Available online at www.sciencedirect.

com

Powder Technology 182 (2008) 81 – 104

www.elsevier.com/locate/powtec

An efficient numerical technique for solving population balance equation

involving aggregation, breakage, growth and nucleation
Jitendra Kumar a,⁎, Mirko Peglow b , Gerald Warnecke a , Stefan Heinrich c
a
Institute for Analysis and Numerics, Otto-von-Guericke University Magdeburg, Universitätsplatz 2, D-39106 Magdeburg, Germany
b
Institute for Process Engineering, Otto-von-Guericke University Magdeburg, Universitätsplatz 2, D-39106 Magdeburg, Germany
c
Institute of Process Equipment and Environmental Technology, Otto-von-Guericke University Magdeburg,
Universitätsplatz 2, D-39106 Magdeburg, Germany
Received 27 February 2007; received in revised form 29 May 2007; accepted 30 May 2007
Available online 6 June 2007

Abstract

A new discretization for simultaneous aggregation, breakage, growth and nucleation is presented. The new discretization is an extension of the cell
average technique developed by the authors [J. Kumar, M. Peglow, G. Warnecke, S. Heinrich, and L. Mörl. Improved accuracy and convergence of
discretized population balance for aggregation: The cell average technique. Chemical Engineering Science 61 (2006) 3327–3342.]. It is shown that
the cell average scheme enjoys the major advantage of simplicity for solving combined problems over other existing schemes. This is done by a
special coupling of the different processes that treats all processes in a similar fashion as it handles the individual process. It is demonstrated that the
new coupling makes the technique more useful by being not only more accurate but also computationally less expensive. At first, the coupling is
performed for combined aggregation and breakage problems. Furthermore, a new idea that considers the growth process as aggregation of existing
particle with new small nuclei is presented. In that way the resulting discretization of the growth process becomes very simple and consistent with first
two moments. Additionally, it becomes easy to combine the growth discretization with other processes. The new discretization of pure growth is a
little diffusive but it predicts the first two moments exactly without any computational difficulties like appearance of negative values or instability etc.
The numerical scheme proposed in this work is consistent only with the first two moments but it can easily be extended to the consistency with any
two or more than two moments. Finally, the discretization of pure and coupled problems is tasted on several analytically solvable problems.
© 2007 Elsevier B.V. All rights reserved.

Keywords: Population balance; Aggregation; Breakage; Growth; Nucleation ; Particle; Batch

1. Introduction are available only for a limited number of simplified problems

In particle technologies population balance equations are
widely used, e.g. in processes such as crystallization, commi-
nution and fluidized bed granulation. They appear also in
mathematical biology as well as in aerosol science. These are
partial integro-differential equations of hyperbolic or parabolic
type. Included are conservation laws with integral fluxes. Inde-
pendent variables are particle properties, especially particle size,
over which a particle number distribution is considered for
which the population balance is formulated. Analytical solutions
⁎ Corresponding author. Tel.: +49 391 6711629; fax: +49 391 6718073.
E-mail address: jitendra.kumar@mathematik.uni-magdeburg.de (J. Kumar).
0032-5910/$ - see front matter © 2007 Elsevier B.V. All rights reserved.
doi:10.1016/j.powtec.2007.05.028
and therefore numerical solutions are frequently needed to solve
a population balance system.
Several numerical techniques including the method of succes-
sive approximations [40], the method of moments [3,28,31], the
finite elements methods [29,33,41], the finite volume scheme
[6–9,32,48] and Monte Carlo simulation methods [19,25,26,44]
can be found in the literature for solving population balance
equations. Besides the approximation of the particle property
distribution, a correct prediction of some selected properties
such as the total number of particles and total mass etc. is mainly
required in many particulate systems. These requirements lead to
the choice of sectional methods. The sectional methods approx-
imate the continuous size distribution by a finite number of
size sections (intervals, cells). A system of ordinary differential
82 J. Kumar et al. / Powder Technology 182 (2008) 81–104
equations describing the change of the number of particles in
each cell is usually obtained and then using any higher order
time integrator is solved. The accuracy of the method certainly
depends on the number of cells. Such methods rely on predicting
some selected properties of the distribution exactly rather than
capturing all details of the distribution. Moreover, simplicity and
low computational cost of these methods make them highly
practical in process engineering.
There are several sectional methods in the literature which
predict the particle property distribution accurately while using a
linear grid, see for example Sutugin and Fuchs [46], Tolfo [47],
Kim and Seinfeld [15], Gelbard et al. [10], Sastry and
Gaschignard [42] as well as Marchal et al. [30]. Linear grid
discretization has a great advantage especially in aggregation
problems since each aggregate finds a new position for its
allocation. This discretization does however have one problem
of extremely high computational cost due to the requirement of
excessively many cells to cover the property domain. Instead, a
geometric type discretization is commonly used to reduce the
computational cost. The difficulty that arises in a non-uniform
grid is, where to allocate newborn particles which do not have
exactly the size of the grid point. Several authors proposed
different methods which can be used using geometric, Gillette
[11], Batterham et al. [5], Marchal et al. [30], Hounslow et al.
[13], or arbitrary grid, Kim and Seinfeld [15], Gelbard et al. [10],
Sastry and Gaschignard [42], Marchal et al. [30], Kumar and
Ramkrishna [22]. Batterham et al. [5] produced numerical
results for aggregation that correctly predicted conservation of
total particles volume but failed to predict the correct evolution
of total numbers. On the other hand, Gelbard et al. [10] predicted
the rate of change of total particle number correctly but lost total
particle volume. Nonetheless, Hounslow et al. [13] have been
the first in the literature to produce a population balance scheme
that correctly predicts the total number and total volume of
particles. They have considered a geometric discretization with a
factor of 2 progression in size. This has the drawback that one
cannot refine the grid. Afterwards, Litster et al. [27] have
extended the technique for adjustable geometric discretizations
with a progression factor of 21/q, where q is an integer greater
than or equal to 1. It has been investigated by Wynn [49] that the
extended formulation is valid only for q b 4. Since then, Wynn
has corrected the formulation. The complex structure of the
formulation and its restriction to specified grids are the major
disadvantages of this method. Wynn [50] has also presented a
more simplified form of Hounslow's discretization to improve
some computational aspects. An evaluation of the various
sectional methods can be found in [4,17,18,34].
The first general formulation consistent with the first two
moments has been proposed as the so called fixed pivot
technique by Kumar and Ramkrishna [22]. The formulation
takes the form of Hounslow's formulation when applied on the
geometric grid with a factor of 2 progression in size. It uses point
masses at a representative location in each cell. Despite the fact
that the numerically calculated moments are fairly accurate, the
fixed pivot technique consistently over-predicts the results of
particle property distribution. The technique focuses on an
accurate calculation of some selected moments instead of
calculation of the whole particle property distribution. However,
the fixed pivot technique can only be applied for aggregation and
breakage problems. The authors have proposed a completely
different concept for combined processes of aggregation or
breakage where additionally growth and nucleation are present.
Nonetheless, in this work we test the performance of the new
scheme by comparing the numerical results calculated by the
fixed pivot technique.
Some of the existing methods overestimate the numerical
results while others are inconsistent with moments. Moreover, of
the existing sectional methods none can be implemented for
solving combined processes effectively. We wish to have a
general, accurate, easy to use, computationally less expensive
and simple scheme. By these standards, none of the previous
schemes we know of are completely satisfactory, although some
are much better than others. This leads to a clear motivation for
the development of the new scheme. Thereby the objective of
this work is to develop a general and simple scheme which
preserves all advantages of the existing schemes and can be used
to solve all processes simultaneously.
A general one-dimensional PBE for a well mixed system is
given as in [12,36]
∂nðt; xÞ Qin Qout ∂½Gðt; xÞnðt; xÞ
¼ nin ð xÞ  nout ð xÞ 
∂t Ṽ Ṽ ∂x
þ Bnuc ðt; xÞ þ Bagg ðt; xÞ  Dagg ðt; xÞ
þ Bbreak ðt; xÞ  Dbreak ðt; xÞ: ð1Þ
Here the aggregation and breakage terms are given as follows
Z
1 x
Bagg ðt; xÞ ¼ bðt; x  ϵ; ϵÞnðt; x  ϵÞnðt; ϵÞdϵ;
2 0
Z l
Dagg ð t; x Þ ¼ n ðt; x Þ bðt; x; ϵÞnðt; ϵÞdϵ;
Z l 0
Bbreak ðt; xÞ ¼ bð x; ϵÞS ðϵÞnðt; ϵÞdϵ;
0
Dbreak ðt; xÞ ¼ S ð xÞnðt; xÞ:
The nature of the process is governed by the coagulation
kernel β representing properties of the physical medium. It is
non-negative and satisfies the symmetry condition β(t, ϵ, x) =
β(t, x, ϵ). The breakage function b(x, ϵ) is the probability
density function for the formation of particles of size x from
particles of size ϵ. The selection function S describes the rate at
which particles are selected to break.
The Eq. (1) must be supplemented with the appropriate initial
and boundary conditions. The parameter x represents the size of a
particle. The first two terms on the right hand side represent the
flow into and out of a continuous process. The symbols Qin and
Qout denote the inlet and outlet flow rates from the system. The
nucleation and growth rates are given by Bnuc(t, x) and G(t, x)
respectively. The terms nuc, agg, and break have been abbre-
viated for nucleation, aggregation and breakage respectively. The
system volume is represented by Ṽ. A batch process has no net
inflow or outflow of particles. Therefore the first two terms on the
right hand side of Eq. (1) can be removed for a batch process. We
will be concerned mainly with batch processes in this work.
 J. Kumar et al. / Powder Technology 182 (2008) 81–104
It is convenient to define moments of the particle size
distribution at this time. The jth moment of the particle size
distribution n is defined as
Z l
lj ¼ x j nðt; xÞdx: ð2Þ
0
The first two moments represent some important properties of
the distribution. The zeroth (j = 0) and first (j = 1) moments are
proportional to the total number and total mass of particles
respectively. In addition to the first two moments, the second
moment of the distribution will be used to compare the
numerical results. The second moment is proportional to the
light scattered by particles in the Rayleigh limit [22,43].
2. New formulation: the cell average technique
The cell average technique has already been formulated for
pure aggregation PBE in J. Kumar [21]. As mentioned earlier,
the objective here is to extend the scheme for coupled problems.
Therefore it is necessary first to present a general idea of the cell
average technique. The entire size domain is divided into a finite
number I of small cells. The lower and upper boundaries of the
ith cell are denoted by xi−1/2 and xi+1/2 respectively. All particles
belonging to a cell are identified by a representative size of the
cell, also called grid point. The representative size of a cell can
be chosen at any position between the lower and upper
boundaries of the cell. In this work the center of the cell, the
arithmetic mean of the cell boundaries, is defined as the
representative size. The average mean volume of the particles or
the geometric mean could be other alternative choice for the
representative. A typical discretized size domain is shown
in Fig. 1. The representative of the ith cell is represented by xi =
(xi+1/2 + xi−1/2) / 2. The width of the ith cell is denoted by Δxi =
xi+1/2 − xi−1/2. The size of a cell can be fixed arbitrarily depend-
ing upon the process of application. In most of the application,
however, geometric type grids are preferred.
We wish to transform the general continuous population
balance equation into a set of I ODEs that can be solved using
any standard ODE solver. Denoting the total number in the ith
cell by Ni, we seek a set of ODEs of the following form
dNi
¼ BCA  DCA
dt |{z}
i
|{z}
i
birth due to particulate events death due to particulate events
The particulate events that may change the number
concentration of particles include breakage, aggregation, growth
and nucleation. The abbreviation CA stands for cell average.
Note that this general formulation is not similar to the traditional
sectional formulation where birth terms corresponding to the
83
each process are summed up to contribute the total birth. Here all
particulate events will be considered in a similar fashion as we
treat individual processes. The concept of combining the
processes will be further elaborated in next sections.
The first step is to compute particle birth and death in each
cell. Consideration of all possible events that lead to the
formation of new particles in a cell provides the birth term.
Similarly all possible events that lead to the loss of a particle
from a cell give the death rate of particles. The new particles in
the cell may either appear at some discrete positions or they may
be distributed continuously in accordance with the distribution
function. For example, in a binary aggregation process particles
appear at discrete points in the cell whereas in breakage process
they are often distributed everywhere according to a continuous
breakage function. Due to non-uniform grids, it is then possible
that the size of a newborn particle in a cell does not match
exactly with the representative size of that cell. The newborn
particles whose sizes do not match with any of the representative
sizes cause an inconsistency of moments in the formulation.
Therefore the aim is to remedy the inconsistency in an efficient
manner. Suppose that we just substitute the total birth and death
of particles, for example, in Eq. (3). Then the prediction of the
zeroth moment by the formulation will be predicted correctly but
the formulation will become inconsistent with respect to the first
moment. In this work we propose the cell average technique
consistent with the first two moments. But the same idea can be
generalized to conserve any two moments.
Let us demonstrate the common concepts of the cell average
technique by the following example shown in Fig. 2. Particle
births Bi1, Bi2,…, BiIi take place at positions yi1, yi2,…, yiIi
respectively due to some particulate processes in the cell i.
Here we considered the purely discrete case but analogous steps
can be performed for continuous appearance of the particles in
the cell. First we compute the total birth of particles in the ith cell
as
X
Ii
Bi ¼ Bij : ð4Þ
j¼1
Since we know the positions of the newborn particles inside
the cell, it is easy to calculate the average volume of newborn
; i ¼ 1; 2; N ; I: particles, v̄ i. It is given by the following formula
ð3Þ X
Ii
yi j B i j
j¼1
v̄i ¼ : ð5Þ
Bi
Now we may assume that Bi particles are sitting at the
position v̄ i ∈ [xi−1/2, xi+1/2]. It should be noted that the averaging
Fig. 1. A discretized size domain.
84 J. Kumar et al. / Powder Technology 182 (2008) 81–104

Fig. 2. Appearance of particles in a cell.

process still maintains consistency with the first two moments. If shown in Fig. 4. Collecting all the birth contributions, the birth
the average volume v̄ i matches with the representative size xi term for the cell average technique is given by
then the total birth Bi can be assigned to the node xi. But this is
rarely possible and hence the total particle birth Bi has to be BCA
i ¼ Bi1 k 
i ðv̄ i1 ÞHðv̄ i1  xi1 Þ þ Bi ki ðv̄ i ÞHðxi  v̄ i Þ
reassigned to the neighboring nodes such that the total number þ B i kþ þ
i ðv̄ i ÞHðv̄ i  xi Þ þ Biþ1 ki ðv̄ iþ1 ÞHðxiþ1  v̄ iþ1 Þ:
and mass remain conserved. Considering that the average ð13Þ
volume v̄ i N xi as shown in Fig. 3, the assignment of particles
must be performed by the following equations The Heaviside step function is a discontinuous function also
known as unit step function and is here defined by
a1 ðv̄ i ; xi Þ þ a2 ðv̄ i ; xiþ1 Þ ¼ Bi : ð6Þ 8
< 1; x N 0
>
1
and H ð xÞ ¼ ; x¼0 ð14Þ
>
:2
xi a1 ðv̄ i ; xi Þ þ xiþ1 a2 ðv̄ i ; xiþ1 Þ ¼ Biv̄ i : ð7Þ 0; x b 0:

Here a1 (v̄ i, xi) and a2 (v̄ i, xi+1) are the fractions of the birth Before we summarize the cell average technique it should be
Bi to be assigned at xi and xi+1 respectively. Solving the above mentioned here that the continuous number density function n(t,
equations we obtain x) can be approximated in terms of Dirac-delta distributions as

v̄ i  xiþ1 X
I
a1 ðv̄ i ; xi Þ ¼ Bi ; ð8Þ nðt; xÞc Ni dðx  xi Þ: ð15Þ
xi  xiþ1
i¼1
and
It is due to the fact that all particles in a cell are assumed to be
v̄ i  xi concentrated at the representative size of the cell. This
a2 ðv̄ i ; xiþ1 Þ ¼ Bi : ð9Þ
xiþ1  xi representation of the number density is useful to calculate the
discrete birth and death rates in a cell. We now summarize the
It is convenient for further simplifications to define a function main steps of the cell average technique discussed above as
λ as follows:
x  xi F1
kF
i ð xÞ ¼ : ð10Þ 1. Computation of birth and death rates: The discrete birth and
xi  xi F1
death rates can be obtained by substituting a Dirac-delta
The fractions can be expressed in terms of λ as representation (15) of the density into the continuous form of
total birth and death rates in a cell. Thereby computations of
a1 ðv̄ i ; xi Þ ¼ Bi kþ
i ðv̄ i Þ; ð11Þ Bi and Di are performed in the first step.
2. Computation of the volume averages: The second step is to
and
calculate the average volume of the particles in each cell, i.e.
a2 ðv̄ i ; xiþ1 Þ ¼ Bi k
iþ1 ðv̄ i Þ: ð12Þ v̄ i. This is a central step of the cell average technique and can
easily be obtained after step 1.
There are 4 possible birth fractions that may add a birth 3. Birth modification: The birth modification, crucial point for
contribution at the node xi: two from the neighboring cells and the sectional methods, is done according to Eq. (13). This
two from the ith cell. All possible birth contributions have been modified birth rate BiCA is consistent with the first two

Fig. 3. Average volume of all newborn particles.

 J. Kumar et al. / Powder Technology 182 (2008) 81–104 85

Fig. 4. Assignment of particles at xi from all possible cells.

moments. Note that there is no need to modify the death term cell average technique over the fixed pivot technique in general.
since particles are just removed from the grid points and In order to demonstrate the idea of the cell average and its
therefore the formulation remains consistent with all improvements over the fixed pivot technique let us consider a
moments due to discrete death. As a result the death term simple example shown in the Fig. 2. As a result of aggregation,
in the cell average formulation DiCA is equal to Di. breakage or any other event, particle births Bi1, Bi2, …, BiIi take
4. Solution of the set of ODEs: Substituting the values of BiCA place at the positions yi1, yi2, …, yiIi respectively in the cell i. As
and DiCA into Eq. (3) we obtain a set of ordinary differential shown in the figure, let us assume that yik b xi and yik+1 N xi. In
equation. It will be then solved by any higher order ODE order to assign them at the nodes xi−1, xi and xi+1 in such a way
solver. An appropriate solver to solve such equations will be that particle number and mass remain conserved, we use the
recommended at the end of this section. fixed pivot and the cell average strategies. The particle number
assigned at xi by the fixed pivot mechanism is given as
A general procedure of the cell average technique has been ! !
Xk
y j
 x  1 X Ii
x þ 1  y j
explained above. Further details and implementation of the cell NiFP ¼ i i j
Bi þ
i i
Bij : ð16Þ
average technique for particular and combined processes will be j¼1
x i  x i  1 j¼kþ1
x i þ 1  x i
presented later.
At this point it is important to comment on the conceptual According to the cell average mechanism the total number of
difference between the fixed pivot and the cell average particles assigned to xi is calculated by either
techniques. The basic difference between the two techniques is   Ii
v̄ i  xi  1 X
the averaging of the volume. The cell average technique first Ni ¼
CA
B j ; if v̄ i V xi ; ð17Þ
collects the total birth of particles in a cell and then distributes it xi  xi  1 j¼1 i
once to the neighboring nodes depending upon the position of
or
the volume average in the cell according to the strategy
  Ii
discussed above. On the other hand each birth which takes xi þ 1  v̄ i X
place in a cell is assigned immediately to the neighboring nodes NiCA ¼ B j; if v̄ i z xi ; ð18Þ
xi þ xi j¼1 i
with the same strategy by the fixed pivot technique. The another
substantial difference as a consequence of the averaging arises depending upon the position of v̄ i in the cell which is defined as
from the domain of particles which contributes to birth at a node.
P
Ii
The particle domain which may contribute a birth at the node xi yij Bij
is shown in Fig. 5 for both techniques. The cell average covers j¼1
v̄ i ¼ : ð19Þ
the two adjacent cells completely while the fixed pivot technique P
Ii
Bij
takes the two adjacent cells only partly into consideration. j¼1
Up to now we have presented the foundation of the cell
average technique and illustrated the difference between the two It is easy to verify that NiCA ≥ NiFP, see Appendix B.1.
techniques. We shall now deliberate about the advantages of the Equality holds if all particles appear at the same side of the

Fig. 5. Particle domain which may contribute a birth at the node xi.
86 J. Kumar et al. / Powder Technology 182 (2008) 81–104

representative xi, Bij = 0 for all j ≥ k + 1 or Bij = 0 for all j ≤ k. Substituting the Dirac-delta mass representation (15) of the
Thus, the number of particles from the ith cell assigned at xi by continuous number density n(t, x) into the above birth and death
the cell average technique is larger than that of the fixed pivot rates, we obtain
technique.
X Z pik
From the preceding discussion it is now evident that the cell
average technique retains more information of the cell, i.e. Bbreak;i ¼ Nk ðtÞSk bðx; xk Þdx; ð22Þ
kzi xi1=2
original particles that belong to the cell, during the assignment
process. The fixed pivot technique tends to spread particles, i.e. and
has more numerical dissipation. In a special case when the
average of the particles is equal to the representative size no Dbreak;i ¼ Si Ni ðtÞ: ð23Þ
distribution to neighboring cells takes place by the cell average
technique. Assignment of particles to neighboring cells for the Here the limit pki is defined as
consistency with moments causes numerical diffusion. Since the
cell average technique maintains consistency with less particle 
xi ; if k ¼ i
distribution to neighboring cells, we expect the technique to be pik ¼ ð24Þ
xiþ1=2 ; otherwise:
more accurate and less diffusive.
We shall now turn our attention to a suitable ODE solver to
The derivation of the discrete birth and death rates is given in
solve the resulting set of ODEs. When integrating the resultant
Appendix B.1.1. The total volume flux as a result of breakage
system (3) using a standard ODE routine, for example ODE45,
into the cell i is given by
ODE15S solvers in MATLAB, this may lead to negative
values for the number density at large sizes. These negative Z xiþ1=2 Z l
values may lead in the sequel to instabilities of the whole Vbreak;i ¼ xbðx; ϵÞSðϵÞnðt; ϵÞdϵ dx: ð25Þ
system. Therefore, one should take care of the positivity of the xi1=2 x

solution by the numerical integration routine. We force the

positivity in our numerical results using an adaptive time step Similar to the discrete birth rate we obtain the discrete volume
Runge–Kutta method. The step size adjustment algorithm is flux as
based on embedded Runge–Kutta formulas, originally
X Z pik
invented by Fehlberg. It uses a fifth-order method with six
Vbreak;i ¼ Nk ðtÞSk xbðx; xk Þdx: ð26Þ
functions evaluation where another combination of the six xi1=2
kzi
functions gives a fourth order method. The difference between
the two estimates is used as an estimate of the truncation error We now compute the volume average of all newborn
to adjust the step size. A more detailed description of the particles. Dividing the total volume birth Vbreak,i by the total
method and information about implementation can be found number birth Bbreak,i, we obtain the volume average v̄ break,i in the
in [38]. ith cell as
Next we present the derivation of the discrete equations of the
cell average technique for several individual and combined Vbreak;i
processes. Numerical results will be obtained for many v̄ break;i ¼ : ð27Þ
Bbreak;i
analytically solvable problems using the cell average technique
and comparisons will be made with numerical results obtained
Now the birth rate for the cell average technique can easily be
by the fixed pivot technique.
obtained by substituting the discrete birth rate (22) and the
volume average (27) into Eq. (13). Finally, the resultant set of
2.1. Pure breakage
ODEs takes the following form
Now we derive the cell average formulation for the pure dNi    
breakage. The birth and death terms of the pure breakage ¼ Bbreak;i1 k
i v̄ break;i1 H v̄ break;i1  xi1
dt    
equation have already been presented in Eq. (1). The total birth þ Bbreak;i k
i v̄ break;i H xi  v̄ break;i
and death rates of particles in the ith cell are calculated by    
þ Bbreak;i kþ
i v̄ break;i H v̄ break;i  xi
integrating the birth and death rates from xi−1/2 to xi+1/2 as    
þ Bbreak;iþ1 kþi v̄ break;iþ1 H xiþ1  v̄ break;iþ1  Si Ni ðt Þ:
Z xiþ1=2 Z l ð28Þ
Bbreak;i ¼ bðx; ϵÞSðϵÞnðt; ϵÞdϵ dx; ð20Þ
xi1=2 x
The set of Eq. (28) is a discrete formulation for solving a
general breakage problem. The form of breakage and selection
and
function and also the type of grids can be chosen arbitrarily. The
Z xiþ1=2 set of Eq. (28) together with an initial condition can be solved
Dbreak;i ¼ SðxÞnðt; xÞdx: ð21Þ with any higher order ODE solver to obtain the number of
xi1=2 particles in a cell Ni.
 J. Kumar et al. / Powder Technology 182 (2008) 81–104 87

2.2. Pure aggregation

In this section we derive the cell average technique for pure

aggregation problems. Following J. Kumar [21], the formulation
in this case takes the following form

X
jzk  
1
Bagg;i ¼ 1  dj;K bj;k Nj Nk : ð29Þ
j; k
2
xi1=2 V ðxj þxk Þ b xiþ1=2

Thus, Bi is the net rate of addition of particles to cell i by

coagulation of particles in lower cells. The net flux of volume Vi
into cell i as a result of these coagulations is therefore given by

X
jzk   Fig. 6. Interaction of disperse phase with continuous phase.
1  
Vagg;i ¼ 1  dj;k bj;k Nj Nk xj þ xk : ð30Þ
j; k
2
xi1=2 V ðxj þxk Þ b xiþ1=2 The growth of a particle can be treated as adherence of small
nuclei on the particle surface. Traditionally, growth process is
Consequently, the average volume of all newborn particles in described as continuous transfer of mass from the continuous
the ith cell v̄ i can be evaluated as phase to the disperse phase, see Fig. 6(a). Here we consider this
process as a combination of two processes. At first, the con-
Vagg;i tinuous phase transforms into small nuclei. Then, these nuclei
v̄ agg;i ¼ : ð31Þ
Bagg;i aggregate discretely with the disperse phase. This is shown in
Fig. 6(b). In this way, the numerical treatment of the growth
Furthermore, the death term is given as process becomes easier.
X
I Thus, the process can be modeled as aggregation of existing
Dagg;i ¼ Ni bi;k Nk : ð32Þ particles with the imaginary nuclei. The aggregation rate will of
k¼1 course be directly proportional to the growth rate of particles. Like
the aggregation population balance equation, the mathematical
Now the final set of discrete equations can be written as model of the growth process then takes the following form
dNi ∂nðt; xÞ
¼ Bagg;i1 k
i ðv̄ i1 ÞH ðv̄ i1  xi1 Þ ¼ bð x  x0 ; x0 Þnðt; x  x0 Þnðt; x0 Þ  bð x; x0 Þnðt; xÞnðt; x0 Þ:
dt ∂t
þ Bagg;i k þ
i ðv̄ i ÞH ðxi  v̄ i Þ þ Bagg;i ki ðv̄ i ÞH ðv̄ i  xi Þ ð34Þ
X
I
þBagg;iþ1 kþ
i ðv̄ iþ1 ÞH ðxiþ1  v̄ iþ1 Þ  Ni bi;k Nk : ð33Þ Here x0 N 0 is the size of the nuclei. In practice we assume that
k¼1 this is a small value. The first term represents the birth at x due to
aggregation of nuclei of size x0 with the particles of size x − x0.
2.3. Pure growth
Similarly the second term describes the death of particles at x
due to aggregation of particles of size x with the nuclei of size x0.
As we have mentioned in the previous section that most of the
Unlike the conventional aggregation population balance
methods for solving growth PBEs suffer from numerical
equation, Eq. (34) does not contain any integral in the birth
diffusion. The finite volume schemes are good alternatives to
and death terms because we have fixed the size of the nuclei and
solve growth PBEs, but they are usually consistent only with a
therefore a single possible event only is responsible to change
single moment. Moreover, the extension of the existing finite
the particles at x. Here we assume that the mass m0 (=x0n(t, x0))
volume schemes to arbitrary grids is extremely complicated. The
of the nuclei has a nonzero limit, i.e. m0 N 0 for x0 → 0. In the
moving mesh techniques are exceptionally good but they are
limit x0 → 0 this equation leads to the classical form of growth
difficult to combine with other processes. We have derived a
PBE. Let us consider Eq. (34) in the form
two-moment consistent discrete formulation for the pure
aggregation and pure breakage PBEs and we do expect the  
∂nðt; xÞ bðx; x0 Þnðt; xÞ  bðx  x0 ; x0 Þnðt; x  x0 Þ
same here. Therefore the aim is to derive a scheme which is ¼ x0 nðt; x0 Þ ;
∂t x0
consistent at least with the first two moments, easy to combine
with other processes and can be applied on general grids. ð35Þ
and take the limit x0 → 0. This gives
2.4. A new approach for modeling the growth process
∂nðt; xÞ ∂ ∂
¼ m0 ðbð x; x0 Þnðt; xÞÞ ¼  ðm0 bð x; x0 Þnðt; xÞÞ:
In order to accomplish the goal stated above, a completely ∂t ∂x ∂x
new perspective to treat the growth process has been introduced. ð36Þ
88 J. Kumar et al. / Powder Technology 182 (2008) 81–104

If we compare this equation directly with the classical PBE Now the birth has to be modified according to the cell average
for growth, we get the following relationship between the strategy as
aggregation kernel β and the mass of the infinitesimally small 
nuclei growth;i ¼ Bgrowth;i1 ki ðv̄ i1 ÞHðv̄ i1  xi1 Þ
BCA
þ Bgrowth;i ki ðv̄ i ÞHðxi  v̄ i Þ
Gðt; xÞ ¼ m0 bðx; x0 Þ: ð37Þ þ Bgrowth;i kþi ðv̄ i ÞHðv̄ i  xi Þ
þ Bgrowth;iþ1 kþ i ðv̄ iþ1 ÞHðxiþ1  v̄ iþ1 Þ: ð44Þ
Now the cell average formulation can easily be derived for
Eq. (34). The numerical discretization of this different form of Since v̄ i ≥ xi, the modified birth term can be simplified further
the classical growth PBE will then become easy, consistent with
 þ
two moments and versatile to apply on any type of grids. growth;i ¼ Bgrowth;i1 ki ðv̄ i1 Þ þ Bgrowth;i ki ðv̄ i Þ:
BCA ð45Þ
We shall now formulate the cell average technique for the PBE
(34). The final formulation of rate of change of number after
substituting the expressions of λ and Bgrowth,i in Eq. (45) takes
2.4.1. Numerical discretization the following form
The cell average formulation is easy to derive for the new  
dNi v̄ i1  xi1
growth PBE. First we calculate the total birth and death rates in ¼ bðxi1 ; x0 ÞNi1 nðt; x0 Þ
dt xi  xi1
the cell i as  
xiþ1  v̄ i
Z xiþ1=2 þ bðxi ; x0 ÞNi nðt; x0 Þ  bðxi ; x0 ÞNi nðt; x0 Þ:
xiþ1  xi
Bgrowth;i ¼ bðx  x0 ; x0 Þnðt; x  x0 Þnðt; x0 Þdx; ð38Þ
xi1=2 ð46Þ

and The aggregation kernel is given by the relationship given in

Z xiþ1=2
Eq. (37). The parameter m0 in Eq. (37) can be adjusted equal to
Dgrowth;i ¼ bðx; x0 Þnðt; xÞnðt; x0 Þdx: ð39Þ 1 by fixing a small value of the size of nuclei x0 and choosing n
xi1=2 (t, x0) = 1/x0. Consequently, the relationship in Eq. (37) changes
to G (t, x) = β(x, x0).
We make the assumption that x0 is small enough so that xi + In the section of numerical results we will see that formula
x0 b xi+1/2 for all i = 1, 2, …, I. This assumption ensures that (46) and the first order upwind discretization produce similar
particles just get larger inside the cells and they do not leave results. The upwind discretization is a first order finite volume
their cells. Then substituting the number density n(t, x) = ∑Ij=1Njδ scheme and can be obtained as follows. Direct integration of
(x − xj) into the above birth and death terms gives pure growth PBE over each cell gives the semi-discrete
formulation
Z xiþ1=2 XI
Bgrowth;i ¼ bðx  x0 ; x0 Þ Nj dðx  x0  xj Þnðt; x0 Þdx dNi ðtÞ    
xi1=2 j¼1
¼ G xi1=2 n t; xi1=2
dt    
¼ bðxi ; x0 ÞNi nðt; x0 Þ; ð40Þ  G xiþ1=2 n t; xiþ1=2 ; i ¼ 1; 2; N ; I: ð47Þ

and Various methods for the numerical solutions of pure growth

PBE can be obtained from different choices of the approxima-
Z xiþ1=2 XI
tion of n(t, xi−1/2) and n(t, xi+1/2) in terms of Ni(t). The easiest
Dgrowth;i ¼ bðx; x0 Þ Nj dðx  xj Þnðt; x0 Þdx approximation n(t, xi+1/2) ≈ Ni(t) / Δx gives the first order upwind
xi1=2 j¼1
difference discretization
¼ bðxi ; x0 ÞNi nðt; x0 Þ: ð41Þ
dNi ðtÞ 1      
The discrete birth and death rate terms are the same in this ¼ G xi1=2 Ni1  G xiþ1=2 Ni : ð48Þ
dt Dx
case. Since we assume that the particles are concentrated only at
the pivots, the mass of the particles in the cell will increase and To conclude, the new form of PBE (34) treats the growth
the total number of particles will remain the same. Similarly the process as aggregation of existing particle with new small nuclei.
volume flux into the ith cell is calculated by The resulting numerical discretization of the growth process
Vgrowth;i ¼ bðxi ; x0 ÞNi nðt; x0 Þðxi þ x0 Þ: ð42Þ becomes very simple and consistent with first two moments.
Additionally, it will be shown later that the new numerical
In order to apply the cell average technique, the volume formulation makes it easy to combine the growth process with
average will simply be computed by other processes. Next we will observe that the new discretization
of the growth is a little diffusive but it predicts the first two
Vgrowth;i
v̄ i ¼ ¼ xi þ x0 : ð43Þ moments exactly without any computational difficulties like
Bgrowth;i appearance of negative values or any other instabilities.
 J. Kumar et al. / Powder Technology 182 (2008) 81–104
It should be noted that the objective of this work is not to
provide a numerical scheme for pure growth problems but to
propose a rather simple and general scheme which can be used to
solve combined processes effectively. Several finite volume and
finite element schemes can be found in the literature which may
be used to solve the growth PBEs but all of them either have
problems with consistency of moments or they are not easy to
combine with other particulate processes. In the next section we
will show that the coupling of the growth process with other
particulate processes using the cell average technique becomes
trivial and considerably better results for combined processes
can be obtained without much extra computational effort.
2.5. Simultaneous aggregation and breakage
This subsection is devoted to the numerical treatment of
combined aggregation and breakage processes. When both
aggregation and breakage processes take place together, the
discrete birth and death rates for combined processes in most of
the numerical schemes are generally considered by algebraically
summing the individual birth and death rates. In other words,
conventionally the easiest way to couple the two processes is
to just add the corresponding discretizations in the following
way.
dNi
¼ BCA
agg;i þ Bbreak;i  Dagg;i  Dbreak;i ;
CA
ð49Þ
dt
This formulation is of course consistent with the first two
moments if this holds for breakage and aggregation separately.
In order to couple the two discretizations more efficiently we
take advantage of the cell average technique. The accuracy can
further be improved by using the idea of cell average for the
combined problem simply by considering the total birth and
death rates from combined processes. Thereby, we shall treat
both processes together in a similar way as we have treated the
individual processes. First we collect all the newborn particles in
a cell independently of the events that make them appear in the
cell. In our case here the events are aggregation and breakage
processes. Then we take the volume average of all newborn
particles in the cell. The remaining steps are the same we used
for the individual processes of aggregation or breakage.
Mathematically, we construct the discrete formulation for the
coupled problem as
dNi
¼ BCA
aggþbreak;i  Daggþbreak;i : ð50Þ
dt
CA
Here the terms Bagg+break,i and Dagg+break,i represent the birth
and death rates of particles in the cell i due to aggregation and
breakage respectively. The total birth and death rates in ith cell is
given by
Baggþbreak;i ¼ Bagg;i þ Bbreak;i ; ð51Þ
and
Daggþbreak;i ¼ Dagg;i þ Dbreak;i : ð52Þ
89
Now the birth rate Bagg+break,i has to be modified to make the
formulation consistent with respect to the first two moments.
Thus, the modified birth term or the cell average birth term will
be then computed as

aggþbreak;i ¼ Baggþbreak;i1 ki ðv̄ i1 ÞHðv̄ i1  xi1 Þ
BCA
þ Baggþbreak;i ki ðv̄ i ÞHðxi  v̄ i Þ
þ Baggþbreak;i kþi ðv̄ i ÞHðv̄ i  xi Þ
þ Baggþbreak;iþ1 kþ i ðv̄ iþ1 ÞHðxiþ1  v̄ iþ1 Þ: ð53Þ
The truly new ingredient is that the average value v̄ i will be
computed as
Vagg;i þ Vbreak;i
v̄ i ¼ : ð54Þ
Bagg;i þ Bbreak;i
The terms Bagg,i, Bbreak,i, Dagg,i, Dbreak,i, Vagg,i, Vbreak,i and
also the λi are the same as given in the previous sections. Next
we will test the coupled formulation on two test problems. The
modified birth term (53) and the death term (52) can be
substituted in Eq. (50) to get the final formulation. We will see
that the new way of coupling is not only more accurate but also
computationally less expensive. This is evident since in the latter
case we are distributing particles once while in the conventional
approach particles are distributed twice; once due to aggregation
and then due to breakage.
2.6. Simultaneous aggregation and nucleation
As in the previous section we have just considered that the
particles are appearing in cells independently of the event that
make them appear and presented a cell average formulation for
combined aggregation and breakage processes. A similar
approach has to be applied here for the case of simultaneous
aggregation and nucleation. The process of nucleation may be
introduced into the modeling in two ways. A common approach
is via a boundary condition at particle size 0 if the nucleation is
mono-disperse. Here we want to consider the alternative of a
source located near particle size 0. In a continuum theory both
are valid approaches.
Before we proceed to the final form of the cell average
technique, we want to comment on the conventional coupling of
these processes. Let us discuss the case of discretization of
Kumar and Ramkrishna [23] for simultaneous aggregation and
nucleation, they considered
Z xiþ1=2
dNi
¼ discretization due to aggregation þ Bnuc ðt; xÞdx:
dt xi1=2
ð55Þ
Note that in the above formulation we may gain or loose
mass if the nucleation does not take place exactly at the
representative sizes xi. It is possible that the particles could
again be distributed to neighboring cells for the consistency of
moments in the fixed pivot technique. We are not interested
here to explore more about this issue. We now present how to
deal with these problems with the cell average techniques
90 J. Kumar et al. / Powder Technology 182 (2008) 81–104

efficiently. Similar to previous section, we wish to have the Here the birth and death terms in above equation are
following discretization given as
dNi 
¼ BCA
aggþnuc;i  Dagg;i : ð56Þ aggþgrowth;i ¼ Baggþgrowth;i1 ki ðv̄ i1 ÞHðv̄ i1  xi1 Þ
BCA
dt
þ Baggþgrowth;i ki ðv̄ i ÞHðxi  v̄ i Þ
Since nucleation is the birth of the particles, there is no þ Baggþgrowth;i kþi ðv̄ i ÞHðv̄ i  xi Þ
nucleation term to be considered in the death term. An analogous þ Baggþgrowth;iþ1 kþ i ðv̄ iþ1 ÞHðxiþ1  v̄ iþ1 Þ; ð62Þ
contribution to the death term would come if we considered
harvesting of certain particle sizes. The numerical treatment and
would be analogous. Now the birth by nucleation and aggregation
Daggþgrowth;i ¼ Dagg;i þ Dgrowth;i ; ð63Þ
can be described as follows
 where
aggþnuc;i ¼ Baggþnuc;i1 ki ðv̄ i1 ÞHðv̄ i1  xi1 Þ
BCA
þ Baggþnuc;i ki ðv̄ i ÞHðxi  v̄ i Þ Baggþgrowth;i ¼ Bagg;i þ Bgrowth;i : ð64Þ
þ Baggþnuc;i kþi ðv̄ i ÞHðv̄ i  xi Þ
þ Baggþnuc;iþ1 kþ i ðv̄ iþ1 ÞHðxiþ1  v̄ iþ1 Þ: ð57Þ The volume average of particles appearing due to growth and
aggregation in the cell i is calculated as
where
Z Vagg;i þ Vgrowth;i
xiþ1=2 v̄ i ¼ : ð65Þ
Baggþnuc;i ¼ Bagg;i þ Bnuc ðt; xÞdx: ð58Þ Bagg;i þ Bgrowth;i
xi1=2
The discrete Eq. (61) has been used in our simulations to
The average values are computed as
Z xiþ1=2 calculate the evolution of particle size distribution and its
Vagg;i þ xBnuc ðt; xÞdx moments.
xi1=2 For comparison we describe an alternative by which the two
v̄ i ¼ Z xiþ1=2 : ð59Þ processes can be coupled in a different manner which is more
Bagg;i þ Bnuc ðt; xÞdx accurate but more expensive. For the simultaneous aggregation
xi1=2
and growth, one may use cell average technique for the
All other notations appearing in the formulation are already aggregation and a Lagrangian approach for the growth to obtain
defined in previous sections. So treating nucleation and a better accuracy of the numerical results. Finally, we solve the
aggregation as particle birth in cells in a similar fashion we following set of ordinary differential equations for the
have gained the consistency with the first two moments. It should simultaneous growth and aggregation
be noted that if the nucleation is mono-disperse, usually the
appearance of smallest particle, and the smallest representative is dNi
¼ BCA
agg;i  Dagg;i ; ð66Þ
chosen such that it matches exactly with the size of the mono- dt
disperse particles, then the above formulation can be rewritten as
together with
dN1
¼ BCA
agg;1  Dagg;1 þ Bnuc ðt; x1 Þ;
dxi
dt ¼ Gðxi Þ; ð67Þ
dt
and
and
dNi
¼ BCA
agg;i  Dagg;i ; i ¼ 2; 3; N ; I: ð60Þ dxiþ1=2  
dt ¼ G xiþ1=2 : ð68Þ
dt
The formulation (56) may be used for a general poly-disperse
nucleation and formulation (60), a particular case of (56), is used The first equation is for the rate of change of particle due to
for mono-disperse nucleation. aggregation while the last two equations describe the motion of
the representatives and boundaries respectively. A similar
2.7. Simultaneous growth and aggregation approach can be used to couple growth process with other
processes like breakage or simultaneous breakage and aggrega-
This section is devoted to the numerical treatment of the tion. In this work we only presented numerical results for the cell
simultaneous growth and aggregation processes. As we have seen average technique where analytical solutions are easily available
that the cell average technique can be applied to solve a pure but it can be formulated in a similar fashion for other com-
growth problem by the application of a different form of growth binations of processes.
PBE. Similar to other combined processes, in this case we like to
have the following form of the cell average formulation 2.8. Simultaneous growth and nucleation

dNi Similar to the case of simultaneous aggregation and

¼ BCA
aggþgrowth;i  Daggþgrowth;i : ð61Þ
dt nucleation, see Section 2.6, it is easy to couple the growth and
 J. Kumar et al. / Powder Technology 182 (2008) 81–104 91

nucleation processes. The cell average formulation in this case is appearing in the analytical average density is computed either
given as analytically if possible otherwise numerically by a low order
integration routine. It should be mentioned that in all following
dNi
¼ BCA
growthþnuc;i  Dgrowth;i : ð69Þ computations, the volume domain is discretized geometrically
dt into small cells by the rule xi+1/2 = 21/qxi−1/2. The parameter q
Now the birth by nucleation and growth can be described as adjusts the coarseness of the mesh.
follows
3.1. Pure breakage

growthþnuc;i ¼ Bgrowthþnuc;i1 ki ðv̄ i1 ÞHðv̄ i1  xi1 Þ
BCA
þ Bgrowthþnuc;i ki ðv̄ i ÞHðxi  v̄ i Þ
Let us first assess the accuracy of the scheme for pure
þ Bgrowthþnuc;i kþi ðv̄ i ÞHðv̄ i  xi Þ
breakage. Here we have considered only two test cases of binary
þ Bgrowthþnuc;iþ1 kþ i ðv̄ iþ1 ÞHðxiþ1  v̄ iþ1 Þ; ð70Þ breakage with linear and quadratic selection functions, see
Appendix A.1. Fig. 7 compares the average number densities
where obtained by the two methods with the analytical one for both test
Z xiþ1=2 cases. The figures have been plotted on a semi-log scale, log in x
Bgrowthþnuc;i ¼ Bgrowth;i þ Bnuc ðt; xÞdx: ð71Þ axis and linear in y axis, to show the performance of the cell
xi1=2 average technique predicting the particle distribution at small
size range. Since particles flow towards the lower size range in
The average values are computed as
the breakage process, it is important to check the performance of
Z xiþ1=2 the scheme to predict these small particles. In this simulation the
Vgrowth;i þ xBnuc ðt; xÞdx number of geometric grid points is taken to be 30. As can be
xi1=2
v̄ i ¼ Z xiþ1=2 : ð72Þ seen, the PSD obtained by the cell average technique is in
Bgrowth;i þ Bnuc ðt; xÞdx excellent agreement with the analytical solution even for a very
xi1=2 coarse grid. On the other hand the fixed pivot technique over-
predicts the numerical results. Here we present only two test
To conclude, we have presented the cell average formulation cases. Other comparisons as well as mathematical analysis will
for some cases of coupling the two selected processes. But the be presented somewhere else.
idea remains the same to couple any number of arbitrary Based on the above results, one can conclude that the
processes. In the next section we proceed to compare the numerically calculated PSDs by the two methods deviate from
numerical results with the analytical solutions in different cases each other significantly. The numerical results calculated from
considered before. the cell average technique on a coarse geometric grid are very
accurate. On the other hand, an over-prediction in the PSD at
3. Numerical results lower volumes has been observed in the fixed pivot technique.
The computation time taken by the cell average technique is
In this section, the performance of the cell average technique slightly more than for the fixed pivot technique.
is evaluated by a direct comparison of the PSD and its moments
calculated by the fixed pivot technique as well as with the 3.2. Pure aggregation
analytical solutions. Several pure and combined processes that
can be solved analytically have been considered in our The performance of the scheme for pure aggregation
comparisons. The analytical solutions in each case together problems has already been published in J. Kumar et al. [21]
with initial condition have been provided in Appendix A. and Kostoglou [16]. Therefore, for completion, we present here
The comparison between numerical and analytical results is only one test case taken from J. Kumar et al. [21].
carried out by means of average number density. Since the total The analytical and numerical results for the mono-disperse
number of particles in each cell is obtained from the numerical initial condition with the sum kernel have been shown in Fig. 8.
techniques, it is fair to compare the average number density. It The analytical solutions of this test case are given in Appendix
can be calculated in the following way A.2. The computation in this case is performed at 0.80° of
aggregation.1 Fig. 8 clearly shows the improvement of the over-
Ni
nnum ¼ ; ð73Þ prediction in PSD at large particle sizes by the cell average
i
Dxi technique over the fixed pivot technique. Fig. 8(a) includes a
where Δxi = xi+1/2 − xi−1/2. Similarly, from the analytical results comparison of the analytical and numerical solutions for the
we get variation of the second moments for the size dependent sum
Z xiþ1=2 kernel. Once again, the agreement is very good by the cell
1
nana ¼ nðt; xÞdx: ð74Þ
i
Dxi xi1=2
1
It is convenient to express the extent of aggregation by a dimensionless
These densities have been plotted at grid points. This criterion form of zeroth moment, also known as degree of aggregation. It is defined as
of comparison is used in the entire work. The integration 1 − μ(t) / μ(0).
92 J. Kumar et al. / Powder Technology 182 (2008) 81–104

upwind discretization. Since the cell average formulation is

consistent with the first two moments, the prediction of the first
two moments using the cell average technique is exact. On the
other hand the upwind discretization shows a diverging behavior
of the first moment. The prediction of the zeroth moment by the
upwind discretization is excellent due to automatic conserving
property of the upwind discretization. The corresponding particle
size distribution at the final time is plotted in Fig. 10(b). The
prediction of the particle size distribution by the two techniques
is quite similar. A small over-prediction of the results by the cell
average technique has been observed at the small as well as at
large volumes. The number density at the small volumes by the
upwind discretization is pretty good whereas it is consistently
over-predicting at the large volumes.
The numerical results for the second test case have been
presented in Fig. 10. For the constant growth rate numerical
results obtained using both techniques are quite similar.
Nevertheless, once more, the prediction of the first moment by
the upwind discretization, shown in Fig. 10(a), is slightly over-
estimated. Whereby both the moments calculated using the cell
average technique are exactly matching with the analytical
results. Fig. 10(b) shows that the numerical results for the

Fig. 7. A comparison of particle size distributions on semi-log scale for mono-

disperse initial condition.

average technique and the same diverging behavior is observed

by the fixed pivot technique. Comparison of size distribution is
made in Fig. 8(b). As for the case of size-independent
aggregation, the prediction is poor at large particle sizes by the
fixed pivot technique. Once more the degree of fit is very high in
the case of the cell average technique.

3.3. Pure growth

We have considered two test cases to access the ability of the

scheme to handle pure growth problems. In the first test case we
take exponentially distributed particle size distribution

N0
nð0; xÞ ¼ expðx=x0 Þ; ð75Þ
x0

with a linear growth rate G(x) = x. The second test case considers
the step function as an initial condition with a constant growth
rate. The analytical results for the number density and different
moments are given in Appendix A.3.
Fig. 9(a) shows numerical results for the first two moments Fig. 8. A comparison of numerical with analytical results for mono-disperse
calculated using the cell average technique and the first order initial condition and sum kernel, q = 1 and Iagg = 0.80.
 J. Kumar et al. / Powder Technology 182 (2008) 81–104 93

S(x) =S0x and constant aggregation kernel β(x, y) =β0. They used
the following two types of initial conditions

 
N0 2 N0
nð0; xÞ ¼ N0 2 x exp 2 x ; ð76Þ
x0 x0

and
 
N0 N0
nð0; xÞ ¼ N0 exp  x : ð77Þ
x0 x0

Note that next to conservation of mass these problems have

the special properties that the processes of aggregation and
breakage balance to conserve total number also. Data (77) are a
steady state solution.
We have discretized the volume domain geometrically using
the rule xi+1/2 = 2xi−1/2. Fig. 11(a) and (b) shows a comparison of
the particle size distributions for the initial condition (76) at time
t = 0.4 and t = 6 respectively. As can be seen from the figures, the
cell average technique is able to approximate the PSD to a very
high degree of accuracy. On the other hand a small over-
prediction is observed at the smaller volumes by the fixed pivot
technique.

Fig. 9. A comparison analytical and numerical results for pure growth using
exponentially distributed initial particle size distribution and linear growth rate,
grid points 130.

particle size distribution are highly diffusive. It is due to the fact

that we have chosen a discontinuous initial condition. Most of
the numerical schemes smear out near discontinuities and higher
gradients. Very fine grids near discontinuities are often required
to capture the complete details of the distribution.
As we have seen the numerical results of the particle size
distribution for pure growth problems by the cell average tech-
nique as well as by the upwind discretization are not very
accurate. A better accuracy in this case can be obtained by
making the formulation consistent with the first three moments,
see for details J. Kumar [20].

3.4. Combined aggregation and breakage

We compared the fixed pivot technique [22] and the cell

average technique with analytically solvable problems proposed
by Patil and Andrews [35]. Later these solutions were slightly
simplified by Lage [24] to the form we use here. A list of
analytical solutions used here has been included in Appendix A.4.
They considered a special case of simultaneous aggregation and Fig. 10. A comparison analytical and numerical results for pure growth using
breakage where the number of particles stays constant with step function as initial particle size distribution and constant growth rate, grid
a uniform binary breakage b(x, y) = 2/y, linear selection function points 95.
94 J. Kumar et al. / Powder Technology 182 (2008) 81–104

numerical and analytical results of PSD has also been reported in

the figure. Due to the simplicity of the problem, uniform binary
breakage, the difference between the two types of coupling is not
so significant. Nevertheless, the cell average type coupling (50)
gives better results which can be observed at the steep part of the
PSD as well as from the L1 error. Another difference has been
observed in computational time. The cell average type coupling
(50) takes about 38 s CPU time while conventional coupling (49)
takes 52 s CPU time on a Pentium-4 machine with 1.5 GHz and
512 MB RAM.
To summarize, the cell average technique provides highly
accurate results already on a very coarse grid. The accuracy can
of course be further increased by making the grid finer. The cell
average technique with the new coupling makes the technique
even more useful for combined aggregation and breakage by
being not only more accurate but also computationally less
expensive.

3.5. Combined aggregation and nucleation

We have tested the formulation for simultaneous nucleation

and aggregation. A simple case of zero initial population with
constant aggregation kernel and mono-disperse nucleation is

Fig. 11. A comparison of particle size distributions for binary breakage and
aggregation with initial condition (76), grid points 38.

Let us now consider the steady state data (77). The results for
this case are presented in Fig. 12. The prediction by the cell
average technique is again excellent in this case while a similar
over-prediction by the fixed pivot technique can be observed.
Actually this over-prediction grows with time, i.e. the fixed pivot
technique is not stable with respect to steady states. Furthermore,
the same numerical results for both the initial conditions at the
final time have been plotted on a log–log scale in Fig. 13. Once
again we can see here that the prediction of PSD by the cell
average technique is superior to the prediction with the fixed
pivot technique.
We shall now demonstrate the difference between conven-
tional coupling (49) of the cell average technique applied
separately to aggregation and breakage and the new cell average
coupling (50). Note that we are not considering the fixed pivot
technique in this comparison. Since the numerical results are
indistinguishable from the analytical results for the steady state
initial condition (77), we have only compared the numerical
results for the unsteady initial condition (76). The numerical
results have been plotted in Fig. 14. The L1 error2 between the

Fig. 12. A comparison of particle size distributions for binary breakage and
2
L1 error = ∑i|Nana
i − Ni
num
|. aggregation with initial condition (77), grid points 38.
 J. Kumar et al. / Powder Technology 182 (2008) 81–104 95

Kiparissides [2] have been included in Appendix A.5.

Formulation (60) is used for the computation in this particular
case.
In Fig. 15(a), a comparison is made between the exact
solutions of the first two moments and the numerical solution
calculated using the cell average formulation (60) for the case
β0 = 1 and Bnuc(t, x1) = 1. The volume domain is discretized
geometrically into 30 cells starting with 10− 6. As can be seen
from the figure the numerical results are in excellent agreement
with the analytical results. It should be mentioned that, after a
certain time, the total number of particles in the system is
invariant. A dynamic equilibrium with respect to the total
number has been reached at that time. In other words the particle
nucleation rate becomes equal to the total particle aggregation
rate. On the other hand, the total mass of the particles which is
proportional to the first moment increases linearly with respect
to the process time. Unfortunately an analytical solution for the
complete PSD is not easily available, only numerical results at
different times have been depicted in Fig. 15(b). A quite
complicated but systematic behavior of the results has been
observed. Clearly the particle size distribution is free from
oscillations.

Fig. 13. A comparison of particle size distributions at t = 6 for binary breakage

and aggregation plotted on log–log scale, grid points 38.

considered. Such systems with zero initial population usually

lead to oscillations in the particle size distribution. But we will
see by numerical results that the cell average technique is
completely free of oscillations. Analytical solutions for the first
two moments of this problem reported by Alexopoulos and

Fig. 14. A comparison of particle size distributions for binary breakage and
aggregation showing the difference between conventional coupling and cell Fig. 15. Numerical results for simultaneous nucleation and aggregation, grid
average coupling, grid points 38. points 30.
96 J. Kumar et al. / Powder Technology 182 (2008) 81–104

3.6. Combined growth and aggregation

In order to demonstrate the effectiveness of the cell average

technique we have applied it to two test problems with a linear
growth rate G(x) = G0x. The constant kernel β(x, y) = β0 and the
sum kernel β(x, y) = β0(x + y) for aggregation are used in the first
and second test problems respectively. An exponentially
decreasing particle size distribution has been used as initial
condition. The initial condition and analytical solutions for the
particle size distribution as well as its moments are given in
Appendix A.6. The particle volume domain was partitioned into
60 non-homogeneous cells with the rule xi+1/2 = 21/qxi−1/2.
Numerical results for the first test problems have been plotted
in Fig. 16. The variation of the first two moments with time is
presented in Fig. 16(a). The numerical and analytical solutions
for moments are in excellent agreement. On the other hand
numerical results for the particle size distribution at final time,
shown in Fig. 16(b), are slightly over-estimated. A better
accuracy, of course, can be achieved by refining the grids.
A similar computation has been performed for the second test
case of size-dependent aggregation kernel. Once more the
numerical solution of the first two moments overlaps the
analytical solutions excellently, see Fig. 17(a). The number

Fig. 17. Simultaneous growth and aggregation using the cell average technique,
β (x, y) = β0(x + y) and G (x) = G0x, grid points 65.

density distribution at final time together with the initial

Fig. 16. Simultaneous growth and aggregation using the cell average technique,
β (x, y) = β0 and G (x) = G0x, grid points 65.
condition is plotted in Fig. 17(b). As can be seen from the
figure a very good agreement between the analytical and
numerical solutions has been obtained.
We can conclude that the cell average technique can be used
for coupled processes effectively. The coupling of growth
processes with aggregation and breakage has been a big issue in
areas related to particulate systems. Different authors proposed
different coupling techniques but all of them either are
complicated [23], less accurate and unstable [13] or applicable
to limited problems [14,37]. The application of the cell average
technique to combined different processes is easy as well as
consistent and produces comparatively good results.
As mentioned before growth processes can be coupled
with other processes by computing growth using Lagrangian
approach and aggregation or breakage using the cell average
technique. In this way numerical diffusion by the growth can
significantly be reduced and considerably better results can be
obtained. In order to illustrate this coupling we have considered
the same test cases as before. Since the prediction of the first two
moments in this case is similar to the previous case, we omitted
them to present here. Only the particle size distributions
corresponding to Figs. 16(b) and 17(b) have been plotted in
 J. Kumar et al. / Powder Technology 182 (2008) 81–104 97

Fig. 18(a) and (b). Clearly, it can be seen from the figures that the
accuracy of the numerical results has been improved in both
cases.

3.7. Combined growth and nucleation

Nucleation with no particles initially present is difficult to

simulate. The analytical solution together with the initial
condition is included in Appendix A.7. The volume domain
has been partitioned into 100 geometrically spaced cells. A
comparison between the analytical and numerical solution is
shown in Fig. 19. As can be seen from the figure the numerical
results are slightly diffusive. Very fine grids are needed to
capture the details near discontinuity.
As we have seen in previous sections the coupling of growth Fig. 19. A comparison of particle size distributions with no particles initially
with other processes using a Lagrangian approach produces present in the system, grid points 120.
excellent results. An attempt in this direction has been made by
Kumar and Ramkrishna [23]. Numerical treatment of simulta-
neous growth and nucleation using the Lagrangian approach is serious difficulties in numerics. The formulation of Kumar and
not trivial because smallest representative sizes together with Ramkrishna [23] is inconsistent with the mass of the particles
their boundaries will move and there may not be a cell to place due to nucleation. It was later improved by Spicer et al. [45] by
subsequently formed nuclei. This definitely forces us to add considering the effect of the mass of new formed nuclei on the
more cells as the PSD moves to the right and therefore it leads to movement of cell representatives.

3.8. From the cell average to the fixed pivot technique

Now we shall divert our attention to the concept of choosing

representative sizes and find out some similarities of the fixed
pivot and cell average schemes. In practice, the representative
size of a cell may be chosen arbitrarily. The center of the cell, an
easy choice, is a common choice for the representative size. On
the basis of the position of the representative sizes we have the
following lemma which tells us about the similarity of two
techniques.

Lemma 3.1. If we choose the representative sizes xi as a

boundary of cells then both the cell average and fixed pivot
schemes become the same, i.e.
 CA  FP
dNi dNi
¼ if xi ¼ xiF1=2 : ð78Þ
dt dt

Proof. Consider some particles n1, n2,…, nm of respective sizes

x1, x2,…, xm appear in the ith cell due to aggregation, breakage
or any other event. Let us suppose that the representative size xi
of this cell lies at the left boundary xi−1/2. In order to make
particles appearance consistent with their number and mass,
only a fraction of them has to be assigned at xi. We calculate the
fraction assigned at xi using the cell average and the fixed pivot
techniques. It is evident that the particles have to be distributed
between xi and xi+1 according to both techniques. The assigned
fraction of particles at xi according to the fixed pivot technique
is given as
     
xiþ1  x1 xiþ1  x2 xiþ1  xm
aFP ¼ n1 þ n2 þ N þ nm :
xiþ1  xi xiþ1  xi xiþ1  xi
Fig. 18. Simultaneous growth and aggregation using the cell average technique
for aggregation and Lagrangian approach for growth, grid points 65. ð79Þ
98 J. Kumar et al. / Powder Technology 182 (2008) 81–104

According to the cell average technique, the fraction is given numerical results of the particle size distribution turn out to be
by very diffusive. Nonetheless, the cell average formulation pre-
  dicts the first two moments extremely accurately without any
xi þ 1  v̄ 1
a CA
¼ ðn1 þ n2 þ N þ nn Þ; ð80Þ computational difficulties.
xi þ 1  xi Furthermore, a new concept for the coupling of different
where the average value v̄ i is computed as processes has been introduced. All processes including growth
and nucleation have been solved using the cell average technique
x 1 n1 þ x 2 n2 þ N þ x m nm by treating them uniformly. It has been demonstrated that the
v̄ i ¼ : ð81Þ
n1 þ n 2 þ N þ nm new approach of coupling is not only more accurate but also
computationally less expensive. This is achieved by collecting
Clearly substitution of the average v̄ i from Eq. (81) into all newborn particles in a cell independently of the events
Eq. (80) gives the required result (79), i.e. aFP = aCA . To (aggregation, breakage, growth and nucleation processes) that
summarize, if the representative size of a cell lies at one of the make them appear in the cell. Then a similar treatment like the
boundaries then taking average and assigning a fraction of all cell average technique provides to individual processes is given.
particles to the neighboring representatives is the same as A substantial success of this coupling is observed for combined
assigning particles of each size separately without taking processes where growth is present because the coupling of
average of them. □ growth with other processes has been a challenging task. The
new approach of coupling makes the numerical discretization
4. Conclusions very simple and more accurate.

This work presents the extension of a numerical scheme, the Nomenclature

cell average technique that we introduced in J. Kumar [21], for a1, a 2, Fraction of particles –
solving population balance equations. The technique is based on a 3, a 4
the fixed pivot technique developed by Kumar and Ramkrishna b Breakage function m− 3
[22]. The cell average technique follows a two step strategy, one B Birth rate m− 6
to calculate average size of the newborn particles in a class and s− 1
m−
3
the other to assign them to neighboring nodes such that the B̃ Binary breakage function
s−
1
properties of interest (moments) are exactly preserved. A com-
parison of the numerical results obtained by the fixed pivot and C Constant –
m−
6
cell average techniques with the analytical results indicates that D Death rate
s−
1
the cell average technique improves the over-prediction of the
number density distribution. E Error –
The cell average technique proposed in this work allows the Gj Growth rate along internal [ij]/s
convenience of using regular (linear, geometric or equal size) and coordinate j
irregular meshes. Furthermore, the cell average scheme is H Heaviside function –
consistent with the zeroth and first moments, the same procedure I Total number of cells –
can be used to provide consistency with any two or more than two Iagg Degree of aggregation –
moments. Consistency with more than two moments can be K Constant –
m−
6
achieved by distributing particles to more grid points. The n Number density function
−3
numerical results have shown the ability of the cell average N Number m
m−
3
technique to predict very well the time evolution of the second N0 Initial number of particles
moment as well as the complete particle size distribution. An pki Limit of integral defined by m 3

important attribute of a numerical scheme, the computational the function (24)

time, has been investigated for numerous problems. It has been q Discretization parameter –
m3 s−
1
found that the computation time taken by the cell average Q Flow rate
technique is comparable to that taken by the fixed pivot technique. r Ratio xi + 1/xi –
s−
1
Moreover, for several aggregation problems it was even less than S Selection function
the computation time taken by the fixed pivot technique. t Time s
A new approach for solving growth population balance Ta Dimensionless time −
equations is presented. The growth process has been treated as for aggregation
aggregation of existing particles in the disperse phase with some Tg Dimensionless time for growth −
imaginary nuclei appearing from the continuous phase. Thus, the Ts Dimensionless time simultaneous −
growth population balance equation is converted to the processes
aggregation population balance equation. The aggregation rate u, v, x Volume of particles m3
is expressed in terms of the growth rate. Consequently, it v¯ Average volume m3
becomes a trivial task to apply the cell average technique to the V˜ System volume m3
transformed aggregation population balance equation. However, v 0, x 0 Initial mean volume m3
 J. Kumar et al. / Powder Technology 182 (2008) 81–104 99

Greek symbols Appendix A. Analytical solutions

α Integer –
β One-dimensional aggregation kernel m3 s− 1 A.1. Pure breakage
βˆ m3 s−
1
Multidimensional dimensional
aggregation kernel Ziff and McGrady [51] provided the analytical solutions for
δ(x) Dirac-delta distribution – various initial conditions and different breakage and selection
δij Kronecker delta – functions. For mono-disperse initial condition n(0, x) = δ(x − L),
Δx Size of a cell – we summarize here only in two different cases.
ϵ Global truncation error –
η Particle distribution function – Case 1 The analytical solution for the case S(x) = x and β(x, y) =
λ Fractions – 2/y, is given as
Λ Discretized domain –
μr The rth moment m3r nðt; xÞ ¼ expðtxÞðdðx  LÞ þ ½2t þ t 2 ðL  xÞhðL  xÞÞ:
m−
3

ð82Þ
μ̃ Logarithmic norm –
ν Parameter in Gaussian-like –
distribution Case 2 For the case S(x) = x2 and β (x, y) = 2/y, the solution
σ Spatial truncation error – takes the following form
τ Age of particles s
ф Limiter function – nðt; xÞ ¼ expðtx2 Þ½dðx  LÞ þ 2tLhðL  xÞ: ð83Þ
ω Constant –
Here the function θ is the step function defined as
Subscripts 
1; x b L
agg Aggregation hðx  LÞ ¼ ð84Þ
0; otherwise:
break Breakage
i, j Index A.2. Pure aggregation
in Inlet
nuc Nucleation The analytical solution for mono-disperse initial condition
out Outlet can be found in Aldous [1]. For the sum β (x, y) = x + y kernel, the
analytical number density is given
Superscripts
nðt; xÞ ¼ expðtÞBð1  expðtÞ; xÞ; with 0 V t b l; ð85Þ
ˇ Constant values
¯ Mean or average values Here the function B is defined as
ˆ Numerical approximations
ana Analytical Bðk; xÞ ¼ ðkxÞx1 expðkxÞ=x!; x ¼ 1; 2; N ; 0 V k V 1: ð86Þ
mod Modified
num Numerical A.3. Pure growth

Acronyms A.3.1. Constant growth rate

CA Cell average Moments: The rate of change of the zeroth and first moment
DPBE Discretized population balance equation is given by
FP Fixed pivot
l0 ðtÞ ¼ l0 ð0Þ: ð87Þ
GA Generalized approximation
GSD Granule size distribution and
MOL Method of line
ODE Ordinary differential equation l1 ðtÞ ¼ l1 ð0Þ þ G0 l0 ð0Þt: ð88Þ
PBE Population balance equation
Number density: The growth rate is constant, so all points on
PPD Particle property distribution
the solution profile will move at the same speed G(x) = G0. If n0
PSD Particle size distribution (x) is the initial solution, then the time evolution of number
density is given by
Acknowledgements
nðt; xÞ ¼ n0 ðx  G0 tÞ: ð89Þ
This work was supported by the Graduiertenkolleg-828,
‘Micro–Macro-Interactions in Structured Media and Particles A.3.2. Linear growth rate
Systems’, Otto-von-Guericke-Universität Magdeburg. The Moments: as we have seen that the rate of change of the
authors gratefully thank for funding through this PhD program. zeroth moment is always zero independently of growth rate.
100 J. Kumar et al. / Powder Technology 182 (2008) 81–104

pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Therefore the zeroth moment remains constant during the problem parameters to satisfy 2S0 l1 =b0 ¼ N ð0Þ. They
growth process i.e. μ0(t) = μ0(0). The rate of change of the first introduced the following dimensionless variables
moment in this case is calculated by
N ðtÞ vN ð0Þ
l1 ðtÞ ¼ l1 ð0ÞexpðG0 tÞ: ð90Þ UðTs Þ ¼ ; g¼ ; ð95Þ
N ð0Þ l1
Number density: For linear growth rate, G(x) = G0x, Kumar
and Ramkrishna [23] have provided the analytical solution for
pure growth PBE
nðv; tÞl1
Ts ¼ N ð0Þb0 t; and /ðg; Ts Þ ¼ : ð96Þ
nðt; xÞ ¼ n0 ðx expðG0 tÞÞexpðG0 tÞ; ð91Þ ½N ð0Þ2

where n0(x) is the initial particle size distribution. Now, Eq. (92) can be written in dimensionless form as
Z
∂/ 1 g
A.4. Simultaneous aggregation and breakage ¼ /ðg  x; Ts Þ/ð x; Ts Þdx  /ðg; Ts ÞUðTs Þ
∂Ts 2 0 Z l
Patil et al. [35] provided an analytical solution for the PBE 2 gh
þ UðlÞ /ð x; Ts Þdx  UðlÞ2 /ðg; Ts Þ;
with simultaneous breakage and coalescence for a special case g 2
where the total number of particles is constant. They considered ð97Þ
the following continuous PBE
Z with the initial condition
∂nðv; tÞ 1 v
¼ bðv  u; uÞnðv  u; t Þnðu; t Þdu
∂t 2 0 Z l /ðg; 0Þ ¼ expðgÞ or /ðg; 0Þ ¼ 4g expð2gÞ; ð98Þ
 nðv; t Þ bðv; uÞnðu; t Þdu pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Z l 0 where UðlÞ¼ ð2S0 l1 =b0 Þ=N ð0Þ and UðTs Þ¼ UðlÞ ¼ 1 .
þ2 S ðuÞnðu; t Þbðv; uÞdu  S ðvÞnðv; t Þ; ð92Þ By the simple use of Laplace and inverse Laplace transform,
v they obtained the following equilibrium solution for the initial
with the constraints condition ϕ(η, 0) = exp(−η),

/ðg; Ts Þ ¼ expðgÞ: ð99Þ

nðv; 0Þ z 0;
0 VZbðu; vÞ ¼ bðv; uÞ;
w If the initial condition is ϕ(η, 0) = 4η exp(−2η), then the
bðv; wÞdv ¼ 1; solution becomes
Z w
0

2vbðv; wÞdv ¼ w:
0 /ðg; Ts Þc
1
FðTs Þ  EðTs Þ ð ½E ðTs Þ2 ½ F ðTs Þ  1exp½E ðTs Þg

Kernels are  ½F ðTs Þ2 ½ E ðTs Þ  1exp½F ðTs Þg : ð100Þ
bðu; vÞ ¼ b0 ;
SðvÞ ¼ S0 v; Here F(Ts) and E (Ts) are the real roots of the polynomial
1
bðv; uÞ ¼ :
u L1 þ L2 P þ L3 p2 þ L4 p3 þ L5 p4 ; ð101Þ
Analytical solutions are obtained for the following two initial where
conditions
  3 1 1
N ð0Þ N ð0Þ L1 ¼ 1 Ts2 þ Ts3 þ Ts2 expðTs Þ;
nðv; 0Þ ¼ N ð0Þ exp  v ; ð93Þ 4 4 4
l1 l1 1 1 1
L2 ¼ 7Ts þ 3 Ts2 þ Ts3 þ T expðTs Þ  Ts2 expðTs Þ;
4 4 4
and 1 2
L3 ¼ 7 þ 10Ts þ 1 Ts þ expðTs Þ  T expðTs Þ;
2
 2   L4 ¼ 9 þ 3Ts  expðTs Þ;
N ð0Þ N ð0Þ
nðv; 0Þ ¼ N ð0Þ 2 v exp 2 v ; ð94Þ L5 ¼ 2p4 :
l1 l1
Obtaining the root of the above polynomial is difficult, there-
where N (t) is the total number of particles and μ1 = ∫0∞ vn (v, t) fore the roots will be obtained numerically. They also mentioned
dv is the first moment of the distribution. It can also be proved that after enforcing constraints on the number and volume of the
that the total number is constant by choosing the values of the floc size distribution, there are only two significant real roots of
 J. Kumar et al. / Powder Technology 182 (2008) 81–104 101

the polynomial. Lage [24] made some simplifications, their final where N0 and v0 are the total number of particles and mean
solution equivalent to the previous one is volume of particles at time t = 0, respectively. The dynamics of
the PSD in particulate systems undergoing combined particle
X2
K1 ðTs Þ þ pi K2 ðTs Þ growth and aggregation will depend on the values of two
/ðg; Ts Þ ¼ expðpi gÞ; 8Ts N 0; ð102Þ
i¼1
L2 ðTs Þ þ 4pi characteristics time constants, namely, the dimensionless
aggregation time, Ta, and the dimensionless growth time, Tg, as
where Pi's are the roots of the polynomial L1(Ts) + L2(Ts)p + 2p2,
and the other parameters are Ta ¼ b0 N0 t;
L1 ðTs Þ ¼ K1 ðTs Þ ¼ 7 þ Ts þ expðTs Þ; and
K2 ðTs Þ ¼ 2  2 expðTs Þ;
L2 ðTs Þ ¼ 9 þ Ts  expðTs Þ: ð103Þ Tg ¼ Gðv0 Þ=tv0 :
The roots of the polynomial can easily be obtained
Let us define some more dimensionless quantities
1 1 pffiffiffiffiffiffiffiffiffiffiffi
p1 ¼ ½expðTs Þ  Ts  9 þ dðTs Þ; ð104Þ L0 ¼ 1  2Kð1 l̃0 Þ=l̃0 ;
4 4
L1 ¼ 1  2Kð1p l̃ 0 Þ=l̃1ffi ;
ffiffiffiffiffiffiffiffiffiffiffi
and
L2 ¼ ðv=v0 Þ ũ1 1 Zu˜0 ;
1 1 pffiffiffiffiffiffiffiffiffiffiffi Ntot ðtÞ 1 l
p2 ¼ ½expðTs Þ  Ts  9  dðTs Þ; ð105Þ ũ0 ðt Þ ¼ ¼ nðv; t Þdv;
4 4 N0 N0 0Z
l
Vtot ðtÞ 1
where ũ1 ðt Þ ¼ ¼ vnðv; t Þdv;
N0 v0 N 0 v0 0
dðTs Þ ¼ Ts2 þ ½10  2expðTs ÞTs þ 25  26 expðTs Þ
þ expð2Ts Þ: and

A.5. Simultaneous aggregation and nucleation K ¼ Tg =Ta :

In this case, analytical solutions are given for moments by Now we will consider two particular cases.
considering the zero initial population with a constant aggrega- Constant aggregation and linear growth: For the case,
tion kernel and a constant nucleation rate. In this case the
analytical solutions for the first two moments are given by b ¼ b0 and G ¼ G0 v=v0 ;

l0 ¼ b̌tanhðb̌s=2Þ; ð106Þ one can easily obtain the following analytical solution
and  
N0 l̃20 l̃0
nðv; t Þ ¼ exp  v ; ð109Þ
l1 ¼ rs; ð107Þ v0 l̃1 l̃1
where where μ̃0 and μ̃1, are expressed as

r ¼ B0 =ðb0 N02 Þ; l̃0 ¼ 1=ð1 þ Ta =2Þ;

s ¼ b0 N0 t; l̃1 ¼ expðTg Þ:

and Sum aggregation and linear growth: For the case that β = β0
pffiffiffiffiffiffiffiffiffi (u + v) and G = G0v/v0, the analytical solution is the following
b̌ ¼ ð2rÞ:
    
N0 l̃0 1 l̃0 v 2 2L2 v
A.6. Simultaneous growth and aggregation nðv; t Þ ¼ exp   1 I1 ; ð110Þ
v0 l̃1 L2 l̃1 v0 l̃0 v0
There are various combinations of growth rates and
where I1 is the modified Bessel function of the first kind of order
aggregation kernels where analytical solutions are available.
one, and
These solutions can be found in Ramabhadran et al. [39].
 
A.6.1. Exponential initial condition 1  expðTg Þ
l̃0 ¼ exp ;
Assuming that the initial condition follows an exponential K
dependence with respect to particle volume, l̃1 ¼ expðTg Þ:

N0 For large values of Ta and x, an asymptotic expression for

nðv; t ¼ 0Þ ¼ expðv=v0 Þ; ð108Þ
v0 modified bessel function can be used to calculate density function.
102 J. Kumar et al. / Powder Technology 182 (2008) 81–104

A.7. Simultaneous growth and nucleation It gives NiCA ≥ NiFP and clearly if Bij = 0 for all j ≥ k + 1 then
equality holds.
For this case, continuous population balance takes the Case 2, v̄ i ≥ xi A similar procedure as before can be followed
following form in this case to obtain
∂nðv; tÞ ∂½GðvÞnðv; tÞ 0 Ii 1
X X Ii
þ ¼ S ðvÞ; ð111Þ j j j !
∂t ∂v B xiþ1 Bi  yi B i C
X
B C k
y j
 x
NiCA  NiFP ¼ B C
j¼1 j¼1 i1
where S is the nucleation rate. Hounslow [12] solved this
B xiþ1 xi C
i
Bij
equation for constant growth rate (G = G0) and mono-disperse @ A j¼1 x i  x i1
nuclei S(v) (=B0δ(v − v0)) of zero size (v0 = 0) with the initial ! !
X I i
xiþ1  yi j
j
X I i
xiþ1  yij
condition n (v, 0) = 0. Using Laplace and inverse Laplace  Bi ¼ Bij
transform, he obtained the following solution j¼kþ1
x iþ1  x i j¼1
x iþ1  x i
  ! !
B0 v X y  xi1
k j X Ii
xiþ1  yij
nðv; t Þ ¼ U t ; ð112Þ  i j
Bi  Bij
G0 G0
j¼1
x i  x i1 j¼kþ1
x iþ1  x i
where U is the unit step function. ! !
X k
xiþ1  yi j
j
X k
yij  xi1
¼ Bi  Bij
Appendix B. Mathematical derivations j¼1
x iþ1  x i j¼1
x i  x i1

Xk
ðxiþ1  xi1 Þðxi  yij Þ j
B.1. Comparison of accuracy of local moments ¼ B :
j¼1
ðxiþ1  xi Þðxi  xi1 Þ i ð117Þ
Case 1, v̄ i ≤ xi The particle birth assigned at xi by the cell
average technique is given as In this case we have proved that NiCA ≥ NiFP and equality
  Ii holds if Bij = 0 for all j ≤ k.
v̄ i  xi  1 X
NiCA ¼ B j ; if v̄ i V xi ; ð113Þ
xi  xi  1 j¼1 i
B.1.1. Discrete birth and death rates for breakage
where Birth rate: The birth rate in the ith interval is given as
XIi
yij Bij Z xiþ1=2 Z l

v̄ i ¼
j¼1
: ð114Þ Bi ¼ bðt; x; ϵÞSðt; ϵÞnðt; ϵÞdϵdx: ð118Þ
P
Ii xi1=2 x
Bij
j¼1
Interchanging the order of integration the above term can be
We want to show that NiCA ≥ NiFP. Thus, we consider the
rewritten as
following expression
  Ii ! Z xiþ1=2 Z ϵ
v̄ i  xi1 X j
Xk
yij  xi1 Bi ¼ bðt; x; ϵÞSðt; ϵÞnðt; ϵÞdxdϵ
Ni  Ni ¼
CA FP
B  Bij
xi  xi1 j¼1 i j¼1
x i  x i1 Z l Z
xi1=2 xi1=2
xiþ1=2
!
XIi j þ bðt; x; ϵÞSðt; ϵÞnðt; ϵÞdxdϵ
xiþ1  yi
þ Bij : ð115Þ Z xiþ1=2 Z ϵ
xiþ1=2 xi1=2
x iþ1  x i
j¼kþ1
¼ bðt; x; ϵÞSðt; ϵÞnðt; ϵÞdxdϵ
xi1=2 xi1=2
We substitute v̄ i into the above equation to get
X I Z xkþ1=2 Z xiþ1=2
0 Ii 1
X j j X Ii þ bðt; x; ϵÞSðt; ϵÞnðt; ϵÞdxdϵ: ð119Þ
B yi B i  j
xi1 Bi C ! k¼iþ1 xk1=2 xi1=2
B C X
k
y j
 x
NiCA  NiFP ¼ B C
j¼1 j¼1 i1
B xi xi1 C
i
Bij
@ A j¼1 x i  x i1 Substituting the number density n(t, x) = ∑k=1
I
Nk(t)δ(x − xk),
we obtain
! ! Z xi
X
Ii
xiþ1  yij j
XIi
yij  xi1
 Bi ¼ Bi j Bi ¼ Ni ðtÞ bðt; x; xi ÞSðt; xi Þdx
j¼kþ1
x iþ1  x i j¼1
x i  x i1 xi1=2
! ! X
I Z xiþ1=2
Xk j
yi  xi1 j
XIi
xiþ1  yij þ Nk ðtÞ bðt; x; xk ÞSðt; xk Þdx: ð120Þ
 Bi  Bij
j¼1
x i  x i1 j¼kþ1
x iþ1  x i k¼iþ1 xi1=2
! !
XIi j
yi  xi1 XI i
xiþ1  yi j We can also combine the above two terms into one as follows
¼ Bi j  Bi j Z
j¼kþ1
x i  x i1 j¼kþ1
x iþ1  x i X
I pik
Bi ¼ Nk ðtÞSðt; xk Þ bðt; x; xk Þdx; ð121Þ
XIi
ðxiþ1  xi1 Þðyij  xi Þ j k¼i xi1=2
¼ B z 0: ð116Þ
j¼kþ1
ðxi  xi1 Þðxiþ1  xi Þ i where pki = xi for k = i and pki = xi+1/2 elsewhere.
 J. Kumar et al. / Powder Technology 182 (2008) 81–104
Death rate: Similarly, we substitute the number density n(t, x)=
∑k=1
I
Nk(t)δ(x−xk) into the death rate to get
Z xiþ1=2 X
I [23] S. Kumar, D. Ramkrishna, On the solution of population balance equations
Di ¼ Sðt; xÞ Nk ðtÞdðx  xk Þdx ¼ Sðt; xi ÞNi ðtÞ:
xi1=2 k¼i
ð122Þ
References
[1] D.J. Aldous, Deterministic and stochastic model for coalescence
(aggregation and coagulation): a review of the mean-field theory for
probabilists, Bernoulli 5 (1999) 48.
[2] A.H. Alexopoulos, C.A. Kiparissides, Part II: dynamic evolution of the
particle size distribution in particulate processes undergoing simultaneous
particle nucleation, growth and aggregation, Chemical Engineering
Science 60 (2005) 4157.
[3] J.C. Barrett, J.S. Jheeta, Improving the accuracy of the moments method
for solving the aerosol general dynamic equation, Journal of Aerosol
Science 27 (1996) 1135.
[4] J.C. Barrett, N.A. Webb, A comparison of some approximate methods for
solving the aerosol general dynamic equation, Journal of Aerosol Science
29 (1998) 31.
[5] R.J. Batterham, J.S. Hall, G. Barton, Pelletizing kinetics and simulation of
full-scale balling circuits, Proceedings of the 3rd International Symposium
on Agglomeration, Nürnberg, Germany, 1981, p. A136.
[6] M.K. Bennett, S. Rohani, Solution of population balance equations with a
new combined Lax-Wendroff/Crank-Nicholson method, Chemical Engi-
neering Science 56 (2001) 6623.
[7] S. Bove, T. Solberg, B.H. Hjertager, A novel algorithm for solving population
balance equations: the parallel parent and daughter classes. Derivation,
analysis and testing, Chemical Engineering Science 60 (2005) 1449.
[8] S.C. Chang, The method of space–time conservation element and solution
element—a new approach for solving the Navier-Stokes and Euler
equations, Journal of Computational Physics 119 (1995) 295.
[9] F. Filbet, P. Laurençot, Numerical simulation of the Smoluchowski coagulation
equation, SIAM Journal on Scientific Computing 25 (2004) 2004.
[10] F. Gelbard, Y. Tambour, J.H. Seinfeld, Sectional representations for simulating
aerosol dynamics, Journal of Colloid and Interface Science 76 (1980) 541.
[11] D.A. Gillette, A study of aging of lead aerosols—II: a numerical model
simulating coagulation and sedimentation of a leaded aerosol in the
presence of an unleaded background aerosol, Atmospheric Environment 6
(1972) 433.
[12] M.J. Hounslow, A discretized population balance for simultaneous
nucleation, growth and aggregation. Phd Thesis, University of Adelaide,
South Australia, 1990.
[13] M.J. Hounslow, R.L. Ryall, V.R. Marshall, A discretized population balance
for nucleation, growth and aggregation, AIChE Journal 38 (1988) 1821.
[14] Q. Hu, S. Rohani, A. Jutan, New numerical method for solving the
dynamic population balance equations, AIChE Journal 51 (2005) 3000.
[15] Y.P. Kim, J.H. Seinfeld, Simulation of multicomponent aerosol dynamics,
Journal of Colloid and Interface Science 149 (1992) 425.
[16] M. Kostoglou, Extended cell average technique for the solution of
coagulation equation, Journal of Colloid and Interface Science 306 (2007) 72.
[17] M. Kostoglou, A.J. Karabelas, Evaluation of zero order methods for
simulating particle coagulation, Journal of Collid and Interface Science
163 (1994) 420.
[18] M. Kostoglou, A.J. Karabelas, Evaluation of numerical methods for
simulating an evolving particle size distribution in growth processes,
Chemical Engineering Communications 136 (1995) 177.
[19] F.E. Kruis, A. Maisels, H. Fissan, Direct simulation Monte Carlo method
for particle coagulation and aggregation, AIChE Journal 46 (2000) 1735.
[20] J. Kumar, Numerical approximations of population balance equations in
particulate systems. Phd Thesis, Otto-von-Guericke-University Magde-
burg, Germany, 2006.
[21] J. Kumar, M. Peglow, G. Warnecke, S. Heinrich, L. Mörl, Improved
accuracy and convergence of discretized population balance for aggregation:
the cell average technique, Chemical Engineering Science 61 (2006) 3327.
103
[22] S. Kumar, D. Ramkrishna, On the solution of population balance equations
by discretization—I. A fixed pivot technique, Chemical Engineering
Science 51 (1996) 1311.
by discretization—III. Nucleation, growth and aggregation of particles,
Chemical Engineering Science 52 (1997) 4659.
[24] P.L.C. Lage, Comments on the “An analytical solution to the population
balance equation with coalescence and breakage—the special case with
constant number of particles” by D.P. Patil and J.R.G. Andrews, Chemical
Engineering Science 57 (2002) 4253.
[25] K. Lee, T. Matsoukas, Simultaneous coagulation and breakage using
constant-N Monte Carlo, Powder Technology 110 (2000) 82.
[26] Y. Lin, K. Lee, T. Matsoukas, Solution of the population balance equation using
constant-number Monte Carlo, Chemical Engineering Science 57 (2002) 2241.
[27] J.D. Litster, D.J. Smit, M.J. Hounslow, Adjustable discretized population
balance for growth and aggregation, AIChE Journal 41 (1995) 591.
[28] G. Madras, B.J. McCoy, Reversible crystal growth-dissolution and
aggregation-breakage: numerical and moment solutions for population
balance equations, Powder Technology 143–144 (2004) 297.
[29] A.W. Mahoney, D. Ramkrishna, Efficient solution of population balance
equations with discontinuities by finite elements, Chemical Engineering
Science 57 (2002) 1107.
[30] P. Marchal, R. David, J.P. Klein, J. Villermaux, Crystallization and
precipitation engineering—I. An efficient method for solving population
balance in crystallization with agglomeration, Chemical Engineering Science
43 (1988) 59.
[31] D.L. Marchisio, R.O. Fox, Solution of population balance equations using
the direct quadrature method of moments, Journal of Aerosol Science 36
(2005) 43.
[32] S. Motz, A. Mitrovic, E.D. Gilles, Comparison of numerical methods for
the simulation of dispersed phase systems, Chemical Engineering Science
57 (2002) 4329.
[33] M. Nicmanis, M.J. Hounslow, A finite element analysis of the steady state
population balance equation for particulate systems: aggregation and
growth, Computers and Chemical Engineering 20 (1996) S261.
[34] I. Nopens, D. Beheydt, P.A. Vanrolleghem, Comparison and pitfalls of
different discretised solution methods for population balance models: a
simulation study, Computers and Chemical Engineering 29 (2005) 367.
[35] D.P. Patil, J.R.G. Andrews, An analytical solution to continuous
population balance model describing floc coalescence and breakage—a
special case, Chemical Engineering Science 53 (1998) 599.
[36] R.H. Perry, D.W. Green (Eds.), Perry's Chemical Engineers' Handbook,
7th edition, McGraw-Hill, New York, USA, 1997.
[37] A. Prakash, A.P. Bapat, M.R. Zachariah, A simple numerical algorithm and
software for solution of nucleation, surface growth and coagulation
problem, Aerosol Science and Technology 37 (2003) 892.
[38] W.H. Press, S.A. Teukolsky, W.T. Vetterling, B.P. Flannery, Numerical
Recipes in C++, 2nd edition, Cambridge University Press, 2002.
[39] T.E. Ramabhadran, T.W. Peterson, J.H. Seinfeld, Dynamics of aerosol
coagulation and condensation, AIChE Journal 22 (1976) 840.
[40] D. Ramkrishna, Population balances, Theory and Applications to Particulate
Systems in Engineering, 1st edition, Academic Press, New York, USA, 2000.
[41] S. Rigopoulos, A.G. Jones, Finite-element scheme for solution of the
dynamic population balance equation, AIChE Journal 49 (2003) 1127.
[42] K.V.S. Sastry, P. Gaschignard, Discretization procedure for the coalescence
equation of particulate processes, Industrial and Engineering Chemistry
Fundamentals 8 (1981) 355.
[43] A. Seifert, K.D. Beheng, A double-moment parameterization for
simulating autocon-version, accretion and selfcollection, Atmospheric
Research 59–60 (2001) 265–281.
[44] M. Smith, T. Matsoukas, Constant-number Monte Carlo simulation of
population balance, Chemical Engineering Science 53 (1998) 1777.
[45] P.T. Spicer, O. Chaoul, S. Tsantilis, S.E. Pratsinis, Titania formation by
TiCl4 gas phase oxidation, surface growth and coagulation, Journal of
Aerosol Science 33 (2002) 17.
[46] A.G. Sutugin, N.A. Fuchs, Formation of condensation aerosols under
repidly changing environmental conditions. Theory and method of
calculation, Journal of Aerosol Science 1 (1970) 287.
104 J. Kumar et al. / Powder Technology 182 (2008) 81–104

[47] F. Tolfo, A simplified model of aerosol coagulation, Journal of Aerosol
Science 8 (1977) 9.
[48] D. Verkoeijen, G.A. Pouw, G.M.H. Meesters, B. Scarlett, Population
balances for particulate processes—a volume approach, Chemical
Engineering Science 57 (2002) 2287.
[49] E. Wynn, Improved accuracy and convergence of discretized population
balance of Lister et al. AIChE Journal 42 (1996) 2084.
[50] E. Wynn, Simulating aggregation and reaction: New Hounslow DPB and
four-parameter summary, AIChE Journal 50 (2004) 578.
[51] R.M. Ziff, E.D. McGrady, The kinetics of cluster fragmentation and
depolymerization, Journal of Physics A: Mathematical and General 18
(1985) 3027.