Researchers at the University of Osaka have used artificial intelligence to gain new insights into the unusual behavior of supercooled water—liquid water that remains unfrozen below its normal freezing point. Water has long puzzled scientists because it exhibits unique properties unlike most other liquids, and understanding its molecular structure has remained a major scientific challenge. By applying AI, the researchers developed a more effective way to analyze these complex molecular patterns.
The research focused on comparing 16 different structural descriptors that scientists use to describe how water molecules are arranged. The AI model was trained on molecular dynamics simulations and learned to identify which descriptors most accurately distinguish between water's two competing liquid forms: high-density liquid (HDL) and low-density liquid (LDL). This provides a standardized framework for evaluating water's microscopic structure, something researchers have struggled to achieve for years.
According to the researchers, the neural network evaluated the descriptors in a way that mimics human pattern recognition, enabling it to identify the most informative measurements of water's molecular organization. This breakthrough improves scientists' understanding of how subtle structural changes influence water's unusual thermodynamic properties, particularly under extremely cold conditions where its behavior becomes even more complex.
The findings, published in Communications Chemistry, could have broad implications for chemistry, materials science, and climate research. A better understanding of water's molecular behavior may lead to improved computational models and provide new insights into natural processes where supercooled water plays a critical role, including cloud formation, ice nucleation, and biological systems. The study also demonstrates how AI is becoming an increasingly valuable tool for solving longstanding scientific problems that are difficult to analyze using conventional methods alone.
