강연제목: Machine Learning Meets Statistical Mechanics: Exploring Phase Behavior of Mixtures Through Local Affinity

초록: Sudden changes in the states of matter, including phase and structural transitions, are fundamental phenomena observed across physical, chemical, and biological systems, yet their underlying mechanisms remain only partially understood, particularly in complex, multiscale, or far-from-equilibrium environments. This work focuses on addressing these challenges through a combination of computational modeling and unsupervised machine learning (UML). I aim to develop new methodologies capable of analyzing liquid-liquid phase separation (LLPS) in complex mixtures such as biocondensates and polyelectrolyte solutions, where conventional theories often fall short. By implementing UML, I construct data-driven phase diagrams and introduce a novel order parameter, local affinity, which encodes phase information as binary vectors to capture emergent behavior at the microscopic level. It is revealed that the method successfully reproduces accurate and precise phase diagrams of mixtures in any types of systems. This bottom-up approach maintains physical interpretability while leveraging machine learning pattern recognition capabilities, offering a framework to understand transitions in soft matter, biological systems, and energy materials.