The Library
A machine learned potential for silicon carbide
Tools
Tunstall, Harry Oliver (2022) A machine learned potential for silicon carbide. PhD thesis, University of Warwick.
|
PDF
WRAP_THESIS_Tunstall_2022.pdf - Submitted Version - Requires a PDF viewer. Download (56Mb) | Preview |
Official URL: http://webcat.warwick.ac.uk/record=b3902129
Abstract
Silicon carbide (SiC) is a prototypical material for high temperature, pressure and radiation applications (e.g. aerospace, automotive, electronics, thermoelectric, nuclear etc.) involving complex nanoscopic processes typically inaccessible to experiments such as grain boundary and amorphous structure interactions. To gain insight into the functional properties of e.g. SiC nanostructures, computationally expensive quantum mechanical methods such as density functional theory (DFT) must be employed. This is because until now less computationally demanding methods are almost always not accurate enough. In fact, similar to silicon and carbon alone, various empirical interatomic potentials have been developed for SiC, such as Tersoff, Tersoff screened, or Stillinger-Weber. These potentials are designed to reproduce specific features of the material, at the expense of transferability to a wider range of functional properties.
This project built a general purpose interatomic potential for SiC using machine-learning regression in the form of Gaussian approximation potentials and neural network potentials. Using a DFT dataset of representative configurations akin to previous work in this field, enables accurate large scale simulations into the amorphous structure.
The final potential matched the DFT radial distribution function (RDF) for liquid, and fast quenches to amorphous. The use of metadynamics yielded structures that matched the experimental RDF, and the potential accurately predicted the DFT energy for these configurations. Whereas, the existing potentials are less accurate at reproducing the amorphous structure.
The methodology for the creation from the ground up, and maintenance of a machine learning database for atomic systems will be discussed, as this is the key to any future work into other systems.
Item Type: | Thesis (PhD) | ||||
---|---|---|---|---|---|
Subjects: | Q Science > Q Science (General) Q Science > QA Mathematics Q Science > QD Chemistry |
||||
Library of Congress Subject Headings (LCSH): | Silicon carbide, Machine learning, Density functionals, Neural networks (Computer science), Gaussian distribution, Thermodynamic potentials | ||||
Official Date: | September 2022 | ||||
Dates: |
|
||||
Institution: | University of Warwick | ||||
Theses Department: | Department of Chemistry | ||||
Thesis Type: | PhD | ||||
Publication Status: | Unpublished | ||||
Supervisor(s)/Advisor: | Sosso, Gabriele ; Kermode, James R. | ||||
Format of File: | |||||
Extent: | x, 144 pages : illustrations, charts | ||||
Language: | eng |
Request changes or add full text files to a record
Repository staff actions (login required)
View Item |