Chemical

Science
EDGE ARTICLE

Can range-separated functionals be optimally

Cite this: Chem. Sci., 2023, 14, 1491
tuned to predict spectra and excited state dynamics
All publication charges for this article
in photoactive iron complexes?†
have been paid for by the Royal Society
of Chemistry J. Patrick Zobel,*a Ayla Kruse, bc Omar Baig, a Stefan Lochbrunner, bc

Sergey I. Bokarev, bd Oliver Kühn, b Leticia González a

and Olga S. Bokareva *be

Density functional theory is an efficient computational tool to investigate photophysical and photochemical
processes in transition metal complexes, giving invaluable assistance in interpreting spectroscopic and
catalytic experiments. Optimally tuned range-separated functionals are particularly promising, as they
were created to address some of the fundamental deficiencies present in approximate exchange-
correlation functionals. In this paper, we scrutinize the selection of optimally tuned parameters and its
influence on the excited state dynamics, using the example of the iron complex [Fe(cpmp)2]2+ with
push–pull ligands. Various tuning strategies are contemplated based on pure self-consistent DFT
protocols, as well as on the comparison with experimental spectra and multireference CASPT2 results.
The two most promising sets of optimal parameters are then employed to carry out nonadiabatic
surface-hopping dynamics simulations. Intriguingly, we find that the two sets lead to very different
relaxation pathways and timescales. While the set of optimal parameters from one of the self-consistent
DFT protocols predicts the formation of long-lived metal-to-ligand charge transfer triplet states, the set
Received 21st October 2022
Accepted 27th December 2022
in better agreement with CASPT2 calculations leads to deactivation in the manifold of metal-centered
states, in better agreement with the experimental reference data. These results showcase the complexity
DOI: 10.1039/d2sc05839a
of iron-complex excited state landscapes and the difficulty of obtaining an unambiguous parametrization
rsc.li/chemical-science of long-range corrected functionals without experimental input.

charge-transfer (MLCT) states via metal-centered (MC) ones to

1 Introduction the ground state.6,7 This relaxation prevents utilizing the initial
Transition metal complexes are widely applied as catalysts, charge separation in MLCT states for further redox reactions. In
photosensitizers, and dyes.1–3 Popular examples are six- order to nd novel complexes with long-lived MLCT states, the
coordinated ruthenium(II) and iridium(III) polypyridine excited-state behavior of such compounds needs to be carefully
complexes that have found numerous applications in rationalized.8–10 An atomistic picture of the excited-state
photocatalysis and light-harvesting technologies, including dye- reactivity can be obtained by simulating the photodynamics of
sensitized solar cells. However, due to the costs of rare metals, transition metal complexes.11 If the reaction coordinates of the
there is a signicant effort to switch to the more Earth- problem at hand are known and comprise only a few degrees of
abundant and economic copper- or iron-based complexes.4,5 freedom, it is possible to simulate excited-state processes with
Unfortunately, the usage of iron in polypyridine complexes is high accuracy using quantum wave packet dynamics.12 By
still limited due to the fast deactivation of metal-to-ligand contrast, if too many degrees of freedom are involved or the
relevant ones cannot be guessed a priori, a widely used strategy
a
is to apply ab initio molecular dynamics methods, such as
Institute of Theoretical Chemistry, Faculty of Chemistry, University of Vienna,
Währingerstr. 19, 1090 Vienna, Austria. E-mail: [email protected] trajectory surface hopping (TSH).13 The latter simulate the
b
Institute of Physics, University of Rostock, Albert-Einstein-Straße 23-24, 18059 photodynamics of a molecule in full dimensionality leveraging
Rostock, Germany. E-mail: [email protected] approximate classical trajectories for the nuclear motion
c
Department of Life, Light and Matter, University of Rostock, 18051 Rostock, Germany instead of quantum wave packets.
d
Chemistry Department, Technical University of Munich, Lichtenbergstr. 4, Garching Regardless of quantum wave packets or classical trajectories,
85748, Germany
e
the motion of the nuclei is governed by the topology of the
Institute of Physics, University of Kassel, Heinrich-Plett-Straße 40, 34132 Kassel,
potential energy surfaces (PESs), so that this will be meaningful,
Germany
† Electronic supplementary information (ESI) available. See DOI:
only if the underlying electronic states are accurately described.
https://doi.org/10.1039/d2sc05839a Transition metal complexes, with their high density of low-lying

© 2023 The Author(s). Published by the Royal Society of Chemistry Chem. Sci., 2023, 14, 1491–1502 | 1491
Chemical Science Edge Article

excited states of different characters, pose intricate challenges to previously in an analogous ruthenium(II) compound.33
electronic structure methods. Due to its moderate computational This complex, recently synthesized and experimentally
costs, efficiency, and “black-box” character, density functional characterized,34 follows two strategies for achieving long-sought
theory (DFT) and its linear-response time-dependent extension long-lived MLCT states:35 forming a close to octahedral
(TDDFT) enjoy great popularity.14 For dynamics simulations, coordination of the central iron that maximizes the ligand-eld
where a large number of single-point calculations are required, splitting and destabilizing the MC states by electron-rich
using formally higher-ranked theoretical methods is not yet an ligands while simultaneously stabilizing the MLCT states by
alternative. However, DFT suffers from errors imbued in the p-acceptor groups, such as in other iron(II) push–pull
approximate nature of the exchange-correlation functionals. complexes.36–40
A common problem of many approximate functionals is the We rst investigate different routes to tune the range-
erroneous description of charge transfer (CT) states due to separation parameters of the LC-BLYP functional18 (Section 2).
unphysical electronic self-interaction.15,16 This problem can be The accuracy of different sets of range-separation parameters is
alleviated by including exact exchange interaction from benchmarked against experimental data and the more accurate
Hartree–Fock theory in the functional. While hybrid functionals multi-reference wave function based method CASPT2 (Section
such as B3LYP do this in a xed manner, a more balanced 3). We then use the two most promising sets of range-separation
inclusion is realized in range-separated functionals. They parameters to parametrize the PESs on which TSH molecular
include the exact exchange weighted with a damping function dynamics simulations are carried out (Section 4). To our
that depends on the inter-electronic distance, ensuring surprise, the simulated photodynamics delivers strikingly
a smooth transition from the short- to the long-range different mechanisms that can be traced back to the different
domain.17–19 This approach aims at minimizing the spurious electronic characters of the states obtained with the two sets of
electron's self-interaction energy20,21 and mitigating the (de) parameters. In order to determine the correct mechanism, we
localization error.22 It remains then only necessary to determine need time-resolved spectroscopic experiments. This study
the steepness of the damping function, a feature that can be highlights how theory alone can struggle to provide the correct
obtained for the individual system by tuning the energetic dynamical picture and how easily one can be led astray.
positions of highest occupied (HOMO) and lowest unoccupied
molecular orbitals (LUMO) to t the ab initio ionization
potential (IP) and electron affinity (EA). 2 Tuning of the DFT functional
Despite the fact that tuned range-separated functionals have 2.1 DSCF tuning for the electronic ground state
been shown to be very useful for describing properties related to The range-separation parameters a and u have been tuned for
fundamental and optical energy gaps,22–32 it can be non-trivial the generalized form (eqn. (1)) of the LC-BLYP functional based
to obtain a unique set of range-separation parameters,25 on the following partitioning of the Coulomb operator:18
particularly if one needs to describe both CT and locally excited
states on the same footing. Furthermore, it is far from obvious 1 1  ½a þ b  erfðður12 ÞÞ a þ b  erfður12 Þ
¼ þ : (1)
whether different parameters would ultimately lead to the same r12 r12 r12
relaxation mechanisms aer light irradiation. Therefore, in this With this, the initial exchange kernel in DFT is complemented
work, we critically examine the performance of several range- with the exact Hartree-Fock exchange, which has the correct
separation parameters on the absorption spectrum, PESs, and asymptotic behavior. The u parameter denes the switching
ultimately on the excited state dynamics of the Fe(II) complex rate between short and long-range exchange, and its inverse
[Fe(cpmp)2]2+ (see Fig. 1), where cpmp = 6,2′′-carboxypyridyl- is proportional to a characteristic interelectron distance.
2,2′-methylamine-pyridyl-pyridine is a push–pull ligand used The second (dimensionless) parameter a sets a global

Fig. 1 (a) Schematic and 3D view of [Fe(cpmp)2]2+ (cpmp = 6,2′′-carboxypyridyl-2,2′-methylamine-pyridyl-pyridine). (b) Active space for
multireference calculations. The occupied and vacant orbitals in the ground state are marked with orange and green boxes, respectively.

1492 | Chem. Sci., 2023, 14, 1491–1502 © 2023 The Author(s). Published by the Royal Society of Chemistry
Edge Article Chemical Science

r12-independent exact-exchange contribution. We assume b = 1 The variational stability of all combinations of parameters
− a to ensure that self-interaction is asymptotically canceled by was investigated and no convergence towards other electronic
the exact exchange. Thus, the range-separated part of the LC- states was observed. Thus, DSCF tuning of the S0 state suggests
BLYP functional depends on the parameters (a, u) that we the global minimum of J* found at (0.00; 0.14) as the optimal
can exploit for tuning. The particular functional LC-BLYP has range-separation parameters. In the following, this will be
been chosen for tuning, as it is the most similar one to the abbreviated as set A. We note, however, that the J* tuning
popular B3LYP that we have applied for the purpose of delivers only a compromise between the J0 (HOMO/IP) and J1
comparison. Note that other functionals like BNL or PBE could (LUMO/EA) tuning. It can be seen from Fig. 2, showing that the
be a possible option too, see e.g. ref. 22 and 29. minimal values of J0 and J1 (black and blue lines) do not
The tuning most commonly follows the DSCF method.21,41,42 coincide.
Here, the energetic positions of the HOMO and LUMO of a given
system are tuned to the ionization potential (IP) and electron 2.2 “MC-tuning” for the electronic ground state
affinity (EA) according to Koopmans' theorem. To nd the
The standard DSCF21,41,42 approach is designed for the
optimal range-separation parameters, we computed the tuning
adjustment of the positions of HOMO and LUMO energies.
function J*, which is the measure of violation of Koopmans'
Thus, for complexes where the frontier orbitals are localized on
theorem for both the HOMO and the LUMO, on a grid of (a, u)
both the metal and ligands, it results in a better description of
pairs using the LC-BLYP functional and 6-31G(d) basis set43,44 at
MLCT states. To better reproduce the transition energies of MC
the B3LYP45/def2TZVP46,47 optimized S0 ground-state geometry.
states, we follow the same approach as above but instead of
Additional computational details are reported in Section S1.1 and
using the LUMO localized on ligands, we use s*dz2 and s*dx2 y2
the general strategy employed in our optimal tuning calculations
orbitals localized on the iron center, see Fig. 1. Using this “MC-
is explained in Section S2. In this study, we focus on the ranges
tuning” leads to the optimal range-separation parameter pairs
[0.0–0.3] for a and [0.00–0.25] for u, which are typical values for
of (0.05, 0.17) and (0.0, 0.17) for both s*-orbitals (see green/blue
organometallic systems with conjugated ligands.48–58
stars in Fig. 2). The sets of parameters for both MC-tuning
The resulting two-dimensional plot of J*(a, u) presented in
variants are quite close to each other. Both points, however,
Fig. 2 displays a global minimum of J* at (0.00; 0.14); exemplary
lie quite apart from the HOMO/IP tuning condition (J0) which
1D-cuts at a = 0 are in Fig. S1.† This minimum is located in the
appears unexpected. It is noteworthy that the inuence of the
minimal valley of points marked with white circles in Fig. 2(a),
LUMO/EA condition (J1) in this approach is much larger than
while other a values correspond to near-optimal (a, u) choices.
for the standard scheme which likely explains the different
We analyzed the deviations from piecewise-linearity59,60 for
HOMO/IP results.
fractional electron charges (Section S2.1) to further scrutinize
different points along the minimal valley on J*(a, u), where
again, the (0.00; 0.14) parameters without short-range exchange 2.3 DSCF tuning of electronic excited states
performed best. The study of photophysical properties requires to have a balanced
description of multiple electronic states. Our optimal tuning
considered so far only the electronic ground state, although we
have used different LUMOs to target electronic states of different
characters. The transferability of the tuned parameters to
electronically excited states is investigated, performing optimal-
tuning calculations for the two lowest triplet excited states of
MC (TMC) and MLCT characters (TMLCT) as well as the lowest
excited quintet state also of MC character (QMC). For the excited
states, we considered only tuning the HOMO/IP condition J0 and
show thus, only the minimum valleys of this function for the
different electronic states in Fig. 2.
As can be seen, the position of optimal (a, u) pairs is
sensitive to both the multiplicity and character of the electronic
state. For the TMC state, the resulting set of (a, u) pairs (green
points) is close to the HOMO/IP curve for the ground state
(black curve). This behavior is due to the HOMO being located
on the central iron atom in the ground state. Analogously, the
Fig. 2 J*(a, u) (measure of violation of Koopmans' theorem for both optimal parameters for TMLCT are close to the LUMO/EA set
the HOMO and the LUMO (see detailed explanation in Section S2) for where the ligand-localized LUMO is considered. For the QMC
[Fe(cpmp)2]2+ in the ground electronic state. White points denote the excited state, a double MC excitation, the results are more
minima of J*(a, u) at constant a values for the ground S0 state. Black,
complicated. Its optimal parameters (dark red curve in Fig. 2)
green, orange and red points denote the corresponding minima for IP-
tuning only for S0, TMC, TMLCT, and QMC states, respectively. Optimal are closer to the HOMO/IP condition than to the LUMO/EA one.
points for triplet tuning and MC-tuning are also presented, see the text This fact might be connected with the higher multiplicity,
for more details. where strong exchange plays a greater role. Importantly, the

© 2023 The Author(s). Published by the Royal Society of Chemistry Chem. Sci., 2023, 14, 1491–1502 | 1493
Chemical Science Edge Article

optimal parameters of the (full) J* tuning (HOMO/IP + LUMO/ Fig. 3(a) shows the ranges of the vertical S1 and T1 excitation
EA) of the electronic ground state fall in the middle of the energies at the S0 geometry as well as the adiabatic TMC, TMLCT,
HOMO/IP valleys of the different excited states, thus and QMC excitation energies over the whole set of parameters.
representing a compromise between multiple electronic states. These energy ranges are plotted against the Fe–N bond length,
which is the main geometric difference between these
2.4 Triplet tuning electronic states. We note that the energies of the TMLCT state
have the smallest dependence on the tested range-separation
As an additional option, we have considered the recently
parameters, whereas both MC states vary substantially. The
suggested “triplet tuning” scheme55 based on the assumption
inuence of the range-separation parameters on the QMC state
that DSCF and TDDFT approaches with the exact exchange-
is even larger than for the TMC state. For a certain domain of
correlation functional should yield the same energy of the
parameters, the QMC is lower than the singlet ground state, see
rst triplet excited state. We located the global minimum at
Fig. S3.† This fact is not surprising since the exchange energy
(0.10; 0.15); see the yellow star in Fig. 2. In general, the nature of
plays a more important role for the MC states than for the MLCT
the lowest triplet states can change depending on the range-
state, as the unpaired spins are located on the same moiety, i.e.,
separation parameters. In the present case, the triplet tuning
the iron center, and not separated by a relatively large distance
parameters are similar to the u value of the HOMO/IP minimum
as in the case of MLCT states.
in the TMLCT state for a = 0.10; compare the position of the
yellow asterisk and nearest green circle in Fig. 2. Fortunately, in
this range of parameters, the MLCT state is the lowest one such 3.2 Comparison to multireference calculations
that the tuning results are consistent with each other.
To evaluate the correctness of the parameters for LC-BLYP, we
use the high-level wave-function based method CASPT2 (Section
3 Performance of range-separation parameters in stationary
S1.2†). The active space (Fig. 1) is able to describe both MC and
calculations
MLCT states. The energies of the lowest singlet and triplet states
3.1 Vertical and adiabatic excitation energies. Using the are summarized in Fig. 4 and compared to the TDDFT results
DSCF method, we obtained an optimal set of range-separation obtained for different parameter pairs from the minimum
parameters of (a = 0.00; u = 0.14) (set A) considering the valley. The points are placed on the graph according to the value
electronic ground state. Applying the DSCF method to different of the constant exact exchange (a); the respective u-values can
electronic states showed that while set A is a good compromise be found in Fig. 2. In addition, we show results obtained with
among electronic states of different characters, for individual B3LYP that uses only a constant amount of exchange of 0.2.
states, other range-separation parameters are optimal. Studies Before comparing the TDDFT and CASPT2 energies of the
of similar iron complexes typically recommended including corresponding MC and MLCT states, we note the inuence of
a certain portion of constant exchange in the functional, i.e., the range-separation parameters on the state characters. For
a s 0.54,56 Furthermore, for photoemission spectra of various this, we show the amount of contribution of MLCT and MC
compounds including copper phthalocyanine, it was shown congurations to the state vector by the red and blue boxes,
that tuned functionals with a > 0 led to better mitigation of respectively, in the upper and lower panels in Fig. 4. As can be
short-range one-electron self-interaction errors and therefore to seen, states assigned to MLCT (upper panel) possess a clear
a better description of spectra.29,61 Thus, it is interesting to dominant MLCT character of 60–80% throughout the different
investigate the inuence of the range-separation parameters on range-separation parameters tested. States of predominant MC
the static properties of [Fe(cpmp)2]2+. character (bottom panel) possess a more diverse mixture with
notable MLCT admixture. Upon increasing the exact exchange
a, however, the MC character increases leading to a clearer
assignment of both singlet and triplet states to MC ones,
highlighting the importance of exact exchange for MC states.
When calculating singlet excited states, CASPT2 predicts the
MC state (blue line) above the MLCT state (red line). This
situation is independent of the IPEA shi,62,63 as discussed in
more detail in Section S3.† The same state ordering as in
CASPT2 is found in TDDFT for a # 0.20, while the best
agreement between TDDFT and CASPT2 is reached for small
exact exchanges a # 0.10. For triplet excited states (right part of
Fig. 4), the situation is different: while CASPT2 predicts the MC
state (blue line) below the MLCT state (red line), this state
ordering is only found for larger exact exchanges a $ 0.20 in
TDDFT calculations, i.e., opposite to the singlet states. Only for
one tested range-separation parameter set, (0.2; 0.08), the
Fig. 3 Changes of adiabatic and vertical (S1 and T1) energies for all order of both singlet and triplet lowest states in the TDDFT
lowest excited states in the whole range of a and u parameters. calculations agrees with CASPT2 results. In consequence, these

1494 | Chem. Sci., 2023, 14, 1491–1502 © 2023 The Author(s). Published by the Royal Society of Chemistry
Edge Article Chemical Science

Fig. 4 Energies of the lowest MC and MLCT states in the singlet (left panel) and triplet (right panel) manifolds predicted by various LC-BLYP
variants, as well as B3LYP, and CASPT2 reference. Additionally, the energies of the corresponding states calculated with tuned LC-BLYP
according to triplet tuning and MC tuning procedures. The percentage of the main character of the states is provided in upper (MLCT) and lower
(MC) parts of both panels. Note the different energy scales of both main panels: the triplet energies are less susceptible to changes in optimal
parameters.

parameters, marked in the subsequent discussion as set B, also 3.3 Comparison to experimental absorption spectrum
provide the best agreement between CASPT2 and TDDFT While the DSCF approach is a non-empirical method for tuning
excitation energies with differences of 0.1–0.3 eV. Meanwhile
functionals, experimental results can also provide valuable
the (0.20; 0.08) parameter set B is also located in the minimal
valley of optimal tuning (see Fig. 2), but its values differ notably
from the optimal-tuned set A of (0.00; 0.14). To describe
spectroscopic observables, three essential but not equivalent
criteria for the functional should be considered: piecewise
linearity, freedom from self-interaction, and an exact
asymptotic potential. While the rst and third conditions are
fullled by optimally tuned functionals by construction, the
second one is achieved only asymptotically.61 For this reason,
the local s-orbitals should suffer from uncompensated self-
interaction error more than delocalized p-orbitals.27,64 Thus,
within the numerical accuracy, among optimally tuned
functionals with comparably good fulllment of Koopmans'
and Janak’ theorems (rst and third conditions), those with
a portion of exact exchange in the short range allow for better
mitigation of self-interaction error and a more balanced
description and energy ordering of differently localized orbitals
and, thus, excited states of different nature.27,61,64 For both
selected sets of parameters A and B, the degree of localization of
orbitals is visually indistinguishable. However, the energetic
position of s orbitals is more sensitive to the portion of exact
exchange in the short range than those of p-orbitals, see
Fig. S2.†
In an effort to investigate further the inuence of the
different parameter sets, we analyze the consequences of
using both sets beyond the Franck–Condon region. First, we
evaluate the agreement of the absorption spectra with the
Fig. 5 Absorption spectrum of [Fe(cpmp)2]2+ in acetonitrile predicted
experimental data. Second, we investigate the behavior of the by various DFT variants as compared to experimental data. The (a, u)
potential energy curves along the important Fe–N bond length. pairs of parameters for tuned LC-BLYP are given in the legend.
And nally, we simulate the excited-state dynamics of the Theoretical spectra are shifted vertically according to the a-parameter.
complex in solution. The lower 4 panels show the density-matrix analysis of the
corresponding excited states computed with two sets of optimal
parameters.

© 2023 The Author(s). Published by the Royal Society of Chemistry Chem. Sci., 2023, 14, 1491–1502 | 1495
Chemical Science Edge Article

guidance for justifying the chosen computational scheme.25 As bohr$(a.m.u.)1/2) shi to lower energies. This result already
excited state properties are of particular interest for such indicates that both sets of (a, u) pairs could lead to different
complexes, the natural choice of the experimental reference is photodynamics due to the character of the lowest excited states,
the absorption spectra. Fig. 5 compares the experimental as it will be investigated next.
absorption spectrum34 against spectra computed with LC-BLYP
and combinations of (a, u) along the minimal valley in the
optimal-tuning plot as well as with the popular B3LYP 4 Performance of range-separation
functional. For set A and set B of optimal parameters, the parameters in nonadiabatic dynamics
transition-density-matrix analysis of singlet and triplet spectra
is also shown. 4.1 Electronic state populations
All theoretical spectra agree well with the experiment, with We showed that the selection of the range-separation
the rst feature around 2.0 eV being better reproduced by LC- parameters signicantly affects the nature of the lowest
BLYP with a smaller constant portion of the exact exchange a. excited states at the Franck–Condon geometry (Fig. 5) as well as
Upon increasing a, all three band maxima are shied to higher the PESs of the breathing mode of [Fe(cpmp)2]2+ (Fig. 6). We
energies, a behavior observed for bright local and charge- now investigate its effect on the photodynamics by using TSH
transfer states previously.48 MC states may exhibit an opposite nonadiabatic simulations on linear vibronic coupling (LVC)
behavior (see above), but usually they possess considerably parametrized potentials65,66 along the 213 vibrational degrees of
smaller oscillator strengths. The B3LYP and LC-BLYP (0.20; freedom of [Fe(cpmp)2]2+. The LVC potentials were set up for the
0.08) functionals with the same portion of global exact exchange range-separation parameters of set A and set B, including 11
give very similar spectra. As the experimental spectrum of singlet/20 triplet states and 9 singlet/14 triplet states,
[Fe(cpmp)2]2+ has rather broad bands without any ne respectively, corresponding to the number of electronic states
structure, it leaves some ambiguity in identifying the best in the energy range of the lowest-energy absorption band (up
computational scheme for [Fe(cpmp)2]2+. to 2.5 eV, Fig. 5). Full computational details are reported in
Section S1.3†.
3.4 Potential energy surfaces The resulting TSH dynamics in the LVC electronic potentials
from both sets of range-separation parameters is shown in
Based on the analysis of Huang-Rhys factors, we identied that
Fig. 7 for the rst 500 fs. We performed SH dynamics for a total
the most important tuning mode corresponds to the Fe–N
of 2 ps (Fig. S5†), but during the rst 500 fs, the dynamics
breathing mode (ground-state frequency is 210 cm−1).
reached a steady state (for set A) or ended all previously started
Therefore, we obtained one-dimensional potential energy
processes (for set B).
proles for this mode for the two sets of range-separation
The time evolution of the adiabatic state populations is
parameters A and B. Interestingly, there is a prominent
shown in Fig. 7(a and b). For set A, the dynamics starts by
difference between both sets of obtained energy proles, as
exciting [Fe(cpmp)2]2+ into higher-lying singlet states SN (N $ 2,
shown in Fig. 6. Most notably, the MLCT states (those with
orange curve), from which most of the population undergoes
energy minima near Q = 0.0) shi to higher energies for the
ultrafast intersystem crossing to higher-lying triplet states TN (N
larger a, whereas the MC states (with a minimum at Q ∼ 0.8
$ 3, violet curve). Within the triplet manifold, further relaxation
to T2 (light–blue curve) and T1 (dark–blue curve) occurs on sub-
100 fs time scales. Aer ca. 300 fs, the population is stable in
the T1, T2, and TN states. As a minor reaction channel,
the population initially decays from the SN to the S1 state
(red curve) on a 2 ps time scale, from where also the TN states
are reached.
By contrast, the dynamics employing parameters of set B is
very different. Despite mostly starting also in the higher-lying
singlet states SN, the adiabatic population bifurcates in a 2 : 1
ratio. The smaller portion decays to the S1 state – also initially
populated by 15% – from where the whole singlet population
relaxes to the adiabatic ground state S0 (green curve), all within
a ca. 100 fs time scale. The majority of the SN population
undergoes intersystem crossing to higher-lying triplet states TN
(N $ 2), which rapidly reach the T1 state. From the T1 state, the
system undergoes back-intersystem crossing to the adiabatic
ground-state S0, however, at a slower rate on a sub-1 ps time
Fig. 6 Potential energy sections along the breathing mode computed
scale. This adiabatic S0 state does not correspond to the diabatic
with optimally tuned LC-BLYP with different range-separation
parameters (a, u) left-hand side: set A (0.0; 0.14); right-hand side: set B S0 state, i.e., the S0 state in the Franck–Condon geometry, at
(0.20; 0.08). Blue solid lines denote singlet states while red dashed least at early simulation times, as discussed in more detail in
curves show triplet states. Section S4.6.†

1496 | Chem. Sci., 2023, 14, 1491–1502 © 2023 The Author(s). Published by the Royal Society of Chemistry
Edge Article Chemical Science

Fig. 7 (a and b) Time evolution of adiabatic electronic state populations (thin lines) and fits based on the mechanisms shown in (c/d) from linear-
vibronic coupling/surface hopping (LVC/SH) dynamics simulations. Higher-lying singlet states SN are combined into one line for N $ 2 and
higher-lying triplet TN into one line for N $ 3 (set A) and N $ 2 (set B). (c and d) Mechanisms of the nonadiabatic dynamics in the basis of adiabatic
electronic states. (e and f) Time evolution of diabatic electronic state populations based on a transition-density matrix analysis.

4.2 Character of the electronic states shown in Fig. 7(f). As can be seen, in addition to the initial 1MLCT
The two sets of range-separation parameters lead to character (orange curve, 55%), there is already a substantial
qualitatively different relaxation mechanisms; see Fig. 7(c and portion of 1MC character (red curve, 25%) in the electronic state
wave functions of the trajectories. Both characters decrease by
d) and also Section S4.2† for a more-detailed derivation of the
about half their initial amounts, and contribution is transferred
mechanism. Most notably, set A leads to the population of long-
with almost equal shares to states of 3MLCT (dark blue) and 3MC
lived triplet states stable within our simulation window, while
(violet) characters. The triplet characters reach a maximum aer
set B leads to deactivation to the ground-state S0 via both the
ca. 100 fs. At later simulation times, the 3MLCT and 3MC
singlet and the triplet manifolds. In order to understand the
characters in the population slowly decrease again, while 1MLCT
different excited-state behaviors, we analyzed the character of
the electronic states involved in the dynamics in a diabatic and 1MC characters rise. Aer 300 fs, a notable increase in the
representation using the transition-density matrix analysis of diabatic S0 character (green curve) is visible.
The initial population and subsequent depopulation of diabatic
the electronic states in the Franck–Condon geometry as 3
MLCT and 3MC states mirror the ultrafast SN to TN intersystem
a reference. The corresponding time evolution of the diabatic
crossing and TN to S0 back-intersystem crossing established for the
state populations is shown in Fig. 7(e and f). Using set A, the
adiabatic electronic states in Fig. 7(d). Interestingly, the population
initially excited states are mainly (74%) of 1MLCT character
of the diabatic S0 population is much smaller than the adiabatic S0
(orange curve in Fig. 7(e)). The remaining character is evenly
population – compare Fig. 7(b) and (f). These states are dened in
distributed over the other singlet excitation characters, i.e.,
1
MC, 1LMCT (ligand-to-metal charge transfer), 1LC (ligand- different manners, and thus, their populations may differ. The
centered), and 1LLCT (ligand-to-ligand charge-transfer). In the adiabatic S0 state collects all states that are the lowest-energy
singlet state in their respective geometry. In contrast, the
adiabatic representation, we observed an ultrafast intersystem
diabatic S0 state refers to the singlet state with the exact electronic
crossing into the triplet manifold. This analysis in the diabatic
character as the adiabatic S0 at the Franck–Condon geometry, and
representation reveals that the populated triplet states are of
the diabatic S0 populations collect their components in the
predominant 3MLCT character (dark blue curve) of up to 55%
trajectories during the dynamics. The difference between the
aer 500 fs. The second largest contribution to the triplet states
adiabatic and diabatic S0 populations can be understood by
is given by 3MC congurations (violet curve) with 15% aer 500
fs. The admixture of 3MLCT and 3MC characters in the diabatic investigating the behavior of the trajectories undergoing
state population is due to state mixing rather than the system deactivation within the singlet manifold (Section S4.6)†.
traversing two different pathways, as is discussed in Section
S4.4 in the ESI†. 4.3 Nuclear motion
The time evolution of the diabatic state populations of the
dynamics using the range-separation parameters of set B is We analyze here the nuclear motion during the nonadiabatic
dynamics simulations, focusing on the motion involving the

© 2023 The Author(s). Published by the Royal Society of Chemistry Chem. Sci., 2023, 14, 1491–1502 | 1497
Chemical Science Edge Article

By contrast, in the simulation using set B, large

displacements of the average Fe–N bond lengths along the
trajectories are visible with respect to their Franck–Condon
reference (Fig. 8(b)). This observation is the result of populating
1
MC and 3MC states. The MC states of closed-shell iron(II)
compounds involve the excitation of an electron from a non-
bonding t2g orbital to an antibonding e*g orbital, which
directly affects the metal–ligand bond lengths. The changes in
the Fe–N bond lengths are up to 0.10–0.15 Å, a typical value for
hexaaza iron(II) compounds.67 Additionally, also during the
intersystem crossing, some Fe–N bonds display pronounced
differences of up to 0.05 Å compared to the Franck–Condon
values. These differences are likely due to the admixture of MC
character to the set B trajectories (recall Fig. 7(f)). In general,
there is substantial motion required to reach intersystem
crossing geometries as is shown by the superposition of all
intersystem crossing geometries of the set B simulation in
Fig. 8(c).
Finally, for set B, we also show the average Fe–N bond
lengths of the geometries, where S1/S*0 transitions occur (orange
lines) in Fig. 8(b). The average Fe–N distances differ to a similar
extent to the values at the intersystem crossing geometries from
their Franck–Condon references. The large variation in the
geometries in this set is also displayed as a superposition of
structures in Fig. 8(d). This highlights the large extent of the
region on the potential energy surface, where polyatomic
molecules can undergo transitions between electronic states.
Note that no average Fe–N bond lengths are shown for the set A
simulations in Fig. 8(a), as the corresponding dynamics did not
exhibit any S1/S*0 transitions. This absence is a direct result of
the different shapes of the potential energy curves obtained
with the different range-separation parameters (Fig. 6).
Fig. 8 (a and b) Time evolution of the average distances Fe–N
between the central iron and the ligating nitrogen atoms from SH/LVC
4.4 Relaxation mechanisms
trajectories (blue curves) compared to the distances in the Franck–
Condon (FC) geometry (red lines), distances in the ISC hopping As we have seen, the two sets of parameters lead to two different
geometries (green lines), and distances in the S1/ S*0 geometries relaxation mechanisms, intimately connected to the character
(orange curves, only set B). Green lines in the top and bottom panels in
of the lowest-excited states originating from the different range-
(b) are hidden behind the orange lines. (c and d) Superposition of
geometries (thin bonds) from the ISC hopping points (c) and S1/S*0 separation parameters. Using set A produces low-lying 3MLCT
geometries (d) on top of the Franck–Condon geometry (thick bonds). states, which leads to long-lived triplet states, while set B leads
to dynamics dominated by re-population of a hot S*0 state
through the singlet manifold as well as the diabatic S0 state via
central iron atom and the directly coordinating ligand atoms. the triplet manifold, both via states involving MC character. The
The evolution of the six different Fe–N bond lengths during the obvious question now is which of the two models describes the
dynamics is displayed in Fig. 8a and b (blue curves), along with behavior of [Fe(cpmp)2]2+ correctly? In order to solve this
the reference bond length in the Franck–Condon geometry (red problem, we leverage transient absorption spectroscopy (TAS).
line). For set A, the average Fe–N bond length stays close to the Our recorded TA spectra (see Section S6†) reveal a single
reference values in the Franck–Condon geometry (Fig. 8(a)). monoexponential decay of the ground-state bleach and, thus,
This behaviour is a direct consequence of the 3MLCT electronic the recovery of the ground state with a time constant of sA = 550
states that control this simulation. These states involve an ps. A similar long-lived signal with s3 = 472 ps is observed in
excitation from a non-bonding t2g orbital of the d6 iron related TAS experiments of [Fe(cpmp)2]2+ in ref. 34. Those
conguration to the p system of the organic ligands, leaving the experiments34 report two additional time constants for the
bonding between the iron center and its ligating nitrogen atoms excited-state decay, i.e., s1 < 200 fs and s2 = 30 − 40 ps. The
unaffected. In addition, Fig. 8(a) shows the average Fe–N bond ultrafast s1 could, in principle, also contribute to the TA signal
distances in the geometries where intersystem crossing occurs in our experiments, which have a time resolution of ca. 100 fs.
(green line). As can be seen, also during intersystem crossing, However, due to a limited signal-to-noise ratio in our
the Fe–N bond lengths remain nearly constant. experiments, a weak sub-200 fs component might not be

1498 | Chem. Sci., 2023, 14, 1491–1502 © 2023 The Author(s). Published by the Royal Society of Chemistry
Edge Article Chemical Science

resolved, in particular, since it is most clearly seen in the near- give good agreement with the experimental absorption spectrum.
infrared region, which is outside the detection range of our More importantly, however, set B provided the only range-
setup. In ref. 34, the s1 component is associated with small separation parameters that predicted the lowest-energy states
spectral changes and ascribed to the lifetime of the 3MLCT state in the same order as multi-reference CASPT2 calculations.
due to the decay into MC states. Note that it was not possible to The mechanism of set B simulations of relaxing via 3MC and
spectroscopically differentiate between triplet and quintet MC possibly 5MC states shows that inserting a bridge between
states. The longer time-scale process described by s2 associated neighboring pyridyl units does not change the deactivation
with the near-infrared region in ref. 34 is outside of the processes found for iron(II)polypyridyl complexes.68–75 Thus, the
present detection range. It is ascribed to cooling and solvent introduction of push–pull ligands results only in a small
reorganization processes when the system is already in MC increase of ligand-eld splitting. Ligand-eld splitting large
states. Finally, ref. 34 ascribes the long-lived signal to the enough to destabilize the MC states to the extent that long-lived
3
lifetime of the lowest-energy MC state, which is in line with our MLCT states can be populated seems to require strong-s
transient absorption spectra (see Section S6†). donating ligands such as NHCs.72,76–80 The [Fe(cpmp)2]2+
At rst glance, the experimental time constants appear complex falls into the line of [Fe(dcpp)(ddpd)]2+, [Fe(dcpp)2]2+
to agree with the results based on set A, where ultrafast and [Fe(ddpd)2]2+ complexes,81 where dcpp and ddpd are
intersystem crossing from the initially excited singlet states and electron-withdrawing and electron-donating tri-pyridyl ligands
subsequent internal conversion populate T1 states of bridged with two CO or NMe units, respectively. Among these
predominant 3MLCT character on a ∼200 fs time scale. The complexes, [Fe(dcpp)(ddpd)]2+ and [Fe(dcpp)2]2+ also showed
T1/3MLCT state is then stable on our ps simulation time. This relaxation to the 5MC states aer photoexcitation. In contrast,
result is seemingly in line with the TAS observation, which whereas in [Fe(ddpd)2]2+ 3MC states were populated, no sign of
shows that the electronic ground state is repopulated within further dynamics into the quintet manifold was seen
only a 500 ps time scale. However, the rapid disappearance of experimentally.81 [Fe(cpmp)2]2+ thereby behaves more similar to
the MLCT signal in ref. 34 suggests that the long-lived state [Fe(dcpp)(ddpd)]2+ and [Fe(dcpp)2]2+, which also feature a CO-
cannot be of MLCT nature, and, thus has to be an MC state. This bridged dcpp ligand, than to the [Fe(ddpd)2]2+ complex
nding contradicts the results of the set A simulation. featuring NMe-bridged ligands. Thus, one can conclude that
The set B simulations also predict fast initial dynamics a more promising design concept to increase ligand-eld
involving intersystem crossing as well as relaxation dynamics splitting in pyridyl-iron(II) complexes is to introduce
within the singlet manifold that can account for the s1 only electron-donating ligands (ddpd) rather than electron-
component. Notably, the simulations of set B feature more withdrawing ligands (dcpp) or combinations that establish
pronounced MC character in the low-lying populated electronic push–pull systems (cpmp or dcpp + ddpd).
states. While the set B simulation shows relaxation to the
lowest-energetic state on a picosecond time scale, this process
does not correspond to the full deactivation of the system. 5 Conclusions and outlook
Instead, in the singlet pathway, a hot S*0 state is populated that
still shares the character of excited singlet states. Furthermore, The theoretical study of the photodynamics of transition-metal
this state stays close in energy to other excited states. Similar complexes is an intricate task. In particular, complexes bearing
behavior is observed for trajectories relaxing via the triplet organic ligands with extended p-systems feature electronic
manifold. At this point, one may wonder about the role that states of very different nature, such as MC and MLCT states,
quintet states play in the excited-state dynamics. Unfortunately, that can be both accessible during excited-state dynamics
it is not possible to include quintet states in surface hopping simulations and need a reliable prediction. This complexity
studies with the presently used setup, as quintet states are not applies not only to their relative energies, but importantly also
accessible from singlet ground states using linear-response to their electronic character that ultimately controls the excited-
TDDFT and, furthermore, spin–orbit couplings involving state dynamics.
quintet states for TDDFT have also not been implemented yet. Here, we assessed the performance of optimally tuned range-
Because of the behavior described above, trajectories in either separated DFT for a prototypical iron compound, [Fe(cpmp)2]2+
singlet or triplet states which remain for extended times in that, having push–pull ligands, exhibits an increased lifetime of
regions of a high density of electronic states could traverse to the MLCT states compared to simple iron polypyridyl complexes
the quintet manifold and end up in low-lying 5MC states. Such such as [Fe(bpy)3]2+. We systematically employed different
states are the lowest-lying excited states, as shown in ref. 34, and tuning schemes, including “classical” ground-state
they have been assigned to the experimentally observed long- fundamental gap tuning, ionization potential tuning in the
lived signal. triplet and quintet excited states, electron affinity tuning
Thus, we conclude that set B represents the best choice to targeting metal-localized s* orbitals, and the so-called triplet
simulate the dynamics of [Fe(cpmp)2]2+. While it did not tuning. We found that the range-separation parameters strongly
represent the optimal choice from the DSCF tuning (set A), the inuence vertical and adiabatic excitation energies and,
range-separation parameters from set B also fall in the minimal importantly, also the nature of the lowest excited states. Thus,
valley in DSCF tuning, thus indicating that it is a good parameter an unambiguous choice of the tuning parameters for such
choice from an ab initio point of view. Furthermore, set B could molecules is not straightforward.

© 2023 The Author(s). Published by the Royal Society of Chemistry Chem. Sci., 2023, 14, 1491–1502 | 1499
Chemical Science Edge Article

Aided by comparison with the experimental absorption performed current and preliminary dynamics simulations, and
spectra and CASPT2 calculations, we found two suitable sets of AK recorded the transient-absorption spectra. OSB, JPZ, and SIB
range-separation parameters that were subsequently used for wrote the manuscript. All authors reviewed and edited the
excited-state dynamics simulations. Using the set of optimal paper.
parameters from the DSCF method (set A) for dynamics simu-
lations results in the population of 3MLCT states that are stable
within our 2 ps simulation time. This result is in contrast to TAS Conflicts of interest
experiments,34 which show the depopulation of MLCT states There are no conicts to declare.
within 200 fs. Using the set which is in agreement with multi-
reference CASPT2 calculations (set B) leads to relaxation
processes within both the singlet and triplet manifolds. The Acknowledgements
participating electronic states in this deactivation mechanism
show pronounced MC character. Throughout the deactivation – This work has been nancially supported by the Deutsche
and even upon reaching the hot S*0 state – the system stays in Forschungsgemeinscha [Priority Program SPP 2102 “Light-
regions of a high density of states, suggesting the possibility for controlled reactivity of metal complexes” (Grants KU952/12-1,
further transfer into quintet states. We thus conclude that only GO 1059/8-2, and LO 714/11-2).
set B, in agreement with multireference calculations, can
describe the electronic states of [Fe(cpmp)2]2+ well. Notes and references
In summary, we illustrate the intricacies of choosing a
suitable DFT computational protocol to model the photody- 1 C. Stephenson and T. Yoon, Acc. Chem. Res., 2016, 49, 2059–
namics of a transition metal complex. Optimal tuning of 2060.
a range-separated DFT functional is not a black-box method,30 2 V. Venkatraman, R. Raju, S. P. Oikonomopoulos and
and the nature of the involved states is unpredictable. The large B. K. Alsberg, J. Cheminf., 2018, 10, 18.
difference in the predicted photodynamics resulting from the 3 R. D. Costa, E. Ortı́, H. J. Bolink, F. Monti, G. Accorsi and
different range-separation parameters can be ascribed to the N. Armaroli, Angew. Chem., Int. Ed., 2012, 51, 8178–8211.
different response of MC vs. MLCT states with changing the 4 H. Junge, N. Rockstroh, S. Fischer, A. Brückner, R. Ludwig,
portion of exact-exchange in the short range. The local S. Lochbrunner, O. Kühn and M. Beller, Inorganics, 2017, 5,
s-orbitals are more suspected to the self-interaction error than 14.
delocalized p-orbitals and, thus, the inclusion of a certain 5 O. S. Wenger, Chem.–Eur. J., 2019, 25, 6043–6052.
portion of global exact exchange leads to a better mitigation of 6 G. Vankó, A. Bordage, M. Pápai, K. Haldrup, P. Glatzel,
self-interaction error.27,61,64 This observation hints towards the A. M. March, G. Doumy, A. Britz, A. Galler, T. Assefa,
importance of a balanced approach. Testing the performance of D. Cabaret, A. Juhin, T. B. van Driel, K. S. Kjær, A. Dohn,
range-separation parameters for describing the excited states of K. B. Møller, H. T. Lemke, E. Gallo, M. Rovezzi, Z. Németh,
transition-metal complexes only for the properties of the E. Rozsályi, T. Rozgonyi, J. Uhlig, V. Sundström,
(bright) MLCT states, e.g., by solely considering absorption M. M. Nielsen, L. Young, S. H. Southworth, C. Bressler and
spectra, will overlook the inuence of (dark) MC states, that can W. Gawelda, J. Phys. Chem. C, 2015, 119, 5888–5902.
drive the outcome of photodynamics as we observed for 7 C. Bressler, C. Milne, V.-T. Pham, A. ElNahhas, R. M. van der
[Fe(cpmp)2]2+. In general, the combination of experimental Veen, W. Gawelda, S. Johnson, P. Beaud, D. Grolimund,
absorption spectra and photorelaxation timescales, multi- M. Kaiser, C. N. Borca, G. Ingold, R. Abela and M. Chergui,
reference calculations, and nonadiabatic dynamics Science, 2009, 323, 489–492.
simulations allowed the identication of a suitable set of tuning 8 I. M. Dixon, G. Boissard, H. Whyte, F. Alary and J.-L. Heully,
parameters. Fortunately, such all-encompassing endeavors are Inorg. Chem., 2016, 55, 5089–5091.
possible nowadays, as time-resolved spectroscopic experiments 9 D. C. Ashley and E. Jakubikova, Coord. Chem. Rev., 2017, 337,
have become main-stream characterization techniques of 97–111.
transition-metal complexes and nonadiabatic dynamics 10 P. Chábera, L. Lindh, N. W. Rosemann, O. Prakash, J. Uhlig,
methods have evolved to be computationally feasible. A. Yartsev, K. Wärnmark, V. Sundström and P. Persson,
Coord. Chem. Rev., 2021, 426, 213517.
Data availability 11 J. P. Zobel and L. González, JACS Au, 2021, 1, 1116–1140.
12 Quantum Chemistry and Dynamics of Excited States: Methods
The parameter for the linear vibronic coupling used in the and Applications, ed. L. González and R. Lindh, John Wiley
dynamics simulations are given in the ESI.† All other data is & Sons, 2021.
available from the corresponding authors upon reasonable request. 13 R. Crespo-Otero and M. Barbatti, Chem. Rev., 2018, 118,
7026–7068.
Author contributions 14 A. D. Laurent and D. Jacquemin, Int. J. Quantum Chem., 2013,
113, 2019–2039.
LG, SIB, and OK conceived the research and acquired funding. 15 A. Dreuw and M. Head-Gordon, Chem. Rev., 2005, 105, 4009–
OSB performed all optimal-tuning calculations, JPZ and OB 4037.

1500 | Chem. Sci., 2023, 14, 1491–1502 © 2023 The Author(s). Published by the Royal Society of Chemistry
Edge Article Chemical Science

16 M. J. G. Peach, P. Beneld, T. Helgaker and D. J. Tozer, J. G. Vankó, C. Bressler and J. K. McCusker, Inorg. Chem.,
Chem. Phys., 2008, 128, 44118. 2019, 58, 9341–9350.
17 A. Savin, in Theoretical and Computational Chemistry, 39 J. Moll, C. Förster, A. König, L. M. Carrella, M. Wagner,
Elsevier, 1996, vol. 4, pp. 327–357. M. Panthöfer, A. Möller, E. Rentschler and K. Heinze,
18 H. Iikura, T. Tsuneda, T. Yanai and K. Hirao, J. Chem. Phys., Inorg. Chem., 2022, 61, 1659–1671.
2001, 115, 3540–3544. 40 L. L. Jamula, A. M. Brown, D. Guo and J. K. McCusker, Inorg.
19 R. Baer, E. Livshits and U. Salzner, Annu. Rev. Phys. Chem., Chem., 2014, 53, 15–17.
2010, 61, 85–109. 41 T. Stein, L. Kronik and R. Baer, J. Am. Chem. Soc., 2009, 131,
20 P. Mori-Sánchez, A. J. Cohen and W. Yang, Phys. Rev. Lett., 2818–2820.
2008, 100, 146401. 42 T. Stein, L. Kronik and R. Baer, J. Chem. Phys., 2009, 131,
21 E. Livshits and R. Baer, Phys. Chem. Chem. Phys., 2007, 9, 244119.
2932–2941. 43 G. A. Petersson, A. Bennett, T. G. Tensfeldt, M. A. Al-Laham,
22 J. Autschbach and M. Srebro, Acc. Chem. Res., 2014, 47, 2592– W. A. Shirley and J. Mantzaris, J. Chem. Phys., 1988, 89, 2193–
2602. 2218.
23 S. Refaely-Abramson, R. Baer and L. Kronik, Phys. Rev. B: 44 G. A. Petersson and M. A. Al-Laham, J. Chem. Phys., 1991, 94,
Condens. Matter Mater. Phys., 2011, 84, 075144. 6081–6090.
24 M. P. Borpuzari and R. Kar, J. Comput. Chem., 2017, 38, 2258– 45 A. D. Becke, J. Chem. Phys., 1993, 98, 5648–5652.
2267. 46 F. Weigend and R. Ahlrichs, Phys. Chem. Chem. Phys., 2005,
25 T. Möhle, O. S. Bokareva, G. Grell, O. Kühn and S. I. Bokarev, 7, 3297.
J. Chem. Theory Comput., 2018, 14, 5870–5880. 47 F. Weigend, Phys. Chem. Chem. Phys., 2006, 8, 1057.
26 K. Hirao, H.-S. Bae, J.-W. Song and B. Chan, J. Phys. Chem. A, 48 S. I. Bokarev, O. S. Bokareva and O. Kühn, J. Chem. Phys.,
2021, 125, 3489–3502. 2012, 136, 214305.
27 S. Refaely-Abramson, S. Sharifzadeh, N. Govind, 49 O. S. Bokareva, G. Grell, S. I. Bokarev and O. Kühn, J. Chem.
J. Autschbach, J. B. Neaton, R. Baer and L. Kronik, Phys. Theory Comput., 2015, 11, 1700–1709.
Rev. Lett., 2012, 109, 226405. 50 S. I. Bokarev, O. S. Bokareva and O. Kühn, Coord. Chem. Rev.,
28 T. Körzdörfer, J. S. Sears, C. Sutton and J.-L. Brédas, J. Chem. 2015, 304–305, 133–145.
Phys., 2011, 135, 204107. 51 S. Fischer, O. S. Bokareva, E. Barsch, S. I. Bokarev, O. Kühn
29 D. A. Egger, S. Weissman, S. Refaely-Abramson, and R. Ludwig, ChemCatChem, 2016, 8, 404–411.
S. Sharifzadeh, M. Dauth, R. Baer, S. Kümmel, 52 O. Bokareva, T. Möhle, A. Neubauer, S. Bokarev,
J. B. Neaton, E. Zojer and L. Kronik, J. Chem. Theory S. Lochbrunner and O. Kühn, Inorganics, 2017, 5, 23.
Comput., 2014, 10, 1934–1952. 53 A. Friedrich, O. S. Bokareva, S.-P. Luo, H. Junge, M. Beller,
30 A. Karolewski, L. Kronik and S. Kümmel, J. Chem. Phys., O. Kühn and S. Lochbrunner, Chem. Phys., 2018, 515, 557–
2013, 138, 204115. 563.
31 A. Ramasubramaniam, D. Wing and L. Kronik, Phys. Rev. 54 G. Prokopiou and L. Kronik, Chem.–Eur. J., 2017, 5173–5182.
Mater., 2019, 3, 084007. 55 Z. Lin and T. Van Voorhis, J. Chem. Theory Comput., 2019, 15,
32 M. Brütting, H. Bahmann and S. Kümmel, J. Chem. Phys., 1226–1241.
2022, 156, 104109. 56 O. S. Bokareva, O. Baig, M. J. Al-Marri, O. Kühn and
33 J. Moll, C. Wang, A. Päpcke, C. Förster, U. Resch-Genger, L. González, Phys. Chem. Chem. Phys., 2020, 22, 27605–27616.
S. Lochbrunner and K. Heinze, Chem.–Eur. J., 2020, 26, 57 P. Dierks, A. Päpcke, O. S. Bokareva, B. Altenburger,
6820–6832. T. Reuter, K. Heinze, O. Kühn, S. Lochbrunner and
34 J. Moll, R. Naumann, L. Sorge, C. Förster, N. Gessner, M. Bauer, Inorg. Chem., 2020, 59, 14746–14761.
L. Burkhardt, N. Ugur, W. Nürnberger, P. Seidel, 58 J. P. Zobel, O. S. Bokareva, P. Zimmer, C. Wölper, M. Bauer
C. Ramanan, M. Bauer and K. Heinze, Chem.–Eur. J., 2022, and L. González, Inorg. Chem., 2020, 59, 14666–14678.
28, e2022ß1858. 59 J. P. Perdew, R. G. Parr, M. Levy and J. L. Balduz, Phys. Rev.
35 O. S. Wenger, Chem.–Eur. J., 2019, 25, 6043–6052. Lett., 1982, 49, 1691–1694.
36 A. K. C. Mengel, C. Förster, A. Breivogel, K. Mack, 60 M. Dauth, F. Caruso, S. Kümmel and P. Rinke, Phys. Rev. B,
J. R. Ochsmann, F. Laquai, V. Ksenofontov and K. Heinze, 2016, 93, 121115.
Chem.–Eur. J., 2015, 21, 704–714. 61 L. Kronik and S. Kümmel, Phys. Chem. Chem. Phys., 2020, 22,
37 A. K. C. Mengel, C. Bissinger, M. Dorn, O. Back, C. Förster 16467–16481.
and K. Heinze, Chem.–Eur. J., 2017, 23, 7920–7931. 62 G. Ghigo, B. O. Roos and P.-Å. Malmqvist, Chem. Phys. Lett.,
38 A. Britz, W. Gawelda, T. A. Assefa, L. L. Jamula, 2004, 396, 142–149.
J. T. Yarranton, A. Galler, D. Khakhulin, M. Diez, 63 J. P. Zobel, J. J. Nogueira and L. González, Chem. Sci., 2017, 8,
M. Harder, G. Doumy, A. M. March, E. Bajnóczi, 1482–1499.
Z. Németh, M. Pápai, E. Rozsályi, D. Sárosiné Szemes, 64 F. Rissner, D. A. Egger, A. Natan, T. Körzdörfer, S. Kümmel,
H. Cho, S. Mukherjee, C. Liu, T. K. Kim, R. W. Schoenlein, L. Kronik and E. Zojer, J. Am. Chem. Soc., 2011, 133, 18634–
S. H. Southworth, L. Young, E. Jakubikova, N. Huse, 18645.

© 2023 The Author(s). Published by the Royal Society of Chemistry Chem. Sci., 2023, 14, 1491–1502 | 1501
Chemical Science Edge Article

65 F. Plasser, S. Gómez, M. F. S. J. Menger, S. Mai and M. M. Nielsen, L. Young, S. H. Southworth, C. Bressler and
L. González, Phys. Chem. Chem. Phys., 2019, 21, 57–69. W. Gawelda, J. Phys. Chem. C, 2015, 119, 5888–5902.
66 J. P. Zobel, M. Heindl, F. Plasser, S. Mai and L. González, Acc. 75 S. M. Fatur, S. G. Shepard, R. F. Higgins, M. P. Shores and
Chem. Res., 2021, 54, 3760–3771. N. H. Damrauer, J. Am. Chem. Soc., 2017, 139, 4493–4505.
67 A. Cannizzo, C. J. Milne, C. Consani, W. Gawelda, 76 L. Liu, T. Duchanois, T. Etienne, A. Monari, M. Beley,
C. Bressler, F. van Mourik and M. Chergui, Coord. Chem. X. Assfeld, S. Haacke and P. C. Gros, Phys. Chem. Chem.
Rev., 2010, 254, 2677–2686. Phys., 2016, 18, 12550–12556.
68 A. Moguilevski, M. Wilke, G. Grell, S. I. Bokarev, S. G. Aziz, 77 L. A. Fredin, M. Pápai, E. Rozsályi, G. Vankó, K. Wärnmark,
N. Engel, A. A. Raheem, O. Kühn, I. Y. Kiyan and E. F. Aziz, V. Sundström and P. Persson, J. Phys. Chem. Lett., 2014, 5,
ChemPhysChem, 2017, 18, 465–469. 2066–2071.
69 W. Zhang, R. Alonso-Mori, U. Bergmann, C. Bressler, 78 P. Chábera, K. S. Kjær, O. Prakash, A. Honarfar, Y. Liu,
M. Chollet, A. Galler, W. Gawelda, R. G. Hadt, L. A. Fredin, T. C. B. Harlang, S. Lidin, J. Uhlig,
R. W. Hartsock, T. Kroll, K. S. Kjær, K. Kubiček, V. Sundström, R. Lomoth, P. Persson and K. Wärnmark, J.
H. T. Lemke, H. W. Liang, D. A. Meyer, M. M. Nielsen, Phys. Chem. Lett., 2018, 9, 459–463.
C. Purser, J. S. Robinson, E. I. Solomon, Z. Sun, 79 Y. Liu, K. S. Kjær, L. A. Fredin, P. Chábera, T. Harlang,
D. Sokaras, T. B. van Driel, G. Vankó, T.-C. Weng, D. Zhu S. E. Canton, S. Lidin, J. Zhang, R. Lomoth,
and K. J. Gaffney, Nature, 2014, 509, 345–348. K.-E. Bergquist, P. Persson, K. Wärnmark and
70 S. G. Shepard, S. M. Fatur, A. K. Rappé and N. H. Damrauer, V. Sundström, Chem.–Eur. J., 2015, 21, 3628–3639.
J. Am. Chem. Soc., 2016, 138, 2949–2952. 80 H. Tatsuno, K. S. Kjær, K. Kunnus, T. Harlang, C. Timm,
71 M. C. Carey, S. L. Adelman and J. K. McCusker, Chem. Sci., M. Guo, P. Chábera, L. A. Fredin, R. W. Hartsock,
2019, 10, 134–144. M. E. Reinhard, S. Koroidov, L. Li, A. A. Cordones,
72 T. Duchanois, T. Etienne, C. Cebrián, L. Liu, A. Monari, O. Gordivska, O. Prakash, Y. Liu, M. G. Laursen, E. Biasin,
M. Beley, X. Assfeld, S. Haacke and P. C. Gros, Eur. J. Inorg. F. B. Hansen, P. Vester, M. Christensen, K. Haldrup,
Chem., 2015, 2469–2477. Z. Németh, D. S. Szemes, E. Bajnóczi, G. Vankó, T. B. Van
73 D. Leshchev, T. Harlang, L. A. Fredin, D. Khakhulin, Y. Liu, Driel, R. Alonso-Mori, J. M. Glownia, S. Nelson,
E. Biasin, M. G. Laursen, G. E. Newby, K. Haldrup, M. Sikorski, H. T. Lemke, D. Sokaras, S. E. Canton,
M. M. Nielsen, K. Wärnmark, V. Sundström, P. Persson, A. O. Dohn, K. B. Møller, M. M. Nielsen, K. J. Gaffney,
K. S. Kjær and M. Wulff, Chem. Sci., 2018, 9, 405–414. K. Wärnmark, V. Sundström, P. Persson and J. Uhlig,
74 G. Vankó, A. Bordage, M. Pápai, K. Haldrup, P. Glatzel, Angew. Chem., Int. Ed., 2020, 59, 364–372.
A. M. March, G. Doumy, A. Britz, A. Galler, T. Assefa, 81 A. K. C. Mengel, C. Förster, A. Breivogel, K. Mack,
D. Cabaret, A. Juhin, T. B. van Driel, K. S. Kjær, A. Dohn, J. R. Ochsmann, F. Laquai, V. Ksenofontov and K. Heinze,
K. B. Møller, H. T. Lemke, E. Gallo, M. Rovezzi, Z. Németh, Chem.–Eur. J., 2015, 21, 704–714.
E. Rozsályi, T. Rozgonyi, J. Uhlig, V. Sundström,

1502 | Chem. Sci., 2023, 14, 1491–1502 © 2023 The Author(s). Published by the Royal Society of Chemistry