Download as pdf or txt
Download as pdf or txt
You are on page 1of 44

CHAPTER SEVEN

Toward accurate spin-state


energetics of transition metal
complexes
*
Mariusz Radon
Faculty of Chemistry, Jagiellonian University, Krakow, Poland
*Corresponding author: e-mail address: mradon@chemia.uj.edu.pl

Contents
1. Introduction 222
2. Advances in methodology 223
2.1 Density functional theory methods 224
2.2 Wave function theory methods 227
3. Challenges in spin-state prediction 231
3.1 Porphyrins and heme-related models 231
3.2 The mystery of hexacyanomolybdate(IV) 237
3.3 Cobalt–nitrosyl complexes 240
4. Mechanistic importance 242
4.1 An example from enzyme studies 243
4.2 Role of spin states in ligand binding 245
5. Benchmarking and environmental effects 249
5.1 Benchmark study of aqua complexes 250
5.2 Other examples of environmental effects 256
6. Conclusions 258
Acknowledgments 259
References 259

Abstract
The ability to conclusively predict relative energies of different spin states for transition
metal complexes and metal sites in enzymes is highly relevant in (bio)inorganic chem-
istry, but poses an outstanding challenge for quantum chemical calculations. We discuss
representative applications of wave function theory (WFT) and density functional theory
(DFT) methods to compelling transition metal complexes and models of active sites in
metalloenzymes, aiming not only to resolve some existing controversies with spin-state
predictions, but also to develop reliable, yet efficient computational protocols for the
problem of spin-state energetics. The presented examples confirm that DFT results
are highly dependent on the choice of exchange–correlation functional and the opti-
mal choice is not universal, even when considering different spin-state gaps in the same
molecule. Mechanistic consequences of these issues are emphasized for spin-forbidden

Advances in Inorganic Chemistry, Volume 73 # 2019 Elsevier Inc. 221


ISSN 0898-8838 All rights reserved.
https://doi.org/10.1016/bs.adioch.2018.10.001
222 Mariusz Rado
n

ligand binding. Among the WFT methods, a high accuracy of the single-reference
coupled cluster CCSD(T) method is confirmed by a number of examples studied, includ-
ing systems with noticeable nondynamic correlation effects. However, we point out that
in some cases controversial results are obtained also from WFT calculations calling for
further benchmarking of quantum chemistry methods with respect to quantitative
experimental data of spin-state energetics. In this regard, the environmental (solvation
or crystal packing) effects on relative spin-state energetics must be accounted for when
comparing theory with experiment.

1. Introduction
Transition metal (TM) complexes with d4 to d7 configurations may
exist in either low-spin (LS) or high-spin (HS) state; for complexes, with
d5 or d6 configuration an intermediate-spin (IS) state is also possible. Differ-
ent spin states have different number of unpaired electrons on the metal
center and which spin state is energetically favored depends on a balance
between two counteracting factors: (a) exchange interactions (tending to
maximize the number of unpaired electrons) and (b) the splitting of the
metal d-orbital energy levels by the ligand field (the larger splitting favors
pairing of electrons in the lower energy levels). An important contribution
comes also from vibrational effects. Since metal–ligand stretching vibrations
have typically lower frequencies in the HS state than in the LS state, the HS
state has lower zero-point energy and higher entropy, contributing to the
free energy difference.1,2 Therefore, if the HS state is only slightly above
the LS state in terms of energy, the phenomenon of thermal spin-crossover
(SCO) may be observed.3
Although the above factors governing stabilization of different spin states
are qualitatively well understood, it is a computationally tough problem to
quantitatively predict spin-state energetics (i.e., relative energies of different
spin states). Usually, the “chemical accuracy” on the order of 1 kcal/mol
(or even higher) is required for studies of reaction mechanisms and conclu-
sive prediction of SCO properties. As will be illustrated by many examples
below, it is a great challenge to systematically achieve this level of accuracy in
quantum chemical calculations of spin-state energetics using both density
functional theory (DFT) and wave function theory (WFT) methods. The
importance of benchmark studies and developing optimized computational
protocols will be emphasized.
Accurate spin-state energetics 223

It must be stressed that accurate computation of the spin-state energetics


is highly relevant not only for prediction of the ground state, magnetic prop-
erties, and the possibility of SCO. It is even more important because
different spin states may give rise to different chemical reactivities4 and
ligand-activation propensities,5 and the ordering of spin states may flip
during chemical reactions. For spin-forbidden reactions,6 the energy contri-
bution related to the change of spin state may directly contribute to thermo-
chemical and/or kinetic parameters. Moreover, the unique possibility of
reversibly switching between different magnetic and structural properties
of TM complexes by simply changing their spin state is very attractive in
the context of materials science.7 Even in the context of geology, the
pressure-induced changes of the iron spin state seem to be an important
factor to understand a physical behavior of the lower mantle rocks.8 The
above examples confirm the relevance of spin-state isomerism in inorganic
chemistry and related fields.9,10 Adding to these the mentioned challenges in
accurate computational treatment of spin states by quantum chemistry
methods (which are related to general difficulties in computational modeling
of TM complexes and still not fully understood), it becomes evident that the
issue of spin-state energetics is of central importance in computational (bio)
inorganic chemistry.
This chapter is aimed to selectively review state of the art in accurate
computation of spin-state energetics for TM complexes and models of enzy-
matic sites. We start with the presentation of available computational
methods, then show a few examples of challenging calculation of spin-state
energetics (mainly from our recent studies) and illustrate that these issues are
of direct importance in studies of reaction mechanisms, to finally underline
the need for quantitative benchmarking of computational methods with
respect to reliable experimental data, where the role of solvation or other
environmental effects must be properly recognized.

2. Advances in methodology
The properties of TM complexes, including spin-state energetics, are
heavily influenced by electron correlation effects. At the level of Hartree–
Fock (HF) theory, where only exchange, but no other (i.e., Coulomb) cor-
relation effects are accounted for, the HS state is strongly overstabilized with
respect to the LS state. The treatment of electron correlation lowers the total
224 Mariusz Rado
n

energies for both states, but more considerably for the LS state (since it con-
tains more electron pairs), correcting for the above bias of HF theory.1
To describe the correlation effects, either DFT or WFT methods can be
used. Whereas the DFT methods are computationally far more efficient—
and hence eligible for large-scale applications in the field of (bio)inorganic
chemistry—the WFT-based approach offers the possibility of systematic
improvements and obtaining a very high accuracy, albeit for the expense
of a much higher computational cost.
The potential energy surfaces obtained from DFT calculations are
typically quite correct, leading to good structures and vibrational
frequencies,11,12 and justifying the common practice of using molecular
geometries optimized at the DFT level for subsequent single-point calcu-
lations with WFT methods. Even considering very optimistic prospects of
applying WFT method to much larger molecules than it is currently fea-
sible in a significantly reduced time (e.g., due to the recent developments
of local correlation methods13,14 and hardware progress), DFT is likely to
remain the method of choice for studying reaction mechanisms, per-
forming conformational analyses or computing vibrational frequencies
of large molecular systems, and for periodic quantum calculations of crys-
talline systems. This being said, one must admit that there exist systems and
problems (e.g., spin-state energetics of TM complexes, the key motif of
this chapter), where the correct choice of exchange–correlation functional
in DFT calculations becomes a critical issue. In such cases, a careful cali-
bration with respect to accurate WFT calculations for suitably designed
small models is necessary to resolve the doubts stemming from the DFT
approximations.

2.1 Density functional theory methods


Built on top of the Hohenberg–Kohn theorems and the Kohn–Sham
construction,15 DFT is a rigorous electronic structure theory (for the ground
state, or more generally, the lowest electronic state with a given spin and
spatial symmetry, allowing to describe alternative spin states of TM com-
plexes). Although the theory is exact, the calculations performed in practice
rely on an approximate form of the exchange–correlation functional. Many
approximations to this functional—usually called “functionals” or “DFT
methods”—have been devised, ranging from semilocal functionals (includ-
ing the local density approximation, gradient functionals and meta-gradient
functionals), through hybrid functionals, up to double-hybrid, local hybrid,
Accurate spin-state energetics 225

and range-separated hybrid functionals.16,17 Since neither semilocal nor


hybrid functionals can deal properly with van der Waals dispersion interac-
tions, the appropriate dispersion corrections were developed by Grimme
and coworkers18 and are widely used nowadays under the acronym DFT-
D (DFT-D3 or earlier versions).
When using DFT methods, the practical question is which functional to
choose for a given problem.19 Whereas some properties (e.g., molecular
geometry) are relatively insensitive to the choice of functional, the com-
puted relative energies of TM complexes (including spin-state energetics,
metal–ligand bond energies, thermochemical and kinetic parameters, d–d
excitation energies) can be strongly functional dependent.11,12,20,21 With
variations of the computed relative energies being as large as 10–20 kcal/
mol, it is not rare that DFT calculations may give inconclusive results. There
is a growing appreciation for the fact that none of the currently available
approximate functionals is flexible enough to be universally recommended
for all systems and properties.
In the context of spin-state energetics, this strong dependence of DFT
results on the choice of exchange–correlation functional has been recognized
for long.20–22 Building on seminal works of Trautwein et al.23 and Reiher
et al.24, a number of authors demonstrated that the main factor affecting
the computed spin-state energetics is the admixture of exact exchange in
hybrid functionals. Hybrid functionals are constructed by admixing a fraction
of exact exchange (from HF theory) to a semilocal exchange functional and
combining it with a semilocal correlation functional,

hybrid
Exc ¼ xExexact + ð1  xÞExsemilocal + Ecsemilocal , (1)

where x describes the admixture of exact exchange. Values of this parameter


vary by functional, e.g.: 25% for PBE0, 20% for B3LYP, 15% for B3LYP*
(a reparametrization of B3LYP by Reiher et al. designed to improve spin-
state energetics of SCO complexes), 10% for TPSSh, and 27% for M06.
Typically, by increasing the amount of exact exchange, a HS state (containing
more unpaired electrons) is stabilized with respect to LS state (containing none
or a smaller number of unpaired electrons). The LS–HS energy difference
depends on x almost linearly with typical slopes on the order of 1 kcal/mol
per 1 percent of exact exchange.25 The x values of about 10%–15% are
appropriate to describe spin-state energetics of SCO complexes,24–26 but
the optimal value is system dependent (see examples in Sections 3–5).
226 Mariusz Rado
n

The sensitivity of DFT spin-state energetics to the exact exchange


admixture is usually rationalized27 by the fact that there are more stabilizing
exchange interactions in the HS state (containing a higher number of elec-
trons with parallel spins) than in the LS state, and hence the HS state is
strongly favored at the level of HF theory, from which the exact exchange
contribution comes. However, the present author has argued that the above
explanation is oversimplified.28 He pointed out examples of TM systems—
including porphyrin, pseudo-octahedral, and metallocene complexes, as
well as TM ions surrounded by point charges—for which the usual sensitiv-
ity of spin-state energetics to the amount of exact exchange is not observed
and, curiously, even the opposite trend (i.e., slight stabilization of LS state
by increasing the admixture of exact exchange) may be observed in some
cases. Based on these considerations, the present author argued that the char-
acteristic dependence of DFT spin-state energetics on the exact exchange
admixture cannot be satisfactorily explained by the notion of metal-centered
exchange interactions, but rather it is intimately connected to description of
the metal–ligand bonding.28
It was explained in Ref. 28 that a transition from the LS to the HS
state typically involves an electron promotion from a mostly nonbonding,
metal-centered d orbital (for octahedral complex: t2g) to an antibonding
metal–ligand orbital (for octahedral complex: eg). Consequently, the metal–
ligand bond has a higher bond order in the LS than in the HS state, i.e., con-
version from the LS to the HS state makes the metal–ligand bonding weaker.
As the result, the LS state benefits more than HS state from charge delocalization
and nondynamical correlation, the effects intrinsic to the metal–ligand bond.
Since both effects are overestimated by semilocal functionals, replacing a part
of semilocal exchange by the exact exchange necessarily leads to stabilization
of the HS state, roughly proportional to the amount of admixed exact
exchange, as indeed observed. This new interpretation is fully consistent with
the typically observed strong dependence of DFT spin-state energetics on the
exact exchange admixture, but it also allows to rationalize why a much
weaker dependence is observed for spin transitions involving electron redistri-
bution only between nonbonding orbitals (so that charge delocalization and
metal–ligand nondynamical correlation effects are comparable for both spin
states).28
In further related study, Pinter et al.29 analyzed changes in electronic
density distributions caused by varying the exact exchange admixture and
ascribed them to correlation effects. Moreover, by comparing two classes
of octahedral complexes with ligands of different σ-donor abilities, these
Accurate spin-state energetics 227

authors further confirmed the key importance of the nondynamic correla-


tion effects related to the metal–ligand bonding/antibonding orbitals in
determining the characteristic dependence of DFT spin-state energetics
on the exact exchange admixture.
Clearly, it is not only the admixture of exact exchange that determines
the spin-state energetics in DFT calculations. Swart30 and Ghosh31,32 with
their respective coworkers demonstrated that among semilocal functionals
those containing the OPTX exchange functional,33 such as OLYP or
OPBE, perform particularly well for spin-state energetics of iron complexes.
Good performance of OPBE has also been noted for SCO complexes of var-
ious TMs in the recent study by Ruiz and coworkers.34 Swart et al. further
observed that, when compared with a classic gradient functional PBE, the
OPBE one is superior for spin-state energetics and reaction barriers, but
simultaneously performs worse for weak interactions.35,36 By considering
how exchange parts of these functionals (OPTX, PBEx) depend on the
dimensionless density gradient (s), they proposed to interpolate between
the OPTX-like behavior for low-s limit and the PBEx-like behavior for
high-s limit, to get the best of these two functionals in a single one. Based
on these considerations, new exchange–correlation functionals (including
dispersion corrections) were constructed: SSB-D36 and more recently
S12g,37 which are promising for TM chemistry.38–40

2.2 Wave function theory methods


The WFT methods of including electron correlation are classified as single-
reference or multireference ones, depending on whether the reference state is a
single Slater determinant or a combination of determinants. Coupled cluster
(CC) methods are the most accurate and reliable single-reference
methods,41,42 and the CCSD(T) method (with full iterative treatment of
singles and doubles, and a perturbational treatment of the triples) is regarded
as the “gold standard” of accuracy in quantum chemistry.43,44 Note that
there are several variations of this method when applied to open-shell sys-
tems, depending on whether spin-restricted or unrestricted orbitals are used
and whether or not the CC wave function is (partially) spin adapted.45
Multireference methods typically start from a CASSCF (complete active
space self-consistent field) wave function. Already at this level, it is possible
to accurately describe the most relevant (static) correlation effects, combined
with optimization of the molecular orbitals. However, to approach the
chemical accuracy, the remaining (dynamic) correlation effect must be
228 Mariusz Rado
n

captured by performing subsequent calculations, either with variational


MRCI (multireference configuration interaction)46 or its spectroscopically
oriented formulation (SORCI),47 or with computationally cheaper second-
order perturbation theory methods: CASPT2 and NEVPT2. The latter two
methods differ in formulation of the zero-order Hamiltonian (H ^ 0 ). For
^
CASPT2, H 0 is a simple one-electron operator, the direct generalization
of the Fock operator used in the single-reference Møller–Plesset perturba-
tion theory.48 In the currently used form it also contains a single parameter,
the IPEA shift, which was introduced to improve relative energies of
open-shell and closed-shell states.49 A more complicated (two-electron)
H^ 0 is used for NEVPT2,50 but this does not necessarily lead to more accurate
results for TM complexes.51 Despite continued efforts to create various
multireference CC theories and some interesting new developments in this
field,52,53 so far none of the resulting methods has attained much use in the
field of (bio)inorganic chemistry.
The critical aspect of any multireference calculations is the choice of
active space.54,55 For a mononuclear TM complex with closed-shell ligands,
the standard active space is obtained by making active the metal d orbitals
along with the correlating d0 orbitals (double d-shell) and ligand orbitals giv-
ing rise to covalent bonding with the metal d orbitals.56 The number of
active orbitals is limited to 15–16, although larger active spaces may be han-
dled by replacing CASSCF with RASSCF (restricted active space SCF)57 or
DMRG (density matrix renormalization group) approaches.58 The selection
of active orbitals is usually performed manually, despite many attempts to
automatize this procedure.59,60 For some problems it may be complicated
to find an active space which is chemically appropriate, numerically stable
with respect to orbital rotations, and small enough to be handled. By
avoiding the difficulty of active space construction, single-reference CC
methods are considered more “user-friendly,” provided that a single deter-
minant is a reasonable reference state.
Another challenge in performing any (single- or multireference) WFT
calculations is the slow convergence of correlation energies with the size
of one-particle basis set. To minimize the basis set incompleteness error,
it is possible to extrapolate the correlation energy to the complete basis
set (CBS) limit based on the results of calculations with at least two different,
systematically convergent basis sets.61,62 It is also possible to use explicitly
correlated (F12) methods63 which (due to incorporation of an inter-
electronic distance into the wave function) drastically improve the conver-
gence rate to the CBS limit.64 For instance, CCSD(T)-F12 is an explicitly
Accurate spin-state energetics 229

correlated version of the CCSD(T) method that seems very promising for
applications by providing a reasonably small basis set incompleteness error
already with the triple-ζ quality basis set.65
A frequently recurring question is whether single-reference CC methods
are appropriate for TM complexes with some sort of multireference charac-
ter. Whereas single-reference methods are known to fail in genuine
multireference scenarios (e.g., homolytic dissociation of a covalent bond,
description of a singlet biradical) when using RHF-type reference,41 and
low-order single-reference methods (e.g., MP2 or CISD) are clearly not
appropriate for TM chemistry, the available data suggest that CCSD(T)
method is reasonably accurate even for molecules with moderate non-
dynamical correlation effects, such as for typical TM complexes near the
equilibrium geometry.44,66,67 Several metrics of multireference character
have been introduced to diagnose problems with nondynamic correlation
effects (including popular T1 and D1 diagnostics, based on amplitudes of sin-
gles in the CCSD wave function, and many others68), but it is not clear how
to translate values of these diagnostics into predictable error bars of a given
method.68,69 Moreover, there are examples of systems (some of which dis-
cussed below), for which CCSD(T) appears to give rather accurate results
despite large values of some of these diagnostics. Finally, as pointed out
by Neese et al.70, the problem with multireference character is sometimes
confused with the problem of choosing the appropriate reference state.
The latter is usually taken as the HF determinant, but in some cases the
shapes of HF orbitals may be inappropriate to describe the state of interest,
and the use of Kohn–Sham orbitals (from DFT calculations) is a reasonable
option.67,71
In view of rather subtle interplay of static and dynamic correlation effects
in TM complexes, many authors (including the present one) attempted to
compare the results obtained from single-reference CCSD(T) and
multireference CASPT2/NEVPT2 calculations. Whereas the good agree-
ment between both approaches is optimistic, it is more problematic to inter-
pret occasionally obtained discrepancies, which may indicate problems with
description of static or dynamic correlation, or both problems together.
Most authors who observed such discrepancies in the context of spin-state
energetics ascribed them to shortcomings of the PT2-based methods.72–75
Based on this experience it seems that CASPT2 calculations—with the
default choice of the IPEA shift parameter (0.25 a.u.) and the standard choice
of the active space—have a tendency to overstabilize the HS state with
respect to the LS state by a few kcal/mol. Tuning the IPEA shift parameter
230 Mariusz Rado
n

was suggested to remedy this problem, although the optimal value is system
dependent.74,75 Recently, Pierloot and coworkers51 observed that these dis-
crepancies between CCSD(T) and CASPT2 arise mostly from description of
correlation effects involving the metal outer-core orbitals (3s3p for first tran-
sition row), whereas for valence-only correlation energies these two
methods give more comparable results. In a subsequent related study, Phung
et al.76 proposed to use the CCSD(T) (with minimal basis set on ligands) to
describe the outer-core correlation effects combined with CASPT2 to cap-
ture the valence-only correlation effects, i.e., the CASPT2/CC method.
The computational cost is always the main factor to be considered when
applying WFT methods. To make the calculations feasible for larger
systems—like TM complexes with organic ligands or models of enzymatic
active sites—it is usually necessary to compromise on the size of basis set,
although this unavoidably deteriorates the quality of results with respect
to the CBS limit. An important goal is thus to develop efficient computa-
tional protocols in order to minimize the basis set error for a moderate com-
putational cost. There are various composite thermochemical approaches
constructed in this spirit by different authors (for review, see Ref. 64 and
references therein); however, many of these protocols available in the liter-
ature target much smaller molecules and higher level of accuracy than those
of relevance in this chapter.
With the focus on heme-related models and the expected accuracy of
roughly 1–3 kcal/mol in relative energies of spin states, the present author
designed an example of efficient CCSD(T) computational protocol.77 It is
based on the splitting of CCSD(T) energy in three contributions: (1) non-
relativistic frozen core CCSD energy, (2) relativity and core correlation
correction to the CCSD energy, (3) correction due to connected triples
(T) (including relativity and core correlation). These contributions have
different requirements on the basis set size and thus it is advantageous to
compute them independently, i.e., contribution (1) was obtained using
CCSD-F12 methodology, whereas contributions (2) and (3) from suitable
CBS extrapolations. To optimize the computational cost and scalability with
the ligand size, composite basis sets were used (generally larger on the metal
than on the ligand atoms, especially those not bonded directly to the metal).
The final choice of basis sets, the F12 ansatz, and extrapolation procedures to
define this computational protocol, was made to best reproduce the highest-
quality CBS-extrapolated results available for 8 small mimics of porphyrin
complexes (with maximum discrepancy of 1.6 kcal/mol, for most cases
below 1 kcal/mol).77 The protocol was subsequently applied to study
Accurate spin-state energetics 231

spin-state energetics of larger heme-related mimics and even to real porphy-


rin complexes sized up to 37 atoms (see Section 3.1). We also note an effi-
cient CCSD(T)-F12 computational protocol by Harvey and coworkers
(proposed in their study of Fenton reaction mechanism),78 which was also
adopted by the present author in some of the subsequent studies.79,80

3. Challenges in spin-state prediction


3.1 Porphyrins and heme-related models
As basic units of heme proteins and important biomimetic models, meta-
lloporphyrin complexes have received justified attention in theoretical stud-
ies. For some metalloporphyrins it is computationally challenging already
to reproduce the experimental ground state. A good example is four-
coordinate FeII porphyrin complex (FeP), experimentally known to have
the triplet ground state and low-energy quintet state (see Ref. 77 and
references therein). The triplet ground state is qualitatively reproduced by
most DFT methods, i.e., they give positive values for the adiabatic electronic
energy difference between the quintet and triplet states, ΔEadiab ¼ E(S ¼ 2) 
E(S ¼ 1), although the actual values of ΔEadiab are strongly functional-
dependent (as an example, the BP86 and B3LYP results differ by 11 kcal/mol;
see Table 1). By contrast, as will be discussed below, most of the WFT calcu-
lations so far published point to the wrong (quintet) ground state.
A few methodology aspects must be clarified before proceeding further.
First, as was mentioned in Section 1, it is not only the electronic energy dif-
ference that determines the experimental ground state. For FeP at room
temperature, the correction to free energy (accounting for vibrational
zero-point and entropic effects) amounts to 2 kcal/mol in favor of the quin-
tet state. Such thermochemical corrections can be conveniently estimated at
the DFT level and the results are practically independent on the choice of
functional (by contrast to relative electronic energies). Second, an excep-
tional feature of FeP is a significant correction due to the spin-orbit coupling
(SOC). The SOC corrections to spin-state energetics are often negligible for
spin-state energetics of first-row TM complexes,80 but in the case of FeP
there is a relatively strong SOC interaction between the lowest triplet
(3A2g under D4h) and the higher energy triplet states (3Eg, 3B2g), which gives
rise to additional stabilization of the lowest triplet state with respect to the
quintet state by 1 kcal/mol.57 Another approximation, made almost rou-
tinely in most theoretical calculations, is the use of unsubstituted porphyrin
232 Mariusz Rado
n

Table 1 Quintet–triplet energy difference for FeP


Method ΔEadiaba Source of data
Exptl lower bound >1 See text
Former works
BP86 16.7 Ref. 81
B3LYP* 8.6 Ref. 81
B3LYP 5.7 Ref. 81
PBE0 0.8 Ref. 81
M06 9.5 Ref. 81
CASPT2(8,11)b 4.8 Ref. 51 (in Ref. 83: 6.5c)
CASPT2(8,11)-CBSb 3.3 Ref. 76
CASPT2(16,15) b
1.8 Ref. 51 (in Ref. 57: +0.9c)
RASPT2(16,19)b 4.3 Ref. 51
DMRG-CASPT2(16,19) b
4.2 Ref. 51
d
CCSD(T)-CBS 2.3 Ref. 77
This work, using common basis sete
CCSD(T)f 3.7 This work
g
CCSD(T)-F12 1.5 This work
CASPT2(8,11)b 8.5 This work
NEVPT2(8,11) b
7.6 This work
MRCISD+Q(8,11) b,h
6.8 This work
a
Quintet minus triplet adiabatic electronic energy difference, all values in kcal/mol.
b
Notation for active spaces used is (Ne, No), where Ne is the number of active electrons and No is the
number of orbitals; (8,11) is standard active space build of Fe d,d’ orbitals and one σ orbital of porphyrin;
(16,15) is (8,11) plus Fe 3s,3p orbitals; (16,19) is (16,15) plus Fe 4s,4p orbitals.
c
The results given in parentheses (from Refs. 83 and 57) suffered from the bug in Molcas ANO-RCC
basis set for C; the corresponding results for correct basis set are quoted after Ref. 51.
d
Composite protocol intended to approach the CBS quality at moderate cost (see Section 2.2 and Ref.
77), using RCCSD(T) formulation.
e
Basis set: cc-pwCVTZ-DK/cc-pVTZ-DK/cc-pVDZ-DK (Fe/N/C,H) with second-order Douglas–
Kroll scalar relativistic Hamiltonian; single-point calculations for B3LYP/def2-TZVP structures from
Ref. 77.
f
Using RCCSD(T) formulation.
g
Nonrelativistic RCCSD(T*)-F12a using cc-pwCVTZ/cc-pVTZ/cc-pVDZ basis set (Fe/N/C,H)
with appropriate auxiliary basis sets chosen analogously as in Ref. 78, with added second-order
Douglas–Kroll (DK) correction for scalar relativistic effects estimated at the RCCSD(T) level.
h
Internally contracted MRCI46 with singles and doubles plus Davidson correction (fixed reference) for
size-extensivity.
Accurate spin-state energetics 233

ring, whereas the experimental data refer to complexes with the substituted
ring, such as Fe(TPP) (TPP ¼ tetraphenylporphyrin).81 The porphyrin side
groups may be important in some cases,82 but apparently not here: the rel-
ative energies of quintet and triplet spin states for FeP and Fe(TPP) agree up
to 0.1 kcal/mol, as can be verified easily at the DFT level, justifying the
neglect of side phenyl groups in the computational model used below.
Taken together, based on the above estimates of free energy corrections
(2 kcal/mol in favor of the quintet) and the SOC correction (1 kcal/mol in
favor of the triplet), and knowing that the experimental ground state is trip-
let, we obtain the following inequality (the lower bound) on the electronic
energy difference: ΔEadiab > 1 kcal/mol. Reproducing this experimentally
derived inequality is a very challenging test for WFT methods (Table 1). All
published CASPT2 and RASPT2 results—including those based on the
standard (8,11) active space and even on more extended ones—still point
to the wrong (quintet) ground state. We note in passing that the results gath-
ered in Table 1 considerably improved over the earliest CASPT2 study,84
where ΔEadiab ¼ 19.2 kcal/mol (!) was reported. The huge error in favor
of the quintet state was mainly due to the problem with inadequate active
space used in Ref. 84, i.e., not accounting for covalency of the Fe–N
σ-bonds with the porphyrin ring,56 and too small basis set used. However,
it seems that even with a reasonably chosen active space (including recently
proposed extensions: making active also Fe 3s,3p outer-core orbitals, possi-
bly in combination with Fe 4s,4p orbitals) and even when extrapolated to
the CBS limit,76 it is not possible to reproduce the correct ground state
of FeP from these perturbational methods. By contrast, the CCSD(T) cal-
culations performed by this author in Ref. 77 gave, correctly, the triplet gro-
und state favored by more than 2 kcal/mol. These calculations were
performed with the computational protocol optimized for spin-state ener-
getics of heme-related models, which was designed to approach the CBS
quality at a reasonable computational cost (see Section 2.2).
A part of the discrepancies between the discussed CCSD(T) and
CASPT2 results may be traced back to the use of different basis sets. In order
to minimize this source of inconsistency, we also give in Table 1 the results
obtained from CCSD(T), CCSD(T)-F12, CASPT2, NEVPT2, and
MRCISD+Q calculations using a common basis set (triple-ζ on Fe and
coordinating N atoms, and double-ζ on the remaining atoms). We note,
first, that a difference of 5 kcal/mol between the CCSD(T) and CCSD
(T)-F12 results confirms the importance of basis set incompleteness error.
The CCSD(T)-F12 result is quite close to the earlier discussed CCSD
234 Mariusz Rado
n

(T)-CBS estimate (0.8 kcal/mol difference) and both CC results satisfy the
experimentally derived inequality on the electronic energy difference. Sec-
ond, both CASPT2 and NEVPT2 methods, when compared against
CCSD(T), tend to overstabilize the quintet with respect to the triplet state
by 4–5 kcal/mol. This is in accord with similar suggestions obtained for sim-
plified heme-related mimics57,72 and other TM complexes.74 Interestingly,
the comparable overstabilization of the triplet state is also observed at the
MRCI level, even with the Davidson correction for size-extensivity. In
agreement with the recent proposal of Pierloot et al.51, the above discrep-
ancy between CASPT2 and CCSD(T) is reduced to  2 kcal/mol when
these methods are compared for valence-only correlation energies,
providing a basis for the CASPT2/CC composite approach by Phung
et al.76 (see Section 2.2). The CASPT2/CC value of the ΔEadiab energy
difference for FeP (+ 0.2 kcal/mol) given by Phung et al.76 is, indeed, a
considerable improvement over the bare CBS-extrapolated CASPT2 result
( 3.3 kcal/mol).
Sticking to the CCSD(T) level of theory, the present author obtained
accurate spin-state energetics for many other heme-related models.77 The
actual CCSD(T) calculations (with the computational protocol described
in Section 2.2) were only carried out for the smallest and most symmetric
porphyrin models: FeP and FeP(Cl). To deal with other, more complicated
porphyrin complexes, the present author introduced an extrapolation method
in which the CCSD(T) estimate for a porphyrin complex can be obtained
from the result of CCSD(T) calculations for its simplified mimic and a series
of DFT calculations for the real complex and the mimic, making us of an
approximately linear relationship between spin-state energetics of both systems. This
approach is illustrated in Fig. 1 for the example of FeP and its small mimic
FeL2 (where L¼C3N2H 5 ). Note that similar or smaller heme mimics were
used before by other authors,85–87 but with no attempts to perform quanti-
tative extrapolation with respect to the model size. The accuracy of this
extrapolation method (1–2 kcal/mol) was verified for FeP and FeP(Cl) by
comparing with the actual CCSD(T) results.77
The present author applied this methodology to study spin-state energet-
ics of several heme-related models, including FeP(Im) (Im ¼ imidazole) and
FeP(SH), which are models of FeII site in myoglobin and FeIII site in the
resting state of P450, respectively.77 Unfortunately, the experimental data
of spin-state energetics for most of these systems are rather scarce (typically,
only the identity of the ground state is known), not allowing us to quanti-
tatively compare theory with experiment (for the experimental data, see
Accurate spin-state energetics 235

A B

C
ΔE for FeP (kcal/mol)


– – –
ΔE for FeL2 (kcal/mol)
Fig. 1 Structures of ferrous porphyrin FeP (A) and its simplified mimic FeL2 (B), and cor-
relations between relative spin-state energetics of both models studied with different
DFT methods (C). Dotted lines show how the results of CCSD(T) calculations for the
mimic are used for making predictions for the large model based on the established
trend lines. The actual CCSD(T) results for FeP are shown as empty squares. Labels 5IS,
3
IS1, 2, and 1LS refer to four considered spin states of FeP: 5A1g, 3A2g (the lowest triplet
state), 3Eg (the other triplet state), and 1A1g (closed-shell singlet). Adapted with permission
from Radon  , M. J. Chem. Theory Comput. 2014, 10, 2306–2321, 10.1021/ct500103h. Copy-
right (2014) American Chemical Society.

Ref. 77 and references therein). Only for one complex studied in Ref. 77,
namely [FeP(CN)], it was possible to obtain a quantitative estimate of spin-
state splitting from the well-documented SCO behavior.88,89 Based on free
energy correction computed at the DFT level for the experimentally
236 Mariusz Rado
n

reported SCO temperature, the electronic energy difference between the


quintet and the singlet state (5HS–1LS) was estimated to be 5 kcal/mol.
The CCSD(T) estimate of the same energy is 2.9 kcal/mol. The obtained
discrepancy of 2 kcal/mol is within the assumed uncertainty of the extrap-
olation procedure. Also environmental effects (not considered in the calcu-
lation, but typically stabilizing the LS state; see Section 5) may contribute
to this.
The lack of quantitative experimental data for other heme models stud-
ied in Ref. 77 allowed for making only a qualitative comparison between
theory and experiment. Both FeP(Im) and FeP(Cl) are experimentally
HS complexes, with the quintet (5HS) or sextet (6HS) ground state, respec-
tively. For both of them, the correct ground state is reproduced at the
CCSD(T) level already by looking at the electronic energies. Accounting
for vibrational zero-point and entropic effects would additionally favor
the HS state in terms of free energy. More problematic results are obtained
for these complexes at the DFT level. Indeed, as also pointed out by other
authors before,38,90 many DFT methods (including those generally rec-
ommended for spin-state energetics: TPSSh and B3LYP*, and to a lesser
extent also OLYP/OPBE) tend to overstabilize the triplet state (3IS) of
FeP(Im). Even accounting for the free energy correction (estimated to
3.5 kcal/mol in favor of the 5HS state), it would be problematic to recover
the correct ground state (5HS) when using these functionals. The similar
overstabilization of the 4IS state with respect to the 6HS state was noted
for FeP(Cl). Interestingly, for this complex the solvation correction to the
4
IS 6HS adiabatic energy difference was estimated by Gruden et al.39 to
be as large as  5 kcal/mol (stabilizing the 4IS excited state), whereas the
free energy correction can be estimated to only + 1.5 kcal/mol (in favor
of the 6HS ground state). Thus, the CCSD(T) calculations, giving the elec-
tronic energy difference of 6.3 kcal/mol (in gas phase),77 are capable of rep-
roducing the correct ground state also in solution, which is not the case of
many DFT methods (including B3LYP*, TPSSh, S12g), having the ten-
dency to overstabilize the 4IS state.39,77
Another system studied in Ref. 77 was FeP(SH), a five-coordinate FeIII
species being a model of P450 resting state.91 In this case the experimental
data are indicative of a thermal SCO between the sextet and doublet states,
with the quartet state lying higher in energy (see Ref. 77 and references
therein). This requires the following ordering of the spin states at the elec-
tronic energy level: 2LS ≲6HS < 4IS. Our CCSD(T) estimates in Ref. 77
Accurate spin-state energetics 237

gave the 4IS state above the 6HS state by 6 kcal/mol. The 2LS and 6HS states
were predicted close in energy (within 2 kcal/mol), but ordered incorrectly
(6HS below 2LS). While this may simply reflect the approximate character of
the extrapolation procedure used to provide these CCSD(T) estimates, it may
also be related to some multireference character observed for this system, and
particularly pronounced in the 2LS state. As discussed in Ref. 77, the above
problem is not remedied by switching from CCSD(T) to CR-CC(2,3)
(a completely renormalized coupled cluster method,92 which is supposed,
in principle, to outperform CCSD(T) for molecules with multireference char-
acter). The use of KS instead of HF orbitals in CCSD(T) calculations improves
the 2LS–6HS splitting for FeL2(SH) mimic by only 0.8 kcal/mol. The
CASPT2 calculations also give the same (incorrect) ordering of the 2LS
and 6HS states for FeP(SH) model.73 Only very recently, Phung et al. obtained
the expected ordering of these close-lying spin states by using their newly pro-
posed composite CASPT2/CC method.76
Concerning the 4IS–6HS energy difference for FeP(SH), the CCSD(T)
estimate agrees well with the experimental state ordering (see above). This is
not the case of many DFT methods—including the functionals normally
recommended for spin-state energetics of TM complexes and performing
well for the 2LS–6HS splitting, like B3LYP*, TPSSh, and OLYP—whose
results are considerably biased in favor of the 4IS state. This is reminiscent
of related problems observed for FeP(Im) and FeP(Cl) as discussed above.
The functional which best reproduces the 4IS–6HS energy difference (com-
pared with the CCSD(T) estimate) is PBE0, although it performs rather
poorly for the 2LS–6HS energy difference.
This example and other similar ones discussed in Ref. 77 demonstrate the
lack of universality in regard to the choice of functional in DFT for the prob-
lem of spin-state energetics. When comparing various energy differences,
not only between different molecules, but also within the same molecule
(e.g., the LS–HS, IS–HS gaps for FeP(Im) or FeP(SH)), it is difficult to find
a single DFT method capable of reproducing all these energy differences
simultaneously correct. A related problem will be also discussed in the con-
text of NO binding to MnII porphyrin (Section 4.2).

3.2 The mystery of hexacyanomolybdate(IV)


The issues with close-lying spin states may also appear for 4d-electron TM
complexes. The present author with coworkers investigated the electronic
238 Mariusz Rado
n

structure of hexacyanomolybdate(IV) anion, [MoIV(CN)6]2, which was


earlier reported to be perfectly octahedral and, at the same time, diamagnetic
at room temperature.93 This combination is paradoxical and totally
unexpected: according to the Hund’s rule, an octahedral complex of MoIV
(a d2 ion), should obviously have the paramagnetic triplet ground state 3T1g.
We then asked in Ref. 94 how such a complex can be diamagnetic.
The calculations at various DFT and WFT levels performed in Ref. 94
confirmed that—when assuming the octahedral geometry from the exper-
imentally reported crystal structure—the triplet state is preferred over the
singlet one by more than 12 kcal/mol. Moreover, additional calculations
assured that for octahedral [Mo(CN)6]2 with the triplet ground state, nei-
ther spin-orbit coupling (SOC) on a single Mo site nor antiferromagnetic
interactions between neighboring sites in the crystal structure, can explain
the lack of paramagnetic properties. The observed diamagnetism can only
be understood when assumed that individual [Mo(CN)6]2 species have sin-
glet ground state, although this is inconsistent with the octahedral geometry!
Rather unexpectedly, our investigations in Ref. 94 revealed that [Mo
(CN)6]2 anion (in either spin state) is unstable as octahedron and it spon-
taneously converts into trigonal prism when unconstrained geometry opti-
mization is performed. Fig. 2 compares both discussed geometries of [Mo
(CN)6]2 with schematic splittings of the d-orbital energy levels. The insta-
bility of the octahedral structure was confirmed for different DFT methods
and by applying a variety of approaches: harmonic analysis, intrinsic reaction
coordinate paths, and molecular dynamics.94 Not only the octahedral geom-
etry was predicted to be unstable (saddle point), but also the total energy gain
from the conversion of octahedral to trigonal prismatic geometry was found
to be large: at least 10 kcal/mol in the triplet state and even larger in the sin-
glet state.
The preference of [Mo(CN)6]2 for trigonal prismatic geometry can be
rationalized94 as second-order Jahn–Teller effect rooted in a tendency to
maximizing σ-bonding interactions between CN anions and an
electron-deficient MoIV center, analogously as for other known trigonal
prismatic ML6 complexes.95–97 For comparison, two other complex anions
were also considered in Ref. 94: [MoIII(CN)6]3 and [MoIVCl6]2. Both of
them were computationally shown to prefer the octahedral geometry, either
due to the higher number of electrons on the MoIII center compared with
the MoIV one, or due to the presence of strong π-bonding interactions with
Cl ligands. The experimentally known [MoIII(CN)6]3 anion is indeed
octahedral.98
Accurate spin-state energetics 239

Fig. 2 Octahedral (A) and trigonal prismatic (B) geometries of [Mo(CN)6]2. The right
panel shows schematically a splitting of the Mo 4d orbitals for both geometries and
orbital occupancies in the triplet ground state for (A) and in close-lying singlet and trip-
let states for (B). Reproduced from Radon , M.; Rejmak, P.; Fitta, M.; Bałanda, M.;
Szklarzewicz, J. Phys. Chem. Chem. Phys. 2015, 17, 14890–14902, 10.1039/c4cp04863f with
permission from the PCCP Owner Societies.

In addition to calculations for the isolated anion, the instability of [Mo


(CN)6]2 in the octahedral geometry was also confirmed by performing
periodic DFT calculations for the previously reported crystal structure of
(Me4N)2[Mo(CN)6]H2O.93 Periodic geometry optimization, starting from
the reported crystal structure, resulted in strongly distorted geometries of the
[Mo(CN)6]2 units, intermediate between the octahedron and trigonal
prism.94 Such strong disagreement between the theoretically optimized
and the previously reported crystal structure indirectly suggests that the crys-
tal structure might have been resolved incorrectly (see below).
As suggested by orbital energy diagrams in Fig. 2, the change of coordi-
nation geometry clearly affects the spin-state energetics of [Mo(CN)6]2.
Whereas the triplet was strongly favored for the octahedral geometry, both
spin states come much closer in energy for the trigonal prism geometry. Still,
most of the methods tested in Ref. 94 (various DFT and CC methods) favor
the triplet state by a few kcal/mol. However, CASPT2 calculations with the
IPEA shift of 0.5 to 0.75 a.u. pointed to the singlet ground state favored over
the triplet one by 1.3–4.6 kcal/mol,94 which appears to be consistent with
240 Mariusz Rado
n

the experimental diamagnetic behavior. Despite the closeness of both spin


states predicted by CASPT2 calculations, the free energy correction for rel-
ative stability of both spin states was found to be insufficient for inducing a
SCO behavior (which is, indeed, not observed in the experiment). Intrigu-
ingly, however, the new magnetic measurements (SQUID) over a broad
range of temperatures revealed a slight increase of magnetic susceptibility
with the increase of temperature. This feature, naturally explained in terms
of Boltzmann population of the higher energy paramagnetic level, is seem-
ingly consistent with the small energy gap between both spin states found
computationally. This interpretation was also supported by a semiquantita-
tive thermodynamical model proposed in Ref. 94.
Our theoretical investigations94 thus have suggested a possible explana-
tion of why the [Mo(CN)6]2 anion is diamagnetic: this becomes possible
only for the trigonal prismatic geometry, but the anion tends to adopt such
geometry. However, a new question appeared. If the calculations unambig-
uously show that [Mo(CN)6]2 should never be isolated as octahedral (but
rather trigonal prismatic), how is it possible that the previously reported
crystal structure “portrayed” [Mo(CN)6]2 units in the ideal octahedral
geometry? In Ref. 94, we were not able to answer this question fully. How-
ever, we suggested that the original crystal structure might have been
resolved incorrectly due to overlooking merohedral twinning, disorder or
superstructure properties. Examples are known from the literature where
a wrong crystal structure with apparently too high symmetry is obtained
due to overlooking the above features (see Ref. 94 and references therein)
and this could happen also for the presently discussed system. Until the crys-
tal structure of this mysterious compound is redetermined correctly (which
is challenging and awaits development of the appropriate structural model to
account for a tentative merohedral twinning or related phenomena), definite
conclusions on the magneto-structural properties of [Mo(CN)6]2 cannot
be reached. However, our work in Ref. 94 is an interesting example of
how theory can be used not only to interpret or explain experimental data,
but also to potentially falsify some of them.

3.3 Cobalt–nitrosyl complexes


Another example of challenging ground state determination is that for
cobalt–nitrosyl complexes, recently studied by Broclawik et al. (with con-
tribution of this author) in the context of NO activation by ammonia-
modified CoII sites in zeolites.5,79,99 Joint experimental and theoretical
Accurate spin-state energetics 241

studies confirmed the formation of Co–NO adducts with variable number


of ammonia coligands (those with 3 and 5 coligands being consistent with
the interpretation of IR spectra).5 Most importantly for our discussion, the
calculations also revealed that these Co–NO species have close-lying singlet
and triplet states, with different propensities for NO activation.5 Fig. 3A
shows structures of two representative models used in the computation of
spin-state energetics.
Relative energies of the two spin states were investigated in Ref. 79 for
models with 0, 2, or 3 NH3 coligands and a single aluminum tetrahedron, T1
¼ [Al(OH)4], to simplistically represent the zeolite framework, as well as
for [Co(NO)(NH3)5]2+ complex, corresponding to the limiting case of
complete ammonia saturation (i.e., with 5 NH3 coligands, cobalt is no lon-
ger coordinated to the zeolite framework). The latter pentaaminenitrosyl
species, in addition to being observed in situ by IR spectroscopy,5 is also
structurally characterized (in the form of chloride) and known to have
the singlet ground state.100 The spin-state energetics of [Co(NO)
(NH3)5]2+ and other related Co–NO models were found to be strongly
method dependent. Most interestingly, significant discrepancies were
observed not only between different DFT methods, but also between
two high-level WFT methods: CASPT2 and CCSD(T); see Fig. 3B.79

A B
CASPT2
S=0
–9.8

S=1

CCSD(T)
S=1

13.4
[T1Co(NO)(NH3)3]+ [Co(NO)(NH3)5]2+
S=0
Fig. 3 (A) Structures of two representative models studied in Refs. 5 and 79: [T1Co(NO)
(NH3)3]+ (T1 ¼ [Al(OH)4]) and [Co(NO)(NH3)5]2+. (B) Adiabatic electronic energies of the
lowest singlet (S ¼ 0) and triplet (S ¼ 1) states of [Co(NO)(NH3)5]2+ obtained from
CASPT2 and CCSD(T) calculations (values in kcal/mol). Based on data from Stępniewski,
A.; Radon, M.; Góra-Marek, K.; Broclawik, E. Phys. Chem. Chem. Phys. 2016, 18 (5),
3716–3729, 10.1039/c5cp07452e.
242 Mariusz Rado
n

By predicting the singlet state almost 10 kcal/mol above the lowest triplet
state, the CASPT2 calculations were found unable to reproduce the exper-
imental (singlet) ground state for [Co(NO)(NH3)5]2+. In contrast,
CCSD(T) calculations pointed to the correct singlet ground state, favored
over the triplet state by more than 10 kcal/mol.
The discrepancy of 20 kcal/mol between the CASPT2 and CCSD(T)
results reported in Ref. 79 deserves attention. It is even more striking
that the single-reference CCSD(T) method appears to work well (and qual-
itatively better than multireference CASPT2) despite a noticeable
multireference character (nondynamic correlation effects) observed for
[Co(NO)(NH3)5]2+ and other related Co–NO complexes.79 Due to the
lack of quantitative experimental data for relative energy of the two spin
states, it is not possible to quantify how good is the singlet–triplet energy
difference obtained from CCSD(T) calculations. However, given that
vibrational effects give rise to relative stabilization of the triplet state in terms
of free energy (see Section 3.1), the singlet state should be at least a few kcal/
mol lower in electronic energy to remain the experimental ground state.
The CCSD(T) energy difference is consistent with these estimates and
(as the present author found later101) it is remarkably close to the energy dif-
ference obtained from multireference CI (MRCISD+Q) calculation. The
good performance of CCSD(T) for the spin-state energetics of these
cobalt–nitrosyl complexes, although potentially surprising at first sight, is
consistent with a number of similar findings in the literature, showing that
CCSD(T) works even for molecules with significant nondynamic correla-
tion effects.43,44,67

4. Mechanistic importance
Accurate computation of spin-state energetics is also essential for stud-
ies of spin-forbidden reaction mechanisms. Even in the optimistic scenario
that the ground states of all reactants and products are known from exper-
iment and correctly reproduced by theory, the energy contribution related
to the change of spin state may be reflected in the thermochemical or kinetic
parameters. Moreover, an accurate description of spin-state energetics may
be required to conclusively find out whether or not the spin-forbidden path-
way is the preferred one. Below, two examples will be presented where
accurate determination of spin-state energetics is of mechanistic importance.
Accurate spin-state energetics 243

4.1 An example from enzyme studies


For studies of enzymatic and catalytic reactions, it is often the question
which of the possible elementary pathways—including spin-conserving
and spin-forbidden ones—is the most favorable one in terms of free energy.
When DFT methods are applied, the answer may be dependent on the
choice of functional and hence ambiguities may arise.
For instance, based on DFT computations with different functionals,
four different mechanisms were proposed for O2 activation by
α-ketoglutarate-dependent oxygenases (αKAO) enzymes.102,103 All the
proposed mechanisms lead in a few steps to formation of reactive ferryl
species, FeIV¼O, but some of them differ already in the initial step, i.e.,
binding of molecular oxygen to the FeII site in αKAO. Three of the
proposed mechanisms (A–C) begin with the formation of end-on FeIII–
O2 initial intermediate, which has a HS FeIII site coupled, either
antiferro- or ferromagnetically, to the superoxide ligand (O 2 ), yielding
close-lying quintet and septet spin states. These mechanisms were suggested
by calculations with the B3LYP hybrid functional containing 20% of exact
exchange. However, based on the calculations with the BP86+10% of
exact exchange functional, a fourth mechanism (D) was alternatively pro-
posed, involving formation of FeIV-alkyl peroxo bridged intermediate
(FeIV–O2 2 ) and switching to the triplet spin state. The [Fe –O2 ] species
3 IV 2

is also favored by the TPSSh functional.103 Structures of the two discussed


electromers are shown in Fig. 4A. The reader is pointed to Refs. 102 and 103
(and references therein) for details of different reaction mechanisms.
In the mentioned work,103 Wójcik and Borowski with some contribu-
tion of the present author were able to resolve the mechanistic doubts by
performing detailed DFT studies of all reaction steps as well as calibration
calculations for different electromers of the initial intermediate. Concerning
the second issue, CCSD(T) and DFT calculations were compared for a sim-
plified mimic shown in Fig. 4B. The triplet state of this mimic corresponds

to the FeIV–O2 2 electromer and the septet state to the Fe –O2 one. Note
III

that the antiferromagnetic [Fe –O2 ] state cannot be described by a single-
5 III

reference CC approach used in Ref. 103 and hence it was not investigated;
however, relative energy of the quintet and septet spin states of FeIII–O 2 is
rather easily obtained from broken–symmetry DFT calculations, and it is less
controversial than the energy difference between the alternative electro-
mers, FeIII–O 2 and Fe –O2 .
IV 2
244 Mariusz Rado
n

A B

Fig. 4 (A) Structures of FeIII-superoxide and FeIV-alkyl peroxo bridged intermediates


after binding of O2 to ferrous center of αKAO enzyme. (B) Simplified mimic used for
CC calculations.

Table 2 Relative energiesa of two electromeric forms of mimic shown in Fig. 4B


B3LYP B3LYP* TPSSh OLYP BP86+10% CCSD(T)b
7
[FeIII–O
2 ] 0 4.5 12.3 5.7 13.5 0
3
[FeIV–O2
2 ] 3.6 0 0 0 0 2.1
a
All values in kcal/mol.
b
Best estimate from CCSD(T)-F12 with relativistic correction (details in Ref. 103).
Source: Adapted from Wójcik, A.; Rado n, M.; Borowski, T. J. Phys. Chem. A 2016, 120 (8),
1261–1274, 10.1021/acs.jpca.5b12311 with Permission. Copyright (2016) American Chemical Society.

As might be expected, results in Table 2 show that relative energies of the


two electromers are strongly functional dependent. Comparison with the
CCSD(T) reference shows that BP86+10% and TPSSh strongly over-

stabilize the FeIV–O22 electromer with respect to the Fe –O2 one. This
III

shows that mechanistic studies using these functionals are incorrectly biased
in favor of the mechanism D (i.e., the one which begins by formation of the
3
[FeIV–O22 ] species) and, in this case, the use of B3LYP functional should
lead to more realistic conclusions. In addition to this argument it was shown
in Ref. 103—by detailed mechanistic calculations at the DFT level—that
the pathway involving formation of the 3[FeIV–O2 2 ] intermediate is
disfavored also in later reaction steps.
This example shows that even for studies of complicated mechanisms,
where high-level WFT methods are computationally too expensive to be
directly applied to realistic models or whole reaction pathways, single-point
calculations with these methods for suitably designed small mimics are
invaluable to calibrate the DFT energetics and thus clarify some mechanistic
controversies.
Accurate spin-state energetics 245

4.2 Role of spin states in ligand binding


Ligand binding to a TM metal site (bond formation) and a reverse process
(bond dissociation) are simple, yet important elementary steps in many reac-
tion mechanisms. Due to a drastic change of the ligand field accompanying
the addition or removal of the ligand, this process is often accompanied by a
change of the metal spin state. An important biological example is binding of
O2, CO, and NO diatomics to heme: the ground spin state changes from HS
(S ¼ 2) in the initial five-coordinate FeII complex to LS in the resulting six-
coordinate complexes (S ¼ 0 for O2 and CO adducts or S ¼ 1/2 for NO
one). It has been shown some time ago by this author and Pierloot83 that the
computed heme–ligand bond energies are strongly method dependent and
CASPT2 calculations gave good agreement with the experimentally derived
data, whereas DFT calculations were less successful. A more detailed analysis
revealed that a part of the challenge in DFT calculations can be traced back
to difficulties in computing accurately the energy needed for conversion of
heme from the initial HS to the final LS state.83
More recently, the present author studied NO binding to MnII and CoII
porphyrins.104 Both resulting nitrosyl complexes have singlet ground state,
but differ in the metal–N–O geometry (cf Fig. 5): linear for Mn–NO com-
plex vs bent for the Co–NO one, which is confirmed by both the calcula-
tions and crystal structures of these complexes (see Ref. 104 and references
therein). As also shown in Fig. 5, the equilibrium Mn–N(O) distance (1.6 Å)
is significantly shorter than the corresponding Co–N(O) one (1.8 Å) and the
Mn–NO stretching frequency is much larger for the Co–NO one, corrob-
orating that NO ligand is more tightly bound to Mn–porphyrin center than
to the corresponding Co one. However, Kubiak and coworkers105 demon-
strated by IR spectroscopy that the following nitrosyl transfer reaction

MnPðNOÞ + CoP ! MnP + CoPðNOÞ (2)

occurs spontaneously upon mixing the reactants. This shows that the Mn–
NO bond is thermodynamically weaker than the corresponding Co–NO one, despite
being shorter! Usually, when comparing two similar bonds, the shorter one (in
terms of bond distance) should also be the stronger one (in terms of bond
energy), but this paradigmatic correlation between bond distance and bond
energy does not hold when comparing CoP(NO) with MnP(NO).
In order to understand this paradox, it was proposed in Ref. 104 to take
into account that the ground state is doublet for CoP, whereas it is sextet for
MnP. This difference has profound consequences for mechanisms of NO
246 Mariusz Rado
n

Fig. 5 Structures of CoP(NO) and MnP(NO) complexes in the singlet ground state (P ¼
porphyrin). Annotated are bond distances in Angstrom for M–NO (bold) and N–O
(italics) bonds, and the corresponding stretching frequencies (M¼Co, Mn), all values
at the DFT:BP86/def2-TZVP level from Ref. 104.

binding in both cases. Binding of NO to CoP (Eq. 3) is an ordinary radical


recombination, occurring on a single potential energy surface. By contrast,
the analogous reaction for MnP (Eq. 4) proceeds by crossing from the HS
energy surface (of the reactants) to the LS energy surface (of the product).
2
CoP + 2 NO ! 1 ½CoPðNOÞ (3)
6
MnP + NO ! ½MnPðNOÞ
2 1 (4)

In view of that difference, it is beneficial to consider reaction (4) in two vir-


tual steps: the first one being promotion of MnP from the ground state to the
excited doublet state (Eq. 5); the second one being radical recombination of
NO with the promoted state of MnP (Eq. 6):
6
MnP ! 2 MnP* (5)
2
MnP* + NO ! ½MnPðNOÞ
2 1
(6)

Note that, in terms of spin couplings, reactions (3) and (6) are analogous rad-
ical recombinations, and their energies should be reflected in the structural
parameters of both nitrosyl complexes and Mn/Co–NO stretching frequen-
cies. However, the experimentally measured Mn–NO bond energy is not
due to reaction (6), but rather reaction (4), i.e., it also contains the compo-
nent related to spin-state conversion energy of MnP (reaction (5)) which is not
reflected in the equilibrium properties of MnP(NO). This explains the lack of
the usual correlation between the bond distance and bond energy when
comparing MnP(NO) with CoP(NO). As was pointed out in Ref. 104,
MnP(NO) resembles isoelectronic FeIII–porphyrin NO complexes, earlier
described by Lehnert and coworkers,106,107 where the nitrosyl is also tightly
bound (from structure and vibrations), although the bond is weak (from
thermodynamics).
Accurate spin-state energetics 247

The problem with the presently discussed Mn–NO and Co–NO por-
phyrins becomes even more challenging when considering it quantitatively
by means of DFT calculations. As first shown by Jaworska and Lodowski,108
DFT calculations predict a reverse order of the bond energies compared
with that implied by the nitrosyl migration experiment (reaction (2)). This
tendency was confirmed by the present author in Ref. 104, where the com-
putational protocol has been considerably improved with respect to the ear-
lier work—by testing more functionals, wherever possible with the
Grimme’s dispersion corrections, including zero-point and thermal vibra-
tional effect, as well as relativistic effects, and exploring broken–symmetry
DFT solutions found for hybrid functionals. The computed bond energies
(ΔEMn–NO, ΔECo–NO) and Gibbs free energy of reaction (2) (ΔGmigr) are
summarized in Fig. 6. Whereas the experiment clearly shows that ΔGmigr
< 0, the majority of DFT(-D3) methods were found to considerably over-
bind the Mn–NO complex with respect to the Co–NO one, leading
ΔE or ΔG (kcal/mol)

Fig. 6 NO binding energies to CoP and MnP, and Gibbs free energy of migration reac-
tion (2) (kcal/mol) computed with a number of DFT methods, including dispersion cor-
rection (-D3) wherever implemented for a given functional; in such cases the effect of
dispersion is shown as the striped part of an M–NO bond energy bar, whereas only the
dispersion-corrected value is annotated. The experimental estimate of the Co–NO bond
energy109 is shown as the horizontal line, whereas the nitrosyl migration experiment
, M. Inorg.
(reaction 2) indicates that ΔGmigr < 0. Reprinted with permission from Radon
Chem. 2015, 54, 5634–5645, 10.1021/ic503109a. Copyright (2015) American Chemical
Society.
248 Mariusz Rado
n

(incorrectly) to positive values of ΔGmigr. Out of tested functionals, only


three give the correct sign of ΔGmigr, namely: PBE0, B3LYP, and M06.
However, the first two (even with the D3 correction) underestimate the
Co–NO bond energy compared with the experimental value of 23 kcal/
mol,109 whereas M06 gives unrealistically small Mn–NO bond energy. It
was thus confirmed in Ref. 104 that, indeed, none of the tested DFT
(-D3) methods give the NO binding energies for these two metal-porphyrin
complexes simultaneously correct. Additional calculations also demon-
strated that this anomaly cannot be explained by improving the computa-
tional models (including side substituents of the porphyrin ring,
accounting for solvation effects).104
As the present author argued in Ref. 104, the change of spin state
coupled to the formation of the Mn–NO bond, makes the Mn–NO bond
energy vulnerable to known shortcomings of DFT methods in the prediction
of spin-state energetics. By probing different exchange–correlation func-
tionals, it may be possible to find a single one which gives the energies of reac-
tions (3) and (6) simultaneously correct, but describing the spin-state
conversion energy of MnP (reaction (5)) is a completely different problem,
requiring a rather different choice of the functional. Indeed, it was shown
in Ref. 104—by comparison with the CCSD(T) benchmark data—that these
functionals which (upon supplementing with the D3 correction) reproduce
the Co–NO bond energy quite well, may either underestimate (e.g., B3LYP*
and TPSSh by almost 12 kcal/mol) or overestimate (e.g., M06 by 18 kcal/
mol) the spin-state conversion energy of MnP. These problems lead to signif-
icant errors (either overbinding or underbinding) on the ΔEMn–NO values
computed by using these functionals. The inclusion of appropriate spin-state
energy (SSE) corrections (for the energy of elementary reaction (5)) greatly
improves the results obtained from B3LYP*-D3, TPSSh-D3, and M06-D3
calculations, leading now to ΔECo–NO > ΔEMn–NO and negative values of
ΔGmigr, i.e., in agreement with the experimental results (Table 3)
Estimates of the Mn/Co–NO bond energies were also obtained from
high-level CCSD(T) calculations. While it was found computationally
too demanding to perform CCSD(T) calculations directly for MnP(NO)
and CoP(NO) complexes, such calculations were performed for their small
mimics, MnL2(NO) and CoL2(NO), and the appropriate extrapolation
method was applied to obtain the CCSD(T) estimates for the porphyrin
complexes (see Section 3.1 for description of the L2 mimics and the extrap-
olation method). We note that the CCSD(T) estimate of the Co–NO bond
energy is close (within 2 kcal/mol) to the experimental value (cf Table 3).
Although the Mn–NO bond energies obtained from different approaches
Accurate spin-state energetics 249

Table 3 Binding energiesa of NO to CoP and MnP, and Gibbs energy of NO migration
after including spin-state energy (SSE) correctionsb for MnP
ΔECo–NO ΔEMn–NO ΔGmigr
B3LYP*-D3 + SSE 21.6 17.4 6.1
TPSSh-D3 + SSE 23.4 23.2 2.1
M06-D3 + SSE 21.7 18.9 4.2
CCSD(T) estimates c
25.1 20.7 6.3
Experiment 23 <0
a
In kcal/mol.
b
The SSE correction (11.8 kcal/mol for B3LYP* and TPSSh, and +18.1 kcal/mol for M06) serves to
correct for an error of a given functional on the spin-state promotion energy of MnP (see text);
uncorrected values can be found in Fig. 6.
c
Estimates based on the small mimics (see text).
Source: Adapted with Permission from Rado n, M. Inorg. Chem. 2015, 54, 5634–5645, 10.1021/
ic503109a. Copyright (2015) American Chemical Society.

gathered in Table 3 slightly differ from each other and the experimental
value is not known quantitatively, it is remarkable that both the SSE-
corrected DFT results (using B3LYP*-D3, TPSSh-D3, and M06 func-
tionals) and the CCSD(T) estimates clearly confirm the experimental fact
that the Mn–NO bond is energetically weaker than the Co–NO one.104
The good agreement of the CCSD(T) estimate with the experimental
Co–NO bond energy is noteworthy because CoL2(NO) has rather signifi-
cant multireference character related to nondynamic correlation effects
in the Co–NO bond.104 For instance, the value of the D1 diagnostic
for CoL2(NO) is 0.43, i.e., considerably beyond the “safe regime” of
D1 < 0.15 often assumed in the literature. This suggests that these nitrosyl
complexes, despite having moderate multireference character, are still rea-
sonably described by the single-reference CCSD(T) method. The latter
conjecture resembles the one made previously for cobalt–nitrosyl complexes
and other similar observations in the literature (see Section 3.3). However,
given the importance and rather complicated nature of this problem, further
reinvestigation of these Mn/Co–NO porphyrin complexes by means of
even more advanced WFT calculations would be in order.

5. Benchmarking and environmental effects


For the majority of TM complexes discussed so far in this chapter, the
experimental data only allow to identify the ground state, but give no quan-
titative information on the relative energies of other spin states. Probing the
250 Mariusz Rado
n

ability to reproduce the experimental ground state is important, but not suf-
ficient for the purpose of method benchmarking. With the latter goal in
mind, it is important to focus on complexes for which quantitative experi-
mental data of spin-state energetics are available.
The appropriate experimental data can be obtained from two sources.
One of them are thermochemical parameters of SCO complexes, giving
information on the adiabatic energy difference between the involved spin
states. The SCO data were used several times to assess the performance of
DFT methods,25,26 but not for WFT methods (in part, due to the size of
experimentally characterized SCO molecules). Moreover, as Hughes and
Friesner110 pointed out, the experimental data available for SCO complexes
are not diverse enough to create a representative benchmark set for the prob-
lem of spin-state energetics in general: majority of known SCO complexes
are not sufficiently diversified (they all have LS ground state and a small gap
to the HS state; most of them contain an FeII ion ligated by six aromatic
nitrogens111). The representability can be increased by including also rela-
tive energies of spin-forbidden d–d excitations, giving information on the
vertical energy difference between the involved spin states. Although such
data are available for many TM complexes with diverse selection of ligands
and different ground states, they have been seldom used in the literature in
the context of spin-state energetics (some notable exceptions are Ref. 110
and a few other works discussed below). The present author currently works
on combining quantitative experimental data of spin-state energetics from
the above two sources in order to create a diverse benchmark set.
When using any kind of experimental data for benchmarking theory, it
may be crucial to account for the effects of environment (solution or crystal
lattice). Although these effects are routinely neglected in many theoretical
studies, our experience shows that they are not necessarily small for relative
spin-state energetics, as will be illustrated by examples below.

5.1 Benchmark study of aqua complexes


Recently, the present author with coworkers studied relative energies of
spin states (and other ligand-field states) for TM aqua complexes.80,112 Since
their optical spectra are rich in d–d bands, including those assigned to spin-
forbidden transitions, and most of the experimental data are well
established,113 aqua complexes seem to be ideal targets for benchmarking
theory. Surprisingly, large discrepancies with experiment were reported
in previous studies with WFT methods.
Accurate spin-state energetics 251

In the work by Yang et al.,114 the lowest d–d excitation energy of FeIII
aqua complex, attributed to the spin-forbidden transition 6A1g !4T1g, was
computed at 2.85 eV (CASPT2) or even at higher energy (CASSCF,
MRCI), compared with the experimental value of 1.56 eV, leading to a
huge computational error of 1.3 eV (or larger)! Related problems with this
excitation energy were observed earlier by other authors: Ghosh and
Taylor,115 and Neese et al.47, who obtained discrepancies with experiment
on the order of 0.5–1 eV when using CASPT2 and CCSD(T) (Ref. 115) or
SORCI method (Ref. 47), respectively. All these studies seemed to indicate
that even best WFT methods available cannot reproduce the vertical sextet–
quartet gap for FeIII aqua complex. This is unexpected and even more dis-
appointing when knowing that simple DFT calculations with the B3LYP
functional seem to reproduce this energy difference almost exactly.115,116
Where is the problem rooted?
Some of the mentioned authors, seemingly confident in high accuracy of
the WFT results, proposed that the problematic d–d band of FeIII aqua com-
plex might be, indeed, situated at the higher energy, close to the computed
values (where some other d–d bands are observed also), whereas the exper-
imental band at the lower energy might be caused by impurities, such like
hydrolysis products115 or small amounts of FeII complex produced by redox
equilibrium.114 However, both possibilities are unlikely when considered in
detail.80 Moreover, a careful examination of the computational results in
Ref. 114 shows that sizable discrepancies also appear for aqua complexes
of other metals (e.g., MnII, CoIII, CoII, VIII), suggesting that the problem
is more general.
As noticed by the present author in Ref. 80, all earlier computational
studies were performed for gaseous [M(H2O)6]n+ models (where Mn+ is a
TM ion). It was assumed that solvation effects on these d–d excitation ener-
gies are negligible, which was also supported by rough computational esti-
mates using continuous solvation models.114,115 By contrast, in our studies
on this topic we used models with explicit water molecules, following earlier
studies on the hydration of metal cations by Uudsemaa and Tamm117 and by
Markham et al.118
Fig. 7A shows the structure of [Fe(H2O)6]3+ complex with added a sec-
ond solvation sphere of 12 water molecules, interacting with the water
ligands by hydrogen bonds. The resulting [Fe(H2O)18]3+ cluster is embed-
ded in the continuous solvation model (COSMO)119 of water to describe
long–range solvation effects. Similar models [M(H2O)18]n+ were con-
structed for aqua complexes of other TM ions.
252 Mariusz Rado
n

A B

Fig. 7 (A) Structure of [Fe(H2O)18]3+ model, showing the network of hydrogen bonds
between the water ligands and second-sphere water molecules. (B) Structure of the
[Fe(H2O)6]3+ core from the optimized [Fe(H2O)18]3+ model (two views). (C) Structure
of [Fe(H2O)6]3+ model optimized in gas phase (two views). Adapted with permission from
Radon , M.; Góra-Marek, K.; Stępniewski, J.; Broclawik, E. J. Chem. Theory Comput. 2016,
12 (4), 1592–1605, 10.1021/acs.jctc.5b01234. Copyright (c) 2016 American Chemical
Society.

By comparing excitation energies obtained for such solvated models


with those for bare unsolvated ones, we demonstrated80,112 that solvation
effects may greatly influence the d–d excitation energies. In regard to the discussed
6
A1g ! 4T1g excitation energy for FeIII aqua complex, we found a difference
of 0.50 eV (4000 cm1, 11.4 kcal/mol) between the CASPT2 results com-
puted for unsolvated [Fe(H2O)6]3+ and solvated [Fe(H2O)18]3+ models. For
RuIII aqua complex the corresponding solvation effect on the lowest d–d
excitation 2T2g !4T1g energy amounts to 0.38 eV (3100 cm1, 8.9 kcal/
mol). Comparable solvation effects were also found at the DFT level with
two different functionals.80 On the example of FeIII complex, these solva-
tion effects were also shown to be important not only for vertical, but also for
adiabatic energies.80
For both FeIII and RuIII aqua complexes, the solvation leads to relative
stabilization of a lower-spin with respect to higher-spin state, i.e., the quartet
with respect to sextet state for FeIII (decreasing the excitation energy) or the
doublet with respect to quartet state for RuIII (increasing the excitation
energy). The observed sign of these effects is consistent with the structural
changes exerted by explicit solvation, namely significant contractions of the
metal–ligand bonds with respect to unsolvated model. In the case of FeIII
complex, as shown in Fig. 7B and C, the solvation effect leads to contraction
of the Fe–O bonds by 0.03 Å coupled with elongation of the O–H bonds in
water ligands. This suggests that the O atoms of water ligands become more
Accurate spin-state energetics 253

basic in the presence of solvent and thus they ligate the central ion more
strongly than in gas phase.80 We also found that when the second layer of
water molecules is removed from the optimized [M(H2O)18]3+ structures,
but keeping the effect it exerts on the first coordination sphere, we still
reproduce more than 85% of the total solvation effect for FeIII and essentially
the whole effect for RuIII complex.80 Moreover, about 50% of the solvation
effect for FeIII can be reproduced for [Fe(H2O)6]3+ model (without explicit
water molecules) by reoptimizing its geometry in the continuous solvation
model80,112 (by contrast to the negligible solvation effect found before in
single-point calculations for the gaseous geometry115,116).
Thus, the solvation effects observed for aqua complexes are mainly of structural
origin: the presence of solvent affects the geometry of the first coordination
sphere, leading (among other changes) to contraction of the metal–ligand
bonds and thus to an increase of the t2g–eg gap. Since the discussed d–d exci-
tations for FeIII and RuIII aqua complexes involve redistribution of electrons
between the t2g and eg levels, their energies should be very sensitive to such
changes, as indeed observed in the calculations, provided that the molecular
geometry is reoptimized using the appropriate solution model. Our work in
Ref. 80 contradicts a widespread presumption that solvation effects should
be small for d–d excitations due to their localized character (i.e., a small effect
on the electronic density, by contrast to charge transfer excitations). These
effects can be large due to these solvent-induced geometry changes, and
therefore a common practice of performing calculations in solution for fixed
gas phase geometry is unjustified in this context.80
A more general survey in Ref. 112, including complexes of various cat-
ions and different types of electronically excited states, revealed that the
discussed solvation effects are more significant for energies of t2g ! eg and
eg ! t2g transitions (like those discussed above for FeIII and RuIII complexes)
than for spin-flip transitions (not involving redistribution of electrons between
t2g and eg levels), and they are typically larger for + 3 than for + 2 cations. In the
case of NiII, negligible solvation effects were found, explaining the success of
earlier calculations for the unsolvated [Ni(H2O)6]2+ model.114,120,121
The treatment of solvation effects beyond the standard continuum
approximation is not the only improvement made in our studies of aqua
complexes80,112 compared with the earlier reports. We also used larger basis
sets and more balanced active spaces (for multireference calculations), and
we correlated the metal outer-core electrons. By making all these improve-
ments, we were able to bring back a good agreement between high-level
WFT calculations and the experiment. To evaluate the accuracy of CASPT2
254 Mariusz Rado
n

and NEVPT2 methods in quantitative terms, we compiled a benchmark set


of vertical excitation energies for 24 spin-allowed (ΔS ¼ 0) and 19 spin-
forbidden transitions (18 with jΔSj ¼ 1 and one with jΔSj ¼ 2) of first-
row TM aqua complexes studied in Ref. 112. When tested against this
benchmark set, the CBS-extrapolated CASPT2 and NEVPT2 excitation
energies have mean absolute errors of only 0.15 and 0.13 eV, and maximum
errors of 0.56 and 0.42 eV, respectively.112 Among outliers responsible for
the largest errors, we identified the already discussed 6A1g ! 4T1g transition
for FeIII aqua complex and three transitions for CoIII aqua complex (in the
latter case, limitations of the presently used solvation model may be partly
responsible112). Note that even when considering these maximum errors,
the accuracy was greatly improved compared with previous studies allowing
to reproduce all the experimental results at least qualitatively correct. This is
a good message not only for theory. Indirectly, it also confirms that the weak
bands experimentally observed for aqua complexes are, indeed, caused by
the occurrence of d–d transitions in their [M(H2O)6]n+ chromophores,112
and not by the presence of impurities or experimental errors, which was
suggested in some of the earlier theoretical studies.114,115
The most important message from our studies of aqua complexes is that
the correct treatment of solvation effects becomes indispensable for fair
benchmarking of quantum chemical methods with respect to the experi-
mental data. This is illustrated in Fig. 8 where the lowest d–d excitation
energies of [Fe(H2O)6]3+ (6A1g !4T1g) and [Ru(H2O)6]3+ (2T2g !4T1g)
are analyzed at various theory levels. Note that computed values shown
in this figure refer to unsolvated [M(H2O)6]3+ models in gas phase, and
hence should not be directly compared with raw experimental values, but
rather with back-corrected experimental values (i.e., after subtracting the
estimates of the solvation effects discussed above).
As shown in Fig. 8A, the sextet–quartet splitting in the case of FeIII com-
plex is considerably method dependent. With strong variations of the DFT
results, some of them are—by pure accident—very close to the uncorrected
experimental value. Naive comparing the results for unsolvated model with
the experimental results obtained in solution would thus lead to an errone-
ous conclusion that OLYP and B3LYP functionals are the most accurate
methods for this problem, whereas all WFT methods considerably over-
estimate the excitation energy. This is, however, an artefact of not treating
the solvation effects appropriately. When comparing with the correct refer-
ence (i.e., back-corrected for solvation effects), it turns out that B3LYP and
OLYP have a tendency to strongly underestimate the excitation energy
Accurate spin-state energetics 255

A B

Fig. 8 Vertical excitation energies (A) 6A1g !4T1g for [Fe(H2O)6]3+ and (B) 2T2g !4T1g for
[Ru(H2O)6]3+ computed with selected DFT and WFT methods. The experimental data are
shown as solid lines and those back-corrected for solvation effects (see text) as dashed
lines. The WFT results were obtained with the triple-ζ (Fe, first sphere O atoms) or
double-ζ (second-sphere O atoms, H atoms) basis set; CASPT2 and NEVPT2 results con-
tain additive correction to approach the CBS limit, estimated from the difference
between CCSD(T)-F12 and standard CCSD(T) calculations. Adapted with Permission
Radon , M.; Góra-Marek, K.; Stępniewski, J.; Broclawik, E. J. Chem. Theory Comput. 2016,
12 (4), 1592–1605, 10.1021/acs.jctc.5b01234. Copyright (2016) American Chemical Society.

(i.e., overstabilize the quartet with respect to the sextet state) and this bias is
even larger for TPSSh and B3LYP* functionals. Although the latter two
functionals are usually recommended for spin-state energetics based on
the experience with SCO complexes,24–26 other functionals, like M06,
M06L, and B2PLYP (double-hybrid), perform significantly better for the
presently discussed case of the sextet–quartet splitting for FeIII aqua complex.
Among the tested WFT methods, the best result is obtained from CCSD
(T)-F12, whereas NEVPT2 and CASPT2 methods (even when corrected
to the CBS limit) give slightly larger errors for this particularly challenging
iron complex. The reliability of the CCSD(T) approach was further
supported by comparison with higher levels of CC theory, up to the
CCSDT(Q) in Ref. 112. This is a rare example where the latter,
256 Mariusz Rado
n

computationally very expensive method could be applied to study spin-state


energetics of a TM complex, and hence obtaining a good agreement with
the standard CCSD(T) calculations is a good message for the accuracy of the
latter method.
Concerning the doublet–quartet splitting for RuIII complex shown in
Fig. 8B, it turns out that variation of results with the choice of method is
smaller than for the FeIII complex. However, nearly all the theoretical results
appear to be underestimated when compared with the uncorrected exper-
imental reference. The agreement is considerably improved when compar-
ing these results with the correct (i.e., solvent back-corrected) experimental
reference.

5.2 Other examples of environmental effects


Significant solvation effects are not limited to aqua complexes. The present
author found that for a typical FeII SCO complex, [Fe(tacn)2]2+ where tacn
is 1,4,7-triazacyclononane, simple reoptimization of geometry in the
COSMO solvent model (water) results in relative stabilization of the LS (sin-
glet) with respect to the HS (quintet) state by nearly 3 kcal/mol in terms of
adiabatic energy.101 Relative stabilization of the LS state agrees with the
effect observed for aqua complexes and is consistent with a slight contraction
of the Fe–N bonds observed in the calculations. However, in variance to the
case of aqua complexes, the change of geometry is not a principal factor
determining the solvation effect for [Fe(tacn)2]2+, since the effect can be
reproduced up to 0.2 kcal/mol by single-point COSMO calculations for
the gaseous structures.101
For related FeIII complex with the same ligands, [Fe(tacn)2]3+, Swart
reported an even larger solvation effect of 6.5 kcal/mol (in methanol,
COSMO model), also leading to relative stabilization of the LS with respect
to the HS state.30 Interestingly, he also found mutually opposite solvation
effects for two similar FeII complexes, [Fe(amp)2Cl2] and [Fe(dmp)2]2+, where
amp ¼ 2-pyridylmethylamine, dmp ¼ di(2-pyridylmethyl)amine. The
behavior of [Fe(dmp)2]2+ might be considered as typical: the solvent (meth-
anol) stabilizes the LS (singlet) state with respect to the HS state by 4.2 kcal/
mol and with respect to the IS (triplet) state by 3.2 kcal/mol. However, for [Fe
(amp)2Cl2] relative stabilization of the HS state was observed: by 2.2 kcal/mol
with respect to the LS state and by 5.4 kcal/mol with respect to the IS state.30
Gruden et al. also reported many examples of TM complexes where
solvent can selectively stabilize either lower-spin or higher-spin states,39
Accurate spin-state energetics 257

showing that these solvation effects are system-specific and not easily
predictable before performing actual calculations. Also, in some cases the
solvation effects can be negligibly small. For ferrocene the energies of
d–d transitions, including those due to spin-forbidden singlet–triplet excita-
tions, are experimentally known to be almost the same in condensed phases
as is gas phase,122,123 and this behavior is confirmed by our recent
calculations.101
Analogous effects of the molecular environment can also be important in
solid state. For [Fe(H2O)6]3+ cation, the present author compared the posi-
tion of the above discussed sextet–quartet band in solution and in the crys-
talline phase of perchlorate trihydrate.101 To this end, the CASPT2
excitation energies were computed for: (a) gaseous [Fe(H2O)6]3+; (b) the
[Fe(H2O)18]3+ model (describing the aqueous solution, see Section 5.1);
and (c) the [Fe(H2O)6(ClO4)6(H2O)6]3 cluster extracted from the crystal
structure. All three structures were earlier optimized the same DFT level of
theory (PBE/pob-TZVP), the crystal one using a periodic DFT code. The
excitation energies were found quite similar for the solution and crystal
structure, and for both of them very different than for the gaseous cation.
Interestingly, the 6A1g !4T1g excitation energy was computed lower by
ca. 500 cm1 in solution than in the crystalline phase, which agrees very well
with the experimentally observed difference of ca. 400 cm1.101
Going back to the case of triplet–quintet adiabatic energy difference for
FeII porphyrin (discussed in Section 3.1), the present author performed peri-
odic DFT calculations for a model of crystal structure of Fe(TPP) in com-
parison with analogous calculations for a single gaseous Fe(TPP)
molecule.101 This revealed the effect of 3.5 kcal/mol relative stabilization
of the triplet state in the crystal structure, which is nicely correlated to con-
traction of the Fe–N distance by 0.02 Å with respect to the isolated mole-
cule. This packing effect may be relevant for the benchmarking of high-level
WFT methods, some of which—when applied to the isolated molecule—
give the quintet state too low in energy (see Table 1). However, even after
considering this additional stabilization of the triplet state by 3.5 kcal/mol
(due to the crystal environment), the results of CASPT2 and RASPT2 cal-
culations in Table 1 can hardly satisfy the experimentally derived inequality
on the triplet–quintet energy difference, by contrast to the CCSD(T) calcu-
lations favoring the triplet state already in gas phase.
In a further related study, Keep found that the use of COSMO model
may relatively stabilize either the triplet or quintet state of FeP depending
on the selection of iron atomic radius.1 He also found that the coordination
258 Mariusz Rado
n

of two tetrahydrofuran solvent molecules as axial ligands added to FeP rel-


atively stabilizes the quintet state by 7 kcal/mol. Indeed, FeP(THF)2 has the
quintet experimental ground state124 and the change of spin state with
respect to FeP suggests that the quintet–triplet splitting of the latter species
should be small, possibly close to the value of  2 kcal/mol obtained from
our CCSD(T) calculations (cf Table 1).
It is clear that somewhat analogous effects of solvent or crystalline envi-
ronment on the structure of TM complexes (or metalloenzyme models) as
discussed here have been recognized in the literature already for some time.
As a remarkable example, Deeth and coworkers125 demonstrated that the
use of solvent-optimized geometries is mandatory to reproduce structural
trends, such as the trans effect and sterically induced bond elongations, in
several series of heteroligand complexes. Also the mentioned works by
Swart and Gruden with their coworkers30,39 fit into this context. Nonethe-
less, our study of aqua complexes and other examples presented in this sec-
tion show that energetic implications of these solvent-induced structural
changes are not fully appreciated in the computational chemistry commu-
nity and it seems to be the “standard approximation” to neglect them in
benchmark studies of spin-state energetics, even recently.126 Moreover, it
is not clear how these environmental effects can be modeled reliably, espe-
cially in the solid state. Our recent studies suggest that these effects are
mainly of structural origin, although the influence of long-range electrostatic
potential from the lattice may also be of some importance.127,128 Ongoing
research in our group is focused on devising appropriate models to account
for these environmental perturbations when benchmarking theory against
quantitative experimental data of spin-state energetics obtained in con-
densed phases. In addition to having this significant benchmark value, the
effort in this direction may be also beneficial to better understand the role
of environmental effects in selective stabilization of spin states for biological
systems and SCO materials (e.g., the cooperativeness of SCO transitions).

6. Conclusions
This chapter was focused on high-level computation of spin-state
energetics for TM complexes and models of TM sites in enzymes,
attempting to illustrate the key importance of computational accuracy for
correct interpretation of experimental data and clarifying the mechanistic
controversies. Even in the most puzzling case of [Mo(CN)6]4 anion, where
we were not able to fully resolve the structure vs spin-state controversy,
Accurate spin-state energetics 259

quantum chemical calculations turned out to be helpful for recasting this dis-
crepancy in a new light. The presented examples confirmed that the optimal
choice of functional in DFT modeling is highly system-dependent. This
underlined the importance of developing accurate, yet computationally effi-
cient protocols based on the WFT method. In our experience, the
CCSD(T) method appears to be particularly promising in terms of accuracy,
even for systems showing appreciable multireference character. Neverthe-
less, additional studies are still required to qualitatively assess the error bars
for computational predictions of spin-state energetics. It would be particu-
larly interesting to analyze, in a systematic way, how the performance of dif-
ferent methods vary for spin-state energetics of TM complexes with
growing amount of nondynamic correlation. Recognizing the rareness of
truly quantitative benchmark studies of spin-state energetics, especially those
accounting properly for solvation or crystal-packing effects, the present
author works on creating a balanced benchmark set of quantitative,
environment-corrected experimental data of spin-state energetics, and using
it for quantifying the accuracy of selected WFT and DFT methods.

Acknowledgments
This work was supported in part by grants from: National Science Center, Poland (2017/26/
D/ST4/00774); Ministry of Science and Higher Education, Poland; Jagiellonian University
(K/DSC/004554); and by PLGrid Infrastructure (computations have been performed in part
on resources provided by ACC Cyfronet AGH/UST). The author is thankful to Prof. Ewa
Broclawik for valuable remarks on the manuscript.

References
1. Kepp, K. P. Coord. Chem. Rev. 2013, 257(1), 196–209. https://doi.org/10.1016/j.
ccr.2012.04.020.
2. Sorai, M. Bull. Chem. Soc. Jpn 2001, 74(12), 2223–2253. https://doi.org/10.1246/
bcsj.74.2223.
3. G€utlich, P.; Goodwin, H. A. In Spin Crossover in Transition Metal Compounds I;
G€utlich, P., Goodwin, H. A., Eds.; Springer: Berlin, Heidelberg, 2004 pp 1–47.
https://doi.org/10.1007/b13527.
4. Shaik, S.; Chen, H.; Janardanan, D. Nat. Chem. 2011, 3, 19–27. https://doi.org/
10.1038/nchem.943.
5. Góra-Marek, K.; Stępniewski, A.; Rado n, M.; Broclawik, E. Phys. Chem. Chem. Phys.
2014, 16, 24089–24098. https://doi.org/10.1039/C4CP03350G.
6. Harvey, J. N. WIREs Comput. Mol. Sci. 2014, 4(1), 1–14. https://doi.org/10.1002/
wcms.1154.
7. Letard, J. F.; Guionneau, P.; Goux-Capes, L. Top. Curr. Chem. 2004, 235, 221–249.
8. Li, J.; Struzhkin, V. V.; Mao, H. K.; Shu, J.; Hemley, R. J.; Fei, Y.; Mysen, B.; Dera, P.;
Prakapenka, V.; Shen, G. Proc. Natl. Acad. Sci. 2004, 101(39), 14027–14030. https://
doi.org/10.1073/pnas.0405804101.
9. Swart, M. Int. J. Quantum Chem. 2013, 113, 2–7. https://doi.org/10.1002/qua.24255.
260 Mariusz Rado
n

10. Costas, M.; Harvey, J. N. Nat. Chem. 2013, 5, 7–9. https://doi.org/10.1038/


nchem.1533.
11. Neese, F. JBIC J. Biol. Inorg. Chem. 2006, 11, 702–711. https://doi.org/10.1007/
s00775-006-0138-1.
12. Cramer, C. J.; Truhlar, D. G. Phys. Chem. Chem. Phys. 2009, 11, 10757–10816.
https://doi.org/10.1039/b907148b.
13. Riplinger, C.; Neese, F. J. Chem. Phys. 2013, 138(3), 034106. https://doi.org/
10.1063/1.4773581.
14. Schwilk, M.; Ma, Q.; K€ oppl, C.; Werner, H. J. J. Chem. Theory Comput. 2017, 13(8),
3650–3675. https://doi.org/10.1021/acs.jctc.7b00554.
15. Koch, W.; Holthausen, M. C. A Chemist’s Guide to Density Functional Theory, 2nd ed.;
WILEY-VCH Verlag GmbH, 2001.
16. Becke, A. D. J. Chem. Phys. 2014, 140(18), 18A301. https://doi.org/
10.1063/1.4869598.
17. Goerigk, L.; Grimme, S. WIREs Comput. Mol. Sci. 2014, 4, 576–600. https://doi.org/
10.1002/wcms.1193.
18. Grimme, S. WIREs Comput. Mol. Sci. 2011, 1, 211–228. https://doi.org/10.1002/
wcms.30.
19. Rappoport, D.; Crawford, N. R. M.; Furche, F.; Burke, K. In Encyclopedia of
Inorganic Chemistry; Scott, R. A. Ed.; Wiley, 2009. https://doi.org/10.1002/
0470862106.ia615.
20. Harvey, J. N. Annu. Rep. Prog. Chem., Sect. C Phys. Chem. 2006, 102, 203–226.
https://doi.org/10.1039/b419105f.
21. Ghosh, A. JBIC J. Biol. Inorg. Chem. 2006, 11, 712–724.
22. Harvey, J. N. Struct. Bonding 2004, 112, 151–183. https://doi.org/10.1007/b97939.
23. Paulsen, H.; Duelund, L.; Winkler, H.; Toftlund, H.; Trautwein, A. X. Inorg. Chem.
2001, 40(9), 2201–2203. https://doi.org/10.1021/ic000954q.
24. Reiher, M.; Salomon, O.; Hess, B. A. Theor. Chem. Acc. 2001, 107(1), 48–55. https://
doi.org/10.1007/s00214-001-0300-3.
25. Kepp, K. P. Inorg. Chem. 2016, 55(6), 2717–2727. https://doi.org/10.1021/acs.
inorgchem.5b02371.
26. Jensen, K. P.; Cirera, J. J. Phys. Chem. A 2009, 113, 10033–10039. https://doi.org/
10.1021/jp900654j.
27. Swart, M. Inorg. Chim. Acta 2007, 360, 179–189. https://doi.org/10.1016/j.
ica.2006.07.073.
28. Rado n, M. Phys. Chem. Chem. Phys. 2014, 16, 14479–14488. https://doi.org/
10.1039/C3CP55506B.
29. Pinter, B.; Chankisjijev, A.; Geerlings, P.; Harvey, J. N.; Proft, F. D. Chem. Eur. J.
2018, 24(20), 5281–5292. https://doi.org/10.1002/chem.201704657.
30. Swart, M. J. Chem. Theory Comput. 2008, 4, 2057–2066. https://doi.org/10.1021/
ct800277a.
31. Conradie, J.; Ghosh, A. J. Phys. Chem. B 2007, 111, 12621–12624. https://doi.org/
10.1021/jp074480t.
32. Conradie, M. M.; Conradie, J.; Ghosh, A. J. Inorg. Biochem. 2011, 105(1), 84–91.
https://doi.org/10.1016/j.jinorgbio.2010.09.010.
33. Handy, N. C.; Cohen, A. J. Mol. Phys. 2001, 99(5), 403–412. https://doi.org/
10.1080/00268970010018431.
34. Ruiz, E.; Martin, A.; Cirera, J. Spin Crossover Complexes: A Challenge from Theory
to Single-Molecule Devices. Presented at 11th WATOC Conference, M€ unich, Germany.
2017.
35. Swart, M.; Solà, M.; Bickelhaupt, F. M. J. Comput. Methods Sci. Eng. 2009, 9(1–2),
69–77. https://doi.org/10.3233/JCM-2009-0230.
Accurate spin-state energetics 261

36. Swart, M.; Solá, M.; Bickelhaupt, F. M. J. Chem. Phys. 2009, 131(9), 094103. https://
doi.org/10.1063/1.3213193.
37. Swart, M. Chem. Phys. Lett. 2013, 580, 166–171. https://doi.org/10.1016/j.
cplett.2013.06.045.
38. Daul, C.; Zlatar, M.; Gruden-Pavlovic, M.; Swart, M. In Spin States in Biochemistry and
Inorganic Chemistry; Swart, M., Costas, M., Eds.; Chapter 2. John Wiley & Sons, Ltd,
2015 pp 7–34. https://doi.org/10.1002/9781118898277.ch2.
39. Gruden, M.; Stepanovic, S.; Swart, M. J. Serbian Chem. Soc. 2015, 80(11), 1399–1410.
https://doi.org/10.2298/JSC150611068G.
40. Swart, M.; Gruden, M. Acc. Chem. Res. 2016, 49(12), 2690–2697. https://doi.org/
10.1021/acs.accounts.6b00271.
41. Paldus, J.; Li, X. In Advanced in Chemical Physics; Prigogine, I., Rice, S. A., Eds.;
Vol. 110; Wiley, 2007; pp 1–175. https://doi.org/10.1002/9780470141694.ch1.
42. Bartlett, R. J.; Musial, M. Rev. Mod. Phys. 2007, 79, 291–352. https://doi.org/
10.1103/RevModPhys.79.291.
43. Lee, T. J.; Scuseria, G. E. In Quantum Mechanical Electronic Structure Calculations with
Chemical Accuracy; Langhoff, S. R. Ed.; Springer Netherlands: Dordrecht, 1995
pp 47–108. https://doi.org/10.1007/978-94-011-0193-6_2.
44. Harvey, J. N. JBIC J. Biol. Inorg. Chem. 2011, 16, 831–839. https://doi.org/10.1007/
s00775-011-0786-7.
45. Knowles, P. J.; Hampel, C.; Werner, H. J. J. Chem. Phys. 1993, 99(7), 5219–5227.
https://doi.org/10.1063/1.465990.
46. Shamasundar, K. R.; Knizia, G.; Werner, H. J. J. Chem. Phys. 2011, 135(5), 054101.
https://doi.org/10.1063/1.3609809.
47. Neese, F.; Petrenko, T.; Ganyushin, D.; Olbrich, G. Coord. Chem. Rev. 2007,
251(3–4), 288–327. https://doi.org/10.1016/j.ccr.2006.05.019.
48. Andersson, K.; Roos, B. O. Int. J. Quantum Chem. 1993, 45(6), 591–607. https://doi.
org/10.1002/qua.560450610.
49. Ghigo, G.; Roos, B.; Malmqvist, P.-Å. Chem. Phys. Lett. 2004, 396, 142–149. https://
doi.org/10.1016/j.cplett.2004.08.032.
50. Angeli, C.; Cimiraglia, R.; Malrieu, J.-P. J. Chem. Phys. 2002, 117(20), 9138–9153.
https://doi.org/10.1063/1.1515317.
51. Pierloot, K.; Phung, Q. M.; Domingo, A. J. Chem. Theory Comput. 2017, 13, 537–553.
https://doi.org/10.1021/acs.jctc.6b01005.
€ Pittner, J. J. Phys. Chem. Lett.
52. Veis, L.; Antalı́k, A.; Brabec, J.; Neese, F.; Legeza, O.;
2016, 7(20), 4072–4078. https://doi.org/10.1021/acs.jpclett.6b01908.
53. Aoto, Y. A.; de Lima Batista, A. P.; K€ ohn, A.; de Oliveira-Filho, A. G. S. J. Chem.
Theory Comput. 2017, 13(11), 5291–5316. https://doi.org/10.1021/acs.jctc.7b00688.
54. Veryazov, V.; Malmqvist, P. Å.; Roos, B. O. Int. J. Quantum Chem. 2011, 111(13),
3329–3338. https://doi.org/10.1002/qua.23068.
55. Pierloot, K. Int. J. Quantum Chem. 2011, 111, 3291–3301. https://doi.org/10.1002/
qua.23029.
56. Pierloot, K. Mol. Phys. 2003, 101(13), 2083–2094. https://doi.org/10.1080/
0026897031000109356.
57. Vancoillie, S.; Zhao, H.; Tran, V. T.; Hendrickx, M. F. A.; Pierloot, K. J. Chem. Theory
Comput. 2011, 7, 3961–3977. https://doi.org/10.1021/ct200597h.
58. Olivares-Amaya, R.; Hu, W.; Nakatani, N.; Sharma, S.; Yang, J.; Chan, G. K.-L.
J. Chem. Phys. 2015, 142(3), 034102. https://doi.org/10.1063/1.4905329.
59. Stein, C. J.; Reiher, M. J. Chem. Theory Comput. 2016, 12(4), 1760–1771. https://doi.
org/10.1021/acs.jctc.6b00156.
60. Bao, J. J.; Dong, S. S.; Gagliardi, L.; Truhlar, D. G. J. Chem. Theory Comput. 2018,
14(4), 2017–2025. https://doi.org/10.1021/acs.jctc.8b00032.
262 Mariusz Rado
n

61. Feller, D.; Peterson, K. A.; Hill, J. G. J. Chem. Phys. 2011, 135(4), 044102. https://doi.
org/10.1063/1.3613639.
62. Jiang, W.; Wilson, A. K. J. Chem. Phys. 2011, 134(3), 034101. https://doi.org/
10.1063/1.3514031.
63. H€attig, C.; Klopper, W.; K€ ohn, A.; Tew, D. P. Chem. Rev. 2012, 112(1), 4–74.
https://doi.org/10.1021/cr200168z.
64. Peterson, K.; Feller, D.; Dixon, D. Theor. Chem. Acc. 2012, 131(1). https://doi.org/
10.1007/s00214-011-1079-51079.
65. Knizia, G.; Adler, T. B.; Werner, H.-J. J. Chem. Phys. 2009, 130(5), 054104. https://
doi.org/10.1063/1.3054300.
66. Bross, D. H.; Hill, J. G.; Werner, H.-J.; Peterson, K. A. J. Chem. Phys. 2013, 139(9).
https://doi.org/10.1063/1.4818725094302.
67. Fang, Z.; Vasiliu, M.; Peterson, K. A.; Dixon, D. A. J. Chem. Theory Comput. 2017, 13,
1057–1066. https://doi.org/10.1021/acs.jctc.6b00971.
68. Jiang, W.; DeYonker, N. J.; Wilson, A. K. J. Chem. Theory Comput. 2012, 8, 460–468.
https://doi.org/10.1021/ct2006852.
69. Sprague, M. K.; Irikura, K. K. Theor. Chem. Acc. 2014, 133(9), 1544. https://doi.org/
10.1007/s00214-014-1544-z.
70. Neese, F.; Liakos, D.; Ye, S. JBIC J. Biol. Inorg. Chem. 2011, 16(6), 821–829. https://
doi.org/10.1007/s00775-011-0787-6.
71. Harvey, J. N.; Tew, D. P. Int. J. Mol. Sci. 2013, 354–355, 263–270. https://doi.org/
10.1016/j.ijms.2013.07.011.
72. Rado n, M.; Broclawik, E.; Pierloot, K. J. Chem. Theory Comput. 2011, 7, 898–908.
https://doi.org/10.1021/ct1006168.
73. Vancoillie, S.; Zhao, H.; Rado n, M.; Pierloot, K. J. Chem. Theory Comput. 2010, 6(2),
576–582. https://doi.org/10.1021/ct900567c.
74. Lawson Daku, L. M.; Aquilante, F.; Robinson, T. W.; Hauser, A. J. Chem. Theory
Comput. 2012, 8(11), 4216–4231. https://doi.org/10.1021/ct300592w. http://pubs.
acs.org/doi/abs/10.1021/ct300592w.
75. Kepenekian, M.; Robert, V.; Le Guennic, B. J. Chem. Phys. 2009, 131, 114702.
https://doi.org/10.1063/1.3211020.
76. Phung, Q. M.; Feldt, M.; Harvey, J. N.; Pierloot, K. J. Chem. Theory Comput. 2018,
14(5), 2446–2455. https://doi.org/10.1021/acs.jctc.8b00057.
77. Rado n, M. J. Chem. Theory Comput. 2014, 10, 2306–2321. https://doi.org/10.1021/
ct500103h.
78. Petit, A. S.; Pennifold, R. C. R.; Harvey, J. N. Inorg. Chem. 2014, 53(13), 6473–6481.
https://doi.org/10.1021/ic500379r.
79. Stępniewski, A.; Rado n, M.; Góra-Marek, K.; Broclawik, E. Phys. Chem. Chem. Phys.
2016, 18(5), 3716–3729. https://doi.org/10.1039/c5cp07452e.
80. Rado n, M.; Ga˛ssowska, K.; Szklarzewicz, J.; Broclawik, E. J. Chem. Theory Comput.
2016, 12(4), 1592–1605. https://doi.org/10.1021/acs.jctc.5b01234.
81. Rado n, M.; Brocławik, E. In Computational Methods to Study the Structure and Dynamics
of Biomolecules and Biomolecular Processes - From Bioinformatics to Molecular Quantum
Mechanics; Liwo, A. Ed.; Springer series on Bio-/Neuroinformatics; Vol. 1;
Springer: Berlin, Heidelberg, 2014; pp 711–782. https://doi.org/10.1007/978-3-
642-28554-7_21.
82. Kepp, K. P. Phys. Chem. Chem. Phys. 2017, 19, 22355–22362. https://doi.org/
10.1039/C7CP03285D.
83. Rado n, M.; Pierloot, K. J. Phys. Chem. A 2008, 112(46), 11824–11832. https://doi.
org/10.1021/jp806075b.
84. Choe, Y.-K.; Nakajima, T.; Hirao, K.; Lindh, R. J. Chem. Phys. 1999, 111(9),
3837–3845. https://doi.org/10.1063/1.479687.
Accurate spin-state energetics 263

85. Ghosh, A.; Persson, B. J.; Taylor, P. R. JBIC J. Biol. Inorg. Chem. 2003, 8, 507–511.
86. Olah, J.; Harvey, J. J. Phys. Chem. A 2009, 113, 7338–7345. https://doi.org/10.1021/
jp811316n.
87. Johansson, M. P.; Sundholm, D. J. Chem. Phys. 2004, 120(7), 3229–3236. https://doi.
org/10.1063/1.1640343.
88. Li, J.; Lord, R.; Noll, B.; Baik, M.-H.; Schulz, C.; Scheidt, W. Angew. Chem. Int. Ed.
2008, 47(52), 10144–10146. https://doi.org/10.1002/anie.200804116.
89. Li, J.; Peng, Q.; Barabanschikov, A.; Pavlik, J. W.; Alp, E. E.; Sturhahn, W.; Zhao, J.;
Sage, J. T.; Scheidt, W. R. Inorg. Chem. 2012, 51(21), 11769–11778. https://doi.org/
10.1021/ic301719v.
90. Swart, M.; G€ uell, M.; Solá, M. In Quantum Biochemistry; Matta, C. F. Ed.; Vol. 2;
Chapter 19. Wiley-Blackwell, 2010; pp 551–583. https://doi.org/10.1002/
9783527629213.ch19.
91. Shaik, S.; Kumar, D.; de Visser, S. P.; Altun, A.; Thiel, W. Chem. Rev. 2005, 105(6),
2279–2328.
92. Ge, Y.; Gordon, M. S.; Piecuch, P.; Włoch, M.; Gour, J. R. J. Phys. Chem. A 2008,
112(46), 11873–11884. https://doi.org/10.1021/jp806029z.
93. Szklarzewicz, J.; Matoga, D.; Niezgoda, A.; Yoshioka, D.; Mikuriya, M. Inorg. Chem.
2007, 46, 9531–9533. https://doi.org/10.1021/ic701395r.
94. Rado n, M.; Rejmak, P.; Fitta, M.; Bałanda, M.; Szklarzewicz, J. Phys. Chem. Chem.
Phys. 2015, 17, 14890–14902. https://doi.org/10.1039/c4cp04863f.
95. Kaupp, M. Chem. Eur. J. 1998, 4(9), 1678–1686. https://doi.org/10.1002/
(SICI)1521-3765(19980904)4:9<1678::AID-CHEM1678>3.0.CO;2-N.
96. Pfennig, V.; Seppelt, K. Science 1996, 271, 626–628. https://doi.org/10.1126/
science.271.5249.626.
97. Seppelt, K. Acc. Chem. Res. 2003, 36(2), 147–153. https://doi.org/10.1021/
ar020052o.
98. Beauvais, L. G.; Long, J. R. J. Am. Chem. Soc. 2002, 124(10), 2110–2111. https://doi.
org/10.1021/ja0175901.
99. Broclawik, E.; Góra-Marek, K.; Rado n, M.; Bučko, T.; Stępniewski, A. J. Mol. Model.
2017, 23(5), 160. https://doi.org/10.1007/s00894-017-3322-z.
100. Pratt, C. S.; Coyle, B. A.; Ibers, J. A. J. Chem. Soc. A 1971, 2146–2151. https://doi.
org/10.1039/J19710002146.
101. Rado n, M.Environmental effects on spin-state energetics of transition metal complexes.n.d
(unpublished results).
102. Blomberg, M. R. A.; Borowski, T.; Himo, F.; Liao, R.-Z.; Siegbahn, P. E. M. Chem.
Rev. 2014, 114(7), 3601–3658. https://doi.org/10.1021/cr400388t.
103. Wójcik, A.; Rado n, M.; Borowski, T. J. Phys. Chem. A 2016, 120(8), 1261–1274.
https://doi.org/10.1021/acs.jpca.5b12311.
104. Rado n, M. Inorg. Chem. 2015, 54, 5634–5645. https://doi.org/10.1021/ic503109a.
105. Zavarine, I. S.; Kini, A. D.; Morimoto, B. H.; Kubiak, C. P. J. Phys. Chem. B 1998,
102(37), 7287–7292. https://doi.org/10.1021/jp9811169.
106. Goodrich, L. E.; Paulat, F.; Praneeth, V. K. K.; Lehnert, N. Inorg. Chem. 2010, 49(14),
6293–6316. https://doi.org/10.1021/ic902304a.
107. Praneeth, V. K. K.; Paulat, F.; Berto, T. C.; George, S. D.; N€ather, C.; Sulok, C. D.;
Lehnert, N. J. Am. Chem. Soc. 2008, 130(46), 15288–15303. https://doi.org/10.1021/
ja801860u.
108. Jaworska, M.; Lodowski, P. Struct. Chem. 2012, 23, 1333–1348. https://doi.org/
10.1007/s11224-012-0053-8.
109. Zhu, X. Q.; Li, Q.; Hao, W. F.; Cheng, J.-P. J. Am. Chem. Soc. 2002, 124(33),
9887–9893.
264 Mariusz Rado
n

110. Hughes, T. F.; Friesner, R. A. J. Chem. Theory Comput. 2011, 7(1), 19–32. https://doi.
org/10.1021/ct100359x.
111. Deeth, R. J.; Anastasi, A. E.; Wilcockson, M. J. J. Am. Chem. Soc. 2010, 132(20),
6876–6877. https://doi.org/10.1021/ja1007323.
112. Rado n, M.; Drabik, G. J. Chem. Theory Comput. 2018, 14, 4010–4027. https://doi.
org/10.1021/acs.jctc.8b00200.
113. Jørgensen, C. K. Absorption Spectra and Chemical Bonding in Complexes; Pergamon Press,
1962.
114. Yang, Y.; Ratner, M. A.; Schatz, G. C. J. Phys. Chem. C 2014, 118(50), 29196–29208.
https://doi.org/10.1021/jp5052672.
115. Ghosh, A.; Taylor, P. R. Curr. Op. Chem. Biol. 2003, 7, 113–124. https://doi.org/
10.1016/S1367-5931(02)00023-6.
116. Yang, Y.; Ratner, M. A.; Schatz, G. C. J. Phys. Chem. C 2013, 117(42), 21706–21717.
https://doi.org/10.1021/jp408066h.
117. Uudsemaa, M.; Tamm, T. J. Phys. Chem. A 2003, 107(46), 9997–10003. https://doi.
org/10.1021/jp0362741.
118. Markham, G. D.; Glusker, J. P.; Bock, C. W. J. Phys. Chem. B 2002, 106(19),
5118–5134. https://doi.org/10.1021/jp020078x.
119. Klamt, A.; Sch€urmann, G. J. Chem Soc, Perkin Trans. 1993, 2(5), 799. https://doi.org/
10.1039/p29930000799.
120. Iuchi, S.; Morita, A.; Kato, S. J. Chem. Phys. 2004, 121(17), 8446–8457. https://doi.
org/10.1063/1.1788654.
121. Landry-Hum, J.; Bussière, G.; Daniel, C.; Reber, C. Inorg. Chem. 2001, 40(11),
2595–2601. https://doi.org/10.1021/ic0010860.
122. Armstrong, A. T.; Smith, F.; Elder, E.; McGlynn, S. P. J. Chem. Phys. 1967, 46(11),
4321–4328. https://doi.org/10.1063/1.1840547.
123. Blackburn, F. R.; Snavely, D. L.; Oref, I. Chem. Phys. Lett. 1991, 178(5–6), 538–542.
https://doi.org/10.1016/0009-2614(91)87016-5.
124. Reed, C. A.; Mashiko, T.; Scheidt, W. R.; Spartalian, K.; Lang, G. J. Am. Chem. Soc.
1980, 102(7), 2302–2306. https://doi.org/10.1021/ja00527a028.
125. Hocking, R. K.; Deeth, R. J.; Hambley, T. W. Inorg. Chem. 2007, 46(20), 8238–8244.
https://doi.org/10.1021/ic701166p.
126. Verma, P.; Varga, Z.; Klein, J. E. M. N.; Cramer, C. J.; Que, L.; Truhlar, D. G. Phys.
Chem. Chem. Phys. 2017, 19, 13049–13069. https://doi.org/10.1039/C7CP01263B.
127. Pierloot, K.; Praet, E. V.; Vanquickenborne, L. G. J. Chem. Phys. 1992, 96(6),
4163–4170. https://doi.org/10.1063/1.461872.
128. Aramburu, J. A.; Garcı́a-Fernández, P.; Garcı́a-Lastra, J. M.; Moreno, M. Inorg. Chem.
2017, 56(15), 8944–8953. https://doi.org/10.1021/acs.inorgchem.7b00932.

You might also like