AI connects ice friction to the heat of sliding

An AI-assisted study accepted by Physical Review Letters on 5 October 2026 connects ice’s increasing slipperiness during motion to contact heating. It models glass sliding over ice.[1]

A neural network learnt atomic forces from quantum calculations. Nanoscale simulations alone predicted friction increasing with speed. Adding heat transfer reversed that trend: warming thickened an existing lubricating film roughly tenfold and reduced its viscosity about 100-fold.[1]

Agreement with 55 held-out measurements reached R² = 0.88 on a logarithmic scale. The four sources included glass contacts and granite curling stones. Separate low-speed data calibrated one parameter; temperatures reference the model’s melting point.[1]

How the layer forms

The layer’s origin is a separate physical question. In a 2025 study, Achraf Atila and colleagues simulated two differently aligned ice crystals sliding against each other. Their molecular structure became disordered as sliding progressed, producing liquid-like material without first heating the interface to melting.[2]

This separates mechanical disruption of a crystal from a heat-driven phase transition. The simulations also found that counterface shape and attraction to water changed friction: curved, water-repelling surfaces generated less resistance than similarly shaped water-attracting surfaces.[2]

These simulations distinguish two jobs for sliding: it can disorder the crystal to create a film, while the contacting material influences how that film carries load and slips.[2]

Measuring the moving interface

A 2019 experiment made that interface measurable while it moved. Researchers attached a glass bead to a tuning fork and oscillated it sideways across ice. A second, much smaller vertical oscillation probed the layer’s mechanical response without stopping the sliding.[3]

The inferred layer was 100–500 nanometres thick and behaved as a viscoelastic material, combining fluid resistance with an elastic response. Its viscosity could reach 100 times that of water. Coating the bead to repel water reduced friction by as much as tenfold, without a corresponding change in layer thickness.[3]

Report an error or suggest a correction