Long Time-Scale Atomistic Simulations

Sadigh, B ; Cai, W ; de Koning, M ; Oppelstrup, T ; Bulatov, V ; Kalos, M

Lawrence Livermore National Laboratory

关键词: Probability;    Abundance;    Minimization;    36 Materials Science;    Atoms;   

DOI  :  10.2172/15014735
RP-ID  :  UCRL-TR-209741
RP-ID  :  W-7405-ENG-48
RP-ID  :  15014735

美国|英语

来源: UNT Digital Library

【 摘 要 】

During the past two years, we have succeeded in implementing an efficient parallel Importance Sampling Monte-Carlo (ISMC) scheme with application towards rarely occurring transition events, of great abundance in materials science and chemistry. The inefficiency of the standard atomistic modeling techniques for these problems may be traced to the extremely low probability of sampling system trajectories, or paths, that lead to a successful transition event. Instead of following the conventional approach of developing smart algorithms for searching transition paths, we tackle this problem by explicitly enhancing the probability of sampling successful transition events by means of an importance function. By selecting it appropriately, one focuses predominantly on the successful transition paths while discarding most irrelevant ones. In this manner, the rare-event problem is reformulated into an optimization problem for finding the best-possible importance function. Utilizing efficient iterative minimization algorithms, our IS approach can now be used to calculate the rate of occurrence of low-probability transition phenomena of short duration (short successful paths), but which involve collective degrees of freedom of many atoms.