Data-driven approaches from ab initio methods in condensed matter to climate science

Munoz, Alexander Reed

Permalink

https://hdl.handle.net/2142/121398 Copy

Description

Title

Data-driven approaches from ab initio methods in condensed matter to climate science

Author(s)

Munoz, Alexander Reed

Issue Date

2023-05-19

Director of Research (if dissertation) or Advisor (if thesis)

Wagner, Lucas K.

Doctoral Committee Chair(s)

Ceperley, David

Committee Member(s)

• Mahmood, Fahad
• Lorenz, Virginia

Department of Study

Physics

Discipline

Physics

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

Ph.D.

Degree Level

Dissertation

Keyword(s)

• physics
• condensed matter
• machine learning
• climate science
• paleoclimate

Abstract

First–principles calculations represent a collection of methods for predicting the properties of quantum mechanical systems with many particles. First–principles approaches to condensed matter are embodied by the development of computational techniques driven by physical knowledge. As such, this dissertation covers ab initio computation, the use of machine learning techniques in climate science, and the use of machine learning for the extension of ab initio computation. Each chapter demonstrates the utility of computation in solving physical problems, and how physical feedback informs the design of computational protocols. The direct application of ab initio techniques to many–body systems is demonstrated by my work on the prediction of elastic neutron scattering experiments. I designed a machine learning workflow for climate science that creates a global isotopic dataset, identifies regions with distinct oceanographic and atmospheric processes, and measures the importance of the processes that determine the isotope’s relationship to water salinity. My ongoing work has been engineering data sets and designing equivariant neural network architectures for the application of machine learning to ab initio electron densities. My work has been focused on the use of physical insights to improve computational models for condensed matter and climate science. The physically informed approach to machine learning accelerates the solution of physical problems and can offer insights into the interpretation of physical systems.

Graduation Semester

2023-08

Type of Resource

Thesis

Copyright and License Information

Copyright 2023 Alexander Munoz

Owning Collections