EAGER: Foundation Models for Rare Events

Rare events play a critical role in fundamental scientific processes, such as diffusion and chemical reactions, which have broad applications in materials science, chemistry, and biology. However, predicting these events remains a major computational challenge due to their rarity and the complexity of high-dimensional potential energy landscapes. This project aims to develop a foundation model for predicting rare events across diverse atomistic systems, significantly reducing computational costs and enabling more efficient scientific discoveries.

By leveraging recent advances in artificial intelligence (AI) and computational materials science, this research aligns with the National Science Foundation’s mission by promoting the progress of science and advancing national prosperity. The project will accelerate the discovery of new materials and chemicals, benefiting industries such as clean energy and sustainability.

Additionally, the open dissemination of models and datasets will foster education and contribute to workforce development in AI and materials science. As part of its outreach efforts, the project will engage students from different levels through educational workshops, mentorship programs, and open-access learning resources, equipping the next generation of researchers with cutting-edge computational tools.

This project focuses on the development of a general-purpose foundation model for predicting rare events in atomistic simulations. Unlike conventional machine learning approaches that require extensive retraining for specific materials, this model leverages advanced AI techniques—such as equivariant Transformers, generative models, and multimodal learning—to enhance prediction accuracy and generalization.

To address data scarcity, the model integrates high-fidelity graph neural network interatomic potentials, large density functional theory databases, and synthetic data from generative models. The proposed workflow enables the prediction of transition states, pathways, and reaction rates for rare events. In its initial phase, the project will focus on diffusional rare events in inorganic solid-state materials, demonstrating applications in energy storage technologies such as batteries and fuel cells.

The outcomes will provide a computational foundation for modeling and predicting rare events across multiple scientific disciplines, accelerating breakthroughs in materials discovery and beyond.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.