AI Unlocks Molecular Mystery Behind Water's Unusual Behavior Below Freezing

Researchers at Osaka University have harnessed artificial intelligence to solve a longstanding scientific puzzle about water's unique physical behavior, especially in the "supercooled" state where water remains liquid below 0 degrees Celsius. Water, the most abundant substance on Earth, behaves differently from most liquids by expanding upon freezing rather than contracting, a phenomenon that has puzzled scientists for years.

The team used a machine learning model to analyze complex simulations of supercooled water, where water molecules exist in two competing structural forms: high-density liquid (HDL) and low-density liquid (LDL). Previously, inconsistent methods for describing these molecular arrangements hindered comparative analysis. By training a neural network through trial and error, the AI identified significant molecular patterns and evaluated 16 different structural description methods to determine the most accurate and effective ones.

This breakthrough, published in Communications Chemistry, establishes a unified framework for comparing microscopic water structures, enabling better understanding of how molecular changes influence water's macroscopic properties. Beyond explaining water's anomalous expansion and other behaviors, the AI-developed tools could advance future technologies involving water manipulation and deepen insights into complex natural materials.

With this AI-generated structural "map," researchers are closer than ever to rewriting fundamental physics textbooks regarding water's properties.