You are on page 1of 12

Theoretical Calculations: Can Gibbs Free Energy

for Intermolecular Complexes Be Predicted Efficiently


and Accurately?

OLEXANDR ISAYEV, LEONID GORB, JERZY LESZCZYNSKI


Computational Center for Molecular Structure and Interactions, Jackson State University,
Jackson, Mississippi 39217

Received 27 October 2005; Revised 12 December 2005; Accepted 28 December 2005


DOI 10.1002/jcc.20696
Published online 5 March 2007 in Wiley InterScience (www.interscience.wiley.com).

Abstract: The theoretical study has been performed to refine the procedure for calculations of Gibbs free energy
with a relative accuracy of less than 1 kcal/mol. Three benchmark intermolecular complexes are examined via sev-
eral quantum-chemical methods, including the second-order Moller-Plesset perturbation (MP2), coupled cluster
(CCSD(T)), and density functional (BLYP, B3LYP) theories augmented by Dunnings correlation-consistent basis
sets. The effects of electron correlation, basis set size, and anharmonicity are systematically analyzed, and the results
are compared with available experimental data. The results of the calculations suggest that experimental accuracy
can be reached only by extrapolation of MP2 and CCSD(T) total energies to the complete basis set. The contribution
of anharmonicity to the zero point energy and TDSint values is fairly small. The new, economic way to reach chemi-
cal accuracy in the calculations of the thermodynamic parameters of intermolecular interactions is proposed. In addi-
tion, interaction energy (De) and free energy change (DA) for considered species have been evaluated by Carr-Parri-
nello molecular dynamics (CPMD) simulations and static BLYP-plane wave calculations. The free energy change
along the reaction paths were determined by the thermodynamic integration/’’Blue Moon Ensemble’’ technique.
Comparison between obtained values, and available experimental and conventional ab initio results has been made.
We found that the accuracy of CPMD simulations is affected by several factors, including statistical uncertainty and
convergence of constrained forces (TD integration), and the nature of DFT (density functional theory) functional.
The results show that CPMD technique is capable of reproducing interaction and free energy with an accuracy of
1 kcal/mol and 2–3 kcal/mol respectively.
q 2007 Wiley Periodicals, Inc. J Comput Chem 28: 1598–1609, 2007

Key words: free energy; interaction energy; intermolecular complex; ab initio; density function theory; Carr-Parri-
nello molecular dynamics; thermodynamic integration

Introduction Another well-established technique to predict the electronic


structure of molecules and molecular systems is the density
Canonical quantum chemistry that is quite often called ‘‘wave functional theory (DFT).5 Since the DFT treatment is usually
function-based quantum chemistry methods’’ is capable of deter- much faster than the application of methods employing tradi-
mining molecular properties at a very high level of accuracy. tional quantum chemistry, the DFT methods are known to lead
The most common area of its application is the study of small- to amazingly good predictions for a wide variety of systems of
or medium-sized systems, especially gas phase phenomena. small, moderate, and large sizes.6–14 However, dramatic failures
There is also plenty of evidence that the wave function-based
quantum chemistry methods are able to predict gas-phase prop-
erties of molecular systems up to virtually any desired accuracy
as a result of systematically improving the correlation treatment Correspondence to: J. Leszczynski; e-mail: jerzy@ccmsi.us
together with the wave function representation. For example, the Contract/grant sponsor: National Science Foundation; contract/grant
modern techniques are capable of reaching about 10 cm1 accu- numbers: NSF CREST HRD-0125484
racy for fundamental vibrational frequencies1 and 0.002–0.003 Å Contract/grant sponsor: Office of Naval Research; contract/grant num-
accuracy in equilibrium bond distances.2–4 bers: ONR N00034-03-1-0116

q 2007 Wiley Periodicals, Inc.


Efficient and Accurate Prediction of Gibbs Free Energy 1599

of this technique are also well documented in the literature.15–19 tosine, the difference between interaction energy and interaction
Among them the most significant, probably, is the incapacity of Gibbs free energy is about 12 kcal/mol.30
most of the DFT methods to correctly describe van der Waals The goal of this study is the evaluation of procedures for pre-
interactions.15,16 This is a serious limitation in the prediction of dictions of thermodynamic parameters (DHint, DSint, DGint) of
the properties of intermolecular complexes. intermolecular interactions by relatively computationally inex-
There is one area where extremely high accuracy is crucially pensive quantum chemical techniques such as DFT5 in conjunc-
important and has significant impact. This is in the determina- tion mostly with the B3LYP functional31,32 (For the develop-
tion of thermodynamic parameters of molecular species and ment of new DFT functionals that are also able to describe ther-
chemical reactions when those species are involved. The avail- modynamics and kinetics of chemical reactions, see ref. 33 We
able techniques represent various compromises between accu- are planning to undertake a special study for the comparison of
racy and computational cost. The most efficient technique in- our approach with the ones suggested in ref. 33), the second
cludes calculations at the CCSD(T) level using extended basis order of Möller-Plesset perturbation theory,34 or the combination
sets or extrapolated to complete basis sets (CBSs), which of these procedures. To elucidate our problem, we have chosen
according to definition should provide extremely accurate three diverse intermolecular systems. They are well character-
results.20 It is estimated that this level of theory, including rela- ized by experimental data. Among them are the well-known val-
tivistic effects and non-Born-Oppenheimer terms, leads to an ues for water dimer interactions at 373 K,35 along with the same
uncertainty of 60.05 kcal/mol in binding energy of water parameters for the formic acid dimer,36 and the HCN. . .Liþ
dimer.21 There are also several less time consuming thermo- complex37 at 298 K. These three systems represent intermolecu-
chemicals protocols such as the ‘‘Gaussian-n’’ (Gn) family of lar complexes that are formed by single and double hydrogen
methods.22–24 These methods first provided accuracy of the ther- bonds, and the ion–molecular species. Their thermodynamic pa-
mochemistry of small molecules in the kilocalories per mole rameters can be considered as a benchmark for our ab initio
range; currently such methods possess an average absolute devi- simulations.
ation of 1.07 kcal/mol. The Gn theory relies on relatively small The evaluation of the thermodynamic parameters of molecu-
basis sets, additivity approximations, and empirical corrections. lar species and chemical reactions in the way described earlier
Similar remarks apply to the CBS family of methods by Peters- uses the static methods of quantum mechanics. The dynamic
son and coworkers,25,26 which involve intricate combinations of methods are incorporated in numerous molecular dynamics (MD)
pair correlation extrapolations and empirical corrections. Among techniques. Computing free energies and entropies at this level
recent developments, it should also be mentioned that there are is one of the main goals and major efforts of MD simulations.
more accurate but computationally more expensive approxima- Unfortunately, these two statistical properties are not easily ac-
tions of the Wn (Weizmann) family27,28 and the HEAT theory29 cessible for simulations as their evaluation in principle requires
which provides accuracy within 1 kJ/mol. These methods in- an infinitely long simulation to scan the entire space. Neverthe-
clude explicit corrections, for anharmonicity, relativistic correc- less, significant progress in studying free-energy differences has
tions and treatment of electron correlation at the coupled-cluster been achieved, showing that it is, in spite of all the sampling
level of theory. problems, possible to obtain reasonable free-energy estimates.38–
45
A common feature of these methods is the use of the QCI, The reason is that, in order to obtain a free-energy difference
CCSD(T), or even CCSDTQ levels of theory at the final step, between two states, the evaluation of the complete partition
which is definitely necessary to reach very high levels of accu- function is not really necessary; the extensive sampling of the
racy. However, such an approach has a very serious disadvant- relevant parts where the two states differ suffices.45,46 In addi-
age since it currently limits the size of the considered systems to tion, the advantage of MD techniques is that the thermal effects,
approximately 6–10 atoms of the second row of the Periodic Ta- such as entropy contribution, are included explicitly. Moreover,
ble. In addition, the majority of available computational proto- most MD approaches, applied at the ab initio level, are BSSE
cols are devoted to the prediction of the enthalpy of formation free. Therefore, for the sake of comparison we have also
and Gibbs free energy of isolated molecules, ions, and radicals. included in this study the results of an evaluation of the relative
In contrast, the related area of an ab initio prediction of the free energies obtained in the framework of a dynamic quantum-
thermodynamic parameters of intermolecular interactions is not chemical approximation.
well represented. To fill this gap, the present study exclusively
focused on the discussion and development of efficient and
accurate protocols for the determination of thermodynamic pa- Computational Methods
rameters of intermolecular complexes.
We also would like to mention that the amount of trustwor- The ab initio linear combination of the atomic-orbitals–molecu-
thy experimental data that include not only the interaction en- lar-orbitals method was used in this study. The calculations were
thalpy (DHint) but also the interaction entropy (DSint) as well as carried out using the GAUSSIAN 03 (ref. 47) code. The total
interaction Gibbs free energy (DGint) values is extremely limited. and interaction energies have been evaluated at the second order
On the other hand, the accurate prediction of DGint is most im- of the Möller-Plesset perturbation theory34 and the DFT level
portant for intermolecular complexes, since in contrast to most using the B3LYP and BLYP functionals31,32 and the counter-
molecules and ions, this value can be significantly different poise-corrected optimization procedure as employed in Gaussian
when compared to DHint. For example, in the case of the forma- 03. The optimization procedure has been described in ref. 48.
tion of a Watson–Crick DNA base pair between guanine and cy- This procedure results in direct optimization of the counterpoise-

Journal of Computational Chemistry DOI 10.1002/jcc


1600 Isayev, Gorb, and Leszczynski • Vol. 28, No. 9 • Journal of Computational Chemistry

corrected interaction energy, which is calculated according to 0.0241888 fs) was used for the numerical integration of the
the equation equations of motions, according to the velocity Verlet algorithm.
In order to determine the properties averaged within the canoni-
X
n X
n cal ensemble, a Nose-Hoover chain thermostat58 with a length
e ¼ Esp 
DCP Emiopt þ ðEmij  Emij Þ (1) of 4 and a characteristic frequency of 500 cm1 was used on the
i¼1 i¼1 nuclear degrees of freedom. Plane wave geometry optimization
was performed with the ODIIS algorithm.59
where DCPe is the counterpoise-corrected interaction energy; Esp Deuterium has been utilized instead of hydrogen to allow a
is the energy of the complex; Em represents the energies of the larger time step. Therefore, in the discussion below (Molecular
individual monomers, with the subscripts ‘‘opt’’ and ‘‘f’’ denot- Dynamics Simulations), it should be kept in mind that all spe-
ing the individually optimized monomers frozen in their super- cies that include hydrogen atoms are in fact deuterium contain-
molecular geometries; the asterisks (*) denote monomers calcu- ing species.
lated with ‘‘ghost’’ orbitals. Two energetic parameters have been evaluated from the MD
The cc-pvdz, cc-pvtz, and aug-cc-pvdz basis sets have been simulation: the interaction energy, which is determined as
used. The vibrational frequencies were always evaluated at the
X
same level as the geometry optimization. Single point calcula- De ¼ hEcomplex i  hEi i (4)
tions have been performed at the CCSD(T) level of theory. i
The MP2 and CCSD(T) total energies were extrapolated to
the CBS values 2-3 (cc-pvdz and cc-pvtz basis sets, respectively) where hEi is the averaged energy values of the intermolecular
according to the expression of Truhlar49 complex and its components and the change of the free energy
DA. To evaluate DA, the phase space along the reaction path has
3 3 been explored by the Blue Moon method,60 keeping the
Etot;1 ¼ EHF;3   EHF;2
3 2  3  2 r(OO), r(C C), or r(LiN) length constrained. After short

3 3 equilibration (2–3 ps) at 300 K by velocity scaling, the equa-
þ  
Ecorr;3   Ecorr;2 ð2Þ
3 2 3  2 tions of motion have been integrated with a time step of 4 a.u.
for a time about 2–3 ps in the NVT ensemble, and the change
in the free energy value between states 1 and 2 has been calcu-
where  is equal to 3.4, and  is equal to 2.2 and 2.4 for MP2 lated using the thermodynamic integration method:
and CCSD(T), respectively, and to the expression of Helgaker
Z
and coworkers50 2 hZ1=2 ½li0
A ¼  f0 d0 ; f 0 ¼ (5)
1 hZ 1=2 i0
EHF;X ¼ EHF;1 þ a expðbXÞ
X3 ðX  1Þ3 where  is the value of the constrained reaction coordinate,
Ecorr;1 ¼ E
3 corr;X
 Ecorr;X1 ð3Þ
X3  ðX  1Þ X3  ðX  1Þ3 and l is the Lagrange multiplier on the constraint which is pro-
portional to the force due to constraint.
Note that all DA values are free energies calculated at constant
where EHF,X, Ecorr,X, and Ecorr,X1 are Hartree-Fock and correla- volume; therefore they are Helmholtz’s free energies. However,
tion energies as a function of the cardinal number X of the cc- for most reactions the volume change of the reaction is fairly
pvXz basis set, X ¼ 3. small (including the interactions, considered in this paper); there-
The thermodynamic parameters of the intermolecular interac- fore, DA is approximately equal to DG under ambient pressure.
tions (interaction enthalpy (DHint), interaction entropy (DSint),
and interaction Gibbs free energy (DGint)) were calculated from
standard expressions using rigid rotor-harmonic oscillator-ideal
Results and Discussion
gas approximations.51
The Car-Parrinello molecular dynamics simulations52 were Dissociation Energies and Interaction Enthalpies
performed with the CPMD program53 using a plane-wave basis
and BLYP functional.31,32 The wave function was expanded at Benchmark calculations of the dissociation energies and interac-
the point in a plane-wave basis set with the kinetic energy cutoff tion enthalpies of intermolecular complexes are presented and
of 70 and 120 Ry respectively. A large 18-Å cubic box was analyzed elsewhere (see, for example, refs. 61–63). The avail-
used under the periodic boundary conditions; this unit cell is able published data include a wide range of the interaction ener-
sufficiently large to ensure negligible overlap of the wave func- gies, starting from dispersion interactions through hydrogen-
tions with the periodic images.54 bonded systems to ion–molecular complexes. We will not pres-
The form of the nonlocal pseudopotential according to Klein- ent here a comprehensive survey of these interactions. However,
man and Bylander was employed,55 and the core electrons were for the purposes of our discussion it is important to highlight the
described by the Troullier and Martins pseudopotentials.56 The following point. As follows from the literature, accurate MP2
Goedecker–Teter–Hutter pseudopotentials57 were used for the and accurate CCSD(T) calculations are able to reproduce the
lithium compounds. The time step 4 a.u. (where 1 a.u. ¼ dissociation energies and interaction enthalpies with experimen-

Journal of Computational Chemistry DOI 10.1002/jcc


Efficient and Accurate Prediction of Gibbs Free Energy 1601

Table 1. DH and D0 (in kcal/mol) for Selected Intermolecular Complexes (Comparison Theoretical
Estimations and Experiment).

Ref. Ref.
System Method Theor. D0(theor) DHint(theor) D0(expt) DHint(expt) expt.

(H2O)2 MP2/aug-cc-pVTZ, CP 64 4.83 5.4460.7 35


MP2/aug-cc-pVQZ, CP 64 4.92
CCSD(T)/cc-pVTZ, CP 64 4.48
(CONH2)2 MP2/D95V**, CP 65 3.24 3.34 66
(HF)2 B3LYP/6-311þG(2df,p), CP 9 2.68 3.0360.02 67,68
MP2/6-311þG(2df,p), CP 2.30
QCISD/6-311þG(2df,p), CP 2.18
MP2/cc-pVTZ//MP2/6-311G(d,p) 69 3.9 4.660.3 (De) 70
W1 71 4.57
(CH3OH)2 MP2/cc-pVTZ//MP2/6-311G(d,p) 69 4.82 4.6 to 5.9 (De) 72,73
MP2/CBS//MP2/6-311G(d,p) 5.58
CCSD(T)/CBS//MP2/6-311G(d,p) 5.45
HCN. . .HF B3LYP/cc-pVTZ 69 6.35 6.9 (De) 74
MP2/cc-pVTZ//MP2/6-311G(d,p) 6.43
CCSD(T)/CBS//MP2/6-311G(d,p) 7.11
CH3COCH3. . .Naþ MP2(full)/6-311þG(2d,2p)//MP2/6-31G* 75 30.0 30.860.5 76
Imidazole. . .Naþ MP2(full)/6-311þG(2d,2p)//MP2/6-31G* 75 34.9 33.7 76
H2O. . .Liþ MP2/6-311þþG(d,p) 77 35.3 32.7 78
MP2/6-311þG(d,p) 79 36.1 34.0 37
CH3CN. . .Kþ B3LYP/6-311þG*(2df,2p) 80 101.9 102.1 81
Ne2 MP2/aug-cc-pV5Z 69 0.05 0.08 82
CCSD(T)/aug-cc-pV5Z 0.07
Ar2 MP2/aug-cc-pV5Z 69 0.30 0.28 82
CCSD(T)/aug-cc-pV5Z 0.26

tal accuracy. This indicates that the difference between the MP2 Tables 2–4. The results obtained at the MP2 level along with
and CCSD(T) or even higher level computational data are insig- cc-pvdz, cc-pvtz, and aug-cc-pvtz basis sets exhibit unsatisfac-
nificant values, and the difference between accurately calculated tory correspondence to the experimental measurements since the
and experimental values also does not exceed 0.5 kcal/mol (or calculated DHint values deviate from the experimental results by
even less). To confirm this conclusion, we have presented in 1.5–3.5 kcal/mol. The DFT results are much better; thus, the
Table1 the selected available data. calculated DHint values lie within a 1–1.5 kcal/mol range of the
Let us now briefly describe the situation with the intermolec- experimental data. Not surprisingly, our data that include extrap-
ular complexes of interest. In the case of a water dimer, a dis- olation to a CBS insignificantly deviate from the experimentally
cussion of the accuracy of the calculated values of interaction observed values. For all cases, the most accurate results are
enthalpies has a long history. Overall, the water dimer is prob- obtained by either the CBS extrapolated MP2 energies from full
ably the most demanded intermolecular complex ever. The most MP2 optimization or the CBS extrapolation of single point MP2
accurate calculations performed at the MP2 and CCSD(T) levels energies from the B3LYP/cc-pvtz geometry.
which use counterpoise-corrected optimizations,61–63,83,84 total An analysis of the data presented in Tables 2–4 allows a few
energy-extrapolated to a CBS,71 and anharmonic values of vibra- important results to be summarized.
tional frequencies85,86 allow to reproduce the experimental value
of the interaction energy35 and the enthalpy of formation with i. Experimental accuracy is reached only at the MP2 and
an uncertainty of about 60.2–0.3 kcal/mol. CCSD(T) levels, if one will use the extrapolation to the
Two other complexes have received less attention. The for- CBS.
mic acid dimer was investigated at the HF, MP2, DFT, and ii. We did not use anharmonic corrections in our calculations;
CCSD(T) levels of theory.69,87,88 The best DFT (B3LYP and however, as follows from the data presented in Tables 2–4,
PW91 with cc-pV5Z) and CCSD(T)/CBS calculated interaction experimental accuracy in the calculation of DHint is reached
enthalpy results are within 0.5 kcal/mol from experimental data. even for vibrational frequencies calculated at the harmonic
To our knowledge, there is only one systematic study of the approximation. A similar conclusion for thermodynamical
HCN. . ..Liþ complex. It was investigated at the B3LYP/6-311 þ evaluations that target an accuracy of 0.25 kcal/mol is sug-
G(d,p) level of theory,89 and the calculated DHint value was 0.8 gested by a few other studies.27,86,90–92 It was also pointed
kcal/mol larger than the available experimental data. out that the anharmonic contributions to the zero point
Calculated and available experimental data for the water energy (ZPE) can be safely neglected, at least for molecules
dimer, formic acid dimmer, and HCN. . ..Liþ are presented in of medium size. To our knowledge, there are only a few

Journal of Computational Chemistry DOI 10.1002/jcc


1602 Isayev, Gorb, and Leszczynski • Vol. 28, No. 9 • Journal of Computational Chemistry

Table 2. Calculated Total Energies E (in a.u.) and Thermodynamical Parameters (in kcal/mol) for Water
Dimer Interaction at 373 K.

E (H2O) E ((H2O)2) DHint TDSint DGint

B3LYP /cc-pvdz 76.42063 152.84867 3.02 7.05 4.04


/cc-pvtz 76.45984 152.92693 2.91 7.49 4.58
/aug-cc-pvdz 76.44464 152.89643 2.86 7.52 4.66

BLYP /cc-pvdz 76.39887 152.80450 2.66 7.27 4.61


/cc-pvtz 76.44146 152.88958 2.59 7.31 4.72

MP2 /cc-pvdz 76.22867 152.46405 2.58 7.02 4.44


/cc-pvtz 76.31866 152.64443 2.86 6.92 4.06
/aug-cc-pvdz 76.26091 152.52894 2.86 7.39 4.52

Truhlar CBS (cc-pvdz–cc-pvtz) 76.37015 152.74844 3.46 MP2/cc-pvdz 3.56


3.50 MP2/cc-pvtz 3.43

Helgaker CBS (cc-pvdz–cc-pvtz) 76.35327 152.71406 3.07 MP2/cc-pvdz 3.95


3.30 MP2/cc-pvtz 3.72

Truhlar CBS //B3LYP/cc-pvtz 76.35335 152.71428 3.58 B3LYP/cc-pvtz 3.92


//B3LYP/aug-cc-pvdz 76.35325 152.71409 3.61 B3LYP/aug-cc-pvdz 3.91

Helgaker CBS //B3LYP/cc-pvtz 76.35335 152.71428 3.12 B3LYP/cc-pvtz 4.37


//B3LYP/aug-cc-pvdz 76.35325 152.71409 3.14 B3LYP/aug-cc-pvdz 4.38

CCSD(T) /cc-pvtz//MP2/cc-pvtz 76.33213 152.67122 2.73 MP2/cc-pvtz 4.77

CCSD(T), Truhlar CBS //B3LYP/aug-cc-pvdz 76.37922 152.76651 3.44 B3LYP/aug-cc-pvdz 4.08


//B3LYP/cc-pvtz 76.37932 152.76668 3.42 B3LYP/cc-pvtz 4.07
//MP2/cc-pvtz 76.37921 152.76642 3.40 MP2/cc-pvdz 4.12

Expt., from ref. 35 3.59 6.93 3.34

references86,90,91 that discuss the specific contribution to The explanation of the results marked as (iii) is quite
ZPE or thermal correction to the enthalpy which came from obvious. One may see that the geometrical parameters predicted
anharmonicity. For example, a recent study of anharmonic- at the MP2/cc-pvtz, and B3LYP/cc-pvtz levels are virtually the
ity effects on zero point energies in weakly bonded molecu- same (see Fig. 1).
lar clusters such as (Ne)n (HF)2, (H2O)2, Ar-H2O, etc.,
finally concluded that ‘‘the potential surfaces of weakly Interaction Entropies and Interaction Gibbs Free Energies
bound complexes often display considerable anharmonicity,
but there can be offsetting effects that make for only small Since the value of DGint can be significantly different when
differences between harmonic and fully anharmonic ZPE.’’86 compared to DHint because of significant contribution of TDSint,
In particular, the difference between calculated harmonic an accurate prediction of DSint is of the same importance as the
and diffusion quantum Monte-Carlo anharmonic ZPE prediction of DHint. However, as already mentioned, the avail-
amounts to 120 cm1 (0.34 kcal/mol) for a water dimer. able experimental data on DSint are rather unreliable. This fact
iii. There is an economic way to reach experimental accuracy limits theoretical estimations also, since theoretical and experi-
in the calculation of a DHint. One may optimize the geome- mental values could not be compared accurately.
try of the intermolecular complex at the B3LYP/cc-pVTZ The literature survey suggests the presence of two types of
level with a counterpoise correction and then perform single data. Some of them are those where the correspondence between
point MP2/cc-pvdz and MP2/cc-pvtz calculations and theoretical and experimental results is good or very good. In
extrapolate the obtained values to the CBS. As follows other words, these calculations reach experimental accuracy.
from the data presented in Tables 2–4, the obtained results Several examples are listed in Table 5. Another group of data
are virtually the same as the ones obtained for the optimiza- (see Table 5 again) relates to the poor correspondence between
tion at the MP2 level using the same extrapolation to the the experimental and calculated results. Comparing the data pre-
CBS. sented in Table 5, one would note that there is both good and

Journal of Computational Chemistry DOI 10.1002/jcc


Efficient and Accurate Prediction of Gibbs Free Energy 1603

Table 3. Calculated Total Energies E (in a.u.) and Thermodynamical Parameters (in kcal/mol) for Formic
Acid Dimer Interaction at 298 K.

E (HCOOH) E ((HCOOH)2) DHint TDSint DGint

B3LYP /cc-pvdz 189.77327 379.57018 13.39 11.07 2.32


/cc-pvtz 189.84066 379.70523 13.56 11.08 2.48
/aug-cc-pvdz 189.79537 379.61473 13.61 11.50 2.11

BLYP /cc-pvdz 189.73750 379.49700 12.24 11.24 1.34


/cc-pvtz 189.80885 379.63978 12.51 11.08 1.43

MP2 /cc-pvdz 189.28132 378.58172 10.17 10.58 0.41


/cc-pvtz 189.47113 378.96511 12.53 10.58 1.94

Truhlar CBS (cc-pvdz–cc-pvtz) 189.58122 379.18959 15.23 MP2/cc-pvdz 4.65


15.23 MP2/cc-pvtz 4.64

Helgaker CBS (cc-pvdz–cc-pvtz) 189.54459 379.11410 13.84 MP2/cc-pvdz 3.25


MP2/cc-pvtz 3.25

Truhlar CBS //B3LYP/cc-pvtz 189.58189 379.19094 15.60 B3LYP/cc-pvtz 4.52


//B3LYP/aug-cc-pvdz 189.58164 379.19034 15.53 B3LYP/aug-cc-pvdz 4.03

Helgaker CBS //B3LYP/cc-pvtz 189.54479 379.11462 14.26 B3LYP/cc-pvtz 3.18


//B3LYP/aug-cc-pvdz 189.54459 379.11411 14.19 B3LYP/aug-cc-pvdz 2.69

CCSD(T) /cc-pvtz//MP2/cc-pvtz 189.49950 379.02281 13.14 MP2/cc-pvtz 2.55

CCSD(T), Truhlar CBS //B3LYP/aug-cc-pvdz 189.60319 379.22749 11.79 B3LYP/aug-cc-pvdz 0.29


//B3LYP/cc-pvtz 189.60347 379.22932 12.59 B3LYP/cc-pvtz 1.51
//MP2/cc-pvtz 189.60439 379.23248 13.43 MP2/cc-pvdz 2.85

Expt, from ref. 36 Via peak height 14.860.5 10.860.6 4.0 61.1
Via band area 13.760.7 2.9
14.161.5 10.7 3.4

poor correspondences among complexes having very similar data presented in ref. 103 for a water dimer at 373 K suggest
structures. It might suggest that some experimental data are just that the difference between the TDSint component that includes
not accurate enough. anharmonic frequencies and the one that includes frequencies
There are six degrees of freedom that are mostly affected by calculated in a harmonic approximation is between 0.56 and
the formation of intermolecular complexes. Three of those are 0.41 kcal/mol. Indeed, an analysis of our data (Tables 2–4) sug-
the relative translational motions of the two monomers. The gests a quite optimistic conclusion: in all considered cases the
other three are rotational degrees of freedom corresponding to MP2 counterpoise-corrected optimization is able to provide the
their relative orientation; three other rotational degrees of free- quality of the TDSint term which is of virtually the same accu-
dom, corresponding to the overall rotations of the complex, do racy as the experimental data. This indicates that for the studied
not affect the association. Upon complex formation these six complexes the rigid rotor-harmonic oscillator-ideal gas approxi-
degrees of freedom convert into ‘‘oscillations’’ (vibrations and/or mation, which is used for the estimation of entropy values at the
hindered rotations) within the potential well of the complex. In harmonic approximation level, is accurate enough for achieving
general, these oscillations are ‘‘soft,’’ i.e., their frequencies are the goals of the present study.
low, with average energies per mode of the order of kT, where k Concluding the discussion on DSint, we would like to empha-
is the Boltzmann’s constant and T the absolute temperature. size that in our study the values of the TDSint term evaluated at
Therefore, we need to return to an analysis of the accuracy of the DFT/B3LYP level are practically of the same quality as the
the vibrational component, since in contrast to ZPE, mostly low ones obtained at the MP2 level; that is, they differ at most by
frequencies will contribute to the TDSint value. about 0.50 kcal/mol when compared with the experimental data
Generally, the situation is quite similar as discussed earlier: for a water dimer (in this case the MP2 data are virtually the
there are only a few papers discussing the difference in the con- same as the experimental data), about 0.3–0.7 kcal/mol for the
tribution of harmonically and anharmonically calculated frequen- formic acid dimer (the MP2 data provide the largest difference
cies in DSint.85,90 The estimations that we performed using the in about 0.1–0.2 kcal/mol), and about 0.8 kcal/mol (the same

Journal of Computational Chemistry DOI 10.1002/jcc


1604 Isayev, Gorb, and Leszczynski • Vol. 28, No. 9 • Journal of Computational Chemistry

þ
Table 4. Calculated Total Energies E (in a.u.) and Thermodynamical Parameters (in kcal/mol) for HCN. . .Li
Interaction at 298 K.

E (HCN) E (Liþ) E(HCN. . .Liþ) DH TDS DG

B3LYP /cc-pvdz 93.43018 7.28511 100.77261 35.25 6.89 28.36


/cc-pvtz 93.46137 7.28538 100.80634 36.58 7.00 29.58
/aug-cc-pvdz 93.43705 7.28514 100.78036 35.70 6.99 28.71

BLYP /cc-pvdz 93.40687 7.27787 100.74100 34.59 6.86 27.73


/cc-pvtz 93.44037 7.27814 100.77681 35.78 6.97 28.80

MP2 (full) /cc-pvdz 92.87848 7.23612 100.17497 33.00 6.88 26.12


/cc-pvtz 93.28116 7.24726 100.58442 34.31 6.96 27.35

Truhlar CBS (cc-pvdz–cc-pvtz) 93.34581 7.25481 100.65798 35.22 MP2/cc-pvdz 28.33


35.15 MP2/cc-pvtz 28.20

Helgaker CBS (cc-pvdz–cc-pvtz) 93.32345 7.25187 100.63219 34.92 MP2/cc-pvdz 28.03


34.86 MP2/cc-pvtz 27.90

Truhlar CBS //B3LYP/cc-pvtz 93.34761 7.25482 100.66008 35.37 B3LYP/cc-pvtz 28.37


//B3LYP/aug-cc-pvdz 93.34746 7.25482 100.65984 35.29 B3LYP/aug-cc-pvdz 28.33

Helgaker CBS //B3LYP/cc-pvtz 93.32398 7.25186 100.63297 35.04 B3LYP/cc-pvtz 28.04


//B3LYP/aug-cc-pvdz 93.32396 7.25186 100.63288 34.96 B3LYP/aug-cc-pvdz 28.01

CCSD(T) (full) /cc-pvtz//MP2 cc-pvtz 93.30332 7.249353 100.60880 34.39 MP2/cc-pvtz 27.43

CCSD(T) (full), Truhlar CBS //B3LYP/cc-pvtz 93.36434 7.25726 100.67938 35.45 B3LYP/cc-pvtz 28.45
//B3LYP/aug-cc-pvdz 93.36397 7.25726 100.67893 35.38 B3LYP/aug-cc-pvdz 28.39
//MP2/cc-pvtz 93.36416 7.25726 100.67912 35.38 MP2/cc-pvtz 28.42

Expt., from ref. 37 36.4 7.7 28.7

difference has been obtained at the MP2 level) for the culation of frequencies at the B3LYP/cc-pvtz level are virtually
HCN. . .Liþ system. the same as the ones using the MP2 optimization. However,
Next we consider the accuracy of the parameter that is gener- optimizations at the B3LYP/aug-ccpvdz level followed by the
ally responsible for the stability of intermolecular complexes, Truhlar extrapolation and calculations of frequencies at the
the Gibbs free energy. Since the best results in the estimation of B3LYP/aug-ccpvdz level result in virtually the same result as
DHint have been obtained using extrapolation of total energy to the experimental value of DGint. The results based on Helgaker
the CBS, we have obtained also quite a satisfactory accuracy of extrapolation [see eq. (3)] are also quite accurate.
DGint using the same approximation. One may see that our best The tendencies that we obtained for analyzed data for water
estimated value for a water dimer differs by only 0.09 kcal/mol and formic acid dimers are confirmed by the results obtained for
from experimental data. As follows from the data presented in the HCN. . .Liþ system. The most accurate (the difference
Table 2, this result has been obtained for the procedure that between the experimental and calculated values is 0.67 kcal/
included the MP2 geometry optimization followed by the CBS mol) are the data that we obtained within the MP2 geometry
interpolation using the Truhlar formula [see eq. (2)] and the optimization followed by CBS interpolation using the Truhlar
MP2/cc-pvtz data for the calculations of vibrational frequencies. formula and the MP2/cc-pvtz data for calculations of vibrational
The deviation of 0.58 kcal/mol has been obtained for the pro- frequencies. Virtually the same results were reached using a ge-
cedure that included geometry optimizations at the B3LYP/cc- ometry optimization at the B3LYP/cc-pvtz level followed by
pvtz level followed by CBS interpolation using the Truhlar for- CBS interpolation using the Truhlar formula and calculations of
mula and calculations of frequencies at the B3LYP/cc-pvtz level. frequencies at the B3LYP/cc-pvtz level. The results obtained
The deviation of 0.65 kcal/mol has been reached for the for- with the aid of the Helgaker technique are almost as accurate as
mic acid dimer using the MP2 geometry optimization followed described earlier.
by CBS interpolation using the Truhlar formula and the MP2/cc- Again, the results in the prediction of DGint obtained at the
pvtz data for calculations of vibrational frequencies. The results MP2 counterpoise-corrected optimization level followed by
of the geometry optimization at the B3LYP/cc-pvtz level fol- extrapolation to the CBS values are practically the same as those
lowed by CBS interpolation using the Truhlar formula and cal- that include the counterpoise-corrected optimization at the B3LYP

Journal of Computational Chemistry DOI 10.1002/jcc


Efficient and Accurate Prediction of Gibbs Free Energy 1605

Figure 1. Optimized geometries (bond length in angstroms and angles in degrees) of considered com-
plexes at (a) MP2/cc-pVTZ CP, (b) B3LYP/cc-pVTZ CP, (c) BLYP/cc-pVTZ CP, and (d) CPMD-
BLYP/PW.

level with further extrapolation of the total energy to the CBS at Molecular Dynamics Simulations
the single point MP2 level. Therefore, we expect that such an
Let us start the presentation of the Carr-Parrinello molecular dy-
approach when applied for the intermolecular complexes of namics (CPMD) data from the analysis of interaction energies
moderate or large size will be able to provide a deviation in ac- which are presented in Table 6. We ran a series of simulations
curacy of DGint within less than 1 kcal/mol. for 2, 5, and 10 ps. As was expected, the results are very con-

Table 5. DS (in cal/mol K) of Selected Intermolecular Complexes (Comparison Theoretical Estimations and
Experiment).

System Method Ref., theor. DSint(theor) DSint(expt) Ref., expt

Good correspondence
(H2O)2 MP2/aug-cc-pVTZ 93 7.82 6.93 35
MP2/aug-cc-pVTZ, anharm. 93 7.41
(HF)2 B3LYP/cc-pVDZ 94 20.86 20.71 94
B3LYP/cc-pVTZ 94 21.41
CH3OH. . .Naþ MP2(full)/6-311þG(2d,2p)//MP2/6-31G* 75 21.8 20.5 62 75
CH3NH2. . .Naþ MP2(full)/6-311þG(2d,2p)//MP2/6-31G* 75 23.4 22.6 62 75
CO(CO2)2Hþ. . .CO2 MP2/6-31G(d,p) 95 22.8 22.7 96
(CO)2Hþ. . .CO MP2/6-31G(d,p) 95 22.1 24 97
G2(MP2) 98 15.9 16.0 99
G2(MP2) 98 17.9 17.3 99
MP2/6-311þþG(2df,2pd) 100 24.15 20.163 101

Poor correspondence
OCHþ. . .CO2 MP2/6-31G(d,p) 95 15.8 21.4 96
(CO)2Hþ. . .CO MP2/6-31G(d,p) 95 22.1 15.0 102
G2(MP2) 98 16.2 13.4 99
MP2/6-311þþG(2df,2pd) 100 13.95 22.363 101

Journal of Computational Chemistry DOI 10.1002/jcc


1606 Isayev, Gorb, and Leszczynski • Vol. 28, No. 9 • Journal of Computational Chemistry

Table 6. Interaction (De)and Free (DA) Energies of Formations (in kcal/mol) Evaluated from CPMD and
Comparison with Available Experimental Data.

De DA DG

System Timescale (ps) MDa TIb PW opt. Expt. TI DAþZPE Expt.

(H2O)2 2 4.0 3.95 4.3 5.0 6 0.1c 4.1 5.7 3.34c


5 4.1
10 4.2

(HCOOH)2 2 11.5 13.1 14.1 13.2d 2.3 0.8 3.4 to 4.0e


5 11.8
10 11.6

HCN. . .Liþ 2 35.5 34.8 36.8 36.4 (DH0)f 26.9 26.1 28.7f
5 35.0 35.9 (De)g
10 

a
MD ¼ results from unconstrained MD.
b
TI ¼ Thermodynamic Integration via internal energy change.
c
From ref. 35.
d
From ref. 104.
e
From ref. 36.
f
From ref. 37.
g
Our estimate (Truhlar 2-3 CBS MP2), this work.

sistent and have deviations between them of 0.5 kcal/mol or First, the quality of the thermodynamic integration has been
less. Since the experimental data on De are not available for the monitored by the constrained force convergence. To analyze the
HCN. . .Liþ complex the value included in Table 6 represents convergence of the constrained force, i.e., the minimum time to
our best estimate calculated at the MP2 level with extrapolation obtain trustworthy values of the average force at particular points
to CBS. It was found that the average error in the De values is of the reaction coordinate, the running average of the Lagrangian
about 0.8 kcal/mol for the water dimer and about 0.9 kcal/mol multiplier hli (see Fig. 2a) is plotted. As can be seen, the force
for the HCN. . .Liþ complex. The error in the De estimation for convergence of the average value is less than 800 fs. In addition
the formic acid dimer of 1.1 kcal/mol is found to be the largest the constraint force autocorrelation function h(0)l(t)i (Fig. 2b)
among all selected molecules. Similar results were obtained for relaxes extremely fast. Since dynamical properties like the auto-
evaluation of De using thermodynamic integration (making the correlation function are very sensitive to changes taking place in
assumption that the internal energy change DhEi during com- the system, they can serve as a convenient criterion to monitor
plexation is equal to the interaction energy) except for the for- the dissociation process. Figure 2 represents just one snapshot for
mic acid dimer where the difference between experimental and the water dimer at r(O O) ¼ 3.5 A; however, all other systems
calculated values is only 0.1 kcal/mol. exhibit the same behavior. Analysis indicates that, while the error
In addition, we have calculated the geometrical parameters decreases with increases in the time steps, any trajectory over
and interaction energies from conventional BLYP/PW geometry 10,000 time steps will usually give convergence to 1 kcal/mol (or
optimizations. The resulting values are listed in Figure 1 and Ta- less). More steps might be needed for considerably large systems
ble 6. As can be seen, the obtained results exhibit practically the or complex reaction coordinates.
same accuracy as the results of the CPMD simulations. Devia- Second, since in the CPMD approach the nuclei are treated
tions among BLYP/plane waves and BLYP/cc-pVTZ are smaller as classical particles, a CPMD free energy is not corrected by
that 0.005 Å for the covalent bonds and within 0.025 Å for the the ZPE value. Therefore, to compare them with the experimen-
weak intermolecular bonds. Differences between BLYP/PW and tal values and the values obtained in previous sections, they
MP2 are within acceptable limits, thus BLYP inter- and intramo- would have to be adjusted by ZPEs and thermal corrections.
lecular bond lengths are deviated up to 0.04 and 0.02 Å when Currently, there is no capable implementation to calculate accu-
compared to MP2 results respectively. Differences between bond rate enough harmonic vibrational frequencies and thermal cor-
angles are within 18 for the all approaches. On the other hand, rections in CPMD. Therefore, we have adjusted the values of
the accuracy of the interaction energies is very close to the static DA by the corresponding calculated (BLYP/cc-pVTZ) values of
results that use the BLYP functional and the Gaussian type tri- thermal correction (Table 6; column marked as DAþZPE).
ple- basis sets (for comparison see the results presented in Analyses of adjusted DA values demonstrate rather poorly the
Tables 2–4). accuracy, with deviations between the calculated and experimen-
Before discussing the CPMD results on the free energy esti- tal free energies for all three systems (about 2.5 kcal/mol). This
mation we would such as to highlight two points. difference is higher than the one for unadjusted DA since the

Journal of Computational Chemistry DOI 10.1002/jcc


Efficient and Accurate Prediction of Gibbs Free Energy 1607

for the species that represent the complexes having single and
double hydrogen bonds and ion–molecular interactions ((H2O)2,
(HCOOH)2, and HCN. . .Liþ) suggests that the MP2 level of ab
initio theory is able to reproduce accurately not only an enthalpy
of the intermolecular interaction but also the Gibbs free energy.
To reach such accuracy, the values of total energies should be
extrapolated to the CBS. Accuracy that is better than 1 kcal/mol
has been obtained using the vibrational frequencies calculated
within the harmonic approximation. It is also important to
emphasize that optimizations at the B3LYP/cc-pvtz level result
in virtually the same geometrical parameters as the one per-
formed at the MP2/cc-pvtz level. Therefore, we recommend the
geometry optimization at the B3LYP/cc-pvtz level followed by
single point MP2 calculations and extrapolation obtained total
energy to CBS value to obtain accurate thermodynamic parame-
ters for the systems of medium and large size.
We found that anharmonic contribution to the entropic and
thermal terms is small for three considered systems. Thus, it can
be safely neglected even for high accuracy calculation of ther-
modynamical parameters. This assumption, in principle, can be
extended to the all similar intermolecular system. However, to
generalize this conclusion one should study extensive set of
intermolecular complexes.
The interaction energy (De) and free energy change (DA) for
considered systems have been obtained from Car-Parrinello ab
initio MD simulations. We found that the accuracy of CPMD
simulations is lower than that of the data obtained at the MP2
level. The reason is, in our opinion, in utilizing the BLYP func-
tional, which is not capable of treating the intermolecular inter-
actions as accurately as does the MP2 theory.

Acknowledgments

We thank the Mississippi Center for Supercomputer Research


(MCSR) and Army High Performance Computer Research Cen-
ter (AHPCRC) for a generous allotment of computer time. We
thank Dr. Axel Kohlmeyer for useful suggestions and stimulat-
Figure 2. (a) Fluctuations of the constraint force and running aver-
ing discussions concerning molecular dynamics. We also thank
age of hli for water dimer  ¼ 3.5 E; (b) The constraint force auto-
correlation function (ACF) h(0)l(t)i for the same point. [Color Dr. David Close for reading the manuscript.
figure can be viewed in the online issue, which is available at
References
www.interscience.wiley.com.]
1. Leszczynski, J.; Goodman, L.; Kwiatkowski, J. Theor Chem Acc 1997,
contribution of the thermal correction is positive. Therefore we 97, 195.
conclude that all calculated free energy values are higher than the 2. Bak, K. L.; Gauss, J.; Jørgensen, P.; Olsen, J.; Helgaker, T.; Stan-
experimental values, i.e., this approach systematically underesti- ton, J. F. J Chem Phys 2001, 114, 6548.
mates the stability of selected species. This tendency is also typi- 3. Xantheas, S. S.; Dunning, T. H., Jr. J Chem Phys 1993, 99, 8774.
cal for the conventional BLYP approach since both techniques uti- 4. Xantheas, S. S. J Chem Phys 1996, 104, 8821.
lize the same functional. The performance of CPMD for the free 5. Kohn, W. Rev Mod Phys 1999, 71, 1253.
energy estimation is also worse than for De, since more statistical 6. Laarsonen, K.; Sprik, M.; Parrinello, M.; Car, R. J Chem Phys
1993, 99, 9080.
uncertainties contribute to its value. On the other hand, the accu-
7. Mattison, A. E.; Schultz, P. A.; Desjarliais, M. P.; Mattsson, T. R.;
racy of the obtained thermodynamic parameters is in agreement
Leung, K. Modelling Simul Mater Sci Eng 2005, 13, R1.
with previous estimates105,106 for bulk water and other species. 8. Civaleri, B.; Garrone, E.; Ugliengo, P. J Mol Struct (THEOCHEM)
1997, 419, 227.
9. Rankin, K. N.; Boyd, R. J. J Comput Chem 2002, 22, 1590.
Conclusions 10. Hobza, P.; Sponer, J.; Reschel, T. J Comput Chem 1995, 11, 1315.
11. Nova, J. J.; Sosa, C. J Chem Phys 1999, 99, 15837.
Analysis of the calculated values of the thermodynamical param- 12. Muller, A.; Losada, M.; Leutwyler, S. J Phys Chem A 2004, 108,
eters (DHint, DSint, and DGint) of the intermolecular interaction 157.

Journal of Computational Chemistry DOI 10.1002/jcc


1608 Isayev, Gorb, and Leszczynski • Vol. 28, No. 9 • Journal of Computational Chemistry

13. Strassner, T.; Taige, M. A. J Chem Theory Comput 2005, 1, 848. J.; Gomperts, R.; Stratmann, R. E.; Yazyev, O.; Austin, A. J.;
14. Lynch, B. J.; Fast, P. L.; Harris, M.; Truhlar, D. G. J Phys Chem A Cammi, R.; Pomelli, C.; Ochterski, J. W.; Ayala, P. Y.; Morokuma,
2000, 104, 4811. K.; Voth, G. A.; Salvador, P.; Dannenberg, J. J.; Zakrzewski, V. G.;
15. Hobza, P.; Sponer, J. Chem Rev 1999, 99, 3247. Dapprich, S.; Daniels, A. D.; Strain, M. C.; Farkas, O.; Malick, D.
16. Rappe, A. K.; Bernstein, E. R. J Phys Chem A 2000, 104, 6117. K.; Rabuck, A. D.; Raghavachari, K.; Foresman, J. B.; Ortiz, J. V.;
17. van Mourik, T.; Gdanitz, R. J. J Chem Phys 2002, 116, 9620. Cui, Q.; Baboul, A. G.; Clifford, S.; Cioslowski, J.; Stefanov, B. B.;
18. Kamiya, M.; Tsuneda, T.; Hirao, K. J Chem Phys 2002, 117, Liu, G.; Liashenko, A.; Piskorz, P.; Komaromi, I.; Martin, R. L.; Fox,
6010. D. J.; Keith, T.; Al-Laham, M. A.; Peng, C. Y.; Nanayakkara, A.;
19. Milet, A.; Korona, T.; Moszynski, R.; Kochanski, E. J Chem Phys Challacombe, M.; Gill, P. M. W.; Johnson, B.; Chen, W.; Wong,
1999, 111, 7727. M. W.; Gonzalez, C.; Pople, J. A. Gaussian 03, Revision C.02,
20. Helgaker, T.; Ruden, T. A.; Jørgensen, P.; Olsen, J.; Klopper, W. Wallingford, CT: Gaussian, 2004.
J Phys Org Chem 2004, 17, 913. 48. Simon, S.; Duran, M.; Dannenberg, J. J. J Chem Phys 1996, 105,
21. Klopper, W.; van Duijneveldt-van de Rijt, J. G. C. M.; van Duijne- 11024.
veldt, F. B. Phys Chem Chem Phys 2000, 2, 2227. 49. Truhlar, D. G. Chem Phys Lett 1998, 294, 45.
22. Fox, D. J.; Pople, J. A.; Raghavachari, K.; Curtis, L. A.; Head- 50. Halkier, A.; Helgaker, T.; Jorgensen, P.; Klopper, W.; Koch, H.;
Gordon, M. J Chem Phys 1990, 93, 2537. Olsen, J.; Wilson, A. K. Chem Phys Lett 1998, 286, 243.
23. Trucks, G. W.; Pople, J. A.; Raghavachari, K.; Curtiss, L. A. J 51. Atkins, P. W. Physical Chemistry; Oxford University Press:
Chem Phys 1991, 94, 7221. Oxford, 1994.
24. Pople, J. A.; Raghavachari, K.; Curtiss, L. A.; Redfern, P. C.; Ras- 52. Car, R.; Parrinello, M. Phys Rev Lett 1985, 55, 2471.
solov, V. J Chem Phys 1999, 110, 4703. 53. CPMD, version 3.9.1; Copyright IBM Corp. (1990–2005), Copy-
25. Ochterski, J. W.; Petersson, G. A.; Montgomery, J. A., Jr. J Chem right MPI fur Festkorperforschung Stuttgart, 1997–2001.
Phys 1996, 104, 2598. 54. Martyna, G. J.; Tuckerman, M. E. J Chem Phys 1999, 110, 2810.
26. Montgomery, J. A., Jr.; Frisch, M. J.; Ochterski, J. W.; Petersson, 55. Kleinman, L.; Bylander, D. M. Phys Rev Lett 1982, 48, 1425.
G. A. J Chem Phys 2000, 112, 6532. 56. Troullier, N.; Martins, J. Phys Rev B 1991, 43, 1993.
27. Martin, J. M. L.; de Oliveira, G. J Chem Phys 1999, 111, 1843. 57. Goedecker, S.; Teter, M.; Hutter, J. Phys Rev B 1996, 54, 1703.
28. Boese, A. D.; Oren, M.; Atasoylu, O.; Martin, J. M. L. J Chem Phys 58. Martyna, G. J.; Klein, M. L.; Tuckerman, M. J Chem Phys 1992,
2004, 120, 4129. 97, 2635.
29. Tajti, A.; Szalay, P. G.; Csaszar, A. G.; Kallay, M.; Gauss, J.; 59. Hutter, J.; Lüthi, H. P.; Parrinello, M. Comput Mater Sci 1994, 2, 244.
Valeev, E. F.; Flowers, B. A.; Vazquez, J.; Stanton, J. F. J Chem 60. Sprik, M.; Ciccotti, G. J Chem Phys 1998, 109, 7737.
Phys 2004, 121, 11599. 61. Rappe, A. K.; Bernstein, E. R. J Phys Chem A 2000, 104, 6117.
30. Gorb, L.; Podolyan, Y.; Dziekonski, P.; Sokalski, W. A.; Leszc- 62. Hobza, P.; Havlas, Z. Theor Chem Acc 1998, 99, 372.
zynski, J. J Am Chem Soc 2004, 126, 10119. 63. Xu, X.; Goddard, W. A., III. J Phys Chem A 2004, 108, 2305.
31. Becke, A. D. Phys Rev A 1988, 38, 3098. 64. Hobza, P.; Bludský, O.; Suhai, S. Phys Chem Chem Phys 1999, 1,
32. Lee, C.; Yang, W.; Parr, R. G. Phys Rev B 1988, 37, 785. 3073.
33. Zhao, Y.; Truhlar, D. G. J Phys Chem A 2005, 109, 5656. 65. Bende, A.; Suhai, S. Int J Quant Chem 2005, 103, 841.
34. Møller, C.; Plesset, M. S. Phys Rev 1934, 46, 618. 66. Defrancois, C.; Periquet, V.; Carles, S.; Schermann, J. P.; Adamo-
35. Curtiss, L. A.; Frutrip, D. J.; Blander, M. J Chem Phys 1979, 71, wicz, L. Chem Phys 1998, 239, 475.
2703. 67. Howard, B. J.; Dyke, T. R.; Klemperer, W. J Chem Phys 1984,
36. Clague, A. D. H.; Bernstein, H. J. Spectrochim Acta A 1969, 25, 81, 5417.
593. 68. Gutowsky, H. S.; Chuang, C.; Keen, J. D.; Klots, T. D.; Emilsson,
37. Woodin, R. L.; Beauchamp, J. L. J Am Chem Soc 1978, 100, 501. T. J Chem Phys 1985, 83, 2070.
38. Rodriguez-Gomez, D.; Darve, E. J Chem Phys 2004, 120, 3563. 69. Tsuzuki, S.; Lüthi, H. P. J Chem Phys 2001, 114, 3949.
39. Straatsma, T. P.; McCammon, J. A. Annu Rev Phys Chem 1992, 70. Pine, A. S.; Howard, B. J. J Chem Phys 1986, 84, 590.
43, 407. 71. Haliker, A.; Klopper, W.; Helgaker, T.; Jørgensen, P.; Taylor,
40. Kollman, P. Chem Rev 1993, 93, 2395. P. R. J Chem Phys 1999, 111, 9157.
41. Kelli, E.; Seth, M.; Ziegler, T. J Phys Chem A 2004, 108, 2167. 72. Tsuzuki, S.; Uchimaru, T.; Matsumura, K.; Mikami, M.; Tanabe,
42. Yang, S.-Y.; Fleurat-Lessard, P.; Hristov, I.; Ziegler, T. J Phys Chem K. J Chem Phys 1999, 110, 11906.
A 2004, 108, 9461. 73. Bizzarri, A.; Stolte, S.; Reuss, J.; van Duijneveldt-van de Rijdt,
43. Den Otter, W. K.; Briels, W. J. J Chem Phys 1998, 109, 4139. J. G. C. M.; van Duijneveldt, F. B. Chem Phys 1990, 143, 423.
44. Carter, E. A.; Ciccotti, G.; Hynes, J. T.; Kapral, R. Chem Phys 74. Wofford, B. A.; Eliades, M. E.; Lieb, S. G.; Bevan, J. W. J Chem
Lett 1989, 156, 472. Phys 1987, 87, 5674.
45. Peter, C.; Oostenbrick, C.; van Dorp, A.; van Gunsteren, W. F. J 75. Hoyau, S.; Norrman, K.; McMahon, T. B.; Ohanessian, G. J Am
Chem Phys 2004, 120, 2652. Chem Soc 1999, 121, 8864.
46. Straatsma, T. P. In Reviews in Computational Chemistry, Vol. 9; 76. Rodgers, M. T.; Armentrout, P. B. Int J Mass Spectrom 1999, 359.
Lipkowitz, K. B.; Boyd, D. B., Eds.; VCH: New York, 1996; p. 81. 185–187.
47. Frisch, M. J.; Trucks, G. W.; Schlegel, H. B.; Scuseria, G. E.; 77. Wu, J.; Polce, M. J.; Wesdemiotis, C. Int J Mass Spectrom, 2001,
Robb, M. A.; Cheeseman, J. R.; Montgomery, J. A., Jr.; Vreven, 204, 125.
T.; Kudin, K. N.; Burant, J. C.; Millam, J. M.; Iyengar, S. S.; 78. Rodgers, M. T.; Armentrout, P. B. J Phys Chem A 1997, 101,
Tomasi, J.; Barone, V.; Mennucci, B.; Cossi, M.; Scalmani, G.; 1238.
Rega, N.; Petersson, G. A.; Nakatsuji, H.; Hada, M.; Ehara, M.; 79. Remko, M.; Rode, B. M. J Mol Struct (THEOCHEM) 2001, 505,
Toyota, K.; Fukuda, R.; Hasegawa, J.; Ishida, M.; Nakajima, T.; 269.
Honda, Y.; Kitao, O.; Nakai, H.; Klene, M.; Li, X.; Knox, J. E.; 80. Lau, J. K.-C.; Wong, C. H. S.; Ng, P. S.; Siu, F. M.; Ma, N. L.;
Hratchian, H. P.; Cross, J. B.; Bakken, V.; Adamo, C.; Jaramillo, Tsang, C. W. Chem Eur J 2003, 9, 3383.

Journal of Computational Chemistry DOI 10.1002/jcc


Efficient and Accurate Prediction of Gibbs Free Energy 1609

81. Davidson, W. R.; Kebarle, P. J Am Chem Soc 1976, 98, 6125. NIST Standard Reference Database No. 101-Release 12; available
82. Ogilvie, J. F.; Wang, F. Y. H. J Mol Struct 1992, 273, 277. at http://srdata.nist.gov/cccbdb.
83. Wieczorek, R.; Haskamp, L.; Dannenberg, J. J. J Phys Chem A 95. Szymczak, J. J.; Roszak, S.; Gora, R. W.; Leszczynski, J. J Chem
2004, 108, 6713. Phys 2003, 119, 6560.
84. Kim, K. S.; Mhin, B. J.; Choi, U.-S.; Lee, K. J Chem Phys 1992, 96. Hiraoka, K.; Yamabe, S. In Dynamics of Excited Molecules in
97, 6649. Studies in Physical and Theoretical Chemistry; Kuchitsu, S., Ed.;
85. Diri, K.; Myshakin, E. M.; Jordan, K. D. J Phys Chem A 2005, Elsevier Science: Amsterdam, 1994; p. 399.
109, 4005. 97. Hiraoka, K.; Mori, T. Chem Phys 1989, 137, 345.
86. Dykstra, C. E.; Shuler, K.; Young, R. A.; Baı̂cić, Z. J Mol Struct 98. Kaczorowska, M.; Roszak, S.; Leszczynski, J. J Chem Phys 2000,
(THEOCHEM) 2002, 591, 11. 113, 3615.
87. Feller, D. J Chem Phys 1992, 96, 6104. 99. Hiraoka, K.; Mori, T. J Chem Phys 1989, 91, 4821.
88. Lüthi, H. P.; Klopper, W. Mol Phys 199, 96, 359. 100. Gora, R.; Roszak, S.; Leszczynski, J. J Chem Phys 2001, 115, 771.
89. Remko, M.; Rode, B. M. J Mol Struct (THEOCHEM) 2000, 505, 101. Hiraoka, K.; Kudaka, I.; Yamabe, S. Chem Phys Lett 1991, 178,
269. 103.
90. Barone, V. J Chem Phys 2004, 120, 3059. 102. Hiraoka, K.; Saluja, P. P. S.; Kebarle, P. Can J Chem 1979, 57,
91. Martin, J. M. L.; de Oliveira, G. J Chem Phys 1999, 111, 1843. 2159.
92. Miller, Y.; Chaban, G. M.; Gerber, R. B. Chem Phys 2005, 313, 103. Munoz-Carom, C.; Nino, A. J Phys Chem A 1997, 101, 4128.
213. 104. Henderson, G. J Chem Educ 1978, 64, 88.
93. Muñoz-Caro, C.; Niño, A. J Phys Chem A 1997, 101, 4128. 105. Sprik, M.; Hutter, J.; Parrinello, M. J Chem Phys 1996, 105, 1142.
94. Johnson, R. D., III, Ed. The NIST Computational Chemistry Com- 106. Vuilleumier, R.; Sprik, M.; Alavi, A. J Mol Struct (THEOCHEM)
parison and Benchmark Database; Gaithersburg, MD: NIST, 2005. 2000, 506, 343.

Journal of Computational Chemistry DOI 10.1002/jcc

You might also like