(692b) Clustering and Measurement Methods with Molecular Dynamics Data for Accurate Identification and Tracking of Polymer Melt Crystallization

The self-assembly of complex crystal structures within an entangled polymeric system represents a compelling but incompletely understood phase behavior. Molecular Dynamics (MD) simulations provide a powerful tool to observe this phenomenon at the granular time and length scales, shedding light on molecular evolution across different temporal phases and deepening our understanding of the microkinetics of crystal growth in real systems. However, acknowledging the spatially and temporally layered complexity of the crystallization process, which unfolds across various levels, necessitates the arduous task of scrutinizing MD-driven data. Building on previous initiatives by the MRAIL group at the University of Tennessee, we have introduced thermodynamic-like variables of local entropy and enthalpy at an atomistic level to delineate and quantify phase phenomena in MD simulations of polymer flows, notably flow-enhanced nucleation and flow-induced crystallization events. By capturing local ordering and energetics at the monomer level rather than globally, these variables offer a more precise approach for detecting and quantifying flow-enhanced nucleation at small spatiotemporal scales, which precede flow-induced crystallization.

The current study demonstrates the accuracy of employing these thermodynamic-like variables to identify crystalline regions, coupled with an original robust and interpretable clustering technique, as a precise and insightful method to investigate system crystallinity over time. Our novel clustering methodology integrates thermodynamic-like variables with a de novo diffusion imputation approach within a Connected Component Analysis (CCA) framework. We validated findings from prior research using quiescent MD simulation data for polyethylene chains comprising 150 united atoms at different quenching temperatures. Our goal was to create a compilation of a set of methods and algorithms used for measuring polymer crystallization alongside standard unsupervised machine learning clustering algorithms. This includes a detailed explanation of the biases and best practices for applying them and a thorough sensitivity/error analysis for each procedure. This innovative approach sets a new benchmark in the field by enabling the precise identification and dynamic tracking of crystalline structures. It showcases unparalleled efficiency and interpretability compared to traditional density-based clustering methods like DBSCAN or graph-based clustering methods like spectral clustering, which may fall short in capturing the intricate patterns of polymer crystallization. Our research underscores the importance of choosing appropriate data inputs (e.g., x, y, z, entropy-like parameter, order parameters) for clustering algorithms, influencing volume, surface, and convexity estimates of identified clusters. Our study has made meaningful strides in elucidating crystal growth mechanisms and quantitatively assessing the distribution and dynamics of crystallization within polymers by leveraging MD-generated data with an effective clustering technique. The results demonstrate significant progress toward our understanding of the underlying processes that govern crystal growth in polymers. This study examines the dynamics of polymer crystallization and provides a foundation for future research into clustering behaviors in MD simulation data.