Data-driven identification and analysis of the glass transition in polymer melts

We propose a data-driven approach based on information about structural fluctuations of polymer chains, which clearly identifies the glass transition temperature $T_g$ of polymer melts of weakly semiflexible chains. We use principal component analysis (PCA) with clustering to distinguish between liquid and glassy states and predict $T_g$ in the asymptotic limit. Our method indicates that for temperatures approaching $T_g$ from above it is sufficient to consider short molecular dynamics simulation trajectories, which just reach into the Rouse-like monomer displacement regime. The first eigenvalue of PCA and participation ratio show sharp changes around $T_g$. Our approach requires minimum user inputs and is robust and transferable.