Distillation of atomistic foundation models across architectures and chemical domains

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Gardner, John L. A., Toit, Daniel F. Thomas du, Mahmoud, Chiheb Ben, Beaulieu, Zoé Faure, Juraskova, Veronika, Paşca, Laura-Bianca, Rosset, Louise A. M., Duarte, Fernanda, Martelli, Fausto, Pickard, Chris J., Deringer, Volker L.
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916792186699776
author Gardner, John L. A.
Toit, Daniel F. Thomas du
Mahmoud, Chiheb Ben
Beaulieu, Zoé Faure
Juraskova, Veronika
Paşca, Laura-Bianca
Rosset, Louise A. M.
Duarte, Fernanda
Martelli, Fausto
Pickard, Chris J.
Deringer, Volker L.
author_facet Gardner, John L. A.
Toit, Daniel F. Thomas du
Mahmoud, Chiheb Ben
Beaulieu, Zoé Faure
Juraskova, Veronika
Paşca, Laura-Bianca
Rosset, Louise A. M.
Duarte, Fernanda
Martelli, Fausto
Pickard, Chris J.
Deringer, Volker L.
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2506_10956
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Distillation of atomistic foundation models across architectures and chemical domains
Gardner, John L. A.
Toit, Daniel F. Thomas du
Mahmoud, Chiheb Ben
Beaulieu, Zoé Faure
Juraskova, Veronika
Paşca, Laura-Bianca
Rosset, Louise A. M.
Duarte, Fernanda
Martelli, Fausto
Pickard, Chris J.
Deringer, Volker L.
Computational Physics
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.
title Distillation of atomistic foundation models across architectures and chemical domains
topic Computational Physics
url https://arxiv.org/abs/2506.10956