Langmuir 1999, 15, 533-544 533

A Molecular Model for Adsorption of Water on Activated

Carbon: Comparison of Simulation and Experiment
C. L. McCallum,† T. J. Bandosz,‡ S. C. McGrother,† E. A. Müller,§ and
K. E. Gubbins*,⊥
School of Chemical Engineering, Cornell University, Ithaca, New York 14853,
Chemistry Department, City College of New York, Convent Avenue at 138th Street,
New York, New York 10031, Departamento de Termodinámica y Fenómenos de Transferencia,
Universidad Simón Bolı́var, Caracas 1080-A, Venezuela, and Department of Chemical
Engineering, North Carolina State University, Raleigh, North Carolina 27695

Received May 19, 1998. In Final Form: September 28, 1998

Experimental and molecular simulation results are presented for the adsorption of water onto activated
carbons. The pore size distribution for the carbon studied was determined from nitrogen adsorption data
using density functional theory, and the density of acidic and basic surface sites was found using Boehm
and potentiometric titration. The total surface site density was 0.675 site/nm2. Water adsorption was
measured for relative pressures P/P0 down to 10-3. A new molecular model for the water/activated carbon
system is presented, which we term the effective single group model, and grand canonical Monte Carlo
simulations are reported for the range of pressures covered in the experiments. A comparison of these
simulations with the experiments show generally good agreement, although some discrepancies are noted
at very low pressures and also at high relative pressures. The differences at low pressure are attributed
to the simplification of using a single surface group species, while those at high pressure are believed to
arise from uncertainties in the pore size distribution. The simulation results throw new light on the
adsorption mechanism for water at low pressures. The influence of varying both the density of surface sites
and the size of the graphite microcrystals is studied using molecular simulation.

1. Introduction reduction, only source (a) above is important, and the

The adsorption behavior of water on activated carbons
is qualitatively different from that of simple fluids, such
as nitrogen, carbon dioxide, or hydrocarbons. This major
difference arises from two sources: (a) for water the fluid-
fluid interaction is much more strongly attractive than
the interaction of water with the carbon surface, in contrast
to the situation for simpler fluids, where the reverse is
the case, and (b) the adsorption behavior for water is
largely controlled by the formation of H-bonds with
oxygenated groups on the surface. Thus, the pore-filling
mechanism for water in activated carbons is distinctly
different from that for most fluids. For simple fluids pore
filling occurs via the formation of a fluid monolayer on the
surface, often followed by a second and possibly further
layers, prior to capillary condensation and pore filling. By
contrast, water molecules first adsorb onto oxygenated
surface sites, and these adsorbed water molecules then
act as nuclei for the formation of larger water clusters;
eventually these clusters connect, either along the surface
or across the pore, and pore filling usually occurs. When
the density of oxygenated sites on the surface is ap-
preciable, the pore filling seems to occur by a continuous
filling process without capillary condensation.
In this paper we address the problem of modeling this
adsorption process at the molecular level. For graphitic
carbons, in which all oxygenated groups have been
removed from the surface through heat treatment or
* To whom correspondence should be addressed.
† Cornell University. Current addresses: C. L. McCallum, Intel
Corporation, 5000 W. Chandler Blvd., Chandler, AZ 85226. S. C.
McGrother, Department of Chemical Engineering, North Carolina
State University, Raleigh, NC 27695.
‡ City College of New York.
§ Simon Bolivar University.
⊥ North Carolina State University.
modeling is relatively straightforward. Such systems have
been successfully modeled using both point charge and
off-center square well models of water in slit graphitic
pores.1-4 Agreement with experiment is good.4 In these
cases there is almost no adsorption of water at low-to-
moderate relative pressures, but there is an onset of pore
filling that occurs suddenly at high relative pressures.
The adsorption behavior is of type V in the IUPAC
classification.5 Such type V behavior can be described even
with simple Lennard-Jones interactions,6 and requires
only a fluid-fluid interaction that is strongly attractive
compared to the fluid-solid interaction.
Activated carbons, in which a significant density of
oxygenated sites are present on the carbon surfaces, are
considerably harder to model and to characterize. Mac-
roscopic phenomenological models have been proposed7,8
which give a reasonable fit to the experimental data
outside the low-pressure region, but these give little insight
into the underlying molecular mechanisms of adsorption
and are mainly useful for data interpolation. A limited
number of attempts to model water on activated carbons
at the molecular level through molecular simulation have
been reported. Segarra and Glandt9 modeled the activated
carbon as made up of graphite platelets having dipoles
(1) Antonchenko, V. Y.; Davidov, A. S.; Ilyin, V. V. Physics of Water;
Naukova Dumka: Kiev, 1991.
(2) Ulberg, D. E.; Gubbins, K. E. Mol. Simul. 1994, 13, 205; Mol.
Phys. 1995, 84, 1139.
(3) Maddox, M.; Ulberg, D. E.; Gubbins, K. E. Fluid Phase Equilibr.
1995, 104, 145.
(4) Müller, E. A.; Rull, L. F.; Vega, L. F.; Gubbins, K. E. J. Phys.
Chem. 1996, 100, 1189.
(5) Sing, K. S. W.; Everett, D. H.; Haul, R. A. W.; Moscou, L.; Pierotti,
R. A.; Rouquérol, J.; Siemineiewska, T. Pure Appl. Chem. 1985, 57, 603.
(6) Balbuena, P. B.; Gubbins, K. E. Langmuir 1993, 9, 1801.
(7) Dubinin, M. M.; Serpinsky, V. V. Carbon 1981, 19, 402.
(8) Talu, O.; Meunier, F. AIChE J. 1996, 42, 809.
(9) Segarra, E. I.; Glandt, E. D. Chem. Eng. Sci. 1994, 49, 2953.

10.1021/la9805950 CCC: $18.00 © 1999 American Chemical Society

Published on Web 12/15/1998
534 Langmuir, Vol. 15, No. 2, 1999 McCallum et al.

smeared out around the edges of the platelets to represent

the surface functional groups. More recent work10 suggests
that this model of the surface sites is insufficiently
inhomogeneous to predict the correct trends in low-
pressure adsorption and in heats of adsorption. Maddox
et al.3 and Müller et al.4 have used models in which the
pores are approximated by slits with graphite walls, on
which are embedded surface sites. In the work of Maddox
et al. the sites were assumed to consist of -COOH groups
and were modeled using optimized potentials for liquid
simulations (OPLS),11 while water was modeled using a
TIP4P potential model;12 these potential models consist
of Lennard-Jones terms for the dispersion and overlap
interactions, plus Coulomb interactions between discrete
charges placed in the molecules to mimic the H-bonding.
Surface sites were placed in a regular array on the graphite
surface. The adsorption isotherms of water at 298 K in a
2 nm pore with various site densities were obtained. A
more detailed study was carried out by Müller et al.4 for
a wide range of surface site densities at 300 K. In their
work the H-bonding between water molecules and between
water and a surface site was modeled using off-center
square well potentials in place of charges. This approach
had two advantages. First, the square well potential is
short-ranged, thus eliminating the need for time-consum-
ing Ewald lattice sums to correct for long-range Coulomb
interactions; this greatly speeds up the simulations.
Second, the square well model allows for highly localized
attractive interactions, which may more realistically model Figure 1. Two possible mechanisms for water adsorption onto
H-bonding. Results were obtained for both regular arrays activated carbon walls, for an intermediate site density. Scheme
of surface sites, and for randomly placed sites. Cooperative A corresponds to sites regularly arranged, and therefore spaced
well apart. Scheme B shows a random array at the same site
adsorption effects were found to be very important, with density. Clearly cooperative bonding is more prevalent in the
water clusters forming readily in surface regions where latter scheme. The schematic adsorption isotherm shown for
two or more sites were placed at a separation suitable for Scheme A assumes that the water-surface site bonding energy
both water-site and water-water bonding. This model is greater than the water-water value, i.e., HB HB
sf > ff . If the
captures the effects of the strongly orientation-dependent, HB HB
reverse holds, i.e., ff > sf , the slope in region I will be less
short-range H-bonds, as well as the heterogeneous nature than that in region II for scheme A. Shaded circles are water
of the surface. However, the intermolecular potential molecules, with newly adsorbed molecules shaded darker; open
parameters used by Müller et al. were not optimized for circles are surface-active sites.
water/activated carbon systems, and no comparison with
experimental data was made. This was due in part to a work we have therefore tried to capture the essential
lack of appropriate data. Although a considerable amount features of hydrogen bonding of water with the surface
of data on water/carbon systems has been published, there sites and with other water molecules, the strongly
are little data in the critically important low-pressure heterogeneous nature of the surfaces, and the distribution
region for well-characterized materials. of pore sizes. Simplifications introduced include the use
In this paper we (a) report new experimental data for of slit-shaped pores, the assumption that the surface sites
water adsorption at low pressure on a carefully charac- can be represented as being of a single species with a
terized carbon, (b) describe a molecular model of water on single H-bonding energy, the use of a single graphite
activated carbons that we believe is more realistic than microcrystal size, and the use of simulations for a few
the ones3,4,9 used previously, and (c) carry out molecular pore widths, together with interpolation between these,
simulations for this model, and compare these model to represent the continuous pore size distribution of the
results with the experimental ones. It should be empha- real carbon.
sized that the complexity of the system is such that some Before describing the methods and results, it is useful
simplifications in the model are essential at this stage. to speculate on the possible adsorption mechanisms for
Complexity in the actual physical system includes poorly this system, particularly at low pressure. For simplicity,
characterized materials, a fairly wide range of pore sizes we assume that only one water molecule can adsorb at a
and shapes, and a lack of precise knowledge of the species given site. Two possible scenarios are illustrated in Figure
and location of the surface groups. Added to these problems 1, for the case of an intermediate site density. In the
are the difficulty of the molecular simulations in a system simplest scenario, A, it is assumed that the surface sites
in which complex arrays of H-bonds are formed; excep- (open circles) are arranged in a regular array on the
tionally long runs on large systems are needed to overcome surface. At the lowest pressures single water molecules
ergodic problems and ensure results that are representa- will adsorb on individual surface sites (stage I), with
tive of a macroscopic system. Such studies are at the limits bonding energy HB sf . As the bulk gas pressure is in-
of capability of current supercomputers. In the current creased, more surface sites are occupied, until at a pressure
corresponding to stage II all sites are occupied by a single
(10) Gordon, P. A.; Glandt, E. D. Langmuir 1997, 13, 4659. water molecule. The pressure region up to stage II is the
(11) Briggs, J. M.; Nguyen, T. B.; Jorgensen, W. L. J. Phys. Chem. Henry’s law region for this heterogeneous system, and
1991, 95, 3315.
(12) Jorgensen, W. L.; Chandrasekhar, J.; Madura, J. D. J. Chem. the adsorption isotherm will be linear with a slope
Phys. 1983, 79, 926. characterized by the value of HB sf . As the pressure is
Adsorption of Water on Activated Carbon
increased further, water molecules will adsorb by forming
H-bonds to preadsorbed water molecules (stage III); the
bonding energy will be HB ff , producing a second linear
region of the adsorption isotherm, with a slope somewhat
different from that of the initial Henry’s law region. A
further pressure increase will lead to the formation of
larger water clusters. As these clusters grow, the op-
portunities for cooperative adsorption effects will increase;
that is, a water molecule adsorbing onto these preexisting
clusters will often be able to position itself so that it forms
two or more H-bonds with neighboring adsorbate mol-
ecules. In this stage (IV) the adsorption energy will be
much greater, leading to a steep rise in the adsorption
isotherm. This stage is followed by a rapid increase in
water adsorption with pressure, with bridging between
adsorbed clusters on the same pore wall, or on opposite
walls. Soon after this the pore fills with water molecules
(stage V).
In scenario B the surface sites are placed randomly on
the edges of the graphite microcrystals. In this case some
sites will be relatively isolated from others; for such sites
the adsorption mechanism will be similar to that for
scenario A. However, some sites will be close together and
can participate in a cooperative adsorption mechanism at
certain pressures. At the lowest possible pressures water
molecules will bond singly to individual sites, as in scenario
A. However, once a few such water molecules are adsorbed
(stage I), there will be opportunities for ensuing water
molecules to engage in cooperative bonding (i.e., to bond
simultaneously to a surface site and to a preadsorbed water
molecule on a neighboring site). Such adsorption produces
an energy release of (HB HB
sf + ff ) and will result in a steep
increase in the slope of the isotherm (stage II). Eventually,
locations where such cooperative bonding can occur will
be saturated, and further adsorption will be by water-
water bonding as in scenario A (stage III). Subsequent
stages in pore filling would be the same as those in scenario
A.
More complex scenarios can be imagined and are likely
to be more typical of real physical systems. Thus, in
practice several species of oxygenated groups are usually
present, and these may have different values of HB sf . In
addition, some of these groups can H-bond to more than
one water molecule. The initial pressure at which coop-
erative bonding occurs, as well as the range of pressures
where this phenomenon is observed, will depend on the
surface density of sites. For higher site densities coopera-
tive bonding is expected to commence at lower pressures.
2. Experimental Method and Characterization
Materials. As an adsorbent Norit activated carbon was chosen
(Sorbonorit 2-A5998); this material is referred to as N2 in this
(and earlier) papers. The precursor for this material was a peat
moss. It was oxidized with 30% hydrogen peroxide at 323 K for
2 h. The details of the procedure applied were described
elsewhere.13-15 We also considered the use of nitric acid as an
alternative oxidizing agent. However, this strong oxidizer was
found to lead to a dramatic reduction in the total pore volume,
and so this approach was not pursued.
Sorption of Nitrogen. The nitrogen adsorption isotherm was
measured at 77 K using a Micromeritics ASAP 2010 sorptometer.
Prior to the adsorption measurements the sample was heated at
393 K and outgassed at this temperature to a constant vacuum
of 10-6 Torr. The isotherm was used to calculate the specific
surface area using the BET method and the micropore volume
(13) Bandosz, T. J.; Jagiello, J.; Schwarz, J. A. Anal. Chem. 1992, 64,
891.
(14) Jagiello, J.; Bandosz, T. J.; Schwarz, J. A. Carbon 1992, 30, 63.
(15) Bandosz, T. J.; Jagiello, J.; Schwarz, J. A.; Krzyzanowski, A.
Langmuir 1996, 12, 6480.
Langmuir, Vol. 15, No. 2, 1999 535
Figure 2. Pore size distribution of the Norit activated carbon
N2, as determined by nitrogen adsorption at 77 K using nonlocal
density functional theory. The vertical dashed lines divide the
PSD into three regions, which form the basis for choosing
discrete pore sizes for study by simulation.
using the Dubinin-Radushkevich (DR) method.16 These were
found to be SBET ) 860 m2 g-1 and Vmic ) 0.426 cm3 g-1,
respectively. The pore size distribution (PSD) was determined
from the adsorption isotherm using the density functional theory
method (DFT),17,18 which has been shown to be more reliable for
small pores than semiempirical methods.19 The PSD determined
in this way is shown in Figure 2 and displays sharp peaks at
approximately 0.65 and 1.35 nm. There is a long tail to the
distribution, which extends to 8.0 nm in Figure 2; pore widths
greater than 8.0 nm account for less than 2.6% of the total pore
volume.
Surface Characterization. The densities of acidic and basic
surface sites were determined using Boehm20 and potentiomet-
ric21,22 titration. The methods and full details of these determi-
nations have been described previously15 and are not repeated
here. The total site density was found to be 0.675 sites/nm2, with
the groups being mainly basic in nature; the oxidation method
used resulted in only small changes in the acidity of the initial
Norit sample. The species present on the surface of the N2 carbon
are characterized by pKa values of 3.3, 4.8, 7.4, and about 10.
Sorption of Water. The water sorption isotherm was obtained
at 298 K using a Micromeritics ASAP 2010 sorptometer equipped
with a vapor sorption kit. Temperature was controlled by a Fisher
Isotemp thermostatted system. Before the measurement the
sample was heated at 393 K and outgassed to a constant vacuum
of 10-6 Torr. HPLC grade water was used as an adsorbate. Prior
to the experiment dissolved gases were removed by consecutively
melting and freezing the water several times, followed by
outgassing. Each point of the isotherm was recorded after
sufficient time to ensure that equilibrium had been reached. The
initial dose amount was 0.1 cm3. It took several days to measure
the sorption at the lowest pressures (P/P0 e 0.02). The desorption
process in this pressure range was not determined because of
the extremely slow equilibration. Thus, it took about 55 h to
desorb water from a relative pressure of 0.02 to 0.01.
3. Model
Pore Size Distribution. The activated carbon is
modeled as being made up of noninterconnected slit pores
(16) Gregg, S. J.; Sing, K. S. W. Adsorption, Surface Area and Porosity,
2nd edition; Academic Press: New York, 1982.
(17) Olivier, J. P.; Conklin, W. B. Determination of pore size distri-
butions from density functional theoretical models of adsorption and
condensation with porous solids. Presented at the 7th International
Conference on Surface and Colloid Science, Compiegne, France, 1991.
(18) Lastoskie, C. M.; Gubbins, K. E.; Quirke, N. J. Phys. Chem.
1993, 97, 4786.
(19) Lastoskie, C. M.; Quirke, N.; Gubbins, K. E. In Equilibria and
Dynamics of Gas Adsorption on Heterogeneous Solid Surfaces; Rudzinski,
W., Steele, W. A., Zgrablich, G., Eds.; Elsevier: Amsterdam, 1996; p
745.
(20) Boehm, H. P. In Advances in Catalysis; Academic Press: New
York, 1966; Vol. 16.
(21) Bandosz, T. J.; Jagiello, J.; Contescu, C.; Schwarz, J. A. Carbon
1993, 31, 1193.
(22) Jagiello, J.; Bandosz, T. J.; Schwarz, J. A. Carbon 1994, 32,
1026.
536 Langmuir, Vol. 15, No. 2, 1999
having a distribution of pore widths given by the experi-
mentally determined PSD shown in Figure 2. The PSD
from experiment gives information on the effective, or
“available” pore width w. This quantity is assumed to be
related to H, the distance separating the planes through
the centers of the first layer of C atoms on opposing walls,
by
w ) H - σc (1)
where σc is the effective diameter of a carbon atom. The
amount of fluid adsorbed at a given pressure may be
written
N) ∫0 f(w) N(w) dw
∞
(2)
where f(w) is the PSD and N(w) is the amount adsorbed
in the pores of width w at the given pressure.
In principle, it is possible to calculate N by carrying out
simulations to determine N(w) for a large number of fixed
pore widths H covering the range of widths shown by the
experimentally determined PSD. The total adsorption can
then be determined from eq 2, using the experimentally
determined PSD, f(w). At low pressures a few pore widths
suffice for such a scheme, since the isotherm changes
slowly with pressure. For higher pressures a much larger
number of pore widths is needed since pore filling occurs,
and the adsorption changes rapidly. Since the simulations
are lengthy, even on the fastest supercomputers, it is not
feasible at present to carry out such calculations for a
large number of pore widths, and some interpolation
scheme must be devised that enables the overall adsorp-
tion to be calculated using results for a small number of
pore widths. We have employed two such schemes, one
suitable for low pressures and the other for higher
pressures.
In the low-pressure region (0 e P/P0 e 0.02) we
approximate the integral of eq 2 by a finite sum over
discrete, but representative, pore widths:
N ) c1N(w1) + c2N(w2) + ... + cnN(wn) (3)
where wi are the selected pore widths and ci are weighting
factors indicative of the total pore volume represented by
the individual pore widths. On the basis of the PSD shown
in Figure 2, three values were chosen for the pore widths
to be studied: H ) 0.99, 1.69, and 4.5 nm. These widths
represent regions 1, 2, and 3 in Figure 2, respectively.
The weighting factors, ci, for these three pore sizes are
taken to be proportional to the area under these three
regions of the PSD: their values are c1 ) 0.518, c2 ) 0.371,
and c3 ) 0.111, respectively.
As mentioned above, this weighted-averaging procedure
breaks down in the pore-filling region because pores of
different width fill at different pressures. Weighted
averaging using a few pore widths necessarily leads to a
stepped isotherm. For these higher pressures, therefore,
we use the semiempirical equation of Talu and Meunier23
to interpolate among the adsorption isotherms for the pore
widths studied. This three-parameter equation was
developed specifically for associating systems. It is
H0Ψ
P) exp(Ψ/Nm) (4)
1 + KΨ
where K is a parameter related to the equilibrium constant
for dimer formation, Nm is the saturation capacity of the
(23) Talu, O.; Meunier, F. AIChE J. 1996, 42, 809.
McCallum et al.
pore (and thus varies with pore width), H0 ) limPf0 dP/
dN is the Henry’s law constant,
-1 + (1 + 4Kξ)1/2
Ψ) (5)
2K
and
NmN
ξ) (6)
Nm - N
where N is the amount adsorbed at pressure P.
To improve this interpolation scheme, a fourth pore
width, H ) 0.79 nm, was studied for the entire pressure
range 0 < P < 1. The four simulated adsorption isotherms,
H ) 0.79, 0.99, 1.69, and 4.5 nm, were fitted to eq 4 to
yield values of the parameters H0, K, and Nm. The resulting
smooth fits showed good agreement with the simulation
data in the steeply rising region of the isotherms; the main
discrepancy between eq 4 and the simulations occurred
in the low-pressure region, where the equation gave
adsorbed amounts that were too low. The parameters H0,
K, and Nm varied smoothly with w and were fitted to simple
relations. Using these relationships, together with eq 4,
enabled us to construct adsorption isotherms for any pore
width; these, together with the experimental PSD, made
it possible to carry out the integration of eq 2.
Surface Groups. Groups such as hydroxyl, carbonyl,
carboxylic, phenolic, carbonyl, lactonic, chromene, and
pyrone have been identified on active carbon surfaces.
The relative and absolute amounts of these groups depend
on the carbon precursor material and on the activation
conditions employed in the creation of the porous sample.
X-ray diffraction studies show that these oxygen-contain-
ing surface groups are located around the edges of the
carbon microcrystallites.24 The experimental character-
ization of surface chemistry can be achieved using a variety
of chemical and physical techniques. Frequently, different
methods lead to quite disparate results, presumably
because of the complexities of the carbon structure.
The carbon used here has a total estimated surface group
density of 0.675 sites/nm2, with about 3 times as many
basic groups (e.g., pyrone) as acidic surface sites (e.g.,
carboxylic); both can bind water, since water can act as
either a Lewis acid or base. To simplify our model, we
assume that the mixture of surface groups can be modeled
by an effective surface in which all oxygenated groups
have the same binding energy, HB sf , with water molecules;
moreover, we assume that each surface group can only
bond with one water molecule. We refer to this as the
“effective single group” model. In reality, there will exist
a mixture of groups of different species on the surface,
with sites which have somewhat different binding energies
and some with more than one binding site. We postulate
that our model can represent this surface reasonably well
with a value of HB
sf that is a suitably weighted mean of all
these site-water interactions.
In our model, surface groups are represented as OH
groups and consist of a Lennard-Jones (LJ) sphere to
represent an oxygen atom, and a single square well site
placed on this that represents the H atom and can interact
with similar sites on water molecules. Surface groups are
placed randomly on a square lattice that is superimposed
on the carbon surfaces, with the restriction than no two
groups can be closer than 21/6σ, where σ is the LJ diameter
(24) Bansal, R. C.; Donnet, J. In Carbon Black; Donnet, J., Bansal,
R. C., Wang, M., Eds.; Marcel Dekker: New York, 1993.
Adsorption of Water on Activated Carbon Langmuir, Vol. 15, No. 2, 1999 537

Table 1. Maximum Site Density n for Different

Microcrystal Edge-Lengths L, for a Random Placement
of Sites
L/nm n/nm-2 L/nm n/nm-2
1.0 9.66 2.5 4.34
1.5 6.89 3.0 3.67
2.0 5.33 10.0 1.15

Table 2. GEMC and NPTMC Simulation Results and

Figure 3. Model of the water molecule used in the simulation.
The oxygen is represented by a Lennard-Jones sphere of
diameter σLJ; the association sites (hydrogen atoms and lone-
pair electrons) are tetrahedrally arranged square well sites of
diameter σHB HB
ff and well depth ff .
for the O atom. The lattice lines define the edges of the
graphite microcrystals, and the length L of a microcrystal
edge is retained as a variable. Values of L between 1 and
10 nm have been reported, on the basis of diffraction
studies; however, most studies have shown values of L in
the range 1-3 nm. Experimental site densities range from
0 to 2.65 sites/nm2. We have chosen L to be 2.0 nm for
most of the calculations in our study (the effect of varying
site density is considered in section 5). The arrangement
of groups on the surface is kept fixed as H is varied, to
eliminate differences due to site location.
Intermolecular Potentials. Water is modeled as a
LJ sphere, with four tetrahedrally arranged square well
sites4,25 (Figure 3). Two sites represent hydrogen atoms,
and two mimic the lone pairs of electrons. Bonding only
occurs between unlike sites. The sites are located at a
distance of 0.42σ from the center of the LJ sphere; this
placement is consistent with the observation of a peak in
the O-H pair correlation function for water at 0.19 nm,
as found from neutron diffraction experiments.26 Thus,
the fluid-fluid pair intermolecular potential energy
between two water molecules can be expressed as
Experimental Data for Coexisting Liquid Water and
Water Vapor at 298 K (The Accuracy Is the Simulation
Value as a Percentage of the Experimental Value)
property experiment simulation accuracy
Fvapor(mmol/m3) 1.2798 1.267 ( 0.004 99.1 ( 0.3%
Fliquid (mmol/m3) 55389 54400 ( 600 98 ( 1%
P0 (atm) 0.03125 0.0307 ( 0.0006 98 ( 2%
a range of values of HBff . These simulations yielded the
vapor pressure and coexisting densities of the gas and
liquid phases. The value of HB ff was then chosen to give
the best agreement with experimental data for these
properties. This procedure gave a value of HBff /k ) 3800
K; agreement with experiment was within 2% for each of
the coexistence properties (see Table 2).
The interaction of a water molecule with the walls is of
the form
usf(r1ω1) ) uCf(z) + uLJ(r) + uHB
sf (r1ω1) (9)
where uCf is the interaction energy of the water with the
carbon atoms, uLJ is the LJ interaction between the LJ
center in the water molecule and the LJ centers that
represent the O atoms in the OH groups attached to the
walls, and uHBsf is the H-bond interaction between the
water molecule and a site on the wall; r1 and ω1 denote
the location and orientation of the water molecule, z is the
distance of the center of the water molecule from the
nearest point on the wall (taken to be the plane through
the centers of the carbon atoms forming the first layer),
and r is the distance between the centers of the LJ sites
in the water molecule and the OH group on the wall. The
interaction between water and the carbon atoms is

[( ) ( ) ]
σff σff
12 6
modeled using the 10-4-3 potential of Steele:27
uff(rω1ω2) ) 4ff - + uHB
ff (rω1ω2) (7)

[( ) ( )
r r
2 σCf 10 σCf 4
where r is the vector joining the centers of the two water uCf(z) ) 2πFCCfσCf2∆ - -
5 z z

molecules, ωi denotes the orientation of water molecule
i, ff and σff are the usual LJ well depth and molecular
diameter parameters, and the H-bond term is given by
uHB HB HB
ff (rω1ω2) ) -ff if rRβ < σff
) 0 otherwise
where rRβ is the distance between the center of a square
well (SW) site R on molecule 1 and a SW site β on molecule
2.
The LJ parameters, and also the HB size parameter,
were taken from previous studies:4 σff ) 0.306 nm; ff/k
) 90.0 K; σHB
ff ) 0.0612 nm. The calculations are sensitive
to the value of the strength of the H-bond, HB ff , and we
therefore carefully fit this parameter to best reproduce
the gas-liquid coexistence properties of bulk water. A
series of Gibbs ensemble and isothermal-isobaric simu-
lations were undertaken at 298 K for this water model for
(25) Müller, E. A.; Gubbins, K. E. Ind. Eng. Chem. Res. 1995, 34,
3662.
(26) Soper, A. K.; Phillips, M. G. Chem. Phys. 1986, 107, 47.
( )]
σCf4
(10)
3∆(z + 0.61∆)3
where the density of the carbon is FC ) 114 nm-3, the
(8) separation of the graphite planes is ∆ ) 0.335 nm, and
Cf and σCf are the LJ parameters for the interaction
between carbon atoms and water. The carbon-carbon
potential parameters are taken as27 Cf/k ) 28 K and σCf
) 0.340 nm. Fluid-solid cross parameters are calculated
using the usual Lorentz-Berthelot rules:
Cf ) (CCff)1/2
σCf ) (σCC + σff)/2 (11)
The H-bond term in eq 9 is the same square well function
given by eq 8, but with subscripts ff replaced by sf. The
active sites on the carbon surface are represented by LJ
(27) Steele, W. A. The Interaction of Gases with Solid Surfaces;
Pergamon Press: Oxford, 1974.
538 Langmuir, Vol. 15, No. 2, 1999
Figure 4. Representation of the activated carbon pore used
in the MC simulations. The structureless walls are decorated
with active sites that are taken to be OH groups; these are
represented as Lennard-Jones oxygen atoms (large open circles)
and square well hydrogen atoms (small filled circles). The pore
width H is the distance between the plane through the centers
of the carbon atoms in the first layer on opposite walls of the
pore. The lower figures (A and B) show typical arrangements
of sites on the surface for n ) 0.675 sites/nm2 for L ) 2.0 (A)
and 1.0 (B) nm.
spheres with an attached single square well bonding site
(Figure 4); these represent hydroxyl groups. The LJ sphere
represents the O atom of the OH group and its center is
placed at a distance of 0.1364 nm above the plane through
the centers of the first-layer carbon atoms. The LJ
parameters for this LJ sphere are the same as those for
the LJ site in the water molecule, i.e.,  ) ff ) 90.0 K and
σ ) σff ) 0.306 nm. The center of the square well site is
placed at a distance of 0.42σ from the center of the LJ
sphere and is placed as far from the wall as possible (i.e.,
on the normal to the wall which passes through the center
of the LJ sphere). The wall-site square well interaction
has the same range as those on the water molecules (i.e.,
σHB
sf ) 0.0612 nm) and mimics a hydrogen atom, so that
it is only able to form bonds with the lone-pair electron
sites on the water molecules. Determination of the
appropriate strength of the wall-fluid hydrogen bond is
discussed below.
4. Molecular Simulation Method
Calculations were carried out using the grand canonical
Monte Carlo (GCMC) method.28 This method is convenient
for adsorption studies, since the chemical potential µ,
temperature T, and volume V are specified and kept fixed
in the simulation. Since µ and T are the same in the bulk
and adsorbed phases at equilibrium, the thermodynamic
state of the bulk phase is known in such simulations. Three
types of molecular moves are attempted: molecular
creation, molecular destruction, and the usual Monte Carlo
translation/rotation moves. Each of these three moves is
attempted with equal frequency. The type of move is
chosen randomly to maintain microscopic reversibility.
The probability of successful creations or destructions is
strongly dependent on the density of the system. The
maximum allowable rotation and displacement of a
molecule are adjusted so that the combined move has an
acceptance probability of about 40%. This value should
ensure the most efficient probing of the phase space
distribution. It is noteworthy that the values of the
(28) Allen, M. P.; Tildesley, D. J. Computer Simulation of Liquids;
Clarendon Press: Oxford, 1987.
McCallum et al.
maximum displacement and rotation are very small
compared to values encountered for nonbonding systems.
In these simulations the number of adsorbed molecules
fluctuates during the simulation. Calculation of the
average of this number for a range of chemical potentials
enables the adsorption isotherm to be constructed.
The walls of the slit pores lie in the x-y plane. Normal
periodic boundary conditions, together with the minimum
image convention, are applied in these two directions. For
low pressures, P/P0 e 0.02, the length of the simulation
cell in the two directions parallel to the walls was
maintained at 10 nm for each of the pore widths studied,
to maintain a sufficient number of adsorbed molecules.
For higher pressures where more water molecules were
present, the minimum cell length in the x and y directions
was 4 nm. A potential cutoff of 5σ was used. For the system
studied HBff /kT ) 12.8. While this value is not high
enough to require biased sampling methods,4 long runs
are needed to ensure ergodicity. Associating water mol-
ecules tend to remain in energetically favorable configu-
rations for many MC steps, and the system thus requires
many steps to reach a true equilibrium state. In our runs
500 million MC steps were used for equilibration, followed
by a further 500 million steps for property averaging.
Shorter runs than this were not adequate to sample
desorption events. The average number of water molecules
in the simulation cell varied from a few molecules at the
lowest pressures to a few hundred or thousands of
molecules when the pores were full; filled pores contained
about 320 molecules for a pore width of 0.79 nm, 460 at
0.99 nm, 830 at 1.69 nm, and 2100 at H ) 4.5 nm.
Calculations were carried out on the Cornell Theory Center
IBM SP2.
In determining the adsorption isotherm, we commenced
with the cell empty; a value of the fugacity corresponding
to a low pressure was chosen and the average adsorption
determined from the simulation. The final configuration
generated at each stage was used as the starting point for
simulations at higher fugacities. The pressure of the bulk
gas corresponding to a given chemical potential was
determined from the ideal gas equation of state. Gas-
phase densities corresponding to the range of chemical
potentials studied were determined by carrying out
simulations of the bulk gas. These were found to agree
with those calculated from the ideal gas equation within
the estimated errors of the simulations.
5. Results
Low-Pressure Region (P/P0 e 0.02). At sufficiently
low pressures the adsorbate molecules interact solely with
the pore walls, and fluid-fluid interactions can be
neglected. The simulations show that this region lies at
relative pressures P/P0 below about 0.003, and this seems
to be confirmed by the experimental results. Experimental
measurements were carried out at relative pressures down
to 0.002, corresponding to 6.3 × 10-5 bar (0.048 Torr),
which is approaching the limit of sensitivity of the
sorptometer. At relative pressures above 0.003 the ex-
perimental adsorption isotherm (Figure 5) shows a sharp
increase in slope. This steeper region persists to relative
pressures of about 0.01, after which the slope decreases
again. The quantity plotted in Figure 5 is the excess
adsorption, Γexcess ) (N - Nbulk)/A, where N is the number
of molecules adsorbed, Nbulk is the number of molecules
in a volume of bulk gas equal to that of the pore at the
same temperature and chemical potential, and A is the
surface area.
The simulation results are sensitive to the value chosen
for HB
sf /k. We first made a rough estimate of this param-
Adsorption of Water on Activated Carbon
Figure 5. Adsorption of water at 298 K at low relative pressure
(P/P0 e 0.02). Open circles are the experimental data (the solid
line is a guide to the eye), and solid points are simulation
results: squares, HB HB
sf /k ) 4800 K; circles, sf /k ) 4900 K;
HB
triangles, sf /k ) 5000 K. Data points are the weighted
average of simulation results for the three individual pore
widths. Error bars denote one standard deviation of the
simulated values.
eter. Typical H-bond strengths obtained by fits of simula-
tions to bulk thermodynamic properties are HB sf /k ) 3643
K for H-bond donor group/water and 2673 K for H-bond
acceptor group/water, respectively. These values were
obtained for acetic acid-water complexes.11,29 These values
are less than the value of 3800 K for water-water
interactions. However, the HB sf /k value for our model is
expected to be larger than this, since it only allows for one
group-water H-bond for each group, whereas the real
carbon surface will have groups that can form multiple
H-bonds with water molecules. Carboxyl groups, for
example, can H-bond with two water molecules, one
through the H-bond donor OH site and the other with the
H-bond acceptor site dO. Similarly, we expect two H-bond
acceptor sites for lactonic groups, one H-bond donor site
for phenolic groups, and so on. In our “effective single
group” model, the appropriate value of HB sf /k will be a
weighted average of the total bonding energies for the
various surface groups for the real carbon. Using the
experimental values of group densities for the various
species present, we estimate the value of the average
H-bond strength to be HB sf /k ) 4949 K. This estimate is
in agreement with our initial simulations, which showed
that a value in the range HB sf /k ) 4800-5000 K gives the
best fit to the low-pressure data.
We have carried out simulations in the pressure range
0 e P/P0 e 0.02 for three pore widths, H ) 0.99, 1.69, and
4.50 nm, and for three values of bonding strength, HB sf /k
) 4800, 4900, and 5000 K. The site density was fixed at
0.675 sites/nm2, and the length of the graphite microc-
rystals was 2 nm in all systems studied. For each of the
three bonding strengths we calculated the isotherm for
the PSD of Figure 2, using eq 3. These results are shown
in Figure 5, along with the experimental data. None of
the three simulated isotherms reproduce the data over
the full pressure range. With HB sf /k ) 5000 K the lowest
pressure part of the isotherm is correctly reproduced, and
the sharp increase in slope is obtained at the correct
pressure (P/P0 ≈ 0.004). However, the higher slope
persists to too high of a pressure, the slope being reduced
only for pressures above about P/P0 ≈ 0.03-0.04, whereas
this reduction in slope occurs at P/P0 ≈ 0.01 in the
(29) Gao, J. J. Phys. Chem. 1992, 96, 6432.
Langmuir, Vol. 15, No. 2, 1999 539
experimental system. The simulations for HB sf /k ) 4800
and 4900 K show similar results, but with less adsorption,
and with the increase in slope delayed to higher pressures.
Snapshots of typical molecular configurations in the
pore reveal that all adsorbed water molecules are bonded
onto a surface site for the pressure range shown in Figure
5; however, at the higher pressures shown such adsorbed
water molecules may also be H-bonded to other water
molecules, where two surface sites are spaced a suitable
distance apart. For pressures below P/P0 ≈ 0.003 water
molecules are adsorbed onto single sites, with no water-
water bonding (Figure 6a). Since all water-surface site
bonds have the same energy in our effective single group
model, this very low pressure region corresponds to
Henry’s law regime, and the isotherm is linear. The
experimental data in this region is sparse and it is not
possible to detect with confidence a linear region. We
believe that even in this low-pressure region the experi-
mental isotherm is likely to show significant nonlinearity,
since different chemical groups on the carbon surface will
interact with water with different bonding energies. At
pressures somewhat above P/P0 ≈ 0.003 the snapshots
show that cooperative adsorption effects are important.
This corresponds to the region of the isotherm with
increased slope. At these pressures newly adsorbed water
molecules seek surface sites on which they can simulta-
neously bond to both the surface site and a preadsorbed
water molecule, thus releasing an amount of energy HB sf
+ HB
ff . An example of this behavior is shown in Figure 6b.
Thus, the adsorption mechanism in our model is similar
to that shown in stages I and II of scheme B of Figure 1.
At still higher pressures, beyond those shown in Figure
5, snapshots of the simulations indicate that cooperative
bonding of the type shown in Figure 6b is complete, and
adsorption involves H-bonding of additional water mol-
ecules to either isolated surface sites or to preadsorbed
water molecules through water-water bonds.
The most noticeable difference between the simulation
results and the experimental ones is the persistence of
the steep region of the isotherm to too high of a pressure
in the case of the model. The effective single group model
apparently provides too many possibilities for cooperative
bonding. A possible reason for this defect may be that on
the real carbon surface many groups have more than one
bonding site, which is not accounted for in our model. In
addition, steric hindrance is likely to limit cooperative
bonding to a greater extent in real carbons than in our
model.
Higher Pressures (0 e P/P0 e 1.0). On the basis of
the results at low and moderate pressures, we adopted a
value of the surface bonding strength of HB sf /k ) 4800 K,
since this seemed to give the best overall agreement with
experiment for relative pressures above 0.01. Simulations
were conducted for pressures over the range 0 e P/P0 e
1 for pores of width H ) 0.79, 0.99, 1.69, and 4.50 nm; site
density and microcrystal length were kept fixed at 0.675
sites/nm2 and 2.0 nm, respectively. The number and
arrangement of the sites were the same for the different
pore widths.
The simulation results for the four pore widths are
shown in Figure 7. Since the pore capacity expressed as
Γexcess varies widely with pore width it is more convenient
to display the results in Figure 7 as the excess density in
the pore, defined as
Γexcess
Fexcess ) (12)
w
540 Langmuir, Vol. 15, No. 2, 1999 McCallum et al.

Figure 6. Snapshots of typical configurations for one of the activated carbon surfaces for a pore of width H ) 1.69 nm for HB
sf /k
) 4800 K: at low pressures, top figure, P/P0 ) 0.003; lower figure, P/P0 ) 0.017. The gray background is the carbon surface; the
black spheres embedded on this surface are the O atoms of the OH surface groups, the attached small white sphere being the square
well bonding site (or H atom). Water molecules are represented as a large white sphere (O atom) with four small attached spheres
arranged tetrahedrally, representing the two H atoms (white spheres) and two lone-pair electrons (gray spheres).

Figure 7. Simulated adsorption isotherms at 298 K for water into activated carbon pores of various widths: H ) 0.79, 0.99, 1.69,
and 4.50 nm. Points denote simulated values, and lines are the fits to the simulated data using the expression of Talu and Meunier
(eq 4).

where w is the effective pore width given by eq 1. As classification,5 and the pore filling pressure and pore
expected, the isotherms are of type V in the usual IUPAC capacity depend strongly on pore width. The narrowest
Adsorption of Water on Activated Carbon
Figure 8. Snapshots of typical configurations for a pore of width H ) 1.69 nm for higher pressures than those shown in Figure
6: upper figure, P/P0 ) 0.20 (showing one of the carbon surfaces); lower figure, P/P0 ) 0.575 (showing both carbon walls). Key
as in Figure 6.
pores, H ) 0.79 and 0.99 nm, display much greater
adsorption at low pressures than the larger pores, and
pore filling occurs at a low relative pressure of P/P0 ≈ 0.2.
In these smaller pores water molecules are able to form
“bridges” between opposing walls at lower pressures than
for larger pores; the establishment of such a link between
the pore walls is quickly followed by pore filling, with the
bridging water molecules acting as the nuclei for further
adsorption.
Snapshots of typical configurations are shown for the
pore of H ) 1.69 nm in Figure 8 for an intermediate
pressure of P/P0 ) 0.20 and a pressure near to pore filling,
P/P0 ) 0.575. At the lower pressure of 0.20 we see water
molecules bonding to preadsorbed water molecules, in
addition to water molecules adsorbed to surface sites and
the cooperative bonding seen in Figure 6b. This corre-
sponds to stage III of scheme B of Figure 1. At a reduced
pressure of 0.575 the water clusters have grown and
bridged across the pore walls (stage IV of scheme B in
Figure 1). Although large water clusters of liquidlike
density are present, there is still considerable empty space
in the pore at this stage. At this stage additional adsorption
of water molecules occurs very easily, with new molecules
forming several water-water bonds on adsorption with
molecules already previously adsorbed. Thus, the pore
fills rapidly, and at a pressure slightly above 0.575 the
pore is full.
For smaller pores the principal difference is that the
formation of bridges of water molecules between opposing
walls can occur more easily, and so appears at lower
pressures. This is seen in the series of snapshots for a
pore of width 0.99 nm in Figure 9, where bridging between
walls occurs at a pressure of P/P0 ) 0.05, and pore filling
has occurred at a reduced pressure of about 0.25.
Langmuir, Vol. 15, No. 2, 1999 541
To calculate the adsorption isotherm for the experi-
mentally determined PSD of Figure 2, the semiempirical
equation of Talu and Meunier23 was used to interpolate
between the simulated isotherms for the four pore widths
shown in Figure 7, as described in section 3. The constants
in this equation were fitted for each of the pore widths,
and were found to vary smoothly with H. These fitted
curves are included in Figure 7. By interpolating between
these fitted values of the constants K, Nm, and H0 it was
possible to estimate adsorption isotherms for a wide range
of pore widths, and to then carry out the integration of eq
2 using the experimentally determined PSD. A comparison
of the isotherm determined in this way from the simulation
results with the experimental one is shown in Figure 10.
The overall agreement is quite good. The pore filling
pressure is predicted well, although the simulated curve
is somewhat steeper than the experimental one in this
region. The predicted adsorption is too low at low
pressures, and the predicted adsorption for the filled pore
is also somewhat too low. The underestimation of the
adsorption at low pressure, and the predicted steepness
in the pore-filling region, are primarily a result of the use
of the Talu-Meunier equation, as seen from the com-
parison of this equation with the simulation results shown
in Figure 7. Errors in the experimentally determined PSD
could also contribute to discrepancies in the pore-filling
region. At the highest pressures, near P/P0 ≈ 1, the
simulated adsorption is approximately 6% too low. The
Talu-Meuinier equation describes this region well. Since
the model used for water predicts the bulk density
accurately, the discrepancy in this region most likely
results from uncertainty in the PSD for large pores; a
542 Langmuir, Vol. 15, No. 2, 1999
Figure 9. Snapshots of typical configurations for a pore of width H ) 0.99 nm: top figure, P/P0 ) 0.010; middle, P/P0 ) 0.050;
bottom, P/P0 ) 0.30. Both carbon walls are shown.
Figure 10. Adsorption of water from experiment and simula-
tion at 298 K. The solid line and open circles represent the
experimental data; the dashed line and filled circles show the
simulated isotherm, obtained from eq 2 using simulated
isotherms for discrete pore widths together with the experi-
mental PSD.
small increase in the fraction of large pores would lead to
a significant increase in the adsorption capacity here.
Influence of Microcrystal Size on Adsorption. For
a constant site density, reducing the length of the
microcrystals leads to an increase in the average spacing
between active sites (see Figure 4). With sites further
apart, we expect less cooperative bonding and less bridging
between adjacent sites. To investigate the importance of
such effects, we carried out simulations for two micro-
crystal lengths, L, of 1.0 and 2.0 nm at 298 K. In these
calculations the site density was fixed at 0.675 sites/nm2,
the pore width was H ) 1.69 nm, and the fluid-wall
bonding strength was chosen to be HB sf /k ) 5000 K. The
simulation cell length was fixed at 10.0 nm, and calcula-
tions were made for the pressure range P/P0 ) 0 to 0.10.
The results are shown in Figure 11. As expected, there
McCallum et al.
Figure 11. Influence of microcrystal length, L, on low-pressure
adsorption, for a pore width H ) 1.69 nm, site density ) 0.675
sites/nm2, T ) 298 K, and HB
sf /k ) 5000 K. Circles show results
for L ) 1.0 nm and squares are for L ) 2.0 nm.
is a substantial decrease (by up to 45%) in the adsorption
as the crystal length is reduced from L ) 2.0 to 1.0 nm.
The study of typical molecular configurations shows that
cooperative bonding of the type shown in Figure 6 is much
less for a microcrystal length of 1.0 nm. For example, if
we consider a point on each isotherm with the same
amount of adsorption, N ≈ 30, the percentage of site-
bonded molecules participating in cooperative bonding is
about 60% for L ) 2.0 nm compared to 33% for L ) 1.0
nm. This observation is part of the reason for the greater
amount of adsorption at a given pressure in the case of
the larger microcrystal. Thus, the microcrystal length has
a relatively large impact on the adsorption at low
pressures, since it strongly affects the relative distances
between sites.
Effect of Site Density on Adsorption. To study the
effect of site density on adsorption, we carried out a series
of simulations in which the number of active sites on the
pore walls was varied. For a pore having a width H ) 1.69
Adsorption of Water on Activated Carbon Langmuir, Vol. 15, No. 2, 1999 543

Figure 12. Effect of active site density on water adsorption

at 298 K in a pore of width 1.69 nm, with HB sf /k ) 5000 K and
a microcrystal length of 1.0 nm. Three site densities are shown:
0.3 (circles), 0.675 (squares), and 1.2 sites/nm2 (diamonds). For
the filled pore (not shown) the adsorption was about 3.3 × 10-2
mmol/m2 for all three cases.

nm and HB sf /k ) 5000 K, three site densities were

investigated: 0.30, 0.675, and 1.20 sites/nm2. The mi-
crocrystal length used for this study was 1.0 nm, and the
simulation cell length was held fixed at 3.0 nm. Adsorption
isotherms at 298 K were generated for the full range of
relative pressures, P/P0 ) 0 to 1.
Adsorption isotherms for the three site densities are
shown in Figure 12 for pressures up to the pore-filling
region. Three conclusions can be drawn. First, the pore-
filling pressure depends strongly on the site density,
occurring at a lower pressure for the higher site densities.
Second, the systems with higher site densities (0.675 and
1.20 sites/nm2) exhibit a gradual transition to pore filling,
while the system with a low site density shows a sharp
transition, which appears to be capillary condensation.
Finally, an increase in site density leads to an increase
in adsorption in the region prior to pore filling. This agrees
with experimental observations that adsorption is en-
hanced by an increase in the degree of oxidation of active
Figure 13. Snapshots of typical configurations just prior to
carbons (up to some limiting level, after which further
pore filling for a pore of width H ) 1.69 nm and HB
sf /k ) 5000
oxidation begins to destroy the porous structure and leads K. Results are shown for three site densities: n ) 0.30 (top
to a reduction in adsorption capacity). Each of these figure) at P/P0 ) 0.81, 0.675 (middle) at P/P0 ) 0.63, and 1.20
observations can be explained by the adsorption mech- (bottom) at P/P0 ) 0.45.
anism, as described below.
An increase in adsorption with site density occurs bond to surface sites, and there is almost no cooperative
naturally because there are more sites available. In bonding. Pore filling appears to occur by capillary con-
addition, at the higher site densities of 0.675 and 1.20 densation, the transition being from a state in which only
sites/nm2, cooperative bonding will come into play and a few water molecules are adsorbed onto active sites (top
lead to significant increases in adsorption. Since the snapshot in Figure 13) to one in which the pore is filled
possibilities for cooperative bonding are similar for these at a liquidlike density. At the higher site densities,
two higher site densities, the adsorption curves are very however, adsorption of single water molecules onto active
similar for relative pressures up to 0.10. For higher sites at low pressure is followed by the buildup of clusters
pressures adsorption of water onto surface sites (without (lower two snapshots in Figure 13), formation of bridges
cooperative bonding) continues up to pressures just below between pore walls, and finally pore filling. This mech-
pore filling, since the wall sites have a higher bonding anism of pore filling is clearly different from that at the
energy than the water-water H-bond. Since the pore with low site density, and the pore filling is a continuous process.
the site density of 1.20 sites/nm2 has more of these sites The results suggest that there may be a limiting site
available, adsorption is higher in this relative pressure density above which capillary condensation no longer
region (0.10-0.43). As a result of this increased adsorption, occurs, somewhere between n ) 0.3 and 0.675 sites/nm2
bridging of adsorbed water molecules between the two for this system.
pore walls occurs more readily for the highest site density.
Such bridging provides many opportunities for cooperative 6. Conclusions
water-water bonding (newly adsorbed water molecules
bonding to more than one preadsorbed molecule), and leads While the experimental isotherm is of class V as
to lower pore-filling pressures. expected, the low-pressure measurements exhibit con-
Snapshots of typical configurations for the three site siderable sensitivity to the surface chemistry, which
densities at pressures up to pore filling show a qualitative manifests itself in changes in slope at particular pressures.
difference in behavior at the lowest site density, 0.3 sites/ Although the model shows a very short Henry’s Law region
nm2 versus the two higher densities, 0.675 and 1.20 sites/ (for relative pressures P/P0 e 0.003), it seems unlikely
nm2 (Figure 13). At 0.3 sites/nm2 single water molecules that a significant Henry’s law region exists for the
544 Langmuir, Vol. 15, No. 2, 1999
experimental system, since a variety of surface groups
are present having different adsorption energies.
The model calculations show good overall agreement
with the experimental data, and we believe they are
significantly more realistic than previous attempts to
model this complex system at the molecular level.
However, the effective single group model as applied here
still has significant defects. In particular it fails to predict
the very low pressure data accurately. While it describes
the increase in slope (seen experimentally at P/P0 ≈ 0.003)
and the later decrease in slope, it predicts too large of a
pressure range for this cooperative bonding region. We
believe this may be due to a neglect of groups that have
multiple bonding sites, and/or to a neglect of groups having
different bonding energies. A further possible cause may
be that the microcrystal length used (L ) 2.0 nm) is
inappropriate. Another unsatisfactory feature is the
method used here to describe the pore size distribution at
the higher pressures. Because the semiempirical equation
used to interpolate between results for different pore
widths does not describe the low-pressure region well,
the resulting calculated isotherm for the experimental
PSD is inaccurate in this region and also has too steep of
a slope in the pore-filling region. This is not a defect of the
model itself, but reflects the difficulty of simulating a
sufficient number of pore widths to cover the PSD for this
system, which is broad. It could be diminished by the use
of a carbon with a narrower PSD.
McCallum et al.
Future experimental work on these systems should focus
on carbons which have a narrow PSD and are well-
characterized. In particular, the use of neutron or X-ray
diffraction to determine the average length (or distribution
of lengths) of the microcrystals would be valuable, as would
the use of spectroscopic techniques such as FTIR to
determine the chemical nature and density of surface
groups. Further work on the molecular model should
incorporate the possibility of groups having multiple
bonding sites, as well as sites of different bonding energies,
to better describe the low-pressure adsorption region.
Finally, these models should be tested against a variety
of activated carbons, having differing surface chemistry
and site density.
Acknowledgment. This work was supported by the
Department of Energy (Grant No. DE-FG02-88ER13974).
The simulations reported here were performed on the IBM-
SP2 at the Cornell Theory Center, with the support of the
National Science foundation through a Metacenter grant
(MCA93S011). S.C.M. also thanks the NSF for a CISE
Postdoctoral Fellowship. International cooperation was
made possible by a NSF/CONICIT U.S.-Venezuela Co-
operative Research grant (No. INT-9602960).
LA9805950