Systematic softening in universal machine learning interatomic potentials
Abstract Machine learning interatomic potentials (MLIPs) have introduced a new paradigm for atomic simulations. Recent advancements have led to universal MLIPs (uMLIPs) that are pre-trained on diverse datasets, providing opportunities for universal force fields and foundational machine learning mode...
Saved in:
| Main Authors: | , , , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Nature Portfolio
2025-01-01
|
| Series: | npj Computational Materials |
| Online Access: | https://doi.org/10.1038/s41524-024-01500-6 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1841544461665959936 |
|---|---|
| author | Bowen Deng Yunyeong Choi Peichen Zhong Janosh Riebesell Shashwat Anand Zhuohan Li KyuJung Jun Kristin A. Persson Gerbrand Ceder |
| author_facet | Bowen Deng Yunyeong Choi Peichen Zhong Janosh Riebesell Shashwat Anand Zhuohan Li KyuJung Jun Kristin A. Persson Gerbrand Ceder |
| author_sort | Bowen Deng |
| collection | DOAJ |
| description | Abstract Machine learning interatomic potentials (MLIPs) have introduced a new paradigm for atomic simulations. Recent advancements have led to universal MLIPs (uMLIPs) that are pre-trained on diverse datasets, providing opportunities for universal force fields and foundational machine learning models. However, their performance in extrapolating to out-of-distribution complex atomic environments remains unclear. In this study, we highlight a consistent potential energy surface (PES) softening effect in three uMLIPs: M3GNet, CHGNet, and MACE-MP-0, which is characterized by energy and force underprediction in atomic-modeling benchmarks including surfaces, defects, solid-solution energetics, ion migration barriers, phonon vibration modes, and general high-energy states. The PES softening behavior originates primarily from the systematically underpredicted PES curvature, which derives from the biased sampling of near-equilibrium atomic arrangements in uMLIP pre-training datasets. Our findings suggest that a considerable fraction of uMLIP errors are highly systematic, and can therefore be efficiently corrected. We argue for the importance of a comprehensive materials dataset with improved PES sampling for next-generation foundational MLIPs. |
| format | Article |
| id | doaj-art-bd0ab6e22c1f4e468aa2579e3eabcd02 |
| institution | Kabale University |
| issn | 2057-3960 |
| language | English |
| publishDate | 2025-01-01 |
| publisher | Nature Portfolio |
| record_format | Article |
| series | npj Computational Materials |
| spelling | doaj-art-bd0ab6e22c1f4e468aa2579e3eabcd022025-01-12T12:32:19ZengNature Portfolionpj Computational Materials2057-39602025-01-011111910.1038/s41524-024-01500-6Systematic softening in universal machine learning interatomic potentialsBowen Deng0Yunyeong Choi1Peichen Zhong2Janosh Riebesell3Shashwat Anand4Zhuohan Li5KyuJung Jun6Kristin A. Persson7Gerbrand Ceder8Department of Materials Science and Engineering, University of CaliforniaDepartment of Materials Science and Engineering, University of CaliforniaDepartment of Materials Science and Engineering, University of CaliforniaCavendish Laboratory, University of CambridgeMaterials Sciences Division, Lawrence Berkeley National LaboratoryMaterials Sciences Division, Lawrence Berkeley National LaboratoryDepartment of Materials Science and Engineering, University of CaliforniaDepartment of Materials Science and Engineering, University of CaliforniaDepartment of Materials Science and Engineering, University of CaliforniaAbstract Machine learning interatomic potentials (MLIPs) have introduced a new paradigm for atomic simulations. Recent advancements have led to universal MLIPs (uMLIPs) that are pre-trained on diverse datasets, providing opportunities for universal force fields and foundational machine learning models. However, their performance in extrapolating to out-of-distribution complex atomic environments remains unclear. In this study, we highlight a consistent potential energy surface (PES) softening effect in three uMLIPs: M3GNet, CHGNet, and MACE-MP-0, which is characterized by energy and force underprediction in atomic-modeling benchmarks including surfaces, defects, solid-solution energetics, ion migration barriers, phonon vibration modes, and general high-energy states. The PES softening behavior originates primarily from the systematically underpredicted PES curvature, which derives from the biased sampling of near-equilibrium atomic arrangements in uMLIP pre-training datasets. Our findings suggest that a considerable fraction of uMLIP errors are highly systematic, and can therefore be efficiently corrected. We argue for the importance of a comprehensive materials dataset with improved PES sampling for next-generation foundational MLIPs.https://doi.org/10.1038/s41524-024-01500-6 |
| spellingShingle | Bowen Deng Yunyeong Choi Peichen Zhong Janosh Riebesell Shashwat Anand Zhuohan Li KyuJung Jun Kristin A. Persson Gerbrand Ceder Systematic softening in universal machine learning interatomic potentials npj Computational Materials |
| title | Systematic softening in universal machine learning interatomic potentials |
| title_full | Systematic softening in universal machine learning interatomic potentials |
| title_fullStr | Systematic softening in universal machine learning interatomic potentials |
| title_full_unstemmed | Systematic softening in universal machine learning interatomic potentials |
| title_short | Systematic softening in universal machine learning interatomic potentials |
| title_sort | systematic softening in universal machine learning interatomic potentials |
| url | https://doi.org/10.1038/s41524-024-01500-6 |
| work_keys_str_mv | AT bowendeng systematicsofteninginuniversalmachinelearninginteratomicpotentials AT yunyeongchoi systematicsofteninginuniversalmachinelearninginteratomicpotentials AT peichenzhong systematicsofteninginuniversalmachinelearninginteratomicpotentials AT janoshriebesell systematicsofteninginuniversalmachinelearninginteratomicpotentials AT shashwatanand systematicsofteninginuniversalmachinelearninginteratomicpotentials AT zhuohanli systematicsofteninginuniversalmachinelearninginteratomicpotentials AT kyujungjun systematicsofteninginuniversalmachinelearninginteratomicpotentials AT kristinapersson systematicsofteninginuniversalmachinelearninginteratomicpotentials AT gerbrandceder systematicsofteninginuniversalmachinelearninginteratomicpotentials |