UC Berkeley

The pursuit of carbon neutrality necessitates improvements in energy storage technologies, with high-performance Li-ion battery cathode materials offering a promising avenue. Modern battery materials can contain many elements with substantial structural complexity, such as configurational disorder that has been shown to be critical for their electrochemical performance. Taking disordered rocksalt cathodes (DRX) as examples, the thesis presents a comprehensive computational modeling study to address the multicomponent complexity through an integrated approach spanning from first-principles calculations to machine learning methods.

The thesis introduces the methodologies required for modeling thermodynamics on lattices from first principles, leveraging density functional theory and lattice models to investigate configurational degrees of freedom. It subsequently demonstrates the application of cluster expansion Monte Carlo simulations to model the short-range order (SRO) in DRXs. The important effect of SRO is demonstrated through the Li-F locking effect in several Mn-based DRXs. A Mg-doping strategy is proposed to increase the capacity by decreasing the amount of Li bound to F. The important role SRO plays in the Li diffusion kinetics is illustrated in several DRX compounds with composition Li$_{x}$Mn$_{0.4}$Nb$_{0.3}$O$_{1.6}$F$_{0.4}$. By quantifying the percolating Li content in the diffusion network, a strategy of introducing cation deficiency is proposed to tune the SRO and improve the high-rate performance.

Subsequent chapters transition to atomistic modeling with charge information, which is crucial for modeling redox reactions and charge transfer phenomena in cathode materials. Two conceptual approaches -- charge-decorated cluster expansion and charge-informed machine learning interatomic potentials -- are introduced. The charge-decorated cluster expansion is applied to study the intercalation chemistry with multi-redox reactions in Li$_{1.3-x}$Mn$_{0.4}$Nb$_{0.3}$O$_{1.6}$F$_{0.4}$, providing a clear demonstration of the Mn and oxygen redox contribution to the redox potential as a function of Li content. The charge-informed interatomic potential is used to study the transition metal migration-induced phase transformation in Li$_{1.1-x}$Mn$_{0.8}$Ti$_{0.1}$O$_{1.9}$F$_{0.1}$ via molecular dynamics. An analysis of structural change and SRO is discussed to reveal the effect on the intercalation chemistry.

Lastly, the thesis introduces a novel approach for directly modeling the electrochemical performance of DRX materials. A comprehensive machine learning model (DRXNet) is introduced as a universal end-to-end model with modular design. A graph neural network and embedding networks are used to encode the chemical and electrochemical conditions, including composition, current density, working voltage window, and battery cycle state. The DRXNet is trained on years of experimental discharge voltage profiles and enables an extensive exploratory search across diverse chemical spaces and test conditions, paving the way for identifying novel cathode materials for next-generation batteries.

By integrating the progress achieved in this thesis with recent advances in the fields of computational physics and AI for Science, the thesis proposes general strategies for advancing computational modeling in energy materials design as future directions.