Presentation Information
[10a-A22-8]AI-Agent for Autonomous Simulation of Time-Dependent Density-Functional Theory
〇Yuran Li1, Siyuan Li1, Arqum Hashmi1, Eiyu Gushiken1, Kenichi L. Ishikawa1 (1.The Univ of Tokyo)
Keywords:
ai agent,autonomous simulation,time-dependent density-functional theory
This presentation reports the design and initial validation of LaserM, a self-developed AI agent for automating simulations of ultrashort-pulse laser–material interactions. LaserM consists of a query loop, tool runtime, context compaction, and memory, and is designed to generate computational conditions, execute jobs, and analyze results. As a benchmark, LaserM autonomously performed linear-response calculations for graphene using SALMON based on time-dependent density-functional theory. After being instructed in natural language to check convergence by tuning parameters, it referred to SALMON source code and examples, adopted a four-atom orthorhombic cell, and completed 28 calculations. By comparing its selected result with a human-prepared calculation, we discuss the reproduction of in-plane isotropy and peak structures, as well as the trade-off between computational accuracy, resource use, and stable convergence.