Elemental augmentation of machine learning interatomic potentials
Machine learning interatomic potentials (MLIPs) bridge the gap between the accuracy of ab initio methods and the computational efficiency needed for large-scale simulations. However, custom-trained MLIPs are often limited to specific materials and lack flexibility for incorporating additional elemen...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-06-01
|
| Series: | Computational Materials Today |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S295046352500002X |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1849325336503779328 |
|---|---|
| author | Haibo Xue Guanjian Cheng Wan-Jian Yin |
| author_facet | Haibo Xue Guanjian Cheng Wan-Jian Yin |
| author_sort | Haibo Xue |
| collection | DOAJ |
| description | Machine learning interatomic potentials (MLIPs) bridge the gap between the accuracy of ab initio methods and the computational efficiency needed for large-scale simulations. However, custom-trained MLIPs are often limited to specific materials and lack flexibility for incorporating additional elements, while universal potentials (UPots), despite covering a wide range of chemical elements, may sacrifice accuracy for generalization. In this work, we propose an elemental augmentation strategy to efficiently expand MLIPs by incorporating new elements into pre-trained models. Using a Bayesian optimization driven active learning framework, we target the configuration space of new elements where the current MLIPs exhibit high uncertainty and demonstrate the addition of up to 10 elements to a pre-trained UPot. The results demonstrate a high tendency for sampling new structures composed of these elements, minimizing sampling requirements. It reduces computational costs by over an order of magnitude compared to training an MLIP from scratch, while preserving accuracy. This strategy offers a scalable pathway to extend MLIP applicability across diverse chemical spaces. |
| format | Article |
| id | doaj-art-387e7c8028734a7e9e1d5b459a687789 |
| institution | Kabale University |
| issn | 2950-4635 |
| language | English |
| publishDate | 2025-06-01 |
| publisher | Elsevier |
| record_format | Article |
| series | Computational Materials Today |
| spelling | doaj-art-387e7c8028734a7e9e1d5b459a6877892025-08-20T03:48:27ZengElsevierComputational Materials Today2950-46352025-06-01610002610.1016/j.commt.2025.100026Elemental augmentation of machine learning interatomic potentialsHaibo Xue0Guanjian Cheng1Wan-Jian Yin2College of Energy, Soochow Institute for Energy and Materials InnovationS (SIEMIS), and Jiangsu Provincial Key Laboratory for Advanced Carbon Materials and Wearable Energy Technologies, Soochow University, Suzhou 215006, ChinaCollege of Energy, Soochow Institute for Energy and Materials InnovationS (SIEMIS), and Jiangsu Provincial Key Laboratory for Advanced Carbon Materials and Wearable Energy Technologies, Soochow University, Suzhou 215006, ChinaCorresponding author.; College of Energy, Soochow Institute for Energy and Materials InnovationS (SIEMIS), and Jiangsu Provincial Key Laboratory for Advanced Carbon Materials and Wearable Energy Technologies, Soochow University, Suzhou 215006, ChinaMachine learning interatomic potentials (MLIPs) bridge the gap between the accuracy of ab initio methods and the computational efficiency needed for large-scale simulations. However, custom-trained MLIPs are often limited to specific materials and lack flexibility for incorporating additional elements, while universal potentials (UPots), despite covering a wide range of chemical elements, may sacrifice accuracy for generalization. In this work, we propose an elemental augmentation strategy to efficiently expand MLIPs by incorporating new elements into pre-trained models. Using a Bayesian optimization driven active learning framework, we target the configuration space of new elements where the current MLIPs exhibit high uncertainty and demonstrate the addition of up to 10 elements to a pre-trained UPot. The results demonstrate a high tendency for sampling new structures composed of these elements, minimizing sampling requirements. It reduces computational costs by over an order of magnitude compared to training an MLIP from scratch, while preserving accuracy. This strategy offers a scalable pathway to extend MLIP applicability across diverse chemical spaces.http://www.sciencedirect.com/science/article/pii/S295046352500002XMachine learning interatomic potentialsElemental augmentationPotential energy surfaceComputational materials science |
| spellingShingle | Haibo Xue Guanjian Cheng Wan-Jian Yin Elemental augmentation of machine learning interatomic potentials Computational Materials Today Machine learning interatomic potentials Elemental augmentation Potential energy surface Computational materials science |
| title | Elemental augmentation of machine learning interatomic potentials |
| title_full | Elemental augmentation of machine learning interatomic potentials |
| title_fullStr | Elemental augmentation of machine learning interatomic potentials |
| title_full_unstemmed | Elemental augmentation of machine learning interatomic potentials |
| title_short | Elemental augmentation of machine learning interatomic potentials |
| title_sort | elemental augmentation of machine learning interatomic potentials |
| topic | Machine learning interatomic potentials Elemental augmentation Potential energy surface Computational materials science |
| url | http://www.sciencedirect.com/science/article/pii/S295046352500002X |
| work_keys_str_mv | AT haiboxue elementalaugmentationofmachinelearninginteratomicpotentials AT guanjiancheng elementalaugmentationofmachinelearninginteratomicpotentials AT wanjianyin elementalaugmentationofmachinelearninginteratomicpotentials |