Home Search Collections Journals About Contact us My IOPscience

Systematic coarse-graining of the dynamics of entangled polymer melts: the road from

chemistry to rheology

This content has been downloaded from IOPscience. Please scroll down to see the full text.

2011 J. Phys.: Condens. Matter 23 233101

(http://iopscience.iop.org/0953-8984/23/23/233101)

View the table of contents for this issue, or go to the journal homepage for more

Download details:

IP Address: 131.238.52.228
This content was downloaded on 25/04/2017 at 21:59

Please note that terms and conditions apply.

You may also be interested in:

Computer simulations of amorphous polymers

Wolfgang Paul and Grant D Smith

Computer simulations of supercooled polymer melts in the bulk and in confinedgeometry

J Baschnagel and F Varnik

Theoretical models for bridging timescales in polymer dynamics

M G Guenza

Entanglement effects in defect-free model polymer networks

Ralf Everaers

Comparison of Various Correlation Times in Polymer Melts by Molecular Dynamics Simulation

E M Pestryaev

Fracture in glassy polymers: a molecular modeling perspective

Jörg Rottler

Glass transition of polymers in bulk, confined geometries, and near interfaces

Simone Napolitano, Emmanouil Glynos and Nicholas B Tito

Molecular dynamics simulations

Kurt Binder, Jürgen Horbach, Walter Kob et al.

Primitive-path statistics of entangled polymers: mapping multi-chain simulations onto single-chain

mean-field models
Rudi J A Steenbakkers, Christos Tzoumanekas, Ying Li et al.
IOP PUBLISHING JOURNAL OF PHYSICS: CONDENSED MATTER
J. Phys.: Condens. Matter 23 (2011) 233101 (17pp) doi:10.1088/0953-8984/23/23/233101

TOPICAL REVIEW

Systematic coarse-graining of the

dynamics of entangled polymer melts: the
road from chemistry to rheology
J T Padding1,2 and W J Briels2
1
Institut de la Matière Condensée et des Nanosciences, Université Catholique de Louvain,
Croix du Sud 1, B-1348 Louvain-la-Neuve, Belgium
2
Computational Biophysics, University of Twente, PO Box 217, NL-7500 AE Enschede,
The Netherlands

E-mail: [email protected] and [email protected]

Received 11 February 2011, in final form 29 March 2011

Published 25 May 2011
Online at stacks.iop.org/JPhysCM/23/233101

Abstract
For optimal processing and design of entangled polymeric materials it is important to establish
a rigorous link between the detailed molecular composition of the polymer and the viscoelastic
properties of the macroscopic melt. We review current and past computer simulation techniques
and critically assess their ability to provide such a link between chemistry and rheology. We
distinguish between two classes of coarse-graining levels, which we term coarse-grained
molecular dynamics (CGMD) and coarse-grained stochastic dynamics (CGSD). In CGMD the
coarse-grained beads are still relatively hard, thus automatically preventing bond crossing. This
also implies an upper limit on the number of atoms that can be lumped together (up to five
backbone carbon atoms) and therefore on the longest chain lengths that can be studied. To reach
a higher degree of coarse-graining, in CGSD many more atoms are lumped together (more than
ten backbone carbon atoms), leading to relatively soft beads. In that case friction and stochastic
forces dominate the interactions, and action must be undertaken to prevent bond crossing. We
also review alternative methods that make use of the tube model of polymer dynamics, by
obtaining the entanglement characteristics through a primitive path analysis and by simulation
of a primitive chain network. We finally review super-coarse-grained methods in which an
entire polymer is represented by a single particle, and comment on ways to include memory
effects and transient forces.

(Some figures in this article are in colour only in the electronic version)

Contents 3.1. Theory of coarse-graining 4

3.2. Classification of coarse-grained dynamics
1. Introduction 2 methods 5
2. Atomistically detailed models 3 4. Lumping a few atoms into beads 5
2.1. Possibilities and limitations 3
4.1. Targeting the structure of real polymer chains 5
2.2. Equilibration schemes for atomistically de-
tailed models 3 4.2. Lattice models 7
2.3. Examples of atomistic simulations of dynamic 4.3. Backmapping: reinserting atomistic details 7
properties 3 4.4. Dynamic properties from coarse-grained molec-
3. Coarse-graining: lumping together the atoms 4 ular dynamics: rescaling of time 7

0953-8984/11/233101+17$33.00 1 © 2011 IOP Publishing Ltd Printed in the UK & the USA
J. Phys.: Condens. Matter 23 (2011) 233101 Topical Review

5. Lumping many atoms into blobs: the role of friction

and uncrossability 8
5.1. Friction and stochastic forces 8
5.2. Uncrossability of bonds and entanglement 9
6. Primitive paths 10
6.1. Primitive path analysis 10
6.2. Primitive chain networks and slip-link models 12
7. Super-coarse-graining: a polymer as a single soft
particle 13
8. Conclusion 14
Acknowledgments 15
References 15

1. Introduction

Ever since the first synthetic polymers were made by

Staudinger and their molecular weights were measured by
Debye and Bueche, polymer systems have received continual
interest from both theorists and experimentalists. Most
processing of polymers takes place in the melt state, when
they are very viscous and have surprising (temporary) elastic
properties. For optimal processing, design and application
of polymeric melts it is important to establish a rigorous
link between their molecular composition and macroscopic
mechanical properties, i.e. between polymer chemistry and
rheology. Figure 1. The dynamics of polymers can be simulated by different
Establishing such a link is by no means an easy task methods which differ in their level of detail describing polymer
and has, despite considerable progress, not yet been fully configurations and entanglement effects. Top: atomistic molecular
achieved. Unlike simple molecules, interactions within and dynamics simulations are usually sufficiently detailed to faithfully
between polymer chains are characterized by a large range predict the dynamic properties of polymers of a specified chemistry.
Second from top: small groups of atoms are lumped together (up to
of different time and length scales. The dynamics of five backbone carbon atoms) into coarse-grained entities; the
atomic bond vibrations are characterized by Ångström and interaction between these ‘beads’ is still hard enough to make bond
sub-picosecond scales, whereas the dynamics of statistically crossing energetically unfavourable. Middle: many more atoms are
independent Kuhn segments are characterized by nanometre lumped together (more than ten backbone carbon atoms) into
and tens of picosecond scales [1]. Beyond this, the coarse-grained entities; the interactions between these ‘blobs’ are so
soft that uncrossability needs to be enforced by additional
connectivity and mutual uncrossability of the segments cause constraints. Second from bottom: polymer chains are forced to
an interdependence between features on a hierarchy of scales. diffuse along primitive paths, representing the temporary topological
The full chain, finally, is characterized by its radius of network which arises because of uncrossability of chains. Bottom:
gyration of 10–100 nm. This may not appear as a very large each polymer chain is represented by just one particle; the
increase from the Kuhn segment scale, but the associated entanglement effect is captured through addition of slow variables
whose deviations from equilibrium lead to transient forces.
timescales increase dramatically, to milliseconds, seconds or
even longer.
The enormously long intrinsic timescales are often
rationalized by viewing polymer systems as temporary rubbery In this review we focus on computer simulation techniques
networks. Such a network arises as a result of mutual that may help in establishing the link between chemistry and
uncrossability of the polymer chains—they are entangled. large-scale dynamics and rheology. Computer simulations
With the advent of the reptation theory of de Gennes, Doi can check the assumptions made in theoretical models, and
and Edwards [1], a new concept was introduced in the theory either accept or reject them [2]. Ideally, one would like to
of polymer dynamics. In reptation theory each polymer use atomistically detailed models because they can accurately
is supposed to move in a tube around a Gaussian path in capture the subtleties in the interactions between the polymers.
space. The tube only serves one purpose, namely to roughly However, because of the intrinsically slow dynamics, a direct
represent the uncrossability of the surrounding chains and to prediction of the large-scale dynamics and rheology from
turn a difficult multi-chain problem into a one-chain problem. such detailed models is computationally unattainable (except
The tube is clearly an idealized mean-field concept, and for the case of unentangled or slightly entangled polymers).
many extensions of the original model have been postulated Traversing the road from chemistry to rheology is usually a
to explain experimental observations; the difficulty is that bumpy ride: the only way to reach the larger scales is by going
without detailed information such postulates often remain through a succession of coarse-graining steps. Different levels
unchecked [2]. of coarse-graining are schematically indicated in figure 1.

2
J. Phys.: Condens. Matter 23 (2011) 233101 Topical Review

A number of excellent reviews on coarse-graining of unentangled polymers and as n 3.4 for entangled polymers [1].
entangled polymers have already appeared in the literature. To measure long-time correlation functions, the simulations
Some of these reviews focus mainly on static properties [3–6], must be performed for at least as long as this timescale.
whereas others focus on the dynamic properties as well [7–9]. Sometimes one is interested in the static properties of a
This review provides an up-to-date overview of past and melt of long polymer chains, or in faster dynamical relaxation
current coarse-graining efforts, with a strong emphasis on our processes on a more local scale. In such cases, one still
own opinion about the correct way to proceed in the future. We needs to make sure that the atomistic conformations in the
therefore focus on methods to handle friction, uncrossability simulation box have sufficiently relaxed to faithfully represent
and memory, all of which are intimately connected to the long- an equilibrium state.
time dynamic properties and rheology.
This review is structured as follows. In section 2 we 2.2. Equilibration schemes for atomistically detailed models
review the work that has been done using atomistically detailed
models. In section 3 we introduce the theory of coarse- So how does one ensure that an atomistically detailed polymer
graining and divide coarse-grained dynamics methods into melt is sufficiently relaxed? A first reaction may be to start
two classes: (1) those with relatively low levels of coarse- with an ensemble of chains with a correct end-to-end distance
graining (up to five backbone carbon atoms), where friction is distribution, arrange them randomly in the simulation box and
not dominant or ignored and where uncrossability is included introduce excluded volume rapidly. However, Auhl et al [11]
automatically by the excluded-volume interactions between showed that this procedure leads to deformations on short
relatively hard beads, and (2) those with relatively high levels length scales, which relax only when the chains have moved
of coarse-graining (more than ten backbone carbon atoms), over their own size, i.e. after one longest intrinsic relaxation
where friction is dominant and measures need to be taken to time. The authors also showed how this local deformation may
ensure uncrossability. Section 4 reviews the work that has been be overcome by first pre-packing the Gaussian chains, reducing
the density fluctuations in the system, followed by a more
undertaken in the first class and section 5 reviews the work
gradual introduction of excluded volume. Another method is
done in the second class. In section 6 we review the attempts
to apply a double-bridging MC algorithm in which new bonds
that have been undertaken at characterizing primitive paths
are formed across a pair of chains, creating two new chains
and the use of primitive path models. In section 7 we focus
substantially different from the original two [11].
on the ultimate form of particulate coarse-graining, where an
The latter method is similar in spirit to the end-
entire polymer is represented by just one particle. Finally, we
bridging (EB) MC method, see figure 2, which has been
conclude in section 8.
developed extensively in the groups of Mavrantzas and
Theodorou [12–22]. A thorough analysis of the geometric
2. Atomistically detailed models formulation and numerical implementation of the original EB
algorithm is given in [12]. Using this method, equilibrated
samples of polyethylene (up to C500 ) [12] and cis-1,4-
2.1. Possibilities and limitations
polyisoprene (up to C200 ) [14, 15] have been generated. Later,
faster methods were developed with directed moves, dubbed
On a microscopic level, molecular dynamics (MD) and Monte
directed internal bridging (DIB) and directed end bridging
Carlo (MC) simulations can be performed, in which each atom
(DEB), which enabled them to equilibrate polyethylene
of a polymer chain is represented separately, see figure 1
samples as long as C6000 [16, 17]. Extensions of the EB
(top) [10]. The atoms are modelled as interacting particles
algorithm to different chain architectures have also been
and their positions are updated according to Newton’s laws
introduced [18].
(MD) or by certain trial moves which are accepted, depending
An exciting development, of relevance to the rheology of
on the statistical ensemble, with a certain probability (MC).
polymer melts, is the use of EB moves in combination with a
Accurate force fields have been constructed to cater for
small tensorial field to obtain well-equilibrated, pre-oriented,
anyone’s research interests, provided they exclude chemical
strained configurations [13, 20, 21]. This way, viscoelastic
reactions and other phenomena of a quantum mechanical
properties can be measured from atomistically detailed models,
nature. Bulk behaviour is simulated by applying periodic
although up to now only simulations with unentangled chains
boundary conditions to the simulation box. Typical MD
have been reported.
simulations cover the motion of a few tens of thousands In light of section 4.3 we already mention that atomistic
of atoms over a period of a few nanoseconds; on current conformations can also be equilibrated efficiently by mapping
computers, such a run would take a week to complete. them onto a coarse-grained model, equilibrating the coarse-
This already sets a limit to the length of polymers that can grained model and finally reinserting the atomistic details. Of
be simulated in atomistic detail. First, the√radius of gyration course, EB moves can also be applied in these coarse-grained
of the polymer chain typically grows as n , where n is the equilibration runs [23–25].
number of monomers per chain. A polymer should not interact
with itself via the periodic boundaries, which means that the
2.3. Examples of atomistic simulations of dynamic properties
volume of the box, and hence the number of particles, should
scale as n 3/2 . Second, the longest intrinsic relaxation time of Given the very steep increase of terminal relaxation times
a polymer chain in a melt scales very fast, usually as n 2 for with molecular weight, it should come as no surprise that

3
J. Phys.: Condens. Matter 23 (2011) 233101 Topical Review

Figure 2. Schematic of the EB move. Left: initial configuration. The red arrow indicates that atom i of chain ich will attack atom j of chain
j ch ; then, trimer ( j a, j b, j c) will be excised from the system. Right: chain configurations after the application of the move. Atoms i and j
are connected through the new trimer bridge ( j a  , j b , j c ). Reprinted from [19], copyright 2006, with permission from Elsevier.

atomistically detailed MD simulations of the dynamics of 3. Coarse-graining: lumping together the atoms
polymer melts have only been performed for relatively short
unentangled or slightly entangled chains. 3.1. Theory of coarse-graining
Polyethylene, because of its simplicity, has been one of In order to increase the time and length scales accessible
the most favourite polymers studied in the past. Starting in in the simulation of polymers, detailed atomistic models are
1995, Paul and co-workers studied the diffusion coefficient replaced by coarse-grained models in which each particle
of unentangled chains, C44 [26] and C100 [27], both by an represents a collection of atomic particles. In this review
explicit atom model and by a united atom model. In this united we focus on the bottom-up (ab initio) approach where the
atom model, which was optimized to reproduce experimental interactions between the coarse-grained particles are directly
P –V –T behaviour, each methyl CHx group is treated as one linked to the atomistic interactions and are aimed at correctly
particle. The viscosity of C100 was investigated by Moore reproducing the structural, thermodynamic and/or dynamical
et al [28]. Later, Harmandaris et al studied the diffusion properties [3, 7, 9, 10, 35].
coefficient of samples of C24 , C78 and C156 polyethylene which When coarse-graining polymers, the polymer chain is
had been previously relaxed by the EB algorithm [29]. The subdivided into subchains, partitioning the degrees of freedom
shear relaxation modulus G(t) of the C24 and C78 samples were into two sets: the coarse-grain coordinates and momenta
also obtained by analysing equilibrium stress fluctuations [13]. {R, P}, which are the centres of mass (or characteristic
Soon after, in 2001, our group reported the mean square atom) positions and momenta of the subchains, and the ‘bath’
displacement, shear relaxation modulus and viscosity of a coordinates and momenta {q, p}, which are the remaining
C120 polyethylene sample [30]. Finally, in 2003 mean square internal coordinates and momenta describing the details of the
displacements and diffusion coefficients for chain lengths C78 configurations. The potential of mean force governing the
to C250 were reported [31]. forces among the coarse-grained coordinates is given by
Other polymer species have been studied as well, notably 
1,4-polybutadiene. Whereas Smith and co-workers in 1999 A(R) = −kB T ln d p dq exp(−β HB ( p, q; R)), (1)
were still limited to 100 carbon atom simulations [32], in 2005
Tsolou and co-workers were able to study the self-diffusion where β = 1/(kB T ) and HB ( p, q; R) is the bath Hamiltonian,
and dynamic structure factor of polybutadienes with up to 400 equal to the sum of kinetic energy TB ( p, q) of the bath
carbon atoms [33]. variables and the potential energy (R, q) of the entire system.
Despite the fast growth in computing power, the extremely In a system containing only coarse-grain particles, employing
fast increase of the longest relaxation time with chain length, the potential of mean force A(R) ensures a correct distribution
n 3.4 , implies that atomistically detailed MD simulations will of R coordinates, as well as correct thermodynamic properties.
remain limited to only slightly entangled samples. Longer Unfortunately, the potential of mean force is generally a
chains can be, and have been, studied, for example Ryckaert complicated function of all coordinates R and invariably
studied a melt of fully atomistic C1000 chains [34], but this can includes complicated multi-body interactions. For practical
only be done for simulation times much shorter than the longest and computational reasons it is impossible to calculate and
relaxation time. store the potential of mean force for all possible multi-body

4
J. Phys.: Condens. Matter 23 (2011) 233101 Topical Review

configurations. Some approximations need to be made. The 3.2. Classification of coarse-grained dynamics methods
most widely used approximation is that the potential of mean
The term ‘coarse-grained dynamics’ is used rather loosely
force can be represented by pairwise and sometimes triplet
in the literature. Many different levels of coarse-graining
terms. We would like to issue a warning that a naive use
are encountered, with different approaches to treating the
of the interactions at the coarse-grained level can lead to
friction. In our view it is important to distinguish between two
incorrect values for some thermodynamic properties. More
main classes, which may be termed coarse-grained molecular
specifically, the thermodynamics of the coarse-grained system
dynamics and coarse-grained stochastic dynamics, see figure 1.
will not correspond to the thermodynamics of the microscopic
The difference is the following:
system if one falsely assumes that one is still dealing with an
atomistic system, where the coarse-grain particles are the only • Coarse-grained molecular dynamics (CGMD) applies to
particles present and the effective pair interactions are the only simulations where a few atoms (up to one chemical repeat
energy terms present. It should be intuitively clear that the unit or five backbone carbon atoms) are coarse-grained to
relatively softer interactions will then lead to a pressure which relatively hard beads. In this case the additional friction
is generally too small and a system which is too compressible. components are often not dominant, and in most cases
The source of this thermodynamic inconsistency can be traced are ignored altogether. Ignoring the friction leads to too
to the state point dependence as well as the pair approximation fast dynamics, which is beneficial for equilibration, but
of the potential of mean force. For a discussion of these requires a rescaling of time when quantitative comparison
matters, the reader is referred to [36–38]. with real dynamics and rheology is desired.
The theory underlying coarse-graining of dynamics is
• Coarse-grained stochastic dynamics (CGSD) applies to
simulations where many atoms (many chemical repeat
the Mori–Zwanzig formalism [10, 35, 39, 40], where a
units or more than ten backbone carbon atoms) are coarse-
projection operator technique leads to a generalized Langevin
grained to relatively soft beads or ‘blobs’. The friction
equation governing the dynamics of the coarse variables.
and stochastic forces dominate the interactions, and cannot
Although the dynamics of the atomistic model is deterministic
simply be ignored but must be included in the equations
and conservative, the dynamics of the generalized Langevin
of motion. Often it is assumed that the stochastic
equation is stochastic and includes dissipation and memory
forces are delta-correlated, i.e. have no memory. Unless
effects:
countermeasures are taken, the soft interactions cannot
d Pn ∂A prevent crossing of the bonds between the coarse-grained
(t) = − (t) + Fn,t
R
dt ∂ Rn particles, leading to unrealistic dynamics for entangled
 t polymers.
− dτ Pm (t − τ )ζmn (τ ; R(t − τ )). (2)
m 0 In the following sections we will treat both levels of coarse-
graining.
Here Pn is the momentum of coarse-grained particle n , A
R
is the above-mentioned potential of mean force, Fn,t is a
random force and ζmn is a time-dependent friction matrix. The 4. Lumping a few atoms into beads
Mori–Zwanzig formalism provides an exact derivation for this
friction matrix, but its interpretation is not straightforward: Let us first focus on CGMD simulations. A coarse-grained
ζmn (τ ; R(t − τ )) is the correlation, at the point where the particle may represent the centre of mass of a small group
of atoms, or the position of a characteristic atom which is
coordinates R equal R(t − τ ), of the random force at time
important in determining the local structure of the polymer
zero on particle m with the random force propagated by a
backbone or a side group. An example is given in figure 3
complicated operator on particle n . Specifically, this operator
where polystyrene chains are coarse-grained to ‘A’ and ‘B’
is exp{(1 − P )iLτ }, where L is the Liouville operator and P
beads representing a part of the backbone and the phenyl ring,
projects onto the coarse-grain coordinates [10]. If the theory
respectively [41].
is followed to the rule (including the exact potential of mean
force), a correct description of the structure and dynamics
follows automatically. However, as the above equation 4.1. Targeting the structure of real polymer chains
shows, the resulting friction on a particle again depends on Most coarse-graining efforts focus on reproducing the structure
all coordinates R and momenta P . Compared to the case of the underlying microscopic chain [3, 5–8, 42].
of the potential of mean force the situation is aggravated, The structure of a polymer melt can be described
because the friction depends also on the coordinates and in terms of distribution functions of distances and angles
momenta in the past. Clearly, for practical and computational between the coarse-grained particles. Once it is decided
reasons approximations need to be made. For example, it is what the position of a coarse-grained particle represents, it
often assumed that the friction is pairwise additive with the is rather straightforward to measure distribution functions
contribution of a pair depending only on the positions and for these coarse-grained coordinates in atomistically detailed
momenta of the two particles involved. Also memory effects MD simulations. Intramolecular distribution functions include
are often ignored. This may or may not be correct. Making the those for the distance between two adjacent coarse-grained
right approximations, often based on physical intuition, is the beads, the angle between three consecutive beads and the
responsibility of the coarse-grainer. dihedral angle between four consecutive beads. Other

5
J. Phys.: Condens. Matter 23 (2011) 233101 Topical Review

Figure 3. Coarse-graining of polystyrene to two types of beads ‘A’

and ‘B’ representing part of the backbone and a phenyl ring, Figure 4. Nonbonded A–A pair distribution function for a
respectively. Dashed lines show the bonds between the polystyrene melt (see figure 3) at T = 463 K, obtained from
coarse-grained beads A and B. Reprinted with permission from [41]. atomistic MD simulations (solid line) and coarse-grained simulation
Copyright 2009 American Chemical Society. (dashed line). Reproduced with permission from [44]. Copyright
2007 Wiley-VCH Verlag GmbH and Co. KGaA.

structural distribution functions include those for distances In some cases a predefined functional form is not
between beads on different chains, and can also include those flexible enough to capture the intricacies of the target
for distances between beads on the same chain but sufficiently radial distribution function. More complicated analytical
far away along the backbone. These distribution functions are potentials, defined piecewise for different ranges, allow
used as target functions for the coarse-grained interactions. more accurate reproduction of the complicated distribution
Usually, the local intramolecular interactions between functions [46–48]. The parameters of these models may
consecutive beads are relatively strong and therefore rather be obtained semi-automatically or fully automatically using
unperturbed by intermolecular interactions. In such a case a simplex optimization [5, 49–52].
simple Boltzmann inversion is enough to obtain the coarse- Another popular technique to obtain the coarse-grained in-
grained interaction [5]. For example, from a distribution teractions is iterative Boltzmann inversion (IBI) [5, 47, 53, 54].
Pbond (r ) of bond lengths r between consecutive coarse-grained Focusing on the nonbonded interactions, for every iteration a
particles (obtained in an atomistic simulation and already one-to-one correspondence between the effects at a distance
corrected for the Jacobian 4πr 2 ), the coarse-grained bond r0 and the nonbonded potential V (r0 ) at the same distance
potential is derived as is assumed. In the beginning, a starting potential V0 (r ) has
to be guessed, which is usually the Boltzmann inversion of
Vbond (r ) = −kB T ln Pbond (r ). (3)
the target distribution function. At each iteration stage n ,
One word of caution is in place here. Factorizing the various a simulation is performed with the last guess Vn (r ) of the
distribution functions into independent parts representing nonbonded potential, and the radial distribution function gn (r )
is measured. This leads to a new guess for the nonbonded
bond distances, bond angles and dihedral angles ignores any
potential:
correlations which may be present between them [43, 44].
The nonbonded interaction, represented by the radial gn (r )
distribution function g(r ) between nonbonded coarse-grained Vn+1 (r ) = Vn (r ) + kB T ln . (4)
gtarget (r )
beads, is, however, much more subtle. It contains both
entropic and enthalpic contributions and a simple Boltzmann The iterations continue until the difference between gn (r ) and
inversion is insufficient because this ignores important packing gtarget is deemed satisfactory. Usually only a few iterations are
effects [4, 5, 42, 45]. Different methods have been proposed necessary [5, 47].
to find the nonbonded interaction which reproduces the target Other techniques have been used as well. Within our
radial distribution function gtarget (r ). One option is to choose a own group we have described coarse-grained interactions
simple functional form for the interaction, based on physical by a number of functions with coefficients which are left
intuition, and optimize the parameters of this function to for automatic optimization by treating them as dynamical
reproduce the complicated radial distribution function as close parameters in an extended ensemble [55]. Notably, Ashbaugh
as possible [7, 8, 44, 46]. For example, the nonbonded et al used a combination of MD simulations and inverse MC
interactions between the coarse-grained polystyrene beads methods to self-consistently map structural correlations from
of figure 3 have been found by optimizing the parameters atomistic simulations of alkane oligomers onto coarse-grained
of a generalized Lennard-Jones-type interaction [44]. As potentials [56].
figure 4 shows, the resulting nonbonded radial distribution A warning is in place here. All techniques described above
functions agree very well with those obtained from atomistic aim at reproducing the structure by means of pair interactions.
simulations. This means that the true potential of mean force, equation (1),

6
J. Phys.: Condens. Matter 23 (2011) 233101 Topical Review

is only approximated and many-body interactions beyond a and polypropylene. For the dynamics local MC moves,
certain number of particles (usually two) are ignored. As a including a crankshaft move, were introduced. Recently
consequence, it is not guaranteed that the thermodynamic state the dynamical updates have been redesigned to study the
is correctly described [5, 47, 55]. mean square displacement in polyethylene melts [70] and in
An option is to include thermodynamic properties in the bidisperse polyethylene melts [71].
optimization scheme. For example Reith et al [47] added
an extra pair potential with an absolute value decreasing 4.3. Backmapping: reinserting atomistic details
linearly to zero at a cutoff range. The amplitude (positive
or negative) and cutoff range were used as new parameters An important task of CGMD models is to quickly generate
for re-optimization. This yielded a radial distribution function well-equilibrated atomistic polymer structures. For this it
which did not deteriorate too strongly, combined with a correct is, of course, necessary to be able to map an equilibrated
pressure. Also, Nielsen et al [57] have developed a coarse- coarse-grained model back onto an atomistic model. Such a
grained model for n-alkanes (polyethylene) by optimizing backmapping procedure is possible, as has been extensively
bond and bend parameters and the nonbonded Lennard-Jones reviewed in [3, 7].
parameters to match observables from atomistic simulations Among the first polymers to be equilibrated in this
as well as experimental surface tension and bulk density way are various polycarbonates [46, 48, 72, 73], but more
data. recently backmapping procedures have also been applied to
As already mentioned, when coarse-graining only a polystyrene [44, 58, 74–76] and polyamide [77, 78] chains.
few atoms to beads the interactions are still relatively An exciting new development, with possible applications
strong. This means that, even though longer simulation to the determination of the rheology, is the extension
times can be reached, equilibration is still sometimes an of backmapping to non-equilibrium situations [79]. In
issue. In such cases the equilibration techniques developed this method deformed conformations are maintained during
for atomistically detailed models can be used with high backmapping by applying position restraints. The method has
efficiency. For example, recently EB techniques have been been demonstrated for atactic polystyrene under steady shear
applied to coarse models of poly(ethylene terephtalate) and flow.
atactic polystyrene [23–25, 58].
In summary, when coarse-graining one should always 4.4. Dynamic properties from coarse-grained molecular
keep an eye on what properties one is interested in. Because in dynamics: rescaling of time
practice one is limited to approximating the potential of mean
force in terms of pair interactions, it is impossible to correctly CGMD simulations have not only been used to accelerate
represent both the structural and thermodynamic properties. the equilibration of polymer melts, but also to study their
Improving the agreement with one will invariably deteriorate dynamics [7, 9]. We have already discussed that additional
the agreement with the other, so compromises need to be friction factors and stochastic forces are inexorably linked to
made. the procedure of coarse-graining. However, when the degree
of coarse-graining is not too large, friction is often ignored,
4.2. Lattice models leading to too fast dynamics. This may optimistically be
referred to as ‘speed-up’. In such cases, when dynamic
Up to this point we have only considered molecular models properties are presented, it is assumed that all important
in continuous space. Atomistic models have also been processes are accelerated by the same factor (usually 2–
mapped to lattice models. One of the most important lattice 5), i.e. that a simple rescaling of time by a single scaling
models for polymers is the MC-based bond fluctuation (BF) factor can recover the real dynamics. Although in most
model [59–61], where each superatom occupied eight sites on cases this assumption has no theoretical justification, it
a cubic lattice. Bond distances and angles can vary between nevertheless seems to be correct for low degrees of coarse-
different discrete values. The advantage of this approach is graining, as may be witnessed by the many successful
that the possible distances and angles can be chosen in such applications of time rescaling that may be found in the
a way that bond crossing becomes impossible. Mapping of literature [41, 44, 53, 74, 75, 80–88].
real polymers onto the BF model has been tried for bisphenol Calibrating the timescale may be done by comparing chain
polycarbonates and polyethylene [62–65]. Coarse-graining diffusion coefficients, although for most realistic polymers
using the BF model is extensively reviewed in [3]. the atomistic simulations do not reach long enough times to
Mattice and co-workers have developed a mapping from observe free chain diffusion. Since the coarse-graining level
a rotational isomeric state (RIS) description of a polymer is relatively modest, there is hardly any difference between a
onto a diamond lattice for the second-nearest-neighbour bead in a relatively short (but still polymeric) chain and a bead
positions of polyolefin backbones [66–69]. The intermolecular in a very long chain. Thus it can be expected that a diffusion-
potential in this model, with different values of the energy based calibration may be done using relatively short chains.
for different ranges of distances between the superatoms, Harmandaris and co-workers have been very successful in
was tuned by aiming at producing a zero second virial consistently determining the time scaling factors by comparing
coefficient (corresponding to theta conditions) as well as a the atomistic and coarse-grained mean square displacements of
correct radius of gyration for polymers such as polyethylene relatively short chains of polystyrene [41, 44, 74]. Figure 5

7
J. Phys.: Condens. Matter 23 (2011) 233101 Topical Review

5.1. Friction and stochastic forces

As already mentioned, the Mori–Zwanzig formalism of
coarse-graining the dynamics leads to a generalized Langevin
equation (GLE), equation (2), in which the friction and
stochastic forces appear with memory [10, 39, 40, 91–93].
Even though this equation is exact, it is very difficult to use
in practice. In the first place, evaluation of the memory terms
requires taking averages over a projected dynamics which one
does not know exactly how to generate. In the second place,
integrating the resulting integro-differential equations is an
extremely challenging numerical task.
Usually one therefore assumes that the timescale
associated with changes in the positions of the coarse-grained
coordinates is much larger than the timescale associated with
the decay of the friction memory. In this so-called Markov
approximation the time dependence of the friction ζmn in
equation (2) is represented by a delta function, resulting in an
ordinary Langevin equation for the dynamics of the coarse-
grained particles. Whether or not the Markov approximation
is valid must be carefully evaluated in each particular case.
Another difficulty remains: the friction and stochastic
forces on a particle still depend on the positions and velocities
of all particles. As already mentioned, it is often assumed
that the friction is pairwise additive with the contribution of
a pair depending only on the distance and relative velocities
between the two particles involved. Doing so, we arrive at
Figure 5. Example of a successful time mapping of coarse-grained the pair friction model employed in the so-called dissipative
(CG) simulations onto chemically realistic united atom (UA) MD particle dynamics (DPD) method [94]. Although this method
simulations of polystyrene. Two melts, with molecular weights 1 and
2 kDa, have been studied, both at T = 463 K. The time scaling factor
is still relatively costly, especially when the frictions are high,
S needed for the chain centre-of-mass mean square displacement (a) it is rather popular since it conserves momentum, which is a
is consistent with the time scaling factor for the time correlation of prerequisite for hydrodynamic behaviour, and it is easy to use
the end-to-end vector (b). Reprinted with permission from [41]. in non-equilibrium simulations. Within our group we have
Copyright 2009 American Chemical Society. studied in detail the pair friction parallel and perpendicular
to the axis connecting the centres of mass of two halves
of a polymer chain in a melt [35]. Also, coarse-grained
shows a recent example [41], where it is found that the forcefields for polyethylene and cis-polybutadiene for use in
mean square displacement of the chain centre-of-mass and DPD have been determined in the group of Rousseau [95, 96].
the end-to-end vector time correlation can both be accurately The Rousseau group discussed the dependence of the friction
reproduced by the same scaling factor S . It is also possible to coefficient on the coarse-graining level in view of the overall
use other shorter local timescales to calibrate the timescale. For scaling of the dynamical properties.
example, one can use the time associated with the relaxation of The use of pair frictions in DPD is necessarily linked
subchains, as sampled by time correlations of the higher Rouse to the use of a second-order propagator. This implies that
modes [1]. one must use low frictions to reach long timescales (high
frictions require too small time steps). The damage an
artificially lowered friction does to the rheological properties
is often counterbalanced by using harder conservative
5. Lumping many atoms into blobs: the role of
interactions [97]. This often leads to unrealistic structure on
friction and uncrossability
the scale of the bead.
Fortunately, in the case of polymer melts it is often
In order to reach even larger time and length scales one admissible to use a scalar friction (i.e. with a static
may combine many more atoms, say 10 monomeric units, background), because the friction may be thought of as being
into one coarse-grained particle. The particle positions caused by the motion of (part of) a chain relative to the rest of
must be updated according to a CGSD scheme because the the material, which to a first approximation may be taken to
friction and stochastic forces often dominate the interactions. be at rest. Propagation of a velocity field as in a normal liquid
Neglect of friction and stochastic forces can lead to erroneous is highly improbable, meaning that hydrodynamic interactions
dynamics [89, 90]. For example, the motion at certain length are screened [1]. The positions of the coarse-grain particles
scales may appear as oscillatory, whereas the real polymer is can then be updated by a first-order Brownian dynamics
overdamped at these scales. propagator, using much larger time steps than are possible with

8
J. Phys.: Condens. Matter 23 (2011) 233101 Topical Review

As more and more atoms are coarse-grained into one

particle, the interactions between them become softer and
softer [88, 112, 113], see figure 6. Beyond a certain
degree of coarse-graining, bonds will be able to cross each
other. Without additional measures the important entanglement
effect, leading to altered and much slower dynamics, is
therefore lost.
An algorithm that can detect and prevent unphysical
bond crossings has been developed by us [112]. Employing
this so-called Twentanglement algorithm in a simulation of
highly coarse-grained polymer chains will reintroduce the
entanglement effect. The principle of the algorithm is depicted
in figure 7. The bonds are considered to be elastic bands
between the bonded particles. The algorithm keeps track of
all (unattached) bond vectors which are close together. For
each bond vector and at each instant of time the following triple
product is calculated:
Figure 6. The potential of mean force between bonded (circles) and
nonbonded (squares) coarse-grained pieces of polyethylene Vi j = (ri − r j ) · [(ri+1 − ri ) × (r j +1 − r j )], (5)
( T = 450 K), each piece representing the centre of mass of 20
carbon groups. These potentials were obtained from distribution where we have used figure 7 as a reference for the indices.
functions measured in atomistically detailed MD simulations of The absolute value of equation (5) is the volume of the
C120 H242 . Note that kB T = 3.74 kJ mol−1 , meaning that without parallelepiped defined by the vectors ri+1,i , r j +1, j , and ri, j .
additional measures bonds can easily cross each other.
Aside from some pathological cases [112], if Vi j changes
sign from one time step to the next, a bond crossing
may have occurred. Additional checks are made to ensure
a second-order propagator. One should be aware, however,
that the crossing is taking place along the physical part
that such a first-order Brownian dynamics propagator is not
of the line segments (the above equation checks if two
Galilean-invariant. Extra care should therefore be taken when
simulating flow of a polymer melt. Often it is assumed infinite lines have crossed). If a real bond crossing has
that the particles flow affinely with the imposed deformation, occurred, an entanglement is created at the crossing point.
although in the case of shear flow this assumption may be Subsequently, the associated volume Vi j will serve to detect
relaxed [98–100]. future disentanglements. If the volume Vi j of the four
objects surrounding an entanglement changes sign, a possible
disentanglement has occurred, i.e. figure 7 may also be read
5.2. Uncrossability of bonds and entanglement
backwards. Usually only a few of the uncrossability constraints
The advantage of lumping together only a few atoms in contribute to an entanglement in the usual sense of long-lasting
CGMD models is that the resulting interactions are still obstacles, slowing down the chain movement. For instance, a
sufficiently strong to prevent bond crossing. In bead–spring C60 H122 chain is generally considered not to be entangled, yet
models, such as the important finitely extensible nonlinear many ‘entanglements’ occur in a coarse-grained simulation.
elastic (FENE) model [101–111], uncrossability is ensured The dynamics of the ‘entanglement’ points is determined by
by the use of relatively hard beads. Such models therefore a balance of forces, as described in detail in [112].
belong to the class of CGMD models. Relatively small Because uncrossability is explicitly taken care of, there
friction and stochastic forces have been added to bead–spring is much freedom to choose the number of monomers per
models, but with the goal of stabilizing the integration of the coarse-grained particle. In our previous work, the following
equations of motion over very long timescales [102], rather considerations were taken into account: (1) the degree
than representing the physical friction caused by the degrees of coarse-graining should be large enough to allow for a
of freedom that have been coarse-grained out. significant increase in the time and length scales accessible to

Figure 7. Principle of the Twentanglement algorithm. (a) Two line segments representing a bond are closing in on each other. (b) At a certain
moment these bonds will touch. An ‘entanglement’ is created at the crossing point X . (c) After this, the bonds are viewed as slippery elastic
bands. The elasticity will slow down the relative speed of the bonds. This sequence of events may also be reversed.

9
J. Phys.: Condens. Matter 23 (2011) 233101 Topical Review

rigid core to the DPD spheres [121]. Note that both these
schemes essentially make the conservative interactions more
rigid, which is opposite to the trend observed when coarse-
graining to higher levels. One of the dire consequences is
the introduction of unwanted structure on the scale of the
DPD particle, which makes these methods unsuitable for high
degrees of coarse-graining.
Uncrossability of chain bonds can alternatively be
introduced by introducting an additional repulsive interaction
which is based on the distance of closest approach between
two bonds [113, 123, 124]. Another possibility to conserve
the topology, at least for Brownian dynamics simulations, is
to simply forbid random displacements that lead to crossing of
bonds [125]. In a similar vein, one of us has very recently
developed an algorithm which efficiently treats collisions
between infinitely thin bonds in such a way that momentum
and energy are conserved locally [126]. This allows the study
Figure 8. The zero-shear viscosity η0 as a function of molecular of the relative importance of topological and hydrodynamic
weight Mw for polyethylene at T = 450 K from coarse-grained interactions. It should be noted that in all methods described
simulations using Twentanglement [114] (solid squares) and in this paragraph the bonds themselves stay rigid. A collision
experiment [119] (dashed lines). The result from a chemically between two such bonds is a hard collision. In real polymer
detailed simulation of C100 [28] is also included (open circle). melts the bonds between consecutive coarse-grain beads are
Observe that the steep increase in viscosity does not start at the
entanglement molecular weight Me but at a critical molecular weight
highly flexible, and the collisions will consequently be much
Mc which is several times higher [114]. softer (e.g. as in Twentanglement). As noted above [81, 120],
the intrinsic stiffness of the bonds may lead to artificially
enhanced entanglement effects. These methods are therefore
simulation, and (2) if one wants to study reptation effects, the more suitable for simulation of semiflexible chains, or for
degree of coarse-graining should not be so large that the size flexible chains at such a low degree of coarse-graining that
of a coarse-grained particle exceeds the typical diameter of the local chain stiffness is still important.
tube in the reptation picture, in other words the entanglement
length. A suitable choice for polyethylene was to coarse- 6. Primitive paths
grain 20 CH2 units to one particle [112]. We have also
been able to measure the required friction on such a coarse- We have seen in the previous section that beyond a certain level
grained particle from a chemically realistic simulation by of coarse-graining it becomes necessary to somehow maintain
analysing the correlations in the constraint force required the information about the topology of the system, otherwise
to keep the location of a coarse-grained particle fixed in the important entanglement effect is lost. One option is to
an atomistically detailed simulation [114]. By using the use algorithms such as Twentanglement. Another option is to
Twentanglement algorithm we have been able to predict the switch to a one-chain model based on the reptation model [1].
dynamics and rheological properties of polyethylene melts This, however, poses several problems. Reptation theory is still
up to C1000 (figure 8) [114–117] and the dynamics of facing many questions which cannot be answered within the
model. For example, to quote a recent review of Likhtman [2],
poly(ethylene-alt-propylene) melts [118]. We also applied
it is still not clear what is the entanglement or tube field
the method to entangled worm-like micelles of a specific
restricting the chain motion. What are the statistical properties
surfactant chemistry and found quantitative agreement with
of the tube—is it semiflexible on the length scale of the tube
experimentally determined rheology [98].
diameter? What is the microscopic basis for the assumptions
Analysing the results of these coarse-grained simulations
used to describe constraint release in the linear and nonlinear
we found that the chain stiffness is an important ingredient
regimes? What happens to the entanglements and the tubes
for the dynamics of polymer chains. Indeed, Faller and
after large deformations or in fast flows? A practical question
Müller-Plathe noted that chain stiffness intensifies the reptation
of relevance for quantitative predictions is how the numerical
characteristics of polymer melt dynamics [81, 120].
parameters of the tube model, such as the entanglement mass
Topological constraints have also recently been introduced
and tube diameter, can be related to the chemical composition
in DPD simulation methods [97, 113, 121]. In standard of the polymer.
DPD, the conservative forces between the polymer segments
decrease linearly with increasing pair distance. The
6.1. Primitive path analysis
interactions are therefore naturally quite soft and polymer
chains built from such soft beads behave as phantom chains The questions posed above have provoked an outburst of
who pass freely through each other [122]. Nikunen et al activities which try to determine the primitive path and
prevented such chain crossing by simply increasing the entanglement characteristics from more detailed simula-
strength of the DPD forces [97], whereas Liu et al added a tions [6, 22, 31, 58, 84, 102, 127–134]. All efforts are

10
J. Phys.: Condens. Matter 23 (2011) 233101 Topical Review

length. The estimated entanglement molecular weight Me

is usually in good agreement with the experimental value
obtained from the plateau modulus G 0 .
The plateau modulus is a mechanical property of the
system. Following a small step strain γ0 , the shear stress
γ0 G(t) initially decays according to local relaxation processes
on length scales smaller than the tube diameter. For well-
entangled chains, after the entanglement time τe , further
relaxation of G(t) is severely delayed by the topological
interactions, leading to an apparent plateau; only for times
much longer than τe can G(t) finally relax to zero. The plateau
modulus may be defined as the modulus at times larger than
τe , but much smaller than the terminal relaxation time. The
relation between the plateau modulus and the entanglement
molecular weight is given by [1]

4 ρ RT
G0 = . (7)
5 Me
Figure 9. Result of a primitive path analysis of a melt of 200 chains
Here ρ is the polymer (mass) density. This expression is
each consisting of 350 beads. The primitive path of one chain is
shown (in red), together with all of those it is entangled with (in very similar (apart from the factor 4/5) to a prediction for
blue). The primitive paths of all other chains are shown as thin lines. the affine model of rubber elasticity [1]. Indeed, the good
From [127]. Reprinted with permission from AAAS. agreement between the entanglement length from the primitive
path analysis and the entanglement length from the plateau
modulus may intuitively be understood because they both
based on Edward’s view [1] of a primitive path as the probe the same (albeit temporary) rubber-like elasticity of the
shortest path remaining when one holds chain ends fixed, underlying primitive network of chains.
while continuously reducing (shrinking) a chain’s contour The agreement with other estimates of the entanglement
without violating uncrossability [130]. The chain contours length is less good, however, especially when these estimates
are reduced simultaneously until topological constraints block are based on characteristic changes in the time dependence
further shrinkage. of quantities which measure the motion of individual chains
Everaers et al introduced one such minimization method within their prospective tubes [61, 116]. It may be argued
to extract the tube diameter and entanglement length from that this disagreement is caused by the dependence on non-
more detailed simulations of linear chains [127, 128], see exact theory in interpreting these measurements [111, 130].
figure 9. On the basis of the tube model, both the real chain This is certainly a factor, but the remarkable observation
and the primitive path beyond some length scale are described is that with reasonable assumptions in most cases time-
as random walk chains with the same mean square end-to- resolved measurements yield an entanglement length which is
end distance R 2 , but with different contour lengths. The consistently larger than the one based on the plateau modulus.
entanglement length and tube diameter d correspond to the For example, for polyethylene, entanglement length estimates
length and end-to-end distance of an entanglement strand, from the dynamic structure factor, the crossover in Rouse mode
which is identified with the Kuhn segment of the primitive path. time correlations and the entanglement time τe (estimated
Thus it is assumed that the following relations hold [1]: directly from the time of the inflection point in G(t)) all give a
R 2  R 2  value which is consistently larger by a factor of 1.5 compared
d= , Ne = N , (6) to the estimate from the plateau modulus [116].
L pp  L pp 2
The disagreement with the ‘plateau modulus’ entangle-
where L pp  is the average primitive path contour length and ment length may have two causes. First, dynamic time-
N is the number of Kuhn segments of the real chain. Ne is resolved quantities are probably more sensitive to details of
the entanglement molecular weight expressed as the number of the statistics (e.g. fluctuations) of the entanglements. Second,
Kuhn segments of the real chain. A refinement of this method dynamic time-resolved quantities may be more sensitive to the
to exclude self-entanglement was introduced later [135]. It finite stiffness of a real polymer chain on the scale of the tube
turned out that entanglements between distant sections of the diameter.
same chain make a negligible contribution to the tube and In a recent article [137], Greco noted that, because the
that the contour length between local self-knots is significantly tube diameter is relatively small, it is very important to
larger than the entanglement length. include fluctuation effects in the description of entanglements.
The above method has been applied, among others, to Unfortunately, at the time of writing this review, there is
polystyrene melts [41], to semidilute solutions of stiffer chains still no definite agreement on the best way to determine
such as actin [136], and to bisphenol-A-polycarbonate [82], the primitive paths and their statistics. Different coarse-
which was found to have a surprisingly low entanglement graining schemes for arriving at the primitive path lead to

11
J. Phys.: Condens. Matter 23 (2011) 233101 Topical Review

Figure 10. Schematic representation of a primitive chain network model. (a) The motion of slip-links (rings) is influenced by the tension in
the chain segments (arrows) and an osmotic force. (b) The motion of the monomers through the slip-links (arrows) results in repetitive
motion. (Picture kindly provided by Y Masubuchi.)

differences in the statistics. For example, Kröger and co- not stopped researchers from trying to predict polymer melt
workers proposed an alternative algorithm which returns a rheology by employing primitive path-based models.
shortest path and related number of entanglements for a given An important development in this direction is the primitive
configuration of polymers [129, 133]. Primitive paths are chain (or slip-link) network model developed by Masubuchi
treated as infinitely thin and tensionless lines rather than and others [142, 143]. In this model polymer chains are coarse-
multibead chains, and excluded volume is taken into account grained to the level of segments between entanglements, see
without a force law. Their implementation allows construction figure 10. Chain coupling is achieved by confining two chains
of an optimal shortest path for 3D systems. The number by a common slip-link placed at a specific position in 3D space.
of entanglements is then obtained from the shortest path as The chains fulfil a force balance at the slip-links, similar to the
either the number of interior kinks, or from the average length treatment of uncrossability constraints in the Twentanglement
of a line segment. This primitive path analysis has been algorithm [112] which act on a somewhat smaller scale. The
applied to polyethylene C24 to C1000 [133, 138, 139], as well Langevin equation for the slip-links contains both the tension
as the FENE model [133], validating analytical predictions in the chain segments emanating from the slip-link and an
of Schieber [140] about the shape of the distribution for osmotic force arising from density fluctuations. The motion of
the number of strands in a chain at equilibrium. It was monomers through the slip-links ultimately generates reptation
as well as tube length fluctuations. When creating new slip-
also concluded that the number of entanglements obtained
links (i.e. new entanglements), a partner chain segment is
by assuming random walk statistics, as is done in the work
chosen randomly among those spatially close to the advancing
of Everaers and co-workers, deviates significantly from the
chain end.
predictions by the algorithm of Kröger and co-workers.
The primitive chain network model correctly predicts the
The random walk approximation partly fails because
scaling of the longest relaxation time and the self-diffusion
real polymers are typically still somewhat stiff at the with chain length [142]. The model was later modified to also
scale of the average distance between two uncrossability include the concept of dynamic tube dilution [144], and good
constraints [130–133]. As a result, the tube diameter cannot be agreement between simulated and real linear and nonlinear
equated to the Kuhn step length of the primitive path. Although rheology was found. Extensions to branched polymers, blends
this fact is well known for semiflexible fibre suspensions, and block copolymers have all been made [145, 146]. Recently
where theoretical expressions for the tube diameter are the model has been subjected to various sudden deformations
different from expressions for the deflection length (average to study the damping function, which is essentially the ratio
distance between successive collisions of the fibre with its of nonlinear stress relaxation to the linear shear relaxation
tube) [141], it seems to be often ignored in the polymer melt modulus [147–149]. It was found that the force balance is
literature. Stiffness effects have been introduced recently in a dominant correction over the basic Doi–Edwards theory
primitive path analyses by Hoy et al [133] and Tzoumanekas as compared with the effect of convective constraint release.
et al [132]. Furthermore, the predicted normal stress ratio in shear,
a quantity which is very sensitive to different modelling
6.2. Primitive chain networks and slip-link models assumptions, was found to be in good agreement with
experiments.
Although the precise characteristics of entanglements and/or In order to facilitate fast computations of the rheology,
confining tubes have remained somewhat elusive, this has various single-chain slip-link models have also been developed

12
J. Phys.: Condens. Matter 23 (2011) 233101 Topical Review

[150–155]. In such models an ensemble of primitive paths Modern applications also include nanoparticles in the melt
is followed in time. Coupling between chains is not explicit, to form composites. To optimize the process parameters for
as in the above primitive chain network model, but virtual such complex flows it is important to understand the interplay
by defining random relations between slip-links on different between time and length scales of the polymer and those
chains. Whenever a slip-link of a chain is destroyed due to tube imposed by the complex geometries. The low frequency linear
renewal at a chain end, the corresponding partner along the and nonlinear rheology of polymer melts is basically concerned
contour of another chain is also eliminated. The same applies with displacements of centres of mass only. Given the need
to the creation of new slip-links. to also reach large length scales, it is therefore natural to just
Very recently, the effectiveness of such a single-chain try to develop simulation models that only keep track of the
slip-link model in describing the dynamics and rheology of dynamics of the centres of mass of all diffusing constituents.
entangled polymers was tested by systematically comparing An early application of this idea was presented by Murat
slip-link with FENE MD results [156]. The parameters and Kremer [159]. These authors developed an efficient
of the model were determined by using the mean square and rather general model in which whole polymer chains
displacements of one particular chain length as a target are represented as soft ellipsoidal particles. The interaction
function. Using the same set of parameters, it was then between two such particles is taken to be proportional to the
tested if the predictions of the mean square displacements for spatial overlap of their monomer density distributions. Since
other chain lengths agreed with the MD calculations. This the internal degrees of freedom of a chain are integrated out,
was followed by a comparison of the shear relaxation moduli many thousands of chains can be simulated within reasonable
G(t). The work identified a limitation of the original slip– computer time. More recently, McCarthy et al introduced an
spring model in describing the static structure of the polymer approach based on the Ornstein–Zernike equations, in which
chain as seen in MD, which was remedied by introducing a polymer is coarse-grained to a soft sphere on the scale of
a pairwise repulsive potential between the monomers in the its radius of gyration [160]. At such a high level of coarse-
chains. Also, poor agreement of the mean square monomer graining typical maximum interaction energies are of the order
displacements at short times could be rectified by the use of of kB T , the thermal energy.
generalized Langevin equations for the dynamics. The latter, All these approaches focus on reproducing the static
of course, somewhat diminishes the computational efficiency structure of a polymer melt, with good to excellent agreement
of the method. with the structure obtained from more detailed simulations.
Slip-link models are generally successful in their overall However, because such models ignore memory and/or
‘rheological’ performance and predictions [130]. They also entanglement effects, they do not have the characteristic slow
offer the possibility to introduce new tube renewal mechanisms dynamics of real polymer chains and therefore cannot be used
and, by turning these on or off at will, probe the effect on the to predict the long-time rheology.
rheology independent of other mechanisms. A disadvantage of One may be tempted to slow down the dynamics of such
both single-chain and multi-chain slip-link models is that the a super-coarse-grained system by using a large friction term.
link with the chemistry is usually distant; if a comparison with Lyubimov et al recently introduced a method to estimate the
a microscopically detailed model is made at all, the parameters friction on a polymer represented by a soft particle [161]. Their
are usually obtained by fitting rather than an ab initio coarse- approach is based on an intermediate mapping of a chemically
graining procedure. detailed chain on a freely rotating chain model and using a
An attempt at systematically bringing in structure from hard-bead approximation to evaluate integrals involving the
a detailed model, albeit again a bead–spring-type model, radial distribution function and dynamic structure factor. This
was introduced by Rakshit and Picu [157]. In their method yields a series of diffusion coefficients in good agreement with
reptation is enforced because the chain inner blobs are results of atomistic simulations of unentangled and slightly
constrained to move along the backbone of the coarse-grained entangled polyethylene melts. No rheological predictions were
chain (the primitive path), while the end blobs move in made.
the three-dimensional embedding space. These end blobs The use of a single large friction term may be questioned,
continuously redefine the diffusion path for the inner blobs. It however, because in real entangled polymeric systems
was shown that various distribution functions, relaxation rates the frictions and random forces have a memory of the
and the diffusion dynamics are properly represented. Recently, configurations the system has gone through in the recent, and
this model was used to study start-up and step strain shear sometimes even the distant, past. A simple Brownian dynamics
flows [158]. The authors showed that their method reproduces propagator with realistic mean forces and uncorrelated, fully
several features observed experimentally such as the overshoot random displacements without memory will not reproduce
during start-up shear flow and shear thinning at large shear correct sequences of configurations of the retained coordinates.
rates. In recent years, our group has introduced and explored a
new efficient framework to reintroduce memory of previous
7. Super-coarse-graining: a polymer as a single soft configurations in soft matter simulations [99, 100, 162–166].
particle In this framework, called responsive particle dynamics
(RaPiD), configurations and forces are described by as few
Industrial processing of polymers usually involves extrusion of variables as possible. In the case of polymer chains,
a hot melt through a die or injection of a melt into a mould. similar to the works mentioned above, we usually restrict

13
J. Phys.: Condens. Matter 23 (2011) 233101 Topical Review

ourselves to three coordinates for the location of each centre of one single computer simulation to calculate the large-
of mass. To circumvent the complicated introduction of scale rheological properties starting from the detailed chemical
memory effects in friction forces and stochastic displacements structure. Instead, some form of coarse-graining is necessary.
(as in the generalized Langevin equation, equation (2)) In this review we have made a distinction between CGMD and
we introduce a relatively small set of additional dynamic CGSD.
variables, symbolically denoted as n(t), which keep track In CGMD a few atoms (up to five backbone carbon atoms)
of the thermodynamic state of the eliminated coordinates are lumped into relatively hard beads. Uncrossability of bonds
for the given values of the retained coordinates. If the is therefore automatically ensured. A large body of work
configuration of retained coordinates is suddenly changed has gone into deriving the effective interactions between the
(e.g. by a sudden change in the distance between the centres relatively hard beads, with IBI emerging as one of the most
of mass of two polymers), the equilibrium values n 0 for the popular techniques. When coarse-graining the interactions,
additional variables change too, but the relaxation to n 0 takes technically one should include additional friction and random
place over a finite time. This gives rise to strong transient forces, but fortunately in CGMD the friction is not dominant.
forces in addition to the thermodynamic forces deriving from If friction is ignored the dynamics will be too fast, usually by a
the potential of mean force. It is possible to show that, to lowest factor 2–5. Combined with the lower number of particles this
order, the transient forces may be thought of as originating leads to a significant computational speed-up of the order of
from a penalty free energy function which is quadratic in the 100. Agreement with detailed atomistic simulations is often
deviations of the additional variables [163]. recovered by a suitable time scaling. Using these techniques it
The particular additional variables and corresponding is possible to predict the rheology of melts of specific polymers
names that are used to describe the transient forces depend on up to one or two entanglement lengths.
the system under study. In some cases, such as for telechelic To reach a higher degree of coarse-graining, in CGSD
polymer networks, the variables are the number of bridging many more atoms are lumped together (more than ten
polymeric chains between nodes of the network [100, 165]. In backbone carbon atoms), leading to relatively soft beads.
other cases, more relevant to this review on polymer melts, the Friction and stochastic forces dominate the interactions.
variables are the number of entanglements shared between a Unless counteractions are taken, the soft interactions cannot
particular polymer chain and its immediate neighbours [162]. prevent crossing of the bonds between the coarse-grained
To some readers it will be rather surprising that a single- particles. Possible counteractions include application of the
particle model would produce rheological results that are most Twentanglement algorithm or prohibition of moves that lead
naturally described by assuming curvilinear tubes surrounding to crossing of bonds. Using such techniques it is possible to
each chain which prohibit the chains to move in any way other predict the rheology of melts of specific polymers up to about
then along the primitive paths of the tubes. A few qualitative ten entanglement lengths.
remarks may be of help at this point [163]. First of all, the main To predict the rheology of polymers of longer polymers
action of the tubes is to conserve the prevailing configurations (say, more than ten entanglement lengths), one is forced
of the chains and their centres of mass in particular. This is to make use of the tube model by obtaining entanglement
exactly what the transient forces do as well. In a sense, the characteristics through a primitive path analysis and simulating
transient forces tie all centres of mass harmonically to their a primitive chain network. Primitive path analysis algorithms
instantaneous positions. Second, beyond the Rouse time, i.e. as allow us to determine the tube diameter and entanglement
soon as the chains perform one-dimensional diffusive motions molecular weight from atomistically detailed or CGMD
along their tube axes, as a result of the Gaussian character simulations. The entanglement molecular weight so obtained
of the tube configurations the centres of mass of all chains is often in good agreement with the entanglement molecular
weight derived from the experimental plateau modulus.
perform simple random displacements. The only thing the tube
However, predictions of time-dependent properties rely on
does is to define the distribution of possible displacements and
details of the interpretation of reptation theory, and often
their averages in particular. Again this is what the transient
seem to be determined by a somewhat different entanglement
forces do as well.
molecular weight. A prerequisite for future computational
The RaPiD method has demonstrated its usefulness by
developments is to have available a better understanding of
correctly reproducing the long-time linear and nonlinear rheol-
the relation between the two apparently different entanglement
ogy of linear polymers [162], telechelic polymers [100, 165],
molecular weights and their relations to the effects of chain
star polymers [164] and polymeric adhesives [166]. The next
stiffness, finite size fluctuations and tube renewal mechanisms.
challenge is to link the few additional parameters introduced
This information needs to be reliably predicted from detailed,
by the method to the chemical details of the system. In some
chemically realistic models. Improvements in understanding
cases, such as for the telechelic polymers [100, 165], such a
will be used to develop more realistic slip-link models. In this
link has already been successfully established.
respect it is important to note that already now slip-link models
are very effective and fast in deriving the rheological properties
8. Conclusion of polymer melts.
Finally we discussed the use of super-coarse-grained
The enormous range of time and length scales associated models, in which a full polymer is represented by a single
with the dynamics of entangled polymers precludes the use particle, which may be used to predict flow and rheology on

14
J. Phys.: Condens. Matter 23 (2011) 233101 Topical Review

industrially relevant large time and length scales. Because [20] Baig C and Mavrantzas V G 2007 Phys. Rev. Lett. 99 257801
coarse-graining has been taken to its limit, conservative [21] Baig C and Mavrantzas V G 2009 Phys. Rev. B 79 144302
[22] Karayiannis N C and Kröger M 2009 Int. J. Mol. Sci.
interactions between full polymer chains are extremely soft
10 5054–89
and memory or entanglement effects are prominent. To [23] Kamio K, Moorthi K and Theodorou D N 2007
efficiently introduce memory effects we have developed the Macromolecules 40 710–22
RaPiD method. [24] Mulder T, Harmandaris V A, Lyulin A V, van der Vegt N F A,
Vorselaars B and Michels M A J 2008 Macromol. Theory
In summary, our search for the link between chemical Simul. 17 290–300
structure and long-time dynamics has led to many innovations [25] Mulder T, Harmandaris V A, Lyulin A V, van der Vegt N F A
during the last two decades. Computer simulations will and Michels M A J 2008 Macromol. Theory Simul.
continue to smooth the bumpy road from chemistry to 17 393–402
[26] Paul W, Yoon D Y and Smith G D 1995 J. Chem. Phys.
rheology. 103 1702–9
[27] Paul W, Smith G D and Yoon D Y 1997 Macromolecules
30 7772–80
Acknowledgments [28] Moore J D, Cui S T, Cochran H D and Cummings P T 2000
J. Non-Newton. Fluid Mech. 93 83
[29] Harmandaris V A, Mavrantzas V G and Theodorou D N 1998
We gratefully acknowledge support by the European Union Macromolecules 31 7934–43
through the Network of Excellence ‘SoftComp’, the Initial [30] Padding J T and Briels W J 2001 J. Chem. Phys. 114 8685–93
Training Network ‘Dynacop’ and NMP ‘Modify’. [31] Harmandaris V A, Mavrantzas V G, Theodorou D N,
Kröger M, Ramirez J, Ottinger H C and
Vlassopoulos D 2003 Macromolecules 36 1376–87
[32] Smith G D, Paul W, Monkenbusch M, Willner L, Richter D,
References Qiu X and Ediger M D 1999 Macromolecules 32 8857–65
[33] Tsolou G, Mavrantzas V G and Theodorou D N 2005
Macromolecules 38 1478–92
[1] Doi M and Edwards S F 1986 The Theory of Polymer
[34] Ryckaert J P 2005 Comput. Phys. Commun. 169 89–94
Dynamics (Oxford: Clarendon)
[2] Likhtman A E 2009 J. Non-Newton. Fluid Mech. 157 158–61 [35] Akkermans R L C and Briels W J 2000 J. Chem. Phys.
[3] Baschnagel J et al 2000 Bridging the gap between atomistic 113 6409–22
and coarse-grained models of polymers: status and [36] Briels W and Akkermans R 2002 Mol. Simul. 28 145–52
perspectives Viscoelasticity, Atomistic Models, Statistical [37] Likos C 2001 Phys. Rep. 348 267–439
Chemistry (Advances in Polymer Science vol 152) [38] Louis A 2002 J. Phys.: Condens. Matter 14 9187–206
pp 41–156 [39] Zwanzig R 1961 Phys. Rev. 124 983
[4] Kremer K and Muller-Plathe F 2002 Mol. Simul. 28 729–50 [40] Kinjo T and Hyodo S 2007 Phys. Rev. E 75 051109
[5] Faller R 2004 Polymer 45 3869–76 [41] Harmandaris V A and Kremer K 2009 Macromolecules
[6] Aleman C, Karayiannis N C, Curco D, Foteinopoulou K and 42 791–802
Laso M 2009 J. Mol. Struct. THEOCHEM 898 62–72 [42] Fritz D, Harmandaris V A, Kremer K and van der Vegt N F A
[7] Muller-Plathe F 2002 ChemPhysChem 3 754–69 2009 Macromolecules 42 7579–88
[8] Muller-Plathe F 2003 Soft Mater. 1 1–31 [43] Fukunaga H, Takimoto J and Doi M 2002 J. Chem. Phys.
[9] Paul W and Smith G D 2004 Rep. Prog. Phys. 67 1117–85 116 8183–90
[10] Padding J T and Briels W J 2007 Ab initio coarse-graining of [44] Harmandaris V A, Reith D, van der Vegt N F A and
entangled polymer systems Nanostructured Soft Matter: Kremer K 2007 Macromol. Chem. Phys. 208 2109–20
Experiment, Theory, Simulation and Perspectives ed [45] Fukunaga H, Aoyagi T, Takimoto J and Doi M 2001 Comput.
A Zvelindovsky (Berlin: Springer) pp 435–58 Phys. Commun. 142 224–6
[11] Auhl R, Everaers R, Grest G S, Kremer K and Plimpton S J [46] Tschop W, Kremer K, Batoulis J, Burger T and Hahn O 1998
2003 J. Chem. Phys. 119 12718–28 Acta Polym. 49 61–74
[12] Mavrantzas V G, Boone T D, Zervopoulou E and [47] Reith D, Putz M and Muller-Plathe F 2003 J. Comput. Chem.
Theodorou D N 1999 Macromolecules 32 5072–96 24 1624–36
[13] Harmandaris V A, Mavrantzas V G and Theodorou D N 2000 [48] Abrams C F and Kremer K 2003 Macromolecules 36 260–7
Macromolecules 33 8062–76 [49] Meyer H, Biermann O, Faller R, Reith D and
[14] Doxastakis M, Mavrantzas V G and Theodorou D N 2001 Muller-Plathe F 2000 J. Chem. Phys. 113 6264–75
J. Chem. Phys. 115 11339–51 [50] Reith D, Meyer H and Muller-Plathe F 2001 Macromolecules
[15] Doxastakis M, Mavrantzas V G and Theodorou D N 2001 34 2335–45
J. Chem. Phys. 115 11352–61 [51] Sun Q and Faller R 2005 Comput. Chem. Eng. 29 2380–5
[16] Uhlherr A, Mavrantzas V G, Doxastakis M and [52] Zhao L, Li Y G, Mi J G and Zhong C L 2005 J. Chem. Phys.
Theodorou D N 2001 Macromolecules 34 8554–68 123 124905
[17] Uhlherr A, Doxastakis M, Mavrantzas V G, Theodorou D N, [53] Faller R and Reith D 2003 Macromolecules 36 5406–14
Leak S J, Adam N E and Nyberg P E 2002 Europhys. Lett. [54] Sun Q and Faller R 2006 J. Chem. Theory Comput. 2 607–15
57 506–11 [55] Akkermans R L C and Briels W J 2001 J. Chem. Phys.
[18] Karayiannis N C, Mavrantzas V G and Theodorou D N 2002 114 1020
Phys. Rev. Lett. 88 105503 [56] Ashbaugh H S, Patel H A, Kumar S K and Garde S 2005
[19] Karayiannis N C and Mavrantzas V G 2006 Atomistic Monte J. Chem. Phys. 122 104908
Carlo methods for the simulation of polymers with a linear [57] Nielsen S O, Lopez C F, Srinivas G and Klein M L 2003
or non-linear molecular architecture Multiscale Modelling J. Chem. Phys. 119 7043–9
of Polymer Properties (Computer-Aided Chemical [58] Spyriouni T, Tzoumanekas C, Theodorou D,
Engineering vol 22) ed M Laso and Mueller-Plathe F and Milano G 2007 Macromolecules
A Perpète (Amsterdam: Elsevier) pp 31–67 40 3876–85

15
J. Phys.: Condens. Matter 23 (2011) 233101 Topical Review

[59] Carmesin I and Kremer K 1988 Macromolecules 21 2819–23 [98] Padding J T, Boek E S and Briels W J 2008 J. Chem. Phys.
[60] Wittmer J, Paul W and Binder K 1992 Macromolecules 129 074903
25 7211–9 [99] van den Noort A and Briels W J 2008 J. Non-Newton. Fluid
[61] Tanaka M, Iwata K and Kuzuu N 2000 Comput. Theor. Polym. Mech. 152 148–55
Sci. 10 299–308 [100] Sprakel J, Spruijt E, van der Gucht J, Padding J T and
[62] Baschnagel J, Binder K, Paul W, Laso M, Suter U W, Briels W J 2009 Soft Matter 5 4748–56
Batoulis I, Jilge W and Bürger T 1991 J. Chem. Phys. [101] Kremer K, Grest G S and Carmesin I 1988 Phys. Rev. Lett.
95 6014–25 61 566–9
[63] Paul W, Binder K, Kremer K and Heermann D W 1991 [102] Kremer K and Grest G S 1990 J. Chem. Phys. 92 5057–86
Macromolecules 24 6332–4 [103] Gao J and Weiner J H 1995 J. Chem. Phys. 103 1614–20
[64] Paul W and Pistoor N 1994 Macromolecules 27 1249–55 [104] Smith S W, Hall C K and Freeman B D 1995 Phys. Rev. Lett.
[65] Tries V, Paul W, Baschnagel J and Binder K 1997 J. Chem. 75 1316–9
Phys. 106 738–48 [105] Gao J P and Weiner J H 1996 Macromolecules 29 6048–55
[66] Rapold R F and Mattice W L 1995 J. Chem. Soc. Faraday [106] Smith S W, Hall C K and Freeman B D 1996 J. Chem. Phys.
Trans. 91 2435–41 104 5616–37
[67] Rapold R F and Mattice W L 1996 Macromolecules [107] Faller R, Putz M and Muller-Plathe F 1999 Int. J. Mod. Phys.
29 2457–66 C 10 355–60
[68] Doruker P and Mattice W L 1997 Macromolecules 30 5520–6 [108] Aoyagi T and Doi M 2000 Comput. Theor. Polym. Sci.
[69] Haliloglu T and Mattice W L 1998 Macromolecules 10 317–21
108 6898–995 [109] Putz M, Kremer K and Grest G S 2000 Europhys. Lett.
[70] Lin H, Mattice W L and Von Meerwall E D 2006 J. Polym. 49 735–41
Sci. B 44 2556–71 [110] Kröger M 2004 Phys. Rep. 390 453–551
[71] Lin H, Mattice W L and von Meerwall E D 2007 [111] Hou J X, Svaneborg C, Everaers R and Grest G S 2010 Phys.
Macromolecules 40 959–66 Rev. Lett. 105 068301
[72] Tschop W, Kremer K, Hahn O, Batoulis J and Burger T 1998 [112] Padding J T and Briels W J 2001 J. Chem. Phys. 115 2846–59
Acta Polym. 49 75–9 [113] Lahmar F, Tzoumanekas C, Theodorou D N and
[73] Hess B, Leon S, van der Vegt N and Kremer K 2006 Soft Rousseau B 2009 Macromolecules 42 7485–94
Matter 2 409–14 [114] Padding J T and Briels W J 2002 J. Chem. Phys. 117 925–43
[74] Harmandaris V A, Adhikari N P, van der Vegt N F A and [115] Padding J T and Briels W J 2003 J. Chem. Phys.
Kremer K 2006 Macromolecules 39 6708–19 118 10276–86
[75] Milano G and Muller-Plathe F 2005 J. Phys. Chem. B [116] Padding J T and Briels W J 2004 J. Chem. Phys.
109 18609–19 120 2996–3002
[76] Santangelo G, Di Matteo A, Muller-Plathe F and [117] Kindt P and Briels W J 2005 J. Chem. Phys. 123 224903
Milano G 2007 J. Phys. Chem. B 111 2765–73 [118] Perez-Aparicio R, Colmenero J, Alvarez F, Padding J T and
[77] Carbone P, Varzaneh H A K, Chen X and Briels W J 2010 J. Chem. Phys. 132 024904
Mueller-Plathe F 2008 J. Chem. Phys. 128 064904 [119] Pearson D S, Ver Strate G, von Meerwall E and Schilling F C
[78] Karimi-Varzaneh H A, Carbone P and Mueller-Plathe F 2008 1987 Macromolecules 20 1133
J. Chem. Phys. 129 154904 [120] Faller R and Muller-Plathe F 2001 ChemPhysChem 2 180–4
[79] Chen X, Carbone P, Santangelo G, Di Matteo A, [121] Liu H, Xue Y H, Qian H J, Lu Z Y and Sun C C 2008
Milano G and Mueller-Plathe F 2009 Phys. Chem. Chem. J. Chem. Phys. 129 024902
Phys. 11 1977–88 [122] Spenley N A 2000 Europhys. Lett. 49 534–40
[80] Hahn O, Delle Site L and Kremer K 2001 Macromol. Theory [123] Kumar S and Larson R G 2001 J. Chem. Phys. 114 6937–41
Simul. 10 288–303 [124] Hoda N and Larson R G 2010 J. Rheol. 54 1061–81
[81] Faller R and Muller-Plathe F 2002 Polymer 43 621–8 [125] Ramanathan S and Morse D C 2007 J. Chem. Phys.
[82] Leon S, van der Vegt N F A, Delle Site L and Kremer K 2005 126 094906
Macromolecules 38 8078–92 [126] Padding J T 2009 J. Chem. Phys. 130 144903
[83] Depa P K and Maranas J K 2005 J. Chem. Phys. 123 094901 [127] Everaers R, Sukumaran S K, Grest G S, Svaneborg C,
[84] Depa P K and Maranas J K 2007 J. Chem. Phys. 126 054903 Sivasubramanian A and Kremer K 2004 Science 303 823–6
[85] Sun Q and Faller R 2006 Macromolecules 39 812–20 [128] Kremer K, Sukumaran S K, Everaers R and Grest G S 2005
[86] Chen C X, Depa P, Sakai V G, Maranas J K, Lynn J W, Comput. Phys. Commun. 169 75–81
Peral I and Copley J R D 2006 J. Chem. Phys. 124 234901 [129] Kröger M 2005 Comput. Phys. Commun. 168 209–32
[87] Chen C, Depa P, Maranas J K and Sakai V G 2008 J. Chem. [130] Tzoumanekas C and Theodorou D N 2006 Curr. Opin. Solid
Phys. 128 124906 State Mater. Sci. 10 61–72
[88] Strauch T, Yelash L and Paul W 2009 Phys. Chem. Chem. [131] Tzoumanekas C and Theodorou D N 2006 Macromolecules
Phys. 11 1942–8 39 4592–604
[89] Oettinger H C 2007 MRS Bull. 32 936–40 [132] Tzoumanekas C, Lahmar F, Rousseau B and Theodorou D N
[90] Qian H J, Liew C C and Mueller-Plathe F 2009 Phys. Chem. 2009 Macromolecules 42 7474–84
Chem. Phys. 11 1962–9 [133] Hoy R S, Foteinopoulou K and Kröger M 2009 Phys. Rev. E
[91] Guenza M 2003 J. Chem. Phys. 119 7568–78 80 031803
[92] Guenza M G 2008 J. Phys.: Condens. Matter 20 033101 [134] Baig C and Mavrantzas V G 2010 Soft Matter 6 4603–12
[93] Hijon C, Espanol P, Vanden-Eijnden E and [135] Sukumaran S K, Grest G S, Kremer K and Everaers R 2005
Delgado-Buscalioni R 2010 Faraday Discuss. 144 301–22 J. Polym. Sci. B 43 917–33
[94] Hoogerbrugge P J and Koelman J M V A 1992 Europhys. Lett. [136] Uchidaa N, Grest G S and Everaers R 2008 J. Chem. Phys.
19 155 128 044902
[95] Guerrault X, Rousseau B and Farago J 2004 J. Chem. Phys. [137] Greco F 2008 Eur. Phys. J. E 25 175–80
121 6538–46 [138] Foteinopoulou K, Karayiannis N C, Mavrantzas V G and
[96] Lahmar F and Rousseau B 2007 Polymer 48 3584–92 Kröger M 2006 Macromolecules 39 4207–16
[97] Nikunen P, Vattulainen I and Karttunen M 2007 Phys. Rev. E [139] Foteinopoulou K, Karayiannis N C, Laso M and
75 036713 Kröger M 2009 J. Phys. Chem. B 113 442–55

16
J. Phys.: Condens. Matter 23 (2011) 233101 Topical Review

[140] Schieber J D 2003 J. Chem. Phys. 118 5162 [151] Doi M and Takimoto J I 2003 Phil. Trans. R. Soc. A
[141] Hinsch H, Wilhelm J and Frey E 2007 Eur. Phys. J. E 24 35 361 641–52
[142] Masubuchi Y, Takimoto J I, Koyama K, Ianniruberto G, [152] Shanbhag S and Larson R G 2004 Macromolecules 37 8160–6
Marrucci G and Greco F 2001 J. Chem. Phys. 115 4387–94 [153] Xu F, Denn M M and Schieber J D 2006 J. Rheol. 50 477–94
[143] Masubuchi Y, Ianniruberto G, Greco F and Marrucci G 2003 [154] Likhtman A E 2005 Macromolecules 38 6128–39
[155] Likhtman A E, Sukumaran S K and Ramirez J 2007
J. Chem. Phys. 119 6925–30
Macromolecules 40 6748–57
[144] Yaoita T, Isaki T, Masubuchi Y, Watanabe H, Ianniruberto G, [156] Sukumaran S K and Likhtman A E 2009 Macromolecules
Greco F and Marrucci G 2004 J. Chem. Phys. 121 12650–4 42 4300–9
[145] Masubuchi Y 2006 Nihon Reoroji Gakkaishi 34 275–82 [157] Rakshit A and Picu R C 2006 J. Chem. Phys. 125 164907
[146] Masubuchi Y, Ianniruberto G, Greco F and Marrucci G 2006 [158] Rakshit A and Picu R C 2008 Rheol. Acta 47 1039–48
J. Non-Cryst. Solids 352 5001–7 [159] Murat M and Kremer K 1998 J. Chem. Phys. 108 4340–8
[147] Furuichi K, Nonomura C, Masubuchi Y, Ianniruberto G, [160] McCarty J, Lyubimov I Y and Guenza M G 2009 J. Phys.
Greco F and Marrucci G 2007 Nihon Reoroji Gakkaishi Chem. B 113 11876–86
35 73–7 [161] Lyubimov I Y, McCarty J, Clark A and Guenza M G 2010
[148] Furuichi K, Nonomura C, Masubuchi Y, Watanabe H, J. Chem. Phys. 132 224903
Ianniruberto G, Greco F and Marrucci G 2008 Rheol. Acta [162] Kindt P and Briels W J 2007 J. Chem. Phys. 127 134901
47 591–9 [163] Briels W J 2009 Soft Matter 5 4401–11
[164] Padding J T, van Ruymbeke E, Vlassopoulos D and Briels W J
[149] Masubuchi Y, Uneyama T, Watanabe H, Ianniruberto G,
2010 Rheol. Acta 49 473–84
Greco F and Marrucci G 2009 Nihon Reoroji Gakkaishi [165] Sprakel J, Padding J T and Briels W J 2011 Europhys. Lett.
37 65–8 93 58003
[150] Tasaki H, Takimoto J I and Doi M 2001 Comput. Phys. [166] Padding J T, Mohite L V, Auhl D, Briels W J and Bailly C
Commun. 142 36–9 2011 Soft Matter 7 5036–46

17