Learning the exchange-correlation functional from nature with fully differentiable density functional theory
Improving the predictive capability of molecular properties in ab initio simulations is essential for advanced material discovery. Despite recent progress making use of machine learning, utilizing deep neural networks to improve quantum chemistry modeling remains severely limited by the scarcity and...
| Main Authors: | , |
|---|---|
| Format: | Journal article |
| Language: | English |
| Published: |
American Physical Society
2021
|
| _version_ | 1826307602656526336 |
|---|---|
| author | Kasim, MF Vinko, SM |
| author_facet | Kasim, MF Vinko, SM |
| author_sort | Kasim, MF |
| collection | OXFORD |
| description | Improving the predictive capability of molecular properties in ab initio simulations is essential for advanced material discovery. Despite recent progress making use of machine learning, utilizing deep neural networks to improve quantum chemistry modeling remains severely limited by the scarcity and heterogeneity of appropriate experimental data. Here we show how training a neural network to replace the exchange-correlation functional within a fully differentiable three-dimensional Kohn-Sham density functional theory framework can greatly improve simulation accuracy. Using only eight experimental data points on diatomic molecules, our trained exchange-correlation networks enable improved prediction accuracy of atomization energies across a collection of 104 molecules containing new bonds, and atoms, that are not present in the training dataset. |
| first_indexed | 2024-03-07T07:05:34Z |
| format | Journal article |
| id | oxford-uuid:f942d4ad-8997-4964-8a0c-0ab551ca5b61 |
| institution | University of Oxford |
| language | English |
| last_indexed | 2024-03-07T07:05:34Z |
| publishDate | 2021 |
| publisher | American Physical Society |
| record_format | dspace |
| spelling | oxford-uuid:f942d4ad-8997-4964-8a0c-0ab551ca5b612022-05-03T10:30:02ZLearning the exchange-correlation functional from nature with fully differentiable density functional theoryJournal articlehttp://purl.org/coar/resource_type/c_dcae04bcuuid:f942d4ad-8997-4964-8a0c-0ab551ca5b61EnglishSymplectic ElementsAmerican Physical Society2021Kasim, MFVinko, SMImproving the predictive capability of molecular properties in ab initio simulations is essential for advanced material discovery. Despite recent progress making use of machine learning, utilizing deep neural networks to improve quantum chemistry modeling remains severely limited by the scarcity and heterogeneity of appropriate experimental data. Here we show how training a neural network to replace the exchange-correlation functional within a fully differentiable three-dimensional Kohn-Sham density functional theory framework can greatly improve simulation accuracy. Using only eight experimental data points on diatomic molecules, our trained exchange-correlation networks enable improved prediction accuracy of atomization energies across a collection of 104 molecules containing new bonds, and atoms, that are not present in the training dataset. |
| spellingShingle | Kasim, MF Vinko, SM Learning the exchange-correlation functional from nature with fully differentiable density functional theory |
| title | Learning the exchange-correlation functional from nature with fully differentiable density functional theory |
| title_full | Learning the exchange-correlation functional from nature with fully differentiable density functional theory |
| title_fullStr | Learning the exchange-correlation functional from nature with fully differentiable density functional theory |
| title_full_unstemmed | Learning the exchange-correlation functional from nature with fully differentiable density functional theory |
| title_short | Learning the exchange-correlation functional from nature with fully differentiable density functional theory |
| title_sort | learning the exchange correlation functional from nature with fully differentiable density functional theory |
| work_keys_str_mv | AT kasimmf learningtheexchangecorrelationfunctionalfromnaturewithfullydifferentiabledensityfunctionaltheory AT vinkosm learningtheexchangecorrelationfunctionalfromnaturewithfullydifferentiabledensityfunctionaltheory |