Molecular dynamics simulations of glassy polymers

Jean-Louis Barrat, Joerg Baschnagel, Alexei Lyulin

To cite this version:

Jean-Louis Barrat, Joerg Baschnagel, Alexei Lyulin. Molecular dynamics simulations of glassy poly-
mers. Soft Matter, 2009, 6, pp.3430. �10.1039/b927044b�. �hal-00455377�

HAL Id: hal-00455377

https://hal.science/hal-00455377
Submitted on 10 Feb 2010

HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est

archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents
entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non,
lished or not. The documents may come from émanant des établissements d’enseignement et de
teaching and research institutions in France or recherche français ou étrangers, des laboratoires
abroad, or from public or private research centers. publics ou privés.
 Molecular dynamics simulations of glassy
polymers

Jean-Louis Barrat∗ Jörg Baschnagel† Alexey Lyulin ‡

February 10, 2010

Abstract
We review recent results from computer simulation studies of polymer glasses,
from chain dynamics around glass transition temperature Tg to the mechanical be-
haviour below Tg . These results clearly show that modern computer simulations
are able to address and give clear answers to some important issues in the field, in
spite of the obvious limitations in terms of length and time scales. In the present
review we discuss the cooling rate effects, and dynamic slowing down of different
relaxation processes when approaching Tg for both model and chemistry-specific
polymer glasses. The impact of geometric confinement on the glass transition is
discussed in detail. We also show that computer simulations are very useful tools
to study structure and mechanical response of glassy polymers. The influence of
large deformations on mechanical behaviour of polymer glasses in general, and
strain hardening effect in particular are reviewed. Finally, we suggest some direc-
tions for future research, which we believe will be soon within the capabilities of
state of the art computer simulations, and correspond to problems of fundamental
interest.

1 Introduction
A polymer is a macromolecular chain resulting from the connection of a large number
of monomeric units. The number of monomers (the chain length N) typically ranges
between 103 and 105 in experimental studies of polymer melts [1]. Recently, this range
of chain lengths has also become accessible in simulations [2, 3]. Such a long chain
has an open structure which is strongly pervaded by other chains in the melt [4]. The
strong interpenetration of the chains has important consequences for the properties
of the melt. For instance, intrachain excluded volume interactions, which swell the
polymer in dilute solution [1, 4], are almost screened so that a chain behaves on large
length scales approximately as a random coil [3]. Furthermore, chain interpenetration
also impacts the polymer dynamics by creating a temporary network of entanglements.
Entanglements strongly slow down the chain relaxation and make the melt viscoelastic
already at high temperature [5, 6].
∗ Université de Lyon; Univ. Lyon I, Laboratoire de Physique de la Matière Condensée et des Nanostruc-

tures; CNRS, UMR 5586, 43 Bvd. du 11 Nov. 1918, 69622 Villeurbanne Cedex, France
† Institut Charles Sadron, Université de Strasbourg, CNRS UPR 22, 23 Rue du Loess, BP 84047, 67034

Strasbourg Cedex 2, France

‡ Group Theory of Polymers and Soft Matter, Eindhoven Polymer Laboratories, Technische Universiteit

Eindhoven, P.O. Box 513 5600 MB Eindhoven, the Netherlands, and Dutch Polymer Institute, P.O. Box 902,
5600 AX Eindhoven, The Netherlands.

1
 glass 0.48 melt

0.46 fast cooling

v
0.44
slow
cooling slow heating
0.42
0 0.2 0.4 0.6 0.8 1
T
semicrystalline polymer

Figure 1: Volume-temperature diagram of a coarse-grained model for poly(vinyl al-

cohol) [7, 8]. The volume per monomer (v) and the temperature (T ) are given in
Lennard-Jones units. In the liquid phase the chains have random-coil-like configura-
tions and the structure of the melt is amorphous. The amorphous structure is preserved
on rapid cooling and the melt eventually undergoes a glass transition at low T . For
slow cooling the melt transforms into a semicrystalline material in which sections of
folded chains order in lamellar sheets that coexist with amorphous regions. When the
crystal is slowly heated up (dashed line), it melts at a higher T than the temperature
where crystallization occurs. This hysteresis is typical of first-order phase transitions.

On cooling toward low temperature the polymer melt eventually transforms into a
solid. Polymeric solids are either semicrystalline or glassy (Fig. 1) [9]. In the semicrys-
talline state amorphous regions are intercalated between crystalline lamellar sheets.
The sheets consist of chains which are folded back on themselves so that chain sec-
tions align parallel to each other [9, 10]. The ability to form crystals crucially depends
on the polymer microstructure. Only chains with regular configurations, e.g. isotatic or
syndiotatic orientations of the sidegroups or chains without sidegroups, can fold into
crystalline lamellae. However, even in these favorable cases full crystallization is hard
to achieve [10].
This intrinsic difficulty to crystallize favors glass formation [11, 12, 13]. Polymer
melts either can be easily supercooled (Fig. 1) or, due to the irregular chain structure,
do not crystallize at all. An example for the former case is bisphenol-A polycarbon-
ate (PC) [14], while examples for the latter involve homopolymers with atactic (bulky)
sidegroups, such as atactic polystyrene (PS), or random copolymers, such as cis-trans-
1,4-polybutadiene (PBD). These polymeric glass formers share with other (intermedi-
ate and fragile) glass-forming liquids the key characteristic feature of the glass transi-
tion; that is, little change of the amorphous structure, but a huge non-Arrhenius-like
slowing down of the dynamics on cooling toward the glass transition temperature Tg
[12, 13, 15]. Understanding the molecular origin of this disproportionate behaviour
represents a great scientific challenge [12, 13, 15, 16, 17]. In addition to this funda-
mental interest in the study of the glass transition, solid polymers are also integral com-
ponents in many modern applications [18]. For instance, glassy polymeric materials
are appreciated because of their spectacular mechanical properties [19, 20, 21]. Instead
of failing abruptly when subject to strong deformations, some polymers, such as poly-
carbonate, may harden for large strains, leading to a tough mechanical response. A
microscopic understanding, elucidating the structure-property relationship, of this be-

2
haviour is still elusive. A further example is provided by thin polymer films, extensively
used in technological applications as protective coatings, optical coatings, adhesives,
etc. In the ongoing quest for progressively smaller structures and devices these films
may attain nanoscopic dimensions, where deviations from the bulk behaviour should
be expected [22]. Indeed, many recent studies suggest that the Tg of thin films is shifted
by spatial confinement, a striking observation which is not well understood yet [23, 24].
This wide array of challenging questions—from the glass transition to the impact
of external stimuli—bestows the theoretical understanding of glassy polymers with
particular significance. Molecular simulations can contribute to this research. Over
the past two decades, computational models and methods have been developed for
simulating these glassy systems. The progress made in the field has regularly been
the subject of topical reviews. For instance, detailed reports of chemically realistic
modeling approaches may be found in Refs. [25, 26, 27], while work on coarse-grained
models is reviewed in Refs. [28, 29, 30, 31, 32, 27]. The purpose of the present article
is to give a brief account of recent work.
We begin our survey by a short introduction to the modeling of polymers in sim-
ulations (section 2). The glass transition temperature naturally splits the following
discussion into two parts, a part devoted to the properties of the model glass formers
above Tg (section 3) and a part addressing sub-Tg phenomena (section 4). In section 3,
our discussion is mainly concerned with the dynamics of weakly supercooled polymer
melts and the impact of spatial confinement on their behaviour. Section 4 focuses on
the response of polymeric glasses to both weak and large external deformation, where
respectively the linear mechanical behaviour of the glassy melt and its approach to
material failure will be explored. The article concludes with a short summary of the
presented results and an outlook on possible future research directions.

2 Computer simulations: models and computational as-

pects
Molecular simulations of glass-forming polymers face the important challenge that
the structure and dynamics of these systems are governed by a large spread of length
and time scales [33, 34]. The relevant length scales extend from the atomic diameter
(∼ 10−10 m) to the chain dimension (∼ 10−7 m for N ∼ 104 ). The spread of time
scales is even larger; it ranges from bond vibrations (∼ 10−13 s) to the slow structural
relaxation close to Tg (∼ 102 s).
From the inspection of these scales, it is clear that simulation approaches have some
limitations and also require simplifications. An obvious limitation are the accessible
time scales. With the currently available computer power the longest time that can
be reached in molecular dynamics (MD) simulations is roughly a few µ s. This time
scale is 8 orders of magnitude smaller than the relaxation time at the experimental Tg ,
implying that the simulated Tg is shifted by about 25 degrees to higher temperature (T )
relative to experiments (where we used the rule of thumb: 1 decade in time ≈ b 3 K). The
simplifications that are currently necessary concern the simulation models which are
obtained by some kind of coarse-graining procedure, designed to eliminate (some) fast
degrees of freedom by incorporating them in effective potentials. This model-building
step can take different levels of complexity [35, 36, 2, 37], but roughly speaking, there
are two families, atomistic models and coarse-grained models. In the following, we
briefly introduce these models and discuss their strengths and weaknesses as we go.

3
 Before doing so, however, we want to point out some general appealing features of
the simulations to balance the reservations expressed above. The simulations offer full
control over the perfectly defined system under study. For instance, it is possible to vary
in a systematic manner parameters, such as chain stiffness, chain length, etc., while
keeping all other defining properties of the model, so as to single out the impact of each
parameter on the properties of the system. Furthermore, the simulations enable one to
study local properties or to explore correlations which are hard (or even impossible) to
access in experiments. This provides microscopic insight and a means to test theoretical
concepts. Examples from the recent literature involve investigations of the potential
energy landscape of glass-forming materials [38, 39, 40], of single-molecule diffusion
in polymeric matrices [41, 42] or of fracture in glassy polymers [27]

Atomistic models. Atomistic models replace electronic degrees of freedom by force

fields, i.e., empirical potentials for bond-length, bond-angle, torsional-angle, and non-
bonded interactions, whose parameters are determined from quantum-chemical cal-
culations and experiments [26, 35]. During the past decades, force fields have been
developed for both explicit atom (EA) models, treating every atom present in nature
as a separate interaction site, and united atom (UA) models which represent a small
number of real atoms (e.g., CH2 or CH3 ) by a single site. The reduction of the num-
ber of interaction sites in the UA model has the computational advantage of allowing
longer simulation times. With a time step of ∼ 10−15 s a few thousand united atoms
can be simulated over several 100 ns, about an order of magnitude longer than an EA
simulation of comparable system size.
Both EA and UA models were employed to study the properties of glassy poly-
mers (see e.g. [25, 26, 27] for reviews). Recent examples include polyisoprene (EA
model [43, 44]), atactic PS (UA model [45, 46, 47, 48, 49]), bisphenol-A PC (UA
model [48, 49]) 1,4-PBD (UA and EA models [26, 50, 51, 52, 53, 54, 55, 56]), and
poly(ethylene oxide) or atactic poly(propylene oxide) (EA models [57]). Certainly, the
main strength of these modeling efforts is that the simulation results allow for a direct
comparison with experiments. In some cases, such a comparison is possible with com-
mercial software packages [56], while a careful fine-tuning of the force field is often
required [50, 54]. Experience from those modeling approaches—not only for poly-
mers but also e.g. for amorphous SiO2 [31, 58]—suggests that the design of a chem-
ically realistic model, aiming at a parameter-free comparison between simulation and
experiment, should involve information about both structural and dynamic properties
[26].

Generic models. Atomistic simulations are ideally suited to study specific polymers,
including their glass transition and the properties of the glassy state. On the other
hand, the strong increase of the relaxation time, which eventually leads to vitrifica-
tion on cooling through Tg , is common to all glass-forming polymers, irrespective of
their chemical composition and architecture. This universal aspect suggests to em-
ploy simplified simulation models which only retain generic features of a polymer.
Various such generic models—on a lattice [31, 30, 28, 28] or in the spatial contin-
uum [31, 30, 32]—have been studied for glass-forming polymers. In the continuum,
the simplest model are highly flexible bead-spring models, where spherical monomers
are tethered together by springs and have nonbonded, Lennard-Jones (LJ) interactions
[31, 30, 32]. More chemical realism can be introduced by making the chains semi-
flexible, through the addition of a bond-angle potential [59] and, possibly also, of a

4
torsional potential [60, 61, 62]. Due to their simplicity, these generic models allow for
an efficient simulation. This is important if one wants to vary the cooling rate [63] or
the strain rate [27] over decades, to explore systematically the impact of model parame-
ters, such as chain length or chain flexibility [59, 60, 61, 62], or to obtain good statistics
for comparison with theory [64]. Computational expedience is also important for ex-
ploratory studies of more complex systems, such as inhomogeneous polymer systems
(e.g., polymer films, nanocomposites, semi-crystalline polymers) or glassy polymer
mixtures (e.g., dynamically asymmetric polymer-polymer mixtures or polymer-solvent
systems). Some examples will be discussed in Sects. 3 and 4.

Hierarchical models. For generic models, the gain in the accessible length and time
scales is obtained at the expense of a loss of correlation to the atomistic conforma-
tion of the polymer. For many problems in materials research, this loss is undesirable.
Therefore, much research efforts currently goes into the development of hierarchical
approaches consisting of interconnected levels of modeling (atomistic, generic, macro-
scopic) [2, 37, 65, 66]. The idea is that each level treats phenomena on its specific
length and time scales and then passes on the results as input to the next, more coarse-
grained level, until the desired materials properties can be predicted. Such multiscale
simulation methods represent a powerful approach whose potential for the modeling of
glassy polymers is beginning to be explored (see e.g. [67]).

Remarks on simulation methods. Sections 3 and 4 will present results from molec-
ular dynamics (MD) simulations, a numerical method to integrate the classical equa-
tions of motion for a many-body system in a given thermodynamic ensemble [68, 69].
Therefore, MD is the natural simulation technique to address dynamical problems,
such as the glass transition. However, the realistic MD dynamics carries an obvious
price: the equilibration time for the system under consideration—for instance, for a
long-chain polymer melt close to its Tg —will exceed the maximum time of a few mi-
croseconds one is currently able to simulate.
Here Monte Carlo (MC) techniques may provide a promising avenue because of
large freedom to design MC moves [68, 70]. The hope is to find an efficient algo-
rithm allowing one to decorrelate the configurations of glassy polymer melts rapidly
For long-chain polymer melts, this demand on the algorithm implies, already at high
T , that the MC move should be nonlocal, i.e., it should modify the chain conformation
at large scales, and it should not require empty space because the melt is a dense liquid.
A promising algorithm satisfying these requirements employs double-bridging moves
which alter the connectivity between two neighbouring chains while preserving the
monodispersity of the chains (for recent reviews see e.g. [3, 2]). Such a connectivity-
altering move drastically changes the conformation of the two chains involved and thus
relaxes the length scales on the order of the chain dimension efficiently. However, it
does not alter the local packing of the monomers. An inherent hazard of the algorithm
therefore is that, if the move is attempted repeatedly, a successful double-bridging event
is likely to annihilate one of its predecessors by performing the transition between two
chains in the reverse direction. To avoid this inefficiency the nonlocal chain updating
should be complemented by a move which efficiently mixes up the local structure of the
melt. At low T , efficient relaxation of the liquid structure calls for a method which al-
leviates the glassy slowing down in general. Thus, any algorithm achieving this aim in
nonpolymeric liquids should also accelerate the equilibration of glassy polymer melts,
provided that it can be generalized to respect chain connectivity. At present, no tech-

5
nique has been established to solve this problem. However, possible candidates could
be parallel tempering [71, 72, 73], Wang-Landau sampling [70] or variants thereof [74],
or transition path sampling methods [75].

3 Glass transition and properties of the supercooled poly-

mer liquid
3.1 Bulk properties
3.1.1 Glass transition temperature
In polymer melts, the transition from the glass to the liquid is accompanied by a strong
decrease in the shear modulus, typically of three to four orders of magnitude [21]. It
is thus clear that the glass transition temperature is a characteristic of high engineering
relevance. Classical methods for the experimental determination of Tg are calorime-
try and dilatometry. Both methods hint at the kinetic features of the glass transition.
The transition occurs when the relaxation time for volume recovery (dilatometry) or
enthalpy recovery (calorimetry) becomes longer than the time scale of the experiment
(i.e., than the cooling or heating rate) [11, 12, 13]. Upon cooling the polymer liquid
falls out of equilibrium close to Tg and freezes below Tg in a glassy state, the properties
of which depend on the details of the cooling process and tend to age physically during
further isothermal equilibration [11, 12, 13]. The glass transition temperature is located
in the T interval where the polymer melt smoothly evolves from the liquid to the solid
state, and can be defined operationally through some prescription, for instance, as the
intersection of straight-line extrapolations from the glassy and liquid branches of the
volume-temperature curve [11, 13].
Similar extrapolation procedures are also applied in simulation studies [26, 63, 30,
47, 48, 77, 78, 76, 79, 80]. The resulting Tg values have dependences comparable to ex-
perimental ones despite the much larger cooling rate employed—typically 1012 K/min
in simulations and 10 K/min in experiments. For instance, Soldera and Metatla find a
linear relationship between numerical and experimental Tg values from atomistic simu-
lations of various vinylic polymers [77]. Figure 2 reveals that Tg decreases nonlinearly
with the logarithm of the cooling rate (ΓT ) [26, 63, 30, 47],
B
Tg (ΓT ) = Tg0 − , (1)
ln(AΓT )
in accordance with experimental observation [81]. Also in agreement with experiment
[11, 82, 83, 84, 85], Tg increases with chain rigidity and chain length [60, 47, 86]. The
chain length dependence can be fitted to the empirical Fox-Flory equation
K
Tg (N) = Tg∞ − , (2)
N
which usually describes experimental data well (if the molecular weight is not too small
[11]), although other forms have recently been discussed in the literature [84, 85] and
can be rationalized theoretically [87].

3.1.2 Dynamics of the supercooled melt

The precursor of the glass transition is the strong slowing down of structural relaxation
processes on approach to the transition from the liquid. This dynamical feature is

6
Tg Tg
PS semiflexible
440
0.5
420
flexible
400
(a) 0.4 (b)
380
10
-2
10
-1
10
0 0 10 20 30
ΓT N
Figure 2: Panel (a): Glass transition temperature (Tg ) versus cooling rate (ΓT ) for
an atomistic model of atactic PS [47]. The simulation box contains 8 chains with
N = 80 monomers each. The solid line shows a fit to Eq. 1 with T0 = 371 K, B = 110
K, and A = 0.23 ps/K. Panel (b): Tg versus chain length (N) for a fully flexible and
a semiflexible (with angular potential) bead-spring model [76]. The simulation box
contains at least 192 chains and the cooling rate is ΓT = 2 × 10−5. The solid lines are
results of a fit to Eq. 2 with Tg∞ = 0.432, K = 0.145 (flexible model) and Tg∞ = 0.525,
K = 0.264 (semiflexible model). All data are given in Lennard-Jones units for the
bead-spring models.

7
 4
10 400 K
3
213 K
10 198 K ~t
0.61
2
Rg
2
gM(t) [Å ]

10
2

2
1 σ
10 2
6rrsc
0
10
-1
10
-2
10 -2 -1 0 1 2 3 4 5 6
10 10 10 10 10 10 10 10 10
t [ps]
Figure 3: Mean-square displacement (MSD) gM (t) versus time t for a united atom
model of 1,4-PBD (results adapted from [53], with permission). The MSD is averaged
over all united atoms in the melt containing 40 chains of 30 repeats units. The temper-
atures shown are (from left to right): T = 400 K, 323 K, 273 K, 240 K, 225 K, 213
K, and 198 K (Tc ≈ 214 K). The horizontal dashed lines indicate the radius of gyration
R2g ≈ 218 Å2 , the average Lennard-Jones diameter of the united atoms σ ≈ 3.8 Å, and
an estimate for the Lindemann localization length rsc ≈ 0.45 Å [43]. The dotted lines
for T = 400 K and 213 K show the power law ∼ t 0.61 , characteristic of Rouse-like
motion.

8
a hallmark of strongly interacting disordered matter, including dense colloids [88],
granular materials [89], and supercooled liquids [12, 16]. In simulated polymer melts,
it is observed in all dynamic correlation functions, e.g., in dynamic structure factors,
conformational correlation functions, or dielectric relaxation [26].
As an example, Fig. 3 shows the mean-square displacement (MSD) gM (t), aver-
aged over all monomers of a chain, for a chemically realistic model of 1,4-PBD [53].
At high temperature, the MSD directly crosses over from ballistic motion (∼ t 2 ) at
short times to subdiffusive motion (∼ t x0 with x0 ≈ 0.61) at intermediate times where
the MSD is bound between the monomer size (σ ) and the end-to-end distance Re of a
chain. This subdiffusive motion does not depend on the strength of the torsional barrier
(cf. Fig. 4), is present even if the torsional potential is absent [52], and thus reflects the
universal Rouse-like dynamics of nonentangled chains in a polymer melt [4]. The term
“Rouse-like” stresses the fact that simulations of nonentangled chains [26, 64, 90, 91]
find both accord with Rouse predictions—e.g., the Rouse modes are (nearly) orthog-
onal for all t—and deviations from them—e.g., the Rouse modes are stretched with
stretching exponents depending on the mode index and the MSD of the chain’s center
of mass increases sublinearly for intermediate times. The origin of these deviations is
not fully understood. However, for short chains the relaxation time of a chain is not
well separated from the (local) α -relaxation so that finite-N corrections to the Rouse
behaviour must be expected [64]. Moreover, even for long chains intermolecular in-
teractions between the polymers are not completely screened on mesoscopic length
scales, which may cause subdiffusive center-of-mass motion [92] or deviations from
reptation theory [6], and lead to corrections to chain ideality [3, 93].
On cooling towards Tg the Rouse-like motion shifts to progressively longer times
due to the appearance of a plateau regime. In this regime, the MSD increases only
very slowly with time and is of the order of 10% of the monomer diameter, reflecting
the temporary localization of a monomer in the cold melt. For polymer models with
intramolecular rotational barriers, such as PBD, this intermittence of large scale motion
stems from two mechanisms of dynamical arrest [26, 51, 55]: monomer caging by
near neighbours in the dense melt and the slowing down of intrachain conformational
transitions occurring through (correlated) torsional motion.
Recent work by Smith and Bedrov [54] suggests that the interplay of both mecha-
nisms can explain the Johari-Goldstein β relaxation in polymers [13]. Figure 4 repro-
duces one of their results. Since experiments reveal that the separation of the α and β
processes occurs only on time scales significantly longer than the multiple microsecond
trajectories generated for the chemically realistic (CR) PBD model, Smith and Bedrov
employ a PBD model with reduced torsional barriers (LB model) but otherwise iden-
tical interactions as for the CR model. While structural properties of the PBD melt
remain unaffected [52, 51], the reduction of the barriers accelerates the dynamics and
shifts the α -β bifurcation into the simulation time window. Figure 4 compares gM (t)
with the decay of the torsional autocorrelation function (TACF). Similar to gM (t), the
TACF relaxes in two steps. To both steps can be associated relaxation times, τβ and
τα , which display an Arrhenius (τβ ) and a non-Arrhenius (τα ) increase with decreas-
ing T , characteristic of β and α processes, respectively. At low T , both processes are
well separated from each other, the beta process corresponding to times for which the
monomers displace, on average, by about 10% of their diameter. This implies that the
cages imposed by the polymer matrix remain largely intact during the β process and
create a potential energy landscape for the underlying conformational transitions. In-
deed, detailed analysis reveals that (nearly) all dihedrals visit all torsional states during
the β relaxation. However, these visits do not occur with the equilibrium probabil-

9
 -1 0 1 2 3 4 5 6
10 10 10 10 10 10 10 10
0
10

TACF(t)
140 K (LB)
130 K (LB)
213 K (CR)
198 K (CR)
2 -1
σ 10
1
10
gM(t) [Å ]
2

0.61
~t
2
0 6rrsc
10
-1 τβ(140) τβ(130)
10
-1 0 1 2 3 4 5 6
10 10 10 10 10 10 10 10
t [ps]
Figure 4: Monomer MSD gM (t) and torsional autocorrelation function (TACF) versus t
for 1,4-PBD. The data for T = 130 K and 140 K (lines) are obtained from a model with
reduced torsional barriers (LB), but which is otherwise the same as in Fig. 3 (results
adapted from [54], with permission). For comparison the symbols show the MSDs at
T = 198 K and 213 K for the chemically realistic (CR) model with the full torsional
potential. These temperatures are approximately at the same distance to Tg (≈ 170
K) as for the LB model (Tg ≈ 102 K) [54]. The horizontal dashed lines indicate the
average Lennard-Jones diameter of the united atoms σ ≈ 3.8 Å and an estimate for
the Lindemann localization length rsc ≈ 0.45 Å [43]. The solid line for T = 140 K
shows the power law ∼ t 0.61 , characteristic of Rouse-like motion. The vertical dotted
lines represent the β -relaxation times (τβ = 716.5 ps for T = 130 K, τβ = 319.4 ps for
T = 140 K).

10
 5
10
4
Tg(PS) Tg(PC) PS
10 PC
τα/τα(T=600K)

3
10 ~exp(Ea/kBT)

2
10
-γ
1 ~(T-Tc)
10
0 Tc(PS) Tc(PC)
10
300 400 500 600
T(K)
Figure 5: Temperature dependence of the α relaxation time (τα ) for united atom models
of atactic PS and bisphenol-A PC. τα is normalized by its value at T = 600 K for both
polymers (PS: τ (600 K) = 1.4 ps, PC: τ (600 K) = 1.3 ps) and is obtained from the
monomer MSD in simulations of 1 chain with N = 80 for PS and 64 chains with N = 10
for PC. For both polymers, the increase of τα at high T is compatible with the MCT
prediction τα ∼ (T − Tc )−γ (PS: Tc ≃ 375 K, γ ≃ 3; PC: Tc ≃ 450 K, γ ≃ 2.2). However,
the divergence at Tc is avoided and a crossover to an Arrhenius behaviour (PS: Ea ≃ 30
kJ/mol; PC: Ea ≃ 16.8 kJ/mol) occurs close to Tg (derived from dilatometry. PS: Tg ≃
375 K; PC: Tg ≃ 433 K). This crossover indicates the transition from the cooperative
translational dynamics above Tg to activated (β process like) hoping below Tg . Results
adapted from [48].

ity because the propensity of a dihedral to return to a preferred conformational state

increases on cooling, due to the stiffening of the matrix. Equilibrium occupancy of
conformational states is only achieved for times comparable to the α time scale when
the “cage effect” of the matrix fully decays (see also [56, 57] for similar results).
In glass physics, the term “cage effect” is intimately connected to the mode-coupling
theory (MCT) of the glass transition, which has certainly been one of the most influen-
tial theoretical approaches in the field during the last twenty years. A comprehensive
review of its foundations and applications was published recently [17]. By correlat-
ing the equilibrium structure to the dynamics of a glass-forming liquid MCT provides
a framework for interpreting spatio-temporal correlations measured in experiment or
simulation. Analytical predictions are derived in the vicinity of an ideal glass transi-
tion which occurs at a critical temperature Tc and is driven by the mutual blocking of
a particle and its neighbours (“cage effect”). Extensive tests by experiments and simu-
lations indicate that, although (the extrapolated) Tc lies above Tg and thus no structural

11
 rl
Hs(r)/Hs(r’)

0
(a)
-1
0 1 2 3
2
rb rmin
Hd(r)/Hd(r’)

0
(b)
-1
0 1 2 3
r
Figure 6: Simulation results for a flexible bead-spring model: Rx (r,t) versus r for
different times from the plateau regime at T = 0.48 (Tc ≃ 0.45). Panel (a) shows Rx (r,t)
for the self-part of the van Hove function and panel (b) for its distinct part. In panel
(b), the pair-distribution function g(r) is also shown (dotted line; rescaled to fit into
the figure). The first peak of g(r) reflects the bond length (≈ 0.97) and the minimum
position of the Lennard-Jones potential (≈ 1.12). In panel (a), rl (= 0.2323) denotes
the√zero of Rs (r,t) and the circles represent a √
Gaussian approximation which has a zero
at 6rsc ≃ 0.2327 and a minimum around 10rsc ≃ 0.3. Here, rsc (≃ 0.095) is the
Lindemann localization length. In both panels, the dash-dotted lines correspond to the
time closest to t ′ which is the least precise because Rx (r,t) is undetermined for t = t ′ .
Adapted from [94, 32].

12
arrest is observed at Tc (cf. Fig. 5), MCT still describes many dynamical features well
[17, 95]. Why this is so, represents a great challenge for the theoretical understanding
[88, 96, 97, 98, 99, 88, 100].
A key prediction of MCT is that structural relaxation functions should obey a fac-
torization property in the plateau regime (also called β regime in MCT). For the self
(Gs ) and distinct (Gd ) parts of the van Hove correlation function [101] this reads

Gx (r,t) = Fx (r) + Hx (r)G(t) (x = s, d) . (3)

The factorization property refers to the fact that the correction to the nonergodicity
parameter Fx (r) splits into two factors, of which G(t) depends only on time (and tem-
perature) and Hx (r) only on r [17, 95]. Therefore, the ratio (r′ = constant) [102, 103]

Gx (r,t) − Gx (r,t ′ ) Hx (r)

Rx (r,t) = = (4)
Gx (r′ ,t) − Gx(r′ ,t ′ ) Hx (r′ )
should be independent of t. For a flexible bead-spring model [94] Fig. 6 confirms this
prediction directly from the simulation data (no fit) and additionally reveals the length
scales pertinent for the dynamics in the plateau regime because distances for which
Hx (r) is zero will not contribute to the relaxation. For the self part of the van Hove
function the dynamics involves displacements up to the monomer diameter, whereas
for the distinct part it includes monomers up to about the forth neighbour shell. This
local character of the relaxation is a direct evidence for the cage effect.
The agreement between MCT and simulation, demonstrated in Fig. 6, is not limited
to the flexible bead-spring model. Simulations of semiflexible bead-spring models
with rotational barriers [61, 62] and of chemically realistic models, including 1,4 PBD
[53, 50, 55], PS and PC [48], generally find that many features of the spatiotemporal
relaxation in weakly supercooled polymer melts (i.e., T & Tc ) are well described by
MCT (see however [57]). The main difference between these models and the flexible
bead-spring model is that two coexisting arrest mechanisms—intramolecular barriers
and monomer caging—determine the structural relaxation of the former, whereas only
caging operates for the flexible model. This interpretation is a key result of Refs. [51,
52] and of the recent work by Bernabei et al [61, 62] who also make the interesting
conjecture that, within MCT, the pertinent theoretical framework for polymer models
with internal rotational barriers are (so-called) higher-order transition scenarios [17],
as it appears to be the case for other systems with distinct arrest mechanisms, such as
dense colloidal suspensions with short-range attraction [104, 105] or polymer mixtures
with strong dynamic asymmetry [106, 107, 108].
A distinctive feature of MCT is that the dynamics (beyond the short time regime)
is fully specified in terms of the liquid structure. This opens the way for an “ab initio”
prediction of the simulated dynamics, solely based on static input obtained from an in-
dependent simulation of the studied glass former. Such an atomistic theory for the slow
relaxation of (nonentangled) polymer melts has been developed [109] and compared to
simulations of a bead-spring model with N = 10 [110, 64]. The comparison gives
semi-quantitative agreement between simulation and theory (for models with large ro-
tational barriers further complications might arise [62]). As an example, Fig. 7 shows
various MSDs, revealing the strengths and weaknesses of the approach [64]. Certainly,
the agreement for the monomer dynamics is very good. MCT describes the dynamics
in the plateau regime, the following polymer-specific subdiffusive increase, gM ∼ t 0.63
which is identified as a finite-N deviation from Rouse behaviour, and the faster motion
of the end monomer (g1 ) relative to the central monomer (g5 ) of the chain. On the

13
 2.0
g1(t)/g5(t) MD: T=0.47
1.5
α-process
1.0 (α+β)-process
3 -8 -6 -4 -2 0 2
2 log10Dt
1 3M
log10 g(t)

0 2
-1 1
-2 0.63 0
~t C
-3 -1
-4 -2
6Dt -3
-4
-8 -7 -6 -5 -4 -3 -2 -1 0 1 2
log10Dt
Figure 7: Simulation results for a flexible bead-spring model: Log-log plot of the
monomer MSD (labeled M, left scale) and the MSD of the chain’s center of mass
(labeled C, right scale) versus Dt with D being the diffusion coefficient of a chain. The
inset shows the ratio g1 (t)/g5 (t) (end monomer MSD over middle monomer MSD).
The circle refer to the MD results at T = 0.47 (Tc ≃ 0.45), the solid lines to the MCT
α master curve, and the dashed lines to the MCT predictions including the MCT β
process. The dash-dotted lines indicate the diffusive motion 6Dt and the dotted line
shows the power law ∼ t 0.63 of Rouse-like motion. Figure taken from [64].

14
other hand, the theory is not so satisfactory for the MSD of the chain’s center of mass
(gC ). Besides underestimating the plateau height it does not reproduce the subdiffusive
center-of-mass motion [92] (gC ∼ t ≈0.8 ) between the plateau and diffusive regimes.
The representation in Fig. 7—that is, plotting the data versus Dt with D being the
chain’s diffusion coefficient—facilitates the comparison of the time dependence in the
β and early α -regimes, but camouflages a systematic deviation between theory and
simulation. MCT predicts that the α relaxation time τα (q) has the same T depen-
dence for all wave vectors q, whereas simulations, not only for flexible bead-spring
models [32] but also for other glass formers [111, 112, 113, 114, 115], find that τα for
wave vectors smaller than the position q∗ of the first maximum of static structure factor
increases on cooling more weakly than for q & q∗ . This difference also implies a vio-
lation of the Stokes-Einstein relation [116, 117, 118], i.e., the product of the diffusion
coefficient (corresponding to the limit q → 0) and τα (q∗ ) is not independent of T .
This decoupling has been interpreted as a signature of increasingly heterogeneous
dynamics in the liquid near its glass transition [117, 118, 119, 120]). The term “hetero-
geneous dynamics” means that a glass former near Tg contains subensembles of parti-
cles with enhanced or reduced mobility relative to the average. To reveal this dynamic
heterogeneity various methods were deployed [119, 120], including filtering techniques
to track slow or fast particles, analysis of ensemble-averaged three- or four-point cor-
relation functions [121, 122], or scrutiny of the self-part of the van Hove correlation
function [111, 112, 115].
Here we briefly discuss the approach of Refs. [111, 112] which has recently been
applied to simulation data of a flexible bead-spring model [123]. Following [111, 112]
we define the probability distribution P(ln r;t) of the logarithm of monomer displace-
ments in time t by
P(ln r;t) = 4π r3Gs (r,t) , (5)
where Gs (r,t) is the self-part of the van Hove function, as before. This probability is
shown in Fig. 8, together with gM (t) and the non-Gaussian parameter [111, 112]
 
gM (t) 1
α2 (t) =
′
−1 , (6)
3 |~r(t) −~r(0)|2

where ~r(t) denotes the position of a monomer at time t. In the cold melt there are
clear deviations from Gaussian behaviour at all but the shortest and longest times. The
non-Gaussian parameter is positive and has a maximum at time t2′ in the late- √α regime.
At this time the distribution P(ln r;t) is very broad, exhibiting small (r & 6rsc ) and
large (r & 1) displacements. This hints at large, non-Gaussian fluctuations in particle
mobility when α2′ peaks. On cooling toward Tc the broad displacement distribution
develops into a double-peak structure, indicative of large disparities in particle mobility.
Apparently, two populations of monomers coexist, “slow” ones which have not moved
much farther than 10% of their diameter in time t2′ , and “fast” ones which have left their
cage and covered a distance of about their diameter or more. This bimodal character
of the structural relaxation is hard to predict from MCT [112], could be responsible for
the absence of the divergence of the relaxation time at Tc , and appears to be a general
feature of materials close to glass or jamming transitions [115].

15
 t
-2 -1 0 1 2 3 4 5 6
10 10 10 10 10 10 10 10 10
gM(t)
, 0
α2(t) 10
-1
2
10
6rrsc t’t2 -2
1 10
√6 rsc
-3
0.8 10
4πr Gs(r,t)

t=10 t=25000
0.6
Gaussian
3

0.4 t’t2 N=64

0.2 T=0.43
0 -1 0 1
10 10 10
r
Figure 8: Simulation results for a flexible bead-spring model: Probability distribution
P(ln r;t) of the monomer displacements (left scale) at T = 0.43 (Tc ≃ 0.415) for t =
10, 100, 250, 500, 1000, 1500, 2500, 5000, 10000, and 25000 (from left to right).
The thick dotted line shows P(ln r;t) at t = t2′ , the peak time of α2′ (t). The dotted
curve for t = 25000 represents the Gaussian approximation for P(ln r;t). The vertical
dashed√line indicates the displacement corresponding to the Lindemann localization
length 6rsc = 0.232. Right scale: Monomer MSD gM (t) and non-Gaussian parameter
α2′ (t) at T = 0.43. The vertical dotted line indicates the peak time t2′ of α2′ (t) and the
2
horizontal dashed line the “plateau value” 6rsc . Figure adapted from [123].

16
3.2 Effects of confinement on glassy polymers
3.2.1 Brief overview of some experimental results
During the past fifteen years the impact of geometric confinement on the glass tran-
sition has received considerable attention. The progress in the field is described in
several comprehensive reviews [23, 124, 125, 126, 24, 127, 128]. The picture emerg-
ing from these studies is that glass formers confined to nanoscopic dimensions may
exhibit deviations from bulk behaviour due to the interplay of spatial restrictions and
interfacial effects. The latter (nonuniversal) effects can result from particle-substrate
interactions, confinement-induced changes of the liquid structure or polymer confor-
mations, density variations, etc. [13, 24], and often appear to dominate the behaviour
of the confined glass former [127].
These interfacial effects have an important impact on the glass transition of thin
polymer films. For films supported on a substrate, many studies [129, 130, 131, 132,
133, 134, 135, 136, 137, 138, 139, 140, 141, 142, 143], though not all [144, 145], find
reductions in Tg with decreasing film thickness (h) if the polymer-substrate attraction is
weak. This is also the case for one of the most extensively studied systems, polystyrene
on a variety of substrates [129, 131, 132, 134, 135, 136, 137, 139, 140, 141, 142,
143]. Here mesurements of surface relaxation after nanodeformation [146] or of the
positional dependence of Tg [139] further indicate that the depression of Tg is related to
a “free-surface effect”: Monomers near the free surface are expected to be more mobile
because they feel less steric contraints than their peers in the bulk. This enhanced
mobility should lead to reductions in Tg .
Measurements of (the average) Tg provide information on the dynamic reponse of
the polymer films on the time scale associated with Tg which depends on experimental
conditions, such as the cooling rate [147]. Additionally, the full structural relaxation
has also been explored by several techniques [126, 24], in particular by dielectric spec-
troscopy [124, 130, 134, 135, 148, 141, 142, 143, 149, 150, 151, 152]. Dielectric spec-
troscopy allows for a simultaneous measurement of Tg and the relaxation spectrum,
even for thin films. A key finding of these studies is that the α process is broadened
relative to the bulk. It is possible to interpret this broadening as a consequence of
spatially heterogeneous dynamics in nanoconfinement. At the interfaces the segmental
dynamics can be enhanced (e.g. at the free surface) or slowed down (e.g. at an attractive
substrate) relative to the bulk. Chain segments in layers adjoining these interfacial lay-
ers should also have their dynamics perturbed, albeit to a lesser extent, which will lead
to a still weaker perturbation for the next layer, and so on. This interpretation implies
that there is a smooth transition from interface-induced perturbations of the dynamics
to bulk behaviour with increasing distance from the interfaces. Recently, this idea has
been exploited to analyze dielectric spectra of nanostructured diblock copolymer melts
[150], and it is also consistent with the positional dependence of Tg found in [139].

3.2.2 Simulation work on geometrically confined glass-forming polymer systems

Simulation studies of confined glass formers support this view of an interface-induced
gradient in relaxation [32]. Many of these studies utilize simple models, e.g. binary
liquids [153, 154, 155, 156] or bead-spring polymer models [157, 158, 159, 160, 161,
162, 163, 164, 165, 166, 167, 168, 169, 170, 171, 172, 173, 174] (see however [175,
176, 177, 178, 179] for work on chemically realistic models), to explore the impact
of confinement in thin films [153, 155, 157, 158, 159, 160, 161, 162, 163, 164, 165,
166, 167, 168, 169, 170, 171, 172, 173, 174, 177, 178, 179], pores [154] or systems

17
 Tc (supported)
1.0
Tg
Tg(h)/T

0.9
Tc (free) Tg (MC: PP)
Tg (PS, low Mw)
0.8
Tc
Tc(h)/T

~ 10 nm Tg (PS, high Mw)

0.7 Tc (supported+capped)

0 1 2
10 10 10
h/h0
Figure 9: Tc (h)/Tc and Tg (h)/Tg (Tc and Tg denote the bulk values) versus rescaled
film thickness h/h0. MD results for supported films (open circles), free-standing films
(shaded circles), and films confined between two smooth repulsive walls (filled circles)
are compared to the glass transition temperatures Tg (h) of three studies: (i) Monte
Carlo simulations of a lattice model for free-standing atactic polypropylene (PP) films
[177] (crosses). (ii) Experiments of supported atactic PS films of low molecular weight
(open squares) [132]. (iii) Experiments of supported, high-molecular weight PS films
[129] (stars). The solid and dashed line show Eqs. Eq. 7 and Eq. 8, respectively. The
vertical dotted line roughly indicates a film thickness of 10 nm. Figure adapted from
[171].

containing nanofillers [180, 181, 156]. The simulations reveal a complex relaxation
behaviour on approach to the glass transition of the confined glass former and also
report shifts of Tg , qualitatively similar to the trends sketched above for experiments.
In the following we illustrate these results by some examples.
Figure 9 compares the reduction of Tg with film thickness found in experiments on
supported PS films of low [132] and high molecular weight [129] with the Tg shifts
obtained from simulations of free-standing films [170, 171, 177], supported films [170,
171], and films confined between two substrates [166, 167]. The simulations study a
chemically realistic model of polypropylene [177] or flexible bead-spring models [166,
167, 170, 171], and use as a substrate completely smooth walls which are either purely
repulsive or weakly attractive. Despite the obvious differences between experimental
and computational systems—different polymers, absence or presence of a substrate,
etc.—a master curve for the reduction of Tg can be constructed, if Tg (h) is scaled by
the bulk value and h by a characteristic thickness h0 that depends on the nature of
the system, but only (very) weakly on molecular weight. This master curve allows
us to compare the different systems, which is instructive in several respects. First,
the close agreement of the PS data for low and high molecular weight suggests that

18
possible changes of the entanglement density [137, 182] or chain conformation [183,
182] in thin films are probably not responsible for the Tg depression. Furthermore,
the experimentally observed Tg shifts have stimulated several attempts to model this
phenomenon theoretically [132, 184, 185, 186, 187, 188, 189, 190, 191, 192, 193]. For
instance, based on a percolation model for slowly relaxing domains Long and Lequeux
[191] derive the formula (solid line in Fig. 9)
  h δ 
0
Tg (h) = Tg 1 − , (7)
h
originally suggested by Keddie et al as an empirical parametrization in their seminal
study on supported PS films [129]. An alternative parameterization was proposed by
Kim et al [136] (dashed line in Fig. 9)
Tg
Tg (h) = , (8)
1 + h0/h
and an attempt was undertaken to justify this formula by a viscoelastic capillary waves
model [132, 184]. Figure 9 shows both formulas and reveals that, for a typical range
of experimental film thicknesses, 15 . h/h0 . 300, it is hard to decide whether Eq. 7
or Eq. 8 is more accurate. However, including the simulation data for h . 10 nm,
Eq. 8 appears to provide the better description of the Tg shift. Therefore, Eq. 8 will be
employed later (Fig. 12) in an analysis of the local Tg of simulated polymer films.
The explanation for the Tg shift in the simulations rests upon the impact that the
boundaries exert, because the average behaviour of the film—and so its Tg —aggregates
contributions from all layers in the film, and these layers have distinct properties. This
can be illustrated by an analysis resolving structure and dynamics as a function of
distance from the boundaries. For flexible bead-spring models Fig. 10 shows two ex-
amples of such an analysis: the layer-resolved incoherent scattering function φqs (t, z) at
q∗ (maximum of S(q); cf. Fig. 11) for a polymer melt surrounding a highly faceted, but
nearly spherical filler particle with a structured and attractive surface (panel (a)) [180],
and φqs∗ (t, z) for a polymer film supported by a smooth, weakly attractive substrate
(panel (b)) [170].
For the polymer melt surrounding the nanofiller the scattering function, averaged
over all monomers (crosses in Fig. 10(a)), displays features unfamiliar from the bulk.
The α -relaxation seems to occur in two steps, as if there were two distinct processes,
a fast one corresponding to a bulk-like phase far away from the filler and a slow one
associated with interfacial relaxation. The layer-resolved analysis reveals that this in-
terpretation is misleading. The strongly stretched tail of the average correlator results
from the smooth gradient in the decay of φqs (t, z) which slows down on approach to the
filler particle. References [180, 181] show that the amplitude of this tail can be tuned
by the monomer-filler interaction. Strong attraction leads to a more pronounced tail;
vanishing attraction suppresses the tail. In the latter case, the shape of the scattering
function is bulk-like.
Additional insight into the slow relaxation of particles in contact with a structured
wall was obtained from studies of the self-part of the van Hove function [155, 194,
195]. For times in the α regime and T ≫ Tc , Gs (r,t) has a clear two-peak structure. The
first peak reflects particles that remain trapped in their cages, while the second peak,
located on a length scale corresponding to the wall structure, reveals that particles have
migrated to neighbouring wells. Evidence for this kind of “hopping motion” is found
for a binary LJ mixture [155], for a polymer melt adsorbed on a structured surface

19
 1.0 layer 1
layer 2
0.8 film
layer 1
φq(t,z)

0.6 ρ(z) layer 2

s

0.4
1

0.2
0
0 1 2 z/Rg
0.0
-2 -1 0 1 2 3 4
10 10 10 10 10 10 10
(a)
t
T=0.42
2 ∆z=2
ρ(z)

1.0
1
0.8
0
0 5 10 15 20 25
φq(t,z)

0.6 z
supported surface
s

0.4
bulk
0.2 z<10
free surface z=11
0.0 z>12
(b) 10
-2
10
-1
10
0 1
10 10 10 10
2 3 4
film
t
Figure 10: (a) Layer-resolved incoherent scattering function φqs (t, z) at T = 0.4 and
q = 7.08 (= maximum of S(q)) for a bead-spring polymer melt surrounding an icosa-
hedral filler particle. The chain length is N = 20, the melt density is ρ = 1 (dashed
horizontal line in the inset), and the filler attracts the monomers more strongly than
they attract each other in the bulk. The solid lines and the squares show φqs (t, z) for
different distances z from the surface of the filler particle. The location of the first
two layers is illustrated in the inset which depicts the monomer density profile ρ (z/Rg )
(Rg ≃ 2.17). The crosses indicate the average over all layers. Figure adapted from [32].
(b) φqs (t, z) at T = 0.42 and q = 6.9 (≈ maximum of S(q)) for a bead-spring polymer
melt in a supported film of thickness h = 20.3 (Tc ≈ 0.392). z denotes the distance from
the (left) wall. φqs (t, z) is obtained as an average
20 over all monomers which remain for
all times shown in a layer of width ∆z = 2 and centered at z. The average behaviour
of the film (average over all layers) is indicated by crosses and the bulk data by filled
circles. Inset: Monomer density profile ρ (z) versus z. The layers for which φqs (t, z) is
shown in the main figure, are indicated. Figure adapted from [170].
 S(q) kBT/γ q
2
0
T=0.44 10
h≈7 -1
q*~2π/σ 10
kBTρκT
-2
3 10
capillary -1 0 1
10 10 10
waves q
2
S(q)

1 bulk
center
free
0 kBTρκT wall
0 5 10 15 20 25
q
Figure 11: Static structure factor S(q) for a bead-spring melt (N = 10) in the bulk
(circles) and in supported film geometry (lines) at T = 0.44 (Tc ≈ 0.405 in the bulk;
Tc (h ≃ 7) ≈ 0.361 in the film [170]). For the film S(q) is shown respectively for layers
in film center, at wall and at the free surface. At the free surface the steep rise of
S(q) for small moduli q of the wave vector is due to capillary waves. For the bulk the
compressibility plateau kB T ρκT is indicated by a horizontal dotted line (κT denotes
the isothermal compressibility). Inset: Same data as in the main figure, but in a log-log
representation. The small-q behaviour expected from capillary wave theory is depicted
by a dashed line (γ is the surface tension).

[194] or for model of liquid toluene confined in cylindrical mesopores [195]. This
relaxation mechanism could be generic when a liquid may lock into registry with the
surface topography, leading to a mechanism of structural slowing down which coexists
with glassy arrest at low T [196, 197]
A similar locking is not possible at the free surface or a smooth, repulsive or weakly
attractive wall. In thin films, one therefore expects enhanced dynamics relative to the
bulk. Figure 10(b) shows that this expectation is borne out for flexible bead-spring
models. For these models the free and smooth interfaces also create enviroments for
nearby monomers which tend to reduce the cage effect: Local spatial correlations on
the scale q∗ of the maximum of S(q) are weaker in the films than in the bulk at the
same temperature (Fig. 11). This is an important contributing factor to the enhanced
monomer dynamics found at both interfaces.
Figure 10(b) also shows that the interface-induced deviations from bulk dynamics
continuously turn into bulk-like relaxation with increasing distance from the bound-

21
 bulk
τα Tc=0.405
4
10

10
3 T=0.42
T=0.44
τα(z,T)

2
T=0.48
10 T=0.52
T=1
1
10

0
10 0 5 10 15 20 25 30
z
Figure 12: Temperature dependence of the layer-resolved α relaxation time τα (z, T ) in
a supported polymer film (N = 10; Tc ≈ 0.405 in the bulk). The simulated system is
the same as in Fig. 10(b). The solid lines at the free surface show Eq. 11 (no adjustable
parameter). The horizontal dashed lines indicate the bulk values for τα at the respective
T . Figure adapted from [171].

22
aries. The range of this crossover grows on cooling. Figure 12 illustrates this point
by an analysis of the temperature dependence of the layer-resolved α relaxation time,
τα (z, T ). At high temperature τα (z, T ) slightly deviates from the bulk value near the
interfaces. With decreasing T the interface-induced enhancement of the dynamics in-
creasingly penetrates into the film and eventually propagates across the entire system
for sufficiently low T . This spatial dependence of τα (z, T ) can be modeled if two as-
sumptions are made: (i) The average Tc (h) of the film is given by Eq. 8 and can be
written as a “democratic average” [193] of the local Tc (z). That is,
Z h/2
Tc 2
Tc (h) = = dz Tc (z) , (9)
1 + h0/h h 0

which gives
Tc (1 + h0/z)
Tc (z) = . (10)
(1 + h0/2z)2
(ii) The second assumption is that the sole effect of the interface is to shift Tc from the
bulk value to Tc (z), whereas all other parameters determining the T dependence of τα
remain the same as in the bulk. Here we model this T dependence by the MCT power
law
τ0bulk
τα (z, T ) = . (11)
(T − Tc (z))γbulk
Figure 12 demonstrates that Eq. 11 yields a reasonable description of the increase of
τα from the free surface to the center of the film (solid lines in the figure) [171]. The
decrease of the α relaxation time on approach to the free surface therefore corresponds
to a decrease of the local glass transition temperature, in qualitative agreement with the
experimental results of [139]
The propagation of enhanced or reduced mobility from the boundary toward the
interior of the film has also been observed in other simulations on freely-standing
[158] and supported polymer films [164]. These studies carried out a cluster analy-
sis, of highly mobile monomers in the case of the freely-standing film and of immobile
monomers for the supported film. In both cases, it was found that clusters start at the
interface and penetrate into the film.
In summary, simulation studies suggest that confined (polymeric) liquids display
complex relaxation behaviour on approach to the glass transition because of the in-
terplay of bulk-like slowing down of the dynamics and interfacial effects. Interfaces
can enhance or retard the relaxation relative to the bulk. Enhanced relaxation may be
expected for smooth or free interfaces, whereas strong particle-substrate attraction or
particle caging in cavities of the substrate tend to slow down the dynamics. These
interface-induced perturbations smoothly transition from the boundaries to the interior
of the confined liquid. The range of this gradient grows on cooling so that the perbur-
bations can propagate across the entire liquid for sufficiently strong confinement or low
T . Similar interfacial effects may also be important for the analysis of other problems,
for instance, for solvent evaporation from (spincoated) polymer films [173, 174] or for
the modeling of the hydrodynamic boundary conditions in microfluidic devices [198].
Therefore, it appears that those theoretical approaches, which treat the interplay of
boundary effects and glassy slowing-down of the dynamics on the same (microscopic)
footing, are most likely to advance our understanding in this field. Folding in boundary
effects is certainly a major challenge [199, 196, 197]. Simulations of model systems—
as those reported here—should be helpful for the development of such theories.

23
4 Studies of the glassy state
Molecular simulations are intrinsically limited in terms of time scales. Therefore study-
ing the glassy state, in which the relaxation time scales are by definition extremely
large, could seem to be out of reach for this kind of numerical approach. Paradoxically,
however, the fact that the intrinsic time scales of the system are large brings the simu-
lations very close to actual experiments. The important fact is that, deep in the glassy
state, segmental relaxation times [200] greatly exceed both the experimental and the
simulation time scales. For both laboratory and computer experiments, a time window
is probed where segmental polymer motions are frozen and hence non-equilibrium phe-
nomena associated with the glassy state are observed. The main issue is then rather a
matter of sample preparation, which is usually done with much faster quenching rates
in computer experiments (cf. Sect. 3.1.1). As we will see below, it turns out that the
mechanical behaviour is not strongly affected, at least at a qualitative level, and that
simulations can therefore be used to analyse the structure and mechanical response of
the glassy state with some confidence.

4.1 Small deformations

We start our discussion by considering the small strain, elastic part of the mechanical
response of a polymer glass, which can be described at the macroscopic scale by linear
elasticity. It is now well recognized that, in spite of their homogeneity in density, glassy
materials are heterogeneous at the nanoscale in terms of their elastic properties. This
heterogeneity is reflected indirectly in some vibrational properties (e.g., the Boson peak
[201]), and is most obviously revealed by simulation, that allows one to compute elas-
tic constants and study elastic response at various scales [202]. For polymer glasses,
the first study of that kind was presented in refs [203, 204], following earlier studies
that indicated size dependent results for the elastic constants of nanometric systems
[205]. This study, which uses a simple definition of local elastic constants based on
a local calculation of the usual fluctuation formulae [204, 206], shows clearly that the
material is inhomogeneous at scales of the order of 5 to 10 monomeric sizes. Small
regions displaying negative elastic constants and therefore would be unstable if they
were not surrounded by regions with large local moduli. These aspects are not specific
to polymer systems, but can be observed in simple molecular or metallic glass formers
[207, 202], and using more sophisticated definitions of the local elastic moduli, which
points to their universal character. However, they are particularly important when the
polymer glass is cast in the form of a nanostructure, as the mechanical failure of such
structures will be strongly affected by the presence of elastic heterogeneities.
While the response of a polymer glass for deformations smaller than a few percent
can be described as elastic from a macroscopic viewpoint, a detailed microscopic stud-
ies show that irreversible rearrangements take place at low temperature even for very
small deformations [208]. This irreversible behaviour is responsible for a dissipative
part or the mechanical response, sometimes described as anelasticity. Ultimately, this
dissipation is associated with strongly localized plastic events that involve only a few
monomers, and can be thought as the precursors of the dissipative processes that take
place under larger deformation.
Two specificities of polymer glasses,already mentioned in Sect. 3.2.2, are the in-
terest for thin films and the ability to introduce various types of additives, from small
molecules to nanoparticle fillers. Relatively few studies are concerned with the thin
film modifications of the glassy state itself. Jain and de Pablo [209] showed that the

24
vibrational density of states is modified compared to the same glass in the bulk. In
particular, the usual excess of low frequency modes observed in glasses (the so called
boson peak) is enhanced in free standing thin films. This increase of the soft modes, low
frequency part of the spectrum is consistent with a negative shift of the glass transition
temperature. As the strength of the boson peak is in general, connected to a decreased
fragility (in the sense of Angell’s classification of glasses, see [12], this points to a
reduction in fragility with film thickness.
Many studies on the other hand have been devoted to the modification of the glassy
state under the influence of nanometric filler particles [180, 181, 210]. As discussed in
Sect. 3.2.2 (see Fig. 10), the vicinity of fillers was shown to modify the dynamics of
the polymer, with a slowing down and a shift of the glass transition temperature in the
case of attractive interactions. From the mechanical point of view, it was also shown
that the presence of attractive fillers results in the presence of a layer with enhanced
mechanical properties (higher local moduli) [211, 210] and modifies the structure of
the entanglement network [212]. The spatial extension of this layer is of the order
of a few monomer diameters and does not depend on the particle radius as soon as
the latter exceeds 2 to 3 monomer size. As a result the mechanical properties of the
composite are enhanced, however for a well dispersed composite this enhancement
remains moderate. Note that the existence of a glassy layer around the nanoparticles is
expected to affect the properties of the nanocomposite mostly in the region of the glass
transition, where the contrast between this layer and the rest of the polymer is maximal
[213, 214]. As for polymer films, the fragility (inferred from boson peak strength and
position) appears to be decreased with respect to the pure polymer.
Further studies of weak deformation include the influence of an external strain on
the microscopic dynamics [215] (in fact the latter study also extends to large, uniax-
ial deformations) , and the evolution of the creep compliance upon aging [216]. It is
found a finite strain rate accelerates the segmental dynamics -measured by the bond
orientation relaxation time- , whether the corresponding strain is positive (dilation) or
negative (compression). This symmetry between compression and dilation is indica-
tive of a stress induced modification of the potential energy landscape, with the applied
stress lowering the local barriers and accelerating the caging dynamics. Free volume
considerations, on the other hand, would not explain such a symmetry [215]. Acceler-
ation of the dynamics under load (sometimes described as ”mechanical rejuvenation”)
is also observed in the aging study of ref; [216], in which the creep compliance J(t,tw )
of a simulated polymer is also shown to be well accounted for by the classical descrip-
tion of Struik [217], in which the creep compliance isdescribed as a scaling function
µ R µ
J(te f f /tw ) where tw is the aging time, and te f f = 0t twt+t
w
′ dt ′ . This result again
shows that, in spite of the wide difference in time scales, the phenomenology of glassy
polymers is reproduced by simulation work. It should also be noticed [218] that the
relation between segmental mobility and strain rate is also dependent on the general
strain history, so that, in contrast to the usual assumptions of Eyring’s theory, no sim-
ple mechanical variable can relate to the local mobility. Recent theoretical approaches
[100], based on nonlinear Langevin equation description of segmental dynamics, go
beyond the simple Eyring description and account for this aging behaviour.
Finally, a rich and promising domain for simulation is the study of the complex
effects of small molecules on the glass transition and on the glassy state. Such small
molecules can be described as solvent, plasticizers or antiplasticizers, and in general
their presence leads to an acceleration of the dynamics and a decrease in the glass
transition More precisely, plasticization of the polymer means that the solvent does not

25
only decrease Tg , but also softens the polymer glass by reducing the elastic moduli.
Besides this normally encountered case, there are also systems in which the solvent
antiplasticizes the polymer. That is, the solvent decreases Tg , but increases the elastic
moduli of the polymer glass. The origin of this antiplasticizing effect has recently been
studied by simulations [219, 220], and appears to be due to a more efficient packing
associated with solvent molecules smaller than the monomers. Small molecule fit into
the ”holes” of the polymer melt, and increase the elastic stiffness. However, their
mobility also facilitates segmental mobility. Both for antiplasticizers and plasticizers,
a strongly heterogeneous dynamics of the solvent molecules has been reported [220,
123]. The corresponding dynamical correlation length is however decreased in the
presence of antiplasticizres, with a fragility of the system that is also decreased.

4.2 Large deformations and strain hardening

The mechanical properties of glassy polymers at large deformations were first inves-
tigated in the pioneering work of Röttler and Robbins [221, 222, 223, 224], and their
results were subsequently reproduced in a number of studies [76, 225]. Beyond the
peak stress, whose actual amplitude depends on the strain rate in a logarithmic way,
as in simple glasses, the stress strain behaviour depends on the type of sollicitation.
In pure shear or in simulations of a tensile test under triaxial conditions, the plastic
flow proceeds through a cavitation of cavities and subsequent formation of fibrils. The
drawing of these fibrils is essentially at constant stress. This phenomenon is discussed
in detail by Röttler [27] Under uniaxial, almost volume conserving conditions (Poisson
ratio close to 1/2), a marked strain hardening is observed, i.e. the plastic stress increases
with strain with a dependency that is close to linear. The corresponding deformation,
however, is very far from elastic, and essentially non recoverable.
This peculiar property of polymer glasses, which strongly contributes to their prac-
tical applications, is the existence of the strain hardening regime at large deformation.
Large strain-hardening effect prevents the strain localization and leads to a tough re-
sponse of the polymeric material; the material breaks only after a significant plastic
strain, as in the case for polycarbonate. For brittle polymers, as atactic polystyrene,
the strain-hardening effect is usually too weak, which leads to a brittle fracture due to
crazes formation within a few percent of strain. Understanding the microscopic ori-
gin of the strain-hardening effect provides a strategy for new tailor-made polymeric
materials.
The microscopic origin of this regime has been clarified only recently, thanks - to a
large extent - to numerical simulations. Both the generic aspects of this phenomenon,
and the way strain-hard and strain-soft polymer materials differ in terms of segmental
mobility and energetics, have been investigated by extensive MD simulations.
Before we discuss simulation results, let us recall the popular rubber elasticity mod-
els [226] which predicts that the hardening modulus Gh varies linearly with tempera-
ture T and entanglement density ρe , Gh = ρe kB T . Experimental values, however, are
two orders of magnitude larger, and, more important, show decrease with increasing
T [227, 228]. To take the dissipative nature of the plastic deformation into account
the rubber elasticity model describes the stress-strain relation in the strain-hardening
regime with the help of two contributions [226].

σ = σY + c2 (λ 2 − λ −1 ) (12)
The first part is a constant dissipative stress σY , due to the presence of energy

26
 initial configuration

uniaxial triaxial
tension tension

z−axis of
deformation

εz =0.5 εz =0.5
Figure 13: Snapshots of three configurations of a polymer glass: initial undeformed
state (up), uniaxially deformed configuration (left), and triaxially deformed configura-
tion (right). Both deformed configurations correspond to a strain of 50% along the z
axis of the simulation box. The simulation results are obtained from a flexible bead-
spring model (N = 100) at T = 0.2 (Tg ≈ 0.43). Figure taken from [76].

27
barriers. The second is the strain-dependent part, which is thought to be described by
rubber-elasticity theory and represents the strain-hardening effect. In this description
the strain-hardening modulus is not affected by thermally activated processes.
However, recent experiments demonstrate that such a description for the strain-
hardening part is invalid; the strain-hardening modulus has characteristics of a thermally-
activated process, and decreases for higher temperatures [229]. Secondly, at higher
strain rate the strain-hardening modulus increases [230], although for some polymers
the dependency on strain rate is rather weak [231]. This increase can be interpreted in a
very elementary manner within a barrier-crossing, Eyring like picture, which however
as discussed above is known to be oversimplified. Finally, the external pressure affects
the strain-hardening modulus as well. A higher external pressure leads to an increase
in the strain-hardening modulus [232]. Again, this behaviour is typical for thermally-
activated processes. All these three observations on the strain-hardening modulus are
not present within the classical rubber theory. The failure of rubber-elasticity theory
is due to the essential difference between a rubbery state, with many possible chain
conformations between cross-links, and a non-ergodic glassy state where chain confor-
mations are practically frozen, and transitions between different conformations are not
possible. Deformation of the polymer glass facilitates these transitions and is accom-
panied with the energy dissipation. The work for this irreversible energy dissipation is
reflected in a dissipative stress, which is absent in the rubber-elasticity theory. The dis-
sipative nature of a polymer strain-hardening is confirmed by experiments. It is found
that for PC and PS at large (> 15 − 30%) strains [233, 20, 234]) more work is dissipated
through heat than converted into internal energy.
The increase in stress for larger strains in combination with the dissipative nature
of the stress in the strain-hardening regime suggest that there is an increase in the rate
of energy dissipation, i.e., more energy per unit of strain is needed for more stretched
samples to stretch them further. This picture is supported by the computer simula-
tions of Hoy and Robbins [235], who showed that the dissipative stress increases with
larger strain and that at zero temperature the stress was directly correlated to the rate of
changes in Lennard-Jones (LJ) binding. Their more recent simulations of polymer toy
models also demonstrated that most of the stress at large strains is due to dissipation.
[235, 236].
If the rubber elasticity theory is invalid, what is then the polymer-specific part of
the strain-hardening modulus? Simulations of the mechanical deformation of simple
molecular glasses show no strain hardening [237]. Hence, for short polymer chains
the amount of strain hardening is expected to be small as well. Moreover, experiments
[228] and simulations [238] show that the strain hardening modulus is positively cor-
related with the entanglement density. Therefore one expects the strain hardening phe-
nomenon to disappear progressively as the chain length falls below the entanglement
threshold. Simulations show that this is indeed the case. In a model polymer glass, Hoy
and Robbins [235] demonstrated that there is a gradual increase in the strain-hardening
modulus as a function of chain length for chains up to about the entanglement length.
For longer chains saturation in the modulus occurs. Similarly, in MD simulations of
atactic polystyrene [48] only a weak strain hardening was observed for chain lengths of
80 monomers, below the experimentally observed entanglement length of about 128–
139 monomers [239, 240].
In the simulations of ref. [235] it was also observed that the strain hardening is
more correlated with the change in the end-to-end distance of the polymer chains than
with the change in the global sample size. If the sample size is decreasing while the
end-to-end distance does not, then the stress does not increase.

28
 After the fundamental aspects of strain hardening have been clarified using simula-
tions of simple, coarse-grained models, more detailed studies using realistic models are
required to understand the influence of chain architecture. For example, it is known that
polymers with a larger persistence length often have a higher strain-hardening modulus
[226]. It was even shown by molecular-dynamics (MD) simulations that if the persis-
tence length of a polyethylene-like polymer is artificially increased by changing the
trans-to-gauche ratio, the strain-hardening modulus of the resulting material increases
as well [241]. That example illustrates that the strain-hardening modulus depends on
the conformation of the chain, which is frozen in the glassy state. The changes in this
conformation are associated with plastic events, that involve a collective nanoscale seg-
mental dynamics of individual or neighbouring chains. Understanding the differences
in local mobility and dynamics for chemically different polymers under large defor-
mations, and how it relates to the different mechanical characteristics, is therefore an
important question, which has been studied in relatively few cases.
As an example of such studies, and of the possibilities offered by simulation, we
consider the extensive studies in refs. [242, 48, 232, 243] o̊n two glassy polymers that
vary greatly in their strain-hardening moduli, viz. polystyrene (PS) and polycarbonate
(PC, of which the modulus is more than a factor of two higher [244]). Molecular-
dynamics simulations [242, 48, 232] have reproduced these experimental findings qual-
itatively, with a strain-hardening modulus of polystyrene that is much lower than that of
polycarbonate, as shown in figure 14. They also allow one to identify the microscopic
mechanisms responsible for the observed difference in the strain-hardening modulus.
In particular, they show the rate of non-affine displacements (or local plastic events)
increases with larger strain, and that the increase is larger for PC. The non-affine dis-
placements of particles are due to restrictions and hindrances, in particular from cova-
lent and steric interactions. These restrictions are extremely important at the scale of
the covalent bond. If particles would displace affinely, then the equilibrium value of
this chemical bond would be excessively disturbed. To circumvent the bond stretch,
the bond vector will not move affinely with the deformation. At much larger length
scales the situation is different. For a long chain the internal conformation can be ad-
justed, while still obeying to the equilibrium length of the covalent bond. The effective
spring constant associated with the end-to-end distance is much less stiff. Moreover,
due to the glassy state the relaxation time of spontaneous rearrangements at the scale of
the whole chain greatly exceeds experimental and simulation time scales . Therefore,
the end-to-end distance cannot adjust back towards the equilibrium value by means of
spontaneous relaxations. Hence the end-to-end distance will follow the imposed defor-
mation much more affinely. So we expect a more affine response, as we probe larger
length scales.
This expectation is borne out by a detailed study of the affine character of the de-
formation of the polymer chain, as a function of the internal distance along the chain
[243]. At the scale of 100 chemical bonds, it is found that the deformation is essen-
tially affine up to about 15% strain for both PS and PC, while it is strongly non affine
already at this strain at the scale of 30 bonds, see Fig. 15. For larger strains the relative
deviation from affinity becomes larger for even long enough (100 backbone bonds)
chains. This effect is present in both PS and PC, the difference being in the magni-
tude of the effect. For an internal distance of 30 monomeric units, the deformation
of the polycarbonate chain is only 30% of the affine value at a strain of 50%, while
in polystyrene under the same conditions the chain deformation is 50% of the affine
value. This is because at the scale of the Kuhn length a chain cannot be stretched any
further. As the Kuhn length of polycarbonate is already larger than that of PS and the

29
 140

120

100
σ (MPa)

80

60

40

20 PS
PC
0
0 0.2 0.4 0.6 0.8 1
ε
eng

Figure 14: The von Mises equivalent true stress σ vs. strain εeng for PS and PC. Solid
lines are fits to Eq. 12. Fit range is εeng = 0.3–0.8. Note that the strain-hardening mod-
ulus Gh for PC (19 MPa) is almost twice that of PS (11 MPa), while their extrapolated
offset yield values σY are about the same (PC: 88 MPa; PS: 86 MPa). Adapted from
ref. [232]

30
total non-affine displacement of PC is to a large extent determined by the backbone
atoms, the increase in the effective stiffness length leads to an increase in non-affine
displacement, more energy dissipation and hence a higher strain-hardening effect. For
polystyrene the Kuhn length is small and the major part of non-affine displacement
is not caused by the backbone. Hence the expected increase in the effective stiffness
length during deformation does not lead to a substantial increase in plastic flow, so that
at moderate strains PS behaves more like a simple glass without strain hardening, as
opposed to polycarbonate.

5 Perspectives
The present review has attempted to describe some aspects of the recent progresses in
our understanding of glassy polymers that have been obtained from simulation work,
with a particular focus on microscopic dynamic, glass transition and mechanical prop-
erties. The results obtained in the past ten years show that the simulations have reached
a state of maturity that allows them to address and clarify important issues in the field,
in spite of the obvious limitations in terms of length and time scales. While the il-
lustrative examples we have chosen were relatively simple, they show that the general
phenomenology of the glass transition and the properties of the glassy state can be
accounted for using molecular simulations, in spite of the relatively small time scale
that can be considered in such simulations. We would like to close the discussion by
suggesting some directions for future research, which we believe will be soon - or
are already - within the capabilities of state of the art simulations, and correspond to
problems of practical or fundamental interest.
A major ingredient of the physics of glassy polymers - and more generally of glassy
systems - is the so called ”time temperature superposition principle” which assumes
that relaxation processes can be rescaled by using a single, temperature dependant re-
laxation time τα (T ). While this ”principle” is widely used to produce frequency depen-
dent relaxation function from data taken at various temperature, it is also well known to
be applicable at best in the frequency region corresponding to the α peak, presumably
only above Tc . It is a challenge to understand the molecular mechanisms underlying
this superposition principle and the deviations thereof [118]. A prominent feature of
polymeric glasses is the observation of secondary relaxations at higher frequency, the
Johari-Goldstein β processes (see Sect. 3.1.2). The associated relaxation times merge
with the α relaxation times typically at temperatures of 0.8Tg , and the associated time
scales are of the order of 10−6-10−7 s at such temperatures. Such time scales are be-
ginning to be in the range attainable by MD simulations, a fact that should allow one
to clarify the molecular origins of secondary relaxations, their relation to the existence
of side groups, to torsional modes, and other interpretations that have been put forward
[245]. Molecular simulation, by the flexibility it offers to block specific motions and
to control chain properties at various scales of coarse graining, should also allow one
to investigate the putative relations between these secondary relaxations, fragility, and
mechanical properties.
Nanocomposites and thin films have already been the subject of a number of simu-
lation studies (see Sect. 3.2.2). Still, the precise mechanisms that give rise to reinforce-
ment and nonlinear behaviour in nanocomposites, and to speed up of the dynamics in
thin films, remain to be elucidated in detail. In particular, for entangled polymers it
can be expected that the presence of interfaces modifies both the characteristics of the
entanglement network and the monomeric friction coefficients. The interplay between

31
atio_max1_P0006PS8x80_300K_X_id=100.mat+4 other ids+2 other
1.7
n=1
1.6 n=30
n=100
1.5
Cn(Heng)/Cn(0)

1.4

1.3

1.2

1.1

0.9
0 0.2 0.4 0.6 0.8 1
Heng

r_ratio__ownPC64x10_300K_X_id=100.mat+4 other ids+0 other de

1.7
n=1
1.6 n=30
n=100
1.5
Cn(Heng)/Cn(0)

1.4

1.3

1.2

1.1

1
32

0.9
0 0.2 0.4 0.6 0.8 1
Heng

Figure 15: The normalized characteristic ratio Cn′ (εeng ) = Cn (εeng )/Cn (0), which gives
these different aspects could, in principle, be elucidated in simulations of entangled
systems. Again, the simulation times needed to explore such phenomena are extremely
large, but, with the development of advanced simulation algorithms—as those alluded
to in Sect. 2—and the increase of computer power they are starting to be within the
attainable range.
Finally, we note that many polymer materials of practical interest display a mixed
structure, being either partially crystalline with an amorphous fraction, or chemically
inhomogeneous with phases (or microphases in the case of block copolymers) display-
ing different mechanical properties, e.g. glassy and rubbery. The properties of the
resulting nanocomposites have, up to now, received little attention from the standpoint
of molecular simulation (see however [108, 107, 246]). They are likely to be strongly
influenced, especially for large deformations, by the interplay between chain architec-
ture and material nanostructure. Another interesting phenomenon of that kind is strain
induced crystallisation, which plays an important role in the strain hardening of some
specific polymer. The capability of molecular simulation to describe polymer crystal-
lization has been recently demonstrated [7, 8, 247], so that the study of strain induced
crystallisation -as well as studies of semi crystalline phases appears to be an interesting
goal for the future.

Acknowledgments The work reported in this article was carried out in fruitful col-
laboration with M. Aichele, N. Balabaev, C. Bennemann, K. Binder, S.-H. Chong, C.
Donati, M. Fuchs, Y. Gebremichael, S. C. Glotzer, O. Lamé, A. Makke, M. Mareschal,
H. Meyer, M.A.J. Michels, S. Napolitano, J. J. de Pablo, G. Papakonstantopoulos, W.
Paul, M. Perez, S. Peter, R. Riggleman, B. Schnell, R. Seemann, F. W. Starr, F. Varnik,
T. Vettorel, B. Vorselaars, and M. Wübbenhorst. It is a great pleasure to thank all of
them. We presented results from the research of W. Paul, G. D. Smith, D. Bedrov
and coworkers. We are grateful that they quickly provided the figures requested. Our
simulations were made possible by generous grants of computer time at the IDRIS in
Orsay and at the NCF in Amsterdam, JB gratefully acknowledges financial support by
the IRTG “Soft condensed matter”.

References
[1] T. P. Lodge and M. Muthukumar. J. Phys. Chem., 100:13275, 1996.
[2] D. N. Theodorou. Chem. Eng. Sci., 67:5697–5714, 2007.
[3] J. P. Wittmer, P. Beckrich, H. Meyer, A. Cavallo, A. Johner, and J. Baschnagel.
Phys. Rev. E, 76:011803, 2007.
[4] M. Rubinstein and R. H. Colby. Polymer Physics. Oxford University Press,
Oxford, 2003.
[5] T. C. B. McLeish. Adv. Phys., 51:1379, 2002.

[6] A. E. Likhtman. J. Non-Newtonian Fluid Mech., 158:158–161, 2009.

[7] H. Meyer and F. Müller-Plathe. J. Chem. Phys., 115:7807–7810, 2001.
[8] H. Meyer and F. Müller-Plathe. Macromolecules, 35:1241–1252, 2002.

33
 [9] G. Strobl. The Physics of Polymers: Concepts for Understanding their Struc-
tures and Behavior. Springer, Berlin–Heidelberg, 1997.
[10] M. Muthukumar. Adv. Chem. Phys., 128:1–63, 2004.
[11] G. B. McKenna. Glass formation and glassy behavior. In C. Booth and C. Price,
editors, Comprehensive Polymer Science, volume 2, pages 311–362. Pergamon,
New York, 1986.
[12] C. A. Angell, K. L. Ngai, G. B. McKenna, P. F. McMillan, and S. W. Martin. J.
Appl. Phys., 88:3113, 2000.
[13] E. Donth. The Glass Transition. Springer, Berlin–Heidelberg, 2001.
[14] P. R. Sundararajan. Can. J. Chem., 63:103–110, 1985.
[15] K. Binder and W. Kob. Glassy Materials and Disordered Solids. World Scien-
tific, New Jersey–London, 2005.
[16] P. G. Debenedetti and F. H. Stillinger. Nature, 410:259, 2001.
[17] W. Götze. Complex Dynamics of Glass-Forming Liquids: A Mode-Coupling
Theory. Oxford University Press, Oxford, 2009.
[18] G. H. Fredrickson. Nature Mat., 7:261, 2008.
[19] J. Ferry. Viscoelastic Properties of Polymers. Wiley, New York, 1980.
[20] R. N. Haward and R. J. Young. The physics of glassy polymers. Chapman &
Hall, London, 2nd edition, 1997.
[21] I. M. Ward and D. W. Hadley. An Introduction to the Mechanical Properties of
Solid Polymers. Wiley, Chichester–New York, 1993.
[22] S. Granick, S. K. Kumar, E. J. Amis, M. Antonietti, A. C. Balazs, A. K.
Chakraborty, G. S. Grest, C. Hawker, P. Janmey, E. J. Kramer, R. Nuzzo, T. P.
Russell, and C. R. Safinya. J. Poly. Sci. B: Polymer Physics, 41:2755–2793,
2003.
[23] J. A. Forrest and K. Dalnoki-Veress. Adv. Coll. Interf. Sci., 94:167, 2001.
[24] M. Alcoutlabi and G. B. McKenna. J. Phys.: Condens. Matter, 17:R461, 2005.
[25] J. H. R. Clarke. Molecular dynamics of glassy polymers. In K. Binder, editor,
Monte Carlo and Molecular Dynamics Simulations in Polymer Science, pages
272–306. Oxford Univ. Press, New York, 1995.
[26] W. Paul and G. D. Smith. Rep. Prog. Phys., 67:1117, 2004.
[27] J. Rottler. J. Phys.: Condens. Matter, 21:463101, 2009.
[28] K Binder, J Baschnagel, W Kob, and W Paul. In Nielaba, P and Mareschal, M
and Ciccotti, G, editor, Bridging Time Scales: Molecular Simulations For The
Next Decade, volume 605 of Lecture Notes In Physics, pages 199–228, 2002.
Conference on Bridging the Time-Scale Gap, Constance, Germany, Sep, 2001.
[29] K. Binder. J. Non-crystalline Solids, 307:1–8, 2002.

34
[30] K. Binder, J. Baschnagel, and W. Paul. Progress In Polymer Science, 28:115–
172, 2003.
[31] K. Binder. Phys. Complex Systems (new Adv. Perspectives), 155:17–82, 2004.
[32] J. Baschnagel and F. Varnik. J. Physics-condensed Matter, 17:R851–R953,
2005.
[33] K. Binder. Introduction: General aspects of computer simulation techniques
and their applications in polymer science. In K. Binder, editor, Monte Carlo
and Molecular Dynamics Simulations in Polymer Science, pages 3–46. Oxford
University Press, New York, 1995.
[34] C. Abrams, L. Delle Site, and K Kremer. Multiscale computer simulations for
polymeric materials in bulk and near surfaces. In P. Nielaba, M. Mareschal, and
G. Cicotti, editors, Bridging Time Scales: Molecular Simulations for the Next
Decade, pages 143–164. Springer, Berlin, 2002.
[35] J. Baschnagel, K. Binder, P. Doruker, A. A. Gusev, O. Hahn, K. Kremer, W. L.
Mattice, F. Müller-Plathe, M. Murat, W. Paul, S. Santos, U. W. Suter, and
V. Tries. Adv. Polym. Sci., 152:41–154, 2000.
[36] F. Müller-Plathe. ChemPhysChem, 3:754, 2002.
[37] C. Peter and K. Kremer. Soft Matter, 5:4357–4366, 2009.
[38] D. G. Tsalikis, N. Lempesis, G. C. Boulougouris, and D. N. Theodorou. J. Phys.
Chem. B, 112:10619–10627, 2008.
[39] D. G. Tsalikis, N. Lempesis, G. C. Boulougouris, and D. N. Theodorou. J. Phys.
Chem. B, 112:10619–10627, 2008.
[40] A. Heuer. J. Phys.: Condens. Matter, 20:373101, 2008.
[41] R. A. L. Vallee, W. Paul, and K. Binder. J. Chem. Phys., 127:154903, 2007.
[42] R. A. L. Vallee, M. Van der Auweraer, W. Paul, and K. Binder. Phys. Rev. Lett.,
97, 2006.
[43] J. Colmenero, F. Alvarez, and A. Arbe. Phys. Rev. E, 65:041804, 2002.
[44] J. Colmenero, A. Arbe, F. Alvarez, M. Monkenbusch, D. Richter, B. Farago, and
B. Frick. J. Phys.: Condens. Matter, 15:S1127, 2003.
[45] A. V. Lyulin, N. K. Balabaev, and M. A. J. Michels. Macromolecules, 35:9595–
9604, 2002.
[46] A. V. Lyulin and M. A. J. Michels. Macromolecules, 35:1463–1472, 2002.
[47] A. V. Lyulin, N. K. Balabaev, and M. A. J. Michels. Macromolecules, 36:8574–
8575, 2003.
[48] A. V. Lyulin, B. Vorselaars, M. A. Mazo, N. K. Balabaev, and M. A. J. Michels.
Europhys. Lett., 71:618–624, 2005.
[49] A. V. Lyulin and M. A. J. Michels. Phys. Rev. Lett., 99:085504, 2007.

35
[50] G. D. Smith, D. Bedrov, and W. Paul. J. Chem. Phys., 121:4961, 2004.
[51] S. Krushev and W. Paul. Phys. Rev. E, 67:021806, 2003.
[52] S. Krushev, W. Paul, and G. D. Smith. Macromolecules, 35:4198, 2002.
[53] W. Paul, D. Bedrov, and G. D. Smith. Phys. Rev. E, 74:021501, 2006.

[54] G. D. Smith and D. Bedrov. J. Polym. Sci. Part B: Polym. Phys., 45:627, 2007.
[55] J. Colmenero, A. Narros, F. Alvarez, A. Arbe, and A. J. Moreno. J. Phys.:
Condens. Matter, 19:205127, 2007.
[56] A. Narros, A. Arbe, F. Alvarez, J. Colmenero, and D. Richter. J. Chem. Phys.,
128:224905, 2008.
[57] M. Vogel. Macromolecules, 41:2949, 2008.
[58] Walter Kob. J. Phys.: Condens. Matter, 11:R85–R115, 1999.
[59] T. C. Dotson, J. V. Heffernan, J. Budzien, K. T. Dotson, F. Avila, D. T. Limmer,
D. T. McCoy, J. D. McCoy, and D. B. Adolf. J. Chem. Phys., 128:184905, 2008.
[60] M. Bulacu and E. van der Giessen. Phys. Rev. E, page 011807, 2007.
[61] M. Bernabei, A. J. Moreno, and J. Colmenero. Phys. Rev. Lett., 101:255701,
2008.
[62] M. Bernabei, A. J. Moreno, and J. Colmenero. arXiv:0909.2128v1.
[63] J. Buchholz, W. Paul, F. Varnik, and K. Binder. J. Chem. Phys., 117:7364–7372,
2002.
[64] S. H. Chong, M. Aichele, H. Meyer, M. Fuchs, and J. Baschnagel. Phys. Rev. E,
76:051806, 2007.
[65] J. McCarty, I. Y. Lyubimov, and M. G. Guenza. J. Phys. Chem. B, 113:11876–
11886, 2009.
[66] T. Murtola, A. Bunker, I. Vattulainen, M. Deserno, and M. Karttunen. Phys.
Chem. Chem. Phys., 11:1869–1892, 2009.
[67] T. Strauch, L. Yelash, and W. Paul. Phys. Chem. Chem. Phys., 11:1942–1948,
2009.
[68] Daan Frenkel and Berend Smit. Understanding Molecular Simulation. Aca-
demic Press, London, 2nd edition, 2002.
[69] K. Binder, J. Horbach, W. Kob, W. Paul, and F. Varnik. J. Physics-condensed
Matter, 16:S429–S453, 2004.
[70] D. P. Landau and K. Binder. A Guide to Monte Carlo Simulations in Statistical
Physics. Cambridge University Press, Cambridge, 2009.
[71] A. Bunker and B. Dünweg. Phys. Rev. E, 63:016701, 2000.
[72] R. Yamamoto and W. Kob. Phys. Rev. E, 61:5473, 2000.

36
[73] C. De Michele and F. Sciortiono. Phys. Rev. E, 65:051202, 2002.
[74] Q. Yan, T. S. Jain, and J. J. de Pablo. Phys. Rev. Lett., 92:235701, 2004.
[75] M. Chopra, R. Malshe, A. S. Reddy, and J. J. de Pablo. J. Chem. Phys.,
128:144104, 2008.
[76] B Schnell. PhD thesis, University of Strasbourg, 2006.
[77] A. Soldera and N. Metatla. Phys. Rev. E, 74:061803, 2006.
[78] N. Metatla and A. Soldera. Macromolecules, 40:9690, 2007.
[79] K. Vollmayr, W. Kob, and K. Binder. J. Chem. Phys., 105:4714, 1996.
[80] K. Vollmayr, W. Kob, and K. Binder. J. Chem. Phys., 54:15808, 1996.
[81] R. Brüning and K. Samwer. Phys. Rev. B, 46:11318, 1992.
[82] Y. Ding, A. Kisliuk, and A. P. Sokolov. Macromolecules, 37:161–166, 2004.
[83] K. Kunal, C. G. Robertson, S. Pawlus, S. F. Hahn, and A. P. Sokolov. Macro-
molecules, 41:7232–7238, 2008.
[84] J. Hintermeyer, A. Herrmann, R. Kahlau, C. Goiceanu, and E. A. Rösler. Macro-
molecules, 41:9335–9344, 2008.
[85] A. L. Agapov and A. P. Sokolov. Macromolecules, 42:2877–2878, 2009.
[86] B. Lobe, J. Baschnagel, and K. Binder. J. Non-crystalline Solids, 172:384–390,
1994.
[87] J. Dudowicz, K. F. Freed, and J. F. Douglas. Adv. Chem. Phys., 137:125–222,
2008.
[88] K. S. Schweizer. Curr. Opinion Coll. & Interf. Sci., 12:297, 2007.
[89] A. S. Keys, A. R. Abate, S. C. Glotzer, and D. J. Durian. Nature Phys., 3:260,
2007.
[90] C. Bennemann, J. Baschnagel, W. Paul, and K. Binder. Computational Theoret-
ical Polymer Science, 9:217–226, 1999.

[91] M. Brodeck, F. Alvarez, A. Arbe, F. Juranyi, T. Unruh, O. Holderer, J. Colmen-

ero, and D. Richter. J. Chem. Phys., 130:094908, 2009.
[92] M. Zamponi, A. Wischnewski, M. Monkenbusch, L. Willner, D. Richter,
P. Falus, B. Farago, and M. Guenza. J. Phys. Chem. B, 112:16220–16229, 2008.
[93] H. Meyer, J. P. Wittmer, T. Kreer, P. Beckrich, A. Johner, J. Farago, and
J. Baschnagel. Eur. Phys. J. E, 26:25–33, 2008.

[94] Martin Aichele and Jörg Baschnagel. Eur. Phys. J. E, 5:229, 2001.
[95] W. Götze. J. Phys.: Condens. Matter, 11:A1, 1999.
[96] P. Mayer, K. Miyazaki, and D. R. Reichman. Phys. Rev. Lett., 97:095702, 2006.

37
 [97] B. Kim and K. Kawasaki. J. Stat. Mech., 2:P02004, 2008.
[98] G. Szamel. J. Chem. Phys., 127:084515, 2007.
[99] A. Andreanov, G. Biroli, and J.-L. Bouchaud. Europhys. Lett., 88:16001, 2009.
[100] K. Chen, E. J. Saltzman, and K. S. Schweizer. J. Phys.: Condens. Matter,
21:503101, 2009.
[101] Jean Pierre Hansen and Ian R. McDonald. Theory of Simple Liquids. Academic
Press, London, 1986.
[102] G. F. Signorini, J.-L. Barrat, and M. L. Klein. J. Chem. Phys., 92:1294–1303,
1990.
[103] Walter Kob and Hans C. Andersen. Phys. Rev. E, 51:4626–4641, 1995.

[104] M. Sperl. Phys. Rev. E, 68:031405, 2003.

[105] E. Zaccarelli and W. C. K. Poon. PNAS, 106:15203–15208, 2009.
[106] A. J. Moreno and J. Colmenero. J. Chem. Phys., 124:184906, 2006.
[107] A. J. Moreno and J. Colmenero. J. Phys.: Condens. Matter, 19:466112, 2007.
[108] J. Colmenero and A. Arbe. Soft Matter, 3:1474, 2007.
[109] S.-H. Chong and M. Fuchs. Phys. Rev. Lett., 88:185702, 2002.
[110] M. Aichele, S. H. Chong, J. Baschnagel, and M. Fuchs. Phys. Rev. E, 69:061801,
2004.
[111] E. Flenner and G. Szamel. Phys. Rev. E, 72:011205, 2005.

[112] E. Flenner and G. Szamel. Phys. Rev. E, 72:031508, 2005.

[113] G. Foffi, W. Götze, F. Sciortino, P. Tartaglia, and Th. Voigtmann. Phys. Rev. E,
69:011505, 2004.
[114] Th. Voigtmann, A. M. Puertas, and M. Fuchs. Phys. Rev. E, 70:061506, 2004.
[115] P. Chaudhuri, L. Berthier, and W. Kob. Phys. Rev. Lett., 99:060604, 2007.
[116] S. Merabia and D. Long. Eur. Phys. J. E, 9:195, 2002.
[117] S. K. Kumar, G. Szamel, and J. F. Douglas. J. Chem. Phys., 124:214501, 2006.
[118] A. P. Sokolov and K. S. Schweizer. Phys. Rev. Lett., 102:248301, 2009. (Com-
ment: Phys. Rev. Lett., 103:159801, 2009. Reply: Phys. Rev. Lett., 103:159802,
2009.).
[119] M. D. Ediger. Annu. Rev. Phys. Chem., 51:99–128, 2000.
[120] S. C Glotzer. J. Non-Cryst. Solids, 274:342–355, 2000.
[121] L. Berthier, G. Biroli, J.-P. Bouchaud, W. Kob, K. Miyazaki, and D. R. Reich-
man. J. Chem. Phys., 126:184503, 2007.

38
[122] L. Berthier, G. Biroli, J.-P. Bouchaud, W. Kob, K. Miyazaki, and D. R. Reich-
man. J. Chem. Phys., 126:184504, 2007.
[123] S. Peter, H. Meyer, and J. Baschnagel. European Phys. J. E, 28:147–158, 2009.
[124] L. Hartmann, K. Fukao, and F. Kremer. Molecular dynamics in thin polymer
films. In F. Kremer and A. Schönhals, editors, Broadband Dielectric Spec-
troscopy, pages 433–473. Springer, Berlin, Heidelberg, New York, 2003.
[125] F. Kremer, A. Huwe, A. Schönhals, and S. A. Różański. Molecular dynamics in
confining space. In F. Kremer and A. Schönhals, editors, Broadband Dielectric
Spectroscopy, pages 171–224. Springer, Berlin, Heidelberg, New York, 2003.
[126] C. B. Roth and J. R. Dutcher. Mobility on different length scales in thin polymer
films. In J. R. Dutcher and A. G. Marangoni, editors, Soft Materials: Structure
and Dynamics, pages 1–38. Marcel Dekker, New York, 2005.
[127] C. Alba-Simionesco, B. Coasne, G. Dosseh, G. Dudziak, K. E. Gubbins, R. Rad-
hakrishnan, and M. Sliwinska-Bartkowiak. J. Phys.: Condens. Matter, 18:R15,
2006.
[128] O. K. C. Tsui. Anomalous dynamics of polymer films. In O. K. C. Tsui and T. P.
Russell, editors, Polymer Thin Films, page 267. World Scientific, Singapore,
2008.

[129] J. L. Keddie, R. A. L. Jones, and R. A. Cory. Europhys. Lett., 27:59, 1994.

[130] J. S. Sharp and J. A. Forrest. Phys. Rev. E, 67:031805, 2003.
[131] J. A. Forrest. Eur. Phys. J. E, 8:261, 2002.
[132] S. Herminghaus, K. Jacobs, and R. Seemann. Eur. Phys. J. E, 5:531, 2001.
[133] Y. Grohens, L. Hamon, G. Reiter, A. Soldera, and Y. Holl. Eur. Phys. J. E, 8:217,
2002.
[134] K. Fukao and Y. Miyamoto. Phys. Rev. E, 61:1743, 2000.
[135] K. Fukao and Y. Miyamoto. Phys. Rev. E, 64:011803, 2001.
[136] J. H. Kim, J. Jang, and W.-C. Zin. Langmuir, 17:2703, 2001.

[137] O. K. C. Tsui and H. F. Zhang. Macromolecules, 34:9139, 2001.

[138] D. S. Fryer, R. D. Peters, E. J. Kim, J. E. Tomaszewski, J. J. de Pablo, P. F.
Nealey, C. C. White, and W. L. Wu. Macromolecules, 34:5627–5634, 2001.
[139] C. J. Ellison and J. M. Torkelson. Nature Mat., 2:695, 2003.
[140] C. Ellison, M. Mundra, and J. Torkelson. Macromolecules, 38:1767, 2005.
[141] R. D. Priestley, L. J. Broadbelt, J. M. Torkelson, and K. Fukao. Phys. Rev. E,
75:061806, 2007.
[142] V. Lupaşcu, S. J. Picken, and M. Wübbenhorst. Journal of Non-Crystalline
Solids, 352:5594, 2006.

39
[143] S. Napolitano and M. Wübbenhorst. J. Phys. Chem. B, 111:9197, 2007.
[144] M. Yu. Efremov, E. A. Olson, M. Zhang, Z. Zhang, and L. H. Allen. Macro-
molecules, 37:4607, 2004.
[145] H. Huth, A. A. Minakov, A. Serghei, F. Kremer, and C. Schick. Eur. Phys. J.
Special Topics, 141:153, 2007.
[146] Z. Fakhraai and J. A. Forrest. Science, 319:600, 2008.
[147] Z. Fakhraai and J. A. Forrest. Phys. Rev. Lett., 95:025701, 2005.
[148] K. Fukao. Euro. Phys. J. E, 12:119–125, 2003.
[149] S. Napolitano, V. Lupaşcu, and M. Wübbenhorst. Macromolecules, 41:1061,
2008.

[150] R. Lund, L. Willner, A. Alegrı́a, J. Colmenero, and D. Richter. Macromolecules,

41:511, 2008.
[151] A. Serghei, M. Tress, and F. Kremer. Macromolecules, 39:9385, 2006.
[152] A. Serghei, L. Hartmann, and F. Kremer. J. Non-Cryst. Solids, 353:4330, 2007.
[153] P. Scheidler, W. Kob, and K. Binder. Europhysics Lett., 59:701–707, 2002.
[154] P. Scheidler, W. Kob, and K. Binder. Computer Simulations Surfaces Interfaces,
114:297–312, 2003.
[155] P. Scheidler, W. Kob, and K. Binder. J. Phys. Chem. B, 108:6673–6686, 2004.
[156] P. Gallo, R. Pellarin, and M. Rovere. Europhys. Lett., 57:212, 2002.
[157] J. A. Torres, P. F. Nealey, and J. J. de Pablo. Phys. Rev. Lett., 85:3221–3224,
2000.
[158] T. S. Jain and J. J. de Pablo. Phys. Rev. Lett., 92:155505, 2004.
[159] T. S. Jain and J. J. de Pablo. Macromolecules, 35:2167–2176, 2002.
[160] T. R. Bohme and J. J. de Pablo. J. Chem. Phys., 116:9939–9951, 2002.
[161] K. Yoshimoto, T. S. Jain, P. F. Nealey, and J. J. de Pablo. J. Chem. Phys.,
122:144712, 2005.
[162] R. A. Riggleman, K. Yoshimoto, J. F. Douglas, and J. J. de Pablo. Phys. Rev.
Lett., 97:045502, 2006.
[163] A. R. C. Baljon, M. H. M. Van Weert, R. Barber DeGraaff, and R. Khare. Macro-
molecules, 38:2391, 2005.
[164] A. R. C. Baljon, J. Billen, and R. Khare. Phys. Rev. Lett., 93:255701, 2004.
[165] H. Morita, K. Tanaka, T. Kajiyama, T. Nishi, and M. Doi. Macromolecules,
39:6233, 2006.
[166] F. Varnik, J. Baschnagel, and K. Binder. Phys. Rev. E, 65:021507, 2002.

40
[167] F. Varnik, J. Baschnagel, and K. Binder. European Phys. J. E, 8:175–192, 2002.
[168] F. Varnik and K. Binder. J. Chem. Phys., 117:6336–6349, 2002.
[169] F. Varnik, J. Baschnagel, K. Binder, and M. Mareschal. European Phys. J. E,
12:167–171, 2003.
[170] S. Peter, H. Meyer, and J. Baschnagel. J. Polymer Science Part B-polymer Phys.,
44:2951–2967, 2006.
[171] S. Peter, H. Meyer, J. Baschnagel, and R. Seemann. J. Physics-condensed Mat-
ter, 19:205119, 2007.
[172] S. Peter, S. Napolitano, H. Meyer, M. Wubbenhorst, and J. Baschnagel. Macro-
molecules, 41:7729–7743, 2008.
[173] S. Peter, H. Meyer, and J. Baschnagel. J. Chem. Phys., 131:014902, 2009.
[174] S. Peter, H. Meyer, and J. Baschnagel. J. Chem. Phys., 131:014903, 2009.
[175] D. Hudzinskyy, A. V. Lyulin, A. R. C. Baljon, N. K. Balabaev, and M. A. J.
Michels. J. Polym. Sci. B, submitted.
[176] O. Alexiadis, V. G. Mavrantzas, R. Khare, J. Beckers, and A. R. C. Baljon.
Macromolecules, 41:987–996, 2008.

[177] G. Xu and W. L. Mattice. J. Chem. Phys., 118:5241, 2003.

[178] K. F. Mansfield and D. N. Theodorou. Macromolecules, 24:6283–6294, 1991.
[179] E. Manias, V. Kuppa, D.-K. Yang, and D. B. Zax. Colloids Surf. A, 187–188:509,
2001.
[180] F. W. Starr, T. B. Schroder, and S. C. Glotzer. Phys. Rev. E, 64:021802, 2001.
[181] F. W. Starr, T. B. Schroder, and S. C. Glotzer. Macromolecules, 35:4481–4492,
2002.
[182] H. Meyer, T. Kreer, A. Cavallo, J.P. Wittmer, and J. Baschnagel. Eur. Phys. J.
Special Topics, 141:167–172, 2006.
[183] A. Cavallo, M. Müller, J. P. Wittmer, A. Johner, and K. Binder. J. Phys.: Con-
dens. Matter, 17:S1697, 2005.
[184] S. Herminghaus. Eur. Phys. J. E, 8:237, 2002.
[185] P. G. de Gennes. Eur. Phys. J. E, 2:201, 2000.
[186] J. D. McCoy and J. G. Curro. J. Chem. Phys., 116:9154, 2002.
[187] T. M. Truskett and V. Ganesan. J. Chem. Phys., 119:1897, 2003.

[188] J. Mittal, P. Shah, and T. M. Truskett. J. Phys. Chem. B, 108:19769, 2004.

[189] T. S. Chow. J. Phys.: Condens. Matter, 14:L333, 2002.
[190] K. L. Ngai. Eur. Phys. J. E, 12:93, 2003.

41
[191] D. Long and F. Lequeux. Eur. Phys. J. E, 4:371, 2001.
[192] S. Merabia, P. Sotta, and D. Long. Eur. Phys. J. E, 15:189, 2004.
[193] J. E. G. Lipson and S. T. Milner. Eur. Phys. J. B, 72:133, 2009.
[194] G. D. Smith, D. Bedrov, and O. Borodin. Phys. Rev. Lett., 90:226103, 2003.
[195] V. Teboul and C. Alba-Simionesco. J. Phys.: Condens. Matter, 14:5699, 2002.
[196] V. Krakoviack. Phys. Rev. E, 75:031503, 2007.
[197] V. Krakoviack. Phys. Rev. E, 79:061501, 2009.
[198] J. Servantie and M. Müller. Phys. Rev. Lett., 101:026101, 2008.
[199] M. Fuchs and K. Kroy. J. Phys.: Condens. Matter, 14:9223, 2002.
[200] C. A. Angell. Polymer, 38:6261, 1997.
[201] A. Tanguy, J. P. Wittmer, F. Leonforte, and J.-L. Barrat. Phys. Rev. B, 66:174205,
2002.
[202] Michel Tsamados, Anne Tanguy, Chay Goldenberg, and Jean-Louis Barrat.
Phys. Rev. E, 80(2, Part 2):026112, 2009.
[203] K Yoshimoto, TS Jain, KV Workum, PF Nealey, and J.J. de Pablo. Mechanical
heterogeneities in model polymer glasses at small length scales. Phys. Rev. Lett.,
93(17):175501, OCT 22 2004.
[204] K Yoshimoto, GJ Papakonstantopoulos, JF Lutsko, and JJ de Pablo. Phys. Rev.
B, 71(18):184108, 2005.
[205] K Van Workum and J.J. de Pablo. Computer simulation of the mechanical prop-
erties of amorphous polymer nanostructures. Nano Letters, 3(10):1405–1410,
OCT 2003.
[206] Jean-Louis Barrat. In Garrido, PL and Hurtado, PI and Marro, J, editor, Model-
ing And Simulation Of New Materials, AIP Conference Proceedings 1091, Gar-
rido et al ed., volume 1091 of AIP Conference Proceedings, pages 79–94. 2009.
10th Granada Seminar, Granada, SPAIN, SEP 15-19, 2008.
[207] S. G. Mayr. Phys. Rev. B, 79(6):060201, 2009.
[208] G. J. Papakonstantopoulos, R. A. Riggleman, J. L. Barrat, and J. J. de Pablo.
Phys. Rev. E, 77:041502, 2008.
[209] T. S. Jain and J. J. de Pablo. J. Chem. Phys., 120:9371–9375, 2004.
[210] G. J. Papakonstantopoulos, M. Doxastakis, P. F. Nealey, J. L. Barrat, and J. J.
de Pablo. Phys. Rev. E, 75:031803, 2007.
[211] G. J. Papakonstantopoulos, K. Yoshimoto, M. Doxastakis, P. F. Nealey, and J. J.
de Pablo. Phys. Rev. E, 72:031801, 2005.
[212] Robert A. Riggleman, Gregory Toepperwein, George J. Papakonstantopoulos,
Jean-Louis Barrat, and Juan J. de Pablo. J. Chem. Phys., 130(24):244903, JUN
28 2009.

42
[213] J Berriot, H Montes, F Lequeux, D Long, and P Sotta. Europhysics Lett.,
64(1):50–56, 2003.
[214] H Montes, F Lequeux, and J Berriot. Macromolecules, 36(21):8107–8118, 2003.
[215] R. A. Riggleman, H. N. Lee, M. D. Ediger, and J. J. De Pablo. Phys. Rev. Lett.,
99:215501, 2007.
[216] M. Warren and J. Rottler. Phys. Rev. E, 76:031802, 2007.
[217] L.C.E. Struik. Physical aging in amorphous polymers and other materials. El-
sevier, Amsterdam, 1978.
[218] Hau-Nan Lee, Robert A. Riggleman, Juan J. de Pablo, and M. D. Ediger. Macro-
molecules, 42(12):4328–4336, 2009.
[219] R. A. Riggleman, J. F. Douglas, and J. J. De Pablo. Phys. Rev. E, 76:011504,
2007.
[220] R. A. Riggleman, J. F. Douglas, and J. J. de Pablo. J. Chem. Phys., 126:234903,
2007.
[221] J. Rottler and M. O. Robbins. Phys. Rev. Lett., 89:148304, 2002.
[222] J Rottler and MO Robbins. Phys. Rev. E, 68:011507, 2003.

[223] J Rottler and MO Robbins. Phys. Rev. E, 68(1, Part 1):011801, 2003.
[224] J Rottler and MO Robbins. Phys. Rev. Letters, 95(22):225504, 2005.
[225] Ali Makke, Michel Perez, Olivier Lame, and Jean-Louis Barrat. J. Chem. Phys.,
131(1):014904, 2009.
[226] R. N. Harward. Macromolecules, 26:5860–5869, 1993.
[227] E. J. Kramer. J. Polym. Sci.: Polym. Phys., 43:3369–3371, 2005.
[228] H. G. H. van Melick, L. E. Govaert, and H. E. H. Meijer. Polymer, 44:2493–
2502, 2003.
[229] L. E. Govaert and T. A. Tervoort. J. Polym. Sci.: Polym. Phys., 42:2041–2049,
2004.
[230] S. S. Sarva, S. Deschanel, M. C. Boyce, and W. N. Chen. Polymer, 48:2208–
2213, 2007.
[231] M. C. Boyce and E. M. Arruda. Polym. Eng. Sci., 30:1288–1298, 1990.
[232] B. Vorselaars, A. V. Lyulin, and M. A. J. Michels. Macromolecules, 42:5829–
5842, 2009.
[233] O. B. Salamatina, S. I. Nazarenko, S. N. Rudnev, and E. F. Oleinik. Mech.
Compos. Mater, 24:721–725, 1988.
[234] E. F. Oleinik, S. N. Rudnev, O. B. Salamatina, S. V. Shenogin, M. I. Kotelyan-
skii, T. V. Paramzina, and S. I. Nazarenko. E-Polymers, 29:029, 2006.

43
[235] R. S. Hoy and M. O. Robbins. Phys. Rev. Lett., 99:117801, 2007.
[236] R. S. Hoy and M. O. Robbins. Phys. Rev. E, 77:031801, 2008.
[237] M Utz, PG Debenedetti, and FH Stillinger. J. Chem. Phys., 114(22):10049–
10057, 2001.
[238] R. S. Hoy and M. O. Robbins. J. Polym. Sci.: Polym. Phys., 44:3487–3500,
2006.
[239] L. J. Fetters, D. J. Lohse, D. Richter, T. A. Witten, and A. Zirkel. Macro-
molecules, 27:4639–4647, 1994.
[240] L. J. Fetters, D. J. Lohse, S. T. Milner, and W. W. Graessley. Macromolecules,
32:6847–6851, 1999.
[241] J. I. McKechnie, R. N. Haward, D. Brown, and J. H. R Clarke. Macromolecules,
26:198–202, 1993.
[242] A. V. Lyulin, N. K. Balabaev, M. A. Mazo, and M. A. J. Michels. Macro-
molecules, 37:8785–8793, 2004.
[243] B. Vorselaars, A. V. Lyulin, and M. A. J. Michels. J. Chem. Phys., 130:074905,
2009.
[244] H. G. H. van Melick, L. E. Govaert, and H. E. H. Meijer. Polymer, 44:3579–
3591, 2003.
[245] J.E. Mark. Physical Properties of Polymers Handbook. Springer, Berlin, 2007.
[246] A. J. Moreno and J. Colmenero. Phys. Rev. Lett., 100:126001, 2008.
[247] T. Vettorel, H. Meyer, J. Baschnagel, and M. Fuchs. Phys. Rev. E, 75:041801,
2007.

44