This research explores how foundation model principles from large language models can be applied to atomistic simulations, enabling more efficient and transferable materials modeling.

• Proposes scaling datasets and architectures for chemistry/materials science
• Develops models that are more robust to out-of-distribution challenges
• Aims to create broadly transferable simulation capabilities across chemical domains
• Leverages pre-training strategies similar to LLMs for improved performance

For engineering applications, this approach promises faster discovery and development of novel materials with desired properties, potentially transforming how we design everything from semiconductors to pharmaceuticals.

Foundation Models for Atomistic Simulation of Chemistry and Materials