Scholarlink

Recent works have highlighted the advantages of algorithm selection to optimise scientific simulations, presenting a range of approaches from classical time-series prediction, to expert-guided, to data-driven. In this work, we present novel variations upon these approaches for Molecular Dynamics simulations, implemented in the algorithm selection particle simulation library AutoPas, and compare them in terms of both performance and practicality. We demonstrate that these approaches can achieve speedups of up to 1.25 compared to an optimal single algorithm without dynamic algorithm selection.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Algorithm Selection in Short-Range Molecular Dynamics Simulations

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Samuel James Newcome,
  • Fabio Alexander Gratl,
  • Manuel Lerchner,
  • Abdulkadir Pazar,
  • Manish Kumar Mishra,
  • Hans-Joachim Bungartz

摘要

Recent works have highlighted the advantages of algorithm selection to optimise scientific simulations, presenting a range of approaches from classical time-series prediction, to expert-guided, to data-driven. In this work, we present novel variations upon these approaches for Molecular Dynamics simulations, implemented in the algorithm selection particle simulation library AutoPas, and compare them in terms of both performance and practicality. We demonstrate that these approaches can achieve speedups of up to 1.25 compared to an optimal single algorithm without dynamic algorithm selection.