Accelerating superconductor discovery through tempered deep learning of the electron-phonon spectral function
Abstract Integrating deep learning with the search for new electron-phonon superconductors represents a burgeoning field of research, where the primary challenge lies in the computational intensity of calculating the electron-phonon spectral function, α 2 F(ω), the essential ingredient of Midgal-Eli...
Saved in:
| Main Authors: | , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Nature Portfolio
2025-01-01
|
| Series: | npj Computational Materials |
| Online Access: | https://doi.org/10.1038/s41524-024-01475-4 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1841544434157617152 |
|---|---|
| author | Jason B. Gibson Ajinkya C. Hire Philip M. Dee Oscar Barrera Benjamin Geisler Peter J. Hirschfeld Richard G. Hennig |
| author_facet | Jason B. Gibson Ajinkya C. Hire Philip M. Dee Oscar Barrera Benjamin Geisler Peter J. Hirschfeld Richard G. Hennig |
| author_sort | Jason B. Gibson |
| collection | DOAJ |
| description | Abstract Integrating deep learning with the search for new electron-phonon superconductors represents a burgeoning field of research, where the primary challenge lies in the computational intensity of calculating the electron-phonon spectral function, α 2 F(ω), the essential ingredient of Midgal-Eliashberg theory of superconductivity. To overcome this challenge, we adopt a two-step approach. First, we compute α 2 F(ω) for 818 dynamically stable materials. We then train a deep-learning model to predict α 2 F(ω), using a training strategy tailored for limited data to temper the model’s overfitting, enhancing predictions. Specifically, we train a Bootstrapped Ensemble of Tempered Equivariant graph neural NETworks (BETE-NET), obtaining an MAE of 0.21, 45 K, and 43 K for the moments derived from α 2 F(ω): λ, $${\omega }_{\log }$$ ω log , and ω 2, respectively, yielding an MAE of 2.5 K for the critical temperature, T c . Further, we incorporate domain knowledge of the site-projected phonon density of states to impose inductive bias into the model’s node attributes and enhance predictions. This methodological innovation decreases the MAE to 0.18, 29 K, and 28 K, respectively, yielding an MAE of 2.1 K for T c . We illustrate the practical application of our model in high-throughput screening for high-T c materials. The model demonstrates an average precision nearly five times higher than random screening, highlighting the potential of ML in accelerating superconductor discovery. BETE-NET accelerates the search for high-T c superconductors while setting a precedent for applying ML in materials discovery, particularly when data is limited. |
| format | Article |
| id | doaj-art-615f817d696543cc948b93922897e12d |
| institution | Kabale University |
| issn | 2057-3960 |
| language | English |
| publishDate | 2025-01-01 |
| publisher | Nature Portfolio |
| record_format | Article |
| series | npj Computational Materials |
| spelling | doaj-art-615f817d696543cc948b93922897e12d2025-01-12T12:32:16ZengNature Portfolionpj Computational Materials2057-39602025-01-0111111010.1038/s41524-024-01475-4Accelerating superconductor discovery through tempered deep learning of the electron-phonon spectral functionJason B. Gibson0Ajinkya C. Hire1Philip M. Dee2Oscar Barrera3Benjamin Geisler4Peter J. Hirschfeld5Richard G. Hennig6Department of Materials Science and Engineering, University of FloridaDepartment of Materials Science and Engineering, University of FloridaDepartment of Materials Science and Engineering, University of FloridaDepartment of Physics, University of FloridaDepartment of Materials Science and Engineering, University of FloridaDepartment of Physics, University of FloridaDepartment of Materials Science and Engineering, University of FloridaAbstract Integrating deep learning with the search for new electron-phonon superconductors represents a burgeoning field of research, where the primary challenge lies in the computational intensity of calculating the electron-phonon spectral function, α 2 F(ω), the essential ingredient of Midgal-Eliashberg theory of superconductivity. To overcome this challenge, we adopt a two-step approach. First, we compute α 2 F(ω) for 818 dynamically stable materials. We then train a deep-learning model to predict α 2 F(ω), using a training strategy tailored for limited data to temper the model’s overfitting, enhancing predictions. Specifically, we train a Bootstrapped Ensemble of Tempered Equivariant graph neural NETworks (BETE-NET), obtaining an MAE of 0.21, 45 K, and 43 K for the moments derived from α 2 F(ω): λ, $${\omega }_{\log }$$ ω log , and ω 2, respectively, yielding an MAE of 2.5 K for the critical temperature, T c . Further, we incorporate domain knowledge of the site-projected phonon density of states to impose inductive bias into the model’s node attributes and enhance predictions. This methodological innovation decreases the MAE to 0.18, 29 K, and 28 K, respectively, yielding an MAE of 2.1 K for T c . We illustrate the practical application of our model in high-throughput screening for high-T c materials. The model demonstrates an average precision nearly five times higher than random screening, highlighting the potential of ML in accelerating superconductor discovery. BETE-NET accelerates the search for high-T c superconductors while setting a precedent for applying ML in materials discovery, particularly when data is limited.https://doi.org/10.1038/s41524-024-01475-4 |
| spellingShingle | Jason B. Gibson Ajinkya C. Hire Philip M. Dee Oscar Barrera Benjamin Geisler Peter J. Hirschfeld Richard G. Hennig Accelerating superconductor discovery through tempered deep learning of the electron-phonon spectral function npj Computational Materials |
| title | Accelerating superconductor discovery through tempered deep learning of the electron-phonon spectral function |
| title_full | Accelerating superconductor discovery through tempered deep learning of the electron-phonon spectral function |
| title_fullStr | Accelerating superconductor discovery through tempered deep learning of the electron-phonon spectral function |
| title_full_unstemmed | Accelerating superconductor discovery through tempered deep learning of the electron-phonon spectral function |
| title_short | Accelerating superconductor discovery through tempered deep learning of the electron-phonon spectral function |
| title_sort | accelerating superconductor discovery through tempered deep learning of the electron phonon spectral function |
| url | https://doi.org/10.1038/s41524-024-01475-4 |
| work_keys_str_mv | AT jasonbgibson acceleratingsuperconductordiscoverythroughtempereddeeplearningoftheelectronphononspectralfunction AT ajinkyachire acceleratingsuperconductordiscoverythroughtempereddeeplearningoftheelectronphononspectralfunction AT philipmdee acceleratingsuperconductordiscoverythroughtempereddeeplearningoftheelectronphononspectralfunction AT oscarbarrera acceleratingsuperconductordiscoverythroughtempereddeeplearningoftheelectronphononspectralfunction AT benjamingeisler acceleratingsuperconductordiscoverythroughtempereddeeplearningoftheelectronphononspectralfunction AT peterjhirschfeld acceleratingsuperconductordiscoverythroughtempereddeeplearningoftheelectronphononspectralfunction AT richardghennig acceleratingsuperconductordiscoverythroughtempereddeeplearningoftheelectronphononspectralfunction |