pubs.acs.

org/JPCB Article

Diffusion Coefficients of Variable-Size Amphiphilic Additives in a

Glass-Forming Polyethylene Matrix
María del Mar Cammarata, Mario D. Contin, R. Martín Negri, and Matias H. Factorovich*
Cite This: https://doi.org/10.1021/acs.jpcb.3c04904 Read Online

ACCESS Metrics & More Article Recommendations *

sı Supporting Information
See https://pubs.acs.org/sharingguidelines for options on how to legitimately share published articles.

ABSTRACT: Diffusion of additives in polymers is an important

issue in the plastics industry since migratory-type molecules are
widely used to tune the properties of polymeric composites.
Predicting the diffusional behavior of new additives can minimize
Downloaded via BOGAZICI UNIV on January 7, 2024 at 13:53:51 (UTC).

the need for repetitive experiments. This work presents molecular

dynamics simulations at the microsecond time scale and uses the
MARTINI force field to estimate self-diffusion coefficients, D, of
six monounsaturated amides and their analogs carboxylic acids in
polyethylene matrices (PE, MW = 5600 Da). The results are
strongly influenced by the glass-forming properties of the PE
matrix, which we characterize by three distinct temperatures. The
metastability region (T < 325 K), the glass transition temperature
(Tg = 256−260 K), and the end of the transition (T ≅ 200 K).
Self-diffusion mechanisms are inferred from the results of the dependence of D on the molecular mass of the additive, observing a
Rouse-like behavior at high temperatures and deviations from it within the metastability region of the matrix. Interestingly, D values
are nonsensitive to the nature of the considered polar head for additives of similar size. The temperature-dependent behavior of D
follows, at fixed additive size, a linear Arrhenius pattern at high temperatures and a super Arrhenius trend at lower temperatures,
which is well represented with a power law equation as predicted by the Mode Coupling Theory (MCT). We offer a conceptual
explanation for the observed super-Arrhenius behavior. This explanation draws on Truhlar and Kohen’s interpretation of the
available energies at both the initial and the transition states along the diffusion pathway. The matrix’s mobility significantly affects
solute self-diffusion, yielding equal activation enthalpies for the Arrhenius region or the same power law parameters for the super-
Arrhenius regime. Finally, we establish a one-to-one time-equivalence of the self-diffusion processes between CG and all-atom
systems for the largest additives and the PE matrix in the high-temperature regime.

I. INTRODUCTION molecular shape, chain length, and branching can modulate the
The presence of additives on the surface of plastic films is diffusion through the matrix.9
crucial for adjusting their properties and making them suitable Additives used for applications such as slip, antiblocking, or
for their numerous applications.1 Additives can be included in antistatic typically exhibit amphiphilic behavior, with a
the polymer formulation through various methods, such as nonpolar long tail and a polar head. The latter causes
drop-casting2 spin-coating,3 and blooming.4 Among these incompatibility between the polymer and the additive,
methods, blooming is commonly used in industrial settings resulting in migration to the surface. Examples of such
for preparing plastics.1 In this process, a mixture of the additives include derivatives of fatty amides, acids, esters,10,11
polymer and the additive is extruded at high temperature, fatty amides such as erucamide7,12 and stearates.13,14
causing the additive to incorporate into the bulk of the material The study of migration, blooming, and/or diffusion of
and then migrate to the surface due to their incompatibility.5 additives has been addressed in the bibliography from different
Eventually, an equilibrium is reached between the additive on points of view. On one side there is the study of diffusion of
the surface and in the bulk. This means that if a certain amount additives in the bulk of the polymer matrix and its migration to
of additive is lost from the surface, it can be replenished by the
remaining additive in the bulk.1 Received: July 20, 2023
Diffusion of these additives from the bulk material to the Revised: December 4, 2023
surface depends on the properties of both the polymer matrix Accepted: December 7, 2023
and the additive. Regarding the polymer, the mobility of the
additive is related to properties, such as degree of crystallinity
and cross-linking.6−8 As for the additive itself, the molar mass,

© XXXX American Chemical Society https://doi.org/10.1021/acs.jpcb.3c04904

A J. Phys. Chem. B XXXX, XXX, XXX−XXX
The Journal of Physical Chemistry B pubs.acs.org/JPCB Article

its surface. This approach is relevant to study the potentiality also accelerate simulation time scales, makes molecular
of different molecules to serve as surface-active additives for dynamics simulation a suitable tool for estimating transport
antiblocking or antistatic applications. Examples of this are the properties of molecules in polymer matrices, without the
works of Quijada-Garrido et al.;15 they studied the diffusion necessity of transport related empirical data as an input for the
and migration of erucamide through isotactic polypropylene, prediction.
finding that the erucamide followed an almost Fickian law of Focusing on the statistical mechanics of polymers, molecular
diffusion in this material. Reynier et al.16 studied the diffusion dynamics simulations were used to test the validity of different
of a broad set of additives into polypropylene, trying to find a theoretically proposed diffusion mechanisms, such as the
correlation between different geometric factors of the Rouse model for unentangled melts, the Zimm model for
molecules and their respective diffusion coefficients. Their diffusion in unentangled dilute solutions,33 and different “tube”
results could not establish a prediction tool for the diffusion of mechanism for long-chain polymers in entangle systems, such
additives in polymer matrices, but provided a qualitative as the reptation34 or the constraint release35 model. Durand et
description of the diffusion mechanism, suggesting crawling for al.36 used molecular dynamics simulations of a bead and spring
long-chain additives and jumps for smaller rigid molecules. model to compute diffusion coefficients of polymer melts,
Works that focus on migration, characterized the time ranging from 2 to 64 beads, and traces of the same nature in a
evolution of the amount of additive in the polymer surface, proportion of 1% to 4%. They found that Rouse’s diffusion
its morphology, and phase state.5,7,9 mechanism was obeyed for all temperatures studied. Shanbhag
Another scenario, where the study of transport properties of et al.37 conducted a comprehensive study on self- and tracer
additives becomes relevant, is in the food,17,18 pharmaceutical, diffusion of polymers, encompassing a range of sizes, including
and medical field,8 where additives are considered a potential those studied by Durand et al. In the case of self-diffusion, they
source of contamination and therefore a threat to human identified two distinct regimes. In systems akin to Durand’s,
health. The work of Hayashi et al.19 is one of the first studies they observed the presence of Rouse-like mechanisms.
where the interaction between plastic packaging and foodstuff However, for longer chains, deviations toward entangled
was analyzed. It presents the diffusion of methyl esters into mechanisms became more prominent, whether using molec-
polypropylene, which serves as a model system to mimic the ular dynamics or Slip-Spring (SS) simulations. As for tracer
penetration of edible oils and fats into plastic containers. The diffusion, the primary findings were derived from SS
results of this work showed a Fickian regime of diffusion. simulations involving long-chain solutes in highly entangled
Another work by Reynier et al.20 shows results of diffusion of matrices, revealing a departure from the Rouse model. Li et
the same additives as in their previous work16 but now into a al.38 studied the dynamic properties of monodisperse polymer
swelling liquid that mimics foodstuff. They found that the melts spanning 12 orders of magnitude in time, using a
swelling liquid has a temperature-like effect, enhancing the free multiscale simulation approach that combined results of united
volume of the matrix and therefore accelerating the diffusion. atoms, coarse grain, and Slip-Spring (SS) simulation. The size
Finally, they outline some key points that models should of the polymers studied spanned from 1 to 34 kDa for the UA
incorporate for predicting diffusion coefficients as a function of or CG approach and reached 200 kDa with the SS model.
different physicochemical properties of the additives and They found what could be interpreted as a smooth change
swelling media. from Rouse dynamics for the smaller polymers to the reptation
Predicting the transport behavior of additives from their regime due to the appearance of entanglements as the
physicochemical properties is of key importance for their polymers got longer.
technological applications and from an economical point of In a work more oriented to the food industry,39 molecular
view. A good model can avoid the necessity of performing dynamics was used to study the diffusion of aldehydes as
repeated diffusion experiments in every new candidate model molecules emulating flavors or additives. The system
molecule that could serve as an additive. Several works tackle was described in a coarse-grain fashion with the MARTINI
this problem, some of them focused on predicting diffusion force field30 and consisted of a polymer matrix above its glass
coefficients by empirical correlations between the experimen- transition, water solvent, and the aldehydes molecules. The
tally measured diffusion coefficients and a variety of parameters diffusion was investigated under gradient conditions, revealing
such as molecular mass, geometry, molecular volume, and that as the chain length increases, diffusion slows down.
activation energy.21−23 Other methods aim to incorporate Additionally, the swelling of the polymer with water had a
experimental data into existing theories, such as the revised substantial impact on the diffusion of the aldehydes.
free volume theory (rFVT),24 in order to test their predictive In this research, we conducted microsecond-scale coarse-
value. Alternatively, some approaches involve retrieving grained (CG) molecular dynamics simulations using the
diffusion coefficients from a more refined microscopic MARTINI force field, to determine the self-diffusion
description of the additives, such as a model of beads and coefficient (D) of additives within a polymeric matrix. We
springs that provides the vibration spectra of the molecule in aimed to assess the capacity of these coarse-grained
question.25 simulations to replicate available experimental data, other
Complementary to diffusion coefficient prediction, other simulations, and theoretical models.
works aim to model the whole migration process by solving the Within this context, the additives are referred to as diffusing
Fickian law of diffusion while incorporating empirical data such compounds, diffusors, diffusants, solutes, or trace molecules.
as partition coefficients, mass transport coefficients, mass The studied additives were amphiphilic molecules with
transfer resistance coefficients, etc.26−28 carboxylic acid or amide groups as polar heads within a
In recent years, computational power has grown in such a polyethylene (PE) matrix. The choice of PE as the matrix and
way that molecular simulation has reached the micro to the inclusion of additives with an amide polar head were
millisecond time scales.29 This, together with the advent of motivated by their significant relevance in the packaging
coarse-grained (CG) force fields and methodologies,30−32 that industry. For instance, erucamide, a member of the amide
B https://doi.org/10.1021/acs.jpcb.3c04904
J. Phys. Chem. B XXXX, XXX, XXX−XXX
The Journal of Physical Chemistry B pubs.acs.org/JPCB Article

Figure 1. (a) Three examples of the parametrized additives with their beads-to-atom correspondence: Light blue: amide (P5) or acid (P3) beads;
Cyan: −(CH2)4− group (C1; m1); Yellow: −(CH2)3− group (C1; m2); Light green: −CH�CH− group (C3). (b) Example configurations of the
simulated systems, consisting of 250 polyethylene chains and 15 additive molecules: Light blue: PE chains; Red: additive.

homologous series, is one of the most widely used sliding MARTINI force field, which has proven in recent years to be
migratory additives. Additionally, we incorporated analogous an suitable tool for studying soft matter related systems and
amphiphilic solutes with carboxylic acid heads for comparative phenomena such as polymeric hydrogel, block copolymer self-
analysis, considering their potential application as new assemble, and microphase segregation of polymers among
additives. other applications.30
We focus on four key factors regarding the behavior of the The matrix consisted of monodisperse branchless polymer
self-diffusion coefficient: (1) its dependence on the nature of chains comprising 100 units (or beads) (MW 5600 Da),
the polar head, by comparing acids with amides; (2) its representing polyethylene. The force field parameters used for
correlation with the molecular weight (MW) of the trace simulating the polymer matrix were adopted from previous
molecules through alterations in the length of the aliphatic works.40,41 This system can be viewed as an approximation to
chain; (3) determining the effect of temperature (T) on D, the National Bureau Standard (NBS standard) polyethylene,
analyzing the different regimes the additives respond to from which has very few branches and low molecular weight
Arrhenius behavior at high temperatures to super-Arrhenius distribution.42
behavior at low temperatures; and (4) comparing the obtained It is important to note that the coarse grain nature of the
self-diffusion coefficients with available experimental data.
force field prevents the PE matrix from crystallizing when
performing simulations below the melting point, even in long
II. METHODOLOGY run times of more than one microsecond. In these conditions,
II.A. Model. We performed molecular simulations of trace full atomistic models of hydrocarbons would easily crystallize.
molecules in a polymeric matrix. All the components of the As a result, we can describe the matrix at all simulated
simulated systems were modeled with the coarse-grained temperatures as a fully amorphous system without any
C https://doi.org/10.1021/acs.jpcb.3c04904
J. Phys. Chem. B XXXX, XXX, XXX−XXX
The Journal of Physical Chemistry B pubs.acs.org/JPCB Article

crystalline domains. This feature is an advantage for the study II.B. Simulation Setup. The simulated system consisted of
because real solid polymer systems are composed of a mixture a polymer matrix composed of 250 PE molecules of 100 C1
of crystal and amorphous regions and it is accepted that the beads each (MW = 5600 Da) and 15 molecules of the additive
diffusion of solutes takes place mostly through the amorphous of interest. Figure 1b presents an example configuration of this
regions.18 system, showing in light green the PE chains and in red the
The diffusing (trace) molecules were monounsaturated additive molecules.
amides and their analogs carboxylic acids comprising two Seven additives were simulated for each homologous series
homologous series. Erucamide, which is a common additive corresponding to the amides or acids. The masses of these
used in the polymer industry, was included in the amide group. additives were comprehended between 101 to 450 g/mol. A
All molecules were parametrized using three types of total of 14 systems were set up and subjected to simulations at
MARTINI beads to account for their different chemical temperatures ranging from 288 to 550 K.
properties. Figure 1a shows a sketch of the parametrization of The setup of the system and the simulations were performed
three of these compounds. Erucamide (X = NH2) or erucic using the tools available in GROMACS 5.1.5.43 The simulation
acid (X = OH), (Z)-9-octadecenamide (X = NH2) or (Z)-9- box was built by randomly inserting 250 PE polymer chains
octadecenoic acid (X = OH), and (Z)-9-tetradecenamide (X = and 15 additive molecules, reaching a low-density system of
NH2) or (Z)-9-tetradecenoic acid (X = OH) are shown. 0.03125 PE molecules/nm3, this was followed by an energy
Hydrocarbon chains were modeled with C1 beads, double minimization. The system was taken to its final equilibrium
bonds with C3 beads, and polar functional groups with P3 state by a two-stage temperature annealing. First from 1000 to
beads for neutral carboxylic acids and P5 beads for amides. 564 K (for 8 ns), then from 564 to 298 K (for 5 ns). Both
The masses of the beads were adjusted to reflect the actual stages were in the isobaric−isothermal (NPT) ensemble at 1
kind and number of atoms included in each bead. Each atm, with a time-step of 5 fs.
MARTINI coarse-grained C1 bead represents a chain of four The production runs consisted of 8 μs simulations in the
single-bonded carbon atoms. However, for molecules that NPT ensemble at 1 atm and the target temperature. To ensure
cannot be divided perfectly in this pattern, a C1 bead is accurate statistical results in determining the diffusion
assigned to a shorter carbon chain with less mass; this is a constant, each temperature was simulated using 16 independ-
common practice when using the MARTINI force field. In ent runs. This was achieved by taking the last frame of the
these homologous series, a group of three single-bonded annealing process and randomly reinitializing the velocities 16
carbon atoms always precedes the double bond and has the times to achieve the desired temperature, one production run
is obtained from each reinitialization. The first microsecond of
same C1 bead designated, but with the mass adjusted to the
these runs was discarded to let the system attain equilibrium at
correct number of carbons. More information regarding the
the new temperature and get diverging trajectories between
intermolecular and intramolecular parameters as well as a full
each replica.
list of the simulated molecules can be found in Section I of the
In all NPT ensemble simulations, temperature and pressure
Supporting Information.
were controlled using the Nosé−Hoover thermostat and the
Nonelectrostatic intermolecular interactions in the MARTI-
MTTK barostat, respectively. Time integration was performed
NI force field are defined according to the Lennard−Jones via the Velocity Verlet algorithm. For the production runs, a
potential, which possesses two parameters. An energetic term large time step of 20 fs was used, allowed by the coarse-grained
(ε), which can be understood as the magnitude of the nature of the force field. The use of the Nosé−Hoover
attractive interaction between two particles, and a character- thermostat plays of key role in determining the diffusion
istic length (σ) which approximates to the radius of the constant in a thermalized ensemble, as is particularly effective
excluded volume between two particles. In this work, we just in preserving the correct transport properties and thermody-
varied the nature of the polar head for the simulated molecules. namic distributions of the simulated system.44
Both groups present the same characteristic length (σX‑PE) and After the annealing and the subsequent 1 μs equilibration
small attraction to the matrix. Nevertheless, the carboxylic period, the PE matrix displays an amorphous structure across
group P3 presents a slightly stronger attraction to the matrix all simulated temperatures, confirmed through visual examina-
(εP3‑PE 15% higher), than the amide group P5. This allowed us tion of the trajectories, as seen in Figure 1b. However, due to
to test how the compatibility between the polar head with the the rapid nature of this annealing, structures below 370 K are
matrix affected the diffusion while keeping the rest of the entrapped in a metastable state, exhibiting continued slow
molecule constant. relaxation throughout the entirety of the production runs. A
To identify the additives, and for the study of size-dependent comprehensive discussion on the implications of this gradual
properties, results are expressed in terms of M/M0 rather than relaxation, specifically in terms of the evolution of the radius of
the number of carbon atoms (Ncarbon) or beads (Nbeads). In this gyration (Rg) and the percentage of free volume (%FreeVol)
case, M0 is the molar mass of the smallest additive molecule during the runs, is provided in Section II of the Supporting
studied, 4-pentanoic amide(acid), Ncarbon = 5, Nbeads = 2, M0 = Information.
101(2) g/mol. This is because the homologous series In brief, although Rg exhibits variations below 370 K during
corresponding to these acids or amides do not always increase production runs, it exerts no discernible influence on other
in a perfectly regular pattern. This is shown in Figure 1a: from properties like %FreeVol, which remains stable throughout the
(Z)-9-tetradecenamide(acid) (N carbon = 14) to (Z)-9- production period. This last property is the most likely to
octadecenamide(acid) (Ncarbon = 18) a C1 bead is added at impact the diffusion process. In line with this last observation,
the end of the molecule, increasing its size after the double there were no significant differences noted in the computation
bond functional group, but going from (Z)-9-octadecenamide- of diffusion coefficients among matrices with different mean Rg
(acid) (Ncarbon = 18) to erucamide (erucic acid) (Ncarbon = 22), values at the same temperature, this is detailed in Section III of
the C1 bead is added before the double bond. the Supporting Information.
D https://doi.org/10.1021/acs.jpcb.3c04904
J. Phys. Chem. B XXXX, XXX, XXX−XXX
The Journal of Physical Chemistry B pubs.acs.org/JPCB Article

II.C. Self-Diffusion Coefficient Calculations. In molec-

ular dynamics simulations, self-diffusion coefficients (from now
on we refer to self-diffusion and diffusion interchangeably for
the sake of simplicity) are readily obtained through the
Einstein diffusion equation:45
N
[ri(t ) ri(0)]2
MSD(t ) = [r(t ) r(0)]2 =
i=1
N
MSD(t )
D = lim
t 6t (1)
where MSD is the mean square displacement, ri is the position
of the particle i at time t, and N is the number of diffusing
particles. The sum in eq 1 can involve different constituents of
a polymeric molecule.46 In this work, we compute two: the
sum over the total number of atoms involved in the diffusion
process (regardless of the additive molecules they belong to), Figure 2. g0 = MSDbeads/6t) and g3 = MSDcom/6t vs t for erucamide at
MSDbeads(t) denoted with the symbol g0, and the sum over the 350 K. Light color curves: computed g0 for 16 replicas of the system.
centers of masses (COM) of these additive chains, which gives Full black thick line: average value of the 16 replicas at each sampled
the MSD(t)COM (g3). At sufficiently long times both time. Dashed blue thick line: Same as the black line but for the COM
computations coincide and it is said that the system has values, computed g3 replicas are not shown. Vertical dashed brown
lines: Region from which Dj and D are computed, the error bar
achieved the free-diffusive regime where Einstein’s equation corresponds to the error of D.
applies.36,46 MSD calculations were performed with the msd
extension that is available in the GROMACS package.
In our simulations, the diffusion coefficients (D) are From this kind of simulation, one can compute the glass
computed by averaging the diffusion constant of 16 replicas, transition temperature by the break of several properties in a
the error in the diffusion values is taken as the 95% confidence property vs T plot.47 In particular, we have chosen to compute
interval of the average from these independent runs: D ± the glass transition from the break in the density (ρ) vs T plot
t(95%,16−1)SDD/161/2. Here, “t” is the statistical Student’s t and from the jump in the heat capacity at constant pressure
function to a 95% confidence, and for n − 1 replicas, SDD (Cp) vs T plot.
denotes the standard deviation linked to the average D. The glass transition temperature (Tg) from the ρ vs T plot
For each replica, the diffusion constant is obtained from an can be determined by extrapolating the two linear parts of the
average value of gx(t)/6t (x = 0 or 3) for all computed times plot before and after Tg or by fitting the results to a hyperbola
after the free-diffusive regime has been achieved, as indicated function.48,49 We have opted for the second option, more
in eq 2. information on the hyperbola fitting function can be found in
Nti > te
gx (ti) ref 49. As for the Cp vs T plot, we identified the glass transition
1 as the midpoint of the transition region between the liquid
Dj =
Nti > te ti > te
6ti (2) state and the glass state.50

Here, Dj refers to the diffusion constant of the j replica, and III. RESULTS AND DISCUSSION
Nti > te is the total number of sampled times after the free
d d

III.A. Approximation to the Glass Transition of the

diffusive regime has been achieved (te). In Figure 2, the Polyethylene Matrix. Polymers are glass-forming systems
previously explained methodology is depicted with the with glass transition temperatures (Tg) typically ranging from
computation of the diffusion coefficient for the Erucamide at 150 to 500 K.51,52 Consequently, the transport properties of
350 K. Only times above 0.01 μs are shown. The black thick polymers, such as viscosity and diffusion, exhibit distinct
line corresponds to the g0/6t averaged over all the 16 replicas, behavior as compared to other solvents that do not exhibit
which are presented as light-colored thin curves in the plot. glass transition.53 This behavior also applies to the diffusion of
Also, the g3/6t curve is presented as a dashed blue thick line. any solute present within these types of matrices.
Both curves can be seen overlapping even before the In this work, we will conduct a systematic study on the
simulation has reached the microsecond, and so attaining the response of the diffusion coefficient of traces in a polymer
free diffusional regime. Between the dashed vertical lines is the matrix to changes in their sizes and the temperature of the
section where the computation for the diffusion coefficient of system. Therefore, as a first step, the glass transition region of
each replica (Dj) takes place. The error bar is a consequence of the polyethylene matrix (MW = 5600 Da) is characterized.
the dispersion of the Dj values, which is computed as explained The dependence of the matrix density (ρ) and the molar
before. enthalpy (H) on temperature (T) is shown in Figure 3a and b,
II.D. Computation of the PE Glass Transition. To better respectively. The derivatives dρ/dT and Cp(T) = dH/dT are
understand how temperature affects the diffusion behavior of also plotted as functions of T. As stated in the methodology
our additive-matrix composites, we determined the glass part, the identification of the glass transition using the density
transition temperature of the pure PE MARTINI model. For data is done by a fitting ρ vs T with a hyperbolic function,49
this purpose, we used a simulation box containing 250 PE of shown in red in Figure 3a. The hyperbolic methodology was
100 beads each, the same as used for the diffusion experiments. chosen over the linear fit extrapolation because it was difficult
We performed a cooling ramp from 600 to 100 K at 0.05 K/ns. to identify a region of constant slope in the low-temperature
E https://doi.org/10.1021/acs.jpcb.3c04904
J. Phys. Chem. B XXXX, XXX, XXX−XXX
The Journal of Physical Chemistry B pubs.acs.org/JPCB Article

From this analysis, the Tg according to the Cp vs T plot is

located at 260 K.
Experimental measurements indicate that linear amorphous
polyethylene exhibits a glass transition at approximately 195 K,
with a range spanning from 190 to 200 K.55 Deviations
between our simulation results and these experiments cannot
be attributed to the smaller size of our polyethylene chains in
comparison to the experimental setup, as this would normally
lead to lower Tg values.33 Given the length of our PE with MW
= 5600 Da and considering the fast convergence of Tg with
MW,33 as well as estimates of Tg from viscosity measure-
ments56 and molecular simulations57 on alkenes with varying
chain lengths, we can reasonably conclude that our poly-
ethylene model should have attained a converged Tg value
relative to the chain length. Thus, any disparities in Tg likely
arise from the intrinsic aspects of the force field’s para-
metrization, where the experimental value of Tg was not taken
into account, as well as variations in the cooling ramp
employed to calculate Tg as compared to experiments.
III.B. Reaching the Free-Diffusive Regime at the
Microsecond Time Scale. As stated in the methodology
part, the superposition of g3 and g0 indicates that a system has
reached the free diffusional regime. For small solutes,
depending on the temperature and the specific considered
matrix, this could be achieved in the picosecond or
nanosecond time scales. Nevertheless, for long-chain additives,
the advent of the free-diffusive regime could be in the order of
tens to hundreds of nanoseconds. Figure 4 shows plots of gx/6t

Figure 3. (a) Black Y-axis corresponds to the density of the polymer

matrix (ρ) vs temperature plots. Black points account for ρ (not all
points are plotted, for clarity). Red line corresponds to the hyperbolic
fit. Tg is indicated with a dashed brown vertical line. Blue Y-axis is the
derivative of ρ. The curve has undergone a smoothing process due to
the high noise associated with deriving scatter data points. (b) Black
Y-axis corresponds to the matrix molar enthalpy (H) vs temperature
plot, presented as a black line. Blue Y-axis corresponds to the molar
heat capacity (Cp) vs temperature represented in the same blue color.
Both plots have been subject to a smoothing process. Red dashed
vertical line marks the beginning of the metastable zone.

zone. The value of Tg recovered from ρ vs T is 256 K, with

ρ(Tg) = 0.894 g/cm3. Another point to highlight in Figure 3a is
the density at 450 K, which was investigated in the original
paper that parametrized the PE with MARTINI.40 In that
work, the calculated density for a 20-bead polymer simulation
box was 0.760 g/cm3 at 450 K, and it has a predicted density of Figure 4. Full lines, g0(t)/6t or dashed lines, g3(t)/6t vs time,
0.773 g/cm3 for an infinite long polymer. In the present work, spanning the whole simulation. Two selected additives ((Z)-9-
the calculated density for 250 PE molecules with 100 beads tetradecenoic acid and (Z)-13-tridecenoic acid) and two temperatures
are shown. Vertical brown dashed line: te, lower time considered for
each was 0.780 g/cm3, less than 1% from the predicted value the computation of D and Dj.
for the model and less than 2% for the experimental value42 of
0.766 g/cm3.
The value of Tg can also be recovered from the plot of Cp vs
T (Figure 3b, blue line) by taking Tg as the midpoint for x = 0 (the beads) and 3 (the center of mass) of the (Z)-9-
temperature of the observed glass transition region.50 Going tetradecenoic acid (Ncarbons = 14; Nbeads = 7; M/M0 = 3.3) and
from higher to lower temperatures, the glass transition begins (Z)-13-tridecenoic acid (Ncarbons = 30; Nbeads = 9; M/M0 = 4.4)
with the peak in the heat capacity of the system (325 K for our at 288 and 390 K. At both temperatures, g3 and g0 exhibit an
system), this point also marks the beginning of the overlap at times reaching the microsecond, as highlighted by
metastability zone and so, below this temperature, the system the vertical dashed brown line in Figure 4. While this
is not at thermodynamic equilibrium anymore, therefore the H superposition is slightly less pronounced for larger solutes,
vs T curve depends on the cooling rate.54 The end of the glass the diffusion constants determined from g3 and g0 are nearly
transition can be identified by a decreasing slope (in absolute identical within the margin of error, D(g0) = 1.8 ± 0.2 nm2/μs
terms) of the Cp vs T curve (around 200 K for our system). and D(g3) = 1.6 ± 0.2 nm2/μs, for M/M0 = 4.4 at 288 K.
F https://doi.org/10.1021/acs.jpcb.3c04904
J. Phys. Chem. B XXXX, XXX, XXX−XXX
The Journal of Physical Chemistry B pubs.acs.org/JPCB Article

We can conclude that, for all our simulations, production for both acids and amides. No significant difference in the
run times were long enough for the systems to achieve the free- values of D is observed for acids (red circles) and amides (blue
diffusive regime, where diffusion coefficients D can be empty triangles) at a given T and M/M0. Results in literature
accurately calculated from Einstein’s relation (eq 1). Never- concerning the diffusion dependence on the hydrophilic nature
theless, it is clear that, for the bigger diffusants, the plateau of of the additive at constant chain length show a variety of
the plot is reached very near the end of the simulations and so behaviors. For example, when comparing alcohols and
diffusion values for 30C at lower temperatures are subject to alkanes27,61 in low-density polyethylene, no significant differ-
more error. The errors arising from not reaching the free- ence for the same number of carbon atoms is observed; this is
diffusion regime are discussed in section III.D in the context of on the same line as our results. Nevertheless, when added
the dependence of the diffusion constant with temperature. results for esterified phenols61,62 in the same polymeric matrix,
It is important to highlight that dynamics in coarse-grained these last ones show a smaller diffusion constant when
models might inherently be accelerated due to the reduced compared with the same number of tail carbons. It has been
degrees of freedom and the smoother energy landscape when hypothesized that the hydrophilic nature of the head of the
compared to experimental systems or full atomistic simu- polymeric additives could have consequences in the diffusion
lations.32,58 To extract accurate dynamical information from constant when aggregates of these molecules could form.16
coarse-grained simulations, it is imperative to account for this Hydrophilic-headed additives in a hydrophobic matrix may
acceleration. tend to aggregate given enough time and concentration. This is
Based on available experimental data, we have estimated the not the case in the considered concentration (15 additives per
scaling factor, β, according to its original definition used in 250 polymer chains). During the simulation time, the solutes
MARTINI,59 βDCG = Dexp, to be approximately β ≅ 1 for remain mostly separated from each other, there are fortuitous
temperatures above 450 K for both the polyethylene matrix encounters of no more than two additive molecules, and they
and the three largest additives. This implies an almost one-to- never stay aggregated. This behavior approximates a very dilute
one time mapping from coarse-grained to all-atom representa- polymer−additive solution.
tions. We are not able to make predictions on the conversion Summarizing, our calculations show that similar values of D
factor at lower temperatures, but it should decrease (CG are recovered for amides and carboxylic acids of equivalent
should accelerate) due to the lower activation free energy length chain, for the considered concentration and range of
associated with coarse-grained dynamics as compared to all- temperatures. The observed behavior corresponds with the
atoms.60 nonpolar nature of the matrix and its interaction with the
An in-depth discussion of how we arrived at this equivalence solutes. From the perspective of force field parametrization,
is presented later in the text, in section III.E, as it builds upon this can be attributed to the fact that while both polar heads
the conclusions derived from the following sections. have slightly different energetic terms (εP5;P3‑PE), the actual
III.C. Influence of the Hydrophilic Group and the magnitude of this parameter for both species is quite low. This
Length of the Chain on the Diffusion Coefficient. In this results in a repulsive effect on the matrix, ultimately leaving the
section, we focus our attention on the dependence of the D on diffusion outcome dependent on the functional group size,
the hydrophilic terminal group and chain length for the which remains equivalent (sharing the same σP5;P3‑PE).
different studied solutes. Figure 5 shows general results of the The dependence of D with M (or N) was fitted using power
recovered values of D as a function of temperature, T, at fixed laws of D ∼ (M/M0)−α or N−α. Theoretically, α can take
M/M0. As expected, D increases with T (at fixed M/M0) and several values depending on the suppositions regarding the
decreases with M/M0 (at fixed T). This behavior is obtained diffusion process,33 and so its magnitude can provide
information about the mechanism of diffusion that is taking
place.
On one hand, the so-called Rouse model describes the
diffusion of relatively large objects in continuous media where
the matrix and the diffusor are not coupled, predicting a value
of α = 1. On the other hand, the Zimm mode applies when the
diffusing object and the matrix media are hydrodynamically
coupled, meaning that, the trace diffusor drags the media,
usually a solvent, within its pervaded volume, in the Zimm
model α takes a value of 0.5.
For motions on entangled polymer melts, different “tube
models” have been proposed. The simplest of these models is
the reptation model, which can describe the motion of a chain
in a fixed network or in a melt of extremely long chains. This
model predicts a value α = 2. Finally, when the diffusor is a
polymer with physical properties similar to those of the matrix,
then, more complex mechanisms can be envisioned, one of
such being constraint release, with a complex dependence of D
Figure 5. Overview of calculated diffusion constants, D, as a function
with M or N.35
of temperature, T, for selected additives: red circles, acids; empty blue The power parameter α can also be a function of the
triangles, amides. Additives show the following: 4-pentanamide(acid) temperature, and it is experimentally observed to be greater
M/M0 = 1; (Z)-9-octadecenamide(acid) M/M0 = 2.8; (Z)-13- than one near the glass transition (Tg), but in general, rapidly
tridecenamide(acid) M/M0 = 4.4. drops as temperature departs from Tg.61
G https://doi.org/10.1021/acs.jpcb.3c04904
J. Phys. Chem. B XXXX, XXX, XXX−XXX
The Journal of Physical Chemistry B pubs.acs.org/JPCB Article

In Figure 6a, we present results for the diffusion coefficient In literature, the power coefficient α exhibits a broad range
as a function of M/M0 in logarithmic scale for both axes at of variation, in both experiments and simulations, often
deviating from the ones proposed by the aforementioned
models or theories. Our results are similar to those observed by
the molecular simulations of Durand et al. for tracer diffusion
in polyethylene,36 but present a mild dependence of α with the
temperature which was not observed in their work. As for
experiments, we observed the same trend as in trace
experiments of alkanes of increasing molecular weight, between
24 and 60 units, in high-density polyethylene.61,63
The temperature-dependent variation of α, which becomes
evident inside the metastability zone (T < 325 K) where the
matrix undergoes a discernible loss of fluidity, could be
attributed to several factors. One possibility is a transition from
a Rouse-like response to a more intricate regime at lower
temperatures. This new regime may involve a combination of
the previously mentioned more complex mechanisms, leading
to dynamic heterogeneities. Another explanation, in alignment
with the findings of Durand et al.36 and Meerwall et al.,63,64
suggests that the alteration in α could be linked to a change in
the friction coefficient between the monomers and the matrix,
rather than a fundamental shift in the underlying mechanisms.
This change in the friction coefficient would exert a more
pronounced influence at lower temperatures.
III.D. Temperature Behavior of the Diffusion Con-
stant. The dependence of the diffusion constant with
temperature (T) for all the amides and acids was investigated
in the range of 288 to 555 K, all above the glass transition
temperature but with some points inside the metastability
region of the polymer matrix. The results for selected
molecules are presented in Figure 7a as Arrhenius plots (D
in natural logarithmic scale vs T in reciprocal scale), the full
data set of additives studied are presented in Supporting
Information, Section V. We also include, as a gray dashed line,
the values of MSD/6t for the polyethylene matrix when its
monomers have traversed a distance equivalent to their square
size, denoted as the segmental relaxation distance, over time,
Figure 6. (a) D vs M/M0 in logarithmic scale for both axes. Selected σ2/6t. In our analysis, we take this characteristic length as the
temperatures are shown: red full circles, acids; open up triangles, parameter σ of PE in the MARTINI force field. This allows us
amides. Full lines are fittings to the power function presented in the to compare the dynamics of the solute with an estimate of the
main text. (b) α vs temperature. The color is the same as for (a). The dynamics of the polymer matrix.
dashed black line highlights α = 1, corresponding to the value It is observed that all Arrhenius plots are not linear,
predicted by the Rouse model.
presenting the convex curvature, referred to as super-Arrhenius
behavior in the literature.36,51,53,65−70 In these cases, the slope
of the Arrhenius plots increases when temperature decreases.
different temperatures ranging from 288 to 550 K. The full Moreover, it seems that this growth in the slope undergoes an
lines represent the fits of the power function D = D∞ + (D1 − enhancement below 350−325 K, which corresponds to the
D∞)·(M/M0)−α. In this equation, D∞ is the diffusion constant onset of the polymer’s metastable region. The linear Arrhenius
for an infinitely large additive and D1 is the value of diffusion behavior is only approached in the high-temperature regime,
for M = M0. D∞ is expected to have a very low value (orders of typically for T greater than 450 K in our simulation. The
magnitude lower than D1 at any temperature), as shown in dynamics of the matrix mirrors that of the solutes, displaying a
Figure S.9 in Section IV of the Supporting Information. comparable linear region and convex pattern. Additionally, the
In Figure 6b, the recovered values for α at each temperature values for σ2PE/6t, associated with a segmental relaxation of a
are presented together with the error bars resulting from the single PE bead, closely resemble those of the smaller solute (2
fits. It can be seen that, for the additives tested here, the values beads) within the matrix. However, it exhibits a more
of α are not significantly different within their uncertainty and pronounced temperature trend, particularly notable at lower
are all comprehended between approximately 0.5 and 1.5 for temperatures.
all the temperatures. Nevertheless, a trend can be discerned, α The temperature above which Arrhenius-linear behavior is
tends to one at high temperatures and has values larger than 1 observed is referred to as TA in the bibliography.53 Hence, in
for temperatures lower than 325 K, which coincide with the our case TA ≅ 450 K, although this value is slightly dependent
beginning of the metastability zone and glass transition region on M/M0, for acids and amides. Above TA, Arrhenius fits,
for the matrix. This also coincides with the aforementioned together with the transition state theory (TST),51,71 can be
expected behavior for the temperature dependence of α.61 used to obtain thermodynamic data associated with the
H https://doi.org/10.1021/acs.jpcb.3c04904
J. Phys. Chem. B XXXX, XXX, XXX−XXX
The Journal of Physical Chemistry B pubs.acs.org/JPCB Article

Figure 7. (a) Diffusion coefficients (logarithmic scale) vs temperature

(inverse reciprocal temperature scale) for selected acids and amides:
circle, acids; up triangles, amides. The colors are specific to a M/M0
series, as shown in the plot. The gray dashed line indicates σ2PE/6t for Figure 8. (a) Activations enthalpy (ΔHa) vs size of the additives (M/
PE; the line is not a fit, but rather connects the simulated data points, M0): red circles, acids; blue up triangles, amides. The horizontal black
which are intentionally omitted for better visual clarity. The vertical dashed line corresponds to the average for all the data as a visual
dashed line indicates the temperature of the peak in the Cp for the guide. The horizontal brown dashed line corresponds to ΔHa
temperature ramp. The error of the data is of the same size as the associated with σ2PE/6t vs T for PE. (b) y-Intercept of the Arrhenius
symbols. (b) Arrhenius scaled plots (same scale as (a)) for selected fits ln(D0) vs size of the additives (M/M0). The horizontal brown
additives in the temperature range where a linear behavior can be dashed line corresponds to the y-intercept associated with σ2PE/6t vs
observed: red dots, acids; blue open-up triangles, amides; gray dots, T for PE. The error bars are confidence intervals to 95% from the
PE; full lines, TST fits, keeping the color reference; dashed lines, Arrhenius fits.
extrapolation of the fits outside the range of temperatures used in the
regression; vertical dashed line, end of the linear region.
In principle, it might seem counterintuitive that larger
additives do not show higher activation enthalpies. However,
diffusion process, by fitting the results for D vs T, with D = D0· this observation actually supports the idea that the diffusion
exp[−ΔHa/(RT)] or its linearized form ln(D) = ln(D0) − pathways are governed by the mobility and available free
ΔHa/(RT). These fits are shown in Figure 7b, which depicts an volume within the polymer matrix. This aligns with the
enhancement of the region for T > 450 K in the same scale as fundamental premise of the free volume theory in explaining
Figure 7a. The full lines are the adjusted curves inside the diffusion. It could be hypothesized that, in the Arrhenius
region of the data point used for the fit whereas break lines are regime zone, the diffusant moves through the available pores of
extrapolations. the polymer matrix, and the temperature-dependent diffusion
The thermodynamic parameters obtained by fitting our data behavior is attributed to changes in the fluidity of the matrix,
to the TST equations are presented in Figure 8a,b for the affecting both the size and the renewal rate of the pores.18,72
amides, the acids, and the segmental relaxation of the matrix Also, the ΔHa link to the segmental relaxation of the matrix is
(brown dashed line). Figure 8a shows the results for ΔHa as a comparable to that associated with the diffusion of the solutes.
function of M/M0. Overall, the results for all the acids and Although slightly higher, it remains statistically equal within
amides show a similar and almost constant value for the the margin of error and might indicate a potential upper bound
activation enthalpy around 19 kJ/mol. The constant nature of for the ΔHa in this system. This further points to the intimate
ΔHa concerning the size of the additive agrees with previous relationship between the dynamics of the matrix and the
experimental63 and simulation36 results. solutes.
I https://doi.org/10.1021/acs.jpcb.3c04904
J. Phys. Chem. B XXXX, XXX, XXX−XXX
The Journal of Physical Chemistry B pubs.acs.org/JPCB Article

The y-intercept of the fit accounts for ln(D0), which is the to be 1/τx = Ax·(T − TC)γ, where τx is the relaxation time for a
diffusion at a hypothetical infinite temperature. This is shown correlation function x, and TC and γ are universal parameters,
in Figure 8b, where ln(D0) diminishes as M/M0 increases. meaning that they are equal for all correlation functions,
According to TST, these values are related to the entropic whereas Ax is a proportionality constant specific for x.77 As a
contribution to the diffusion process. It could be interpreted consequence of the universality of TC and γ, the temperature
that smaller diffusers are capable of accessing a greater number dependence of the diffusion constant (D) should also obey this
of sectors in the matrix (a larger number of configurations in power law, as it is a particular case of the incoherent
phase space) compared to larger ones, even at high intermediate scattering function.77 Hence, D = 1/τD = AD·(T −
temperatures. The intercept for the segmental relaxation of TC)γ, which is the fitting function in Figure 9. It is worth
PE exceeds the rest of the ln(D0) values due to its higher noticing that, although MCT is rigorously developed for pure
activation enthalpy as compared to solute diffusion in the glass-forming substances, here it is used to adjust the diffusion
matrix. Additionally, the ln(D0) value for the PE bead results for the tracer molecules in the polymer matrix. This is
relaxation is not far from the two-bead solute M/M0 = 1, not uncommon in literature for solutes of different natures in a
again showing a possible upper bound to the value of ln(D0). glass-forming media.36,78,79 Also, theoretical approaches have
We now turn our attention to the departure from the linear been made to extend MCT to solutions.80
Arrhenius behavior, characterized by the emergence of a super- The dependence of the recovered parameters with M/M0 is
Arrhenius or convex Arrhenius trend. It is important to note presented in Figure 10a for TC and (b) for γ. AD is left apart
that, in order to reach the correct super Arrhenius behavior at because it is just a proportionality constant.
low temperatures, sufficient long times must be simulated.36,73 According to MCT, Tc is the predicted “theoretical” glass
Also, other more sophisticated methodologies can be used to transition temperature, it is a temperature of divergence where
obtain the correct diffusion coefficient from short simula- the relaxation time τx for all correlation functions becomes
tions.69,70 Otherwise, at low temperatures, short simulations infinitely long, and the system freezes. Obviously, this behavior
tend to give errors by excess in the calculation of the diffusion
constant, and so the deviations from linear Arrhenius trends
are not observed.
The super-Arrhenius trend is usually fitted by two models.
The Vogel−Fulcher−Tammann (VFT) model, which can be
derived from generalized entropy theory,53,68 or the so-called
power-law model, which is usually discussed within the frame
of the Mode Coupling Theory (MCT).74
We will further focus on the MCT, which was able to better
adjust to our data in the whole temperature range. This is
shown in Figure 9, as Arrhenius plots (D in logarithmic scale vs

Figure 9. Arrhenius scaled plots (same scale as in Figure 7) for

selected additives: red dots, acids; blue open-up triangles, amides. Full
lines represent the MCT power law fits keeping the color reference.

T in reciprocal scale) for selected diffuser molecules and for

the segmental relaxation of the polyethylene monomers, which
is shown as a horizontal brown dashed line.
MCT predicts that a glass-forming system approaching its Figure 10. (a) TC vs M/M0 for all the additives studied: red circles,
glass transition has three distinctive relaxation regimes with acids; blue up triangles, amides. The black dashed line corresponds to
different characteristic time scales, identified as fast relaxation, the average for M/M0 > 1. The brown dashed line corresponds to TC
β-relaxation, and α-relaxation.75,74,76,77 In particular, for any for the segmental relaxation of the matrix. (b) Power coefficient γ vs
time correlation function in the α-relaxation regime, the M/M0 for all the additives studied; symbols and coloring are the same
temperature dependence of its characteristic time is predicted as for (a).

J https://doi.org/10.1021/acs.jpcb.3c04904
J. Phys. Chem. B XXXX, XXX, XXX−XXX
The Journal of Physical Chemistry B pubs.acs.org/JPCB Article

Figure 11. Schemes associated with linear, concave, and convex Arrhenius plots. (a) Linear-Arrhenius plots. (b) Concave Arrhenius, showing the
case of only two competitive diffusion paths (c) and (b). (c) Convex Arrhenius (usually referred to as Super-Arrhenius) illustrates the case of only
two initial states, I1 and I2.

is not actually seen in experiments or simulations and the real the tracer in our system lies between 1.03 and 1.07. In
glass transition does not correspond to Tc. Actually, the bibliography, the available data is almost exclusively found for
computed Tc usually lies between the real Tg and the start of pure systems. Experiments in different polymers show values of
the Arrhenius behavior TA. Nevertheless, this temperature is Tc/Tg between 1.06 to 1.35,83,84 whereas for simulations the
associated with the beginning of a region where real distinctive values are around 1.06 to 1.1.36,76 Again, the discrepancy
changes occur in the time-dependent properties of the glass- between experiments and simulations might be associated with
forming system, before the actual glass transition.53 Around Tc, the overestimation in Tg by simulations due to the much faster
for example, a breakdown in the Stokes−Einstein relation cooling times.
between the diffusion coefficient and the fluid shear viscosity81 The power coefficient γ controls the temperature evolution
is observed, and the clear bifurcation of structural relaxation of the α-relaxation (how fast τx grows or D decreases as T
times into a fast and a slow one.82 Our results show that, for all approaches Tc), but it actually contains information on the
the diffusers, Tc ≅ 268−273 K, without any dependence with shape of the time decay of both β and fast relaxation, showing
M/M0, except for particular anomalous low values for the that the three relaxation regimes are intimately related.74,75,77,85
smaller diffuser. All results are above the calculated glass The results in Figure 10b depict a slight dependence for γ with
transition, Tg ≅ 256−260 K. The fact that Tc remains constant the mass of the diffusers, which except for the lower value of
concerning M/M0 parallels the constant activation enthalpy in M/M0 could be considered under the uncertainties of our
the Arrhenius regime. The result for the PE matrix follows the results. For M/M0 = 1, γ ≅ 2.35, whereas for the other values
same trend as the solutes with a Tc ≅ 272 K. This is further the range is between 2.25 to 2.12. The value of γ associated
evidence that the dynamics of the whole system should be with the segmental relaxation of the matrix is slightly larger
conditioned by the polymeric matrix. The quotient Tc/Tg for than that for the solutes. This emphasizes the significant
K https://doi.org/10.1021/acs.jpcb.3c04904
J. Phys. Chem. B XXXX, XXX, XXX−XXX
The Journal of Physical Chemistry B pubs.acs.org/JPCB Article

influence of temperature on the matrix. At elevated temper- initial states. Considering, just for simplicity, only two initial
atures, σ2PE/6t displays values akin to those of the smallest states, the model can be presented in a potential−barrier plot,
solutes. However, as temperatures decrease, the curve diverges, as in Figure 11c. In this figure, two different initial possibilities,
resulting in values comparable to those of the solutes of larger I1 and I2, are considered, with E1 < E2 (thus, I1 is the trapped
mass. situation), where E1 and E2 are the thermodynamic equilibrium
Regarding the distinct behavior of the smallest molecules energies associated with each situation, respectively. The
(M/M0 = 1), we can hypothesize that they are capable of diffusant, starting from I1 or I2, diffuses toward a final situation
moving through the network even when it has entered the through the potential curve. This model must include the
regime of glass transition, thus exhibiting the smallest Tc value. constraint, already indicated by Truhlar and Kohen,65 that the
In contrast, the other molecules (M/M0 > 1) are highly thermodynamic average energy at the top of the barrier must
influenced by the freezing of the network, resulting in them change less with T than does E1 and E2. In the scheme of
showing the same Tc value, which matches that of the matrix Figure 11c, this constraint is incorporated by assuming that the
itself. potential energy at the transition state (top of the potential
Experiments in pure polymers and other glass-forming curve), referred to as E3, is similar for diffusers that started
liquids show a big dispersion in the values of γ ranging from from I1 or I2 (in the notation of Truhlar and Kohen: E3 = E̿ ,
3.6 to 2.2.85−89 In simulations, a dispersion in the values is while E1, E2 = E̅ ). This scheme is consistent with the super-
observed according to the coarse-grained nature of the model. Arrhenius behavior: when T decreases, the diffuser is mainly
All-atom models tend to have greater γ values.89 Here our located at the trapped state I1 (E1 < E2), and it must surpass a
polymeric matrix−traces system with the MARTINI force field relatively larger activation barrier equal to E3 − E1. At a higher
is similar to the bead−spring simulations of polymers, which temperature, the trace diffuser is distributed between I1 and I2,
show values for γ ≅ 2−3.78,89 As γ is a parameter related to the and a larger proportion of diffusers have energy E2; therefore, it
relaxation times of the system, discrepancy between all-atom must surpass a relatively lower activation barrier (E3 − E2).
and coarse-grained simulations could be associated with the The result is that the observed activation energy, ∂ ln(D)/∂(1/
accelerated nature of the last ones, which becomes more T), is higher at the lowest temperatures, which corresponds to
evident at lower temperatures.60 the super-Arrhenius behavior.
MCT is capable of predicting the super-Arrhenius (convex III.E. Estimate of the Coarse-Grained to All-Atom
Arrhenius deviations) behavior by modeling the different Time Equivalence. Several methods exist to extract accurate
relaxation times of the system as it approaches the glass dynamical information from coarse-grained simulations.32,58
transition. However, there are other approaches to describe They vary in complexity, but all require calibration from
this behavior in terms of the energetics of the system, such as experimental or all-atom simulation data. One such approach is
the work of Truhlar and Kohen,65 which explains both concave time-remapping, which involves determining the speed-up
and convex deviations, from which here we elaborate factor of coarse-grained simulations by comparing them to
conceptually. available experimental or full atomistic data and then making
The concave deviations are not seen in this work and are corresponding adjustments. Another method, energy renorm-
also not generally observed with the temperature dependence alization, is based on the theoretical framework of generalized
of transport properties in glass-forming systems. This deviation entropy theory. It postulates that time rescaling within a given
is consistent with the possibility of many possible paths of temperature range can be achieved by fine-tuning the energy
diffusion, connecting an initial state, I, and a final state, F. parameter (ε) of the force field potential relative to
Hypothetically, describing the diffusion in our system, I and F temperature, using a sigmoidal function. It necessitates full
would be associated with the initial and final locations of the all-atom or experimental data within the temperature range of
trace diffusant in the phase space. As an example of this system, interest.
one can imagine two possible paths, denoted A and B, as Additionally, more sophisticated techniques incorporate the
shown in Figure 11b. These two paths are competitive introduction of friction coefficients into the motion equations
processes with activation energies EAa > EBa . When T is to counteract the ’smoothing’ effect of coarse-graining. This
decreased, the more probable path corresponds to the process involves employing Langevin dynamics and introducing
with the lower activation energy, thus the slope of ln(D) vs 1/ supplementary ’friction terms’ that bridge the dynamics from
T decreases when T decreases. In other words, ∂ ln(D)/∂(1/ coarse-grained to full atomistic representation. These friction
T) → EBa when T decreases, while ∂ ln(D)/∂(1/T) → EAa when terms may either remain constant throughout the simulations
T increases, generating a concave Arrhenius plot, as shown in or vary over time (in the form of memory terms, resulting in
Figure 11b, right. Within this context, a linear-Arrhenius plot is non-Markovian dynamics) and can be obtained with brief all-
a particular case that corresponds to the situation of only one atom simulations.58 This area is currently undergoing
possible path connecting I and F, with activation energy Ea significant development, but is marked by notable complexity,
(Figure 11a). particularly when incorporating time-dependent friction
However, although this picture provides an interpretation coefficients (memory terms) that make the dynamics non-
for linear and concave Arrhenius plots, it cannot predict the Markovian.
convex behavior observed in our system (super-Arrhenius The time-scaling factor (β) is found to be temperature-
plots). Therefore, the super-Arrhenius behavior must be dependent. This phenomenon has been both empirically
rationalized on a different basis. For instance, Debenedetti observed through time-remapping90 and theoretically pre-
and Stillinger67 in studies of self-diffusion of supercooled dicted by Energy Renormalization. This prediction is based on
liquids, associated the super-Arrhenius behavior with the the disparity in activation-free energies associated with
possibility of several local minima for the energy (cool traps) dynamical processes between coarse-grained models and all-
of the trace diffusers. This idea can be adapted to Truhlar and atom or experimental systems. Notably, the study by Xia et
Kohen’s interpretation by assuming a distribution of localized al.60 showed that the scaling factor approximately doubles
L https://doi.org/10.1021/acs.jpcb.3c04904
J. Phys. Chem. B XXXX, XXX, XXX−XXX
The Journal of Physical Chemistry B pubs.acs.org/JPCB Article

Table 1. Diffusion Coefficient for Acids and Amides at 450 K and Experimental Data from Von Meerwall et al.63a
simulation Nbeads Dsim × 1011 (m2/s) β = Dexp/DCG Dexp × 1011 (m2/s) exp
b
M/M0 = 3.3; C22 7 51
avg (C22; C26); C24 7.5c 47 0.88 42 C24H50
M/M0 = 3.9; C26 8 43b
avg (C26;C30); C28 8.5c 41 0.86 36 C28H50
M/M0 = 4.4; C30 9 39b
a
The data from simulations is averaged to compare with experiments at the same number of carbon atoms. bResults are averages from Amides and
Acids. chypothetical number of beads that results from the average.

when transitioning from the high-temperature Arrhenius The results in Table 1 indicate again, nearly one-to-one
regime to temperatures below the glass transition region. correspondence between the diffusion values of the coarse-
However, it remains constant within these respective regions. grained model and the experimental results.
Building upon the preceding discussion, we estimate β = The established MARTINI scaling factor, as determined
Dexp/DCG by performing time remapping with the available from water diffusion studies,59,91 results in a value of 4. This
experimental data. Other more sophisticated approaches means that, to bridge the dynamics results from coarse-grain to
should be left for future investigations due to the amount of experiments or all-atom simulations, it should be necessary to
simulated data in this work. multiply the CG results by 4. The discrepancy observed with
We will be comparing our results with available NMR field these large polymer simulations may be attributed to the
gradient studies that assess self-diffusion. This experimental coarse-graining procedure of water in MARTINI. This involves
the amalgamation of four independent water molecules into
technique is most compatible with the determination of self-
one. Consequently, when calculating the diffusion coefficient
diffusion coefficients from MSD in simulations, as opposed to
of CG water, one is essentially examining the dynamics of the
other gradient-driven experiments for determining diffusion center of mass of a body equivalent to 4 water molecules. In
coefficients. contrast, for larger molecules, coarse-graining occurs at subunit
Person et al.42 utilized spin echo NMR to evaluate the levels, and so, the individual molecules are preserved. From
diffusion coefficients of highly monodisperse linear poly- this, we can hypothesize that the larger the molecule, the less
ethylene standards, spanning a molecular weight range from influence coarse-graining exerts on the long-term dynamics,
200 to 120000 Da. Figure 6 of their work presents the results such as the self-diffusion coefficient. The scaling factor 4, might
for the diffusion coefficient at 448 K. From this figure, we be recovered at shorter time scales if studying, for example,
extracted the interpolated diffusion coefficient for a 5600 Da dynamics of individual beads and comparing directly to atoms
PE molecule, Dexp = 6.02 × 10−12 m2/s. in all atom simulations.
In our work, the polyethylene matrix does not reach the free- Based on the preceding analysis, it can be expected that
diffusive regime where g0 = g3. Nevertheless, a good estimate of within the Arrhenius regime (above 450 K), both the
the diffusion coefficient can be computed from g3/6t at high polyethylene matrix and the largest solutes exhibit a time
temperatures, because it has reached an almost constant value. scaling factor of approximately β = 0.9−1.1. To fully
We show this in Figure S12 of the Supporting Information at understand the time behavior of the coarse-grained systems a
450 K. In this figure, g3/6t remains almost constant above 1 μs, more systematic study complemed with atomistic simulations
varying by less than 5%. Hence, we estimate D with the should be done. This is especially crucial for the smaller
converged value of g3/6t = DPE(450 K) = 5.4 × 10−12 m2/s. solutes, where, as discussed, coarse-graining is anticipated to
exert a more pronounced effect on dynamics. Additionally, it is
This gives a ratio β = Dexp/DCG = 1.1, which indicates an
essential at lower temperatures where greater deviations in
almost one-to-one correlation between the model and the
dynamics are expected due to the lower activation energy in
experiment. A similar correspondence is observed for a smaller coarse-grained systems, resulting in faster processes as
size PE (MW = 2240 g/mol) in a work by Khan et. al, also compared to all-atom ones. This investigation should be left
modeled with MARTINI. for future work.
Regarding solutes in polymer matrices, the available data
pertains to the trace diffusion of alkanes. In section III.C we IV. CONCLUSIONS
showed that, for our solutes, the nature of their polar heads
appears to not influence their self-diffusion constant inside the We characterized the glass transition temperature for the pure
polymer matrix. This could be attributed to both the low PE with two properties, the dependences of density and heat
capacity with temperature, getting an agreement within 2%
concentration of the traces and the intrinsic nonpolar nature of
between both methodologies. However, showing significant
the matrix. Therefore, we may draw comparisons between our
differences between the simulated and experimental values, we
data and the available data for alkanes. attributed this to intrinsic aspects of the force field’s
In particular, the work of Von Meerwall et al.63 presents parametrization, where the experimental value of Tg was not
diffusion coefficients for traces of alkanes in a monodisperse taken into account, and/or variations in the cooling ramp were
polyethylene matrix of 33 kDa at 453 K, obtained via spin− employed to calculate Tg.
echo NMR. We can then compare our data for our three All simulations successfully achieved the self-diffuse regime
largest amide and acid solutes (M/M0 = 3.3, 3.9, and 4.4), for all considered molecules and temperatures, within the level
which have equivalent carbon counts of 22, 26, and 30, of uncertainty addressed in this study. A noteworthy point of
respectively, with the C28H58 and C24H50 experimental emphasis lies in the advantages of utilizing the gx/6t vs t
diffusions. These findings are presented in Table 1. representation in a log−log format for accurately calculating
M https://doi.org/10.1021/acs.jpcb.3c04904
J. Phys. Chem. B XXXX, XXX, XXX−XXX
The Journal of Physical Chemistry B pubs.acs.org/JPCB Article

self-diffusion coefficients. This method provides a lucid visual value between 0.9 to 1.1, which contrasts with the standard
indication of the degree to which the free diffusive regime has MARTINI value of 4 obtained from CG water simulations. We
been reached. hypothesize that the coarse-graining nature of the force field
No significant differences were observed in the recovered might not affect long-term dynamics, such as the self-diffusion
values of D between acids and amides of the same carbon coefficient of large molecules. This result holds promising
length (at a given temperature). Our findings align with implications for the accurate prediction of self-diffusion
experimental data comparing the diffusion of alcohols to coefficients using these force fields. Nevertheless, further
alkanes within a PE matrix. This similarity can be attributed to analysis is necessary to generalize this result and also to
the low interaction between the polar heads and the matrix, or,
determine the scaling factor in the lower temperature range
in simpler terms, to the inherently nonpolar nature of PE.
and for the other, smaller additives, conditions where coarse-
The dependence of D on the mass of the additive was
accurately modeled by the power-law relationship, D ∝ (M/ graining time acceleration should have a greater effect.
M0)−α, with equivalent α values for both acids and amides at a
given temperature, within the range of error considered. Our
simulations revealed a temperature-dependent relationship for
■ ASSOCIATED CONTENT
* Supporting Information
sı
α. Specifically, α ≅ 1 was observed for temperatures outside
The Supporting Information is available free of charge at
the metastability zone of the matrix, indicative of a diffusion
https://pubs.acs.org/doi/10.1021/acs.jpcb.3c04904.
behavior compatible with the Rouse mechanism. Conversely, α
values slightly greater than 1 were obtained for temperatures S.I: A detailed description of all the simulated additives
within the glass transition region (metastable zone of the PE as well as the intramolecular and intermolecular
matrix), coinciding with the loss of mobility in the matrix. This parameter values in order to perform the simulations,
result might be explained by a combination of diffusion also references to the MARTINI web page for further
mechanisms (dynamic heterogeneities), or a change in the details. S.II: Analysis of time convergence for the radius
friction coefficient of the monomer, which exerts more of gyration (Rg) and % of Free Volume at different
influence at these lower temperatures. temperatures for the PE matrix. S.III: Results for self-
Regarding the self-diffusion constant dependence on diffusion coefficients of additives computed on the
temperature, we observed an Arrhenius behavior at elevated matrix at the same temperatures but with different Rg.
temperatures (T > 450 K), transitioning into a convex S.IV: Results for the Arrhenius fits for all the additives
Arrhenius trend as temperature decreases. This trend was studied in this work. S.V: Results for Mode Coupling
consistent for both the self-diffusion of additives in the free- Theory fits for all the additives studied in this work.
diffusing regime and the segmental relaxation of the monomers S.VI: MSD plot for PE matrix at 450 K (PDF)
of the PE matrix (σ2PE/6t). The activation enthalpy derived
from the high-temperature regime was found to be
independent of solute mass, and closely aligned with the
value obtained for the segmental relaxation of the matrix.
■ AUTHOR INFORMATION
Corresponding Author
Although the latter was slightly higher, this difference falls
within the bounds of our calculated uncertainties. The Matias H. Factorovich − Departamento de Química
equivalence in activation enthalpy for the solutes and the Inorgánica, Analítica y Química Física/INQUIMAE,
similarity to the segmental relaxation value in the Arrhenius Facultad de Ciencias y Naturales, Universidad de Buenos
regime suggests that the diffusion of the additives is Aires, Ciudad Universitaria, Buenos Aires C1428EHA,
constrained by the mobility and free volume of the polymer Argentina; orcid.org/0000-0001-5611-8751;
matrix, in accordance with a key premise of the free volume Email: [email protected]
theory. Authors
The whole temperature range of the diffusion data (additives
María del Mar Cammarata − Departamento de Química
and matrix) was accurately fitted by the power law model
Inorgánica, Analítica y Química Física/INQUIMAE,
based on MCT, despite being rigorously developed for pure
Facultad de Ciencias y Naturales, Universidad de Buenos
substance. From these fits we recovered the parameters γ and
Aires, Ciudad Universitaria, Buenos Aires C1428EHA,
Tc, this last one, resulted to be higher than the calculated Tg for
Argentina
the PE matrix, as expected by the theory. The values of Tc were
Mario D. Contin − Departamento de Ciencias Química,
equal for the additives with M/M0 > 1 and also equal to Tc for
Catedra de Química Analítica. Facultad de Farmacia y
the segmental relaxation of the matrix. This led to the
Bioquímica, Universidad de Buenos Aires, Buenos Aires
interpretation that the smallest solute is capable of moving C1113AAD, Argentina
through the network even when it has entered the regime of R. Martín Negri − Departamento de Química Inorgánica,
glass transition, while the larger ones are highly influenced by Analítica y Química Física/INQUIMAE, Facultad de
the motion of the matrix. Both parameters coincide with Ciencias y Naturales, Universidad de Buenos Aires, Ciudad
results from coarse-grained models but not so much with the Universitaria, Buenos Aires C1428EHA, Argentina;
all-atom dynamics, highlighting the difference in relaxation orcid.org/0000-0003-1427-7927
times that may arise, at these near Tg temperatures, between
these kinds of simulations. Complete contact information is available at:
Finally, by employing experimental data, we estimate the https://pubs.acs.org/10.1021/acs.jpcb.3c04904
time scaling factor (β = Dexp/DCG) between coarse-grained
(CG) and all-atom systems for the three largest additives and Notes
the PE matrix at temperatures above 450 K. This results in a The authors declare no competing financial interest.
N https://doi.org/10.1021/acs.jpcb.3c04904
J. Phys. Chem. B XXXX, XXX, XXX−XXX
The Journal of Physical Chemistry B pubs.acs.org/JPCB Article

■ ACKNOWLEDGMENTS
We thank Prof. Rocio Semino for valuable discussions. We are
(18) Fang, X.; Vitrac, O. Predicting Diffusion Coefficients of
Chemicals in and through Packaging Materials. Crit. Rev. Food Sci.
Nutr. 2017, 57 (2), 275−312.
also grateful to the reviewers for worthy comments. We (19) Hayashi, H.; Sakai, H.; Matsuzawa, S. Diffusion of Methyl
acknowledge the Center of High-Performance Computing Esters of Higher Fatty Acid in Polypropylene. J. Appl. Polym. Sci.
CECAR and LOGAN from the University of Buenos Aires for 1994, 51 (13), 2165−2173.
the allocation of computing time and technical support. (20) Reynier, A.; Dole, P.; Feigenbaum, A. Additive Diffusion
MdMC acknowledges CONICET-AMPACET for a cofi- Coefficients in Polyolefins. II. Effect of Swelling and Temperature on
nanced doctoral fellowship. the D = f(M) Correlation. J. Appl. Polym. Sci. 2001, 82 (10), 2434−

■ REFERENCES
(1) Patel, P.; Savargaonkar, N. A Review of Additives for Plastics:
2443.
(21) Vitrac, O.; Lézervant, J.; Feigenbaum, A. Decision Trees as
Applied to the Robust Estimation of Diffusion Coefficients in
Slips and Antiblocks. Plast. Eng. 2007, 63 (1), 48−51. Polyolefins. J. Appl. Polym. Sci. 2006, 101 (4), 2167−2186.
(2) Huang, Y.; Xiong, Y.; Liu, C.; Li, L.; Xu, D.; Lin, Y.-H.; Nan, C.- (22) Martínez-López, B.; Gontard, N.; Peyron, S. Worst Case
W. Single-Crystalline 2D Erucamide with Low Friction and Enhanced Prediction of Additives Migration from Polystyrene for Food Safety
Thermal Conductivity. Colloids Surf. Physicochem. Eng. Asp. 2018, 540, Purposes: A Model Update. Food Addit. Contam. Part A 2018, 35 (3),
29−35. 563−576.
(3) Gubała, D.; Fox, L. J.; Harniman, R.; Hussain, H.; Robles, E.; (23) Pinte, J.; Joly, C.; Dole, P.; Feigenbaum, A. Diffusion of
Chen, M.; Briscoe, W. H. Heads or Tails: Nanostructure and Homologous Model Migrants in Rubbery Polystyrene: Molar Mass
Molecular Orientations in Organised Erucamide Surface Layers. J. Dependence and Activation Energy of Diffusion. Food Addit. Contam.
Colloid Interface Sci. 2021, 590, 506−517. Part A 2010, 27 (4), 557−566.
(4) Allan, A. J. G. Surface Properties of Polyethylene: Effect of an (24) Zhu, Y.; Welle, F.; Vitrac, O. A Blob Model to Parameterize
Amphipathic Additive. J. Colloid Sci. 1959, 14 (2), 206−221. Polymer Hole Free Volumes and Solute Diffusion. Soft Matter 2019,
(5) Gubała, D.; Taylor, N.; Harniman, R.; Rawle, J.; Hussain, H.; 15 (43), 8912−8932.
Robles, E.; Chen, M.; Briscoe, W. H. Structure, Nanomechanical (25) Martinez-Lopez, B.; Huguet, P.; Gontard, N.; Peyron, S.
Properties, and Wettability of Organized Erucamide Layers on a Developing a Macroscopic Mechanistic Model for Low Molecular
Polypropylene Surface. Langmuir 2021, 37 (21), 6521−6532. Weight Diffusion through Polymers in the Rubbery State. Ind. Eng.
(6) Schlotter, N. E.; Furlan, P. Y. A Review of Small Molecule Chem. Res. 2016, 55 (17), 5078−5089.
Diffusion in Polyolefins. Polymer 1992, 33 (16), 3323−3342. (26) Dole, P.; Voulzatis, Y.; Vitrac, O.; Reynier, A.; Hankemeier, T.;
(7) Dulal, N.; Shanks, R.; Chalmers, D.; Adhikari, B.; Gill, H. Aucejo, S.; Feigenbaum, A. Modelling of Migration from Multi-Layers
Migration and Performance of Erucamide Slip Additive in High- and Functional Barriers: Estimation of Parameters. Food Addit.
Density Polyethylene Bottle Caps. J. Appl. Polym. Sci. 2018, 135 (43), Contam. 2006, 23 (10), 1038−1052.
46822. (27) Vitrac, O.; Mougharbel, A.; Feigenbaum, A. Interfacial Mass
(8) Nouman, M.; Saunier, J.; Jubeli, E.; Yagoubi, N. Additive Transport Properties Which Control the Migration of Packaging
Blooming in Polymer Materials: Consequences in the Pharmaceutical Constituents into Foodstuffs. J. Food Eng. 2007, 79 (3), 1048−1064.
and Medical Field. Polym. Degrad. Stab. 2017, 143, 239−252. (28) Gavriil, G.; Kanavouras, A.; Coutelieris, F. A. Food-Packaging
(9) Dulal, N.; Shanks, R.; Gengenbach, T.; Gill, H.; Chalmers, D.; Migration Models: A Critical Discussion. Crit. Rev. Food Sci. Nutr.
Adhikari, B.; Pardo Martinez, I. Slip-Additive Migration, Surface 2018, 58 (13), 2262−2272.
Morphology, and Performance on Injection Moulded High-Density (29) Schlick, T.; Portillo-Ledesma, S. Biomolecular Modeling
Polyethylene Closures. J. Colloid Interface Sci. 2017, 505, 537−545. Thrives in the Age of Technology. Nat. Comput. Sci. 2021, 1 (5),
(10) Maier, C.; Calafut, T. Additives. In Polypropylene; Maier, C., 321−331.
Calafut, T., Eds.; William Andrew Publishing: Norwich, NY, 1998; pp (30) Alessandri, R.; Grünewald, F.; Marrink, S. J. The Martini Model
27−47. DOI: 10.1016/B978-188420758-7.50008-4. in Materials Science. Adv. Mater. 2021, 33 (24), 2008635.
(11) Murphy, J. Modifying Processing Characteristics: Lubricants, (31) Singh, N.; Li, W. Recent Advances in Coarse-Grained Models
Mould Release Agents, Anti-Slip and Anti-Blocking. In Additives for
for Biomolecules and Their Applications. Int. J. Mol. Sci. 2019, 20
Plastics Handbook, 2nd ed.; Murphy, J., Ed.; Elsevier Science:
(15), 3774.
Amsterdam, 2001; pp 205−218. DOI: 10.1016/B978-185617370-
(32) Schmid, F. Understanding and Modeling Polymers: The
4/50019-1.
Challenge of Multiple Scales. ACS Polym. Au 2023, 3 (1), 28−58.
(12) Coelho, F.; Vieira, L. F.; Benavides, R.; da Silva Paula, M. M.;
(33) Rubinstein, M.; Colby, R. H. Polymer Physics; OUP Oxford,
Bernardin, A. M.; Magnago, R. F.; da Silva, L. Synthesis and
Evaluation of Amides as Slip Additives in Polypropylene. Int. Polym. 2003; Chapter 8, pp 309−351.
Process. 2015, 30 (5), 574−584. (34) Rubinstein, M.; Colby, R. H. Polymer Physics; OUP Oxford,
(13) Adriana; Jalal, R.; Yuniati. Antistatic Effect of Glycerol 2003; Chapter 9, pp 361−364 and 387−389.
Monostearate on Volume Resistivity and Mechanical Properties of (35) Green, P. F.; Kramer, E. J. Matrix Effects on the Diffusion of
Nanocomposite Polystyrene-Nanocrystal Cellulose. AIP Conf. Proc. Long Polymer Chains. Macromolecules 1986, 19 (4), 1108−1114.
2018, 030047. (36) Durand, M.; Meyer, H.; Benzerara, O.; Baschnagel, J.; Vitrac,
(14) Sakhalkar, S. S.; Walters, K. B.; Hirt, D. E.; Miranda, N. R.; O. Molecular Dynamics Simulations of the Chain Dynamics in
Roberts, W. P. Surface Characterization of LLDPE Film Containing Monodisperse Oligomer Melts and of the Oligomer Tracer Diffusion
Glycerol Monostearate. J. Plast. Film Sheeting 2002, 18 (1), 33−43. in an Entangled Polymer Matrix. J. Chem. Phys. 2010, 132 (19),
(15) Quijada-Garrido, I.; Barrales-Rienda, J. M.; Frutos, G. Diffusion 194902.
of Erucamide (13-Cis-Docosenamide) in Isotactic Polypropylene. (37) Shanbhag, S.; Wang, Z. Molecular Simulation of Tracer
Macromolecules 1996, 29 (22), 7164−7176. Diffusion and Self-Diffusion in Entangled Polymers. Macromolecules
(16) Reynier, A.; Dole, P.; Humbel, S.; Feigenbaum, A. Diffusion 2020, 53 (12), 4649−4658.
Coefficients of Additives in Polymers. I. Correlation with Geometric (38) Li, W.; Jana, P. K.; Behbahani, A. F.; Kritikos, G.; Schneider, L.;
Parameters. J. Appl. Polym. Sci. 2001, 82 (10), 2422−2433. Polińska, P.; Burkhart, C.; Harmandaris, V. A.; Müller, M.; Doxastakis,
(17) Cecon, V. S.; Da Silva, P. F.; Curtzwiler, G. W.; Vorst, K. L. M. Dynamics of Long Entangled Polyisoprene Melts via Multiscale
The Challenges in Recycling Post-Consumer Polyolefins for Food Modeling. Macromolecules 2021, 54 (18), 8693−8713.
Contact Applications: A Review. Resour. Conserv. Recycle. 2021, 167, (39) Lin, E.; You, X.; Kriegel, R. M.; Moffitt, R. D.; Batra, R. C.
No. 105422. Interdiffusion of Small Molecules into a Glassy Polymer Film via

O https://doi.org/10.1021/acs.jpcb.3c04904
J. Phys. Chem. B XXXX, XXX, XXX−XXX
The Journal of Physical Chemistry B pubs.acs.org/JPCB Article

Coarse-Grained Molecular Dynamics Simulations. Polymer 2017, 115, (60) Xia, W.; Song, J.; Jeong, C.; Hsu, D. D.; Phelan, F. R., Jr.;
273−284. Douglas, J. F.; Keten, S. Energy-Renormalization for Achieving
(40) Panizon, E.; Bochicchio, D.; Monticelli, L.; Rossi, G. MARTINI Temperature Transferable Coarse-Graining of Polymer Dynamics.
Coarse-Grained Models of Polyethylene and Polypropylene. J. Phys. Macromolecules 2017, 50 (21), 8787−8796.
Chem. B 2015, 119 (25), 8209−8216. (61) Diffusion of Aromatic Solutes in Aliphatic Polymers above
(41) Polyethylene Force-Field. http://cgmartini.nl/index.php/force- Glass Transition Temperature | Macromolecules. https://pubs.acs.
field-parameters/polymers2/442-polymers.html?dir=Linear&lipid=PE org/doi/full/10.1021/ma3022103 (accessed 2022−08−30).
(accessed 2023−03−31). (62) Möller, K.; Gevert, T. An FTIR Solid-State Analysis of the
(42) Pearson, D. S.; Ver Strate, G.; Von Meerwall, E.; Schilling, F. C. Diffusion of Hindered Phenols in Low-Density Polyethylene (LDPE):
Viscosity and Self-Diffusion Coefficient of Linear Polyethylene. The Effect of Molecular Size on the Diffusion Coefficient. J. Appl.
Macromolecules 1987, 20 (5), 1133−1141. Polym. Sci. 1994, 51 (5), 895−903.
(43) Van Der Spoel, D.; Lindahl, E.; Hess, B.; Groenhof, G.; Mark, (63) von Meerwall, E. D.; Lin, H.; Mattice, W. L. Trace Diffusion of
A. E.; Berendsen, H. J. C. GROMACS: Fast, Flexible, and Free. J. Alkanes in Polyethylene: Spin-Echo Experiment and Monte Carlo
Comput. Chem. 2005, 26 (16), 1701−1718. Simulation. Macromolecules 2007, 40 (6), 2002−2007.
(44) Basconi, J. E.; Shirts, M. R. Effects of Temperature Control (64) von Meerwall, E.; Beckman, S.; Jang, J.; Mattice, W. L.
Algorithms on Transport Properties and Kinetics in Molecular Diffusion of Liquid N-Alkanes: Free-Volume and Density Effects. J.
Dynamics Simulations. J. Chem. Theory Comput. 2013, 9 (7), Chem. Phys. 1998, 108 (10), 4299−4304.
2887−2899. (65) Truhlar, D. G.; Kohen, A. Convex Arrhenius Plots and Their
(45) Frenkel, D.; Smit, B. Understanding Molecular Simulation: From Interpretation. Proc. Natl. Acad. Sci. U. S. A. 2001, 98 (3), 848−851.
Algorithms to Applications; Elsevier, 2001; Chapter 4, pp 87−97. (66) Angell, C. A. Formation of Glasses from Liquids and
(46) Pestryaev, E. M. Comparison of Various Correlation Times in Biopolymers. Science 1995, 267 (5206), 1924−1935.
Polymer Melts by Molecular Dynamics Simulation. J. Phys. Conf. Ser. (67) Debenedetti, P. G.; Stillinger, F. H. Supercooled Liquids and
2011, 324 (1), No. 012031. the Glass Transition. Nature 2001, 410 (6825), 259−267.
(47) Yang, Q.; Chen, X.; He, Z.; Lan, F.; Liu, H. The Glass (68) Saylor, D. M.; Jawahery, S.; Silverstein, J. S.; Forrey, C.
Transition Temperature Measurements of Polyethylene: Determined Communication: Relationship between Solute Localization and
by Using Molecular Dynamic Method. RSC Adv. 2016, 6 (15), Diffusion in a Dynamically Constrained Polymer System. J. Chem.
12053−12060. Phys. 2016, 145 (3), No. 031106.
(48) Afzal, M. A. F.; Browning, A. R.; Goldberg, A.; Halls, M. D.; (69) Elder, R. M.; Saylor, D. M. Predicting Solute Diffusivity in
Gavartin, J. L.; Morisato, T.; Hughes, T. F.; Giesen, D. J.; Goose, J. E. Polymers Using Time−Temperature Superposition. J. Phys. Chem. B
High-Throughput Molecular Dynamics Simulations and Validation of 2022, 126 (20), 3768−3777.
(70) Elder, R. M.; Saylor, D. M. Relations Between Dynamic
Thermophysical Properties of Polymers for Various Applications. ACS
Localization and Solute Diffusion in Polymers. J. Phys. Chem. B 2021,
Appl. Polym. Mater. 2021, 3 (2), 620−630.
125 (32), 9372−9383.
(49) Patrone, P. N.; Dienstfrey, A.; Browning, A. R.; Tucker, S.;
(71) Jeong, C.; Douglas, J. F. Mass Dependence of the Activation
Christensen, S. Uncertainty Quantification in Molecular Dynamics
Enthalpy and Entropy of Unentangled Linear Alkane Chains. J. Chem.
Studies of the Glass Transition Temperature. Polymer 2016, 87, 246−
Phys. 2015, 143 (14), 144905.
259.
(72) Cohen, M. H.; Turnbull, D. Molecular Transport in Liquids
(50) Rossi, G.; Giannakopoulos, I.; Monticelli, L.; Rostedt, N. K. J.;
and Glasses. J. Chem. Phys. 1959, 31 (5), 1164−1169.
Puisto, S. R.; Lowe, C.; Taylor, A. C.; Vattulainen, I.; Ala-Nissila, T. A (73) de Souza, V. K.; Wales, D. J. Correlation Effects and Super-
MARTINI Coarse-Grained Model of a Thermoset Polyester Coating. Arrhenius Diffusion in Binary Lennard-Jones Mixtures. Phys. Rev. B
Macromolecules 2011, 44 (15), 6198−6208. 2006, 74 (13), No. 134202.
(51) Xu, W.-S.; Douglas, J. F.; Sun, Z.-Y. Polymer Glass Formation: (74) Reichman, D. R.; Charbonneau, P. Mode-Coupling Theory. J.
Role of Activation Free Energy, Configurational Entropy, and Stat. Mech. Theory Exp. 2005, 2005 (05), No. P05013.
Collective Motion. Macromolecules 2021, 54 (7), 3001−3033. (75) Berthier, L.; Biroli, G. Theoretical Perspective on the Glass
(52) Agapov, A. L.; Wang, Y.; Kunal, K.; Robertson, C. G.; Sokolov, Transition and Amorphous Materials. Rev. Mod. Phys. 2011, 83 (2),
A. P. Effect of Polar Interactions on Polymer Dynamics. Macro- 587−645.
molecules 2012, 45 (20), 8430−8437. (76) Binder, K.; Baschnagel, J.; Paul, W. Glass Transition of Polymer
(53) Dudowicz, J.; Freed, K. F.; Douglas, J. F. Generalized Entropy Melts: Test of Theoretical Concepts by Computer Simulation. Prog.
Theory of Polymer Glass Formation. Advances in Chemical Physics; Polym. Sci. 2003, 28 (1), 115−172.
John Wiley & Sons, Ltd, 2007; pp 125−222. DOI: 10.1002/ (77) Kob, W. Course 5: Supercooled Liquids, the Glass Transition,
9780470238080.ch3. and Computer Simulations. In Slow Relaxations and nonequilibrium
(54) Moore, E. B.; Molinero, V. Structural Transformation in dynamics in condensed matter; Barrat, J.-L., Feigelman, M., Kurchan, J.,
Supercooled Water Controls the Crystallization Rate of Ice. Nature Dalibard, J., Eds.; Les Houches-É cole d’É té de Physique Theorique;
2011, 479 (7374), 506−508. Springer: Berlin, Heidelberg, 2003; pp 199−269. DOI: 10.1007/978-
(55) Shalaby, S. W. Thermoplastic Polymers. In Thermal Character- 3-540-44835-8_5.
ization of Polymeric Materials; Turi, E. A., Ed.; Academic Press, 1981; (78) Vallée, R. A. L.; Paul, W.; Binder, K. Probe Molecules in
Chapter 3, pp 235−364. DOI: 10.1016/B978-0-12-703780-6.50008- Polymer Melts near the Glass Transition: A Molecular Dynamics
0. Study of Chain Length Effects. J. Chem. Phys. 2010, 132 (3),
(56) Miller, A. A. Kinetic Interpretation of the Glass Transition: No. 034901.
Glass Temperatures of n-Alkane Liquids and Polyethylene. J. Polym. (79) Egorov, S. A. Anomalous Nanoparticle Diffusion in Polymer
Sci. A-2 Polym. Phys. 1968, 6 (1), 249−257. Solutions and Melts: A Mode-Coupling Theory Study. J. Chem. Phys.
(57) Zerze, H. Nucleation and Growth of Crystals inside 2011, 134 (8), No. 084903.
Polyethylene Nano-Droplets. J. Chem. Phys. 2022, 157 (15), 154901. (80) Das, S. P. Mode-Coupling Theory and the Glass Transition in
(58) Klippenstein, V.; Tripathy, M.; Jung, G.; Schmid, F.; van der Supercooled Liquids. Rev. Mod. Phys. 2004, 76 (3), 785−851.
Vegt, N. F. A. Introducing Memory in Coarse-Grained Molecular (81) Ediger, M. D. Spatially Heterogeneous Dynamics in Super-
Simulations. J. Phys. Chem. B 2021, 125 (19), 4931−4954. cooled Liquids. Annu. Rev. Phys. Chem. 2000, 51 (1), 99−128.
(59) Marrink, S. J.; de Vries, A. H.; Mark, A. E. Coarse Grained (82) Napolitano, S.; Glynos, E.; Tito, N. B. Glass Transition of
Model for Semiquantitative Lipid Simulations. J. Phys. Chem. B 2004, Polymers in Bulk, Confined Geometries, and near Interfaces. Rep.
108 (2), 750−760. Prog. Phys. 2017, 80 (3), No. 036602.

P https://doi.org/10.1021/acs.jpcb.3c04904
J. Phys. Chem. B XXXX, XXX, XXX−XXX
The Journal of Physical Chemistry B pubs.acs.org/JPCB Article

(83) Utracki, L. A.; Sammut, P. Molten Polystyrene Structures above

the Glass Transition, T > Tg. J. Polym. Sci., Part B: Polym. Phys. 2011,
49 (19), 1369−1380.
(84) Novikov, V. N.; Sokolov, A. P. Universality of the Dynamic
̀
Crossover in Glass-Forming Liquids: A ̀ magic” Relaxation Time. Phys.
Rev. E 2003, 67 (3), No. 031507.
(85) Bergman, R.; Börjesson, L.; Torell, L. M.; Fontana, A.
Dynamics around the Liquid-Glass Transition in Poly(Propylene-
Glycol) Investigated by Wide-Frequency-Range Light-Scattering
Techniques. Phys. Rev. B 1997, 56 (18), 11619−11628.
(86) Frick, B.; Farago, B.; Richter, D. Temperature Dependence of
the Nonergodicity Parameter in Polybutadiene in the Neighborhood
of the Glass Transition. Phys. Rev. Lett. 1990, 64 (24), 2921−2924.
(87) Kisliuk, A.; Mathers, R. T.; Sokolov, A. P. Crossover in
Dynamics of Polymeric Liquids: Back to Tll? J. Polym. Sci., Part B:
Polym. Phys. 2000, 38 (21), 2785−2790.
(88) Horbach, J.; Kob, W. Relaxation Dynamics of a Viscous Silica
Melt: The Intermediate Scattering Functions. Phys. Rev. E 2001, 64
(4), No. 041503.
(89) Khairy, Y.; Alvarez, F.; Arbe, A.; Colmenero, J. Applicability of
Mode-Coupling Theory to Polyisobutylene: A Molecular Dynamics
Simulation Study. Phys. Rev. E 2013, 88 (4), No. 042302.
(90) Peters, B. L.; Salerno, K. M.; Agrawal, A.; Perahia, D.; Grest, G.
S. Coarse-Grained Modeling of Polyethylene Melts: Effect on
Dynamics. J. Chem. Theory Comput. 2017, 13 (6), 2890−2896.
(91) Monticelli, L.; Kandasamy, S. K.; Periole, X.; Larson, R. G.;
Tieleman, D. P.; Marrink, S.-J. The MARTINI Coarse-Grained Force
Field: Extension to Proteins. J. Chem. Theory Comput. 2008, 4 (5),
819−834.

Q https://doi.org/10.1021/acs.jpcb.3c04904
J. Phys. Chem. B XXXX, XXX, XXX−XXX