Agent-MD: LLM-Driven Framework for Molecular Simulation Campaigns
Agent-MD is a groundbreaking approach that blends reasoning from large language models (LLMs) into detailed molecular simulation processes, particularly focusing on grand canonical Monte Carlo-molecular dynamics (GCMC-MD) workflows. It cleverly integrates LLMs during the setup of simulation campaigns and when specific events prompt reviews. Meanwhile, a consistent rule-based agent manages routine tasks like simulation execution, data analysis, and archiving, all while following established guidelines and maintaining accurate state records. This system addresses challenges in simulations needing repeated continuations and adaptive evaluations. It was tested with water-vapor desorption across five montmorillonite systems and three relative humidity levels (RH = 0.9-0.3-0.1), completing 120 simulation cycles. You can find the full paper on arXiv using the identifier 2608.07637.
Key facts
- Agent-MD is a framework that places LLM reasoning selectively at campaign construction and event-triggered review.
- Routine simulation, analysis, continuation, archiving, and state progression are handled by a persistent rule-based campaign agent.
- The framework was demonstrated in a GCMC-MD water-vapor desorption campaign.
- The campaign comprised five montmorillonite systems and three sequential relative-humidity states (RH = 0.9-0.3-0.1).
- Across 15 system-RH states, the workflow completed 120 segmented simulation cycles.
- The workflow used state-specific sampling lengths and provenance-aware tracking.
- The paper is available on arXiv under identifier 2608.07637.
- The framework addresses challenges in long-running molecular simulation campaigns.
Entities
Institutions
- arXiv