Learning pair potentials using differentiable simulations
<jats:p> Learning pair interactions from experimental or simulation data is of great interest for molecular simulations. We propose a general stochastic method for learning pair interactions from data using differentiable simulations (DiffSim). DiffSim defines a loss function based on structur...
| Main Authors: | , , , |
|---|---|
| Other Authors: | |
| Format: | Article |
| Language: | English |
| Published: |
AIP Publishing
2023
|
| Online Access: | https://hdl.handle.net/1721.1/150447 |
| _version_ | 1826188005126176768 |
|---|---|
| author | Wang, Wujie Wu, Zhenghao Dietschreit, Johannes CB Gómez-Bombarelli, Rafael |
| author2 | Massachusetts Institute of Technology. Department of Materials Science and Engineering |
| author_facet | Massachusetts Institute of Technology. Department of Materials Science and Engineering Wang, Wujie Wu, Zhenghao Dietschreit, Johannes CB Gómez-Bombarelli, Rafael |
| author_sort | Wang, Wujie |
| collection | MIT |
| description | <jats:p> Learning pair interactions from experimental or simulation data is of great interest for molecular simulations. We propose a general stochastic method for learning pair interactions from data using differentiable simulations (DiffSim). DiffSim defines a loss function based on structural observables, such as the radial distribution function, through molecular dynamics (MD) simulations. The interaction potentials are then learned directly by stochastic gradient descent, using backpropagation to calculate the gradient of the structural loss metric with respect to the interaction potential through the MD simulation. This gradient-based method is flexible and can be configured to simulate and optimize multiple systems simultaneously. For example, it is possible to simultaneously learn potentials for different temperatures or for different compositions. We demonstrate the approach by recovering simple pair potentials, such as Lennard-Jones systems, from radial distribution functions. We find that DiffSim can be used to probe a wider functional space of pair potentials compared with traditional methods like iterative Boltzmann inversion. We show that our methods can be used to simultaneously fit potentials for simulations at different compositions and temperatures to improve the transferability of the learned potentials. </jats:p> |
| first_indexed | 2024-09-23T07:53:08Z |
| format | Article |
| id | mit-1721.1/150447 |
| institution | Massachusetts Institute of Technology |
| language | English |
| last_indexed | 2024-09-23T07:53:08Z |
| publishDate | 2023 |
| publisher | AIP Publishing |
| record_format | dspace |
| spelling | mit-1721.1/1504472024-01-19T19:05:49Z Learning pair potentials using differentiable simulations Wang, Wujie Wu, Zhenghao Dietschreit, Johannes CB Gómez-Bombarelli, Rafael Massachusetts Institute of Technology. Department of Materials Science and Engineering <jats:p> Learning pair interactions from experimental or simulation data is of great interest for molecular simulations. We propose a general stochastic method for learning pair interactions from data using differentiable simulations (DiffSim). DiffSim defines a loss function based on structural observables, such as the radial distribution function, through molecular dynamics (MD) simulations. The interaction potentials are then learned directly by stochastic gradient descent, using backpropagation to calculate the gradient of the structural loss metric with respect to the interaction potential through the MD simulation. This gradient-based method is flexible and can be configured to simulate and optimize multiple systems simultaneously. For example, it is possible to simultaneously learn potentials for different temperatures or for different compositions. We demonstrate the approach by recovering simple pair potentials, such as Lennard-Jones systems, from radial distribution functions. We find that DiffSim can be used to probe a wider functional space of pair potentials compared with traditional methods like iterative Boltzmann inversion. We show that our methods can be used to simultaneously fit potentials for simulations at different compositions and temperatures to improve the transferability of the learned potentials. </jats:p> 2023-04-06T18:25:51Z 2023-04-06T18:25:51Z 2023-01-28 2023-04-06T18:07:58Z Article http://purl.org/eprint/type/JournalArticle https://hdl.handle.net/1721.1/150447 Wang, Wujie, Wu, Zhenghao, Dietschreit, Johannes CB and Gómez-Bombarelli, Rafael. 2023. "Learning pair potentials using differentiable simulations." The Journal of Chemical Physics, 158 (4). en 10.1063/5.0126475 The Journal of Chemical Physics Creative Commons Attribution 4.0 International license https://creativecommons.org/licenses/by/4.0/ application/pdf AIP Publishing American Institute of Physics (AIP) |
| spellingShingle | Wang, Wujie Wu, Zhenghao Dietschreit, Johannes CB Gómez-Bombarelli, Rafael Learning pair potentials using differentiable simulations |
| title | Learning pair potentials using differentiable simulations |
| title_full | Learning pair potentials using differentiable simulations |
| title_fullStr | Learning pair potentials using differentiable simulations |
| title_full_unstemmed | Learning pair potentials using differentiable simulations |
| title_short | Learning pair potentials using differentiable simulations |
| title_sort | learning pair potentials using differentiable simulations |
| url | https://hdl.handle.net/1721.1/150447 |
| work_keys_str_mv | AT wangwujie learningpairpotentialsusingdifferentiablesimulations AT wuzhenghao learningpairpotentialsusingdifferentiablesimulations AT dietschreitjohannescb learningpairpotentialsusingdifferentiablesimulations AT gomezbombarellirafael learningpairpotentialsusingdifferentiablesimulations |