Chemical Engineering Science 145 (2016) 233–244

Contents lists available at ScienceDirect

Chemical Engineering Science

journal homepage: www.elsevier.com/locate/ces

Gas–liquid mass transfer enhancement by catalyst

particles, a modelling study
P.W.A.M. Wenmakers a, J.A.A. Hoorn a, J.A.M. Kuipers b, N.G. Deen c,n
a
DSM Ahead – Research and Development, Advanced Chemical Engineering Solutions (ACES), P.O. Box 18, 6160 MD Geleen, The Netherlands
b
Multiphase Reactors Group, Department of Chemical Engineering and Chemistry, Eindhoven University of Technology, P.O. Box 513, 5600 MB Eindhoven,
The Netherlands
c
Multiphase and Reactive Flows Group, Department of Mechanical Engineering, Eindhoven University of Technology, P.O. Box 513, 5600 MB Eindhoven, the
Netherlands

H I G H L I G H T S

The presence of catalyst particles on the enhancement of gas absorption is studied.

The enhancement factor is numerically calculated as function of all key parameters.

These parameters are catalyst concentration, particle size and diffusion coefficients.

A homogeneous model is derived that can adequately describe the results.

art ic l e i nf o a b s t r a c t

Article history: In this paper we studied the effect of catalyst particles on the enhancement of the gas–liquid mass
Received 4 August 2015 transfer due to a first order heterogeneously catalyzed reaction. The enhancement factor was obtained by
Received in revised form solving the liquid side unsteady reaction-diffusion equation accounting for the presence of the particles.
12 January 2016
The results are compared with an analytical model that was derived assuming a homogeneous dis-
Accepted 16 January 2016
Available online 9 February 2016
tribution of the particles in the liquid film. It was found that the enhancement factor increases with
increasing catalyst concentration and increasing rate of reaction, as expected. As the particle diameter is
Keywords: increased, the enhancement factor decreases due to an increasing degree of diffusion limitation.
Enhancement factor The results of the homogeneous model match well with the results of the numerical simulations
Numerical simulation
within a 10% error. Therefore it is concluded that the homogeneous model can be used to predict the
Chemically enhanced mass transfer
enhancement of the mass transfer due to the presence of catalytic particles.
Three-phase transport phenomena
& 2016 Elsevier Ltd. All rights reserved.

1. Introduction
In chemical industry the importance of multiphase reactors is
eminent already for decades. Still today there are enough challenges
left to further improve performance of these reactors to maintain
profitability and also to meet more strict (future) environmental
targets. Especially for reactor systems in large bulk chemical plants
every small fraction in reactor performance increase can be a sub-
stantial overall improvement. Not only in increased reactor yield as
such but also separation downstream can be positively affected.
Studies to support the fundamental understanding of the mechan-
isms taking place in the reactors and at the same time extending
n
Corresponding author.
E-mail address: N.G.Deen@tue.nl (N.G. Deen).
engineering skills are still very worthwhile to do. One of the more
complicated multiphase systems is the so-called gas–liquid–solid
reactor where gaseous and liquid components are allowed to react by
means of a solid (supported) catalyst. A proper understanding of the
hydrodynamics, mass transfer, heat transfer, and kinetics is required
for the proper design of such a reactor. For slurry reactors, where the
small catalyst particles are dispersed throughout the liquid phase, the
gas to liquid mass transfer is one of the critical aspects.
It is known that the rate of gas–liquid mass transfer can be sig-
nificantly influenced by the presence of small particles in the liquid,
either being gas-absorbing or reactive particles (Ramachandran, 2007;
Ruthiya et al., 2004; Wimmers et al., 1984, 1988; Wimmers and For-
tuin, 1988b; Holstvoogd et al., 1988; Holstvoogd and Van Swaaij, 1990).
The presence of the small particles near the gas–liquid interface can
increase the liquid side concentration gradient of the transferred

http://dx.doi.org/10.1016/j.ces.2016.01.043
0009-2509/& 2016 Elsevier Ltd. All rights reserved.
234 P.W.A.M. Wenmakers et al. / Chemical Engineering Science 145 (2016) 233–244
component, thereby increasing the mass transfer rate. Depending on
the lyophobicity the particles can actually adhere to the gas–liquid
surface (Wimmers and Fortuin, 1988a; Omota et al. 2006a,b; Verrelli
et al., 2011), thus reducing the diffusion distance between the
gas–liquid interface and the (catalyst) particles. The enhancement of
mass transfer due to the presence of catalyst particles near the gas–
liquid surface has been investigated in a couple of earlier papers, see
e.g. Ruthiya et al. (2004, 2005), Ramachandran (2007). Especially
Ramachandran (2007) presents a thorough review on the modelling
approaches followed. Typically a continuum approach is adopted
where the steady state diffusion-reaction equation is solved. In some
cases also the effect of internal diffusion is taken into account via the
effectiveness factor. However, the presence of the particles themselves
and the reaction occurring within the particles will result in the pre-
sence of spatial concentration gradients in all directions, not only
perpendicular to the gas–liquid interface. The particles themselves can
also be seen as an obstruction within the fluid, thus hindering mass
transfer towards the bulk of the liquid. These aspects are typically not
taken into account within the available models. Also, as stated earlier,
the internal diffusion is approximated by the effectiveness factor
approach which per definition does not account for local concentra-
tion variations around the catalyst particle.
Nowadays, the computing power and the availability of com-
mercial software allows for a much more detailed consideration of
transport phenomena and related problems. The integration
between software packages facilitates the accurate representation
of more complex geometrical structures and to arrange for a
sequence of simulations. Therefore, a large number of simulations
can be conducted and run even on a standard workstation.
Therefore, the effect of particles near the gas–liquid interface can
now be studied numerically in more detail.
In the present study the enhancement of gas–liquid mass
transfer was studied using 3D unsteady simulations using the
COMSOL Multiphysics software environment. The diffusion of the
gaseous component through a suspension of catalyst particles was
studied for various cases, viz. effect of:
1. particle configuration, i.e. randomly distributed particles versus
closely packed particles,
2. particle diameter,
3. particle volume fraction,
4. rate of reaction, i.e. different reaction rate coefficients (this also
directly relates to the degree of internal diffusion limitation), and
5. liquid diffusivity.
In addition to the 3D simulations a homogenous model was
derived that incorporates the combined effects of the catalyst
particles and internal diffusion limitation on the overall rate of
mass transfer through the gas–liquid interface. The results of the
homogeneous model have been compared with the rigorous
simulations in COMSOL Multiphysics.
The outline of our paper is as follows: first the adapted mod-
elling approach for gas–liquid mass transfer through a suspension
of catalyst particles is described. Subsequently the derivation of
the homogeneous model for the enhancement factor for gas–
liquid mass transfer is presented, followed by a comparison of the
results obtained from the homogeneous and detailed models.
Finally it is shown that the homogeneous model is capable of
providing adequate estimates of the enhancement factor using
either graphs or (relatively) simple algebraic expressions.
The presented graphs and expressions can be used to estimate
the degree of enhancement on gas–liquid mass transfer for a ple-
thora of reactor systems, since no assumptions are made on this
part. Using knowledge (or assumptions) on the local catalyst con-
centration and the kinetics thus allows the engineer to account for
gas–liquid mass transfer enhancement, resulting in a more accurate
design. Heterogeneously catalyzed reactions are abundant in
industry for e.g. oxidation, hydrogenation, and Fischer–Tropsch, see
e.g. Youssef et al. (2013).
2. Theory
The enhancement of gas absorption into the liquid phase by
means of heterogeneous catalyst particles has already been stu-
died for many decades, see e.g. Ruthiya et al. (2005), Ramachan-
dran (2007), Wimmers et al. (1984), both experimentally and
theoretically. In these publications a number of modeling
approaches have been followed to describe the enhancement of
gas absorption. Ramachandran (2007) gives an overview of the
various methodologies applied. All of these models, in one way or
another, describe the enhancement of the absorption by con-
centration depletion within the film layer. The different models
vary in level of complexity from pseudo-homogeneous models to
more complex cell models which take into account various dis-
tances of the catalyst particle from the gas–liquid interface. Some
of these models employ the classical resistances in series approach
(see Fig. 1) to describe the observed rate of absorption:


1 1 1  1 pA
RAv ¼ þ þ ð1Þ
kgl agl kls als ηkr HA
while others start from the stationary reaction-diffusion equation. In
some studies even the effect of internal diffusion limitation by means
of the so-called effectiveness factor (η) is accounted for. These studies
typically have derived their relations on (pseudo) steady state
assumptions. In this work two approaches have been followed to
study the enhancement of the gas absorption by catalyst particles
residing inside the liquid-side mass transfer boundary layer:
1. Detailed unsteady simulation of diffusion into a liquid film filled
with reactive catalyst particles using COMSOL Multiphysics.
2. Derivation of an unsteady homogeneous model for diffusion
into a liquid containing catalyst particles.
2.1. Numerical model
COMSOL Multiphysics 4.3a has been used to model the diffu-
sion of the gas into a liquid film filled with catalyst particles. For
practical reasons only a small part of a gas–liquid interface has
been modelled, therefore curvature of the gas–liquid interface is
neglected, i.e. the gas–liquid interface has been modelled as a flat
interface. The catalyst particles are assumed spherical and homo-
geneous, i.e. no pores have been modelled explicitly where the
(effective) diffusion coefficient throughout the solid is assumed to
be constant. The presence of the solid catalyst particles within the
liquid film has been represented in two ways:
1. The particles are randomly distributed.
Fig. 1. Schematic representation of the concentration profile for a heterogeneously
catalyzed gas–liquid reaction, showing the resistances in series.
 P.W.A.M. Wenmakers et al. / Chemical Engineering Science 145 (2016) 233–244 235

2. The particles are closely packed in a hexagonal packing near the continuum (Crank, 1975, p. 271):
gas–liquid interface. This can be considered as an extreme limit

1 þ 2θ
of the particle configuration. Ds ¼ Dl ð4Þ
1θ
In this way two extreme cases have been taken into account, with
either the particles are distributed evenly throughout the liquid or

Dp  Dl
the particles collect and organize themselves in a close packing θ ¼ εp ð5Þ
near the gas–liquid interface. If catalyst particles are lyophilic they Dp þ 2Dl
have the tendency to remain distributed throughout the liquid, In this way both the diffusivity of the solid catalyst particles
whereas if the catalyst particles are lyophobic they have a higher and the diffusivity of the liquid have been accounted for. In
tendency to adhere to the gas–liquid interface (Ruthiya et al., experiments, typically a weight fraction xp is used instead of a
2005, 2004). By assuming that the catalyst particles stack in a volume fraction. The volume fraction εp is obtained using:
hexagonal close packing the most extreme case has been taken
into account with respect to catalyst concentration near the gas– xp
liquid interface. Fig. 2 shows examples of the (a) random and ρp
εp ¼ ð6Þ
(b) hexagonal distribution of the particles in space. xp 1  xp
þ
For both cases the diffusion and reaction is modelled in the ρp ρl
same way using the general reaction-diffusion equation:
Now, using the following set of initial and boundary conditions:
∂C
¼ ∇  ðD∇C Þ  kr C ð2Þ t ¼ 0; z Z 0; C¼0 ð7Þ
∂t
where D is Dl for the liquid and Dp for the solid, whereas the
z ¼ 0; t 4 0; C ¼ C0 ð8Þ
reaction only takes places inside the solid, viz. kr ¼0 for the
liquid phase.
z¼1 t 4 0; C¼0 ð9Þ
2.2. Homogeneous model the solution of Eq. (3) reads (Crank, 1975, p. 334):
pffiffiffi
pffiffiffiffiffi 1 pffiffiffik
pffiffiffiffiffi
For the homogenous model the liquid and solid parts are not C 1  z Dk z z
¼ e s erfc pffiffiffiffiffiffiffi  kt þ ez Ds erfc pffiffiffiffiffiffiffi þ kt
considered separately. It is assumed that the reaction occurs in a C0 2 2 Ds t 2 2 Ds t
homogeneous medium where no distinction is made between the ð10Þ
liquid and solid phase. The homogeneous medium is assumed to 1
In case of no reaction, i.e. k ¼ 0 s , this expression reduces to
have constant effective properties throughout space and to be
the well-known solution for diffusion into a semi-infinite med-
semi-infinite. Only the transport by means of diffusion into the z-
ium:
direction perpendicular to the gas–liquid interface is considered.


The reaction-diffusion equation now reads: C z
¼ erfc pffiffiffiffiffiffiffi ð11Þ
∂C ∂2 C C0 2 Ds t
¼ Ds 2  kC ð3Þ
∂t ∂z The degree of enhancement of the gas absorption rate in case of
where Ds is the effective diffusion coefficient of the transferred a chemical reaction is typically quantified by means of the so-
component in the liquid–solid suspension. The value for Ds can be called enhancement factor. The enhancement factor is described as
obtained using the relation for dilute dispersed particles in a the ratio of gas flux into the liquid with and without a reaction

Fig. 2. Schematic representation of the different particle configurations used in the simulations; (a) random distribution of the particles in the fluid and (b) hexagonal
packing near the gas–liquid interface.
236 P.W.A.M. Wenmakers et al. / Chemical Engineering Science 145 (2016) 233–244

occurring in the liquid (Westerterp et al., 1984): reaction taking place):

 

J reaction z ¼ 0 Flux without reaction; with particles
E¼  ð12Þ R¼ ð23Þ
J no reaction z ¼ 0 Flux without reaction; without particles
where the flux at the gas–liquid interface is defined as: The retention factor describes how much the rate of mass
 transfer through the gas–liquid interface is reduced by the pre-
 ∂C 
J z ¼ 0 ¼  Ds  ð13Þ sence of the particles. This is due to the lower diffusivity of the
∂z z ¼ 0
solid particles. The expression for R is readily obtained using
The flux for the case of reaction and no reaction can now Eq. (18) using Ds and Dl in the presence and absence of particles,
readily be obtained by combining Eqs. (10) and (11) with Eq. (13): respectively, in combination with Eq. (23):
0sffiffiffiffiffi 1 sffiffiffiffiffi
 k pffiffiffiffiffi expð  ktÞ Ds
J reaction z ¼ 0 ¼ C 0 Ds @ erf kt þ pffiffiffiffiffiffiffiffiffiffi A ð14Þ R¼ ð24Þ
Ds π Ds t Dl

rffiffiffiffiffi The enhancement factors E and E0 can be related using R as

 Ds
J no  ¼ C ð15Þ follows:
reaction z¼0 0
πt
E0 ¼ E  R ð25Þ
For the enhancement factor the time-averaged flux is used
rather than the instantaneous flux, which is defined as: Two additional aspects of the catalyst particles still have to be
Z taken into account. First, the reaction-diffusion equation of the
 1 τ
J ¼ J A j z ¼ 0 dt ð16Þ homogeneous model, as shown in Eq. (3), is based on the total volume
τ 0
whereas the reaction only takes part in the particles. This can easily be
Integration of Eqs. (14) and (15) yields the following expres- corrected for by multiplication with the volume fraction of the parti-
sions for the time-averaged flux: cles. Secondly, the particles can exhibit (partial) diffusion limitation.
  pffiffiffiffiffiffiffiffi
1
 pffiffiffiffiffi expð  kτÞ Therefore, the k also needs to be corrected for the effect of internal
J reaction z ¼ 0 ¼ C 0 kDs 1 þ erfð kτÞ þ pffiffiffiffiffiffiffiffiffi ð17Þ diffusion limitation. This is done using the regular reaction engineer-
2kτ π kτ
ing approach by multiplying with the effectiveness factor, η. Thus, k
rffiffiffiffiffiffi
  Ds can now be defined as the effective rate coefficient:
J no  ¼ 2C 0 ð18Þ
reaction z¼0 πτ k ¼ ηkr εp ð26Þ
Combining Eqs. (17) and (18) with the definition of the For spherical particles and first order kinetics the effectiveness
enhancement factor (Eq. (12)) yields the following expression for factor is obtained using (Levenspiel, 1999):
E:


 pffiffiffiffiffi expð  kτÞ
1 1 1
1pffiffiffiffiffiffiffiffiffi 1 η¼  ð27Þ
E¼ π kτ 1 þ erf ð kτÞ þ pffiffiffiffiffiffiffiffiffi ð19Þ ϕ tanhð3ϕÞ 3ϕ
2 2kτ π kτ
with ϕ, the Thiele modulus, defined as:
This is the enhancement factor for the case of catalytic (reac- sffiffiffiffiffiffiffiffi
tive) particles with respect to a suspension of similar inert (non- Rp kr
reactive) particles. In addition to that, one can also compare the ϕ¼ ð28Þ
3 Deff
flux through the gas–liquid interface under reactive conditions
with particles (Eq. (17)) to the flux through the gas–liquid inter- Since in the simulations no explicit correction is made for the
face in the absence of particles, i.e. Eq. (18) where Ds is replaced by effective diffusion coefficient, Deff, in terms of tortuosity or por-
Dl. This now yields: osity it is assumed that the effective diffusivity for the analytical
sffiffiffiffiffi
 model is the same as the particle diffusivity as applied in the
1pffiffiffiffiffiffiffiffiffi Ds 1 pffiffiffiffiffi expð  kτÞ numerical simulations.
0
E ¼ π kτ 1þ erf ð kτÞ þ pffiffiffiffiffiffiffiffiffi ð20Þ
2 Dl 2kτ π kτ
Eq. (19) only depends on the parameter kτ and Eq. (20) also
takes into account the difference in diffusivity through the para- 3. Methodology
meter DDsl . Depending on whether mass transfer data is available for
the case with particles or without particles either Eq. (19) or Eq. 3.1. Geometry
(20) can be used to calculate the expected enhancement of mass
transfer due to reaction. In addition, Eq. (20) allows one to study The geometry of the computational domain consists of two parts,
the effect of catalyst concentration on the observed mass transfer the continuous liquid phase and the dispersed spherical catalyst par-
enhancement since the effect of catalyst concentration is incor- ticles. The overall geometry is based on a rectangular box throughout
porated in both Ds (Eq. (4)) and k (see Eq. (26)). We note that a which spherical particles have been dispersed (see Fig. 2). For the
more rigorous derivation of Eq. (19) can be found in Appendix A. particles two cases in terms of spatial configuration have been
The two definitions of the enhancement factor, E and E0 , are considered:
formally defined as:

 1. The particles are randomly dispersed throughout the liquid
Flux with reaction; with particles medium (see Fig. 2(a)).
E¼ ð21Þ
Flux without reaction; with particles 2. The particles are stacked in a hexagonal close packing near the

 gas–liquid interface (see Fig. 2(b)).
Flux with reaction; with particles
E0 ¼ ð22Þ
Flux without reaction; without particles
For the simulations the geometry was constructed using a
These definitions of the enhancement factor are related combination of Matlab R2014a and COMSOL Multiphysics 4.3a.
through a retention factor, R, which relates the flux through the The geometry and the simulation settings were constructed using
gas–liquid interface with and without particles (both without Matlab and were then transferred to the COMSOL environment. In
 P.W.A.M. Wenmakers et al. / Chemical Engineering Science 145 (2016) 233–244 237

this way, both the random and hexagonal configuration of parti-

cles could be created using the flexibility of Matlab.
For the randomly distributed particles the rectangular domain
was set to a fixed size of 0.3  0.3  0.5 mm3, except for the case
with the lowest concentrations and the largest particle diameter
(xw ¼1 wt.% and dp ¼ 100 μm) where the domain size was
increased to 0.8  0.8  0.5 mm 3 to ensure for a sufficient number
of particles within the domain. This corresponds to 6dp  6dp 
10dp for the 50 μm particles and to 3dp  3dp  5dp or 8dp  8dp 
5dp for the 100 μm particles. For the hexagonal distributed particles
the size of the domain was calculated on basis of the size and Fig. 3. Schematic representation of the gas–liquid interface for a gas bubble with a
number of particles and ranged from 0.25  0.19  0.27 mm3 to R¼ 1.5 mm radius and dp ¼ 50 μm catalyst particles. The dashed box with a height
0.50  0.37  0.54 mm3. Fig. 3 shows a 2D representation of the of h ¼ 300 μm shows the calculation domain as applied for the numerical simula-
tions. The error in the surface area due to neglecting the curvature is 1.4%.
catalyst particles near the gas–liquid interface for a 3 mm gas
bubble and 50 μm catalyst particles. The figure shows that the
pffiffiffi
 
1
approximation of a flat gas–liquid interface is valid, due to the large yi;j;k ¼ 3 jþ modðk; 2Þ þ 1 Rp þ ðj þ 1Þδ ð34Þ
3
difference in diameter of the gas bubble and the catalyst particles.
pffiffiffi !
2 6
3.1.1. Randomly configured particles zi;j;k ¼ k þ1 Rp þkδ þ δgp ð35Þ
3
The position of the randomly configured particles was con-
structed using a uniform random number generator. Three ran- where the particle indices i, j, and k range from 0 to np;x  1,
dom numbers are needed to define the position in space. The np;y  1, and np;z  1, respectively. Here an additional correction has
random numbers are generated on the interval [0,1]. The numbers been included for the x-position to ensure for a non staggered
were subsequently scaled in such a way that a particle cannot stacking for those cases where modðj; 2Þ and modðk; 2Þ equals 1.
cross the boundary of the computational domain: The size of the rectangular block domain is calculated on basis of



xr ¼ Rp þ δ þ X W  2 Rp þ δ ð29Þ the number of particles in the x, y, and z directions:


 



 W ¼ 2 np;x  1 þ 3 Rp þ np;x þ 2 δ ð36Þ
yr ¼ Rp þ δ þ X H 2 Rp þ δ ð30Þ

pffiffiffi
 



 1

zr ¼ Rp þ δ þ X L  2 Rp þ δ ð31Þ H¼ 3 np;y  1 þ þ1 Rp þ np;y þ 1 δ þ Rp ð37Þ
3
where X are uniformly distributed random numbers. At first a pffiffiffi !
6

particle is randomly positioned in space. For any subsequently L¼ 2 np;z  1 þ 1 Rp þnp;z δ þ δgp þ Rp ð38Þ
generated particle, with updated random position, it is checked 3
whether or not the particles overlap. This is done by ensuring that
the distance between the newly generated particle and the nearest
3.2. Boundary and initial conditions
particle is greater than 1/25th of the particle diameter. If the dis-
tance is less a new position is generated. The distance of 1/25th of
The exterior walls of the rectangular simulation domain are
a particle diameter is chosen in such a way that it prevents the
positioned perpendicular to the gas–liquid interface and are sub-
occurrence of too small mesh elements between particles. The
jected to a zero flux boundary condition:
procedure is repeated until the required number of particles is
!
generated. The required number of particles is calculated on basis x ¼ wall; 8 t; y; z D∇C  n ¼ 0 ð39Þ
of the specified volume fraction of particles:
!
V y ¼ wall; 8 t; x; z D∇C  n ¼ 0 ð40Þ
np ¼ εp d ð32Þ
Vp The concentration at the gas–liquid interface has been set to a
constant value of C0:
where Vp is the particle volume and Vd is the volume of the
simulation domain. Fig. 2(a) shows an example of the resulting C
z ¼ 0; 8 t; x; y ¼1 ð41Þ
geometry for the case of a particle diameter of 50 μm and xp ¼10 C0
wt%. The figure shows the random distribution through space The boundary condition of the wall opposing the gas–liquid
without any overlap of particles with each other nor with the interface, has been set to a zero flux condition:
confining walls.
!
z ¼ L; 8 t; x; y D∇C  n ¼ 0 ð42Þ
3.1.2. Hexagonal configured particles The initial condition for the complete domain has been set to 0:
The position of the hexagonal configured particles is calculated
on basis of a modified coordinate system for a hexagonal close C
t ¼ 0; 8 x; y; z ¼0 ð43Þ
packing. An additional space between the particles has been C0
introduced to ensure sufficient space between the particles to For the surfaces between the liquid and the solid continuity of
allow for proper meshing. Similar as for the case of the randomly the species holds:
distributed particles the space between the particles is set to ! !
 Dl ∇C l  n ¼ Dp ∇C p  n ð44Þ
1/25th of the particle diameter. Accounting for the extra space in
the x, y, and z directions this yields the following set of coordinates
for hexagonally stacked spherical particles:
3.2.1. Mesh and solver
xi;j;k ¼ ð2iþ modðj; 2Þ þ modðk; 2Þ þ 1ÞRp


  The geometry was meshed using an unstructured tetrahedral
 modðj; 2Þmodðk; 2Þ dp þ δ þ δ þ ði þ 1Þδ ð33Þ mesh. The total number of mesh elements ranged from 0.32  106 to
238 P.W.A.M. Wenmakers et al. / Chemical Engineering Science 145 (2016) 233–244

Fig. 4. 2D representation of mesh elements as used in the simulations for the case of random particles of 50 μm and 10 wt%. The figures show the random tetrahedral mesh
and the more structured boundary mesh near the gas–liquid interface and near the particle surface, the blue region represents the liquid phase and the red and purple
regions represent the particle phase. (a) Detail of the gas–liquid interface and showing a number of particles, this region represents approximately 50% of the domain; (b) a
close up of the marked region in (a) showing the boundary layer mesh near the gas–liquid interface and at the particle surface. (For interpretation of the references to color
in this figure caption, the reader is referred to the web version of this paper.)

3.7  106 elements for the randomly distributed particles and was
approximately 2.7  106 for the hexagonally distributed particles. To
account for the potentially steep gradients near the gas–liquid
interface and near the solid surface the mesh was refined at these
locations. At the gas–liquid interface a boundary layer mesh was set
of 8 elements in depth. For the particles a boundary layer mesh of
8 elements in depth was constructed on both the inside and outside
of the particle. Fig. 4 shows a 2D example of the mesh used for the
computations. Fig. 4(a) shows the random structure of the tetra-
hedral mesh and the dense boundary layer mesh at the gas–liquid
interface and near the surface of the particles. Fig. 4(b) shows a
detail of the boundary layer mesh near the gas–liquid interface and
also shows the mesh refinement near the particle surface.
The mesh quality was tested using two cases. The quality of the
boundary mesh near the gas–liquid interface was evaluated using
diffusion into a semi-infinite medium as a base case. Using the said
mesh conditions the results of COMSOL coincided with the analytical
solution. For the boundary layer near the catalyst particles a first
Fig. 5. Schematic representation of the surface domain used for averaging the flux.
order reaction was simulated with a high diffusivity in the liquid
phase. Also here the results between COSMOL and the textbook 0:2 o t r 1Δt ¼ 0:1 s ð46Þ
solutions matched well. Since the gradients will be steepest in the
catalyst particle due to the presence of a reaction no issues are
1 o t r 10Δt ¼ 1 s ð47Þ
expected with respect to the liquid side mesh of the catalyst particles.
Eq. (2) has been solved on the generated mesh using the GMRES The time stepping was set to strictly ensure that the results were
iterative solver incorporated in COSMOL Multiphysics. A relative evaluated at the specified time steps, whereas the solver is still free
tolerance of 0.01 and an absolute, scaled, tolerance of 0.001 was to use adaptive time steps between the specified time steps.
applied. COSMOL applies an adaptive time stepping method for
solving the model. However, the solver was forced to generate 3.3. Data analysis
output at the selected time steps shown below. In this way the
solver was forced to evaluate the solution at the desired time step, After COMSOL obtained the solution of the boundary value
preventing the need of interpolating the data. The solver applied an problem (Eq. (2)), the temporal concentration profiles can readily
adaptive time step method for the intermediate time steps. be derived. On basis of this data also the flux though the gas–liquid
interface can be determined. To account for potential spatial var-
0 o t r 0:2Δt ¼ 0:01 s ð45Þ iations of the flux at the gas–liquid interface the spatially averaged
 P.W.A.M. Wenmakers et al. / Chemical Engineering Science 145 (2016) 233–244 239

Fig. 6. Example of a numerically computed concentration profile for (a) random and (b) hexagonal particle configuration. The blue circles/line corresponds to dp ¼ 50 μm and
the red squares/dashed line to dp ¼ 100 μm; k ¼1 1/s, Dl ¼ 10  8 m2/s, xp ¼10 wt%, t ¼10 s. (For interpretation of the references to color in this figure caption, the reader is
referred to the web version of this paper.)

Table 1
Parameters used in simulations.

Parameter Values studied

Particle configuration [–] random, hexagonal

dp ½μm 50, 100
k [1/s] 0.1, 1, 5, 10
Dl [m2/s] 6  10  9, 1  10  8
Dp [m2/s] 1  10  10
xp [wt %] 1, 5, 10

flux is determined:
ZZ
  1 ∂C
J ðtÞz ¼ 0 ¼  D dA ð48Þ
A A ∂z
To exclude any wall effects from the xz and yz oriented walls only a
part of the gas–liquid interface was used (see Fig. 5). A region was
selected which is one particle distance, dp, away from the sides. During
the time-averaging of the flux, which was done according to Eq. (16), it
was found that the flux for the first two time steps (viz. t¼ 0 and
0.01 s) was incorrect. This is a result of the large concentration gra-
dient in the very first steps of the simulation, where the boundary
value of the concentration at the gas–liquid interface is set to 1, while Fig. 7. Enhancement factor as a function of time for random (xp ¼ 10%) and hex-
the initial condition in the rest of the domain is 0. This issue has been agonal packing of particles for k¼ 10 1/s, D¼ 10  8 m2/s. The symbols correspond to
resolved by replacing the values of the (progressive) average flux for the data obtained from the numerical simulations and the lines to the analytical
the first two time steps by those for the theoretical value of diffusion model. (For interpretation of the references to color in this figure caption, the
reader is referred to the web version of this paper.)
into a semi-infinite medium as described by Eq. (18).

3.4. Model parameters
Eq. (2) has been solved for the two particle configurations and
various values of the particle diameter, reaction rate coefficient,
liquid diffusivity, and catalyst concentration (see Table 1). The density
of the liquid and the bulk density of the particles were set to fixed
values of 1000 and 400 kg/m3, respectively, as reported by Merck for
10% Pd/C. The value of the effective particle diffusivity was set to a
fixed value of 10  10 m2/s, in this way the effect of internal diffusion
limitation could be studied by varying the reaction rate coefficient.
The values for the reaction rate coefficient were varied between
0.1 and 10 1/s to account for a range of degrees of internal diffusion
limitation (viz. η  0:96 to η  0:18). The catalyst concentration was
varied between 1 and 10 wt%. In this way both dilute and more
concentrated systems are taken into account (see e.g. Kantarci et al.,
2005). Also, in case of dilute systems particles can collect in the wake
of a rising gas bubble thus resulting in a locally higher catalyst
concentration. For the diffusion coefficient a value of 6  10  9 and
1  10  8 m2/s was used, which spans reasonable values for the dis-
solved gas diffusion coefficients (see e.g. Ruthiya, 2005) and allows
for reasonable calculation times. For the particle diameter values of
50 and 100 μm have been used, which are on the larger side for a
slurry catalyst, and thus more prone to diffusion limitation. The
applied set of parameters is summarized in Table 1.
Simulations have been conducted using all combinations of the
parameters presented in Table 1. For the case of the hexagonal
particle assembly the effect of catalyst concentration is omitted
since the hexagonal packing implies a fixed concentration corre-
sponding to a packing density of around 0.74.
240 P.W.A.M. Wenmakers et al. / Chemical Engineering Science 145 (2016) 233–244
4. Results and discussion
The time dependent concentration profiles for all cases have
been obtained by solving Eq. (2) for the fluid and catalyst particles.
Fig. 6 shows the concentration profile in the liquid and the catalyst
particles at t ¼10 s for (a) randomly dispersed and (b) hexagonal
configured catalyst particles of 50 μm diameter with k ¼1 s  1,
Dl ¼ 10  8 m2 =s, xp ¼10 wt%. The plot shows the concentration
profile along the depth direction (viewed from the gas–liquid
interface) of the liquid domain. The plot also shows that the
concentration in the liquid phase is quite homogeneous in the x
and y directions, showing that the diffusion is rapid enough to
overcome any disturbance posed by the presence of the catalyst
particles. In addition to the concentration gradient in the liquid
phase the plot also clearly reveals the presence of strong con-
centration gradients within the particles.
4.1. Enhancement factor
The transient concentration profiles obtained from the simu-

lations were used to calculate the flux J ðtÞ through the gas–liquid
 rated in k (note that k ¼ ηkr εp ). The graph indeed shows that all
interface using Eq. (48). The obtained values for J ðtÞ, the
instantaneous spatially averaged flux, were subsequently time-
averaged according to Eq. (16) for the calculation of E. The
obtained average flux was then used to calculate E from the
numerical simulations using Eq. (12).
Fig. 7 shows E as a function of time obtained from the numerical
simulations (symbols) for a representative case for both the random
and hexagonal particle assembly. The blue circles and the green tri-
angles correspond to the particle size of 50 μm, the red squares and
the purple triangle correspond to the particle size of 100 μm. The
graph shows that E increases monotonically over time for all cases.
The graph nicely shows the effect of the packing density. E is the
highest for the hexagonal packing. This is of course due to the high
particle concentration near the gas–liquid interface. Also the effect of
the particle size is revealed in the graph. The data shown in Fig. 7 is
for the cases where significant internal diffusion limitation is present.
Therefore, the observed rate of absorption decreases with increasing
particle size. This effect can be clearly seen in Fig. 7 where the 50 μm
particle show higher values for E than the 100 μm particles for both
the random and the hexagonal configured particles.
In addition to the data for E as obtained from the numerical
simulations Fig. 7 also shows the results of the analytical model
(lines). The graph shows that the data obtained from the approx-
imate analytical model agree well with the data obtained from the
numerical simulations. The model shows a slight overprediction of
E for the data represented in the graph. For the 50 μm particles
with a random configuration, the model shows a very close match
with E obtained from the numerical simulations.
Fig. 8 shows E as a function of time for the random configured
particles of (a) 50 μm and (b) 100 μm. The graphs show the results
for all studied cases of xp, kr and Dl. The legend of the figures is
shown in Table 2. The graphs show that E increases as the overall
reaction rate increases, either by increasing the reaction rate
coefficient ðblue 4 magenta 4 green 4 redÞ or by increasing the
catalyst concentration ð△ 4 □ 4 ◯Þ. Comparing graphs 8a&b also
reveals the clear effect of the particle diameter (viz. internal dif-
fusion limitation). That is, E is larger for smaller particle diameters.
For lower overall reaction rate coefficients (viz. low kr and xp) E is
close to unity. For larger reaction rates E extends to 2.6 and 1.8 for
50 μm and 100 μm particles, respectively.
Fig. 9 shows E as a function of time for all studied cases of the
hexagonal configured particles. The legend of the figure is given in
Table 2. Also in this graph, the effect of the reaction rate coefficient
is clearly seen ðblue 4 magenta 4green 4 redÞ, where a low reac-
tion rate coefficient results in a E value of approximately unity.
Increasing kr to 10 s  1 results in values of E up to E4.6 for 50 μm
particles. As expected, also for the hexagonal particle assembly the
effect of particle size is clear. The 50 μm particles (circles) show
significantly higher values for E than the 100 μm particles.
Both Figs. 8 and 9 show the results for numerical (symbols) and the
analytical model (lines). The graphs show that the analytical model
accurately describes the data obtained from the numerical simulations.
The analytical model is capable of describing the trends as well as the
quantitative values of E for the studied parameter set. This can also be
seen in Fig. 10 which shows a parity plot for the analytical model ðEanal: Þ
and the numerical simulations (Ecomsol). The parity plot shows that
almost all the values for the analytical model are within 10% deviation
of the numerical simulations. The parity plot also reveals that in all
cases the analytical model tends to slightly overpredict the value of E.
Eq. (19) shows that, according to the homogeneous model, the
enhancement factor is only dependent on the parameter combi-
nation kτ. Therefore, the curves shown in Figs. 7–9 should reduce
to a singular curve when E is plotted as a function of kτ. Fig. 11
shows E as a function of kτ, where all the effects of internal dif-
fusion limitation and particle concentration have been incorpo-
the curves collapse to a single line, which is then described very
well by the analytical model. This allows, in essence, for an easy
estimation of the enhancement factor for any given hetero-
geneously catalyzed (first order) reaction.
Fig. 12 shows the enhancement factor E0 according to Eq. (20) as a
function of ηkr εp τ. Here the flux in case of reactive particles is
compared to the flux in the absence of particles. The graph in Fig. 12
(a) shows E0 for various values of DDsl allowing for an easy estimation of
the enhancement factor for a given value of DDsl and ηkr εp τ. Fig. 12 also
shows the data as obtained from the numerical simulations. It can
clearly be seen that there is a good agreement between the numer-
ical data and the homogeneous model as presented in Eq. (20).
5. Conclusions
We have derived a method in which the homogeneous model
can be used to predict the enhancement of the mass transfer due
to the presence of catalytic particles in a relative easy manner
using Eqs. (19) and (20). More elaborate simulations take a lot of
effort but do not improve the predictions substantially.
We studied the effect of catalyst particles on the enhancement
of gas absorption due to a first order heterogeneously catalyzed
reaction. This has been done by numerically solving an unsteady
reaction-diffusion equation where individual catalyst particles
have been modeled, as well as through a simplified analytic model
based on the assumption of a homogeneous medium. The effect of
the following aspects was studied:

Catalyst concentration ðxp ¼ 1–10 wt%Þ.

Catalyst particle size ðdp ¼ 50 μm and 100 μmÞ:

Geometrical configuration of the catalyst particle (randomly
distributed versus hexagonal close packing).

Liquid phase diffusion coefficient (Dl ¼ 6  10  9 m2 =s and 10  8 m2 =s).

Different degrees of internal diffusion limitation (η ¼0.18–0.96).
Both the numerical model and the homogeneous model show
that the enhancement factor:

Increases with increasing catalyst concentration.

Increases with increasing rate of reaction. However, E approa-
ches unity for slow reactions.

Decreases with increasing particle diameter (viz. increasing
degree of diffusion limitation).

Increases monotonically over time.
 P.W.A.M. Wenmakers et al. / Chemical Engineering Science 145 (2016) 233–244 241

Fig. 8. Enhancement factor as a function of time for particles with a random configuration for all studied cases; (left) 50 μm and (right) 100 μm. The legend of the graphs is
shown in Table 2, the symbols correspond to the numerical simulations and the lines to the analytical model. (For interpretation of the references to color in this figure
caption, the reader is referred to the web version of this paper.)

Table 2
Legend for Figs. 8 and 9.

kr [s  1]

dp ½μm xcat [wt. %] 0.1 1 5 10

Random particle configuration (Fig. 8)

50/100 1
50/100 5
50/100 10

Hexagonal particle configuration (Fig. 9)

50 E74
100 E74

Fig. 10. Parity plot of the enhancement factor; the solid line represents the parity
line, the dashed lines the þ/  5% lines, and the dotted lines the þ /  10% lines. The
symbols correspond to the legend entries given in Table 2. (For interpretation of the
references to color in this figure caption, the reader is referred to the web version of
this paper.)

The enhancement factor was found to be the highest for the

Fig. 9. Enhancement factor as a function of time for particles with a hexagonal
configuration. The legend of the graph is given in Table 2, the symbols correspond
to the numerical simulations and the lines to the analytical model. (For inter-
pretation of the references to color in this figure caption, the reader is referred to
the web version of this paper.)
catalyst particles configured in a hexagonal closed packing. This is
due to the high (maximum) catalyst concentration ðεp  0:74Þ.
The results show that the homogeneous model is adequately
capable of describing the results of the detailed numerical simu-
lations. The deviation is mostly within 10% error. Comparison
between the homogeneous model and the numerical model data
shows that the homogeneous model tends to overpredict E.
Depending on the definition of the enhancement factor (E or E0 )
either one or two parameters are needed to describe the
enhancement factor, viz. ηkr εp τ and DDsl . This is in agreement with
the numerical model data which either collapses to a single line
for E, or corresponds to the simulated lines for the various values
242 P.W.A.M. Wenmakers et al. / Chemical Engineering Science 145 (2016) 233–244

6 C0 concentration at the gas–liquid interface (mol m  3)

Cp particle phase concentration (mol m  3)
Cl liquid phase concentration (mol  3)
D diffusion coefficient (m2 s  1)
5 Dl liquid phase diffusion coefficient (m2 m  1)
Dp particle phase diffusion coefficient (m2 s  1)
dp particle diameter (m)
Ds diffusion coefficient of suspension (m2 s  1)
4 Deff effective diffusion coefficient (m2 s  1)
E enhancement factor with particles present as
E [−]

reference (–)
E0 enhancement factor with no particles present as
3
reference (–)
H height of computational domain (m)
HA Henry coefficient of transferred component
(Pa m3 mol  1)
2 
J average flux (mol m  2 s  1)
 
J comsol flux through the gas–liquid interface as obtained
by COSMOL (mol m  2 s  1)
 
1 J no reaction average flux in case of no chemical reaction
0 5 10 15 20 25 (mol m  2 s  1)
 
η kr εp τ [−] J reaction average flux in the presence of a chemical reaction
(mol m  2 s  1)
Fig. 11. Enhancement factor as a function of kτ ¼ ηkr εp τ for all studied cases. The symbols J flux (mol m  2 s  1)
correspond to the legend entries given in Table 2, the solid line corresponds to Eq. (19). (For k effective reaction rate coefficient (s  1)
interpretation of the references to color in this figure caption, the reader is referred to the
web version of this paper.)
kr intrinsic reaction rate coefficient (s  1)
kgl agl volumetric gas–liquid mass transfer coefficient (s  1)
kls als volumetric liquid–solid mass transfer coefficient (s  1)
L depth of computational domain (m)
! normal vector (–)
n
np number of particles in computational domain (–)
np;x number of particles in the x-direction of the com-
putational domain (–)
np;y number of particles in the y-direction of the com-
putational domain (–)
np;z number of particles in the z-direction of the com-
putational domain (–)
pA partial pressure (Pa)
R retention factor (–)
Rp particle radius (m)
RAv volumetric reaction rate (mol m  3 s  1)
t time (s)
Vd volume of the computational domain (m3)
Vp volume of a particle (m3)
W width of computational domain (m)
X uniformly distributed random variable (m)
xp weight fraction of particles (–)
xr x-coordinate of randomly distributed particle (m)
xi;j;k x-coordinate of hexagonal distributed particle (m)
Fig. 12. Enhancement factor ðE0 Þ according to Eq. (20) as a function of ηkr εp τ. The
yr y-coordinate of randomly distributed particle (m)
figure shows the solutions for E0 for various values of DDsl along with the data
obtained from the numerical simulations. (For interpretation of the references to
yi;j;k y-coordinate of hexagonal distributed particle (m)
color in this figure caption, the reader is referred to the web version of this paper.) z spatial coordinate (m)
zr z-coordinate of randomly distributed particle (m)
zi;j;k z-coordinate of hexagonal distributed particle (m)
of DDsl for E0 . Therefore, we conclude that the homogeneous model
can be used to predict the enhancement of the mass transfer due Greek symbols
to the presence of catalytic particles in a relative easy manner
using Eqs. (19) and (20) without the need to resort to elaborate δ minimum distance between particles (m)
simulations. δgp minimum distance between gas–liquid interface
and the shell of the nearest particle (m)
Nomenclature εp particle volume fraction (–)
ϕ Thiele modulus (–)
Roman symbols η effectiveness factor (–)
A area (m ) 2 ρl liquid phase density (kg m  3)
C concentration (mol m  3) ρp bulk density of a particle (kg m  3)
 P.W.A.M. Wenmakers et al. / Chemical Engineering Science 145 (2016) 233–244 243

τ contact time (s) The time-averaged flux at the gas–liquid interface in case of a
θ dimensionless parameter for calculation of sus- reaction now becomes:
pension diffusion coefficient (–) 0sffiffiffiffiffi 1
Z pffiffiffiffiffi expð  ktÞ
 1 τ k
J ¼ C D @ erf ð kt Þ þ pffiffiffiffiffiffiffiffiffiffi A dt ðA:11Þ
τ 0 0 s Ds π Ds t

Acknowledgments 0sffiffiffiffiffi 1
Z τ pffiffiffiffiffi Z τ
  C 0 Ds k expð  ktÞ A
J ¼ @ erf ð kt Þ dt þ p ffiffiffiffiffiffiffiffiffiffi dt ðA:12Þ
Patrick Wenmakers and Niels Deen thank Hans Kuipers and τ Ds 0 0 π Ds t
Johan Hoorn for making it possible to do this work in the context
2sffiffiffiffiffi pffiffiffiffiffi 3
of a research exchange programme between DSM and TU

 pffiffiffiffiffi τ expð  kτÞ erf ð kτ Þ
  C 0 Ds k 1
Eindhoven. J ¼ 4 τ erfð kτÞ þ pffiffiffiffiffiffiffiffiffi þ pffiffiffiffiffiffiffiffi 5
τ Ds 2k π kτ kDs
ðA:13Þ
Appendix A. Analytical solution 2 qffiffiffiffi pffiffiffiffiffi 3
 pffiffiffiffiffiffiffiffi

1
 pffiffiffiffiffi expð  kτÞ Ds
erfð kτÞ
J ¼ C 0 4 kDs 5
k
1 erf ð kτÞ þ pffiffiffiffiffiffiffiffiffi þ
A.1. Enhancement factor 2kτ π kτ τ

In this appendix the derivation of the enhancement factor for ðA:14Þ

the homogeneous model (Eq. (19)) is presented in more detail. The combining the error function terms yields:
flux through the gas–liquid interface in case of no reaction is
 pffiffiffiffiffiffiffiffi
1 1
 pffiffiffiffiffi expð kτÞ
obtained using Eqs. (13) and (11): J ¼ C 0 kDs 1  þ erf ð kτÞ þ pffiffiffiffiffiffiffiffiffi ðA:15Þ
2kτ kτ π kτ
 rffiffiffiffiffi
2  rffiffiffiffiffi
 ∂C  

 pffiffiffiffiffi expð  kτÞ
 Ds z Ds
J A z ¼ 0 ¼  Ds  ¼ C0 exp  ¼ C0 ðA:1Þ  pffiffiffiffiffiffiffiffi 1
∂z z ¼ 0 πt 4Ds t  πt J ¼ C 0 kDs 1þ erfð kτÞ þ pffiffiffiffiffiffiffiffiffi ðA:16Þ
z¼0 2kτ π kτ
And the time-averaged flux is now obtained using Eq. (16): The enhancement can now be written as the ratio of Eqs. (A.15)
Z rffiffiffiffiffi rffiffiffiffiffiffi and (A.2):
 1 τ Ds Ds
J ¼
τ 0
C0
πt
dt ¼ 2C 0
πτ
ðA:2Þ pffiffiffiffiffiffiffiffi
1

pffiffiffiffiffi expð kτÞ

C 0 kDs 1 þ erf ð kτÞ þ pffiffiffiffiffiffiffiffiffi
2kτ π kτ
In case of a reaction taking place, the flux at any z position in E¼ rffiffiffiffiffiffiffi ðA:17Þ
Ds
the domain is: 2C 0
πτ
∂C 
 pffiffiffiffiffi expð  kτÞ
1
J A ¼  Ds ¼  C 0 Ds expð  azÞð  a erfcðbz  cÞ 1pffiffiffiffiffiffiffiffiffi 1
∂z 2 E¼ π kτ 1 þ erf ð kτÞ þ pffiffiffiffiffiffiffiffiffi ðA:18Þ
2 2kτ π kτ
þ a expð2azÞ erfcðbz þ cÞ ðA:3Þ

2b   A.2. Limit of small times

 pffiffiffiffi expð2az ðbz þ cÞ2 Þ þexpð  ðc  bzÞ2 Þ ðA:4Þ
π
The limiting case for the analytical solution for the enhance-
where ment factor can be obtained by slightly rewriting Eq. (19):
sffiffiffiffiffi pffiffiffiffi
pffiffiffiffiffi 
k π 1 pffiffiffiffiffi expð  kτÞ
a¼ ðA:5Þ E¼ kτ þ pffiffiffiffiffi erf ð kτÞ þ pffiffiffiffi ðA:19Þ
Ds 2 2 kτ π
pffiffiffiffiffi
Now by applying the Taylor series expansion for erf ð kτÞ and
1
b ¼ pffiffiffiffiffiffiffi ðA:6Þ expð  kτÞ around kτ ¼ 0 and only using the first term the following
2 Ds t expressions are obtained:
pffiffiffiffiffi pffiffiffiffiffi 2 pffiffiffiffiffi
c¼ kt ðA:7Þ lim erfð kτÞ  pffiffiffiffi kτ ðA:20Þ
kτ-0 π
At the gas–liquid interface (z¼0) this becomes:

 lim expð  kτÞ  1  kτ ðA:21Þ
 1 2b
 kτ-0
J A z ¼ 0 ¼  C 0 Ds  a erfcð  cÞ þ a erfcðcÞ  pffiffiffiffi expð  c2 Þ þ expð  c2 Þ pffiffiffiffiffi
2 π Substituting the truncated expressions of erfð kτÞ and expð  kτÞ
ðA:8Þ into Eq. (A.19) gives:
pffiffiffiffi
pffiffiffiffiffi 
using the fact that the error function is an odd function the first π 1 2 pffiffiffiffiffi 1  kτ
lim E  kτ þ pffiffiffiffiffi pffiffiffiffi kτ þ pffiffiffiffi ðA:22Þ
two terms within the brackets can be combined to: kτ-0 2 2 kτ π π
a erfcð  cÞ þa erfcðcÞ ¼  2a erfðcÞ ðA:9Þ rewriting yields:
pffiffiffiffi

π 1 2 1  kτ
substituting and replacing the expressions for a, b and c yields for lim E  kτ þ pffiffiffiffi þ pffiffiffiffi ðA:23Þ
the flux:
kτ-0 2 2 π π
0sffiffiffiffiffi 1 taking the limit for τ-0:
 k pffiffiffiffiffi expð ktÞ pffiffiffiffi

J A z ¼ 0 ¼ C 0 Ds @ erf kt þ p ffiffiffiffiffiffiffiffiffiffi A ðA:10Þ π 1 1
Ds π Ds t E pffiffiffiffi þ pffiffiffiffi ¼ 1 ðA:24Þ
2 π π
244 P.W.A.M. Wenmakers et al. / Chemical Engineering Science 145 (2016) 233–244

A similar reasoning and approach can be applied to Eq. (20), Ramachandran, P.A., 2007. Gas absorption in slurries containing fine particles:
which after rewriting reads: review of models and recent advances. Ind. Eng. Chem. Res. 46 (10), 3137–3152.
sffiffiffiffiffi
Ruthiya, K.C., 2005. Mass Transfer and Hydrodynamics in Catalytic Slurry Reactors
 pffiffiffiffiffi expð  kτÞ
1pffiffiffiffi Ds pffiffiffiffiffi
(Ph.D. thesis), Technische Universiteit Eindhoven.
1
E0 ¼ π kτ þ pffiffiffiffiffi erfð kτÞ þ pffiffiffiffi ðA:25Þ Ruthiya, K.C., Van Der Schaaf, J., Kuster, B.F.M., Schouten, J.C., 2004. Modeling the
2 Dl 2 kτ π effect of particle-to-bubble adhesion on mass transport and reaction rate in a
stirred slurry reactor: influence of catalyst support. Chem. Eng. Sci. 59 (22–23),
inserting the expressions (A.20) and (A.21) and taking the limit for 5551–5558.
τ-0, E0 now becomes: Ruthiya, K.C., Van Der Schaaf, J., Kuster, B.F.M., Schouten, J.C., 2005. Model to
sffiffiffiffiffi describe mass-transfer enhancement by catalyst particles adhering to a gas-
liquid interface. Ind. Eng. Chem. Res. 44 (16), 6123–6140.
0 Ds
lim E ¼ ðA:26Þ Verrelli, D.I., Koh, P.T.L., Nguyen, A.V., 2011. Particle-bubble interaction and
kt-0 Dl attachment in flotation. Chem. Eng. Sci. 66 (23), 5910–5921, cited By 14.
Westerterp, K.R., van Swaaij, W.P.M., Beenackers, A.A.C.M., 1984. Chemical Reactor
Design and Operation. John Wiley & Sons Ltd.
Wimmers, O.J., de Sauvage Nolting, H.J.J., Fortuin, J.M.H., 1988. The effect of the size
of catalyst particles adhering to gas bubbles on the enhancement of gas
References absorption in slurry reactors. Chem. Eng. Sci. 43 (8), 2155–2159, cited By 10.
Wimmers, O.J., Fortuin, J.M.H., 1988a. The use of adhesion of catalyst particles to
Crank, J., 1975. The Mathematics of Diffusion, 2nd ed. Clarendon Press, Oxford gas bubbles to achieve enhancement of gas absorption in slurry reactors—i.
[England]. investigation of particle-to-bubble adhesion using the bubble pick-up method.
Holstvoogd, R.D., Van Swaaij, W.P.M., 1990. The influence of adsorption capacity on Chem. Eng. Sci. 43 (2), 303–312, cited By 22.
enhanced gas absorption in activated carbon slurries. Chem. Eng. Sci. 45 (1), Wimmers, O.J., Fortuin, J.M.H., 1988b. The use of adhesion of catalyst particles to
151–162, cited By 20. gas bubbles to achieve enhancement of gas absorption in slurry reactors—ii.
Holstvoogd, R.D., van Swaaij, W.P.M., van Dierendonck, L.L., 1988. The absorption of determination of the enhancement in a bubble-containing slurry reactor.
gases in aqueous activated carbon slurries enhanced by adsorbing or catalytic Chem. Eng. Sci. 43 (2), 313–319, cited By 34.
particles. Chem. Eng. Sci. 43 (8), 2181–2187, cited By 43. Wimmers, O.J., Paulussen, R., Vermeulen, D.P., Fortuin, J.M.H., 1984. Enhancement
Kantarci, N., Borak, F., Ulgen, K.O., 2005. Bubble column reactors. Process Biochem. of absorption of a gas into a stagnant liquid in which a heterogeneously cata-
40 (7), 2263–2283. lysed chemical reaction occurs. Chem. Eng. Sci. 39 (9), 1415–1422.
Levenspiel, O., 1999. Chemical Reaction Engineering, 3rd ed. John Wiley & Sons, Youssef, A.A., Hamed, M.E., Grimes, J.T., Al-Dahhan, M.H., Dudukovic, M.P., 2013.
New York. Hydrodynamics of pilot-scale bubble columns: effect of internals. Ind. Eng.
Omota, F., Dimian, A.C., Bliek, A., 2006a. Adhesion of solid particles to gas bubbles. Chem. Res. 52 (1), 43–55.
part 1: modelling. Chem. Eng. Sci. 61 (2), 823–834, cited By 11.
Omota, F., Dimian, A.C., Bliek, A., 2006b. Adhesion of solid particles to gas bubbles.
part 2: experimental. Chem. Eng. Sci. 61 (2), 835–844, cited By 13.