Anda di halaman 1dari 10

Accurate thermochemistry for transition metal complexes from first-

principles calculations
Nathan J. DeYonker, T. Gavin Williams, Adam E. Imel, Thomas R. Cundari, and Angela K. Wilson

Citation: J. Chem. Phys. 131, 024106 (2009); doi: 10.1063/1.3160667


View online: http://dx.doi.org/10.1063/1.3160667
View Table of Contents: http://jcp.aip.org/resource/1/JCPSA6/v131/i2
Published by the American Institute of Physics.

Additional information on J. Chem. Phys.


Journal Homepage: http://jcp.aip.org/
Journal Information: http://jcp.aip.org/about/about_the_journal
Top downloads: http://jcp.aip.org/features/most_downloaded
Information for Authors: http://jcp.aip.org/authors

Downloaded 18 May 2012 to 136.159.235.223. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/about/rights_and_permissions
THE JOURNAL OF CHEMICAL PHYSICS 131, 024106 共2009兲

Accurate thermochemistry for transition metal complexes


from first-principles calculations
Nathan J. DeYonker,a兲 T. Gavin Williams, Adam E. Imel, Thomas R. Cundari, and
Angela K. Wilson
Department of Chemistry, Center for Advanced Scientific Computing and Modeling (CASCaM),
University of North Texas, Denton, Texas 76203-5070, USA
共Received 16 April 2009; accepted 9 June 2009; published online 8 July 2009兲

The “correlation consistent Composite Approach” or ccCA is an ab initio model chemistry based on
the single reference MP2 level of theory. By adjusting the basis set and level of theory of the core
valence additive correction, ccCA is capable of reliable thermochemical predictions of inorganic and
organometallic transition metal-containing molecules, as well as achieving chemical accuracy on
main group species, with a mean absolute deviation of 0.89 kcal mol−1 against the 147 enthalpies
of formation in the G2/97 test set. For a set of 52 complexes containing elements Sc–Zn, ranging in
size from diatomics to Ni共PF3兲4 and Fe共C5H2兲2, ccCA on average predicts enthalpies of formation
to within ⫾3 kcal mol−1 of the experimental result with a mean absolute deviation of
2.85 kcal mol−1 and a root mean square deviation of 3.77 kcal mol−1. The ccCA methodology is a
significant step toward quantitative theoretical modeling of transition metal thermodynamics.
© 2009 American Institute of Physics. 关DOI: 10.1063/1.3160667兴

I. INTRODUCTION be to develop first principles computational techniques to


calculate their energetic content reliably, accurately, and ef-
The energetic content of a material or the energetic bal-
ficiently. Herein, we describe a methodology that provides a
ance of a chemical process is arguably the most important
physical factor in determining its suitability. For example, substantial improvement in accuracy over previous MP2-
explosives must detonate with sufficient energy to achieve based techniques and in efficiency over coupled cluster-
their intended purpose but must be stable enough to be safely based techniques, and achieves the aforementioned goals for
handled. Catalysts must be active enough to allow reactants large, chemically relevant transition metal species without
to swiftly traverse reaction coordinates to produce the many resorting to ad hoc parameters.
products needed by our society—fuels, drugs, commodity A host of theoretical hurdles have thwarted efforts at
and bulk chemicals, etc.—but not so high in energy that they using single reference ab initio methods for quantitative tran-
react in unintended ways 共i.e., display low selectivity兲. sition metal computational chemistry. To name a few, these
Societal progress has long been marked by the mastery include strongly interacting valence electrons, increased im-
of metals. Among these, the d-block 共or transition兲 metals portance of core-core and core-valence electron correlation, a
occupy a pivotal position in numerous technical fields. These diversity of low-lying excited electronic states, and relativis-
include current technologies that form the foundation of tic effects. The evolution of density functional theory 共DFT兲,
modern society 共e.g., the fixation of dinitrogen by iron cata- electron pseudopotentials, and multireference techniques to
lysts to make ammonia for fertilizers兲1,2 and necessary new treat heavy element chemistry has resulted in many success-
technologies 共e.g., photosplitting of water to produce ful and synergistic avenues between experimental and com-
hydrogen3 or materials4 to more efficiently convert sunlight
putational chemistry. However, the aforementioned problems
to electricity兲. The past decades have seen a revolution in
with transition metals, exacerbated by the nonsystematic na-
how scientific research is conducted, whereby modeling and
simulation now complement the traditional pursuits of theory ture of treating electron correlation via DFT5–8 and multiref-
and experiment. Despite rapid advances, modeling and simu- erence methods9–13 mean that quantitative transition metal
lation of transition metal containing entities—at all scales 共TM兲 thermochemistry remains a lofty goal.
from molecular to device level—lag behind comparable There has been some success using single reference
methods for nonmetal species, in large part due to the ability coupled cluster theory to study inorganic chemical problems.
of transition metals to stabilize many disparate chemical and Nielsen and co-workers14–16 examined the enthalpies of for-
physical environments. It is, however, the same aptitude that mation 共⌬H f 兲 for chromium compounds. Dixon and
has made transition metals an important ally in technological co-authors17–20 used a coupled cluster-based model chemis-
advancement. To the extent that modeling and simulation try to examine binding and electronic excitation energies of
will play an important role in identifying new materials and some transition metal oxides. Hyla-Kryspin and Grimme21
processes for technological advances, an essential goal must used spin-component scaled MP2 and MP3 methods to ob-
tain metal-carbonyl bond dissociation energies 共BDEs兲. Gen-
a兲
Electronic mail: ndeyonk@unt.edu. erally, however, the quality, quantity, and reliability of re-

0021-9606/2009/131共2兲/024106/9/$25.00 131, 024106-1 © 2009 American Institute of Physics

Downloaded 18 May 2012 to 136.159.235.223. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/about/rights_and_permissions
024106-2 DeYonker et al. J. Chem. Phys. 131, 024106 共2009兲

search involving high-accuracy calculation of transition While the newest hybrid metaexchange DFT functionals de-
metal thermodynamic properties greatly lag behind that of veloped by Truhlar and co-worker39 have a mean absolute
their main group counterparts. deviation 共MAD兲 in calculated versus experimental energet-
Ab initio methods should ideally be calibrated against a ics of 5.6– 5.7 kcal mol−1 for 48 different metal-bonding
set of well-established experimental energetic quantities situations, these functionals may contain more than a dozen
and/or the most sophisticated theoretical results that are tech- optimized parameters.
nologically possible. While there are enormous test sets for While developing ccCA, we sought a HLC-free ap-
main group/organic energetic quantities,22–24 the same cannot proach with the target of a uniform ab initio recipe for com-
be said for inorganic molecules. In part, appropriate compari- puting accurate and efficient energies across the entire peri-
son of transition metal-containing molecules with experi- odic table without utilization of large basis set coupled
mental values contained in popular thermochemical compen- cluster computations. In earlier work, DeYonker et al.40 used
dia is problematic because the uncertainties 共when these are ccCA to compute enthalpies of formation for 18 small
even reported兲 in transition metal thermochemical quantities transition-metal-containing molecules and compared them to
are often significantly larger than those for main group spe- an accurate coupled-cluster based model chemistry 关CBS-
cies, certainly outside of the usual definition of “chemical CCSD共T兲兴. Agreement between ccCA and CBS-CCSD共T兲
accuracy” 共i.e., ⫾1 kcal/mol兲, and in extreme cases literature was satisfactory. Over a subset of eight enthalpies of forma-
values 关cf. the ⌬H f共g兲 of titanocene dichloride, Ti共C5H5兲2Cl2, tion for which the latter was computationally feasible, the
listed in the NIST webbook兴25 can differ by over ccCA MAD was 3.4 versus 3.1 kcal mol−1 for CBS-
70 kcal mol−1. The largest sets of transition metal thermody- CCSD共T兲. Our laboratories sought a target “transition metal
namic data available so far are from DFT studies by Furche chemical accuracy” of ⫾3.0 kcal mol−1, this choice reflect-
and Perdew,7 Rinaldo et al.,26 and Riley and Merz.6 Except ing the aforementioned increased uncertainties vis-à-vis TM
for the Riley and Merz study, most comparisons of experi- thermochemistry.
mental and theoretical data have focused on DFT perfor- However, as the “ccCA-TM” method was further tested
mance on TM dissociation energies. The Riley and Merz on larger inorganic complexes, serious problems emerged.
study contains the largest number of enthalpies of formation For example, the ccCA-predicted enthalpy of formation for
benchmarked so far. However, all 12 of the density function- Ni共CO兲4 was in error by 30– 100 kcal mol−1 when using
als investigated, including the functional that performed the various ccCA schemes. For the CBS-CCSD共T兲 model chem-
best 共TPSS1KCIS兲, showed inconsistent performance from istry, the error was only 5.5 kcal mol−1 from the experimen-
the early to middle to late elements in the 3d block.6 tal value of −144⫾ 2.5 kcal mol−1. Upon further investiga-
Another major impediment for quantitative transition tion, it was discovered that the MP2-based additive
metal thermochemistry has been the lack of high-quality ba- correction for core-valence and core-core electron correlation
sis sets. While basis sets have been created for elements was quickly deteriorating in quality with increasing molecu-
Sc–Zn,27–30 the TM basis sets that would allow systematic lar size. By improving the treatment of correlation for the
convergence of thermochemical properties with respect to core-valence ccCA step from MP2 to MP3 or CCSD, vastly
increasing basis set size simply did not exist until recently, improved results were found for carbonyl complexes and
when Peterson and Balabanov31,32 expanded the suite of cor- larger molecules such as ferrocene. However, the computa-
relation consistent basis sets to include elements Sc–Zn. tion of core-valence electron interactions became the bottle-
A focus of the research in our laboratories has been the neck of the ccCA algorithm and greatly increased the com-
development of the correlation consistent Composite Ap- putational requirements. Because the ccCA methodology is
proach 共ccCA兲, an efficient MP2-based model chemistry that not based on semiempirical parameters, it is possible to in-
uses correlation consistent basis sets. Tested against a variety vestigate changes to the level of theory and basis set used in
of main group energetic properties, ccCA has been shown to the core-valence term. In this manuscript, we propose a
be comparable to the most accurate formulations of the MP4- modification of ccCA based on improved physics rather than
based G3 methods 共G3X兲 and improved accuracy versus the ad hoc parameters that retains chemical accuracy for main
MP2-based G4 variant 关G4共MP2兲兴.33 Where MP2-based group molecules, as well as predicts accurate enthalpies of
model chemistries and even some coupled cluster-based formation for smaller inorganic radicals and larger inorganic
model chemistries have failed, ccCA has been able to be and organometallic species.
used to reliably compute energies and thermodynamic prop-
erties, viz., alkali and alkaline earth metal oxides and II. METHODS
hydroxides,34,35 hypervalent inorganics,36 sulfur-containing
molecules with substantial multireference character,37 and There are five major technical modifications to the new
heavy p-block molecules.38 ccCA-TM model chemistry versus that previously
It is worth noting that the most recent formulation of reported.33,36,40 First, equilibrium geometries and harmonic
ccCA contains no empirical parameters. Even the harmonic vibrational frequencies are computed using the B3LYP/cc-
vibrational scale factor in ccCA, used to account for anhar- pVTZ level of theory. Second, for single point energies, all
monic vibrational effects, is calibrated based on theoretical computations utilize the Douglas–Kroll basis sets and em-
zero-point energies 共ZPEs兲. Comparatively, the G3 methods ploy the spin-free, one-electron Douglas–Kroll–Hess
have a “high level correction” or HLC utilizing four opti- Hamiltonian.41,42 Third, the basis set level for the additive
mized energy corrections, and G4 methods contain six. correction accounting for core valence correlation effects is

Downloaded 18 May 2012 to 136.159.235.223. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/about/rights_and_permissions
024106-3 Thermochemistry for inorganic complexes J. Chem. Phys. 131, 024106 共2009兲

TABLE I. Formulation of the previously recommended ccCA for main group atoms vs the more “pan-periodic” formulation used in this study. For atoms
Na–Ar, cc-pV共X + d兲Z basis sets are always used with ccCA.

ccCAa ccCA-TM

Optimization B3LYP cc-pVTZ B3LYP cc-pVTZ


Harmonic frequencies B3LYP cc-pVTZ scaled by 0.989 B3LYP cc-pVTZ scaled by 0.989
HF CBS HF aug-cc-pVTZ HF aug-cc-pVTZ-DK
HF aug-cc-pVQZ HF aug-cc-pVQZ-DK
HF CBS fit E共n兲 = EHF-CBS + B exp共−1.63n兲 E共n兲 = EHF-CBS + B exp共−1.63n兲
MP2 CBS MP2 aug-cc-pVDZ MP2 aug-cc-pVDZ-DK
MP2 aug-cc-pVTZ MP2 aug-cc-pVTZ-DK
MP2 aug-cc-pVQZ MP2 aug-cc-pVQZ-DK
MP2 CBS fit “P ” = E共x兲 = ECBS + B exp关−共x − 1兲兴 + C exp关−共x − 1兲2兴 “P ” = E共x兲 = ECBS + B exp关−共x − 1兲兴 + C exp关−共x − 1兲2兴
B
“S3 ” = E共lmax兲 = ECBS +
共lmax兲3
“PS3 ” = 1 / 2ⴱ共“P ” + “ S3”兲
⌬共CC兲 CCSD共T兲 cc-pVTZ CCSD共T兲 cc-pVTZ-DK
⫺ MP2 cc-pVTZ ⫺MP2 cc-pVTZ-DK
⌬共CV兲 MP2共FC1兲 aug-cc-pCVTZ CCSD共T,FC1兲 aug-cc-pCVDZ-DK
⫺ MP2 aug-cc-pVTZ ⫺CCSD共T兲 aug-cc-pCVDZ-DK
⌬共DK兲 MP2 cc-pVTZ-DK Folded into all energy calculations
⫺ MP2 cc-pVTZ
⌬共SO兲 Experimental atomic values Experimental atomic values
FOCI Stuttgart ECP for molecules
a
From Ref. 33.

computed using basis sets of double-zeta quality, but the computed using Stuttgart “small-core” effective core poten-
wave function is computed using CCSD共T兲. Fourth, a small tials 共ECP10MDF兲 for TM elements43 and cc-pVTZ basis
basis set CCSD共T兲 computation is performed to avoid double sets for main group elements.44 Table I shows the additive
counting the extra basis polarization in the core-valence ad- corrections of the new ccCA formulation compared to that in
ditive correction. Overall, the core valence additive correc- Ref. 33. All computations employing an unrestricted Hartree-
tion is Fock 共UHF兲 reference were performed using GAUSSIAN 03,45
while all computations employing a restricted open shell
⌬E共CV兲 = E关CCSD共T,FC1兲/aug-cc-pCVDZ-DK兴
Hartree-Fock 共ROHF兲 reference wave function were per-
− E关CCSD共T,valence兲/aug-cc-pCVDZ-DK兴. formed using MOLPRO 2006.1.46
In the usual notation, “FC1” indicates that all electrons of
III. RESULTS
first-row atoms are correlated, all electrons of atoms Na–Ar
are correlated except the 1s orbitals, and all electrons of To assess the modified ccCA method against a set of
atoms K–Kr are correlated except the 1s2s2p orbitals, i.e., transition metal-containing species, it is first necessary to
the inner noble gas core molecular orbitals remain frozen. reexamine the performance for main group species. A pri-
Fifth, molecular spin-orbit coupling energy corrections are mary ethos of ccCA is to develop a chemically accurate com-

TABLE II. MSDs and MADs 共in kcal mol−1兲 of ccCA against the G2/97 enthalpies of formation.

UHF ccCA-PS3 UHF ccCA-TM ROHF ccCA-TM

MSDs
Enthalpies of formation 共147兲 0.20 0.08 ⫺0.24
Nonhydrogens 共34兲 0.20 0.27 ⫺0.27
Hydrocarbons 共22兲 ⫺0.25 ⫺0.77 ⫺1.16
Substituted hydrocarbons 共47兲 0.52 0.31 ⫺0.17
Inorganic hydrides 共15兲 0.99 1.12 0.95
Radicals 共29兲 ⫺0.38 ⫺0.40 ⫺0.24

MADs
Enthalpies of formation 共147兲 0.81 0.91 0.89
Nonhydrogens 共34兲 0.86 0.80 0.92
Hydrocarbons 共22兲 0.75 1.03 1.22
Substituted hydrocarbons 共47兲 0.83 0.92 0.80
Inorganic hydrides 共15兲 0.99 1.12 0.95
Radicals 共29兲 0.68 0.84 0.72

Downloaded 18 May 2012 to 136.159.235.223. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/about/rights_and_permissions
024106-4 DeYonker et al. J. Chem. Phys. 131, 024106 共2009兲

TABLE III. Deviations of ten enthalpies of formation from initial ccCA transition metal benchmark set using ccCA and the CBS-CCSD共T兲 model chemistry
described in Ref. 40. All units are in kcal mol−1.

Old UHF ROHF Old UHF ROHF-TM


CBS-CCSD共T兲 ccCA-S4 共Q5兲 ccCA-TM ccCA-TM CBS-CCSD共T兲 ccCA-S4 共Q5兲 ccCA-TM ccCA
Expt. ⌬H f ⌬H f ⌬H f ⌬H f deviation deviation deviation deviation

TiF2 −164.5⫾ 10 ⫺131.0 ⫺138.0 ⫺130.5 ⫺132.7 ⫺33.5 ⫺26.5 ⫺34.0 ⫺31.8
TiF3 −284⫾ 10 ⫺259.7 ⫺267.1 ⫺260.4 ⫺265.6 ⫺24.3 ⫺16.9 ⫺23.6 ⫺18.4
VO 31.8⫾ 2 32.3 27.7 38.0 32.8 ⫺0.5 4.1 ⫺6.2 ⫺1.0
VO2 −41.6⫾ 3.3 ⫺38.4 ⫺44.6 ⫺30.3 ⫺39.8 ⫺3.2 3.0 ⫺11.3 ⫺1.8
CrO3 −77.3⫾ 1.0 ⫺63.3 ⫺82.0 ⫺69.3 ⫺69.3 ⫺14.0 4.7 ⫺8.0 ⫺8.0
FeCl2 −31.7⫾ 0.2 ⫺32.1 ⫺35.2 ⫺22.1 ⫺32.5 0.4 3.5 ⫺9.6 0.8
CoCl3 −39.1⫾ 2.5 ⫺35.7 ⫺35.0 ⫺38.3 ⫺34.6 ⫺3.4 ⫺4.1 ⫺0.8 ⫺4.5
NiCl2 −17.4⫾ 1.6 ⫺14.6 ⫺18.4 ⫺14.1 ⫺14.1 ⫺2.8 1.0 ⫺3.3 ⫺3.3
CuF −3.1⫾ 2.0 2.3 ⫺3.0 0.2 0.3 ⫺5.4 ⫺0.1 ⫺3.3 ⫺3.4
CuF2 −63.8⫾ 3.0 ⫺64.7 ⫺59.8 ⫺61.9 ⫺67.3 0.9 ⫺4.1 ⫺1.9 3.5
Ni共PF3兲4 −953.4⫾ 2.4 ⫺1034.4 ⫺957.7 ⫺957.0 81.0 4.3 3.6
Cr共CO兲6 −240⫾ 1.1 ⫺269.7 ⫺238.7 ⫺237.8 29.7 ⫺1.3 ⫺2.2
Mn共CO兲5Cl −219.5⫾ 3.1 ⫺253.8 ⫺218.3 ⫺217.4 34.3 ⫺1.2 ⫺2.1
Fe共CO兲5 −173.0⫾ 1.6 ⫺221.7 ⫺179.6 ⫺179.1 48.7 6.6 6.1
Ni共CO兲4 −144⫾ 2.5 ⫺136.7a ⫺189.6 ⫺143.4 143.3 ⫺7.4 45.6 ⫺0.6 ⫺0.7
Fe共C5H5兲2 57.9⫾ 0.6 30.2 55.0 55.7 27.7 2.9 2.2

MSD all 14.48 ⫺5.71 ⫺3.81


MSD w/o
larger
species 3.56 ⫺8.59 ⫺3.54 ⫺10.20 ⫺6.79
MSD w/o
TiFx 1.95 ⫺3.50 1.00 ⫺5.55 ⫺2.21
MSD w/o
TiFx and
CrO3 2.09 ⫺2.01 0.47 ⫺5.20 ⫺1.38
Standard deviation
w/o TiFx and
CrO3 2.31 3.42 3.98 2.78

MAD all 20.94 7.43 5.84


MAD w/o
larger
species 8.84 6.79 10.20 7.65
MAD w/o
TiFx 3.82 3.06 5.55 3.29
MAD w/o TiFx and
CrO3 2.37 2.83 5.20 2.62
a
For the CBS-CCSD共T兲 computations on Ni共CO兲4, cc-pVxZ basis sets were used in the CBS extrapolations instead of aug-cc-pVxZ sets.

posite ab initio method that is pan periodic, whereby one In our initial study of TM thermochemistry,40 ten enthal-
formulation results in acceptable accuracy for molecules pies of formation were compared using both ccCA and a
across the entire periodic table. Table II shows the original CBS-CCSD共T兲 based model chemistry. In the previous work,
recommended formulation of ccCA 共denoted as ccCA-PS3兲 the enthalpies of formation of five carbonyl-containing com-
compared to the new formula using both unrestricted and plexes and ferrocene were computed using ccCA-S4 共Q5兲.
restricted Hartree–Fock wave functions for open-shell atoms The average error for these six larger molecules was
and molecules 共UHF ccCA-TM and ROHF ccCA-TM, re- +44.5 kcal mol−1. For the smaller molecules, Table III
spectively兲 and their performance against the 147 main shows the deviations of the ten small TM-containing mol-
group enthalpies of formation contained in the G2/97 train- ecules from our initial investigation using the “old”
ing set.47 While there is a slight decrease in overall accuracy, ccCA-TM method 关共ccCA-S4 共Q5兲兴 and the new ccCA-TM
the new ccCA-TM methods show little systematic bias and method, again the latter with both UHF and ROHF reference
the performance is still within the boundaries of main group wave functions. Six out of ten of the CBS-CCSD共T兲 values
chemical accuracy. The ROHF-ccCA-TM method has a are within two times the experimentally quoted uncertainties.
mean signed deviation 共MSD兲 of −0.24 kcal mol−1 and a Previously, the enthalpies of formation for TiF2, TiF3, and
MAD of 0.89 kcal mol−1 compared to a MSD of CrO3 were recognized as being significant outliers and theo-
0.20 kcal mol−1 and MAD of 0.81 kcal mol−1 for the previ- retical results cast significant doubt on the quality of the
ous formulation 共UHF-based ccCA-PS3兲. literature values.40 In the case of CuF, although the value

Downloaded 18 May 2012 to 136.159.235.223. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/about/rights_and_permissions
024106-5 Thermochemistry for inorganic complexes J. Chem. Phys. 131, 024106 共2009兲

TABLE IV. Listing of molecules in the ccCA-TM test set along with ground electronic states used to compute
energies.

Ground state Ground state Ground state

ScO 2
⌺+ MnCl 7
⌺+ NiCl 2

ScF 1
⌺+ MnCl2 6
⌺+ NiCl2 3
⌺−g
ScF3 1
A⬘1 MnF ⌺
7 +
NiF2 ⌺g
3 −

ScCl3 1
A1 MnS ⌺
6 +
Ni共CO兲4 1
A1
Mn共CO兲5Cl 1
A1 Ni共PF3兲4 1
A1
TiO 3
⌬ Mn共CO兲5 2
A1
1
TiO2 A1
TiF2 3
⌬g FeCl 6
⌬ CuO 2

TiF3 2
A⬘1 FeCl2 5
⌬g CuF 1
⌺+
TiCl2 3
⌬g FeCl3 6
A⬘1 CuF2 2
兺g
TiF4 1
A1 FeCO ⌺
3 −
CuCl ⌺
1 +

Fe共CO兲2 兺g
3 −
Cu2 ⌺
1 +

VO 4
⌺− Fe共CO兲3 3
⌬2 CuH ⌺
1 +
2 3
VO2 A1 Fe共CO兲4 B1
VF5 1
A⬘1 Fe共CO兲5 1
A⬘1 Zn共CH3兲2 1
A1
VCl 5
⌬ Fe共C5H5兲2 1
A1g Zn共CH2CH3兲2 1
Ag
Fe共C5H5兲 1
A⬘ Zn共CH2CH2CH3兲2 1
A
CrO3 1
A1 ZnCl ⌺
2 +

CrCl 6
⌺+ CoCl3 5
A1 ZnCl2 兺g
1 +

Cr共CO兲3 1
A1 CoCl2 3
⌬g ZnH ⌺
2 +

Cr共CO兲4 1
A1 CoF2 4
⌬g
1 1
Cr共CO兲5 A1 Co共CO兲4H A1
1 2
Cr共CO兲6 A1g Co共CO兲4 A1

from the Yungman compendium48 has smaller error bars, In light of these positive results, enthalpies of formation
both the CBS-CCSD共T兲 and ccCA results are within the error were computed using UHF and ROHF-based modified ccCA
bars of the JANAF value 共1.1⫾ 3 kcal mol−1兲. After remov- methods for a larger set of 58 transition metal-containing
ing TiF2, TiF3, and CrO3 from this test set the MSDs of the molecules. This training set includes larger carbonyl-
CBS-CCSD共T兲 and ROHF ccCA are ⫺2.01 and containing species such as Ni共CO兲4, Mn共CO兲5Cl, and
−1.38 kcal mol−1, respectively. The MSD of the old ccCA Cr共CO兲6, for which the previous version of ccCA for transi-
enthalpies for this subset has the smallest magnitude versus tion metals gave poor result deviations from experiment on
the new formulations, but deviations in the original ccCA the order of several dozen kcal mol−1. Molecules containing
method were usually of opposite sign to the CBS-CCSD共T兲 elements from Sc–Zn are represented, with the majority of
values. The agreement of ROHF ccCA-TM and CBS- molecules in the test set being halogenated species and car-
CCSD共T兲 MSDs suggests a more favorable comparison. The bonyl species, due to the relative amount of reliable thermo-
MADs of CBS-CCSD共T兲 and ROHF ccCA-TM are 2.37 and chemical data in existence for these ligand families. Table IV
2.62 kcal mol−1, and these two methods are more reliable lists the molecules contained in our overall test set along
than the previous formulation of ccCA for transition metals. with the electronic ground states used in our computations.
Because of the test set used in the initial ccCA-TM paper, it Table V lists the MSDs and MADs for the test set. Experi-
was difficult to make a clear distinction between the perfor- mental values for the enthalpies of formation used in our
mance of ccCA-S4 共Q5兲 and the newer UHF- and ROHF- study, along with atomic ccCA energies, deviations, and
based ccCA-TM model chemistry for smaller radical inor- atomic/molecular spin-orbit corrections are in the supple-
ganic species. However, the large positive deviations of the mentary material.49
larger classical inorganic complexes are evidence of system- As an example, Fig. 1 shows the effect of ccCA additive
atic error, which is eliminated by incorporating a more so- corrections upon Ni共CO兲4 for the previous ccCA implemen-
phisticated treatment of core-valence electron correlation. tation from Ref. 40 关ccCA-S4 共Q5兲兴 compared to the current

TABLE V. MSDs and MADs 共in kcal mol−1兲 of various ccCA-TM test sets.

No. of Average of MSD MAD MSD MAD


Test set molecules uncertainties UHF UHF ROHF ROHF

Overall 共set I兲 58 2.67 ⫺3.89 5.99 ⫺2.50 4.78


w/o TiFx and CrO3 共set II兲 55 2.43 ⫺2.91 5.13 ⫺1.58 3.98
w/o CrCOx FeC5H5 共set III兲 51 2.17 ⫺2.69 4.40 ⫺1.24 3.20
w/JANAF CrO3 and Ricca Fe共CO兲x 共set IV兲 52 2.22 ⫺2.58 4.17 ⫺1.15 2.85

Downloaded 18 May 2012 to 136.159.235.223. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/about/rights_and_permissions
024106-6 DeYonker et al. J. Chem. Phys. 131, 024106 共2009兲

FIG. 2. Scatter plot of experimental vs theoretical ROHF ccCA-TM enthal-


pies of formation for the 52 molecules in set IV 共in kcal mol−1兲.

FIG. 1. Contributions of additive corrections to the enthalpy of formation of TiF3, and CrO3 from our first ccCA-TM study. The ROHF
Ni共CO兲4 共in kcal mol−1兲 using methodology from Ref. 40 关ccCA-S4共Q5兲兴 vs ccCA-TM deviations from reported experimental values for
new ccCA-TM with ROHF reference. TiF2 and TiF3 are ⫺31.8 and −18.4 kcal mol−1, respectively.
These two enthalpies both depend on an enthalpy of subli-
implementation 共ROHF ccCA-TM兲. A complete basis set mation for TiF3 determined experimentally by Zmbov and
treatment at the MP2 level of theory is not sufficient for Margrave.50 On the other hand, the experimental value for
accurate thermochemistry, as the MP2 CBS ⌬H f values are TiF4, which is obviously not reliant on TiF3 vapor pressure
off from experiment by 203– 244 kcal mol−1. For both measurements, has a much more precise experimental value.
implementations, a treatment of extended electron correla- The ROHF ccCA-TM deviation of the TiF4 enthalpy is only
tion 关via the CCSD共T兲 cc-pVTZ computation兴 contributes 3.1 kcal mol−1. When the enthalpies of TiF2, TiF3, and CrO3
almost 190 kcal mol−1 to the enthalpy of formation. How- are removed 共set II兲, the MSD of ROHF ccCA-TM is
ever, the ccCA-S4 共Q5兲 core-valence addition correction −1.58 kcal mol−1 and the MAD of ROHF ccCA-TM is
overcorrects at the MP2 aug-cc-pCVTZ level of theory, con- 3.98 kcal mol−1. Further refining the set, we have removed
tributing 37.0 kcal mol−1 to the ⌬H f versus only the tabulated values of Cr共CO兲x, x = 3 – 5. These values are
9.2 kcal mol−1 using CCSD共T兲 aug-cc-pCVDZ in the new from a review by Simoes posted on the NIST webbook25
ccCA implementation. Enthalpies of formation using the new using the JANAF literature value of Cr共CO兲6, which is
ROHF ccCA-TM are too low, but corrected accordingly via 22 kcal mol−1 higher than the recommended value from the
ZPE and 共in the case of closed shell carbonyl and metal- Yungman compendium 共and also ⬃20 kcal mol−1 higher
locene complexes兲 atomic spin-orbit corrections. Once all of than the ROHF ccCA-TM value兲. Also, the NIST webbook
the additive corrections are included, the ⌬H f value of value of the FeC5H5 enthalpy of formation does not include
Ni共CO兲4 is within the experimental error bars of references justifying the specific recommended value.25 Due
⫾2.5 kcal mol−1. The trends shown in Fig. 1 are general for to controversy in the free cyclopentadienyl ⌬H f value and
the larger molecules in our test set; emphasis on the treat- the rather large multireference character of the sextet ground
ment of electron correlation rather than basis set size is cru- state of FeC5H5 radical, we have also removed this value
cial for appropriate consideration of core-core and core- from our test set. When these four molecules are removed
valence correlation. from the test set 共set III兲, the error of ROHF ccCA-TM de-
In the complete set of 58 molecules 共set I兲, the ROHF creases substantially with a MSD of −1.24 kcal mol−1 and a
ccCA-TM shows significant bias with a MSD of MAD of 3.20 kcal mol−1.
−2.50 kcal mol−1 and a large mean absolute error with a By adopting the JANAF value for CrO3, which has a
MAD of 4.78 kcal mol−1. The performance of UHF-based larger uncertainty but agrees very well with CBS-CCSD共T兲
ccCA-TM 共with a MAD of 5.99 kcal mol−1兲 is worse than and ccCA values, and by using theoretical CBS-CCSD共T兲
the results obtained using a ROHF reference wave function values obtained by Ricca51 for Fe共CO兲x, x = 1 – 4, a final sub-
for set I and all subsequent pruned test sets. Clearly spin set of 52 molecules 共set IV兲 is created. The MSD for ROHF
contamination is more pronounced for many transition metal ccCA-TM is −1.15 kcal mol−1 with a MAD of
species. For example, the difference between the UHF and 2.85 kcal mol−1 and a root mean square deviation of
ROHF enthalpies of formation for VO2 is 9.5 and 3.77 kcal mol−1. This MAD is near the average of the ex-
25.7 kcal mol−1 for CuO. A ROHF reference wave function perimental uncertainties 共2.22 kcal mol−1兲 in set IV. Al-
is therefore recommended for inorganic studies and all sub- though this is a rough statistical metric, it shows that the
sequent discussion will focus on ROHF ccCA-TM results. ROHF ccCA-TM model is already at the threshold of mean-
As the test set is slightly pruned 共vida infra), the statistical ingful accuracy for transition metal complexes. The maxi-
differences between ROHF and UHF based ccCA-TM will mum errors in set IV are both open-shell unsaturated carbo-
become more apparent. nyl systems: Mn共CO兲5 and Co共CO兲4 with deviations of
Set I includes some previously discussed outliers, TiF2, ⫺11.2 and −10.1 kcal mol−1, respectively. The enthalpy of

Downloaded 18 May 2012 to 136.159.235.223. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/about/rights_and_permissions
024106-7 Thermochemistry for inorganic complexes J. Chem. Phys. 131, 024106 共2009兲

FIG. 3. Histogram of signed deviations for set IV 共in kcal mol−1兲.

FIG. 5. MADs of set IV 共in kcal mol−1兲 categorized by metal-ligand bond-


formation of Mn共CO兲5 is another potentially controversial ing type.
molecule in our test set. The NIST webbook value has a
larger uncertainty 共−171⫾ 5.3 kcal mol−1兲 共Ref. 25兲 but
agrees much better with the ccCA value 共with a deviation of containing molecules in our test set is slightly worse with a
only −2.8 kcal mol−1兲. However, this molecule shows sig- MAD of 4.25 kcal mol−1. Looking at trends in chemical
nificant spin contamination with an 具S2典 value of 0.26 higher bonding 共Fig. 5兲, all bonding categories are also within the
than the optimal doublet value. In the absence of either clear transition metal chemical accuracy, except for the carbonyls,
information about the origin of the NIST literature values or owing to the large errors in Co共CO兲4 and Mn共CO兲5. With a
coupled-cluster based theoretical values, further study is war- MAD of 2.70 kcal mol−1 for the eight metal-oxide bonds
ranted to resolve major discrepancies between literature val- and a MAD of 1.93 kcal mol−1 for the four organometallic
ues and ccCA. Overall, however, the new ccCA performs species in our test set, ccCA shows promising capabilities in
extremely well for large TM-containing systems. For ex- the regimes of combustion and catalysis research for 3d met-
ample, the MAD for 16 molecules in set IV that contain als.
more than five atoms is only 3.39 kcal mol−1. In Table VI, carbonyl BDEs are given using ccCA. Re-
Figure 2 shows the scatter of experimental versus theo- sults are compared to DFT computations performed by
retical values for test set IV. With an R2 value of 0.999 and a Furche and Perdew7 as well as CCSD共T兲-based calculations
nearly linear slope, ROHF ccCA-TM is a reliable predictor performed by Ricca and Bauschlicher.52 These BDEs are
of gas phase enthalpies of formation without significant sys- known to provide an extremely stringent test of computa-
tematic error. For the 52 molecules in set IV, 29 molecules in tional methods, as FeCO and FeCp are known to have high
our test set are within the transition metal chemical accuracy multireference character. For instance, Ricca and Bauschli-
of ⫾3.0 kcal mol−1 using ccCA, with 16 of those enthalpies cher showed that CCSD共T兲 and multireference configuration
of formation within the chemical accuracy of interaction methods generally give the wrong electronic
⫾1.0 kcal mol−1 from the experimental value. Figure 3 ground state for FeCO.52 The delicate balance between dy-
shows the histogram of signed deviations for set IV. Most of namical and nondynamical electron correlation in this mol-
the errors cluster either very close to the experimental value ecule is akin to the difficulties encountered in applying
or deviate in the range of −6.0 – − 2.0 kcal mol−1 from the CCSD共T兲 and MRCISD methods to investigate FeCN and
experimental values. To further break down the enthalpies of FeNC.53 For the first four carbonyl dissociations of Fe共CO兲5,
formation in our training set, Fig. 4 shows the MADs for ccCA performs very well and is within or very close to being
each of the ten 3d elements. Unlike the inconsistencies seen within the range of experimental uncertainty. However, the
in DFT benchmarking, ccCA is better than or near transition dissociation of excited state FeCO to Fe+ CO shows the
metal chemical accuracy when computing enthalpies of for- same difficulties that previous theoretical studies have
mation from Ti–Zn. The performance on the four scandium- highlighted.51,52 The ccCA method also predicts an incorrect
ground electronic state 共 5⌺−兲 for FeCO, compared to the ex-
perimental observation of Leopold and co-worker 共 3⌺−兲.54
The first CO dissociation of Cr共CO兲6 is only
−4.2 kcal mol−1 from the experimental value of 39⫾ 3,
while the third CO dissociation is in error by 37 kcal mol−1
Note that the recommended literature value of the Cr共CO兲6
→ Cr共CO兲3 + 3CO reaction uses the questionable value of the
Cr共CO兲6 enthalpy of formation in the NIST webbook.25
Since all species involved in this reaction are closed shell, it
is likely that the ccCA value is much closer to the true value.
BDEs for ferrocene are also listed in Table VI. The dis-
sociation of 1/2 molar equivalent of ferrocene to cyclopenta-
dienyl radical and half a molar equivalent of iron atom de-
FIG. 4. MADs of set IV categorized by TM element 共in kcal mol−1兲. viates from experiment by −5.7 kcal mol−1. However, the

Downloaded 18 May 2012 to 136.159.235.223. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/about/rights_and_permissions
024106-8 DeYonker et al. J. Chem. Phys. 131, 024106 共2009兲

TABLE VI. Dissociation energies 共in kcal mol−1兲 for various carbonyl and ferrocene reactions.

ROHF ccCA-TM CCSD共T兲 B3LYP TPSSh Expt.

Fe共CO兲5 → Fe共CO兲4 + CO 41.6 36.9 a


45.7 41.5⫾ 2 b
Fe共CO兲4 → Fe共CO兲3 + CO 36.5 31.1 27.9⫾ 8.8 c
Fe共CO兲3 → Fe共CO兲2 + CO 31.6 32.8 29.1⫾ 5.8 c
Fe共CO兲2 → FeCO+ CO 40.5 33.7 36.7⫾ 3.5 c
5
FeCO→ 5Fe + CO 1.3 共2.2兲d 7.0 12.4 19.2 8.1⫾ 3.5 c
Cr共CO兲6 → Cr共CO兲5 + CO 43.2 35.6 42 39⫾ 3 b
Cr共CO兲6 → Cr共CO兲3 + 3CO 136.9 108.9 131.4 99⫾ 10 e
1 / 2Fe共Cp兲2 → 1 / 2Fe+ Cp 87.1 67.9 88 81.4⫾ 1 f 共88.2兲g
Fe共Cp兲2 → Fe+ 2Cp 174.2 158.4h 共172兲g
a
All DFT results are from Ref. 7.
b
Reference 57.
c
Reference 51.
d
Reference 52. Value in parenthesis is without ad hoc correction of experimental 3⌺− − 5⌺− energy splitting.
e
Reference 25.
f
Reference 58.
g
Values in parenthesis are BDEs utilizing corrected C5H5 enthalpy of formation from Ref. 56.
h
Reference 59.

literature value of this BDE comes via the usage of an en- thermochemical values, such as the “higher” enthalpy of for-
thalpy of formation for C5H5 radical of 58⫾ 2 kcal mol−1.55 mation of cyclopentadienyl radical determined by Nunes et
This value has been recently revisited by Nunes et al.56 using al.56
time-resolved photoacoustic calorimetry and a newly recom- Previous DFT benchmark studies highlight the pitfalls
mended value of 64.8⫾ 2.0 kcal mol−1 was obtained 共in and nonsystematic nature of choosing the “correct” func-
agreement with the ROHF-ccCA-TM value of tional. Cognizance is necessary in utilization of DFT to pre-
65.4 kcal mol−1兲. Using this newly recommended value of dict energetic properties of open-shell and/or coordinatively
⌬H f 共C5H5兲 results in a ROHF-ccCA-TM value approxi- unsaturated inorganic molecules. The ccCA is reliable
mately within 1 – 2 kcal mol−1 of the experimental BDE for enough to provide a guideline for the creation of improved
ferrocene to iron+ cyclopentadienyl radical. density functionals for TM thermochemistry when 共as is of-
ten the case兲 such data are sparse or nonexistent, and perhaps
will also help thermochemical research place the emphasis
back on wave function-based methods to resolve controver-
IV. CONCLUSIONS sial quantities and to instigate investigation of new chemical
problems in the realm of quantitative inorganic chemistry.
By focusing on extrapolation to a complete basis set
This research also highlights the need for a “gold-standard”
limit, in tandem with flexible correlation consistent basis
共but computationally feasible兲 methodology in TM thermo-
sets, efficient MP2-based single reference wave function
chemistry akin to CBS-CCSD共T兲 methods for important
methods can be employed to obtain quantitatively accurate
chemical issues in which pertinent experimental/
thermochemical values. Our improved version of the ccCA
computational data are conflicting. However, such large basis
has been created such that not only it achieves main group
set coupled cluster computations are a long way from being
chemical accuracy for the G2/97 enthalpies of formation
tractable for molecules as large as Ni共PF3兲4 and beyond, but
共with a MAD of 0.89 kcal mol−1 over a set of 147 mol-
still they represent a worthy computational goal.
ecules兲 but also exceeds transition metal chemical accuracy
for small transition metal-containing radicals as well as
ACKNOWLEDGMENTS
larger carbonyl and organometallic complexes 共with a MAD
of 2.85 kcal mol−1 over a set of 52 molecules whose enthal- We would like to thank Professor Dave A. Dixon and
pies of formation are reasonably well established兲. While Professor Kirk A. Peterson for helpful discussions and K. A.
DFT studies have shown that certain functionals display an P. for providing unpublished cc-pCVDZ basis sets for Sc–
erratic behavior when used to compute energetics of early Zn. CASCaM is supported by a grant from the U.S. Depart-
versus late transition metal containing species, ccCA is con- ment of Energy 共Grant No. BER-08ER64603兲. This research
sistently accurate for the entire 3d block. Unlike our first is also partially supported by the National Science Founda-
reports, the quality of the present ccCA methodology does tion via CAREER Award Nos. CHE-0239555 and CHE-
not degrade when tested against molecules with fairly high 0809762 共for A.K.W.兲 and by a grant from the United States
multireference character nor as a function of increasing mo- Department of Energy, Office of Basic Energy Sciences 共for
lecular size. Generally, BDEs computed with ccCA are reli- A.K.W. and T.R.C.兲 under Grant No. DE-FG02-03ER15387.
able. We have identified a few molecules for which the ex- Computational resources were provided via the National Sci-
perimental gas phase enthalpies of formation should clearly ence Foundation 共Grant Nos. CHE-0342824 and CHE-
be revisited such as FeCO, TiF2, TiF3, and Cr共CO兲3, as well 0741936兲 and by the Teragrid Alliance under Grant No. TG-
as amounted increasing evidence for the adoption of certain CHE010021 共for A.K.W.兲 and utilized the NCSA IBM p690

Downloaded 18 May 2012 to 136.159.235.223. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/about/rights_and_permissions
024106-9 Thermochemistry for inorganic complexes J. Chem. Phys. 131, 024106 共2009兲

26
and the SGI Altix. Additional support was provided by the D. Rinaldo, L. Tian, J. N. Harvey, and R. A. Friesner, J. Chem. Phys.
129, 164108 共2008兲.
University of North Texas Academic Computing Services for 27
H. Partridge, J. Chem. Phys. 90, 1043 共1989兲.
the use of the UNT Research Cluster. CASCaM is supported 28
F. Weigend, F. Furche, and R. Ahlrichs, J. Chem. Phys. 119, 12753
by a grant from the U.S. Department of Education. AEI was 共2003兲.
supported by the NSF-REU site in Chemistry at UNT 共Grant
29
F. Weigend and R. Ahlrichs, Phys. Chem. Chem. Phys. 7, 3297 共2005兲.
30
No. CHE-0648843兲. N. J. Mayhall, K. Raghavachari, P. C. Redfern, L. A. Curtiss, and V.
Rassolov, J. Chem. Phys. 128, 144122 共2008兲.
31
1 N. B. Balabanov and K. A. Peterson, J. Chem. Phys. 123, 064107 共2005兲.
J. M. Smith, A. R. Sadique, T. R. Cundari, K. R. Rodgers, G. Lukat- 32
N. B. Balabanov and K. A. Peterson, J. Chem. Phys. 125, 074110 共2006兲.
Rodgers, R. J. Lachicotte, C. J. Flaschenriem, J. Vela, and P. L. Holland, 33
N. J. DeYonker, B. R. Wilson, A. W. Pierpont, T. R. Cundari, and A. K.
J. Am. Chem. Soc. 128, 756 共2006兲. Wilson, Mol. Phys. 107, 1107 共2009兲.
2
J. M. Smith, R. J. Lachicotte, K. A. Pittard, T. R. Cundari, G. Lukat- 34
D. S. Ho, N. J. DeYonker, A. K. Wilson, and T. R. Cundari, J. Phys.
Rodgers, K. R. Rodgers, and P. L. Holland, J. Am. Chem. Soc. 123, 9222
Chem. A 110, 9767 共2006兲.
共2001兲. 35
3 N. J. DeYonker, D. S. Ho, A. K. Wilson, and T. R. Cundari, J. Phys.
N. S. Lewis and D. G. Nocera, Proc. Natl. Acad. Sci. U.S.A. 103, 15729
Chem. A 111, 10776 共2007兲.
共2006兲. 36
4 N. J. DeYonker, T. Grimes, S. Yockel, A. Dinescu, B. Mintz, T. R.
T. A. Betley, Q. Wu, T. Van Voorhis, and D. G. Nocera, Inorg. Chem. 47,
Cundari, and A. K. Wilson, J. Chem. Phys. 125, 104111 共2006兲.
1849 共2008兲. 37
5 T. G. Williams and A. K. Wilson, J. Sulfur Chem. 29, 353 共2008兲.
T. R. Cundari, H. A. R. Leza, T. Grimes, G. Steyl, A. Waters, and A. K. 38
N. J. DeYonker, B. Mintz, T. R. Cundari, and A. K. Wilson, J. Chem.
Wilson, Chem. Phys. Lett. 401, 58 共2005兲.
6 Theory Comput. 4, 328 共2008兲.
K. E. Riley and K. M. Merz, J. Phys. Chem. A 111, 6044 共2007兲. 39
7
F. Furche and J. P. Perdew, J. Chem. Phys. 124, 044103 共2006兲. Y. Zhao and D. G. Truhlar, Theor. Chem. Acc. 120, 215 共2008兲.
40
8
Y. Zhao and D. G. Truhlar, J. Chem. Phys. 124, 224105 共2006兲. N. J. DeYonker, K. A. Peterson, G. Steyl, A. K. Wilson, and T. R.
9
J. F. Harrison, Chem. Rev. 100, 679 共2000兲. Cundari, J. Phys. Chem. A 111, 11269 共2007兲.
41
10
M. Brynda, L. Gagliardi, and B. O. Roos, Chem. Phys. Lett. 471, 1 B. A. Hess, Phys. Rev. A 33, 3742 共1986兲.
42
共2009兲. B. A. Hess, Phys. Rev. A 32, 756 共1985兲.
43
11
T. R. Cundari, A. Dinescu, and A. B. Kazi, Inorg. Chem. 47, 10067 M. Dolg, U. Wedig, H. Stöll, and H. Preuss, J. Chem. Phys. 86, 866
共2008兲. 共1987兲.
44
12
C. W. Bauschlicher, S. P. Walch, and H. Partidge, J. Chem. Phys. 76, T. H. Dunning, Jr., J. Chem. Phys. 90, 1007 共1989兲.
45
1033 共1982兲. M. J. Frisch, G. W. Trucks, H. B. Schlegel et al., GAUSSIAN 03, Revision
13
S. R. Langhoff and J. C. W. Bauschlicher, Annu. Rev. Phys. Chem. 39, D.02, Gaussian, Inc., Wallingford, CT, 2004.
46
181 共1988兲. MOLPRO, a package of ab initio programs designed by H.-J. Werner and P.
14
I. M. B. Nielsen and M. A. Allendorf, J. Phys. Chem. A 109, 928 共2005兲. J. Knowles, Version 2006.1, R. Lindh, F. R. Manby, M. Schütz et al.
47
15
I. M. B. Nielsen and M. D. Allendorf, J. Phys. Chem. A 110, 4093 L. A. Curtiss, K. Raghavachari, P. C. Redfern, V. Rassolov, and J. A.
共2006兲. Pople, J. Chem. Phys. 109, 7764 共1998兲.
16 48
E. J. Opila, D. L. Myers, N. S. Jacobson, I. M. B. Nielsen, D. F. Johnson, V. S. Yungman, Thermal Constants of Substances Wiley, New York,
J. K. Olminsky, and M. D. Allendorf, J. Phys. Chem. A 111, 1971 2001兲.
49
共2007兲. See EPAPS supplementary material at http://dx.doi.org/10.1063/
17
S. G. Li and D. A. Dixon, J. Phys. Chem. A 112, 6646 共2008兲. 1.3160667 for experimental enthalpies of formation and atomic/
18 molecular ccCA energies.
H. J. Zhai, S. Li, D. A. Dixon, and L. S. Wang, J. Am. Chem. Soc. 130,
50
5167 共2008兲. K. F. Zmbov and J. L. Margrave, J. Phys. Chem. 71, 2893 共1967兲.
51
19
S. G. Li and D. A. Dixon, J. Phys. Chem. A 110, 6231 共2006兲. A. Ricca, Chem. Phys. Lett. 350, 313 共2001兲.
52
20
S. G. Li and D. A. Dixon, J. Phys. Chem. A 111, 11908 共2007兲. A. Ricca and C. W. Bauschlicher, Theor. Chem. Acc. 106, 314 共2001兲.
21 53
I. Hyla-Kryspin and S. Grimme, Organometallics 23, 5581 共2004兲. N. J. DeYonker, Y. Yamaguchi, W. D. Allen, C. Pak, H. F. Schaefer, and
22
L. A. Curtiss, P. C. Redfern, and K. Raghavachari, J. Chem. Phys. 123, K. A. Peterson, J. Chem. Phys. 120, 4726 共2004兲.
54
124017 共2005兲. P. W. Villalta and D. G. Leopold, J. Chem. Phys. 98, 7730 共1993兲.
23 55
J. Cioslowski, M. Schimeczek, G. Liu, and V. Stoyanov, J. Chem. Phys. J. P. Puttemans, G. P. Smith, and D. M. Golden, J. Phys. Chem. 94, 3226
113, 9377 共2000兲. 共1990兲.
24 56
D. Feller, K. A. Peterson, and D. A. Dixon, J. Chem. Phys. 129, 204105 P. M. Nunes, F. Agapito, B. J. Costa Cabral, R. M. Borges dos Santos,
共2008兲. and J. A. Martinho Simões, J. Phys. Chem. A 110, 5130 共2006兲.
25 57
H. Y. Afeefy, J. F. Liebman, and S. E. Stein, “Neutral Thermochemical K. E. Lewis, D. M. Golden, and G. P. Smith, J. Am. Chem. Soc. 106,
Data” in NIST Chemistry WebBook, NIST Standard Reference Database 3905 共1984兲.
58
Number 69, edited by P. J. Linstrom and W. G. Mallard 共National Insti- M. F. Ryan, J. R. Eyler, and D. E. Richardson, J. Am. Chem. Soc. 114,
tute of Standards and Technology, Gaithersburg, MD, 2005兲, p. 20899, 8611 共1992兲.
59
http://webbook.nist.gov. M. L. McKee, J. Phys. Chem. 96, 1683 共1992兲.

Downloaded 18 May 2012 to 136.159.235.223. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/about/rights_and_permissions

Anda mungkin juga menyukai