Machine-learning models explain structural transitions in disordered silicon
The properties of matter often change drastically under pressure, as bonds between atoms are broken and new ones are formed. The mechanisms behind such structural changes have been persistent challenges for even the most advanced experimental and simulation techniques.
In this week's issue of Nature, Volker Deringer and collaborators report new insight into the structural and electronic transitions of disordered phases of silicon, obtained from computer simulations that are driven by atomistic machine-learning methods. The team were able to describe the behaviour of a system of 100,000 silicon atoms with quantum-mechanical accuracy. Their simulations reveal a complex series of transformations under pressure – finally leading to a nanocrystalline phase. In a wider picture, the authors’ work provides an example for the power of emerging machine-learning methods to solve challenging problems in physics, chemistry, and materials science.