Article

pubs.acs.org/JPCA

Benchmarking Compound Methods (CBS-QB3, CBS-APNO, G3, G4,

W1BD) against the Active Thermochemical Tables: A Litmus Test for
Cost-Effective Molecular Formation Enthalpies
John M. Simmie* and Kieran P. Somers
Combustion Chemistry Centre & School of Chemistry, National University of Ireland, Galway, Ireland
*
S Supporting Information

ABSTRACT: The theoretical atomization energies of some 45 CxHyOz molecules

present in the Active Thermochemical Tables compilation and of particular interest to
the combustion chemistry community have been computed using five composite model
chemistries as titled. The species contain between 1−8 “heavy” atoms, and a few are
conformationally diverse with up to nine conformers. The enthalpies of formation at 0
and 298.15 K are then derived via the atomization method and compared against the
recommended values. In general, there is very good agreement between our averaged
computed values and those in the ATcT; those for 1,3-cyclopentadiene exceptionally
differ considerably, and we show from isodesmic reactions that the true value for 1,3-
cyclopentadiene is closer to 134 kJ mol−1 than the reported 101 kJ mol−1. If one is
restricted to using a single method, statistical measures indicate that the best methods are
in the rank order G3 ≈ G4 > W1BD > CBS-APNO > CBS-QB3. The CBS-x methods do
on average predict ΔfH⊖(298.15 K) within ≈5 kJ mol−1 but are prone to occasional
lapses. There are statistical advantages to be gained from using a number of methods in tandem, and all possible combinations
have been tested. We find that the average formation enthalpy coming from using CBS-APNO/G4, CBS-APNO/G3, and G3/G4
show lower mean signed and mean unsigned errors, and lower standard and root-mean-squared deviations, than any of these
methods in isolation. Combining these methods also leads to the added benefit of providing an uncertainty rooted in the
chemical species under investigation. In general, CBS-APNO and W1BD tend to underestimate the formation enthalpies of
target species, whereas CBS-QB3, G3, and G4 have a tendency to overestimate the same. Thus, combining CBS-APNO with a
G3/G4 combination leads to an improvement in all statistical measures of accuracy and precision, predicting the ATcT values to
within 0.14 ± 4.21 kJ mol−1, thus rivalling “chemical accuracy” (±4.184 kJ mol−1) without the excessive cost associated with
higher-level methods such as W1BD.

■ INTRODUCTION
The determination of accurate formation enthalpies (ΔfH⊖) of
stable, radical, and ionic species is central to the science of
combustion modeling, serving as essential input information for
a multitude of numerical simulations. Such is the importance of
this fundamental property that great effort has been invested in
their tabulation and refinement, with the JANAF,1 CODATA,2
and Third Millenium databases,3 and the Active Thermochem-
ical Tables (ATcT),4−6 serving the combustion community well
for the last number of decades.
Although experimental determination (primarily calorimetry)
is the ideal method to obtain accurate ΔfH⊖, it is both laborious
and time-consuming. Add to this the number of species which
may comprise a chemical kinetic mechanism7 (100s−1000s),
many of which are radicals which may be experimentally
inaccessible, it is clear that obtaining experimental ΔfH⊖ for all
species of interest in combustion modeling is effectively
impossible. However, it is now accepted that quantum
chemistry, either in isolation or in tandem with experiment,
can offer an effective means with which to obtain such
information on feasible timescales.
The accuracy of any ΔfH⊖ derived from quantum chemistry
depends largely on the model chemistry applied and the size of
the molecular system, with the theoretical thermodynamicist in
constant pursuit of “chemical accuracy”  that is formation
enthalpies computed within 1 kcal mol−1 (4.184 kJ mol−1) of
the true formation enthalpy. The current “gold-standard” in
model chemistries is the coupled-cluster method with
perturbative quadruples, CCSDT(Q), with subsequent extrap-
olation of the computed energies to the basis-set limit; see for
example, work on the C2H5 + O2 reaction.8 This is rarely
achievable, and more practical standards are the CCSD(T)
methods9 with perturbative inclusion of triples. The Wn,10−12
high-accuracy extrapolated ab initio thermochemistry
(HEAT),13−15 focal-point extrapolation,16 and Feller-Peter-
Special Issue: 100 Years of Combustion Kinetics at Argonne: A
Festschrift for Lawrence B. Harding, Joe V. Michael, and Albert F.
Wagner
Received: November 14, 2014
Revised: January 5, 2015
Published: January 12, 2015

© 2015 American Chemical Society 7235 DOI: 10.1021/jp511403a

J. Phys. Chem. A 2015, 119, 7235−7246
The Journal of Physical Chemistry A
son-Dixon17 methods have also shown accuracy to within ≈2 kJ
mol−1 when compared with well-known experimental for-
mation enthalpies. However, despite advances in computer
software and hardware, these methods remain computationally
expensive and are somewhat restricted in their application to
molecules containing some five or so nonhydrogen atoms and,
as a consequence, can rarely treat the full panoply of molecules
which are to be encountered in detailed chemical kinetic
models.7
Compound methods offer a cost-effective alternative to the
above, albeit at the expense of accuracy. These methods consist
of a series of less expensive geometry optimization, frequency,
and single-point energy calculations with empirical corrections
used to overcome deficiencies when the methods are
benchmarked against formation enthalpies of well-known
atomic and small-molecule standards. Numerous compound
methods have entered widespread use within the combustion
community in recent times; in particular, the CBS-x18,19 (x =
QB3 and APNO) and Gaussian-x20−24 (x = 1, 2, 3, and 4)
methods are regularly employed in thermodynamic and kinetics
computations of relevance to combustion and atmospheric
chemistry.
Variants of the G3 method, such as G3B325 and G3MP226
are also popular, with the former frequently employed for ΔfH⊖
computations by Burcat et al. as part of the Third Millennium
Database, with an uncertainty of ±8 kJ mol−1 typically quoted
therein when they have applied this method.
“Standard” methods have been tweaked on numerous
occasions to remedy perceived difficulties and/or with the
aim of improving the computational cost; for example, variants
of G4,27,28 of W129,30 and W3,31,32 etc. It appears that divergent
evolution is the order of the day.
In order to overcome the large uncertainties in ΔfH⊖ when
compound methods are employed, the isodesmic reaction
method33 is frequently employed to reduce errors and
uncertainties in the computation. However, a number of
problems arise when employing this method. Well-known
formation enthalpies are required from experiment or high-level
theory to create these hypothetical working reactions; the
computed ΔfH⊖ is very much dependent on the quality of the
isodesmic reaction framed (given that there is no unique
answer), and the computational cost is increased as numerous
quantum chemical calculations must be carried out to
determine the theoretical enthalpy of all the chaperone
molecules. However, provided good working reactions can be
framed the method can deliver excellent results; as an example,
see previous work on cyclic esters.34
The empirical corrections employed in compound methods
also reduce their predictive capability, as the databases against
which they are benchmarked, and any empirical corrections
thus derived tend to consist of a limited number of small
molecule targets. Molecules of interest to combustion scientists
can be large, conformationally complex, species extending well
outside the benchmarked test-set in terms of size and the true
uncertainty in any ΔfH⊖ computed from these methods is
somewhat unknown.
Thus, we arrive at the goals of this work: to benchmark a
range of popular compound methods against a database of well-
known formation enthalpies, namely, the Active Thermochem-
ical Tables of Ruscic.6 This database has been chosen as it is
based on either direct experiment or experiment in
combination with high-level theory and is undergoing constant
development and refinement. A brief description of the
7236
Article
methodology behind ATcT and discussion of uncertainty
quantification in general has been recently given by Ruscic.35
The isodesmic reaction method is not employed in this work
as part of our benchmarking, owing to the problems discussed
above, rather we use the atomization method alone to derive
ΔfH⊖ throughout. Thus, the formation enthalpies derived
herein are representative of the “pure” method, and are
independent of experimental formation enthalpies required for
isodesmic reactions. The results aim to provide an assessment
of the performance of each method against a wide-range of
stable and, in the future, radical species.
Recently, Simmie et al.36 carried out a thermochemical study
of furan derivatives, and they found that a ΔfH⊖ computed by
taking an average ΔfH⊖ from CBS-QB3, CBS-APNO, and G3
atomization energies recreated those computed from well-
framed isodesmic working reactions to within ≈4 kJ mol−1,
despite the large uncertainties associated with the atomization
calculations. The reasons for this interesting behavior were not
investigated at the time, but Somers37 later ascribed it to a
tendency of the G3 method to consistently overpredict ΔfH⊖,
for the CBS-APNO method to consistently underpredict ΔfH⊖,
and for the CBS-QB3 method to arrive at a similar answer to
the isodesmic working reactions.
A second goal of this work is therefore to assess whether a
combination of compound methods can provide a more
rigorous assessment of ΔfH⊖ than any of these methods in
isolation, thus offering a cost-effective alternative to high-level
methods. A primary reason for using two or more model
chemistries, quite apart from the statistical gain, is that the final
result is then less-dependent upon the actual geometrical
structure if the optimization and frequency calculations, inbuilt
into each composite method, differ. Of course if the two
optimizations result in distinctly different geometries and/or
electronic states then this approach would be compromised, but
in our experience this has rarely happened for the species under
consideration here.
■
COMPUTATIONAL METHODOLOGY
Five compound methods have been chosen for benchmarking
as part of this work: CBS-QB3, CBS-APNO, G3, G4, and
W1BD,38 as embedded within the application Gaussian.39
For the computation of 0 K enthalpies of formation, ΔfH0,
we commence by calculating the theoretical atomization
energy, TAE0, for the reaction:
Cx HyOz → x 3C + y 2 H + z 3O
which is given by
TAE0 = xH0(3C) + yH0(2 H) + zH0(3O) − H0(Cx HyOz )
(1)
where H0 is the 0 K enthalpy of an atom or molecule and is the
sum of the electronic and zero-point energies. Zero-point
energies are automatically computed, adjusted by a built-in
scale factor and added to the 0 K electronic energy by each
compound method as part of its predefined series of
computations. The ΔfH0 of the molecule then follows knowing
the theoretical atomization energy and the experimentally
known formation enthalpies of the component atoms in their
gaseous state:
DOI: 10.1021/jp511403a
J. Phys. Chem. A 2015, 119, 7235−7246
The Journal of Physical Chemistry A
Δf H0(Cx HyOz ) = [xΔf H0(3C) + yΔf H0(2 H)
+ zΔf H0(3O)] − TAE0
⊖
For formation enthalpies at 298.15 K, ΔfH , the same
equations are employed, with thermal enthalpy corrections
added to H0 in eq 1 to account for translational, rotational, and
vibrational degrees of freedom of each molecule and atom.
Standard state formation enthalpies of each gaseous atom,
ΔfH⊖, also replace the 0 K counterparts in eq 2.
Note that the influence of vibrational anharmonicities,
rotational nonrigidities, and torsional degrees of freedom are
not accounted for in the predefined routine of calculations in
any of the compound methods employed, and therefore, in the
final formation enthalpies reported herein. The 0 and 298.15 K
formation enthalpies of atomic species used in the above
equations are given in Table 1 and are adopted from the
ATcT.6
Table 1. Atomic Formation Enthalpies (kJ mol−1)
T (K) C (3P) H (2S1/2) O (3P2)
0 711.38 216.034 246.844
298.15 716.87 217.998 249.229
In those cases where conformational diversity exists the
Gibbs free energies, ΔG⊖ (298.15 K) were determined for each
conformer and the Boltzmann distribution computed, taking
due account of degeneracies, σ (eq 3). The contribution, xi,
made by each conformer to the overall enthalpy of formation
for species X is then calculated from
n
xi = σi exp( −ΔGm⊖(i)/RT )/∑ [σi exp( −ΔGm⊖(i)/RT )]
i=1
n
Δf Hm⊖(X) = ∑ [xiΔf Hm⊖(i)]
i=1
Thus, in the case of n-hexane nine conformers (out of a
possible 12 as shown schematically by Morini et al.)40 with
accompanying degeneracies or symmetry numbers were
identified: ttt(1), gtt(4), tgt(2), tgg(4), gtg(2), g+t+g−(2),
ggg(2), g+x−t+(4), and t+g+x−(4). These give rise to populations
of 17.1, 45.0, 11.8, 13.6, 3.2, 5.9, 1.2, 1.4, and 0.8%, respectively,
from G4 calculations.
Each individual enthalpy of formation was then weighted
with respect to the population to calculate a final value for that
particular molecule (eq 4). In some cases, individual con-
formers are present in the ATcT such as syn- and anti-formic
acid, and in these cases, no distributions are needed.
In general, the literature was consulted for possible
conformers of a particular molecule, but in any event, these
were then checked by using the application Spartan and
utilizing a modest basis set and B3LYP functional to generate
the distributions.41 Only those which contributed to a
significant extent were then retained for further computation.
■ RESULTS AND DISCUSSION
The results at 0 K are shown in Table 2; although interesting,
the primary focus of our attention will be on the results at
298.15 K, which are much more important for chemical kinetic
simulations and modeling purposes.
Article
298.15 K. We display in Table 3, the results of computed
formation enthalpies for all five model chemistries in
(2) comparison to ATcT values.
Problem Molecules. Some molecules in the ATcT
compendium are problematic in the sense that the composite
methods under discussion here are unable to attain a
satisfactory solution to their structure. Thus, for example,
oxirene, a three-membered heterocyclic ring containing a CC
bond resists optimization (or rather optimizes to a structure
with an imaginary frequency) with the B3LYP functional, which
is of course an essential component of CBS-QB3, G4, and
W1BD. The basis sets used by these three methods are
predefined at 6-311G(2d,d,p), 6-31G(2df,p), and cc-pVTZ+d;
note that using a 6-311++G(d,p) basis set leads to a successful
convergence but just not to a C2v structure.
This molecule has a long and checkered history; in a telling
phrase it has been described as “hovering on the edge of
reality”, or it has been subtitled with a Shakespearean quotation
“to be or not to be”, with some theoretical methods concluding
that it is a saddle point and others a local minimum on the
potential energy surface.42,43 Recent extremely expensive
coupled-cluster interference-corrected explicitly correlated
second-order perturbation theory results44 ascribe a total
atomization energy of 1827.8 kJ mol−1 at 0 K, which implies
an enthalpy of formation of 277.0 kJ mol−1, whereas equally
expensive W4 calculations45 give 272.1 and 268.2 kJ mol−1 at 0
and 298.15 K, respectively. Viewed from this perspective, the
CBS-APNO and G3 0 K results of 274.5 and 275.4 kJ mol−1,
respectively, are surprisingly good.
We should also take a moment to note if there are any
considerable differences between any of our computations and
those values recommended in the ATcT. The most obvious
example is 1,3-cyclopentadiene where the ATcT value of
ΔfH⊖(298.15K) = 101.30 ± 2.5 differs considerably from our
result of 137.1 ± 6.7 kJ mol−1. In the compendium
(3)
Thermochemical Data of Organic Compounds, Pedley et al.
quote46 a value of 134.3 ± 1.5 kJ mol−1, whereas the NIST
WebBook includes determinations47,48 of 139 and 133.4 kJ
(4) mol−1. In a thermochemical and kinetic analysis of addition
reactions to 1,3-cyclopentadiene Zhong and Bozzelli49 use a
value of 131 kJ mol−1. More recent photoacoustic spectroscopy
and quantum chemical studies by Agapito et al.50 imply a heat
of formation of 134 kJ mol−1.
From a consideration of isodesmic reactions involving (i)
furan, propane and dimethyl ether, (ii) ethyne, 2 molecules of
ethene and methane as chaperones, it can be shown that the
computed reaction enthalpies of −90.30 ± 1.58 and 275.53 ±
4.71 kJ mol−1 lead to values for ΔfH (298.15 K) of 135.21 ±
1.80 and 132.47 ± 4.76 kJ mol−1, respectively, from which a
grand-weighted average of 134.9 ± 1.7 kJ mol−1 results. Thus, it
appears that the ATcT value for 1,3-cyclopentadiene is
incorrect.
There are similar, but less severe, disparities for oxalic acid.
We note that an isodesmic reaction:
(COOH)2 + C2H6 = 2CH3COOH
can be framed whose average reaction enthalpy is −43.86 ±
0.96 kJ mol−1, from which the predicted formation enthalpy of
the dominant aa conformer of oxalic acid is −739.8 ± 1.2 kJ
mol−1 where we have adopted the ATcT numbers for the
chaperones ethane and cis acetic acid. This is in very good
agreement with the atomization value of −735.5 ± 5.3 kJ mol−1
for the same conformer, thus indicating that the ATcT value for
7237 DOI: 10.1021/jp511403a
J. Phys. Chem. A 2015, 119, 7235−7246
The Journal of Physical Chemistry A Article

Table 2. Final Computed Formation Enthalpies of Compounds at 0 K/kJ mol−1

species QB3 APNO G3 G4 W1BD ATcT
1,3-butadiyne 468.41 462.06 462.03 458.61 458.08 458.39
1,3-cyclopentadiene 161.39 148.19 155.65 153.43 144.16 118.20
1-butene (gauche) 28.49 15.49 22.23 23.05 15.39 20.30
1-butene (cis) 29.32 16.30 23.07 24.11 15.85 20.93
1-butyne 188.37 178.03 181.67 181.33 176.13 178.97
2,2,4-trimethylpentane −147.58 −178.69 −159.19 −156.69 −168.22
gauche −160.09 −191.51 −171.70 −169.26 −181.87 −171.20
2,3-butanedione −315.19 −319.76 −313.36 −311.53 −318.11 −310.25
2-butyne 165.90 155.80 160.26 160.51 154.99 158.72
2-propanol (gauche) −249.47 −257.53 −249.86 −247.65 −257.09
2-propanol (trans) −247.99 −256.49 −248.78 −246.28 −255.95
acetic acid (cis) −422.67 −422.63 −418.61 −416.90 −423.43 −419.59
acetone −200.49 −−207.47 −200.81 −199.61 −206.14 −199.39
allene 202.25 196.48 196.62 197.07 194.26 197.63
benzene 107.21 91.80 103.83 102.26 92.65 100.70
carbon monoxide −115.57 −113.78 −114.71 −117.59 −114.07 −113.81
cis-2-butene 21.79 8.40 16.17 16.91 8.30 13.94
cyclobutene 186.71 175.51 183.12 180.29 172.02 173.90
cyclohexane −74.02 −97.33 −82.55 −80.66 −93.44 −82.52
cyclopropane 76.69 65.64 73.82 71.92 65.95 70.76
cyclopropene 298.87 292.28 295.82 293.58 288.91 292.64
dimethyl ether −171.42 −173.81 −167.66 −166.80 −171.77 −166.51
dioxirane 0.42 5.71 9.70 6.86 8.66 9.00
ethanal −156.89 −158.95 −155.71 −155.49 −158.85 −154.98
ethane −66.45 −74.99 −69.15 −67.35 −73.12 −68.13
ethene 64.11 59.75 60.55 60.84 58.11 61.08
ethylene glycol no. 1 −375.41 −374.64 −370.04 −368.17 −378.03
ethylene glycol no. 2 −373.05 −372.87 −367.84 −365.49 −376.18
ethylene glycol no. 3 −372.82 −376.46 −366.17 −364.65 −375.53
ethyne 234.82 233.54 230.90 229.18 228.61 228.89
ethynol 95.86 97.37 95.93 95.82 92.19 94.80
formaldehyde −110.24 −107.06 −107.15 −108.04 −107.36 −105.33
formic acid (syn) −377.05 −372.59 −371.39 −371.35 −374.83 −371.62
formic acid (anti) −360.12 −356.14 −354.84 −354.93 −358.01 −355.28
glyoxal (trans) −214.26 −209.65 −209.62 −210.61 −210.77 −206.85
glyoxal (cis) −194.85 −190.70 −190.25 −192.15 −192.13 −188.42
isobutane −100.31 −116.91 −105.72 −103.79 −113.44 −106.76
isobutene 11.26 −1.83 5.36 5.60 −1.81 3.46
ketene −45.49 −47.19 −47.14 −−45.51 −47.47 −45.47
methane −66.31 −70.91 −67.78 −66.61 −68.60 −66.56
methanediol −385.18 −381.72 −378.43 −378.13 −385.35 −379.21
methanol −193.15 −193.46 −190.06 −189.87 −194.75 −189.82
methyl formate (cis Z) −354.08 −351.59 −348.58 −347.45 −350.69 −344.54
methyl hydroperoxide −123.01 −117.57 −112.96 −114.09 −117.97 −114.99
n-butane (trans) −92.35 −109.03 −97.89 −95.63 −106.21 −98.54
n-butane (gauche) −89.57 −106.14 −95.06 −92.69 −103.17 −98.54
n-hexane ttt −119.90 −144.30 −128.06 −125.53 −140.24
gtt −117.02 −141.50 −125.39 −122.61 −137.14
tgt −117.10 −141.41 −125.38 −122.65 −137.18
tgg −114.34 −139.78 −123.75 −120.02 −134.17
gtg −114.07 −138.68 −122.71 −119.69 −134.04
g+t+g− −113.52 −138.25 −122.20 −119.07 −133.46
ggg −112.14 −138.09 −122.10 −117.97 −131.84
g+x−t+ −107.68 −132.51 −116.30 −113.51 −127.90
t+g+x− −107.19 −132.15 −116.18 −112.80 −127.20
o-benzyne 477.32 463.97 470.99 467.35 464.34 469.60
oxalic acid (c/c) −716.78 −710.04 −707.19 −704.49 −711.63
oxalic acid (t/t) −734.03 −726.40 −724.48 −719.94 −728.90
oxalic acid (c/t) −723.06 −715.92 −713.53 −709.83 −717.78
oxirane (1A1) −43.62 −44.31 −39.64 −41.35 −44.17 −40.17
oxirene (singlet) 270.38 274.54 275.44 269.63 270.16 276.60

7238 DOI: 10.1021/jp511403a

J. Phys. Chem. A 2015, 119, 7235−7246
The Journal of Physical Chemistry A Article

Table 2. continued
species QB3 APNO G3 G4 W1BD ATcT
propane −78.96 −91.57 −82.99 −80.87 −89.26 −82.13
propene 40.32 31.34 35.52 35.95 30.35 35.36
propyne 198.64 191.56 193.73 192.74 189.92 192.86
succinic acid (aAa) −806.20 −805.87 −799.30 −793.72 −806.69
sGs −810.18 −809.03 −802.93 −798.32 −808.41
sGa −804.18 −802.68 −796.99 −792.09 −802.00
gAs −801.13 −800.04 −793.68 −789.33 −800.58
toluene 80.45 60.99 74.99 73.56 63.80 73.69
trans-2-butene 16.56 3.38 10.73 11.40 3.36 9.39
vinyl alcohol (anti) −108.62 −110.86 −109.00 −108.14 −113.89
vinyl alcohol (syn) −112.65 −114.81 −112.72 −111.93 −118.27 −112.45

oxalic acid of −721.4 ± 2.1 is probably somewhat on the high
side.
In a similar manner, an isodesmic reaction targeting
cyclobutene can be constructed with ethane, cyclobutane, and
ethene as chaperons. This reaction is, not unexpectedly, very
nearly thermoneutral at 1.74 ± 1.16 kJ mol−1 and based on a
formation enthalpy for cyclobutane,46 which is absent from the
ATcT, of 28.4 ± 0.6 kJ mol−1; this results in ΔfH⊖(298.15 K) =
163.0 ± 1.3 kJ mol−1, which is somewhat higher than the ATcT
value of 156.9 ± 1.6 but in excellent agreement with our
computed average of 163.9 ± 5.9 kJ mol−1.
Statistical analyses in subsequent sections therefore do not
include the molecules 1,3-cyclopentadiene and oxalic acid but
cyclobutene is included in our later calculations.
Conformers. Some of the values in ATcT v1.110 are
composites, for example 1-butene is quoted at −0.03 ± 0.48 kJ
mol−1, but in reality this molecule is comprised of two
conformers, namely cis and gauche, with CCCC dihedral angles
of ≈0° and ≈120°, respectively. The degenerate gauche
conformer is dominant at 78.5% ± 1.5%, but the differences
in enthalpy are slight, so the overall enthalpy of formation at
298.15 K is +0.78 ± 5.59, in very good agreement with the
listed ATcT value of −0.03 ± 0.48 kJ mol−1. The results for 1-
butene and other systems are tabulated in Table 4.
In general, most of the conformer distributions are consistent
across all five composite methods, but the treatment of those
compounds whose geometries are very sensitive to slight
changes in dihedral angles, such as the gylcols and dibasic acids,
can be much less successful.
For oxalic acid, (COOH)2, three conformers prevail with the
OC−CO always anti and the H−OC−C angles either
anti or syn (Figure 1).
In the case of the higher acid succinic, (CH2COOH)2, 4
principal conformers were identified in which three dihedral
angles differ. These are classified in the order shown, Figure 2,
as syn s, anti a, or gauche g, with the middle dihedral
capitalized. Thus, the all trans configuration is denoted as aAa,
etc. These lowest energy conformers all exhibit syn H−O−C
O dihedrals. In this, we follow the notation and conclusions of a
recent study by Vogt et al.51
The optimization procedures inbuilt in CBS-QB3 and G4
converge to a structure with an imaginary frequency if a C2h
symmetry is imposed; this is a problem with the functional and
basis sets used for these two cases. The incorporation of one
(or more) diffuse terms in the 6-311G(2d,d,p) basis set, which
is the default option in the model chemistry CBS-QB3, so that
it now becomes CBS-QB3(+)52, does yield a nonimaginary
solution with C2h symmetry.
The conformer distribution outlined in Table 4 are
consistent except for the W1BD results which give less
prominence to the sGs conformer than to the aAa. These
were among the most challenging calculations undertaken
(Table 5), and perhaps a relaxation, in the case of the sGs
conformer, of the convergence-related options for the Berny
algorithm is responsible for this anomaly.
Computational Cost. A crude example of the computa-
tional cost in terms of job times is shown in Table 5 from which
it can be seen that G3 emerges as a relatively cheap yet, to be
described below, effective method.
Comparison of Methods with ATcT. We benchmark each
theoretical method against the formation enthalpies recom-
mended in the ATcT via descriptive statistics, and the results
are shown in Tables 6 and 7. In order to assess how well a
combination of different compound methods may perform,
unweighted average formation enthalpies for various combina-
tions of the CBS-QB3, CBS-APNO, G3, and G4 methods have
also been computed. Note that combinations involving the
W1BD method have generally been excluded, owing to its
expense, but it is included in a combination of “all methods” as
shown in subsequent figures and tables.
For each method, or combination of methods, the difference
between the ATcT recommendation and our computed
formation enthalpy for species, i, is denoted as Φi = ΔfH⊖ ATcT
− ΔfH⊖ i , and the mean signed errors (MSE) and mean
unsigned errors (MUE) are computed as follows where n is the
number of species in the benchmark set:
MSE = ∑ (Φi)/n (5)
MUE = ∑ |Φ|i /n (6)
In terms of uncertainties, the maximum absolute deviation
(|Φmax|) from the ATcT is determined from eq 6 and is
essentially an indicator of “worst-case scenario” performance for
each method; it is rarely used in thermochemical work as a true
indicator of statistical accuracy or precision. In line with the
recommendations of Ruscic, uncertainties are considered for
95% confidence intervals via computation of the population
standard deviation (σMSE) and root-mean-square deviations
(RMSD):
σMSE = 2 × ∑ (Φi − MSE)2 /n (7)
RMSD = 2 × ∑ (Φi)2 /n (8)

7239 DOI: 10.1021/jp511403a

J. Phys. Chem. A 2015, 119, 7235−7246
The Journal of Physical Chemistry A Article

Table 3. Final Computed Formation Enthalpies of Compounds at 298.15 K/kJ mol−1

species QB3 APNO G3 G4 W1BD ATcT ±
1,3-butadiyne 471.07 464.24 463.89 461.47 460.74 460.11 0.87
1,3-cyclopentadiene 145.90 132.59 140.30 138.00 128.72 101.30 2.50
1-butene 8.28 −4.76 2.24 2.96 −4.83 −0.03 0.48
1-butyne 175.81 165.16 169.07 168.97 163.64 165.39 0.85
2,2,4-trimethylpentane −210.74 −242.31 −221.92 −219.66 −232.28 −223.70 1.50
2,3-butanedione −329.50 −334.43 −327.74 −325.78 −332.37 −326.81 0.98
2-butyne 155.35 144.95 149.65 150.01 144.48 145.76 0.79
2-propanol −272.13 −280.26 −272.30 −270.09 −279.53 −272.81 0.37
acetic acid −434.23 −434.25 −430.05 −428.31 −434.87 −433.71 0.49
acetone −214.95 −224.33 −215.26 −214.24 −220.67 −216.09 0.37
allene 195.66 189.77 190.04 190.47 187.70 190.15 0.37
benzene 91.65 76.16 88.47 86.81 77.14 83.18 0.26
carbon monoxide −111.44 −109.62 −110.56 −113.43 −109.91 −110.53 0.03
cis-2-butene 2.10 −11.22 −3.24 −2.78 −11.32 −7.33 0.53
cyclobutene 171.05 159.78 167.55 164.67 156.42 156.90 1.60
cyclohexane −111.55 −134.71 −119.53 −117.87 −130.86 −122.08 0.68
cyclopropane 60.53 49.45 57.73 55.82 49.84 53.61 0.53
cyclopropene 291.06 284.32 287.98 285.72 281.14 283.91 0.59
dimethyl ether −188.22 −190.50 −184.27 −183.56 −188.43 −184.02 0.44
dioxirane −5.95 −0.72 3.30 0.55 2.36 1.30 1.20
ethanal −166.36 −168.38 −165.00 −164.90 −168.26 −165.46 0.32
ethane −81.61 −90.15 −84.21 −82.48 −88.23 −83.79 0.17
ethene 56.24 51.84 52.70 53.00 50.26 52.56 0.15
ethylene glycol −393.53 −394.68 −386.99 −384.86 −394.21 −389.42 0.49
ethyne 234.81 233.21 230.69 229.44 228.62 228.33 0.15
ethynol 94.53 95.57 94.34 94.75 90.96 92.70 1.40
formaldehyde −113.24 −110.05 −110.13 −111.01 −110.33 −109.17 0.11
formic acid −383.10 −378.60 −377.30 −377.30 −380.80 −378.94 0.27
glyoxal −218.25 −213.70 −213.57 −214.51 −214.65 −212.48 0.59
isobutane −127.920 −144.48 −133.05 −131.24 −140.92 −135.36 0.40
isobutene −8.92 −22.01 −14.57 −14.56 −21.95 −17.60 0.53
ketene singlet −47.50 −49.37 −49.19 −47.46 −49.38 −48.58 0.15
methane −73.95 −78.54 −75.41 −74.22 −76.21 −74.53 0.06
methanediol (sc,sc) −398.12 −394.59 −391.18 −390.87 −398.03 −392.61 0.96
methanol −203.44 −203.71 −200.26 −200.09 −204.88 −200.71 0.18
methyl formate (cis Z) −365.61 −363.24 −360.10 −358.88 −362.14 −357.80 0.59
methyl hydroperoxide −134.33 −129.09 −124.40 −125.60 −129.37 −127.73 0.91
n-butane −118.72 −135.75 −124.35 −121.86 −132.42 −125.85 0.38
n-hexane −155.63 −180.08 −163.49 −161.05 −175.67 −166.94 0.48
o-benzyne 470.39 457.41 464.76 460.69 457.45 460.70 1.40
oxalic acid −742.52 −734.26 −732.26 −728.21 −737.19 −721.40 2.10
oxirane (1A1) −55.01 −55.77 −51.02 −52.70 −55.51 −52.72 0.44
oxirene (singlet) 267.82 272.45 273.50 267.18 267.60 275.90 3.10
propane −100.52 −113.11 −104.35 −102.36 −110.74 −104.41 0.29
propene 26.05 17.07 21.40 21.71 16.12 20.35 0.33
propyne 192.48 185.07 187.42 186.75 183.79 185.80 0.38
succinic acid −829.62 −827.47 −820.98 −816.80 −826.02 −817.75 0.61
toluene 60.59 41.08 55.48 53.85 44.02 50.41 0.37
trans-2-butene −3.18 −16.34 −8.74 −8.31 −16.33 −11.18 0.51
vinyl alcohol −121.93 −123.86 −121.56 −121.16 −127.54 −123.76 0.91

Note first the 2-fold multiplier in the above formulas which
provides us with 95% confidence, and second that there is a
subtle difference in our two measures of dispersion, σMSE and
the RMSD. The former assumes the average deviation from the
ATcT (MSE) is the central value for the distribution, whereas
the latter assumes the central line is zero deviation from the
ATcT. Both measures ultimately highlight the same trends in
the performance of each method. In the text, uncertainties
typically refer to the MSE ± σMSE.
7240
We will first consider the performance of individual
compound methods against the ATcT benchmark. As a general
trend, the G3, G4, and CBS-QB3 methods tend to overpredict
the formation enthalpies of the compounds studied, thus their
negative MSE. Conversely, the W1BD and CBS-APNO
methods underpredict ΔfH⊖ leading to a positive MSE. The
computed MSEs and MUEs imply the following trend with
respect to accuracy: G3 ≈ G4 > W1BD > CBS-APNO > CBS-
QB3 (Figure 3). These results are somewhat surprising, with
DOI: 10.1021/jp511403a
J. Phys. Chem. A 2015, 119, 7235−7246
The Journal of Physical Chemistry A Article

Table 4. Composition-Averaged Enthalpies of Formation at 298.15 K/kJ mol−1

composition QB3 APNO G3 G4 W1BD ATcT
1-Butene
gauche 0.788 0.781 0.786 0.805 0.763
cis 0.212 0.219 0.214 0.195 0.237
final 8.28 −4.76 2.24 2.96 −4.83 −0.03 ± 0.48
2,2,4-Trimethylpentane
gauche 0.9981 0.9981 0.9978 0.9980 0.9980
transa 0.0019 0.0019 0.0022 0.0020 0.0022
final −210.74 −242.31 −221.90 −219.66 −232.28 −223.7 ± 1.5
2-Propanol
gauche 0.787 0.755 0.755 0.776 0.755
trans 0.213 0.245 0.246 0.224 0.245
final −272.13 −280.26 −272.30 −270.09 −279.53 −272.81 ± 0.37
Ethylene Glycol
no. 1 0.609 0.316 0.547 0.492 0.298
no. 2 0.206 0.144 0.211 0.144 0.127
no. 3 0.185 0.540 0.242 0.364 0.575
final −393.53 −394.68 −386.99 −384.86 −394.21 −389.42 ± 0.49
n-Butane
trans 0.593 0.756 0.750 0.614 0.625
gauche 0.407 0.244 0.250 0.386 0.375
final −118.72 −135.75 −124.35 −121.86 −132.42 −125.85 ± 0.38
n-Hexane
ttt 0.174 0.161 0.146 0.171 0.191
gtt 0.460 0.423 0.423 0.450 0.462
tgt 0.122 0.102 0.105 0.118 0.120
tgg 0.124 0.179 0.185 0.136 0.111
gtg 0.031 0.033 0.035 0.032 0.030
g+t+g- 0.059 0.066 0.068 0.059 0.055
ggg 0.010 0.018 0.019 0.012 0.009
g+x-t+ 0.012 0.010 0.010 0.014 0.014
t+g+x- 0.008 0.009 0.009 0.008 0.008
final −155.63 −180.08 −163.49 −161.05 −175.67 −166.94 ± 0.48
Oxalic Acid
syn/syn 0.003 0.007 0.001 0.007 0.005
anti/anti 0.966 0.919 0.936 0.950 0.966
syn/anti 0.031 0.074 0.063 0.044 0.030
final −742.52 −734.26 −732.26 −728.21 −737.19 −721.4 ± 2.1
Succinic Acid
aAa 0.166 0.128 0.108 0.136 0.480
sGs 0.733 0.765 0.778 0.782 0.433
sGa 0.059 0.056 0.065 0.042 0.033
gAs 0.042 0.052 0.049 0.040 0.053
final −829.62 −827.47 −820.98 −816.80 −826.02 −817.8 ± 0.61
Vinyl Alcohol
anti 0.195 0.229 0.255 0.209 0.169
syn 0.805 0.772 0.745 0.791 0.832
final −121.93 −123.86 −121.56 −121.16 −127.54 −123.76 ± 0.91
a
One imaginary frequency of 26 cm−1.

Figure 2. Key dihedrals of succinic acid.

Figure 1. Conformers of oxalic acid.

both Gx methods outperforming the considerably more
expensive W1BD.
7241 DOI: 10.1021/jp511403a
J. Phys. Chem. A 2015, 119, 7235−7246
The Journal of Physical Chemistry A Article

Table 5. Crude Timings (Minutes) of Calculations Steps for species, this result manifesting itself as a positive MSE. As a
sGs Succinic Acid; 8 Heavy Atoms, with Total result, combining CBS-APNO with either G3/G4, CBS-QB3/
Computational Time G3, or CBS-QB3/G4 leads to improvements in the computed
MSEs when the results are compared with individual methods.
no. step QB3 APNO G3 G4 W1BD
In fact, adding the CBS-APNO to the G3/G4 combination
1 Opt 19 7 3 81 142 leads to improvements in all statistical indicators of accuracy
2 Freq 17 4 1 45 86 and precision when compared to a G3/G4 combination. As the
3 Opt − 1579 9 − − CBS-QB3, G3, and G4 methods all tend to overpredict the
4 SPE 127 1907 73 79 435 formation enthalpies of the ATcT (negative MSE), combining
5 SPE 6 158 53 53 14760 these methods offers no improvement on a G3/G4
6 SPE 24 143 362 359 56340 combination.
7 SPE − 317 193 277 23280 Also shown in Table 7 are results from a combination of all
8 SPE − − − 355 39240 methods, both including and excluding W1BD computations.
9 SPE − − − 3480 − Adding the CBS-QB3 method to a CBS-APNO/G3/G4
Total Simulation Times combination offers nothing in terms of reducing absolute
hours 3.2 68.6 11.6 78.8 2238 deviations and uncertainties. In turn, incorporating our W1BD
relative 1.0 21.3 3.6 24.5 695.8 results into the average reduces the MUE, RMSD, and standard
deviation, but only marginally, and perhaps not significantly
Rayne and Forrest have shown in a study of a number of enough to warrant the added expense.
compounds, not just of type CxHyOz but also containing Surprisingly, the W1BD method shows poorer performance
nitrogen, sulfur, fluorine, and chlorine, with up to 4 “heavy” than many so-called “cheaper” methods against this test-set of
atoms that W1BD ΔfH⊖(298.15 K) are systematically lower formation enthalpies. On the basis of the present results, the
than comparable G4 ones.53 Some of these results have been following methods, or combination of methods, could be
questioned,54,55 but for all the 18 species in common between argued to offer more accurate and precise results than W1BD at
our work and Rayne’s and Forrest’s, there are only very small significant cost savings: G3, G4, G3/G4, CBS-APNO/G3,
differences in the reported formation enthalpies. CBS-APNO/G4, CBS-APNO/G3/G4, CBS-QB3/CBS-
The CBS-APNO and CBS-QB3 methods tend to perform APNO/G4, CBS-QB3/CBS-APNO/G4, CBS-QB3/CBS-
worst of all, and while on average they tend to predict the ΔfH⊖ APNO/G3/G4. The added advantages of using multiple
to within ≈5 kJ mol−1, in some instances these methods deviate methods which were alluded to earlier would also apply.
absolutely from the ATcT by >10 kJ mol−1, with respective The question also arises as to whether the accuracy of any
|Φmax| of 18.61 and 14.10 kJ mol−1 computed. |Φmax| in the given method deteriorates with the size of the system under
cases of G3, G4, and W1BD are computed as 10.60, 8.70, and study. Plots of MSE versus molecular weight (MW) (Figure 5)
8.82 kJ mol−1, respectively. These trends are visible in Figure 4, show that over the admittedly narrow range explored here,
where statistical measures of variation (2σ about the MSE, and increasing MW leads to increasing MSEs for both CBS-APNO
twice the RMSD) show the following trends in terms of and W1BD. Conversely, a slight, perhaps negligible, decreasing
precision: G3 ≈ G4 > W1BD > CBS-APNO > CBS-QB3. MSE is observed for CBS-QB3 computations as f(MW), and
The CBS-x methods are therefore more prone to producing almost no change for G3 and G4. Figure 5 also illustrates the
an extremely erroneous result, a certain warning flag should one general trend of increasing variation in ΔfH⊖ across methods as
MW increases.

attempt to use either of these methods in isolation for
thermochemical work. Thus, Figures 3 and 4 also delineate the
performance of combinations of two or more different
compound methods in terms of accuracy and precision.
For combinations of two compound methods, six combina-
tion are possible using the CBS-QB3, CBS-APNO, G3, and G4
methods. An average formation enthalpy from any combination
of the CBS-APNO, G3, and G4 methods shows the lowest
MSE and MUE, and RMSD and 2σ deviations, of all individual
and combined methods. Thus, combining any two of these
methods may offer an improvement over their individual
implementation. Likewise, the CBS-QB3 and CBS-APNO
methods are shown to perform quite poorly in isolation, but
a combination of these methods outperforms either method in
isolation in terms of accuracy and precision.
In terms of computational time, Table 5, a combination of
CBS-QB3 and the CBS-APNO methods is of a similar expense
to a combination of the G3 and G4 methods, so a combination
of the latter two methods offers increased accuracy and
precision for roughly the same cost. Combining CBS-APNO
with G4 does lead to improvements from the CBS-APNO
method in isolation, but combining CBS-APNO with G3, or
G3 with G4, would provide a more cost-effective approach.
Excluding W1BD, CBS-APNO was the only method which
tended to under-predict the formation enthalpy of the target
7242
■
CONCLUSION AND RECOMMENDATIONS
In identifying the combination CBS-APNO/G3, CBS-APNO/
G4, and/or G3/G4 as the best combinations of methods for
computing atomization energies, we are conscious of the fact
that the different approaches used by these two methods are
poles apart. Thus, for example, G3 utilizes HF/6-31G(d) for
geometry optimization and frequency determination, whereas
G4 employs B3LYP/6-31G(2df,p) for that purpose. Subse-
quent single-point energy calculations are also considerably
different as is the fact that G3 reoptimizes the structure at
MP2(Full)/6-31G(d) before embarking on the single-point
corrections. Thus, the results from this combination are to
some extent insulated from errors arising from specific
contributions, and very good data can be obtained. The fact
that the model chemistry CBS-APNO can only be applied to
first row atoms does militate against its more widespread
application, although a CBS-APNO/G3/G4 combination offers
formation enthalpies which rival chemical accuracy (0.14 ±
4.21 kJ mol−1). These uncertainties are expected to be lower
should one employ this trio of methods in combination with
well-framed isodesmic reactions.
For a discussion of the statistical advantages and the relevant
equations, applied in that case to a series of isodesmic reactions
DOI: 10.1021/jp511403a
J. Phys. Chem. A 2015, 119, 7235−7246
The Journal of Physical Chemistry A Article

Table 6. Calculated Differences (Φi, kJ mol−1) for the Five Compound Methods Tested with ATcT Recommendations and
Uncertaintiesa
molecule QB3 APNO G3 G4 W1BD ATcT ±
1,3-butadiyne −10.99 −4.09 −3.79 −1.39 −0.59 460.11 0.87
1,3-cyclopentadienea −44.60 −31.30 −39.00 −36.70 −27.40 101.30 2.50
1-butene −8.33 4.77 −2.23 −3.03 4.77 −0.03 0.48
1-butyne −10.41 0.19 −3.71 −3.61 1.79 165.39 0.85
2,2,4-trimethylpentane −12.96 18.61 −1.80 −4.04 8.58 −223.70 1.50
2,3-butanedione 2.69 7.59 0.89 −1.01 5.59 −326.81 0.98
2-butyne −9.54 0.76 −3.84 −4.24 1.26 145.76 0.79
2-propanol −0.71 7.49 −0.51 −2.71 6.69 −272.81 0.37
acetic acid 0.49 0.49 −3.61 −5.41 1.19 −433.71 0.49
acetone −1.09 8.21 −0.79 −1.89 4.61 −216.09 0.37
allene −5.55 0.35 0.15 −0.35 2.45 190.15 0.37
benzene −8.42 6.98 −5.32 −3.62 6.08 83.18 0.26
carbon monoxide 0.88 −0.93 0.07 2.88 −0.63 −110.53 0.03
cis-2-butene −9.43 3.87 −4.13 −4.53 3.97 −7.33 0.53
cyclobutene −14.10 −2.90 −10.60 −7.80 0.50 156.90 1.60
cyclohexane −10.48 12.62 −2.58 −4.18 8.82 −122.08 0.68
cyclopropane −6.89 4.11 −4.09 −2.19 3.81 53.61 0.53
cyclopropene −7.19 −0.39 −4.09 −1.79 2.81 283.91 0.59
dimethyl ether 4.18 6.48 0.28 −0.42 4.38 −184.02 0.44
dioxirane 7.20 2.00 −2.00 0.80 −1.10 1.30 1.20
ethanal 0.94 2.94 −0.46 −0.56 2.84 −165.46 0.32
ethane −2.19 6.31 0.41 −1.29 4.41 −83.79 0.17
ethene −3.64 0.76 −0.14 −0.44 2.26 52.56 0.15
ethylene glycol 4.08 5.28 −2.42 −4.52 4.78 −389.42 0.49
ethyne −6.47 −4.87 −2.37 −1.07 −0.27 228.33 0.15
ethynol −1.80 −2.90 −1.60 −2.10 1.70 92.70 1.40
formaldehyde 4.03 0.93 0.93 1.83 1.13 −109.17 0.11
formic acid 4.16 −0.34 −1.64 −1.64 1.86 −378.94 0.27
glyoxal 5.72 1.22 1.12 2.02 2.12 −212.48 0.59
isobutane −7.46 9.14 −2.36 −4.16 5.54 −135.36 0.40
isobutene −8.70 4.40 −3.00 −3.00 4.40 −17.60 0.53
ketene singlet −1.08 0.82 0.62 −1.08 0.82 −48.58 0.15
methane −0.53 3.97 0.87 −0.33 1.67 −74.53 0.06
methanediol (sc,sc) 5.49 1.99 −1.41 −1.71 5.39 −392.61 0.96
methanol 2.69 2.99 −0.41 −0.61 4.19 −200.71 0.18
methyl formate (cis Z) 7.80 5.40 2.30 1.10 4.30 −357.80 0.59
methyl hydroperoxide 6.57 1.37 −3.33 −2.13 1.67 −127.73 0.91
n-butane −7.15 9.85 −1.55 −3.95 6.55 −125.85 0.38
n-hexane −11.34 13.16 −3.44 −5.84 8.76 −166.94 0.48
o-benzyne −9.70 3.30 −4.10 0.00 3.30 460.70 1.40
oxalic acida 21.10 12.90 10.90 6.80 15.80 −721.40 2.10
oxirane (1A1) 2.28 3.08 −1.72 −0.02 2.78 −52.72 0.44
oxirene (singlet) 8.10 3.40 2.40 8.70 8.30 275.90 3.10
propane −3.91 8.69 −0.01 −2.01 6.29 −104.41 0.29
propene −5.65 3.25 −1.05 −1.35 4.25 20.35 0.33
propyne −6.70 0.70 −1.60 −0.90 2.00 185.80 0.38
succinic acid 11.87 9.75 3.25 −0.95 8.25 −817.75 0.61
toluene −10.19 9.31 −5.09 −3.39 6.41 50.41 0.37
trans-2-butene −7.98 5.12 −2.48 −2.88 5.12 −11.18 0.51
vinyl alcohol −1.86 0.14 −2.16 −2.56 3.74 −123.76 0.91
a
Species are omitted from statistical analyses.

but capable of much wider interpretation, see earlier work on
alkyl hydroperoxides by Simmie et al.56
Usually transferable uncertainties are used by theoreticians
developing new methodologies; that is, they benchmark their
method against an existing database and then transfer that
uncertainty to new species. But there is no evidence that such
an approach has any grounding in reality.
7243
A model chemistry which combines the best functional and
basis set for optimization and frequency calculations and not
too expensive higher-order corrections is badly needed for
usage by inexpert computational chemists. It should not have
two geometry optimization routines, as in G3 or CBS-APNO
for example, because occasional failures can occur in the second
optimization which is perforce starting from the previous
DOI: 10.1021/jp511403a
J. Phys. Chem. A 2015, 119, 7235−7246
The Journal of Physical Chemistry A Article

Table 7. Summary Statistics (kJ mol−1) Based on Results

shown in Table 6
model chemistry MUE MSE |Φmax| ±2σ 2RMSD
G3 2.27 −1.71 10.60 4.76 5.86
G4 2.46 −1.74 8.70 5.15 6.21
W1BD 3.85 3.74 8.82 5.13 8.82
APNO 4.55 3.86 18.61 9.21 12.02
QB3 6.08 −2.78 14.10 13.02 14.16
2 Methods
APNO/G4 2.05 1.06 7.29 4.71 5.17
APNO/G3 2.24 1.08 8.41 5.51 5.91
G3/G4 2.27 −1.72 9.20 4.56 5.72
QB3/APNO 3.06 0.54 10.81 7.50 7.58
QB3/G3 3.85 −2.24 12.35 8.33 9.46
QB3/G4 3.98 −2.26 10.95 8.33 9.47
3 Methods
APNO/G3/G4 1.58 0.14 7.10 4.21 4.22
QB3/APNO/G4 2.37 −0.22 8.27 5.87 5.89
QB3/APNO/G3 2.45 −0.21 9.20 6.21 6.22
QB3/G3/G4 3.23 −2.07 10.83 6.83 7.99
4+ Methods
Figure 4. Measures of dispersion (±2σ from MSE, 2 times root-mean-
QB3/APNO/G3/G4 2.17 −0.59 8.85 5.40 5.53
squared deviation) for individual compound methods, and combina-
all methods 1.80 0.28 6.98 4.75 4.79 tions thereof, upon comparison with the ATcT.

Figure 5. Signed error in standard state formation enthalpy versus

Figure 3. Mean signed errors (MSE), mean unsigned errors (MUE),
molecular mass for compounds studied. Lines are fitted via least-
and maximum absolute deviations (|Φmax|) for individual compound
squares regression. Consult online version of manuscript for
methods, and combinations thereof, upon comparison with the ATcT.
interpretation of color in figure.

geometry. This is particularly the case in the location of

transition states where energy minimization failures are rather
more common. Some progress in this direction has been made
by da Silva who has combined the M06-2X functional within a
G3-type framework to create G3X-K theory.57
Three out of the five model chemistries surveyed here
employ the B3LYP functional, which although not as popular
now as it was at the time of a 2007 review by Sousa et al.,58 is
still very common. A range of better functionals are available,
see for example recent comprehensive reviews,59,60 but have
not yet come into widespread use.
The treatment of vibrational and rotational partition
functions is crucial because from an end-user perspective
thermochemistry at 0 K is of a much lesser value than the same
data at 298.15 K. The harmonic oscillator approximation is
7244
commonly used to determine vibrational frequencies, but this
necessitates scale factors to correct the resultant zero-point
energies and vibrational partition functions. Unfortunately,
separate factors seem to be required to scale the frequencies
(“low” frequencies being treated differently than “high”
frequencies)61 to compute zero-point energies, entropies and
enthalpies, and worse still, these latter are functions of
temperature.62 An effort is needed here to simplify without
losing essential details perhaps by way of universal scale
factors.63 The incorporation of anharmonic effects has not been
seriously attempted for moderately sized molecules for reasons
discussed by Irikura.64
In order to reach the wider computational chemistry
community, the composite method should be embedded or
DOI: 10.1021/jp511403a
J. Phys. Chem. A 2015, 119, 7235−7246
The Journal of Physical Chemistry A
easily interfaced with all popular computational codes including
those which carry a licensing fee.
Although the ATcT probably represents the current state of
the art for thermochemical tabulations, and the results
presented herein reinforce the large majority of recommenda-
tions in this database, there remains a certain lack of
transparency in how some formation enthalpies reported
therein have been derived, 1,3-cyclopentadiene, oxalic and
succinic acids, and cyclobutene being exemplary. A minor but
important improvement to benefit auditability of the
compendium would be direct referencing of any relevant
experiment or theory which were primarily used to derive a
given value, as done in the Third Millennium Database.3
■
*
ASSOCIATED CONTENT
S Supporting Information
0 K Zero-point energy-corrected electronic energies of all
compounds studied and 298.15 K zero-point and thermal
energy-corrected electronic energies of all compounds studied.
This material is available free of charge via the Internet at
http://pubs.acs.org.
■
AUTHOR INFORMATION
Corresponding Author
*E-mail: john.simmie@nuigalway.ie.
Notes
The authors declare no competing financial interest.
■
ACKNOWLEDGMENTS
Computational resources were provided by the Irish Centre for
High-End Computing, ICHEC. We are grateful to Alin Elena
(ICHEC), C. Franklin Goldsmith (Brown), Amir Karton
(Western Australia), Wim Klopper (Karlsruhe), Errol Lewars
(Trent), Jan Martin (Weizmann Institute), and Andrew Yeung
(Texas A & M) for their considerable assistance. We thank the
reviewers for their useful comments.
■
REFERENCES
(1) NIST-JANAF Thermochemical Tables, 4th ed.; Chase, M. W. Jr.,
Ed.; J. Phys. Chem. Ref. Data Monogr. 9; American Institute of
Physics and American Chemical Society: New York, 1998.
(2) Cox, J. D.; Wagman, D. D.; Medvedev, V. A. CODATA Key
Values for Thermochemistry; Hemisphere, New York, 1989.
(3) Goos, E.; Burcat, A.; Ruscic, B. Ideal Gas Thermochemical
Database with Updates from Active Thermochemical Tables. http://
garfield.chem.elte.hu/Burcat/burcat.html (accessed January 20, 2015).
(4) Ruscic, B.; Pinzon, R. E.; Morton, M. L.; von Laszevski, G.;
Bittner, S. J.; Nijsure, S. G.; Amin, K. A.; Minkoff, M.; Wagner, A. F.
Introduction to Active Thermochemical Tables: Several “Key”
Enthalpies of Formation Revisited. J. Phys. Chem. A 2004, 108 (45),
9979−9997.
(5) Ruscic, B.; Pinzon, R. E.; von Laszewski, G.; Kodeboyina, D.;
Burcat, A.; Leahy, D.; Montoya, D.; Wagner, A. F. Active
Thermochemical Tables: Thermochemistry for the 21st Century. J.
Phys. Conf. Ser. 2005, 16, 561−570.
(6) Ruscic, B. Active Thermochemical Tables. ATcT.anl.gov
(updated Active Thermochemical Tables (ATcT) values based on v
1.110 of the Thermochemical Network (2012); last update Nov 30,
2013.)
(7) Simmie, J. M. Detailed Chemical Kinetic Models for the
Combustion of Hydrocarbon Fuels. Prog. Energy Combust. Sci. 2003,
29 (6), 599−634.
(8) Wilke, J. J.; Allen, W. D.; Schaefer, H. F. Establishment of the
C2H5+O2 Reaction Mechanism: A Combustion Archetype. J. Chem.
Phys. 2008, 128 (7), 074308.
7245
Article
(9) Raghavachari, K.; Trucks, G. W.; Pople, J. A.; Head-Gordon, M.
A 5th-Order Perturbation Comparison of Electron Correlation
Theories. Chem. Phys. Lett. 1989, 157 (6), 479−483.
(10) Martin, J. M. L.; de Oliveira, G. Towards Standard Methods for
Benchmark Quality Ab Initio Thermochemistry: W1 and W2 Theory.
J. Chem. Phys. 1999, 111 (5), 1843−1856.
(11) Boese, A. D.; Oren, M.; Atasoylu, O.; Martin, J. M. L.; Kallay,
M.; Gauss, J. W3 Theory: Robust Computational Thermochemistry in
the kJ/mol Accuracy Range. J. Chem. Phys. 2004, 120 (9), 4129−4141.
(12) Karton, A.; Rabinovich, E.; Martin, J. M. L.; Ruscic, B. W4
Theory for Computational Thermochemistry: In Pursuit of Confident
Sub-kJ/mol Predictions. J. Chem. Phys. 2006, 125 (14), 144108.
(13) Tajti, A.; Szalay, P. G.; Császár, A. G.; Kallay, M.; Gauss, J.;
Valeev, E. F.; Flowers, B. A.; Vazquez, J.; Stanton, J. F. HEAT: High
Accuracy Extrapolated Ab Initio Thermochemistry. J. Chem. Phys.
2004, 121 (23), 11599−11613.
(14) Bomble, Y. J.; Vazquez, J.; Kallay, M.; Michauk, C.; Szalay, P. G.;
Császár, A. G.; Gauss, J.; Stanton, J. F. High-Accuracy Extrapolated Ab
Initio Thermochemistry. II. Minor Improvements to the Protocol and
a Vital Simplification. J. Chem. Phys. 2006, 125 (6), 064108.
(15) Harding, M. E.; Vazquez, J.; Ruscic, B.; Wilson, A. K.; Gauss, J.;
Stanton, J. F. High-Accuracy Extrapolated Ab Initio Thermochemistry.
III. Additional Improvements and Overview. J. Chem. Phys. 2008, 128
(11), 114111.
(16) Császár, A. G.; Allen, W. D.; Schaefer, H. F. In Pursuit of the Ab
Initio Limit for Conformational Energy Prototypes. J. Chem. Phys.
1998, 108 (23), 9751−9764.
(17) Feller, D.; Peterson, K. A.; Dixon, D. A. Further Benchmarks of
a Composite, Convergent, Statistically Calibrated Coupled-Cluster-
Based Approach for Thermochemical and Spectroscopic Studies. Mol.
Phys. 2012, 110 (19−20), 2381−2399.
(18) Montgomery, J. A.; Frisch, M. J.; Ochterski, J. W.; Petersson, G.
A. A Complete Basis Set Model Chemistry. VII. Use of the Minimum
Population Localization Method. J. Chem. Phys. 2000, 112 (15),
6532−6542.
(19) Ochterski, J. W.; Petersson, G. A.; Montgomery, J. A. A
Complete Basis Set Model Chemistry. V. Extensions to Six or More
Heavy Atoms. J. Chem. Phys. 1996, 104 (7), 2598−2619.
(20) Pople, J. A.; Head-Gordon, M.; Fox, D. J.; Raghavachari, K.;
Curtiss, L. A. Gaussian-1 Theory: A General Procedure for Prediction
of Molecular Energies. J. Chem. Phys. 1989, 90, 5622.
(21) Curtiss, L. A.; Jones, C.; Trucks, G. W.; Raghavachari, K.; Pople,
J. A. Gaussian-1 Theory of Molecular-Energies for 2nd-Row
Compounds. J. Chem. Phys. 1990, 93 (4), 2537−2545.
(22) Curtiss, L. A.; Raghavachari, K.; Trucks, G. W.; Pople, J. A.
Gaussian-2 Theory for Molecular-Energies of 1st-Row and 2nd-Row
Compounds. J. Chem. Phys. 1991, 94 (11), 7221−7230.
(23) Curtiss, L. A.; Raghavachari, K.; Redfern, P. C.; Rassolov, V.;
Pople, J. A. Gaussian-3 (G3) Theory for Molecules Containing First
and Second-Row Atoms. J. Chem. Phys. 1998, 109 (18), 7764−7776.
(24) Curtiss, L. A.; Redfern, P. C.; Raghavachari, K. Gaussian-4
Theory. J. Chem. Phys. 2007, 126, 084108.
(25) Baboul, A. G.; Curtiss, L. A.; Redfern, P. C.; Raghavachari, K.
Gaussian-3 Theory Using Density Functional Geometries and Zero-
Point Energies. J. Chem. Phys. 1999, 110 (7), 7650−7657.
(26) Curtiss, L. A.; Redfern, P. C.; Raghavachari, K.; Rassolov, V.;
Pople, J. A. Gaussian-3 Theory Using Reduced Möller-Plesset Order. J.
Chem. Phys. 1999, 110 (10), 4703−4709.
(27) Chan, B.; Coote, M. L.; Radom, L. G4-SP, G4(MP2)-SP, G4-
SC, and G4(MP2)-SC: Modifications to G4 and G4(MP2) for the
Treatment of Medium-Sized Radicals. J. Chem. Theory Comput. 2010,
6 (9), 2647−2653.
(28) Chan, B.; Deng, J.; Radom, L. G4(MP2)-6x: A Cost-Effective
Improvement to G4(MP2). J. Chem. Theory Comput. 2011, 7 (1),
112−120.
(29) Chan, B.; Radom, L. W1x-1 and W1x-2: W1-Quality Accuracy
with an Order of Magnitude Reduction in Computational Cost. J.
Chem. Theory Comput. 2012, 8 (11), 4259−4269.
DOI: 10.1021/jp511403a
J. Phys. Chem. A 2015, 119, 7235−7246
The Journal of Physical Chemistry A
(30) Karton, A.; Martin, J. M. L. Explicitly Correlated Wn Theory:
W1-F12 and W2-F12. J. Chem. Phys. 2012, 136 (12), 124114.
(31) Chan, B.; Radom, L. W3x: A Cost-Effective Post-CCSD(T)
Composite Procedure. J. Chem. Theory Comput. 2013, 9 (11), 4769−
4778.
(32) Karton, A.; Gruzman, D.; Martin, J. M. L. Benchmark
Thermochemistry of the CnH2n+2 Alkane Isomers (N = 2−8) and
Performance of DFT and Composite Ab Initio Methods for
Dispersion-Driven Isomeric Equilibria. J. Phys. Chem. A 2009, 113
(29), 8434−8447.
(33) Wheeler, S. E.; Houk, K. N.; Schleyer, P. V. R.; Allen, W. D. A
Hierarchy of Homodesmotic Reactions for Thermochemistry. J. Am.
Chem. Soc. 2009, 131 (7), 2547−2560.
(34) Würmel, J.; Simmie, J. M. Thermochemistry and Kinetics of
Angelica and Cognate Lactones. J. Phys. Chem. A 2014, 118, 4172−
4183.
(35) Ruscic, B. Uncertainty Quantification in Thermochemistry,
Benchmarking Electronic Structure Computations, and Active
Thermochemical Tables. Int. J. Quantum Chem. 2014, 114 (17),
1097−1101.
(36) Simmie, J. M.; Somers, K. P.; Metcalfe, W. K.; Curran, H. J.
Substituent Effects in the Thermochemistry of Furans: A Theoretical
(CBS-QB3, CBS-APNO and G3) Study. J. Chem. Thermodyn. 2013,
58, 117−128.
(37) Somers, K. P. On the Pyrolysis and Combustion of Furans:
Quantum Chemical, Statistical Rate Theory and Chemical Kinetic
Modelling Studies. Ph.D. Thesis, National University of Ireland,
Galway, 2014.
(38) Barnes, E. C.; Petersson, G. A.; Montgomery, J. A.; Frisch, M. J.;
Martin, J. M. L. Unrestricted Coupled Cluster and Brueckner Doubles
Variations of W1 Theory. J. Chem. Theory Comput. 2009, 5 (10),
2687−2693.
(39) Frisch, M. J.; Trucks, G. W.; Schlegel, H. B.; Scuseria, G. E.;
Robb, M. A.; Cheeseman, J. R.; Scalmani, G.; Barone, V.; Mennucci,
B.; Petersson, G. A.; et al. Gaussian 09, revision D.01; Gaussian, Inc.:
Wallingford, CT, 2009.
(40) Morini, F.; Knippenberg, S.; Deleuze, M. S.; Hajgato, B.
Quantum Chemical Study of Conformational Fingerprints in the
Photoelectron Spectra and (e, 2e) Electron Momentum Distributions
of n-Hexane. J. Phys. Chem. A 2010, 114 (12), 4400−4417.
(41) Spartan ‘08, v 1.0.0; Wavefunction Inc., Irvine, CA, 2008.
(42) Vacek, G.; Galbraith, J. M.; Yamaguchi, Y.; Schaefer, H. F.;
Nobes, R. H.; Scott, A. P.; Radom, L. Oxirene: To Be or Not to Be. J.
Phys. Chem. 1994, 98 (35), 8660−8665.
(43) Lewars, E. Modeling Marvels, Chapter 3; Springer: Amsterdam,
2008.
(44) Vogiatzis, K. D.; Haunschild, R.; Klopper, W. Accurate
Atomization Energies from Combining Coupled-Cluster Computa-
tions with Interference-Corrected Explicitly Correlated Second-Order
Perturbation Theory. Theor. Chem. Acc. 2014, 133 (3), 1−12.
(45) Karton, A.; Daon, S.; Martin, J. M. L. W4−11: A High-
Confidence Benchmark Dataset for Computational Thermochemistry
Derived from First-Principles W4 Data. Chem. Phys. Lett. 2011, 510
(4−6), 165−178.
(46) Pedley, J. B.; Naylor, R. D.; Kirby, S. P. Thermochemical Data of
Organic Compounds, 2nd ed.; Chapman & Hall: London, 1986.
(47) Roth, W. R.; Adamczak, O.; Breuckmann, R.; Lennartz, H. W.;
Boese, R. Resonance Energy Calculation: The MM2ERW Force-Field.
Chem. Ber. 1991, 124 (11), 2499−2521.
(48) Furuyama, S.; Golden, D. M.; Benson, S. W. Thermochemistry
of Cyclopentene and Cyclopentadiene from Studies of Gas-Phase
Equilibria. J. Chem. Thermodyn. 1970, 2 (2), 161−169.
(49) Zhong, X.; Bozzelli, J. W. Thermochemical and Kinetic Analysis
of the H, OH, HO2, O, and O2 Association Reaction with
Cyclopentadienyl Radical. J. Phys. Chem. A 1998, 102, 3537−3555.
(50) Agapito, F.; Nunes, P. A.; Cabral, B. J. C.; dos Santos, R. M. B.;
Simoes, J. A. M. Energetic Differences between the Five- and Six-
Membered Ring Hydrocarbons: Strain Energies in the Parent and
Radical Molecules. J. Org. Chem. 2008, 73 (16), 6213−6223.
7246
Article
(51) Vogt, N.; Abaev, M. A.; Rykov, A. N.; Shishkov, I. F.
Determination of Molecular Structure of Succinic Acid in a Very
Complex Conformational Landscape: Gas-Phase Electron Diffraction
(GED) and Ab Initio Studies. J. Mol. Struct. 2011, 996 (1−3), 120−
127.
(52) Parthiban, S.; de Oliveira, G.; Martin, J. M. L. Benchmark Ab
Initio Energy Profiles for the Gas-Phase SN2 Reactions Y− + CH3X →
CH3Y + X− (X,Y = F,Cl,Br). Validation of Hybrid DFT Methods. J.
Phys. Chem. A 2001, 105 (5), 895−904.
(53) Rayne, S.; Forest, K. Estimated Gas-Phase Standard State
Enthalpies of Formation for Organic Compounds Using the Gaussian-
4 (G4) and W1BD Theoretical Methods. J. Chem. Eng. Data 2010, 55
(11), 5359−5364.
(54) Dorofeeva, O. V. Comments on “Estimated Gas-Phase Standard
State Enthalpies of Formation for Organic Compounds Using the
Gaussian-4 (G4) and W1BD Theoretical Methods” (Rayne, S.; Forest,
K. J. Chem. Eng. Data 2010, 55, 5359−5364). J. Chem. Eng. Data 2011,
56 (3), 682−683.
(55) Rayne, S.; Forest, K. Reply to Comments by O. V. Dorofeeva on
J. Chem. Eng. Data 2010, 55, 5359−5364. J. Chem. Eng. Data 2011, 56
(3), 684−685.
(56) Simmie, J. M.; Black, G.; Curran, H. J.; Hinde, J. P. Enthalpies of
Formation and Bond Dissociation Energies of Lower Alkyl Hydro-
peroxides and Related Hydroperoxy and Alkoxy Radicals. J. Phys.
Chem. A 2008, 112 (22), 5010−5016.
(57) da Silva, G. G3X-K Theory: A Composite Theoretical Method
for Thermochemical Kinetics. Chem. Phys. Lett. 2013, 558, 109−113.
(58) Sousa, S. F.; Fernandes, P. A.; Ramos, M. J. General
Performance of Density Functionals. J. Phys. Chem. A 2007, 111
(42), 10439−10452.
(59) Peverati, R.; Truhlar, D. G. Quest for a Universal Density
Functional: The Accuracy of Density Functionals across a Broad
Spectrum of Databases in Chemistry and Physics. Philos. Trans. R. Soc.,
A 2014, 372, 20120476.
(60) Zhang, I. Y.; Xu, X. Doubly Hybrid Density Functional for
Accurate Description of Thermochemistry, Thermochemical Kinetics
and Nonbonded Interactions. Int. Rev. Phys. Chem. 2011, 30 (1), 115−
160.
(61) Laury, M. L.; Boesch, S. E.; Haken, I.; Sinha, P.; Wheeler, R. A.;
Wilson, A. K. Harmonic Vibrational Frequencies: Scale Factors for
Pure, Hybrid, Hybrid Meta, and Double-Hybrid Functionals in
Conjunction with Correlation Consistent Basis Sets. J. Comput.
Chem. 2011, 32 (11), 2339−2347.
(62) Merrick, J. P.; Moran, D.; Radom, L. An Evaluation of
Harmonic Vibrational Frequency Scale Factors. J. Phys. Chem. A 2007,
111 (45), 11683−11700.
(63) Alecu, I. M.; Zheng, J. J.; Zhao, Y.; Truhlar, D. G.
Computational Thermochemistry: Scale Factor Databases and Scale
Factors for Vibrational Frequencies Obtained from Electronic Model
Chemistries. J. Chem. Theory Comput. 2010, 6 (9), 2872−2887.
(64) Irikura, K. K. Anharmonic Partition Functions for Polyatomic
Thermochemistry. J. Chem. Thermodyn. 2014, 73, 183−189.
DOI: 10.1021/jp511403a
J. Phys. Chem. A 2015, 119, 7235−7246