Perspective: Fundamental Aspects of Time-Dependent Density Functional Theory

Neepa T. Maitra1
1
Department of Physics and Astronomy, Hunter College and the Physics Program at the Graduate
Center of the City University of New York, 695 Park Avenue, New York, New York 10065, USA
(Dated: May 31, 2016)
In the thirty-two years since the birth of the foundational theorems, time-dependent density func-
tional theory has had a tremendous impact on calculations of electronic spectra and dynamics in
chemistry, biology, solid-state physics, and materials science. Alongside the wide-ranging applica-
tions, there has been much progress in understanding fundamental aspects of the functionals and
the theory itself. This Perspective looks back to some of these developments, reports on some recent
progress and current challenges for functionals, and speculates on future directions to improve the
accuracy of approximations used in this relatively young theory.

The wavefunction rose from the early days of quan- dent Erich Runge, were particularly motivated by the
tum mechanics as the central player, the provider of all question of what is the time-dependent potential acting
observable properties of atoms, molecules and solids. on an electron when it scatters from ions whose nuclei
Within a few years it was however recognized that find- are treated classically. Given that Hohenberg, Kohn, and
ing this complex-valued high-dimensional function was Sham had derived such a potential for an electron in a
impossible for all but very few problems, even if we ground-state twenty years earlier, it was natural to won-
just focus on electrons alone. Almost 40 years later, the der whether such a formulation could be extended to the
Hohenberg-Kohn theorem [1] proved that the ground- time-dependent case, yielding the exact time-dependent
state density alone provides all observable properties of potential acting on an electron. This led to the birth of
any static system. This is an astonishing result given the the foundational theorem of TDDFT, the Runge-Gross
simplicity of the density, the probability of finding any theorem [2].
one electron at a given point in space, compared with What Runge and Gross considered, were non-
the wavefunction, a function of all electronic coordi- equilibrium systems of electrons treated non-
nates. Even more astounding, in 1984 Erich Runge and relativistically, described by the following universal
Hardy Gross proved that for time-dependent (TD) sys- Hamiltonian:
tems evolving from a given initial wavefunction, all TD
N   X N XN
properties can be extracted from the time-evolving den- X 1 1
sity [2]. Their theory, Time-Dependent Density Func- Ĥ(t) = − ∇2i + vext (ri , t) + (1)
i
2 i>j j
|ri − rj |
tional Theory (TDDFT) has become one of the most pop-
ular and successful methods for computing electronic
We use atomic units (e2 = ~ = me = 1) through-
excitation spectra today, with increasing applications for
out. What distinguishes different N -electron systems
non-perturbative electron dynamics, and is the subject
is the “external” potential, vext (r, t), that represents the
of this Perspective. I focus here mostly on fundamental
potential the electrons experience due to the nuclear
aspects underlying the theory and its functionals, while
attraction and due to any field applied to the system.
applications enter throughout as a guide and inspira-
The time-dependence in this term can arise through nu-
tion.
clear motion (with nuclei treated as classical, but pos-
The very first TD density functional calculation, to sibly dynamic, point particles) and/or through exter-
my knowledge, in fact predates not only the Runge- nally applied (scalar) fields. Of course, solving the
Gross theorem, but also the Hohenberg-Kohn theorem: TD Schrödinger equation (TDSE), Ĥ(t)Ψ(t) = i∂t Ψ(t)
Felix Bloch developed a time-dependent extension of for the many-electron wavefunction Ψ(t) is not com-
Thomas-Fermi theory [3, 4] to theoretically model the putationally feasible for more than a few electrons, as
stopping power of a fast charged particle in matter. In it runs into an exponential wall very quickly for non-
the 1970’s, the first time-dependent Kohn-Sham calcu- perturbative fields. Moreover Ψ(t) provides usually far
lations were run [5–7], and in the early 1980’s proofs of more information than we would be interested in. Typ-
one-to-one density-potential mappings for certain cases ically, observables of interest involve integrated quan-
began to appear, for time-periodic potentials [8] and adi- tities, in particular one- and two-body probability den-
abatic processes[9, 10]. The growing interest in running sities. The idea of density-functional theories, broadly
time-dependent Kohn-Sham calculations did not have speaking, is to obtain such quantities directly without
to wait much longer before they could stand on a formal, calculating Ψ(t).
and general, foundation. The Institute for Theoretical Time-dependence opens up an ever-increasing wealth
Physics at the Goethe University Frankfurt in the 1980’s of interesting physical and chemical phenomena that
was especially known for nuclear physics and atomic tend to be classified into two groups: spectroscopy,
scattering. Hardy Gross, a postdoc at the time, and stu- which gives an electronic “fingerprint” of the atom,
molecule, or solid (Sec. III), and real-time non- considered potentials be time-analytic around the ini-
perturbative dynamics where the electronic system tial time, and used their Taylor expansion expressions
roams far from any ground-state (Secs. II and IV). In in the equation of motion for time-derivatives of the
this Perspective, I cover only in passing a third group of current-density, evaluated about the initial time: Essen-
phenomena arising from relaxation and dissipation in tially, the difference ∂tl j(r, t)|t=0 − ∂tl j0 (r, t)|t=0 depends
very large or extended systems, as induced by electron- on the difference between the k<lth Taylor coefficients
electron viscosity. A fourth group of applications rele- of the two potentials. If the two potentials are physically
vant to the ground-state correlation energy, which ex- distinct, then there must exist a k for which their spatial-
ploit the fluctuation-dissipation theorem to fold in long- dependences differ, from which the equation shows that
range effects in a natural way [11], is not discussed. the resulting currents will differ. In a second step, the
Further, I have brazenly ignored all-important compu- equation of continuity, ∂t n(r, t) = −∇ · j(r, t), is used to
tational concerns, numerical aspects, details, and algo- transform the one-to-one current-potential mapping to
rithms. a one-to-one density-potential mapping.
The top part of Figure 1 illustrates this map. The ini-
tial states labelling the upper ellipses have the prop-
I. DENSITY-POTENTIAL MAPPING
erty that they have the same n(r, 0) and ∂t n(r, 0) as
each other, and the symbols inside the ellipses are ex-
The Runge-Gross theorem proves the one-to-one amples of possible density-evolutions n(r, t) stemming
mapping from them. (The phase of the initial wavefunction deter-
1−1 mines ∂t n(r, 0), via the equation of continuity). Differ-
Ψ(0) : n ←→ vext (2) ent initial states can lead to the same density evolution
where Ψ(0) is the initial wavefunction and the density in different potentials, so some density symbols appear
n(r, t) is the probability of finding any one electron of in both. The Runge-Gross theorem states that no two
any spin σ at point r in space at time t: lines from the same ellipse may lead to the same vext
in the central ellipse, which is the space of all one-body
X Z
n(r, t) = N d3 r2 ..d3 rN |Ψ(rσ, r2 σ2 ...rN σN , t)|2 . potentials, but lines from different densities in two dif-
σ,σ2 ...σN ferent ellipses may point to either the same or different
(3) points in the central ellipse. The open symbols corre-
As in the ground-state case, an elegant aspect of the spond to “non-v-representable” densities, meaning that
original proof of this density-potential mapping, is that these densities cannot be generated from evolving un-
it can be completely followed from basic principles der any Hamiltonian of the form Eq. 1. We come back to
learned in the first year of a quantum mechanics course. the lower part of the figure shortly.
Instead of the variational principle for energy minimiza-
tion that leads to the Hohenberg-Kohn theorem, the TD
case follows from Heisenberg equations of motion for
the current-density, and the continuity equation. The
basic idea of the proof is outlined here, and readers are
referred to Refs. [2, 12, 13] for the details.
Consider two systems of N electrons, both initially in
the same many-body state Ψ(0). Can two, physically
0
distinct potentials, vext (r, t) and vext (r, t) be found such
that the two systems evolve in a Hamiltonian of the form
Eq. (1) with exactly identical densities n(r, t) = n0 (r, t)
at all times? Physically distinct potentials are those
that differ by more than just a purely TD function, i.e.
vext (r, t) and vext (r, t) + α(t) are not physically distinct,
because they lead to wavefunctions that differ only by
a purely-TD phase factor, which cancels out when ex-
pectation values of Hermitian operators are taken in cal-
culating observables. Now, if the answer to the above FIG. 1. Illustration of the one-to-one density-potential map in
question can be shown to be “no”, then it implies that TDDFT. The upper (lower) ellipses contain density evolutions
the density n(r, t) can only be produced by one external of interacting (non-interacting) systems arising from the ini-
tial state labelling the ellipse. The four initial states shown all
potential, i.e. it proves the arrow pointing to the right in
have the same density as each other, and first time-derivative
Eq. (2). (The arrow pointing to the left, that the poten- of the density as each other. The potentials in the central ellipse
tial points to a unique density follows from the usual are one-body potentials, representing vext (vS ) when pointed
assumption that the TDSE yields a unique Ψ(t), and to from the upper (lower) ellipses.
hence via Eq. (3) a unique density). To prove the an-
swer is in fact “no”, Runge and Gross required that the

2
 A. The TD Kohn-Sham system The initial-state dependence of the 1-1 mapping for
the interacting and non-interacting systems endows the
The statement in Eq. (2) implies that all observables xc potential with a functional-dependence on the initial
are functionals of the initial state Ψ(0) and the TD den- states Ψ(0) and Φ(0), as manifest in Eq. (5), a feature that
sity n(r, t): O[n; Ψ(0)]. This is because it states that, for has no precedent in ground-state DFT. In a large class of
a given initial state, knowledge of the density identifies applications (e.g. calculations of response and spectra),
the external potential, and therefore the entire Hamil- the initial true and KS states are ground states, which, by
tonian Eq. (1), and hence the wavefunction Ψ(t) (up virtue of the Hohenberg-Kohn theorem, are themselves
to a physically irrelevant purely TD phase). But, as in functionals of their density; initial-state dependence is
the ground-state case, it is difficult to find these func- then redundant. But it is generally unknown what the
tionals, and so one resorts to mapping the system to a functional dependence of the xc potential is on the initial
non-interacting system of orbitals, the TD Kohn-Sham states. The exact xc potential is known to depend on the
(KS) system, which reproduces the exact density. Hav- past also through the history of the density, n(t0 < t),
ing orbitals eases the job of the functionals in many and this dependence together with the dependence on
cases; in particular, the kinetic energy, approximated as Ψ(0) and Φ(0) is called memory-dependence [16, 17]. In the
2 vast majority of applications today, an adiabatic approx-
the orbital functional i hφi | − ∇2 |φi i does a far bet-
P
imation is used, where the TD density is inserted into a
ter job than available explicit kinetic-energy density-
ground-state approximation, audaciously neglecting all
functionals. The Runge-Gross argument holds for non-
memory-dependence.
interacting electrons (in fact, for any particle-particle in-
teraction), as illustrated in the lower part of Figure 1. Note that the TD KS equations do not arise from a
Again, no two lines from the same lower ellipse can variational principle: all that was needed above is the
lead to the same one-body potential in the central el- reassurance that a given density evolution beginning
lipse, representing now vS (r, t), and again, lines from the from a given initial state can be produced by at most
same symbol appearing in both lower ellipses leading to one potential, and the assumption of non-interacting
two different potentials, reflects the fact that two differ- v−representability (see more shortly). Still, thinking
ent initial states can yield the same density evolution in of the ground-state case, where many approximations
different potentials. A density evolution in both an up- are derived directly for the xc energy, whose func-
per ellipse and a lower ellipse means that the density is tional derivative then gives the ground-state potential, it
“non-interacting v-representable” (more in Sec I B). might be similarly useful if there is an analog for the TD
To run a TDDFT calculation for a given problem then, case. TD variational principles are based on the action,
one first picks an initial state to be used for the non- but, if one attempted to write vXC [n, Ψ(0), Φ(0)](r, t) as
interacting KS propagation, Φ(0), that is compatible δAXC [n, Ψ(0), Φ(0)]/δn(r, t) one encounters a predica-
with the true initial state of the physical problem Ψ(0) ment: δvXC [n, Ψ(0), Φ(0)](r, t)/δn(r0 , t0 ) would equal
in that it has the same density and first time-derivative δ 2 AXC [n, Ψ(0), Φ(0)]/δn(r, t)δn(r0 , t0 ) which is symmet-
of the density. Typically a Slater determinant of single- ric in t and t0 , while in reality density-changes at times
particle orbitals φi (r, 0) is chosen. Even when the initial t0 > t cannot affect the xc potential at earlier times. This
density is stationary, there are infinitely many Slater de- causality violation was first pointed out in Ref. [18] and
terminants to choose from that reproduce a given den- several different possible solutions have been proposed:
sity [14, 15]. The KS orbitals evolve via a single-particle defining the action on the Keldysh contour [19], using
TDSE whose potential, vS [n; Φ(0)] is defined such that it Liouville space pathways [20], or including boundary
guarantees the KS one-body density reproduces the ex- terms with the usual real-time contour [21].
PN
act density: i=1 |φi (r, t)|2 = n(r, t) where
∇2
 
B. Solving an Existential Crisis Leads to Alternative
− + vS (r, t) φi (r, t) = i∂t φi (r, t) (4) Mapping Proofs
2
The KS potential vS (r, t) = vS [n; Φ(0)](r, t) is usually
How do we even know that the time-evolving density
written as
of an interacting electronic system can be reproduced by
vS (r, t) = vext (r, t) + vH [n](r, t) + vXC [n; Ψ0 , Φ0 ](r, t) (5) a non-interacting system? The Runge-Gross proof en-
0 sures the uniqueness of the potential that yields a given
where vH [n](r, t) = d3 r0 n(r ,t)
R
|r−r0 | is the Hartree poten- density evolution from a given initial state but it does
tial, and vXC [n; Ψ0 , Φ0 ](r, t) is the exchange-correlation not prove its existence. The question of existence of a
(xc) potential. In practical calculations, approximations KS system for a given density is called “non-interacting
enter first when selecting an initial KS state (the exact v-representability”.
initial density is generally only known approximately, In Ref. [22], Robert van Leeuwen showed that a
e.g. via a ground-state DFT calculation, or a correlated unique TD KS potential exists for a prescribed TD den-
many-body calculation), and second in approximating sity generated in an interacting system, under some con-
the xc functional. ditions, and provided a method for its construction. His

3
method also provides an alternate proof of the density- to the true system, one needs a magnetic field in a non-
potential mapping theorem of the statement (2). Like interacting system for it to reproduce the current. That
the Runge-Gross theorem, the van Leeuwen proof re- is, the KS current in TDDFT is, generally, not equal to
quires that the potential be time-analytic, but addition- the true physical curent. The longitudinal component
ally the latter assumes the density is time-analytic. Now of the TDDFT KS current is equal to that of the physical
the time-analyticity of the potential covers most physi- current, thanks to the continuity equation, but they may
cal cases of interest (especially with the piece-wise ana- differ in their transverse component.
lytic extension [16]), but time-analyticity of the density But even stepping back to the very first stage, the
is a more restrictive condition. The coupling of space question of whether one can find an initial Slater de-
and time-derivatives in the TDSE means that spatial sin- terminant that reproduces a given initial density and
gularities in the potential can lead to non-analyticities in current-density of an interacting system is highly non-
time in the wavefunction and density, even when the po- trivial, unless the velocity j(r, 0)/n(r, 0) is curl-free; re-
tential is time-analytic. Technically the proof in Ref. [22] cent results can be found in Ref. [36].
unfortunately does not apply to Coulomb potentials. A more useful one-to-one map exists between current-
Limiting consideration to the linear response regime, al- densities and vector potentials [18, 37–40], yield-
ternate proofs of the density-potential mapping how- ing what is known today by TD current-density
ever have no time-analyticity requirements [23, 24]. functional theory (TDCDFT). Here non-interacting A-
The search for a general proof free of the time- representability was proven under conditions of time-
analyticity requirements has seen a resurgence of inter- analyticity [41], similar to the van Leeuwen proof of
est in recent years. Ref. [25] reformulated the question of v−representability in the TDDFT case. Closely related
the density-potential mapping and v−representability is TD deformation functional theory [42], which can be
in terms of uniqueness and existence, respectively, of a viewed as a Lagrangian version of TDCDFT, with the
particular TD nonlinear Schrödinger equation (NLSE). deformation metric tensor as basic variable. In fact, Ilya
However the nature of the particular NLSE involves Tokatly has illuminated the relationships between the
fourth-order derivatives of the wavefunction and be- NLSEs of TDDFT, TDCDFT, and that underlying TD de-
cause of this together with the singularity in the formation functional theory [43].
Coulomb potential, existence and uniqueness results are
Even when the external potential is purely scalar,
not known. Refs. [26–28] proposed a proof based on
there can be advantages to use TDCDFT instead of
global fixed-point iteration, which further yields an al-
TDDFT when approximations are involved. Going be-
gorithm to numerically construct the exact KS poten-
yond the adiabatic approximation was a motivating fac-
tial for a prescribed density-evolution from a given ini-
tor for Giovanni Vignale and Walter Kohn, in deriving
tial state. The convergence is technically not assured
the “Vignale-Kohn” (VK) approximation in TDCDFT.
for Coulombic systems, but demonstrations on systems
Consider trying to incorporate memory-dependence in
with smooth potentials have already led to interesting
TDDFT. The xc potential at time t depends on the den-
insights on features that exact xc potentials have [29–31].
sity at earlier times t0 < t, when a small volume element
Ref. [32] provides a detailed review of the mathematical
which is now at r was, at that earlier time, at r0 [44],
aspects of the density-potential mapping, delving also
suggesting that memory resides with the fluid element
into uniqueness and existence of solutions of the TDSE
rather than with the local density. This entanglement
when Coulomb interactions are present.
of spatial- and time- nonlocality was first brought to
Discretizing space, conditions for existence and
light by John Dobson in Ref. [45], where he showed that
uniqueness of the density-potential mapping and
simply modifying ALDA by using the finite-frequency
KS system in “lattice-TDDFT” have been proven in
response of the uniform gas while retaining only local
Refs. [33, 34]. For practical computations on a real-space
density-dependence as in the Gross-Kohn approxima-
grid, one might be tempted argue that the lattice-TDDFT
tion [46], violated exact conditions, such as the harmonic
is what is relevant, however without well-posedness of
potential theorem. Vignale showed this approximation
the continuum limit, there is no guaranteed convergence
also violates the zero-force theorem (Sec. II) [47]. Any
with respect to grid size.
nonadiabatic xc functional must depend in a strongly
spatially-non-local way on the density, even when in
the limit of spatially-slowly-varying densities. A the-
C. TD Current Density Functional Theory ory local in space and non-local in time is instead pos-
sible in terms of the current-density [40, 48–50], and
The first part of the RG theorem actually proves that TDCDFT with the VK approximation has proven suc-
no two scalar potentials yield the same one-body cur- cessful for a number of applications where the stan-
rent, for a fixed initial state, implying a one-to-one dard approximations in TDDFT perform poorly. Due
mapping between the external potential and the cur- to its inclusion of retardation effects, it has had some
rent. However, currents of interacting systems are gen- success in capturing relaxation due to electron-electron
erally not non-interacting v-representable, as shown in interaction in large or periodic systems [51, 52], vis-
Ref. [35]: even if no external magnetic field is applied cous forces responsible for the stopping power in elec-

4
tron liquids [53], spin-Coulomb drag [54], the frequency notice that the components from each orbital can indi-
and linewidth of intersubband transitions in quantum vidually be very different, but they sum to the same vS .
wells [55], and many-body dynamical corrections to It is clear that the KS potential has a complicated struc-
conductance in molecular junctions [56]. Spatially local ture coming from all contributions, going beyond just a
functionals of the current incorporate non-local density- static screening of the “nuclear charge” in vext .
dependence, which allows VK to correct the overestima-
tion in ALDA/GGA of polarizabilities for many poly-
mer chains [57] and to improve the dielectric function
in semiconductors to some extent [58]. However its
general application has been limited, partly because it
spuriously damps finite systems [52], and, even if only
applied to extended systems, the electron gas response
functions that enter the VK functional are not known
with certainty at general frequencies and wavevectors.

II. EXCHANGE-CORRELATION POTENTIALS: WHAT

DO WE KNOW EXACTLY?

The TDDFT theorems say that the dynamics of the

electrons in any atom, molecule, or solid can be exactly
extracted from completely removing the electron repul-
sion while distorting the attractive potential of the nu-
clei. What kind of crazy distortion is required for this to
happen? Here, we discuss some general aspects about
FIG. 2. Exact KS potential for field-free evolution of the 50:50
what is known about the exact potential.
superposition state example (see text). The upper plots show
The exact TD KS potential By inverting the TD KS the three contributions to Eq. (7) from one orbital at two differ-
equation, we find an expression for the exact KS po- ent times, while the lower plots show those of the other orbital.
tential in terms of the density ni (r, t) and phase αi (r, t)
of
p any one of the occupied KS orbitals φi (r, t) =
Although Eq. (6) and Eq. (7) give an explicit orbital-
ni (r, t)eiαi (r,t) :
functional for the KS potential, the functional itself is
p utterly useless! Without input (knowledge) of the ex-
∇2 ni (r, t) (∇αi (r, t))2
vS (r, t) = p − − ∂t αi (r, t) , (6) ternally applied potential, it clearly cannot be used to
2 ni (r, t) 2 propagate the orbital. (Nevertheless Eq. (6) and Eq. (7)
where ∇ · (ni ∇αi (r, t)) = −∂t (ni (r, t)) can be instructive in analysing/understanding features
of the exact xc potential, e.g. step features arising from
expresses the continuity equation for orbital φi . In one- the integral of the local acceleration [59, 60], see Sec IV).
dimension (1D), this simplifies in terms of ni , the orbital- In fact, if one did try to propagate using Eq. (6) or (7),
velocity ui = ji /ni and the orbital-acceleration ∂t ui , of one immediately runs into a problem: At the initial time,
any one of the occupied orbitals: the choice of the initial KS state specifies the orbitals, so
that, on the right-hand-side of Eq. (7) for example, ni
x
p
∇2 ni (x, t) 1
Z
and ui are known, however the orbital-acceleration ∂t ui
vS (x, t) = − ui (x, t)2 − ∂t ui (x0 , t)dx0
is not determined by the initial state. In fact, the exter-
p
2 ni (x, t) 2
(7) nal potential determines the acceleration, and is hiding
As an illustration, in Fig. 2, we plot the exact KS po- in the functional-dependence in this term. It is essen-
tential for a model problem: field-free evolution of the tial to realize that the external potential is given by the
√ problem at hand, and is not to be treated as a density-
initial state Ψ(0) = (Ψ0 + Ψ1 )/ 2 where Ψ0,1 are the
ground and first excited singlet states in a 1D He atom. functional [61]; otherwise one runs into misunderstand-
The initial KS state Φ(0) has the same configuration ings concerning the predictivity of the TD KS evolution.
but in terms of non-interacting Slater determinants [31], With the external potential vext (r, t), initial many-body
with the KS orbitals φ0 and φ1 that form these determi- state Ψ(0) and initial choice of KS state Φ(0), as inputs,
nants being eigenstates of some non-interacting Hamil- the exact TD KS propagation is predictive: extracting an
tonian. The top panels show the KS potential and its exact expression for the xc potential alone shows that it
decomposition into the three terms on the right-hand- depends only on times past [12, 61] as we next show.
side of Eq. (7), labelled as adia, veloc, and acc, respec- The exact xc potential If we differentiate the continuity
tively, evaluated using φ0 (φ1 ) in the upper(lower) pan- equation with respect to time for each of the interact-
els, at the initial time, and about half a period later. We ing systems and KS systems, requiring the density to be

5
identical, we obtain [22, 62–64] state xc potential corresponding to the instantaneous
density may be found. In this way, the AE isolates er-
∇·(n(r, t)∇vXC (r, t)) = ∇·(aS (r, t) − a(r, t) − n(r, t)∇vH (r, t))rors due to the adiabatic approximation alone, and is the
(8) “best” that an adiabatic approximation could do.
where a(r, t) = −ihΨ(t)|[ĵ(r), T̂ + Ŵ |Ψ(t)i and aS (r, t) = If a ground-state orbital functional, such as exact-
−ihΦ(t)|[ĵ(r), T̂ |Φ(t)i. Here T̂ is the kinetic energy op- exchange, or self-interaction corrected LDA (SIC-LDA),
g.s.
erator (first term on the right in Eq. (1)), and Ŵ is the is used as vXC then it contains some memory: a func-
electron-electron interaction (third term in Eq. (1)). At tional that depends on instantaneous orbitals depends
the initial time, the right-hand-side involves expectation on the history of the density and the initial KS state
values in the given initial states, showing that vXC (r, t = since the orbitals are themselves implicit functionals of
0) can be constructed as a functional of the initial states these. The xc potential is technically given by the time-
alone (unlike vS and vext ), and therefore together with dependent optimized effective potential (OEP) equa-
the vext prescribed by the problem at hand, and vH com- tions [19, 67]. Although not always [68, 69], these are
puted from the density, vS can be constructed and the most often solved within the KLI approximation, at the
propagation can proceed. sacrifice of the zero-force theorem [70].
Now, Eq. (8) is an exact expression for the xc potential, Known exact conditions The satisfaction of exact condi-
which is instructive to rewrite as: tions has played, and continues to play, an important
h1
role in the development of ground-state functionals (see
(∇0 − ∇) ∇2 − ∇02 ρ1c (r0 , r, t)|r0 =r


∇ · (n∇vXC ) = ∇ ·
4Z Ref. [71] for a recent advance). When such functionals
i are adopted in an adiabatic functional for TDDFT, these
+ n(r, t) nXC (r0 , r, t)∇w(|r0 − r|)d3 r0 , (9) exact conditions can endow v A with desirable proper-
XC

0
ties also for dynamics and excitations. An important ex-
where ρ1c = ρ1 − ρ1,S with ρ1 (r , r, t) = ample of this is the asymptotic behavior of the poten-
N σ1 ..σN d3 r2 ..d3 rN Ψ∗ (r0 σ1 ..rN σN ; t)Ψ(rσ1 ..rN σN ; t)
P R
tial far from an atom or molecule: vXC (r) → −1/r [72].
is the spin-summed one-body density-matrix of the This is perhaps more important in the TD case than the
true system of electrons with two-body interaction ground-state case, because the density is more likely to
0 0
potential w(|r − r |), ρ1,S (r , r, t) is the one-body density- “sample” regions far from the nuclei over time, and cer-
matrix for the KS system, and nXC (r0 , r, t) is the xc tainly the accuracy of energies of higher-lying excita-
0
hole,
P defined R via0 the pair density, P (r ,2r,3t) = 3N (N − tions depend on this decay behavior (see also Sec. III).
1) σ1 ..σN |Ψ(r σ1 , rσ2 , r3 σ3 ..rN σN ; t)| d r3 ..d rN = Relatedly, via Koopman’s theorem, H = −I, the highest
n(r, t) (n(r0 , t) + nXC (r0 , r, t)) . Eq. (9) is a Sturm- occupied molecular orbital (HOMO) energy determines
Liouville equation for vXC that has a unique solution the onset of continuous absorption in the spectrum.
for a given boundary condition [22]. The first term rep- Several exact conditions that are specific to the TD
resents a kinetic contribution to the potential, and the case are known, and some have shaped the develop-
second term is an interaction component. The problem ment of non-adiabatic functional approximations. For
of approximating the xc potential as a functional of the example, the harmonic potential theorem [45] and the
density and initial-states is therefore here re-cast as a zero-force theorem [47] were instrumental in the discov-
problem of modeling the interacting density-matrix and ery that a spatially-local but time-non-local approxima-
xc hole as functionals of the density and initial-states. tion is not possible in terms of the density, but is possi-
The adiabatically-exact approximation In most practi- ble in terms of the current-density (Sec. I C). The har-
cal applications, an adiabatic approximation is used monic potential theorem, or generalized Kohn’s theo-
for the xc potential, which inserts the instanta- rem, states that the electron density in a harmonic poten-
neous density into a ground-state approximation, tial subject to a uniform TD selectric field rigidly follows
A g.s.
vXC [n; Ψ(0), Φ(0)](r, t) = vXC [n(t)](r). Even if the in- the classical oscillation of the center of mass. The zero-
stantaneous density happens to be a density of some force theorem is an expression of Newton’s third law,
ground state, the underlying interacting and KS states requiring the total force exerted on the system by the xc
A
typically are not ground states, and so use of vXC can and Hartree potentials to vanish. For TDCDFT, the anal-
lead to significant error. There are two sources of er- ogous theorem for the xc torque is also satisfied [49]. Im-
A
ror in vXC ; one is in neglecting all memory-dependence, posing these conditions were key in leading to the non-
and the other is in the choice of the ground-state func- adiabatic VK approximation of TDCDFT [40, 48] (Sec
tional approximation. To disentangle the two sources, II.C).
the “adiabatically-exact” (AE) approximation is a useful Exact conditions unique to the TD problem, such
analytic tool, defined by using the exact ground-state xc as the zero-force and harmonic potential theorems,
potential [65, 66]: naturally involve the memory-dependence (time-non-
AE
vXC [n; Ψ(0), Φ(0)](r, t) = vXC exactg.s.
[n(t)](r) (10) locality) of the xc functional. Another such condition
is the “memory-condition” which entangles the initial-
Obtaining such an approximation is clearly only possi- state dependence and history-dependence of the xc po-
ble for simple model systems where the exact ground- tential [16]. Essentially it states that one can consider any

6
fixed time t0 along an evolution as the initial time, and
P
where n̂(r) = i δ(r − rˆi ) is the density operator, ΩI =
so vXC [nt0 ; Ψ(t0 ), Φ(t0 )](r, t > t0 ), where nt0 is the density EI − E0 is the excitation frequency of the Ith excited
function from time t0 onwards, must be the same for all state, and the sum goes over all interacting states ΨI .
choices of such t0 . This is a very hard condition to satisfy The notation c.c.(−ω) means the complex conjugate of
for any memory-dependent functional, and is actually the first term evaluated at −ω.
violated by VK.
Due to its complete lack of memory, the adiabatic ap-
proximation satisfies the memory-condition, the zero-
force and harmonic potential theorems. But it vio-
lates the “resonance condition” [73], with severe conse-
quences for resonantly-driven processes, and possibly
for general non-equilibrium dynamics. This is a condi-
tion on the density-response function of a system in a
general non-stationary state, and expresses the fact that
the excitation frequency of a given transition is indepen-
dent of the state. This involves a delicate cancellation
of time-dependences between the KS response function
around the KS non-stationary state and shifts induced
by the generalized xc kernel such that the poles of the
true density-response function remain at the same fre-
quency, and is generally hard to satisfy.

III. EXCITATION SPECTRA: LINEAR RESPONSE

The vast majority of applications of TDDFT today

lie in spectroscopy, to compute the linear response and
excitation energies of atoms, molecules, and solids.
TDDFT’s rise to fame is largely due to its unprecedented
balance between accuracy and efficiency for these ap-
plications. Figure 3 gives a pictorial representation of
a few of the most recent successes; these are chosen
to illustrate the wide range of applications in which
TDDFT today is giving useful results. I refer the reader
to Refs. [74–80] for recent reviews that give more detail
on the state-of-the-art performance for energies and ex-
cited state properties and instead here I highlight chal-
lenges for the field after first giving a brief outline of the
linear response TDDFT formalism.
Consider perturbing a system initially in its ground-
state Ψ0 with density n0 , with a weak field: vext (r, t) =
(0) (0)
vext (r) + δvext (r, t), where vext represents the field-free
unperturbed external potential, and δvext (r, t) the per-
turbation, turned on at t = 0. The Fourier trans-
form of the first-order response n(1) (r, t) to the δvext (r, t)
yields excitation energies and oscillator strengths of the
unperturbed system. Polarizabilities can be obtained
from the linear response of the dipole to electric fields
while higher-order response yields hyperpolarizabili-
ties. The density-density response function is defined
as χ[n0 ](r, r0 , t − t0 ) = δn(r, t)/δvext (r0 , t0 ), and, for finite
systems, has the Lehmann (or spectral) representation in
frequency-space:

X hΨ0 |n̂(r)|ΨI ihΨI |n̂(r0 )|Ψ0 i

χ(r, r0 , ω) = + c.c.(−ω) ,
ω − ΩI + i0+
I
(11)

7
 density when perturbed via the corresponding KS per-
turbation δvS (r, t), all response quantities can be directly
obtained from the KS system. The KS density-density
response function, χS [n0 ](r, r0 , t−t0 ) = δn(r, t)/δvS (r0 , t0 )
naturally has a Lehmann representation in frequency-
space for finite systems, but the poles lie at KS single-
excitations, i.e. orbital-energy differences between or-
bitals that are occupied and unoccupied in the ground-
state. Further, the residue of the pole ωia = a − i re-
duces to simple orbital product: φi (r)φa (r)φi (r0 )φa (r0 ).
A Dyson-like equation relates χS to the interacting χ [93,
94]: using ? to denoteRspatial convolution integrals such
as χS (ω) ? fHXC (ω) = d3 r1 χS (r, r1 , ω)fHXC (r1 , r0 , ω), we
have

χ(ω) = χS (ω) + χS (ω) ? fHXC (ω) ? χ(ω) (12)

This equation features the indispensable Hartree-xc ker-

nel, fHXC (ω) = fH + fXC (ω) where fH (r, r0 ) = 1/|r − r0 |
and

fXC [n0 ](r, r0 , t − t0 ) = δvXC [n](r, t)/δn(r0 , t0 )|n=n0 (13)

The kernel plays a crucial role in dressing the KS re-

sponse function, shifting the KS poles to the true exci-
tation energies, and mixing the KS oscillator strengths
to form the true oscillator strengths. With the ubiquitous
A
adiabatic approximation, since vXC [n](r, t) depends only
on n(t) at the present time, the right-hand-side of Eq. 13
FIG. 3. “Splash”-plot illustrating the range of TDDFT appli- becomes proportional to δ(t − t0 ) and the Fourier trans-
cations today for excitations and response: (a) In Ref. [81], form frequency-independent, fXC A
(r, r0 , ω) = fXC (r, r0 ).
the rate of yellowing of Leonardo da Vinci’s self-portrait be- Equation (12) is the basis of TDDFT linear response.
gan to be ascertained by comparing non-destructive optical There are evidently two ingredients: first the elements
reflectance measurements with TDDFT calculations of the rel- of the KS density-response function, involving all occu-
evant chromophoric spectra. The image here is from the pied and unoccupied KS orbitals living in the ground-
Wikimedia Commons page for Leonardo da Vinci [Public do- (0)
main]. (b) In Ref. [82], Published by the PCCP Owner So- state KS potential vS . Second, the xc kernel, which
cieties, the absorption spectrum of the complete chlorophyll introduces so-called dynamical effects. Technically, the
network of the light-harvesting complex from green plants, approximation used for the ground-state xc potential
was computed using TDDFT, yielding implications for in- should be linked to that used for the xc kernel since they
trinsic energy transfer pathways within the LHC-II complex. are generated from one and the same TD xc potential,
Reproduced under a Creative Commons Attribution 3.0 Un- (0)
vXC [n](r, t) = vXC (r) + δvXC . But in practise, the two ap-
ported Licence. (c) The bound exciton in the absorption spec- proximations are often treated independently.
trum of LiF was successfully captured by several approxi-
Typically in quantum chemistry codes, the equation is
mate functionals developed from quite different approaches
(see also text). Here the experiment is from Ref. [83], while re-cast in the form of matrix equations derived by Mark
the TDDFT results are from: the nanoquanta kernel [84], Casida, based on considering poles and residues of the
the dynamical long-range corrected kernel (dynLRC) [85], the frequency-dependent polarizability [95, 96]. The eigen-
bootstrap kernel on top of an exact-exchange OEP ground- values of the matrix
state (BO oep) [86], the jellium-with-gap kernel on top of the √ qq 0
scissors-corrected LDA ground-state (JGM lda sx) [87], the Rqq0 (ω) = ωq2 δqq0 + 4 ωq ωq0 fHXC (ω) , where (14)
guided iteration bootstrap kernel on top of scissors-corrected
qq 0
(ω) = d3 rd3 r0 φi (r)φa (r)fHXC (r, r0 , ω)φi (r0 )φa (r0 ),
R
LDA (GIBO lda sx) [88], screened exchange on top of scissors- fHXC
corrected LDA (SXX lda sx) [89], and the optimally-tuned lie at the squares of the excitation energies, and oscil-
range-separated hybrid (OTSRSH) [90]. (d) Table of errors lator strengths can be obtained from the eigenvectors.
from different functionals for 103 vertical excitation energies
Here q = (i, a) represents a double-index, with i la-
of singlet valence excitations in 28 small and medium-size or-
ganic molecules from data of Refs. [75, 91], as compiled in
belling an occupied orbital and a an unoccupied one.
Ref. [92]. Other equivalent derivations and formulations appear
in Refs. [97, 98]; note that the derivation in Ref. [97] is
valid only when an adiabatic approximation is assumed
Now, because the TD KS system gives the same TD from the outset.

8
 The matrix formulation is only valid for discrete spec- brids produce the requisite −1/r: the essential idea is to
tra, so used mostly for atoms and molecules, while separate the Coulomb repulsion using a parameter γ as,
Eq. (12) is usually solved when dealing with continuous e.g.,
spectra as in solids.
To obtain the continuous part of spectra of atoms and 1 1 − erf (γr) erf (γr)
= + (15)
molecules, in particular resonance widths and positions, r r r
the Sternheimer approach is often used [99–101]. But where the first term dominates at short-range and is
even for bound excitations, the Sternheimer approach, treated via a semi-local functional, and the second
also known as density functional perturbation theory, or term dominates at longer ranges and is treated via ex-
coupled perturbed KS, can be more efficient, if just a few act (Hartree-Fock) exchange. The result yields a self-
low-lying excitations are of interest; it avoids solving for interaction-free description at large electron-electron
many unoccupied states by considering instead pertur- separations, while retaining the balanced description of
bations of the occupied KS orbitals in frequency-space. exchange and correlation at short range.
Response can also be obtained from real-time propaga- Double excitations It was recognized soon after the
tion of occupied orbitals [102] under a weak perturba- derivation of the linear response equations, that states
tion; often a δ-kick is applied so to uniformly stimulate of double-excitation character were missing from the
all excitations, and the system is then evolved freely for approximate TDDFT spectra; this time the fault lies in
a time long enough such that Fourier transform of the the adiabatic approximation to the kernel [105, 120]. It
dipole yields well-enough resolved peaks at the reso- is clear they are absent in the non-interacting response
nant frequencies. This is, for example, how the absorp- function χS : considering Slater determinants in the nu-
tion spectrum of such a large system as that in Fig 3b merator of Eq. (11), doubly-excited KS states give a zero
could be computed [82]. contribution to χS because the matrix elements of the
density-operator vanish between Slater determinants
that differ by more than one orbital due to its one-body
A. Challenging Excitations property. This is just the mathematical statement of the
fact that exciting two electrons in a non-interacting sys-
The successes of linear response TDDFT have been tem requires two photons, so is beyond linear response.
celebrated in a number of reviews cited earlier. Its ef- Excited states, as well as the ground-state, of the inter-
ficiency and overall reasonable reliability have led to it acting system, on the other hand, are superpositions of
becoming an easy-to-use “go to” method for simulating single, double, and higher-electron-number excitations,
spectra that, for other methods of comparable accuracy, and so the numerator hΨ0 |n̂(r)|ΨI i remains finite from
are not easily computationally accessible. However, one contributions between components of the two many-
must be cautious in treating it completely as a black-box: body wavefunctions that differ by one orbital. A sim-
there are some notorious problems where either a care- ple number-counting argument [121, 122] shows that
ful choice of functional is needed or, where the results no adiabatic kernel can introduce doubly-excited con-
with currently available functional approximations are figurations. The strong frequency-dependence of the
simply unacceptable. We discuss some of these here. exact kernel near a state of double-excitation charac-
Rydberg excitations Here the problem lies primarily ter was deduced [121, 123] by in effect inverting the
with the ground-state approximation, rather than the Dyson equation Eq (12) for a model where one KS sin-
kernel: Because LDA and GGA ground-state KS po- gle and one double excitation interact, from which a
tentials depend locally or semilocally on the density, practical approximation was proposed. More rigorous
they asymptotically approach zero exponentially with derivations using many-body theory subsequently es-
r, much faster than the −1/r fall-off of the exact KS sentially verified this form [124–127] (see Ref. [78] for
potential. Without the −1/r tail, they cannot gener- a recent review), and the approximate kernel, dubbed
ate a Rydberg series. The frequencies of the highest- “dressed LR-TDDFT”, has been implemented in some
lying excitations that are bound within these approxi- codes (Niedoida and deMon2k), enabling extensive
mations tend to be underestimated, while even higher tests [125, 128, 129]. In particular, Ref. [125] found that
excitations are absent, with their oscillator strengths be- the non-adiabatic dressing has best performance when
ing squeezed into the continuum albeit with accurate used in conjunction with a global hybrid functional.
optical intensity [103–107]. Different solutions include Long-range charge-transfer excitations As illustrated in
asymptotically-corrected functionals [104, 106, 108], Eqs. (12)–(14), TDDFT excitations are obtained from
exact-exchange methods [109, 110], and hybrids [111– shifting and mixing single-excitations of the KS system
113], including range-separated hybrids [114–116]. Con- via matrix elements of fXC , often referred to as dynam-
ventional hybrid functionals, formally justified by gen- ical corrections. When there is little spatial overlap be-
eralized KS theory [117–119], mix in a fraction CX of tween the occupied and unoccupied orbitals relevant in
Hartree-Fock exchange into otherwise semi-local func- the transition, the dynamical corrections from conven-
tionals, which means their potentials decay asymptoti- tional xc-kernels vanish, and the TDDFT excitation en-
cally as −CX /r. On the other hand, range-separated hy- ergy reduces to KS orbital energy difference. Excita-

9
tions of charge-transfer character over long range are shell fragments [142, 143]. At the root of the diffi-
a notorious example of this, and was first brought to culty in this case is the KS ground-state description: the
light in Ref. [112]. In the large separation limit, the HOMO (and LUMO) straddle both fragments, delocal-
exact lowest charge-transfer excitation energy tends to ized over the whole molecule, quite distinct from the
exact
ωCT → I D − AA − 1/R, where I D is the ionization Heitler-London-like nature of the physical wavefunc-
energy of the donor, AA is the electron affinity of the tion. Although in this case the HOMO and LUMO have
acceptor, and −1/R is the first electrostatic correction finite overlap, their orbital energy difference tends to
between the species after the charge-transfer. As ana- zero as the molecule stretches (for either the exact func-
lyzed in Refs. [130–133] however, TDDFT with LDA or tional or approximations), a sign of static correlation in
GGA yields the orbital energy difference between the the KS system, and fHXC must be responsible for the en-
LUMO of the acceptor and the HOMO of the donor, tire charge-transfer energy. It can be shown that the ex-
L (A) − H (D). With approximate ground-state func- act kernel not only diverges with interatomic separation
tionals H underestimates the true ionization energy, but but also is strongly frequency-dependent.
even with the exact ground-state functional L lacks re- Conical intersections Photoinduced dynamics is heav-
laxation contributions to the electron affinity, and fur- ily influenced by conical intersections, seams of exact
ther the −1/R tail is missing. A useful diagnostic of degeneracy between electronic potential energy surfaces
the performance of a given functional based on con- that are ubiquitous in the landscape of nuclear config-
sidering the overlaps was given in Ref. [115]. Signif- urations. They funnel nuclear wavepackets between
icant effort has been devoted to solving this problem, surfaces, providing a rapid pathway for non-radiative
e.g. Refs. [113, 116, 130, 134–139], especially because decay, and even affect nuclear trajectories that encircle
charge-transfer excitations play an important role in the intersection at large distances, by generating a geo-
many topical applications today that involve systems metric phase. Unfortunately, current approximations in
large enough that few practical alternatives to TDDFT TDDFT typically provide a poor description of the sur-
exist, e.g. biomolecules, photovoltaics, conductance. faces near conical intersections with the ground-state. In
Most modify the ground-state functional to correct the some cases the shape is dramatically exaggerated (e.g.
underestimation of the ionization potential, and mix H3 ), while in other cases, the dimensionality of the seam
in some degree of Hartree-Fock exchange, especially is wrong (e.g. H2 0) [144]. There is trouble here for
via range-separated hybrids [117–119], since they dis- both the ground-state approximation and the xc ker-
play the −1/R asymptotic behavior. The amount of nel approximation, and, conceptually, the issues are not
long-range exchange needed to give reasonable answers unlike the case of charge-transfer excitations between
varies widely depending on the system, which moti- open-shell fragments. Static correlation resulting from
vated the development of the non-empirical “optimally- nearly degenerate determinants near the conical inter-
tuned” range-separated hybrids of Refs. [137, 138]. The section creates unusual features in the exact ground-
range-separation parameter is tuned so that H is as state xc functional in order to compensate for the KS
close as possible to the negative ionization potential Slater determinant description being so far from the true
computed by the same functional from total energy correlated ground-state. The xc kernel then must be
differences, typically for both the neutral and anionic strongly frequency-dependent because of the double-
system. Size-consistency violation and other prob- excitation to the near-lying determinant. In practise,
lems [140] arising from system-dependence of these the very small HOMO-LUMO orbital energy difference
functionals means caution must be applied for general leads to an instability in the solution of the LR equa-
usage; the range-separation parameter can have very tions, that can be tamed by making the Tamm-Dancoff
different values for the individual subsystems and the approximation [145]. In fact, in Ref. [145], where ring-
combined system, and varies along the internuclear sep- opening in oxirane was studied, it was argued that al-
aration, leading to poor binding energies and sometimes though TDDFT gets the dimension of seam wrong, giv-
erratic potential energy surfaces [140]. ing slightly interpenetrating cones rather than touch-
The xc-kernel in pure (i.e. not generalized) KS DFT ing ones, that the resulting dynamics (using surface-
must diverge exponentially with internuclear separa- hopping) still yielded useful results regarding reaction
tion [133], and it has been argued that discontinu- pathways, although rate constants were likely not accu-
ities of the xc-kernel with respect to particle number rate. Also, spin-flip TDDFT gets around the problem of
play a crucial role [141] whenever the KS orbital over- the ground-state degeneracy by “viewing” the conical
lap is too small. The exact-exchange approximation intersection from a high-spin reference state [146].
(within OEP) captures these features when the charge- Optical response of solids Simple ALDA accurately cap-
transfer occurs between closed-shell fragments, when tures plasmon dispersion curves in simple metals, with
used in its fully non-adiabatic form [139, 141]. The self- the xc kernel providing an improved description of the
interaction-corrected LDA, applied within a generalized plasmon dispersion and linewidth compared to the ran-
OEP scheme, has also been shown to work [69]. dom phase approximation that uses the Hartree ker-
However, one case where none of the above sugges- nel only [147]. For non-metallic systems, ALDA pro-
tions works is that of charge-transfer between open- vides a good description of electron energy loss spec-

10
tra (EELS) [80, 148], and for the optical response at finite free broadening of the spectra [161]. In yet another di-
photon wavevector q [149], but does a poor job for the rection, a kernel built on a jellium-with-gap system [87]
spectrum at vanishing q. There are two main problems: has been developed. Meta-GGA functionals also dis-
one is that the optical gap, defined by the onset of con- play the requisite α/q 2 behavior [162], as do hybrid
tinuous absorption, can be underestimated by as much functionals [163], and optimally-tuned screened range-
as 30-50%, and the second is that the excitonic structure separated hybrids have recently been applied [90]. Re-
is missing from the spectrum. For both of these prob- latedly, screened exact-exchange [89], which formally
lems, a long-ranged xc kernel is necessary [50, 80, 150– falls into generalized KS theory, has been proposed as
152]: the xc kernel must have a long-wavelength be- a bridge between TDDFT and BSE; it is computationally
havior of 1/q 2 as q → 0. In fact, this behavior can much cheaper than BSE yet of comparable accuracy. Fi-
be deduced from exact conditions satisfied by the xc nally, Ref. [164, 165] showed how to directly obtain the
kernel, such as the zero-force theorem, that inextrica- excitonic binding energies instead of the full spectra by
bly link nonlocality in time and space [50, 153]. ALDA, adapting the Casida-equation formalism.
on the other hand, being local in space, is constant in Molecular Conductance An area of great interest is in
q. Further, to open the gap, the xc kernel must have simulating charge transport at the nanoscale. Most ab
a non-vanishing imaginary part [12, 50] which, due to initio descriptions are in the steady-state using DFT
Kramers-Kronig relations, implies that it must have a within a Landauer formalism, but give poor agreement
frequency-dependence, i.e. be non-adiabatic. Instead, with experiment due to problems with both the ground-
today a common workaround is to use LDA bandstruc- state xc potential and the dynamical fXC corrections. Re-
ture with a scissors operator [154, 155]. garding the former, the lack of derivative discontinu-
A non-adiabatic kernel is not necessary to capture the ity [166] in LDA and GGA leads to a large overestimate
excitonic peak, but the 1/q 2 behavior is essential [150]. of conductance [167, 168] and artificial level broadening,
(The excitonic Rydberg series however does appear to and is intimately related to Coulomb blockade, a fully
require a frequency-dependent kernel [156]). Interest- dynamical picture of which was provided in Ref. [169].
ingly, although 1/q 2 is the Fourier-transform of −1/r, The discontinuity also needs to be accounted for in the
this required 1/q 2 -dependence of the kernel is not a con- xc kernel, where, in addition, viscous effects can be im-
sequence of the long-rangedness of the Coulomb inter- portant [56, 170]. An exactly solvable impurity model
action, and it has been dubbed an “ultranonlocal” effect has demonstrated the importance of long-range non-
since it is even present in systems arbitrarily close to the adiabatic terms in the xc potential [171]. The exact the-
homogeneous liquid [50, 153]. oretical framework for transport in TDDFT for strong
Much progress has been made to correct this defi- and weak bias, requiring a real-time description beyond
ciency of ALDA, along several different avenues, as evi- Landauer, was formulated in Refs. [172, 173]. To in-
denced by the success in Fig. 3c of capturing the strongly clude dissipation and decoherence induced by the envi-
bound exciton in LiF from a variety of different mod- ronment, several approaches to extend TDDFT to open
ern functional approximations. Exact exchange initially quantum-systems were developed [174–178], relevant
seemed to give promising results [157] until it was real- both in and out of the transport context.
ized that the calculations only worked because of an ar- The lack of derivative discontinuity in (semi-)local
tificial screening induced by a long-wavelength cut-off functionals is sometimes called the “gap problem”,
in the numerics [158]. Kernels derived from the Bethe- however it is worth realizing that several distinct gaps
Salpeter equation, such as the nanoquanta kernel, have can be defined, and functionals have no problem with
seen much success [84, 150, 159]. The long-range cor- some of these. The KS HOMO-LUMO energy differ-
rected (LRC) kernel [85, 150, 160], α/q 2 , with α deter- ence approximates neutral excitations such that, once the
mined empirically from the static dielectric constant of TDDFT fXC corrections are applied, one obtains the low-
the crystal, can be viewed as a simple scalar approxi- est optical excitation, or optical gap; (semi-)local func-
mation to the nanoquanta kernel; a dynamical version tionals give pretty accurate results for molecules. The
uses also the plasma frequency. The “bootstrap” kernel fundamental gap, on the other hand, I−A, involves elec-
proposed in Ref. [86] uses instead self-consistent itera- tron addition and removal, and in particular L should
tions of Eq. (12) with an approximate expression for the not be interpreted as an affinity level. When a fraction
(static) xc kernel in terms of the dielectric function; go- of electron is added to a neutral system, the exact KS
ing beyond the scalar approximation, it includes local- potential jumps up by a positive constant equal to the
field-effects. A guided-iteration version has been pre- derivative discontinuity, which then shifts the orbital en-
sented in Ref. [88] where the sensitivity of the boot- ergy up to the affinity level [166]; such a shift is not cap-
strap approach to details such as choice of ground- tured by semi-local functional approximations that are
state orbitals and energies from which to build χS (usu- smooth functions of particle number [179, 180]. (That
ally scissors-shifted LDA, sometimes instead with GW- it is in fact possible to extract derivative discontinuities
corrected energies, sometimes OEP), and treatment of from LDA and GGA has been pointed out in Refs. [181,
the local field effects, has been stressed. The VK approxi- 182], but this is not how these functionals are usu-
mation has recently been shown to provide a parameter- ally used.) Because Hartree-Fock virtual orbitals rep-

11
resent addition-energy levels rather than neutral excita- relation effects missed in the single-active-electron ap-
tions, hybrid functionals evaluated within the general- proximation. Because the electrons make large excur-
ized Kohn-Sham scheme effectively open the gap, espe- sions away from the nuclei, exact-exchange or self-
cially the range-separated ones [183]. Lack of the discon- interaction-corrected-LDA are often used, implemented
tinuity is partly responsible for the so-called level align- within a KLI approximation to the OEP [193]. TDDFT
ment problems of standard functionals at molecule- revealed interesting interference effects between con-
metal, molecule-semiconductor, and molecule-molecule tributions from different molecular orbitals in high-
interfaces in applications relevant to transport as well harmonic generation and multiphoton ionization in
as to photovoltaic modeling and to photocatalysis [184– molecules, e.g. Refs. [194–197], where, additionally,
187]. For metal-molecule interfaces non-local polariza- orientation effects have been studied. Time-resolved
tion effects also contribute significantly to the renormal- Coulomb explosion imaging of molecular torsion dy-
ization of the frontier energy levels, for which “image- namics in complex molecules has been shown to give
charge” models have been helpful to capture [184, 185]. results consistent with experiment [198]. TDDFT has
begun to elucidate the mechanism of charge-resonance
enhanced-ionization in triatomic molecules [199], iden-
IV. FULLY NON-LINEAR DYNAMICS tifying which molecular orbitals are responsible; to-
gether with Ehrenfest dynamics for the nuclei, en-
Notwithstanding the successes of TDDFT for linear hanced ionization and orientation dependence has been
response, TDDFT has particular relevance in the regime analyzed in acetylene and ethylene [200]. The time-
of non-perturbative time-resolved dynamics. There is dependent electron localization function introduced in
really no alternative practical method for accurately de- Ref. [201] has been used to understand time-resolved
scribing correlated electron dynamics when more than a dynamics of chemical bonds in scattering and excita-
few electrons are involved, yet many phenomena which tion processes. Recently attosecond “self-streaking” has
are fascinating in themselves and also have technologi- been proposed and demonstrated on the endohedral
cal applications, lie in this regime. Although no strong fullerene Ne@C60 , from which one can directly observe
fields are present during simulations of photoinduced electronic density oscillations in real-time [202]. TDDFT
coupled electron-ion dynamics (e.g. in light-harvesting has facilitated the analysis of photoelectron spectra and
and artificial photosynthesis) and other photo-chemical angular distributions in molecules and clusters after ir-
or -physical processes, the electrons do not remain close radiation by strong pulses; a review can be found in
to any ground state, so also lie beyond the perturbative Ref. [203].
regime. Optimal control theory has been formulated with
TDDFT has provided useful predictions for a vari- TDDFT [204], and used already to obtain desired tar-
ety of strong-field processes and below I select some to gets in an array of different scenarios, from selectively
briefly describe. The examples are chosen to illustrate cleaving a molecular bond [205] to enhancing a partic-
the wide range of applications of TDDFT in this regime, ular harmonic in high-harmonic generation [206]. Very
and emphasise recent achievements over older ones; in- recently, the possibility for control of spin-orbit medi-
evitably such a short description is a little random. ated ultrafast demagnetization via lasers in solids has
been demonstrated [207].
In solids, coherent phonon generation produced by
intense laser pulses has been simulated [188], capturing Alongside this rosy picture, however, studies on small
the transition from the impulsively stimulated Raman molecules where numerically exact or high-level wave-
scattering mechanism to the displacive excitation mech- function methods can be applied, have shown that ap-
anism and the associated phase change of the generated proximate TDDFT functionals can yield significant er-
phonon. Dielectric breakdown in crystalline silicon has rors in their predictions of the dynamics, sometimes
been studied, where the importance of including both even failing qualitatively. We next discuss these prob-
electric and magnetic fields self-consistently has been lems.
stressed; the authors developed a multiscale model solv-
ing Maxwell’s equations alongside TDKS [189]. The
same method showed that TDDFT calculations of laser- A. Challenges
induced ablation by very short, intense laser pulses
in α-quartz gives reasonable agreement with experi- The problems fall into two main categories: the inac-
ment [190]. TDDFT has elucidated the possible mecha- curacy of the approximate xc potential functional, and
nisms of the experimentally-observed laser-induced ul- the need for additional functionals to extract observ-
trafast demagnetization and details of spin-dynamics in ables of interest from the KS system. For the first, we fo-
bulk ferromagnets in Ref. [191]. cus here mostly on problems associated with making the
Multiphoton processes have been studied in atoms adiabatic approximation, rather than the specific choice
and molecules, with the first calculations of high- of ground-state functional that goes into such an ap-
harmonic generation spectra and ionization rates in proximation. Exactly-solvable model systems are useful
atoms emerging in the mid-1990’s [192] capturing cor- to explore the impact of these problems, and to compare

12
approximate functionals, especially the AE, with the ex-
act one.
Memory appears to generally play a more important
role in non-perturbative time-resolved dynamics com-
pared to the linear response regime [52, 59, 63, 65, 68, 73,
208–212]. The problems appear to require a memory-
dependence beyond what is contained in the presently-
available orbital-functionals. To investigate the effects
of the adiabatic approximation, comparison of exact xc
functionals with the AE of Eq. 10 has proven quite in-
structive [59, 64–66, 211, 213–218]. It was shown that
FIG. 4. The xc potential (black), plotted at a quarter of a Rabi
for strong field ionization, the xc potential appears not period, in a model He atom driven by a weak resonant field
to have significant non-adiabatic dependence in some that causes Rabi oscillations between the ground and first-
cases, but this depends on the details of the field. Some- excited state [59]. The two soft-Coulomb √ interacting electrons
times the observables of interest are not so much af- live in the potential vext (x, t) = −2/ x2 + 1 + Axsin(ωt)
fected by non-adiabatic features present in the xc po- where A = 0.0067a.u., and ω = 0.533a.u., resonant with the
tential, e.g. integrating over regions of space to find to- lowest singlet excitation. The density is shown in red, while
tal ionization probabilites can wash over differences in the AE potential is in blue. Linked here (Multimedia view)
the density variations. In fact, the famous knee in the is a movie, directed by Peter Elliott, showing the xc potential
double-ionization curve that signals a switch from a cor- evolve in time, over the duration of about one Rabi period,
related to sequential ionization mechanism, can be cap- TR = 854.86au.
tured by a non-adiabatic approximation [219] provided
it has a derivative discontinuity.
But in the general case, it appears that neglect of thing, but having them continuously drift in time is
memory-dependence in the xc-potential can lead to more serious.
large errors, and part of the problem appears to come Turning now to the second category of problem: the
from the lack of dynamical step features that the exact xc need for “observable-functionals” when the observable
functional possesses, that have a non-adiabatic depen- of interest is not directly related to the density. Al-
dence on the density [59, 60, 64, 211, 220]. In the dynam- though in principle all properties of interest can be ex-
ics of charge-transfer beginning in the ground-state, the tracted from the KS system, it is not generally known
lack of these features have a drastic effect [211, 217, 221– how to extract them. Only in some cases can one use
223]; even the AE fails to transfer charge even though the usual quantum mechanical operators acting directly
its predictions for the charge-transfer excitations them- on the KS wavefunction, e.g. high-harmonic genera-
selves can be accurate [213, 214]. The non-adiabatic tion spectra are measured from the dipole moment of
step structures have been shown to appear quite gener- the system. But for general observables, e.g. cross-
ically in all sorts of dynamics, and are not associated to sections in atomic collisions [229], ionization probabil-
ionization or fractional charge as other observed time- ities [230–233], or momentum-distributions [231], sub-
dependent steps are [141, 169, 219]. They appear in the stituting the KS wavefunction into the usual quantum
presence or absence of a field, for few electrons [59] or mechanical operator typically gives poor results. For ex-
many electrons [60], whether one uses a Slater determi- ample, in ion-recoil upon ionization of a model He atom,
nant or not [31] for the KS system (see also Fig. 2). Un- the KS momentum distributions were found to be dras-
like step features that appear in ground-state potentials, tically wrong, displaying a single maximum instead of
they are not a signature of static correlation, and their the characteristic two-hump structure, and with a signif-
relationship with time-dependent natural orbital occu- icantly overestimated magnitude [231]. More sophisti-
pation numbers is non-trivial [64]. The AE potential cated functionals have been developed for a few partic-
completely lacks these steps, as illustrated in Figure 4 ular cases; momentum-densities for the above problem
(Multimedia view) which, together with a movie, plots in Ref. [231], and double-ionization probabilities, that
the xc potential in a model He atom driven by a weak “lower” the knee, in Ref. [234].
resonant field that causes Rabi oscillations between the The one-body nature of the KS Hamiltonian is some-
ground and first-excited state. what of a double-edged sword in the non-perturbative
Violation of the resonance condition mentioned at regime. On the one hand, it makes impressive calcula-
the end of Sec. II can wreak havoc and confusion tions on large systems doable, but on the other hand, an
in analysing time-resolved spectra as is now exper- initial Slater determinant remains a Slater determinant
imentally accessible via ultrafast pump-probe spec- throughout the evolution, even when the interacting
troscopy [224, 225]. Several researchers had noted the wavefunction significantly changes its degree of correla-
peaks in their spectra would unphysically shift when tion [235]. Even if initially the physical system is weakly
evolving far from the ground-state [221–223, 226–228]; correlated, and adiabatic approximations with the usual
having peak positions that are somewhat off are one ground-state approximations initially work well, over

13
time excited states get populated with the degree of cor- (although approximate methods that use the bare KS
relation possibly significantly changing, as reflected in surfaces have also been used [243], with notable applica-
the natural orbital occupation numbers. If one perfects tions in photovoltaics [244]), and make jumps between
a time-dependent non-adiabatic functional approxima- them based on Tully’s fewest-switches surface-hopping
tion (for either the xc functional or an observable) to algorithm [245]. Such calculations have shed light into
work in the weakly-correlated regime, it will not work the photochemical reactions going on in vitamin D syn-
well for general dynamics. thesis [246], for example, and combined with local con-
trol theory to design pulses to enhance proton trans-
fer reaction in 4-hydroxyacridine [247]. The algorithm
B. Coupled electron-ion dynamics requires non-adiabatic couplings (NACs) between sur-
faces, and it has been shown that these are rigorously
Dynamics of electrons in molecules usually cannot be available from linear response TDDFT for NACs involv-
considered in isolation from the dynamics of the ions ing the ground-state [248–252]. But NACs between ex-
to which they are coupled and the phenomena aris- cited states, which are clearly most relevant for photo-
ing from the correlated dynamics of electrons and ions induced dynamics, where typically the simulation starts
even without worrying about electron-electron interac- in an electronic excited state, are strictly not available
tion themselves form a very rich field. In simulating from linear response. They can be obtained instead
photo-induced processes, it is often the case that the nu- from quadratic response theory, however it was recently
clear motion drives the electronic dynamics after the ini- shown that the adiabatic approximation yields unphys-
tial excitation. ical divergences in the couplings whenever the energy-
Typically, although not always, a classical treatment difference between two excited states equals another ex-
of nuclear motion is coupled to the electronic dynam- citation energy [253–256].
ics. The most common methods are Ehrenfest dynamics Besides this, the accuracy of these methods are clearly
and trajectory surface-hopping, and here we restrict our limited by the accuracy of the forces obtained from the
discussion to an analysis of their use specifically in con- TDDFT surfaces, so the problems of Sec III enter. In
junction with TDDFT, rather than giving a more exten- particular, the dynamics probes nuclear configurations
sive discussion of their relative merits and pitfalls. away from equilibrium, so even in situations where
It is well known that the mean-field nature of Ehren- these problems are not so important at equilibrium con-
fest dynamics is inadequate when branching of nuclear figurations, they can plague the dynamics, e.g. the
trajectories occurs, e.g. in electron-transfer reactions, re- case has been made that without adequate treatment of
actions near metal surfaces, and photochemistry gener- doubly-excited states, global potential energy surfaces
ally speaking. However, as a formalism, Ehrenfest is are unlikely to be accurate [144].
well-suited to be used in conjunction with TDDFT: the
classical force on the Ith nucleus is obtained via
V. OUTLOOK
M R̈I (t) = −hΨ(t)|∇I Hel |Ψ(t)i − ∇R Vnn (16)
From some of the discussion above, the features of the
where Vnn is the classical nuclear-nuclear potential, and exact xc functional and the subtleties of extracting ob-
Hel is the Hamiltonian from Eq. (1) with R appearing servables, one might be tempted to conclude that some
in vext evaluated at the instantaneous nuclear positions. magic is required to make the ingredients and goals of
The only term that the gradient in Eq. (16) acts on is the TDDFT get along, and that TDDFT actually stands for
PNn PN
electron-nuclear attraction Ven = − I=1 i=1 ZI /|ri − Thaumaturge-Dependent DysFunctional Theory. But
RI | (where Nn is the number of nuclei and ZI one should bear in mind that TDDFT is a relatively
their charge), and so the first term, the quantum young field; indeed the success it has had already is re-
“backreaction” on the nuclear
R 3 trajectories, becomes markable. The variety and size of the systems studied
−hΨ(t)|∇I Hel |Ψ(t)i = d rn(r, t)∇I Ven ({ri }, {RI }). these days in TDDFT is impressive, as both Fig. 3 and
Ehrenfest therefore needs nothing but the electronic the non-linear calculations described at the beginning
density, which can be directly obtained from the TD of Sec. IV suggest. Its impact in materials science and
KS evolution. Notably, Ehrenfest was recently used quantum chemistry is incontrovertible, and evidenced
to demonstrate the relevance of coherent coupled elec- by its increasing use in an ever-widening range of appli-
tronic and vibronic motion in driving the charge- cations, in both the response regime and the fully non-
transfer in a prototypical light-harvesting system [236, perturbative regime.
237]. Nevertheless, the results on the model systems where
Surface-hopping on the other hand, is able to capture exact potentials are available more often than not sug-
wavepacket branching but its marriage with TDDFT is a gest that, in the non-perturbative regime, the currently
little more stormy due to its need for non-adiabatic cou- used functional approximations may have significant er-
plings. Trajectories evolve on Born-Oppenheimer sur- rors. The model systems do probe perhaps the worst-
faces obtained from TDDFT linear response [238–242] case scenarios for approximate TDDFT, and the system-

14
size scaling of the xc features missing in approxima- imations. The expression for the exact xc potential
tions should be explored. Characterizing when and in Eq. (8), approximated to yield an orbital-dependent
why the current approximations, particularly the adia- functional, may be a useful starting point.
batic ones, work in a given situation remains an open Alongside new functional approaches, I have no
problem today, but an important one in establishing doubt that the applications of TDDFT will continue to
the reliability of the results in different situations. In grow, taking on ever-larger systems possibly with the
the linear response regime, such characterization has help of embedding- and fragment-based methods [258–
greatly improved the predictability of TDDFT for exci- 262], and adventuring into new regimes such as the
tation spectra, and our understanding of when the re- recently-introduced quantum electrodynamical density
sults can be trusted. The adiabatic approximation triv- functional theory that includes coupling to quantized
ially satisfies many of the exact conditions, some of radiation fields [263–266], and to quantum ions in a mul-
which were mentioned in Section II, which might be be- ticomponent formalism [267]. It will be exciting to see
hind the unexpected fact that an approximation based where TDDFT will be in another thirty years.
on ground-states can give useful results for problems
far beyond the ground-state, as the applications partic-
ularly in Sec. IV have shown.
New developments for functionals might take advan-
ACKNOWLEDGMENTS
tage of the initial-state dependence of the Runge-Gross
theorem. For a given initial state of the physical interact-
ing system, there is an infinite choice of allowed KS ini- I warmly thank Johanna Fuks for generating the po-
tial state, and whether one can make a judicious choice tentials for Fig. 2 and for help with some of the plots,
for the best one to use when stuck with an adiabatic xc and Soeren Nielsen and Michael Ruggenthaler for the
functional has only begun to be explored [29, 31, 215]. use of their global fixed point iteration code used to find
Even in the linear response regime, using excited states the potentials in Fig 2. Much appreciation goes to Peter
as alternative references around which perturbations Elliott for making the movie linked to Fig. 4. Part (c) of
yield excitation frequencies could be very useful; for Fig. 3 used data generously sent to me by Carsten Ull-
example, a double-excitation is a single-excitation of rich, Francesco Sottile, Silvana Botti, Sangeeta Sharma,
a singly-excited state, so the need for non-adiabaticity Santiago Rigamonti, Andrea Marini, Fabio Della Salla,
could be avoided. Conical intersections involving the Valerio Olevano, Sivan Refaely-Abramson, Leeor Kro-
ground-state would appear simply as a degenerate de- nik, and their co-authors. I thank Dmitrij Rappoport and
excitation from an excited state. Already, calculations of Jürg Hutter for allowing me to use the data they com-
excited state absorption in molecules have appeared by piled in the table in Fig. 3(d). Financial support from the
kicking excited-state densities obtained from linear re- National Science Foundation CHE-1152784, and Depart-
sponse [257]. ment of Energy, Office of Basic Energy Sciences, Divi-
It may be time to consider cutting the apron strings sion of Chemical Sciences, Geosciences and Biosciences
from ground-state DFT, and try starting points other under Award DE-SC0008623 and DE-SC0015344, over
than the adiabatic approximation for building approx- the past few years, are gratefully acknowledged.

[1] P. Hohenberg and W. Kohn, Phys. Rev. 136, B864 (1964). dependent density functional theory, edited by M. A. Mar-
[2] E. Runge and E. K. U. Gross, Phys. Rev. Lett. 52, 997 ques, N. T. Maitra, F. M. Nogueira, E. K. Gross, and
(1984). A. Rubio, pp. 53–97.
[3] F. Bloch, Zeitschrift für Physik 81, 363 (1933). [13] C. A. Ullrich, Time-dependent density-functional theory:
[4] S. Lundqvist, in Theory of the Inhomogeneous Electron Gas, concepts and applications (Oxford University Press, 2011).
edited by S. Lundqvist and N. H. March (Springer US, [14] J. E. Harriman, Phys. Rev. A 24, 680 (1981).
Boston, MA, 1983) pp. 149–188. [15] G. Zumbach and K. Maschke, Phys. Rev. A 28, 544 (1983).
[5] T. Ando, Zeitschrift für Physik B Condens. Matter 26, 263 [16] N. T. Maitra, K. Burke, and C. Woodward, Phys. Rev.
(1977). Lett. 89, 023002 (2002).
[6] T. Ando, Solid State Communications 21, 133 (1977). [17] N. T. Maitra, in Fundamentals of time-dependent density
[7] V. Peuckert, J. Phys. C: Solid State Phys. 11, 4945 (1978). functional theory, edited by M. A. Marques, N. T. Maitra,
[8] B. M. Deb and S. K. Ghosh, J. Chem. Phys. 77 (1982). F. M. Nogueira, E. K. Gross, and A. Rubio, pp. 167–184.
[9] L. J. Bartolotti, Phys. Rev. A 24, 1661 (1981). [18] E. K. U. Gross, J. F. Dobson, and M. Petersilka, in Den-
[10] L. J. Bartolotti, Phys. Rev. A 26, 2243 (1982). sity Functional Theory II: Relativistic and Time Dependent
[11] J. F. Dobson, in Fundamentals of Time-Dependent Density Extensions, edited by R. F. Nalewajski (Springer Berlin
Functional Theory, edited by M. A. Marques, N. T. Maitra, Heidelberg, Berlin, Heidelberg, 1996) pp. 81–172.
F. M. Nogueira, E. Gross, and A. Rubio (Springer Berlin [19] R. van Leeuwen, Phys. Rev. Lett. 80, 1280 (1998).
Heidelberg, Berlin, Heidelberg, 2012) pp. 417–441. [20] S. Mukamel, Phys. Rev. A 71, 024503 (2005).
[12] E. K. Gross and N. T. Maitra, in Fundamentals of time- [21] G. Vignale, Phys. Rev. A 77, 062511 (2008).

15
[22] R. van Leeuwen, Phys. Rev. Lett. 82, 3863 (1999). [58] J. A. Berger, P. L. de Boeij, and R. van Leeuwen, Phys.
[23] T. K. Ng and K. S. Singwi, Phys. Rev. Lett. 59, 2627 (1987). Rev. B 75, 035116 (2007).
[24] R. van Leeuwen, Int. J. Mod. Phys. B 15, 1969 (2001). [59] P. Elliott, J. I. Fuks, A. Rubio, and N. T. Maitra, Phys.
[25] N. T. Maitra, T. N. Todorov, C. Woodward, and K. Burke, Rev. Lett. 109, 266404 (2012).
Phys. Rev. A 81, 042525 (2010). [60] J. Ramsden and R. Godby, Phys. Rev. Lett. 109, 036402
[26] M. Ruggenthaler and R. van Leeuwen, Europhys. Lett. (2012).
95, 13001 (2011). [61] N. T. Maitra, R. van Leeuwen, and K. Burke, Phys. Rev.
[27] M. Ruggenthaler, K. Giesbertz, M. Penz, and R. van A 78, 056501 (2008).
Leeuwen, Phys. Rev. A 85, 052504 (2012). [62] M. Ruggenthaler and D. Bauer, Phys. Rev. A 80, 052502
[28] M. Penz and M. Ruggenthaler, J. Phys. A: Mathematical (2009).
and Theoretical 44, 335208 (2011). [63] N. T. Maitra and K. Burke, Phys. Rev. A 63, 042501 (2001).
[29] S. Nielsen, M. Ruggenthaler, and van Leeuwen, Euro- [64] K. Luo, J. I. Fuks, E. D. Sandoval, P. Elliott, and N. T.
phys. Lett. 101, 33001 (2013). Maitra, J. Chem. Phys 140, 18A515 (2014).
[30] M. Ruggenthaler, S. E. Nielsen, and R. Van Leeuwen, [65] P. Hessler, N. T. Maitra, and K. Burke, J. Chem. Phys.
Phys. Rev. A 88, 022512 (2013). 117 (2002).
[31] J. I. Fuks, S. Nielsen, M. Ruggenthaler, and N. Maitra, [66] M. Thiele, E. K. U. Gross, and S. Kümmel, Phys. Rev.
Phys. Chem. Chem. Phys. , (2016). Lett. 100, 153004 (2008).
[32] M. Ruggenthaler, M. Penz, and R. van Leeuwen, J. Phys. [67] C. A. Ullrich, U. J. Gossmann, and E. K. U. Gross, Phys.
Condens. Matter 27, 203202 (2015). Rev. Lett. 74, 872 (1995).
[33] I. V. Tokatly, Phys. Rev. B 83, 035127 (2011). [68] H. O. Wijewardane and C. A. Ullrich, Phys. Rev. Lett.
[34] M. Farzanehpour and I. V. Tokatly, Phys. Rev. B 86, 100, 056404 (2008).
125130 (2012). [69] D. Hofmann, T. Körzdörfer, and S. Kümmel, Phys. Rev.
[35] R. D’Agosta and G. Vignale, Phys. Rev. B 71, 245103 Lett. 108, 146401 (2012).
(2005). [70] M. Mundt, S. Kümmel, R. van Leeuwen, and P.-G. Rein-
[36] E. H. Lieb and R. Schrader, Phys. Rev. A 88, 032516 hard, Phys. Rev. A 75, 050501 (2007).
(2013). [71] J. Sun, A. Ruzsinszky, and J. P. Perdew, Phys. Rev. Lett.
[37] B.-X. Xu and A. K. Rajagopal, Phys. Rev. A 31, 2682 115, 036402 (2015).
(1985). [72] C.-O. Almbladh and U. von Barth, Phys. Rev. B 31, 3231
[38] T. K. Ng, Phys. Rev. Lett. 62, 2417 (1989). (1985).
[39] S. K. Ghosh and A. K. Dhara, Phys. Rev. A 38, 1149 [73] J. I. Fuks, K. Luo, E. D. Sandoval, and N. T. Maitra, Phys.
(1988). Rev. Lett. 114, 183002 (2015).
[40] G. Vignale and W. Kohn, Phys. Rev. Lett. 77, 2037 (1996). [74] C. Adamo and D. Jacquemin, Chem. Soc. Rev. 42, 845
[41] G. Vignale, Phys. Rev. B 70, 201102 (2004). (2013).
[42] I. V. Tokatly, Phys. Rev. B 75, 125105 (2007). [75] D. Jacquemin, V. Wathelet, E. A. Perpéte, and C. Adamo,
[43] I. Tokatly, Chem. Phys. 391, 78 (2011). J. Chem. Theory and Comput. 5, 2420 (2009).
[44] J. F. Dobson, M. J. Bünner, and E. K. U. Gross, Phys. Rev. [76] P. Elliott, F. Furche, and K. Burke, in Reviews in Com-
Lett. 79, 1905 (1997). putational Chemistry, edited by K. B. Lipkowitz and T. R.
[45] J. F. Dobson, Phys. Rev. Lett. 73, 2244 (1994). Cundari (Wiley, Hoboken, NJ, 2009) pp. 91–165.
[46] E. K. U. Gross and W. Kohn, Phys. Rev. Lett. 55, 2850 [77] M. Casida and M. Huix-Rotllant, Ann. Rev. Phys. Chem.
(1985). 63, 287 (2012).
[47] G. Vignale, Phys. Rev. Lett. 74, 3233 (1995). [78] M. E. Casida and M. Huix-Rotllant, in Density-Functional
[48] G. Vignale, C. A. Ullrich, and S. Conti, Phys. Rev. Lett. Methods for Excited States, edited by N. Ferré, M. Fila-
79, 4878 (1997). tov, and M. Huix-Rotllant (Springer International Pub-
[49] G. Vignale and W. Kohn, in Electronic Density Functional lishing, Cham, 2016) pp. 1–60.
Theory: Recent Progress and New Directions, edited by J. F. [79] C. A. Ullrich and Z.-h. Yang, Brazilian Journal of Phys.
Dobson, G. Vignale, and M. P. Das (Springer US, Boston, 44, 154 (2013).
MA, 1998) pp. 199–216. [80] S. Botti, A. Schindlmayr, R. D. Sole, and L. Reining, Re-
[50] G. Giuliani and G. Vignale, Quantum Theory of the Elec- ports on Progress in Phys. 70, 357 (2007).
tron Liquid, Masters Series in Physics and Astronomy [81] A. M. Conte, O. Pulci, M. C. Misiti, J. Lojewska,
(Cambridge University Press, 2005). L. Teodonio, C. Violante, and M. Missori, App. Phys.
[51] R. D’Agosta and G. Vignale, Phys. Rev. Lett. 96, 016405 Lett. 104, 224101 (2014).
(2006). [82] J. Jornet-Somoza, J. Alberdi-Rodriguez, B. F. Milne,
[52] C. A. Ullrich, J. Chem. Phys. 125, 234108 (2006). X. Andrade, M. A. L. Marques, F. Nogueira, M. J. T.
[53] V. U. Nazarov, J. M. Pitarke, Y. Takada, G. Vignale, and Oliveira, J. J. P. Stewart, and A. Rubio, Phys. Chem.
Y.-C. Chang, Phys. Rev. B 76, 205103 (2007). Chem. Phys. 17, 26599 (2015).
[54] I. D’Amico and C. A. Ullrich, Phys. Rev. B 74, 121303 [83] D. M. Roessler and W. C. Walker, J. Opt. Soc. Am. 57, 835
(2006). (1967).
[55] C. A. Ullrich and G. Vignale, Phys. Rev. B 58, 15756 [84] A. Marini, R. Del Sole, and A. Rubio, Phys. Rev. Lett. 91,
(1998). 256402 (2003).
[56] N. Sai, M. Zwolak, G. Vignale, and M. Di Ventra, Phys. [85] S. Botti, A. Fourreau, F. m. c. Nguyen, Y.-O. Renault,
Rev. Lett. 94, 186810 (2005). F. Sottile, and L. Reining, Phys. Rev. B 72, 125203 (2005).
[57] M. van Faassen, P. L. de Boeij, R. van Leeuwen, J. A. [86] S. Sharma, J. K. Dewhurst, A. Sanna, and E. K. U. Gross,
Berger, and J. G. Snijders, Phys. Rev. Lett. 88, 186401 Phys. Rev. Lett. 107, 186401 (2011).
(2002). [87] P. E. Trevisanutto, A. Terentjevs, L. A. Constantin, V. Ol-

16
 evano, and F. D. Sala, Phys. Rev. B 87, 205143 (2013). [117] A. Savin, in Recent Developments and Applications of Mod-
[88] S. Rigamonti, S. Botti, V. Veniard, C. Draxl, L. Reining, ern Density Functional Theory, edited by J. M. Seminario
and F. Sottile, Phys. Rev. Lett. 114, 146402 (2015). (Elsevier, 1996) pp. 327–354.
[89] Z.-h. Yang, F. Sottile, and C. A. Ullrich, Phys. Rev. B 92, [118] T. Leininger, H. Stoll, H.-J. Werner, and A. Savin, Chem.
035202 (2015). Phys. Lett. 275, 151 (1997).
[90] S. Refaely-Abramson, M. Jain, S. Sharifzadeh, J. B. [119] A. Seidl, A. Görling, P. Vogl, J. A. Majewski, and
Neaton, and L. Kronik, Phys. Rev. B 92, 081204 (2015). M. Levy, Phys. Rev. B 53, 3764 (1996).
[91] L. Goerigk, J. Moellmann, and S. Grimme, Phys. Chem. [120] C. Jamorski, M. E. Casida, and D. R. Salahub, J. Chem.
Chem. Phys. 11, 4611 (2009). Phys. 104, 5134 (1996).
[92] D. Rappoport and J. Hutter, in Fundamentals of Time- [121] N. T. Maitra, F. Zhang, R. J. Cave, and K. Burke, J. Chem.
Dependent Density Functional Theory, edited by M. A. Phys. 120 (2004).
Marques, N. T. Maitra, F. M. Nogueira, E. Gross, and [122] P. Elliott, S. Goldson, C. Canahui, and N. T. Maitra,
A. Rubio (Springer Berlin Heidelberg, Berlin, Heidel- Chem. Phys. 391, 110 (2011).
berg, 2012) pp. 317–336. [123] R. J. Cave, F. Zhang, N. T. Maitra, and K. Burke, Chem.
[93] M. Petersilka, U. J. Gossmann, and E. K. U. Gross, Phys. Phys. Lett. 389, 39 (2004).
Rev. Lett. 76, 1212 (1996). [124] M. E. Casida, J. Chem. Phys. 122, 054111 (2005).
[94] M. Petersilka and E. K. U. Gross, Int. J. Quant. Chem. 60, [125] M. Huix-Rotllant, A. Ipatov, A. Rubio, and M. E. Casida,
1393 (1996). Chem. Phys. 391, 120 (2011).
[95] M. Casida, in Recent Advances in Density Functional Meth- [126] P. Romaniello, D. Sangalli, J. A. Berger, F. Sottile, L. G.
ods, Part I, edited by D. Chong (World Scientific, Singa- Molinari, L. Reining, and G. Onida, J. Chem. Phys. 130,
pore, 1995). 044108 (2009).
[96] M. E. Casida, in Recent Developments and Applications of [127] O. V. Gritsenko and E. Jan Baerends, Phys. Chem. Chem.
Modern Density Functional Theory, edited by J. M. Semi- Phys. 11, 4640 (2009).
nario (Elsevier, Amsterdam, 1996) p. 391. [128] G. Mazur, M. Makowski, R. Wldarczyk, and Y. Aoki, Int.
[97] R. Bauernschmitt and R. Ahlrichs, Chem. Phys. Lett. 256, J. Quant. Chem. 111, 819 (2011).
454 (1996). [129] G. Mazur and R. Wlodarczyk, J. Comput. Chem. 30, 811
[98] T. Grabo, M. Petersilka, and E. Gross, Journal of Molec- (2009).
ular Structure: THEOCHEM 501, 353 (2000). [130] A. Dreuw, J. L. Weisman, and M. Head-Gordon, J.
[99] R. M. Sternheimer, Phys. Rev. 96, 951 (1954). Chem. Phys. 119, 2943 (2003).
[100] X. Andrade, S. Botti, M. A. L. Marques, and A. Rubio, J. [131] D. J. Tozer, J. Chem. Phys. 119, 12697 (2003).
Chem. Phys. 126, 184106 (2007). [132] A. Dreuw and M. Head-Gordon, J. Am. Chem. Soc. 126,
[101] D. A. Strubbe, L. Lehtovaara, A. Rubio, M. A. Marques, 4007 (2004).
and S. G. Louie, in Fundamentals of Time-Dependent Den- [133] O. Gritsenko and E. J. Baerends, J. Chem. Phys. 121, 655
sity Functional Theory, edited by M. A. Marques, N. T. (2004).
Maitra, F. M. Nogueira, E. Gross, and A. Rubio (Springer [134] Y. Tawada, T. Tsuneda, S. Yanagisawa, T. Yanai, and
Berlin Heidelberg, 2012) pp. 139–166. K. Hirao, J. Chem. Phys. 120, 8425 (2004).
[102] K. Yabana, T. Nakatsukasa, J.-I. Iwata, and G. Bertsch, [135] M. A. Rohrdanz, K. M. Martins, and J. M. Herbert, J.
Physica Status Solidi (b) 243, 1121 (2006). Chem. Phys. 130, 054112 (2009).
[103] M. E. Casida, C. Jamorski, K. C. Casida, and D. R. [136] J. Autschbach, ChemPhysChem 10, 1757 (2009).
Salahub, J. Chem. Phys. 108 (1998). [137] T. Stein, L. Kronik, and R. Baer, J. Am. Chem. Soc. 131,
[104] D. J. Tozer and N. C. Handy, J. Chem. Phys. 109 (1998). 2818 (2009).
[105] D. J. Tozer and N. C. Handy, Phys. Chem. Chem. Phys. [138] R. Baer, E. Livshits, and U. Salzner, Ann. Rev. Phys.
2, 2117 (2000). Chem. 61, 85 (2010).
[106] M. E. Casida and D. R. Salahub, J. Chem. Phys. 113 [139] A. Heßelmann, A. Ipatov, and A. Görling, Phys. Rev. A
(2000). 80, 012507 (2009).
[107] A. Wasserman, N. T. Maitra, and K. Burke, Phys. Rev. [140] A. Karolewski, L. Kronik, and S. Kümmel, J. Chem.
Lett. 91, 263001 (2003). Phys. 138, 204115 (2013).
[108] R. van Leeuwen and E. J. Baerends, Phys. Rev. A 49, 2421 [141] M. Hellgren and E. K. U. Gross, Phys. Rev. A 85, 022514
(1994). (2012).
[109] A. Görling, Phys. Rev. Letters 83, 5459 (1999). [142] N. T. Maitra, J. Chem. Phys. 122, 234104 (2005).
[110] F. Della Sala and A. Görling, Int. J. Quant. Chem. 91, 131 [143] N. T. Maitra and D. G. Tempel, J. Chem. Phys. 125,
(2003). 184111 (2006).
[111] K. B. Wiberg, R. Stratmann, and M. J. Frisch, Chem. [144] B. G. Levine, C. Ko, J. Quenneville, and T. J. Martinez,
Phys. Lett. 297, 60 (1998). Mol. Phys. 104, 1039 (2006).
[112] D. J. Tozer, R. D. Amos, N. C. Handy, B. O. Roos, and [145] E. Tapavicza, I. Tavernelli, U. Rothlisberger, C. Filippi,
L. Serrano-Andres, Mol. Phys. 97, 859 (1999). and M. E. Casida, J. Chem. Phys. 129, 124108 (2008).
[113] Y. Zhao, , and D. G. Truhlar, J. Phys. Chem. A 110, 13126 [146] X. Zhang and J. M. Herbert, J. Chem. Phys. 141, 064104
(2006). (2014).
[114] M. J. G. Peach, A. J. Cohen, and D. J. Tozer, Phys. Chem. [147] A. A. Quong and A. G. Eguiluz, Phys. Rev. Lett. 70, 3955
Chem. Phys. 8, 4543 (2006). (1993).
[115] M. J. G. Peach, P. Benfield, T. Helgaker, and D. J. Tozer, [148] G. Onida, L. Reining, and A. Rubio, Rev. Mod. Phys. 74,
J. Chem. Phys. 128, 044118 (2008). 601 (2002).
[116] T. Yanai, D. P. Tew, and N. C. Handy, Chem. Phys. Lett. [149] H.-C. Weissker, J. Serrano, S. Huotari, F. Bruneval, F. Sot-
393, 51 (2004). tile, G. Monaco, M. Krisch, V. Olevano, and L. Reining,

17
 Phys. Rev. Lett. 97, 237602 (2006). [179] A. J. Cohen, P. Mori-Sánchez, and W. Yang, Science 321,
[150] L. Reining, V. Olevano, A. Rubio, and G. Onida, Phys. 792 (2008).
Rev. Lett. 88, 066404 (2002). [180] A. J. Cohen, P. Mori-Sánchez, and W. Yang, Chem. Rev.
[151] I. V. Tokatly and O. Pankratov, Phys. Rev. Lett. 86, 2078 112, 289 (2012).
(2001). [181] E. Kraisler and L. Kronik, Phys. Rev. Lett. 110, 126403
[152] C. A. Ullrich and Z.-H. Yang, in Density-Functional Meth- (2013).
ods for Excited States, Vol. 368, edited by N. Ferré, M. Fila- [182] A. Görling, Phys. Rev. B 91, 245120 (2015).
tov, and M. Huix-Rotllant (Springer International Pub- [183] L. Kronik, T. Stein, S. Refaely-Abramson, and R. Baer, J.
lishing, Cham, 2016) pp. 185–17. Chem. Theory and Comput. 8, 1515 (2012).
[153] G. Vignale, in Fundamentals of Time-Dependent Density [184] J. B. Neaton, M. S. Hybertsen, and S. G. Louie, Phys.
Functional Theory, edited by M. A. Marques, N. T. Maitra, Rev. Lett. 97, 216405 (2006).
F. M. Nogueira, E. Gross, and A. Rubio (Springer Berlin [185] D. A. Egger, Z.-F. Liu, J. B. Neaton, and L. Kronik, Nano
Heidelberg, 2012) pp. 457–469. Letters 15, 2448 (2015).
[154] Z. H. Levine and D. C. Allan, Phys. Rev. Lett. 63, 1719 [186] F. D. Angelis, Acc. Chem. Res. 47, 3349 (2014).
(1989). [187] A. Migani, D. J. Mowbray, A. Iacomino, J. Zhao, H. Pe-
[155] X. Gonze, P. Ghosez, and R. W. Godby, Phys. Rev. Lett. tek, and A. Rubio, J. Am. Chem. Soc. 135, 11429 (2013).
74, 4035 (1995). [188] Y. Shinohara, K. Yabana, Y. Kawashita, J.-I. Iwata,
[156] Z.-h. Yang, Y. Li, and C. A. Ullrich, J. Chem. Phys. 137, T. Otobe, and G. F. Bertsch, Phys. Rev. B 82, 155110
014513 (2012), http://dx.doi.org/10.1063/1.4730031. (2010).
[157] Y.-H. Kim and A. Görling, Phys. Rev. Lett. 89, 096402 [189] K. Yabana, T. Sugiyama, Y. Shinohara, T. Otobe, and
(2002). G. F. Bertsch, Phys. Rev. B 85, 045134 (2012).
[158] F. Bruneval, F. Sottile, V. Olevano, and L. Reining, J. [190] S. A. Sato, K. Yabana, Y. Shinohara, T. Otobe, K.-M. Lee,
Chem. Phys. 124, 144113 (2006). and G. F. Bertsch, Phys. Rev. B 92, 205413 (2015).
[159] F. Sottile, V. Olevano, and L. Reining, Phys. Rev. Lett. 91, [191] K. Krieger, J. K. Dewhurst, P. Elliott, S. Sharma, and
056402 (2003). E. K. U. Gross, J. Chem. Theory and Comput. 11, 4870
[160] S. Botti, F. Sottile, N. Vast, V. Olevano, L. Reining, H.-C. (2015).
Weissker, A. Rubio, G. Onida, R. Del Sole, and R. W. [192] C. A. Ullrich and A. D. Bandrauk, in Fundamentals
Godby, Phys. Rev. B 69, 155112 (2004). of Time-Dependent Density Functional Theory, edited by
[161] J. A. Berger, Phys. Rev. Lett. 115, 137402 (2015). M. A. Marques, N. T. Maitra, F. M. Nogueira, E. Gross,
[162] V. U. Nazarov and G. Vignale, Phys. Rev. Lett. 107, and A. Rubio (Springer Berlin Heidelberg, 2012) pp. 351–
216402 (2011). 371.
[163] L. Bernasconi, S. Tomić, M. Ferrero, M. Rérat, R. Or- [193] S.-I. Chu, J. Chem. Phys. 123, 062207 (2005).
lando, R. Dovesi, and N. M. Harrison, Phys. Rev. B 83, [194] E. Penka Fowe and A. D. Bandrauk, Phys. Rev. A 84,
195325 (2011). 035402 (2011).
[164] V. Turkowski, A. Leonardo, and C. A. Ullrich, Phys. Rev. [195] D. Dundas and J. Rost, Phys. Rev. A 71, 013421 (2005).
B 79, 233201 (2009). [196] X. Chu and S.-I. Chu, Phys. Rev. A 64, 063404 (2001).
[165] Z.-h. Yang and C. A. Ullrich, Phys. Rev. B 87, 195204 [197] K. Nobusada and K. Yabana, Phys. Rev. A 70, 043411
(2013). (2004).
[166] J. P. Perdew, R. G. Parr, M. Levy, and J. L. Balduz, Phys. [198] J. Haruyama, C. Hu, and K. Watanabe, Phys. Rev. A 85,
Rev. Lett. 49, 1691 (1982). 062511 (2012).
[167] C. Toher, A. Filippetti, S. Sanvito, and K. Burke, Phys. [199] I. Bocharova, R. Karimi, E. F. Penka, J.-P. Brichta, P. Las-
Rev. Lett. 95, 146402 (2005). sonde, X. Fu, J.-C. Kieffer, A. D. Bandrauk, I. Litvinyuk,
[168] M. Koentopp, C. Chang, K. Burke, and R. Car, J. Phys. J. Sanderson, and F. m. c. Légaré, Phys. Rev. Lett. 107,
Condens. Matter 20, 083203 (2008). 063201 (2011).
[169] S. Kurth, G. Stefanucci, E. Khosravi, C. Verdozzi, and [200] A. Russakoff, S. Bubin, X. Xie, S. Erattupuzha, M. Kitzler,
E. K. U. Gross, Phys. Rev. Lett. 104, 236801 (2010). and K. Varga, Phys. Rev. A 91, 023422 (2015).
[170] S. Kurth and G. Stefanucci, Phys. Rev. Lett. 111, 030601 [201] T. Burnus, M. A. L. Marques, and E. K. U. Gross, Phys.
(2013). Rev. A 71, 010501 (2005).
[171] P. Schmitteckert, M. Dzierzawa, and P. Schwab, Phys. [202] G. Wachter, S. Nagele, S. A. Sato, R. Pazourek, M. Wais,
Chem. Chem. Phys. 15, 5477 (2013). C. Lemell, X.-M. Tong, K. Yabana, and J. Burgdörfer,
[172] G. Stefanucci and C.-O. Almbladh, Phys. Rev. B 69, Phys. Rev. A 92, 061403 (2015).
195318 (2004). [203] P. Wopperer, P. Dinh, P.-G. Reinhard, and E. Suraud,
[173] G. Stefanucci and C.-O. Almbladh, Europhys. Lett. 67, 14 Phys. Reports 562, 1 (2015).
(2004). [204] A. Castro, J. Werschnik, and E. K. U. Gross, Phys. Rev.
[174] K. Burke, R. Car, and R. Gebauer, Phys. Rev. Lett. 94, Lett. 109, 153603 (2012).
146803 (2005). [205] K. Krieger, A. Castro, and E. Gross, Chem. Phys. 391, 50
[175] J. Yuen-Zhou, D. Tempel, C. Rodriguez-Rosario, and (2011).
A. Aspuru-Guzik, Phys. Rev. Lett. 104, 043001 (2010). [206] A. Castro, A. Rubio, and E. K. U. Gross, Eur. Phys. J. B
[176] M. di Ventra and R. d’Agosta, Phys. Rev. Lett. 98, 226403 88, 1 (2015).
(2007). [207] P. Elliott, K. Krieger, J. Dewhurst, S. Sharma, and
[177] X. Zheng, F. Wang, C. Yam, Y. Mo, and G.-H. Chen, E. Gross, New Journal of Phys. 18, 013014 (2016).
Phys. Rev. B 75, 195127 (2007). [208] C. Verdozzi, Physical review letters 101, 166401 (2008).
[178] D. Neuhauser and K. Lopata, J. Chem. Phys. 129, 134106 [209] C. Verdozzi, D. Karlsson, M. P. von Friesen, C.-O. Alm-
(2008). bladh, and U. von Barth, Chem. Phys. 391, 37 (2011).

18
[210] M. Ruggenthaler and D. Bauer, Phys. Rev. Lett. 102, and C. Lienau, Science 344, 1001 (2014).
233001 (2009). [238] E. Tapavicza, I. Tavernelli, and U. Rothlisberger, Phys.
[211] J. I. Fuks, P. Elliott, A. Rubio, and N. T. Maitra, J. Phys. Rev. Lett. 98, 023001 (2007).
Chem. Lett. 4, 735 (2013). [239] B. F. Curchod, U. Rothlisberger, and I. Tavernelli,
[212] M. Provorse and C. Isborn, Int. J. Quant. Chem. 116, 739 ChemPhysChem 14, 1314 (2013).
(2016). [240] E. Tapavicza, G. D. Bellchambers, J. C. Vincent, and
[213] J. I. Fuks and N. T. Maitra, Phys. Chem. Chem. Phys. 16, F. Furche, Phys. Chem. Chem. Phys. 15, 18336 (2013).
14504 (2014). [241] R. Mitric, U. Werner, and V. Bonacic-Koutecky, J. Chem.
[214] J. I. Fuks and N. T. Maitra, Phys. Rev. A 89, 062502 (2014). Phys. 129, 164118 (2008).
[215] P. Elliott and N. T. Maitra, Phys. Rev. A 85, 052510 (2012). [242] M. Barbatti and R. Crespo-Otero, in Density-Functional
[216] M. Thiele and S. Kümmel, Phys. Chem. Chem. Phys. 11, Methods for Excited States, edited by N. Ferré, M. Fila-
4631 (2009). tov, and M. Huix-Rotllant (Springer International Pub-
[217] R. Requist and O. Pankratov, Phys. Rev. A 81, 042519 lishing, Cham, 2016) pp. 415–444.
(2010). [243] C. F. Craig, W. R. Duncan, and O. V. Prezhdo, Phys. Rev.
[218] L. Mancini, J. D. Ramsden, M. J. P. Hodgson, and R. W. Lett. 95, 163001 (2005).
Godby, Phys. Rev. B 89, 195114 (2014). [244] O. V. Prezhdo, W. R. Duncan, and V. V. Prezhdo,
[219] M. Lein and S. Kümmel, Phys. Rev. Lett. 94, 143003 Progress in Surface Science 84, 30 (2009).
(2005). [245] J. C. Tully and R. Preston, J. Chem. Phys. 55, 562 (1971).
[220] M. J. P. Hodgson, J. D. Ramsden, J. B. J. Chapman, [246] E. Tapavicza, A. M. Meyer, and F. Furche, Phys. Chem.
P. Lillystone, and R. W. Godby, Phys. Rev. B 88, 241102 Chem. Phys. 13, 20986 (2011).
(2013). [247] B. F. E. Curchod, T. J. Penfold, U. Rothlisberger, and
[221] S. Raghunathan and M. Nest, J. Chem. Theory and Com- I. Tavernelli, ChemPhysChem 16, 2127 (2015).
put. 7, 2492 (2011). [248] I. Tavernelli, B. F. E. Curchod, and U. Rothlisberger, J.
[222] R. Ramakrishnan and M. Nest, Phys. Rev. A 85, 054501 Chem. Phys. 131, 196101 (2009).
(2012). [249] V. Chernyak and S. Mukamel, J. Chem. Phys. 112 (2000).
[223] S. Raghunathan and M. Nest, J. Chem. Theory and Com- [250] R. Baer, Chem. Phys. Lett. 364, 75 (2002).
put. 8, 806 (2012). [251] C. Hu, H. Hirai, and O. Sugino, J. Chem. Phys. 127,
[224] F. Krausz and M. Ivanov, Rev. Mod. Phys. 81, 163 (2009). 064103 (2007).
[225] E. Perfetto and G. Stefanucci, Phys. Rev. A 91, 033416 [252] R. Send and F. Furche, J. Chem. Phys. 132, 044107 (2010).
(2015). [253] Z. Li and W. Liu, J. Chem. Phys. 141, 014110 (2014).
[226] B. F. Habenicht, N. P. Tani, M. R. Provorse, and C. M. [254] Z. Li, B. Suo, and W. Liu, J. Chem. Phys. 141, 244105
Isborn, J. Chem. Phys. 141, 184112 (2014). (2014).
[227] M. J. T. Oliveira, B. Mignolet, T. Kus, T. A. Papadopoulos, [255] X. Zhang and J. M. Herbert, J. Chem. Phys. 142, 064109
F. Remacle, and M. J. Verstraete, J. Chem. Theory and (2015).
Comput. 11, 2221 (2015). [256] Q. Ou, G. D. Bellchambers, F. Furche, and J. E. Subotnik,
[228] U. De Giovannini, G. Brunetto, A. Castro, J. Walken- J. Chem. Phys. 142, 064114 (2015).
horst, and A. Rubio, ChemPhysChem 14, 1363 (2013). [257] S. A. Fischer, C. J. Cramer, and N. Govind, J. Chem. The-
[229] N. Henkel, M. Keim, H. LÃ 14 dde, and T. Kirchner, Phys. ory and Comput. 11, 4294 (2015).
Rev. A 80, 032704 (2009). [258] A. Krishtal, D. Ceresoli, and M. Pavanello, J. Chem.
[230] D. G. Lappas and R. van Leeuwen, J. Phys. B: At., Mol. Phys. 142, 154116 (2015).
and Opt. Phys. 31, L249 (1998). [259] M. Pavanello, J. Chem. Phys. 138, 204118 (2013).
[231] F. Wilken and D. Bauer, Phys. Rev. A 76, 023409 (2007). [260] J. Neugebauer, J. Chem. Phys. 126, 134116 (2007).
[232] C. Ullrich and E. Gross, Comments on At. and Mol. [261] M. E. Casida and T. A. Wesolowski, Int. J. Quantum
Phys. 33, 211 (1997). Chem. 96, 577 (2004).
[233] M. Petersilka and E. Gross, LASER PHYSICS- [262] M. A. Mosquera, D. Jensen, and A. Wasserman, Phys.
LAWRENCE- 9, 105 (1999). Rev. Lett. 111, 023001 (2013).
[234] F. Wilken and D. Bauer, Phys. Rev. Lett. 97, 203001 (2006). [263] I. V. Tokatly, Phys. Rev. Lett. 110, 233001 (2013).
[235] A. K. Rajam, I. Raczkowska, and N. T. Maitra, Phys. Rev. [264] C. Pellegrini, J. Flick, I. V. Tokatly, H. Appel, and A. Ru-
Lett. 105, 113002 (2010). bio, Phys. Rev. Lett. 115, 093001 (2015).
[236] C. A. Rozzi, S. M. Falke, N. Spallanzani, A. Ru- [265] J. Flick, M. Ruggenthaler, H. Appel, and A. Rubio, Proc.
bio, E. Molinari, D. Brida, M. Maiuri, G. Cerullo, Natl. Acad. Sci. 112, 15285 (2015).
H. Schramm, J. Christoffers, et al., Nature Communica- [266] M. Ruggenthaler, J. Flick, C. Pellegrini, H. Appel, I. V.
tions 4, 1602 (2013). Tokatly, and A. Rubio, Phys. Rev. A 90, 012508 (2014).
[237] S. M. Falke, C. A. Rozzi, D. Brida, M. Maiuri, M. Amato, [267] O. Butriy, H. Ebadi, P. L. de Boeij, R. van Leeuwen, and
E. Sommer, A. De Sio, A. Rubio, G. Cerullo, E. Molinari, E. K. U. Gross, Phys. Rev. A 76, 052514 (2007).

19