FULL PAPER

DOI: 10.1002/chem.201200497

Supramolecular Binding Thermodynamics by Dispersion-Corrected Density

Functional Theory

Stefan Grimme*[a]

Abstract: The equilibrium association
free enthalpies DGa for typical supra-
molecular complexes in solution are
calculated by ab initio quantum chemi-
cal methods. Ten neutral and three pos-
itively charged complexes with experi-
mental DGa values in the range 0 to
21 kcal mol1 (on average 6 kcal
mol1) are investigated. The theoretical
approach employs a (nondynamic)
single-structure model, but computes
the various energy terms accurately
without any special empirical adjust-
ments. Dispersion corrected density
functional theory (DFT-D3) with ex-
tended basis sets (triple-z and quadru-
ple-z quality) is used to determine
structures and gas-phase interaction en-
ergies (DE), the COSMO-RS continu-
um solvation model (based on DFT
data) provides solvation free enthalpies
and the remaining ro-vibrational en-
thalpic/entropic contributions are ob-
tained from harmonic frequency calcu-
lations. Low-lying vibrational modes
are treated by a free-rotor approxima-
tion. The accurate account of London
dispersion interactions is mandatory
with contributions in the range 5 to
60 kcal mol1 (up to 200 % of DE).
Inclusion of three-body dispersion ef-
fects improves the results considerably.
Keywords: density functional calcu-
lations dispersion corrections
noncovalent interactions theoreti-
cal chemistry thermodynamics
A semilocal (TPSS) and a hybrid densi-
ty functional (PW6B95) have been
tested. Although the DGa values result
as a sum of individually large terms
with opposite sign (DE vs. solvation
and entropy change), the approach
provides unprecedented accuracy for
DGa values with errors of only 2 kcal
mol1 on average. Relative affinities
for different guests inside the same
host are always obtained correctly. The
procedure is suggested as a predictive
tool in supramolecular chemistry and
can be applied routinely to semirigid
systems with 300400 atoms. The vari-
ous contributions to binding and en-
thalpyentropy compensations are dis-
cussed.

Introduction on the binding mechanisms prevails.[6, 7] One of the reasons

The noncovalent interactions between atoms and molecules
play an important role in structural biology and supramolec-
ular chemistry. They control the structures of proteins and
DNA, hostguest systems, enzymesubstrate binding, anti-
genantibody recognition, or the orientation of molecules
on surfaces or in molecular films.[1, 2] Because of their ubiqui-
tous role in diverse fields, the investigation and understand-
ing of these weak interactions has become one of the major
goals of modern chemistry. Noncovalent interactions[35]
form the basis of what is usually called a supramolecular
interaction. However, quantitative description of its ther-
modynamics by means of ab initio electronic structure
theory is not well developed and a phenomenological view
[a] Prof. Dr. S. Grimme
Mulliken Center for Theoretical Chemistry
Institut fr Physikalische und Theoretische Chemie
Universitt Bonn, Beringstr. 4, 53115 Bonn (Germany)
E-mail: grimme@thch.uni-bonn.de
Supporting information for this article is available on the WWW
under http://dx.doi.org/10.1002/chem.201200497. It contains Cartesian
coordinates of all molecules and complexes, all thermodynamic data
and a detailed description of the SCC-DFTB-D3H and PM6-D3H
methods used for semi-empirical frequency calculations.
for this is that already the smallest experimentally consid-
ered systems often consist of a hundred or more atoms and
that solvation as well as entropic effects play an important
role in the binding. These three factors contribute to large
and often unaffordable computational costs.
It is shown in this work for the first time in a systematic
study, that the free enthalpy of association of a host and
a guest molecule to form a supramolecular complex in solu-
tion (DGa, equilibrium association constant Ka) can be com-
puted with reasonable accuracy from well-established, stan-
dard quantum chemical methods without any special adjust-
ments. This has some perspective for an ab initio computa-
tion of absolute proteinligandACHTUNGRE(drug) binding constants. In
a pilot study on proteinligand model complexes, including
also the use of accurate wave-function-based methods,[8] we
have already shown that quantum chemical methods based
on London dispersion[9, 10] corrected density functional
theory (DFT-D3[11]) provide very accurate interaction ener-
gies for large complexes (for recent benchmarks on smaller
complexes see references [12, 13]). The typical DFT-D3
error for the basic gas-phase interaction energy DE has
been estimated to be < 5 % in several studies,[8, 13, 14] which
would enable an accurate prediction of total DGa values
within 12 kcal mol1 if the other contributions (solvation
and entropy change[15]) could be obtained within similar

Chem. Eur. J. 2012, 00, 0 0  2012 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim &1&
These are not the final page numbers!
error limits. That this is indeed possible will be shown
herein. The major advantage of such an ab initio treatment
of supramolecular interactions as opposed to empirical
force-field methods (for two examples see referen-
ces [16, 17]) is that systems of any composition including
transition metals can be studied and that an unbiased pic-
ture and understanding of the fundamental physical interac-
tion mechanisms can be obtained. For a critical evaluation
of force-field methods for noncovalent interactions see ref-
erence [18].
We have investigated the following systems (see Figure 1):
a benzene dimer (1), two tweezer complexes with tetra-
cyanoquinone (TCNQ) and 1,4-dicyanobenzene (2 a and 2 b
in CHCl3) investigated in the group of Klrner,[19] two
pincer complexes of organic p systems (3 a and 3 b with
guests denoted as 3 c and 3 d in CH2Cl2) that have been in-
vestigated recently also theoretically,[20] fullerenes C60 and
C70 in a so-called buckycatcher (4 a and 4 b in toluene)
from reference [21], an amide macrocycle (mcycle) that
binds benzoquinone (BQ) and glycine anhydride (GLH)
from the work of Hunter et al.[22] (5 a and 5 b in CHCl3) and
the cucurbit[6]uril (CB6) cation complexes of butylammoni-
um (BuNH4) and propylammonium (PrNH4), respectively,
from the work of Mock and Shih[23] (6 a and 6 b in an 1:1
mixture of formic acid and water). The cucurbit[7]uril
(CB7) host forms an extremely stable inclusion complex 7 a
with the dicationic ferrocene derivative bis(trimethylammo-
niomethyl)ferrocene (FECP) with an equilibrium associa-
tion constant of 3  1015 m1, which represents some kind of
world record for synthetic systems in water.[17] The corre-
sponding neutral complex 7 b with 1-hydroxyadamantane
(ADOH) as guest in water from the same work is investigat-
ed here for comparison.
Figure 1. Gas-phase structure of investigated complexes.
&2& www.chemeurj.org  2012 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
These are not the final page numbers!
These systems cover a broad range of typical supramolec-
ular interactions ranging from nonpolar-dispersion-dominat-
ed ones for complexes 1 and 2 and p-stacked structures in 3
and 4 to hydrogen bonding in 5 and 7 b and cationdipolar
binding modes in 6 and 7 a. For each of the six hosts, two
complexes with different guest molecules are investigated.
This should shed light on the ability of the approach to cal-
culate accurate relative affinities, which is in some applica-
tions even more important than the absolute magnitude of
DGa. The benzene dimer in water 1[24] is included as a small
system for which accurate gas-phase interaction energies are
known theoretically[2528] and to demonstrate that the ap-
proach is completely general and works for small as well as
extended complexes. It also represents a limiting case with
a very small (near zero) binding affinity.
Experimentally observed DGa values for supramolecular
complexes consisting of a few hundred atoms are typically
in the range 1 to 10 kcal mol1.[2] Complexes 7 a and 7 b
with values of 21 and 14 kcal mol1, respectively, repre-
sent extremes. In any case this is almost an order of magni-
tude less than the computed DE values for complexes of this
size.[8, 21, 20] This indicates that solvent and entropic correc-
tions are large and positive and are effectively quenching
the many, in total rather strong intermolecular, noncovalent
interactions as noted already many years ago.[15] This situa-
tion in which a small quantity results from addition of sever-
al large terms with opposite signs is the worst-case scenario
for any nonempirical approach. Systematic error cancella-
tion cannot be expected and each term individually must be
computed rather accurately to obtain reliable DGa values.
With this in mind one could consider this project successful
already if the order of magnitude and affinity trends for dif-
ferent complexes are correct. How difficult the problem is
Chem. Eur. J. 0000, 00, 0 0
Supramolecular Binding Thermodynamics
has been exemplified in a DFT study for the binding of
a neutral guest inside a charged [2]pseudorotaxane with ex-
perimental values of DHa = 4.9 kcal mol1 and DGa =
1.6 kcal mol1.[29] Very strongly deviating theoretical DHa
values between 28 and + 23 kcal mol1 (in MeCN using
a continuum solvation model, with and without counterions)
have been computed using various (dispersion-uncorrected)
density functional methods.[29] Before we show that one can
significantly improve the accuracy, the general theoretical
approach is outlined highlighting a few theoretical aspects
which require special attention.
Computational Methods
General approach to free enthalpies of association in solution: The free
enthalpy of association of molecules A and B (e.g., host and guest) to
form a noncovalently bound complex C at temperature T in solvent X is
given by Equation (1) in which DE is the gas-phase association energy as
calculated in the super-molecule approach DE = E(C)E(A)E(B) (E
being the total electronic energies), GTRRHO is the sum of corrections from
energy to free enthalpy in the rigid-rotor-harmonic-oscillator approxima-
tion (RRHO) also including zero-point-vibrational energy for each spe-
cies in the gas phase, and dGTsolv(X) is their corresponding free solvation
enthalpy.
DGa DE DGTRRHO DdGTsolv X 1 previous quantum chemical studies on molecular tweezer or clip
In the following we use the short notation DGa (and similar for DHa and
the other terms, i.e., drop the T and X indices), but note that these
values refer to specific solvents and temperature for each of the systems.
The prefix D refers to differences regarding the reaction A + B!C, while
d indicates that the quantity implicitly represents a difference between
thermodynamic states (i.e., gassolution).
The last term in Equation (1) is taken unmodified from the COSMO-RS
continuum solvation model,[3032] which is based on DFT (BP86,[33]/
TZVP[34]) single-point energy computations for the electrostatic solute
solvent interaction energy (as expressed by the screening-charges through
the so-called s-profiles[30]) and including further molecular surface pro-
portional corrections to directly yield free solvation energies. The corre-
sponding enthalpies dHsolv are computed from the temperature depend-
ence of the free enthalpies, but are used here merely for analysis purpos-
es and do not represent fundamental quantities. The dGsolv terms ob-
tained from COSMO-RS refer to the 1 mol gas/1 mol solvent frame and
can directly be used in Equation (1) to compute standard state data.
Note that very strongly and specifically coordinating solvent molecules
can be included in a quantum chemical supermolecular system and
should not be treated by COSMO-RS (which, however, is not considered
here). The COSMO-RS model normally provides accurate dGsolv values
with an estimated error of about 510 % (< 0.51 kcal mol1 for neutral
systems[35, 36]). Because the error for the two other terms in Equation (1)
is of the same order of magnitude (or even somewhat larger) and be-
cause other means to account for solvation effects quantitatively are still
not routine, we take this continuum model as is and concentrate on an
accurate evaluation of the DE and DGRRHO terms. A further advantage of
COSMO-RS is that solvent mixtures can be treated on equal footing. For
a thorough discussion of the current status and performance of this and
related solvation models see also Cramer and Truhlar.[3739] The depend-
ency of the performance of COSMO-RS on the underlying DFT method
(here: B86/TZVP//TPSS-D3/def2-TZVP) has recently been investigated
in detail,[40] but is found to be very weak.
The gas-phase complexation energy DE is expected to be the dominant
contribution in Equation (1); for a further discussion of this point see
below. It is very specific for the mode of interaction and its accurate com-
putation is mandatory in order to obtain reliable results. For large sys-
Chem. Eur. J. 2012, 00, 0 0  2012 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
FULL PAPER
tems it can be calculated now routinely by well-established dispersion-
corrected DFT (atom pair-wise DFT-D3[11, 41] or density-dependent DFT-
nonlocal[42, 43] (NL) variants). Only DFT-D3 is considered in this work
(for a recent overview of various other dispersion-corrected DFT meth-
ods[4448] see reference [49]). We have tested two different density func-
tionals (the TPSS meta-GGA[50] and the PW6B95 meta-hybrid[51]) to
obtain an estimate for the error of DE. The DGRRHO contribution is com-
puted by lower-level quantum chemical methods (SCC-DFTB[52, 53] and
PM6[54, 55]) in order to make the entire approach routinely applicable to
systems with about 300400 atoms (although this is not a prerequisite
and these semi-empirical MO methods could be avoided entirely).
No special adjustments for the systems considered here are made to the
published procedures that are employed to compute the individual parts
in Equation (1). It should be stressed that the theoretical approach fol-
lowed is ab initio in a sense that all physically relevant terms are consid-
ered and computed as accurately as possible, but are not adjusted empiri-
cally for the present purpose. However, the methods themselves necessar-
ily contain a few empirical parameters that are essential to treat large
systems (several hundreds of atoms) with reasonable accuracy. A similar
approach as used here has been proposed very recently by Jacquemin
and co-workers[20] and has been applied to 3 a and 3 b. These authors also
employ DFT-D for the interaction energy, but use the PCM solvation
model[35] and a relatively small AO basis set for structure optimization.
Three-body dispersion effects as well as the low-frequency mode problem
(see below) were not considered. However, good agreement with the ex-
perimental DGa values (errors of only 13 kcal mol1) has been obtained.
The catcher complex 4 a was investigated previously by Zhao and Truhlar
using DFT without explicit dispersion corrections.[56] Beside DE and
DGRRHO values, only estimates for the solvation effects were given. For
complexes see, for example, references [5761].
Equation (1) is based on the fundamental assumption that the compo-
nents in the association reaction are distinct chemical species with a well-
defined molecular structure. It would easily be possible to apply the ap-
proach for a few well-defined conformers (the corresponding quantities
have to be thermodynamically averaged), but extremely flexible, highly
dynamic systems cannot be treated in this way (for the thermodynamic
treatment of very flexible, but smaller systems see reference [62]). Force-
field studies by Gilson and co-workers[17, 63] for similar complexes have
shown that only the few lowest-lying conformers contribute and that the
herein pursued nondynamic, single-structure approach may introduce
errors of merely about 1 kcal mol1. Note, in this context, that the present
study represents a proof-of-principle and that thermodynamic averaging
over a few conformers is possible in principle also at an ab initio level.
The focus here is clearly on the basic intermolecular contacts and the
electronic structure aspect, which is complementary to more dynamic
treatments.[17, 63, 64]
The examples have been taken either from our previous work[21, 61] or are
more or less randomly chosen from a standard textbooks or reviews on
supramolecular chemistry.[2, 7] No biased selection of complexes for which
the methods work well has been attempted. The only prerequisite was
a certain molecular rigidity in addition to the coverage of different bind-
ing types and a wide range of affinities.
Computation of (gas-phase) interaction energies: The species A, B, and
C were fully optimized in the gas phase at the TPSS-D3/def2-TZVP level
and these structures were also used for the COSMO-RS treatment. Scans
for low-lying conformers and pre-optimizations were performed at the
semi-empirical level, but always only the lowest energy structure found
was further considered for DFT-D3. The semilocal TPSS[50] functional
represents a robust, well-tested DFT method which provides good molec-
ular structures[41] at low computational cost. For some years it has been
the default functional for structure optimizations in our group (for the
most recent example see reference [65]).
The molecular electronic energies E were computed by dispersion-cor-
rected DFT given by Equation (2), in which EDFT is the (all-electron) KS-
DFT SCF energy for a particular density functional, E2 disp is the standard
atom pair-wise London dispersion energy from D3 theory[11] (using
BeckeJohnson damping[41, 47, 48]), and E3
disp is a three-body dispersion term
www.chemeurj.org &3&
These are not the final page numbers!

(of AxilrodTellerMutto type[66, 67]), which was calculated as described in
reference [11] using our freely available program dftd3.[68]
E EDFT E2 3
disp Edisp 2
Note that this last term is independent of the chosen density functional
in our approach. These three-body corrections to the dispersion interac-
tion of small complexes (< 1020 atoms) are negligible for, for example,
the S66 benchmark set[69] (average E3 1
disp corrections are 0.06 kcal mol
1
for an average total interaction energy of 6 kcal mol ; similar results
have been reported in references [70, 71]). However, as already noted in
reference [11], three-body dispersion effects become more important in
larger and dense systems. As will be shown below for the first time, its
account improves computed supramolecular interaction energies consid-
erably (the more commonly used three-body uncorrected energy E =
EDFT + E2
disp is also discussed). Furthermore, for a consistent description of
small and concomitantly large complexes it is essential to describe the
asymptotic 1/R6 dependence of the interaction energy (in which R is
the intermolecular distance) correctly also in the large distance regime,
which is not possible with for example, uncorrected, (semilocal) meta-
hybrid functionals like M06-2X[72] (for more details see reference [13]).
The relatively large systems considered here should be used in the future
as benchmarks to investigate this point also for other dispersion correc-
tions in more detail.
The dependence of the results for DGa on the chosen density functional
stems from the DE part. We investigate this by single-point calculations
with the PW6B95[51] meta-hybrid functional developed by Zhao and
Truhlar, which has proven to be very accurate for a wide range of appli-
cations when dispersion corrections are properly taken into account.[13, 73]
In all geometry optimizations we apply the large def2-TZVP AO basis
set,[34, 74] which yields small basis set superposition errors (BSSE) on the
order of 5 % of DE for larger complexes as considered here. Computa-
tionally demanding counterpoise corrections are hence not applied and
we prefer to investigate the basis set dependence with the huge def2-
QZVP basis set (def2-QZVP[74, 75] with discarded g- and f-functions on
non-hydrogen and hydrogen atoms, respectively; this is denoted in short
QZ in the following). In one case the full def2-QZVP and cc-pVQZ[76]
basis has been applied for convergence checking purposes and differen-
ces of only < 0.2 kcal mol1 have been observed. Remaining (tiny) BSSE
effects have been absorbed into the short-range D3 parameters as deter-
mined in original work[11, 41] at a similar basis set level. Note that the QZ
basis already contains semidiffuse AO basis functions that are known to
be sufficient to describe molecular polarizeabilities at the GGA level.[11]
Diffuse functions from reference [77] (e.g., def2-TZVPD) have been
tested, but often lead to serious SCF convergence problems due to near-
linear dependencies in the basis and are generally not recommended
(and not necessary for simulations of large systems in solution).
The largest SCF calculation performed here consists of 8250 contracted
Gaussian AO basis functions, but can be computed nowadays routinely
on conventional PC clusters. Note that this technical level is much more
sophisticated than in many previous applications for systems of similar
size (e.g., references [16, 56, 78]) that employ only double-z basis sets. For
a careful basis set study on 3 see reference [20].
All DFT calculations were performed with TURBOMOLE.[79] The RI
approximation[80] for the Coulomb integrals is always applied, which
speeds up the computations (in particular for TPSS) significantly without
any significant loss of accuracy. Matching default auxiliary basis sets are
used.[81] The numerical quadrature grid in TURBOMOLE denoted as m4
(m5 for the final single-point energies) has been employed for the inte-
gration of the exchange-correlation contributions. The computations of
the harmonic vibrational frequencies at the DFT-D3 level for comparison
were performed seminumerically in parallel using the TURBOMOLE
script NumForce.
Computation of internal (gas-phase) entropies: Low-lying vibrational fre-
quencies are notoriously inaccurate in the harmonic approximation and
numerical noise in typical quantum chemical calculations causes further
errors. This will be considered in the following and a completely black-
box type procedure is proposed to remedy the problem. For previous ap-
&4& www.chemeurj.org  2012 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
These are not the final page numbers!
S. Grimme
proaches that require use of internal coordinates (and that have never
been extended to large systems) see reference [82]. A related but compu-
tationally more involved alternative is the harmonic mode scanning
method, see reference [64].
The vibrational entropy of an harmonic oscillator with frequency w is
given by Equation (3) in which k is Boltzmanns constant, h is Plancks
constant, T is the absolute temperature and R is the usual gas constant.
 
hw
SV R  ln1  ehw=kT 3
ke hw=kT  1
Because the first term on the right-hand side of Equation (3) asymptoti-
cally approaches infinity for w!0 we replace the contribution of low-
lying modes to the entropy by a corresponding rotational entropy (quasi-
RRHO approach). For any normal mode we compute the moment of in-
ertia m for a free-rotor with the same frequency [Eq. (4)].
h
m 4
8p2 w
In order to define an effective moment of inertia and to restrict it to
a reasonable value, we introduce the average molecular moment of iner-
tia Bav as a limiting value for very small w (large m) by taking the follow-
ing reduced value from Equation (5).
mBav
m0 5
m Bav
The entropy for a low-frequency mode in this approximation is then
given by Equation (6).
" ( 1=2 )#
8p3 m0 kT
SR R 1=2 ln 6
h2
In order to continuously interpolate between the rotational and harmonic
vibrational approximations, SV and SR are combined by using a weighting
function of w as given by Equation (7) in which w(w) is given by the
Head-Gordon damping function [Eq. (8)],[83] with a = 4 (very similar re-
sults have been obtained by a Fermi-weighting function that has also
been tested).
S wwSv 1  wwSR 7
1
ww 8
1 w0 =wa
This interpolates between the harmonic vibrational entropy for w @ w0
and a pure (and always finite) rotational entropy for small w. We chose
a value of w0 = 100 cm1 as a default (about 1/2kT at room temperature).
Equation (7) then effectively replaces the vibrational entropy for all
modes with frequencies < 100 cm1 by a corresponding free-rotor entro-
py. A similar cut-off value has been used by Truhlar and co-workers re-
cently.[56, 84] In the Supporting Information it is shown that the computed
entropies are only weakly dependent on the value of w0 if chosen within
reasonable limits (50150 cm1). Figure 2 shows the entropy for one
mode as a function of w at T = 298 K and for a typical value of Bav =
1044 kg m2. As can be seen the average entropy given by Equation (7) re-
mains reasonable even for very low frequencies of < 23 cm1.
This allows us to use computed entropies (free enthalpies) in a black-
box fashion, even in cases when considerable numerical noise in the fre-
quency calculations cannot be avoided for technical reasons. As will be
discussed further below, the differences between the results from Equa-
tions (3) and (7) for bimolecular reaction (association) entropies for mol-
ecules with only about 100 atoms already is often 12 kcal mol1. Test cal-
culations on model systems with 300400 atoms have shown that the
errors from Equation (3) can easily reach 34 kcal mol1, because very
small w values on the order of 510 cm1 are quite common in such sys-
tems. Note that for mid-sized molecules without low-frequency modes
(for which experimental data would be available) the difference to the
Chem. Eur. J. 0000, 00, 0 0
Supramolecular Binding Thermodynamics
FULL PAPER
frequencies for DFTB-D3H and PM3-D3H methods has been considered
but optimum values (in comparison to DFT-D3/def2-TZVP computed
free enthalpies at 298 K for a set of 14 organic molecules) are close to
unity and hence frequencies are finally taken unscaled. Change of a fre-
quency scale factor by  5 % typically changes computed Ga values by
less than  0.2 kcal mol1.

Results

Gas-phase interaction energies and RRHO free enthalpy

contributions: The rotational part of the entropy depends
on the molecular symmetry of the host, guest, and the com-
plex. For highly symmetric molecules the effect of the sym-
metry number s is significant (12 kcal mol1) and requires
Figure 2. Computed entropies for one vibrational mode as a function of some consideration. For a thorough discussion of this issue
w for T = 298.15 K, Bav = 1044 kg m2, and w0 = 100 cm1. see reference [86]. We mention only cases in which symme-
try plays a role. The cucurbit[6]uril host has D6d symmetry,
standard treatment is negligible so that direct tests of the quality of the
while both hostguest complexes have only CS symmetry.
approximation cannot be made. However, as shown below, the better The cucurbit[7]uril is only close to D7d or D7h symmetry and
agreement of computed and experimental DGa values for some of the hence s = 1 is taken (the effect on DGRRHO is 1.3 kcal mol1).
here investigated systems strongly supports the reliability of the above The pincer host and glycineanhydride have C2 symmetry.
approach. The tweezer host is C2v-symmetric and benzoquinone be-
Computation of harmonic vibrational frequencies: The zero-point vibra- longs to the D2h group. The fullerenes C60 and C70 are treat-
tional energy, translational and rotational contributions to enthalpy and
free enthalpy were calculated in the usual rigid-rotor-harmonic-oscillator
ed in Ih and D5h groups, respectively, and the C60 complex
approximation (RRHO) using the equilibrium structure and seminumeri- 4 a is C2v-symmetric. The benzene dimer is close to C2h sym-
cally computed vibrational frequencies. Their computation with DFT-D3 metry, but the true minimum is found, in agreement with
is without any theoretical problems possible, but requires a huge amount previous studies, to have no symmetry.[45] Because the corre-
of computational resources for systems with several hundreds of atoms. sponding hypersurfaces are extremely flat the complex can
In fact the DFT-D3/large basis set frequency calculations would repre-
sent by far the computational bottleneck of the entire procedure. We
be considered as having effectively symmetry. Nevertheless,
hence propose to compute those by lower-level SCF methods that are, we consistently use s = 1, but expect a relatively large inher-
however, dispersion-corrected and hence can treat weakly bound and ent error of the DGRRHO term in this case.
supramolecular systems at very modest computational cost. In a few The data for DFT-D3 interaction energies, two- and
cases we compare the results from these more approximate methods to three-body dispersion contributions and RRHO free enthal-
full DFT-D3 frequency calculations. In all cases the rotational constants
computed with DFT-D3 structures have been employed for the rotational
py contributions are given in Table 1. We compare DGRRHO
part of the partition functions. values from PM6-D3H and DFTB-D3H methods in selected
We have tested D3-dispersion-correct-
ed semi-empirical methods PM6[54, 55]
and SCC-DFTB,[52] which are known Table 1. Computed gas-phase interaction energies DE (i.e., not including three-body terms), two-body DE2 disp
to be reasonably accurate for common (TPSS short-range damping parameters) and three-body DE3 disp dispersion contributions to the interaction
noncovalent interactions.[85] However, energy, and gas-phase RRHO free enthalpy contributions. For PM6 the values without the low-frequency-free-
these methods do not perform well for rotor approximation are given in parenthesis. All methods are D3-dispersion-corrected, but the D3-label has
hydrogen-bonded systems, which re- been omitted for clarity. All values in kcal mol1.
quire further, special corrections [see
Complex DE2
disp DE3
disp DE DGRRHO
the Supporting Information for a de-
TPSS/TZ TPSS/QZ PW6B95/QZ PM6 DFTB
tailed description of the corrections
that are used here and that are denot- 1 benzene dimer 5.1 0.1 3.0 2.9 3.0 7.4 (3.1) 8.8
ed D3H (zero-damping D3-disper- 2a TCNA@tweezer 38.2 1.8 33.5 33.0 32.1 16.7 (15.5) 16.5
sion correction plus hydrogen-bond 2b DCB@tweezer 27.7 1.2 22.2 21.8 21.7 14.7 (12.8) 15.0
correction)]. DFTB has sometimes 3a 3c@pincer 31.3 1.8 26.8 25.8 26.2 17.1 (16.6) 17.6[a]
problems with very polar structures 3b 3d@pincer 16.8 0.7 20.6 19.9 21.1 16.4 (15.9) [b]
(e.g. 6 or 7 b) and PM6 is known to 4a C60@catcher 55.1 3.2 36.9 36.5[c] 35.0 16.1 (13.7) 16.5[d]
yield too short alkylalkyl contacts.[85] 4b C70@catcher 58.6 3.5 38.6 38.1 36.1 16.9 (14.4) 17.7
The ferrocene part in 7 a is somewhat 5a GLH@mcyle 23.7 1.0 34.2 32.9 32.9 17.3 (17.7) 17.7[e]
distorted by PM6 and represents a bor- 5b BQ@mcyle 22.1 1.0 23.1 22.1 21.4 16.0 (16.3) 17.2
derline case for its applicability. After 6a BuNH4@CB6 25.1 2.2 84.2 83.2 84.4 16.0 (16.9) 17.6
some testing it was concluded that 6b PrNH4@CB6 21.0 1.8 80.7 79.8 80.7 15.1 (15.6) 18.1
semi-empirically computed frequencies 7a FECP@CB7 52.9 4.6 132.8 130.3 134.2 23.4 (26.5) [b]
could be used but other options (e.g., 7b ADOH@CB7 35.5 3.3 28.5 27.3 29.0 17.4 (16.5) 18.6
DFT-D3/very small basis set) should [a] The TPSS-D3 value is 16.8 kcal mol1. [b] Not computed due to missing SCC-DFTB parametrization for
be tested in the future. Applying scale some atoms. [c] 36.7 kcal mol1 using the cc-pVQZ basis. [d] The TPSS-D3 value is 16.8 kcal mol1. [e] The
factors for the harmonic vibrational TPSS-D3 value is 17.4 kcal mol1.

Chem. Eur. J. 2012, 00, 0 0  2012 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim www.chemeurj.org &5&
These are not the final page numbers!
 S. Grimme

cases with those from a full TPSS-D3 frequency calculation. (1). In all these cases substantially better agreement be-
Table 1 also contains data in the standard formalism tween the theoretical and experimental DGa values is ob-
[Eq. (3)]. The DE values are given for the semilocal TPSS tained when the low-frequency vibrations are treated in the
and PW6B95 hybrid meta-GGA functionals and for TPSS free-rotor approximation as proposed above.
with def2-TZVP (TZ) and def2-QZVPACHTUNGRE(-g,-f) (QZ) AO basis
sets. Total association free enthalpies and comparison with exper-
Interaction energies at the EDFT + E2 disp level range from imental data: In Table 2 the total gas-phase interaction ener-
3 kcal mol1 for the benzene dimer to large values of 20 gies including three-body terms with the best functional
to 30 kcal mol1 for the other neutral complexes.
The largest values of about 80 and 130 kcal
mol1 are computed for the cationic systems 6 and
Table 2. Best estimate total gas-phase interaction energies (PW6B95-D3/QZ//TPSS-
7 a, respectively. The two-body dispersion correc- D3/TZ including three-body dispersion), solvent free enthalpy contributions
tions DE2
disp to the raw DFT interaction energy are (COSMO-RS), total enthalpies, entropies and free enthalpies of association, and cor-
huge, often on the order of about 100 % of DE and responding experimental values. If not stated otherwise the data refer to standard
for the catcher complexes even almost 200 %. Al- state conditions in the solvents mentioned in the text.
though DE2 disp is a model (density functional)-depen-
Complex DE DdGsolv DHa,calcd TDSa,calcd DGa,calcd DGa,exptl
[a]
dent quantity and only the total DE represents the 1 benzene dimer 2.8 5.8 0.7 0.6 1.3 0.1
basic term, this very clearly shows that an accurate 2 a TCNA@tweezer 30.3 9.0 8.7 4.1 4.6 4.2
2 b DCB@tweezer 20.5 4.4 6.4 4.9 1.5 1.4
account of the dispersion energy is mandatory for
3 a 3c@pincer 24.4 5.4 7.8 6.0 1.8 2.3
obtaining overall reliable supramolecular interac- 3 b 3d@pincer 20.4 3.0 8.3 7.2 1.0 1.3
tions. 4 a C60@catcher[b] 31.8 6.5 14.1 4.9 9.2 5.3
The three-body dispersion term DE3 is always 4 b C70@catcher[b] 32.6 7.3 13.7 5.2 8.4 5.1
disp
31.9 5.4 5.5 8.3
repulsive and roughly an order of magnitude small- 5 a GLH@mcyle 9.2 0.1
5 b BQ@mcyle 20.4 3.9 5.4 4.8 0.6 3.3
er. However, the contributions are mostly 13 kcal 6 a BuNH @CB6 82.2 54.5 6.8 4.9 11.7 6.9
4
1
mol , which is very significant in relation to the 6 b PrNH4@CB6 78.9 56.2 0.2 7.3 7.5 5.7
magnitude of the DGa target quantity which is typi- 7 a FECP@CB7 129.6 86.9 8.5 10.7 19.2 21.1
cally in the range 0 to 8 kcal mol1. In the case of 7 b ADOH@CB7 25.7 8.9 21.3 4.1 17.2 14.1
3 a and 3 b even the affinity difference for the same [a] At 303 K. [b] At 293 K.
host is influenced by 1.1 kcal mol1 due to three-
body effects. The complexes with the biggest DE3 disp
terms are the large 4 and 7 and with an exceptionally high (PW6B95) and the largest basis set are given together with
contribution of 4.6 kcal mol1 for the iron-containing 7 a. solvation enthalpies, free enthalpies, and experimental data.
Basis set effects (i.e., difference between TZ and QZ A comparison of computed and experimental DGa values is
values) as estimated at the TPSS level are weak in most also displayed graphically in Figure 3. In this plot we include
cases (< 1 kcal mol1), which gives confidence that the inter- estimated theoretical error bars, which have been obtained
action energies with the largest QZ basis are sufficiently by adding the following three main estimates (absolute
converged for the present purpose. In addition, the differen- values): 1) the difference between PM6 and DFTB comput-
ces between the two density functionals (using the same ge- ed RRHO corrections (if available), 2) the difference be-
ometries) are small, mostly less than 1 kcal mol1 (at most
3.8 kcal mol1 corresponding to 3 % for the most strongly
bound 7 a), which is very encouraging. Calculations using
lower-level optimized structures reveal insignificant error
due to the single-point approximation (i.e., that TPSS-D3/
TZ structures are used for PW6B95/QZ).
Finally, we shall discuss methodological issues regarding
the calculation of the DGRRHO values. These contributions
are always positive (complex destabilizing) and show moder-
ate variation (with the exception of 1 due to its small size
and of 7 a) for the different complexes. In the three cases
for which we also computed TPSS-D3/TZ frequencies (un-
scaled), good mutual agreement between the semi-empirical
results within 0.5 kcal mol1 is observed. However, a notable
impact of the proposed free-rotor approximation for the
low-frequency vibrations is found. The difference to the
standard treatment (at the PM6-D3H level) can reach size-
able values of 22.5 kcal mol1 (for 2 a and 4 b), 3.1 kcal Figure 3. Comparison of experimental and computed total free associa-
mol1 for 7 a, and even 4.3 kcal mol1 for the benzene dimer tion enthalpies.

&6& www.chemeurj.org  2012 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim Chem. Eur. J. 0000, 00, 0 0

These are not the final page numbers!

Supramolecular Binding Thermodynamics
tween TPSS/QZ and PW6B95/QZ results, and 3) 3 % of the
DdGsolv term.
Before discussing the values in some detail it should be
noted that the overall agreement between theory and exper-
imental data can be considered as exceptionally good. For 2
and 3 the theoretical and experimental data almost exactly
match and for 5 b and 6 b the deviations are only about
1 kcal mol1. Neglecting the three-body-dispersion energy
would cause systematic overbinding by about 2 kcal mol1.
Overall we obtain a respectable mean absolute deviation
(MAD) and mean (signed) deviation (MD) of 2.1 and
0.8 kcal mol1, respectively (3.4 and 3.1 kcal mol1 with-
out DE3
disp). However, these statistical data should be taken
with some care because of the small number of comparisons
(13) and a few outliers that strongly influence these statis-
tics.
Beside this very good agreement of the absolute values
for complexes of very different structure we furthermore
note a good relative accuracy. The DDGa values for the two
guests in the same host are always predicted qualitatively
right and often also quantitative agreement with the experi-
mental difference is obtained. As already mentioned, the
benzene dimer has a very flat potential energy surface for
which the present approach is limited. Nevertheless by using
the free-rotor approximation for the low-frequency modes,
the error of about 1.4 kcal mol1 for DGa in water (0.5 kcal
mol1 using DFTB-D3H frequencies) seems acceptable. The
here proposed approach also yields much better affinities
for 3 (1.8 and 1.0 kcal mol1 compared to 2.3 and
1.3 kcal mol1 experimentally) than reported in refer-
ence [20]. Somewhat surprisingly very reasonable results are
also obtained for the cationic cucurbituril complexes for
which the cancellation between large gas-phase DE values
and the solvation contribution is very pronounced. This in
particular holds for the dicationic complex 7 a, for which
DdGsolv = 87 kcal mol1, while DE = 130 kcal mol1. These
results indicate the high accuracy and reliability of the
COSMO-RS treatment although this should be verified for
other charged complexes, in particular if multiply charged.
Inspection of Figure 3 shows that there are basically only
three supramolecular systems which might be considered as
outliers. The hydrogen-bonded complex 5 a with a mea-
sured Ka = 106 m1 is computed to be too weakly bound
(about 2.8 kcal mol1 error). The deviation for the similar
complex 5 b is of about the same size, so that some systemat-
ic problems for this system (maybe also in the experiment)
seem to be present. For the overbinding of the complex 6 a
(by 4.8 kcal mol1) there is also no evident reason, as 6 b is
described well and the structural difference between the two
guests (propyl vs. butyl moiety) is small. Qualitatively the
computed difference between 6 a and 6 b also seems reason-
able (better binding of the larger guest in 6 a and more re-
pulsive solvent contribution of the smaller cation guest in
6 b). The error for 6 a is also difficult to understand in light
of the excellent results for the similar cucurbit[7]uril com-
plexes 7. This in particular holds for 7 a, which not only is
a dication with the already mentioned huge compensation
Chem. Eur. J. 2012, 00, 0 0  2012 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
FULL PAPER
between the gas-phase interaction energy and DdGsolv, but
also contains an electronically difficult transition-metal com-
plex.
The other problematic system seems to be the catcher
complexes 4, which are overbound by 34 kcal mol1. Care-
ful searches for conformers with different orientations of
the guest than those reported in reference [21] were success-
ful, but revealed only minor energy changes (about 0.5 kcal
mol1 for DE compared to the minimum structure reported
previously). The computed gas-phase interaction energies of
about 32 kcal mol1 for 4 (including three-body dispersion)
are lower than the plain M06-2X value from reference [56]
(26.4 kcal mol1 for 4 a). This result can be expected be-
cause attractive long-range dispersion is not included in the
M06 functionals. Accounting for these by using D3 (zero-
damping parameters for M06-2X taken from reference [14]),
a value of 35.2 kcal mol1 is obtained, which is in excellent
agreement with our DE(PW6B95-D3) result of 36 kcal
mol1. Hence, because the gas-phase interaction seems to be
rather certain we can only speculate that the treatment of
a nonpolar solute in a nonpolar solvent (toluene) in
COSMO-RS is responsible for the problem. Again as noted
above, the similar affinity of the host for C60 and C70 (with
a slight preference for the former) is nicely reproduced by
the calculations (computed DDGa = 0.8 vs. exptl. 0.2 kcal
mol1).
Thermodynamic analysis and discussion of the binding
mechanisms: Beside the free enthalpy also DHa and TDSa
are observable and have been measured for a variety of
supramolecular complexes.[2, 7] Because the experimental
values sometimes vary considerably also for structurally sim-
ilar complexes and are not available for most of the here
considered compounds, a detailed comparison of these ex-
perimental and theoretical data is beyond the scope of this
work. Nevertheless it should be mentioned that the DHa and
TDSa values given in Table 2 tend to be individually larger
than observed in most experiments meaning that enthalpy
entropy compensation effects appear stronger theoretically.
Similar observations have been made in a force-field study
on cyclodextrin complexes of small molecules[63] (for
a recent discussion see also reference [87]). However,
a basic discussion of this point requires further studies on
specifically selected complexes for which very solid experi-
mental data are available and therefore we will discuss here
only the implications from the present theoretical results.
Figure 4a shows a comparison of computed gas-phase in-
teraction energies and solution free association enthalpies.
This may shed light on the important question of how much
supramolecular interactions are influenced by secondary
effects like solvation and change of entropy.
We notice some correlation between DE and DGa values.
Of course it cannot be used to quantitatively estimate en-
thalpies from energies because the structural dependency of
the other contributions is substantial. Nevertheless, the in-
teraction energy can be considered as the basic driving force
for binding in most cases and should be employed as guiding
www.chemeurj.org &7&
These are not the final page numbers!

Figure 4. Comparison of computed a) gas-phase interaction energies with
free association enthalpies in solution (the solid line shows the result of
a linear regression with a correlation coefficient of 0.79) and b) associa-
tion solution enthalpies with corresponding entropies (multiplied by the
measurement temperature, the regression correlation coefficient is 0.45).
principle when new supramolecular systems are developed.
Note that the DE discussed here does not represent some
model-dependent quantity as in empirical methods, but rep-
resents (within its error limit) a physically well-defined and
fundamental quantity. This is an important advantage of the
present approach. Although strong energy binders can
suffer from large entropy penalty (see below), it is unlikely
that high binding affinities can be obtained without energet-
ically strong interactions. Although this seems to be a some-
what trivial conclusion, it is based here for the first time on
a set of fundamentally correct ab initio data for large sys-
tems.
Figure 4b shows a comparison of computed association
enthalpies and the corresponding entropy penalties TDSa.
Although the plot shows a lot of scatter in the data, a rough
enthalpyentropy compensation can be deduced; that is,
strong binding causes freezing of internal and solvent de-
grees of freedom. Interestingly, for most neutral complexes
(except the very nonpolar systems) the association enthal-
pies DHa fall in a rather narrow range of 6 to 10 kcal
mol1. Hence, the substantial variations in DGa for the neu-
tral complexes are also strongly influenced by entropic (in-
&8& www.chemeurj.org  2012 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
These are not the final page numbers!
S. Grimme
ternal and solvation) factors. In contrast to most other sys-
tems, the cationic complexes have strong stabilizing entropic
contributions. This can be explained by the fact that the
host does not need to be ordered in the complexation (in
contrast to the solvent molecules) and the corresponding
solvent entropy is released. This is in qualitative agreement
with experimental observations for many cationic supra-
molecular complexes.[2, 7] The only other supramolecular
complex with a positive TDSa value is the strongly hydro-
gen-bonded 4 a.
The most striking result of the analysis of the thermody-
namic data is the neutral cucurbit[7]uril complex with 1-hy-
droxyadamantane 7 b that shows a very large affinity of
DGa = 14.1 kcal mol1 (calcd 17.2 kcal mol1), but only
a moderate DE value of 26 kcal mol1. It is clearly recog-
nized in Figure 4a as an outlier. According to the data in
Table 2 and in comparison to, for example, the energetically
similarly bound complex 5 a, favorable solvation contribu-
tions (enthalpic and entropic) seem to be the descisive
factor. Tentatively it can be assigned to the semipolar char-
acter (relatively large hydrocarbon with a small polar OH
side group) of the guest (similar to the case of the benzene
dimer in water). Upon complexation this releases the sol-
ventsolvent interactions in water, but nevertheless, the
guest matches perfectly with the host regarding pure ener-
getics. It can be forseen that valuable insight of this kind
will lead in the future to specifically tailored hostguest sys-
tems.
Conclusion
A nonempirical quantum chemical procedure to compute
free enthalpies of association for supramolecular complexes
in solution (common organic solvents or water at ambient
temperature) is proposed. The approach can be applied in
the present form routinely to complexes of about 300400
atoms and the only prerequisite is a certain conformational
rigidity. In contrast to classical force-field based approaches,
the method can handle almost any type of electronic struc-
ture including charged systems without special adjustments.
The unprecedented accuracy obtained here with errors for
absolute DGa values of typically only about 12 kcal mol1
(average deviation of 2 kcal mol1 with a maximum of 4 kcal
mol1) is rooted in the high accuracy of the DFT-D3 method
to compute gas-phase interaction energies DE. It has been
previously estimated to provide errors for DE of about 5 %
or less, which is consistent with the findings reported here.
In passing it is noted that the accuracy of DFT-D3 (and this
holds for the many other published dispersion corrections)
has never been tested for large molecular systems in com-
parison to experimental data. The very good results ob-
tained here for the first time in such a real-life situation
are encouraging and support the basic ideas in the D3 devel-
opment. However, a new and important outcome of the
present work is that three-body dispersion effects (which are
unimportant in small benchmark systems and that have not
Chem. Eur. J. 0000, 00, 0 0
Supramolecular Binding Thermodynamics
been considered in great detail so far) must be considered
for accurate results. Neglecting those would lead to over-
binding for the majority of the investigated complexes by
typically 13 kcal mol1 (with a maximum value of 4.6 kcal
mol1). The other hitherto overlooked and important effect
in the calculations is the entropy due to the ubiquitous low-
frequency vibrational modes in large molecules. If treated in
the standard harmonic approximation, they can cause unpre-
dictable errors for entropy differences and often introduce
significant noise into the total free enthalpy results. This
problem has been practically solved by applying a simple
free-rotor approximation. The third main conclusion con-
cerns the COSMO-RS solvation model that provides excep-
tionally accurate solvation free energies. From the present
results and analysis it seems as if all three main factors
(DFT-D3, vibrational entropies, solvation) contribute rough-
ly with the same uncertainty to the theoretical results, at
least for neutral complexes. For charged systems the solva-
tion model likely becomes the accuracy determining factor.
Beside the astonishingly good agreement between theory
and experiment for absolute values it should be mentioned
that the relative affinities for different guests are obtained
accurately as well. In this context we want to highlight here
only the compensation effects in the example of cucur-
bit[7]uril, for which the two complexes have similarly huge
experimental affinities of 21 and 14 kcal mol1, respec-
tively (computed to be 19.2 and 17.2 kcal mol1) that
result from totally different contributions, for example, the
DDGsolv terms of 87 and 9 kcal mol1 and the gas-phase in-
teraction energies of 130 and 26 kcal mol1, respectively.
Similar analysis of the different contributions to relative
DGa values in the other cases shows, that DE is the decisive
factor for favorable association although to a varying
degree. In all of the investigated pairs of hostguest com-
plexes a larger (absolute) DE value also leads to a better af-
finity. Enthalpyentropy compensation plays a role, but its
structural and solvent dependence is not fully clear and
should be investigated in more detail further. In any case,
the future for an ab initio based understanding of noncova-
lent interactions in large supra- and biomolecular systems is
bright.
Acknowledgements
The author thanks the team of principal investigators of the SUPRA-
TEC DFG excellence initiative for stimulating discussions that have
triggered the present investigations and D. Jacquemin for providing coor-
dinates of structures 3.
[1] J. M. Lehn, Supramolecular Chemistry, Concepts and Perspectives,
VCH, Weinheim, 1995.
[2] J. L. Atwood, J. W. Steed, Supramolecular Chemistry. Wiley, 2009.
[3] K. S. Kim, P. Tarakeshwar, J. Y. Lee, Chem. Rev. 2000, 100, 4145.
[4] P. Hobza, Wiley Interdiscip. Rev. Comput. Mol. Sci. 2011, 1, 3.
[5] E. G. Hohenstein, C. D. Sherrill, Wiley Interdiscip. Rev. Comput.
Mol. Sci. 2012, 2, 304.
Chem. Eur. J. 2012, 00, 0 0  2012 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
FULL PAPER
[6] E. A. Meyer, R. K. Castellano, F. Diederich, Angew. Chem. 2003,
115, 1244; Angew. Chem. Int. Ed. 2003, 42, 1210.
[7] H.-J. Schneider, Angew. Chem. 2009, 121, 3982; Angew. Chem. Int.
Ed. 2009, 48, 3924.
[8] J. Antony, S. Grimme, D. G. Liakos, F. Neese, J. Phys. Chem. A
2011, 115, 11210.
[9] S. C. Wang, Phys. Z. 1927, 28, 663.
[10] F. London, Z. Phys. 1930, 63, 245.
[11] S. Grimme, J. Antony, S. Ehrlich, H. Krieg, J. Chem. Phys. 2010,
132, 154104.
[12] L. A. Burns, A. Vazquez-Mayagoitia, B. G. Sumpter, C. D. Sherrill,
J. Chem. Phys. 2011, 134, 084107.
[13] L. Goerigk, H. Kruse, S. Grimme, ChemPhysChem 2011, 12, 3421.
[14] L. Goerigk, S. Grimme, Phys. Chem. Chem. Phys. 2011, 13, 6670.
[15] B. Tidor, M. Karplus, J. Mol. Biol. 1994, 238, 405.
[16] K. N. Houk, S. Menzer, S. P. Newton, F. M. Raymo, J. F. Stoddart,
D. J. Williams, J. Am. Chem. Soc. 1999, 121, 1479.
[17] S. Moghaddam, C. Yang, M. Rekharsky, Y. H. Ko, K. Kim, Y. Inoue,
M. K. Gilson, J. Am. Chem. Soc. 2011, 133, 3570.
[18] C. D. Sherrill, B. G. Sumpter, M. O. Sinnokrot, M. S. Marshall, E. G.
Hohenstein, R. C. Walker, I. R. Gould, J. Comput. Chem. 2009, 30,
2187.
[19] M. Kamieth, U. Burkert, P. S. Corbin, S. J. Dell, S. C. Zimmerman,
F.-G. Klrner, Eur. J. Org. Chem. 1999, 2741.
[20] J. Graton, J.-Y. Le Questel, B. Legouin, P. Uriac, P. van de Weghe,
D. Jacquemin, Chem. Phys. Lett. 2012, 522, 11.
[21] C. Mck-Lichtenfeld, S. Grimme, L. Kobryn, A. Sygula, Phys.
Chem. Chem. Phys. 2010, 12, 7091.
[22] C. Allott, H. Adams, P. L. Bernad Jr., C. A. Hunter, C. Rotger, J. A.
Thomas, Chem. Commun. 1998, 2449.
[23] W. L. Mock, N.-Y. Shih, J. Am. Chem. Soc. 1989, 111, 2697.
[24] E. E. Tucker, S. D. Christian, J. Phys. Chem. 1979, 83, 426.
[25] C. D. Sherrill, Rev. Comput. Chem. 2009, 26, 1.
[26] M. O. Sinnokrot, E. F. Valeev, C. D. Sherrill, J. Am. Chem. Soc.
2002, 124, 10887.
[27] P. Jurecka, J. Sponer, J. Cerny, P. Hobza, Phys. Chem. Chem. Phys.
2006, 8, 1985.
[28] T. Takatani, E. G. Hohenstein, M. Malagoli, M. S. Marshall, C. D.
Sherrill, J. Chem. Phys. 2010, 132, 144104.
[29] D. Benitez, E. Tkatchouk, I. Yoon, J. F. Stoddart, W. A. God-
dard III, J. Am. Chem. Soc. 2008, 130, 14928.
[30] A. Klamt, J. Phys. Chem. 1995, 99, 2224.
[31] F. Eckert, A. Klamt, AIChE J. 2002, 48, 369.
[32] F. Eckert, A. Klamt, COSMOtherm, Version C2.1, Release 01.11;
COSMOlogic GmbH & Co. KG, Leverkusen (Germany), 2010.
[33] J. P. Perdew, Phys. Rev. B 1986, 33, 8822.
[34] A. Schfer, C. Huber, R. Ahlrichs, J. Chem. Phys. 1994, 100, 5829.
[35] J. Tomasi, B. Mennucci, R. Cammi, Chem. Rev. 2005, 105, 2999.
[36] A. Klamt, B. Mennucci, J. Tomasi, V. Barone, C. Curutchet, M.
Orozco, F. Luque, Acc. Chem. Res. 2009, 42, 489.
[37] C. J. Cramer, Essentials of Compuational Chemistry, Wiley, New
York, 2002.
[38] C. J. Cramer, D. G. Truhlar, Acc. Chem. Res. 2008, 41, 760.
[39] C. J. Cramer, D. G. Truhlar, Acc. Chem. Res. 2009, 42, 493.
[40] R. Franke, B. Hannebauer, Phys. Chem. Chem. Phys. 2011, 13,
21344.
[41] S. Grimme, S. Ehrlich, L. Goerigk, J. Comput. Chem. 2011, 32, 1456.
[42] O. A. Vydrov, T. Van Voorhis, J. Chem. Phys. 2010, 133, 244103.
[43] W. Hujo, S. Grimme, J. Chem. Theory Comput. 2011, 7, 3866.
[44] M. Dion, H. Rydberg, E. Schrder, D. C. Langreth, B. I. Lundqvist,
Phys. Rev. Lett. 2004, 92, 246401.
[45] J. Grfenstein, D. Cremer, J. Chem. Phys. 2009, 130, 124105.
[46] A. Krishtal, D. Geldorf, K. Vanommeslaeghe, C. VanAlsenoy, P.
Geerlings, J. Chem. Theory Comput. 2012, 8, 125.
[47] A. D. Becke, E. R. Johnson, J. Chem. Phys. 2005, 122, 154104.
[48] E. R. Johnson, A. D. Becke, J. Chem. Phys. 2005, 123, 024101.
[49] S. Grimme, Wiley Interdiscip. Rev. Comput. Mol. Sci. 2011, 1, 211.
[50] J. Tao, J. P. Perdew, V. N. Staroverov, G. E. Scuseria, Phys. Rev. Lett.
2003, 91, 146401.
www.chemeurj.org &9&
These are not the final page numbers!

[51] Y. Zhao, D. G. Truhlar, J. Phys. Chem. A 2005, 109, 5656.
[52] M. Elstner, D. Porezag, G. Jungnickel, J. Elsner, M. Haugk, T.
Frauenheim, S. Suhai, G. Seifert, Phys. Rev. B 1998, 58, 7260 7268.
[53] T. Frauenheim, DFTB + (Density Functional based Tight Binding)
2008, see http://www.dftb.org/.
[54] J. J. P. Stewart, J. Mol. Model. 2007, 13, 1173.
[55] J. J. P. Stewart, Stewart Computational Chemistry, Version 11.038 L,
see http://OpenMOPAC.net.
[56] Y. Zhao, D. G. Truhlar, Phys. Chem. Chem. Phys. 2008, 10, 2813.
[57] F. G. Klrner, J. Panitzky, D. Preda, L. T. Scott, J. Mol. Model. 2000,
6, 318.
[58] F. G. Klrner, B. Kahlert, Acc. Chem. Res. 2003, 36, 919.
[59] C. E. Chang, M. K. Gilson, J. Am. Chem. Soc. 2004, 126, 13156.
[60] S. P. Brown, T. Schaller, U. P. Seelbach, F. Koziol, C. Ochsenfeld,
F. G. Klrner, H. W. Spiess, Angew. Chem. 2001, 113, 740; Angew.
Chem. Int. Ed. 2001, 40, 717.
[61] M. Parac, M. Etinski, S. Grimme, J. Chem. Theory Comput. 2005, 1,
1110.
[62] T. Yu, J. Zheng, D. G. Truhlar, Phys. Chem. Chem. Phys. 2012, 14,
482.
[63] W. Chen, C.-E. Chang, M. K. Gilson, Biophys. J. 2004, 87, 3035.
[64] C.-E. Chang, M. J. Potter, K. Gilson, J. Phys. Chem. B 2003, 107,
1048.
[65] T. Wiegand, H. Eckert, O. Ekkert, R. Frhlich, G. Kehr, G. Erker, S.
Grimme, J. Am. Chem. Soc. 2012, 134, 4236.
[66] B. M. Axilrod, E. Teller, J. Chem. Phys. 1943, 11, 299.
[67] J. Mutto, Proc. Phys. Math. Soc. Japan 1943, 17, 629.
[68] See http://www.thch.uni-bonn.de/.
[69] J. Rezc, K. E. Riley, P. Hobza, J. Chem. Theory Comput. 2011, 7,
2427.
&10& www.chemeurj.org  2012 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
These are not the final page numbers!
S. Grimme
[70] A. Tkatchenko, O. A. von Lilienfeld, Phys. Rev. B 2008, 78, 045116.
[71] O. A. von Lilienfeld, A. Tkatchenko, J. Chem. Phys. 2010, 132,
234109.
[72] Y. Zhao, D. G. Truhlar, Theor. Chem. Acc. 2008, 120, 215.
[73] L. Goerigk, S. Grimme, J. Chem. Theory Comput. 2011, 7, 291.
[74] F. Weigend, R. Ahlrichs, Phys. Chem. Chem. Phys. 2005, 7, 3297.
[75] F. Weigend, F. Furche, R. Ahlrichs, J. Chem. Phys. 2003, 119, 12753.
[76] T. H. Dunning Jr., J. Chem. Phys. 1989, 90, 1007.
[77] D. Rappoport, F. Furche, J. Chem. Phys. 2010, 133, 134105.
[78] M. Murata, S. Maeda, Y. Morinaka, Y. Murata, K. Komatsu, J. Am.
Chem. Soc. 2008, 130, 15800.
[79] R. Ahlrichs, M. Br, M. Hser, H. Horn, C. Klmel, Chem. Phys.
Lett. 1989, 162, 165.
[80] K. Eichkorn, O. Treutler, H. hm, M. Hser, R. Ahlrichs, Chem.
Phys. Lett. 1995, 240, 283.
[81] K. Eichkorn, F. Weigend, O. Treutler, R. Ahlrichs, Theor. Chem.
Acc. 1997, 97, 119.
[82] P. Y. Ayala, H. B. Schlegel, J. Chem. Phys. 1998, 108, 2314.
[83] J.-D. Chai, M. Head-Gordon, Phys. Chem. Chem. Phys. 2008, 10,
6615.
[84] R. F. Ribeiro, A. V. Marenich, C. J. Cramer, D. G. Truhlar, J. Phys.
Chem. B 2011, 115, 14556.
[85] J. Rezc, P. Hobza, J. Chem. Theory Comput. 2012, 8, 141.
[86] M. K. Gilson, K. K. Irikura, J. Phys. Chem. B 2010, 114, 16304.
[87] A. Choutko, W. F. van Gunsteren, P. H. Hnenberger, ChemPhys-
Chem 2011, 12, 3214.
Received: February 15, 2012
Published online: && &&, 0000
Chem. Eur. J. 0000, 00, 0 0
Supramolecular Binding Thermodynamics
Unprecedented accuracy for computed
association free enthalpies of supra-
molecular hostguest complexes (some
examples are shown here) in solution
has been achieved by a combination of
Chem. Eur. J. 2012, 00, 0 0  2012 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
FULL PAPER
Theoretical Chemistry
S. Grimme* . . . . . . . . . . . . . . . . . . . &&&&&&&&
Supramolecular Binding Thermody-
namics by Dispersion-Corrected
Density Functional Theory
high-level quantum chemical proce-
dures. The approach is quite general,
includes all basic physical effects quan-
titatively and requires no special
empirical adjustments.
www.chemeurj.org &11&
These are not the final page numbers!