Chemical Engineering Science 110 (2014) 185199

Contents lists available at ScienceDirect

Chemical Engineering Science

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

Improved kinetic Monte Carlo simulation of chemical

composition-chain length distributions in polymerization processes
P.H.M. Van Steenberge, D.R. Dhooge, M.-F. Reyniers n, G.B. Marin
Laboratory for Chemical Technology (LCT), Ghent University, Krijgslaan 281 (S5), B-9000 Gent, Belgium

H I G H L I G H T S

Kinetic Monte Carlo modeling of radical polymerization processes.

Introduction of composite binary trees to calculate bivariate distribution.

Modeling of diffusional limitations on the micro-scale.

art ic l e i nf o

a b s t r a c t

Article history:
Received 30 April 2013
Received in revised form
9 January 2014
Accepted 17 January 2014
Available online 28 January 2014

Composite binary trees are introduced for an improved kinetic Monte Carlo (kMC) calculation of
chemical composition-chain length distributions (CC-CLDs) in polymerization processes, such as the
bivariate copolymer composition-CLD (CoC-CLD). For the calculation of the CC-CLD, each leaf node of the
main tree, which reects the number of macromolecules with a given chain length, serves as a root node
for a sub-tree containing information on the CC distribution for the macromolecules with the selected
chain length. For low maximum chain lengths of 1000, the improvement consists already in a reduction
of the kMC operations by a factor between 103 and 106. The approach is illustrated for the calculation of
the CoC-CLD in free and atom transfer radical copolymerization of methyl methacrylate and styrene
while accounting for potential diffusional limitations. Main focus is on the capability of the algorithm to
ensure an accurate calculation of the average copolymer composition including high chain lengths.
& 2014 Elsevier Ltd. All rights reserved.

Keywords:
Chemical composition distribution
Chain length distribution
Diffusion
Modeling
Multivariate

1. Introduction
Polymerization is an important chemical process allowing the
production of a wide range of materials used in daily-life and for
high-tech applications (Matyjaszewski and Davis, 2002; Wang
et al., 2005; Dompazis et al., 2005; Liu et al., 2009). Depending
on the polymerization technique (e.g., radical or coordination
polymerization), the comonomers (e.g., styrene, (meth)acrylates),
the reactor conguration, and the operating conditions (e.g.,
(semi)batch) a polymer with a given microstructure can be
obtained. This polymer microstructure is codetermined by the
number of monomer units incorporated per chain, i.e., the chain
length distribution (CLD). In particular, the polymer microstructure can be related to the contribution of the different comonomer
types and the presence of short/long chain branches (S/LCBs) and
functional groups. These microstructural characteristics inuence
the chemical, rheological, mechanical and physical properties and,

Corresponding author.
E-mail address: mariefrancoise.reyniers@ugent.be (M.-F. Reyniers).

0009-2509/$ - see front matter & 2014 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.ces.2014.01.019

hence, determine the application range of the polymer product

(Kiparissides, 2006). It is therefore of paramount importance to
control the polymer microstructure and to understand its link
with the polymerization kinetics.
Typically the polymer microstructure is described by the chemical
composition-chain length distribution (CC-CLD) (Asua, 2007; Krallis
et al., 2010; Doremaele et al., 1992). This bivariate distribution
represents the fraction of macromolecules with a given chain
length and given chemical composition, i.e., number of comonomer
units of a given type, number of SCBs or LCBs. Frequently CC-CLD
denotes the copolymer composition-CLD, whereas different acronyms are used to denote other bivariate distributions. For example,
Meimaroglou et al. (2007) introduced the SCB-CLD and LCB-CLD to
describe the conditional (i.e., per value of the chain length) branching
distribution in the production of low density polyethylene (LDPE).
Similarly, Konkolewicz et al. (2011) highlighted the importance of the
chain branches-CLD (CB-CLD), in which CB is the sum of the number
of SCBs and LCBs, in the controlled radical polymerization (CRP) of nbutyl acrylate (nBuA). These authors showed that careful selection of
the mediating agent allows to suppress branch formation compared
to conventional free radical polymerization (FRP). In this work, CCD

186

P.H.M. Van Steenberge et al. / Chemical Engineering Science 110 (2014) 185199

refers to the general chemical composition distribution, whereas

CoCD denotes the copolymer composition distribution in particular.
It should be stressed that control over the CC-CLD during the
polymerization is important to attain the targeted polymer microstructure, since both the CCD and CLD codetermine, among
others, the strength, toughness and the softening/decomposition
temperature of the nal polymer product (Sperling, 1986; Young
and Lovell, 1991). For example, for polyethylene products, high
chain lengths and low branching contents are necessary to ensure
a high strength (Aggarwal and Sweeting, 1957). LDPE has a high
number of branches and a low crystallinity, whereas high density
polyethylene (HDPE) possesses fewer branches and thus results in
a stronger polymeric material for a similar CLD (White et al., 1974).
Furthermore, a small change in the comonomer (feed) composition in copolymerization processes can already induce compositional drifting and lead to off-spec material properties.
For (optical) applications, a copolymer composition drift of 35%
is undesired (Asua, 2007), since incompatibility issues result in
case polymer chains with a different composition are formed
leading to phase separation and a change in the refraction index.
For instance, Schiers and Priddy (Schiers and Priddy, 2003)
indicated that styrene acrylonitrile (SAN) copolymers characterized by a compositional drift cannot be used for the fabrication of
an optically transparent product. On the other hand, for gradient
index (GRIN) lenses, a controlled dynamic evolution of the CoCD is
needed (Asua, 2007) and compositional drift has shown to be
benecial to increase the fracture energy of polymerpolymer
interfaces (Benkowski et al., 2001). Hence, depending on the
application, it is desired to ensure a constant CC-CLD with
conversion or that the CC-CLD follows a pre-xed path.
Unfortunately, the CC-CLD is difcult to obtain experimentally
(Cools et al., 1996; Soares, 2004). For example, the measurement of
the branching content by the identication and quantication of
quaternary carbon atoms by nuclear magnetic resonance (NMR)
involves an intensive experimental procedure (Ahmad et al., 2009)
and does not allow to discriminate between identical branches on
molecules with different chain lengths. Furthermore, Soares
(Soares, 2004) indicated that the CoCD/SCB-CCD in coordination
copolymerization of ethylene and -olens can only be obtained
indirectly. The distribution of crystallization temperatures (CTD) has
to be measured and correlated to the CoCD, since it can be expected
that higher crystallization temperatures correspond to lower -olen
contents, i.e., a lower number of SCBs. Additionally, it has been shown
that the measurement of the CoC-CLD in FRP is not straightforward
(Tacx et al., 1988) and requires the combined use of size exclusion
chromatography (SEC) and molecular spectroscopy, e.g., infrared
spectroscopy, at high resolution. More recently, more advanced 2Dchromotagraphy methods have been applied (Baumgaertel et al.,
2012; Ginzburg et al., 2013; Weidner et al., 2012.) but it remains a
tedious task to obtain reliable experimental data.
Hence, taking into account the industrial demand to control the
CC-CLD on-line (Richards and Congalidis, 2006), which allows a
safe operation and efcient switch to different grades, it is
desirable to dispose of a reliable modeling tool which can calculate
a variety of bivariate distributions as a function of polymerization
time and process conditions in an efcient manner.
The deterministic two dimensional xed pivot technique (2DFPT) and the stochastic kinetic Monte Carlo (kMC) technique
(Krallis et al., 2008; Narkchamnan et al., 2011; Szymanski and
Sosnowski, 2012; Van Steenberge et al., 2012: Van Steenberge
et al., 2013) have been widely used to calculate the C(o)C-CLD.
In 2D FPT, the composition domain, representing the number of
A and B comonomer units in the copolymer chains, is discretized
leading to a reduced number of differential equations. As shown
by Krallis et al. (2010) a discretization in a forty by forty grid on a
logarithmic scale for each monomer type is necessary to calculate the

CoC-CLD in the FRP of styrene and methyl methacrylate (MMA). In the

kMC technique, following the algorithm of Gillespie (1977), a large
control volume is needed for an accurate calculation of distributions of
polymer molecules, i.e., a sufciently high initial number of molecules
has to be considered, in agreement with kMC simulation results of
other research groups (Al-Harthi et al., 2009; Vinu et al., 2012; Wang
and Broadbelt, 2010; Van Steenberge et al., 2011). Similar conclusions
for the 2D-FPT and kMC technique were formulated for the calculation
of the LCB-CLD of poly(vinyl acetate) and LPDE via FRP (Tobita and
Hatanaka, 1996; Kiparissides et al., 2010).
Recently, Hamzehlou et al. (2012) applied the kMC technique to
model the copolymerization of hydroxyethyl methacrylate and nBuA
and indicated that this technique possesses a higher potential to
retrieve microstructural information of the copolymer than deterministic solvers, albeit at a higher computational cost. The latter was
shortly afterwards conrmed by Van Steenberge et al. (2012) and
Toloza Porras et al. (2013) for the evaluation of respectively the (linear)
gradient and block quality of copolymers made by atom transfer
radical polymerization (ATRP). Furthermore, Szymanski (2009) highlighted the robustness of stochastic simulations by demonstrating that
the shape and evolution of the CoC-CLD in living polymerization
depends on the involved intrinsic homopropagation rate coefcients
and monomer reactivity ratios.
Importantly, the advent of computer cluster architectures, recent
advances in compiler technology and improved implementation of
algorithms have led to increases of multiple orders of magnitude in
calculation speed for the kMC technique, closing in on simulation
times offered by deterministic solvers as for instance included in the
software package PREDICI (Wulkow, 2009). For instance, ChaffeyMillar et al. (2007) combined the use of binary trees and the Gillespie
algorithm (Gillespie, 1977) to store and retrieve the chain lengths of
reacting macroradicals. Such binary trees were shown to be very
suitable to describe reactions of chain length distributed species in
radical polymerization and, hence, to calculate the CLD. However, for
copolymerizations or when branch formation occurs, a similarly fast
kMC calculation method is currently lacking, despite the importance
of the CC-CLD to optimize polymerization processes.
In this work, a exible and easy-to-implement kMC method is
therefore proposed based on the work of Chaffey-Millar et al. (2007)
to efciently calculate the CoC-CLD, SCB-CLD and LCB-CLD using
composite binary trees, i.e., the main tree contains the chain length
information and the sub-trees (for each chain length) contain the
(conditional) composition information. The proposed approach is
illustrated for the calculation of the CoC-CLD in bulk FRP and CRP
with styrene and MMA as comonomers, as the reduction in the
number of kMC operations is most pronounced in case the CoC is
tracked as second variate besides the chain length.
The potential inuence of diffusional limitations on the individual chemical reaction steps is accounted for and it is shown that
these limitations codetermine the size of the binary trees and thus
the efciency of the proposed method. In particular, for the CRP case
for the rst time the recently introduced coupled parallel encounter
pair model (Dhooge et al., 2013) is implemented in the kMC
algorithm to describe the inuence of diffusional limitations on the
activation/deactivation process. The proposed composite binary tree
approach can be extended in a later stage to multivariate calculations, such as the trivariate SCB-LCB-CLD or CB-CoC-CLD.

2. Material and methods

2.1. Reactions and intrinsic rate coefcients
In this work, bulk free radical polymerization (FRP) and controlled radical polymerization (CRP) of styrene (comonomer A;
STY) and methyl methacrylate (comonomer B; MMA) are modeled

P.H.M. Van Steenberge et al. / Chemical Engineering Science 110 (2014) 185199

in a laboratory scale batch reactor under isothermal polymerization

conditions (reference temperature: 80 1C). The principle of both
radical polymerization techniques is given in Scheme 1 with atom
transfer radical polymerization (ATRP) as model CRP technique.
The common reactions in ATRP and FRP are propagation and
termination reactions, the latter leading to the formation of dead
polymer molecules (P). In FRP, the polymerization is started by the
dissociation of a conventional radical initiator I2 in I radicals,
whereas in ATRP the initial presence of activator molecules (MtLyX)
allows the activation of an ATRP initiator (R0X) generating R0
radicals (Matyjaszewski and Xia, 2001). R0 and I may initiate

187

chains via propagation, forming R1 macroradicals. In addition, in

ATRP, the initiator radicals (R0) and macroradicals (Ri) can be
temporarily deactivated in dormant species (RiX (iZ0)), i.e., endgroup functionality (X; EGF) can be incorporated in the polymer
chains. Hence, in FRP two populations exist, i.e., the population of
macroradicals (R) and dead polymer molecules (P), whereas in
ATRP there are three populations, i.e., the two populations present
in FRP and the population of dormant polymer molecules (RX).
An overview of the intrinsic chemical rate coefcients is given
in Table 1 (Beuermann et al., 1997; Buback et al., 1995; Brandup
et al., 1999; Moad and Solomon, 1995; Heberger and Fischer, 1993).
Note that, as explained further, for termination no such intrinsic
rate coefcients are needed. Since in this work a polymerization
temperature of 80 1C is selected, (chain) transfer reactions can be
neglected to a rst approximation (Krallis et al., 2008; D'hooge
et al., 2012; Bentein et al., 2011). For simplicity, it is assumed that
the chemical reactivity is determined by the last monomer unit
only, i.e., a so-called terminal model is applied (Asua, 2007). For
the FRP process, 2,2-azobisisobutyronitrile (AIBN) is selected as
conventional radical initiator, whereas in the ATRP process a
typical secondary alkyl halide is used.
2.2. Apparent rate coefcients
The increase of viscosity with monomer conversion and hence
the potential inuence of diffusional limitations on the reaction
steps is accounted for by the calculation of apparent rate coefcients (Gilbert, 1992; Barner-Kowollik and Russell, 2009;
Delgadillo-Velazquez et al., 2002; Bentein et al., 2012, D'hooge
et al., 2013; Payne et al., 2013).
For termination, a population weighted apparent termination
rate coefcient is introduced, which accounts for conversion and
chain length dependencies (D'hooge et al., 2009):

Scheme 1. Principle of (a) free radical polymerization (FRP) and (b) atom transfer
radical polymerization (ATRP); for styrene/MMA homopolymerization termination
by recombination/disproportionation; chain initiation: propagation with a conventional radical initiator fragment (I) or ATRP initiator radical (R0); FRP: conventional
radical initiator (I2) and monomer (M) initially present; ATRP: ATRP initiator (R0X),
M and activator (MtLyX) initially present; Mt transition metal; X halogen atom;
L ligand; n oxidation number; f conventional radical initiator efciency; ka,da,dis,
p,t rate coefcient for activation, deactivation; dissociation, propagation, termination; i 0 (initiator related); i40 (chain length).

ij

kt;app Ri Rj
kt;app
i

!2
Ri
i

in which Ri and Rj are the concentration of macroradicals with

chain length i and j, and kijt,app is the corresponding individual

Table 1
Intrinsic chemical rate coefcients (kchem) for free radical polymerization (FRP) and atom transfer radical polymerization (ATRP) of styrene (A; STY) and methyl methacrylate
(B; MMA) at 80 1C; I2 conventional radical initiator; R0X ATRP initiator; I conventional radial initiator fragment; R0 ATRP initiator radical; i chain length; terminal
model for chemical reactivities.
Reactiona

kchem
I2b

Dissociation of
Propagation I with A
Propagation I with B
Propagation R0 with A
Propagation R0 with B
Propagation Ri,A with A
Propagation Ri,A with B
Propagation Ri,B with B
Propagation Ri,B with A
ATRP Activation R0X
ATRP Activation Ri,AX
ATRP Activation Ri,BX
ATRP Deactivation R0
ATRP Deactivation Ri,A
ATRP Deactivation Ri,B
Termination reactionsc

kdis,chem.
kp,chem,0A
kp,chem,0B
kp,chem,0A
kp,chem,0B
kp,chem,AA
kp, chem,AB
kp, chem,BB
kp, chem,BA
ka,chem,0X
ka,chem,AX
ka,chem,BX
kda,chem,0
kda,chem,A
kda,chem,B

Value ((L mol  1) s  1)

4

1.7  10
6.6  103
4.5  103
5.0  104
1.5  105
6.6  102
1.6  103
1.3  103
2.2  103
1.0
1.0
1.0
1.0  107
1.0  107
1.0  107
N/A

Reference
Brandup et al. (1999)
Heberger and Fischer (1993)
Heberger and Fischer (1993)
Van Steenberge et al. (2012)
Van Steenberge et al. (2012)
Buback et al. (1995)
Asua (2007)
Beuermann et al. (1997)
Asua (2007)
Van Steenberge et al. (2012)
Van Steenberge et al. (2012)
Van Steenberge et al., 2012)
Van Steenberge et al. (2012)
Van Steenberge et al.(2012)
Van Steenberge et al. (2012)
Johnston-Hall and Monteiro (2008)

a
For termination diffusional limitations are accounted for via the composite kt model with parameters from the RAFT-CLD-T technique (Barner-Kowollik and Russell,
2009; Johnston-Hall and Monteiro, 2008), i.e., measured apparent rate coefcients are directly used; for the other reaction steps the apparent rate coefcients are calculated
based on a parallel encounter pair model (Dhooge et al., 2013; Gilbert, 1992; Delgadillo-Velazquez et al., 2002) (Eqs. (2), (5) and (6)).
b
fchem 0.75 (Moad and Solomon, 1995); fapp via parallel encounter pair model approach (Eq. (4)).
c
No intrinsic rate coefcients needed due to RAFT-CLD-T model.

188

P.H.M. Van Steenberge et al. / Chemical Engineering Science 110 (2014) 185199

apparent termination rate coefcient, which is obtained via a composite kt model (Barner-Kowollik and Russell, 2009; Johnston-Hall
and Monteiro, 2008) with model parameters from literature
(Johnston-Hall and Monteiro, 2008). These parameters were determined by regression of experimental data obtained after applying
the reversible addition fragmentation chain transfer polymerization-chain length dependent-termination (RAFT-CLD-T) technique
using different power law models. Note that no intrinsic rate
coefcients have to be known explaining the absence of intrinsic
termination rate coefcients in Table 1. For AB cross-termination,
the geometric mean of the corresponding homo-termination apparent rate coefcients is used in agreement with related kinetic
modeling studies (Van Steenberge et al., 2012; Toloza Porras et al.,
2013).
For propagation reactions, a classical parallel encounter pair
model is employed instead (D'hooge et al., 2013; Gilbert, 1992;
Achilias and Kiparissides, 1992):
1
1
1

kapp;p kchem;p kdif f ;p

in which kchem,p and kdiff,p are respectively the intrinsic chemical

and diffusion propagation coefcient. The latter is calculated via
the Smoluchowski model (Smoluchowski, 1917):
kdif f 4sN A D

in which s is the reaction distance, NA the Avogadro constant and

D the mutual diffusion coefcient, which for propagation can be
approximated by the translational diffusion coefcient of the
monomer Dm. The latter is calculated by the Vrentas and Duda
free volume theory (Vrentas and Duda, 1977a, 1997b; Vrentas
et al., 1984; Vrentas and Vrentas, 2003). For the model parameters,
the reader is referred to Bentein et al. (2011) and D'hooge et al.
(2009).
For conventional radical initiation, the following formally
analogous equation is used to calculate the apparent initiator
efciency fapp (Achilias and Kiparissides, 1992; De Roo et al. 2005):
1
1
1

f app f chem 8sN A DI

in which fchem is the chemical conventional radical initiator efciency and DI the translational diffusion coefcient of a conventional radical initiator fragment. A factor 8 appears since two
conventional radical initiator fragments are formed upon
dissociation.
However, for ATRP activation and deactivation reactions
(Scheme 1), as recently indicated by D'hooge et al. (D'hooge
et al., 2013), the parallel' encounter pair model equations introduced above cannot be used, since they do not take into account
that activation and deactivation are reversible reactions
(Scheme 1). Instead, the following so-called coupled parallel
encounter pair model equations have to be used:
1
1
1
1

ka;app ka;dif f ka;chem K eq kda;dif f

1
1
1
K eq

kda;app kda;dif f kda;chem ka;dif f

in which Keq is the equilibrium coefcient of the activation/

deactivation process:
K eq

ka;chem
kda;chem

and ka,diff and kda,diff are the corresponding diffusional rate coefcients, which are again calculated with the Smoluchowski model
(Smoluchowski, 1917) and in which D can now be approximated by
respectively the translational activator and deactivator diffusion

coefcient. For the calculation of the ATRP specic translational

diffusion coefcients, the reader is referred to Toloza Porras et al.
(2013) and D'hooge et al. (2009).
2.3. Kinetic Monte Carlo model for simulation of the CoC-CLD
Scheme 2 presents the basic steps of the kinetic Monte Carlo
(kMC) model used in this work to simulate a radical copolymerization while accounting for diffusional limitations on the microscale. The algorithm can be seen an extension of the algorithm of
Gillespie(1977).
First, the initial number of molecules are specied per type, the
diffusion model parameters are provided and for the considered
polymerization temperature the macroscopic intrinsic rate coefcients (in s  1 or L mol  1 s  1) are calculated. The controlling
volume is subsequently calculated and the (population weighted)
microscopic or Monte Carlo apparent rate coefcients (s  1) are
obtained.
The stochastic reaction time is then calculated (Gillespie,
1977) based on a random number r1 ([0,1]) and the overall reaction
rate, i.e., the sum of the microscopic apparent reaction rates (Rj's)
per considered reaction channel j (for j 1,,N) is obtained as:

lnr 1
N
J 1 Rj

In the FRP case, conventional radical dissociation, chain initiation

with A/B, propagation with A/B and termination, either by recombination or by disproportionation, are considered as reaction
channels, whereas in the ATRP case activation of the ATRP initiator,
deactivation/termination of the ATRP initiator radical, and (de)
activation involving a macroradical with as terminal monomer
unit A and B are accounted for besides the FRP reaction channels,
except conventional radical initiation (Scheme 1). To sample
reactions of the distributed species lumped microscopic reaction
rates (i.e., the sum of the individual reaction rates) are calculated.
This implies the calculation of population weighted rate coefcients. For example, for the calculation of the termination by
recombination rate between radicals having respectively A and B
as terminal monomer unit, a summation of the radical concentrations is made with respect to the chain length of the corresponding radical populations accounting for the individual reactivities.
Next, a reaction channel is sampled using a second random
number r2 ([0,1]) while considering the reaction rates of the
considered reaction channels in a cumulative way (Gillespie,
1977):
1

j1

j1

Pj o r 2 r Pj

For the reaction channel satisfying the above criterion, the

numbers/concentrations and apparent reaction rates/probabilities
are adjusted. For reactions involving non-distributed species,
this update can be performed immediately, whereas for reactions involving a distributed species the reactants are selected by
the generation of one or two additional random numbers (r3 and/
or r4). As will be explained in Section 3.1, in this work composite
binary trees are used for the fast execution of such reaction
events.
Furthermore, for the description of the conventional radical
initiation, an extra random number ranging between [0,1] is used
to select whether a conventional radical fragment propagates or
not using the apparent conventional initiator efciency as the
probability of forming a radical which will lead to chain initiation.
In an additional step, derived properties can be calculated, such
as the monomer conversion, the number/mass average chain
length and EGF. The previous steps are repeated until the nal

P.H.M. Van Steenberge et al. / Chemical Engineering Science 110 (2014) 185199

189

Scheme 2. Basic steps for the kinetic Monte Carlo simulation for the description of radical copolymerization while accounting for apparent kinetics; main algorithm:
Gillespie's algorithm (Gillespie, 1977); macroscopic rate coefcients are converted into microscopic rate coefcients while verifying whether the reaction is unimolecular
or bimolecular (for bimolecular reactions a division by VNA with V the volume and NA the Avogadro constant is performed); lumped reaction rates are used for reactions
involving distributed species; if a reaction involving a distributed species is selected, then one or two values of the chain length are sampled to select which distributed
species react (binary trees are used; see Section 3).

polymerization time or monomer conversion is reached while

updating the (population) weighted apparent rate coefcients, the
number of molecules/concentrations and the (cumulative) apparent reaction rates.
2.4. Computer specications
All simulations were executed on a Dell computer with an Intel
CPU (Core i7-3770-3.40 GHz4 cores-BS:1600 MHz) and 16 GB
RAM (DDR3 1600 MHz NO-ECC 4  4 GB) using Windows 7 (Enterprise SP1, 64 bit). The source code was written in Fortran and
compiled using Intel(R) Visual Fortran Composer XE 2011 (Update
8 Integration for Microsoft Visual Studio 2010, 12.1.3520.2010).

3. Results and discussion

It is clear that one of the most intensive steps in the kinetic
Monte Carlo (kMC) simulation (Scheme 2) is the identication of
distributed species and the update of the individual/cumulative
apparent reaction rates/probabilities. In what follows, based on the
work of Chaffey-Millar et al. (2007), an improved method is
presented using composite binary trees to lower the computational cost of the identication process of distributed species in
polymerizations for which a CC-CLD has to be calculated.
Both bivariate and trivariate distributions are considered in the
theoretical derivation of composite binary trees and a comparison
is made with other kMC methods with respect to the number of
kMC operations to be executed. The proposed method is subsequently applied to calculate the bivariate CoCD-CLD for FRP and
ATRP of styrene and MMA while illustrating the inuence of
diffusional limitations on the dynamic behavior of the CoC-CLDs.

3.1. Development of a kMC method for CC-CLD calculations based on

composite binary trees
First, it is explained how, for a given macromolecule type,
(composite) binary trees can be used for the efcient calculation
of the univariate CLD and the bivariate CCD-CLD. Next, it is
discussed how a molecule can be identied in these binary trees
and it is explained in detail how these binary trees have to be
adapted for the main reactions in the FRP and ATRP copolymerization of MMA and STY, i.e., focus is on the calculation of the
CoC-CLD. A similar approach is formulated for the calculation of
the SCB-CLD and LCB-CLD, two other bivariate distributions.
Finally, an extension is presented for the calculation of the
trivariate SCB-CoC-CCD.
3.1.1. Representation of a macromolecule via binary trees
3.1.1.1. Univariate representation: CLD. A macromolecule (e.g., a
dead polymer molecule) is characterized in the rst place by its
chain length, i.e., the rst variate. As can be derived from Fig. 1 and
as introduced by Chaffey-Millar et al. (2007), a binary tree can be
easily contained in vectors/arrays, nodes being numbered from left
to right and from top to bottom. This results in the utmost left
node numbers being powers of two and the utmost right node
numbers being powers of two diminished by unity. The height of
this tree is denoted kmax.
In the same gure, it is shown that this binary tree can be
used to represent the chain length of a macromolecule. Consider
e.g., a macromolecule possessing the maximum chain length
imax 2kmax  1 (typically kmax 414 for FRP) which can be represented. For each such macromolecule, the value contained by the
leaf node 2kmax  1, which is located at the utmost right of the
bottom row, is increased by unity (bold lines). In other words, if
node 2kmax  1 contains the number ve, ve macromolecules of

190

P.H.M. Van Steenberge et al. / Chemical Engineering Science 110 (2014) 185199

Fig. 1. Binary tree to calculate the chain length distribution (CLD) of a macromolecule type; node numbers range from 1 to 2kmax  1 (kmax 41); three height
(number of rows) kmax; bold lines are related to the representation of a macromolecule with maximal chain length imax 2kmax  1 which can be represented;
node 1: root node; bottom row: leaf nodes.

the considered type are present possessing the maximum chain

length. Each parent node is binary connected to two underlying
child nodes.
Note that by denition, the root node, which is the top node,
corresponds to the total number of macromolecules of the
considered type (e.g., dead polymer molecule) and the leaf
nodes (bottom row) represent the number chain length distribution (CLD) after normalization of the values they contain. To
ensure a reliable calculation of the CLD, the three height should
be selected so that the leaf node 2kmax 1 is not used, implying
that the actually formed maximal chain length in the polymerization process is lower than 2kmax  1.
3.1.1.2. Bivariate representation: CC-CLD. The representation of the
chain length of a macromolecule of a certain type (Fig. 1) can be
extended to a bivariate description in a straightforward manner by
considering that the leaf node of the main tree represents a total
number of macromolecules with xed chain length and can
function as the root node of a sub-tree describing different
chemical compositions e.g., the copolymer composition, the
number of SCBs or the number of LCBs (Fig. 2). Hence, a
composite binary tree results consisting of the chain length
distribution and the (conditional, i.e., for given chain length)
chemical composition distribution.
As shown in Fig. 2, the previous CLD calculation using a tree for
the macromolecule chain length domain with height kmax is
extended to a CC-CLD calculation by constructing a sub-tree for
the second variate with height lmax 1, since one node is shared
with the main tree. The height of this second tree depends on the
maximum value pmax of the second variate. For instance, to
account for the number of SCBs in the CRP of acrylates, this
second variate pmax may be as low as eight (Konkolewicz et al.,
2011; D'hooge et al., 2010) and only three rows are needed, i.e.,
lmax 1 is much lower than kmax.
However, for copolymerization, p is the number of monomer
A units in the chain and may be as high as the chain length itself
and, hence, the height of the second tree can be as large as the
height for the rst tree (lmax 1 kmax) leading to a total height of
2kmax 1. To be more precise, the copolymer composition of a
macromolecule with for instance a chain length equal to in is
located in the sub-tree of node in 2kmax  11 (red node in Fig. 2).

Fig. 2. Composite binary tree to calculate the chemical composition-chain length

distribution (CC-CLD); i chain length; p number describing chemical composition (e.g., number of A comonomer units in a given polymer chain); node numbers
in the tree range from 1 to 2kmax lmax  1 (kmax 41; lmax 40); leaf node of the rst
tree (height kmax) contains a sub-tree with height (lmax 1; sharing one node with
main tree); total height of tree kmax lmax.

If this macromolecule possesses zero monomer A units (p 0),

then the function value of the rst node of the corresponding subtree is increased by unity, as well as the values in its parent
nodes. By convention it is thus assumed that the rst leaf node of
each sub-tree corresponds to a homopolymer B chain.

3.1.1.3. Distinction between populations. If different types of

populations of macromolecules exist during a polymerization
process, then a (composite) tree is necessary for each type of
population. For FRP homopolymerization, two trees, for the
macroradicals and for the dead homopolymer molecules, are
necessary. For FRP copolymerization, two trees for macroradicals
are necessary for a terminal reactivity model (this work) and four
trees for a penultimate reactivity model. In case of ATRP, additional
trees are necessary for the dormant macrospecies.
In all cases, the root node of the tree represents the total
number of macromolecules with the same reactivity (e.g., radicals
ending in A, dormant species ending in BX, etc.). The number
contained by the root node allows the calculation of the total
propagation rate, total termination rate and total (de)activation
rate, i.e., the lumped reaction rates (Scheme 2).

3.1.2. Identication of macromolecule in a binary tree

In this subsection, it is discussed how the kMC algorithm allows
to traverse up and down the tree to locate efciently macromolecules with given specications. The latter operations will be
extensively used to execute reaction events (see further). For
illustration purposes, the operations performed are limited to
nding macroradicals with a given chain length, i.e., only the
main tree is traversed.
As mentioned before, binary trees can conveniently be stored in
a vector/array. A movement up the tree may be simply performed
by dividing the node number by two and rounding down to the
nearest integer. A movement down the tree is performed by
multiplying the node number by two and choosing left or right
by adding respectively 0 or 1. For example, a macromolecule with
the maximum chain length can be found in the tree by the
following pseudo-code in which n is the node number (e.g.,
Fig. 3; green arrows; total number of macroradicals 39; maximum

P.H.M. Van Steenberge et al. / Chemical Engineering Science 110 (2014) 185199

191

Fig. 3. Operations to locate a macromolecule possessing the maximum chain length which can represented (green arrows) in the main tree; left: node numbers n; right:
function values m(n), i.e., number of molecules contained in the corresponding nodes on the left; total number of macroradicals 39; maximum chain length 8; tree
height kmax 4 (For interpretation of the references to color in this gure legend, the reader is referred to the web version of this article).

Fig. 4. Example trajectory to locate a macroradical (left: node numbers (Fig. 1 with kmax 4); right: function values); orange arrows: selection of macroradical 29 of the 39
macroradicals in total after a random number generation (0.74) when ranking these macroradicals from low to high chain length; maximum chain length 8; tree
height kmax 4 (For interpretation of the references to color in this gure legend, the reader is referred to the web version of this article).

chain length8; tree height 4):

n1
DO while n is lower than 2kmax
n 2n
n n1
ENDDO

10

In general, random numbers are generated between 0 and 1 to

determine which macromolecule is reacting (cf. Scheme 2). In Fig. 4
such identication process is illustrated by the orange arrows. Here,
the 29th macroradical, when ranking the radicals with respect to
their length from low to high, is selected since the random number
r is assumed to have a value of 0.74 so that after multiplication with
the total number of macroradicals, i.e., the function value for the
root node (here m(1)39) a value of 29 is obtained after rounding
up, i.e., mtarget 29.
For the identication of this 29th radical in the binary tree, rst
the function value for the second node (m(2)31) is compared
with mtarget 29 (right gure rst orange arrow). Since 29 is
smaller than 31, a step toward the left is taken by multiplying
the current node number n 1 by two, yielding the new node
number n 2 (left gure rst orange arrow). Subsequently, 29 is
compared again to the left child node which contains 19 (right
gure next orange arrow). Since 19 is smaller than 29, a step
toward the right is taken by multiplying the current node
number (1) by two and adding one, arriving at the new node
number (n 5) (left gure next orange arrow). At the same time,
target value of 29 for m is reduced by 19, i.e., mtarget becomes 10,
since a step toward the right is taken. The new target value is
compared with the value in node n 5 (12), which is greater, so
again a step toward the right must be taken. The node number
n 5 is thus again multiplied by two and increased by one, arriving
at the desired leaf node n 11, which represents a chain length
of 4 for the reacting macroradical. For more mathematical details
the reader is referred to Section 1.2 of the Supporting Information.
3.1.3. Update of bivariate CoC-CLD for a given reaction event in FRP
and ATRP
In this subsection, propagation of initiator radicals is taken as
exemplary reaction. A distinction can be made between propagation

with A and B. For the other reactions, the reader is referred to the
Supporting Information.
Dissociation of I2 or activation of R0X is treated conventionally
according to Gillespie's algorithm (Gillespie, 1977). In a next step,
propagation of the formed initiator radical (I or R0) with monomer
B forms the rst element of the population of macroradicals. Only
steps toward the left must be taken and the node value of all
parent nodes of the node 2kmax  1 must be increased by unity as
well as this last node. In addition, the sub-tree, which describes
the copolymer composition has to be updated similarly, taking
into account that by convention (Fig. 2) the rst leaf node of each
sub-tree represents the number of homopolymer B chains.
In case the conventional/ATRP initiator radical propagates with
monomer A, then the last left step should be replaced by a right
step because the corresponding node represents a macroradical
with chain length one and one monomer A (Section 1.3.1.1.2 in the
Supporting Information).

3.1.4. Update of bivariate SCB/LCB-CLD and trivariate SCB-CoC-CLD

In case homopolymerizations are considered, in which chain
length and amount of SCBs or LCBs are the variates, earlier
introduced kMC operations can be applied to calculate the
bivariate distribution.
For the S/LCB-CLD, switching between trees of macroradicals is
related to the occurrence of inter- or intramolecular H-abstraction
reactions leading to the formation of mid-chain macroradicals and
subsequent propagation reforming end-chain radicals (Hamzehlou
et al., 2012; Junkers and Barner-Kowollik, 2008).
Note that in case both the number of SCBs and LCBs per chain
length are desired, a second sub-tree has to be dened, i.e., Fig. 2
has to be extended with a third three with height mmax 1, as
shown in Fig. 5 for the general case of a SCB-CC-CLD. This
extension, however, does not lead to a much higher computational
cost, since the maximum of SCB/LCBs is very low, as explained
above. In total four populations exist in FRP, namely the dead
polymer molecules, the end-chain macroradicals, the mid-chain
radicals with the radical center located near the chain end (e.g.,
pen-penultimate monomer unit) and the mid-chain macroradicals
with the radical center in the middle region of the chain.
For the update of reactions related to branch formation, the
reader is referred to Section 1.4 of the Supporting Information.

192

P.H.M. Van Steenberge et al. / Chemical Engineering Science 110 (2014) 185199

Fig. 5. Composite binary tree to calculate the short chain branch-chemical composition-chain length distribution (SCB-CoC-CLD); i chain length; p number describing
chemical composition (e.g., number of A comonomer units in a given polymer chain or number of LCBs); q number describing chemical composition (number of SCBs);
node numbers in the tree range from 1 to 2kmax lmax mmax  1 (kmax 41; lmax 40; mmax 40); leaf node of the rst tree (height kmax) contains a sub-tree with height (lmax 1;
sharing one node with main tree); total height of tree kmax lmax mmax.

Table 2
The strength of composite binary trees for kMC simulations illustrated via a comparison of the number of operations for low maximum chain lengths. imax maximum chain
length; pmax maximum value for second variate; qmax maximum value for third variate.
Developer(s)

Gillespie

Chaffey-Millar et al.

This work

This work

Data structure

Vectors

Non-composite binary trees

composite binary trees

(bivariate)

composite binary trees

(trivariate)

Algorithm type

Linear

Logarithmic

Logarithmic

Logarithmic

Log(2,2048) 11

N/A

N/A

8  Log(2, 2048) 8  1188

N/A

1024  Log(2, 2048)

1024  11 11,264

Log(2, 2048  16)

11 4
Log(2,2048  2048)
11 11 22

Log(2,2048) 
8  8 11  8  8792

Log(2,2048  16)  8
(11 4)  8 120

# of operations for CLD calculation with

1024
chain lengths up to 1024
1024  8 9192
# of operations for SCB-CLD calculation with
chain lengths up to 1024 and up to 8 SCBs
1024  1024 1,048,576
# of operations for operations for CoC-CLD
calculation for chain lengths up to 1024 and up to
1024 monomer A units
1024  8  8 65,536
# of operations for operations for SCB-LCB-CLD
calculation for chain lengths up to 1024
and up to 8 SCBs and up to 8 LCBs
1024  1024  8 8,388,608
# of operations for operations for CB-CoC-CLD
calculation for chain lengths up to 1024
and up to 8 CBs

Log(2,2048) 
Log(2,2048  2048) 
1024  8 11  1024  890,112 8(11 11)  8176

N/A

Log(2,2048  16  16)
11 4 4 19
Log
(2,2048  2048  16)
11 11 4 26

For each binary tree the number of nodes is equal to twice the maximum value of the variate.

3.1.5. Comparison with other kMC methods

The strength of the proposed composite binary tree method
can be deduced from Table 2 in which this method is compared
with more traditional methods, such as the original linear vector
based method of Gillespie (1977) and the non-composite binary
tree method of Chaffey-Millar et al. (2007), using the number of
kMC operations. A comparison of methods in such way is valid
regardless of the computer conguration (Wang and Broadbelt,
2010). It follows that, already for low maximal chain lengths
(103), an enhancement in number of operations is obtained with
a factor between 103 and 106. The effect is most pronounced in
case the copolymer composition has to be tracked as one of the
multivariates besides the chain length.

3.2. Application of composite binary tree structures for the

calculation of the CoCD-CLD in radical polymerization
In this section, the above introduced method is applied to
describe the copolymerization of STY and MAA under free and
controlled radical conditions. It is shown that the developed
method allows to capture the dynamic behavior of both polymerization processes, which inherently lead to different CoC-CLDs, and
that diffusional limitations greatly affect the evolution of these
CoC-CLDs with increasing conversion and thus inuence the sizes
of the binary trees and the efciency of the method introduced
above. Additionally, a comparison with experimental data is
included in Supporting Information.

P.H.M. Van Steenberge et al. / Chemical Engineering Science 110 (2014) 185199

193

Fig. 6. Conversion prole and number average chain length (xn) and polydispersity index (PDI) as a function of conversion in the free radical polymerization (FRP) of styrene
(STY) and methyl methacrylate (MMA); 80 1C; [STY]0 [MMA]0 4.5 mol L  1; [I2]0:([STY]0 [MMA]0) 0.005; dashed blue lines: intrinsic model (no diffusional limitations
accounted for); full green line: only diffusional limitations on termination (gel-effect); red dotted line: only diffusional limitations on termination and radical initiation (geleffect and cage-effect); dashed dotted black lines: diffusional limitations on termination, propagation and conventional radical initiation (gel-effect, glass-effect and cageeffect); initial number of monomers 2  109 (For interpretation of the references to color in this gure legend, the reader is referred to the web version of this article).

Fig. 7. Population weighted apparent termination, apparent propagation rate coefcient and apparent conventional radical initiator efciency in the free radical
polymerization (FRP) of styrene (STY) and methyl methacrylate (MMA) as a function of conversion; 80 1C; [S]0 [MMA]0 4.5 mol L  1; [I2]0:([S]0 [MMA]0) 0.005; dashed
blue lines: intrinsic model (no diffusional limitations accounted for); full green line: only diffusional limitations on termination (gel-effect); red dotted line: only diffusional
limitations on termination and radical initiation (gel-effect and cage-effect); dashed dotted black lines: diffusional limitations on termination, propagation and conventional
radical initiation (gel-effect, glass-effect and cage-effect); termination shown for recombination of macroradicals with STY as terminal monomer unit; propagation shown for
STY homopropagation; initial number of monomers 2  109 (For interpretation of the references to color in this gure legend, the reader is referred to the web version of
this article).

3.2.1. FRP of STY and MMA

3.2.1.1. Conversion and control over chain length. For the isothermal
(80 1C) FRP of styrene and MMA using equimolar initial amounts
of both comonomers and 0.01 mol% of conventional radical
initiator with respect to monomer, Fig. 6(a) shows the simulated
conversion prole with (black dashed dotted line) and without
(dashed blue line) accounting for diffusional limitations on
termination, propagation and conventional radical initiation.
In the same gure, the conversion proles are also given in case
only diffusional limitations on (i) termination (full green line;
intrinsic propagation and fapp fchem) and (ii) on termination and
propagation (dotted red line; fapp fchem) are accounted for so that
the individual effect of diffusional limitations on termination,
propagation and conventional radical initiation can be evaluated.
The corresponding changes of the population weighted apparent
rate coefcients are given in Fig. 7. For termination, termination by
recombination of macroradicals having STY as terminal unit is
selected.
Clearly, diffusional limitations have a signicant effect on
the polymerization rate. Due to diffusional limitations on termination the polymerization is accelerated from low conversions
onwards, a phenomenon known as the gel- or Trommsdorff effect
(Trommsdorff et al., 1948; Dube et al., 1997) (green full line vs. blue
dashed line in Fig. 6(a)). Radicals are less prone to termination and
can thus take up a higher number of monomer units before they
are transformed into dead polymer molecules. Although, at high
conversion this acceleration is counteracted by diffusional limitations on propagation, a so-called glass effect (Achilias, 2007), and

by diffusional limitations on conventional radical initiation, also

denoted as a (apparent) cage effect (Achilias, 2007; Kurdikar and
Peppas, 1994; Achilias and Kiparissides, 1992) (green full line vs.
red dotted and black dashed line in Figs. 6(a) and 7(b) and (c)).
As can be seen in Fig. 6(b) and (c) the number average chain
length (xn) and polydispersity index (PDI) are also inuenced by
diffusional limitations. The gel-effect allows a longer growth of
macroradicals before they are terminated leading to higher xn
values (dashed blue line vs. full green line in Fig. 6(b)). It can be
seen that in the absence of diffusional limitations a xn of circa 100
would be obtained but due to diffusional limitations on termination a circa eightfold increase is obtained for xn clearly, showing
that for a real polymerization process the binary trees are large.
In the absence of a cage- and glass-effect (full green line), xn
would somewhat drop at high conversion, since under such
conditions new polymer chains would still be formed as chain
initiation and propagation can still occur. As at such conversions,
the monomer concentration is much lower, these polymer chains
possess a much lower chain length explaining the decrease of the
xn prole (full green line, Fig. 6(b)). At the same time, the PDI
increases as a wide spectrum of chain lengths is formed. In reality,
the decrease of xn and the increase of PDI are attenuated, since the
cage- and glass-effect are occurring, as evidenced by the only
slight drop of the corresponding xn and the slight increase of the
PDI at high conversion (red dotted and black dashed dotted line).
The latter can also be deduced from Fig. 8 in which the number
CLD is depicted at a conversion of 0.10, 0.50 and 0.90 in case
the gel-, glass- and cage-effect are all accounted for (top row) and

194

P.H.M. Van Steenberge et al. / Chemical Engineering Science 110 (2014) 185199

Fig. 8. Number chain length distribution (CLD) at a monomer conversion of 0.10, 0.60 and 0.90 in the free radical polymerization (FRP) of styrene (STY) and methyl
methacrylate (MMA); 80 1C; [S]0 [MMA]0 4.5 mol L  1; [I2]0:([S]0 [MMA]0) 0.005: top row: diffusional limitations on termination, propagation and conventional radical
initiation accounted for; bottom row: intrinsic model; initial number of monomers 2  109.

an intrinsic model (bottom row) would be used instead. It follows

that, at low to intermediate conversions, the number CLD shifts to
higher chain lengths as compared to the prediction with the
intrinsic model and at high conversions a limited movement of
the number CLD is obtained to lower chain lengths.
3.2.1.2. Evolution of the CoC-CLD with conversion. Fig. 9 shows the
corresponding number CoC-CLDs for Fig. 8. Again the top row
corresponds with the full kinetic model accounting for diffusional
limitations on termination, propagation and initiation, whereas
the bottom row relates to the intrinsic model. In agreement with
the xn proles, the number CoC-CLD stretches out to higher chain
lengths for the full kinetic model. Whereas according to the
intrinsic kinetic model the CoC-CLD is static (bottom row of
subgures), the evolution of the CoC-CLD using the full model is
dynamic.
In particular, at very high conversion, as explained above, dead
polymer molecules with a lower chain length are formed, leading
to an additional movement of the number CoC-CLD (middle vs.
right top subgure). It can be seen that these newly formed dead
polymer molecules contain a high amount of monomer A (STY).
STY is less reactive than MMA (Table 1), explaining this deviating
composition of the formed copolymer chains at high conversions.
It further follows from Fig. 9 (top vs. bottom subgures) that
diffusional limitations on termination also imply a more narrow
number CoCD per chain length, i.e., a more narrow number CCD is
expected. Indeed, Fig. 10 (top vs. bottom subgures) shows that,
due to diffusional limitation on termination, the polymer chains
possess a more similar composition as longer chains are obtained.
In other words, in FRP copolymerization a similar trend is
observed as predicted by the analytical Stockmayer equation
derived for coordination copolymerization (Stockmayer, 1945).
From this equation it can be derived that more narrow number
CoCDs result for higher average chain lengths. In Fig. S1 in
Section 2 of the Supporting Information, it is illustrated that, in
case the conventional initiator loading is reduced with a factor 10,
i.e., higher averages chain lengths are targeted, an even more

narrow number CoCD results as higher chain lengths are obtained,

further conrming this statement.
It should be noted that for lower initiator loadings the maximum chain length can become too high (4 20,000) to capture
explicitly the CoC-CLD also for the highest chain lengths using the
currently employed computer resources (16 GB RAM). However,
simulations revealed that the number fraction of these species is
always low (o 0.05) and reliable simulation results can thus still
be obtained. For example, for 0.01 mol% of conventional radical
initiator with respect to monomer, the contribution of high chain
lengths ( 415,000) in the CoC-CLD, as shown in Fig. S2 in the
Supporting Information, is still low, allowing an accurate
representation of this bivariate distribution and its marginal
distributions. It should however be stressed that for simulations
characterized by a very strong gel-effect leading it can be expected
that the contribution of the long macrospecies becomes too high
and thus accurate simulations can only be obtained in case the
computer resources are improved.
3.2.2. ATRP of STY and MMA
3.2.2.1. Conversion and control over chain length. For the isothermal
(80 1C) ATRP of styrene and MMA, using equimolar initial amounts
of both comonomers and 1 mol% of ATRP initiator and activator
with respect to monomer, Fig. 11(a) shows the simulated
conversion prole with (full black line) and without (dashed
blue line), accounting for diffusional limitations on termination,
propagation and activation/deactivation. No deactivator is initially
present and for simplicity it is assumed that no precipitation of the
(de)activator occurs along the ATRP.
As for FRP (Fig. 5(a)), from low conversion onwards, a rate
acceleration takes place due to diffusional limitations on termination, in agreement with the signicant drop of the corresponding population weighted apparent termination rate coefcient
(Fig. 12(a) for termination by recombination involving STY terminal units). Furthermore, at higher conversion, it follows from
Fig. 12(c) and (d) that (de)activation becomes diffusion controlled.
At a conversion of circa 0.80, both the activation and deactivation

P.H.M. Van Steenberge et al. / Chemical Engineering Science 110 (2014) 185199

195

Fig. 9. Number copolymer composition-chain length distribution (CoC-CLD) at a monomer conversion of 0.10, 0.60 and 0.90 in the free radical polymerization (FRP) of
styrene (STY) and methyl methacrylate (MMA); 80 1C; [S]0 [MMA]0 4.5 mol L  1; [I2]0:([S]0 [MMA]0) 0.005: top row: diffusional limitations on termination, propagation
and conventional radical initiation accounted for; bottom row: intrinsic model; initial number of monomers 2  109.

Fig. 10. Number copolymer composition distribution (CoCD) at a monomer conversion of 0.10, 0.60 and 0.90 in the free radical polymerization (FRP) of styrene (STY) and
methyl methacrylate (MMA); 80 1C; [S]0 [MMA]0 4.5 mol L  1; [I2]0:([S]0 [MMA]0) 0.005; top row: diffusional limitations on termination, propagation and conventional
radical initiation accounted for; bottom row: intrinsic model; for termination shown for recombination involving only STY terminal monomer units; initial number of
monomers 2  109.

apparent rate coefcient become almost zero and the polymerization stops. Since no high conversions are reached, no cage-effect is
obtained, as can be seen in Fig. 12(b), in which a constant apparent
propagation rate coefcient is depicted as a function of conversion.
Fig. 11(b) and (d) shows that the control over the polymer
properties improves due to diffusional limitations on termination.
In an ideal ATRP, each ATRP initiator molecule eventually leads
to one dormant macrospecies possessing the same chain length.

Due to diffusional limitations on termination less dead polymer molecules are formed and thus a higher EGF is obtained
(Delgadillo-Velazquez et al., 2002; D'hooge et al., 2013; JohnstonHall and Monteiro, 2010). In Fig. 11(d), EGF is dened as the
fraction of polymer molecules being dormant instead of dead
(Scheme 2). It can be seen that values close to 1 are obtained in
case the full kinetic model is used, i.e., less termination reaction
events take place. Furthermore, the PDI is closer to 1, i.e., the CLD

196

P.H.M. Van Steenberge et al. / Chemical Engineering Science 110 (2014) 185199

Fig. 11. Conversion prole and number average chain length (xn), polydispersity index (PDI), end-group functionality (EGF) as a function of conversion in the atom transfer
radical polymerization (ATRP) of styrene (STY) and methyl methacrylate (MMA); 80 1C; [STY]0 [MMA]0 4.5 mol L  1; [R0X]0:([STY]0 [MMA]0):[Activator]0 0.01:0.01; full
blue lines: diffusional limitations accounted for; blue dashed lines: intrinsic model; [Deactivator]0 0; EGF [RX]:([P] [RX]) (Scheme 1); initial number of monomers 107
(For interpretation of the references to color in this gure legend, the reader is referred to the web version of this article).

Fig. 12. Population weighted apparent termination rate coefcients and apparent propagation, activation and deactivation coefcient as a function of conversion in the atom
transfer radical polymerization (ATRP) of styrene (STY) and methyl methacrylate (MMA); 80 1C; [STY]0 [MMA]0 4.5 mol L  1; [R0X]0:([STY]0 [MMA]0):[Activator]0
0.01:0.01; full blue lines: diffusional limitations accounted for; blue dashed lines: intrinsic model; [Deactivator]0 0; Initial number of monomers 107 (For interpretation of
the references to color in this gure legend, the reader is referred to the web version of this article).

is more narrow (Fig. 13; conversion 0.20, 0.50 and 0.80). Due
to diffusional limitation on termination, mainly dormant species
are obtained, which all take up concurrently monomer units

diminishing the chain length gap that is obtained at low conversion due to non-instantaneous ATRP initiation (Fig. S3 in Section 2
of the Supporting Information).

P.H.M. Van Steenberge et al. / Chemical Engineering Science 110 (2014) 185199

197

Fig. 13. Number chain length distribution (CLD) at a monomer conversion of 0.20, 0.50 and 0.80 in the atom transfer radical polymerization (ATRP) of styrene (STY) and
methyl methacrylate (MMA); 353 K; [STY]0 [MMA]0 4.5 mol L  1); 80 1C; [STY]0 [MMA]0; [R0X]0:([S]0 [MMA]0):[Activator]0 0.01:1:0.01; full black lines: diffusional
limitations accounted for; full blue lines: intrinsic model; [Deactivator]0 0; Initial number of monomers 108 (For interpretation of the references to color in this gure
legend, the reader is referred to the web version of this article).

Fig. 14. Number copolymer composition-chain length distribution (CoC-CLD) at a monomer conversion of 0.20, 0.50 and 0.80 in the atom transfer radical polymerization
(ATRP) of styrene (STY) and methyl methacrylate (MMA); 80 1C; [STY]0 [MMA]0 4.5 mol L  1; [R0X]0:([S]0 [MMA]0):[Activator]0 0.01:1:0.01; top row: diffusional
limitations on termination and (de)activation accounted for; bottom row: intrinsic model; initial number of monomers 107.

3.2.2.2. Evolution of the CoCD-CLD with conversion. Fig. 14 shows

the corresponding number CoC-CLDs for Fig. 13. Again the top
row corresponds with the full kinetic model, i.e., accounting

for diffusional limitations on termination, propagation and

activation/deactivation, whereas the bottom row relates to the
intrinsic model. In contrast to the FRP case, in ATRP both

198

P.H.M. Van Steenberge et al. / Chemical Engineering Science 110 (2014) 185199

models lead to a dynamic evolution of the CoC-CLD with

conversion.
However, as explained above, less termination products are
formed due to diffusional limitations on termination, leading to
different number CoC-CLDs per conversion studied (top vs. bottom
subgure). It follows that, for the intrinsic model, a broader
number CoC-CLD results as in the low chain length range dead
polymer molecules can be easily identied. In other words, due to
diffusional limitations on termination, better shaped CoC-CLDs
result, reecting the inherent capability of CRP processes to obtain
well-tailored polymer microstructures.

4. Conclusions
An improved kinetic Monte Carlo (kMC) approach is presented based on composite binary trees to calculate chemical
composition-chain length distributions (CC-CLDs), such as the
bivariate copolymer composition-CLD (CoC-CLD). For each chain
length, which is linked with a leaf node of a main tree, additional
structural characteristics are stored and updated in sub-trees
connected to this leaf node.
The approach has been successfully illustrated for the calculation of the bivariate copolymer composition-CLD (CoC-CLD) for
free radical and atom transfer radical copolymerization of STY and
MMA, as the reduction of the number of kMC operations is the
highest in case the CoC is tracked as second variate. It is clear that
the number CoC-CLD is not constant in both polymerization
processes. In the FRP case, a narrowing of the number CoCD
results as higher chain lengths are obtained due to diffusional
limitation on termination. However, at high conversion compositional drift can be obtained due the diffusional limitation on
propagation and conventional radical initiation. In ATRP, diffusional limitations on termination allow to obtain a more narrow
number CoC-CLD as the formation of dead polymer molecules in
the low chain length range is reduced.
The approach could also be applied to trivariate distributions
and to other polymerization techniques provided that the intrinsic
kinetic and diffusion parameters are known. In addition, the
developed modeling tool will allow to control the evolution of
the CoC-CLD with respect to the targeted polymer microstructure.
The improved calculation of trivariate distributions and the
development of model based controlling strategies are envisaged
in future research work starting from available experimental data.

Acknowledgments
The authors acknowledge nancial support from the Long Term
Structural Methusalem Funding by the Flemish Government, the
Interuniversity Attraction Poles ProgrammeBelgian StateBelgian
Science Policy, and the Fund for Scientic Research Flanders (FWO).

Appendix A. Supporting information

Supplementary data associated with this article can be found in
the online version at http://dx.doi.org/10.1016/j.ces.2014.01.019.
References
Achilias, D.S., Kiparissides, C., 1992. Development of a general mathematical
framework for modeling diffusion-controlled free-radical polymerization reactions. Macromolecules 25, 3739.
Achilias, D., 2007. A review of modelling of diffusion controlled polymerization
reactions. Macromol. Theory Simul. 16, 319347.
Aggarwal, S.L., Sweeting, O.J., 1957. Polyethylene: preparation, structure and
properties. Chem. Rev. 57, 665742.

Ahmad, N.M., Charleux, B., Farcet, C., Ferguson, C.J., Gaynor, S.G., Hawkett, B.S.,
Heatley, F., Klumperman, B., Konkolewicz, D., Lovell, P.A., Matyjaszewski, K.,
Venkatesh, R., 2009. Chain transfer to polymer and branching in controlled
radical polymerizations of n-butyl acrylate. Macromol. Rapid Commun. 30,
20022021.
Al-Harthi, M., Khan, M.J., Abbasi, S.H., Soares, J.B.P., 2009. Macromol. React. Eng. 3,
148.
Asua, J., 2007. In: Asua, J.M. (Ed.), Polymer Reaction Engineering. Blackwell Publishing,
Oxford
Barner-Kowollik, C., Russell, G.T., 2009. Chain-length-dependent termination in
radical polymerization: subtle revolution in tackling a long-standing challenge.
Prog. Polym. Sci. 34, 12111259.
Baumgaertel, A., Altuntas, E., Schubert, U.S., 2012. Recent developments in the
detailed characterization of polymers by multidimensional chromatography. J.
Chromatogr. A 1240, 120.
Benkowski, J., Fredrickson, G., Kramer, E., 2001. American Physical Society, Annual
March Meeting, March 1216, 2001, Washington State Convention Center
Seattle, Washington Meeting ID: MAR01, abstract #C19.004.
Bentein, L., Dhooge, D.R., Reyniers, M.-F., Marin, G.B., 2011. Kinetic modeling as a
tool to understand and improve the nitroxide mediated polymerization of
styrene. Macromol. Theory Simul. 20, 238265.
Bentein, L., Dhooge, D.R., Reyniers, M.-F., Marin, G.B., 2012. Kinetic modeling of
miniemulsion nitroxide mediated polymerization of styrene: effect of particle
diameter and nitroxide partitioning up to high conversion. Polymer 53,
681693.
Beuermann, S., Buback, M., Davis, T.P., Gilbert, R.G., Hutchinson, R.A., Olaj, O.F.,
Russell, G.T., Schweer, J., van Herk, A.M., 1997. Critically evaluated rate
coefcients for free-radical polymerization, 2. Propagation rate coefcients
for methyl methacrylate. Macromol. Chem. Phys. 198, 15451560.
Brandup, J., Immergut, E.H., Grulke, E.A., Abe, A., Bloch, D.R. (Eds.), 1999. Polymer
Handbook, 4th edition John Wiley & Sons, New York
Buback, M., Gilbert, R.G., Hutchinson, R.A., Klumperman, B., Kutcha, F.-D., Manders, B.G.,
ODriscoll, K.F., Russell, G.T., Schweer, J., 1995. Critically evaluated rate coefcients
for free-radical polymerzation, 1. Propagation rate coefcients for styrene. Macromol. Chem. Phys. 196, 32673280.
Chaffey-Millar, H., Stewart, D., Chakravarty, M.M.T., Keller, G., Barner-Kowollik, C.,
2007. A parallelised high performance Monte Carlo simulation approach for
complex polymerisation kinetics. Macromol. Theory Simul. 16 (6), 575592.
Cools, P.J.C.H., Maesen, F., Klumperman, B., van Herk, A.M., German, A.L., 1996.
Determination of the chemical composition distribution of styrene and butadiene by gradient polymer elution chromatography (GPEC). J. Chromatogr. A
736, 125.
Delgadillo-Velazquez, O., Vivaldo-Lima, E., Quintero-Ortega, I.A., Zhu, S., 2002.
Effects of diffusion-controlled reactions on atom-transfer radical polymerization. AIChE J. 48, 25972608.
De Roo, T., Wieme, J., Heyndericks, G., Marin, G.B., 2005. Estimation of intrinsic rate
coefcients in vinyl chloride suspension polymerization. Polymer 46,
83408354.
Dhooge, D.R., Reyniers, M.-F., Marin, G.B., 2009. Methodology for kinetic modeling
of atom transfer radical polymerization. Macromol. React. Eng. 3, 185209.
Dhooge, D.R., Reyniers, M.-F., Stadler, F.J., Dervaux, B., Bailly, C., Du Prez, F.E., Marin,
G.B., 2010. Atom transfer radical polymerization of isobornyl acrylate: a kinetic
modeling study. Macromolecules 43, 87668781.
Dhooge, D.R., Konkolewicz, K., Reyniers, M.-F., Marin, G.B., Matyjaszewski, K., 2012.
Tuning polymer properties by competitive equilibria. Macromol. Theory Simul.
21, 5269.
Dhooge, D.R., Reyniers, M.-F., Marin, G.B., 2013. The crucial role of diffusional
limitations in controlled radical polymerization. Macromol. React. Eng. 7,
362379.
Dompazis, G., Kanellopoulos, V., Kiparissides, C., 2005. A multi-scale modeling
approach for the prediction of molecular and morphological properties in
multi-site catalyst, olen polymerization reactors. Macromol. Mater. Eng. 290,
525536.
Doremaele, G.H.J., van Herk, A.M., German, A.L., 1992. Modelling of emulsion
copolymer microstructure. Polym. Int. 27, 95108.
Dube, M., Soares, J.B.P., Penlidis, A., Hamielec, A.E., 1997. Mathematical modelling of
multicomponent chain-growth polymerization in batch, semi-batch and continuous reactors: a review. Ind. Eng. Chem. Res. 36, 9661015.
Gillespie, D.T., 1977. Exact stochastic simulation of coupled chemical reactions. J.
Phys. Chem. 81 (25), 23402361.
Gilbert, R.G., 1992. Pure Appl. Chem. 64, 1563.
Ginzburg, A., Macko, T., Dolle, V., Bruell, R., 2013. Multidimensional hightemperature liquid chromatography: a new technique to characterize the
chemical heterogeneity of ZieglerNatta-based bimodal HDPE. J. Appl. Polym.
Sci. 129 (4), 18971906.
Hamzehlou, S., Reyes, Y., Leiza, J.R., 2012. Detailed microstructure investigation of
acrylate/methacrylate functional copolymers by kinetic Monte Carlo simulation. Macromol. React. Eng. 6 (8), 319329.
Heberger, K., Fischer, H., 1993. Rate constants for the addition of the 2-cyano-2propyl radical to alkenes in solution. Int. J. Chem. Kinet. 25, 249263.
Johnston-Hall, G., Monteiro, M.J., 2008. Bimolecular radical termination: new
perspectives and insights. J. Polym. Sci. Part A: Polym. Chem. 46, 31553173.
Johnston-Hall, G., Monteiro, M.J., 2010. Kinetic simulations of atom transfer radical
polymerization (ATRP) in light of chain length dependent termination. Macromol. Theory Simul. Chem. 19 (7), 387393.

P.H.M. Van Steenberge et al. / Chemical Engineering Science 110 (2014) 185199

Junkers, T., Barner-Kowollik, C., 2008. The role of mid-chain radicals in acrylate free
radical polymerization: branching and scission. J. Polym. Sci. Part A: Polym.
Chem. 46, 75857605.
Kiparissides, C., 2006. Challenges in particulate polymerization reactor modeling
and optimization: a population balance perspective. J. Process Control 16,
205224.
Kiparissides, C., Krallis, A., Meimaroglou, D., Pladis, P., Baltsas, A., 2010. From
molecular to plant-scale modeling of polymerization processes: a digital highpressure low-density polyethylene production paradigm. Chem. Eng. Technol.
33, 17541766.
Konkolewicz, D., Sosnowski, S., D'hooge, D.R., Szymanski, R., Reyniers, M.-F., Marin,
G.B., Matyjaszewski, K., 2011. Origin of the difference between branching in
acrylates polymerization under controlled and free radical conditions: a
computational study of competitive processes. Macromolecules 44, 83618373.
Krallis, A., Meimaroglou, D., Kiparissides, C., 2008. Dynamic prediction of the
bivariate molecular weightcopolymer composition distribution using
sectional-grid and stochastic numerical methods. Chem. Eng. Sci. 63,
43424360.
Krallis, A., Pladis, P., Kanellopoulos, V., Kiparissides, C., 2010. Development of
advanced computer-aided software tools for design, simulation and optimization of polymerization processes. Macromol. React. Eng. 4, 303308.
Kurdikar, D.L., Peppas, N.A., 1994. Method of determination of initiator efciency:
application to UV-polymerizations using 2,2-dimethoxy-2-phenyl acetophenone. Macromolecules 27, 733738.
Liu, S., Cao, K., Yao, Z., Bogeng, B., Zhu, S., 2009. Model development for
semicontinuous production of ethylene and norbornene copolymers having
uniform composition. AIChE J. 55, 663674.
Matyjaszewski, K., Xia, J., 2001. Atom transfer radical polymerization. Chem. Rev.
101, 29212990.
Matyjaszewski, K., Davis, P., 2002. Handbook of Radical Polymerization. WileyInterscience, Hoboken.
Meimaroglou, D., Krallis, A., Saliakas, V., Kiparissides, C., 2007. Prediction of the
bivariate molecular weightlong chain branching distribution in highly
branched polymerization systems using Monte Carlo and sectional grid
methods. Macromolecules 40, 22242234.
Moad, G., Solomon, D.H., 1995. The Chemistry of Free Radical Poylmerization.
Elsevier Science Ltd., Oxford
Narkchamnan, K., Anantawaraskul, S., Soares, J.B.P., 2011. Bimodality criterion for
the chemical composition distribution of ethylene/1-olen copolymers: theoretical development and experimental validation. Macromol. React. Eng. 5,
198210.
Payne, K.A., Dhooge, D.R., Van Steenberge, P.H.M., Reyniers, M.-F., Cunningham, M.F.,
Hutchinson, R.A., Marin, G.B., 2013. ARGET ATRP of butyl methacrylate: utilizing
kinetic modeling to understand experimental trends. Macromolecules 46,
38283840.
Richards, J.R., Congalidis, J.P., 2006. Measurement and control of polymerization
reactors. Comput. Chem. Eng. 30, 14471463.
Schiers, J., Priddy, D., 2003. Modern Styrenic Polymers: Polystyrenes and Styrenic
Copolymers.
Smoluchowski, M., 1917. Attempt for a mathematical theory of kinetic coagulation
of colloid solutions. Z. Phys. Chem. 92, 129168.
Soares, J.B.P., 2004. In: Kroschwitz, J.I. (Ed.), Encyclopedia of Polymer Science and
Technology. Wiley-VCH, Weinheim.
Sperling, L.H., 1986. Introduction to Physical Polymer Science.
Stockmayer, W.H., 1945. Distribution of chain lengths and compositions in
copolymers. J. Chem. Phys. 13 (1), 199207.

199

Szymanski, R., 2009. On the determination of the ratios of the propagation rate
constants on the basis of the MWD of copolymer chains: a new Monte Carlo
algorithm. e-Polymer (art no. 044).
Szymanski, R., Sosnowski, S., 2012. Kinetic Monte Carlo studies on the importance
of the reaction scheme in segmental exchange of copolymer chains. Macromol.
Theory Simul. 21, 411427.
Tacx, J.C.J., Ammerdorffer, J.L., German, A.L., 1988. Chemical-composition distribution of styrene ethyl methacrylate copolymers studied by means of TLC FID
effect of high conversion in various polymerization processes. Polymer 29,
20872094.
Tobita, H., Hatanaka, K., 1996. Branched structure formation in free radical
polymerization of vinyl acetate. J. Polym. Sci. Part B: Polym. Phys. 34, 671681.
Toloza Porras, C., Dhooge, D.R., Van Steenberge, P.H.M., Reyniers, M.-F., Marin, G.B.,
2013. A theoretical exploration of the potential of ICAR ATRP for one- and twopot synthesis of well-dened diblock copolymers. Macromol. React. Eng. 7,
311326.
Trommsdorff, E., Khle, H., Lagally, P., 1948. Polymerization of methyl methacrylates. Makromol. Chem. 1, 169198.
Van Steenberge, P.H.M., Vandenbergh, J., Dhooge, D.R., Reyniers, M.-F., Adriaensens,
P.J., Lutsen, L., Vanderzande, D.J.M., Marin, G.B., 2011. Kinetic Monte Carlo
modeling of the sulnyl precursor route for poly(para-phenylene vinylene)
synthesis. Macromolecules 44, 87168726.
Van Steenberge, P.H.M., Dhooge, D.R., Wang, Y., Zhong, M., Reyniers, M.-F.,
Konkolewicz, D., Matyjaszewski, K., Marin, G.B., 2012. Linear gradient quality
of ATRP copolymers. Macromolecules 45, 8519.
Van Steenberge, P.H.M., Dhooge, D.R., Vandenbergh, J., Reyniers, M.-F., Adriaensens,
P.J., Vanderzande, D.J.M., Marin, G.B., 2013. Kinetic Monte Carlo modeling of the
sulnyl precursor route for poly(para-phenylene vinylene) synthesis. Macromol. Theory Simul. 22, 246255.
Vinu, R., Levine, S.E., Wang, L., Broadbelt, L.J., 2012. Detailed mechanistic modeling
of poly(styrene peroxide) pyrolysis using kinetic Monte Carlo simulation. Chem.
Eng. Sci. 69, 456471.
Vrentas, J.S., Duda, J.L., 1977a. Diffusion in polymersolvent systems. I. Reexamination of the free-volume theory. J. Polym. Sci. Part B: Polym. Phys. 15, 403416.
Vrentas, J.S., Duda, J.L., 1997b. Diffusion in polymersolvent systems. II. A predictive
theory for the dependence of diffusion coefcients on temperature, concentration, and molecular weight. J. Polym. Sci. Part B: Polym. Phys. 15, 417439.
Vrentas, J.S., Duda, J.L., Ling, H.-C., 1984. Self-diffusion in polymer-solvent-solvent
systems. J. Polym. Sci. Part B: Polym. Phys. 22, 459469.
Vrentas, J.S., Vrentas, C.M., 2003. Evaluation of the free-volume theory of diffusion.
J. Polym. Sci. Part B: Polym. Phys. 41, 501507.
Wang, W., Hutchinson, R.A., Cunningham, M.F., 2005. A semi-batch process for
nitroxide mediated radical polymerization. Macromol. Mater. Eng. 290,
230241.
Wang, L., Broadbelt, L.J., 2010. Kinetics of segment formation in nitroxide-mediated
controlled radical polymerization: comparison with classic theory. Macromolecules 43 (5), 22282235.
Weidner, S.M., Falkenhagen, J., Bressler, I., 2012. Copolymer composition determined by LCMALDI-TOF MS coupling and MassChrom2D data analysis.
Macromol. Chem. Phys. 213 (22, Special Issue), S2404S2411.
White, J.L., Dharod, K.C., Clark, E.S., 1974. Interaction of melt spinning and drawing
variables on the crystalline morphology and mechanical properties of highdensity and low-density polyethylene ber. J. Appl. Polym. Sci. 18, 25392568.
Wulkow, M., 2009. Computer aided modeling of polymer reaction engineeringthe
status of Predici. ISimulation. Macromol. React. Eng. 2, 461494.
Young, R.J., Lovell, P.A., 1991. Introduction to Polymers. CRC Press., Boca Raton, FL.