AI Decodes Hidden Molecular Structure of Supercooled Water

Researchers at the University of Osaka have utilized artificial intelligence to decode the complex molecular structure of supercooled water, a state where water remains liquid below its freezing point. By employing a neural network to evaluate 16 different structural descriptors, the team successfully identified the most effective methods for characterizing the transition between high-density liquid (HDL) and low-density liquid (LDL) states. The study, published on July 6, 2026, in Communications Chemistry, addresses the long-standing scientific challenge of measuring water's microscopic structural changes. The authors, including Kohei Yoshikawa and Kang Kim, demonstrate that AI can effectively classify structural data from molecular dynamics simulations, providing a consistent framework to compare previously disparate measurement scales. This research provides a critical advancement in understanding water's anomalous thermodynamic properties, such as its expansion upon freezing. By determining which descriptors best capture structural information, the framework offers a new tool for scientists to map the relationship between microscopic molecular configurations and macroscopic physical behavior. The findings offer significant practical value for materials science and chemistry, enabling more precise modeling of water-based systems. Future applications include leveraging this AI-driven methodology to explore other complex substances, potentially unlocking further secrets regarding the fundamental nature of liquid states and phase transitions.