Universally Accurate or Specifically Inadequate? Stress-Testing General Purpose Machine Learning Interatomic Potentials

Title data

Jakob, Konstantin S. ; Reuter, Karsten ; Margraf, Johannes T.:
Universally Accurate or Specifically Inadequate? Stress-Testing General Purpose Machine Learning Interatomic Potentials.
In: Advanced Intelligent Discovery. (2025) . - 202500031.
ISSN 2943-9981
DOI: https://doi.org/10.1002/aidi.202500031

Official URL:

Abstract in another language

Machine learning interatomic potentials (MLIPs) have revolutionized the field of atomistic materials simulation, both due to their remarkable accuracy and their computational efficiency compared to established ab initio methods. Very recently, several general purpose MLIPs have been reported, which are broadly applicable across the periodic table. These represent a fascinating opportunity for materials discovery, provided that they are robust and transferable. In order to stress test current general purpose MLIPs, we evaluate the performance of M3GNet and MACE models in element-substitution based structure prediction workflows for a diverse range of inorganic, crystalline materials. Importantly, these results are compared with a full density functional based workflow, shifting the focus from merely evaluating single-point energy and force predictions of MLIPs toward an end-to-end perspective. We find that general purpose MLIPs are in general well-suited to accelerate computational materials discovery and structure prediction, but also display certain systematic biases. To address these, a simple metric to quantify MLIP reliability for materials discovery is introduced. As a by-product, we also predict novel ground state structures for 15 out of 100 analyzed compositions.

Further data

Item Type:	Article in a journal
Refereed:	Yes
Keywords:	element substitution; inorganic materials discovery; machine learning potentials
Institutions of the University:	Faculties > Faculty of Biology, Chemistry and Earth Sciences > Department of Chemistry > Chair Physical Chemistry V - Theory and Machine Learning Faculties > Faculty of Biology, Chemistry and Earth Sciences > Department of Chemistry > Chair Physical Chemistry V - Theory and Machine Learning > Chair Physical Chemistry V - Theory and Machine Learning - Univ.-Prof. Dr. Johannes Theo Margraf Research Institutions > Central research institutes > Research Center for AI in Science and Society
Result of work at the UBT:	Yes
DDC Subjects:	500 Science > 540 Chemistry
Date Deposited:	10 Nov 2025 09:16
Last Modified:	10 Nov 2025 09:16
URI:	https://eref.uni-bayreuth.de/id/eprint/95159