You are on page 1of 11

Polyhedron 208 (2021) 115399

Contents lists available at ScienceDirect

Polyhedron
journal homepage: www.elsevier.com/locate/poly

Spin-state energetics of manganese spin crossover complexes: Comparison


of single-reference and multi-reference ab initio approaches☆
Maria Drosou a, Christiana A. Mitsopoulou a, Dimitrios A. Pantazis b, *
a
Inorganic Chemistry Laboratory, National and Kapodistrian University of Athens, Panepistimiopolis, Zografou 15771, Greece
b
Max-Planck-Institut für Kohlenforschung, Kaiser-Wilhelm-Platz 1, 45470 Mülheim an der Ruhr, Germany

A R T I C L E I N F O A B S T R A C T

Keywords: Manganese spin crossover (SCO) complexes form a small but ever expanding family of compounds with ther­
Spin-state energetics mally accessible states of different electronic configuration and total spin. Accurate prediction of spin-state
Spin crossover energetics is essential for the theoretical description of these systems. However, this represents a challenging
Coupled cluster theory
problem that necessitates recourse to correlated wave function methods rather than the more approximate
Multireference calculations
DFT
density functional theory (DFT). Here we present a detailed study of spin-state energetics for eight Mn(III) and
Mn(II) SCO complexes using the domain-based local pair natural orbital approach to coupled cluster theory with
singles, doubles, and perturbative triples, DLPNO-CCSD(T). The effects of reference determinants, basis set,
triples excitations, and pair natural orbitals (PNO) thresholds are evaluated and analysed in detail, enabling us to
propose a robust and efficient computational protocol based on a combined and balanced mix of extrapolation to
the complete basis set and infinite PNO space limits. The results are subsequently used to evaluate multireference
wavefunction-based (CASSCF/NEVPT2) and DFT approaches, highlighting their inability to provide a balanced
description of spin-state energetics for these complexes. The DLPNO-CCSD(T) protocol proposed in this study can
serve as a generally applicable reference-quality quantum chemical method for studying spin crossover systems.

1. Introduction entropy of the HS state. The thermally induced LS → HS spin transition is


an entropy driven process. The degrees of freedom are higher in the HS
Spin crossover (SCO) transition metal complexes undergo spin state state, attributed to the increase in spin multiplicity and in the intra­
changes induced by external stimuli, such as temperature, light, and molecular vibrations [4]. It follows that SCO systems with transition
pressure, as a result of redistribution of electrons in partially filled temperatures around 100–300 K have a LS ground state at low tem­
d shells along with geometric rearrangements. SCO materials can be peratures, i.e. the LS state has a lower electronic energy and the HS state
used as molecular switches with significant applications in data storage, has to be a few kcal mol− 1 higher.
display devices, sensors and MRI contrast agents [1–3]. A molecule can In the matter of design, the electronic configuration of the metal
show SCO behaviour if the spin states involved are energetically close. center and the ligand field strength determine the multiplicity in the
The spin transition can be considered as a phase transition accompanied ground state as well as the spin transition temperature. In contrast to
with a change of the Gibbs free energy ΔG = ΔG − TΔS. The SCO other first row transition metals such as Fe(II) [5], Fe(III) [6,7], Co(II)
behaviour of a complex is characterized by its transition temperature [8] and Co(III) [9], Mn SCO complexes are rare. SCO-active octahedral
T1/2 , which is the temperature at which the high-spin (HS) and low-spin Mn(III) complexes (d4 electron configuration) have three possible spin
(LS) states coexist in 1:1 ratio, i.e. states: HS with S = 2, intermediate spin (IS) with S = 1 and LS with S = 0,
( ) and they switch between S = 1 and S = 2 (IS → HS). Mn(II) SCO-active
ΔGHS− LS = GHS − GLS = H HS − T1/2 SHS − H LS − T1/2 SLS complexes (d5) can be observed from the S = 5/2 HS state to the S = 1/2
= EHS LS HS LS
el − E el + E vib − E vib + T1/2 S
LS
− T1/2 SHS = 0 (1) LS state. d4 and d5 octahedral complexes show the tendency to adopt the
HS state, because the loss of exchange energy, which would stabilize the
where EHS
el the electronic energy, Evib the vibrational energy and S
HS HS
the system in the LS state due to pair formation with parallel electron spins,


Part of the special issue dedicated to the element manganese, entitled “Manganese: A Tribute to Chemical Diversity”
* Corresponding author.
E-mail address: dimitrios.pantazis@kofo.mpg.de (D.A. Pantazis).

https://doi.org/10.1016/j.poly.2021.115399
Received 25 April 2021; Accepted 26 July 2021
Available online 30 July 2021
0277-5387/© 2021 Elsevier Ltd. All rights reserved.
M. Drosou et al. Polyhedron 208 (2021) 115399

usually cannot compensate for the ligand field energy [9]. Therefore, Mn Here we investigate the application of the DLPNO-CCSD(T) method
SCO-active complexes must bear stronger field ligands. In addition, both to the calculation of spin-splitting energies of Mn(III) and Mn(II) SCO
states involved in the SCO process for Mn(II) and Mn(III) complexes are complexes. We also evaluate multireference CASSCF/NEVPT2 and a
paramagnetic, which may complicate the interpretation of their mag­ number of popular DFT functionals and compare them to the reference
netic properties due to the possibility of (anti)ferromagnetic exchange coupled cluster results. There exist 25 Mn SCO complexes that have been
and states admixing [10]. The SCO-properties of Mn(III) and Mn(II) experimentally characterized and found to exhibit thermal spin cross­
complexes have been recently reviewed [9,10]. over behaviour (Figure S1 and Table S1). They can be divided into four
Knowledge of accurate spin state-splitting energies is essential in subsets: a) complexes that bear hexadentate salen-type ligands of
order to rationalize the magnetic properties of existing SCO complexes substituted salicylaldehyde units and tetraamines, b) complexes that
and to design novel SCO-active materials with specific properties. have tridentate N2O ligands, c) complexes with N donor ligands, and d)
Magnetic susceptibility measurements are applied on either crystalline Mn(II) manganocenes. In our study we selected the eight complexes
or ground samples, thus the SCO behaviour depends on several envi­ shown in Fig. 1, which are representative of all four types of Mn SCO
ronmental effects such as crystal lattice, counterion and the amount and complexes.
nature of solvent in the crystal lattice [10,11]. Computational studies
describe the electronic structure of the isolated complex, which in­ 2. Computational details
troduces a significant approximation. Thus, semi-quantitative criteria
are typically adopted in computational studies. A difference of up to 10 All calculations in this work were carried out using the Orca program
kcal mol− 1 between the electronic energy of the HS and LS forms in their suite [37].
respective ground state geometries is an indicator of potential SCO ac­
tivity [12]. The goal of molecular computational chemistry is therefore 2.1. Geometry optimizations
to define a reliable, efficient, and transferable method for preliminary
screening to identify potential candidates for SCO complexes to be Geometry optimizations of the Mn SCO complexes C1–C8 were
synthesized and studied experimentally. performed starting from structures taken from the Cambridge Structural
There is general consensus that different SCO systems are optimally Database (CSD) [38]. Optimizations were conducted separately in the
described by different density functional theory (DFT) methods, so a respective spin states using the TPSSh functional [39], which has been
universal approach is still lacking [13]. Recently, other groups have found to perform well on reproducing both the geometrical parameters
presented DFT benchmark studies on mononuclear SCO complexes [12], and relative energies of different spin states of Mn complexes
Mn(III) SCO complexes [14] and Mn(II) manganocenes [15]. Investi­ [12,14,40]. The resolution of identity (RI) approximation was used
gation of spin-state energetics involves calculation of energy differences along with the chain-of-spheres (COSX) approximation [41] to exact
between states with different numbers of unpaired electrons, a situation exchange in order to accelerate the SCF step. The calculations were
where systematic errors in DFT methods do not cancel out. In order to performed within a spin unrestricted formalism for open-shell com­
avoid systematic bias in a spin-state splitting calculation, recovery of plexes, and tight convergence and integration criteria (TightSCF, Grid5,
nearly 100% of the correlation energy using wavefunction methods is and GridX7 in Orca convention) were used throughout. The zero order
required [16]. For this level of accuracy, the method of choice is coupled regular approximation (ZORA)[42–44] was used to account for scalar
cluster with singles, doubles and perturbative triples excitations, CCSD relativistic effects. The all-electron ZORA-TZVP [45] basis set was used
(T), which is viewed as the “gold standard” of computational chemistry. on all atoms. The most important structural parameters for optimized
CCSD(T) most accurately reproduces spin-splitting energies and is structures of complexes C1-C8 are shown in Table S2.
therefore used to benchmark less expensive methods, but its application
is limited to small systems due to its high cost [17–19]. Multireference
2.2. DLPNO-CCSD(T) calculations
complete active space self-consistent field (CASSCF) calculations with
subsequent treatment of dynamical correlation with a second-order
The DLPNO-CCSD(T) correlation energy can be expressed as a sum of
perturbation method such as CASPT2[20] and NEVPT2[21,22] have
electron-pair correlation energy contributions, εij , where i and j are
been used with some success for Fe and Co SCO complexes and are
localized orbitals. The approximations made in the DLPNO approach is
considered applicable for systems that either have strong multi­
that only a portion of the electron pairs, ij, are treated with canonical
configurational character or are too large for CCSD(T) [13,23]. Even
CCSD and that only the most important PNOs are considered. The
though considerable progress has been achieved in both DFT and
determination of those pairs is based on correlation energies calculated
wavefunction theory methods, there is not yet a general-purpose pro­
with local second-order many-body perturbation theory (LMP2):
tocol for the prediction of SCO properties of transition metal complexes
[24]. EDLPNO− CCSD = ESP(CCSD) + EWP(LMP2) (2)
The domain-based pair natural orbital (DLPNO) approach to CCSD
(T) is a modern method that exploits the locality of electron correlation where ESP(CCSD) denotes the energy contribution from the pairs charac­
to yield a near linear scaling implementation of coupled cluster theory, terized as “strong”, which is calculated employing canonical CCSD, and
while retaining an accuracy close to the canonical CCSD(T) level EWP(LMP2) is the correlation contribution from the weak pairs, derived
[25–28]. Starting with localized orbitals for the occupied space and from LMP2. The correlation domain of each occupied orbital is a group
projected pair atomic orbitals (PAOs) for the virtual space, it uses fast, of atoms on which the orbital has a significant amplitude, which is set by
lower level calculations to screen electron pair correlation energies and the TCutMKN threshold. The extended domains of each pair ij consist of
pair natural orbital (PNO) occupation numbers, in order to select –based the union of the respective orbital domains and additionally includes all
on several cutoffs– a number of pairs that will be treated with canonical atoms of the domains of all orbitals k, which fulfil the condition that
CCSD(T), pairs that will be treated only at a lower level of theory, and none of the three pairs ij, ik and kj has been dropped from the calcula­
pairs that will be ignored from the correlation treatment. In a recent tion. In Orca three default sets of collective thresholds are available,
study Flöser et al. [16] showed that DLPNO-CCSD(T) can produce results which are derived from extensive benchmarking, namely “LoosePNO”,
comparable to canonical CCSD(T), provided that full iterative pertur­ “NormalPNO” and “TightPNO” settings, with “TightPNO” being the
bative triples (T1) are employed, TightPNO settings are applied and most accurate [28]. The most important individual thresholds are the
relativistic effects are treated. DLPNO-CCSD(T) has been employed to TCutPNO that controls the number of PNOs per electron pair, and the
investigate the SCO behaviour of several systems [16,29–36], but it has TCutPairs that controls the number of pairs that will be treated with ca­
not yet been used to study energetics of manganese SCO complexes. nonical CCSD (denoted as “strong pairs”), while the rest of the electron

2
M. Drosou et al. Polyhedron 208 (2021) 115399

Fig. 1. The manganese spin crossover complexes C1–C8 included in the present study.

pairs will be treated only with LMP2 (denoted as “weak pairs”) or √̅̅̅ √̅̅
ignored, which depends on lower level estimates of their correlation ea X
E(X)
corr − e
a Y (Y)
Ecorr
ECBS
HF =
√̅̅̅ √̅̅ (7)
energy. ea X − ea Y
As reference for the DLPNO-CCSD(T) calculations, both unrestricted
for the SCF energy and
Hartree-Fock (UHF) and Kohn-Sham (UKS) orbitals calculated with the
M06 [46], TPSS0, TPSSh [39], and BP86[47,48] functionals were used. X b E(X) b (Y)
corr − Y E corr
Both full iterative perturbative triples (T1)[27] and semi-canonical ECBS
corr = (8)
Xb − Y b
perturbative triples (T0)[26] corrections were employed. During
exploratory calculations we also used CASSCF starting orbitals for the correlation energy basis set extrapolation. In the above re­
(Table S12), and we observed that this leads to underestimation of lationships a = 7.88, b = 2.97 and X and Y are the corresponding suc­
perturbative triples (T) corrections contributions to the relative energies cessive cardinal numbers of the Mn basis set (X = 3 for TZ/TZ and Y = 4
of the two spin states. The default “NormalPNO” settings were used for QZ/TZ basis sets) [50].
except for calculations based on HF determinants, where “TightPNO” For the ligand CBS limit extrapolation, the TZ/TZ and TZ/DZ basis
settings were applied. Two-point extrapolation to the infinite PNO space sets were used, where TZ/DZ: ZORA-def2-TZVP on Mn and ZORA-def2-
was performed using the results of TCutPNO thresholds of 10-6 and SVP on all ligand atoms. Extrapolation was carried out according to Eqs.
3.33×10-7, based on the following equations: (7) and (8), with X = 2, Y = 3, a = 10.39 and b = 2.40.
The term δCBS
corr was defined as:
Ex = E∞ + A∙x− β
(3)
δCBS CBS (X)
corr = E corr − E corr (9)
where Ex is the correlation energy calculated with TCutPNO = 10-x, E∞ is
the extrapolated correlation energy at the PNO space limit and A and β and was used to estimate the CBS and PNO space extrapolated energies
are constants. A constant F can be defined according to the relationship: according to the formula:
yβ (10)
CBS CBS QZ/TZ
F= (4) E = ECBS ∞ ∞
HF + E SD + δSD + E (T1 ) + δ(T1 ) + E LMP2
yβ − xβ
This expression leads to an extrapolation equation analogous to the where E∞
SD and E(T1 ) are the SD and (T1) correlation energies extrapolated

two-point extrapolation scheme for the CBS limit:[49] to the complete PNO space limit. The LMP2 correlation energy compo­
nent was not extrapolated neither to the complete PNO space limit nor to
E∞ = yβ ∙Ey − xβ ∙Ex the CBS limit, because this value should vanish as the PNO space ex­
(5)
yβ − xβ pands and more pairs are characterized as “strong pairs” and thus
treated with canonical CCSD. Therefore, three DLPNO-CCSD(T) calcu­
so that Eq. (3) can be expressed as:
lations were carried out for each structure, as shown in detail in
E∞ = Ex + F∙(Ey − Ex ) (6) Tables S3-S8; i) basis set TZ/TZ and TCutPNO threshold 10-6, ii) basis set
TZ/TZ and TCutPNO threshold 3.33×10-7 and iii) basis set QZ/TZ and
After extensive benchmarking, the optimal value of F was suggested
TCutPNO threshold 3.33×10-7. Results from i) and ii) were used to
to be 1.5 [49], and we use this value in the present work.
perform the extrapolation with respect to the TCutPNO cutoff to the PNO
Extrapolation to the complete basis-set (CBS) limit with respect to
space limit at the TZ/TZ level and iii) was used to extract the δCBS
corr and
Mn was carried out using two basis set combinations, TZ/TZ: ZORA-
ELMP2 values.
QZ/TZ
def2-TZVP on all atoms except H, for which ZORA-def2-SVP was used,
and QZ/TZ: ZORA-def2-QZVPP on Mn and ZORA-def2-TZVP on all For the DLPNO-CCSD(T) calculations of complex C4, the t-butyl
ligand atoms except H (ZORA-def2-SVP). Extrapolation to the CBS limit moieties were replaced with hydrogen atoms, in order to reduce the size
with respect to Mn (CBS [3:4]/TZ) was accomplished separately for the of the system. This has negligible impact (ca. 0.2 kcal mol− 1) on
SCF energy and for the correlation energy, using the two-point extrap­ computed spin-splitting energies.
olation schemes:

3
M. Drosou et al. Polyhedron 208 (2021) 115399

2.3. CASSCF/NEVPT2 calculations (Figure 1, C3), synthesized by Liu et al in 2008 [79]. This complex un­
dergoes thermally induced incomplete SCO from the triplet ground state
For the CASSCF calculations, Kohn-Sham natural orbitals from to the quintet state with T1/2 250 K. In 2014 Shongwe et al. [80] re­
TPSSh/def2-TZVP calculations were used as reference orbitals. All ported the complex mer-[MnIII(3,5-tBuNPy)]+ (Figure 1, C4) with
CASSCF calculations were state-specific and the QZ/TZ basis set com­ T1/2 80 K. C4 is the first example of a hydraxone-based Mn-SCO complex
bination was used. The active space was constructed based on the and the only that does not have O donor atoms in trans- arrangement.
standard rules for transition metal complexes established by Pierloot [Mn(trp)] (Figure 1, C5), where trp3− =tris[1-(2-azolyl)-2-azabuten-
[51,52] and Roos et al. [53]. For all multireference calculations the 4-yl]amine, was the first Mn(III) SCO complex, discovered by Sim and
active space included five orbitals with 3d character and two ligand Sinn in 1981 [81]. It changes from the HS S = 2 to the IS state with S = 1
σ-bonding orbitals, to account for non-dynamic correlation effects upon cooling at 40–50 K. Contrary to Mn complexes with N4O2 ligands,
associated with covalent Mn-L interactions. The “extorbs doubleshell” C5 has a Jahn–Teller elongation axis in the HS (S = 2) state.
keyword was used in the CASSCF calculations to automatically detect The square-pyramidal, five-coordinate nitrosyl species [Mn(tmtaa)
the double shell 3d’ Mn orbitals and rotate them to be the first virtual (NO)] (C6), where tmtaa = dibenzotetramethyltetraaza-14-annulene,
orbitals. The double shell orbitals were included in the subsequent has a LS (S = 0) ground state at temperatures < 200 K [82] and upon
CASSCF/NEVPT2 calculations. We also performed CASSCF calculations heating it transits either to the IS (S = 1) state or to the HS (S = 2) state, a
including the double shell orbitals in the active space and subsequent problem that is still a matter of debate [14]. Boguslawski et al. [83]
NEVPT2 calculations to address the effect of orbital optimization with studied the performance of DFT on reproducing the spin densities of iron
larger active spaces. Both strongly contracted SC-NEVPT2[21,22,54] nitrosyl complexes. They concluded that nitrosyl complexes are a
and fully internally contracted FIC-NEVPT2[55,56] calculations were particularly challenging case for DFT, because the spin density distri­
carried out. butions are sensitive to the chosen approximate exchange–correlation
functional. In C6, the unpaired electron from the neutral NO species and
the five d electrons of Mn(II) can be distributed over the combinations of
2.4. DFT calculations
the Mn 3d and NO π*orbitals, which means that the valence of Mn is
ambiguous; there are two main electronic configurations, which corre­
Single point DFT calculations were performed on the TPSSh opti­
spond to Mn(II) − NO0 and Mn(III) − NO− . Despite the multireference
mized structures using a selection of fourteen popular functionals: the
character of nitrosyl complexes, CCSD(T) calculations have been used
GGA functionals PBE [57,58], BLYP [47,59], BP86 [47,48], OLYP
successfully to describe iron nitrosyl complexes with porphyrin ligands
[59,60] and OPBE [57,60], the meta-GGA TPSS [61] and M06L [46,62],
[18,19]. During initial exploratory calculations, we calculated the
the hybrid GGA functionals B3LYP [47,59,63], B3LYP*[64] (B3LYP
singlet and quintet states relative energy using DLPNO-CCSD(T) to be
functional with 15% exact exchange, EEX) and PBE0 [65,66], the hybrid
26 kcal mol− 1 in favor of the singlet state, while TPSSh also gave a
meta-GGA TPSSh [39], TPSS0 (TPSSh functional with 25% EEX), M06
difference of 24 kcal mol− 1, close to 22 kcal mol− 1 reported by Ama­
[46], and the double-hybrid functional B2PLYP [67]. The ZORA-def2-
bilino and Deeth [14]. Those values are too large, even considering
TZVP basis set was employed for all atoms. In the cases of GGA and
potential sources of error in the computational methods, so the S = 0 to
meta-GGA functionals, the RI approximation was employed, while with
S = 2 SCO thermally induced transition is unlikely. Therefore, we
the hybrid functionals the COSX approximation was used also. Calcu­
considered only the singlet and triplet spin states for our computational
lations with the double-hybrid B2PLYP functional employed the RI-MP2
studies.
approximation.
Some Mn(II) d5 manganocene derivatives of general formula
We define the adiabatic relative energies of the different spin states
[MnII(R-Cp)2] are SCO-active at low temperatures (Figure S1) [84–87].
as:
For the simplest manganocene [MnCp2] (C7 in Fig. 1), a thermal SCO
ΔEHL = EHS − ELS (11) equilibrium between the doublet (LS) and sextet (HS) states was
confirmed by magnetic resonance studies in solution [85]. Andersen
therefore positive values of ΔEHL mean that the low spin form is more et al. [88] presented a systematic experimental investigation of the ef­
stable than the high spin form. fect of bulky substituents on the magnetic properties of manganocenes.
They observed that electron donating moieties promote the LS state,
3. Results and discussion which means that they activate SCO behaviour and that, on the contrary,
bulky substituents favor the HS state. A recent computational study of
3.1. Mn SCO complexes the SCO behaviour of manganocenes using the OPBE functional showed
that addition of an electron donor substituent on the Cp raises the energy
The most in depth investigated class of Mn(III) SCO complexes are of the frontier orbitals, thus stabilizing the LS state and increasing the
octahedral Mn(III) complexes bearing N4O2 hexadentate ligands. In the T1/2, while the steric congestion induced by a bulky moiety has the
HS (S = 2) state these complexes have a distorted compressed octahedral opposite effect, which means that one has to consider both electronic
geometry with a Jahn-Teller compression axis along the O-Mn-O bonds and steric effects to control the SCO temperature in such systems [15].
and zero field splitting parameter D > 0. The first example of this type of The [MnII(tBu-Cp)2] [15,88] (C8) shows larger Mn-Cp distances
complex was discovered by Morgan et al. in 2006 [68], who synthesized (Table S2), but the presence of the electron donating tBu group leads to a
the ligand 3-MeO-Sal-323, derived from two 3-methoxysalicylaldehyde higher transition temperature (Table S1).
units and a tetraamine moiety and pointed out that thermal SCO In the following we use a uniform notation, ΔEHL defined in Eq. (11),
behaviour may be induced when the phenolate oxygen donors are trans to refer to the spin-states that participate in the SCO process for each
to each other, which can be achieved by adjusting the length of the alkyl complex, thus for C1–C5 ΔEHL refers to the adiabatic spin-splitting en­
chains in the starting tetraamine. Since this discovery, a series of ergy between the S = 2 and S = 1 states, for C6 ΔEHL describes the S = 1
[MnIII(R-Sal-323)]+ Schiff base complexes [11,68–78] (for R = H see C1, to S = 0 transition and for C7 and C8 it describes the S = 5/2 to S = 1/2
for R = Cl see C2 in Fig. 1) with various substituents (-R) on the phenyl transition.
groups of the Sal-323 ligand have been found to exhibit thermal SCO
(Figure S1).
Another class of Mn(III) SCO complexes contain tridentate N2O 3.2. Spin state energetics from DLPNO-CCSD(T)
donor ligands. The first system of this type was fac-trans-
[MnIII(Py2O)2]+, where Py2O = 2-(2-Pyridylmethyleneamino) pyridine The results of the DLPNO-CCSD(T1) calculations of the HS and LS-

4
M. Drosou et al. Polyhedron 208 (2021) 115399

state splitting energies, ΔEHL , of complexes C1-C8, extrapolated to the orbitals of different functionals, we looked at the mean and max
complete PNO space limit and to the CBS[3:4]/TZ limit, based on numbers of PNOs per pair calculated for each reference determinant,
different reference determinants are shown in Fig. 2 (see also which are plotted in Figure S2. When the calculation starts from HF
Table S11). Clearly, HF-based calculations overstabilize the HS states, determinants, the mean number of PNOs per pair is smaller than for KS
except from C6, for which DLPNO-CCSD(T) results are almost inde­ determinants. In addition, the differences between the various KS de­
pendent of the reference orbitals. KS orbitals form a correlated single- terminants may stem from differences in the maximum number of PNOs
determinantal description of the system and provide an inexpensive per pair, which indicates why the BP86 orbitals provide the best refer­
alternative to the HF reference. CCSD(T) calculations based on KS or­ ence. GGA exchange functionals overestimate non-dynamic correlation
bitals are reported to give improved results particularly for open shell effects due to the self-interaction error (SIE), while hybrid functionals
systems, and lower T1 diagnostic [89,90], because they minimize the cover part of the SIE in order to balance these effects. Orbitals obtained
contribution of single excitations in the CCSD expansion [91–94]. under the influence of electron correlation are more diffuse [97]. We
Expectedly, the same has been shown for DLPNO-CCSD(T) [36]. Here thus tentatively attribute the strong reference orbital dependence of the
we must note that in a converged canonical CCSD(T) calculation the DLPNO-CCSD(T) to differences in the shape of the reference orbitals
quality of the reference orbitals, in principle, has a rather negligible produced by different density functionals. In the case of DLPNO coupled
effect on the computed total energy. Indeed, studies by Radon et al. cluster, the shape of the starting orbitals plays a crucial role during the
[95,96] report weak correlation between the choice of reference orbitals initial stages of domain construction. The DLPNO-CCSD(T) calculations
and observed CCSD(T) errors. However, when approximations are presented in the rest of this work are based on BP86 starting orbitals.
introduced, the minimization of excitation amplitudes induced by using
correlated KS reference orbitals can be an advantage.
The deviations among DLPNO-CCSD(T1) ΔEHL values with reference 3.3. Approaching the basis set limit
KS orbitals obtained with different DFT functionals lie in a range of 4
kcal mol− 1, which is an indicator of the degree of convergence of the Attempting to reach the basis set limit of the DLPNO-CCSD(T) en­
method. The calculations based on reference orbitals derived from the ergies, calculations at triple and quadruple zeta levels were conducted
GGA functional BP86 seem to give the most accurate results, because together with TZ/TZ and QZ/TZ extrapolation to the CBS[3:4]/TZ limit.
they correctly predict the ground state (LS) for all studied systems. In Since it has been reported that in CCSD(T) calculations the quality of the
order to evaluate the accuracy of the DLPNO-CCSD(T1) approach with basis set on the metal influences the spin splitting energies to a much
respect to canonical CCSD(T), we performed CCSD(T) calculations for larger extent than the basis set size on the ligands [19], complete basis
C7, the only one among the studied complexes for which these calcu­ set extrapolation was carried out only with respect to Mn. The results are
lations are feasible. The CCSD(T) ΔEHL value after extrapolation to the shown in Fig. 3. It can be seen that as the basis set of the Mn center is
CBS[3:4]/TZ limit is 9.60 kcal mol− 1, only 2.12 kcal mol− 1 higher than expanded, the LS form becomes more favourable for all complexes. The
the DLPNO-CCSD(T1) with BP86 orbitals (Fig. 2). Therefore, the DLPNO- same behaviour is observed both for the DLPNO-CCSD (red bars) and
CCSD(T1) with BP86 orbitals provides the best agreement with canoni­ after taking into account the (T1) contributions {DLPNO-CCSD(T1),
cal CCSD(T), using the same basis set scheme, among the DLPNO-CCSD green bars}. To investigate whether the results were converged with
(T1) protocols with different starting orbitals. Notably, in a previous respect to the ligand basis sets, additional calculations were carried out
study of DLPNO-CCSD(T) spin-splitting energetics of Co(III) complexes for all complexes using the def2-SVP basis sets on all ligand atoms and
calculations based on BP86 orbitals outperformed calculations using HF def2-TZVP basis set on Mn. The results of the TZ/DZ and TZ/TZ
orbitals on the same level of theory [31]. In an attempt to rationalize the extrapolation to the CBS limit (Figure S3) show insignificant differences
observed differences between the results obtained with the KS reference in the computed ΔEHL values, i.e. relative energies obtained with TZ/TZ
basis set are very close (within 1 kcal mol− 1) to the TZ/CBS values,

Fig. 2. DLPNO-CCSD(T1) derived ΔEHL for complexes C1–C8 calculated with different reference orbitals. The values are extrapolated to the complete PNO space
limit and to the CBS[3:4]/TZ limit. Calculations for complex C4 could not be converged when starting with HF orbitals and are not reported.

5
M. Drosou et al. Polyhedron 208 (2021) 115399

Fig. 3. ΔEHL values calculated using DLPNO-CCSD (red) and DLPNO-CCSD(T1) (green) with TZ/TZ and QZ/TZ basis sets and extrapolation of the SCF and correlation
energies to the complete basis set limit with respect to Mn (CBS[3:4]/TZ), for complexes C1–C8. ((Colour online.))

which shows that basis set size extrapolation with respect to Mn basis set for complexes C1-C8 is presented in Fig. 4. In general, significant de­
is adequate to achieve CBS convergence of the coupled cluster method. viations of (T0) with respect to (T1) are observed, especially for open
The LS geometry has typically shorter metal–ligand bond lengths shell systems and systems with low-lying electronic states, but explor­
than the HS geometry, which leads to greater overlap between localized atory calculations are recommended to test the difference between the
orbitals in the LS state. Consequently, the LS structures have a larger two approaches, due to the increased cost of (T1) [16]. The (T0) semi­
number of mean number of PNOs per pair and of maximum number of canonical approximation neglects off-diagonal contributions from the
PNOs per pair than the HS state. In addition, LS states have more paired internal Fock matrix, in order to avoid the potentially expensive itera­
electrons than high-spin states. Given that the intra-orbital electron-pair tive process, whereas (T1) follows the full-iterative solution, but takes
repulsion energy is much larger than the interorbital interaction, into account only the most important subset of triple excitations.
retrieval of the full correlation energy is more important for the lower [26,27] For complexes C1–C5, the semicanonical triples recover only
spin states. Indeed, as we will describe in the following, increasing the 94% of the full-iterative perturbative triples correlation energy, which
accuracy of the method usually favors the LS state. projects to deviations of ca. 4 kcal mol− 1 in the spin-state splitting
attributed to the triples contribution. Notably, the percentage of (T0)
correlation energy recovered relative to (T1) is consistently ~ 1% lower
3.4. Perturbative triples excitations contributions for LS states, as has been also observed in Fe(II) complexes [16]. This
means that (T1) favours the LS state in all systems studied herein. For
A comparison of the semicanonical (T0) and full-iterative (T1) per­ complex C6, (T0) recovers only 92% of (T1) and its contribution to the
turbative triples excitation contributions to the computed ΔEHL values HS–LS energy difference is ~ 2 kcal mol− 1. The largest contribution of
(T1) to the ΔEHL is observed for C7 and C8, where (T0) recovers 96% of
(T1) correlation energy for the HS state and only 91% for the LS state
which leads to an HS-LS energy difference of 8 kcal mol− 1.

3.5. PNO thresholds

The influence of the thresholds used to determine the degree of


approximation of the DLPNO method to the canonical CCSD(T) result on
the spin-state splitting energies was also investigated. In the calculations
presented in this work we use NormalPNO settings (NPNO) with TCutPNO
thresholds of 10-6 and 3.33 × 10-7 (default value for NPNO) and carry
out extrapolation of the CCSD correlation energy of the strong pairs and
of the (T1) correlation energy at the complete PNO space limit, using
Eqs. (3)–(6). For complexes C1, C6 and C8 additional calculations were
Fig. 4. Contributions of semicanonical (T0) (blue) and full-iterative (T1) performed applying TightPNO settings (TPNO). For this investigation
(green) perturbative triples excitations to ΔEHL in DLPNO-CCSD(T) calculations M06 reference orbitals were used. The default TCutPNO threshold for
for complexes C1–C8 using the TZ/TZ basis sets. ((Colour online.)) TPNO settings is 10-7, but we performed the PNO space extrapolation

6
M. Drosou et al. Polyhedron 208 (2021) 115399

using TCutPNO thresholds of 10-6 and 3.33 × 10-7. In Fig. 5 (see also Table 1
Table S13), the correlation energies of both the HS and LS forms of C1, Computed adiabatic spin splitting energies ΔEHL (kcal mol− 1) at the FIC-
C6 and C8 and the corresponding calculated ΔEHL values are plotted for NEVPT2 level with the QZ/TZ basis sets with different choices of the active
NPNO (yellow) and TPNO (purple) settings for different values of the space.
exponent x of TCutPNO = 10− x, as well as after PNO space extrapolation at Complex (8,7) (8,12)a (10,13) frozen 3s3p (10,13)
both NPNO (orange) and TPNO (dark purple). Along with tightening of C1 3.75 2.99 3.10 3.77
the TCutPNO threshold, the absolute amount of correlation energy C2 3.82 3.16 5.99 5.88
recovered is increased, which affects the spin-splitting energies. As we C3 − 2.69 − 4.04 − 7.46 − 6.62
observe in complex C8, where calculations with TCutPNO 10-7 were (8,7) (8,12)a (8,12) frozen 3s3p (8,12)
C4 2.29 − 0.06 − 1.35 − 0.88
feasible, PNO space size extrapolated NPNO correlation energies (or­ C5 2.58 0.13 − 0.90 − 0.20
ange horizontal lines) approximate TPNO energies with TCutPNO 10-7. (14,11) (14,15)a b b

Altun et al. [49] found that two point extrapolation of the TCutPNO = C6 45.50 45.99
10− 5 and TCutPNO = 10− 6 calculations provides an accuracy between that (9,7) (9,12)a (9,12) frozen 3s3p (9,12)
C7 9.08 − 14.47 23.06 25.15
of TCutPNO = 10− 6 and of TCutPNO = 10− 7, while the TCutPNO = 10− 6 and
C8 10.07 − 14.02 20.08 22.15
TCutPNO = 10− 7 extrapolation provides results that are very close to those
a
obtained using TCutPNO = 10− 8, when TPNO settings are used. As for the Double shell Mn d’ orbitals were extracted using the “extorbs doubleshell”
spin-state splitting energies, after the PNO space extrapolations the feature and used without further orbital optimization. b CASSCF calculations
were not converged with (14,15) active space and were not feasible with larger
differences between the NPNO and TPNO derived ΔEHL values are from
active space.
1 to 1.5 kcal mol− 1. Even though this is not a negligible difference, those
findings indicate that NPNO settings can be used as a cost effective
alternative to TPNO settings, considering the higher cost of TPNO cal­ d orbitals and two bonding Mn-L orbitals. In C1–C5 the dz2 and dx2 − y2
culations in relatively large transition metal complexes with highly orbitals are antibonding combinations to the Mn-L σ-bonding orbitals
conjugated ligands. Contrary to the observations above for the effects of and in C7 and C8 the dxz and dyz orbitals have antibonding character
the reference orbitals, the method used for the perturbative triple ex­ with respect to the Mn–π interaction. For the nitrosyl complex C6, an
citations and the Mn CBS extrapolation, when we compare NPNO and active space composed of 14 electrons in 11 orbitals, i.e. five Mn 3d
TPNO derived ΔEHL values, the LS states are not consistently favoured. orbitals, two σ-bonding orbitals with respect to the Mn-ligand bonds and
This is attributed to the LMP2 contribution of the “weak” pairs to the the four NO π and π* orbitals, according to the observations reported by
correlation energy, which sometimes leads to a slight overshooting of Boguslawski et al. [83]. A strong covalent interaction is evident between
the total correlation energy, leading to overstabilization of the LS state. the Mn dxz , dyz , and NO π and π* orbitals, which results in two bonding
{dxz , 1πNO * } and {dyz , 2πNO * } orbitals, two antibonding {dxz , 1π∗NO }∗ and
3.6. CASSCF and NEVPT2 calculations {dyz , 2π∗NO }∗ orbitals. The NEVPT2 calculations that correspond to the
minimal active space sizes predict the correct spin state and SCO
Table 1 shows the ΔEHL values for complexes C1–C8 computed with behaviour for all complexes except from C3 and C6.
FIC-NEVPT2 using the QZ/TZ basis sets with various active space Increasing the active space to include the Mn 3d’ double shell or­
compositions. The minimal active spaces for the octahedral complexes bitals consistently favours the HS state. In those calculations, the
C1–C5 and for the manganocenes C7 and C8 consist of the five Mn extended active spaces for C4 and C5 were (8,12) and for C7 and C8

Fig. 5. Comparison of correlation energies and ΔEHL values calculated with different TCutPNO cutoffs for complexes C1, C6 and C8 using NormalPNO (NPNO, yellow)
and TightPNO (TPNO, purple) settings, using the TZ/TZ basis sets. ((Colour online.))

7
M. Drosou et al. Polyhedron 208 (2021) 115399

(9,12). For complexes C1, C2, and C3 the Mn-O π-bonding orbital was use of the coupled cluster approach, and it is possible that improved
also included, resulting in a (10,13) active space. The active orbitals for description of the zeroth-order wave function, such as including addi­
complex C1 are shown in Fig. 6 and for C2–C8 in Figures S4–S10. For tional π* orbitals in the active space, might be needed to achieve better
complex C6 only four double shell 3d’ Mn orbitals were made active for accuracy, which would rapidly increase the computational cost.
the subsequent NEVPT2 calculations; The Mn dx2 − y2 double shell orbital
was excluded due to computational limits and because additional ligand 3.7. Evaluation of density functional theory
orbitals would be needed to be made active [83]. After CASSCF orbital
optimization of the extended active space the correct ground spin state is The adiabatic spin-state splittings of Mn SCO complexes C1–C8 were
predicted by NEVPT2 only for C1, C2, C7 and C8, while for the man­ computed using fourteen of the most popular density functionals that
ganocenes it overshoots the stability of the LS state. For complex C6 it is cover the full range of “Jacob’s ladder” [103], i.e. simple GGAs, meta-
clear that a larger active space is needed, possibly including additional GGAs, hybrids, meta-GGA hybrids, up to the double-hybrid B2PLYP
ligand π* orbitals, but this cannot be achieved using regular CASSCF and functional, in order to assess their performance. Inspection of Fig. 7 (see
additional approximations must be employed, such as the density matrix also Table S15 for complete results) reveals that the magnitude of the
renormalization group (DMRG) approach [98–101]. The CASSCF/FIC- energy difference between HS and LS forms depends strongly on the
NEVPT2 calculated ΔEHL values for the largest active spaces studied functional with a large scattering of ΔEHL values of more than 30 kcal
are presented in Fig. 7. mol− 1. The double-hybrid B2PLYP functional disfavours the LS states by
In general, the effect of increasing the size of the active space is not about 6 kcal mol− 1 for complexes C1–C5, while for C6 it gives a very
systematic, in line with results reported in previous studies [96,102]. large ΔEHL value of 47 kcal mol− 1 (not shown in Fig. 7) and for C7 and
This lack of systematic stabilization of the high- or low-spin state over C8 it disfavours the LS state by 4 kcal mol− 1. Notably, the quintet state of
the other contrasts with both DLPNO-CCSD(T), where the relative en­ complex C6 is calculated 65 kcal mol− 1 higher than the singlet ground
ergies converge with increasing the level of theory, and even with DFT, state with the B2PLYP functional, therefore this large inconsistency is
where this equilibrium depends on the exact exchange percentage, as not a result of inversion of the S = 1 and S = 2 state stabilities. Hybrid
will be shown below. The effect of the frozen core approximation, where functionals typically show a consistent overstabilization of the HS form
core electrons are not included in the perturbation treatment, was also with increasing the exact exchange percentage. This failure is attributed
investigated. In almost all cases the frozen core approximation leads to a to the fact that Fermi correlation is included in the exact (Hartree–Fock)
small error (<2 kcal mol− 1) in favor of the LS state. The respective SC- exchange, while Coulomb correlation is not [64]. This behaviour is
NEVPT2 results are shown in Table S14. The strongly contracted (SC-) reproduced here, with a variation of ~ 20 kcal mol− 1 found between
and fully internally contracted (FIC-, also called partially contracted) TPSSh, which has 10% EEX, and M06, with 27% EEX. The TPSSh
NEVPT2 variants have been reported [56] to perform almost equiva­ functional shows the best performance for complexes C1-C6, with
lently on the calculation of vertical excitation energies, with the former, respect to the DLPNO-CCSD(T1) derived relative energies. Several
perhaps counterintuitively, giving results systematically closer to studies have suggested that exact exchange admixture of ~ 15% is
experiment. Overall, compared to DLPNO-CCSD(T), multireference optimal for the calculation of spin state splitting energies [104], even
wavefunction theory methods studied in this work have lower compu­ though it cannot be expected to apply to all compound types [105]. In
tational cost, but they exhibit non-systematic behaviour, lack the ease of the case of Mn-based complexes however, the TPSSh functional was
reported to best reproduce spin-state energetics [12,14] and exchange
coupling parameters [40]. Here, B3LYP* (15% EEX) also works well, but
disfavours the LS states by 2–5 kcal mol− 1. The manganocenes C7 and
C8 show different behaviour, since OPBE better reproduces their ΔEHL
values, while the TPSSh functional overstabilizes the LS states. The
functionals PBE, BLYP, BP86 and TPSS yield excessively large ΔEHL
values for C7 and C8 (greater than 20 kcal mol− 1, see Table S15). In
contrast, the successful description of the SCO behaviour of man­
ganocenes by the OPBE functional, compared to the TPSSh, has been
reported previously by Cirera and Ruiz [15]. However, it can be
observed that non-hybrid functionals generally produce worse results.
BLYP and PBE have been found to disfavour HS states in iron complexes,
whereas OLYP and OPBE perform better [106]. The same pattern is
observed in the present study. The over-stabilization of the LS states
with respect to the HS states by pure GGA functionals has been attrib­
uted the overestimation of non-dynamic correlation arising from the
mixing of metal d orbitals and M− L bonding orbitals [107,108]. In
summary, our results confirm once more the general inconsistency of
DFT functionals, in that all functionals overstabilize either the HS or the
LS forms.

4. Conclusions

Rational design of SCO complexes requires accurate prediction of


spin state energetics with a fast and reliable method that allows efficient
screening of large databases of transition metal complexes. In the pre­
sent study we employed the DLPNO-CCSD(T) method to extract highly
accurate estimates of spin state energetics of experimentally character­
ized mononuclear Mn(III) and Mn(II) spin crossover complexes. The
results of the DLPNO-CCSD(T) calculations were extrapolated to the
complete basis set limit with respect to the metal center and to the
Fig. 6. Contour plots of CASSCF natural orbitals for complex C1. infinite PNO space limit. The impact of different reference orbitals on

8
M. Drosou et al. Polyhedron 208 (2021) 115399

Fig. 7. ΔEHL values for C1-C8 computed using DLPNO-CCSD(T1), FIC-NEVPT2 and various density functional approximations. For compounds that fall outside the
scale of the graph for specific functionals, see Tables 1 and S15.

the energetics was investigated. It appears that Kohn–Sham reference significant multireference character and provides systematically spin-
orbitals, particularly from the BP86 functional, are superior to Har­ splitting energies consistent with the SCO behavior of the studied
tree–Fock orbitals, in agreement with previous studies on the CCSD(T) complexes. After initial screening, the method has capacity for general
method, and additionally because they lead to an expanded PNO space improvement by modification of the most important cutoffs, i.e. TCutPNO
compared to a Hartree–Fock reference. Consideration of triples excita­ and TCutPairs, and by assessing the validity of various approximations.
tions is essential for the present problem. In terms of the triples ap­ Overall, for fast initial pre-screening of Mn complexes with potential
proximations in the coupled cluster expansion, even though (T1) is more SCO behaviour we may recommend the use of the TPSSh or B3LYP*
expensive than (T0), in the present systems we report large ΔEHL dif­ functionals (and possibly OPBE or OLYP for manganocenes), and the
ferences from 2 up to 8 kcal mol− 1 attributed to the perturbative triples DLPNO-CCSD(T) methodology presented in this work as a reliable
contributions to ΔEHL computed using the two methods, which is method to deliver quantitative results, as well as to accurately describe
indicative of a problem where (T1) becomes mandatory. On the con­ potential energy surfaces of the corresponding spin-states. Dispersion
trary, applying NormalPNO settings instead of the more expensive corrections and entropy should also be considered in order to estimate
TightPNO settings, leads to an error in the order of 1.5 kcal mol− 1, the corresponding transition temperature of such systems, according to
provided that extrapolation with respect to the PNO space is performed. Eq. (1). These contributions are usually small and can be obtained at the
This leads to the conclusion that NormalPNO settings combined with DFT level without introducing considerable errors. A combination of
PNO extrapolation can be used as a reliable protocol, which is very DLPNO-CCSD(T) derived electronic energy differences between the spin
important for extensively conjugated systems where the orbitals do­ states involved in the SCO process, ΔEHL , along with dispersion cor­
mains can become very large. CASSCF/NEVPT2 calculations did not rections and thermochemical parameters obtained using lower-level
predict the correct ground states for all studied complexes and did not methods, is proposed to provide an efficient computational protocol
show systematic behaviour or convergence with respect to the size of the with the maximum currently attainable accuracy for large transition
active space. Therefore, although they can have advantages in specific metal based SCO systems.
settings, they cannot be considered in general superior to single-
reference correlated approaches for the description of SCO systems, CRediT authorship contribution statement
while their use requires significant user intervention and case-by-case
optimization. At the DFT level, despite the expected non-universality Maria Drosou: Methodology, Investigation, Data curation, Formal
problem of DFT with respect to spin-state energetics, the hybrid func­ analysis, Writing - original draft. Christiana A. Mitsopoulou: Super­
tionals TPSSh and B3LYP* show good performance, with TPSSh giving vision, Writing - review & editing. Dimitrios A. Pantazis: Conceptu­
ΔEHL values closest to the DLPNO-CCSD(T1) results for all complexes alization, Methodology, Writing - review & editing.
except the manganocenes C7 and C8, for which OPBE and OLYP give the
most accurate values with respect to ab initio estimates. Declaration of Competing Interest
Therefore, we can conclude that the DLPNO-CCSD(T) approach
represents an efficient method that is applicable even to systems with The authors declare that they have no known competing financial

9
M. Drosou et al. Polyhedron 208 (2021) 115399

interests or personal relationships that could have appeared to influence [39] V.N. Staroverov, G.E. Scuseria, J. Tao, J.P. Perdew, J. Chem. Phys. 119 (2003)
12129–12137.
the work reported in this paper.
[40] M. Orio, D.A. Pantazis, T. Petrenko, F. Neese, Inorg. Chem. 48 (2009) 7251–7260.
[41] F. Neese, F. Wennmohs, A. Hansen, U. Becker, Chem. Phys. 356 (2009) 98–109.
Acknowledgements [42] E. van Lenthe, E.J. Baerends, J.G. Snijders, J. Chem. Phys. 99 (1993) 4597–4610.
[43] E. van Lenthe, E.J. Baerends, J.G. Snijders, J. Chem. Phys. 101 (1994)
9783–9792.
DAP acknowledges support by the Max Planck Society. MD ac­ [44] C. van Wüllen, J. Chem. Phys. 109 (1998) 392–399.
knowledges support by the Hellenic Foundation for Research and [45] D.A. Pantazis, X.-Y. Chen, C.R. Landis, F. Neese, J. Chem. Theory Comput. 4
(2008) 908–919.
Innovation (HFRI) under the HFRI PhD Fellowship grant (Fellowship
[46] Y. Zhao, D.G. Truhlar, Theor. Chem. Account 120 (2008) 215–241.
Number: 16199). [47] A.D. Becke, Phys. Rev. A 38 (1988) 3098–3100.
[48] J.P. Perdew, Phys. Rev. B 33 (1986) 8822–8824.
[49] A. Altun, F. Neese, G. Bistoni, J. Chem. Theory Comput. 16 (2020) 6142–6149.
Appendix A. Supplementary data [50] F. Neese, E.F. Valeev, J. Chem. Theory Comput. 7 (2011) 33–43.
[51] T.R. Cundari (Ed.), Computational Organometallic Chemistry, Marcel Dekker,
Supplementary data to this article can be found online at https://doi. New York, 2001.
[52] K. Pierloot, Mol. Phys. 101 (2003) 2083–2094.
org/10.1016/j.poly.2021.115399.
[53] V. Veryazov, P.Å. Malmqvist, B.O. Roos, Int. J. Quantum Chem. 111 (2011)
3329–3338.
References [54] C. Angeli, R. Cimiraglia, J.-P. Malrieu, Chem. Phys. Lett. 350 (2001) 297–305.
[55] R.W.A. Havenith, P.R. Taylor, C. Angeli, R. Cimiraglia, K. Ruud, The J. Chem.
Phys. 120 (2004) 4619–4625.
[1] J.-F. Létard, P. Guionneau, L. Goux-Capes, in Spin Crossover in Transition Metal
[56] I. Schapiro, K. Sivalingam, F. Neese, J. Chem. Theory Comput. 9 (2013)
Compounds III, Springer-Verlag, Berlin/Heidelberg, 2004, pp. 221–249.
3567–3580.
[2] R.N. Muller, L. Vander Elst, S. Laurent, J. Am. Chem. Soc. 125 (2003) 8405–8407.
[57] J.P. Perdew, K. Burke, M. Ernzerhof, Phys. Rev. Lett. 77 (1996) 3865–3868.
[3] G. Molnár, S. Rat, L. Salmon, W. Nicolazzi, A. Bousseksou, Adv. Mater. 30 (2018)
[58] J.P. Perdew, K. Burke, M. Ernzerhof, Phys. Rev. Lett. 1998, 80, 891–891.
1703862.
[59] C. Lee, W. Yang, R.G. Parr, Phys. Rev. B 37 (1988) 785–789.
[4] P. Gütlich, H. A. Goodwin, in Spin Crossover in Transition Metal Compounds I
[60] N.C. Handy, A.J. Cohen, Mol. Phys. 99 (2001) 403–412.
(Eds.: P. Gütlich, H.A. Goodwin), Springer Berlin Heidelberg, Berlin, Heidelberg,
[61] J. Tao, J.P. Perdew, V.N. Staroverov, G.E. Scuseria, Phys. Rev. Lett. 91 (2003),
2004, pp. 1–47.
146401.
[5] M.A. Halcrow, Polyhedron 26 (2007) 3523–3576.
[62] Y. Zhao, D.G. Truhlar, J. Phys. Chem. 125 (2006), 194101.
[6] P.J. Koningsbruggen, Y. Maeda, H. Oshio, in Spin Crossover in Transition Metal
[63] A.D. Becke, J. Chem. Phys. 98 (1993) 1372–1377.
Compounds I (Eds.: P. Gütlich, H.A. Goodwin), Springer Berlin Heidelberg,
[64] M. Reiher, O. Salomon, B. Artur Hess, Theor. Chem. Acc. 107 (2001) 48–55.
Berlin, Heidelberg, 2004, pp. 259–324.
[65] J.P. Perdew, M. Ernzerhof, K. Burke, J. Chem. Phys. 105 (1996) 9982–9985.
[7] D.J. Harding, P. Harding, W. Phonsri, Coord. Chem. Rev. 313 (2016) 38–61.
[66] C. Adamo, V. Barone, J. Chem. Phys. 110 (1999) 6158–6170.
[8] H.A. Goodwin, in Spin Crossover in Transition Metal Compounds II, Springer
[67] S. Grimme, J. Chem. Phys. 124 (2006), 034108.
Berlin Heidelberg, Berlin, Heidelberg, 2004, pp. 23–47.
[68] G.G. Morgan, K.D. Murnaghan, H. Müller-Bunz, V. McKee, C.J. Harding, Angew.
[9] Y. Garcia, P. Gütlich, in Spin Crossover in Transition Metal Compounds II,
Chem. Int. Ed. 45 (2006) 7192–7195.
Springer Berlin Heidelberg, Berlin, Heidelberg, 2004, pp. 49–62.
[69] C. Gandolfi, T. Cotting, P.N. Martinho, O. Sereda, A. Neels, G.G. Morgan,
[10] J. Olguín, Coord. Chem. Rev. 407 (2020), 213148.
M. Albrecht, Dalton Trans. 2011 (1855) 40.
[11] S. Wang, Y.-H. Li, W. Huang, Eur. J. Inorg. Chem. 2015 (2015) 2237–2244.
[70] K. Pandurangan, B. Gildea, C. Murray, C.J. Harding, H. Müller-Bunz, G.
[12] J. Cirera, M. Via-Nadal, E. Ruiz, Inorg. Chem. 57 (2018) 14097–14105.
G. Morgan, Chem. Eur. J. 18 (2012) 2021–2029.
[13] J. Cirera, E. Ruiz, Comments Mod. Chem. A Comments Inorg. Chem. 39 (2019)
[71] P.N. Martinho, B. Gildea, M.M. Harris, T. Lemma, A.D. Naik, H. Müller-Bunz, T.
216–241.
E. Keyes, Y. Garcia, G.G. Morgan, Angew. Chem. 124 (2012) 12765–12769.
[14] S. Amabilino, R.J. Deeth, Inorg. Chem. 56 (2017) 2602–2613.
[72] B. Gildea, M.M. Harris, L.C. Gavin, C.A. Murray, Y. Ortin, H. Müller-Bunz, C.
[15] J. Cirera, E. Ruiz, Inorg. Chem. 57 (2018) 702–709.
J. Harding, Y. Lan, A.K. Powell, G.G. Morgan, Inorg. Chem. 53 (2014)
[16] B.M. Flöser, Y. Guo, C. Riplinger, F. Tuczek, F. Neese, J. Chem. Theory Comput.
6022–6033.
16 (2020) 2224–2235.
[73] A.J. Fitzpatrick, E. Trzop, H. Müller-Bunz, M.M. Dîrtu, Y. Garcia, E. Collet, G.
[17] L.M. Lawson Daku, F. Aquilante, T.W. Robinson, A. Hauser, J. Chem. Theory
G. Morgan, Chem. Commun. 51 (2015) 17540–17543.
Comput. 8 (2012) 4216–4231.
[74] S. Wang, Y.-J. Li, F.-F. Ju, W.-T. Xu, K. Kagesawa, Y.-H. Li, M. Yamashita,
[18] R. Lo, D. Manna, R. Zbořil, D. Nachtigallová, P. Hobza, J. Phys. Chem. C 123
W. Huang, Dalton Trans. 46 (2017) 11063–11077.
(2019) 23186–23194.
[75] A.V. Kazakova, A.V. Tiunova, D.V. Korchagin, G.V. Shilov, E.B. Yagubskii, V.
[19] J. Oláh, J.N. Harvey, J. Phys. Chem. A 113 (2009) 7338–7345.
N. Zverev, S.C. Yang, J. Lin, J. Lee, O.V. Maximova, A.N. Vasiliev, Chem. Eur. J.
[20] K. Andersson, P. Malmqvist, B.O. Roos, J. Chem. Phys. 96 (1992) 1218–1226.
25 (2019) 10204–10213.
[21] C. Angeli, R. Cimiraglia, S. Evangelisti, T. Leininger, J.-P. Malrieu, J. Chem. Phys.
[76] I.A. Kühne, A. Barker, F. Zhang, P. Stamenov, O. O’Doherty, H. Müller-Bunz,
114 (2001) 10252–10264.
M. Stein, B.J. Rodriguez, G.G. Morgan, J. Phys. Condens. Matter 32 (2020),
[22] C. Angeli, R. Cimiraglia, J.-P. Malrieu, J. Chem. Phys. 117 (2002) 9138–9153.
404002.
[23] C. Sousa, C. de Graaf, in Spin States in Biochemistry and Inorganic Chemistry
[77] V.B. Jakobsen, E. Trzop, L.C. Gavin, E. Dobbelaar, S. Chikara, X. Ding, K. Esien,
(Eds.: M. Swart, M. Costas), John Wiley & Sons, Ltd, Oxford, UK, 2015, pp. 35–57.
H. Müller-Bunz, S. Felton, V.S. Zapf, E. Collet, M.A. Carpenter, G.G. Morgan,
[24] M. Swart, in New Directions in the Modeling of Organometallic Reactions (Eds.:
Angew. Chem. 132 (2020) 13407–13414.
A. Lledós, G. Ujaque), Springer International Publishing, Cham, 2020, pp.
[78] S. Ghosh, S. Bagchi, S. Kamilya, A. Mondal, Dalton Trans. 50 (2021) 4634–4642.
191–226.
[79] Z. Liu, S. Liang, X. Di, J. Zhang, Inorg. Chem. Commun. 11 (2008) 783–786.
[25] C. Riplinger, F. Neese, J. Chem. Phys. 138 (2013), 034106.
[80] M.S. Shongwe, K.S. Al-Barhi, M. Mikuriya, H. Adams, M.J. Morris, E. Bill, K.
[26] C. Riplinger, B. Sandhoefer, A. Hansen, F. Neese, J. Chem. Phys. 139 (2013),
C. Molloy, Chem. Eur. J. 20 (2014) 9693–9701.
134101.
[81] P.G. Sim, E. Sinn, J. Am. Chem. Soc. 103 (1981) 241–243.
[27] Y. Guo, C. Riplinger, U. Becker, D.G. Liakos, Y. Minenkov, L. Cavallo, F. Neese,
[82] F. Franceschi, J. Hesschenbrouck, E. Solari, C. Floriani, N. Re, C. Rizzoli,
J. Chem. Phys. 148 (2018), 011101.
A. Chiesi-Villa, J. Chem. Soc. Dalton Trans. (2000) 593–604.
[28] D.G. Liakos, M. Sparta, M.K. Kesharwani, J.M.L. Martin, F. Neese, J. Chem.
[83] K. Boguslawski, C.R. Jacob, M. Reiher, J. Chem. Theory Comput. 7 (2011)
Theory Comput. 11 (2015) 1525–1539.
2740–2752.
[29] A. Altun, M. Saitow, F. Neese, G. Bistoni, J. Chem. Theory Comput. 15 (2019)
[84] M.E. Switzer, R. Wang, M.F. Rettig, A.H. Maki, J. Am. Chem. Soc. 96 (1974)
1616–1632.
7669–7674.
[30] R. Ghafarian Shirazi, F. Neese, D.A. Pantazis, G. Bistoni, J. Phys. Chem. A 123
[85] F.H. Koehler, B. Schlesinger, Inorg. Chem. 31 (1992) 2853–2859.
(2019) 5081–5090.
[86] M.L. Hays, D.J. Burkey, J.S. Overby, T.P. Hanusa, S.P. Sellers, G.T. Yee, V.
[31] S.E. Neale, D.A. Pantazis, S.A. Macgregor, Dalton Trans. 49 (2020) 6478–6487.
G. Young, Organometallics 17 (1998) 5521–5527.
[32] A. Antalík, D. Nachtigallová, R. Lo, M. Matoušek, J. Lang, Ö. Legeza, J. Pittner,
[87] S. Scheuermayer, F. Tuna, M. Bodensteiner, M. Scheer, R.A. Layfield, Chem.
P. Hobza, L. Veis, Phys. Chem. Chem. Phys. 22 (2020) 17033–17037.
Commun. 48 (2012) 8087.
[33] R.G. Shirazi, D.A. Pantazis, F. Neese, Mol. Phys. 118 (2020), e1764644.
[88] M.D. Walter, C.D. Sofield, C.H. Booth, R.A. Andersen, Organometallics 28 (2009)
[34] D. Manna, R. Lo, P. Hobza, Dalton Trans. 49 (2020) 164–170.
2005–2019.
[35] P. Comba, D. Faltermeier, S. Krieg, B. Martin, G. Rajaraman, Dalton Trans. 49
[89] T.J. Lee, P.R. Taylor, Int. J. Quantum Chem. 36 (2009) 199–207.
(2020) 2888–2894.
[90] W. Jiang, N.J. DeYonker, A.K. Wilson, J. Chem. Theory Comput. 8 (2012)
[36] M. Feldt, Q.M. Phung, K. Pierloot, R.A. Mata, J.N. Harvey, J. Chem. Theory
460–468.
Comput. 15 (2019) 922–937.
[91] G.J.O. Beran, S.R. Gwaltney, M. Head-Gordon, Phys. Chem. Chem. Phys. 5 (2003)
[37] F. Neese, F. Wennmohs, U. Becker, C. Riplinger, J. Chem. Phys. 152 (2020),
2488.
224108.
[92] J.N. Harvey, M. Aschi, Faraday Disc. 124 (2003) 129.
[38] C.R. Groom, I.J. Bruno, M.P. Lightfoot, S.C. Ward, Acta Crystallogr. B Struct. Sci.
[93] J.N. Harvey, D.P. Tew, Int. J. Mass Spectrom. 354–355 (2013) 263–270.
Cryst. Eng. Mater. 72 (2016) 171–179.

10
M. Drosou et al. Polyhedron 208 (2021) 115399

[94] L.W. Bertels, J. Lee, M. Head-Gordon, J. Chem. Theory Comput. 17 (2021) [102] K. Pierloot, Q.M. Phung, A. Domingo, J. Chem. Theory Comput. 13 (2017)
742–755. 537–553.
[95] G. Drabik, J. Szklarzewicz, M. Radoń, Phys. Chem. Chem. Phys. 23 (2021) [103] J.P. Perdew, in AIP Conference Proceedings, AIP, Antwerp (Belgium), 2001, pp.
151–172. 1–20.
[96] M. Radoń, Phys. Chem. Chem. Phys. 21 (2019) 4854–4870. [104] J.N. Harvey, in Principles and Applications of Density Functional Theory in
[97] D. Cremer, Mol. Phys. 99 (2001) 1899–1940. Inorganic Chemistry I, Springer Berlin Heidelberg, Berlin, Heidelberg, 2004, pp.
[98] R. Olivares-Amaya, W. Hu, N. Nakatani, S. Sharma, J. Yang, G.-K.-L. Chan, 151–184.
J. Chem. Phys. 142 (2015), 034102. [105] R. Poli, J.N. Harvey, Chem. Soc. Rev. 32 (2003) 1–8.
[99] C.J. Stein, D.A. Pantazis, V. Krewald, J. Phys. Chem. Lett. 10 (2019) 6762–6770. [106] M. Swart, A.R. Groenhof, A.W. Ehlers, K. Lammertsma, J. Phys. Chem. A 108
[100] A. Baiardi, M. Reiher, J. Chem. Phys. 152 (2020), 040903. (2004) 5479–5483.
[101] L. Freitag, M. Reiher, in Quantum Chemistry and Dynamics of Excited States [107] B. Pinter, A. Chankisjijev, P. Geerlings, J.N. Harvey, F. De Proft, Chem. Eur. J. 24
(Eds.: L. González, R. Lindh), Wiley, 2020, pp. 205–245. (2018) 5281–5292.
[108] M. Radoń, Phys. Chem. Chem. Phys. 16 (2014) 14479–14488.

11

You might also like