Coordination Chemistry Reviews 257 (2013) 196209

Contents lists available at SciVerse ScienceDirect

Coordination Chemistry Reviews

journal homepage: www.elsevier.com/locate/ccr

Review

Consistent descriptions of metalligand bonds and spin-crossover in inorganic

chemistry
Kasper P. Kepp
Technical University of Denmark, DTU Chemistry, DK 2800 Kongens Lyngby, Denmark

Contents

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
2. General methodological considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
2.1. Electron correlation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
2.2. Complex density functionals as a path toward accuracy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
2.3. Energies and basis sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
2.4. Identifying the congurations that are lowest in energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
3. Systematic effects in electronic structure calculations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
3.1. HF exchange . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
3.2. Self-interaction and non-locality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
3.3. Dispersion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
3.4. Relativistic effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
3.5. Solvent effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
3.6. Zero-point energies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
3.7. Entropy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
4. Insight gained from benchmarking of metalligand bonds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
4.1. Correlation effects in metalligand bonds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
4.2. Bond strengths of M L diatomics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
5. Metalligand bonds and spin crossover . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
5.1. General aspects of spin crossover . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
5.2. HF exchange and spin states . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
6. Larger-system effects not described in this review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206
7. Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206
List of abbreviations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208

a r t i c l e i n f o a b s t r a c t

Article history: Density functional theory (DFT) is today the unchallenged tool for routinely obtaining molecular infor-
Received 30 November 2011 mation on chemical stability, reactivity, and electronic structure across the Periodic Table. The chemical
Received in revised form 12 April 2012 bond is the fundamental unit of molecular structure and reactivity, and thus, large-scale DFT studies of
Accepted 15 April 2012
inorganic systems in catalysis and bioinorganic chemistry rely directly on the ability to balance corre-
Available online 21 April 2012
lation effects in the involved bonds across the s-, p-, and d-blocks. This review concerns recent efforts
to describe such bonds accurately and consistently across the s-, p-, and d-blocks. Physical effects and
Keywords:
ingredients in functionals, their systematic errors, and approaches to deal with them are discussed, in
DFT
Systematic effects
order to identify broadly applicable methods for inorganic chemistry.
Metalligand bonds 2012 Elsevier B.V. All rights reserved.
Electron correlation
Spin
Catalysis
Bioinorganic chemistry

E-mail address: kpj@kemi.dtu.dk

0010-8545/$ see front matter 2012 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.ccr.2012.04.020
 K.P. Kepp / Coordination Chemistry Reviews 257 (2013) 196209
1. Introduction
The last two decades have been characterized by the massive
diffusion of density functional theory (DFT) [13] into all elds of
chemistry, a development that already in 1998 led to Walter Kohn
sharing the Nobel Prize in chemistry with John Pople for their pio-
neering efforts in DFT and general quantum chemistry, respectively
[4].
The current status 14 years later is that DFT is the unchal-
lenged method for describing electronic structure in medium- and
large-size molecular systems [5,6]. Its unique combination of com-
putational speed and accuracy affects most areas of molecular
science from molecular physics and chemistry to biochemistry,
nanochemistry, and materials science. As an illustration, state-of-
the-art DFT is an invaluable complementary and supporting tool for
spectroscopic methods that have greatly contributed to shaping the
eld of bioinorganic chemistry [7,8].
DFT can describe molecular systems with an accuracy that
often competes with experimental data, e.g. for local geometries
of metal centers in proteins [9,10]. However, due to the formula-
tion of DFT, there is no rigorous hierarchy of accuracy of functionals
corresponding to the many-electron space of conguration state
functions in wave function theory (WFT), although DFT is varia-
tional for each functional separately. Thus, no universal functional
is available in algorithm to accurately describe any given electron
density [1,3]. Scientists today rely on a relatively large arsenal of
functionals with various strengths and weaknesses, depending on
molecular systems and electronic properties studied, as reviewed
before [6,1123].
Functionals can be categorized according to their design and
mathematical complexity [14]. In terms of complexity, one can
distinguish (i) local uniform functionals (local spin density approx-
imations [24]) with the energy being a functional of the electron
density [2527]; (ii) functionals that depend also on the gradient
of the density (generalized gradient approximation, GGA) [2830];
(iii) functionals that also depend on more or higher derivatives,
e.g. the Laplacian of the density [31] or the kinetic energy density
(meta functionals) [32]; (iv) functionals that substitute part of the
exchange functional with non-local exchange from a determinant,
typically from a HartreeFock (HF) procedure (hybrid functionals)
[33]; and (v) functionals that substitute also part of the correla-
tion energy for WFT correlation, e.g. from MP2 (double hybrids)
[34]. Further levels of complexity also occur, e.g. range-corrected
hybrids [35,36].
Fig. 1. Papers published 19902011 containing keywords corresponding to various electronic-structure methods; from Thomson Web of Science. Left: total counts of general
methods: DFT, MP2, CCSD, and PM3. Center: DFT methods. Right: DFT methods without B3LYP.
197
One can also distinguish a priori and a posteriori functionals: the
a priori functionals fulll physical bounds on the energy and density
or its derivatives without violating fundamental laws [3739], risk-
ing poor accuracy for specic, perhaps important classes of systems
in exchange for a sound platform for further improvement. Rep-
resentative functionals include PBE [40], TPSS [41], and to some
extent their hybrid counterparts, PBE0 [42] and TPSSh (contain-
ing only one parameter, the amount of HF exchange) [43]. The a
posteriori functionals use several (>1) tted parameters and prag-
matically and efciently reproduce vital observables, guaranteeing
good accuracy for systems and properties within the data sets used
for developing the functionals, but possibly restricted from expand-
ing this accuracy beyond their parameterization range [44]. Some
representative functionals are B3LYP [33,4547] of the hybrid type
and M06 [48], a meta hybrid.
Fig. 1 shows the occurrence of topical words in papers from
Thomsons Web of Science 19902011 as a very rough estimate
of the overall trends in use of various theoretical methods (the
absolute numbers may be affected by search strategy). To the
left, various general methods are compared: while semi-empirical
methods such as PM3 [49,50] are in decline due to DFT outcom-
peting them for widely studied medium-size systems (10100
atoms), MP2, with its N5 size-restriction, maintains a steady use
of 1000/year due to its accuracy and stringent interpretation. DFT
passed MP2 in the nineties and has grown super-linearly since then
to reach 6000/year. Coupled-cluster methods (CCSD keyword)
display modestly increasing use, but a declining market share due
to the massive increase in the use of DFT the last ten years.
The center and right parts of Fig. 1 show trends of a few
individual functionals (they were all searched as method syn-
onym + dft, reducing total hits against dft alone to eliminate
bias from other acronym uses). B3LYP has for many years been
almost synonymous with DFT and remains by far the most applied
functional, although during the last 23 years, new functionals are
quickly conquering market shares. PBE was a rst contender, which
now seems to be overrun by the Minnesota functionals (M0x series)
[51], but the trends are not completely established yet.
The chemistry community currently faces a transition from an
almost uniform standard during the golden age of B3LYP to a new,
refreshingly vibrant, but highly fragmented and confused market
place of density functionals with numerous persuasive vendors. No
new golden standard will arise soon, as judging from Fig. 1, and
this could damage the credibility of theoretical chemists to their
experimental colleagues and may divert productivity away from
198 K.P. Kepp / Coordination Chemistry Reviews 257 (2013) 196209
real problems of science. Thus, it is necessary to consciously eval-
uate new approaches, if not in the hope for a universal functional,
then at least for a state-of-the-art for DFT, just as Pople dened a
model chemistry many years ago [52].
On the path toward new standards, many critical assessments
of DFT have been performed by comparison to experimental prob-
lems [6]. The situation was reviewed in 2007 [14], concluding that
there is no consensus to deduce from these benchmarks. The all-
dominating B3LYP in its various modications [53] performs only
slightly better than average when viewing these benchmarks in
total [14], and worse than average in a recent benchmark even for
main-group systems [54]. Still, the restricted nature of benchmarks,
the number and types of functionals included, the systems stud-
ied, the properties chosen, and the specics of calculation, render
comparisons very hard. How does one normalize such studies to
the same basis sets, correct for zero-point energies (ZPEs), thermal
effects, dispersion, or relativistic effects, identify use of erroneous
electronic congurations, and remedy electronic bias of any spe-
cic choice of benchmark systems, e.g. iron complexes with nearly
half-full d-shells?
Whereas there are excellent reviews of DFT applications in inor-
ganic chemistry [6,1214,16,22,23], this mini-reviews aim is to
provide a detailed discussion of specic systematic effects often
ignored in mainstream DFT modeling. Emphasis is on metalligand
bonds, the core entity of coordination chemistry. As shown, neglect
of systematic effects may cause erroneous conclusions about chem-
ical energies and the adequacy of functionals, and a good practice
of theoretical inorganic chemistry is suggested.
2. General methodological considerations
2.1. Electron correlation
Electron correlation is the correlated movement of electrons
beyond simple mean-eld behavior, and its realistic treatment is
required for computational accuracy, also in inorganic systems,
where electron correlation is typically highly complex.
In terms of spatial correlation of electrons, one can distinguish
three components of the electron correlation: radial, angular, and
leftright correlation [55]. Radial correlation reects correlated elec-
tron movement along the radial coordinate, i.e. toward and away
from the nuclei, and is allowed by using diffuse basis functions
with small exponents and thus, long tails. Angular correlation
is correspondingly achieved via higher-angular momentum basis
functions. Leftright correlation is the third component of the elec-
tron correlation that reduces the probability of having two bonding
electrons in the same part of the bond. This distinction is important
when choosing proper basis functions.
Another distinction of relevance to inorganic chemistry is
between dynamic and static (non-dynamic) correlation, aris-
ing from close-range electron encounters or from quantum-
mechanical interference of energetically close-lying electronic
congurations, respectively. Substantial changes in static corre-
lation often occur upon bond cleavage and in transition metal
systems with close-lying d- or f-orbitals [56].
Consider a metalligand bond that is broken homolytically, e.g.
the organometallic Co C bond of cobalamins: leftright correlation
may weaken the bond as it destabilizes the bond by increasing the
probability of having electrons on opposite sides of the molecule,
e.g. by moving electron density from a metalligand -MO to a *-
MO. Also, static correlation affects the metalligand bond cleavage
as it is often larger in the dissociated states, or on the path toward
dissociation, where electronic congurations have more similar
energies. Furthermore, close-lying electronic congurations with
variable d-orbital occupations will display more static correlation
energy in some cases, e.g. in high-spin complexes with small ligand
eld splitting, compared to systems where electrons are paired in
MOs subject to strong splitting.
A third distinction is that of Coulomb and Fermi correlations,
which are also critical to inorganic chemistry, as they emphasize
the role of electron spin: Coulomb correlation is the part of the total
correlation that is not due to electron spin, whereas Fermi correla-
tion arises from the Pauli repulsion between electrons, which are
S = fermions. In HF theory, although often said to be uncorrelated
(in Lwdins sense [57]), some Fermi correlation is described by the
Slater determinant (Eq. (1)), with N electrons in N orbitals [58]:
 
 1 (1) 1 (2) 1 (N) 
 
1  2 (1) 2 (2) 2 (N) 

 = (1)
N!  
.. .. ..
. . . 
 (N) 
N (1) N (2) N
The energy lowering obtained in this way is called exchange,
because the determinant exchanges (permutes) electrons among
all orbitals to ensure an anti-symmetric wave function as required
by the Pauli Principle [58]. Changing the number of unpaired elec-
trons in a process, as often observed in inorganic chemistry, e.g.
high-spin or odd-electron systems, is particularly challenging to
describe theoretically, as discussed below.
2.2. Complex density functionals as a path toward accuracy
One signicant difference between WFT and the KohnSham
(KS) procedure [2] used in most DFT is that WFT keeps the
Hamiltonian H simple while expanding the one-electron and many-
electron spaces, as seen in Eq. (2), whereas DFT keeps to the
one-electron space while expanding the complexity of the Hamil-
tonian (i.e. functional, fxc which can depend in various ways on the
electron density), and integrating only over space d3 r, as seen in Eq.
(3) [59].

N
 = ai i , 
E =  |H| (2)
i

Exc [ ,  ] = d3 rfxc ( ,  ,  ,  , . . .) (3)
Whereas for WFT, the path toward universally accurate energies
is well-dened but slowly converging in many-electron space, in
DFT, the path is less straight-forward but tends to converge quickly
due to ad hoc inclusion of the most relevant effects in the func-
tional. The difference arises because physical interactions between
electrons (i.e. the electronelectron cusp) affect much more the
energy than the precise electronic distribution beyond some accu-
racy. Thus, in DFT, the correlation energy converges much faster
with the one-electron basis set than in WFT, because more of the
important correlation is obtained directly from the functional: the
He-atom is a good example, where s-functions are enough to obtain
the correct correlation energy with the exact density functional
[60].
It is thus more apparent why current DFT research focuses on
expanding the complexity of functionals: they progress from the
local density functionals that only depend on the electron densi-
ties of spin-up ( ) and spin-down ( ) electrons (Eq. (4)), toward
GGA functionals that depend on the gradients (  ,  ) (Eq. (5)),
or meta functionals that depend on higher terms, e.g. the kinetic
energy densities  and  (Eq. (6)) [61].

LOCAL
Exc [ ,  ] = d3 r(r)fxc
LOCAL
( ,  ) (4)
 K.P. Kepp / Coordination Chemistry Reviews 257 (2013) 196209

GGA
Exc [ ,  ] = d3 r(r)fxc
GGA
( ,  ,  ,  )
 tronic conguration of lowest energy mimicking the ground state
meta
Exc [ ,  ] = d3 r(r)fxc
meta
( ,  ,  ,  ,  ,  , . . .)
As discussed in this review, while expanding the complexity of
these functionals further, one should assess the impact of other
effects usually added as corrections to the calculations, since they
often affect accuracy more than incorporation and parameteriza-
tion of new terms in functionals. Experimental structural, energetic,
and spectroscopic data are critical for validating the accuracy of
functionals for given tasks, and should be practiced whenever
experimental data of any kind are available; if such data are unavail-
able, correlated ab initio methods may provide insight into the
accuracy of the energies and properties obtained by a given func-
tional.
2.3. Energies and basis sets
As in WFT, molecular geometries are less sensitive to the method
and can be modeled quite accurately by most GGA and hybrid func-
tionals in wide use, with mean absolute errors (MAE) of 0.02 A
for bonds and 12 for angles [1315,6265]. However, geometries
of systems with very weak bonds or signicant dispersion interac-
tions [66] may be subject to larger functional-dependent errors and
require special attention. Classical bioinorganic examples are the
soft Co Nax bond in cobalamin enzymes [67] or the soft Cu S bonds
in blue copper proteins [68], whereas strong -donating ligands
usually exhibit smaller geometric errors but equally large effects
on energy [67].
When modeling accurately energies of bonds and transition
metal systems, to allow for the necessary electronic freedom, triple-
zeta basis sets with polarization and diffuse functions are required
[52,69]. While moving to quadruple-zeta is important in post-HF ab
initio WFT, it has little effect on DFT-computed bond dissociation
enthalpies (BDE) of most metalligand bonds [54]. However, dif-
fuse functions can be important due to the differential correlation
effects in the compact (bound) state not present in the dissociated
state [62], and use of unbalanced basis sets is error-prone as elec-
trons favor to reside in the part of space covered mostly by basis
functions.
Polarization functions on hydrogens are required if the electron
density changes around hydrogen atoms, as e.g. in hydrogen-bond-
breaking or H-abstraction reactions, and more diffuse functions are
required if electron-rich or excited electronic congurations are
involved. Electron afnities, excitation energies, and some softly
bound systems should routinely be studied with diffuse functions
that enhance the radial correlation of the electrons to allow unbi-
ased reproduction of both states of the real process. When this is
done and other effects such as solvation are taken into account, rel-
ative reduction potentials of electronically similar systems can be
obtained with almost chemical accuracy, as evidenced by studies
of rst-coordination sphere single-mutants of ironsulfur proteins
[70].
Thus, a state-of-the-art basis set is necessary but not suf-
cient to provide electrons with the freedom to distribute optimally
in space in the involved states of the process: the second criti-
cal requirement is how the functional describes short-, medium-,
and long-range Fermi- and Coulomb correlation and thus effec-
tuates this distribution, and the third critical requirement is
proper account of corrections for other physical effects not directly
included in the functional (entropy, ZPE, relativistic effects, disper-
sion, etc.), which is the main focus of this review.
199
2.4. Identifying the congurations that are lowest in energy
(5)
However, even before that, one must correctly identify the elec-
(6) density. Most standard DFT approaches rest on the KS approach
where the KS-determinant is an effective approximation to the
true (mixed) state. Therefore, KS molecular orbitals (MOs) are not
strictly representative of congurations as in WFT, but more of
approximate states. Furthermore, as in UHF, spin contamination
will be observed in most non-closed-shell systems, with S not
being a good quantum number in unrestricted calculations where
the Fermi correlation polarizes the - and -spaces differently
[71]. Both hybrids and non-hybrids can display spin contamina-
tion, which will tend to be larger in hybrids due to the HF exchange
that enhances spin polarization.
Despite this, any conguration of the S, MS quantum numbers
is approximately accessible by local convergence or spin-down-
coupling: one may obtain both closed-shell singlets (S, MS = 0, 0)
that are pure, i.e. similar to the true single-congurational state,
and impure open-shell singlets (S, MS 1, 0) that are only one of
the several congurations of the true multi-congurational state.
The latter is usually obtained from the input MOs of the pure triplet
conguration (S, MS = 1, 1) by changing the spin of the highest singly
occupied molecular -electron in the calculation input. To honor
the KS effective determinant principle, it is absolutely necessary
to obtain the correct (i.e. lowest in energy) of these congurations.
It is quite common to see excited congurations resembling lower
spin polarization and higher energy: a typical example is O2 binding
to heme, where the closed-shell conguration is excited, whereas
the correct conguration is spin-polarized reecting some Fe(III)
character [72].
In clusters with coupling between several metal sites, the
maximally spin-polarized congurations are often referred to as
broken symmetry congurations that best resemble the true
anti-ferromagnetic state [73]. A conguration with spin densities
close to 1 and 1 in a coupled iron(III) cluster is thus a single-
conguration-approximation to an excited multi-congurational
state (namely, an impure S, MS = 1, 0 state). If the congurations
of lowest energies are not obtained by careful procedures both in
benchmarks and in routine studies, the states and processes inves-
tigated will not be accurately described. This is true particularly in
systems with large static correlation such as transition metal sys-
tems and potential energy surfaces. Careful, manual sampling of
the electronic congurations is necessary.
The problem of static correlation also haunts the use of coupled-
cluster methods or indeed, any post-HF ab initio method based
on a single HF reference determinant [74]: coupled-cluster with-
out explicit triples excitation operator (i.e. CCSD(T) or less) from
such a reference is unreliable in systems with static correlation
[75,76], e.g. transition metal systems and bond breaking reactions
[77,78]. Time-consuming CCSDT can recover signicantly more
of the static correlation, as can active-space CCSDT or hybrids
thereof [79]. Sometimes, using a DFT or CASSCF/CASPT2 [80] refer-
ence may drastically improve otherwise erroneous CCSD(T) results
[81]. CASSCF/CASPT2 wave functions are not biased toward a
single HF reference but are fully optimized within their active
space and are thus important tools for statically correlated tran-
sition metal systems and for properties such as excitation energies
[82]. The known one-determinant bias of single-reference post-
HF methods has not prevented some researchers from blindly
using UHF- or even ROHF-based CCSD(T) energies with strong
bias toward some HF conguration for systems with substan-
tial static correlation. The failure of the single-determinant HF
reference for describing pseudo-degenerate states is also the rea-
son why perturbation theory (e.g. MPn) is not valid in these
situations.
200 K.P. Kepp / Coordination Chemistry Reviews 257 (2013) 196209

Still, ab initio methods such as CASPT2 and coupled-cluster 3.1. HF exchange

methods can offer critically important insight into statically cor-
related inorganic systems and constitute the only validation of
density functionals when experimental data are poor or absent [83].
Much work has been done using CASPT2 in inorganic chemistry
[84], including problems with substantial static correlation such as
charge-transfer states [85,86] and biologically relevant tetrapyrrole
systems [8791]. Also, substantial work has been done with e.g.
equation-of-motion coupled-cluster methods, e.g. Fe-, Co-, and Ni-
carbenes with nearly degenerate ground states [92,93], considering
various reference wave functions to assess one-conguration bias.
Both types of studies may accurately account for the lowest-lying
electronic states in statically correlated inorganic systems and thus
provide important data in the future benchmarking of inorganic
DFT energies.
3. Systematic effects in electronic structure calculations
In addition to basis set, choice of functional, model system, and
efcient conguration optimization, several other aspects need to
be addressed. Table 1 outlines these additional effects that should
ideally be addressed in DFT model chemistry. As will be discussed,
many of these effects are not always appreciated although they
are often systematic and predictable and can be generalized to
favor either the state with the more compact or the looser electron
density.
On one hand, there are real, physical effects often with system-
atic behavior such as dispersion interactions, relativistic effects,
entropy, ZPE, and solvent effects, which must be included as a
correction to the electronic energy. Of these, dispersion and rela-
tivistic effects are electronic in nature, whereas entropy and ZPE are
mainly related to the vibrational states of the system, although the
electronic congurations also affect them via the electronic state
function and the vibrational coupling. Thus, their effect on observ-
ables such as bond strengths is indirect (and their effects marked by
in Table 1), as they are added after convergence of the electronic
structure, but are generally systematic as the electronic effects. Sol-
vent effects deal with the systems surroundings. On the other hand,
there are some features of the functionals themselves, notably HF
exchange and self interaction, which are also often of a systematic
nature.
It was early thought that HF exchange from the Slater determi-
nant (Eq. (1)) might improve DFT [11], and the hybrid functional
B3LYP showed that a fraction of HF exchange can improve descrip-
tions of many properties [33]. Due to the permutation of all
indistinguishable (i.e. same-spin) electrons caused by the deter-
minant, HF exchange has an important ability to mimic the real
long-range (non-local) component of electron correlation that is
very hard to mimic by other simple means. Some hybrid func-
tionals, notably B3LYP with 20% HF exchange, perform well for
structures and many types of chemical reactions involving s- and
p-block elements [1,5]. This success has made B3LYP the most used
functional in chemistry, including inorganic chemistry, as men-
tioned Section 1 [14,21,22].
HF provides a decent description of some atomic and ionic con-
gurations where errors tend to cancel, although the absence of
dynamic correlation renders it very inaccurate for electron-rich
systems, as observed e.g. in the increasing errors in 3dn1 4s1 3dn
excitation energies going from Sc+ to Cu+ [11]. HF dramatically
underbinds in many cases [95,96]. In the d- and f-blocks, electronic
congurations are closer in energy than the s- and p-blocks and
thus enhance static correlation, which is also absent in HF [97].
The optimal amount of HF exchange depends on the electron cou-
pling via the adiabatic connection formula [98]: near-degenerate
ground states are typically more accurately described by less than
20% HF exchange [42], because a single HF Slater determinant is
then biased. Non-hybrid exchange functionals are not biased by the
HF exchange of a single, in such cases misleading, conguration,
and thus tend to better describe systems with substantial static
correlation often found in the d-block [62].
When the electronic structure changes qualitatively, Fermi cor-
relation constitutes the majority of the differential correlation
energy, as witnessed from the impact of HF exchange on BDEs [99],
and 20% HF exchange is excessive [54,99]. As observed already by
Reiher [100] and Paulsen et al. [101], relative multiplet energies
dependent almost linearly on the amount of HF exchange, which
thus favors high-spin (HS) over low-spin (LS) [102,103]. This also
leads to weaker bonds with lower BDEs of 30 kJ/mol per 10%
of HF exchange for metalligand bonds of diatomics in vacuum,
where dispersion, solvent effects, and entropy errors are negligible
[62,99].

Table 1
Some systematic effects in density functional theory.

Systematic effect Typical effect on  Typical effect on bonds Examples Representative references

Ingredients in functionals
HF exchange Looser Longer/weaker HF and hybrid functionals [11,62,100,104,127,160]
underbind and favor HS, up to
30 kJ/mol per 10% HF exchange
Self-interaction error (SIE) Looser Weaker in Articial low dissociation limits of [108,113115]
odd-electron systems X2 + ; favors delocalization of
SOMOs, errors can be >50 kJ/mol
Physical effects
Dispersion More compact Shorter/stronger Dispersion increases BDE, favors [126,128]
LS, often >10 kJ/mol
Relativistic effects s, p-contraction, valence Stronger, compact Favor stronger -bonds; but favor [62,136,139]
expansion -bonds; weaker atomic states in middle d-block
bonds in middle d- and 40 kJ/mol in some cases
p-blocks
Solvent effects Depends on polarity and Signicant in charged, E.g. 0.020.03 A shorter M L [94,144,145,147,151]
charge of molecule polar, weakly bonded, bonds, favors LS. Polar reactions
and unsaturated are highly sensitive to solvent
systems (often 100 kJ/mol or more)
Zero-point energies (ZPEs) Indirectly loosening Longer/weaker ZPE favors HS, typically by [62,99,160]
515 kJ/mol
Entropy Indirectly loosening Longer/weaker Entropy lowers BDE and favors HS, [126,163]
typically by 1030 kJ/mol per
metal site per bond/metal
 K.P. Kepp / Coordination Chemistry Reviews 257 (2013) 196209
Down-toning the HF exchange component of B3LYP improves
accuracy in d-block chemistry, rst done via the B3LYP* functional
(15% HF exchange) [104,105] and later via a BP86 hybrid with
10% HF exchange giving better agreement with spectroscopic data
[106]. Newer hybrid functionals display various HF exchange com-
ponents in the range from 10% (TPSSh) and up to approximately
50% (B2PLYP, M06-2X). The impact of HF exchange on results for
late transition metal systems has also been discussed in detail in a
review that also encouraged the use of meta GGA functionals [107].
3.2. Self-interaction and non-locality
The HF exchange, so far necessary although problematic in DFT,
is accompanied by other effects that are often systematic, notably
the self-interaction energy (SIE) of electrons [3,108,109]. Whereas
this energy is strictly zero in WFT, where K and J integrals cancel
completely, in DFT, due to the above-mentioned separate treatment
of exchange and correlation functionals, it is not zero except by
special design. Constraining the SIE to zero damages the accuracy
of current functionals dramatically [110,111], showing that the SIE
is coupled to other effects, as will be discussed below.
SIE increases with the distance between orbital centers and is
thus important in processes that separate single electrons over
delocalized MOs, e.g. dissociation of odd-electron systems [112],
near-degenerate ground states as seen in the d-block, some open-
shell transition states, or radicals in delocalized -systems.
SIE is often illustrated by dissociation of X2 + : at shorter dis-
tances, the - and *-MOs are separated and the electron resides
in the -MO. SIE corresponds to a localized electron due to the
signicant energy gap to *, thus always carrying its SIE with it.
However, as X2 + dissociates, the unpaired electron is delocalized by
static correlation. In WFT, this is reected by multi-congurational
character, whereas in DFT, it is reected by polarization of the
singly occupied molecular orbital (SOMO) in response to the func-
tionals attempt to mimic the static correlation. Upon dissociation,
the splitting is reduced and the ground state becomes a quantum-
superposition of two ionic states with energy E(X+ ) + E(X). The exact
energy is a linear function of any fractional electronic occupa-
tion of the MOs. This situation is correctly described by HF/WFT,
which is SIE-free, due to 100% HF exchange. However in standard
DFT, this cancelation is incomplete, and the energy curves down-
wards between integer electron systems, due to the lowering of
SIE obtained from partial electron delocalization. The problem is
always the differential SIE of the process, which in this case equals
the difference between the symmetric (half-integer electron) and
the ionic (full electron) dissociation limits and can be massive
[113,114].
Normally, functionals will enhance resonance and delocaliza-
tion of electrons between close-lying congurations as a means
to lower the SIE. Therefore, SIE is sometimes said to mimic static
correlation [115,116]. Thus, SIE is, as spin contamination, an arti-
cial by-product of the functionals attempt to optimally correlate
the electrons. In coordination complexes, the error will be smaller
because the MOs made from the d-orbitals are usually localized, but
cases have been reported where differential SIE signicantly affects
exchange coupling constants [117] and reaction energies [118,119]
of transition metal systems.
3.3. Dispersion
Dispersion, a second-order perturbation of instantaneously
induced dipoles in closely interacting electron densities, is gen-
erally not accounted for in standard functionals, even meta
functionals, although it could possibly be accounted for in higher-
order meta functionals as local density gradient products. The
absence of dispersion can be remedied e.g. by corrections such as
201
DFT-D [120,121], dDsC [122] or the non-empirical corrections by
Tkatchenko and Schefer [123]. These methods develop rapidly: as
an example, the newest of the DFT-D corrections, DFT-D3 [124],
differs substantially from the previous DFT-D1 and -D2 methods,
in particular by using new physical cutoff radii for the dispersion
interaction and specic atom-pair coefcients, rendering it less
empirical [124,125]. DFT-D2 improved DFT-D1, e.g. by providing
more realistic geometric-mean combination rules for estimat-
ing the dispersion interaction from individual atomic coefcients
[121].
Dispersion in a purely attractive form favors closer interaction
between electrons in different MOs and produces more compact
electron densities, stronger binding, and shorter bonds. LS states
are favored by shorter metalligand bonds (see Section 5), and the
stabilization of LS by such dispersion corrections can be large, as
realized in recent work on axially substituted iron(III)porphines
[126]. Dispersion also increases BDEs of bulky molecules such as the
Co C of cobalamins, and the good performance of GGA functionals
for this bond [127] may be due to neglect of dispersion [128]. How-
ever, the D3 correction is substantially smaller than D2 in these
cases, and the models used for AdoCbl neglected many strongly
interacting side chains of the corrin ring, suggesting a cancelation
of two errors in the study, although the general conclusion remains
valid. Dispersion is also necessary to describe relative energies of
isomers and conformations with different steric crowding, as seen
in e.g. alkanes [129].
A natural step forward is to include dispersion corrections
directly in benchmark studies once these are sufciently trusted
[54,130], which, as other systematic effects, will change the overall
conclusions from many previous benchmarks [14]. Overall perfor-
mance is markedly better than for old benchmarks of DFT without
dispersion [131], conrming the utility of such methods. They
should therefore be routinely applied as good practice, to test its
impact on computed results.
3.4. Relativistic effects
Relativistic effects generally stabilize and contract the most rel-
ativistic core s- and p-electrons, and in response all outer s-shells
due to their overlap and reduced repulsion [132]. This often leads to
strengthening of late d-block metalligand -bonds, e.g. Au L vs.
Ag L [133]. Importantly, these systematic relativistic contractions
are often >0.02 A already for Cu L (Zn L are less contracted) [132]
and are important already for the electronic properties of transition
metal systems from the rst row of the d-block [81,134].
Because of the resulting, enhanced shielding of the nuclei, the
diffuse, higher angular momentum MOs (d- and f-type orbitals
and anti-bonding MOs in the valence shell) are typically expanded
and destabilized [135], leading to polarization of electrons and
weakened bonds, especially of higher bond order. Furthermore,
spinorbit coupling can contribute substantially to weakening
of bonds by favoring the dissociated states that often exhibit
higher spin and angular quantum numbers [138]. The effects are
comparable in magnitude and often roughly proportional to Z2
[132].
Sometimes, effective core potentials can account for most of
the relativistic effects [81]. They confer a reasonably accurate con-
tracted atomic core and reduced effective nuclear charge, thus
mimicking the above-mentioned effects. For isodesmic reactions
where the quantum numbers are qualitatively unchanged, the
effective core potential often sufces. However, in processes where
valence electronic structures change qualitatively, differential rel-
ativistic effects are important, both due to corevalence correlation
effects not imitated by a rigid core and due to intrinsic effects of the
valence shell [81]. To remedy this, one can compute the scalar rela-
tivistic effects directly on all explicit electrons and include them in
202 K.P. Kepp / Coordination Chemistry Reviews 257 (2013) 196209
Fig. 2. Scalar-relativistic effects in 80 rst-row d-block M L diatomic bond enthalpies (gure based on data from Ref. [62]). Notice the behavior of Cr and Cu due to half-
and full d-shell congurations 3d5 4s1 and 3d10 4s1 sacricing the 4s2 inert pair stabilization for lower electron repulsion.
benchmarks of rst-row [62,99] and especially second- and third-
row elements [136].
In the rst-row of the d-block, scalar-relativistic effects aris-
ing especially from changes in electronic structure (i.e. quantum
numbers) contribute signicantly to reaction energies. As seen in
Fig. 2, on average, scalar-relativistic effects weaken M L bonds
by 7 kJ/mol with large variations [62], and the effects are largest
in the middle of the d-block, where more unpaired electrons, i.e.
highest spin, are generated in the atoms. Another cause of the rel-
ativistic destabilization of the bonds is the 4s2 inert pair of the
atomic states. This pair is absent in the half- and full-d-shell Cr
and Cu (3d5 4s1 and 3d10 4s1 ) where normal shell-wide lowering of
electron repulsion dominates over the relativistic 4s stabilization,
causing overall M L -bond stabilization as commonly seen for
heavy metal halides [137]. These effects (e.g. +13 kJ/mol for Cu H)
are comparable to full relativistic considerations (+8 kJ/mol) [138].
However, in all diatomics except Cu L and Cr L, scalar relativistic
effects favor the dissociated congurations.
These non-negligible relativistic effects on M L bonds in the
rst row of the d-block are consistent with second-row p-block
systems, e.g. 20 kJ/mol scalar-relativistic destabilization already in
SF6 , increasing with higher substitution (corresponding to coor-
dinative saturation in the d-block) [139]. Both scalar-relativistic
and spinorbit effects, which favor higher spin and angular
momentum in dissociated states, systematically favor bond weak-
ening. Bauschlicher showed this in the case of scalar-relativistic
weakening of Ga Cln bonds, increasing quickly with n and reach-
ing 25 kJ/mol of bond weakening with n = 3 [140]. Martin and
co-workers identied systematic bond weakening from scalar-
relativistic effects in the order of 15 kJ/mol for the rst- and
second-row of the p-block, but increasing rapidly with size [141].
A discussion of the reduction in p-block bond energies by
relativistic effects has been presented by Wang et al. [142]: interest-
ingly, this destabilization of bonds is not simply correlated to bond
elongation as in non-relativistic situations. Thus, bond contraction
will often be observed, as in many -bonds described in Pyykkos
review [132], while destabilization of the bond energy occurs [142].
This is also mentioned by Pyykko, giving the example of AuCl [132].
The consequence of scalar-relativistic bond weakening is that
studies that neglect scalar relativistic corrections in non-isodesmic
reactions (enthalpies of formation, bond breaking, spin crossover,
etc.) of third- and fourth period can be biased by typically
525 kJ/mol toward methods that compensate this, e.g. hybrids
with large percentages (>15%) of HF exchange.
3.5. Solvent effects
Solvent effects play a key role in two forms: via uniform
electrostatic screening in the condensed phase, obtainable from
macroscopic continuum models, and via heterogeneous screening
and explicit solvation, obtainable from microscopic solvent struc-
ture.
The effect of electrostatic screening is largest in molecules with
very heterogeneous charge distributions, non-zero total charges,
soft bonds or loose conformations exposed to solvent, and/or
explicit hydrogen bonding. In such cases, geometries can also
be affected, and hydrogen-bonded systems have a dispersion-
component that also needs to be addressed [143]. For ironsulfur
[144] and metallothionein [145] clusters, which are 2 or even 3
charged, optimizing geometries in a dielectric continuum model
such as Cosmo [146] leads to a substantial (0.020.03 A) shortening
of the M S bonds by screening the repulsion between negatively
charged sulfur atoms. This profoundly inuences the electronic
structure of these biological metal clusters. The rst 5 units of
from homogeneous screening captures most of the effect because
of the inverse relation of on the Coulomb interaction (Eq. (7)),
rendering discussions of = 10 or 20 more academic [145].
Q1 Q2
E= (7)
4r12
Heterogeneous solvent effects and the accuracy of cluster mod-
els have been discussed in detail by Siegbahn and Himo [147,148]
and by Ryde and co-workers [149]. As has been observed by
performing full DFT optimization of small proteins [150], the
heterogeneous and homogeneous effects on geometry can be
quite similar, i.e. a large part of explicit solvation is obtainable
from the homogeneous component in dielectric continuum mod-
els. However, this is not always true: if the system is subject
to specic, non-uniform interactions from hydrogen-bonding or
metal-coordinating solvents such as water, acetonitrile, or tetrahy-
drofuran, explicit solvation can affect structure and energies.
Energies of polar reactions often exhibit solvation effects of the
order of 100 kJ/mol, since the hydration energy of monovalent
ions is typically several hundred kJ/mol [151]. This is e.g. seen
 K.P. Kepp / Coordination Chemistry Reviews 257 (2013) 196209
Table 2
Effects of homogeneous solvation, Cosmo radii of Fe, dispersion, and explicit THF solvation on the stability of high-spin vs. intermediate-spin in iron(II)porphine (in kJ/mol).
Energy gaps (kJ/mol) Vacuum
EHSIS (BP86 no dispersion) 70.3
EHSIS (BP86-D3) 70.8
EHSIS (B3LYP-D3) 23.0
in polar bond cleavage reactions [152], in neutral-molecule-ion
substitutions [153], and in SN 2 reactions with charged reactants
or products such as in the cobalt-containing methionine synthase
[154], to mention a few examples.
As an illustrative example, coordinatively unsaturated planar
porphyrins are affected by explicit solvation not captured by dielec-
tric models: from Table 2, THF coordination to iron(II)porphine
moves HS 2535 kJ/mol closer to intermediate spin (IS), although
the HS state is still an excited state with this method. The effect
originates from oxygen lone-pairs in THF that destabilize IS, which
has one more electron in solvent-exposed occupied d-orbitals with
a z-component (either dz 2 or dxz /dyz ) compared to HS.
In this case, dispersion does not change the relative energy
because the porphine is rigid, but it would if axial ligands were
present, via the differential ring-ligand dispersion interactions in
longer-bond HS and shorter-bond LS [126]. Notice also the sig-
nicant effect of changing the Cosmo radius of iron, allowing the
solvation probe to perturb the two states differently. The experi-
mental ground state is a triplet state [155,156], but the energy to
the quintet state is not known, although CASPT2 computes them
to be very close in energy [157]. A microscopic solvation effects of
25 kJ/mol could change conclusions, as is also seen experimentally
when changing solvent from e.g. CH2 Cl2 to THF [158], and coordi-
natively unsaturated systems in condensed phase thus need critical
validation.
3.6. Zero-point energies
Zero-point energy, the quantum-mechanical vibrational energy
at the bottom of the potential energy curve, does not affect the
computed electron density or bond directly, because it is added as
a correction after the convergence of the electronic structure, as
is entropy. Still, ZPE can affect calculated heats of formation and
bond strengths signicantly to the same nal effect as a loosen-
ing of the density or bond. Thus, ZPE and entropy are marked with
in Table 1, as their effect on observables is after nuclear correc-
tions. Thus, their neglect can cause error cancelation with effects
that directly inuence the electron density. Reasonably, functional-
dependent scale factors should be used in such studies [159].
ZPE tends to favor the HS states that generally have weaker M L
bonds and therefore smaller ZPEs than the LS states. Neglect of
ZPE in spin crossover, even when entropies are absent, is therefore
a source of systematic errors, typically 515 kJ/mol bias against
the HS state [105,160]. With a GGA functional that itself is biased
toward compact electron densities (LS, strong bonds), neglect of
ZPE will make errors even larger. For hybrid functionals such as
B3LYP, which intrinsically favor HS, neglecting ZPE will cause two
systematic errors to partly cancel, even more so if entropy is also
neglected (vide infra).
When computing reaction energies, dispersion effects that favor
the compact electronic structure (LS or bound state) work oppo-
site to ZPE corrections and HF exchange, and neglect of dispersion
in larger complexes, in particular with bulky groups, will lead to
an articial bias toward looser states, something that has been
common until the advent of dispersion-corrected DFT methods
[120,121].
In the calculation of M L BDEs, ZPE is even more important as
the dissociated state does not have ZPE from the dissociated bond
203
Cosmo, = 25, rFe = 2.0 A Cosmo, = 25, rFe = 1.5 A THF2
76.3 66.9 41.9
76.6 67.4 41.3
32.7 23.2
at all. For example, in cobalamin coenzymes, the effect of ZPE on
the Co C bond was 20 kJ/mol, however canceling with several
other effects, leaving the BP86 result close to the real Co C BDE
[127]. This is probably due to cancelation of errors from absence
of both dispersion and HF exchange with SIE, as suggested from
Table 1 and observed partly by adding dispersion to B3LYP [128].
Most researchers include ZPE as state-of-the-art, although it is still
sometimes ignored, causing the systematic errors mentioned.
3.7. Entropy
Observed ground states in condensed phase at thermal equi-
librium are the result of free energies, not electronic energies or
enthalpies. Therefore, any conclusions regarding the chemical equi-
librium or the thermally averaged ground state need to take
entropy into account [126].
In chemical reactions where bonds break, which include the
majority of interesting catalytic processes, entropy is highly impor-
tant due to the additional translational and rotational degrees of
freedom. For example, when identifying partially ligand-bound
intermediates in (bioinorganic) catalysis [161], entropy is critical,
as evidenced e.g. in the identication of catalytic intermediates of
di-iron enzymes with differing O2 -binding geometries [162]. The
differential entropy increases more or less linearly with the num-
ber of metal centers in clusters, up to 60 kJ/mol for isomers of
di-iron enzymes [162].
Neglect of entropy is particularly grave because
entropyenthalpy compensation, as in many other aspects of
physics and chemistry, is a major force in molecular interac-
tions: a strong bond favored by enthalpy will usually display
unfavorable entropy because of the smaller contribution of more
widely separated vibrational energy levels to the state func-
tion, whereas the vibrational levels are accessible in the weaker
bond. This entropyenthalpy compensation is signicant even
in reactions where bonds do not break, as is well illustrated by
spin-crossover complexes (vide infra) for which the spin transition
is an equilibrium with temperature usually favoring HS [163].
Under typical conditions, entropy contributes 1030 kJ/mol to
the stability of HS in mononuclear coordination complexes [163].
The reason for this entropy effect in spin-crossover is that the
occupation of d-orbitals pointing toward the ligands (the eg set
in Oh symmetry) weakens metalligand bonds in the HS state and
renders its entropy larger. Many studies neglect this large compen-
sation effect when hypothesizing on the nature of ground states.
Neglecting entropy will typically, because its neglect indirectly
favors electronic structure methods giving compensatory, looser
electron densities, tend to cancel with errors in hybrids having too
much HF exchange [126].
Calculations of entropy contributions are associated with sub-
stantial uncertainties, in particular when low-energy modes are
involved, which are affected by the environment of the molecule
[100]. Still, entropies within 10 kJ/mol of experimental data are
attainable for mononuclear complexes using standard harmonic
vibrations and subsequent thermodynamic analysis [105]. Thus,
the entropy effect is signicant and computable, although the
uncertainty should be acknowledged, and its neglect will lead
to wrong conclusions of the performance of electronic structure
204 K.P. Kepp / Coordination Chemistry Reviews 257 (2013) 196209
methods if the property used for benchmarking is related to the
free energy, not energy or enthalpy [164].
4. Insight gained from benchmarking of metalligand
bonds
4.1. Correlation effects in metalligand bonds
The M L diatomics (M = metal of the d-block, L = s-block
or p-block atom) have been the focus of recent benchmarks
[62,99,119,136] for several reasons, other than being the funda-
mental entity of many inorganic chemical reactions: (i) there are
a large number of experimental bond enthalpies available; (ii)
these data are from vacuum, eliminating random errors associ-
ated with use of solvent models; (iii) entropies are not required
to be estimated, eliminating a further cause of error, including a
large random error component of at least 5 kJ/mol; (iv) disper-
sion effects in diatomics in vacuum are of the order of 1 kJ/mol
even with larger metals, reducing this source of error; and (v) the
large number of data for the entire d-block eliminates commonly
observed bias toward certain electronic systems, as required in
universal benchmarking. None of the other commonly used bench-
mark data sets fulll all these criteria, and therefore, the M L
diatomic bonds are an important part of the benchmark tool box in
this authors opinion. Elimination of bias to electronic structures is
incidentally also a motivation of the articial molecule data sets
by Grimme and co-workers [188].
An all-purpose functional claimed to work well across chemistry
should balance long- and short-range Fermi- and Coulomb corre-
lation for both loose (atomic) and compact (molecular) states. In
Section 3, it was shown how various effects that are often partially
ignored affect energies. In the quest for a state-of-the-art of inor-
ganic DFT modeling, M L BDEs are critically challenging because
they depend on two very different limits of interaction: bond break-
ing leads to redistribution of electrons in new, more atom-centered
MOs (AOs) under substantial changes in correlation energy. Mov-
ing from atoms to diatomics to saturated systems, orbitals change
from being more atomic to become more molecular, residing in
pairs in the bonding regions. Functionals may apparently perform
well for coordinatively saturated systems only because of cancela-
tion of errors, e.g. dispersion, thermal or relativistic effects that are
more prominent in larger systems.
For the rst row of the d-block, both GGA functionals and
hybrids are known to give accurate bond lengths, typically
with uncertainties of 0.02 A [62,63]. For larger, coordinatively
Fig. 3. Effect of functional on metalligand bond polarity: differences in Mulliken dipole moment (a.u.) of diatomics computed with B3LYP and BP86 (positive means that
the B3LYP dipole moment is larger). The main reason for the systematic difference is HF exchange (see text).
Figure produced using data from Ref. [62].
saturated complexes, hybrids such as PBE0, B3LYP, or TPSSh give
slightly more accurate M L bond lengths than GGA functionals
across the three rows of the d-block, although these differences
are not critical to the reaction energies [6264,165].
As other explicit correlation effects, HF exchange favors delo-
calization of electrons into MOs of higher angular, spin, and radial
quantum numbers, including anti-bonding MOs that lowers bond
orders. This polarization is witnessed in M L dipole moments,
which are substantially (typically 0.5 a.u.) and systematically
larger when computed with B3LYP than with BP86 without HF
exchange (see Fig. 3) [62]. The only exceptions (TiC, TiN and ZnO)
could be due to differences in BP86 and B3LYP electronic struc-
tures. The increased dipole moments caused by HF exchange is
another manifestation of the polarization of the electron density
also responsible for weakening of bonds. Given the importance
of HF exchange in most properties, testing the variation of pos-
sible outcomes using two functionals such as B3LYP and BP86 (or
another non-hybrid GGA functional) is a reasonable way to rou-
tinely perform a reality check of results, which should in this
authors opinion always be practiced.
Electronic polarization also increases spin contamination in
unrestricted computations (except in pure congurations of
MS = S) and is sometimes called mixing-in of higher multiplet
determinants although it is the composition of the MOs that
change. It illustrates the KS effective conguration principle in
action, and spin contamination is thus a natural by-product of this
optimization under the inuence of the functional [166].
4.2. Bond strengths of M L diatomics
For the d-block, BDEs for M L with L = H, F, Cl, Br, I, O, N, C, S,
and Se are in many cases known experimentally in vacuum and
constitute a benchmark database with the advantages described
above. Furthermore, BDEs are useful proxies for reaction barriers
of M L bond-breaking reactions in coordination chemistry.
Common non-hybrid GGA functionals or hybrids with HF
exchange of 20% or more (B3LYP, PBE0) exhibit mean errors in BDEs
of 4050 kJ/mol. These errors are largest for M L bonds with L = O,
S, N, C, i.e. bonds with higher bond order and more complex elec-
tronic structures. Early d-block M L bonds (Sc, Ti and V) exhibit
larger errors in BDEs [62]. However, mean errors are largely due
to systematic errors of over-binding in non-hybrid GGA function-
als and under-binding in B3LYP and PBE0 [62]. This suggested that
systematic errors may be reduced by reducing HF exchange, as was
previously observed in spin splitting [100].
 K.P. Kepp / Coordination Chemistry Reviews 257 (2013) 196209
Fig. 4. DFT-computed bond dissociation energies plotted against experimental data.
Figure reproduced with permission from Ref. [99]. Copyright 2008: The American
Chemical Society.
The meta hybrid TPSSh was developed from the non-empirical
TPSS, designed to fulll fundamental constraints, with 10% HF
exchange [41,43]. Thus, it was a plausible candidate for balancing
the systematic errors mentioned above. TPSSh improves perfor-
mance over GGA functionals for rst-row transition metal systems
[119]. It removes most of the systematic error of BDEs, as evi-
dent from a linear regression of experimental vs. computed BDEs
(Fig. 4) [99]. For the second- and third-row d-block, the coefcient
is 0.99 [136], i.e. the systematic error component is eliminated
in the whole d-block, not obtained by any other method so far
investigated. Uniform accuracy across the d-block is valuable upon
evaluation of compounds e.g. for catalyst design. This universal-
ity indicates the potential future success of a priori functionals in
inorganic chemistry. In a benchmark of binding energies with M05
which also included TPSSh [167], while M05 had a lower mean
absolute error than TPSSh (33 vs. 41 kJ/mol), TPSSh in fact com-
fortably gave the smallest signed error (5 kJ/mol) of all functionals
studied. This could suggest that TPSSh eliminates most of the sys-
tematic error also for saturated complexes.
Despite the elimination of the systematic error component,
large random errors of 40 kJ/mol remain in absolute BDEs across
the d-block when modeled with TPSSh (>50 kJ/mol with B3LYP, PBE,
BP86, etc.). A large part of this random error is due to pathologi-
cal cases confronting all functionals, e.g. the previously mentioned
4s1 dq+1 4s2 dq promotion energies that embody many of the
effects discussed above [11], notably the inert pair effect, the SIE
(if the d-conguration is JahnTeller instable), and HF exchange.
Therefore, these promotion energies are particularly interesting to
understand in future work.
MAEs of 40 kJ/mol for M L BDEs may sound discouraging and
are substantially larger than the average MAEs of main group reac-
tion energies, typically in the range of 525 kJ/mol for hybrids
and double hybrids [52]. However, the M L BDEs are extremely
challenging cases with substantial differential correlation energies.
In larger, substitution-saturated systems, energy differences and
absolute errors tend to be down-scaled due to the higher density
of states.
Many reactions are isodesmic, e.g. conformational and isomer-
ization energies that have modest differential correlation energies,
and many other reactions can be studied via their relative effects
based on a reference reaction, thus eliminating many of the hid-
den systematic errors mentioned above e.g. differential energies
of the type:
M L + M L M L + M L (8)
205
where systematic errors will cancel to an extent that goes well
beyond the cancelation achieved with 10% HF exchange [168].
Such energies were used to conclude that tetrapyrrole systems
have co-evolved with their respective, native metal ions (i.e. LS/HS
Fe(II/III) with porphyrin vs. LS Co(I/II/III) with corrin), based on
relative thermodynamics of metal exchange reactions that quan-
tify how tetrapyrrole cavity sizes and electrostatics are optimal
for each metal ion and multiplicity [169]. Using specially designed
isodesmic reference reactions such as Eq. (8) greatly increases accu-
racy by cancelation of otherwise obscure errors, and should, even
when practicing the inclusion of all effects in Table 1, be applied
when possible.
5. Metalligand bonds and spin crossover
5.1. General aspects of spin crossover
When bonds break in inorganic chemistry, it is common that
the overall spin changes. Together with bond dissociation of M L
bonds, spin crossover (SCO) is a challenging process to describe,
due to substantial changes in electron correlation, in particular
Fermi correlation. SCO is an important process found in widely
different areas of chemistry [170,171]. It occurs during many
M L bond dissociations, e.g. ligand binding to metalloproteins
[72,172], and is of great interest in molecular magnetism, catal-
ysis, and molecular storage [173176]. Thus, there have been
many attempts to provide accurate theoretical predictions of SCO
[101103,160,175,177181].
SCO occurs at thermal equilibrium and is subject to a very
high degree of entropyenthalpy compensation: typical values
are TSSCO 1030 kJ/mol and HSCO 1030 kJ/mol [163]. Thus,
using only the enthalpies or electronic energies to judge whether
a functional predicts the observed ground state (which includes
entropy) introduces a systematic error of this magnitude and such
benchmarks will articially favor methods that are HS-biased, e.g.
post-HF methods and hybrids with >15% HF exchange. As described,
both entropy and ZPE, added as corrections after electronic con-
vergence, almost uniformly favor the looser HS state, because
the occupied ligand-directed (anti-bonding) d-orbitals in the HS
states weaken and elongate M L bonds, lower ZPE, and increase
vibrational entropy [126]. SCO enthalpies for benchmarking are
interesting because of available, accurate experimental data [163]
and because the process depends critically on balanced Fermi cor-
relation that all functionals should strive to achieve. However, the
data are restricted to dq congurations in the center of the d-block
(q = 47), which precludes probing for universality.
5.2. HF exchange and spin states
As described above, HF exchange causes hybrid function-
als to favor HS whereas non-hybrids favor LS too much
[100,101,180,181]. Thus, in SCO complexes, use of B3LYP* with
15% [178] or B3LYP** [179] or TPSSh [160] with 10% HF exchange
provides better results than B3LYP. As the electronic HSLS gap is
nearly linear in HF exchange [100], it is not surprising that TPSSh
with 10% HF exchange usually falls in between non-hybrids and
B3LYP [126,160]. For nine iron and cobalt complexes (see Fig. 5),
TPSSh achieved a mean absolute error of 11 kJ/mol, probably close
to the limit of absolute errors attainable from theoretical chemistry
for these complex systems [160]. For purely main-group molecular
systems, TPSSh retains an accuracy similar to B3LYP for the G3/99
set [43,159].
Diverse spin states are particularly widespread in the vast eld
of porphyrin chemistry, of substantial importance in bioinorganic,
synthetic, and materials chemistry [20,182]. Thus, porphyrins
are important and challenging test cases for the reliable use of
206 K.P. Kepp / Coordination Chemistry Reviews 257 (2013) 196209
Fig. 5. Enthalpy differences between high- and low-spin states of some iron-
and cobalt complexes computed with different functionals, including thermal and
relativistic effects. Red bars signify experimental ranges. The acronyms refer to
complexes dened in Ref. [160].
Copyright 2009: The American Chemical Society.
modern electron-structure methods. In a recent work [183], DFT-
calculated energies of a number of electronic congurations of ve
iron(III)porphine complexes were compared to the experimentally
observed ground states [184] of these complexes. Given the above
results, it was surprising to observe that TPSSh is less accurate
for these systems [126]. However, these conclusions rested on the
premise that direct electronic DFT-energies could probe thermally
observed ground states, and inclusion of entropy, ZPE, and disper-
sion rendered TPSSh the most accurate functional, consistent with
other results [160].
Another example of erroneous conclusions from neglect of
entropy is the conclusion that B2PLYP [34] performs better than
TPSSh for spin-crossover systems [185]. This conicts with the
results cited above [126,160] and is also anomalous given the
large component (53%) of HF exchange in B2PLYP, since HF
exchange linearly affects spin splitting energies with 20% being
commonly too large [100,101,104,105]. Again, directly calculated
electronic energies of the relevant congurations were used as
proxies for experimentally observed ground states, neglecting
entropy [185]. It is known both experimentally and theoretically
[105,126,163,186] that HS states are favored by entropy by typi-
cally 20 kJ/mol per metal site, and its neglect leads to erroneous
appraisals of HF exchange and conclusions regarding the relative
merits of functionals.
For the archetypical spin-crossover complex Fe(1,10-
phenantroline)2 -(NCS)2 , electronic energies alone would show
TPSSh favoring LS by 17 kJ/mol [185], but upon entropy cor-
rection, the estimates will be very close to SCO as observed
experimentally, whereas B2PLYP will over-stabilize the HS state,
as expected from its large HF exchange component. After cor-
recting by 20 kJ/mol as a reasonable entropy, TPSSh is the most
accurate functional also for this data set, removing the conict
between that study and others. A dispersion correction would
work to the opposite effect, probably by less than the 515 kJ/mol
seen for substituted phenyl iron-porphine complexes [126], still
rendering TPSSh substantially more accurate than previously
found, and such new calculations might in fact be encouraged.
In conclusion, the practice of DFT prediction of thermal ground
states from electronic energies alone while neglecting entropy
should be discouraged, and 10% HF exchange seems appropriate
in these cases when the effects of Table 1 are accounted for.
Despite this, the general experience with B2PLYP has been
positive: variations of the functional perform well for main-
group thermochemistry involving also cases with some static
correlation, although the best version depends on reaction
type [187]. B2PLYP-D3, i.e. with the latest dispersion correction by
Grimme, also performed well in a benchmark of articial molecules
with less bias toward specic electronic structures, when compared
against CCSD(T) calculations [188]. B2PLYP accurately describes
some electronic excitation energies [189] and is a very promising
candidate for further development among the new functionals
that are steadily challenging B3LYPs market dominance [190].
6. Larger-system effects not described in this review
Although this paper has been concerned with the systematic
effects in Table 1 and how their corrections affect studies and
benchmarking of methods, the eld of inorganic chemistry is of
course much larger than the chemistry of individual M L bonds.
Many reviews only mentioned in the passing deal with larger
molecular systems and broad ranges of properties and should be
consulted for these purposes [5,6,1023].
Some notable effects occurring in larger systems are dispersion,
which as discussed favors compact states, and coordinative satura-
tion that increases the density of states and thus static correlation,
but also reduces apparent errors by scaling down energy differ-
ences. This usually means that the errors observed in M L BDEs
(with discouraging, random error components of up to 40 kJ/mol)
are larger than in saturated complexes, unless large systematic
errors from e.g. dispersion are present, but it also easily leads to
partial cancelation of errors and obscures the insight that is more
clear in simpler systems.
For larger systems, choosing a realistic model system that
includes the most signicant physical interactions becomes critical
[10,13,147,149]. Another important non-systematic effect rarely
dealt with is the sampling or local minimum problem, not of
electronic congurations (Section 2.4.), but of conformations and
isomers in coordination complexes and clusters, as investigated in
a few cases such as di-iron enzymes [162] and inorganic clusters
[191,192]. The sampling of isomers in these cases clearly affects
the identication of species that are considered probes of the
experimental system. This problem is even more substantial in the
modeling of proteins and cannot currently be dealt with, except in
relatively small systems.
Also, in larger systems, the chemical complexity may increase
e.g. by the presence of metalmetal bonds, which have also not
been discussed here. Metalmetal bonds are often electronically
even more complex due to involvement of more orbitals, rendering
electron correlation quite distinct from that in most metalligand
bonds, giving other trends than those discussed here [193]: for
example, non-hybrids (such as TPSS) perform better than most
hybrids, probably due to the massive static correlation that reduces
the accuracy of single-conguration HF exchange [167].
When it comes to molecular properties, the situation is equally
discouraging as for molecular systems, with different properties
requiring different choices of methods [6,1416,48]. However, as
attempted in this review, using the simplest possible systems to
reduce errors from e.g. dispersion, electronic conguration bias,
sampling, entropy, and solvent might be a path forward for each
property separately.
7. Concluding remarks
The Pareto principle (or 80-20 rule) states that in many situ-
ations, 80% of the results can be achieved from 20% of the input,
if the underlying dynamics are governed by power-law behavior
(exact numbers depend on the power-law exponent) [194]. Plausi-
bly, modern computational chemistry, as indeed many other elds
of modeling, suffers from its own 80-20 rule: while the exact
distribution depends on methodology and systems studied, a very
large part of the electron correlation energy is generally obtained
 K.P. Kepp / Coordination Chemistry Reviews 257 (2013) 196209 207

by a small fraction of the computational work, whereas the remain- B3LYP
ing small fraction of the correlation energy requires the majority of
the total efforts. This is clearly true in wave function theory, where
one- and many-electron spaces rapidly restrict accuracy due to B3LYP* B3LYP functional with 15% HF exchange instead of
size restrictions. While this distribution is less extreme in DFT, it is
still apparent from the increased algorithmic complexity associated BDE
with new functionals. BP86
However, as the central message of this review, additional
effects, often systematic and partly neglected, play comparable CASPT2 CASSCF subject to second-order perturbation the-
or even larger roles: ZPE, entropy, HF exchange, some relativistic
effects (notably scalar relativistic effects and spinorbit coupling on
higher-angular momentum bonds), and electronic self-interaction
tend to weaken bonds either directly (HF exchange, relativistic
effects, and SIE) or indirectly, as a correction to the converged elec- CASSCF complete active space self-consistent eld, a multi-
tronic structure (ZPE, entropy). They mimic static correlation and
thus favor higher spin-, radial-, and angular quantum numbers,
with resulting bias toward high-spin, excited, atomic, and disso-
ciated states throughout chemistry. In contrast, dispersion, some
solvent effects, and relativistic effects in low angular-momentum
bonded systems such as M-X -bonds, will often work to the oppo- CCSD(T) coupled-cluster with singlets, doublets, and parameter-
site effect of favoring compact electron densities.
As seen in Table 1, errors are often of comparable magnitude and CCSDT coupled-cluster with singlets, doublets, and triplets exci-
usually larger than 10 kJ/mol each, often summing to 50 kJ/mol
for non-isodesmic reaction energies. Lack of attention to any of CI
them may lead to systematic errors and erroneous conclusions. For
example, studies that neglect entropy, scalar-relativistic effects,
and ZPE may erroneously favor methods that resemble the HF
determinant to higher extent, e.g. B3LYP. Thus, accounting for DFT
these systematic effects, preferably in benchmarks that elucidate GGA
them carefully, may be worthwhile before complicating density HF
functionals further.
As many researchers have testied [13,22,62,81,167], there
is no one-size-ts-all in current DFT, although the inclusion of HS high-spin
all the mentioned systematic effects should render it more clear IS
which functional to use for a given property. However, geometries KS
can be computed with any standard functional of at least GGA-
type, once the necessary effects (e.g. solvation in charged states) LS
are included [6264]. For energies, some researchers including M0x
the present author hold that HF exchange should be 1015% for
GGA- and meta-hybrids as in B3LYP* and TPSSh, whereas non-
hybrid GGA functionals tend to perform worse [81,99,126,128]. MO
Scalar-relativistic effects and dispersion e.g. in the form of DFT-D3 MP2
[124] should be evaluated routinely. For barriers used to derive
rate constants, no general consensus exists, although SIE is often PBE
a larger problem that suggests use of more HF exchange as e.g. PBE0
in B2PLYP [187,195197]. Finally, new functionals such as double PM3
hybrids should be considered valid alternatives, although better
experience is warranted. ROHF
Most important is arguably the need to pick the low-hanging
fruit by eliminating the discussed systematic errors, to generally
use several functionals for reality checking results, to always com-
pare results to experimental data wherever possible during the SCF
project, to use advanced correlated methods to validate and under-
stand DFT energies, and to design carefully problems to incorporate
large extents of error cancelation, e.g. only quantifying energies rel- SCO
ative to a reference. Hopefully, such a good practice may remove SIE
some of the confusion that currently resides within the inorganic
chemistry community regarding the performance and choices of TPSS
methods for particular modeling tasks.
TPSSh
UHF
List of abbreviations
B2PLYP recently developed double hybrid by Grimme and co-
workers
3-parameter hybrid GGA functional by Becke, using func-
tionals developed by Colle-Salvetti, Lee, Yang, Parr, and
others, with 20% HF exchange
20%
bond dissociation enthalpy
combination of Becke 88 exchange and Perdew 1986 cor-
relation functionals
ory to dynamically correlate a CASSCF wave func-
tion: a useful method in inorganic chemistry since
it systematically includes both static and dynamic
correlation
conguration method by Roos and co-workers that
performs a full CI optimization of all congurations
within an active space of the molecules MOs, thus
describing static correlation not obtained in post-HF
single-reference ab initio methods
ized triplets
tations included
conguration interaction, wave function theory with use
of complex wave functions that are superpositions of
several electronic congurations, typically Slater deter-
minants, to incorporate electron correlation
density functional theory
generalized gradient approximation
HartreeFock, simple wave function method that uses
one antisymmetrized Slater determinant as wave func-
tion
intermediate-spin
KohnSham procedure, using a HF-like SCF procedure to
optimize MOs under the inuence of the functional
low-spin
Minnesota-type functionals, developed by Truhlar and
co-workers (examples include M05, M06-L, M06-2X,
M08-HX)
molecular orbital
MllerPlesset second order perturbation theory on a HF
reference determinant
Perdew Burke Ernzerhof a priori GGA functional
25% hybrid version of PBE
parameterized model 3, a semi-empirical model by Stew-
art
restricted open-shell HartreeFock (also applies to DFT),
meaning that all MOs are required to be doubly occupied
except sometimes the highest occupied MOs, depending
on the state
self-consistent eld, a concept used in HF theory of single
electrons moving in an effective, iteratively solved eld
of other electrons
spin crossover
self-interaction error (or self-interaction energy as this is
an energy)
Tao Perdew Staroverov, Scuseria a priori meta GGA func-
tional
10% hybrid version of TPSS
unrestricted HartreeFock (also applies to DFT, where it
is called UKS for unrestricted KohnSham), meaning that
- and -spin electrons are treated as separate spaces,
i.e. MOs are singly occupied. This allows spin polariza-
tion of core electrons and provides a better description of
correlation in most cases
208 K.P. Kepp / Coordination Chemistry Reviews 257 (2013) 196209
WFT wave function theory, usually based on a HartreeFock
reference wave function
ZPE zero point energy
Acknowledgements
The Danish Center for Scientic Computing (DCSC) and the Dan-
ish Science Research Council (FNU, Steno Grant: 272-08-0041) are
acknowledged for supporting this work.
References
[1] W. Kohn, A.D. Becke, R.G. Parr, J. Phys. Chem. 100 (1996) 1297412980.
[2] W. Kohn, L.J. Sham, Phys. Rev. 140 (1965) A1133A1138.
[3] R.G. Parr, W. Yang, Density Functional Theory of Atoms and Molecules, Oxford
University Press, New York, 1989.
[4] http://www.nobelprize.org/nobel prizes/chemistry/laureates/1998/.
[5] P. Geerlings, F. De Proft, W. Langenaeker, Chem. Rev. 103 (2003) 17931873.
[6] C.J. Cramer, D.G. Truhlar, Phys. Chem. Chem. Phys. 11 (2009) 1075710816.
[7] E.I. Solomon, M.D. Lowery, Science 259 (1993) 15751581.
[8] E.I. Solomon, T.C. Brunold, M.I. Davis, J.N. Kemsley, S.-K. Lee, N. Lehnert, F.
Neese, A.J. Skulan, Y.-S. Yang, J. Zhou, Chem. Rev. 100 (2000) 235349.
[9] U. Ryde, K. Nilsson, J. Am. Chem. Soc. 125 (2003) 1423214233.
[10] U. Ryde, Dalton Trans. (2007) 607625.
[11] M.C. Holthausen, J. Comput. Chem. 26 (2005) 15051518.
[12] T. Ziegler, Chem. Rev. 91 (1991) 651667.
[13] P.E.M. Siegbahn, J. Biol. Inorg. Chem. 11 (2006) 695701.
[14] S.F. Sousa, P.A. Fernandes, M.J. Ramos, J. Phys. Chem. A 111 (2007)
1043910452.
[15] K.E. Riley, B.T. Opt Holt, K.M. Merz Jr., J. Chem. Theory Comput. 3 (2007)
407433.
[16] F. Neese, Coord. Chem. Rev. 253 (2009) 526563.
[17] L. Hermosilla, P. Calle, J.M.G. de la Vega, C. Sieiro, J. Phys. Chem. A 109 (2005)
11141124.
[18] S. Li, D.A. Dixon, J. Phys. Chem. A 111 (2007) 1190811921.
[19] E. Oldeld, Phil. Trans. R. Soc. B 360 (2005) 13471361.
[20] A. Ghosh, J. Biol. Inorg. Chem. 11 (2006) 712724.
[21] P.E.M. Siegbahn, M.R.A. Blomberg, Chem. Rev. 100 (2000) 421437.
[22] J.N. Harvey, Annu. Rep. Prog. Chem. C 102 (2006) 203226.
[23] C. Daniel, Coord. Chem. Rev. 238239 (2003) 143166.
[24] J.C. Slater, Phys. Rev. 81 (1951) 385390.
[25] E. Fermi, Rend. Accad. Naz. Lincei 6 (1927) 602607.
[26] L.H. Thomas, Proc. Camb. Philos. Soc. 23 (1927) 542548.
[27] P.A.M. Dirac, Proc. Camb. Philos. Soc. 26 (1930) 376.
[28] J.P. Perdew, Phys. Rev. Lett. 55 (1985) 1665.
[29] J.P. Perdew, Phys. Rev. B 33 (1986) 88228824.
[30] A.D. Becke, Phys. Rev. A 38 (1988) 30983100.
[31] P. Jemmer, P.J. Knowles, Phys. Rev. A 51 (1995) 35713575.
[32] A.D. Becke, J. Chem. Phys. 104 (1996) 1040.
[33] A.D. Becke, J. Chem. Phys. 98 (1993) 56485652.
[34] S. Grimme, J. Chem. Phys. 124 (2006) 034108.
[35] Y. Tawada, T. Tsuneda, S. Yanagisawa, T. Yanai, K. Hirao, J. Chem. Phys. 120
(2004) 8425.
[36] J.D. Chai, M. Head-Gordon, J. Chem. Phys. 128 (2008) 084106.
[37] D. Joubert, M. Levy, Phys. Rev. A 54 (1996) 961963.
[38] M. Levy, A. Grling, Phys. Rev. A 51 (1995) 28512856.
[39] K. Burke, J.P. Perdew, M. Levy, Phys. Rev. A 53 (1996) R2915R2917.
[40] J.P. Perdew, K. Burke, M. Ernzerhof, Phys. Rev. Lett. 77 (1996) 38653868.
[41] J. Tao, J.P. Perdew, V.N. Staroverov, G.E. Scuseria, Phys. Rev. Lett. 91 (2003)
146401.
[42] J.P. Perdew, M. Ernzerhof, K. Burke, J. Chem. Phys 105 (1996) 99829985.
[43] V.N. Staroverov, G.E. Scuseria, J. Tao, J.P. Perdew, J. Chem. Phys. 119 (2003)
1212912137.
[44] S. Kurth, J.P. Perdew, P. Blaha, Int. J. Quantum Chem. 75 (1999) 889909.
[45] P.J. Stephens, F.J. Devlin, C.F. Chabalowski, M.J. Frisch, J. Phys. Chem. 98 (1994)
1162311627.
[46] C. Lee, W. Yang, R.G. Parr, Phys. Rev. B 37 (1988) 785789.
[47] S.H. Vosko, L. Wilk, M. Nusair, Can. J. Phys. 58 (1980) 12001211.
[48] Y. Zhao, D.G. Truhlar, Theor. Chem. Acc. 120 (2008) 215241.
[49] J.J.P. Stewart, J. Comput. Chem. 10 (1989) 209220.
[50] J.J.P. Stewart, J. Comput. Chem. 12 (1991) 320341.
[51] Y. Zhao, N.E. Schultz, D.G. Truhlar, J. Chem. Phys. 123 (2005) 161103.
[52] J.A. Pople, in: D.W. Smith (Ed.), Proceedings of the Summer Research Confer-
ence on Theoretical Chemistry, Energy Structure and Reactivity, John Wiley
& Sons, Ltd., New York, 1973.
[53] R.H. Hertwig, W. Koch, Chem. Phys. Lett. 268 (1997) 345351.
[54] L. Goerigk, S. Grimme, Phys. Chem. Chem. Phys. 13 (2011) 66706688.
[55] M.J.G. Peach, D.J. Tozer, N.C. Handy, Int. J. Quantum Chem. 111 (2011)
563569.
[56] G. Frenking, N. Frhlich, Chem. Rev. 100 (2000) 717774.
[57] P.-O. Lwdin, Adv. Chem. Phys. 2 (1959) 207322.
[58] J.C. Slater, Phys. Rev. 34 (1929) 12931322.
[59] A. Ruzsinszky, J.P. Perdew, Comput. Theor. Chem. 963 (2011) 26.
[60] C.W. Bauschlicher Jr., A. Ricca, H. Patridge, in: D.P. Chong (Ed.), Recent
Advances in Density Functional Methods, World Scientic, 1997, p. 165
(Chapter 6).
[61] J.P. Perdew, A. Ruzsinszky, Int. J. Quantum Chem. 110 (2010) 28012807.
[62] K.P. Jensen, B.O. Roos, U. Ryde, J. Chem. Phys. 126 (2007) 014103.
[63] M. Bhl, C. Reimann, D.A. Pantazis, T. Bredow, F. Neese, J. Chem. Theory Com-
put. 4 (2008) 14491459.
[64] M. Bhl, H. Kabrede, J. Chem. Theory Comput. 2 (2006) 12821290.
[65] S.F. Sousa, E.S. Carvalho, D.M. Ferreira, I.S. Tavares, P.A. Fernandes, M.J. Ras-
mos, J.A.N.F. Gomes, J. Comput. Chem. 30 (2009) 27522763.
[66] Y. Zhao, D.G. Truhlar, J. Phys. Chem. A 108 (2004) 69086918.
[67] K.P. Jensen, U. Ryde, J. Mol. Struct. (THEOCHEM) 585 (2002) 239255.
[68] M.H. Olsson, U. Ryde, B.O. Roos, Protein Sci. 7 (1998) 26592668.
[69] P.J. Hay, J. Chem. Phys. 66 (1977) 43774384.
[70] K.P. Jensen, B.L. Ooi, H.E.M. Christensen, Inorg. Chem. 46 (2007) 87108716.
[71] J. Grfenstein, D. Cremer, Mol. Phys. 99 (2001) 981989.
[72] K.P. Jensen, U. Ryde, J. Biol. Chem. 279 (2004) 1456114569.
[73] J.G. Norman Jr., P.B. Ryan, L. Noodleman, J. Am. Chem. Soc. 102 (1980)
42794282.
[74] J.F. Stanton, J. Gauss, Adv. Chem. Phys. 125 (2003) 101146.
[75] J. Paldus, X. Li, Adv. Chem. Phys. 110 (1999) 1175.
[76] P. Piecuch, K. Kowalski, I.S.O. Pimienta, S.A. Kucharski, in: M.R. Hoffmann,
K.G. Dyall (Eds.), Low-lying Potential Energy Surfaces, ACS Symp. Ser. 828,
Washington, DC, 2002.
[77] P. Piecuch, L. Adamowicz, J. Chem. Phys. 102 (1995) 898904.
[78] A.I. Krylov, Acc. Chem. Res. 39 (2006) 8391.
[79] J. Shen, Z. Kou, E. Xu, S. Li, J. Chem. Phys. 133 (2010) 234106.
[80] B.O. Roos, Acc. Chem. Res. 32 (1999) 137144.
[81] F. Neese, D.G. Liakos, S. Ye, J. Biol. Inorg. Chem. 16 (2011) 821829.
[82] B.O. Roos, K. Andersson, M.P. Flscher, P.-. Malmqvist, L. Serrano-Andres,
Adv. Chem. Phys. 93 (1996) 219331.
[83] T.K. Todorova, L. Gagliardi, J.R. Walensky, K.A. Miller, W.J. Evans, J. Am. Chem.
Soc. 132 (2010) 1239712403.
[84] K. Pierloot, Mol. Phys. 101 (2003) 20832094.
[85] N.B. Amor, S. Zlis, C. Daniel, Int. J. Quantum Chem. 106 (2006) 24582469.
[86] S. Zlis, N.B. Amor, C. Daniel, Inorg. Chem. 43 (2004) 79787985.
[87] A. Ghosh, B.J. Persson, P.R. Taylor, J. Biol. Inorg. Chem. 8 (2003) 507511.
[88] A. Ghosh, P.R. Taylor, J. Chem. Theory Comput. 1 (2005) 597600.
[89] K.P. Jensen, J. Phys. Chem. B 109 (2005) 1050510512.
[90] K.P. Jensen, B.O. Roos, U. Ryde, J. Inorg. Biochem. 99 (2005) 4554.
[91] K. Pierloot, Int. J. Quantum Chem. 111 (2011) 32913301.
[92] S. Villaume, A. Strich, C.A. Ndoye, C. Daniel, S.A. Perera, R.J. Bartlett, J. Chem.
Phys. 126 (2007) 154318.
[93] S. Villaume, C. Daniel, A. Strich, S.A. Perera, R.J. Bartlett, J. Chem. Phys. 126
(2005) 044313.
[94] A. Vlcek Jr., S. Zlis, Coord. Chem. Rev. 251 (2007) 258287.
[95] J.P. Perdew, M. Ernzerhof, A. Zupan, K. Burke, in: J.M. Seminario, P.-O. Lwdin,
J.R. Sabin, M.C. Zerner (Eds.), Advances in Density Functional Theory, vol. 33,
Academic Press, San Diego, 1999.
[96] C.J. Barden, J.C. Rienstra-Kiracofe, H.F. Schaefer III, J. Chem. Phys. 113 (2000)
690700.
[97] S.R. Langhoff, C.W. Bauschlicher Jr., Annu. Rev. Phys. Chem. 39 (1988)
181212.
[98] J. Harris, R.O. Jones, J. Phys. F 4 (1974) 11701186.
[99] K.P. Jensen, Inorg. Chem. 47 (2008) 1035710365.
[100] M. Reiher, Inorg. Chem. 41 (2002) 69286935.
[101] H. Paulsen, L. Duelund, H. Winkler, H. Toftlund, A.X. Trautwein, Inorg. Chem.
40 (2001) 22012203.
[102] S. Zein, S.A. Borshch, P. Fleurat-Lessard, M.E. Casida, H. Chermette, J. Chem.
Phys. 126 (2007) 014105.
[103] L.M.L. Daku, A. Vargas, A. Hauser, A. Fouqueau, M.E. Casida, ChemPhysChem
6 (2005) 13931410.
[104] M. Reiher, O. Salomon, B.A. Hess, Theor. Chem. Acc. 107 (2001) 4855.
[105] H. Paulsen, A.X. Trautwein, Top. Curr. Chem. 235 (2004) 197219.
[106] G. Schenk, M.Y.M. Pau, E.I. Solomon, J. Am. Chem. Soc. 126 (2004) 505515.
[107] M.M. Quintal, A. Karton, M.A. Iron, A.D. Boese, J.M.L. Martin, J. Phys. Chem. A
110 (2006) 709716.
[108] V. Polo, E. Kraka, D. Cremer, Mol. Phys. 100 (2002) 17711790.
[109] O.A. Vydrov, Correcting the self-interaction error of approximate density
functionals, Ph.D. Thesis, Rice University, Houston, TX, 2007.
[110] O.A. Vydrov, G.E. Scuseria, J. Chem. Phys. 121 (2004) 8187.
[111] P. Mori-Sanchez, A.J. Cohen, W. Yang, J. Chem. Phys. 125 (2006) 201102.
[112] Y. Zhang, W. Yang, J. Chem. Phys. 109 (1998) 26042608.
[113] J. Grfenstein, E. Kraka, D. Cremer, J. Chem. Phys. 120 (2004) 524539.
[114] M. Lundberg, P.E.M. Siegbahn, J. Chem. Phys. 122 (2005) 224103.
[115] D. Cremer, Mol. Phys. 99 (2001) 18991940.
[116] J. Grfenstein, D. Cremer, Theor. Chem. Acc. 123 (2009) 171182.
[117] E. Ruiz, S. Alvarez, J. Cano, V. Polo, J. Chem. Phys. 123 (2005) 164110.
[118] A.J. Johansson, M.R.A. Blomberg, P.E.M. Siegbahn, J. Chem. Phys. 129 (2008)
154301.
[119] F. Furche, J.P. Perdew, J. Chem. Phys. 124 (2006) 044103.
[120] S. Grimme, J. Comput. Chem. 25 (2004) 14631473.
[121] S. Grimme, J. Comput. Chem. 27 (2006) 17871799.
[122] S.N. Steinmann, C. Corminboeuf, J. Chem. Theory Comput. 7 (2011)
35673577.
 K.P. Kepp / Coordination Chemistry Reviews 257 (2013) 196209
[123] A. Tkatchenko, M. Schefer, Phys. Rev. Lett. 192 (2009) 073005.
[124] S. Grimme, J. Antony, S. Ehrlich, H. Krieg, J. Chem. Phys. 132 (2010)
154104.
[125] S. Grimme, S. Ehrlich, L. Goerigk, J. Comput. Chem. 32 (2011) 14561465.
[126] K.P. Kepp, J. Inorg. Biochem. 105 (2011) 12861292.
[127] K.P. Jensen, U. Ryde, J. Phys. Chem. A 107 (2003) 75397545.
[128] P.E.M. Siegbahn, M.R.A. Blomberg, S.-L. Chen, J. Chem. Theor. Comput. 6 (2010)
20402044.
[129] M.D. Wodrich, C. Corminboeuf, P. von Rage Schleyer, Org. Lett. 8 (2006)
36313634.
[130] L.A. Burns, A. Vazquez-Mayagoitia, B.G. Sumpter, C.D. Sherrill, J. Chem. Phys.
134 (2011) 084107.
[131] Y. Zhao, D.G. Truhlar, J. Chem. Theory Comput. 1 (2005) 415432.
[132] P. Pyykko, Chem. Rev. 88 (1988) 563594.
[133] D.J. Gorin, F.D. Toste, Nature 446 (2007) 395403.
[134] F. Neese, E.I. Solomon, Inorg. Chem. 37 (1998) 65686582.
[135] P.A. Christiansen, W.C. Ermler, K.S. Pitzer, Annu. Rev. Phys. Chem. 36 (1985)
407432.
[136] K.P. Jensen, J. Phys. Chem. A 113 (2009) 1013310141.
[137] M. Hargittai, Acc. Chem. Res. 42 (2009) 453462.
[138] I.B. Bersuker, Electronic Structure and Properties of Transition Metal Com-
pounds, Wiley & Sons, New Jersey, 2010.
[139] D.J. Grant, M.H. Matus, J.R. Switzer, D.A. Dixon, J.S. Francisco, K.O. Christe, J.
Phys. Chem. A 112 (2008) 31453156.
[140] C.W. Bauschlicher Jr., Theor. Chem. Acc. 101 (1999) 421425.
[141] J.M.L. Martin, A. Sundermann, P.L. Fast, D.G. Truhlar, J. Chem. Phys. 113 (2000)
13481358.
[142] S.-G. Wang, W. Liu, W.H.E. Schwarz, J. Phys. Chem. A 106 (2002) 795803.
[143] K.S. Thanthiriwatte, E.G. Hohenstein, L.A. Burns, C.D. Sherrill, J. Chem. Theory
Comput. 7 (2011) 8896.
[144] K.P. Jensen, J. Inorg. Biochem. 102 (2008) 87100.
[145] K.P. Jensen, M. Rykr, Dalton Trans. 39 (2010) 96849695.
[146] A. Klamt, G.J. Schrmann, J. Chem. Soc. Perkin Trans. 2 (1993) 799805.
[147] P.E.M. Siegbahn, F. Himo, J. Biol. Inorg. Chem. 14 (2009) 643651.
[148] P.E.M. Siegbahn, F. Himo, Wiley Interdiscip. Rev.: Comput. Mol. Sci. 1 (2011)
323336.
[149] L. Hu, J. Eliasson, J. Heimdal, U. Ryde, J. Phys. Chem. A 113 (2009) 1179311800.
[150] K.P. Kepp, J. Inorg. Biochem. 107 (2012) 1524.
[151] A. Warshel, J. qvist, Annu. Rev. Biophys. Biophys. Chem. 20 (1991) 267298.
[152] M. Naja, M. Zahedi, E. Klein, Comput. Theor. Chem. 978 (2011) 1628.
[153] T. Dudev, Y.-I. Lin, M. Dudev, C. Lim, J. Am. Chem. Soc. 125 (2003) 31683180.
[154] K.P. Jensen, U. Ryde, J. Am. Chem. Soc. 125 (2003) 1397013971.
[155] H. Goff, G.N. La Mar, C.A. Reed, J. Am. Chem. Soc. 99 (1977) 36413646.
[156] J.P. Collman, J.L. Hoard, N. Kim, G. Lang, C.A. Reed, J. Am. Chem. Soc. 97 (1975)
26762681.
[157] M. Radon, K. Pierloot, J. Phys. Chem. 112 (2008) 1182411832.
[158] T. Kitagawa, J. Teraoka, Chem. Phys. Lett. 63 (1979) 443446.
[159] L.A. Curtiss, K. Raghavachari, P.C. Redfern, J.A. Pople, Chem. Phys. Lett. 270
(1997) 419426.
[160] K.P. Jensen, J. Cirera, J. Phys. Chem. A 113 (2009) 1003310039.
209
[161] G. Roos, P. Geerlings, J. Messens, J. Phys. Chem. B 113 (2009) 1346513475.
[162] K.P. Jensen, C.B. Bell III, M.D. Clay, E.I. Solomon, J. Am. Chem. Soc. 131 (2009)
1215512171.
[163] J.W. Turner, F.A. Schultz, Coord. Chem. Rev. 219221 (2001) 8197.
[164] G. Brewer, M.J. Olida, A.M. Schmiedekamp, C. Viragh, P.Y. Zavalij, Dalton Trans.
(2006) 56175629.
[165] M.P. Waller, H. Braun, N. Hojdis, M. Bhl, J. Chem. Theory Comput. 3 (2007)
22342242.
[166] J.P. Perdew, A. Ruzsinszky, L.A. Constantin, J. Sun, G.I. Csonka, J. Chem. Theory
Comput. 5 (2009) 902908.
[167] Y. Zhao, D.G. Truhlar, J. Chem. Phys. 124 (2006) 224105.
[168] K.P. Jensen, in: E.I. Solomon, R.A. Scott, R. Bruce King (Eds.), Computational
Inorganic and Bioinorganic Chemistry, John Wiley & Sons, Ltd., Chichester,
UK, 2009, pp. 373386.
[169] K.P. Jensen, U. Ryde, ChemBioChem 4 (2003) 413424.
[170] J.A. Real, A.B. Gaspar, M.C. Munoz, Dalton Trans. 12 (2005) 20622079.
[171] H.A. Goodwin, Coord. Chem. Rev. 18 (1976) 293325.
[172] N. Strickland, J.N. Harvey, J. Phys. Chem. B 111 (2007) 841852.
[173] E. Knig, K. Madeja, Inorg. Chem. 7 (1967) 18481855.
[174] P. Gtlich, A. Hauser, H. Spiering, Angew. Chem. Int. Ed. Engl. 33 (1994)
20242054.
[175] R.J. Deeth, A.E. Anastasi, M.J. Wilcockson, J. Am. Chem. Soc. 132 (2010)
68766877.
[176] A.V. Postnikov, G. Bihlmayer, S. Blugel, Comput. Mater. Sci. 36 (2006) 9195.
[177] A. Sorkin, M.A. Iron, D.G. Truhlar, J. Chem. Theory Comput. 4 (2008) 307315.
[178] J. Conradie, A. Ghosh, J. Phys. Chem. B 111 (2007) 1262112624.
[179] I. Respondek, L. Bressel, P. Saalfrank, H. Kampf, A. Grohmann, Chem. Phys. 347
(2008) 514522.
[180] M. Swart, A.R. Groenhof, A.W. Ehlers, K. Lammertsma, J. Phys. Chem. A 108
(2004) 54795483.
[181] R.J. Deeth, N. Fey, J. Comput. Chem. 25 (2004) 18401848.
[182] W.R. Scheidt, C.A. Reed, Chem. Rev. 81 (1981) 543555.
[183] M.M. Conradie, J. Conradie, A. Ghosh, J. Inorg. Biochem. 105 (2011) 8491.
[184] M.K. Kadish, E. Van Caemelbecke, E. Gueletii, S. Fukuzumi, K. Miyamoto, T.
Suenobu, A. Tabard, R. Guilard, Inorg. Chem. 37 (1998) 17591766.
[185] S. Ye, F. Neese, Inorg. Chem. 49 (2010) 772774.
[186] G. Baranovic, Chem. Phys. Lett. 369 (2003) 668672.
[187] A. Karton, A. Tarnopolsky, J.-F. Lamere, G.C. Schatz, J.M.L. Martin, J. Phys. Chem.
A 112 (2008) 1286812886.
[188] M. Korth, S. Grimme, J. Chem. Theory Comput. 5 (2009) 9931003.
[189] S. Grimme, F. Neese, J. Chem. Phys. 127 (2007) 154116.
[190] T. Schwabe, S. Grimme, Acc. Chem. Res. 41 (2008) 569579.
[191] A.N. Alexandrova, A.I. Boldyrev, J. Chem. Theory Comput. 1 (2005) 566580.
[192] D.Y. Zubarev, A.N. Alexandrova, A.I. Boldyrev, L.F. Cui, X. Li, L.S. Wang, J. Chem.
Phys. 124 (2006) 124305.
[193] S. Yanagisawa, T. Tsuneda, K. Hirao, J. Chem. Phys. 112 (2000) 545553.
[194] A. Grosfeld-Nir, B. Ronen, N. Kozlovsky, Int. J. Prod. Res. 45 (2007) 23172325.
[195] H.C. Fang, Z.H. Li, K.N. Fan, Phys. Chem. Chem. Phys. 13 (2011) 1335813369.
[196] T.R. Ramadhar, R.A. Batey, Comput. Theor. Chem. 976 (2011) 167182.
[197] S. Goel, A.E. Masunov, Int. J. Quantum Chem. 111 (2011) 42764287.