The Bethe-Salpeter Equation Formalism:

From Physics to Chemistry

Xavier Blase,∗,† Ivan Duchemin,‡ Denis Jacquemin, ¶ and Pierre-François Loos∗,§

†Université Grenoble Alpes, CNRS, Institut NEEL, F-38042 Grenoble, France

‡Université Grenoble Alpes, CEA, IRIG-MEM-L Sim, 38054 Grenoble, France
arXiv:2006.09440v2 [physics.chem-ph] 8 Aug 2020

¶Université de Nantes, CNRS, CEISAM UMR 6230, F-44000 Nantes, France

§Laboratoire de Chimie et Physique Quantiques, Université de Toulouse, CNRS, UPS, France

E-mail: [email protected]; [email protected]

Abstract
The Bethe-Salpeter equation (BSE) formalism is steadily asserting itself as a new efficient and accurate
tool in the ensemble of computational methods available to chemists in order to predict optical excitations in
molecular systems. In particular, the combination of the so-called GW approximation, giving access to reliable
ionization energies and electron affinities, and the BSE formalism, able to model UV/Vis spectra, has shown to
provide accurate singlet excitation energies with a typical error of 0.1–0.3 eV. With a similar computational cost
as time-dependent density-functional theory (TD-DFT), BSE is able to provide an accuracy on par with the most
accurate global and range-separated hybrid functionals without the unsettling choice of the exchange-correlation
functional, resolving further known issues (e.g., charge-transfer excitations). In this Perspective article, we
provide a historical overview of BSE, with a particular focus on its condensed-matter roots. We also propose a
critical review of its strengths and weaknesses in different chemical situations.

Graphical TOC Entry

Kohn-Sham DFT
 
∇2
− + vext + V Hxc φKp S = εKp S φKp S
2

Fundamental gap
GW approximation Ionization potentials (Inverse)
Electron affinities
photoemission
εGW
p = εpK S + φpK S ΣGW (εGW
p ) − V xc φpK S spectroscopy
Excitonic effect
Bethe-Salpeter equation Optical excitations Optical
spectroscopy
    
R C Xm Xm
= Ωm
−C ∗ −R∗ Ym Ym

Introduction. In its press release announcing the attri- are “much too complicated to be soluble”, urging scien-
bution of the 2013 Nobel prize in Chemistry to Karplus, tists to develop “approximate practical methods”. This is
Levitt, and Warshel, the Royal Swedish Academy of Sci- where the electronic structure community stands, attempting
ences concluded by stating “Today the computer is just as to develop robust approximations to study with increasing
important a tool for chemists as the test tube. Simulations accuracy the properties of ever more complex systems. The
are so realistic that they predict the outcome of traditional study of optical excitations (also known as neutral excitations
experiments.” 1 Martin Karplus’ Nobel lecture moderated in condensed-matter systems), from molecules to extended
this statement, introducing his presentation by a 1929 quote solids, has witnessed the development of a large number of
from Dirac emphasizing that laws of quantum mechanics such approximate methods with numerous applications to

1
a large variety of fields, from the prediction of the colour affinities) of the system
of precious metals for jewellery, 2 to the understanding,
e.g., of the basic principles behind organic photovoltaics,
Õ fs (x) fs∗ (x 0)
G(x, x 0; ω) = , (3)
photocatalysis and DNA damage under irradiation. 3–5 The s
ω − εs + iη × sgn(εs − µ)
present Perspective aims at describing the current status and
upcoming challenges for the Bethe-Salpeter equation (BSE) where µ is the chemical potential, η is a positive infinites-
formalism 6,7 that, while sharing many features with time- imal, εs = EsN +1 − E0N for εs > µ, and εs = E0N − EsN −1
dependent density-functional theory (TD-DFT), 8 including for εs < µ. Here, EsN is the total energy of the sth excited
computational scaling with system size, relies on a very dif- state of the N-electron system. The fs ’s are the so-called
ferent formalism, with specific difficulties but also potential Lehmann amplitudes that reduce to one-body orbitals in the
solutions to known TD-DFT issues. 9 case of single-determinant many-body wave functions (see
below). Unlike KS eigenvalues, the poles of the Green’s
function {εs } are proper addition/removal energies of the
N-electron system, leading to well-defined ionization poten-
Theory. The BSE formalism 6,7,10–13 belongs to the fam-
tials and electronic affinities. In contrast to standard ∆SCF
ily of Green’s function many-body perturbation theories
techniques, the knowledge of G provides the full ionization
(MBPT) 14–16 together with, for example, the algebraic-
spectrum, as measured by direct and inverse photoemission
diagrammatic construction (ADC) techniques 17 or the polar-
spectroscopy, not only that associated with frontier orbitals.
ization propagator approaches (like SOPPA 18 ) in quantum
Using the equation-of-motion formalism for the cre-
chemistry. While the one-body density stands as the basic
ation/destruction operators, it can be shown formally that
variable in density-functional theory (DFT), 19,20 the pillar
G verifies
of Green’s function MBPT is the (time-ordered) one-body
Green’s function ∂
  ∫
− h(r 1 ) G(1, 2) − d3 Σ(1, 3)G(3, 2) = δ(1, 2), (4)
∂t1
G(xt, x 0t 0) = −i hN |T ψ̂(xt)ψ̂ † (x 0t 0) |Ni , (1)
 

where we introduce the composite index, e.g., 1 ≡ (x 1 t1 ).

where |Ni is the N-electron ground-state wave function. The Here, δ is Dirac’s delta function, h is the one-body Hartree
operators ψ̂(xt) and ψ̂ † (x 0t 0) remove and add an electron (re- Hamiltonian and Σ is the so-called exchange-correlation (xc)
spectively) in space-spin-time positions (xt) and (x 0t 0), while self-energy operator. Using the spectral representation of G
T is the time-ordering operator. For t > t 0, G provides the [see Eq. (3)], dropping spin variables for simplicity, one gets
amplitude of probability of finding, on top of the ground-state the familiar eigenvalue equation, i.e.,
Fermi sea (i.e., higher in energy than the highest-occupied
energy level, also known as Fermi level), an electron in (xt)
∫
h(r) fs (r) + d r 0 Σ(r, r 0; εs ) fs (r 0) = εs fs (r), (5)
that was previously introduced in (x 0t 0), while for t < t 0 the
propagation of an electron hole (often simply called a hole)
is monitored. which formally resembles the KS equation 20 with the differ-
This definition indicates that the one-body Green’s func- ence that the self-energy Σ is non-local, energy-dependent
tion is well suited to obtain “charged excitations", more com- and non-hermitian. The knowledge of Σ allows to access the
monly labeled as electronic energy levels, as obtained, e.g., true addition/removal energies, namely the entire spectrum
in a direct or inverse photo-emission experiments where an of occupied and virtual electronic energy levels, at the cost
electron is ejected or added to the N-electron system. In par- of solving a generalized one-body eigenvalue equation.
ticular, and as opposed to Kohn-Sham (KS) DFT, the Green’s
function formalism offers a more rigorous and systematically
improvable path for the obtention of the ionization potential The GW self-energy. While the equations reported
I N = E0N −1 − E0N , the electronic affinity A N = E0N − E0N +1 , above are formally exact, it remains to provide an expres-
and the experimental (photoemission) fundamental gap sion for the xc self-energy operator Σ. This is where Green’s
function practical theories differ. Developed by Lars Hedin in
Egfund = I N − A N (2)
1965 with application to the interacting homogeneous elec-
tron gas, 14 the GW approximation 15,21 follows the path of
of the N-electron system, where E0N corresponds to its
linear response by considering the variation of G with re-
ground-state energy. Since these energy levels are key input
spect to an external perturbation (see Fig. 1). The resulting
quantities for the subsequent BSE calculation, we start by
equation, when compared with the equation for the time-
discussing these in some details.
evolution of G [see Eq. (4)], leads to a formal expression for
the self-energy
∫
Charged excitations. A central property of the one- Σ(1, 2) = i d34 G(1, 4)W(3, 1+ )Γ(42, 3), (6)
body Green’s function is that its frequency-dependent (i.e.,
dynamical) spectral representation has poles at the charged
where W is the dynamically-screened Coulomb potential and
excitation energies (i.e., the ionization potentials and electron
Γ is the so-called “vertex" function. The notation 1+ means

2
that the time t1 is taken at t1+ = t1 + 0+ for sake of causal- In practice, the input G and χ0 required to initially build
ity, where 0+ is a positive infinitesimal. The neglect of the ΣGW are chosen as the “best” Green’s function and suscepti-
vertex, i.e., Γ(42, 3) = δ(23)δ(24), leads to the so-called GW bility that can be easily computed, namely the KS or Hartree-
approximation of the self-energy Fock (HF) ones where the {ε p, fp } of Eq. (3) are taken to
be KS (or HF) eigenstates. Taking then (ΣGW − V xc ) as a
ΣGW (1, 2) = i G(1, 2)W(2, 1+ ), (7) correction to the KS xc potential V xc , a first-order correction
to the input KS energies {ε KSp } is obtained by solving the
that can be regarded as the lowest-order perturbation in terms so-called quasiparticle equation
of the screened Coulomb potential W with
∫ ω = ε KS KS GW
p + φp Σ (ω) − V xc φKS
p . (10)
W(1, 2) = v(1, 2) + d34 v(1, 3) χ0 (3, 4)W(4, 2), (8a)
As a non-linear equation, the self-consistent quasiparticle
χ0 (1, 2) = −iG(1, 2+ )G(2, 1+ ), (8b) equation (10) has various solutions associated with different
spectral weights. The existence of a well defined quasipar-
where χ0 is the independent electron susceptibility and v the ticle energy requires a solution with a large spectral weight,
bare Coulomb potential. Equation (8a) can be recast as i.e., close to unity, a condition not always fulfilled for states
∫ far away from the fundamental gap. 22
W(1, 2) = v(1, 2) + d34 v(1, 3) χ(3, 4)v(4, 2), (9) Such an approach, where input KS energies are corrected
to yield better electronic energy levels, is labeled as the
where χ is the interacting susceptibility. In this latter expres- single-shot, or perturbative, G0W0 technique. This simple
sion, (v χv) represents the field created in (2) by the charge scheme was used in the early GW studies of extended semi-
rearrangement of the N-electron system generated by a (unit) conductors and insulators, 23–26 and surfaces, 27–29 allowing
charge added in (1). As such, this term contains the effect to dramatically reduce the errors associated with KS eigenval-
of dielectric screening (or polarization in a quantum chemist ues in conjunction with common local or gradient-corrected
language). As in a standard ∆SCF calculation, the GW for- approximations to the xc potential. In particular, the well-
malism contains the response of the N-electron system to an known “band gap" problem, 30,31 namely the underestimation
electron added (removed) to any virtual (occupied) molecular of the occupied to unoccupied bands energy gap at the local-
orbital, but without the restriction that only frontier orbitals density approximation (LDA) KS level, was dramatically re-
can be tackled. This explains that the GW one-electron ener- duced, bringing the agreement with experiment to within a
gies are proper addition/removal energies. few tenths of an eV with a computational cost scaling quarti-
cally with the system size (see below). A compilation of data
BSE KS-DFT for G0W0 applied to extended inorganic semiconductors can
be found in Ref. 32.
W (ω) & εGW

Although G0W0 provides accurate results (at least for

εKS

weakly/moderately correlated systems), it is strongly starting-

point dependent due to its perturbative nature. For exam-
G = G0 + G0 ΣG
Σ G ple, the quasiparticle energies, and in particular the HOMO-
LUMO gap, depends on the input KS eigenvalues. Tuning
Γ=
Γ

the starting point functional or applying a self-consistent GW

iGW

1+
1)

scheme are two different approaches commonly employed to

(Γ =

δ Σ GG
δG
Σ=

tackle this problem. We will comment further on this par-

G

ticular point below when addressing the quality of the BSE

−iG

W Γ
W optical excitations.
P=

=
v+
vP GG
Γ Another important feature compared to other perturbative
W −i techniques, the GW formalism can tackle finite and peri-
=
P
P odic systems, and does not present any divergence in the
limit of zero gap (metallic) systems. 33 However, remaining
Figure 1: Hedin’s pentagon connects the Green’s function G, its a low-order perturbative approach starting with a single-
non-interacting analog G0 , the irreducible vertex function Γ, the irre- determinant mean-field solution, it is not intended to explore
ducible polarizability P, the dynamically-screened Coulomb poten- strongly correlated systems. 34
tial W , and the self-energy Σ through a set of five integro-differential
equations known as Hedin’s equations. 14 The path made of black
arrows shows the GW process which bypasses the computation of
Γ (gray arrows). As input, one must provide KS (or HF) orbitals Neutral excitations. Like TD-DFT, BSE deals with the
and their corresponding energies. Depending on the level of self- calculations of optical (or neutral) excitations, as measured
consistency in the GW calculation, only the orbital energies or both by optical (e.g., absorption) spectroscopy, However, while
the orbitals and their energies are corrected. As output, GW pro-
TD-DFT starts with the variation of the charge density ρ(1)
vides corrected quantities, i.e., quasiparticle energies and W , which
with respect to an external local perturbation U(1), the BSE
can then be used to compute the BSE optical excitations of the sys-
tem of interest.
formalism considers a generalized four-point susceptibility,
or two-particle correlation function, that monitors the varia-

3
 opt
Figure 2: Definition of the optical gap Eg and fundamental gap Egfund . E B is the electron-hole or excitonic binding energy, while I N and A N
are the ionization potential and the electron affinity of the N -electron system. EgKS and EgGW are the KS and GW HOMO-LUMO gaps. See
main text for the definition of the other quantities

tion of the one-body Green’s function G(1, 10) with respect where it is customary to neglect the derivative (∂W/∂G) that
to a non-local external perturbation U(2, 20): 7 introduces again higher orders in W. 35–37 At that stage, the
BSE kernel is fully dynamical, i.e., it explicitly depends on the
DFT ∂ ρ(1) ∂G(1, 10)
BSE frequency ω. Taking the static limit, i.e., W(ω = 0), for the
χ(1, 2) = → L(1, 2; 10, 20) = .
∂U(2) ∂U(20, 2) screened Coulomb potential, that replaces the static DFT xc
(11) kernel, and expressing Eq. (13) in the standard product space
The formal relation χ(1, 2) = −iL(1, 2; 1+, 2+ ) with ρ(1) = {φi (r)φ a (r 0)} [where (i, j) are occupied spatial orbitals and
−iG(1, 1+ ) offers a direct bridge between the TD-DFT and (a, b) are unoccupied spatial orbitals], leads to an eigenvalue
BSE worlds. The equation of motion for G [see Eq. (4)] can problem similar to the so-called Casida equations in TD-
be reformulated in the form of a Dyson equation DFT: 38    m  m
R C X X
G = G0 + G0 (v H + U + Σ)G, (12) = Ω m , (18)
−C ∗ −R∗ Y m Ym
that relates the full (interacting) Green’s function, G, to its with electron-hole (eh) eigenstates written as
non-interacting version, G0 , where v H and U are the Hartree Õ
ψmeh
(r e, r h ) = m
φi (r h )φ a (r e ) + Yia
m
φi (r e )φ a (r h ) ,

and external potentials, respectively. The derivative with Xia
respect to U of this Dyson equation yields the self-consistent ia
Bethe-Salpeter equation (19)
where m indexes the electronic excitations. The {φi/a } are
typically the input (KS) eigenstates used to build the GW self-
L(1, 2; 10, 20) = L0 (1, 2; 10, 20)+
∫ energy. They are here taken to be real in the case of finite-size
d3456 L0 (1, 4; 10, 3)ΞBSE (3, 5; 4, 6)L(6, 2; 5, 20), (13) systems. In the case of a closed-shell singlet ground state,
the resonant and coupling parts of the BSE Hamiltonian read
where L0 (1, 2; 10, 20) = G(1, 20)G(2, 10) is the non-interacting  
4-point susceptibility and Rai,b j = εaGW − εiGW δi j δab + κ(ia| jb) − Wi j,ab, (20)
Cai,b j = κ(ia|bj) − Wib,a j , (21)
∂Σ(3, 4)
i ΞBSE (3, 5; 4, 6) = v(3, 6)δ(34)δ(56) + i (14)
∂G(6, 5) with κ = 2 or 0 if one targets singlet or triplet excited states
(respectively), and
is the so-called BSE kernel. This equation can be compared
to its TD-DFT analog ∬
∫ Wi j,ab = d r d r 0 φi (r)φ j (r)W(r, r 0; ω = 0)φ a (r 0)φb (r 0),
χ(1, 2) = χ0 (1, 2) + d34 χ0 (1, 3)ΞDFT (3, 4) χ(4, 2), (15) (22)
where we notice that the two occupied (virtual) eigenstates
where are taken at the same position of space, in contrast with the
∂V xc (3) (ia| jb) bare Coulomb term defined as
ΞDFT (3, 4) = v(3, 4) + (16)
∂ ρ(4) ∬
is the TD-DFT kernel. Plugging now the GW self-energy (ia| jb) = d r d r 0 φi (r)φ a (r)v(r − r 0)φ j (r 0)φb (r 0). (23)
[see Eq. (7)], in a scheme that we label BSE@GW, leads to
an approximate version of the BSE kernel Neglecting the coupling term C between the resonant term
R and anti-resonant term −R∗ in Eq. (18), leads to the well-
i ΞBSE (3, 5; 4, 6) known Tamm-Dancoff approximation (TDA).
= v(3, 6)δ(34)δ(56) − W(3+, 4)δ(36)δ(45), (17) As compared to TD-DFT: i) the GW quasiparticle energies
GW
{εi/a } replace the KS eigenvalues, and ii) the non-local

4
screened Coulomb matrix elements replaces the DFT xc where EB accounts for the excitonic effect, that is, the stabi-
kernel. We emphasize that these equations can be solved at lization induced by the attraction of the excited electron and
exactly the same cost as the standard TD-DFT equations once its hole left behind (see Fig. 2).
the quasiparticle energies and screened Coulomb potential W Such a residual gap problem can be significantly im-
are inherited from preceding GW calculations. This defines proved by adopting xc functionals with a tuned amount
the standard (static) BSE@GW scheme that we discuss in of exact exchange 59,60 that yield a much improved KS
this Perspective, highlighting its pros and cons. gap as a starting point for the GW correction. 56,61–63 Al-
ternatively, self-consistent approaches such as eigenvalue
self-consistent (evGW) 24 or quasiparticle self-consistent
(qsGW) 64 schemes, where corrected eigenvalues, and possi-
Historical overview. Originally developed in the frame-
bly orbitals, are reinjected in the construction of G and W,
work of nuclear physics, 6 the BSE formalism has emerged
have been shown to lead to a significant improvement of the
in condensed-matter physics around the 1960’s at the tight-
quasiparticle energies in the case of molecular systems, with
binding level with the study of the optical properties of simple
the advantage of significantly removing the dependence on
semiconductors. 37,39,40 Three decades later, the first ab initio
the starting point functional. 62,65–69 As a result, BSE singlet
implementations, starting with small clusters, 11,41 extended
excitation energies starting from such improved quasiparticle
semiconductors, and wide-gap insulators 12,42,43 paved the
energies were found to be in much better agreement with ref-
way to the popularization in the solid-state physics commu-
erence calculations. For sake of illustration, an average error
nity of the BSE formalism.
of 0.2 eV was found for the well-known Thiel set 57 gathering
Following pioneering applications to periodic polymers
ca. 200 representative singlet excitations from a large variety
and molecules, 44–47 BSE gained much momentum in quan-
of representative molecules. 50,51,55,56 This is equivalent to the
tum chemistry 48 with, in particular, several benchmarks 49–56
best TD-DFT results obtained by scanning a large variety of
on large molecular sets performed with the very same pa-
hybrid functionals with various amounts of exact exchange.
rameters (geometries, basis sets, etc) than the available
higher-level reference calculations. 57 Such comparisons were
grounded in the development of codes replacing the plane- Charge-transfer excited states. A very remarkable
wave paradigm of solid-state physics by Gaussian basis sets, success of the BSE formalism lies in the description of
together with adequate auxiliary bases when resolution-of- charge-transfer (CT) excitations, a notoriously difficult prob-
the-identity (RI) techniques 58 were used. lem for TD-DFT adopting standard (semi-)local function-
An important conclusion drawn from these calculations als. 70 Similar difficulties emerge in solid-state physics for
was that the quality of the BSE excitation energies is strongly semiconductors where extended Wannier excitons, charac-
correlated to the deviation of the preceding GW HOMO- terized by weakly overlapping electrons and holes (Fig. 3),
LUMO gap cause a dramatic deficit of spectral weight at low energy. 71
EgGW = εLUMO
GW
− εHOMO
GW
, (24) These difficulties can be ascribed to the lack of long-range
electron-hole interaction with local xc functionals. It can be
with the experimental (photoemission) fundamental gap de-
cured through an exact exchange contribution, a solution that
fined in Eq. (2).
explains the success of (optimally-tuned) range-separated
Standard G0W0 calculations starting with KS eigenstates
hybrids for the description of CT excitations. 59,60 The anal-
generated with (semi)local functionals yield much larger
ysis of the screened Coulomb potential matrix elements in
HOMO-LUMO gaps than the input KS gap
the BSE kernel [see Eq. (17)] reveals that such long-range
EgKS = εLUMO
KS KS
− εHOMO, (25) (non-local) electron-hole interactions are properly described,
including in environments (solvents, molecular solid, etc.)
but still too small as compared to the experimental value, i.e., where the screening reduces the long-range electron-hole
interactions. The success of the BSE formalism to treat CT
EgKS  EgG0W0 < Egfund . (26) excitations has been demonstrated in several studies, 66,72–79
opening the way to the modeling of key applications such as
Such a residual discrepancy has been attributed by several doping, 80 photovoltaics or photocatalysis in organic systems.
authors to “overscreening", namely the effect associated with
building the susceptibility χ based on a grossly underesti-
mated (KS) band gap. This leads to a spurious enhancement
of the screening or polarization and, consequently, to an un- Combining BSE with PCM and QM/MM models.
derestimated G0W0 gap as compared to the (exact) fundamen- The ability to account for the effect on the excitation energies
tal gap. More prosaically, the G0W0 approach is constructed of an electrostatic and dielectric environment (an electrode,
as a first-order perturbation theory, so by correcting a very a solvent, a molecular interface. . . ) is an important step to-
“bad" zeroth-order KS system one cannot expect to obtain wards the description of realistic systems. Pioneering BSE
an accurate corrected gap. Such an underestimation of the studies demonstrated, for example, the large renormalization
fundamental gap leads to a similar underestimation of the of charged and neutral excitations in molecular systems and
opt
optical gap Eg , i.e., the lowest optical excitation energy: nanotubes close to a metallic electrode or in bundles. 74,81,82
Recent attempts to merge the GW and BSE formalisms with
opt
Eg = E1N − E0N = Egfund + EB, (27) model polarizable environments at the PCM or QM/MM lev-

5
 Mo'(Wannier.
(a) ronment, is that the renormalization of the electron-electron
and electron-hole interactions by the reaction field captures
+
both linear-response+"and state-specific contributions 90 to the
W = V/e solvatochromic shift of the optical lines, allowing to treat on
W=v/ε
the same footing local (Frenkel) and CT excitations. This is
µe4 an important advantage as compared to, e.g., TD-DFT where
EB =
2~2 ✏2 - linear-response and state-specific
EB =
µe4 effects have to be explored
with different formalisms. 2~2 ✏2 #"
To date, environmental effects on fast electronic excita-
tions are only included by considering the low-frequency
optical response of the polarizable medium (e.g., consider-
ing the ∞ ' 1.78 macroscopic dielectric constant of water
(b) Frenkel in the optical range), neglecting the frequency dependence
Frenkel.
+ of the dielectric constant in the
+" optical range. Generaliza-
tion to fully frequency-dependent polarizable properties of
- the environment would allow to#" explore systems where the
relative dynamics of the solute and the solvent are not de-
coupled, i.e., situations where
CT. neither the adiabatic limit nor
+ +" are
the anti-adiabatic limits expected to be valid (for a recent
CT discussion, see Ref. 91).
We now leave the description of successes to discuss
difficulties and future directions of developments and im-
- provements. #"

Figure 3: Symbolic representation of (a) extended Wannier exciton

with large electron-hole average distance, and (b) Frenkel (local) and
The computational challenge. As emphasized above,
charge-transfer (CT) excitations at a donor-acceptor interface. Wan- the BSE eigenvalue equation in the single-excitation space
nier and CT excitations require long-range electron-hole interaction [see Eq. (18)] is formally equivalent to that of TD-DFT or TD-
accounting for the host dielectric constant. In the case of Wannier HF. 92 Searching iteratively for the lowest eigenstates exhibits
the same O(Norb 4 matrix-vector multiplication computational
excitons, the binding energy E B can be well approximated by the )
standard hydrogenoid model where µ is the effective mass and  is cost within BSE and TD-DFT. Concerning the construction
the dielectric constant. of the BSE Hamiltonian, it is no more expensive than build-
ing its TD-DFT analogue with hybrid functionals, reducing
again to O(Norb4 operations with standard RI techniques. Ex-
)
els 83–89 paved the way not only to interesting applications plicit calculation of the full BSE Hamiltonian in transition
but also to a better understanding of the merits of these ap- space can be further avoided using density matrix perturba-
proaches relying on the use of the screened Coulomb potential tion theory, 72,93 not reducing though the O(Norb 4
) scaling, but
designed to capture polarization effects at all spatial ranges. sacrificing further the knowledge of the eigenvectors. Ex-
As a matter of fact, dressing the bare Coulomb potential with ploiting further the locality of the atomic orbital basis, the
the reaction field matrix [v(r, r 0) −→ v(r, r 0)+v reac (r, r 0; ω)] BSE absorption spectrum can be obtained with O(Norb 3
) op-
in the relation between the screened Coulomb potential W 94
erations using such iterative techniques. With the same
and the independent-electron susceptibility [see Eq. (8a)] al- restriction on the eigenvectors, a time-propagation approach,
lows to perform GW and BSE calculations in a polarizable similar to that implemented for TD-DFT, 95 combined with
environment at the same computational cost as the corre- stochastic techniques to reduce the cost of building the BSE
sponding gas-phase calculation. The reaction field operator Hamiltonian matrix elements, allows quadratic scaling with
v reac (r, r 0; ω) describes the potential generated in r 0 by the systems size. 96
charge rearrangements in the polarizable environment in- In practice, the main bottleneck for standard BSE calcula-
duced by a source charge located in r, where r and r 0 lie in tions as compared to TD-DFT resides in the preceding GW
the quantum mechanical subsystem of interest. The reaction calculation that scales as O Norb 4


with system size using
field is dynamical since the dielectric properties of the envi- plane-wave basis sets or RI techniques, but with a rather
ronment, such as the macroscopic dielectric constant  M (ω), large prefactor. The field of low-scaling GW calculations is
are in principle frequency dependent. Once the reaction however witnessing significant advances. While the sparsity
field matrix is known, with typically O(Norb NMM 2
) operations of, for example, the overlap matrix in the atomic orbital
(where Norb is the number of orbitals and NMM the number of basis allows to reduce the scaling in the large size limit, 97,98
polarizable atoms in the environment), the full spectrum of efficient real-space grids and time techniques are bloom-
GW quasiparticle energies and BSE neutral excitations can ing, 99,100 borrowing in particular the well-known Laplace
be renormalized by the effect of the environment. transform approach used in quantum chemistry. 101 Together
A remarkable property 87 of the scheme described above, with a stochastic sampling of virtual states, this family of
which combines the BSE formalism with a polarizable envi-

6
techniques allow to set up linear scaling GW calculations. 102 evaluated within the ACFDT framework, 117 was mostly dis-
The separability of occupied and virtual states summations carded from the set of tested techniques due to instabilities
lying at the heart of these approaches are now spreading (negative frequency modes in the BSE polarization propa-
fast in quantum chemistry within the interpolative separa- gator) and replaced by an approximate (RPAsX) approach
ble density fitting (ISDF) approach applied for calculating where the screened-Coulomb potential matrix elements was
with cubic scaling the susceptibility needed in random-phase removed from the resonant electron-hole contribution. 113,121
approximation (RPA) and GW calculations. 103–105 These on- Moreover, it was also observed in Ref. 115 that, in some
going developments pave the way to applying the GW@BSE cases, unphysical irregularities on the ground-state PES ap-
formalism to systems containing several hundred atoms on pear due to the appearance of discontinuities as a function of
standard laboratory clusters. the bond length for some of the GW quasiparticle energies.
Such an unphysical behavior stems from defining the quasi-
particle energy as the solution of the quasiparticle equation
with the largest spectral weight in cases where several so-
The triplet instability challenge. The analysis of the
lutions can be found [see Eq. (10)]. We refer the interested
singlet-triplet splitting is central to numerous applications
reader to Refs. 22,122–125 for detailed discussions.
such as singlet fission or thermally activated delayed fluores-
cence (TADF). From a more theoretical point of view, triplet
instabilities that often plague the applicability of TD-DFT
are intimately linked to the stability analysis of restricted The challenge of analytical nuclear gradients. The
closed-shell solutions at the HF 106 and KS 107 levels. While features of ground- and excited-state potential energy sur-
TD-DFT with range-separated hybrids can benefit from tun- faces (PES) are critical for the faithful description and a
ing the range-separation parameter(s) as a mean to act on deeper understanding of photochemical and photophysical
the triplet instability, 108 BSE calculations do not offer this processes. 126 For example, chemoluminescence and fluo-
pragmatic way-out since the screened Coulomb potential that rescence are associated with geometric relaxation of ex-
builds the kernel does not offer any parameter to tune. cited states, and structural changes upon electronic excita-
Benchmark calculations 109,110 clearly concluded that tion. 127 Reliable predictions of these mechanisms, which
triplets are notably too low in energy within BSE and have attracted much experimental and theoretical interest
that the use of the TDA was able to partly reduce this error. lately, require exploring the ground- and excited-state PES.
However, as it stands, the BSE accuracy for triplets remains From a theoretical point of view, the accurate prediction of
rather unsatisfactory for reliable applications. An alternative excited electronic states remains a challenge, 128 especially
cure was offered by hybridizing TD-DFT and BSE, that for large systems where state-of-the-art computational tech-
is, by adding to the BSE kernel the correlation part of the niques (such as multiconfigurational methods 129 ) cannot be
underlying DFT functional used to build the susceptibility afforded. For the last two decades, TD-DFT has been the
and resulting screened Coulomb potential W. 111 go-to method to compute absorption and emission spectra in
large molecular systems.
In TD-DFT, the PES for the excited states can be easily and
efficiently obtained as a function of the molecular geometry
The challenge of the ground-state energy. In con-
by simply adding the ground-state DFT energy to the excita-
trast to TD-DFT which relies on KS-DFT as its ground-state
tion energy of the selected state. One of the strongest assets of
analog, the ground-state BSE energy is not a well-defined
TD-DFT is the availability of first- and second-order analytic
quantity, and no clear consensus has been found regarding its
nuclear gradients (i.e., the first- and second-order deriva-
formal definition. Consequently, the BSE ground-state for-
tives of the excited-state energy with respect to atomic dis-
malism remains in its infancy with very few available studies
placements), which enables the exploration of excited-state
for atomic and molecular systems. 88,112–115
PES. 130
A promising route, which closely follows RPA-type for-
A significant limitation of the BSE formalism, as compared
malisms, 116 is to calculate the ground-state BSE energy
to TD-DFT, lies in the lack of analytical nuclear gradients
within the adiabatic-connection fluctuation-dissipation the-
for both the ground and excited states, preventing efficient
orem (ACFDT) framework. 117 Thanks to comparisons with
studies of many key excited-state processes. While calcu-
both similar and state-of-art computational approaches, it
lations of the GW quasiparticle energy ionic gradients is
was recently shown that the ACFDT@BSE@GW approach
becoming increasingly popular, 131–135 only one pioneering
yields extremely accurate PES around equilibrium, and can
study of the excited-state BSE gradients has been published
even compete with high-order coupled cluster methods in
so far. 136 In this seminal work devoted to small molecules
terms of absolute ground-state energies and equilibrium
(CO and NH3 ), only the BSE excitation energy gradients
distances. 115 However, their accuracy near the dissociation
were calculated, with the approximation that the gradient of
limit remains an open question. 112,113,118–120 Indeed, in the
the screened Coulomb potential can be neglected, computing
largest available benchmark study 113 encompassing the total
further the KS-LDA forces as its ground-state contribution.
energies of the atoms H–Ne, the atomization energies of
the 26 small molecules forming the HEAT test set, and the
bond lengths and harmonic vibrational frequencies of 3d
transition-metal monoxides, the BSE correlation energy, as

7
Beyond the static approximation. Going beyond the ment of the strengths and weaknesses of the BSE formalism
static approximation is a difficult challenge which has been, of many-body perturbation theory. To do so, we have briefly
nonetheless, embraced by several groups. 7,137–145 As men- reviewed the theoretical aspects behind BSE, and its in-
tioned earlier in this Perspective, most of BSE calcula- timate link with the underlying GW calculation that one
tions are performed within the so-called static approxi- must perform to compute quasiparticle energies and the
mation, which substitutes the dynamically-screened (i.e., dynamically-screened Coulomb potential; two of the key in-
frequency-dependent) Coulomb potential W(ω) by its static put ingredients associated with the BSE formalism. We have
limit W(ω = 0) [see Eq. (22)]. It is important to mention that then provided a succinct historical overview with a particular
diagonalizing the BSE Hamiltonian in the static approxima- focus on its condensed-matter roots, and the lessons that the
tion corresponds to solving a linear eigenvalue problem in the community has learnt from several systematic benchmark
space of single excitations, while it is, in its dynamical form, studies on large molecular systems. Several success stories
a non-linear eigenvalue problem (in the same space) which is are then discussed (charge-transfer excited states and combi-
much harder to solve from a numerical point of view. In com- nation with reaction field methods), before debating some of
plete analogy with the ubiquitous adiabatic approximation in the challenges faced by the BSE formalism (computational
TD-DFT, one key consequence of the static approximation cost, triplet instabilities, ambiguity in the definition of the
is that double (and higher) excitations are completely absent ground-state energy, lack of analytical nuclear gradients, and
from the BSE optical spectrum, which obviously hampers limitations due to the static approximation). We hope that,
the applicability of BSE as double excitations may play, in- by providing a snapshot of the ability of BSE in 2020, the
directly, a key role in photochemistry mechanisms. Higher present Perspective article will motivate a larger commu-
excitations would be explicitly present in the BSE Hamil- nity to participate to the development of this alternative to
tonian by “unfolding” the dynamical BSE kernel, and one TD-DFT which, we believe, may become a very valuable
would recover a linear eigenvalue problem with, nonetheless, computational tool for the physical chemistry community.
a much larger dimension. Corrections to take into account
the dynamical nature of the screening may or may not recover
ACKNOWLEDGMENTS
these multiple excitations. However, dynamical corrections
PFL thanks the European Research Council (ERC) under
permit, in any case, to recover, for transitions with a domi-
the European Union’s Horizon 2020 research and innovation
nant single-excitation character, additional relaxation effects
programme (Grant agreement No. 863481) for financial sup-
coming from higher excitations.
port. Funding from the “Centre National de la Recherche
From a more practical point of view, dynamical effects have
Scientifique” is also acknowledged. This work has also been
been found to affect the positions and widths of core-exciton
supported through the EUR Grant NanoX ANR-17-EURE-
resonances in semiconductors, 36,37 rare gas solids, and tran-
0009 in the framework of the “Programme des Investisse-
sition metals. 146 Thanks to first-order perturbation theory,
ments d’Avenir”. DJ acknowledges the Région des Pays de
Rohlfing and coworkers have developed an efficient way of
la Loire for financial support.
taking into account the dynamical effects via a plasmon-pole
approximation combined with TDA. 137–139,147 With such a
scheme, they have been able to compute the excited states of
biological chromophores, showing that taking into account REFERENCES
the electron-hole dynamical screening is important for an ac-
curate description of the lowest n → π ∗ excitations. 138,139,147 (1) The Royal Swedish Academy of Sciences, The Nobel
Studying PYP, retinal and GFP chromophore models, Ma Prize in Chemistry 2013. Press release, 2013.
et al. found that “the influence of dynamical screening (2) Prandini, G.; Rignanese, G.-M.; Marzari, N. Photore-
on the excitation energies is about 0.1 eV for the lowest alistic Modelling of Metals from First Principles. npj
π → π ∗ transitions, but for the lowest n → π ∗ transitions Comput. Mater. 2019, 129.
the influence is larger, up to 0.25 eV.” 139 Zhang et al. have (3) Kippelen, B.; Brédas, J.-L. Organic photovoltaics. En-
studied the frequency-dependent second-order BSE kernel ergy Environ. Sci. 2009, 2, 251–261.
and they have observed an appreciable improvement over (4) Improta, R.; Santoro, F.; Blancafort, L. Quantum Me-
configuration interaction with singles (CIS), time-dependent chanical Studies on the Photophysics and the Photo-
Hartree-Fock (TDHF), and adiabatic TD-DFT results. 143 chemistry of Nucleic Acids and Nucleobases. Chem.
Rebolini and Toulouse have performed a similar investiga- Rev. 2016, 116, 3540–3593.
tion in a range-separated context, and they have reported a (5) Wu, X.; Choudhuri, I.; Truhlar, D. G. Computational
modest improvement over its static counterpart. 144 In these Studies of Photocatalysis with Metal–Organic Frame-
two latter studies, they also followed a (non-self-consistent) works. Energy Environ. Mat. 2019, 2, 251–263.
perturbative approach within TDA with a renormalization of (6) Salpeter, E. E.; Bethe, H. A. A Relativistic Equation
the first-order perturbative correction. for Bound-State Problems. Phys. Rev. 1951, 84, 1232.
(7) Strinati, G. Application of the Green’s Functions
Method to the Study of the Optical Properties of Semi-
Conclusion. Although far from being exhaustive, we conductors. Riv. Nuovo Cimento 1988, 11, 1–86.
hope that this Perspective provides a concise and fair assess- (8) Runge, E.; Gross, E. K. U. Density-Functional Theory

8
 for Time-Dependent Systems. Phys. Rev. Lett. 1984, Quasiparticle Energies. Phys. Rev. B 1986, 34, 5390–
52, 997–1000. 5413.
(9) Blase, X.; Duchemin, I.; Jacquemin, D. The Bethe– (25) Godby, R. W.; Schlüter, M.; Sham, L. J. Self-Energy
Salpeter Equation in Chemistry: Relations with TD- Operators and Exchange-Correlation Potentials in
DFT, Applications and Challenges. Chem. Soc. Rev. Semiconductors. Phys. Rev. B 1988, 37, 10159–
2018, 47, 1022–1043. 10175.
(10) Albrecht, S.; Reining, L.; Del Sole, R.; Onida, G. Ab (26) von der Linden, W.; Horsch, P. Precise Quasiparticle
Initio Calculation of Excitonic Effects in the Optical Energies and Hartree-Fock Bands of Semiconductors
Spectra of Semiconductors. Phys. Rev. Lett. 1998, 80, and Insulators. Phys. Rev. B 1988, 37, 8351–8362.
4510–4513. (27) Northrup, J. E.; Hybertsen, M. S.; Louie, S. G. Many-
(11) Rohlfing, M.; Louie, S. G. Electron-Hole Excitations body Calculation of the Surface-State Energies for
in Semiconductors and Insulators. Phys. Rev. Lett. Si(111)2×1. Phys. Rev. Lett. 1991, 66, 500–503.
1998, 81, 2312–2315. (28) Blase, X.; Zhu, X.; Louie, S. G. Self-Energy Effects
(12) Benedict, L. X.; Shirley, E. L.; Bohn, R. B. Optical on the Surface-State Energies of H-Si(111)1×1. Phys.
Absorption of Insulators and the Electron-Hole In- Rev. B 1994, 49, 4973–4980.
teraction: An Ab Initio Calculation. Phys. Rev. Lett. (29) Rohlfing, M.; Krüger, P.; Pollmann, J. Efficient
1998, 80, 4514–4517. Scheme for GW Quasiparticle Band-Structure Cal-
(13) van der Horst, J.-W.; Bobbert, P. A.; Michels, M. A. J.; culations with Aapplications to Bulk Si and to the
Brocks, G.; Kelly, P. J. Ab Initio Calculation of the Si(001)-(2×1) Surface. Phys. Rev. B 1995, 52, 1905–
Electronic and Optical Excitations in Polythiophene: 1917.
Effects of Intra- and Interchain Screening. Phys. Rev. (30) Perdew, J. P.; Levy, M. Physical Content of the Exact
Lett. 1999, 83, 4413–4416. Kohn-Sham Orbital Energies: Band Gaps and Deriva-
(14) Hedin, L. New Method for Calculating the One- tive Discontinuities. Phys. Rev. Lett. 1983, 51, 1884–
Particle Green’s Function with Application to the 1887.
Electron-Gas Problem. Phys. Rev. 1965, 139, A796. (31) Sham, L. J.; Schlüter, M. Density-Functional Theory
(15) Onida, G.; Reining, L.; and, A. R. Electronic Excita- of the Energy Gap. Phys. Rev. Lett. 1983, 51, 1888–
tions: Density-Functional Versus Many-Body Green’s 1891.
Function Approaches. Rev. Mod. Phys. 2002, 74, 601– (32) Shishkin, M.; Kresse, G. Self-Consistent GW Calcu-
659. lations for Semiconductors and Insulators. Phys. Rev.
(16) Martin, R.; Reining, L.; Ceperley, D. Interacting Elec- B 2007, 75, 235102.
trons: Theory and Computational Approaches; Cam- (33) Campillo, I.; Pitarke, J. M.; Rubio, A.; Zarate, E.;
bridge University Press, 2016. Echenique, P. M. Inelastic Lifetimes of Hot Electrons
(17) Dreuw, A.; Wormit, M. The Algebraic Diagrammatic in Real Metals. Phys. Rev. Lett. 1999, 83, 2230–2233.
Construction Scheme for the Polarization Propagator (34) Verdozzi, C.; Godby, R. W.; Holloway, S. Evaluation
for the Calculation of Excited States. Wiley Interdiscip. of GW Approximations for the Self-Energy of a Hub-
Rev. Comput. Mol. Sci. 2015, 5, 82–95. bard Cluster. Phys. Rev. Lett. 1995, 74, 2327–2330.
(18) Packer, M. K.; Dalskov, E. K.; Enevoldsen, T.; (35) Hanke, W.; Sham, L. J. Many-Particle Effects in the
Jensen, H. J.; Oddershede, J. A New Implementation Optical Spectrum of a Semiconductor. Phys. Rev. B
of the Second-Order Polarization Propagator Approx- 1980, 21, 4656.
imation (SOPPA): The Excitation Spectra of Benzene (36) Strinati, G. Dynamical Shift and Broadening of Core
and Naphthalene. J. Chem. Phys. 1996, 105, 5886– Excitons in Semiconductors. Phys. Rev. Lett. 1982, 49,
5900. 1519.
(19) Hohenberg, P.; Kohn, W. Inhomogeneous Electron (37) Strinati, G. Effects of Dynamical Screening on Reso-
Gas. Phys. Rev. 1964, 136, B864–B871. nances at Inner-Shell Thresholds in Semiconductors.
(20) Kohn, W.; Sham, L. J. Self-Consistent Equations In- Phys. Rev. B 1984, 29, 5718.
cluding Exchange and Correlation Effects. Phys. Rev. (38) Casida, M. E. In Time-Dependent Density Functional
1965, 140, A1133–A1138. Response Theory for Molecules; Chong, D. P., Ed.;
(21) Golze, D.; Dvorak, M.; Rinke, P. The GW Com- Recent Advances in Density Functional Methods;
pendium: A Practical Guide to Theoretical Photoe- World Scientific, Singapore, 1995; pp 155–192.
mission Spectroscopy. Front. Chem. 2019, 7, 377. (39) Sham, L. J.; Rice, T. M. Many-Particle Derivation of
(22) Veril, M.; Romaniello, P.; Berger, J. A.; Loos, P. F. the Effective-Mass Equation for the Wannier Exciton.
Unphysical Discontinuities in GW Methods. J. Chem. Phys. Rev. 1966, 144, 708–714.
Theory Comput. 2018, 14, 5220. (40) Delerue, C.; Lannoo, M.; Allan, G. Excitonic and
(23) Strinati, G.; Mattausch, H. J.; Hanke, W. Dynamical Quasiparticle Gaps in Si Nanocrystals. Phys. Rev. Lett.
Correlation Effects on the Quasiparticle Bloch States 2000, 84, 2457–2460.
of a Covalent Crystal. Phys. Rev. Lett. 1980, 45, 290– (41) Onida, G.; Reining, L.; Godby, R. W.; Del Sole, R.;
294. Andreoni, W. Ab Initio Calculations of the Quasipar-
(24) Hybertsen, M. S.; Louie, S. G. Electron Correlation ticle and Absorption Spectra of Clusters: The Sodium
in Semiconductors and Insulators: Band Gaps and Tetramer. Phys. Rev. Lett. 1995, 75, 818–821.

9
(42) Albrecht, S.; Onida, G.; Reining, L. Ab initio Cal- (58) Ren, X.; Rinke, P.; Blum, V.; Wieferink, J.;
culation of the Quasiparticle Spectrum and Excitonic Tkatchenko, A.; Sanfilippo, A.; Reuter, K.;
Effects in Li2 O. Phys. Rev. B 1997, 55, 10278–10281. Scheffler, M. Resolution-of-identity Approach to
(43) Rohlfing, M.; Louie, S. G. Excitons and Optical Spec- Hartree–Fock, Hybrid Fensity Functionals, RPA, MP2
trum of the Si(111) − (2 × 1) Surface. Phys. Rev. Lett. and GW with Numeric Atom-Centered Orbital Basis
1999, 83, 856–859. Functions. New J. Phys. 2012, 14, 053020.
(44) Rohlfing, M.; Louie, S. G. Optical Excitations in Con- (59) Stein, T.; Kronik, L.; Baer, R. Reliable Prediction of
jugated Polymers. Phys. Rev. Lett. 1999, 82, 1959– Charge Transfer Excitations in Molecular Complexes
1962. Using Time-Dependent Density Functional Theory. J.
(45) van der Horst, J.-W.; Bobbert, P. A.; Michels, M. A. J.; Am. Chem. Soc. 2009, 131, 2818–2820.
Brocks, G.; Kelly, P. J. Ab Initio Calculation of the (60) Kronik, L.; Stein, T.; Refaely-Abramson, S.; Baer, R.
Electronic and Optical Excitations in Polythiophene: Excitation Gaps of Finite-Sized Systems from Opti-
Effects of Intra- and Interchain Screening. Phys. Rev. mally Tuned Range-Separated Hybrid Functionals. J.
Lett. 1999, 83, 4413–4416. Chem. Theory Comput. 2012, 8, 1515–1531.
(46) Puschnig, P.; Ambrosch-Draxl, C. Suppression of (61) Bruneval, F.; Marques, M. A. L. Benchmarking
Electron-Hole Correlations in 3D Polymer Materials. the Starting Points of the GW Approximation for
Phys. Rev. Lett. 2002, 89, 056405. Molecules. J. Chem. Theory Comput. 2013, 9, 324–
(47) Tiago, M. L.; Northrup, J. E.; Louie, S. G. Ab Initio 329.
Calculation of the Electronic and Optical Properties (62) Rangel, T.; Hamed, S. M.; Bruneval, F.; Neaton, J. B.
of Solid Pentacene. Phys. Rev. B 2003, 67, 115212. Evaluating the GW Approximation with CCSD(T) for
(48) See Table 1 of Ref. 9 for an exhaustive list of applica- Charged Excitations Across the Oligoacenes. J. Chem.
tions to molecular systems. Theory Comput. 2016, 12, 2834–2842.
(49) Boulanger, P.; Jacquemin, D.; Duchemin, I.; Blase, X. (63) Knight, J. W.; Wang, X.; Gallandi, L.; Dolgo-
Fast and Accurate Electronic Excitations in Cyanines unitcheva, O.; Ren, X.; Ortiz, J. V.; Rinke, P.; Körzdör-
with the Many-Body Bethe–Salpeter Approach. J. fer, T.; Marom, N. Accurate Ionization Potentials
Chem. Theory Comput. 2014, 10, 1212–1218. and Electron Affinities of Acceptor Molecules III: A
(50) Jacquemin, D.; Duchemin, I.; Blase, X. Benchmarking Benchmark of GW Methods. J. Chem. Theory Com-
the Bethe–Salpeter Formalism on a Standard Organic put. 2016, 12, 615–626.
Molecular Set. J. Chem. Theory Comput. 2015, 11, (64) van Schilfgaarde, M.; Kotani, T.; Faleev, S. Quasi-
3290–3304. particle Self-Consistent GW Theory. Phys. Rev. Lett.
(51) Bruneval, F.; Hamed, S. M.; Neaton, J. B. A System- 2006, 96, 226402.
atic Benchmark of the ab initio Bethe-Salpeter Equa- (65) Rostgaard, C.; Jacobsen, K. W.; Thygesen, K. S. Fully
tion Approach for Low-Lying Optical Excitations of Self-Consistent GW Calculations for Molecules. Phys.
Small Organic Molecules. J. Chem. Phys. 2015, 142, Rev. B 2010, 81, 085103.
244101. (66) Blase, X.; Attaccalite, C. Charge-Transfer Excitations
(52) Jacquemin, D.; Duchemin, I.; Blase, X. 0–0 Ener- in Molecular Donor-Acceptor Complexes within the
gies Using Hybrid Schemes: Benchmarks of TD-DFT, Many-Body Bethe-Salpeter Approach. Appl. Phys.
CIS(D), ADC(2), CC2, and BSE/GW formalisms for Lett. 2011, 99, 171909.
80 Real-Life Compounds. J. Chem. Theory Comput. (67) Ke, S.-H. All-Electron G W Methods Implemented
2015, 11, 5340–5359. in Molecular Orbital Space: Ionization Energy and
(53) Hirose, D.; Noguchi, Y.; Sugino, O. All- Electron Affinity of Conjugated Molecules. Phys. Rev.
Electron GW+Bethe-Salpeter Calculations on Small B 2011, 84, 205415.
Molecules. Phys. Rev. B 2015, 91, 205111. (68) Kaplan, F.; Harding, M. E.; Seiler, C.; Weigend, F.;
(54) Jacquemin, D.; Duchemin, I.; Blase, X. Is the Bethe– Evers, F.; van Setten, M. J. Quasi-Particle Self-
Salpeter Formalism Accurate for Excitation Energies? Consistent GW for Molecules. J. Chem. Theory Com-
Comparisons with TD-DFT, CASPT2, and EOM- put. 2016, 12, 2528–2541.
CCSD. J. Phys. Chem. Lett. 2017, 8, 1524–1529. (69) Caruso, F.; Dauth, M.; van Setten, M. J.; Rinke, P.
(55) Krause, K.; Klopper, W. Implementation of the Bethe- Benchmark of GW Approaches for the GW100 Test
Salpeter equation in the TURBOMOLE program. J. Set. J. Chem. Theory Comput. 2016, 12, 5076.
Comp. Chem. 2017, 38, 383–388. (70) Dreuw, A.; Head-Gordon, M. Failure of Time-
(56) Gui, X.; Holzer, C.; Klopper, W. Accuracy Assess- Dependent Density Functional Theory for
ment of GW Starting Points for Calculating Molecular Long-Range Charge-Transfer Excited States:
Excitation Energies Using the Bethe–Salpeter Formal- The Zincbacteriochlorin-Bacteriochlorin and
ism. J. Chem. Theory Comput. 2018, 14, 2127–2136. Bacteriochlorophyll-Spheroidene Complexes. J. Am.
(57) Schreiber, M.; Silva-Junior, M. R.; Sauer, S. P. A.; Chem. Soc. 2004, 126, 4007–4016.
Thiel, W. Benchmarks for Electronically Excited (71) Botti, S.; Sottile, F.; Vast, N.; Olevano, V.; Rein-
States: CASPT2, CC2, CCSD and CC3. J. Chem. ing, L.; Weissker, H.-C.; Rubio, A.; Onida, G.;
Phys. 2008, 128, 134110. Del Sole, R.; Godby, R. W. Long-Range Contribu-
tion to the Exchange-Correlation Kernel of Time-

10
 Dependent Density Functional Theory. Phys. Rev. B (86) Varsano, D.; Caprasecca, S.; Coccia, E. Theoretical
2004, 69, 155112. Description of Protein Field Effects on Electronic Ex-
(72) Rocca, D.; Lu, D.; Galli, G. Ab Initio Calculations of citations of Biological Chromophores. J. Phys.: Cond.
Optical Absorption Spectra: Solution of the Bethe– Matt. 2016, 29, 013002.
Salpeter Equation Within Density Matrix Perturbation (87) Duchemin, I.; Guido, C. A.; Jacquemin, D.; Blase, X.
Theory. J. Chem. Phys. 2010, 133, 164109. The Bethe–Salpeter Formalism with Polarisable Con-
(73) Cudazzo, P.; Attaccalite, C.; Tokatly, I. V.; Ru- tinuum Embedding: Reconciling Linear-Response
bio, A. Strong Charge-Transfer Excitonic Effects and and State-Specific Features. Chem. Sci. 2018, 9, 4430–
the Bose-Einstein Exciton Condensate in Graphane. 4443.
Phys. Rev. Lett. 2010, 104, 226804. (88) Li, J.; Drummond, N. D.; Schuck, P.; Olevano, V.
(74) Garcia-Lastra, J. M.; Thygesen, K. S. Renormaliza- Comparing Many-Body Approaches Against the He-
tion of Optical Excitations in Molecules Near a Metal lium Atom Exact Solution. SciPost Phys. 2019, 6, 040.
Surface. Phys. Rev. Lett. 2011, 106, 187402. (89) Tirimbò, G.; Sundaram, V.; Çaylak, O.;
(75) Baumeier, B.; Andrienko, D.; Rohlfing, M. Frenkel Scharpach, W.; Sijen, J.; Junghans, C.; Brown, J.;
and Charge-Transfer Excitations in Donor–Acceptor Ruiz, F. Z.; Renaud, N.; Wehner, J. et al. Excited-state
Complexes From Many-Body Green’s Functions The- electronic structure of molecules using many-body
ory. J. Chem. Theory Comput. 2012, 8, 2790–2795. Green’s functions: Quasiparticles and electron–hole
(76) Duchemin, I.; Deutsch, T.; Blase, X. Short-Range excitations with VOTCA-XTP. J. Chem. Phys. 2020,
to Long-Range Charge-Transfer Excitations in the 152, 114103.
Zincbacteriochlorin-Bacteriochlorin Complex: A (90) Cammi, R.; Corni, S.; Mennucci, B.; Tomasi, J.
Bethe-Salpeter Study. Phys. Rev. Lett. 2012, 109, Electronic Excitation Energies of Molecules in So-
167801. lution: State Specific and Linear Response Methods
(77) Sharifzadeh, S.; Darancet, P.; Kronik, L.; Neaton, J. B. for Nonequilibrium Continuum Solvation Models. J.
Low-Energy Charge-Transfer Excitons in Organic Chem. Phys. 2005, 122, 104513.
Solids from First-Principles: The Case of Pentacene. (91) Phan Huu, D. K. A.; Dhali, R.; Pieroni, C.;
J. Phys. Chem. Lett. 2013, 4, 2197–2201. Di Maiolo, F.; Sissa, C.; Terenziani, F.; Painelli, A.
(78) Cudazzo, P.; Gatti, M.; Rubio, A.; Sottile, F. Frenkel Antiadiabatic View of Fast Environmental Effects on
versus Charge-Transfer Exciton Dispersion in Molec- Optical Spectra. Phys. Rev. Lett. 2020, 124, 107401.
ular Crystals. Phys. Rev. B 2013, 88, 195152. (92) Dreuw, A.; Head-Gordon, M. Single-Reference Ab
(79) Yin, H.; Ma, Y.; Mu, J.; Liu, C.; Rohlfing, M. Charge- Initio Methods for the Calculation of Excited States of
Transfer Excited States in Aqueous DNA: Insights Large Molecules. Chem. Rev. 2005, 105, 4009–4037.
from Many-Body Green’s Function Theory. Phys. Rev. (93) Nguyen, N. L.; Ma, H.; Govoni, M.; Gygi, F.; Galli, G.
Lett. 2014, 112, 228301. Finite-Field Approach to Solving the Bethe-Salpeter
(80) Li, J.; D’Avino, G.; Pershin, A.; Jacquemin, D.; Equation. Phys. Rev. Lett. 2019, 122, 237402.
Duchemin, I.; Beljonne, D.; Blase, X. Correlated (94) Ljungberg, M. P.; Koval, P.; Ferrari, F.; Foerster, D.;
electron-hole mechanism for molecular doping in or- Sánchez-Portal, D. Cubic-Scaling Iterative Solution of
ganic semiconductors. Phys. Rev. Materials 2017, 1, the Bethe-Salpeter Equation for Finite Systems. Phys.
025602. Rev. B 2015, 92, 075422.
(81) Rohlfing, M. Redshift of Excitons in Carbon Nan- (95) Yabana, K.; Bertsch, G. F. Time-Dependent Local-
otubes Caused by the Environment Polarizability. Density Approximation in Real Time. Phys. Rev. B
Phys. Rev. Lett. 2012, 108, 087402. 1996, 54, 4484–4487.
(82) Spataru, C. D. Electronic and Optical Gap Renormal- (96) Rabani, E.; Baer, R.; Neuhauser, D. Time-Dependent
ization in Carbon Nanotubes Near a Metallic Surface. Stochastic Bethe-Salpeter Approach. Phys. Rev. B
Phys. Rev. B 2013, 88, 125412. 2015, 91, 235302.
(83) Baumeier, B.; Rohlfing, M.; Andrienko, D. Electronic (97) Foerster, D.; Koval, P.; Sánchez-Portal, D. An O(N3)
Excitations in Push–Pull Oligomers and Their Com- Implementation of Hedin’s GW Approximation for
plexes with Fullerene from Many-Body Green’s Func- Molecules. J. Chem. Phys. 2011, 135, 074105.
tions Theory with Polarizable Embedding. J. Chem. (98) Wilhelm, J.; Golze, D.; Talirz, L.; Hutter, J.;
Theory Comput. 2014, 10, 3104–3110. Pignedoli, C. A. Toward GW Calculations on Thou-
(84) Duchemin, I.; Jacquemin, D.; Blase, X. Combining sands of Atoms. J. Phys. Chem. Lett. 2018, 9, 306–
the GW Formalism with the Polarizable Continuum 312.
Model: A State-Specific Non-Equilibrium Approach. (99) Rojas, H. N.; Godby, R. W.; Needs, R. J. Space-Time
J. Chem. Phys. 2016, 144, 164106. Method for Ab Initio Calculations of Self-Energies
(85) Li, J.; D’Avino, G.; Duchemin, I.; Beljonne, D.; and Dielectric Response Functions of Solids. Phys.
Blase, X. Combining the Many-Body GW Formal- Rev. Lett. 1995, 74, 1827.
ism with Classical Polarizable Models: Insights on (100) Liu, P.; Kaltak, M.; Klimeš, J. c. v.; Kresse, G. Cu-
the Electronic Structure of Molecular Solids. J. Phys. bic Scaling GW: Towards Fast Quasiparticle Calcula-
Chem. Lett. 2016, 7, 2814–2820. tions. Phys. Rev. B 2016, 94, 165109.

11
(101) Häser, M.; Almlöf, J. Laplace Transform Techniques Connection Fluctuation Dissipation Theorem Ap-
in Møller–Plesset Perturbation Theory. J. Chem. Phys. proach. J. Chem. Theory Comput. 2011, 7, 3116–
1992, 96, 489–494. 3130.
(102) Vlček, V.; Rabani, E.; Neuhauser, D.; Baer, R. (117) Furche, F.; Van Voorhis, T. Fluctuation-Dissipation
Stochastic GW Calculations for Molecules. J. Chem. Theorem Density-Functional Theory. J. Chem. Phys.
Theory Comput. 2017, 13, 4997–5003. 2005, 122, 164106.
(103) Lu, J.; Thicke, K. Cubic Scaling Algorithms for RPA (118) Caruso, F.; Rohr, D. R.; Hellgren, M.; Ren, X.;
Correlation Using Interpolative Separable Density Fit- Rinke, P.; Rubio, A.; Scheffler, M. Bond Break-
ting. J. Comput. Phys. 2017, 351, 187 – 202. ing and Bond Formation: How Electron Correla-
(104) Duchemin, I.; Blase, X. Separable resolution-of-the- tion Is Captured in Many-Body Perturbation Theory
identity with all-electron Gaussian bases: Applica- and Density-Functional Theory. Phys. Rev. Lett. 2013,
tion to cubic-scaling RPA. J. Chem. Phys. 2019, 150, 110, 146403.
174120. (119) Colonna, N.; Hellgren, M.; de Gironcoli, S. Correla-
(105) Gao, W.; Chelikowsky, J. R. Accelerating Time- tion Energy Within Exact-Exchange Adiabatic Con-
Dependent Density Functional Theory and GW Calcu- nection Fluctuation-Dissipation Theory: Systematic
lations for Molecules and Nanoclusters with Symme- Development and Simple Approximations. Phys. Rev.
try Adapted Interpolative Separable Density Fitting. B 2014, 90, 125150.
J. Chem. Theory Comput. 2020, 16, 2216–2223. (120) Hellgren, M.; Caruso, F.; Rohr, D. R.; Ren, X.; Ru-
(106) Seeger, R.; Pople, J. A. SelfâĂŘConsistent Molecular bio, A.; Scheffler, M.; Rinke, P. Static Correlation and
Orbital Methods. XVIII. Constraints and Stability in Electron Localization in Molecular Dimers from the
Hartree–Fock Theory. J. Chem. Phys. 1977, 66, 3045– Self-Consistent RPA and G W Approximation. Phys.
3050. Rev. B 2015, 91, 165110.
(107) Bauernschmitt, R.; Ahlrichs, R. Stability Analysis for (121) Maggio, E.; Kresse, G. Correlation Energy for the Ho-
Solutions of the Closed Shell Kohn–Sham Equation. mogeneous Electron Gas: Exact Bethe-Salpeter Solu-
J. Chem. Phys. 1996, 104, 9047–9052. tion and an Approximate Evaluation. Phys. Rev. B
(108) Sears, J. S.; Koerzdoerfer, T.; Zhang, C.-R.; Brédas, J.- 2016, 93, 235113.
L. Orbital Instabilities and Triplet States fFrom Time- (122) van Setten, M. J.; Caruso, F.; Sharifzadeh, S.;
Dependent Density Functional Theory and Long- Ren, X.; Scheffler, M.; Liu, F.; Lischner, J.; Lin, L.;
Range Corrected Functionals. J. Chem. Phys. 2011, Deslippe, J. R.; Louie, S. G. et al. GW 100: Bench-
135, 151103. marking G0 W 0 for Molecular Systems. J. Chem. The-
(109) Jacquemin, D.; Duchemin, I.; Blondel, A.; Blase, X. ory Comput. 2015, 11, 5665–5687.
Benchmark of Bethe-Salpeter for Triplet Excited- (123) Maggio, E.; Liu, P.; van Setten, M. J.; Kresse, G. GW
States. J. Chem. Theory Comput. 2017, 13, 767–783. 100: A Plane Wave Perspective for Small Molecules.
(110) Rangel, T.; Hamed, S. M.; Bruneval, F.; Neaton, J. B. J. Chem. Theory Comput. 2017, 13, 635–648.
An Assessment of Low-Lying Excitation Energies (124) Loos, P. F.; Romaniello, P.; Berger, J. A. Green
and Triplet Instabilities of Organic Molecules with an Functions and Self-Consistency: Insights From the
Ab Initio Bethe-Salpeter Equation Approach and the Spherium Model. J. Chem. Theory Comput. 2018, 14,
Tamm-Dancoff Approximation. J. Chem. Phys. 2017, 3071–3082.
146, 194108. (125) Duchemin, I.; Blase, X. Robust Analytic-Continuation
(111) Holzer, C.; Klopper, W. A Hybrid Bethe– Approach to Many-Body GW Calculations. J. Chem.
Salpeter/Time-Dependent Density-Functional-Theory Theory Comput. 2020, 16, 1742–1756.
Approach for Excitation Energies. J. Chem. Phys. (126) Olivucci, M. Computational Photochemistry; Elsevier
2018, 149, 101101. Science: Amsterdam; Boston (Mass.); Paris, 2010.
(112) Olsen, T.; Thygesen, K. S. Static Correlation Be- (127) Navizet, I.; Liu, Y.-J.; Ferre, N.; Roca-
yond the Random Phase Approximation: Dissociat- Sanjun, D.; Lindh, R. The Chemistry of Biolumi-
ing H2 With the Bethe-Salpeter Equation and Time- nescence: An Analysis of Chemical Functionalities.
Dependent GW. J. Chem. Phys. 2014, 140, 164116. ChemPhysChem 2011, 12, 3064–3076.
(113) Holzer, C.; Gui, X.; Harding, M. E.; Kresse, G.; Hel- (128) Loos, P. F.; Scemama, A.; Jacquemin, D. The Quest
gaker, T.; Klopper, W. Bethe-Salpeter Correlation En- for Highly-Accurate Excitation Energies: a Computa-
ergies of Atoms and Molecules. J. Chem. Phys. 2018, tional Perspective. J. Phys. Chem. Lett. 2020, submit-
149, 144106. ted.
(114) Li, J.; Duchemin, I.; Blase, X.; Olevano, V. Ground- (129) Roos, B. O.; Andersson, K.; Fulscher, M. P.;
state correlation energy of beryllium dimer by the Malmqvist, P.-A.; Serrano-Andrés, L. In Multiconfig-
Bethe-Salpeter equation. SciPost Phys. 2020, 8, 20. urational Perturbation Theory: Applications In Elec-
(115) Loos, P.-F.; Scemama, A.; Duchemin, I.; tronic Spectroscopy; Prigogine, I., Rice, S. A., Eds.;
Jacquemin, D.; Blase, X. Pros and Cons of the Bethe- Adv. Chem. Phys.; Wiley, New York, 1996; Vol.
Salpeter Formalism for Ground-State Energies . 2020, XCIII; pp 219–331.
(116) Angyan, J. G.; Liu, R.-F.; Toulouse, J.; Jansen, G. (130) Furche, F.; Ahlrichs, R. Adiabatic Time-Dependent
Correlation Energy Expressions from the Adiabatic-

12
 Density Functional Methods for Excited State Proper- Frequency-Dependent Second-Order Bethe-Salpeter
ties. J. Chem. Phys. 2002, 117, 7433. Correlation Kernel. J. Chem. Phys. 2016, 144,
(131) Lazzeri, M.; Attaccalite, C.; Wirtz, L.; Mauri, F. Im- 094107.
pact of the Electron-Electron Correlation on Phonon (145) Olevano, V.; Toulouse, J.; Schuck, P. A formally
Dispersion: Failure of LDA and GGA DFT Function- exact one-frequency-only Bethe-Salpeter-like equa-
als in Graphene and Graphite. Phys. Rev. B 2008, 78, tion. Similarities and differences between GW+BSE
081406. and self-consistent RPA. J. Chem. Phys. 2019, 150,
(132) Faber, C.; Janssen, J. L.; Côté, M.; Runge, E.; Blase, X. 084112.
Electron-Phonon Coupling in the C60 Fullerene within (146) Ankudinov, A. L.; Nesvizhskii, A. I.; Rehr, J. J. Dy-
the Many-body GW Approach. Phys. Rev. B 2011, 84, namic Screening Effects in X-Ray Absorption Spectra.
155104. Phys. Rev. B 2003, 67, 115120.
(133) Yin, Z. P.; Kutepov, A.; Kotliar, G. Correlation- (147) Baumeier, B.; Andrienko, D.; Ma, Y.; Rohlfing, M.
Enhanced Electron-Phonon Coupling: Applications Excited States of Dicyanovinyl-Substituted Oligothio-
of GW and Screened Hybrid Functional to Bis- phenes from Many-Body Green’s Functions Theory.
muthates, Chloronitrides, and Other High-Tc Super- J. Chem. Theory Comput. 2012, 8, 997–1002.
conductors. Phys. Rev. X 2013, 3, 021011.
(134) Monserrat, B. Correlation Effects on Electron-Phonon
Coupling in Semiconductors: Many-Body Theory
Along Thermal Lines. Phys. Rev. B 2016, 93, 100301.
(135) Li, Z.; Antonius, G.; Wu, M.; da Jornada, F. H.;
Louie, S. G. Electron-Phonon Coupling from Ab Ini-
tio Linear-Response Theory within the GW Method:
Correlation-Enhanced Interactions and Superconduc-
tivity in Ba1−x Kx BiO3 . Phys. Rev. Lett. 2019, 122,
186402.
(136) Ismail-Beigi, S.; Louie, S. G. Excited-State Forces
within a First-Principles Green’s Function Formalism.
Phys. Rev. Lett. 2003, 90, 076401.
(137) Rohlfing, M.; Louie, S. G. Electron-hole Excitations
and Optical Spectra from First Principles. Phys. Rev.
B 2000, 62, 4927–4944.
(138) Ma, Y.; Rohlfing, M.; Molteni, C. Excited States
of Biological Chromophores Studied Using Many-
Body Perturbation Theory: Effects of Resonant-
Antiresonant Coupling and Dynamical Screening.
Phys. Rev. B 2009, 80, 241405.
(139) Ma, Y.; Rohlfing, M.; Molteni, C. Modeling the
Excited States of Biological Chromophores within
Many-Body Green’s Function Theory. J. Chem. The-
ory. Comput. 2009, 6, 257–265.
(140) Romaniello, P.; Sangalli, D.; Berger, J. A.; Sottile, F.;
Molinari, L. G.; Reining, L.; Onida, G. Double Exci-
tations in Finite Systems. J. Chem. Phys. 2009, 130,
044108.
(141) Sangalli, D.; Romaniello, P.; Onida, G.; Marini, A.
Double Excitations in Correlated Systems: A
Many–Body Approach. J. Chem. Phys. 2011, 134,
034115.
(142) Huix-Rotllant, M.; Ipatov, A.; Rubio, A.;
Casida, M. E. Assessment of Dressed Time-
Dependent Density-Functional Theory for the Low-
Lying Valence States of 28 Organic Chromophores.
Chem. Phys. 2011, 391, 120–129.
(143) Zhang, D.; Steinmann, S. N.; Yang, W. Dynami-
cal second-order Bethe-Salpeter equation kernel: A
method for electronic excitation beyond the adiabatic
approximation. J. Chem. Phys. 2013, 139, 154109.
(144) Rebolini, E.; Toulouse, J. Range-Separated Time-
Dependent Density-Functional Theory with a

13