Accurate machine learning force fields via experimental and simulation data fusion
Abstract Machine Learning (ML)-based force fields are attracting ever-increasing interest due to their capacity to span spatiotemporal scales of classical interatomic potentials at quantum-level accuracy. They can be trained based on high-fidelity simulations or experiments, the former being the com...
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Nature Portfolio
2024-04-01
|
| Series: | npj Computational Materials |
| Online Access: | https://doi.org/10.1038/s41524-024-01251-4 |
| _version_ | 1797219679461179392 |
|---|---|
| author | Sebastien Röcken Julija Zavadlav |
| author_facet | Sebastien Röcken Julija Zavadlav |
| author_sort | Sebastien Röcken |
| collection | DOAJ |
| description | Abstract Machine Learning (ML)-based force fields are attracting ever-increasing interest due to their capacity to span spatiotemporal scales of classical interatomic potentials at quantum-level accuracy. They can be trained based on high-fidelity simulations or experiments, the former being the common case. However, both approaches are impaired by scarce and erroneous data resulting in models that either do not agree with well-known experimental observations or are under-constrained and only reproduce some properties. Here we leverage both Density Functional Theory (DFT) calculations and experimentally measured mechanical properties and lattice parameters to train an ML potential of titanium. We demonstrate that the fused data learning strategy can concurrently satisfy all target objectives, thus resulting in a molecular model of higher accuracy compared to the models trained with a single data source. The inaccuracies of DFT functionals at target experimental properties were corrected, while the investigated off-target properties were affected only mildly and mostly positively. Our approach is applicable to any material and can serve as a general strategy to obtain highly accurate ML potentials. |
| first_indexed | 2024-04-24T12:37:29Z |
| format | Article |
| id | doaj.art-97b4414b3a964afd9b80b5ea9b560376 |
| institution | Directory Open Access Journal |
| issn | 2057-3960 |
| language | English |
| last_indexed | 2024-04-24T12:37:29Z |
| publishDate | 2024-04-01 |
| publisher | Nature Portfolio |
| record_format | Article |
| series | npj Computational Materials |
| spelling | doaj.art-97b4414b3a964afd9b80b5ea9b5603762024-04-07T11:24:49ZengNature Portfolionpj Computational Materials2057-39602024-04-0110111010.1038/s41524-024-01251-4Accurate machine learning force fields via experimental and simulation data fusionSebastien Röcken0Julija Zavadlav1Department of Engineering Physics and Computation, Multiscale Modeling of Fluid Materials, TUM School of Engineering and Design, Technical University of MunichDepartment of Engineering Physics and Computation, Multiscale Modeling of Fluid Materials, TUM School of Engineering and Design, Technical University of MunichAbstract Machine Learning (ML)-based force fields are attracting ever-increasing interest due to their capacity to span spatiotemporal scales of classical interatomic potentials at quantum-level accuracy. They can be trained based on high-fidelity simulations or experiments, the former being the common case. However, both approaches are impaired by scarce and erroneous data resulting in models that either do not agree with well-known experimental observations or are under-constrained and only reproduce some properties. Here we leverage both Density Functional Theory (DFT) calculations and experimentally measured mechanical properties and lattice parameters to train an ML potential of titanium. We demonstrate that the fused data learning strategy can concurrently satisfy all target objectives, thus resulting in a molecular model of higher accuracy compared to the models trained with a single data source. The inaccuracies of DFT functionals at target experimental properties were corrected, while the investigated off-target properties were affected only mildly and mostly positively. Our approach is applicable to any material and can serve as a general strategy to obtain highly accurate ML potentials.https://doi.org/10.1038/s41524-024-01251-4 |
| spellingShingle | Sebastien Röcken Julija Zavadlav Accurate machine learning force fields via experimental and simulation data fusion npj Computational Materials |
| title | Accurate machine learning force fields via experimental and simulation data fusion |
| title_full | Accurate machine learning force fields via experimental and simulation data fusion |
| title_fullStr | Accurate machine learning force fields via experimental and simulation data fusion |
| title_full_unstemmed | Accurate machine learning force fields via experimental and simulation data fusion |
| title_short | Accurate machine learning force fields via experimental and simulation data fusion |
| title_sort | accurate machine learning force fields via experimental and simulation data fusion |
| url | https://doi.org/10.1038/s41524-024-01251-4 |
| work_keys_str_mv | AT sebastienrocken accuratemachinelearningforcefieldsviaexperimentalandsimulationdatafusion AT julijazavadlav accuratemachinelearningforcefieldsviaexperimentalandsimulationdatafusion |