Scholarlink

<p>The integration of large language models (LLMs) with domain-specific computational tools provides a pathway to streamline and enhance materials science workflows. This paper introduces MatSciAgent, a multi-agent framework supporting tasks such as materials data retrieval, continuum simulation, crystal structure generation, and molecular dynamics simulation. At its core is a master agent that interprets user queries, identifies the task type, and delegates to task-specific agent(s) equipped with tools. Leveraging databases such as Materials Project and MatWeb, the framework retrieves and summarizes materials data with grounded, factual responses, addressing limitations of vanilla LLMs. When a target material is absent from databases, a generative agent can propose plausible crystal structures. For simulations, specialized agents extract parameters to perform continuum and molecular dynamics simulations using existing software or custom code. MatSciAgent demonstrates stability, with parameter extraction achieving 100% success across five runs and materials extraction consistent in 9 of 10 runs. Its modular design ensures seamless extensibility to evolve as new capabilities are integrated.</p><p></p>

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

Modular large language model agents for multi-task computational materials science

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年3月26日

查看全文
  • Akshat Chaudhari,
  • Janghoon Ock,
  • Amir Barati Farimani

摘要

The integration of large language models (LLMs) with domain-specific computational tools provides a pathway to streamline and enhance materials science workflows. This paper introduces MatSciAgent, a multi-agent framework supporting tasks such as materials data retrieval, continuum simulation, crystal structure generation, and molecular dynamics simulation. At its core is a master agent that interprets user queries, identifies the task type, and delegates to task-specific agent(s) equipped with tools. Leveraging databases such as Materials Project and MatWeb, the framework retrieves and summarizes materials data with grounded, factual responses, addressing limitations of vanilla LLMs. When a target material is absent from databases, a generative agent can propose plausible crystal structures. For simulations, specialized agents extract parameters to perform continuum and molecular dynamics simulations using existing software or custom code. MatSciAgent demonstrates stability, with parameter extraction achieving 100% success across five runs and materials extraction consistent in 9 of 10 runs. Its modular design ensures seamless extensibility to evolve as new capabilities are integrated.