AtomAgents: Alloy design and discovery through physics-aware multi-modal multi-agent artificial intelligence

TL;DR

AtomAgents presents a physics-aware, multi-agent AI framework that orchestrates LLMs, perception, and atomistic simulations to automate alloy design and analysis. By integrating multi-modal data, literature retrieval, and physics-based computation within autonomous agent workflows, it demonstrates end-to-end tasks such as property calculation, defect analysis, and fracture-toughness screening across alloy systems. The results show accurate property predictions, dislocation-core characterization, and hypothesis-driven discoveries, highlighting reduced human intervention and enhanced design throughput. The work points to significant impact in multi-objective materials optimization and lays groundwork for expanding to complex systems like high-entropy alloys and beyond.

Abstract

The design of alloys is a multi-scale problem that requires a holistic approach that involves retrieving relevant knowledge, applying advanced computational methods, conducting experimental validations, and analyzing the results, a process that is typically reserved for human experts. Machine learning (ML) can help accelerate this process, for instance, through the use of deep surrogate models that connect structural features to material properties, or vice versa. However, existing data-driven models often target specific material objectives, offering limited flexibility to integrate out-of-domain knowledge and cannot adapt to new, unforeseen challenges. Here, we overcome these limitations by leveraging the distinct capabilities of multiple AI agents that collaborate autonomously within a dynamic environment to solve complex materials design tasks. The proposed physics-aware generative AI platform, AtomAgents, synergizes the intelligence of large language models (LLM) the dynamic collaboration among AI agents with expertise in various domains, including knowledge retrieval, multi-modal data integration, physics-based simulations, and comprehensive results analysis across modalities that includes numerical data and images of physical simulation results. The concerted effort of the multi-agent system allows for addressing complex materials design problems, as demonstrated by examples that include autonomously designing metallic alloys with enhanced properties compared to their pure counterparts. Our results enable accurate prediction of key characteristics across alloys and highlight the crucial role of solid solution alloying to steer the development of advanced metallic alloys. Our framework enhances the efficiency of complex multi-objective design tasks and opens new avenues in fields such as biomedical materials engineering, renewable energy, and environmental sustainability.

Figures (11)

• Figure 1: Multi-model multi-agent approach as a flexible modeling strategy for materials discovery, modeling, and prediction. Multi-agent modeling can extend the power of large-language models by enabling the integration of multimodal data from diverse sources, including simulations, experiments, materials databases, and theoretical models.
• Figure 2: AtomAgents, a physics-based generative multi-agent model for automating alloy discovery and analysis with atomistic simulations. The structure of AtomAgents comprises a team of agents constructing the core who collaborate to solve complex alloy design tasks with the help of a set of tools for different purposes described in the image from knowledge retrieval to coding to image analysis. Each tool is composed of a set of AI agents that collaborate to solve the query received from the User and return the results to the core agents. Each individual AI agent in AtomtAgents is assigned a distinct profile that defines its role and may be powered by a general purpose large language model from the OpenAI GPT family. The entire process is automated, providing a robust framework for solving challenging tasks in alloy design and analysis with minimal or no human intervention.
• Figure 2: The analyzes returned by the "plot analyzer" agent for the plot of critical fracture toughness versus Nb concentration created by the multi-agent collaboration.
• Figure 3: Overview of the multi-agent collaboration to solve the complex task posed in Experiment I. After receiving the task from the user, the core agents call the "planning" tool to create a plan for the task. Then the core agents start executing the plan by using "computation" tool to compute the material properties and "knowledge retrieval" tool to retrieve the material properties from a set of scientific papers or other documents. Finally, all the data are collected and sent to "Coding" tool to save them in a comma-separated values (CSV) file.
• Figure 3: The results of the analyze tool from the analyzes of the plot generated by the multi-agent collaboration in experiment IV.