For decades, chemists have puzzled over water's quirks. Why does it become easier to compress as it cools? Why does it reach its maximum density at 4°C rather than at its freezing point? Now, a team of researchers has captured something that previous scientists could only theorize about: direct molecular-level evidence that liquid water is not a uniform substance, but a constantly shifting mixture of two distinct microscopic structures.
The study, published in Nature Physics, marks a breakthrough in a debate that has simmered in the scientific community for years. Professor Xiao Cheng Zeng from City University of Hong Kong, the corresponding author, has been pursuing this question since his graduate school days. "I started theoretical research on freezing of liquids when I was a postdoc, but I was always hoping to study freezing of water one day," he told Phys.org. "Since then, I have been particularly interested in the topic of liquid-liquid transition in water."
The two-state model has long suggested that liquid water contains two interconvertible structures: a denser, more disordered form and a less dense, more ordered one. But the model remained controversial because no one had seen direct molecular evidence of these structures—neither a "pure A" nor "pure B" liquid water had ever been observed.
The breakthrough came through artificial intelligence. Zeng and his team trained an unsupervised deep learning autoencoder on approximately 74 million local water-molecule configurations drawn from molecular dynamics simulations using the TIP4P/Ice model, a widely trusted computational model of water. About 17 percent of that training data came from the liquid-liquid phase transition region, where water exists in a deeply supercooled state.
Traditional methods, which measured density and energy differences between molecules, had failed to cleanly separate the two structures. What was needed was a way to let the data reveal its own hidden patterns—without human assumptions about what those patterns should look like. "We need AI's help to learn and uncover these hidden physical characteristics," Zeng said.
The results offer a new lens for understanding not just water, but the fundamental phase transitions that shape the physical world. And for Zeng, who spent decades waiting for this moment, the discovery fulfills a long-held scientific dream.