Ultimate self-learning metabasin escape algorithm for supercooled liquids and glasses

ORAL

Abstract

A generic history-penalized metabasin escape algorithm is presented in this work without any predetermined parameters. The configuration space location and volume of imposed penalty functions are determined in self-learning processes as the complete 3N-dimensional potential energy surface is sampled. The computational efficiency is demonstrated using the binary Lennard-Jones liquid supercooled to the glass transition temperature, which shows an exponential enhancement over previous algorithm implementations.

–

Authors

  • Penghui Cao

    Department of Mechanical Engineering, Boston University

  • Harold S. Park

    Department of Mechanical Engineering, Boston University

  • Xi Lin

    Boston University, Department of Mechanical Engineering and Division of Materials Science and Engineering, Boston University, Department of Mechanical Engineering, Boston University, Division of Materials Science and Engineering, Boston University