it to (2), thus gaining the advantage of diagrammatic
expansion of a Hermitian interaction, while retaining Dyson's insight that
when ^4 = 0, the low-momentum spin waves "slide past" one another.
The physical equivalence of a bosonic to a fermionic system raises
an apparent paradox: the commutator (7) is always different from the
bosonic commutator no matter what the value of G, and yet this com-
mutator is the zero-time value of the fourier transform of the propagator
which reflects the physics of the system. This contradiction is resolved
by noting that the bosonic propagator for any fixed energy converges as
G -> +oo to the fermionic propagator, but that this convergence is not
uniform in the energy. Thus the fourier transform converges to agree-
ment for any value of time other than zero. The one anomalous point is
that of measure zero and does not affect the physics.
Bose condensation in (2) can take place for a < 4 (yo > 0); the corre-
sponding (4) with A > 0 has "x-y-like" anisotropy, and the condensation
involves breaking the symmetry about the z-axis. For A < 0 this sym-
metry is not broken, and the corresponding (2) with a > 4 and 70 < 0
has a two-body attraction strong enough to bind, so that the system
will separate, "Ising-like," into a packed liquid and a gas, with no Bose
condensation.
568 R. Friedberg, T. D. Lee and H. C. Ren
Bose condensation in a dense system corresponds to the anisotropic
Heisenberg model with A > 0 and \h\ not too large; then there is
a large spontaneous symmetry breaking represented by the transverse
components of < (X, > . Matsubara and Matsuda [1] studied the dense
Bose condensate by exploiting this connection and subjecting (4) to a
rotation:
- • HHeis = e2 JHHeiSe 2
J. (11)

The parameter 9 is chosen so that Hjjeis leads to a zero value of < erf > .
Then, by keeping only those terms of Hfjeis quadratic in Cj and cj,
they obtained an approximate spectrum of excitations in the presence
of the condensate. This spectrum is phonon-like at low momentum;
the phonons are the Goldstone bosons associated with the breaking of
azimuthal symmetry by setting 9 ^= 0.
We have made (2) the starting point of a systematic expansion by using
the connection between (2) and (4) a second time in reverse; that is, we
replace i/# ds by a new bosonic Hamiltonian H obtained by replacing
c\ by a new boson operator b\ and adding a term G ^ . 6fft,-&,-&,• where
G -> +00 [2]. Because of the 0-rotation, < ft, >= 0; 9, not < ft >, is
the order parameter for Bose condensation. The new Hamiltonian H
contains terms in btbj and ftjft; so that the ground state has nonzero
values of < ft, ft; > and < ftjft; >, etc. These and other quantities
(speed of sound, spectrum, ground state energy) can be calculated from
diagrams as a power series in A. For furthur details, see Ref. [2].

References
[1] T. Matsubara and H. Matsuda, Prog. Theor. Phys. 16, 569 (1956).
[2] R. Friedberg, T.D. Lee and H.C. Ren, Ann. Phys. (N. Y.) 228, 52
(1993).
[3] T.D. Lee and C.N. Yang, Phys. Rev. 105, 1119 (1957); 113, 1165
(1959).
[4] F. Dyson, Phys. Rev. 102, 1217, 1230 (1956).
 34
Universal Behaviour within the
Nozieres-Schmitt-Rink Theory
F. Pistolesi and G. C. Strinati
Scuola Normale Superiore
1-56126 Pisa
Italy

Abstract

We show that the natural variable to follow the crossover from Cooper-pair-
based superconductivity to Bose-Einstein condensation within the model of
Nozieres and Schmitt-Rink is the product kp£>, where kp is the Fermi wave
vector and £, is the coherence length for two-electron correlation. In terms
of this product, the results of the model do not depend on the detailed
form of the (separable) pairing potential, and the crossover turns out to be
restricted to the universal region n~l< kp£, < 2n. Experimental estimates
indicate that kpt, « 10 (> 2%) for high-Tc superconductors.

Evolution from weak to strong coupling superconductivity has been

considered by Nozieres and Schmitt-Rink [1] (hereafter referred to as
NSR) following the pioneering work by Leggett [2]. After the discovery
of high- Tc superconductivity, the interest in this problem has grown, and
many papers on this subject have appeared [3]. In the present work, we
show that working within the simplified treatment by NSR, it is already
possible to isolate the essential features of the crossover.
Central to the work of NSR and Leggett is the argument [4] that
the BCS wave function has the Bose-Einstein condensation (BEC) built
in as a limiting case. (See the review by Randeria in this volume.)
NSR study the evolution from BCS to BEC through the increase of the
coupling strength associated to an effective fermionic attractive potential,
and conclude that the evolution is "smooth." Although this result is
appealing from a theoretical point of view, it does not allow for a direct
comparison with the experimental data since the coupling strength of the
effective fermionic attraction is not a quantity that is easily obtained from
experiments. Thus it is not possible to know whether a given physical

569
570 F. Pistolesi and G. C. Strinati

system is either in the BCS limit or in the BEC limit, or in a genuine

intermediate regime.
It would clearly be useful to have a more accessible variable to track.
The new variable should satisfy the following criterion: the evolution
from BCS to BEC should be universal, that is, independent of the details
of the interaction potential and of the single-particle density of states.
In the following, we propose to identify the product kpZ, as the desired
parameter.
We briefly review the NSR model. We introduce the model fermionic
Hamiltonian

H =
k,<7

a
/2,Ta -k+q/2,i a -k' + q/2,l^ +q /2,T
^,k'4+ q /2,T
k,k',q

where ako. is the destruction operator for fermions with wave vector k
and spin <r, ek is a single-particle dispersion relation, and Vk# is an
"effective" fermionic attraction. In its simplest form, ek = k2/2m* — fi,
where m* is an effective particle mass and \i is the chemical potential.
The variational calculation with the ground state BCS wave function

leads to the two coupled equations

+ (1 - 2vk2) Y, VkM(pk> = 0, (3)

where n is the particle density, Q is the quantization volume, and cpk =

2ukvk = Ak/£k with Ak = —J2k VKKuvvkf and Ek = We£ + A£ [5].
2
Provided that vk <C 1 for all k, (3) reduces to the Schrodinger equation
for the relative motion of two particles with equal mass m*, interacting
via Fk,ks and with eigenvalue 2/x. In this limit, complete bosonization of
bound-electron pairs is achieved.
Solution of (3) and (4) is simplified with the NSR choice of a separable
 Universal Behaviour within the Nozieres-Schmitt-Rink Theory 571

pairing potential:

Vk,k> = Vwkwk>, wk = (5)

VI + (k/feo)2
where F ( < 0) and fco are parameters specifying the interaction. With
this potential, Ak = Aovvt and the two-body eigenvalue problem has
(in three dimensions) the eigenvalue 2JLI = — eo = — (/CQ/m*)(G — I) 2 for
G > 1, where G = — VQm*ko/4n > 0, while the bound-state radius has
the asymptotic behavior r 0 ~ fc^G"1/2 for G -» oo [6]. This latter result
should be contrasted with that of NSR, that ro —• fc^"1,which led NSR
to conclude that bosonization can be achieved only in the "dilute" limit.
In agreement with NSR, we find that the solutions (Ao, fi) of (3) and (4)
are smooth functions of G for a given fco, although (Ao, fi) both depend
strongly on fco for given G.
One way of introducing a more physical variable in place of G stems
from the observation that, for strong coupling, the (T=0) coherence
length £ is expected to be quite short. This length £ can be calculated
from the pair correlation function (with opposite spins)

(6)

where ^ ( r ) is the fermion field operator, using the expression

2 = /<frg(r)r2 = EklVkffkl2
Jdrg(r) £ k l<Pk| 2 '

Our definition (7) recovers the Pippard expression £o = (dek/dk)kF/nAkF

in the weak-coupling limit (£ -> ^n/2yfl = 1.1 l^o when G < 1) and
the bound-state radius ro in the strong-coupling limit. The behavior of
£ versus G turns out to be "smooth", although it strongly depends on
the value of the parameter fco of the interaction. We attempt to eliminate
the coupling constant G by replacing it by a dimensionless parameter
containing £. Since fc^1 is the only other independent physical length
scale in the problem, we replace the original pair of variables (G,fco) by
the alternative pair (kf^ko) and study the crossover from BCS to BEC
as a function of kF£, for given fco.
In Fig. 1 the chemical potential fi is plotted versus kF£ for different
values of the reduced density n/k\ (= 10a with a = —5, —4,...,+4).
Positive values of \i have been normalized by the Fermi energy eF
(equal to k\/2m*, by our definition), while negative values of JLL have
been normalized by half of the eigenvalue eo of the two-body problem.
572 F. Pistolesi and G. C. Strinati

1.0

o
A
0.5

00 BEC region BCS-like region .

v

-1.0

10' 10°

Fig. 1. Chemical potential \i versus kF^ (at zero temperature). For the normal-
ization of pL and the meaning of the different curves see the text. The two limiting
curves corresponding to the values 10~5 and 104 of n/kl are marked with arrows.

The two curves shown for the reported extreme values of n/k$ m a v ^>e
regarded as limiting curves for all practical purposes. The remarkable
feature of Fig. 1 is that, when expressed in terms of kf£, the behavior
of the chemical potential is universal (i.e., independent of the parameter
fe0 of the interaction potential), but possibly for an "intermediate" range
rc"1- kF£ i 2n [7]. This universal behavior of \i versus kF£ suggests that
kF£ is the appropriate variable to follow the evolution from BCS to BEC.
Note also from Fig. 1 that \i is pinned to (about) the normal-state
value eF when kpt; > 2n, and that \i drops rather abruptly from eF at
kr€ ^ 2TT. Fig. 1 thus shows that, when the coherence length £ equals
the Fermi wavelength X? = 2n/kp, the system becomes unstable against
bosonization and the Fermi surface disappears. We expect that the
instability of the Cooper-pair-based superconductivity when kF£ ~ In
should persist beyond the limits of validity of the procedure we have
adopted to establish it. The stability criterion kF£ > In could then
be regarded as the analog of the Ioffe-Regel criterion for transport in
disordered systems.
One of us (F.R) gratefully acknowledges partial research support from
Europa Metalli-LMI S.p.A.
 Universal Behaviour within the Nozieres-Schmitt-Rink Theory 573
References
[1] P. Nozieres and S. Schmitt-Rink, J. Low. Temp. Phys. 59, 195 (1985).
[2] A.J. Leggett, in Modern Trends in the Theory of Condensed Matter,
A. Pekalski and J. Przstawa, eds. (Springer-Verlag, Berlin, 1980), p.
13.
[3] S. Schmitt-Rink, C M . Varma, and A.E. Ruckenstein, Phys. Rev.
Lett. 63, 445 (1989); R. Friedberg and T.D. Lee, Phys. Rev. B 40,
6745 (1989); M. Randeria, J. Duan, and L. Shieh, Phys. Rev. Lett.
62, 981 (1989) and Phys. Rev. B 41, 327 (1990); L. Belkhir and M.
Randeria, Phys. Rev. B 45, 5087 (1992); M. Randeria, N. Trivedi,
A. Moreo and R.T. Scalettar, Phys. Rev. Lett. 69, 2001 (1992); R.
Cote and A. Griffin, Phys. Rev. B 48, 10404 (1993).
[4] Cf., e.g., J.R. Schrieffer, Theory of Superconductivity (Benjamin, New
York, 1964), Chapter 2.
[5] We have eliminated the Hartree-Fock-like terms by setting Fk,k = 0
for the diagonal components. This choice will not invalidate our
results, since these terms turn out to be irrelevant in the parameter
region of interest.
[6] NSR (Ref. [1]) state that r 0 ~ fcj^1 for G -> oo. Since bosonization
can be achieved only when r$ <C rT1, NSR are able to follow the
evolution from BCS to BEC as a function of G only in the "dilute
limit" n/feg <C 1 for the reduced (three-dimensional) density, that is,
for given density n only when fco > kp. This limitation prevented
NSR from connecting the two limits (BCS and BEC) irrespective of
fco. Our conclusion that ro ~ fcj^G"1/2, on the other hand, enables
us to satisfy the bosonization condition (W//CQ)/G 3 / 2 <C 1 even in the
"dense limit" n/k$ > 1, provided G is large enough.
[7] We have verified that the universal behavior of Fig. 1 for kf£ > 2n
and kft; < TT"1 is independent of the choice of the single-particle
dispersion relation 6k and of the shape of the interaction potential
Wk (barring pathological cases).
 35
Bound States and Superfluidity in
Strongly Coupled Fermion Systems
G. Ropke
Fachbereich Physik
Universitat Rostock
Universitdtsplatz 3
D 18051 Rostock
Germany

Abstract

The two-particle spectrum in a dense Fermion system is treated using a

thermodynamic Green function approach. A self-consistent description of
possible bound states and a superfluid condensate in a correlated medium
is given.

A detailed understanding of superfluidity and superconductivity in corre-

lated Fermion systems, especially the transition from the Cooper-paired
state (weak-coupling limit) to the Bose condensed state of tightly bound
pairs of Fermions (strong coupling limit) [1], is of great interest for very
different physical systems. The problem of a unified treatment of Bose-
Einstein condensation (BEC) and the Bardeen-Cooper-Schrieffer (BCS)
phase arises not only in describing the electron structure of strongly
correlated electron superfluids such as superconductors [2], the electron-
hole system in semiconductors [3], spin-polarized hydrogen and liquid
3
He [4], but also in the theory of nuclear matter [5] and quark-gluon
systems [6].
Recently, there have been several new approaches to this stimulating
problem. A Monte-Carlo simulation of a finite He system at zero
temperature has been performed in Ref. [7]. The microscopic theory
of strongly coupled quantum fluids has been treated within the Jastrow
approximation (cf. Ref. [8]) to obtain the ground state and low-lying
excited states of strongly correlated boson quantum fluids. The crossover
from weak to strong coupling superconductivity has been considered
using a functional integral representation [2] (see also Ref. [7]). The
critical temperature for the onset of a superfluid state has been obtained
within the Matsubara Green function approach neglecting the interaction

574
 Bound States and Superfluidity in Fermion Systems 575
between pairs [9] (see also Ref. [10] for pairing in two dimensions). A
finite temperature Green function approach describing both two-particle
(Brueckner ladders) and pairing correlations has been considered in
Ref. [11], but a self-consistent solution of the gap equation has not been
obtained so far (see Zimmermann [3]).
Clearly one would like an improvement of the Hartree-Fock-Gorkov
mean-field description. As is well known, the Gorkov equation corre-
sponds to a Hartree-Fock approximation in the normal state. A thermo-
dynamic Green's function approach has been developed for the normal
state which is able to describe the formation of bound states and their
dissolution at higher densities (Mott effect, see Ref. [9]). We will give an
approach which includes both correlations and bound state formation in
the superfluid state. In this way, the equation of state is obtained for the
normal state and the superfluid state in a consistent way. Furthermore,
a unified description of BEC and BCS can be given, with a generalized
Gorkov equation.
The fermion system is described by the Hamiltonian

H = ^£(l)a+fli + \ ] T V(m'2')a+a+a2,ar (1)

with 1 = {p\(7i} denoting both momentum and internal quantum num-

bers such as spin. Due to the interaction V(pcc,kP;p'(xf,kfpf), which is
assumed to be proportional to SarfSpp'Sp+kj+k', correlations and bound
states can arise. We restrict ourselves in this paper to two-particle cor-
relations. We note, however, that the generalization to higher order
correlations is straightforward.
In general, the many-particle system described by (1) should be treated
within a nonequilibrium approach. To derive kinetic equations for the
reduced density matrices, we follow the scheme cited in Ref. [12] based
on the Zubarev approach [13] to the nonequilibrium statistical operator.
In Ref. [12], a Boltzmann-type equation is derived in the normal state
for the single-particle diagonal distribution function
y ). (2)
Here, we also include the time evolution of the pair amplitude
F(12;t) and the two-particle distribution function
c(12,1'2', t\ (3)
where the index ex denotes antisymmetrization. For simplicity, we restrict
ourselves to the special case of a homogeneous system (see Eq. (2)).
576 G. Ropke
Furthermore, we consider only singlet pairing with total momentum 2q,
and thus
F(12, t) = F(p, 0^'ap(r)x(12)^1+P2,2,, (4)
with #(12) denoting the spin function, and p the relative momentum.
We introduce new operators b,b+ which are defined by the time-
dependent Bogoliubov-Valatin transformation

a
q-pi =

where

21"/
with

p
l
The anomalous (or off-diagonal) mean values {bibi)1 vanish.
Expressing H = H — J2i ^ i a ^ i i n the ^-representation, we have

- (MF) + # off , (7)

where the mean-field (MF) contributions are contained in h(l). Ad-

ditional "off-diagonal" contributions involving b,b+ such as 62^1 are
collected in // off . In the ^-representation, the modified interaction is (for
example)

= V{q + pi \9q - p2 I;q + p\ \,q - p'2

><l+P2\>(l+ Pi t p ^ i
. (8)
Similar expressions are obtained for other spin combinations. In terms
of these new operators, in place of the original distribution function
given in (2)-(4), we can use the mean values < b\b\ >l, < b^b^b^br >l
to describe the system. We will not derive the kinetic equations for
the corresponding quantities N(l,t), 5(12,0, C(12,l/2/,t) (see Ref. [12]),
but will only discuss these expressions for thermal equilibrium. A sys-
tematic perturbative treatment can be given of the statistical operator
 Bound States and Superfluidity in Fermion Systems 577
Z o * exp[—PH] with the help of the usual Matsubara Green function
technique. As pointed out above, we assume that only single-particle
and two-particle states are of relevance. In the cluster-Hartree-Fock
approximation (i.e., interactions involving the single-particle and two-
particle states in the medium are taken in the Born approximation), we
obtain the following diagrammatic representation for the single-particle
and two-particle Green functions (compare Ref. [14] for the normal state
case):

IiL

ex

ex

(9)

In (9), a line with an arrow signifies the mean-field single-particle Green

function, while the broken line denotes the effective interaction W, such
as given by (8). One can show that it is consistent to neglect terms arising
from Hoff if we work in the Born approximation.
The solution of the Bethe-Goldstone equation (9) can be given via a
spectral representation of the single-frequency, two-particle Green func-
tion, where the two-particle states cpnp (12) and eigenvalues Enp follow
from the Schrodinger-like wave equation
[h(l) + h(2)-EnP]cpnP(12) (10)
l, 1'2')(1 - N(l) - N(2)) + AiJ2(12,1'2')] = 0.

In addition to the mean-field contributions in h(l) and the Pauli blocking

terms, additional "in-medium" effects are described by AH2. For simplic-
ity, we consider only two-particle correlations in the singlet spin channel,
578 G. Ropke
so that (pnp(12) = (pn(p)x(12)dPl+P2f. Evaluating the diagrams in (9), we
find in the low-density limit

AH2(12,1'2') = J ^ ^(1234, 1'2'3'4')C(3'4', 34), (11)

343'4'

with

4, 1'2'3'4') = W(13, lf3%xd4,2d42,

2 ' ) ^ 3 i ' - W(U93'4')52>2S3i> + (1 <-> 2,3 «- 4), (12)
C(34,3'4') = ^
nP

where (5np(12) = (5Pl+P2,p^n,p, and g(E) = [exp(/?E) — I ] " 1 is the Bose

distribution function. Having the two-particle Green function (or T-
matrix T2) at our disposal, the single-particle density can be obtained via
the self-energy according to (9), as discussed in Ref. [9].
Before evaluating the pair amplitude F(12, t) (see Eq. (4)), we briefly
discuss the general nonequilibrium case (compare Ref. [12] for the case
of the normal state). A relevant statistical operator exp(—S(t)) can be
introduced in such a fashion that the mean values (2)-(4) are correctly
reproduced at any value of t. After a cluster decomposition of S(t) and
a perturbative expansion, the approximation characterized by the class
of diagrams (9) can be used to evaluate these mean values. In particular,
we obtain for the pair amplitude F(12, t) the kinetic equation
o

at
= lTn eJJ[
—00

s + i t n t / h
^ % " ^ " (13)

To find the stationary solution F(12, t) ~ exp(—icot), we can perform a

perturbative expansion with respect to H' = H — S/p. In zeroth order,
the t! integral can be immediately performed and yields a nonvanishing
result in the limit e —> 0 only if the condition hco = ji\ + [i2 is fulfilled.
We find

l'2', t)
VI!

- ^2 ^ ( 1 3 , lr3')C(13, Yl') - 7(13,1'3')C*(13,1'3')]

1'33'
+(1 ^ 2) = 0, (14)
 Bound States and Superfluidity in Fermion Systems 579
where we have defined

= V(q + k T, q - P i; q + k' | , q - pf l)vku*pUp>uk>

+V(q + k t , q - kf j ; q + p ] , q - pf l)vkvpUj/V*k,
(15)

7(13, 1'3') follows from 7(13,1'3'), with k T<-> -k |, p |<-> -/? |, and

In this manner, together with the treatment of the two-particle prop-

agator (9), a self-consistent description of the stationary state for a
superfluid system with two-particle correlations is obtained from (10)
and (14). At a given temperature T, the remaining parameters \i\ and q
are fixed by the particle and current densities. In the trivial case C = 0,
the Gorkov equation is obtained from (14), and the BCS theory results
in the weak-coupling limit. In the strong-coupling limit, we find Bose
condensation of bound states. In the limit of low temperatures and low
densities, the gap equation (14) only has solutions if fi\ + \ii is equal to
the bound state energy.
The main theoretical advance given here is to take into account the
influence of "in-medium" correlations and bound states, as described by
the gap equation (14) and the modification of the two-particle properties
(10) in the superfluid state. In contrast to the Gorkov approach, which
corresponds to a mean-field theory, in the normal state (F(12, i) = 0) the
equation of state of a correlated many-particle system [9] is obtained
where the two-particle states are described according to (10).
To illustrate the modification of the two-particle spectrum in a simple
exploratory calculation, we consider the one-dimensional case, where
only a short-range <5-like interaction is assumed to act between opposite
spins:

The parameter s is the binding energy of the isolated two-particle

system. The two-particle energies are shown in Fig. 1 as a function
of the total momentum P at given values of T, N/Lo, q. The shifts of
bound state energies Eb(P) and the continum edge of scattering states
ESC(P) = 2E\(P/2) are determined in this case by a Hartree term, which
is constant, and the Pauli blocking term. The Pauli blocking term, which
is large for two-particle states with total momentum P near 2q, leads to a
deformation of the two-particle spectrum. It should be noted out that the
580 G. Ropke

E/€

Fig. 1. Two-particle energy spectrum in a strongly correlated superfluid system

as a function of the total momentum P: Eb(P) is the energy of the bound state;
E^iP) is the energy of the continuum edge of scattering states; Econd(2q) = 2\i
is the energy of the superfluid condensate. Parameter values: temperature
p~l = 0.1s, density N/Qo = 0.5(ms)l/2/h, q = 02(ms)l/2/h; e is the binding
energy of the isolated two-particle system.

two-particle energies lie well above the energy of the Bose condensate at
Econd(2q) = 2JU. Furthermore, the two-particle binding energy (difference
between the continuum edge of scattering states and the bound state
energy) depends on the total momentum and decreases with increasing
density. The Mott effect arises when the bound state energies merge with
the continuum of scattering states. This illustrates that our approach
to strongly coupled superfluidity corresponds, in the normal phase, not
only to a Hartree-Fock approximation, but also to a self-consistent
description of two-particle correlations [9], In particular, our present
approach includes the formation of bound states in a correlated medium
and their disappearance (or unbinding) at high densities. The Pauli
blocking is responsible not only for the dissolution of the bound states,
but it also arises as the essential mechanism to establish the superfluid
condensate in the strong-coupling limit.
An interesting result of our approach is the description of the transition
 Bound States and Superfluidity in Fermion Systems 581

/T\.
1
3.0 - BCS — 1
BEC 1

\
2.5 • 85ft.:." 1
1
2.0
T/e ; /

1.5

1.0
0.5

no
-2.0 -1.5 -1.0 -0.5 0.5 1.0 1.5

Fig. 2. Phase transition region to the superfluid state for a Fermionic system
with separable interaction (17), y = 6.25. The solution of the gap equation
(14) including correlations in the medium is compared with the solution of the
Gorkov equation for vanishing gap (BCS) and the Bose-Einstein condensation of
noninteracting bound states (BEC). Bound states are "blocked out" for densities
above the Mott line.

from the normal state to the superfluid state. We present a calculation

for a simple separable interaction of the Yamaguchi type

xg(Pi - P2) g(pi - P2) S*,-p (17)

where g(p) = (h2p2/ms + y2)"1. For large values of the parameter 7,

the Hartree energy shifts are small compared with the Fock terms. In
Fig. 2, the transition temperature from the normal state to the superfluid
state is shown for the Gorkov equation (BCS, no correlation in the
medium), for non-interacting bosons (BEC) and for the solution of the
gap equation (14), including self-consistent correlations in the medium. A
value y = 6.25 was choosen corresponding to the interaction in the triplet
(deuteron) channel in nuclear matter. The Mott line shows the densities
above which the bound states are "blocked out". A homogeneous state
has been considered, neglecting first-order phase transitions (see Ref. [9]).
The Gorkov equation overestimates the transition temperature in the
low density region, where bound states are of importance. Because few
bound states are excited, the medium is described in terms of free-
quasiparticle states which are relatively weakly populated. In the high
density region, the bound states are shifted into the continuum of scatter-
ing states and are blocked out (Mott effect). Here, the usual BCS picture
582 G. Ropke
is applicable. In general, the correlations included in the gap equation
(14) leads to a reduction of the region where superfluidity occurs.
In contrast, the BEC scenario of non-interacting bound states leads
to small values of the transition temperature in the low density region,
but overestimates the transition temperature in the higher density region.
The interaction between the bosonic bound states according to the gap
equation (14), especially the Pauli blocking, stabilizes the superfluid phase
in the low density region, but reduces the transition temperature at high
densities.
The present formalism can be extended to describe higher order cor-
relations, such as a-condensation in nuclear matter.

References
[1] P. Nozieres and S. Schmitt-Rink, J. Low Temp. Phys. 59, 159 (1985);
M. Randeria, this volume.
[2] M. Drechsler and W. Zwerger, Ann. Physik 1, 15 (1992); AJ.
Leggett, in Modern Trends in the Theory of Condensed Matter,
A. Pekalski and J. Przystawa, eds. (Springer, Berlin, 1980). M.
Randeria, J.M. Duan, and L.Y. Shieh, Phys. Rev. Lett. 62, 981
(1989); Phys. Rev. B 41, 327 (1990); see also M. Randeria, this
volume.
[3] H. Haug and S. Schmitt-Rink, Progress in Quantum Electronics 9,
3 (1984); R. Zimmermann, Many-Particle Theory of Highly Excited
Semiconductors (Teubner, Leipzig, 1987).
[4] I.F. Silvera, J. Low Temp. Phys. 89, 287 (1992); D. Vollhardt and
P. Wolfle, The Superfluid Phases of Helium-3 (Taylor and Francis,
London, 1990).
[5] R.K. Su, S.D. Yang and T.T.S. Kuo, Phys. Rev. C 35, 1539 (1987);
J.M.C. Chen, J.W. Clark, E. Krotscheck, and R.A. Smith, Nucl.
Phys. A 451, 509 (1986); T. Aim, G. Ropke, and M. Schmidt, Z.
Phys. A 337, 355 (1990).
[6] W. Weise and U. Vogl, Progr. Part. Nucl. Phys. 27, 195 (1991);
S. Klevanski, Rev. Mod. Phys. 64, 649 (1992).
[7] S.A. Chin and E. Krotschek, Phys. Rev. Lett. 65, 2658 (1990); Phys.
Rev. B 33, 3158 (1986); see also D.M. Ceperley and E.L. Pollock,
Phys. Rev. Lett. 56, 351 (1986); Phys. Rev. B 36, 8343 (1987).
[8] C.E. Campbell and B.E. Clements, in Elementary Excitations in
Quantum Fluids, K. Ohbayashi and M. Watanabe, eds. (Springer,
Berlin, 1989).
 Bound States and Superfluidity in Fermion Systems 583
[9] M. Schmidt, G. Ropke, and H. Schulz, Ann. Phys. (N.Y.) 202, 57
(1990).
[10] S. Schmitt-Rink, C M . Varma, and A.E. Ruckenstein, Phys. Rev.
Lett. 63, 445 (1989).
[11] W.H. Dickhoff, Phys. Lett. B 210, 15 (1988); W.H. Voderfecht,
C.C. Gearhart, W.H. Dickhoff, A. Polls and A. Ramos, Phys. Lett.
253, 1 (1991).
[12] G. Ropke and H. Schulz, Nucl. Phys. A 477, 472 (1988).
[13] D.N. Zubarev, Nonequilibrium Statistical Thermodynamics (Plenum,
New York, 1974).
[14] G. Ropke, T. Seifert. H. Stolz, and R. Zimmermann, Phys. Stat. Sol.
(b) 100, 215 (1980).
 36
Onset of Superfluidity in Nuclear Matter
A. Hellmich, G. Ropke, A. Schnell, and H. Stein
Fachbereich Physik and MPG-AG "Theoretische Vielteilchenphysik"
Universitat Rostock
PF 999, D 18051 Rostock
Germany

Abstract
Within the Gorkov approach, the onset of superfluidity in nuclear matter
is considered in the spin-singlet (S=0) and triplet (S=l) channel. In the
triplet channel a transition from Bose-Einstein condensation (BEC) of
deuterons at low densities to a BCS state at high densities is obtained. It
is shown that correlations and bound state formation in the medium should
be included to improve the Gorkov approach. The drastic change in the
composition of the system due to the Mott effect is investigated.

1 Introduction
Nuclear matter is an interesting example of strongly coupled fermions
showing a transition to a superfluid state at densities n ^ no = 0.17 fm~3
and temperatures of the order T ~ 1 MeV. We consider symmetric
nuclear matter containing equal ratios of protons and neutrons. The
nucleon-nucleon interaction is taken in the form of a separable poten-
tial [1]

~~ J(*'), (1)
2S+1
where f^(fe) denotes the form factor in the channel a = Lj for angu-
lar momentum L, tffj are coupling constants .

Calculations are performed for a rank n = 1 potential of the Yam-

aguchi type describing scattering states and a bound state (deuteron) in
the S-channel. Nuclear matter is an example of a system showing Bose-
Einstein condensation (BEC) of bound states (deuterons) at low densities

584
 Onset of Superfluidity in Nuclear Matter 585
and Cooper pairing (BCS) at higher densities. (See also the article by
Randeria in this volume.) Therefore, we expect a smooth transition from
strong coupling in the BEC regime to weak coupling in the BCS regime.
We apply the formalism given by Nozieres and Schmitt-Rink [2] (see
also [3]) to nuclear matter.
In Section 2 the critical temperature is derived using two alternative
approaches, the equation of motion for thermodynamic Green functions
and the T-matrix approach. For the gap and the critical temperature
model, calculations using the Yamaguchi potential are given in Section
3. The equation of state and the composition of nuclear matter, the Mott
effect and the onset of a superfluid phase are discussed in Section 4. For
convenience we put h, c, kg =1.

2 Determination of the Critical Temperature for Superfluidity in

Fermionic Systems
2.1 Gap Equation from BCS Theory
The gap equation can be derived by the real-time finite temperature Green
function method (see [4]). Starting with a Hamiltonian H — H — fiN,

{tj V
\ E I I lmWajamal > (2)

the single-particle retarded Green function Gn(t — t') = ((ai(t),a%(t')))

and the off-diagonal Green function Fn(t — t') = ((a^(t\a^(tf))) are
introduced. Gorkov-type coupled equations for G and F are obtained
from the equations of motion after decoupling in the abnormal pair cut-
off approximation. From the solution for F12, a temperature dependent
gap equation results:

tanh (±Jel,+A(k'A , (3)

with sk = (/c2/2m-/z) + g£fc,(/c,/c' I V I Kk')nk,-Y,k,(Kkr \V\

nk = (fl£atf/c,a) a n d g = (2s + 1)(2T + 1) denoting single-particle spin and
isospin degeneracies.

2.2 T-matrix Approach

Bound and scattering states in nuclear matter are obtained from a non-
relativistic Bethe-Goldstone equation T = V + VG°2T for the T-matrix.
586 A. Hellmich, G. Ropke, A. Schnell, and H. Stein
The two-fermion propagator has the form

where k is the relative and K the total momentum and Ae(k,K) is the
angle averaged quasiparticle energy shift due to the interactions with the
medium. In this approximation, the effects of the medium enter solely
through the phase-space occupation factors /(k, K) and the quasiparticle
energy shift As(k,K).

With a separable interaction (1), the T-matrix equation reduces to an

algebraic one (we consider uncoupled channels L = L') [3]:

Ta(kk'K,z) = 5^M*)[1 ~ UK,* T,z)]^ a ; **rf), (5)

ijk

where

dkk2
A f (6)

The critical temperature Tc for the onset of pairing is given by the

Thouless criterion that a pole at z — fii + ^2 arises for the two-particle
T-matrix at K = 0. For the T-matrix, this pole condition reads

det[l - ReJa(X = 0, T = TC9 z = /n + W )]y = 0. (7)

If the pole corresponds to a two-body bound state, the imaginary part

of J a vanishes. Furthermore, at this particular energy z = fi\ + ^ the
imaginary part of Ja vanishes when it lies in the continuum of scattering
states, because the Pauli blocking factor (1 — f\ — fi) contained in (4)
vanishes in this case. Thus, the pole condition (7) for the onset of
superfluidity holds for positive as well as negative values of pi\ + fi2- It
coincides with the solution of the BCS theory for vanishing gap. This is
most easily demonstrated for the a=15o channel (see Eq. (11) below), but
it holds also in other channels as well as in isospin asymmetric nuclear
matter [5, 6].
 Onset of Superfluidity in Nuclear Matter 587
3 Model Calculation
3.1 The Separable Yamaguchi Nucleon—Nucleon Interaction
In the following calculation, we use the separable Yamaguchi potential
of rank n = 1 [7]:
(k | V | k'> = -Aai;(kMk'), (8)
with the form factor
(9)

The coupling of the 3Si state to the 3D\ state is neglected. The parameters
/?, ks (singlet channel) and Xt (triplet channel) are fitted to the empirical
nucleon-nucleon scattering phase shifts and the deuteron binding energy
El = -2.22 MeV.
With the inverse potential range /?=1.4488 fin"1, empirical values for the
potential parameters 2a are

= 4263 MeV fin"1,

mnL
Xs = 2994 MeV fin"1,
where m = 938.9 MeV denotes the nucleon mass.

This simple rank 1 potential has been choosen to illustrate the many-
particle effects in nuclear matter. In order to obtain the effects of
a repulsive core, more realistic nucleon-nucleon interaction should be
considered. Starting from a general BCS theory for pairing in arbitrary
channels [8] this is done in [6] for the Graz-II potential which also
considers the coupling of the 3S\ - 3D\ channel. The fact that different
realistic potentials yield very similar results has been discussed in Refs. [9,
10].

3.2 Solution of the Gap Equation

For the separable potential (8), the solution of the gap equation (3) reads

(10)

where C contains the summation over k'. Fig. 1 shows the shape of the gap
which is proportional to the form factor of the potential. At the critical
588 A. Hellmich, G. Ropke, A. Schnell, and H. Stein

A[ MeV]

Fig. 1. Gap function A(/c) for triplet and singlet channel at /xeff = 2 MeV and
T = 0.3 MeV.

temperature Tc, the gap vanishes, which corresponds to C(/z, T) = 0. To

determine Tc, we find from (3):

+00

1= (11)

where
(12)

We have introduced the ^-dependent Hartree-Fock (HF) shift of quasi-

particle energy

where f(k) is the Fermi distribution function which contains the shift
self-consistently.

We introduce an effective chemical potential /ieff = V — As(0) so that

the continuum of scattering states starts at /xeff = 0 for k = 0. Thus (12)
reads

As shown in Fig. 2, the critical temperature in the 35i channel is larger

than that in the % channel due to the stronger attraction. In both
channels, the maximum of the critical temperature reduces remarkably if
the Hartree-Fock shift is considered self-consistently (see also [6]).
 Onset of Superfluidity in Nuclear Matter 589

-10 0 10 20 30 40 50 60 70

Fig. 2. Critical temperature in the 3Si and 1So channel without (dashed) and with
(bold) selfconsistent HF shifts.

3.3 On the Relation between Bose Condensation and Cooper Pairing

In the case of ^o pairing, the condition for the critical temperature

(7) yields non trivial solutions only for /ieff > 0. For 3S\ pairing, the
superfluid region shown in Fig. 2 extends down to negative values until
jneS = E®/2 at Tc = 0. This limit represents the formation and Bose
condensation of deuterons.
In order to understand the difference between Bose condensation of
deuteron-like bound states and the pairing phenomenon described by
BCS theory, we discuss the two cases /xeff < 0 and jueff > 0 separately. A
pole in the T-matrix at £& = 2/ieff < 0 defines the critical temperature Tc
for Bose condensation of a bound-state with in-medium binding energy
Eb. In the low-density limit, Tc tends to 0 and ^eff to E$/2. In this
limit, the phase-space occupation and the self-energy effects in (4) can
be neglected and nuclear matter is an ideal gas of structureless bosons
(strong coupling limit).
For /ieff > 0, we find the formation of Cooper pairs according to BCS
theory, which leads to the familiar expression for the critical temperature
in the weak-coupling limit. Of special interest is the region around
/Jeff = 0, where the in-medium bound states with K = 0 disappear due
to the Mott effect [3]. Here the pole of the T-matrix, which defines the
critical temperature, moves continuously from negative to positive energy
values.
The smooth transition between the two limiting cases of Cooper pairing
and Bose condensation has already been discussed in a more general
context [2, 11, 12].
590 A. Hellmich, G. Ropke, A. Schnell, and H. Stein
4 Mott Effect and Composition of Nuclear Matter
We discuss the problem of bound state formation and the composition
in a dense medium. Within a Green function approach, the solution
of the non-relativistic Bethe-Salpeter equation for the two-particle T-
matrix is related to the solution of a Schrodinger-type wave equation
where the interaction term contains the neutron-proton interaction as
well as contributions of the medium,
/ /
2) =0. (14)

The solution gives a shift of the binding energies and a modification of

the wave function. Effects of the medium enter via the HF-shift (13) in
£i,2 and the Pauli blocking factor [1 - fn(K/2 + k) - fp(K/2 - k)], which
accounts for the phase-space occupation (k: relative momentum). Due to
the Pauli blocking, the deuteron binding energy Ej, becomes dependent
on temperature, chemical potential and total momentum K and increases
with increasing density until the bound state disappears (Mott effect) at
the so-called Mott density. Increasing either the temperature or total
momentum shifts the Mott density to higher values. For each density
higher then the Mott density at K = 0, bound states exist only for K
above a minimum total momentum KmoU. The question of effective two-
body scattering in a medium in a T-matrix formalism has been treated
in Refs. [13, 14].
We consider the composition of the nuclear matter system to obtain an
equation of state for the normal phase which is applicable until the crit-
ical temperature given by (11) is reached. The importance of correlated
pairs in this temperature and density region is investigated by splitting
the total density in a contribution due to free nucleons, a deuteron den-
sity and a contribution due to scattering states:

) + 2ncon(fl, T)

with

Wcorr(jK, T) = ndQUt(fl, T) + n scat (//, T).

The different contributions are derived from a Green function ap-

proach [3] so that a generalized Beth-Uhlenbeck formula is obtained,
where
 Onset of Superfluidity in Nuclear Matter 591

0.8

g 0.6

Fig. 3. Correlated pair abundancies showing dependence on total density wtot

using the classical (dashed) and generalized (bold) Beth-Uhlenbeck formula at
two temperatures.

"free = 4

= 3

r
*scat = -3

J i""*

-3 E)

with a degeneracy factor 3 for the triplet and singlet channels, / is the
Fermi, g is the Bose distribution function, £COnt = K2/4m + 2As(K,k = 0)
is the continuum edge and Sa(E9K9fi,T) are the generalized scattering
phase shifts in the 3S\ and ^o channels. Binding energies Eb are calculated
according to (14). Here the quasiparticle shifts are considered only in a
rigid shift approximation (effective mass m* = m), which represents the
same level of approximation as in Ref. [2] or [3].
As expected from the discussion of the binding energy the Mott effect
already leads, for lower temperatures, to a strong suppression of the
correlated pair density at low densities. Fig. 3 compares this with a
classical Beth-Uhlenbeck calculation (nucleons and deuterons considered
as ideal Boltzmann particles), where the contribution of correlated density
is monotonically increasing with total density. Because the Green's
592 A. Hellmich, G. Ropke, A. Schnell, and H. Stein

Fig. 4. Temperature-density plane of nuclear matter showing lines of equal

concentration of correlated pairs 2ncon/ntot (dashed). For comparison the critical
temperatures in the 35i channel (bold) and the Bose-Einstein condensation curve
(long dashed) are plotted.

function treatment was performed in the normal state, the composition

for T = 5 MeV is shown only for densities ntot < 0.06 fm~3.
In Fig. 4 lines of equal concentration of correlated pair density are
plotted in the temperature-density plane outside the superfluid region
according to (15). Due to the break up of deuterons (Mott effect)
and medium-dependent scattering, the isolines of composition have a
maximum. The critical temperature in the 3Si channel from (11) is
plotted versus total density (compare with the upper curve in Fig. 2).
It shows clearly the transition from the Bose-Einstein condensation of
ideal bosons (strong-coupling limit) in the low-density, low-temperature
region to the BCS pairing (weak-coupling limit) in a fermionic medium
at higher densities. The negative slope of the critical line in a small range
of densities around 0.02 fm~3 reflects the lack of a unique solution for
Mntot)- ^n ^ a t r e gi°n bound states disappear (the Mott effect), which
may lead to a phase instability.

5 Conclusion
It has been shown that the solutions of both the Gorkov equation and
the T-matrix equation coincide at the critical temperature for the onset
of superfluidity. In the low-density limit the latter is identical with the
Schrodinger equation for bound states so that we obtain the correct
description in the low-density limit (BEC) as well as in the high-density
 Onset of Superfluidity in Nuclear Matter 593
limit (BCS). However, the Gorkov gap equation is not correct in the
entire temperature-density plane because it uses a HF type decoupling
procedure for the higher Green functions, equivalent to the simple ladder
summation with account taken of H F single particle Green functions in
the T-matrix approach. This approximation means all correlations in the
medium are neglected. However, in a system such as nuclear matter, we
have strongly interacting fermions which are correlated. In particular,
they can form bound states (deuterons). Therefore, a H F type approach
is not always justified.
Thus, we use a generalized Beth-Uhlenbeck formula as an equation
of state for the normal phase to calculate the composition of the system.
It has been found that bound states are of major importance in the
region of low temperatures and low densities. With increasing density,
bound states disappear due to the Pauli blocking (Mott effect) so that
at high densities compared with the Mott density, the approximation of
uncorrelated quasiparticles is possible.
To consider bound states in the medium, the Gorkov equation needs
to be improved. A possible approach with Green functions could be a
cluster Hartree-Fock calculation as performed in [14]. The correlated
medium enters via effective phase-space occupation and cluster-Hartree-
Fock shifts. In the region considered, a liquid-gas phase transition
occurs, as shown in [14], which is important for the investigation of
phase instabilities. For more realistic calculations of nuclear matter,
the Yamaguchi potential has to be replaced by a more sophisticated
nucleon-nucleon interaction. Further investigations should also include
larger clusters such as tritons and a-particles.
The authors thank T. Aim for his valuable suggestions and support of
this work.

References
[1] J. Haidenbauer and W. Plessas, Phys. Rev. C 30, 1822 (1984).
[2] P. Nozieres and S. Schmitt-Rink, J. Low Temp. Phys. 59, 195 (1985).
[3] M. Schmidt, G. Ropke, and H. Schulz, Ann. Phys. 202, 57 (1990).
[4] R.K. Su, S.D. Yang, and T.T.S. Kuo, Phys. Rev. C 35, 1539 (1987).
[5] T. Aim, GSI-93-07 Report (dissertation), (1993).
[6] T. Aim, B. Friman, G. Ropke, and H. Schulz, Nucl. Phys. A 551, 45
(1993).
[7] Y. Yamaguchi, Phys. Rev. 95, 1628 (1954).
[8] R. Tamagaki, Prog. Theor. Phys. 44, 905 (1970).
594 A. Hellmich, G. Ropke, A. Schnell, and H. Stein
[9] M. Baldo, J. Cugnon, A. Lejeune, and U. Lombardo, Nucl. Phys. A
515, 409 (1990).
[10] M. Baldo, I. Bombaci, and U. Lombardo, Phys. Lett. B 283, 8
(1992).
[11] M. Randeria, J.M. Duan, and L.Y. Shieh, Phys. Rev. Lett. 62, 981
(1989).
[12] M. Randeria, this volume.
[13] R.F. Bishop, H.B. Ghassib, and M.R. Strayer, Phys. Rev. A 13, 1570
(1976).
[14] G. Ropke, M. Schmidt, L. Miinchow, and H. Schulz, Nucl. Phys. A
379, 536 (1982) and A 399, 587 (1983).
 Appendix. BEC 93 Participant List
International Workshop on Bose-Einstein Condensation
(Levico Terme, 31 May - 4 June, 1993)

Claudio Albanese, ETH, Zurich, Switzerland

A.S. Alexandrov, University of Cambridge, UK
Gordon Baym, University of Illinois at Urbana-Champaign, USA
Aleksandar Belie, ICTP, Italy
David Brink, Universita di Trento, Italy
G.E. Brown, SUNY at Stony Brook, USA
Christoph Bruder, Universitat Karlsruhe, Germany
Herbert Capellmann, RWTH Aachen, Germany
Yvan Castin, Laboratoire de Physique de TENS, France
Eddie Cheng, Universite Paris-Sud, Orsay, France
Maria Luisa Chiofalo, Scuola Normale Superiore, Pisa, Italy
Eric Cornell, University of Colorado, USA
Franco Dalfovo, Universita di Trento, Italy
Nguyen Dinh Dang, Universita di Catania, Italy
E. Fortin, University of Ottawa, Canada
Frank Y. Fradin, Argonne National Laboratory, USA
Richard Friedberg, Columbia University, USA
Stefano Giorgini, Universita di Trento, Italy
S. Giovanazzi, Universita di Trento, Italy
Tom Greytak, MIT, USA
Allan Griffin, University of Toronto, Canada
Mashahiro Hasuo, University of Tokyo, Japan
H. Haug, Universitat Frankfurt am Main, Germany
Anke Hellmich, Universitat Rostock, Germany
Kristian Helmerson, NIST (Gaithersburg), USA
T.W. Hijmans, University of Amsterdam, The Netherlands
Kerson Huang, MIT, USA
F. Iachello, Yale University, USA
A. Ivanov, Universitat Frankfurt am Main, Germany
Simo Jaakkola, University of Turku, Finland
Y.M. Kagan, Kurchatov Institute, Russia
G. Kavoulakis, University of Illinois at Urbana-Champaign, USA
L.V. Keldysh, Lebedev Physical Institute, Russia

595
596 Appendix. BEC 93 Participant List
Franck Laloe, Laboratoire de Physique de TENS, France
Andrea Lastri, Universita di Trento, Italy
Anthony Leggett, University of Illinois at Urbana-Champaign, USA
R. Leonardi, Universita di Trento, Italy
Jia Ling Lin, University of Illinois at Urbana-Champaign, USA
Bennett Link, Los Alamos National Laboratory USA
Akira Matsubara, Kyoto University, Japan
S.A. Moskalenko, Academy of Sciences of Moldova, Moldova
A. Mysyrowicz, Ecole Polytechnique, France
Nobukata Nagasawa, University of Tokyo, Japan
Fabio Pistolesi, Scuola Normale Superiore, Pisa, Italy
Ludovic Pricavpenko, Universite Paris-Sud, Orsay, France
Mohit Randeria, Argonne National Laboratory, USA
Julius Ranninger, CNRS-CRTBT, Grenoble, France
John D. Reppy, Cornell University, USA
Meritt Reynolds, University of Amsterdam, The Netherlands
Mannque Rho, Saclay, France
Gerd Ropke, Universitat Rostock, Germany
G.V. Shlyapnikov, Kurchatov Institute, Russia
L.D. Shvartsman, The Hebrew University of Jerusalem, Israel
Yu.M. Shvera, Academy of Sciences of Moldova, Moldova
Isaac Silvera, Harvard University, USA
David Snoke, The Aerospace Corporation, Los Angeles, USA
Paul E. Sokol, Penn. State University, USA
Holger Stein, Universitat Rostock, Germany
H.T.C. Stoof, Eindhoven University of Technology, The Netherlands
Sandro Stringari, Universita di Trento, Italy
Eric Svensson, AECL Research, Canada
E. Tiesinga, Eindhoven University of Technology, The Netherlands
Sergei Tikhodeev, General Physics Institute, Russia
Jacques Treiner, Universite Paris-Sud, Orsay, France
D. van der Marel, University of Groningen, The Netherlands
Silvio A. Vitiello, Universita degli Studi di Milano, Italy
Jook T.M. Walraven, University of Amsterdam, The Netherlands
Tilo Wettig, S.U.N.Y. at Stony Brook, USA
Jim Wolfe, University of Illinois at Urbana-Champaign, USA
 Index

abnormal distribution function, 514-515 of biexcitons, see biexcitons,

AlSb, 534 condensation of
Anderson localization, 26 of bipolarons, see bipolarons,
Anderson mode, 380, 388 condensation of
anomalous Green's functions, 264, 407 of deuterons, 584
anomalous self energy, 266, 271 of excitons, see excitons, condensation of
antivortices, 35 of finite systems, 9, 418-422, 436, 574
atomic cesium, 166, 175, 177, 179, 188, 190, of 4 He, see 4 He, superfluid
195, 226, 227, 465-470 of hydrogen, see spin-polarized
Cs-Cs interaction, 467 hydrogen, condensation of
hyperfine splitting, 466-470 of ideal gas, 15, 17, 31, 39, 41, 53-55, 89,
s-wave scattering length, 469 100, 101, 117, 168, 203-204,
atomic fountain, 226 234-236, 294, 298, 487
attractive Fermi gas, 359-367, 373-376, 569 of kaons, 3, 438^50
of nucleons, 429^32
of pairs, 122-126, 335, 363-367, 507
BCS theory, 3, 5, 24, 126, 243, 249, 261, of pions, 3
334, 355-388, 394-395, 409, 422, 453, of polaritons, see polaritons,
585, 586 condensation of
and BEC, see BEC to BCS crossover of positronium, see positronium,
BEC condensation of
compressibility, 40, 89 of quarks, 3, 574
deformed, 430, 434 of system with finite lifetime, 7, 202, 228,
direct evidence for ("smoking gun"), 4, 6, 241, 298, 336, 524-530
52, 53, 68, 81, 320, 444 of weakly interacting gas, see weakly
ferromagnetism analogy, 127, 229-231, interacting Bose gas
234, 452-455, 461 order parameter, 33, 34, 205, 232-236,
in a potential minimum, 6, 15, 147-153, 241, 453, 455, 568
168, 203, 223-224, 311, 320-325, sum rules, 86-97, 543, 553, 556
472-474 timescale for onset of, 7, 9, 149, 202-224,
in boson-fermion mixture, 396, 401-407, 226-243, 298, 310, 336, 497, 527
434^36 to BCS crossover, 5, 125-126, 334,
in one dimension, 87-90 355-388, 569-572, 574-582, 585, 589
in random potential, 10, 26, 42-48, 93 to laser crossover, 9, 197, 255-257
in two dimensions, 15, 34, 46, 87-90, 163, Belyaev method, 103
203, 270, 411 Beth-Uhlenbeck formula, 99, 116, 365, 590,
infrared divergence, 55, 61, 62, 64, 76, 591, 593
88-90, 117-118, 127, 270, 294 biexcitons, 7, 202, 247, 249-252, 262-277,
of atomic gas, 100, 188-190, 197, 227, 287, 326, 332-334, 358, 487, 496-505,
465^70 507-511

597
598 Index
biexciton-biexciton interaction, 335-336, charged bosons, 377, 380-381, 395, 411,
504 532, 544-547
binding energy, 333 chiral symmetry, 3, 439
condensation of, 265, 337, 340-346, coherence length, 18, 357, 371, 395, 546
487^96, 505, 507, 510 coherent pairing, 507, 510-511
creation, 333-334, 345 coherent potential approximation, 367
Lamb shift, 497-504 collisions, 204, 208, 214, 215, 228
luminescence, 337-344, 488, 511 atoms, 103, 190, 465^70
momentum distribution, 338 biexcitons, 249
s-wave scattering length, 249, 270, 494 excitons, 295, 296, 300, 308, 312, 318, 513
4
single-photon absorption, 333 He, 66
two-photon absorption, 265, 333-334, positronium, 561-564
339,489,496,508,511 spin-polarized hydrogen, 134-136,
two-photon recombination, 342 139-140, 143, 145, 147, 148, 161,
biholes, 532-539 227, 228
Bil 3 , 533 composite bosons, 1, 4, 24-25, 247, 253,
binary collision approximation, 101 283, 355, 356, 358, 371, 387, 423^25,
bipolarons, 22, 358, 395^14, 541-547 465, 532
condensation of, 395, 397, 400, 407-413, condensate
542-547 amplitude fluctuations, 20, 205, 210, 243
bismuth films, 28 in momentum distribution, 55, 58, 61, 68,
bismuth oxide superconductors, 397 100, 206, 220, 234, 294, 302
black holes, 444-450 phase coherence, 21, 22, 25, 32-35
Bogoliubov approximation, 21, 44, 101, phase fluctuations, 8, 34, 35, 46, 203, 205,
335, 379 217, 220, 243
Bogoliubov dispersion law, 92, 209, 220, wavefunction, 32, 33, 205, 234, 456
243, 374, 544 Cooper pairs, 5, 249, 258, 334, 355-358,
Bogoliubov inequality, 87, 89, 91 370, 373, 375, 387, 388, 394, 395, 421,
Bogoliubov sum rule, 87 453, 455^58, 507, 532, 572, 585
Bogoliubov transformation, 40, 466, 510, in disordered alloys, 27
correlation length, 163, 205
576
CP violation, 226
Bohr solution of hydrogen atom, 282,
critical exponents, 4, 46, 48, 79
331-332
Cu 2 O, 6, 21, 253, 281-326, 346-353, 439,
Boltzmann equation, 202, 204, 206-208,
442, 519-523, 525, 526
227, 228, 236, 239, 277, 544, 575
band structure constants, 283
Bose gas, see weakly interacting Bose gas,
exciton lifetime, 286, 288, 346-347
hard sphere Bose gas
excitonic relaxation processes, 300
Bose glass, 26, 27, 43, 46
optical phonons, 286
Bose liquid, 18, 22, 58, 160, 265, 358,
CuCl, 7, 336-346, 487-495, 504, 508, 511
551-557 cuprate superconductors, 358, 382, 397,
condensate fraction, 57 414, 542, 547
driven, 264 CuO 2 planes, 397^01, 413, 541
hydrodynamic equations, 221
phonons, 89, 552, 554, 555
DeBroglie wavelength, 53, 105, 137, 145,
Bose narrowing, 6, 293
190, 198, 235, 293, 466
Bose-Einstein distribution, 204, 292, 293,
dilute Bose gas, see weakly interacting
302, 305, 314, 340, 347, 488, 524 Bose gas
bosonization, 9, 418^36, 570, 571 Doppler cooling, 176-177, 179
broken gauge symmetry, 1-3, 7, 21, 32-35,
52, 93, 234, 236, 241, 368, 372, 376,
electron-hole droplets, see electron-hole
377, 387, 439, 452-462, 567
liquid
BSCO, 397, 412, 547
electron-hole exchange, 262, 286, 335, 346
electron-hole insulator, 254
cesium, see atomic cesium electron-hole liquid, 247, 249-252, 254,
charge density wave, 252, 361 287, 335, 358, 528
charge transfer gap, 397-399 electron-hole plasma, 249
 Index 599
electron-phonon coupling, 394-395 Fermi-Dirac distribution, 254, 261
electronic phase transition, 252 of localized bosons, 8, 546
Eliashberg theory, 395 four-wave mixing, 265, 273-274, 496
entropy loss, 312, 442 phase conjugation, see optical phase
exchange effects, 17, 103-110, 119-122, 425, conjugation
435 fractional quantum Hall effect, 90
excitonic crystal, 249 fullerenes, 397, 414, 541
excitonic gas, 247-255, 263, 281-284, 288,
295-325, 438 GaAs, 500, 502, 508, 534, 535, 537
excitonic insulator, 249, 250, 257 Galitsky method, 103
excitonic molecule, see biexcitons GaP, 534
excitons, 7, 10, 18, 21, 24, 25, 82, 100, 202, GaSb, 534
246-277, 281-326, 330-353, 355, 426, Ge, 250, 253, 293, 323, 525, 526, 528, 534,
439, 442, 487, 519-530 538
Auger process, 296, 300, 308, 311, 315, Gell-Mann-Levy model, 442
325 Ginzburg-Landau theory, 33-35, 442, 527
binding energy (excitonic Rydberg), 248, time-dependent, 238, 368 -372
282-283, 285, 331 Goldstone bosons, 568
Bohr radius, 248, 282, 331 gravitational atomic cavity, 194-197
condensation of, 6, 39, 100, 203, 254, Gross-Pitaevsky equation, see nonlinear
302-305, 307, 312, 316-320, 332, Schrodinger equation
334-353, 357, 377, 507, 510, 520-523 gun, smoking, see BEC, direct evidence for
creation, 253, 265, 284, 295, 296, 332 hard core Bose gas, see hard sphere Bose
diffusion, 289-291, 296, 320, 332, 349, gas
520, 528 hard sphere Bose gas, 23, 38^42, 44, 56, 64,
dumbbell, 8 100, 102, 117-118, 284, 297, 361, 379,
exciton-exciton interaction, 263-264, 553, 565
269, 296, 297, 312, 320, 326, 335, Hartree-Fock approximation, 17-19, 103,
501, 507-510, 513-516, 526 121, 237, 577, 580, 593
exciton-phonon interaction, 288-291, healing length, 148, 149
298-300, 305, 310, 320, 336, 352, 3
He, 18, 485
526, 529 superfluid, 5, 355, 357, 453, 455, 457, 574
in a stress field, 291, 313-325 4
He, 6, 35-38, 42, 51-82, 92, 160, 162, 168,
in two dimensions, 8, 270 248, 254, 286, 355, 478, 550, 551
luminescence, 284^325, 347-349 gas, 250
momentum distribution, 286, 298, in a porous medium, 10, 27, 43, 46
301-320 in two dimensions, 8
ortho, 21, 286-325, 332, 346-349, 507 lambda transition, 42
para, 21, 286-325, 332, 346, 349, 507, momentum distribution, 4, 53, 58-82,
508, 519, 525 286, 551
phonon-assisted spin relaxation, 288, correlated basis function calculation,
300, 302, 314 60
photovoltaic detection, 349-352, 519 Green's function Monte Carlo
polarized, 508, 510 calculation, 60-62, 64, 74, 76, 77,
"quantum saturation", 21, 223, 307-312, 80
314^315, 325, 349 path integral Monte Carlo calculation,
recombination, 9, 252, 253, 284-289, 298 60, 62, 70, 73, 77, 79, 81
s-wave scattering length, 297, 312 shadow wave function calculation, 60
single-photon absorption, 331-332 variational calculation, 60, 61, 64
superfluidity of, 7, 39, 203, 252-253, 305, phase diagram, 56
311, 334-335, 349-353, 520-523, solid, 23, 56, 80
528, 529 superfluid, 1, 9, 135, 142, 161, 221, 229,
two-body spin relaxation, 21, 300, 308 355, 459
virtual, 262 condensate fraction, 62, 64, 68-81, 101
maxons, 96
Fermi gas, see attractive Fermi gas phonons, 35, 36, 60, 61, 63, 64, 96
Fermi liquid, 247, 249, 259, 358-367, 404 rotons, 63, 96
600 Index
Higgs boson, 3 magnetoexcitons, 8
Hohenberg-Mermin-Wagner theorem, maser, 132
87-90 Maxwell-Boltzmann distribution, 54, 288,
holons, 35 289, 298, 299, 302, 314, 338, 347, 488
Hubbard model, 409, 413 mean-field approximation, 23, 34, 100, 101,
attractive, 360-361, 367, 373, 377, 379, 121, 259, 260, 269, 271, 419, 575
381, 387 Mott transition, 26, 27, 575, 581, 590, 592
repulsive, 358, 381
hydrogen, see spin-polarized hydrogen neutrinos, 441, 447
molecules, 142, 332-333, 335 neutron scattering, 4, 53, 62-81, 550-557
under high pressure, 359 final-state effects, 66-68, 71, 73, 74
hyper-Raman scattering, 511 impulse approximation, 63, 64, 66,
550-552, 555, 556
InAs, 534
neutron stars, 439, 444
InP, 534
nonequilibrium Green's functions
InSb, 534
interacting boson model, 418, 426, 427 electron-hole, 257-262
Ising model, 23, 567 exciton, 264-269
nonequilibrium phase transition, 255, 514
Josephson effect, 21 nonlinear optics, 255, 257, 262, 263, 265,
Josephson junction, 454, 455, 458 277
nonlinear Schrodinger equation, 34, 103,
kaons, 438-450 205, 207, 242, 371, 545
kaon-nucleon interaction, 439-440 nuclear ferromagnetism, 19
kaonic atoms, 439 nuclear matter, 418, 574, 582, 584-593
Kosterlitz-Thouless transition, 4, 8, 18, 35, equation of state, 359, 585
38, 163, 203, 368 nuclear spin waves, 104, 132
of 4 He, 8 nucleons, 3, 418-436
of excitons, 8 nucleon-nucleon interaction, 584, 587
of spin-polarized hydrogen, 8, 168-170, nucleosynthesis, 448-449
478^85
off-diagonal long-range order, 1, 4, 8, 33,
Levy flights, 194 57, 59, 206, 220-223, 356, 368, 387
lambda transition, see 4 He, lambda one-body density matrix, 32, 57, 60, 94,
transition 100, 104, 109, 122
Landau-Ginzburg theory, see Onsager-Feynman quantization, 33
Ginzburg-Landau theory optical lattices, 180, 186-187
laser, 4, 9, 255
optical molasses, 176, 179, 185-188, 194
and BEC, see BEC to laser crossover
optical phase conjugation, 344-346,
excitonic, 255-257
488-495
neutrino, 9
laser cooling of atoms, 100, 166, 173-198,
202, 466 Pauli exclusion, 253, 254, 292, 379, 387,
Doppler cooling, 154 425, 579-582, 590
recoil limit, 190 phase locking, see condensate, phase
Sisyphus effect, 179, 182-186, 189 coherence
velocity-selective trapping, 190-194 phase transition
laser generation of excitons, see excitons, first order, 32, 41, 42
creation second order, 34, 41, 79, 229-232, 443
linear response theory, 36, 38 phase-space filling, 262
lithium, 227 phonon wind, 528-530
local density approximation, 152 phonons, 220, 222
long-range phase coherence, see in Bose liquid, see Bose liquid, phonons
off-diagonal long-range order in crystal lattice, see electron-phonon
LSCO, 397 coupling, exciton-phonon
interaction
magneto-optical trap, 166, 177-179, in superfluid helium, see 4 He, superfluid,
188-190 phonons
 Index 601
in weakly interacting Bose gas, see self-consistent field approximation, 266,
weakly interacting Bose gas, 515, 516
phonons self-phase modulation, 265
plasmons, 377, 381, 388, 559 shell model, 419, 426
polariton bottleneck, 256 Si, 250, 253, 293, 323
polaritons, 255-262, 268-269, 336, 338, 342, SiGe, 538
497-504, 513-516 Silvera-Reynolds number, 525
condensation of, 9, 257, 513-516 single-particle density matrix, see one-body
dispersion law, 255, 339 density matrix
polarons, 395, 398 smoking gun, see BEC, direct evidence for
positronium, 7, 283, 285, 287, 300, 331 solitons, 353
condensation of, 558-564 spin density wave, 252
molecule, 333, 559, 564 spin gaps, 358, 382-387, 414, 542
momentum distribution, 563 spin waves, 565-568
s-wave scattering length, 562 spin-polarized hydrogen, 7, 18, 19, 100,
superfluidity of, 564 131-157, 160-171, 202, 203, 226-228,
two-body spin relaxation, 562 241, 295, 442, 453, 525, 574
two-photon recombination, 563 compression, 136
"precondensate", 211 condensation of, 39, 100, 131-156,
pseudogap, 407, 408, 410, 413, 414, 543 160-161, 163-167, 227, 228,
pseudopotential method, 101 231-232, 236, 241
evaporative cooling, 138-147, 154, 163,
QCD theory, 439 165, 227, 236, 475
quantum heterostructures, 8, 533 hyperfine splitting, 133, 161, 227
quantum lattice model, 565 in two dimensions, 162-163, 168-170,
quantum Monte Carlo method, 381, 382 478^85
quantum tunneling, 454 magnetic trapping of, 134-157, 161, 227,
quantum well, 8, 10, 270, 500, 508, 509, 533 472-475
quantum wire, 502 microwave trapping of, 163-167
quasicondensate, 8, 163, 168, 203, 205, relaxation explosion, 151, 164, 165,
208-211, 218, 220, 223, 243, 320, 527 472-475
quasiequilibrium, 213, 253, 267, 295, 300, s-wave scattering length, 145, 148-149,
336 227
"quasipairs", 126 superfluidity of, 39, 132, 160
quasiparticles, 35, 36, 38, 40, 41, 119, 122, three-body recombination, 132, 134, 137,
242, 378, 591 161, 165, 169, 210-211, 472, 479-484
two-body spin relaxation, 135, 140, 143,
Rabi oscillations, 257, 275, 277 147, 162, 165, 300
radiation pressure, 175 spontaneous symmetry breaking, see
random phase approximation, 204-206, broken gauge symmetry
377, 378, 380, 381, 383 stimulated emission, 6
stimulated scattering, 6, 273, 292, 344, 352
s-d boson model, see interacting boson strangeness condensation, 443
model strongly interacting Bose gas, 160
s-wave scattering length, 39, 126, 211, 237, superconductivity, 3, 5, 24, 25, 249, 355,
360, 362, 371 356, 361, 374, 394, 403, 421, 426, 453,
atomic cesium, see atomic cesium, 532, 574
s-wave scattering length collective excitations, 377-381
biexcitons see - biexcitons, s-wave high T c , 5, 35, 357-358, 382, 386-387,
scattering length, 249 394-414, 462, 541-547, 569
excitons, see excitons, s-wave scattering and bipolarons, see bipolarons,
length condensation of
positronium, see positronium, s-wave in disordered system, 10, 358
scattering length Schafroth, 395, 396
spin-polarized hydrogen, see type II, 10
spin-polarized hydrogen, s-wave superfluid density, see two-fluid model,
scattering length superfluid density
602 Index
superfluid gas, 128 biexcitons, see biexcitons, two-photon
superfluidity, 3-5, 8, 21, 23-25, 27, 33, recombination
35-38, 4 2 ^ 8 , 51, 59, 87, 101, 127, 160, excitons, 285
163, 206, 207, 217, 220, 223, 355, 356, positronium, 285, 559
387, 407, 453, 455-462, 574-582,
584-593 ultrafast optics, 298
supernova 1987A, 440, 449, 450 "universal phase standard", 455, 459
supersolids, 23 Ursell operators, 102-103, 110-128
symmetry breaking, see broken gauge
symmetry vortices, 21, 34, 35, 206, 217, 219, 220, 222,
223

thermodynamic limit, 9, 16, 117-118, 228, weakly interacting Bose gas, 5-7, 82, 91-93,
235, 243, 452, 457, 526 95, 100-128, 148-149, 160, 163, 168,
topological long-range order, 206, 217-220, 203, 204, 226, 284, 291-294, 325, 336,
222 371, 372, 380, 438, 524, 526
two-body density matrix, 94-97, 101 blue shift, 320, 494
two-body spin relaxation equation of state, 110-118
of excitons, see excitons, two-body spin far from equilibrium, 203-224, 227-243,
relaxation 514
of hydrogen, see spin-polarized phonons, 206, 218, 220-223, 388
hydrogen, two-body spin relaxation speed of sound, 149, 218, 221, 379
of positronium, see positronium, weakly interacting Fermi gas, 357, 360
two-body spin relaxation Wigner crystal, 380
two-fluid model, 4, 36-38, 221-222
superfluid density, 41, 43, 45-47, 81, 368 YBCO, 382, 386, 387, 397, 399^01, 412,
two-photon absorption 542-545
by biexcitons, see biexcitons, two-photon
absorption ZnP 2 , 533
by hydrogen, 154-157 ZnSe, 508
two-photon recombination ZnTe, 534