Theoretical and Machine Learning Modeling and DFT Simulations in Energy-Related Materials and Devices

Abstract

Modern energy-related materials often exhibit complex, multiscale interactions involving strong coupling between mechanical, electronic, and diffusion degrees of freedom. This dissertation addresses key challenges in understanding and predicting the behavior of such systems through an integrated approach that combines analytical modeling, first-principles simulations, experimental interpretation, and machine learning techniques. We begin by investigating solid-state batteries (SSBs) interfaces, where interfacial degradation and dendrite formation limit performance and reliability. A continuum model is developed to describe how strain-induced suppression of diffusivity leads to a self-limiting reaction mechanism. The model unifies previously fragmented insights into a general theoretical framework that explains self-limiting reaction behavior in a wide range of materials and interface types by quantitative modeling works. Large-scale simulations further connect local self-limiting kinetics with experimentally observed degradation morphology patterns. This work broadens the design space for next-generation batteries from the perspective of dynamic stability design, emphasizing interface engineering and transport dynamics coupled with reaction induced local strain-stress field as critical levers for achieving long-term stability and reliability. In the context of high-temperature superconductors, the group’s material-dependent DFT computations reveal strong lattice–charge–magnetic coupling in cuprate superconductors. By analyzing anharmonic phonon modes and charge redistribution patterns, dynamic charge fluxes patterns were extracted that unveils their correlation with superconducting T_c. Building on this insight, my work focuses on constructing a quantum theoretical framework where flux fluctuations—characterized by wavevector oscillation and coherence width—mediate an attractive interaction kernel. Theoretical analysis within the random phase approximation (RPA) leads to a derived scaling law for T_c, showing strong dependence on material-specific anharmonicity and coupling strength. These results suggest that flux-mediated interactions, emerging from tightly coupled lattice–charge–spin dynamics, may provide a pathway toward understanding and enhancing unconventional superconductivity. Next, we study Na-layered oxides and unveil critical structural details in NaxCrO2 electrochemical evolution. Combining experimental XRD with DFT data, we develop a structure optimization framework capable of identifying complex Na-vacancy ordering patterns and explain the existence of Na density wave ordering patterns. The result reveals novel nano-stripe-like vacancy domains under low Na composition and resolves puzzles about charge–discharge asymmetry and metastability, offering new principles for cathode design with improved reversibility and structural stability. Finally, in the field of battery data science, we present a 2D image-based machine learning framework for battery performance prediction. By encoding cycling curves as binary images and training on various models including deep residual networks end-to-end, we demonstrate enhanced predictive accuracy, robustness and interpretability of 2D representation compared to traditional 1D approaches. The work opens new directions for leveraging advanced vision models to electrochemical degradation and lifetime forecasting for batteries. Together, these studies highlight the power of multi-method modeling in revealing hidden mechanisms, guiding material selection, and proposing novel design strategies for energy related materials.