cs.AIApr 1, 2026

From Data to Theory: Autonomous Large Language Model Agents for Materials Science

Authors: Samuel Onimpa Alfred, Veera Sundararaghavan

Organizations: Department of Aerospace Engineering, University of Michigan, Ann Arbor, MI 48109

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

We present an autonomous large language model (LLM) agent for end-to-end, data-driven materials theory development. The model can choose an equation form, generate and run its own code, and test how well the theory matches the data without human intervention. The framework combines step-by-step reasoning with expert-supplied tools, allowing the agent to adjust its approach as needed while keeping a clear record of its decisions. For well-established materials relationships such as the Hall-Petch equation and Paris law, the agent correctly identifies the governing equation and makes reliable predictions on new datasets. For more specialized relationships, such as Kuhn's equation for the HOMO-LUMO gap of conjugated molecules as a function of length, performance depends more strongly on the underlying model, with GPT-5 showing better recovery of the correct equation. Beyond known theories, the agent can also suggest new predictive relationships, illustrated here by a strain-dependent law for changes in the HOMO-LUMO gap. At the same time, the results show that careful validation remains essential, because the agent can still return incorrect, incomplete, or inconsistent equations even when the numerical fit appears strong. Overall, these results highlight both the promise and the current limitations of autonomous LLM agents for AI-assisted scientific modeling and discovery.

Figures & tables

Figure 1: Schematic overview of the three-component autonomous fitting framework, illustrating the closed-loop iterative workflow. The Reasoning Engine, Tool Registry, and Agent State interact through a closed-loop iterative workflow comprising Thought, Action, and Observation steps.
Figure 2: Structured response format . At each iteration, the LLM responds with a Thought , an Action specifying a tool and its input, and an Observation that captures the tool’s output.
Figure 3: Sample tool entry for the generate_function tool .Each tool in the registry is defined by a description, an input schema, and an execution function.
ToolDescriptionInput SchemaPurpose
load_dataLoad experimental data{"filename": "path"}Data ingestion
generate_functionGenerate physical equation from LLM’s knowledge{}Symbolic function generation with no-fallback
fit_modelPerform nonlinear least squares fitting{"initial_params": ...}Parameter optimization
test_functionValidate generated function with sample inputs{"test_params": [values], ...}Prevents invalid functions
validate_fitCalculate goodness-of-fit metrics{}Computes R², RMSE, residuals
create_plotsGenerate visualizations{}Produces diagnostic plots
Table 1: Core tools in the agentic framework. Each tool implements a specific operation required for empirical model fitting.
Figure 4: Sample task-specific system prompt. The LLM is initialized with a prompt that defines its role, the target physical context, and the expected reasoning structure.
Figure 5: Logical substructures of the agent’s state . The state is organized into modular fields that are updated by tool executions.
Figure 6: Agent-generated Hall-Petch fitting results .Experimental data points (blue circles) are shown together with the fitted Hall-Petch line (red line)).
Figure 7: Agent-generated Paris law fitting results. Experimental data points (blue circles) are shown together with the fitted Paris line (red line)).
Figure 8: DFT(LDA) HOMO-LUMO gaps for helicenes as a function of chain length. The solid blue curve represents the fit provided by the GPT-5 powered agent using the Kuhn’s equation.
SourceGPT-4 PoweredGPT-5 Powered
Knowledge recallh28mL2N+V0\displaystyle\frac{h^{2}}{8mL^{2}}N+V_{0}h28mL2(N+1)+V0\displaystyle\frac{h^{2}}{8mL^{2}}(N+1)+V_{0}
Literatureh28mL2(N+1)+V0\displaystyle\frac{h^{2}}{8mL^{2}}(N+1)+V_{0}h28mL2(N+1)+V0(1−1N)\displaystyle\frac{h^{2}}{8mL^{2}}(N+1)+V_{0}\left(1-\frac{1}{N}\right)
Table 2: Equations generated by GPT-4 and GPT-5 powered agents across knowledge recall and literature extraction tasks.
Figure 9: Reasoning trace showing PDF extraction failure and subsequent successful equation extraction from HTML.
Figure 10: Prompt provided to the agent to model strain effects. The LLM is initialized with a prompt that defines its role, the physical context (modifying the Kuhn equation for strain effects), and the expected reasoning-action-observation workflow.
Figure 11: Representative output plot from multiple agent runs showing the strain-modified Kuhn equation predictions.
More in this topic

Large Language Model-Guided Scientific Discovery

Large language model-guided scientific discovery uses AI to propose, test, and refine hypotheses, models, experiments, and designs.

See all 274 papers

Explore similar work

CardsList
Jun 29, 2026

Optimizing Expert-Designed Crystal Graph Networks for Band-Gap Prediction with an Autonomous LLM Research Loop

Predicting a material's properties from its structure is a central, fast-advancing problem in computational materials science. A decade of work has produced standard public benchmarks and many published machine-learning models for the task (Dunn et al., 2020). The task's fixed metric and these baselines make it a natural setting for autonomous agent research (Karpathy, 2026). On the MatBench band-gap benchmark (>>100k crystals), a general-purpose coding agent autonomously built the most accurate model trained without external pretraining, ahead of all seventeen expert-designed models reported for the task. A closer analysis shows it reached this by implementing known methods: either already standard in crystal neural-network models, or borrowed from other areas of machine learning. The contributing implementations include element-pair features on each message-passing edge and a crystal space-group embedding. The work not only demonstrates that LLM-agent autonomous research can optimize an expert-designed machine learning model for material property prediction, but also investigates the limitations of such autonomous research.

Chenmu Zhang, Boris I. Yakobson

Department of Materials Science and NanoEngineering Rice University Houston, TX 77005, USA

Graph Neural NetworksLarge Language Model-Guided OptimizationMaterials Property PredictionAutonomous Research Agents
AskPDF
Apr 15, 2026

MIND: AI Co-Scientist for Material Research

Large language models (LLMs) have enabled agentic AI systems for scientific discovery, but most approaches remain limited to textbased reasoning without automated experimental verification. We propose MIND, an LLM-driven framework for automated hypothesis validation in materials research. MIND organizes the scientific discovery process into hypothesis refinement, experimentation, and debate-based validation within a multi-agent pipeline. For experimental verification, the system integrates Machine Learning Interatomic Potentials, particularly SevenNet-Omni, enabling scalable in-silico experiments. We also provide a web-based user interface for automated hypothesis testing. The modular design allows additional experimental modules to be integrated, making the framework adaptable to broader scientific workflows. The code is available at: https://github.com/IMMS-Ewha/MIND, and a demonstration video at: https://youtu.be/lqiFe1OQzN4.

Geonhee Ahn, Donghyun Lee, Hayoung Doo +3

Institute for Multiscale Matter and Systems (IMMS), Ewha Womans University, Seoul 03760, Republic of Korea

AI Agents for Scientific DiscoveryAutonomous ExperimentationMaterials ScienceMulti-Agent LLM Systems
AskPDF
Dec 12, 2025

AGAPI-Agents: An Open-Access Agentic AI Platform for Accelerated Materials Design on AtomGPT.org

Agentic AI systems increasingly connect large language models (LLMs) to external scientific tools, yet whether and when tool access improves prediction accuracy remains uncharacterized. We present AGAPI (AtomGPT.org API), an open access platform integrating eight open-source LLMs with 18 REST endpoints (28 agent tools, 50 web apps) spanning materials databases, force fields, tight-binding band structures, X-ray diffraction, and protein structure. A three-evaluation residual decomposition on JARVIS-Leaderboard electronic-structure test sets separates agent pipeline fidelity from inherited density functional theory (DFT) functional bias. For bulk modulus and bandgap the agent reproduces JARVIS-DFT entries to numerical precision, so the experimental-reference degradation is functional bias, not agentic malfunction. On memorization-resistant test sets (57 defective supercells, 60 hypothetical compositions), tool-augmented mean absolute error (MAE) is below 0.005 eV versus 1.25 to 1.86 eV tool-free, confirming tools are indispensable where parametric knowledge is unavailable. We further demonstrate autonomous multi-step workflows including 10-operation defect-engineering pipelines. AGAPI is available at https://github.com/atomgptlab/agapi.

Jaehyung Lee, Justin Ely, Kent Zhang +3

Department of Materials Science and Engineering, Johns Hopkins University, Baltimore, MD 21218, USA · Department of Electrical and Computer Engineering, Johns Hopkins University, Baltimore, MD 21218, USA · Department of Physics and Astronomy, West Virginia University, Morgantown, WV 26506, USA

AI Agents for Scientific DiscoveryLLM Agent EvaluationMaterials Property PredictionTool-Using AgentsComputational Materials Science
AskPDF