Water may appear perfectly uniform, yet new research suggests it could shift between two microscopic arrangements that constantly transform into one another. Using artificial intelligence and massive simulations, scientists at City University of Hong Kong found fresh evidence for a long-debated two-state model of liquid water.
The team analyzed about 74 million local molecular configurations generated with the TIP4P/Ice water model. Instead of telling the AI exactly what to search for, the researchers used an unsupervised deep-learning approach, allowing the system to detect patterns on its own.
What the AI revealed
The model separated the data into two distinct clusters. One matched a denser, more disordered arrangement, while the other reflected a less dense and more ordered structure. The system also mapped how water molecules move between these states, showing that the transition is more complex than a simple switch.
According to the study, the pathways change depending on temperature and pressure. In some conditions, the shift follows a simpler route; near the boundary where both liquid forms compete, the process becomes a more intricate loop with multiple transition states.
Researchers say this could help explain several unusual properties of water, including why it reaches its maximum density at 4°C and behaves unusually under pressure. The findings also strengthen the idea that liquid water may contain two interconverting local structures rather than one fixed microscopic form.
The work, published in Nature Physics, is still based on simulations, so the newly identified hidden variables now need experimental confirmation. If future studies verify them, the result could reshape how scientists understand water in biology, geology, and advanced materials research. It may open a clearer future view of one of nature's most essential liquids.