Advanced Review

Density functional theory with

London dispersion corrections
Stefan Grimme∗

Dispersion corrections to standard Kohn–Sham density functional theory (DFT)

are reviewed. The focus is on computationally efficient methods for large sys-
tems that do not depend on virtual orbitals or rely on separated fragments. The
recommended approaches (van der Waals density functional and DFT-D) are
asymptotically correct and can be used in combination with standard or slightly
modified (short-range) exchange–correlation functionals. The importance of the
dispersion energy in intramolecular cases (conformational problems and thermo-
chemistry) is highlighted. C 2011 John Wiley & Sons, Ltd. WIREs Comput Mol Sci 2011 1 211–228
DOI: 10.1002/wcms.30

INTRODUCTION inary development stage, or currently not applicable

to large systems.
T he development of approximate density func-
tional theory (DFT) approaches that accurately
model the physically and chemically very important
Because all current DFT approaches that accu-
rately account for dispersion include empirical ele-
London dispersion interactions1, 2 is a very active field ments in various ways, solid benchmarking on reliable
of research (for recent papers with some review char- (nonempirical) reference data [mostly of coupled
acter, see Refs 3–6). Dispersion interactions can be cluster singles doubles with perturbative triples
empirically defined as the attractive part of the van (CCSD(T)) quality] is mandatory. We will use in-
der Waals (vdW)-type interaction potential between termolecular interactions of small-to-medium-sized
atoms and molecules that are not directly bonded to complexes, intramolecular dispersion effects in con-
each other, although the terms ‘dispersion’ and ‘vdW’ formational problems, and their influence on chemical
are often used synonymously. It has now become very reaction energies as examples. An extension of our so-
clear especially for the chemistry and physics of large called GMTKN24 database,7 which is a collection of
systems, e.g., in bio- or nanoarchitectures, that inclu- 24 previously published or newly developed bench-
sion of these interactions in theoretical simulations mark sets for general main group thermochemistry,
is indispensable in order to reach so-called chemical kinetics, and noncovalent interactions, is used for this
accuracy. Because the discussed effects are due to om- purpose.
nipresent electron correlations, they also influence the
accuracy of theoretical (reaction) thermodynamics.
The scope of this work is to review the existing, THEORY
well-established methods for including dispersion in- The Failure of Standard Functionals
teractions into DFT. We concentrate on approaches For more than a decade, it is known that commonly
that work with standard or slightly modified density used DFs do not describe the long-range dispersion in-
functionals (DFs). Virtual orbital-dependent (i.e., ran- teractions correctly.8–11 Originally, this was noted for
dom phase approximation (RPA)); or fragment-based rare-gas dimers (rediscovered, e.g., in Ref 12) but later
methods (e.g., DFT–symmetry adapted perturbation noticed also in base pair stacking13 or N2 dimers.14
theory (SAPT)); are excluded because either they are Some confusion in these early days arose because the
not completely generalized, in some kind of prelim- problem is highly functional dependent. If one mainly
considers equilibrium distances for common weakly
bound complexes, some DFs such as PW9115 provide
∗
Correspondence to: [email protected] at least qualitatively correct interaction potentials,
Theoretische Organische Chemie, Organisch-Chemisches Institut whereas other DFs, e.g., the popular BLYP16, 17 or
der Universität Münster, Münster, Germany B3LYP18, 19 approximations were found to be purely
DOI: 10.1002/wcms.30 repulsive. Nowadays, it is clear that all semilocal

Volume 1, March/April 2011 

c 2011 John Wiley & Sons, Ltd. 211
Advanced Review wires.wiley.com/wcms

0.1 Reference 0.4

B3LYP
PBE 0.2

0 0

Reference
E (kcal/mol)

E (kcal/mol)
B3LYP
PBE

3.5 4 4.5 5 5.5 6 6.5 7 7.5 8 8.5 9 9.5 10 3.5 4 4.5 5 5.5 6 6.5 7 7.5 8
R (Å) R (Å)

F I G U R E 1 | Potential energy curves for the Kr2 (left) and the benzene dimer (right, D 6h symmetry) with two different density functional
approximations in comparison with accurate CCSD(T) reference data.20, 21

DFs and conventional hybrid functionals (that include and j → b (on B), (ia| jb) is a two-electron integral,
nonlocal Fock exchange) asymptotically cannot pro- and  are the corresponding orbital energies. This
vide the correct −C6 /R6 dependence of the disper- is also illustrated schematically in Figure 2. Note that
sion interaction energy on the interatomic (molecular) A and B do not necessarily have to be clearly separated
distance R. This is different for intermediate distances atoms or molecules but can also be molecular frag-
where the fragment electron densities overlap and ments leading to intramolecular dispersion effects.
semilocal DFs may yield bound states. The induced dipole moments on one of the frag-
In Figure 1, two typical examples are shown ments have its origin in ‘charge fluctuations’ on the
that illustrate the problem. For both dimers, B3LYP other, but this process should be better viewed as
is overrepulsive and yields no binding at all. For the instantaneous electron correlations. In a more pre-
Perdew-Burke-Ernzerhoff (PBE)22 functional, a mini- cise picture, electromagnetic zero-point energy fluc-
mum is found for Kr2 that occurs, however, at a too tuations in the vacuum lead to ‘virtual’ excitations
long distance and with an unrealistically small inter- to allowed atomic or molecular electronic states. The
action energy. For the benzene dimer, the PBE func- corresponding (pseudo)densities interact electrostat-
tional also yields no minimum. Note that in the case of ically (with exchange-type modifications at smaller
Kr2 , which is very strongly dominated by dispersion distances). They are not represented by conventional
interactions, for both DFs, the interaction potential is (hybrid) functionals that only consider electron ex-
exponentially decaying and almost zero for R > 6 Å, change but do not employ virtual orbitals (i.e., use
whereas the decay of the −C6 /R6 term is rather slow, electronic charge but no transition density). In pass-
so that the reference potential is significantly bound ing, it is noted that dispersion is ‘transmitted’ by
up to about 8 Å. electromagnetic radiation, so that screening effects in
The failure of standard DFs is easily under- dense materials (as described by an effective dielectric
stood by considering the ‘true’, wave function (WF)- constant) appear.
based origin of the dispersion energy. For example, in Most of the current dispersion-corrected DFT
second-order Møller–Plesset perturbation theory,23 it approaches (and especially the accurate ones) in-
is given by the Coulomb and exchange interactions of clude in various ways empirical components. The
single-electron transition densities centered on inter- basic reason for this is rooted in the fact that dis-
acting fragments A and B persion is a special kind of electron correlation
problem operating merely on long-range scales (see
(2)
  (ia| jb)[(ia| jb) − (ib| ja)] Figure 3). At short electron–electron distances, the
Edisp = − (1) standard functionals describe the corresponding ef-
ia jb
a + b − i −  j
fects rather well because of their deep relation to the
where the sum is over all possible single-particle hole corresponding electron density changes. Thus, any
excitations between orbitals i → a (localized on A) dispersion including DFT approach is faced with the

212 
c 2011 John Wiley & Sons, Ltd. Volume 1, March/April 2011
 WIREs Computational Molecular Science Density functional theory with London dispersion corrections

Single excitations = electron fluctuations

}
Transition density

(Transition) dipole
moment

Coloumb and
exchange
interaction
A B
F I G U R E 2 | Schematic description of the dispersion interaction for two interacting fragments A and B (e.g., helium atoms) at long distance.

(Hybrid) functional (density, occ. orbitals) Dispersion

corrections
Problematic region

overlap dispersive) Density based C6 based Effective one−electron

potential
V = V KS + V NL V = V KS E = E KS + Epair V = V KS + V 1e
Semiclassical (1/R6) vdW−DF Parameterized DCACP
DFT−D
asymptotic region DF LAP/DCP

0 F I G U R E 4 | Overview of currently used dispersion corrections in

Short range Long range
density functional theory (DFT). E KS and VKS correspond to the bare
Electron correlation length Kohn–Sham total energies and potentials, respectively.

F I G U R E 3 | Schematic classification of the correlation and

dispersion problems on different electron correlation length scales. (1ePOT, called DCACP31 or in local variants LAP32
or DCP5 ). In the following subsections, these methods
are briefly described and the relevant key references
problem to merge in a seamless fashion the short- and are given. Their pros and cons are summarized in
long-range asymptotic regions that are fairly well un- Table 1, which also provide some guidelines for prac-
derstood separately. The electron correlations in this tical purpose. The relevant properties that are men-
problematic region are difficult to classify but they tioned in Table 1 are discussed in the corresponding
often have the typical WF signatures of dispersion- method sections.
type interactions and are nowadays usually termed
‘medium-range’ correlation.24 They are of particular
importance for the equilibrium structures of many The vdW-DF and Related Methods
vdW complexes and also the thermodynamic proper- The vdW-DF method in its currently most widely used
ties of larger molecules. form25 is a nonempirical way to compute the disper-
sion energy for arbitrary systems, solely based on their
electron density. As also in the other approaches, a
Overview and Classification of Methods supermolecular calculation of the total energy of the
The various approaches currently in use are grouped complex and the fragments is performed to obtain the
(into four classes) in Figure 4 and characterized in interaction energy. For the relevant total exchange–
Table 1. This overview includes nonlocal vdW- correlation energy Exc of a system, the following ap-
DFs,25, 26 ‘pure’ [semilocal (hybrid)] DFs, which proximation is employed in all vdW-DF schemes
are highly parameterized forms of standard meta-
hybrid approximations (e.g., the M0XX family Exc = ExLDA/GGA + EcLDA/GGA + EcNL (2)
of functionals27 ), DFT-D methods (atom pairwise
sum over −C6 R−6 potentials28–30 ), and dispersion- where standard exchange and correlation compo-
correcting atom-centered one-electron potentials nents of local density approximation (LDA) or

Volume 1, March/April 2011 

c 2011 John Wiley & Sons, Ltd. 213
 Advanced Review wires.wiley.com/wcms

T A B L E 1 Properties of Current Methods (grouped in four classes) to Account for London Dispersion
Interactions in DFT Calculations

vdW-DF DFa DFT-D 1ePOT

Property I II III IV

Correct R −6 Yes No Yes No

Good thermochemistry ? Yes Yes ?
Numerical complexity High Medium Low Low
Simple forces No Yes Yes Yes
System dependency Yes Yes Yesb Noc
Electronic effectd Yes Yes No Yes
Empiricism Low Medium Medium High
Analysis/insight ? No Good ?

a
Specially developed semilocal (hybrid) functionals that recover medium-range correlation effects.
b
This only holds for the most recent DFT-D variants (e.g., Refs 6, 28, 33, 34). Older DFT-D versions with fixed atomic C6 coefficients
(e.g., Refs 30, 35) do not include any specific system dependency.
c
The potential parameters are determined only once for one element and are not specifically computed for each system. For DCACP (but
not for LAP or DCP), due to the nonlocality of the potential, a small system dependency is included.
d
Implies that the dispersion correction affects the WF or density of the system.

semilocal (GGA) type are used for the short-ranged where ρ is the electron density and r and r  denote
parts and EcNL represents the nonlocal term describ- electron coordinates. The different variants of vdW-
ing the dispersion energy. This term is in the mod- DF that are currently on the market only differ in
ern versions undamped and also contributes at short the choice of the nonlocal correlation kernel φ(r, r  ).
electron–electron distances to the correlation energy. These kernels are physically based on local approxi-
Hence, corrections of this type will also affect ‘nor- mations to the (averaged) dipole polarizability at fre-
mal’ (covalent) thermochemistry. Earlier versions of quency ω [i.e., α(r, ω)], which when integrated, yields
vdW-DF (dubbed ALL96; Ref 36) are derived for the total polarizability α,
nonoverlapping densities and require empirical damp- 
ing functions to ensure finite values for short inter- α(ω) = α(r, ω)dr. (4)
fragment distances. A new version termed vdW-DF2
(Ref 26) goes beyond local approximations in the Knowing the polarizability at all (imaginary) fre-
kernel used to compute EcNL , which gives improved quencies leads automatically via the famous Casimir–
results and in particular a more consistent descrip- Polder relationship42 to the long-range part of the
tion of interaction energies and complex geometries dispersion energy. The C6 dispersion coefficient for
(intermolecular distances37 ). Typically, EcNL is com- interacting fragments A and B is given by
puted non-self-consistently, i.e., it is simply an add- 
3 ∞
on to the self-consistent field (SCF)-DF energy simi- C6AB = α(iω)A α(iω)B dω. (5)
π 0
lar as in DFT-D (see below). Although this has often
only minor effects on computed energies, it hampers This integral also forms the basis of modern DFT-
the computation of forces (for geometry optimiza- D approaches to the dispersion problem (see below).
tion) such that efficient SCF implementations for EcNL Using a dielectric function (e.g., the Drude model for
have been developed.38 It is also found39 that vdW- a metal)
GGA
DF works better with short-range Exc components ωp
that are basically repulsive such as Hartree–Fock (e.g., (ω) = 1 − (6)
ω
revPBE40 is preferred over PBE). Empirical optimiza-
tion of the short-range GGA parts further improves the local polarizability α(r ) can be related via a local
the results.39, 41 plasma frequency ω p (r ) to the density, i.e,

In the general vdW-DF framework, the nonlocal ω p (r ) = ρ(r ). (7)
correlation (dispersion) energy takes the form of a
double-space integral For systems with a finite highest occupied molecu-
lar orbital–lowest unoccupied molecular orbital gap,
 
1 modified dielectric functions that include approx-
EcNL = ρ(r )φ(r, r  )ρ(r  )dr dr  (3) imations to local band gaps are used.43, 44 The
2

214 
c 2011 John Wiley & Sons, Ltd. Volume 1, March/April 2011
 WIREs Computational Molecular Science Density functional theory with London dispersion corrections

different vdW-DF versions currently in use are vdW- Conventional and Parameterized
DF (2004),25 vdW-DF (2010)26 (also called vdW- Functionals (DFs)
DF2), VV09,43 and VV10.45 The older damped ver- If no special corrections for dispersion effects are in-
sion is also still in use, e.g., for a recent application of cluded, all current Kohn–Sham density functional ap-
ALL96 to the benzene dimer potential energy surface proximations that are based solely on the electron
(PES), see Ref 4. density and occupied orbitals do not accurately ac-
In the context of vdW-DF, the terms ‘local’ and count for the long-range interactions in the weakly
‘nonlocal’ are often used in the literature, and in the overlapping regime. However, as long as only equi-
DFT community, the dispersion energy is understood librium structures of not too large molecules are con-
as an inherently nonlocal property. Hence, it must be sidered, nondispersion-corrected DF can provide in
described by a kernel φ(r, r  ) in Eq. (3), which depends some cases fairly accurate results. Some of such stud-
on two electron coordinates simultaneously. How- ies are mentioned below.
ever, this statement makes only sense in a DFT lan- The hybrid density functional Becke half-and-
guage wherein ρ(r ) and derived exchange–correlation half reproduces binding energies and potential en-
potentials vxc (r ) are the central (local) quantities. In ergy surfaces for π -stacked geometries of substituted
a WF picture, long-range dispersion has no nonlocal benzenes and pyridines, as well as pyrimidine and
(in the sense Fock exchange is nonlocal) component DNA bases46 within ±0.5 kcal/mol of MP2 and/or
(e.g., the MP2 energy then only contains Coulomb in- CCSD(T) reference data. This result is presumably
tegrals). This is confusing for readers that for example due to error cancellation, as the authors state, and hy-
know Stone’s book1 in which the term ‘nonlocal’ does drogen bonding interaction energies are significantly
not occur anywhere in the chapter on the dispersion overestimated with this functional. For vdW com-
energy. Physically, dispersion is the Coulomb inter- plexes of fluorine-containing organic molecules, only
action between (local, fragment centered) transition the PBE density functional yields some binding, how-
densities. These can be plotted like ‘normal’ densi- ever, too weak as compared with quite accurate MP2
ties ρ(r ) and have no ‘mysterious’ nonlocal charac- results.47 The X3LYP functional improves the accu-
ter (except that virtual orbitals are needed for their racy of hybrid GGA methods for rare-gas dimers48
construction). The Coulomb interactions are modi- and the water dimer49 significantly, but fails qualita-
fied only at short-range by ‘true’ exchange [second tively for stacking,50 which contradicts the original
term in the enumerator in Eq. (1]. This is also com- claims of its inventors that it is well suited for non-
pletely analogous to the classical electronic Coulomb bonded interactions. The Wilson–Levy correlation
energy Vee (which is, inconsistently, also in DFT not functional together with Hartree–Fock exchange re-
called ‘nonlocal’). This WF-based picture is also in produces binding trends for selected rare-gas dimers,
line with our basic understanding of long-range dis- isomers of the methane dimer, benzene dimer, naph-
persion. Many simple schemes with R−6 -dependent thalene dimer, and stacked base-pair structures.51
terms and employing atom- and group-additive dis- The Tao-Perdew-Staroverov-Scuseria (TPSS)
persion coefficients (see DFT-D section) only work ac- and TPSS hybrid (TPSSh) functionals52 produce
curately because dispersion is basically a local, atom- vdW bonds in 10 rare-gas dimers with Z ≤ 36
like property. and correct the overbinding of the local spin den-
The biggest advantage of vdW-DF methods over sity approximation.53 Binding energy curves for the
all other approaches mentioned here is that disper- ground-state rare-gas diatomics Ne2 and Ar2 and for
sion effects are naturally included via the charge den- the alkaline–earth diatomic Be2 in reasonable agree-
sity so that charge-transfer (atomic oxidation state) ment with those from experiment are found for PBE
dependence of dispersion is automatically included and TPSS, but they have an incorrect asymptotic be-
in a physically sound manner. If performed self- havior for large internuclear separation.54 Adamo and
consistently, the correction in turn also changes the Barone55 could achieve an improved description of
density. Both effects are not fully accounted for in He2 and Ne2 interaction potentials with still accurate
class III and IV methods (see Table 1). What is cur- results for other properties by refitting the Perdew–
rently not known is whether double-counting effects Wang exchange functional and using it in a hybrid
of correlation at short range are present in the men- model called mPW1PW.
tioned vdW-DFs. These functionals have not been Some highly parameterized meta-GGAs incor-
tested so far on intramolecular or thermochemical porating kinetic energy density have been assessed
problems (see sections on thermochemistry and con- to quantitatively account for dispersion effects. Zhao
formations). and Truhlar56 describe a test of 18 DFs for the

Volume 1, March/April 2011 

c 2011 John Wiley & Sons, Ltd. 215
Advanced Review wires.wiley.com/wcms

calculation of bond lengths and binding energies of the 1970s in the context of Hartree–Fock theory65, 66
rare-gas dimers, alkaline–earth metal dimers, zinc (for more recent Hartree–Fock dispersion models, see
dimer, and zinc–rare-gas dimers. The authors con- Refs 67, 68). The method has been forgotten for al-
clude from the combined mean percentage unsigned most 30 years and was rediscovered about 10 years
error in geometries and energies that M05-2X57 and ago69, 70 as the DFT problems became more evident. It
MPWB1K58 are the overall best methods for the pre- is usually termed DFT-D (or sometimes DFT + disp).
diction of vdW interactions in the 17 metal and rare- Although this name is quite unspecific, it has been
gas vdW dimers studied. A set of 13 complexes with accepted meanwhile generally and will also be used
biological relevance is considered in a study of newly here. Probably, the first published paper in which a
developed DFT methods that give reasonable results standard DFT calculation has been combined with a
for the stacked arrangements in the DNA base pairs damped dispersion energy is the work of Gianturco
and amino acid pairs, in contrast to previous DFT et al.71 (for the special case of Ar–CO). An even earlier
methods, which fail to describe interactions in stacked approach along the same lines, however, not employ-
complexes.59 In Ref 60, multicoefficient extrapolated ing a standard DF but LDA-based expressions for
DFT methods are used to calculate the interaction the repulsive short-range part (the Gordon–Kim–Rae
energy of benzene dimers. The M06-2X functional27 model) combined with a damped interatomic pairwise
is probably the most accurate dispersion-uncorrected treatment can be found in the paper of Cohen and
functional that gives good results for the S22 set Pack72 , which likely represents the first ‘true’ DFT-D
as well as stacked aromatic structures.61 One seri- work.
ous problem of this and related highly parameterized Meanwhile, many modifications of the DFT-D
functionals with many terms in a power series expan- approach have been published. All are based on an
sion is numerical instability that can lead to artificial atom pairwise additive treatment of the dispersion
vdW minima and ‘noisy’ potential energy curves.62 energy (for extensions to include three-body nonad-
Recently, it has been reported independently by ditive dispersion effects, see Refs 28, 73). The general
two groups39, 63 that DFs that more closely resem- form for the dispersion energy (which is simply added
ble the repulsive behavior of Hartree–Fock (‘disper- to the Kohn–Sham DFT energy) is
sionless’ DF) perform better in dispersion-corrected
  CnAB
DFT treatments than over-repulsive (B88 Ref 16) DFT−D
Edisp =− sn n fdamp (RAB ). (8)
or overattractive (PBE Ref 22) constructions. This RAB
AB n=6,8,10,...
idea seems physically sound and the two approaches
termed Dl09 (Ref 63; based on the M06 family where, the sum is over all atom pairs in the system,
of functionals) and a reparameterization of PW86 CnAB denotes the averaged (isotropic) nth-order dis-
exchange39 are currently further tested for noncova- persion coefficient (orders n = 6, 8, 10, . . .) for atom
lent interactions.64 pair AB, and RAB is their internuclear distance. Global
In summary, one can state that modern (highly (DF- dependent) scaling factors sn are typically used
parameterized) DFs can ‘emulate’ medium-range, to adjust the correction to the repulsive behavior of
nonlocal dispersion effects rather accurately by the chosen DF.29 If this is done only for n > 6 (as
semilocal exchange–correlation potentials. They are in DFT-D3, Ref 28), asymptotic exactness is fulfilled
purely density based and thus share some good prop- when the C6AB are exact. Note that the contribution
erties with the vdW-DF (class I). However, asymp- of the higher-ranked multipole terms n > 6 is more
totically, the dispersion energy in these approaches short ranged and rather strongly interferes with the
is zero and, therefore, they cannot be recommended (short-ranged) DF description of electron correlation.
for extended systems (e.g., solids or biomolecules) The higher Cn terms can be used to adapt the potential
in which long-range (asymptotic) contributions are specifically to the chosen DF in this mid-range region.
important. The question how many higher-order terms are nec-
essary is not completely clear at present. Although C8
and C10 contribute significantly in equilibrium regions
Semiclassical Corrections (DFT-D) (roughly 50% of Edisp for heavier atoms), owing to
The idea to treat the (quantum mechanically) dif- the huge values of these coefficients, their correspond-
ficult dispersion interactions semiclassically and to ing errors are grossly amplified and make the correc-
combine the resulting potential with a quantum tion somewhat unstable.28 Some kind of consensus
chemical approach (a kind of quantum mechanical– has been reached in that C6 alone is not sufficient to
molecular mechanical hybrid scheme) goes back to describe medium/short-range dispersion.28, 64, 74, 75

216 
c 2011 John Wiley & Sons, Ltd. Volume 1, March/April 2011
 WIREs Computational Molecular Science Density functional theory with London dispersion corrections

In order to avoid near singularities for small R 0

and double-counting effects of correlation at inter-
mediate distances, damping functions fdamp are used,
which determine the range of the dispersion correc-
tion. For a discussion of general damping functions,
see Ref 76. If only noncovalent interactions are con-
sidered, the results are only weakly dependent on the

Edisp(kcal/mol)
specific choice of the function. Typical expressions are
given in Ref 77
Total, TPSS-D3
1 Total, Becke–Johnson
fdamp (RAB ) = , (9)
1 + 6(RAB /(sr,n R0AB ))−γ C6R

or Ref 35
1
fdamp (RAB ) = , (10)
e−γ (RAB /sr,n R0 −1)
AB
1+
1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6 6.5 7 7.5 8
where R0AB is a cutoff radius for atom pair AB, sr,n is a R(C-—C) [Å]
DF-dependent (global) scaling factor (as introduced in
Ref 30), and γ is a global constant that determines the F I G U R E 5 | Dispersion correction for two carbon atoms (dispersion
steepness of the functions for small R. For the cutoff coefficients from Ref 28) with the zero- and finite-damping
radii, often (averaged) empirical atomic vdW radii are (Becke–Johnson) methods in comparison with the undamped R −6
term.
used. The most sophisticated ab initio approach to
determine pair-specific values is described in Ref 28.
Currently, the most widely used DFT-D method
A fundamental difference between the ex-
is our version from 2006 (Ref 35; now termed DFT-
isting damping approaches is their behavior for
D2), which represents an update of DFT-D1 from
small R. Although in most methods (e.g., for the
2004.29 The method has recently been refined regard-
above given formulas) the damping function (and
ing higher accuracy, broader range of applicability,
thus Edisp ) approaches zero for R → 0, Becke and
and less empiricism (called DFT-D328 ). The main new
coworkers64, 78–80 use rational damping in the form
ingredients are atom pairwise-specific dispersion co-
 C6 efficients and a new set of cutoff radii, both computed
Edisp = − . (11)
6
RAB + const. from first principles. The coefficients for eighth-order
AB
dispersion terms are computed using established re-
This leads to a constant contribution of Edisp to the cursion relations.82, 83 System (geometry)-dependent
total correlation energy from each bonded atom pair. information is used for the first time in a DFT-D-
Although this seems theoretically justified,81 it basi- type approach by employing the new concept of frac-
cally requires special adjustment of the standard cor- tional coordination numbers. This allows one to dis-
relation DF used, whereas the ‘zero-damping’ method tinguish the different hybridization states of atoms in
works very well with standard functionals and also molecules in a differentiable way, which in particular
for thermochemistry.28, 35 The differences between for the first two rows of the periodic table have quite
the two philosophies are illustrated schematically different dispersion coefficients. The method only re-
in Figure 5. Note that a disadvantage of the zero- quires adjustment of two global parameters for each
damping approach is that at small and medium dis- density functional is asymptotically same for a gas
tances (‘left’ of the minimum in Figure 5), the atoms of weakly interacting neutral atoms and easily allows
experience a repulsive force that may lead in some the computation of atomic forces. Accurate disper-
(rare and special) situations to longer interatomic dis- sion coefficients and cutoff radii are available for all
tances with dispersion correction than without disper- elements up to Z = 94. The revised DFT-D3 method
sion correction. Note that the minimum in the TPSS- can be used as a general tool for the computation
D3 curve in Figure 5 has the physical meaning that of the dispersion energy in molecules and solids (see,
it marks the onset of the DF description of electron e.g., also Refs 84, 85) of any kind with DFT and
correlation. Nowadays, distance values smaller than related (low-cost) electronic structure methods for
this (atom pair-dependent) minimum are considered large systems. Figure 6 displays DFT-D3-computed
as belonging to the ‘short- to-medium-range correla- molecular dispersion coefficients in comparison with
tion region’. experimental values. The mean average deviation is

Volume 1, March/April 2011 

c 2011 John Wiley & Sons, Ltd. 217
 Advanced Review wires.wiley.com/wcms

only about 5%, which is the limiting accuracy of any

DFT-D3, MAD = 4.74% ‘-D3’ method asymptotically. DFT-D3 is currently the
best ‘simple’ way to compute such data for arbitrary
systems.
1000
There are three big advantages of DFT-D3 com-
pared with other methods. First, it can easily be cou-
C6 (calc.) (a.u.)

pled with any standard DF (45+ have been param-

eterized, see Ref 86) without any significant loss of
accuracy (see Table 2 below). Second, it also easily
100 allows the calculation of energy gradients for efficient
geometry optimization, which is one of the main pur-
poses of the method. The third point concerns the
analysis of the results. In DFT-D, the total dispersion
energy can easily be attributed to contributions from
individual atom pairs, parts of a molecule, or for a
10
particular distance range (see, e.g., the example given
in Ref 87).
10 100 1000
C6 (exp.) (a.u.)
Related to DFT-D3 are approaches that employ
Eq. (8) and also compute the C6 coefficients specifi-
F I G U R E 6 | Comparison of experimental (dipole oscillator strength cally for each atom pair in a molecule and so include
distribution (DOSD)) and theoretical molecular C6 coefficients (1225 system dependency. This fundamentally distinguishes
cases, new compilation by A. Tkatchenko, FHI, Berlin, Germany, these modern DFT-D methods from older variants
private communication). Note the logarithmic scale and variation of that employ fixed values (e.g., Refs 29, 30, 35, 69,
the coefficients over three orders of magnitude. The test set contains 70, 74, 88). We mention here the work of Tkatchenko
small to medium-sized, ‘normal’ inorganic and organic molecules and Scheffler34 (C6 from atomic volumes), Sato et al.6
(H2 to C8 H18 ).

T A B L E 2 Mean Absolute Deviations (MADs, in kcal/mol) for the S22 Benchmark Set for Common
Dispersion-Corrected DFT Methods

Method Type of DF Class MAD Ref

ωB97X-D Hybrid GGA III 0.22 77

BLYP-D3 GGA III 0.23 28

vdW-DF (optB88) GGA + nonlocal I 0.23 41

LC-BOP + LRD Hybrid GGA III 0.27 6

B2PLYP-D3 Double hybrida III 0.29 28

VV09(PBE, rPW86) GGA + nonlocal I 0.29 110

PW86PBE + XDM GGA III 0.31 64

M06-2X Meta-hybrid GGA II 0.41 61

TPSS-d3 Meta-GGA III 0.45 28

B3LYP-D3 Hybrid GGA III 0.46 28

revPBE + LAP GGA IV 0.57 32

FN-DMC WF 0.68 111

MP2/CBS WF 0.78 112

vdW-DF (PBE) GGA +nonlocal I 1.19 38

PBE GGA No disp. 2.61 28

TPSS GGA No disp. 3.49 28

B3LYP Hybrid GGA No disp. 3.82 28

BLYP GGA No disp. 4.81 28

For comparison, values for two wave function (WF)-based methods and standard functionals without correction are also given.
a
Double-hybrid functionals partially account for dispersion by virtual orbital-dependent (perturbation) terms113, 114 but nevertheless
require in their current forms additional dispersion corrections. By using a scaling factor s6 < 1 in Eq. (8), which in all other cases is
unity, correct asymptotic behavior is achieved and double-counting effects can be avoided.

218 
c 2011 John Wiley & Sons, Ltd. Volume 1, March/April 2011
 WIREs Computational Molecular Science Density functional theory with London dispersion corrections

0 0

Ne Ne

–0.0001 –0.0001

Edisp / Eh
Edisp / Eh

–0.0002 In- plane Stacked –0.0002

In- plane Stacked
Stacked, MP2 Stacked, MP2
In-plane, MP2 In-plane, MP2
Stacked, TPSS-D3 Stacked, TPSS-D3
In-plane, TPSS-D3 In-plane, TPSS-D3
–0.0003 –0.0003

10 15 20 25 15 20 25 30 35
R CMA / Bohr R CMA / Bohr

F I G U R E 7 | Asymptotic region of the interaction between a neon atom and adenine (left) and two benzene molecules (right). The
intermolecular distance is given by the center-of-mass distance. The supermolecular MP2 correlation energies shown are taken as an approximation
to the dispersion energy and are computed by aug-cc-pVXZ (X = 2,3) basis set extrapolation.

(local atomic response function in the spirit of vdW- anisotropy is correctly accounted for because the
DF), and the dipole-exchange hole model (XDM) of C6 coefficients are atom-centered quantities. Their
Becke and Johnson78–80, 89 (see Refs 64, 75, 90 for spatial distribution basically reflects the dispersion
extensions). For the DFT-D (XDM) approach, SCF anisotropy of the system and, indeed in DFT-D, the
and gradient implementations have been reported dispersion energy between two molecules depends on
recently.91 their mutual orientation. This is demonstrated nu-
The atom pairwise additive schemes (as DFT- merically for two examples (neon–adenine and the
D) to compute the dispersion energy [that is used benzene dimer) in Figure 7. As it is clearly seen from
similarly in almost all empirical force fields (FFs)] the data, the D3 dispersion energy qualitatively cor-
has recently been criticized for its insufficient descrip- rectly follows the corresponding MP2 values for in-
tion of the anisotropy of the molecular dispersion plane and stacked arrangements. Note the inversion
energy.92 Because this is also an important point for between in-plane and stacked arrangements regard-
understanding of noncovalent interactions, it is briefly ing the lower dispersion energy in the two examples,
considered here. Molecular polarizabilities and the which is also right in DFT-D3. The absolutely lower
derived dispersion coefficients are tensorial quanti- values from MP2 can be explained by a systematic
ties and hence are different for different spatial di- overestimation of dispersion for unsaturated systems
rections. This anisotropy leads to a dependence of (see Ref 93 and references therein). In summary, it can
the intermolecular dispersion energy on the relative be stated that the description of dispersion anisotropy
orientation of the interacting molecular fragments. It in atom pairwise additive schemes is qualitatively cor-
was argued92 that DFT-D and FFs employ isotrop- rect and the remaining errors are on the order of the
ically averaged dispersion coefficients and therefore typical errors of such nonelectronic approaches. For
cannot described the dispersion anisotropy correctly. the effect of using nonisotropic, atomic dispersion co-
This conclusion was mainly based on the systematic efficients in DFT-D, see Ref 90.
over/underbinding of the FF-computed dispersion en-
ergy for stacked/H-bonded uracil dimers compared
with accurate DFT–SAPT data. One-Electron Corrections (1ePOT)
At least for large intermolecular distances, the The dispersion energy inherently is an electron corre-
argumentation is basically flawed. Anisotropic disper- lation effect and thus has a many-electron origin. In
sion coefficients are only necessary if the intermolec- WF theory, it reduces mathematically to a contrac-
ular dispersion energy is expanded in a molecule- tion of two-electron integrals with electron excitation
centered coordinate system with, e.g., a set of different (collision) amplitudes. It can be modeled by a nonlo-
(tensorial) C6 coefficients for each of the x, y, and cal, two-particle-dependent kernel that acts on elec-
z directions. In DFT-D and related methods, this tron densities (see section on vdW-DF). Keeping this

Volume 1, March/April 2011 

c 2011 John Wiley & Sons, Ltd. 219
Advanced Review wires.wiley.com/wcms

in mind, it seems rather odd to describe dispersion by interaction energies between hydrocarbons composed
an effective 1ePOT. However, beside computational of sp3 , sp2 , and sp hybridized carbon atoms in Ref
reasons, some support for this ansatz comes from the 101 reveals that the ethane dimer (with a small C6
fact that dispersion coefficients (polarizabilities) of value) is underbound by about 26%, whereas the cor-
molecules are represented rather well by adding lo- responding ethene and ethyne complexes very closely
cal, atom-like quantities (as in DFT-D). In contrast match the reference values.
to DFT-D, however, this correction is defined by a Compared with the latest version of DFT-D, the
potential and so produces changes in the electronic mentioned 1ePOT-type methods are more empirical
charge density. because each element (atom type) requires fitting of
von Lilienfeld et al.94–96 first employed this at least two potential parameters on reference inter-
concept in their method. It uses optimized, atom- action energies. In DFT-D3, the necessary atom pair-
centered nonlocal potentials (DCACP) that are nor- wise data (cutoff radius and C6 coefficient) are com-
mally used in the context of pseudopotentials for puted by DFT and are not fitted at all. The number
core electrons.97 The application of this method for of free parameters in DFT-D3 is only two, whereas it
modeling attractive long-range vdW forces is illus- equals the number of different atoms in the system in
trated for argon–argon, benzene–benzene, graphite– DCACP.
graphite, argon–benzene complexes; Arn Krm (n +
m ≤ 4) vdW clusters; and cyclooctatetraene and hy-
drogen bromide dimers. The approach has also been
applied to calculate interaction energies for pol-
INTERMOLECULAR INTERACTION
yaromatic hydrocarbon molecules from monocyclic
benzene up to hexabenzocoronene98, 99 and the ad-
ENERGY BENCHMARKS
sorption of Ar on graphite.100 For the most recent The ‘de facto’ standard benchmark for intermolec-
applications of DCACP, see Refs 101–104. ular interaction energies is the so-called S22 set of
The idea has later been used also by DiLabio and Hobza and coworkers107 (for a recent revision of the
coworkers5, 105 and by Sun et al.32 The basic differ- reference values, see Ref 108). Although the set does
ence with DCACP is that only the local part of stan- not contain the typical rare-gas dimers and molecules
dard effective core potentials is retained. These meth- comprising heavier atoms than HCNO, it is rather
ods perform well for equilibrium structures of organic representative for many types of weak interactions
molecules in standard benchmarks (e.g., S22 set) but and therefore extensively used in the literature. Note
little is known how they perform for intramolecular that in its original form it contains only equilibrium
cases or thermochemistry. Similar to parameterized structures (MP2 optimized). Hence, benchmarking on
DFs (class II), it seems difficult to extract reason- this set provides only a limited information about the
able insight about the dispersion effects from such performance of a method for shorter or longer inter-
calculations. molecular distances. This is of particular importance
The most serious disadvantage of these ap- for the dispersion energy in large systems that is domi-
proaches is, however, that the currently used po- nated by the asymptotic (long-range) behavior, which
tentials do not show the correct asymptotic R−6 is not exact in all methods. For recent developments
behavior and decay too fast (exponentially) with in- of this and related sets, see Ref 109.
teratomic distance (as in class II methods). As a result, Table 2 shows mean absolute deviations
graphene sheets are underbound by about 20% of De (MADs) for common dispersion-corrected DFT meth-
with DCACP.101 Furthermore, the numerical and hu- ods taken from the literature. To put these values
man effort to determine the potential parameters [two into some broader perspective, values for the widely
(DCACP) to four (DCP) per element] is high and up used MP2 method [at the estimated complete ba-
to now only a small fraction of the periodic table is sis set (CBS) limit] and electronic fixed-node diffu-
covered.106 The approach is also flawed by the fact sion Monte Carlo (FN-DMC) are given. Both meth-
that the atomic parameters are fixed for each element ods are potentially applicable (as is DFT) to large
and therefore do not reflect the changes of dispersion systems (about 100–200 atoms currently). The av-
coefficients with the hybridization or oxidation state erage dissociation energy of the S22 complexes is
of an atom in a molecule or solid. At present, it is about 7 kcal/mol. By comparing the original107 and
not clear whether these effects are really accounted more recent108 (and better) CCSD(T) reference data
for by the nonlocality of the DCACP (DCP and LAP and including estimates for the inherent CCSD(T) er-
definitely do not include it). A DCACP study of the ror, one can estimate that the residual ‘noise’ in the

220 
c 2011 John Wiley & Sons, Ltd. Volume 1, March/April 2011
 WIREs Computational Molecular Science Density functional theory with London dispersion corrections

reference values might lead to a limiting MAD of and additional dispersion corrections are required.114
about 0.1 kcal/mol for any method. Without these, B2PLYP (although performing better
As can be seen, this accuracy is almost attained than plain B3LYP) would not be competitive to the
by various variants of dispersion-corrected DFT. In other methods for the S22 set in Table 2 (i.e, the MAD
particular, the ‘simple’ DFT-D method as in ωB97X- of dispersion-uncorrected B2PLYP is 1.84 kcal/mol).
D and BLYP-D3 performs extraordinarily well. The
vdW-DF approach when used with empirically ad-
justed GGA parts [in the two different flavors vdW-
INTRAMOLECULAR DISPERSION
DF(optB88) and VV09] is also very successful. Similar
accuracy is also obtained by the most recent DFT-
EFFECTS FOR THERMOCHEMISTRY
D variant of Kannemann and Becke,64 who used Because dispersion is an electron correlation effect,
the Becke–Johnson XDM model for the computa- it operates also intramolecularly (between atoms or
tion of the C6 coefficients. All these methods (that are functional groups that are not directly bonded to
asymptotically correct) provide small MADs (about each other). This contributes to the internal energy
0.3 kcal/mol or less), which leads to errors of typ- of (mainly larger) molecules and must be considered
ically less than 5% for De . The heavily parameter- for accurate thermochemical computations.
ized M06-2X functional on the contrary contains In 2009, we published the so-called GMTKN24
a reasonable description of dispersion only in the database, which is a collection of 24 previously pub-
medium range (low but finite density regions). Asymp- lished or newly developed benchmark sets for general
totically, it does not provide any dispersion energy main group thermochemistry, kinetics, and nonco-
(exponentially decaying density). Although an MAD valent interactions.7 Very recently, it was extended
of 0.41 kcal/mol for M06-2X is quite good, it is by six additional sets and dubbed GMTKN30.117 It
worse than properly corrected methods (class I and covers atomization energies, electron affinities, ion-
III). This indicates that for the interaction energies ization potentials, proton affinities, self-interaction
of relatively small molecules, the asymptotic behav- error-related problems, barrier heights, various re-
ior is significant. This conclusion is supported by the action energies, particularly difficult cases for DFT
large MAD for the revPBE + LAP class IV method methods, relative energies between conformers, and
(0.57 kcal/mol), which also lacks this property. In inter- and intramolecular noncovalent interactions
passing, we note that all fully nonempirical methods problems. Reference values are based on highly ac-
[FN-DMC, MP2, and vdW-DF(PBE)] do not perform curate theoretical or experimental data. GMTKN30
very well (although much better than a dispersion- makes it possible to thoroughly evaluate existing
uncorrected GGA such as BLYP, which provides as methods and fosters the development of new DFT
disastrous value of 4.81 kcal/mol). In the case of the approaches.
vdW-DF, this can clearly be attributed to double- As handling and discussion of the large number
counting effects of electron correlation at intermedi- of statistical values for such a database can be un-
ate distance. Corrected standard functionals without practical, we defined a so-called weighted total MAD
further/special adjustment (e.g., B3LYP or TPSS) typ- (WTMAD), which combines all 30 MADs into one
ically yield MAD values of about 0.4 kcal/mol, which final number. For every subset, the size and ‘diffi-
are sufficient for many practical purposes. A similar culty’ is taken into account by a factor with which
accuracy is obtained by revPBE + LAP (an 1ePOT ap- each MAD is scaled. Finally, the average is taken for
proach), whereas S22 data with the DCACP method these scaled MADs. With the help of WTMADs, a
have not been reported yet. For a larger benchmark set direct comparison between different DFT methods is
of biomolecular complexes similar to S22 (Ref 115), easy to carry out. In previous investigations, we al-
BLYP–DCACP seems to be slightly less accurate than ready pointed out that including a proper description
BLYP-D2.116 of London dispersion effects has a positive impact on
The double-hybrid DFs that overall perform the final WTMAD values.7, 87, 117
best (see also next sections) deserve a special com-
ment. Because they account for dispersion by virtual
orbital-dependent (perturbation) terms113 at all in- Chemical Reactions
terelectronic distances, they do not strictly belong to For the present context, we consider two different
any of the classes considered in this review. How- aspects of the GMTKN30 set. The first 20 subsets
ever, in their current form, the nonlocal correlation deal with thermochemistry and kinetics—mainly re-
terms used for overall optimum performance are too action energies and barrier heights—the remaining
small to accurately describe noncovalent interactions 10 with noncovalent interactions. For the first 20

Volume 1, March/April 2011 

c 2011 John Wiley & Sons, Ltd. 221
Advanced Review wires.wiley.com/wcms

9 8.8

DFT
8 DFT-D3
7.1 7.2
7 6.8 6.9
6.4
6.1
WTMAD / (kcal/mol)

5 4.8

4 3.8
3.5
3.3
3 2.7 2.8
2.2
2

0
TPSS

PW6B95
PBE
BLYP

B2GPPLYP
B3LYP

B2PLYP
F I G U R E 8 | Weighted total mean absolute deviation (WTMAD) for thermochemical benchmarks (noncovalent interactions excluded) with and
without dispersion correction for a series of representative density functionals.

subsets, we calculated a WTMAD, with the same tide, alkane, sugar, and cysteine conformers. The re-
factors as described in Ref 117, for seven different spective WTMADs are shown in Figure 9. Similar
functionals from the GGA to the double-hybrid level. to the case of the S22 set, uncorrected DFs provide
Figure 8 shows these WTMADs for uncorrected func- large errors. The advantages of including dispersion
tionals, and when combined with the latest atom is more obvious than for ‘normal’ chemical reactions.
pairwise London dispersion correction (−D3). Two The reason for this is rooted in the large changes of the
effects can be observed. First of all, the WTMADs de- intramolecular dispersion energy at medium and long
crease with increasing sophistication of the DF (GGA ranges for different conformers. This is often related
> meta-GGA > hybrid > double hybrid). Second, in to drastic changes of the three-dimensional structure
all cases, the WTMAD of an uncorrected DF is re- (e.g., folded vs. unfolded conformations). The same
duced by adding the dispersion correction. BLYP and effects are also present in many chemical reactions
B3LYP benefit the most from adding the correction but are smaller on a percentage basis and often not
(improvements by 1.7 and 1.6 kcal/mol). The WT- fully recognized. In any case, one can conclude that
MAD of PBE improves by 0.4 kcal/mol, of TPSS118 the relevant noncovalent interactions between organic
by 0.8, and of PW6B95119 by 0.5 kcal/mol. Although molecules (intermolecular) and in different conform-
double hybrids include a portion of nonlocal corre- ers (intramolecular) are physically similar.
lation, including the dispersion correction is still im- The WTMAD of BLYP is reduced by 2 kcal/mol
portant (reduction by 0.8 kcal/mol for B2PLYP113 at the DFT-D3 level. With an WTMAD of
and 0.6 kcal/mol for B2GPPLYP120 ). These results 0.5 kcal/mol, BLYP-D3 can even compete with
clearly show how important the inclusion of disper- the hybrids B3LYP-D3 and PW6B95-D3 (0.3 and
sion effects for normal thermochemistry is. For recent 0.4 kcal/mol). The improvement from B3LYP to
examples including also transition metal complexes, B3LYP-D3 is also about 2 kcal/mol, whereas pure
see Refs 87, 121, 122. PW6B95 already has a WTMAD of 0.7 kcal/mol.
BLYP-D3 is much better than the other (meta-)GGAs
PBE-D3 and TPSS-D3 (0.9 and 0.8 kcal/mol), for
Conformational Energies which the WTMADs are reduced by more than
Finally, the same analysis is repeated with four subsets 1 kcal/mol. As for thermochemistry and kinetics, the
of GMTKN30 that cover relative energies of tripep- double hybrids also require an additional dispersion

222 
c 2011 John Wiley & Sons, Ltd. Volume 1, March/April 2011
 WIREs Computational Molecular Science Density functional theory with London dispersion corrections

2.5
2.5
DFT
2.2 2.2
DFT-D3

2.0
2
WTMAD / (kcal/mol)

1.5

1.1
1 0.9
0.8 0.8
0.7

0.5
0.5 0.4
0.3
0.2
0.1
0
TPSS

PW6B95
PBE
BLYP

B2GPPLYP
B3LYP

B2PLYP
F I G U R E 9 | Weighted total mean absolute deviation (WTMAD) for conformational benchmarks with and without dispersion correction and
employing a series of representative density functionals.

correction. The WTMADs are then reduced from 1.1 for generally high accuracy seems to be a correct de-
(B2PLYP) and 0.7 kcal/mol (B2GPPLYP) to 0.2 and scription of the asymptotic −C6 /R6 behavior of the
0.1 kcal/mol. The reference data for these sets are interaction energy. However, the chemically and
based on estimated CCSD(T)/CBS values, for which physically most interesting region of interatomic dis-
an accuracy of ±0.1 kcal/mol for these types of inter- tances in and between molecules is typically about
actions is expected. Thus, double hybrids lie within 3–4 Å. It is also clear that a balanced treatment
this accuracy. But the performance of dispersion- of correlation in this overlapping region is neces-
corrected GGA and hybrid DF is so good that typical sary for high accuracy. This requirement currently
problems of structural biochemistry can be treated leads in all accurate dispersion-corrected DFT meth-
routinely. ods to the inclusion of some empiricism (a few global-
fit parameters). The best corrected DFT methods
are relatively close to the WF-based ‘gold standard’
CCSD(T) for dispersion-dominated noncovalent in-
CONCLUSIONS teractions. Remaining deficiencies are probably more
The computationally most efficient basic approaches related to other DFT problems than to dispersion
to account for London dispersion effects in DFT cal- (e.g., overdelocalization, self-interaction error). Al-
culations have been reviewed. These do not include though being the simplest of the discussed correc-
any information from virtual orbitals but instead use tions, DFT-D in the most recent forms is truly com-
only electron density-based information (vdW-DF, petitive with the other methods. Furthermore, it is
parameterized DF), rely on a semiclassical atom pair- computationally much faster, numerically very stable,
wise potential (DFT-D), or employ effective 1ePOT. and provides physical insight. It is the recommended
Solid thermochemical benchmarks for typical vdW method in combination with standard functionals
complexes, conformational processes in (bio)organic and for large-scale geometry optimizations. Although
molecules, and for standard chemical reactions en- the ‘dispersion problem’ in DFT has not been com-
ergies (for not too small molecules) show that in- pletely solved, accurate and practical solutions are
clusion of dispersion effects (nonlocal, medium-to- now well established so that the future of DFT for the
long-range correlation energy) is mandatory to obtain computation of large nano/biosystems seems to be
so-called chemical accuracy. An essential ingredient bright.

Volume 1, March/April 2011 

c 2011 John Wiley & Sons, Ltd. 223
Advanced Review wires.wiley.com/wcms

ACKNOWLEDGEMENTS
This work was supported by the Deutsche Forschungsgemeinschaft in the framework of the
SFB 858. I want to thank my students L. Goerigk and S. Ehrlich for providing data and
performing some calculations presented in this work.

REFERENCES
1. Stone AJ. The Theory of Intermolecular Forces. Ox- 14. Couronne O, Ellinger Y. An ab initio and DFT study
ford: Oxford University Press; 1997. of (N2 )2 dimers. Chem Phys Lett 1999, 306:71–77.
2. Kaplan IG, Intermolecular Interactions. Chichester: 15. Perdew JP. In: Ziesche P, Eschrig H, eds. Proceed-
John Wiley & Sons; 2006. ings of the 21st Annual International Symposium on
3. Grimme S, Antony J, Schwabe T, Mück-Lichtenfeld the Electronic Structure of Solids. Akademie Verlag,
C. Density Functional Theory with Dispersion Cor- Berlin, Germany; 1991.
rections for Supramolecular Structures, Aggregates, 16. Becke AD. Density-functional exchange-energy ap-
and Complexes of (Bio)Organic Molecules. Org proximation with correct asymptotic behavior. Phys
Biomol Chem 2007, 5:741–758. Rev A 1988, 38:3098–3100.
4. Gräfenstein J, Cremer D. An efficient algorithm for 17. Lee C, Yang W, Parr RG. Development of the Colle-
the density-functional theory treatment of dispersion Salvetti correlation-energy formula into a functional
interactions. J Chem Phys 2009, 130:124105. of the electron density. Phys Rev B 1988, 37:785–
5. Johnson ER, Mackie ID, DiLabio GA. Dispersion in- 789.
teractions in density-functional theory. J Phys Org 18. Becke AD. Density-Functional Thermochemistry. III.
Chem 2009, 22:1127–1135. The Role of Exact Exchange. J Chem Phys 1993,
6. Sato T, Nakai H. Local response dispersion method. 98:5648–5652.
II. Generalized multicenter interactions. J Chem Phys 19. Stephens PJ, Devlin FJ, Chabalowski CF, Frisch MJ.
2009, 131:224104. Ab initio calculation of vibrational absorption and
7. Goerigk L, Grimme S. A General Database for Main circular dichroism spectra using density functional
Group Thermochemistry, Kinetics, and Noncovalent force fields. J Phys Chem 1994, 98:11623–11627.
Interactions - Assessment of Common and Repa- 20. Slavicek P, Kalus R, Paska P, Odvarkova I, Hobza
rameterized (meta)GGA Density Functionals. J Chem P, Malijevsky A. State-of-the-art correlated ab initio
Theor Comput 2010, 6:107–126. potential energy curves for heavy rare gas dimers: Ar2 ,
8. Kristyán S, Pulay P. Can (semi)local density func- Kr2 , and Xe2 . J Chem Phys 2003, 119:2102–2118.
tional theory account for the London dispersion 21. Sherrill CD, Takatani T, Hohenstein EG. An As-
forces? Chem Phys Lett 1994, 229:175–180. sessment of Theoretical Methods for Nonbonded In-
9. Hobza P, Sponer J, Reschel T. Density Functional teractions: Comparison to Complete Basis Set Limit
Theory and Molecular Clusters. J Comp Chem 1995, Coupled-Cluster Potential Energy Curves for the Ben-
16:1315–1325. zene Dimer, the Methane Dimer, Benzene-Methane,
10. Pérez-Jordá JM, Becke AD. A density-functional and Benzene-H2S CH/π , and SH/π Interactions.
study of van der Waals forces: rare gas diatomics. J Phys Chem A 2009, 113:10146–10159.
Chem Phys Lett 1995, 233:134–137. 22. Perdew JP, Burke K, Ernzerhof M. Generalized Gra-
11. Pérez-Jordá JM, San-Fabián E, Pérez-Jiménez AJ. A dient Approximation Made Simple. Phys Rev Lett
density-functional study of van der Waals forces on 1996, 77:3865–3868.
rare-gas diatomic: Hartree-Fock exchange. J Chem 23. Cremer D. Møller-Plesset Perturbation Theory.
Phys 1999, 110:1916–1920. WIREs Comput Mol Sci 2011, Accepted.
12. Kurita N, Sekino H. Ab initio and DFT studies for 24. Grimme S. Seemingly simple stereoelectronic effects in
accurate description of van der Waals interaction be- alkane isomers and the implications for Kohn–Sham
tween He atoms. Chem Phys Lett 2001, 348:139– density functional theory. Angew Chem Int Ed 2006,
146. 45:4460–4464.
13. Sponer J, Leszczynski J, Hobza P. Base stacking in 25. Dion M, Rydberg H, Schröder E, Langreth DC,
cytosine dimer. A comparison of correlated ab ini- Lundqvist BI. Van der Waals Density Functional for
tio calculations with three empirical potential models General Geometries. Phys Rev Lett 2004, 92:246401.
and density functional theory calculations. J Comput 26. Lee K, Murray ED, Kong L, Lundqvist BI, Langreth
Chem 1996, 17:841–850. DC. A Higher-Accuracy van der Waals Density

224 
c 2011 John Wiley & Sons, Ltd. Volume 1, March/April 2011
 WIREs Computational Molecular Science Density functional theory with London dispersion corrections

Functional. Phys Rev B 2010, 82:081101(R). arXiv: 42. Casimir HBG, Polder D. The Influence of Retardation
1003.5255v2 [cond-mat.mtrl-sci]. on the London-van der Waals Forces. Phys Rev 1948,
27. Zhao Y, Truhlar DG. Density Functionals with Broad 73:360–372.
Applicability in Chemistry. Acc Chem Res 2008, 43. Vydrov OA, Van Vorhis T. Nonlocal van der Waals
41:157–167. Density Functional Made Simple. Phys Rev Lett
28. Grimme S, Antony J, Ehrlich S, Krieg H. A consis- 2009, 103:063004.
tent and accurate ab initio parametrization of density 44. Vydrov OA, Van Voorhis T. Dispersion interactions
functional dispersion correction (DFT-D) for the 94 from a local polarizability model. Phys Rev A 2010,
elements H-Pu. J Chem Phys 2010, 132:154104. 81:062708.
29. Grimme S. Accurate description of van der Waals 45. Vydrov OA. Simple nonlocal correlation function-
complexes by density functional theory including em- als incorporating dispersion interactions. In: CECAM
pirical corrections. J Comput Chem 2004, 25:1463– Workshop on vdW Interactions in DFT, Lausanne;
1473. July 2010.
30. Jurecka P, Cerny J, Hobza P, Salahub DR. Density 46. Waller MP, Robertazzi A, Platts JA, Hibbs DE,
Functional Theory Augmented with an Empirical Dis- Williams PA. Use of DFT for prediction of π -stacking
persion Term. Interaction Energies and Geometries of interactions: Applications to benzenes, pyridines, and
80 Noncovalent Complexes Compared with Ab Initio DNA bases. J Comp Chem 2006, 27:491–504.
Quantum Mechanics Calculations. J Comput Chem 47. Hyla Kryspin I, Haufe G, Grimme S. Weak hydrogen
2007, 28:555–569. bridges: a systematic theoretical study on the nature
31. von Lilienfeld OA, Tavernelli I, Röthlisberger U, und strength of C H· · ·F C interactions. Chem Eur
Sebastiani D. Optimization of effective atom centered J 2004, 10:3411–3422.
potentials for London dispersion forces in density 48. Xu X, Goddard, WA III. The X3LYP extended den-
functional theory. Phys Rev Lett 2004, 93:153004. sity functional for accurate descriptions of nonbond
32. Sun YY, Kim Y-H, Lee K, Zhang SB. Accurate and interactions, spin states, and thermochemical prop-
efficient calculation of van der Waals interactions erties. Proc Natl Acad Sci U S A 2004, 101:2673–
within density functional theory by local atomic po- 2677.
tential approach. J Chem Phys 2008, 129:154102.
49. Xu X, Goddard, WA III. Bonding properties of the
33. Johnson ER, Becke AD. A post-Hartree-Fock model water dimer: A comparative study of density func-
of intermolecular interactions: inclusion of higher- tional theories. J Phys Chem A 2004, 108:2305–
order corrections. J Chem Phys 2006, 124:174104. 2313.
34. Tkatchenko A, Scheffler M. Accurate Molecular Van 50. Černý J, Hobza P. The X3LYP Extended Density
Der Waals Interactions from Ground-State Electron Functional Accurately Describes H-Bonding but Fails
Density and Free-Atom Reference Data. Phys Rev Completely for Stacking. Phys Chem Chem Phys
Lett 2009, 102:073005. 2005, 7:1624–1626.
35. Grimme S. Semiempirical GGA-Type Density Func- 51. Walsh TR. Exact exchange and Wilson–Levy correla-
tional Constructed with a Long-Range Dispersion tion: a pragmatic device for studying complex weakly-
Correction. J Comput Chem 2006, 27:1787–1799. bonded systems. Phys Chem Chem Phys 2005, 7:443–
36. Andersson Y, Langreth DC, Lundqvist BI. van der 451.
Waals Interactions in Density-Functional Theory. 52. Staroverov VN, Scuseria GE, Tao J, Perdew JP.
Phys Rev Lett 1996, 76:102–105. Comparative assessment of a new nonempirical den-
37. Berland K, Borck Ø, Hyldgaard P. van der Waals sity functional: Molecules and hydrogenbonded com-
density functional calculations of binding in molecu- plexes. J Chem Phys 2003, 119:12129–12137.
lar crystals. arXiv:1007.3305v1. 53. Tao J, Perdew JP. Test of a nonempirical density
38. Gulans A, Puska MJ, Nieminen RM. Linear-scaling functional: Short-range part of the van der Waals in-
self-consistent implementation of the van der Waals teraction in rare-gas dimmers. J Chem Phys 2005,
density functional. Phys Rev B 2009, 79:201105. 122:114102.
39. Murray ED, Lee K, Langreth DC. Investigation of 54. Ruzsinszky A, Perdew JP, Csonka GI. Binding En-
Exchange Energy Density Functional Accuracy for In- ergy Curves from Nonempirical Density Function-
teracting Molecules. J Chem Theory Comput 2009, als II. Van der Waals Bonds in Rare-gas Dimers
5:2754–2762. and Alkaline-Earth Diatomics. J Phys Chem A 2005,
40. Zhang Y, Yang W. Comment on “Generalized Gra- 109,11015–11021.
dient Approximation Made Simple”. Phys Rev Lett 55. Adamo C, Barone V. Exchange functionals with im-
1998, 80:890. proved long-range behavior and adiabatic connection
41. Klimes J, Bowler DR, Michaelides A. Chemical accu- methods without adjustable parameters: The mPW
racy for the van der Waals density functional. J Phys and mPW1PW models. J Chem Phys 1998, 108664–
Condens Matter 2010, 22:022201. 675.

Volume 1, March/April 2011 

c 2011 John Wiley & Sons, Ltd. 225
Advanced Review wires.wiley.com/wcms

56. Zhao Y, Truhlar DG. Comparative DFT Study 69. Elstner M, Hobza P, Frauenheim T, Suhai S,
of van der Waals Complexes: Rare-Gas Dimers, Kaxiras E. Hydrogen bonding and stacking in-
Alkaline-Earth Dimers, Zinc Dimer, and Zinc-Rare- teractions of nucleic acid base pairs: A density-
Gas Dimers. J Phys Chem A 2006, 110:5121–5129. functional-theory based treatment. J Chem Phys
57. Zhao Y, Schultz NE, Truhlar DG. Design of Density 2001, 114:5149–5155.
Functionals by Combining the Method of Constraint 70. Wu Q, Yang W. Empirical correction to density func-
Satisfaction with Parametrization for Thermochem- tional theory for van der Waals interactions. J Chem
istry, Thermochemical Kinetics, and Noncovalent In- Phys 2002, 116:515–524.
teractions. J Chem Theory Comput. 2006, 2:364– 71. Gianturco FA, Paesani F, Laranjeira MF, Vassilenko
382. V, Cunha MA. Intermolecular forces from density
58. Zhao Y, Truhlar DG. Hybrid Meta Density Func- functional theory: III a multiproperty analysis for
tional Theory Methods for Thermochemistry, Ther- the Ar(1S)-CO(1 ) interaction. J Chem Phys 1999,
mochemical Kinetics, and Noncovalent Interactions: 110:7832.
The MPW1B95 and MPWB1K Models and Compar- 72. Cohen JS, Pack RT. Modified statistical method for
ative Assessments for Hydrogen Bonding and van der lar potentials. Combining rules for higher Van der
Waals Interactions. J Phys Chem A 2004, 108:6908– Waals coefficients. J Chem Phys 1974, 61:2372–
6918. 2382.
59. Zhao Y, Truhlar DG. How well can new-generation 73. von Lilienfeld OA, Tkatchenko A. Two- and three-
density functional methods describe stacking interac- body interatomic dispersion energy contributions to
tions in biological systems? Phys Chem Chem Phys binding in molecules and solids. J Chem Phys 2010,
2005, 7:2701–2705. 132:234109.
60. Zhao Y, Truhlar DG. Multicoefficient Extrapolated 74. Steinmann SN, Csonka G, Corminboeuf C. Unified
Density Functional Theory Studies of π · · ·π Inter- Inter- and Intramolecular Dispersion Correction For-
actions: The Benzene Dimer. J Phys Chem A 2005, mula for Generalized Gradient Approximation Den-
109:4209–4212. sity Functional Theory. J Chem Theory Comput.
61. Zhao Y, Truhlar DG. A Prototype for Graphene Ma- 2009, 5:2950–2958.
terial Simulation: Structures and Interaction Poten-
75. Steinmann SN, Corminboeuf C. A System-Dependent
tials of Coronene Dimers. J Phys Chem C 2008,
Density-Based Dispersion Correction. J Chem Theory
112:4061–4067.
Comput 2010, 6:1990–2001.
62. Hohenstein EG, Chill ST, Sherrill CD. Assessment
76. Liu Y, Goddard WA III. A Universal Damping Func-
of the Performance of the M05-2X and M06-2X
tion for Empirical Dispersion Correction on Den-
Exchange-Correlation Functionals for Noncovalent
sity Functional Theory. Mat Trans 2009, 50:1664–
Interactions in Biomolecules. J Chem Theory Com-
1670.
put 2008, 4:1996–2000.
77. Chai J-D, Head-Gordon M. Long-range corrected hy-
63. Pernal K, Podeszwa R, Patkowski K, Szalewicz K.
brid density functionals with damped atom–atom dis-
Dispersionless Density Functional Theory. Phys Rev
persion corrections. Phys Chem Chem Phys 2000,
Lett 2009, 103:263201.
106615–6620.
64. Kannemann FO, Becke AD. van der Waals Interac-
78. Becke AD, Johnson ER. Exchange-hole dipole mo-
tions in Density-Functional Theory: Intermolecular
ment and the dispersion interaction. J Chem Phys
Complexes. J Chem Theory Comput 2010, 6:1081–
2005, 122:154104.
1088.
65. Hepburn J, Scoles G, Penco R. A simple but reliable 79. Becke AD, Johnson ER. A density-functional model
method for the prediction of intermolecular poten- of the dispersion interaction. J Chem Phys 2005,
tials. Chem Phys Lett 1975, 36:451–456. 123:154101.
66. Ahlrichs R, Penco R, Scoles G. Intermolecular 80. Becke AD, Johnson ER. Exchange-hole dipole mo-
forces in simple systems. Chem Phys 1977, 19:119– ment and the dispersion interaction: High-order dis-
130. persion coefficients. J Chem Phys 2006, 124:014104.
67. Clementi E, Corongiu G. Van der Waals interaction 81. Koide A. A new expansion for dispersion forces and
energies of helium, neon, and argon with naphtha- its application. J Phys B 1976, 9:3173.
lene. J Phys Chem A 2001, 105:10379–10383. 82. Starkschall G, Gordon RG. Calculation of Coeffi-
68. Gonzalez C, Lim EC. Evaluation of the Hartree–Fock cients in the Power Series Expansion of the Long-
Dispersion (HFD) Model as a Practical Tool for Prob- Range Dispersion Force between Atoms. J Chem Phys
ing Intermolecular Potentials of Small Aromatic Clus- 1972, 56:2801–2806.
ters: Comparison of the HFD and MP2 Intermolec- 83. Thakkar AJ, Hettema H, Wormer PE. Ab initio dis-
ular Potentials. J Phys Chem A 2003, 107:10105– persion coefficients for interactions involving rare gas
10110. atoms. J Chem Phys 1992, 97:3252–3257.

226 
c 2011 John Wiley & Sons, Ltd. Volume 1, March/April 2011
 WIREs Computational Molecular Science Density functional theory with London dispersion corrections

84. Moellmann J, Grimme S. Importance of London dis- 97. Cao X, Dolg M. Pseudopotentials and modelpoten-
persion effects for the packing of molecular crystals: a tials. WIREs Comput Mol Sci 2011, 1:200–210.
case study for intramolecular stacking in bigthiophene 98. von Lilienfeld OA, Andrienko D. Coarse-grained in-
derivative. Phys Chem Chem Phys 2010, 12:8500– teraction potentials for polyaromatic hydrocarbons.
8504. J Chem Phys 2006, 124:054307.
85. Yousaf KE, Brothers EN. Applications of Screened 99. Cinacchi G. Comment on “Coarse-grained interac-
Hybrid Density Functionals with Empirical Disper- tion potentials for polyaromatic hydrocarbons”. J
sion Corrections to Rare Gas Dimers and Solids. J Chem Phys 2006, 125:057101.
Chem Theory Comput 2010, 6:864–872. 100. Tkatchenko T, von Lilienfeld OA. Adsorption of Ar
86. Goerigk L, Grimme S. Phys Chem Chem Phys DOI: on graphite using London dispersion forces corrected
10.1039/C0CP02984J. http://www.uni-muenster.de/ Kohn-Sham density functional theory. Phys Rev B
Chemie.oc/grimme/. 2006, 73:153406.
87. Huenerbein R, Schirmer B, Moellmann J, Grimme 101. Tapavicza E, Lin I, von Lilienfeld OA, Tavernelli
S. Effects of London dispersion on the isomeriza- I, Coutinho-Neto MD, Rothlisberger U. Study of
tion reactions of large organic molecules: a density weakly bonded carbon compounds using dispersion
functional benchmark study. Phys Chem Chem Phys corrected density functional theory. J Chem Theory
2010, 12:6940–6948. Comput 2007, 3:1673–1679.
88. Zimmerli U, Parrinello M, Koumoutsakos P. Dis- 102. Lin I, von Lilienfeld OA, Coutinho-Neto MD,
persion corrections to density functionals for water Tavernelli I, Rothlisberger U. Predicting noncova-
aromatic interactions. J Chem Phys 2004, 120:2693– lent interactions between aromatic biomolecules with
2699. London-dispersion-corrected DFT. J Phys Chem B
89. Becke AD, Johnson ER. A unified density-functional 2007, 111:14346–14354.
treatment of dynamical, nondynamical, and disper- 103. Lin I, Seitsonen AP, Coutinho-Neto MD, Tavernelli I,
sion correlations. J Chem Phys 2007, 127:124108. Rothlisberger U. Importance of weak interactions in
90. Krishtal A, Vanommeslaeghe K, Olasz A, Veszpremi liquid water. J Phys Chem B 2009, 113:1127–1131.
T, VanAlsenoy C, Geerlings P. Accurate Interaction 104. Arey JS, Aeberhard PC, Lin I, Rothlisberger U. Hy-
Energies at DFT Level by means of Efficient Disper- drogen bonding described using dispersion-corrected
sion Correction. J Chem Phys 2009, 130:174101. density functional theory. J Phys Chem B 2009,
91. Kong J, Gan Z, Proynov E, Freindorf M, Furlani 113:4726–4732.
TR. Efficient computation of the dispersion interac- 105. DiLabio GA. Accurate treatment of van der Waals
tion with density-functional theory. Phys Rev A 2009, interactions using standard density functional theory
79:042510. methods with effective core-type potentials: Applica-
92. Zgarbova M, Otyepka M, Sponer J, Hobza P, Jurecka tion to carbon-containing dimers. Chem Phys Lett
P. Large-scale compensation of errors in pairwise- 2008, 455:348–353.
additive empirical force fields: comparison of AMBER 106. Cascella M, Lin I, Tavernelli I, Rothlisberger U.
intermolecular terms with rigorous DFT-SAPT cal- Dispersion Corrected Atom-Centered Potentials for
culations. Phys Chem Chem Phys 2010, 12:10476– Phosphorus. J Chem Theory Comput 2009, 5:2930–
10493. 2934.
93. Grimme S, Mück-Lichtenfeld C, Antony J. Analysis 107. Jurecka P, Sponer J, Cerny J, Hobza P. Benchmark
of non-covalent interactions in (bio)organic molecules database of accurate (MP2 and CCSD(T) complete
using orbital-partitioned localized MP2. Phys Chem basis set limit) interaction energies of small model
Chem Phys 2008, 10:3327–3334. complexes, DNA base pairs, and amino acid pairs.
Phys Chem Chem Phys 2006, 8:1985–1993.
94. von Lilienfeld OA, Tavernelli I, Röthlisberger U,
Sebastiani D. Optimization of effective atom centered 108. Takatani T, Hohenstein EG, Malagoli M, Marshall
potentials for London dispersion forces in density MS, Sherrill CD. Basis Set Consistent Revision of the
functional theory. Phys Rev Lett 2004, 93:153004. S22 Test Set of Noncovalent Interaction Energies. J
Chem Phys 2010, 132:144104.
95. von Lilienfeld OA, Tavernelli I, Röthlisberger U,
Sebastiani D. Performance of optimized atom cen- 109. Hobza P, Rezac J, eds. http://www.begdb.com/.
tered potentials for weakly bonded systems using 110. Vydrov OA, Van Voorhis T. Implementation and as-
density functional theory. Phys Rev B 2005, 71: sessment of a simple nonlocal van der Waals density
195119. functional. J Chem Phys 2010, 132:164113.
96. von Lilienfeld OA, Tavernelli I, Röthlisberger U, 111. Korth M, Grimme S. Mindless DFT Benchmarking. J
Sebastiani D. Variational optimization of effective Chem Theory Comput 2009, 5:993–1003.
atom centered potentials for molecular properties. J 112. Antony J, Grimme S. Is Spin-Component Scaled
Chem Phys 2005, 122:014113. Second-Order Møller–Plesset Perturbation Theory an

Volume 1, March/April 2011 

c 2011 John Wiley & Sons, Ltd. 227
Advanced Review wires.wiley.com/wcms

Appropriate Method for the Study of Noncovalent 118. Tao J, Perdew JP, Staroverov VN, Scuseria GE.
Interactions in Molecules? J Phys Chem A 2007, Climbing the Density Functional Ladder: Nonempir-
111:4862–4868. ical Meta–Generalized Gradient Approximation De-
113. Grimme S. Semiempirical hybrid density functional signed for Molecules and Solids. Phys Rev Lett 2003,
with perturbative second-order correlation. J Chem 91:146401.
Phys 2006, 124:034108. 119. Zhao Y, Truhlar DG. Design of Density Function-
114. Schwabe T, Grimme S. Double-hybrid density func- als That Are Broadly Accurate for Thermochem-
tionals with long-range dispersion corrections: higher istry, Thermochemical Kinetics, and Nonbonded
accuracy and extended applicability. Phys Chem Interactions. J Phys Chem A 2005, 109:5656–
Chem Phys 2007, 9:3397–3406. 5667.
115. Antony J, Grimme S. Density functional theory in- 120. Karton A, Tarnopolsky A, Lamere JF, Schatz GC,
cluding dispersion corrections for intermolecular in- Martin JML. Highly Accurate First-Principles Bench-
teractions in a large benchmark set of biologically mark Data Sets for the Parametrization and Valida-
relevant molecules. Phys Chem Chem Phys 2006, tion of Density Functional and Other Approximate
8:5287–5293. Methods. Derivation of a Robust, Generally Applica-
116. Lin I, Rothlisberger U. Describing weak interac- ble, Double-Hybrid Functional for Thermochemistry
tions of biomolecules with dispersion-corrected den- and Thermochemical Kinetics. J Phys Chem A 2008,
sity functional theory. Phys Chem Chem Phys 2008, 112:12868–12886.
10:2730–2734. 121. Schwabe T, Grimme S, Djukic JP. Noncovalent
117. Goerigk L, Grimme S. Efficient and Accurate Double- Metal-Metal Interactions: The Crucial Role of Lon-
Hybrid-Meta-GGA Density Functionals–Evaluation don Dispersion in a Bimetallic Indenyl System. J Am
with the Extended GMTKN30 Database for General Chem Soc 2009, 131:14156–14157.
Main Group Thermochemistry, Kinetics, and Non- 122. Grimme S, Djukic JP. The Crucial Role of Dispersion
convalent Interactions. J Chem Theor Comput 2011, of Nonbridged Binuclear Os?Cr and Os?W Adducts.
7:291–309. Inorg Chem 2010, 49:2911–2919.

228 
c 2011 John Wiley & Sons, Ltd. Volume 1, March/April 2011