Towards a transferable fermionic neural wavefunction for molecules
Abstract Deep neural networks have become a highly accurate and powerful wavefunction ansatz in combination with variational Monte Carlo methods for solving the electronic Schrödinger equation. However, despite their success and favorable scaling, these methods are still computationally too costly f...
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Nature Portfolio
2024-01-01
|
| Series: | Nature Communications |
| Online Access: | https://doi.org/10.1038/s41467-023-44216-9 |
| _version_ | 1797363133593944064 |
|---|---|
| author | Michael Scherbela Leon Gerard Philipp Grohs |
| author_facet | Michael Scherbela Leon Gerard Philipp Grohs |
| author_sort | Michael Scherbela |
| collection | DOAJ |
| description | Abstract Deep neural networks have become a highly accurate and powerful wavefunction ansatz in combination with variational Monte Carlo methods for solving the electronic Schrödinger equation. However, despite their success and favorable scaling, these methods are still computationally too costly for wide adoption. A significant obstacle is the requirement to optimize the wavefunction from scratch for each new system, thus requiring long optimization. In this work, we propose a neural network ansatz, which effectively maps uncorrelated, computationally cheap Hartree-Fock orbitals, to correlated, high-accuracy neural network orbitals. This ansatz is inherently capable of learning a single wavefunction across multiple compounds and geometries, as we demonstrate by successfully transferring a wavefunction model pre-trained on smaller fragments to larger compounds. Furthermore, we provide ample experimental evidence to support the idea that extensive pre-training of such a generalized wavefunction model across different compounds and geometries could lead to a foundation wavefunction model. Such a model could yield high-accuracy ab-initio energies using only minimal computational effort for fine-tuning and evaluation of observables. |
| first_indexed | 2024-03-08T16:16:55Z |
| format | Article |
| id | doaj.art-fd5c1b3624c04040bef7b41aa201b45a |
| institution | Directory Open Access Journal |
| issn | 2041-1723 |
| language | English |
| last_indexed | 2024-03-08T16:16:55Z |
| publishDate | 2024-01-01 |
| publisher | Nature Portfolio |
| record_format | Article |
| series | Nature Communications |
| spelling | doaj.art-fd5c1b3624c04040bef7b41aa201b45a2024-01-07T12:35:33ZengNature PortfolioNature Communications2041-17232024-01-0115111210.1038/s41467-023-44216-9Towards a transferable fermionic neural wavefunction for moleculesMichael Scherbela0Leon Gerard1Philipp Grohs2Faculty of Mathematics, University of ViennaResearch Network Data Science, University of ViennaFaculty of Mathematics, University of ViennaAbstract Deep neural networks have become a highly accurate and powerful wavefunction ansatz in combination with variational Monte Carlo methods for solving the electronic Schrödinger equation. However, despite their success and favorable scaling, these methods are still computationally too costly for wide adoption. A significant obstacle is the requirement to optimize the wavefunction from scratch for each new system, thus requiring long optimization. In this work, we propose a neural network ansatz, which effectively maps uncorrelated, computationally cheap Hartree-Fock orbitals, to correlated, high-accuracy neural network orbitals. This ansatz is inherently capable of learning a single wavefunction across multiple compounds and geometries, as we demonstrate by successfully transferring a wavefunction model pre-trained on smaller fragments to larger compounds. Furthermore, we provide ample experimental evidence to support the idea that extensive pre-training of such a generalized wavefunction model across different compounds and geometries could lead to a foundation wavefunction model. Such a model could yield high-accuracy ab-initio energies using only minimal computational effort for fine-tuning and evaluation of observables.https://doi.org/10.1038/s41467-023-44216-9 |
| spellingShingle | Michael Scherbela Leon Gerard Philipp Grohs Towards a transferable fermionic neural wavefunction for molecules Nature Communications |
| title | Towards a transferable fermionic neural wavefunction for molecules |
| title_full | Towards a transferable fermionic neural wavefunction for molecules |
| title_fullStr | Towards a transferable fermionic neural wavefunction for molecules |
| title_full_unstemmed | Towards a transferable fermionic neural wavefunction for molecules |
| title_short | Towards a transferable fermionic neural wavefunction for molecules |
| title_sort | towards a transferable fermionic neural wavefunction for molecules |
| url | https://doi.org/10.1038/s41467-023-44216-9 |
| work_keys_str_mv | AT michaelscherbela towardsatransferablefermionicneuralwavefunctionformolecules AT leongerard towardsatransferablefermionicneuralwavefunctionformolecules AT philippgrohs towardsatransferablefermionicneuralwavefunctionformolecules |