You are on page 1of 12

ARTICLE

pubs.acs.org/JPCC

Ab Initio, Physically Motivated Force Fields for CO2 Adsorption


in Zeolitic Imidazolate Frameworks
Jesse G. McDaniel,† Kuang Yu,† and J. R. Schmidt*
Theoretical Chemistry Institute and Department of Chemistry, University of Wisconsin—Madison, Madison, Wisconsin 53706,
United States
bS Supporting Information
Downloaded via TATA INST OF FUNDAMENTAL RESEARCH on September 8, 2022 at 09:55:40 (UTC).

ABSTRACT: We present an entirely ab initio methodology, based on symmetry adapted


See https://pubs.acs.org/sharingguidelines for options on how to legitimately share published articles.

perturbation theory (SAPT), for constructing force-fields to study CO2 adsorption in


nanoporous zeolitic imidazolate frameworks (ZIFs). Our approach utilizes the SAPT energy
decomposition to generate physically motivated force fields for the CO2-ZIF interaction, with
explicit terms representing exchange, electrostatic, induction and dispersion interactions. Each
of these terms is fit to the corresponding term in the SAPT energy decomposition, yielding a
force field entirely free of empirical parameters. This approach was utilized to construct force
fields describing the CO2 interaction with both ZIF-8 and ZIF-71. In conjunction with our
existing CO2CO2 force field, parametrized in a consistent manner, we validate our force
fields using grand canonical Monte Carlo simulations and obtain good agreement with the
corresponding experimental CO2 adsorption isotherms. Furthermore, the explicit correspon-
dence between force field terms and fundamental interaction types (dispersion, electrostatics,
and induction) allows for an analysis of the underlying physics controlling ZIF gas adsorption
that is far more direct and well-defined than with the generic force fields that had been previously utilized to study these systems.
As our force fields are free from empirical parameters, these results demonstrate the potential for computationally screening novel
ZIFs for flue gas separation applications with near quantitative accuracy.

’ INTRODUCTION interactions between different gas molecules and ZIF frame-


Concern about the increasing production of carbon dioxide works.
(CO2) and its role in global warming has led to significant in- Molecular dynamics (MD) and Monte Carlo (MC) computer
terest in postcombustion CO2 capture and separation. As coal simulations are important tools for studying gas adsorption
fired power plants are one of the major point-source emitters of isotherms and diffusion in porous materials such as ZIFs. The
CO2, they are the obvious targets for the implementation of use of accurate and physically appropriate force fields in these
carbon capture and sequestration (CCS) technologies. One simulations can provide valuable insight into the important inter-
inherent challenge is that the flue gas emitted from such a power actions between gas and framework that dictate adsorption capacity
plant is relatively dilute in CO2, with the major component being and/or selectivity. There has been a significant amount of previous
nitrogen gas. Nanoporous metalorganic frameworks (MOFs) computational work modeling the uptake and diffusion of various
have been shown to exhibit high CO2 uptake with excellent gases in different ZIFs. Much of this computational work has utilized
adsorption selectively over nitrogen and other gases, and have standard “generic” force fields such as UFF41 or Drieding42 to
thus received a large amount of research interest as potential CO2 describe the interactions of gas molecules with the ZIF framework,
separation materials. Here zeolitic imidazolate frameworks (ZIFs),135 yielding mixed results and raising concerns regarding the applica-
consisting of metal cations bridged by organic imidazolate anions, have bility of these standard force fields to these systems. Nonetheless,
received particular attention due to their selective CO2 adsorption,1,2,7,8 many previous grand canonical Monte Carlo (GCMC) and MD
excellent chemical and thermal stability,6,36 and potential for tunability studies have examined the uptake and separation of various gases
via functionalization of the organic linker groups.2,4,7,8 including CO2, N2, CH4, and H2 with ZIFs, characterizing adsorp-
Uptake and selective adsorption of CO2, N2, and CH4 in ZIFs tion isotherms of these gases,5,38,4357 separation and selective
is relevant to many applications, including natural gas recovery adsorption of gas mixtures,4446,49 diffusion,4345,48,49 and impor-
and CCS,13,5,7,8,3740 and a wide variety of ZIFs with different tant interaction sites in the ZIFs.38,43,58
topologies, pore/aperture sizes, and chemical functionality have In particular, ZIF-68 and ZIF-69 have been the subject of
previously been developed.111,15,16,18,20,24,25,35 This tunability numerous computational studies targeting CO2 adsorption and
has led to the description of ZIF synthesis as structural “design”
as opposed to materials “discovery”.7 Yet in order to develop and Received: September 27, 2011
synthesize ZIFs optimized for a particular application, it is Revised: December 9, 2011
essential to have a fundamental understanding of the detailed Published: December 15, 2011

r 2011 American Chemical Society 1892 dx.doi.org/10.1021/jp209335y | J. Phys. Chem. C 2012, 116, 1892–1903
The Journal of Physical Chemistry C ARTICLE

separation due to their excellent capacity and selectively.2 “unphysical” parameters. We illustrate this principle via a simple
Various force fields have been used in these studies with mixed example, CO2 adsorption in ZIF-8. Using the EPM2 model60 for
conclusions. For example, describing both CO2CO2 interac- CO2 in conjunction with UFF ZIF parameters and density
tions and CO2ZIF interactions using the UFF force field yields functional theory (DFT) derived charges for the ZIF-8 frame-
reasonable agreement with the experimentally measured CO2 work (from ref 38), we locate three dramatically different param-
adsorption,50 although it is well-known that the UFF force field eter sets that all reproduce the same experimental adsorption
does a poor job of describing bulk CO2CO2 interactions. It is isotherm. For the first parameter set, we used the scaling de-
therefore perhaps not surprising that utilizing either the UFF/ veloped by Perez-Pellitero et al.,38 a uniform scaling of σ1scale =
Drieding parameters for the CO2ZIF interactions in conjunc- 0.95σUFF and ε1scale = 0.69εUFF. For the second parameter set,
tion with the (accurate) TraPPE59 force field for CO2CO2 we use a different uniform scaling of the LJ parameters of σ2scale =
interactions led to dramatic over prediction of CO2 adsorption.48 0.9σUFF and ε2scale = 0.9εUFF, and artificially modify the role of
Other authors resorted to scaling the UFF parameters describing electrostatics by scaling all charges by q2scale
i = 0.25qoriginal
i , where
CO2ZIF interactions so that interactions were less attractive, qioriginal
are the original charges developed by Perez-Pellitero et al.
then yielding good agreement with experimental isotherms.46 Note that this extreme scaling is certainly unphysical and is
Recently it was also noted that by blocking small pores that were merely meant to illustrate the diverse parameter sets that are
deemed “inaccessible” to gas molecules, either a UFF or Drieding consistent with the experimental data. The third parameter set
description of gas-framework interaction could reproduce ex- utilizes different scalings for the imidazolate ring atoms and the
perimental isotherms to reasonable accuracy.47 methyl functional group attached to the ring, thus modifying the
Other notable applications of standard force fields to ZIFs importance of the functional group in comparison to the ring
include the simulations by Perez-Pellitero et al., in which adsorp- ring = 0.95σUFF and εring = 0.85εUFF for the ring
atoms: σ3scale 3scale

tion isotherms for CO2, CH4, and N2 in ZIF-8, ZIF-76, and ZIF- atoms and σmeth = 0.75σUFF and ε3scale
3scale
meth = 0.5εUFF for the methyl
69 were found to agree with experiment only after introduc- atoms. The complete example parameter sets used in these
ing empirical scaling factors of 0.69 for ε values and 0.95 for simulations are listed in Table 1. As shown in Figure 1, these
UFF σ values,38 indicating a very significant empirical correction. parameter sets yield very similar isotherms as predicted by
Battisti et al. utilized this same scaling of the UFF gas-framework GCMC. However, they suggest dramatically different interpreta-
parameters to systematically study the adsorption and dynamics tions as to the importance of electrostatic and dispersive inter-
of CO2, CH4, N2, and H2 and mixtures thereof in ZIF-2 through actions, as well as the role of the methyl functional group in
ZIF-10.49 Morris et al. examined ZIFs 25, 71, 93, 96, and 97,5 determining the CO2 uptake, depending on the scaling method
using the EPM2 force field60 for CO2 and the UFF force field to utilized. Furthermore, these algorithms would likely yield very
describe the framework and were for the most part able to repro- different and inconsistent results when applied to other func-
duce the experimental trends in CO2 adsorption with function- tionalized ZIFs. This inherent ambiguity when empirically
alization of the imidazolate linker groups, with the exception of optimizing existing force fields motivates our present efforts to
ZIF-96. In the latter case they found that by using OPLS-ua develop parameter-free, ab initio force fields for nanoporous
parameters61 to describe the nitrile and amine groups they were systems; nonetheless, approximate generic force fields are still
able to obtain better agreement with the experimental isotherm extremely valuable for use in computational screenings, where
for this system. the availability of parameters for arbitrary systems is a principle
In contrast, the literature contains few examples of ab initio- concern.
based force fields to describe gas adsorption/diffusion processes These considerations motivated us to develop entirely first-
in MOFs. Schmid and co-workers have developed ab initio force principles based force fields for CO2ZIF interactions. In this
fields to describe framework flexibility in a variety of MOFs,6264 work, we present a methodology for the development of such
this being of potential importance in understanding gas diffusion in force fields and, utilizing our recently developed CO2 model for
these systems. Goddard and co-workers have parametrized func- adsorbateadsorbate interactions,67 have carried out GCMC
tional group-H2 interactions based on ab initio calculations,53,54,65,66 simulations to compute CO2 adsorption isotherms. We validate
enabling first principles calculation of H2 uptake in MOFs doped our approach via calculations on ZIF-8 and ZIF-71, as these ZIFs
with metal cations. They have also used this methodology to contain relatively “simple” linker groups. However the meth-
investigate H2 adsorption in ZIFs.53,54 Thus very little work has odologies developed here are equally applicable to a vast array of
focused on developing force fields tailored specifically to ZIFs, and existing and novel ZIFs. As in our previous work,67 the force
in nearly all cases force field construction has been limited to fields developed here rely on the use of symmetry adapted
deriving atomic charges from ab initio calculations and coupling perturbation theory (SAPT)68 to compute the interaction en-
these with Lennard-Jones (LJ) parameters of a generic force field ergies between two species. SAPT has the advantage of not only
such as UFF or Dreiding.5,38,43,47,48,50 To our knowledge, there have high accuracy, but it also naturally decomposes the interaction
not been any attempts to develop rigorous ab intio force fields for energy into the fundamental components of electrostatic, ex-
the interaction of ZIFs with species relevant to flue gas separation. change, induction and dispersion interactions. This allows for the
Thus overall, previous simulations of gas uptake and selective utilization of separate, physically appropriate terms to represent
adsorption have often been forced to resort to arbitrary scaling each of these interaction types. As they include essentially all of
factors parametrized to prior experimental data and a patchwork the relevant “physics”, these physically motivated force fields
of standard force fields to achieve semi-quantitative agreement yield not only extremely accurate results but also a natural energy
with experiment. This is somewhat unsatisfying as this scaling decomposition allowing for greater interpretation of thermo-
procedure may not capture the correct underlying physics. In dynamic predictions (adsorption isotherms) in terms of the under-
other words, there are likely many possible scaling procedures lying fundamental interactions. In order to ensure robust, phys-
(and thus many force fields) that reproduce the same isotherm, ical, and asymptotically valid parameters in our force field, we obtain
leading to an underspecified fitting problem and the possibility of parameters from monomer calculations (polarizabilities, charge
1893 dx.doi.org/10.1021/jp209335y |J. Phys. Chem. C 2012, 116, 1892–1903
The Journal of Physical Chemistry C ARTICLE

Table 1. Scaled UFF Parameter Sets Used to Compute the CO2 Isotherms Shown in Figure 1a

parameter set 1 parameter set 2 parameter set 3

q ε (kJ/mol) σ (Å) q ε (kJ/mol) σ (Å) q ε (kJ/mol) σ (Å)

C1 0.640 0.3033 3.259 0.160 0.3954 3.088 0.640 0.3734 3.259


C2 0.080 0.3033 3.259 0.020 0.3954 3.088 0.080 0.3734 3.259
C3 0.670 0.3033 3.259 0.168 0.3954 3.088 0.670 0.2197 2.573
H2 0.144 0.1271 2.440 0.036 0.1657 2.314 0.144 0.1565 2.440
H3 0.144 0.1271 2.440 0.036 0.1657 2.314 0.144 0.0920 1.928
N 0.540 0.1993 3.097 0.135 0.2598 2.935 0.540 0.2454 3.097
Zn 1.100 0.3582 2.338 0.275 0.4669 2.215 1.100 0.4410 2.338
a
The atom types are shown in the corresponding methyl imidazolate ring pictured in Figure 2.

Figure 2. Imidazolate atom types.

Next we discuss the various monomer calculations and fitting pro-


cedures that we use to obtain the force-field parameters. Finally,
we present the GCMC simulations and the computed CO2
adsorption isotherms and discuss possible applications of this
Figure 1. The experimental CO2/ZIF-8 isotherm at 303 K38 is shown
approach to ZIF screening.
as the solid black dots. The GCMC predicted isotherm resulting from
the first set of parameters is depicted by the red squares and solid red ’ MODEL SYSTEM CONSTRUCTION
line; second set of parameters depicted by the green diamonds and green
dashed line; third set of parameters depicted by the blue triangles and SAPT was used to compute the interaction energies necessary
blue dotted-dashed line. to fit parameters in our force field. This technique is not
amenable to periodic systems, and thus our first step was to
distribution, etc.) whenever possible, and fitting remaining param- develop a finite model system that was large enough to be
eters that rely on charge-density overlap to the SAPT interaction representative of the bulk crystal, but small enough to enable
energies. tractable SAPT calculations. Our goal in constructing this model
We should note that many previous interaction potentials have system was to reproduce the full three-dimensional electron
been derived based on SAPT interaction energy calculations,6975 density on the imidazolate rings and surrounding Zn2+ cations as
and we briefly comment on the primary differences between our compared to the corresponding bulk. In creating this model
methodology and techniques used in previous works. Almost all system, we considered a ZIF system with the simplest possible
previous works have fit the total interaction energy between linker group, an unfunctionalized imidazolate ring. The periodic
molecules,69,70,72,74,75 while we choose to independently fit each ZIF crystal used as a precursor was created by obtaining the XRD
individual interaction energy component. This not only allows for crystal structure of ZIF-8 from the Cambridge Structural Data-
greater physical interpretation when such a force field is used in a base (CSD), replacing the methyl groups with hydrogen. Bader
computer simulation, but also reduces the possibility of error charge analysis76 was employed to characterize the electron den-
cancelation and helps to ensure physically meaningful parameters, sity of the imidazolate rings in both the periodic system and the
especially in larger molecules. Our procedure is probably most model fragment computed utilizing the PBE density functional
similar to that of Misquitta et al.,73 with the major difference being with a plane wave basis (500 eV energy cutoff). The electron
that we abandon the use of explicit anisotropic force field terms so density for both the model fragment and the periodic system
that our final force field has a form amenable to standard simulation was computed using the VASP77,78 package. From the peri-
packages. odic system, we constructed a Zn2Im7Li63+ fragment, where two
This paper is organized as follows: We first consider our choice Zn ions were each attached to three outer imidazolate rings, and
of ZIF model system, examining the transformation from an connected through a common imidazolate ring (see the Support-
infinite periodic ZIF crystal, to a medium sized model system ing Information). The outer six imidazolate rings were capped
fragment (Zn2Im7Li63+), and finally to a single Li-capped imidazo- with Li ions. This system was constructed by cutting out the
late type ring that we utilize to obtain force-field parameters. appropriate fragment from the periodic system and capping the
1894 dx.doi.org/10.1021/jp209335y |J. Phys. Chem. C 2012, 116, 1892–1903
The Journal of Physical Chemistry C ARTICLE

Table 2. Comparison of Bader Charges on Imidazolate Rings


of Periodic System and Zn2Im7Li63+ Fragmenta

fragment

atom periodic central ring outer rings (average)

C1 1.062 1.075 1.061


N 1.292 1.305 1.274
N 1.292 1.300 1.353 Figure 3. Structures of Li2mIm ring (left) and Li2dcIm ring (right).
C2 0.588 0.596 0.596
C2 0.676 0.660 0.621 done previously,67 mid bond basis functions were used, as these are
H1 0.024 0.027 0.025 essential to converge dispersion energies.97 Three sites were used for
H2 0.180 0.155 0.119
the mid bond functions, with the sites placed between the CO2
molecule and the three closest imidazolate rings, and the locations
H2 0.180 0.168 0.149
determined using the method of Podeszwa et al.74
Total 0.594 0.570 0.593
a
We seek a smaller model system to reproduce the SAPT
The ring atom types are shown in Figure 2. interaction energies between the Zn2Im7Li63+ fragment and CO2
molecule (and thus the bulk-CO2 interaction). We considered
dangling nitrogen lone pairs with Li ions to mimic dative ligand three different rings: an imidazole ring, imidazolate ring (anion),
bonding. The LiN bond distances were adjusted until the and a Li-capped imidazolate ring (net cation), where the LiN
Bader charges on the fragment rings matched those on the bond distance for this last ring was taken to be the same as that
periodic system. Shown in Table 2 is a comparison of the Bader used in the Zn2Im7Li63+ fragment. As before, GROMACS was
charges of the imidazolate rings in the Zn2Im7Li63+ fragment to used to run NVT simulations at 3000 K to generate configura-
the charges of the imidazolate rings in the corresponding periodic tions for each case. Approximately 700 single-point DF-DFT-
crystal. Comparing the electron density difference and Bader SAPT calculations were carried out using the same methodology
charges for ring atoms (see the Supporting Information for as before. In contrast to the large model system, these smaller
electron density difference isosurface), it is evident that the SAPT calculations took only several processor-hours, thus
Zn2Im7Li63+ fragment does an excellent job of reproducing the allowing for many of these calculations to be done for various
electron density of the periodic system in the central region, systems. Using the methodology described in the next section,
although differences obviously arise near the terminating Li force fields were developed for each of these three ring systems.
groups. Finally, a constrained geometry optimization was carried We find that the force field generated from the small Li-
out on the fragment, where all metalorganic bonds were kept imidazolate (Li2Im) SAPT calculations, in combination with
fixed. The main purpose for this optimization was to generate charges for the Zn ions determined using the methodology
physical CH, CC, and CN bond distances and compensate described below, was able to accurately reproduce the SAPT
for the finite resolution of the XRD structure. interaction energies between CO2 and the larger Zn2Im7Li63+
We subsequently carried out approximately 200 SAPT calcu- fragment (see the Supporting Information). We thus utilize Li-
lations for different configurations of CO2 interacting with the capped imidazolate-type rings in all subsequent calculations to
frozen Zn2Im7Li63+ fragment. The configurations used were efficiently generate force fields for various bulk ZIFs.
taken from snapshots of an NVT MD simulation using a box
with a CO2 molecule and the Zn2Im7Li63+ fragment. The
GROMACS package79,80 was used to generate these conforma- ’ FORCE FIELD DEVELOPMENT
tions. A temperature of 3000 K was utilized to generate sampling in Based on the above results, we developed corresponding small
repulsive regions, and the intermolecular interactions were described model systems for ZIF-8 and ZIF-71. For ZIF-8, a Li-capped
using the OPLS force field61 for the fragment, and the TraPPE59 force methyl-imidazolate ring (Li2mIm) was used and for ZIF-71 a Li-
field for CO2; the results do not depend sensitively on the choice of capped dichloro-imidazolate ring (Li2dcIm) was employed
force field due to the high temperature, and our goal is to sample a (see Figure 3). As described above, ∼700 single-point DF-DFT-
wide variety of representative intermolecular conformations. The SAPT calculations with different CO2 configurations were carried
SAPT calculations were extremely time-consuming (45 days per out for each imidazolate-type ring system. We utilized a dimer cen-
point on 8 processors, without computing dispersion interactions), tered AVTZ/def2-TZVPP basis plus midbond functions, yielding
and are infeasible for more than one type of ZIF. However, the results SAPT interaction energies closely approximating the basis set limit.
provide benchmark interaction energies against which we evaluate The midbond basis consisted of the same basis functions as before,
smaller model systems. The details for the calculations were almost located at the center of mass of the dimer. Unless explicitly stated,
identical to those done in our previous work.67 The density fitting only parameters corresponding to imidazolate ring/substituent
DFT-SAPT (DF-DFT-SAPT) method8191 was used as implemen- atoms were fit, while the CO2 parameters were taken from our pre-
ted in the MolPro 2009 package.92 The PBE exchange-correlation vious work.67 Using a similar methodology as before,67 we separately
functional93,94 was used with an asymptotic correction, which is es- fit each physically distinct term in our force field to the correspond-
sential for accurate interaction energies.86,91 Ionization potentials ing SAPT term in our calculations. We believe that this approach not
required for the asymptotic correction were computed at the aug-cc- only results in a more robust force field with physically accurate
pVTZ (AVTZ)/PBE0 level. We used a dimer-centered basis set parameters, but also allows for greater insight and analysis when
consisting of Dunning style aug-cc-pVDZ (AVDZ) basis95 for organic such a force field is used in simulations.
atoms and def2-SVP basis96 for metal atoms (AVDZ/def2-SVP). For All fitting in this paper, unless otherwise explicitly stated,
twelve of these points, dispersion interactions were calculated, and as utilized a least mean square error approach, in which parameters
1895 dx.doi.org/10.1021/jp209335y |J. Phys. Chem. C 2012, 116, 1892–1903
The Journal of Physical Chemistry C ARTICLE

were explicitly constrained to exhibit the appropriate symmetry where the index i runs over all atoms on the imidazolate-type ring
of the ring, with χ2 = ∑ni W(ESAPT,tot
i )(ESAPT,comp
i  EMM,comp
i )2 and j runs over the three atoms of the CO 2 molecule. The
SAPT,comp
where Ei is the desired SAPT interaction energy com- exponents,Bij, were determined using the procedure used
ponent for data point i, EMM,comp
i is the corresponding energy previously.67 Briefly, initial pairwise exponents, B0ij, were deter-
component predicted by the force field, and W(ESAPT,tot i ) mined using atomic ionization potentials which describe the
is a weighting function for each data point depending on the asymptotic exponential decay of atomic electron densities. Then,
total interaction energy for that configuration. The weighting the prefactors A exch ij along with a scaling factor, λ, such that
function was used to make the fit insensitive to high-energy B ij = λB 0ij, were fit to the SAPT first-order exchange interac-
repulsive points (which are unlikely to be sampled at relevant tion energies. Note that for the Aexch ij prefactors (and all other
temperatures) and was of the Fermi-Dirac form similar Buckingham prefactors used in the other interaction energy
terms), only atomic ring parameters Aexch ii were fit, and a standard
1 combination rule, Aexch = (Aexch exch 1/2
WðEÞ ¼ ij ii Ajj ) , was used to generate the
expððE  μÞ=kB TÞ þ 1 pairwise parameters. For CO2, the Ajj parameters were taken from
our previous work.67 As before,67 a scale factor of λ ≈ 1.2 was found,
where μ was typically chosen to be 5 mHartree and kBT was and the final exponents Bij were used accordingly in all other
chosen to be 1 mHartree (∼300 K). Buckingham terms for the remaining energy components.
We have taken many steps to reduce the correlation of param- The electrostatic term, E(1) pol , was fit using a combination of screened
eters during force field fitting, which is necessary to ensure the atom-centered point charges and Buckingham-type terms to account
resulting parameters are physically meaningful and transferable. for charge penetration. The atom-centered point charges were
Using physical considerations and obtaining parameters from obtained as follows. First, the electron density of the imidazolate-type
monomer calculations whenever possible, we rely on SAPT inter- ring was computed using the same level of theory as used for the
action energies only for fitting terms explicitly involving electron SAPT calculations, and a distributed multipole analysis (DMA)98,99
overlap (e.g., exponential, Buckingham-like terms, vide infra). was carried out using the Molpro package.92 Using the method
Nonetheless, we have found cases where there is coupling between proposed by Ferenczy,100 atom centered point charges were fit to
Buckingham coefficients on neighboring atoms when fitting inter- represent these distributed multipole moments. Fitting atomic charges
action energies. Therefore, while fitting all Buckingham coefficients to DMA generated multipole moments rather than the electrostatic
on functionalized imidazolate rings, we have introduced weak potential (ESP) leads to negligible differences in reproducing the ESP,
harmonic constraints for carbon, nitrogen, hydrogen, and lithium but has the advantage of resulting in more transferable charges and
atoms. The values around which parameters were constrained were prevents unphysical charges on “buried” atoms.101,102 The final
determined from corresponding fits to SAPT calculations for a Li- functional form used to fit the electrostatic SAPT term was
capped unfunctionalized imidazolate ring interacting with a CO2
qi qj
molecule. For each energy component, only four atomic parameters
were used, namely those for the four atom types mentioned above.
ð1Þ
Epol = ∑
i, j
f1 ðBij , rij Þ
rij ∑
þ
i, j
ij expð  Bij rij Þ
Aelec
In order to determine the uniqueness of the unfunctionalized
imidazolate ring fits, we analyzed the fit of Buckingham terms for the For the screening function, the TangToennies functional
exchange energy. We constructed a covariance matrix, describing form103 was used
the sensitivity of the fit of the data points to percentage changes in n ðλrÞm
the parameters. Small eigenvalues of this covariance matrix indicate fn ðλ, rÞ ¼ 1  eλr ∑
m¼0 m!
that the parameters are under-determined due to coupling, and the
corresponding eigenvectors describe the coupling of these param- where the parameters required for the screening function were
eters. We therefore characterized the change in rms error of the fit taken to be the exponents used in all Buckingham terms, λ = Bij,
with a change in parameter space along the eigenvector of the as both screening and Pauli repulsion arise from electron cloud
covariance matrix corresponding to the smallest eigenvalue. We overlap. The atomic ring parameters Aelec ii , required to generated
found a 5% increase in rms error with an average change of 2025% the pairwise Aelec parameters, were then fit to reproduce the
ij
of each parameter along this eigenvector. Therefore, we expect these SAPT first-order electrostatic energies. As the penetration terms
parameters to be qualitatively accurate and physically meaningful for for electrostatics were attractive in nature, the combination rule
use as atomic base values, and thus all subsequent fits utilized these Aelec = (Aelec elec 1/2
ij ii Ajj ) was used.
resulting parameters to center the harmonic constraints for similar The second order polarization/induction energy, E(2) (2)
pol = Eind +
atom types. The magnitude of the constraint is very weak, leading to Eindexch, was fit using an explicitly polarizable treatment for CO2.
(2)
rms errors in the energy fits that were negligibly different for This explicit polarization was accomplished with the use of a shell
constrained and unconstrained fits. model, and is described in our previous work.67 All intermolecular
The SAPT interaction energy decomposition is as follows: chargecharge interactions, whether static or shell, were screened
ð1Þ ð1Þ ð2Þ ð2Þ ð2Þ using the Tang-Toennies function as described above. The screen-
Eint ¼ Epol þ Eexch þ Eind þ Eind  exch þ Edisp ing parameters for the shells were taken to be the same as those used
ð2Þ
for the corresponding atoms. All intramolecular chargecharge
þ Edisp  exch þ Eδhf interactions (due to polarization) were screened with a Thole-type
function104 as described previously.67 No explicit polarization was
To fit the first order exchange repulsion energy, E(1) exch, we used for the ZIF framework. The total functional form used to
utilized a pairwise additive Buckingham functional form model the induction energy is
ð1Þ
Eexch = ∑i, j Aexch
ij expð  Bij rij Þ
ð2Þ
Epol = Ushell þ ∑i, j Aindij expð  BijrijÞ
1896 dx.doi.org/10.1021/jp209335y |J. Phys. Chem. C 2012, 116, 1892–1903
The Journal of Physical Chemistry C ARTICLE

Here, Ushell is defined as the difference between the energy of the The final functional form used to reproduce the SAPT
system with optimized shell positions, and that of the corresponding dispersion energies is
nonpolarizable system. The atomic ring parameters, Aind jj , required
Cij
ij parameters were fit to reproduce the SAPT
for the pairwise Aind
second-order polarization/induction energy, E(2)
ð2Þ
EVdW = ∑i, j Adisp
ij expð  Bij rij Þ  ∑ ∑ fn ðβij , rij Þ nn
rij
pol . As with electro- n ¼ 6, 8, 10 i, j
statics, it was found that the penetration terms for induction were
ij = (Aii Ajj )
attractive in nature, and the combination rule Aind ind ind 1/2 Here, we have included two scale factors such that βij = λdampBij
was used to generate the pairwise parameters. Since the shell and Cijn = λasympCij0n . Both of these scale factors, as well as atomic
model can exhibit overpolarization in strong electric fields,105 Adisp
ii ring parameters, were fit to the SAPT imidazolate-type
and the capping Li atoms are formally cations, configurations of ring/CO2 dispersion interaction energies and thus are not
CO2 close to Li exhibit induction energies that are incapable of empirically determined. Similar to the treatments of electro-
being fit with the shell model; we omit these data points (using a statics and induction, the pairwise combination rule Adisp ij =
methodology described below) in our fits to yield more physical (Adisp disp 1/2
ii Ajj ) was used. Because of the noted sensitivity of
Aind
ij parameters. It is important to note that this problem with
the CO2 isotherms to dispersion parameters, we added a penalty
the shell model only arises due to the ionic nature of the Li- function to the typical least mean square error fitting approach.
capped imidazolate-type rings that we use to generate param- This penalty function was proportional to the absolute mean
eters for the ZIF systems, and as the ZIF crystals contain only deviation of the fitted energies to the SAPT values, and the
buried cationic centers we do not anticipate these problems in purpose of this penalty function was to prevent any systematic
simulations of such systems. error in the fit. Also, to ensure that the correct asymptotic
The second-order dispersion energy, E(2) (2) (2)
VdW = Edisp + Edispexch,
behavior was preserved, rather than constructing the χ2 function
was fit using a combination of steps to arrive at an accurate and using the absolute energy differences, we used the percentage
physically meaningful parameter set; here a na€ive fit directly to energy deviations. This ensured that small dispersion energies
the SAPT energies yields ambiguous results due to the long- were fit accurately; although small in magnitude, these weak
range nature of dispersion and thus strong coupling between interactions contribute significantly to bulk adsorption due to the
atomic parameters. We thus employ the approach developed by corresponding increase in the volume element at large distances.
Williams and Stone,106 as implemented in the CAMCASP Because our previous CO2 model67 used a different treatment
program,107110 to compute distributed imaginary frequency- for dispersion, we obtained new atomic parameters for CO2
dependent polarizabilities and obtain atomic C6, C8, and C10 consistent with this new functional form. This was done by
dispersion parameters. For these monomer calculations we used refitting the previously computed SAPT CO2CO2 dispersion
an AVTZ basis with an asymptotically corrected PBE0 exchange- interaction data67 using the functional form and methodology
correlation functional. The linear response employs an ALDAX + discussed above. The final dispersion energy fits utilized a scale
CHF kernel, meaning 25% exact HartreeFock exchange was factor λasymp that was typically ∼1.05, indicating that the
used, consistent with the amount present in the PBE0 functional Cijn parameters used in the force field were typically 5% greater than
(see CAMCASP manual for further discussion). This methodol- those predicted by the CAMCASP calculations (possibly due to
ogy was used to compute atomic C6, C8, and C10 dispersion basis set size as well as order of polarizability tensors utilized109).
parameters for CO2 and the Li2dcIm and Li2mIm imidazolate- The scale factor λdamp used to scale the “exponents” for the
type rings (our previous bulk CO2 force field67 did not utilize TangToennies damping function was typically ∼0.9, which in
C8 or C10 terms). The pairwise dispersion parameters, Cij0 n were
combination with the original ∼1.2 scaling of the exponents from
then generated using the combination rule Cij0 ii0 jj0 1/2
n = (Cn Cn ) , their asymptotic values, led to damping exponents very close to
where the atomic parameters Cii0 n were those computed using
those predicted by the exact asymptotic atomic densities.
CAMCASP. A notable exception to this rule was for hydrogen- The delta HartreeFock terms, Eδhf, have been described
heavy atom pairwise parameters. Here, consistent with literature previously67,86 and were fit using exponential functions
recommendation,109 only atomic C6 parameters were computed
for hydrogen atoms, and therefore the nonzero C8 parameter for
Eδhf = ∑i, j Aδhf
ij expð  Bij rij Þ
such pairs was given by CHydHeavy0
8 = (CHyd0
6 CHeavy0
10 )1/2, which
is a consistent definition when noting the rank of the polariz- The atomic ring parameters Aδhf jj were fit to reproduce the
ability tensors involved in computing each Cn term. calculated Eδhf interaction energies, using a slightly different
As with the induction energy fit, we found it best to remove combination rule to generate the pairwise parameters Aδhf ij . The
points from the SAPT data set in which the CO2 molecule was combination rule used here was Aδhf δhf δhf 1/2
ij = (|Aii ||Ajj |) , where
“too close” to either Li atom of the imidazolate type ring. The the sign of Aδhf δhf δhf
ij was determined to be positive if Aii Ajj was
data points were removed by estimating the dispersion energy of positive, and negative otherwise.
each configuration using the parameters from the CAMCASP All fits for the Li2mIm and Li2dcIm ring systems are given in
calculations; data points were removed if either Li atom of the the Supporting Information (sections 3 and 4) as are the
imidazolate type ring was the main contributor to the dispersion complete set of parameters generated by these fits.
interaction energy. For all other energy components, only Finally, we note one possible potential improvement in our
Buckingham coefficients were fit, accounting for charge pen- force field fitting methodology. For intermolecular configura-
etration effects. The dispersion energy, in addition to fitting tions that are closer than the potential minimum (i.e., on the
Buckingham coefficients, was also used to fit two scale factors, repulsive “wall”), the total interaction energy results from
one of which adjusts the magnitude of the asymptotic C6C10 cancellation of substantially larger individual interaction energy
terms. As the C6C10 treatment of dispersion becomes unphy- components. Thus small errors (<5%) in individual energy
sical in charge penetration regions, we found it best to remove components can result in non-trivial errors in total energy. As
said data points for this fit. these configurations involve significant charge penetration, the
1897 dx.doi.org/10.1021/jp209335y |J. Phys. Chem. C 2012, 116, 1892–1903
The Journal of Physical Chemistry C ARTICLE

associated errors are likely related to our approximations in The resulting parameters for both ZIF-8 and ZIF-71 are sum-
accounting for complex charge penetration effects in individual marized in the tables below.
interaction terms. Note that we would almost always expect these Finally, it is worth emphasizing that, despite the technical
configurations to be unimportant in a typical simulation since nature of the fitting procedure outlined above, the resulting force
they involve short inter-molecular distances and thus typically field is of the relatively simple function form
high total energies. However, at high temperatures or for specific
qq
ZIF topologies it is conceivable that these points may contribute
non-trivially to the Boltzmann-weighted ensemble. In this case,
Utot ¼ ∑i, j f1 ðBij, rij Þ riij j
we suggest that it may be preferable to revert to utilizing a !
piecewise-defined potential, utilizing a different functional form Cij
to describe these very shortest distances under the constraint of þ ∑i, j Atot
ij expð  Bij rij Þ  ∑
n ¼ 6, 8, 10
fn ðβij , rij Þ nn
rij
continuity and differentiability of the total potential.
Force fields for bulk ZIF-8 and ZIF-71 were constructed by
treating Zn atoms as bare point charges (since they are fully þ Ushell
coordinated, buried, and cationic), and using the parameters
developed for the Li2 mIm and Li2 dcIm ring systems respec- consisting of electrostatics, Buckingham (exponential) short-
tively, for the remaining organic atoms. Atomic charges are range repulsion, damped dispersion, and Drude oscillator polar-
obtained using DMA analysis on slightly larger models ization. This makes it straightforward to implement in existing
systems. Starting with the Zn2 Im7 Li6 3+ fragment, we con- standard simulation packages, such as GROMACS79,80 or
structed Zn 2 mIm7 Li6 3+ and Zn2 dcIm7 Li6 3+ fragment sys- Towhee.111
tems easily by functionalizing the imidazolate rings. A sub-
sequent DMA followed by a fit to point charges (as before) Table 4. Force Field Parameters for ZIF-71a
was carried out for both Zn 2mIm7 Li 6 3+ and Zn2 dcIm 7Li6 3+
atom type i C1 C2 Cl H1 N1 Zn
fragments. Note that the charges resulting from the DMA
naturally yield a nearly perfect electrically neutral bulk q(e) 0.5934 0.3886 0.1139 0.0609 0.9992 1.5894
stoichiometric material, and thus no adjustment was neces-
Carbon
sary. These charges were then used for the simula-
tions involving ZIF-8 and ZIF-71 respectively (Figure 4). Atot (kJ/mol)/104 6.381 5.051 17.227 1.144 10.650 0.000
B (Å1) 3.222 3.222 3.351 3.392 3.448 3.917
C6 [kJ/(mol Å6)]/102 10.501 17.440 23.721 5.642 9.286 0.000
C8 [kJ/(mol Å8)]/103 7.715 19.351 15.327 3.642 14.922 0.000
C10 [kJ/(mol Å10)]/104 2.807 10.636 4.878 0.000 11.760 0.000
Oxygen
tot 4
A (kJ/mol)/10 12.125 9.064 23.952 2.421 19.259 0.000
B (Å1) 3.396 3.396 3.554 3.603 3.672 4.293
C6 [kJ/(mol Å6)]/102 8.734 14.505 19.729 4.692 7.723 0.000
C8 [kJ/(mol Å8)]/103 4.326 10.851 8.595 1.334 8.367 0.000
C10 [kJ/(mol Å10)]/104 1.028 3.894 1.786 0.000 4.306 0.000
a
After the charges are listed, the cross-parameters are listed for the
Figure 4. Atom types for imidazolate rings in ZIF-8 (left) and ZIF-71 carbon atom of CO2, followed by the cross-parameters for the oxygen
(right). atoms of CO2.

Table 3. Force Field Parameters for ZIF-8a


atom type i C1 C2 C3 H2 H3 N Zn

q(e) 0.6786 0.1198 0.1498 0.1182 0.0682 1.0135 1.6352


Carbon
Atot (kJ/mol)/104 4.474 6.636 6.129 1.076 1.199 10.573 0.000
B (Å1) 3.222 3.222 3.222 3.392 3.392 3.448 3.917
C6 [kJ/(mol Å6)]/102 10.033 16.076 18.381 2.344 2.384 10.448 0.000
C8 [kJ/(mol Å8)]/103 12.097 15.223 15.286 1.513 1.539 14.607 0.000
C10 [kJ/(mol Å10)]/104 7.355 7.063 6.165 0.000 0.000 9.914 0.000
Oxygen
Atot (kJ/mol)/104 7.859 10.483 8.823 1.405 1.844 19.296 0.000
B (Å1) 3.396 3.396 3.396 3.603 3.603 3.672 4.293
C6 [kJ/(mol Å6)]/102 8.344 13.371 15.288 1.950 1.983 8.690 0.000
C8 [kJ/(mol Å8)]/103 6.783 8.536 8.571 0.554 0.563 8.191 0.000
C10 [kJ/(mol Å10)]/104 2.693 2.586 2.257 0.000 0.000 3.630 0.000
a
After the charges are listed, the cross-parameters are listed for the carbon atom of CO2, followed by the cross-parameters for the oxygen atoms of CO2.

1898 dx.doi.org/10.1021/jp209335y |J. Phys. Chem. C 2012, 116, 1892–1903


The Journal of Physical Chemistry C ARTICLE

Figure 5. CO2 adsorption isotherm in ZIF-8 at 303 K. Experimental Figure 6. CO2 adsorption isotherm in ZIF-71 at 298 K. Experimental
results from ref 38 are shown as black squares. Simulation results from results from ref 5 are shown as the black squares. Simulation results
this work are shown as the red circles. utilizing the UFF force field also from ref 5 are shown as the blue squares.
Simulation results from this work are shown as the red circles.

’ RESULTS AND DISCUSSION 10 Å. For the long-range part of the Buckingham potential, we use a
GCMC simulations were carried out to compute CO2 adsorp- cutoff of 14 Å  for ZIF-71 and 16 Å  for ZIF-8, these cutoffs being
tion isotherms in ZIF-8 and ZIF-71 and validate our parameter- dictated by the size of the unit cell (or super cell); no additional long-
free force fields. Our recently developed CO2 model, which has range correction is used due to the inhomogeneous nature of the sys-
been derived in a similar fashion and benchmarked against tem. We utilize a conjugate gradient algorithm115 to optimize the shell
various bulk CO2 properties, was used to describe CO2CO2 positions at every step. Particle insertions and deletions were attempt-
interactions.67 The unit cell of ZIF-8 was obtained from the CSD ed with equal probability, and the combination of these moves
crystal structure, and the refined unit cell of ZIF-71 was obtained accounted for 20% of all trial moves, with the remaining 80% of
from the authors of ref 5. For ZIF-8, a box consisting of 2  2  2 moves being hybrid MD/MC type displacements.
unit cells was used so as to allow for a longer real space cutoff for The calculated CO2 adsorption isotherms for ZIF-8 and ZIF-
the calculation of dispersion terms. It has been shown that using 71 are shown in Figures 5 and 6, respectively. The nonideality
experimental ZIF XRD crystal structures in GCMC simulations dependence of the chemical potential of CO2 at various pressures
leads to results comparable to using DFT optimized structures, as was accounted for using the PengRobinson equation of state.
long as the XRD crystal structure does not contain ambiguous For both ZIFs, we find good agreement with the experimental
atom positions.48 We utilize a rigid treatment of ZIF atom isotherms. It is important to note that these force fields have been
positions for both simulations, which we believe is justified since generated entirely from ab initio calculations, demonstrating the
we are computing room temperature isotherms at relatively low potential to predict CO2 adsorption in ZIFs completely from first
pressures. We note that previous observations of ZIF-8 frame- principles without resorting to arbitrary “scaling” to fit previously
work structure changes due to gas adsorption have occurred at measured experimental results.
either much higher pressures or much lower temperatures.112 We can trivially analyze the contributions of the distinct types
Framework flexibility may play an important role in transport of physical interactions to the average interaction energy con-
properties, particularly with bulky adsorbates, although we tributing to CO2 adsorption in these ZIFs under representative
examine strictly equilibrium properties in the present work. conditions. This energy decomposition is an intrinsic feature of
Since we use a shell model for CO2, all shell positions have to the physically motivated force field, with distinct terms in our
be reoptimized whenever any CO2 molecule is moved. This force field used to represent these different types of physical
makes single-molecule translation/rotation moves extremely interactions. Shown in Figure 7 are the percentage contributions
inefficient. We thus utilize a GCMC simulation with hybrid to the total interaction energies for ZIF-8 and ZIF-71 at condi-
MD/MC translational/rotational moves.113 In this hybrid ap- tions close to STP. It is evident that both these systems are
proach, we randomly select center-of-mass and angular velocities dispersion dominated, emphasizing the importance of obtaining
from a MaxwellBoltzmann distribution for each attempted the correct physical description for this term in the force fields.
move, compute the forces and torques on all CO2 molecules, and The fact that electrostatics (and thus induction) plays a more
numerically integrate Newton’s equations of motion to generate minor role in the adsorption is in part due to the fact that both
a new trial move in which all CO2 molecules in the system have ZIF-8 and ZIF-71 are constructed with symmetrically function-
been displaced. Here the only criterion is that the integrator must alized imidazolate linker rings. As pointed out by Morris et al.,5
be reversible (energy conservation is not required), and we utilize ZIFs with symmetrically functionalized linker groups tend to
the velocity Verlet method, with a typical time step of ∼50 fs. In produce weaker electrostatic fields in their cavities than do
conjunction with these hybrid translation/rotation moves, we similar ZIFs with asymmetric functionalization. Therefore, we
utilize standard single particle insertion/deletion techniques to believe that electrostatic and induction interactions will be of
complete the GCMC approach. The particle-mesh Ewald (PME)114 greater importance in other, asymmetric ZIFs, and we believe
technique is used to treat electrostatics, with a real space cutoff of that our use of DMA to obtain physical atomic charges and the
1899 dx.doi.org/10.1021/jp209335y |J. Phys. Chem. C 2012, 116, 1892–1903
The Journal of Physical Chemistry C ARTICLE

explicit treatment of many-body induction effects through our simulation are presented in Figure 9 (generated using the VMD
shell model for CO2 will be increasingly important for such cases. software116). We find that this density plot is very similar to the
A comparison of the volumetric CO2 uptakes in ZIF-8 and one computed by Perez-Pellitero et al.38 (as can be verified by
ZIF-71 given by our simulations is shown in Figure 8 (note these comparing the (100) perspective to the identical perspective in
are at slightly different temperatures corresponding to the ref 38). The prominent adsorption sites are on either side of the
experimental conditions). From these isotherms, we gain insight six-membered ring aperture between cages in the ZIF crystal.
on the effect of functionalization of the imidazolate rings on the Three of the six methyl-imidazolate rings comprising the aper-
CO2 uptake. At pressures less than ∼3 bar, the isotherms show ture have their methyl groups oriented along the aperture, and a
that ZIF-71, with imidazolate rings “functionalized” with chlorine typical configuration shows a CO2 molecule close (∼3 Å) to one
atoms, exhibits greater CO2 uptake than ZIF-8 which has of these methyl groups. To analyze the energetics controlling the
imidazolate rings “functionalized” with a single methyl group. binding at such sites, a short NVT-ensemble simulation was run
At low pressures, gas uptake should be dominated by the at 200K with one CO2 molecule initially positioned at this
magnitude of the binding energies at attractive sites, and indeed interaction site. This low temperature is chosen so that the
our simulations give average adsorption energies of 14.14 and molecule would not desorb from this site, thus allowing us to
19.41 kJ/mol for a CO2 molecule in ZIF-8 and ZIF-71, analyze the binding energetics at this single site. We find an
respectively (at conditions of ∼300 K, 1 bar). At pressures average total adsorption energy of 1617 kJ/mol for such sites,
greater than ∼5 bar we find that ZIF-8 starts to exhibit slightly with attraction dominated by dispersion (roughly 2.5:1/disper-
higher uptake than ZIF-71, most likely due to the former’s higher sion to electrostatics). This is consistent with the findings of
free volume (experimental BET surface areas are ∼650 m2/g for Perez-Pellitero et al.,38 where it was noted that the main adsorp-
ZIF-715 and ∼1450 m2/g for ZIF-8, with similar topologies;38 tion sites match the regions of maximal “van der Waals” inter-
note that FZIF8/FZIF71 ≈ 0.8). actions. However note that our electrostatic energy accounts for
CO2 density plots in ZIF-8, showing the most CO2 density (and charge penetration effects, whereas Perez-Pellitero’s treatment
thus the most probable adsorption sites) from the Monte Carlo does not. It is also interesting to note that these prominent
interaction sites are the same as those found for methane
adsorption in ZIF-8 by Wu et al.40 using neutron diffraction
(as can be verified by comparing the (111) perspective to the
identical perspective in ref 40). As the electrostatic interaction
between methane and the ZIF-8 framework is weak, the fact that

Figure 7. Percentage contributions of distinct physical terms with


respect to the total interaction energy under representative conditions,
including all CO2CO2 interactions and CO2framework interactions.
Simulations were run at ∼1 bar, ∼300 K for both ZIFs. ZIF-8 is shown in
blue, and ZIF-71 is shown in red. Components which are attractive are
shown as positive percentage contributions, while repulsive contribu- Figure 9. CO2 density plots in ZIF-8: left (100) perspective; right (111)
tions are represented as negative percentages. perspective.

Figure 8. Comparison of ZIF-8 and ZIF-71 computed isotherms. The ZIF-8 isotherm (left) is at 303 K, while the ZIF-71 isotherm (right) is at 298 K.
Note the slight difference in temperatures, corresponding to the experimental conditions. Both isotherms show absolute gas uptakes.

1900 dx.doi.org/10.1021/jp209335y |J. Phys. Chem. C 2012, 116, 1892–1903


The Journal of Physical Chemistry C ARTICLE

The primary limitation of the present approach is that it


requires parametrization for each individual ZIF/MOF system,
which would make their use in computational screening quite
challenging. In this sense, it is important to note that transferable
“generic” force fields do play an extremely important role in the
computational screening of MOFs. Due to their simplicity and
the fact that no new parameters need to be developed, such force
fields can be used to quickly screen many MOFs for a particular
gas adsorption application.117,118 Although the predictions from
these force fields may not always be accurate (as seen in ZIF-8
with unscaled UFF parameters), they provide initial “first pass”
information that can be valuable to narrow down either further
computation investigation with more accurate models or experi-
Figure 10. CO2 density plots in ZIF-71: left (100) perspective; right mental characterization. It is thus important to recognize the
(111) perspective. limitations of both standard “generic” force fields (e.g., accuracy)
and system-specific ab inito force fields (e.g., transferability).
the ZIF-8 adsorption sites are the same for both methane and We are currently working to bridge this divide by developing
CO2 further supports the conclusion that this process is disper- transferable ab initio force fields tailored to specific classes of
sion dominated. materials, e.g. functionalized ZIFs, thus incorporating the best
The ZIF-71 cages have faces (apertures) consisting of four, six, strengths of both approaches; these efforts will be the subject of
or eight imidazolate-type rings, and the CO2 density plots shown future publications.
in Figure 10 indicate that all of these faces act as important
interaction sites. However, similar to ZIF-8, it is seen that the
principle adsorption sites are on either side of the six ring ’ CONCLUSION
apertures. We characterize the adsorption at these specific sites We have developed parameter-free ab initio force fields
via a 200 K simulation, as before, yielding an average total tailored to describe the interactions between CO2 and ZIFs,
adsorption energy of 2425 kJ/mol. At these sites, the interac- specifically ZIF-8 and ZIF-71. These physically motivated force
tion is even more dispersive, roughly 4:1/dispersion to electro- fields are novel in that they contain separate terms to explicitly
statics. This somewhat surprising absence of enhanced elec- account for the distinct types of physical interactions, namely
trostatics is further illustrated by examining the DMA derived exchange, electrostatic, induction, and dispersion. We have made
atomic charges on the ring atoms (given in the Supporting the effort to ensure physically meaningful parameters in our force
Information), where we see a partial charge on the chlorine fields, obtaining as many parameters as possible based on isolated
atoms of only ∼ 0.1 au This result can be explained by the molecular properties, and relying on SAPT interaction energy
symmetry of the functionalization as stated before, as well as calculations to obtain parameters that are related to the overlap of
possibly by the proximity of the chlorine atoms to the nitrogen charge densities. Using these novel force fields, we find excellent
atoms of the imidazolate ring. Comparing these important agreement with experimental CO2 adsorption isotherms for both
absorption sites in ZIF-71 to those in ZIF-8, we see the same of these systems, with both systems being dominated by disper-
correlation between dispersion dominated interactions and the sion interactions, and electrostatic interactions playing a smaller
preference for interaction with six ring aperture sites. We also role partially due to the symmetry of the functionalized imida-
note that the six ring sites have significantly greater adsorption zolate linker rings in each ZIF.
energies (a difference of ∼8 kJ/mol) in ZIF-71 compared to Overall, the approach outlined in the present work offers
ZIF-8, which might be expected due to the chlorine atoms on the several compelling advantages over the use of standard, empirical
imidazolate rings of ZIF-71. The fact that this enhancement is force fields. Most significantly, our methodology generates force
primarily due to dispersive interactions might be somewhat fields entirely from ab initio SAPT calculations with absolutely no
unexpected, however. In this context the ability to rigorously empirical parameters. Thus although the present work focuses on
separate different interaction components through the use of a two minimally functionalized ZIFs, the methodology outlined
physically decomposed force field is essential for the accurate here is entirely general and easily extensible to novel and/or
interpretation of the effects of functionalization. highly functionalized ZIFs (or even adsorption of other gases,
The observed semiquantitative to quantitative agreement with e.g., N2). In contrast, previous approaches have often relied on a
the corresponding experimental isotherms validates the accuracy patchwork of force fields or empirical scale factors based on
of our force fields and the general, parameter-free approach to postfacto comparison with experiment. Furthermore, our newly
generating them. In conjunction with our previous correspond- generated force fields are extremely accurate, yielding semiquan-
ing results for bulk CO2, they also reaffirm our broader metho- titative to quantitative accuracy, surpassing the level required for
dology for obtaining robust and physically motivated force field purposes of computational screening. Finally, we note that the
parameters. Although the ZIFs studied in this work are relatively physically motivated nature of these force fields admits a natural
“simple” (i.e., minimally functionalized), they provide proof of energy decomposition, expressing the interaction energy in
concept for the force field development methods that we have “chemical” terms such as exchange, electrostatics, polarization,
used. The resulting natural energy decomposition yields chemi- and dispersion. By allowing for the analysis of simulation results
cal insight into the properties of ZIFs that contribute to high CO2 under representative conditions in chemically intuitive terms, the
adsorption. Given the extensive tunability of ZIFs, this enhanced present developments will lay the groundwork for intelligent
understanding of physical interactions is crucial in order to synthesis and design of novel ZIFs for CO2 sequestration and
efficiently develop new and better systems for CCS. other gas adsorption applications.
1901 dx.doi.org/10.1021/jp209335y |J. Phys. Chem. C 2012, 116, 1892–1903
The Journal of Physical Chemistry C ARTICLE

’ ASSOCIATED CONTENT (17) Masciocchi, N.; Bruni, S.; Cariati, E.; Cariati, F.; Galli, S.; Sironi, A.
Inorg. Chem. 2001, 40, 5897.
bS Supporting Information. Figure comparing the electron (18) Huang, X. C.; Zhang, J. P.; Chen, X. M. Chin. Sci. Bull. 2003,
density of the imidazolate rings of the Zn2Im7Li63+ fragment 48, 1531.
and those of the corresponding periodic system. Figures show- (19) Zhang, H.-X.; Wang, F.; Yang, H.; Tan, Y.-X.; Zhang, J.; Bu, X.
ing SAPT energy fits for the Zn2Im7Li63+ fragment and a CO2 J. Am. Chem. Soc. 2011, 133, 11884.
molecule. Tables giving parameters used in CO2-ZIF-8 force (20) Zhang, J.; Wu, T.; Zhou, C.; Chen, S.; Feng, P.; Bu, X. Angew.
Chem. Int. Ed. Engl. 2009, 48, 2542.
field. Figures showing SAPT energy fits for a Li2mIm ring and a
(21) Lehnert, R.; Seel, F. Z. Anorg. Allgem. Chem. 1980, 464, 187.
CO2 molecule used to derive CO2-ZIF-8 force field parameters. (22) Sturm, M.; Brandl, F.; Engel, D.; Hoppe, W. Acta Crystallogr.
Tables giving parameters used in CO2-ZIF-71 force field. Sect. B 1975, 31, 2369.
Figures showing SAPT energy fits for a Li2dcIm ring and a (23) Masciocchi, N.; Ardizzoia, G. A.; Brenna, S.; Castelli, F.; Galli,
CO2 molecule used to derive CO2-ZIF-71 force field param- S.; Maspero, A.; Sironi, A. Chem. Commun. 2003, 2018.
eters. This material is available free of charge via the Internet at (24) Tian, Y. Q.; Chen, Z. X.; Weng, L. H.; Guo, H. B.; Gao, S.; Zhao,
http://pubs.acs.org. D. Y. Inorg. Chem. 2004, 43, 4631.
(25) Huang, X. C.; Zhang, J. P.; Lin, Y. Y.; Yu, X. L.; Chen, X. M.
’ AUTHOR INFORMATION Chem. Commun. 2004, 1100.
(26) Rettig, S. J.; Storr, A.; Summers, D. A.; Thompson, R. C.;
Author Contributions Trotter, J. J. Am. Chem. Soc. 1997, 119, 8675.
† (27) Rettig, S. J.; Sanchez, V.; Storr, A.; Thompson, R. C.; Trotter, J.
These authors contributed equally to this work.
J. Chem. Soc., Dalton Trans. 2000, 3931.
(28) Liu, Y. L.; Kravtsov, V. C.; Eddaoudi, M. Angew. Chem., Int. Ed.
’ ACKNOWLEDGMENT 2008, 47, 8446.
(29) Muller-Buschbaum, K. Z. Naturforsch. Sect. B 2006, 61, 792.
We thank Brian Laird for providing us with simulation data for (30) Fu, Y. M.; Zhao, Y. H.; Lan, Y. Q.; Wang, Y.; Qiu, Y. Q.; Shao,
various ZIFs (including ZIF-71). This research was supported by K. Z.; Su, Z. M. Inorg. Chem. Commun. 2007, 10, 720.
the National Science Foundation through TeraGrid resources (31) Lorente, M. A. M.; Dahan, F.; Sanakis, Y.; Petrouleas, V.;
provided by Purdue University and via Grant CHE-0840494.119 Bousseksou, A.; Tuchagues, J. P. Inorg. Chem. 1995, 34, 5346.
Additional computing resources were provided via the Center for (32) Han, J. Y.; Fang, J.; Chang, H. Y.; Dong, Y.; Liang, S. Acta
High Throughput Computing at the University of Wisconsin120 Crystallogr. Sect. E 2005, 61, M2667.
and by National Science Foundation Grant CHE-0840494. This (33) Lehnert, R.; Seel, F. Z. Anorg. Allgem. Chem. 1978, 444, 91.
research was supported by DOE (DE-FG02-09ER16059). (34) Spek, A. L.; Duisenberg, A. J. M.; Feiters, M. C. Acta Crystallogr.
Sect. C 1983, 39, 1212.
(35) Wu, T.; Bu, X. H.; Liu, R.; Lin, Z. E.; Zhang, J.; Feng, P. Y. Chem.
’ REFERENCES —Eur. J. 2008, 14, 7771.
(1) Banerjee, R.; Furukawa, H.; Britt, D.; Knobler, C.; O’Keeffe, M.; (36) Kusgens, P.; Rose, M.; Senkovska, I.; Frode, H.; Henschel, A.;
Yaghi, O. M. J. Am. Chem. Soc. 2009, 131, 3875. Siegle, S.; Kaskel, S. Microporous Mesoporous Mater. 2009, 120, 325.
(2) Banerjee, R.; Phan, A.; Wang, B.; Knobler, C.; Furukawa, H.; (37) Nune, S. K.; Thallapally, P. K.; Dohnalkova, A.; Wang, C. M.;
O’Keeffe, M.; Yaghi, O. M. Science 2008, 319, 939. Liu, J.; Exarhos, G. J. Chem. Commun. 2010, 46, 4878.
(3) Hayashi, H.; Cote, A. P.; Furukawa, H.; O’Keeffe, M.; Yaghi, (38) Perez-Pellitero, J.; Amrouche, H.; Siperstein, F. R.; Pirngruber,
O. M. Nat. Mater. 2007, 6, 501. G.; Nieto-Draghi, C.; Chaplais, G.; Simon-Masseron, A.; Bazer-Bachi,
(4) Morris, W.; Doonan, C. J.; Furukawa, H.; Banerjee, R.; Yaghi, D.; Peralta, D.; Bats, N. Chem.—Eur. J. 2010, 16, 1560.
O. M. J. Am. Chem. Soc. 2008, 130, 12626. (39) Venna, S. R.; Carreon, M. A. J. Am. Chem. Soc. 2010, 132, 76.
(5) Morris, W.; Leung, B.; Furukawa, H.; Yaghi, O. K.; He, N.; (40) Wu, H.; Zhou, W.; Yildirim, T. J. Phys. Chem. C 2009,
Hayashi, H.; Houndonougbo, Y.; Asta, M.; Laird, B. B.; Yaghi, O. M. 113, 3029.
J. Am. Chem. Soc. 2010, 132, 11006. (41) Rappe, A. K.; Casewit, C. J.; Colwell, K. S.; Goddard, W. A.;
(6) Park, K. S.; Ni, Z.; Cote, A. P.; Choi, J. Y.; Huang, R. D.; Skiff, W. M. J. Am. Chem. Soc. 1992, 114, 10024.
Uribe-Romo, F. J.; Chae, H. K.; O’Keeffe, M.; Yaghi, O. M. Proc. Natl. (42) Mayo, S. L.; Olafson, B. D.; Goddard, W. A. J. Phys. Chem. 1990,
Acad. Sci. U.S.A. 2006, 103, 10186. 94, 8897.
(7) Phan, A.; Doonan, C. J.; Uribe-Romo, F. J.; Knobler, C. B.; (43) Sirjoosingh, A.; Alavi, S.; Woo, T. K. J. Phys. Chem. C 2010,
O’Keeffe, M.; Yaghi, O. M. Acc. Chem. Res. 2010, 43, 58. 114, 2171.
(8) Wang, B.; Cote, A. P.; Furukawa, H.; O’Keeffe, M.; Yaghi, O. M. (44) Keskin, S. J. Phys. Chem. C 2011, 115, 800.
Nature 2008, 453, 207. (45) Krishna, R.; van Baten, J. M. J. Membr. Sci. 2010, 360, 323.
(9) Huang, X. C.; Lin, Y. Y.; Zhang, J. P.; Chen, X. M. Angew. Chem., (46) Liu, B.; Smit, B. J. Phys. Chem. C 2010, 114, 8515.
Int. Ed. 2006, 45, 1557. (47) Babarao, R.; Dai, S.; Jiang, D. E. J. Phys. Chem. C 2011,
(10) Tian, Y. Q.; Zhao, Y. M.; Chen, Z. X.; Zhang, G. N.; Weng, 115, 8126.
L. H.; Zhao, D. Y. Chem.—Eur. J. 2007, 13, 4146. (48) Rankin, R. B.; Liu, J. C.; Kulkarni, A. D.; Johnson, J. K. J. Phys.
(11) Wu, T.; Bu, X. H.; Zhang, J.; Feng, P. Y. Chem. Mater. 2008, Chem. C 2009, 113, 16906.
20, 7377. (49) Battisti, A.; Taioli, S.; Garberoglio, G. Microporous Mesoporous
(12) Liu, Y. L.; Kravtsov, V. C.; Larsen, R.; Eddaoudi, M. Chem. Mater. 2011, 143, 46.
Commun. 2006, 1488. (50) Liu, D. H.; Zheng, C. C.; Yang, Q. Y.; Zhong, C. L. J. Phys. Chem.
(13) Zhang, J. P.; Chen, X. M. Chem. Commun. 2006, 1689. C 2009, 113, 5004.
(14) Rettig, S. J.; Storr, A.; Summers, D. A.; Thompson, R. C.; (51) Assfour, B.; Leoni, S.; Seifert, G. J. Phys. Chem. C 2010,
Trotter, J. Can. J. Chem. 1999, 77, 425. 114, 13381.
(15) Tian, Y. Q.; Cai, C. X.; Ren, X. M.; Duan, C. Y.; Xu, Y.; Gao, S.; (52) Guo, H. C.; Shi, F.; Ma, Z. F.; Liu, X. Q. J. Phys. Chem. C 2010,
You, X. Z. Chem.—Eur. J. 2003, 9, 5673. 114, 12158.
(16) Tian, Y. Q.; Cai, C. X.; Ji, Y.; You, X. Z.; Peng, S. M.; Lee, G. H. (53) Han, S. S.; Choi, S. H.; Goddard, W. A. J. Phys. Chem. C 2010,
Angew. Chem., Int. Ed. 2002, 41, 1384. 114, 12039.

1902 dx.doi.org/10.1021/jp209335y |J. Phys. Chem. C 2012, 116, 1892–1903


The Journal of Physical Chemistry C ARTICLE

(54) Han, S. S.; Choi, S. H.; Goddard, W. A. J. Phys. Chem. C 2011, A. J.; McNicholas, S. J.; Meyer, W.; Mura, M. E.; Nicklass, A.; Palmieri,
115, 3507. P.; Pfluger, K.; Pitzer, R.; Reiher, M.; Schumann, U.; Stoll, H.; Stone,
(55) Liu, Y.; Liu, H. L.; Hu, Y.; Jiang, J. W. J. Phys. Chem. B 2009, A. J.; Tarroni, R.; Thorsteinsson, T.; M.Wang; Wolf, A. MOLPRO, a
113, 12326. package of ab initio programs, version 2009.1; see http://www.molpro.net.
(56) Zhou, M.; Wang, Q.; Zhang, L.; Liu, Y. C.; Kang, Y. J. Phys. (93) Perdew, J. P.; Burke, K.; Ernzerhof, M. Phys. Rev. Lett. 1996,
Chem. B 2009, 113, 11049. 77, 3865.
(57) Pantatosaki, E.; Pazzona, F. G.; Megariotis, G.; Papadopoulos, (94) Perdew, J. P.; Burke, K.; Ernzerhof, M. Phys. Rev. Lett. 1997,
G. K. J. Phys. Chem. B 2010, 114, 2493. 78, 1396.
(58) Hou, X. J.; Li, H. Q. J. Phys. Chem. C 2010, 114, 13501. (95) Dunning, T. H. J. Chem. Phys. 1989, 90, 1007.
(59) Potoff, J. J.; Siepmann, J. I. AIChE J. 2001, 47, 1676. (96) Weigend, F.; Ahlrichs, R. Phys. Chem. Chem. Phys. 2005,
(60) Harris, J. G.; Yung, K. H. J. Phys. Chem. 1995, 99, 12021. 7, 3297.
(61) Jorgensen, W. L.; Tiradorives, J. J. Am. Chem. Soc. 1988, (97) Williams, H. L.; Mas, E. M.; Szalewicz, K.; Jeziorski, B. J. Chem.
110, 1657. Phys. 1995, 103, 7374.
(62) Tafipolsky, M.; Schmid, R. J. Phys. Chem. B 2009, 113, 1341. (98) Stone, A. J. Chem. Phys. Lett. 1981, 83, 233.
(63) Tafipolsky, M.; Amirjalayer, S.; Schmid, R. J. Phys. Chem. C (99) Stone, A. J.; Alderton, M. Mol. Phys. 1985, 56, 1047.
2010, 114, 14402. (100) Ferenczy, G. G. J. Comput. Chem. 1991, 12, 913.
(64) Tafipolsky, M.; Amirjalayer, S.; Schmid, R. J. Comput. Chem. (101) Chipot, C.; Angyan, J. G.; Ferenczy, G. G.; Scheraga, H. A.
2007, 28, 1169. J. Phys. Chem. 1993, 97, 6628.
(65) Han, S. S.; Goddard, W. A. J. Am. Chem. Soc. 2007, 129, 8422. (102) Winn, P. J.; Ferenczy, G. G.; Reynolds, C. A. J. Phys. Chem. A
(66) Han, S. S.; Goddard, W. A. J. Phys. Chem. C 2008, 112, 13431. 1997, 101, 5437.
(67) Yu, K.; McDaniel, J. G.; Schmidt, J. R. J. Phys. Chem. B 2011, (103) Tang, K. T.; Toennies, J. P. J. Chem. Phys. 1984, 80, 3726.
115, 10054. (104) Thole, B. T. Chem. Phys. 1981, 59, 341.
(68) Jeziorski, B.; Moszynski, R.; Szalewicz, K. Chem. Rev. 1994, (105) Yu, H. B.; Whitfield, T. W.; Harder, E.; Lamoureux, G.;
94, 1887. Vorobyov, I.; Anisimov, V. M.; MacKerell, A. D.; Roux, B. J. Chem.
(69) Bukowski, R.; Sadlej, J.; Jeziorski, B.; Jankowski, P.; Szalewicz, K. Theory Comput. 2010, 6, 774.
J. Chem. Phys. 1999, 110, 3785. (106) Williams, G. J.; Stone, A. J. J. Chem. Phys. 2003, 119, 4620.
(70) Bukowski, R.; Szalewicz, K.; Chabalowski, C. F. J. Phys. Chem. A (107) Misquitta, A. J.; Stone, A. J. J. Chem. Theory Comput. 2008, 4, 7.
1999, 103, 7322. (108) Misquitta, A. J.; Stone, A. J.; Price, S. L. J. Chem. Theory
(71) Li, X.; Volkov, A. V.; Szalewicz, K.; Coppens, P. Acta Crystallogr. Comput. 2008, 4, 19.
Sect. D 2006, 62, 639. (109) Misquitta, A. J.; Stone, A. J. Mol. Phys. 2008, 106, 1631.
(72) Mas, E. M.; Szalewicz, K.; Bukowski, R.; Jeziorski, B. J. Chem. (110) Nayak, S. K.; Nooijen, M.; Bernasek, S. L.; Blaha, P. J. Phys.
Phys. 1997, 107, 4207. Chem. B 2001, 105, 164.
(73) Misquitta, A. J.; Welch, G. W. A.; Stone, A. J.; Price, S. L. Chem. (111) http://towhee.sourceforge.net/.
Phys. Lett. 2008, 456, 105. (112) Fairen-Jimenez, D.; Moggach, S. A.; Wharmby, M. T.; Wright,
(74) Podeszwa, R.; Bukowski, R.; Szalewicz, K. J. Phys. Chem. A P. A.; Parsons, S.; Duren, T. J. Am. Chem. Soc. 2011, 133, 8900.
2006, 110, 10345. (113) Duane, S.; Kennedy, A. D.; Pendleton, B. J.; Roweth, D. Phys.
(75) Podeszwa, R.; Bukowski, R.; Rice, B. M.; Szalewicz, K. Phys. Lett. B 1987, 195, 216.
Chem. Chem. Phys. 2007, 9, 5561. (114) Essmann, U.; Perera, L.; Berkowitz, M. L.; Darden, T.; Lee, H.;
(76) Henkelman, G.; Arnaldsson, A.; Jonsson, H. Comput. Mater. Sci. Pedersen, L. G. J. Chem. Phys. 1995, 103, 8577.
2006, 36, 354. (115) Lindan, P. J. D.; Gillan, M. J. J. Phys.: Condens. Matter 1993,
(77) Kresse, G.; Furthmuller, J. Phys. Rev. B 1996, 54, 11169. 5, 1019.
(78) Kresse, G.; Joubert, D. Phys. Rev. B 1999, 59, 1758. (116) Humphrey, W.; Dalke, A.; Schulten, K. J. Mol. Graph. 1996,
(79) Hess, B.; Kutzner, C.; van der Spoel, D.; Lindahl, E. J. Chem. 14, 33.
Theory Comput. 2008, 4, 435. (117) Wilmer, C. E.; Leaf, M.; Lee, C. Y.; Farha, O. K.; Hauser, B. G.;
(80) Van der Spoel, D.; Lindahl, E.; Hess, B.; Groenhof, G.; Mark, Hupp, J. T.; Snurr, R. Q. Nat. Chem, in press.
A. E.; Berendsen, H. J. C. J. Comput. Chem. 2005, 26, 1701. (118) Yazaydin, A. O.; Snurr, R. Q.; Park, T. H.; Koh, K.; Liu, J.;
(81) Hesselmann, A.; Jansen, G. Chem. Phys. Lett. 2002, 362, 319. LeVan, M. D.; Benin, A. I.; Jakubczak, P.; Lanuza, M.; Galloway, D. B.;
(82) Hesselmann, A.; Jansen, G. Chem. Phys. Lett. 2002, 357, 464. Low, J. J.; Willis, R. R. J. Am. Chem. Soc. 2009, 131, 18198.
(83) Hesselmann, A.; Jansen, G. Chem. Phys. Lett. 2003, 367, 778. (119) TeraGrid: Analysis of Organization, System Architecture, and
(84) Hesselmann, A.; Jansen, G. Phys. Chem. Chem. Phys. 2003, Middleware Enabling New Types of Applications; Catlett, C., Ed.; IOS
5, 5010. Press: Amsterdam, 2007.
(85) Hesselmann, A.; Jansen, G.; Schutz, M. J. Chem. Phys. 2005, (120) Litzkow, M.; Livney, M.; Mutka, M. Condor - A Hunter of Idle
122. Workstations; IEEE: New York, 1988.
(86) Williams, H. L.; Chabalowski, C. F. J. Phys. Chem. A 2001,
105, 646.
(87) Bukowski, R.; Podeszwa, R.; Szalewicz, K. Chem. Phys. Lett.
2005, 414, 111.
(88) Misquitta, A. J.; Jeziorski, B.; Szalewicz, K. Phys. Rev. Lett.
2003, 91.
(89) Misquitta, A. J.; Podeszwa, R.; Jeziorski, B.; Szalewicz, K.
J. Chem. Phys. 2005, 123.
(90) Misquitta, A. J.; Szalewicz, K. Chem. Phys. Lett. 2002, 357, 301.
(91) Misquitta, A. J.; Szalewicz, K. J. Chem. Phys. 2005, 122.
(92) Werner, H.-J.; Knowles, P. J.; Lindh, R.; Manby, F. R.; Schutz,
M.; Celani, P.; Korona, T.; Mitrushenkov, A.; Rauhut, G.; Adler, T. B.;
Amos, R. D.; Bernhardsson, A.; Berning, A.; Cooper, D. L.; Deegan,
M. J. O.; Dobbyn, A. J.; Eckert, F.; Goll, E.; Hampel, C.; Hetzer, G.;
€ oppl, C.; Liu, Y.; Lloyd, A. W.; Mata, R. A.; May,
Hrenar, T.; Knizia, G.; K

1903 dx.doi.org/10.1021/jp209335y |J. Phys. Chem. C 2012, 116, 1892–1903

You might also like