Mathematical

Surveys
and
1SRSKVETLW
Volume 228

Hilbert Schemes of
4SMRXWERH-R½RMXI
Dimensional Lie Algebras

Zhenbo Qin
Hilbert Schemes of
Points and Infinite
Dimensional Lie Algebras
Mathematical
Surveys
and
Monographs
Volume 228

Hilbert Schemes of
Points and Infinite
Dimensional Lie Algebras

Zhenbo Qin
 EDITORIAL COMMITTEE
Robert Guralnick Benjamin Sudakov
Michael A. Singer, Chair Constantin Teleman
Michael I. Weinstein
The author was supported in part by a Collaboration Grant for Mathematicians (Award
Number: 268702) from the Simons Foundation and by a Research Council Grant (Award
Number: URC-17-084-n) from the University of Missouri.

2010 Mathematics Subject Classification. Primary 14C05, 17B65;

Secondary 14F43, 14J60, 14N35, 17B69.

For additional information and updates on this book, visit

www.ams.org/bookpages/surv-228

Library of Congress Cataloging-in-Publication Data

Names: Qin, Zhenbo, author.
Title: Hilbert schemes of points and infinite dimensional lie algebras / Zhenbo Qin.
Description: Providence, Rhode Island: American Mathematical Society, [2018] | Series: Mathe-
matical surveys and monographs; volume 228 | Includes bibliographical references and index.
Identifiers: LCCN 2017036491 | ISBN 9781470441883 (alk. paper)
Subjects: LCSH: Hilbert schemes. | Schemes (Algebraic geometry) | Lie algebras. | AMS: Alge-
braic geometry – Cycles and subschemes – Parametrization (Chow and Hilbert schemes). msc
| Nonassociative rings and algebras – Lie algebras and Lie superalgebras – Infinite-dimensional
Lie (super)algebras. msc | Algebraic geometry – (Co)homology theory – Other algebro-
geometric (co)homologies (e.g., intersection, equivariant, Lawson, Deligne (co)homologies).
msc | Algebraic geometry – Surfaces and higher-dimensional varieties – Vector bundles on sur-
faces and higher-dimensional varieties, and their moduli. msc | Algebraic geometry – Projective
and enumerative geometry – Gromov-Witten invariants, quantum cohomology, Gopakumar-
Vafa invariants, Donaldson-Thomas invariants. msc | Nonassociative rings and algebras –
Lie algebras and Lie superalgebras – Vertex operators; vertex operator algebras and related
structures. msc
Classification: LCC QA564 .Q56 2018 | DDC 516.3/5–dc23
LC record available at https://lccn.loc.gov/2017036491

Copying and reprinting. Individual readers of this publication, and nonprofit libraries acting
for them, are permitted to make fair use of the material, such as to copy select pages for use
in teaching or research. Permission is granted to quote brief passages from this publication in
reviews, provided the customary acknowledgment of the source is given.
Republication, systematic copying, or multiple reproduction of any material in this publication
is permitted only under license from the American Mathematical Society. Permissions to reuse
portions of AMS publication content are handled by Copyright Clearance Center’s RightsLink
service. For more information, please visit: http://www.ams.org/rightslink.
Send requests for translation rights and licensed reprints to [email protected].
Excluded from these provisions is material for which the author holds copyright. In such cases,
requests for permission to reuse or reprint material should be addressed directly to the author(s).
Copyright ownership is indicated on the copyright page, or on the lower right-hand corner of the
first page of each article within proceedings volumes.
2018
c by the American Mathematical Society. All rights reserved.
Printed in the United States of America.

∞ The paper used in this book is acid-free and falls within the guidelines
established to ensure permanence and durability.
Visit the AMS home page at http://www.ams.org/
10 9 8 7 6 5 4 3 2 1 23 22 21 20 19 18
 Contents

Preface ix

Part 1. Hilbert schemes of points on surfaces 1

Chapter 1. Basic results on Hilbert schemes of points 3
1.1. Partitions 3
1.2. The ring of symmetric functions 5
1.3. Symmetric products 7
1.4. Hilbert schemes of points 10
1.5. Incidence Hilbert schemes 17
Chapter 2. The nef cone and flip structure of (P2 )[n] 19
2.1. Curves homologous to βn 19
2.2. The nef cone of (P2 )[n] 28
2.3. Curves homologous to β − (n − 1)βn 32
2.4. A flip structure on (P2 )[n] when n ≥ 3 35

Part 2. Hilbert schemes and infinite dimensional Lie algebras 41

Chapter 3. Hilbert schemes and infinite dimensional Lie algebras 43

3.1. Affine Lie algebra action of Nakajima 43
3.2. Heisenberg algebras of Nakajima and Grojnowski 46
3.3. Geometric interpretations of Heisenberg monomial classes 55
3.4. The homology classes of curves in Hilbert schemes 58
3.5. Virasoro algebras of Lehn 61
3.6. Higher order derivatives of Heisenberg operators 63
3.7. The Ext vertex operators of Carlsson and Okounkov 69
Chapter 4. Chern character operators 73
4.1. Chern character operators 73
4.2. Chern characters 80
4.3. Characteristic classes of tautological bundles 87
4.4. W algebras and Hilbert schemes 91

Chapter 5. Multiple q-zeta values and Hilbert schemes 99

5.1. Okounkov’s conjecture 99
5.2. The series Fkα11,...,k
,...,αN
(q) 102
N
 L1 
5.3. The reduced series chk1 · · · chLN
kN 118

Chapter 6. Lie algebras and incidence Hilbert schemes 121

6.1. Heisenberg algebra actions for incidence Hilbert schemes 121
v
vi CONTENTS

6.2. A translation operator for incidence Hilbert schemes 129

6.3. Lie algebras and incidence Hilbert schemes 137

Part 3. Cohomology rings of Hilbert schemes of points 139

Chapter 7. The cohomology rings of Hilbert schemes of points

on surfaces 141
7.1. Two sets of ring generators for the cohomology 141
7.2. The Hilbert ring 146
7.3. Approach of Lehn-Sorger via graded Frobenius algebras 149
7.4. Approach of Costello-Grojnowski via Calogero-Sutherland
operators 154

Chapter 8. Ideals of the cohomology rings of Hilbert schemes 157

8.1. The cohomology ring of the Hilbert scheme (C2 )[n] 157
8.2. Ideals in H ∗ (X [n] ) for a projective surface X 161
8.3. Relation with the cohomology ring of the Hilbert scheme (C2 )[n] 164
8.4. Partial n-independence of structure constants for X projective 166
8.5. Applications to quasi-projective surfaces with the S-property 171

Chapter 9. Integral cohomology of Hilbert schemes 175

9.1. Integral operators 175
9.2. Integral operators involving only divisors in H 2 (X) 180
9.3. Integrality of mλ,α for integral α 184
9.4. Unimodularity 185
9.5. Integral bases for the cohomology of Hilbert schemes 190
9.6. Comparison of two integral bases of H ∗ ((P2 )[n] ; Z) 191
∗
Chapter 10. The ring structure of Horb (X (n) ) 203
10.1. Generalities 203
10.2. The Heisenberg algebra 205
10.3. The cohomology classes ηn (γ) and Ok (α, n) 206
10.4. Interactions between Heisenberg algebra and Ok (γ) 209
∗
10.5. The ring structure of Horb (X (n) ) 212
10.6. The W algebras 214

Part 4. Equivariant cohomology of the Hilbert schemes

of points 217

Chapter 11. Equivariant cohomology of Hilbert schemes 219

11.1. Equivariant cohomology rings of Hilbert schemes 219
11.2. Heisenberg algebras in equivariant setting 224
11.3. Equivariant cohomology and Jack polynomials 225

Chapter 12. Hilbert/Gromov-Witten correspondence 231

12.1. A brief introduction to Gromov-Witten theory 232
12.2. The Hilbert/Gromov-Witten correspondence 233
12.3. The N -point functions and the multi-point trace functions 238
12.4. Equivariant intersection and τ -functions of 2-Toda hierarchies 241
12.5. Numerical aspects of Hilbert/Gromov-Witten correspondence 244
 CONTENTS vii

12.6. Relation to the Hurwitz numbers of P1 247

Part 5. Gromov-Witten theory of the Hilbert schemes

of points 251
Chapter 13. Cosection localization for the Hilbert schemes of points 253
13.1. Cosection localization of Kiem and J. Li 253
13.2. Vanishing of Gromov-Witten invariants when pg (X) > 0 257
13.3. Intersections on some moduli space of genus-1 stable maps 261
13.4. Gromov-Witten invariants of the Hilbert scheme X [2] 265
Chapter 14. Equivariant quantum operator of Okounkov-Pandharipande 271
14.1. Equivariant quantum cohomology of the Hilbert scheme (C2 )[n] 271
14.2. Equivalence of four theories 275
14.3. The quantum differential equation of Hilbert schemes of points 278
Chapter 15. The genus-0 extremal Gromov-Witten invariants 283
15.1. 1-point genus-0 extremal Gromov-Witten invariants 283
15.2. 2-point genus-0 extremal invariants of J. Li and W.-P. Li 294
15.3. The structure of the genus-0 extremal Gromov-Witten invariants 301
Chapter 16. Ruan’s Cohomological Crepant Resolution Conjecture 307
16.1. The quantum corrected cohomology ring Hρ∗n (X [n] ) 308
16.2. The commutator [G̃k (α), a−1 (β)] 310
16.3. Ruan’s Cohomological Crepant Resolution Conjecture 322
Bibliography 325
Index 335
 Preface

Since the pioneering work of Grothendieck [Grot], Hilbert schemes, which pa-
rametrize subschemes in algebraic varieties, have been studied extensively as evi-
denced in the survey [Iar2]. Viewed also as the simplest moduli spaces of sheaves,
they are fundamental objects in algebraic geometry, play essential roles in various
important enumerative problems, and provide testing grounds for many interesting
conjectures in algebraic geometry and its interplay with theoretical physics. Apart
from the Hilbert schemes of curves on 3-folds over which the Donaldson-Thomas
theory is defined and investigated [MNOP1, MNOP2], much attention has been
concentrated on Hilbert schemes of points, i.e., those Hilbert schemes parametriz-
ing 0-dimensional closed subschemes. When the dimension of the variety is at least
three, the corresponding Hilbert schemes of points are in general singular. When
the variety is a smooth curve, the corresponding Hilbert schemes of points coincide
with the symmetric products of the curve. When X is a smooth algebraic surface,
the Hilbert scheme X [n] parametrizing length-n 0-dimensional closed subschemes
of X is irreducible and smooth [Bri, Fog1, Iar1]. Moreover, the Hilbert-Chow
morphism ρn from X [n] to the symmetric product X (n) , which sends an element
in X [n] to its support (counted with multiplicities) in X (n) , is a crepant desin-
gularization of X (n) . Many fundamental aspects of X [n] such as its cohomology
groups, Chow groups, motive, cobordism class, and relations with algebraic com-
binatorics, the McKay correspondence and integrable systems have been analyzed
[CoG, dCM, EGL, ES1, Got1, Hai1, Hai2, Mar1].
In the seminal papers [Groj, Nak3] which were motivated by [Nak1, Nak2,
Nak4] regarding the construction of representations of affine Lie algebras on the
homology groups of the moduli spaces of instantons on ALE spaces (equivalently,
on the homology groups of quiver varieties), Grojnowski and Nakajima geomet-
rically constructed Heisenberg algebra actions on the cohomology of the Hilbert
schemes X [n] , where X denotes a smooth algebraic surface. Their geometric con-
structions started a whole new chapter investigating interplays between the Hilbert
schemes of points and infinite dimensional Lie algebras. Subsequently, using the
Hilbert schemes X [n] , Lehn [Leh1] geometrically constructed the Virasoro algebras
and the boundary operator, W.-P. Li, W. Wang and the author [LQW1, LQW4]
constructed the Chern character operators and the W algebras, and Carlsson and
Okounkov [Car1, Car2, CO] constructed the Ext vertex operators. As noted in
[FW, Introduction] and [Mat, Section 5], Lehn’s boundary operator is a version
of the bosonized Calogero-Sutherland operator. The Chern character operators are
vertex operators of higher conformal weights, and the W algebras are higher-spin
generalizations of Lehn’s Virasoro algebras. The Ext vertex operators of Carlsson
and Okounkov are motivated by the study of Nekrasov partition functions (which

ix
x PREFACE

arise from supersymmetric quantum gauge theory) in the setting of Hilbert schemes
of points.
All of these operators and algebras are powerful tools in understanding the
finer geometric properties of the Hilbert schemes X [n] . Indeed, the Heisenberg
algebras serve as a fundamental language in describing the homology and cohomol-
ogy classes of X [n] , which is crucial in studying the Gromov-Witten theory of X [n]
[Cheo2, ELQ, HLQ, LL, LQ1, LQ5, OP3, OP4]. The Virasoro operators and the
Chern character operators determine the cup products and ring structures of the
cohomology of X [n] [CoG, LQW1, LQW2, LQW3, LQW5, LS1, LS2]. The Ext
vertex operators provide a very useful approach in understanding the intersection
theory of X [n] when the tangent bundle of X [n] is involved [Car1, Car2, QY].
Two excellent books on Hilbert schemes of points exist. The first one is
Göttsche’s book [Got2] which includes fundamental facts about X [n] , the Betti
numbers of X [n] when X is a surface, and the computation of the homology and
Chow rings of Hilbert schemes. The second is Nakajima’s book [Nak5] which
contains his Heisenberg algebra constructions, symmetric products (in the Hilbert
scheme X [n] ) of an embedded curve in a surface X, and interactions with singu-
larities, symplectic geometry and the ring of symmetric functions. We also refer to
[EG, Leh2] for surveys on Hilbert schemes of points, the Heisenberg algebras and
the Virasoro algebras, and to [Nak7] for a survey on the equivariant cohomology
of the Hilbert schemes, the Heisenberg algebras, the Virasoro algebras and some
interesting applications to algebraic combinatorics.
The purpose of this book is to present a detailed survey of the developments,
in the context of interactions between Hilbert schemes of points and infinite dimen-
sional Lie algebras, that appeared after the book [Nak5]. This book contains 5
parts consisting of 16 chapters. Part 1 deals with the basics of the Hilbert schemes
of points and some geometry of the Hilbert schemes of points on the projective
plane. Part 2 is devoted to the constructions of various infinite dimensional Lie al-
gebra actions on the cohomology of the Hilbert schemes of points on surfaces, and
to the connections with multiple q-zeta values. It includes Nakajima’s affine Lie
algebra actions on the homology of quiver varieties as a motivation and background
material, the Heisenberg algebras of Grojnowski and Nakajima, the boundary op-
erator and the Virasoro algebras of Lehn, the Ext vertex operators of Carlsson and
Okounkov, and the Chern character operators and the W algebras of W.-P. Li,
W. Wang and the author. Part 3 studies the cohomology ring structure of these
Hilbert schemes, such as the ring generators and the ideals, the approach of Lehn
and Sorger when the canonical divisor of the surface is numerically trivial, the ap-
proach of Costello and Grojnowski in terms of the Calogero-Sutherland operators
and the Dunkl-Cherednik operators, the integral basis for the cohomology group,
and the orbifold cohomology ring of the symmetric product of a surface. Part 4
is about the equivariant cohomology of the Hilbert schemes via Jack polynomials,
the Hilbert/Gromov-Witten correspondence between the equivariant cohomology
of the Hilbert schemes of points on the affine plane and the Gromov-Witten theory
of curves, and applications to the Hurwitz numbers of the projective line. Part 5
includes the cosection localization technique of Kiem and J. Li, the Gromov-Witten
 PREFACE xi

theory of the Hilbert schemes of points, the equivariant quantum corrected bound-
ary operator and the quantum differential equations of Okounkov and Pandhari-
pande, the quantum corrected boundary operator of J. Li and W.-P. Li, and the
proof of Ruan’s Cohomological Crepant Resolution Conjecture.
This book is suitable for graduate students and researchers in algebraic ge-
ometry, representation theory, algebraic combinatorics, topology, number theory
and theoretical physics. Furthermore, a semester of an advanced course on Hilbert
schemes of points and infinite dimensional Lie algebras may be organized from se-
lected chapters in this book, e.g., Chapter 1, Chapter 3, Chapter 4, Chapter 7,
Chapter 8, Chapter 11 and Chapter 12.
The author would like to thank all of his collaborators on works related to
the Hilbert schemes of points, without whom this book would be impossible: Dan
Edidin, Jianxun Hu, Wei-Ping Li, Yuping Tu, Weiqiang Wang, Fei Yu and Qi
Zhang. In addition, thanks are due to Lie Fu, Wei-Ping Li, Hiraku Nakajima, Boiar
Qin, Wenzer Qin, Weiqiang Wang and the two anonymous referees for carefully
reading the manuscript and providing useful comments which have greatly improved
the exposition of the book.
The author also thanks Sergei Gelfand, Christine M. Thivierge and the Editorial
Committee of the American Mathematical Society for their editorial guidance.

December 14, 2017

Zhenbo Qin
 Part 1

Hilbert schemes of
points on surfaces
 CHAPTER 1

Basic results on Hilbert schemes of points

This chapter collects basic definitions and results regarding partitions, the ring
of symmetric functions, symmetric products, Hilbert schemes of points, and inci-
dence Hilbert schemes. Details may be found in Chapter I of [Mac2] and Chapters 1
and 2 of [Got2]. The focuses of this chapter are various symmetric functions and
the Hilbert schemes of points on a smooth projective complex surface. Throughout
the book, all the homology and cohomology are with C-coefficients, unless otherwise
specified.

1.1. Partitions
Let n ≥ 1. A partition λ of n, denoted by λ  n, is an infinite sequence
λ = (λ1 ≥ λ2 ≥ λ3 ≥ . . .)

 a finite sequence λ = (λ1 ≥ . . . ≥ λ )

of nonnegative integers λi , or equivalently
of positive integers λi , such that n = i λi . In this case, is the length of λ and
denoted by (λ), and n is the size of λ and denoted by |λ|. A partition may be
rewritten as
λ = (1m1 2m2 3m3 · · · ).
Put

(1.1) zλ = (imi mi !).
i≥1

Definition 1.1. Let λ = (λ1 ≥ · · · ≥ λk > 0) and λ̃ = (λ̃1 ≥ · · · ≥ λ̃ > 0) be

partitions. We define λ ≺ λ̃ if one of the three following conditions holds:
(i) |λ| < |λ̃|.
(ii) |λ| = |λ̃| and k < ,
(iii) |λ| = |λ̃|, k = , and λr < λ̃r for some r but λi = λ̃i for all i = r +1, . . . , k.
This is a total ordering on the set of all partitions. Another useful ordering is
the dominance (partial) ordering ≤.
Definition 1.2. Let λ = (λ1 ≥ λ2 ≥ · · · ) and λ̃ = (λ̃1 ≥ λ̃2 ≥ · · · ) be
partitions. We define λ ≤ λ̃ if
λ1 + . . . + λi ≤ λ̃1 + . . . + λ̃i
for all i ≥ 1.
A partition is also referred to as a 2-dimensional partition.
Definition 1.3. (i) Let d ≥ 2 and n ≥ 0. A d-dimensional partition λ of n,
denoted by λ  n, is an array
λ = (λi1 ,...,id−1 )i1 ,...,id−1
3
4 1. BASIC RESULTS ON HILBERT SCHEMES OF POINTS

of nonnegative integers λi1 ,...,id−1 indexed by the tuples (i1 , . . . , id−1 ) ∈ (Z≥0 )d−1
such that

λi1 ,...,id−1 = n,
i1 ,...,id−1

and λi1 ,...,id−1 ≥ λj1 ,...,jd−1 if i1 ≤ j1 , . . . , id−1 ≤ jd−1 .

(ii) We define Pd (n) to be the number of d-dimensional partitions of n.

One sees immediately that the generating function for P2 (n) is given by


+∞ 
+∞
1
(1.2) P2 (n)q n =
n=0 n=1
1 − qn

where q is a formal variable. A well-known result of McMahon [And] states that


+∞ 
+∞
1
(1.3) P3 (n)q n = .
n=0 n=1
(1 − q n )n

There is no analogous formula for Pd (n) when d > 3.

Geometrically, a (2-dimensional) partition can be represented by a Young dia-
gram. Here we adopt the convention of drawing diagrams upside down (so equiva-
lently, a partition can be represented by stacking unit squares in the first quadrant).

Definition 1.4. Let Dλ denote the Young diagram associated to a partition

λ. For a cell  in Dλ , let (),  (), a() and a () be respectively the numbers
of cells in Dλ to the south, north, east and west of  (excluding  itself). The
numbers () and  () are called the leg length and the leg colength respectively,
and a() and a () are called the arm length and the arm colength respectively.
The hook length h() and the content c() are defined to be a() + () + 1 and
a () −  () respectively.

Example 1.5. Let λ = (5, 5, 5, 2, 1) = (11 21 53 ). Then λ is a partition of 18.

Its Young diagram Dλ


contains 18 cells. For the cell ♠ ∈ Dλ , we have (♠) = 1, (♠) = 2, a(♠) = 3,
a (♠) = 1, h(♠) = 5, and c(♠) = −1.

Similarly, a 3-dimensional partition can be represented by stacking unit cubes

at a corner.

Definition 1.6. A generalized partition of an integer n is of the form

λ = (· · · (−2)m−2 (−1)m−1 1m1 2m2 · · · )

 1.2. THE RING OF SYMMETRIC FUNCTIONS 5

that part i ∈ Z has multiplicity mi , mi = 0 for only finitely many i’s, and
such
n = i imi . Define

(λ) = mi ,
i

|λ| = imi ,
i

s(λ) = i2 mi ,
i

!
λ = mi !.
i

Let −λ be the generalized partition whose multiplicity of i ∈ Z is m−i . The set of


all generalized partitions is denoted by P.

1.2. The ring of symmetric functions

Let x1 , . . . , xn be formal variables. The n-th symmetric group Sn acts on the
polynomial ring Z[x1 , . . . , xn ] by permuting the variables x1 , . . . , xn . Let
Λn = Z[x1 , . . . , xn ]Sn
be the subring of symmetric polynomials in x1 , . . . , xn . For r ≥ 0, let Λrn be the
additive group of homogeneous symmetric polynomials of degree r. Then,

+∞
Λn = Λrn .
r=0

By setting xn+1 = 0, there is a natural surjective homomorphism Λn+1 → Λn which

maps Λrn+1 surjectively to Λrn . Let
Λr = lim Λrn .
← −
n

Then the ring of symmetric functions in countably many independent variables

x1 , x2 , . . . is defined to be

+∞
Λ= Λr .
r=0
Note that Λ is not the inverse limit of the rings Λn relative to the homomorphisms
Λn+1 → Λn . For instance, the infinite product +∞ i=1 (1 + xi ) is not contained in Λ
since it is not a finite sum of homogeneous symmetric functions.
For each integer r ≥ 0, the r-th elementary symmetric function er is the sum
of all products of r distinct variables xi , so that e0 = 1 and

er = x i1 · · · x ir
i1 <...<ir

for r ≥ 1. For a partition λ = (λ1 ≥ λ2 ≥ . . .), define the elementary symmetric

function eλ by
eλ = eλ1 eλ2 · · · .
For a sequence α = (α1 , α2 , . . .) of nonnegative integers, define

1 x2 · · · .
xα = xα1 α2
6 1. BASIC RESULTS ON HILBERT SCHEMES OF POINTS

For a partition λ = (λ1 ≥ λ2 ≥ . . .), define the monomial symmetric function by


mλ = xα
α

where α runs over all distinct permutations of the sequence (λ1 , λ2 , . . .).
For each integer r ≥ 1, the r-th power sum is defined to be

+∞
pr = xri .
i=1

For a partition λ = (λ1 ≥ λ2 ≥ . . .), define the power symmetric function pλ by

p λ = p λ1 p λ2 · · · .
Since the monomial symmetric function m(1n ) is the same as the n-th elementary
symmetric function, we conclude from formula (2.14 ) in Chapter I of [Mac2] that
 (−1)n−(μ) pμ
(1.4) m(1n ) = .
zμ
|μ|=n

Moreover, it is a classical result that


(1.5) p i mλ = aλ,μ mμ
μ

where the summation is over certain partitions μ of |λ| + i, which are obtained as
follows:
(i) add i to a part in λ, say λk (possibly 0), and then
(ii) arrange the parts in descending order.
The coefficient aλ,μ is equal to the number of elements in { | μ = λk + i}.
To define the Schur functions, we will work in Z[x1 , . . . , xn ] for the moment.
For a sequence α = (α1 , . . . , αn ) of nonnegative integers, let
α
aα = det(xi j )1≤i,j≤n .
When δ = (n − 1, n − 2, . . . , 1, 0), aδ is the Vandermonde determinant and equal to
i<j (xi − xj ). Note that aα = 0 if two of the integers αi are equal. So up to sign,
we may assume that α1 > . . . > αn ≥ 0. For 1 ≤ i ≤ n, define λi by αi = λi + n − i.
Put λ = (λ1 , . . . , λn ). Then λ is a partition of length at most n, and α = λ + δ.
The determinant aα = aλ+δ is divisible by aδ in the polynomial ring Z[x1 , . . . , xn ],
and
sλ (x1 , . . . , xn ) = aλ+δ /aδ
is a symmetric homogeneous polynomial of degree |λ|. Moreover,
sλ (x1 , . . . , xn , 0) = sλ (x1 , . . . , xn ).
Therefore, there exists a unique element
sλ ∈ Λ
that reduces to sλ (x1 , . . . , xn ) when xn+1 , xn+2 , . . . are all set equal to zero, for
every n ≥ (λ). This symmetric function sλ is the Schur function corresponding
to λ.
 1.3. SYMMETRIC PRODUCTS 7

Next, we review the Jack symmetric functions. Fix a positive real number α.
Let Q(α) be the subfield of R generated by α over Q. Let ·, · α be the inner product
on Λ ⊗Z Q(α) defined by declaring
pμ , pν α = α(μ) zμ δμ,ν .
(α)
For a fixed positive real number α, the Jack symmetric functions Pλ form a
basis of Λ ⊗Z Q(α) indexed by partitions λ and are characterized by the following
properties (1.6) and (1.7):
(α)
 (α)
(1.6) P λ = mλ + uλ,μ mμ
μ<λ
(α)
where uλ,μ are real numbers and < denotes the dominance ordering, and
(α)
(1.7) Pλ , Pμ(α) α =0
if λ = μ.
It is well-known that each of the sets {eλ }λ , {mλ }λ and {sλ }λ is a Z-linear
basis of Λ. In addition, the set
{ei = m(1i ) }i≥1
generates the integral ring Λ. The set {pλ }λ is a Q-linear basis of Λ ⊗Z Q, but not
(α)
a Z-linear basis of Λ. For a fixed positive real number α, Pλ λ is a linear basis
of Λ ⊗Z Q(α).
Finally, there is an involution ω on Λ satisfying
(1.8) ω(pλ ) = (−1)|λ|−(λ) pλ .
The forgotten symmetric function fλ associated to a partition λ is defined by
(1.9) fλ = ω(mλ ).
The set {fλ }λ is another Z-linear basis of Λ.

1.3. Symmetric products

Let X be a compact polyhedron, and X n be the n-th Cartesian product of
X. The n-th symmetric group Sn acts on X n by permuting the factors. The n-th
symmetric product X (n) is the quotient space X n /Sn . An element in X n is denoted
by a tuple (x1 , . . . , xn ). An element in X (n) is denoted by x1 + . . . + xn where the
points x1 , . . . , xn are not necessarily distinct. Equivalently, an element in X (n) can
also be denoted by n1 x1 + . . . + nk xk where the points x1 , . . . , xk ∈ X are distinct,
n1 , . . . , nk are positive, and n1 + . . . + nk = n.
Let Hi (X) be the i-th homology group of X with complex coefficients, and
bi (X) = dim Hi (X) be the i-th Betti number of X. The Poincaré polynomial of X
is defined to be

+∞
Pt (X) = bi (X) ti .
i=0
Macdonald [Mac1] proved that the generating function for the Poincaré polynomial
of X (n) is given by

+∞
(1 + tq)b1 (X) (1 + t3 q)b3 (X) · · ·
(1.10) Pt (X (n) ) q n = .
n=0
(1 − q)b0 (X) (1 − t2 q)b2 (X) (1 − t4 q)b4 (X) · · ·
8 1. BASIC RESULTS ON HILBERT SCHEMES OF POINTS

Now let X be a quasi-projective complex variety. To see whether the symmetric

product X (n) exists as an algebraic variety, we introduce the following.
Definition 1.7. Let Y be a quasi-projective complex variety with an action of
a group G. A variety Z together with a surjective morphism π : Y → Z is defined
to be a quotient of Y by G if
(i) the fibers of π are the orbits of G,
(ii) π̃ : Y → W factors through π whenever π̃ is a G-equivariant morphism
and maps every G-orbit in Y to a single point in W .
If a quotient exists, then it is unique up to isomorphisms and denoted by Y /G.
In general, it is not easy to see whether a quotient exists, and the question is
somewhat related to geometric invariant theory. On the other hand, when the
group G is finite, the quotient Y /G always exists.
Theorem 1.8. Let Y be a quasi-projective complex variety with an action of a
finite group G. Then the quotient Y /G exists as a quasi-projective variety.
Proof. First of all, we assume that Y is affine. Let
Y /G = Spec A[Y ]G
where A[Y ] denotes the affine coordinate ring of Y . To prove that A[Y ]G is finitely
generated over C, note that we can write A[Y ] in the form C[x1 , . . . , xm ]/I(Y )
where G acts on x1 , . . . , xm as permutations. Moreover, we have a surjection
G
C[x1 , . . . , xm ]G → C[x1 , . . . , xm ]/I(Y ) .
G
Indeed, an element α ∈ C[x1 , . . . , xm ]/I(Y ) corresponds to an element α ∈
C[x1 , . . . , xm ] congruent to every element g ∗ (α) mod I(Y ). Then the average
1  ∗
g (α) ∈ C[x1 , . . . , xm ]
|G|
g∈G

is G-invariant and will be mapped to α. Thus it suffices to prove that C[x1 , . . . , xm ]G

is finitely generated over C. We have
C[x1 , . . . , xm ]Sm ⊂ C[x1 , . . . , xm ]G ⊂ C[x1 , . . . , xm ].
By the fundamental theorem on symmetric polynomials,
C[x1 , . . . , xm ]Sm = C[ẽ1 , . . . , ẽm ]
where ẽi = ei (x1 , . . . , xm , 0, . . .) with 1 ≤ i ≤ m are the elementary symmetric poly-
nomials with the m variables x1 , . . . , xm . Since C[x1 , . . . , xm ] is a finitely generated
module over C[ẽ1 , . . . , ẽm ], so is C[x1 , . . . , xm ]G . It follows that C[x1 , . . . , xm ]G is a
finitely generated algebra over C.
Next we prove that the points of Y /G correspond to the G-orbits on Y . To see
that π(p) = π(q) for two points p, q ∈ Y not in the same G-orbit, observe that we
can find a function f ∈ A[Y ] such that f (p) = 0 but f (g(q)) = 0 for every g ∈ G.
The product g∈G g ∗ (f ) gives rise to a function in A[Y ]G which vanishes at π(p)
but not at π(q). To see that π is surjective, let m = (h1 , . . . , hk ) be a maximal
ideal in A[Y ]G . We claim that m · A[Y ] = A[Y ]. If m · A[Y ] = A[Y ], then
1 = a1 h1 + . . . + ak hk
 1.3. SYMMETRIC PRODUCTS 9

for some a1 , . . . , ak ∈ A[Y ]. Applying g ∗ for every g ∈ G and summing up yield

⎛ ⎞ ⎛ ⎞
 
|G| = ⎝ g ∗ (a1 )⎠ h1 + . . . + ⎝ g ∗ (ak )⎠ hk .
g∈G g∈G

Thus 1 ∈ m which is absurd. Now let m be a maximal ideal of A[Y ] containing

m · A[Y ]. Then m = m ∩ A[Y ]G .
To prove the universal property of π, we may assume that W is affine. Then
π̃ : Y → W corresponds to a homomorphism α̃ : A[W ] → A[Y ]. It suffices to prove
that α̃(A[W ]) ⊂ A[Y ]G . Let f ∈ A[W ]. For every g ∈ G and p ∈ Y , we have

g ∗ (α̃(f )) (p) = (α̃(f ))(g(p)) = f π̃(g(p)) = f π̃(p) = (α̃(f ))(p)

by the assumptions on π̃. So g ∗ (α̃(f )) = α̃(f ) for every g ∈ G, and α̃(f ) ∈ A[Y ]G .
Finally, if Y is not affine, then there exists an open cover of Y by G-invariant
affine open subsets Yi , and Y /G is the gluing of the quasi-projective varieties Yi /G.
We refer to [Harr, p. 127] for the proof that Y /G is quasi-projective. 

By Theorem 1.8, if X is a quasi-projective complex variety, then so is the sym-

metric product X (n) . Note that the symmetric product X (n) may not be smooth
even if X itself is smooth.

Example 1.9. (i) Note that C[x1 , . . . , xn ]Sn = C[ẽ1 , . . . , ẽn ] where

ẽi = ei (x1 , . . . , xn , 0, . . .)

with 1 ≤ i ≤ n are the elementary symmetric polynomials with the n variables

x1 , . . . , xn . It follows that (A1 )(n) = An . Similarly, (P1 )(n) = Pn .
(ii) If X is a smooth curve, then locally in the étale topology X is isomorphic
to A1 . In view of (i), the symmetric product X (n) is smooth.

Example 1.10. The symmetric product (A2 )(2) is not smooth. In fact,

C[u, v, w]
(1.11) (A2 )(2) ∼
= A2 × Spec .
uw − v 2

To prove (1.11), let x1 , y1 and x2 , y2 be the coordinates on the two factors of A2 ×A2 .
Let s = x1 − x2 , t = y1 − y2 , s = x1 + x2 and t = y1 + y2 . Denote the transposition
in S2 by τ . Then, τ (s) = −s, τ (t) = −t, τ (s ) = s , and τ (t ) = t . So

C[x1 , y1 , x2 , y2 ]S2 = C[s, t, s , t ]S2 = C[s, t]S2 [s , t ].

Note that τ (si tj ) = (−1)i+j si tj . Thus, C[s, t]S2 is the subalgebra generated by
s2 , st, t2 . Putting u = s2 , v = st and w = t2 , we see that
C[u, v, w]
C[s, t]S2 ∼
= .
uw − v 2
Hence
C[u, v, w]
(A2 )(2) = Spec C[x1 , y1 , x2 , y2 ]S2 ∼
= A2 × Spec .
uw − v 2
10 1. BASIC RESULTS ON HILBERT SCHEMES OF POINTS

1.4. Hilbert schemes of points

Let X be a projective complex scheme and OX (1) be an ample line bundle on
X. Consider the contravariant functor
HilbX : [Schemes] → [Sets]
from the category of complex schemes to the category of sets defined by
HilbX (U ) = {Z ⊂ U × X|Z is a closed subscheme and flat over U } .
Namely, HilbX is the functor associating a complex scheme U to the set of flat
families of closed subschemes in X parametrized by U . Let π : Z → U be the
restriction of the natural projection U × X → U . For u ∈ U , we define the Hilbert
polynomial
Pu (m) = χ OZu ⊗ OX (m)
where Zu = π ∗ (u). For each polynomial P , let HilbP
X be the subfunctor of HilbX
associating a complex scheme U to the set of flat families of closed subschemes in X
parametrized by U such that P is the Hilbert polynomial of the closed subschemes.
Grothendieck [Grot] proved the following fundamental result.
Theorem 1.11. The functor HilbP
X is representable by a projective scheme
HilbP
X .
In particular, there exists a closed subscheme Z ⊂ HilbP X × X such that Z →
HilbPX is a universal family in the following sense: if Z is a flat family of closed
subschemes in X parametrized by U such that P is the Hilbert polynomial of the
closed subschemes, then there exists a unique morphism f : U → HilbP X such that

Z = (f × IdX )∗ Z.
Furthermore, if Y is an open subscheme of X, then there is a corresponding
open subscheme HilbP P
Y of HilbX parametrizing closed subschemes in Y . In partic-
P
ular, HilbY is defined for a quasi-projective scheme Y .
If [Z] ∈ HilbP
X is a point corresponding to a closed subscheme Z of X, then the
Zariski tangent space of HilbPX at [Z] is canonically isomorphic to

(1.12) Hom(IZ , OZ ).
Since every homomorphism IZ → OZ of OX -modules factors uniquely through
IZ /IZ2 , the Zariski tangent space of HilbP
X at [Z] is also canonically isomorphic to

(1.13) Hom(IZ /IZ2 , OZ ).

Definition 1.12. Let n be a positive integer. We define X [n] to be HilbP X where
P is the constant polynomial given by P (m) = n for all m ∈ Z, and call it the Hilbert
scheme of n points in X. An element ξ ∈ X [n] represents a 0-dimensional closed
subscheme of X with length (ξ) = h0 (X, Oξ ) = n. For convenience, we define X [0]
to be a reduced point.
Example 1.13. (i) Note that a length-1 0-dimensional closed subscheme of a
smooth scheme X is simply a point of X. So X [1] = X. The universal closed
subscheme in X [1] × X = X × X is the diagonal in X × X.
(ii) Let ξ be the closed subscheme of a smooth scheme X consisting of distinct
points x1 , . . . , xn ∈ X, denoted by ξ = x1 +. . .+xn . Then, ξ ∈ X [n] . These elements
 1.4. HILBERT SCHEMES OF POINTS 11

ξ form a smooth open (but not necessarily dense) subset of X [n] of dimension
dim(X) · n.
(iii) A length-n 0-dimensional closed subscheme of Ad corresponds to an ideal
I ⊂ C[x1 , . . . , xd ] of colength-n, i.e.,
dim C[x1 , . . . , xd ]/I = n.
Therefore, as sets,
[n]
Ad = I ⊂ C[x1 , . . . , xd ]| I is a colength-n ideal of C[x1 , . . . , xd ] .
Theorem 1.14. Let X be a smooth irreducible variety with dim(X) ≤ 2. Then
X [n] is a smooth irreducible variety of dimension dim(X) · n.
Proof. Since X is connected, X [n] is connected as well. In view of Exam-
ple 1.13 (ii), it remains to prove that for every element ξ ∈ X [n] , the dimension of
the Zariski tangent space Tξ,X [n] of X [n] at ξ is at most dim(X) · n. By (1.12), we
have
Tξ,X [n] = Hom(Iξ , Oξ ).
When dim(X) = 1, we obtain Iξ = OX (−ξ) and Hom(Iξ , Oξ ) = ∼ H 0 (X, Oξ ).
So dim Tξ,X [n] = (ξ) = n and we are done.
Let dim(X) = 2. Applying the functor Hom(·, Oξ ) to the exact sequence
0 → Iξ → OX → Oξ → 0
yields an exact sequence
0 → Hom(Oξ , Oξ ) → Hom(OX , Oξ ) → Hom(Iξ , Oξ ) → Ext1 (Oξ , Oξ ).
Since
dim Hom(Oξ , Oξ ) = (ξ) = n = dim Hom(OX , Oξ ),
we have an injection
0 → Hom(Iξ , Oξ ) → Ext1 (Oξ , Oξ ).
So it remains to prove that
(1.14) dim Ext1 (Oξ , Oξ ) = 2n.
We have
dim Ext2 (Oξ , Oξ ) = dim Hom(Oξ , Oξ ⊗ KX ) = n
by the Serre duality and dim Hom(Oξ , Oξ ) = n. By the Riemann-Roch Theorem,

2
(−1)i · dim Exti (Oξ , Oξ ) = χ(Oξ , Oξ ) = 0.
i=0

It follows that dim Ext1 (Oξ , Oξ ) = 2n. This proves (1.14). 

Remark 1.15. When dim X = 2, Theorem 1.14 is due to Fogarty [Fog1,
Theorem 2.4]. In a different direction, one can also show that if X is a smooth
variety, then X [n] is smooth and irreducible for all n ≤ 3 [FG2, Remark 7.2.5].
Example 1.16. Let X be a smooth quasi-projective variety. An element ξ ∈
X [2] is either a set {x1 , x2 } of distinct points in X or a length-2 closed subscheme
supported at one point x ∈ X. In the latter case, let m be the ideal of x. Then,
m2 ⊂ Iξ ⊂ m. So Iξ is given by a codimension-1 subspace of m/m2 , i.e., by a point
in P(Tx,X ) = Pd−1 where d denotes the dimension of X. Hence an element in X [2]
12 1. BASIC RESULTS ON HILBERT SCHEMES OF POINTS

is either a set of two distinct points or a point together with a tangent direction at
that point. In fact, the Hilbert scheme X [2] can be described globally as follows.
Let X 2 be the blow-up of X 2 along the diagonal. The action of S2 on X 2 extends

to X 2 . Then X [2] = X2 /S2 .

Let X be a smooth quasi-projective variety. Sending an element ξ ∈ X [n] to

the element 
Oξ,x x ∈ X (n)
x∈X
defines the Hilbert-Chow morphism
(1.15) ρn : X [n] → X (n) .
A detailed construction of ρn can be found in Section 5.4 of [MF].
Definition 1.17. (i) For a closed subset S of X, define Mn (S) ⊂ X [n] to be
the closed subset 
Mn (S) = (ρn )−1 (nx)
x∈S
with the reduced induced subscheme structure.
(ii) The punctual Hilbert scheme at a point x ∈ X is defined to be Mn (x).
Analytically, the punctual Hilbert scheme Mn (x) is independent of the point
x ∈ X and the smooth variety X as long as the dimension of X is fixed. In fact,
let d = dim X,
R = C[[x1 , . . . , xd ]]
be the local ring of formal power series in the variables x1 , . . . , xd , and m =
(x1 , . . . , xd ) be the maximal ideal of R. Then,
Mn (x) ∼
= Hilbn (R)
which denotes the Hilbert scheme parametrizing all colength-n ideals in R/mn .
The punctual Hilbert scheme Hilbn (R) has been studied extensively (e.g. [Bri,
Iar1, Iar2]). The following is due to Briancon [Bri].
Theorem 1.18. The punctual Hilbert scheme Hilbn (C[[x, y]]) is irreducible of
dimension (n − 1). Moreover, generic elements in Hilbn (C[[x, y]]) correspond to
curvi-linear ideals of the form
y − a1 x − . . . − an−1 xn−1 , xn
where a1 , . . . , an−1 ∈ C.
Remark 1.19. Let X be a smooth surface, and let n1 x1 + . . . + n x ∈ X (n)
where x1 , . . . , x are distinct and n1 , . . . , n > 0. Then


(ρn ) −1
(n1 x1 + . . . + n x ) ∼
= (ρni )−1 (ni xi )
i=1


∼
= Hilbni (C[[x, y]]).
i=1
−1
By Theorem 1.18, (ρn ) (n1 x1 + . . . + n x ) is irreducible of dimension (n − ).
 1.4. HILBERT SCHEMES OF POINTS 13

Remark 1.20. Fix a point x ∈ X. For m ≥ 0 and n > 0, define

(1.16) Mm,m+n (x) = {(ξ, ξ  )| ξ ⊂ ξ  } ⊂ Mm (x) × Mm+n (x).
Ellingsrud and Strømme [ES3] proved that Mm,m+1 (x) is irreducible with
(1.17) dim Mm,m+1 (x) = m.
Proposition 1.21. Let X be a smooth quasi-projective variety.
(i) If dim X = 1, then the Hilbert-Chow morphism ρn is an isomorphism.
(ii) If dim X = 2, then ρn is a resolution of singularities.
Proof. (i) Since dim X = 1, every ideal sheaf is invertible. Thus for every
x ∈ X and every positive integer k, there exists a unique length-k 0-dimensional
closed subscheme ξ supported at x. This unique subscheme ξ is defined by the
ideal sheaf OX (−kx), and ρk (ξ) = kx. It follows that the Hilbert-Chow morphism
ρn : X [n] → X (n) is bijective. By Theorem 1.14 and Example 1.9 (ii), both X [n] and
X (n) are smooth of dimension n. Hence ρn is an isomorphism in view of Zariski’s
Main Theorem.
(n)
(ii) Let X0 be the open subset of X (n) consisting of sums of n distinct points.
(n) (n)
Then X0 is smooth, and ρn is an isomorphism over X0 . By Theorem 1.14,
X [n] is smooth, irreducible and of dimension 2n. Since X (n) is irreducible and of
dimension 2n, it remains to prove that the singular locus Sing X (n) of X (n) is
(n)
X (n) \X0 .
Let π : X n → X (n) be the quotient map. The big diagonal of X n is

Δ= Δi,j
1≤i<j≤n

where
Δi,j = {(x1 , . . . , xn ) ∈ X n |xi = xj }.
Let Δ0 be the open subset of Δ where precisely two of the xi ’s coincide. Let p ∈ Δ0 .
Without loss of generality, we may assume that p ∈ Δ1,2 . By Example 1.10, a
formal neighborhood of the point π(p) in X (n) (i.e., the complete local ring of X (n)
at π(p)) is isomorphic to
C[[u, v, w, s , t , x3 , y3 , . . . , xn , yn ]]
Spec .
uw − v 2
Thus all the points in π(Δ0 ) are singular points of X (n) . So π(Δ0 ) is contained in
the singular locus Sing X (n) . Since Sing X (n) is closed, the closure of π(Δ0 ) is
(n)
contained in Sing X (n) . Since the closure of π(Δ0 ) is X (n) \X0 and
(n)
Sing X (n) ⊂ X (n) \X0 ,
(n)
we conclude that Sing X (n) is equal to X (n) \X0 . 

The generating function for the Euler characteristic χ X [n] of the Hilbert
scheme X [n] was calculated in [Che1] (the case d = 1 also follows from Proposi-
tion 1.21 (i) and (1.10) by setting t = −1, and the case d = 2 is due to Göttsche
and follows from Theorem 1.23 below by setting t = −1).
14 1. BASIC RESULTS ON HILBERT SCHEMES OF POINTS

Proposition 1.22. Let X be a smooth quasi-projective variety of dimension

d, and let Mn (O) be the punctual Hilbert scheme at the origin O ∈ Ad . Then,
 +∞ χ(X)

+∞ 
[n] n n
(1.18) χ X q = χ Mn (O) q
n=0 n=0
 +∞ χ(X)

n
= Pd (n)q
n=0

where Pd (n) denotes the number of d-dimensional partitions of n.

Proof. The first equality follows from the standard stratifications of X [n] and
(n)
X together with a careful analysis of the fibers of the Hilbert-Chow morphism
ρn . We refer to [Che1] for the details. For the second equality in (1.18), it suffices
to prove
(1.19) χ Mn (O) = Pd (n).
Note that the d-dimensional torus T = (C∗ )d acts on Ad via
(1.20) (t1 , . . . , td )(x1 , . . . , xd ) = (t1 x1 , . . . , td xd )
where x1 , . . . , xd denote the affine coordinates of Ad , and (t1 , . . . , td ) ∈ T. This
action lifts to a T-action on (Ad )[n] , and Mn (O) is invariant. By the Theorem 4.1
in [LY],
(1.21) χ Mn (O) = χ Mn (O)T
where Mn (O)T denotes the set of fixed points under the T-action. Using Exam-
ple 1.13 (iii), the elements in Mn (O)T are in one-to-one correspondence with T-
invariant colength-n ideals I of C[x1 , . . . , xd ]. These ideals must be generated by
monomials, and are in one-to-one correspondence with d-dimensional partitions of
n. Indeed, given a d-dimensional partition (λi1 ,...,id−1 )i1 ,...,id−1 ≥0 of n, the ideal of
C[x1 , . . . , xd ] generated by the monomials
i ni1 ,...,id−1
xi11 · · · xd−1
d−1
xd
has colength-n. Conversely, given a colength-n ideal I of C[x1 , . . . , xd ] generated
by monomials, we obtain a d-dimensional partition (λi1 ,...,id−1 )i1 ,...,id−1 ≥0 of n by
putting
 
id−1 id
λi1 ,...,id−1 = min id | xi11 · · · xd−1 xd ∈ I .

Therefore, χ Mn (O) = Pd (n). This completes the proof of (1.19). 

In particular, when dim X = 2, we see from (1.18) and (1.2) that


+∞ 
+∞
1
(1.22) χ X [n] q n = .
n=0 n=1
(1 − q n )χ(X)

When dim X = 3, we see from (1.18) and (1.3) that


+∞
(1.23) χ X [n] q n = M (q)χ(X)
n=0
 1.4. HILBERT SCHEMES OF POINTS 15

where

+∞
1
M (q) = .
n=1
(1 − q n )n
For the finer invariants of X [n] such as its Betti numbers, we have the following
elegant formula [Got1, GS] when X is a surface.
Theorem 1.23. Let X be a smooth quasi-projective surface. Then,

+∞ 
+∞ 4  (−1)i bi (X)
[n] n 1
Pt (X )q = .
n=0 n=1 i=0
1 − (−t)2n−2+i q n

It will be clear from Chapter 3 that Theorem 1.23 is important in the interplay
between the Hilbert schemes of points and infinite dimensional Lie algebras.
Let Zn ⊂ X [n] × X be the universal codimension-2 closed subscheme:
(1.24) Zn = (ξ, x) ⊂ X [n] × X | x ∈ Supp (ξ) .
Let p : Zn → X [n] and q : Zn → X be the natural morphisms. For a sheaf F on
X, let
(1.25) F [n] = p∗ q ∗ F.
Since p is a degree-n flat morphism, if F is locally free of rank r, then F [n] is locally
free of rank rn; in this case, F [n] is called a tautological bundle over X [n] . We define
the subset Bn of the Hilbert scheme X [n] by putting
 
(1.26) Bn = ξ ∈ X [n] | | Supp(ξ)| < n .

Let X be a projective surface. Then Bn is codimension-1 in X [n] and called

the boundary of the Hilbert scheme X [n] . For a real-surface C in X, define
 
(1.27) DC = ξ ∈ X [n] | Supp(ξ) ∩ C = ∅ .

Note that DC is a codimension-1 subvariety of X [n] when C is a smooth algebraic

curve in X. We extend the notion DC to all divisors C in X by linearity.
Theorem 1.24. Let n > 1 and X be a smooth projective surface. Then,
(i) det (OX (C))[n] = DC − Bn /2 for every divisor C in X. Moreover, if
C is a smooth curve in X and C [n] is regarded as a closed subscheme of
X [n] , then
 
(1.28) cn (OX (C))[n] = C [n] ,
(1.29) NC [n] ⊂X [n] = (OX (C))[n] |C [n]
where NC [n] ⊂X [n] denotes the normal bundle of C [n] in X [n] .
(ii) KX [n] = ρ∗n KX (n) = DKX .
(iii) Pic(X [n] ) ∼
= (Pic(X n ))Sn ⊕ (Bn /2) Z.
Theorem 1.24 (i), (ii) and (iii) were proved in [AIK, BSG], [Bea1] and [BSG,
Fog2] respectively. When H 1 (X, OX ) = 0, it follows from Theorem 1.24 (iii) that
(1.30) Pic(X [n] ) ∼
= Pic(X) ⊕ (Bn /2) Z.
Under this isomorphism, the divisor DC ∈ Pic(X [n] ) corresponds to C ∈ Pic(X).
16 1. BASIC RESULTS ON HILBERT SCHEMES OF POINTS

Remark 1.25. Let n > 1 and X be a smooth projective surface. By Proposi-

tion 1.21 (ii) and Theorem 1.24 (ii), the Hilbert-Chow morphism ρn : X [n] → X (n)
is a crepant resolution.
Definition 1.26. Let L be a line bundle on a smooth projective surface X.
Then,
(i) L is nef if L · C ≥ 0 for every curve C in X.
(ii) L is nef and big if it is nef and L2 > 0.
(iii) L is k-very ample if the restriction
H 0 (X, L) → H 0 (X, Oξ ⊗ L)
is surjective for every ξ ∈ X [k+1] .
The concept of k-very ampleness relates the Hilbert scheme X [k+1] to a Grass-
mannian as follows. The surjective map in Definition 1.26 (iii) represents an element
in Grass(H 0 (X, L), k + 1). So we get a morphism
(1.31) ϕk (L) : X [k+1] → Grass(H 0 (X, L), k + 1)
when L is k-very ample. Let h = h0 (X, L) and
p : Grass(Ch , k + 1) → P((∧h−(k+1) Ch )∗ )
be the Plücker embedding. Then by the Appendix of [BSG], we have
(1.32) (p ◦ ϕk (L))∗ H = OX [k+1] Dc1 (L) − Bk+1 /2
where H is the hyperplane line bundle over P((∧h−(k+1) Ch )∗ ).
Theorem 1.27. Let L be a line bundle on a smooth projective surface X.
(i) If L is nef and big, and if L2 ≥ 4k +5, then either KX +L is k-very ample,
or there exists an effective divisor D such that L − 2D is Q-effective, D
contains some 0-cycle of degree ≤ (k +1) where the k-very ampleness fails,
and
L · D − k − 1 ≤ D2 < L · D/2 < (k + 1).
(ii) If L is (k + 1)-very ample, then the morphism ϕk (L) in (1.31) is an
embedding.
Definition 1.26 (iii) and Theorem 1.27 (i) are from [BSG], and Theorem 1.27 (ii)
is from [CG].
Remark 1.28. Let k ≥ 1 be an integer. By Theorem 1.27 (i), there exists
a very ample divisor C on X such that the line bundle L = OX (C) is (k + 1)-
very ample. Then by Theorem 1.27 (ii), the morphism ϕk (L) is an embedding.
Therefore, by (1.32), the divisor (DC − Bk+1 /2) on X [k+1] is very ample.
Let X be a smooth projective surface. A basic intersection formula ([ES3,
Theorem 1.1] and [Nak5, Theorem 9.20]) for X [n] states that
(1.33) [Mn (A1 )] · [Mn (A2 )] = (−1)n−1 n(A1 · A2 )
where A1 and A2 are subvarieties of X with dim(A1 ) + dim(A2 ) = 2.
For n ≥ 2, let βn ∈ H2 (X [n] ; Z) be the homology class in H2 (X [n] ) represented
by a curve of the form:
(1.34) M2 (x) + η
 1.5. INCIDENCE HILBERT SCHEMES 17

where x ∈ X and η ∈ X [n−2] are fixed such that x ∈ Supp(η). Then,

(1.35) βn , Bn = [M2 (x)] · [M2 (X)] = −2
where we have used (1.33) in the last step.
Remark 1.29. Using (1.35), we now prove the formula
KX [n] = ρ∗n KX (n)
in Theorem 1.24 (ii). Since ρn : X [n] → X (n) is the desingularization of X (n) and
the exceptional divisor Bn is irreducible,
KX [n] = ρ∗n KX (n) + aBn
for some rational number a. So it suffices to show a = 0. To see this, we have
a = −KX [n] · βn /2 = −KX [2] · β2 /2
by (1.35). Recall from Example 1.16 that X [2] = X 2 /S2 where X 2 is the blow-up
2
of X along the diagonal Δ. Let E be the exceptional divisor in X 2 , and let f be
a fiber of the P -bundle E → Δ. Let π1 : X
1 2 → X be the blowing-up morphism,
2


and let π2 : X → X be the quotient map. Then, KX
2 [2] ∗
2 = π1 KX 2 + E and
∗ ∗ ∗
KX 2 = π2 KX [2] + E. Thus, π1 KX 2 = π2 KX [2] . Finally, since π2∗ f = β2 , we obtain

a = −KX [2] · π2∗ f /2 = −π2∗ KX [2] · f /2 = −π1∗ KX 2 · f /2 = 0

where the projection formula has been used twice.

1.5. Incidence Hilbert schemes

In this section, let X be a smooth projective surface. The incidence Hilbert
scheme X [n,n+1] is defined by
(1.36) X [n,n+1] = {(ξ, η) | ξ ⊂ η} ⊂ X [n] × X [n+1] .
It is known [Che2, Tik] that X [n,n+1] is irreducible, smooth and of dimension
2(n + 1). Define
(1.37) Mn,n+1 (X) = {(ξ, η) ∈ X [n,n+1] | Supp(η) = {x} for some x ∈ X}.
Notation 1.30. We fix the following notations for various morphisms:
fn+1 : X [n,n+1] → X [n] with fn+1 (ξ, η) = ξ.
gn+1 : X [n,n+1] → X [n+1] with gn+1 (ξ, η) = η.
ψn+1 : X [n,n+1] → Zn+1 with ψn+1 (ξ, η) = η, Supp(Iξ /Iη ) .
ρ : X [n,n+1] → X with ρ(ξ, η) = Supp(Iξ /Iη ).
The following theorem was proved in [ES3].
Theorem 1.31. (i) The morphism ψn+1 : X [n,n+1] → Zn+1 is canoni-
cally isomorphic to the projectification P(ωZn+1 ) → Zn+1 where ωZn+1 is
the dualizing sheaf of Zn+1 .
(ii) The morphism (fn+1 , ρ) : X [n,n+1] → X [n] × X is canonically isomorphic
to the blowing-up of X [n] × X along Zn . The exceptional locus En+1 is
equal to
(ξ, η) ∈ X [n,n+1] | Supp(ξ) = Supp(η) .
18 1. BASIC RESULTS ON HILBERT SCHEMES OF POINTS

(iii) As cycles in the Chow group,

1
g ∗ [Mn+1 (X)] = [Mn,n+1 (X)].
n + 1 n+1
Put fX = fn+1 × IdX , gX = gn+1 × IdX and ρX = ρ × IdX . Let ΔX be the
diagonal in X × X. Then there exists an exact sequence over X [n,n+1] × X:
(1.38) 0 → ρ∗X OΔX ⊗ p∗1 OX [n,n+1] (−En+1 ) → gX
∗ ∗
OZn+1 → fX OZn → 0
where p1 : X [n,n+1] × X → X [n,n+1] is the first projection. Let F be a locally free
sheaf on X. Applying the functor p1∗ (· ⊗ p∗1 ρ∗ F ) to (1.38) yields an exact sequence:
(1.39) 0 → ρ∗ F ⊗ OX [n,n+1] (−En+1 ) → gn+1
∗ ∗
F [n+1] → fn+1 F [n] → 0.
In particular, setting F = OX and using Theorem 1.24 (i), we obtain
∗ ∗
(1.40) gn+1 Bn+1 − fn+1 Bn = 2En+1 .
The generating function for the Poincaré polynomials of X [n,n+1] has a closed
formula due to Cheah [Che3, Section 0.3]:

+∞
(1.41) Pt X [n,n+1] q n
n=0
+∞ 4  (−1)
Pt (X)  
i
bi (X)
1
= ·
1 − t2 q n=1 i=0 1 − (−t)2n−2+i q n

Pt (X) 
+∞
= · Pt X [n] q n .
1 − t2 q n=0
Formula (1.41) will be important in Chapter 6 which deals with the relations
between the incidence Hilbert schemes of points and infinite dimensional Lie alge-
bras.
We refer to [Can, Che2, Che3] for generalizations from incidence Hilbert
schemes to nested Hilbert schemes of points on surfaces, and to [GSY1, GSY2] for
generalizations to nested Hilbert schemes of points and curves on surfaces which
are related to localized Donaldson-Thomas theory.
 CHAPTER 2

The nef cone and flip structure of (P2 )[n]

The Hilbert schemes of points on the complex projective plane P2 provide

excellent and simple examples illustrating some geometric aspects of the Hilbert
schemes of points on general surfaces. In this chapter, we will study the nef cone
and birational geometry of the Hilbert schemes (P2 )[n] . We will begin with the
description of certain curves which are contracted by the Hilbert-Chow morphism
(1.15) for a simply connected surface X. Then for the Hilbert scheme (P2 )[n] ,
we will describe its nef cone and the cone spanned by all the effective curves,
and study the curves having minimal degree with respect to a certain very ample
divisor. We will end with the construction of a flip structure on (P2 )[n] for n ≥ 3.
The main results in this chapter appeared in the announcement [LQZ], and have
been partially generalized in [QT]. We refer to [ABCH, LZ] for further birational
geometry of (P2 )[n] in connection with the Bridgeland stability.

2.1. Curves homologous to βn

In this section, for simplicity, X stands for a simply connected smooth projec-
tive surface. We will determine all the curves in the Hilbert scheme X [n] homologous
to βn defined in (1.34), and study their moduli space. The understanding of these
curves plays an important role in studying the extremal Gromov-Witten invariants
of the Hilbert scheme X [n] which will be dealt with in Chapters 14 and 15.
First of all, we determine bases of H2 (X [n] ) and H4n−2 (X [n] ). For a real-surface
C in X, define βC to be the homology class in H2 (X [n] ) represented by a curve of
the form:
(2.1) {x + η| x ∈ C}
where η ∈ X [n−1] is fixed such that C ∩ Supp(η) = ∅. Recall the notations Bn and
DC from (1.26) and (1.27) respectively.
Lemma 2.1. Let n ≥ 2 and X be a simply connected smooth projective surface.
Let C1 , . . . , Cs be real-surfaces in X such that their homology classes form a basis
of H2 (X). Then,
(i) a basis of H2 (X [n] ) consists of the homology classes
βn , βC1 , . . . , βCs ;
[n]
(ii) a basis of H4n−2 (X ) consists of the homology classes
Bn , DC1 , . . . , DCs .
1 , . . . , C
Proof. Let C s be real-surfaces in X such that
1 ∼ C1 , . . . , C
C s ∼ C
s ,

19
20 2. THE NEF CONE AND FLIP STRUCTURE OF (P2 )[n]

where the symbol “A1 ∼ A2 ” means that A1 and A2 are homologous. We further
1 , . . . , C
assume that C1 , . . . , Cs , C s are in general position. It follows that
(2.2) βC , D  = C i , C j
i Cj

for all i, j = 1, . . . , s. By (1.35), βn , Bn = −2. Also, it is clear that for all

j = 1, . . . , s, we have
(2.3) βn , DCj = 0.
Thus, the matrix consisting of the intersection pairings between βn , βC1 , . . . , βCs
and Bn , DC1 , . . . , DCs is nondegenerate. By Theorem 1.23, the second Betti number
of X [n] is equal to 1 + b2 (X) = 1 + s. It follows that βn , βC1 , . . . , βCs form a basis
of H2 (X [n] ) and Bn , DC1 , . . . , DCs form a basis of H4n−2 (X [n] ). 
Next, we will study the curves γ in the punctual Hilbert scheme Mn (x) ⊂ X [n]
such that γ ∼ βn . We will establish some notations, and prove three technical
lemmas. Throughout this section, we identify Mn (x) with the punctual Hilbert
scheme Mn (O) of C2 at the origin O. Let
R = OC2 ,O
be the local ring of C at the origin, and m = u, v be the maximal ideal of R
2

where u, v form a coordinate system of C2 at the origin. Let η ∈ Mn (O). It is

known from [Iar1] that Iη,O ⊃ mn and there is an embedding
(2.4) τ : Mn (O) → Grass(R/mn , n)
where R/mn is considered as a C-vector space of dimension n+1
2 , and τ maps
η ∈ Mn (O) to the quotient
R/mn → R/Iη,O = Oη,O → 0.
For simplicity, put
G = Grass(R/mn , n),
 
n+1
1 = .
2
Then over G, there are a universal quotient bundle Q, a universal subbundle S and
an exact sequence:
(2.5) 0 → S → (R/mn ) ⊗ OG → Q → 0.
It is known that det(Q) induces the Plücker embedding
(2.6) p : G → P((∧1 −n C1 )∗ ).
So if H is the hyperplane line bundle over P((∧1 −n C1 )∗ ), then det(Q) = p∗ H.
Lemma 2.2. Let n ≥ 2. Fix a point x ∈ X and a curve γ ⊂ Mn (x). Regard
γ ⊂ Mn (O) via the identification Mn (x) ∼ = Mn (O). Then γ ∼ βn if and only if
γ · Bn = −2 if and only if p ◦ τ (γ) is a line.
[n]
Proof. Consider the rank-n tautological bundle OX over X [n] . Following
the similar argument as in the proof of the Claim 2 in [BSG, Appendix], we can
show that OX |Mn (x) is a quotient of the trivial bundle (OMn (x) )⊕1 . The fiber
[n]

[n]
of OX |Mn (x) over η ∈ Mn (x) ⊂ X [n] is H 0 (X, Oη ) = Oη,x . Since the fiber of
∗
τ Q over η ∈ Mn (O) is Oη,O , we see from the universality of the quotient bundle
 2.1. CURVES HOMOLOGOUS TO βn 21

that OX |Mn (x) ∼

= τ ∗ Q. By Theorem 1.24 (i), c1 (OX ) = −Bn /2. Combining with
[n] [n]

det(Q) = p∗ H yields
[n]
(2.7) γ · Bn = −2γ · c1 (OX )
[n]
= −2γ · c1 (OX |Mn (x) )
= −2γ · c1 (τ ∗ Q)
= −2γ · τ ∗ p∗ c1 (H)
= −2(p ◦ τ )(γ) · c1 (H).
Next, since γ ⊂ Mn (x), we have γ · DC = 0 = βn · DC for every real-surface
C in X. By Lemma 2.1, γ ∼ βn if and only if γ · Bn = −2 = βn · Bn . By (2.7),
γ · Bn = −2 if and only if p ◦ τ (γ) is a line. 
Lemma 2.3. Let k = n(n − 1)/2, and L ⊂ G = Grass(R/mn , n) be a curve.
Then, L is mapped to a line by p if and only if L is of the form
(2.8) L = Span(f 1 , . . . , f k−1 , λf k + μf k+1 ) | λ, μ ∈ C, |λ| + |μ| = 0
where f1 , . . . , fk+1 ∈ R are polynomials such that f 1 , . . . , f k+1 are linearly indepen-
dent in the vector space R/mn ∼ = Cn+k .
Proof. From the definition of p, it is clear that if L is of the form (2.8), then
p(L) is a line. Conversely, assume that p(L) is a line. We see from det(Q) = p∗ H
that det(Q|L ) ∼
= OL (1). Restricting (2.5) to L, we obtain an exact sequence:
(2.9) 0 → S|L → (R/mn ) ⊗ OL → Q|L → 0.
So det(S|L ) ∼
= OL (−1). From the injection S|L → (R/mn ) ⊗ OL , we obtain

k
S|L ∼
= OL (ai )
i=1
k
with i=1 ai = −1 and ai ≤ 0. Thus

S|L ∼
⊕(k−1)
= OL ⊕ OL (−1).
⊕(k−1)
Now the injection OL → (R/mn ) ⊗ OL gives rise to linearly independent
elements f 1 , . . . , f k−1 in R/mn . The injection OL (−1) → (R/mn ) ⊗ OL gives rise
to two linearly independent elements f k and f k+1 in R/mn such that
f 1 , . . . , f k−1 , λf k + μf k+1
are linearly independent when |λ| + |μ| = 0. So L is of the form (2.8). 
Lemma 2.4. Let x ∈ X and η ∈ Mn+1 (x). Let ξ1 , ξ2 ∈ Mn (x) and ξ1 = ξ2 . If
ξ1 , ξ2 ⊂ η, then Iη,x ⊃ mnx where mx is the maximal ideal of OX,x .
Proof. We have Iη,x ⊂ Iξ1 ,x ∩ Iξ2 ,x ⊂ Iξ2 ,x and Iξ1 ,x ∩ Iξ2 ,x = Iξ2 ,x . Since the
colength of Iη,x is n + 1 and the colength of Iξ2 ,x is n, we obtain Iη,x = Iξ1 ,x ∩ Iξ2 ,x .
Since mnx ⊂ Iξ1 ,x and mnx ⊂ Iξ2 ,x , we get mnx ⊂ Iξ1 ,x ∩ Iξ2 ,x = Iη,x . 
Now we are able to characterize all the curves γ in the punctual Hilbert scheme
Mn (x) which is homologous to βn . By Theorem 1.31 (i), the fiber (ψn+1 )−1 (η, x)
over (η, x) ∈ Zn+1 ⊂ X [n+1] × X is isomorphic to P(ωηx ⊗ Cx ).
22 2. THE NEF CONE AND FLIP STRUCTURE OF (P2 )[n]

Proposition 2.5. Let n ≥ 2, and X be simply connected. Let x ∈ X, and γ

be a curve in Mn (x). Then, γ ∼ βn if and only if γ = fn+1 (C) where C is a line
in the fiber (ψn+1 )−1 (η, x) ∼
= P(ωη ⊗ Cx ) for some η ∈ Mn+1 (x).
Proof. (i) Fix the identification Mn (x) ∼
= Mn (O). Let C be a line in
−1 ∼ P(ωη ⊗ Cx )
(ψn+1 ) (η, x) =
for some η ∈ Mn+1 (x). By Lemma 2.4, we have Iη,x ⊃ mn . Choose polynomials
f1 , . . . , fk−1 ∈ Iη such that f 1 , . . . , f k−1 form a basis of the (k − 1)-dimensional
subspace Iη,x /mn ⊂ R/mn . Now, Iη,x = Cf1 + . . . + Cfk−1 + mn . So
Iη = Iη,x ∩ C[u, v] = Cf1 + . . . + Cfk−1 + u, v n .
Next, since P(ωη ⊗ Cx ) ∼
= P(Hom(Cx , Oη )∗ ), the line C in
(ψn+1 )−1 (η, x) ∼
= P(Hom(Cx , Oη )∗ )
is generated by two different points [1 ], [2 ] ∈ P(Hom(Cx , Oη )∗ ) (so that the two
morphisms 1 , 2 ∈ Hom(Cx , Oη ) = HomR (R/m, Oη ) are linearly independent).
Choose fk , fk+1 ∈ R such that [fk ] = 1 (1) and [fk+1 ] = 2 (1) in R/Iη,x . Note that
f 1 , . . . , f k+1 ∈ R/mn are linearly independent.
Let ξ ∈ γ = fn+1 (C), and assume that the point (ξ, η) ∈ C corresponds to
[λ1 + μ2 ] ∈ P(Hom(Cx , Oη )∗ ). Then we have the exact sequence:
λ 1 +μ 2
0 −→ Cx −→ Oη −→ Oξ −→ 0.
So
Iξ = Iη + C(λfk + μfk+1 )
= Cf1 + . . . + Cfk−1 + C(λfk + μfk+1 ) + u, v n .
Thus τ (γ) = L which is of the form (2.8). By Lemma 2.3 and Lemma 2.2, γ ∼ βn .
(ii) Conversely, assume γ ∼ βn . By Lemma 2.2 and Lemma 2.3, τ (γ) = L which
is of the form (2.8). Let u : R/mn → R/mn be the map defined by multiplication
with u. Let
ϕu : R/mn ⊗ OL → R/mn ⊗ OL
be the induced morphism, and consider the restriction of ϕu to the subbundle
S|L ∼
⊕(k−1)
= OL (−1) ⊕ OL in (2.9). Since L = τ (γ),
Span(f 1 , . . . , f k−1 , λf k + μf k+1 )
is an ideal of R/mn whenever |λ| + |μ| = 0. So we see from the proof of Lemma 2.3
⊕(k−1)
that ϕu maps OL (−1) ⊕ OL to itself. Since Hom(OL , OL (−1)) = 0, ϕu maps
⊕(k−1) ⊕(k−1)
OL to OL . The same statement holds when we replace u by v. Hence,
Span(f 1 , . . . , f k−1 ) is an ideal of R/mn , and Cf1 + . . . + Cfk−1 + mn is an ideal of
R with colength n + 1. Let η ∈ Mn+1 (x) be defined by the ideal
Iη = Cf1 + . . . + Cfk−1 + u, v n .
Since Hom(OL (−1), OL (−1)) = C · IdOL (−1) , we have ϕu |OL (−1) = c · IdOL (−1)
for some c ∈ C. Then (ϕu |OL (−1) )n = cn · IdOL (−1) . However, (ϕu )n = 0 since
⊕(k−1) ⊕(k−1)
un ∈ mn . So c = 0, and ϕu maps OL (−1) ⊕ OL to OL . In particular,
uf k , uf k+1 ∈ Span(f 1 , . . . , f k−1 ). Hence ufk , ufk+1 ∈ Iη . Similarly, vfk , vfk+1 ∈
Iη .
 2.1. CURVES HOMOLOGOUS TO βn 23

Define two R-morphisms R → Oη by sending 1 ∈ R to [fk ], [fk+1 ] ∈ Oη respec-

tively. They induce two R-morphisms 1 , 2 : Cx = R/m → Oη since ufk , ufk+1 ,
vfk , vfk+1 ∈ Iη . Since f 1 , . . . , f k−1 , f k , f k+1 ∈ R/mn are linearly independent, so
are the two morphisms 1 , 2 ∈ Hom(Cx , Oη ). Let C be the line in
(ψn+1 )−1 (η, x) ∼ = P(Hom(Cx , Oη )∗ )
generated by the points [1 ], [2 ] ∈ P(Hom(Cx , Oη )∗ ). As in the last paragraph in
part (i) above, the points in fn+1 (C) correspond to the ideals
Cf1 + . . . + Cfk−1 + C(λfk + μfk+1 ) + u, v n , λ, μ ∈ C
with |λ| + |μ| = 0. These same ideals define the points in γ since τ (γ) = L is given
by (2.8). So γ = fn+1 (C). 
Next, we study an arbitrary curve γ ⊂ X [n] homologous to βn .
Lemma 2.6.
(i) A curve γ ⊂ X [n] homologous to βn is a smooth rational curve;
(ii) A curve γ ⊂ X [n] is homologous to βn if and only if γ = γk + ξ where
2 ≤ k ≤ n, γk is a curve in the punctual Hilbert scheme Mk (x) for some
x ∈ X, γk ∼ βk in X [k] , and ξ ∈ X [n−k] is a fixed point with x ∈ Supp(ξ).
Proof. (i) By Remark 1.28, (DC − Bn /2) is very ample for some divisor C on
X. Since γ · (DC − Bn /2) = βn · (DC − Bn /2) = 1, γ is a smooth rational curve.
(ii) First of all, assume γ ∼ βn . Since γ is irreducible and γ · DC = βn ·
DC= 0, ρn (γ) is a point in X (n) . Let ρn (γ) = n1 x1 + . . . + nm xm where ni ≥
m
1, i=1 ni = n, and x1 , . . . , xm are distinct points. Then γ is contained in the
product i=1 Mni (xi ) ⊂ X [n] . Since Mni (xi ) ∼
m
= Mni (O), H1 (M m ni (xi ); Z) = 0 by
the Theorem 1.1 (i) of [ES1]. By the Künneth formula, γ ∼ i=1 (pi (γ) + ξi ) in
m m
i=1 Mni (xi ), and ξi ∈ X
[n−ni ]
i=1 Mni (xi ), where pi is the i-th projection of
with Supp(ξi ) ∩ Supp pi (γ) = ∅. So

m
1 = (DC − Bn /2) · γ = (DC − Bn /2) · (pi (γ) + ξi ).
i=1

Since (DC − Bn /2) is very ample, there exists exactly one i such that pi (γ) is a
curve in X [ni ] and pj (γ) is a point in X [nj ] for all j = i. We may assume i = 1.
Let k = n1 and x = x1 . Then γ = γk + ξ where γk ⊂ Mk (x) and ξ ∈ X [n−k] is
fixed with x ∈/ Supp(ξ). It remains to prove γk ∼ βk in X [k] .
Claim. Let x̃ ∈ X be a fixed point, and α ⊂ X [n−1] be a curve such that x̃ ∈
/
Supp(ξ) for every ξ ∈ α. Then, (α + x̃) · Bn = α · Bn−1 .
Proof. Define
 = { (ξ, ξ + x̃) | ξ ∈ α} ⊂ X [n−1,n] .
α
Since x̃ ∈/ Supp(α), the restriction gn |α : α  → α + x̃ is an isomorphism. So
 = α + x̃. Similarly, (fn )∗ α
(gn )∗ α  = α. Let
En = {(ξ, η) ∈ X [n−1,n] | Supp(ξ) = Supp(η), ξ ⊂ η}.
Since x̃ ∈  ∩ En = ∅. By (1.40),
/ Supp(α), we obtain α
(gn )∗ Bn − (fn )∗ Bn−1 = 2En .
24 2. THE NEF CONE AND FLIP STRUCTURE OF (P2 )[n]

 · ((gn )∗ Bn − (fn )∗ Bn−1 ) = 0. Therefore, we obtain

So α
 · (gn )∗ Bn = α
 · Bn = α
(α + x̃) · Bn = (gn )∗ α  · (fn )∗ Bn−1 .
By the projection formula,
(α + x̃) · Bn = (fn )∗ α
 · Bn−1 = α · Bn−1 .
This proves the claim. 

By deforming ξ in X [n−k] , we can assume that ξ is x̃1 + . . . + x̃n−k where

x̃1 , . . . , x̃n−k are distinct points different from x. From the Claim, we get
(2.10) (γk + x̃1 + . . . + x̃n−k ) · Bn = (γk + x̃1 + · · · + x̃n−k−1 ) · Bn−1
= ...
= γ k · Bk .
So γk · Bk = γ · Bn = βn · Bn = −2. Since γk ⊂ Mk (x), γk ∼ βk by Lemma 2.2.
Conversely, suppose γ = γk + ξ where 2 ≤ k ≤ n, γk ⊂ Mk (x) for some x ∈ X,
γk ∼ βk in X [k] , and ξ ∈ X [n−k] is fixed with x ∈ Supp(ξ). We may assume that ξ
consists of (n − k) distinct points. By (2.10),
γ · Bn = γk · Bk = βk · Bk = −2.
Then the same argument as in the proof of Lemma 2.2 yields γ ∼ βn . 

Theorem 2.7. Let n ≥ 2, and X be a simply connected smooth projective

surface. Then, a curve γ in X [n] is homologous to βn if and only if γ = fn+1 (C)
where C is a line in the projective space (ψn+1 )−1 (η, x) for some (η, x) ∈ Zn+1 .
Moreover, in this case, (η, x) and C are uniquely determined by γ.
Proof. (i) First, assume γ ∼ βn . By Lemma 2.6 (ii), we know that γ =
γk + ξ where γk ⊂ Mk (x) for some x ∈ X, ξ ∈ X [n−k] is fixed with x ∈ Supp(ξ),
and γk ∼ βk in X [k] . By Proposition 2.5, γk = fk+1 (C) where C  is a line in
−1
(ψk+1 ) (ηk+1 , x) for some ηk+1 ∈ Mk+1 (x). Since Supp(ηk+1 ) = {x} ⊂ Supp(ξ),
we can define η = ηk+1 + ξ ∈ X [n+1] . Then, (η, x) ∈ Zn+1 . By the definition of
ψn+1 , we have
(ψn+1 )−1 (η, x) = {(ζ, η) ∈ X [n,n+1] | ζ ⊂ η and Supp(Iζ /Iη ) = x}
= {(ζk + ξ, ηk+1 + ξ) ∈ X [n,n+1] | ζk ⊂ ηk+1 }
∼
= {(ζk , ηk+1 ) ∈ X [k,k+1] | ζk ⊂ ηk+1 }
= (ψk+1 )−1 (ηk+1 , x).
Put
C = 
{(ζk + ξ, ηk+1 + ξ)|(ζk , ηk+1 ) ∈ C}
= 
{(ζk + ξ, η)|(ζk , ηk+1 ) ∈ C},
 in
i.e., let C be the line in (ψn+1 )−1 (η, x) corresponding to the line C
(ψk+1 )−1 (ηk+1 , x)
via the above isomorphism (ψn+1 )−1 (η, x) ∼
= (ψk+1 )−1 (ηk+1 , x). Then, we see that
 + ξ = fn+1 (C).
γ = γk + ξ = fk+1 (C)
 2.1. CURVES HOMOLOGOUS TO βn 25

Conversely, suppose γ = fn+1 (C) where C is a line in (ψn+1 )−1 (η, x) for some
(η, x) ∈ Zn+1 . Assume (ηx ) = k + 1. Let ηk+1 = ηx ∈ X [k+1] . Reversing the
argument in the preceding paragraph and using Lemma 2.6 (ii), we get γ ∼ βn .
(ii) To show the uniqueness of (η, x) and C, let ξ1 , ξ2 ∈ γ with ξ1 = ξ2 . Then,
ξ1 , ξ2 ⊂ η. Since (η) = 1 + (ξ1 ) = 1 + (ξ2 ), we get Iη = Iξ1 ∩ Iξ2 . So η and hence
the point x are uniquely determined by γ. Since
fn+1 |(ψn+1 )−1 (η,x) : (ψn+1 )−1 (η, x) → fn+1 ((ψn+1 )−1 (η, x))
is an isomorphism, C is also uniquely determined by γ. 
Our next goal is to give a global description of the moduli space M(βn ) of all
the curves in X [n] homologous to βn . Note that M(βn ) is the union of certain
irreducible components in the Hilbert scheme of curves in X [n] . We will use the
concept of the relative Grassmannian scheme from [Sim].
Theorem 2.8. Let n ≥ 2, X be a simply connected smooth projective surface,
and ωZn+1 be the dualizing sheaf of Zn+1 ⊂ X [n+1] × X. Then, there exists a
bijective morphism from the relative Grassmannian Grass(ωZn+1 , 2) over Zn+1 to
the moduli space M(βn ) of all the curves in X [n] homologous to βn .
 = Grass(ωZ , 2) for simplicity. Define
Proof. Put G n+1

Γ = [C], (ξ, η) ∈ G × X [n,n+1] | C is a line in (ψn+1 )−1 (η, x)

and (ξ, η) ∈ C with Supp(Iξ /Iη ) = {x} .
By Theorem 1.31 (i), P(ωZn+1 ) ∼
= X [n,n+1] over Zn+1 . So Γ is the universal sub-
  Consider
scheme in G ×Zn+1 P(ωZn+1 ) which is flat over G.
 ×fn+1
IdG
 ×Z
α:G  × X [n,n+1]
X [n,n+1] ⊂ G −→  × X [n] .
G
n+1

 over (η, x) ∈ Zn+1 is G|

Note that the fiber of G  (η,x) = Grass(ωη ⊗ Cx , 2). So
x

(2.11) Γ|(η,x)  (η,x) × P(ωZ )|(η,x)

⊂ G| n+1

= Grass(ωηx ⊗ Cx , 2) × P(ωηx ⊗ Cx )
= Grass(ωηx ⊗ Cx , 2) × (ψn+1 )−1 (η, x).
Since fn+1 maps (ψn+1 )−1 (η, x) isomorphically to fn+1 ((ψn+1 )−1 (η, x)), we see
that α|Γ : Γ → α(Γ) is bijective. Moreover, assuming that q ∈ G  is mapped to

(η, x) via the natural projection G → Zn+1 and that q corresponds to the line
{q} × C = Γ|q ⊂ {q} × P(ωηx ⊗ Cx ) = {q} × (ψn+1 )−1 (η, x)
via the universal subscheme Γ ⊂ G ×Z P(ωZn+1 ), then we see from (2.11) that
n+1

the same point q ∈ G also corresponds to the curve
α(Γ)|q = α(Γ|q ) = α({q}×C) = {q}×fn+1 (C) ⊂ {q} × fn+1 ((ψn+1 )−1 (η, x))
via the subscheme α(Γ) ⊂ G  × X [n] . By Theorem 2.7, we conclude that via the

subscheme α(Γ) ⊂ G × X [n]  the relative Grassmannian G
which is flat over G, 
[n]
parameterizes curves in X homologous to βn . By the universal property of the
Hilbert scheme of curves, we obtain a morphism Λ : G  → M(βn ). Moreover, we see
from Theorem 2.7 that the morphism Λ is bijective. 
26 2. THE NEF CONE AND FLIP STRUCTURE OF (P2 )[n]

Next, we study an open subset in Grass(ωZn+1 , 2). Let x ∈ X and

ξ = x1 + . . . + xn−2 ∈ X [n−2]
be fixed such that x, x1 , . . . , xn−2 are distinct points. Then the curve M2 (x) + ξ is
homologous to βn . The set of all such curves has dimension
#(moduli of x) + #(moduli of ξ) = 2 + 2(n − 2) = 2n − 2.
We now give a more detailed geometric description of this set as follows. Note that
m2x is the only colength-3 ideal in OX,x whose minimal number of generators is 3.
Let θ(x) be the unique length-3 0-dimensional subscheme of X corresponding to
m2x . Let
Δn−1 = {(x, x1 , . . . , xn−2 ) ∈ X n−1 |x, x1 , . . . , xn−2 are not distinct}.
Define a map Ψ : X n−1 − Δn−1 → X [n+1] by putting

n−2
Ψ(x, x1 , . . . , xn−2 ) = θ(x) + xi .
i=1

Let θ(x) + ξ ∈ Im(Ψ) where ξ denotes x1 + . . . + xn−2 . Then, (πn+1 )−1 (θ(x) + ξ)
consists of (n − 1) distinct points in Zn+1 :
(θ(x) + ξ, x), (θ(x) + ξ, xi ), 1 ≤ i ≤ n − 2.
Note that
Grass(ωZn+1 , 2)|(θ(x)+ξ,x) = Grass(ωθ(x) ⊗ Cx , 2) ∼
= Grass(C2 , 2) = pt.
In fact, Grass(ωZn+1 , 2)|(θ(x)+ξ,x) is the point in Grass(ωZn+1 , 2) corresponding to
the curve M2 (x) + ξ in X [n] . Also,
Grass(ωZ , 2)|(θ(x)+ξ,x ) ∼
n+1 = Grass(C, 2) = ∅.
i

So Grass(ωZn+1 , 2)|(πn+1 )−1 (Im(Ψ)) is exactly the subset of Grass(ωZn+1 , 2) described

at the beginning of this paragraph. This proves (i) of the following proposition.
Proposition 2.9. (i) The open subset Grass(ωZn+1 , 2)|(πn+1 )−1 (Im(Ψ)) of
Grass(ωZn+1 , 2) has dimension (2n − 2);
(ii) Let γ = M2 (x) + x1 + . . . + xn−2 where x, x1 , . . . , xn−2 are distinct points
in X, and let Nγ⊂X [n] be the normal bundle of γ in X [n] . Then,
Nγ⊂X [n] ∼
= Oγ⊕(2n−2) ⊕ Oγ (−2),
TX [n] |γ ∼
= Oγ⊕(2n−2) ⊕ Oγ (−2) ⊕ Oγ (2).
Proof. To prove (ii), note that in an analytic open neighborhood of γ, X [n]
splits into X [2] × X [n−2] . So
Nγ⊂X [n] ∼
= N ⊕ O⊕(2n−4)
γ

where N is isomorphic to the normal bundle N of M2 (x) in X [2] via γ ∼

= M2 (x).
 ∼ ⊕2
Since N = OM2 (x) ⊕ OM2 (x) (−2), we obtain

Nγ⊂X [n] ∼
= Oγ⊕(2n−2) ⊕ Oγ (−2).
Next, we have an exact sequence
0 → Tγ → TX [n] |γ → Nγ⊂X [n] → 0.
 2.1. CURVES HOMOLOGOUS TO βn 27

Since Tγ = Oγ (2), we obtain the exact sequence

0 → Oγ (2) → TX [n] |γ → Oγ⊕(2n−2) ⊕ Oγ (−2) → 0
which must split. Therefore, we conclude that
TX [n] |γ ∼
= O⊕(2n−2) ⊕ Oγ (−2) ⊕ Oγ (2).
γ
This completes the proof of the proposition. 
By Proposition 2.9 (i) and Theorem 2.8, the dimension of the moduli space
M(βn ) is at least (2n − 2), one more than the expected dimension
c1 (TX [n] ) · βn + dim X [n] − dim Aut(P1 ) = (2n − 3).
Therefore the moduli space of curves in X [n] homologous to βn is obstructed. The
obstruction may be understood via Proposition 2.9 (ii). In Lemma 15.4 below,
we will use Proposition 2.9 (ii) to determine the obstruction sheaf of the 1-point
genus-0 extremal Gromov-Witten theory of the Hilbert scheme X [n] .
Finally, we construct some special curves in the punctual Hilbert schemes.
Lemma 2.10. Fix n ≥ 2. Let Mn (O) be the punctual Hilbert scheme at the
origin O ∈ C2 , and u, v be the coordinates of C2 . Then,
H2 (Mn (O); Z) ∼
= Z.
Moreover, a generator of H2 (Mn (O); Z) is given by
(2.12) σn = {(λu + μv n−1 , u2 , uv, v n ) | λ, μ ∈ C with |λ| + |μ| = 0}.
Proof. The first statement follows from [ES1, Theorem 1.1]. To prove the
second statement, take a C∗ -action on C2 given by
t · (u, v) = (t−α u, t−β v)
with β  α. For ξ ∈ Mn (O), we use the ideal Iξ ⊂ C[u, v] to represent ξ. Then
the C∗ -invariant ideal in C[u, v] corresponding to a generator σn of H2 (Mn (O); Z)
is (v n−1 , uv, u2 ). Therefore σn is the closure of the cell
 
I ∈ C[u, v] | (C[u, v]/I) = n, lim (t · I) = (v n−1 , uv, u2 )
t→0
= {(v n−1 + au, uv, u2 ) | a ∈ C}
∼
= C.
Finally, notice that if a = 0, then (v n−1 + au, uv, u2 ) = (v n−1 + au, v n ). So letting
a → ∞, we see that the ideal (u, v n ) is also contained in σn . Thus,
σn = {(v n−1 + au, uv, u2 ) | a ∈ C} ∪ {(u, v n )}
which is the same as {(λu + μv n−1 , u2 , uv, v n ) | λ, μ ∈ C with |λ| + |μ| = 0}. 
Proposition 2.11. Identify Mn (x) with Mn (O), and regard σn as a curve in
Mn (x) ⊂ X [n] . Then as a curve in X [n] , σn is homologous to βn .
Proof. By Lemma 2.2, it suffices to show that the image (p ◦ τ )(σn ) is a line.
Fix a basis for the C-vector space R/mn :
1, u, u2 , u v, u3 , u2 v, u v 2 , . . . , un−1 , un−2 v, . . . , u v n−2 , v, . . . , v n−1 .
Note the special ordering of this basis. Recall from (2.12) that for any η ∈ σn ⊂
Mn (O),
Iη,0 = (λu + μv n−1 , u2 , uv, v n )
28 2. THE NEF CONE AND FLIP STRUCTURE OF (P2 )[n]

for some λ, μ ∈ C with |λ| + |μ| = 0. So a basis for the subspace Iη,0 /mn ⊂ R/mn
can be chosen as
λu + μv n−1 , u2 , u v, u3 , u2 v, u v 2 , . . . , un−1 , un−2 v, . . . , u v n−2 ,
and the matrix representation of Iη,0 /mn is given by the n2 × n+1
2 -matrix:
⎡ ⎤
0 λ 0 ... 0 0 ... 0 μ
⎢ 0 0 1 ... 0 0 ... 0 0 ⎥
⎢ ⎥
⎢ .. .. .. .. .. .. .. .. ⎥ .
⎣ . . . . . . ... . . ⎦
0 0 0 ... 1 0 ... 0 0
Thus, (p ◦ τ )(η) = [0, . . . , 0, λ, 0, . . . , 0, μ, 0, . . . , 0] where the positions of λ and μ
are independent of η ∈ σn . So the image (p ◦ τ )(σn ) is a line. 
Note that the flat limits of the elements (λu + v, v n ), λ ∈ C∗ in Mn (O) as
λ → 0 and λ → ∞ are equal to (v, un ) and (u, v n ) respectively. So in the punctual
Hilbert scheme Mn (O), we have the projective curve:
(2.13) σ̃n = {(λu + v, v n ) | λ ∈ C∗ } ∪ {(v, un ), (u, v n )}.
n
Lemma 2.12. As a curve in X [n] , σ̃n is homologous to 2 βn .
Proof. It suffices to show that σ̃n is homologous to n2 σn in H2 (Mn (O); Z).
By (2.13), if η ∈ σ̃n − {(v, un ), (u, v n )}, then a basis for Iη,0 /mn ⊂ R/mn consists
of
λu + v, λu2 + u v, λu v + v 2 , . . . ,
n−1 n−2
λu +u v, λun−2 v + un−3 v 2 , . . . , λu v n−2 + v n−1 .
As in the proof of Lemma 2.11, we see that the degree of
" #
(p ◦ τ ) σ̃n − {(v, un ), (u, v n )}

is n2 . So (p ◦ τ )(σ̃n ) has degree n2 . By Lemma 2.10, there exists an integer d

such that σ̃n ∼ dσn in H2 (Mn (O); Z). Since (p ◦ τ )(σn ) is a line, d = n2 . 

2.2. The nef cone of (P2 )[n]

In this section, we will determine the nef cone of the Hilbert scheme (P2 )[n]
and the cone NE (P2 )[n] spanned by all the effective curves in (P2 )[n] . We remark
that as a general fact, these two cones are dual to each other.
First of all, recall the concept of k-very ampleness from Definition 1.26 (iii).
This concept will be applied to produce nef divisors and very ample divisors on the
Hilbert scheme (P2 )[n] .
Lemma 2.13. Let X = P2 . Then the line bundle OX (n) is n-very ample.
Proof. Let be a line on P2 , C = (n + 3) , and k = n. Then, KX + C = n
and C · C = (n + 3)2 > 4n + 5. Suppose OX (n) = OX (KX + C) is not n-very
ample. By Theorem 1.27 (i), there exists an effective divisor D on X such that
C · D − n − 1 ≤ D · D < C · D/2 < n + 1.
Let D ∼ d . Then, d ≥ 0 and
(n + 3)d − n − 1 ≤ d2 < (n + 3)d/2 < n + 1.
 2.2. THE NEF CONE OF (P2 )[n] 29

Since d2 < (n + 3)d/2, d > 0. Now, (n + 3)d/2 < (n + 1) implies d < 2. So d = 1

which contradicts (n + 3)d − n − 1 ≤ d2 . Hence, OX (n) is n-very ample. 
Note that for k ≥ 1, k-very ampleness implies (k−1)-very ampleness. Therefore,
by Lemma 2.13, we get the following two morphisms:
ϕn−1 OX (n − 1) : X [n] → Grass H 0 X, OX (n − 1) , n ,
ϕn−1 OX (n) : X [n] → Grass H 0 X, OX (n) , n .
Put
(2.14) ϕ1 = ϕn−1 OX (n − 1) : X [n] → Grass H 0 X, OX (n − 1) , n ,
(2.15) ϕ2 = ϕn−1 OX (n) : X [n] → Grass H 0 X, OX (n) , n .
For i = 1 or 2, let
 
n+i
i = h0 X, OX (n − 2 + i) = ,
2
Ni = dim(∧i −n Ci ) − 1, and
pi : Grass(Ci , n) → PNi
be the Plücker embedding. Let
Φi = pi ◦ ϕi : X [n] → PNi ,
and Hi be the hyperplane line bundle over PNi . By (1.32),
(Φ1 )∗ H1 = OX [n] (n − 1)D − Bn /2 ,
(2.16) (Φ2 )∗ H2 = OX [n] (nD − Bn /2).
Lemma 2.14. Let X = P2 . Then the morphism ϕ2 in ( 2.15) is an embedding
of the Hilbert scheme X [n] , and the divisor (nD − Bn /2) is very ample.
Proof. By Lemma 2.13, OX (n) is n-very ample. So by Theorem 1.27 (ii), ϕ2
is an embedding. By (2.16), (nD − Bn /2) is very ample. 
Next, we study ((n − 1)D − Bn /2). We begin with the construction of a curve
which lies on the boundary of the cone NE (P2 )[n] . Let
ξ = x1 + . . . + xn−1 ∈ X [n−1]
where x1 , . . . , xn−1 are distinct points in the line ⊂ X. Let ( + ξ) be the closure
of − Supp(ξ) + ξ in X [n] . Alternatively, consider:
˜ξ ⊂ ξ
X ⊂ X [n−1,n]
gn
→ X [n]
⏐ ⏐ ⏐
⏐ ⏐ ⏐
(2.17) % % %φn
ξ× ⊂ ξ×X ⊂ X [n−1]
×X
where ˜ξ and Xξ are the strict transforms of ξ × and ξ ×X in X [n−1,n] respectively.
Since φn is the blowup of X [n−1] × X along Zn−1 , X ξ is isomorphic to the blowup
∼
of ξ × X = X at x1 , . . . , xn−1 . Let Ei be the exceptional divisor in X ξ over xi .
Then,

n−1
(φn |Xξ )∗ (ξ × ) = ˜ξ + Ei
i=1
30 2. THE NEF CONE AND FLIP STRUCTURE OF (P2 )[n]

ξ ). Notice that gn ( ˜ξ ) = ( + ξ) and

in the Chow group A1 (X
gn (Ei ) = M2 (xi ) + x1 . . . + xi−1 + xi+1 + . . . + xn−1 .
ξ → gn (X
In fact, since gn |Xξ : X ξ ) is an isomorphism, we have

(2.18) (gn |Xξ )∗ ( ˜ξ ) = ( + ξ) and (gn |Xξ )∗ (Ei ) = βn

in A1 (X [n] ).
Lemma 2.15. With the above notations, then
( + ξ) = β − (n − 1)βn
[n]
in A1 (X ).
 
Proof. Choose another line such that Supp(ξ) ∩ = ∅. Then in A1 (X [n] ),
(gn |Xξ )∗ (φn |Xξ )∗ (ξ × ) = (gn |Xξ )∗ (φn |Xξ )∗ (ξ ×  ) = (  + ξ) = β = β ,
n−1
(gn |Xξ )∗ (φn |Xξ )∗ (ξ × ) = (gn |Xξ )∗ ( ˜ξ + i=1 Ei ) = ( + ξ) + (n − 1)βn
where we have used (2.18). So ( + ξ) = β − (n − 1)βn in A1 (X [n] ). 
Proposition 2.16. Let X = P2 , and ϕ1 be the morphism defined in ( 2.14).
(i) If n ≥ 2, then the divisor ((n − 1)D − Bn /2) on X [n] is nef but not ample.
(ii) If n = 2, then
ϕ1 : X [2] → Grass H 0 X, OX (1) , 2 = (P2 )∗
is isomorphic to the relative Hilbert scheme Hilb2 (P((T(P2 )∗ )∗ )/(P2 )∗ ) over
the dual space (P2 )∗ .
Proof. (i) We have the following intersection numbers:
(2.19) β · D = 1, β · (−Bn /2) = 0, βn · D = 0, βn · (−Bn /2) = 1.
It follows that
(n − 1)D − Bn /2 · (β − (n − 1)βn ) = 0.
By Lemma 2.15, (β − (n − 1)βn ) is rationally equivalent to an effective curve. So
((n − 1)D − Bn /2) is not ample. By (2.16), the divisor ((n − 1)D − Bn /2) is nef.
(ii) When n = 2, we have
ϕ1 = ϕ1 OX (1) : X [2] → Grass H 0 X, OX (1) , 2 = (P2 )∗ .
An element ξ ∈ X [2] determines a unique line ξ in X and ϕ1 (ξ) = ξ ∈ (P2 )∗ . So
ϕ1 is a surjection whose fiber over a line ∈ (P2 )∗ is [2] ∼
= (2) ∼ = P2 . Note that
{(x, ) ∈ P2 × (P2 )∗ | x ∈ } ∼
= P((T(P2 )∗ )∗ ).
So ϕ1 : X [2] → (P2 )∗ is isomorphic to the relative Hilbert scheme
Hilb2 (P((T(P2 )∗ )∗ )/(P2 )∗ )
over (P2 )∗ . 
Theorem 2.17. Let n ≥ 2 and be a line in X = P2 . Let NE(X [n] ) be the
cone spanned by all the effective curves on X [n] .
(i) The nef cone of X [n] is the cone spanned by D and ((n − 1)D − Bn /2);
(ii) The cone NE(X [n] ) is spanned by βn and (β − (n − 1)βn );
 2.2. THE NEF CONE OF (P2 )[n] 31

(iii) Let β be the homology class of an effective curve in X [n] . Then,

β · (nD − Bn /2) = 1

if and only if either β = βn or β = (β − (n − 1)βn ).

Proof. (i) Note that D is nef and big, but not ample. By Proposition 2.16,
((n − 1)D − Bn /2) is also nef but not ample. By Theorem 1.24 (iii), (−Bn /2) and
D form a basis for A2n−1 (X [n] ) ∼
= H4n−2 (X [n] ; Z). So the nef cone of the Hilbert
scheme X is the cone spanned by D and ((n − 1)D − Bn /2).
[n]

(ii) Since (−Bn /2) and D form a basis for H4n−2 (X [n] ; Z), we see from (2.19)
and the Poincaré duality that {β , βn } is a basis for H2 (X [n] ; Z). By Lemma 2.15,
both βn and (β − (n − 1)βn ) are numerical equivalence classes of effective curves.
By (2.19) again,

βn · D = 0,
(β − (n − 1)βn ) · D = 1,
βn · ((n − 1)D − Bn /2) = 1,
(β − (n − 1)βn ) · ((n − 1)D − Bn /2) = 0.

It follows from (i) that the cone NE(X [n] ) is spanned by βn and (β − (n − 1)βn ).
(iii) Let β = aβn + b(β − (n − 1)βn ). Intersecting β with the nef divisors
D and ((n − 1)D − Bn /2), we see that a and b are nonnegative integers. Since
(nD − Bn /2) is very ample and

βn · (nD − Bn /2) = (β − (n − 1)βn ) · (nD − Bn /2) = 1,

we conclude that β ·(nD −Bn /2) = 1 if and only if β = βn or β = (β −(n−1)βn ). 

Theorem 2.17 can be generalized to the following, and we refer to [QT] for
further details.

Theorem 2.18. Let n ≥ 2, and let X be a smooth projective surface X sat-

isfying h1 (X, OX ) = 0. Assume that the nef cone of X is the span of the di-
visors F1 , . . . , Ft , and the cone NE(X) is the span of the curves C1 , . . . , Ct with
Fi · Cj = δi,j for all i and j. Assume further that


t
OX (n − 1) Fi
i=1

is an (n − 1)-very ample line bundle. Then,

(i) the nef cone of the Hilbert scheme X [n] is spanned by


t
(2.20) DF1 , ..., DFt , (n − 1) DFi − Bn /2;
i=1

(ii) the cone NE(X [n] ) is spanned by the classes

(2.21) βC1 − (n − 1)βn , ..., βCt − (n − 1)βn , βn .

32 2. THE NEF CONE AND FLIP STRUCTURE OF (P2 )[n]

2.3. Curves homologous to β − (n − 1)βn

In this section, we will continue to take X = P2 . We will characterize all the
curves in X [n] homologous to the class β − (n − 1)βn . In addition, we will compute
the normal bundles of these curves in X [n] , and describe their moduli space.
First of all, we describe some curves in X [n] homologous to β − (n − 1)βn .
Lemma 2.19. Let C be a line in X = P2 , and consider C (n) = C [n] ⊂ X [n] .
(i) Every line in C (n) ∼ = Pn is homologous to β − (n − 1)βn ;
(ii) OX [n] (D )|C (n) = OC (n) (1).
Proof. (i) Let x1 , . . . , xn−1 ∈ C be distinct, and put γ = C +(x1 +. . .+xn−1 ).
Then γ is a line in the projective space C (n) ∼ = Pn . By Lemma 2.15, γ ∼ β −
(n) ∼ n
(n − 1)βn . So every line in C = P is homologous to β − (n − 1)βn .
(ii) Follows from γ · D |C (n) = γ · D = (β − (n − 1)βn ) · D = 1. 
Next we will prove the converse of Lemma 2.19. Let γ be a curve in X [n]
homologous to β − (n − 1)βn . Then, we have
γ · (nD − Bn /2) = (β − (n − 1)βn ) · (nD − Bn /2) = 1.
By Lemma 2.14, (nD − Bn /2) is very ample. So γ is a smooth rational curve.
Let p1 be the projection from X n to the first factor. Let Cγ be the union of all
the curves in p1 υn−1 ρn (γ) where υn : X n → X (n) is the quotient map. The
properties of Cγ will be studied in the next two lemmas.
Lemma 2.20. With the notations as above, Cγ is a line in X = P2 .
Proof. Suppose deg (Cγ ) ≥ 2. Take a point ξ ∈ γ and a smooth point x ∈ Cγ
such that x ∈ / Supp(ξ). Since x ∈ Cγ ⊂ p1 υn−1 ρn (γ) , there exists ξx ∈ γ such
that ρn (ξx ) = nx x + ηx where nx ≥ 1, ηx ∈ X (n−nx ) , and x ∈
/ Supp(ηx ). Choose
a line in X missing Supp(ηx ) ∪ Supp(ξ), passing through x, and intersecting Cγ
transversally. Then the intersection ∩ Cγ is a finite set. Since deg(Cγ ) ≥ 2, ∩ Cγ
contains one more point y = x. Hence there exists ξy ∈ γ with y ∈ Supp(ξy ). Thus
ξx , ξy ∈ γ ∩ D . Since y = x, y ∈ and misses Supp(ηx ), we have
y∈
/ {x} ∪ Supp(ηx ) = Supp(ξx ).
So ξx = ξy . Since Supp(ξ) ∩ = ∅, we have ξ ∈ D . Since ξ ∈ γ, γ is not contained
in D . So γ ∩ D is a finite set of points. Since ξx , ξy ∈ γ ∩ D and ξx = ξy , we
obtain γ · D ≥ 2, a contradiction to γ · D = 1. 
Lemma 2.21. With the notations as above, let pγ be a linear homogeneous
polynomial defining the line Cγ ⊂ X = P2 . Then, any degree-(n − 1) homogeneous
polynomial divisible by pγ must vanish at ξ for every ξ ∈ γ.
Proof. Let Vpγ be the subspace of H 0 X, OX (n − 1) consisting of all the
degree-(n − 1) homogeneous polynomials divisible by pγ . Recall from (2.14) that
ϕ1 sends ξ ∈ X [n] to the n(n − 1)/2-dimensional subspace H 0 (X, OX (n − 1) ⊗ Iξ )
of H 0 X, OX (n − 1) . Since
γ · (n − 1)D − Bn /2 = (β − (n − 1)βn ) · (n − 1)D − Bn /2 = 0,
γ is contracted by ϕ1 . So the subspaces H 0 (X, OX (n−1)⊗Iξ ) ⊂ H 0 X, OX (n−1)
are independent of ξ ∈ γ. Let Vγ = H 0 (X, OX (n − 1) ⊗ Iξ ) where ξ ∈ γ. It suffices
to prove that Vγ = Vpγ as subspaces of H 0 X, OX (n − 1) .
 2.3. CURVES HOMOLOGOUS TO β − (n − 1)βn 33

Let s ∈ Vγ . Then s vanishes at every ξ ∈ γ by definition. Since Cγ ⊂

p1 υn−1 ρn (γ) , s vanishes at every point in Cγ . So the polynomial s is divisi-
ble by pγ . Thus, s ∈ Vpγ , and Vγ ⊂ Vpγ . Since
dim Vpγ = h0 (X, OX (n − 2)) = n(n − 1)/2 = h0 (X, OX (n − 1) ⊗ Iξ ) = dim Vγ
where ξ ∈ γ, we have Vγ = Vpγ . 

Theorem 2.22. Let n ≥ 2, and be a line in X. Then, a curve γ ⊂ X [n] is

homologous to β − (n − 1)βn if and only if there is a line C in X such that γ is a
line in C [n] ⊂ X [n] . Moreover, C is uniquely determined by the curve γ.
Proof. By Lemma 2.19 (i), it remains to prove the “only if” part.
Let C = Cγ where Cγ is from Lemma 2.20. First, we claim γ ⊂ C [n] . Indeed,
let ξ ∈ γ. With the same notations as those in the proof of Lemma 2.21, we choose
s = pγ · q ∈ Vpγ where q ∈ H 0 X, OX (n − 2) and q doesn’t vanish at any point
in Supp(ξ). By Lemma 2.21, s vanishes at ξ. So pγ must vanish at ξ. Therefore,
ξ ⊂ Cγ = C. Hence, ξ ∈ C [n] = C (n) for every ξ ∈ γ. So γ ⊂ C [n] ⊂ X [n] . Next,
OX [n] (D )|C [n] = OC [n] (1)
by Lemma 2.19 (ii). So viewing γ as a curve in C [n] , we obtain
γ · c1 (OC [n] (1)) = γ · D = 1.
Thus γ is a line in C [n] .
Finally, the uniqueness of the line C follows from the observation that if ξ ∈ X [n]
and n ≥ 2, then ξ is contained in at most one line in X = P2 . 

In the following, we will study the normal bundle of a curve γ in X [n] homol-
ogous to β − (n − 1)βn . By Theorem 2.22, there exists a line C in X = P2 such
that γ is a line in C (n) = C [n] ⊂ X [n] . In particular,
Nγ⊂C (n) ∼
= Oγ (1)⊕(n−1) .
So we have the following exact sequence of normal bundles:
(2.22) 0 → Oγ (1)⊕(n−1) → Nγ⊂X [n] → NC (n) ⊂X [n] |γ → 0.

Proposition 2.23. Let n ≥ 2, and C and be lines in X = P2 . Let γ ⊂ X [n]

be a curve homologous to β − (n − 1)βn . Then,
∼ OX (1)[n] |C (n) ∼
(i) NC (n) ⊂X [n] = = O⊕2(n) ⊕ OC (n) (−1)⊕(n−2) ;
C
(ii) Nγ⊂X [n] = Oγ (1)⊕(n−1) ⊕ Oγ⊕2 ⊕ Oγ (−1)⊕(n−2) ;
(iii) the moduli space M(β − (n − 1)βn ) of all the curves in X [n] homolo-
gous to (β − (n − 1)βn ) is unobstructed, i.e., is smooth with the expected
dimension.
Proof. (i) The isomorphism
NC (n) ⊂X [n] ∼
= OX (1)[n] |C (n)
follows from (1.29). To prove the second isomorphism, let Zn be the universal
subscheme in C (n) × C ∼
= Pn × P1 . Then, we have
OX (1)[n] |C (n) ∼
= π̃n∗ q̃n∗ OP1 (1)|Zn
34 2. THE NEF CONE AND FLIP STRUCTURE OF (P2 )[n]

where π̃n∗ : Zn → Pn and q̃n : Pn × P1 → P1 are the projections. It is known that
Zn ⊂ Pn × P1 is defined by the equation
a0 U n + a1 U n−1 V + . . . + an V n = 0
where a0 , a1 , . . . , an and U, V are the homogeneous coordinates on Pn and P1 re-
spectively. So the line bundle OPn ×P1 (Zn ) is of type (1, n) in Pic(Pn × P1 ) ∼
= Z ⊕ Z.
Let p̃n : Pn × P1 → Pn be the projection. Applying p̃n∗ to the exact sequence
0 → q̃n∗ OP1 (1) ⊗ OPn ×P1 (−Zn ) → q̃n∗ OP1 (1) → q̃n∗ OP1 (1)|Zn → 0,
we obtain
⊕2 ∗ ⊕(n−2)
0 → OC n → OC (n) (−1)
(n) → π̃n∗ q̃n OP1 (1)|Z → 0.
Since this exact sequence splits, we conclude the second isomorphism.
(ii) Follows immediately from (2.22) and the isomorphisms in (i).
(iii) By (ii), H 1 (γ, Nγ⊂X [n] ) = 0 for any curve γ homologous to (β −(n−1)βn ).
By deformation theory, M(β − (n − 1)βn ) is unobstructed. 
In the next theorem, we further identify M(β − (n − 1)βn ) with a certain
Grassmannian bundle.
Theorem 2.24. Let n ≥ 2, X = P2 , and be a line in X. Then the moduli space
M(β − (n − 1)βn ) of curves in X [n] homologous to (β − (n − 1)βn ) is isomorphic
to the Grassmannian bundle Grass(Symn ((T(P2 )∗ )∗ ), 2) over (P2 )∗ .
Proof. First of all, we give a global description of the union Wn of all the
subsets C [n] ⊂ X [n] with C being lines in X = P2 . Consider
F = {(x, [C]) ∈ P2 × (P2 )∗ | x ∈ C} ⊂ P2 × (P2 )∗ .
We know that F ∼
= P((T(P2 )∗ )∗ ). Then we have
Hilbn (P((T(P2 )∗ )∗ )/(P2 )∗ ) ∼
= Hilbn (F/(P2 )∗ )
⊂ Hilbn (P2 × (P2 )∗ /(P2 )∗ )
∼
= X [n] × (P2 )∗
of relative Hilbert schemes over (P2 )∗ . Let
π : Hilbn (P((T(P2 )∗ )∗ )/(P2 )∗ ) → (P2 )∗
be the projection. Consider the map
α : Hilbn (P((T(P2 )∗ )∗ )/(P2 )∗ ) → X [n] × (P2 )∗ → X [n] .
Note that via the isomorphism
Hilbn (P((T(P2 )∗ )∗ )/(P2 )∗ ) ∼
= Hilbn (F/(P2 )∗ ),
the fiber of the map π over a point [C] ∈ (P2 )∗ is isomorphic to C [n] . So Im(α) =
Wn . Moreover, since n ≥ 2, an element ξ ∈ X [n] is contained in at most one line
C in X = P2 . So if α(ξ  ) = α(η  ) for ξ  , η  ∈ Hilbn (P((T(P2 )∗ )∗ )/(P2 )∗ ), then there
exists a unique line C such that α(ξ  ) = α(η  ) ⊂ C, i.e., ξ  and η  are contained in
the fiber π −1 ([C]) over the point [C] ∈ (P2 )∗ . Since α maps this fiber isomorphically
to C [n] ⊂ X [n] , we must have ξ  = η  . Also
∼ P(Symn ((T(P2 )∗ )∗ ))
Hilbn (P((T(P2 )∗ )∗ )/(P2 )∗ ) =
 2.4. A FLIP STRUCTURE ON (P2 )[n] WHEN n ≥ 3 35

by the Theorem 4.1.11 in [Got2]. Thus, α gives the isomorphism

(2.23) P(Symn ((T(P2 )∗ )∗ )) ∼
= Wn .
So the set of all the lines in C [n] ⊂ X [n] with C being lines in X = P2 is
parameterized by the Grassmannian bundle Grass(Symn ((T(P2 )∗ )∗ ), 2) over (P2 )∗ .
By Theorem 2.22, from the universal properties of the Hilbert schemes and the
Grassmannians, we obtain a bijective morphism
Λ : Grass(Symn ((T(P2 )∗ )∗ ), 2) → M(β − (n − 1)βn ).
Since both Grass(Symn ((T(P2 )∗ )∗ ), 2) and M(β − (n − 1)βn ) are smooth, the bijec-
tivity of Λ implies that Λ is an isomorphism. 
By Lemma 2.14 and (2.16), the morphism
Φ2 = p2 ◦ ϕ2 : X [n] → PN2
is an embedding associated to the complete linear system |nD − Bn /2|. By The-
orem 2.17 (iii), γ ⊂ X [n] is mapped to a line by Φ2 if and only if either γ ∼ βn or
γ ∼ (β − (n − 1)βn ). Therefore, regarding X [n] as a closed subvariety of PN2 , then
the Hilbert scheme of lines in X [n] is the disjoint union
&
M(βn ) M(β − (n − 1)βn ).

2.4. A flip structure on (P2 )[n] when n ≥ 3

Let X = P2 . This section is devoted to the birational geometry of the Hilbert
scheme (P2 )[n] . We will study the structure of the morphism ϕ1 defined in (2.14)
when n ≥ 3, and construct a flip for ϕ1 .
We recall from Proposition 2.16 (ii) that the structure of ϕ1 has been deter-
mined for n = 2. For n ≥ 3, we begin with a lemma.
Lemma 2.25. Let n ≥ 3, X = P2 , and Wn be the contraction locus of the
morphism ϕ1 : X [n] → ϕ1 (X [n] ) defined by ( 2.14). Then,
(i) a fiber of ϕ1 is either a point or the subset C (n) ⊂ X [n] for some line
C ⊂ X;
(ii) the restriction ϕ1 |Wn : Wn → ϕ1 (Wn ) is isomorphic to the projective
bundle P(Symn ((T(P2 )∗ )∗ )) over (P2 )∗ ;
(iii) N ∼= (ϕ1 |Wn )∗ M ⊗ O(−1) where N = NWn ⊂X [n] is the normal bundle of
Wn in X [n] , M is some vector bundle over ϕ1 (Wn ) ∼ = (P2 )∗ , and O(1)
stands for the tautological line bundle over the projective bundle ϕ1 |Wn :
Wn → ϕ1 (Wn ).
Proof. (i) Fix ξ ∈ X [n] . By definition,
ϕ1 (ξ) ∈ ϕ1 (X [n] ) ⊂ Grass H 0 X, OX (n − 1) , n
corresponds to the n(n − 1)/2-dimensional subspace H 0 X, Iξ ⊗ OX (n − 1) in
H 0 X, OX (n − 1) . So η ∈ (ϕ1 )−1 (ϕ1 (ξ)) if and only if
H 0 X, Iη ⊗ OX (n − 1) = H 0 X, Iξ ⊗ OX (n − 1) .
Let d ⊂ |(n − 1) | be the linear system corresponding to the subspace H 0 X, Iξ ⊗
OX (n−1) in H 0 X, OX (n−1) . Then, either d cuts out a 0-dimensional subscheme
ζ with ζ ⊃ ξ, or d has a fixed component C.
36 2. THE NEF CONE AND FLIP STRUCTURE OF (P2 )[n]

Assume that d cuts out a 0-dimensional subscheme ζ with ζ ⊃ ξ. Then,

H 0 X, Iζ ⊗ OX (n − 1) = H 0 X, Iξ ⊗ OX (n − 1) ,
and (ϕ1 )−1 (ϕ1 (ξ)) ⊂ {η ∈ X [n] | η ⊂ ζ}. Conversely, if η ∈ X [n] and η ⊂ ζ, then
H 0 X, Iη ⊗ OX (n − 1) ⊃ H 0 X, Iζ ⊗ OX (n − 1)
= H 0 X, Iξ ⊗ OX (n − 1) .
So
H 0 X, Iη ⊗ OX (n − 1) = H 0 X, Iξ ⊗ OX (n − 1)
since they have the same dimension. Thus η ∈ (ϕ1 )−1 (ϕ1 (ξ)), and
{η ∈ X [n] | η ⊂ ζ} ⊂ (ϕ1 )−1 (ϕ1 (ξ)).
Hence (ϕ1 )−1 (ϕ1 (ξ)) = {η ∈ X [n] | η ⊂ ζ}. Now we claim
(ϕ1 )−1 (ϕ1 (ξ)) = {ξ}.
Indeed, if (ϕ1 )−1 (ϕ1 (ξ)) contains more than one point, then there is an irreducible
projective curve Γ in (ϕ1 )−1 (ϕ1 (ξ)) = {η ∈ X [n] | η ⊂ ζ} since the closed subset
{η ∈ X [n] | η ⊂ ζ} of X [n] is known to be connected. Moreover, the projective
curve Γ is contracted to points by both the Hilbert-Chow morphism ρn and the
morphism ϕ1 . So Γ · D = 0, and Γ · Bn = 0 in view of (2.16). By Theorem 2.17 (i),
the intersection of Γ with every divisor on X [n] is zero. This is impossible. So
(ϕ1 )−1 (ϕ1 (ξ)) = {ξ}.
Next, assume that d has a fixed component C. Let C ∼ d where d ≥ 1. Then,
n(n − 1)/2 − 1 = dim(d) ≤ dim |(n − 1 − d) | = (n + 1 − d)(n − d)/2 − 1.
It follows that d = 1, C is a line in X = P2 , and ξ ⊂ C (i.e. ξ ∈ C (n) ⊂ X [n] ). So
(ϕ1 )−1 (ϕ1 (ξ)) ⊂ C (n) . On the other hand, letting p(C) be a linear homogeneous
polynomial defining the line C, we see that for every η ∈ C (n) ⊂ X [n] , the subspace
H 0 X, Iη ⊗ OX (n − 1) in H 0 X, OX (n − 1) consists of all the degree-(n − 1)
homogeneous polynomials divisible by p(C). In particular,
H 0 X, Iη ⊗ OX (n − 1) = H 0 X, Iξ ⊗ OX (n − 1) .
So C (n) ⊂ (ϕ1 )−1 (ϕ1 (ξ)). Hence (ϕ1 )−1 (ϕ1 (ξ)) = C (n) .
(ii) By (i), the contraction locus Wn is the union of all the subsets C (n) ⊂ X [n]
with C being lines in X = P2 . So (ii) follows from (2.23) and (i).
(iii) Fix a line C in X = P2 . By (ii), C (n) is a fiber in the fiber bundle ϕ1 |Wn :
⊕2
Wn → ϕ1 (Wn ). So we have NC (n) ⊂Wn = OC (n) . Thus by Proposition 2.23 (i), we

obtain an exact sequence of normal bundles:

⊕2 ⊕2 ⊕(n−2)
0 → OC (n) → OC (n) ⊕ OC (n) (−1) → (NWn ⊂X [n] )|C (n) = N |C (n) → 0.
It follows that N |C (n) ∼
= OC (n) (−1)⊕(n−2) for every line C in X = P2 . Since these
C , C being lines in X = P2 , are precisely the fibers of the projective bundle
(n)

ϕ1 |Wn : Wn → ϕ1 (Wn ), we conclude that N ∼ = (ϕ1 |Wn )∗ M ⊗ O(−1) for some

∼
rank-(n − 2) vector bundle M over ϕ1 (Wn ) = (P2 )∗ . 

Let n ≥ 3. By Lemma 2.25 (ii), we have

dim(Wn ) = (n + 2) ≤ dim(X [n] ) − 1.
 2.4. A FLIP STRUCTURE ON (P2 )[n] WHEN n ≥ 3 37

So ϕ1 : X [n] → ϕ1 (X [n] ) is a birational morphism. Since KX [n] = −3D where is

a line in X = P2 , (β − (n − 1)βn ) · KX [n] = −3 < 0. This says that
R+ · (β − (n − 1)βn )
is an extremal ray since (−KX [n] ) is nef and the Picard number of X [n] is 2. Thus
ϕ1 is the contraction of the ray R+ · (β − (n − 1)βn ) since
((n − 1)D − Bn /2) · (β − (n − 1)βn ) = 0.
Moreover, when n ≥ 4, dim(Wn ) ≤ dim(X [n] ) − 2, so ϕ1 is a small contraction.
Theorem 2.26. Let n ≥ 4 and X = P2 . Then, the flip for the small contraction
ϕ1 : X [n] → ϕ1 (X [n] ) defined via ( 2.14) exists.
Proof. We will explicitely construct a commutative diagram
(ϕ+ )−1 ◦ϕ1
X [n]  (X [n] )+
 ϕ1
ϕ+

ϕ1 (X [n] )

satisfying the Flip Conjecture 5-1-10 in [KMM], by proving that (X [n] )+ is smooth,
ϕ+ is a birational morphism, and K(X [n] )+ is ϕ+ -ample. The proof is a little long,
but may be divided into two steps. In step one, we construct two additional bira-
tional morphisms ϕ − and ϕ+ . In step two, we construct the birational morphism
ϕ+ .
Let
 :X
ϕ ' [n] → X [n]
−

' '
n ⊂ X
[n] [n]
be the blowing-up of X along the contraction locus Wn of ϕ1 , and W
[n]
be the exceptional divisor. Let N be the normal bundle of Wn in X . Then over
'
W ∗
n = P(N ), we have the tautological surjection

(2.24) − |W
(ϕ  n
)∗ (N ∗ ) → OW
 n
(1) → 0

where OW

'
(1) is the tautological line bundle over W ∗
n = P(N ). In view of
n
Lemma 2.25 (iii),
N∼ = (ϕ1 |W )∗ M ⊗ OW (−1)
n n

where M is some rank-(n − 2) vector bundle over ϕ1 (Wn ) = ∼ (P2 )∗ , and OW (1)
n
stands for the tautological line bundle over the projective bundle ϕ1 |Wn : Wn →
ϕ1 (Wn ). So from (2.24), we obtain a surjection:
(ϕ1 |Wn ◦ ϕ
− |W
 n
)∗ (M ∗ ) → (ϕ
− |W
 n
)∗ OWn (−1) ⊗ OW
 n
(1) → 0.

By the Proposition 7.12 in Chapter II of [Hart], the map

(ϕ1 |Wn ◦ ϕ
− |W
 ): '
W n → ϕ1 (Wn )
n

factors through the natural projection P(M ∗ ) → ϕ1 (Wn ). In fact, the induced
morphism ϕ :W' ∗
n → P(M ) is a P -bundle. So the fibers of
n

(ϕ1 |Wn ◦ ϕ
− |W
 ): '
W n → ϕ1 (Wn )
n
38 2. THE NEF CONE AND FLIP STRUCTURE OF (P2 )[n]

are naturally Pn × Pn−3 . Moreover, for the projective spaces Pn in the fibers Pn ×
Pn−3 , we have OX '
(W ∼
n )|Pn = OPn (−1), using the facts that OX

'
(W ∼
n )|Pn ×Pn−3 =
[n] [n]

O(a, −1) for some integer a and that

⊗ OPn (1) ∼
⊕(n−2)
OPn = N ∗ |Pn
( )
∼
= − |Pn ×Pn−3 )∗ OX
(ϕ (−W'n )| Pn ×Pn−3
[n]

∼
= OPn
⊕(n−2)
⊗ OPn (−a).

In other words, the restriction OX '

(−W n )|W
 is a tautological line bundle of
[n] n

:W
ϕ ' n → P(M ∗
). By the results in [Naka, FN], there exists a morphism

ϕ '
+ : X [n] → (X [n] )
+

such that (X [n] )+ is smooth and contains P(M ∗ ). Moreover, ϕ

+ |W
 n
 and ϕ
=ϕ + is
∗
the blowing-up of (X )+ along P(M ).
[n]

Now we claim that

− )∗ ((n − 1)D − Bn /2) = (ϕ

(ϕ + )∗ H+

for some divisor H+ on (X [n] )+ . Indeed, recall from (2.16) that ((n−1)D −Bn /2) =
(ϕ1 )∗ H for some very ample divisor H on ϕ1 (X [n] ). Consider the divisor

(2.25) − )∗ ((n − 1)D − Bn /2) = (ϕ1 ◦ ϕ

(ϕ − )∗ H
'
on X [n] . Since

'
Z · [W '
− )∗ Cl(X [n] ) = Cl(X ' + )∗ Cl((X [n] )+ )
n ] ⊕ (ϕ
[n] ) = Z · [W
n ] ⊕ (ϕ

where Cl(·) stands for the divisor class group, we conclude that

(2.26) (ϕ '
− )∗ ((n − 1)D − Bn /2) = λW + )∗ H+
n + (ϕ

for some integer λ and some divisor H+ on (X [n] )+ . Let C  ⊂ X ' [n] be a line

contained in a fiber P of the projective bundle ϕ

n
:W' ∗
n → P(M ). Then, ϕ
 =
+ (C)
 ∗
 C) is a point in P(M ) ⊂ (X )+ , and (ϕ1 ◦ ϕ
ϕ( [n] 
− )(C) is a point in ϕ1 (Wn ) ⊂
 to both sides of (2.26) and using (2.25), we obtain
ϕ1 (X [n] ). So applying ·C

− )∗ H · C
0 = (ϕ1 ◦ ϕ

− )∗ ((n − 1)D − Bn /2) · C
= (ϕ
= (λW' n + (ϕ
∗ 
+ ) H+ ) · C
'
= λ · (W 
n · C)

= λ · (c1 (OPn (−1)) · C)
= −λ

where we have used the fact that OX '

(W n )|Pn = OPn (−1) when P is a fiber of
n
[n]

'
:W
the projective bundle ϕ ∗
n → P(M ). So λ = 0 and

(2.27) − )∗ ((n − 1)D − Bn /2) = (ϕ

(ϕ + )∗ H+ .
 2.4. A FLIP STRUCTURE ON (P2 )[n] WHEN n ≥ 3 39

Since ϕ+ is the blowing-up morphism of the smooth variety (X [n] )+ , we see
∗
+ )∗ |H+ |. So combining with (2.25) and (2.27), we obtain
+ ) H+ | = (ϕ
that |(ϕ
(2.28) − )∗ H| = |(ϕ
|(ϕ1 ◦ ϕ − )∗ ((n − 1)D − Bn /2)|
= |(ϕ+ )∗ H+ |
= (ϕ+ )∗ |H+ |.
Since ϕ1 is the contraction of the extremal ray R = R+ ·(β −(n−1)βn ), it is known
from [KMM, Remark 5-1-1] that ϕ1 (X [n] ) has at most rational singularities. So
− )∗ H| = (ϕ1 ◦ ϕ
|(ϕ1 ◦ ϕ − )∗ |H|.
In view of (2.28),
− )∗ |H| = |(ϕ1 ◦ ϕ
(ϕ1 ◦ ϕ − )∗ H| = (ϕ
+ )∗ |H+ |.
Since H is a very ample divisor on ϕ1 (X [n] ), (ϕ1 ◦ ϕ − )∗ |H| is base-point-free. It
∗ ∗
+ ) |H+ | = (ϕ1 ◦ ϕ
follows from (ϕ − ) |H| that |H+ | is also base-point-free. So |H+ |
induces a morphism
ϕ+ : (X [n] )+ → ϕ+ ((X [n] )+ ).
Moreover, since (ϕ+ )∗ |H+ | = (ϕ1 ◦ ϕ− )∗ |H|, we conclude that ϕ1 ◦ ϕ − = ϕ+ ◦ ϕ +
up to an isomorphism between ϕ1 (X [n] ) and ϕ+ ((X [n] )+ ). So putting ϕ− = ϕ1 , we
obtain a commutative diagram:
'
X [n]

ϕ− ϕ+
[n]
(2.29) X (X [n] )+ .
 ϕ−
ϕ+

ϕ1 (X [n] )
Note that ϕ+ : (X [n] )+ − P(M ∗ ) → ϕ1 (X [n] ) − ϕ1 (Wn ) is an isomorphism.
It is standard to check that K(X [n] )+ is ϕ+ -ample. So the commutative triangle
− )−1
+ ◦(ϕ
ϕ
X [n]  (X [n] )+
(2.30)  ϕ−
 ϕ+

ϕ1 (X [n] )
is the flip for ϕ− = ϕ1 . 
Remark 2.27. The proof of Theorem 2.26 shows that for n = 3, the morphism
ϕ1 : X [n] → ϕ1 (X [n] )
is the blowing-up along ϕ1 (Wn ) ∼
= (P2 )∗ , and Wn is the exceptional locus.
 Part 2

Hilbert schemes and infinite

dimensional Lie algebras
 CHAPTER 3

Hilbert schemes and infinite

dimensional Lie algebras

In this chapter, we will survey various infinite dimensional Lie algebra actions
on the cohomology of the Hilbert schemes of points on a smooth projective complex
surface. As a motivation and background material, we will begin with Nakajima’s
affine Lie algebra action on the homology of quiver varieties [Nak1, Nak2, Nak4].
We will review the Heisenberg algebras of Nakajima and Grojnowski [Groj, Nak3]
which provide an important language for describing the homology and cohomology
classes of the Hilbert schemes, determine the geometric interpretations of Heisen-
berg monomial classes, and study the homology classes of curves in the Hilbert
schemes. Then we will present the Virasoro algebras and boundary operator of
Lehn [Leh1] as well as the Ext vertex operators of Carlsson and Okounkov [Car1,
Car2,CO]. As evidenced by [LQW1,LQW2,LQW3,LQW5,LQW7,LS1,LS2],
Lehn’s boundary operator and the Heisenberg algebras of Nakajima and Grojnowski
play essential roles in determining the cohomology ring structures of the Hilbert
schemes. The Ext vertex operators of Carlsson and Okounkov are very powerful
tools in understanding the intersection theory of the Hilbert schemes coupled with
the Chern classes of their tangent bundles, and in investigating the connection be-
tween the multiple q-zeta values and the Hilbert schemes of points [Ok2,QY]. These
operators also have applications in studying the AGT correspondence [AGT, Neg].
In addition, we will prove a structure theorem (Theorem 3.31) for the higher order
derivatives of the Heisenberg operators.

3.1. Affine Lie algebra action of Nakajima

In this section, we will briefly review Nakajima’s construction [Nak1, Nak2,
Nak4] of affine Lie algebra action on the homology of quiver varieties. When a
complex surface X0 is the minimal resolution of a Kleinian (or simple surface)
singularity, the Hilbert schemes of points on it can be described as quiver varieties
of affine type. Therefore, Nakajima’s work provides a strong motivation for the
construction [Groj, Nak3] of Heisenberg algebra action on the cohomology of the
Hilbert schemes of points on complex surfaces. To avoid getting into technicalities,
we will follow the presentations in [Nag] and concentrate on the case of the affine
Kac-Moody algebra sl  .
Let (I, H) be a quiver , i.e, I is a set of vertices and H is a set of oriented edges.
Assume that there is a subset Ω ⊂ H such that Ω ∪ Ω = H and Ω ∩ Ω = ∅, where
Ω means reversing orientations of the edges in Ω. Assume further that Ω has no
cycle. Fix a dimension vector v ∈ (Z+ )I and a framing vector w ∈ (Z≥0 )I . Let

ζ = (ζC , ζR ) ∈ CI ⊕ RI .
43
44 3. HILBERT SCHEMES AND INFINITE DIMENSIONAL LIE ALGEBRAS

Let V, W be I-graded vector spaces such that dim Vi = vi and dim Wi = wi . Put
   
 
M(v, w) = Hom(Vout(h) , Vin(h) ) ⊕ Hom(Wi , Vi ) ⊕ Hom(Vi , Wi )
h∈H i∈I

where h ∈ H is drawn from out(h) to in(h). An element of M(v, w) is called an

ADHM (Atiyah, Drinfel’d, Hitchin and Manin) datum. Let

GLv = GL(Vi ),
i∈I

glv = gl(Vi ).
i∈I

Then GLv acts on M(v, w) by

" #
−1
(gi ) · (Bh , ai , bi ) = gin(h) Bh gout(h) , gi ai , bi gi−1

for gi ∈ GL(Vi ), Bh ∈ Hom(Vout(h) , Vin(h) ), ai ∈ Hom(Wi , Vi ), and bi ∈ Hom(Vi , Wi ).

The moment map μC is defined by
⎛ ⎞
 
μC (B, a, b) = ⎝ (h)Bh Bh + ai bi ⎠ ∈ gl(Vi ) = glv
in(h)=i i∈I

where (h) = 1 and (h) = −1 for h ∈ Ω. Note that CI is canonically identified with
the center of glv . The element ζC = (ζC,i ) ∈ CI is identified with ⊕i∈I ζC,i IdVi ∈ glv .
The quiver variety Mζ (v, w) associated to v, w and ζ is defined to be
μ−1
ss
C (ζC ) //GLv
where the index “ss” means the subset of ζR -semistable objects in μ−1 C (ζC ).
Let A = (ai,j )i,j∈I be the adjacency matrix of the quiver (I, H), that is to say
ai,j is the number of oriented edges in H which are drawn from i to j. Let
C = 2 · Id − A
be the associated Cartan matrix . Then we consider the set of positive roots
R+ = θ ∈ (Z≥0 )I |θ t Cθ ≤ 2 − {0}.
For v ∈ (Z+ )I , define
R+ (v) = θ = (θi ) ∈ R+ |θi ≤ vi ∀i ∈ I .
An element ζ ∈ CI ⊕ RI is called generic with respect to v if for any θ ∈ R+ (v),
ζ ∈ Dθ ⊗ R3 ⊂ RI ⊗ R3 ∼
= CI ⊕ RI
where * +

Dθ = η = (ηi ) ∈ R | I
ηi θi = 0 .
i∈I
Note that
ζ0 = (0, ζR )
is generic if ζR ∈ C0 where
C0 = η = (ηi ) ∈ RI |ηi < 0 ∀i ∈ I .
 3.1. AFFINE LIE ALGEBRA ACTION OF NAKAJIMA 45

It is known that Mζ (v, w) is smooth if ζ is generic with respect to v. Since


RI ⊗ R3 − Dθ ⊗ R3
θ∈R+ (v)

is connected, Mζ (v, w) and Mζ  (v, w) are diffeomorphic for generic ζ and ζ  . More-
over, for generic ζ, Mζ (v, w) is a fine moduli space of representations of the quiver
[King, Section 3].
In the rest of this section, let (I, H) be the quiver of type Â−1 with cyclic
orientation. Fix w0 = (1, 0, . . . , 0) ∈ ZI , and ζ0 = (0, ζR ) with ζR ∈ C0 . Let
ei denote the i-th coordinate vector of ZI . For fixed i and v, define the Hecke
correspondence to be the following subvariety of Mζ0 (v, w0 ) × Mζ0 (v + ei , w0 ):
(3.1) Bi (v) = (J1 , J2 )|J1 ∈ Mζ0 (v, w0 ) is
a subrepresentation of J2 ∈ Mζ0 (v + ei , w0 ) .
By [Nak4, Theorem 5.7], Bi (v) is Lagrangian in Mζ0 (v, w0 ) × Mζ0 (v + ei , w0 ).
Next, let sl be the Lie algebra consisting of all complex × -matrices of trace
zero. It has Chevalley generators
ei = Ei,i+1 , fi = Ei+1,i , hi = Ei,i − Ei+1,i+1
with 1 ≤ i ≤ − 1, where Ei,j denotes the matrix which has a 1 on the i-th row
  is given by
and j-th column and 0 elsewhere. The affine Kac-Moody algebra sl
  = sl [t, t−1 ] ⊕ Cc ⊕ Cd
sl
[Hern, Subsection 4.2]. In [Nak4, Section 9], a geometric representation of sl   on
the homology of the quiver varieties Mζ0 (v, w0 ) indexed by v is constructed. The
images of hi and d are certain multiples of the fundamental classes of the diagonals
in the products Mζ0 (v, w0 ) × Mζ0 (v, w0 ). The image of ei is the operator Ei
defined by
Ei (α) = (−1)vi−1 +vi p1∗ p∗2 α · Bi (v)
where pj is the j-th projection on Mζ0 (v, w0 ) × Mζ0 (v + ei , w0 ), while the image
of fi is the operator Fi defined by
Fi (α) = (−1)vi +vi+1 p2∗ p∗1 α · Bi (v) .
The quiver variety Mζ0 (v, w0 ) is related to the Hilbert schemes of points as
follows. Regard the cyclic group Z/ Z as the subgroup of SL2 (C) consisting of the
diagonal matrices diag(i , −i ), 0 ≤ i < where  is a primitive -th root of unity.
Then, Z/ Z acts on the affine plane C2 . Let X0 denote the minimal resolution of
C2 /(Z/ Z). Then there exists an element
ζ∞ = (0, ζR ) ∈ CI ⊕ RI
generic with respect to v such that Mζ∞ (v, w0 ) is naturally identified with the
Hilbert scheme (X0 )[n] where n = dim Mζ∞ (v, w0 ) /2. Since both ζ0 and ζ∞ are
generic with respect to v, Mζ∞ (v, w0 ) is diffeomorphic to Mζ0 (v, w0 ). Therefore,
the above geometric action of the affine Lie algebra sl   on the homology of the
  on the
quiver varieties Mζ0 (v, w0 ) with varying v induces naturally an action of sl
homology of the Hilbert schemes (X0 ) , n ≥ 0.
[n]

The preceding discussion indicates strongly that for an arbitrary complex sur-
face X, there should exist, via certain Hecke correspondence, geometric action of
infinite dimensional Lie algebras on the (co)homology groups of the Hilbert schemes
46 3. HILBERT SCHEMES AND INFINITE DIMENSIONAL LIE ALGEBRAS

X [n] , n ≥ 0 (here we have by-passed the identification of quiver varieties with moduli
spaces of torsion-free sheaves on ALE spaces). Indeed, Nakajima and Grojnowski
[Groj, Nak3] constructed action of Heisenberg algebras on the (co)homology of
these Hilbert schemes. This leads us to the next section.
Remark 3.1. It is unclear whether the Hecke correspondence Bi (v) defined in
(3.1) for the quiver varieties could be transferred directly to the Hilbert schemes
(X0 )[n] . On the other hand, by the theorem in [Nag, Subsection 6.2], the above
induced action of the affine Lie algebra sl   on the homology of the Hilbert schemes
(X0 )[n] is related to the action of the Heisenberg algebra (to be constructed in the
next section) via the Frenkel-Kac construction [Kac, Section 5.6].

3.2. Heisenberg algebras of Nakajima and Grojnowski

In this section, we will define the Fock space as the sum of the cohomology
groups of all the Hilbert schemes X [n] of points on a smooth projective complex
surface X. Then we will construct the Heisenberg algebras of Nakajima and Gro-
jnowski, and discuss the Heisenberg monomial classes.
By our convention, all homology and cohomology groups are with complex
coefficients, unless otherwise specified.
Definition 3.2. Let X be a smooth projective complex surface.
(i) Let Hn,i i
X denote H (X
[n]
), and

4n
HnX = H ∗ (X [n] ) = Hn,i
X .
i=0

Let

+∞ 
HX = HnX = Hn,i
X
n=0 n,i≥0

denote the bi-graded vector space. The element 1 in H 0 (X [0] ) = C is

called the vacuum vector and denoted by |0 ;
(ii) Define the degree of α ∈ H ∗ (X [n] ) to be |α| = s if α ∈ H s (X [n] );
(iii) An operator f ∈ End(HX ) is homogeneous of bi-degree ( , m) if
f(Hn,i n+,i+m
X ) ⊂ HX ;
(iv) For linear operators f and g ∈ End(HX ) of bi-degrees ( , m) and ( 1 , m1 )
respectively, define the Lie superalgebra bracket [f, g] by putting
[f, g] = fg − (−1)mm1 gf.
(v) The non-degenerate super-symmetric bilinear form ·, · on HX is induced
from the standard one on HnX = H ∗ (X [n] ) defined by
,
a, b = a·b
X [n]

for a, b ∈ H ∗ (X [n] ). For an operator f ∈ End(HX ) of bi-degree ( , m), we

define its adjoint operator f† ∈ End(HX ) by putting
f(a), b = (−1)m·|a| · a, f† (b) .
 3.2. HEISENBERG ALGEBRAS OF NAKAJIMA AND GROJNOWSKI 47

Let f ∈ End(HX ) be of bi-degree ( , m). Then the bi-degree of f† is (− , m−4 ).

Also, for g, h ∈ End(HX ) of bi-degrees ( 1 , m1 ), ( 2 , m2 ) respectively, we have
(3.2) (fg)† = (−1)mm1 · g† f† ,
(3.3) [f, g]† = (−1)mm1 · [g† , f† ] = −[f† , g† ],
(3.4) [f, [g, h]] = [[f, g], h] + (−1)mm1 [g, [f, h]],
(3.5) [f, gh] = [f, g]h + (−1)mm1 g[f, h].
We recall Nakajima’s definition of the Heisenberg operators [Groj,Nak3,Nak5].
For m ≥ 0 and n > 0, let Q[m,m] = ∅ and define Q[m+n,m] to be
 
(3.6) (ξ, x, η) ∈ X [m+n] × X × X [m] | ξ ⊃ η and Supp(Iη /Iξ ) = {x} .

Definition 3.3. Fix a cohomology class α ∈ H ∗ (X). Let n > 0. The linear
operator a−n (α) ∈ End(HX ) is defined by
" #
a−n (α)(a) = p̃1∗ [Q[m+n,m] ] · ρ̃∗ α · p̃∗2 a

for a ∈ H ∗ (X [m] ), where p̃1 , ρ̃, p̃2 are the projections of X [m+n] × X × X [m] to
X [m+n] , X, X [m] respectively. Define an (α) ∈ End(HX ) to be (−1)n times the
operator obtained from the definition of a−n (α) by switching the roles of p̃1 and
p̃2 . We often refer to a−n (α) (respectively, an (α)) as the creation (respectively,
annihilation) operator. We also set a0 (α) = 0.
Lemma 3.4. Let m ≥ 0 and n > 0. Then, dim Q[m+n,m] = 2m + n + 1.
Proof. Every element (ξ, x, η) ∈ Q[m+n,m] is of the form
η = ηx + η  , ξ = ξx + η 
where Supp(ηx ) = Supp(ξx ) = {x}, x ∈ Supp(η  ) and ηx ⊂ ξx . Fix the length
= (ηx ). If = 0, then these elements (ξ, x, η) ∈ Q[m+n,m] form an open subset
of Q[m+n,m] with dimension equal to
#moduli{η  } + #moduli{x} + #moduli{ξx } = 2m + 2 + (n − 1)
= 2m + n + 1
where we have used dim Mn (x) = n − 1 from Theorem 1.18. If > 0, then the
number of moduli of these elements (ξ, x, η) ∈ Q[m+n,m] is at most
#moduli{η  } + #moduli{x} + #moduli{ηx } + #moduli{ξx }
≤ 2(m − ) + 2 + ( − 1) + (n + − 1)
= 2m + n.
It follows that the dimension of Q[m+n,m] is equal to 2m + n + 1. 

Corollary 3.5. Let n = 0. Then the bi-degree of a−n (α) is (n, 2n − 2 + |α|).
Proof. Follows immediately from Lemma 3.4 and Definition 3.3. Note from
the proof of Lemma 3.4 that Q[m+n,m] has a unique irreducible component of di-
mension 2m+n+1. Although Q[m+n,m] may (or may not) contain other irreducible
components of lower dimension, it does not matter since only the fundamental class
[Q[m+n,m] ] is used in Definition 3.3. 
48 3. HILBERT SCHEMES AND INFINITE DIMENSIONAL LIE ALGEBRAS

Remark 3.6. Let n > 0. The geometric meaning of the operator a−n (α)
is roughly the following. Let a ∈ H ∗ X [m] . Represent α and a by the cycles
Γα ⊂ X and Γa ⊂ X [m] respectively such that a generic element x ∈ Γα and a
generic element η ∈ Γa satisfy x ∈ Supp(η). Then a−n (α) is represent by the cycle
Γ ⊂ X [m+n] whose generic elements are of the form η + ξx where η ∈ Γa , x ∈ Γα
but x ∈ Supp(η), and ξx ∈ Mn (x).
Lemma 3.7. a−n (α)† = (−1)n · an (α).
Proof. The argument is completely formal. We may assume that n > 0. Let
a ∈ H ∗ X [m] and b ∈ H ∗ X [m+n] . We must prove that
   
(3.7) a−n (α)(a), b = (−1)(2n−2+|α|) |a| · a, (−1)n · an (α)(b)
noting that the bi-degree of a−n (α) is (n, 2n − 2 + |α|).
Let notations be the same as in Definition 3.3. By Definition 3.3,
,
 
a−n (α)(a), b = a−n (α)(a) · b
[m+n]
,X " #
= p̃1∗ [Q[m+n,m] ] · ρ̃∗ α · p̃∗2 a · b
[m+n]
,X " #
(3.8) = p̃1∗ [Q[m+n,m] ] · ρ̃∗ α · p̃∗2 a · p̃∗1 b
X [m+n]
where we have used the projection formula in the last step. Similarly, for the
right-hand-side of (3.7), we see from the definition of an (α) that
 
(−1)(2n−2+|α|) |a| · a, (−1)n · an (α)(b)
, " #
|α| |a|
= (−1) · p̃2∗ p̃∗2 a · [Q[m+n,m] ] · ρ̃∗ α · p̃∗1 b
X [m]
, " #
= p̃2∗ [Q[m+n,m] ] · ρ̃∗ α · p̃∗2 a · p̃∗1 b .
X [m]
Combining this with (3.8), we conclude that (3.7) holds. 
The following theorem is proved by Nakajima [Nak3, Section 3] and Gro-
jnowski [Groj, Section 3].
Theorem 3.8. The operators an (α) ∈ End(HX ) with α ∈ H ∗ (X) and n ∈ Z
satisfy the following Heisenberg algebra commutation relation:
(3.9) [am (α), an (β)] = −m · δm,−n · (α, β) · IdHX
where we have used δm,−n to denote 1 if m = −n and 0 otherwise. Moreover,
the space HX is an irreducible module over the Heisenberg algebra generated by the
operators an (α) with a highest weight vector |0 = 1 ∈ H 0 (X [0] ).
It follows that HX is linearly spanned by all the Heisenberg monomial classes:
(3.10) a−n1 (α1 ) · · · a−nk (αk )|0
where k ≥ 0, n1 , . . . , nk > 0, and the cohomology classes α1 , . . . , αk run over a fixed
linear basis of H ∗ (X). In view of Corollary 3.5, we have
(3.11) a−n1 (α1 ) · · · a−nk (αk )|0 ∈ H m (X [n] )

where n = n1 + . . . + nk and m = ki=1 (2ni − 2 + |αi |).
 3.2. HEISENBERG ALGEBRAS OF NAKAJIMA AND GROJNOWSKI 49

Next, we will introduce the notion am1 · · · amk (τk∗ α) and study its properties.
This notion will be useful in writing down the Chern character operators and the
higher derivatives of the Heisenberg operators.
Definition 3.9. For k ≥ 1, define τk∗ : H ∗ (X) → H ∗ (X k ) to be the map
induced by the diagonal embedding τk : X → X k , and am1 · · · amk (τk∗ α) to be

am1 (αj,1 ) · · · amk (αj,k )
j

when

(3.12) τk∗ α = αj,1 ⊗ · · · ⊗ αj,k
j

via the Künneth decomposition of H ∗ (X k ).

Lemma 3.10. The notion am1 · · · amk (τk∗ α) is well-defined in the sense that it
is independent of the decomposition (3.12).
Proof. Assume that via the Künneth decomposition of H ∗ (X k ), we have
 
(3.13) τk∗ α = αj,1 ⊗ · · · ⊗ αj,k = βs,1 ⊗ · · · ⊗ βs,k .
j s
∗
Let {γ } be a linear basis of H (X). For j, s and 1 ≤ t ≤ k, put
 
αj,t = cj,t, γ , βs,t = ds,t, γ
 

where cj,t, , ds,t, ∈ C. In view of (3.13), we get

⎛ ⎞
 
⎝ cj,1,1 · · · cj,k,k ⎠ γ1 ⊗ · · · ⊗ γk
1 ,...,k j
 
 
= ds,1,1 · · · ds,k,k γ 1 ⊗ · · · ⊗ γ k .
1 ,...,k s

Therefore,  
cj,1,1 · · · cj,k,k = ds,1,1 · · · ds,k,k
j s
for all 1, . . . , k. It follows that

am1 (αj,1 ) · · · amk (αj,k )
j
⎛ ⎞
 
= ⎝ cj,1,1 · · · cj,k,k ⎠ am1 (γ1 ) · · · amk (γk )
1 ,...,k j
 
 
= ds,1,1 · · · ds,k,k am1 (γ1 ) · · · amk (γk )
1 ,...,k s

= am1 (βs,1 ) · · · amk (βs,k ).
s

Hence, am1 · · · amk (τk∗ α) is independent of the decomposition (3.13). 

50 3. HILBERT SCHEMES AND INFINITE DIMENSIONAL LIE ALGEBRAS

The following topological lemma is elementary, but very efficient in organizing

the Heisenberg monomial classes.

Lemma 3.11. Let k, u ≥ 1 and α, β ∈ H ∗ (X). Assume that


τk∗ (α) = αi,1 ⊗ . . . ⊗ αi,k
i

under the Künneth decomposition of H ∗ (X k ). Then for 0 ≤ j ≤ k, we have

 k " #
τk∗ (αβ) = (−1)|β|· =j+1 |αi, |
· ⊗j−1
s=1 αi,s ⊗ (αi,j β) ⊗ ⊗t=j+1 αi,t
k

i
 k
,
τ(k−1)∗ (αβ) = (−1)|β| =j+1 |αi, |
αi,j β · ⊗1≤s≤k,s=j αi,s
i X
 " j−1 #
τ(k+u−1)∗ (α) = ⊗s=1 αi,s ⊗ (τu∗ αi,j ) ⊗ ⊗kt=j+1 αi,t .
i

Proof. The basic idea is to use the projection formula. Let pj be the projec-
tion of X k to the j-th factor. Then, we have
 k " #
(−1)|β|· =j+1 |αi, |
· ⊗j−1
s=1 αi,s ⊗ (αi,j β) ⊗ ⊗t=j+1 αi,t
k

i
 

= αi,1 ⊗ . . . ⊗ αi,k · p∗j (β)
i
= τk∗ (α) · p∗j (β)
= τk∗ (α · (pj ◦ τk )∗ (β))
= τk∗ (αβ).

This proves the first formula.

For the second formula, let p̃j : X k → X k−1 be the projection by ignoring the
j-th factor. Then, we get

 k
,
(−1)|β| =j+1 |αi, |
αi,j β · ⊗1≤s≤k,s=j αi,s
i X
 
  " #
|β|· k=j+1 |αi, |
= p̃j∗ (−1) · ⊗j−1
s=1 αi,s ⊗ (αi,j β) ⊗ ⊗kt=j+1 αi,t
i
= p̃j∗ τk∗ (αβ)
= τ(k−1)∗ (αβ)

where we have used the first formula in the second step. This proves the second
formula.
For the third formula, let

p̂j = (IdX )j−1 × τu × (IdX )k−j : X k → X k+u−1 .

 3.2. HEISENBERG ALGEBRAS OF NAKAJIMA AND GROJNOWSKI 51

Then, we obtain
 " j−1 #
⊗s=1 αi,s ⊗ (τu∗ αi,j ) ⊗ ⊗kt=j+1 αi,t
i
 

= p̂j∗ αi,1 ⊗ . . . ⊗ αi,k
i
= p̂j∗ (τk∗ (α))
= τ(k+u−1)∗ (α).

This proves the third formula. 

In the lemma below and throughout the book, the products of Heisenberg
operators are understood to be in the increasing order of the parametrizing indices
from the left to the right, e.g.,

ans = an1 an2 . . . anj−1
1≤s<j

and

anu = an1 . . . ant−1 ant+1 . . . ank .
1≤u≤k,u=t

Lemma 3.12. Let k, s ≥ 1, n1 , . . . , nk , m1 , . . . , ms ∈ Z and α, β ∈ H ∗ (X).

(i) The commutator [an1 · · · ank (τk∗ α), am1 · · · ams (τs∗ β)] is equal to
⎛ ⎞
k s 
j−1   s
− nt δnt ,−mj · ⎝ am anu am ⎠ (τ(k+s−2)∗ (αβ));
t=1 j=1 =1 1≤u≤k,u=t =j+1

(ii) Let j satisfy 1 ≤ j < k. Then, an1 · · · ank (τk∗ α) is equal to

⎛ ⎞
 
⎝ ans · anj+1 anj · ans ⎠ (τk∗ α)
1≤s<j j+1<s≤k

− nj δnj ,−nj+1 ans (τ(k−2)∗ (eX α))
1≤s≤k
s=j,j+1

where eX is the Euler class of X.

Proof. To simplify the signs, we will assume that all the cohomology classes
have even degrees.
(i) Let

τk∗ α = αa,1 ⊗ · · · ⊗ αa,k ,
a

τs∗ β = βb,1 ⊗ · · · ⊗ βb,s .
b
52 3. HILBERT SCHEMES AND INFINITE DIMENSIONAL LIE ALGEBRAS

By (3.5), we have

(3.14) [an1 · · · ank (τk∗ α), am1 · · · ams (τs∗ β)]


= [an1 · · · ank (τk∗ α), am1 (βb,1 ) · · · ams (βb,s )]
b

 
s j−1 
s
= am (βb, ) · [an1 · · · ank (τk∗ α), amj (βb,j )] · am (βb, ).
b j=1 =1 =j+1

By (3.5) again and Theorem 3.8, we get

[an1 · · · ank (τk∗ α), amj (βb,j )]


= [an1 (αa,1 ) · · · ank (αa,k ), amj (βb,j )]
a
 
k t−1 
k
= anu (αa,u ) · [ant (αa,t ), amj (βb,j )] · anu (αa,u )
a t=1 u=1 u=t+1


k , 
= − nt δnt ,−mj · αa,t βb,j · anu (αa,u ).
t=1 a X 1≤u≤k,u=t

By the second formula in Lemma 3.11, we obtain

[an1 · · · ank (τk∗ α), amj (βb,j )]

⎛ ⎞
 k 
= − nt δnt ,−mj · ⎝ anu ⎠ τ(k−1)∗ (αβb,j ).
t=1 1≤u≤k,u=t

Combining with (3.14), we see that [an1 · · · ank (τk∗ α), am1 · · · ams (τs∗ β)] is equal to


k 
s  j−1

− nt δnt ,−mj · am (βb, )
t=1 j=1 b =1

⎛ ⎞
 
s
·⎝ anu ⎠ τ(k−1)∗ (αβb,j ) · am (βb, ).
1≤u≤k,u=t =j+1

By the first and third formulas in Lemma 3.11, we conclude that

[an1 · · · ank (τk∗ α), am1 · · · ams (τs∗ β)]

⎛ ⎞
 k  s 
j−1  
s
= − nt δnt ,−mj · ⎝ am anu am ⎠ (τ(k+s−2)∗ (αβ)).
t=1 j=1 =1 1≤u≤k,u=t =j+1

(ii) Let

τ(k−1)∗ α = αa,1 ⊗ · · · ⊗ αa,k−1 ,
a

τ2∗ αa,j = βa,j,b,1 ⊗ βa,j,b,2
b
 3.2. HEISENBERG ALGEBRAS OF NAKAJIMA AND GROJNOWSKI 53

where we may assume that b is independent of a and j. By the third formula in

Lemma 3.11,

τk∗ α = ⊗j−1
s=1 αa,s ⊗ βa,j,b,1 ⊗ βa,j,b,2 ⊗ ⊗s=j+1 αa,s .
k−1

a,b

Combining with Theorem 3.8, we obtain

an1 · · · an (τk∗ α)
  k 
= ans (αa,s ) · anj (βa,j,b,1 )anj+1 (βa,j,b,2 ) · ans+1 (αa,s )
a,b 1≤s<j j<s≤k−1
  
= ans (αa,s ) · anj+1 (βa,j,b,2 )anj (βa,j,b,1 ) · ans+1 (αa,s )
a,b 1≤s<j j<s≤k−1
,  
−nj δnj ,−nj+1 βa,j,b,1 βa,j,b,2 · ans (αa,s ) · ans+1 (αa,s )
a,b X 1≤s<j j<s≤k−1
⎛ ⎞
 
= ⎝ ans · anj+1 anj · ans ⎠ (τk∗ α)
1≤s<j j+1<s≤k
,  
−nj δnj ,−nj+1 βa,j,b,1 βa,j,b,2 · ans (αa,s ) · ans+1 (αa,s ).
a,b X 1≤s<j j<s≤k−1

Note that 
βa,j,b,1 βa,j,b,2 = eX αa,j .
b
Therefore,
an · · · ank (τk∗ α)
⎛1 ⎞
 
= ⎝ ans · anj+1 anj · ans ⎠ (τk∗ α)
1≤s<j j+1<s≤k
,  
−nj δnj ,−nj+1 eX αa,j · ans (αa,s ) · ans+1 (αa,s )
a X 1≤s<j j<s≤k−1
⎛ ⎞
 
= ⎝ ans · anj+1 anj · ans ⎠ (τk∗ α)
1≤s<j j+1<s≤k

−nj δnj ,−nj+1 ans (τ(k−2)∗ (eX α))
1≤s≤k
s=j,j+1

where at the last step, we have used the second formula in Lemma 3.11. 
In the following, we will define some convenient operation regarding Heisenberg
monomial classes. It enables us to decompose cohomology classes. In Section 15.3,
we will see that this operation plays an interesting role in describing the structure
of the 3-point genus-0 extremal Gromov-Witten invariants of the Hilbert scheme
X [n] .
Definition 3.13. Let A = a−n1 (α1 ) · · · a−n (α )|0 where n1 , . . . , n > 0.
(i) If B = a−m1 (β1 ) · · · a−ms (βs )|0 with m1 , . . . , ms > 0, then we define
(3.15) A ◦ B = a−n1 (α1 ) · · · a−n (α )a−m1 (β1 ) · · · a−ms (βs )|0 .
54 3. HILBERT SCHEMES AND INFINITE DIMENSIONAL LIE ALGEBRAS

(ii) We use the symbol B ⊂ A if B = a−ni1 (αi1 ) · · · a−nis (αis )|0 with 1 ≤
A
i1 < . . . < is ≤ . In this case, we use A/B or AB −1 or to denote the
B
cohomology class obtained from A by deleting the factors
a−ni1 (αi1 ), . . . , a−nis (αis ).
Note that when B ⊂ A, then we have the identity (A/B) ◦ B = A.
Lemma 3.14. Assume that the odd Betti numbers of X are equal to zero. Let
A = a−n1 (α1 ) · · · a−n (α )|0 where n1 , . . . , n > 0. Let p be a monomial of Heisen-
berg creation operators. Then,
  A
(3.16) p† A = p|0 , B · .
B
B⊂A

Proof. Let
p = a−p1 (β1,1 ) · · · a−p1 (β1,k1 ) · · · a−ps (βs,1 ) · · · a−ps (βs,ks )
where 0 < p1 < . . . < ps . We use induction on s. Let s = 1. If p1 = ni for every
i = 1, . . . , , then both sides of (3.16) are equal to 0. Next, let p1 = ni for some
i ∈ {1, . . . , }. We may assume that i = 1 with
p1 = n1 = . . . = nt < nt+1 ≤ nt+2 . . . ≤ n .
By Lemma 3.7 and Theorem 3.8,
p† A
= (−1)k1 p1 ap1 (β1,1 ) · · · ap1 (β1,k1 )a−n1 (α1 ) · · · a−n (α )|0
 (−n1 )k1 · β1,1 , ασ(j1 ) · · · β1,k1 , ασ(jk1 ) · A
= (−1)k1 p1
1≤j <...,<j ≤
a−nj1 (αj1 ) · · · a−njk (αjk1 )|0
1 k1 1
σ∈Per{j1 ,...,jk }
1
  A
= p|0 , B ·
B
B⊂A

where Per{j1 , . . . , jk1 } denotes the permutation group of {j1 , . . . , jk1 }. Therefore,
(3.16) is true when s = 1. Next, let s > 1 and assume that (3.16) holds for
p̃ = a−p1 (β1,1 ) · · · a−p1 (β1,k1 ) · · · a−ps−1 (βs−1,1 ) · · · a−ps−1 (βs−1,ks−1 ).
†
Note from (3.2) that p† = a−ps (βs,1 ) · · · a−ps (βs,ks ) p̃† . By induction,
p† A
†
= a−ps (βs,1 ) · · · a−ps (βs,ks ) p̃† A
†    A
= a−ps (βs,1 ) · · · a−ps (βs,ks ) p̃|0 , B1 ·
B1
B1 ⊂A
      A/B1
= a−ps (βs,1 ) · · · a−ps (βs,ks )|0 , B2 · p̃|0 , B1 ·
B2
B1 ⊂A B2 ⊂A/B1
   
If a−ps (βs,1 ) · · · a−ps (βs,ks )|0 , B2 · p̃|0 , B1 = 0, then B1 and B2 do not share
any common Heisenberg factors, and
     
a−ps (βs,1 ) · · · a−ps (βs,ks )|0 , B2 · p̃|0 , B1 = p|0 , B1 ◦ B2 .
 3.3. GEOMETRIC INTERPRETATIONS OF HEISENBERG MONOMIAL CLASSES 55

Hence, putting B = B1 ◦ B2 , we obtain

  A
p† A = p|0 , B · .
B
B⊂A

This completes the proof of (3.16). 

3.3. Geometric interpretations of Heisenberg monomial classes

By (3.10), the Heisenberg monomial classes linearly span the cohomology group
of the Hilbert scheme X [n] . In this section, we will describe the geometric repre-
sentations of the Heisenberg monomial classes
a−n1 (α1 ) · · · a−nk (αk )|0 ∈ H ∗ (X [n] )
where n = n1 + . . . + nk .
We begin with the construction of a certain closed subset in the Hilbert scheme
X [n] and the determination of its main component. Fix closed real cycles X1 , . . . , Xk
of X in general position in the sense that any subset of the Xi ’s meet transversally
in the expected dimension. Define
W (n1 , X1 ; . . . ; nk , Xk ) ⊂ X [n]
to be the closed subset consisting of all ξ ∈ X [n] which admit filtrations
ξ = ξk ⊃ . . . ⊃ ξ1 ⊃ ξ0 = ∅
with (ξi ) = (ξi−1 ) + ni and
(3.17) Supp(Iξi−1 /Iξi ) = xi ∈ Xi
for 1 ≤ i ≤ k. Let
W (n1 , X1 ; . . . ; nk , Xk )0 ⊂ W (n1 , X1 ; . . . ; nk , Xk )
be the open subset consisting of all ξ ∈ W (n1 , X1 ; . . . ; nk , Xk ) such that the points
x1 , . . . , xk in (3.17) are distinct.
Lemma 3.15. Denote W (n1 , X1 ; . . . ; nk , Xk )0 and W (n1 , X1 ; . . . ; nk , Xk ) by
W and W respectively. Let W c = W − W 0 . Then,
0


k
dimR (W c ) < dimR (W 0 ) = 2ni − 2 + dimR (Xi ) .
i=1

Proof. Note that the points ξ ∈ W 0 are of the form

ξ = η1 + . . . + ηk
with (ηi ) = ni and Supp(ηi ) = xi ∈ Xi for every i. So

k
dimR (W 0 ) = 2ni − 2 + dimR (Xi ) .
i=1

Next, without loss of generality, we may assume that a point ξ ∈ W c satisfies

x1 = . . . = xk 1 ,
...,
xk1 +...+kr−1 +1 = . . . = xk1 +...+kr .
56 3. HILBERT SCHEMES AND INFINITE DIMENSIONAL LIE ALGEBRAS

Here r ≥ 1, k1 ≥ 2, k2 , . . . , kr ≥ 1, k1 + . . . + kr = k, and the points y1 , . . . , yr are

distinct where
y1 = x1 = . . . = xk1 ,
...,
yr = xk1 +...+kr−1 +1 = . . . = xk1 +...+kr .
Since the cycles X1 , . . . , Xk are in general position, these ξ’s either do not exist at
all, or form a subset of dimension at most
- k  .
 1 k1
2 ni − 1 + dimR (Xi ) − 4(k1 − 1) + . . .
i=1 i=1
- k  .
 r 
kr
+ 2 nk1 +...+kr−1 +i − 1 + dimR (Xk1 +...+kr−1 +i ) − 4(kr − 1)
i=1 i=1

k 
r
= (2ni − 2 + dimR (Xi )) − 2(ki − 1)
i=1 i=1

k
< (2ni − 2 + dimR (Xi )).
i=1

It follows that

k
dimR (W c ) < 2ni − 2 + dimR (Xi ) = dimR (W 0 ).
i=1

This completes the proof of the lemma. 

We conclude from Lemma 3.15 that W (n1 , X1 ; . . . ; nk , Xk ) is a union of irre-

ducible closed subsets among which the closure of W (n1 , X1 ; . . . ; nk , Xk )0 is the
only irreducible component having the largest dimension.
Proposition 3.16. Let , k ≥ 0, si ≥ 0 (1 ≤ i ≤ ), ni > 0 (1 ≤ i ≤ k). Let

4
α1 , . . . , αk ∈ H i (X)
i=1

be homogeneous classes represented by the real cycles X1 , . . . , Xk ⊂ X respectively

such that X1 , . . . , Xk are in general position. Then, the Heisenberg monomial class
   k 
 a−i (1X )si 
(3.18) a−ni (αi ) |0
i=1
si ! i=1

is represented by the closure of

(3.19) W (1, X; . . . ; 1, X ; . . . ; , X; . . . ; , X ; n1 , X1 ; . . . ; nk , Xk )0 .
/ 01 2 / 01 2
s1 times s times
 k
Proof. We use induction on i=1 isi + i=1 ni . When

 
k
isi + ni = 0,
i=1 i=1
 3.3. GEOMETRIC INTERPRETATIONS OF HEISENBERG MONOMIAL CLASSES 57

the conclusion is trivially true. In the following, assuming that the conclusion is
 
true whenever i=1 isi + ki=1 ni < n, we will prove that it still holds when


 
k
isi + ni = n.
i=1 i=1

There are two cases: si > 0 for some i, or ni > 0 for some i. Since the proof of
these two cases are similar, we only prove the first case.
So let si > 0 for some i. Without loss of generality, we may assume that
i = , i.e., s > 0. Denote (3.18) and (3.19) by As and Ws0 respectively. By the
induction hypothesis, As −1 is represented by the closure Ws0 −1 . By Definition 3.3,
a− (1X )(As −1 ) is represented by
" #
] ∩ p∗2 Ws0 −1
[n,n−]
p1∗ [QX

where p1 , p2 are the projections of X [n] × X [n−] , and

 
[n,n−]
QX = (ξ, η) ∈ X [n] × X [n−] | ξ ⊃ η and Supp(Iη /Iξ ) is a point .

Since a− (1X )(As −1 ) = s As , we see that As is represented by

1 " #
] ∩ p∗2 Ws0 −1 .
[n,n−]
(3.20) · p1∗ [QX
s
" #
∩ p−1
[n,n−]
Note that the image p1 QX 2 Ws0 −1 is contained in the closed subset

W (1, X; . . . ; 1, X ; . . . ; , X; . . . ; , X ; n1 , X1 ; . . . ; nk , Xk ; , X)
/ 01 2 / 01 2
s1 times s −1 times

whose real dimension is precisely the cohomology degree of As , and

Ws0 = W (1, X; . . . ; 1, X ; . . . ; , X; . . . ; , X ; n1 , X1 ; . . . ; nk , Xk ; , X)0 .
/ 01 2 / 01 2
s1 times s −1 times
" #
] ∩ p∗2 Ws0 −1
[n,n−]
By Lemma 3.15, p1∗ [QX is a multiple of Ws0 . By (3.20), it
remains to show that the multiplicity is equal to s . Indeed, take a general element
ξ ∈ Ws0 . Then


 
si 
k
ξ= ξi,j + ξi
i=1 j=1 i=1

where (ξi,j ) = i, (ξi ) = ni , Supp(ξi,j ) = xi,j , Supp(ξi ) = xi , all the points xi,j , xi
are distinct, xi ∈ Xk if and only if i = k, and xi,j ∈ Xk for all i, j, k. Now,
" #
(p1 )−1 (ξ) ∩ QX ∩ p−1
[n,n−]
2 Ws0 −1

consists of exactly s different pairs (ξ, ξ − ξ,j ), 1 ≤ j ≤ s . 

Let 1X [n] be the fundamental class of X [n] . Recall the notations Bn and DC
from (1.26) and (1.27) respectively.
58 3. HILBERT SCHEMES AND INFINITE DIMENSIONAL LIE ALGEBRAS

Corollary 3.17. Let n ≥ 2 and C be a real-surface on X. Then,

1
(3.21) 1X [n] = · a−1 (1X )n |0 ,
n!
1
(3.22) Bn = · a−2 (1X )a−1 (1X )n−2 |0 ,
(n − 2)!
1
(3.23) DC = · a−1 (C)a−1 (1X )n−1 |0
(n − 1)!
where by abusing notations, we have also used C to denote its Poincaré dual in
H 2 (X).
Proof. These follows from Proposition 3.16, (1.26) and (1.27). 
Definition 3.18. For a nonnegative integer n, we define the operator:
1
1−n = · a−1 (1X )n
n!
so that 1X [n] = 1−n |0 . For simplicity, we set 1−n = 0 when n < 0.

3.4. The homology classes of curves in Hilbert schemes

In this section, with the help of Heisenberg monomial classes, we will present
two lemmas dealing with the homology classes of certain curves in the Hilbert
schemes X [n] . The results are useful in studying the Gromov-Witten theory of
X [n] .
First of all, we describe some elements in H2 (X [n] ). Let {α1 , . . . , αs } be a
basis of H 2 (X, ) and {αs+1 , . . . , αs+t } be a basis of H 3 (X), where every class αi is
homogeneous. By (3.10), a basis of the cohomology group H 4n−2 (X [n] ) consists of
the classes:
a−2 (x)a−1 (x)n−2 |0 ,
a−1 (αi )a−1 (x)n−1 |0 ,
a−1 (αj )a−1 (αk )a−1 (x)n−2 |0
where |αi | = 2, |αj | = |αk | = 3 with j < k, and by abusing notations, we have used
x to denote the cohomology class Poincaré dual to a point x ∈ X. The homology
class βn defined by (1.34) is Poincaré dual to a−2 (x)a−1 (x)n−2 |0 . Let
(3.24)  2 (X [n] ) ⊂ H2 (X [n] )
H
be the linear span of the Poincaré duals of the basis classes
a−1 (αj )a−1 (αk )a−1 (x)n−2 |0
in the above list. Moreover, for a homology class γ ∈ H2 (X), let βγ ∈ H2 (X [n] ) be
the Poincaré dual of a−1 (α)a−1 (x)n−1 |0 , where α is the Poincaré dual of γ. Note
that when C is a real-surface in X, the homology class βC is represented by (2.1).
Let Γ be an irreducible curve in the Hilbert scheme X [n] . In view of the
preceding paragraph,
(3.25) [Γ] = βγ + dβn (mod H  2 (X [n] ))
for some homology class γ ∈ H2 (X) and some integer d. To determine γ, recall the
universal codimension-2 closed subscheme Zn ⊂ X [n] × X from (1.24). Define
(3.26) ZΓ = Γ ×X [n] Zn ⊂ Zn ⊂ X [n] × X.
 3.4. THE HOMOLOGY CLASSES OF CURVES IN HILBERT SCHEMES 59

Let π̃1 , π̃2 be the restrictions to Zn of the two natural projections on X [n] × X.
The following lemma and its proof are from [Obe, Lemma 1] (similar statement
also appeared in [HLQ, Lemma 5.1]).
Lemma 3.19. Let n ≥ 2. Let Γ be an irreducible curve in the Hilbert scheme
X [n] . Then,
(3.27) [Γ] = βπ̃2∗ [ZΓ ] + dβn  2 (X [n] ))
(mod H
for some integer d.
Proof. Since π̃1 is flat, π̃2∗ [ZΓ ] = π̃2∗ π̃1∗ [Γ]. By (3.25), it remains to prove
π̃2∗ π̃1∗ βn = 0,
π̃2∗ π̃1∗ βγ = γ,
π̃2∗ π̃1∗ PD a−1 (αj )a−1 (αk )a−1 (x)n−2 |0 = 0
for γ ∈ H2 (X) and |αj | = |αk | = 3. Representing βn by (1.34), we see that
π̃2∗ π̃1∗ βn = 0.
Next, to prove π̃2∗ π̃1∗ βγ = γ, represent the homology class γ by a real smooth
surface S in X. Let y1 , . . . , yn−1 ∈ X be distinct with {y1 , . . . , yn−1 } ∩ S = ∅. Let
U = {(x1 , . . . , xn ) ∈ X n |x1 , . . . xn are mutually distinct},
and let σ : U → X [n] be the morphism sending (x1 , . . . , xn ) to x1 + . . . + xn . Define
 = U ×X [n] Zn . Let π̃1 : U
U  → U be the natural map. Then we have a diagram of
morphisms:

U → Z
π̃
→2 X
⏐ ⏐n
⏐  ⏐
%π̃1 %π̃1
σ
U → X [n] .
 → Zn →2 X be the composition. Then, βγ = σ∗ [y1 × · · · × yn−1 × S]. So
Let f : U
π̃

π̃2∗ π̃1∗ βγ = π̃2∗ π̃1∗ σ∗ [y1 × · · · × yn−1 × S]

= f∗ (π̃1 )∗ [y1 × · · · × yn−1 × S]
= [S]
= γ.
Finally, to prove π̃2∗ π̃1∗ PD a−1 (αj )a−1 (αk )a−1 (x)n−2 |0 = 0, represent αj , αk
by real smooth curves Cj , Ck ⊂ X respectively such that Cj ∩ Ck = ∅. Let
y1 , . . . , yn−2 ∈ X be distinct with {y1 , . . . , yn−2 } ∩ (Cj ∪ Ck ) = ∅. Then,
π̃2∗ π̃1∗ PD a−1 (αj )a−1 (αk )a−1 (x)n−2 |0
= π̃2∗ π̃1∗ σ∗ [y1 × · · · × yn−2 × Cj × Ck ]
= f∗ (π̃1 )∗ [y1 × · · · × yn−2 × Cj × Ck ]
= 0.
This completes the proof of the lemma. 

Next, we study the homology classes of curves in C (n) ⊂ X [n] where C denotes
a smooth curve in X. Let gC be the genus of C. Assume that gC ≥ 1. We recall
60 3. HILBERT SCHEMES AND INFINITE DIMENSIONAL LIE ALGEBRAS

some standard facts about C (n) from [ACGH, BT]. For a fixed point p ∈ C, let Ξ
denote the divisor p + C (n−1) ⊂ C (n) . Let
AJ : C (n) → Jacn (C)
be the Abel-Jacobi map sending an element ξ ∈ C (n) to the corresponding degree-n
divisor class in Jacn (C). For an element δ ∈ Jacn (C), the fiber AJ−1 (δ) is the
complete linear system |δ|. Let
Zn (C) ⊂ C (n) × C
be the universal divisor, and let π̃1 , π̃2 be the two projections on C (n) × C. By the
Lemma 2.5 on p.340 of [ACGH] and the Proposition 2.1 (iv) of [BT], we have
(3.28) c1 π̃1∗ OZn (C) = (1 − gC − n)Ξ + Θ
where Θ is the pull-back via AJ of a Theta divisor on Jacn (C).
Lemma 3.20. Let n ≥ 2 and X be a smooth projective surface. Let C ⊂ X be
a smooth curve with genus gC ≥ 1, and Γ ⊂ C (n) be a curve. Then, [Γ] is equal to
(3.29) (Ξ · Γ)βC + − (n + gC − 1)(Ξ · Γ) + (Θ · Γ) βn  2 (X [n] )).
(mod H
In addition, for every line Γ0 in a positive-dimensional fiber AJ−1 (δ), we have
(3.30) [Γ0 ] = βC − (n + gC − 1)βn  2 (X [n] )).
(mod H
 2 (X [n] ) from (3.24). By Lemma 3.19,
Proof. (i) Recall the subspace H
(3.31) [Γ] = βγ + dβn  2 (X [n] ))
(mod H
for some effective cycle γ and some integer d.
To determine d and γ, let Zn be the universal codimension-2 subscheme of
X [n] × X, and let π1 , π2 be the two projections on X [n] × X. Then,
π1∗ OZn |C (n) = π̃1∗ OZn (C) .
By Theorem 1.24 (i), c1 π1∗ OZn = −Bn /2. Combining with (3.31) and (3.28), we
obtain
(3.32) d = (−Bn /2) · Γ
= c1 π1∗ OZn |C (n) · Γ
= c1 π̃1∗ OZn (C) · Γ
= (1 − gC − n)(Ξ · Γ) + (Θ · Γ).
Next, let α ∈ H (X). Then, Dα |C (n) = (C · α)Ξ. Thus, we get
2

Dα · Γ = (C · α)(Ξ · Γ).
By (3.31),
α · γ = Dα · Γ = (C · α)(Ξ · Γ).
It follows that γ = (Ξ · Γ) C.
(ii) Note that Θ · Γ0 = 0. Also, it is known that Ξ|AJ−1 (δ) = OAJ−1 (δ) (1). So
Ξ · Γ0 = 1. Now (3.30) follows from (3.29). 
Remark 3.21. In view of the proof of Lemma 2.19 (i), we conclude that when
the genus gC of the smooth curve C is equal to 0, Lemma 3.20 still holds if we
delete the term (Θ · Γ) in (3.29).
 3.5. VIRASORO ALGEBRAS OF LEHN 61

3.5. Virasoro algebras of Lehn

There is a standard construction which associates a Virasoro algebra to a
Heisenberg algebra. In this section, using the Heisenberg algebras constructed in
Section 3.2, we will introduce the Virasoro algebras and boundary operator of Lehn
[Leh1], and discuss their properties. Up to a factor of (−1/2), the boundary oper-
ator is defined as the cup product by the boundary divisors of the Hilbert schemes
X [n] . These operators play an essential role in understanding the cohomology ring
structure of the Hilbert scheme X [n] . Lehn’s work [Leh1] is the first step in this
direction.
Definition 3.22. The normally ordered product : am1 am2 : is defined by
3
am1 am2 , m1 ≤ m2
: am1 am2 : =
am2 am1 , m1 ≥ m2 .
For n ∈ Z, define the Virasoro operator Ln : H ∗ (X) → End(HX ) by putting
1 
Ln = − · : am an−m : τ2∗ .
2
m∈Z

By (3.9), [am (α), an−m (β)] = 0 if n = 0. It follows that

⎧
⎪ 1 
⎨ −2 ·
⎪ am an−m τ2∗ , if n = 0,
(3.33) Ln =  m∈Z
⎪
⎪
⎩ − a−m am τ2∗ , if n = 0.
m>0

Moreover, a geometric interpretation of the Virasoro operator Ln with n ≥ 0 is

given by [Leh1, Theorem 3.5].
Definition 3.23. Define the boundary operator d ∈ End(HX ) by

d= (−Bn /2)∪
n

where the class (−Bn /2) acts on HnX = H ∗ (X [n] ) by the cup product. For a linear
operator f ∈ End(HX ), define its derivative f by
f = ad(f) = [d, f].
The higher order derivative f(k) of f is defined inductively by
f(k) = [d, f(k−1) ].
The operators Ln (α) and d and the following theorem are due to Lehn [Leh1,
Theorems 3.3 and 3.10].
Theorem 3.24. Let X be a smooth projective complex surface with the canon-
ical class KX and the Euler class eX . Let n, m ∈ Z and α, β ∈ H ∗ (X). Then,
(i) [Lm (α), an (β)] = −n · am+n (αβ).
n(|n| − 1)
(ii) an (α) = n · Ln (α) − · an (KX α).
2
(iii) The operators Ln (α) satisfy the Virasoro algebra commutation relation:
,
m3 − m
[Lm (α), Ln (β)] = (m − n) · Lm+n (αβ) + δm,−n (eX αβ) · IdHX .
12 X
62 3. HILBERT SCHEMES AND INFINITE DIMENSIONAL LIE ALGEBRAS

Proof. We refer to [Leh1] for the proofs of (ii) and (iii) since the argu-
ments are very technical. To prove (i), assume first that m = 0. By (3.33) and
Lemma 3.12 (i),
1 
[Lm (α), an (β)] = − · [ai am−i (τ2∗ α), an (β)]
2
i∈Z
1 
= · iδi,−n am−i (αβ) + (m − i)δm−i,−n ai (αβ)
2
i∈Z
1
= · − nam+n (αβ) − nam+n (αβ)
2
= −n · am+n (αβ).

The case when m = 0 can be similarly proved. 

Next, we address the issue of how to identify two linear operators in End(HX ).
We begin with a lemma.

Lemma 3.25. Let n ≥ 1, α ∈ H ∗ (X), and f ∈ End(HX ) with f = 0. Then,

1  
[f, a−(n+1) (α)] = − · [[f, a−1 (1X )] , a−n (α)] + [a−1 (1X ), [f, a−n (α)]] .
n
Proof. By the Jacobi identity (3.4), we have

[f, a−1 (1X )] = [f , a−1 (1X )] + [f, a−1 (1X ) ]
= [f, a−1 (1X ) ]
= −[a−1 (1X ) , f]

since f = 0. Applying (3.4) one more time, we see that

[a−1 (1X ), [f, a−n (α)]] = [[a−1 (1X ), f], a−n (α)] + [f, [a−1 (1X ), a−n (α)]]
= −[[f, a−1 (1X )] , a−n (α)] + [f, [−L−1 (1X ), a−n (α)]]
= −[[f, a−1 (1X )] , a−n (α)] − [f, na−(n+1) (α)]

where Theorem 3.24 (ii) and (i) have been used. So the lemma holds. 

The following proposition allows us to identify two linear operators in End(HX ).

Proposition 3.26. For i = 1, 2, let fi ∈ End(HX ) such that fi · |0 = 0 and

fi = 0. If [f1 , a−1 (α)] = [f2 , a−1 (α)] for every α ∈ H ∗ (X), then f1 = f2 .

Proof. Recall that the space HX is linearly spanned by the Heisenberg mono-
mial classes (3.10). Therefore, it suffices to show that

(3.34) [f1 , a−n (α)] = [f2 , a−n (α)]

for every n ≥ 1 and α ∈ H ∗ (X). By assumption, (3.34) holds for n = 1. So (3.34)

follows from Lemma 3.25 and an induction on n. 

We end this section with a lemma which will play an important role in the proof
of Lemma 3.29 below. It enables us to use induction on the order of derivatives.
 3.6. HIGHER ORDER DERIVATIVES OF HEISENBERG OPERATORS 63

Lemma 3.27. Let k ≥ 1 and α ∈ H ∗ (X). Then (an1 · · · ank (τk∗ α)) is equal to


k
ni 
− · an1 · · · ani−1 : am1 am2 : ani+1 · · · ank (τ(k+1)∗ α)
i=1
2 m1 +m2 =ni


k
ni (|ni | − 1)
− · an1 · · · ank (τk∗ (KX α)).
i=1
2

Proof. Let

τk∗ α = αj,1 ⊗ · · · ⊗ αj,k
j

via the Künneth decomposition of H ∗ (X k ). By (3.5),

(an1 · · · ank (τk∗ α))
 
= an1 (αj,1 ) · · · ank (αj,k )
j


k  i−1
 
k
= ans (αj,s ) · ani (αj,i ) · ans (αj,s ).
i=1 j s=1 s=i+1

By Theorem 3.24 (ii),

(an1 · · · ank (τk∗ α))

k
ni   i−1
 
k
= − · ans (αj,s )· : am1 am2 : (τ2∗ αj,i ) · ans (αj,s )
i=1
2 m1 +m2 =ni j s=1 s=i+1

k
ni (|ni | − 1)  
i−1 k
− · ans (αj,s ) · ani (KX αj,i ) · ans (αj,s )
i=1
2 j s=1 s=i+1


k
ni 
= − · an1 · · · ani−1 : am1 am2 : ani+1 · · · ank (τ(k+1)∗ α)
i=1
2 m1 +m2 =ni


k
ni (|ni | − 1)
− · an1 · · · ank (τk∗ (KX α))
i=1
2

where we have used the third and first formulas in Lemma 3.11 at the last step. 

3.6. Higher order derivatives of Heisenberg operators

The derivative of a linear operator in End(HX ) is defined in Definition 3.23.
In this section, we will determine the structure of the higher order derivatives
(k)
of Heisenberg operators. We will prove that an (α) has a universal expression,
depending only on the canonical class KX and Euler class eX of the surface X.
The result is essential to the understanding of the Chern character operators in
Section 4.1.
Recall the generalized partitions from Definition 1.6.

Definition 3.28. Let X be a smooth projective complex surface.

64 3. HILBERT SCHEMES AND INFINITE DIMENSIONAL LIE ALGEBRAS

(i) Let α ∈ H ∗ (X), and λ = (· · · (−2)m−2 (−1)m−1 1m1 2m2 · · · ). Define

 m
(3.35) aλ (α) = ai (α) i ,
i
 

mi
(3.36) aλ (τ∗ α) = (ai ) (τ(λ)∗ α)
i
mi
where the product i (ai ) is understood to be
m m
· · · a−2−2 a−1−1 am
1 a2 · · · .
1 m2

(ii) Define
AX = α ∈ H ∗ (X)|KX α = 0
which is an ideal in the cohomology ring H ∗ (X).
(k)
The following lemma determines an (α) when α ∈ AX . The main idea in its
proof is to use induction on k, Lemma 3.27, and Lemma 3.12 (ii).
(k)
Lemma 3.29. Let k ≥ 0, n ∈ Z, and α ∈ AX . Then, an (α) is equal to
⎛ ⎞
 1  s(λ) − 1
(−n)k k! ⎝ aλ (τ∗ α) − aλ (τ∗ (eX α))⎠ .
λ! 24λ!
(λ)=k+1,|λ|=n (λ)=k−1,|λ|=n

Proof. Use induction on k. When k = 0 or 1, the lemma is trivially true.

(k) (k+1) (k) 
Next, assume that the lemma is true for an (α). Since an (α) = an (α) and
(k+1)
KX α = 0, we see from Lemma 3.27 and Lemma 3.12 (ii) that an (α) is of the
form
 
f˜1X (λ)aλ (τ∗ α) + f˜eX (λ)aλ (τ∗ (eX α)).
(λ)=k+2,|λ|=n (λ)=k,|λ|=n

Fix a generalized partition λ = (· · · (−2)m−2 (−1)m−1 1m1 2m2 · · · ) with |λ| = n. In

the following, we compute the coefficients f˜1X (λ) and f˜eX (λ) separately.
(i) We start with the computation of f˜1X (λ). Note that (λ) = k + 2. Going
(α) = (an (α)) , we see that those terms in an (α) whose
(k) (k+1) (k) (k)
from an (α) to an
(k+1)
derivatives contain the term aλ (τ∗ α) in an (α) are of the form aλi,j (τ∗ α) where
i ≤ j, mi ≥ 1, mj ≥ 1, and λi,j stands for the generalized partition obtained
from λ by subtracting 1 from the multiplicities of i and j and by adding 1 to the
multiplicity of (i + j). For fixed i and j, {aλi,j (τ∗ α)} is equal to
 
i+j
− (mi+j + 1)(2 − δi,j )aλ (τ∗ α)
2
 
+ dμ aμ (τ∗ α) + dμ aμ (τ∗ (eX α))
μ=λ (μ)=k,|μ|=n
(μ)=k+2,|μ|=n

(k)
for some constant dμ . By induction, the coefficient of aλi,j (τ∗ α) in an (α) is

(−n)k k! (−n)k k! mi (mj − δi,j )

= · .
(λi,j )! λ! mi+j + 1
 3.6. HIGHER ORDER DERIVATIVES OF HEISENBERG OPERATORS 65

(k+1)
Combining all of these, we see that the coefficient of aλ (τ∗ α) in an (α) is
 (−n)k k! mi (mj − δi,j )  
i+j
· − (mi+j + 1)(2 − δi,j )
λ! mi+j + 1 2
i≤j
⎛ ⎞
(−n)k k! ⎝  
= − mi mj (i + j) − mi (mi − 1)i⎠
λ! i<j i
⎛ ⎞
(−n)k k! ⎝ 1  
= − mi mj (i + j) + mi i⎠
λ! 2 i,j i

(−n)k+1 (k + 1)!
=
λ!
 
where we have used i mi = (λ) = k + 2 and i imi = |λ| = n in the last step.
(ii) Before going into the proof, let us use one example to illustrate the idea.
Take k = 2, n = −3, and λ = ((−5)2). To figure out the coefficient of aλ (τ∗ (eX α))
(3) (2)
in a−3 (α), we apply the induction hypothesis to a−3 (α) and take the derivative
one more time. In view of Lemma 3.12 (ii) and Lemma 3.27, there are exactly two
ways to get aλ (τ∗ (eX α)). The first one is similar to (i) above, and comes from the
term {aλ−5,2 (τ∗ (eX α))} where λ−5,2 = ((−3)), i.e., the generalized partition λ−5,2
is obtained from λ by subtracting 1 from the multiplicities of (−5) and 2 and by
adding 1 to the multiplicity of (−3) = (−5) + 2. The second source comes from
three terms {aλ̃ (τ∗ α)} where λ̃ = ((−5)12 ), ((−5 + j)(−j)2) with j = 1, 2. For
instance, when λ̃ = ((−5 + j)(−j)2) with j = 1 or 2, {aλ̃ (τ∗ (α))} contains
a−5 aj a−j a2 (τ4∗ α) = a−5 a−j aj a2 (τ4∗ α) + (−j)a−5 a2 (τ2∗ (eX α))
= a−5 a−j aj a2 (τ4∗ α) + (−j)aλ (τ∗ (eX α)).
Keeping track of the coefficients in these derivatives
{aλ−5,2 (τ∗ (eX α))} , {aλ̃ (τ∗ α)} ,
(3)
we obtain the coefficient of aλ (τ∗ (eX α)) in a−3 (α).
Now we go back to the proof and compute f˜eX (λ). In this case, (λ) = k. We
(k+1)
will prove that the coefficient of aλ (τ∗ (eX α)) in an (α) is equal to
(−n)k+1 (k + 1)!
·c
λ!
where

1 − s(λ) 1 − i i2 mi
c= = .
24 24
In view of Lemma 3.27 and Lemma 3.12 (ii), there are exactly two sources con-
(k+1)
tributing to the term aλ (τ∗ (eX α)) in an (α) from the derivatives of the terms in
(k)
an (α). In the following, we handle these two sources separately.
The first is similar to (i) above, and comes from the derivatives {aλi,j (τ∗ (eX α))}
where i ≤ j, mi ≥ 1, mj ≥ 1. For fixed i and j, {aλi,j (τ∗ (eX α))} is equal to
  
i+j
− (mi+j + 1)(2 − δi,j )aλ (τ∗ (eX α)) + dμ aμ (τ∗ (eX α))
2 μ=λ
(μ)=k,|μ|=n
66 3. HILBERT SCHEMES AND INFINITE DIMENSIONAL LIE ALGEBRAS

(k)
for some dμ . By induction, the coefficient of aλi,j (τ∗ (eX α)) in an (α) is
 
(−n)k k! 1 − s(λi,j ) (−n)k k! mi (mj − δi,j ) ij
· = · · c − .
(λi,j )! 24 λ! mi+j + 1 12
 
Combining these and noting i mi = k and i imi = n, we see that the total
(k+1)
contribution of the first source to the coefficient of aλ (τ∗ (eX α)) in an (α) is

 (−n)k k! mi (mj − δi,j )  ij


i+j

(3.37) · · c− − (mi+j + 1)(2 − δi,j )
λ! mi+j + 1 12 2
i≤j
  
(−n)k k!  ij i+j
= m i (m j − δ i,j ) c − − (2 − δi,j )
λ! 12 2
i≤j
⎛ ⎞
   
(−n)k k! ⎝  ij  i2
= − mi mj c − (i + j) − mi (mi − 1) c − i⎠
λ! i<j
12 i
12
⎛ ⎞
   
(−n)k k! ⎝ 1  ij  i2
= − mi mj c − (i + j) + mi c − i⎠
λ! 2 i,j 12 i
12
 
(−n) k!k
n  2 1  3
= nc − knc + i mi − i mi .
λ! 12 i 12 i

The second source comes from the derivatives {aλ̃i,j (τ∗ α)} where i = 0, 0 <
2j ≤ |i|, mi ≥ 1, and λ̃i,j is defined as follows. There are two cases, depending on
whether i > 0 or i < 0. For i > 0, define λ̃i,j to be the generalized partition obtained
from λ by subtracting 1 from the multiplicity of i and by adding 1 to the multiplic-
ities of j and (i − j). Then, {aλ̃i,j (τ∗ α)} contains a term (. . . aj . . . ai−j . . .) which
contains a term (. . . aj . . . a−j ai . . .) after expanding the derivative ai−j . When we
(k+1)
switch aj with a−j , we get the term aλ (τ∗ (eX α)) in an (α). For i < 0, define λ̃i,j
to be the generalized partition obtained from λ by subtracting 1 from the multiplic-
ity of i and by adding 1 to the multiplicities of −j and (i + j). Then {aλ̃i,j (τ∗ α)}
contains a term (. . . ai+j . . . a−j . . .) which contains a term (. . . ai aj . . . a−j . . .) after
expanding ai+j . Switching aj with a−j , we get aλ (τ∗ (eX α)) in an
(k+1)
(α).
More precisely, for fixed i > 0 and j, the derivative {aλ̃i,j (τ∗ α)} is

(mi−j + 1)(mj + 1 + δi,2j )

j(i − j) · · aλ (τ∗ (eX α)) +
1 + δi,2j
 
+ dμ aμ (τ∗ (eX α)) + dμ aμ (τ∗ α)
μ=λ (μ)=k+2,|μ|=n
(μ)=k,|μ|=n

(k)
for some constant dμ . By induction, the coefficient of aλ̃i,j (τ∗ (eα)) in an (α) is

(−n)k k! (−n)k k! mi
= !
· .
(λ̃i,j ) ! λ (m i−j + 1)(m j + 1 + δi,2j )
 3.6. HIGHER ORDER DERIVATIVES OF HEISENBERG OPERATORS 67

It follows that the total contribution of the second source with i > 0 to the coeffi-
(k+1)
cient of the term aλ (τ∗ (eX α)) in an (α) is equal to

  (−n)k k! mi
(3.38) !
· ·
i>0
λ (mi−j + 1)(mj + 1 + δi,2j )
0<2j≤i
(mi−j + 1)(mj + 1 + δi,2j )
·j(i − j) ·
1 + δi,2j
(−n)k k!   j(i − j)
= mi
λ! i>0
1 + δi,2j
0<2j≤i
⎛ ⎞
(−n)k k!  1 
= mi ⎝ j(i − j)⎠
λ! i>0
2 0<j<i

(−n)k k!  i3 − i
= m i ·
λ! i>0
12
(−n)k k! 1  3
= · (i mi − imi ).
λ! 12 i>0

Similarly, we see that the total contribution of the second source with i < 0 to the
(k+1)
coefficient of aλ (τ∗ (eX α)) in an (α) is equal to

(−n)k k! 1  3
(3.39) · (i mi − imi ).
λ! 12 i<0

Next, we see from (3.37), (3.38) and (3.39) that the coefficient of the term
(k+1)
aλ (τ∗ (eX α)) in the derivative an (α) is equal to
 
(−n)k k! n  2 1  3
nc − knc + i mi − i mi
λ! 12 i 12 i
(−n)k k! 1  3
+ · (i mi − imi ).
λ! 12 i

 (k+1)
Since n = i imi , the coefficient of the term aλ (τ∗ (eX α)) in an (α) is equal to
 
(−n)k k! n  2 n
nc − knc + i mi −
λ! 12 i 12
(−n)k k!
= (nc − knc − 2nc)
λ!
1 − s(λ)
= (−n)k+1 (k + 1)! .
24λ!
(k+1)
Combining (i) and (ii), we have proved the lemma for an (α). 
68 3. HILBERT SCHEMES AND INFINITE DIMENSIONAL LIE ALGEBRAS

Remark 3.30. Let α ∈ AX . From the proof of Lemma 3.29, we can read that
⎛ ⎞
 1  s(λ) − 1
⎝ aλ (τ∗ α) − aλ (τ∗ (eX α))⎠
λ! 24λ!
(λ)=k+1,|λ|=n (λ)=k−1,|λ|=n
⎛ ⎞
 1  s(λ) − 1
= −n(k + 1) ⎝ aλ (τ∗ α) − aλ (τ∗ (eX α))⎠ .
λ! 24λ!
(λ)=k+2,|λ|=n (λ)=k,|λ|=n

Note that when n = 0, this formula is not covered by Lemma 3.29. Applying this
formula twice in the second equality below, we obtain that for a fixed constant d,
⎛ ⎞
 1  s(λ) + d
⎝ aλ (τ∗ α) − aλ (τ∗ (eX α))⎠
λ! 24λ!
(λ)=k+1,|λ|=n (λ)=k−1,|λ|=n
⎛ ⎞
 1  s(λ) − 1
= ⎝ aλ (τ∗ α) − aλ (τ∗ (eX α))⎠
λ! 24λ!
(λ)=k+1,|λ|=n (λ)=k−1,|λ|=n
⎛ ⎞
d+1⎝  1
− aλ (τ∗ (eX α))⎠
24 λ!
(λ)=k−1,|λ|=n
⎛ ⎞
 1  s(λ) − 1
= −n(k + 1) ⎝ aλ (τ∗ α) − aλ (τ∗ (eX α))⎠
λ! 24λ!
(λ)=k+2,|λ|=n (λ)=k,|λ|=n

n(k − 1)(d + 1)  1
+ · aλ (τ∗ (eX α)).
24 λ!
(λ)=k,|λ|=n

Therefore, we have
⎛ ⎞
 1  s(λ) + d
⎝ aλ (τ∗ α) − aλ (τ∗ (eX α))⎠
λ! 24λ!
(λ)=k+1,|λ|=n (λ)=k−1,|λ|=n
⎛ ⎞
 1  s(λ) + d
= −n(k + 1) ⎝ aλ (τ∗ α) − aλ (τ∗ (eX α))⎠
λ! 24λ!
(λ)=k+2,|λ|=n (λ)=k,|λ|=n

n(d + 1)  1
− · aλ (τ∗ (eX α)).
12 λ!
(λ)=k,|λ|=n

We will need this formula in the proof of Lemma 4.3.

Now we describe a formula of an (α) for general α ∈ H ∗ (X).
(k)

Theorem 3.31. Let k ≥ 0, n ∈ Z, and α ∈ H ∗ (X). Then, an (α) is equal to

(k)
⎛ ⎞
 1  s(λ) − 1
(−n)k k! ⎝ aλ (τ∗ α) − aλ (τ∗ (eX α))⎠
λ! 24λ!
(λ)=k+1,|λ|=n (λ)=k−1,|λ|=n
  f (λ)
+ aλ (τ∗ (α))
2 } (λ)=k+1−| |/2,|λ|=n
λ!
∈{KX ,KX

where all the numbers f (λ) are independent of X and α.

 3.7. THE EXT VERTEX OPERATORS OF CARLSSON AND OKOUNKOV 69

Proof. We see from Lemma 3.27 and Lemma 3.12 (ii) that
  f (λ)
(3.40) a(k)
n (α) = aλ (τ∗ (α))
λ!
∈{1X ,eX ,KX ,KX } (λ)=k+1−| |/2,|λ|=n
2

where all the coefficients f (λ) are independent of X and α.

To determine f1X (λ) and feX (λ), let X be a K3 surface and α = 1X . Then,
(k)
KX α = 0 but eX α = 0. By Lemma 3.29, an (α) is equal to
⎛ ⎞
 1  s(λ) − 1
(−n)k k! ⎝ aλ (τ∗ α) − aλ (τ∗ (eX α))⎠ .
λ! 24λ!
(λ)=k+1,|λ|=n (λ)=k−1,|λ|=n

Comparing with (3.40) and noting that the set of all the Heisenberg monomials
aλ (τ∗ α), aλ̃ (τ∗ (eX α)) with (λ) = k + 1, |λ| = n, (λ̃) = k − 1, |λ̃| = n is linearly
independent, we see that f1X (λ) = (−n)k k! and
(−n)k k!(s(λ) − 1)
feX (λ) = − .
24
This completes the proof of the theorem. 

3.7. The Ext vertex operators of Carlsson and Okounkov

The Nekrasov partition function in supersymmetric quantum gauge theory is
mathematically formulated as an equivariant integral over certain moduli spaces
of sheaves on a complex surface. Okounkov and Nekrasov [NO] showed that this
function, together with the related correlation function, may be studied using the
representation theory of the infinite wedge representation and vertex operators.
Subsequently, Carlsson and Okounkov [Car1, Car2, CO] defined a new vertex op-
erator out of the Chern class of a vector bundle on a pair of Hilbert schemes of
points on the same surface. When the pair consists of the same Hilbert scheme, the
restriction of this vector bundle to the diagonal of the pair coincides with the tan-
gent bundle of the Hilbert scheme. In this section, we will review the construction
of Carlsson and Okounkov, and survey the properties of their Ext vertex operators
which play important roles in understanding the intersection theory of the Hilbert
schemes X [n] when the Chern classes of the tangent bundles TX [n] are involved.
Let X be either a smooth projective complex surface, or a smooth quasi-
projective complex surface with a C∗ -action whose fixed point set is compact. Let
L be a line bundle over the smooth surface X. Let EL be the virtual vector bundle
on X [k] × X [] whose fiber at (ξ, η) ∈ X [k] × X [] is given by
(3.41) EL |(ξ,η) = χ(O, L) − χ(Iη , Iξ ⊗ L)
where Iξ and Iη denote the ideal sheaves of the 0-dimensional closed subschemes ξ
and η respectively, and we have used the notation

2
χ(F, G) = (−1)i Exti (F, G).
i=0

By the Riemann-Roch Theorem, the virtual rank of EL is

rk(EL ) = k + .
70 3. HILBERT SCHEMES AND INFINITE DIMENSIONAL LIE ALGEBRAS

Let Lm be the trivial line bundle on X with a scaling action of C∗ of character

z , and let Δn be the diagonal in X [n] × X [n] . Then,
m

(3.42) ELm |Δn = TX [n] ,m ,

the tangent bundle TX [n] with a scaling action of C∗ of character z m .
Definition 3.32. Let z be a formal variable. The Ext operator
W(L, z) : HX [[z, z −1 ]] → HX [[z, z −1 ]]
is the linear operator defined by
,
(3.43) W(L, z)η, ξ = (η ⊗ ξ) ck+ (EL )
X [k] ×X []

for η ∈ H ∗ (X [k] ) and ξ ∈ H ∗ (X [] ).

Lemma 3.33. Let n be the number-of-points operator, i.e., n|H ∗ (X [n] ) = n Id.
Let TX [n] ,m be the tangent bundle TX [n] with a scaling action of C∗ of character z m .
Then,

+∞
(3.44) q n χ TX [n] ,m = Tr q n W(Lm , z).
n=0

Proof. We will show that the coefficients of q n on both sides of (3.44) are
equal. Let {ej }j be a linear basis of H ∗ (X [n] ). Then the fundamental class of the
diagonal Δn in X [n] × X [n] is given by

(3.45) [Δn ] = (−1)|ej | ej ⊗ e∗j
j

where {e∗j }j is the linear basis of H ∗ (X [n] ) dual to {ej }j in the sense that ej , e∗j  =
δj,j  . By the definition of W(Lm , z),
  
Tr q n W(Lm , z) = q n (−1)|ej | W(Lm , z)ej , e∗j
j
 ,
|ej |
= q n
(−1) (ej ⊗ e∗j ) c2n (ELm )
j X [n] ×X [n]
,
= qn [Δn ] c2n (ELm )
X [n] ×X [n]
,
= qn c2n (ELm |Δn ).
Δn

By (3.42), we have c2n ELm |Δn = c2n TX [n] ,m . Therefore,

,
Tr q n W(Lm , z) = q n c2n TX [n] ,m = q n χ TX [n] ,m .
X [n]
This completes the proof of the lemma. 
To describe the operator W(L, z), we need to define the operators Γ± (L, z).
By abusing notations, we also use L to denote its first Chern class. Define
 
 z ∓n
(3.46) Γ± (L, z) = exp a±n (L) .
n>0
n
 3.7. THE EXT VERTEX OPERATORS OF CARLSSON AND OKOUNKOV 71

Remark 3.34. There is a sign difference between the Heisenberg commutation

relations used in [Car1] (see p.3 there) and in (3.9). So for n > 0, our Heisenberg
operators a−n (L) and an (−L) are equal to the Heisenberg operators a−n (L) and
an (L) in [Car1] respectively. Accordingly, the operators Γ− (L, z) and Γ+ (−L, z)
in (3.46) are equal to the operators Γ− (L, z) and Γ+ (L, z) in [Car1] respectively.
Lemma 3.35. The following commutation relations hold
(3.47) [Γ+ (L, x), Γ+ (L , y)] = [Γ− (L, x), Γ− (L , y)] = 0,

(3.48) Γ+ (L, x)Γ− (L , y) = (1 − y/x)L,L  Γ− (L , y)Γ+ (L, x).
Proof. Formula (3.47) follows immediately from the Heisenberg commutation
relations (3.9). To prove (3.48), we notice from (3.9) again that for n > 0,
 −n   n 
x y
exp an (L) · exp a−n (L )
n n
 1  x−n i  n
y
j

= an (L) · a−n (L )
i! · j! n n
i,j≥0
 1   yn j−r  −n
x
i−r

= a−n (L ) · an (L)
i! · j! n n
i,j≥0 r≤i,j
   8 −n 9r
i j x yn 
· r! an (L), a−n (L )
r r n n
 n   −n   
y  x 1 " y #n 
= exp a−n (L ) · exp an (L) · exp − L, L .
n n n x
It follows that
Γ+ (L, x)Γ− (L , y)
 
 1 " y #n
= exp − L, L · Γ− (L , y)Γ+ (L, x)
n>0
n x

= (1 − y/x)L,L  Γ− (L , y)Γ+ (L, x).
This completes the proof of the lemma. 

The following two results are proved in [Car1,CO], but we follow the notations
in [Car1, Theorem 1] adjusted in view of Remark 3.34.
Theorem 3.36. Let L be a line bundle over the smooth surface X. Then,
(3.49) W(L, z) = Γ− (L − KX , z) Γ+ (−L, z).
We outline the proof of Theorem 3.36, and refer to [Car1,Car2,CO] for further
details. The first step is to use certain universality results and reduce to the special
case of the equivariant cohomology of a smooth toric surface X. Then for a smooth
toric surface X, choose a torus action on X such that the fixed points are isolated.
Some decomposition argument with respect to the isolated fixed points further
reduces the proof to the cases of X = C2 and P1 × P1 . Finally, the cases of X = C2
and P1 × P1 can be verified directly by using equivariant localization.
72 3. HILBERT SCHEMES AND INFINITE DIMENSIONAL LIE ALGEBRAS

Corollary 3.37. Let TX [n] ,m and Lm be the tangent bundle TX [n] and the
trivial line bundle on X respectively with a scaling action of C∗ of character z m .
Let q be a real number with |q| < 1. Then,

+∞
Lm ,KX −Lm −χ(X)
(3.50) q n χ TX [n] ,m = (q; q)∞
n=0
where (a; q)n = (1 − a)(1 − aq) · · · (1 − aq n ).
Proof. By Lemma 3.33, it suffices to prove that
Lm ,KX −Lm −χ(X)
(3.51) Tr q n W(Lm , z) = (q; q)∞ .
By Theorem 3.36, we obtain
Tr q n W(Lm , z) = Tr q n Γ− (Lm − KX , z) Γ+ (−Lm , z)
= Tr Γ− (Lm − KX , zq) q n Γ+ (−Lm , z)
= Tr q n Γ+ (−Lm , z) Γ− (Lm − KX , zq).
By (3.48) and since |q| < 1, we get
Tr q n W(Lm , z)
= (1 − q)Lm ,KX −Lm  Tr q n Γ− (Lm − KX , zq) Γ+ (−Lm , z)
Lm ,KX −Lm 
= (q; q)∞ Tr q n Γ− (Lm − KX , 0) Γ+ (−Lm , z)
Lm ,KX −Lm 
= (q; q)∞ Tr q n
Lm ,KX −Lm −χ(X)
= (q; q)∞
where we have used (1.22) in the last step. 
 CHAPTER 4

Chern character operators

Understanding intersection theory over moduli spaces of sheaves is a fascinat-

ing aspect of research in both mathematics and theoretical physics, with topics in-
cluding Donaldson theory, Donaldson-Thomas theory, S-duality, and the Nekrasov
partition function, to name a few. The Hilbert schemes of points on surfaces can
be viewed as the basic models for moduli spaces of sheaves. To study intersection
over the Hilbert schemes of points on surfaces, one key step is to obtain meaningful
ring generators for their cohomology rings, which can be achieved (Section 7.1)
via Chern character operators. This chapter is devoted to these Chern charac-
ter operators which were introduced in [LQW1] and studied further in [LQW4].
We will define these operators, and prove a universal formula expressing them in
terms of the Heisenberg operators, the canonical class KX , and the Euler class eX .
As applications, we will obtain certain universality of the intersections of Chern
characters on X [n] , and describe a connection between the W algebras and the
Hilbert scheme X [n] . Moreover, we will address the characteristic classes of the
tautological bundles over X [n] . Theorem 4.2, Theorem 4.7, Theorem 4.15 and their
proofs are from [LQW1, Section 5], [LQW4, Subsection 4.2], [LQW3, Section 4]
respectively. The proof of Theorem 4.25 is outlined, and the details can be found
in [LQW4, Section 6].

4.1. Chern character operators

In this section, we will define the Chern character operators Gk (γ) which are
the generalization of Lehn’s boundary operator d introduced in Definition 3.23
(see also Definition 4.17 below, which is from [Leh1]). Then we will compute the
commutator [Gk (γ), a−1 (α)], and determine the universal structure of Gk (γ) in
terms of the Heisenberg operators.
Definition 4.1. Let X be a smooth projective complex surface.
(i) For γ ∈ H ∗ (X), we define G(γ, 0) = 0.
(ii) For γ ∈ H ∗ (X) and n > 0, define
(4.1) G(γ, n) = p1∗ ch(OZn ) · p∗2 td(X) · p∗2 γ ∈ H ∗ (X [n] )
where ch(OZn ) is the Chern character of the sheaf OZn , td(X) is the Todd
class of X, and p1 , p2 are the natural projections of X [n] × X to X [n] , X
respectively. Define the Chern character operator G(γ) ∈ End(HX ) by

+∞
G(γ) = G(γ, n) ∪
n=0

where G(γ, n) acts on HnX = H ∗ (X [n] ) by the cup product.

73
74 4. CHERN CHARACTER OPERATORS

(iii) For i ∈ Z and γ ∈ H s (X), define Gi (γ, n) to be the component of

G(γ, n) in H s+2i (X [n] ). Define the i-th Chern character operator Gi (γ) ∈
End(HX ) by

+∞
Gi (γ) = Gi (γ, n) ∪
n=0

where again Gi (γ, n) acts on HnX = H ∗ (X [n] ) by the cup product.

Notice that Gi (γ) ∈ End(HX ) is homogeneous of bi-degree (0, |γ| + 2i), and
(4.2) Gi (γ) = 0 and Gi (γ)† = Gi (γ).
Also,

G(γ, n) = Gi (γ, n).
i∈Z

Therefore, by the definitions of the operators G(γ) and Gi (γ), we get


G(γ) = Gi (γ).
i∈Z

When γ = 1X , we have
G(1X , n) = p1∗ (ch(OZn ) · p∗2 td(X)).
So we see from the Grothendieck-Riemann-Roch Theorem [Hart, Theorem 5.3 in
Appendix A] that
(4.3) G(1X , n) = ch(p1∗ OZn ) = ch (OX )[n] .
Recall that (OX )[n] is a locally free sheaf of rank-n on X [n] . By Theorem 1.24 (i),
we have c1 (OX )[n] = −Bn /2. It follows that
 
(4.4) G0 (1X ) = G0 (1X , n) ∪ = (n · IdX [n] ) = L0 (1X ),
n n
 
(4.5) G1 (1X ) = G1 (1X , n) ∪ = c1 (p1∗ OZn ) ∪ = d.
n n

Theorem 4.2. Let γ, α ∈ H ∗ (X). Then, we have

(4.6) [G(γ), a−1 (α)] = exp(ad(d))(a−1 (γα)),
or equivalently, using the component Gk (γ) of G(γ), we obtain
1 (k)
(4.7) [Gk (γ), a−1 (α)] = · a (γα).
k! −1
Proof. Let notations be as in Section 1.5. By the exact sequence (1.38),
∗ ∗
(4.8) gX ch(OZn ) − fX ch(OZn−1 ) = ρ∗X ch(OΔX ) · p∗1 exp(−[En ])
where we have used the fact that
ch(ρ∗X OΔX ⊗ p∗1 OX [n,n−1] (−En )) = ρ∗X ch(OΔX ) · p∗1 ch(OX [n,n−1] (−En ))
= ρ∗X ch(OΔX ) · p∗1 exp(−[En ]).

Claim. gn∗ G(γ, n) = fn∗ G(γ, n − 1) + ρ∗ γ · exp(−[En ]).

 4.1. CHERN CHARACTER OPERATORS 75

Proof. For i ≥ 1, let pi,1 and pi,2 be the projections of X [i] × X to X [i] and
X respectively. From the definition of G(γ, n), we see that
gn∗ G(γ, n) = gn∗ (pn,1 )∗ (ch(OZn ) · (pn,2 )∗ td(X) · (pn,2 )∗ γ)
∗
= p1∗ gX (ch(OZn ) · (pn,2 )∗ td(X) · (pn,2 )∗ γ)
∗
= p1∗ (gX ch(OZn ) · p∗2 td(X) · p∗2 γ)
where p2 is the projection of X [n,n−1] × X to X. Similarly,
fn∗ G(γ, n − 1) = p1∗ (fX
∗
ch(OZn−1 ) · p∗2 td(X) · p∗2 γ).
So combining these with (4.8) above, we conclude that
gn∗ G(γ, n) − fn∗ G(γ, n − 1)
∗ ∗
= p1∗ ((gX ch(OZn ) − fX ch(OZn−1 )) · p∗2 td(X) · p∗2 γ)
= p1∗ (ρ∗X ch(OΔX ) · p∗1 exp(−[En ]) · p∗2 td(X) · p∗2 γ)
= p1∗ (ρ∗X ch(OΔX ) · p∗2 td(X) · p∗2 γ) · exp(−[En ])
= p1∗ ρ∗X (ch(OΔX ) · (p1,2 )∗ td(X) · (p1,2 )∗ γ) · exp(−[En ])
= ρ∗ (p1,1 )∗ (ch(OΔX ) · (p1,2 )∗ td(X) · (p1,2 )∗ γ) · exp(−[En ]).
Applying the Grothendieck-Riemann-Roch Theorem to the diagonal embedding
τ2 : X → X × X, we get
ch(OΔX ) · (p1,2 )∗ td(X) = τ2∗ [X].
So
(p1,1 )∗ (ch(OΔX ) · (p1,2 )∗ td(X) · (p1,2 )∗ γ) = (p1,1 )∗ (τ2∗ [X] · (p1,2 )∗ γ) = γ.
It follows that
gn∗ G(γ, n) − fn∗ G(γ, n − 1) = ρ∗ γ · exp(−[En ]).
This completes the proof of the claim. 
We continue the proof of the theorem. By Definition 3.3, for a ∈ H ∗ (X [n−1] ),
a−1 (α)(a) = p̃1∗ ([Q[n,n−1] ] · ρ̃∗ α · p̃∗2 a).
Define ι : X [n,n−1] → X [n] × X × X [n−1] by putting ι(ξ, η) = (ξ, ρ(ξ, η), η). Then,
ι is an embedding. Moreover, ι∗ [X [n,n−1] ] = [Q[n,n−1] ]. Thus, we have
a−1 (α)(a) = p̃1∗ (ι∗ [X [n,n−1] ] · ρ̃∗ α · p̃∗2 a)
= p̃1∗ ι∗ ([X [n,n−1] ] · (ι∗ ◦ ρ̃∗ )α · (ι∗ ◦ p̃∗2 )a)
= gn∗ ([X [n,n−1] ] · ρ∗ α · fn∗ a).
Combining this with the above Claim, we conclude that
G(γ)a−1 (α)(a) = G(γ, n) · gn∗ ([X [n,n−1] ] · ρ∗ α · fn∗ a)
= gn∗ (gn∗ G(γ, n) · [X [n,n−1] ] · ρ∗ α · fn∗ a)
= gn∗ (fn∗ G(γ, n − 1) · [X [n,n−1] ] · ρ∗ α · fn∗ a)
+gn∗ (ρ∗ γ · exp(−[En ]) · [X [n,n−1] ] · ρ∗ α · fn∗ a)
= gn∗ ([X [n,n−1] ] · fn∗ G(γ, n − 1)ρ∗ α · fn∗ a)
+gn∗ ((exp(−[En ]) · [X [n,n−1] ]) · ρ∗ (γα) · fn∗ a).
76 4. CHERN CHARACTER OPERATORS

Thus,
[G(γ), a−1 (α)](a) = gn∗ ((exp(−[En ]) · [X [n,n−1] ]) · ρ∗ (γα) · fn∗ a)
which is equal to exp(ad(d))(a−1 (γα))(a) by the Lemma 3.9 in [Leh1]. 
Recall from Definition 3.28 (ii) that
AX = α ∈ H ∗ (X)|KX α = 0 .
The next lemma determines Gk (α) when α ∈ AX .
Lemma 4.3. Let k ≥ 0, and α ∈ AX . Then, Gk (α) is equal to
 1  s(λ) − 2
(4.9) − !
aλ (τ∗ α) + aλ (τ∗ (eX α)).
λ 24λ!
(λ)=k+2,|λ|=0 (λ)=k,|λ|=0

Proof. Denote the expression (4.9) by f2 . We will apply Proposition 3.26 to

the two operators f1 = Gk (α) and f2 . Note that Gk (α)|0 = 0 and Gk (α) = 0.
From Theorem 4.2 and Lemma 3.29, we see that
[Gk (α), a−1 (β)]
1 (k)
= · a (αβ)
k! −1
 1  s(λ) − 1
= aλ (τ∗ (αβ)) − aλ (τ∗ (eX αβ)).
λ! 24λ!
(λ)=k+1,|λ|=−1 (λ)=k−1,|λ|=−1

Next we need to check the corresponding properties for f2 . Indeed, f2 · |0 = 0

from the definition of f2 . Also, f2 = 0 by Remark 3.30. By Lemma 3.12 (i),
[f2 , a−1 (β)]
 1  s(λ) − 2
= aλ (τ∗ (αβ)) − aλ (τ∗ (eX αβ))
λ! 24(λ1 )! 1
(λ)=k+1,|λ|=−1 (λ)=k,|λ|=0
 1  s(λ1 ) − 1
= aλ (τ∗ (αβ)) − aλ1 (τ∗ (eX αβ))
λ! 24(λ1 )!
(λ)=k+1,|λ|=−1 (λ1 )=k−1,|λ1 |=−1

= [Gk (α), a−1 (β)]

where for a generalized partition λ with (λ) = k, the generalized partition λ1 is
obtained from λ by subtracting 1 from the multiplicity of 1.
Thus, we conclude from Proposition 3.26 that Gk (α) = f2 . 
To understand the operator Gk (α) for a general α ∈ H ∗ (X), we need the notion
of universal linear combinations.
Definition 4.4. Let X be a projective surface, s ≥ 1, and α1 , . . . , αs ∈ H ∗ (X).
Let k1 , . . . , ks ≥ 0, and ni,j ∈ Z with 1 ≤ i ≤ s and 1 ≤ j ≤ ki . A universal linear
combination of ani,1 · · · ani,ki (τki ∗ (αi )), 1 ≤ i ≤ s is a linear combination of the
form
 s
fi (ki , ni,1 , . . . , ni,ki ) ani,1 · · · ani,ki (τki ∗ (αi ))
i=1
where the coefficients fi (ki , ni,1 , . . . , ni,ki ) are independent of X, α1 , . . . , αs . A uni-
versal linear combination of ani,1 · · · ani,ki (τki ∗ (αi ))|0 , 1 ≤ i ≤ s is defined in a
similar way.
 4.1. CHERN CHARACTER OPERATORS 77

Lemma 4.5. Let k ≥ 0, n ∈ Z, and α, β ∈ H ∗ (X). Then, [Gk (α), an (β)] is a

universal linear combination of expressions of the form
am1 · · · amk+1−r (τ(k+1−r)∗ (KX
r
αβ))
where 0 ≤ r ≤ 2 and m1 + . . . + mk+1−r = n.
Proof. First of all, note that a0 (β) = 0. So we may let n = 0. Since Gk (α)† =
Gk (α) and an (β)† = (−1)n a−n (β), we see from (3.3) that
[Gk (α), an (β)]† = −[Gk (α)† , an (β)† ] = (−1)1+n [Gk (α), a−n (β)].
By (3.2), (am1 · · · amk+s−r (τ(k+s−r)∗(KX
r
αβ)))† is equal to
(−1)m1 +...+mk+s−r · a−mk+s−r · · · a−m1 (τ(k+s−r)∗(KX
r
αβ)).
Thus it suffices to prove the lemma for n ≤ −1. When n = −1,
1 (k)
[Gk (α), an (β)] = · a (αβ)
k! −1
by Lemma 4.2. Therefore, the lemma for n = −1 follows by repeatedly applying
Lemma 3.27. When n ≤ −2, we see from Lemma 3.25 that
[Gk (α), an (β)]
1
= · [[Gk (α), a−1 (1X )] , an+1 (β)] + [a−1 (1X ), [Gk (α), an+1 (β)]]
n+1
1
= · [[Gk (α), a−1 (1X )] , an+1 (β)]
n+1
+[a−1 (1X ), [Gk (α), an+1 (β)]] − [a−1 (1X ), [Gk (α), an+1 (β)] ] .
Therefore, the lemma for n ≤ −2 follows from induction, Lemma 3.12 (i), and
Lemma 3.27. 
The formulas in the following proposition are referred to as the transfer prop-
erties.
Proposition 4.6. (The Transfer Properties) Let α, β ∈ H ∗ (X).
(i) Let k, s ≥ 1, n1 , . . . , nk , m1 , . . . , ms ∈ Z. Then,
[an1 · · · ank (τk∗ α), am1 · · · ams (τs∗ β)]
= [an1 · · · ank (τk∗ 1X ), am1 · · · ams (τs∗ (αβ))];
(ii) Let k ≥ 0 and n ∈ Z. Then,
[Gk (α), an (β)] = [Gk (1X ), an (αβ)] = [Gk (αβ), an (1X )].
Proof. The formulas in (i) and (ii) follow from Lemma 3.12 (i) and Lemma 4.5
respectively. 
Now we present a formula of Gk (α) for a general α ∈ H ∗ (X).
Theorem 4.7. Let k ≥ 0 and α ∈ H ∗ (X). Then, Gk (α) is equal to
 1  s(λ) − 2
− aλ (τ ∗ α) + aλ (τ∗ (eX α))
λ! 24λ!
(λ)=k+2,|λ|=0 (λ)=k,|λ|=0
  g (λ)
+ aλ (τ∗ (α))
2 } (λ)=k+2−| |/2,|λ|=0
λ!
∈{KX ,KX

where all the numbers g (λ) are independent of X and α.

78 4. CHERN CHARACTER OPERATORS

Proof. Clearly we may assume α = 0. By (4.4) and the definition of L0 (1X ),

 1
G0 (1X ) = L0 (1X ) = − aλ (τ∗ 1X ).
λ!
(λ)=2,|λ|=0

So our theorem is true for k = |α| = 0. In the following, assume k + |α| > 0.
 we write Gk (α) ∈ End(HX ) as a linear combination
Next,

of Heisenberg mono-
mials (ai (αi,1 ) · · · ai (αi,mi )). For degree reasons, i imi = 0 and
i
 
(4.10) −2 mi + |αi,j | = 2k + |α|.
i i,j

In other words, we regard Gk (α) as an element in the completion of the universal

enveloping algebra of the Heisenberg algebra. We can always do so because HX is
irreducible as a representation of the Heisenberg algebra. Now rewrite Gk (α) as

(4.11) Gk (α) = gλ
|λ|=0

where for each fixed generalized partition λ = (· · · (−2)m−2 (−1)m−1 1m1 2m2 · · · )
with |λ| = 0, the operator
gλ stands for the component in Gk (α) containing all the
expressions of the form (ai (αi,1 ) · · · ai (αi,mi )).
i
Fix n ∈ Z and β ∈ H ∗ (X). By Lemma 4.5 and Lemma 3.12 (ii),

(4.12) [Gk (α), an (β)] = d(, μ)aμ (τ∗ (αβ))
∈{1X ,KX ,K 2 ,eX }
X
(μ)=k+1−||/2,|μ|=n

where all the coefficients d(, μ) ∈ Q are independent of the surface X and the
cohomology classes α, β ∈ H ∗ (X). It follows that the generalized partitions λ in
(4.11) must satisfy k ≤ (λ) ≤ (k + 2). Moreover, for a fixed generalized partition
λ with |λ| = 0 and k ≤ (λ) ≤ (k + 2), and for every i with m−i > 0, we have
⎧
⎨ d(1X , λ−i )aλ−i (τ∗ (αβ)), if (λ) = k + 2
(4.13) [gλ , ai (β)] = d(KX , λ−i )aλ−i (τ∗ (KX αβ)), if (λ) = k + 1
⎩
aλ−i (τ∗ ω), if (λ) = k
where λ−i is the generalized partition obtained from λ by subtracting 1 from
2 2
the multiplicity of (−i), and ω = d(eX , λ−i )eX + d(KX , λ−i )KX αβ. Note that
d(, λ−i ) is independent of X, α and β.
By choosing special surfaces X and α, β ∈ H ∗ (X), we now claim that
2
d(1X , λ−i ) d(KX , λ−i ) d(eX , λ−i ) d(KX , λ−i ) 2
(4.14) , , eX + KX
im−i im−i im−i im−i
are independent of i as long as m−i > 0. Indeed, to prove that the 3rd item in (4.14)
is independent of i as long as m−i > 0, we  may assume that gλ = 0 by (4.13). Note
that |λ| = 0 and (λ) = k. By (4.10), i,j |αi,j | = 4k + |α| for every Heisenberg

monomial (ai (αi,1 ) · · · ai (αi,mi )) contained in gλ . On the other hand, |αi,j | ≤ 4
i
and thus i,j |αi,j | ≤ 4 (λ) = 4k. It follows that |α| = 0 and |αi,j | = 4 for every
i and j. So gλ = b · aλ (τ∗ [x])where b is a nonzero number, and [x] ∈ H 4 (X) is
 4.1. CHERN CHARACTER OPERATORS 79

the cohomology class corresponding to a point x ∈ X. Since α = 0 and |α| = 0,

α = b̃ · 1X for some nonzero number b̃. Choosing β = 1X , we see from (4.13) that
m−i ib · aλ−i (τ∗ [x]) = [gλ , ai (1X )]
= b̃ · aλ−i (τ∗ (d(eX , λ−i )eX + d(KX
2 2
, λ−i )KX )).
So b[x] = b̃ · (d(eX , λ−i )eX + d(KX
2 2
, λ−i )KX )/(im−i ). Thus
2
d(eX , λ−i ) d(KX , λ−i ) 2
eX + KX
im−i im−i
is independent of i as long as m−i > 0. Similarly, by choosing α ∈ H 2 (X) with
KX , α = 0 (respectively, α ∈ H 4 (X) − {0}), we see that d(KX , λ−i )/(im−i )
(respectively, d(1X , λ−i )/(im−i )) is independent of i as long as m−i > 0. This
proves (4.14).
Now, for an arbitrary X and α ∈ H ∗ (X), put
  d(, λ−i )
f= aλ (τ∗ (α)).
im−i
∈{1X ,KX ,KX ,eX } (λ)=k+2−| |/2,|λ|=0
2

We claim that Gk (α) = f. Indeed, for every n ∈ Z − {0} and β ∈ H ∗ (X), we have
[f, an (β)]
  d(, λ−i )
= nm−n · aλ−n (τ∗ (αβ))
2 ,e } (λ)=k+2−| |/2,|λ|=0
im−i
∈{1X ,KX ,KX X
 
= d(, λ−n ) · aλ−n (τ∗ (αβ))
2 ,e } (λ)=k+2−| |/2,|λ|=0
∈{1X ,KX ,KX X

= [Gk (α), an (β)]

where we have used (4.14) in the second equality, and (4.11) and (4.13) in the last
equality. Since HX is irreducible, we see from Schur’s lemma that (Gk (α) − f) must
be a scalar multiple of the identity operator. Recall that the bi-degree of (Gk (α)−f)
is (0, 2k + |α|) which is nontrivial since k + |α| > 0. Thus, (Gk (α) − f) = 0 and
  d(, λ−i )
Gk (α) = f = aλ (τ∗ (α)).
im−i
∈{1X ,KX ,KX ,eX } (λ)=k+2−| |/2,|λ|=0
2

Finally, since the numbers d(, λ−i )/(im−i ) are independent of X and α ∈
H ∗ (X), by using Lemma 4.3 and an argument similar to that in the proof of
Theorem 3.31, we conclude that d(1X , λ−i )/(im−i ) = −1/λ! and
d(eX , λ−i ) s(λ) − 2
= .
im−i 24λ!
In particular, d(eX , λ−i )/(im−i ) is independent of i whenever m−i > 0. So by
(4.14), d(K 2 , λ−i )/(im−i ) is also independent of i whenever m−i > 0. Now denoting
d(, λ−i )/(im−i ) by g (λ)/λ! for  = KX and KX 2
, we have proved the theorem. 
It turns out that the operator G1 (α) can be determined completely. To state
the result, we need the notation : a3 :0 (τ3∗ α). So fix α ∈ H ∗ (X). Define a vertex
operator (i.e. a field) a(α)(z) by putting

(4.15) a(α)(z) = an (α)z −n−1 .
n∈Z
80 4. CHERN CHARACTER OPERATORS

If 
τp∗ (α) = αi,1 ⊗ αi,1 ⊗ . . . ⊗ αi,p ∈ H ∗ (X)⊗p ,
i
then we define

: a(z)p : (τ∗ α) = : a(αi,1 )(z)a(αi,2 )(z) · · · a(αi,p )(z) : .
i

Note that : a(z)p : (τ∗ α) is of conformal weight p. So we can rewrite


(4.16) : a(z)p : (τ∗ α) = : ap :m (τ∗ α) z −m−p ,
m

where the operator : a :m (τ∗ α) ∈ End(HX ) is the coefficient of z −m−p (i.e.

p

the m-th Fourier component of the field : a(z)p : (τ∗ α)), and maps H ∗ (X [n] ) to
H ∗ (X [n+m] ). We have

(4.17) : a3 :0 (τ3∗ α) = a−k a− ak+ + a−k− ak a (τ3∗ α).
k,>0

Proposition 4.8. Let α ∈ H ∗ (X). Then,

1  n−1
(4.18) G1 (α) = − : a3 :0 (τ3∗ α) − : an a−n : (τ2∗ (KX α)).
6 n>0
2

Proof. Observe that both sides of (4.18) annihilate the vacuum vector |0 . To
prove (4.18), it suffices to show that the commutators of both sides of (4.18) with the
operators an (β), n ∈ Z, β ∈ H ∗ (X), coincide. By Lemma 4.5 and Theorem 3.24 (ii),
[G1 (α), an (β)] = [G1 (1X ), an (αβ)]
= an (αβ)
n(|n| − 1)
= n · Ln (αβ) − · an (KX αβ).
2
Using (4.17) and Lemma 3.12 (i), we can easily compute the commutator of the
right hand side of (4.18) with an (β). These two commutators indeed coincide. 
In particular, setting α = 1X , we obtain from (4.18) that
1  n−1
(4.19) d = G1 (1X ) = − : a3 :0 (τ3∗ 1X ) − : an a−n : (τ2∗ KX ).
6 n>0
2
This is equivalent to Theorem 3.24 (ii) of Lehn.
By the definition of Chern character operators (Definition 4.1 (iii)), we have
Gk (α, n) = Gk (α)1X [n] where 1X [n] denotes the fundamental class of the Hilbert
scheme X [n] . It follows that the cohomology class Gk (α, n) can be understood via
Theorem 4.7. This leads us to the next section.

4.2. Chern characters

In this section, using the results from the previous section, we will study the
structure of the cohomology class Gk (α, n) on the Hilbert scheme X [n] . A universal
expression of Gk (α, n) in terms of the Heisenberg monomial classes will be derived.
This further leads to the cup product and intersection number among the classes
Gk (α, n).
We begin with an elementary observation.
 4.2. CHERN CHARACTERS 81

Lemma 4.9. Fix a, b with 1 ≤ a ≤ b. Let g ∈ End(HX ) be of bi-degree ( , s),

and A = a−n1 (β1 ) · · · a−nb (βb )|0 . Then, g(A) is equal to the sum of the following
two terms:
 
a−1
(4.20) ± a−n (β )[[· · · [g, a−nσi (1) (βσi (1) )], · · · ], a−nσi (i) (βσi (i) )]|0
i=0 σi ∈σi0
 a−1 k  
(4.21) + (−1) k=0 (s+ =1 |βσa () |) σa (k)<j<σa (k+1) |βj | a−n (β ) ·
σa 1
∈σa

·[[· · · [g, a−nσa (1) (βσa (1) )], · · · ], a−nσa (a) (βσa (a) )] a−n (β )|0
2
∈σa

where for each fixed i with 0 ≤ i ≤ a, σi runs over all the maps
{ 1, . . . , i } → { 1, . . . , b }
satisfying σi (1) < · · · < σi (i). Moreover,
σi0 = { | 1 ≤ ≤ b, = σi (1), . . . , σi (i)},
σa1 = { | 1 ≤ < σa (a), = σa (1), . . . , σa (a)},
σa2 = { | σa (a) < ≤ b}.
Proof. Note that for all i with 0 ≤ i < a and for all the above σi , we move
[· · · [g, a−nσi (1) (βσi (1) )], · · · ], a−nσi (i) (βσi (i) )]
all the way to the right. This produces (4.20). In doing so, we obtain (4.21) by
repeatedly applying the fact that
g1 g2 = [g1 , g2 ] + (−1)s1 s2 g2 g1
for two operators g1 , g2 ∈ End(HX ) of bi-degrees ( 1 , s1 ), ( 2 , s2 ) respectively. 
Theorem 4.10. Let n ≥ 1, k ≥ 0, and α ∈ H ∗ (X). Then, Gk (α, n) is equal to
  (−1)|λ|−1
· 1−(n−j−1) a−λ (τ∗ α)|0
λ! · |λ|!
0≤j≤k
λ(j+1)
(λ)=k−j+1

  (−1)|λ| |λ| + s(λ) − 2

+ · · 1−(n−j−1) a−λ (τ∗ (eX α))|0
λ! · |λ|! 24
0≤j≤k λ(j+1)
(λ)=k−j−1

  (−1)|λ| g (λ + (1j+1 ))
+ · 1−(n−j−1) a−λ (τ∗ (α))|0
λ! · |λ|!
∈{KX ,K 2 } λ(j+1)
X (λ)=k−j+1−||/2
0≤j≤k

where the universal function g is from Theorem 4.7, and λ + (1j+1 ) is the partition
obtained from λ by adding (j + 1) to the multiplicity of 1.
Proof. By (3.21),
1
· a−1 (1X )n |0 .
1X [n] =
n!
From the definition of Chern character operators, we see that Gk (α)|0 = 0 and
Gk (α, n) = Gk (α)1X [n] . Also, by Theorem 4.7 and Lemma 3.12 (i), we have
[. . . [Gk (α), an1 (α1 )], . . .], ank+2 (αk+2 )] = 0
82 4. CHERN CHARACTER OPERATORS

whenever n1 , . . . , nk+2 < 0. Combining all of these observations, we obtain

Gk (α, n)
1
= · Gk (α)a−1 (1X )n |0
n!
k+1  
1  n
= · a−1 (1X )n−u [· · · [Gk (α), a−1 (1X )], · · · ], a−1 (1X )] |0
n! u=1 u / 01 2
u times

k
1
= · 1−(n−j−1) [· · · [Gk (α), a−1 (1X )], · · · ], a−1 (1X )] |0 .
(j + 1)! / 01 2
j=0
j+1 times

Now our theorem follows immediately from Theorem 4.7. 

Remark 4.11. Fix a cohomology class γ ∈ H ∗ (X). Then we have
(4.22) G0 (γ, n) = 1−(n−1) a−1 (γ)|0
by Theorem 4.10. Using Proposition 4.8, we obtain
(4.23) G1 (γ, n) = G1 (γ)1X [n]
1
= G1 (γ) a−1 (1X )n |0
n!
1
= − · 1−(n−2) a−2 (γ)|0 .
2
Next, we will study the cup product of the cohomology classes Gk (α, n). Our
goal is to express the cup product

s
Gki (αi , n)
i=1

as a universal linear combination of the Heisenberg monomial classes.

Definition 4.12. Let s ≥ 1 and α1 , . . . , αs ∈ H ∗ (X) be homogeneous. For a
partition π = {π1 , . . . , πj } of the set {1, . . . , s}, we fix the orders of the elements
listed in each subset πi (1 ≤ i ≤ j) once and for all, and define (π) = j, απi =
m∈πi αm , and sign(α, π) by


j 
s
απi = sign(α, π) · αi .
i=1 i=1

Lemma 4.13. Let n ≥ 1, s ≥ 0, k1 , . . . , ks ≥ 0, and α1 , . . . , αs ∈ H ∗ (X) be


s
homogeneous. Then, Gki (αi , n) is a finite linear combination of expressions:
i=1

(4.24) sign(α, π) · 1   a−ni,j (τ(mi −ri )∗ (i απi )) · |0
−ri
 mi
(π)
− n− ni,j 1≤i≤(π)
i=1 j=1 1≤j≤mi −ri

whose coefficients are independent of X, α1 , . . . , αs , and the integer n. Here π runs

over all partitions of {1, . . . , s}, i ∈ {1X , KX , KX
2
, eX },

ri = |i |/2 ≤ mi ≤ 2 + kj ,
j∈πi
 4.2. CHERN CHARACTERS 83

m
i −ri 
0 < ni,1 ≤ . . . ≤ ni,mi −ri , ni,j ≤ (kj + 1) for each i, and
j=1 j∈πi

⎛ ⎞

(π) i −ri
m 
s
(4.25) ⎝mi − 2 + ni,j ⎠ = ki .
i=1 j=1 i=1

Proof. Use induction on s. When s = 1, the statement is true by Theo-


s
rem 4.10. Next, let s ≥ 2. By induction, Gki (αi , n) is a linear combination of
i=2
expressions:

⎛ ⎛ ⎞ ⎞

(σ) i −ri
m
sign(α, σ) · 1−(n−ñ) ⎝ ⎝ a−ni,j ⎠ (τ(mi −ri )∗ (i ασi ))⎠ |0
i=1 j=1
⎛ ⎛ ⎞ ⎞
 m i −ri(σ)
sign(α, σ)
= · a−1 (1X )n−ñ ⎝ ⎝ a−ni,j ⎠ (τ(mi −ri )∗ (i ασi ))⎠ |0
(n − ñ)! i=1 j=1

where σ runs over all partitions of {2, . . . , s}, i ∈ {1X , KX , KX

2
, eX },


ri = |i |/2 ≤ mi ≤ 2 + kj ,
j∈σi

m
i −ri   m
(σ) i −ri
0 < ni,1 ≤ . . . ≤ ni,mi −ri , ni,j ≤ (kj + 1), and ñ = ni,j . More-
j=1 j∈σi i=1 j=1
over, the coefficients in the linear combination are independent of X, α2 , . . . , αs and
s
n. Now apply Gk1 (α1 ) to Gki (αi , n), and move Gk1 (α1 ) to the right by using
i=2
Lemma 4.9. Note that |τ(mi −ri )∗ (i ασi )| ≡ |ασi | (mod 2). By Theorem 4.7,

[. . . [Gk1 (α1 ), a1 (β1 )], . . .], ak1 +2 (βk1 +2 )] = 0

when 1 , . . . , k1 +2 < 0. Since Gk1 (α1 )|0 = 0 and

 

s 
s
Gki (αi , n) = Gk1 (α1 ) Gki (αi , n) ,
i=1 i=2
84 4. CHERN CHARACTER OPERATORS


s
we see from Lemma 4.9 that Gki (αi , n) is a universal linear combination of
i=1
expressions:

  |α1 |

|ασi |+

u 
|ασi ||ασw |
sign(α, σ) n − ñ v
(4.26) · (−1) 1≤i≤(σ),i∈U v=1 w>iv ,w∈U
·
(n − ñ)! t
⎛ ⎛ ⎞ ⎞
 i −ri
m
⎜ ⎝ ⎟
·a−1 (1X )n−ñ−t ⎝ a−ni,j ⎠ (τ(mi −ri )∗ (i ασi ))⎠ ·
1≤i≤(σ) j=1
i∈U
⎡
mi1 −ri1

⎣
· · · · [Gk1 (α1 ), a−1 (1X )], · · · ], a−1 (1X )], a−ni1 ,j (τ(mi1 −ri1 )∗ (i1 ασi1 ))],
/ 01 2
j=1
t times
⎛ ⎞ ⎤
miu −riu

· · · ], ⎝ a−niu ,j ⎠ (τ(miu −riu )∗ (iu ασiu ))⎦ |0
j=1

where 0 ≤ t ≤ (k1 + 1), u ≥ 0, (t + u) ≥ 1,

U = {i1 , . . . , iu } ⊂ {1, . . . , (σ)}

with i1 < . . . < iu . Let π be the partition

 of{1, . . . , s} consisting of all the σi with
& &
1 ≤ i ≤ (σ) and i ∈ U , and {1} σi . Then, απ(π) = α1 ασi1 · · · ασiu and
i∈U
(4.27)
 
u 
|α1 | |ασi |+ |ασi ||ασw |
v
sign(α, π) = sign(α, σ) · (−1) 1≤i≤(σ),i∈U v=1 w>iv ,w∈U
.


s
In view of (4.26) and (4.27), Gki (αi , n) is a linear combination of
i=1

⎛ ⎛ ⎞ ⎞
 i −ri
m
(4.28) sign(α, π) · 1−(n−ñ−t) ⎝ ⎝ a−ni,j ⎠ (τ(mi −ri )∗ (i απi ))⎠ ·
1≤i<(π) j=1
⎡
mi1 −ri1

⎣
· · · · [Gk1 (α1 ), a−1 (1X )], · · · ], a−1 (1X )], a−ni1 ,j (τ(mi1 −ri1 )∗ (i1 ασi1 ))],
/ 01 2
j=1
t times
⎛ ⎞ ⎤
miu −riu

· · · ], ⎝ a−niu ,j ⎠ (τ(miu −riu )∗ (iu ασiu ))⎦ |0 .
j=1

with all the coefficients being independent of X, α1 , . . . , αs and n. Also notice that
in the expression (4.28), the only factor depending on n is 1−(n−ñ−t) .
 4.2. CHERN CHARACTERS 85

By Theorem 4.7 and Lemma 3.12, (4.28) is a universal linear combination of

⎛ ⎛ ⎞ ⎞
 mi −ri

sign(α, π) · 1−(n−ñ−t) ⎝ ⎝ a−ni,j ⎠ (τ(mi −ri )∗ (i απi ))⎠ ·

1≤i<(π) j=1

·a−n1 · · · a−nm−r (τ(m−r)∗ (α1 ασi1 · · · ασiu ))|0

⎛ ⎛ ⎞ ⎞
 mi −ri

= sign(α, π) · 1−(n−ñ−t) ⎝ ⎝ a−ni,j ⎠ (τ(mi −ri )∗ (i απi ))⎠ ·

1≤i<(π) j=1

·a−n1 · · · a−nm−r (τ(m−r)∗ (απ(π) ))|0

which is of the form (4.24). Here

m = (k1 − t + 2) + mi1 + . . . + miu − 2u ≤ 2 + kj ,
j∈π(π)

 ∈ {1X , KX , KX
2
, eX }, r = ||/2 ≤ m, 0 < n1 ≤ . . . ≤ nm−r , and
i −ri
 m 
n1 + . . . + nm−r = t + ni,j ≤ (kj + 1)
i∈U j=1 j∈π(π)

since t ≤ (k1 + 1).


s
s
Finally, the cohomology degree of Gki (αi , n) is equal to i=1 (2ki + |αi |).
i=1
Comparing this with the cohomology degree of (4.24), we obtain (4.25). 

The following lemma determines the leading term in the cup product

s
Gki (αi , n).
i=1


s
Lemma 4.14. Let notations be the same as in Lemma 4.13, and n0 = (ki +1).
i=1

 m
(π) i −ri
(i) If an expression from ( 4.24) satisfies ni,j = n0 , then it is equal
i=1 j=1
to

s
1−(n−n0 ) a−(ki +1) (αi ) · |0 .
i=1
s 
s
(ii) The coefficient of 1−(n−n0 ) a−(ki +1) (αi ) · |0 in Gki (αi , n) is
i=1 i=1

s
(−1)ki
.
i=1
(ki + 1)!

Proof. (i) We may let s ≥ 1. Since

(π) mi −ri
  
s
ni,j = (ki + 1),
i=1 j=1 i=1
86 4. CHERN CHARACTER OPERATORS

we see from Lemma 4.13 that

i −ri
m 
ni,j = (kj + 1)
j=1 j∈πi

for every i. By (4.25), we obtain


(π)
(4.29) (mi − 2 + |πi |) = 0
i=1

where |πi | stands for the number of elements in the subset πi . Note that for every
i with 1 ≤ i ≤ (π), we have (mi − ri ) ≥ 1 since
i −ri
m 
ni,j = (kj + 1) ≥ 1.
j=1 j∈πi

So for every i, mi ≥ 1 and ri = 0 if mi = 1. By (4.29), mi = |πi | = 1 for

1 ≤ i ≤ (π). Thus for 1 ≤ i ≤ (π), we have ri = 0, i = 1X , and ni,1 = (kj + 1) if
πi = {j}.
Now let πi = {ti } for 1 ≤ i ≤ (π) = s. Then the expression (4.24) is
 s 

sign(α, π) · 1−(n−n0 ) a−(kti +1) (αti ) |0
i=1
 

s
= 1−(n−n0 ) a−(ki +1) (αi ) |0 .
i=1

(ii) The idea is to use induction on s and track the proof of Lemma 4.13 more
carefully. When s = 1, the statement is true by Theorem 4.10. Next, let s≥ 2
s s
and ñ0 = (kj + 1). Assume that the coefficient of 1−(n−ñ0 ) a−ni (αi ) |0
j=2 i=2

s s
(−1)ki
in the cup product Gki (αi , n) is equal to . Tracking the proof
i=2 i=2
(ki + 1)!
of Lemma  4.13 and applying  Theorem 4.7, we conclude that the coefficient of
s s s
(−1)ki
1−(n−n0 ) a−(ki +1) (αi ) |0 in Gki (αi , n) is equal to . 
i=1 i=1 i=1
(ki + 1)!

We now apply Lemma 4.13 to study certain intersection numbers in the Hilbert
scheme X [n] . For this purpose, we establish the notation
,
(4.30) ω1 , . . . , ωk = ω1 · · · ωk .
Y

where ω1 , . . . , ωk ∈ H ∗ (Y ) and Y stands for a smooth projective variety.

Theorem 4.15. Let n, s ≥ 1, k1 , . . . , ks ≥ 0, and let α1 , . . . , αs ∈ H ∗ (X) be
s
homogeneous cohomology classes. Assume that (2ki + |αi |) = 4n. Then,
i=1
< =

s
Gki (αi , n)
i=1
 4.3. CHARACTERISTIC CLASSES OF TAUTOLOGICAL BUNDLES 87


(π)
is a finite linear combination of sign(α, π) · i απi where π runs over all par-
i=1
titions of {1, . . . , s}, i ∈ {1X , KX , KX
2
, eX }. Moreover, all the coefficients in this
linear combination are independent of X, α1 , . . . , αs and n.
Proof. Note that a−1 (x)n |0 is a generator of H 4n (X [n] ) ∼
< C.
= So in view =

s
of Lemma 4.13, an expression (4.24) nontrivially contributing to Gki (αi , n)
i=1
must satisfy:
(i) ni,j = 1 for all 1 ≤ i ≤ (π) and 1 ≤ j ≤ mi − ri ,
(ii) i απi = i απi · x for all 1 ≤ i ≤ (π), and

(π)
(iii) n − (mi − ri ) = 0.
i=1
Since
/ ⊗ ·01
τk∗ (x) = x · · ⊗ x2
k times
for all k ≥ 0, our conclusion follows immediately from (4.24). 

s
Remark 4.16. Assume that (ki + 2) = 2n. By Theorem 4.2,
i=1

[Gki (x), a−1 (x)] = 0

for all 1 ≤ i ≤ s. Using Theorem 4.7 and Theorem 4.10 repeatedly, we see that
< s =
  s  (−1)|λi |−1
Gki (x, n) = .
i=1 0≤j ≤k ,...,0≤js ≤ks i=1 λ (j +1)
(λi )! |λi |!
1 1 i i
(j1 +1)+...+(js +1)=n (λi )=ki −ji +1

Note that this number is independent of the projective surface X.

4.3. Characteristic classes of tautological bundles

A basic and important class of sheaves over the Hilbert scheme X [n] consists
of the tautological bundles defined by (1.25). This section is devoted to the char-
acteristic classes of these tautological bundles. We will state Lehn’s formula for
the generating series of the total Chern class of the rank-n tautological bundle L[n]
where L denotes a line bundle on X. Moreover, we will prove its generalization by
inserting a formal variable .
To begin with, for a smooth quasi-projective complex variety Y , let K(Y ) be
the Grothendieck group generated (over C) by locally free sheaves, or equivalently
by arbitrary coherent sheaves.
Let X be a smooth projective complex surface. Recall from (1.25) that if F is
a rank-r locally free sheaf on X, then F [n] is a rank-rn locally free sheaf on X [n] .
If
0 → F1 → F2 → F3 → 0
is an exact sequence of locally free sheaves on X, then we have an exact sequence
0 → (F1 )[n] → (F2 )[n] → (F3 )[n] → 0
88 4. CHERN CHARACTER OPERATORS

of locally free sheaves on X [n] . Therefore, we obtain a well-defined group homo-

morphism:
−[n] : K(X) → K(X [n] ).
Definition 4.17. Let u ∈ K(X). Define c(u) and ch(u) ∈ End(HX ) to be
the linear operators which act on the n-th component H ∗ (X [n] ) by cup product
with the total Chern class c(u[n] ) and the Chern character ch(u[n] ) respectively.
Define ci (u) and chi (u) ∈ End(HX ) to be the i-th components of c(u) and ch(u)
corresponding to the cup products by ci (u[n] ) and chi (u[n] ) respectively.
The definition above and theorem below are from [Leh1, Section 4]. Defi-
nition 4.1 and Theorem 4.2 are the generalizations of Definition 4.17 and (4.31)
respectively.
Theorem 4.18.
(i) Let u ∈ K(X) be a class of rank r, and α ∈ H ∗ (X). Then,
(4.31) [ch(u), a−1 (α)] = exp(ad(d))(a−1 (ch(u)α)),
 r − k  (v)
(4.32) c(u) · a−1 (α) · c(u)−1 = a−1 ck (u)α .
v
v,k≥0

(ii) Let C(u) be the right-hand-side of (4.32) with α = 1X . Then,


+∞
(4.33) c(u[n] ) = exp(C(u)) · |0 .
n=0
In particular, for a line bundle L on X, we have
⎛ ⎞

+∞  (−1)r−1
(4.34) c(L[n] ) = exp ⎝ a−r (c(L))⎠ · |0 .
n=0
r
r≥1

Proof. By the Grothendieck-Riemann-Roch Theorem and (4.1),

G(ch(u), n) = ch(u[n] ).
So G(ch(u)) = ch(u), and (4.31) follows from (4.6). The proof of (4.32) is similar
to the proof of Theorem 4.2. Next, recall that 1X [n] = 1/n! · a−1 (1X )n |0 . Thus,

+∞ 
+∞
c(u[n] ) = c(u) 1X [n]
n=0 n=0
= c(u) exp a−1 (1X ) · |0
= c(u) exp a−1 (1X ) c(u)−1 · |0
= exp c(u)a−1 (1X )c(u)−1 · |0
= exp(C(u)) · |0
where we have used (4.32) in the last step. This proves (4.33). Formula (4.34) and
its proof will be generalized in Proposition 4.19 and its proof below. 
We remark that (4.34) was conjectured by Göttsche. Assume that C is a smooth
curve in the surface X. Then, C [n] is an n-dimensional closed subscheme of X [n] .
Let L = OX (C). By (1.28),
 [n] 
C = cn L[n] .
 4.3. CHARACTERISTIC CLASSES OF TAUTOLOGICAL BUNDLES 89

Reading the top Chern classes cn L[n] from (4.34), we recover the following formula
of Nakajima [Nak5, Formula (9.16)]:
⎛ ⎞

+∞
 [n]   (−1)r−1
(4.35) C = exp ⎝ a−r ([C])⎠ · |0 .
n=0
r
r≥1

Proposition 4.19. Let  and w be two formal variables, and let


n
c (L[n] ) = ci (L[n] )i
i=0
where L denotes a line bundle on the surface X. Then,
⎛ ⎞

+∞  (−)r−1
(4.36) c (L[n] )wn = exp ⎝ a−r (c (L))wr ⎠ · |0 .
n=0
r
r≥1

Proof. We modify the proof of (4.34) in [Leh1] by inserting the formal vari-
able  in suitable positions. First of all, for a class u ∈ K(X), put

c (u) = ci (u)i .
i

Then the modification of (4.32) is

 r − k  (v)
−1
(4.37) c (u) · a−1 (α) · c (u) = a−1 ck (u)α v+k .
v
v,k≥0

where r denotes the rank of u. In particular,

(4.38) c (L) · a−1 (1X ) · c (L)−1 = C (L)
where
C (L) = a−1 (c (L)) + a−1 (1X ).
As in the proof of (4.33), we obtain

+∞
(4.39) c (L[n] )wn = c (L) exp a−1 (1X )w · |0 .
n=0

Since c (L) · a−1 (1X ) = C (L) · c (L), we have

 +∞ 
d  [n] n
c (L )w
dw n=0
d " #
= c (L) exp a−1 (1X )w · |0
dw
= c (L) · a−1 (1X ) · exp a−1 (1X )w · |0
= C (L) · c (L) · exp a−1 (1X )w · |0

+∞
= C (L) · c (L[n] )wn .
n=0

So (4.39) is the solution to the initial-value-problem:

d
(4.40) X = C (L) · X, X(0) = |0 .
dw
90 4. CHERN CHARACTER OPERATORS

Next, define two operators:

 (−)r−1
R (w) = a−r (c (L))wr ,
r
r≥1
⎛ ⎞
 (−)r−1
S (w) = exp ⎝ a−r (c (L))wr ⎠ = exp (R (w)) .
r
r≥1

Then, S (0)|0 = |0 , and

d 
(4.41) S (w)|0 = S (w) · (−)r a−r−1 (c (L))wr · |0 .
dw
r≥0

Moreover, by Theorem 3.24 (i) and (ii),

(4.42) [a−1 (1X ), R (w)] = −[L−1 (1X ), R (w)]

= (−)r a−r−1 (c (L))wr .
r≥1

In particular, [a−1 (1X ), R (w)] commutes with R (w). It follows that
[a−1 (1X ), R (w)m ] = mR (w)m−1 [a−1 (1X ), R (w)].
Therefore, we have
[C (L), S (w)] = [a−1 (1X ), exp (R (w))]
= exp (R (w)) · [a−1 (1X ), R (w)]

= S (w) · (−)r a−r−1 (c (L))wr
r≥1

where we have used (4.42) in the last step. Combining with a−1 (1X )|0 = 0, we
obtain
C (L)S (w)|0

= S (w)C (L)|0 + S (w) · (−)r a−r−1 (c (L))wr · |0
r≥1

= S (w)a−1 (c (L))|0 + S (w) · (−)r a−r−1 (c (L))wr · |0
r≥1

= S (w) · (−) a−r−1 (c (L))wr · |0
r

r≥0
d
= S (w)|0
dw
where we have used (4.41) in the last step. Hence, S (w)|0 is also a solution to
(4.40), and is equal to (4.39). 
Corollary 4.20. Let L be a line bundle on the surface X, and let L[n]∗ be the
dual of L[n] . Then,
⎛ ⎞

+∞ 1
(4.43) c(L[n]∗ )wn = exp ⎝ a−r (c(L∗ ))wr ⎠ · |0 .
n=0
r
r≥1

Proof. This follows from (4.36) by setting  = −1. 

 4.4. W ALGEBRAS AND HILBERT SCHEMES 91

The generating series


+∞ 
+∞
c(u[n] ), ch(u[n] )
n=0 n=0

and similar generating series for the tangent bundles TX [n] of the Hilbert schemes
X [n] have been further investigated in [Boi1, BN, QY].

4.4. W algebras and Hilbert schemes

The geometric construction of Lehn’s Virasoro algebras acting on the coho-
mology of the Hilbert schemes of points has been presented in Section 3.5. The
Chern character operators constructed in Section 4.1 are vertex operators of higher
conformal weights. Another distinguished class of vertex algebras, which are called
W algebras and are higher-spin generalizations of Virasoro algebras, often appears
in connection with conformal field theories and with representation theory of affine
Kac-Moody Lie algebras [FB]. Among W algebras, a well-known example is given
by the so-called W1+∞ algebra which is the central extension of the Lie algebra of
differential operators on the circle [FKRW, Kac]. In this section, we will study
the interplay between W algebras and the Hilbert schemes X [n] . By introducing
the linear operators Jpn (α), we will construct geometrically a W algebra action on
the Fock space HX . The proof of the main result (Theorem 4.25) will be sketched.
We begin with the definition of W algebras. Let

d
A= An
n=0

(where Ad ∼ = C) be a finite dimensional graded complex vector space. We write

|α| = n if α ∈ An , and define a super structure (i.e. the Z2 -grading) on A by letting

A0̄ = An ,
n∈2Z+

A1̄ = An
n∈1+2Z+

(here Z+ stands for the set of all nonnegative integers). We assume that A affords an
algebra structure which is super commutative and compatible with the Z2 -grading.
We further assume that there exists a linear operator Tr : A → C which is zero on
the graded subspace An unless n is the top degree. This defines a supercommutative
bilinear form ·, · : A × A → C by
α, β = Tr(αβ).
d
Let t be an indeterminate and let ∂t = . Let Was be the associative algebra
dt
of regular differential operators on the circle S . Denote by W(A)as the associative
1

superalgebra Was ⊗ A. It has a linear basis given by

Jkp (α) = −tp+k (∂t )p ⊗ α, p ∈ Z+ , k ∈ Z, α ∈ A.
A different linear basis of W(A)as is given by
Lpk (α) = −tk Dp ⊗ α, p ∈ Z+ , k ∈ Z, α ∈ A,
92 4. CHERN CHARACTER OPERATORS

where D = t∂t . Note that f (D)t = tf (D + 1) for every f (w) ∈ C[w]. Hence,
Jkp (α) = −tk D(D − 1) · · · (D − p + 1) ⊗ α.
Let W(A) denote the Lie superalgebra obtained from the associative superal-
gebra W(A)as by taking the usual super bracket of operators:
[X ⊗ α, Y ⊗ β] = (XY − Y X) ⊗ (αβ).
Assuming that α and β are of homogeneous degree, we see that
[X ⊗ α, Y ⊗ β] = (−1)1+|α|·|β| [Y ⊗ β, X ⊗ α].
The commutation relation in W(A) is given by
[tr f (D) ⊗ α, ts g(D) ⊗ β]
(4.44) = tr+s (f (D + s)g(D) − f (D)g(D + r)) ⊗ (αβ).
When A = C (hence d = 0), we will simply write the Lie superalgebra W(A)
as W, which is the usual Lie algebra of differential operator on the circle [FKRW,
Kac]. The algebra W affords a universal central extension which is usually referred
to as the W1+∞ algebra. In W1+∞ , the operators L0k generate a Heisenberg algebra,
while the operators L1k generate a Virasoro algebra.
The algebra W(A) has a natural weight filtration
W(A)0 ⊂ W(A)1 ⊂ W(A)2 ⊂ . . . ⊂ W(A)
by letting the weight of Lpk (α) be p. We have

[W(A)p , W(A)q ] ⊂ W(A)p+q−1 .

Denote the associated graded algebra by GW(A). The leading term (according to
the weight filtration) of Jkp (α) is just Lpk (α), and so Jkp (α) and Lpk (α) give rise to
the same element in the graded algebra GW(A) which is denoted by Lpk (α). We
derive the following commutation relation from (4.44):

[Lpm (α), Lqn (β)] = (qm − pn) · Lp+q−1

m+n (αβ).

>
The W (super)algebra W(A) is a central extension of the Lie superalgebra
GW(A) by a one-dimensional center with a specified generator C:

(4.45) >
0 −→ CC −→ W(A) −→ GW(A) −→ 0,
>
such that the commutators in W(A) = GW(A) + CC are given by:
3
mδm,−n Tr(αβ) · C, if p = q = 0,
(4.46) [Lpm (α), Lqn (β)] =
(qm − pn) · Lp+q−1
m+n (αβ), otherwise.
>
In this section, we will be mainly interested in the W (super)algebra W(A)
∗
when A is (a subring of) the cohomology ring H (X) of a projective surface X,
with the trace defined by
,
Tr(α) = − α
X
for α ∈ H ∗ (X). First of all, we introduce the following definitions.
 4.4. W ALGEBRAS AND HILBERT SCHEMES 93

Definition 4.21. Let X be a smooth projective complex surface.

(i) For p ≥ 0, n ∈ Z and α ∈ H ∗ (X), define Jpn (α) ∈ End(HX ) to be
⎛ ⎞
 1  s(λ) + n 2
− 2
p! · ⎝− aλ (τ∗ α) + aλ (τ∗ (eX α))⎠ ;
λ! 24λ!
(λ)=p+1,|λ|=n (λ)=p−1,|λ|=n

(ii) We define WX to be the linear span of the identity operator IdHX and the
operators Jpn (α) in End(HX ), where p ≥ 0, n ∈ Z and α ∈ H ∗ (X).
Some of the operators Jpn (α) can be identified with the familiar ones, by using
the above definition, Theorem 4.7 and Theorem 3.31. For example, for α ∈ H ∗ (X),
we see that
J0n (α) = −an (α),
J1n (α) = Ln (α).
For α ∈ AX , we obtain
Jp0 (α) = p! · Gp−1 (α),
(p)
Jp−1 (α) = −a−1 (α).
In general, we have the following.
Lemma 4.22. For p ≥ 0, α ∈ AX and β ∈ H ∗ (X), the following identity holds:
n p
[Gp (α), an (β)] = · J (αβ).
p! n
Proof. Follows from Definition 4.21 (i), Theorem 4.7, and Lemma 3.12 (i). 
By Lemma 4.22, we may regard the linear operators Jpn (α) ∈ End(HX ) as
being geometric since both the Chern character operators Gp (α) and the Heisenberg
operators an (β) are geometric by their constructions in Section 4.1 and Section 3.2
respectively.
We are interested in the commutation relation among the operators Jpm (α). To
this end, it is useful to adopt the vertex algebra language [Bor, FB, Kac] for our
present setup. Our convention for fields is to write them in a form

φ(z) = φn z −n−Δ
n

where Δ is the conformal weight of the field φ(z). Define


∂φ(z) = (−n − Δ)φn z −n−Δ−1
n

which is called the derivative field of φ(z). We set


φ− (z) = φn z −n−Δ ,
n≥0

φ+ (z) = φn z −n−Δ .
n<0

If ψ(z) is another vertex operator, we define a new vertex operator, which is called
the normally ordered product of φ(z) and ψ(z), to be:
: φ(z)ψ(z) : = φ+ (z)ψ(z) + (−1)φψ ψ(z)φ− (z)
94 4. CHERN CHARACTER OPERATORS

where (−1)φψ is −1 if both φ(z) and ψ(z) are odd (i.e. fermionic) fields and 1
otherwise. Inductively we can define the normally ordered product of k vertex
operators φ1 (z), φ2 (z), . . . , φk (z) from right to left by

: φ1 (z)φ2 (z) · · · φk (z) : = : φ1 (z)(: φ2 (z) · · · φk (z) :) : .

For α ∈ H ∗ (X), recall the field a(α)(z) from (4.15), and the notations : a(z)p :
(τ∗ α) and : ap :m (τ∗ α) from (4.16). For r ≥ 1, we can similarly define the field
: (∂ r a(z))a(z)p−1 : (τ∗ α), and define the operator : (∂ r a)ap−1 :m (τ∗ α) as the
coefficient of z −m−r−p in : (∂ r a(z))a(z)p−1 : (τ∗ α).
We remark that when the variable z or the cohomology class α is clear or
irrelevant in the context, we will drop z or α from the notations of fields. For
instance, we will sometimes use : ap : (τ∗ α) or : ap :, : (∂ r a)ap : (τ∗ α) or : (∂ r a)ap :,
etc., to stand for : a(z)p : (τ∗ α), : (∂ r a(z))a(z)p : (τ∗ α), etc., respectively.

Lemma 4.23. In terms of fields, the operator Jpm (α) can be rewritten as:

1
(4.47) − : ap+1 :m (τ∗ α)
(p + 1)
1
+ p(m2 − 3m − 2p) : ap−1 :m (τ∗ (eX α))
24
1
+ p(p − 1) : (∂ 2 a) ap−2 :m (τ∗ (eX α)).
24
Proof. First of all, by the definition of : ap+1 :m (τ∗ α), we have
 1
p! · aλ (τ∗ α)
λ!
(λ)=p+1,|λ|=m
1 
= : ai1 · · · aip+1 : (τ∗ α)
(p + 1) i
1 +...+ip+1 =m

1
= : ap+1 :m (τ∗ α).
(p + 1)

Next, recalling the definition of s(λ), we obtain

 s(λ) + m2 − 2
p! · aλ (τ∗ (eX α))
λ!
(λ)=p−1,|λ|=m
p−1 
 
= p· i2b + m2 − 2 : ai1 · · · aip−1 : (τ∗ (eX α))
i1 +...+ip−1 =m b=1
p−1 
 
= p· (ib + 1)(ib + 2) + m − 3m − 2p ·
2

i1 +...+ip−1 =m b=1

: ai1 · · · aip−1 : (τ∗ (eX α))

= p(p − 1) : (∂ 2 a) ap−2 :m (τ∗ (eX α)) + p(m2 − 3m − 2p) : ap−1 :m (τ∗ (eX α)).

Now the lemma follows from the definition of the operator Jpm (α). 
 4.4. W ALGEBRAS AND HILBERT SCHEMES 95

Remark 4.24. More explicitly, we can rewrite the operator Jpm (α) as the m-th
Fourier component of a vertex operator as follows:
Jpm (α)
1 p
= − : ap+1 :m (τ∗ α) + (∂ 2 : ap−1 :)m (τ∗ (eX α))
(p + 1) 24
(p + 1)p p(p2 − p − 2) p−1
+ (∂ : ap−1 :)m (τ∗ (eX α)) + :a :m (τ∗ (eX α))
12 24
1
+ p(p − 1) : (∂ 2 a)ap−2 :m (τ∗ (eX α)).
24
In practice, (4.47) suffices for the purpose of computations below. Also, since
Jp+1
0 (α) = (p + 1)! · Gp (α)
for α ∈ AX , we conclude that the Chern character operator Gp (α) with α ∈ AX is
the zero-mode of a vertex operator.

m,n for m, n, p, q ∈ Z as follows:

Next, define an integer Ωp,q
(4.48) Ωp,q
m,n = mp3 n2 + 3mp2 n2 q − p2 nq + p2 qn3 − 3mp2 n2 + pnq
+3m2 pnq − 3mpn2 q − m3 q 2 p − pqn3 − mpq + m3 pq
+mpq 2 + 2mpn2 − 3m2 pnq 2 − 2m2 nq + 3m2 nq 2 − m2 nq 3 .
Theorem 4.25. The space WX is closed under the Lie bracket. More explicitly,
for m, n ∈ Z and α, β ∈ H ∗ (X), the following commutation relation holds:
p+q−1 Ωp,q
m,n
[Jpm (α), Jqn (β)] = (qm − pn) · Jm+n (αβ) − · Jp+q−3
m+n (eX αβ)
12
where (p, q) ∈ Z2+ except for the unordered pairs (0, 0), (1, 0), (2, 0) and (1, 1). In
addition, for these four exceptional cases, the following identities hold:
,
[J0m (α), J0n (β)] = −mδm,−n (αβ) · IdHX ,
X
[J1m (α), J0n (β)] = −n · 0
Jm+n (αβ),
,
m3 − m
[J2m (α), J0n (β)] = −2n · J1m+n (αβ) + δm,−n (eX αβ) · IdHX ,
6 X
,
m3 − m
[J1m (α), J1n (β)] = (m − n) · J1m+n (αβ) + δm,−n (eX αβ) · IdHX .
12 X

Note that our algebra WX contains as subalgebras the Heisenberg algebra (Sec-
tion 3.2) of Nakajima and Grojnowski generated by the operators
J0m (α) = −am (α),
and the Virasoro algebra (Section 3.5) of Lehn generated by the operators
J1m (α) = Lm (α).
We denote
BX = {α ∈ H ∗ (X)|eX α = KX α = 0}
which is an ideal in the ring H ∗ (X). Obviously, BX ⊂ AX . Denote by
WX
B
96 4. CHERN CHARACTER OPERATORS

the linear span of Jpn (α), where p ≥ 0, n ∈ Z and α ∈ BX . From the commutation
relation in Theorem 4.25 we observe that WX B
is a Lie subalgebra of the Lie (su-
>
per)algebra WX . Recall the (super)algebra W(A) introduced in (4.45) for a general
ring A. The next theorem follows from comparing the commutator (4.46) and the
ones given in Theorem 4.25.
Theorem 4.26. Let X be a smooth projective complex surface. Then, the
> X ) is isomorphic to the Lie (super)algebra W B by sending
Lie (super)algebra W(B X
C → IdHX and Ln (α) → Jpn (α), where p ≥ 0, n ∈ Z, and α ∈ BX .
p

Remark 4.27. Let X be a smooth projective complex surface.

(i) If the canonical class KX and Euler class eX are numerically trivial, then
the ideal BX is the entire cohomology ring H ∗ (X), and thus the algebra
WX B
coincides with WX . In addition, Theorem 4.25 implies that the field

Jp (α)(z) = Jpn (α)z −n−p−1
n∈Z

is a primary field of conformal weight (p + 1) with respect to the Virasoro

field.
(ii) In general, we note that the leading term in the commutators of the al-
gebra WX given in Theorem 4.25 is precisely the W algebra associated
to the ring H ∗ (X). Therefore, we are justified to regard the W algebras
WX and WX B
in general as certain topological deformations of the W1+∞
algebra in the framework of Hilbert schemes.
The proof of Theorem 4.25 is very technical and uses the operator product
expansion (OPE) technique. In the rest of this section, we will outline the proof,
and refer to [LQW4, Section 6] for further details. Denote the three terms in (4.47)
by 1 Jpm (α), 2 Jpm (α) and 3 Jpm (α) respectively, that is,
Jpm (α) = 1 Jpm (α) + 2 Jpm (α) + 3 Jpm (α).
Note that [i Jpm (α), j Jqn (β)] = 0 (i, j = 2, 3) by Lemma 3.12 (i) and e2X = 0. Thus,
(4.49) [Jpm (α), Jqn (β)]
= [1 Jpm (α), 1 Jqn (β)] + [1 Jpm (α), 2 Jqn (β)] + [1 Jpm (α), 3 Jqn (β)]
+[2 Jpm (α), 1 Jqn (β)] + [3 Jpm (α), 1 Jqn (β)].
Following [Kac, FB], we see that the commutation relations of the Heisenberg
generators can be recast equivalently in terms of the operator product expansion
(OPE) as:
?
− X (αβ)
(4.50) a(α)(z) a(β)(w) ∼ .
(z − w)2
Here and below, ∼ means that the regular terms with respect to (z − w) on the
right hand side of the OPEs are omitted; it is well known that the meromorphic
terms carry all the information about the corresponding commutation relations.
We derive from (4.50) the following two OPEs:
? ?
−2 X (αβ) −6 X (αβ)
(4.51) a(α)(z) ∂a(β)(w) ∼ , a(α)(z) ∂ a(β)(w) ∼
2
.
(z − w)3 (z − w)4
 4.4. W ALGEBRAS AND HILBERT SCHEMES 97

Our goal is to compute the commutator [Jpm (α), Jqn (β)] by using the OPE tech-
nique. To this end, we need to compute the following OPEs
(4.52) (: a(z)p+1 : (τ∗ α))(: a(w)q+1 : (τ∗ β)),
(4.53) (: a(z)p+1 : (τ∗ α))(: a(w)q−1 : (τ∗ (eX β))),
(4.54) (: a(z)p+1 : (τ∗ α))(: ∂ 2 a(w)a(w)q−2 : (τ∗ (eX β))),
which correspond to the first three commutators on the right-hand-side of (4.49)
respectively. Then we will be able to rewrite the terms appearing in these OPEs as
combinations of (the derivative fields of) the four fields
: a(z)p+q : (τ∗ (αβ)),
: a(z)p+q−2 : (τ∗ (eX αβ)),
: ∂ 2 (a(z))a(z)p+q−3 : (τ∗ (eX αβ)),
: ∂ 3 (a(z))a(z)p+q−3 : (τ∗ (eX αβ)),
by using the following simple lemma.
Lemma 4.28. As in the previous section, write the fields ar (z), ∂ s a(z), etc.,
simply as ar , ∂ s a, etc., respectively. Then for N ≥ 0, we have the following:
(i) ∂ 2 : aN := N : (∂ 2 a)aN −1 : +N (N − 1) : (∂a)2 aN −2 :
(ii) ∂ 3 : aN : is equal to
   
N N
N : (∂ 3 a)aN −1 : +6 : (∂a)(∂ 2 a)aN −2 : +6 : (∂a)3 aN −3 :
2 3
(iii) ∂ : aN ∂ 2 a :=: (∂ 3 a)aN : +N : (∂a)(∂ 2 a)aN −1 :
(iv) ∂ : (∂a)2 aN := 2 : (∂a)(∂ 2 a)aN : +N : (∂a)3 aN −1 :.
Proof. Using the Leibnitz rule, we obtain
∂ 2 : aN : = ∂(N : (∂a)aN −1 :)
= N : (∂ 2 a)aN −1 : +N (N − 1) : (∂a)2 aN −2 : .
This proves part (i). The proofs of the remaining formulas are similar. 
Another ingredient in calculating various OPEs is the Wick Theorem [Kac,
Theorem 3.3]. The Wick Theorem in the vertex algebra literature is a very powerful
tool for calculating the OPE of two normally ordered products of free fields.
Theorem 4.29. (Wick Theorem) Let a1 (z), . . . , aM (z) and b1 (z), . . . , bN (z) be
two collections of free fields such that the following properties hold:
 
(i) [ai (z)− , bj (w)], ck (z) = 0 for all i, j, k, and c = a or b,
(ii) [ai (z)± , bj (w)± ] = 0 for all i and j.
Let [ai bj ] = [ai (z)− , bj (w)] be the contraction of ai (z) and bj (w). The one has the
following OPE in the domain |z| > |w|:
: a1 (z) · · · aM (z) : : b1 (w) · · · bN (w) :

min(M,N )
 "
= ± [ai1 bj1 ] · · · [ais bjs ]
s=0 i1 <···<is
j1 =···=js
#
: a1 (z) · · · aM (z)b1 (w) · · · bN (w) :(i1 ,...,is ;j1 ,...,js )
98 4. CHERN CHARACTER OPERATORS

where the subscript (i1 , . . . , is ; j1 , . . . , js ) means that the fields

ai1 (z), . . . , ais (z), bj1 (w), . . . , bjs (w)
are removed, and the sign ± is obtained by the usual super rule: each permutation
of the adjacent odd fields changes the sign.
The Wick Theorem has to be modified in order to apply to our setup. First
of all, we naturally incorporate the transfer property, Lemma 3.12 (i). Next, by
Lemma 3.12 (ii), the contraction of two pairs of free fields will give rise to a term
depending on eX αβ. So the contraction of more than two pairs of fields will be
zero thanks to e2X = 0.
 CHAPTER 5

Multiple q-zeta values and Hilbert schemes

As applications of the results in Chapter 3 and Chapter 4, in this chapter

we will compute the generating series for the intersection pairings between the
total Chern classes of the tangent bundles of the Hilbert schemes of points on a
smooth projective complex surface and the Chern characters of tautological bundles
over these Hilbert schemes. As pointed out in [Car1], these generating series
may be regarded as the generalization of the Nekrasov partition function and the
related correlation function from the equivariant setting to the non-equivariant
setting. Modulo the lower weight term, we will verify Okounkov’s conjecture [Ok2,
Conjecture 2] which connects the Hilbert schemes of points on surfaces and multiple
q-zeta values. In addition, this conjecture is completely proved when the surface
is abelian. Okounkov’s conjecture is another example of the interplay of algebraic
geometry, representation theory and analytic number theory. The main approach is
to use the Ext vertex operators of Carlsson and Okounkov discussed in Section 3.7
and the structure of the Chern character operators proved in Section 4.1. This
chapter follows the presentation of [QY], and we refer to [QY, Section 6] for further
applications which determine some universal constants in the sense of Boissière and
Nieper-Wisskirchen [Boi1, BN] regarding the total Chern classes of the tangent
bundles of these Hilbert schemes.

5.1. Okounkov’s conjecture

In this section, we will review Okounkov’s conjecture from [Ok2]. For integers
k1 , . . . , kN ≥ 0 and cohomology classes α1 , . . . , αN ∈ H ∗ (X) on a smooth projective
complex surface X, we will define the generating series Fkα11,...,k ,...,αN
N
(q) and express
it in terms of the Ext vertex operators of Carlsson and Okounkov discussed in
Section 3.7.
In the region Re s > 1, the Riemann zeta function is defined by

+∞
1
ζ(s) = .
n=1
ns
The integers s > 1 give rise to a sequence of special values of the Riemann zeta
function. Multiple zeta values are series of the form
 1
ζ(s1 , . . . , sk ) =
n
n >···>n >0 1
s1
· · · nskk
1 k

where n1 , . . . , nk denote positive integers, and s1 , . . . , sk are positive integers with

s1 > 1. Their Q-linear span is a subalgebra of R and denoted by MZV. Multiple
q-zeta values ζq (s1 , . . . , sk ) are q-deformations of ζ(s1 , . . . , sk ), which may take
different forms as indicated by [Bra1, Bra2, OT, Zhao, Zud]. In this chapter, we
99
100 5. MULTIPLE q-ZETA VALUES AND HILBERT SCHEMES

adopt the convention that ζq (s1 , . . . , sk ) is a multiple q-zeta value if

lim (1 − q)s1 +...+sk ζq (s1 , . . . , sk ) ∈ MZV.
q→1

The weight of ζq (s1 , . . . , sk ) is defined to be s1 + . . . + sk .

Next, we switch to the setting of Hilbert schemes of points. Let X be a smooth
projective complex surface, and let L be a line bundle on X. Let chk (L[n] ) be the
[n]
k-th Chern character of the tautological  rank-n bundle  L over the Hilbert scheme
X . We define the generating series chk1 · · · chkN by
[n] L1 LN

,
 L1   n [n] [n]
(5.1) chk1 · · · chL
kN
N
= q chk1 (L1 ) · · · chkN (LN ) · c(TX [n] )
n≥0 X [n]

where 0 < q < 1, and c TX [n] is the total Chern class of the tangent bundle TX [n] .
By (1.22),
1 = (q; q)−χ(X)
∞

where

n
(a; q)n = (1 − aq i ).
i=0
 
The reduced series chL
k1 · · · chL
kN is defined by
 L1   L1 
(5.2) chk1 · · · chLN
kN = chk1 · · · chL
kN / 1
N

 
= (q; q)χ(X)
∞ · chLk1 · · · chkN .
1 LN

In [Car1, Car2], for X = C2 with a suitable C∗ -action and L = OX , the

 L 

series chLk1 · · · chk in the equivariant setting has been studied, and proved to be
quasi-modular in q [Car2, Theorem 2 on p.2904]. Using this as evidence, Okounkov
[Ok2, Conjecture 2] proposed the following interesting conjecture.
 
conjecture 5.1. The reduced series chL k1 · · · chkN
L
is a multiple q-zeta value
N
of weight i=1 (ki + 2).

 A motivation  for Conjecture 5.1 is Okounkov’s observation that the series

chLk1 · · · chkN
LN
1
can be written as the trace over the Fock space of a product of
a certain vertex operator (that is, the Ext operator) and a certain integral of mo-
tion of the second quantized trigonometric Calogero-Sutherland system. This will
be carried out via (5.41) and Lemma 5.2 below in terms of the Chern character
operators (the relation with Calogero-Sutherland systems is the subject of Sec-
tion 7.4). The structure (Theorem 4.7) of these Chern character operators will help
   
with understanding chL k1 · · · chkN k1 · · · chkN
LN
1
and hence the reduced series chL L
.
Conjecture 5.1 may lead to further understanding of the graded algebra of multiple
q-zeta values, and provides another example of the interplay of algebraic geometry,
representation theory, and analytic number theory.
We will partially verify Conjecture 5.1. To achieve this, for cohomology classes
α1 , . . . , αN ∈ H ∗ (X) and integers k1 , . . . , kN ≥ 0, we define the series
N 
α1 ,...,αN
 , 
n
(5.3) Fk1 ,...,kN (q) = q Gki (αi , n) c TX [n]
n X [n] i=1
 5.1. OKOUNKOV’S CONJECTURE 101

where the classes Gki (αi , n) are defined in Definition 4.1 (iii). Note that
 , 
F (q) = 1 = qn c TX [n] = χ(X [n] )q n = (q; q)−χ(X)
∞
n X [n] n

by (1.22). Recall the line bundle Lm and Ext operator W(L, z) from Section 3.7.
The following lemma is similar to Lemma 3.33.

Lemma 5.2. Let n be the number-of-points operator, i.e., n|H ∗ (X [n] ) = n Id.
Then,


N
(5.4) Fkα11,...,k
,...,αN
N
(q) = Tr q n W(L1 , z) Gki (αi ).
i=1

Proof. We use the same arguments as in the proof of Lemma 3.33. We will
show that the coefficients of q n on both sides of (5.4) are equal. Let {ej }j be a
linear basis of H ∗ (X [n] ). By (3.45),

[Δn ] = (−1)|ej | ej ⊗ e∗j
j

where {e∗j }j is the linear basis of H ∗ (X [n] ) dual to {ej }j in the sense that ej , e∗j  =
δj,j  . By the definitions of W(L, z) and Gk (α), we have


N
Tr q n W(L1 , z) Gki (αi )
i=1
< N  =
 
= qn (−1)|ej | W(L1 , z) Gki (αi ) ej , e∗j
j i=1
,  N  
 
|ej |
= q n
(−1) Gki (αi ) ej ⊗ e∗j c2n (EL1 )
j X [n] ×X [n] i=1
,  N  
 
|ej |
= q n
(−1) Gki (αi , n) ej ⊗ e∗j c2n (EL1 )
j X [n] ×X [n] i=1
, N 
 
|ej |
= q n
(−1) (ej ⊗ e∗j ) p∗1 Gki (αi , n) c2n (EL1 )
j X [n] ×X [n] i=1
, N 

= q n
[Δn ] p∗1 Gki (αi , n) c2n (EL1 )
X [n] ×X [n] i=1

where p1 : X [n] × X [n] → X [n] denotes the first projection. By (3.42), we have
c2n EL1 |Δn = c TX [n] . Therefore,
, N 

N 
n n
Tr q W(L1 , z) Gki (αi ) = q Gki (αi , n) c TX [n] .
i=1 X [n] i=1

This completes the proof of the lemma. 

102 5. MULTIPLE q-ZETA VALUES AND HILBERT SCHEMES

5.2. The series Fkα11,...,k

,...,αN
N
(q)
In this section, we will first determine the structure of the trace

N
a λ(i) (τ∗ αi )
(5.5) Tr q n W(L1 , z)
i=1
λ(i)!

where λ(1) , . . . , λ(N ) are generalized partitions, and the notation aλ(i) (τ∗ αi ) is de-
fined in (3.36). Then, the structure of the generating series Fkα11,...,k
,...,αN
N
(q) will follow
from Lemma 5.2, Theorem 4.7 and the structure of the trace (5.5).
We begin with four technical lemmas. To explain the ideas behind these lem-
mas, note from (3.49) that the trace (5.5) is equal to


N
a λ(i) (τ∗ αi )
(5.6) Tr q n Γ− (L1 − KX , z) Γ+ (−L1 , z) .
i=1
λ(i)!
 n 
aλ (τ∗ α) z
Lemma 5.3 deals with the commutator between and exp a−n (γ) .
λ! n
It enables us to eliminate Γ− (L1 − KX , z) from (5.6) in Lemma 5.4, and allows
us to eliminate Γ+ (−L1 , z) from (5.6) in Lemma 5.5. Lemma 5.7 determines the
structure of the trace
N
aλ(i) (τ∗ αi )
Tr q n .
i=1
λ(i)!
The proofs of these lemmas are standard but lengthy.
Recall from Definition 1.6 that P  denotes the set of generalized partitions. If
λ = · · · (−2)s−2 (−1)s−1 1s1 2s2 · · · and μ = · · · (−2)t−2 (−1)t−1 1t1 2t2 · · · , let

λ − μ = · · · (−2)s−2 −t−2 (−1)s−1 −t−1 1s1 −t1 2s2 −t2 · · ·

with the convention that λ − μ = ∅ if si < ti for some i.

Lemma 5.3. Let λ ∈ P. Assume that γ ∈ H even (X). Then,

 n 
aλ (τ∗ α) z
(5.7) exp a −n (γ)
λ! n
 n  
z (−z n )i aλ−(ni ) (τ∗ (γ i α))
= exp a−n (γ) · ,
n i! (λ − (ni ))!
i≥0
 n 
z aλ (τ∗ α)
(5.8) exp an (γ) ·
n λ!
 (−z n )i aλ−((−n)i ) (τ∗ (γ i α))  n 
z
= exp an (γ) .
i! (λ − ((−n)i ))! n
i≥0

Proof. Note that the adjoint of am (β) is equal to (−1)m a−m (β). So (5.8)
follows from (5.7) by taking the adjoint on both sides of (5.7) and by making
suitable adjustments. To prove (5.7), put
 n 
aλ (τ∗ α) z
A= exp a−n (γ) .
λ! n
 α ,...,αN
5.2. THE SERIES Fk 1,...,k (q) 103
1 N

Then,
 t
aλ (τ∗ α)  1 zn
A = a−n (γ)
λ! t! n
t≥0
 t t  
1  1 zn  t t−i  
= a−n (γ) · · · [aλ (τ∗ α), a−n (γ)], . . . , a−n (γ) .
λ! t! n i / 01 2
t≥0 i=0
i times

Let λ = · · · (−2)
 (−1) 1 2 · · · . We conclude
s−2 s−1 s1 s2
 from Lemma 3.12 (i) that
the commutator · · · [aλ (τ∗ α), a−n (γ)], . . . , a−n (γ) is equal to
/ 01 2
i times

sn (sn − 1) · · · (sn + 1 − i) (−n)i aλ−(ni ) (τ∗ (γ i α))

where by our convention, λ − (ni ) = ∅ if sn < i. So A is equal to
 t t  
1  1 zn  t t−i
a−n (γ) ·
λ! t! n i=0
i
t≥0

·sn (sn − 1) · · · (sn + 1 − i) (−n)i aλ−(ni ) (τ∗ (γ i α)).

Simplifying this, we complete the proof of our formula (5.7). 
+ = P be the subset of P consisting of the usual partitions, and P
Let P − be the

subset of P consisting of generalized partitions of the form · · · (−2) (−1)s−1 .
s−2

Lemma 5.4. Let λ(1) , . . . , λ(N ) ∈ P be generalized partitions, and α1 , . . . , αN ∈

∗
H (X). Then, the trace

N
a λ(i) (τ∗ αi )
(5.9) Tr q n W(L1 , z)
i=1
λ(i)!
is equal to
  (−(zq s )n )mn(i,s)
(5.10) (i,s)
·

μ(i,s) ∈P +
1≤i≤N mn !
s,n≥1
1≤i≤N,s≥1

 (i,s)
N a (i) 
λ − s≥1 μ(i,s) τ∗ ((1X − KX )
mn
s,n≥1 αi )
·Tr q n Γ+ (−L1 , z)  !
i=1 λ(i) − s≥1 μ(i,s)
where μ(i,s) = 1m1
(i,s) (i,s)
· · · n mn + for 1 ≤ i ≤ N and s ≥ 1.
··· ∈ P
Proof. For simplicity, denote the trace (5.9) by Q1 . By (3.49),

N
a λ(i) (τ∗ αi )
Q1 = Tr q Γ− (L1 − KX , z) Γ+ (−L1 , z)
n

i=1
λ(i)!

N
a λ(i) (τ∗ αi )
(5.11) = Tr Γ− (L1 − KX , zq) q n Γ+ (−L1 , z)
i=1
λ(i)!

N
aλ(i) (τ∗ αi )
= Tr q n Γ+ (−L1 , z) · Γ− (L1 − KX , zq).
i=1
λ(i)!
104 5. MULTIPLE q-ZETA VALUES AND HILBERT SCHEMES

By (3.46) and applying (5.7) repeatedly, we obtain


N
aλ(i) (τ∗ αi )
· Γ− (L1 − KX , zq)
i=1
λ(i)!
 

N  (zq)n
a λ(i) (τ∗ αi )
= · exp a−n (L1 − KX )
i=1
λ(i)! n>0
n
  (−(zq)n )mn(i,1)
= Γ− (L1 − KX , zq) (i,1)
·

μ(i,1) ∈P +
1≤i≤N mn !
n≥1
1≤i≤N


(i,1)
N
aλ(i) −μ(i,1) τ∗ ((1X − KX ) n≥1 mn
αi )
· !
.
i=1 λ(i) − μ(i,1)
(i,1) (i,1)
where μ(i,1) = 1m1 · · · n mn · · · . Therefore, Q1 is equal to
  (−(zq)n )m(i,1)
n
Tr q n Γ+ (−L1 , z)Γ− (L1 − KX , zq) (i,1)
·

μ(i,1) ∈P +
1≤i≤N mn !
n≥1
1≤i≤N



N
aλ(i) −μ(i,1) τ∗ ((1X − KX ) n≥1 m(i,1)
n αi )
· !
.
i=1 λ(i) − μ(i,1)
Since L1 , L1 − KX = 0, we see from (3.48) that Q1 is equal to
  (−(zq)n )mn(i,1)
Tr q Γ− (L1 − KX , zq)Γ+ (−L1 , z) ·
n
(i,1)
·

μ(i,1) ∈P +
1≤i≤N mn !
n≥1
1≤i≤N



N
aλ(i) −μ(i,1) τ∗ ((1X − KX ) n≥1 m(i,1)
n αi )
· !
.
i=1 λ(i) − μ(i,1)
Repeat the above process beginning at line (5.11) s times. Then, Q1 is equals to
  (−(zq r )n )mn
(i,r)

Tr q n Γ− (L1 − KX , zq s )Γ+ (−L1 , z) · (i,r)

·

μ(i,r) ∈P +
1≤i≤N mn !
1≤r≤s,n≥1
1≤i≤N,1≤r≤s

s 
 m(i,r)
N
aλ(i) −sr=1 μ(i,r) τ∗ ((1X − KX ) r=1 n≥1 n αi )
·  !
i=1 λ(i) − sr=1 μ(i,r)
(i,r) (i,r)
where μ(i,r) = 1m1 · · · n mn · · · . Letting s → +∞ proves our lemma. 

Lemma 5.5. Let λ̃(1) , . . . , λ̃(N ) ∈ P be generalized partitions, and α̃1 , . . . , α̃N ∈
∗
H (X). Then, the trace

N
a λ̃(i) (τ∗ α̃i )
(5.12) Tr q n Γ+ (−L1 , z)
i=1
λ̃(i)!
 α ,...,αN
5.2. THE SERIES Fk 1,...,k (q) 105
1 N

is equal to
  (z −1 q t−1 )nm̃n(i,t)
(i,t)
N (i) |−
 m̃n !
i=1 (|λ̃ t≥1 |μ̃(i,t) |)=0 1≤i≤N
t,n≥1

N a (i) 
λ̃ − t≥1 μ̃(i,t) (τ∗ α̃i )
·Tr q n 
(i) − (i,t) !
i=1 λ̃ t≥1 μ̃

where μ̃(i,t) = · · · (−n)m̃n

(i,t)
· · · (−1)m̃1
(i,t)
− for 1 ≤ i ≤ N and t ≥ 1.
∈P
Proof. For simplicity, denote the trace (5.12) by Q2 . By (3.46),
  N
 z −n  a (i) (τ∗ α̃i )
λ̃
Q2 = Tr q n exp an (−L1 ) .
n>0
n i=1
λ̃(i)!
Applying (5.8) repeatedly, we see that Q2 is equal to
  z −nm̃n(i,1) 
N
aλ̃(i) −μ̃(i,1) (τ∗ α̃i )
(i,1)
· Tr q n !
· Γ+ (−L1 , z)

μ̃(i,1) ∈P −
1≤i≤N m̃n ! i=1 λ̃(i) − μ̃(i,1)
n≥1
1≤i≤N

where μ̃(i,1) = · · · (−n)m̃n

(i,1)
· · · (−1)m̃1
(i,1)
− . Now
∈P
Q2
  z −nm̃(i,1)
n N (i,1)
|−|λ̃(i) |)
= (i,1)
·q i=1 (|μ̃


μ̃(i,1) ∈P −
1≤i≤N m̃n !
n≥1
1≤i≤N


N
aλ̃(i) −μ̃(i,1) (τ∗ α̃i )
·Tr !
· q n Γ+ (−L1 , z)
i=1 λ̃(i) − μ̃(i,1)
  z −nm̃(i,1)
n N (i,1)
|−|λ̃(i) |)
= (i,1)
·q i=1 (|μ̃


μ̃(i,1) ∈P −
1≤i≤N m̃n !
n≥1
1≤i≤N


N
aλ̃(i) −μ̃(i,1) (τ∗ α̃i )
·Tr q n Γ+ (−L1 , z) !
.
i=1 λ̃(i) − μ̃(i,1)
By degree reasons, Tr q n Γ+ (−L1 , z)aμ (β) = 0 if |μ| > 0. If |μ| = 0, then we have
Tr q n Γ+ (−L1 , z)aμ (β) = Tr q n aμ (β).
So Q2 is equal to
  z −nm̃n(i,1) N (i,1)
|−|λ̃(i) |)
(i,1)
·q i=1 (|μ̃

N (i) |−|μ̃(i,1) |)<0 1≤i≤N m̃n !

i=1 (|λ̃ n≥1


N
aλ̃(i) −μ̃(i,1) (τ∗ α̃i )
·Tr q n Γ+ (−L1 , z) !
i=1 λ̃(i) − μ̃(i,1)

  z −nm̃(i,1)
n 
N
aλ̃(i) −μ̃(i,1) (τ∗ α̃i )
+ (i,1)
· Tr q n !
.
N (i) |−|μ̃(i,1) |)=0 1≤i≤N m̃n ! i=1 λ̃(i) − μ̃(i,1)
i=1 (|λ̃ n≥1
106 5. MULTIPLE q-ZETA VALUES AND HILBERT SCHEMES

Repeating the process in the previous paragraph t times, we conclude that

Q2 = U (t) − V (t)
where U (t) is given by
⎛ ⎞
 t
⎜  z −nm̃(i,r)
n N r (i) ⎟
· q i=1 ( =1 |μ̃ |−|λ̃ |) ⎠ ·
(i,)
(5.13) ⎝ (i,r)
N t m̃n !
i=1 (|λ̃
(i) |−
r=1 |μ̃(i,r) |)<0 r=1 1≤i≤N
n≥1

N a 
λ̃(i) − tr=1 μ̃(i,r) (τ∗ α̃i )
(5.14) ·Tr q Γ+ (−L1 , z)
n
t
(i) − (i,r) !
i=1 λ̃ r=1 μ̃

with μ̃(i,r) = · · · (−n)m̃n

(i,r) (i,r)
· · · (−1)m̃1 ∈P − , and V (t) is given by
⎛ ⎞
 ⎜  z
t −n m̃ (i,r)
n N r (i) ⎟
· q i=1 ( =1 |μ̃ |−|λ̃ |) ⎠ ·
(i,)
⎝ (i,r)
N  m̃n !
i=1 (|λ̃(i) |− t |μ̃(i,r) |)=0 r=1
r=1
1≤i≤N
n≥1

N a 
λ̃(i) − tr=1 μ̃(i,r) (τ∗ α̃i )
·Tr q n t .
(i) − (i,r) !
i=1 λ̃ r=1 μ̃

 (t). Since
Denote line (5.13) by U

N 
t
(|λ̃(i) | − |μ̃(i,r) |) < 0
i=1 r=1

and |μ̃(i,r) | < 0, U (t) is a polynomial in q with coefficients being bounded in terms of
N  (t). Line (5.14) is bounded in terms of the generalized
− i=1 |λ̃ |. Moreover, q t |U
(i)

partitions λ̃ . Since 0 < q < 1, U (t) → 0 as t → +∞. Letting t → +∞, we see

(i)

that Q2 equals
⎛ ⎞
  ⎜  z −nm̃n(i,t) N t (i) ⎟
· q i=1 ( =1 |μ̃ |−|λ̃ |) ⎠ ·
(i,)
⎝ (i,t)
N  m̃n !
(|λ̃(i) |−
i=1 |μ̃(i,t) |)=0 t≥1
t≥1
1≤i≤N
n≥1

N a (i) 
λ̃ − t≥1 μ̃(i,t) (τ∗ α̃i )
·Tr q n  .
(i) − (i,t) !
i=1 λ̃ t≥1 μ̃
N t N 
Replacing q i=1 ( =1 |μ̃(i,) |−|λ̃(i) |)
by q − i=1 ≥t+1 |μ̃(i,) |
proves our lemma. 

We remark that the assumption 0 < q < 1 was used in the above proof.
However, for most parts of this chapter (e.g. in (5.4)), q can also be viewed as a
formal variable.
Let

N
a λ(i) (τ∗ αi )
(5.15) AN = Tr q n
i=1
λ(i)!
 α ,...,αN
5.2. THE SERIES Fk 1,...,k (q) 107
1 N

 are generalized partitions, and α1 , . . . , αN ∈ H ∗ (X) are

where λ(1) , . . . , λ(N ) ∈ P
homogeneous. Our Lemma 5.7 below deals with the trace AN . To illustrate its
proof, let us consider the following example.
Example 5.6. Let N = 3, α1 = α3 = 1X , α2 = x, λ(1) = (−4), λ(2) = (4), and
(3)
λ = ((−5)5) in (5.15). Then,
A3 = Tr q n a−4 (1X )a4 (x)a((−5)5) (τ2∗ 1X ).
To evaluate A3 , we first apply Case 2 in the proof of Lemma 5.7 to eliminate
a−4 (1X )a4 (x). By (5.21) with i0 = 1 and n0 = 4,
q4
A3 = Tr q n [a4 (x), a−4 (1X )]a((−5)5) (τ2∗ 1X )
1 − q4
(−4)q 4
= · x, 1X · Tr q n a((−5)5) (τ2∗ 1X ).
1 − q4
Then we apply Case 1 in the proof of Lemma 5.7 to eliminate a((−5)5) (τ2∗ 1X ). By
(5.20) with n1 = 5,
(−4)q 4 −χ(X) (−5)q 5
A3 = · x, 1X · (q; q) ∞ · eX , 1X ·
1 − q4 1 − q5
4
4q 5q 5
= (q; q)−χ(X)
∞ · x, 1X · eX , 1X · · .
1 − q4 1 − q5
Lemma 5.7. Let λ(1) , . . . , λ(N ) ∈ P be generalized partitions, and α1 , . . . , αN ∈
∗
H (X) be homogeneous. Then, the trace (5.15) can be computed by an induction
on N , and is a linear combination of expressions of the form:
< = v
 u   nki q ni
−χ(X)
(5.16) (q; q)∞ · Sign(π) · mi
eX , αj · i

i=1 j∈πi i=1

1 − q ni
N
where 0 ≤ v ≤ i=1 (λ )/2, mi ≥ 0, ni > 0, all the integers involved and
(i)

the partition {π1 , . . . , πu } of {1, . . . , N } depend only on λ(1) , . . . , λ(N ) , and Sign(π)
depends on the parity of the degrees of α1 , . . . , αN and is the sign defined by
u 
(5.17) αj = Sign(π) · α1 · · · αN .
i=1 j∈πi

Moreover, the coefficients of this linear combination are independent of q, αi , ni , X.

Proof. Note that the bi-degree of aλ(i) (τ∗ αi ) is equal to
" #
(5.18) |λ(i) |, 2( (λ(i) ) − 2) + |αi | .
It follows that AN = 0 unless

N
|λ(i) | = 0.
i=1
N
In the rest of the proof, we will assume i=1 |λ(i) | = 0. We will divide the proof
into two cases.

Case 1: |λ(i) | = 0 for every 1 ≤ i ≤ N . Then, (λ(i) ) ≥ 2 for every i. Since the
bi-degree of aλ(i) (τ∗ αi ) is given by (5.18), AN = 0 unless (λ(i) ) = 2 and |αi | = 0
for all 1 ≤ i ≤ N . Assume that (λ(i) ) = 2 and |αi | = 0 for all 1 ≤ i ≤ N . Then
108 5. MULTIPLE q-ZETA VALUES AND HILBERT SCHEMES

for every 1 ≤ i ≤ N , we have λ(i) = ((−ni )ni ) for some ni > 0. We further assume
that n1 = . . . = nr for some 1 ≤ r ≤ N and ni = n1 if r < i ≤ N . Let α1 = a1X
and 
τ2∗ 1X = (−1)|βj | βj ⊗ γj
j

with βj , γj  = δj,j  . Then, AN is equal to

 
N
a (−1)|βj | Tr q n a−n1 (βj ) an1 (γj ) aλ(i) (τ∗ αi )
j i=2

 
N
= aq n1 (−1)|βj | Tr a−n1 (βj ) q n an1 (γj ) aλ(i) (τ∗ αi )
j i=2

 
N
= aq n1 Tr q n an1 (γj ) aλ(i) (τ∗ αi ) · a−n1 (βj ).
j i=2

Moving the operator a−n1 (βj ) to the middle, we see that AN is equal to
 
N
aq n1 Tr q n an1 (γj )a−n1 (βj ) aλ(i) (τ∗ αi )
j i=2


r 
i−1
+ aq n1 Tr q n an1 (γj ) aλ(k) (τ∗ αk )
j i=2 k=2


N
·[aλ(i) (αi ), a−n1 (βj )] · aλ(k) (τ∗ αk ).
k=i+1

By Lemma 3.12 (i), AN is equal to the sum of the expressions

< =
k1
(−n1 )k1 q n1 
(5.19) eX , α1 αji · · Tr q n
aλ(i) (τ∗ αi )
1 − q n1
i=1 i∈{2,...,N }−{j1 ,...,jk1 }

where 0 ≤ k1 ≤ r −1, {j1 , . . . , jk1 } ⊂ {2, . . . , r}, every factor in (−n1 )k1 comes from
a commutator of type [an1 (·), a−n1 (·)], and the coefficients of this linear combination
depend only on λ(1) , . . . , λ(N ) . In particular, we have
(−n1 )q n1
(5.20) A1 = Tr q n aλ(1) (α1 ) = (q; q)−χ(X)
∞ · eX , α1 · .
1 − q n1
Combining with (5.19), we see that our lemma holds in this case.
N
i=1 |λ | = 0 but |λ | = 0 for some i0 . Then N ≥ 2, and we may
(i) (i0 )
Case 2:
assume that |λ | < 0. To simplify the expressions, we further assume that every
(i0 )

αi has an even degree. Note that AN can be rewritten as

a (i ) (τ∗ αi )  aλ(i) (τ∗ αi )
Tr q n λ 0 (i )! 0
λ 0 1≤i≤N
λ(i)!
i=i0

0 −1
i 
r−1 8 9 
aλ(i) (τ∗ αi ) aλ(r) (τ∗ αr ) aλ(i0 ) (τ∗ αi0 ) aλ(i) (τ∗ αi )
+ Tr q n
· , · .
r=1 i=1
λ(i)! λ(r)! λ(i0 )! r+1≤i≤N
λ(i)!
i=i0
 α ,...,αN
5.2. THE SERIES Fk 1,...,k (q) 109
1 N

Since
q n aλ(i0 ) (τ∗ αi0 ) = q −|λ |
(i0 )
aλ(i0 ) (τ∗ αi0 )q n ,
we see that AN is equal to
aλ(i0 ) (τ∗ αi0 ) n  aλ(i) (τ∗ αi )
q −|λ |
(i0 )
Tr q
λ(i0 )! 1≤i≤N
λ(i)!
i=i0

0 −1
i 
r−1 8 9 
aλ(i) (τ∗ αi ) aλ(r) (τ∗ αr ) aλ(i0 ) (τ∗ αi0 ) aλ(i) (τ∗ αi )
+ Tr q n · , · .
r=1 i=1
λ(i)! λ(r)! λ(i0 )! r+1≤i≤N
λ(i)!
i=i0

Since
aλ(i0 ) (τ∗ αi0 ) n  aλ(i) (τ∗ αi )  aλ(i) (τ∗ αi ) a (i0 ) (τ∗ αi )
Tr q = Tr q n
· λ (i )! 0 ,
λ(i0 )! 1≤i≤N
λ(i)! 1≤i≤N
λ(i)! λ 0
i=i0 i=i0

AN is equal to
 a (i) (τ∗ αi ) a (i0 ) (τ∗ αi )
q −|λ |
(i0 )
Tr q n λ
(i)!
· λ (i )! 0
1≤i≤N
λ λ 0
i=i0

0 −1
i 
r−1 8 9 
aλ(i) (τ∗ αi ) aλ(r) (τ∗ αr ) aλ(i0 ) (τ∗ αi0 ) aλ(i) (τ∗ αi )
+ Tr q n
· , · .
r=1 i=1
λ(i)! λ(r)! λ(i0 )! r+1≤i≤N
λ(i)!
i=i0

Note that
 a (i) (τ∗ αi ) a (i0 ) (τ∗ αi )
Tr q n λ
(i)!
· λ (i )! 0
1≤i≤N
λ λ 0
i=i0

is equal to
AN


N  8 9  N
aλ(i) (τ∗ αi ) aλ(r) (τ∗ αr ) aλ(i0 ) (τ∗ αi0 ) aλ(i) (τ∗ αi )
+ Tr q n
· , · .
r=i0 +1 1≤i≤r−1
λ(i)! λ(r)! λ(i0 )! i=r+1
λ(i)!
i=i0

Therefore, we conclude that (1 − q −|λ |

(i0 )
)AN is equal to


N  aλ(i) (τ∗ αi )
q −|λ |
(i0 )
Tr q n
r=i0 +1 1≤i≤r−1,i=i0
λ(i)!
8 9  N
aλ(r) (τ∗ αr ) aλ(i0 ) (τ∗ αi0 ) aλ(i) (τ∗ αi )
· , ·
λ(r)! λ(i0 )! i=r+1
λ(i)!
0 −1
i  a (i) (τ∗ αi ) 8 a (r) (τ∗ αr ) a (i0 ) (τ∗ αi ) 9
r−1  aλ(i) (τ∗ αi )
+ Tr q n λ
(i)!
· λ (r)! , λ (i )! 0 · .
r=1 i=1
λ λ λ 0 r+1≤i≤N
λ(i)!
i=i0
110 5. MULTIPLE q-ZETA VALUES AND HILBERT SCHEMES

Put n0 = −|λ(i0 ) | > 0. It follows that AN is equal to

q n0 
N  aλ(i) (τ∗ αi )
Tr q n
1 − q n0 r=i0 +1
λ(i)!
1≤i≤r−1,i=i0
8 9  N
aλ(r) (τ∗ αr ) aλ(i0 ) (τ∗ αi0 ) aλ(i) (τ∗ αi )
· , ·
λ(r)! λ(i0 )! i=r+1
λ(i)!

1 0 −1
i  aλ(i) (τ∗ αi ) 8 aλ(r) (τ∗ αr ) a (i0 ) (τ∗ αi ) 9
r−1
(5.21) + Tr q n
· , λ (i )! 0
1 − q n0 r=1 i=1
λ (i)! λ (r)! λ 0
 aλ(i) (τ∗ αi )
· .
r+1≤i≤N,i=i
λ(i)!
0

By Lemma 3.12 (i) and (ii), our lemma holds in this case as well. 

The following theorem provides the structure of the trace


N
a λ(i) (τ∗ αi )
Tr q n W(L1 , z) .
i=1
λ(i)!

Theorem 5.8. For 1 ≤ i ≤ N , let

(i) (i) (i) (i)
λ(i) = · · · (−n)m̃n · · · (−1)m̃1 1m1 · · · nmn · · · )
and αi ∈ H ∗ (X) be homogeneous. Then, the trace

N
a λ(i) (τ∗ αi )
(5.22) Tr q n W(L1 , z)
i=1
λ(i)!

is equal to
N 
N
  (i) 
|λ(i) |
z i=1 · (q; q)−χ(X)
∞ · (1X − KX ) n≥1 mn , αi
i=1
 
 (−1)mn
(i) (i)
q nmn 1 1
· ,
+W
(i) (i) (i)
1≤i≤N,n≥1 mn ! (1 − q n )mn m̃(i)
n ! (1 − q ) n
n m̃

and the lower weight term W  is a linear combination of expressions of the form:
< =
N 
u
 
z i=1 |λ | · (q; q)−χ(X)
(i) r
(5.23) ∞ · Sign(π) · ri i
KX eX , αj
i=1 j∈πi


v
q ni wi pi
·
i=1
(1 − q ni )wi
v N 
i=1 (λ ), the integers u, v, ri , ri ≥ 0, ni > 0, wi > 0, pi ∈
(i)
where i=1 wi <
{0, 1} and the partition π = {π1 , . . . , πu } of {1, . . . , N } depend only on the general-
ized partitions λ(1) , . . . , λ(N ) , and Sign(π) is from (5.17). Moreover, the coefficients
of this linear combination are independent of q, α1 , . . . , αN and X.
 α ,...,αN
5.2. THE SERIES Fk 1,...,k (q) 111
1 N

Proof. For simplicity, denote the trace (5.22) by Trλ . Combining Lemma 5.4
and Lemma 5.5, we conclude that Trλ is equal to
  (−(zq s )n )mn(i,s)  (z −1 q t−1 )nm̃n(i,t)
(i,s)
· (i,t)
N  
(|λ(i) |− s≥1 |μ(i,s) |− t≥1 |μ̃(i,t) |)=0 1≤i≤N mn ! 1≤i≤N m̃n !
i=1 s,n≥1 t,n≥1
μ (i,s) 
∈P+ , μ̃ (i,t) 
∈P −

 (i,s)
N a (i)  
 λ − s≥1 μ(i,s) − t≥1 μ̃(i,t) τ∗ ((1X − KX )
s,n≥1 mn αi )
·Tr q n   !
i=1 λ(i) − s≥1 μ(i,s) − t≥1 μ̃(i,t)
(i,s) (i,s) (i,t) (i,t)
where μ(i,s) = 1m1 · · · nmn · · · and μ̃(i,t) = · · · (−n)m̃n · · · (−1)m̃1 . The
N
sum of all the exponents of z is i=1 |λ(i) |. So Trλ is equal to
N 
z i=1 |λ | ·
(i)

N  
(|λ(i) |− s≥1 |μ(i,s) |− t≥1 |μ̃(i,t) |)=0
i=1
 , μ̃(i,t) ∈P
μ(i,s) ∈P 
+ −

 (−q sn )mn(i,s)  q (t−1)nm̃n(i,t)

(i,s)
· (i,t)
·
1≤i≤N mn ! 1≤i≤N m̃n !
s,n≥1 t,n≥1

 (i,s)
N a (i)  
 λ − s≥1 μ(i,s) − t≥1 μ̃(i,t) τ∗ ((1X − KX )
s,n≥1 mn αi )
(5.24) ·Tr q n
  !
.
i=1 λ(i) − s≥1 μ(i,s) − t≥1 μ̃(i,t)
By our convention,
 
μ(i,s) + μ̃(i,t) ≤ λ(i)
s≥1 t≥1

for every 1 ≤ i ≤ N . In the rest of the proof, we group the possible terms in the
above summand into two cases: Case A and Case B.
 
Case A: s≥1 μ
(i,s)
+ t≥1 μ̃(i,t) = λ(i) for every 1 ≤ i ≤ N . Then line (5.24) is
equal to

N
  (i,s) 
Tr q n · (1X − KX ) s,n≥1 mn
, αi
i=1

N
  (i,s) 
= (q; q)−χ(X)
∞ (1X − KX ) s,n≥1 mn , αi .
i=1

Therefore, the contribution C1 of this case to Trλ is equal to

N 
N
  
|λ(i) | m(i)
z i=1 · (q; q)−χ(X)
∞ (1X − KX ) n≥1 n , αi ·
i=1

  (−q sn )mn(i,s)   q (t−1)nm̃n(i,t)

· (i,s)
· (i,t)
.
 (i,s) (i) 1≤i≤N mn !  (i,t) (i) 1≤i≤N m̃n !
s≥1 mn =mn
s,n≥1 t≥1 m̃n =m̃n
t,n≥1
1≤i≤N, n≥1 1≤i≤N, n≥1
112 5. MULTIPLE q-ZETA VALUES AND HILBERT SCHEMES

Rewrite q sn as q (s−1)n q n . Then the contribution C1 is equal to

N 
N
   
|λ(i) | m(i)
· (q; q)−χ(X)
(i)
z i=1
∞ (1X − KX ) n≥1 n , αi · (−q n )mn ·
i=1 1≤i≤N,n≥1

  q (i,s)
(s−1)nmn   q (t−1)nm̃n(i,t)
· (i,s)
· (i,t)
.
 (i,s) (i) 1≤i≤N mn !  (i,t) (i) 1≤i≤N m̃n !
s≥1 mn =mn
s,n≥1 t≥1 m̃n =m̃n
t,n≥1
1≤i≤N, n≥1 1≤i≤N, n≥1

Since
  (q (s−1)n )is,n  1 1

= ,
 is,n ! in ! (1 − q n )in
s≥1 is,n =in , n≥1
s,n≥1 n≥1

C1 is equal to
N 
N
  
|λ(i) | m(i)
(5.25) z i=1 · (q; q)−χ(X)
∞ (1X − KX ) n≥1 n , αi ·
i=1
   
 (−1)mn
(i)
q nmn
(i)
 1 1
· (i) (i)
· (i) (i)
.
1≤i≤N,n≥1 mn ! (1 − q n )mn 1≤i≤N,n≥1 m̃n ! (1 − q n )m̃n

 
Case B: s≥1 μ(i,s) + t≥1 μ̃(i,t) < λ(i) for some 1 ≤ i ≤ N . Without loss of
generality, we may assume that
 
μ(i,s) + μ̃(i,t) = λ(i)
s≥1 t≥1

for every 1 ≤ i ≤ N1 where N1 < N , and

 
μ(i,s) + μ̃(i,t) < λ(i)
s≥1 t≥1

for every N1 + 1 ≤ i ≤ N . For N1 + 1 ≤ i ≤ N , put

  (i) (i) (i) (i)
μ(i,s) + μ̃(i,t) = λ̃(i) = · · · (−n)p̃n · · · (−1)p̃1 1p1 · · · npn · · · .
s≥1 t≥1

An argument similar to that in the previous paragraph shows that for the fixed
generalized partitions λ̃(i) with N1 + 1 ≤ i ≤ N , the contribution C2 of this case to
Trλ is equal to
N 
N1
  (i) 
|λ(i) |
(5.26) z i=1 · (1X − KX ) n≥1 mn , αi
i=1
 
 (−1) m(i)
n q nmn
(i)
1 1
· (i) (i) (i) (i)
1≤i≤N1 mn ! (1 − q n )mn m̃n ! (1 − q n )m̃n
n≥1
 
 (i)
(−1)pn q npn
(i)
1 1
· (i) (i) (i) (i)
N1 +1≤i≤N pn ! (1 − q n )pn p̃n ! (1 − q n )p̃n
n≥1


N
aλ(i) −λ̃(i) τ∗ ((1X − KX ) n≥1 p(i)
n αi )
·Tr q n
!
.
i=N1 +1 λ(i) − λ̃(i)
 α ,...,αN
5.2. THE SERIES Fk 1,...,k (q) 113
1 N

By Lemma 5.7, C2 is a linear combination of expressions of the form:

N 
N1
  (i) 
|λ(i) |
z i=1 · (1X − KX ) n≥1 mn , αi
i=1
 
 (−1) m(i)
n q nmn
(i)
1 1
· (i) (i) (i) (i)
1≤i≤N1 mn ! (1 − q n )mn m̃n ! (1 − q n )m̃n
n≥1

 
 (−1)pn
(i) (i)
q npn 1 1
· (i) (i) (i)
·
N1 +1≤i≤N pn ! (1 − q n )pn p̃(i)
n ! (1 − q ) n
n p̃
n≥1

< =

u   (j) 
v
q ni
(q; q)−χ(X)
∞ · Sign(π) · em
X ,
i
(1X − KX ) n≥1 pn αj ·
i=1 j∈πi i=1
1 − q ni
N
where v < i=N1 +1 (λ
(i)
− λ̃(i) ), ni > 0, mi ≥ 0, {π1 , . . . , πu } is a parti-
tion of {N1 + 1, . . . , N }, and Sign(π) compensates the formal difference between
u
i=1 j∈πi αj and αN1 +1 · · · αN . The coefficients of this linear combination are
independent of q, α1 , . . . , αN and X, and depend only on the partitions λ(i) − λ̃(i) .
Note that for nonnegative integers a and b, the pairing
eaX , (1X − KX )b β = eaX (1X − KX )b , β
is a linear combination of eaX KX
c
, β , 0 ≤ c ≤ b. In addition, we have
  
N
m(i) (i)
n + m̃n + p(i) (i)
n + p̃n +v < (λ(i) )
1≤i≤N1 ,n≥1 N1 +1≤i≤N,n≥1 i=1

regarding the weights in C2 . It follows that C2 is a linear combination of the

expressions (5.23). Combining with (5.25) completes the proof of our theorem. 

Remark 5.9. When N = 1, we can work out the lower weight term W  in
Theorem 5.8 by examining its proof more carefully and by using (5.20). To state
 For n1 ≥ 1 with
the result, let λ = (· · · (−n)m̃n · · · (−1)m̃1 1m1 · · · nmn · · · ) ∈ P.
mn1 · m̃n1 ≥ 1, define mn1 (n1 ) = mn1 − 1, m̃n1 (n1 ) = m̃n1 − 1, and mn (n1 ) = mn
and m̃n (n1 ) = m̃n if n = n1 . Then,
aλ (τ∗ α)
Tr q n W(L1 , z)
λ!
is equal to the sum

(5.27) z |λ| · (q; q)−χ(X)
∞ · (1X − KX ) n≥1 mn , α ·

 (−1) n m

q nmn 1 1
·
mn ! (1 − q n )mn m̃n ! (1 − q n )m̃n
n≥1
 n1 q n1
+ z |λ| · (q; q)−χ(X)
∞ · eX , α · ·
1 − q n1
n1 ≥1 with mn1 ·m̃n1 ≥1
  (−1)mn (n1 ) q nmn (n1 )
1 1

·
mn (n1 )! (1 − q n )mn (n1 ) m̃n (n1 )! (1 − q n )m̃n (n1 )
n≥1
114 5. MULTIPLE q-ZETA VALUES AND HILBERT SCHEMES

where the first two lines come from Case A in the proof of Theorem 5.8, and the
last two lines come from Case B in the proof of Theorem 5.8.
Example 5.10. Let λ = ((−5)(−4)5). Then, we see from (5.27) with n1 = 5
that Tr q n W(L1 , z) a((−5)(−4)5) (τ3∗ α) is equal to
q4 −q 5 q5
z −4 · (q; q)−χ(X)
∞ · 1X − KX , α · ·
1 − q 1 − q 1 − q5
4 5

5q 5 1
+ z −4 · (q; q)−χ(X)
∞ · eX , α · ·
1 − q 1 − q4
5

where the first line is the leading term and the second line is the lower weight term.
The next lemma is used to organize the leading term in Theorem 5.8.
Lemma 5.11. For α ∈ H ∗ (X) and k ≥ 0, define Θαk (q) to be
   
(5.28) − (1X − KX ) n≥1 in , α
(λ)=k+2,|λ|=0
  (−1)in q nin 1 1

·
in ! (1 − q n )in ĩn ! (1 − q n )ĩn
n≥1

where λ = (· · · (−n)ĩn · · · (−1)ĩ1 1i1 · · · nin · · · ). Then,

Θα α
k (q) = Coeff z 0 Θk (q, z)
which denotes the coefficient of z 0 in Θα
k (q, z) defined by

  a a
(−1)si  1
b
(5.29) − (1X − KX ) i=1 si , α ·
a,s1 ,...,sa ,b,t1 ,...,tb ≥1 i=1
si ! t !
j=1 j
a 
s + b t =k+2
i=1 i j=1 j

 a
(qz)ni si  
b
z −mj tj
· · .
n1 >···>na >0 i=1
(1 − q ni )si m (1 − q mj )tj
1 >···>mb >0 j=1

Proof. Put   
A = (1X − KX ) n≥1 in , α

which implicitly depends on n≥1 in . Rewrite |λ| and (λ) in terms of the integers
in and ĩn . Then, Θαk (q) is equal to
   (−1)in q nin 1 1

(5.30) − A .
  in ! (1 − q n )in ĩn ! (1 − q n )ĩn
in +
n≥1 ĩn =k+2
n≥1 n≥1
 
n≥1 nin = n≥1 nĩn >0

Denote the positive integers in the ordered list {i1 , . . . , in , . . .} by sa , . . . , s1 respec-

tively (e.g., if the ordered list {i1 , . . . , in , . . .} is {2, 0, 5, 4, 0, . . .}, then a = 3 with
s3 = 2, s2 = 5, s1 = 4). We have a ≥ 1. Similarly, denote the positive integers
in
 the ordereda list {ĩ1 , . . . , ĩn , . . .} by tb , . . . , t1 respectively. Then b ≥ 1. Since
i
n≥1 n = i=1 si , we get
 a 
A = (1X − KX ) i=1 si , α .
Rewriting (5.30) in terms of sa , . . . , s1 and tb , . . . , t1 , we see that
Θα α
k (q) = Coeff z 0 Θk (q, z).
This completes the proof of the lemma. 
 α ,...,αN
5.2. THE SERIES Fk 1,...,k (q) 115
1 N

We remark that the multiple q-zeta value Θα

k (q, z) has weight (k + 2).

Theorem 5.12. For 1 ≤ i ≤ N , let ki ≥ 0 and αi ∈ H ∗ (X) be homogeneous.

Then,
N 

(5.31) Fk1 ,...,kN (q) = (q; q)−χ(X)
α1 ,...,αN
∞ · Coeff z10 ···zN
0
α
Θkii (q, zi ) + W1 ,
i=1

and the lower weight term W1 is an infinite linear combination of the expressions:
< = v
u
   q ni wi pi
−χ(X) ri ri
(5.32) (q; q)∞ · Sign(π) · KX eX , αj ·
i=1 j∈π i=1
(1 − q ni )wi
i
v N
where i=1 wi < i=1 (ki + 2), and the integers u, v, ri , ri ≥ 0, ni > 0, wi > 0, pi ∈
{0, 1} and the partition π = {π1 , . . . , πu } of {1, . . . , N } depend only on the integers
ki . Moreover, the coefficients of this linear combination are independent of q, αi , X.
Proof. By Lemma 5.2,

N
Fkα11,...,k
,...,αN
N
(q) = Tr q n W(L1 , z) Gki (αi ).
i=1

Combining with Theorem 4.7 and Theorem 5.8, we conclude that

(5.33) Fkα11,...,k
,...,αN
N
(q) = Fkα11,...,k
,...,αN
N
(q) + W1,1
where W1,1 is an infinite linear combination of the expressions (5.32), and the term
F α1 ,...,αN (q) is defined to be
k1 ,...,kN

 
N
a λ(i) (τ∗ αi )
(5.34) (−1)N · Tr q n W(L1 , z) .
i=1
λ(i)!
(λ(i) )=ki +2,|λ(i) |=0
1≤i≤N

Applying Theorem 5.8 again, we see that Fkα11,...,k

,...,αN
N
(q) is equal to
 
N
  (i) 
(5.35) (−1)N (q; q)−χ(X)
∞ · (1X − KX ) n≥1 mn , αi ·
(λ(i) )=ki +2,|λ(i) |=0 i=1
1≤i≤N
 
 (−1)mn
(i) (i)
q nmn 1 1
· (i) (i) (i) (i)
+ W1,2
1≤i≤N,n≥1 mn ! (1 − q n )mn m̃n ! (1 − q n )m̃n
where the lower weight term W1,2 is an infinite linear combination of the expressions
(i) (i) (i) (i)
(5.32), and we have put λ(i) = · · · (−n)m̃n · · · (−1)m̃1 1m1 · · · nmn · · · ). So
(5.36) Fkα11,...,k
,...,αN
N
(q)

N
= (q; q)−χ(X)
∞ · Θα
ki (q) + W1,2
i

i=1
N 

= (q; q)−χ(X)
∞ · Coeff z10 ···zN
0 Θα
ki (q, zi )
i
+ W1,2
i=1

by Lemma 5.11. Putting W1 = W1,1 + W1,2 completes the proof of (5.31). 

116 5. MULTIPLE q-ZETA VALUES AND HILBERT SCHEMES

In order to relate the lower weight term W1,2 in (5.35) and (5.36) to multiple q-
zeta values (with additional variables z1 , . . . , zN inserted), we will assume eX αi = 0
for all 1 ≤ i ≤ N . The following lemma strengthens Lemma 5.7, and follows from
the same arguments.
 and α1 , . . . , αN ∈ H ∗ (X) be homoge-
Lemma 5.13. Let λ(1) , . . . , λ(N ) ∈ P,

neous. Assume that eX αi = 0 for every 1 ≤ i ≤ N , and N i=1 |λ | = 0. Put
(i)


N
aλ(i) (τ∗ αi )
AN = Tr q n .
i=1
λ(i)!

(i) If (λ(i) ) ≥ 2 for every 1 ≤ i ≤ N , then AN = 0.

(ii) If AN = 0, then AN is a linear combination of the expressions:
< = ˜
u   (−ñi )q ñi p̃i
−χ(X)
(q; q)∞ · Sign(π) · 1X , αj ·
i=1 j∈πi i=1
1 − q ñi
< = 
u   (−n )wi q ni pi
−χ(X)
= (q; q)∞ · Sign(π) · 1X , αj · i


i=1 j∈π i=1

(1 − q ni )wi
i

where

N 

˜= (λ(i) )/2 = wi ,
i=1 i=1
p̃i ∈ {0, 1}, 0 ≤ pi ≤ wi , the partition π = {π1 , . . . , πu } of {1, . . . , N } de-
pend only on λ(1) , . . . , λ(N ) , the integers ñ1 , . . . , ñ˜ are the positive parts
(repeated with multiplicities) in λ(1) , . . . , λ(N ) , the integers n1 , . . . , n de-
note the different integers in ñ1 , . . . , ñ˜, and each ni appears wi times in
ñ1 , . . . , ñ˜.
Theorem 5.14. For 1 ≤ i ≤ N , let ki ≥ 0 and αi ∈ H ∗ (X) be homogeneous.
Assume that eX αi = 0 for every 1 ≤ i ≤ N . Then,
N 

(5.37) F 1
α ,...,αN
(q) = (q; q)−χ(X)
k1 ,...,kN ∞ · Coeff z0 ···z0
1
α
Θ i (q, zi ) + W1,2 ,
N ki
i=1
χ(X)
and (q; q)∞ · W1,2 is a linear combination of the coefficients of z10 · · · zN 0
in some
N
multiple q-zeta values (with variables z1 , . . . , zN inserted) of weights < i=1 (ki +2).
Moreover, the coefficients in this linear combination are independent of q.
We now sketch the proof of Theorem 5.14, and refer to the proof of [QY,
Theorem 4.10] for further details. Recall that Fkα11,...,k
,...,αN
N
(q) is defined in (5.34),
and that (5.37) is just (5.36). From the proofs of (5.36) and Theorem 5.8, we
see that the lower weight term W1,2 in (5.37) is the contribution of Case B in the
proof of Theorem 5.8 to the right-hand-side of (5.34). A careful analysis of these
contributions together with Lemma 5.13 completes the proof.
We will end this section with three propositions about Fkα (q), which provide
some insight into the lower weight term W1 in Theorem 5.12. Proposition 5.15
deals with F0α (q) for an arbitrary α ∈ H ∗ (X). Proposition 5.17 calculates F1α (q)
by assuming eX α = 0. Proposition 5.18 computes Fkα (q), k ≥ 2 by assuming
eX α = KX α = 0.
 α ,...,αN
5.2. THE SERIES Fk 1,...,k (q) 117
1 N

Proposition 5.15. The generating series F0α (q) is equal to

 qn
(5.38) (q; q)−χ(X)
∞ · 1X − KX , α ·
n>0
(1 − q n )2
 nq n
+ (q; q)−χ(X)
∞ · eX , α · .
n>0
1 − qn

Proof. By Lemma 5.2,

Fkα (q) = Tr q n W(L1 , z) Gk (α).
By Theorem 4.7, we have

G0 (α) = − (a−n an )(τ2∗ α).
n>0

Now (5.38) follows from Remark 5.9. 

Remark 5.16. By (5.38),
 nq n d
F01X (q) = (q; q)−χ(X)
∞ · χ(X) · = q (q; q)−χ(X)
∞ .
n>0
1 − q n dq

Proposition 5.17. Let α ∈ H ∗ (X) be a homogeneous class satisfying eX α = 0.

Then, the generating series F1α (q) is the coefficient of z 0 in

KX − K X 2
,α  (n − 1)q n
−χ(X)
(q; q)∞ · ·
2 n>0
(1 − q n )2
 
 (qz)n  z −2m  z −m1 z −m2
+ · +2 .
n>0
1 − qn m>0
(1 − q m )2 m >m >0
1 − q m1 1 − q m2
1 2

Proof. We have
F1α (q) = Tr q n W(L1 , z) G1 (α).
By (4.18),
 aλ (τ3∗ α)  n − 1
G1 (α) = − − (a−n an )(τ2∗ (KX α)).
λ! n>0
2
(λ)=3,|λ|=0

Applying Remark 5.9 to

 n−1
− Tr q n W(L1 , z) (a−n an )(τ2∗ (KX α))
n>0
2
yields the weight-2 terms in our proposition. Again by Remark 5.9, the trace
aλ (τ3∗ α)
Tr q n W(L1 , z)
λ!
with (λ) = 3 and |λ| = 0 contains only weight-3 terms (i.e., does not contain lower
weight terms). So the proof of Theorem 5.12 shows that
 aλ (τ3∗ α)
− Tr q n W(L1 , z) = (q; q)−χ(X)
∞ · Coeff z0 Θα
1 (q, z).
λ!
(λ)=3,|λ|=0

Expanding Coeff z0 Θα
1 (q, z) yields the weight-3 terms in our proposition. 
118 5. MULTIPLE q-ZETA VALUES AND HILBERT SCHEMES

Proposition 5.18. Let α ∈ H ∗ (X) be homogeneous satisfying KX α = eX α =

0.
(i) If |α| < 4, then Fkα (q) = 0 for every k ≥ 0;
(ii) Let |α| = 4 and k ≥ 0. Then, Fkα (q) is the coefficient of z 0 in
 a
(−1)si  1
b
−(q; q)−χ(X)
∞ · 1X , α · ·
a,s1 ,...,sa ,b,t1 ,...,tb ≥1 i=1
si ! t !
j=1 j
a 
s + b t =k+2
i=1 i j=1 j

 a
(qz)ni si  
b
z −mj tj
(5.39) · · .
n1 >···>na >0 i=1
(1 − q ni )si m (1 − q mj )tj
1 >···>mb >0 j=1

In particular, if 2 |k, then Fkα (q) = 0.

Proof. Since KX α = eX α = 0, we conclude from Theorem 4.7 that
 aλ (τ∗ α)
Gk (α) = − .
λ!
(λ)=k+2,|λ|=0

As in the proof of Proposition 5.17, Remark 5.9 and the proof of Theorem 5.12
yield
Fkα (q) = (q; q)−χ(X)
∞ · Coeff z0 Θα
k (q, z).

By the definition of Θα k (q, z) in (5.29), we see that (i) holds and that our formula
for Fkα (q) with |α| = 4 and k ≥ 0 holds. Note that line (5.39) can be rewritten as
 a
(qz 2 )ni si /2  b
(qz −2 )mj tj /2
· .
n1 >···>na >0 i=1
(1 − q ni )si m (1 − q mj )tj
1 >···>mb >0 j=1

Therefore, if |α| = 4 and 2  k, then the roles of a, s1 , . . . , sa and b, t1 , . . . , tb in the

above formula of Fkα (q) are anti-symmetric; so Fkα (q) = 0. 
 
LN 
k1 · · · chkN
5.3. The reduced series chL1

In this section, using the calculations from the previous section, we will prove
Conjecture 5.1 modulo the lower weight term. Moreover, for abelian surfaces, we
will verify Conjecture 5.1. The main idea is to express the Chern character of the
tautological bundle L[n] in terms of the classes Gi (α, n).
Let L be a line bundle on the smooth projective surface X. We see from (4.1)
and the Grothendieck-Riemann-Roch Theorem that
(5.40) ch(L[n] ) = p1∗ (ch(OZn ) · p∗2 ch(L) · p∗2 td(X))
= p1∗ (ch(OZn ) · p∗2 (1X + L + L2 /2) · p∗2 td(X))
= G(1X , n) + G(L, n) + G(L2 /2, n).
Since the cohomology degree of Gi (α, n) is 2i + |α|, we have
(5.41) chk (L[n] ) = Gk (1X , n) + Gk−1 (L, n) + Gk−2 (L2 /2, n).
 
Next, we recall the generating series chL k1 · · · chkN
LN
1
and its reduced version
 L1 
chk1 · · · chL
kN
N
from (5.1) and (5.2) respectively.
  L1 L 
5.3. THE REDUCED SERIES chk · · · chk N 119
1 N

Theorem 5.19. Let L1 , . . . , LN be line bundles over X, and k1 , . . . , kN ≥ 0.

Then,
N 
 L1  
LN 
(5.42) chk1 · · · chkN = Coeff z10 ···zN 0
1X
Θki (q, zi ) + W,
i=1

and the lower weight term W is an infinite linear combination of the expressions:
u @ A  v

ri ri   q ni wi pi
KX eX , L1i,1 · · · LNi,N ·
i=1 i=1
(1 − q ni )wi
v N
where i=1 wi < i=1 (ki + 2), and the integers u, v, ri , ri , i,j ≥ 0, ni > 0, wi >
0, pi ∈ {0, 1} depend only on k1 , . . . , kN . Furthermore, all the coefficients of this
linear combination are independent of q, L1 , . . . , LN and X.
Proof. We conclude from (5.1), (5.2), (5.41) and (5.3) that
 L 
chk1 · · · chL
kN = (q; q)χ(X)
∞ · Fk11X,...,k
,...,1X
N
(q) + (q; q)χ(X)
∞ ·A
where A is the sum of the series Fkα1,...,k
,...,αN
 such that for every 1 ≤ i ≤ N ,
1 N

(αi , ki ) ∈ {(1X , ki ), (Li , ki − 1), (L2i /2, ki − 2)},

N N
and i=1 ki < i=1 ki . Now our result follows from Theorem 5.12. 

Theorem 5.19 proves Conjecture 5.1, modulo the lower weight term W . Note
that the leading term
N 

1X
Coeff z10 ···zN
0 Θki (q, zi )
i=1
 
in chL1
k1 · · · chLN
kN has weight

N
(ki + 2),
i=1

and is a multiple of KX , KX N whose coefficient depends only on k1 , . . . , kN and

is independent of the line bundles L1 , . . . , LN and the surface X.
In general, it is unclear how to organize the lower weight term W in Theo-
rem 5.19 into multiple q-zeta values. On the other hand, we have the following
result which, together with Theorem 5.19, verifies Conjecture 5.1 when X is an
abelian surface.
Theorem 5.20. Let L1 , . . . , LN be line bundles over an abelian surface X, and
k1 , . . . , kN ≥ 0. Then, the lower weight term W in (5.42) is a linear combination of
the coefficients of z10 · · · zN0
in some multiple q-zeta values (with additional variables

z1 , . . . , zN inserted) of weights < N i=1 (ki + 2). Moreover, the coefficients in this
linear combination are independent of q.
Proof. Since eX = KX = 0,
Fk̃α̃1,...,
,...,α̃N
k̃
= Fk̃α̃1,...,
,...,α̃N
k̃
1 N 1 N

by Lemma 5.2, Theorem 4.7 and (5.34). By Theorem 5.14 and the proof of Theo-
rem 5.19, our theorem follows. 
120 5. MULTIPLE q-ZETA VALUES AND HILBERT SCHEMES

We remark that some of the multiple q-zeta values mentioned in Theorem 5.20
are in the generalized sense, i.e., in the following form:
 

(−ni )wi q ni pi fi (z1 , . . . , zN )ni
n1 >···>n >0 i=1
(1 − q ni )wi

where 0 ≤ pi ≤ wi , and each fi (z1 , . . . , zN ) is a monomial of z1±1 , . . . , zN ±1

. As
indicated in [Ok2, Section 2], the factors (−ni )wi in the above expression may be
d
related to the operator q .
dq  
Our next two propositions compute the series chL k completely, and should
offer some insight
 into the lower weight term W in Theorem 5.19. Proposition 5.21
calculates chL by assuming eX = 0, while Proposition 5.22 deals with the series
 L 1
chk , k ≥ 2 by assuming eX = KX = 0 (i.e., by assuming that X is an abelian
surface). Note from (5.41) that when χ(X) = 0, we have
 L   L  1 L2
(5.43) chk = chk = Fk1X (q) + Fk−1 L
(q) + · Fk−2 (q).
2
Proposition 5.21. Let L be a line  bundle over a smooth projective surface X
L  0
with eX = 0. Then, the series ch1 is the coefficient of z in
 qn KX , KX  (n − 1)q n
− KX , L · − ·
n>0
(1 − q )
n 2 2 n>0
(1 − q n )2
 
KX , KX  (qz)n  z −2m  z −m1 z −m2
− · · +2 .
2 n>0
1 − qn m>0
(1 − q m )2 m >m >0
1 − q m1 1 − q m2
1 2

Proof. Our formula follows from (5.43), (5.38) and Proposition 5.17. 
Proposition 5.22. Let L be a line bundle over an abelian surface X. If 2  k,
   
then chL
k = 0. If 2|k, the generating series chL
k is the coefficient of z 0 in

L, L  a
(−1)si  1
b
− · ·
2 ≥1 i=1
a,s1 ,...,sa ,b,t1 ,...,tb
si ! t !
j=1 j
a 
s + b t =k
i=1 i j=1 j

 a
(qz)ni si  
b
z −mj tj
· · .
n1 >···>na >0 i=1
(1 − q ni )si m (1 − q mj )tj
1 >···>mb >0 j=1

Proof. Follows immediately from (5.43) and Proposition 5.18. 

 CHAPTER 6

Lie algebras and incidence Hilbert schemes

Let X be a smooth projective complex surface. It is known [Che2, Theorem on

p.43] that besides the Hilbert scheme X [n] , the incidence Hilbert scheme X [n,n+1]
is the only class of (generalized or nested) Hilbert schemes of points on smooth
varieties of dimension bigger than one which are smooth for all n. We refer to
Section 1.5 for some basic properties of X [n,n+1] . Inspired by the generating series
(1.41) of the Poincaré polynomial of X [n,n+1] and the Heisenberg algebra actions
(in Section 3.2) on the cohomology of the Hilbert schemes X [n] , it is natural to
consider the sum of the cohomology group of X [n,n+1] over all n ≥ 0:

+∞
(6.1) X =
H H ∗ X [n,n+1] .
n=0

 X has richer structures than the Fock space

The space H

+∞
HX = H ∗ X [n] .
n=0

 X is an H ∗ (X)-module which is induced by the natural morphisms

For instance, H
X [n,n+1]
→ X, and explains the factor Pt (X) in (1.41). In addition to the Heisen-
berg algebra action, H X admits the action of a translation operator t̃, which ex-
plains the factor 1/(1 − t2 q) in (1.41). This chapter follows the presentation of
[LQ2]. Furthermore, we refer to [LQ2, Section 6] for potential applications to
Donaldson-Thomas theory, where the infinite dimensional Lie algebra actions on
the cohomology of the moduli space of ideal sheaves on certain 3-folds were con-
structed. We also refer to [LQ4] for the equivariant cohomology of incidence Hilbert
schemes (C2 )[n,n+1] and its relations with infinite dimensional Lie algebras.

6.1. Heisenberg algebra actions for incidence Hilbert schemes

In this section, we will define the Heisenberg operators acting on the space H X.
We will prove that these operators satisfy the Heisenberg commutation relations,
and are H ∗ (X)-linear. Moreover, via the natural morphisms X [n,n+1] → X [n] and
X [n,n+1] → X [n+1] , we will compare these Heisenberg operators with the Heisenberg
operators acting on the Fock space HX .
We begin with the definition of the Heisenberg operators. For m ≥ 0 and
n > 0, let Q  [m,m] = ∅ and define Q [m+n,m] to be the following closed subset of
X [m+n,m+n+1]
×X ×X [m,m+1]
:


Q [m+n,m]
= (ξ, ξ ), x, (η, η  ) |

ξ ⊃ η, ξ  ⊃ η  , Supp(Iη /Iξ ) = {x},

Supp(Iξ /Iξ ) = Supp(Iη /Iη ) .

121
122 6. LIE ALGEBRAS AND INCIDENCE HILBERT SCHEMES

 [m+n,m] is given by the following lemma.

The dimension of the subset Q
 [m+n,m] = 2m + n + 3.
Lemma 6.1. For m ≥ 0 and n > 0, dim Q
[m+n,m] . Let
Proof. Take an element (ξ, ξ  ), x, (η, η  ) in Q
{t} = Supp(Iξ /Iξ ) = Supp(Iη /Iη )
for some t ∈ X. First of all, assume that x = t. Then η can be decomposed as
η = η0 + ηx + ηt
where ηx ∈ Mi (x) for some i ≥ 0, ηt ∈ Mj (t) for some j ≥ 0, η0 ∈ X [m−i−j] , and
x, t ∈ Supp(η0 ). Then, η  , ξ and ξ  can be written as
η = η0 + ηx + ηt ,
ξ = η0 + ξx + ηt ,
ξ = η0 + ξx + ηt ,
with
(ηt , ηt ) ∈ Mj,j+1 (t), (ηx , ξx ) ∈ Mi,i+n (x).
When i = j = 0, we conclude from (1.17) that the number of moduli of these triples
 [m+n,m] is equal to
(ξ, ξ  ), x, (η, η  ) ∈ Q
(6.2) #(moduli of η0 ) + #(moduli of ξx ) + #(moduli of t)
= 2m + (n − 1) + 4
= 2m + n + 3.
In general, when i ≥ 1 or j ≥ 1, we see from (1.17) again that the number of
 [m+n,m] is at most
moduli of these triples (ξ, ξ  ), x, (η, η  ) ∈ Q
(6.3) #(moduli of η0 ) + #(moduli of ηx ) +
#(moduli of ηt ⊂ ηt ) + #(moduli of ξx )
= 2(m − i − j) + max(i − 1, 0) + j + (i + n − 1) + 4
< 2m + n + 3.
Next, let x = t. This time we decompose the 0-cycle η into
η = η0 + ηx
where ηx ∈ Mi (x) for some i ≥ 0, η0 ∈ X [m−i] , and x ∈ Supp(η0 ). Then,
η = η0 + ηx ,
ξ = η0 + ξx ,
ξ = η0 + ξx ,
with
(ηx , ηx ) ∈ Mi,i+1 (x), (ηx , ξx ) ∈ Mi,i+n (x),
(ξx , ξx ) ∈ Mi+n,i+n+1 (x), (ηx , ξx ) ∈ Mi+1,i+n+1 (x).
[m+n,m] is at most
So the number of moduli of these triples (ξ, ξ  ), x, (η, η  ) ∈ Q
(6.4) #(moduli of η0 ) + #(moduli of ηx ⊂ ηx ) + #(moduli of ξx ⊂ ξx )
= 2(m − i) + i + (i + n) + 2
< 2m + n + 3.
 6.1. HEISENBERG ALGEBRA ACTIONS FOR INCIDENCE HILBERT SCHEMES 123

[m+n,m] = 2m + n + 3.
Combining (6.2), (6.3) and (6.4), we get dim Q 

 X from (6.1). Now we introduce the operators on H

Recall the notation H X

induced by the correspondence of Q [m+n,m]
.

Definition 6.2. (i) Let α ∈ H ∗ (X). Define ã0 (α) = 0 in End H X .

∗
(ii) Let n > 0 and α ∈ H (X). Define ã−n (α) ∈ End HX by 
" #
(6.5) ã−n (α)(A)  [m+n,m] ] · ρ̃∗ α · p̃∗2 A
 = p̃1∗ [Q 

 ∈ H ∗ (X [m,m+1] ), where p̃1 , ρ̃, p̃2 are the projections of

for A
X [m+n,m+n+1] × X × X [m,m+1]
to X [m+n,m+n+1] , X, X [m,m+1] respectively.
(iii) Define ãn (α) ∈ End H X to be (−1)n times the operator obtained from
the definition of ã−n (α) by switching the roles of p̃1 and p̃2 .

A non-degenerate super-symmetric bilinear form ·, · on H  X is induced from

the standard one on H ∗ (X [n,n+1] ) defined by
,
 B
A,  = A B,
  B
A,  ∈ H ∗ (X [n,n+1] ).
X [n,n+1]

X
This allows us to define the adjoint f̃† ∈ End H  X . As in
for f̃ ∈ End H
Lemma 3.7, we have
(6.6) ãn (α) = (−1)n · ã−n (α)† .
Let n = 0. By (6.5) and Lemma 6.1, ã−n (α) has bi-degree (n, 2n − 2 + |α|), i.e.,
ã−n (α) : H r X [m,m+1] → H r+2n−2+|α| X [m+n,m+n+1] .
 X is an H ∗ (X)-module. The module structure is induced
Next, notice that H
by the natural morphisms ρ̃n : X [n,n+1] → X.
X → H
Lemma 6.3. The map ã−n (α) : H  X is an H ∗ (X)-module homomor-
phism.
Proof. We will only prove the lemma for n > 0 since the proof for n < 0 is
 ∈ H ∗ X [m,m+1] . We need to show that
similar. Let n > 0, β ∈ H ∗ (X), and A

(6.7)  = (−1)|α||β| ρ̃∗m+n β · ã−n (α)(A).

ã−n (α) ρ̃∗m β · A 

By definition,

ã−n (α) ρ̃∗m β · A  [m+n,m] ] · ρ̃∗ α · p̃∗2 (ρ̃∗m β · A)
= p̃1∗ [Q 
= (−1)|α||β| p̃1∗ [Q 
[m+n,m] ] · p̃∗2 ρ̃∗m β · ρ̃∗ α · p̃∗2 A

= (−1)|α||β| p̃1∗ τ∗ ((ρ̃m ◦ p̃2 ◦ τ )∗ β) · ρ̃∗ α · p̃∗2 A
 [m+n,m] → X [m+n,m+n+1] × X × X [m,m+1] is the inclusion map.
where τ : Q
 [m+n,m] ,
From the proof of Lemma 6.1, we see that for (ξ, ξ  ), x, (η, η  ) ∈ Q
Supp(Iη /Iξ ) = Supp(Iη /Iξ ) = {x}.
124 6. LIE ALGEBRAS AND INCIDENCE HILBERT SCHEMES

Therefore, ρ̃m ◦ p̃2 ◦ τ = ρm+n ◦ p̃1 ◦ τ . Hence, we obtain


ã−n (α)(ρ̃∗m β · A) = (−1)|α||β| p̃1∗ τ∗ ((ρ̃m+n ◦ p̃1 ◦ τ )∗ β) · ρ̃∗ α · p̃∗2 A 
 [m+n,m] ] · p̃∗1 ρ̃∗m+n β · ρ̃∗ α · p̃∗2 A
= (−1)|α||β| p̃1∗ [Q 
= (−1) |α||β| ∗
ρ̃ 
β · p̃1∗ [Q [m+n,m] ∗ ∗
] · ·ρ̃ α · p̃ A 
m+n 2

by the projection formula. It follows immediately that (6.7) holds. 

There are two natural morphisms from X [m,m+1] to X [m] and X [m+1] respec-
tively:
gm+1
X [m,m+1] −→ X [m+1]
↓ fm+1
X [m] .
Since there are Heisenberg operators a−n (α) on HX , it is natural to ask whether
the following diagram is commutative or not:
a−n (α)
H ∗ X [m] −−−−−→ H ∗ X [m+n]
⏐ ⏐
⏐f ∗ ⏐f ∗
(6.8) % m+1 % m+n+1
ã−n (α)
H ∗ X [m,m+1] −−−−−→ H ∗ X [m+n,m+n+1] .
Lemma 6.4. The diagram ( 6.8) is commutative.
This lemma not only relates different Heisenberg operators on different spaces,
but also is used in the proof of Proposition 6.5 to determine a constant.
Proof. Again, we will only prove the lemma for n > 0.
Let n > 0 and A ∈ H ∗ X [m] . By Definition 3.3 and (6.5), we obtain
∗ ∗
(6.9) fm+n+1 a−n (α)(A) = fm+n+1 p1∗ ([Q[m+n,m] ] · ρ∗ α · p∗2 A),
(6.10) ∗
ã−n (α)fm+1  [m+n,m] ] · ρ̃∗ α · p̃∗2 fm+1
(A) = p̃1∗ ([Q ∗
(A)).
Let g = IdX [m+n,m+n+1] ×X × fm+1 and h = fm+n+1 × IdX×X [m] :
g: X [m+n,m+n+1] × X × X [m,m+1] → X [m+n,m+n+1] × X × X [m] ,
h: X [m+n,m+n+1] × X × X [m] → X [m+n] × X × X [m] ,
and let
p̃1 : X [m+n,m+n+1] × X × X [m] → X [m+n,m+n+1]
be the first projection. By (6.9) and base change, we conclude that
∗
(6.11) fm+n+1 a−n (α)(A)
= (p̃1 )∗ h∗ ([Q[m+n,m] ] · ρ∗ α · p∗2 A)
= (p̃1 )∗ h∗ [Q[m+n,m] ] · (ρ ◦ h)∗ α · (p2 ◦ h)∗ A .
Similarly, using (6.10), p̃1 = p̃1 ◦ g, and the projection formula, we obtain
∗
(6.12) ã−n (α)fm+1 (A)
" #

= (p̃1 )∗ g∗ [Q [m+n,m]
] · g ∗ (ρ ◦ h)∗ α · (p2 ◦ h)∗ A
[m+n,m] ] · (ρ ◦ h)∗ α · (p2 ◦ h)∗ A .
= (p̃1 )∗ g∗ [Q
 6.1. HEISENBERG ALGEBRA ACTIONS FOR INCIDENCE HILBERT SCHEMES 125

 [m+n,m] ]. Define the

Next, we compare the two cycles h∗ [Q[m+n,m] ] and g∗ [Q
subset
W = (ξ, ξ  ), x, η | ξ ⊃ η and Supp(Iη /Iξ ) = {x}
 [m+n,m] ) are
of X [m+n,m+n+1] × X × X [m] . Note that both h−1 (Q[m+n,m] ) and g(Q
equal to W . A typical element (ξ, x, η) ∈ Q[m+n,m]
is of the form:
(x1 + . . . + xm + ξx , x, x1 + . . . + xm )
where x1 , . . . , xm , x ∈ X are distinct, and ξx ∈ Mn (x). Hence,
(6.13) h∗ [Q[m+n,m] ] = [W ].
 [m+n,m] is of the form:
On the other hand, a typical element (ξ, ξ  ), x, (η, η  ) ∈ Q
(ξ0 + ξx , ξ0 + xm+1 + ξx ) , x, (ξ0 , ξ0 + xm+1 )
where x1 , . . . , xm+1 , x ∈ X are distinct, ξ0 = x1 +. . .+xm , and ξx ∈ Mn (x). Hence,
(6.14) [m+n,m] ] = [W ].
g∗ [Q
The combinations of (6.11), (6.12), (6.13) and (6.14) prove the lemma. 
We remark that in the next section, another comparison of the Heisenberg
∗
operators a−n (α) and ã−n (α) will be made by using the pullbacks gm+1 .
In the rest of this section, we will determine the commutation relations among
the operators
ãn (α), n ∈ Z, α ∈ H ∗ (X).
The following proposition is the analogue of (3.9). Its proof is a bit long, and will
be divided into many cases. The main idea in the proof is to find the geometric
representation (with the correct dimension) of the composition of two Heisenberg
operators ã−n (α) and ã−k (β), and then compare the geometric representation of
ã−n (α)ã−k (β) with that of ã−k (β)ã−n (α).
Proposition 6.5. The operators ãn (α), n ∈ Z, α ∈ H ∗ (X) satisfy the following
Heisenberg algebra commutation relation:
,
(6.15) [ãn (α), ãk (β)] = −n δn,−k (αβ) · IdH X .
X

Proof. In view of (6.6), formula (6.15) is equivalent to the formulas:

(6.16) [ã−n (α), ã−k (β)] = 0,
,
(6.17) [ã−n (α), ãk (β)] = n δn,k (αβ) · IdH
X
X
where n, k > 0. We will prove these two formulas separately.
Proof of ( 6.16):
Let n, k > 0. Then the operator ã−n (α)ã−k (β) is induced by the class
"  [m+n+k,m+k]  ∗  [m+k,m] #
(6.18) ∗
w = π1245∗ π123 
Q · π345 Q

in the Chow group A2m+k+n+4 (W  ), and W  = π1245 (W ) with

−1  [m+n+k,m+k] −1  [m+k,m]
(6.19) W = π123 Q ∩ π345 Q .
Note that W is a closed subset of the ambient space
X [m+n+k,m+n+k+1] × X × X [m+k,m+k+1] × X × X [m,m+1] .
126 6. LIE ALGEBRAS AND INCIDENCE HILBERT SCHEMES

The map πi1 ,...,ik in (6.18) denotes the projection of this ambient space to the
product of its i1 -th, . . ., ik -th factors.
In the sequel, the ambient spaces in similar situations will not be explicitly
presented since they can be written out easily from the context.
From Lemma 6.1, we know that the expected dimension of the intersection W
should be 2m + n + k + 4. The subset W may have many irreducible components.
Those that are mapped by π1245 to subsets of dimension less than 2m + n + k + 4
will not contribute to the cycle w. The aim of the computations below is to pick
out those components with dimension no less than the expected dimension.
Any element ((ξ, ξ  ), x, (η, η  ), t, (ζ, ζ  )) in W must satisfy the conditions:
ζ ⊂ ζ  , η ⊂ η  , ξ ⊂ ξ  , ζ ⊂ η ⊂ ξ, ζ  ⊂ η  ⊂ ξ  ,
Supp(Iη /Iξ ) = {x}, Supp(Iζ /Iη ) = {t},
Supp(Iζ /Iζ  ) = Supp(Iη /Iη ) = Supp(Iξ /Iξ ) = {p}
for some p ∈ X. In the following, we consider four different cases separately. We
use Wi to denote the subset of W consisting of all the points satisfying Case i, and
put
wi = dim π1245 (Wi )
(similar notations such as Ui , ui , Vi , vi will be used throughout the rest of this
chapter).
Case 1: x, t, p are distinct. We have the following decompositions:
ζ = ζ0 + ζx + ζt + ζp , ζ  = ζ0 + ζx + ζt + ζp ,
η = ζ0 + ζx + ηt + ζp , η  = ζ0 + ζx + ηt + ζp ,
ξ = ζ0 + ξx + ηt + ζp , ξ  = ζ0 + ξx + ηt + ζp ,
where Supp(ζ0 ) ∩ {x, t, p} = ∅, (ζx ) = i, (ζt ) = j, (ζp ) = , and ζx , ζt , . . . are
supported at {x}, {t}, . . . respectively. Then,
w1 = #(moduli of ζ0 ) + #(moduli of ζx ) + #(moduli of ζt ) +
+ #(moduli of ηt ) + #(moduli of ξx ) + #(moduli of ζp ⊂ ζp )
= 2(m − i − j − ) + max(i − 1, 0) + 2 + max(j − 1, 0) + 2 +
+ (j + k − 1) + (i + n − 1) + + 2
= (2m + k + n + 4) − i + max(i − 1, 0) − j + max(j − 1, 0) − .
Case 1.1: If i = j = = 0, then we have w1 = (2m + k + n + 4).
Case 1.2: If one of the integers i, j, is positive, then w1 < (2m + k + n + 4).
The three remaining cases are listed by:
Case 2: x = t = p;
Case 3: x = p = t (by symmetry, this also covers the case t = p = x);
Case 4: x = p = t.
For these three cases, we skip the arguments which are in the same style as in
Case 1. They all have dimension less than the expected dimension (2m + n + k +
4). Therefore, only Case 1.1 has contribution to the cohomological operation. In
this subcase, it is not difficult to show that the intersection (6.19) along W1 is
transversal. Moreover, π1245 (W1 ) consists of all the points of the form:
(ζ0 + ξx + ηt , ζ0 + ξx + ηt + p), x, t, (ζ0 , ζ0 + p)
 6.1. HEISENBERG ALGEBRA ACTIONS FOR INCIDENCE HILBERT SCHEMES 127

in X [m+n+k,m+n+k+1] × X × X × X [m,m+1] . So the contribution of this subcase to

the operator ã−n (α)ã−k (β) coincides with the corresponding contribution for the
operator (−1)|α||β| · ã−k (β)ã−n (α). In other words, we obtain the identity:

ã−n (α)ã−k (β) = (−1)|α||β| · ã−k (β)ã−n (α).

This completes the proof of the commutation relation (6.16).
Proof of ( 6.17):
Let n, k > 0. Let
τm,m+k : X [m+k,m+k+1] × X × X [m,m+1] → X [m,m+1] × X × X [m+k,m+k+1]

be the isomorphism switching X [m,m+1] and X [m+k,m+k+1] . Then the operator

ã−n (α)ãk (β) is induced by the class
"  [m+n,m]  ∗  #
(6.20) ∗
u = π1245∗ π123 
Q  [m+k,m] )
· π345 τm,m+k (Q

in the Chow group A2m+k+n+4 (U  ), and U  = π1245 (U ) with

−1  [m+n,m] −1  [m+k,m] ) .
(6.21) U = π123 Q ∩ π345 τm,m+k (Q
Any element ((ξ, ξ  ), x, (η, η  ), t, (ζ, ζ  )) in U must satisfy the conditions
ζ ⊂ ζ  , η ⊂ η  , ξ ⊂ ξ  , ξ ⊃ η ⊂ ζ, ξ  ⊃ η  ⊂ ζ  ,
Supp(Iη /Iξ ) = {x}, Supp(Iη /Iζ ) = {t},
Supp(Iζ /Iζ  ) = Supp(Iη /Iη ) = Supp(Iξ /Iξ ) = {p}
for some p ∈ X. In the following, we consider four different cases separately.
Case 1: x, t, p are distinct. We have the following decompositions:
η = η0 + ηx + ηt + ηp , η  = η0 + ηx + ηt + ηp ,
ξ = η0 + ξx + ηt + ηp , ξ  = η0 + ξx + ηt + ηp ,
ζ = η0 + ηx + ζt + ηp , ζ  = η0 + ηx + ζt + ηp
where Supp(η0 ) ∩ {x, t, p} = ∅, (ηx ) = i, (ηt ) = j, and (ηp ) = . Then,
u1 = #(moduli of η0 ) + #(moduli of ξx ) + #(moduli of ζt ) +
+ #(moduli of ηx ) + #(moduli of ηt ) + #(moduli of ηp ⊂ ηp )
= 2(m − i − j − ) + (i + n − 1) + (j + k − 1) +
+ max(i − 1, 0) + max(j − 1, 0) + + 6
= (2m + k + n + 4) − i + max(i − 1, 0) − j + max(j − 1, 0) − .
Case 1.1: If i = j = = 0, then we have u1 = (2m + k + n + 4).
Case 1.2: If one of the integers i, j, is positive, then u1 < (2m + k + n + 4).
The three remaining cases are listed by:
Case 2: x = t = p;
Case 3: x = p = t (by symmetry, this also covers the case t = p = x);
Case 4: x = p = t.
128 6. LIE ALGEBRAS AND INCIDENCE HILBERT SCHEMES

In these three cases, all the dimensions are smaller than the expected dimension
(2m + n + k + 4). So only Case 1.1 contributes to the class u in (6.20).
Next we consider the operator ãk (β)ã−n (α). This is the only case where there
are two components for V below with the expected dimension. One of the com-
ponents will cancel out with the one from ã−n (α)ãk (β), and the other is a non-
transversal intersection which carries a multiplicity . Using Lemma 6.4 which
compares ãn (α) with the Heisenberg operators an (α) on HX , we determine the
multiplicity.
More precisely, the operator ãk (β)ã−n (α) is induced by the class:
(6.22) "   ∗  [m+k+n,m+k] #
∗
v = π1245∗ π123 τm+n,m+k+n (Q  [m+k+n,m+n] ) · π345 
Q
in A2m+k+n+4 (V  ), where V  = π1245 (V ) and V is given by
(6.23) −1
V = π123 τm+n,m+k+n (Q [m+k+n,m+k] .
 [m+k+n,m+n] ) ∩ π −1 Q
345

Any element ((ξ, ξ  ), x, (η, η  ), t, (ζ, ζ  )) in (6.23) must satisfy the conditions:
ζ ⊂ ζ  , η ⊂ η  , ξ ⊂ ξ  , ξ ⊂ η ⊃ ζ, ξ  ⊂ η  ⊃ ζ  ,
Supp(Iξ /Iη ) = {x}, Supp(Iζ /Iη ) = {t},
Supp(Iζ /Iζ  ) = Supp(Iη /Iη ) = Supp(Iξ /Iξ ) = {p}
for some p ∈ X. In the following, we consider four different cases separately.
Case 1 : x, t, p are distinct. We have the following decompositions:
ζ = ζ0 + ζx + ζt + ζp , ζ  = ζ0 + ζx + ζt + ζp ,
η = ζ0 + ζx + ηt + ζp , η  = ζ0 + ζx + ηt + ζp ,
ξ = ζ0 + ξx + ηt + ζp , ξ  = ζ0 + ξx + ηt + ζp ,
where Supp(ζ0 ) ∩ {x, t, p} = ∅, (ξx ) = i, (ζt ) = j, and (ζp ) = . Then,
v1 = #(moduli of ζ0 ) + #(moduli of ζx ) + #(moduli of ζt ) +
+ #(moduli of ηt ) + #(moduli of ξx ) + #(moduli of ζp ⊂ ζp )
= 2(m − i − j − ) + (i + k − 1) + max(j − 1, 0) +
+ (j + n − 1) + max(i − 1, 0) + + 6
= (2m + k + n + 4) − i + max(i − 1, 0) − j + max(j − 1, 0) − .

Case 1 .1: If i = j = = 0, then we have v1 = (2m + k + n + 4).
Case 1 .2: If one of the integers i, j and is positive, then v1 < (2m + k + n + 4).

Case 2 : x = t = p. We have the following decompositions:
ζ = ζ0 + ζx + ζp , ζ  = ζ0 + ζx + ζp ,
η = ζ0 + ηx + ζp , η  = ζ0 + ηx + ζp ,
ξ = ζ0 + ξx + ζp , ξ  = ζ0 + ξx + ζp ,
where Supp(ζ0 ) ∩ {x, p} = ∅, (ξx ) = i, (ζx ) = j, and (ζp ) = . Note that
(6.24) k + i = (ηx ) = n + j.

Case 2 .1: i = j = = 0. By (6.24), we have k = n. Thus,
v2 = 2(m + k) + 4 = 2m + k + n + 4.
 6.2. A TRANSLATION OPERATOR FOR INCIDENCE HILBERT SCHEMES 129

Case 2 .2: If one of the integers i, j and is positive, then v2 < (2m + k + n + 4).
The two remaining cases are listed by:
Case 3 : x = p = t (by symmetry, this also covers the case t = p = x);
Case 4 : x = p = t.
In these cases, the dimensions are smaller than (2m + k + n + 4). Therefore these
two cases have no contributions to the class v in (6.22).
Finally, note that the contribution of Case 1.1 to the class u in (6.20) and the
contribution of Case 1 .1 to the class v in (6.22) cancel out in the commutation:
(6.25) [ã−n (α), ãk (β)].
So [ã−n (α), ãk (β)] = 0 when n = k. This proves (6.17) when n = k.
When n = k, Case 2 .1 also contributes to the operator ãk (β)ã−n (α), and hence
to (6.25). Note from Case 2 .1 that π1245 (V2 ) consists of all the points of the form:
(ζ0 , ζ0 + p), x, x, (ζ0 , ζ0 + p)
in X [m+n,m+n+1] × X × X × X [m+n,m+n+1] , where x = p and p ∈ Supp(ζ0 ). Thus,
,
[ã−n (α), ãn (β)]|H ∗ (X [m+n,m+n+1] ) = c · (αβ) · IdH ∗ (X [m+n,m+n+1] )
X
for some constant c. Now we conclude from Lemma 6.4 and (3.9) that c = n. This
completes the proof of the commutation relation (6.17) when n = k. 

6.2. A translation operator for incidence Hilbert schemes

In this section, we will introduce a new operator t̃ on H  X , called the transla-
tion operator. The operator t̃ is constructed via a correspondence and has many
nice properties. Indeed, we will show that it is an H ∗ (X)-module homomorphism,
commutes with the Heisenberg operators, and has a left inverse. These properties
imply that it is responsible for the factor 1/(1 − t2 q) in (1.41).
From Lemma 6.4, we may tend to infer that the Heisenberg operators ãn (α) are
no different from the pullback of the Heisenberg operators an (α) of Nakajima and
Grojnowski. This is almost the case when we compare these two types of operators
via the map
fm+1 : X [m,m+1] → X [m] .
However, we will see that when we compare them via the map
gm+1 : X [m,m+1] → X [m+1] ,
they are far from being the same. The difference is measured by the new translation
operator. Once we are presented with the fact that the space H  X is a highest weight
module of the algebra generated by the operators ãn (α), the cohomology H ∗ (X)
and the translation operator t̃, we will realize that the naive choice of the pullback
of Heisenberg algebras on Hilbert schemes either by the map fm+1 or gm+1 won’t
provide the right algebra. In this regard, the new algebra we are going to construct
is subtler and richer than the Heisenberg algebra on the Hilbert scheme X [m] . The
fundamental difference between these two algebras is the translation operator t̃.
To define the translation operator, let Q  m be the closed subset:
m =
Q (ξ  , ξ  ), (ξ, ξ  ) | Supp(Iξ /Iξ ) = Supp(Iξ /Iξ )
⊂ X [m+1,m+2] × X [m,m+1] .
130 6. LIE ALGEBRAS AND INCIDENCE HILBERT SCHEMES

As in Lemma 6.1, we can verify that

 m = 2m + 3.
dim Q
X
Definition 6.6. The translation operator is the linear operator t̃ ∈ End H
defined by
" #
(6.26)  = p̃1∗ [Q
t̃(A) 
m ] · p̃∗2 A

 ∈ H ∗ (X [m,m+1] ), where p̃1 , p̃2 are the two projections of X [m+1,m+2] ×

for A
[m,m+1]
X .
The bi-degrees of t̃ and its adjoint t̃† are (1, 2) and (−1, −2) respectively.
Lemma 6.7. (i) The operator (−t̃† ) is the left inverse of t̃;
X → H
(ii) The maps t̃, t̃† : H  X are H ∗ (X)-module homomorphisms.

Proof. We skip the proof of (ii) since it is similar to the proof of Lemma 6.3.
In the following, we prove (i). Note that the operator t̃† t̃ is induced by the class:
"   ∗ #
(6.27) w = π13∗ π12∗  m ) · π23
τ (Q m ∈ A2m+2 (W  )
Q

where τ : X [m+1,m+2] × X [m,m+1] → X [m,m+1] × X [m+1,m+2] is the isomorphism

switching the two factors, W  = π13 (W ), and W is given by
(6.28) −1
W = π12  m ) ∩ π −1 Q
τ (Q m .
23

The expected dimension of the intersection W is 2m + 2.

Any element (η, ξ  ), (ξ  , ξ  ), (ξ, ξ ) in W must satisfy the conditions:
η, ξ ⊂ ξ  ⊂ ξ  ,
Supp(Iη /Iξ ) = Supp(Iξ /Iξ ) = Supp(Iξ /Iξ ) = {p}
for some p ∈ X. We have the following decompositions:
ξ = ξ0 + ξp , ξ  = ξ0 + ξp ,
ξ  = ξ0 + ξp , η = ξ0 + ηp ,
where p ∈ Supp(ξ0 ) and (ξp ) = i. Note that (ηp ) = i. Fix the integer i.
If i > 0, then the projection of the subset of W , consisting of all the elements
(η, ξ  ), (ξ  , ξ  ), (ξ, ξ  ) ,
to X [m,m+1] × X [m,m+1] has dimension at most equal to:
#(moduli of ξ0 ) + #(moduli of ξp ⊂ ξp ) + #(moduli of ηp )
= 2(m − i) + i + (i − 1) + 2
< (2m + 2).
So the case i > 0 does not contribute to the class w in (6.27).
If i = 0, then ξp = p, ξp ∈ M2 (p), and ξp = ηp = ∅. So η = ξ = ξ0 . The
projection of this part of W to X [m,m+1] × X [m,m+1] consists of elements of the
form:
(ξ0 , ξ0 + p), (ξ0 , ξ0 + p) .
 6.2. A TRANSLATION OPERATOR FOR INCIDENCE HILBERT SCHEMES 131

It follows that w = c[Δ] for some constant c, where Δ denotes the diagonal in
X [m,m+1] × X [m,m+1] . Hence, we get
t̃† t̃ = c IdH ∗ (X [m,m+1] ) .
To determine c, note that we can split off the factor ξ0 from our consideration,
i.e., we can simply consider the case m = 0. Then we have the morphism:
π13 : X × X [1,2] × X → X × X.
Fix a point p ∈ X. Since [Δ] · [{p} × X] = 1, we see from (6.27) that
c = w · [{p} × X]
"   ∗  #
= π13∗ π12 ∗  0 ) · π23
τ (Q  0 · [{p} × X]
Q
   
∗
= π12  0 ) · π23
τ (Q ∗ 
Q0 · [{p} × X [1,2] × X]
= [{p} × Up × X] · [{p} × Q 0 ]
where Up = (p, ξp )| ξp ∈ M2 (p) . It follows that
(6.29) 0 ] = [Up ] · φ1∗ [Q
c = [Up × X] · [Q 0 ] = [Up ] · [φ1 (Q
 0 )]
where φ1 : X [1,2] × X → X [1,2] is the projection. We have
 0 ) = {(x, ξx )| x ∈ X and ξx ∈ M2 (x) .
φ1 ( Q
Recall the natural morphism g2 : X [1,2] → X [2] . By (1.40),
 0 )] = 1 g2∗ [M2 (X)]
[φ1 (Q
2
where M2 (X) = ∪x∈X M2 (x). Hence, we see from (6.29) that
1
c = [Up ] · g2∗ [M2 (X)]
2
1
= g2∗ [Up ] · [M2 (X)]
2
1
= [g2 (Up )] · [M2 (X)]
2
1
= [M2 (p)] · [M2 (X)]
2
= −1
where we have used the fact that [M2 (p)] · [M2 (X)] = −2 from (1.35). 
The translation operator t̃ may look similar to the creation operators at the
first glimpse. However, we now show that it differs from the creation operators in
the essential way in that it commutes with all the annihilation operators.
Proposition 6.8. The translation operator t̃ and its adjoint t̃† commute with
the Heisenberg operators ã−n (α) for all n and α, i.e.,
[t̃, ã−n (α)] = [t̃† , ã−n (α)] = 0.
Proof. Let n > 0. Then Proposition 6.8 is decomposed into two parts:
(6.30) [t̃, ã−n (α)] = 0,
(6.31) [t̃, ãn (α)] = 0.
We prove them separately, and will compare the proof with that of Proposition 6.5.
132 6. LIE ALGEBRAS AND INCIDENCE HILBERT SCHEMES

Proof of ( 6.30):
Let n > 0. This part is similar to the proof of (6.16).
The operator ã−n (α)t̃ is induced by the class:
"  [m+n+1,m+1]  ∗  #
(6.32) w = π124∗ π123∗
Q m ∈ A2m+n+4 (W  )
· π34 Q

where W  = π124 (W ) and the subset W is defined by

−1  [m+n+1,m+1] −1 
(6.33) W = π123 Q ∩ π34 Qm .
The expect dimension of the intersection W is 2m + n + 4.
Any element (η, η  ), x, (ξ  , ξ  ), (ξ, ξ  ) in (6.33) must satisfy the conditions
η ⊂ η , ξ ⊂ ξ  ⊂ ξ  , ξ  ⊂ η, ξ  ⊂ η  ,
Supp(Iξ /Iη ) = {x},
Supp(Iη /Iη ) = Supp(Iξ /Iξ ) = Supp(Iξ /Iξ ) = {p}
for some p ∈ X. In the following, we consider two cases separately.
Case 1: x = p. We have the following decompositions:
ξ = ξ0 + ξx + ξp , ξ  = ξ0 + ξx + ξp , ξ  = ξ0 + ξx + ξp ,
η = ξ0 + ηx + ξp , 
η = ξ0 + ηx + ξp ,
where Supp(ξ0 ) ∩ {x, p} = ∅, (ξx ) = i, and (ξp ) = j.
Case 1.1: If i = j = 0, then we have w1 = (2m + n + 4).
Case 1.2: If i > 0 or j > 0, then we obtain w1 < (2m + n + 4).
Case 2: x = p. We have w2 < (2m + n + 4).
In summary, only Case 1.1 contributes to the class w. In this subcase, the in-
tersection (6.33) along W1 is transversal, and π1245 (W1 ) consists of all the elements:
(ξ0 + ηx + p, ξ0 + ηx + ξp ), x, (ξ0 , ξ0 + p)
in X [m+n+1,m+n+2] × X × X [m,m+1] . So the contribution of this subcase to the
operator ã−n (α) t̃ coincides with the corresponding contribution for the operator
t̃ ã−n (α). This completes the proof of the commutation relation (6.30).
Proof of ( 6.31):
This part looks similar to the proof of (6.17). However, it is fundamentally dif-
ferent. Recall that the subset V in (6.23) whose correspondence defines ãn (α)ã−n (β)
has one more component with the expected dimension than the subset for the op-
erator ã−n (α)ãn (β). That doesn’t occur here. If there were two components, they
would come from the operator ãn (α)t̃. So pay a special attention to Case 2 below
and compare it with Case 2 .1 in the proof of Proposition 6.5. In order to illustrate
this subtle difference, we present the full content of the proof.
Let n > 0. Then the operator ãn (α)t̃ is induced by the class:
"   ∗  #
(6.34) u = π124∗ π123 ∗
τm+1,m+n+1 (Q [m+n+1,m+1] ) · π34  m+n
Q

in A2m+n+4 (U  ), where U  = π124 (U ) and the subset U is defined by

(6.35) −1
U = π123 τm+1,m+n+1 (Q m+n .
 [m+n+1,m+1] ) ∩ π −1 Q
34
 6.2. A TRANSLATION OPERATOR FOR INCIDENCE HILBERT SCHEMES 133

Any element (η, η  ), x, (ξ  , ξ  ), (ξ, ξ  ) in (6.35) must satisfy the conditions

η ⊂ η , ξ ⊂ ξ  ⊂ ξ  , η ⊂ ξ  , η  ⊂ ξ  ,
Supp(Iη /Iξ ) = {x},
Supp(Iη /Iη ) = Supp(Iξ /Iξ ) = Supp(Iξ /Iξ ) = {p}
for some p ∈ X. In the following, we consider two cases separately.
Case 1: x = p. We have the following decompositions:
ξ = ξ0 + ξx + ξp , ξ  = ξ0 + ξx + ξp , ξ  = ξ0 + ξx + ξp ,
η = ξ0 + ηx + ξp , η  = ξ0 + ηx + ξp ,
where Supp(ξ0 ) ∩ {x, p} = ∅, (ηx ) = i, and (ξp ) = j.
Case 1.1: If i = j = 0, then we have u1 = (2m + n + 4).
Case 1.2: If i > 0 or j > 0, then we obtain
u1 ≤ #(moduli of ξ0 ) + #(moduli of ξx ) + #(moduli of ηx ) +
+ #(moduli of ξp ) + #(moduli of ξp ⊂ ξp )
= 2(m − i − j) + (i + n − 1) + max(i − 1, 0) +
+max(j − 1, 0) + (j + 1) + 4
< (2m + n + 4).
Case 2: x = p. We have the following decompositions:
ξ = ξ0 + ξx , ξ  = ξ0 + ξx , ξ  = ξ0 + ξx ,
η = ξ0 + ηx , η  = ξ0 + ηx ,
where x ∈ Supp(ξ0 ) and (ηx ) = i. In this case, we see that
u2 ≤ #(moduli of ξ0 ) + #(moduli of ξx ⊂ ξx ) + #(moduli of ηx ⊂ ηx )
≤ 2(m + 1 − i) + (i + n − 1) + i + 2
< (2m + n + 4).
In summary, only Case 1.1 contributes to the class u. The contribution of this
subcase to the operator ãn (α) t̃ coincides with that of the operator t̃ ãn (α). This
completes the proof of the commutation relation (6.31). 

Next, we compare the Heisenberg operators ãn (α) with the pull-back of the
Heisenberg operators an (α) via the morphism
gm+1 : X [m,m+1] → X [m+1] .
The comparison of annihilation operators is similar to Lemma 6.4. The proof of
the following lemma is omitted since it is similar to the proof of Lemma 6.4.
Lemma 6.9. Let n > 0 and α ∈ H ∗ (X). Then, we have a commutative diagram:
an (α)
H ∗ X [m+1] ←−−−− H ∗ X [m+n+1]
⏐ ⏐
⏐g ∗ ⏐g∗
% m+1 % m+n+1
ãn (α)
H ∗ X [m,m+1] ←−−−− H ∗ X [m+n,m+n+1] .
134 6. LIE ALGEBRAS AND INCIDENCE HILBERT SCHEMES

Lemma 6.9 will not hold for the creation operators. However, Proposition 6.11
below provides an explicit formula relating the creation operators. In order to prove
Proposition 6.11, we begin with a technical lemma.
Let m ≥ 0. Define Q  m,0 to be the diagonal of X [m,m+1] × X [m,m+1] . For n ≥ 1,

define Qm,n to be the closed subset of X [m+n,m+n+1] × X [m,m+1] :

Qm,n = (ξ, ξ  ), (η, η  ) | ξ  ⊃ ξ ⊃ η  ⊃ η,

Supp(Iξ /Iξ ) = Supp(Iη /Iξ ) = Supp(Iη /Iη ) .

m,n has dimension (2m + n + 2), and contains exactly one irreducible
Note that Q
component of dimension (2m + n + 2) whose generic elements are of the form:
(6.36) (η + ξx , η + ξx ), (η, η + x) , x ∈ Supp(η).
Lemma 6.10. The restriction of t̃n to H ∗ (X [m,m+1] ) is given by the cycle
 m,n ].
[Q

Proof. Use induction on n. Note that Q  m,1 = Qm by definition. So the

Lemma is trivially true when n = 0, 1. Next, assume that n ≥ 2 and the restriction
of t̃n−1 to H ∗ (X [m,m+1] ) is given by the cycle [Q m,n−1 ]. Then we see that the
n n−1 ∗ [m,m+1]
restriction of t̃ = t̃ t̃ to H (X ) is given by the cycle
" #
∗  ∗ 
(6.37) w = π13∗ π12 [Qm+n−1 ] · π23 [Qm,n−1 ] ∈ A2m+n+2 (W  )

where W  = π13 (W ) and the subset W is given by

−1  −1 
(6.38) W = π12 (Qm+n−1 ) ∩ π23 (Qm,n−1 ).
Any element (ξ, ξ  ), (θ, θ  ), (η, η  ) in (6.38) must satisfy:
ξ  ⊃ ξ = θ  ⊃ θ ⊃ η  ⊃ η,
Supp(Iξ /Iξ ) = Supp(Iθ /Iθ ) = Supp(Iη /Iθ ) = Supp(Iη /Iη ) = {x}
for some point x ∈ X. We have the following decompositions:
η = η0 + ηx , η  = η0 + ηx , θ = η0 + θx ,

θ = ξ = η0 + ξx , ξ  = η0 + ξx ,
where x ∈ Supp(η0 ) and ηx ⊂ ηx ⊂ θx ⊂ ξx ⊂ ξx . Put = (ηx ). When ≥ 1, the
relation ηx ⊂ ξx imposes a nontrivial condition on the points
(ηx , ηx ), (ξx , ξx ) ∈ M,+1 (x) × M+n,+n+1 (x).
Hence the subset consisting of the images (ξ, ξ  ), (η, η  ) of those points
(ξ, ξ  ), (θ, θ  ), (η, η  )
in (6.38) with = (ηx ) ≥ 1 has dimension at most
2(m − ) + 2 + [ + (n + ) − 1] = (2m + n + 1)
which is less than the expected dimension (2m + n + 2). So the case ≥ 1 does not
contribute to the cycle w in (6.37). When = 0, we have
η = η0 , η  = η0 + x, θ = η0 + θx ,

θ = ξ = η0 + ξx , ξ  = η0 + ξx
 6.2. A TRANSLATION OPERATOR FOR INCIDENCE HILBERT SCHEMES 135

where x ∈ Supp(η0 ) and θx ⊂ ξx ⊂ ξx . The images (η0 + ξx , η0 + ξx ), (η0 , η0 + x)

form a subset of dimension (2m + n + 2) in X [m+n,m+n+1] × X [m,m+1] . From the
m,n , we conclude that
description (6.36) of the generic points in Q
(6.39)  m,n ]
w = c · [Q
where c is the intersection multiplicity of (6.38) along the generic points.
To determine c, choose a primitive integral class A  ∈ H ∗ (X [m,m+1] ), i.e.,
 ∈ H ∗ (X [m,m+1] ; Z)/Tor ⊂ H ∗ (X [m,m+1] ).
A
 = cB
By (6.39), t̃n (A)  for some integral class B.
 By Lemma 6.7 (i), we see that
 = (−t̃† )n t̃n (A)
A  = c(−t̃† )n (B).


Since A  is an integral class, we must have c = 1.

 is primitive and (t̃† )n (B) 

Proposition 6.11. Let n > 0, α ∈ H ∗ (X), and A ∈ H ∗ X [m+1] . Then,

(i) gn∗ a−n (α)|0 = n · t̃n−1 ρ̃∗0 (α);
∗ ∗
(ii) gm+n+1 a−n (α)(A) = ã−n (α)(gm+1 A) + n · t̃n−1 ρ̃∗m+1 (α) · fm+2
∗
(A) .
Proof. (i) It suffices to prove the formula for α = 1X . The cohomology
class a−n (1X )|0 is represented by the subscheme Mn (X). Recall Mn−1,n (X) from
(1.37). By Theorem 1.31 (iii),
1 ∗
g [Mn (X)] = [Mn−1,n (X)].
n n
By Lemma 6.10, t̃n−1 (1X [0,1] ) is also represented by Mn−1,n (X). Therefore,
gn∗ a−n (1X )|0 = n · t̃n−1 ρ̃∗0 (1X ).
(ii) By Definition 3.3 and (6.5), we obtain
∗ ∗
(6.40) gm+n+1 a−n (α)(A) = gm+n+1 p1∗ ([Q[m+n+1,m+1] ] · ρ∗ α · p∗2 A),
(6.41) ∗
ã−n (α)gm+1  [m+n,m] ] · ρ̃∗ α · p̃∗2 gm+1
(A) = p̃1∗ ([Q ∗
(A)).
Let g = IdX [m+n,m+n+1] ×X × gm+1 and h = gm+n+1 × IdX×X [m+1] :
g: X [m+n,m+n+1] × X × X [m,m+1] → X [m+n,m+n+1] × X × X [m+1] ,
h: X [m+n,m+n+1] × X × X [m+1] → X [m+n+1] × X × X [m+1] .
Let
p̃1 : X [m+n,m+n+1] × X × X [m+1] → X [m+n,m+n+1]
be the first projection. By (6.40) and base change, we conclude that
∗
(6.42) gm+n+1 a−n (α)(A)
= (p̃1 )∗ h∗ ([Q[m+n+1,m+1] ] · ρ∗ α · p∗2 A)
= (p̃1 )∗ h∗ [Q[m+n+1,m+1] ] · (ρ ◦ h)∗ α · (p2 ◦ h)∗ A .
Similarly, using (6.41), p̃1∗ = p̃1 ◦ g and the projection formula, we obtain
∗
(6.43) ã−n (α)gm+1 (A)
" #

= (p̃1 )∗ g∗ [Q [m+n,m]
] · g ∗ (ρ ◦ h)∗ α · (p2 ◦ h)∗ A
[m+n,m] ] · (ρ ◦ h)∗ α · (p2 ◦ h)∗ A .
= (p̃1 )∗ g∗ [Q
136 6. LIE ALGEBRAS AND INCIDENCE HILBERT SCHEMES

Next, we compare h∗ [Q[m+n+1,m+1] ] and g∗ [Q [m+n,m] ]. Let U1 be the closure

of the subset of X [m+n,m+n+1]
×X ×X [m+1]
consisting of all the elements of the
form:
 m  
 
m+1 
m+1
xi + ξx , xi + ξx , x, xi
i=1 i=1 i=1

where x1 , . . . , xm+1 , x ∈ X are distinct, and ξx

∈ Mn (x) is curvi-linear. Let U2
be the closure of the subset of X [m+n,m+n+1] × X × X [m+1] consisting of all the
elements:
m+1  
 
m+1 
m+1

(6.44) xi + ξ x , xi + ξx , x, xi
i=1 i=1 i=1

where x1 , . . . , xm+1 , x ∈ X are distinct, ξx

∈ Mn (x) is curvi-linear, and ξx ⊂ ξx .
Note that general elements in Mn (x) are curvi-linear. Moreover, since ξx ∈ Mn (x)
is curvi-linear, ξx uniquely determines ξx with ξx ⊂ ξx .
A typical element (ξ  , x, η  ) ∈ Q[m+n+1,m+1] is:
(x1 + . . . + xm+1 + ξx , x, x1 + . . . + xm+1 )
where x1 , . . . , xm+1 , x ∈ X are distinct, and ξx ∈ Mn (x) is curvi-linear. Hence,
(6.45) h∗ [Q[m+n+1,m+1] ] = [U1 ] + n[U2 ].
 [m+n,m] is of the form:
On the other hand, a typical element (ξ, ξ  ), x, (η, η  ) ∈ Q
 m   
 
m+1 m 
m+1
 
xi + ξ x , xi + ξx , x, xi , xi
i=1 i=1 i=1 i=1

where x1 , . . . , xm+1 , x ∈ X are distinct, and ξx ∈ Mn (x) is curvi-linear. Hence,

(6.46) [m+n,m] ] = [U1 ].
g∗ [Q
Combining (6.42), (6.43), (6.45) and (6.46), we obtain
∗ ∗
(6.47) gm+n+1 a−n (α)(A) − ã−n (α)gm+1 (A)
= n · (p̃1 )∗ [U2 ] · (ρ ◦ h)∗ α · (p2 ◦ h)∗ A .
Let φ = IdX [m+n,m+n+1] × ρ̃m+1 × fm+2 be the morphism:
X [m+n,m+n+1] × X [m+1,m+2] → X [m+n,m+n+1] × X × X [m+1] .
From the descriptions (6.36) and (6.44), we conclude that
 m+1,n−1 ].
[U2 ] = φ∗ [Q
By (6.47), the projection formula and Lemma 6.10, we obtain
∗ ∗
gm+n+1 a−n (α)(A) − ã−n (α)gm+1 (A)
" #

= n · (p̃1 ◦ φ)∗ [Q m+1,n−1 ] · (ρ ◦ h ◦ φ)∗ α · (p2 ◦ h ◦ φ)∗ A
" #
= n · π1∗ [Q m+1,n−1 ] · π2∗ ρ̃∗m+1 (α) · fm+2
∗
(A)
= n · t̃n−1 ρ̃∗m+1 (α) · fm+2
∗
(A)
where π1 and π2 are the two projections of X [m+n,m+n+1] × X [m+1,m+2] . 
 6.3. LIE ALGEBRAS AND INCIDENCE HILBERT SCHEMES 137

6.3. Lie algebras and incidence Hilbert schemes

With all the results obtained in previous sections, we are ready to formulate
and prove the main theorem.
Definition 6.12. (i) Define h̃X to be the Heisenberg algebra generated by the
operators ãn (α), n ∈ Z, α ∈ H ∗ (X) and the identity operator IdH X ;
(ii) Define a Lie algebra structure on
 X = h̃X ⊕ H ∗ (X) ⊕ Ct̃
H
by declaring
(6.48) [ãn (α), β] = 0, [β, γ] = 0, [t̃, ãn (α)] = 0, [t̃, β] = 0
for operators ãn (α) ∈ h̃X and cohomology classes β, γ ∈ H ∗ (X).
Theorem 6.13. The space H  X is a representation of the Lie algebra H
 X with
a highest weight vector being the vacuum vector
|0 = 1X ∈ H 0 (X) = H 0 (X [0,1] )
where 1X denotes the fundamental cohomology class of X.
Proof. By Lemma 6.3, Proposition 6.5, Lemma 6.7 (ii) and Proposition 6.8,
 X is a representation of the Lie algebra H
we see that the space H X.
Next, for each 0 ≤ i ≤ 4, we fix a linear basis
{αi,1 , . . . , αi,bi (X) }
of H (X). By Proposition 6.5 and the existence of the left inverse t̃† of the operator
i

t̃ in Lemma 6.7 (i), the following cohomology classes:

 
(6.49) t̃m · ã−ni,ji (αi,ji )mi,ji · ã−ni,ji (αi,ji )mi,ji · αk,jk |0
i even i odd
 
= t̃ ·
m
ã−ni,ji (αi,ji ) mi,ji
· ã−ni,ji (αi,ji )mi,ji · αk,jk
i even i odd
are linearly independent, where 0 ≤ i, k ≤ 4, 0 ≤ ji ≤ bi (X), 0 ≤ jk ≤ bk (X),
ni,ji > 0, m ≥ 0, mi,ji ≥ 0 for all the i and ji , and mi,ji = 0 or 1 for odd i.
Since the bi-degrees of the operators t̃ and ã−n (α) are (1, 2) and (n, 2n − 2 + |α|)
respectively, we conclude from (1.41) that cohomology classes in (6.49) form a linear
 X . Therefore, the representation is the highest weight representation. 
basis of H
 Part 3

Cohomology rings of
Hilbert schemes of points
 CHAPTER 7

The cohomology rings of Hilbert

schemes of points on surfaces

Let X be a smooth projective complex surface. One way to understand the

intersection theory on the Hilbert scheme X [n] is to construct a meaningful set of
generators for the cohomology ring of X [n] . In this chapter, we will construct two
sets of generators for the cohomology ring of the Hilbert scheme X [n] , which are
related to the Mumford principle. For a moduli space of sheaves, the conjectural
Mumford principle asserts that the Künneth components of the Chern character of
the universal sheaf form a set of generators for the cohomology ring of the moduli
space. Regarding the Hilbert schemes X [n] as the moduli spaces parametrizing the
ideal sheaves, the results in Section 7.1 imply that the Mumford principle holds for
X [n] . Moreover, using the stability theorem (Theorem 7.7), we will define a Hilbert
ring HX associated to the projective surface X. The Hilbert ring HX encodes all
the information about the cohomology ring of X [n] for every n ≥ 1. However,
these descriptions of the cohomology ring of the Hilbert scheme X [n] are implicit
in terms of the multiplication operators or explicit generators with implicitly given
relations. The approaches of Lehn-Sorger [LS2] and Costello-Grojnowski [CoG]
provide explicit descriptions of the cohomology ring of the Hilbert scheme X [n] via
the cohomogy ring of the surface X. Both approaches are based on the observation
[Leh1,LQW1] that the ring structure of H ∗ (X [n] ), n ≥ 1 is uniquely determined by
the theorem of Nakajima and Grojnowski (Theorem 3.8), Lehn’s boundary operator
d (Theorem 3.24 (ii) and (4.19)), and Theorem 4.2. The approach of Lehn and
Sorger works only when the canonical divisor of X is numerically trivial, while the
approach of Costello and Grojnowski uses the Calogero-Sutherland operators and
the Dunkl-Cherednik operators. We will survey both approaches in this chapter.
Theorem 7.5 is from [LQW1, LQW2], while Theorem 7.7, Theorem 7.11 and their
proofs are from [LQW3].

7.1. Two sets of ring generators for the cohomology

In this section, we will construct two sets of generators for the cohomology
ring of the Hilbert scheme X [n] . The first set consists of the cohomology classes
Gi (α, n) where 0 ≤ i < n and α runs over a fixed linear basis of H ∗ (X). The
cohomology classes Gi (α, n) are closely related to the Grothendieck-Riemann-Roch
Theorem [Hart, Theorem 5.3 in Appendix A], and are thus easier to handle. The
second set of generators consists of the cohomology classes Bi (α, n) which are more
geometric (Definition 7.3). Moreover, we will prove a stability theorem regarding
the cup product in the cohomology of X [n] .
We begin with a technical lemma which expresses a Heisenberg monomial class
in terms of the cohomology classes Gi (α, n).
141
142 7. THE COHOMOLOGY RINGS OF HILBERT SCHEMES OF POINTS

Lemma 7.1. Fix n, s ≥1, n1 , . . . , ns ≥ 1, and α1 , . . . , αs ∈ H ∗ (X). Put

s 
s
n0 = ni . Then, 1−(n−n0 ) a−ni (αi ) |0 is a finite linear combination of
i=1 i=1


t
(7.1) Gmj (βj , n)
j=1

whose coefficients are independent of X, α1 , . . . , αs and n. Here


t
(mj + 1) ≤ n0 ,
j=1

and β1 , . . . , βt depend only on eX , KX , α1 , . . . , αs and τi∗ with 1 ≤ i ≤ n0 .

Proof. We use induction on n0 . When n0 = 1, s = n1 = 1. By (4.22),
G0 (α1 , n) = 1−(n−1) a−1 (α1 )|0 .
So the lemma holds for n0 = 1. Next, let n0 > 1. Let ki = ni − 1. Then,

s
ki ≥ 0 for every i. By Lemma 4.13, Gki (αi , n) is a finite linear combination of
i=1
expressions of the form (4.24) such that the coefficients in this linear combination
are independent of X, α1 , . . . , αs and n. Note that
(π) mi −ri
  
(π)

s
ni,j ≤ (kj + 1) = (ki + 1) = n0 .
i=1 j=1 i=1 j∈πi i=1

By induction, those expressions (4.24) with

(π) mi −ri
 
ni,j < n0
i=1 j=1

are linear combinations of the form (7.1) where


t
(mj + 1) ≤ (n0 − 1),
j=1

and β1 , . . . , βt depend only on eX , KX , α1 , . . . , αs , τi∗ with 1 ≤ i ≤ (n0 − 1). More-

over, the coefficients in these linear combinations are independent of X, α1 , . . . , αs
and n. Now our lemma follows from Lemma 4.14. 
Remark 7.2. By Lemma 7.1, an expression (7.1) in
 s 

1−(n−n0 ) a−ni (αi ) |0
i=1


t
satisfies (mj + 1) ≤ n0 . In fact, we see from the proof of Lemma 7.1 that an
j=1

t
expression (7.1) satisfies the upper bound (mj + 1) = n0 if and only if it is equal
j=1

s 
s
to Gni −1 (αi , n), whose coefficient is ((−1)ni −1 ni !) by Lemma 4.14 (ii).
i=1 i=1
 7.1. TWO SETS OF RING GENERATORS FOR THE COHOMOLOGY 143

Definition 7.3. For 0 ≤ i < n and α ∈ H ∗ (X), define

(7.2) Bi (α, n) = 1−(n−i−1) a−(i+1) (α)|0 ∈ H ∗ (X [n] ).
Lemma 7.4. The Chern character Gk (α, n) is a finite linear combination of
products
t
Bmj (βj , n)
j=1


t
whose coefficients are independent of X, α and n. Here mj ≤ k, and β1 , . . . , βt
j=1

t
depend only on eX , KX and α. In addition, mj = k if and only if the product
j=1

t
Bmj (βj , n) is equal to Bk (α, n), whose coefficient is equal to (−1)k /(k + 1)!.
j=1

Proof. Use induction on k. When k = 0, we have G0 (α, n) = B0 (α, n) by

(4.22). Next, we assume that the lemma is true for 0, . . . , k − 1 for some fixed
k ≥ 1. We will prove that the lemma holds for k as well. We apply Lemma 7.1 and
Remark 7.2 to Bk (α, n) = 1−(n−k−1) a−(k+1) (α)|0 . We see that
(−1)k
Gk (α, n) − · Bk (α, n)
(k + 1)!

u
is a finite linear combination of Gnj (γj , n) where
j=1


u
(nj + 1) < (k + 1),
j=1

and γ1 , . . . , γu depend only on eX , KX and α. Moreover, the coefficients in this

linear combination are independent of X, α and n. Note that nj < k for all 1 ≤
j ≤ u. So by induction hypothesis, the lemma holds for k. 

Theorem 7.5. For n ≥ 1, the cohomology ring H ∗ (X [n] ) is generated by the

cohomology classes Gi (α, n) (respectively, the cohomology classes Bi (α, n)) where
0 ≤ i < n and α runs over a fixed linear basis of H ∗ (X).
Proof. By Lemma 7.1, the cohomology ring H ∗ (X [n] ) is generated by the
cohomology classes Gi (α, n) where 0 ≤ i < n and α runs over a fixed linear basis of
H ∗ (X). By Lemma 7.4, H ∗ (X [n] ) is generated by the cohomology classes Bi (α, n)
where 0 ≤ i < n and α runs over a fixed linear basis of H ∗ (X). 

The following corollary asserts that the cohomology ring of the Hilbert scheme
X [n] is uniquely determined by the cohomology ring of the surface X and its canon-
ical class KX .
Corollary 7.6. Let X and Y be smooth projective complex surfaces. Assume
that there exists a ring isomorphism Φ : H ∗ (X) → H ∗ (Y ) with Φ(KX ) = KY . Then
for every n ≥ 1, the cohomology rings H ∗ (X [n] ) and H ∗ (Y [n] ) are isomorphic.
144 7. THE COHOMOLOGY RINGS OF HILBERT SCHEMES OF POINTS

Proof. Define a linear map Ψ : H ∗ (X [n] ) → H ∗ (Y [n] ) by

Ψ a−n1 (α1 ) · · · a−nk (αk )|0 = a−n1 Φ(α1 ) · · · a−nk Φ(αk ) |0 .
Then Ψ is an isomorphism of vector spaces. Note that Φ(eX ) = eY . By Lemma 4.13,
Ψ Gk1 (α1 , n) · · · Gks (αs , n) = Gk1 Φ(α1 ), n · · · Gks Φ(αs ), n .
Since the classes Gk (α, n) generate H ∗ (X [n] ), Ψ is a ring isomorphism. 

Assume that the canonical class KX of the surface X is numerically trivial.

Then by Corollary 7.6, the cohomology ring H ∗ (X [n] ) is uniquely determined by
the cohomology ring H ∗ (X). This observation is the foundation of the approach of
Lehn and Sorger [LS2] to be discussed in Section 7.3 where the cohomology ring
H ∗ (X) will be explicitly constructed from the cohomology ring H ∗ (X).
Next, we study the cup product of Heisenberg monomial classes
⎛ ⎞

ki
1−(n−ki n ) ⎝ a−ni,j (αi,j )⎠ |0
j=1 i,j
j=1

where 1 ≤ i ≤ s. The following result is known as the stability theorem.

Theorem 7.7. Let s ≥ 1 and ki ≥ 1 for 1 ≤ i ≤ s. Fix ni,j ≥ 1 and

ki
αi,j ∈ H ∗ (X) for 1 ≤ j ≤ ki , and fix n with n ≥ ni,j for all 1 ≤ i ≤ s. Then
j=1
the cup product
⎛ ⎛ ⎞ ⎞

s 
ki
(7.3) ⎝1 ki ⎝ a−ni,j (αi,j )⎠ |0 ⎠
−(n− j=1 ni,j )
i=1 j=1

in H ∗ (X [n] ) is equal to a finite linear combination of monomials of the form

N 

(7.4) 1−(n−N p=1 mp )
a−mp (γp ) |0
p=1

whose coefficients are independent of X, αi,j and n. Here


N 
s 
ki
mp ≤ ni,j ,
p=1 i=1 j=1

and γ1 , . . . , γN depend only on eX , KX , αi,j , 1 ≤ i ≤ s, 1 ≤ j ≤ ki . In addition, the

expression ( 7.4) satisfies the upper bound

N 
s 
ki
mp = ni,j
p=1 i=1 j=1

if and only if it is equal to

⎛ ⎞

s 
ki
1−(n−s  ki ⎝ a−ni,j (αi,j )⎠ |0 ,
i=1 j=1 ni,j )
i=1 j=1

whose coefficient is 1.
 7.1. TWO SETS OF RING GENERATORS FOR THE COHOMOLOGY 145

 ki
Proof. Put Ni = j=1 ni,j for 1 ≤ i ≤ s. For each i, we see from Lemma 7.1
that ⎛ ⎞

ki
1−(n−ki ⎝ a−ni,j (αi,j )⎠ |0
j=1 ni,j )
j=1


ti

ti
is a finite linear combination of products Gmi,j (βi,j , n) where (mi,j + 1) ≤
j=1 j=1
Ni , and βi,1 , . . . , βi,ti depend only on eX , KX , αi,1 , . . . , αi,ki and τj∗ with 1 ≤
j ≤ Ni . Moreover, the coefficients in the linear combinations are independent of
X, αi,1 , . . . , αi,ki and n. By Remark 7.2, the product satisfies the upper bound
ti ki
(mi,j + 1) = Ni if and only if it is equal to Gni,j −1 (αi,j , n). Furthermore,
j=1 j=1

ki
the coefficient of Gni,j −1 (αi,j , n) in
j=1
⎛ ⎞

ki
1−(n−ki ⎝ a−ni,j (αi,j )⎠ |0
j=1 ni,j )
j=1


ki
is equal to ((−1)ni,j −1 ni,j !).
j=1
So (7.3) is a universal finite linear combination of products

s 
ti
Gmi,j (βi,j , n)
i=1 j=1

where

s 
ti 
s 
s 
ki
(mi,j + 1) ≤ Ni = ni,j .
i=1 j=1 i=1 i=1 j=1

Also,

s 
ti 
s 
ki
(mi,j + 1) = ni,j
i=1 j=1 i=1 j=1

if and only if the product


s 
ti
Gmi,j (βi,j , n)
i=1 j=1

is equal to

s 
ki
Gni,j −1 (αi,j , n)
i=1 j=1

whose coefficient in (7.3) is equal to


s 
ki
((−1)ni,j −1 ni,j !).
i=1 j=1
146 7. THE COHOMOLOGY RINGS OF HILBERT SCHEMES OF POINTS

It follows from Lemma 4.13 that (7.3) is a universal linear combination of

expressions (7.4). The statement for the expression (7.4) reaching the upper bound


N 
s 
ki
mp = ni,j
p=1 i=1 j=1

follows from Lemma 4.14. 

Theorem 7.7 indicates that the cup product on the Hilbert scheme X [n] is
independent of n in an appropriate sense. Furthermore, it provides an explicit
form of the leading term in the cup product. This result enables us to construct
the Hilbert ring in the next section.

7.2. The Hilbert ring

In this section, using the stability theorem (Theorem 7.7), we will introduce and
determine the Hilbert ring HX associated to a smooth projective complex surface X.
On one hand, the Hilbert ring HX encodes all the information about the cohomology
ring of the Hilbert scheme X [n] for every n ≥ 1, and thus offers an approach to
describing the cohomology ring of X [n] . On the other hand, the ring structure of
HX itself is implicit since the structure constants of the product are implicit. This
makes HX less appealing.
Given a finite set S which is a disjoint union of subsets S0 and S1 , we denote
by P(S) the set of partition-valued functions ρ = (ρ(c))c∈S on S such that for every
c ∈ S1 , the partition ρ(c) is required to be strict in the sense that

ρ(c) = (1m1 (c) 2m2 (c) . . .)

with mr (c) = 0 or 1 for all r ≥ 1.

Now let us take a linear basis

S = S0 ∪ S1

of H ∗ (X) such that 1X ∈ S0 , S0 ⊂ H even (X), and S1 ⊂ H odd (X). If we write

ρ = (ρ(c))c∈S and

ρ(c) = (r mr (c) )r≥1 = (1m1 (c) 2m2 (c) . . .),

then we put
 
(ρ) = (ρ(c)) = mr (c),
c∈S c∈S,r≥1
 
ρ = |ρ(c)| = r · mr (c),
c∈S c∈S,r≥1
Pn (S) = {ρ ∈ P(S) | ρ = n}.

Given
ρ = (ρ(c))c∈S = (r mr (c) )c∈S,r≥1 ∈ P(S)
 7.2. THE HILBERT RING 147

and n ≥ 0, we define

a−ρ(c) (c) = a−r (c)mr (c) = a−1 (c)m1 (c) a−2 (c)m2 (c) · · ·
r≥1

aρ (n) = 1−(n−ρ) a−ρ(c) (c) · |0 ∈ H ∗ (X [n] )
c∈S

where we fix the order of the elements c ∈ S1 appearing in once and for all.
c∈S
Note from Definition 3.18 that aρ (n) = 0 for 0 ≤ n < ρ .
As ρ runs over all partition-valued functions on S with ρ ≤ n, the correspond-
ing aρ (n) linearly span H ∗ (X [n] ) as a corollary to Theorem 3.8. By Theorem 7.7
(for s = 2), we have the cup product

(7.5) aρ (n) · aσ (n) = aνρσ aν (n)
ν
∗
in H (X ), where ν ≤ ρ + σ and the structure coefficients aνρσ are indepen-
[n]

dent of n. Even though the cohomology classes aν (n) with ν ≤ n in H ∗ (X [n] )

are not linearly independent in general, we have the following.
Lemma 7.8. The structure constants aνρσ in the formula ( 7.5) are uniquely
determined by the requirement that they are independent of n.
Proof. Assume that there exist a finite subset I ⊂ P(S) and some constants
cν ∈ Q independent of n such that for all n ≥ 0, we have

(7.6) cν aν (n) = 0.
ν∈I
By Theorem 3.8, the Heisenberg monomial classes

a−ρ(c) (c) · |0 ,
c∈S

where ρ = (ρ(c))c ∈ Pn (S), are linearly independent in the cohomology H ∗ (X [n] ).

Therefore, by the definition of aν (n), we may assume in (7.6) that any two distinct
ν and ν̃ in I satisfy
ν(1X ) = (1m1 2m2 · · · ),
ν̃(1X ) = (1m1 +ν̃|−ν 2m2 · · · ),
and ν̃(c) = ν(c) for c = 1X (here we assume for definiteness that n ≥ ν̃ > ν ).
In this case, we have
(n − ν̃ )!
aν (n) = · aν̃ (n).
(n − ν )!
Letting n → +∞, we see from (7.6) that cν̃ = 0 for the ν̃ ∈ I with the largest size
ν̃ . So all the constants cν are zero. 
Now we are ready to introduce the Hilbert ring.
Definition 7.9. The Hilbert ring associated to a smooth projective surface X,
denoted by HX , is defined to be the ring with a linear basis formed by the symbols
aρ , ρ ∈ P(S) and with the multiplication defined by

(7.7) aρ · aσ = aνρσ aν
ν
where the structure constants aνρσ are from the relations (7.5).
148 7. THE COHOMOLOGY RINGS OF HILBERT SCHEMES OF POINTS

Note that the Hilbert ring does not depend on the choice of a linear basis S of
H ∗ (X) containing 1X since the operator an (α) depends on the cohomology class
α ∈ H ∗ (X) linearly. It follows from the super-commutativity and associativity
of the cohomology ring H ∗ (X [n] ) that the Hilbert ring HX itself is also super-
commutative and associative.

Theorem 7.10. (Stability) Let X be a smooth projective complex surface.

(i) The cohomology rings H ∗ (X [n] ), n ≥ 1, give rise to a Hilbert ring HX
which is super-commutative and associative.
(ii) Conversely, the Hilbert ring HX completely determines the cohomology
ring structure of H ∗ (X [n] ) for each n ≥ 1.

Proof. (i) Follows from Definition 7.9 and the observations in the preceding
paragraph.
(ii) Recall that as ρ runs over all partition-valued functions on S with ρ ≤ n,
the corresponding classes aρ (n) linearly span H ∗ (X [n] ). Given the Hilbert ring HX ,
the cup products (7.5) of these cohomology classes aρ (n) ∈ H ∗ (X [n] ) can be read
from the multiplications (7.7). 

We also have the following result on the structure of the Hilbert ring HX . For
convenience, in the case when (ρ) = 1, that is, when the partition ρ(c) is a one-part
partition (r) for some element c ∈ S and is empty for all the other elements in S,
we will simply write aρ = ar,c and aρ (n) = ar,c (n).

Theorem 7.11. The Hilbert ring HX is isomorphic to the tensor product P ⊗E,
where P is the polynomial algebra generated by

ar,c , c ∈ S0 , r ≥ 1,

and E is the exterior algebra generated by

ar,c , c ∈ S1 , r ≥ 1.

Proof. Note that

ar,c (n) = 1−(n−r) a−r (c)|0 = Br−1 (c, n).

By Lemma 7.1 and Lemma 7.4, the ring HX is generated by the elements ar,c ,
where c ∈ S = S0 ∪ S1 and r ≥ 1. By the super-commutativity of HX , we have
a2r,c = 0 for c ∈ S1 and r ≥ 1. It remains to show that as ρ = (r mr (c) )s∈S,r≥1 runs
over P(S), the monomials

am r (c)
r,c
c∈S,r≥1

are linearly independent in HX . Assume

  mi (c)
ai ar,cr =0
i∈I c∈S,r≥1
 7.3. APPROACH OF LEHN-SORGER VIA GRADED FROBENIUS ALGEBRAS 149

i
where ai ∈ C and ρi = (r mr (c) )c∈S,r≥1 runs over a finite set I of distinct elements
in P(S). Since 1−(n−r) a−r (c)|0 = ar,c (n), we obtain
  mir (c)
(7.8) ai · 1−(n−r) a−r (c)|0
i∈I c∈S,r≥1
  i
= ai · ar,c (n)mr (c)
i∈I c∈S,r≥1
= 0
by the definition of the structure constants in HX .
Take an integer n large enough such that
def

n ≥ ni = rmir (c)
r,c

for all i ∈ I. From Theorem 7.7, we see that (7.8) can be rewritten as
⎛ ⎞
 
ai ⎝1−(n−ni ) (a−r (c))mr (c) · |0 + wi ⎠ = 0
i
(7.9)
i∈I c∈S,r≥1

where each wi is a finite linear combination of


N
1−(n− N
p=1 mp ) a−mp (γp ) · |0
p=1
N
with p=1 mp < ni and γp ∈ S for every p. Recall that 1−k = 1/k! · a−1 (1X )k ,
k ≥ 0. If we multiply (7.9) by n!, and locate in the resulting summation those
terms whose coefficients contain the largest power of n, then we see that
  i
(7.10) ai · 1−(n−ni ) (a−r (c))mr (c) · |0 = 0
i c∈S,r≥1

where i satisfies ni = max{nj |j ∈ I}. Since all the integers ni in (7.10) are equal,
the Heisenberg monomials in (7.10) are linearly independent as a corollary to The-
orem 3.8. Thus all the coefficients ai in (7.10) are zero. By repeating the above
argument, we conclude that ai = 0 for all i ∈ I. 
In Section 7.1 and the present section, the cohomology ring structure of the
Hilbert scheme X [n] has been studied and described. However, these descriptions
are very implicit in terms of the multiplication operators or explicit generators
with implicitly given relations. In [LS2, CoG], two new approaches to study the
cohomology ring of the Hilbert scheme X [n] appeared. In the next two sections, we
will survey these two papers respectively.

7.3. Approach of Lehn-Sorger via graded Frobenius algebras

This section is devoted to [LS2]. The main idea in the approach of Lehn and
Sorger is to construct a certain algebraic model from a graded Frobenius algebra.
In the end, this algebraic model will be applied to the cohomology ring H ∗ (X) of a
smooth projective complex surface X, viewed naturally as a graded Frobenius alge-
bra. We remark that the description given in [LS2] works only when the canonical
divisor KX of the surface X is numerically trivial. The numerical vanishing of KX
simplifies the description of the cohomology ring of X [n] .
150 7. THE COHOMOLOGY RINGS OF HILBERT SCHEMES OF POINTS

We start with the definition of graded Frobenius algebras.

Definition 7.12. A graded Frobenius algebra of degree d is a finite dimensional
graded (complex) vector space
d
A= Ai
i=−d
with a graded commutative and associative multiplication A ⊗ A → A of degree d
and unit element 1 (necessarily of degree −d) together with a linear form
T :A→C
of degree −d such that the induced symmetric bilinear form
a, b = T (ab)
is nondegenerate (and of degree 0).
It follows from 1·1 = 1 that d must be an even number. The degree conventions
are chosen in such a way that A is centered around degree 0. The degree of an
element a ∈ A will be denoted by |a|.
The tensor product A⊗n is again a graded Frobenius algebra of degree nd with
product
(a1 ⊗ · · · ⊗ an ) · (b1 ⊗ · · · ⊗ bn ) = (a, b) · (a1 b1 ) ⊗ · · · ⊗ (an bn )
where (a, b) is the sign resulting from reordering the a’s and b’s. The linear form
on A⊗n is given by
T (a1 ⊗ · · · ⊗ an ) = T (a1 ) · · · T (an ).
The n-th symmetric group Sn acts on A⊗n by
π(a1 ⊗ · · · ⊗ an ) = (π, a) · aπ−1 (1) ⊗ · · · ⊗ aπ−1 (n)
where (π, a) is the sign resulting from reordering the a’s.
It will be useful to extend the definition of the n-fold tensor product to allow
arbitrary unordered finite indexing sets. Let I be a finite set with n elements and
let {Ai }i∈I be a family of copies of A. Denote the set {1, . . . , n} by [n]. Define
⎛ ⎞
⎜  ⎟
A⊗I = ⎝ Af (1) ⊗ · · · ⊗ Af (n) ⎠ /Sn
∼
=
f :[n]→I

where the direct sum inherits a Frobenius algebra structure by component-wise

defined operations, and Sn acts on this direct sum via the induced operation on
∼
= ∼
=
the set of bijections [n] → I. So for each bijection f : [n] → I, there is a canonical
isomorphism
∼
A⊗n → A⊗I
=

which will again be denoted by f .

Let n1 , . . . , nk be positive integers, and let n = n1 + . . . + nk . Define the ring
homomorphism
ϕn• ,k : A⊗n → A⊗k
by mapping a1 ⊗ · · · ⊗ an to
(a1 · · · an1 ) ⊗ · · · ⊗ (an1 +···+nk−1 +1 · · · an ).
 7.3. APPROACH OF LEHN-SORGER VIA GRADED FROBENIUS ALGEBRAS 151

To generalize this to get a graded ring homomorphism ϕ∗ : A⊗I → A⊗J for any
surjective map ϕ : I → J of finite sets of cardinality n = |I| and k = |J|, choose a
bijection g : [k] → J and let ni = |ϕ−1 (g(i))|. Then there is a bijection f : [n] → I
such that for each i ∈ [k], we have
ϕ−1 (g(i)) = f (n1 + · · · + ni−1 + 1), . . . , f (n1 + · · · + ni ) .
The composition
f −1 ϕn g
ϕ∗ : A⊗I −→ A⊗n −→
•
A⊗k −→ A⊗J
,k

is well-defined and independent of the choices of f and g. Moreover, if ϕ : I → J

and ψ : J → K are surjections, then
ψ ∗ ◦ ϕ∗ = (ψ ◦ ϕ)∗ .
In a different direction, for a surjective map ϕ : I → J, define
ϕ∗ : A⊗J → A⊗I
to be the linear map adjoint to ϕ∗ with respect to the bilinear forms on A⊗I and
A⊗J . Note that ϕ∗ is a ring homomorphism, but ϕ∗ is not. However, the projection
formula holds:
ϕ∗ (a · ϕ∗ (b)) = ϕ∗ (a) · b
for b ∈ A⊗I and a ∈ A⊗J .
Next, consider the composition map
Δ
∗
A −→ A ⊗ A −→ A
where the first map is the co-product and the second map is the product. The
image of 1 under the composition map is called the Euler class of A and is denoted
by e = e(A). If dim(A−d ) = 1, then by duality we have dim(Ad ) = 1 as well. Hence
there is a unique element vol ∈ Ad such that T (vol) = 1. In this case
e(A) = χ(A) · vol
 i i
where χ(A) = i (−1) dim(A ) is the Euler-Poincaré characteristic of A.
For a function ν : I → {0, 1, 2, . . .}, define
eν = ⊗i∈I eν(i) ∈ A⊗I .
Let  denote the co-Cartesian diagram of finite sets and surjective maps:
α
I⏐ −→ K
⏐
⏐ ⏐
%β %γ
δ
J −→ L.
The associated diagram
α∗
AB⊗I −→ A⊗K
B
⏐ ⏐
⏐β∗ ⏐ γ∗
δ∗
A⊗J −→ A⊗L
of linear maps is not commutative. In fact, by [LS2, Lemma 2.4], we have
(7.11) α∗ β∗ (a) = γ∗ eν · δ ∗ (a)
152 7. THE COHOMOLOGY RINGS OF HILBERT SCHEMES OF POINTS

for a ∈ A⊗J , where the function ν : L → {0, 1, 2, . . .} is defined by

ν ( ) = 1 − |δ −1 ( )| − |γ −1 ( )| + |(δ ◦ β)−1 ( )|.
The degree of a permutation π ∈ Sn of cycle type given by a partition λ =
(λ1 ≥ . . . ≥ λ > 0) is defined to be
(7.12) d(π) = d(λ) = n − .
Since π → sgn(π) = (−1)d(π) is a homomorphism, the degree defect
1
d(π1 ) + . . . + d(πt ) − d(π1 · · · πt )
a(π1 , . . . , πt ) =
2
is a non-negative integer. For a subgroup H ⊂ Sn and an H-stable subset B ⊂ [n],
let H\B denote the orbit space for the induced action. For π, ρ ∈ Sn , the graph
defect
g(π, ρ) : π, ρ \[n] → Q
is defined by
1
(7.13) g(π, ρ)(B) = |B| + 2 − | π \B| − | ρ \B| − | πρ \B| .
2
By [LS2, Lemma 2.7], the graph defect g takes value in the non-negative integers.
As in the case of the degree defect, the graph defect allows a natural multivariant
extension: for π1 , . . . , πt ∈ Sn and B ∈ π1 , . . . , πt \[n], define
⎛ ⎞
1⎝ 
g(π1 , . . . , πt )(B) = |B| + 2 − | πj \B| − | π1 · · · πt \B|⎠ .
2 j

Define the graded vector space A{Sn } by putting


(7.14) A{Sn } = A⊗(π\[n]) · π
π∈Sn

where the grading of an element a · π is |a · π| = |a|. The symmetric group Sn acts

on A{Sn } as follows: the action of σ ∈ Sn on [n] induces a bijection
σ : π \[n] → σπσ −1 \[n], x → σx
for each π and hence an isomorphism
σ̃ : A{Sn } → A{Sn }, a · π → σ ∗ (a) · σπσ −1 .
Define
A[n] = (A{Sn })Sn
to be the subspace of the fixed elements.
Next, we will define a ring structure on A{Sn } such that A[n] becomes a com-
mutative subring. Note that any inclusion H ⊂ K of subgroups of Sn leads to a
surjection H\[n]  K\[n] of orbit spaces and hence to maps
f H,K : A⊗H\[n] → A⊗K\[n]
and
fK,H : A⊗K\[n] → A⊗H\[n] .
For π, ρ ∈ Sn , define
mπ,ρ : A⊗π\[n] ⊗ A⊗ρ\[n] → A⊗πρ\[n]
 7.3. APPROACH OF LEHN-SORGER VIA GRADED FROBENIUS ALGEBRAS 153

by
(7.15) mπ,ρ (a ⊗ b) = fπ,ρ,πρ f π,π,ρ (a) · f ρ,π,ρ (b) · eg(π,ρ) .
where g(π, ρ) is the graph defect defined in (7.13). The next result is [LS2, Propo-
sitions 2.13 and 2.14], and follows from (7.15) and repeated applications of (7.11).
Proposition 7.13. (i) The product A{Sn } × A{Sn } → A{Sn } defined
by
aπ · bρ = mπ,ρ (a ⊗ b)πρ
is associative, Sn -equivariant, and homogeneous of degree nd.
(ii) For any two homogeneous elements aπ, bρ ∈ A{Sn }, we have the following
(non)commutativity relation:
aπ · bρ = (−1)|a|·|b| π ∗ (b)πρπ −1 · aπ = (−1)|a|·|b| π̃ ∗ (bρ) · aπ.
By Proposition 7.13, A[n] is a subring of the center of A{Sn }. Let
T : A{Sn } → C
be defined by *
T (a), if π = 1,
T(aπ) =
0, else.
Then T induces a non-degenerate and Sn -invariant bilinear pairing on A{Sn }.
Therefore, the restriction of T to A[n] defines the structure of a graded Frobenius
algebra of degree nd on A[n] .
Let
V(A) = Sym(A ⊗ t−1 C[t−1 ])
be the bosonic Fock space modeled on the graded vector space A. Then V(A) is
bi-graded by degree and weight, where an element a ⊗ t−m ∈ A ⊗ t−m is given
degree |a| and weight m. The component of V(A) of weight n is the graded vector
space C
V(A)n ∼= Symαi A
αn i
where α = (1α1 2α2 · · · ) runs through all partitions of n. Let
f : {1, . . . , N } → π \[n]
be an enumeration of the orbits of π ∈ Sn , and let i = |f (i)|. Define Φ : A⊗N →
V(A) by
1
a1 ⊗ · · · ⊗ aN → (a1 ⊗ t−1 ) · · · (aN ⊗ t−N ),
n!
and let

+∞
Φ: A{Sn } → V(A)
n=0
⊗π\[n]
be given on the summand A · π by the composition
f −1 Φ
A⊗π\[n] −→ A⊗N −→ V(A).
Note that Φ is surjective and invariant under the Sn -action, so that its restriction
to A[n] is also surjective. Moreover, the vector spaces A[n] and V(A)n have the
same dimension. These observations lead to the following [LS2, Proposition 2.11].
154 7. THE COHOMOLOGY RINGS OF HILBERT SCHEMES OF POINTS

Proposition 7.14. Φ induces an isomorphism of graded vector spaces

A[n] → V(A)n .
Finally, the above constructions are applied to A = H ∗ (X)[2] for a smooth
projective complex surface X, where the linear form T : A = H ∗ (X)[2] → C is
given by
,
(7.16) T (a) = − a.
X
The following is from [LS2, Theorem 3.2].
Theorem 7.15. Let X be a smooth projective complex surface with a numer-
ically trivial canonical divisor. Then there is a canonical isomorphism of graded
rings
[n] ∼
H ∗ (X)[2] −→ H ∗ (X [n] )[2n].
=
(7.17)
The linear form T : H ∗ (X [n] )[2n] → C is given by
,
T (a) = (−1)n · a
X [n]
which is consistent with (7.16) when n = 1. We now sketch the main ideas in the
proof of Theorem 7.15, and refer to Sections 3 and 4 in [LS2] for further details.
Put A = H ∗ (X)[2]. By Theorem 3.8, there is an isomorphism of graded vector
spaces

+∞
Ψ : V(A) → H ∗ (X [n] )[2n]
n=0
sending (a1 t−n1 ) · · · (as t−ns ) to a−n1 (a1 ) · · · a−ns (as )|0 . Note that Ψ maps the n-
th component V(A)n to H ∗ (X [n] )[2n]. It was observed in [Leh1, LQW1] that the
ring structure of H ∗ (X [n] )[2n], n ≥ 1 is uniquely determined by the linear isomor-
phism Ψ, Theorem 3.24 (ii) (also (4.19)), and Theorem 4.2. Another observation is
the relation [FW] between Goulden’s operator [Gou, Section 3] on the symmetric
groups and Lehn’s boundary operator (4.19) in the Hilbert schemes. Exploiting
these observations shows that the restriction of Ψ ◦ Φ to A[n] is the desired isomor-
phism (7.17).

7.4. Approach of Costello-Grojnowski via

Calogero-Sutherland operators
The approach of Lehn and Sorger [LS2] described in Section 7.3 fails when the
canonical divisor of the smooth projective complex surface X is not numerically
trivial. In [CoG], Costello and Grojnowski came up with a different approach to
deal with an arbitrary smooth projective complex surface X. In this section, we
will outline the main ideas in [CoG]. To avoid the formalism and technicalities, our
presentation follows that of [CoG, Introduction].
The starting point of [CoG] is the same as what is outlined for the proof of
Theorem 7.15: the ring structure of H ∗ (X [n] ), n ≥ 1 is uniquely determined by the
theorem of Nakajima and Grojnowski (Theorem 3.8), Lehn’s boundary operator d
(Theorem 3.24 (ii) and (4.19)), and Theorem 4.2.
So let X be a smooth projective complex surface. Put
H = H ∗ (X).
 7.4. APPROACH OF COSTELLO AND GROJNOWSKI 155

Regard H as a graded Frobenius algebra over C. By Theorem 3.8, there is a

canonical identification

+∞
(7.18) F = Sym(xH[x]) ∼
= HX = H ∗ (X [n] )
n=0

as modules for the Heisenberg algebra modeled on H. Write


+∞
F= Fn
n=0

where F is the n-th eigenspace of the energy operator

n

∂
∂=x .
∂x
Via (7.18), F n is linearly identified with H ∗ (X [n] ). Cup product with an element

+∞
a∈F = Fn
n=0

defines a linear operator on F. For reasonable a, such a linear operator is a differen-

tial operator. This differential operator commutes with Lehn’s boundary operator
d since cohomology is a (super)-commutative ring. Hence there are at least as many
differential operators on F which commute with d as the dimension of the space F.
Thinking of d as a Hamiltonian, this is precisely saying that d defines an in-
tegrable system. It rephrases the problem of determining the ring structure on F
as the problem of finding integrals of motion, i.e., determining the centralizer of
d in an appropriate ring of differential operators on F, and showing that the cen-
tralizer is commutative and in bijection with F. As noted in [FW, Introduction]
and [Mat, Section 5], d is a version of the bosonized Calogero-Sutherland Hamil-
tonian. So the problem of describing the cohomology ring structure of the Hilbert
scheme X [n] becomes an algebraic problem, that of solving a deformed version of
the Calogero-Sutherland system.
To carry out the observations in the preceding paragraph, a ring, denoted by
Diff H,K  F,
of continuous differential operators on the Fock space F is defined. It depends on
the ring structure on H = H ∗ (X) and an additional parameter
K  ∈ H 0 (X) ⊕ H 2 (X) ⊕ H 4 (X).
It is a limit of finitely generated algebras and much smaller than the ring of all
differential operators. Moreover, Diff H,K  F contains the Calogero-Sutherland op-
erator
LK 
which is a second order differential operator and coincides with d if K  is equal to the
canonical divisor KX of X. The Calogero-Sutherland operator LK  can be written
as the inverse limit of r-particle generalized Calogero-Sutherland Hamiltonians.
Each r-particle system is integrable via the methods of Dunkl, Cherednik and
Opdam.
For generic K  , define
IMK 
156 7. THE COHOMOLOGY RINGS OF HILBERT SCHEMES OF POINTS

to be the centralizer in Diff H,K  F of LK  and the energy operator ∂. Since IMK 
commutes with ∂, it projects to give a subalgebra of End(F n ) for each n ≥ 1.
An explicit construction of IMK  as a polynomial algebra is given. The algebra
IMKX is obtained by degenerating
K  = λ + KX
to KX as λ → 0. A description of IMKX in terms of degenerating families of
Dunkl-Cherednik operators is then obtained.
The following is the main result [CoG, Theorem 1.0.1].
Theorem 7.16. IMKX is a commutative algebra, and for every n the image
of IMKX in End(F n ) is precisely the algebra of left multiplication operators (i.e.,
the image of the natural map H ∗ (X [n] ) → End(F n ), a → a∪).
Theorem 7.16 together with the explicit structure of IMKX provides a new
description of the cohomology ring of X [n] . We refer to [CoG] for further details.
 CHAPTER 8

Ideals of the cohomology rings of Hilbert schemes

In Chapter 7, we studied the cohomology ring of the Hilbert scheme X [n] of

points on a smooth projective complex surface X. However, the situation changes
dramatically when X is quasi-projective. The understanding of the cohomology
ring H ∗ (X [n] ) for quasi-projective X has been rather limited with the exception
of the affine plane [ES2, Leh1, LS1, Vas] and the minimal resolution of a simple
singularity C2 /Γ where Γ is a finite group of SL2 (C) [QW3, Wan3]. Besides these
minimal resolutions, typical important examples of quasi-projective surfaces include
the cotangent bundle of a smooth projective curve and the surface obtained from a
smooth projective surface by deleting a point. All these surfaces are among a class
of quasi-projective surfaces which satisfy the S-property (Definition 8.20). In this
chapter, we will establish some general results regarding the cohomology rings of
the Hilbert schemes of points for such a large class of quasi-projective surfaces. As
a preparation, for a smooth projective complex surface X, we will define the ideal
I [n] of the cohomology ring H ∗ (X [n] ) associated to an ideal I of the cohomology
ring H ∗ (X). We will prove that when


4
I= H  (X),
=1

the quotient ring H ∗ (X [n] )/I [n] is isomorphic to the cohomology ring H ∗ ((C2 )[n] ) of
the Hilbert schemes of points on the affine plane C2 . For this purpose, we will review
the cohomology ring H ∗ ((C2 )[n] ), following [ES2,Leh1,LS1,Vas]. Moreover, when
I = H 4 (X), we will show that modulo the ideal I [n] , the structure constants in
certain cup products of the Hilbert schemes X [n] are independent of n. The main
results and their proofs in this chapter are from [LQW5].

8.1. The cohomology ring of the Hilbert scheme (C2 )[n]

In this section, we will address the issue of Heisenberg algebra actions on the
cohomology of the Hilbert scheme X [n] when X is a smooth quasi-projective com-
plex surface. Then following [Leh1, LS1, Vas], we will see that the cohomology
ring of the Hilbert scheme (C2 )[n] can be identified with the class algebra (with
the cup product) of the symmetric group Sn . This part of the discussion may be
viewed as a simplified version of Section 7.3 with A = C.
We begin with some discussions about the Heisenberg algebra actions when the
complex surface is quasi-projective. The constructions of the Heisenberg algebra
actions in Section 3.2 are carried out for smooth projective surfaces. According to
[Nak3, Section 3], with minor modifications, these constructions hold for a smooth
quasi-projective surface X. Let H∗BM (X) be the Borel-Moore homology of X,
157
158 8. IDEALS OF THE COHOMOLOGY RINGS OF HILBERT SCHEMES

which is defined by
H∗BM (X) = H∗ (X̂, ∞)
where X̂ = X ∪ {∞} is the one-point compactification of X, and H∗ (X̂, ∞) is the
ordinary relative homology of the pair (X̂, ∞). Then the creation operators are
modeled on the Borel-Moore homology H∗BM (X), while the annihilation operators
are modeled on the ordinary homology H∗ (X). The Fock space of the Heisenberg
algebra is taken to be the direct sum

+∞
H∗BM (X [n] )
n=0

of the Borel-Moore homology groups H∗BM (X [n] ). Let Hc∗ (X) be the cohomology
with compact support. Using the the Poincaré dualities
PD : H 4−i (X) → HiBM (X)
PD : Hc4−i (X) → Hi (X),
we can regard the creation operators a−n (α) with n > 0 as being modeled on H ∗ (X)
(i.e., α ∈ H ∗ (X)), while we can regard the annihilation operators an (β) with n > 0
as being modeled on Hc∗ (X) (i.e., β ∈ Hc∗ (X)). Accordingly, with the help of the
Poincaré duality
PD : H 4n−i (X [n] ) → HiBM (X [n] ),
from now on we can take the Fock space to be the direct sum

+∞
(8.1) HX = H ∗ (X [n] )
n=0

of the ordinary cohomology groups H ∗ (X [n] ).

Let X be a smooth projective surface compactifying the smooth quasi-projective
surface X, and let ι : X → X be the inclusion map. Then we have induced embed-
[n]
dings ιn : X [n] → X for n ≥ 0, and induced ring homomorphisms
[n]
ι∗n : H ∗ (X ) → H ∗ (X [n] ).
The maps ι∗ and ι∗n are related by the following.
Lemma 8.1. Let notations be as in the preceding paragraph. Then,
(8.2) ι∗n a−n1 (α1 ) · · · a−nk (αk )|0 = a−n1 (ι∗ α1 ) · · · a−nk (ι∗ αk )|0
where k ≥ 0, n1 , . . . , nk > 0, and n1 + . . . + nk = n.
Proof. For n ≥ 0, let
[n]
(ι∗n )BM : HiBM (X ) → HiBM (X [n] )
[n]
be the natural map induced by the embedding ιn : X [n] → X . Then, it is
well-known that
(ι∗n )BM ◦ PD = PD ◦ (ι∗n )BM .
Combining with the construction of the creation operators, we obtain
ι∗m1 +m2 a−m2 (α)(A) = a−m2 (ι∗ α)(ι∗m1 A)
[m1 ]
for m1 , m2 > 0 and A ∈ H ∗ (X ). Applying this repeatedly, we get (8.2). 
 8.1. THE COHOMOLOGY RING OF THE HILBERT SCHEME (C2 )[n] 159

In the rest of this section, we consider the affine plane X = C2 . Since H ∗ (C2 ) =
C · 1C2 , we see that the Fock space

+∞
HC2 = H ∗ (C2 )[n]
n=0

has a linear basis consisting of the Heisenberg monomial classes

(8.3) a−n1 (1C2 )r1 · · · a−nk (1C2 )rk |0
where k ≥ 0 and n1 , . . . , nk , r1 , . . . , rk > 0.
Let C[q1 , q2 , . . .] be the polynomial ring in countably infinitely many variables.
Then there is a natural isomorphism of Fock spaces:
(8.4) Ψ1 : C[q1 , q2 , . . .] → HC2
defined by
(8.5) Ψ1 qnr11 · · · qnrkk = a−n1 (1C2 )r1 · · · a−nk (1C2 )rk |0 .
Under this linear isomorphism, the creation operators a−n (1C2 ), n > 0 on HC2 cor-
respond to the multiplication by qn on C[q1 , q2 , . . .], while the annihilation operators
on HC2 correspond to the operators
∂
(8.6) −m
∂qm
on C[q1 , q2 , . . .].
As for the ring structure on H ∗ ((C2 )[n] ), note that the Chern character opera-
tors Gk (1C2 ) from Definition 4.1 (iii) are well-defined on the Fock space HC2 .
Theorem 8.2. Let ∂m = m ∂q∂m .
(i) Via the linear isomorphism Ψ1 in (8.4),
(−1)k 
Gk (1C2 ) = qn +...+nk ∂n0 · · · ∂nk .
(k + 1)! n ,...,n >0 0
0 k

(ii) For each n, the cohomology ring H ∗ ((C2 )[n] ) is generated over C by the
Chern characters chk (OC2 )[n] .
Theorem 8.2 (i) is from [Leh1, Theorem 4.10], while Theorem 8.2 (ii) is a
classical result that follows from [ES2, Theorem 1.1]. Intuitively, Theorem 8.2 (i)
also follows from Theorem 4.7. Indeed, since KC2 = eC2 = 0, Theorem 4.7 says
that Gk (1C2 ) is equal to
 1
(8.7) − aλ (τ∗ 1C2 ).
λ!
(λ)=k+2,|λ|=0

Note that aλ (τ∗ 1C2 ) is not zero only when the generalized partition λ has exactly
one negative part. Moreover, when λ has exactly one negative part, via the linear
isomorphism Ψ1 , the operator aλ (τ∗ 1C2 ) is of the form
qn0 +...+nk (−∂n0 ) · · · (−∂nk )
where n0 , . . . , nk > 0 (see (8.6)). Recalling the definition of λ! from Definition 1.6
and keeping track of the coefficient of qn0 +...+nk (−∂n0 ) · · · (−∂nk ), we conclude that
Theorem 8.2 (i) is a consequence of (8.7).
160 8. IDEALS OF THE COHOMOLOGY RINGS OF HILBERT SCHEMES

In [LS1, Vas], the cohomology H ∗ ((C2 )[n] ) as a C-algebra is further identified

with the class algebra (or the algebra of class functions)
C(Sn )
of the symmetric
 group Sn . Let C[Sn ] be the group algebra of Sn over C. An
element π∈Sn aπ π may be regarded as a function f : Sn → C defined by f (π) =
aπ . The convolution product  on C[Sn ] is defined by

(8.8) (f ∗ g)(π) = f (πσ −1 )g(σ).
σ∈Sn

For each partition λ = (λ1 ≥ . . . ≥ λ ) of n, define


(8.9) Cλ = π
π of type λ

where π runs through all permutations with cycle type λ, i.e., those having a disjoint
cycle decomposition with cycle lengths λ1 , . . . , λ .
The class algebra C(Sn ) is the center of C[Sn ], and has a linear basis
{Cλ |λ  n}.
The product on C(Sn ) is the cup product ∪ defined by
*
σ  π, if d(σ) + d(π) = d(σπ),
σ∪π =
0, otherwise
where d(π) is defined in (7.12). Put

+∞
(8.10) C= C(Sn ).
n=0

In [FJW], an isomorphism of vertex algebras

(8.11) Ψ2 : C → C[q1 , q2 , . . .]
was defined by putting
  1 " qi #mi 
(8.12) Ψ2 (Cλ ) =
mi ! i
i≥1

for λ = (1 m1 m2
2 · · · ). The following is from [LS1, Theorem 1.1].
Theorem 8.3. The composite isomorphism of Fock spaces
Ψ Ψ
C −→
2
C[q1 , q2 , . . .] −→
1
HC2
induces an isomorphism of C-algebras
C(Sn ) −→ H ∗ (C2 )[n]
for each n ≥ 0.
The main idea in the proof of Theorem 8.3 is to use the alternating character
εn ∈ C(Sn ) defined by
 
εn = (−1)d(π) π = sgn(π) π.
π∈Sn π∈Sn
 8.2. IDEALS IN H ∗ (X [n] ) FOR A PROJECTIVE SURFACE X 161

It is known that
 
  m
n m−1 z
(8.13) Ψ2 (εn )z = exp (−1) qm .
m>0
m
n≥0

On the other hand, it follows from Theorem 8.2 (i) that

 
  m
−1 n m−1 z
Ψ1 γn z = exp (−1) qm .
m>0
m
n≥0

where γn denotes the total Chern class of the tautological bundle (OC2 )[n] . There-
fore, Ψ1 Ψ2 maps the alternating character εn ∈ C(Sn ) to the total Chern class of
(OC2 )[n] in H ∗ (C2 )[n] . In addition, via some technical computation, one proves
that
Ψ1 Ψ2 (εn ∪ w) = γn ∪ Ψ1 Ψ2 (w)
for every w ∈ C(Sn ). Now Theorem 8.3 follows since by Theorem 8.2 (ii), the
cohomology ring H ∗ (C2 )[n] is generated by the Chern classes of (OC2 )[n] .

8.2. Ideals in H ∗ (X [n] ) for a projective surface X

In this section, X denotes a smooth projective complex surface. We will define
the ideal I [n] of the cohomology ring H ∗ (X [n] ) associated to an ideal I of the
cohomology ring H ∗ (X). Moreover, we will study the properties of I [n] , including
its generators.
The first lemma is elementary.
Lemma 8.4. Let I be an ideal in the cohomology ring H ∗ (X). Let α ∈ I and
k ≥ 2. Then, the pushforward τk∗ α can be written as

αj,1 ⊗ · · · ⊗ αj,k ∈ H ∗ (X k )
j

such that for each fixed j, there exists some with αj, ∈ I.
Proof. First of all, note from the third formula in (3.11) that if

τk∗ α = αj,1 ⊗ αj,2 ⊗ · · · ⊗ αj,k ,
j

then 
τ(k+1)∗ α = (τ2∗ αj,1 ) ⊗ αj,2 ⊗ · · · ⊗ αj,k .
j
Therefore, by induction, it suffices to prove the lemma for k = 2. Now the case
k = 2 follows from the first formula in (3.11) that if we write

τ2∗ (1X ) = αi ⊗ βi ,
i

then 
τ2∗ (α) = (ααi ) ⊗ βi
i
with (ααi ) ∈ I. 
In view of the preceding lemma, we introduce the following important definition.
162 8. IDEALS OF THE COHOMOLOGY RINGS OF HILBERT SCHEMES

Definition 8.5. Let X be a smooth projective complex surface, and I be an

ideal in the cohomology ring H ∗ (X). For n ≥ 1, define I [n] to be the subset of
H ∗ (X [n] ) consisting of the linear spans of Heisenberg monomials of the form
a−n1 (α1 ) · · · a−nb (αb )|0

where αi ∈ I for some i, and n1 , . . . , nb are positive integers with  n = n.
Lemma 8.6. Let I be an ideal in the cohomology ring H ∗ (X). Then,
(i) the linear subspace I [n] in H ∗ (X [n] ) is an ideal.
(ii) Gk (α, n) ∈ I [n] if α ∈ I.
Proof. (i) By Theorem 7.5, the cohomology ring H ∗ (X [n] ) is generated by
the cohomology classes Gk (α, n). So it suffices to prove
(8.14) Gk (α)a−n1 (α1 ) · · · a−nb (αb )|0 ∈ I [n]

whenever α1 ∈ I, and n1 , . . . , nb are positive with  n = n. For simplicity, put
g = Gk (α). Then, the operator g is of bi-degree (0, 2k + |α|), and satisfies
[[· · · [g, am1 (β1 )], · · · ], amk+2 (βk+2 )] = 0
whenever mi < 0 for all i by Theorem 4.7. Now we see from Lemma 4.9 that
Gk (α)a−n1 (α1 ) · · · a−nb (αb )|0 is a linear combination of expressions of the form
⎛ ⎞

(8.15) ⎝ a−n (α )⎠ [[· · · [g, a−nσ (1) (ασi (1) )], · · · ], a−nσ (i) (ασi (i) )]|0
i i
∈σi0

where 0 ≤ i ≤ k + 1, σi maps the set { 1, . . . , i } to the set { 1, . . . , b } with σi (1) <

· · · < σi (i), and
σi0 = { | 1 ≤ ≤ b, = σi (1), . . . , σi (i)}.
If 1 ∈ σi , then (8.15) is contained in I [n] . In the following, we assume 1 ∈ σi0 .
0

So 1 = σi (1) since σi (1) < · · · < σi (i). By Theorem 4.7 and Lemma 3.12 (i), (8.15)
is a linear combination of expressions of the form
⎛ ⎞

(8.16) ⎝ a−n (α )⎠ a−λ (τ∗ (αασi (1) · · · ασi (i) ))|0
∈σi0

where  ∈ {1X , eX , KX , KX2
}, λ  ij=1 nσi (j) , and (λ) = k + 2 − ||/2 − i. By
Lemma 8.4, the expression (8.16) is contained in I [n] since ασi (1) = α1 ∈ I. It
follows that (8.15) is contained in I [n] . This proves (8.14).
(ii) This follows from Theorem 4.10 and Lemma 8.4. 
The following technical definition is similar to Definition 4.4.
Definition 8.7. Let X be a smooth projective complex surface, s ≥ 1, and
t1 , . . . , ts ≥ 1. Fix mi,j ≥ 0 and βi,j ∈ H ∗ (X) for 1 ≤ i ≤ s and 1 ≤ j ≤ ti . Then, a
ti
universal linear combination of Gmi,j (βi,j , n), 1 ≤ i ≤ s is a linear combination
j=1
of the form

s 
ti
fi Gmi,j (βi,j , n)
i=1 j=1
where the coefficients fi are independent of X and n.
 8.2. IDEALS IN H ∗ (X [n] ) FOR A PROJECTIVE SURFACE X 163

The next result deals with the generators of the ideal I [n] .

Theorem 8.8. Let X be a smooth projective complex surface, and I be an ideal

in the cohomology ring H ∗ (X). If I is homogeneous (i.e., I = ⊕4i=0 (I ∩ H i (X))),
then the ideal I [n] is generated by the classes Gk (α, n) with α ∈ I.

Proof. Note that every Heisenberg monomial class in I [n] can be written as
 s 

A = 1−(n−n0 ) a−ni (αi ) |0
i=1


s
where s ≥ 1, n1 , . . . , ns ≥ 1, n0 = ni , and α is contained in I and homogeneous
i=1
for some . By Lemma 8.6 (ii), it suffices to show that A ∈ I [n] is a universal finite
linear combination of expressions of the form


t
(8.17) Gmj (βj , n)
j=1


t
where (mj + 1) ≤ n0 , and β ∈ I for some .
j=1
Use induction on n0 . When n0 = 1, s = n1 = 1. So

A = 1−(n−1) a−1 (α1 )|0 = G0 (α1 , n)

by (4.22). Hence the statement in the previous paragraph holds for n0 = 1.

Next assume n0 > 1. Let ki = ni − 1 for 1 ≤ i ≤ s. Then, ki ≥ 0 for every i.
s
By Lemma 4.13 and Lemma 4.14, the cup product Gki (αi , n) is equal to
i=1
 
s
(−1)ki
A
i=1
(ki + 1)!

(defined to be the leading term) plus a universal finite linear combination of expres-
sions
 s̃ 

1−(n−ñ0 ) a−ñi (α̃i ) |0
i=1

where α̃ ∈ I for some , s̃ ≥ 1, ñ1 , . . . , ñs̃ ≥ 1, and


s̃ 
s
ñi = ñ0 < (ki + 1) = n0 .
i=1 i=1

By induction, A is a universal finite linear combination of expressions of the form

(8.17). 

Remark 8.9. The assumption in Theorem 8.8 that the ideal I ⊂ H ∗ (X) is
homogeneous can be dropped when the surface X is simply connected.
164 8. IDEALS OF THE COHOMOLOGY RINGS OF HILBERT SCHEMES

Remark 8.10.
(i) Let X be a smooth projective complex surface, and I be a homogeneous
ideal in the cohomology ring H ∗ (X). Assume further that for every k ≥ 1
and α ∈ I, the pushforward τk∗ α can be written as

αj,1 ⊗ · · · ⊗ αj,k ∈ H ∗ (X k )
j

such that αj,s ∈ I for all j and s. Let

 s 

A = 1−(n−n0 ) a−ni (αi ) |0
i=1


s
where s ≥ 1, n1 , . . . , ns ≥ 1, n0 = ni , and α1 , . . . , αs ∈ I. Then the
i=1
same proof of Theorem 8.8 shows that A can be written as a polynomial
of the classes Gk (α, n), k ≥ 0 and α ∈ I. Moreover, the coefficients in the
polynomial are independent of n.
(ii) Under certain conditions, the conclusion in (i) can be sharpened. For
instance, let λ  n0 , α ∈ H ∗ (X) with |α| = 2, and m ≥ 1.
(ii-a) The class 1−(n−n0 ) a−λ (x)|0 can be written as a polynomial of the
classes Gk (x, n), k ≥ 0. Moreover, the coefficients are independent of
n and X.
(ii-b) If the odd Betti numbers of the surface X are equal to zero, then
1−(n−n0 −m) a−λ (x)a−m (α)|0

= α, KX · F1 (n) + Gki (α, n) · F2,i (n)
i

where F1 (n) and F2,i (n) are polynomials of the classes Gk (x, n), k ≥
0. Moreover, the coefficients of F1 (n) and F2,i (n) are independent of
n, α and X.
The statements (ii-a) and (ii-b) follow from the same proof of Theorem 8.8
by setting I = C · x ⊂ H ∗ (X) and I = C · x + C · α ⊂ H ∗ (X) respectively.
A natural problem to consider is to interpret the quotient ring
H ∗ (X [n] )/I [n]
for an ideal I of the cohomology ring H ∗ (X). In the next two sections, we will do
this for I = ⊕4=1 H  (X) and I = H 4 (X) respectively.

8.3. Relation with the cohomology ring of the Hilbert scheme (C2 )[n]
Let X denote a smooth projective complex surface. In this section, we will
study the quotient ring H ∗ (X [n] )/I [n] when

4
I= H  (X),
=1

and prove that it is isomorphic to the cohomology ring H ∗ ((C2 )[n] ) which has been
dealt with in Section 8.1.
 8.3. RELATION WITH THE COHOMOLOGY RING OF (C2 )[n] 165

First of all, note that H ∗ (X [n] )/I [n] has a linear basis consisting of Heisenberg
monomials of the form
a−n1 (1X )r1 · · · a−nk (1X )rk |0
k
where r1 , . . . , rk ≥ 1, and 0 < n1 < . . . < nk with =1 r n = n. Therefore, we
have an isomorphism of vector spaces

(8.18) Φ: H ∗ (X [n] )/I [n] → C[q1 , q2 , . . .]
n≥0

defined by
Φ a−n1 (1X )r1 · · · a−nk (1X )rk |0 = qnr11 · · · qnrkk .
Setting the degree of the variable qi to be i, we see that Φ maps H ∗ (X [n] )/I [n] to
the homogeneous component of C[q1 , q2 , . . .] of degree n.
Lemma 8.11. Let

4
I= H  (X).
=1
∗ [n] [n]
Then, the quotient ring H (X )/I is generated by the Chern character classes
Gk (1X , n), k = 0, 1, . . . , n − 1. Moreover,
(−1)k
(8.19) Gk (1X , n) ≡ · 1−(n−k−1) a−(k+1) (1X )|0 (mod I [n] ).
(k + 1)!
Proof. Since the ring H ∗ (X [n] ) is generated by the classes Gk (α, n) with
0 ≤ k < n and α ∈ H ∗ (X), the first statement follows from Lemma 8.6 (ii). To
prove (8.19), we note from Theorem 4.10 that the leading term in Gk (1X , n) is
(−1)k
· 1−(n−k−1) a−(k+1) (1X )|0
(k + 1)!
corresponding to j = k, λ  (j + 1) = k + 1, and (λ) = k − j + 1 = 1. The other
terms in Gk (1X , n) contain τi∗ () with either i ≥ 2 or  = eX , KX , KX
2
∈ I, and
hence are contained in I . This proves (8.19).
[n]

Theorem 8.12. Let X be a smooth projective complex surface, and

4
I= H  (X).
=1
∗
Then, the quotient H (X [n]
)/I [n]
is isomorphic to the cohomology ring H ∗ ((C2 )[n] ).
Proof. By Lemma 8.11, the quotient ring H ∗ (X [n] )/I [n] is generated by the
classes Gk (1X , n), k = 0, 1, . . . , n − 1. So by Theorem 8.2, it suffices to show that
via the isomorphism Φ in (8.18), the linear operator gk on C[q1 , q2 , . . .] induced by
the operator Gk (1X ) on H ∗ (X [n] ) is given by
n≥0

(−1)k 
(8.20) gk = qn1 +...+nk+1 ∂n1 · · · ∂nk+1 .
(k + 1)! n
1 ,...,nk+1 >0

Indeed, let
A = a−n1 (1X ) · · · a−nb (1X )|0 ∈ H ∗ (X [n] )
166 8. IDEALS OF THE COHOMOLOGY RINGS OF HILBERT SCHEMES


where n1 , . . . , nb > 0 with  n = n. By Lemma 4.9, Gk (1X )(A) is equal to
(8.21)
 
k+1
a−n (1X ) · [[· · · [Gk (1X ), a−nσi (1) (1X )], · · · ], a−nσi (i) (1X )]|0
i=0 σi ∈σi0

where for each fixed i, σi runs over all the maps { 1, . . . , i } → { 1, . . . , b } satisfying
σi (1) < · · · < σi (i), and
σi0 = { | 1 ≤ ≤ b, = σi (1), . . . , σi (i)}.
Note from Lemma 3.12 that
[[· · · [at1 · · · atr (τr∗ α), a−nσi (1) (1X )], · · · ], a−nσi (i) (1X )]|0 ∈ I [n]
if i ≤ r − 2 or α ∈ I. Hence by (8.21) and Theorem 4.7, Gk (1X )(A) equals
⎛ ⎞
 
⎝ a−n (1X )⎠ ·
σk+1 0
∈σk+1
⎡⎡ ⎡ ⎤ ⎤ ⎤
 1
⎣⎣· · · ⎣− aλ (τ∗ 1X ), a−nσk+1 (1) (1X )⎦ , · · · ⎦ , a−nσk+1 (k+1) (1X )⎦ |0
λ!
(λ)=k+2,|λ|=0

modulo I . By Lemma 3.12 (i) again, the preceding expression is equal to

[n]

 k+1 ⎛ ⎞
  
− (−nσk+1 () ) ⎝ a−n (1X )⎠ a−nσk+1 (1) −...−nσk+1 (k+1) (1X )|0 .
σk+1 =1 0
∈σk+1

Therefore, we conclude that for the induced operator gk , we have

gk (qn1 · · · qnb )
 ⎛ ⎞
 k+1  
= (−1)k nσk+1 () ⎝ qn ⎠ qnσk+1 (1) +...+nσk+1 (k+1)
σk+1 =1 0
∈σk+1
 ⎛ ⎞
(−1)k 
 k+1 
= nσ̃k+1 () ⎝ qn ⎠ qnσ̃k+1 (1) +...+nσ̃k+1 (k+1)
(k + 1)! σ̃ 0
k+1 =1 ∈σ̃k+1

where σ̃k+1 runs over all injective maps { 1, . . . , k + 1 } → { 1, . . . , b } and

0
σ̃k+1 = { | 1 ≤ ≤ b, = σ̃k+1 (1), . . . , σ̃k+1 (k + 1)}
(so the only difference between σ̃k+1 and σk+1 is that we have dropped the condition
σk+1 (1) < · · · < σk+1 (k + 1)). Finally, the above formula for gk (qn1 · · · qnb ) is
equivalent to (8.20). 

8.4. Partial n-independence of structure constants for X projective

Let X be a smooth projective complex surface. In this section, we will define
a linear basis of H ∗ (X [n] ) consisting of the classes bρ (n) where ρ denotes certain
type of partitions. Let I = H 4 (X). The main theorem states that modulo the ideal
I [n] , all the structure constants in the cup product bρ (n) · bσ (n) are independent
of n.
 8.4. PARTIAL n-INDEPENDENCE OF STRUCTURE CONSTANTS 167

First of all, take a linear basis

S = S0 ∪ S1
∗
of H (X) such that 1X , x ∈ S0 , S0 ⊂ H even (X) and S1 ⊂ H odd (X). Let notations
such as P(S) be the same as in Section 7.2. If we write ρ = (ρ(c))c∈S and
ρ(c) = (r mr (ρ(c)) )r≥1 = (1m1 (ρ(c)) 2m2 (ρ(c)) . . .),
then we put  
(ρ) = (ρ(c)) = mr (ρ(c))
c∈S c∈S,r≥1
and
 
ρ = |ρ(c)| = r · mr (ρ(c)).
c∈S c∈S,r≥1

Given ρ ∈ P(S) and n ≥ ρ + (ρ(1X )), we define ρ̃ ∈ P(S) by putting

⎧
⎪
⎨mr (ρ(c)), if c ∈ S − {1X },
mr (ρ̃(c)) = mr−1 (ρ(1X )), if c = 1X and r ≥ 2,
⎪
⎩
n − ρ − (ρ(1X )), if c = 1X and r = 1.
Note that ρ̃ = n. We define bρ (n) ∈ H ∗ (X [n] ) by
1 
(8.22) bρ (n) = a−r (c)mr (ρ̃(c)) · |0
r≥1 (r mr (ρ̃(1X )) m r (ρ̃(1X ))!) c∈S r≥1

1−(n−ρ−(ρ(1X ))) 
(8.23) = mr (ρ̃(1X )) m (ρ̃(1 ))!)
a−r (c)mr (ρ̃(c)) · |0 .
r≥2 (r r X c∈S,r≥1
c=1X or r>1

where we fix the order of the elements c ∈ S1 appearing in the product once
c∈S
and for all. For 0 ≤ n < ρ + (ρ(1X )), we set bρ (n) = 0. This is consistent with
(8.23) and Definition 3.18. We remark that the only part in bρ (n) involving n is
the factor 1−(n−ρ−(ρ(1X ))) in (8.23) when n ≥ ρ + (ρ(1X )).
Remark 8.13. The Heisenberg monomial class bρ (n) is different from the
Heisenberg monomial class aρ (n) defined in Section 7.2.
By Theorem 3.8, H ∗ (X [n] ) has a linear basis consisting of the classes:
(8.24) bρ (n), ρ ∈ P(S) and ρ + (ρ(1X )) ≤ n.
Fix a positive integer n and choose ρ, σ ∈ P(S) satisfying ρ + (ρ(1X )) ≤ n and
σ + (σ(1X )) ≤ n. Then we can write the cup product bρ (n) · bσ (n) as

(8.25) bρ (n) · bσ (n) = bνρσ (n) bν (n)
ν∈P(S)

where we have used bνρσ (n) ∈ C to denote the structure constants.

Proposition 8.14. Let X be a smooth projective complex surface. The struc-
ture constants bνρσ (n) of the cohomology H ∗ (X [n] ) are polynomials in n of degree at
most
(8.26) ( ρ + (ρ(1X ))) + ( σ + (σ(1X ))) − ( ν + (ν(1X ))).
168 8. IDEALS OF THE COHOMOLOGY RINGS OF HILBERT SCHEMES

Proof. For simplicity, let


f (ρ) = (r mr (ρ̃(1X )) mr (ρ̃(1X ))!).
r≥2

Then, f (ρ) is independent of n. By Theorem 7.7, the cup product

f (ρ)bρ (n) · f (σ)bσ (n)
is a linear combination of expressions of the form:
(n − ν − (ν(1X )))!
· f (ν)bν (n),
(n − ν − (ν(1X )) − i)!
such that i ≥ 0,
( ν + (ν(1X )) + i) ≤ ( ρ + (ρ(1X ))) + ( σ + (σ(1X ))),
and all the coefficients in this linear combination are independent of n. It follows
that all the structure constants bνρσ (n) are polynomials in n of degree at most
(8.26). 

In the following, we will study various cup products in the cohomology of X [n]
modulo the ideal I [n] with I = H 4 (X). We begin with an elementary topological
lemma.
Lemma 8.15. If k ≥ 2 and α is homogeneous, then

τk∗ α = αj,1 ⊗ · · · ⊗ αj,k
j

where for each j, either |αj, | = 4 for some , or 0 < |αj, | < 4 for every .
Proof. Assume |αj, | < 4 for every . If |αj, | = 0 for some , then

k
4(k − 1) + |α| = |τk∗ α| = |αj,i | ≤ 3(k − 1).
i=1

So (k − 1) + |α| ≤ 0, contradicting k ≥ 2. 
s
Lemma 8.16. Let I = H 4 (X), s ≥ 0, n1 , . . . , ns > 0, ñ = =1 n , and
n ≥ ñ. Let α, α1 , . . . , αs ∈ H ∗ (X) be homogeneous. Assume that k + |α| ≥ 1 and
n + |α | ≥ 2 for every . Then modulo I [n] , the cup product
Gk (α, n) · 1−(n−ñ) a−n1 (α1 ) · · · a−ns (αs )|0
is a universal linear combination of the basis ( 8.24).
Proof. By Lemma 8.6 (i) and (ii), the statement is trivial if one of the classes
α, α1 , . . . , αs ∈ H ∗ (X) is contained in I. So in the rest of the proof, we assume
that none of the classes α, α1 , . . . , αs is contained in I.
Our argument is similar to the proof of Lemma 8.6 (i). Put g = Gk (α) and
B = Gk (α, n) · 1−(n−ñ) a−n1 (α1 ) · · · a−ns (αs )|0 .
Then, B is equal to
1
ga−1 (1X )n−ñ a−n1 (α1 ) · · · a−ns (αs )|0 .
(n − ñ)!
 8.4. PARTIAL n-INDEPENDENCE OF STRUCTURE CONSTANTS 169

By Lemma 4.9, B is a universal finite linear combination of expressions of the form:

  
1 n − ñ
(8.27) a−1 (1X )n−ñ−j a−n (α )
(n − ñ)! j 0 ∈σi

·[[· · · [[[[g, a−1 (1X )], . . .], a−1 (1X )], a−nσi (1) (ασi (1) )], · · · ], a−nσi (i) (ασi (i) )]|0
/ 01 2
j times

where 0 ≤ j ≤ n − ñ, 0 ≤ i ≤ s, 0 ≤ j + i ≤ k + 1, σi maps the set { 1, . . . , i } to

the set { 1, . . . , s } with σi (1) < · · · < σi (i), and
σi0 = { | 1 ≤ ≤ s, = σi (1), . . . , σi (i)}.
By Theorem 4.7 and Lemma 3.12 (i), we conclude that
[[· · · [[[[g, a−1 (1X )], . . .], a−1 (1X )], a−nσi (1) (ασi (1) )], · · · ], a−nσi (i) (ασi (i) )]|0
/ 01 2
j times

is a universal finite linear combination of expressions

a−λ (τ∗ (αασi (1) · · · ασi (i) ))|0 ,
where  ∈ {1X , eX , KX , KX
2
}, λ  j + nσi (1) + · · · + nσi (i) and
(λ) = k + 2 − ||/2 − (j + i).
So B is a universal finite linear combination of expressions
⎛ ⎞

(8.28) 1−(n−ñ−j) ⎝ a−n (α )⎠ a−λ (τ∗ (αασi (1) · · · ασi (i) ))|0
∈σi0

where λ  j + nσi (1) + · · · + nσi (i) and (λ) = k + 2 − ||/2 − (j + i).

To prove our lemma, we see from (8.23) that it suffices to show that modulo
I [ñ+j] , the part
⎛ ⎞

⎝ a−n (α )⎠ a−λ (τ∗ (αασi (1) · · · ασi (i) ))|0
∈σi0

in (8.28) does not contain a−1 (1X ). Since n +|α | ≥ 2 for every , this is equivalent
to show that modulo I [(λ)] , the part a−λ (τ∗ (αασi (1) · · · ασi (i) ))|0 in (8.28) does
not contain a−1 (1X ). By Lemma 8.15, this is true if (λ) ≥ 2. So let (λ) = 1.
Then, we have
a−λ (τ∗ (αασi (1) · · · ασi (i) ))|0 = a−t (αασi (1) · · · ασi (i) )|0
where t = |λ| = j + nσi (1) + · · · + nσi (i) and k + 2 − ||/2 − (j + i) = 1.
If a−λ (τ∗ (αασi (1) · · · ασi (i) ))|0 contains a−1 (1X ), then we must have t = 1 and
|αασi (1) · · · ασi (i) | = 0. So j + nσi (1) + · · · + nσi (i) = 1,  = 1X , and
|α| = |ασi (1) | = . . . = |ασi (i) | = 0.
Thus, either j = 0, i = 1, nσ1 (1) = 1 and |ασ1 (1) | = 0, or j = 1 and i = 0. The first
case contradicts nσ1 (1) + |ασ1 (1) | ≥ 2. In the second case, we see from k + 2 − ||/2 −
(j + i) = 1 that k = 0, contradicting k + |α| ≥ 1. So a−λ (τ∗ (αασi (1) · · · ασi (i) ))|0
can not contain a−1 (1X ). 
170 8. IDEALS OF THE COHOMOLOGY RINGS OF HILBERT SCHEMES

Lemma 8.17. Let I = H 4 (X), s ≥ 1, k1 , . . . , ks ≥ 0, ki + |αi | ≥ 1 for every i.

Put

s
n0 = (ki + 1).
i=1

s
(i) Modulo I [n] , Gki (αi , n) is a universal linear combination of ( 8.24).
i=1

s
(ii) When n ≥ n0 , the leading term in Gki (αi , n) is equal to
i=1
   s 
s
(−1)ki 
1 a (αi ) |0
i=1
(ki + 1)! −(n−n0 ) i=1 −(ki +1)
which is equal to a universal multiple of bν (n) for some ν ∈ P(S).
Proof. (i) Use induction on s. When s = 1, our statement follows from
Lemma 8.16 (take the integer s there to be 0). Now, assume that the statement is

s−1
true for s − 1 with s ≥ 2, i.e., Gki (αi , n) is a universal finite linear combination
i=1
of the basis classes bν (n), ν ∈ P(S) and ν + (ρ(1X )) ≤ n. Note from (8.23) that
up to some universal factor, every basis class bν (n) is of the form
1−(n−ñ) a−n1 (β1 ) · · · a−ns (βs )|0 ,
where n1 , . . . , ns > 0, ñ = n1 + . . . + ns , n ≥ ñ, and n + |β | ≥ 2 for every . So
by Lemma 8.16, our statement for s follows.
(ii) The statement about the leading term comes from the proof of Theorem 8.8,
while the other statement about the universal multiple follows from (8.23). 

The next lemma studies the basis element bρ (n) modulo I [n] when I = H 4 (X).
Lemma 8.18. Let I = H 4 (X). Then modulo I [n] , the basis element bρ (n) is

t
a universal finite linear combination of products of the form Gmj (βj , n) where
j=1
mj + |βj | ≥ 1 for each j, and

t
(mj + 1) ≤ ρ + (ρ(1X )).
j=1

Proof. By (8.23), up to some universal factor, bρ (n) is of the form

 s 

1−(n−n0 ) a−(ki +1) (αi ) |0 ,
i=1
def s
where n ≥ n0 = ρ + (ρ(1X )) = i=1 (ki + 1), and ki ≥ 0 and ki + |αi | ≥ 1 for
every i. Now our result follows from an induction on n0 the same way as in the
proof of Theorem 8.8, and Lemma 8.17 (i) and (ii). 
To state our main result in this section, let I = H 4 (X) and
SI = {[x]} ⊂ S.
 8.5. APPLICATIONS TO QUASI-PROJECTIVE SURFACES 171

Regard P(S − SI ) ⊂ P(S). Then, (8.25) implies that


(8.29) bρ (n) · bσ (n) ≡ bνρσ (n) bν (n) (mod I [n] ).
ν∈P(S−SI )

Theorem 8.19. Let X be a smooth projective complex surface, and I = H 4 (X).

Then, all the structure constants bνρσ (n) in ( 8.29) are independent of n.
Proof. By Lemma 8.18, bρ (n) is a universal finite linear combination of ex-
pressions of the form
 t1
Gm1,j (β1,j , n)
j=1
where m1,j + |β1,j | ≥ 1 for every j. Similarly, bσ (n) is a universal finite linear
combination of expressions of the form

t2
Gm2,j (β2,j , n)
j=1

where m1,j + |β1,j | ≥ 1 for every j. Therefore, bρ (n) · bσ (n) is a universal finite
linear combination of expressions of the form

t
Gmj (βj , n)
j=1

where mj + |βj | ≥ 1 for every j. Now our result follows from Lemma 8.17 (i). 

8.5. Applications to quasi-projective surfaces with the S-property

In this section, we will apply our results in previous sections to smooth quasi-
projective surfaces satisfying the S-property. Our terminology “quasi-projective”
here means “quasi-projective but not projective”.
Definition 8.20. A smooth quasi-projective complex surface X satisfies the
S-property if it can be embedded in a smooth projective complex surface X such
that the induced ring homomorphism H ∗ (X) → H ∗ (X) is surjective.
Example 8.21. Let X be a projective surface and let X be the quasi-projective
surface obtained from X with a point removed. It is easy to see that this smooth
quasi-projective surface X satisfies the S-property.
Example 8.22. Let Γ be a finite subgroup of SL2 (C). Let X be the minimal
resolution of the simple singularity C2 /Γ. It is known that this smooth quasi-
projective surface X satisfies the S-property. Moreover, KX = 0.
Example 8.23. (The cotangent bundle of a smooth projective curve) Consider
the ruled surface
X = P(OC (−KC ) ⊕ OC )
where C is a smooth projective curve. Let σ be the section (to the projection
X → C) corresponding to the natural surjection
OC (−KC ) ⊕ OC → OC (−KC ) → 0,
and put X = X − σ. Then, X is the total space of the cotangent bundle of C, and
KX = 0. We claim that X satisfies the S-property. In fact, the following general
172 8. IDEALS OF THE COHOMOLOGY RINGS OF HILBERT SCHEMES

statement is true. Let X = P(L1 ⊕ L2 ) where L1 and L2 are two invertible sheaves
over C. Let σ (respectively, σ  ) be the section of X → C corresponding to the
natural surjection L1 ⊕ L2 → L1 → 0 (respectively, L1 ⊕ L2 → L2 → 0). Put
X = X − σ. Then X satisfies the S-property. To see this, let X  = X − σ  , and
notice that X and X  are affine bundles over C. Hence X is homotopic to C, and
H i (X) ∼= H i (C) for every i. Therefore, to verify the S-property of X, it remains
to verify the surjectivities of the induced homomorphisms ri : H i (X) → H i (X) for
i = 1, 2. Consider the relative cohomology group H 2 (X, X). We have
H 2 (X, X) ∼= H 2 (X − σ  , X − σ  ) = H 2 (X  , X  − σ)
by the excision theorem. By the Thom isomorphism, since X  is an affine bundle
over C with σ being the zero section, we have
H 2 (X  , X  − σ) ∼
= H 0 (C) ∼
= C.
Hence H 2 (X, X) ∼
= C. Now consider the exact sequence
r δ r
H 1 (X) −→ 1
H 1 (X) −→ H 2 (X, X) −→ H 2 (X) −→2
H 2 (X).
Since H 2 (X) = ∼ H 2 (C) =
∼ C and H 2 (X) ∼
= C ⊕ C, we conclude that the map δ must
be zero and the map r2 must be surjective. Therefore, r1 is also surjective.
In the following, let X be a smooth quasi-projective surface satisfying the S-
property. So X can be embedded in a smooth projective surface X such that the
induced ring homomorphism ι∗ : H ∗ (X) → H ∗ (X) is surjective where ι : X → X is
the inclusion map. Let I = ker(ι∗ ). Fix a linear basis S of H ∗ (X) as in Section 8.4
such that S contains a linear basis SI of I and ι∗ (S −SI ) is a linear basis of H ∗ (X).
Put
SX = ι∗ (S − SI )
By Lemma 8.1, ker(ι∗n ) = I [n] which is defined in Definition 8.5. Also, a linear basis
of H ∗ (X [n] ) is given by
(8.30) bρX (n), ρX ∈ P(SX ) and ρX + (ρX (1X )) ≤ n
where bρX (n) is defined in a similar way as in (8.22) and (8.23). So for ρX , σX ∈
P(SX ), we can write the cup product bρX (n) · bσX (n) as

(8.31) bρX (n) · bσX (n) = bνρX
X σX
(n) bνX (n)
νX ∈P(SX )

where bνρX
X σX
(n) ∈ C stands for the structure constants.
Theorem 8.24. Let X be a smooth quasi-projective surface satisfying the S-
property. Then the structure constants bνρX
X σX
(n) in ( 8.31) are independent of n.
Proof. Let notations be as above. Note that ι∗ (S − SI ) = SX . Define ρ ∈
P(S) by *
mr (ρX (ι∗ c)), if c ∈ (S − SI ),
mr (ρ(c)) =
0, if c ∈ SI .
Similarly, define σ ∈ P(S) by
*
mr (σX (ι∗ c)), if c ∈ (S − SI ),
mr (σ(c)) =
0, if c ∈ SI .
 8.5. APPLICATIONS TO QUASI-PROJECTIVE SURFACES 173

By Theorem 8.19,
 [n]
bρ (n) · bσ (n) ≡ bνρσ (n) bν (n) (mod I )
ν∈P(S−{[x]})

where I = H 4 (X) and all the bνρσ (n) are independent of n. Since I ⊂ I,

(8.32) bρ (n) · bσ (n) ≡ bνρσ (n) bν (n) (mod I [n] )
ν∈P(S−SI )

where all the structure constants bνρσ (n) are independent of n. By Lemma 8.1,
ι∗n (bρ (n)) = bρX (n) and ι∗n (bσ (n)) = bσX (n). Therefore, applying ι∗n to (8.32), we
see that all the structure constants bνρX X σX
(n) in (8.31) are independent of n. 
Definition 8.25. Let X be a smooth quasi-projective surface satisfying the
S-property. We define the FH ring GX associated to X to be the ring with a linear
basis given by the symbols bρX , ρX ∈ P(SX ), with the multiplication given by

bρX · bσX = bνρX
X σX
bνX
νX ∈P(SX )

where the structure constants bνρX

X σX
come from (8.31).
Theorem 8.26. Let X be a smooth quasi-projective surface which satisfies the
S-property. Then the FH ring GX is isomorphic to the tensor product P ⊗ E, where
P is the polynomial algebra generated by
br,c , c ∈ SX ∩ H even (X), r ≥ 1,
and E is the exterior algebra generated by
br,c , c ∈ SX ∩ H odd (X), r ≥ 1.
The proof of Theorem 8.26 is similar to the proof of Theorem 7.11. We refer
to [LQW5, Section 4] for the details.
 CHAPTER 9

Integral cohomology of Hilbert schemes

The cohomology H ∗ (X [n] ) of the Hilbert scheme X [n] with complex coefficients
has been investigated in previous chapters. By Theorem 3.8, it is linearly spanned
over C by the Heisenberg monomial classes
a−n1 (α1 ) · · · a−nk (αk )|0
where n1 , . . . , nk > 0, n1 + . . . + nk = n, and the cohomology classes α1 , . . . , αk run
over a fixed linear basis of H ∗ (X). Its ring structure was discussed in Chapters 7
and 8. A natural problem is to study the integral cohomology of the Hilbert scheme
X [n] , its integral linear basis, and its integral generators. This chapter is devoted to
these issues. We will define integral operators, show that some familiar operators are
integral, and construct some integral operators by using integral classes in H 2 (X).
Moreover, we will prove that the intersection matrix of certain integral classes in
the middle cohomology H 2n (X [n] ) has determinant ±1. Then we will write down
a linear basis of the integral cohomology of X [n] , modulo the torsion part. As an
application, we will compare two integral bases of the integral cohomology of the
Hilbert scheme (P2 )[n] of points on the complex projective plane P2 , and prove that
when the basis elements are arranged properly, the transition matrix is a triangular
matrix. Further applications can also be found in [Kap, KaM]. Theorem 9.12,
Theorem 9.22 and their proofs are from [QW2], while Theorem 9.14, Theorem 9.23
and their proofs are from [LQ3]. The proof of Theorem 9.35 follows the presentation
of [Mur], while we refer to [Mur, Subsection 3.3] for the proof of Proposition 9.32.

9.1. Integral operators

In this section, we will introduce the concept of integral operators, and discuss
the methods to construct these operators. Moreover, we will prove that many of
the familiar operators, such as 1/zλ · a−λ (1X ), are integral operators.
Let X be a smooth projective complex surface. Recall from Definition 3.2 (i)
that

+∞
HX = H ∗ (X [n] ).
n=0

Definition 9.1. (i) A class A ∈ H ∗ (X [n] ) is integral if it is contained in

H ∗ (X [n] ; Z)/Tor ⊂ H ∗ (X [n] )
where by abusing notation, Tor denotes the torsion part of H ∗ (X [n] ; Z).
(ii) A linear basis of H ∗ (X [n] ) is integral if its members are integral classes and
form a Z-basis of the lattice H ∗ (X [n] ; Z)/Tor;
(iii) A linear operator f ∈ End(HX ) is integral if f(A) ∈ HX is an integral class
whenever A ∈ HX is an integral cohomology class.
175
176 9. INTEGRAL COHOMOLOGY OF HILBERT SCHEMES

A linear basis of H ∗ (X [n] ) is integral if and only if its members are integral
classes and the matrix formed by the pairings of its members is unimodular. In the
next two lemmas, we prove that the Heisenberg operators an (α) and the operators
1−n are integral.
Lemma 9.2. (i) If f ∈ End(HX ) is integral, then so is its adjoint f† ;
(ii) The Heisenberg operators an (α), n ∈ Z are integral if α ∈ H ∗ (X) is integral.
Proof. (i) Note that a class A ∈ HX is integral if and only if A, B is an
integer whenever B ∈ HX is an integral class. It follows that f ∈ End(HX ) is
integral if and only if its adjoint operator f† ∈ End(HX ) is integral.
(ii) Recall that a0 (α) = 0. Next, fix n > 0. By Definition 3.3, a−n (α) is
integral. By Lemma 3.7 and (i), an (α) is integral as well. 
Lemma 9.3. For n ≥ 0, the operator 1−n is integral.
Proof. Recall from Definition 3.18 that 1−n = 1/n! · a−1 (1X )n . Fix any
integer m ≥ 0 and an integral class A ∈ H j (X [m] ). For i = m, . . . , m + n − 1,
recall the subscheme Q[i+1,i] of X [i+1] × X × X [i] defined by (3.6), and let Qi be
the image of Q[i+1,i] under the natural projection
X [i+1] × X × X [i] → X [i+1] × X [i] .
Set-theoretically,
Qi = {(ξi+1 , ξi ) ∈ X [i+1] × X [i] | ξi+1 ⊃ ξi }.
Let φi,1 and φi,2 be the two projections of X [i+1] × X [i] . Let
Y = X [m+n] × · · · × X [m+1] × X [m] .
By Definition 3.3,
a−1 (1X )(A) = (φm,1 )∗ ([Qm ] · φ∗m,2 A).
Repeating this process and using the projection formula, we conclude that
a−1 (1X )n (A) = (φ̃m+n )∗ ([Q] · φ̃∗m A)
where φ̃i denotes the projection of Y to X [i] and
Q = {(ξm+n , . . . , ξm+1 , ξm ) ∈ Y | ξm+n ⊃ · · · ⊃ ξm+1 ⊃ ξm }
(the scheme structure on Q can be described similarly as that on Qi ). To show that
1−n (A) ∈ H j (X [m+n] ) is integral, it suffices to prove that the intersection number
(φ̃m+n )∗ [Q] · φ̃∗m A · B
is divisible by n! for any integral class B ∈ H 4(m+n)−j (X [m+n] ).
Represent the integral class A by a piecewise smooth cycle WA ⊂ X [m] . Let
QA = {(ξm+n , . . . , ξm+1 , ξm ) ∈ Y | ξm+n ⊃ · · · ⊃ ξm+1 ⊃ ξm and ξm ∈ WA }.
Then, the integral class [Q] · φ̃∗m A is represented by QA . Note that
φ̃m+n |QA : QA → φ̃m+n (QA )
is generically finite, and a generic element in φ̃m+n (QA ) is of the form ξm + x1 +
. . . + xn where ξm ∈ WA is generic, the points x1 , . . . , xn are distinct, and
Supp(ξm ) ∩ {x1 , . . . , xn } = ∅.
 9.1. INTEGRAL OPERATORS 177

Represent B by a piecewise smooth cycle WB such that WB and φ̃m+n (QA )

intersect transversely at generic points P1 , . . . , Ps . Write Pi as ξm + x1 + . . . + xn .
Let
ξm = η1 + . . . + ηt + x1 + . . . + xu + x1 + . . . + xv
where each ηi is supported at one point with (ηi ) ≥ 2, the following subsets of X:
Supp(η1 ), . . . , Supp(ηt ), {x1 }, . . . , {xu }, {x1 }, . . . , {xv }
are mutually disjoint, and only the points x1 , . . . , xu can move in Zariski open
subsets of the surface X (In other words, if we put
U = X − Supp(η1 ) ∪ . . . ∪ Supp(ηt ) ∪ {x1 , . . . , xu , x1 , . . . , xv },
then for all distinct points x̃1 , . . . , x̃u in U , we have
η1 + . . . + ηt + x̃1 + . . . + x̃u + x1 + . . . + xv ∈ WA ).
−1
So φ̃m+n |QA (Pi ) consists of (u + n)(u + n − 1) · · · (u + 1) distinct points. It
follows that the intersection number (φ̃m+n )∗ ([Q] · φ̃∗m A) · B is divisible by n!. 

Let A ∈ H ∗ (X [n] ). By (3.10), A is a linear combination of classes of the form

a−1 (1X )m a−m1 (α1 ) · · · a−m (α )|0
D
where α1 , . . . , α ∈ i≥1 H i (X), m ≥ 0, ≥ 0, and m1 , . . . , m > 0. By (3.9),

a1 (x) a−m1 (α1 ) · · · a−m (α )|0 = 0.

It follows that the class A ∈ H ∗ (X [n] ) can be written as

(9.1) A = 1−n1 (A1 ) + . . . + 1−nk (Ak )
where 0 ≤ n1 < . . . < nk , and Ai ∈ H ∗ (X [n−ni ] ) with a1 (x)(Ai ) = 0 for every i.

Lemma 9.4. Let A ∈ H ∗ (X [n] ) be expressed as in ( 9.1). Then,

(i) the classes Ai are uniquely determined by A;
(ii) A is integral if and only if all the classes Ai are integral.
Proof. First of all, by Lemma 9.3, A is integral if all the classes Ai are integral.
So it remains to prove (i) and the “only if” part of (ii).
Next, applying the operator a1 (x)nk to both sides of (9.1), we obtain
a1 (x)nk (A) = a1 (x)nk 1−n1 (A1 ) + . . . + 1−nk (Ak ) = (−1)nk · Ak .
Thus the class Ak is uniquely determined by A. Moreover, by Lemma 9.2 (ii), if A
is an integral class, then so is the class Ak .
Finally, repeating the above process to the class A = (A − 1−nk (Ak )), we
conclude that (i) and the “only if” part of (ii) hold for all the classes Ai . 

The next results enables us to construct integral operators.

Proposition 9.5. Let A ∈ HX be an integral class. Write A as aA |0 where
aA is a unique polynomial of creation operators. Then, aA is an integral operator.
178 9. INTEGRAL COHOMOLOGY OF HILBERT SCHEMES

Proof. Let A ∈ H ∗ (X [n] ). Fix an integral class B ∈ H ∗ (X [m] ). We want to

show that the cohomology class aA (B) ∈ H ∗ (X [m+n] ) is still integral.
First of all, decomposing A as in (9.1), we see from Lemma 9.4 (ii) and
Lemma 9.3 that we may assume a1 (x)(A) = 0. Similarly, write

B = 1−m1 (B1 ) + . . . + 1−mk (Bk )

as in (9.1), where 0 ≤ m1 < . . . < mk ≤ m, and Bi ∈ H ∗ (X [m−mi ] ) with

a1 (x)(Bi ) = 0 for every i. By Lemma 9.4 (ii), each class Bi is integral. Now


k 
k
aA (B) = aA 1−mi (Bi ) = 1−mi aA (Bi ) .
i=1 i=1

It follows from Lemma 9.3 that we may assume a1 (x)(B) = 0 as well.

By Lemma 9.3 again, we have two integral classes 1−m A, 1−n B ∈ H ∗ (X [m+n] ).
By Theorem 7.7, the cup product (1−m A) · (1−n B) equals

aA (B) + 1−i (Ai )
i≥1

where a1 (x)(Ai ) = 0 for every i ≥ 1. By our assumption on A and B, we have

a1 (x) aA (B) = 0. Hence aA (B) ∈ H ∗ (X [m+n] ) is integral by Lemma 9.4. 

In view of Proposition 9.5, to obtain integral operators, we need to construct

integral classes. We begin with the Chern classes of some tautological vector bundles
over the Hilbert scheme X [n] . Recall the notation L[n] from (1.25). For a partition
λ = (1m1 2m2 · · · ) and a cohomology class α ∈ H ∗ (X), define

(9.2) a−λ (α) = a−r (α)mr .
r≥1

Lemma 9.6. Let 0 ≤ i ≤ n. Let L be a line bundle on X with α = c1 (L). Then,

 (−1)i−(μ) a−λ (1X )a−μ (α)|0

(9.3) ci (L[n] ) = .
zλ zμ
|λ|+|μ|=n
(λ)=n−i

Proof. Let

n
c (L[n] ) = ci (L[n] )i
i=0

be the generating function for the total Chern classes of L[n] , where  is a formal
variable. By Proposition 4.19,
⎛ ⎞

+∞  (−)r−1
(9.4) c (L[n] )wn = exp ⎝ a−r (c (L))wr ⎠ · |0
n=0
r
r≥1
 9.1. INTEGRAL OPERATORS 179

where w is another formal variable. Expanding the right-hand-side yields

⎛ ⎞
 (−)r−1
(9.5) exp ⎝ a−r (c (L))wr ⎠ · |0
r
r≥1
⎛ ⎞ ⎛ ⎞
 (−1)r−1  (−1)r−1
= exp ⎝ a−r (1X )r−1 wr ⎠ · exp ⎝ a−r (α)r wr ⎠ · |0
r r
r≥1 r≥1
 (−1)|λ|−(λ) a−λ (1X )  (−1)|μ|−(μ) a−μ (α)
= |λ|−(λ) w|λ| · |μ| w|μ| |0
zλ μ
zμ
λ
 (−1)|λ|+|μ|−(λ)−(μ) a−λ (1X )a−μ (α)|0
= |λ|+|μ|−(λ) w|λ|+|μ| .
zλ zμ
(λ,μ)

Now the Chern class ci (L[n] ) is the coefficient of i wn in (9.5). 

Lemma 9.7. For every partition λ, we have

(i) 1/zλ · a−λ (1X )|0 is an integral class;
(ii) 1/zλ · a−λ (1X ) is an integral operator.

Proof. (ii) follows from (i) and Proposition 9.5. To prove (i), we let n = |λ|
and use induction on n. Our result is trivially true when n = 0, 1. In the following,
assume that 1/zμ · a−μ (1X )|0 is integral whenever
 |μ| < n.
Let λ = (1m1 2m2 3m3 · · · ) so that n = r rmr . First of all, assume mr > 0
for at least two different r’s. Then, rmr < n for every r. Putting μr = (r mr )
for r ≥ 1 and applying induction to the partitions μr , we obtain integral classes
1/zμr · a−μr (1X )|0 . Put
1
Ar = · a−μr (1X )|0 .
zμ r
By Proposition 9.5, the operators aAr = 1/zμr · a−μr (1X ) are integral. Thus,

1/zλ · a−λ (1X )|0 = 1/zμr · a−μr (1X ) · |0
r≥1

is integral.
We are left with the case when mr > 0 for a unique r. In this case, |λ| = rmr =
n and (λ) = mr . Applying (9.3) to L = OX and i = n − mr , we have

[n]
 a−ν (1X )|0
(9.6) ci (OX ) = (−1)i ·
zν
|ν|=rmr , (ν)=mr

a−λ (1X )|0  a−ν (1X )|0

= (−1)i · + (−1)i · .
zλ zν
ν=λ,|ν|=rmr
(ν)=mr

Note that for any partition ν = (1t1 2t2 · · · ) satisfying ν = λ, |ν| = rmr and (ν) =
mr , there are at least two i’s with ti > 0. By the previous paragraph, 1/zν ·
a−ν (1X )|0 is integral. Hence, 1/zλ · a−λ (1X )|0 is integral by (9.6). 
180 9. INTEGRAL COHOMOLOGY OF HILBERT SCHEMES

9.2. Integral operators involving only divisors in H 2 (X)

In this section, we will define the linear operator mλ,α ∈ End(HX ) for a parti-
tion λ and an arbitrary class α ∈ H 2 (X). We will show that when α is a divisor,
mλ,α is an integral operator.
To begin with, let C be a smooth irreducible curve in the smooth projective
complex surface X. By abusing notations, we also use C to denote the correspond-
ing divisor and cohomology class. For every partition λ = (λ1 ≥ λ2 ≥ . . .) of n, we
define the subvariety
(9.7) Lλ C = (ρn )−1 (Sλn C)
of X [n] , where
* +

(9.8) Sλn C = λi xi ∈ C (n) | xi ∈ C, xi = xj for i = j ,
i

and ρn is the Hilbert-Chow morphism. By Remark 1.19, Lλ C is irreducible and of

dimension n. For n ≥ 0, let
Hn,C
be the C-linear span of all the classes a−λ (C)|0 , λ  n. By the Theorem 9.14 in
[Nak5], the integral class [Lλ C] ∈ H 2n (X [n] ; Z) is contained in Hn,C ⊂ H ∗ (X [n] );
moreover,

(9.9) a−i (C)[Lλ C] = aλ,μ [Lμ C]
μ

where μ and aλ,μ are the same as defined in (1.5).

Define

(9.10) HC = Hn,C .
n≥0

Next, we recall some notations from Section 1.2. Let Λ be the ring of symmetric
functions in infinitely many variables, and
ΛC = Λ ⊗Z C.
Let Λn and ΛnC be the degree-n parts in Λ and ΛC respectively. For a partition λ,
let pλ , mλ and sλ be the power-sum symmetric function, the monomial symmetric
function and the Schur function respectively.
In [Nak5, (9.13)], Nakajima defined a linear isomorphism

(9.11) ΦC : ΛC = ΛnC → HC
n≥0

which satisfies the following two properties:

(9.12) ΦC (pλ ) = a−λ (C)|0 , ΦC (mλ ) = [Lλ C].
By Proposition 9.5, the operator a[Lλ C] ∈ End(HX ) is integral. Define
mλ,C = a[Lλ C] ∈ End(HX ).
By (9.12), mλ,C is a polynomial of the creation Heisenberg operators a−i (C), i > 0
with rational coefficients.
 9.2. INTEGRAL OPERATORS INVOLVING ONLY DIVISORS IN H 2 (X) 181

Definition 9.8. For a partition λ and an arbitrary class α ∈ H 2 (X), we define

mλ,α ∈ End(HX )
by replacing the creation operators a−i (C) in mλ,C by the corresponding creation
operators a−i (α). Define the cohomology class [Lλ α] ∈ H ∗ X [|λ|] by putting
(9.13) [Lλ α] = mλ,α |0 .
Similarly, we can define the subspaces Hn,α and Hα of HX as in (9.10). More-
over, in view of (9.11) and (9.12), we have a linear isomorphism
(9.14) Φα : ΛC → Hα
which satisfies the following two properties:
(9.15) Φα (pλ ) = a−λ (α)|0 , Φα (mλ ) = [Lλ α] = mλ,α |0 .
Applying Φα to both sides of (1.4) and using (9.15), we obtain
n  (−1)n−(μ) a−μ (α)|0
(9.16) m(1n ),α |0 = [L(1 ) α] = .
zμ
|μ|=n

Our goal in this section is to prove that mλ,α ∈ End(HX ) is an integral operator
if α is a divisor. In view of (9.13) and Proposition 9.5, it suffices to show that [Lλ α]
is an integral class. In the next three lemmas, we will study the properties of the
class [Lλ α].
Lemma 9.9. Let α ∈ H 2 (X) and i > 0. Then, we have

(9.17) a−i (α)[Lλ α] = aλ,μ [Lμ α]
μ

where μ and aλ,μ are the same as defined in (1.5).

Proof. Fix a smooth irreducible curve C in X. Define a linear map
ΨC,α : HC → Hα
by sending the basis elements a−λ (C)|0 of HC to the elements a−λ (α)|0 in Hα .
Note that the creation operators a−j (C) and a−j (α) preserve HC and Hα respec-
tively. Moreover,
ΨC,α ◦ a−j (C) = a−j (α) ◦ ΨC,α .
It follows from the definition of mλ,α that ΨC,α ◦ mλ,C = mλ,α ◦ ΨC,α . Thus
(9.18) ΨC,α [Lλ C] = ΨC,α ◦ mλ,C |0 = mλ,α ◦ ΨC,α |0 = mλ,α |0 = [Lλ α].
Now applying ΨC,α to both sides of (9.9), we obtain (9.17). 
λ λ
Lemma 9.10. If [L α] is integral for every λ, then so is [L (−α)].
Proof. Fix a smooth irreducible curve C in X, and let notations be the same
as in the proof of Lemma 9.9. Recall from (9.12) that the linear isomorphism ΦC
sends mλ and pλ to [Lλ C] and a−λ (C)|0 respectively. Put

(9.19) mλ = dμ pμ
|μ|=|λ|

where dμ ∈ C. Applying ΨC,α ◦ ΦC to both sides and using (9.18), we obtain


[Lλ α] = dμ a−μ (α)|0 .
|μ|=|λ|
182 9. INTEGRAL COHOMOLOGY OF HILBERT SCHEMES

It follows from the definition of [Lλ (−α)] that


(9.20) [Lλ (−α)] = dμ (−1)(μ) a−μ (α)|0 .
|μ|=|λ|

Next, recall from (1.8) the involution ω on Λ and from (1.9) the forgotten sym-
metric function fλ = ω(mλ ) associated to a partition λ. The forgotten symmetric
function fλ is an integral linear combination of the monomial symmetric functions
mμ , μ  |λ|. Thus ΨC,α ◦ ΦC (fλ ) is an integral linear combination of the classes
ΨC,α ◦ ΦC (mλ ) = [Lμ α], μ  |λ|, and hence is an integral class. By (9.19),

(9.21) ΨC,α ◦ ΦC (fλ ) = ΨC,α ◦ ΦC ◦ ω(mλ )

⎛ ⎞

= ΨC,α ◦ ΦC ◦ ω ⎝ dμ pμ ⎠
|μ|=|λ|
⎛ ⎞

= ΨC,α ◦ ΦC ⎝ dμ (−1)|μ|−(μ) pμ ⎠
|μ|=|λ|

= (−1)|λ| · dμ (−1)(μ) a−μ (α)|0 .
|μ|=|λ|

Combining (9.20) and (9.21), we conclude that [Lλ (−α)] is integral as well. 

Lemma 9.11. Let α1 , α2 ∈ H 2 (X), and α = α1 + α2 . Then,


(9.22) [Lλ α] = mλ1 ,α1 mλ2 ,α2 |0
(λ1 ,λ2 ):λ1 ∪λ2 =λ

where (λ1 , λ2 ) stands for ordered pairs of partitions.

Proof. We start with some notations. For the partitions μ obtained from λ
as in (9.17), we will denote μ = λ ↑i . If we specify further that such a μ is obtained
from adding i to a part of λ equal to j (here j is allowed to be 0), then we denote
μ = λ ↑ij . Given a partition λ, we denote by mk (λ) the multiplicity of the parts of
λ equal to k. In these notations, the coefficient aλ,μ in Lemma 9.9 for μ = λ ↑ij is
simply equal to mi+j (μ). Denote the right-hand-side of (9.22) by Rλ (α).

Claim. a−i (α)Rλ (α) = aλ,μ Rμ (α).
μ=λ↑i

Proof. The right-hand-side in the Claim is equal to

 
(9.23) aλ,μ mμ1 ,α1 mμ2 ,α2 |0 .
μ=λ↑i (μ1 ,μ2 ):μ1 ∪μ2 =μ
 9.2. INTEGRAL OPERATORS INVOLVING ONLY DIVISORS IN H 2 (X) 183

On the other hand, by Lemma 9.9, the left-hand-side in the claim is

a−i (α)Rλ (α)

= (mλ2 ,α2 a−i (α1 )mλ1 ,α1 |0 + mλ1 ,α1 a−i (α2 )mλ2 ,α2 |0 )
(λ1 ,λ2 ):λ1 ∪λ2 =λ
 
(9.24) = aλ1 ,ρ1 mρ1 ,α1 mλ2 ,α2 |0
(λ1 ,λ2 ):λ1 ∪λ2 =λ ρ1 =λ1 ↑i
 
(9.25) + aλ2 ,ρ2 mλ1 ,α1 mρ2 ,α2 |0 .
(λ1 ,λ2 ):λ1 ∪λ2 =λ ρ2 =λ2 ↑i

Note that the partitions ρ1 ∪λ2 and λ1 ∪ρ2 associated to λ1 , λ2 , ρ1 and ρ2 appearing
above are of the form λ ↑i . Thus, the same types of terms appear on both sides of
the Claim. It remains to identify the coefficients of a given term.
Fix μ = λ ↑ij for some part j of λ, and fix μ1 , μ2 such that μ1 ∪μ2 = μ. From the
above computation, the contributions to the term mμ1 ,α1 mμ2 ,α2 |0 in the left-hand-
side of the Claim come from two places: the term in (9.24) for ρ1 = μ1 , λ2 = μ2
whose coefficient is mi+j (μ1 ), and the term in (9.25) for λ1 = μ1 , ρ2 = μ2 whose
coefficient is mi+j (μ2 ). Therefore in view of (9.23), the coefficients of the term
mμ1 ,α1 mμ2 ,α2 |0 in both sides of the Claim coincide thanks to
(9.26) mi+j (μ1 ) + mi+j (μ2 ) = mi+j (μ1 ∪ μ2 ) = mi+j (μ) = aλ,μ .
This completes the proof of the above Claim. 

Next, we continue the proof of (9.22) by using induction on n and the reverse
dominance ordering of partitions λ of n (see Definition 1.2 for the dominance or-
dering). For n = 1, formula (9.22) is clear. Assume that formula (9.22) holds for
all partitions of size less than n.
For λ = (n), (9.22) holds since mλ,α = a−n (α) and [Lλ α] = a−n (α)|0 .
 the
For a general partition λ of n with a part equal to, say, i, we denote by λ
partition obtained from λ with a part equal to i removed. Now replacing λ in (9.17)
and the above Claim by λ  respectively, we obtain
 
(9.27) a−i (α)[Lλ α] = aλ,μ μ
 [L α],
 i
μ=λ↑
 
(9.28) a−i (α)Rλ (α) = aλ,μ μ
 R (α).
 i
μ=λ↑

Note that λ appears among the above μ’s as the maximum in the reverse dominance
ordering. By induction hypothesis, the left-hand-sides of (9.27) and (9.28) coincide,
and all the terms on the right-hand-sides of (9.27) and (9.28) involving μ not equal
 = mi (λ) = 0.
to λ coincide. Thus (9.22) follows since aλ,λ 

Theorem 9.12. For every divisor α on X and every partition λ, [Lλ α] is an

integral class and mλ,α ∈ End(HX ) is an integral operator.
Proof. Every divisor α can be written as C1 −C2 for some very ample divisors
C1 and C2 . Represent C1 and C2 by smooth irreducible curves. Then the operators
mμ,Ci are integral. By Lemma 9.10 and Lemma 9.11, [Lλ α] is an integral class. By
Proposition 9.5, mλ,α ∈ End(HX ) is an integral operator. 
184 9. INTEGRAL COHOMOLOGY OF HILBERT SCHEMES

Theorem 9.12 is not sufficient in applications. In the next section, we will

extend it to cover the case of an arbitrary integral class α ∈ H 2 (X).

9.3. Integrality of mλ,α for integral α

In this section, we will prove that if α ∈ H 2 (X) is integral, then the class [Lλ α]
and the operator mλ,α are integral.
We begin with the case when λ = (1n ).
Lemma 9.13. If α ∈ H 2 (X) is integral, then so is the operator m(1n ),α .

Proof. Let p1 and p2 be the projections of X [n] ×X to X [n] and X respectively.

By the Theorem on p.20 in [AH], p1 induces a homomorphism
0
(p1 )! : Ktop (X [n] × X) → Ktop
0
(X [n] )
0
where Ktop (·) denotes the ring of complex vector bundles; moreover, for an element
V ∈ Ktop (X [n] × X), the Riemann-Roch formula holds:
0

(9.29) ch (p1 )! (V ) = (p1 )∗ ch(V ) · p∗2 td(X) .

Next, by the Theorem E.5 in [FU], there exists a complex (not necessarily
algebraic) line bundle Lα on X with c1 (Lα ) = α. Put

Vα = OZn ⊗ p∗2 Lα ∈ Ktop

0
(X [n] × X)
where Zn is the universal codimension-2 subscheme. Applying (9.29) to the element
V = Vα , we obtain
(9.30) ch (p1 )! (Vα ) = p1∗ ch(OZn ) · p∗2 td(X) · p∗2 ch(Lα )
= G(1X , n) + G(α, n) + α, α /2 · G(x, n)
where we have used the definition of G(α, n) and the fact that
ch(Lα ) = 1X + α + α, α /2 · x.
In particular, we conclude that the integral class ci (p1 )! (Vα ) is a polynomial Pi
of Gi1 (1X , n), Gi2 (α, n), α, α and Gi3 (x, n) with 0 ≤ i1 , i2 , i3 ≤ 2n, i.e.,
(9.31) ci (p1 )! (Vα )
" E #
= Pi Gi1 (1X , n), Gi2 (α, n), α, α , Gi3 (x, n)E 0 ≤ i1 , i2 , i3 ≤ 2n .

Note that the coefficients in Pi are independent of α and X.

When α is a divisor, Lα is an algebraic line bundle on X, and the element
(p1 )! (Vα ) ∈ Ktop
0
(X [n] )

is represented precisely by the tautological rank-n vector bundle (Lα )[n] defined in
(1.25). Therefore, we obtain from Lemma 9.6 that
 (−1)i−(μ) a−λ (1X )a−μ (α)|0
(9.32) ci (p1 )! (Vα ) = .
zλ zμ
|λ|+|μ|=n
(λ)=n−i
 9.4. UNIMODULARITY 185

Finally, we conclude from (9.31) and Lemma 4.13 that (9.32) holds for a general
integral class α ∈ H 2 (X). Setting i = n in (9.32) and using (9.16), we obtain the
integral class
 (−1)n−(μ) a−μ (α)|0 n
cn (p1 )! (Vα ) = = [L(1 ) α].
zμ
|μ|=n

By Proposition 9.5 and (9.13), the operator m(1n ),α is integral. 

Theorem 9.14. Let the class α ∈ H 2 (X) be integral. Then for every partition
λ, the class [Lλ α] and the operator mλ,α are integral.
Proof. Recall from Section 1.2 that the monomial symmetric functions
m(1i ) , i≥1
generate the integral ring Λ. So the monomial symmetric function mλ can be
written as an integral polynomial P of the functions m(1i ) , i ≥ 1, i.e.,
E
mλ = P m(1i ) E i ≥ 1 .
Combining this with the two formulas in (9.15), we see that
E E
(9.33) [Lλ α] = Φα (mλ ) = Φα P m(1i ) E i ≥ 1 = P m(1i ),α E i ≥ 1 |0 .
Hence, the class [Lλ α] is integral by Lemma 9.13. Finally, we conclude from Propo-
sition 9.5 and (9.13) that the operator mλ,α is integral. 

9.4. Unimodularity
Now Lemma 9.4, Lemma 9.7 and Theorem 9.14 have provided us with three
types of integral operators:
1
a−λ (1X ), a−μ (x), mν,α
zλ
where α ∈ H 2 (X) is integral. These operators enable us to write down integral
classes for the Hilbert scheme X [n] . To ensure that a certain set of these integral
classes is an integral linear basis, we must show that their intersection matrix is
unimodular. In this section, we will prove that if the intersection matrix of the
cohomology classes α1 , · · · , αk ∈ H 2 (X) has determinant ±1, then so does the
intersection matrix of the following cohomology classes in H 2n (X [n] ):
mλ1 ,α1 · · · mλk ,αk |0 , |λ1 | + . . . + |λk | = n.
First of all, we begin with some linear algebra preparations. Let V be a k-
dimensional complex vector space with a symmetric bilinear form
·, · : V × V → C.
Fix a linear basis v = {v1 , . . . , vk } of V , and let
Mv = vi , vj 1≤i,j≤k

be the matrix formed by the pairings vi , vj . The following is elementary.

Lemma 9.15. Let ṽ = {ṽ1 , . . . , ṽk } be another basis of V . Let T be the transition
matrix from ṽt to vt (i.e., vt = T ṽt ). Then, Mv = T Mṽ T t . 
186 9. INTEGRAL COHOMOLOGY OF HILBERT SCHEMES

For a given n ≥ 0, the symmetric power S n (V ) admits a monomial basis

(9.34) {vi1 · · · vin | 1 ≤ i1 ≤ . . . ≤ in ≤ k}.
Define a bilinear form, still denoted by ·, · , on S n (V ) by letting
 
n
vi1 · · · vin , vj1 · · · vjn = via , vjσ(a)
σ∈Sn a=1

where Sn denotes the n-th symmetric group. Denote by

Mn,v
the matrix of the pairings of the monomial basis elements of S n (V ) (so we have
M1,v = Mv ). The following lemma expresses det Mn,v in terms of det Mv .
Lemma 9.16. For some constant c(n, k), we have
n+k−1
(9.35) det Mn,v = c(n, k) · (det Mv )( k ) .

Proof. First we prove the formula under the assumption that

(9.36) det Mvi = 0 for 1 ≤ i ≤ k − 1
where vi = {v1 , . . . , vi } for 1 ≤ i ≤ k (so vk = v).
Use induction on k. When k = 1, our formula is clearly true.
Fix k > 1. Since det Mvk−1 = 0, there is a vector
ṽk = e1 v1 + . . . + ek−1 vk−1 + vk
such that e1 , . . . , ek−1 ∈ C and v1 , ṽk = . . . = vk−1 , ṽk = 0. Let
ṽ = {v1 , . . . , vk−1 , ṽk }.
Then the transition matrix from the basis ṽt to the basis vt is lower triangular with
all diagonal entries being 1. By Lemma 9.15, we have
(9.37) det Mv = det Mṽ = det Mvk−1 · ṽk , ṽk .
Next, we apply Lemma 9.15 to S n (V ) for the two monomial bases:
i
{v1i1 · · · vk−1 vk | i1 + . . . + ik−1 + ik = n},
k−1 ik

i
{v1i1 · · · vk−1 ṽk | i1 + . . . + ik−1 + ik = n}.
k−1 ik

Since the transition matrix from the second basis to the first basis is lower triangular
with all diagonal entries being 1, we conclude that
(9.38) det Mn,v = det Mn,ṽ .
ik−1 ik j
Now note that v1i1 · · · vk−1 ṽk , v1j1 · · · vk−1
k−1 jk
ṽk is equal to
i j
v1i1 · · · vk−1
k−1
, v1j1 · · · vk−1
k−1
· δik ,jk · ik ! · ṽk , ṽk ik
.
It follows that
Mn,ṽ = diag · · · , Mn−m,vk−1 · m! · ṽk , ṽk m
, ···
n−m+k−2
where m runs from 0 to n. The matrix Mn−m,vk−1 has k−2 rows. So
n "
 #
m·(n−m+k−2 )
det Mn,ṽ = c1 (n, k) · det Mn−m,vk−1 · ṽk , ṽk k−2

m=0
 9.4. UNIMODULARITY 187

for some constant c1 (n, k). Applying induction to det Mn−m,vk−1 yields
(9.39) det Mn,ṽ
n 
 
(n−m+k−2 ) m·(n−m+k−2 )
= c(n, k) · det Mvk−1 k−1
· ṽk , ṽk k−2

m=0
n n
(n−m+k−2 ) m·(n−m+k−2 )
= c(n, k) · det Mvk−1 m=0 k−1
· ṽk , ṽk m=0 k−2

(n+k−1 )
= c(n, k) · det Mvk−1 · ṽk , ṽk k

for some constant c(n, k), where we have used the combinatorial identities:
n    n    
n−m+k−2 n−m+k−2 n+k−1
(9.40) = m· = .
m=0
k−1 m=0
k−2 k

Indeed, all the three terms in (9.40) compute the dimension of the space of degree-
(n − 1) homogeneous polynomials in (k + 1)-variables.
Combining (9.38), (9.39) and (9.37), we conclude that
(n+k−1 )
det Mn,v = c(n, k) · det Mvk−1 · ṽk , ṽk k

(n+k−1 )
= c(n, k) · det Mv k
.
Finally, we come to the general case. Assume that the bilinear form ·, · on V
is not identically zero (otherwise the lemma trivially holds). Set zij = vi , vj for
1 ≤ i ≤ j ≤ k. Then both sides of (9.35) are easily seen to be polynomials in the
variables zij , 1 ≤ i ≤ j ≤ k. Under the assumption (9.36), the proof above implies
that the (polynomial) identity (9.35) holds for a Zariski open subset (zij )1≤i≤j≤k
of Ck(k+1)/2 . Thus, (9.35) holds for an arbitrary (zij )1≤i≤j≤k . 

For r ≥ 1, we denote by V [r] a copy of V with bilinear form

(9.41) ·, · r = (−1)r−1 r · ·, · .
We will denote by ·, · r as well the induced bilinear form on the symmetric power
S ∗ (V [r]). In particular, V [1] = V with ·, · 1 = ·, · .
For a partition μ = (r mr )r≥1 , we form the vector space
C
S μV = S mr (V [r])
r≥1

with a bilinear form ·, · μ given by


⊗r uIr , ⊗r vJr μ = uIr , vJr r
r

for uIr , vJr ∈ S mr (V [r]). Denote by Mμ,v the matrix of the pairings of the induced
monomial basis for S μ V (see (9.34) for the monomial basis of S n (V )).
Lemma 9.17. For some constant c(μ, k) and some integer d(μ, k) ≥ 1, we have
d(μ,k)
det Mμ,v = c(μ, k) · det Mv .
Proof. Follows immediately from our definitions and Lemma 9.16. 
188 9. INTEGRAL COHOMOLOGY OF HILBERT SCHEMES

Now let α1 , · · · , αk ∈ H 2 (X) be linearly independent classes, and let Mα be

the intersection matrix of α1 , · · · , αk . Fix a positive integer n. Let
Mn,α
be the intersection matrix of the classes in H 2n (X [n] ):
(9.42) mλ1 ,α1 · · · mλk ,αk |0 , |λ1 | + . . . + |λk | = n.
Let
V = Cα1 ⊕ · · · ⊕ Cαk
with the pairing induced from the one on H 2 (X). Let Hn,V ⊂ H ∗ (X [n] ) be the
C-linear span of the classes (9.42). Then, Hn,V has two linear bases:
(9.43) mn,α = {mλ1 ,α1 · · · mλk ,αk |0 },
(9.44) an,α = {a−λ1 (α1 ) · · · a−λk (αk )|0 }
where λ1 , . . . , λk are partitions satisfying |λ1 | + · · · + |λk | = n.
Our goal is to compute det Mn,α . To accomplish this, we will first take an
orthogonal basis {β1 , · · · , βk } of V , and calculate det Mn,β in Lemma 9.19 below.
Via the elementary Lemma 9.18, the matrix Mn,α is expressed in terms of Mn,β and
certain transition matrices A and B. Then (det A)2 is determined in Lemma 9.21.
Summarizing all these yields det Mn,α in Theorem 9.22 which is the main result in
this section.
So take an orthogonal basis {β1 , · · · , βk } of V with βi , βi = ±1 for every
i. We similarly define bases mn,β and an,β for Hn,V . Observe that the transition
matrix B from the basis mn,α to an,α is independent of the basis {α1 , · · · , αk } of
H 2 (X), that is, it is the same as the transition matrix from the basis mn,β to an,β .
Denote by A the transition matrix from the basis an,β to an,α . Denote by Mn,β
the intersection matrix of the basis mn,β . Since the transition matrix from mn,β to
mn,α is B −1 AB, we obtain from Lemma 9.15 the following.
Lemma 9.18. Mn,α = (B −1 AB) Mn,β (B −1 AB)t . 
Next, we prove a special case of Theorem 9.22 for the basis {β1 , · · · , βk }.
Lemma 9.19. det Mn,β = ±1.
Proof. As complex vector spaces, we have a linear isomorphism:
C
k
(9.45) Hn,V ∼
= Hni ,Cβi
n i=1

where the sum is over n = (n1 , · · · , nk ) ∈ (Z≥0 )k such that n1 + . . . + nk = n.

Denote by Mni ,βi the matrix of pairings among the basis mλi ,βi |0 | λi  ni for
Hni ,Cβi . For given 1 ≤ i ≤ k and λi , the operator mλi ,βi is a linear combination
of the operators a−μi (βi ), μi  |λi |. Since the classes β1 , · · · , βk are orthogonal, we
see from (3.9) that
 
mλ1 ,β1 · · · mλk ,βk |0 , mμ1 ,β1 · · · mμk ,βk |0

k
 
= δ|λi |,|μi | · mλi ,βi |0 , mμi ,βi |0 .
i=1
 9.4. UNIMODULARITY 189

This orthogonality together with (9.45) implies by standard linear algebra that

k
dim Hnj ,Cβj
det Mn,β = (det Mni ,βi ) 1≤j≤k,j=i .
n i=1

Thus it suffices to prove our lemma for k = 1.

Put β = β1 and V = Cβ. By (9.43) and (9.44), Hn,V has two linear bases:
mn,β = mλ,β |0 | λ  n ,
an,β = a−λ (β)|0 | λ  n .

Since β, β = ±1, we obtain by Lemma 3.7 and (3.9) that

 
(9.46) a−λ (β)|0 , a−μ (β)|0 = ±δλ,μ · zλ · β, β (λ) = ±δλ,μ · zλ .

Recall that the transition matrix from the basis an,β to mn,β is B −1 . So we conclude
from Lemma 9.15 and (9.46) that
Mn,β = B −1 · diag(· · · , ±zλ , · · · ) · (B −1 )t
where λ runs over all partitions of n. Hence we have

(9.47) det Mn,β = ±(det B)−2 · zλ .
λn

By (9.12), Lemma 9.9, and (1.5), we see that the transition matrix from the
basis {pλ | λ  n} of ΛnC ⊂ ΛC to the basis {mλ | λ  n} is B −1 as well. Introduce
the standard bilinear form ·, · on ΛnC by letting
(9.48) pλ , pμ = δλ,μ · zλ .
The matrix Mn formed by the pairings of the monomial symmetric functions
mλ , λ  n is unimodular, and thus det Mn = ±1. By Lemma 9.15 and (9.48),
Mn = B −1 · diag(· · · , zλ , · · · ) · (B −1 )t
where λ runs over all partitions of n. It follows immediately that

(9.49) (det B)−2 · zλ = det Mn = ±1.
λn

Combining this with (9.47), we finally obtain det Mn,β = ±1. 

For μ  n, let Hμ,α ⊂ Hn,V be the span of a−μ1 (α1 ) · · · a−μk (αk )|0 where
μ , . . . , μk are partitions such that μ1 ∪ · · · ∪ μk = μ. Note that
1

Hμ,α ⊂ Hn,V ⊂ H 2n (X [n] ).

So Hμ,α carries a pairing induced from the one on H 2n (X [n] ).

Lemma 9.20. For every partition μ, we have an isometry Hμ,α ∼
= S μ (V ).
Proof. Follows from the definitions, (9.41), and (3.9). 

Lemma 9.21. If det Mα = ±1, then (det A)2 = ±1.

190 9. INTEGRAL COHOMOLOGY OF HILBERT SCHEMES

Proof. By (3.9), a−μ1 (∗) · · · a−μk (∗) and a−ν 1 (∗) · · · a−ν  (∗)|0 are orthogonal
unless μ1 ∪ · · · ∪ μk = ν 1 ∪ · · · ∪ ν  . Here, ∗’s denote unspecified classes in H 2 (X).
Thus, we have an orthogonal direct sum

Hn,V = Hμ,α .
μn

By Lemma 9.20, the intersection matrix Mn,α of the basis an,α for Hn,V is given
by the diagonal block matrix whose diagonal consists of Mμ,α , μ  n. So

(9.50) n,α =
det M det Mμ,α .
μn

Similarly, by repeating the above with the α’s replaced by the β’s, we obtain

(9.51) n,β =
det M det Mμ,β .
μn

By assumption, det Mα = ±1 and det Mβ = ±1. Thus by Lemma 9.17,

det Mμ,α = ± det Mμ,β for every μ. Therefore, we see from (9.50) and (9.51) that

(9.52) det M n,β .

n,α = ± det M

n,α = A M
By Lemma 9.15 and the definition of the matrix A, we have M n,β At .
Combining this with (9.52) yields (det A)2 = ±1. 

Theorem 9.22. Let α1 , · · · , αk ∈ H 2 (X) be linearly independent classes, and

let Mα be the intersection matrix of α1 , · · · , αk . Fix a positive integer n. Let Mn,α
be the intersection matrix of the classes in H 2n (X [n] ):
mλ1 ,α1 · · · mλk ,αk |0 , |λ1 | + . . . + |λk | = n.
If det Mα = ±1, then we have det Mn,α = ±1 as well.
Proof. Follows immediately from Lemmas 9.18, 9.19, and 9.21. 

9.5. Integral bases for the cohomology of Hilbert schemes

The following theorem is the main result in this chapter.
Theorem 9.23. Let X be a smooth projective surface with vanishing odd Betti
numbers. Let α1 , · · · , αk be an integral basis for H 2 (X; Z)/Tor. Then the classes

1 
k
(9.53) a−λ (1X )a−μ (x)mν 1 ,α1 · · · mν k ,αk |0 , |λ| + |μ| + |ν i | = n
zλ i=1
∗
are integral, and furthermore, they form an integral basis for H (X [n] ; Z)/Tor.
Proof. By Lemma 9.4, Lemma 9.7 and Theorem 9.14, we have three types of
integral operators:
1
a−λ (1X ), a−μ (x), mνi ,αi
zλ
where λ, μ, νi stand for partitions. Therefore, we obtain the integral classes (9.53).
Since H 1 (X) = H 3 (X) = 0, we see from (3.10) that the number of classes in (9.53)
 9.6. COMPARISON OF TWO INTEGRAL BASES OF H ∗ ((P2 )[n] ; Z) 191

is equal to the rank of H ∗ (X [n] ; Z)/Tor. Now consider the intersection number of
two classes from (9.53):
1
a−λ (1X )a−μ (x)mν 1 ,α1 · · · mν k ,αk |0 ,
zλ
1
a (1X )a−μ̃ (x)mν̃ 1 ,α1 · · · mν̃ k ,αk |0 .
zλ̃ −λ̃
By the Heisenberg algebra commutation relation (3.9), the intersection number is
 
±δλ,μ̃ · δλ̃,μ · mν 1 ,α1 · · · mν k ,αk |0 , mν̃ 1 ,α1 · · · mν̃ k ,αk |0 .
By Theorem 9.22, the intersection matrix formed by the pairings of the integral
classes in (9.53) is unimodular. Therefore, the classes in (9.53) form an integral
basis of H ∗ (X [n] ; Z)/Tor. 
Remark 9.24. Under the same assumptions as in Theorem 9.23, one can fur-
ther prove that the integral ring H ∗ (X [n] ; Z)/Tor is generated over Z by all the
integral classes:
[n]
ci OX , 1−(n−j) m(1j ),αs |0 , 1−(n−j) a−j (x)|0
where 1 ≤ i ≤ n − 1, 1 ≤ j ≤ n, and 1 ≤ s ≤ k. We refer to [LQ3, Section 4] for
the details.
Remark 9.25. When X is a K3 surface, the ring generators for the integral
cohomology ring of the Hilbert scheme X [n] are also obtained by Markman [Mar2].

9.6. Comparison of two integral bases of H ∗ ((P2 )[n] ; Z)

In this section, as an application of Theorem 9.23, we will compare two integral
bases of the integral cohomology of the Hilbert scheme (P2 )[n] of points on the
complex projective plane P2 . One integral basis comes from the torus action on
(P2 )[n] , and the other one comes from Theorem 9.23. We will prove that when the
basis elements are arranged properly, the transition matrix is a triangular matrix.
First of all, we recall the integral basis of H ∗ ((P2 )[n] ; Z) from [ES1, Sections 1
and 2]. Let
T0 , T1 , T2
be a system of homogeneous coordinates for P2 . Let
H = {T2 = 0} ⊂ P2
be a coordinator line. Let G ⊂ SL3 (C) be the maximal torus consisting of all
diagonal matrices g = diag(g0 , g1 , g2 ) with g0 , g1 , g2 ∈ C∗ and g0 g1 g2 = 1. The
group G acts on P2 via
g(Ti ) = gi Ti , i = 0, 1, 2.
On points (a0 , a1 , a2 ) ∈ P2 , this action is given by
(9.54) g(a0 , a1 , a2 ) = (g0−1 a0 , g1−1 a1 , g2−1 a2 ).
The fixed points are
P0 = (1, 0, 0), P1 = (0, 1, 0), P2 = (0, 0, 1).
Let
F0 = {P0 }, F1 = H − {P0 }, F2 = P2 − H.
192 9. INTEGRAL COHOMOLOGY OF HILBERT SCHEMES

Then, F0 , F1 , F2 define a cellular decomposition of P2 . The one-parameter sub-

groups
(9.55) φ : C∗ → G
inducing this cellular decomposition are those of the type
(9.56) φ(t) = diag(tw0 , tw1 , tw2 )
with w0 < w1 < w2 and w0 + w1 + w2 = 0.
The fixed points of the G-action on (P2 )[n] are in one-to-one correspondence
with ordered triples (λ, μ, ν) of partitions satisfying |λ| + |μ| + |ν| = n. More
precisely, the fixed points are of the form
ξλ + ξμ + ξν
where ξλ ∈ (P2 )[|λ|] , ξμ ∈ (P2 )[|μ|] , ξν ∈ (P2 )[|ν|] are fixed points supported at
P0 , P1 , P2 respectively. Define
(9.57) ξλ,μ,ν = ξλ + ξμ + ξν .
If u and v are the affine coordinates at P0 and if λ = (λ1 ≥ λ2 ≥ · · · ≥ λk ), then
the 0-dimensional closed subscheme ξλ is defined by the following ideal in C[u, v]:
 
(9.58) Iξλ = v λ1 , uv λ2 , . . . , uk−1 v λk , uk .
Take a generic one-parameter subgroup φ in (9.55). Then, the fixed points of
φ are the same as those of G, and (P2 )[n] has a cellular decomposition with cells
 
(9.59) Cλ,μ,ν = ξ ∈ (P2 )[n] | lim φ(t)ξ = ξλ,μ,ν
t→0

where |λ| + |μ| + |ν| = n. It follows that the integral cohomology H ∗ ((P2 )[n] ; Z) is
a Z-module freely generated by the fundamental classes of the closures:
(9.60) [C λ,μ,ν ], |λ| + |μ| + |ν| = n.
Moreover, it is known that
(9.61) dim(C0,μ,0 ) = |μ|;
if (w2 − w0 )  (w2 − w1 ), then
(9.62) dim(C0,0,ν ) = |ν| + (ν);
if (w2 − w0 )  (w1 − w0 ), then
(9.63) dim(Cλ,0,0 ) = |λ| − (λ).
Since the integral cohomology H ∗ ((P2 )[n] ; Z) is torsion free, we conclude from
Theorem 9.23 that another integral basis of H ∗ ((P2 )[n] ; Z) consists of the classes
1
(9.64) a−λ (1)mμ,H a−ν (x)|0 , |λ| + |μ| + |ν| = n
zλ
where 1 denotes 1P2 and x ∈ H 4 (P2 ; Z) is the class corresponding to a point in P2 .
We will study the transition matrix between the two integral bases (9.60) and
(9.64) of H ∗ ((P2 )[n] ; Z). To begin with, we will express the three special types of
classes [C 0,μ,0 ], [C 0,0,ν ] and [C λ,0,0 ] in terms of the basis (9.64) . For the cohomology
classes [C 0,μ,0 ], we put
T0 T2
u= , v= .
T1 T1
Then u, v form a system of affine coordinates at P1 .
 9.6. COMPARISON OF TWO INTEGRAL BASES OF H ∗ ((P2 )[n] ; Z) 193

Proposition 9.26. [C 0,μ,0 ] = mμ,H |0 for every partition μ.

Proof. Let n = |μ|. By the definition of mμ,H , we have mμ,H |0 = [Lμ H]. By
(9.61), we have
dim(C0,μ,0 ) = |μ| = n.
Now both Lμ H and C 0,μ,0 are irreducible and have the same dimension. It suffices
to show that generic elements ξ ∈ Lμ H are contained in the cell C0,μ,0 , i.e.,
(9.65) lim φ(t)ξ = ξ0,μ,0 .
t→0

Assume that μ = (μ1 ≥ μ2 ≥ . . . ≥ μk ). The fixed point ξ0,μ,0 , regarded as a

point in (C2 )[n] , is defined by the following ideal in C[u, v]:
(9.66) Iξμ = v μ1 , uv μ2 , . . . , uk−1 v μk , uk .
In view of (9.7), a generic element ξ ∈ Lμ H is defined by an ideal of the form
k @
 A
(u − a0 ) + b1 v + . . . + bμi −1 v μi −1 , v μi
(i) (i) (i)
I =
i=1
(1) (k) (i) (i)
where are distinct points in C, and b1 , . . . , bμi −1 ∈ C for every i. For
a0 , . . . , a0
each j with 0 ≤ j < k, the ideal I contains the polynomials:
⎡ ⎤
j 
s−1 k
⎣ (u − a(i)0 )·
(i)
(u − a0 ) · (u − a0 ) + fs (u, v) v ⎦ v μs
(i)

i=1 i=j+1 i=s+1

where s = j + 1, . . . , k, and fs (u, v) denotes polynomials of u and v. Since μj+1 ≥

. . . ≥ μk , we see that I contains the polynomials:
⎡ ⎤

j 
s−1 k
(9.67) ⎣ (u − a0 ) · (u − a0 ) + fs (u, v) v ⎦ v μj+1
(i) (i) (i)
(u − a0 ) ·
i=1 i=j+1 i=s+1

where s = j + 1, . . . , k. The (k − j) degree-(k − j − 1) polynomials


s−1
(i)

k
(i)
(u − a0 ) · (u − a0 ), s = j + 1, . . . , k
i=j+1 i=s+1

are linearly independent. Hence a certain linear combination of them is equal to 1.

It follows from (9.67) that there exists a polynomial f (u, v) of u and v such that
- j .
 (i)
(9.68) (u − a0 ) + f (u, v) v v μj+1 ∈ I.
i=1

Note from (9.56) that φ(t)u = tw0 −w1 u with w1 − w0 > 0, and φ(t)v = tw2 −w1 v
with w2 − w1 > 0. Choosing (w2 − w1 )  (w1 − w0 ), we see from (9.68) that
uj v μj+1 ∈ lim φ(t)I
t→0

for with 0 ≤ j < k. Similar, we have u ∈ limt→0 φ(t)I. Since μ1 +. . .+μk = |μ| = n
k

and the flat limit limt→0 φ(t)I is of colength-n in C[u, v], we must have
lim φ(t)I = v μ1 , uv μ2 , . . . , uk−1 v μk , uk .
t→0

In view of (9.66), we obtain limt→0 φ(t)I = Iξμ . This proves (9.65). 

194 9. INTEGRAL COHOMOLOGY OF HILBERT SCHEMES

Next, to handle the classes [C 0,0,ν ], we let

T1 T0
u= , v=
T2 T2
(note that the proof of the Proposition 4.1 in [ES1] took u = T0 /T2 and v = T1 /T2 ).
Then u and v form a system of affine coordinates at P2 . Let (w2 −w0 )  (w2 −w1 ).
The following lemma determines generic elements in the cell C0,0,ν .
Lemma 9.27. Let ν = (ν1 ≥ . . . ≥ νk )  n. Then generic elements in C0,0,ν
are:

ν1 
νk
(9.69) (a1 , b1,i ) + . . . + (ak , bk,i ) ∈ (C2 )[n] ⊂ (P2 )[n]
i=1 i=1

where a1 , b1,1 . . . , b1,ν1 , . . . , ak , bk,1 . . . , bk,νk ∈ C are distinct.

Proof. By (9.62),
dim(C0,0,ν ) = |ν| + (ν) = n + k.
So it suffices to show that every element (9.69) is contained in the cell C0,0,ν , i.e.,
(9.70) lim φ(t)I = v ν1 , uv ν2 , . . . , uk−1 v νk , uk
t→0

where I denotes the ideal in C[u, v] defining the element (9.69). Note that
I
ν1 
νk
= u − a1 , v − b1,i · · · u − ak , v − bk,i
i=1 i=1
   
= u − a1 , (v − b1,1 ) · · · (v − b1,ν1 ) · · · u − ak , (v − bk,1 ) · · · (v − bk,νk ) .
Since (u − a1 ) · · · (u − ak ) ∈ I, we get uk ∈ limt→0 φ(t)I. Next, fix j with 0 ≤ j ≤
(k − 1). Then the ideal I contains the (k − j) polynomials:

j 
νs 
(u − ai ) · (v − bs,i ) · (u − ai )
i=1 i=1 j+1≤i≤k, i=s

where j + 1 ≤ s ≤ k. Since νj+1 ≥ · · · ≥ νk , we see that there exist polynomials

fs (v), j + 1 ≤ s ≤ k, in v of degree less than νj+1 such that

j 
(9.71) (u − ai ) · (v νj+1 + fs (v)) · (u − ai ) ∈ I
i=1 j+1≤i≤k, i=s

where j + 1 ≤ s ≤ k. Since aj+1 , . . . , ak ∈ C are distinct, a certain linear combina-

tion of the (k − j) degree-(k − j − 1) polynomials

(u − ai ), j+1≤s≤k
j+1≤i≤k, i=s

is equal to 1. It follows from (9.71) that there exists a polynomial f (u, v) with

j
(9.72) (u − ai ) · (v νj+1 + f (u, v)) ∈ I
i=1
 9.6. COMPARISON OF TWO INTEGRAL BASES OF H ∗ ((P2 )[n] ; Z) 195

and with the degree of f (u, v) in v being less than νj+1 . Since (w2 −w0 )  (w2 −w1 ),
we conclude that uj v νj+1 ∈ limt→0 φ(t)I for 0 ≤ j ≤ (k − 1). In summary,
v ν1 , uv ν2 , . . . , uk−1 v νk , uk ⊂ lim φ(t)I.
t→0
ν1 ν2 k−1 νk k
Since both the ideals v , uv , . . . , u v , u and limt→0 φ(t)I have co-length
n in the polynomial ring C[u, v], we see that (9.70) holds. 

The next lemma studies the intersection between the integral classes a−λ (x)|0
and [C 0,0,ν ] where λ, ν  n.
Lemma 9.28. (i) Let λ, ν  n. If the intersection pairing between the cohomol-
ogy classes a−λ (x)|0 and [C 0,0,ν ] is nonzero, then λ = ν;
(ii) The intersection pairing between a−ν (x)|0 and [C 0,0,ν ] is positive.
Proof. (i) Let λ = (λ1 ≥ . . . ≥ λ ) and ν = (ν1 ≥ . . . ≥ νk ). Let
a1 , b1 , . . . , a , b ∈ C be different numbers. Put
x1 = (a1 , b1 ), . . . , x = (a , b ) ∈ C2 .
Then, the points x1 , . . . , x are distinct. Now a−λ (x)|0 is represented by the sub-
variety
Mλ1 (x1 ) × · · · × Mλ (x )
whose dimension is equal to (n − ). Since dim(C0,0,ν ) = n + k, we must have
= k. Since the pairing between a−λ (x)|0 and [C 0,0,ν ] is nonzero, the intersection
between Mλ1 (x1 ) × · · · × Mλ (x ) and C 0,0,ν is nonempty. Let ξ be a point in
the intersection. Then in view of (9.69), ξ = ξ1 + . . . + ξk where for each i with
1 ≤ i ≤ k,
ξi ∈ Mi (xi ) and λi = (ξi ) = νji
for some ji such that {j1 , . . . , jk } = {1, . . . , k}. Therefore, λ = ν.
(ii) Let notations be the same as in the previous paragraph. Then the only
point in the intersection between Mν1 (x1 ) × · · · × Mνk (xk ) and C 0,0,ν is
ξ = ξ1 + . . . + ξk
where each element ξi ∈ Mνi (xi ) is defined by the ideal u − ai , (v − bi )νi in the
polynomial ring C[u, v]. In particular, a−ν (x)|0 , [C 0,0,ν ] > 0. 

Definition 9.29. For n ≥ 1, define I [n] to be the subset of H ∗ ((P2 )[n] ; Z)

consisting of the integral linear spans of the basis classes:
1
a−λ (1)mμ,H a−ν (x)|0
zλ
where |λ| + |μ| + |ν| = n, and either |μ| > 0 or |ν| > 0.
As in Lemma 8.6 (i), I [n] is an ideal of the cohomology ring H ∗ ((P2 )[n] ; Z). In
fact,
I [n] = (I0 )[n] ∩ H ∗ ((P2 )[n] ; Z)
where I0 = H 2 (P2 )⊕H 4 (P2 ) and (I0 )[n] is from Definition 8.5. Lemma 9.28 enables
us to determine [C 0,0,ν ] modulo the ideal I [n] .
196 9. INTEGRAL COHOMOLOGY OF HILBERT SCHEMES

Proposition 9.30. For every partition ν, we have

f˜ν
(9.73) [C 0,0,ν ] ≡ (−1)|ν|−(ν) a−ν (1)|0 (mod I [|ν|] )
zν
where f˜ν is a positive integer.
Proof. Let n = |ν|. By (9.64), [C 0,0,ν ] is an integral linear combination of the
classes:
1
a−λ (1)mμ,H a−ν̃ (x)|0 , |λ| + |μ| + |ν̃| = n.
zλ
By Lemma 9.28 (i), the coefficient of 1/zλ a−λ (1)|0 is 0 if λ = ν. In view of
Lemma 9.28 (ii), the coefficient of 1/zν a−ν (1)|0 is (−1)|ν|−(ν) f˜ν for some positive
integer f˜ν . 

Next, we determine the class [C 0,0,ν ] when ν = (1n ) or (1n−2 2).

Lemma 9.31. (i) If ν = (1n ), then
1
[C 0,0,ν ] = · a−ν (1)|0 = 1(P2 )[n] ;
zν
(ii) If ν = (1n−2 2) where n ≥ 2, then we have
1 1
(9.74) [C 0,0,ν ] = (n − 1) · a−(1n−1 ) (1) m(1),H |0 − a−ν (1)|0 .
z(1n−1 ) zν
Proof. (i) Recall that 1(P2 )[n] = 1/n! · a−1 (1)n |0 . By (9.62), the dimension
of the cell C0,0,ν is 2n. So C 0,0,ν = (P2 )[n] and [C 0,0,ν ] = 1(P2 )[n] .
(ii) A basis of H 2 ((P2 )[n] ; Z) consists of the two classes:
1 1
a n−1 (1) m(1),H |0 , a−ν (1)|0 .
z(1n−1 ) −(1 ) zν

So the divisor [C 0,0,ν ] is an integral linear combination of these two classes. Using
the proof of Proposition 9.30 (ii), we see that f˜ν = 1. So it remains to show that
 
(9.75) [C 0,0,ν ], a−1 (H)a−1 (x)n−1 |0 = (n − 1).
Fix a line H  in P2 such that H  , viewed as a line in C2 , is not parallel to the
v-axis. Choose distinct numbers a1 , b1 , . . . , an−1 , bn−1 ∈ C such that none of the
(n − 1) points xi = (ai , bi ), 1 ≤ i ≤ (n − 1) in C2 is contained in the line H  . Then
the cohomology class a−1 (H)a−1 (x)n−1 |0 is represented by the curve:
H  + x1 + . . . + xn−1 ⊂ (P2 )[n] .
For 1 ≤ i ≤ (n − 1), let Hi be the line in P2 defined by u = ai when Hi is viewed
as a line in C2 . Using the description (9.69) of generic elements in C 0,0,ν , we see
that the intersection C 0,0,ν ∩ (H  + x1 + . . . + xn−1 ) consists of (n − 1) elements:
(H  ∩ Hi ) + x1 + . . . + xn−1 , 1 ≤ i ≤ (n − 1).
Moreover, the intersection is transversal at these points. Therefore, the pairing
between the classes [C 0,0,ν ] and a−1 (H)a−1 (x)n−1 |0 is equal to (n − 1). 
 9.6. COMPARISON OF TWO INTEGRAL BASES OF H ∗ ((P2 )[n] ; Z) 197

To study the classes [C λ,0,0 ], a system of affine coordinates at P0 is chosen to

be
T1 T2
u= , v= .
T0 T0
Let (w2 − w0 )  (w1 − w0 ). Since the closure C λ,0,0 is supported at one point, we
conclude that [C λ,0,0 ] is a linear combination of the integral classes
a−ν (x)|0
where ν denotes partitions with ν  n and (ν) = (λ). Moreover, in view of (9.64),
all the coefficients in the linear combination are integers. Hence

(9.76) [C λ,0,0 ] = eλν a−ν (x)|0
|λ|=|ν|, (λ)=(ν)

where ∈ Z. Recall from Definition 1.1 the total ordering ≺ on the set of all
eλν
partitions.
Proposition 9.32. For every partition λ, there exist integers eλν ∈ Z for par-
titions ν with |λ| = |ν|, (λ) = (ν), λ ≺ ν such that

(9.77) [C λ,0,0 ] = eλλ a−λ (x)|0 + eλν a−ν (x)|0 .
|λ|=|ν|, (λ)=(ν)
λ≺ν

Moreover, eλλ is a positive integer.

The proof of Proposition 9.32 is straightforward but extremely lengthy, and
follows from similar arguments as in the proofs of Lemma 9.27 and Proposition 9.30.
We refer to [Mur, Subsection 3.3] for details.
Now we compare the two integral bases (9.60) and (9.64) of H ∗ ((P2 )[n] ; Z). Let
n = |λ| + |μ| + |ν|. Then there exist eλ,μ,ν λ ,μ ,ν  ∈ Z such that
 λ,μ,ν 1
(9.78) [C λ,μ,ν ] = eλ ,μ ,ν  a−ν  (1)mμ ,H a−λ (x)|0
  
zν 
λ ,μ ,ν
  
where |λ | + |μ | + |ν | = n. Assume that
(9.79) eλ,μ,ν
λ ,μ ,ν  = 0

for some λ , μ , ν  .
The cell Cλ,μ,ν ⊂ (P2 )[n] consists of elements of the form
(9.80) ξ1 + ξ2 + ξ3
with ξ1 ∈ Cλ,0,0 , ξ2 ∈ C0,μ,0 and ξ3 ∈ C0,0,ν . In particular, Supp(ξ1 ) = F0 = {P0 },
Supp(ξ2 ) ⊂ F1 = H − {P0 }, and Supp(ξ2 ) ⊂ F2 = P2 − H.
Choose a partition μ̃ with |μ̃| = |μ |. Let H̃ ⊂ P2 be a line different from H.
Fix distinct points x1 , . . . , x(ν  ) ∈ P2 not lying on the lines H and H̃. Let
Zλ ,μ̃,ν  ⊂ (P2 )[n]
be the subset consisting of the elements of the form
(λ ) (ν  )
 
(9.81) η1,i + η2 + η3,i
i=1 i=1
198 9. INTEGRAL COHOMOLOGY OF HILBERT SCHEMES

where η3,i ∈ Mνi (xi ), η2 ∈ Lμ̃ H̃, and η1,i ∈ Mλi (yi ) such that y1 , . . . , y(λ ) ∈ P2
are distinct and not in the subset {x1 , . . . , x(ν  ) } ∪ H̃. Let Z λ ,μ̃,ν  be the closure
 
of Zλ ,μ̃,ν  . Let λ = (1m1 (λ ) 2m2 (λ ) · · · ). As in Proposition 3.16, we have
(λ )
1  
(9.82) [Z λ ,μ̃,ν  ] = imi (λ ) ·
a−λ (1)mμ̃,H̃ a−ν  (x)|0 .
i=1
zλ 
 
By (9.78) and (9.82), the pairing [C λ,μ,ν ], [Z λ ,μ̃,ν  ] is equal to
(λ )
    
|−(ν  ))
(9.83) imi (λ ) · (−1)(|λ |−(λ ))+(|ν
i=1
< =

· eλ,μ,ν
λ ,μ ,ν  mμ ,H |0 , mμ̃,H̃ |0 .
μ

The next lemma determines the leading coefficient in the expression (9.78). The
main idea in its proof is to exploit the non-emptiness of the intersecton C λ,μ,ν ∩
Z λ ,μ̃,ν  for a suitable μ̃, whenever eλ,μ,ν
λ ,μ ,ν  = 0.

Lemma 9.33. Let λ, μ, ν be three partitions. Then, [C λ,μ,ν ] is equal to

(−1)|ν|−(ν) eλλ f˜ν
a−ν (1)mμ,H a−λ (x)|0
zν
 eλ,μ,ν
λ ,μ ,ν 
+ a−ν  (1)mμ ,H a−λ (x)|0
  
zν 
λ ,μ ,ν

where eλ,μ,ν∈ Z, |λ | + |μ | + |ν  | = |λ| + |μ| + |ν|, and either |ν  | < |ν|, or ν  = ν
λ ,μ ,ν 
and |μ | < |μ|, or ν  = ν, μ = μ, (λ ) = (λ) and λ ≺ λ .


Proof. Recall the expression (9.78) and other notations introduced in the
  
previous paragraphs. As in (9.79), assume eλ,μ,ν λ ,μ ,ν  = 0 for some λ , μ , ν . By

Theorem 9.22, there exists a partition μ̃ such that |μ̃| = |μ | and
< =
 λ,μ,ν
(9.84) eλ ,μ ,ν  mμ ,H |0 , mμ̃,H̃ |0 = 0.
μ
 
By (9.83), [C λ,μ,ν ], [Z λ ,μ̃,ν  ] = 0. So C λ,μ,ν ∩ Z λ ,μ̃,ν  = ∅. Let
ξ ∈ C λ,μ,ν ∩ Z λ ,μ̃,ν  .
Since {x1 , . . . , x(ν  ) } ∩ H = ∅, we see from (9.80) and (9.81) that
ξ = ξ  + ξ3
where Supp(ξ  ) ⊂ H, Supp(ξ3 ) ∩ H = ∅, (ξ  ) ≥ |λ| + |μ|, and |ν| ≥ (ξ3 ) ≥ |ν  |. If
|ν| > |ν  |, we are done. In the following, we assume |ν| = |ν  |.
Since |ν| = |ν  |, we obtain (ξ3 ) = |ν| = |ν  |, and (ξ  ) = |λ| + |μ|. Moreover,
(ν  )

ξ3 ∈ C 0,0,ν ∩ Mνi (xi ).
i=1
 9.6. COMPARISON OF TWO INTEGRAL BASES OF H ∗ ((P2 )[n] ; Z) 199

From the proof of Lemma 9.28 (i), we see that ν = ν  . Also, by the proof of
(ν)
Lemma 9.28 (ii), C 0,0,ν and i=1 Mνi (xi ) intersect at a unique point which will be
denoted by ξ3 for simplicity. Now, ξ ∈ C λ,μ,ν ∩ Z λ ,μ̃,ν if and only if

ξ = ξ  + ξ3

with ξ  ∈ C λ,μ,0 ∩ Z λ ,μ̃,0 . This decomposition allows us to split (P2 )[n] locally into
the product (P2 )[|λ|+|μ|] × (P2 )[|ν|] , i.e., we obtain
     
(9.85) [C λ,μ,ν ], [Z λ ,μ̃,ν ] = [C λ,μ,0 ], [Z λ ,μ̃,0 ] · [C 0,0,ν ], a−ν (x)|0 .
 
By Proposition 9.30, [C 0,0,ν ], a−ν (x)|0 = f˜ν . Therefore,
   
(9.86) [C λ,μ,ν ], [Z λ ,μ̃,ν ] = f˜ν [C λ,μ,0 ], [Z λ ,μ̃,0 ] .

Note that (9.86) is independent of the nonvanishing condition (9.84), i.e., it holds
as long as |μ̃| = |μ |. Combining (9.86) and (9.83), we see that for every partition
μ̃ with |μ̃| = |μ |, the number f˜ν [C λ,μ,0 ], [Z λ ,μ̃,0 ] is equal to
(λ )
   
(9.87) imi (λ ) · (−1)(|λ |−(λ ))+(|ν|−(ν))
i=1
< =

· eλ,μ,ν
λ ,μ ,ν  mμ ,H |0 , mμ̃,H̃ |0 .
μ
 
In the rest of the proof, we work with [C λ,μ,0 ], [Z λ ,μ̃,0 ] , i.e., both ν and ν 
are empty partitions. Now (9.80) and (9.81) are simplified to:
(λ )

ξ1 + ξ2 , η1,i + η2
i=1

where ξ1 ∈ Cλ,0,0 , ξ2 ∈ C0,μ,0 , η2 ∈ Lμ̃ H̃, and η1,i ∈ Mλi (yi ) such that the points
y1 , . . . , y(λ ) ∈ P2 are distinct and not on the line H̃. Assuming (9.84), we see from
(9.87) that
 
[C λ,μ,0 ], [Z λ ,μ̃,0 ] = 0.

So C λ,μ,0 ∩ Z λ ,μ̃,0 = ∅. Since P0 ∈ H̃, ξ ∈ C λ,μ,0 ∩ Z λ ,μ̃,0 is of the form:

ξ = ξ1 + ξ2

where Supp(ξ1 ) = {P0 }, P0 ∈ Supp(ξ2 ), (ξ1 ) ≥ |λ|, and |μ| ≥ (ξ2 ) ≥ |μ̃|. Since
|μ̃| = |μ |, we get |μ| ≥ |μ |. If |μ| > |μ |, we are done.
Assume that |μ| = |μ |. Then, we have |μ| = |μ̃| = (ξ2 ) and |λ| = |λ | = (ξ1 ).
Moreover, we see that ξ ∈ C λ,μ,0 ∩ Z λ ,μ̃,0 if and only if

ξ = ξ1 + ξ2

where ξ1 ∈ C λ,0,0 ∩ Z λ ,0,0 and ξ2 ∈ C 0,μ,0 ∩ Lμ̃ H̃. As in (9.85), we obtain
     
(9.88) [C λ,μ,0 ], [Z λ ,μ̃,0 ] = [C λ,0,0 ], [Z λ ,0,0 ] · [C 0,μ,0 ], [Lμ̃ H̃]
200 9. INTEGRAL COHOMOLOGY OF HILBERT SCHEMES

for every μ̃ with |μ̃| = |μ|. Note that [C 0,μ,0 ] = mμ,H |0 by Proposition 9.26, and
[Lμ̃ H̃] = mμ̃,H̃ |0 = mμ̃,H |0 . Combining with (9.87) and (9.88) yields

(λ )
< =
 
mi (λ ) (|λ |−(λ ))+(|ν|−(ν))
i · (−1) · eλ,μ,ν
λ ,μ ,ν  mμ ,H |0
 , mμ̃,H̃ |0
i=1 μ
   
= f˜ν [C λ,0,0 ], [Z λ ,0,0 ] · mμ,H |0 , mμ̃,H |0 .

In view of (9.82), we conclude for all the partitions μ̃ with |μ̃| = |μ| that
< =
 λ,μ,ν
(|λ |−(λ ))+(|ν|−(ν))
(9.89) (−1) · eλ ,μ ,ν  mμ ,H |0 , mμ̃,H̃ |0
μ
F G
1  
= f˜ν [C λ,0,0 ], a−λ (1)|0 · mμ,H |0 , mμ̃,H |0 .
zλ 

Let L ⊂ H 2n ((P2 )[n] ; Z) be the Z-span of all the classes mμ̃,H |0 , μ̃  |μ|. Since
H, H = 1, we see from Theorem 9.22 that L, together with the bilinear form ·, · ,
is an integral lattice, and the classes mμ̃,H |0 , μ̃  |μ| form a basis of L. It follows
from (9.89) that
   λ,μ,ν
(−1)(|λ |−(λ ))+(|ν|−(ν)) · eλ ,μ ,ν  mμ ,H |0
μ
F G
1
= f˜ν [C λ,0,0 ], a−λ (1)|0 · mμ,H |0 .
zλ 
Therefore, we must have μ = μ and
F G
  1
(−1)(|λ |−(λ ))+(|ν|−(ν)) · eλ,μ,ν ˜
λ ,μ,ν = fν [C λ,0,0 ], a−λ (1)|0 .
zλ 
In particular, the right-hand-side is nonzero. By Proposition 9.32, (λ) = (λ ) and
λ # λ . If λ ≺ λ , then we are done. If λ = λ , then
F G
(−1)(|λ|−(λ))+(|ν|−(ν)) · eλ,μ,ν = ˜ν [C λ,0,0 ], 1 a−λ (1)|0 .
f
λ,μ,ν
zλ

By Proposition 9.32 again, the right-hand-side is equal to (−1)|λ|−(λ) eλλ f˜ν . Hence
|ν|−(ν) λ ˜
we obtain eλ,μ,ν
λ,μ,ν = (−1) eλ fν . 

Inspired by Lemma 9.33, we introduce the following definition.

Definition 9.34. Let (λ, μ, ν) and (λ , μ , ν  ) be 3-tuples of partitions. We

define (λ , μ , ν  ) ≺ (λ, μ, ν) if one of the three following conditions holds:
(i) ν  ≺ ν,
(ii) ν  = ν and μ ≺ μ,
(iii) ν  = ν, μ = μ and λ ≺ λ .

This defines a total ordering for all 3-tuples (λ, μ, ν) of partitions. Combining
with Lemma 9.33 yields the main result in this section.
 9.6. COMPARISON OF TWO INTEGRAL BASES OF H ∗ ((P2 )[n] ; Z) 201

Theorem 9.35. Let λ, μ, ν be three partitions. Then, [C λ,μ,ν ] is equal to

(−1)|ν|−(ν)
a−ν (1P2 )mμ,H a−λ (x)|0
zν
 eλ,μ,ν
λ ,μ ,ν 
+ a−ν  (1)mμ ,H a−λ (x)|0
  
zν 
λ ,μ ,ν

     
where eλ,μ,ν
λ ,μ ,ν  ∈ Z, |λ | + |μ | + |ν | = |λ| + |μ| + |ν|, and (λ , μ , ν ) ≺ (λ, μ, ν).

Proof. By Lemma 9.33, it remains to show that eλλ = f˜ν = 1. Let n =

|λ| + |μ| + |ν|. Consider the two integral bases of of H ∗ ((P2 )[n] ; Z):
An = [C λ,μ,ν ]||λ| + |μ| + |ν| = n ,
3 H
1
Bn = a−ν (1)mμ,H a−λ (x)|0 | |λ| + |μ| + |ν| = n .
zν
Order the basis elements in both An and Bn by using the total ordering for all
3-tuples of partitions defined in Definition 9.34. Let T be the transition matrix
between the ordered bases An and Bn . By Lemma 9.33, T is an integral triangular
matrix whose diagonal entries are (−1)|ν|−(ν) eλλ f˜ν with |λ|+|ν| ≤ n. Since det T =
±1 and both eλλ and f˜ν are positive integers, we must have eλλ = f˜ν = 1. 

We refer to [Mur, Section 5] for the transition matrices between the two integral
bases (9.60) and (9.64) of H ∗ ((P2 )[n] ; Z) in the special cases of n = 2, 3.
conjecture 9.36. [C λ,0,0 ] = a−λ (x)|0 for every partition λ.
By Theorem 9.35, Conjecture 9.36 holds when λ = (1n ), (1n−2 2), or (n).
Corollary 9.37. Let C∗ act on the affine coordinates u and v of C2 by
t(u, v) = (tw1 −w2 u, tw0 −w2 v), t ∈ C∗ and w1  w0 .
For ν  n, let Cν be the cell in (C2 )[n] corresponding to the fixed point ξν . Then,
1
(9.90) a−ν (1C2 )|0 = (−1)|ν|−(ν) [C ν ]
zν
in the integral cohomology H ∗ ((C2 )[n] ; Z) of the Hilbert scheme (C2 )[n] .
Proof. We see from (8.2) and (9.73) that in H ∗ ((C2 )[n] ; Z), we have
1
a−ν (1C2 )|0 = (−1)|ν|−(ν) ι∗n [C 0,0,ν ] .
zν
Now (9.90) follows from ι∗n [C 0,0,ν ] = [ι−1
n (C 0,0,ν )] = [C ν ]. 

Definition 9.38. Fix a point x on an arbitrary smooth projective surface X.

Identify Mn (x) with
Mn (P0 ) ⊂ (C2 )[n] ⊂ (P2 )[n] ,
where P0 denotes the origin of C2 . For λ  n, we define M λ (x) to be the closed
subvariety of Mn (x) ⊂ X [n] corresponding to C λ,0,0 ⊂ Mn (P0 ) under this identifi-
cation.
202 9. INTEGRAL COHOMOLOGY OF HILBERT SCHEMES

Corollary 9.39. Let x be a point on a smooth projective surface X. Then,


(9.91) [M λ (x)] = a−λ (x)|0 + eλν a−ν (x)|0 .
|λ|=|ν|, (λ)=(ν)
λ≺ν

Moreover, the coefficients eλν are integers and independent of X and x.

Proof. Since the elements in M λ (x) are supported at x, we see as in (9.76)
that

(9.92) [M λ (x)] = eλν (x) a−ν (x)|0 , eλν (x) ∈ Z.
|λ|=|ν|, (λ)=(ν)

To analyze the coefficients eλν (x), put n = |λ|. For ν  n with (ν) = (λ), let
Z ν (X) ⊂ X [n] be the closure of the subset consisting of the elements of the form

(ν)
ξi
i=1
where ξi ∈ Mνi (xi ), and x1 , . . . , x(ν) ∈ X are distinct. Then,

(ν)
1
(9.93) [Z ν (X)] = imi · a−ν (1)|0
i=1
zν
where ν = (1m1 2m2 · · · ). Note that the pairing [M λ (x)], [Z ν (X)] depends only
on an analytic neighborhood of x, which can be chosen to be independent of X and
x. Setting X = P2 and x = P0 , we obtain from Theorem 9.35 that

(ν)
(9.94) [M λ (x)], [Z ν (X)] = (−1)n−(ν) imi · eλν
i=1

where eλλ = 1 and eλν = 0 for ν ≺ λ. On the other hand, by (9.92) and (9.93),

(ν)
[M λ (x)], [Z ν (X)] = (−1) n−(ν)
imi · eλν (x).
i=1

Now it follows immediately from (9.94) that eλν (x) = eλν . 

 CHAPTER 10

∗
The ring structure of Horb (X (n) )

By Remark 1.25, the Hilbert scheme X [n] of points on a surface X is a crepant

resolution of singularities of the symmetric product X (n) = X n /Sn where Sn is
the n-th symmetric group. Inspired by orbifold string theory [DHVW, VW], the
geometry of the Hilbert scheme X [n] should be “equivalent” to the Sn -equivariant
geometry of X n . It is stated in the footnote 3 of [Groj] that the direct sum

+∞
KSn (X n ) ⊗ C
n=0

of equivariant K-groups can be identified with a Fock space of a Heisenberg algebra

[Seg, Wan1]. In particular, the size of the cohomology group H ∗ (X [n] ) coincides
with that of the equivariant K-group KSn (X [n] ) or the orbifold cohomology group
of X (n) . (Equivariant K-groups are related to orbifold cohomology groups by a
decomposition theorem due to [BC] and independently [Kuhn]). However the
“equivalence” on the level of ring structures is more subtle.
An orbifold is a topological space locally modeled on the quotient of Euclidean
space by a finite group. In [CR2, Section 4], Chen and Ruan introduced the orb-
∗
ifold cohomology ring Horb (·). For the symmetric product X (n) which is an orbifold,
∗
the ring structure of Horb (X (n) ) has been obtained in [FG1, Uri]. Combined with
[LS2], which was surveyed in Section 7.3, the results in [FG1, Uri] yield the proof
of Ruan’s Cohomological Crepant Resolution Conjecture [Rua2, Section 2] iden-
tifying the cohomology ring of the Hilbert scheme X [n] of points on a surface X
∗
with a numerically trivial canonical class and the orbifold ring Horb (X (n) ). The
isomorphisms in [FG1,LS2,Uri] are quite indirect. To deal with Ruan’s conjecture
in Chapter 16 for an arbitrary smooth projective complex surface X, an alterna-
∗
tive approach is needed to obtain a direction connection between Horb (X (n) ) and
vertex algebras. This is achieved in [QW1] via Heisenberg algebra actions. In this
∗
chapter, we will review Horb (M/G) for a global orbifold M/G and recall the results
from [QW1]. Since the results here are parallel to those in previous chapters for
the Hilbert schemes X [n] , many proofs will be omitted.

10.1. Generalities
∗
In this section, we will review the orbifold cohomology Horb (M/G) of the global
orbifold M/G where M is a complex manifold of complex dimension d with a finite
group G action.
First of all, we introduce the space
&
M $ G = {(g, x) ∈ G × M | gx = x} = M g,
g∈G

203
 ∗
204 10. THE RING STRUCTURE OF Horb (X (n) )

and G acts on M $ G naturally by h.(g, x) = (hgh−1 , hx). As a vector space, we

define H ∗ (M, G) to be the cohomology group of M $ G with complex coefficients,
or equivalently, 
H ∗ (M, G) = H ∗ (M g ).
g∈G
The space H ∗ (M, G) has a naturally induced G action, which is denoted by
−1
ad h : H ∗ (M g ) → H ∗ (M hgh ).
As a vector space, the orbifold cohomology group
∗
Horb (M/G)
is the G-invariant part of H ∗ (M, G), which is isomorphic to

H ∗ (M g /Z(g))
[g]∈G∗

where G∗ denotes the set of conjugacy classes of G and Z(g) = ZG (g) denotes the
centralizer of g in G.
For g ∈ G and x ∈ M g , write the eigenvalues of the action of g on the complex
tangent space Tx,M to be μk = e2πirk , where 0 ≤ rk < 1 and k = 1, . . . , d. The
degree shift number (or age in the terminology of Ito and Reid) is the rational
number
d
Fxg = rk .
k=1
It depends only on the connected component Z which contains x, so we can de-
note it by FZg . Then associated to a cohomology class in H r (Z), we assign the
corresponding element in H ∗ (M, G) (and thus in Horb ∗
(M/G)) the degree r + 2FZg .
∗
A ring structure on Horb (M/G) was introduced in [CR2, Section 4]. This was
subsequently clarified in [FG1] by introducing a ring structure on H ∗ (M, G) first
∗
and then passing to Horb (M/G) by restriction. We will use • to denote this product.
The ring structure on H ∗ (M, G) is degree-preserving, and has the property that
α • β lies in H ∗ (M gh ) for α ∈ H ∗ (M g ) and β ∈ H ∗ (M h ).
For 1 ∈ G,
H ∗ (M 1 /Z(1)) ∼
= H ∗ (M/G),
and thus we can regard α ∈ H ∗ (M/G) to be α ∈ Horb ∗
(M/G) by this isomorphism.
Also, given 
a= ag g
g∈G

in the group algebra C[G] (respectively, C[G]G ), we may regard a as an element in

H ∗ (M, G) (respectively, Horb
∗
(M/G)) whose component in each H ∗ (M g ) is
ag · 1M g ∈ H 0 (M g ).
If K is a subgroup of G, then we can define the restriction map from H ∗ (M, G)
to H ∗ (M, K) by projection to the component ⊕g∈K H ∗ (M g ) which, when restricted
to the G-invariant part, induces naturally a degree-preserving linear map
∗ ∗
K : Horb (M/G) → Horb (M/K).
ResG
We define the induction map
∗ ∗
K : H (M, K) → Horb (M/G)
IndG
 10.2. THE HEISENBERG ALGEBRA 205

by sending α ∈ H ∗ (M h ), where h ∈ K, to
1 
IndG
K (α) = adg(α).
|K|
g∈G

Note that IndGK (α) is clearly G-invariant. When restricted to the invariant part,
we obtain a degree-preserving linear map
∗ ∗
K : Horb (M/K) → Horb (M/G).
IndG

We often write the restriction and induction maps as ResK , Res and IndG , Ind,
when the groups involved are clear from the context. In particular, when M is a
∗
point, Horb (pt/G) reduces to the Grothendieck ring RC (G) of G, and we recover
the induction and restriction functors in the theory of finite groups.

10.2. The Heisenberg algebra

In this section, we will discuss the Heisenberg algebra action on the orbifold
∗
cohomology Horb (X (n) ) of the symmetric products X (n) . We will construct the
linear operators pn (α) (n ∈ Z, α ∈ H ∗ (X)), and write down their Heisenberg
commutation relations.
Let X be a closed complex manifold of complex dimension d. Our main objects
∗
are the ring Horb (X (n) ) and the non-commutative ring H ∗ (X n , Sn ). We denote


+∞
∗
FX = Horb (X (n) ).
n=0

We introduce a linear map

ωn : H ∗ (X) → Horb
∗
(X (n) )

as follows. Let σn be any permutation in the conjugacy class [n] ∈ (Sn )∗ which
consists of the n-cycles, and let τn : H ∗ (X) → H ∗ ((X n )σn ) be the obvious isomor-
phism. Given α ∈ H ∗ (X), we define

ωn (α) = τn (nα).
r+d(n−1)
Note that if α ∈ H r (X), then ωn (α) ∈ Horb (X (n) ). We also define
∗
chn : Horb (X (n) ) → H ∗ (X)

as the composition of the isomorphism H ∗ ((X n )σn ) ∼

= H ∗ (X) with the projection
∗ (n) ∗ n σn
from Horb (X ) to H ((X ) ).
Let α ∈ H ∗ (X). Set p0 (α) = 0. For n > 0, we define the creation operator
p−n (α) ∈ End(FX ) given by the composition (k ≥ 0):
ωn (α)⊗· C
∗ ∗ ∗
Horb (X k /Sk ) −→ Horb (X n /Sn ) Horb (X k /Sk )
∼
= ∗
−→ Horb (X n+k /(Sn × Sk ))
Ind ∗
−→ Horb (X n+k /Sn+k ),
 ∗
206 10. THE RING STRUCTURE OF Horb (X (n) )

and the annihilation operator pn (α) ∈ End(FX ) given by the composition (k ≥ 0):
∗ Res ∗
Horb (X n+k /Sn+k ) −→ Horb (X n+k /(Sn × Sk ))
∼ C
= ∗ ∗
−→ Horb (X n /Sn ) Horb (X k /Sk )
C
H ∗ (X) ∗
ch
−→
n
Horb (X k /Sk )
(α,·) ∗
−→ Horb (X k /Sk ).
Theorem 10.1. The operators pn (α) ∈ End(FX ) (n ∈ Z, α ∈ H ∗ (X)) generate
a Heisenberg (super)algebra with commutation relations given by
[pm (α), pn (β)] = mδm,−n (α, β) · IdFX
where n, m ∈ Z, α, β ∈ H ∗ (X). Furthermore, FX is an irreducible representation
of the Heisenberg algebra with the vacuum vector |0 = 1 ∈ H ∗ (pt) ∼
= C.
This theorem can be proved in the same way as an analogous theorem formu-
lated by using the equivariant K-group KSn (X n ) ⊗ C. This analogous theorem
was established in [Seg] (we refer to [Wan1, Theorem 4] and its proof for the de-
tails). Note that there is a fundamental sign difference in the two commutators of
Theorems 3.8 and Theorems 10.1.
|y|
In particular, for a given y ∈ Horb (X (n−1) ), by the definition of p−1 (α) (where
α ∈ H |α| (X)) and the induction map, we can write that
1 
p−1 (α)(y) = adg (α ⊗ y)
(n − 1)!
g∈Sn

(−1)|α|·|y| 
= adg (y ⊗ α).
(n − 1)!
g∈Sn
∗
For 0 ≤ i < n, we introduce the following cohomology class in Horb (X (n) ):
1
(10.1) Pi (α, n) = · p−i−1 (α)p−1 (1X )n−i−1 |0 .
(n − i − 1)!
This is the analogue of the class Bi (α, n) defined in Definition 7.3. The analogue
of the class Gi (α, n) (in the setting of Hilbert schemes) will be studied in the next
section.

10.3. The cohomology classes ηn (γ) and Ok (α, n)

In this section, we will define the classes ηn (γ) and Ok (α, n) in the orbifold
∗
cohomology Horb (X (n) ). The classes ηn (γ) play the role of the total Chern classes
c(L ) of the dual tautological bundles L[n]∗ over the Hilbert scheme X [n] , where
[n]∗

L denotes a line bundle over X. The classes Ok (α, n) play the role of the classes
Gk (α, n) for the Hilbert scheme X [n] which are defined in Definition 4.1.
Let
+∞
ch : C(Sn ) −→ ΛC = Λ ⊗Z C
n=0
be the Frobenius characteristic map from the direct sum of class algebras of the
symmetric group Sn to the ring ΛC of symmetric functions in infinitely many vari-
ables (with complex coefficients). Denote by ηn and εn the trivial and alternating
characters of Sn . Then ch sends ηn and εn to the n-th complete and elementary
 10.3. THE COHOMOLOGY CLASSES ηn (γ) AND Ok (α, n) 207

symmetric functions in ΛC respectively. We denote by pr the r-th power sum sym-

metric function.
Following [Juc] and [Mu, p. 287], we define the Jucys-Murphy elements ξj;n
of the symmetric group Sn to be the sums of transpositions:

ξj;n = (i, j), j = 1, . . . , n.
i<j

When it is clear from the text, we may simply write ξj;n as ξj . Put
Ξn = {ξ1 , . . . , ξn }.
Then the k-th elementary symmetric function ek (Ξn ) of Ξn is equal to the sum of
all permutations in Sn having exactly (n − k) cycles. Therefore, we obtain
 n 
n
(10.2) εn = (−1)d(σ) σ = (−1)k ek (Ξn ) = (1 − ξi )
σ∈Sn k=0 i=1

where d(σ) is the degree of σ defined by (7.12). Define


n
εn () = (1 − ξi ),
i=1
where  is a formal parameter here and below. We have
⎛ ⎞

+∞  p
ch(εn ())z n = exp ⎝ (−)r−1 z r ⎠ .
r
(10.3)
n=0
r
r≥1

Noting that εn (1) (respectively, εn (−1)) coincides with the alternating character εn
(respectively, the trivial character ηn ), we obtain two classical identities involving
ηn , εn , and pr by setting  = ±1 in (10.3). See also (8.13).
In the rest of this chapter, we will assume that X is a closed complex manifold
of even complex dimension d. Given γ ∈ H ∗ (X), we denote
γ (i) = 1⊗i−1 ⊗ γ ⊗ 1⊗n−i ∈ H ∗ (X n ),
and regard it to be a cohomology class in H ∗ (X n , Sn ) associated to the identity
conjugacy class. We define
ξi (γ) = ξi + γ (i) ∈ H ∗ (X n , Sn ).
We sometimes write ξi (γ) as ξi;n (γ) to specify its dependence on n when necessary.
Definition 10.2. Given γ ∈ H ∗ (X), we define ηn (γ) (respectively, εn (γ)) to
∗
be the cohomology class in Horb (X (n) ) whose component associated to an element
σ in the conjugacy class of partition λ of n is given by
γ ⊗(λ) ∈ H ∗ ((X n )σ ) ∼
= H ⊗(λ)
(respectively, by (−1)d(λ) γ ⊗(λ) ). We further define an operator η(γ) (respectively,
∗
ε(γ)) in End(FX ) by letting it act on Horb (X (n) ) by the orbifold product with
ηn (γ) (respectively, εn (γ)) for every n.
This definition is motivated by its counterpart in terms of equivariant K-groups
[Seg, Wan1]. Parallel to the Proposition 4 in [Wan1], we have
⎛ ⎞

+∞ 1
ηn (γ)z n = exp ⎝ p−r (γ)z r ⎠ · |0 .
n=0
r
r≥1
 ∗
208 10. THE RING STRUCTURE OF Horb (X (n) )

Proposition 10.3. Given γ ∈ H ∗ (X), the orbifold cup product of the n ele-
ments ξi (γ) (i = 1, . . . , n) in H ∗ (X n , Sn ) lies in Horb
∗
(X (n) ), and furthermore the
following identity holds:

n
(10.4) ηn (γ) = ξi (γ) = ξ1 (γ) • ξ2 (γ) • . . . • ξn (γ).
i=1

Proof. It suffices to prove (10.4), since the first claim follows from (10.4) and
the fact that ηn (γ) is Sn -invariant.
A typical monomial on the right-hand side of (10.4) is of the form
(ξi1 · · · ξik ) • (γ (j1 ) · · · γ (jn−k ) )
where i1 < . . . < ik , j1 < . . . < jn−k , and
{i1 , . . . , ik , j1 , . . . , jn−k } = {1, . . . , n}.
Here we have used the observation that ξi1 • · · · • ξik is just the usual multiplication
of permutations ξi1 · · · ξik and γ (j1 ) • · · · • γ (jn−k ) is just the ordinary cup product
γ (j1 ) · · · γ (jn−k ) in H ∗ (X n ) ∼
= H ∗ (X)⊗n . Note that every cycle of each permutation
σ appearing in ξi1 · · · ξik has exactly one number which does not belong to i1 , . . . , ik ,
and in addition, (σ) = n − k and d(σ) = k. Using the definition of the orbifold cup
product, we see that the product σ•γ (j1 ) · · · γ (jn−k ) does not involve the obstruction
bundles (or the group defects are trivial in the terminology of Lehn-Sorger) and is
equal to
∼ H ∗ ((X n )σ ).
γ ⊗(σ) ∈ (H ∗ (X))⊗(σ) =
This proves (10.4). 

Put

n
εn (γ, ) = (γ (i) − ξi ).
i=1
Then we have
⎛ ⎞

+∞  p−r (γ)
(10.5) εn (γ, )z n = exp ⎝ (−)r−1 zr ⎠ .
n=0
r
r≥1

∗
Regarding ξi = ξi (0) ∈ H (X , Sn ), we denote
n

ξi•k = ξi • . . . • ξi ∈ H ∗ (X n , Sn ),
/ 01 2
k times

and define
 1
e−ξi = (−ξi )•k ∈ H ∗ (X n , Sn ).
k!
k≥0

Definition 10.4. For a homogeneous class α ∈ H |α| (X), we define the class
∗
Ok (α, n) ∈ Horb (X (n) ) to be

1 
n
(−ξi )•k • α(i) ∈ Horb (X (n) ).
dk+|α|
(10.6) Ok (α, n) =
k! i=1
 10.4. INTERACTIONS BETWEEN HEISENBERG ALGEBRA AND Ok (γ) 209

We extends the notion Ok (α, n) linearly to all α ∈ H ∗ (X), and put

 
n
O(α, n) = Ok (α, n) = e−ξi • α(i) ,
k≥0 i=1

O (α, n) =  Ok (α, n).
k

k≥0

We further define the operator

Ok (α) ∈ End(FX )
(respectively, O(α), or O (α)) to be the orbifold cup product with Ok (α, n) (re-
∗
spectively, O(α, n), or O (α, n)) in Horb (X (n) ) for every n ≥ 0.
Remark 10.5. We can see that Ok (α, n) ∈ H ∗ (X n , Sn ) is Sn -invariant (and
∗
thus lies in Horb (X (n) )) as follows. For γ ∈ H ∗ (X), note that ej (ξ1 (γ), . . . , ξn (γ))
∗
lies in Horb (X ), where ej (ξ1 (γ), . . . , ξn (γ)) (1 ≤ j ≤ n) is the j-th elementary
(n)
∗
symmetric function in ξi (γ)’s. So Horb (X (n) ) contains all symmetric functions in
ξi (γ)’s. In particular,
 
O(e−γ , n) = (e−ξi • (e−γ )(i) ) = e−ξi (γ) ∈ Horb
∗
(X (n) ).
i i
∗
Letting γ vary, we see that O(α, n) and, similarly, Ok (α, n) lie in Horb (X (n) ).
In particular, O1 (1X ) ∈ End(FX ) is the generalized Goulden’s operator [Gou,
Section 3]. The reason for the convention of putting the sign in front of ξi in (10.6) is
to make the comparison with the Hilbert scheme side easier. Also, the power sums
of Jucys-Murphy elements have been studied in [LT, Section 4] which corresponds
to the present case when X is a point.
Note that the generalized Goulden’s operator O1 (1X ) is defined to be the orb-
∗
ifold cup product with O1 (1X , n) in Horb (X (n) ). Given an operator f ∈ End(FX ),
we define
f{1} = f = [O1 (1X ), f ],
and f{k+1} = (f{k} ) . We have the following.
1 3
Theorem 10.6. O1 (1X ) = − : p :0 (τ∗ 1X ).
6
Remark 10.7. This theorem is a counterpart of (4.18). The proof is essentially
the same as the proof in the case when X is a point, which was treated in [Gou,
Section 3] and [FW, Section 3] (also [LS2]). For example, if we look at the proof
of [LS2, Proposition 4.4], the Δ∗ and e there should be replaced by our τ∗ (which
equals −Δ∗ ) and −e respectively, since we are using the orbifold cup product of
[CR2].

10.4. Interactions between Heisenberg algebra and Ok (γ)

In this section, we will prove that the commutator [Ok (γ), p−1 (α)] is equal
{k}
to 1/k! · p−1 (γα). Moreover, we will study the relations between the operators
η(γ), ε(γ) and the operator p−1 (α).
The following is parallel to formula (4.7).
 ∗
210 10. THE RING STRUCTURE OF Horb (X (n) )

Theorem 10.8. Let γ, α ∈ H ∗ (X). Then we have

[O (γ), p−1 (α)] = exp( · ad O1 (1X ))(p−1 (γα))

1 {k}
[Ok (γ), p−1 (α)] = p (γα), k ≥ 0.
k! −1
Proof. For simplicity of signs, we assume that the cohomology classes γ, α
are of even degree. It suffices to prove the second identity.
∗
Recall that for y ∈ Horb (X (n−1) ), we have

1 
p−1 (α)(y) = adg (y ⊗ α).
(n − 1)!
g∈Sn

Regarding Sn−1 as the subgroup Sn−1 × 1 of Sn , we introduce an injective ring

homomorphism

ι : H ∗ (X n−1 , Sn−1 ) → H ∗ (X n , Sn )

by sending ασ to ασ ⊗ 1X , where σ ∈ Sn−1 . Thus

(n − 1)! [Ok (γ), p−1 (α)](y)

= (n − 1)! [Ok (γ) · p−1 (α)(y) − p−1 (α) · Ok (γ)(y)]
 
= Ok (γ, n) • adg (y ⊗ α) − adg ((Ok (γ, n − 1) • y) ⊗ α)
g∈Sn g∈Sn

= adg [(Ok (γ, n) − ι(Ok (γ, n − 1))) • (y ⊗ α)]
g

where we used the fact that Ok (γ, n) is Sn -invariant. By definition, we have

1
Ok (γ, n) − ι(Ok (γ, n − 1)) = · (−ξn;n )•k • γ (n) .
k!

Thus, we obtain

1 
(n − 1)! [Ok (γ), p−1 (α)](y) = · adg [(−ξn;n )•k • γ (n) • (y ⊗ α)]
k! g
1 
= · adg [(−ξn;n )•k • (y ⊗ γα)].
k! g

It remains to prove that

 {k}
(10.7) adg [(−ξn;n )•k • (y ⊗ γα)] = (n − 1)! p−1 (γα)(y).
g∈Sn

We will prove this by induction. It is clearly true for k = 0. Note that

O1 (1X , n) − ι(O1 (1X , n − 1)) = −ξn;n .

 10.4. INTERACTIONS BETWEEN HEISENBERG ALGEBRA AND Ok (γ) 211

Under the assumption that the formula (10.7) is true for k, we have

adg [(−ξn;n )•(k+1) • (y ⊗ γα)]
g

= adg [(O1 (1X , n) − ι(O1 (1X , n − 1))) • (−ξn;n )•k • (y ⊗ γα)]
g

= O1 (1X , n) • adg [(−ξn;n )•k • (y ⊗ γα)]
g

− adg [ι(O1 (1X , n − 1)) • (−ξn;n )•k • (y ⊗ γα)],
g

since O1 (γ, n) is Sn -invariant. By using the induction assumption twice, we get


adg [(−ξn;n )•(k+1) • (y ⊗ γα)]
g
{k}
= (n − 1)! O1 (1X , n) • p−1 (γα)(y)

− adg [(−ξn;n )•k • ((O1 (1X , n − 1) • y) ⊗ γα)]
g
{k} {k}
= (n − 1)! [O1 (1X ) · p−1 (γα)(y) − p−1 (γα)(O1 (1X , n − 1) • y)]
{k+1}
= (n − 1)! p−1 (γα)(y).

So by induction, we have established (10.7) and thus the theorem. 

We also have a theorem concerning the operator η(γ). It generalizes [LS2,

Proposition 4.6] which corresponds to the special case when γ = 1X (the assumption
there that X is a surface is unnecessary).

Theorem 10.9. Let γ, α ∈ H ∗ (X). We further assume that γ can be written

as a sum of classes of even degree. Then we have the following operator identities:

η(γ) · p−1 (α) = p−1 (γα) · η(γ) − p−1 (α) · η(γ),

ε(γ) · p−1 (α) = p−1 (γα) · ε(γ) + p−1 (α) · ε(γ).

Proof. The proof of the second formula is similar to that of the first, so we
will prove the first one only. For simplicity of signs in the proof, we assume that
the cohomology class α is of even degree.
By definition,

n
ηn (γ) = (ξi;n + γ (i) ).
i=1

It follows that

ηn (γ) − ηn−1 (γ) ⊗ γ = ξn;n • ι(ηn−1 (γ)).

 ∗
212 10. THE RING STRUCTURE OF Horb (X (n) )

∗
Given y ∈ Horb (X (n−1) ), we have
(n − 1)! [η(γ)·p−1 (α)(y) − p−1 (γα) · η(γ)(y)]
 
= ηn (γ) • adg (y ⊗ α) − adg [(ηn−1 (γ) • y) ⊗ γα]
g∈Sn g∈Sn

= adg [(ηn (γ) − ηn−1 (γ) ⊗ γ) • (y ⊗ α)]
g

= adg [ξn;n • ι(ηn−1 (γ)) • (y ⊗ α)]
g

= adg [ξn;n • ((ηn−1 (γ) • y) ⊗ α)]
g

= −O1 (1X , n) • adg [(ηn−1 (γ) • y) ⊗ α]
g

+ adg [(O1 (1X , n − 1) • ηn−1 (γ) • y) ⊗ α]
g
= (n − 1)! [−O1 (1X ) · p−1 (α) · η(γ)(y) + p−1 (α) · O1 (1X ) · η(γ)(y)]
= −(n − 1)! p−1 (α) · η(γ)(y).
This finishes the proof. 

∗
10.5. The ring structure of Horb (X (n) )
Using Theorem 10.1, Theorem 10.6 and Theorem 10.8, we can follow the ap-
proaches in Chapter 3, Chapter 4 and Chapter 7 to study the ring structure of
∗
Horb (X (n) ). Note that the terms involving the canonical class KX of X in various
formulas for the Hilbert schemes X [n] will disappear because there is no KX -term
in Theorem 10.6. Also, the fact e2X = 0 was used in previous chapters (where X
is a surface and eX is the Euler class of X). Thus, when dealing with the ring
∗
Horb (X (n) ) in this section, we sometimes need to treat separately and carefully the
case when X is a point (i.e. when e2X = 0).
Recall the classes Pi (α, n) and Oi (α, n) from (10.1) and (10.6) respectively.
The next result is parallel to Theorem 7.5.
Theorem 10.10. Let X be a closed complex manifold of even dimension. Then,
∗
the ring Horb (X (n) ) is generated by the cohomology classes Oi (α, n) (respectively,
the cohomology classes Pi (α, n)) where 0 ≤ i < n and α runs over a fixed linear
basis of H ∗ (X).
Theorem 10.11. Let X be a closed complex manifold of even dimension. Then
∗
the ring Horb (X (n) ) is determined uniquely by the ring H ∗ (X).
We refer to Theorem 10.11, and Proposition 10.12 and Proposition 10.13 be-
∗
low as the universality of the ring Horb (X (n) ). Theorem 10.11 follows from the
more quantitative descriptions of the orbifold cup product of ring generators of
∗
Horb (X (n) ) in Proposition 10.12 and Proposition 10.13. It also follows from the
results in [FG1]. Let 1−k = 0 if k < 0, and
1
1−k =p−1 (1X )k
k!
if k ≥ 0. The following proposition is parallel to Lemma 4.13.
 ∗
10.5. THE RING STRUCTURE OF Horb (X (n) ) 213

Proposition 10.12. (Universality) Let X be a closed complex manifold of

even dimension d > 0. Let n, s ≥ 1, k1 , . . . , ks ≥ 0, and let α1 , . . . , αs ∈ H ∗ (X) be
∗
homogeneous. Then, the orbifold product Ok1 (α1 , n) • · · · • Oks (αs , n) in Horb (X (n) )
is a finite linear combination of expressions of the form
⎛ ⎞
(π) mi −ri
 
sign(α, π) · 1  (π) −ri
 ⎝ p−ni,j ⎠ (τ(mi −ri )∗ (i απi )) · |0
 mi
− n− ni,j i=1 j=1
i=1 j=1

where π runs over all partitions of {1, . . . , s}, i ∈ {1X , eX },


ri = |i |/d ≤ mi ≤ 2 + kj ,
j∈πi

m
i −ri 
0 < ni,1 ≤ . . . ≤ ni,mi −ri , ni,j ≤ (kj + 1) for every i, and
j=1 j∈πi
⎛ ⎞

(π) i −ri
m 
s
⎝mi − 2 + ni,j ⎠ = ki .
i=1 j=1 i=1

Moreover, all the coefficients in this linear combination are independent of the man-
ifold X, the cohomology classes α1 , . . . , αs , and the integer n.
For the case d = 0 (i.e., X is a point), we adopt the simplified notations pm
and Ok (n) for pm (1X ) and Ok (1X , n) respectively. We have the following analog of
Proposition 10.12.
Proposition 10.13. Let n, s ≥ 1, k1 , . . . , ks ≥ 0. Then, Ok1 (n) • · · · • Oks (n)
is a finite linear combination of expressions of the form
⎛ ⎞
(π) mi −2ri
 
1  (π) −2ri
 ⎝ p−ni,j ⎠ · |0
 mi
− n− ni,j i=1 j=1
i=1 j=1

where π runs over all partitions of {1, . . . , s}, mi , ri ∈ Z+ such that


2ri ≤ mi ≤ 2 + kj ,
j∈πi

mi
−2ri 
0 < ni,1 ≤ . . . ≤ ni,mi −2ri , ni,j ≤ (kj + 1) for every i, and
j=1 j∈πi
⎛ ⎞

(π) i −2ri
m 
s
⎝mi − 2 + ni,j ⎠ = ki .
i=1 j=1 i=1

Moreover, all the coefficients in this linear combination are independent of n.

∗
The stability of the ring Horb (X (n) ) is given by the following which is parallel
to Theorem 7.7.
Theorem 10.14. Let X be a closed complex manifold of even dimension d. Let
s ≥ 1 and ki ≥ 1 for 1 ≤ i ≤ s. Fix ni,j ≥ 1 and αi,j ∈ H ∗ (X) for 1 ≤ j ≤ ki , and
 ∗
214 10. THE RING STRUCTURE OF Horb (X (n) )


ki
fix n with n ≥ ni,j for all 1 ≤ i ≤ s. Then the orbifold cup product
j=1
⎛ ⎞

s 
ki
⎝1 ki p−ni,j (αi,j ) · |0 ⎠
−(n− j=1 ni,j )
i=1 j=1

∗
in Horb (X (n) ) is equal to a finite linear combination of monomials of the form

N
1−(n− N
a=1 ma ) p−ma (γa ) · |0
a=1

where

N 
s 
ki
ma ≤ ni,j ,
a=1 i=1 j=1
and γ1 , . . . , γN depend only on eX , αi,j , 1 ≤ i ≤ s, 1 ≤ j ≤ ki . Moreover, the coeffi-
cients in this linear combination are independent of αi,j and n. These coefficients
are also independent of X provided d > 0.
Parallel to Section 7.2, we can use Theorem 10.14 to define a ring RX , called
∗
the stable ring, which completely encodes the ring structure of Horb (X (n) ) for each
∗
n. The stable ring RX depends only on the cohomology ring H (X). We refer to
[QW1, Subsection 4.4] for details.

10.6. The W algebras

In this section, we will assume that X is a closed complex manifold of even
dimension d > 0. Results here are counterparts of those in Section 4.4. However,
some signs have been modified due to the sign difference between the two Heisenberg
algebra commutators in the setups of Hilbert schemes and symmetric products (see
Theorem 3.8 and Theorem 10.1).
For p ≥ 0, m ∈ Z and α ∈ H ∗ (X), define Jpm (α) ∈ End(FX ) to be
⎛ ⎞
 1  s(λ) + m 2
− 2
p! · ⎝ pλ (τ∗ α) + pλ (τ∗ (eX α))⎠
λ! 24λ!
(λ)=p+1,|λ|=m (λ)=p−1,|λ|=m

where the λ’s are generalized partitions. Note that  J0m (α) = pm (α). We define
WX to be the linear span of the identity operator IdFX and the operators 
 Jpm (α)
∗
in End(FX ), where p ≥ 0, m ∈ Z and α ∈ H (X).
The following theorem describes the operator Ok (α) in terms of the Heisenberg
generators explicitly. It is a counterpart of Theorem 4.7.
Theorem 10.15. Let k ≥ 0, and α ∈ H ∗ (X). Then, Ok (α) is equal to
⎛ ⎞
 1  s(λ) − 2
(−1)k · ⎝ pλ (τ∗ α) + pλ (τ∗ (eX α))⎠ .
λ! 24λ!
(λ)=k+2,|λ|=0 (λ)=k,|λ|=0

In view of Theorem 10.15 and the definition of 

Jpm (α) ∈ End(FX ), we have
(−1)k k+1
Ok (α) = ·J (α).
(k + 1)! 0
 10.6. THE W ALGEBRAS 215

Since Ok (α, n) = Ok (α)p−1 (1X )n |0 /n!, we conclude from Theorem 10.15 that
Ok (α, n) is equal to
⎛
⎜  1
(10.8) (−1)k · ⎝ · 1−(n−j−1) p−λ (τ∗ α)|0
0≤j≤k,λ(j+1)
λ! · |λ|!
(λ)=k−j+1

⎞
  1 |λ| + s(λ) − 2 ⎟
+ · · 1−(n−j−1) p−λ (τ∗ (eX α))|0 ⎠ .
λ! · |λ|! 24
0≤j≤k λ(j+1)
(λ)=k−j−1

In terms of vertex operators, the operator 

Jpm (α) can be rewritten as:

1 1
(10.9) : pp+1 :m (τ∗ α) + p(m2 − 3m − 2p) : pp−1 :m (τ∗ (eX α))
(p + 1) 24
p(p − 1)
+ : (∂ 2 p) pp−2 :m (τ∗ (eX α)).
24

If we want the coefficients above to be independent of m, we can further rewrite

 1 p
Jpm (α) = : pp+1 :m (τ∗ α) + (∂ 2 : pp−1 :)m (τ∗ (eX α))
(p + 1) 24
(p + 1)p
+ (∂ : pp−1 :)m (τ∗ (eX α))
12
p(p2 − p − 2) p−1
+ :p :m (τ∗ (eX α))
24
p(p − 1)
+ : (∂ 2 p)pp−2 :m (τ∗ (eX α)).
24

The operators Op (α), pn (α), and 

Jpn (α) are related in the following way.

Proposition 10.16. Given p ≥ 0, α, β ∈ H ∗ (X), we have

(−1)p p
[Op (α), pn (β)] = −n · · Jn (αβ).
p!

Next, recall the number Ωp,q

m,n defined by (4.48). The following is parallel to
Theorem 4.25.

Theorem 10.17. Let X be a closed complex manifold of even dimension d > 0.

X is closed under the Lie bracket. More explicitly, for m, n ∈ Z,
The vector space W
and α, β ∈ H ∗ (X), we have

Ωp,q
[
Jpm (α), 
Jqn (β)] = (qm − pn) · p+q−1 m,n p+q−3
Jm+n (αβ) + · Jm+n (eX αβ)
12
 ∗
216 10. THE RING STRUCTURE OF Horb (X (n) )

where (p, q) ∈ Z2+ except for the unordered pairs (0, 0), (1, 0), (2, 0) and (1, 1). In
addition, for these four exceptional cases, we have
,
J0m (α), 
[ J0n (β)] = mδm,−n (αβ) · IdFX ,
X
J1m (α), 
[ J0n (β)] = −n · 
J0m+n (αβ),
,
   m3 − m
[Jm (α), Jn (β)] = −2n · Jm+n (αβ) +
2 0 1
δm,−n (eX αβ) · IdFX ,
6 X
,
   m3 − m
[Jm (α), Jn (β)] = (m − n) · Jm+n (αβ) +
1 1 1
δm,−n (eX αβ) · IdFX .
12 X

Remark 10.18. The W algebra W X should be viewed as a generalization of

the W∞ algebra. The assumption d > 0 above ensures that e2X = 0. The case when
d = 0 (i.e. X is a point) has been treated in [LT].
 Part 4

Equivariant cohomology of the

Hilbert schemes of points
 CHAPTER 11

Equivariant cohomology of Hilbert schemes

There exist fundamental and deep connections between the geometry of the
Hilbert schemes X [n] of points on a (quasi-)projective surface X and combinatorics
of symmetric functions. The monomial symmetric functions can be realized as cer-
tain ordinary cohomology classes of the Hilbert schemes associated to an embedded
curve in a surface ((9.12) and [Nak5, (9.13)]). Nakajima [Nak7] further showed
that the Jack polynomials whose Jack parameter is a positive integer γ can be re-
alized as certain T-equivariant cohomology classes of the Hilbert schemes of points
on the surface X(γ) which is the total space of the line bundle OP1 (−γ) over the
complex projective line P1 . Here T stands for the one-dimensional complex torus,
and the Jack parameter is interpreted as minus the self-intersection number of the
zero-section in X(γ). With very different motivations, Haiman [Hai1, Hai2, Hai3]
developed connections between the Macdonald polynomials and the geometry of
Hilbert schemes, and in particular realized the Macdonald polynomials as certain
T-equivariant K-homology classes of the Hilbert schemes of points on the affine
plane C2 . Moreover, the equivariant cohomology of Hilbert schemes are related to
Toda hierarchies, Hurwitz theory, equivariant (local) Gromov-Witten theory, and
equivariant (local) Donaldson-Thomas theory. These connections will be covered
in Chapters 12 and 14.
In this chapter, we will deal with the equivariant cohomology of the Hilbert
schemes X [n] where X is either the complex affine plane C2 or the total space of
a line bundle over P1 , and is equipped with a torus action. As in Section 3.2,
Heisenberg algebra actions on the equivariant cohomology of the Hilbert schemes
X [n] will be constructed. These actions enable the interpretation of the equivariant
cohomology of X [n] in terms of the ring of symmetric functions. In addition, the
equivariant cohomology classes corresponding to the Jack symmetric functions will
be identified. This chapter follows the presentation of [LQW7] which generalizes
the results in [Nak6, Vas]. We refer to [Nak7] for an excellent expository on
further topics such as geometric proofs of the norm formula and Pieri formula
of Jack symmetric functions, and a representation of the Virasoro algebra on the
equivariant cohomology of the Hilbert schemes (C2 )[n] (which is a special case of
Lehn’s result, Theorem 3.24).

11.1. Equivariant cohomology rings of Hilbert schemes

In this section, we will review the basics of equivariant cohomology and the
localization theorem. Let X be either the complex affine plane C2 or the total
space of a line bundle over P1 . We will set up the torus actions on X, the fixed
points of the induced actions on the Hilbert scheme X [n] , and the distinguished
equivariant cohomology classes associated to the fixed points. Moreover, we will
219
220 11. EQUIVARIANT COHOMOLOGY OF HILBERT SCHEMES

introduce a nondegenerate bilinear form on the space


+∞
HX = Hn
n=0

where Hn denotes the middle equivariant cohomology HT2n (X [n] ).

We begin with a brief review of equivariant cohomology and the localization
theorem. Let G be a connected Lie group, and let EG → BG be the universal
principal G-bundle where the total space EG is contractible. Let M be a topological
space with a G-action. Then the equivariant cohomology of X is defined to be
∗
HG (M ) = H ∗ (M ×G EG).
∗
Note that HG (pt) = H ∗ (BG) where pt denotes a point. Via the trivial map
∗
M → pt, we see that HG (M ) is an H ∗ (BG)-module.
∗ n
Let T = (C ) . For an algebraic variety M with a T-action, the equivariant
cohomology HT∗ (M ) is a C[t1 , . . . , tn ]-module if we identify HT∗ (pt) with C[t1 , . . . , tn ]
(each ti is an element of degree 2). For a T-equivariant and proper morphism
f : N → M of algebraic varieties, there is a Gysin homomorphism
f! : HT∗ (N ) → HT∗ (M )
of T-equivariant cohomology groups. If N is a T-equivariant codimension-k closed
subvariety of M and ι : N → M is the inclusion map, define
[N ] = ι! (1N ) ∈ HT2k (M )
where 1N ∈ HT0 (N ) is the unit of the algebra HT∗ (N ).
Let M be a smooth algebraic variety with a T-action. It is known that the
fixed point set M T is a union of smooth connected components Zj . Let
ιj : Zj → M
be the inclusion map, and Nj be the normal bundle of Zj in M . Since Nj is an
equivariant vector bundle, it has an equivariant Euler class:
eT (Nj ) ∈ HT∗ (Zj ).
Then for every α ∈ HT∗ (Zj ), we have
ι∗j (ιj! (α)) = α ∪ eT (Nj ).
Theorem 11.1. (Localization Theorem) There is an isomorphism
∼ 
HT∗ (X) ⊗ C(t1 , . . . , tn ) −→ HT∗ (Zj ) ⊗ C(t1 , . . . , tn )
=

induced by the map α → ι∗j α/eT (Nj ) j .

Moreover, the inverse is induced by

(αj )j → j ιj! (αj ). In particular, for every α ∈ HT∗ (X) ⊗ C(t1 , . . . , tn ), we have
  ι∗j α 
α= ιj! .
j
eT (Nj )

In the rest of this chapter, we will let

T = C∗ ,
and θ be the 1-dimensional standard T-module. We will consider two types of
surfaces with T-actions and the Hilbert schemes of points on these surfaces. The
first one is the complex plane C2 , while the second is the total space of a line
 11.1. EQUIVARIANT COHOMOLOGY RINGS OF HILBERT SCHEMES 221

bundle over P1 . The T-actions on these surfaces are specified in the following two
examples.
Example 11.2. Fix two nonzero integers α and β with the same signs. Let
u, v be the standard coordinate functions on C2 . We define the action of T on C2
by
(11.1) s · (u, v) = (sα u, s−β v), s ∈ T.
The origin of C2 is the only fixed point, which will be denoted by x. Let Σ and
Σ be the u-axis and v-axis respectively in C2 . As T-modules, we have Tx,Σ = θ −α
and Tx,Σ = θ β . By the localization theorem, we get
(11.2) [Σ] = −α−1 t−1 [x], [Σ ] = β −1 t−1 [x].
Example 11.3. Fix an integer γ > 1. Let X(γ) be the total space of the line
bundle OP1 (−γ) over P1 . The quasi-projective surface X(γ) can be regarded as the
quotient space of C × (C2 − {0}) by the C∗ -action defined by
(11.3) s · (b, b1 , b2 ) = (s−γ b, sb1 , sb2 ), s ∈ C∗ .
We use [(b, b1 , b2 )] to denote the equivalence class. Define a T-action on X(γ) by
(11.4) s · [(b, b1 , b2 )] = [(sb, s−1 b1 , b2 )], s ∈ T.
For i = 1, 2, let Xi be the open subset of X(γ) given by
(11.5) X1 = {[(b, b1 , b2 )] | b2 = 1}, X2 = {[(b, b1 , b2 )] | b1 = 1}.
Then X1 and X2 form an affine open cover of X(γ). Moreover, each Xi is T-
invariant. For simplicity, denote the point [(b, b1 , 1)] ∈ X1 by (b, b1 ). Similarly,
denote [(b, 1, b2 )] ∈ X2 by (b, b2 ). Then T acts on the points of X1 by
s · (b, b1 ) = (sb, s−1 b1 ),
i.e., T acts on the coordinate functions u1 and v1 of X1 by
(11.6) s · (u1 , v1 ) = (s−1 u1 , sv1 ), s ∈ T.
Similarly, T acts on the coordinate functions u2 and v2 of X2 by
(11.7) s · (u2 , v2 ) = (sγ−1 u2 , s−1 v2 ), s ∈ T.
Let xi be the origin of Xi . Then X(γ)T = {x1 , x2 }. Let
ρ : X(γ) → P1
be the projection sending [(b, b1 , b2 )] to [b1 , b2 ]. Let Σ0 ∼
= P1 be the zero section of
ρ, and
(11.8) Σ1 = ρ−1 ([0, 1]), Σ2 = ρ−1 ([1, 0]).
Then as T-modules, we have
Tx1 ,Σ1 = θ, Tx1 ,Σ0 = θ −1 , Tx2 ,Σ0 = θ, Tx2 ,Σ2 = θ 1−γ .
By the localization theorem, we get
(11.9) [Σ1 ] = t−1 [x1 ], [Σ0 ] = −t−1 [x1 ] + t−1 [x2 ], [Σ2 ] = (1 − γ)−1 t−1 [x2 ].
222 11. EQUIVARIANT COHOMOLOGY OF HILBERT SCHEMES

Remark 11.4. (i) Let X(1) be the total space of OP1 (−1). Using [(b, b1 , b2 )] ∈
X(1) to denote the equivalence class defined by (11.3), we define a T-action on
X(1) by
s · [(b, b1 , b2 )] = [(sb, s−2 b1 , b2 )], s ∈ T.
Then the methods and results below apply to X(1) as well.
(ii) Other T-actions on the surfaces X(γ), γ ≥ 1, with isolated fixed points can
be treated similarly.
In the rest of this chapter, let X be a surface in Example 11.2 or Exam-
ple 11.3. Next we will define some distinguished equivariant cohomology classes
for the Hilbert scheme X [n] . The T-action on X induces a T-action on X [n] . The
support of a T-fixed point in X [n] is contained in X T . As in Section 9.6, the T-fixed
points of X [n] are isolated and parametrized in terms of (multi-)partitions.
First of all, let X = C2 as in Example 11.2. The T-fixed points of X [n] are
supported in X T = {x} and indexed by partitions λ of n. We use ξλ to denote
the fixed point in (X [n] )T corresponding to a partition λ of n, and use Tξλ ,X [n] to
denote the T-equivariant tangent space of X [n] at the fixed point ξλ . As in (9.58),
we adopt the convention that ξλ is defined by the ideal
 
(11.10) Iξλ = v λ1 , uv λ2 , . . . , uk−1 v λk , uk
when λ = (λ1 ≥ λ2 ≥ · · · ≥ λk ). Then by [Nak6, Lemma 6.2] (also [ES1, Got2]),
for each λ  n, we have
 " #
(11.11) Tξλ ,X [n] = θ α(()+1)+βa() ⊕ θ −α()−β(a()+1)
∈Dλ

where Dλ is the Young diagram associated to the partition λ,  is a cell in Dλ , ()

is the leg length of , and a() is the arm length of  (we refer to Definition 1.4
for these notations). So
(11.12) eT (Tξλ ,X [n] ) = (−1)n cλ (α, β)cλ (α, β)t2n
where eT (·) stands for the equivariant Euler class and

(11.13) cλ (α, β) = α( () + 1) + βa() ,
∈Dλ

(11.14) cλ (α, β) = α () + β(a() + 1) .
∈Dλ

Note that [ξλ ] ∈ HT4n (X [n] ). We define the following distinguished class:
(−1)n −n
(11.15) [λ] = t [ξλ ].
cλ (α, β)
Now let X = X(γ) as in Example 11.3. In view of (11.1) and (11.6), there is
a T-equivariant identification between X1 and the complex plane in Example 11.2
with α = β = −1. Similarly, there is a T-equivariant identification between X2
and the complex plane in Example 11.2 with α = γ − 1 and β = 1. Then the
T-fixed points of X [n] are of the form ξλ1 + ξλ2 where λ1 and λ2 are partitions with
|λ1 | + |λ2 | = n, and ξλ1 and ξλ2 are defined in the previous paragraph as we identify
X1 and X2 with C2 respectively. For simplicity, put
ξλ1 ,λ2 = ξλ1 + ξλ2 .
 11.1. EQUIVARIANT COHOMOLOGY RINGS OF HILBERT SCHEMES 223

We have a T-equivariant splitting of tangent spaces:

Tξ ∼
,X [n] = T [|λ1 |] ⊕ T [|λ2 |] .
λ1 ,λ2 ξλ1 ,X1 ξλ2 ,X2

By (11.12),
eT (Tξλ1 ,λ2 ,X [n] ) = (−1)n cλ1 (−1, −1)cλ1 (−1, −1)cλ2 (γ − 1, 1)cλ2 (γ − 1, 1)t2n .
Also, as in (11.15), we introduce the distinguished class
(−1)n
(11.16) [λ1 , λ2 ] = t−n [ξλ1 ,λ2 ].
cλ1 (−1, −1)cλ2 (γ − 1, 1)
In the following, we study bilinear forms on the equivariant cohomology of the
Hilbert schemes. Define
(11.17) Hn = HT2n (X [n] ).
Since X [n] admits a cell decomposition, its odd Betti numbers are equal to zero. By
calculating the dimensions of the cells, we conclude that H k (X [n] ) = 0 for k > 2n.
Therefore, the spectral sequence associated with the fibration X [n] ×T ET → BT
degenerates at the E2 -term. We now have
HT2k (X [n] ) = tk−n ∪ HT2n (X [n] )
for k ≥ n, which shows that the classes defined in (11.15) and (11.16) are contained
in Hn . Furthermore, following [Vas, Section B], we define a product structure  on
Hn by
(11.18) tn ∪ (A  B) = A ∪ B ∈ HT4n (X [n] )
for A, B ∈ HT2n (X [n] ). We see that (Hn , ) is a ring.
Let
HT∗ (·) = HT∗ (·) ⊗C[t] C(t)
be the localization, and let

+∞
(11.19) HX = HT∗ (X [n] ) .
n=0

Let
ι : (X [n] )T → X [n]
be the inclusion map. By abusing notations, we also use ι! to denote the induced
Gysin map on the localized equivariant cohomology groups:
(11.20) ι! : HT∗ ((X [n] )T ) → HT∗ (X [n] ) ,
which is an isomorphism by the localization theorem.
Define a bilinear form ·, · : HT∗ (X [n] ) × HT∗ (X [n] ) → C(t) by putting
(11.21) A, B = (−1)n p! ι−1
! (A ∪ B)

where p is the projection (X [n] )T → pt. This induces a bilinear form ·, · on HX .
Remark 11.5. Passing from HT∗ (X [n] ) to Hn does not lead to loss of infor-
mation since we can recover the cup product and bilinear form on HT∗ (X [n] ) from
those on Hn . Thus, understanding the ring Hn is the same as understanding the
equivariant intersection theory on X [n] .
224 11. EQUIVARIANT COHOMOLOGY OF HILBERT SCHEMES

Next, we study the restriction of the bilinear form ·, · to Hn = HT2n (X [n] ).

When X = C2 , we see from the projection formula and (11.12) that
(11.22) [ξλ ] ∪ [ξμ ] = iλ! (1ξλ ) ∪ iμ! (1ξμ )
= iλ! 1ξλ ∪ i∗λ iμ! (1ξμ )
= δλ,μ eT (Tξλ ,X [n] )[ξλ ]
= δλ,μ (−1)n cλ (α, β)cλ (α, β)t2n [ξλ ]
where iλ : ξλ → X is the inclusion map. It follows from (11.15) and (11.21) that
for λ, μ  n, we have
cλ (α, β)
(11.23) [λ], [μ] = δλ,μ .
cλ (α, β)
By the localization theorem, we see that the classes [λ], λ  n form a linear basis
of the C-vector space Hn . Similarly, when X = X(γ) is from Example 11.3,
cλ2 (1 − γ, −1)
(11.24) [λ1 , λ2 ], [μ1 , μ2 ] = δλ1 ,μ1 δλ2 ,μ2
cλ2 (1 − γ, −1)
= [λ1 ], [μ1 ] · [λ2 ], [μ2 ] ,
and the classes [λ1 , λ2 ], where |λ1 | + |λ2 | = n, form a linear basis of Hn .
It follows that the restriction to Hn of the bilinear form ·, · on HT∗ (X [n] ) is
a nondegenerate bilinear form ·, · : Hn × Hn → C. This induces a nondegenerate
bilinear form
·, · : HX × HX → C
where the space HX is defined by

+∞
(11.25) HX = Hn .
n=0

11.2. Heisenberg algebras in equivariant setting

In this section, following [Vas], we will define Heisenberg algebra actions on
the equivariant cohomology HX of the Hilbert schemes X [n] .
Let X be a surface in Example 11.2 or Example 11.3. Fix a positive integer i.
For a T-invariant closed subscheme Y ⊂ X, we define
 
Yn,i = (ξ, η) ∈ X [n+i] × X [n] | η ⊂ ξ, Supp(Iη /Iξ ) = {y}, y ∈ Y

where Iη and Iξ are the sheaves of ideals corresponding to η and ξ respectively. Let
p1 and p2 be the projections of X [n+i] × X [n] to the two factors. As in [Vas, B.4],
we define a linear operator
p−i ([Y ]) ∈ End(HX )
by
" #
(11.26) p−i ([Y ])(A) = p1! p∗2 A ∪ [Yn,i ]

for A ∈ HT∗ (X [n] ) . Note that the restriction of p1 to Yn,i is proper. We define
pi ([Y ]) ∈ End(HX )
 11.3. EQUIVARIANT COHOMOLOGY AND JACK POLYNOMIALS 225

to be the adjoint operator of p−i ([Y ]) with respect to the bilinear form ·, · on HX .
For A ∈ HT∗ (X [n] ) , we have
" #
(11.27) pi ([Y ])(A) = (−1)i p2! ι × Id)−1 ∗
! (p1 A ∪ [Yn−i,i ]

where p2 is the projection of (X [n] )T × X [n−i] to X [n−i] . Finally we put

p0 ([Y ]) = 0.
In the following, assume that Y ⊂ X denotes a T-invariant closed curve. By
the definition of p−i ([Y ]) for i > 0, its restriction to HX gives a linear operator in
End(HX ), denoted by p−i ([Y ]) as well. Next, we recall from the previous section
that there is a nondegenerate bilinear form
·, · : HX ⊗C HX → C,
which is the restriction of the bilinear form ·, · on HX . Thus, the restriction of
pi ([Y ]) to HX is the adjoint operator of p−i ([Y ]) with respect to the bilinear form
·, · on HX , and hence is an operator in End(HX ) which will also be denoted by
pi ([Y ]).
By (11.2) and (11.9), HT2 (X) is linearly spanned by the classes [Y ] where Y
denotes T-invariant closed curves in X. So we can extend the notion pk ([Y ]) linearly
to obtain the operator pk (ω) ∈ End(HX ) for an arbitrary class ω ∈ HT2 (X). Note
from Example 11.2 (respectively, Example 11.3) that Σ and Σ (respectively, Σ0 and
Σi where i = 1 or 2) intersect transversely at one point. This simple but crucial
observation, together with an argument parallel to the proof of [Vas, Lemma 1],
leads to the following.
Proposition 11.6. The operators pk (ω), k ∈ Z and ω ∈ H1 = HT2 (X), acting
on HX satisfy the following Heisenberg commutation relation:
(11.28) [pk (ω1 ), p (ω2 )] = kδk,− ω1 , ω2 Id.
Furthermore, HX becomes the Fock space over the Heisenberg algebra modeled on
HT2 (X) with the unit |0 ∈ HT0 (X [0] ) of HT∗ (X [0] ) being a highest weight vector. 
As in the non-equivariant setting, the Heisenberg algebra actions on the Fock
space HX can be used to study its structure. In the next section, we will describe
the ring structure of HX via the ring of symmetric functions.

11.3. Equivariant cohomology and Jack polynomials

Let X be either the complex affine plane C2 as in Example 11.2 or the total
space of a line bundle over P1 as in Example 11.3. In this section, we will interpret
the Fock space HX in terms of the ring of symmetric functions. In particular,
we will identify the equivariant cohomology classes corresponding to the monomial
symmetric functions and the Jack symmetric functions.

11.3.1. The case of the complex plane C2 . In this subsection, we consider

X = C2 from Example 11.2. Let pi = pi ([Σ]) for i ∈ Z. Given a partition
λ = (λ1 , . . . , λ ) = (1m1 2m2 . . .), define
1  mi
(11.29) p̃−λ = p−i
zλ
i≥1
226 11. EQUIVARIANT COHOMOLOGY OF HILBERT SCHEMES

where zλ is from (1.1). By (11.2), (11.15) and (11.23), we have

[Σ], [Σ] = β/α.
So by Proposition 11.6,
@ A 1
(11.30) p̃−λ |0 , p̃−μ |0 = δλ,μ (β/α)(λ) .
zλ
Let X (n) (respectively, Σ(n) ) be the n-th symmetric product of X (respectively,
of Σ). Let ρn : X [n] → X (n) be the Hilbert-Chow morphism. As in (9.8), for λ  n,
define
*  +

Sλn Σ = λi yi ∈ Σ(n) | yi ∈ Σ, and yi = yj if i = j .
i=1

As in (9.7), let L Σ be the closure of ρ−1

λ n
n (Sλ Σ) in X
[n]
. An argument parallel to
the proof of the Corollary 6.10 in [Nak6] (also [Vas, Lemma 2]) shows that

(11.31) [Lλ Σ] = [λ] + cλ,μ [μ]
μ<λ

where cλ,μ ∈ C and “<” denotes the dominance partial ordering of partitions.
Recall from Section 1.2 the ring Λ of symmetric functions in infinitely many
variables, the space Λn of degree-n symmetric functions, the monomial symmetric
function mλ associated to a partition λ, the power sum symmetric function pλ
(α)
associated to λ, and the Jack symmetric function Pλ . Put
ΛC = Λ ⊗Z C,
ΛnC = Λn ⊗Z C,
1
(11.32) p̃λ = pλ .
zλ
The symmetric functions p̃λ form a linear basis of ΛC since {pλ }λ is a linear basis
of ΛC . So we can define a bilinear form ·, · on ΛC by
1
(11.33) p̃λ , p̃μ = δλ,μ (β/α)(λ) .
zλ
We introduce a ring structure ◦ on ΛnC defined by
(β/α) (β/α) (β/α)
Pλ Pμ P
(11.34) ◦ = δλ,μ  λ .
cλ (α, β) cμ (α, β) cλ (α, β)
Theorem 11.7. There exists a linear isomorphism φ : HX → ΛC preserving
bilinear forms such that
φ(p̃−λ |0 ) = p̃λ ,
φ([Lλ Σ]) = mλ ,
(β/α)
φ([λ]) = Pλ .
Furthermore, the restriction φn of φ to Hn is an isomorphism of rings, i.e.,
φn : (Hn , ) ∼
= (ΛnC , ◦).
 11.3. EQUIVARIANT COHOMOLOGY AND JACK POLYNOMIALS 227

Proof. The linear isomorphism φ : HX → ΛC is defined by mapping p̃−λ |0

to p̃λ . So
(11.35) φ(p−i A) = pi φ(A), i > 0, A ∈ HX .
Next, an argument similar to the proof of (9.9) verifies that for all i > 0, we
have the identity

(11.36) p−i · [Lλ Σ] = aλ,μ [Lμ Σ]
μ

where μ and aλ,μ are the same as defined in (1.5). We conclude from an induction,
(11.36) and (11.35) that
φ([Lλ Σ]) = mλ .
By (11.30) and (11.33), φ preserves the bilinear forms. So we see from (11.23)
that
c (α, β) c (1, β/α)
φ([λ]), φ([μ]) = δλ,μ λ = δλ,μ λ .
cλ (α, β) cλ (1, β/α)
By (11.31), we have

[λ] = [Lλ Σ] + dλ,μ [Lμ Σ].
μ<λ
It follows that

φ([λ]) = mλ + dλ,μ mμ .
μ<λ

By the characterization of the Jack symmetric functions in Section 1.2,

(β/α)
φ([λ]) = Pλ .
Finally, by (11.22) and the definition of , we get
[λ] [μ] [λ]
 = δλ,μ , λ, μ  n.
cλ (α, β) cμ (α, β) cλ (α, β)
In view of (11.34), φn : (Hn , ) → (ΛnC , ◦) is an isomorphism of rings. 

11.3.2. The case of the surface X(γ). Let X = X(γ) be the surface from
Example 11.3. Recall that there is a T-equivariant identification between the affine
open subset X1 of X and the complex plane in Example 11.2 with α = β = −1;
similarly for X2 and the complex plane in Example 11.2 with α = γ − 1 and β = 1.
By Proposition 11.6, HX is the Fock space of the Heisenberg algebra generated by
pi ([Σ1 ]) and pi ([Σ2 ]) with i ∈ Z, where Σ1 and Σ2 are the two T-equivariant fibers
in X defined by (11.8). Note also that HX1 is the Fock space of the Heisenberg
algebra generated by pi ([Σ1 ]) for i ∈ Z, where Σ1 is considered as a T-equivariant
closed curve in the affine open subset X1 ⊂ X. Similarly, HX2 is the Fock space of
the Heisenberg algebra generated by pi ([Σ2 ]) for i ∈ Z. To avoid confusion, we use
pX
i
1
to denote the operators pi ([Σ1 ]) acting on HX1 , pX i
2
to denote the operators
pi ([Σ2 ]) acting on HX2 , and pi ([Σ1 ]) and pi ([Σ2 ]) for the Heisenberg operators acting
on HX . As in (11.29), for a partition λ = (1m1 2m2 . . .) and for j = 1 or 2, we define
1 
p̃−λ ([Σj ]) = p−i ([Σj ])mi .
zλ
i≥1
228 11. EQUIVARIANT COHOMOLOGY OF HILBERT SCHEMES

Define a linear map Ψ : HX1 ⊗C HX2 → HX as follows:

(11.37) Ψ([λ1 ] ⊗ [λ2 ]) = [λ1 , λ2 ].
The linear map Ψ is an isomorphism since the classes [λ1 ], [λ2 ], [λ1 , λ2 ] defined in
(11.15) and (11.16) form linear bases of HX1 , HX2 , HX respectively. In addition, we
see from (11.24) that Ψ preserves the bilinear forms.
Lemma 11.8. The isomorphism Ψ commutes with Heisenberg operators, i.e.,

−i ⊗ Id
Ψ ◦ pX 1
= p−i ([Σ1 ]) ◦ Ψ,
Ψ ◦ Id ⊗ pX
−i
2
= p−i ([Σ2 ]) ◦ Ψ.
Proof. By the symmetry between Σ1 and Σ2 , we need only to prove the first
identity. Also, since Ψ preserves the bilinear forms and pi is the adjoint operator
of p−i , it suffices to prove the first identity for i > 0.
Since the two fibers Σ1 and Σ2 do not intersect, we see that for partitions λ1
and λ2 with |λ1 | + |λ2 | = n, the T-equivariant closed subvariety
1 2
Lλ Σ1 × Lλ Σ2 ⊂ X [n]
" 1 #
|λ | |λ2 |
is the closure of ρ−1
n Sλ1 Σ1 × Sλ2 Σ2 in X [n] , where ρn : X [n] → X (n) denotes
the Hilbert-Chow morphism. We conclude that
" 1 2
# 1 2
Ψ [Lλ Σ1 ] ⊗ [Lλ Σ2 ] = [Lλ Σ1 × Lλ Σ2 ]
1 2
by writing [Lλ Σ1 × Lλ Σ2 ] in terms of [λ1 , λ2 ] and [μ1 , μ2 ] where μ1 < λ1 or
1 2
μ2 < λ2 , similar to (11.31). It implies that the classes [Lλ Σ1 ×Lλ Σ2 ] form a linear
1 2
basis of HX . By (11.36) and a similar computation for p−i ([Σ1 ])[Lλ Σ1 × Lλ Σ2 ],
we have
" #
λ1 λ2
Ψ pX −i ⊗ Id [L Σ1 ] ⊗ [L Σ2 ]
1

" #
λ1 λ2
= Ψ pX 1
−i [L Σ 1 ] ⊗ [L Σ 2 ]
1 2
= p−i ([Σ1 ])[Lλ Σ1 × Lλ Σ2 ]
1 2
= p−i ([Σ1 ]) Ψ [Lλ Σ1 ] ⊗ [Lλ Σ2 ] .
It follows that Ψ ◦ pX
−i ⊗ Id = p−i ([Σ1 ]) ◦ Ψ for i > 0.
1


Let ΛC,1 be the ring ΛC of symmetric functions with the bilinear form
1
p̃λ , p̃μ = δλ,μ ,
zλ
ΛC,2 be the same ring ΛC equipped with a different bilinear form
1 (λ)
p̃λ , p̃μ = δλ,μ 1/(γ − 1) ,
zλ
and ΛnC,i
i
be the space of degree-ni symmetric polynomials in ΛC,i . The tensor
product ΛC,1 ⊗C ΛC,2 has an induced bilinear form. Let

(ΛC,1 ⊗C ΛC,2 )n = ΛnC,1
1
⊗C ΛnC,2
2
.
n1 +n2 =n
 11.3. EQUIVARIANT COHOMOLOGY AND JACK POLYNOMIALS 229

We define a ring structure on (ΛC,1 ⊗C ΛC,2 )n by declaring the elements

(1) (1/(γ−1))
Pλ1 P 2
⊗ λ ∈ (ΛC,1 ⊗C ΛC,2 )n ,
cλ1 (−1, −1) cλ2 (γ − 1, 1)
where |λ1 | + |λ2 | = n, to be idempotents.
Theorem 11.9. There exists a linear isomorphism
Φ : HX → ΛC,1 ⊗C ΛC,2
preserving bilinear forms such that
Φ p̃−λ1 ([Σ1 ])p̃−λ2 ([Σ2 ])|0 = p̃λ1 ⊗ p̃λ2 ,
1 2
(λ (λ )
Φ [Σ1 × Σ2 ] = mλ 1 ⊗ mλ 2 ,
(1) (1/(γ−1))
1
Φ([λ , λ ]) = 2
Pλ1 ⊗ Pλ2 .
Furthermore, the restriction of the map Φ to the component Hn is a ring isomor-
phism onto (ΛC,1 ⊗C ΛC,2 )n .
Proof. The statements follow from Lemma 11.8 and arguments similar to the
proof of Theorem 11.7. 
 CHAPTER 12

Hilbert/Gromov-Witten correspondence

It is evident from Part 2 and Part 3 that the Chern character operators intro-
duced in Definition 4.1 (ii) are very essential to the understanding of the cohomology
ring structure and intersection theory of the Hilbert scheme of points on a surface.
In this chapter, we will discuss their analogues in the equivariant setting. More
precisely, using the setups and results from Chapter 11, we will define equivariant
Chern character operators on the equivariant cohomology of the Hilbert schemes
(C2 )[n] of points on the complex affine plane C2 . These operators give rise to a mas-
ter operator H(z) acting on certain Fock space. In the study of the class algebras
of symmetric groups, a counterpart of this operator has also played a distinguished
role [LT, Wan4]. Via the standard boson-fermion correspondence, the operator
H(z) turns out to be related to another operator ε0 (z), which has played a key role
in the study of stationary Gromov-Witten theory of curves [OP1, OP2], by the
following formula:

 
1 1
(12.1) H(z) = ε0 (z) − I .
e − e−z/2
z/2 ez/2 − e−z/2

Formula (12.1) defines a Gromov-Witten/Hilbert correspondence. Combining this

with the Gromov-Witten/Hurwitz correspondence in [OP1, OP2] gives us a Hur-
witz/Hilbert correspondence. In fact, the same combinatorics of the symmetric
groups underlies these distinct geometric studies.
We will begin this chapter with a brief introduction to Gromov-Witten theory.
Then using the equivariant Chern character operators, we will define the N -point
function, the multi-point trace function and the τ -functions as the generating se-
ries of suitable equivariant intersection theory on (C2 )[n] , and reformulate them
in terms of the operator formalism. We will compute these functions, and prove
that the τ -functions satisfy the 2-Toda hierarchies. In addition, we will study
the numerical aspects of the Hilbert/Gromov-Witten correspondence by proving
that certain structure constants in the cup products of the cohomology ring of the
Hilbert schemes are related to the 1-point disconnected series of the stationary
Gromov-Witten theory of the projective line P1 relative to 0, ∞ ∈ P1 . We will
also establish a precise identity between some general structure constants under
the cup product in H ∗ ((C2 )[n] ) and certain Hurwitz numbers for P1 . We refer to
[QW3] for generalizations to the Hilbert schemes of points on the minimal reso-
lution of the quotient C2 /Γ where Γ is a finite cyclic subgroup of SL2 (C) and to
their connections with the class algebras of wreath products. Theorem 12.5, Theo-
rem 12.7, Theorem 12.10, Theorem 12.12 and their proofs are from [LQW6]; while
Theorem 12.13, Theorem 12.16 and their proofs are from [LQW8].
231
232 12. HILBERT/GROMOV-WITTEN CORRESPONDENCE

12.1. A brief introduction to Gromov-Witten theory

In this section, we will give a brief introduction to Gromov-Witten theory. We
will establish the basic notations and definitions in Gromov-Witten theory, and
summarize some properties regarding the virtual fundamental class.
Let Y be a smooth projective variety. A k-pointed stable map to Y consists of
a complete nodal curve D with k distinct ordered smooth points p1 , . . . , pk and a
morphism μ : D → Y such that the data (μ, D, p1 , . . . , pk ) has only finitely many
automorphisms. In this case, the stable map is denoted by
[μ : (D; p1 , . . . , pk ) → Y ].
For a fixed homology class β ∈ H2 (Y, Z), let
Mg,k (Y, β)
be the coarse moduli space parameterizing all the stable maps [μ : (D; p1 , . . . , pk ) →
Y ] such that μ∗ [D] = β and the arithmetic genus of D is g. Then, we have the i-th
evaluation map :
(12.2) evi : Mg,k (Y, β) → Y
defined by
evi ([μ : (D; p1 , . . . , pk ) → Y ]) = μ(pi ) ∈ Y.
It is known that the coarse moduli space Mg,k (Y, β) is projective [FP, Theorem 1]
and has a virtual fundamental class [LT1, LT2, Beh1, BF1]
[Mg,k (Y, β)]vir ∈ Ad (Mg,k (Y, β))
where
(12.3) d = −(KY · β) + (dim(Y ) − 3)(1 − g) + k
is the expected (complex) dimension of Mg,k (Y, β), and Ad (Mg,k (Y, β)) is the Chow
group of d-dimensional cycles in the moduli space Mg,k (Y, β).
The Gromov-Witten invariants are defined by using the virtual fundamental
class [Mg,k (Y, β)]vir . Recall that an element


2 dimC (Y )
α ∈ H ∗ (Y, C) = H j (Y, C)
j=0

is homogeneous if α ∈ H j (Y, C) for some j; in this case, we take |α| = j. Let

α1 , . . . , αk ∈ H ∗ (Y, C) such that every αi is homogeneous and

k
(12.4) |αi | = 2d.
i=1

Then, we have the k-point Gromov-Witten invariant defined by:

,
(12.5) α1 , . . . , αk g,β = (ev1 × · · · × evk )∗ (α1 ⊗ . . . ⊗ αk ).
[Mg,k (Y,β)]vir

The Fundamental Class Axiom of Gromov-Witten theory asserts that

(12.6) α1 , . . . , αk−1 , 1Y g,β =0
 12.2. THE HILBERT/GROMOV-WITTEN CORRESPONDENCE 233

if either k + 2g ≥ 4 or β = 0 and k ≥ 1. The Divisor Axiom states that

,
(12.7) α1 , . . . , αk−1 , αk g,β = αk · α1 , . . . , αk−1 g,β
β

if αk ∈ H 2 (Y ), and if either k + 2g ≥ 4 or β = 0 and k ≥ 1. A special case of the

Composition Law [KM, (3.3) and (3.6)] states that
(12.8) α1 α2 , α3 , α4 0,β + α1 , α2 , α3 α4 0,β
 
+ α1 , α2 , Δa 0,β1 Δa , α3 , α4 0,β2
β1 +β2 =β, β1 ,β2 =0 a

= α1 α3 , α2 , α4 0,β + α1 , α3 , α2 α4 0,β
 
+ α1 , α3 , Δa 0,β1 Δa , α2 , α4 0,β2
β1 +β2 =β, β1 ,β2 =0 a

where α1 , α2 , α3 , α4 ∈ H ∗ (Y ) are cohomology classes of even degrees, {Δa }a de-

notes a basis of H ∗ (Y ), and {Δa }a is the basis of H ∗ (Y ) dual to {Δa }a with respect
to the intersection pairing of Y .
Next, we summarize certain properties concerning the virtual fundamental
class. To begin with, we recall that the excess dimension is the difference be-
tween the dimension of Mg,k (Y, β) and the expected dimension d in (12.3). For
0 ≤ i < k, use
(12.9) fk,i : Mg,k (Y, β) → Mg,i (Y, β)
to stand for the forgetful map obtained by forgetting the last (k − i) marked points
and contracting all the unstable components. It is known that fk,i is flat when
β = 0 and 0 ≤ i < k. The following can be found in [LT1, Theorem 4.3], [Get,
Proposition 2.5] and [CK, Proposition 7.1.8].
Proposition 12.1. Let β ∈ H2 (Y, Z) and β = 0. Let e be the excess dimension
of Mg,k (Y, β), and M ⊂ Mg,k (Y, β) be a closed subscheme. Then,
(i) [Mg,k (Y, β)]vir = (fk,0 )∗ [Mg,0 (Y, β)]vir ;
(ii) [Mg,k (Y, β)]vir = ce (R1 (fk+1,k )∗ (evk+1 )∗ TY ) if R1 (fk+1,k )∗ (evk+1 )∗ TY is
a rank-e locally free sheaf over the moduli space Mg,k (Y, β);
(iii) [Mg,k (Y, β)]vir |M = ce ((R1 (fk+1,k )∗ (evk+1 )∗ TY )|M ) if there exists an open
subset O of Mg,k (Y, β) such that M ⊂ O (i.e, O is an open neighborhood
of M) and the restriction (R1 (fk+1,k )∗ (evk+1 )∗ TY )|O is a rank-e locally
free sheaf over O.

12.2. The Hilbert/Gromov-Witten correspondence

In this section, we will consider the action of T = C∗ on X = C2 by
(12.10) s · (u, v) = (su, s−1 v), s ∈ T.
This is a special case of Example 11.2 with α = β = 1. We will establish
the Hilbert/Gromov-Witten correspondence by identifying some operators on the
equivarian cohomology of the Hilbert schemes X [n] with some operators of Ok-
ounkov and Pandharipande which play a fundamental role in the study of the
stationary Gromov-Witten invariants of curves [OP1, OP2].
234 12. HILBERT/GROMOV-WITTEN CORRESPONDENCE

We will adopt the notations from Chapter 11 relevant to X = C2 . In particular,

we have the Heisenberg operators pi , i ∈ Z from Subsection 11.3.1 acting on the
Fock space (11.25):

+∞ 
+∞
HX = Hn = HT2n (X [n] ),
n=0 n=0

and satisfying the commutation relations (11.28):

(12.11) [pk , p ] = kδk,− · Id.
Next, for a partition λ = (λ1 ≥ . . . ≥ λ ) of n, the corresponding fixed point
ξλ ∈ X [n] is defined by the ideal
 
(12.12) Iξλ = uλ1 , uλ2 v, . . . , uλ v −1 , v  .
For simplicity, let O = OX . The tautological rank-n vector bundle O[n] over
X is T-equivariant. The fiber of O[n] over the fixed point ξλ ∈ X [n] is canonically
[n]
 
(12.13) O[n] |ξλ = H 0 Oξλ = C[u, v]/ uλ1 , uλ2 v, . . . , uλ v −1 , v  .
It follows that as a T-module, we have

(12.14) O[n] |ξλ = θ c()
∈Dλ

where c() is the content of the cell  in the Young diagram Dλ , and is defined in
Definition 1.4. Denote by
[n]
chk,T
the k-th T-equivariant Chern character of O[n] . In particular the zero-th Chern
[n]
character ch0,T equals the rank of the vector bundle O[n] , which is n. Then
[n] 1 
chk,T |ξλ = (c()t)k .
k!
∈Dλ

By the projection formula, we have in HT∗ (X [n] ) that

⎛ ⎞
1  1 
chk,T ∪[ξλ ] = iλ! ⎝ c()k tk ⎠ =
[n]
(12.15) c()k tk [ξλ ].
k! k!
∈Dλ ∈Dλ

Let k be a nonnegative integer. Denote

(12.16)  [n] = tn−k ch[n] ∈ Hn

ch k k,T

where Hn is from (11.17).

 [n] , 0 ≤ k < n,
Remark 12.2. The Theorem 5.10 in [Wan4] implies that chk
form a set of ring generators of Hn = HT2n (X [n] ).
Recall from (11.18) the ring product  on Hn . We define an operator G (re-
spectively, Gk ) in End(HX ) by sending a ∈ Hn to
 [n]
a 
chk
k≥0
 12.2. THE HILBERT/GROMOV-WITTEN CORRESPONDENCE 235

 [n] ) in Hn for each n. Similarly, we define an operator Gz by

(respectively, to a  ch k
sending a ∈ Hn to

a  [n]
z k chk
k≥0
for each n, where z is a variable. By definition, we have

Gz = Gk z k .
k≥0

Formula (12.15) is equivalent to


(12.17) Gz ([λ]) = ezc() · [λ]
∈Dλ

where [λ] is defined by (11.15) with α = β = 1.

Lemma 12.3. Let ς(z) = ez/2 − e−z/2 and λ = (λ1 ≥ λ2 ≥ . . .). Then,
+∞ 
1  1
z(λi −i+1/2)
Gz ([λ]) = e − · [λ].
ς(z) i=1 ς(z)

Proof. The contents c()’s of λ are:

−i + 1, −i + 2, . . . , −i + λi , i = 1, . . . , (λ).
Noting that
ez(λi −i+1) − ez(−i+1)
ez(−i+1) + . . . + ez(−i+λi ) = ,
ez − 1
formula (12.17) can be written as

1  z(λi −i+1/2)
+∞
Gz ([λ]) = (e − ez(−i+1/2) ) · [λ]
ς(z) i=1
where we have used λi = 0 for i > (λ). In view of the identity

+∞
1
ez(−i+1/2) = ,
i=1
ς(z)
our lemma follows. 
Next, denote by Om the T-equivariant line bundle over X = C2 associated to
the T-character tm , where m ∈ Z. Let
M(m, n)
be the moduli space which parameterizes all rank-1 subsheaves of Om such that
the quotients are supported at finitely many points of X and have length n. Given
I ∈ X [n] , then Om ⊗ I is an element in M(m, n). Define
H(m)
n = HT2n (M(m, n)).
As before, the study of the equivariant cohomology ring HT∗ (M(m, n)) leads to the
(m)
ring Hn whose product is denoted by . The natural identification
M(m, n) ∼
= X [n]
236 12. HILBERT/GROMOV-WITTEN CORRESPONDENCE

leads to the natural identification of the rings Hn

(m) ∼
= Hn , which induces a bilinear
(m) (m)
form ·, · n on Hn from Hn . We introduce

+∞
F (m) = H(m)
n , m∈Z
n=0

F = F (m) .
m∈Z
(0)
In particular we identify M(0, n) = X [n] and Hn = Hn . We denote by S the
isomorphism
S : H(m)
n → H(m+1)
n .
This induces isomorphisms (which will be denoted by S again)
S:F →F
and
S : F (m) → F (m+1)
(m) (m)
for all m ∈ Z. The bilinear form on F induced from ·, · n on Hn will be again
denoted by ·, · .
By the standard boson-fermion correspondence in [MJD, Theorem 5.2], F
can be identified with the fermionic Fock space (or equivalently, the infinite wedge
space). The operator S is exactly the shift operator on the fermionic Fock space.
Given an operator f ∈ End(F), define the vacuum expectation f of f by
f = |0 , f|0 .
It is well-known that the completed infinite-rank general linear Lie algebra sl I∞
(whose standard basis is denoted by Ei,j , i, j ∈ Z + 1/2) acts on the fermionic Fock
space. As will become clear below, the study of the equivariant intersection theory
on Hilbert schemes naturally leads to the following operator in End(F):
1  kz
H(z) = e Ek,k .
ς(z) 1
k∈Z+ 2

The operator is further expanded as


+∞
H(z) = Hk z k .
k=0
We remark that the operator H(z) in a somewhat different form has appeared
in the study of the class algebras of the symmetric groups and wreath products
[LT, Wan4], and it affords a compact expression in terms of vertex operators.
On the other hand, the following operators
 δr,0
(12.18) εr (z) = ez(k−r/2) Ek−r,k + , r∈Z
1
ς(z)
k∈Z+ 2

have been introduced by Okounkov and Pandharipande. These operators have

played a fundamental role in the study of the stationary Gromov-Witten invariants
of curves [OP1, OP2], where stationary means that every marked point of the
stable maps is mapped to some prescribed point on the curve. To state the results,
for two partitions λ and μ with |λ| = |μ|, let
•
(12.19) Fλ,μ (z1 , . . . , zN )
 12.2. THE HILBERT/GROMOV-WITTEN CORRESPONDENCE 237

be the N -point disconnected series of stationary Gromov-Witten invariants of P1

relative to 0, ∞ ∈ P1 , where disconnected means that the domain curves of the
stable maps to P1 are allowed to be disconnected. Then, given two partitions λ and
μ of n, we have
•
(12.20) Fλ,μ (z1 , . . . , zN ) = p̃−λ , ε0 (z1 ) . . . ε0 (zN )p̃−μ n

where the notation p̃−λ is introduced in (11.29). Moreover, define the q-trace Trq (f)
of an operator f ∈ End(HX ) by

(12.21) Trq (f) = zλ p̃−λ , f(p̃−λ ) q |λ| ,
λ

and let
FE (z1 , . . . , zn ; q)
be the N -point disconnected series (with an additional variable q inserted) of the
stationary Gromov-Witten invariants of an elliptic curve E. Since Gromov-Witten
invariants are deformation invariant, FE (z1 , . . . , zn ; q) is independent of the elliptic
curve E. We have

(12.22) FE (z1 , . . . , zn ; q) = Trq (ε0 (z1 ) · · · ε0 (zN )).

The operators εr (z), r ∈ Z, satisfy the following identities:

(12.23) [pk , εr (z)] = ς(kz)εk+r (z)

(12.24) S −1 ε0 (z)S = ez ε0 (z)

+∞
(12.25) ε0 (z)([λ]) = ez(λi −i+1/2) · [λ]
i=1

for all partitions λ = (λ1 ≥ λ2 ≥ . . .) (here we recall [λ] ∈ HX = F (0) ).

Lemma 12.4. The following identities hold for operators on F:

 
1 1
(12.26) H(z) = ε0 (z) − Id
ς(z) ς(z)
e −1
mz
(12.27) S −m H(z)S m = emz H(z) + Id, m ∈ Z.
ς(z)2
Proof. The first identity follows from the definitions. The second one follows
from the first one and (12.24). 

Theorem 12.5. As operators on F (0) = HX , the following identification holds

Gz = H(z).

Proof. Follows from Lemma 12.3, formula (12.25) and Lemma 12.4. 

In light of the interpretation of H(z) in Theorem 12.5 and the role of ε0 (z)
in the study of the stationary Gromov-Witten invariants (12.20) and (12.22), the
identity (12.26) defines the Hilbert/Gromov-Witten correspondence.
238 12. HILBERT/GROMOV-WITTEN CORRESPONDENCE

12.3. The N -point functions and the multi-point trace functions

Let X = C2 and consider the T-action (12.10). In view of Remark 11.5 and
Remark 12.2, we see that one way to understand the equivariant intersection theory
on X [n] is to study the products
 [n]  ch
ch  [n]  . . .  ch
 [n]
k1 k2 kN
for arbitrary nonnegative integers k1 , . . . , kN , N ≥ 1. For partitions λ and μ of n,
we define the intersection numbers
F G
[n] [n] [n] [n] [n]
(12.28)     
λ, chk1 · · · chkN , μ n = p̃−λ , chk1  chk2  . . .  chkN  p̃−μ
n
In this section, we will form two types of generating functions (namely, the N -point
function and the multi-point trace function) of these intersection numbers and then
reformulate them in terms of the operator formalism.
We begin with the N -point function. Given two partitions λ and μ of n, the
N -point function Gλ,μ (z1 , . . . , zN ) is defined by

(12.29) Gλ,μ (z1 , . . . , zN ) = z1k1 . . . zN
kN  [n] · · · ch
λ, ch  [n] , μ n .
k1 kN
k1 ,...,kN

This can be reformulated in an operator form:

Gλ,μ (z1 , . . . , zN ) = p̃−λ , Gz1 . . . GzN p̃−μ n
= p̃−λ , H(z1 ) . . . H(zN )p̃−μ n .
Recall from (12.20) that
•
Fλ,μ (z1 , . . . , zN ) = p̃−λ , ε0 (z1 ) . . . ε0 (zN )p̃−μ n .
•
Note that, in particular for N = 0, we have defined Gλ,μ () and Fλ,μ (). By (12.11),
we have
• δλ,μ
Gλ,μ () = Fλ,μ () = .
zλ
It follows from (12.26) and Theorem 12.5 that the 1-point function is given by
 
1 • 1 •
(12.30) Gλ,μ (z) = Fλ,μ (z) − Fλ,μ ()
ς(z) ς(z)
 
1 • δλ,μ
= Fλ,μ (z) − .
ς(z) zλ ς(z)
• ◦
What remains is computing Fλ,μ (z). The somewhat simpler connected series Fλ,μ
rather than the disconnected series has been computed in [OP1, Subsection 3.2].
We introduce some notations on partitions. For two partitions λ = (1m1 2m2 . . .)
and μ = (1n1 2n2 . . .), define the partition
λ + μ = (1m1 +n1 2m2 +n2 . . .).
We say λ ⊂ μ if mi ≤ ni for all i, and in this case, define the partition
μ − λ = (1n1 −m1 2n2 −m2 . . .).
For λ = (λ1 ≥ . . . ≥ λr ) where r = (λ), put
ς(λ, z) = ς(λ1 z) · · · ς(λr z).
Given a subset U ⊂ r = {1, . . . , r}, we denote by λU the subpartition of λ which
consists of the parts λi , i ∈ U .
 12.3. THE N -POINT FUNCTIONS AND MULTI-POINT TRACE FUNCTIONS 239

Proposition 12.6. Let λ = (λ1 ≥ . . . ≥ λr ) and μ = (μ1 ≥ . . . ≥ μs ) be

partitions of n, where r = (λ) and s = (μ). Then,
 ς(λU , z) ς(λU + μ − λ, z)
•
Fλ,μ (z) =
zλ zλU +μ−λ ς(z)
U
summed over subsets U ⊂ r such that λ ⊂ λU + μ.
Proof. Let [pλ,U , ε0 (z)] denote the multi-commutator
[· · · [pλb , [pλa , ε0 (z)]] · · · ],
if we write U = {a, b, . . .}. Note that the multi-commutator is independent of the
ordering of elements in U since the pk ’s (k > 0) commute with each other.
By moving ε0 (z) to the left whenever possible, we have
 
(12.31) pλr . . . pλ1 ε0 (z) = [pλ,U , ε0 (z)] pλi
U⊂r i∈r\U
 
= ς(λU , z)ε|λU | (z) pλi
U⊂r i∈r\U

where we have repeatedly used (12.23). It follows that

• 1
(12.32) Fλ,μ (z) = pλr . . . pλ1 ε0 (z)p−μ1 . . . p−μs
zλ zμ
< =
1  
= ς(λU , z) ε|λU | (z) pλi · p−μ1 . . . p−μs
zλ zμ
U⊂r i∈r\U

Denote by λ− U to be the partition which consists of the parts λi , i ∈ r\U .

Apparently, the vacuum expectation
< =

ε|λU | (z) pλi · p−μ1 . . . p−μs = 0 unless λ−
U ⊂ μ.
i∈r\U

Ifλ−
U ⊂ μ, or equivalently if λ ⊂ λU + μ, then μ − λ− U = (λU + μ) − λ, and we can
show by induction that
 zμ
(12.33) pλi · p−μ1 . . . p−μs |0 = p−μa1 . . . p−μat |0
zλU +μ−λ
i∈r\U

where we have denoted μ − λ− U = (μa1 , . . . , μat ).

By (12.32) and (12.33), we have
• 1  zμ  
(12.34) Fλ,μ (z) = ς(λU , z) ε|λU | (z)p−μa1 . . . p−μat .
zλ zμ zλU +μ−λ
U⊂r
 
Similar to (12.31), we now move ε|λU | (z) to the right in ε|λU | (z)p−μa1 . . . p−μat
whenever possible. Note that if εK (z) (K > 0) results from such a move, then the
corresponding vacuum expectation is zero. Therefore
   
(12.35) ε|λU | (z)p−μa1 . . . p−μat = [· · · [ε|λU | (z), p−μa1 ], · · · p−μat ]
= ς(λU + μ − λ, z) ε0 (z)
= ς(λU + μ − λ, z)/ς(z).
Now the proposition follows from (12.34) and (12.35). 
240 12. HILBERT/GROMOV-WITTEN CORRESPONDENCE

Theorem 12.7. The 1-point function Gλ,μ (z) is given by

 
1  ς(λU , z) ς(λU + μ − λ, z)
Gλ,μ (z) = − δλ,μ
zλ ς(z)2 zλU +μ−λ
U

where U runs over the subsets of (λ) such that λ ⊂ λU + μ.

Proof. Follows from (12.30) and Proposition 12.6. 
In general, by Theorem 12.5, we have the following.
Proposition 12.8. We have
1  (−1)N −|U| •
Gλ,μ (z1 , . . . , zN ) = N
Fλ,μ (zU )
i=1 ς(zi ) U⊂N i∈N \U ς(zi )

where < =

•
Fλ,μ (zU ) = p̃−λ , ε0 (zi )p̃−μ .
i∈U n

Remark 12.9. The strategy used in Proposition 12.6 could be generalized to

•
compute the N -point disconnected series Fλ,μ (z1 , . . . , zN ), although the notations
would be a bit involved. Then the computation of Gλ,μ (z1 , . . . , zN ) follows by
Proposition 12.8.
Next, we study the multi-point trace function. Recall from (12.21) the q-trace
Trq (f) of an operator f ∈ End(HX ). In particular, for the identity operator Id on
HX , we have
1
Trq (Id) = ,
(q; q)∞
where
(12.36) (a; q)n = (1 − a)(1 − aq)(1 − aq 2 ) · · · (1 − aq n ).
Our main object here is the multi-point trace function
Trq (Gz1 · · · GzN )
of a product of the operators Gzj , which encodes information about the intersection
numbers (12.28). Here z1 , . . . , zN are independent variables.
We define Θ(z) = Θ(z; q) by
 (m+1/2)2
Θ(z) = Θ(z; q) = η(q)−3 (−1)m q 2 e(m+1/2)z
m∈Z

where η(q) = q 1/24 (q; q)∞ is the Dedekind eta function. By the Jacobi triple prod-
uct identity, we obtain
Θ(z) = (ez/2 − e−z/2 )(qez ; q)∞ (qe−z ; q)∞ /(q; q)2∞ .
We further define
dk
Θ(k) (z) = Θ(z), k ≥ 0.
dz k
We agree that Θ(k) (z) = 0 for k < 0. Given a positive integer N , we denote
N = {1, 2, . . . , N }. Given a finite set U , we denote by SU the symmetric group
on U . In particular, SN = SN . Given U = {u1 , . . . , uk } ⊂ N with u1 < . . . < uk
 12.4. τ -FUNCTIONS OF 2-TODA HIERARCHIES 241

and a permutation σ ∈ SU , we denote by MU,σ the k × k matrix whose (i, j)-th

entry is
Θ(j−i+1) (zσu1 · · · zσuk−j )/(j − i + 1)!.
Put
ΘU,σ = Θ(zσu1 )Θ(zσu1 zσu2 ) · · · Θ(zσu1 · · · zσuk ).
By the Theorem 0.5 in [BO],
1  det MN ,σ
(12.37) Trq (ε0 (z1 ) · · · ε0 (zN )) = .
(q; q)∞ ΘN ,σ
σ∈SN

In view of (12.22), formula (12.37) also computes the stationary Gromov-Witten

invariants of an elliptic curve.
Theorem 12.10. We have

1  (−1)N −|U| σ∈SU
det MU,σ
ΘU,σ
Trq (Gz1 · · · GzN ) = N
.
(q; q)∞ i=1 ς(zi ) U⊂N i∈N \U ς(zi )

Proof. By Lemma 12.4 and Theorem 12.5, we have

1  (−1)N −|U| Trq ( ε0 (zi ))
Trq (Gz1 · · · GzN ) = N
i∈U
.
i=1 ς(zi ) U⊂N i∈N \U ς(zi )

Now the theorem follows by applying (12.37) by replacing the set N by U . 

12.4. Equivariant intersection and τ -functions of 2-Toda hierarchies

The relations between the cohomology rings of Hilbert schemes of points on
surfaces and W algebras (Section 4.4) through vertex operators have led to the
following question (Question 5 in [QW1, Section 6]): what is the precise connec-
tion between Hilbert schemes and integrable hierarchies? The connection between
the geometry of Hilbert schemes and vertex operators is made through the Chern
character operators. However the Chern character operators are usually nilpotent
for (cohomology) degree reasons. This presents a serious difficulty for a sensible
answer to the above question in the framework of ordinary cohomology theory. In
this section, we will deal with this question in the equivariant setting. We will
continue to consider the T-action (12.10) on X = C2 , and define τ -functions as the
generating series of certain equivariant intersection numbers on X [n] . Then we will
prove that these τ -functions satisfy the 2-Toda hierarchies.
Let
t = (t1 , t2 , . . .),
s = (s1 , s2 , . . .),
x = (x1 , x2 , . . .)
be three sequences of indeterminates. Define the following half vertex operators:
 

Γ± (t) = exp tk p±k /k .
k>0

Given a partition μ = (μ1 ≥ μ2 ≥ . . .), we write tμ = tμ1 tμ2 · · · .

242 12. HILBERT/GROMOV-WITTEN CORRESPONDENCE

Define the τ -function τ (x, t, s) to be the following generating function for the
equivariant intersection numbers λ, ch  [n] . . . ch
 [n] , μ n from (12.28):
k1 kN
< +∞  =
   [n]
τ (x, t, s) = tλ sμ λ, exp 
xk ch ,μ
k
n |λ|=|μ|=n k=0 n

Note that
 
Γ− (s) = tλ p̃−λ ,
n≥0 |λ|=n

Γ+ (t) = Γ− (t)† .
From the definition of Hk and Theorem 12.5, we see that the τ -function affords an
operator formulation:
< +∞  =

τ (x, t, s) = Γ+ (t) exp xk Hk Γ− (s) .
k=0

Next, we define the Chern character operators from the space M(m, n). We
have a universal exact sequence:
0 → Jm → π2∗ Om → Qm → 0
[n]
where π1 , π2 are the projections of M(m, n) × X to the two factors. Denote by Om
the T-equivariant rank-n vector bundle over M(m, n) given by the push-forward
π1∗ (Qm ), whose fiber over a point ξλ ∈ M(m, n) ∼ = X [n] is given by
C
(12.38) Om[n]
|ξλ = Om O[n] |ξλ .

In the same way as defining the operators Gz and Gk (k ≥ 0) acting on HX = F (0) ,

(m) (m)
we can define the operators Gz and Gk (k ≥ 0) acting on F (m) using the cup
 [n] [n] (m)
products with k≥0 tn−k chk (Om )z k and tn−k chk (Om ) respectively on Hn . For
technical reasons below, we introduce the following modification
 k (O[n] ) = tn−k chk (O[n] ) + c(m) tn
ch m m k
(m)
where the constant ck is defined by
emz − 1  (m)
2
= mz −1 + ck z k .
ς(z)
k≥0

Equivalently, if we define
 (m) = G(m) + e − 1 Id.
mz
(12.39) G z z
ς(z)2
 (m)   (m) z k , then
and further write G z = mz −1 Id + k≥0 G k

 (m) = G(m) + c(m) Id,

G k≥0
k k k

and G  (m) acts on H(m)

n by the product with ch k (Om
[n]
). When m = 0, we have
k
(0)  (0) (0)
ck = 0 and Gk = Gk = Gk for all k.
Similarly we define the equivariant intersection numbers for M(m, n), denoted
(m)
by · n . When m = 0, they reduce to the ones defined earlier.
 12.4. τ -FUNCTIONS OF 2-TODA HIERARCHIES 243

Now define the τ -function τ (x, t, s, m) to be the following generating function

of the equivariant intersection numbers on M(m, n):
< +∞  =(m)
  
τ (x, t, s, m) = tλ sμ λ, exp  [n]
xk chk (Om ) , μ
n |λ|=|μ|=n k=0 n

In particular we have τ (x, t, s, 0) = τ (x, t, s).

Lemma 12.11. As operators on F (m) , the identification holds:
 (m) = H(z).
G z

Proof. Under the identification

∼
Sm : Hn = H(0)
n → Hn ,
(m)

we denote by [λ](m) the image of [λ]. By the identification of toric action (12.38),
the same proof as in Lemma 12.3 implies that
 +∞  
emz  1 emz
− 1
 ([λ] ) =
G(m) (m) z(λ
e i −i+1/2)
− + · [λ](m) .
z
ς(z) i=1 ς(z) ς(z)2

By (12.27) and S m ([λ]) = [λ](m) , we have

emz − 1 m
H(z)([λ](m) ) = H(z)S m ([λ]) = emz S m H(z)([λ]) + S ([λ]).
ς(z)2
This can be rewritten as
 +∞  
emz  z(λi −i+1/2) 1 emz − 1
(m)
H(z)([λ] ) = e − + · [λ](m)
ς(z) i=1 ς(z) ς(z)2

by Lemma 12.3 and Theorem 12.5. This finishes the proof. 

Theorem 12.12.
(i) The function τ (x, t, s, m) can be reformulated as:
< +∞  =

−m m
τ (x, t, s, m) = S Γ+ (t) exp xk Hk Γ− (s)S .
k=0

(ii) The function τ (x, t, s, m), m ∈ Z, satisfies the 2-Toda hierarchy of Ueno-
Takasaki [UT, Section 1]. The lowest equation among the hierarchy reads:
∂2 τ (t, s, x, m + 1) τ (t, s, x, m − 1)
(12.40) ln τ (t, s, x, m) = .
∂t1 ∂s1 τ (t, s, x, m)2
Proof. (i) follows from Lemma 12.11 and the definition of τ (x, t, s, m), while
I ∞.
(ii) is standard since the operator Hk lies in sl 

By setting t2 = t3 = · · · = s2 = s3 = · · · = 0 and x2 = x3 = · · · = 0 in τ (x, t, s),

we obtain the following generating function of the intersection numbers on Hilbert
schemes:
 F G
1 n 1
(t1 s1 ex0 )n p−1 , exp (x1 G1 ) · pn−1
n
n! n! n
244 12. HILBERT/GROMOV-WITTEN CORRESPONDENCE

thanks to the fact that G0 ([λ]) = |λ| · [λ]. Setting u = x0 + ln(t1 s1 ), we denote
the above generating function by τ (u, x1 ). A simple computation reduces the Toda
equation (12.40) to the following:
∂2 τ (u + x1 , x1 ) τ (u − x1 , x1 )
e−u 2
ln τ (u, x1 ) = .
∂u τ (u, x1 )2
It is interesting to observe that this τ function can also be interpreted [Ok1, Sub-
section 2.9] as generating functions of certain Hurwitz numbers.

12.5. Numerical aspects of Hilbert/Gromov-Witten correspondence

In this section, X denotes an arbitrary smooth projective complex surface.
We will study the numerical aspect of the Hilbert/Gromov-Witten correspondence
discovered in Theorem 12.5. More precisely, we will prove that certain structure
constants in the cup products of the cohomology ring of the Hilbert scheme X [n] are
•
related to the 1-point disconnected series Fλ,μ (z) of the stationary Gromov-Witten
theory of the projective line P relative to 0, ∞ ∈ P1 .
1

Recall the notations I [n] and a−λ (α) from Definition 8.5 and (9.2) respectively.
For a partition λ = (1m1 2m2 · · · ) and a cohomology class α ∈ H ∗ (X), define
1 1 
(12.41) ã−λ (α) = a−λ (α) = a−r (α)mr .
zλ zλ
r≥1

The following theorem relates the construct constant cλk,μ to the 1-point discon-
•
nected series Fλ,μ (z).

4
Theorem 12.13. Let 2 ≤ k ≤ n, μ  n, and I = H i (X). Then,
i=1

(12.42) a−(1n−k k) (1X )|0 ∪ a−μ (1X )|0


≡ cλk,μ a−λ (1X )|0 (mod I [n] )
λn
(λ)=(μ)+1−k

where cλk,μ is the coefficient of z k−1 in

(n − k)! k! zμ •
Fλ,μ (z).
ς(z)
Proof. In view of (12.41), we note that (12.42) is equivalent to
(12.43) ã−(1n−k k) (1X )|0 ∪ ã−μ (1X )|0

≡ c̃λk,μ ã−λ (1X )|0 (mod I [n] )
λn
(λ)=(μ)+1−k

where c̃λk,μ is the coefficient of z k−1 in

(k − 1)! zλ •
Fλ,μ (z).
ς(z)
Next, we outline our main idea used in the proof of (12.43). Using the results
in Section 8.3, we will convert (12.43) to a cup product in the cohomology ring
H ∗ ((C2 )[n] ) of the Hilbert scheme (C2 )[n] . Via the geometric realization of Heisen-
berg monomial classes given by Proposition 3.16, the cup product in H ∗ ((C2 )[n] )
 12.5. NUMERICAL ASPECTS OF HILBERT/GW CORRESPONDENCE 245

is related to the cup product in HT∗ ((C2 )[n] ). Then, we will apply (12.30) which
is a consequence of Theorem 12.5 where an algebraic relation between the equi-
variant cohomology ring of (C2 )[n] and the Gromov-Witten theory of P1 have been
established.
To convert (12.43) to a cup product in H ∗ ((C2 )[n] ), define ã−μ (1C2 ) as in
(12.41). By Theorem 8.12, the quotient ring H ∗ (X [n] )/I [n] is isomorphic to the co-
homology ring H ∗ ((C2 )[n] ). Moreover, we see from (8.18), (8.19), (8.4) that via this
isomorphism, the cohomology classes ã−(1n−k k) (1X )|0 (mod I [n] ) and ã−μ (1X )|0
[n]
(mod I [n] ) correspond to the classes (−1)k−1 (k − 1)! chk−1 and ã−μ (1C2 )|0 in
H ∗ ((C2 )[n] ) respectively. Here, chk−1 stands for the (k − 1)-th component of the
[n]

Chern character of the rank-n tautological vector bundle (OC2 )[n] . Thus, proving
(12.43) is equivalent to showing that
[n]
(12.44) (−1)k−1 (k − 1)! chk−1 ∪ã−μ (1C2 )|0

= c̃λk,μ ã−λ (1C2 )|0 .
λn
(λ)=(μ)+1−k

Next we rewrite the left-hand-side of (12.44) by using the equivariant cohomol-

ogy HT∗ ((C2 )[n] ) where T acts on C2 as in (12.10). Recall the equivariant Heisenberg
operator pi (1C2 ) from Section 11.2. Define p̃−μ (1C2 ) similarly as in (12.41). It is
well-known that HT∗ ((C2 )[n] ) is a C[t]-module, and there is a ring isomorphism
HT∗ ((C2 )[n] )/ t ∼
= H ∗ ((C2 )[n] ).
Let
Ψ : HT∗ ((C2 )[n] ) → HT∗ ((C2 )[n] )/ t ∼
= H ∗ ((C2 )[n] )
be the natural projection. Assume that μ = (1s1 2s2 . . . s ). As in Proposition 3.16,
the Heisenberg monomial class
  
zμ  a−r (1C2 )sr
ã−μ (1C2 )|0 = |0 ∈ H ∗ ((C2 )[n] )
r≥1 s r ! r=1
s r !

is represented by the closure Wμ0 ⊂ (C2 )[n] of

W (1, C2 ; . . . ; 1, C2 ; . . . ; , C2 ; . . . ; , C2 )0 .
/ 01 2 / 01 2
s1 times s times

Similarly, Wμ0 is a T-equivariant closed subscheme of (C2 )[n] , and

zμ
p̃−μ (1C2 )|0 ∈ HT∗ ((C2 )[n] )
r≥1 sr !

is the equivariant cohomology class corresponding to Wμ0 . So

Ψ(p̃−μ (1C2 )|0 ) = ã−μ (1C2 )|0
and
[n]
(12.45) (−1)k−1 (k − 1)! chk−1 ∪ã−μ (1C2 )|0
[n]
= (−1)k−1 (k − 1)! Ψ chk−1,T ∪p̃−μ (1C2 )|0 .
246 12. HILBERT/GROMOV-WITTEN CORRESPONDENCE

Note that 1C2 = [C2 ] = −t−1 [{0} × C] in HT∗ ((C2 )[n] ). It follows that
(12.46) p̃−μ (1C2 ) = (−t−1 )(μ) p̃−μ ([{0} × C]) = (−t)−(μ) p̃−μ
where the last p̃−μ denotes p̃−μ ([{0} × C]) for simplicity. By (12.45),
[n]
(12.47) (−1)k−1 (k − 1)! chk−1 ∪ã−μ (1C2 )|0
" #
= (−1)k−1−(μ) (k − 1)! Ψ t−(μ) chk−1,T ∪p̃−μ |0 .
[n]

Recall from Section 11.2 that {p̃−λ |0 | λ  n} is a linear basis of HT2n ((C2 )[n] ).
Put

(12.48)  [n]  p̃−μ |0 =
ch dλk,μ p̃−λ |0
k−1
λn

where all the coefficients dλk,μ are independent of t. So by (12.16) and (11.18),
[n]
[n]
chk−1,T ∪p̃−μ |0 
= t−n+(k−1) ch
(12.49) k−1,T ∪ p̃−μ |0
[n]

= tk−1 ch k−1,T  p̃−μ |0

= t k−1
dλk,μ p̃−λ |0 .
λn

Combining this with (12.47) and (12.46), we conclude that

[n]
(12.50) (−1)k−1 (k − 1)! chk−1 ∪ã−μ (1C2 )|0
 

= (−1)k−1−(μ) (k − 1)! Ψ tk−1−(μ) dλk,μ p̃−λ |0
λn
 

= (k − 1)! Ψ (−t) k−1−(μ)+(λ)
dλk,μ p̃−λ (1C2 )|0 .
λn

Since Ψ(p̃−λ (1C2 )|0 ) = ã−λ (1C2 )|0 = 0 for each λ  n, we obtain from (12.50)
that
[n]
(12.51) (−1)k−1 (k − 1)! chk−1 ∪ã−μ (1C2 )|0

= (k − 1)! dλk,μ ã−λ (1C2 )|0 .
λn
(λ)=(μ)+1−k

By (12.51), to finish the proof of (12.44), it remains to show that the structure
constant dλk,μ in (12.48) is the coefficient of z k−1 in the expansion of the series
zλ •
(12.52) F (z).
ς(z) λ,μ
By (12.29) and (12.30), we have
  
 [n]  p̃−μ |0 1 • δλ,μ
z i p̃−λ |0 , chi n = Fλ,μ (z) − .
i
ς(z) zλ ς(z)
Combining these observations with (12.48) and (11.30) with α = β = 1, we get
[n]

= zλ p̃−λ |0 , ch
dλk,μ k−1  p̃−μ |0 n
  
zλ • δλ,μ
= Coeffzk−1 Fλ,μ (z) − .
ς(z) zλ ς(z)
 12.6. RELATION TO THE HURWITZ NUMBERS OF P1 247

Finally, since (λ) = (μ)+1−k and k ≥ 2, we have λ = μ. Therefore, the structure

constant dλk,μ in (12.51) is the coefficient of z k−1 in (12.52). 


4
Remark 12.14. Let n ≥ 2, μ  n, and I = H i (X). Theorem 12.13 handles
i=1
the cup product a−(1n−k k) (1X )|0 ∪a−μ (1X )|0 (mod I [n] ) for the partition (1n−k k)
of n. By Lemma 8.11, the quotient ring H ∗ (X [n] )/I [n] is generated by the (n − 1)
classes a−(1n−k k) (1X )|0 (mod I [n] ), 2 ≤ k ≤ n. It follows that the cup product
a−ν (1X )|0 ∪ a−μ (1X )|0 (mod I [n] )
for an arbitrary partition ν  n can be obtained by repeatedly applying Theo-
rem 12.13. In other words, the structure constants of a−ν (1X )|0 ∪ a−μ (1X )|0
(mod I [n] ) should also be related to the Gromov-Witten theory of P1 .

12.6. Relation to the Hurwitz numbers of P1

In view of Theorem 8.12, Theorem 12.13 states that in H ∗ ((C2 )[n] ), we have

a−(1n−k k) (1C2 )|0 ∪ a−μ (1C2 )|0 = cλk,μ a−λ (1C2 )|0
λn
(λ)=(μ)+1−k

where cλk,μ is the coefficient of z k−1 in

(n − k)! k! zμ •
Fλ,μ (z).
ς(z)
In this section, we will establish a precise identity between some general structure
constants under the cup product in H ∗ ((C2 )[n] ) and certain Hurwitz numbers for
the projective line P1 .
The Hurwitz theory of a nonsingular curve C counts the number of covers of C
with specified ramification. The ramifications are determined by the profile of the
cover over the branch points. For Hurwitz theory, we will only consider covers
π:D→C
where D is a nonsingular (possibly disconnected) curve and π is dominant on each
component of D. Let n > 0 be the degree of π. The profile of π over a point c ∈ C
is the partition λ of n obtained from the multiplicities of π −1 (c). The profile of π
over c is the partition (1n ) if and only if π is unramified over c.
Let λ1 , . . . , λk be partitions of n assigned to k distinct points c1 , . . . , ck ∈ C.
A Hurwitz cover of C of genus g, degree n and monodromy λi at the point ci is a
morphism
π:D→C
such that
(i) D is a nonsingular (possibly disconnected) curve of genus g;
(ii) π has profile λi at the point ci ;
(iii) π is unramified over C − {c1 , . . . , ck }.
The Riemann-Hurwitz formula provides the genus constraint of a Hurwitz cover:

k
(12.53) 2g − 2 + n(2 − 2gC ) = (n − (λi ))
i=1
248 12. HILBERT/GROMOV-WITTEN CORRESPONDENCE

which is valid for both connected and disconnected Hurwitz covers. In disconnected
theory, the domain genus may be negative. Two Hurwitz covers π : D → C and
π  : D → C are defined to be isomorphic if there exists an isomorphism of curves
f : D → D satisfying π  ◦ f = π. Up to isomorphism, there are only finitely many
Hurwitz covers of C of genus g, degree n and monodromy λi at ci . Each cover π
has a finite group of automorphisms Aut(π). The Hurwitz number
(12.54) HnC (λ1 , . . . , λk )
is defined to be the weighted count of the distinct, possibly disconnected Hurwitz
covers π with the prescribed data. Each such cover is weighted by 1/|Aut(π)|. We
refer to [GJV] for more references on (single) Hurwitz numbers and some interesting
development which relates double Hurwitz numbers to Gromov-Witten theory.
The enumeration of Hurwitz covers of the projective line P1 is classically known
to be equivalent to multiplication in the class algebra of the symmetric group. Fix
(k + 1) partitions λ0 , λ1 , . . . , λk of n. Hurwitz covers with profile λi over ci ∈ P1
canonically yield (k + 1)-tuples of permutations (s0 , . . . , sk ), in the n-th symmetric
group Sn , defined up to conjugation satisfying:
(i) si is of cycle type λi for each i;
(ii) s0 s1 . . . sk = 1.
The elements si are determined by the monodromies of π around the points ci .
This leads to the following well-known lemma (e.g. [OP1, Subsection 0.3.3]).

Lemma 12.15. The Hurwitz number HnP (λ0 , . . . , λk ) is equal to the number of
1

(k +1)-tuples of permutations (s0 , . . . , sk ) in the n-th symmetric group Sn satisfying

the two conditions below divided by n!:
(i) si is of cycle type λi for each i;
(ii) s0 s1 . . . sk = 1.
Next, we consider the cup product a−λ1 (1C2 )|0 ∪ · · · ∪ a−λk (1C2 )|0 in the
0
cohomology ring of (C2 )[n] . Denote by cλλ1 ,...,λk the coefficient of a−λ0 (1C2 )|0 in

a−λ1 (1C2 )|0 ∪ · · · ∪ a−λk (1C2 )|0

when expressed as a linear combination of the Heisenberg monomial classes. By
cohomology degree reasons, we have

k
(12.55) (n − (λ0 )) = (n − (λi )).
i=1

Setting g = 1 − (λ0 ), we see that n, g, λ0 , λ1 , . . . , λk satisfy the identity (12.53).

Theorem 12.16. Let λ0 , λ1 , . . . , λk be partitions of n satisfying ( 12.55). Then,

k
zλi · HnP (λ0 , λ1 , . . . , λk ).
0 1
(12.56) cλλ1 ,...,λk =
i=1
0
Proof. Recall the notation ã−λ (1C2 ) from (12.41). Let c̃λλ1 ,...,λk be the coeffi-
cient of ã−λ0 (1C2 )|0 in
ã−λ1 (1C2 )|0 ∪ · · · ∪ ã−λk (1C2 )|0
 12.6. RELATION TO THE HURWITZ NUMBERS OF P1 249

when expressed as a linear combination of the Heisenberg monomial classes. Then

0 
k
1 0
cλλ1 ,...,λk = zλ i · · c̃λ1 k,

i=1
zλ0 λ ,...,λ
and (12.56) is equivalent to
c̃λλ1 ,...,λk = zλ0 · HnP (λ0 , λ1 , . . . , λk ).
0 1
(12.57)
Let Ψ1 , Ψ2 and Cλ be from (8.4), (8.11) and (8.9) respectively. By (8.5) and
(8.12), we obtain
Ψ1 Ψ2 (Cλ ) = ã−λ (1C2 )|0 .
0
By the assumption (12.55), we see from Theorem 8.3 that c̃λλ1 ,...,λk is equal to
the number of conjugacy class Cλ0 in the cup product of the conjugacy classes
Cλ1 , . . . , Cλk in the class algebra C(Sn ), denoted by

k
[Cλ0 ] C λi .
i=1
Since the conjugacy class Cλ0 consists of n!/zλ0 elements, we have

k
n! k
n! λ0
[C(1n ) ]Cλ0 C λi = · [Cλ0 ] C λi = · c̃ 1 k.

i=1
zλ 0 i=1
zλ0 λ ,...,λ
On the other hand, we conclude from Lemma 12.15 that
1 k
HnP (λ0 , λ1 , . . . , λk ) =
1
· [C(1n ) ]Cλ0 C λi .
n! i=1
Now (12.57) follows from putting the above together. 
 Part 5

Gromov-Witten theory of the

Hilbert schemes of points
 CHAPTER 13

Cosection localization for the

Hilbert schemes of points

The Heisenberg algebra of Grojnowski and Nakajima offers a basic tool for
writing down the cohomology classes of the Hilbert schemes of points on a sur-
face. The geometric representation (Proposition 3.16) of the Heisenberg mono-
mial classes provides a convenient way to analyze the Gromov-Witten invariants of
these Hilbert schemes. The study of the Gromov-Witten theory of the Hilbert
schemes of points on a surface began with the work [LQ1] where the 1-point
genus-0 extremal Gromov-Witten invariants are computed. The results in [LQ1]
give the first evidence to Ruan’s Cohomological Crepant Resolution Conjecture
[Rua2, Section 2]. In [ELQ], the genus-0 extremal Gromov-Witten invariants
of the Hilbert scheme of 3 points on the complex projective plane P2 are calcu-
lated. Motivated by the Gromov-Witten and Donaldson-Thomas correspondence
conjectured in [MNOP1, MNOP2], Okounkov and Pandharipande [OP3, OP4]
investigated the equivariant Gromov-Witten theory of the Hilbert schemes of points
on the affine plane C2 , and proved its equivalence with the local Gromov-Witten
theory of the 3-fold C2 × P1 . Using the technique of cosection localization and
the equivariant quantum corrected boundary operator of Okounkov and Pandhari-
pande [OP3], Jun Li and W.-P. Li [LL] determined the 2-point genus-0 extremal
Gromov-Witten invariants of the Hilbert scheme X [n] of points on an arbitrary
surface X.
In this chapter, we will address the technique of cosection localization developed
in [KL1,KL2], which provides a powerful tool for studying the virtual fundamental
cycles. We will apply it to the Gromov-Witten theory of the Hilbert schemes of
points on surfaces, and review the reduction results from [LL] when the images
of the stable maps to the Hilbert schemes are contracted by the Hilbert-Chow
morphism. When the smooth projective surface X has a non-trivial holomorphic 2-
form, we will prove some vanishing results regarding the Gromov-Witten invariants
of the Hilbert scheme X [n] . Moreover, we will apply the results to the Hilbert
scheme X [2] , and calculate its genus-1 Gromov-Witten invariant. Proposition 13.3
is from [LL], while Proposition 13.15, Theorem 13.19 and their proofs are from
[HLQ].

13.1. Cosection localization of Kiem and J. Li

In this section, we will recall from [KL1, KL2] the technique of cosection lo-
calization and its application to the Gromov-Witten theory. This technique was
used in [LL] to study stable maps to the Hilbert scheme X [n] , whose images in X [n]
are contracted by the Hilbert-Chow morphism to points in the symmetric product
253
254 13. COSECTION LOCALIZATION FOR THE HILBERT SCHEMES

X (n) . We will present the reduction results from [LL] regarding the corresponding
virtual fundamental cycles.
First of all, we review the technique of cosection localization. Following the con-
cept of a localized Gysin map, a localized virtual fundamental cycle is constructed
by Kiem and J. Li [KL2, Theorem 1.1].
Theorem 13.1. Let M be a Deligne-Mumford stack endowed with a perfect
obstruction theory. Suppose the obstruction sheaf ObM admits a surjective homo-
morphism σ : ObM |U → OU over an open U ⊂ M. Let
M(σ) = M − U.
be the degeneracy locus. Then (M, σ) has a localized virtual fundamental cycle

loc ∈ A∗ M(σ) .
[M]vir
This cycle has the usual properties of the virtual fundamental cycles, and relates to
the usual virtual fundamental cycle [M]vir via

loc ∈ A∗ M
[M]vir = ι∗ [M]vir
where ι : M(σ) → M is the inclusion map.
To apply the technique of cosection localization to Gromov-Witten theory, let
Y be a smooth quasi-projective variety with a non-trivial holomorphic 2-form θ̃ ∈
H 0 (Y, Ω2Y ). Let M = Mg,r (Y, β) be the moduli space of r-pointed genus-g stable
maps to Y with homology class β. The 2-form θ̃ on Y induces a cosection σ of the
obstruction sheaf on M as follows. Let f : C → Y and π : C → M be the universal
family of M. Let S be the Artin stack of genus-g connected nodal curves. Then
the relative obstruction sheaf of the standard relative obstruction theory of M/S is
ObM/S = R1 π∗ f ∗ TY .
Now regard the holomorphic 2-form θ̃ ∈ H 0 (Y, Ω2Y ) as an anti-symmetric homo-
morphism
(13.1) θ̂ : TY −→ ΩY , θ̂(v), v = 0.
Then θ̂ defines the first arrow in the following sequence of homomorphisms
(13.2) R1 π∗ f ∗ TY −→ R1 π∗ f ∗ ΩY −→ R1 π∗ ΩC/M −→ R1 π∗ ωC/M
where the second is induced by f ∗ ΩY → ΩC/M , and the third is induced by the
tautological map ΩC/M → ωC/M . Since R1 π∗ ωC/M ∼= OM , the composite of this
sequence provides
(13.3) σ rel : R1 π∗ f ∗ TY = ObM/S −→ OM .
The obstruction sheaf of M is the cokernel of p∗ TS → ObM/S where p : M → S is
the projection. Using the universal family f and R1 π∗ f ∗ TY = Ext1π (f ∗ ΩY , OC ), we
obtain the exact sequence
(13.4) Ext1π (ΩC/M , OC ) −→ Ext1π (f ∗ ΩY , OC ) −→ ObM −→ 0
where the first arrow is induced by f ∗ ΩY → ΩC/M . It can be proved that the
composition
σ rel
Ext1π (ΩC/M , OC ) −→ Ext1π (f ∗ ΩY , OC ) −→ OM
 13.1. COSECTION LOCALIZATION OF KIEM AND J. LI 255

is the zero homomorphism. Combining with (13.4), we obtain a cosection of the

obstruction sheaf ObM :
(13.5) σ: ObM −→ OM .
The vanishing criterion asserts that the degeneracy locus of σ consists of stable
maps [ϕ : C → Y ] satisfying
(13.6) θ̂(ϕ∗ TCreg ) = 0
where Creg denotes the smooth locus of the curve C (these stable maps are called
θ̃-null stable maps).
Next, we assume that X denotes a smooth projective complex surface with a
non-trivial holomorphic differential 2-form
θ ∈ H 0 (X, Ω2X ) = H 0 (X, OX (KX )).
Using θ, Beauville [Bea2, Section 5] (also [Bea1]) constructed a holomorphic 2-
form on the Hilbert scheme X [n] as follows. Let
X [n] ∗ = {ξ ∈ X [n] | | Supp(ξ)| ≥ n − 1},
(n)
X∗ = ρn (X [n] ∗ ),
= ηn−1 (X∗ ),
(n)
X∗n
Δn∗ = {(x1 , . . . , xi , . . . , xj , . . . , xn ) ∈ X∗n | xi = xj for some i < j}
where ρn : X [n] → X (n) is the Hilbert-Chow morphism, and ηn : X n → X (n) is the
quotient map. Let
πn : ' X∗n → X∗n
be the blowing-up morphism of X∗n along the big diagonal Δn∗ . Then we have a
commutative diagram of morphisms:
'
X n πn
−→ X n
⏐∗ ⏐∗
⏐ ⏐
(13.7) %η̃n %ηn
ρn (n)
X [n] ∗ −→ X∗
where η̃n : '
X∗n → X' n
∗ /Sn = X
[n]
∗ is the quotient map. Let pi : X → X be the
n
∗
n ∗ ' n
i-th projection. Then the holomorphic 2-form πn ( i=1 pi θ) on X ∗ is Sn -invariant,
[n]
and thus induces a holomorphic 2-form θ∗ on X ∗ . Note that X [n] − X [n] ∗ is a
[n]
[n]
codimension-2 closed subscheme of X [n] . By Hartogs’ Theorem, θ∗ extends to a
[n] [n]
holomorphic 2-form θ on X :
(13.8) θ [n] ∈ H 0 (X [n] , Ω2X [n] ).
Now apply the construction of the cosection (13.5) to θ̃ = θ [n] and Y = X [n] .
Put
M = Mg,0 (X [n] , dβn ).
Then, θ [n] defines a regular cosection of the obstruction sheaf ObM of M:
(13.9) σ : ObM → OM .
Moreover, if M(σ) ⊂ M denotes the degeneracy locus of points over which σ van-
ishes, then
(13.10) [M]vir ∈ H∗ (M(σ))
256 13. COSECTION LOCALIZATION FOR THE HILBERT SCHEMES

by Theorem 13.1. By abusing notations, denote a point [ϕ : C → X [n] ] ∈ M by

ϕ. By the vanishing criterion (13.6), σ vanishes at a point ϕ ∈ M if the image of
ϕ∗ : TCreg → TX [n] lies in the null space of
(13.11) θ [n] : TX [n] → ΩX [n] .
To understand such ϕ, note that [ϕ(C)] = dβn . Thus the composition ρn ◦ ϕ is a
constant map. Let
Spt : M → X (n)

be the induced map. If Spt(ϕ) = i=1 ni xi ∈ X (n) , then the morphism ϕ factors
through the product of punctual Hilbert schemes:


(13.12) ϕ = (ϕ1 , . . . , ϕ ) : C → Mni (xi ) ⊂ X [n]
i=1
where ϕi is a morphism from C to Mni (xi ). The collection
(13.13) ϕ = (ϕ1 , . . . , ϕ )
is defined to be the standard decomposition of ϕ, and the point xi is called the
support of ϕi . Note that the collection {ϕ1 , . . . , ϕ } is unique up to the ordering of
the ϕi ’s. The following lemma and its proof are from [LL, Lemma 3.2].
Lemma 13.2. Let Λθ ⊂ M = Mg,0 (X [n] , dβn ) be the subset consisting of the
points
ϕ = (ϕ1 , . . . , ϕ ) ∈ M
such that for each i, either ϕi is a constant map or its support xi = Spt(ϕi ) lies in
the vanishing locus of θ. Then,
M(σ) ⊂ Λθ .
Proof. Assume that ϕ = (ϕ1 , . . . , ϕ ) ∈ M lies in the degeneracy locus M(σ)
of σ. By the vanishing criterion (13.6), ϕ∗ (TCreg ) is contained in the null space of
θ [n] . Note that if ξ = ξ1 + . . . + ξ with ξi ∈ Mni (xi ), then


Tξ,X [n] ∼
= Tξi ,X [ni ] .
i=1

Thus, θ [n] is the direct sum of the forms θ [ni ] , and ϕ∗ (TCreg ) is contained in the null
space of θ [n] if and only if ϕi∗ (TCreg ) is contained in the null space of θ [ni ] for each i.
According to the work of Beauville, the form θ [ni ] is non-degenerate along Mni (x)
if x ∈ θ −1 (0). Applying this to the support xi of ϕi , we see that ϕ ∈ Λθ . 
Lemma 13.2 is sufficient if we have a regular section θ ∈ H 0 (X, OX (KX )). For
a general surface, we take a meromorphic section θ of OX (KX ). Let D0 and D∞ be
the vanishing and pole divisors of θ respectively. Viewed as a meromorphic section
of Ω2X , θ induces a meromorphic section θ [n] of Ω2X [n] and hence a meromorphic
homomorphism
σ : E → OM .
Here, E is a certain locally free sheaf admitting a surjection onto R1 π∗ f ∗ TX [n] ,
and f : C → X [n] and π : C → M are the universal family of M. Define the
degeneracy locus of σ to be the subset Deg(σ) ⊂ M consisting of all the points
ϕ ∈ M such that either σ is undefined or vanishes at ϕ. Adopting the proof of
 13.2. VANISHING OF GROMOV-WITTEN INVARIANTS 257

Lemma 13.2, one sees that Deg(σ) is contained in the subset of M consisting of
all those ϕ = (ϕ1 , . . . , ϕ ) ∈ M such that either for some i the support of ϕi is
contained in D0 ∪ D∞ , or for each i the map ϕi is constant.
vir
The following reduction result states that the virtual fundamental cycle M
is supported in a much smaller subset than Deg(σ).
Proposition 13.3. Let Λθ ⊂ M = Mg,0 (X [n] , dβn ) be the subset consisting of
the points
(13.14) ϕ = (ϕ1 , . . . , ϕ ) ∈ M
such that for each i, either ϕi is a constant map or the support xi = Spt(ϕi ) lies
vir
in D0 ∪ D∞ . Then M is supported in Λθ .
The proof of Proposition 13.3 is based on the observation that when ϕ ∈ M is
decomposed into maps into smaller Hilbert schemes as in (13.13), the obstruction
sheaf can be decomposed into the direct sum of factors. We refer to [LL, Section 3]
for the details.
Proposition 13.3 deals with stable maps whose images are contracted by the
Hilbert-Chow morphisms, and only requires the presence of a meromorphic section
θ of OX (KX ). Therefore, it is very useful in studying extremal Gromov-Witten
invariants of the Hilbert scheme of points on an arbitrary surface, as we will see
in Chapter 15. However, to apply the technique of cosection localization to handle
stable maps whose images are not necessarily contracted by the Hilbert-Chow mor-
phisms, we will have to assume the existence of a non-trivial holomorphic section
of OX (KX ). This leads us to the next section.

13.2. Vanishing of Gromov-Witten invariants when pg (X) > 0

In this section, using the technique of cosection localization, we will study the
stable maps to the Hilbert scheme X [n] , whose images in X [n] are not necessarily
contracted by the Hilbert-Chow morphism to points in the symmetric product X (n) .
When pg (X) = h0 (X, OX (KX )) > 0, we will prove some vanishing results regarding
the Gromov-Witten invariants of the Hilbert scheme X [n] .
Let X be a smooth projective complex surface. For simplicity, put
M = Mg,r (X [n] , β).
Assume that the surface X admits a non-trivial holomorphic 2-form
θ ∈ H 0 (X, Ω2X ) = H 0 (X, OX (KX )).
As we have seen in Section 13.1, θ induces a holomorphic 2-form θ [n] of the Hilbert
scheme X [n] which can also be regarded as a map θ [n] : TX [n] → ΩX [n] . In turn,
θ [n] induces a regular cosection
(13.15) σ: ObM −→ OM
of the obstruction sheaf ObM of M. The degeneracy locus
(13.16) M(σ)
of σ is the subset of M consisting of all the stable maps u : Γ → X [n] such that the
composition
(13.17) u∗ (θ [n] ) ◦ du : TΓreg → u∗ TX [n] |Γreg → u∗ ΩX [n] |Γreg
258 13. COSECTION LOCALIZATION FOR THE HILBERT SCHEMES

is trivial over the regular locus Γreg of Γ. Moreover, the cosection σ is surjective
away from the degeneracy locus M(σ), and there exists a localized virtual cycle
loc ∈ A∗ (M(σ)) such that
[M]vir
(13.18) loc ∈ A∗ (M)
[M]vir = ι∗ [M]vir
where ι : M(σ) → M stands for the inclusion map.
Our next lemma is parallel to Lemma 13.2. On one hand, it applies to an
arbitrary β ∈ H2 (X [n] ). On the other hand, it provides less information than
Lemma 13.2 does.
Lemma 13.4. Let C0 be the zero divisor of θ. Let u : Γ → X [n] be a stable map
in the degeneracy locus M(σ), and let Γ0 be an irreducible component of Γ with
non-constant restriction u|Γ0 . Then there exists ξ1 ∈ X [n0 ] for some n0 such that
Supp(ξ1 ) ∩ C0 = ∅ and
(13.19) u(Γ0 ) ⊂ ξ1 + ξ2 | Supp(ξ2 ) ⊂ C0 .
Proof. For notational convenience, we assume that Γ = Γ0 is irreducible.
Then there exist a nonempty open subset U ⊂ Γ and an integer n0 ≥ 0 such that
U is smooth and for every element p ∈ U , the image u(p) is of the form
(13.20) u(p) = ξ1 (p) + ξ2 (p)
where ξ1 (p) ∈ X with Supp(ξ1 (p)) ∩ C0 = ∅ and Supp(ξ2 (p)) ⊂ C0 . This
[n0 ]

induces a decomposition u|U = (u1 , u2 ) where the morphisms u1 : U → X [n0 ] and

u2 : U → X [n−n0 ] are defined by sending p ∈ U to ξ1 (p) and ξ2 (p) respectively.
Since (13.17) is trivial over the regular locus Γreg of Γ, the composite
(13.21) u∗1 (θ [n0 ] ) ◦ du1 : TU → u∗1 TX [n0 ] |U → u∗1 ΩX [n0 ] |U
is trivial. Note that the holomorphic 2-form θ [n0 ] on X [n0 ] is non-degenerate at
ξ1 (p), p ∈ U since Supp(ξ1 (p)) ∩ C0 = ∅. Thus, du1 = 0 and u1 is a constant
morphism. Setting ξ1 = ξ1 (p) = u1 (p), p ∈ U proves the lemma. 
In the rest of this chapter, we will assume that X is simply connected. By
(1.30),
Pic(X [n] ) ∼
= Pic(X) ⊕ Z · (Bn /2).
Under this isomorphism, the divisor DC ∈ Pic(X [n] ) corresponds to C ∈ Pic(X).
Let {α1 , . . . , αs } be a linear basis of H 2 (X, C). Then,
(13.22) {Dα1 , . . . , Dαs , Bn }
is a linear basis of H 2 (X [n] , C). Represent α1 , . . . , αs by real-surfaces C1 , . . . , Cs ⊂
X respectively. Then a linear basis of H2 (X [n] , C) is given by
(13.23) {βC1 , . . . , βCs , βn }
where βCi is defined by (2.1).
Lemma 13.5. Let the surface X be simply connected. Assume that C0,1 , . . . , C0,s
are the irreducible components (with reduced scheme structures) of the zero divi-
sor of θ. Let M = Mg,r (X [n] , β). If the degeneracy locus M(σ) from (13.16) is
nonempty, then
s
β= di βC0,i − dβn
i=1
for some integers d1 , . . . , ds ≥ 0 and d.
 13.2. VANISHING OF GROMOV-WITTEN INVARIANTS 259

Proof. Let u : Γ → X [n] be a stable map in M(σ), and let Γ1 , . . . , Γt be the

irreducible components of Γ. Then,

t
(13.24) β = u∗ [Γ] = deg(u|Γi ) [u(Γi )].
i=1

Assume that the restriction u|Γi is non-constant. Put

C0 = C0,1 ∪ · · · ∪ C0,s .
By Lemma 13.4, there exists ξ1 ∈ X [n0 ] for some n0 such that Supp(ξ1 ) ∩ C0 = ∅
and
u(Γi ) ⊂ ξ1 + {ξ2 | Supp(ξ2 ) ⊂ C0 }.
By Lemma 3.19,

s
[u(Γi )] = di βC0,i − d βn
i=1

for some integers d1 , . . . , ds ≥ 0 and d . Now our lemma follows immediately from
(13.24). 

Theorem 13.6. Let X be a simply connected smooth projective complex surface

admitting a holomorphic 2-form θ, and let C0,1 , . . . , C0,s be the irreducible compo-
nents (with reduced scheme structures) of the zero divisor of θ. If

s
β= di βC0,i − dβn
i=1

for some integers d1 , . . . , ds ≥ 0 and d, then all the Gromov-Witten invariants of

X [n] defined via the moduli space Mg,r (X [n] , β) vanish.

Proof. By Lemma 13.5, the degeneracy locus M(σ) from (13.16) is empty. It
follows from (13.18) that
[Mg,r (X [n] , β)]vir = 0.
Therefore, all the Gromov-Witten invariants of X [n] defined via the moduli space
Mg,r (X [n] , β) vanish. 

Corollary 13.7. Let X be a simply connected smooth projective complex sur-

face admitting a holomorphic 2-form with irreducible zero divisor. If β = d0 βKX −
dβn for some integer d and rational number d0 ≥ 0, then all the Gromov-Witten
invariants of X [n] defined via the moduli space Mg,r (X [n] , β) vanish.
Proof. Let θ ∈ H 0 (X, Ω2X ) = H 0 (X, OX (KX )) be the holomorphic 2-form
whose zero divisor C0 is irreducible (but possibly non-reduced). Then,
(13.25) KX = C0 = m(C0 )red
for some positive integer m, and the corollary follows from Theorem 13.6. 

Remark 13.8. From the proof of Corollary 13.7, we see that if m is from
(13.25), then the rational number d0 in Corollary 13.7 may be required to be of the
form d0 /m for some integer d0 ≥ 0.
260 13. COSECTION LOCALIZATION FOR THE HILBERT SCHEMES

Recall from Theorem 1.24 (ii) that KX [n] = DKX . Thus, if β = d0 βKX − dβn
for some rational number d0 ≥ 0 and integer d, then the expected dimension of
Mg,r (X [n] , β) is
(13.26) d = −KX [n] · β + (dim X [n] − 3)(1 − g) + r
= −d0 KX
2
+ (2n − 3)(1 − g) + r.
Our first corollary deals with the case when X is an elliptic surface.
Corollary 13.9. Let X be a simply connected (minimal) elliptic surface with-
out multiple fibers and with positive geometric genus. Let n ≥ 2 and β = 0. Then all
the Gromov-Witten invariants without descendant insertions defined via the moduli
space Mg,r (X [n] , β) vanish, except possibly when 0 ≤ g ≤ 1 and β = d0 βKX − dβn
for some integer d and rational number d0 ≥ 0.
Proof. Since X is a simply connected elliptic surface without multiple fibers,
KX = (pg − 1)f where pg ≥ 1 is the geometric genus of X and f denotes a smooth
fiber of the elliptic fibration. By Corollary 13.7, it remains to consider the case
when β = d0 βKX − dβn for some integer d and rational number d0 ≥ 0. By (13.26)
2
and KX = 0, the expected dimension of the moduli space Mg,r (X [n] , β) is equal to
d = (2n − 3)(1 − g) + r.
By the Fundamental Class Axiom, all the Gromov-Witten invariants without de-
scendant insertions are equal to zero if g ≥ 2. 
Our second corollary concentrates on the case when X is of general type.
Corollary 13.10. Let X be a simply connected minimal surface of general type
admitting a holomorphic 2-form with irreducible zero divisor. Let n ≥ 2 and β = 0.
Then all the Gromov-Witten invariants without descendant insertions defined via
Mg,r (X [n] , β) vanish, except possibly in the following cases
(i) g = 0 and β = dβn for some integer d > 0;
(ii) g = 1 and β = dβn for some integer d > 0;
(iii) g = 0 and β = d0 βKX − dβn for some integer d and rational number
d0 > 0.
Proof. In view of Corollary 13.7, it remains to consider the case when β =
d0 βKX − dβn for some integer d and rational number d0 ≥ 0.
When d0 = 0 and β = dβn with d > 0, we see from (13.26) that the expected
dimension of the moduli space Mg,r (X [n] , β) is equal to
d = (2n − 3)(1 − g) + r.
If g ≥ 2, then all the Gromov-Witten invariants without descendant insertions
defined via Mg,r (X [n] , β) vanish by the Fundamental Class Axiom.
Next, assume that d0 > 0. Since KX 2
≥ 1, we see from (13.26) that
d < (2n − 3)(1 − g) + r.
By the Fundamental Class Axiom, all the Gromov-Witten invariants without de-
scendant insertions vanish except possibly in the case when g = 0. 
In Section 13.4 below, for the Hilbert scheme X [2] where X is a simply con-
nected minimal surface of general type admitting a non-trivial holomorphic 2-form
 13.3. INTERSECTIONS ON MODULI SPACE OF GENUS-1 STABLE MAPS 261

with irreducible zero divisor, we will study the exceptional cases listed in Corol-
lary 13.10 (i), (ii) and (iii). The computation will involve certain integrals over
some moduli space of genus-1 stable maps. These integrals will be calculated in the
next section.

13.3. Intersections on some moduli space of genus-1 stable maps

In this section, we will compute certain intersection numbers on the moduli
space of genus-1 stable maps to P(V ) where V is a rank-2 vector bundle over a
smooth projective curve C. The results will be used in the next section.
We begin with some notations.
Notation 13.11. Let V be a rank-2 bundle over a smooth projective variety B.
(i) f denotes a fiber of the ruling π : P(V ) → B or its cohomology class.
(ii) H = (f1,0 )∗ ω is the rank-1 Hodge bundle over M1,0 (P(V ), df ) where ω is
the relative dualizing sheaf for the forgetful map
f1,0 : M1,1 (P(V ), df ) → M1,0 (P(V ), df ).
(iii) λ = c1 (H).
Let d ≥ 1. If u = [μ : D → P(V )] ∈ M1,0 (P(V ), df ), then μ(D) is a fiber of the
ruling π : P(V ) → B. Therefore, there exists a natural morphism
(13.27) φ: M1,0 (P(V ), df ) → B
whose fiber over b ∈ B is
M1,0 π −1 (b), d[π −1 (b)] ∼
= M1,0 (P1 , d[P1 ]).
So the moduli space M1,0 (P(V ), df ) is smooth (as a stack) with dimension
dim M1,0 (P1 , d[P1 ]) + dim(B) = 2d + dim(B).
By (12.3), the expected dimension of M1,0 (P(V ), df ) is 2d. Since d ≥ 1, the sheaf
R1 (f1,0 )∗ ev∗1 OP(V ) (−2) on M1,0 (P(V ), df ) is locally free of rank-2d. Moreover,
(13.28) λ2 = 0
in view of [Mum, (5.2)] regarding the Chern character of the Hodge bundles and
the proof of [FPa, Proposition 1].
Lemma 13.12. Let d ≥ 1. Let V be a rank-2 bundle over B0 × C where B0 and
C are smooth projective curves. Let Vb = V |{b}×C for b ∈ B0 . Then,
,
(13.29) λ · c2d R1 (f1,0 )∗ ev∗1 OP(Vb ) (−2)
[M1,0 (P(Vb ),df )]

is independent of the points b ∈ B0 .

Proof. This follows from the observation that (13.29) is equal to
,
φ∗ [{b} × C] · λ · c2d R1 (f1,0 )∗ ev∗1 OP(V ) (−2)
[M1,0 (P(V ),df )]

where the morphism φ : M1,0 (P(V ), df ) → B0 × C is from (13.27). 

Formula (13.30) below is probably well-known, but we could not find a refer-
ence.
262 13. COSECTION LOCALIZATION FOR THE HILBERT SCHEMES

Lemma 13.13. Let d be a positive integer. Then, we have

,
(13.30) c2d R1 (f1,0 )∗ ev∗1 OP1 (−2) = 0,
[M1,0 (P1 ,d[P1 ])]
,
1
(13.31) λ · c2d−1 R1 (f1,0 )∗ ev∗1 OP1 (−2) = − .
[M1,0 (P1 ,d[P1 ])] 12d
Proof. We begin with the proof of (13.30). Choose a K3 surface S which
contains a smooth rational curve C. Then, C 2 = −2, TS |C = OC (2) ⊕ OC (−2),
and dC is the only element in the complete linear system |dC|. So we have
,
c2d R1 (f1,0 )∗ ev∗1 OP1 (−2)
[M1,0 (P1 ,d[P1 ])]
,
= c2d R1 (f1,0 )∗ ev∗1 (TS |C )
[M1,0 (C,d[C])]
,
= c2d R1 (f1,0 )∗ ev∗1 TS .
[M1,0 (S,d[C])]

Note that R1 (f1,0 )∗ ev∗1 TS is a rank-2d bundle on the 2d-dimensional moduli space
M1,0 (S, d[C]) whose virtual dimension is 0. By Proposition 12.1,
,
c2d R1 (f1,0 )∗ ev∗1 OP1 (−2) = deg [M1,0 (S, d[C])]vir .
[M1,0 (P1 ,d[P1 ])]

Since [Mg,r (S, β)]vir = 0 whenever β = 0, we obtain

,
c2d R1 (f1,0 )∗ ev∗1 OP1 (−2) = 0.
[M1,0 (P1 ,d[P1 ])]

To prove (13.31), we apply (f1,0 )∗ ev∗1 to the exact sequence

0 → OP1 (−2) → OP1 (−1)⊕2 → OP1 → 0.
Since (f1,0 )∗ ev∗1 OP1 = OM1,0 (P1 ,d[P1 ]) and R1 (f1,0 )∗ ev∗1 OP1 = H∗ , we get the exact
sequence
0 → OM1,0 (P1 ,d[P1 ]) → R1 (f1,0 )∗ ev∗1 OP1 (−2)
→ R1 (f1,0 )∗ ev∗1 OP1 (−1)⊕2 → H∗ → 0.
Calculating the total Chern class and using (13.28), we see that
(13.32) c R1 (f1,0 )∗ ev∗1 OP1 (−2)
= c R1 (f1,0 )∗ ev∗1 OP1 (−1)⊕2 /c H∗
= c R1 (f1,0 )∗ ev∗1 OP1 (−1)⊕2 · (1 + λ).
Thus, the top Chern class c2d R1 (f1,0 )∗ ev∗1 OP1 (−2) is equal to
c2d R1 (f1,0 )∗ ev∗1 OP1 (−1)⊕2 + λ · c2d−1 R1 (f1,0 )∗ ev∗1 OP1 (−1)⊕2 .
By the Proposition 2 in [GP] and (13.30), we conclude that
,
1
(13.33) λ · c2d−1 R1 (f1,0 )∗ ev∗1 OP1 (−1)⊕2 = − .
[M1,0 (P1 ,d[P1 ])] 12d
By (13.32) again,
λ · c2d−1 R1 (f1,0 )∗ ev∗1 OP1 (−2) = λ · c2d−1 R1 (f1,0 )∗ ev∗1 OP1 (−1)⊕2 .
Combining this with (13.33), we obtain our formula (13.31). 
 13.3. INTERSECTIONS ON MODULI SPACE OF GENUS-1 STABLE MAPS 263

The following lemma deals with the case when V is a rank-2 bundle over P1 .
The main idea in its proof is to reduce to the special cases
OP1 ⊕ OP1 , OP1 (2) ⊕ OP1 (−1)
depending on the parity of the degree of V .
Lemma 13.14. Let d ≥ 1, and V be a rank-2 bundle over P1 . Then,
,
deg(V )
(13.34) λ · c2d R1 (f1,0 )∗ ev∗1 OP(V ) (−2) = .
[M1,0 (P(V ),df )] 12d

Proof. First of all, assume that deg(V ) = 2k for some integer k. Then V can
be deformed to
OP1 (k) ⊕ OP1 (k) = OP1 ⊕ OP1 ⊗ OP1 (k).
By Lemma 13.12,
,
λ · c2d R1 (f1,0 )∗ ev∗1 OP(V ) (−2)
[M1,0 (P(V ),df )]
, " #
= λ · c2d R1 (f1,0 )∗ ev∗1 OP1 ×P1 (−2) ⊗ OP1 ×P1 (−2kf ) .
[M1,0 (P1 ×P1 ,df )]

Note that
M1,0 (P1 × P1 , df ) ∼
= P1 × M1,0 (P1 , d[P1 ]),
Thus, we obtain
,
λ · c2d R1 (f1,0 )∗ ev∗1 OP(V ) (−2)
[M1,0 (P(V ),df )]
, " #
= π2∗ λ · c2d π2∗ R1 (f1,0 )∗ ev∗1 OP1 (−2) ⊗ π1∗ OP1 (−2k)
[P1 ×M1,0 (P1 ,d[P1 ])]

where π1 and π2 are the projection on P1 × M1,0 (P1 , d[P1 ]). Hence,
,
(13.35) λ · c2d R1 (f1,0 )∗ ev∗1 OP(V ) (−2)
[M1,0 (P(V ),df )]
,
= π2∗ λ · π2∗ c2d−1 R1 (f1,0 )∗ ev∗1 OP1 (−2) · π1∗ c1 OP1 (−2k)
[P1 ×M1,0 (P1 ,d[P1 ])]
deg(V )
=
12d
where we have used formula (13.31) in the last step.
Next, assume that deg(V ) = 2k+1 for some integer k. Then V can be deformed
to OP1 (2) ⊕ OP1 (−1) ⊗ OP1 (k). As in the previous paragraph, we have
,
λ · c2d R1 (f1,0 )∗ ev∗1 OP(V ) (−2)
[M1,0 (P(V ),df )]
, " #
= λ · c2d R1 (f1,0 )∗ ev∗1 OS (−2) ⊗ OS (−2kf )
[M1,0 (S,df )]
,
2k
= λ · c2d R1 (f1,0 )∗ ev∗1 OS (−2) +
[M1,0 (S,df )] 12d
264 13. COSECTION LOCALIZATION FOR THE HILBERT SCHEMES

where S = P OP1 (2) ⊕ OP1 (−1) . Let F1 be the blown-up of P2 at a point p, and
σ be the exceptional curve. Then, TF1 |σ ∼= OP1 (2) ⊕ OP1 (−1). So
,
(13.36) λ · c2d R1 (f1,0 )∗ ev∗1 OP(V ) (−2)
[M1,0 (P(V ),df )]
,
2k
= λ · c2d R1 (f1,0 )∗ ev∗1 OP(TF1 |σ ) (−2) +
[M1,0 (P(TF1 |σ ),df )] 12d
,
2k
= φ∗ [σ] · λ · c2d R1 (f1,0 )∗ ev∗1 OP(TF1 ) (−2) +
[M1,0 (P(TF1 ),df )] 12d
where the morphism φ : M1,0 P(TF1 ), df → F1 is from (13.27). Let f0 be a fiber
of the ruling F1 → P1 , and C be a smooth conic in P2 such that p ∈ C. We use C
to denote its strict transform in F1 . Then, [σ] = [C]/2 − [f0 ]. By (13.36),
,
λ · c2d R1 (f1,0 )∗ ev∗1 OP(V ) (−2)
[M (P(V ),df )]
,1,0
1
= · φ∗ [C] · λ · c2d R1 (f1,0 )∗ ev∗1 OP(TF1 ) (−2)
2 [M1,0 (P(TF1 ),df )]
,
2k
− φ∗ [f0 ] · λ · c2d R1 (f1,0 )∗ ev∗1 OP(TF1 ) (−2) +
[M1,0 (P(TF1 ),df )] 12d
,
1
= · λ · c2d R1 (f1,0 )∗ ev∗1 OP(TF1 |C ) (−2)
2 [M1,0 (P(TF1 |C ),df )]
,
2k
− λ · c2d R1 (f1,0 )∗ ev∗1 OP(TF1 |f0 ) (−2) + .
[M1,0 (P(TF1 |f0 ),df )] 12d
Note that deg TF1 |C = 6 and deg TF1 |f0 = 2. By (13.35),
,
λ · c2d R1 (f1,0 )∗ ev∗1 OP(V ) (−2)
[M1,0 (P(V ),df )]
1 6 2 2k
= · − +
2 12d 12d 12d
deg(V )
= .
12d
This completes the proof of the lemma. 
The following is the main result in this section. The ideas in its proof are
to deform the rank-2 vector bundle V over a curve C to a direct sum of two line
bundles over C, reduce the situation to a rank-2 bundle over P1 , and then apply
Lemma 13.14.
Proposition 13.15. Let d be a positive integer. Assume that V is a rank-2
vector bundle over a smooth projective curve C. Then, we have
,
deg(V )
(13.37) λ · c2d R1 (f1,0 )∗ ev∗1 OP(V ) (−2) = .
[M1,0 (P(V ),df )] 12d
Proof. It is well-known that there exist a rank-2 bundle V over P1 × C and
two points b1 , b2 ∈ P1 such that
V|{b1 }×C = V,
V|{b2 }×C = (OC ⊕ M ) ⊗ N
 13.4. GROMOV-WITTEN INVARIANTS OF THE HILBERT SCHEME X [2] 265

where M and N are line bundles on C with M being very ample. As in the proof
of Lemma 13.14, we conclude from Lemma 13.12 that
,
(13.38) λ · c2d R1 (f1,0 )∗ ev∗1 OP(V ) (−2)
[M1,0 (P(V ),df )]
,
2 deg(N )
= λ · c2d R1 (f1,0 )∗ ev∗1 OP(OC ⊕M ) (−2) + .
[M1,0 (P(OC ⊕M ),df )] 12d
Since M is very ample, there exists a morphism α : C → P1 such that M =
α∗ OP1 (1). Then,
OC ⊕ M = α∗ OP1 ⊕ OP1 (1) .
This induces an isomorphism
M1,0 (P(OC ⊕ M ), df ) ∼
= C ×P1 M1,0 P(OP1 ⊕ OP1 (1)), df .
Let
 : M1,0 (P(OC ⊕ M ), df ) → M1,0 P(OP1 ⊕ OP1 (1)), df
α
be the projection. Then
,
λ · c2d R1 (f1,0 )∗ ev∗1 OP(OC ⊕M ) (−2)
[M1,0 (P(OC ⊕M ),df )]
, " #
= ∗ λ · c2d α
α ∗ R1 (f1,0 )∗ ev∗1 O (−2)
P OP1 ⊕OP1 (1)
[M1,0 (P(OC ⊕M ),df )]
,
α) ·
= deg( λ · c2d R1 (f1,0 )∗ ev∗1 OP(OP1 ⊕OP1 (1)) (−2) .
[M1,0 (P(OP1 ⊕OP1 (1)),df )]

By Lemma 13.14 and noticing deg( α) = deg(α) = deg(M ), we get

,
deg(M )
(13.39) λ · c2d R1 (f1,0 )∗ ev∗1 OP(OC ⊕M ) (−2) = .
[M1,0 (P(OC ⊕M ),df )] 12d
Now our formula (13.37) follows immediately from (13.38) and (13.39). 

13.4. Gromov-Witten invariants of the Hilbert scheme X [2]

In this section, a continuation of Section 13.2, we will study the Gromov-Witten
invariants of the Hilbert scheme X [2] when X is a simply connected minimal surface
of general type admitting a non-trivial holomorphic 2-form with irreducible zero
divisor.
To begin with, we obtain the following from Corollary 13.10.
Proposition 13.16. Let X be a simply connected minimal surface of general
type admitting a non-trivial holomorphic 2-form with irreducible zero divisor. Let
β = 0. Then all the Gromov-Witten invariants without descendant insertions de-
fined via Mg,r (X [2] , β) vanish, except possibly in the following cases
(i) g = 0 and β = dβ2 for some integer d > 0;
(ii) g = 1 and β = dβ2 for some integer d ≥ 2;
(iii) KX2
= 1, g = 0 and β = βKX − dβ2 for some integer d.
Proof. Cases (i) and (ii) follow from Corollary 13.10 (i) and (ii) respectively.
In the case of Corollary 13.10 (iii), we have g = 0 and β = d0 βKX − dβ2 for some
266 13. COSECTION LOCALIZATION FOR THE HILBERT SCHEMES

rational number d0 > 0 and some integer d. We see from (13.26) that the expected
dimension of the moduli space M0,r (X [2] , β) is equal to
d = −d0 KX
2
+ 1 + r.
2
Since d0 KX must be a positive integer, we conclude from the Fundamental Class
Axiom that all the Gromov-Witten invariants without descendant insertions defined
via M0,r (X [2] , β) vanish except possibly when d0 KX
2
= 1. Now write KX = mC0

where C0 is an irreducible and reduced curve, and m ≥ 1 is an integer. By Re-
mark 13.8, d0 = d0 /m for some integer d0 ≥ 1. Therefore, we obtain
2
1 = d0 KX = d0 m(C0 )2 .
It follows that d0 = m = (C0 )2 = 1. Hence KX
2
= 1 and d0 = 1. 
Case (i) in Proposition 13.16 will be dealt with in Example 15.10. By the
Divisor Axiom, Case (ii) in Proposition 13.16 can be reduced to the invariant
X [2]
(13.40) 1 1,dβ2 .

Similarly, Case (iii) in Proposition 13.16 can be reduced to the invariant

X [2]
(13.41) 1 0, βKX −dβ2 .

The invariants (13.40) and (13.41) have been computed in [HLQ, Section 6].
In the rest of this section, we will present the calculation of (13.40) for an
arbitrary smooth projective surface X. Let d ≥ 2 be an integer. For simplicity, we
put
(13.42) Mg,r,d = Mg,r (X [2] , dβ2 ).
The next lemma determines the obstruction sheaf over the moduli space M1,0,d and
the corresponding virtual fundamental class.
Lemma 13.17. (i) The obstruction sheaf Ob = R1 (f1,0 )∗ (ev∗1 TX [2] ) over
the moduli space M1,0,d is locally free of rank 2d + 2;
(ii) [M1,0,d ]vir = c2d+2 (Ob) ∩ [M1,0,d ].
Proof. (i) Let ev1 : M1,1,d → X [2] and f1,0 : M1,1,d → M1,0,d be the evalu-
ation map and the forgetful map respectively. Let u = [μ : D → X [2] ] ∈ M1,0,d .
Then
H 1 f −1 (u), (ev∗ TX [2] )| −1 ∼ H 1 (D, μ∗ TX [2] ) = H 1 D, μ∗ (TX [2] |μ(D) ) .
=
1,0 1 f1,0 (u)

Since d ≥ 1, μ(D) = M2 (x) ∼

= P1 for some point x ∈ X. By Proposition 2.9 (ii),
TX [2] |μ(D) = Oμ(D) (2) ⊕ Oμ(D) (−2) ⊕ Oμ(D) ⊕ Oμ(D) .
Since the curve D is of genus 1,
h1 D, μ∗ (TX [2] |μ(D) ) = 2d + 2.
It follows that the sheaf R1 (f1,0 )∗ (ev∗1 TX [2] ) is locally free of rank 2d + 2.
(ii) First of all, note that there exists a natural morphism
φ : M1,0,d → X
sending an element u = [μ : D → X [2] ] ∈ M1,0,d to x ∈ X if μ(D) = M2 (x). The
fiber φ−1 (x) over x ∈ X is
∼ M1,0 (P1 , d[P1 ]).
M1,0 (M2 (x), d[M2 (x)] =
 13.4. GROMOV-WITTEN INVARIANTS OF THE HILBERT SCHEME X [2] 267

So the moduli space M1,0,d is smooth (as a stack) with dimension

dim M1,0,d = dim M1,0 (P1 , d[P1 ]) + 2 = 2d + 2.
By (13.26), the expected dimension of M1,0,d is 0. Thus, the excess dimension of
M1,0,d is 2d + 2. By (i) and Proposition 12.1,
[M1,0,d ]vir = c2d+2 (Ob) ∩ [M1,0,d ].
This completes the proof of the lemma. 
Via the inclusion map B2 → X [2] , the evaluation map ev1 : M1,1,d → X [2]
 1 : M1,1,d → B2 . Also, B2 ∼
factors through a morphism ev = P(TX ∗
). Let ρ :
B2 → X be the canonical projection. Then, there exists a commutative diagram of
morphisms:
1
ev
M1,1,d → B2
(13.43) ↓f1,0 ↓ρ
φ
M1,0,d → X.
The obstruction sheaf over the moduli space M1,0,d is further investigated in the
following lemma.
Lemma 13.18. (i) Let H be the Hodge bundle over M1,0,d . Then,
 ∗ TB ∼
R1 (f1,0 )∗ ev 1 = H ∗ ⊗ φ ∗ TX ;
2

(ii) There exists an exact sequence of locally free sheaves:

(13.44)  ∗1 TB2 → Ob → R1 (f1,0 )∗ ev
0 → R1 (f1,0 )∗ ev  ∗1 OB2 (−2) → 0.
Proof. (i) Let TB2 /X be the relative tangent sheaf for the projection ρ : B2 →
 ∗1 and (f1,0 )∗ to the exact sequence
X. Applying the functors ev
0 → TB2 /X → TB2 → ρ∗ TX → 0
of locally free sheaves, we obtain an exact sequence
(13.45)  ∗1 TB2 /X → R1 (f1,0 )∗ ev
R1 (f1,0 )∗ ev  ∗1 TB2
→ R1 (f1,0 )∗ ev ∗1 (ρ∗ TX ) → 0.
 ∗1 TB2 /X = 0 since f1,0 is of relative dimension 1.
where we have used R2 (f1,0 )∗ ev
∗
1
 1 TB2 /X = 0. Indeed, let u = [μ : D → X [2] ] ∈ M1,0,d ,
We claim that R (f1,0 )∗ ev
and assume that μ(D) = M2 (x). Since TB2 /X |M2 (x) = TM2 (x) = OM2 (x) (2),
−1
H 1 f1,0  ∗1 TB2 /X |f −1 (u) ∼
(u), ev = H 1 (D, μ∗ OM2 (x) (2)) = 0.
1,0

By (13.45),
 ∗1 TB2 ∼
R1 (f1,0 )∗ ev  ∗1 (ρ∗ TX ).
= R1 (f1,0 )∗ ev
 1 = φ ◦ f1,0 , we get
Since ρ ◦ ev
 ∗ TB ∼
R1 (f1,0 )∗ ev 1 2 = R1 (f1,0 )∗ f ∗ (φ∗ TX )
1,0
∼
= R1 (f1,0 )∗ OM1,1,d ⊗ φ∗ TX
∼
= H ∗ ⊗ φ ∗ TX .
 1 , we see from Lemma 13.17 (i) that
(ii) Since ev1 factors through ev
 ∗1 (TX [2] |B2 ) .
Ob = R1 (f1,0 )∗ (ev∗1 TX [2] ) = R1 (f1,0 )∗ ev
268 13. COSECTION LOCALIZATION FOR THE HILBERT SCHEMES

Since B2 is a smooth divisor in X [2] and OB2 (B2 ) = OB2 (−2), we have

0 → TB2 → TX [2] |B2 → OB2 (−2) → 0.

 ∗1 and (f1,0 )∗ , we obtain an exact sequence

Applying the functors ev

(13.46)  ∗1 OB2 (−2) → R1 (f1,0 )∗ ev

(f1,0 )∗ ev  ∗1 TB2 → Ob

 ∗1 OB2 (−2) → 0
→ R1 (f1,0 )∗ ev

 ∗1 TB2 = 0 since f1,0 is of relative dimension 1.

where we have used R2 (f1,0 )∗ ev
∗
 1 OB2 (−2) = 0. Indeed, let
We claim that (f1,0 )∗ ev

u = [μ : D → X [2] ] ∈ M1,0,d ,

and assume that μ(D) = M2 (x). Then, since OB2 (−2)|M2 (x) = OM2 (x) (−2),
−1
H 0 f1,0  ∗1 OB2 (−2)|f −1 (u) ∼
(u), ev = H 0 (D, μ∗ OM2 (x) (−2)) = 0.
1,0

 ∗1 OB2 (B2 ) = 0. It follows that the exact sequence (13.46) is simplified

So (f1,0 )∗ ev
to the exact sequence (13.44). 

Theorem 13.19. Let d ≥ 2. Let X be a smooth projective surface. Then,

2
X [2] KX
1 1,dβ2 = .
12d
Proof. By Lemma 13.17 (ii),

X [2]
1 1,dβ2 = deg[M1,0,d ]vir = deg c2d+2 (Ob) ∩ [M1,0,d ] .

 ∗1 OB2 (−2) on the moduli space M1,0,d are of

The Hodge bundle H and R1 (f1,0 )∗ ev
ranks 1 and 2d respectively. Therefore, by Lemma 13.18 and (13.28),

1 X [2]
1,dβ2  ∗1 OB2 (−2)
= c2 H∗ ⊗ φ∗ TX · c2d R1 (f1,0 )∗ ev
=  ∗1 OB2 (−2)
λ2 + φ∗ KX · λ + φ∗ c2 (TX ) · c2d R1 (f1,0 )∗ ev
=  ∗1 OB2 (−2) .
φ∗ KX · λ + φ∗ c2 (TX ) · c2d R1 (f1,0 )∗ ev

Let χ(X) be the Euler characteristic of X. By (13.30), we obtain

 ∗1 OB2 (−2)
φ∗ c2 (TX ) · c2d R1 (f1,0 )∗ ev
,
= χ(X) · c2d R1 (f1,0 )∗ ev∗1 OP1 (−2)
[M1,0 (P1 ,d[P1 ])]
= 0.

Hence,

1 X [2]
1,dβ2  ∗1 OB2 (−2) .
= φ∗ KX · λ · c2d R1 (f1,0 )∗ ev
 13.4. GROMOV-WITTEN INVARIANTS OF THE HILBERT SCHEME X [2] 269

Choose two smooth irreducible curves C1 and C2 satisfying [C1 ]−[C2 ] = KX . Since
B2 ∼
= P(TX∗
), we see that
1 X [2]
1,dβ2  ∗1 OB2 (−2)
= φ∗ ([C1 ] − [C2 ]) · λ · c2d R1 (f1,0 )∗ ev
,
= λ · c2d R1 (f1,0 )∗ ev∗1 OP(TX∗ |C1 ) (−2)
∗|
[M1,0 (P(TX C1 ),df )]
,
− λ · c2d R1 (f1,0 )∗ ev∗1 OP(TX∗ |C2 ) (−2)
∗|
[M1,0 (P(TX C2 ),df )]
∗ ∗
deg(TX |C1 ) deg(TX |C2 )
= −
12d 12d
2
KX
=
12d
where we have used Proposition 13.15 in the third step. 
Problem 13.20. Let n ≥ 3 and d ≥ 2. Let X be a smooth projective complex
[n]
surface. What is the genus-1 extremal Gromov-Witten invariant 1 X
1,dβn ?
 CHAPTER 14

Equivariant quantum operator of

Okounkov-Pandharipande

Motivated by the Gromov-Witten and Donaldson-Thomas correspondence con-

jectured in [MNOP1,MNOP2], Okounkov and Pandharipande [OP3,OP4] stud-
ied the equivariant Gromov-Witten theory of the Hilbert schemes of points on
the affine plane C2 . This chapter surveys their work on the equivariant quantum
corrected boundary operator, the equivariant quantum cohomology of the Hilbert
schemes of points in the affine plane C2 , and the quantum differential equation for
the Hilbert schemes of points. The equivariant quantum corrected boundary opera-
tor is the q-deformation of the quantum-mechanical Calogero-Sutherland operator,
parallel to those Calogero-Sutherland operators discussed in Section 7.4 for the
non-equivariant setting. We will determine the structure of the small equivariant
quantum cohomology of the Hilbert scheme (C2 )[n] , and see that the equivariant
quantum cohomology of (C2 )[n] is equivalent to the equivariant Gromov-Witten
theory of C2 × P1 . This equivalence is one edge of the tetrahedron of equivalences
among four theories: the (equivariant) quantum cohomology of (C2 )[n] , the (equi-
variant) Gromov-Witten theory of C2 × P1 , the (equivariant) Donaldson-Thomas
theory of C2 × P1 , and the (equivariant) quantum orbifold cohomology of the sym-
metric product (C2 )(n) . The quantum differential equation for the Hilbert schemes
of points is defined by the equivariant quantum corrected boundary operator, and
closely related to the Gromov-Witten theory, Donaldson-Thomas theory and Mac-
donald polynomials. Its behavior at the two special values q = 0 and q = −1
will be discussed. All the results in this chapter are from [OP3, OP4] which have
been generalized to the Hilbert scheme of points on An -resolutions by Maulik and
Oblomkov [Mau, MO1, MO2].

14.1. Equivariant quantum cohomology of the Hilbert scheme (C2 )[n]

In this section, we will review the equivariant quantum corrected boundary
operator of Okounkov and Pandharipande [OP3] which will be used in Sections 14.3
and 15.2. We will sketch the calculation of this important operator. Moreover, we
will see that the small equivariant quantum cohomology of the Hilbert scheme
(C2 )[n] of points in the affine plane is generated by the boundary divisor.
Define the Fock space F which is freely generated over C by commuting creation
operators α−k , k > 0, acting on the vacuum vector |0 . The annihilation operators
αk , k > 0 kill the vacuum αk |0 = 0, k > 0 and satisfy the commutation relations
(14.1) [αk , α ] = kδk,− Id.
Next, let
T = (C∗ )2
271
272 14. EQUIVARIANT QUANTUM OPERATOR

and HT∗ (pt) = C[t1 , t2 ]. Similar to Example 11.2, the 2-dimensional algebraic torus
T acts on C2 by
(14.2) (s1 , s2 ) · (u, v) = (s1 u, s2 v), s 1 , s 2 ∈ C∗
where u and v denote the coordinate functions of C2 . This action lifts to a T-action
on the Hilbert schemes (C2 )[n] . The study of the equivariant cohomology of the
Hilbert schemes (C2 )[n] in terms of the Heisenberg algebra action can be carried
out as in Section 11.1 and Section 11.2. The Heisenberg algebra is generated by the
operators pk ([Y ]) defined in (11.26) and (11.27), where Y denotes a T-invariant
closed subscheme of C2 . The analogue of Proposition 11.6 gives the following
commutation relations:
(14.3) [pk ([C2 ]), p ([x])] = −kδk,− Id.
It allows us to define a linear isomorphism of Fock spaces:

+∞
(14.4) Ψ: F ⊗C C[t1 , t2 ] → HT∗ (C2 )[n]
n=0
by putting
(14.5) Ψ α−k1 · · · α−kr |0 = p−k1 ([C2 ]) · · · p−kr ([C2 ])|0
for k1 , . . . , kr > 0.
Lemma 14.1. Let k > 0, and x denote the origin of C2 . Then via the iso-
morphism Ψ, the operators α−k and αk correspond to the operators p−k ([C2 ]) =
1/(t1 t2 ) · p−k ([x]) and −pk ([x]) respectively.
Proof. Note that [C2 ] = 1/(t1 t2 )·[x]. By (14.5), the operator α−k corresponds
to the operator p−k ([C2 ]) = 1/(t1 t2 )·p−k ([x]). In view of the commutation relations
(14.1) and (14.3), αk corresponds to −pk ([x]). 
For a partition λ with parts λi , put
1 
|λ = α−λi · |0 ,
zλ i
|λ([x]) = (t1 t2 )(λ) · |λ
where zλ is defined by (1.1). Under the identification (14.4), the elements |λ , λ 
n form a linear basis of the equivariant cohomology HT∗ (C2 )[n] of (C2 )[n] . By
Lemma 14.1, |λ([x]) corresponds to the class
1 
(14.6) p−λi ([x]) · |0 .
zλ i
For simplicity, we will denote the classes |λ and |λ([x]) by λ and λ([x]) respectively
if no confusion arises.
The standard inner product on the equivariant cohomology induces the follow-
ing nonstandard inner product on Fock space after an extension of scalars:
  (−1)|μ|−(μ) · δμ,ν
μ, ν = |μ , |ν = .
(t1 t2 )(μ) · zμ
The equivariant Gromov-Witten theory can be defined as in Section 12.1. Let
βn be the homology class of (C2 )[n] defined in (1.34). The T-fixed points ξλ of
(C2 )[n] are in one-to-one correspondence with the partitions λ  n. Throughout this
 14.1. EQUIVARIANT QUANTUM COHOMOLOGY OF (C2 )[n] 273

chapter, as in [OP3, OP4], we adopt the convention that ξλ is defined by the ideal
(12.12). By the localization theorem, the equivariant cohomology classes [ξλ ] form
a linear basis of the localized equivariant cohomology HT∗ (C2 )[n] ⊗C C(t1 , t2 ). The
locus of the stable maps in Mg,k ((C2 )[n] , dβn ) meeting a T-fixed point is compact.
The T-equivariant Gromov-Witten invariants of (C2 )[n] are defined by putting
@ A ,
[ξλ1 ], . . . , [ξλk ] = ev∗1 ([ξλ1 ]) ∪ · · · ∪ ev∗k ([ξλk ])
g,dβn [Mg,k ((C2 )[n] ,dβn )]vir

where ev1 , . . . , evk are the evaluation maps on Mg,k ((C2 )[n] , dβn ), and λ1 , . . . , λk
are partitions of n.
The following is [OP3, Theorem 1].
Theorem 14.2. Let w1 , w2 ∈ HT∗ (C2 )[n] ⊗C C(t1 , t2 ), and let M(q, t1 , t2 ) be
  k (−q)k + 1 1 (−q) + 1 
(14.7) (t1 + t2 ) − α−k αk
2 (−q)k − 1 2 (−q) − 1
k>0
1 
+ (t1 t2 αk+ α−k α− − α−k− αk α ).
2
k,>0
Then,
 [n]   
w1 , c1 OC2 , w2 0,dβn q d = w1 , M(q, t1 , t2 ) w2 .
d≥0
[n]
Note that c1 OC2 = −Bn /2 where Bn ⊂ (C2 )[n] is the boundary divisor
defined by (1.26). The operator M(q, t1 , t2 ) is known as the equivariant quantum
[n]
corrected boundary operator. Since βn · c1 OC2 = 1, Theorem 14.2 implies that
   @ A
d w1 , w2 0,dβ q d = w1 , M(q, t1 , t2 ) − M(0, t1 , t2 ) w2
n
d≥1

which computes the equivariant 2-point genus-0 extremal Gromov-Witten invari-

ants of (C2 )[n] . In view of Lemma 14.1, we have
(14.8)  t1 , t2 ) − M(0,
M(q, t1 , t2 ) − M(0, t1 , t2 ) = M(q,  t1 , t2 )
 t1 , t2 ) is defined to be
where the operator M(q,
 
t1 + t2  k (−q)k + 1 1 (−q) + 1
(14.9) − − p−k ([x])pk ([x]).
t1 t2 2 (−q)k − 1 2 (−q) − 1
k>0

The structure of the small equivariant quantum cohomology ring QHT∗ (C2 )[n]
of (C2 )[n] is given by the following [OP3, Corollary 1].
[n]
Corollary 14.3. The divisor class c1 OC2 generates the small equivariant
quantum cohomology ring QHT∗ (C2 )[n] of (C2 )[n] over C(q, t1 , t2 ).
Proof. Since the limiting operator
1   k (−q)k + 1 1 (−q) + 1 
−1
lim M(q, t, t ) = − α−k αk
t→∞ t 2 (−q)k − 1 2 (−q) − 1
k>0

is diagonal with distinct eigenvalues, M(q, t1 , t2 ) has distinct eigenvalues for generic
values of the parameters. By the Localization Theorem (Theorem 11.1), the equi-
variant cohomology ring HT∗ (C2 )[n] of (C2 )[n] is semisimple after localization.
274 14. EQUIVARIANT QUANTUM OPERATOR

Thus the equivariant quantum cohomology ring QHT∗ (C2 )[n] of (C2 )[n] is also
semisimple after localization. So

QHT∗ (C2 )[n] = C(q, t1 , t2 ) · wi
i

where {wi }i is the (finite) set of idempotents of QHT∗ (C2 )[n] satisfying wi · wj =
δi,j ·wi . Note that the idempotents wi are eigenvectors of quantum multiplication by
[n]
c1 OC2 . So they are eigenvectors under the action of the operator M(q, t1 , t2 ). The
element 1/n! · α−1

n
|0 represents the unit of QHT∗ (C2 )[n] . Since the unit is equal to
i wi , the action of M(q, t1 , t2 ) on 1/n!·α−1 |0 generates the n-eigenvalue subspace
n

of Fock space. Hence, c1 OC2 generates QHT∗ (C2 )[n] over C(q, t1 , t2 ).
[n]

In the rest of this section, we will outline the proof of Theorem 14.2. The first
step is to introduce a T-equivariant reduced obstruction theory for stable maps (of
positive degrees) to (C2 )[n] relative to the moduli space of pointed genus-g curves.
The new obstruction theory differs from the standard theory by a 1-dimensional
obstruction space L with c1 (L) = (t1 + t2 ). Therefore, when d > 0, we have
(14.10) [Mg,k ((C2 )[n] , dβn )]vir = (t1 + t2 ) · [Mg,k ((C2 )[n] , dβn )]vir
red

where [Mg,k ((C2 )[n] , dβn )]vir

red denotes the virtual fundamental cycle from the re-
duced theory. Formula (14.10) is the analogue of the reduction lemma (Lemma 13.2)
in the equivariant setting.
The second step in the proof of Theorem 14.2 is to establish some additive
property. For simplicity, denote the invariant w1 , · · · , wk 0,dβn by w1 , · · · , wk d .
By (14.10), when d > 0, we get
 [n]  [n]
μ([x]), c1 OC2 , ν d = (t1 + t2 ) · μ([x]), c1 OC2 , ν d
where (·, ·, ·)d on the right-hand-side denotes the reduced invariant of (C2 )[n] . If
[n]
(μ) = (ν), the integral μ([x]), c1 OC2 , ν d is a nonequivariant constant. Let

ξ= μi xi ∈ (C2 )(n)
i

where the points x1 , . . . , x(μ) ∈ C2 are distinct. By Proposition 3.16, we can

replace the equivariant class μ([x]) in the integrand by the nonequivariant class
1
μ(ξ) = · ρ−1 (ξ)
zλ n
where ρn is the Hilbert-Chow morphism (C2 )[n] → (C2 )(n) . Every rational curve
in (C2 )[n] is contracted by ρn . The moduli space of maps connecting the locus
μ(ξ) and the cycle ν = |ν is isomorphic to the moduli space of stable maps to the
product

(μ)
Mμi (xi )
i=1
if μ = ν and empty otherwise, where Mμi (xi ) ⊂ (C2 )[μi ] denotes the punctual
Hilbert scheme of C2 at xi . The moduli space of maps to the product,
⎛ ⎞

(μ)
(14.11) M0,3 ⎝ Mμi (xi ), d⎠
i=1
 14.2. EQUIVALENCE OF FOUR THEORIES 275

has components corresponding to the different distributions of the total degree d

among the factors. Consider a component
⎛ ⎞

(μ)
M[j, k] ⊂ M0,3 ⎝ Mμi (xi ), d⎠ ,
i=1

for which the degree splitting has at least two non-zero terms corresponding to
two different points xj and xk . The component M[j, k] has a standard obstruction
theory obtained from the standard obstruction theory of M0,3 ((C2 )[n] , dβn ). The
standard obstruction theory of M[j, k] has a 2-dimensional quotient obtained from
the 2-dimensional family of holomorphic symplectic forms (at xj and xk ). So we
obtain a doubly reduced obstruction theory by reducing the obstruction space by
the 2-dimensional quotient. The nonequivariant integral of the (singly) reduced
theory over such a component vanishes since the singly reduced theory contains an
additional 1-dimensional trivial factor. Therefore, the only components of (14.11)
[n]
which contribute to the integral μ([x]), c1 OC2 , μ d are those for which the degree
d is distributed entirely to a single factor of the product.
In summary, the following properties have been verified:
 [n] 
(i) μ, c1 OC2 , ν d = 0 if d > 0 and μ = ν;
 
(ii) μ, c1 OC2 , μ d = γn,d · (t1 t2 )−(μ) · (t1 + t2 ) for some γn,d ∈ Q;
[n]

(iii) the addition formula for d > 0,

   
[n]
μ, c1 OC2 , μ d  μi , c1 OC[n]2 , μi
d
(14.12) =
μ, μ i
μ i , μ i

where by abusing notations, the μi ’s on the right-hand-side denote the

partition with one part μi .
Note that the operator M(q, t1 , t2 ) also satisfies the corresponding properties, in-
cluding the addition formula for d > 0,
   
μ, M(q, t1 , t2 )μ d  μi , M(q, t1 , t2 )μi
d
(14.13) = .
μ, μ i
μi , μi
Now Theorem 14.2 follows from inductions on both n and d. We refer to
[OP3, Section 3] for further details.

14.2. Equivalence of four theories

As an application of the results in Section 14.1, Okounkov and Pandharipande
[OP3, Subsection 4.3] obtained an equivalence between the quantum cohomology of
the Hilbert scheme (C2 )[n] and the local Gromov-Witten theory of C2 × P1 . In this
section, we will survey this equivalence which leads to a tetrahedron of equivalences
among four theories.
To begin with, let (P1 , x1 , . . . , xr ) be the complex projective line with r distinct
marked points. Fix partitions λ1 , . . . , λr of n. Let
•
Mh (P1 , λ1 , . . . , λr )
denote the moduli space of (possibly disconnected) relative stable maps from genus
h curves to P1 with prescribed ramification λi at xi . The prescribed ramification
276 14. EQUIVARIANT QUANTUM OPERATOR

points on the domain are unmarked, and the maps are required to be nonconstant
on all connected components.
The partition function ZGW (C2 × P1 )n[P1 ],λ1 ,...,λr of the local Gromov-Witten
theory of C2 × P1 is defined to be
 ,
u 2h−2
· •
e − R• π∗ f ∗ (OP1 ⊕ OP1 )
h∈Z [Mh (P1 ,λ1 ,...,λr )]vir
•
where π and f are from the universal diagram for Mh (P1 , λ1 , . . . , λr ):
f
U
⏐ −→ P1
⏐
π%
•
Mh (P1 , λ1 , . . . , λr ).
The shifted generating function GW∗n (C2 × P1 )λ1 ,...,λr is defined to be
r
(λj )
(−iu)n(2−r)+ j=1 · ZGW (C2 × P1 )n[P1 ],λ1 ,...,λr .
(C2 )[n]
Fix a point ξ ∈ M0,r . Define the multipoint invariant λ1 , . . . , λr ξ to be

+∞ ,
qd · ev∗1 (λ1 ) ∪ · · · ∪ ev∗r (λr ) ∪ π̃ ∗ (ξ)
d=0 [M0,r ((C2 )[n] ,dβn )]vir

where π̃ : M0,r ((C2 )[n] , dβn ) → M0,r is the T-equivariant forgetful map, and
ev1 , . . . , evr are the evaluation maps on M0,r ((C2 )[n] , dβn ). The following corre-
spondence theorem [OP3, Theorem 2] relates the local Gromov-Witten theory of
(C2 )[n]
C2 × P1 to the genus-0 invariants λ1 , . . . , λr ξ of the Hilbert scheme (C2 )[n] .
Theorem 14.4. After the variable change eiu = −q,
(C2 )[n]
(14.14) GW∗n (C2 × P1 )λ1 ,...,λr = (−1)n · λ1 , . . . , λr ξ .
Proof. A direct comparison of the formulas of Theorem 14.2 and [BP, Theo-
rem 6.5] yields the result in case r = 3 and λ1 = (1n−2 2). Moreover, a verification
shows that the degeneration formula of local Gromov-Witten theory is compatible
via the correspondence (14.14) with the splitting formula for genus-0 fixed mod-
uli invariants of (C2 )[n] . So the correspondence (14.14) follows from the observa-
tion that by Corollary 14.3 and the reconstruction result in [BP, Appendix], both
sides of (14.14) are canonically determined by the 3-point case with one 2-cycle
(1n−2 2). 
It would be interesting to compare the present Hilbert/Gromov-Witten cor-
respondence Theorem 14.4 with the Hilbert/Gromov-Witten correspondence dis-
cussed in Chapter 12.
The Gromov-Witten and Donaldson-Thomas theories of C2 × P1 are related
by the correspondence conjectured in [MNOP1, MNOP2] and refined for the
equivariant context in [BP]. The proof of this equivalence has been completed in
[OP5].
In the context of the Crepant Resolution Conjecture, Bryan and Graber [BG]
proved that the equivariant quantum cohomology of the Hilbert scheme (C2 )[n]
is equivalent to the equivariant quantum orbifold cohomology of the symmetric
product (C2 )(n) . Related material will be presented in Chapter 16.
 14.2. EQUIVALENCE OF FOUR THEORIES 277

In summary, there exists a tetrahedron of equivalences among the (equivariant)

quantum cohomology of (C2 )[n] , the (equivariant) Gromov-Witten theory of C2 ×
P1 , the (equivariant) Donaldson-Thomas theory of C2 × P1 , and the (equivariant)
quantum orbifold cohomology of (C2 )(n) :

Quantum cohomology of (C2 )[n]

GW theory of C2 × P1 DT theory of C2 × P1

Quantum orbifold cohomology of (C2 )(n) .

In [PTse, Tse], Pandharipande and Tseng generalized this tetrahedron of

equivalences to the higher genus case:

Genus-g GW theory of (C2 )[n]

GW theory of (C2 × Cg,r )/Mg,r DT theory of (C2 × Cg,r )/Mg,r

Genus-g quantum orbifold theory of (C2 )(n)

where Cg,r denotes the universal curve over the moduli space Mg,r of genus-g stable
curves with r marked points, and (C2 × Cg,r )/Mg,r is treated as a family of 3-folds
parametrized by Mg,r . In establishing the above vertical equivalence between the
278 14. EQUIVARIANT QUANTUM OPERATOR

genus-g Gromov-Witten theory of (C2 )[n] and the genus-g quantum orbifold theory
of (C2 )(n) , a key role was played by the quantum differential equations for which
the genus-0 case is the subject of the next section.

14.3. The quantum differential equation of Hilbert schemes of points

This section covers the work of Okounkov and Pandharipande [OP4] on the
quantum differential equation for the Hilbert schemes of points in the affine plane.
The behavior at the special values q = 0 and q = −1, which are related to
the Gromov-Witten and Donaldson-Thomas theories respectively, will be stud-
ied. The residue at q = 0 is given in terms of the value of the equivariant
quantum corrected boundary operator M(q, t1 , t2 ) at q = 0, which is essentially
the quantum-mechanical Calogero-Sutherland operator. The appearance of the
Calogero-Sutherland operator here is the equivariant analogue of those Calogero-
Sutherland operators discussed in Section 7.4 in the non-equivariant setting. The
monodromy at q = −1 will be investigated, and the connection problem between
q = 0 and q = −1 will be solved via Macdonald polynomials.
The quantum differential equation (QDE) for the Hilbert scheme of points is
the differential equation
d
(14.15) q Ψ = M Ψ, Ψ ∈ F
dq
where the operator M = M(q, t1 , t2 ) is the equivariant quantum corrected boundary
operator defined by (14.7). An equivalent version is the differential equation
d
(14.16) q Ψ = M Ψ, Ψ∈F
dq
where the operator M = M(q, t1 , t2 ) is defined by
 k (−q)k + 1 1 
(t1 + t2 ) α−k αk + (t1 t2 αk+ α−k α− − α−k− αk α )
2 (−q) − 1
k 2
k>0 k,>0
t1 + t2  (−q) + 1
= M(q, t1 , t2 ) + α−k αk .
2 (−q) − 1
k>0

Since M commutes with the operator k>0 α−k αk , the equation splits into a direct
sum of finite-dimensional ODEs.
The quantum-mechanical Calogero-Sutherland operator
 2
1 ∂  1
(14.17) HCS = zi + θ(θ − 1)
2 i ∂zi i<j
|zi − zj |2

describes particles moving on the torus |zi | = 1 interacting via the potentials |zi −
zj |−2 . The parameter θ adjusts the strength of the interaction. The function

φ(z) = (zi − zj )θ
i<j

is an eigenfunction of HCS , and the operator φHCS φ−1 preserves the space of sym-
metric polynomials in the variables z1 , z2 , z3 , . . .. Identify the space of symmetric
functions in the variables z1 , z2 , z3 , . . . with the Fock space F via the linear map
pμ → zμ |μ
 14.3. QUANTUM DIFFERENTIAL EQUATIONS 279

where pμ is the power symmetric function defined in Section 1.2. Via this identifi-
cation, the operator φHCS φ−1 acts on the Fock space F.
A direct computation shows that the operator φHCS φ−1 is equal to
1−θ  1 
ΔCS = kα−k αk + (θαk+ α−k α− + α−k− αk α )
2 2
k>0 k,>0
 ∂
modulo scalars and a multiple of the momentum operator i zi ∂z i
. We have
(·)+1 −(·)
(14.18) M(0) = −t1 ΔCS |θ=−t2 /t1 t1
where (·) is the function mapping the basis |μ to (μ). This implies that the
behavior of (14.16) near the regular singularities q = 0, ∞ is described by the
Schrödinger equation for (14.17).
For Ψ in the n-th component of F, the equation (14.15) is a linear first order
ODE in P2 (n) unknowns, where P2 (n) is the number of partitions of n. In view of
(14.7), it has regular singularities. These are q = 0, ∞, and the solutions ξ of
(−ξ)k = 1, k = 2, . . . , n
excluding q = −1. Using L’Hospital’s rule, we see that for k > 0,
 
k (−q)k + 1 1 (−q) + 1
(14.19) lim − = 0.
q→−1 2 (−q)k − 1 2 (−q) − 1
Therefore, q = −1 is a nonsingular point of (14.15).
Two special values of q play a special role. These are q = −1 and q = 0, which
may be called the Gromov-Witten and Donaldson-Thomas points, respectively. On
one hand, the point q = 0 is a regular singularity of (14.15), and (14.18) relates
the residue of (14.15) at q = 0 to the Calogero-Sutherland operator. In particular,
the eigenfunctions of M(0) are, up to normalization, the Jack symmetric functions.
(α)
More precisely, recall the Jack symmetric function Pλ ∈ F defined in Section 1.2
with α = 1/θ. Then the eigenfunctions of M(0) are
|λ| (·) (α)
J λ = t2 t1 Pλ |α=−t1 /t2 .
Note that geometrically, M(0) is the equivariant boundary operator
 [n]
c1 OC2 ∪,
n

and up to normalization, Jλ corresponds to the class of a T-fixed point in the

equivariant cohomology of the Hilbert schemes of points in C2 (see Chapter 11 and
Section 12.2 for details). In particular, we see from (12.13) that the eigenvalue
corresponding to the eigenfunction Jλ is

−c(λ) = −c(λ; t1 , t2 ) = a ()t1 +  ()t2
∈Dλ
 
where a () and () are the arm colength and the leg colength respectively of the
cell  in the Young diagram Dλ , and are defined in Definition 1.4. By the general
theory of ODEs, there exists a solution
(14.20) Ψ = Yλ q −c(λ) , Yλ ∈ C[[q]]
of (14.15) which converges for |q| < 1 and satisfies
Yλ (0) = Jλ .
280 14. EQUIVARIANT QUANTUM OPERATOR

Let Y(q) denote the matrix formed by the vectors Yλ . Then

(14.21) Y(q) q −c(·)
is a fundamental solution to the QDE (14.15).
On the other hand, q = −1 is a nonsingular point of (14.15), and the mon-
odromy at this point is studied. Let
xtx
g(x, t) = , x>0
Γ(tx)
where Γ denotes the Gamma function. Let GGW = GGW (t1 , t2 ) and Γ = Γ(t1 , t2 )
be the diagonal operators defined by

GGW (t1 , t2 )|μ = g(μj , t1 )g(μj , t2 ) · |μ ,
j

(2πi)(μ)
Γ(t1 , t2 )|μ = GGW (t1 , t2 )|μ .
j μj
Define the connection
 
d
%Γ = Γ q − M(q, t1 , t2 ) Γ−1 .
dq
The monodromy at q = −1 is given by [OP4, Theorem 3].
Theorem 14.5. The monodromy of %Γ based at q = −1 is a Laurent polyno-
mial in
T1 = e2πit1 , T2 = e2πit2 .
Moreover, it is unitary with respect to the Hermitian form defined by
δμ,ν  μ /2 −μ /2 μ /2 −μ /2
μ, ν = (T1 i − T1 i )(T2 i − T2 i ) .
z(μ) i

Next, the connection problem for the QDE (14.15) between the special points
q = −1 and q = 0 is solved. The connection problem is to find the value of the
matrix Y(q) at q = −1. It is proved in [OP4] that up to normalization, the value
of the matrix Y(q) at q = −1 is given in terms of Macdonald polynomials with
parameters e2πit1 and e2πit2 . More precisely, let Pλ ∈ F ⊗ Q(q, t) be the monic
Macdonald polynomial as defined in [Mac2]. In studying the K-theory of the
Hilbert scheme, Haiman [Hai3] used the polynomial

Hλ (q, t) = tn(λ) 1 − q a() t−()−1 · ΥPλ (q, t−1 )
∈Dλ

where n(λ) = i (i − 1)λi , and Υ is the diagonal operator defined by

Υ|μ = (1 − t−μi )−1 · |μ .
i
Haiman’s parameters q and t are identified with the parameters T1 and T2 in The-
orem 14.5 respectively. Define the operator O(a) in K-theory by
O(a)Hλ = e−2πiac(λ) Hλ .
Let GDT = GDT (t1 , t2 ) be the diagonal operator defined by
 1
GDT (t1 , t2 )|λ = q −c(λ) · |λ
w
Γ(w + 1)
 14.3. QUANTUM DIFFERENTIAL EQUATIONS 281

where w runs over the exponents

 
t1 a() + 1 − t2 (), −t1 a() + t2 () + 1
∈Dλ
(i.e. the tangent weights) in (11.11) by replacing α and β there by t2 and −t1
respectively. Our settings (14.2) and (12.12) for this chapter are compared with
the settings (11.1) and (11.10) respectively.
The following is [OP4, Theorem 4].
Theorem 14.6. Let H be the matrix with columns Hλ . Then,
1
Γ−1 YGDT |q=−1 = O(1/2)H
(2πi)|·|
where | · | is the degree function mapping |μ to |μ|.
Finally, notice that
M(q −1 ) = −(−1)(·) M(q)(−1)(·) .
It follows that
(14.22) (−1)(·) Y(q −1 ) q c(·)
is another fundamental solution to the QDE (14.15). It is natural to ask for the
relation between the two fundamental solutions (14.21) and (14.22). Since, as
q → 0, ∞, the QDE becomes the Calogero-Sutherland system, the q → q −1 trans-
formation of the fundamental solution can be naturally interpreted as scattering by
the nonstationary terms. Moreover, since q = −1 is a fixed point of the involution
q → q −1 , the evaluation at q = −1 can be used to compare the two solutions. By
Theorem 14.6, the scattering transformation essentially amounts to the action of
the operator (−1)(·) .
 CHAPTER 15

The genus-0 extremal Gromov-Witten invariants

Motivated by Ruan’s Cohomological Crepant Resolution Conjecture [Rua2,

Section 2], [LQ1, LL, LQ5] studied the genus-0 extremal Gromov-Witten theory
of the Hilbert schemes of points on a smooth projective complex surface. This
chapter surveys the main results and ideas in these papers. We will begin with the
computation of 1-point genus-0 extremal Gromov-Witten invariants by calculating
the restriction of the obstruction sheaf to a Zariski open subset of the moduli space
of stable maps. Then, we will review the computations of J. Li and W.-P. Li
[LL] regarding the quantum corrected boundary operator and the 2-point genus-0
extremal Gromov-Witten invariants of the Hilbert schemes of points on surfaces.
The establishment of the quantum corrected boundary operator by J. Li and W.-
P. Li is one of the major steps toward the proof of Ruan’s Cohomological Crepant
Resolution Conjecture to be presented in Chapter 16. The main idea of [LL] is
to use the technique of cosection localization discussed in Section 13.1 and the
equivariant quantum corrected boundary operator of Okounkov and Pandharipande
discussed in Section 14.1. In addition, we will sketch the argument from [LQ5]
regarding the structure of the 3-point genus-0 extremal Gromov-Witten invariants
of the Hilbert scheme X [n] . Once again, the technique of cosection localization plays
an essential role. Moreover, the notion of Hilbert schemes of α-points from [LiJ,
Section 1] enables us to obtain finer analysis of the relevant virtual fundamental
cycle. Theorem 15.9 and its proof are from [LQ1]. The results of Section 15.2 are
from [LL], while we will refer to [LL] for the detailed proofs (except for the proof
of Proposition 15.16 which is slightly different).

15.1. 1-point genus-0 extremal Gromov-Witten invariants

In this section, as a warm-up to the results in later sections, we will compute
the 1-point genus-0 extremal Gromov-Witten invariants of the Hilbert scheme X [n]
for an arbitrary smooth projective complex surface X. The main idea is to find a
non-empty Zariski open (not necessarily dense) subset of the moduli space of stable
maps, over which the obstruction sheaf can be determined. The calculation of the
obstruction sheaf here is similar to what has been presented in Section 13.4.

15.1.1. Bases of H2 (X [n] ) and H 4n−4 (X [n] ). Let X be a smooth projective

complex surface. For simplicity, we will assume that X is simply connected. Recall
the homology classes βn , βC ∈ H2 (X [n] ) from (1.34) and (2.1) respectively. A basis
of the homology group H2 (X [n] ) is presented by Lemma 2.1 (i). In this subsection,
we will write down a basis of the cohomology group H 4n−4 (X [n] ).
283
284 15. THE GENUS-0 EXTREMAL GROMOV-WITTEN INVARIANTS


k 
k
Note from (3.11) that if n = ni and m = (2ni − 2 + |αi |), then
i=1 i=1

(15.1) a−n1 (α1 ) . . . a−nk (αk )|0 ∈ H m (X [n] ).

 ∈ H 2 (X [n] ). Then, we define the following classes:
Definition 15.1. Let α, α
sn,1 = a−1 (1X )a−1 (x)n−1 |0
sn,2 = a−2 (x)a−2 (x)a−1 (x)n−4 |0
sn,3 = a−3 (x)a−1 (x)n−3 |0
sα,1 = a−1 (α)a−2 (x)a−1 (x)n−3 |0
sα,2 = a−2 (α)a−1 (x)n−2 |0
sα,
α α)a−1 (x)n−2 |0
= a−1 (α)a−1 (
where x denotes the cohomology class of a point in X.
By (15.1), we have
α ∈H
4n−4
sn,1 , sn,2 , sn,3 , sα,1 , sα,2 , sα, (X [n] ).
To discuss the geometric interpretations of sn,1 , sn,2 , sn,3 , sα,1 , sα,2 , sα,
α , we recall
Proposition 3.16 and let PD stand for the Poincaré dual. Represent α, α  ∈ H 2 (X [n] )
 of X in general position. For x1 , x2 ∈
respectively by real-2-dimensional cycles C, C
X and η ∈ X [n−4]
satisfying x1 = x2 and x1 , x2 ∈ Supp(η), we have
(15.2) PD(sn,2 ) ∼ M2 (x1 ) + M2 (x2 ) + η.
For x ∈ X and η ∈ X [n−3]
with x ∈ C ∪ Supp(η) and Supp(η) ∩ C = ∅, we get
(15.3) PD(sn,3 ) ∼ M3 (x) + η,

(15.4) PD(sα,1 ) ∼ C + M2 (x) + η.

For a fixed η ∈ X [n−2] satisfying Supp(η) ∩ C = ∅, we have
(15.5) PD(sα,2 ) ∼ M2 (C) + η.
For η = x1 + . . . + xn−1 ∈ X [n−1] where x1 , . . . , xn−1 are distinct, we obtain
(15.6) PD(sn,1 ) ∼ “the closure of (X\ Supp(η)) + η in X [n] ”.
Alternatively, consider the following commutative diagram:
η
X ⊂ X⏐[n−1,n]
gn
−→ X [n]
⏐
⏐ ⏐
(15.7) % %(fn , ρ)
η×X ⊂ X [n−1] × X
where fn , gn are from Notation 1.30, (fn , ρ) is the blowing-up of X [n−1] × X along
Zn−1 by Theorem 1.31 (ii), and X η stands for the strict transform of η × X. So

Xη is isomorphic to the blown-up of X at the (n − 1) distinct points x1 , . . . , xn−1 .
Moreover,
gn |Xη : X η → gn (X η )
η ) is precisely the closure of (X\ Supp(η)) + η in the
is an isomorphism and gn (X
[n]
Hilbert scheme X . So in view of (15.6), we conclude that
(15.8) η ).
PD(sn,1 ) ∼ gn (X
 15.1. 1-POINT GENUS-0 EXTREMAL GROMOV-WITTEN INVARIANTS 285

η ) are
Note that the (n − 1) exceptional curves in the surface gn (X
(15.9) M2 (xi ) + (η\{xi }), i = 1, . . . , n − 1.
Finally, choose η ∈ X such that Supp(η) ∩ (C ∪ C)
[n−2]  = ∅. Then according to
Proposition 3.16, sα,
α is represented by the closure of the subset
(15.10)  and x = x̃} ⊂ X [n] .
{x + x̃ + η| x ∈ C, x̃ ∈ C,
Via the above notations, the following elementary lemma lists the basis elements
of the cohomology groups H 2 (X [n] ), H 4n−2 (X [n] ) and H 4n−4 (X [n] ).
Lemma 15.2. Let n ≥ 2 and X be simply connected. Let {α1 , . . . , αs } be a
basis of H 2 (X) represented by real-2-dimensional cycles {C1 , . . . , Cs } respectively.
Then,
(i) a basis of H 2 (X [n] ) consists of the cohomology classes
Bn , a−1 (αi )1−(n−1) |0 (i = 1, . . . , s).
(ii) a basis of H 4n−2 (X [n] ) consists of the cohomology classes
a−2 (x)a−1 (x)n−2 |0 , a−1 (αi )a−1 (x)n−1 |0 (i = 1, . . . , s).
(iii) a basis of H 4n−4 (X [n] ) consists of the cohomology classes
sn,1 , sn,2 , sn,3 , sαi ,1 (i = 1, . . . , s),
sαi ,2 (i = 1, . . . , s), sαi ,αj (i, j = 1, . . . , s).
Proof. Note that (i) and (ii) follow from Lemma 2.1. To prove (iii), fix a
point x ∈ X. Expand the basis {α1 , . . . , αs } of H 2 (X) to the basis
{α0 = 1X , α1 , . . . , αs , αs+1 = x}
∗
of H (X) = H (X) ⊕ H (X) ⊕ H 4 (X). By (15.1), a basis of H 4n−4 (X [n] ) consists
0 2

of the classes
(15.11) a−n1 (αm1 ) . . . a−nk (αmk )|0

k
satisfying ni ≥ 1, ni = n, and
i=1


k
(2ni − 2 + |αmi |) = 4n − 4.
i=1

Without loss of generality, we may assume that |αm1 | ≤ . . . ≤ |αmk |. We have


k
(−2 + |αmi |) = 2n − 4 ≥ 2k − 4.
i=1

Thus,

k
|αmi | ≥ 4k − 4.
i=1
Since X is simply connected, |αmi | ∈ {0, 2, 4} for every i. It follows that |αmi | = 4
for at least (k − 2) many i’s, i.e., αm3 = . . . = αmk = x. Then,
|αm1 | + |αm2 | = 2(n − k) + 4 ≥ 4.
286 15. THE GENUS-0 EXTREMAL GROMOV-WITTEN INVARIANTS

Assume that |αm2 | = 4, that is, αm2 = x. Then, |αm1 | = 2(n−k). If |αm1 | = 4,

n−2
then αm1 = x, k = n − 2, and ni = n. In this case, (15.11) is equal to either
i=1
sn,2 or sn,3 . If |αm1 | = 2, i.e., αm1 = αi for some 1 ≤ i ≤ s, then k = n − 1 and

n−1
ni = n. In this case, (15.11) is equal to either sαi ,1 or sαi ,2 . If |αm1 | = 0, then
i=1

n
αm1 = 1X , k = n, and ni = n. In this case, (15.11) is equal to sn,1 .
i=1
Finally, assume that |αm2 | ≤ 2. Since |αm1 | ≤ |αm2 | and
|αm1 | + |αm2 | = 2(n − k) + 4 ≥ 4,

n
we must have |αm1 | = |αm2 | = 2. So k = n and ni = n. In this case, (15.11) is
i=1
equal to sαi ,αj for some i and j satisfying 1 ≤ i, j ≤ s. 

15.1.2. The obstruction sheaf. In this subsection, we will construct a non-

empty Zariski open subset U0 of the moduli space M0,0 (X [n] , dβn ). Then we will
determine the restriction of the obstruction sheaf to the open subset U0 .
We begin with several notations.
Notation 15.3. Let σ : X n → X (n) be the natural quotient map. Put
X [n] ∗ = {ξ ∈ X [n] | | Supp(ξ)| ≥ n − 1},
(n)
X∗ = ρn (X [n] ∗ ),
= σ −1 (X∗ ),
(n)
X∗n
B = Bn = {ξ ∈ X [n] | | Supp(ξ)| < n},
B∗ = {ξ ∈ X [n] | | Supp(ξ)| = n − 1},
(n)
Xs∗ = ρn (B∗ ),
&
Δn∗ = σ −1 (ρn (B)) ∩ X∗n = Δi,j
n∗
1≤i<j≤n

where Δi,j
n∗ = {(x1 , . . . , xi , . . . , xj , . . . , xn ) ∈ X∗ | xi = xj } for 1 ≤ i < j ≤ n.
n

When we compute the 1-point Gromov-Witten invariants α 0,dβn , only X [n] ∗

is involved in most of the cases. Even though the Hilbert scheme X [n] is very
complicated, the open subset X [n] ∗ has a very simple description given below. Let
(15.12) πn : '
X∗n → X∗n
be the blowing-up morphism of X∗n along the big diagonal Δn∗ . The action of Sn
'
on X∗n lifts to an action on X n
∗ and

'
X [n] ∗ = X n
∗ /Sn .

'
Let σ̃ : X ∗ → X
n [n] i,j '
∗ be the quotient map. Let E∗ ⊂ X∗ be the exceptional locus
n
i,j
over Δn∗ . Consider the following morphisms:
(15.13) p1,2 n∗ −→ X,
: Δ1,2 (x, x, x3 , . . . , xn ) → x,
(n)
(15.14) j2 : Xs∗ −→ X, 2x + x3 + . . . + xn → x.
 15.1. 1-POINT GENUS-0 EXTREMAL GROMOV-WITTEN INVARIANTS 287

∗
Since the normal bundle of Δ1,2 n
n∗ in X∗ is isomorphic to p1,2 TX , we have

E∗1,2 ∼
= P(p∗1,2 TX
∗
).
The subgroup S2 × Sn−2 ⊂ Sn acts on Δ1,2 n∗ with the S2 -factor acting trivially on
Δ1,2 1,2 1,2
n∗ . The action of S2 × Sn−2 on Δn∗ lifts to an action on E∗ . It is easy to see
that
(n)
Xs∗ = Δ1,2n∗ /(S2 × Sn−2 )
and
B∗ = E∗1,2 /(S2 × Sn−2 ).
Regard p1,2 : Δ1,2n∗ → X as an S2 × Sn−2 -equivariant morphism where S2 × Sn−2
acts on X trivially. Then, S2 × Sn−2 acts on p∗1,2 TX ∗
, and the isomorphism E∗1,2 ∼
=
P(p∗1,2 TX
∗
) is S2 × Sn−2 -equivariant. So we get an isomorphism
j1 : B∗ = E∗1,2 /(S2 × Sn−2 ) ∼
= P(p∗1,2 TX
∗
)/(S2 × Sn−2 ) ∼
= P(j2∗ TX
∗
).
where the last isomorphism is due to the fact that the S2 -factor acts trivially on
p∗1,2 TX and the Sn−2 -factor commutes with the morphism p1,2 .
Next, we study OB∗ (B∗ ). Since
⎛ ⎞

σ̃ ∗ OX [n] ∗ (B∗ ) ∼
= OX [n] ∗ ⎝2 E∗i,j ⎠
1≤i<j≤n

and E∗i,j ∩ E∗1,2 = ∅ if and only i = 1 and j = 2, we conclude that

(15.15) (σ̃|E∗1,2 )∗ OB∗ (B∗ ) ∼
= σ̃ ∗ OX [n] ∗ (B∗ )|E∗1,2
∼
= O 1,2 (2E 1,2 )
E∗ ∗
∼
= OP(p∗ ∗ (−2)
1,2 TX )

where we have used the fact that OE∗1,2 (E∗1,2 ) ∼

= OP(p∗1,2 TX∗ ) (−1) via the isomorphism
∼ P(p∗ T ∗ ). Note that
E∗1,2 = 1,2 X

OP(p∗1,2 TX∗ ) (−2) = τ ∗ OP(j2∗ TX∗ ) (−2)

where τ : P(p∗1,2 TX
∗
) → P(j2∗ TX ∗
) is the natural morphism. Moreover, j1 ◦(σ̃|E∗1,2 ) = τ
1,2 ∼
via the isomorphism E∗ = P(p∗1,2 TX ∗
). Combining with (15.15), we obtain
" #
(σ̃|E∗1,2 )∗ OB∗ (B∗ ) ∼= (σ̃|E∗1,2 )∗ j1∗ OP(j2∗ TX∗ ) (−2) .
Since Pic(B∗ ) has no torsion, we conclude that
(15.16) OB (B∗ ) =∼ j ∗ OP(j ∗ T ∗ ) (−2).
∗ 1 2 X

Consider the open subset U0 of M0,0 (X [n] , dβn ) consisting of stable maps
[μ : D → X [n] ] such that μ(D) ⊂ X [n] ∗ . Similarly, take the open subset U1 of
M0,1 (X [n] , dβn ) consisting of stable maps [μ : (D; p) → X [n] ] such that μ(D) ⊂
−1
X [n] ∗ . Clearly U1 = f1,0 (U0 ). Let [μ : (D; p) → X [n] ] ∈ U1 . Since μ∗ (D) ∼ dβn , we
must have
μ(D) = M2 (x2 ) + x3 + . . . + xn
for some distinct points x2 , . . . , xn ∈ X. Hence μ(D) ⊂ B∗ . Moreover, the compos-
ite ρn ◦ ev1 sends the stable map [μ : (D; p) → X [n] ] to the point 2x2 + x3 + . . . + xn ,
which is independent of the marked point p on D. Hence ev1 induces a morphism φ
288 15. THE GENUS-0 EXTREMAL GROMOV-WITTEN INVARIANTS

 1 = ev1 |U1 and f˜1,0 = f1,0 |U1 , we have the following

from U0 to ρn (B∗ ). Putting ev
commutative diagram:
1 j1
U1
ev
→ B∗ ∼
= P(j2∗ TX
∗
)
˜
(15.17) ↓f1,0 ↓ρn ↓ π
φ (n) j2
U0 → ρn (B∗ ) = Xs∗ → X

where π : P(j2∗ TX
∗ (n)
) → Xs∗ is the natural projection of the P1 -bundle.
Note that the fiber φ−1 (2x2 +x3 +. . .+xn ) over a fixed point 2x2 +x3 +. . .+xn ∈
ρn (B∗ ) is simply
M0,0 M2 (x2 ) + x3 + . . . + xn , d[M2 (x2 ) + x3 + . . . + xn ]
which is isomorphic to the moduli space M0,0 (P1 , d[P1 ]) via the isomorphism
M2 (x2 ) + x3 + . . . + xn ∼
= P1 .
Hence the complex dimension of U0 is equal to
dim M0,0 (P1 , d[P1 ]) + 2(n − 1) = 2n − 3 + 2d − 1.
Since KX [n] ·dβn = 0, the expected dimension of M0,0 (X [n] , dβn ) is 2n−3 according
to (12.3) . Hence the excess dimension of U0 is e = (2d − 1).
Lemma 15.4. With notations as above, the restriction of the direct image sheaf
R1 (f1,0 )∗ (ev∗1 TX [n] ) to U0 is a locally free sheaf of rank (2d − 1).
Proof. Take a stable map u = [μ : D → X [n] ] in U0 , and consider
H 1 (f −1 (u), (ev∗ TX [n] )| −1 ) ∼
1,0 1 = H 1 (D, μ∗ TX [n] ).
f1,0 (u)

Since μ(D) = M2 (x2 ) + x3 + . . . + xn ∼

= P1 for some distinct points x2 , . . . , xn , we
have
(2n−2)
TX [n] |μ(D) = OP1 (2) ⊕ OP1 (−2) ⊕ OP1
by Proposition 2.9 (ii). Thus
H 1 (D, μ∗ TX [n] ) ∼
= H 1 (D, μ∗ OP1 (−2))
which has dimension equal to the excess dimension e = (2d − 1). Hence the direct
image sheaf R1 (f1,0 )∗ (ev∗1 TX [n] ) over U0 is locally free of rank (2d − 1). 

Suppose that M1 is a closed subset of M0,1 (X [n] , dβn ) contained in U1 and

M0 = f1,0 (M1 ) ⊂ U0 ⊂ M0,0 (X [n] , dβn ).
By Proposition 12.1 (i) and (iii), we have
(15.18) [M0,1 (Y, β)]vir |M1 = (f˜1,0 )∗ c2d−1 (R1 (f1,0 )∗ (ev1 )∗ TX [n] )|M0 .
Hence it is crucial to determine the sheaf V = R1 (f1,0 )∗ (ev1 )∗ TX [n] over U0 .
Lemma 15.5. Let V denote the restriction of R1 (f1,0 )∗ (ev1 )∗ TX [n] to U0 . Then,
(i) V ∼  1 )∗ OB∗ (B∗ ) ∼
= R1 (f˜1,0 )∗ (ev  1 )∗ OP(j2∗ TX∗ ) (−2).
= R1 (f˜1,0 )∗ (j1 ◦ ev
(ii) the locally free sheaf V sits in the exact sequence
(15.19) 0 → (j2 ◦ φ)∗ OX (−KX ) → V
 1 )∗ ((j2 ◦ π)∗ TX ⊗ OP(j2∗ TX∗ ) (−1)) → 0.
→ R1 (f˜1,0 )∗ (j1 ◦ ev
 15.1. 1-POINT GENUS-0 EXTREMAL GROMOV-WITTEN INVARIANTS 289

Proof. (i) Since ev1 (U1 ) ⊂ B∗ , we have

((ev1 )∗ TX [n] )|U1 = (ev
 1 )∗ TX [n] |B∗
∗

and
V = (R1 (f1,0 )∗ (ev1 )∗ TX [n] )|U0
= R1 (f˜1,0 )∗ ((ev1 )∗ TX [n] )|U 1

 1 )∗ TX [n] |B∗ .
= R1 (f˜1,0 )∗ (ev
Since B∗ is a smooth codimension-1 subvariety of X [n] , we obtain the exact sequence
(15.20) 0 → TB∗ → TX [n] |B∗ → OB∗ (B∗ ) → 0.
∗
 1 ) and (f˜1,0 )∗ to the exact sequence (15.20), we get
Applying (ev
 1 )∗ TB∗ → V → R1 (f˜1,0 )∗ (ev
R1 (f˜1,0 )∗ (ev  1 )∗ OB∗ (B∗ ) → 0
 1 )∗ TB∗ = 0 since f˜1,0 is of relative dimension 1.
where we have used R2 (f˜1,0 )∗ (ev
If [μ : D → X ] is a stable map in U0 , then
[n]

μ(D) = M2 (x2 ) + x3 + . . . + xn .
Hence the normal bundle of μ(D) in B∗ is trivial since μ(D) is a fiber of the P1 -
bundle P(j2∗ TX
∗ (n)
) over Xs∗ . Thus

TB∗ |μ(D) ∼
⊕(2n−2)
= Oμ(D) (2) ⊕ Oμ(D) .
Therefore,
H 1 (D, μ∗ TB∗ ) ∼
⊕(2n−2)
= H 1 (D, μ∗ (Oμ(D) (2) ⊕ Oμ(D) )) = 0,

 1 )∗ TB∗ = 0. So in view of (15.16), we obtain

and R1 (f˜1,0 )∗ (ev
V∼  1 )∗ OB∗ (B∗ ) ∼
= R1 (f˜1,0 )∗ (ev  1 )∗ OP(j2∗ TX∗ ) (−2).
= R1 (f˜1,0 )∗ (j1 ◦ ev
(ii) For simplicity, we denote P(j2∗ TX
∗
) by P. Consider the natural surjection
π ∗ (j2∗ TX
∗
) → OP (1) → 0.
The kernel of this surjection is a line bundle. By comparing the first Chern classes,
we get the following exact sequence:
(15.21) 0 → π ∗ OX (n) (j2∗ KX ) ⊗ OP (−1) → π ∗ (j2∗ TX
∗
) → OP (1) → 0.
s∗

Tensoring (15.21) with π ∗ OX (n) (−j2∗ KX ) ⊗ OP (−1), we get

s∗

(15.22) 0 → OP (−2) → (j2 ◦ π)∗ (TX

∗
⊗ OX (−KX )) ⊗ OP (−1)
∗
→ (j2 ◦ π) OX (−KX ) → 0.
∗
Note that TX ⊗ OX (−KX ) ∼  1 )∗ to (15.22) yields
= TX . Applying (j1 ◦ ev
(15.23)  1 )∗ OP (−2) → (j1 ◦ ev
0 → (j1 ◦ ev  1 )∗ ((j2 ◦ π)∗ TX ⊗ OP (−1))
 1 )∗ OX (−KX ) → 0.
→ (j2 ◦ π ◦ j1 ◦ ev
By (15.17), we have
 1 )∗ = (j2 ◦ φ ◦ f˜1,0 )∗ = (f˜1,0 )∗ ◦ (j2 ◦ φ)∗ .
(j2 ◦ π ◦ j1 ◦ ev
290 15. THE GENUS-0 EXTREMAL GROMOV-WITTEN INVARIANTS

So rewriting the 3rd term in the exact sequence (15.23), we obtain

(15.24)  1 )∗ OP (−2) → (j1 ◦ ev
0 → (j1 ◦ ev  1 )∗ ((j2 ◦ π)∗ TX ⊗ OP (−1))
→ (f˜1,0 )∗ ((j2 ◦ φ)∗ OX (−KX )) → 0.
Applying (f˜1,0 )∗ to the above exact sequence and using part (i), we have
0 → (f˜1,0 )∗ (f˜1,0 )∗ ((j2 ◦ φ)∗ OX (−KX )) → V
→ R1 (f˜1,0 )∗ (j1 ◦ ev 1 )∗ ((j2 ◦ π)∗ TX ⊗ OP (−1))
→ R1 (f˜1,0 )∗ (f˜1,0 )∗ ((j2 ◦ φ)∗ OX (−KX ))
where we have used
 1 )∗ ((j2 ◦ π)∗ TX ⊗ OP (−1)) = 0.
(f˜1,0 )∗ (j1 ◦ ev
Note that (f˜1,0 )∗ OU1 ∼ = OU0 and R1 (f˜1,0 )∗ OU1 = 0. So we get
(f˜1,0 )∗ (f˜1,0 )∗ ((j2 ◦ φ)∗ OX (−KX )) ∼ = (j2 ◦ φ)∗ OX (−KX ) ⊗ (f˜1,0 )∗ OU1
∼
= (j2 ◦ φ)∗ OX (−KX )
by the projection formula. Similarly,
R1 (f˜1,0 )∗ (f˜1,0 )∗ ((j2 ◦ φ)∗ OX (−KX )) = 0.
Therefore, the locally free sheaf V sits in the exact sequence (15.19). 
Remark 15.6. Fix distinct points x2 , . . . , xn on X. Via the isomorphism
φ−1 (2x2 + x3 + . . . + xn ) ∼ = M0,0 (P1 , d[P1 ]),
the restriction of
 1 )∗ (j2 ◦ π)∗ TX ⊗ OP (−1)
R1 (f˜1,0 )∗ (j1 ◦ ev
to φ−1 (2x2 + x3 + . . . + xn ) is isomorphic to
R1 (f1,0 )∗ (ev1 )∗ (OP1 (−1) ⊕ OP1 (−1))
where by abusing notations, we still use f1,0 and ev1 to denote the forgetful map
and the evaluation map from M0,1 (P1 , d[P1 ]) to M0,0 (P1 , d[P1 ]) and P1 respectively.
15.1.3. The 1-point genus-0 extremal Gromov-Witten invariants. In
this subsection, we will compute all the genus-0 1-point extremal Gromov-Witten
invariants γ 0,dβn for the Hilbert schemes X [n] . Recall from (12.4) and (12.3) that
|γ| = 4n − 4. In view of Lemma 15.2 (iii), we need only to compute γ 0,dβn when
γ is one of the cohomology classes
sn,1 , sn,2 , sn,3 , sα1 ,1 , sα1 ,2 , sα1 ,α2
where α1 , α2 ∈ H (X). These six cases will be treated in two lemmas.
2

Lemma 15.7. Let d ≥ 1, and α1 , α2 ∈ H 2 (X).

(i) If γ = sn,1 , sα1 ,α2 or sα1 ,1 , then γ 0,dβn = 0.
(ii) If γ = sα1 ,2 , then γ 0,dβn = 2 KX , α1 /d2 .
Proof. (i) Represent the cohomology classes α1 , α2 ∈ H 2 (X) respectively
by real-2-dimensional cycles C1 , C2 ⊂ X in general position. Fix distinct points
x1 , . . . , xn−1 ∈ X which are not contained in C1 ∪ C2 .
Let γ = sn,1 . By (15.8),
η ) ∼
PD(sn,1 ) ∼ gn (X =Xη
 15.1. 1-POINT GENUS-0 EXTREMAL GROMOV-WITTEN INVARIANTS 291

where Xη is the blow up of X along η = x1 + . . . + xn−1 . Moreover, the exceptional

η ) are
curves in gn (X
ρ−1
n (x1 + . . . + xi−1 + 2xi + xi+1 + . . . + xn−1 )

where 1 ≤ i ≤ n − 1. Let M1 be the subset of M0,1 (X [n] , dβn ) consisting of all

the stable maps [μ : (D; p) → X [n] ] such that μ(p) ∈ gn (X η ). In this case, μ(D)

is one of the exceptional curves in gn (Xη ) ⊂ B∗ . In particular, the stable maps
[μ : (D; p) → X [n] ] are contained in U1 , and
&
M1 = (f˜1,0 )−1 (φ−1 (x1 + . . . + xi−1 + 2xi + xi+1 + . . . + xn−1 )).
1≤i≤n−1

So as algebraic cycles, we have


n−1
[M1 ] = (f˜1,0 )∗ φ∗ [x1 + . . . + xi−1 + 2xi + xi+1 + . . . + xn−1 ].
i=1

By Lemma 15.5 (ii), we get c2d−1 (V) = −(j2 ◦ φ)∗ KX · c2d−2 (E) where
 1 )∗ ((j2 ◦ π)∗ TX ⊗ OP(j2∗ TX∗ ) (−1)).
E = R1 (f˜1,0 )∗ (j1 ◦ ev
In view of (12.5) and (15.18), we conclude that
,
γ 0,dβn = (ev1 )∗ γ
[M0,1 (Y,β)]vir

= [M1 ] · [M0,1 (Y, β)]vir

= [M1 ] · [M0,1 (Y, β)]vir |M1
= [M1 ] · (f˜1,0 )∗ (c2d−1 (V))

n−1 " #
= − (f˜1,0 )∗ φ∗ ([x1 + . . . + 2xi + . . . + xn−1 ] · j2∗ KX ) · c2d−2 (E)
i=1
= 0.
Next let γ = sα1 ,α2 . Assume that C1 and C2 intersect transversally at the
points y1 , . . . , ym . By (15.10), sα1 ,α2 is represented by the closure of
{x + x̃ + x1 + . . . + xn−2 | x ∈ C1 , x̃ ∈ C2 , and x = x̃} ⊂ X [n] .
Let M1 ⊂ M0,1 (X [n] , dβn ) consist of all the stable maps [μ : (D; p) → X [n] ] such
that μ(p) ∈ sC1 ,C2 . In this case,
ρn (μ(D)) = 2yk + x1 + . . . + xn−2
for some k with 1 ≤ k ≤ m. Therefore,
μ(D) = ρ−1
n (2yk + x1 + . . . + xn−2 ).

Hence the stable map [μ : (D; p) → X [n] ] is contained in U1 . So M1 ⊂ U1 is the

disjoint union of
(f˜1,0 )−1 (φ−1 (2yk + x1 + . . . + xn−2 )), 1≤k≤m
with ± orientations. By the same computations as in the previous paragraph, we
obtain γ 0,dβn = 0.
For the case of sα1 ,1 , the proof is similar to the cases of sn,1 and sα1 ,α2 .
292 15. THE GENUS-0 EXTREMAL GROMOV-WITTEN INVARIANTS

(ii) Let η̃ = x1 + . . . + xn−2 . By (15.5),

PD(sα1 ,2 ) ∼ M2 (C1 ) + η̃ = ρ−1
n (2C1 + η̃).

Thus, γ = PD(ρ−1
n (2C1 + η̃)). So we see from (15.18) and Lemma 15.5 (ii) that

(15.25) γ 0,dβn
,
= (ev1 )∗ γ
[M0,1 (Y,β)]vir
,
= (ev1 )∗ γ
−(f˜1,0 )∗ (j2 ◦φ)∗ KX ·(f˜1,0 )∗ c2d−2 (E)
,
= − (ev1 )∗ γ
 1 )∗ (ρ∗
(ev ∗ ˜ ∗
n j2 KX )·(f1,0 ) c2d−2 (E)
,
= −  1 )∗ PD(ρ−1
(ev n (2C1 + η̃) · c1 (OB∗ (B∗ )))
 1 )∗ (ρ∗
(ev ∗ ˜ ∗
n j2 KX )·(f1,0 ) c2d−2 (E)
, " " # #
= −  1 )∗ PD ρ−1
(ev n (j ∗
2 XK ) · (2C 1 + η̃) · c 1 (OB∗ (B ∗ ))
(f˜1,0 )∗ c2d−2 (E)
,
= 2 KX , C 1 ·  1 )∗ PD(ξ)
(ev
(f˜1,0 )∗ c2d−2 (E)

where ξ ∈ ρ−1 −1
n (2x + η̃) = ρn (2x + x1 + . . . + xn−2 ) is a fixed point for some fixed
point x ∈ C1 . Also, we have used the isomorphism (15.16) in the last step.
Let M1 ⊂ M0,1 (X [n] , dβn ) be the subset consisting of all the stable maps
[μ : (D; p) → X [n] ] with μ(p) = ξ. If [μ : (D; p) → X [n] ] ∈ M1 , then
ρn (μ(D)) = ρn (μ(p)) = 2x + x1 + . . . + xn−2 .
So μ(D) = ρ−1 ˜
n (2x + x1 + . . . + xn−2 ). Thus the restriction of the forgetful map f1,0
 −1
to M1 gives a degree-d morphism from M1 to φ (2x + x1 + . . . + xn−2 ). Put
M0 = φ−1 (2x + x1 + . . . + xn−2 ).
Now, as algebraic cycles, we have
(f˜1,0 )∗ [M1 ] = d[M0 ] = d · φ∗ [2x + x1 + . . . + xn−2 ].
By (15.25), we obtain
(15.26) γ 0,dβ2 = 2 KX , C1 · [M1 ] · (f˜1,0 )∗ c2d−2 (E)
= 2 KX , C1 · (f˜1,0 )∗ [M ] · c2d−1 (V)
1
∗
= 2d KX , C1 · φ [2x + x1 + . . . + xn−2 ] · c2d−2 (E)
= 2d KX , C1 · c2d−2 (E|φ−1 (2x+x1 +...+xn−2 ) ).
By Remark 15.6,
E|φ−1 (2x+x1 +...+xn−2 ) ∼
= R1 (f1,0 )∗ (ev1 )∗ (OP1 (−1) ⊕ OP1 (−1))
where f1,0 and ev1 denote the forgetful map and the evaluation map from the moduli
space M0,1 (P1 , d[P1 ]) to M0,0 (P1 , d[P1 ]) and P1 respectively. By the Theorem 9.2.3
in [CK],
c2d−2 (R1 (f1,0 )∗ (ev1 )∗ (OP1 (−1) ⊕ OP1 (−1))) = 1/d3 .
 15.1. 1-POINT GENUS-0 EXTREMAL GROMOV-WITTEN INVARIANTS 293

So we have
c2d−2 (E|φ−1 (2x+x1 +...+xn−2 ) )
= c2d−2 (R1 (f1,0 )∗ (ev1 )∗ (OP1 (−1) ⊕ OP1 (−1))
= 1/d3 .
Combining this with (15.26), we conclude that γ 0,dβn = 2 KX , C1 /d2 . 
Lemma 15.8. Let d ≥ 1. If γ = sn,2 or sn,3 , then γ 0,dβn = 0.
Proof. Since similar argument works for sn,2 , we will only prove the lemma
for sn,3 . So assume γ = sn,3 . Let x1 , . . . , xn−2 ∈ X be fixed distinct points on
X contained in a small analytic open subset U of X. We may assume that U is
independent of the smooth surface X. Let U1 ⊂ M0,1 (X [n] , dβn ) be the analytic
open subset consisting of all stable maps [μ : (D; p) → X [n] ] with μ(p) ∈ U [n] . Since
μ∗ (D) ∼ dβn , we see that Supp(μ(D)) = Supp(μ(p)) for
[μ : (D; p) → X [n] ] ∈ M0,1 (X [n] , dβn ).
So μ(D) ⊂ U [n] , and U1 is independent of X.
Next, recall from (15.3) that sn,3 is represented by M3 (x1 ) + x2 + . . . + xn−2 .
Let M1 ⊂ M0,1 (X [n] , dβn ) be the closed subset consisting of all stable maps
[μ : (D; p) → X [n] ] with μ(p) ∈ M3 (x1 ) + x2 + . . . + xn−2 . Then, M1 ⊂ U1 since
M3 (x1 ) + x2 + . . . + xn−2 ⊂ U [n] . In addition, since Supp(μ(D)) = Supp(μ(p)), we
must have
μ(D) ⊂ M3 (x1 ) + x2 + . . . + xn−2
for every [μ : (D; p) → X [n] ] ∈ M1 . So M1 is independent of X. Thus the pull-back
ev∗1 (γ) is also independent of X.
In summary, M1 ⊂ U1 , U1 is analytic open in M0,1 (X [n] , dβn ), and M1 and U1
are independent of X. It follows from the constructions of the virtual fundamental
cycle in [LT2, LT3] that the restriction [M0,1 (X [n] , dβn )]vir |M1 is independent of
the smooth surface X. So the 1-point Gromov-Witten invariant γ 0,dβn , which is
defined to be [M0,1 (X [n] , dβn )]vir · ev∗1 (γ) with ev∗1 (γ) being independent of X, is
independent of X as well. Since all the Gromov-Witten invariants γ1 , . . . , γk 0,β
with β = 0 for a K3 surface are zero, we conclude that γ 0,dβn = 0 for d ≥ 1. 
The following is the main result in this section.
Theorem 15.9. Let X be a simply connected smooth projective surface. Let
n ≥ 2, d ≥ 1, and α1 , α2 ∈ H 2 (X).
(i) If γ = sn,1 , sα1 ,α2 , sα1 ,1 , sn,2 or sn,3 , then
γ 0,dβn = 0.
(ii) If γ = sα1 ,2 , then
2
γ =
KX , α1 .
0,dβn
d2
Proof. Follows immediately from Lemma 15.7 and Lemma 15.8. 
Example 15.10. Let X be a simply connected smooth projective surface. Let
us consider all the genus-0 extremal Gromov-Witten invariants of X [2] :
(15.27) γ1 , . . . , γk 0,dβ2
294 15. THE GENUS-0 EXTREMAL GROMOV-WITTEN INVARIANTS

where k ≥ 0, d > 0, and γ1 , . . . , γk ∈ H ∗ (X [2] ) are homogeneous classes. By the

Fundamental Class Axiom (12.6) and the Divisor Axiom (12.7), we may assume
|γi | ≥ 4 for every i. In view of the degree constraint (12.4) and the expected
dimension (12.3), we have
1 
k
|γi | = − KX [2] · dβ2 + dim(X [2] ) − 3 + k = 1 + k.
2 i=1
It follows that k = 1. In other words, all the genus-0 extremal Gromov-Witten
invariants of X [2] are reduced to the 1-point genus-0 extremal Gromov-Witten in-
variants γ 0,dβn which is given by Theorem 15.9.
The above calculation of the 1-point genus-0 extremal Gromov-Witten invari-
ants of the Hilbert scheme X [n] follows a somewhat standard process. However, for
k ≥ 2, the determination of the k-point genus-0 extremal Gromov-Witten invari-
ants of X [n] requires completely new ideas. The case r = 2 is worked out in [LL]
which will be presented in the next section.

15.2. 2-point genus-0 extremal invariants of J. Li and W.-P. Li

In this section, X denotes a simply connected smooth projective complex sur-
face. We will survey the results in [LL] regarding the 2-point genus-0 extremal
Gromov-Witten invariants of the Hilbert scheme X [n] and the quantum corrected
boundary operator. We will refer to [LL] for the detailed proofs of these results ex-
cept for the proof of Proposition 15.16, which is presented here in a slightly different
way.
We begin with some notations. As in the paragraph after Lemma 13.2, we fix a
non-trivial meromorphic section θ of OX (KX ) whose vanishing and pole divisors are
denoted by D0 and D∞ respectively. Let (μ1 , μ2 , μ3 ) denote a triple of partitions
with |μ1 | + |μ2 | + |μ3 | = n. Let r = (μ1 ), s = (μ2 ) and t = (μ3 ). For cohomology
classes c1 , . . . , cs ∈ H 2 (X), define the class Aμc ∈ H ∗ (X [n] ) by

r 
s 
t
(15.28) Aμc = a−μ1i (x) · a−μ2j (cj ) · a−μ3k (1X )|0 .
i=1 j=1 k=1

Note from (3.11) that the cohomology degree of Aμc is given by


r 
s 
t
(15.29) |Aμc | = (2(μ1i − 1) + 4) + (2(μ2j − 1) + 2) + 2(μ3k − 1)
i=1 j=1 k=1

= 2(n + (μ ) − (μ )).1 3

By Proposition 3.16, up to a scalar multiple, the cohomology class Aμc is geometri-

cally represented by the closure Wcμ of the subset
(15.30) W (μ11 , q1 ; . . . ; μ1r , qr ; μ21 , C1 ; . . . ; μ2s , Cs ; μ31 , X; . . . ; μ3t , X)0
where q1 , . . . , qr are points and C1 , . . . , Cs are Riemann surfaces Poincaré dual to
c1 , . . . , cs ∈ H 2 (X) respectively such that
D0 , D∞ , q1 , . . . , qr , C1 , . . . , Cs
are in general position. Similarly, we have the class Aλe ∈ H ∗ (X [n] ) and its
geometric representation Weλ for another triple (λ1 , λ2 , λ3 ) of partitions and co-
homology classes e1 , . . . , eb ∈ H 2 (X) with |λ1 | + |λ2 | + |λ3 | = n, a = (λ1 ),
 15.2. THE 2-POINT GENUS-0 EXTREMAL GROMOV-WITTEN INVARIANTS 295

b = (λ2 ) and c = (λ3 ). The analogues to q1 , . . . , qr and C1 , . . . , Cs are denoted by

p1 , . . . , pa and E1 , . . . , Eb , where E1 , . . . , Eb are Riemann surfaces Poincaré dual to
e1 , . . . , eb ∈ H 2 (X) respectively. We assume that
D0 , D∞ , q1 , . . . , qr , C1 , . . . , Cs , p1 , . . . , pa , E1 , . . . , Eb
are in general position.
The goal is to calculate the 2-point genus-0 extremal Gromov-Witten invariants
 λ μ
Ae , Ac 0,dβn .
 λ μ
Assume that Ae , Ac 0,dβn = 0. By (12.3), the expected dimension of the moduli
space M0,2 (X [n] , dβn ) is 2n − 1. So
2(n + (λ1 ) − (λ3 )) + 2(n + (μ1 ) − (μ3 )) = 2(2n − 1)
in view of (12.4) and (15.29). Therefore, we must have
(15.31) ( (λ3 ) − (μ1 )) + ( (μ3 ) − (λ1 )) = 1.
 
Moreover, by Proposition 13.3, Aλe , Aμc 0,dβn = 0 implies that

(15.32) Λθ ∩ ev−1 λ −1
1 (We ) ∩ ev2 (Wc ) = ∅
μ

where evi is the i-th evaluation map from (12.2). A detailed analysis involving
(15.32), (15.30) and the definition of Λθ in Proposition 13.3 leads to the following.
 
Proposition 15.11. Let d ≥ 1. Assume that Aλe , Aμc 0,dβ = 0. Then,
n

3 1
(15.33) (λ ) = (μ ) + δ
(15.34) 3
(μ ) = (λ1 ) + (1 − δ)
where δ = 0 or 1. If δ = 0, then λ3 = μ1 , and there exists an integer = μ3i = λ2j
for some i and j such that the partition λ1 is obtained from μ3 with deleted, and
the partition μ2 is obtained from λ2 with deleted.
λ−λ2j 2
Let Ae be the cohomology class in H ∗ (X [n−λj ] ) obtained from Aλe with the
λ−λ1j λ−λ2i −λ2j
factor a−λ2j (ej ) deleted. We similarly define Ae , Ae , etc.

Lemma 15.12. Let notations be as in Proposition 15.11 with δ = 0. Then there

exists a universal function c,d depending only on and d such that
 λ μ   λ−λ2 μ−μ3 
(15.35) Ae , Ac 0,dβ = Ae j , Ac i · KX , ej · c,d .
n
=μ3i =λ2j
 
In fact, the proof of Proposition 15.11 shows that Aλe , Aμc 0,dβn is equal to
  λ−λ2j μ−μ3i   
(15.36) Ae , Ac · a− (ej )|0 , a− (1X )|0 0,dβ .

=μ3i =λ2j

By the Lemma 4.3 in [LL], there exists a universal function c,d depending only on
and d such that for every e ∈ H 2 (X), we have
 
(15.37) a− (e)|0 , a− (1X )|0 0,dβ = KX , e · c,d .


Combining (15.36) and (15.37) yields formula (15.35).

296 15. THE GENUS-0 EXTREMAL GROMOV-WITTEN INVARIANTS

By the Divisor Axiom (12.7) and Theorem 1.24 (i), we have

 λ [n]  [n]  
Ae , c1 OX , Aμc 0,dβ = c1 OX · dβn · Aλe , Aμc 0,dβ
 λ μ
n n

= − Bn /2 · dβn · Ae , Ac 0,dβn
 
= d · Aλe , Aμc 0,dβn
if d ≥ 1. Hence we obtain the following corollary from Lemma 15.12.
Corollary 15.13. Let λ and μ be as in Proposition 15.11 with δ = 0. Then
 [n] 
Aλe , c1 OX , Aμc 0,dβn q d
d≥0
 [n]    λ−λ2j μ−μ3i  
= Aλe , c1 OX , Aμc + Ae , Ac · KX , e j · d c,d q d .
1≤i≤t d≥1
1≤j≤b
μ3 =λ2 =
i j

Next, recall the notation τk∗ from Definition 3.9. Consider the operator
(15.38) M(q)
  k (−q)k + 1 1 (−q) + 1

= − a−k ak (τ2∗ KX )
2 (−q)k − 1 2 (−q) − 1
k>0
1 
− a−k a− ak+ + a−k− ak a (τ3∗ [X])
2
k,>0
  k(−q)k q
  k−1
= − a−k ak (τ 2∗ K X ) − a−k ak (τ2∗ KX )
(−q)k − 1 1 + q 2
k>0 k>0
1 
− a−k a− ak+ + a−k− ak a (τ3∗ [X]).
2
k,>0

By (4.19),
M(0) = d.
In view of this and Theorem 15.17 below, M(q) is defined to be the quantum
corrected boundary operator. The operator M(q) is closely related to the operator
M(q, t1 , t2 ) introduced in Theorem 14.2.
To compute M(q)(Aμc ), it suffices to calculate
(15.39) a−k ak (τ2∗ KX )(Aμc ),
(15.40) a−k a− ak+ (τ3∗ 1X )(Aμc ),
(15.41) a−k− ak a (τ3∗ 1X )(Aμc )
where k, > 0. Using Lemma 3.12, we see that a−k ak (τ2∗ KX )(Aμc ) is equal to
 μ−μ2
 μ−μ3
(15.42) (−μ2i )a−k (KX ci )Ac i + (−μ3i )a−k (KX )Ac i .
1≤i≤s 1≤i≤t
μ2 =k μ3 =k
i i
 
Proposition 15.14. If Aλe , M(q)(Aμc ) is not a constant function of q, then
(15.43) (λ3 ) = (μ1 ) + δ
(15.44) (μ3 ) = (λ1 ) + (1 − δ)
 15.2. THE 2-POINT GENUS-0 EXTREMAL GROMOV-WITTEN INVARIANTS 297

where δ = 0 or 1. If δ = 0, then λ3 = μ1 , and there exists an integer = μ3i = λ2j

for some i and j such that the partition λ1 is obtained from μ3 with deleted, and
the partition μ2 is obtained from λ2 with deleted.

Proposition 15.14 is parallel to Proposition 15.11. While the proof of Proposi-

tion 15.11 is geometric, the proof of Proposition 15.14 is algebraic, and follows from
(15.42) and the Heisenberg commutation relation (3.9). The following is parallel to
Corollary 15.13.

Corollary 15.15. Let the partitions

 λ and μ be as in Proposition 15.14 with
δ = 0. Then Aλe , M(q) − M(0) (Aμc ) is equal to
  λ−λ2 μ−μ3   
(−q) q
Ae j
, Ac i
· KX , ej · (−1)  2
− .
1≤i≤t
(−q) − 1 1 + q
1≤j≤b
μ3 =λ2 =
i j

Corollary 15.13 and Corollary 15.15 will be connected via the complex projec-
tive plane P2 . To achieve this, let X = P2 , T = (C∗ )3 and HT∗ (pt) = C[t1 , t2 , t3 ].
The group T acts on a point [z1 , z2 , z3 ] ∈ X by

(s1 , s2 , s3 ) · [z1 , z2 , z3 ] = [s1 z1 , s2 z2 , s3 z3 ]

for (s1 , s2 , s3 ) ∈ T. There are three T-fixed points:

q1 = [1, 0, 0], q2 = [0, 1, 0], q3 = [0, 0, 1].

At q1 , under the identification [z1 , z2 , z3 ] = [1, z2 /z1 , z3 /z1 ], T acts as

(s1 , s2 , s3 ) · (z2 /z1 , z3 /z1 ) = (s−1 −1

1 s2 z2 /z1 , s1 s3 z3 /z1 ).

Let Ni be the normal bundle of qi in X. Then as a T-module,

N1 ∼
= T1−1 T2 ⊕ T1−1 T3
where Ti is the 1-dimensional representation given by (s1 , s2 , s3 ) → si . Similarly,

N2 ∼
= T2−1 T1 ⊕ T2−1 T3 ,
N3 ∼
= T3−1 T1 ⊕ T3−1 T1 .

Arguments parallel to (11.22) show that for 1 ≤ i, j ≤ 3, we have


(15.45) [qi ] · [qj ] = δi,j [qi ] · (tk − ti ).
k∈{1,2,3}−{i}

Let L1 ⊂ X be the line passing through q2 and q3 , L2 ⊂ X be the line passing

through q1 and q3 , and L3 ⊂ X be the line passing through q1 and q2 . The line L3
is defined by the equation z3 /z1 = 0 near q1 and the equation z3 /z2 = 0 near q2 .
Therefore by the localization theorem, we have
(t2 − t1 )[q1 ] (t1 − t2 )[q2 ]
(15.46) [L3 ] = +
(t2 − t1 )(t3 − t1 ) (t1 − t2 )(t3 − t2 )
[q1 ] [q2 ]
= +
t3 − t1 t3 − t2
298 15. THE GENUS-0 EXTREMAL GROMOV-WITTEN INVARIANTS

in HT∗ (X) = HT∗ (X) ⊗C[t1 ,t2 ,t3 ] C(t1 , t2 , t3 ). Similarly, we obtain
[q2 ] [q3 ]
(15.47) [L1 ] = + ,
t1 − t2 t1 − t3
[q1 ] [q3 ]
(15.48) [L2 ] = + ,
t2 − t1 t2 − t3
[q1 ] [q2 ] [q3 ]
(15.49) [X] = + + .
(t2 − t1 )(t3 − t1 ) (t1 − t2 )(t3 − t2 ) (t1 − t3 )(t2 − t3 )
Using −KX = [L1 ] + [L2 ] + [L3 ] and the above formulas, we see that −KX is equal
to
(t2 + t3 − 2t1 )[q1 ] (t1 + t3 − 2t2 )[q2 ] (t1 + t2 − 2t3 )[q3 ]
(15.50) + + .
(t2 − t1 )(t3 − t1 ) (t1 − t2 )(t3 − t2 ) (t1 − t3 )(t2 − t3 )
Let Y1 , . . . , Yk be T-equivariant subvarities of X. Put
an1 · · · ank ([Y1 × . . . × Yk ]) = an1 ([Y1 ]) · · · ank ([Yk ]).
Proposition 15.16. Let L be a line in the projective plane X = P2 . Then,
 [n] 
(15.51) a−n ([L])|0 , c1 OX , a−n ([X])|0 0,dβ q d
n
d≥0
 
= a−n ([L])|0 , M(q)(a−n ([X])|0 ) .
[n]
Proof. Note that M(0) = d = c1 OX ∪. Thus, we have
 [n]   
a−n ([L])|0 , c1 OX , a−n ([X])|0 = a−n ([L])|0 , M(0)(a−n ([X])|0 ) .
Therefore (15.51) is equivalent to the following:
 [n] 
(15.52) a−n ([L])|0 , c1 OX , a−n ([X])|0 0,dβn q d
d≥1
 
= a−n ([L])|0 , (M(q) − M(0))(a−n ([X])|0 ) .
Recall from Section 11.2 that the equivariant Heisenberg operators are denoted by
pi (α). Take L = L3 . For d ≥ 1, we have
 [n] 
a−n ([L3 ])|0 , c1 OX , a−n ([X])|0 0,dβ
  n

= d a−n ([L3 ])|0 , a−n ([X])|0 0,dβn

 
= d p−n ([L3 ])|0 , p−n ([X])|0 0,dβ .
n
 [n] 
By (15.46) and (15.49), a−n ([L3 ])|0 , c1 OX , a−n ([X])|0 0,dβn is equal to
F  
[q1 ] [q2 ]
d p−n + |0 ,
t3 − t1 t3 − t2
  G
[q1 ] [q2 ] [q3 ]
p−n + + |0 .
(t2 − t1 )(t3 − t1 ) (t1 − t2 )(t3 − t2 ) (t1 − t3 )(t2 − t3 ) 0,dβn
 
A simple geometric argument shows that p−n ([qi ])|0 , p−n ([qj ])|0 0,dβ = 0 if
 [n]  n

i = j. Thus for d > 0, a−n ([L3 ])|0 , c1 OX , a−n ([X])|0 0,dβn is equal to
d  
p−n ([q1 ])|0 , p−n ([q1 ])|0 0,dβ
(t2 − t1 )(t3 − t1 ) 2 n
 15.2. THE 2-POINT GENUS-0 EXTREMAL GROMOV-WITTEN INVARIANTS 299

d  
(15.53) + p−n ([q2 ])|0 , p−n ([q2 ])|0 0,dβn .
(t1 − t2 )(t3 − t2 ) 2

 
The term p−n ([qi ])|0 , p−n ([qi ])|0 0,dβ has been computed by Theorem 14.2,
n
(14.8) and (14.9):
  
(15.54) d p−n ([qi ])|0 , p−n ([qi ])|0 0,dβn q d
d≥1
 [n] 
= p−n ([qi ])|0 , c1 OC2 , p−n ([qi ])|0 0,dβn q d
d≥1
 
= p−n ([qi ])|0 , M  i (0) p−n ([qi ])|0
 i (q) − M

where the operators M i (q) are defined by

  k (−q)k + 1 1 (−q) + 1 
 1 (q) = − (t2 + t3 − 2t1 )
M − p−k ([q1 ])pk ([q1 ]),
(t2 − t1 )(t3 − t1 ) 2 (−q)k − 1 2 (−q) − 1
k>0
  k (−q)k + 1 1 (−q) + 1 
 2 (q) = − (t1 + t3 − 2t2 )
M − p−k ([q2 ])pk ([q2 ]),
(t1 − t2 )(t3 − t2 ) 2 (−q)k − 1 2 (−q) − 1
k>0
 
 (t1 + t2 − 2t3 )  k (−q)k + 1 1 (−q) + 1
M3 (q) = − − p−k ([q3 ])pk ([q3 ]).
(t1 − t3 )(t2 − t3 ) 2 (−q)k − 1 2 (−q) − 1
k>0

By (15.53) and (15.54),

 [n] 
(15.55) a−n ([L])|0 , c1 OX , a−n ([X])|0 0,dβ q d
n
d≥1
1  
= p−n ([q1 ])|0 , M  1 (0) p−n ([q1 ])|0
 1 (q) − M
(t2 − t1 )(t3 − t1 ) 2

1  
+  2 (q) − M
p−n ([q2 ])|0 , M  2 (0) p−n ([q2 ])|0 .
(t1 − t2 )(t3 − t2 ) 2

On the other hand, by (15.50), p−k pk (τ2∗ KX ) is equal to

(t2 + t3 − 2t1 )
(15.56) − p−k ([q1 ])pk ([q1 ])
(t2 − t1 )(t3 − t1 )

(t1 + t3 − 2t2 ) (t1 + t2 − 2t3 )

− p−k ([q2 ])pk ([q2 ]) − p−k ([q3 ])pk ([q3 ])
(t1 − t2 )(t3 − t2 ) (t1 − t3 )(t2 − t3 )
noting that τ2∗ ([qi ]) = [qi ] × [qi ]. By (15.46) and (15.49),
 
a−n ([L])|0 , (M(q) − M(0))(a−n ([X])|0 )
 
= p−n ([L3 ])|0 , (M(q) − M(0))(p−n ([X])|0 )
F   
[q1 ] [q2 ] [q1 ]
= p−n + |0 , (M(q) − M(0))p−n
t3 − t1 t3 − t2 (t2 − t1 )(t3 − t1 )
 G
[q2 ] [q3 ]
+ + |0 .
(t1 − t2 )(t3 − t2 ) (t1 − t3 )(t2 − t3 )
300 15. THE GENUS-0 EXTREMAL GROMOV-WITTEN INVARIANTS

Combining this with (15.38) and (15.56), we conclude that

 
a−n ([L])|0 , (M(q) − M(0))(a−n ([X])|0 )
1  
=  1 (q) − MM
p−n ([q1 ])|0 , M  1 (0) p−n ([q1 ])|0
(t2 − t1 )(t3 − t1 ) 2

1  
+  2 (q) − M
p−n ([q2 ])|0 , M  2 (0) p−n ([q2 ])|0 .
(t1 − t2 )(t3 − t2 )2
In view of (15.55), we have proved the desired formula (15.52). 

The following determines all the 2-point genus-0 extremal Gromov-Witten in-
variants of X [n] .
Theorem 15.17. Let X be a simply connected smooth projective complex sur-
face. Then
 [n]   
(15.57) Aλe , c1 OX , Aμc 0,dβn q d = Aλe , M(q)Aμc .
d≥0

We now sketch the proof of Theorem 15.17, and refer to [LL, Section 7] for the
details. Since  [n]
M(0) = d = c1 OX ∪,
n

Proposition 15.11 and Proposition 15.14 reduce (15.57) to the case when λ3 = μ1
and (μ3 ) = (λ1 ) + 1. By Corollary 15.13 and Corollary 15.15, the proof of (15.57)
is further reduced to the proof of the following identity:
  
(−q) q
(15.58) d c,d q d = (−1) 2 − .
(−q) − 1 1 + q
d≥1

To prove (15.58), consider the special case when X = P2 , Aλe = a− ([L]) and
Aμc = a− ([X]) where L denotes a line in X. By Corollary 15.13,
 [n]   [n]  
Aλe , c1 OX , Aμc 0,dβ q d = Aλe , c1 OX , Aμc − 3 d c,d q d .
n
d≥0 d≥1

Combining with Proposition 15.16, we conclude that


−3 d c,d q d
d≥1
 
= a− ([L])|0 , M(q) − M(0) (a− ([X])|0 )
 
(−q) q  
= − a− ([L])|0 , a− a (τ2∗ KX )(a− ([X])|0 )
(−q) − 1 1 + q

 
(−q) q  
= 3 − a− ([L])|0 , a− ([L])|0 ) .
(−q) − 1 1 + q

 
Since a− ([L])|0 , a− ([L])|0 ) = (−1)−1 , formula (15.58) follows.
Remark 15.18. Note that Theorem 15.9 is a special case of Theorem 15.17.
However unlike the proof of Theorem 15.17, the proof of Theorem 15.9 is indepen-
dent of Theorem 14.2 and is based on partial computation of the obstruction sheaf
on the moduli space of stable maps to the Hilbert scheme X [n] .
 15.3. THE STRUCTURE OF THE GENUS-0 EXTREMAL INVARIANTS 301

15.3. The structure of the genus-0 extremal Gromov-Witten invariants

In this section, we will sketch the argument in [LQ5] which leads to the struc-
ture of 3-point genus-0 extremal Gromov-Witten invariants of the Hilbert scheme
X [n] . It involves finer analysis of the virtual fundamental cycle using the method
in [LiJ] and the cosection localization technique in Section 13.1.
First of all, we recall the concept of Hilbert schemes of α-points from [LiJ,
Section 1]. Let Λ be a finite set with |Λ| = n. We define
X [Λ] = X [n] ,
X (Λ) = X (n) ,
and for accounting purpose, denote
X Λ = {(xa )a∈Λ | xa ∈ X}.
Let
ρΛ = ρn : X [Λ] → X (Λ)
be the Hilbert-Chow morphism. Using ρΛ , we define the Hilbert scheme of Λ-points
to be
(15.59) X [[Λ]] = X [Λ] ×X (Λ) X Λ .
These spaces X [[Λ]] can be thought of as Hilbert schemes of ordered points.
Let PΛ be the set of partitions or equivalence relations on Λ. When α ∈ PΛ
consists of l equivalence classes α1 , . . . , αl , we write α = (α1 , . . . , αl ). For such α,
put

l
X (α) = X (αi ) ,
i=1

l
X [α] = X [αi ] .
i=1
[[α]]
Define the Hilbert scheme X of α-points by

l
(15.60) X [[α]] = X [[αi ]] .
i=1
Note that
X [[α]] = X [α] ×X (α) X Λ .
The spaces X [[α]] and X [[β]] are birational. To make this precise, we first fix the
convention on a partial ordering on PΛ . We define
“α ≥ β” ⇐⇒ “a ∼β b ⇒ a ∼α b”.
Namely, α ≥ β if β is finer than α. When β = (β1 , . . . , βr ), we put
α ∧ β = (α1 ∩ β1 , . . . , αl ∩ βr )
which is the largest element among all that are less than or equal to both α and β.
Note that PΛ contains a maximal and a minimal element. The maximal element
is Λ consisting of a single equivalence class Λ; the minimal element is 1Λ whose
equivalence classes are single element sets.
For α > β ∈ PΛ , define
β = {x ∈ X | ∃ a, b ∈ Λ so that xa = xb , a ∼α b, a ∼β b}.
Ξα Λ
302 15. THE GENUS-0 EXTREMAL GROMOV-WITTEN INVARIANTS

β
For α = β ∈ PΛ , define Ξα β = Ξα∧β ∪ Ξα∧β . The discrepancy between X
α [[α]]
and
[[β]] [[α]]
X (in X ) and its complement are defined to be
[[α]]
Ξβ = X [[α]] ×X Λ Ξα
β,
[[α]] [[α]]
Xβ = X [[α]] − Ξβ
respectively. More precisely, by Lemma 1.2 in [LiJ], there exists a functorial open
embedding
[[α]]
ζαβ : Xβ → X [[β]]
induced by the universal property of the respective moduli spaces such that
Im(ζαβ ) = Xα[[β]] .
[[α]] ∼
= [[β]]
Thus we obtain an isomorphism (equivalence) ζαβ : Xβ −→ Xα . We define
"& #
(15.61) X [[≤α]] = X [[β]] / ∼,
β≤α

[[β]] [[γ]]
where the equivalence is by identifying Xγ ⊂ X [[β]] and Xβ ⊂ X [[γ]] via ζβγ for all
β, γ ≤ α. Note that X [[≤α]] is non-separated (except when α = 1Λ ), and contains
the spaces X [[≤β]] , β ≤ α, as open subschemes.
Next, let
X [n,d] = M0,3 (X [n] , dβn )
be the moduli space of 3-pointed genus-0 degree-dβn stable maps to X [n] . Recall
from (13.12) that every stable map (ϕ, C) ∈ X [n,d] has a standard decomposition
ϕ = (ϕ1 , . . . , ϕl ) ∈ X [n,d]
where the stable reduction ϕst
i is contained in X
[ni ,di ]
for some ni and di ,
ρni (Im(ϕi )) = ni xi ,

the points x1 , . . . , xl are distinct, and ϕ(p) = i ϕi (p) for all p ∈ C. We use the
ideas from [LiJ, Section 1] to index the support of

ρn (Im(ϕ)) = ni xi ∈ X (n) .
i

This is done by introducing the notion of 3-pointed genus-0 degree-δ α-stable maps
to X [n] , where α = (α1 , · · · , αl ) denotes a partition of the set [n] = {1, . . . , n} and
δ = (δ1 , · · · , δl ) with δi ’s being nonnegative integers. The set of such pairs (α, δ)
with i δi = d is denoted by P[n],d . The techniques in [LiJ, Section 1] and the
product formula [Beh2, (1)] for Gromov-Witten invariants enable us to express the
virtual fundamental cycle [X [n,d] ]vir in terms of certain discrepancy cycles
[Θ[[α,δ]] ], (α, δ) ∈ P[n],d .
In fact, one of the key points in [LQ5] is to study such decomposition of [X [n,d] ]vir
as a sum of cycles indexed by the partition type of ρn (Im(ϕ)) ∈ X (n) . However, this
cannot be done on the moduli space X [n,d] . The technique to overcome this impasse
is to use the Hilbert scheme X [[α]] of α-points defined in (15.60) and the space
X [[≤α]] defined in (15.61). Then the cycle ev∗ ([X [n,d] ]vir ) is a sum of various [Θ[[α,δ]] ]
in (X [[≤[n]]] )3 . Combining with the cosection localization theory in Section 13.1,
 15.3. THE STRUCTURE OF THE GENUS-0 EXTREMAL INVARIANTS 303

pairings with [Θ[[α,δ]] ] can be studied via C ∞ -maps from X to the Grassmannians.
For d ≥ 1, we assemble those
[Θ[[α,δ]] ], (α, δ) ∈ P[n],d
with δi > 0 for every i into a homology class
Zn,d ∈ H∗ ((X [n] )3 ).
Note that we are back to the original Hilbert scheme X [n] . Now the structure
of the 3-point genus-0 extremal Gromov-Witten invariants of X [n] is given by the
following two theorems from [LQ5, Section 1].
Theorem 15.19. Let X be a simply connected smooth projective surface. Let
A1 , A2 , A3 ∈ H ∗ (X [n] ) be Heisenberg monomial classes, and πm,i be the i-th pro-
jection on (X [m] )3 . Then, A1 , A2 , A3 0,dβn is equal to
< =
  
3
∗
(15.62) A1,1 , A2,1 , A3,1 · Zm,d , πm,i Ai,2 .
m≤n A1,1 ◦A1,2 =A1 i=1
A2,1 ◦A2,2 =A2
A3,1 ◦A3,2 =A3

Theorem 15.20. Let X be a simply connected smooth projective surface. Let

A1 , A2 , A3 ∈ H ∗ (X [n] ) be Heisenberg monomial classes.
(i) If Ai contains a factor a−j (x) for some i, then
< =

3
∗
Zn,d , πn,i Ai = 0.
i=1

(ii) For 1 ≤ i ≤ 3, let

Ai = a−λ(i) (1X )a−ni,1 (αi,1 ) · · · a−ni,ui (αi,ui )|0
where ui ≥ 0 and |αi,1 | = . . . = |αi,ui | = 2. Then,
< =

3 3  ui
∗
(15.63) Zn,d , πn,i Ai = KX , αi,j · p
i=1 i=1 j=1

where p is a polynomial in KX , KX whose degree is at most


(n − ni,j )/2,
i,j

and whose coefficients depend only on d, n, λ(i) , ni,j (and hence are inde-
pendent of the surface X and the classes αi,j ).
We refer to Definition 3.13 for the operation ◦ appearing in (15.62), and to
Definition 3.28 (i) for the notation a−λ(i) (1X ) appearing in Theorem 15.20 (ii).
Geometrically, we may think of the pairing
< =

3
∗
Zm,d , πm,i Ai,2
i=1

in (15.62) as the contributions of the non-constant components ϕi in the standard

decomposition of ϕ = (ϕ1 , . . . , ϕl ) ∈ X [n,d] , while those constant components ϕi
contribute to the factor A1,1 , A2,1 , A3,1 in (15.62).
304 15. THE GENUS-0 EXTREMAL GROMOV-WITTEN INVARIANTS

Corollary 15.21. Let d ≥ 1, and let A1 , A2 , A3 ∈ H ∗ (X [n] ) be Heisenberg

monomial classes.
(i) If A1 = a−1 (1X )n−1 a−1 (α)|0 , then
< =
3
∗
Zn,d , πn,i Ai = 0.
i=1
n−1−|λ|
(ii) If A1 = a−1 (1X ) a−1 (α)a−λ (x)|0 for some λ, then
A1 , A2 , A3 0,dβn = 0.
Proof. (i) First of all, if α = x, then
< =

3
∗
Zn,d , πn,i Ai = 0.
i=1

by Theorem 15.20 (i).

Next, let α = 1X . Use induction on n. Since d ≥ 1, the conclusion is trivially
true when n = 1. Let n > 1. Recall that 1/n! · A1 is the fundamental class 1X [n]
of X [n] . By Theorem 15.19 and the Fundamental Class Axiom of Gromov-Witten
theory,
< =
n  
3
∗
A1,1 , A2,1 , A3,1 · Zm,d , πm,i Ai,2 = 0.
m=2 A1,1 ◦A1,2 =A1 i=1
A2,1 ◦A2,2 =A2
A3,1 ◦A3,2 =A3

Since A1 = a−1 (1X )n |0 , we have A1,2 = a−1 (1X )m |0 . By induction,

< =

3
∗
Zm,d , πm,i Ai,2 = 0
i=1

if 2 ≤ m ≤ n − 1. It follows that
< =

3
∗
Zn,d , πn,i Ai = 0.
i=1

Now let |α| = 2. By A1 , βn = 0 and the Divisor Axiom of Gromov-Witten

theory, we have A1 , A2 , A3 0,dβn = 0. Using an argument similar to the one in the
previous paragraph, we conclude that
< =

3
∗
Zn,d , πn,i Ai = 0.
i=1

(ii) We compute A1 , A2 , A3 0,dβn by using (15.62). Note that the class A1,2
in (15.62) is equal to a−1 (1X )m |0 , or is equal to a−1 (1X )m−1 a−1 (α)|0 , or con-
tains a factor a−i (x) for some i > 0. By (i) and Theorem 15.20 (i), we get
A1 , A2 , A3 0,dβn = 0. 

Problem 15.22. Let n ≥ 3, k ≥ 4 and d ≥ 1. Let X be a simply connected

smooth projective complex surface, and let A1 , . . . , Ak ∈ H ∗ (X [n] ) be Heisenberg
monomial classes. What is the structure of the genus-0 extremal Gromov-Witten
invariant A1 , . . . , Ak 0,dβn ?
 15.3. THE STRUCTURE OF THE GENUS-0 EXTREMAL INVARIANTS 305

We remark that when k ≥ 4, the arguments in [LQ5, Section 5] do not go

through since the moduli space M0,k of genus-0 k-pointed stable curves has positive
dimension.
 CHAPTER 16

Ruan’s Cohomological Crepant

Resolution Conjecture

The celebrated Crepant Resolution Conjecture of Bryan-Graber [BG, Conjec-

ture 1.2], Coates-Corti-Iritani-Tseng [CCIT, Conjecture 1.3], Coates-Ruan [CR,
Conjecture 4.1], and Ruan [Rua2, Section 2] (also Batyrev-Borisov [BB], Etingof-
Ginzburg [EtG, Section 1] and Reid [Reid]) has been extensively investigated
from various aspects in recent years. It originated from the orbifold string theory
in physics (notably, the work of Dixon-Harvey-Vafa-Witten [DHVW, Zas]). An
orbifold Z is a topological space locally modeled on the quotient of Euclidean space
by a finite group. Physicists attempted to formulate a string theory on Calabi-Yau
orbifolds which could be equivalent to the string theory on its crepant resolutions
(here a resolution ρ : W → Z is crepant if KW = ρ∗ KZ ). A consequence of this orb-
ifold string theory is the prediction that orbifold quantum cohomology of an orbifold
is isomorphic to the quantum cohomology of its crepant resolutions. Using the orb-
∗
ifold cohomology Horb (Z) developed in [CR1,CR2] (we also refer to [AGV] for the
algebraic setting), Ruan [Rua2, Section 2] formulated a more specific consequence
of this orbifold string theory conjecture, known as the Cohomological Crepant Res-
olution Conjecture. It eventually involved into the Crepant Resolution Conjecture
after the work of Bryan-Graber [BG, Conjecture 1.2], Coates-Corti-Iritani-Tseng
[CCIT, Conjecture 1.3] and Coates-Ruan [CR, Conjecture 4.1]. Roughly speak-
ing, the Crepant Resolution Conjecture predicts that the orbifold Gromov-Witten
theory of an orbifold Z is ring isomorphic to (in the sense of analytic continuations,
symplectic transformations and change of variables of type q = ceiu ) the ordinary
cohomology ring of its crepant resolution W plus those quantum corrections on W
which are related to curves contracted by the crepant resolution ρ. We refer to
[Coa] and the references there for excellent examples confirming this conjecture.
In this chapter, we will concentrate on Ruan’s Cohomological Crepant Resolu-
tion Conjecture for the Hilbert scheme X [n] where X denotes a smooth projective
complex surface. We will prove Ruan’s conjecture for the Hilbert-Chow morphism
ρn : X [n] → X (n)
which is a crepant resolution of the global orbifold X (n) by Remark 1.25. The
∗
orbifold cohomology ring Horb (X (n) ) has been studied in Chapter 10. We will
define the quantum corrected cohomology ring Hρ∗n (X [n] ) for the Hilbert scheme
X [n] . As vector spaces, Hρ∗n (X [n] ) and H ∗ (X [n] ) are the same. We will introduce
the cohomology classes G  k (α, n) ∈ Hρ∗ (X [n] ) and the operators G k (α), which are
n
∗
the analogues of the cohomology classes Gk (α, n) ∈ H (X ) and the operators
[n]

Gk (α) defined in Definition 4.1. From previous chapters, it is clear that the coho-
mology classes Gk (α, n) ∈ H ∗ (X [n] ) and the operators Gk (α) play significant roles
307
308 16. RUAN’S COHOMOLOGICAL CREPANT RESOLUTION CONJECTURE

in understanding the ordinary cohomology ring H ∗ (X [n] ). Similarly, the classes

G  k (α) will play essential roles in studying
 k (α, n) ∈ Hρ∗ (X [n] ) and the operators G
n

the quantum corrected cohomology ring Hρ∗n (X [n] ). We will prove the identity
1 {k}
 k (α), a−1 (β)] =
[G a (αβ).
k! −1
Following [LQ5, Section 6], an axiomatization originated from [Leh1, LQW1] and
formulated in [QW1, Subsection 2.6] completes the proof of Ruan’s conjecture.
Theorem 16.5 and its proof are from [LL], while Theorem 16.20, Theorem 16.22
and their proofs are from [LQ5].

16.1. The quantum corrected cohomology ring Hρ∗n (X [n] )

In this section, we will define the quantum corrected product w1 ·ρn w2 and
the quantum corrected cohomology ring Hρ∗n (X [n] ) for the Hilbert scheme X [n] of
points on a smooth projective complex surface X. Moreover, we will define the
cohomology classes G  k (α).
 k (α, n) ∈ Hρ∗ (X [n] ) and the operators G
n

Recall the homology class βn ∈ H2 (X ; Z) from (1.34). An irreducible curve

[n]

C ⊂ X [n] is contracted to a point by the Hilbert-Chow morphism

ρn : X [n] → X (n)
if and only if C ∼ dβn for some positive integer d. Let q be a formal variable. Define
a 3-point function with variable q for cohomology classes w1 , w2 , w3 ∈ H ∗ (X [n] ):

(16.1) w1 , w2 , w3 ρn (q) = w1 , w2 , w3 0,dβn q d
d≥0

where w1 , w2 , w3 0,dβn is the 3-point Gromov-Witten invariant defined by (12.5).

The following definition is adopted from [Rua2, Section 2] for the setting of
the Hilbert schemes X [n] .
Definition 16.1. The quantum corrected cohomology ring Hρ∗n (X [n] ) is the
group H ∗ (X [n] ) together with the quantum corrected product w1 ·ρn w2 defined by
(16.2) w 1 ·ρ n w 2 , w 3 = w 1 , w 2 , w 3 ρn (−1).

Remark 16.2. A priori, it is not clear why the power series (16.1) should be
convergent at q = −1 so that (16.2) makes sense. The convergence of (16.1) at
q = −1 will follow from the proofs in Subsection 16.2 which reduce the convergence
for an arbitrary surface X to the case of smooth toric surfaces. The convergence of
(16.1) at q = −1 for smooth toric surfaces is a consequence of [Cheo2, Theorem 0.2]
where a stronger version of Ruan’s Cohomological Crepant Resolution Conjecture
was proved for smooth toric surfaces.
Note that as complex vector spaces, Hρ∗n (X [n] ) and H ∗ (X [n] ) are the same.
Thus,

+∞ 
+∞
HX = H ∗ (X [n] ) = Hρ∗n (X [n] )
n=0 n=0
as complex vector spaces. Moreover, we see from (16.2), (16.1) and the Fundamental
Class Axiom (12.6) that
w ·ρn 1X [n] = w
for every w ∈ Hρ∗n (X [n] ).
 16.1. THE QUANTUM CORRECTED COHOMOLOGY RING Hρ∗n (X [n] ) 309

In the setting of the quantum corrected product, the Chern character opera-
tor Gi (α) ∈ End(HX ) and the cohomology class Gi (α, n) ∈ H ∗ (X [n] ) defined in
Definition 4.1 (iii) must be modified as follows.
Definition 16.3. Let k ≥ 0 and α ∈ H ∗ (X).
 k (α, n) ∈ Hρ∗ (X [n] ) to be
(i) Define the cohomology class G n

  (−1)|λ|−1
(16.3) · 1−(n−j−1) a−λ (τ∗ α)|0
λ! · |λ|!
0≤j≤k λ(j+1)
(λ)=k−j+1

  (−1)|λ| |λ| + s(λ) − 2

+ · · 1−(n−j−1) a−λ (τ∗ (eX α))|0 .
λ! · |λ|! 24
0≤j≤k λ(j+1)
(λ)=k−j−1

 k (α) ∈ End(HX ) to be the operator on the space

(ii) Define G

+∞
HX = Hρ∗n (X [n] )
n=0

which acts on the n-th component Hρ∗n (X [n] ) by the quantum corrected
 k (α, n)·ρ .
product G n

Remark 16.4. We have

 0 (α, n) = 1−(n−1) a−1 (α)|0 = G0 (α, n),
G
 1 (α, n) = − 1 1−(n−2) a−2 (α)|0 = G1 (α, n).
G
2
In general, we see from Theorem 4.10 that the class G  k (α, n) consists of those terms
in Gk (α, n) which do not contain the canonical divisor KX .
Note from the definition of the operator G  k (α) that

(16.4) G k (α)w1 , w2 = G
 k (α, n) ·ρ w1 , w2 = G  k (α, n), w1 , w2 ρn (−1)
n

for w1 , w2 ∈ Hρ∗n (X [n] ). For convenience, we introduce the operator G  k (α; q) by


(16.5)  k (α; q)w1 , w2 =
G G k (α, n), w1 , w2 0,dβ q d .
n
d≥0

From now on, let the surface X be simply connected. By Remark 16.4,
 1 (1X , n) = G1 (1X , n) = c1 O[n] .
G X
The following was proved in [LL, Subsection 7.2].
Theorem 16.5. We have
 1
 1 (1X ) =
G
[n]
c1 OX ·ρn = − : a3 :0 (τ3∗ 1X ).
n
6
Proof. First of all, recall from (14.19) that
 
k (−q)k + 1 1 (−q) + 1
(16.6) lim − = 0.
q→−1 2 (−q)k − 1 2 (−q) − 1
Next, by (16.2) and Theorem 15.17,
 [n]
c1 OX ·ρn = M(−1).
n
310 16. RUAN’S COHOMOLOGICAL CREPANT RESOLUTION CONJECTURE

So it remains to prove that

1 3
M(−1) = − : a :0 (τ3∗ 1X ).
6
In view of (4.17), this is equivalent to proving
1 
(16.7) M(−1) = − a−k a− ak+ + a−k− ak a (τ3∗ 1X ).
2
k,>0

Finally, (16.7) follows from (15.38) and (16.6). 

 1 (1X ) for the quantum corrected product on H ∗ (X [n] ) is par-
The operator G ρn
allel to the boundary operator d for the ordinary cup product on H ∗ (X [n] ). The
following definition is parallel to Definition 3.23.
Definition 16.6. For a linear operator f ∈ End(HX ), put f{0} = f. For k ≥ 1,
define f{k} inductively by
 1 (1X ), f{k−1} ].
f{k} = [G
The next lemma follows from Theorem 16.5 and the same proof of Theo-
rem 3.31.
{k}
Lemma 16.7. Let k ≥ 0, m ∈ Z, and α ∈ H ∗ (X). Then, am (α) is equal to
⎛ ⎞
 1  s(λ) − 1
(−m)k k! ⎝ aλ (τ∗ α) − aλ (τ∗ (eX α))⎠ .
λ! 24λ!
(λ)=k+1,|λ|=m (λ)=k−1,|λ|=m

{k}
Comparing with Theorem 3.31, we see that the operator am (α) consists of
(k)
those terms in am (α) which do not contain the canonical divisor KX .

16.2. The commutator [G̃k (α), a−1 (β)]

In this section, we will prove
1 {k}
 k (α), a−1 (β)] =
(16.8) [G a (αβ)
k! −1
for every simply connected smooth projective complex surface X. First of all, using
[Cheo2], we will show that (16.8) holds for a smooth projective toric surface X.
Then for an arbitrary surface X, using the results from Section 15.3 about the
structure of 3-point genus-0 extremal Gromov-Witten invariants of X [n] , we will
prove (16.8) in three separate cases: α = x is the cohomology class of a point,
|α| = 2, and α = 1X .
We begin with the following lemma which verifies (16.8) when X is a smooth
projective toric surface.
Lemma 16.8. Let X be a smooth projective toric surface. Then
 k (α), a−1 (β)] = 1 {k}
(16.9) [G a (αβ).
k! −1
Proof. Recall that P2 and the Hirzebruch surfaces Fa are smooth toric sur-
faces, and admit T = (C∗ )2 -actions. By the Proposition in Subsection 2.5 of [Ful],
X is obtained from P2 or Fa by a succession of blow-ups at T-fixed points.
 16.2. THE COMMUTATOR [G̃k (α), a−1 (β)] 311

∗,T
Now let aTm (α), Hρ∗,T n
(X [n] ) and pTm (α), Horb (X (n) ) be the equivariant versions
of am (α), Hρ∗n (X [n] ) and pm (α), Horb∗
(X (n) ) respectively. Following [Cheo2, Sub-
section 3.1], there exists a ring isomorphism
∗,T
ΨTn : Horb (X (n) ) → Hρ∗,T
n
(X [n] )
defined by
√ n1 +...+ns −s T
ΨTn −1 p−n1 (α1 ) · · · pT−ns (αs )|0
= aT−n1 (α1 ) · · · aT−ns (αs )|0 .
Note that up to a scalar factor which depends only on the partition λ = (n1 , . . . , ns )
and the tuple −→
α = (α1 , . . . , αs ), our notation pT−n1 (α1 ) · · · pT−ns (αs )|0 coincides
with the notation λ −→
α used in [Cheo2]. Also, our notation aTm (α) coincides with
the notation pm (α) used in [Cheo2]. The integer n1 + . . . + ns − s is the age.
Passing the map ΨTn to the ordinary cohomology, we obtain a ring isomorphism
∗
Ψn : Horb (X (n) ) → Hρ∗n (X [n] )
with
√ n1 +...+ns −s
Ψn −1 p−n1 (α1 ) · · · p−ns (αs )|0
= a−n1 (α1 ) · · · a−ns (αs )|0 .
√ k
Using (10.8) and (16.3), we see that Ψn −1 Ok (α, n) = G  k (α, n).
∗
Next, let A = a−n1 (α1 ) · · · a−ns (αs )|0 ∈ H (X [n−1]
). By definition,
[G  k (α)a−1 (β)A − a−1 (β)G
 k (α), a−1 (β)]A = G  k (α)A
 k (α, n) ·ρ a−1 (β)A − a−1 (β) G
= G  k (α, n − 1) ·ρ A .
n n−1

Put P = p−n1 (α1 ) · · · p−ns (αs )|0 and a = n1 + . . . + ns − s. Let • denote the
orbifold ring product. Then,
" √ k √ a #
Ψn p−1 (β) −1 Ok (α, n − 1) • −1 P
"√ k √ a #
= a−1 (β)Ψn −1 Ok (α, n − 1) • −1 P

=  k (α, n − 1) ·ρ
a−1 (β) G A ,
n−1

√ a
and Ψn  k (α), a−1 (β)]A is equal to
−1 p−1 (β)P = a−1 (β)A. So [G
√ k √ a
Ψn −1 Ok (α, n) • −1 p−1 (β)P
√ k √ a
− Ψn p−1 (β) −1 Ok (α, n − 1) • −1 P .
Since Ok (α, n) • p−1 (β)P = Ok (α)p−1 (β)P , we obtain
√
 k (α), a−1 (β)]A = −1k+a · Ψn [Ok (α), p−1 (β)]P .
[G

By Theorem 10.1 (iii), we conclude that [G k (α), a−1 (β)]A is equal to
⎛ ⎞
√ k+a ⎜  1  s(λ) − 1 ⎟
−1 ·(−1)k ·Ψn ⎝ p (τ (αβ))P +
! λ ∗ !
pλ (τ∗ (eX αβ))P ⎠ .
(λ)=k+1
λ (λ)=k−1
24λ
|λ|=−1 |λ|=−1
312 16. RUAN’S COHOMOLOGICAL CREPANT RESOLUTION CONJECTURE

 k (α), a−1 (β)]A is equal to

Finally, by the definition of Ψn and Lemma 16.7, [G
 1  s(λ) − 1 1 {k}
aλ (τ∗ (αβ))A − aλ (τ∗ (eX αβ))A = a (αβ)A.
(λ)=k+1
λ! (λ)=k−1
24λ! k! −1
|λ|=−1 |λ|=−1

Therefore,

 k (α), a−1 (β)] = 1 {k}

[G · a (αβ).
k! −1
This completes the proof of the lemma. 

To prove (16.8) for every simply connected smooth projective surface X, for
simplicity, we put
w1 , w2 , w3 d = w1 , w2 , w3 0,dβn .

In addition, w1 , w2 and w3 will stand for Heisenberg monomial classes.

Let α, β ∈ H ∗ (X). Then (16.8) is equivalent to
   
(16.10)  k (α), a−1 (β)]w1 , w2 = 1 a{k} (αβ)w1 , w2 .
[G −1
k!

for all cohomology classes w1 ∈ Hρ∗n (X [n−1] ) = H ∗ (X [n−1] ) and w2 ∈ Hρ∗n (X [n] ) =
H ∗ (X [n] ). Put
   
(16.11)  k (α; q), a−1 (β)]w1 , w2 − 1 a{k} (αβ)w1 , w2
Dβα (w1 , w2 ; q) = [G −1
k!
where we have omitted k in Dβα (w1 , w2 ; q) since it will be clear from the context.
The following lemma gives an explicit expression to Dβα (w1 , w2 ; q), which en-
ables us to analyze Dβα (w1 , w2 ; q).

Lemma 16.9. The difference Dβα (w1 , w2 ; q) is equal to

  (−1)|λ|−1  " 
1−(n−j−1) a−λ (τ∗ α)|0 , a−1 (β)w1 , w2 d
λ! · |λ|!
0≤j≤k λ(j+1) d≥1
(λ)=k−j+1
  #
− 1−(n−j−2) a−λ (τ∗ α)|0 , w1 , a−1 (β)† w2 d q d
   
+ f˜| | (λ) · aλ (τ∗ (αβ))w1 , w2
∈{KX ,KX
2 } (λ)=k+1−||/2
|λ|=−1
  " 
− g̃| | (λ) · 1−(n−j−1) a−λ (τ∗ (α))|0 , a−1 (β)w1 , w2
∈{KX ,K 2 } λ(j+1)
X (λ)=k−j+1−||/2
0≤j≤k
 #
− 1−(n−j−2) a−λ (τ∗ (α))|0 , w1 , a−1 (β)† w2

where a−1 (β)† = −a1 (β) is the adjoint operator of a−1 (β), and the functions f˜| | (λ)
and g̃| | (λ) depend only on k, || and λ.
 16.2. THE COMMUTATOR [G̃k (α), a−1 (β)] 313

 
Proof. By (16.5), [G  k (α; q), a−1 (β)]w1 , w2 is equal to
   
(16.12) G k (α; q) a−1 (β)w1 , w2 − a−1 (β)G  k (α; q)(w1 ), w2
   
= G  k (α; q) a−1 (β)w1 , w2 − G  k (α; q)(w1 ), a−1 (β)† w2
 " 
= G k (α, n), a−1 (β)w1 , w2
d
d≥0
  # d
− G k (α, n − 1), w1 , a−1 (β)† w2 q .
d

If d ≥ 1, then we see from (16.3) and Corollary 15.21 (ii) that

 
(16.13) G k (α, n), a−1 (β)w1 , w2
d
  (−1)|λ|−1  
= ! · |λ|!
1−(n−j−1) a−λ (τ∗ α)|0 , a−1 (β)w1 , w2 d .
λ
0≤j≤k λ(j+1)
(λ)=k−j+1
 
Similarly, if d ≥ 1, then G k (α, n − 1), w1 , a−1 (β)† w2 is equal to
d
  (−1)|λ|−1  
(16.14) 1−(n−j−2) a−λ (τ∗ α)|0 , w1 , a−1 (β)† w2 d .
λ · |λ|!
!
0≤j≤k λ(j+1)
(λ)=k−j+1

Next, we study the two terms with d = 0 in (16.12). By (16.3) and Theo-
 k (α, n) is equal to
rem 4.10, G
 
Gk (α, n) − g̃| | (λ) · 1−(n−j−1) a−λ (τ∗ (α))|0
∈{KX ,K 2 } λ(j+1)
X (λ)=k−j+1−||/2
0≤j≤k

where g̃| | (λ) depends only on k, || and λ. Note that

Gk (α, n), a−1 (β)w1 , w2 − Gk (α, n − 1), w1 , a−1 (β)† w2
= Gk (α, n) · a−1 (β)w1 , w2 − Gk (α, n − 1) · w1 , a−1 (β)† w2
= Gk (α)a−1 (β)w1 , w2 − a−1 (β)Gk (α)w1 , w2

= [Gk (α), a−1 (β)]w1 , w2 .
By (4.7), Theorem 3.31 and Lemma 16.7, we obtain
Gk (α, n), a−1 (β)w1 , w2 − Gk (α, n − 1), w1 , a−1 (β)† w2
1 (k)
= a (αβ)w1 , w2
k! −1
1 {k}    
= a−1 (αβ)w1 , w2 + f˜| | (λ) · aλ (τ∗ (αβ))w1 , w2 .
k!
∈{KX ,KX }
2 (λ)=k+1−||/2
|λ|=−1
   
Thus, G k (α, n), a−1 (β)w1 , w2 − G
 k (α, n − 1), w1 , a−1 (β)† w2 is equal to
1 {k}    
a−1 (αβ)w1 , w2 + f˜| | (λ) · aλ (τ∗ (αβ))w1 , w2
k!
∈{KX ,KX }
2 (λ)=k+1−||/2
|λ|=−1
  " 
− g̃| | (λ) · 1−(n−j−1) a−λ (τ∗ (α))|0 , a−1 (β)w1 , w2
∈{KX ,K 2 } λ(j+1)
X (λ)=k−j+1−||/2
0≤j≤k
 #
− 1−(n−j−2) a−λ (τ∗ (α))|0 , w1 , a−1 (β)† w2 .
314 16. RUAN’S COHOMOLOGICAL CREPANT RESOLUTION CONJECTURE

Combining this with (16.12), (16.13) and (16.14), we prove the lemma. 

In the following, the proof of (16.8) will be divided into three cases:
• α = x is the cohomology class of a point,
• |α| = 2,
• α = 1X .
We start with the simplest case when α = x and β is arbitrary.
Proposition 16.10. If α = x is the cohomology class of a point, then ( 16.8)
is true.
Proof. By Corollary 15.21 (ii), every term in Lemma 16.9 is equal to zero.
So Dβx (w1 , w2 ; q) = 0. In particular, Dβx (w1 , w2 ; −1) = 0. By (16.11),

 k (x; −1), a−1 (β)] = 1 {k}

[G a (xβ).
k! −1
 k (x) = G
In view of (16.4) and (16.5), G  k (x; −1). Hence (16.8) is true when α =
x. 

Next, we deal with the case |α| = 2. We will study Dβα (w1 , w2 ; q) when |α| = 2.
We begin with two lemmas about the structures of the intersections in H ∗ (X [n] ).
Lemma 16.11. Let λ be a partition with |λ| ≤ n. For i = 1 and 2, let
(16.15) wi = a−λ(i) (x)a−μ(i) (1X )a−ni,1 (αi,1 ) · · · a−ni,ui (αi,ui )|0

where |αi,j | = 2 for all i and j. Then,

 
(16.16) a−1 (1X )n−|λ| a−λ (x)|0 , w1 , w2
 
u1
= δu1 ,u2 · α1,i , α2,σ(i) · p(σ)
σ∈Perm{1,...,u1 } i=1

where p(σ) depends only on σ, n, λ and all the λ(i) , μ(i) , ni,j .

Proof. By Remark 8.10 (ii-a), a−1 (1X )n−|λ| a−λ (x)|0 is a polynomial of the
classes Gk (x, n), k ≥ 0 whose coefficients are independent of X. In addition, the
integers k involved depend only on λ. Note that
 
Gk1 (x, n) · · · Gkl (x, n), w1 , w2
 
= Gk1 (x) · · · Gkl (x)w1 , w2
 
= a−λ(1) (x)a−n1,1 (α1,1 ) · · · a−n1,u1 (α1,u1 )Gk1 (x) · · · Gkl (x)a−μ(1) (1X )|0 , w2 .
 
So by Theorem 4.7 and Theorem 3.8, Gk1 (x, n) · · · Gkl (x, n), w1 , w2 is equal to
 
u1
(16.17) δu1 ,u2 · α1,i , α2,σ(i) · p̃(σ)
σ∈Perm{1,...,u1 } i=1

where p̃(σ) depends only on σ, n, k1 , . . . , kl and all the λ(i) , μ(i) , ni,j . 
 16.2. THE COMMUTATOR [G̃k (α), a−1 (β)] 315

Lemma 16.12. Let n0 ≥ 1, |α| = 2, and λ be a partition.  Let w1 and w2 be

given by ( 16.15). Then, 1−(n−|λ|−n0 ) a−λ (x)a−n0 (α)|0 , w1 , w2 is equal to
 
u1
(16.18) KX , α · δu1 ,u2 · α1,i , α2,σ1 (i) · p1 (σ1 )
σ1 ∈Perm{1,...,u1 } i=1

u1 
+ α, α1,j · δu1 −1,u2 · α1,i , α2,σ2 (i) · p2 (σ2 )
j=1 σ2 i=j


u2  u1
+ α, α2,j · δu1 ,u2 −1 · α1,i , α2,σ3 (i) · p3 (σ3 )
j=1 σ3 i=1

where σ2 runs over all bijections {1, . . . , u1 } − {j} → {1, . . . , u2 }, σ3 runs over all
bijections {1, . . . , u1 } → {1, . . . , u2 } − {j}, and p1 (σ1 ) (respectively, p2 (σ2 ), p3 (σ3 ))
depend only on σ1 (respectively, σ2 , σ3 ), n, n0 , λ and all the λ(i) , μ(i) , ni,j .

Proof. By Remark 8.10 (ii-b), 1−(n−|λ|−n0 ) a−λ (x)a−n0 (α)|0 can be written
as

KX , α · F1 (n) + Gki (α, n) · F2,i (n)
i

where F1 (n) and F2,i (n) are polynomials of Gk (x, n), k ≥ 0 whose coefficients are
independent of n and α. Moreover, the integers k and ki depend only on λ and n0 .
Thus,
 
(16.19) 1−(n−|λ|−n0 ) a−λ (x)a−n0 (α)|0 , w1 , w2
   
= KX , α · F1 (n), w1 , w2 + Gki (α, n) · F2,i (n), w1 , w2 .
i
 
As in the proof of Lemma 16.11, F1 (n), w1 , w2 is of the form
 
u1
(16.20) δu1 ,u2 · α1,i , α2,σ1 (i) · p̃1,1 (σ1 )
σ1 ∈Perm{1,...,u1 } i=1

where p̃1,1 (σ1 ) depends only on σ1 , n, n0 , λ and all the λ(i) , μ(i) , ni,j . Also,
 
Gki (α, n)Gs1 (x, n) · · · Gsl (x, n), w1 , w2
 
= Gs1 (x) · · · Gsl (x)Gki (α)w1 , w2 .
 
By Theorem 4.7 and Lemma 3.12, Gki (α, n)Gs1 (x, n) · · · Gsl (x, n), w1 , w2 is equal
to
 
u1
(16.21) KX , α · δu1 ,u2 · α1,i , α2,σ1 (i) · p̃1,2 (σ1 )
σ1 ∈Perm{1,...,u1 } i=1

u1 
+ α, α1,j · δu1 −1,u2 · α1,i , α2,σ2 (i) · p̃2 (σ2 )
j=1 σ2 i=j


u2  u1
+ α, α2,j · δu1 ,u2 −1 · α1,i , α2,σ3 (i) · p̃3 (σ3 )
j=1 σ3 i=1
316 16. RUAN’S COHOMOLOGICAL CREPANT RESOLUTION CONJECTURE

where σ2 runs over all the bijections {1, . . . , u1 } − {j} → {1, . . . , u2 }, and σ3 runs
over all the bijections {1, . . . , u1 } → {1, . . . , u2 } − {j}. Hence
 
Gki (α, n) · F2,i (n), w1 , w2
i
is of the form (16.21) as well. Combining with (16.19) and (16.20), we obtain
formula (16.18). 
Now we introduce the notion of universal polynomials
P (KX , S1 , S2 )
in KX , KX of degree at most m and of type (u1 , u2 ), and prove a vanishing lemma.
Definition 16.13. Fix three integers m, u1 , u2 ≥ 0. Then a universal polyno-
mial P (KX , S1 , S2 ) in KX , KX of degree at most m and of type (u1 , u2 ) is of the
form
  
(16.22) KX , α1,i · KX , α2,i
1≤j1 <...<js ≤u1 i∈{j1 ,...,js } i∈{l1 ,...,ls }
1≤l1 <...<ls ≤u2

 
s
· α1,ji , α2,σ(li ) · p(j1 , . . . , js ; l1 , . . . , ls ; σ)
σ∈Perm{l1 ,...,ls } i=1

where Si = {αi,1 , . . . , αi,ui } ⊂ H 2 (X), and p(j1 , . . . , js ; l1 , . . . , ls ; σ) is a polynomial

in KX , KX whose degree is at most m and whose coefficients are independent of
X and the classes αi,j .
Lemma 16.14. Fix m, u1 , u2 ≥ 0. Let P (KX , S1 , S2 ) be a universal polynomial
in KX , KX of degree at most m and of type (u1 , u2 ). Assume that
P (KX , S1 , S2 ) = 0
whenever X is a smooth projective toric surface. Then P (KX , S1 , S2 ) = 0 for every
smooth projective surface X.
Proof. Let r  m + u1 + u2 , and let Xr be a smooth toric surface obtained
from P2 as an r-fold blown-up. Let L0 be a line in P2 , and let e1 , . . . , er be the
exceptional divisors. Then,
KXr = −3L0 + e1 + . . . + er .
For fixed j1 , . . . , js , l1 , . . . , ls and σ, let
α1,i | i ∈ {1, . . . , u1 } − {j1 , . . . , js }} = {−e1 , . . . , −eu1 −s ,
α2,i | i ∈ {1, . . . , u2 } − {l1 , . . . , ls }} = {−eu1 −s+1 , . . . , −eu1 −s+u2 −s ,
and
α1,ji = α2,σ(li ) = eu1 −s+u2 −s+2i − eu1 −s+u2 −s+2i−1
for i = 1, . . . , s. Then,
(16.23) 0 = P (KXr , S1 , S2 ) = (−2)s · p(j1 , . . . , js ; l1 , . . . , ls ; σ)
by (16.22). It follows that p(j1 , . . . , js ; l1 , . . . , ls ; σ) = 0 for all the surfaces Xr with
r  m + u1 + u2 . Since p(j1 , . . . , js ; l1 , . . . , ls ; σ) is a polynomial in KXr , KXr
whose degree is at most m, we conclude that as polynomials,
p(j1 , . . . , js ; l1 , . . . , ls ; σ) = 0.
Therefore, P (KX , S1 , S2 ) = 0 for every smooth projective surface X. 
 16.2. THE COMMUTATOR [G̃k (α), a−1 (β)] 317

Our next lemma is about the structure of certain 3-pointed extremal Gromov-
Witten invariants, and provides the motivation for Definition 16.13.
Lemma 16.15. Let d, n0 ≥ 1 and |α| = 2. Let w1 and w2 be given by ( 16.15).
Then,
 
(16.24) 1−(n−|λ|−n0 ) a−λ (x)a−n0 (α)|0 , w1 , w2 d
= KX , α · P (KX , S1 , S2 )
where S1 = {α1,1 , . . . , α1,u1 }, S2 = {α2,1 , . . . , α2,u2 }, and P (KX , S1 , S2 ) is a uni-
versal polynomial in KX , KX of degree at most (n − n0 )/2 and of type (u1 , u2 ).
Proof. For simplicity, let
w0 = 1−(n−|λ|−n0 ) a−λ (x)a−n0 (α)|0 .
Also, for i = 1 and 2, let
i = a−μ(i) (1X )a−ni,1 (αi,1 ) · · · a−ni,ui (αi,ui )|0 .
w
 
We compute w0 , w1 , w2 d by using (15.62). Consider the following term from
(15.62):
F      G
∗ w0 ∗ w1 ∗ w2
(16.25) B0 , B1 , B2 · Zm,d , πm,1 · πm,2 · πm,3
B0 B1 B2
where m ≤ n, B0 , B1 , B2 ∈ H ∗ (X [n−m] ), B0 ⊂ w0 , B1 ⊂ w1 , and B2 ⊂ w2 . By
Theorem 15.20 (i) and Corollary 15.21 (i), such a term is nonzero only if
B0 = a−1 (1X )j a−λ (x)|0 ,
B1 = a−λ(1) (x)B1 ,
B2 = 2
a−λ(2) (x)B

with j ≤ (n − |λ| − n0 ), B 1 ⊂ w 1 , and B2 ⊂ w 2 . In this situation, (16.25) can be

rewritten as
 
(16.26) a−1 (1X )j a−λ (x)|0 , a−λ(1) (x)B1 , a−λ(2) (x)B 2
F      G
∗ 1−(n−|λ|−n0 ) a−n0 (α)|0 ∗ 1
w ∗ w2
(16.27) · Zm,d , πm,1 · πm,2 · πm,3 .
a−1 (1X )j |0 
B1 B2
Note that
1 = a−ν (1) (1X )a−n (α1,j ) · · · a−n (α1,j )|0
B 1,j1 1 1,js s

for some 1 ≤ j1 < . . . < js ≤ u1 and some sub-partition ν (1) of μ(1) (i.e., every part
of ν (1) is a part of μ(1) ). Similarly,
2 = a−ν (2) (1X )a−n (α2,l ) · · · a−n (α2,l )|0
B 2,l1 1 2,lt t

for some 1 ≤ l1 < . . . < lt ≤ u2 and some sub-partition ν (2) of μ(2) . By

Lemma 16.11, (16.26) is equal to
 
s
(16.28) δs,t · α1,ji , α2,σ(li ) · p1 (j1 , . . . , js ; l1 , . . . , ls ; σ)
σ∈Perm{l1 ,...,ls } i=1
318 16. RUAN’S COHOMOLOGICAL CREPANT RESOLUTION CONJECTURE

where p1 is a number independent of the surface X and the classes αi,j . By Theo-
rem 15.20, we see that the factor (16.27) is equal to
 
KX , α · KX , α1,i · KX , α2,i · p2 (j1 , . . . , js ; l1 , . . . , ls ; σ)
i∈{j1 ,...,js } i∈{l1 ,...,ls }

where p2 is a polynomial in KX , KX whose degree is at most (m − n0 )/2 ≤ (n −

n0 )/2, and whose coefficients are independent of the surface X and the classes αi,j .
Combining this with (16.25), (16.26), (16.27) and (16.28), we obtain (16.24). 

Proposition 16.16. If |α| = 2, then ( 16.8) is true.

Proof. Recall that (16.8) is equivalent to (16.10), and Dβα (w1 , w2 ; q) from
(16.11) is computed by Lemma 16.9. Let w1 and w2 be given by (16.15). Let
u1 = δ2,|β| + u1 and S2 = {α2,1 , . . . , α2,u2 }. Let S1 = {α1,1 , . . . , α1,u1 } if |β| = 2,
and S1 = {β, α1,1 , . . . , α1,u1 } if |β| = 2.
By Lemma 16.8 and Lemma 16.14, it suffices to prove that
(16.29) Dβα (w1 , w2 ; −1) = KX , α · P (KX , S1 , S2 )
where P (KX , S1 , S2 ) is a universal polynomial in KX , KX of degree at most
(n − 1)/2 and of type (u1 , u2 ). This follows if we can prove that

(16.30) Dβα (w1 , w2 ; q) = KX , α · P (KX , S1 , S2 ; d) q d
d≥0

where every P (KX , S1 , S2 ; d) is a universal polynomial in KX , KX of degree at

most (n − 1)/2 and of type (u1 , u2 ). We remark that d has been inserted into the
notation P (KX , S1 , S2 ; d) to emphasis its dependence on d.
In the rest of the proof, we will show that the contribution of every term in
Lemma 16.9 is of the form P (KX , S1 , S2 ; d) for a suitable d ≥ 0. Note that in
H ∗ (X i ),
τi∗ (α) = α ⊗ x ⊗ · · · ⊗ x + x ⊗ α ⊗ x ⊗ · · · ⊗ x + . . . + x ⊗ · · · ⊗ x ⊗ α.
 
Thus, by Lemma 16.15, 1−(n−j−1) a−λ (τ∗ α)|0 , a−1 (β)w1 , w2 d is equal to
KX , α · P1 (KX , S1 , S2 ; d)
where P1 (KX , S1 , S2 ; d) is a universal polynomial in KX , KX of degree at most
(n − 1)/2 and of type (u1 , u2 ). Similarly, since a−1 (β)† w2 = −a1 (β)w2 , we see from
Theorem 3.8 and Lemma 16.15 that
 
1−(n−j−2) a−λ (τ∗ α)|0 , w1 , a−1 (β)† w2 d = KX , α · P2 (KX , S1 , S2 ; d).
 
Next, we move to the term aλ (τ∗ (αβ))w1 , w2 in Lemma 16.9, where  ∈
{KX , KX2
}. Such a term is zero unless  = KX and |β| = 0. In this case, we may
assume that β = 1X . So let  = KX and β = 1X . Then,
   
aλ (τ∗ (αβ))w1 , w2 = KX , α · aλ (x)w1 , w2
= KX , α · P3 (KX , S1 , S2 ; 0).
by Theorem 3.8, where P3 (KX , S1 , S2 ; 0) is a universal polynomial in KX , KX of
degree 0 (i.e., KX , KX does not appear) and of type (u1 , u2 ).
 16.2. THE COMMUTATOR [G̃k (α), a−1 (β)] 319

Finally, τ∗ (α) is zero unless  = KX . Let  = KX . By Lemma 16.11,

 
1−(n−j−1) a−λ (τ∗ (α))|0 , a−1 (β)w1 , w2
 
= KX , α · 1−(n−j−1) a−λ (x)|0 , a−1 (β)w1 , w2
= KX , α · P4 (KX , S1 , S2 ; 0)
where P4 (KX , S1 , S2 ; 0) is a universal polynomial in KX , KX of degree 0 and of
type (u1 , u2 ). Similarly, since a−1 (β)† w2 = −a1 (β)w2 , we obtain
 
1−(n−j−2) a−λ (τ∗ (α))|0 , w1 , a−1 (β)† w2 = KX , α · P5 (KX , S1 , S2 ; 0).
This completes the proof of our proposition. 
We are left with the case α = 1X . We start with two lemmas.
Lemma 16.17. Let d, n0 ≥ 1. Let w1 and w2 be given by ( 16.15). Then,
 
(16.31) 1−(n−|λ|−n0 ) a−λ (x)a−n0 (1X )|0 , w1 , w2 d = P (KX , S1 , S2 )
where S1 = {α1,1 , . . . , α1,u1 }, S2 = {α2,1 , . . . , α2,u2 }, and P (KX , S1 , S2 ) is a uni-
versal polynomial in KX , KX of degree at most n/2 and of type (u1 , u2 ).
Proof. This follows from the proof of Lemma 16.15 by replacing α by 1X (and
then by noticing that the factor KX , α there will not appear here). 
Lemma 16.18. Let d ≥ 1 and |λ| ≤ n. Let w1 and w2 be given by ( 16.15).
Then,
 
(16.32) 1−(n−|λ|) a−λ (τ∗ 1X )|0 , w1 , w2 d = P (KX , S1 , S2 )
where S1 = {α1,1 , . . . , α1,u1 }, S2 = {α2,1 , . . . , α2,u2 }, and P (KX , S1 , S2 ) is a uni-
versal polynomial in KX , KX of degree at most n/2 and of type (u1 , u2 ).
Proof. For i = 1 and 2, let
i = a−μ(i) (1X )a−ni,1 (αi,1 ) · · · a−ni,ui (αi,ui )|0 .
w
Note that if the Künneth decomposition of τ2∗ 1X ∈ H ∗ (X 2 ) is given by

τ2∗ 1X = x ⊗ 1X + 1X ⊗ x + γj,1 ⊗ γj,2
j

where |γj,1 | = |γj,2 | = 2, then up to permutations of factors, a typical term in the

Künneth decomposition of τi∗ 1X ∈ H ∗ (X i ) with i ≥ 3 is either x ⊗ · · · ⊗ x ⊗ 1X or
x ⊗ · · · ⊗ x ⊗ γj,1 ⊗ γj,2 . In view of Lemma 16.17, it suffices to verify that
 
(16.33) 1−ñ a−λ̃ (x)a−n1 (γj,1 )a−n2 (γj,2 )|0 , w1 , w2 d = P1 (KX , S1 , S2 )
j

where ñ = n − |λ̃| − n1 − n2 , and P1 (KX , S1 , S2 ) is a universal polynomial in

KX , KX of degree at most n/2 and of type (u1 , u2 ). For simplicity, let
w0 = 1−ñ a−λ̃ (x)a−n1 (γj,1 )a−n2 (γj,2 )|0 .
We see from (15.62) that to prove (16.33), it suffices to show that
 F      G
∗ w0 ∗ w1 ∗ w2
(16.34) B0 , B1 , B2 · Zm,d , πm,1 · πm,2 · πm,3
j
B 0 B 1 B 2

is equal to P2 (KX , S1 , S2 ), where m ≤ n, B0 , B1 , B2 ∈ H ∗ (X [n−m] ), B0 ⊂ w0 ,

B1 ⊂ w1 , and B2 ⊂ w2 . By Theorem 15.20 (i) and Corollary 15.21 (i), such
320 16. RUAN’S COHOMOLOGICAL CREPANT RESOLUTION CONJECTURE

a term is nonzero only if B1 = a−λ(1) (x)B 1 with B 1 ⊂ w 2

1 , B2 = a−λ(2) (x)B

with B2 ⊂ w 2 , and B0 = a−1 (1X ) a−λ̃ (x)|0 or a−1 (1X ) a−λ̃ (x)a−n1 (γj,1 )|0 or
s s

a−1 (1X )s a−λ̃ (x)a−n2 (γj,2 )|0 where s ≤ ñ. In the following, we assume that (16.34)
is nonzero. By symmetry, we need only to consider two cases for B0 :
B0 = a−1 (1X )s a−λ̃ (x)|0
or
B0 = a−1 (1X )s a−λ̃ (x)a−n1 (γj,1 )|0 .
We begin with the case B0 = a−1 (1X )s a−λ̃ (x)|0 . Then (16.34) becomes
 
1 , a−λ(2) (x)B
a−1 (1X )s a−λ̃ (x)|0 , a−λ(1) (x)B 2
j
F      G
∗ 1−ñ a−n1 (γj,1 )a−n2 (γj,2 )|0 ∗ 1
w ∗ 2
w
· Zm,d , πm,1 · πm,2 · πm,3 .
a−1 (1X ) |0
s 1
B 2
B
Applying the same arguments as in the computations of (16.26) and (16.27), we
conclude that the term (16.34) is equal to

KX , γj,1 · KX , γj,2 · P3 (KX , S1 , S2 )
j

where P3 (KX , S1 , S2 ) is a universal polynomial in KX , KX of degree at most

(m − n1 − n2 )/2 and of type (u1 , u2 ). Note that for β1 , β2 ∈ H 2 (X), we have

(16.35) β1 , γj,1 · β2 , γj,2 = β1 , β2 .
j

Therefore, (16.34) is equal to KX , KX · P3 (KX , S1 , S2 ) which is a universal poly-

nomial in KX , KX of degree at most m/2 ≤ n/2 and of type (u1 , u2 ).
Next, let B0 = a−1 (1X )s a−λ̃ (x)a−n1 (γj,1 )|0 . This time, (16.34) becomes
 
1 , a−λ(2) (x)B
a−1 (1X )s a−λ̃ (x)a−n1 (γj,1 )|0 , a−λ(1) (x)B 2
j
F      G
∗ 1−ñ a−n2 (γj,2 )|0 ∗ 1
w ∗ 2
w
· Zm,d , πm,1 · πm,2 · πm,3 .
a−1 (1X ) |0
s 
B1 2
B
Using Lemma 16.12, Theorem 15.20 and (16.35), we conclude that (16.34) is equal
to P4 (KX , S1 , S2 ) which is a universal polynomial in KX , KX of degree at most
(m − n2 )/2 + 1 ≤ ((n − n1 ) − n2 )/2 + 1 ≤ n/2
and of type (u1 , u2 ). This completes the proof of (16.32). 
Proposition 16.19. If α = 1X , then ( 16.8) is true.
Proof. We adopt the same notations and approaches as in the proof of Propo-
sition 16.16. By Lemma 16.8 and Lemma 16.14, it suffices to prove that
(16.36) Dβ1X (w1 , w2 ; −1) = P (KX , S1 , S2 )
where P (KX , S1 , S2 ) is a universal polynomial in KX , KX of degree at most
(n + 1)/2 and of type (u1 , u2 ). This follows if we can prove that

(16.37) Dβ1X (w1 , w2 ; q) = P (KX , S1 , S2 ; d) q d
d≥0
 16.2. THE COMMUTATOR [G̃k (α), a−1 (β)] 321

where P (KX , S1 , S2 ; d) is a universal polynomial in KX , KX of degree at most

(n + 1)/2 and of type (u1 , u2 ). In the following, we will show that the contribution
of every term in Lemma 16.9 is of the form P (KX , S1 , S2 ; d) for a suitable d ≥ 0.
First of all, when d ≥ 1, we conclude from Lemma 16.18 that
 
1−(n−j−1) a−λ (τ∗ 1X )|0 , a−1 (β)w1 , w2 d
 
− 1−(n−j−2) a−λ (τ∗ 1X )|0 , w1 , a−1 (β)† w2 d
is equal to P1 (KX , S1 , S2 ; d) which is a universal polynomial in KX , KX of degree
at most n/2 and of type (u1 , u2 ).
Next, consider
   
aλ (τ∗ (αβ))w1 , w2 = aλ (τ∗ (β))w1 , w2
from Lemma 16.9, where  ∈ {KX , KX 2
}. It is zero unless  = KX 2
and β = 1X
(when |β| = 0, we let β = 1X ), or  = KX and |β| = 2, or  = KX and β = 1X . If
2
 = KX and β = 1X , then
   
aλ (τ∗ (β))w1 , w2 = KX , KX · aλ (x)w1 , w2
= KX , KX · P2 (KX , S1 , S2 ; 0)
by Theorem 3.8, where P2 (KX , S1 , S2 ; 0) is a universal polynomial in KX , KX of
degree 0 and of type (u1 , u2 ). If  = KX and |β| = 2, then
   
aλ (τ∗ (β))w1 , w2 = KX , β · aλ (x)w1 , w2
= KX , β · P3 (KX , S1 , S2 ; 0)
which is a universal polynomial in KX , KX of degree 0 and of type (u1 , u2 ). If
 = KX and β = 1X , then we obtain
   
aλ (τ∗ (β))w1 , w2 = aλ (τ∗ KX )w1 , w2
which again is a universal polynomial in KX , KX of degree 0 and of type (u1 , u2 ).
Finally, let  ∈ {KX , KX 2
}. We have τ∗ (α) = τ∗ . Let I be the difference
 
1−(n−j−1) a−λ (τ∗ )|0 , a−1 (β)w1 , w2
 
− 1−(n−j−2) a−λ (τ∗ )|0 , w1 , a−1 (β)† w2
2
from Lemma 16.9. When  = KX , we see from Lemma 16.11 that
 
I = KX , KX · 1−(n−j−1) a−λ (x)|0 , a−1 (β)w1 , w2
 
− KX , KX · 1−(n−j−2) a−λ (x)|0 , w1 , a−1 (β)† w2
= KX , KX · P4 (KX , S1 , S2 ; 0)
where P4 (KX , S1 , S2 ; 0) is a universal polynomial in KX , KX of degree 0 and of
type (u1 , u2 ). When  = KX , we see from Lemma 16.12 that
 
I = 1−(n−j−1) a−λ (τ∗ KX )|0 , a−1 (β)w1 , w2
 
− 1−(n−j−2) a−λ (τ∗ KX )|0 , w1 , a−1 (β)† w2
is a universal polynomial in KX , KX of degree at most 1 and of type (u1 , u2 ). 

The following is the main result in this section.

322 16. RUAN’S COHOMOLOGICAL CREPANT RESOLUTION CONJECTURE

Theorem 16.20. Let X be a simply connected smooth projective complex sur-

face. Then
 k (α), a−1 (β)] = 1 a{k} (αβ).
[G
k! −1
Proof. Follows from Propositions 16.10, 16.16 and 16.19. 
At this point, all the necessary properties (Theorem 3.8, Theorem 16.5 and
Theorem 16.20) on the Hilbert scheme side in order to prove Ruan’s Cohomological
Crepant Resolution Conjecture for the Hilbert-Chow morphisms have been estab-
lished. We are therefore led to the final section of the book.

16.3. Ruan’s Cohomological Crepant Resolution Conjecture

In this section, we will review Ruan’s Cohomological Crepant Resolution Con-
jecture for the Hilbert-Chow morphisms, and present a proof of it following the
axiomatization approach formulated in [QW1, Subsection 2.6].
Let X denote a smooth projective complex surface. By Remark 1.25, the
Hilbert-Chow morphism
ρn : X [n] → X (n)
is a crepant resolution of the global orbifold X (n) . In this setting, Ruan’s Cohomo-
logical Crepant Resolution Conjecture states as follows.
conjecture 16.21. (Ruan’s Cohomological Crepant Resolution Conjecture)
Let X be a smooth projective complex surface. Then, the quantum corrected coho-
mology ring Hρ∗n (X [n] ) is ring isomorphic to Horb
∗
(X (n) ).
Conjecture 16.21 was first confirmed for n = 2, following Theorem 15.9 and
Example 15.10. When the smooth projective surface X has a numerically trivial
canonical class KX , Conjecture 16.21 was proved in [FG1, LS2, Uri] (the case
of certain quasi-projective surfaces X was also considered in [LQW5, Section 5]
and [QW3, Subsection 3.4]). In [Cheo2, Corollary 0.3], Conjecture 16.21 was
verified when X is a smooth projective toric surface (in fact, a stronger version
of Conjecture 16.21 was proved there). The main idea in [Cheo2] is to take a
toric action on X with finitely many fixed points x1 , . . . , xs . Each point xi has an
open neighborhood Xi such that Xi ∼ = C2 and xi ∈ Xi is identified with the origin
of C . In terms of the Heisenberg monomial classes (3.10), a product formula is
2

proved, which reduces the 3-point genus-0 extremal Gromov-Witten invariants of

X [n] to those of Xi ∼
= C2 . There is also a similar product formula on the orbifold
side. Therefore, when X is a smooth projective toric surface, Conjecture 16.21 is
reduced to the case of the affine plane C2 in the equivariant setting. The case of
the affine plane C2 in the equivariant setting has been verified earlier in [BG].
An axiomatization originated from [Leh1, LQW1] and formulated in [QW1,
Subsection 2.6] was used in [LQ5, Section 6] to prove Conjecture 16.21. This
axiomatization involves Heisenberg algebra actions and vertex operator techniques
pioneered in [Groj, Nak3]. We recall from Definition 7.12 that a graded Frobenius
algebra over C is a finite dimensional graded (complex) vector space A with a
graded associative multiplication A ⊗ A → A and unit element 1A together with
a linear form T : A → C such that the induced bilinear form a, b = T (ab) is
nondegenerate. For k ≥ 1, the k-th co-product
τk∗ : A → A⊗k
 16.3. RUAN’S COHOMOLOGICAL CREPANT RESOLUTION CONJECTURE 323

is defined by requiring
τk∗ (a), b1 ⊗ · · · ⊗ bk = T (ab1 · · · bk ).
Now the axiomatization in [QW1, Subsection 2.6] states that the algebra structure
on each A[n] in a sequence of graded Frobenius algebras A[n] (n ≥ 0) is determined
if
(A1) the direct sum

+∞
A[n]
n=0
affords the structure of the Fock space of a Heisenberg algebra modeled
on A which denotes A[1] .
(A2) There exists a sequence of elements G k (α, n) ∈ A[n] depending on α ∈ A
(linearly) and a non-negative integer k. Define the operators G  k (α) on


+∞
A[n]
n=0

which act on the component A[n] via multiplication by G  k (α, n) ∈ A[n] .

 k (α) and the Heisenberg generators satisfy:
The operators G

 1 (1A ) = 1 3
(16.38) G − : a :0 (τ3∗ 1A ),
6
 k (α), a−1 (β)] = 1 {k}
(16.39) [G a (αβ)
k! −1
where : a3 :0 is the zero mode in the normally ordered product : a3 :, and
{k} {0}
a−1 (α) denotes the k-th derivative with a−1 (α) = a−1 (α) and
{k}  1 (1A ), a{k−1} (α)]
a−1 (α) = [G −1

for k ≥ 1.
When (A1) and (A2) are satisfied, the algebra A[n] is generated by the elements
 k (α, n) ∈ A[n] ,
G α ∈ A, k ≥ 0.
In addition, the product is determined by (16.38) and (16.39).
On one hand, with
∗
A[n] = Horb (X (n) )
∗
(viewed as an algebra over C) and the sequence of classes Ok (α, n) ∈ Horb (X (n) ), we
see from Theorem 10.1, Theorem 10.6 and Theorem 10.8 that (A1) and (A2) hold
∗
for the orbifold cohomology rings Horb (X (n) ). On the other hand, by Theorem 3.8,
Theorem 16.5 and Theorem 16.20, the quantum corrected cohomology rings
A[n] = Hρ∗n (X [n] )

also satisfy (A1) and (A2) with A = A[1] = H ∗ (X) and the sequence of classes
 k (α, n) ∈ Hρ∗ (X [n] ). This leads to the following result.
G n

Theorem 16.22. Conjecture 16.21 holds when X is a simply connected smooth

projective surface.
324 16. RUAN’S COHOMOLOGICAL CREPANT RESOLUTION CONJECTURE

Proof. Recall the operators pn (α) from Section 10.2. Note that the shift
number (or the age) of p−n1 (α1 ) · · · p−ns (αs )|0 is equal to n1 + . . . + ns − s. Define
a linear isomorphism

+∞ 
+∞
∗
(16.40) Ψ: Horb (X (n) ) −→ Hρ∗n (X [n] )
n=0 n=0
by
"√ n +...+n −s #
−1 p−n1 (α1 ) · · · p−ns (αs )|0 = a−n1 (α1 ) · · · a−ns (αs )|0 .
1 s
Ψn
This induces a linear isomorphism
∗
Ψn : Horb (X (n) ) −→ Hρ∗n (X [n] ) = H ∗ (X [n] )
for each n. Moreover, Ψ1 is simply the identity map on the cohomology group of
the surface X.
By Theorem 16.5 and Theorem 16.20, the two formulas (16.38) and (16.39)
hold for A[n] = Hρ∗n (X [n] ). By the proof of Theorem 4.7, we have
 1  s(λ) − 2
 k (α) = −
G aλ (τ∗ α) + aλ (τ∗ (eX α)).
λ ! 24λ!
(λ)=k+2,|λ|=0 (λ)=k,|λ|=0

Combining this with Theorem 10.15, we check directly that

"√ k √ n1 +...+ns −s #
Ψn −1 Ok (α, n) • −1 p−n1 (α1 ) · · · p−ns (αs )|0
"√ k+n +...+n −s #
= Ψn −1 Ok (α)p−n1 (α1 ) · · · p−ns (αs )|0
1 s

 k (α)a−n (α1 ) · · · a−n (αs )|0

= G 1 s

 k (α, n) ·ρ a−n (α1 ) · · · a−n (αs )|0

= G n 1 s

where n1 + . . . + ns = n. In particular, letting s = n, n1 = . . . = ns = 1 and

α1 = . . . = αs = 1X , we obtain
√ k
Ψn −1 Ok (α, n) = G  k (α, n).
Thus,
"√ k √ n1 +...+ns −s
#
Ψn −1 Ok (α, n) • −1 p−n1 (α1 ) · · · p−ns (αs )|0
√ k
= Ψn −1 Ok (α, n) ·ρn a−n1 (α1 ) · · · a−ns (αs )|0
√ k √ n1 +...+ns −s
= Ψn −1 Ok (α, n) ·ρn Ψn −1 p−n1 (α1 ) · · · p−ns (αs )|0 .
Since the classes Ok (α, n) with k ≥ 0, α ∈ H ∗ (X) generate the ring Horb
∗
(X (n) ), we
conclude that
∗
Ψn : Horb (X (n) ) −→ Hρ∗n (X [n] )
is a ring isomorphism. 
Remark 16.23. Using Heisenberg monomial classes, one checks that the ring
∗
isomorphism Ψn preserves the pairings on Horb (X (n) ) and Hρ∗n (X [n] ) = H ∗ (X [n] ).
Remark 16.24. The assumption in Theorem 16.22 that the smooth projec-
tive complex surface X is simply connected is not essential. This assumption is
for the convenience that no odd degree cohomology classes appear. The proof of
Theorem 16.22 should work for every smooth projective complex surface X.
 Bibliography

[AGV] D. Abramovich, T. Graber, and A. Vistoli, Algebraic orbifold quantum products, Orb-
ifolds in mathematics and physics (Madison, WI, 2001), Contemp. Math., vol. 310,
Amer. Math. Soc., Providence, RI, 2002, pp. 1–24, DOI 10.1090/conm/310/05397.
MR1950940
[AGT] L. F. Alday, D. Gaiotto, and Y. Tachikawa, Liouville correlation functions from
four-dimensional gauge theories, Lett. Math. Phys. 91 (2010), no. 2, 167–197, DOI
10.1007/s11005-010-0369-5. MR2586871
[AIK] A. B. Altman, A. Iarrobino, and S. L. Kleiman, Irreducibility of the compactified Jaco-
bian, Real and complex singularities (Proc. Ninth Nordic Summer School/NAVF Sym-
pos. Math., Oslo, 1976), Sijthoff and Noordhoff, Alphen aan den Rijn, 1977, pp. 1–12.
MR0498546
[And] G. E. Andrews, The theory of partitions, Addison-Wesley Publishing Co., Reading,
Mass.-London-Amsterdam, 1976. Encyclopedia of Mathematics and its Applications,
Vol. 2. MR0557013
[ACGH] E. Arbarello, M. Cornalba, P. A. Griffiths, and J. Harris, Geometry of algebraic curves.
Vol. I, Grundlehren der Mathematischen Wissenschaften [Fundamental Principles of
Mathematical Sciences], vol. 267, Springer-Verlag, New York, 1985. MR770932
[ABCH] D. Arcara, A. Bertram, I. Coskun, and J. Huizenga, The minimal model program for
the Hilbert scheme of points on P2 and Bridgeland stability, Adv. Math. 235 (2013),
580–626, DOI 10.1016/j.aim.2012.11.018. MR3010070
[AH] M. F. Atiyah and F. Hirzebruch, Vector bundles and homogeneous spaces, Proc. Sym-
pos. Pure Math., Vol. III, American Mathematical Society, Providence, R.I., 1961,
pp. 7–38. MR0139181
[BBC] D. H. Bailey, J. M. Borwein, and R. E. Crandall, Computation and theory of extended
Mordell-Tornheim-Witten sums, Math. Comp. 83 (2014), no. 288, 1795–1821, DOI
10.1090/S0025-5718-2014-02768-3. MR3194130
[BB] V. V. Batyrev and L. A. Borisov, Mirror duality and string-theoretic Hodge numbers,
Invent. Math. 126 (1996), no. 1, 183–203, DOI 10.1007/s002220050093. MR1408560
[BBM] P. Baum, J.-L. Brylinski, and R. MacPherson, Cohomologie équivariante délocalisée
(French, with English summary), C. R. Acad. Sci. Paris Sér. I Math. 300 (1985),
no. 17, 605–608. MR791098
[BC] P. Baum and A. Connes, Chern character for discrete groups, A fête of topology, Aca-
demic Press, Boston, MA, 1988, pp. 163–232, DOI 10.1016/B978-0-12-480440-1.50015-
0. MR928402
[Bea1] A. Beauville, Variétés Kähleriennes dont la première classe de Chern est nulle
(French), J. Differential Geom. 18 (1983), no. 4, 755–782 (1984). MR730926
[Bea2] , Variétés kählériennes compactes avec c1 = 0 (French), Astérisque 126
(1985), 181–192. Geometry of K3 surfaces: moduli and periods (Palaiseau, 1981/1982).
MR785234
[Beh1] K. Behrend, Gromov-Witten invariants in algebraic geometry, Invent. Math. 127
(1997), no. 3, 601–617, DOI 10.1007/s002220050132. MR1431140
[Beh2] , The product formula for Gromov-Witten invariants, J. Algebraic Geom. 8
(1999), no. 3, 529–541. MR1689355
[BF1] K. Behrend and B. Fantechi, The intrinsic normal cone, Invent. Math. 128 (1997),
no. 1, 45–88, DOI 10.1007/s002220050136. MR1437495

325
326 BIBLIOGRAPHY

[BF2] K. Behrend and B. Fantechi, Symmetric obstruction theories and Hilbert schemes
of points on threefolds, Algebra Number Theory 2 (2008), no. 3, 313–345, DOI
10.2140/ant.2008.2.313. MR2407118
[BSG] M. Beltrametti and A. J. Sommese, Zero cycles and kth order embeddings of smooth
projective surfaces, Problems in the theory of surfaces and their classification (Cortona,
1988), Sympos. Math., XXXII, Academic Press, London, 1991, pp. 33–48. With an
appendix by Lothar Göttsche. MR1273371
[BT] A. Bertram and M. Thaddeus, On the quantum cohomology of a symmetric product
of an algebraic curve, Duke Math. J. 108 (2001), no. 2, 329–362, DOI 10.1215/S0012-
7094-01-10825-9. MR1833394
[BO] S. Bloch and A. Okounkov, The character of the infinite wedge representation, Adv.
Math. 149 (2000), no. 1, 1–60, DOI 10.1006/aima.1999.1845. MR1742353
[Boi1] S. Boissière, Chern classes of the tangent bundle on the Hilbert scheme of points on
the affine plane, J. Algebraic Geom. 14 (2005), no. 4, 761–787, DOI 10.1090/S1056-
3911-05-00412-1. MR2147349
[Boi2] , On the McKay correspondences for the Hilbert scheme of points on the
affine plane, Math. Ann. 334 (2006), no. 2, 419–438, DOI 10.1007/s00208-005-0738-z.
MR2207706
[BN] S. Boissière and M. A. Nieper-Wisskirchen, Generating series in the cohomology of
Hilbert schemes of points on surfaces, LMS J. Comput. Math. 10 (2007), 254–270,
DOI 10.1112/S146115700000139X. MR2320831
[Bor] R. E. Borcherds, Vertex algebras, Kac-Moody algebras, and the Monster, Proc.
Nat. Acad. Sci. U.S.A. 83 (1986), no. 10, 3068–3071, DOI 10.1073/pnas.83.10.3068.
MR843307
[Bra1] D. M. Bradley, Multiple q-zeta values, J. Algebra 283 (2005), no. 2, 752–798, DOI
10.1016/j.jalgebra.2004.09.017. MR2111222
[Bra2] , On the sum formula for multiple q-zeta values, Rocky Mountain J. Math. 37
(2007), no. 5, 1427–1434, DOI 10.1216/rmjm/1194275927. MR2382894
[Bri] J. Briançon, Description de Hilbn C{x, y}, Invent. Math. 41 (1977), no. 1, 45–89, DOI
10.1007/BF01390164. MR0457432
[BG] J. Bryan and T. Graber, The crepant resolution conjecture, Algebraic geometry—
Seattle 2005. Part 1, Proc. Sympos. Pure Math., vol. 80, Amer. Math. Soc., Providence,
RI, 2009, pp. 23–42, DOI 10.1090/pspum/080.1/2483931. MR2483931
[BP] J. Bryan and R. Pandharipande, The local Gromov-Witten theory of curves, J. Amer.
Math. Soc. 21 (2008), no. 1, 101–136, DOI 10.1090/S0894-0347-06-00545-5. With an
appendix by Bryan, C. Faber, A. Okounkov and Pandharipande. MR2350052
[Can] M. B. Can, Nested Hilbert schemes and the nested q, t-Catalan series (English, with
English and French summaries), 20th Annual International Conference on Formal Power
Series and Algebraic Combinatorics (FPSAC 2008), Discrete Math. Theor. Comput.
Sci. Proc., AJ, Assoc. Discrete Math. Theor. Comput. Sci., Nancy, 2008, pp. 61–69.
MR2721442
[Car1] E. Carlsson, Vertex operators and moduli spaces of sheaves, ProQuest LLC, Ann Arbor,
MI, 2008. Thesis (Ph.D.)–Princeton University. MR2711527
[Car2] , Vertex operators and quasimodularity of Chern numbers on the Hilbert scheme,
Adv. Math. 229 (2012), no. 5, 2888–2907, DOI 10.1016/j.aim.2011.10.003. MR2889150
[CO] E. Carlsson and A. Okounkov, Exts and vertex operators, Duke Math. J. 161 (2012),
no. 9, 1797–1815, DOI 10.1215/00127094-1593380. MR2942794
[dCM] M. A. A. de Cataldo and L. Migliorini, The Douady space of a complex surface, Adv.
Math. 151 (2000), no. 2, 283–312, DOI 10.1006/aima.1999.1896. MR1758249
[CG] F. Catanese and L. Gœttsche, d-very-ample line bundles and embeddings of
Hilbert schemes of 0-cycles, Manuscripta Math. 68 (1990), no. 3, 337–341, DOI
10.1007/BF02568768. MR1065935
[CL] H.-l. Chang and J. Li, Semi-perfect obstruction theory and Donaldson-Thomas in-
variants of derived objects, Comm. Anal. Geom. 19 (2011), no. 4, 807–830, DOI
10.4310/CAG.2011.v19.n4.a6. MR2880216
[Che1] J. Cheah, On the cohomology of Hilbert schemes of points, J. Algebraic Geom. 5 (1996),
no. 3, 479–511. MR1382733
 BIBLIOGRAPHY 327

[Che2] , Cellular decompositions for nested Hilbert schemes of points, Pacific J. Math.
183 (1998), no. 1, 39–90, DOI 10.2140/pjm.1998.183.39. MR1616606
[Che3] , The virtual Hodge polynomials of nested Hilbert schemes and related varieties,
Math. Z. 227 (1998), no. 3, 479–504, DOI 10.1007/PL00004387. MR1612677
[CR1] W. Chen and Y. Ruan, Orbifold Gromov-Witten theory, Orbifolds in mathematics and
physics (Madison, WI, 2001), Contemp. Math., vol. 310, Amer. Math. Soc., Providence,
RI, 2002, pp. 25–85, DOI 10.1090/conm/310/05398. MR1950941
[CR2] , A new cohomology theory of orbifold, Comm. Math. Phys. 248 (2004), no. 1,
1–31, DOI 10.1007/s00220-004-1089-4. MR2104605
[Cheo1] Wan Keng Cheong, Orbifold quantum cohomology of the symmetric product of Ar .
Preprint. arXiv:0910.0629
[Cheo2] W. K. Cheong, Strengthening the cohomological crepant resolution conjecture for
Hilbert-Chow morphisms, Math. Ann. 356 (2013), no. 1, 45–72, DOI 10.1007/s00208-
012-0833-x. MR3038121
[ChG] W. K. Cheong and A. Gholampour, Orbifold Gromov-Witten theory of the symmetric
product of Ar , Geom. Topol. 16 (2012), no. 1, 475–527, DOI 10.2140/gt.2012.16.475.
MR2916292
[Coa] T. Coates, On the crepant resolution conjecture in the local case, Comm. Math. Phys.
287 (2009), no. 3, 1071–1108, DOI 10.1007/s00220-008-0715-y. MR2486673
[CCIT] T. Coates, H. Iritani, and H.-H. Tseng, Wall-crossings in toric Gromov-Witten
theory. I. Crepant examples, Geom. Topol. 13 (2009), no. 5, 2675–2744, DOI
10.2140/gt.2009.13.2675. MR2529944
[CR] T. Coates and Y. Ruan, Quantum cohomology and crepant resolutions: a conjecture
(English, with English and French summaries), Ann. Inst. Fourier (Grenoble) 63 (2013),
no. 2, 431–478. MR3112518
[Cos] K. Costello, Higher genus Gromov-Witten invariants as genus zero invariants of sym-
metric products, Ann. of Math. (2) 164 (2006), no. 2, 561–601, DOI 10.4007/an-
nals.2006.164.561. MR2247968
[CoG] K. Costello, I. Grojnowski, Hilbert schemes, Hecke algebras and the Calogero-
Sutherland system. Preprint. math.AG/0310189.
[CK] D. A. Cox and S. Katz, Mirror symmetry and algebraic geometry, Mathematical Sur-
veys and Monographs, vol. 68, American Mathematical Society, Providence, RI, 1999.
MR1677117
[DHVW] L. Dixon, J. A. Harvey, C. Vafa, and E. Witten, Strings on orbifolds, Nuclear Phys. B
261 (1985), no. 4, 678–686, DOI 10.1016/0550-3213(85)90593-0. MR818423
[ELQ] D. Edidin, W.-P. Li, and Z. Qin, Gromov-Witten invariants of the Hilbert scheme of 3-
points on P2 , Asian J. Math. 7 (2003), no. 4, 551–574, DOI 10.4310/AJM.2003.v7.n4.a7.
MR2074891
[EG] G. Ellingsrud and L. Göttsche, Hilbert schemes of points and Heisenberg algebras,
School on Algebraic Geometry (Trieste, 1999), ICTP Lect. Notes, vol. 1, Abdus Salam
Int. Cent. Theoret. Phys., Trieste, 2000, pp. 59–100. MR1795861
[EGL] G. Ellingsrud, L. Göttsche, and M. Lehn, On the cobordism class of the Hilbert scheme
of a surface, J. Algebraic Geom. 10 (2001), no. 1, 81–100. MR1795551
[EL] G. Ellingsrud and M. Lehn, Irreducibility of the punctual quotient scheme of a surface,
Ark. Mat. 37 (1999), no. 2, 245–254, DOI 10.1007/BF02412213. MR1714770
[ES1] G. Ellingsrud and S. A. Strømme, On the homology of the Hilbert scheme of points
in the plane, Invent. Math. 87 (1987), no. 2, 343–352, DOI 10.1007/BF01389419.
MR870732
[ES2] , Towards the Chow ring of the Hilbert scheme of P2 , J. Reine Angew. Math.
441 (1993), 33–44. MR1228610
[ES3] , An intersection number for the punctual Hilbert scheme of a surface, Trans.
Amer. Math. Soc. 350 (1998), no. 6, 2547–2552, DOI 10.1090/S0002-9947-98-01972-2.
MR1432198
[EtG] P. Etingof and V. Ginzburg, Symplectic reflection algebras, Calogero-Moser space, and
deformed Harish-Chandra homomorphism, Invent. Math. 147 (2002), no. 2, 243–348,
DOI 10.1007/s002220100171. MR1881922
[FPa] C. Faber and R. Pandharipande, Hodge integrals and Gromov-Witten theory, Invent.
Math. 139 (2000), no. 1, 173–199, DOI 10.1007/s002229900028. MR1728879
328 BIBLIOGRAPHY

[FG1] B. Fantechi and L. Göttsche, Orbifold cohomology for global quotients, Duke Math. J.
117 (2003), no. 2, 197–227, DOI 10.1215/S0012-7094-03-11721-4. MR1971293
[FG2] , Local properties and Hilbert schemes of points, Fundamental algebraic ge-
ometry, Math. Surveys Monogr., vol. 123, Amer. Math. Soc., Providence, RI, 2005,
pp. 139–178. MR2223408
[Fog1] J. Fogarty, Algebraic families on an algebraic surface, Amer. J. Math 90 (1968), 511–
521, DOI 10.2307/2373541. MR0237496
[Fog2] J. Fogarty, Algebraic families on an algebraic surface. II. The Picard scheme of the
punctual Hilbert scheme, Amer. J. Math. 95 (1973), 660–687, DOI 10.2307/2373734.
MR0335512
[FU] D. S. Freed and K. K. Uhlenbeck, Instantons and four-manifolds, 2nd ed., Mathemat-
ical Sciences Research Institute Publications, vol. 1, Springer-Verlag, New York, 1991.
MR1081321
[FB] E. Frenkel and D. Ben-Zvi, Vertex algebras and algebraic curves, Mathematical Sur-
veys and Monographs, vol. 88, American Mathematical Society, Providence, RI, 2001.
MR1849359
[FKRW] E. Frenkel, V. Kac, A. Radul, and W. Wang, W1+∞ and W(glN ) with central charge
N , Comm. Math. Phys. 170 (1995), no. 2, 337–357. MR1334399
[FJW] I. B. Frenkel, N. Jing, and W. Wang, Vertex representations via finite groups and
the McKay correspondence, Internat. Math. Res. Notices 4 (2000), 195–222, DOI
10.1155/S107379280000012X. MR1747618
[FW] I. B. Frenkel and W. Wang, Virasoro algebra and wreath product convolution, J. Algebra
242 (2001), no. 2, 656–671, DOI 10.1006/jabr.2001.8860. MR1848963
[Fu] Y. Fu, Quantum cohomology of a Hilbert scheme of a Hirzebruch surface, ProQuest
LLC, Ann Arbor, MI, 2010. Thesis (Ph.D.)–University of Illinois at Urbana-Champaign.
MR2941780
[FN] A. Fujiki and S. Nakano, Supplement to “On the inverse of monoidal transformation”,
Publ. Res. Inst. Math. Sci. 7 (1971/72), 637–644, DOI 10.2977/prims/1195193401.
MR0294712
[FO] K. Fukaya and K. Ono, Arnold conjecture and Gromov-Witten invariant for general
symplectic manifolds, The Arnoldfest (Toronto, ON, 1997), Fields Inst. Commun.,
vol. 24, Amer. Math. Soc., Providence, RI, 1999, pp. 173–190. MR1733575
[Ful] W. Fulton, Introduction to toric varieties, Annals of Mathematics Studies, vol. 131,
Princeton University Press, Princeton, NJ, 1993. The William H. Roever Lectures in
Geometry. MR1234037
[FP] W. Fulton and R. Pandharipande, Notes on stable maps and quantum cohomology,
Algebraic geometry—Santa Cruz 1995, Proc. Sympos. Pure Math., vol. 62, Amer.
Math. Soc., Providence, RI, 1997, pp. 45–96, DOI 10.1090/pspum/062.2/1492534.
MR1492534
[Get] E. Getzler, Intersection theory on M1,4 and elliptic Gromov-Witten invariants, J.
Amer. Math. Soc. 10 (1997), no. 4, 973–998, DOI 10.1090/S0894-0347-97-00246-4.
MR1451505
[GSY1] A. Gholampour, A. Sheshmani, S.-T. Yau, Nested Hilbert schemes on surfaces: virtual
fundamental class. Preprint. arXiv.1701.08899.
[GSY2] A. Gholampour, A. Sheshmani, S.-T. Yau, Localized Donaldson-Thomas theory of sur-
faces. Preprint. arXiv.1701.08902.
[Got1] L. Göttsche, The Betti numbers of the Hilbert scheme of points on a smooth pro-
jective surface, Math. Ann. 286 (1990), no. 1-3, 193–207, DOI 10.1007/BF01453572.
MR1032930
[Got2] , Hilbert schemes of zero-dimensional subschemes of smooth varieties, Lecture
Notes in Mathematics, vol. 1572, Springer-Verlag, Berlin, 1994. MR1312161
[Got3] , On the motive of the Hilbert scheme of points on a surface, Math. Res. Lett.
8 (2001), no. 5-6, 613–627, DOI 10.4310/MRL.2001.v8.n5.a3. MR1879805
[GS] L. Göttsche and W. Soergel, Perverse sheaves and the cohomology of Hilbert
schemes of smooth algebraic surfaces, Math. Ann. 296 (1993), no. 2, 235–245, DOI
10.1007/BF01445104. MR1219901
 BIBLIOGRAPHY 329

[Gou] I. P. Goulden, A differential operator for symmetric functions and the combinatorics of
multiplying transpositions, Trans. Amer. Math. Soc. 344 (1994), no. 1, 421–440, DOI
10.2307/2154724. MR1249468
[GJV] I. Goulden, D. Jackson, R. Vakil, Towards the geometry of double Hurwitz numbers,
math.AG/0309440.
[GP] T. Graber and R. Pandharipande, Localization of virtual classes, Invent. Math. 135
(1999), no. 2, 487–518, DOI 10.1007/s002220050293. MR1666787
[Groj] I. Grojnowski, Instantons and affine algebras. I. The Hilbert scheme and vertex oper-
ators, Math. Res. Lett. 3 (1996), no. 2, 275–291, DOI 10.4310/MRL.1996.v3.n2.a12.
MR1386846
[Grot] A. Grothendieck, Techniques de construction et théorèmes d’existence en géométrie
algébrique. IV. Les schémas de Hilbert (French), Séminaire Bourbaki, Vol. 6, Soc.
Math. France, Paris, 1995, pp. Exp. No. 221, 249–276. MR1611822
[Hai1] M. Haiman, Hilbert schemes, polygraphs and the Macdonald positivity conjecture, J.
Amer. Math. Soc. 14 (2001), no. 4, 941–1006, DOI 10.1090/S0894-0347-01-00373-3.
MR1839919
[Hai2] , Vanishing theorems and character formulas for the Hilbert scheme of points
in the plane, Invent. Math. 149 (2002), no. 2, 371–407, DOI 10.1007/s002220200219.
MR1918676
[Hai3] , Combinatorics, symmetric functions, and Hilbert schemes, Current develop-
ments in mathematics, 2002, Int. Press, Somerville, MA, 2003, pp. 39–111. MR2051783
[Harr] J. Harris, Algebraic geometry, Graduate Texts in Mathematics, vol. 133, Springer-
Verlag, New York, 1992. A first course. MR1182558
[Hart] R. Hartshorne, Algebraic geometry, Springer-Verlag, New York-Heidelberg, 1977. Grad-
uate Texts in Mathematics, No. 52. MR0463157
[Hern] D. Hernandez, An introduction to affine Kac-Moody algebras. DEA. Lecture notes from
CTQM Master Class, Aarhus University, Denmark, October 2006.
[HLQ] J. Hu, W.-P. Li, and Z. Qin, The Gromov-Witten invariants of the Hilbert schemes of
points on surfaces with pg > 0, Internat. J. Math. 26 (2015), no. 1, 1550009, 26, DOI
10.1142/S0129167X15500093. MR3313654
[Iar1] A. A. Iarrobino, Punctual Hilbert schemes, Mem. Amer. Math. Soc. 10 (1977), no. 188,
viii+112, DOI 10.1090/memo/0188. MR0485867
[Iar2] A. Iarrobino, Hilbert scheme of points: overview of last ten years, Algebraic geometry,
Bowdoin, 1985 (Brunswick, Maine, 1985), Proc. Sympos. Pure Math., vol. 46, Amer.
Math. Soc., Providence, RI, 1987, pp. 297–320. MR927986
[IN] Y. Ito and I. Nakamura, McKay correspondence and Hilbert schemes, Proc. Japan
Acad. Ser. A Math. Sci. 72 (1996), no. 7, 135–138. MR1420598
[Juc] A.-A. A. Jucys, Symmetric polynomials and the center of the symmetric group ring,
Rep. Mathematical Phys. 5 (1974), no. 1, 107–112. MR0419576
[Kac] V. Kac, Vertex algebras for beginners, 2nd ed., University Lecture Series, vol. 10, Amer-
ican Mathematical Society, Providence, RI, 1998. MR1651389
[Kap] S. Kapfer, Computing cup products in integral cohomology of Hilbert schemes of
points on K3 surfaces, LMS J. Comput. Math. 19 (2016), no. 1, 78–97, DOI
10.1112/S1461157016000012. MR3542518
[KaM] S. Kapfer, G. Menet, Integral cohomology of the generalized Kummer fourfold. Preprint.
[KMM] Y. Kawamata, K. Matsuda, and K. Matsuki, Introduction to the minimal model prob-
lem, Algebraic geometry, Sendai, 1985, Adv. Stud. Pure Math., vol. 10, North-Holland,
Amsterdam, 1987, pp. 283–360. MR946243
[KL1] Y. Kiem, J. Li, Gromov-Witten invariants of varieties with holomorphic 2-forms.
Preprint. arXiv:0707.2986
[KL2] Y.-H. Kiem and J. Li, Localizing virtual cycles by cosections, J. Amer. Math. Soc. 26
(2013), no. 4, 1025–1050, DOI 10.1090/S0894-0347-2013-00768-7. MR3073883
[King] A. D. King, Moduli of representations of finite-dimensional algebras, Quart. J. Math.
Oxford Ser. (2) 45 (1994), no. 180, 515–530, DOI 10.1093/qmath/45.4.515. MR1315461
[KM] M. Kontsevich and Yu. Manin, Gromov-Witten classes, quantum cohomology, and enu-
merative geometry, Comm. Math. Phys. 164 (1994), no. 3, 525–562. MR1291244
330 BIBLIOGRAPHY

[Kuhn] N. J. Kuhn, Character rings in algebraic topology, Advances in homotopy theory

(Cortona, 1988), London Math. Soc. Lecture Note Ser., vol. 139, Cambridge Univ.
Press, Cambridge, 1989, pp. 111–126. MR1055872
[LT] A. Lascoux and J.-Y. Thibon, Vertex operators and the class algebras of
symmetric groups (English, with English and Russian summaries), Zap.
Nauchn. Sem. S.-Peterburg. Otdel. Mat. Inst. Steklov. (POMI) 283 (2001),
no. Teor. Predst. Din. Sist. Komb. i Algoritm. Metody. 6, 156–177, 261, DOI
10.1023/B:JOTH.0000024619.77778.3d; English transl., J. Math. Sci. (N. Y.) 121
(2004), no. 3, 2380–2392. MR1879068
[LY] H. B. Lawson Jr. and S. S.-T. Yau, Holomorphic symmetries, Ann. Sci. École Norm.
Sup. (4) 20 (1987), no. 4, 557–577. MR932798
[Leh1] M. Lehn, Chern classes of tautological sheaves on Hilbert schemes of points on surfaces,
Invent. Math. 136 (1999), no. 1, 157–207, DOI 10.1007/s002220050307. MR1681097
[Leh2] , Lectures on Hilbert schemes, Algebraic structures and moduli spaces, CRM
Proc. Lecture Notes, vol. 38, Amer. Math. Soc., Providence, RI, 2004, pp. 1–30.
MR2095898
[LS1] M. Lehn and C. Sorger, Symmetric groups and the cup product on the cohomology of
Hilbert schemes, Duke Math. J. 110 (2001), no. 2, 345–357, DOI 10.1215/S0012-7094-
01-11026-0. MR1865244
[LS2] , The cup product of Hilbert schemes for K3 surfaces, Invent. Math. 152 (2003),
no. 2, 305–329, DOI 10.1007/s00222-002-0270-7. MR1974889
[LZ] C. Li, X. Zhao, The MMP for deformations of Hilbert schemes of points on the pro-
jective plane. Algebraic Geometry (to appear).
[LiJ] J. Li, Zero dimensional Donaldson-Thomas invariants of threefolds, Geom. Topol. 10
(2006), 2117–2171, DOI 10.2140/gt.2006.10.2117. MR2284053
[LL] J. Li and W.-P. Li, Two point extremal Gromov-Witten invariants of Hilbert schemes
of points on surfaces, Math. Ann. 349 (2011), no. 4, 839–869, DOI 10.1007/s00208-
010-0542-2. MR2777035
[LT1] J. Li and G. Tian, Virtual moduli cycles and Gromov-Witten invariants of algebraic
varieties, J. Amer. Math. Soc. 11 (1998), no. 1, 119–174, DOI 10.1090/S0894-0347-98-
00250-1. MR1467172
[LT2] , Virtual moduli cycles and Gromov-Witten invariants of general symplectic
manifolds, Topics in symplectic 4-manifolds (Irvine, CA, 1996), First Int. Press Lect.
Ser., I, Int. Press, Cambridge, MA, 1998, pp. 47–83. MR1635695
[LT3] , Comparison of algebraic and symplectic Gromov-Witten invariants, Asian J.
Math. 3 (1999), no. 3, 689–728, DOI 10.4310/AJM.1999.v3.n3.a7. MR1793677
[LQ1] W.-P. Li and Z. Qin, On 1-point Gromov-Witten invariants of the Hilbert schemes of
points on surfaces, Turkish J. Math. 26 (2002), no. 1, 53–68. MR1892799
[LQ2] W.-P. Li, Z. Qin, Incidence Hilbert schemes and infinite dimensional Lie algebras.
Proceedings of Fourth International Congress of Chinese Mathematicians (Hangzhou,
2007), 408-441, Vol II.
[LQ3] , Integral cohomology of Hilbert schemes of points on surfaces, Comm. Anal.
Geom. 16 (2008), no. 5, 969–988, DOI 10.4310/CAG.2008.v16.n5.a3. MR2471364
[LQ4] , Equivariant cohomology of incidence Hilbert schemes and infinite dimensional
Lie algebras, Manuscripta Math. 133 (2010), no. 3-4, 519–544, DOI 10.1007/s00229-
010-0387-8. MR2729266
[LQ5] , The cohomological crepant resolution conjecture for the Hilbert-Chow mor-
phisms, J. Differential Geom. 104 (2016), no. 3, 499–557. MR3568629
[LQW1] W.-p. Li, Z. Qin, and W. Wang, Vertex algebras and the cohomology ring structure of
Hilbert schemes of points on surfaces, Math. Ann. 324 (2002), no. 1, 105–133, DOI
10.1007/s002080200330. MR1931760
[LQW2] , Generators for the cohomology ring of Hilbert schemes of points on surfaces,
Internat. Math. Res. Notices 20 (2001), 1057–1074, DOI 10.1155/S1073792801000502.
MR1857595
[LQW3] W.-P. Li, Z. Qin, and W. Wang, Stability of the cohomology rings of Hilbert
schemes of points on surfaces, J. Reine Angew. Math. 554 (2003), 217–234, DOI
10.1515/crll.2003.007. MR1952174
 BIBLIOGRAPHY 331

[LQW4] , Hilbert schemes and W algebras, Int. Math. Res. Not. 27 (2002), 1427–1456,
DOI 10.1155/S1073792802110129. MR1908477
[LQW5] , Ideals of the cohomology rings of Hilbert schemes and their applications, Trans.
Amer. Math. Soc. 356 (2004), no. 1, 245–265, DOI 10.1090/S0002-9947-03-03422-6.
MR2020031
[LQW6] , Hilbert schemes, integrable hierarchies, and Gromov-Witten theory, Int. Math.
Res. Not. 40 (2004), 2085–2104, DOI 10.1155/S1073792804140452. MR2064317
[LQW7] , The cohomology rings of Hilbert schemes via Jack polynomials, Algebraic
structures and moduli spaces, CRM Proc. Lecture Notes, vol. 38, Amer. Math. Soc.,
Providence, RI, 2004, pp. 249–258. MR2096149
[LQW8] , Hilbert scheme intersection numbers, Hurwitz numbers, and Gromov-Witten
invariants, Infinite-dimensional aspects of representation theory and applications, Con-
temp. Math., vol. 392, Amer. Math. Soc., Providence, RI, 2005, pp. 67–81, DOI
10.1090/conm/392/07354. MR2189871
[LQZ] W.-P. Li, Z. Qin, and Q. Zhang, Curves in the Hilbert schemes of points on
surfaces, Vector bundles and representation theory (Columbia, MO, 2002), Con-
temp. Math., vol. 322, Amer. Math. Soc., Providence, RI, 2003, pp. 89–96, DOI
10.1090/conm/322/05681. MR1987741
[Mac1] I. G. Macdonald, The Poincaré polynomial of a symmetric product, Proc. Cambridge
Philos. Soc. 58 (1962), 563–568. MR0143204
[Mac2] , Symmetric functions and Hall polynomials, 2nd ed., Oxford Mathematical
Monographs, The Clarendon Press, Oxford University Press, New York, 1995. With
contributions by A. Zelevinsky; Oxford Science Publications. MR1354144
[Mar1] E. Markman, Generators of the cohomology ring of moduli spaces of sheaves on sym-
plectic surfaces, J. Reine Angew. Math. 544 (2002), 61–82, DOI 10.1515/crll.2002.028.
MR1887889
[Mar2] , Integral generators for the cohomology ring of moduli spaces of
sheaves over Poisson surfaces, Adv. Math. 208 (2007), no. 2, 622–646, DOI
10.1016/j.aim.2006.03.006. MR2304330
[Mat] Y. Matsuo, Matrix theory, Hilbert scheme and integrable system, Modern Phys. Lett.
A 13 (1998), no. 34, 2731–2742, DOI 10.1142/S0217732398002904. MR1660859
[Mau] D. Maulik, Gromov-Witten theory of An -resolutions, Geom. Topol. 13 (2009), no. 3,
1729–1773, DOI 10.2140/gt.2009.13.1729. MR2496055
[MNOP1] D. Maulik, N. Nekrasov, A. Okounkov, and R. Pandharipande, Gromov-Witten theory
and Donaldson-Thomas theory. I, Compos. Math. 142 (2006), no. 5, 1263–1285, DOI
10.1112/S0010437X06002302. MR2264664
[MNOP2] , Gromov-Witten theory and Donaldson-Thomas theory. II, Compos. Math. 142
(2006), no. 5, 1286–1304, DOI 10.1112/S0010437X06002314. MR2264665
[MO1] D. Maulik and A. Oblomkov, Quantum cohomology of the Hilbert scheme of points on
An -resolutions, J. Amer. Math. Soc. 22 (2009), no. 4, 1055–1091, DOI 10.1090/S0894-
0347-09-00632-8. MR2525779
[MO2] , Donaldson-Thomas theory of An × P 1 , Compos. Math. 145 (2009), no. 5,
1249–1276, DOI 10.1112/S0010437X09003972. MR2551996
[MJD] T. Miwa, M. Jimbo, and E. Date, Solitons, Cambridge Tracts in Mathematics, vol. 135,
Cambridge University Press, Cambridge, 2000. Differential equations, symmetries and
infinite-dimensional algebras; Translated from the 1993 Japanese original by Miles Reid.
MR1736222
[Mod] L. J. Mordell, On the evaluation of some multiple series, J. London Math. Soc. 33
(1958), 368–371, DOI 10.1112/jlms/s1-33.3.368. MR0100181
[Mum] D. Mumford, Towards an enumerative geometry of the moduli space of curves, Arith-
metic and geometry, Vol. II, Progr. Math., vol. 36, Birkhäuser Boston, Boston, MA,
1983, pp. 271–328. MR717614
[MF] D. Mumford and J. Fogarty, Geometric invariant theory, 2nd ed., Ergebnisse der Math-
ematik und ihrer Grenzgebiete [Results in Mathematics and Related Areas], vol. 34,
Springer-Verlag, Berlin, 1982. MR719371
[Mu] G. E. Murphy, A new construction of Young’s seminormal representation of the sym-
metric groups, J. Algebra 69 (1981), no. 2, 287–297, DOI 10.1016/0021-8693(81)90205-
2. MR617079
332 BIBLIOGRAPHY

[Mur] R. Murphy, A transition matrix for two bases of the integral cohomology of the Hilbert
scheme of points in the projective plane, ProQuest LLC, Ann Arbor, MI, 2010. Thesis
(Ph.D.)–University of Missouri - Columbia. MR2982352
[Nag] K. Nagao, Quiver varieties and Frenkel-Kac construction, J. Algebra 321 (2009),
no. 12, 3764–3789, DOI 10.1016/j.jalgebra.2009.03.012. MR2517812
[Nak1] H. Nakajima, Instantons on ALE spaces, quiver varieties, and Kac-Moody alge-
bras, Duke Math. J. 76 (1994), no. 2, 365–416, DOI 10.1215/S0012-7094-94-07613-8.
MR1302318
[Nak2] , Gauge theory on resolutions of simple singularities and simple Lie alge-
bras, Internat. Math. Res. Notices 2 (1994), 61–74, DOI 10.1155/S1073792894000085.
MR1264929
[Nak3] , Heisenberg algebra and Hilbert schemes of points on projective surfaces, Ann.
of Math. (2) 145 (1997), no. 2, 379–388, DOI 10.2307/2951818. MR1441880
[Nak4] , Quiver varieties and Kac-Moody algebras, Duke Math. J. 91 (1998), no. 3,
515–560, DOI 10.1215/S0012-7094-98-09120-7. MR1604167
[Nak5] , Lectures on Hilbert schemes of points on surfaces, University Lecture Series,
vol. 18, American Mathematical Society, Providence, RI, 1999. MR1711344
[Nak6] , The Heisenberg algebra and Hilbert schemes of points on surfaces (Japanese),
Sūgaku 50 (1998), no. 4, 385–398. MR1690690
[Nak7] , More lectures on Hilbert schemes of points on surfaces, Development of moduli
theory—Kyoto 2013, Adv. Stud. Pure Math., vol. 69, Math. Soc. Japan, [Tokyo], 2016,
pp. 173–205. MR3586508
[Naka] S. Nakano, On the inverse of monoidal transformation, Publ. Res. Inst. Math. Sci. 6
(1970/71), 483–502, DOI 10.2977/prims/1195193917. MR0294710
[Neg] A. Neguţ, Exts and the AGT relations, Lett. Math. Phys. 106 (2016), no. 9, 1265–1316,
DOI 10.1007/s11005-016-0865-3. MR3533570
[NO] N. A. Nekrasov and A. Okounkov, Seiberg-Witten theory and random partitions, The
unity of mathematics, Progr. Math., vol. 244, Birkhäuser Boston, Boston, MA, 2006,
pp. 525–596, DOI 10.1007/0-8176-4467-9 15. MR2181816
[Nie1] M. A. Nieper-Wisskirchen, Characteristic classes of the Hilbert schemes of points on
non-compact simply-connected surfaces, JP J. Geom. Topol. 8 (2008), no. 1, 7–21.
MR2444822
[Nie2] , Twisted cohomology of the Hilbert schemes of points on surfaces, Doc. Math.
14 (2009), 749–770. MR2578804
[Obe] G. Oberdieck, Gromov-Witten invariants of the Hilbert scheme of points of a K3 sur-
face. Preprint. arXiv:1406.1139
[Ok1] A. Okounkov, Toda equations for Hurwitz numbers, Math. Res. Lett. 7 (2000), no. 4,
447–453, DOI 10.4310/MRL.2000.v7.n4.a10. MR1783622
[Ok2] A. Yu. Okunkov, Hilbert schemes and multiple q-zeta values (Russian, with Russian
summary), Funktsional. Anal. i Prilozhen. 48 (2014), no. 2, 79–87; English transl.,
Funct. Anal. Appl. 48 (2014), no. 2, 138–144. MR3288178
[OP1] A. Okounkov and R. Pandharipande, Gromov-Witten theory, Hurwitz theory, and
completed cycles, Ann. of Math. (2) 163 (2006), no. 2, 517–560, DOI 10.4007/an-
nals.2006.163.517. MR2199225
[OP2] , The equivariant Gromov-Witten theory of P1 , Ann. of Math. (2) 163 (2006),
no. 2, 561–605, DOI 10.4007/annals.2006.163.561. MR2199226
[OP3] , Quantum cohomology of the Hilbert scheme of points in the plane, Invent.
Math. 179 (2010), no. 3, 523–557, DOI 10.1007/s00222-009-0223-5. MR2587340
[OP4] , The quantum differential equation of the Hilbert scheme of points in the
plane, Transform. Groups 15 (2010), no. 4, 965–982, DOI 10.1007/s00031-010-9116-
3. MR2753265
[OP5] , The local Donaldson-Thomas theory of curves, Geom. Topol. 14 (2010), no. 3,
1503–1567, DOI 10.2140/gt.2010.14.1503. MR2679579
[OT] J.-i. Okuda and Y. Takeyama, On relations for the multiple q-zeta values, Ramanujan
J. 14 (2007), no. 3, 379–387, DOI 10.1007/s11139-007-9053-5. MR2357443
[PTse] R. Pandharipande, H.-H. Tseng, Higher genus Gromov-Witten theory of the Hilbert
scheme of points of the plane and CohFTs associated to local curves. Preprint.
arXiv:1707.01406
 BIBLIOGRAPHY 333

[QT] Z. Qin and Y. Tu, The nef cones of and minimal-degree curves in the Hilbert schemes
of points on certain surfaces, Pacific J. Math. 284 (2016), no. 2, 439–453, DOI
10.2140/pjm.2016.284.439. MR3544309
[QW1] Z. Qin and W. Wang, Hilbert schemes and symmetric products: a dictionary, Orb-
ifolds in mathematics and physics (Madison, WI, 2001), Contemp. Math., vol. 310,
Amer. Math. Soc., Providence, RI, 2002, pp. 233–257, DOI 10.1090/conm/310/05407.
MR1950950
[QW2] , Integral operators and integral cohomology classes of Hilbert schemes, Math.
Ann. 331 (2005), no. 3, 669–692, DOI 10.1007/s00208-004-0602-6. MR2122545
[QW3] , Hilbert schemes of points on the minimal resolution and soliton equations, Lie
algebras, vertex operator algebras and their applications, Contemp. Math., vol. 442,
Amer. Math. Soc., Providence, RI, 2007, pp. 435–462, DOI 10.1090/conm/442/08541.
MR2372578
[QY] Z. Qin, F. Yu, On Okounkov’s conjecture connecting Hilbert schemes of points and mul-
tiple q-zeta values. Intern. Math. Res. Notices (to appear). Published online: December
16, 2016.
[Reid] M. Reid, La correspondance de McKay, Astérisque 276 (2002), 53–72. Séminaire Bour-
baki, Vol. 1999/2000. MR1886756
[Rua1] Y. Ruan, Topological sigma model and Donaldson-type invariants in Gromov the-
ory, Duke Math. J. 83 (1996), no. 2, 461–500, DOI 10.1215/S0012-7094-96-08316-7.
MR1390655
[Rua2] , The cohomology ring of crepant resolutions of orbifolds, Gromov-Witten theory
of spin curves and orbifolds, Contemp. Math., vol. 403, Amer. Math. Soc., Providence,
RI, 2006, pp. 117–126, DOI 10.1090/conm/403/07597. MR2234886
[Seg] G. Segal, Equivariant K-theory and symmetric products. Lecture at Cambridge, July
1996.
[Sim] C. T. Simpson, Moduli of representations of the fundamental group of a smooth pro-
jective variety. I, Inst. Hautes Études Sci. Publ. Math. 79 (1994), 47–129. MR1307297
[Tik] A. S. Tikhomirov, Standard bundles on a Hilbert scheme of points on a surface, Al-
gebraic geometry and its applications (Yaroslavl, 1992), Aspects Math., E25, Friedr.
Vieweg, Braunschweig, 1994, pp. 183–203. MR1282029
[Tor] L. Tornheim, Harmonic double series, Amer. J. Math. 72 (1950), 303–314, DOI
10.2307/2372034. MR0034860
[Tse] H.-H. Tseng, A tale of four theories. Talk at the conference “Geometry of Moduli
Spaces”. University of California at San Diego, May 2017.
[UT] K. Ueno and K. Takasaki, Toda lattice hierarchy, Group representations and systems
of differential equations (Tokyo, 1982), Adv. Stud. Pure Math., vol. 4, North-Holland,
Amsterdam, 1984, pp. 1–95. MR810623
[Uri] B. Uribe, Orbifold cohomology of the symmetric product, Comm. Anal. Geom. 13
(2005), no. 1, 113–128. MR2154668
[VW] C. Vafa and E. Witten, A strong coupling test of S-duality, Nuclear Phys. B 431 (1994),
no. 1-2, 3–77, DOI 10.1016/0550-3213(94)90097-3. MR1305096
[Vas] E. Vasserot, Sur l’anneau de cohomologie du schéma de Hilbert de C2 (French, with
English and French summaries), C. R. Acad. Sci. Paris Sér. I Math. 332 (2001), no. 1,
7–12, DOI 10.1016/S0764-4442(00)01766-3. MR1805619
[Wan1] W. Wang, Equivariant K-theory, wreath products, and Heisenberg algebra, Duke Math.
J. 103 (2000), no. 1, 1–23, DOI 10.1215/S0012-7094-00-10311-0. MR1758236
[Wan2] , Algebraic structures behind Hilbert schemes and wreath products, Recent devel-
opments in infinite-dimensional Lie algebras and conformal field theory (Charlottesville,
VA, 2000), Contemp. Math., vol. 297, Amer. Math. Soc., Providence, RI, 2002, pp. 271–
295, DOI 10.1090/conm/297/05102. MR1919822
[Wan3] , The Farahat-Higman ring of wreath products and Hilbert schemes, Adv. Math.
187 (2004), no. 2, 417–446, DOI 10.1016/j.aim.2003.09.003. MR2078343
[Wan4] , Vertex algebras and the class algebras of wreath products, Proc. London Math.
Soc. (3) 88 (2004), no. 2, 381–404, DOI 10.1112/S0024611503014382. MR2032512
[Wit] E. Witten, On quantum gauge theories in two dimensions, Comm. Math. Phys. 141
(1991), no. 1, 153–209. MR1133264
334 BIBLIOGRAPHY

[Zas] E. Zaslow, Topological orbifold models and quantum cohomology rings, Comm. Math.
Phys. 156 (1993), no. 2, 301–331. MR1233848
[Zhao] J. Zhao, Uniform approach to double shuffle and duality relations of various q-analogs
of multiple zeta values via Rota-Baxter algebras. Preprint. arXiv:1412.8044
[Zud] V. V. Zudilin, Algebraic relations for multiple zeta values (Russian, with
Russian summary), Uspekhi Mat. Nauk 58 (2003), no. 1(349), 3–32, DOI
10.1070/RM2003v058n01ABEH000592; English transl., Russian Math. Surveys 58
(2003), no. 1, 1–29. MR1992130
 Index

N -point function, 238 expected dimension, 232

W algebra, 92 Ext operator, 70
τ -function, 242
k-very ample, 16 FH ring, 173
q-trace, 237 forgetful map, 233
forgotten symmetric function, 7
ADHM datum, 44 Fundamental Class Axiom, 232
adjoint operator, 46
affine Kac-Moody algebra sl   , 45 graded Frobenius algebra, 150, 322
graph defect, 152
alternating character, 160
Gromov-Witten invariants, 232
arm colength, 4
Hecke correspondence, 45
bi-degree, 46
Heisenberg monomial classes, 48
Borel-Moore homology, 157
Heisenberg operators, 47
boson-fermion correspondence, 236
Hilbert ring, 147
boundary of Hilbert schemes, 15
Hilbert scheme of α-points, 301
boundary operator, 61
Hilbert scheme of Λ-points, 301
Calogero-Sutherland operator, 155 Hilbert scheme of n points, 10
Cartan matrix, 44 Hilbert scheme, punctual, 12
Chern character operator, 73 Hilbert-Chow morphism, 12
Chevalley generators, 45 Hilbert/Gromov-Witten correspondence,
class algebra, 160 237, 276
Cohomological Crepant Resolution Hodge bundle, 261
Conjecture, 322 hook length, 4
content, 4 Hurwitz cover, 247
convolution product, 160 Hurwitz number, 248
Crepant Resolution Conjecture, 307
incidence Hilbert scheme, 17
curvi-linear ideals, 12
induction map, 204
degeneracy locus, 254, 257 integral basis, 175
degree defect, 152 integral class, 175
degree of a permutation, 152 integral operator, 175
degree shift number, 204 Jack symmetric function, 7
Divisor Axiom, 233 Jucys-Murphy elements, 207
dominance ordering, 3
leg colength, 4
elementary symmetric function, 5 leg length, 4
equivariant cohomology, 220 Lie superalgebra bracket, 46
equivariant quantum corrected boundary local Gromov-Witten theory, 276
operator, 273
Euler class, 151 monomial symmetric function, 6
Euler-Poincaré characteristic, 151 multi-point trace function, 240
evaluation map, 232 multiple q-zeta value, 99
excess dimension, 233 multiple zeta value, 99

335
336 INDEX

Mumford principle, 141

nef, 16
nef and big, 16
normally ordered product, 61, 93

operator product expansion, 96

orbifold, 203
orbifold cohomology group, 204
orbifold cohomology ring, 204

partition, 3
partition, d-dimensional, 3
partition, generalized, 4
Poincaré polynomial, 7
power symmetric function, 6

quantum corrected boundary operator, 296

quantum corrected cohomology ring, 308
quantum corrected product, 308
quantum differential equation, 278
quantum-mechanical Calogero-Sutherland
operator, 278
quiver, 43
quiver variety, 44

Riemann zeta function, 99

ring of symmetric functions, 5

S-property, 171
Schur function, 6
stable map, 232
standard decomposition, 256
symmetric product, 7

tautological bundle, 15
transfer properties, 77
translation operator, 130

universal linear combination, 76, 162

vacuum expectation, 236

Virasoro operator, 61
virtual fundamental class, 232

weight, 100

Young diagram, 4
Selected Published Titles in This Series
228 Zhenbo Qin, Hilbert Schemes of Points and Infinite Dimensional Lie Algebras, 2018
227 Roberto Frigerio, Bounded Cohomology of Discrete Groups, 2017
226 Marcelo Aguiar and Swapneel Mahajan, Topics in Hyperplane Arrangements, 2017
225 Mario Bonk and Daniel Meyer, Expanding Thurston Maps, 2017
224 Ruy Exel, Partial Dynamical Systems, Fell Bundles and Applications, 2017
223 Guillaume Aubrun and Stanislaw J. Szarek, Alice and Bob Meet Banach, 2017
222 Alexandru Buium, Foundations of Arithmetic Differential Geometry, 2017
221 Dennis Gaitsgory and Nick Rozenblyum, A Study in Derived Algebraic Geometry,
2017
220 A. Shen, V. A. Uspensky, and N. Vereshchagin, Kolmogorov Complexity and
Algorithmic Randomness, 2017
219 Richard Evan Schwartz, The Projective Heat Map, 2017
218 Tushar Das, David Simmons, and Mariusz Urbański, Geometry and Dynamics in
Gromov Hyperbolic Metric Spaces, 2017
217 Benoit Fresse, Homotopy of Operads and Grothendieck–Teichmüller Groups, 2017
216 Frederick W. Gehring, Gaven J. Martin, and Bruce P. Palka, An Introduction to
the Theory of Higher-Dimensional Quasiconformal Mappings, 2017
215 Robert Bieri and Ralph Strebel, On Groups of PL-homeomorphisms of the Real Line,
2016
214 Jared Speck, Shock Formation in Small-Data Solutions to 3D Quasilinear Wave
Equations, 2016
213 Harold G. Diamond and Wen-Bin Zhang (Cheung Man Ping), Beurling
Generalized Numbers, 2016
212 Pandelis Dodos and Vassilis Kanellopoulos, Ramsey Theory for Product Spaces,
2016
211 Charlotte Hardouin, Jacques Sauloy, and Michael F. Singer, Galois Theories of
Linear Difference Equations: An Introduction, 2016
210 Jason P. Bell, Dragos Ghioca, and Thomas J. Tucker, The Dynamical
Mordell–Lang Conjecture, 2016
209 Steve Y. Oudot, Persistence Theory: From Quiver Representations to Data Analysis,
2015
208 Peter S. Ozsváth, András I. Stipsicz, and Zoltán Szabó, Grid Homology for Knots
and Links, 2015
207 Vladimir I. Bogachev, Nicolai V. Krylov, Michael Röckner, and Stanislav V.
Shaposhnikov, Fokker–Planck–Kolmogorov Equations, 2015
206 Bennett Chow, Sun-Chin Chu, David Glickenstein, Christine Guenther, James
Isenberg, Tom Ivey, Dan Knopf, Peng Lu, Feng Luo, and Lei Ni, The Ricci Flow:
Techniques and Applications: Part IV: Long-Time Solutions and Related Topics, 2015
205 Pavel Etingof, Shlomo Gelaki, Dmitri Nikshych, and Victor Ostrik, Tensor
Categories, 2015
204 Victor M. Buchstaber and Taras E. Panov, Toric Topology, 2015
203 Donald Yau and Mark W. Johnson, A Foundation for PROPs, Algebras, and
Modules, 2015
202 Shiri Artstein-Avidan, Apostolos Giannopoulos, and Vitali D. Milman,
Asymptotic Geometric Analysis, Part I, 2015
201 Christopher L. Douglas, John Francis, André G. Henriques, and Michael A.
Hill, Editors, Topological Modular Forms, 2014

For a complete list of titles in this series, visit the

AMS Bookstore at www.ams.org/bookstore/survseries/.
Hilbert schemes, which parametrize subschemes in algebraic varieties, have been
extensively studied in algebraic geometry for the last 50 years. The most interesting
class of Hilbert schemes are schemes X [n] of collections of n points (zero-dimensional
subschemes) in a smooth algebraic surface X . Schemes X [n] turn out to be closely
related to many areas of mathematics, such as algebraic combinatorics, integrable
systems, representation theory, and mathematical physics, among others.
This book surveys recent developments of the theory of Hilbert schemes of points
on complex surfaces and its interplay with infinite dimensional Lie algebras. It starts
with the basics of Hilbert schemes of points and presents in detail an example of
Hilbert schemes of points on the projective plane. Then the author turns to the study
of cohomology of X [n] , including the construction of the action of infinite dimen-
sional Lie algebras on this cohomology, the ring structure of cohomology, equivariant
cohomology of X [n] and the Gromov–Witten correspondence. The last part of the book
presents results about quantum cohomology of X [n] and related questions.
The book is of interest to graduate students and researchers in algebraic geometry,
representation theory, combinatorics, topology, number theory, and theoretical physics.

For additional information

and updates on this book, visit
[[[EQWSVKFSSOTEKIWWYVZ

SURV/228
www.ams.org