You are on page 1of 11

Backbone oriented anisotropic coarse

grains for efficient simulations of polymers


Cite as: J. Chem. Phys. 153, 214901 (2020); https://doi.org/10.1063/5.0019945
Submitted: 26 June 2020 . Accepted: 11 November 2020 . Published Online: 02 December 2020

Florent Goujon, Nicolas Martzel, Alain Dequidt, Benoit Latour, Sébastien Garruchet, Julien Devémy,
Ronald Blaak, Étienne Munch, and Patrice Malfreyt

ARTICLES YOU MAY BE INTERESTED IN

Coarse graining molecular dynamics with graph neural networks


The Journal of Chemical Physics 153, 194101 (2020); https://doi.org/10.1063/5.0026133

Reflections on electron transfer theory


The Journal of Chemical Physics 153, 210401 (2020); https://doi.org/10.1063/5.0035434

Generalized mode-coupling theory of the glass transition. II. Analytical scaling laws
The Journal of Chemical Physics 153, 214506 (2020); https://doi.org/10.1063/5.0026979

J. Chem. Phys. 153, 214901 (2020); https://doi.org/10.1063/5.0019945 153, 214901

© 2020 Author(s).
The Journal
ARTICLE scitation.org/journal/jcp
of Chemical Physics

Backbone oriented anisotropic coarse grains


for efficient simulations of polymers
Cite as: J. Chem. Phys. 153, 214901 (2020); doi: 10.1063/5.0019945
Submitted: 26 June 2020 • Accepted: 11 November 2020 •
Published Online: 2 December 2020

Florent Goujon,1,a) Nicolas Martzel,2 Alain Dequidt,1 Benoit Latour,2 Sébastien Garruchet,2
1 1
Julien Devémy, Ronald Blaak, Étienne Munch, and Patrice Malfreyt1
2

AFFILIATIONS
1
Université Clermont Auvergne, CNRS, SIGMA Clermont, Institut de Chimie de Clermont-Ferrand, F-63000 Clermont-Ferrand,
France
2
Manufacture Française des Pneumatiques Michelin, Site de Ladoux, 23 Place des Carmes Déchaux, France Cedex 9,
63040 Clermont-Ferrand, France

a)
Author to whom correspondence should be addressed: florent.goujon@uca.fr

ABSTRACT
Despite the fact that anisotropic particles have been introduced to describe molecular interactions for decades, they have been poorly used for
polymers because of their computing time overhead and the absence of a relevant proof of their impact in this field. We first report a method
using anisotropic beads for polymers, which solves the computing time issue by considering that beads keep their principal orientation
alongside the mean local backbone vector of the polymer chain, avoiding the computation of torques during the dynamics. Applying this
method to a polymer bulk, we study the effect of anisotropic interactions vs isotropic ones for various properties such as density, pressure,
topology of the chain network, local structure, and orientational order. We show that for different classes of potentials traditionally used in
molecular simulations, those backbone oriented anisotropic beads can solve numerous issues usually encountered with isotropic interactions.
We conclude that the use of backbone oriented anisotropic beads is a promising approach for the development of realistic coarse-grained
potentials for polymers.
Published under license by AIP Publishing. https://doi.org/10.1063/5.0019945., s

I. INTRODUCTION Top-down approaches derive parameters from macroscopic proper-


ties (compressibility, diffusion, surface tension, density, and radius
The field of computer simulations in polymer science has been of gyration).10–12 Bottom-up approaches use the configurations at
at the origin of many developments of coarse-grained (CG) mod- the atomistic level obtained by using molecular dynamics or Monte
els and corresponding mesoscopic simulation methods.1,2 Indeed, Carlo simulation methods to develop interaction forces and param-
polymer chains are natural candidates for coarse-graining due to eters for the realistic CG models.13–18 A widely used bottom-up tech-
their slow relaxation, the self-similarity of chains that are formed nique is the Iterative Boltzmann Inversion (IBI) method,19 which is
by covalent bonding of a large number of identical monomers, and based on the structural characterization and consists in determin-
the universality of some properties at the large scale through scaling ing an effective pair potential for an ensemble of identical particles
laws. A number of generic CG models3–9 with low resolutions have from the corresponding radial distribution function (RDF) via the
been applied in the past to reproduce various properties of polymer potential of mean force approach.20 An alternative to the IBI method
melts, but the quantitative prediction of properties as a function of is the idea of Force Matching (FM),21–26 which is based on the
polymer chemistry remains a challenging task. analysis of the instantaneous forces exerted on the beads. Recently,
The extension by including more molecular details leads to Dequidt et al. developed a new bottom-up method, Statistical Tra-
the so-called realistic CG models and can be achieved through jectory Matching (STM),27–31 which is based on a Bayesian approach
both the top-down and bottom-up parameterization schemes. to obtain the parameters of CG models.

J. Chem. Phys. 153, 214901 (2020); doi: 10.1063/5.0019945 153, 214901-1


Published under license by AIP Publishing
The Journal
ARTICLE scitation.org/journal/jcp
of Chemical Physics

Nevertheless, coarse-graining suffers from some major draw- that are often induced by soft mesoscopic potentials. We analyze
backs when applied to polymer simulation because of the implicit the local positional structure of the chains by separating the radial
assumption of spherical grains. Spherical grains leave low energy distribution function and compute it in two domains defined with
saddle points along the polymer chain and hence result in unreal- respect to the bead directors. Finally, the local orientational order
istic chain crossing topology violations.32 Different strategies have is investigated by calculating the nematic order parameter at dif-
been introduced to solve this issue, such as segmental repulsion,33,34 ferent distances. We conclude with discussing the efficiency of our
slip links,35 or use of rouse modes.36 All these methodologies add method and its ability to reduce the known weaknesses of current
new degrees of freedom to those of the grains. Another prob- coarse-graining methods.
lem of coarse-grained polymer simulations is the anomalous com-
pressibility prediction.37 Anisotropic interactions, or equivalently
ellipsoidal grains stretched along the backbone of the chain, could
solve both issues, by closing the saddle points, and allowing a II. METHODOLOGY
better packing, without the need for extra degrees of freedom. A. Anisotropic potential
Ellipsoidal representation of grains is also supported by results
showing that they are close to the real physical shape of coarse In this paper, we consider mesoscopic beads bonded to each
grains.38,39 other to form linear polymer chains. The anisotropic shape of the
Since the introduction of the Berne potential,40 a lot of effort beads is chosen to be an ellipsoid of revolution defined by the parallel
has been invested in the representation of the potential of non- and perpendicular components of its gyration tensor. The paral-
spherical atomic groups41–45 leading to generic methods to compute lel component σ ∥ associated with the main orientation is along the
them, either from abstract geometric representations46,47 or from director of the bead. The perpendicular component σ – is used to
multipole developments.48–50 However, the physical implications represent the squared radius of gyration along the two other princi-
of the introduction of anisotropic interactions in the mesoscopic pal axes. Both values σ ∥ and σ – are fixed and could, for instance,
behavior of coarse-grained systems have not been studied in the lit- stem from the average values of these quantities obtained for the
erature, even though accuracy improvements are expected.51,52 One corresponding segment length in an atomistic simulation. For sim-
of the underlying reasons is that the implementation of anisotropic plicity, we will consider here only one type of bead for all polymer
interactions is associated with a potentially considerable increase in chains. A representation of the shape of one bead and its director
the computation time. Additionally, a relevant proof of interest of used in this work is shown in Fig. 1(a). By using this description
anisotropic interactions for the simulation of polymers remains to for the chains, we can deduce the directors of each bead from the
be found as the addition of spherical beads is often sufficient to avoid local orientation of the polymer chain. Given a bead i located at
the complications introduced by anisotropic particles. Similar issues position pi , its director u is determined by what we define as the
are encountered for the simulation of liquid crystals, and several backbone mean orientation, i.e., it is the normalized vector taken
solutions involving spherical or anisotropic particles have been stud- between the preceding (i − 1) and the next bead (i + 1) in the
ied.53–55 As a consequence, the first step in the design of interesting chain,
anisotropic CG potentials is to be able to correct the weaknesses of
pi+1 − pi−1
spherical potential while keeping the central processing unit (CPU) u= . (1)
time as low as possible. ∣pi+1 − pi−1 ∣
In this work, we report a method using anisotropic beads for
polymers by introducing the following approximation: the orienta- For the terminal beads, we simply use the outer bond as the orien-
tion of the polymer beads is kept along the mean backbone orien- tation of the director. An example of the resulting representation is
tation of the chain. This results in fast calculations as the torques shown in Fig. 1(b) for a 30-bead long polymer chain. Figure 1(c)
do not need to be calculated. This is the first step in the design shows a typical simulation box (detailed in Sec. II B), with bead
of realistic CG potentials that include anisotropy. In this work, directors emphasized by displaying unwrapped chains.
we aim at testing the influence of the bead anisotropy on vari- Using this approximation to calculate the bead orientations, we
ous properties of interest in order to check if the classical issues only need to consider the forces acting on the center-of-mass of the
of spherical potentials can be improved by a single bead shape beads for the dynamics without the calculation of torques because
parameter. the new directors are recalculated every step after the bead posi-
Section II describes the model developed for this work to deal tions have been updated. The orientation of the beads along the
with interacting ellipsoids applied to the case of polymer chains average local direction of the chain is a strong hypothesis, which is
with only a slight increase in computing time. By using only one supposed to be true only for large levels of coarse-graining or for spe-
additional parameter (the anisotropy factor), we are able to sim- cific monomers. The shape and orientation of cis-PB at λ = 5 have
ulate prolate-shaped particles with orientations that are automati- been shown to match this hypothesis.38 The additional calculation
cally deduced from the positions of the chain beads. We apply this caused by the anisotropic interactions has to be straightforward and
methodology in dissipative particle dynamics (DPD) simulations of fast enough that it does not significantly affect the global computa-
a bulk polymer melt at constant volume or pressure. Section III tion time. All that remains to be done is to calculate the force acting
presents the results obtained by our simulations using two com- between two ellipsoidal shaped beads.
monly used interaction potentials in coarse-grained simulations. We introduce a modified distance R = Mr between two ellip-
We study the effect that the grain anisotropy has on thermody- soidal beads by means of the dimensionless matrix M. Considering
namic properties and the ability to reduce the topology violations two beads with directors u and v, respectively, we define the radius

J. Chem. Phys. 153, 214901 (2020); doi: 10.1063/5.0019945 153, 214901-2


Published under license by AIP Publishing
The Journal
ARTICLE scitation.org/journal/jcp
of Chemical Physics

FIG. 1. (a) Schematic representation of


the gyration tensor and the director used
in this work to model the anisotropy of
a bead. (b) Visualization of the bead
directors in a single chain as obtained
from the backbone mean orientation. (c)
Snapshot of an equilibrated simulation
box with random chain coloring.

of gyration tensor S(u) and S(v) [see Eq. (A1)]. We then impose the and the transformation
following form:
[S(u) + S(v)]−1 r = {[A′0 + A′′0 (u ⋅ v)2 ]r
1
∣S(u) + S(v)∣
M 2 = 2(σ⊥2 σ∥ ) [S(u) + S(v)]−1
1
3 (2)
+ [A1 (u ⋅ r) + A2 (u ⋅ v)(v ⋅ r)]u
+ [A1 (v ⋅ r) + A2 (u ⋅ v)(u ⋅ r)]v}, (7)
such that in the limit of spherical beads, the correct distance is
obtained, i.e., R • R = r • r. In the particular case of two aligned ellip-
soidal beads u = v, this transformation maps the surface as described where D′ , D′′ , A′0 , A′′0 , A1 , and A2 are known constants that depend
by the shape of the beads on a sphere with the same volume. This on σ ∥ and σ – and only need to be calculated once. Details of the
modified distance can be used to construct an anisotropic potential calculation of these constants can be found in the Appendix. It
W(r, u, v) in a simple fashion by should be noted that the choices made for the normalization of
the determinant and fixing the prefactor of the interaction poten-
tial are not unique and that the softness of the interactions could be
W(r, u, v) = ∣M∣V(∣Mr∣) (3) manipulated further.
In order to compare the different shapes with spherical beads,
we define σ ∥ and σ – such that the volume of the bead remains con-
for any particular choice of isotropic interaction potential V(|R|). stant and introduce the anisotropy factor h = σ ∥ /σ – , with h = 1
The normalizing determinant |M| ensures the conservation of for spherical beads, as a convenient characteristic. Therefore, the
energy, product σ∥ σ⊥2 is kept constant for a given value of h, which is the
only additional input in the simulation. For two beads located at
∫ W(r, u, v)dr = ∫ V(∣R∣)dR. (4) a distance of r = 1, we call R the effective distance between ellip-
Ω MΩ
soids. It represents the distance between two Gaussian distributions
of the bead masses inside the ellipsoid. Values of R are shown in
In addition, we find W(r) = V(r) as the correct limit of spherical Fig. 2 for different angles between the directors and the line joining
beads. The force of the anisotropic potential yields the beads in the case of a real distance of r = 1 between the bead
positions.
M2 r As expected, the effective distance R is lower than r when the
F(r, u, v) = −∇W(r, u, v) = −∣M∣V ′ (R) . (5) directors lie along the connectivity line, whereas R > r when the
R
beads are perpendicular to it. This effect is enhanced when the
anisotropy h of the beads is increased. These reference values can
These computations are simplified by the realization that the be used to optimize the cutoff radius for anisotropic interactions.
two main terms have simple functional forms. We have a determi- Since in the implementation of neighbor lists or cell lists in simula-
nant tions the calculation of the distances is based on the real distance r,
a larger cutoff radius needs to be considered due to the presence of
∣S(u) + S(v)∣ = D′ + D′′ (u ⋅ v)2 (6) the aligned pair-interactions. Within our approach for anisotropic

J. Chem. Phys. 153, 214901 (2020); doi: 10.1063/5.0019945 153, 214901-3


Published under license by AIP Publishing
The Journal
ARTICLE scitation.org/journal/jcp
of Chemical Physics

between beads i and j. In this work, we focus on two expressions


widely used to model mesoscopic interaction: the Groot–Warren
potential (GW), which is a linear repulsive force decreasing to zero
at the cutoff radius rc ,58,59 and a truncated and shifted Lennard-Jones
potential with soft exponents 6 − 3 (TS63). Both energy and force are
considered to be zero for R > rc . For R ≤ rc , the expressions for these
potentials are given by

a
VijGW (Rij ) = rc (1 − Rij /rc )2 , (8)
2

VijTS63 (Rij ) = V LJ63 (Rij ) − V LJ63 (rc ), (9)

with
b 1
V LJ63 (r) = − . (10)
(r + rs )6 (r + rs )3
FIG. 2. Effective distance R between two ellipsoids as a function of their orientation We used a cutoff value of rc = 1 in reduced units, and the
along their joining line for different values of the anisotropy factor h. The distance parameters were fixed to b = 0.2 and rs = 0.05 for TS63 and
between the bead positions is always 1.0 (the same as for h = 1).
a = 40.0 for GW. The force for TS63 is also shifted so that at the
cutoff, the force becomes F TS63 (rc ) = 0. Since we do not attempt to
develop a realistic CG potential in this work, the interaction param-
eters were chosen to match the widely used qualitative potentials.
TABLE I. Relative simulation time as a function of the anisotropy factor h using the
spherical interactions as a reference. The system is the one described in Sec. II B The GW parameter a is chosen from other works on polymers using
(GW potential at constant-NVT), simulated during 104 steps using a replicated data qualitative interactions potentials.33,60 We have chosen to use the
MPI parallelization on 12 nodes. GW potential as it is the simplest and most documented potential
in DPD. We can also study the effect of our model on bond cross-
h t/t iso h t/t iso ings that often occur with such soft potential. The TS63 parameters
b and rs were adjusted such that the repulsive part starts at rc /2
1.00 1.08 1.75 1.31 and that the same time step and imposed temperature can be safely
1.10 1.12 2.00 1.38 applied without any divergence due to the relative softness of the
1.25 1.17 2.50 1.49 potential.
1.50 1.23 3.00 1.65 In addition to the conservative force, a dissipative force F D ,
which models the local energy dissipation between beads, and a ran-
dom force F R , representing the energy fluctuations due to Brownian
motion, are added between each pair of beads i and j,
particles, this is the main cause of an increase in computational
cost. The cutoff itself scales with the anisotropy factor as h1/3 . The F Dij = −γωD (Rij )(r̂ij ⋅ v ij )R̂ij , (11)
results presented in this work are from a homemade DPD code
designed for this purpose. The bead anisotropy and the backbone 1
mean orientation were also implemented in LAMMPS and gave sim- F Rij = σωR (Rij )ζij √ R̂ij , (12)
ilar results and performances on the test systems. The use of two dt
cutoff radii, one to detect pairs and one to apply forces, must be
checked carefully, though. where v ij = v i − v j is the relative velocity of beads i and j and ζ ij
Table I shows the relative time obtained with our system for 104 is the random number with a Gaussian distribution, zero mean, and
steps as a function of h using isotropic interactions as a reference. unit variance. The dissipative and random parameters γ and σ as
The increase of 8% for h = 1 is due to the additional calculations in well as the weighting functions ωD (Rij ) and ωR (Rij ) are related by the
the inner loop only as isotropic interactions and anisotropic interac- fluctuation–dissipation theorem and produce a correct sampling of
tions with h = 1 give the same trajectory. The largest increase in CPU the canonical ensemble61 if the following conditions are satisfied:
time is 65%, which is excellent for a system composed by ellipsoids
σ2 2
only. γ= and ωD (r) = [ωR (r)] , (13)
2kB T
B. Simulation protocol
where kB is the Boltzmann constant and T is the imposed tempera-
We use the Dissipative Particle Dynamics (DPD)56,57 method to ture. We used a standard soft repulsive expression similar to the GW
model a bulk system of linear polymer chains. A conservative force potential given by ωD (r) = 1 − R/rc and γ = 25.0 in reduced units.
F Cij defined by Eq. (5) is used to model the conservative interaction The sum of the three forces (conservative, dissipative, and random)

J. Chem. Phys. 153, 214901 (2020); doi: 10.1063/5.0019945 153, 214901-4


Published under license by AIP Publishing
The Journal
ARTICLE scitation.org/journal/jcp
of Chemical Physics

is the total force used in Eq. (5) to model the non-bonded anisotropic
interactions.
In order to maintain the connectivity in our model polymer
chains, we use a harmonic potential62,63 between bonded beads i and
j and a bending potential for triplets of bonded beads i, j, and k,

k
Vijbond = (rij − r0 )2 , (14)
2

angle
Vijk = A(1 + cos θijk ), (15)

with k = 100.0, r0 = 0.6, and A = 5.0 in reduced units. The bend-


ing potential is used as a slight correction to the ideal Gaussian
chain behavior and controls the persistence length as is observed for
realistic coarse-grained potentials.17,64 The 1–2 conservative interac-
tions between bonded beads are excluded in the GW simulations,
and both 1–2 and 1–3 interactions are excluded in the TS63 sim-
ulations. The use of an angular potential may cause an excessive
stiffness in the chains, which is a choice in this work as we want
to ensure that the backbone mean orientation hypothesis is valid for
the description of the polymer chains. However, this is a key issue
when dealing with real monomers as the resulting CG beads may
not orient correctly. This effect is supposed to vanish as the level of
coarse-graining increases, which is the kind of systems we want to
study.
Reduced units are used throughout the remainder of this work,
with the cutoff radius taken as the reference value for the reduced
distance of r∗ = r/rc . The mass of each particle is taken as unity,
the reference energy ϵ is used so that the reduced temperature is
T ∗ = kB T/ϵ, the time is t ∗ = t(ϵ/mrc2 )1/2 , and the pressure is
p∗ = p(rc3 /ϵ). The star notation is omitted in the following for
simplification as we use the reduced notation only.
A system of 400 chains of length 60 is simulated in a cubic vol-
ume of V = 20 × 20 × 20 with the usual periodic boundary conditions
at a constant temperature of T = 1.0. The forces are integrated using
a velocity-Verlet algorithm with a time step of δt = 0.010 for the sim- FIG. 3. (a) Pressure (at constant-NVT) and (b) density (at constant-NpT) as a
ulations of the GW potential, whereas the smaller value of δt = 0.005 function of anisotropy factor h for simulations using the GW and TS63 interaction
potentials. The fluctuations are smaller than 0.25 for both potentials (error bars are
is used for the TS63 potential. An equilibration period of t equ = 2000
smaller than the symbols).
is followed by an acquisition period of t acq = 5000 during which 5000
configurations are stored for the post-treatment analysis. Starting
from a randomly generated configuration, we conduct two separate
series of simulations in the constant-NVT and constant-NpT ensem-
bles, respectively. Separate runs using GW and TS63 potentials are by 10% compared to the spherical case for the very soft GW poten-
performed for each series. In both ensembles, we vary the anisotropy tial. On the contrary, the TS63 potential shows a decrease of 40%
factor h from 1.0 to 3.0, using the case of isotropic interactions of caused by the sharp repulsive part of the potential. It is interest-
h = 1 as a reference. ing to note that the pressure follows the same evolution for both
potentials.
In order to confirm these results, we plot in Fig. 3(b) the density
III. RESULTS AND DISCUSSION as a function of h for simulations at constant pressure. The pres-
sure is imposed by using a Berendsen barostat with p = 65.0 for GW
A. Thermodynamic properties and p = 15.0 for TS63. These values are chosen from the constant-
We first focus on the bulk pressure of the system as a function of NVT simulations to keep acceptable density values close to ρ = 3.0
anisotropy h, which is presented in Fig. 3(a) for both potentials GW for h = 1. Since this is a good reference value for using the GW
and TS63. Although by construction a change in h does not affect potential,58 we have designed the TS63 potential such that it gives a
the volume of each bead, the anisotropy of the beads decreases the similar density. The anisotropy of the particles results in an increase
global excluded volume of the chains, and as expected, the pressure in the density, again caused by a smaller excluded volume around
drops for elongated particles. The pressure for h = 3 is decreased the chains. This is one of the expected features of introducing the

J. Chem. Phys. 153, 214901 (2020); doi: 10.1063/5.0019945 153, 214901-5


Published under license by AIP Publishing
The Journal
ARTICLE scitation.org/journal/jcp
of Chemical Physics

bead anisotropy exposed in the introduction because coarse-grained crossings for h = 3 in all cases. The systems at constant pressure
potentials have a systematic tendency to underestimate the target exhibit more crossings than those at constant volume, but this effect
density by using spherical interactions. is merely due to the larger density of the former. The effectiveness
of anisotropy to prevent bond crossings is clearly dependent on the
B. Topology of the polymer network potential that is used, mainly because of the coupled contribution
of the non-bonded and bonded potentials. Additionally, we have to
One of the main issues caused by a large degree of coarse-
consider the role played by the random force that can cause cross-
graining is the increased length that is allowed for the bonds between
ings by itself by the random movements of the particles it creates. We
neighboring beads. This can result in the occurrence of chain cross-
must not, however, try to reach zero topology violations as false pos-
ings, which violate the topology of the chain network. This unphys-
itive cases still exist whatever the detection method. Two consecutive
ical behavior may decrease the real number of chain entanglements
crossings of the same chains in reverse directions do not also cause
dramatically, causing a wrong long-term behavior when calculating
any topology change but count as two crossings. This is why we only
the dynamic properties of polymer melts. In this section, we want
aim at decreasing the relative number of crossings using spherical
to examine the ability of the anisotropic potential to decrease the
interactions as a reference.
number of crossings.
Prevention of chain crossing is usually achieved in mesoscopic
We can detect the crossing of two bonds by the change in their
simulations by adding repulsive interactions to the system such as
relative positions during a time step, i.e., between t and t + δt. A way
segment repulsion or extra particles located within each bond seg-
of doing so is to keep track of every distance vector between bonds
ment.33,60,65,66 However, these methods have two main drawbacks.
and compare them after each time step by calculating their scalar
The first is that they increase the computing time because the addi-
product: a crossing is detected if the angle between the two distance
tional centers of force lead to an effective decrease in the coarse-
vectors is larger than π/2. A detailed description of this method and
grain level. The second problem is that these repulsive forces also
others can be found in a previous article.33 This manner of deter-
add conservative interactions that modify the thermodynamic prop-
mining whether a crossing has occurred is not perfect because events
erties. These deficiencies are undesirable and should be avoided
can be missed as well as false positives may be detected. An example
when developing a realistic coarse-grained potential. In the case
of such a situation is when a crossing in one direction is followed
of a realistic potential, the bond crossings are rarely checked as
by a subsequent crossing in the opposite direction, leading to two
the interactions are supposed to be repulsive enough to keep the
crossings being counted despite the fact that no change in the sys-
topology. However, this may not be the case for high levels of
tem topology has occurred. Therefore, the goal is not to decrease the
coarse-graining. Furthermore, the case of nonequilibrium simula-
number of detected crossings down to zero but to obtain a relative
tions has to be specifically considered, for example, an imposed flux
decrease in crossings due to the anisotropy of the beads. The relative
may cause additional topology violations. We show that even for
number of crossings is shown in Fig. 4 for each set of simulations,
harder potentials such as TS63, there is a significant decrease in the
where we have used the number of crossings at h = 1 as the reference
bond crossings when considering anisotropic beads. The use of non-
value.
spherical shapes for polymer beads seems to be a very efficient way to
We observe a net decrease in the overall bond crossings for
reduce the number of crossings of the chains without any additional
increasing values of h, which leads to more than 80% reduction in
interaction or substantially increasing the required computational
time.

C. Structure
The increase in the density as a function of h shown in Sec. III A
is clearly caused by the stacking of the chains closer to each other
that is made possible by the decrease in the tube excluded volume. In
this section, we investigate this phenomenon by examining the local
structure and order of the polymer beads. The definition of a director
via the polymer backbone, including for the case of spherical parti-
cles, allows us to have a more detailed examination of the radial dis-
tribution function. For each particle, we split the local environment
into two regions, as shown in Fig. 5.
If we consider the director u of the reference bead, region A
contains the two cones corresponding to the parallel direction of the
bead, i.e., the region where the rest of the chain has the highest prob-
ability to be found. Region B is the part of space around the chain
direction and will be occupied more likely by other chains. In order
to have the same statistics when calculating both radial distribution
functions g [A] (r) and g [B] (r), we impose the same volume for A and
FIG. 4. The relative number of bond crossings as a function of the anisotropy factor B by choosing an angle of 60○ between u and the cone surface.
h for simulations in the constant-NVT and constant-NpT ensembles and for both
Figure 6(a) shows the contributions of regions A and B to the
GW and TS63 interaction potentials.
radial distribution function (RDF) for spherical beads simulated

J. Chem. Phys. 153, 214901 (2020); doi: 10.1063/5.0019945 153, 214901-6


Published under license by AIP Publishing
The Journal
ARTICLE scitation.org/journal/jcp
of Chemical Physics

with the TS63 potential. Since the bonded particles are excluded
from the analysis, region B shows an obvious lack of particles in the
first peak. It is interesting to note that the second neighbor peak is,
in fact, the combination of a maximum in region A, correspond-
ing to the second bonded beads, and a minimum in region B that
is mainly due to the non-bonded interactions falling to zero at the
cutoff.
The resulting RDFs of regions A and B for different values
of the anisotropy h are shown in Figs. 6(b) and 6(c), respectively.
For region A, only a slight vertical shift is observed in the range of
r ∈ [0.7, 0.9]. This is due to the fact that region A contains mainly
particles from the same chain because even for distances larger than
the cutoff radius, the chain structure is imposed by the intramolec-
ular bonding and bending potentials. The change in the local struc-
ture for increasing values of h is much more significant in region B
around the chains, with a noticeable horizontal shift that allows for
smaller distances between particles of different chains. This might
also be responsible for the vertical shift seen in the A region, where
FIG. 5. Schematic representation of the A and B regions used to split the radial the first neighbors of the bonded particles can get closer as well when
distribution for a particular bead. the anisotropy is increased. Finally, the first peak of g [B] (r) spreads

FIG. 6. (a) Contributions of the A and B regions to the total radial distribution function for non-bonded spherical beads using the TS63 potential. [(b) and (c)] Radial distribution
function of non-bonded beads located in the A and B regions, respectively, for different values of the anisotropy parameter h. (d) Zoom of g(r) at the shortest distances for
h = 1 and h = 3 using the GW potential.

J. Chem. Phys. 153, 214901 (2020); doi: 10.1063/5.0019945 153, 214901-7


Published under license by AIP Publishing
The Journal
ARTICLE scitation.org/journal/jcp
of Chemical Physics

a little for larger values of h due to the variety in relative orienta- long ranges, by separating the calculation of S2 in four contributions
tions of the ellipsoids that allow for a wider range of distances to corresponding to different intervals for distances rij between beads.
be possible. Interestingly, it seems that the short distances are more Figure 7(a) shows the different intervals chosen from the RDF using
probable with high values of h despite the fact that the number of different colors for each zone: 1 includes the first neighbors only, 2
bond crossings decreases as h increases. Figure 6(d) shows the com- is for close particles, 3 represents an intermediate range, and 4 is for
parison of the total RDF for h = 1 and h = 3 on a logarithmic scale: large distances rij > 4rc . Figure 7(b) shows the different values of S2
√ distances occur more often for h = 1, more precisely for
the smallest as a function of h for each of the four zones as calculated for every
r < r0 / 2 ≈ 0.5. This value corresponds to the most symmetrical set of simulations using the TS63 potential.
configuration for two orthogonal bonds during crossing. The order parameter is non-zero for short distances even for
The conclusion that the chains stack more with each other for spherical particles because of the underlying chain structure. How-
increasing h can also be supported by examining the relative orien- ever, the value of S2 increases with h, which is in line with the parallel
tations of the beads. The orientational order of the system can be stacking hypothesis. The orientational order decreases rapidly for
quantified by means of the nematic order parameter S2 , distances larger than 1.5rc , and already, zone 3 shows almost no
order. This is a good result because it is undesirable to have long-
3 1 range order in a polymer melt. Similar results are obtained for the
S2 = ⟨ cos2 θij − ⟩, (16) GW potential. The difference that is observed between constant-
2 2
NVT and constant-NpT simulations as function of h is, as was men-
where θij is the angle between the directors of beads i and j. In order tioned before, due to the differences in density when comparing
to avoid the systematic correlation between the directors of bonded the two types of ensemble and the subsequential effect on the local
particles, the ensemble average ⟨⋯⟩ is performed only between pairs orientational order.
of non-bonded beads. We distinguish the behavior at the short and

IV. CONCLUSION
We have developed a model that uses anisotropic interactions
for the case of coarse-grained polymer chains. Using the backbone
mean orientation, we are able to compute the orientation of poly-
mer beads at each time step from the bead positions only. A simple
calculation of the force between ellipsoids allows us to model non-
spherical particles with only a small increase in computing time. We
introduce the anisotropy factor h = σ ∥ /σ – as the only parameter to
describe the shape of elongated particles designed to increase the
performance of coarse-grained potentials for large degrees of coarse-
graining. We have demonstrated by using two particular interaction
potentials that the elongation of the beads results in an increase in
the density and results in the excluded volume of the chains to have
a shape closer to that of a tube.
Despite allowing closer distances between non-bonded beads,
the increase in anisotropy significantly suppresses the probability
of unphysical crossings between chains. This is a very promising
feature of our model because the elimination of this artificial pro-
cess encountered in simulations of polymer chains is a key require-
ment for developing models with high levels of coarse-graining and
realistic softer mesoscopic potentials. We also plan to address the
effectiveness of our model on a sheared system as the topology
of the polymer network is crucial in these systems. Additionally,
the short-range induced order by anisotropy can be of interest for
coarse-grained simulations of semi-crystalline polymers, where it
may facilitate crystallization.
The fact that only one parameter is added in the model with
no additional interactions makes this work a good starting point
to include anisotropy in the design of realistic coarse-grain poten-
FIG. 7. (a) Different zones based on the radial distribution functions used to cal- tials. The anisotropy factor may be deduced from the average radius
culate the order parameter: (1) very close to, (2) close to, (3) far from, and (4) of a gyration matrix calculated from an all-atom reference tra-
very far from the central bead. (b) The nematic order parameter S2 as a function jectory and then included in various coarse-graining methods as
of the anisotropy parameter h in each of the four zones for the case of the TS63 an input. We are currently developing such a method to include
potential. Solid lines are for constant-NVT simulations, and dashed lines are for
anisotropy as a key parameter for interactions at the mesoscopic
constant-NpT simulations.
scale.

J. Chem. Phys. 153, 214901 (2020); doi: 10.1063/5.0019945 153, 214901-8


Published under license by AIP Publishing
The Journal
ARTICLE scitation.org/journal/jcp
of Chemical Physics

ACKNOWLEDGMENTS REFERENCES
1
This work was performed in SimatLab, a joint public-private K. Binder, Monte Carlo and Molecular Dynamics Simulations in Polymer Science
laboratory dedicated to the modeling of polymer materials. This (Oxford University Press, 1995).
2
laboratory is supported by Michelin, Clermont Auvergne Univer- M. Kotelyanskii and D. N. Theodorou, Simulation Methods for Polymers (Marcel
sity (UCA), SIGMA Clermont, and CNRS. SimatLab acknowledges Dekker, 2004).
3
G. S. Grest and K. Kremer, Phys. Rev. A 33, 3628–3631 (1986).
the support received from the Agence Nationale de la Recherche 4
A. Baumgärter and K. Binder, J. Chem. Phys. 75, 2994–3005 (1981).
of the French government through the program “Investissements 5
K. Binder, Computational Modelling of Polymers (Marcel Dekker, 1992).
d’Avenir” (Grant No. 16-IDEX-0001 CAP 20-25). 6
G. S. Grest, Curr. Opin. Colloid Interface Sci. 2, 271–277 (1997).
7
T. Kreer, M. H. Müser, K. Binder, and J. Klein, Langmuir 17, 7804–7813
(2001).
8
J. T. Padding and W. J. Briels, J. Chem. Phys. 115, 2846–2859 (2001).
APPENDIX: CALCULATION
OF D ′ , D ′′ , A′0 , A′′0 , A 1 , AND A 2
9
J. T. Padding and W. J. Briels, J. Chem. Phys. 117, 925–943 (2002).
10
R. D. Groot and K. L. Rabone, Biophys. J. 81, 725–736 (2001).
The radius of gyration tensor S(u) of an ellipsoidal bead with 11
A. Ghoufi and P. Malfreyt, Phys. Rev. E 83, 051601 (2011).
12
an orientation described by the unit-vector u is given by A. Ghoufi and P. Malfreyt, J. Chem. Theory Comput. 8, 787–791 (2012).
13
T. Spyriouni, C. Tzoumanekas, D. Theodorou, F. Müller-Plathe, and G. Milano,
Macromolecules 40, 3876–3885 (2007).
S(u) = σ⊥ 1 + [σ∥ − σ⊥ ]π(u), (A1) 14
V. A. Harmandis, N. P. Adhikari, N. F. A. van der Vegt, and K. Kremer,
Macromolecules 39, 6708–6719 (2006).
15
V. A. Harmandis and K. Kremer, Macromolecules 42, 791–802 (2009).
where παβ (u) = uα uβ is the projector operator along u and σ – and σ ∥ 16
T. Mulder, V. A. Harmandaris, A. V. Lyulin, N. F. A. van der Vegt, K. Kremer,
are the eigenvalues associated with the short and long axis, respec- and M. A. J. Michels, Macromolecules 42, 384–391 (2009).
tively. It can easily be derived or checked that the determinant of the 17
G. Maurel, B. Schnell, F. Goujon, M. Couty, and P. Malfreyt, J. Chem. Theory
matrix S(u) + S(v) is given by Comput. 8, 4570–4579 (2012).
18
L. Lu, J. F. Dama, and G. A. Voth, J. Chem. Phys. 139, 121906 (2013).
19
D. Reith, M. Pütz, and F. Müller-Plathe, J. Comput. Chem. 24, 1624–1636
∣S(u) + S(v)∣ = D′ + D′′ (u ⋅ v)2 , (A2) (2003).
20
J. G. Kirkwood, J. Chem. Phys. 3, 300–313 (1935).
21
F. Ercolessi and J. B. Adams, Europhys. Lett. 26, 583–588 (1994).
with 22
S. Izvekov, M. Parrinello, C. J. Burnham, and G. A. Voth, J. Chem. Phys. 120,
10896 (2004).
D′ = 2σ⊥ (σ∥ + σ⊥ ) ,
2 23
C. Hijón, P. Español, E. Vanden-Eijnden, and R. Delgado-Buscalioni, Faraday
(A3) Discuss. 144, 301–322 (2010).
D′′ = −2σ⊥ (σ∥ − σ⊥ ) .
2 24
Z. Li, X. Bian, X. Yang, and G. E. Karniadakis, J. Chem. Phys. 145, 044102
(2016).
25
Likewise, the inverse can be derived with the aid of the Cayley– C. A. Lemarchand, M. Couty, and B. Rousseau, J. Chem. Phys. 146, 074904
(2017).
Hamilton theorem to yield 26
W. G. Noid, J. Chem. Phys. 139, 090901 (2013).
27
A. Dequidt and J. G. Solano Canchaya, J. Chem. Phys. 143, 084122 (2015).
28
[S(u) + S(v)]−1 =
1 J. G. Solano Canchaya, A. Dequidt, F. Goujon, and P. Malfreyt, J. Chem. Phys.
∣S(u) + S(v)∣ 145, 054107 (2016).
29
K. Kempfer, J. Devémy, A. Dequidt, M. Couty, and P. Malfreyt, Macromolecules
× {A′0 + A′′0 (u ⋅ v)2 + A1 [π(u) + π(v)] 52, 2736–2747 (2019).
30
J. G. S. Canchaya, A. Dequidt, S. Garruchet, B. Latour, N. Martzel, J. Devémy,
+ A2 [π(u)π(v) + π(v)π(u)]}, (A4)
F. Goujon, R. Blaak, B. Schnell, E. Munch, N. Seeboth, and P. Malfreyt, J. Chem.
Phys. 151, 064703 (2019).
31
K. Kempfer, J. Devémy, A. Dequidt, M. Couty, and P. Malfreyt, Soft Matter 16,
with the constants 1538–1547 (2020).
32
C. Tzoumanekas and D. Theodorou, Curr. Opin. Solid State Mater. Sci. 10,
A′0 = (σ∥ + σ⊥ ) ,
2
61–72 (2006).
33

A′′0 = −(σ∥ − σ⊥ ) ,
2 F. Goujon, P. Malfreyt, and D. J. Tildesley, J. Chem. Phys. 129, 034902 (2008).
34
N. Iwaoka, K. Hagita, and H. Takano, J. Chem. Phys. 149, 114901 (2018).
(A5) 35
A1 = −(σ∥2 − σ⊥2 ), 36
A. E. Likhtman, Macromolecules 38, 6128–6139 (2005).
A. Korolkovas, P. Gutfreund, and J.-L. Barrat, J. Chem. Phys. 145, 124113
2
A2 = (σ∥ − σ⊥ ) . (2016).
37
N. J. H. Dunn and W. G. Noid, J. Chem. Phys. 143, 243148 (2015).
38
K. Kempfer, J. Devémy, A. Dequidt, M. Couty, and P. Malfreyt, ACS Omega 4,
DATA AVAILABILITY 5955–5967 (2019).
39
N. Martzel, A. Dequidt, J. Devémy, R. Blaak, S. Garruchet, B. Latour, F. Goujon,
The data that support the findings of this study are available E. Munch, and P. Malfreyt, Adv. Theory Simul. 3, 2000124 (2020).
40
from the corresponding author upon reasonable request. B. J. Berne and P. Pechukas, J. Chem. Phys. 56, 4213–4216 (1972).

J. Chem. Phys. 153, 214901 (2020); doi: 10.1063/5.0019945 153, 214901-9


Published under license by AIP Publishing
The Journal
ARTICLE scitation.org/journal/jcp
of Chemical Physics

41 54
C. K. Lee, C. C. Hua, and S. A. Chen, J. Chem. Phys. 133, 064902 (2010). Z. E. Hughes, L. M. Stimson, H. Slim, J. S. Lintuvuori, J. M. Ilnytskyi, and M. R.
42 Wilson, Comput. Phys. Commun. 178, 724–731 (2008).
C. K. Lee, C. C. Hua, and S. A. Chen, J. Chem. Phys. 136, 084901 (2012).
43 55
M. Babadi, R. Everaers, and M. R. Ejtehadi, J. Chem. Phys. 124, 174708 (2006). J. S. Lintuvuori and M. R. Wilson, J. Chem. Phys. 128, 044906 (2008).
44 56
A. F. Tillack, L. E. Johnson, B. E. Eichinger, and B. H. Robinson, J. Chem. Theory P. J. Hoogerbrugge and J. M. V. A. Koelman, Europhys. Lett. 19, 155–160
Comput. 12, 4362–4374 (2016). (1992).
45 57
O. Hahn, L. D. Site, and K. Kremer, Macromol. Theory Simul. 10, 288–303 J. M. V. A. Koelman and P. J. Hoogerbrugge, Europhys. Lett. 21, 363–368
(2001). (1993).
46 58
L. Paramonov and S. N. Yaliraki, J. Chem. Phys. 123, 194111 (2005). R. D. Groot and P. B. Warren, J. Chem. Phys. 107, 4423–4435 (1997).
47 59
N.-V. Buchete, J. E. Straub, and D. Thirumalai, J. Mol. Graphics Modell. 22, P. B. Warren, Curr. Opin. Colloid Interface Sci. 3, 620–624 (1998).
60
441–450 (2004). F. Goujon, P. Malfreyt, and D. J. Tildesley, Macromolecules 42, 4310–4318
48 (2009).
J. Wu, X. Zhen, H. Shen, G. Li, and P. Ren, J. Chem. Phys. 135, 155104 (2011).
49 61
H. Shen, Y. Li, P. Ren, D. Zhang, and G. Li, J. Chem. Theory Comput. 10, P. Español and P. Warren, Europhys. Lett. 30, 191–196 (1995).
62
731–750 (2014). Y. Kong, C. W. Manke, W. G. Madden, and A. G. Schlijper, Int. J. Thermophys.
50 15, 1093–1101 (1994).
G. Li, H. Shen, D. Zhang, Y. Li, and H. Wang, J. Chem. Theory Comput. 12,
63
676–693 (2016). A. G. Schlijper, P. J. Hoogerbrugge, and C. W. Manke, J. Rheol. 39, 567–579
51 (1995).
G. G. Rondina, M. C. Böhm, and F. Müller-Plathe, J. Chem. Theory Comput.
64
16, 1431–1447 (2020). K. Kempfer, J. Devémy, A. Dequidt, M. Couty, and P. Malfreyt, ACS Appl.
52 Polym. Mater. 1, 969–981 (2019).
M. K. Meinel and F. Müller-Plathe, J. Chem. Theory Comput. 16, 1411–1419
65
(2020). S. Kumar and R. G. Larson, J. Chem. Phys. 114, 6937–6941 (2001).
53 66
M. R. Wilson, Chem. Soc. Rev. 36, 1881–1888 (2007). S. P. Holleran and R. G. Larson, Rheol. Acta 47, 3–17 (2008).

J. Chem. Phys. 153, 214901 (2020); doi: 10.1063/5.0019945 153, 214901-10


Published under license by AIP Publishing

You might also like