Machine Learning Clustering of Chain Conformations Resolves Glass Transition Temperature (Tg) Shifts in Polymer Thin Films

Polymer thin films exhibit pronounced deviations from bulk glass-transition behavior, most notably a reduction in T_g that grows as the film becomes thinner. Understanding this requires T_g estimates resolved across distinct regions of the film, yet conventional bilinear fitting pins T_g to an arbitrary crossover of two linear regimes, an ambiguity that worsens for local, data-limited regions. To overcome it, we introduce a machine learning approach that estimates T_g directly from chain conformations. Applying the density-based clustering algorithm (DBSCAN) to temperature-dependent radius of gyration (R_g) or Rouse mode (X_p) data, we detect a sharp transition between distinct conformational states, marking the onset of the glass transition. By eliminating fitting-induced ambiguity, the method delivers robust T_g estimates even for localized regions, enabling spatially resolved analysis of confinement effects. Our approach provides mechanistic insight into how enhanced surface mobility propagates into the film interior, explaining the thickness-dependent T_g suppression characteristic of polymer thin films.