You are on page 1of 11

JOURNAL OF COLLOID AND INTERFACE SCIENCE 183, 57–67 (1996)

ARTICLE NO. 0518

Derivation and Application of a Jovanovic–Freundlich Isotherm Model


for Single-Component Adsorption on Heterogeneous Surfaces
IGOR QUIÑONES * AND GEORGES GUIOCHON† ,1
*Departamento de Tecnologı́a, Centro de Quı́mica Farmacéutica (CQF), Calle 200 y 21, Atabey, Playa; P.O. Box 16042, La Habana, Cuba,
C.P. 11600; and †Department of Chemistry, University of Tennessee, Knoxville, Tennessee 37996-1600 and Division of Chemical
and Analytical Sciences, Oak Ridge National Laboratory, Oak Ridge, Tennessee 37831-6120

Received January 16, 1996; accepted May 21, 1996

ponents are required, there are methods to derive reason-


A new Jovanovic–Freundlich isotherm model is derived for de- able approximations of these competitive isotherms from
scribing single-component adsorption equilibria on heterogeneous single-component isotherms. These methods allow con-
surfaces. The equation is obtained by assuming that the rate of
siderable reduction in the time and cost required by the
decrease of the fraction of the surface unoccupied by the adsorbate
molecules is proportional to a certain power of the partial pressure data acquisition. Their accuracy depends greatly on the
of the adsorbate. The equation reduces to the Jovanovic equation availability of accurate correlation for the single-compo-
when the surface becomes homogeneous. At low pressures, the nent isotherm data ( 1, 2 ) . However, except in rare cases
equation reduces to the Freundlich isotherm but at high pressures, involving homogeneous surfaces ( e.g., graphitized carbon
a monolayer coverage is achieved. This model has been applied black, zeolites ) , the single-component isotherms cannot
successfully to the description of the adsorption behavior of a be calculated from first principles but must be measured.
series of chlorinated hydrocarbons on a microporous silicagel, at Then arises the problem of selecting a proper isotherm
different temperatures. The monolayer capacity and the heteroge- model for correlating these data ( 2 ) .
neity parameter exhibit a weak temperature dependence. The third Theoretical models of single-component adsorption iso-
parameter of the model decreases exponentially with increasing
therms have been derived using a variety of kinetic, statisti-
temperature. The fit of the experimental data to the new model
described is shown to be better than the comparable fits to classical
cal mechanical, or thermodynamic approaches (2, 3). These
isotherms used for heterogeneous surfaces. The energy distribution fundamental studies assume the solid surface to be homoge-
function corresponding to the model for Langmuir local adsorption neous. Most of the adsorbents used in current practice, such
behavior was derived using the Sips procedure and evaluated nu- as activated carbons, silica gels, alumina, and even zeolites
merically in a few selected cases. This distribution is an exponen- exhibit surface and structural heterogeneities (3). For such
tial decay. q 1996 Academic Press, Inc. surfaces, isotherm models are derived from a fundamental
Key Words: adsorption; Freundlich isotherm; isotherm; isotherm integral equation relating the experimental isotherm, the
model; Jovanovic isotherm. adsorption energy distribution, and the local isotherm (3).
Independently, a number of empirical and semi-empirical
isotherm models have been suggested to account for the
INTRODUCTION adsorption behavior on heterogeneous surfaces (3). Single-
component isotherms have been derived from a differential
The use of adsorption processes for large-scale separa- equation relating surface coverage and pressure. This ap-
tions or purifications is becoming common in industry. proach was first suggested by Schmidt (4), later expanded
The recent development of industrial-scale preparative by Tóth (5) and Misra (6), and recently applied again by
chromatography is contributing to further increase the im- Tóth (7). It allows the derivation of thermodynamically con-
portance of adsorption-based processes. The design and sistent isotherms for both homogeneous and heterogeneous
the optimization of an implementation of an adsorption- surfaces.
based separation process require the ability to characterize In the present paper, we apply the latter approach to derive
accurately the adsorption equilibria involved and their de- a semiempirical Jovanovic–Freundlich isotherm model for
pendence on the experimental conditions ( 1 ) . Although, the monolayer, single-component adsorption on heteroge-
in principle, the competitive isotherms of all the feed com- neous surfaces. We use this model to account for experimen-
tal adsorption data published recently, regarding chlorinated
1
Author to whom correspondence should be sent, at the University of hydrocarbons on silica gel, at different temperatures (8, 9),
Tennessee. and we compare the quality of the representation obtained
57 0021-9797/96 $18.00
Copyright q 1996 by Academic Press, Inc.
All rights of reproduction in any form reserved.

AID JCIS 4406 / 6g16$$$181 09-25-96 13:47:31 coida AP: Colloid


58 QUIÑONES AND GUIOCHON

TABLE 1
Selection of Adsorption Isotherm Models

Model UÅ l(u, p) Eq. no. Ref.

Henry Kp K/(1 0 Kp) [2] (2)


Langmuir Kp/(1 / Kp) K/(1 / Kp) [3] (13)
Jovanovic 1 0 e0Kp K [4] (14)
Quadratic (Kp / K2hp2)/(1 / 2Kp / K2hp2) K(1 / 2Khp / K2hp2)/[1 / 3Kp / K2(2 / h)p2 / K3hp3] [5] (15)
Fowler Kp/(e0xu / Kp) Ku/(Kp / xu2e0xu) [6] (16)
Freundlich (ap)n an(ap)n01/[1 0 (ap)n] [7] (17)
Langmuir–Freundlich (ap)n/[1 / (ap)n] n(ap)n01/[1 / (ap)n] [8] (18)
Tóth Kp/[1 / (Kp)n]1/n K{1 0 (Kp)n[1 / (Kp)n]01}/{[1 / (Kp)n]1/n 0 Kp} [9] (5)
Misra 1 0 [1 / (k 0 1)Kp]1/(k01) K/[1 / (k 0 1)Kp] [10] (6)
Radke–Prausnitz Kp/[1 / (Kp)n] [K / (Kp)n(K 0 n)]/{[1 / (Kp)n] [1 / (Kp)n 0 Kp]} [11] (19)
2 2
Dubinin–Radushkevich e0a[ln(b/p)] 2a ln(b/p)/{ea[ln(b/p)] 0 1} [12] (20)

Note. For homogeneous-surface models, U is the local surface coverage, u; for the other models, it is the overall coverage, ut ; see Eq. [1]. K is the
Henry constant. h and x are the adsorbate–adsorbate interaction parameters. n is the heterogeneity parameter. a, b, and k are other parameters which
appear in some models.

with a variety of other theoretical and semiempirical iso- (2, 3). A last approach involves the assumption of a local
therm models previously reported (5, 10, 11). and an overall isotherm model (3). It is also arbitrary, as
are the choices of the two models.
THEORY We note that Langmuir (13) had already proposed de-
An heterogeneous surface is characterized by a distribu- scribing the adsorption on an heterogeneous surface as the
tion of adsorption sites which have different adsorption ener- sum of a finite number of classical Langmuir isotherms (Eq.
gies. For homotattic sites, a local adsorption isotherm model [3], Table 1), each of them characterizing adsorption on an
is assumed. This model refers to an homogeneous surface. homogeneous patch of surface. This, in effect, introduces a
The experimental or global isotherm is obtained (3) by the discrete energy distribution. The biLangmuir isotherm, intro-
integral equation duced by Graham (21) and used by Laub (22) in gas chro-
matography, is a particular case of this approach. It has

*
`
q proven to be extremely successful in the description of the
ut (p) Å Å u(p, e )F( e )d e, [1] adsorption of enantiomers on chiral phases (23, 24). This
qs 0
result is explained by the physical nature of these interac-
where ut (p) is the overall surface coverage (excess adsorp- tions. Both enantiomers undergo the same nonselective inter-
tion) of the heterogeneous surface by the monolayer of ad- actions with most patches of surface, while enantioselective
sorbate, a function of the partial pressure p of the adsorbate interactions are highly selective, but take place in well-de-
in the bulk, q is the amount of adsorbate adsorbed at equilib- fined sites which are localized and relatively far from each
rium per unit amount of adsorbent, qs is the monolayer capac- other, ensuring the lack of adsorbate–adsorbate interactions
ity, u(p, e ) is the local adsorption isotherm for homottatic on the enantioselective sites. Furthermore, adsorption energ-
sites with an adsorption energy e, and F( e ) is the adsorption ies on these sites are high, so saturation is achieved at low
energy distribution. Equation [1] is the fundamental equation values of the concentration and the activity coefficients re-
in the theory of adsorption on heterogeneous surfaces. Be- main practically constant.
cause this Fredholm integral equation is ill-posed, it does A significant number of overall isotherms of adsorption
not have a unique solution (3). Attempts at the derivation on heterogeneous surfaces were originally proposed for em-
of the adsorption energy distribution from the experimental pirical or semiempirical reasons, with the sole aim of de-
isotherm, using numerical solutions and an assumed local scribing accurately and as simply as possible the adsorption
isotherm, have not led to convincing results but have demon- behavior of real systems. Examples of these models are the
strated the difficulties, usually underestimated, of collecting Freundlich (17), the Langmuir–Freundlich (18), the Tóth
proper experimental data (12). Another approach consists (5), the Radke–Prausnitz (19), and the Dubinin–Radush-
in deriving closed-form integrals of Eq. [1], assuming simple kevich (20) equations. These isotherm models were derived
functional dependences for u(p, e ) and F( e ) and fitting ex- to use Eq. [1] to calculate the adsorption energy distribution
perimental results to these equations in an effort to identify from a combination of a local and an overall adsorption
the relevant parameters. A number of such solutions are isotherm. Most studies concerning physical adsorption on
listed in Table 1. This list is by no means comprehensive heterogeneous surfaces were carried out using the Langmuir

AID JCIS 4406 / 6g16$$$182 09-25-96 13:47:31 coida AP: Colloid


NEW JOVANOVIC–FREUNDLICH ISOTHERM MODEL 59
model (13) to describe the local isotherm, i.e., the adsorption tained the following overall adsorption isotherm for a Jova-
behavior on homotattic sites (3). Many empirical equations novic local isotherm, using a g-function to characterize the
for the overall isotherm can be obtained with local Langmuir distribution of the Henry constant, K, on the heterogeneous
behavior if an appropriate choice is made for the adsorption surface
energy distribution. To account for lateral adsorbate–ad-
sorbate interactions, however, the Langmuir model should
be replaced as the local isotherm in Eq. [1] by another model ut (p) Å 1 0 F G r
r/p
g/1
e 0Kp , [15]
which takes into account this additional source of nonideal
behavior, such as the Fowler (16), Moreau (15), or Ruthven where r and g are the parameters of the energy distribution.
(2) isotherm models. This model, however, has not been widely applied.
Two important overall equations are the Tóth (5) and the Recently, Hines et al. (11) developed an overall adsorp-
UNILAN (10) models. For correlation purposes, we have tion isotherm model in which the local isotherm is given by
used in this work the equation of the Tóth model (5) as the Jovanovic equation (Eq. [4], Table 1) and the energy
applied by Valenzuela and Myers (1) distribution is represented by a Morse type distribution. They
obtained the equation
p

F G
ut (p) Å , [13]
(b / p n) 1 / n K1K3 1 K2
ut (p) Å 1 0 0 , [16]
K3 0 K1K2 K1 / p K3 / p
where b is an adjustable parameter and n is the heterogeneity
parameter [3]. The UNILAN equation is based on a uniform where K1 , K2 , and K3 are the parameters of the energy distri-
distribution of the adsorption energies and a local Langmuir bution. K2 is usually very small. We note that if we neglect
isotherm (10). Its equation is it, Eq. [16] becomes identical with the Langmuir isotherm,
with K Å 1/K1 .
ut (p) Å
1
2n
ln F
C / pe /n
C / pe 0 n G , [14]
The derivation of adsorption isotherm equations from dif-
ferential equations has been reported for a long time. In spite
of several important papers published this century (4–7), it
where C is an adjustable parameter, which, in principle, is not a most popular approach. Initially, Schmidt (4) pro-
should be equal to 1/K, where K is the low-pressure equilib- posed a semi-empirical treatment of physical adsorption and
rium constant or Henry constant. Note that the heterogeneity derived a single-component adsorption isotherm using the
parameter, n, is a parameter of the adsorption isotherm but relationship
is also a parameter of the adsorption energy distribution (3)
and, thus, is not the same for different models. For the mod- dx
Å D(S 0 x), [17]
els which are reducible to the Langmuir model (e.g., Lang- dC
muir–Freundlich, Tóth, generalized Freundlich, Redlich–
Peterson models), n is between 0 and 1. For other models where x is the amount of component adsorbed at equilibrium,
(e.g., Freundlich, UNILAN, Dubinin) which can fit also type C is the concentration of the component in the bulk, S is the
III isotherm data, n may take negative values. For some maximum amount adsorbable, and D is a constant. Integra-
models (e.g., Freundlich, Langmuir–Freundlich, Tóth, Jova- tion of Eq. [17] gives
novic–Freundlich, Redlich–Peterson models), the surface
is homogeneous when n is equal to 1. For other models x Å S(1 0 e 0DC ). [18]
(e.g., UNILAN), the surface is homogeneous when n is
equal to 0. Later, Tóth (5) derived several adsorption isotherm models
Equations [13] and [14] reduce to the Langmuir equation using one of the equations
for values of n equal to 1 and 0, respectively. Misra (25)
used the Jovanovic model (14) to represent the local adsorp- du u
tion isotherm in the solution of Eq. [1], in combination with Å [19]
dp [ F1 ( u ) / 1]p
different energy distributions. At low pressures, or for small
values of n, the solutions reduce to the Shlygin–Frumkin du u
Å , [20]
(26) or Freundlich (17) isotherms, respectively. It seems dp [ F2 (p) / 1]p
that the use of either the Jovanovic or the Langmuir models
to account for the local behavior of the adsorbate permits where F1 ( u ) and F2 (p) are functions of the surface coverage
one to obtain similar overall isotherms for analogous energy and the adsorbate partial pressure, respectively. For example,
distributions (3). Jaroniec and Piotrowska (27) have ob- Eq. [13] was derived from Eq. [19].

AID JCIS 4406 / 6g16$$$182 09-25-96 13:47:31 coida AP: Colloid


60 QUIÑONES AND GUIOCHON

Later, Misra (6) derived several other adsorption isotherm du


Å F( u, p), [26]
models from the differential equation dp

du where F( u, p) is a function of both the surface coverage


Å K(1 0 u ) k , [21]
dp and the partial pressure of the adsorbent. Accordingly, we
consider the following equation
where k is a constant. The general solution of Eq. [21] is Eq.
[10] (Table 1). When k § 1, certain theoretical conditions of F( u, p) Å (1 0 u ) l ( u, p). [27]
monomolecular adsorption are met. However, Eq. [21] is
empirical. When k assumes values of 0, 1, or 2, Eq. [21] This relationship can be rewritten as
reduces to the Henry (2), Jovanovic (14), or Langmuir
(13) isotherms, respectively. It is worth noting the similarity 0 d(1 0 u )
between Eq. [18] and the Jovanovic model (14), considering Å l ( u, p). [28]
(1 0 u )dp
the proportionality between q and x on the one hand, between
qs and S on the other. However, the concept of monolayer
adsorption was first introduced by Langmuir (13) while Eq. Equation [28] was selected because l ( u, p) has a physical
[4] was originally derived by Jovanovic (14), using a kinetic meaning. It is the rate of decrease of the fraction of the
approach, and later by Jaroniec (28), using statistical ther- surface which is unoccupied by the adsorbate molecules.
modynamics. Most of the classical isotherms can be analyzed using Eq.
Misra (6) considered that the functional dependence of [28]. The functions l ( u, p) were derived for these isotherms
the variation of the surface coverage with respect to the on homogeneous surfaces (Henry (2), Langmuir (13), Jova-
adsorbate partial pressure at constant temperature is given novic (14), quadratic (15), and Fowler (16)) and on hetero-
by geneous surfaces (Freundlich (17), Langmuir–Freundlich
(18), Tóth (5), Radke–Prausnitz (19), and Dubinin–Ra-
du dushkevich (20)). They are listed in Table 1. Among these
Å K f (u ), [22] equations, we can distinguish those which do not take the
dp adsorbate–adsorbate interactions into account (Henry,
Langmuir, and Jovanovic) and those which do (Fowler and
or rather, considering the proportion of uncovered surface quadratic). We also observe that for all models which reduce
area, to the Henry law at low partial pressure, the L-function,
l ( u, p), tends toward K when p tends toward 0. Thus, the
du
Å K f * (1 0 u ). [23] limit of the L-function for low values of p (hence of u ) is
dp K, the low pressure equilibrium constant. This conclusion
is also obvious from Eq. [28].
Equation [21] is one of the simplest forms of Eq. [23]. Isotherms which are explicit with respect to pressure give
Recently, Tóth (7) proposed the derivation of thermodynam- values of l ( u, p) which depend only on the adsorbate partial
ically consistent isotherm equations for homogeneous or het- pressure. Implicit models (e.g., the Fowler isotherm) give
erogeneous surfaces from one of the differential equations values of l ( u, p) which are functions of both u and p. For
several models (Langmuir, Fowler, quadratic, and Misra),
du u l ( u, p) decreases with increasing partial pressure of the
Å [24]
dp p c( u ) adsorbate. However, in the case of the Jovanovic isotherm,
l ( u, p) remains constant and equal to K. For the other
or isotherms, the relationship is more complex and depends on
the specific values of the parameters of these models. Thus,
du u this aspect must be discussed for specific cases. It will be
Å , [25]
dp p f(p) analyzed later.
Consideration of the L-functions in Table 1 suggests try-
where c( u ) and f(p) are functions of the surface coverage ing the following function
ratio or the partial pressure, respectively, which are derived
from an incorrect isotherm model, previously obtained em- l ( u, p) Å an(ap) n01 , [29]
pirically (7). It is clear that Eqs. [19] and [20] can be
reduced to Eqs. [24] and [25], respectively. Consideration an expression which is also found in the numerator of the
of Eqs. [24] and [25] suggests that a more general relation- Freundlich equation. On the other hand, the numerator of the
ship is Langmuir–Freundlich equation is also similar to the RHS of

AID JCIS 4406 / 6g16$$$183 09-25-96 13:47:31 coida AP: Colloid


NEW JOVANOVIC–FREUNDLICH ISOTHERM MODEL 61
Eq. [29]. Substitution of Eq. [29] into Eq. [28] and integra- The relationship between the functions F( j ) and C( z ) is
tion gives defined as
n
ut (p) Å 1 0 e 0 ( ap ) . [30] F( ze 0ip) 0 F( ze ip)
C( z ) Å . [33]
2ip
This model can be considered as a combination of the Jova-
novic and the Freundlich isotherm models. For n Å 1, it In the version of Sips procedure considered here (3), the
reduces to the Jovanovic isotherm for homogeneous sur- position of the energy distribution function is characterized
faces. For large values of the partial pressures, ut (p) tends by a minimum adsorption energy, ea . The parameter j is
toward unity; a monolayer coverage is achieved. For low defined as
values of the partial pressures, the isotherm equation is
equivalent to the Freundlich isotherm (17). Accordingly, 1
the model cannot account for a Henry law (2) region. On jÅ1/ , [34]
ap
the other hand, the coverage ratio is given as an explicit
function of the partial pressure of the adsorbate, which is a where the constant a is the same parameter as in Eq. [30].
considerable advantage for the determination of isotherms Its temperature dependence is defined as
by the chromatographic methods of frontal analysis or pulse
on a plateau (29) or for their use in the calculation of elution a Å K 0e ea/ RT , [35]
profiles or other concentration signals at high concentrations,
under nonlinear conditions (30). Note that Eq. [30] is equiv-
where K 0 is the preexponential factor, R the ideal gas con-
alent to a cumulative Weibull distribution (42).
stant, and T the absolute temperature. The parameter z is
This new model can be used to represent experimental
defined as
data concerning adsorption isotherms on heterogeneous sur-
faces composed of several subsurfaces which are themselves
z Å e (e0 ea) / RT 0 1. [36]
heterogeneous, so the overall isotherm is the sum of several
expressions such as Eq. [30], each one written for one of
the subsurfaces. Like other models developed for the de- The relationship between the global adsorption isotherm and
the function F( j ) is given by
scription of monolayer adsorption on heterogeneous surfaces

S D
without lateral interactions, this model can, in principle, be
modified to describe monolayer, single-gas adsorption with 1
F ( j ) Å ut , [37]
lateral interactions, multilayer single-gas adsorption without a( j 0 1)
or with lateral interactions, the adsorption of multicompo-
nent gas mixtures, or the adsorption of multisolute dilute while the energy distribution function is related to the func-
solutions (3). For example, the extension of the model to tion C( z ) by
describe competitive adsorption can be accomplished using
models of the adsorbed solution theory (31) or the method C( z ) Å RTF( e ). [38]
described by Jaroniec et al. (32).
The integral equation [1] can be solved analytically with Considering Eqs. [30], [34], and [37], we see that the
respect to the energy distribution function in the particular function F( j ) for the Jovanovic–Freundlich isotherm model
case in which one assumes a Langmuir local adsorption is
isotherm and a global adsorption isotherm given by Eq. [30].
0n
The solution is derived using the procedure suggested by F( j ) Å 1 0 e 0 ( j01 ) . [39]
Sips (18, 33) and was recently summarized by Jaroniec (3).
In this case, Eq. [1] is rewritten as Combining Eqs. [33] and [39] gives the value of the func-
tion C( z ),

*
`
pK( e )
ut (p) Å F( e )d e. [31] ip01 ) 0 n 0 ip01 ) 0 n
0 1 / pK( e ) e 0(ze 0 e 0(ze
C( z ) Å . [40]
2ip
Sips (18, 33) transformed Eq. [31] into the following inte-
gral, known as the Stieltjes transform: Using Euler’s formula, Eq. [40] can be transformed into

*
` 0 ncos pn
C( z ) e 0 ( z /1 ) sin[( z / 1) 0 nsin pn]
F( j ) Å dz . [32] C( z ) Å . [41]
0 j/z p

AID JCIS 4406 / 6g16$$$183 09-25-96 13:47:31 coida AP: Colloid


62 QUIÑONES AND GUIOCHON

TABLE 2 tion to compare the performance of these models in account-


Summary of the Adsorption Data Analyzed in This Study ing for these data with that of the Jovanovic–Freundlich
model (Eq. [30]). The Tóth and Unilan models were chosen
T Number of in this work because they are considered as the best all-
System Adsorbate (K) data pointse Ref.
around three-constant isotherm models for microporous ad-
1 CH3Cl 288 15 (8) sorbents and they were those selected for the systematic
2 CH3Cl 293 16 (8) correlations of the large data bank of adsorption isotherms
3 CH3Cl 298 21 (8) on heterogeneous surfaces (1). The Hines model is one of
4 CH2Cl2 288 15 (8) the most recent models for heterogeneous adsorbents which
5 CH2Cl2 293 16 (8)
6 CH2Cl2 298 15 (8) assumes that the local adsorption isotherm on homotattic
7 CHCl3 288 12 (8) sites is described by the Jovanovic model (14).
8 CHCl3 293 12 (8) The nonlinear regression analysis of the models was car-
9 CHCl3 298 11 (8) ried out using a fitting procedure based on Marquardt algo-
10 CCl4 288 12 (8) rithm (34), which minimizes the residual sum of the squares
11 CCl4 293 12 (8)
12 CCl4 298 14 (8) of the differences between the experimental data and the
13 C2H3Cl3 288 15 (9) model calculations. The estimates of the parameters present
14 C2H3Cl3 293 17 (9) in the models are given at the asymptotic 95% confidence
15 C2H3Cl3 298 18 (9) interval. Conventional use of the Fisher distribution is not
16 C2Cl4 288 15 (9) possible in this case because there is only one data point for
17 C2Cl4 293 15 (9)
18 C2Cl4 298 18 (9) each value of the partial pressure. As suggested in (35), for
each model and each set of experimental data, the Fisher
a
Number of data points in the original work plus the coordinate origin, parameter was calculated according to
minus one outlier in the case of system 2.

m 0 l ( imÅ1 (qexp,i 0 qexp ) 2


Fcalc Å , [44]
Substituting Eqs. [36] and [38] into Eq. [41] gives the final m 0 1 ( imÅ1 (qexp,i 0 qt ,i ) 2
expression for the adsorption energy distribution
F( e ) where

e 0[e
(e0 ea) /RT ] 0 ncos pn
sin{e 0 [ e
(e0 ea) /RT ] 0 n
sin pn )} qexp,i are the experimental values of the solid phase con-
Å . [42] centration of the adsorbate for a given system,
pRT
qexp is the mean value of the data, qexp,i for a given system,
This distribution includes an exponential decay at high ad- qt ,i is the estimate of the solid phase concentration of the
sorption energies and a maximum value for e Å ea which is adsorbate by a given model,
equal to l is the number of adjusted parameters in the model, and
m is the number of experimental data for a given system.
e 0cos pnsin[sin ( pn /e)]
F( ea ) Å . [43]
pRT Equation [ 44 ] is different from the conventional Fisher
equation. The second factor in its RHS contains the sum
RESULTS of residuals in the denominator. Thus, the higher Fcalc , the
better the model correlates the experimental data. The first
The adsorption data used in the present work to illustrate factor in the RHS of Eq. [ 44 ] decreases with increasing
the new isotherm model described above, in Eq. [30], and number of parameters of the model. This equation allows
to show its potential usefulness were reported by Hines et the comparison of models having different numbers of
al. (8, 9).2 These data regard the adsorption of several chlori- parameters ( 35 ) .
nated hydrocarbons on a microporous silica gel at different Tables 3–6 summarize the results of the nonlinear regres-
temperatures (see Table 2). The surface should be consid- sion analysis of the models evaluated in this study. These
ered as heterogeneous because the heat of adsorption tends tables report the parameter estimates, their confidence inter-
to decrease with increasing concentration of the compounds vals, and the Fisher values for each set of experimental data.
(8, 9). These experimental data were fitted to the Tóth (5), For systems 13 to 18, we calculated the average absolute
UNILAN (10), and Hines (11) models of isotherm adsorp- deviations (AAD) produced by the Jovanovic–Freundlich
model, in order to compare these deviations with those gen-
2
The point corresponding to the isotherm origin was added to the original erated by the fit of the data to the Langmuir model (13) and
experimental data to perform the fitting of the data to the different isotherm the Vacancy Solution Model based on the use of the Flory–
models used in this work. Huggins equation (36) for the activity coefficient reported

AID JCIS 4406 / 6g16$$$184 09-25-96 13:47:31 coida AP: Colloid


NEW JOVANOVIC–FREUNDLICH ISOTHERM MODEL 63
TABLE 3a TABLE 5a
Regression Results for the Jovanovic–Freundlich Modelb Regression Results for the UNILAN Model

System System
no. qs a 1 102 n Fcalc no. qs n c Fcalc

1 5.96 { 0.26 0.42 { 0.05 0.74 { 0.04 1517.0 1 7.4 { 1.2 1.6 { 0.9 226 { 96 531.13
2 5.77 { 0.38 0.38 { 0.06 0.77 { 0.06 850.1 2 7.1 { 1.2 1.1 { 1.1 243 { 96 460.66
3 5.46 { 0.28 0.36 { 0.05 0.80 { 0.05 852.5 3 6.8 { 0.9 1.0 { 0.9 268 { 82 538.61
4 5.26 { 0.29 2.4 { 0.4 0.66 { 0.06 401.7 4 6.3 { 1.1 2.0 { 1.1 35 { 19 164.48
5 5.12 { 0.33 2.0 { 0.4 0.71 { 0.08 239.6 5 6.0 { 1.1 1.6 { 1.3 42 { 20 119.62
6 5.02 { 0.42 1.6 { 0.4 0.70 { 0.10 152.6 6 5.9 { 1.2 1.6 { 1.6 51 { 30 82.13
7 3.67 { 0.14 8.3 { 1.3 0.60 { 0.06 439.0 7 4.0 { 0.4 2.1 { 1.0 8 { 2.7 171.22
8 3.59 { 0.17 7.2 { 1.4 0.62 { 0.07 332.8 8 4.0 { 0.5 2.0 { 1.0 9 { 3.6 151.40
9 3.51 { 0.19 5.8 { 1.2 0.65 { 0.09 229.6 9 3.9 { 0.5 1.8 { 1.4 11 { 5.1 97.88
10 2.91 { 0.072 13.0 { 1.3 0.64 { 0.04 1154.6 10 3.2 { 0.2 1.7 { 0.7 5.3 { 1.2 379.79
11 2.78 { 0.086 11.6 { 1.4 0.76 { 0.06 547.8 11 3.1 { 0.2 1.3 { 1.0 6.0 { 1.5 227.25
12 2.72 { 0.058 9.6 { 0.8 0.69 { 0.04 923.2 12 3.0 { 0.19 1.4 { 0.7 7.2 { 1.4 287.98
13 3.35 { 0.041 42 { 4.1 0.60 { 0.06 392.8 13 3.45 { 0.04 0.4 { 1.7 1.4 { 0.11 645.32
14 3.33 { 0.062 25 { 2.8 0.57 { 0.06 250.6 14 3.48 { 0.08 1.63 { 0.6 2.3 { 0.3 246.37
15 3.33 { 0.052 14 { 1.2 0.58 { 0.04 489.1 15 3.54 { 0.10 1.9 { 0.5 4.0 { 0.5 290.06
16 2.95 { 0.062 260 { 28 0.37 { 0.03 1530.0 16 3.03 { 0.06 3.4 { 0.4 0.16 { 0.015 1336.01
17 2.87 { 0.047 208 { 17 0.41 { 0.03 1638.7 17 2.97 { 0.07 3.0 { 0.4 0.22 { 0.02 928.19
18 2.76 { 0.090 144 { 24 0.48 { 0.06 160.2 18 2.9 { 0.15 2.6 { 0.8 0.35 { 0.08 106.58

a
Units: qs , mmole/g; a, (mm Hg)01, n and Fcalc , dimensionless. a
Dimensions: as in Table 3, with the addition of c, mm Hg.
b
Model equation, Eq. [30].

Table 7 reports the AAD values calculated according to


by Hines (9). The AAD were calculated according to the Eq. [45] for the selected models. Figure 1 shows the best
expression fit of the Jovanovic–Freundlich model to the adsorption data
of methyl chloride. Table 8 and Fig. 2 illustrate the tempera-
Éqexp,i 0 qt ,i É ture dependence of the parameter a in Eq. [30], according
AAD Å 100 . [45] to the linear form of Eq. [35] for the six chlorinated hydro-
qexp,i
carbons studied. Figure 3 shows the pressure dependence of
the function l ( u, p) for the Jovanovic–Freundlich (Eq.
TABLE 4a [30]) and Tóth (Eq. [13]) models in the case of chlorometh-
Regression Results for the Toth Model ane at 298 K. Figure 4 represents the energy distribution
System functions calculated according to Eq. [42] for dichlorometh-
no. qs b n Fcalc ane at 288, 293, and 298 K using the parameters of Tables 2,
3, and 8. Figure 5 shows the adsorption energy distributions
1 8.1 { 1.5 30 { 33 0.69 { 0.19 607.46 calculated according to Eq. [42] for the six chlorinated hy-
2 7.5 { 1.6 66 { 108 0.80 { 0.3 480.92
drocarbons studied at 288 K.
3 7.1 { 1.1 85 { 110 0.83 { 0.2 555.46
4 7.0 { 1.7 5 { 4.5 0.59 { 0.2 177.61
5 6.7 { 1.7 8 { 10 0.66 { 0.3 124.11 DISCUSSION
6 6.6 { 2 9 { 14 0.65 { 0.4 84.69
7 4.4 { 0.8 2 { 1.5 0.60 { 0.24 162.12
8 4.3 { 0.8 2.5 { 2 0.63 { 0.26 144.90
The results presented in Tables 3–6 show that the parame-
9 4.1 { 0.9 3.5 { 4 0.68 { 0.36 95.00 ters of the Jovanovic–Freundlich isotherm model were iden-
10 3.4 { 0.4 2.1 { 0.9 0.69 { 0.17 378.28 tified with a smaller error than those of all the other models
11 3.2 { 0.4 3.4 { 2 0.81 { 0.2 225.79 studied. The Fisher values obtained for all the systems stud-
12 3.2 { 0.3 3.4 { 1.8 0.76 { 0.18 287.39 ied are larger for the regression of the Jovanovic–Freundlich
13 3.45 { 0.06 1.4 { 0.3 0.99 { 0.13 643.86
14 3.6 { 0.15 1.4 { 0.5 0.77 { 0.14 235.90
isotherm model than for those of the Tóth and UNILAN
15 3.7 { 0.2 1.7 { 0.6 0.69 { 0.12 269.36 models, with the exception of 1,1,1-trichloroethane at 288
16 3.3 { 0.2 0.19 { 0.010 0.45 { 0.09 892.09 K, showing that the data fit better to the first than to the
17 3.2 { 0.2 0.22 { 0.018 0.49 { 0.10 560.05 other two models. Note that the fit of the data for 1,1,1-
18 3.1 { 0.4 0.29 { 0.051 0.53 { 0.18 90.17 trichloroethane at 288 K to the Tóth and UNILAN models
a
Dimensions: as in Table 3, with the addition of b, dimensionless.
is also better than their fit to Hines model. Comparison of
the results obtained with the Jovanovic–Freundlich and the

AID JCIS 4406 / 6g16$$$185 09-25-96 13:47:31 coida AP: Colloid


64 QUIÑONES AND GUIOCHON

TABLE 6a
Regression Results for the Hines et al. Model

System
no. qs K1 K2 1 104 K3 Fcalc

1 6.78 { 0.03 200 { 4.5 2.9 { 1.0 01.5 { 0.7 51725


2 6.69 { 0.07 224 { 9 2.0 { 0.19 02.2 { 0.8 14943
3 6.49 { 0.06 252 { 8 1.34 { 0.15 01.9 { 0.4 10373
4 5.79 { 0.25 37 { 10 10 { 52 00.4 { 2 461
5 5.77 { 0.33 45 { 15 8 { 40 00.5 { 3 236
6 5.69 { 0.45 59 { 26 0 { 57 0.05 { 4 149
7 3.88 { 0.13 9 { 3.7 49 { 51 00.82 { 1.5 547
8 3.86 { 0.18 12 { 5.4 33 { 112 00.36 { 1.7 300
9 3.78 { 0.20 13 { 6 30 { 45 00.92 { 1.5 229
10 3.11 { 0.03 5.7 { 0.5 34 { 5 00.42 { 0.2 4813
11 3.03 { 0.09 6.6 { 1.5 12 { 9 00.22 { 0.5 564
12 2.96 { 0.05 7.9 { 1.0 13 { 7 00.25 { 0.3 1244
13 3.45 { 0.06 1.5 { 3.9 0 { 11000 0.0 { 6 600
14 3.46 { 0.06 3.5 { 2.0 0 { 590 0.00 { 0.5 350
15 3.48 { 0.04 6.0 { 1.4 0 { 150 0.00 { 0.6 942
16 2.98 { 0.05 0.04 { 0.04 023 { 29b 1.09 { 0.6 1807
17 2.91 { 0.02 0.00 { 0.04 0 { 4900b 0.81 { 0.2 5834
18 2.85 { 0.08 0.00 { 0.13 0 { 264000b 0.88 { 0.6 233

a
Dimensions: as in Table 3, with the addition of K1 , K3 , mm Hg. K2, dimensionless.
b
In this case the parameter K2 is not multiplied by 10,000.

Hines models shows that the performance of the Hines model than those reported (9) for the Langmuir model and for the
is superior for a majority of systems (1 to 3, 10, 12 to 15, Vacancy Solution Model (36). The only exception is for
17, and 18). By contrast, the values of the Fisher parameters 1,1,1-trichloroethane at 288 K. In this case, the adsorption
are close for several other systems (4 to 9, 11, and 16), data fit better to the Langmuir model. All these results dem-
indicating similar model performance. It seems that the ori- onstrate the suitability of the Jovanovic–Freundlich iso-
gin of the isotherm was not included in the initial regression therm model to correlate the adsorption data studied here.
of the data (8, 9). If we include this point in the data set, The Tóth, UNILAN, and Hines models reduce to the
however, our estimates of the parameters of the Hines iso- Henry law at low pressures, whereas the Jovanovic–Freund-
therm model differ from those reported initially (8, 9). It
should be noted that the errors associated with the parameter
estimates are higher for the Hines model, especially for K2
and K3 , than for the Jovanovic–Freundlich model. This latter
model is also simpler than the former, having three parame-
ters instead of four. The results of Table 7 show that the
AAD values are lower for the Jovanovic–Freundlich model

TABLE 7
Average Absolute Deviations Observed with the Jovanovic–
Freundlich Model (Eq. [30]), the Langmuir Model (Eq. [3] and
Ref. (9)), and the Vacancy Solution Model Based on the Flory–
Huggins Activity Coefficient Equation (VSM–FH, Ref. (9))

System
no. Eq. [30] Eq. [3] VSH–FH

13 1.22 1.06 2.05


14 1.90 2.54 3.13
15 1.52 3.21 2.18
FIG. 1. Adsorption isotherms of methyl chloride on silica gel at differ-
16 0.61 3.25 2.33
17 0.67 3.44 2.22 ent temperatures. Symbols denote experimental data from Ref. (8). Solid
18 1.88 3.87 4.47 lines were calculated with Eq. [30] using the values of the parameters
given in Table 3.

AID JCIS 4406 / 6g16$$$185 09-25-96 13:47:31 coida AP: Colloid


NEW JOVANOVIC–FREUNDLICH ISOTHERM MODEL 65
TABLE 8
Regression Results for the Temperature Dependence
of the a Parameter (Eq. [35])

Standard Regression
Adsorbate ln K0 ea error coefficient

CH3Cl 010 { 9 12 { 23 0.0178 0.988626


CH2Cl2 016 { 10 29 { 24 0.0187 0.997822
CHCl3 013 { 17 26 { 42 0.0326 0.992086
CCl4 011 { 15 21 { 38 0.0295 0.990129
C2H3Cl3 033 { 7 78 { 17 0.0130 0.999858
C2Cl4 017 { 30 43 { 71 0.0556 0.991744

lich model does not. Apparently, the experimental data avail-


able do not allow sufficient coverage of the Henry law re-
gion. The success of the Jovanovic–Freundlich model to
better correlate the data may be associated with the fact
that these data are concentrated mainly in the regions of FIG. 3. L-functions, l ( u, p), calculated for the Tóth and the Jovanovic–
intermediate and high coverage ratio, so the main drawback Freundlich models using the parameters of the adsorption of methyl chloride
of the Jovanovic–Freundlich model, its inability to correctly at 298 K.
reproduce the Henry law region, has little influence on the
quality of the data fit observed. Besides, it is known that
Fritz–Schlünder (39) models. None of them reduces to the
any three-constant isotherm equation is unable to provide a
Henry law at low pressures.
precise fit of the experimental data at both high and low
The impossibility of accounting for the low partial pres-
surface coverages (1). The fact that the Jovanovic–Freund-
sure, Henry law region of the isotherm is a drawback of
lich model does not reduce to the Henry law is a drawback
these isotherm models which cannot be ignored nor corrected
which has a certain importance for a variety of reasons,
for single-component problems. However, there is an im-
although it shares this property with other important isotherm
portant application of single-components in which this diffi-
models widely applied to describe adsorption isotherms on
culty can be alleviated; it is the modeling of competitive
heterogeneous surfaces. Such is the case, among others, of
isotherms. In this case, a new approach has been recently
the Freundlich (17), Langmuir–Freundlich (18), general-
proposed (40), based on models of the adsorbed solution
ized Freundlich (33), Marczewski–Jaroniec (37), Dubi-
theory. It allows ignoring single-component behavior in the
nin–Radushkevich (20), Dubinin–Astakhov (38), and
Henry law region and gives results which are as good as the
classical ones. Thus, it removes the sensitivity of adsorbed
solution theory models to the type of isotherm models used
to fit single-component equilibrium data in the Henry law
region. In compensation, the method introduces new parame-
ters that must be evaluated from binary equilibrium data,
thus eliminating the possibility of predicting multicompo-
nent equilibrium isotherms from single-component data.
This limitation is relative, however, because multicomponent
data must always be generated, as tests of the model and,
in the case of actual adsorbed solution theory models, for
correlating adsorbed phase activity coefficients (31).
In this case, there are two possibilities, either to measure
single-component isotherms down to the Henry law region or
to measure binary equilibrium under constant total pressure.
From a practical point of view, the second solution is often
better, especially when strongly adsorbed compounds are
concerned. Furthermore, the experimental methods of deter-
mination of multicomponent equilibrium isotherms for both
FIG. 2. Temperature dependence of the parameter a of Eq. [30] for gas mixtures and solutions have been significantly improved
the different adsorbates. in recent years, mainly because of the availability of com-

AID JCIS 4406 / 6g16$$$185 09-25-96 13:47:31 coida AP: Colloid


66 QUIÑONES AND GUIOCHON

(9), as shown in Table 7. Nevertheless, the surface is not


homogeneous, as indicated by the decreasing trend of the
values of the isosteric heats of adsorption (9). The statistical
similarity between the values of the heterogeneity parame-
ters determined from the Jovanovic–Freundlich and the Tóth
isotherm models (see Tables 3 and 4) should be noted. The
heterogeneity parameters of the Jovanovic–Freundlich
model are nearly independent of the temperature. The tem-
perature dependence of the heterogeneity parameter is not
clearly defined and the temperature dependence of the ad-
sorption energy distribution is also small (3). Many empiri-
cal and semiempirical adsorption isotherm equations for het-
erogeneous surfaces produce adsorption energy distributions
that are temperature independent (3).
As shown in Fig. 4, there is a great similarity between
the adsorption energy distributions derived for dichlorometh-
ane at the different temperatures studied. This result obtained
FIG. 4. Adsorption energy distributions calculated for dichloromethane with the Jovanovic–Freundlich model was also obtained by
at 288, 293, and 298 K. Hines et al. (8). The values obtained for the heterogeneity
parameter of the different compounds are close. The hetero-
puter data acquisition and processing (41). Moreover, when geneity parameter of the Jovanovic–Freundlich model
data have been acquired in the Henry law region, it is possi- shows near independence of the nature of the compound.
ble to correlate them separately. This confirms that silicagel exhibits the same degree of het-
The importance of models that fit single-component ad- erogeneity for the majority of the chlorinated hydrocarbons
sorption data well at high concentrations stems from the studied, in agreement with previous results regarding their
common observation that competitive adsorption behavior adsorption energy distribution (8). The distributions derived
is rather unimportant at low concentrations (1) and that in this work, using Eq. [42], are similar in shape for five
the different theories now available for the calculation of of the six compounds (Fig. 5). There is a shift along the
multicomponent isotherms provide good estimates at low energy axis corresponding to the different values obtained
coverages, while no model is completely satisfactory at high for ea (Table 8). This result is again similar to the one
concentrations (31). Thus, high concentrations is the range reported previously (8). The values obtained for the hetero-
in which it is desirable to find models accounting well for geneity parameter of tetrachloroethylene are different from
single-component data, provided that these models can be those of all the other chlorinated hydrocarbons, which may
used within the framework of the adsorbed solution theory be explained by the fact that it is the only unsaturated com-
models.
The results of the regression analyses presented in Tables
3–6 show a slight variation (usually a decrease but, in a few
cases, an increase) of the monolayer capacity with increasing
temperature for all the compounds studied. However, the
significant error observed in the determination of this param-
eter does not allow a definitive conclusion regarding this
effect. This situation is the same as for the results previously
reported (8, 9). However, the Jovanovic–Freundlich iso-
therm gives the smallest value of all isotherm models for
the monolayer capacity.
The temperature dependence of the heterogeneity parame-
ter is not well defined, given the values of the estimates
obtained at the different temperatures and those of the errors
made. The only exception is the trend exhibited by the het-
erogeneity parameter determined from the fit of the data for
1,1,1-trichloroethane to the Tóth model. At 288 K, the value
identified is close to unity, suggesting that the Langmuir
model may correlate these data as well as more complex FIG. 5. Adsorption energy distributions calculated for the six chlorohy-
models. In fact, this result was obtained previously by Hines drocarbons studied at 288 K.

AID JCIS 4406 / 6g16$$$185 09-25-96 13:47:31 coida AP: Colloid


NEW JOVANOVIC–FREUNDLICH ISOTHERM MODEL 67
pound studied. The same result is derived from the analysis 5. Tóth, J., Acta Chim. Hung. 32, 31 (1962).
of the values derived from the regression of the Tóth and 6. Misra, D. N., J. Colloid Interface Sci. 79, 543 (1980).
7. Tóth, J., J. Colloid Interface Sci. 163, 299 (1994).
UNILAN models to the adsorption data of this compound. 8. Kuo, S.-L., Hines, A. L., and Dural, N. H., Sep. Sci. Technol. 26, 1077
The energy distribution calculated for this compound ac- (1991).
cording to Eq. [42] does not have the same shape as the 9. Kuo, S.-L., and Hines, A. L., J. Chem. Eng. Data 37, 1 (1992).
distributions of the other five compounds (Fig. 5). 10. Honig, J. M., and Reyerson, L. H., J. Phys. Chem. 56, 140 (1952).
The parameter a decreases exponentially with increasing 11. Hines, A. L., Kuo, S.-L., and Dural, N. H., Sep. Sci. Technol. 25, 869
(1990).
temperature, as seen in Table 8 and Fig. 2. Large values 12. Stanley, B., and Guiochon, G., Langmuir 11, 1735 (1995).
of the correlation coefficients were obtained in all cases. 13. Langmuir, I., J. Am. Chem. Soc. 40, 1361 (1918).
The standard error of the estimation ( correlation ) is small. 14. Jovanovic, D. S., Kolloid-Z.Z. Polym. 235, 1203 (1969).
However, the errors made on the estimates of K 0 and ea 15. Moreau, M., Valentin, P., Vidal-Madjar, C., Lin, B. C., and Guiochon,
are rather large because the correlations of these two pa- G., J. Colloid Interface Sci. 141, 127 (1991).
16. Fowler, R., and Guggenheim, E. A., ‘‘Statistical Thermodynamics.’’
rameters are based on the use of three experimental data Cambridge Univ. Press, Cambridge, 1965.
points, resulting in only one degree of freedom. The values 17. Freundlich, ‘‘Kapillarchemie.’’ Akademische Verlagsgesellschaft,
of K 0 are low, as expected ( 3 ) . For each compound, the Leipzig, 1922.
value of ea derived from the regression is close to the 18. Sips, R., J. Chem. Phys. 16, 490 (1948).
minimum value of the isosteric heat of adsorption ( 8, 9 ) , 19. Radke, C. J., and Prausnitz, J. M., Ind. Eng. Chem. (Fundam.) 11, 445
(1972).
as required by theory ( 3 ) . 20. Dubinin, M. M., and Radushkevich, L. V., Dokl. Akad. Nauk SSR Ser.
The L-function is plotted in Fig. 3 versus the partial pres- Khim. 55, 331 (1947).
sure of the adsorbate. It is higher for the Jovanovic–Freund- 21. Graham, D., J. Phys. Chem. 57, 665 (1953).
lich isotherm than for the Tóth isotherm within the range 22. Laub, R. J., ACS Symp. Ser. 297, 1 (1986).
studied. On the other hand, the function l ( u, p) decreases 23. Jacobson, S., Golshan-Shirazi, S., and Guiochon, G., AIChE J. 37, 836
(1991).
faster with increasing pressure for the Tóth model than for 24. Charton, F., Jacobson, S. C., and Guiochon, G., J. Chromatogr. 630,
the Jovanovic–Freundlich isotherm. At low pressures, the 21 (1993).
function tends toward a limit equal to the low pressure equi- 25. Misra, D. N., J. Colloid Interface Sci. 43, 85 (1973).
librium constant for the Tóth isotherm, while it tends toward 26. Shlygin, A., and Frumkin, A., Acta Physicochim. URSS. 3, 791 (1935).
` for the Jovanovic–Freundlich isotherm (see Eq. [29] and 27. Jaroniec, M., and Piotrowska, J., Chem. Zvesti 40, 65 (1986).
28. Jaroniec, M., Colloid Polym. Sci. 254, 601 (1976).
numerical values of a and n in Table 3). The fact that the 29. Dondi, F., Gonnord, M.-F., and Guiochon, G., J. Colloid Interface Sci.
function l ( u, p) for the Jovanovic–Freundlich model tends 62, 316 (1977).
toward infinity when the pressure tends toward 0 may be a 30. Rouchon, P., Schonauer, M., Valentin, P., Vidal-Majdar, C., and Guio-
result of the fact that this equation does not reduce to the chon, G., J. Phys. Chem. 89, 2076 (1985).
Henry law at low pressures. However, the L-function de- 31. Scholl, S., Schachtl, M., Sievers, W., Schweighart, P., and Mersmann,
A., Chem. Eng. Technol. 14, 311 (1991).
creases with increasing pressure in a similar way for both 32. Jaroniec, M., Narkiewicz, J., and Rudzinski, W., J. Colloid Interface
models. Sci. 65, 9 (1978).
33. Sips, R., J. Chem. Phys. 18, 1024 (1950).
ACKNOWLEDGMENTS 34. Marquardt, D. W., J. Soc. Appl. Math. 11, 431 (1963).
35. Ajnazarova, S. L., and Kafarov, V. V., ‘‘Metodi Optimisatsi Eksperi-
This work was supported in part by UNDP project CUB/91/001. I.Q. menta v Khimicheskoy Teknologui.’’ Vishaia Shkola, Moskva, SSSR,
thanks the Laboratoire de Génie Chimique (URA CNRS 192) of INP- 1985.
ENSIGC, Toulouse, France for support, and especially Prof. Dr. A. M. 36. Cochran, T. W., Kabel, R. L., and Danner, R. P., AIChE J. 31, 268
Wilhelm, and acknowledges Eng. René Legrà from CQF, Habana, Cuba (1985).
for his help in the preparation of computer graphs. 37. Marczewski, A. W., and Jaroniec, M., Monatsh. Chem. 114, 711
(1983).
38. Dubinin, M. M., and Astakhov, V. A., Izv. Akad. Nauk SSSR Ser. Khim.
REFERENCES 71, 5 (1971).
39. Fritz, W., and Schlünder, E. U., Chem. Eng. Sci. 29, 1279 (1974).
1. Valenzuela, D. P., and Myers, A. L., ‘‘Adsorption Equilibrium Data 40. Gamba, G., Rota, R., Storti, G., Carra, S., and Morbidelli, M., AIChE
Handbook.’’ Prentice Hall, Englewood Cliffs, 1989. J. 35, 959 (1989).
2. Ruthven, D. M., ‘‘Principles of Adsorption and Adsorption Processes.’’ 41. Guiochon, G., Golshan Shirazi, S., and Katti, A. M., ‘‘Fundamentals of
Wiley–Interscience, New York, 1984. Nonlinear and Preparative Chromatography.’’ Academic Press, Boston,
3. Jaroniec, M., and Madey, R., ‘‘Physical Adsorption on Heterogeneous 1994.
Solids.’’ Elsevier, Amsterdam, 1998. 42. Burlington, R. S. D., and May, D. C., ‘‘Handbook of Probability and
4. Schmidt, G. C., Z. Phys. Chem. 77, 641 (1911). Statistics with Tables.’’ McGraw–Hill, New York, 1970.

AID JCIS 4406 / 6g16$$$185 09-25-96 13:47:31 coida AP: Colloid

You might also like