Exploring model complexity in machine learned potentials for simulated properties
Abstract Machine learning (ML) enables the development of interatomic potentials with the accuracy of first principles methods while retaining the speed and parallel efficiency of empirical potentials. While ML potentials traditionally use atom-centered descriptors as inputs, different...
| 主要な著者: | , , , , , , , |
|---|---|
| その他の著者: | |
| フォーマット: | 論文 |
| 言語: | English |
| 出版事項: |
Springer International Publishing
2023
|
| オンライン･アクセス: | https://hdl.handle.net/1721.1/152387 |
| _version_ | 1826209959278280704 |
|---|---|
| author | Rohskopf, A. Goff, J. Sema, D. Gordiz, K. Nguyen, N. C. Henry, A. Thompson, A. P. Wood, M. A. |
| author2 | Massachusetts Institute of Technology. Department of Mechanical Engineering |
| author_facet | Massachusetts Institute of Technology. Department of Mechanical Engineering Rohskopf, A. Goff, J. Sema, D. Gordiz, K. Nguyen, N. C. Henry, A. Thompson, A. P. Wood, M. A. |
| author_sort | Rohskopf, A. |
| collection | MIT |
| description | Abstract
Machine learning (ML) enables the development of interatomic potentials with the accuracy of first principles methods while retaining the speed and parallel efficiency of empirical potentials. While ML potentials traditionally use atom-centered descriptors as inputs, different models such as linear regression and neural networks map descriptors to atomic energies and forces. This begs the question: what is the improvement in accuracy due to model complexity irrespective of descriptors? We curate three datasets to investigate this question in terms of ab initio energy and force errors: (1) solid and liquid silicon, (2) gallium nitride, and (3) the superionic conductor Li
$$_{10}$$
10
Ge(PS
$$_{6}$$
6
)
$$_{2}$$
2
(LGPS). We further investigate how these errors affect simulated properties and verify if the improvement in fitting errors corresponds to measurable improvement in property prediction. By assessing different models, we observe correlations between fitting quantity (e.g. atomic force) error and simulated property error with respect to ab initio values.
Graphical abstract |
| first_indexed | 2024-09-23T14:37:17Z |
| format | Article |
| id | mit-1721.1/152387 |
| institution | Massachusetts Institute of Technology |
| language | English |
| last_indexed | 2024-09-23T14:37:17Z |
| publishDate | 2023 |
| publisher | Springer International Publishing |
| record_format | dspace |
| spelling | mit-1721.1/1523872024-01-23T18:38:31Z Exploring model complexity in machine learned potentials for simulated properties Rohskopf, A. Goff, J. Sema, D. Gordiz, K. Nguyen, N. C. Henry, A. Thompson, A. P. Wood, M. A. Massachusetts Institute of Technology. Department of Mechanical Engineering Massachusetts Institute of Technology. Department of Aeronautics and Astronautics Abstract Machine learning (ML) enables the development of interatomic potentials with the accuracy of first principles methods while retaining the speed and parallel efficiency of empirical potentials. While ML potentials traditionally use atom-centered descriptors as inputs, different models such as linear regression and neural networks map descriptors to atomic energies and forces. This begs the question: what is the improvement in accuracy due to model complexity irrespective of descriptors? We curate three datasets to investigate this question in terms of ab initio energy and force errors: (1) solid and liquid silicon, (2) gallium nitride, and (3) the superionic conductor Li $$_{10}$$ 10 Ge(PS $$_{6}$$ 6 ) $$_{2}$$ 2 (LGPS). We further investigate how these errors affect simulated properties and verify if the improvement in fitting errors corresponds to measurable improvement in property prediction. By assessing different models, we observe correlations between fitting quantity (e.g. atomic force) error and simulated property error with respect to ab initio values. Graphical abstract 2023-10-06T15:49:52Z 2023-10-06T15:49:52Z 2023-09-18 2023-09-24T03:14:37Z Article http://purl.org/eprint/type/JournalArticle https://hdl.handle.net/1721.1/152387 Rohskopf, A., Goff, J., Sema, D., Gordiz, K., Nguyen, N. C. et al. 2023. "Exploring model complexity in machine learned potentials for simulated properties." PUBLISHER_CC en https://doi.org/10.1557/s43578-023-01152-0 Creative Commons Attribution https://creativecommons.org/licenses/by/4.0/ The Author(s) application/pdf Springer International Publishing Springer International Publishing |
| spellingShingle | Rohskopf, A. Goff, J. Sema, D. Gordiz, K. Nguyen, N. C. Henry, A. Thompson, A. P. Wood, M. A. Exploring model complexity in machine learned potentials for simulated properties |
| title | Exploring model complexity in machine learned potentials for simulated properties |
| title_full | Exploring model complexity in machine learned potentials for simulated properties |
| title_fullStr | Exploring model complexity in machine learned potentials for simulated properties |
| title_full_unstemmed | Exploring model complexity in machine learned potentials for simulated properties |
| title_short | Exploring model complexity in machine learned potentials for simulated properties |
| title_sort | exploring model complexity in machine learned potentials for simulated properties |
| url | https://hdl.handle.net/1721.1/152387 |
| work_keys_str_mv | AT rohskopfa exploringmodelcomplexityinmachinelearnedpotentialsforsimulatedproperties AT goffj exploringmodelcomplexityinmachinelearnedpotentialsforsimulatedproperties AT semad exploringmodelcomplexityinmachinelearnedpotentialsforsimulatedproperties AT gordizk exploringmodelcomplexityinmachinelearnedpotentialsforsimulatedproperties AT nguyennc exploringmodelcomplexityinmachinelearnedpotentialsforsimulatedproperties AT henrya exploringmodelcomplexityinmachinelearnedpotentialsforsimulatedproperties AT thompsonap exploringmodelcomplexityinmachinelearnedpotentialsforsimulatedproperties AT woodma exploringmodelcomplexityinmachinelearnedpotentialsforsimulatedproperties |