WIREs Comput Mol Sci 2014 Goerigk Doublehybrid Density Functionals

Advanced Review
Double-hybrid density functionals

Lars Goerigk1,†∗ and Stefan Grimme2
Double-hybrid density functionals (DHDFs) are reviewed in this study. In DHDFs

parts of conventional density functional theory (DFT) exchange and correlation
are replaced by contributions from nonlocal Fock-exchange and second-order
perturbative correlation. The latter portion is based on the well-known MP2
wave-function approach in which, however, Kohn–Sham orbitals are used to cal-
culate its contribution. First, related methods preceding this idea are reviewed,
followed by a thorough discussion of the first modern double-hybrid B2-PLYP.
Parallels and differences between B2-PLYP and its various successors are then
outlined. This discussion is rounded off with representative thermochemical
examples demonstrating that DHDFs belong to the most robust and accurate DFT
approaches currently available. This analysis also presents hitherto unpublished
results for recently developed DHDFs. Finally, how double-hybrids can be com-
bined with linear-response time-dependent DFT is also outlined and the value of
this approach for electronically excited states is shown. © 2014 John Wiley & Sons, Ltd.
How to cite this article:

WIREs Comput Mol Sci 2014, 4:576–600. doi: 10.1002/wcms.1193
INTRODUCTION follows on a ladder, with each higher rung offering a

means to obtain improved results, albeit at a higher
ohn–Sham density functional theory1 (KS-DFT)
K has become the most important methodology in
routine computational chemistry, but despite being an
computational cost:
exact theory, one has to rely on approximations to the (1) The local-density approximation (LDA)
unknown ‘true’ functional in practical computations. depends on the electron density as the only
As a consequence, there is ongoing immense effort to variable. LDA is the basis of many higher-rung
develop new and improved approximations. However, functionals, but when applied in its pure form it
this also comes with the disadvantage that the users cannot be recommended for general-chemistry
of DFT are faced with a plethora of methods (see e.g., applications, as LDA functionals have been
Ref 2 for a current overview). Perdew’s Jacob’s Ladder derived from the uniform-electron-gas model
analogy allows to bring order to this ‘functional that has a constant density at every point in
zoo’.3,4 It starts in the ‘Hartree hell’ and points space. This consequently causes problems in
toward the desired ‘heaven of chemical accuracy’, the the description of inhomogenous atoms and
ultimate goal desired in any calculation. According molecules.
to this classification scheme, density functionals (DFs) (2) The generalized gradient approximation (GGA)
are assigned to five different categories or rungs as corrects the LDA through an additional den-
sity gradient correction, making the functional
∗ Correspondence to: [email protected] more flexible. Thus, the exchange-correlation
1 School of Chemistry, The University of Sydney, Sydney, New South energy and potential are more realistically
Wales, Australia described in inhomogenous electronic systems.
2 Mulliken Center for Theoretical Chemistry, Institut für Physikalis-
che und Theoretische Chemie, Universität Bonn, Bonn, Germany (3) The meta-generalized gradient approximation
† Present address: School of Chemistry, The University of Mel- (meta-GGA) modifies an underlying GGA
bourne, Parkville, Victoria, Australia method by additionally taking into account
Additional Supporting Information may be found in the online the kinetic-energy density or the density Lapla-
version of this article.
Conflict of interest: The authors have declared no conflicts of interest cian, therefore offering even more flexibility
for this article. compared to GGAs.
576 © 2014 John Wiley & Sons, Ltd. Volume 4, November/December 2014
17590884, 2014, 6, Downloaded from https://wires.onlinelibrary.wiley.com/doi/10.1002/wcms.1193 by University Of Wisconsin - Madison, Wiley Online Library on [21/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
WIREs Computational Molecular Science Double-hybrid density functionals
(4) Hybrid DFs are formally derived from the Double−hybrid

adiabatic-connection formula5 which linearly functionals
links the noninteracting KS-reference system
with the real system of fully interacting elec- Good for
TM chemistry Non−DFT regime
trons. In hybrid functionals, a portion of
Good for main−group
the (semi-)local exchange energy is substituted
thermochemistry
by the corresponding nonlocal Fock-exchange
contribution. LDA, GGA Good for barriers HF, MP2
(5) Functionals containing a virtual KS-orbital
contribution offer an intriguing way of tak- 0 50 100
ing nonlocal electron correlation effects into % Fock−exchange
account, in contrast to the lower-rung function- Decreasing self−interaction error
als that attempt to treat electron correlation at
a (semi-)local level. Better accout of static correlation
improves quality of reference for PT2
A closer inspection of rung five reveals many
possibilities of how to include virtual orbitals into FIGURE 1 | Functional principle of double-hybrids in the DFT and
the KS-DFT framework. Examples are Görling–Levy WFT ‘worlds’.
perturbation theory (GLPT),6,7 the closely related
optimized-effective-potential (OEP) methods,8,9
the random-phase approximation (RPA),10–15 of a good description of dynamic correlation effects.
and generally all methods that use KS orbitals in Double-hybrids therefore have to maintain the right
post-self-consistent-field (post-SCF) treatments origi- balance between those two extrema and usually their
nally developed for wave-function theory (WFT).16–19 amount of Fock-exchange ranges between 50% and
This review focuses on a specific class of 70%. The first DHDF B2-PLYP has been shown to
fifth-rung methodologies, namely the double-hybrid overcome many problems connected with hybrid DFs
density functionals (DHDFs). DHDFs represent an such as the popular hybrid B3-LYP,22,23 and it indeed
efficient approximation to GLPT and they have offers a balanced description of the above explained
gained considerable attention in the community since effects.20,24
the publication of the first double-hybrid B2-PLYP The success of B2-PLYP has sparked the develop-
in 2006.20 The motivation behind B2-PLYP is sim- ment of a large number of double-hybrid functionals,
ilar to that of a hybrid DF. While in the latter only particularly since the year 2010, and an overview of
a portion of the DFT exchange is substituted by those methods is given in Table 1.
Fock-exchange, B2-PLYP additionally replaces part It is now well-established that DHDFs belong
of the DFT correlation functional with a nonlocal to the most accurate and universally applicable meth-
correlation contribution based on a Møller–Plesset ods for electronic ground- and excited-state energetic
(MP) type21 second-order perturbative treatment properties as well as for molecular structures.67–71
(PT2). In fact, one can say that B2-PLYP combines the This review provides a detailed overview of DHDFs
best of both the DFT and WFT worlds. The effects of presented in a historical form. We begin with a
merging these two worlds for an electronic structure short introduction of methods preceding B2-PLYP
description are depicted in Figure 1. that also combine DFT and WFT ideas. Then a
On the left of Figure 1 we have the pure DFT detailed description of the B2-PLYP methodology
world (LDA, GGA, and meta-GGA) and on the right is given, followed by short reviews on subsequent
the non-DFT regime of HF and MP2. In between lie advances, including very recent theoretical insights
hybrid and double-hybrid functionals. While function- based on the adiabatic-connection formula. After
als with very low amounts of Fock-exchange may be this theoretical overview, the overall applicability
good in transition-metal chemistry, higher amounts of DHDFs to general main-group thermochemistry
improve the description of main-group thermochem- (and one transition-metal case)—including ‘real-life’
istry. Larger nonlocal exchange contributions dimin- examples—is demonstrated. Therein, also hitherto
ish the notorious self-interaction error (SIE), which unpublished results for recently developed DHDFs are
consequently leads to a better description of, e.g., reac- presented. We conclude this review by discussing how
tion barrier heights. Functionals with low amounts of the DHDF idea can be successfully applied to elec-
nonlocal correlation on the other hand allow a better tronic excited states within a time-dependent density
implicit description of static correlation, but at the cost functional theory (TD-DFT) scheme.
Volume 4, November/December 2014 © 2014 John Wiley & Sons, Ltd. 577
Advanced Review wires.wiley.com/wcms
TABLE 1 Overview of Various Double-Hybrid Density Functionals with Their Year of Publication, a Short Description, and Their Respective
DFT-Exchange and -Correlation (DFTX/C ) Components1
Functional Year Description DFTX DFTC

20 25
B2-PLYP 2006 First modern double-hybrid with a PT2 contribution replacing parts of the DFT B88 LYP26,27
correlation
mPW2-PLYP24 2006 Similar to B2-PLYP, but with different DFT exchange mPW28 LYP
29
B2K-PLYP, B2T-PLYP 2008 Reparametrized B2-PLYP versions for kinetics and thermochemistry B88 LYP
mPW2K-PLYP29 2008 Reparametrized mPW2-PLYP version for kinetics mPW LYP
30
B2-P3LYP 2008 B2-PLYP variant with long-range corrected PT2 B88 LYP
B2-OS3LYP30 2008 Like B2-P3LYP, but with spin-opposite-scaled31 PT2 (SOS-PT2) contribution B88 LYP
B2GP-PLYP32 2008 Reparametrized general-purpose B2-PLYP version B88 LYP
XYG333 2009 Uses B3-LYP orbitals to obtain double-hybrid energies B88 LYP
34
ROB2-PLYP 2009 Reparametrized restricted-open-shell version of B2-PLYP B88 LYP
B2-𝜋PLYP35 2009 Reparametrized B2-PLYP version for conjugated 𝜋-systems B88 LYP
𝜔B97X-236 2009 Based on a range-separated hybrid functional; Spin-component-scaled37 PT2 𝜔B97X38 B97
(SCS-PT2)
DSD-BLYP39 2010 Modified B2-PLYP version with SCS-PT2; Optimized with DFT-D240 dispersion B88 LYP
correction
PTPSS41 2011 Reoptimized DFT exchange and correlation components; SOS-PT2; optimized TPSS43 TPSS
with DFT-D342 dispersion correction
PWPB9541 2011 Reoptimized DFT exchange and correlation components; SOS-PT2; optimized PW9144 B9545
with DFT-D3 dispersion correction
DH-BLYP46 2011 B2-PLYP version with one empirical parameter derived from B88 LYP
adiabatic-connection formula
PBE0-DH47 2011 Similarly derived as 1DH-BLYP; parameter obtained from theoretical PBE PBE
considerations
DuT,PoT48 2011 Reparametrized DSD-BLYP versions for triple-𝜁 basis sets B88 LYP
LS1-DH49 2011 Modified 1DH-BLYP variant B88 LYP
50 2 51
DSD-PBEP86 2011 SCS-PT2; optimized with DFT-D3 dispersion correction PBE P8652,53
XYGJ-OS54 2011 XYG3 variant with SOS-PT2 B88 LYP
55
B2-PPW91 2012 Parametrized for response properties B88 PW9144
PBE0-256 2012 Modified PBE0-DH version PBE PBE
57
xDH-PBE0 2012 PBE0-DH variant based on PBE0 orbitals and SOS-PT2 PBE PBE
lrc-XYG358 2013 XYG3 variant with long-range corrected PT2 B88 LYP
59 60
revPBE0-DH 2013 Similar to PBE0-DH revPBE PBE
TPSS0-DH59 2013 Similar to PBE0-DH TPSS TPSS
OO-PBE0-DH,OO-PBE0-261 2013 PBE0-DH and PBE0-2 variants with orbital-optimized62,63 PT2 contribution PBE PBE
DSD-PBEhB9564 2013 Similar to DSD-PBEP862 PBE-hole65 B95
DH-TPSS, LS1-TPSS66 2014 Similar to 1DH-BLYP and LS1-DH TPSS TPSS
1 Related methods preceding B2-PLYP are mentioned separately in the text.
2 Other combinations of DFT exchange and correlation functionals have also been tested.
METHODS PRECEDING THE FIRST ̂𝜆

functions. In GLPT, the perturbed Hamiltonian H
DOUBLE-HYBRID reads
Görling–Levy Perturbation Theory H ̂ +V

̂𝜆 = T ̂ ext + 𝜆V
̂ ee + ̂
v𝜆 [𝜌] , (1)
Already in 1993 and 1994, Görling and Levy
introduced a perturbation theory (GLPT)6,7 which where T̂ is the kinetic-energy operator, V
̂ ext is the
resembles that of MP21 for ordinary electronic wave external (normally nuclear-electron) potential,
and V ̂ ee is the Coulomb operator scaled by the Equation (4) can be obtained as the functional
coupling-constant 𝜆, which connects the KS Hamilto- derivative of the perturbation energy components.
nian of a system of noninteracting electrons (𝜆 = 0) to This computationally very demanding step is avoided
the exact Hamiltonian of the real system (𝜆 = 1). The in double-hybrid functionals, i.e., the nonlocal corre-
local external operator ̂ v𝜆 ensures that the electron lation contribution only appears in the energy expres-
density 𝜌 remains constant and exact for all 𝜆. The sion but not in the potential.
zeroth-order Hamiltonian is H ̂ =T ̂ + v𝜆 [𝜌] and the One of the first applications of adiabatic-
̂ ̂
perturbation defined by H = Vext + 𝜆V
′
̂ ee − ̂
v𝜆 [𝜌]. If connection second-order-perturbation DFT was pub-
this ansatz is expanded in the usual way, the following lished by Ernzerhof in 1996; however, the method
first two energy terms of the series are obtained from (based on GGA orbitals) showed large errors for
the KS reference determinant Φs as atomization energies.74 It would be interesting to see
what results would look like if exact-exchange or
E0 + E1 = Ts + Eext + J + Ex , (2)
RPA orbitals were used in such a treatment instead.
However, reviews of this kind have to the best of our
where T s and Eext are their noninteracting kinetic
energy and external potential expectation values, and knowledge not been published yet.
J and Ex are the corresponding electronic Coulomb
(J) and exchange (Ex ) energies, respectively. This is in Methods Combining DFT with MP2
close analogy to the MP case with two notable differ-
In 2004 and 2005, Truhlar and co-workers pro-
ences. First, the KS-determinant (and orbitals) instead
posed multi-coefficient correlation methods by merg-
of the Hartree-Fock (HF) determinant is taken in the
ing scaled wave-function contributions with scaled
expectation value. Secondly, because the expression
DFT energies and they coined those approaches as
of the so-called exact exchange term Ex contains the
‘doubly-hybrid’ DFT.75,76 Different methods were dis-
orbitals in a nonlocal form, GLPT that is already
cussed containing varying higher-order terms with
in first-order requires the solution of additional
some of them including MP4(DQ),77 QCISD,78 or
equations to derive the corresponding local potential
coupled-cluster79 contributions.
v1 = vJ + vx by, e.g., the OEP method. It provides a
Such approaches can either be understood as
procedure to transform the nonlocal HF exchange
extensions to multi-coefficient wave-function meth-
operator into a local variationally optimized exchange
ods, such as Gn methods,80 or as extensions of DFT
potential, for details and approximations to it see Refs
methods to include electron correlation in terms of
8, 9 and 72. The GLPT represents one way of deriving
unoccupied orbitals. In the context of this review, we
the OEP method. The connection here for the DHDFs
follow the latter interpretation and we also restrict
(and conventional hybrids) is that GLPT employs KS
our discussion to multi-coefficient methods containing
orbitals to evaluate Ex . The second-order perturbative
second-order perturbative contributions only, as those
correlation obtains an electron correlation term
are most closely related to modern double-hybrid
∞ ||⟨Φ |V
̂ −̂ |2
v1 |Φs,i ⟩| functionals. For more details on methods including
∑ | s,0 ee |
E2 = (3) terms of higher orders, we refer the reader to the orig-
i=1
Es,0 − Es,i inal publication.76 The energy expression of Truhlar’s
approach with MP2 contributions reads:
where Es,i is the expectation value of the ith excited
determinant with the zeroth-order Hamiltonian, Φs is ( ) DFT∕TZ
∑N EMC3-DFT = c2 ESAC-MP2∕DZ + 1 − c2 EXC
the corresponding determinant, and v1 = j=1 ̂ vJ (j) + [ ]
̂
vx (j) (N is the number of electrons). = c2 EHF∕DZ + c1 EcMP2∕DZ
The infinite sum in Eq. (3) runs (in a complete ( ) DFT∕TZ
+ 1 − c2 EXC (5)
one-particle basis set) over all excited KS determinants
which couple to the ground state by V ee − v1 . As in
the MP case these are the singly and doubly excited where ESAC - MP2/DZ is the energy expression for the
ones so that in the end an MP2-analogous energy scaled-all-correlation MP2 (SAC-MP2) approach,81
expression results. Note, that single excitations do not in which one attempts to overcome basis-set related
necessarily vanish in GLPT opposed to MP2 but their deficiencies in MP2 calculations with a small double-𝜁
effect is considered to be small and they are typically (DZ) basis set by adding a scaled (parameter c1 )
neglected for DHDFs (or related methods73 ). second-order perturbative correlation contribution
MP2∕DZ
Ec to the underlying HF energy EHF/DZ . The
𝛿Ej [𝜌] SAC-MP2 energy is then scaled by the parameter
vj (r) = (4) c2 in Eq. (5) and combined with the result of a
𝛿𝜌 (r)
TABLE 2 DFT Exchange and Correlation Components and Scale Loosely related methods combining DFT and
Parameters in the Four MC3-DFT Methods WFT information in various ways were proposed
MC3BB75 MC3MPW75 MC3MPWB9576 MC3TS76
around the same time, see e.g. Refs 17–19 and 88.
DFTX B88 mPW mPW TPSS
DFTC B95 PW91 B95 KCIS85–87 THE FIRST MODERN DOUBLE-HYBRID
c1 1.332 1.339 1.196 1.256 AND SUBSEQUENT VERSIONS
c2 0.205 0.266 0.257 0.336
The term ‘double-hybrid’—as it is used
X 0.39 0.38 0.38 0.31 nowadays—was for the first time used in Ref 70
and it referred to Grimme’s B2-PLYP functional,
which can be seen as a semi-empirical approximation
separate DFT calculation carried out at the triple-𝜁 to second-order Görling–Levy perturbation theory.20
DFT∕TZ
level (EXC ). This DFT contribution is scaled down From a different point of view, it can also be clas-
by 1 − c2 . The DFT component can either be based sified as an extension of Becke’s hybrid-DFT idea.89
on a hybrid-GGA or hybrid-meta-GGA approach While in hybrid DFT a portion of the DFT exchange
and the scale factor X for the exact exchange in functional is substituted by ‘exact’(HF) exchange, in
these hybrid functionals is a third adjustable param- B2-PLYP some part of the DFT correlation functional
eter in this methodology. The approach was also is additionally replaced by a second-order pertur-
called MC3-DFT and its three parameters were bative correlation (PT2). This latter contribution is
fitted to atomization energies and barrier heights. based on KS-orbitals and all parts contributing to the
Four flavors of this multi-coefficient method were total B2-PLYP energy are based on the same basis set,
developed with varying DFT exchange and corre- in contrast to MC3-DFT methods. The general energy
lation contributions (see Table 2). Overall, these expression for B2-PLYP (and related DHDFs) reads:20
methods were shown to be more accurate than
( )
pure, wave-function-based multi-coefficient meth- EDHDFT
XC
= 1 − aX EDFT
X
+ aX EHF
X
ods and other DFT methods for various chemical ( ) DFT
problems,75,76,82,83 however, they have also been + 1 − aC EC + aC EPT2C
(6)
outperformed by newer DFs.83,84
At this point, we would like to emphasize where EDFTX
is the DFT-exchange energy, EHF X
the
that this approach is characterized by three main HF-exchange energy, EDFT C
the DFT-correlation
features. First, the results of two different calcula- energy, and EPT2 C
the nonlocal second-order perturba-
tions for two different basis sets are combined. The tive energy. EHFX
and EPT2
C
are scaled by the parameters
higher-order wave-function contribution is obtained aX and aC , respectively. In the spirit of the popular
with a double-𝜁 basis to mainly lower the overall B3-LYP22,23 hybrid functional, B2-PLYP uses Becke88
computational cost of the entire approach. Second, (B88) exchange25 for EDFT X
and Lee–Yang–Parr (LYP)
26,27 DFT
the MP2 part is based on HF- and the DFT part on correlation for EC .
KS-orbitals.a Third, the entire hybrid-DFT energy Practically, the B2-PLYP energy is obtained
is scaled down, that means it does not distinguish in two steps. First, the hybrid part of the func-
between the hybrid-DFT exchange and correlation tional, i.e., the first three terms in Eq. (6), is solved
contributions. As we will see below, these points self-consistently. Subsequently an MP2-type calcu-
differentiate this approach from what is nowadays lation based on the converged KS-orbitals is car-
called ‘double-hybrid’ DFT. ried out and the resulting correlation energy EPT2 C
is
Another approach combining DFT with MP2 added to the hybrid-DFT result. The resulting orbitals
treatments has been proposed by Ángyáan et al. are therefore of hybrid-DFT character, in contrast
in 2005.16 In the so-called RSH+MP2 method, a to GLPT2. Note that contrary to standard MP2,
range-separated hybrid DF calculation is carried out single-excitations from the KS-determinant would
and the resulting KS-orbitals and eigenvalues are have to be included in this treatment, since Brilloiun’s
used in a long-range-corrected MP2 treatment. The theorem is not valid. However, in B2-PLYP and related
main intention behind this approach was to tackle approaches these terms are neglected and implicitly
the London-dispersion problem of DFT. Although accounted for by the scaling parameter aC in Eq. (6).
KS-orbitals are used for the perturbative treatment, From a practical point of view, it is straight-
RSH+MP2 differs from the B2-PLYP functional (see forward to implement B2-PLYP into any quantum
next section), as it uses both 100% of DFT and MP2 chemical program that includes DF and MP2 theories.
correlation. Although the cost of a double-hybrid calculation is
TABLE 3 Scale Factors for Exact Exchange and Perturbative benchmark studies carried out on the G2/97 and
Correlation for Various Double-Hybrid Functionals of the Form as G3/0594 test sets reported B2-PLYP results with at that
Presented in Eq. (6) time unprecedented accuracy for a DFT method.20,24
Functional aX aC In particular, it was demonstrated that the error
distributions for B2-PLYP were significantly narrower
B2-PLYP 0.53 0.27
than for B3-LYP and other popular DFs. Subsequent
B2K-PLYP 0.72 0.42 studies showed the superior performance of B2-PLYP
B2T-PLYP 0.60 0.31 (or double-hybrids in general) for the description
B2GP-PLYP 0.65 0.36 of cycloaddition reactions,95 ruthenium-catalyzed
ROB2-PLYP 0.59 0.28 olefine-metathesis reaction,96 radical-stabilization
B2-𝜋PLYP 0.602 0.2731 energies,97 substituent effects in alkanes,98 organic
isomerization reactions,99,100 isodesmic reactions,101
mPW2-PLYP2 0.55 0.25
2
and magnetic coupling constants.102 Also, thor-
mPW2K-PLYP 0.72 0.42 ough benchmark studies revealed an overall better
B2-PPW913 0.2 0.1 performance of B2-PLYP than standard hybrid func-
1 In this version, the scale factor for the LYP correlation energy does not obey tionals and MP2 methods.41,67,103,104 B2-PLYP was
the restriction made in Eq. (6) and a value of 0.795 was proposed. also helpful in joint experimental and theoretical
2 Based on mPW exchange.
3 Based on PW91 correlation. studies investigating, for instance, the synthesis of
alkyl and aryl phosphanes105 or the activation of
comparable to that of MP2, B2-PLYP has been shown small molecules with frustrated Lewis-pairs.106 The
to be much more accurate and robust.67 Moreover, derivation of analytical gradients70 and Hessians107
efficient quantum chemical (QM) programs allow to also allowed to perform geometry optimizations and
speed up the MP2 treatment by making use of the frequency calculations with high accuracy.71
resolution-of-the-identity approximation,90 which is The success of B2-PLYP inspired researchers
the reason why B2-PLYP calculations are practically to further explore this methodology. In most cases,
not much more expensive than standard B3-LYP in subsequent developments were based on the same
routine applications even for a few hundred of atoms. DFT exchange and correlation functionals, but only
The two B2-PLYP mixing parameters aX and with slight variations in, e.g., the parameter val-
aC in Eq. (6) have been fitted to small-molecule ues or the way the orbitals are generated for the
heats of formation of the G2/9791 test set and they PT2-type treatment. Examples are the B2K-PLYP,29
amount to 0.53 and 0.27, respectively (see also B2T-PLYP,29 B2GP-PLYP32 ROB2-PLYP,34 XYG3,33
Table 3). As already mentioned in Section Introduc- and B2-𝜋PLYP35 methods. As shown in Table 3,
tion, too low values of aX combined with too high the parameters in those methods vary from 0.53
values of aC would lead to double-counting effects to 0.72 for aX and from 0.27 to 0.43 for aC .
of static electron-correlation effects caused by the The B2GP-PLYP functional is a refitted version
DFT exchange functional and the PT2 portion. On of B2-PLYP with slightly higher amounts of exact
the other hand high values for aX may be a solution exchange and perturbative correlation and it was
to the notorious SIE problem of DFT, but when proposed to be used for general-purpose (GP) appli-
combined with low values for aC , important (static cations. In fact, in the B2GP-PLYP publication it
and dynamic) correlation effects are neglected. The was shown that double-hybrids do not necessar-
empirically obtained parameters for B2-PLYP also ily have one well-defined global minimum in their
make sense from a theoretical point of view and allow two-dimensional empirical parameter space, but that
an optimal fusion of the best of both the DFT and they rather have an optimal ‘channel’ depending
the MP2 worlds. Recently, a rigorous derivation of on the chemical problem under consideration.32 In
double-hybrid DFT with the adiabatic-connection a variety of benchmark studies—mostly involving
formula confirmed that aC should be lower than aX main-group chemical problems—B2GP-PLYP was
and resulted in values very similar to the ones fitted for shown to give even better results than B2-PLYP.32,41,67
B2-PLYP.92 This topic is discussed further in Section ROB2-PLYP is a reparametrized version for the
One-Parameter and ‘Parameter-Free’ DHDFs. treatment of radicals with the restricted-open-shell
B2-PLYP has become a well-established methodology, while B2-𝜋PLYP was suggested for the
quantum chemical method and the first B2-PLYP treatment of delocalized 𝜋-systems.
publication has been cited meanwhile 650 times.93 Furthermore, double-hybrids were developed
This review can of course only mention a few of that follow the same methodology as B2-PLYP
the many successful B2-PLYP applications. In 2006, in Eq. (6), but use different DFT exchange and
correlation contributions. The mPW2-PLYP method 0.4

B2-PLYP
makes use of the modified-Perdew-Wang (mPW) B2-PLYP-D3
exchange28 rather than of B88 exchange in order to CCSD(T)/CBS
improve the description of noncovalent interactions 0.2
ΔE / (kcal / mol)
at an electronic level. It is in fact the second developed
double-hybrid and although it showed a very good
performance at the beginning, this functional has not
0.0
been used very often. The mPW2K-PLYP method is
a slight modification for the description of kinetics.29
The B2-PPW9155 functional uses Perdew-Wang
(PW91) correlation44 instead of LYP and it has been –0.2
developed for the calculation of response properties.

3.0 3.5 4.0 4.5 5.0
rAr–Ar / Å
DOUBLE-HYBRIDS
AND LONDON-DISPERSION FIGURE 2 | Dissociation curve of the argon dimer with B2-PLYP and
B2-PLYP-D3. Results are based on the aug-cc-pVQZ110,111 AO basis set
Methods that also take into account virtual orbitals which provides results quite close to the basis set limit for this property.
are in principle able to describe important long-range As a comparison CCSD(T)112 /CBS results are shown (based on
correlation effects, such as London-dispersion. This aug-cc-pVTZ/aug-cc-pVQZ extrapolations). ΔE is the interaction energy
also holds true for double-hybrid functionals, but it between the two argon atoms. All calculations were carried out with
has to be kept in mind that they only contain a TURBOMOLE.113
scaled-down portion of PT2 correlation resulting in
dispersion corrections already during the fitting pro-
an incomplete description of these effects. In other
cedures of their functional parameters, such as the
words, DHDFs still suffer from the same ‘missing
PWPB9541 and DSD39,50,64 methods discussed in later
dispersion energy problem’ as other functionals,108
sections.
albeit to a smaller extent, and consequently they
Various dispersion corrections for DFT have
do also benefit from dispersion-correction schemes.
been developed and we refer the interested reader
This was first addressed by Schwabe and Grimme in
to other publications that have tested and compared
2007 in a study in which the authors combined the
those with each other, such as Refs 108, 116, and
B2-PLYP and mPW2-PLYP methods with the additive
117. A widely used approach in dispersion-corrected
DFT-D240 dispersion correction leading to signifi-
DFT is the atomic pair-wise DFT-D3 scheme,42 which
cantly improved results for systems in which disper-
allows to take London-dispersion effects accurately
sion effects play a major role.109 Note, that the concept
into account at a negligible additional computational
of the London dispersion energy that is derived from
cost. DFT-D3 has been published in two flavors.42,114
a multipole approximation together with perturbation
The first version is often referred to as ‘DFT-D3
theory, is only valid in the nonoverlapping distance
with zero-damping’.42 The second version employs
regime. For shorter interatomic (fragment) distances
a rational damping function according to Becke
the term ‘medium-range correlation’ (or alternatively
and Johnson.114,118–120 It is therefore often called
‘overlap-dispersion’) could be used. However, in the
DFT-D3(BJ). DFT-D3(BJ) has been recommended as
following we employ the wording dispersion for both,
the method of choice114 and we will only review
long- and short-range correlation effects and explicitly
this version herein. Whenever the context is clear we
state when a distinction is appropriate.
will skip the suffix BJ in our discussion and refer to
The importance of adding a dispersion correc-
this approach as DFT-D3. With the rational damping
tion to a double-hybrid result is demonstrated in
function the DFT-D3 energy expression reads:
Figure 2 for the dissociation of the argon dimer. How-
ever, we also want to emphasize that dispersion effects
are not only important in noncovalently bound com- -D3 = − 1∑ ∑ CnAB
EDFT s ( √ )n ,
plexes or in conformers of organic- and biochemical disp 2 A≠Bn=6,8 n n AB AB
RAB + a1 C8 ∕C6 + a2
systems, but that it has also been established that they
decisively influence the thermodynamics of chemical (7)
reactions of larger systems in general.67,114,115 There-
fore, we suggest to always use some sort of dispersion where the sum is over all atom pairs AB in the sys-
correction, also in combination with DHDFs. In tem. RAB is the internuclear distance and n equals 6
fact, new DHDFs have been developed that included and 8 which means that DFT-D3 takes into account
the R−6 long-range and R−8 medium-range behav- London-dispersion effects. The double integral in
AB AB
iors of the atomic-pair contributions to the disper- Eq. (8) evaluates the electron density 𝜌 over two
sion energy. The CnAB ’s are the averaged (isotropic) different points in space r and r′ by incorporat-
nth-order dispersion coefficients for atom pair AB ing the nonlocal kernel 𝜙(r, r′ ). The specific form of
that are calculated on-the-fly from atomic disper- 𝜙(r, r′ ) can be found in Ref 125. The original VV10
sion coefficients derived from linear-response TD-DFT functional uses the rPW86PBE51,126 GGA-functional
calculations121,122 . for EDFTXC
. The attenuation parameter b in Eq. (8)
While the R− 6 term provides the correct poten- prevents double-counting effects in the short-range
tial in the long-range region, where practically no con- regime and it is fitted to interaction energies of non-
tribution from the underlying DFT method exists, this covalently bound complexes. Hujo and Grimme fit-
particular form of the damping function also assures ted this parameter to two additional GGA and three
that there is a constant dispersion contribution in hybrid-GGA functionals.127 Recently, the DFT-NL
regions in which two atoms lie spatially very close scheme has also been adjusted to the DHDFs B2-PLYP,
to each other, as theoretically outlined by Koide.123 revPBE0-DH, and TPSS0-DH.59
The parameters a1 and a2 influence the extent of this Note that besides using dispersion corrections it
constant contribution. The medium-range region of is also possible to treat long-range correlation effects
this correction is less well-defined and a specific form by employing a range-separated ansatz for the PT2
of extrapolation is chosen to obtain a seamless con- calculation (see below).
nection between the theoretically justified short- and There is large body of evidence that
long-range potentials. dispersion-corrected DHDFs are powerful meth-
The parameters sn in Eq. (7) are global scal- ods often providing results with an accuracy
ing factors. DFs below the fifth rung of Jacob’s higher than other DFT and MP2 approaches,
ladder have an s6 of unity to assure the correct and thus they have helped to gain insight in vari-
long-range asymptotic behavior of the dispersion ous applications concerning for instance chemical
energy. Double-hybrids, however, need a value reactions or the stability of noncovalently bound
smaller than unity commensurate with their nonlocal aggregates.35,39,41,50,59,64,67,100,109,114,128–136 For rela-
correlation component. It can be obtained by com- tive energies of conformers it has even been shown
paring the dispersion energy contribution of the that dispersion-corrected DHDF results come close
uncorrected double-hybrid in rare-gas dimers with or lie even within the error margin of the CCSD(T)
CCSD(T) reference data.41 The s6 is then equal to complete-basis-set (CBS) limit.41,67,114,137–141
the relative amount of missing dispersion energy. For
current double-hybrids the s6 varies between 0.42 SPIN-COMPONENT-SCALED
(DSD-PBEP86) and 0.82 (PWPB95-D3).41,50,64,67 DOUBLE-HYBRIDS
The scale factor s8 and the parameters a1 and
a2 in Eq. (7) are determined by a least-squares-fit In 2003, the spin-component-scaled MP2 (SCS-MP2)
to accurate binding energies of noncovalently bound method was developed, in which a separate scaling of
complexes and relative energies of conformers. For the correlation-energy contributions of electron-pairs
double-hybrids it was found that a1 and s8 are very with same and opposite spin was proposed.37 The
small and hence they are often set to zero.50,64,67 motivation behind this was that same-spin correlation
A different approach to treat London-dispersion is already described at the HF level and that it is there-
effects in DFT goes back to the van-der-Waals DF, fore overestimated by a conventional MP2 treatment.
in which nonlocal effects evaluated with the help of Indeed, SCS-MP2 turned out to be an improvement
the actual electron density.124 The van-Vorhis-Vydrov over MP2 (see Ref 142). The first method making use
(VV10) variant of this idea, also called DFT-NL, of this approach in a double-hybrid context is Chai
reads125 : and Head-Gordon’s 𝜔B97X-2 functional,36 which is
[ [ ]3∕4 discussed again in a later section as the main focus of
DFT-NL DFT 1 3 this functional lies on a long-range correction to its
EXC = EXC + dr𝜌 (r)
∫ 32 b2 hybrid-DFT part. Martin and co-workers have con-
] ducted thorough studies on combining SCS-MP2 and
1 ′ ( ′) ( ′ ) DHDFs since 2010 and they have suggested function-
+ dr 𝜌 r 𝜙 r, r , b . (8)
2∫ als of the form:39
DSD-DFT
( )
Therein, a standard exchange-correlation func- EXC = 1 − aX EDFTX
+ aX EHF
X
+ cC EDFT
C
tional EDFT is combined with a nonlocal corre-
XC
lation contribution that allows the description of + c EOS-PT2 + c ESS-PT2 + E
O C S C
(9)
disp
TABLE 4 Parameters for Various DSD Methods (Obeying Eq. (9)) and the Dispersion-Corrected PTPSS and PWPB95 Functionals (Eq. (10))1
Functional aX cC cO cS s6 s8 a2
2
DSD-BLYP 0.69 0.54 0.46 0.37 0.50 0.2130 6.0519
3
DuT 0.70 0.59 0.47 0.36 — — —
PoT3 0.68 0.63 0.46 0.27 — — —
DSD-PBEP86 0.70 0.43 0.53 0.25 0.418 0.0 5.65
DSD-PBEhP86 0.69 0.44 0.52 0.23 0.46 0.0 5.6
DSD-PBEB95 0.66 0.55 0.46 0.09 0.61 0.0 6.2
DSD-PBEhB95 0.66 0.55 0.47 0.09 0.58 0.0 6.2
PTPSS4 0.50 0.6255 0.375 0.00 0.75 0.2804 6.5745
5
PWPB95 0.50 0.7315 0.269 0.00 0.82 0.2904 7.3141
1 The last three columns show the parameters for the DFT-D3 correction with Becke-Johnson damping. The parameter a1 (Eq. (7)) is zero in all cases.
2 The DFT-D2 correction was employed for the optimization of this functional and the DFT-D3 parameters were determined later.67
3 DFT-D3 parameters for the Becke-Johnson variant have not been determined yet.
4 A previous version of the DFT-D3 correction (with zero damping) was employed for the optimization of this functional and the DFT-D3 parameters for the
Becke-Johnson variant were determined later.67

5 c = 1 − c , in contrast to the DSD methods.
C O
where the DFT- and HF-related terms are the same as melatonin, and tetrapeptide conformers.50,137–139
SS-PT2 -PT2 are the same- and
in Eq. (6) and EC and EOS
C DSD-PBEhB95 was published in 2013 and showed
opposite-spin contributions to the PT2 energy, scaled promising improvements over DSD-PBEP86, but
by the parameters cS and cO . Here, the scale factor applications of this method are still scarce.64,140
cC for the DFT correlation is independent from the The parameter values for some DSD func-
two PT2 parameters. Additionally, all parameters are tionals are shown in Table 4, but details about
fitted in the presence of a dispersion correction Edisp . other DSD approaches can be found in the original
Therefore, this approach was dubbed DSD-DFT publications.50,64
(‘Dispersion-corrected, Spin-component-scaled One topic that has not been addressed before
Double-hybrid’). The first functional of this type in this review is that double-hybrid functionals
uses B88 exchange and LYP correlation and is show a more pronounced basis-set dependence than
hence called DSD-BLYP.39 DSD-BLYP was first lower-rung DFT methods due to the inclusion of
optimized together with Grimme’s older DFT-D240 the PT2 term, although this dependence is nor-
dispersion correction. Results for various chemical mally smaller than that for pure MP2.67 To obtain
problems such as total atomization energies, barrier the best possible results at a triple-𝜁 level, Chan
heights, Grubbs-catalyzed olefine metathesis, and and Radom have reparametrized the DSD-BLYP
noncovalently bound dimers showed a significant functional specifically for the aug′ -cc-pVTZ+d and
improvement over B2-PLYP (similar to the way in 6–311+G(3df,2p)+d basis sets and dubbed these
which SCS-MP2 is an improvement over MP2). methods DuT and PoT.48 For discussions of basis-set
In a subsequent benchmark study, DSD-BLYP was convergence in DHDFs, see also Refs 143 and 144.
combined with the DFT-D3 correction and it was
shown to be the most accurate DF out of almost 50
functionals for general main-group thermochemistry, SPIN-OPPOSITE-SCALED
kinetics, and noncovalent interactions.67
DOUBLE-HYBRIDS
Later, Kozuch and Martin tested various com-
binations of exchange and correlation functionals In 2004, Head-Gordon and co-workers showed that
to determine the best DSD-DFT methods.50,64 the SCS-MP2 method can be considerably sped up
It turned out that the best combinations were by completely ignoring the same-spin contribution
DSD-PBEP86 and DSD-PBEhB95, where PBEh is to the MP2 correlation energy and by consider-
the PBE-exchange-hole65 functional. ing just electron pairs with opposite spins. This
DSD-PBEP86 was published in 2011 and it has allows evoking a Laplace-transformation145 algo-
been tested successfully for various problems such rithm that brings down the formal scaling of the
as total atomization energies, hydrogen-bonding, resulting SOS-MP2 (spin-opposite-scaled) approach
noncovalently bound complexes, barrier heights, from N5 for (SCS-)MP2 to N4 , with N represent-
pericyclic reactions, and relative energies of alkane, ing the system size or number of basis functions,
respectively.31 Methods which include Coulomb-only this can be assigned to the fixed functional-inherent
correlation diagrams (i.e., opposite-spin in SCS lan- parameters.
guage) and ignore the corresponding exchange com- The least-squares-fit of the DFT parameters and
ponents have a deep relation to RPA and so-called the scale factor cO for PTPSS and PWPB95 was carried
JMP2 methods.14,15,19 out in the presence of the DFT-D3 dispersion correc-
SOS-type approaches can consequently also be tion (with zero damping) on a training set containing
applied to double-hybrids and Head-Gordon and total atomization energies, reaction energies, and non-
co-workers used it for their B2-OS3LYP functional covalent interaction. The resulting values for cO are
in 2008.30 Because B2-OS3LYP additionally employs 0.375 for PTPSS and 0.269 for PWPB95.
range-separation to calculate the PT2 energy, it is A subsequent analysis of both functionals car-
further discussed in the section about long-range ried out with the large GMTKN30 database for
corrected double-hybrids. general main-group thermochemistry, kinetics and
The idea of using spin-opposite-scaling in noncovalent interactions41,104 showed that both func-
double-hybrids has not been pursued for the fol- tionals provide a higher accuracy than their predeces-
lowing two years until in early 2011 Goerigk and sor B2-PLYP, but that overall PWPB95 is more robust
Grimme published two new double-hybrids of the than PTPSS.41 In fact, PWPB95 turned out to be the
general form41 : second-best functional out of 50 on a quadruple-𝜁
basis-set level for GMTKN30 and it was only slightly
ESOS -DHDFT = (1 − a ) EDFT + a EHF surpassed by DSD-BLYP.67 Also it was shown that
XC X X X X
( ) DFT PWPB95 is the only double-hybrid functional with
+ 1 − cO EC + cO EOS -PT2 , (10)
C a basis-set dependence similar to that of lower-rung
functionals and that results are nearly converged for
where the various terms have the same meaning as triple-𝜁 basis sets, making PWPB95 the best functional
in Eqs (6) and (9). The two functionals obeying at this basis-set level in the aforementioned study of
Eq. (10) are called PTPSS and PWPB95. PTPSS uses 50 methods.67 The small basis-set dependence can be
TPSS exchange and correlation, while PWPB95 evokes explained by the only small percentage of the SOS-PT2
the PW91 exchange and B95 correlation functionals. contribution.
They are also one of the early double-hybrids not using A study on dissociation reactions in 3d-metal
LYP correlation. Abandoning LYP in any SOS-PT2 carbonyl compounds showed that PWPB95 gave the
treatment makes perfect sense, as this functional does best results compared to other double-hybrids. That
not include any same-spin correlation and hence an PWPB95 is indeed a promising approach for transition
SOS-DHDF method based on LYP is expected to give metal chemistry was later confirmed by Steinmetz and
only mediocre performance. Grimme in a separate benchmark study focusing on
PTPSS and PWPB95 distinguish themselves in nickel- and palladium-catalyzed reactions.136 Further-
three points from DSD-BLYP, which was developed more, its accuracy and broad applicability was shown
at around same time, and subsequent DSD variants. for strain energies in cyclophanes,132 relative energies
First, the amount of Fock-exchange was not fitted but of biologically relevant tetrapeptide conformers,138
limited to the ‘adiabatic-connection’ value of 50% and and in applications to large systems, such as frustrated
thus lower than the optimized values for DSD methods Lewis-pairs, the dissociation energy of a substituted
that range around 70% (see Table 4). This limitation dimeric hydroquinone, and ligand-exchange and dis-
to 50% was seen as a reasonable compromise to suf- sociation in a dimeric palladium compound.141
fice the need of high amounts of Fock-exchange for
main-group chemistry without deteriorating results
for transition-metal chemistry too much (compare FURTHER DEVELOPMENTS
Figure 1). Second, the scale factors for DFT and
perturbative correlation are coupled (see Eq. (10)). One-Parameter and ‘Parameter-Free’
Third, the inherent parameters of the DFT contribu- Double-Hybrid Functionals
tions (seven for TPSS and five for PW91-B95) were In 2011, Sharkas et al.46 analyzed DHDFs with the
also adjusted, as such an approach had been shown help of the adiabatic-connection formula5 which is
to be successful for GGA and hybrid functionals in similar to that used to derive GLPT (see Eq. (1)).
the past.28,60,104,146 Kozuch and Martin have also Considering that the Hartree and exchange
explored the spin-opposite-scaling idea, but the result- contributions depend linearly on the perturba-
ing DOD-DFT methods turned out to be less accurate tion parameter 𝜆 and the correlation contribution
than their DSD-DFT counterparts.39,50,64 Tentatively nonlinearly, the authors were able to derive an
energy expression for a double-hybrid functional emphasize that the parameters have not been fitted.
that only depends on 𝜆 (see Refs 46 and 147 for However, one has to admit that the choice of 𝜆 still
more details). Because this functional also contains a allows for some arbitrariness as can be seen for
correlation contribution
[ ] that depends on a 𝜆-scaled PBE0-DH and PBE0-2. For similar approaches see
density (EDFT
C
𝜌 1∕𝜆 ), the resulting approximation was also Refs 57 and 59.
dubbed ‘density-scaled one-parameter double-hybrid’ Inspired by the success of the meta-GGA
(DS1-DH) approximation:46 double-hybrids PTPSS and PWPB95 in 2011, Souvi
et al. also suggested to use TPSS-exchange and corre-
EDS1 -DHDFT = (1 − 𝜆) EDFT [𝜌] + 𝜆EHF + EDFT [𝜌] lation in Eqs (12) and (13) resulting in the 1DH-TPSS
XC X X C
[ ] and LS1DH-TPSS methods.66
− 𝜆2 EDFT 𝜌 1∕𝜆 + 𝜆2 PT2
E . (11)
C C The number of one-parameter or parameter-free
functionals obeying the above equations has thus
By neglecting the density-scaled
[ ] term
become large, however, applications of them are lim-
through the assumption EDFT 𝜌 1∕𝜆 ≈ EDFT
[𝜌] one
C C ited. Also, with only two exceptions,59,133 the com-
derives the ‘one-parameter double-hybrid’ (1DH)
bination of these methods with dispersion-corrections
approximation46 :
has not yet been investigated. Studies involving geome-
E1DH -DFT = (1 − 𝜆) EDFT [𝜌] + 𝜆EHF tries, atomization energies, barrier heights and also an
XC X X
( ) DFT analysis of parts of the GMTKN30 test set (all without
+ 1 − 𝜆 EC [𝜌] + 𝜆2 EPT2
2
C
(12) dispersion corrections) have revealed that these meth-
ods are definitely better than the underlying B3-LYP
This equation is identical to the original and PBE0148,149 hybrids,150,151 but a closer inspection
double-hybrid formulation in Eq. (6) with the only of the results shows that the more empirical B2-PLYP,
difference that the scale parameters for Fock-exchange B2GP-PLYP, PWPB95, and the DSD functionals seem
and nonlocal correlation are now coupled with each to be overall more robust and accurate.41,42 Never-
other. Practically, 1DH-DFT calculations are car- theless, the herein reviewed derivation is of academic
ried out the same way as the previously discussed value and helps emphasizing the theoretical justifica-
functionals and no additional code modification is tion for double-hybrid functionals.
needed. The 1DH-BLYP functional is based on Eq.
(12). It contains B88 exchange and LYP correlation
and a fit to a small training set of total atomization Long-Range Corrected Double-Hybrids
energies and barrier heights resulted in 𝜆 = 0.65 and In 2008, Head-Gordon applied the idea of
𝜆2 = aC ≃ 0.42, which are values very close to the range-separation, which is well known for the
B2GP-PLYP functional. exchange part of hybrid DFs,38,152,153 to the
Shortly afterwards, the same authors and Adamo nonlocal correlation portion of the B2-PLYP func-
reanalyzed the above derivation from the point of tional to improve the description of van-der-Waals
view of a high-density limit and proposed the ‘lin- complexes:30
early scaled one-parameter double-hybrid’ (LS1-DH)
approximation, which is technically the same as before EB2 -P3LYP = 0.47EB88 + 0.53EHF
XC X X
but now introduces a cubic dependence of 𝜆:49 ( )
+ 0.73ELYP
C
+ aC EPT2
C
𝜔, c𝜔 . (14)
E1DH -DFT = (1 −𝜆) EDFT
[𝜌] + 𝜆EHF
XC X X
( ) This so-called B2-P3LYP functional uses the
+ 1 − 𝜆3 EDFT
C
[𝜌] + 𝜆3 EPT2
C
(13) same exchange and DFT correlation parts as B2-PLYP
with the same scale factors, however it introduces a
A fit of the coupling-constant yielded 𝜆=0.75 modified PT2 correlation term of the form:
and 𝜆3 = aC = 0.42 for the LS1-DH method, again with
( )2
B88 exchange and LYP correlation. occ virt || ia |̂ | ja ||
( ) 1 ∑ ∑ | |g 𝜔 (r) | | ,
Equation (13) also served as a motivation for EPT2 𝜔, c𝜔 = (15)
a PBE-based functional called PBE0-DH.47 How-
C 4 ij ab 𝜖i + 𝜖j − 𝜖a − 𝜖b
ever, instead of obtaining 𝜆 from an empirical fit,
its value was set to 𝜆 = 0.5, inspired by Becke’s in which i and j denote occupied and a and b
HalfandHalf-hybrid approach.89 Chai and Mao sug- denote virtual orbitals with their respective eigen-
gested to set 𝜆3 to 0.5 instead, which resulted in values 𝜖. In contrast to conventional (second-order)
the PBE0-2 functional.56 These latter two methods (MP) perturbation theory, the integral ia ||̂
g𝜔 (r)|| ja
were also called ‘parameter-free’ double-hybrids to is not evaluated with the standard two-electron
operator ̂g (r) = 1r , with r being the interelectronic OO-MP2 and OO-SCS-MP2 methods in which
distance, but with a range-separated operator the (SCS-)MP2 energies are minimized with respect
of the form to the MP2 amplitudes and orbital rotations.62,63 The
same approach has been applied to the PBE0-DH
1 erf (𝜔r) and PBE0-2 functionals, and although it is slightly
̂
g𝜔 (r) = + c𝜔 , (16)
r r more expensive than conventional DHDFs, it has
been shown to be particularly useful for open-shell
in which erf is the error-function. c𝜔 determines the molecules and symmetry-breaking problems in
weight and 𝜔 controls the asymptotic limit of the cor- potential-energy or first-derivative surfaces.61
relation contribution. These two parameters and the
overall scale factor aC were determined empirically to
be 1.5, 0.2 and 0.27. A variant called B2-OS3LYP con-
tains only SOS-PT2 correlation with the parameters APPLICATIONS
c𝜔 = 1.6, 𝜔 = 0.3, and aC = 0.32. Assessment of Double-Hybrids for General
In 2009, Chai and Head-Gordon used their
Main-Group Chemistry
𝜔B97X long-range corrected hybrid functional38 as
Carrying out thorough benchmark studies of newly
a basis for a new double-hybrid called 𝜔B97X-2, in
developed quantum chemical methods has become a
which the 𝜔B97X part is reparametrized and com-
crucial element in the assessment of their universal-
bined with the SCS-PT2 idea.36 Note that this func-
ity and robustness before applying them to unknown
tional does not make use of a range-separated ansatz
problems. The large GMTKN30 database for general
for the nonlocal correlation as described in Eq. (15).
Until now, applications of range-separated main-group thermochemistry, kinetics, and noncova-
double-hybrids have been scarce. B2-P3LYP behaves lent interactions has been shown to be very use-
similarly to B2-PLYP for atomization energies and ful in this context.41,104 It comprises 841 reference
reaction energies, slightly worse than B2-PLYP for energies—based on experimental or high-level theo-
barrier heights and better than dispersion-uncorrected retical reference data—ordered into 30 different test
B2-PLYP for binding energies of weakly bound sets. These test sets can be divided into the three cat-
complexes.30 B2-OS3LYP turned out to give slightly egories ‘basic properties’ (e.g., atomization energies,
larger deviations from accurate reference values than electron affinities or reaction barrier heights), ‘reac-
B2-P3LYP, probably because of the aforementioned tion energies’ (mostly of organic-chemical nature),
problem of the lack of same-spin correlation in and ‘noncovalent interactions’ (inter- and intramolec-
LYP-based SOS-DHDFs. 𝜔B97X-2 was shown to be ular). To facilitate the analysis of GMTKN30, the
particularly good for systems where SIEs are known authors suggested to combine the separate mean abso-
to be severe.36 lute deviations (MADs) of the 30 test sets to one
A long-range-corrected variant of the XYG3 value, a so-called weighted total mean absolute devi-
functional was recently tested for 15 of the 30 sub- ation (WTMAD). This WTMAD takes also the size
sets of the GMTKN30 database of general main-group and ‘difficulty’ of each test set into account (see Ref
thermochemistry, kinetics, and noncovalent interac- 41 for more details). WTMADs only are discussed in
tions, and shown to be an improvement over XYG3 the remainder of this section.
and dispersion-corrected B3-LYP.58 However, when GMTKN30 was first used in 2011 for the
comparing those with separately published results for assessment of the at that time newly developed
dispersion-corrected double-hybrids41 and the same PWPB95-D3 double-hybrid in comparison with four
benchmark sets, one only observes a slight improve- other dispersion-corrected DHDFs.41 Later in the
ment in four cases. same year, those five DHDFs were then compared
Including long-range corrections in with 42 lower-rung dispersion-corrected DFs: two
double-hybrids is an intriguing approach and it is LDAs, 14 GGAs, three meta-GGAs, and 23 hybrids.67
expected that more developments will be published in One of the main outcomes of that particular study
the future; for recent derivations and developments in is that the results nicely reproduce the Jacob’s Lad-
this area see also Refs 154 and 155. der scheme. WTMADs averaged over all methods
belonging to each specific rung are shown in Figure 3.
LDA functionals show the largest errors with an aver-
Orbital-Optimized Double-Hybrids aged WTMAD of 12.0 kcal/mol. GGAs present a sig-
Peverati and Head-Gordon have very recently deter- nificant improvement to 5.3 kcal/mol with a further
mined DHDF energies within a fully orbital-optimized decrease to 4.4 kcal/mol for meta-GGAs. On average,
(OO) framework.61 This idea stems from the hybrid functionals show a WTMAD of 3.3 kcal/mol
TABLE 5 Weighted Total Mean Absolute Deviations (WTMADs) in

kcal/mol for the Complete GMTKN30 Database and its Three Categories
Double-
Hybrid 1.8 Basic Reaction Noncovalent Complete
Method Properties Energies Interactions GMTKN30
DSD-BLYP1 2.0 1.3 0.37 1.3
Hybrid 3.3 2
DSD-PBEP86 2.0 1.4 0.39 1.3
meta- DSD-PBEB952 1.8 1.4 0.60 1.3
GGA 4.4 B2GP-PLYP-D3 1
2.4 1.5 0.45 1.5
PWPB95-D31 2.2 2.0 0.63 1.7
GGA 5.3
1
B2-PLYP-D3 2.7 2.0 0.50 1.8
LS1-DH-D32 3.7 1.5 0.47 2.0
2
PBE0-2-D3 3.6 2.7 0.59 2.4
PBE0-DH-D32 5.3 3.6 0.81 3.4
3
LDA 12.0 MP2/CBS 5.7 3.6 0.90 3.6
SCS-MP2/CBS3 5.1 1.8 1.15 2.9
FIGURE 3 | Jacob’s Ladder for GMTKN30. The values shown are Values were obtained with the (aug′ )-def2-QZVP111,156 basis set.
1 Values are taken from Ref 41 and reassessed with the newer DFT-D3 variant
weighted total mean absolute deviations in kcal/mol averaged for each
with Becke-Johnson damping.
rung of the ladder and obtained at the quadruple-𝜁 basis-set level (see 2 Calculations were carried out with ORCA 3.0.1.157
3 Values are taken from Ref 67.
Ref 67).
and they are clearly surpassed by double-hybrids with PBE0-DH and we reassess this method for the entire
an average value of 1.8 kcal/mol, thus nicely show- GMTKN30 set. DFT-D3 parameters have also been
ing that DHDFs as fifth-rung functionals do indeed obtained for LS1-DH and PBE0-2 (see Supporting
provide improved results. As already mentioned in Information for more details). Finally, we also reassess
the above sections, dispersion-corrected DSD-BLYP the B2-PLYP, B2GP-PLYP, DSD-BLYP, and PWPB95
and PWPB95 were the best functionals in this par- approaches together with the newer DFT-D3 variant.
ticular study. It is also noteworthy that the range of As mentioned above, the term DFT-D3 in this
WTMADs on the fifth-rung was smaller than for the article refers to the version with Becke-Johnson damp-
lower rungs, thus most DHDFs can be recommended ing, unless noted otherwise.
over hybrids for applications. The MADs for all test sets of GMTKN30
The GMTKN30 study in 2011 was carried and all nine methods are shown in the Supporting
out with the first version of the DFT-D3 correc- Information. All results have been obtained at the
tion with zero damping. Since then the latest ver- same def2-QZVP156 quadruple-𝜁 basis-set level as
sion with Becke-Johnson damping (see Eq. (7)) has the study from 2011, with additional diffuse func-
been published and, moreover, more DHDFs have tions where appropriate (see Ref 67). We restrict
been developed. Herein, we take both developments our discussion to WTMADs only. Table 5 shows
into consideration and present an updated analysis WTMADs for the three categories of GMTKN30
of GMTKN30 with a total of nine DHDFs, all dis- and for the entire database. Also shown are WTM-
persion corrected with the Becke-Johnson variant of DADs for the MP2 and SCS-MP2 approaches at
DFT-D3. Four of these nine methods have not been the CBS limit taken from Ref 67. Overall, DHDFs
assessed for GMTKN30 before: LS1-DH, PBE0-2, outperform both MP2 approaches, with the excep-
DSD-PBEP86, and DSD-PBEB95. The latter has been tion of PBE0-DH-D3, for which the WTMAD for
used instead of the recommended DSD-PBEhB95 for the complete GMTKN30 set is with 3.4 kcal/mol
technical reasons. However, a closer inspection of the higher than for SCS-MP2/CBS (2.9 kcal/mol) and
original paper shows that both functionals yield sim- very close to MP2/CBS (3.6 kcal/mol). Particularly for
ilar results and that also their parameters are similar PBE0-DH-D3 we notice a large range in the MADs
(see Table 4).64 Additionally, the PBE0-DH functional (see Supporting Information) which is an indication
is assessed. In a recent study it has only been tested for that it seems to be less robust than the other DHDFs.
parts of GMTKN30 and dispersion-corrections have A comparison between all nine DHDFs shows that
been neglected.151 For our purpose, we have fitted the the two ‘parameter-free’ approaches PBE0-DH-D3
DFT-D3 correction with Becke-Johnson damping for and PBE0-2-D3 and the one-parameter functional
TABLE 6 Results (def2-QZVP AO Basis Set) for the Three Chemical Reactions in kcal/mol
DHCH DIMPD FLPH2
no D3 D3 no D3 D3 no D3 D3
B3-LYP −14.6 1.4 −11.9 −41.5 −18.6 −19.4
1DH-BLYP 15.0 — −34.1 — −19.5 —
B2GP-PLYP 12.1 18.0 −31.5 −42.9 −20.2 −20.4
B2-PLYP 4.4 12.7 −27.4 −42.8 −19.3 −19.6
B2-PPW91 −8.6 — −22.8 — −19.2 —
DSD-BLYP 23.7 29.4 −39.8 −50.9 −19.0 −19.3
DSD-PBEB95 16.1 21.7 −31.9 −41.9 −19.8 −20.1
DSD-PBEP86 22.6 27.1 −35.5 −43.8 −20.3 −20.8
LS1-DH 17.3 22.1 −33.4 −42.8 −21.0 −20.8
PBE0-2 27.3 30.3 −41.5 −47.3 −22.8 −22.6
PBE0-DH 1.7 9.8 −28.8 −44.0 −24.0 −24.3
PWPB95 6.9 12.6 −28.3 −39.1 −19.4 −19.3
Reference 17 ± 2 −32 ± 3 −21.6 ± 1
LS1-DH-D3 are all outperformed by the empirically dimeric hydroquinone derivative (DHCH), the ligand
fitted methods. This mirrors previously reported find- exchange and dissociation reaction of a dimeric palla-
ings for GGA and meta-GGA functionals that showed dium species (DIMPD), and the activation of H2 by a
that nonempirical functionals are overall not neces- so-called frustrated Lewis Pair (FLP). The results for
sarily better than empirical ones for broad chemistry all tested functionals are given in Table 6, the reaction
applications.104 Note that GMTKN30 is by far big- formulas and structures are shown in Figure 4. In two
ger than the respective training sets used in the tested cases zero-point energy exclusive, gas phase reaction
methods’ development and that therefore this analysis energies ΔE are taken as reference (obtained from
is a reliable cross-validation. back-corrected experimental free reaction energies141 )
The WTMADs for the complete GMTKN30 and these values can be compared directly to com-
database for the remaining six DHDFs are all in puted energy values. In the FLP example, a theoretical
a range between 1.8 kcal/mol (B2-PLYP-D3) and DLPNO-CCSD(T)158 value are taken as reference.
1.3 kcal/mol (DSD functionals). It is interesting to The systems are good examples for the importance
see that averaged over a large number of chemical of exchange-correlation (and SIE) effects on various
problems all three DSD methods have the same length scales. They also demonstrate that an accuracy
WTMAD. Also for the three categories of GMTKN30 of 2–3 kcal/mol or better poses a challenge for even
the three methods lie very close to each other, with
the most sophisticated DFs when large systems are
the only exception of DSD-PBEB95 that has a higher
involved. In these examples the reaction energy is
WTMAD for noncovalent interactions (0.60 kcal/mol)
strongly influenced by London-dispersion interaction
than DSD-BLYP (0.37 kcal/mol) and DSD-PBEP86
which is typical for molecules with 50–100 atoms.
(0.39 kcal/mol).
Eleven different DHDFs together with the well-known
Overall these results show that the empirical
B3-LYP functional for comparison have been applied
DHDFs can all be recommended for general chemistry
applications as the error margins of all methods are to the reactions.
rather small compared to other functionals from the It was recently shown experimentally159 that the
2011 study. 2,6-di-tert-butyl-4-metoxyphenoxyl radical dimerises
in solution and in the solid state (reaction a). In the
same study, a bond dissociation Gibbs free energy
Applications to Thermochemistry of Larger around zero (−0.2 ± 0.1 kcal/mol) was measured
Molecules leading to a ΔE value of 17 ± 2 kcal/mol. In this
In this section, we discuss DHDF results for three case the DFT description of the monomers can suf-
‘real-life’ chemical applications involving rather fer from spin-overdelocalization, which leads to an
large systems: the dissociation of a substituted, overstabilization of the monomers with respect to the
O
(a) t-Bu
t-Bu t-Bu t-Bu
O
O
2
O
O
t-Bu
t-Bu O
(b)
t-Bu t-Bu
N N
Cl Cl
Pd Pd + 2 PPh3 2 Pd
Cl PPh3
N
t-Bu
t-Bu
(c) C6F5 t-Bu C6F5 t-Bu

– +
B P + H2 B H H P
C6F5 t-Bu C6F5 t-Bu
C6F5 t-Bu C6F5 t-Bu
FIGURE 4 | Reaction formulas and optimized structures of the three reaction energy examples.
dimer. This makes the system very challenging also In the second example a Pd–Pd bond is
for DHDFs. quenched by a triphenylphosphane ligand yielding the
The DFT results for the dissociation are given corresponding mono-palladium complex. The best
in Table 6. The strong effect of dispersion and estimate experimental reference value for the reaction
SIE is obvious from the B3-LYP and B3-LYP-D3 energy is ΔE = −32 ± 3 kcal/mol.141 The relatively
results, respectively, which deviate by about 30 and large uncertainty of ±3 kcal/mol reflects the problems
15 kcal/mol from the reference value. It is clear that associated with obtaining reliable reference values for
only methods with a large fraction of Fock-exchange large systems, and a relative error of about ±5%–10%
can yield good results in this case. Most DHDFs is a realistic estimate when data measured in solution
perform much better, although we note a tendency are used for back-correction.
for overestimation of the reaction energy. Good Most DHDFs overshoot the reaction energy in
DF performers are the dispersion-corrected B2-PLYP, particular when dispersion corrected. Because such
B2GP-PLYP, and PWPB95 functionals. double-counting effects are normally not observed for
main-group thermochemical problems, we attribute In their approach excitation energies are

the deviations to the effect of the metal for which obtained in two steps, similar to DHDFs for elec-
perturbation theory can be problematic. In this case tronic ground states. First, a standard linear-response
MP2 itself shows a large error (MP2/CBS: ΔE = − 54, time-dependent density functional theory (TD-DFT)
SCS-MP2/CBS: ΔE = − 41) which transfers directly treatment is carried out for the hybrid portion of
(but scaled) to the DHDF result. This view is sup- the double-hybrid by solving the familiar RPA-type
ported by the B2-PPW91 functional which yields the eigenvalue problem170,171 :
absolutely smallest value of all uncorrected methods ( )( ) ( )( )
and which has only a small perturbation contribution A B X 1 0 X
= ΔETD-DFT , (17)
of 10%. B A Y 0 -1 Y
The so-called FLP, i.e., pairs of Lewis acids and
bases, that do not quench each other due to the steric where ΔETD - DFT is the excitation energy, X and
bulk of their substituents can heterolytically split the Y are solution vectors describing single excitations
H2 -molecule.160,161 The quantum-chemical descrip- and de-excitations, and A and B are related to the
tion of FLP reactions has attracted a lot of interest corresponding Hamiltonian matrices between these
recently.162–167 Here we study the reaction energy of (de-)excitations.
the original [B(C6 F5 )3 ]/[P(tBu)3 ] system168 with H2 . The resulting excitation energy is then per-
Because the thermodynamic properties have never turbatively corrected with a CIS(D)-type172 energy
been measured accurately (the reaction is practically contribution Δ(D) . CIS(D) has originally been devel-
irreversible in common solvents, i.e., ΔG ≪ 0) we oped as a second-order perturbation correction of
take as reference the value of −21.6 kcal/mol from a configuration-interaction-singles (CIS) excitation
DLPNO-CCSD(T)/CBS treatment, which should be energies and it involves single and double excita-
accurate to about 1–2 kcal/mol.141 Note that the reactions with respect to a CIS determinant or in other
tion energy is calculated relative to the weakly bound words double and triple excitations with respect to
[B(C6 F5 )3 ]/[P(tBu)3 ] donor–acceptor complex and the underlying HF determinant. In the framework of
separate H2 and not with respect to all free reactants. double-hybrid DFT this procedure is applied to the KS
As can be seen from Table 6, the DFDH results determinant of the hybrid portion and in spin-orbital
are relatively close to each other and also to the form it is described as:
reference value. Although the structural changes are ( )2
ij
large, i.e., splitting of a strong single bond and for- ∑ ∑ uab
1
mation of a zwitterionic structure with two new bond Δ(D) = via tai −
ia
4 ijab 𝜖a + 𝜖b − 𝜖i − 𝜖j − ΔETD-DFT
types, it seems that the FLP reaction is electronically
unproblematic. The span of the values from −19.6 to (18)
−20.8 kcal/mol for the dispersion-corrected DHDFs is with
much smaller than in the previous example. Within 1∑ j ik
the mentioned 1–2 kcal/mol error bar, the values are vai = ⟨jk||bc⟩[tbi tca
jk
+ taj tcb
ik
+ 2tb tac ]
2 jkbc
in good agreement with the DLPNO-CCSD(T)/CBS
ij
∑
reference energy. However, also in this case in which uab = ⟨ab||cj⟩tci − ⟨ab||cj⟩tcj
long-range dispersion makes only a small contribution c
(due to the choice of the reference point for the reac- ∑
− ⟨ka||ij⟩tbk − ⟨kb||ij⟩tak . (19)
tion energy calculation), the DHDFs perform mostly
k
better than standard B3-LYP. A single but notable
outlier in the list of DHDFs is PBE0-DH that already Again, i, j, and k refer to occupied orbitals
dispersion-uncorrected overshoots the reaction energy in the reference determinant, a, b, c to virtual
by 2.4 kcal/mol. orbitals, and t stands for single and double exci-
tation amplitudes, respectively. In the framework
of double-hybrids, these amplitudes are taken from
DOUBLE-HYBRIDS FOR ELECTRONIC the TD-DFT treatment (Eq. (17)). Thus, the overall
excitation energy can be written as:
EXCITED STATES
So far, we have outlined developments that treat ΔETD-DHDFT = ΔETD-DFT + aC Δ(D) , (20)
electronic ground-state properties. However, already
in 2007 Grimme and Neese suggested a way to treat where aC is the well-known scale parameter of
electronic excited states within a DHDF scheme.169 ground-state DHDFs (see Eq. (6)). This approach has
CF3
N O O
+H2N O NH2
O 0.51 0.50
H
N F N 0.5 MD
N
MAD
F N
B C6F5 0.4
H
O
C6F5 0.31
0.3
R O O R 0.23 0.24 0.24
0.20
MD and MAD / eV
0.2 0.16 0.17
N N
0.1
R O O R
R=H 0.0
(C2H5)2N O O O
O –0.01 0.18 0.23 0.00
S –0.1
N
–0.09 –0.11
S
N O –0.2
O –0.22
–0.3
–0.4
HN O NH N
–0.5
–0.49 –0.48
BLYP
PBE
B3-LYP
PBE0
B2-PLYP
B2GP-PLYP
PBE0-DH
PBE0-2
CC2
Cl
N
N
N
Cl
FIGURE 5 | Structures of the dye test set (left).68,69 Mean (MDs) and mean absolute deviations (MADs) in eV for the entire test set for various
TD-DFT methods and for the CC2 approach (right). Values are taken from Refs 69 and 175 and calculated herein for TD-PBE and TD-PBE0 (with
TURBMOLE113 ). All values have been obtained with the def2-TZVPP156 AO basis set.
been first developed and tested for the B2-PLYP169 these methods are more accurate than the underly-
and B2GP-PLYP68 functionals with exactly the same ing wave-function based, perturbative CIS(D)-type
scale factors as for their ground-state counterparts. approaches, thus paralleling what is known from
Note that this approach is a correction for excitation ground-state thermochemistry when comparing
energies only and that transition-dipole moments are double-hybrids with MP2 methods.69 A benchmark
the same as for the underlying TD-DFT calculation. If study on 142 singlet-singlet excitations in small
one neglects the matrix B in Eq. (17), one obtains the organic molecules showed a significantly narrower
Tamm-Dancoff approximation (TDA),173,174 which error distribution for TD-B2-PLYP and TD-B2GP-
has been tested in 2007 for B2-PLYP and very recently PLYP than for TD-B3-LYP and other common
also for the PBE0-DH and PBE0-2 functionals.175 functionals.68 Also singlet-triplet and doublet-doublet
TD-DHDFT calculations can be carried out with excitations were described more accurately than with
any program that has TD-DFT and CIS(D) algorithms conventional TD-DFT methods.169
implemented. Practically, however, TD-DHDFs have Double-hybrids also cure the notorious problem
only been tested with a group-own program in the of spurious intruder or ‘ghost’ states, i.e., artificial
Grimme group and they are currently available for low-lying states due to the SIE. This problem can also
the scientific community in the TDA form in the free be solved at the hybrid level by using functionals with
program ORCA.157 Results for this approach, both in high amounts of Fock-exchange or range-separated
its TD and TDA forms,169 were shown to be more functionals; however, these tend to sometimes
accurate than for standard TD-DFT treatment, which overestimate excitation energies. Double-hybrids
is why we would like to raise awareness of this method also benefit from higher amounts of Fock-exchange,
to encourage program developers to also make this but at the same time the nonlocal CIS(D) contribution
method available in their codes. successfully prevents severe overestimation of the
The success of TD-DHDFs over conventional energies. Therefore, this approach has shown to be
methods has been shown in several studies. In general, particularly useful in theoretical UV/Vis and electronic
circular dichroism (ECD) spectroscopy.169,176,177 For dispersion-corrected by default) is still widely used
example the TD-B2GP-PLYP functional correctly and its performance is competitive to other modern
computed an exciton-coupled ECD spectrum of DFs of similar numerical complexity. The basic reason
a large merocyanine dimer178 for which conven- for this success story is that double-hybrids have a
tional TD-DFT (including range-separated hybrids) very solid theoretical foundation (GLPT) and that
failed.179 they really introduce new physical effects. In the
Closely connected to self-interaction or same way that the exchange energy has a nonlocal
delocalization-error problems is the description component (as first shown by A. Becke), also the
of the two lowest-lying states in polycyclic aromatic correlation energy contains inherently nonlocal parts.
hydrocarbons (La and Lb states) and excitations This is most clearly seen from the importance of
in linear cyanine dyes. These cases are difficult to its long-range (London dispersion) part to binding,
describe with conventional TD-DFT180–182 , whereas i.e., the failure of standard (hybrid-)functionals to
double-hybrids showed superior accuracy.169,183,184 provide any reasonable description of van-der-Waals
With the exception of electronic transitions with complexes. This problem and many short-range cor-
prominent charge-transfer character, double-hybrids relation effects in molecules are improved by inclusion
also provide very good results for large organic of orbital-dependent correlation terms. However, the
dyes.68,69 The structures and averaged results for a drawback of computing it perturbatively as in all
benchmark set containing 12 systems are shown as an double-hybrids is obvious in electronically compli-
example in Figure 5. Results are depicted for BLYP cated situations when this approach can fail (and
and PBE based functionals, including GGA (BLYP and actually does, e.g., in more strongly correlated, metal-
PBE), hybrid (B3-LYP and PBE0), and double-hybrids lic systems, see the DIMPD example discussed above
(B2-PLYP, B2GP-PLYP, PBE0-DH, and PBE0-2) and e.g., Ref 136). In this respect its computation
results.69,175 The PBE0-DH and PBE0-2 numbers are by infinite-order (RPA) methods may be the right
based on TDA-DFT calculations and taken from Ref way to go in the future and these methods eventually
175, while PBE and PBE0 results were recalculated will replace double-hybrid functionals as ‘everyday’
herein based on full TD-DFT (see Supporting Infor- electronic structure methods. However, as B2-PLYP,
mation). All other results are taken from Ref 69. PWPB95 and similar functionals stand they are com-
All results have been obtained with the def2-TZVPP putationally efficient, well-tested, and for many cases
triple-𝜁 basis set.156 As can be seen the Jacob’s Ladder (insulators) of sufficient high accuracy that is not
picture is also reproduced for electronic excitations too far away from the CCSD(T) quantum chemistry
with double-hybrid functionals giving mean devia- ‘gold standard’. As demonstrated by the success
tions (MDs) close to zero and the smallest MADs. of the theory also for excited states, applications
The BLYP based double-hybrids provide more accu- to the computation of molecular properties other
rate results than the PBE-based methods. The most than ground-state energy and structure might yield
accurate double-hybrid is B2GP-PLYP with an MAD improved results compared to those from standard
of 0.16 eV, which can even compete with the ‘gold functionals. Studies in this direction as well as some
standard’ single-reference method for electronic kind of consolidation process in the huge number of
transitions in large systems, the CC2185 method already existing double-hybrids to a few robust and
(MAD = 0.17 eV). For the tested systems, B2GP-PLYP generally recommended variants is desirable.
comes close to the target accuracy of ±0.1 eV desired
for large chromophores with at least 20 nonhydrogen
atoms. NOTE
a
One reviewer of this manuscript pointed out that
CONCLUSIONS Truhlar presented a multi-coefficient method with
MP2 contributions based on Kohn–Sham orbitals in
DHDFs do not only conceptually climb up one conference talks in Atlanta (March 2006) and Tokyo
further step on Jacob’s ladder, but since their first (May 2006), i.e., 2 and 4 months after online pub-
appearance 8 years ago, a large number of bench- lication of the B2-PLYP article. Therein he reported
mark and application studies used various versions that the results of such an approach were similar to
of this approach and they have successfully shown multi-coefficient DFT methods with MP2 contribu-
its higher accuracy compared to hybrid function- tions based on HF orbitals.
als. Even the first B2-PLYP functional (nowadays
ACKNOWLEDGMENTS
L. G. acknowledges funding from the German Academy of Sciences Leopoldina (grant number LPDS 2011-11).
We also thank Dr. M. Steinmetz for conducting some of the calculations.
REFERENCES
1. Kohn W, Sham LJ. Self-consistent equations including 13. Paier J, Janesko BG, Henderson TM, Scuseria GE,
exchange and correlation effects. Phys Rev 1965, Grüneis A, Kresse G. Hybrid functionals includ-
140:A1133–A1138. ing random phase approximation correlation and
2. Peverati R, Truhlar DG. The quest for a universal second-order screened exchange. J Chem Phys 2010,
density functional: the accuracy of density functionals 132:094103.
across a broad spectrum of databases in chemistry and 14. Eshuis H, Yarkoni J, Furche F. Fast computation
physics. Phil Trans R Soc A 2013, 372:20120476. of molecular random phase approximation correla-
3. Perdew JP, Schmidt K. Jacob’s ladder of density function energies using resolution of the identity and
tional approximations for the exchange-correlation imaginary frequency integration. J Chem Phys 2010,
energy. In: Van Doren VE, Van Alsenoy C, Geerlings 132:234114.
P, eds. Density Functional Theory and Its Applications 15. Eshuis H, Bates JE, Furche F. Electron correlation
to Materials, AIP Conference Proceedings, vol. 577. methods based on the random phase approximation.
Melville, NY: American Institute of Physics; 2001, Theor Chem Acc 2012, 131:1084.
1–20. 16. Ángyán JG, Gerber IC, Savin A, Toulouse J. Van
4. Perdew JP, Ruzsinszky A, Tao J, Staroverov VN, der Waals forces in density functional theory: pertur-
Scuseria GE, Csonka G. Prescription for the design and bational long-range electron-interaction corrections.
selection of density functional approximations: more Phys Rev A 2005, 72:012510.
constraint satisfaction with fewer fits. J Chem Phys 17. Bartlett RJ, Grabowski I, Hirata S, Ivanov S. The
2005, 123:062201. exchange-correlation potential in ab initio density
5. Harris J. Adiabatic-connection approach to Kohn- functional theory. J Chem Phys 2005, 122:034104.
Sham theory. Phys Rev A 1984, 29:1648–1659. 18. Schweigert IV, Lotrich VF, Bartlett RJ. Ab initio corre-
6. Görling A, Levy M. Correlation-energy functional lation functionals from second-order perturbation the-
and its high-density limit obtained from a coupling- ory. J Chem Phys 2006, 125:104108.
constant perturbation expansion. Phys Rev B 1993, 19. Janesko BG, Scuseria GE. Coulomb-only second-order
47:13105–13113. perturbation theory in long-range-corrected hybrid
7. Görling A, Levy M. Exact Kohn-Sham scheme density functionals. Phys Chem Chem Phys 2009,
based on perturbation theory. Phys Rev A 1994, 11:9677–9686.
50:196–204. 20. Grimme S. Semiempirical hybrid density functional
8. Bonetti AF, Engel E, Schmid RN, Dreizler RM. with perturbative second-order correlation. J Chem
Investigation of the correlation potential from Phys 2006, 124:034108.
Kohn-Sham perturbation theory. Phys Rev Lett 2001, 21. Møller C, Plesset MS. Note on an approximation
86:2241–2244. treatment for many-electron systems. Phys Rev 1934,
9. Mori-Sanchez P, Wu Q, Yang W. Orbital-dependent 46:618–622.
correlation energy in density-functional theory based 22. Becke AD. Density-functional thermochemistry. III.
on a second-order perturbation approach: success and The role of exact exchange. J Chem Phys 1993,
failure. J Chem Phys 2005, 123:062204. 98:5648–5652.
10. Langreth DC, Perdew JP. Exchange-correlation energy 23. Stephens PJ, Devlin FJ, Chabalowski CF, Frisch MJ.
of a metallic surface: wave-vector analysis. Phys Rev B Ab initio calculation of vibrational absorption and
1977, 15:2884–2901. circular dichroism spectra using density functional
11. Kurth S, Perdew JP. Density-functional correction force fields. J Phys Chem 1994, 98:11623–11627.
of random-phase-approximation correlation with 24. Schwabe T, Grimme S. Towards chemical accuracy for
results for jellium surface energies. Phys Rev B 1999, the thermodynamics of large molecules: new hybrid
59:10461. density functionals including non-local correlation
12. Toulouse J, Gerber IC, Jansen G, Savin A, Ángyán effects. Phys Chem Chem Phys 2006, 8:4398–4401.
JG. Adiabatic-connection fluctuation-dissipation 25. Becke AD. Density-functional exchange-energy
density-functional theory based on range separation. approximation with correct asymptotic behavior.
Phys Rev Lett 2008, 102:096404. Phys Rev A 1988, 38:3098–3100.
26. Lee C, Yang W, Parr RG. Development of the 39. Kozuch S, Gruzman D, Martin JML. DSD-BLYP:
Colle-Salvetti correlation-energy formula into a func- a general purpose double hybrid density functional
tional of the electron density. Phys Rev B 1988, including spin component scaling and dispersion cor-
37:785–789. rection. J Phys Chem C 2010, 114:20801–20808.
27. Miehlich B, Savin A, Stoll H, Preuss H. Results 40. Grimme S. Semiempirical GGA-type density functional
obtained with the correlation energy density function- constructed with a long-range dispersion correction. J
als of Becke and Lee, Yang and Parr. Chem Phys Lett Comput Chem 2006, 27:1787–1799.
1989, 157:200–206. 41. Goerigk L, Grimme S. Efficient and accurate
28. Adamo C, Barone V. Exchange functionals with double-hybrid-meta-GGA density functionals—
improved long-range behavior and adiabatic con- evaluation with the extended GMTKN30 database
nection methods without adjustable parameters: the for general main group thermochemistry, kinetics, and
mPW and MPW1PW models. J Chem Phys 1998, noncovalent interactions. J Chem Theory Comput
108:664–675. 2011, 7:291–309.
29. Tarnopolsky A, Karton A, Sertchook R, Vuzman D, 42. Grimme S, Antony J, Ehrlich S, Krieg H. A consistent
Martin JML. Double-hybrid functionals for thermo- and accurate ab initio parametrization of density
chemical kinetics. J Phys Chem A 2008, 112:3–8. functional dispersion correction (DFT-D) for the 94
elements H-Pu. J Chem Phys 2010, 132:154104.
30. Benighaus T, DiStasio RA Jr, Lochan RC, Chai
JD, Head-Gordon M. Semiempirical double-hybrid 43. Tao J, Perdew JP, Staroverov VN, Scuseria GE.
density functional with improved description of Climbing the density functional ladder: nonempirical
long-range correlation. J Phys Chem A 2008, 112: meta-generalized gradient approximation designed for
2702–2712. molecules and solids. Phys Rev Lett 2003, 91:146401.
31. Jung Y, Lochan RC, Dutoi AD, Head-Gordon M. 44. Perdew JP, Ziesche P, Eschrig H. In: Proceedings of
Scaled opposite-spin second order Møller-Plesset cor- the 21st Annual International Symposium on the Elec-
relation energy: an economical electronic structure tronic Structure of Solids, Akademie Verlag, Berlin,
method. J Chem Phys 2004, 121:9793–9802. 1991, 11.
32. Karton A, Tarnopolsky A, Lamere JF, Schatz GC, 45. Becke AD. Density-functional thermochemistry. IV. A
Martin JML. Highly accurate first-principles bench- new dynamical correlation functional and implica-
mark data sets for the parametrization and valida- tions for exact-exchange mixing. J Chem Phys 1996,
tion of density functional and other approximate 104:1040–1046.
methods. Derivation of a robust, generally appli- 46. Sharkas K, Toulouse J, Savin A. Double-hybrid
cable, double-hybrid functional for thermochemistry density-functional theory made rigorous. J Chem Phys
and thermochemical kinetics. J Phys Chem A 2008, 2011, 134:064113.
112:12868–12886. 47. Bremond E, Adamo C. Seeking for parameter-free
33. Zhang IY, Xu X, WA Goddard III. Doubly hybrid double-hybrid functionals: The PBE0-DH model. J
density functional for accurate descriptions of Chem Phys 2011, 135:024106.
nonbond interactions, thermochemistry, and ther- 48. Chan B, Radom L. Obtaining good performance with
mochemical kinetics. Proc Nat Acad Sci USA 2009, triple- 𝜁 -type basis sets in double-hybrid density func-
106:4963–4968. tional theory procedures. J Chem Theory Comput
34. Graham DC, Menon AS, Goerigk L, Grimme S, 2011, 7:2852–2863.
Radom L. Optimization and basis-set dependence of a 49. Toulouse J, Sharkas K, Bremond E, Adamo C. Ratio-
restricted-open-shell form of B2-PLYP double-hybrid nale for a new class of double-hybrid approxima-
density functional theory. J Phys Chem A 2009, tions in density-functional theory. J Chem Phys 2011,
113:9861–9873. 135:101102.
35. Sancho-García JC, Pérez-Jiménez AJ. Assessment of 50. Kozuch S, Martin JML. DSD-PBEP86: In search of the
double-hybrid energy functionals for 𝜋-conjugated best double-hybrid DFT with spin-component scaled
systems. J Chem Phys 2009, 131:084108. MP2 and dispersion corrections. Phys Chem Chem
36. Chai JD, Head-Gordon M. Long-range corrected Phys 2011, 13:20104–20107.
double-hybrid density functionals. J Chem Phys 2009, 51. Perdew JP, Burke K, Ernzerhof M. Generalized gradi-
131:174105. ent approximation made simple. Phys Rev Lett 1996,
37. Grimme S. Improved second-order Møller-Plesset per- 77:3865–3868.
turbation theory by separate scaling of parallel- and 52. Perdew JP. Density-functional approximation for the
antiparallel-spin pair correlation energies. J Chem correlation energy of the inhomogeneous electron gas.
Phys 2003, 118:9095–9102. Phys Rev B 1986, 33:8822–8824.
38. Chai JD, Head-Gordon M. Systematic optimization 53. Perdew JP. Erratum: Density-functional approxima-
of long-range corrected hybrid density functionals. J tion for the correlation energy of the inhomogeneous
Chem Phys 2008, 128:084106. electron gas. Phys Rev B 1986, 34:7406.
54. Zhang IY, Xu X, Jung Y, Goddard WA III. A fast molecules with double-hybrid density functionals.
doubly hybrid density functional method close to Phys Chem Chem Phys 2009, 11:4611–4620.
chemical accuracy using a local opposite spin ansatz. 69. Goerigk L, Grimme S. Assessment of TD-DFT meth-
Proc Natl Acad Sci USA 2011, 108:19896–19900. ods and of various spin scaled CIS(D) and CC2 ver-
55. Mohajeri A, Alipour M. B2-PPW91: a promising sions for the treatment of low-lying valence excitations
double-hybrid density functional for the electric of large organic dyes. J Chem Phys 2010, 132:184103.
response properties. J Chem Phys 2012, 136:124111. 70. Neese F, Schwabe T, Grimme S. Analytic derivatives
56. Chai JD, Mao SP. Seeking for reliable double-hybrid for perturbatively corrected “double hybrid” density
density functionals without fitting parameters: the functionals: theory, implementation, and applications.
PBE0-2 functional. Chem Phys Lett 2012, 538: J Chem Phys 2007, 126:124115.
121–125. 71. Grimme S, Steinmetz M. Effects of London dispersion
57. Zhang IY, Su NQ, Bremond E, Adamo C, Xu X. correction in density functional theory on the struc-
Doubly hybrid density functional xDH-PBE0 from a tures of organic molecules in the gas phase. Phys Chem
parameter-free global hybrid model PBE0. J Chem Chem Phys 2013, 15:16031–16042.
Phys 2012, 136:174103. 72. Della Sala F, Görling A. Efficient localized
58. Zhang IY, Xu X. Reaching a uniform accuracy Hartree–Fock methods as effective exact-exchange
for complex molecular systems: long-range-corrected Kohn-Sham methods for molecules. J Chem Phys
XYG3 doubly hybrid density functional. J Phys Chem 2001, 115:5718–5732.
Lett 2013, 4:1669–1675. 73. Fromager E, Jensen HJA. Self-consistent many-body
59. Arago J, Orti E, Sancho-Garcia JC. Nonlocal van der perturbation theory in range-separated density-
Waals approach merged with double-hybrid density functional theory: a one-electron reduced-density-
functionals: toward the accurate treatment of non- matrix-based formulation. Phys Rev A 2008, 78:
covalent interactions. J Chem Theory Comput 2013, 022504.
9:3437–3443. 74. Ernzerhof M. Construction of the adiabatic connec-
60. Zhang Y, Yang W. Comment on “Generalized gradient tion. Chem Phys Lett 1996, 263:499–506.
approximation made simple”. Phys Rev Lett 1998, 75. Zhao Y, Lynch BJ, Truhlar DG. Doubly hybrid
80:890–890. meta DFT: new multi-coefficient correlation and
61. Peverati R, Head-Gordon M. Orbital optimized density functional methods for thermochemistry
double-hybrid density functionals. J Chem Phys 2013, and thermochemical kinetics. J Phys Chem A 2004,
139:024110. 108:4786–4791.
62. Lochan RC, Head-Gordon M. Orbital-optimized 76. Zhao Y, Lynch BJ, Truhlar DG. Multi-coefficient
opposite-spin scaled second-order correlation: an extrapolated density functional theory for thermo-
economical method to improve the description of chemistry and thermochemical kinetics. Phys Chem
open-shell molecules. J Chem Phys 2007, 126:164101. Chem Phys 2005, 7:43–52.
63. Neese F, Schwabe T, Kossmann S, Schirmer B, Grimme 77. Raghavachari K, Pople JA. Approximate 4th-order
S. Assessment of orbital-optimized, spin-component perturbation-theory of electron correlation energy. Int
scaled second-order many-body perturbation theory J Quantum Chem 1978, 14:91–100.
for thermochemistry and kinetics. J Chem Theory 78. Pople JA, Head-Gordon M, Raghavachari K.
Comput 2009, 5:3060–3073. Quadratic configuration interaction—a general tech-
64. Kozuch S, Martin JML. Spin-component-scaled dou- nique for determining electron correlation energies. J
ble hybrids: an extensive search for the best fifth-rung Chem Phys 1987, 87:5968–5975.
functionals blending DFT and perturbation theory. J 79. Bartlett RJ, Purvis GD III. Many-body perturbation-
Comput Chem 2013, 34:2327–2344. theory, coupled-pair many-electron theory, and impor-
65. Ernzerhof M, Perdew JP. Generalized gradient approx- tance of quadruple excitations for the correlation
imation to the angle- and system-averaged exchange problem. Int J Quantum Chem 1978, 14:561–581.
hole. J Chem Phys 1998, 109:3313–3320. 80. Curtiss LA, Redfern PC, Raghavachari K. Gn theory.
66. Souvi SMO, Sharkas K, Toulouse J. Double-hybrid WIREs Comput Mol Sci 2011, 1:810–825.
density-functional theory with meta-generalized- 81. Gordon MS, Thruhlar DG. Scaling all correlation
gradient approximations. J Chem Phys 2014, 140: energy in perturbation theory calculations of bond
084107. energies and barrier heights. J Am Chem Soc 1986,
67. Goerigk L, Grimme S. A thorough benchmark of den- 108:5412–5419.
sity functional methods for general main group ther- 82. Zhao Y, Truhlar DG. Assessment of model chemistries
mochemistry, kinetics, and noncovalent interactions. for noncovalent interactions. J Chem Theory Comput
Phys Chem Chem Phys 2011, 13:6670–6688. 2006, 2:1009–1018.
68. Goerigk L, Moellmann J, Grimme S. Computation 83. Zheng J, Zhao Y, Truhlar DG. The DBH24/08
of accurate excitation energies for large organic database and its use to assess electronic structure
model chemistries for chemical reaction barrier an assessment of contemporary theoretical procedures.
heights. J Chem Theory Comput 2009, 5:808–821. J Phys Chem A 2007, 111:13638–13644.
84. Dahlke EE, Orthmeyer MA, Thruhlar DG. Assessment 98. Grimme S, Steinmetz M, Korth M. Stereoelectronic
of multicoefficient correlation methods, second-order substituent effects in saturated main group molecules:
Møller-Plesset perturbation theory, and density severe problems of current Kohn-Sham density func-
functional theory for H3 O+ (H2 O)n (n = 1–5) and tional theory. J Chem Theory Comput 2007, 3:42–45.
OH− (H2 O)n (n = 1–4). J Phys Chem B 2008, 112: 99. Grimme S, Steinmetz M, Korth M. How to compute
2372–2381. isomerization energies of organic molecules with
85. Rey J, Savin A. Virtual space level shifting and correla- quantum chemical methods. J Org Chem 2007,
tion energies. Int J Quantum Chem 1998, 69:581–590. 72:2118–2126.
86. Krieger JB, Chen J, Iafrate GJ, Savin A. Construction 100. Huenerbein R, Schirmer B, Moellmann J, Grimme
of an accurate self-interaction-corrected correlation S. Effects of London dispersion on the isomerization
energy functional based on an electron gas with a gap. reactions of large organic molecules: a density func-
In: Gonis A, Kioussis N, eds. Electron Correlations tional benchmark study. Phys Chem Chem Phys 2010,
and Materials Properties. Plenum: New York; 1999, 12:6940–6948.
463–477.
101. Grimme S. n-Alkane isodesmic reaction energy errors
87. Toulouse J, Savin A, Adamo C. Validation and in density functional theory are due to electron corre-
assessment of an accurate approach to the correlation effects. Org Lett 2010, 12:4670–4673.
lation problem in density functional theory: the
Krieger-Chen-Iafrate-Savin model. J Chem Phys 2002, 102. Schwabe T, Grimme S. Calculation of magnetic cou-
117:10465–10473. plings with double-hybrid density functionals. J Phys
Chem Lett 2010, 1:1201–1204.
88. Tuma C, Sauer J. Treating dispersion effects in
extended systems by hybrid MP2:DFT calculations— 103. Korth M, Grimme S. Mindless DFT benchmarking. J
protonation of isobutene in zeolite ferrierite. Phys Chem Theory Comput 2009, 5:993–1003.
Chem Chem Phys 2006, 8:3955–3965. 104. Goerigk L, Grimme S. A general database for main
89. Becke AD. A new mixing of Hartree–Fock and group thermochemistry, kinetics, and noncovalent
local density-functional theories. J Chem Phys 1993, interactions—assessment of common and reparame-
98:1372–1377. terized (meta-)GGA density functionals. J Chem The-
ory Comput 2010, 6:107–126.
90. Weigend F, Häser M. RI-MP2: first derivatives
and global consistency. Theor Chem Acc 1997, 97: 105. Vaillard SE, Mueck-Lichtenfeld C, Grimme S, Studer
331–340. A. Homolytic substitution at phosphorus for the syn-
thesis of alkyl and aryl phosphanes. Angew Chem Int
91. Curtiss LA, Raghavachari K, Redfern PC, Pople JA.
Ed 2007, 46:6533–6536.
Assessment of Gaussian-2 and density functional the-
ories for the computation of enthalpies of formation. J 106. Greb L, Ona-Burgos P, Schirmer B, Grimme S,
Chem Phys 1997, 106:1063–1079. Stephan DW, Paradies J. Metal-free catalytic olefin
92. Cornaton Y, Franck O, Teale AM, Fromager E. Analy- hydrogenation: low-temperature H2 activation by
sis of double-hybrid density functionals along the adi- frustrated Lewis pairs. Angew Chem Int Ed 2012,
abatic connection. Mol Phys 2013, 111:1275–1294. 51:10164–10168.
93. Web of Science Core Collection, Thomson Reuters. 107. Biczysko M, Panek P, Scalmani G, Bloino J, Barone
Available at: http://apps.webofknowledge.com. V. Harmonic and anharmonic vibrational frequency
(Accessed June 20, 2014). calculations with the double-hybrid B2PLYP method:
analytic second derivatives and benchmark studies. J
94. Curtiss LA, Redfern PC, Raghavachari K. Assessment
Chem Theory Comput 2010, 6:2115–2125.
of Gaussian-3 and density-functional theories on the
G3/05 test set of experimental energies. J Chem Phys 108. Grimme S. Density functional theory with London
2005, 123:124107. dispersion corrections. WIREs Comput Mol Sci 2011,
1:211–228.
95. Grimme S, Mück-Lichtenfeld C, Würthwein EU,
Ehlers AW, Goumans TPM, Lammertsma K. 109. Schwabe T, Grimme S. Double-hybrid density func-
Consistent theoretical description of 1,3-dipolar tionals with long-range dispersion corrections: higher
cycloaddition reactions. J Phys Chem A 2006, accuracy and extended applicability. Phys Chem Chem
110:2583–2586. Phys 2007, 9:3397–3406.
96. Piacenza M, Hyla-Kryspin I, Grimme S. A com- 110. Dunning TH Jr. Gaussian basis sets for use in cor-
parative quantum chemical study of the ruthenium related molecular calculations. I. The atoms boron
catalyzed olefin metathesis. J Comput Chem 2007, through neon and hydrogen. J Chem Phys 1989,
28:2275–2285. 90:1007–1023.
97. Menon AS, Wood GPF, Moran D, Radom L. Bond 111. Kendall RA, Dunning TH Jr, Harrison RJ. Electron
dissociation energies and radical stabilization energies: affinities of the first-row atoms revisited. Systematic
basis sets and wave functions. J Chem Phys 1992, 127. Hujo W, Grimme S. Performance of the van der Waals
96:6796–6806. density functional VV10 and (hybrid)GGA variants
112. Raghavachari K, Trucks GW, Pople JA, Head-Gordon for thermochemistry and noncovalent interactions. J
M. A fifth-order perturbation comparison of elec- Chem Theory Comput 2011, 7:3866–3871.
tron correlation theories. Chem Phys Lett 1989, 157: 128. Mück-Lichtenfeld C, Grimme S. Structure and binding
479–483. energies of the porphine dimer. Mol Phys 2007,
113. Furche F, Ahlrichs R, Hättig C, Klopper W, Sierka M, 105:2793–2798.
Weigend F. Turbomole. WIREs Comput Mol Sci 2014, 129. Schwabe T, Grimme S. The benzenium-ethene
4:91–110. complex: a fundamental problem for standard
114. Grimme S, Ehrlich S, Goerigk L. Effect of the damping second-order Møller-Plesset theory. J Phys Chem A
function in dispersion corrected density functional 2009, 113:3005–3008.
theory. J Comput Chem 2011, 32:1456–1465. 130. Krieg H, Grimme S. Thermochemical benchmark-
115. Grimme S, Huenerbein R, Ehrlich S. On the impor- ing of hydrocarbon bond separation reaction ener-
tance of the dispersion energy for the thermody- gies: Jacob’s ladder is not reversed!. Mol Phys 2010,
namic stability of molecules. Chem Phys Chem 2011, 108:2655–2666.
12:1258–1261. 131. Goerigk L, Kruse H, Grimme S. Benchmarking density
116. Klimes J, Michaelides A. Perspective: advances and functional methods against the S66 and S66x8 datasets
challenges in treating van der Waals dispersion forces for non-covalent interactions. Chem Phys Chem 2011,
in density functional theory. J Chem Phys 2012, 12:3421–3433.
137:120901. 132. Grimme S, Mück-Lichtenfeld C. Accurate compu-
117. Goerigk L. How do DFT-DCP, DFT-NL, and DFT-D3 tation of structures and strain energies of cyclo-
compare for the description of London-dispersion phanes with modern DFT methods. Isr J Chem 2012,
effects in conformers and general thermochemistry? J 52:180–192.
Chem Theory Comput 2014, 10:968–980. 133. Sancho-Garcia JC. Extending the applicability of the
118. Becke AD, Johnson ER. A density-functional model PBE0-DH double-hybrid model to weak interactions.
of the dispersion interaction. J Chem Phys 2005, Chem Phys Lett 2012, 535:136–139.
123:154101. 134. Waller MP, Kruse H, Mück-Lichtenfeld C, Grimme
119. Johnson ER, Becke AD. A post-Hartree–Fock model S. Investigating inclusion complexes using quan-
of intermolecular interactions. J Chem Phys 2005, tum chemical methods. Chem Soc Rev 2012,
123:024101. 41:3119–3128.
120. Johnson ER, Becke AD. A post-Hartree–Fock model of 135. Karton A, Martin JML. Explicitly correlated bench-
intermolecular interactions: inclusion of higher-order mark calculations on C8 H8 isomer energy separations:
corrections. J Chem Phys 2006, 124:174104. how accurate are DFT, double-hybrid, and composite
121. van Gisbergen SJA, Snijders JG, Baerends EJ. A den- ab initio procedures? Mol Phys 2012, 110:2477–2491.
sity functional theory study of frequency-dependent 136. Steinmetz M, Grimme S. Benchmark study of the per-
polarizabilities and van der Waals dispersion coeffi- formance of density functional theory for bond acti-
cients for polyatomic molecules. J Chem Phys 1995, vations with (Ni,Pd)-based transition-metal catalysts.
103:9347–9354. ChemistryOpen 2013, 2:115–124.
122. Osinga VP, van Gisbergen SJA, Snijders JG, Baerends 137. Fogueri UR, Kozuch S, Karton A, Martin JML.
EJ. Density functional results for isotropic and The melatonin conformer space: benchmark and
anisotropic multipole polarizabilities and C6 , C7 and assessment of wave function and DFT methods
C8 van der waals dispersion coefficients for molecules. for a paradigmatic biological and pharmacological
J Chem Phys 1997, 106:5091–5101. molecule. J Phys Chem A 2013, 117:2269–2277.
123. Koide A. A new expansion for dispersion forces and 138. Goerigk L, Karton A, Martin JML, Radom L. Accu-
its application. J Phys B 1976, 9:3173–3183. rate quantum chemical energies for tetrapeptide con-
124. Dion M, Rydberg H, Schröder E, Langreth DC, formations: why MP2 data with an insufficient basis
Lundqvist BI. Van der Waals density functional for set should be handled with caution. Phys Chem Chem
general geometries. Phys Rev Lett 2004, 92:246401. Phys 2013, 15:7028–7031.
125. Vydrov OA, Van Voorhis T. Nonlocal van der Waals 139. Martin JML. What can we learn about dispersion from
density functional: the simpler the better. J Chem Phys the conformer surface of n-pentane? J Phys Chem A
2010, 133:244103. 2013, 117:3118–3132.
126. Murray ED, Lee K, Langreth DC. Investigation of 140. Kozuch S, Bachrach SM, Martin JML. Conforma-
exchange energy density functional accuracy for inter- tional equilibria in butane-1,4-diol: a benchmark
acting molecules. J Chem Theory Comput 2009, of a prototypical system with strong intramolecular
5:2754–2762. H-bonds. J Phys Chem A 2014, 118:293–303.
141. Steinmetz M, Hansen A, Ehrlich S, Risthaus T, 156. Weigend F, Ahlrichs R. Balanced basis sets of split
Grimme S. Accurate thermochemistry for large valence, triple zeta valence and quadruple zeta valence
molecules with modern density functionals. Top Curr quality for H to Rn: design and assessment of accuracy.
Chem (Epub ahead of print; April 24, 2014). doi: Phys Chem Chem Phys 2005, 7:3297–3305.
10.1007/128_2014_543. 157. Neese F. The ORCA program system. WIREs Comput
142. Grimme S, Goerigk L, Fink RF. Spin-component- Mol Sci 2012, 2:73–78.
scaled electron correlation methods. WIREs Comput 158. Riplinger C, Sandhoefer B, Hansen A, Neese F. Nat-
Mol Sci 2012, 2:886–906. ural triple excitations in local coupled cluster calcu-
143. Karton A, Martin JML. Basis set convergence of lations with pair natural orbitals. J Chem Phys 2013,
explicitly correlated double-hybrid density functional 130:134101.
theory calculations. J Chem Phys 2011, 135:144119.
159. Wittman JM, Hayoun R, Kaminsky W, Coggins MK,
144. Chan B, Radom L. Accurate quadruple-𝜁 basis-set Mayer JM. A C-C bonded phenoxyl radical dimer with
approximation for double-hybrid density functional a zero bond dissociation free energy. J Am Chem Soc
theory with an order of magnitude reduction in com- 2013, 135:12956–12959.
putational cost. Theor Chem Acc 2014, 133:1426.
160. Stephan DW, Erker G. Frustrated Lewis pairs:
145. Almlöf J. Elimination of energy denominators in metal-free hydrogen activation and more. Angew
Møller-Plesset perturbation theory by a Laplace trans- Chem Int Ed 2010, 49:46–76.
form approach. Chem Phys Lett 1991, 181:319–320.
161. Stephan DW. “Frustrated Lewis pairs”: a concept for
146. Zhao Y, Truhlar DG. Design of density functionals new reactivity and catalysis. Org Biomol Chem 2008,
that are broadly accurate for thermochemistry, ther- 6:1535–1539.
mochemical kinetics, and nonbonded interactions. J
Phys Chem A 2005, 109:5656–5667. 162. Rokob TA, Hamza A, Papai I. Rationalizing the
reactivity of frustrated Lewis pairs: thermodynamics
147. Sancho-Garcia JC, Adamo C. Double-hybrid den-
of H2 activation and the role of acid-base properties. J
sity functionals: merging wavefunction and density
Am Chem Soc 2009, 131:10701–10710.
approaches to get the best of both worlds. Phys Chem
Chem Phys 2013, 15:14581–14594. 163. Grimme S, Kruse H, Goerigk L, Erker G. The
mechanism of dihydrogen activation by frustrated
148. Ernzerhof M, Scuseria GE. Assessment of the
Lewis pairs revisited. Angew Chem Int Ed 2010,
Perdew-Burke-Ernzerhof exchange-correlation func-
49:1402–1405.
tional. J Chem Phys 1999, 110:5029–5036.
164. Schirmer B, Grimme S. Quantum chemistry of FLPs
149. Adamo C, Barone V. Toward reliable density func-
and their activation of small molecules: methodologi-
tional methods without adjustable parameters: the
cal aspects. Top Curr Chem 2013, 332:213–230.
PBE0 model. J Chem Phys 1999, 110:6158–6170.
150. Su NQ, Adamo C, Xu X. A comparison of geometric 165. Rokob TA, Hamza A, Stirling A, Soós T, Pápai I.
parameters from PBE-based doubly hybrid density Turning frustration into bond activation: a theoretical
functionals PBE0-DH, PBE0-2, and xDH-PBE0. J mechanistic study on heterolytic hydrogen splitting by
Chem Phys 2013, 139:174106. frustrated Lewis pairs. Angew Chem Int Ed 2008,
47:2435–2438.
151. Bousquet D, Bremond E, Sancho-Garcia JC, Ciofini I,
Adamo C. Is there still room for parameter free double 166. Rokob TA, Bako I, Stirling A, Hamza A, Papai I.
hybrids? Performances of PBE0-DH and B2PLYP over Reactivity models of hydrogen activation by frus-
extended benchmark sets. J Chem Theory Comput trated Lewis pairs: synergistic electron transfers or
2013, 9:3444–3452. polarization by electric field? J Am Chem Soc 2013,
135:4425–4437.
152. Yanai T, Tew DP, Handy NC. A new hybrid
exchange-correlation functional using the Coulomb- 167. Pu M, Privalov T. Ab initio dynamics trajectory study
attenuating method (CAM-B3LYP). Chem Phys Lett of the heterolytic cleavage of H2 by a Lewis acid
2004, 393:51–57. B(C6 F5 )3 and a Lewis base P(tBu)3 . J Chem Phys 2013,
138:154305.
153. Vydrov OA, Scuseria GE. Assessment of a long-range
corrected hybrid functional. J Chem Phys 2006, 168. Welch GC, Juan RRS, Masuda JD, Stephan DW.
125:234109. Reversible, metal-free hydrogen activation. Science
154. Cornaton Y, Stoyanova A, Jensen HJA, Fromager E. 2006, 314:1124–1126.
Alternative separation of exchange and correlation 169. Grimme S, Neese F. Double-hybrid density functional
energies in range-separated density-functional pertur- theory for excited electronic states of molecules. J
bation theory. Phys Rev A 2013, 88:022516. Chem Phys 2007, 127:154116.
155. Cornaton Y, Fromager E. Double hybrid density- 170. Casida ME. Time-dependent density functional
functional theory using the Coulomb-attenuating response theory for molecules. In: Chong DP, ed.
method. Int J Quantum Chem (Epub ahead of print; Recent Advances in Density Functional Methods, vol.
April 17, 2014). doi: 10.1002/qua.24682. 1. Singapore: World Scientific; 1995, 155.
171. Bauernschmitt R, Ahlrichs R. Treatment of electronic supramolecular systems. In: Berova N, Polavarapu
excitations within the adiabatic approximation of time PL, Nakanishi K, Woody RW, eds. Comprehensive
dependent density functional theory. Chem Phys Lett Chiroptical Spectroscopy, Vol 1: Instrumentation,
1996, 256:454–464. Methodologies and Theoretical Simulations. Hobo-
172. Head-Gordon M, Rico RJ, Oumi M, Lee TJ. A doubles ken, NJ: Wiley; 2012, 643–673.
correction to electronic excited states from configu- 179. Goerigk L, Grimme S. Quantum chemical investiga-
ration interaction in the space of single substitutions. tion of exciton coupling: super-molecular calculations
Chem Phys Lett 1994, 219:21–29. of a merocyanine dimer aggregate. Chem Phys Chem
173. Fetter AL, Walecka JD. Quantum Theory of 2008, 9:2467–2470.
Many-Particle Systems. New York: McGraw-Hill; 180. Grimme S, Parac M. Substantial errors from
1971. time-dependent density functional theory for the
174. Hirata S, Head-Gordon M. Time-dependent density calculation of excited states of large 𝜋-systems. Chem
functional theory within the Tamm-Dancoff approx- Phys Chem 2003, 4:292–295.
imation. Chem Phys Lett 1999, 314:291–299. 181. Parac M, Grimme S. A TDDFT study of the lowest
175. Di Meo F, Troiullasa P, Adamo C, Sancho-García excitation energies of polycyclic aromatic hydrocar-
JC. Application of recent double-hybrid density func- bons. Chem Phys 2003, 292:11–21.
tionals to low-lying singlet-singlet excitation energies 182. Schreiber M, Buß V, Fülscher MP. The electronic
of large organic compounds. J Chem Phys 2013, spectra of symmetric cyanine dyes: a CASPT2 study.
139:164104. Phys Chem Chem Phys 2001, 3:3906–3912.
176. Goerigk L, Grimme S. Calculation of electronic 183. Goerigk L, Grimme S. Double-hybrid density func-
circular dichroism spectra with time-dependent tionals provide a balanced description of excited 1 La
double-hybrid density functional theory. J Phys Chem and 1 Lb states in polycyclic aromatic hydrocarbons. J
A 2009, 113:767–776. Chem Theory Comput 2011, 7:3272–3277.
177. Vintonyak VV, Warburg K, Kruse H, Grimme S, 184. Send R, Valsson O, Filippi C. Electronic excitations of
Huebel K, Rauh D, Waldmann H. Identification simple cyanine dyes: reconciling density functional and
of thiazolidinones spiro-fused to indolin-2-ones as wave function methods. J Chem Theory Comput 2011,
potent and selective inhibitors of the mycobacterium 7:444–455.
tuberculosis protein tyrosine phosphatase B. Angew 185. Christiansen O, Koch H, Jørgensen P. The
Chem Int Ed 2010, 49:5902–5905. second-order approximate coupled cluster singles
178. Goerigk L, Kruse H, Grimme S. Theoretical electronic and doubles model CC2. Chem Phys Lett 1995,
circular dichroism, spectroscopy of large organic and 243:409–418.

WIREs Comput Mol Sci 2014 Goerigk Doublehybrid Density Functionals

Uploaded by

Copyright:

Available Formats

WIREs Comput Mol Sci 2014 Goerigk Doublehybrid Density Functionals

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

WIREs Comput Mol Sci 2014 Goerigk Doublehybrid Density Functionals

Uploaded by

Copyright:

Available Formats

Advanced Review

Double-hybrid density functionals

Double-hybrid density functionals (DHDFs) are reviewed in this study. In DHDFs

How to cite this article:

INTRODUCTION follows on a ladder, with each higher rung offering a

(4) Hybrid DFs are formally derived from the Double−hybrid

Functional Year Description DFTX DFTC

METHODS PRECEDING THE FIRST ̂𝜆

Görling–Levy Perturbation Theory H ̂ +V

correlation contributions. The mPW2-PLYP method 0.4

improve the description of noncovalent interactions 0.2

developed for the calculation of response properties.

Becke-Johnson variant were determined later.67

TABLE 5 Weighted Total Mean Absolute Deviations (WTMADs) in

(c) C6F5 t-Bu C6F5 t-Bu

main-group thermochemical problems, we attribute In their approach excitation energies are

You might also like