Development of transferable equivariant graph neural network forcefields for enhanced exploration of molten salt systems

Despite the growing interest in molten salt reactors and thermal storage systems, our understanding of the physicochemical properties of molten salts remains incomplete, partly due to challenges in performing experiments involving extreme temperatures, strict impurity control, and corrosion management, and partly due to the limited length-scale and time-scale of first-principles calculations. In this thesis, a modernized method for fabricating a transferable equivariant graph neural network forcefield for a model molten salt system using minimal DFT simulations is presented. Using this transferable machine learned forcefield, the thermal conductivity, radial distribution function, and self-intermediate scattering function of LiF-NaF was computed at various chemical ratios. Results show compelling agreement with first-principles computations, and the ability to interpolate and extrapolate various chemical ratios.

Name thesis:degree_name

M.S.

Level thesis:degree_level

Thesis

Discipline thesis:degree_discipline

Nuclear, Plasma, Radiolgc Engr

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2024

Author dc:creator

• Murg, Luca

Contributors dc:contributor

• Zhang, Yang
• Vergari, Lorenzo

• Transferable Equivariant Graph Neural Networks Forcefields
• Flina
• Lif
• Naf

Statement dc:rights

• Copyright 2024 Luca Murg

Language dc:language

en, eng

Handle dc:identifier

https://hdl.handle.net/2142/124426

OAI record oai:identifier

https://www.ideals.illinois.edu/oai-pmh?verb=GetRecord&metadataPrefix=etdms&identifier=oai%3Awww.ideals.illinois.edu%3A2142%2F124426