2025/06/26 by Małgorzata J. Zimoń, Fausto Martelli, Zimoń, Małgorzata J. +1
Computer Science · Materials Science · #Chemical Physics (physics.chem-ph) #Computational Physics (physics.comp-ph) #FOS: Physical sciences #Machine Learning in Materials Science #Neural Networks and Reservoir Computing #Topological and Geometric Data Analysis
paper · pdf · doi:10.48550/arxiv.2506.21821
openalex publication_date 2025/06/26 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Thermodynamics, introduced over two centuries ago, remains foundational to our understanding of physical, chemical, biological, and engineering systems. Its principles are traditionally grounded in the statistical mechanics framework, which explains macroscopic behavior from microscopic states. In this work, we propose an alternative approach that interprets thermodynamic behavior through the lenses of time series analysis, an approach commonly used in other fields, including finance, climate, and signal processing. We perform classical molecular dynamics simulations of liquid water, the most complex, anomalous, and important substance known, over a wide range of its phase diagram. By examining the temporal evolution of the hydrogen bond network (HBN) topology, we demonstrate that the dynamics of microscopic topological motifs populating the HBN encode the system's macroscopic thermodynamic behavior. Furthermore, our approach enables the prediction of thermodynamic properties in regions beyond those directly sampled in our simulations. We achieve this result by leveraging artificial intelligence to uncover patterns in temporally resolved data that are often lost through conventional averaging. This work offers new insights into the fundamental behavior of water and network-forming materials more broadly, establishing a new paradigm for understanding material properties beyond the classical confines of statistical mechanics.