Crystal graph attention networks for the prediction of stable materials
(2021) In Science Advances 7(49).- Abstract
- Graph neural networks for crystal structures typically use the atomic positions and the atomic species as input. Unfortunately, this information is not available when predicting new materials, for which the precise geometrical information is unknown. We circumvent this problem by replacing the precise bond distances with embeddings of graph distances. This allows our networks to be applied directly in high-throughput studies based on both composition and crystal structure prototype without using relaxed structures as input. To train these networks, we curate a dataset of over 2 million density functional calculations of crystals with consistent calculation param-eters. We apply the resulting model to the high-throughput search of 15... (More)
- Graph neural networks for crystal structures typically use the atomic positions and the atomic species as input. Unfortunately, this information is not available when predicting new materials, for which the precise geometrical information is unknown. We circumvent this problem by replacing the precise bond distances with embeddings of graph distances. This allows our networks to be applied directly in high-throughput studies based on both composition and crystal structure prototype without using relaxed structures as input. To train these networks, we curate a dataset of over 2 million density functional calculations of crystals with consistent calculation param-eters. We apply the resulting model to the high-throughput search of 15 million tetragonal perovskites of compo-sition ABCD2. As a result, we identify several thousand potentially stable compounds and demonstrate that transfer learning from the newly curated dataset reduces the required training data by 50%. (Less)
Please use this url to cite or link to this publication:
https://lup.lub.lu.se/record/52b2378b-46b5-4c38-a3b7-e77972f7cf00
- author
- Schmidt, Jonathan ; university, Love ; Verdozzi, Claudio LU ; Botti, Silvana and Marques, Miguel A. L.
- organization
- publishing date
- 2021-12-03
- type
- Contribution to journal
- publication status
- published
- subject
- in
- Science Advances
- volume
- 7
- issue
- 49
- article number
- 7948
- pages
- 11 pages
- publisher
- American Association for the Advancement of Science (AAAS)
- external identifiers
-
- pmid:34860548
- scopus:85120707749
- ISSN
- 2375-2548
- DOI
- 10.1126/sciadv.abi7948
- language
- English
- LU publication?
- yes
- id
- 52b2378b-46b5-4c38-a3b7-e77972f7cf00
- date added to LUP
- 2021-12-06 20:14:47
- date last changed
- 2022-04-27 06:22:18
@article{52b2378b-46b5-4c38-a3b7-e77972f7cf00,
abstract = {{Graph neural networks for crystal structures typically use the atomic positions and the atomic species as input. Unfortunately, this information is not available when predicting new materials, for which the precise geometrical information is unknown. We circumvent this problem by replacing the precise bond distances with embeddings of graph distances. This allows our networks to be applied directly in high-throughput studies based on both composition and crystal structure prototype without using relaxed structures as input. To train these networks, we curate a dataset of over 2 million density functional calculations of crystals with consistent calculation param-eters. We apply the resulting model to the high-throughput search of 15 million tetragonal perovskites of compo-sition ABCD2. As a result, we identify several thousand potentially stable compounds and demonstrate that transfer learning from the newly curated dataset reduces the required training data by 50%.}},
author = {{Schmidt, Jonathan and university, Love and Verdozzi, Claudio and Botti, Silvana and Marques, Miguel A. L.}},
issn = {{2375-2548}},
language = {{eng}},
month = {{12}},
number = {{49}},
publisher = {{American Association for the Advancement of Science (AAAS)}},
series = {{Science Advances}},
title = {{Crystal graph attention networks for the prediction of stable materials}},
url = {{http://dx.doi.org/10.1126/sciadv.abi7948}},
doi = {{10.1126/sciadv.abi7948}},
volume = {{7}},
year = {{2021}},
}