View Article Online / Journal Homepage / Table of Contents for this issue

PCCP Dynamic Article Links

Cite this: Phys. Chem. Chem. Phys., 2011, 13, 13683–13689

www.rsc.org/pccp PAPER
Validation of electronic structure methods for isomerization reactions
of large organic molecules
Published on 04 July 2011. Downloaded by Universidad de Salamanca on 5/3/2021 5:26:15 PM.

Sijie Luo,a Yan Zhaob and Donald G. Truhlar*a

Received 20th March 2011, Accepted 25th May 2011
DOI: 10.1039/c1cp20834a

In this work the ISOL24 database of isomerization energies of large organic molecules presented
by Huenerbein et al. [Phys. Chem. Chem. Phys., 2010, 12, 6940] is updated, resulting in the
new benchmark database called ISOL24/11, and this database is used to test 50 electronic
model chemistries. To accomplish the update, the very expensive and highly accurate
CCSD(T)-F12a/aug-cc-pVDZ method is first exploited to investigate a six-reaction subset of the
24 reactions, and by comparison of various methods with the benchmark, MCQCISD-MPW is
confirmed to be of high accuracy. The final ISOL24/11 database is composed of six reaction
energies calculated by CCSD(T)-F12a/aug-cc-pVDZ and 18 calculated by MCQCISD-MPW. We
then tested 40 single-component density functionals (both local and hybrid), eight doubly hybrid
functionals, and two other methods against ISOL24/11. It is found that the SCS-MP3/CBS
method, which is used as benchmark for the original ISOL24, has an MUE of 1.68 kcal mol 1,
which is close to or larger than some of the best tested DFT methods. Using the new benchmark,
we find oB97X-D and MC3MPWB to be the best single-component and doubly hybrid
functionals respectively, with PBE0-D3 and MC3MPW performing almost as well. The best
single-component density functionals without molecular mechanics dispersion-like terms are
M08-SO, M08-HX, M05-2X, and M06-2X. The best single-component density functionals
without Hartree–Fock exchange are M06-L-D3 when MM terms are included and M06-L
when they are not.

1. Introduction but progress is being made in extending the capabilities to

larger systems.
It is important to test electronic structure model chemistries1 It has been pointed out4 that ‘‘a good place to start is the
against a wide variety of data. Many validation studies have energetics of isomerization reactions since this allows direct
been based on small molecules where accurate data is most comparison on the performance for differences in bonding,
readily available. But some theoretical methods that work well conjugation, and steric effects.’’ Recently, Huenerbein et al.5
for small molecules are not so successful for large ones. very constructively pushed the limit of what can be done in this
Further tests are necessary, but extracting accurate electronic regard by using spin-component-scaled third-order Møller–
structure test data from experiment is problematic for large Plesset perturbation theory (SCS-MP3) to calculate reference
molecules because electronic structure calculations directly values of the isomerization energy (DE) of 24 large molecules
predict Born–Oppenheimer energies (E), but experiment yields containing 24–81 atoms. They estimated uncertainties to be
enthalpies (H) or free energies (G). Extracting E from H or G typically smaller than 2 kcal mol 1 with maximum uncertainties
requires the removal of conformational-vibrational-rotational of up to 4–5 kcal mol 1. The reference values were used to test
energy, which becomes progressively more complicated as 22 density functional methods that were found to have mean
molecular size increases.2 Furthermore experimental H and unsigned deviations from the reference data of 2.5–14.8 kcal mol 1.
G data becomes sparser as molecular size increases. In recent Their reference database is called ISOL24.5 The ISOL24
years though an alternative approach has become more affordable, reactions are shown in Fig. 1, which numbers each reaction
namely the direct calculation of E by high-level wave function (from 1 to 24) as in ref. 5, for convenience.
theory.3 This also becomes harder as molecular size increases, The objective of the present work is twofold: (1) to improve
the accuracy of the reference data and (2) to test 50 additional
a
Department of Chemistry and Supercomputing Institute, electronic structure methods that are affordable for the entire
University of Minnesota, Minneapolis, MN 55455-0431, USA benchmark suite. An electronic structure method is a short name
b
Commercial Print Engine Lab., HP Laboratories,
Hewlett-Packard Co., 1501 Page Mill Road, Palo Alto, for an electronic structure model chemistry,1 which is a combi-
CA 94304, USA nation of a wave function level (e.g., MP3 or Hartree–Fock) or

This journal is c the Owner Societies 2011 Phys. Chem. Chem. Phys., 2011, 13, 13683–13689 13683
Published on 04 July 2011. Downloaded by Universidad de Salamanca on 5/3/2021 5:26:15 PM. View Article Online

Fig. 1 The 24 reactions of ISOL24 and ISOL24/11.

an approximate density functional with a one-electron basis In order to validate affordable methods for improving the
set or a one-electron-basis-set extrapolation procedure. A method reference results, we first carried out CCSD(T)-F12a/aDZ
is denoted, as usual, as L/B where L denotes the wave function wave function calculations on the six smallest reactions (reactions
level or density functional, and B denotes the basis set or basis-set 3, 9, 10, 13, 14, and 20); these reactions have 24–35 atoms.
extrapolation procedure. Here CCSD denotes coupled cluster theory with single and
double excitations,15 (T) denotes a quasiperturbative treatment
of connected triple excitations,16 and F12a denotes a simpli-
2. Methods
fied version17 of the F1218 method that includes functions
In the present work we directly use three basis sets: aDZ6 of interelectronic distance (r12) to accelerate convergence. The
(shorthand for aug-cc-pVDZ), MG3S,7 and MG3SXP,8 where CCSD(T)-F12a/aDZ calculations are expected to be close to
aDZ is an augmented valence double zeta plus polarization the CBS limit of CCSD(T).17,19 We then compare these
set, MG3S is a minimally augmented valence triple zeta plus calculations to CCSD-F12a/aDZ, SCS-MP3/CBS, and seven
double polarization set, and MG3SXP is MG3S plus extra MCCMs, which are explained next.
polarization functions. We also consider some calculations extra- MCG3/3,11 G3SX(MP3),10 and BMC-CCSD12 are MCCMs
polated to a complete basis set (CBS), and we consider some based entirely on wave function components. MCG3/3 and
multi-coefficient correlation methods9–13 (MCCMs) involving G3SX(MP3) each involve an expensive CCSD(T) step, but
different basis sets for different components, as defined in the BMC-CCSD involves no step more expensive than CCSD and
original references, which are given below. thus is more affordable for large systems.
All geometries were optimized by Huenerbein et al.5 with MCG3-MPW13 and MCG3-MPWB13 are MCCMs that combine
the B97-D/TZVP method. Since the same geometries are used wave function and density functional components with the
for the reference calculations as for the tested methods, the highest-order wave function component being CCSD(T).
comparisons are a direct test of the ability of the tested methods MCQCISD/MPW13 and MCQCISD/MPWB13 are similar
to predict relative energies (DE) at pre-specified geometries. but the highest-order wave function component is the less
But since the geometries predicted by B97-D/TZVP are expected expensive quadratic configuration interaction with single and
to be reasonably close to the accurate equilibrium geometries, double excitations (QCISD). MC3MPW13 and MC3MPWB13
we interpret the results as Born–Oppenheimer isomerization are even cheaper because the highest level of correlation
energies. included is based on the MP2 method. Since the Hartree–Fock

13684 Phys. Chem. Chem. Phys., 2011, 13, 13683–13689 This journal is c the Owner Societies 2011
 View Article Online

Table 1 Density functional applied to all 24 reactions

Functional Year Ref. Typea

BP86 1986–8 20, 21 GGA
B3LYP 1993–4 22, 23 Global-hybrid GGA
PBE 1996 24 GGA
PBE0 1996–8 25, 26 Global-hybrid GGA
tHCTHhyb 2002 27 Global-hybrid meta-GGA
TPSS 2003 28 Meta-GGA
TPSSh 2003 29 Global-hybrid meta-GGA
BMK 2004 30 Global-hybrid meta-GGA
Published on 04 July 2011. Downloaded by Universidad de Salamanca on 5/3/2021 5:26:15 PM.

MPW3LYP 2004 31 Global-hybrid GGA

MC3MPW 2005 13 Doubly hybrid
MC3MPWB 2005 13 Doubly hybrid
MCQCISD-MPW 2005 13 Doubly hybrid
MCQCISD-MPWB 2005 13 Doubly hybrid
MOHLYP 2005 32 Global-hybrid GGA
PW6B95 2005 46 Global-hybrid meta-GGA
M05 2005 33 Global-hybrid meta-GGA
LC-oPBE 2006 34 RS-hybrid GGA
B97-D 2006 35 GGA + MM
M05-2X 2006 36 Global-hybrid meta-GGA
M06-L 2006 37 Meta-GGA
M06-HF 2006 38 Global-hybrid meta-GGA
B2-PLYP 2006 39 Doubly hybrid
mPW2-PLYP 2006 40 Doubly hybrid
M06 2008 41 Global-hybrid meta-GGA
M06-2X 2008 41 Global-hybrid meta-GGA
M08-HX 2008 8 Global-hybrid meta-GGA
M08-SO 2008 8 Global-hybrid meta-GGA
oB97 2008 42 RS-hybrid GGA
oB97X 2008 42 RS-hybrid GGA
oB97X-D 2008 43 RS-hybrid GGA + MM
B2GP-PLYP 2008 47 Doubly hybrid
M06-L-D 2009 37, 44 Global-hybrid meta-GGA+MM
M06-D 2009 41, 44 Global-hybrid meta-GGA+MM
BP86-D3 2010 20, 21, 45 GGA + MM
B3LYP-D3 2010 22, 23, 45 Global-hybrid GGA+MM
PBE-D3 2010 24, 45 GGA + MM
PBE0-D3 2010 25, 26, 45 Global-hybrid GGA + MM
TPSS-D3 2010 28, 45 Meta-GGA+MM
B2P-PLYP-D3 2010 39, 45 Doubly hybrid + MM
B97-D3 2010 45 GGA + MM
B2GP-PLYP-D3 2011 47, 48 Doubly hybrid + MM
TPSSh-D3 2011 29, 49 Global-hybrid meta-GGA + MM
PW6B95-D3 2011 46, 49 Global-hybrid meta-GGA + MM
M05-D3 2011 33, 49 Global-hybrid meta-GGA + MM
M05-2X-D3 2011 36, 49 Global-hybrid meta-GGA + MM
M06-L-D3 2011 37, 49 Meta-GGA + MM
M06-HF-D3 2011 38, 49 Global-hybrid meta-GGA + MM
M06-D3 2011 41, 49 Global-hybrid meta-GGA + MM
M06-2X-D3 2011 41, 49 Global-hybrid meta-GGA + MM
a
‘‘GGA’’ denotes generalized gradient approximation; ‘‘RS’’ denotes range-separated; ‘‘+ MM’’ denotes the addition of a post-SCF empirical
molecular mechanics term.

orbitals on which CCSD(T), QCISD, and MP2 are built and and two of such combined doubly-hybrid-functional-plus-MM
the generalized Kohn–Sham orbitals on which some density methods are also included in the test. All density functionals
functional components are built are both functionals of the considered in this article are listed and explained in Table 1.
electron density, these four methods are all doubly hybrid14 The table contains 40 single-component density functionals
density functional methods. and nine doubly hybrid functionals. The 40 single-component
The other density functionals8,20–49 considered in this work functionals consist of 22 without MM terms and 18 with them.
are all applied to all 24 reactions. For comparison we also The nine multi-component functionals consist of seven without
include three other doubly hybrid density functional methods, MM and two with MM.
namely, B2-PLYP,39 mPW2-PLYP40 and B2GP-PLYP,47 in
which the MP2-like correlation is combined with a DFT calcu-
3. Reference values
lation. An empirical molecular-mechanics (MM) dispersion-like
term can be added to these functionals in a post-SCF fashion Table 2 compares the isomerization energies calculated by
(just as in the single-component functionals discussed below), 11 methods for the six smallest systems in the ISOL24 database.

This journal is c the Owner Societies 2011 Phys. Chem. Chem. Phys., 2011, 13, 13683–13689 13685
 View Article Online

Table 2 Isomerization energies (kcal mol 1) for a 6-reaction subset of the ISOL24 database

Reaction CCSD(T)- CCSD- MCG3- MCG3- MCQCISD- MCQCISD- SCS- BMC- M06-L/ M06-L-
no. F12a/aDZ F12/aDZ MPW MPWB MG3/3 MPW MPWB MP3/CBS CCSD MG3SXP DFa
3 9.77 10.98 11.05 10.67 10.57 11.30 11.02 11.70 11.16 4.58 4.58
9 21.76 21.34 21.21 20.93 22.44 20.90 20.33 22.30 20.20 19.05 19.03
10 6.82 5.93 6.34 6.25 6.92 6.30 6.10 7.90 8.57 2.72 2.70
13 33.52 32.75 33.52 33.48 34.57 33.47 33.30 35.10 34.75 32.44 32.42
14 5.30 4.19 5.22 5.04 5.19 5.43 5.18 5.20 5.33 3.46 3.45
20 4.66 4.70 4.86 4.76 4.69 4.63 4.47 4.50 4.75 3.14 3.15
MUE 0.00b 0.74 0.43 0.45 0.46 0.52 0.66 0.90 1.01 2.74 2.75
Published on 04 July 2011. Downloaded by Universidad de Salamanca on 5/3/2021 5:26:15 PM.

Cost 3000 2600 1900 1900 1700 370 370 190 520 21 12
a b
Same basis but with density fitting. By definition.

The last row of the table gives the ‘‘cost,’’ which is defined as four of them with the MG3 basis set. As for the doubly hybrid
the sum of the computer times (in processor hours) for single- functionals, B2-PLYP, B2GP-PLYP, and mPW2-PLYP with
point energy calculations on reactions 10 and 20 (the two and without MM terms are also tested with the MG3SXP
largest systems in Table 2) divided by the sum of the computer basis set, but MC3MPW and MC3MPWB have different
times for single-point MP2/6-31G calculations on the same two basis sets for different level of calculations. Finally we tested
reactions, run with the same software on the same computer. MCQCISD-MPWB, BMC-CCSD, and SCS-MP3/CBS. In
We recognize that such timings depend on the software, the every case we computed the mean unsigned error (MUE).
computer and its load, and the parallelization, but nevertheless These MUEs are given in Tables 4–6.
they are usually meaningful when they differ by more than a Table 4 compares the results of M06, M06-2X and M06-HF
factor of about 1.5–2. All costs are rounded to two significant with MG3S and MG3SXP basis sets. MG3SXP (where XP
figures. All coupled cluster and QCISD results were calculated denotes ‘‘extra polarization’’) differs from MG3S in that it
with Molpro50 and all density functional calculations were run adds an extra polarization function to MG3S basis. In particular,
with Gaussian 09.51 the 2df polarization functions of MG3S on Li–Ne are replaced
The second last line of Table 1 is the mean unsigned error by a 3df set, and the 3d2f polarization functions on Al–Ar are
(MUE) of each method as compared to the CCSD(T)-F12a/aDZ replaced by a 4d2f set. It has been shown in previous research
results. The BMC-CCSD, SCS-MP3/CBS, and CCSD-F12/ that the extra polarization functions of MG3SXP is necessary
aDZ methods have MUEs of 1.01, 0.90, and 0.74 kcal mol 1 for obtaining accurate results in certain systems containing
respectively, and the four doubly hybrid methods have MUEs fluorine atoms, and considering that there are three reactions
in the range 0.43–0.66 kcal mol 1. The most accurate method (reaction 12, 20 and 22) in the ISOL24/11 database involving
that is affordable for the entire database of 24 reactions is fluorine, it is beneficial to clarify the necessity of extra polari-
MCQCISD-MPW. This method was applied to the other 18 reac- zation. The MUEs of M06 and M06-2X decrease by 1.27 and
tions, and a new reference database was created containing 1.15 kcal mol 1 after adding the extra polarization, while the
CCSD(T)-F12a/aDZ results for the six reactions of Table 2 result of M06-HF with MG3SXP is unexpectedly worse than
and MCQCISD-MPW results for the other 18. This new set of MG3S. In general we believe that MG3SXP provides validating
reference data is called ISOL24/11 where the last two digits results by properly treating the fluorine-containing systems
signify the update of 2011. The isomerization energies of the with the extra polarization functions, and the quality of various
ISOL24/11 database are listed in Table 3. density functionals can be appropriately compared by use of
this basis set.
Table 5 compares the performance of various single-component
4. Testing density functionals and doubly hybrid density functionals against the ISOL24/11
All 40 single-component density functionals were tested database. Empirical dispersion-like MM terms (D-correction)
against ISOL24/11 with the MG3SXP basis set. We also tested can also be added to the DFT calculations in a post-SCF
fashion to improve accuracy for weak interactions, and here
Table 3 The ISOL24/11 database (kcal mol 1) we include three versions of them. The label ‘‘D’’ refers to the
D-correction in ref. 44 except for two cases: in oB97X-D it
Reaction DE Reaction DE refers to ref. 42 and in B97-D, it refers to ref. 35. On the other
1 69.17 13 33.52 hand, ‘‘D3’’ refers to the D-correction in ref. 45 and 46.
2 37.54 14 5.30
3 9.77 15 3.06
4 66.43 16 22.78 Table 4 Mean unsigned error (kcal mol 1) for density functionals
5 32.84 17 10.33 tested with both MG3S and MG3SXP basis sets
6 25.51 18 22.57
7 17.37 19 18.25 MUE
8 22.34 20 4.66 Functionals MG3S MG3SXP
9 21.76 21 11.21
10 6.82 22 0.77 M06-2X 3.65 2.46
11 37.87 23 23.43 M06 4.29 3.06
12 0.20 24 14.94 M06-HF 3.56 4.47

13686 Phys. Chem. Chem. Phys., 2011, 13, 13683–13689 This journal is c the Owner Societies 2011
 View Article Online

Table 5 Mean unsigned errors (kcal mol 1) for 48 density functionals. The only exception is that M06-L-D has a MUE of approxi-
All calculations are performed with MG3SXP basis except for mately 1 kcal mol 1 smaller than M06-L. We attribute the
MCQCISD-MPWB, MC3MPW, and MC3MPWB, which have method-
specific basis sets for each of their components above observations to the fact that the whole family of
Minnesota functionals has incorporated reasonably accurate
Functional MUE Functional MUE Functional MUE medium-range correlation energy by virtue of their functional
MCQCISD-MPWB 0.47 PW6B95-D3 3.09 TPSS-D3 5.91 forms and their parameterization process and adding post-SCF
B2GP-PLYP-D3 1.36 oB97 3.42 M06-L 5.94 dispersion-like corrections thus does not improve their perfor-
MC3MPWB 1.65 PBE0 3.43 B3LYP-D3 6.02 mance significantly. If we restrict ourselves to functionals without
MC3MPW 1.79 BMK 3.46 PBE 6.19
oB97X-D 1.90 mPW2-PLYP 3.74 M05 6.49 Hartree–Fock exchange (which is an important consideration
Published on 04 July 2011. Downloaded by Universidad de Salamanca on 5/3/2021 5:26:15 PM.

PBE0-D3 1.91 BP86-D3 4.20 B97-D 6.71 from the point of view of cost, especially for extended systems),
the best functional without MM terms is M06-L and the best
M08-SO 2.19 B2-PLYP 4.39 TPSSh 6.78 with MM terms is M06-L-D3.
M08-HX 2.27 PBE-D3 4.41 BP86 7.58
B2GP-PLYP 2.33 TPSSh-D3 4.41 TPSS 8.37 We then proceed to consider the more expensive doubly
M06-2X-D3 2.34 PW6B95 4.44 MPW3LYP 8.48 hybrid functionals, in which an MP2-like or QCISD-like term
M05-2X 2.35 M06-HF 4.47 B3LYP 8.98 is added as a post-SCF term to the DFT or weighted DFT
M06-2X 2.46 LC-oPBE 4.73 MOHLYP 13.49
result.
M05-2X-D3 2.48 M06-HF-D3 4.77 Because MP2 and QCISD are more expensive than DFT for
M06-D3 2.59 M06-L-D 4.82 a given basis set for large systems, these methods-when used
oB97X 2.77 B97-D3 5.17 with a single basis set-have the computational cost of MP2
M06-D 2.81 M05-D3 5.61
B2-PLYP-D3 2.86 M06-L-D3 5.73 or QCISD rather than that of DFT; however, this does not
M06 3.06 tHCTHhyb 5.87 affect MCQCISD-type and MC3M-type methods because
they involve two basis sets. It should be noticed that the cost
of MP2 steps can be reduced by using RI-MP2 with appro-
Table 6 Mean unsigned errors (kcal mol 1) for other methods priate auxiliary basis sets, but this is not adopted in this paper.
For all the doubly hybrid functionals based on MP2 as the
Method MUE WFT component, which excludes those in Table 2, but
BMC-CCSD 1.46 includes B2-PLYP, B2GP-PLYP, mPW2-PLYP, MC3MPW,
SCS-MP3/CBS 1.68 and MC3MPWB, the average cost is approximately one to
two orders of magnitude higher than for single-component
functionals. However, these methods are still quite affordable
Although Table 1 contains 49 density functionals, Table 5 with presently available computational resources.
contains only 48 because it omits MCQCISD-MPW, which is Although the MUE of MCQCISD-MPWB is only
used for some of the benchmark values. 0.46 kcal mol 1, which is the smallest of all the tested methods,
We first look at the ‘‘pure’’ density functionals without MM this is partly due to the fact that it is very similar to
corrections. (These are the density functionals that do not MCQCISD-MPW, which is the method we use to calculate
have a suffix-D or -D3.) The popular B3LYP functional has an 18 of the 24 reactions. On the other hand, the value is probably
enormous MUE of 8.98 kcal mol 1, while the best single- reasonable since it is close to the values in Table 2, which are
component functional without D-correction, namely, M08-SO, based on comparison to the very accurate results. MC3MPWB
achieves a MUE of 2.19 kcal mol 1. Several other Minnesota and MC3MPW both achieve good performance without MM
functionals, namely, M08-HX, M05-2X and M06-2X also perform terms, achieving MUEs of 1.79 and 1.65 kcal mol 1, which out-
quite well with MUEs of 2.27, 2.35 and 2.46 kcal mol 1 perform all the single-component functionals. B2GP-PLYP also
respectively. Furthermore, we observe no improvement performs well, and its accuracy is further improved by adding
when comparing PBE with its meta-GGA counterpart TPSS, the MM term.
but the hybrid PBE0 almost halves the error of PBE. The Finally in Table 6 we compare the results for the methods whose
range-separated functional oB97X achieves an MUE of costs scale as N6, namely, SCS-MP3/CBS and BMC-CCSD.
2.77 kcal mol 1, but the other two RS-hybrid functionals, SCS-MP3/CBS is the method used in ref. 5 to construct the
namely, oB97 and LC-oPBE, do not perform as well. This is original ISOL24 database, and we find its MUE against the
well understood; the advantage of oB97X is that it is not updated ISOL24/11 to be 1.68 kcal mol 1. This result is
constrained to have zero Hartree–Fock exchange at small inter- consistent with the error estimation of 2 kcal mol 1 in ref. 5,
electronic distances. and also implies that the original ISOL24 is not accurate
On the other hand, the best MM-corrected functional, enough for evaluation of methods such as B2GP-PLYP-D3
which is also the best single-component functional overall, is or MC3MPW, for which the MUE is even smaller than
oB97X-D with an MUE of 1.90 kcal mol 1. Also surprising is SCS-MP3/CBS. Second, BMC-CCSD is found to have an
the good performance of PBE0-D3, which is almost equally MUE of 1.46 kcal mol 1, which is only slightly worse than
good as oB97X-D and also outperforms all the other single- the best method in Table 5 if we exclude MCQCISD-MPW. In
component functionals in our test. Moreover, most of the spite of the fact that it scales as N6, its most expensive step
Minnesota functionals do not significantly improve by adding only requires a CCSD calculation with a polarized valence
the MM corrections, while other MM-corrected functionals in double-zeta basis set, which should be affordable for a large
general significantly improve over their non-corrected counterparts. number of applications.

This journal is c the Owner Societies 2011 Phys. Chem. Chem. Phys., 2011, 13, 13683–13689 13687
 View Article Online

Table 7 Mean unsigned errors (kcal mol 1) of the 6-reaction subset The results differ from those in Tables 5 and 6 primarily
calculated by all the methods used in this paper with the results because this subset of reactions is smaller and less diverse.
obtained by CCSD(T)-F12a/aDZ taken as reference data

Functional MUE Functional MUE Functional MUE

Acknowledgements
MCG3-MPW 0.43 B2-PLYP 1.28 M05-D3 2.43
MCG3-MPWB 0.45 M06-D3 1.36 M06-HF-D3 2.43 This work was supported in part by the National Science
MG3/3 0.46 oB97 1.48 MPW3LYP 2.46 Foundation under Grant No. CHE03-49122. This research
MCQCISD-MPW 0.52 PBE0 1.50 B97-D3 2.52 was performed in part using the MSCF in EMSL, a national
M08-HX 0.60 M06-2X-D3 1.53 B3LYP 2.63
B2G-PLYP-D3 0.64 M06-2X 1.53 M06-L-D3 2.73 scientific user facility sponsored by the US Department of
Published on 04 July 2011. Downloaded by Universidad de Salamanca on 5/3/2021 5:26:15 PM.

MCQCISD-MPWB 0.66 oB97X 1.55 M06-L 2.74 Energy Office of Basic Energy Research and located at PNNL
and also using a computer time grant from Minnesota Super-
B2G-PLYP 0.67 M06-D 1.67 M05 2.75
CCSD-F12a/aDZ 0.74 B97D 1.75 TPSSh-D3 2.76
computing Institute.
SCS-MP3/CBS 0.90 PW6B95-D3 1.75 TPSSh 3.12
B2-PLYP-D3 1.00 BMK 1.81 TPSS-D3 3.34
BMC-CCSD 1.01 tHCTHhyb 1.82 TPSS 3.67 References
MC3MPW 1.03 PBE-D3 1.89 MOHLYP 4.32
1 J. A. Pople, Rev. Mod. Phys., 1999, 71, 1267.
mPW2-PLYP 1.11 LC-oPBE 1.91
2 J. Zheng, T. Yu, E. Papajak, I. M. Alecu, S. L. Mielke and
D. G. Truhlar, Phys. Chem. Chem. Phys., 2011, 13, 10885.
MC3MPWB 1.14 BP86-D3 1.99 3 See, e.g., A. Karton, E. Rabinovich, J. M. L. Martin and B. Ruscic,
PBE0-D3 1.14 PW6B95 2.00 J. Chem. Phys., 2006, 125, 144108.
oB97X-D 1.14 PBE 2.02
4 K. W. Sattelmeyer, J. Tirado-Rives and W. L. Jorgensen, J. Phys.
M08-SO 1.17 M06-L-D 2.09
Chem. A, 2006, 110, 13551.
M05-2X 1.23 B3LYP-D3 2.14
5 R. Huenerbein, B. Schirmer, J. Moellmann and S. Grimme, Phys.
M06 1.26 BP86 2.30
Chem. Chem. Phys., 2010, 12, 6940.
M05-2X-D3 1.27 M06-HF 2.42 6 R. A. Kendall, T. H. Dunning Jr. and R. J. Harrison, J. Chem.
Phys., 1992, 96, 6796.
7 (a) R. Krishnan, J. S. Binkley, R. Seeger and J. A. Pople, J. Chem.
5. Concluding remarks Phys., 1980, 72, 650; (b) T. Clark, J. Chandrasekhar, G. W.
Spitznagel and P. v. R. Schleyer, J. Comput. Chem., 1983, 4, 294;
We have updated the ISOL24 database constructed in ref. 5 by (c) L. A. Curtiss, K. Raghavachari, C. Redfern, V. Rassolov and
methods of higher accuracy, and the new ISOL24/11 database J. A. Pople, J. Chem. Phys., 1998, 109, 7764; (d) P. L. Fast, M. L.
is composed of six reactions calculated by CCSD(T)/aDZ and Sanchez and D. G. Truhlar, Chem. Phys. Lett., 1999, 306, 407;
(e) B. J. Lynch, Y. Zhao and D. G. Truhlar, J. Phys. Chem. A,
18 reactions by MCQCISD-MPW. This is validated by a 2003, 107, 1384.
careful investigation of a subset of the smallest six reactions 8 Y. Zhao and D. G. Truhlar, J. Chem. Theory Comput., 2008,
by the very accurate CCSD(T)-F12a/aDZ method, which should 4, 1849.
be very close to the CCSD(T) results with a complete basis set, 9 P. L. Fast, J. C. Corchado, M. L. Sanchez and D. G. Truhlar,
J. Phys. Chem. A, 1999, 103, 5129.
and MCQCISD-MPW is then found to be the most accurate 10 L. A. Curtiss, P. C. Redfern, K. Raghavachari and J. A. Pople,
method we can afford for the remaining 18 reactions. The new J. Chem. Phys., 2001, 114, 108.
ISOL24/11 database is recommended for future tests and para- 11 B. J. Lynch and D. G. Truhlar, J. Phys. Chem. A, 2003, 107, 3898.
12 B. J. Lynch, Y. Zhao and D. G. Truhlar, J. Phys. Chem. A, 2005,
meterization of methods aimed at treatment of chemical reactions 109, 1643.
of large size. 13 Y. Zhao, B. J. Lynch and D. G. Truhlar, Phys. Chem. Chem. Phys.,
We have tested 48 single-component and doubly hybrid 2005, 7, 43.
functionals with and without MM terms, together with 2 other 14 Y. Zhao, B. J. Lynch and D. G. Truhlar, J. Phys. Chem. A, 2004,
108, 4786.
methods, against the ISOL24/11 database, and we conclude 15 G. D. Purvis III and R. J. Bartlett, J. Chem. Phys., 1982, 76, 1910.
this paper with recommendations as follows. 16 K. Raghavachari, G. W. Trucks, J. A. Pople and M. Head-Gordon,
(a) oB97X-D and PBE0-D3 are the most highly recommended Chem. Phys. Lett., 1989, 157, 479.
17 G. Knizia, T. B. Adler and H.-J. Werner, J. Chem. Phys., 2009,
single-component functionals with MM corrections, while
130, 054104.
Minnesota functionals with high HF exchange, namely, M08-SO, 18 J. Noga, S. Kedzuch, J. Simunek and S. Ten-no, J. Chem. Phys.,
M08-HX, M05-2X and M06-2X perform the best even without 2008, 128, 174103.
MM terms (D-correction). 19 E. Papajak and D. G. Truhlar, J. Chem. Theory Comput., 2011,
7, 10.
(b) For the more expensive doubly hybrid functionals, whose 20 J. P. Perdew, Phys. Rev., 1986, 33, 8862.
most expensive component is an MP2 step, B2GP-PLYP-D3 21 A. D. Becke, Phys. Rev. A, 1988, 38, 3098.
and MC3MPWB are recommended. 22 A. D. Becke, J. Chem. Phys., 1993, 98, 5648.
(c) The even more expensive MCG3-MPW, MCQCISD-MPW, 23 P. J. Stephens, F. J. Devlin, C. F. Chabalowski and M. J. Frisch,
J. Phys. Chem., 1994, 98, 11623.
and BMC-CCSD methods are also recommended for benchmark 24 J. P. Perdew, K. Burke and M. Ernzerhof, Phys. Rev.Lett., 1996,
of large reactions involving weak-interactions, whenever they are 77, 3865.
affordable. 25 J. P. Perdew, M. Ernzerhof and K. Burke, J. Chem. Phys., 1996,
105, 9982.
26 C. Adamo and V. Barone, Chem. Phys. Lett., 1998, 298, 113.
27 A. D. Boese and N. C. Handy, J. Chem. Phys., 2002, 116, 9559.
Appendix 28 J. Tao, J. P. Perdew, V. N. Staroverov and G. E. Scuseria,
Phys. Rev. Lett., 2003, 91, 146401.
Table 7 shows a comparison of the methods based only on 29 V. N. Starverov, G. E. Scuseria, J. Tao and J. P. Perdew, J. Chem.
the six reactions for which we have the most accurate results. Phys., 2003, 119, 12129.

13688 Phys. Chem. Chem. Phys., 2011, 13, 13683–13689 This journal is c the Owner Societies 2011
 View Article Online

30 A. D. Boese and J. M. L. Martin, J. Chem. Phys., 2004, 121, 49 L. Goerigk and S. Grimme, http://toc.uni-muenster.de/DFTD3/
3405. functionals.html (accessed Jan. 15, 2011).
31 Y. Zhao and D. G. Truhlar, J. Phys. Chem. A, 2004, 108, 50 H.-J. Werner, P. J. Knowles, F. R. Manby, M. Schütz, P. Celani,
6908. G. Knizia, T. Korona, R. Lindh, A. Mitrushenkov, G. Rauhut,
32 N. Schultz, Y. Zhao and D. G. Truhlar, J. Phys. Chem. A, 2005, T. B. Adler, R. D. Amos, A. Bernhardsson, A. Berning,
109, 11127. D. L. Cooper, M. J. O. Deegan, A. J. Dobbyn, F. Eckert,
33 Y. Zhao, N. E. Schultz and D. G. Truhlar, J. Chem. Phys., 2005, E. Goll, C. Hampel, A. Hesselmann, G. Hetzer, T. Hrenar,
123, 161103. G. Jansen, C. Köppl, Y. Liu, A. W. Lloyd, R. A. Mata,
34 O. A. Vydrov, J. Heyd, A. V. Krukau and G. E. Scuseria, J. Chem. A. J. May, S. J. McNicholas, W. Meyer, M. E. Mura,
Phys., 2006, 125, 074106. A. Nicklaß, P. Palmieri, K. Pflüger, R. Pitzer, M. Reiher,
35 S. Grimme, J. Comput. Chem., 2006, 27, 1787. T. Shiozaki, H. Stoll, A. J. Stone, R. Tarroni, T. Thorsteinsson,
Published on 04 July 2011. Downloaded by Universidad de Salamanca on 5/3/2021 5:26:15 PM.

36 Y. Zhao, N. Schultz and D. G. Truhlar, J. Chem. Theory Comput., M. Wang and A. Wolf, Molpro, version 2010, University College
2006, 2, 364. Cardiff, Cardiff, 2010.
37 Y. Zhao and D. G. Truhlar, J. Chem. Phys., 2006, 125, 194101. 51 M. J. Frisch, G. W. Trucks, H. B. Schlegel, G. E. Scuseria,
38 Y. Zhao and D. G. Truhlar, J. Phys. Chem. A, 2006, 110, 13126. M. A. Robb, J. R. Cheeseman, G. Scalmani, V. Barone,
39 S. Grimme, J. Chem. Phys., 2006, 124, 034108. B. Mennucci, G. A. Petersson, H. Nakatsuji, M. Caricato, X. Li,
40 T. Schwabe and S. Grimme, Phys. Chem. Chem. Phys., 2006, H. P. Hratchian, A. F. Izmaylov, J. Bloino, G. Zheng,
8, 4398. J. L. Sonnenberg, M. Hada, M. Ehara, K. Toyota, R. Fukuda,
41 Y. Zhao and D. G. Truhlar, Theor. Chem. Acc., 2008, 120, 215. J. Hasegawa, M. Ishida, T. Nakajima, Y. Honda, O. Kitao,
42 J.-D. Chai and M. Head-Gordon, J. Chem. Phys., 2008, H. Nakai, T. Vreven, J. A. Montgomery Jr., J. E. Peralta,
128, 084106. F. Ogliaro, M. Bearpark, J. J. Heyd, E. Brothers, K. N. Kudin,
43 J.-D. Chai and M. Head-Gordon, Phys. Chem. Chem. Phys., 2008, V. N. Staroverov, R. Kobayashi, J. Normand, K. Raghavachari,
10, 6615. A. Rendell, J. C. Burant, S. S. Iyengar, J. Tomasi, M. Cossi,
44 A. Karton, D. Gruzman and J. M. L. Martin, J. Phys. Chem. A, N. Rega, J. M. Millam, M. Klene, J. E. Knox, J. B. Cross,
2009, 113, 8434. V. Bakken, C. Adamo, J. Jaramillo, R. Gomperts, R. E. Stratmann,
45 S. Grimme, J. Antony, S. Ehrlich and H. Krieg, J. Chem. Phys., O. Yazyev, A. J. Austin, R. Cammi, C. Pomelli, J. W. Ochterski,
2010, 132, 154104. R. L. Martin, K. Morokuma, V. G. Zakrzewski, G. A. Voth,
46 Y. Zhao and D. G. Truhlar, J. Phys. Chem. A, 2005, 109, 5656. P. Salvador, J. J. Dannenberg, S. Dapprich, A. D. Daniels,
47 A. Karton, A. Tarnopolsky, J.-F. Lamère, G. C. Schatz and O. Farkas, J. B. Foresman, J. V. Ortiz, J. Cioslowski and D. J. Fox,
J. M. L. Martin, J. Phys. Chem. A, 2008, 112, 12868. GAUSSIAN 09 (Revision A.02), Gaussian, Inc., Wallingford CT,
48 L. Georigk and S. Grimme, J. Chem. Theory Comput., 2011, 7, 291. 2009.

This journal is c the Owner Societies 2011 Phys. Chem. Chem. Phys., 2011, 13, 13683–13689 13689