An unsupervised machine learning (ML) method is proposed to analyse the local structure around an atom, using a novel two-step locality-preserving projection. CONQUEST is used in linear scaling mode to produce MD data on 1,000 atoms of silicon and a silicon-germanium alloy to fit the ML forces.

The paper is published in Phys. Rev. B 105, 075107 (2022).

This entry was posted on 2022/2/3.