Polymorphism of syndiotactic polystyrene crystals from multiscale simulations Chan Liu,1 Kurt Kremer,1 and Tristan Bereau1, a) Max Planck Institute for Polymer Research, Ackermannweg 10, 55128 Mainz, Germany (Dated: 4 May 2019) Syndiotactic polystyrene (sPS) exhibits complex polymorphic behavior upon crystallization. Computational modeling of polymer crystallization has remained a challenging task because the relevant processes are slow on the molecular time scale. We report herein a detailed characterization of sPS-crystal polymorphism by means of coarse-grained (CG) and atomistic (AA) modeling. The CG model, parametrized in the melt, shows remarkable transferability properties in the crystalline phase. Not only is the transition temperature in good agreement with atomistic simulations, it stabilizes the main α and β polymorphs, observed experimentally. We compare in detail the propensity of polymorphs at the CG and AA level and discuss finite-size as well as box-geometry effects. All in all, we demontrate the appeal of CG modeling to efficiently characterize polymer-crystal polymorphism at large scale.

I. INTRODUCTION bead B

The crystallization of molecular assemblies often in- volves the stabilization of a variety of distinct conform- ers, denoted polymorphism.1 The impact of polymorphs, may it be thermodynamic, structural, or physiological, makes their understanding and prediction essential for (a) a broad area of research, from hard condensed mat- bead A ter to organic materials. The computational modeling of crystalline materials has rapidly developed in recent years, from crystallization2,3 to predicting polymorphs and their relative stability4–7. These developments are remarkable given the complexity of the system: from the (b) (c) (d) molecular architecture to sampling challenges at low tem- peratures to the significant timescales involved. For this FIG. 1. (a) Representation of syndiotactic polystyrene, in- reason, past computational studies focused on crystalliza- cluding the CG mapping scheme: each monomer is mapped tion and/or polymorphism have been limited to molec- onto two coarse-grained beads. CG bead A is the center of ular crystals—based on small molecules. Here instead, mass of the backbone connecting two sidechains, while bead we extend the computational characterization of poly- B is the center of mass of the phenyl group. Polymorphism: mers crystals,8–11 by modeling for the first time its self (b) α, (c) β, (d) δ forms. assembly and resulting polymorphs. Syndiotactic polystyrene (sPS) is known for its unusual crystal polymorphism. Five different crystalline forms the limiting ordered β00 model is obtained by crystalliza- have been reported experimentally. Fig. 1 shows three tion from solution, when the solvent is rapidly removed main forms studied in this work. The unit cells are pro- at higher temperatures above 150◦C16. Two of the heli- jected onto the a b plane, and the backbones of the − cal crystalline phases (δ and ε) can only be obtained by chains are along the c axis. Experimentally, two types guest removal from co-crystalline phases. The δ form of crystalline phases have been identified: the zig-zag- e 12–15 16,17 is transformed into the solvent-free γ form by annealing chain forming α and β appear upon thermal an- above 130◦C18,20,26. These findings illustrate that the nealing of a melt, whereas the other three helix-forming 18,19 20–22 23–25 experimentally-observed structures depend on the exper- arXiv:1805.10147v1 [cond-mat.soft] 25 May 2018 crystalline phases, γ , δ and ε are obtained imental processing, making conclusions about their ther- by solution processing. The α and β forms of sPS are modynamic equilibrium difficult. further classified into the limiting disordered forms (α0 and β0) and limiting ordered forms (α00 and β00)12,14,16,26. Several molecular simulation studies have been per- In particular, melt crystallization procedures generally formed to better understand the nanoporous cavity struc- 00 tures formed by crystalline syndiotactic polystyrene. produce the limiting ordered α and limiting disordered 27 β0 models26. The limiting disordered α0 model is ob- Tamai and his co-workers studied the size, shape and tained by annealing the amorphous sample26, whereas connectivity of the cavities in the crystal α, β and δ forms. Some other properties were also studied, such as diffusion of gases28,29, reorientational motion of guest solvents30,31, and sorption of small molecules32–34. While a) Electronic mail: [email protected] these studies helped understand the behavior of specific 2 forms of sPS crystals, they did not address the crys- More importantly, this finding illustrates the remarkable tallization process or the relative stability of the poly- transferability properties of the CRW method. morphs. By studying sPS on the nanosecond timescale, In this paper, we report both AA and CG simulations they could not observe self assembly and spontaneous of the temperature-driven phase transition between crys- polymorph interconversion, due to their metastability. tal and melt. Annealing simulations led to highly-ordered Yamamoto observed the onset of isotactic polypropylene conformations, but exhibiting significant hysteretic be- crystallization, but leading to a smectic mesophase in- havior. Replica-exchange molecular dynamics (REMD) stead of crystal polymorphs35. The simulation of poly- simulations helped the conformational sampling and the mer crystallization remains challenging because these identification of the transition temperature. Backmap- processes are slow on the molecular time scale. In ad- ping of CG snapshots provided starting AA configura- dition, the simulated chain lengths must be large enough tions. For both resolutions, we consider several poly- to link to experimentally realistic situations36. As an al- morphs, α, β and δ, observed experimentally. We find ternative to atomistic (AA) simulations, coarse-grained that the two models stabilize α and β, but not δ. We (CG) models37–39 combine a number of atoms into su- discuss the impact of finite-size effects and box geome- peratoms or beads to significantly speed up the simula- tries. To alleviate these artefacts, we use the CG model tions, while potentially retaining enough chemical detail to construct a larger system and discuss the convergence to differentiate polymorphs. Its capability to reach longer of polymorph populations. length and timescales is strengthened by an observation we present below: the relatively small systems we can afford at the AA level display finite-size effects. II. METHODS Several CG models for polystyrene have been re- A. Model ported, but so far focusing solely on the amorphous phase. In several models one CG bead represents a sin- gle monomer40–43. Milano and Mueller-Plathe40,41 cen- 1. AA Model ter the beads on methylene carbons. The model of Qian and coworkers42 places the bead on the center of mass of We used the atomistic (AA) model of polystyrene from the monomer. Both of these models use different bead Mueller-Plathe49. In this model, the phenyl groups are types to represent different types of diads, thus keep- described using parameters from benzene, while the pa- ing information about the chain stereosequences. Sun rameters for the aliphatic carbons and hydrogens were and Faller43 center the beads on the backbone carbons identical to aliphatic polymers. The rotation of the to which the phenyl rings are attached. In this model, phenyl rings is left free. All bond lengths were con- each bead represents one PS monomer without distin- strained by the LINCS method50. guishing between different types of diads. Single-bead representations for a polystyrene monomer represent sig- nificant issues to describe polymorphism, as they lack 2. CG Model the necessary degrees of freedom to distinguish between the relevant forms. Models of higher resolution were also The CG polystyrene force field of Fritz et al.46 is de- developed: Harmandaris and coworkers44,45 devised two rived from the above-mentioned AA model. It maps each different two-bead-per-monomer models. In their first monomer onto two CG beads of different types, denoted 44 model , the CH2 group of the backbone chain is rep- A for the chain backbone and B for the phenyl ring (see resented by one CG bead, whereas the remaining CH Fig. 1a). We stress that the CG model represents PS group of the monomer in the backbone and its pen- by a linear chain, where the beads are connected by CG dant phenyl ring are represented by another CG bead. bonds A–B. There are no bonds between the A beads, In their second model45, the phenyl ring is represented and the close connection between them is reproduced in- by one CG bead, while the other bead represents the directly by the angular potentials θABA. The tacticity of CH2 of the backbone as well as a contribution from each polystyrene is encoded in the potential parameters. Here one of the two neighboring CH groups. By comparison, we only use the parameters of syndiotactic polystyrene the second mapping scheme better reproduces the local whose chains only consist of racemic diads. chain conformations and melt packing observed in atom- The CG force field includes the bonded and nonbonded istic simulations. Based on this mapping scheme, Fritz interaction potentials in a tabulated form and are derived and coworkers46 developed a new CG force field using separately. Potentials for bonded degrees of freedom of the conditional reversible work (CRW) method47,48, a the CG model are obtained by direct Boltzmann inver- thermodynamic-based CG method that samples isolated sion of distributions obtained from atomistic simulations atomistic chains or pairs of oligomers in vacuum. In this of single chains in vacuum using stochastic dynamics. paper, we find that this CG model can successfully crys- The nonbonded potentials are derived by the condi- tallize sPS. The bonded force field seems to adequately tional reversible work (CRW) method47,48. They are ob- reproduce the local chain conformations of syndiotactic tained from constraint dynamics runs with the all-atom polystyrene so that the sterics allow for crystallization. model of two trimers (or fourmers) in vacuum. In these 3 runs the atoms mapping to a pair of beads A or B were fixed. We have found that anisotropic coupling led to held at fixed distance r. The pair potential of mean force large fluctuations close to the phase transition. (PMF), VPMF, was calculated by Z r VPMF(r) = ds [ fc s] + 2kBT ln r, (1) C. Order parameter rm h i where kB is the Boltzmann constant, T is the tempera- Orientational order parameters are useful indicators of ture, fc is the constraint force between the two CG map- crystal formation and crystal growth. In this work, calcu- ping points, s is a constrained ensemble average, and lating order parameters in both AA and CG simulations h·i rm is the maximum distance between the two mapping is based on geometries at the CG level, and we take one points. racemic diad (two monomers) as one characterizing unit. i The effective, nonbonded A–A interaction potential is In a polymer chain, for a side-chain bead B1 in unit i (see next obtained from Fig. 2), we identify an orientation vector which points i i+1 from B1 to B1 and normalize it to a unit vector AA AA excl,AA Veff (r) = VPMF(r) VPMF (r) (2) − ri+1 ri ei = − . (3) excl,AA ri+1 ri where the second PMF, VPMF , is along the same co- | − | ordinate r but excludes all direct A–A atomistic inter- i actions while maintaining all other interactions with and Similar orientation vectors can be obtained for bead A1, i i e between neighboring parts of the oligomers. A similar A2 and B2. For two given orientation vectors i and e procedure is applied to A–B and B–B. j, the order parameter between them is calculated as follows55: One advantage of this approach is that it is computa- tionally inexpensive. More importantly, this approach is 3 2 1 Pij = (ei ej) . (4) widely different from many other CG methods that use 2 · − 2 information on condensed-phase properties of the melt state as input in the development of the effective poten- The order parameter of the whole system is the average tials. This crucial difference enables the greater transfer- over all possible Pij in the system. ability reported herein. Note that this order parameter has some limitations: it can only recognize trans-planar chain conformations. As mentioned in the introduction, syndiotactic polystyrene B. Replica exchange simulations has five different crystalline forms. To further study the phase transition between these different forms, we devel- oped a procedure to characterize the different forms in 51 Replica exchange molecular dynamics (REMD) is these ordered states. used to enhance the sampling of computer simulations, especially when metastable states are separated by rela- tively high energy barriers. It involves simulating multi- ple replicas of the same system at different temperatures D. Form characterization and randomly exchanging the complete state of two repli- cas at regular intervals with a Metropolis criterion. In analogy to the order parameter, form characteriza- All simulations reported in this study were performed tion in both AA and CG simulations is based on confor- using the molecular dynamics package GROMACS 4.652. mations at the CG level, and we take two monomers as The integration time step of the AA model was 1 fs. AA- one characterizing unit. RE simulations were performed with 48 (56 for the β There are two types of single-chain conformations (see form) temperatures from 250 K to 550 K (700 K for the Fig. 2): trans-planar chains (α and β) and helical chains β form), and replica exchange was attempted every 1 ps. (γ, δ and ε). First, we identify these two different crys- Simulations were carried out in the NPT ensemble using talline chain conformations from amorphous chains. In a 53 AA the stochastic velocity rescaling thermostat (τT = 0.2 polymer chain, for unit i, we identify a backbone vector 54 AA i i ps) and the Berendsen barostat (τP = 2 ps). For the bi which points from bead A2 to bead A1. When the −3 ◦ CG model, the integration time step was 10 τ, where τ angle between vector bi and vector bi+1 is around 0 , defines the reduced unit of time in the CG model. CG-RE the unit i and i + 1 are trans-planar chain segments (see simulations were performed with 48 temperatures from Fig. 2 (a)). When the angle between vector bi and vec- ◦ 0 250 K to 550 K, and replica exchange was attempted ev- tor bi+1 is around 70 and the angle between vector b i i+1 i ◦ ery τ. In CG simulations, we also employed the stochas- (from bead A1 to bead A2) and vector bi is around 25 , CG tic velocity rescaling thermostat (τT = 0.2 τ) and the the unit i and i + 1 are helical chain segments (see Fig. 2 CG Berendsen barostat (τP = 2 τ). Note that in all replica (b)). The remaining segments are considered amorphous. exchange simulations, we use isotropic pressure coupling, For trans-planar chain segments, we discriminate crys- which means that the angles of the simulation box are tal regions between different chain-stretching directions. 4

Thermally induced Solvent-induced 1 α and β γ, δ and ε 0.8 { y Planar zigzag Helix 0.6 i i B1 A1 0.4 x z Cooling

i Order Parameter unit i i bi Heating bi 0.2 A2 B2 bi0 Recooling ei bi+1 Reheating unit i + 1 0 i+1 bi+1 250 300 350 400 450 500 B1 T [K]

FIG. 3. Evolution of a representative structural order pa- rameter (where a value of 1 is completely ordered and 0 is disordered) during the cooling, heating, recooling and reheat- ing CG-MD simulations.

including the melt packing and the density between 400 and 520 K, we use this model to perform annealing simu- lations to test its ability to crystallize. The initial system is an sPS melt that contains 9 chains, each comprising 10 monomers. To crystallize the system, annealing sim- (a) (b) ulation is performed from T = 500 K to T = 250 K, with a total cooling time of 2.5 105 τ. This is then fol- × FIG. 2. Five different crystalline forms: two crystalline phases lowed by a heating procedure back to T = 500 K at the (α and β) with trans-planar chains; three crystalline phases same rate. The evolution of a commonly-used structural (γ, δ and ε) with the s(2/1)2 helical chains. order parameter during the cooling, heating, recooling and reheating processes is displayed in Fig. 3. Note that the simulation box is rectangular under isotropic pressure Then in each region and each single layer, which is per- coupling. It shows that highly-ordered conformations can pendicular to the chain stretching direction, we measure indeed be obtained from annealing simulations, but the the distances between all beads of type A in the same fast annealing rate leads to significant hysteresis. layer. When the distances between two such beads is smaller than 0.6 nm, we associate them into the same group. In each group, we measure the orientations of side 2. Replica exchange simulations from CG and AA models chains to identify α and β forms (see Fig. 1(b)(c)). If the orientations are all parallel, the group is a “full β” group; To improve the sampling efficiency and accurately if the angles of the orientations are all around 120◦, the characterize the transition behavior, we employed replica group is a “full α” group. Some groups have features of exchange molecular dynamics (REMD)51 in the CG and both α and β forms, in which case we call them “mix” AA simulations. Simulating a crystallization process us- (see Fig. 7). In case only a single characterizing unit is ing an atomistic model is difficult. Backmapping is a found in the group, we call it “defect.” good strategy to reconstruct an atomistic configuration In this work, we focus on the main α and β polymorphs from a CG snapshot. We use backmapping to generate found in annealing experiments, while all helical chain a highly ordered atomistic configuration, which is taken segments are herein called δ for convenience. as the initial structure in the AA simulations. Fig. 4(a) shows that both AA and CG replica exchange simulations display a phase transition, and the transition tempera- III. RESULTS AND DISCUSSION tures are both around 450 K. This indicates that this CG model can reproduce the phase transition of sPS roughly at the correct temperature. This overall hints at remark- A. Crystallization of the CG model able transferability of the CG model between the melt and the crystal. 1. CG annealing simulations However, some differences appear between these two models at low temperatures. Fig. 4 (b) and (c) show Since the CG model of Fritz et al.46 derived using CRW multiple peaks in the order-parameter distributions of method from AA simulations can predict melt properties, the AA model, while only two peaks are found in the 5

1 an α00 form, as determined by X-ray diffraction experi- (a) ments14. The limiting-ordered α00 form is characterized 0.8 by a specific positioning of triplets of chains, e.g., one triplet is oriented in one direction, while the other two 0.6 in the opposite directions. The limiting disordered α0 form, on the other hand, has a statistical disorder be- 0.4 tween these two orientations of triplets of chains. Fig. 6 Order Parameter 0.2 (b) shows a representative snapshot obtained from CG- AA-REMD RE simulations at 250 K. We find that all the triplets CG-REMD 0 of chains sampled from the CG model always display the 200 250 300 350 400 450 500 550 600 same arrangement of triplets—characteristic of α0—while T [K] α00 does not seem stable in the CG model. 0.16 0.14 (b) AA (c) CG 250K 400.00K 0.12 294.59K 416.37K 2. β form 348.08K 433.52K 0.1 412.17K 451.68K 0.08 489.01K 470.88K 532.94K 482.92K Fig. 5 (b) shows the initial β configuration, containing 0.06

Probability 12 chains of 10 monomers each. We performed RE sim- 0.04 ulations for the CG and AA models for 2 106 τ and 0.02 200 ns, respectively. The CG model predominently× sta- 0 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 bilizes a mixture of α and β forms below the transition Order Parameter Order Parameter temperature (see Fig. 5 (b-1)). While many snapshots are made of large homogeneous regions of α or β poly- FIG. 4. (a) Structural order parameter as a function of morphs, we also find a significant population that mix temperature obtained from AA and CG REMD simulations. the two, denoted “mix” (Fig. 7). The presence of α/β Structural order-parameter distributions as a function of tem- mixtures breaks the long-range order expected in homo- perature: (b) AA and (c) CG. geneous phases and resembles a form of intermediate be- tween the two polymorphs. This mixing has not been reported experimentally. CG model. The CG model fails to display the structural Fig. 8(a) shows the order parameter as a function of heterogeneity of its AA counterpart. temperature obtained from CG and AA REMD simula- tions. Comparing the two resolutions, we find that the transition temperatures agree when simulated in the α- B. Crystal polymorphs compatible box shape, but not for β. The AA model shows a much more stable β form than the other scenar- ios: a transition temperature around 600 K, compared to To further explore crystallization of sPS, we consider 430 450 K otherwise. In the CG simulations, the transi- the different crystal forms observed experimentally. The tion− temperature of the β form is around 430 K, roughly crystalline polymorphs studied in this paper are listed in 20 K lower than the α form. Experimentally, crystalliza- Tab. I. We perform atomistic and coarse-grained replica tion of a melt leads preferentially to the α (β) crystals exchange simulations from these different crystal struc- when cooled from low (high) temperatures56. The AA tures. Fig. 5 shows that three different initial configu- model is thus in qualitative agreement with the preferen- rations lead to different thermodynamics because of size tial behavior for β found experimentally. effects and box shapes. In the following we analyze the To further study the β form in CG simulations, we results in detail. compare β configurations that are representative of our CG-RE simulations with the β00 form observed experi- mentally17. Fig. 8 (b) shows there are two kinds of bi- 1. α form layers, indicated as A(/) and B( ). The two bilayers are characterized by different orientations\ of the lines con- Fig. 5 (a) shows the initial α form configuration, which necting two adjacent phenyl rings inside each chain. The contains 9 chains of 10 monomers each. RE simulations regular succession of bilayers ABAB gives rise to the were performed for both the CG and AA models for ordered β00 form, also showing that these two orienta- 1.2 106 τ and 300 ns, respectively. Fig. 5 (a-1) and tions are not parallel to the layer lines. On the other (a-2)× shows the polymorph distributions in the tempera- hand, CG configurations have bilayers parallel to the ture range 250 550 K. At both resolutions in this box layer lines(Fig. 8 (c)). geometry, the α−form is not only stable but also predom- The densities obtained from AA and CG simulations inant, with a transition temperature around 460 K. are listed in Tab. II, compared with experimental data. Fig. 6 (a) shows the initial CG configuration, set up in While the temperature dependence shown in Fig. 8 (a) 6

TABLE I. Main crystal structures of s-PS studied in this work, along with the number of crystal units, number of chains and number of monomers for each chain in the MD simulation. ◦ Crystal a (A)˚ b (A)˚ c (A)˚ γ ( ) Space group Conformation Nunit Nchain Nmon α14 25.82 26.26 5.03 119.9 P 3 TTTT 1 1 5 9 10 × 17 × × × β 8.79 28.61 5.04 90.0 P 212121 TTTT 3 1 5 12 10 20 × × × δ 17.38 11.73 7.81 115.0 P 21/a (T T GG)2 3 2 3 12 12 × × × ↵

(a) (b) (c) 1 CG-RE (a-1) (b-1) (c-1) ↵ form 0.8 form defect amorphous 0.6 form

0.4 Probability 0.2

0

1 AA-RE (a-2) (b-2) (c-2) 0.8

0.6

0.4 Probability

0.2

0 200 300 400 500 600 700 200 300 400 500 600 700 200 300 400 500 600 700 T [K] T [K] T [K]

FIG. 5. From top to bottom: Initial crystal structures for (a) α, (b) β and (c) δ polymorphs; distributions of polymorphs as a function of temperature obtained from CG and AA REMD simulations. resulted from the average effect of different phases, here limited modeling of π-stacking interactions between side we discriminate between them. The results show that the chains, as well as the packing of phenyl hydrogens (e.g., AA simulations reproduce the experimental values satist- T-like stacking). This structural deficiency, leading to factorily. The CG results are somewhat larger: While the reduced stabilization of the β form, may well explain the α form in CG simulations remains in reasonable agree- transition-temperature offset. ment with both references, the β form is markedly higher. These results hint at structural deficiencies for the β form in the CG model, both in terms of limited pack- ing orientations (8) and higher density. We hint at the 7

1 Triplet A ↵AA ↵CG 0.8 AA CG 0.6

0.4

Triplet B Triplet B Order Parameter (a) 0.2

0 Triplet A (a) 200 300 400 500 600 700 800 T [K]

B( ) A( ) B( ) A( ) Triplet A Triplet A (b) B( ) (b) (c) FIG. 6. (a) The initial α00 configuration. (b) A representative 0 CG-RE snapshot sampled at 250 K, belonging to the α form. FIG. 8. (a) Structural order parameter as a function of tem- perature obtained from CG and AA REMD simulations. (b) The initial β00 configuration. (c) Representative snapshot ob- tained from CG-RE simulations at 250 K. (a) CG-RE at 250 K 1 ↵ form TABLE II. Density of s-PS polymorphs (g/cm3). form 0.8 defect Crystal experiment AA simulation CG simulation full or full ↵ amorphous α 1.034 1.075 1.098 form β 1.08 1.078 1.208 0.6 δ 0.977 - - mix 0.4 Probability 3. δ form 0.2 Fig. 5(c) shows the initial δ form configuration, con- 0 1800 1850 1900 1950 2000 taining 12 chains with 12 monomers each. RE sim- 3 ulations at the CG and AA leve were performed for Time [10 ⌧] 1.6 106 τ and 100 ns, respectively. Fig. 5 (c-1) shows (b) Full (c) Full ↵ (d) Mix that× the δ form is not stable at low temperatures at the CG level—α and β form coexist instead. This phe- nomenon is consistent with experimental studies, since we are not modeling the cosolvent necessary to stabilize the δ form. At 250 K, the system favors the α form over β, possibly because the geometry of the unit cell is more congruent with the former. At the AA level, we observed no crystallization (Fig. 5 (c-2)). While we do expect the system to eventually crys- tallize, we suppose that our simulation times–even using FIG. 7. (a) Distributions of polymorphs in the last 2 × 105 τ RE—did not allow to probe long-enough timescales to of the CG-RE simulations at 250 K. (b) Full β, (c) full α, (d) reach any reasonable equilibrium. While AA simulations and mixture at 250K. in the α and β geometries did stabilize crystals, we ob- served little dynamical exchange among them, further 8 suggesting the sampling challenge of polymer crystals at 1 ↵ form an AA resolution. form 0.8 defect amorphous form 0.6 4. Melt structure ↵ 0.4 Probability At high temperatures the system consists of amor- mix 0.2 phous, defects, and a minor population of δ-type con- 0 figurations. Note that δ-type phases here refer to the (a) (b) 200 300 400 500 600 presence of helical chain segments, while defect corre- T [K] spond to chain segments that are planar but falling out 1 0.3 of the α and β forms. A comparison of the melt popu- 0.25 250.00K 0.8 308.36K lations between the two resolutions shows the larger oc- 364.63K 0.2 currence of planar conformations (i.e., defect) at the CG 0.6 383.88K 404.40K level. On the other hand, amorphous and helical confor- 0.15 464.55K 0.4 mations have a somewhat higher population at the AA Probability 0.1 Order Parameter 0.2 level. This indicates that the CG model of sPS tends to 12 chains 0.05 favor planar arrangements. Here again, the packing of 96 chains 0 0 the phenyl rings may play a role, in accordance with the (c) 200 300 400 500 600 (d) 00.20.40.60.81 high density found for β. T [K] Order Parameter

FIG. 9. CG simulations: (a) Representative snapshot made C. Toward larger systems from 96 chains at 250K; (b) CG Distributions of polymorphs as a function of temperature; (c) Structural-order parameter as a function of temperature obtained (comparison between In the previous sections, we demonstrated that the 12 chains and 96 chains); and (d) Structural order parameter thermodynamic-based CG model applied here can crys- distributions at different temperatures. tallize and stabilize major polymorphs. Given the sam- pling difficulties already met at the AA level, we con- sider larger systems only with the CG model: 96 chains observed here for the larger system. Experimental re- of 10 monomers each. The CG-RE simulations ran for sults have suggested that formation of the α form may 7 105 τ. be a kinetically-controlled process.26 Our results, while ×Fig. 9 (a) shows one snapshot starting from the β form, devoid of non-equilibrium aspects, hint at a templating where α, β and the mixed phases exist simultaneously in mechanism that favors the formation of α due to the en- the crystalline state. Full α or β forms, observed in the vironment. small systems, are not found here. At low temperatures, the proportion of α is a bit higher than β (Fig. 9 (b)). It is worth mentioning that we also performed another simu- lation starting from the α form (72 chains), and obtained IV. CONCLUSION virtually the same proportion of α and β forms. These converging results thus indicate that they do not depend We report the first computational studies of polymer- on the chosen box geometry, unlike the abovementioned crystal polymorphism. Replica exchange molecular dy- findings on the smaller systems. Unfortunately, this re- namics simulations are performed using atomistic (AA) sult is not consistent with experiments, which associates and coarse-grained (CG) models to model the ther- a higher stability to the β form. This stands as another mal crystallization of syndiotactic polystyrene (sPS). Re- evidence of the structural deficiency of the β form in this markably, the thermodynamic-based CG model not only CG model. reproduces melt properties, but also transfers to the crys- Fig. 9(c) shows that, compared with the small system, tals. Furthermore, we find that the CG model stabilizes the transition temperature of the large system is a bit the main polymorphs α and β, while the δ form was not lower: 400 K. Fig. 9(d) shows the order-parameter dis- observed due to the lack of solvent. These results are tributions at different temperatures, which show more in qualitative agreement with experiments. Our simula- structural heterogeneity than found earlier on (Fig. 4 tions suggest the role of templating mechanisms to ra- (c)). We conclude that the larger system self-assembles tionalize the experimentally-observed kinetic control of more diverse crystalline phases that dynamically inter- the α form. Because the CG model markedly speeds convert, leading to an overall reduction in the transition up the simulations, it can be used to simulate signifi- temperature compared to the small systems. cantly larger systems, better suited to study crystalliza- Interestingly, while the α form was predominantly sam- tion without finite-size effects. We thus show that CG pled when simulating a smaller box congruent with the models are powerful tools to investigate the polymorphic α unit cell (Fig. 5), this preferential stabilization was not behavior of polymers. The choices of mapping scheme 9 and force field are important to stabilize and distinguish 17Chatani Y, Shimane Y, Ijitsu T, et al. Structure study on syndio- the main polymorphs. In spite of remaining shortcom- tactic polystyrene: 3. Crystal structure of planar form I. Polymer, ings, the ability of the CG model (without additional 1993, 34(8): 1625-1629. 18Rizzo P, Lamberti M, Albunia A R, et al. Crystalline orienta- tuning) to reproduce the crystallization transition and tion in syndiotactic polystyrene cast films. Macromolecules, 2002, polymorphism of polymer is remarkable. We expect CG 35(15): 5854-5860. simulations to become a major tool for polymer-crystal 19Rizzo P, Albunia A R, Guerra G. Polymorphism of syndiotactic polymorphism studies. polystyrene: γ phase crystallization induced by bulky non-guest solvents. Polymer, 2005, 46: 9549-9554. 20De Rosa C, Guerra G, Petraccone V, et al. Crystal structure of the emptied clathrate form (δe form) of syndiotactic polystyrene. V. ACKNOWLEDGMENT Macromolecules, 1997, 30(14): 4147-4152. 21Milano G, Venditto V, Guerra G, et al. Shape and volume of cavities in thermoplastic molecular sieves based on syndiotactic We thank Christine Peter, Omar Valsson, and Clau- polystyrene. Chemistry of Materials, 2001, 13(5): 1506-1511. 22 dio Perego for critical discussions and reading of the Gowd E B, Shibayama N, Tashiro K. Structural changes in ther- mally induced phase transitions of uniaxially oriented δe form of manuscript. T.B. acknowledges funding from the Emmy syndiotactic polystyrene investigated by temperature-dependent Noether program of the Deutsche Forchungsgemeinschaft measurements of X-ray fiber diagrams and polarized infrared (DFG). spectra. Macromolecules, 2006, 39(24): 8412-8418. 23Rizzo P, D’Aniello C, De Girolamo Del Mauro A, et al. Thermal

1 transitions of ε crystalline phases of syndiotactic polystyrene. Bernstein J. Polymorphism in molecular crystals. Oxford Univer- Macromolecules, 2007, 40(26): 9470-9474. sity Press, 2002. 24 2 Petraccone V, Ruiz de Ballesteros O, Tarallo O, et al. Perego C, Salvalaglio M, Parrinello M. Molecular dynamics sim- Nanoporous polymer crystals with cavities and channels. Chem- ulations of solutions at constant chemical potential. The Journal istry of Materials, 2008, 20(11): 3663-3668. of Chemical Physics, 2015, 142(14): 144113. 25 3 Tarallo O, Schiavone M M, Petraccone V, et al. Channel clathrate Radu M, Kremer K. Enhanced crystal growth in binary Lennard- of syndiotactic polystyrene with p-nitroaniline. Macromolecules, Jones mixtures. Physical Review Letters, 2017, 118(5): 055702. 4 2010, 43(3): 1455-1466. Beran G J O. Modeling polymorphic molecular crystals with elec- 26Guerra G, Vitagliano V M, De Rosa C, et al. Polymorphism tronic structure theory. Chemical Reviews, 2016, 116(9): 5567- in melt crystallized syndiotactic polystyrene samples. Macro- 5613. 5 molecules, 1990, 23(5): 1539-1544. Price S L. Predicting crystal structures of organic compounds. 27Tamai Y, Fukuda M. Nanoscale molecular cavity in crystalline Chemical Society Reviews, 2014, 43(7): 2098-2111. 6 polymer membranes studied by molecular dynamics simulation. Marom N, DiStasio Jr. R A, Atalla V, et al. Many-Body dis- Polymer, 2003, 44: 3279-3289. persion interactions in molecular crystal polymorphism. Ange- 28Milano G, Guerra G, Mueller-Plathe F. Anisotropic diffusion wandte Chemie, 2013, 52: 6629-6632. 7 of small penetrants in the δ crystalline phase of syndiotactic Reilly A M, Tkatchenko A. Role of dispersion interactions in the polystyrene: A molecular dynamics simulation study. Chemistry polymorphism and entropic stabilization of the Aspirin crystal. of Materials, 2002, 14(7): 2977-2982. Physical Review Letters, 2014, 113(5): 055701. 29 8 Tamai Y, Fukuda M. Fast one-dimensional gas transport in Hu W, Frenkel D. Polymer crystallization driven by anisotropic molecular capillary embedded in polymer crystal. Chemical interactions. Interphases and Mesophases in Polymer Crystalliza- Physics Letters, 2003, 371(1): 217-222. tion III, 2005, 191: 1-35. 30 9 Tamai Y, Fukuda M. Reorientational dynamics of aromatic Sommer J U, Luo C. Molecular dynamics simulations of semicrys- molecules clathrated in δ form of crystalline syndiotactic talline polymers: crystallization, melting, and reorganization. polystyrene. Chemical Physics Letters, 2003, 371(1): 620-625. Journal of Polymer Science Part B: Polymer Physics, 2010, 31Tamai Y, Tsujita Y, Fukuda M. Reorientational relaxation of 48(21): 2222-2232. 10 aromatic molecules in molecular cavity of crystalline syndiotactic Yamamoto T. Molecular dynamics of polymer crystallization re- polystyrene studied by molecular dynamics simulation. Chemical visited: Crystallization from the melt and the glass in longer Physics Letters, 2005, 739(1): 33-40. polyethylene. The Journal of Chemical Physics, 2013, 139(5): 32Figueroa-Gerstenmaier S, Daniel C, Milano G, et al. Storage of 054903. 11 hydrogen as a guest of a nanoporous polymeric crystalline phase. Welch P M. Examining the role of fluctuations in the early stages Physical Chemistry Chemical Physics, 2010, 12(1): 5369-5374. of homogenous polymer crystallization with simulation and sta- 33Figueroa-Gerstenmaier S, Daniel C, Milano G, et al. Hydro- tistical learning. The Journal of Chemical Physics, 2017, 146(4): gen adsorption by δ and ε crystalline phases of syndiotactic 044901. 12 polystyrene aerogels. Macromolecules, 2010, 43(20): 8594-8601. De Rosa C, Guerra G, Petraccone V, et al. Crystal structure of 34Sanguigno L, Cosentino F, Larobina D, et al. Molecular simula- the α-form of syndiotactic polystyrene. Polymer Journal, 1991, tion of carbon dioxide sorption in nanoporous crystalline phase of 23(12): 1435-1442. 13 syndiotactic polystyrene. Soft Materials, 2011, 9(2-3): 169-182. Corradini P, De Rosa C, Guerra G, et al. Conformational and 35Yamamoto T. Molecular dynamics of crystallization in a heli- packing energy of the crystalline α modification of syndiotac- cal polymer isotactic polypropylene from the oriented amorphous tic polystyrene. European Polymer Journal, 1994, 30(10): 1173- state. Macromolecules, 2014, 47: 3192-3202. 1177. 36 14 Strobl G. The physics of polymers: Concepts for Understanding De Rosa C. Crystal structure of the trigonal modification (α Their Structures and Behavior, 3rd ed. Springer: Berlin, 2007. form) of syndiotactic polystyrene. Macromolecules, 1996, 29(26): 37Mueller-Plathe F. Coarse-graining in polymer simulation: from 8460-8465. 15 00 the atomistic to the mesoscopic scale and back. ChemPhysChem, Cartier L, Okihara T, Lotz B. The α ”superstructure” of syn- 2002, 3(9): 754-769. diotactic polystyrene: A frustrated structure. Macromolecules, 38Voth G A. Coarse-graining of condensed phase and biomolecular 1998, 31(10): 3303-3310. 16 systems. CRC press, 2008. De Rosa C, Rapacciuolo M, Guerra G, et al. On the crystal struc- 39Peter C, Kremer K. Multiscale simulation of soft matter systems. ture of the orthorhombic form of syndiotactic polystyrene. Poly- Faraday discussions, 2010, 144: 9-24. mer, 1992, 33(7): 1423-1428. 10

40Milano G, Mueller-Plathe F. Mapping atomistic simulations to trostatics – A proof of concept study on weakly polar organic mesoscopic models: a systematic coarse-graining procedure for molecules. The Journal of Chemical Theory and Computation, vinyl polymer chains. The Journal of Physical Chemistry B, 2005, 2017, 13: 6158-6166. 109(39): 18609-18619. 49Mueller-Plathe F. Local structure and dynamics in solvent- 41Spyriouni T, Tzoumanekas C, Theodorou D, et al. Coarse- swollen polymers. Macromolecules, 1996, 29(13): 4782-4791. grained and reverse-mapped united-atom simulations of long- 50Hess B, Bekker H, Berendsen H J C, et al. LINCS: A Linear Con- chain atactic polystyrene melts: structure, thermodynamic prop- straint Solver for Molecular Simulations. The Journal of Compu- erties, chain conformation, and entanglements. Macromolecules, tational Chemistry, 1997, 18(12): 1463-1472. 2007, 40(10): 3876-3885. 51Sugita Y, Okamoto Y. Replica-exchange molecular dynamics 42Qian H, Carbone P, Chen X, et al. Temperature-transferable method for protein folding. Chemical Physics Letters, 1999, coarse-grained potentials for ethylbenzene, polystyrene, and their 314(1): 141-151. mixtures. Macromolecules, 2008, 41(24): 9919-9929. 52Hess B, Kutzner C, van der Spoel D, et al. GROMACS 4: Algo- 43Sun Q, Faller R. Crossover from unentangled to entangled rithms for highly efficient, load-balanced, and scalable molecular dynamics in a systematically coarse-grained polystyrene melt. simulation. The Journal of Chemical Theory and Computation, Macromolecules, 2006, 39(2): 812-820. 2008, 4(3): 435-447. 44Harmandaris V A, Adhikari N P, van der Vegt N F A, et al. 53Bussi G, Donadio D, Parrinello M. Canonical sampling through Hierarchical modeling of polystyrene: from atomistic to coarse- velocity rescaling. The Journal of Chemical Physics, 2007, 126(1): grained simulations. Macromolecules, 2006, 39(19): 6708-6719. 014101. 45Harmandaris V A, Reith D, van der Vegt N F A, et al. Com- 54Berendsen H J C, Postma J P M, van Gunsteren W F, et al. parison between coarse-graining models for polymer systems: Molecular dynamics with coupling to an external bath. The Jour- two mapping schemes for polystyrene. Macromolecular Chem- nal of Chemical Physics, 1984, 81(8): 3684-3690. istry and Physics, 2007, 208: 2109-2120. 55Liu C, Muthukumar M. Langevin dynamics simulations of early- 46Fritz D, Harmandaris V A, Kremer K, et al. Coarse-grained stage polymer nucleation and crystallization. The Journal of polymer melts based on isolated atomistic chains: simulation of Chemical Physics, 1998, 109(6): 2536-2542. polystyrene of different tacticities. Macromolecules, 2009, 42(19): 56Woo E M, Sun Y S, Yang C P. Polymorphism, thermal behavior, 7579-7588. and crystal stability in syndiotactic polystyrene vs. its miscible 47Brini E, Marcon V, van der Vegt N F A. Conditional reversible blends. Progress in Polymer Science, 2001, 26: 945-983. work method for molecular coarse graining applications. Physical Chemistry Chemical Physics, 2011, 13(1): 10468-10474. 48Diechmann G, van der Vegt N F A. Conditional reversible work coarse-graining models of molecular liquids with coulomb elec-