Unified graph neural network force-field for the periodic table: Solid state applications

Kamal Choudhary; Brian DeCost; Lily Major; Keith Butler; Jeyan Thiyagalingam; Francesca Tavazza

doi:10.1039/d2dd00096b

Unified graph neural network force-field for the periodic table: Solid state applications

Kamal Choudhary^*, Brian DeCost, Lily Major, Keith Butler, Jeyan Thiyagalingam, Francesca Tavazza

^*Corresponding author for this work

Department of Computer Science

Research output: Contribution to journal › Article › peer-review

32 Citations (Scopus)

39 Downloads (Pure)

Abstract

Classical force fields (FFs) based on machine learning (ML) methods show great potential for large scale simulations of solids. MLFFs have hitherto largely been designed and fitted for specific systems and are not usually transferable to chemistries beyond the specific training set. We develop a unified atomisitic line graph neural network-based FF (ALIGNN-FF) that can model both structurally and chemically diverse solids with any combination of 89 elements from the periodic table. To train the ALIGNN-FF model, we use the JARVIS-DFT dataset which contains around 75 000 materials and 4 million energy-force entries, out of which 307 113 are used in the training. We demonstrate the applicability of this method for fast optimization of atomic structures in the crystallography open database and by predicting accurate crystal structures using a genetic algorithm for alloys.

Original language	English
Pages (from-to)	346-355
Number of pages	10
Journal	Digital Discovery
Volume	2
Issue number	2
DOIs	https://doi.org/10.1039/d2dd00096b https://doi.org/10.6084/m9.figshare.21667874
Publication status	Published - 23 Jan 2023

Access to Document

10.1039/d2dd00096bLicence: CC BY
10.6084/m9.figshare.21667874Licence: CC BY

ArticleFinal published version, 1.06 MBLicence: CC BY

https://jarvis.nist.gov/Licence: Unclear 1

Permanent link

Persistent link

Cite this

@article{469bdc1ed2f1491ba0c48e4017819a83,

title = "Unified graph neural network force-field for the periodic table: Solid state applications",

abstract = "Classical force fields (FFs) based on machine learning (ML) methods show great potential for large scale simulations of solids. MLFFs have hitherto largely been designed and fitted for specific systems and are not usually transferable to chemistries beyond the specific training set. We develop a unified atomisitic line graph neural network-based FF (ALIGNN-FF) that can model both structurally and chemically diverse solids with any combination of 89 elements from the periodic table. To train the ALIGNN-FF model, we use the JARVIS-DFT dataset which contains around 75 000 materials and 4 million energy-force entries, out of which 307 113 are used in the training. We demonstrate the applicability of this method for fast optimization of atomic structures in the crystallography open database and by predicting accurate crystal structures using a genetic algorithm for alloys.",

author = "Kamal Choudhary and Brian DeCost and Lily Major and Keith Butler and Jeyan Thiyagalingam and Francesca Tavazza",

note = "Funding Information: K. C. thanks the National Institute of Standards and Technology for funding and computational support. In addition, K. C thanks Ruth Pachter and Kiet Ngyuen from the Air Force Research Laboratory for helpful discussions. This work was also partially supported by Wave 1 of the UKRI Strategic Priorities Fund under the EPSRC grant EP/T001569/1, particularly the {\textquoteleft}AI for Science{\textquoteright} theme within that grant, by the Alan Turing Institute ad UKRI AnIMLAC CDT, grant no. EP/S023992/1. Publisher Copyright: {\textcopyright} 2023 The Author(s). Published by the Royal Society of Chemistry.",

year = "2023",

month = jan,

day = "23",

doi = "10.1039/d2dd00096b",

language = "English",

volume = "2",

pages = "346--355",

journal = "Digital Discovery",

publisher = "Royal Society of Chemistry",

number = "2",

}

TY - JOUR

T1 - Unified graph neural network force-field for the periodic table

T2 - Solid state applications

AU - Choudhary, Kamal

AU - DeCost, Brian

AU - Major, Lily

AU - Butler, Keith

AU - Thiyagalingam, Jeyan

AU - Tavazza, Francesca

N1 - Funding Information: K. C. thanks the National Institute of Standards and Technology for funding and computational support. In addition, K. C thanks Ruth Pachter and Kiet Ngyuen from the Air Force Research Laboratory for helpful discussions. This work was also partially supported by Wave 1 of the UKRI Strategic Priorities Fund under the EPSRC grant EP/T001569/1, particularly the ‘AI for Science’ theme within that grant, by the Alan Turing Institute ad UKRI AnIMLAC CDT, grant no. EP/S023992/1. Publisher Copyright: © 2023 The Author(s). Published by the Royal Society of Chemistry.

PY - 2023/1/23

Y1 - 2023/1/23

N2 - Classical force fields (FFs) based on machine learning (ML) methods show great potential for large scale simulations of solids. MLFFs have hitherto largely been designed and fitted for specific systems and are not usually transferable to chemistries beyond the specific training set. We develop a unified atomisitic line graph neural network-based FF (ALIGNN-FF) that can model both structurally and chemically diverse solids with any combination of 89 elements from the periodic table. To train the ALIGNN-FF model, we use the JARVIS-DFT dataset which contains around 75 000 materials and 4 million energy-force entries, out of which 307 113 are used in the training. We demonstrate the applicability of this method for fast optimization of atomic structures in the crystallography open database and by predicting accurate crystal structures using a genetic algorithm for alloys.

AB - Classical force fields (FFs) based on machine learning (ML) methods show great potential for large scale simulations of solids. MLFFs have hitherto largely been designed and fitted for specific systems and are not usually transferable to chemistries beyond the specific training set. We develop a unified atomisitic line graph neural network-based FF (ALIGNN-FF) that can model both structurally and chemically diverse solids with any combination of 89 elements from the periodic table. To train the ALIGNN-FF model, we use the JARVIS-DFT dataset which contains around 75 000 materials and 4 million energy-force entries, out of which 307 113 are used in the training. We demonstrate the applicability of this method for fast optimization of atomic structures in the crystallography open database and by predicting accurate crystal structures using a genetic algorithm for alloys.

UR - http://www.scopus.com/inward/record.url?scp=85169305642&partnerID=8YFLogxK

U2 - 10.1039/d2dd00096b

DO - 10.1039/d2dd00096b

M3 - Article

VL - 2

SP - 346

EP - 355

JO - Digital Discovery

JF - Digital Discovery

IS - 2

ER -

Unified graph neural network force-field for the periodic table: Solid state applications

Abstract

Access to Document

Permanent link

Other files and links

Fingerprint

Cite this