You are on page 1of 12

Eur. Phys. J.

E 28, 147–158 (2009)


DOI 10.1140/epje/i2008-10372-9
THE EUROPEAN
PHYSICAL JOURNAL E
Regular Article

MD simulation of concentrated polymer solutions: Structural


relaxation near the glass transition

S. Peter, H. Meyer, and J. Baschnagela

Institut Charles Sadron, CNRS, 23 rue du Loess - BP 84047, 67034 Strasbourg Cedex 2, France

Received 27 May 2008


Published online: 15 October 2008 – 
c EDP Sciences / Società Italiana di Fisica / Springer-Verlag 2008

Abstract. We examine by molecular dynamics simulations the relaxation of polymer-solvent mixtures close
to the glass transition. The simulations employ a coarse-grained model in which polymers are represented
by bead-spring chains and solvent particles by monomers. The interaction parameters between polymer
and solvent are adjusted such that mixing is favored. We find that the mixtures have one glass transition
temperature Tg or critical temperature Tc of mode-coupling theory (MCT). Both Tg and Tc (> Tg ) decrease
with increasing solvent concentration φS . The decrease is linear for the concentrations studied (up to
φS = 25%). Above Tc we explore the structure and relaxation of the polymer-solvent mixtures on cooling.
We find that, if the polymer solution is compared to the pure polymer melt at the same T , local spatial
correlations on the length scale of the first peak of the static structure factor S(q) are reduced. This
difference between melt and solution is largely removed when comparing the S(q) of both systems at similar
distance to the respective Tc . Near Tc we investigate dynamic correlation functions, such as the incoherent
intermediate scattering function φsq (t), mean-square displacements of the monomers and solvent particles,
two non-Gaussian parameters, and the probability distribution P (ln r; t) of the logarithm of single-particle
displacements. In accordance with MCT we find, for instance, that φsq (t) obeys the time-temperature
superposition principle and has α relaxation times τqs which are compatible with a power law increase close
(but not too close) to Tc . In divergence to MCT, however, the increase of τqs depends on the wavelength
q, small q values having weaker increase than large ones. This decoupling of local and large-length scale
relaxation could be related to the emergence of dynamic heterogeneity at low T . In the time window of the
α relaxation an analysis of P (ln r; t) reveals a double-peak structure close to Tc . The first peak correponds
to “slow” particles (monomer or solvent) which have not moved much farther than 10% of their diameter in
time t, whereas the second occurs at distances of the order of the particle diameter. These “fast” particles
have succeeded in leaving their nearest-neighbor cage in time t. The simulation thus demonstrates that
large fluctuations in particle mobility accompany the final structural relaxation of the cold polymer solution
in the vicinity of the extrapolated Tc .

PACS. 61.25.H- Macromolecular and polymers solutions; polymer melts – 61.20.Ja Computer simulation
of liquid structure

1 Introduction free-volume model or the Gibbs-DiMarzio entropy the-


ory (see [2] for a review), provide Tg as a function of
Mixing a small-molecule solvent with a glass-forming poly- φS . Recently, however, it has been pointed out that these
mer melt usually leads to a depression of the glass transi- expressions require re-examination. The glass transition
tion temperature Tg relative to the value of the neat poly- reflects the segmental mobilities in the mixture, which
mer melt [1,2]. This plasticization effect [3] has elicited should be determined by the local instead of the aver-
strong interest in deriving predictive expressions for Tg in age concentration [6]. Differences between the local con-
terms of fundamental compositional parameters (see [2,4], centration and φS arise from “self-concentration” effects
references therein, and [5]). An important compositional —due to chain connectivity the local environment around
parameter is the concentration of solvent φS , and com- a monomer is enriched in monomers relative to the aver-
monly employed theoretical expressions, based e.g. on the age macroscopic concentration— and composition fluctu-
ations. It is possible to turn these ideas into a computa-
a
e-mail: baschnag@ics.u-strasbg.fr tional model [5]. The results of this model compare well
148 The European Physical Journal E

with differential-scanning calorimetry (DSC) experiments namics in a supercooled concentrated solution at a fixed
for polystyrene-solvent mixtures, even if the DSC traces composition [24,25]. Here we want to complement these
display two distinct Tg ’s whose position and width shift studies by MD simulations of a bead-spring model for con-
with φS [5]. centrated polymer solutions in which the interaction pa-
The approach sketched above focuses on modeling the rameters are adjusted to mimic polystyrene in toluene.
composition dependence of Tg . It does not directly ad- Due to the similarity of the styrene monomer and toluene
dress the segmental dynamics. Different relaxation times the difference between the constituents of the solution re-
of the polymer and solvent are rather a consequence of the sults from the interaction strength and not from size dis-
composition-dependent Tg . It is also imaginable to tackle parity (contrary to the work of Refs. [10,11,24,25]). For
the problem vice versa. The composition dependence of this system we explore the dependence of the structural
Tg results from the modeling of the dynamics of the mix- relaxation on composition as the mixture is cooled toward
ture. This is the approach naturally taken in computer the critical temperature Tc of MCT.
simulations of model glass formers [7–11] or in theoret- Our article is arranged as follows. We introduce the
ical treatments based e.g. on the mode-coupling theory simulation model in Section 2. In Section 3 we present re-
(MCT) [12]. sult for dynamic correlation functions, such as the incoher-
Within MCT the effect of composition changes on the ent intermediate scattering function φsq (t), mean-square
structural relaxation of a binary mixture of hard spheres displacements of the monomers and solvent particles, two
has been explored [12]. For size ratios δ of the small and non-Gaussian parameters, and the probability distribu-
large components varying between 0.5 and 1 two scenar- tion P (ln r; t) of the logarithm of single-particle displace-
ios are found. For small-size disparity (e.g. for δ = 0.8), ments, which show that the solvent plasticizes the poly-
increasing the concentration of small particles leads to a mers. The final section (Sect. 4) summarizes our results.
slowing down of the relaxation. In this case, mixing favors
glass formation relative to the pure system. For larger size
disparity (δ  0.65), however, mixing impedes glass for- 2 Simulation with explicit solvent: Model and
mation and a plasticizing effect is found. Physically, this technique
plasticization is interpreted as a consequence of a deple-
tion attraction induced between the large particles by the 2.1 Description of the simulation model
small ones, similar to the behavior that can be observed
in colloid-polymer mixtures [13–15]. We examine a coarse-grained model of polymer-solvent
These theoretical predicitions for small to moderate mixtures, which is an extension of the model studied previ-
size disparity are confirmed by molecular dynamics (MD) ously for undiluted polymer melts and polymer films [26–
simulations [7]. Recently, these numerical investigations 28]. The polymers are represented by monodisperse bead-
have been extended to a binary (soft-sphere) mixture with spring chains each having N monomers of mass mPP , and
a very large size disparity (δ = 0.4) [9]. This large disparity the solvent is modeled as single monomers of mass mSS .
entails strongly different relaxation times for both compo- Non-bonded monomer-monomer as well as monomer-
nents in the mixture, the small particles being the fast solvent and solvent-solvent interactions are given by a
ones, and leads to anomalous relaxation behavior. Unlike truncated and shifted Lennard-Jones (LJ) potential with
the typical two-step relaxation found for smaller-size dis- cutoff radius rc ,
parity [7,12] and in other glass-forming liquids [16–19] the  
long-time decay of dynamic correlation functions may be σAB 12  σAB 6
ULJ (r) = 4AB
AB
− − C, r ≤ rc .
very stretched, even logarithmic, for both small and large r r
particles depending on composition. For the small par- (1)
ticles these anomalies are interpreted as a result of two Here C is a constant added so that the potential vanishes
competing mechanisms for dynamic arrest: the bulk-like continuously at r = rc . The indices A and B refer to the
slowing down of the small particles due to the proximity different types of particles, monomers (A = P) and solvent
of their glass transition and the confinement imposed by (B = S).
the much slower large particles. In the following, we present our data in LJ units rel-
Reference [9] suggests that these mechanisms also op- ative to the polymer potential, i.e., we set PP = 1,
erate in other systems. For instance, dynamic anoma- σPP = 1, and mPP = 1. Then, temperature T is measured
lies similar to those of the binary mixture with δ = 0.4 in units of PP /kB (Boltzmann constant kB = 1), and time
are also reported for thermodynamically miscible polymer t in units of τLJ = (mPP σPP2
/PP )1/2 . In these units, our
blends [10,11] if the asymmetry in segmental mobilities is choice for the solvent-solvent and solvent-polymer inter-
large (see [20] for a recent review). Since a polymer so- action parameters reads: σSS = σSP = 1, SS = 0.5 and
lution in a good solvent is a miscible blend in which one SP = 0.8. Furthermore, we set mSS = 1. These model
component has a low molecular weight, it is interesting parameters are chosen because they mimic a real system,
to study the dynamics of polymer-solvent mixtures as the polystyrene in toluene (cf. App. A). The fact that the sol-
mixture is cooled toward Tg . There has been recent sim- vent particles have the same size as the monomers implies
ulation work in this direction, focusing, for instance, on that the cutoff distance is the same for all A and B. As in
the tracer diffusion in a polymer melt close to Tg (i.e., on previous work [26–28] we take rc = 2.3  2rmin , with rmin
the very dilute solvent regime) [21–23] or on the slow dy- being the minimum of the LJ potential.
S. Peter et al.: MD simulation of concentrated polymer solutions near the glass transition 149

For nearest-neighbor monomers along the chain the LJ -4.5


interaction is not included. These monomers are connected 0%
to each other by a harmonic potential, 10%
15%
k -5.0
Ubond (r) = (r − r0 )2 , (2)

ENB
pol
2
with equilibrium bond length r0 = 0.967 and spring con- -5.5
stant k = 1111. The value of k is large enough so that the
chains cannot cut through each other in the simulation, Tg
which allows the formation of entanglements. The entan-
glement length of bulk polymer melt is Ne ≈ 32 [29,30]. -6.0
0.2 0.3 0.4 0.5
In the following, we present results for N = 64.
T
pol
Fig. 1. Lennard-Jones energy of non-bonded monomers (ENB )
2.2 Simulation aspects versus temperature (T ) for three number fractions of solvent,
pol
φS = 0%, 10%, 15%. ENB is the energy per monomer and all
We perform MD simulations to study polymer-solvent results refer to N = 64. The solid lines for φS = 0% indi-
mixtures in the bulk. The equations of motion are inte- cate the method used to determine Tg by the intersection of
grated by the velocity-Verlet algorithm [31] with a step linear extrapolations from the low-T (glass) and high-T (liq-
pol
size of Δt = 0.01. The simulations are carried out at con- uid) branches of ENB . The vertical lines show the resulting
stant temperature (DPD thermostat [32]) and constant Tg ’s: Tg (φS = 0%)  0.404, Tg (φS = 10%)  0.378, and
pressure p which we take to be p = 0 as in previous Tg (φS = 15%)  0.365.
work [26–28].
The simulated systems contain a total of Ntot = we determine the glass transition temperature Tg (φS ) of
N n + NS particles, where N = 64 is the chain length, n polymer-solvent mixtures for various φS from continuous
is the number of chains, and NS is the number of solvent cooling runs with rate ΓT = 2 × 10−5 . To this end, we
particles. We specify the solvent content by the number monitor the T dependence of the LJ energy of a monomer
concentration (fraction) φS = NS /Ntot . The simulations given by
have Ntot = 3072, with 320 (φS ≈ 10%), 448 (φS ≈ 15%)
nN N n+N
and 768 (φS = 25%) solvent particles.
pol 1   S AB
To prepare the polymer-solvent mixtures we start from ENB = ULJ (|ri − rj |), (3)
nN i=1 j>i
equilibrated configurations of the pure polymer melt at
T = 0.5 and “dissolve” a number of (randomly chosen)
chains by redefining their monomers as solvent particles where ULJ AB
is defined in equation (1). Figure 1 depicts
pol pol
so as to obtain the desired solvent concentration. The ENB (T ) at different φS . We see that ENB (T ) continuously
resulting systems are equilibrated at T = 0.5 until the decreases on cooling. From the decrease we define, fol-
solvent distribution stabilizes. Starting from these equili- lowing previous work [26, 33], Tg by the intersection of
brated configurations the polymer solutions are cooled to linear extrapolations from the low-T (glass) and high-T
lower temperatures with a rate of ΓT = 2 × 10−5 accord- (liquid) branches of the LJ energy (see lines in Fig. 1 for
ing to the cooling schedule T (t) = 0.5 − ΓT t [26]. After φS = 0%). Certainly, the precise position and size of the
further equilibration we carried out production runs with temperature interval used in the analysis lead to a system-
a typical length of 100000 τLJ . atic error in Tg . This error can be rather large (cf. [26]),
but does not affect the qualitative dependence of Tg on
φS provided the same T intervals on the liquid and glass
3 Results branches are chosen for all systems. Furthermore, we con-
firm by an analysis of equilibrium data in Section 3.2 (see
In this section we explore the influence of the solvent on Fig. 4) that the qualitative trend obtained here is correct.
the properties of the polymer solution. We begin by a dis- We find that the solvent reduces Tg and thus acts as a
cussion of Tg obtained from non-equilibrium simulations plasticizing agent. The reduction of Tg with φS is approx-
with finite cooling rate (Sect. 3.1). Then, we present re- imately linear, as will be further discussed in Section 3.2
sults for various dynamic correlation functions in thermal (Fig. 4).
equilibrium (Sects. 3.2–3.4), such as the incoherent inter-
mediate scattering function or the van Hove correlation
function. 3.2 Mean-square displacements

We consider the mean-square displacements (MSDs) of the


3.1 Glass transition temperature monomers and solvent particles. The MSD is defined by
Nx
We expect that the solvent modifies the mobility of the 1   
g0,x (t) = |ri (t) − ri (0)|2 (x = P, S), (4)
monomers. To obtain a first idea about this modification, Nx i=1
150 The European Physical Journal E

T-Tc
2 N=64 ~t polymer 25%
10 15% 0.01 0.1
1
T=0.45
10 10% γ=2 10
4
0%
0 g0,x(τx)=1.5 solvent 25%
g0,x(t)

10

τS, τP
0.5 15% 3
-1 ~t 10
10 2 10%
6rr0,c
-2 0% 2
10 10% 10
4 10
-3 2 25%
10 3Tt 100%

2.3τS, τP
3 N=64
-2 -1 0 1 2 3 4 5 10 bulk
10 10 10 10 10 10 10 10
t
2
Fig. 2. Log-log plot of the MSD of all monomers (solid 10
lines; x = P) and of the solvent (dashed lines; x = S) at
p = 0 and T = 0.45. The chain length of the polymers is 1.5 2 2.5 3
N = 64. We compare different solvent concentrations: 0%
1/T
(= pure polymer melt; circles), 10% (black lines), 25% (grey
lines), and 100% (pure solvent; squares). The dotted lines in- Fig. 3. Lower panel: relaxation times (Eq. (5)) versus 1/T
dicate early-time ballistic motion (∼ t2 ), sub-diffusive motion for different solvent concentrations. The filled symbols refer to
due to chain connectivity (∼ t1/2 ), and free diffusion at late the monomers (τP ) and the open symbols to the solvent (τS ).
times (∼ t). The dashed horizontal lines show the criterion used τP of the pure melt (0%) is also shown (stars). τS is multiplied
to define the relaxation time τx (Eq. (5)), and the “plateau by 2.3; this factor makes the solvent relaxation time equal to τP
2
value” 6r0,c = 0.054. The latter was determined in previous at T = 0.5 for all finite φS . The lines through the polymer data
work [19] for N = 10. We assume that such local motion, are guides to the eye. Upper panel: same data plotted versus
r0,c ≈ 10% of the monomer diameter, depends only weakly T − Tc (φS ). Tc (φS ) is obtained from a fit of equation (6) to τP
on N , if at all. with γ = 2 (solid line through the polymer data for φS = 0%).
The dashed line shows the result of a fit with variable γ to the
(last 3 points of the) solvent data for φS = 25%. This gives
where ri (t) is the position of particle i at time t and the γ = 1.703.
sum runs either over NP = nN monomers or NS solvent
particles.
Figure 2 shows the MSDs for polymer-solvent mixtures system. (The dynamics of the pure solvent is much faster
of two solvent concentrations at T = 0.45. The data (lines than that of the pure polymer melt due to the difference
in Fig. 2) are compared with the MSD of the pure sol- in density ρ. For the solvent ρ = 0.61, while ρ = 1.02
vent (squares) and the monomer MSD of the pure polymer for the melt. This difference is caused by the smaller LJ
melt (circles) at the same T . The MSD of the pure melt interaction —SS = 0.5 versus PP = 1— and the chain
has the following features: At short times, the monomer length, the dominant factor being the LJ interaction.) In
moves ballistically, g0 (t) = 3T t2 . The regime of ballis- the plateau regime, g0,S (t) closely follows the monomer
tic motion is succeeded by a “plateau regime”, where the displacement, albeit g0,S (t) > g0,P (t). As time increases,
MSD increases only slowly with time. There, g0 (t) is of solvent and monomer MSDs separate more and more from
the order of 10% of the monomer diameter. This reflects each other. Due to chain connectivity g0,P (t) ∼ t1/2 , while
the temporary caging of a monomer by its neighbors. The the solvent MSD directly crosses over to diffusive motion,
plateau regime may thus be identified with the MCT β g0,S (t) ∼ t.
process [16,34]. For longer times, the MSD is determined In order to further quantify the relaxation and to dis-
by chain connectivity. We find g0 (t) ∼ tx , with x = 1/2. cuss its temperature dependence, we introduce a relax-
(For shorter chains, such as N = 10, larger exponents are ation time for the polymer (τP ) and solvent (τS ) by
found, x ≈ 0.63, due to small-N effects [35].) This sub-
diffusive motion agrees with the prediction of the Rouse 3
model [36,37] for intermediate times that are shorter than g0,x (t = τx (T, φS )) = , (x = P, S). (5)
2
the time a monomer takes to move across a distance com-
parable to the chain size. τx measures the time a monomer or a solvent particle take
Figure 2 reveals that the addition of solvent strongly to cover a distance comparable to its own size. This time
affects the monomer dynamics in the plateau regime and therefore characterizes the final structural (α) relaxation
beyond. The larger φS , the more the plateau disappears of the polymer solution. Figure 3 shows τP and τS as a
and the faster g0,P (t) increases. Apparently, the solvent function of temperature and composition. Both τP and τS
weakens the nearest-neighbor cage relative to the undi- increase —in a non-Arrhenius fashion— with decreasing
lute polymer melt. In turn, the solvent is also subject T , but the increase is weaker for τS than for τP . Relative
to caging by the monomers. The solvent dynamics is to the bulk τP is reduced and, as expected, the reduction
substantially slowed down compared to the pure solvent increases with increasing φS .
S. Peter et al.: MD simulation of concentrated polymer solutions near the glass transition 151

Following the prediction of ideal MCT for the bulk [34] 0.42
we attempt to fit τx (T, φS ) by a power law 3

aP
0.40
ax (φS ) 2

Tc, Tg
τx (T, φS ) = , (x = P, S). (6)
[T − Tc (φS )]γ(φS ) 0.38
0 0.1 0.2
φs
This equation needs some comments. 0.36
i) It provides two MCT parameters, the critical tem- Tc
perature Tc (φS ) and the exponent γ(φS ). From the the- 0.34 Tg
0 0.1 0.2
ory one would expect that Tc (φS ) and γ(φS ) depend on
the thermodynamic state of the system —and thus on
φs
composition— but not on the studied quantity “x”. (The Fig. 4. Main figure: critical temperature of MCT Tc and glass
prefactor ax (φS ), however, depends on “x”.) The latter transition temperature Tg (from Fig. 1) versus solvent concen-
expectation is not fully borne out. We obtain Tc (φS ) and tration φS . Inset: prefactor aP (φS ) from the fit of the monomer
γ(φS ) from fitting equation (6) to τP (T, φS ) (see below). relaxation time to equation (6). The dashed lines in the main
When fixing Tc (φS ) and fitting equation (6) to the data figure and the inset represent linear fits to the φS dependence of
for τS , we find γ values that are smaller than those deter- Tc (φS ) (= 0.4165 − 0.27φS ) and aP (φS ) (= 3.0731 − 4.3846φS ).
mined from τP and decrease with increasing φS (Fig. 3,
upper panel).
ii) Many simulation studies show that equation (6) 3.3 Incoherent intermediate scattering function
is only applicable in a limited T interval above Tc (see,
e.g., [38,39] or [19] and references therein). For large T and The incoherent intermediate scattering function measures
very close to Tc equation (6) is expected to break down. the decorrelation of the positions of a single particle with
For T  Tc the asymptotic formula (6) cannot be applied time on length scale 1/q, where q is the modulus of the
yet, whereas for T very close to Tc decay processes which wave vector q. Here we introduce this function for the
are not included in the ideal MCT allow the system to re- monomers of the polymers (NP = nN ); it is defined by
lax after a large but finite time. (Deviations from Eq. (6)
near Tc are clearly visible in Fig. 3.) This introduces an NP
1 
uncertainty in the determination of Tc because the choice φsq,P (t) = exp{−iq · [ri (t) − ri (0)]} . (7)
of the interval for which equation (6) may be expected to NP i=1
hold cannot be made unambiguously. For the present fit
we used the largest T interval where equation (6) can be In the following we will examine the monomer dynamics
applied to the monomer relaxation time τP (T, φS ). for three wave vectors, the maximum q ∗  7 of the static
iii) It is often found that the fit parameters Tc (φS ) and structure factor S(q) and a wave vector smaller (q = 4)
γ(φS ) are correlated (see, e.g., [19,38,39]). For instance, a and larger (q = 12.8) than q ∗ .
lower value of Tc can be compensated by a higher value Before addressing the dynamics we briefly discuss S(q).
of γ without significantly changing the quality of the fit. The static structure factor is defined by
Here we encountered the same problem. Nevertheless, the NP
 +NS

analysis suggested that τP (T, φS ) can be fitted by an ex- 1


S(q) = e−iq·(ri −rj ) , (8)
ponent that varies very weakly with φS and is close to the NP + N S i,j=1
result for the pure melt, γ = 2 [27]. Thus, we decided to
determine Tc (φS ) under the constraint γ(φS ) = 2.
where the sum runs over all monomers (NP ) and solvent
Figure 4 shows the results of this analysis. We find particles (NS ). S(q) measures the collective density fluc-
that Tc —and also the prefactor aP (φS )— decrease lin- tuations of the system. Figure 5 depicts S(q) for the pure
early with increasing solvent concentration. The linear de- polymer melt and a polymer-solvent mixture with 25% of
crease of Tc is fairly robust. Even if we allow γ to vary with solvent. The temperatures shown are close to the Tc of the
φS , the linear dependence of Tc on φS is preserved. There- respective system. We see that S(q) displays the typical
fore, despite the weaknesses of the fit summarized above, features of a liquid. The structure factor is small at low q,
we believe that the result of Figure 4 would hold up under reflecting the small compressibility of the system. Then,
further scrutiny, for instance, by quantitative MCT calcu- S(q) increases with increasing q toward a maximum, the
lations based on static input from the simulation, as done “amorphous halo”, before converging to 1 in an oscillatory
recently for the pure polymer melt [35]. fashion as q → ∞. The position q ∗ and in particular the
A depression of Tc is also found in simulations of amplitude S(q ∗ ) of the amorphous halo grow on cooling.
another polymer-solvent mixture [24] and of a miscible This indicates that the intermolecular packing on the lo-
blend [11]. There is also evidence from experiments for cal length scale of the particles —2π/q ∗ is of the order of
a linear decrease of Tg with increasing φS [5], although the diameter of a monomer or solvent particle— becomes
such a decrease is typically limited to fairly small solvent tighter due to the increase of density with decreasing T .
concentrations (see [2] for an overview). The tightening of the nearest-neighbor cages is, according
152 The European Physical Journal E

4 4
10
γ=2.05
q* 1 10
3

3.5

S(q*)
2
10

τq
s
φS=0%
4 3
0.8 q=7 10
1 γ=2.34
φS=0.25% 0 γ=2.5
10
2.5 0.6 10
-2
10
-1

φq(t)
3 -3
10
-2
10
-1
10
T-Tc

s
T-Tc q=12.8
0.4 q=4
N=64
S(q)

2 φS=0%
T=0.45 φS=0.25% 0.2
T=0.35 φS=0.25% b=0.6
1 0 -5 -4 -3 -2 -1 0 1
T=0.55 φS=0% 10 10 10 10 10 10 10
s
T=0.43 φS=0% t/ττq
0
5 20
5
10 15 10
q 1 10
4 γ=2

τq,P
10
Fig. 5. Main figure: static structure factor S(q) versus the

s
q=7 γ=2.38
modulus of the wave vector q for the pure polymer melt (φS = 0.8 10
2

γ=2.55
0; symbols) and a polymer-solvent mixture containing φS = 10
1

25% of solvent (lines). For both systems two temperatures are φq,P (t) 0.6 0.01 0.1
shown: T = 0.55 and T = 0.43 for the pure melt (Tc  0.415); q=12.8
T-Tc
s

T = 0.45 and T = 0.35 for the polymer solution (Tc  0.347). 0.4 q=4
The position q ∗ of the first maximum of S(q) is indicated. N=64
Inset: S(q ∗ ) versus T − Tc for the pure melt (circles) and the 0.2 φS=25%
polymer-solvent mixture (stars).
b=0.6
0 -6 -5 -4 -3 -2 -1 0 1
10 10 10 10 10 10 10 10
to MCT, the driving mechanism for the glass-like slowing s
t/ττq,P
down of the relaxation in supercooled liquids [16,34].
In this respect, the comparison between the pure melt Fig. 6. Incoherent intermediate scattering function at q =
s s
and the polymer solution reveals an interesting result. If 4, 7, 12.8 versus rescaled time t/τq,P where τq,P is defined by
we compare both systems at roughly the same T , such equation (9). The upper panel depicts the results for the pure
as T = 0.43 for the melt and T = 0.45 for the solution, polymer melt (φsq (t)), the lower panel for the polymers (φsq,P (t))
we find that the structure factors differ, in particular in in a polymer-solvent mixture with a solvent content of 25%. For
the q range around q ∗ where the S(q) of the solution is the pure system the following temperatures are shown (from
smaller than in the melt. The structure factor of the solu- right to left): T = 0.55, 0.5, 0.45, and 0.43 (Tc  0.415); for
tion at T = 0.45 —this temperature is at ΔT  0.103 the polymer-solvent mixture the temperatures are (also from
from Tc  0.347— rather agrees with the S(q) of the right to left): T = 0.42, 0.4, 0.37, and 0.35 ( Tc ). The data for
pure system at T = 0.55, i.e. with the S(q) at a temper- T = 0.35 are represented by dashed lines due to violations of
ature that is at a comparable distance to Tc , ΔT  0.135 the TTSP at this T . In both panels the dotted lines indicate a
fit to equation (11) and the dashed lines a fit for q = 12.8 to the
(Tc  0.415 for the pure melt). Thus, addition of solvent sK
KWW function (Eq. (10)) with βq,P = 0.6. The insets show τqs
to the polymer melt delays the tightening of the nearest- s
(upper panel: pure system) and τq,P (lower panel: mixture) as
neighbor cages —and along with that, the slowing down a function of T − Tc (q = 4: stars, q = 7: circles, and q = 12.8:
of the dynamics— to lower temperatures. This impact of squares). The solid lines are fits to equation (6); the resulting
solvent on the structure of the polymer solution appears values for γ are indicated.
to be a key factor to understand its plasticizing effect.
We now resume the discussion of φsq,P (t). Figure 6 com-
pares the monomer dynamics in a polymer solution with time τq,P
s
through the condition that φsq,P (t) has decayed
φS = 25% (lower panel), the highest solvent concentration to 10% of its initial value,
investigated here, to the monomer dynamics in the pure
polymer melt (upper panel), as differences between these φsq,P (t = τq,P
s
) = 0.1. (9)
two cases should be largest.
According to the ideal MCT, the α process should Figure 6 demonstrates that the incoherent scattering func-
obey a time-temperature superposition principle (TTSP) tions indeed collapse onto a single master curve when plot-
for T → Tc+ [40–43]. TTSP means that the simulation re- ted versus t/τq,P
s
. Thus, we find that the TTSP holds, ex-
sults for φsq,P (t), measured at different T , collapse onto a cept for φq,P (t) of the polymer-solvent mixture at T = 0.35
s

temperature-independent master curve if time is rescaled (dashed lines in Fig. 6; the deviations are most strongly
by the α relaxation time. Here we define the α relaxation visible at the highest q-value). This temperature is very
S. Peter et al.: MD simulation of concentrated polymer solutions near the glass transition 153

close to Tc  0.347. Similar deviations from the TTSP The observation that relaxation times characterizing
are also observed in other simulations [7,44,45]. Within the dynamics on large length scales increase on cooling
MCT they are attributed to additional, not well under- more weakly than those on scales q  q ∗ , is not particu-
stood relaxation mechanisms which lead to violations of lar to our polymer model. Similar observations were also
the predictions from the idealized theory [46,47]. So we made in simulations of other glass-forming liquids [7,38,
exclude T = 0.35 in the subsequent analysis. 39,45,50,51]. In particular, the difference in the slowing-
No closed expression for the α master curve is known down of the dynamics at large and small length scales
from ideal MCT. However, the master curve may be fitted implies the violation of the Stokes-Einstein relation, i.e.,
well by a Kohlrausch-Williams-Watts (KWW) function the product of the diffusion coefficient (corresponding to
a relaxation time in the limit q → 0) and the α relaxation
φsq,P (t)  fq,P
sK
exp −(t/τq,P )
sK
sK βq,P
, (10) time (τqs∗ ,P ) is not independent of T .
The decoupling of diffusion and structural relaxation
on cooling has been interpreted as one signature of in-
except for times of the early α processes where φsq,P (t) creasingly heterogeneous dynamics in the liquid near its
relaxes out of the plateau regime. There, systematic devi- glass transition (for reviews on “dynamic heterogeneity”
ations are expected because the short-time expansion of see [52–55]). The term “heterogeneous dynamics” circum-
equation (10) does not agree with the short-time expan- scribes the idea that a glass former near Tg contains ag-
sion of the α master curve. The latter is given by gregates (“subensembles”) of particles having enhanced
s b or reduced mobility relative to the average. To evidence
φsq,P (t) = fq,P
sc
− Aq,P t/τq,P , (t τq,P
s
). (11) this dynamic heterogeneity, various methods have been
deployed [19,52–55], such as filtering techniques to track
Here fq,Psc
is the “non-ergodicity parameter”, Aq,P is an slow or fast particles, analysis of ensemble-averaged three-
amplitude, and b is the “von Schweidler exponent”. Near or four-point correlation functions [56,57], or scrutiny of
Tc , fq,P
sc
, Aq,P , and b should be independent of tempera- the distribution of single-particle displacements [38,39,
ture. Equation (11) shows that βq,P sK
= b in general [48]. 51]. Here we pursue the latter approach following the work
However, there is a special case, the limit q → ∞. In this by Flenner and Szamel [38,39].
limit, it was proved [49] that βq,P
sK
→ b and fq,P sK
→ fq,P
sc
.
We test these predictions by fitting equations (11)
and (10) to the simulation results, following the prescrip- 3.4 Probability distribution of single-particle
tion given in [19]. The fit to equation (11) reveals that we displacements and non-Gaussian parameters
can describe the early α relaxation of φsq,P (t) for both the
Following references [38, 39] we introduce the probability
pure melt and polymer solution by the same von Schwei-
distribution Px (ln r; t) of the logarithm of monomer (x =
dler exponent, b = 0.6, for all q. The non-ergodicity pa-
P) or solvent particle (x = S) displacements in time t.
rameter, on the other hand, depends on q. As expected
This probability distribution is related to the self-part of
from theory [43] and other simulations (see, e.g., [16,19]
the van Hove correlation function [58], Gs,x (r, t), by
and references therein), it increases with decreasing q. Fur-
thermore, we find that for the largest q value studied, Px (ln r; t) = 4πr3 Gs,x (r, t), (x = P, S), (12)
q = 12.8, a fit to a KWW function with βq,P sK
= b = 0.6
is possible. Apparently, this wave vector is large enough where Gs,x (r, t) is given by
so that (some features of) the asymptotic large-q behavior
Nx
may be observed (see [19] for a critical discussion of this 1 
point). Gs,x (r, t) = δ(r − [ri (t) − ri (0)]) (13)
Nx i=1
The preceding discussion shows that the fits yield re-
sults which are compatible with the expectation from ideal and Nx = nN for the monomers or Nx = NS for the sol-
MCT. However, Figure 6 also reveals a point of disagree- vent particles. A special feature of Px (ln r; t) is that its
ment, the T dependence of the relaxation time τq,P s
(in- shape is independent of time if Gs,x (r, t) is Gaussian —as
sets in the figure). Following MCT τq,P should increase on
s
is the case in the short-time, ballistic and long-time, dif-
cooling according to τq,P s
∼ (T − Tc )−γ , where γ is fully fusive regimes. For Gaussian displacementsPx (ln r; t) ex-
determined by the von Schweidler exponent [16,34]. For hibit a single peak whose height is equal to 54/πe−3/2 
b = 0.6 we expect γ = 2.406 [19]. However, if we fix Tc 0.925. Deviations from this value can therefore be used
at the values obtained before (Fig. 4) and determine γ by as a diagnostic for non-Gaussian particle motion. Fur-
performing the fit over the largest T interval compatible thermore, Px (ln r; t) allows one to detect dynamic het-
with a power law increase of τq,P
s
, Figure 6 shows that γ de- erogeneities. For a glass-forming binary liquid close to
creases with decreasing q and approaches, for the smallest Tc [38,39] a double-peak structure of Px (ln r; t) has been
q-values studied, the exponent γ = 2 found in Section 3.2 observed, indicating the “coexistence” of weakly moving
from the MSD. The latter finding can be understood if we (“slow”) and far moving (“fast”) particles at the same
take into account that φsq,P (t)  1 − 16 q 2 g0,P (t) for q → 0. time. (Similar observations are also made for a model of a
Therefore, an analysis of φsq,P (t) at small q must yield the colloidal gel close to the gel transition [59].) These findings
same result as the MSD. prompted us to analyze Px (ln r; t) here too.
154 The European Physical Journal E

In conjunction with Px (ln r; t) the studies [38,39] also t


-2 -1 0 1 2 3 4 5 6
determine two other quantities, the classical non-Gaussian 10 10 10 10 10 10 10 10 10
parameter [58] g0(t)
0
α2(t) 10
Nx
3 1    ,
α2(t) 10
-1
α2,x (t) = |ri (t) − ri (0)|4 − 1, (14)
5g0,x (t)2 Nx i=1 2
6rr0,c t2 t’t2 -2
1 10
√6 r0,c
—for our pure polymer melt α2 (t) was studied in detail -3
0.8 10
in [60]— and a new non-Gaussian parameter defined by

4πr Gs(r,t)
t=10 t=25000
0.6
Nx  
 g0,x (t) 1  1 t2 Gaussian
α2,x − 1. 0.4

3
(t) = (15) t’t2 N=64
3 Nx i=1 |ri (t) − ri (0)|2
0.2 T=0.43

These definitions guarantee that α2,x (t) and α2,x (t) van-
0 -1 0 1
ish for very short and long times, where particle displace- 10 10 10
ments are Gaussian. However, for intermediate times com- r
prising the β and (early) α processes both non-Gaussian Fig. 7. Left ordinate: probability distribution P (ln r; t) of the
parameters are found to be positive [38,39]. A positive logarithm of monomer displacements in the pure polymer melt
value of α2,x (t) implies that the mean-quartic displace- at T = 0.43 for times t = 10, 100, 250, 500, 1000, 1500,
ment of a particle is larger than expected from ordinary 2500, 5000, 10000, and 25000 (from left to right). The thick
diffusion. Since the quartic moment is significantly influ- dashed line represents P (ln r; t) at t = t2 , the peak time of
enced by particles that move far in time t, α2,x (t) is par- α2 (t); the thick dotted line shows P (ln r; t) at t = t2 , the
ticularly sensitive to “fast” particles. On the other hand, peak time of α2 (t); and the dotted curve for t = 25000 cor-
the new parameter α2,x 
(t) will be large if there is both re- responds to the Gaussian approximation for P (ln r; t) using
duced and enhanced mobility at the same time. Through g0 (t = 25000) = 2.678355 from the upper panel. The vertical

1/|ri (t) − ri (0)|2 , α2,x (t) weighs strongly the “slow” par- dashed line indicates the square root of the “plateau value”
2
6r0,c = 0.054 of the monomer MSD (see Fig. 2). Right or-
ticles which have not moved as far as expected for Gaus-
dinate: monomer MSD g0 (t), non-Gaussian parameter α2 (t)
sian displacements, and through g0,x (t) the “fast” parti-
defined in equation (14), and modified non-Gaussian parame-
cles which have moved farther.
ter α2 (t) defined in equation (15) for the pure polymer melt
Figure 7 depicts P (ln r; t) and both non-Gaussian pa- at T = 0.43. The vertical dashed and dotted lines indicate,
rameters for the pure polymer melt at T = 0.43. (For the respectively, the peak times t2 and t2 of α2 (t) and α2 (t). The
pure melt we drop again the index “x”.) We find clear devi- horizontal dashed line shows the “plateau value” 6r0,c2
= 0.054
ations from Gaussian behavior at all but the shortest and of g0 (t) from Figure 2.
longest times. The non-Gaussian parameters are positive
and have a maximum for times where the monomers leave
the plateau regime of the MSD. However, as pointed out t2 for α2,x (t) and t2 for α2,x

(t) with t2 < t2 as before.
in [38,39], the peak occurs for α2 (t) at a later time than The peak amplitudes, however, are different, those of the
for α2 (t). At the peak time t2 of α2 (t) the distribution of solvent being larger than the corresponding ones for the
logarithmic√ displacements has a pronounced maximum at monomer. Appealing back to the interpretation of α2,x we
only r ≈ 6r0,c . This illustrates the difficulty a monomer conclude that there are more far moving solvent particles
has in exceeding the distance of about 10% of its diameter than monomers at the peak time t2 of α2,x .
in the cold melt. On the other hand, at the peak time t2 of Figure 9 supports this interpretation. At t2 , PS (ln r; t)
α2 (t) √
the distribution is very broad, encompassing small has a pronounced shoulder for r ≈ 1 contrary to
(r  6r0,c ) and large (r  1) displacements. This hints PP (ln r; t). This shoulder develops into a second peak with
at large, non-Gaussian fluctuations in particle mobility at increasing time. For times around the peak time t2 of

this time. Our results thus support the conclusion of [38] α2,x , Px (ln r; t) has —in accord with the findings of refer-
that the peak position of α2 (t) is better suited than that of ence [38]— a clear double-peak structure. This allows us
the conventional non-Gaussian parameter to detect these to identify two populations of particles, “slow” ones which
fluctuations and the attendant dynamic heterogeneity ac- have not moved much farther than 10% of their diameter
companying the structural relaxation in the cold liquid. in time t2 , and “fast” ones which have left their cage and
Nevertheless, P (ln r; t) does not display a double-peak covered a distance of about their diameter or more. The
structure, presumably because T = 0.43 is still too far existence of two populations of particles in the interme-
above Tc ( 0.415). This is suggested by the work of ref- diate time window where structural relaxation occurs has
erence [38] and by an analysis of our polymer solution with also been demonstrated recently by a direct analysis of
25% of solvent at T = 0.35 (Tc  0.347). We turn to this the van Hove function [51].
analysis now. For t > t2 the disparity between monomer and solvent
Figure 8 depicts the monomer and solvent MSDs for displacements increases. This is expected. The solvent par-
the solution and the non-Gaussian parameters, α2,x (t) and ticles move much further because they begin to cross over

α2,x (t). We see that the non-Gaussian parameters of the to free diffusion, whereas the monomer motion is bound
solvent and monomers peak at roughly the same time, by chain connectivity before free diffusion can set in.
S. Peter et al.: MD simulation of concentrated polymer solutions near the glass transition 155

t 1
-2 -1 0 1 2 3 4 5 √6 r0,c
10 10 10 10 10 10 10 10 Gaussian
2 0.8
10 φS=25% t2
t’

4πr Gs,P(r,t)
1 t=10 t=25000
10 N=64 0.6
g0,x(t)

0
10 ~t2
-1
10 0.4

3
10
-2
6 T=0.35
-3
0.2 N=64
10 5 φS=25%

α2,x(t)
pure polymer T=0.43 4 0
polymer T=0.35 -1 0 1

α2,x(t), α’
α2,x(t)
α’ 10 10 10
solvent T=0.35 3 r
2 1 T=0.35
Gaussian
α2,x(t) solvent √6 r0,c
1 0.8 φ =25%
S

4πr Gs,S(r,t)
0
-2 -1 0 1 2 3 4 5 0.6 ~t2 ~tt’2
10 10 10 10 10 10 10 10
t
0.4 t=10 t=25000

3
Fig. 8. Upper panel: MSD of the monomers in the pure poly-
mer melt at T = 0.43 (filled circles), of the monomers in a
0.2
binary polymer-solvent mixture with φS = 25% solvent at
T = 0.35 (solid line), and of the solvent particles in the same
mixture (dashed line). Lower panel: non-Gaussian parameters,
0 -1 0 1
 10 10 10
α2,x (t) (Eq. (14)) and α2,x (t) (Eq. (15)) for the pure polymer r
melt at T = 0.43 and the polymer solution with φS = 25%
of solvent at T = 0.35. α2,x (t) is represented by filled circles Fig. 9. Probability distribution Px (ln r; t) of the logarithm
(monomers in the pure melt), a thick solid line (monomers in of single-particle displacements in a polymer-solvent mixture

the solution), and a thick dashed line (solvent particles). α2,x (t) with φS = 25% at T = 0.35. The upper panel refers to the
is shown by open circles (monomers in the pure melt), a thin monomers, the lower to the solvent particles. For the monomers
solid line (monomers in the solution), and a thin dashed line and solvent particles the following times are shown (from left
(solvent particles). to right): t = 10, 100, 250, 1000, 2500, 5000, 10000, and 25000.
In both panels the thick dashed lines represent Px (ln r; t) for
t close to the peak time t2 of α2 (t); the thick dotted lines
show Px (ln r; t) for t close to the peak time t2 of α2 (t); and
4 Summary the dotted curves for t = 25000 correspond to the Gaussian
approximation for Px (ln r; t) using g0,P (t = 25000) = 1.7785
By means of molecular dynamics simulations we explored for the monomer and g0,S (t = 25000) = 8.8399. The values of
structural and dynamic features of model polymer-solvent these MSDs are read off from Figure 8. The vertical dashed
mixtures on cooling toward the glass transition. Our lines in both panels indicate the square root of the “plateau
2
model may arguably be considered as an archetypal model value” 6r0,c = 0.054 of the monomer MSD in the pure melt
for polymer-solvent mixtures. Polymers are represented by (Fig. 2).
bead-spring chains and solvent particles by Lennard-Jones
(LJ) spheres having the same size as the monomers. The
various LJ interactions are adjusted such that mixing is indeed find such an effect (Fig. 5). Moreover, if the struc-
favored at low temperatures T . The focus of the present ture factors of the solution and melt are compared not at
work are the equilibrium properties of polymer solutions the same T but for the same distance to the respective Tc ,
above the glass transition temperature Tg . Our main re- they closely agree with one another.
sults may be summarized as follows: This suggests to perform a more quantitative compar-
The polymer solution has one Tg (Fig. 1) and the sol- ison between simulation and MCT. We base this compar-
vent acts as a plasticizer. The glass transition temperature ison on predictions from ideal MCT, which are supposed
and the critical temperature Tc of mode-coupling theory to be valid close to Tc . In agreement with the theory, we
(MCT) decrease linearly with increasing solvent concen- find that i) dynamic correlation functions, such as mean-
tration φS , at least up to φS = 25%, the highest solvent square displacements or the incoherent intermediate scat-
content studied (Fig. 4). tering function φsq (t), relax in two steps close to Tc , ii) the
Viewed from the perspective of (ideal) MCT the shift α relaxation of φsq (t) obeys the time-temperature super-
of Tc should be accompanied by changes in the structure. position principle (for T close but not “too close” to Tc ;
In particular, local spatial correlation at the scale of the cf. Fig. 6), iii) the initial decay of the resulting α master
wavelength 1/q ∗ , q ∗ being the peak position of the static curve can be fitted by a von Schweidler law, and iv) the
structure factor S(q), should be weaker in the polymer so- increase of the α relaxation time on cooling is compatible
lution than in the pure polymer melt at the same T . We with a power law (Figs. 3 and 6).
156 The European Physical Journal E

Certainly, our analysis is less than ideal in the respect Appendix A. Choice of the Lennard-Jones
that it hinges on a fit procedure which has to optimize sev- parameters
eral parameters simultaneously after having done an “ap-
propriate” choice for the time interval of the fit. There is The magnitude of the LJ parameters in equation (1) deter-
some latitude in this choice, leading to uncertainties in the mines the compatibility of both species and thus the phase
estimated MCT parameters, which are hard to avoid (see, behavior of the polymer-solvent mixture [61]. We are in-
e.g., [19] for further discussion). However, MCT provides terested in a situation where mixing is energetically favor-
an avenue around this problem. The full time dependence able (“good solvent conditions” [36]). The parameters AB
of dynamic correlation functions can be calculated from and σAB have to be chosen accordingly. One way to guide
structural input determined a priori from the simulation. this choice is to adjust the parameters to a real polymer
This opens the possibility for a direct comparison between in good solvent, such as polystyrene (PS) in toluene. This
predicted and simulated dynamics without the need to re- approach is sometimes taken in the literature [62], and we
sort to fitting procedures. We have recently applied this follow the procedure of references [62–65] in the present
approach to the pure polymer melt [35]. The results ob- modeling. We emphasize, however, that our aim is not
tained suggest that it might be worthwhile to extend the to develop a faithful coarse-grained model for PS-toluene
analysis to the polymer-solvent mixtures studied here. solutions. Rather we utilize the mapping to PS-toluene
solutions as a tool to make an educated guess for the LJ
The possibility of directly comparing theory and sim- parameters of equation (1).
ulation is certainly the main strength of ideal MCT. How- We identify solvent molecules with LJ monomers and
ever, the theory also has a weakness which is well known. determine the LJ parameters by equating the critical
It predicts vitrification at Tc , but the expected divergence temperature and density of the liquid-gas transition of
of the structural relaxation time is not observed (Figs. 3 toluene —Tc = 591.8 K and ρc = 288 kg/m3 [66]— with
−3
and 6). Particles do eventually escape from their cages as Tc = 0.999 SS /kB and ρc = 0.32 σSS [64] of the liquid-gas
Tc is approached from above. In simulations these devia- transition of LJ monomers interacting by equation (1).
tions are often accompanied by a decoupling of the relax- This gives SS = 8.18 × 10−21 J and σSS = 5.54 × 10−10 m.
ation times on local (q ≈ q ∗ ) and large (q q ∗ ) length For the LJ parameters of the polymer we follow ref-
scales (Fig. 6). This decoupling resembles the violation of erence [65] and match the T dependence of the monomer
the Stokes-Einstein relation found in experiments close to density of the bead-spring polymer melt with that of PS.
Tg and is attributed to the existence of spatially hetero- To this end, we employ a theoretical model developed by
geneous dynamics. Long and Lequeux for the thermodynamics of polymer
melts above Tg [67]. Long and Lequeux suggest that the
We looked for such dynamic heterogeneities by ex- free energy of a high-density polymer melt close to Tg can
amining the distribution P (ln r; t) of the logarithm be decomposed into two terms, an energetic term due to
of single-particle displacements, as suggested in refer- the van der Waals attraction between polymer segments
ences [38,39]. The analysis reveals strong fluctuations in and an entropic term penalizing densities which approach
the size of the particle displacements on the time scale close packing. From this free energy they derive the de-
of the α process (Fig. 7). For the polymer solution we pendence of the monomer density ρ(T ) on temperature
find that P (ln r; t) has, very close to Tc , a double-peak   
structure. The peaks are roughly of the same height on ρ0 T
ρ(T ) = 1+ 1− , Tg ≤ T T c , (A.1)
the time scale t2 beyond the plateau regime, where a new 2 Tc
non-Gaussian parameter, suggested in [38], has its maxi-
mum. The double-peak structure of P (ln r; t) reflects large where Tc is the hypothetical temperature above which air
disparities in particle mobility and thus the dynamic het- is a good solvent for the polymer and ρ0 is the monomer
erogeneity of the system. It implies that there are two pop- density at T = 0. Although this prediction can only
ulations of particles, “slow” ones which have not moved be approximately valid —the development of an accu-
much farther than 10% of their diameter in time t2 , and rate equation of state for polymer systems is a difficult
“fast” ones which have left their cage and covered a dis- problem [65,68,69] (see, e.g. [61] for a review on recent
tance of about their diameter or more (Fig. 9). This bi- progress)— reference [67] suggests that equation (A.1)
modal character of the structural relaxation is hard to provides a good description of experimental data for dif-
predict from ideal MCT [39], could be responsible for the ferent polymers including PS. Therefore, we adjust equa-
absence of the divergence of the relaxation time at Tc , and tion (A.1) to the ρ(T ) data obtained from simulations of
appears to be a general feature of materials close to glass an N = 64 bead-spring melt at pressure p = 0 (Fig. 10).
or jamming transitions [51]. This gives ρ0 = 1.125/σPP 3
and Tc = 1.355PP /kB . Us-
ing the results for PS from [67], i.e. ρ0 = 3.57 × 1027 /m3
and Tc = 1225 K  3.3Tg , we find PP = 1.25 × 10−20 J
and σPP = 6.81 × 10−10 m. When comparing these re-
We acknowledge financial support from the European Commu- sults with those derived above for the solvent, we see that
nity’s “Marie-Curie Actions” under contract MRTN-CT-2004- SS  0.66 PP and σSS  0.81 σPP .
504052 (POLYFILM). J. B. acknowledges financial support by This analysis shows that the polymer-polymer at-
the IUF. traction is about twice as large as the solvent-solvent
S. Peter et al.: MD simulation of concentrated polymer solutions near the glass transition 157

References
1.1 ρ0=1.12
Tc=1.35 1. J. Ferry, Viscoelastic Properties of Polymers (Wiley, New
York, 1980).
1.0 2. G.B. McKenna, in Comprehensive Polymer Science, edited
ρ(T)

by C. Booth, C. Price, Vol. 2 (Pergamon, New York, 1986)


pp. 311–362.
0.9 3. More precisely, plasticization of the polymer means that
N=64 Tg the solvent does not only decrease Tg , but also softens the
bulk polymer glass by reducing the elastic moduli [1]. Besides
this normally encountered case, there are also systems in
0.8
0.2 0.4 0.6 0.8 1 which the solvent “antiplasticizes” the polymer. That is,
the solvent decreases Tg , but increases the elastic moduli
T of the polymer glass. The origin of this antiplasticizing
Fig. 10. Density ρ(T ) of a bulk polymer melt (N = 64) melt effect has recently been studied by simulations for both
versus temperature T at pressure p = 0 (circles). The black line bulk systems and polymer films [24, 25].
is a fit to equation (A.1) for T > Tg . The resulting fit parame- 4. T.S. Chow, Macromolecules 13, 362 (1980).
ters, Tc and ρ0 , are indicated in the figure. The vertical dotted 5. J.E.G. Lipson, S.T. Milner, J. Polym. Sci. Part B 44, 3528
line represents the glass transition temperature Tg (= 0.404) (2006).
of the melt (see Sect. 3.1). 6. T.P. Lodge, T.C.B. McLeish, Macromolecules 33, 5278
(2000).
7. G. Foffi, W. Götze, F. Sciortino, P. Tartaglia, T. Voigt-
mann, Phys. Rev. E 69, 011505 (2004).
8. U.K. Rößler, H. Teichler, Phys. Rev. E 61, 394 (2000).
attraction, whereas the particle diameters are roughly
9. A.J. Moreno, J. Colmenero, Phys. Rev. E 74, 021409
the same for polymer and solvent. This suggests to pose
(2006).
SS = 0.5 and σSS = 1, which is the choice made in
10. A.J. Moreno, J. Colmenero, J. Chem. Phys. 124, 184906
Section 2.1.
(2006).
There are two additional parameters to determine, the 11. A.J. Moreno, J. Colmenero, J. Phys.: Condens. Matter 19,
characteristic energy and length scales, PS and σPS , of the 466112 (2007).
polymer-solvent interaction. It is generally assumed that 12. W. Götze, T. Voigtmann, Phys. Rev. E 67, 021502 (2003).
these cross-interaction parameters are related by simple 13. W.C.K. Poon, J. Phys.: Condens. Matter 14, R859 (2002).
mixing rules to those of the like components. For σPS the 14. A.M. Puertas, M. Fuchs, M.E. Cates, Phys. Rev. Lett. 88,
most common choice is the “Lorentz rule” [58] 098301 (2002).
15. K. Dawson, G. Foffi, M. Fuchs, W. Götze, F. Sciortino,
1 M. Sperl, P. Tartaglia, T. Voigtmann, E. Zaccarelli, Phys.
σPS = (σPP + σSS ), (A.2) Rev. E 63, 011401 (2000).
2
16. W. Götze, J. Phys.: Condens. Matter 11, A1 (1999).
17. W. Kob, J. Phys.: Condens. Matter 11, R85 (1999).
which is exact for hard spheres. We adopt this choice here,
18. K.S. Schweizer, Curr. Opin. Colloid Interface Sci. 12, 297
and with σPP = σSS = 1 we find σPS = 1. For PS we take
(2007).
√ 19. J. Baschnagel, F. Varnik, J. Phys.: Condens. Matter 17,
PS = ξ PP SS , (A.3) R851 (2005).
20. J. Colmenero, A. Arbe, Soft Matter 3, 1474 (2007).
which deviates from the frequently used “Berthelot 21. R.A.L. Vallée, M. Van der Auweraer, W. Paul, K. Binder,
rule” [58] that is obtained for ξ = 1. In fact, the value Phys. Rev. Lett. 97, 217801 (2006).
of ξ has a pronounced impact on the phase diagram of 22. R.A.L. Vallée, W. Paul, K. Binder, J. Chem. Phys. 127,
the polymer-solvent mixture [61,64]. For ξ > 1 the poly- 154903 (2007).
mer and solvent attract each other more than they at- 23. J.V. Heffernan, J. Budzien, A.T. Wilson, R.J. Baca, V.J.
tract themselves, and mixing is energetically favorable. Aston, F. Avila, J.D. McCoy, D.B. Adolf, J. Chem. Phys.
126, 184904 (2007).
For ξ < 1, however, polymer and solvent “like” them-
24. R.A. Riggleman, K. Yoshimoto, J.F. Douglas, J.J. de
selves more than each other. This net repulsion between
Pablo, Phys. Rev. Lett. 97, 045502 (2006).
the species may lead to demixing. As toluene is a good
25. R.A. Riggleman, J.F. Douglas, J.J. de Pablo, Phys. Rev.
solvent for PS, we expect that ξ > 1. Therefore, we take E 76, 011504 (2007).
ξ = 1.1314 which gives PS = 0.8 from equation (A.3) 26. S. Peter, H. Meyer, J. Baschnagel, J. Polym. Sci. B 44,
with PP = 1 and SS = 0.5. 2951 (2006).
Finally, for the MD simulation we still have to specify 27. S. Peter, H. Meyer, J. Baschnagel, R. Seemann, J. Phys.:
the masses of the monomer and solvent particles. As the Condens. Matter 19, 205119 (2007).
molar masses of the styrene (mS ) and toluene (mT ) are 28. S. Peter, S. Napolitano, H. Meyer, M. Wübbenhorst, J.
very similar (mS /mT  1.13), we choose mSS = mPP = 1 Baschnagel, to be published in Macromolecules.
in the simulation. 29. K. Kremer, G.S. Grest, J. Chem. Phys. 92, 5057 (1990).
158 The European Physical Journal E

30. M. Pütz, K. Kremer, G.S. Grest, Europhys. Lett. 49, 735 50. W. Kob, in Slow relaxations and nonequilibrium dynamics
(2000), see also the Comment and Reply in Europhys. Lett. in condensed matter, edited by J.-L. Barrat, M. Feigel-
52, 719 (2000); 721. mann, J. Kurchan, J. Dalibard (EDP Sciences/Springer,
31. K. Binder, J. Horbach, W. Kob, W. Paul, F. Varnik, J. Les Ulis/Berlin, 2003) pp. 201–269.
Phys.: Condens. Matter 16, S429 (2004). 51. P. Chaudhuri, L. Berthier, W. Kob, Phys. Rev. Lett. 99,
32. T. Soddemann, B. Dünweg, K. Kremer, Phys. Rev. E 68, 060604 (2007).
046702 (2003). 52. H. Sillescu, J. Non-Cryst. Solids 243, 81 (1999).
33. J. Buchholz, W. Paul, F. Varnik, K. Binder, J. Chem. 53. M.D. Ediger, Annu. Rev. Phys. Chem. 51, 99 (2000).
Phys. 117, 7364 (2002). 54. S.C. Glotzer, J. Non-Cryst. Solids 274, 342 (2000).
34. S.-H. Chong, M. Fuchs, Phys. Rev. Lett. 88, 185702 55. R. Richert, J. Phys.: Condens. Matter 14, R703 (2002).
(2002). 56. L. Berthier, G. Biroli, J.-P. Bouchaud, W. Kob, K.
35. S.-H. Chong, M. Aichele, H. Meyer, M. Fuchs, J. Miyazaki, D.R. Reichman, J. Chem. Phys. 126, 184503
Baschnagel, Phys. Rev. E 76, 051806 (2007). (2007).
36. M. Rubinstein, R.H. Colby, Polymer Physics (Oxford Uni- 57. L. Berthier, G. Biroli, J.-P. Bouchaud, W. Kob, K.
versity Press, Oxford, 2003). Miyazaki, D.R. Reichman, J. Chem. Phys. 126, 184504
37. M. Doi, S.F. Edwards, The Theory of Polymer Dynamics (2007).
(Oxford University Press, Oxford, 1986). 58. J.P. Hansen, I.R. McDonald, Theory of Simple Liquids
38. E. Flenner, G. Szamel, Phys. Rev. E 72, 011205 (2005). (Academic Press, London, 1986).
39. E. Flenner, G. Szamel, Phys. Rev. E 72, 031508 (2005). 59. A.M. Puertas, M. Fuchs, M.E. Cates, J. Chem. Phys. 121
40. W. Götze, Condens. Matter Phys. 1, 873 (1998). (2004).
41. W. Götze, in Proceedings of the Les Houches Summer 60. M. Aichele, Y. Gebremichael, F.W. Starr, J. Baschnagel,
School of Theoretical Physics, Les Houches 1989, Session S.C. Glotzer, J. Chem. Phys. 119, 5290 (2003), publisher’s
LI, edited by J.P. Hansen, D. Levesque, J. Zinn-Justin note: J. Chem. Phys. 120, 6798 (2004).
(North-Holland, Amsterdam, 1991) pp. 287–503. 61. K. Binder, M. Müller, P. Virnau, L.G. MacDowell, Adv.
42. T. Franosch, M. Fuchs, W. Götze, M.R. Mayr, A.P. Singh, Polym. Sci. 173, 1 (2005).
Phys. Rev. E 55, 7153 (1997). 62. M. Müller, G.D. Smith, J. Polym. Sci.: Part B 43, 934
43. M. Fuchs, W. Götze, M.R. Mayr, Phys. Rev. E 58, 3384 (2005).
(1998). 63. P. Virnau, M. Müller, L. Gonzalez MacDowell, K. Binder,
44. A. Rinaldi, F. Sciortino, P. Tartaglia, Phys. Rev. E 63, Comput. Phys. Commun. 147, 378 (2002).
061210 (2001). 64. P. Virnau, M. Müller, L.G. MacDowell, K. Binder, J.
45. T. Voigtmann, A.M. Puertas, M. Fuchs, Phys. Rev. E 70, Chem. Phys. 121, 2169 (2004).
061506 (2004). 65. L. Yelash, M. Müller, W. Paul, K. Binder, J. Chem. Phys.
46. W. Götze, T. Voigtmann, Phys. Rev. E 61, 4133 (2000). 123, 14908 (2005).
47. C.Z.-W. Liu, I. Oppenheim, Physica A 235, 369 (1997). 66. M. de Podesta, Understanding the Properties of Matter
48. M. Fuchs, I. Hofacker, A. Latz, Phys. Rev. A 45, 898 (CRC Press, 2002).
(1992). 67. D. Long, F. Lequeux, Eur. Phys. J. E 4, 371 (2001).
49. M. Fuchs, J. Non-Cryst. Solids 172-174, 241 (1994). 68. L. González MacDowell, M. Müller, C. Vega, K. Binder, J.
Chem. Phys. 113, 419 (2002).
69. K.S. Schweizer, E.J. Saltzman, J. Chem. Phys. 121, 1984
(2004).

You might also like