Distillation of atomistic foundation models across architectures and chemical domains

TL;DR

This paper demonstrates how distillation with synthetic data can significantly improve the efficiency of atomistic foundation models, enabling faster and more resource-efficient simulations across various chemical and materials domains.

Contribution

It introduces a distillation method that transfers knowledge between different architectures and domains, greatly reducing computational costs of atomistic models.

Findings

Speed-ups of over 10x by distilling between graph-network architectures

Speed-ups of over 100x using atomic cluster expansion framework

Applicable across diverse chemical and materials systems

Abstract

Machine-learned interatomic potentials have transformed computational research in the physical sciences. Recent atomistic `foundation' models have changed the field yet again: trained on many different chemical elements and domains, these potentials are widely applicable, but comparably slow and resource-intensive to run. Here we show how distillation via synthetic data can be used to cheaply transfer knowledge from atomistic foundation models to a range of different architectures, unlocking much smaller, more efficient potentials. We demonstrate speed-ups of $> 10\times$ by distilling from one graph-network architecture into another, and $> 100\times$ by leveraging the atomic cluster expansion framework. We showcase applicability across chemical and materials domains: from liquid water to hydrogen under extreme conditions; from porous silica and a hybrid halide perovskite solar-cell material to modelling organic reactions. Our work shows how distillation can support the routine and computationally efficient use of current and future atomistic foundation models in real-world scientific research.