View Article Online / Journal Homepage / Table of Contents for this issue
Dynamic Article Links
RSC Advances
Cite this: RSC Advances, 2012, 2, 4382–4396
PAPER
www.rsc.org/advances
Metal–organic frameworks and related materials for hydrogen purification:
Interplay of pore size and pore wall polarity{
Michael Fischer, Frank Hoffmann and Michael Fröba*
Published on 30 March 2012. Downloaded on 11/26/2023 10:46:19 AM.
Received 5th December 2011, Accepted 29th February 2012
DOI: 10.1039/c2ra01239a
The separation of hydrogen from other weakly adsorbing gases is a topic of high industrial relevance.
Microporous materials, such as zeolites, metal–organic frameworks (MOFs), and nanoporous
molecular crystals, hold much promise as adsorbent materials for adsorption-based hydrogen
separation units. However, most experimental and theoretical studies that have been reported so far
have focused on relatively few gas mixtures (mainly CO2/H2 and CH4/H2). In this work, the suitability
of five materials (zeolite: silicalite; MOFs: Mg-formate, Zn(dtp), Cu3(btc)2; porous molecular crystal:
cucurbit[6]uril) for the adsorption-based separation of carbon monoxide/hydrogen and oxygen/
hydrogen mixtures is assessed using force-field based grand-canonical Monte Carlo simulations. The
simulations are employed to predict single-component and mixture isotherms, as well as adsorption
selectivities. Moreover, a detailed analysis of the solid-fluid interactions is carried out on an atomistic
level. The choice of materials is motivated by their structural properties: four systems contain
relatively narrow channels (diameters , 6.5 Å), but differ in pore wall composition and polarity. The
fifth system possesses coordinatively unsaturated metal sites, which can act as preferential adsorption
sites for some guest molecules. The role of electrostatic interactions is fundamentally different for the
two mixtures considered: for CO/H2 separation, the employment of polar adsorbents is beneficial due
to the enhanced electrostatic interaction with carbon monoxide. On the contrary, an increased
polarity of the pore wall tends to reduce the O2/H2 selectivity, because electrostatic interactions
favour hydrogen over oxygen due to its larger quadrupole moment. In general, materials with narrow
channels perform best in the separation of hydrogen from weakly adsorbing species, because the
dispersive interactions are maximized in the channels. Moreover, they provide little space for the
co-adsorption of hydrogen.
1. Introduction
Due to the potential of hydrogen as a ‘‘clean’’ energy carrier,
much scientific attention has been directed towards the development and optimization of hydrogen production, storage, and
utilization technologies.1 Currently, hydrogen is mostly produced by steam-reforming of methane or higher hydrocarbons.
For these large-scale plants, efficient technologies have been
developed to purify the produced hydrogen, making use of
pressure-swing adsorption (PSA).2 This technology relies on a
preferred adsorption of the impurities over hydrogen in porous
adsorbents, typically using a combination of activated carbons
and zeolites in the adsorbent bed. In order to use hydrogen as an
energy carrier to propel vehicles, it may be advantageous to
replace the production in large-scale chemical plants by other
Institute of Inorganic and Applied Chemistry, Department of Chemistry,
University of Hamburg, Martin-Luther-King-Platz 6, D-20146, Hamburg,
Germany. E-mail: froeba@chemie.uni-hamburg.de;
Fax: (+49)-40-42838-6348; Tel: (+49)-40-42838-3100
{ Electronic Supplementary Information (ESI) available. See DOI:
10.1039/c2ra01239a/
4382 | RSC Adv., 2012, 2, 4382–4396
technologies that can be implemented in smaller, decentralized
units, in order to avoid the necessity to transport the produced
hydrogen over large distances.1,3 In this context, it may be
necessary to develop new purification technologies that are able
to deliver hydrogen of the required purity in these small-scale
units. Both adsorption-based separation processes and membrane-based separation processes could be employed for the
removal of undesired byproducts from the hydrogen feed. In the
former case, the adsorption selectivity and the working capacity
are key quantities that define the suitability of a material.4,5 In
the latter case, the (ideal) membrane selectivity corresponds to
the product of the adsorption selectivity and the ratio of the selfdiffusivities.6,7 The following discussion will focus exclusively on
the adsorption selectivity, concentrating on the usage of
materials in adsorption-based processes. It should, however, be
mentioned that significant advances in the preparation of
membranes that could be employed in the separation of
hydrogen from other gases have also been reported.3,8–11
The development of new classes of ordered microporous
materials, namely metal–organic frameworks (MOFs)12 and
nanoporous molecular crystals (NMCs),13 has stimulated
This journal is ß The Royal Society of Chemistry 2012
Published on 30 March 2012. Downloaded on 11/26/2023 10:46:19 AM.
View Article Online
intensive research efforts aiming at the employment of these
materials in gas separation processes. Recent reviews have
summarized the most significant advances in the field of MOFs,
both from the experimental14–17 and the theoretical18 side. With
regard to hydrogen-containing mixtures, most attention has been
paid to the separation of CO2/H2 and CH4/H2 mixtures, whereas
experimental studies of CO/H2 or O2/H2 separation are very
scarce.
The direct experimental study of mixture adsorption requires
a tedious experimental setup.19 Therefore, most publications
report the measurement of single-component isotherms, and
then apply an appropriate model, such as ideal adsorbed solution
theory (IAST),20 to predict mixture adsorption isotherms that
permit a calculation of the adsorption selectivity. For example,
experimental measurements of single-component adsorption,
with a subsequent analysis using IAST, delivered high CO2/H2
and CO2/N2 selectivities for Mg-MOF-74, a system with
coordinatively unsaturated Mg sites.21,22 An even simpler,
formally exact approach to calculate the selectivity in the limit
of zero coverage is given by the ratio of the Henry constants.20
One of the few published experimental studies that explicitly
addresses CO/H2 mixture separation reports a Henry’s law
selectivity of 12.8 for an activated carbon material, and a much
higher selectivity of 125 for zeolite 5A.2
In contrast to the problems encountered in experiments, a
prediction of mixture adsorption isotherms using grand-canonical Monte Carlo (GCMC) simulations is straightforward. In
several modelling publications, this method was used to assess
the adsorption selectivity of various MOFs with regard to CO2/
H2 and/or CH4/H2 mixtures.23–32 While some of these papers
also include N2/H2 mixtures,25,27 the separation of hydrogen
from other weakly adsorbing species such as oxygen or carbon
monoxide has not yet been widely studied. This is quite
surprising, as the removal of these impurities constitutes an
important step in hydrogen production and, possibly, prior to
utilization (see below). A notable exception is the simulation
study by Karra and Walton:33 these authors performed GCMC
simulations to predict the CO/H2 selectivity of Cu3(btc)2, and
observed an important influence of the different electrostatic
properties of the molecules (dipolar carbon monoxide vs.
quadrupolar hydrogen) on the selectivity. The resulting CO/H2
adsorption selectivities ranged from 20 to 50 at pressures up to
20 bar, and were thus considerably higher than the selectivities
obtained in comparable GCMC studies of carbon-based systems,
e.g. carbon nanotubes, which typically range below 10.34
In comparison to MOFs, experimental studies of nanoporous
molecular crystals that aim at gas separation applications are as
yet relatively rare, particularly with regard to mixtures containing hydrogen. However, the potential of some of these materials
has been assessed with regard to a selective adsorption of CO2
over nitrogen,35 methane,36 and both carbon monoxide and
methane.37 A computational study of CO2/H2, CH4/H2, and
N2/H2 separation in the organic crystal 3,39,4,49-tetra-(trimethylsilylethynyl)biphenyl has been reported very recently.38
In this work, force-field based grand-canonical Monte Carlo
simulations are employed to predict the CO/H2 and O2/H2
adsorption selectivity of three MOFs, one porous molecular
crystal, and one all-silica zeolite (as a reference system from the
field of ‘‘classical’’ adsorbents). The removal of CO impurities
This journal is ß The Royal Society of Chemistry 2012
from the H2 gas feed is particularly important with regard to the
use of proton exchange membrane fuel cells, because even trace
amounts of carbon monoxide (as low as 10 ppm) can poison the
platinum catalyst. Moreover, CO is a byproduct of hydrogen
production processes.2 As mentioned above, carbon monoxide
has quite different electrostatic properties when compared to
hydrogen, having a small dipole moment and a relatively large
quadrupole moment, whereas hydrogen is only weakly quadrupolar. While the separation of oxygen from hydrogen is of
lesser importance for industrial processes (although its importance may increase when hydrogen generation from water
electrolysis becomes more widely used), it is of some interest
from a fundamental point of view: O2 is a weakly interacting
molecule with a very small quadrupole moment. Thus, O2/H2
constitutes a model system for the separation of a mixture of
hydrogen and another weakly adsorbed, non-polar species.
The following five adsorbents were considered in this study:
N silicalite, an all-silica zeolite with MFI topology and two
different types of channels;39
N a-Mg-formate, a MOF with very narrow, one-dimensional
channels;40
N Zn(dtp) (with H2dtp = 2,3-di-1H-tetrazole-5-ylpyrazine), a
MOF with a nitrogen-rich linker and helical channels;41
N Cu3(btc)2 (with H3btc = 1,3,5-benzene-tricarboxylic acid), a
MOF with a bimodal pore size distribution and coordinatively
unsaturated copper sites;42
N microporous cucurbit[6]uril (CB[6]), a molecular crystal in
which the three-dimensional arrangement of CB[6] macrocycles
generates one-dimensional, trigonal channels.43
All systems are visualized in the electronic supplementary
information{ (ESI). Four of these materials contain relatively
narrow channels. Clearly, the low free pore volumes of these
systems cause serious limitations with regard to the working
capacity, and, potentially, limit the applicability of these
materials in real separation processes. On the other hand, the
strong solid–fluid interactions in the narrow channels often lead
to large differences in affinity towards different molecules, and,
consequently, may permit high adsorption selectivities. In the
context of this study, these materials were chosen despite their
limited capacities, because the main scope is the detailed
assessment of the influence of the pore wall composition,
particularly varying pore wall polarity, on the separation
behaviour. Furthermore, Cu3(btc)2 was included in order to
assess whether coordinatively unsaturated metal sites have a
beneficial influence on the selectivity. In addition to the
prediction of adsorption selectivities for different conditions, a
strong emphasis is put on an analysis of the contribution of
different interaction types (dispersive vs. electrostatic interactions), as well as a detailed investigation of the interaction energy
distribution in the pores.
2. Models and methods
2.1 Computational details
Force-field based GCMC simulations of single-component (H2,
CO, O2) and binary mixture (CO/H2, O2/H2) adsorption at T =
298 K (room temperature) were carried out using the SORPTION
module included in the Accelrys ‘‘Materials Studio’’ package.44
The theoretical background of GCMC simulations is described in
RSC Adv., 2012, 2, 4382–4396 | 4383
Published on 30 March 2012. Downloaded on 11/26/2023 10:46:19 AM.
View Article Online
detail elsewhere.45,46 The simulations covered a pressure range
from 0.1 to 20 bar. In addition to equimolar mixtures, CO/H2 and
O2/H2 ratios of 1 : 9 were also considered. Furthermore,
calculations were carried out for a fixed total pressure of 1 bar,
the composition varying from 19 : 1 to 1 : 19. It should be pointed
out that actual hydrogen purification involves much more extreme
molar ratios, e.g. in the removal of trace amounts of carbon
monoxide. However, these conditions are difficult to model due to
the large number of required simulation steps.
The simulations of adsorption isotherms involved at least
2.5 million equilibration steps and 5 million production steps.
More steps were used for the calculation of mixture isotherms, or
in cases when the calculated isotherms indicated insufficient
equilibration. The reported uptake values, which are given in
mmol g21, correspond to absolute uptakes (nabs). Only for cases
were experimental high-pressure adsorption data is available,
the excess uptakes nexc were also calculated, using the constant
volume approximation, i.e. nexc = nabs 2 Vp?r. Here, Vp
corresponds to the free pore volume, which was calculated
according to a previously described procedure.47,48 The bulk gas
density r was obtained from the ideal gas law. While this might
appear as a rather crude approximation, the deviations with
respect to the actual gas density obtained from thermophysical
reference data49 were found to be negligible (,1.5%) for the
conditions considered.
For a given pressure, the adsorption selectivity a was
calculated from the binary mixture isotherm according to:50
a~
nabs ðAÞ=nabs ðBÞ
yðAÞ=yðBÞ
(1)
The numerator corresponds to the ratio of the amounts
adsorbed in the material (in molar units), and the denominator
corresponds to the ratio of the concentrations in the gas phase.
For an ideal gas mixture, this is the ratio of the partial pressures.
In addition to the adsorption isotherms, the Henry constants
were calculated for the three species for a temperature range
from 273 K to 373 K, using at least 50 million insertion steps.
The Henry constant, which is expressed in mmol g21 bar21, is
defined for the Henry’s law regime according to:
KH ~ lim
p?0
nabs
p
(2)
It has been shown that the ratio of the Henry constants is
identical to the adsorption selectivity in the limit of low loading
(Henry’s law selectivity).20 Therefore, the calculation of the
Henry constants provides an efficient means to estimate the
selectivity, and to comment on its evolution on changing
temperature.
Finally, simulations for H2, CO, and O2 using an increased
number of simulation steps (100 million production steps) were
carried out for a constant loading of one molecule per cell in
order to calculate the isosteric heat of adsorption (qst) at low
coverage. The contributions of dispersive and electrostatic
interactions to the total interaction were also derived from these
calculations. In addition, three-dimensional plots of the potential
energy were created, using a resolution of 0.25 Å.
4384 | RSC Adv., 2012, 2, 4382–4396
2.2 Models of fluid molecules
In all simulations, dispersive and electrostatic interactions were
taken into account. Dispersive interactions were modeled using a
Lennard-Jones (LJ) 12-6 potential. Parameters to represent the
interaction between different atom types were calculated using
Lorentz–Berthelot mixing rules. A cutoff radius of 12.5 Å was
employed for dispersive interactions. Electrostatic interactions
were modeled by assigning point charges to the atomic sites, and
Ewald summation was used to account for the periodicity of the
systems. The choice of the parameters used for the fluid molecules
is discussed in this subsection, and the parameters are tabulated in
the ESI (Tables S-1 and S-2{). Some key quantities for the three
fluid molecules are summarized in Table 1.
For the H2 molecule, a single-site model was used to represent
dispersive interactions. The parameters published by Buch were
employed in this united-atom description.51 This model has been
used in various earlier studies of hydrogen adsorption in MOFs
and related compounds.48,52–56 The LJ parameters, which are
placed at the center of mass of the H2 molecule, were combined
with a partial charge model consisting of three point charges, a
negative charge of 22q located at the center of mass, and
positive charges of +q located at the atomic positions. The exact
values of these charges were calculated from the quadrupole
moment of H2, as described in earlier work.48 It is commonly
agreed that quantum effects play a non-negligible role when
modelling hydrogen adsorption at low temperatures due to the
low mass of the H2 molecule. Appropriate correction schemes to
account for these effects have been described previously.48,53–56
At room temperature, however, the magnitude of the quantum
correction is negligible, and it is possible to use the Buch
parameters without further corrections.55
For carbon monoxide, a variety of Lennard-Jones parameter
sets have been published. The most simple model consists of a
single interaction site at the center of mass. Parameters for this
model were proposed by Gu et al.34 A two-site model with
identical parameters for both sites (located at the positions of the
two atoms) was derived by Stoll et al.57 Two-site models with
different parameters for the carbon and oxygen atom were
used by Piper et al.,58 and by Straub and Karplus.59 Further
complexity is added by the inclusion of electrostatic interactions,
as the CO molecule possesses both a dipole and a quadrupole
moment. The model of Stoll et al. approximates the electrostatic
properties using a point dipole, which is larger than the actual
dipole moment to account for the ‘‘missing’’ quadrupole
moment.57 Straub and Karplus proposed a model consisting of
three point charges located at the atom positions and at the
center of mass,59 whereas Piper et al. placed one of the charges at
the carbon position, and the other two charges at non-atomic
sites. For the comparison of potential models, the models of Stoll
et al., Piper et al., and Straub and Karplus were taken ‘‘as is’’.
Table 1 Kinetic diameter dkin, dipole moment m, quadrupole moment h,
polarizability a, critical temperature Tcrit, and critical pressure pcrit of the
three gases considered in this work61,14
H2
CO
O2
dkin (Å)
m (eÅ)
h (eÅ2)
a (Å3)
Tcrit (K)
2.89
3.69
3.47
—
0.0233
—
0.1288
20.5911
20.0975
0.787
1.953
1.562
32.98
132.85
154.58
pcrit (bar)
12.93
34.94
50.43
This journal is ß The Royal Society of Chemistry 2012
Published on 30 March 2012. Downloaded on 11/26/2023 10:46:19 AM.
View Article Online
Additionally, the LJ parameters proposed by Gu et al. were
combined with the point charges from the model of Straub and
Karplus to account for electrostatic interactions.
To assess the performance of the CO potential models, the
pressure–density relationship of carbon monoxide at room
temperature was derived from GCMC simulations of CO for
an empty cell of sufficient size. The results are shown in Fig. 1,
together with experimental data.49 Quite interestingly, the model
of Straub and Karplus underestimates the density at pressures
above 100 bar, while the model of Piper et al. significantly
overestimates the density at pressures above 50 bar. The other
two models show a similar performance, with a rather accurate
prediction up to 200 bar, and a tendency to underestimate the
density above this pressure. In total, the LJ parameters of Gu
et al. together with the point charges from the model of Straub
and Karplus perform best. Therefore, this model is used in all
calculations of carbon monoxide adsorption in this work.
For oxygen, a model consisting of two LJ sites at the atomic
positions, and three point charges (2q at the oxygen positions,
+2q at the center of mass) was proposed by Zhang and
Siepmann.60 This model was used in all simulations of oxygen
adsorption reported in this work. However, the magnitude of the
charges was slightly adjusted to better match the best available
quadrupole moment from the computational chemistry comparison and benchmark database (CCCBDB).61
2.3 Models of adsorbents
The structures of the five adsorbent materials were taken from
experimental data: silicalite,62 Mg-formate,63 Zn(dtp),41
Cu3(btc)2,42 and cucurbit[6]uril.43 Where necessary, the structures
were idealized by removing all remnant solvent molecules,
including the water molecules coordinated to the metal sites of
Cu3(btc)2. The unit cell of Cu3(btc)2 is sufficiently large to dispense
with a supercell. For the other systems, appropriately sized
supercells were employed: 2 6 2 6 2 supercells were constructed
for silicalite, Mg-formate, and Zn(dtp), whereas a 1 6 1 6 2
supercell was used for cucurbit[6]uril. The intramolecular cavities
of the CB[6] macrocycle, which are expected to be inaccessible for
adsorbed molecules, were blocked by non-interacting spheres to
avoid an artificial adsorption inside these cavities.
The LJ parameters to represent the framework oxygen atoms of
silicalite were derived from the parameters representing the oxygen–
methane interaction in the work of Dubbeldam et al.64 As it will be
shown below, these parameters provide for good agreement of
the calculated adsorption isotherms with experimental data. For
the other four adsorbents, the LJ parameters were taken from the
Universal Force Field (UFF).65 This force-field is very attractive
due to its broad applicability, as it provides parameters for the
whole periodic system. The UFF has been successfully employed in
several earlier GCMC studies of gas adsorption and separation in
MOFs.24–26,28–32,48,66 While the DREIDING force-field67 has also
frequently been used in modelling studies of Zn-based MOFs, it is
not a suitable alternative in this case, as it does not contain
parameters for magnesium and copper.
The partial charges to represent the electrostatic potential inside
the pores were derived from DFT calculations, using the ESP
method as described by Singh and Kollman.68 The computational
details of these calculations, as well as the obtained charge values,
are provided in the ESI.{ For Mg-formate, Cu3(btc)2, and
cucurbit[6]uril, the same ESP charges have already been used in
an earlier study.66 For silicalite, a small cluster containing 12 Si
atoms and 16 bridging oxygens was extracted from the periodic
structure, and saturated with hydroxyl groups. For Zn(dtp), a
[Zn3(dtp)(tetraz)6]22 (tetraz = tetrazolate) cluster was used as a
representative structural fragment.
In order to assess the pore size of the adsorbents, two
quantities were derived from calculations of the solvent
accessible volume using the ‘‘Atom Volumes and Surfaces’’ tool
included in Accelrys ‘‘Materials Studio’’.44 This tool uses the
insertion of a spherical probe molecule of a given diameter to
determine which areas of the structure can accommodate a
molecule of this size. By varying the probe molecule diameter
from 2.8 Å, the diameter of the smallest pore window, dw, as well
as the diameter of the largest pore, dlp (corresponding to the
diameter of the largest sphere that can be accommodated at any
point of the framework), were computed. While the former
quantity gives information on the size of a molecule that can
actually diffuse through the structure, the latter permits an
ordering of the solids according to their maximal pore size. It
should be noted that very similar quantities were derived by
Foster et al. for 165 zeolite framework types.69
The results of the calculations are given in Table 2. Mgformate is the system with the smallest maximal pore diameter.
Moreover, it has very narrow windows, whose diameter of 3.3 Å
is actually smaller than the kinetic diameter of oxygen and
carbon monoxide (Table 1). However, experimental studies have
shown that the pores of this material are accessible to both
Table 2 Diameter of the smallest pore window, dw, and diameter of the
largest pore, dlp, for all five adsorbents considered. The values were
determined by insertion of probe molecules of varying diameter. For
silicalite, the diameter of the windows bordering the straight channels
and the sinusoidal channels are given. For Cu3(btc)2, both quantities are
included for the small and the large pores
Fig. 1 Bulk properties of carbon monoxide at room temperature
obtained with different potential models. Experimental data from ref. 49.
This journal is ß The Royal Society of Chemistry 2012
dw (Å)
dlp (Å)
Silicalite
Mg-formate
Zn(dtp)
4.7/4.5
6.1
3.3
4.5
6.1
6.3
Cu3(btc)2
3.7/6.5
5.3/12.8
Cucurbit[6]uril
5.6
6.5
RSC Adv., 2012, 2, 4382–4396 | 4385
Published on 30 March 2012. Downloaded on 11/26/2023 10:46:19 AM.
View Article Online
nitrogen, which has a similar kinetic diameter as carbon
monoxide (dkin(N2) = 3.68 Å), and even methane (dkin(CH4) =
3.76 Å).40,63 For silicalite, the diameters of the narrowest
windows of the two channel types (straight channels running
along the b-axis and sinusoidal channels running along the
a-axis) are given, the straight channels being slightly wider. The
maximal pore size amounts to 6.1 Å. A very similar value of dlp =
6.3 Å is obtained for Zn(dtp). This value is only 0.2 Å larger than
the diameter of the narrowest pore window in this structure,
reflecting the homogeneous width of the helical channels. The
pore topology of cucurbit[6]uril is more difficult to assess using
the rather simple quantities employed here: while the maximal
pore diameter is even slightly larger than in Zn(dtp), the
structural arrangement of the CB[6] molecular units leads to
the formation of lateral cavities with a diameter of approximately 4 Å that are surrounded by numerous framework atoms.
Finally, Cu3(btc)2 has a bimodal pore size distribution, and dw
and dlp are given for both pore types in Table 2. With dlp =
12.8 Å, the maximal pore size of this material is a factor of two
larger than in all other materials considered in this study.
In contrast to the pore size, the pore wall polarity cannot
be assessed quantitatively in a comparably simple fashion.
Nevertheless, a reasoning based on chemical intuition provides a
pathway to rank the adsorbents according to their pore wall
polarity. The chemical bonds in silicalite are dominantly covalent,
leading to a low degree of charge separation and behaviour as a
hydrophobic, non-polar adsorbent.39,71 Compared to this material,
the bonds to the metal centers in the three MOFs are highly polar.
In Mg-formate, the density of the negatively polarized oxygen
atoms located at the pore walls is very high, as the linker molecule
is very short. In Zn(dtp), the presence of the heteroaromatic linker
contributes to the charge separation, and the pore walls are mainly
constituted by negatively polarized nitrogen atoms.41 On these
grounds, a similar degree of pore wall polarity can be expected for
these two systems. The situation is more complicated for Cu3(btc)2:
on the one hand, the coordinatively unsaturated copper sites are
both strongly positively polarized and well accessible to adsorbed
molecules, which is why they can act as preferential adsorption
sites for different species, such as CO and H2.72 On the other hand,
the phenyl ring of the linker has a relatively homogeneous charge
distribution, implying that a large fraction of the pore surface
exhibits a rather low polarity. Finally, the portals of the pumpkinshaped cucurbit[6]uril molecule are surrounded by negatively
polarized carbonyl groups. These polar groups, however, mainly
participate in the hydrogen bonds that build up the supramolecular
arrangement, whereas the pore walls of cucurbit[6]uril are mainly
decorated by weakly polar CH and CH2 groups. In summary,
silicalite and cucurbit[6]uril can be expected to act as relatively
non-polar adsorbents, while Mg-formate and Zn(dtp) have a high
pore wall polarity. The electrostatic potential in the pores of
Cu3(btc)2 is highly position-dependent.
adsorption at T = 298 K and pressures up to 20 bar were carried
out. The results are shown in Fig. 2, together with experimental
data.71,73 While the calculated oxygen and carbon monoxide
isotherms are in excellent agreement with experiment results, the
simulated hydrogen adsorption isotherm range is below the
published experimental values after the excess correction is
applied. Possibly, the magnitude of the excess correction, which
amounts to almost 50% of the total loading in the case of H2, is
overestimated for this compound, and it can be hypothesized
that the constant volume approximation loses its validity in this
system with narrow pores.
The uptake values at p = 1 bar and p = 20 bar are given in
Table 3, together with the isosteric heats of adsorption, and the
relative contributions of dispersive and electrostatic interactions.
Silicalite takes up very little H2, whereas appreciable amounts of
carbon monoxide and oxygen are adsorbed. The molar uptake of
both gases is quite similar, with CO being only slightly favoured.
This is also reflected by the calculated isosteric heats. The results
correspond reasonably well with the experimental values, which
amount to 6.0 kJ mol21 for H2, 16.7 kJ mol21 for CO, and
16.3 kJ mol21 for O2.71,73 The role of electrostatic interactions is
negligible for all three molecules, contributing only 5% to the
total potential energy for carbon monoxide, the most polar
molecule. This is in accordance with the experimental observation that silicalite acts as a non-polar adsorbent.71
From the results of the Henry constant and mixture isotherm
calculations, which are given in Fig. 3 and 4 and in Table 4, it is
apparent that the predicted selectivities of silicalite are modest for
both gas mixtures, with a # 20 for the CO/H2 mixture, and a # 12
for the O2/H2 mixture. The selectivities do not show any
pronounced dependence on pressure or composition, although a
slight decrease of the selectivity on increasing pressure is
observable for an equimolar CO/H2 mixture. For the case of a
CO/H2 mixture, experimental Henry’s law selectivities can be
derived from the Henry constants obtained by Golden and Sircar
for two different temperatures.71 The simulation results compare
very favourably with these data: at T = 305 K and T = 342 K the
experimental selectivities amount to a = 17.8 and a = 11.6,
respectively, whereas the simulations deliver a = 19.2 and a = 12.5.
The calculated potential energy maps for H2, CO, and O2 are
displayed in Fig. 5. Sections through both channel types are
shown. The highest value of the potential energy amounts to
approximately 7 kJ mol21 for hydrogen, 218 kJ mol21 for
carbon monoxide, and 215 kJ mol21 for oxygen, numbers that
are in good correspondence with the isosteric heats of adsorption. For all three species, the potential energy is most favourable
in the narrow areas where the channels pass through the rings
constituted by 10 silicon and 10 oxygen atoms. Compared to the
windows, the interaction energy is reduced considerably at the
wider channel intersections.
3. Results and discussion
3.2 Mg-formate
3.1 Silicalite
In order to assess the performance of the chosen parameter
combination for predictions of the adsorption of different gases,
GCMC simulations of oxygen adsorption at T = 306 K and
pressures up to 1 bar, and of hydrogen and carbon monoxide
4386 | RSC Adv., 2012, 2, 4382–4396
The calculated single-component adsorption isotherms are
shown in Fig. 6. Experimental data for the whole pressure range
are available for hydrogen, only.70 The simulation results are in
fair agreement with these data, with the calculated excess
isotherm ranging somewhat below the experimental hydrogen
loading. As discussed above, this could be related to a
This journal is ß The Royal Society of Chemistry 2012
Published on 30 March 2012. Downloaded on 11/26/2023 10:46:19 AM.
View Article Online
Fig. 2 Simulation results for silicalite: a) calculated oxygen adsorption isotherm at T = 306 K. Experimental data are shown for comparison (open
symbols).73 b) Calculated hydrogen, carbon monoxide, and oxygen adsorption isotherms at T = 298 K. For hydrogen and carbon monoxide,
experimental data (obtained at temperatures y300 K) are shown as open symbols.71 To enable a direct comparison of simulation results with
experimental data, the calculated excess isotherms are included as solid lines.
Table 3 Results of simulations of H2, CO, and O2 single-component adsorption: Gas uptakes n (in mmol g21), isosteric heats of adsorption qst derived
at low loading (in kJ mol21), and relative contributions of dispersive and electrostatic interactions to the total interaction energy
n(H2), p = 1 bar
n(H2), p = 20 bar
qst(H2)
Edisp/Ees(H2)
n(CO), p = 1 bar
n(CO), p = 20 bar
qst(CO)
Edisp/Ees(CO)
n(O2), p = 1 bar
n(O2), p = 20 bar
qst(O2)
Edisp/Ees(O2)
Silicalite
Mg-formate
Zn(dtp)
Cu3(btc)2
Cucurbit[6]uril
0.016
0.29
7.1
99%/1%
0.27
1.76
17.5
95%/5%
0.18
1.73
15.0
Ees , 1%
0.018
0.34
8.4
92%/8%
0.56
2.51
21.8
84%/16%
0.24
2.07
17.1
Ees , 1%
0.025
0.46
7.0
94%/6%
0.45
2.58
17.8
82%/18%
0.21
2.24
14.0
Ees , 1%
0.062
1.16
5.7
98%/2%
0.69
6.85
18.1
61%/39%
0.42
4.68
14.6
Ees , 1%
0.018
0.32
8.0
97%/3%
0.45
1.77
19.2
92%/8%
0.28
1.81
16.6
Ees , 1%
breakdown of the constant volume approximation in materials
with narrow pores. For the case of oxygen, experimental
measurements have been reported for pressures up to 1 bar.63
A comparison of the calculated isotherm with these values is
shown in the ESI,{ revealing a tendency to moderately
overestimate the amount of O2 adsorbed.
Comparing the different species, it is noteworthy that the
affinity of Mg-formate towards carbon monoxide is remarkably
higher than towards oxygen: at p = 1 bar, the amount of CO
adsorbed is more than twice as large as the amount of O2. Of the
five adsorbents considered in this work, Mg-formate exhibits the
highest isosteric heats of adsorption at low coverage (Table 3) for
all three gases. Experimental values have been reported for
hydrogen.70 Ranging between 6.5 and 7 kJ mol21, these values
are lower than the qst derived from the simulations. The
contribution of electrostatic interactions to the total potential
energy is insignificant for oxygen. For hydrogen and carbon
monoxide, the electrostatic contribution amounts to 8% and
16%, respectively, showing that the charge-quadrupole and (in
the case of CO) charge–dipole interactions in the narrow
channels of Mg-formate are responsible for a significant fraction
of the total interaction.
The evolution of a determined from the mixture isotherms
(Fig. 3 and 4) reveals that the selectivity does not exhibit a
This journal is ß The Royal Society of Chemistry 2012
pronounced dependency on pressure for both mixtures, although
a very slight increase of the CO/H2 selectivity with pressure is
observable. Similarly, the variations of a as a function of the
mixture composition do not show any clear trend. Overall, the
selectivity towards a CO/H2 mixture ranges above 30, and is thus
significantly higher than in silicalite. In contrast, the O2/H2
selectivity is only modest, and very similar to the selectivity
obtained for the zeolitic system.
The calculated potential energy maps are shown in Fig. 7. As
for silicalite, the calculated energy values in the regions of
strongest interaction are in good correspondence with the
isosteric heats of adsorption. For all three molecules, the maps
show elongated minima of the potential energy, connected by
window-like regions of weaker interaction. These apertures are
bordered by three formate moieties that point into the channel,
thereby narrowing the channel diameter to 3.3 Å. While the
potential energy at the windows is only slightly reduced for
hydrogen, the smallest molecule, it is approximately halved for
carbon monoxide. It is quite enlightening to note the differences
between Mg-formate and silicalite: in silicalite, the preferential
adsorption occurs in the narrowest areas of the channels, which
have a diameter of approximately 4.5 Å, because the overlap of
the interaction potentials from the surrounding framework
atoms is maximized in these regions. In contrast to this, the
RSC Adv., 2012, 2, 4382–4396 | 4387
Published on 30 March 2012. Downloaded on 11/26/2023 10:46:19 AM.
View Article Online
Fig. 3 CO/H2 selectivity: summary of results. a) Henry’s law selectivity. The vertical grey line marks T = 298 K. The CO/H2 selectivities of silicalite
and Zn(dtp) are virtually identical over the whole temperature range. b) Adsorption selectivities derived from binary mixture isotherm calculations for
varying compositions of the gas phase at p = 1 bar. The x-axis corresponds to the content of CO in relation to the total pressure. c) Adsorption
selectivities derived from binary mixture isotherm calculations for an equimolar CO/H2 mixture. d) Adsorption selectivities derived from binary mixture
isotherm calculations for a 1 : 9 CO/H2 mixture.
highest interaction energies in Mg-formate are reached in the
wider parts of the channels (which also have a diameter of 4.5 Å),
whereas the energy is reduced at the apertures. Because the pore
windows of Mg-formate are very narrow, a molecule that passes
through the window comes so close to some of the surrounding
atoms that the interatomic distance is smaller than the
equilibrium distance of the Lennard-Jones potential, thereby
leading to a reduction of the interaction energy on decreasing
distance. Ultimately, this effect would lead to a prevention of the
diffusion of CO molecules through the structure. However, as
discussed previously, it has been shown that the pores of Mgformate are fully accessible to molecules of similar size, maybe
due to a certain influence of structural flexibility.40,63 In some
other MOFs with very narrow pores with diameters ,4 Å, it has
been observed that the materials can adsorb appreciable
amounts of hydrogen, but very little carbon monoxide (at least
at T = 77 K, as no measurements at room temperature have been
reported).74–76 Such a size exclusion of CO could also be
exploited for separation applications.
4388 | RSC Adv., 2012, 2, 4382–4396
Finally, it must be remarked that each of the energetically
most favourable regions visible in Fig. 7 cannot be occupied by
more than one molecule for geometric reasons. It can be
imagined that the molecules may assume a relatively ordered
arrangement in these regions, depending on the energetically
preferred orientation, as it has been discussed in detail in
previous work for the case of acetylene.66 Particularly for carbon
monoxide, such an ordering could induce attractive electrostatic
interactions between adjacent CO molecules, which could
explain the slight increase of the CO/H2 selectivity at high CO
loadings.
3.3 Zn(dtp)
The calculated adsorption isotherms for Zn(dtp), displayed in
Fig. 6, show a similar evolution to those obtained for Mgformate. While a slightly lower uptake for CO and O2 is
observed at low pressures, the adsorption capacities at 20 bar are
somewhat higher for all gases due to the slightly larger pore
volume of Zn(dtp). The isosteric heats of adsorption are
This journal is ß The Royal Society of Chemistry 2012
Published on 30 March 2012. Downloaded on 11/26/2023 10:46:19 AM.
View Article Online
Fig. 4 O2/H2 selectivity: summary of results. a) Henry’s law selectivity. The vertical grey line marks T = 298 K. b) Adsorption selectivities derived
from binary mixture isotherm calculations for varying compositions of the gas phase at p = 1 bar. The x-axis corresponds to the content of O2 in
relation to the total pressure. c) Adsorption selectivities derived from binary mixture isotherm calculations for an equimolar O2/H2 mixture. d)
Adsorption selectivities derived from binary mixture isotherm calculations for a 1 : 9 O2/H2 mixture.
Table 4 Comparison of adsorption selectivities. The Henry’s law selectivities and the selectivities derived from binary mixture isotherm calculations
(equimolar composition) at two different pressures are given
KH(CO)/KH(H2)
a (CO/H2), p = 1 bar
a (CO/H2), p = 20 bar
KH(O2)/KH(H2)
a (O2/H2), p = 1 bar
a (O2/H2), p = 20 bar
Silicalite
Mg-formate
Zn(dtp)
Cu3(btc)2
Cucurbit[6]uril
21.1
21.2
18.6
11.6
11.6
11.5
35.1
30.5
31.8
14.0
12.2
12.1
21.9
20.6
15.8
9.1
8.9
9.0
12.7
12.1
10.4
7.7
7.2
5.6
38.2
35.0
23.8
17.7
17.6
17.8
significantly lower than for Mg-formate, and relatively similar to
the qst values obtained for silicalite. With an isosteric heat of
14.0 kJ mol21, the affinity towards oxygen is the lowest of all
systems considered. The relative contribution of electrostatic
interactions to the total energy is similar to Mg-formate, which is
in line with the aforementioned expectation that the two
frameworks have a similar polarity.
The calculated adsorption selectivities of Zn(dtp) (Fig. 3 and
4) reveal a relatively modest separation performance, despite the
presence of narrow channels in the structure. The CO/H2
This journal is ß The Royal Society of Chemistry 2012
selectivity is comparable to silicalite, with a # 20 at low and
intermediate coverages. In the case of an equimolar mixture, the
selectivity decreases significantly at pressures above 5 bar,
whereas only a slight decrease is observed for the 1 : 9 mixture.
At a constant pressure of 1 bar, the selectivity shows no
dependence on mixture composition. The O2/H2 selectivity is
lower than for silicalite, ranging near a # 9, and exhibits no
significant changes on varying pressure or composition.
The potential energy maps, shown in Fig. 7, reveal
pronounced energy minima in the lateral areas at the outside
RSC Adv., 2012, 2, 4382–4396 | 4389
Published on 30 March 2012. Downloaded on 11/26/2023 10:46:19 AM.
View Article Online
Fig. 5 Potential energy maps derived from simulations of hydrogen, carbon monoxide, and oxygen adsorption in silicalite. The straight channels
running along the b-axis are shown on the left-hand side (section || (100) plane), the sinusoidal channels running along the a-axis on the right-hand side
(section || (010) plane). The diameters of the framework atoms correspond to the van der Waals diameters. Different energy ranges are displayed for the
three molecules to visualize all features of the interaction energy distribution.
of the helical channels. The highest potential energies amount to
approximately 28 kJ mol21 for H2, 220 kJ mol21 for CO, and
215 kJ mol21 for O2, and are thus higher than the isosteric heats
of adsorption. This difference indicates that adsorption in other
regions of the framework, where the interaction is weaker, also
contributes to the total adsorption, even at low loadings. The
lateral areas are closely surrounded by three tetrazole and two
pyrazine rings, which leads to maximization of the dispersive
interactions in this region. The energy maps show that the
interaction with the framework is strongest in these areas for all
three molecules, regardless of the sign and magnitude of the
molecular quadrupole moment. Because the wall-to-wall
4390 | RSC Adv., 2012, 2, 4382–4396
distance along the c-axis amounts to approximately 4.5 Å in
the lateral areas, all molecules located within these areas are
small enough to assume an orientation that maximizes attractive
electrostatic interactions.
3.4 Cu3(btc)2
Due to its larger free pore volume, Cu3(btc)2 adsorbs much higher
amounts of the three gases than the other adsorbents considered
(Fig. 8). The hydrogen adsorption isotherm predicted by the
simulations is in excellent agreement with experimental data.56
The isosteric heat of hydrogen adsorption is slightly lower than
This journal is ß The Royal Society of Chemistry 2012
Published on 30 March 2012. Downloaded on 11/26/2023 10:46:19 AM.
View Article Online
Fig. 6 Calculated hydrogen, carbon monoxide, and oxygen adsorption isotherms for Mg-formate (left) and Zn(dtp) (right). For Mg-formate,
experimental hydrogen adsorption data are shown as open symbols.70 To enable a direct comparison, the excess H2 uptake derived from the simulations
is shown as a solid line.
typical experimental values, which range between 6 and 7 kJ mol21.
As far as the adsorption of carbon monoxide is concerned, a
relatively good agreement between the simulated CO isotherm and
experimental data is observed, with a tendency to underestimate the
loading.77 As it will be discussed later, this might be related to the
interaction of carbon monoxide with the unsaturated copper sites.
The isosteric heats of CO and O2 adsorption are lower than for
Mg-formate, but slightly higher when compared to Zn(dtp).
Interestingly, the contribution of electrostatic interactions is very
high for carbon monoxide, amounting to 40% of the total potential
energy. In contrast to this, electrostatic interactions are negligible
for both hydrogen and oxygen.
As it is visible from Fig. 3 and 4, respectively, the selectivities
towards both mixtures are the lowest of all systems considered:
at 1 bar, the CO/H2 selectivity amounts to a # 12, and the O2/H2
selectivity ranges around a # 7. A slight decrease of a on
increasing pressure is observed. It is more pronounced for the
equimolar mixture than for the 1 : 9 mixture. Moreover, a slight
dependence of the selectivity on gas phase composition is
detectable, with a decreasing on decreasing H2 content. These
observations are in line with the common expectation that the
selectivity decreases with increasing pressure, and with increasing
concentration of the more strongly adsorbed species.19 In this
context, it is interesting to note that similar GCMC simulations
by Karra and Walton predicted a sharp rise of the CO/H2
selectivity at p . 20 bar for a CO-rich and an equimolar mixture,
while the selectivity for an H2-rich mixture remained unaffected.33 This was rationalized with the complete occupation of
the cell by CO molecules, which prevented a significant coadsorption of hydrogen. While the conditions under which this
behaviour is most pronounced were not considered in the
computations reported here, there are no indications that a
comparable rise of the CO/H2 selectivity at high pressures could
be reproduced. The differences between the modelling results
reported in ref. 33, and those obtained in the context of this
study, are most probably related to a different choice of forcefield parameters (in the work of Karra and Walton, the model of
Piper et al. was used for CO,58 and the framework parameters
were specifically adjusted for each species). The impact of the
different choice of parameters is also reflected by a considerably
This journal is ß The Royal Society of Chemistry 2012
higher CO/H2 selectivity at low coverage reported by Karra and
Walton, who obtained a Henry’s law selectivity of a # 24, as
compared to a value of 13 calculated in this study.
The potential energy maps derived from the simulations are
displayed in Fig. 9. There are notable differences between the
results for hydrogen and oxygen on the one hand, and carbon
monoxide on the other hand. For H2 and O2, the interaction
strength is highest inside the small pores, which are surrounded by
four phenyl rings and six Cu2 paddle wheels. For both molecules,
the potential energies in these areas are 1 to 2 kJ mol21 higher than
the isosteric heats of adsorption. For carbon monoxide, the
central areas of the small pores also correspond to regions of
increased interaction. However, the interaction energies in these
regions are only slightly higher than for oxygen. More
pronounced, sharp energy minima are located at the unsaturated
copper centers, at a distance of approximately 2.7 Å from the Cu
atoms. The potential energy in these areas exceeds 225 kJ mol21,
and is thus drastically higher than the isosteric heat of adsorption.
Due to the small number of interaction partners at a similar
distance, the environment of the metal centers in Cu3(btc)2 is not
particularly favourable as long as only dispersive interactions are
considered. However, as pointed out in the description of the pore
wall polarity in the different systems, the copper centers are well
accessible, strongly positively polarized interaction sites. Clearly,
the preferential adsorption of CO at these sites must be attributed
to electrostatic (charge–dipole and charge–quadrupole) interactions. This interpretation is also corroborated by the large
contribution of electrostatic effects to the total energy. In this
context, it is worth noting that experimental evidence for a
relatively strong interaction of carbon monoxide molecules with
the copper centers has been obtained in a combined IR and
XANES study on CO-loaded samples.72
As a final point, it must be emphasized that no specific
adjustments were made in order to accurately represent the
interaction of CO with the copper centers. This is in contrast to
previous work, where the parameters representing the interaction
of these sites with hydrogen, acetylene, and carbon dioxide were
derived from density functional theory calculations.66,78
Although the potential energy distribution shown in Fig. 9
reveals that a strong, localized Cu–CO interaction is qualitatively
RSC Adv., 2012, 2, 4382–4396 | 4391
Published on 30 March 2012. Downloaded on 11/26/2023 10:46:19 AM.
View Article Online
Fig. 7 Potential energy maps derived from simulations of hydrogen, carbon monoxide, and oxygen adsorption in Mg-formate (left, section || (100)
plane) and Zn(dtp) (right, section || (010) plane). The coordinate systems displayed at the top of the figure indicate the approximate orientation of the
crystallographic axes with respect to the section.
reproduced in the simulations, this does not automatically imply
that the interaction strength predicted by the chosen parameters
is also quantitatively correct. It is quite likely that the differences
between the experimental and the calculated CO isotherm that
are visible in Fig. 8 are due to an inaccurate representation of the
Cu–CO interaction strength. Further computational work will be
necessary to address this issue in more detail. In this context, it
should be mentioned that a detailed computational study of the
interaction of carbon monoxide with the accessible metal sites of
Mg-MOF-74 has been reported by Valenzano et al.79
3.5 Cucurbit[6]uril
Of the three gases considered in this work, experimental
adsorption measurements on microporous cucurbit[6]uril have
been performed for carbon monoxide, only, for a pressure range
4392 | RSC Adv., 2012, 2, 4382–4396
up to 1 bar.37 The comparison of the experimental data with the
simulation results for this range is shown in the ESI.{ It is
noteworthy that the simulation overestimates the amount
adsorbed by a factor of 2 to 3. It cannot be elucidated whether
this discrepancy is related to experimental issues, or to problems
with an adequate description of the solid–fluid interactions in the
simulation. However, previous work has shown that the UFF
parameters together with ESP charges are able to deliver a
reliable prediction of the C2H2 and CO2 uptake in this system.66
Possibly, the diffusion of CO in the narrow channels is so limited
that some areas of the structure remain inaccessible in real
samples, thus leading to a reduced carbon monoxide uptake.
The simulated single-component isotherms for all gases are
shown in Fig. 8. At p = 1 bar, cucurbit[6]uril exhibits the highest
oxygen uptake of all systems except Cu3(btc)2, which has a much
larger free pore volume. At 20 bar, the storage capacities for all
This journal is ß The Royal Society of Chemistry 2012
Published on 30 March 2012. Downloaded on 11/26/2023 10:46:19 AM.
View Article Online
Fig. 8 Calculated hydrogen, carbon monoxide, and oxygen adsorption isotherms for Cu3(btc)2 (left) and cucurbit[6]uril (right). For Cu3(btc)2,
experimental hydrogen (ref. 56) and carbon monoxide (ref. 77) adsorption data are given as open symbols. To enable a direct comparison, the excess
CO and H2 uptakes derived from the simulations are shown as solid lines.
gases are very modest due to the limited pore volume. The isosteric
heats of adsorption are relatively high for hydrogen and oxygen,
ranging only 0.4 kJ mol21 below the corresponding values obtained
for Mg-formate. For carbon monoxide, however, the affinity is
considerably lower, with qst being 2.6 kJ mol21 lower than for Mgformate. Compared to the three MOFs, the contribution of
electrostatic interactions to the total potential energy is considerably reduced, but it is still slightly higher than for silicalite.
Interestingly, cucurbit[6]uril exhibits the highest Henry’s law
selectivities of all five systems towards both gas mixtures (Fig. 3
and 4). Although the increases in comparison to Mg-formate are
relatively modest in absolute terms, a relative increase of the
O2/H2 selectivity by 30% is quite remarkable. Similar to the
observations made for other systems, the selectivity towards a
CO/H2 mixture decreases on increasing total pressure, and on
increasing carbon monoxide content. For both compositions, the
drop in selectivity is most pronounced at low pressures. The
selectivity towards an O2/H2 mixture is practically unaffected by
total pressure and mixture composition.
Sections through the calculated potential energy are displayed
in Fig. 9. The energy maps reveal elongated regions of high
interaction strength in the lateral cavities of the channels. These
energy minima have an arrow-like shape, which is why the values
of highest energy are slightly displaced from the section. For each
visible minimum, there are two other minima in the same channel
at equal z-coordinates generated by the threefold rotation axis.
Each of these lateral cavities is surrounded by four CB[6] moieties,
two of which are approximately located in plane with the section
(and thus well visible in Fig. 9), whereas the other two lie above
and below the section. The cavities are mainly surrounded by CH
and CH2 groups, but there are also two carbonyl oxygens at
relatively close distance. For hydrogen and carbon monoxide, the
highest energy values are in good correspondence with the
isosteric heat of adsorption. For oxygen, however, the highest
energy values observed in the potential energy distribution exceed
219 kJ mol21, and are thus considerably higher than qst.
4. Discussion
Taken together, the observations made for the five different
systems provide some interesting insights into the structural
This journal is ß The Royal Society of Chemistry 2012
origins of the observed separation behaviour. As a general
observation, it is noteworthy that all systems with narrow
channels of diameters ¡6.5 Å exhibit a higher adsorption
selectivity than Cu3(btc)2, the only system with relatively large
pores. This can be explained straightforwardly with the increased
overlap of the dispersive atom–atom contributions stemming
from the pore walls. Because the enhancement of the dispersive
interactions is always more significant for the more strongly
adsorbed molecule, i.e. for carbon monoxide and oxygen,
respectively, the systems with narrow channels provide for a
higher selectivity over hydrogen. In addition to this general
observation, which is in line with well-known relationships, the
detailed analysis of the contribution of dispersive and electrostatic interactions, as well as the calculated potential energy
maps, permits more detailed insights into the structural origins
of the separation behaviour. The key features of each system will
be briefly discussed in the following.
Silicalite is the least polar of all the adsorbents considered, with
electrostatic interactions contributing less than 1 kJ mol21 for
carbon monoxide, the most polar molecule. The fact that the CO/
H2 selectivity is approximately a factor of 2 higher than the O2/H2
selectivity is therefore practically exclusively caused by the
stronger dispersive interaction with carbon monoxide, which is
due to the higher polarizability of CO (Table 1). Fig. 5 clearly
shows that CO and O2 are strongly favoured in the areas where the
channels pass through the 10-ring windows. However, the
differences in interaction strength in the wider intersections of
the channels are much less pronounced. This leads to a
considerable co-adsorption of H2, resulting in a modest selectivity.
In contrast to silicalite, a co-adsorption of the more weakly
adsorbing species is efficiently prevented in the case of Mgformate: as is visible from Fig. 7, the distribution of the
interaction energy in the accessible parts of the channels is
relatively homogeneous, and there are no regions of distinctly
weaker interaction where the co-adsorption could occur.
Therefore, the fact that the CO/H2 selectivity is considerably
increased when compared to silicalite can be explained with the
stronger dispersive interactions on the one hand (due to the
narrower pores), and the reduced co-adsorption of hydrogen on
the other hand. Both these explanations would also hold for the
O2/H2 mixture. However, the calculated selectivities are nearly
RSC Adv., 2012, 2, 4382–4396 | 4393
Published on 30 March 2012. Downloaded on 11/26/2023 10:46:19 AM.
View Article Online
Fig. 9 Potential energy maps derived from simulations of hydrogen, carbon monoxide, and oxygen adsorption in Cu3(btc)2 (left, section || (110) plane)
and cucurbit[6]uril (right, section || (110) plane).
identical for Mg-formate and silicalite. This qualitatively
different behaviour can be understood when electrostatic
contributions are considered: the pore walls of Mg-formate,
which are mainly constituted by oxygen atoms, are relatively
polar. Because the CO molecule is dipolar and has a larger
quadrupole moment than the H2 molecule, the electrostatic
interactions with the pore walls will lead to an increased affinity
for CO over H2. On the contrary, the quadrupole moment of O2
is smaller than the quadrupole moment of H2: while dispersive
interactions will favour oxygen due to its higher polarizability,
electrostatic interactions will favour hydrogen. Thus, electrostatic interactions actually have a negative effect on the O2/H2
adsorption selectivity, which is why Mg-formate and silicalite
exhibit practically identical selectivities despite the more favourable pore size of Mg-formate.
The pore walls of Zn(dtp) mainly consist of negatively
polarized nitrogen atoms, which provide for a comparable
4394 | RSC Adv., 2012, 2, 4382–4396
electrostatic contribution to the total interaction energy as in
Mg-formate. Fig. 7 reveals that the more strongly interacting
molecules are energetically favoured in lateral areas of the
channels, but that there are also extended regions at the channel
center where the interaction energy is considerably decreased. A
co-adsorption of hydrogen in these regions is possible, which is
why the selectivity remains modest. Again, it is quite interesting
to compare Zn(dtp) with its polar pore walls to the non-polar
silicalite, which has a slightly smaller maximal pore diameter: the
CO/H2 selectivity of both systems is nearly identical, because the
stronger dispersive interactions in Silicalite are compensated by
a higher electrostatic contribution in Zn(dtp). In contrast, the
O2/H2 selectivity of the more polar Zn(dtp) is lower, because the
O2/H2 selectivity decreases with increasing importance of
electrostatic interactions.
As mentioned above, the unsaturated metal sites of Cu3(btc)2
act as strongly preferred CO adsorption sites due to the
This journal is ß The Royal Society of Chemistry 2012
Published on 30 March 2012. Downloaded on 11/26/2023 10:46:19 AM.
View Article Online
electrostatic interactions of the carbon monoxide molecules with
the copper centers. Nonetheless, Cu3(btc)2 exhibits the lowest
adsorption selectivity of all systems towards both CO/H2 and
O2/H2 mixtures, because the large free pore volume permits the
co-adsorption of a considerable amount of hydrogen. When
aiming at applications in PSA processes, however, it must be
considered that the adsorption selectivity is not the only criterion
determining the optimal adsorbent material, but that the
working capacity is another important aspect. The large free
pore volume of Cu3(btc)2 leads to high total uptakes, and thus
renders this material most promising from the point of view of
the working capacity, which is determined by the difference in
uptake (‘‘delta loading’’) between the adsorption and desorption
pressure.32 While the actual value of the working capacity
depends on the process design (e.g. mixture composition,
pressure range),17 some estimations can be made from the
mixture isotherms calculated for equimolar compositions,
assuming an adsorption pressure of 20 bar and a desorption
pressure of 1 bar: for these conditions, the working capacity for
Cu3(btc)2 amounts to approximately 4 mmol g21 for CO and
2.5 mmol g21 for O2, whereas it ranges below 2 mmol g21 (for
CO) and below 1.5 mmol g21 (for O2) for all other adsorbents.
Finally, the porous molecular crystal cucurbit[6]uril exhibits a
CO/H2 selectivity that is comparable to Mg-formate, despite the
much weaker electrostatic interactions in this system, whose pore
walls, decorated mainly by CH and CH2 groups, have a low
polarity. The selectivity decreases rapidly on increasing loading,
which may be related to the fact that the interaction energy
shows relatively pronounced variations throughout the channels
(Fig. 9): at higher loadings, it becomes more and more probable
that some carbon monoxide molecules will occupy regions where
the interaction strength is not maximal, thereby reducing the
selectivity over hydrogen. In contrast to this, the interaction
energy in the channels of Mg-formate exhibits extended areas of
similar interaction strength, and there is no significant reduction
of the CO/H2 selectivity on increasing loading for the pressure
range considered. Concerning the separation of O2/H2 mixtures,
cucurbit[6]uril has the highest selectivity of all five adsorbents
considered. While dispersive interactions with the oxygen
molecules are maximized in the lateral cavities, electrostatic
interactions do not play a role. As discussed above, electrostatic
interactions favour hydrogen over oxygen, thereby decreasing
the O2/H2 selectivity. Therefore, the absence of these interactions
is a crucial factor that is responsible for the superior O2/H2
separation performance of cucurbit[6]uril when compared to the
other systems.
5. Conclusions
In this work, force-field based GCMC simulations have been
employed to predict the potential of five different adsorbent
materials for the separation of CO/H2 and O2/H2 mixtures. As a
general observation, it was found that higher adsorption
selectivities can be reached in materials with narrow pores,
whereas the presence of a relatively large available pore space
will inevitably lead to a significant co-adsorption of hydrogen,
reducing the selectivity. Therefore, Cu3(btc)2, the system with the
largest pores, exhibits the lowest selectivity for both mixtures,
despite the presence of unsaturated metal sites which interact
This journal is ß The Royal Society of Chemistry 2012
strongly with carbon monoxide. On the other hand, Cu3(btc)2 is
the material that could provide the highest working capacity of
all materials considered. For a real application, it is necessary to
find an adsorbent having the optimal balance between adsorption selectivity and working capacity in the pressure range of
interest.4 As it was highlighted recently, GCMC simulations
provide an efficient way to determine both quantities simultaneously at a moderate computational expense.32
Concerning the CO/H2 adsorption selectivity, all systems
except Cu3(btc)2 show a superior performance to BPL activated
carbon as a representative of typical carbon materials, for which
a Henry’s law selectivity of a = 12.8 has been obtained
experimentally.2 Mg-formate emerges as the material with the
most promising behaviour, maintaining a high selectivity of a .
30 over the whole pressure range. An analysis of the different
contributions of dispersive and electrostatic interactions, as well
as the interaction energy inside the pores was carried out. It
was found that the high selectivity arises from a favourable
combination of strong dispersive interactions in the narrow
pores, and strong electrostatic interactions of adsorbed CO
molecules with the polar pore walls. In order to develop
adsorbents with a further improved selectivity, materials that
combine narrow channels with specific sites that provide for a
localized interaction with carbon monoxide (unsaturated metal
sites, extra-framework cations) appear to be most promising. In
this context, it should be mentioned that a very high Henry’s law
selectivity of a = 125 has been reported from experimental
measurements on zeolite 5A.2 However, the evolution of the
selectivity on pressure was not included in the publication, and it
is possible that the behaviour at very low coverages is affected by
the adsorption at structural defects.
Finally, the results obtained for the adsorption of O2/H2
mixtures show that the adsorption selectivity decreases with
increasing polarity of the pore wall, because electrostatic
interactions favour hydrogen over oxygen (due to the larger
quadrupole moment of H2), whereas dispersive interactions
favour oxygen. Therefore, cucurbit[6]uril, a material with narrow
channels and non-polar pore walls, is the system that exhibits the
highest selectivity. From these results, it can be concluded that
the screening of potential materials for O2/H2 separation should
concentrate on non-polar adsorbents with narrow pores, such as
all-silica zeolites with very small pore diameters, or porous
molecular crystals that do not contain polar functional groups.
Acknowledgements
We thank Dr Garikoitz Beobide (Bilbao) for helpful comments.
Financial support by the DFG Interdisciplinary Graduate
School 611 ‘‘Design and Characterisation of Functional
Materials’’ is gratefully acknowledged.
References
1 N. Armaroli and V. Balzani, ChemSusChem, 2011, 4, 21–36.
2 S. Sircar and T. Golden, Sep. Sci. Technol., 2000, 35, 667–687.
3 J. W. Phair and S. P. S. Badwal, Sci. Technol. Adv. Mater., 2006, 7,
792–805.
4 R. Kumar, Ind. Eng. Chem. Res., 1994, 33, 1600–1605.
5 R. Krishna and J. R. Long, J. Phys. Chem. C, 2011, 115,
12941–12950.
6 R. Krishna and J. M. van Baten, Chem. Eng. J., 2007, 133, 121–131.
RSC Adv., 2012, 2, 4382–4396 | 4395
Published on 30 March 2012. Downloaded on 11/26/2023 10:46:19 AM.
View Article Online
7 R. Krishna and J. M. van Baten, J. Membr. Sci., 2010, 360, 323–333.
8 S. Adhikari and S. Fernando, Ind. Eng. Chem. Res., 2006, 45,
875–881.
9 N. W. Ockwig and T. M. Nenoff, Chem. Rev., 2007, 107, 4078–4110.
10 A. Huang, H. Bux, F. Steinbach and J. Caro, Angew. Chem., Int. Ed.,
2010, 49, 4958–4961.
11 M. Yu, H. H. Funke, R. D. Noble and J. L. Falconer, J. Am. Chem.
Soc., 2011, 133, 1748–1750.
12 G. Férey, Chem. Soc. Rev., 2008, 37, 191–214.
13 N. B. McKeown, J. Mater. Chem., 2010, 20, 10588–10597.
14 J.-R. Li, R. J. Kuppler and H.-C. Zhou, Chem. Soc. Rev., 2009, 38,
1477–1504.
15 S. Keskin, T. M. van Heest and D. S. Sholl, ChemSusChem, 2010, 3,
879–891.
16 B. Chen, S. Xiang and G. Qian, Acc. Chem. Res., 2010, 43,
1115–1124.
17 Y.-S. Bae and R. Q. Snurr, Angew. Chem., Int. Ed., 2011, 50,
11586–11596.
18 D. Liu and C. Zhong, J. Mater. Chem., 2010, 20, 10308–10318.
19 O. Talu, Chem. Ing. Tech., 2011, 83, 67–82.
20 A. L. Myers and J. M. Prausnitz, AIChE J., 1965, 11, 121–127.
21 Z. R. Herm, J. A. Swisher, B. Smit, R. Krishna and J. R. Long,
J. Am. Chem. Soc., 2011, 133, 5664–5667.
22 J. A. Mason, K. Sumida, Z. R. Herm, R. Krishna and J. R. Long,
Energy Environ. Sci., 2011, 4, 3030–3040.
23 Q. Yang and C. Zhong, J. Phys. Chem. B, 2006, 110, 17776–17783.
24 R. Babarao and J. Jiang, J. Am. Chem. Soc., 2009, 131, 11417–11425.
25 C. Xue, Q. Yang and C. Zhong, Mol. Simul., 2009, 35, 1249–1255.
26 J. Liu and J. K. Johnson, J. Low Temp. Phys., 2009, 157, 268–276.
27 S. Keskin and D. S. Sholl, Ind. Eng. Chem. Res., 2009, 48, 914–922.
28 S. Keskin, J. Phys. Chem. C, 2010, 114, 13047–13054.
29 R. Babarao, M. Eddaoudi and J. W. Jiang, Langmuir, 2010, 26,
11196–11203.
30 Y. Chen and J. Jiang, ChemSusChem, 2010, 3, 982–988.
31 B. Liu, C. Sun and G. Chen, Chem. Eng. Sci., 2011, 66, 3012–3019.
32 R. Krishna and J. M. van Baten, Phys. Chem. Chem. Phys., 2011, 13,
10593–10616.
33 J. R. Karra and K. S. Walton, Langmuir, 2008, 24, 8620–8626.
34 C. Gu, G.-H. Gao, Y.-X. Yu and T. Nitta, Fluid Phase Equilib., 2002,
194–197, 297–307.
35 Y. Jin, B. A. Voss, A. Jin, H. Long, R. D. Noble and W. Zhang,
J. Am. Chem. Soc., 2011, 133, 6650–6658.
36 M. Mastalerz, M. W. Schneider, I. M. Oppel and O. Presly, Angew.
Chem., Int. Ed., 2011, 50, 1046–1051.
37 H. Kim, Y. Kim, M. Yoon, S. Lim, S. M. Park, G. Seo and K. Kim,
J. Am. Chem. Soc., 2010, 132, 12200–12202.
38 W. Li, J. Zhang, H. Guo and G. Gahungu, J. Phys. Chem. C, 2011,
115, 4935–4942.
39 E. M. Flanigen, J. M. Bennett, R. W. Grose, J. P. Cohen, R. L.
Patton, R. M. Kirchner and J. V. Smith, Nature, 1978, 271, 512–516.
40 J. A. Rood, B. C. Noll and K. W. Henderson, Inorg. Chem., 2006, 45,
5521–5528.
41 J.-R. Li, Y. Tao, Q. Yu, X.-H. Bu, H. Sakamoto and S. Kitagawa,
Chem.–Eur. J., 2008, 14, 2771–2776.
42 S. S.-Y. Chui, S. M.-F. Lo, J. P. H. Charmant, A. G. Orpen and I. D.
Williams, Science, 1999, 283, 1148–1150.
43 S. Lim, H. Kim, N. Selvapalam, K.-J. Kim, S. J. Cho, G. Seo and K.
Kim, Angew. Chem., Int. Ed., 2008, 47, 3352–3355.
44 Materials Studio Version 5.5, Accelrys Inc., San Diego, 2011.
45 M. P. Allen and D. J. Tildesley. Computer simulation of liquids,
Clarendon Press, Oxford, UK, 1st edn, 1987.
46 D. Frenkel and B. Smit, Understanding molecular simulation: from
algorithms to applications, Academic Press, San Diego, California,
USA, 2nd edn, 2002.
4396 | RSC Adv., 2012, 2, 4382–4396
47 A. L. Myers, J. A. Calles and G. Calleja, Adsorption, 1997, 3,
107–115.
48 M. Fischer, F. Hoffmann and M. Fröba, ChemPhysChem, 2009, 10,
2647–2657.
49 E. W. Lemmon, M. O. McLinden and D. G. Friend, in NIST
Chemistry WebBook, NIST standard reference database number 69,
ed. P. J. Linstrom and W. G. Mallard, National Institute of
Standards and Technology, Gaithersburg MD, 2010.
50 S. Keskin and D. S. Sholl, Langmuir, 2009, 25, 11786–11795.
51 V. Buch, J. Chem. Phys., 1994, 100, 7610–7629.
52 R. F. Cracknell, Phys. Chem. Chem. Phys., 2001, 3, 2091–2097.
53 G. Garberoglio, A. I. Skoulidas and J. K. Johnson, J. Phys. Chem. B,
2005, 109, 13094–13103.
54 S. E. Wenzel, M. Fischer, F. Hoffmann and M. Fröba, Inorg. Chem.,
2009, 48, 6559–6565.
55 B. Kuchta, L. Firlej, R. Cepel, P. Pfeifer and C. Wexler, Colloids
Surf., A, 2010, 357, 61–66.
56 J. Liu, J. T. Culp, S. Natesakhawat, B. C. Bockrath, B. Zande, S. G.
Sankar, G. Garberoglio and J. K. Johnson, J. Phys. Chem. C, 2007,
111, 9305–9313.
57 J. Stoll, J. Vrabec and H. Hasse, J. Chem. Phys., 2003, 119,
11396–11407.
58 J. Piper, J. A. Morrison and C. Peters, Mol. Phys., 1984, 53,
1463–1480.
59 J. E. Straub and M. Karplus, Chem. Phys., 1991, 158, 221–248.
60 L. Zhang and J. I. Siepmann, Theor. Chem. Acc., 2006, 115, 391–397.
61 R. D. Johnson III, NIST Computational Chemistry Comparison and
Benchmark Database, NIST Standard Reference Database Number
101, Release 15b, 2011, cccbdb.nist.gov/.
62 H. van Koningsveld, H. van Bekkum and J. C. Jansen, Acta
Crystallogr., Sect. B: Struct. Sci., 1987, 43, 127–132.
63 D. G. Samsonenko, H. Kim, Y. Sun, G.-H. Kim, H.-S. Lee and K.
Kim, Chem.–Asian J., 2007, 2, 484–488.
64 D. Dubbeldam, S. Calero, T. J. H. Vlugt, R. Krishna, T. L. M.
Maesen, E. Beerdsen and B. Smit, Phys. Rev. Lett., 2004, 93, 088302.
65 A. K. Rappe, C. J. Casewit, K. S. Colwell, W. A. Goddard and
W. M. Skiff, J. Am. Chem. Soc., 1992, 114, 10024–10035.
66 M. Fischer, F. Hoffmann and M. Fröba, ChemPhysChem, 2010, 11,
2220–2229.
67 S. L. Mayo, B. D. Olafson and W. A. Goddard, J. Phys. Chem., 1990,
94, 8897–8909.
68 U. C. Singh and P. A. Kollman, J. Comput. Chem., 1984, 5, 129–145.
69 M. D. Foster, I. Rivin, M. M. J. Treacy and O. Delgado Friedrichs,
Microporous Mesoporous Mater., 2006, 90, 32–38.
70 B. Schmitz, I. Krkljus, E. Leung, H. W. Höffken, U. Müller and M.
Hirscher, ChemSusChem, 2010, 3, 758–761.
71 T. C. Golden and S. Sircar, J. Colloid Interface Sci., 1994, 162,
182–188.
72 C. Prestipino, L. Regli, J. G. Vitillo, A. Damin, F. Bonino, A.
Zecchina, C. Lamberti, P. L. Solari, K. O. Kongshaug and S.
Bordiga, Chem. Mater., 2006, 18, 1337–1346.
73 J. A. Dunne, R. Mariwala, M. Rao, S. Sircar, R. J. Gorte and A. L.
Myers, Langmuir, 1996, 12, 5888–5895.
74 M. Dinca and J. R. Long, J. Am. Chem. Soc., 2005, 127, 9376–9377.
75 S. Ma, X.-S. Wang, C. D. Collier, E. S. Manis and H.-C. Zhou, Inorg.
Chem., 2007, 46, 8499–8501.
76 S. Ma, D. Yuan, X.-S. Wang and H.-C. Zhou, Inorg. Chem., 2009,
48, 2072–2077.
77 J. R. Karra and K. S. Walton, J. Phys. Chem. C, 2010, 114,
15735–15740.
78 M. Fischer, B. Kuchta, L. Firlej, F. Hoffmann and M. Fröba,
J. Phys. Chem. C, 2010, 114, 19116–19126.
79 L. Valenzano, B. Civalleri, S. Chavan, G. T. Palomino, C. O. Areán
and S. Bordiga, J. Phys. Chem. C, 2010, 114, 11185–11191.
This journal is ß The Royal Society of Chemistry 2012