Rapid, Accurate and Reproducible Prediction of the Glass Transition Temperature Using Ensemble-Based Molecular Dynamics Simulation

For the computational design of new polymeric materials, accurate methods for determining the glass transition temperature (<i>T</i>_g) are required. We apply an ensemble approach in molecular dynamics (MD) and examine its predictions of <i>T</i>_g and their associated uncertainty. We separate the uncertainty into the aleatoric contributions arising from dynamical chaos and that due to the computational scenarios chosen to compute <i>T</i>_g. We propose a new scenario for computing <i>T</i>_g, where the density-temperature behavior is computed by running all temperatures concurrently, rather than invoking a sequential approach, thereby significantly reducing wall-clock time from days to several hours without increasing the aleatoric uncertainty. On comparing concurrent and sequential scenarios on six highly cross-linked epoxy resins cured with aromatic amines, we find excellent agreement with our experimentally determined <i>T</i>_g using dynamical mechanical analysis for both scenarios. The confidence intervals are found to scale as <i>N</i>^-0.5, where <i>N</i> is the number of members in the ensemble, implying that ensembles comprised of at least ten replicas are required to predict <i>T</i>_g using MD with 95% confidence intervals of less than 20 K. The optimal MD simulation protocol is 4 ns of burn-in time followed by 2 ns of production run time.