Chinese scientists unravel atomic structure of pre-melt layer in ice with AI

Researchers at Universidade of Pequim have revealed the atomic structure of the pre-melt layer in the ice using a combination of artificial intelligence and atomic force microscopy. The study, published in Physical Review X in December 2025, identified a solid amorphous layer at subzero temperatures. Essa discovery solves a riddle observed by Michael Faraday over 170 years ago.
The layer forms even in extreme cold conditions and gradually evolves into a near-liquid state as the temperature rises. The innovative technique overcomes limitations of traditional methods and allows complex three-dimensional structures to be reconstructed.
Researchers at Universidade of Pequim were able to map the ice surface at the atomic level by integrating atomic force microscopy (AFM) with machine learning algorithms. The phenomenon of pre-melt, a thin liquid-like layer on the surface of ice below 0°C, has intrigued scientists since observations of Michael Faraday in the 19th century.
The innovation made it possible to reconstruct three-dimensional structures from incomplete AFM data. The algorithm was trained with molecular dynamics simulations that included real noise from experiments.
- The technique detects minute variations in force with a sensitive microscopic tip.
- Realistic simulations helped interpret challenging signals on uneven surfaces.
- The method overcomes barriers of conventional techniques for disordered interfaces.
Historical origin of the enigma
Michael Faraday noticed more than 170 years ago that blocks of ice stick together even at subzero temperatures due to a moist surface layer. Esse pre-melt influences processes such as friction in ice, chemical reactions in the atmosphere and tissue preservation at low temperatures.
Despite its relevance to fields such as cryopreservation and atmospheric chemistry, the molecular structure of this layer remained unknown until now. Previous Estudos confirmed the existence of the layer, but did not reveal its precise atomic organization.

Methodology used in the study
Scientists used high-resolution AFM to scan ice surfaces at controlled temperatures. They then applied machine learning trained on thousands of molecular simulations.
Training included realistic variations to deal with experimental noise. Isso allowed predicting and reconstructing atomic details invisible in direct AFM images.
The experiments took place in temperature ranges between -152°C and -93°C. At Nesses levels, the surface showed gradual transition from solid crystalline to disordered state.
Identified amorphous layer
The main discovery was an amorphous surface layer, where water molecules lose their typical crystalline order, but maintain solid behavior. Essa layer exhibits strong topological disorder and solid-like dynamics.
As the temperature increases, it evolves into an almost liquid layer. The amorphous layer appears before the traditionally described quasiliquid layer.
This intermediate phase resets the pre-melt process. Ela explains how the ice surface gradually transforms under extreme conditions.
Details of the observed transition
At very low temperatures, ice maintains a hexagonal crystalline structure. Acima of certain thresholds, the amorphous layer is formed with disorganized molecules at the atomic level.
Thickness and disorder grow with warming. Próximo to the melting point, more fluid behavior appears.
- Intense topological disorder in the amorphous layer.
- Solid dynamics preserved despite loss of crystallinity.
- Continuous evolution to quasiliquid state.
- Impact on properties such as adhesion and friction.
Applications of the new technique
The machine learning framework for AFM opens doors to studying complex interfaces beyond ice. Ele allows you to investigate phase transitions and defects in materials with atomic precision.
Researchers highlight potential in areas such as catalysis, functional materials and biological systems. The approach can analyze ion solvation and biomolecular recognition.
The tool represents a significant advance in the characterization of disordered surfaces. Ela facilitates studies at previously inaccessible atomic scales.
Relevance to everyday phenomena
Pre-melting explains why ice is slippery even at subzero temperatures, like when skating. The surface layer reduces friction by acting as a natural lubricant.
Understanding its molecular structure helps in modeling glacial and atmospheric processes. Isso contributes to more accurate predictions about climate change in polar regions.
The discovery reinforces the importance of hybrid techniques between experiments and computation. Ela demonstrates how AI accelerates resolutions of historical scientific problems.






