Analyzing and improving electronic structure calculations of catalytic interfaces using density functional theory and machine learning
Author(s)
Sahoo, Sushree Jagriti
Advisor(s)
Editor(s)
Collections
Supplementary to:
Permanent Link
Abstract
Density functional theory (DFT) plays an important role in heterogeneous catalysis by enabling first-principles study of large and periodic systems due to its accuracy and low computational cost. It provides detailed insights into the catalyst activity and selectivity at atomic scale. These calculations require an input for the exchange-correlation (xc) functionals, which accounts for the multi-particle interactions in DFT but its universal form is unknown. The most accurate xc functional typically depends on the type of chemical system, making it challenging to choose a functional for systems that contain interfaces between different phases of matter or where multiple types of chemical bonding are important. Semilocal functionals are typically used to calculate chemisorption energies due to their low cost, but they differ from experimental values by as much as 1 eV, which can lead to quantitatively and qualitatively incorrect conclusions in the analysis of surface reaction systems. In this thesis, we first explore the typical model space of xc functionals: hybrid and generalized gradient approximation (GGA) level functionals in DFT to investigate the convergence of surface properties and electronic gap of rutile titania nanoparticles with particle size. The geometric and electronic finite-size effects in surface energy are deconvoluted and the influence of defects on electronic gap are evaluated. Further, we explore a novel approach for xc functional design using the multipole (MP) descriptor family to describe the local electronic environments in chemical systems. MP descriptors for electron density provide a set of complete, translationally, and 3D-rotationally invariant convolutional descriptors. Utilizing the MP descriptors, we propose a data-driven framework that uses energies from two different levels of theory to predict the gas-phase corrections in heterogeneous catalytic systems. Next, we introduce a new method to construct xc functionals using convolutions of arbitrary kernels with electron density. We derive the variational derivative of these functionals and provide equations for variational derivatives based on MP descriptors from convolutional kernels. A proof-of-concept functional, PBEq which allows a single functional to use different GGAs at different spatial points in a system is implemented. Testing of PBEq functional on small molecules, bulk metals, and surface catalysts suggests that this approach is a promising route to simultaneously optimize multiple properties of interest. Finally, we propose a framework for developing surrogate hybrid functionals using MP descriptors, at the cost of semilocal functionals. This framework highlights the challenges related to satisfaction of physical constraints while linking exact exchange to semilocal functionals. This approach, combined with the derivation of variational derivatives, has the potential to pave the way for new strategies for functional design and integration of self-consistent ML functionals within DFT.
Sponsor
Date
2025-01-16
Extent
Resource Type
Text
Resource Subtype
Dissertation