[Paper Review] AutoMOOSE: An Agentic AI for Autonomous Phase-Field Simulation

Sukriti Manna, Henry Chan+1 more|arXiv (Cornell University)|Mar 22, 2026

Solidification and crystal growth phenomena0 citations

TL;DR

AutoMOOSE is an open-source agentic framework that autonomously orchestrates the full MOOSE phase-field simulation workflow from a natural-language prompt, including input generation, execution, failure recovery, and analysis.

ABSTRACT

Multiphysics simulation frameworks such as MOOSE provide rigorous engines for phase-field materials modeling, yet adoption is constrained by the expertise required to construct valid input files, coordinate parameter sweeps, diagnose failures, and extract quantitative results. We introduce AutoMOOSE, an open-source agentic framework that orchestrates the full simulation lifecycle from a single natural-language prompt. AutoMOOSE deploys a five-agent pipeline in which the Input Writer coordinates six sub-agents and the Reviewer autonomously corrects runtime failures without user intervention. A modular plugin architecture enables new phase-field formulations without modifying the core framework, and a Model Context Protocol (MCP) server exposes the workflow as ten structured tools for interoperability with any MCP-compatible client. Validated on a four-temperature copper grain growth benchmark, AutoMOOSE generates MOOSE input files with 6 of 12 structural blocks matching a human expert reference exactly and 4 functionally equivalent, executes all runs in parallel with a 1.8x speedup, and performs an end-to-end physical consistency check spanning intent, finite-element execution, and Arrhenius kinetics with no human verification. Grain coarsening kinetics are recovered with R^2 = 0.90-0.95 at T >= 600 K; the recovered activation energy Q_fit = 0.296 eV is consistent with a human-written reference (Q_fit = 0.267 eV) under identical parameters. Three runtime failure classes were diagnosed and resolved autonomously within a single correction cycle, and every run produces a provenance record satisfying FAIR data principles. These results show that the gap between knowing the physics and executing a validated simulation campaign can be bridged by a lightweight multi-agent orchestration layer, providing a pathway toward AI-driven materials discovery and self-driving laboratories.

Motivation & Objective

• Demonstrate autonomous, end-to-end orchestration of a MOOSE phase-field simulation from natural-language intent.
• Eliminate manual input-file construction and iterative debugging by introducing agent-based workflow management.
• Provide a physics-agnostic plugin architecture to extend to new phase-field formulations without core changes.
• Ensure reproducibility and FAIR data principles via self-describing provenance and structured records.
• Showcase end-to-end validation on a four-temperature copper grain-growth benchmark with autonomous convergence recovery.

Proposed method

• Define a five-agent pipeline (Architect, Input Writer, Runner, Reviewer, Visualization) that translates a natural-language prompt into a validated MOOSE input file and runs parallel sweeps.
• Use a modular two-function plugin contract (generate_input, parse_results) to separate physics specifics from orchestration.
• Implement autonomous failure diagnosis and correction by routing convergence failures to the Reviewer, which adjusts parameters and regenerates inputs.
• Capture complete run provenance in self-contained directories with structured metadata to satisfy FAIR data principles.
• Validate on a four-temperature copper polycrystalline grain-growth benchmark, recovering kinetics and Arrhenius activation energy without human verification.

Figure 1: AutoMOOSE agentic pipeline. Five claude-sonnet-4-20250514 agents ( $f_{1}$ – $f_{5}$ ) transform a natural-language prompt into a completed MOOSE phase-field simulation. Architect ( $f_{1}$ ) parses the user prompt and constructs the structured simulation plan $\mathcal{P}$ (Eq. ( 9 )), en

Experimental results

Research questions

• RQ1 Can an agentic AI synthesize a correct MOOSE input file from natural language and execute a full phase-field simulation autonomously?
• RQ2 Does autonomous failure recovery (timesteps, mesh, tolerances) converge to valid results without human intervention?
• RQ3 How does the autonomous workflow perform in parallel temperature sweeps and reproduce known kinetic laws and Arrhenius parameters?
• RQ4 Is the pipeline sufficiently modular to extend to additional phase-field formulations via plugins?

Key findings

• The pipeline autonomously generates syntactically valid MOOSE inputs with high block match (6 of 12 blocks exact, 4 functionally equivalent) to a human reference.
• Four parallel runs execute with an 1.8× wall-clock speedup over serial execution.
• End-to-end consistency checks recover expected kinetics and Arrhenius behavior, with R^2 in grain coarsening between 0.90 and 0.95 for T≥600 K.
• Recovered Arrhenius activation energy Q_fit = 0.296 eV aligns with specified Q = 0.23 eV under identical mesh and parameters (and matches a human reference run).
• Three convergence-failure classes are autonomously diagnosed and resolved within a single correction cycle.
• Each run produces a self-documenting provenance record encoding simulation plan fields and execution metadata.

Figure 2: AutoMOOSE run directory structure. Each run directory is timestamped and self-contained, comprising: grain_growth.i (complete MOOSE input file), grain_growth.csv (tabulated grain count time series $N(t)$ , Eq. ( 18 )), run.log (full solver stdout), metadata.json (structured provenance reco