Researchers at Osaka University have used artificial intelligence to better understand the unusual behavior of supercooled water. The study aims to explain how water changes at the microscopic level when cooled below its freezing point without turning into ice.
Scientists published the findings in the journal Communications Chemistry. The research focuses on water’s unique structural behavior and how artificial intelligence can help identify patterns that traditional methods struggle to analyze.
Scientists study water below freezing point
Water remains one of the most unusual liquids on Earth because it expands when it freezes. Scientists believe these rare properties come from changes in water’s microscopic structure caused by temperature and pressure.
Normally, water molecules form an ordered crystal structure when freezing into ice. This process usually begins around impurities or scratches inside a container that act as nucleation points for ice formation.
However, in smooth and clean containers, water can remain liquid even below its freezing point. Scientists call this process supercooling.
During supercooling, water shows stronger signs of its unusual behavior. Researchers explain this through two competing liquid states known as high density liquid and low density liquid.
At the molecular level, hydrogen bonds constantly shift and reshape the structure of water. Warmer temperatures increase the presence of high density liquid structures, while colder conditions favor low density liquid arrangements.
AI compares water structure patterns
To understand these structural changes, scientists previously developed several measurement systems called structural descriptors. These descriptors help researchers study local water structures and molecular arrangements.
However, the different methods often use separate scales and formats, making comparisons difficult.
The Osaka University research team used artificial intelligence to evaluate and compare 16 different structural descriptors within one unified framework.
Researchers trained a neural network model using molecular simulation data from supercooled water. The AI system learned through repeated pattern analysis and identified how effectively each descriptor distinguished between different water states.
Lead researcher Kang Kim said machine learning has already shown strong potential in analyzing complex structural data. The team wanted to test whether AI could evaluate water structures in a way similar to human understanding.
Senior researcher Nobuyuki Matubayasi explained that the AI system compared low density and high density liquid structures across different temperatures to identify the most effective descriptors.
Study may improve understanding of water behavior
The findings could help scientists better understand the connection between water’s microscopic structure and its physical properties.
Also read Emmad Irfani Celebrates 13 Years in Acting With Emotional Message
Researchers believe the study may explain why water behaves differently from most liquids and could improve future scientific models used to analyze liquids and phase transitions.
The research also highlights the growing role of artificial intelligence in scientific discovery. AI systems continue to help researchers process large amounts of complex data more efficiently across multiple fields.
Scientists hope the new framework will support future studies on water behavior, thermodynamics, and molecular science. The research may also contribute to advancements in chemistry, climate science, and material engineering.
