Glass Transition and Structure of Organic Polymers from All-Atom Molecular Simulations

Molecular dynamics simulations of polymer samples with all-atom resolution provide important insight into the relationship between the atom-level structure and macroscopic properties of polymers. The computational setup of molecular simulations in such a case deserves to be validated, paying attention not to overlook various aspects or inferior settings or postprocessing analyses that have the potential to distort the simulation outcome or at least to make the simulated ensemble too incompatible with its experimental counterparts, such as their polydispersity, initial conformation, or thermal history of the samples. The accuracy of the simulation results obtained from existing all-atom nonpolarizable force fields for three selected polymers is independently benchmarked with respect to the polymer densities and glass transition temperatures. Errors of structural or thermodynamic properties arising due to insufficient sample equilibration or inadequate simulation setup are quantified. Special attention is paid to the selection of reference literature data for polymer systems that are well characterized and as similar as possible to the computationally treated samples. Very different performances of predictions of the glass transition temperatures occur among the individual target polymers, with both their sampling uncertainty and errors from reference experimental data ranging from acceptable below 10 K to highly unsatisfactory 100 K in individual cases. The accuracy of the predicted glass transition temperature is found to be higher for polymers exhibiting faster internal dynamics and distinct trend shifts between the glass and the liquid. On the contrary, when the glass transition occurs gradually over a wider temperature range, it becomes very difficult to capture the adequate transition temperature within molecular simulations, regardless of the evaluation protocol used. Bulk density proves to be the most reliable observable for subsequent trend shift analyses, which typically yield similar results regardless of performing equilibrium or nonequilibrium simulations and adopting the bilinear or hyperbolic regression analyses.