Ab Initio & Machine Learning Combined Study of the Lattice Thermal Conductivity of of Boron Arsenide including Anharmonic Effects

Efficient thermal management has become a primary bottleneck in the development of high-performance power electronics and next-generation microchips. Boron Arsenide (BAs) has emerged as a landmark material in this field due to its exceptional room-temperature thermal conductivity ($> 1000$ W/mK), which rivals that of diamond and significantly exceeds that of traditional semiconductors like Silicon or Gallium Nitride. This thesis provides a comprehensive theoretical and computational investigation of the vibrational and thermal transport properties of BAs, bridging the gap between standard ab initio methods and advanced non-perturbative treatments of anharmonicity. The research begins with a fundamental characterization of the system using Density Functional Theory (DFT). Systematic convergence tests for k-point sampling and energy cutoffs were conducted to establish a reliable ground-state reference, followed by the optimization of the lattice parameter. Electronic band structures and the density of states (DOS) were computed, providing the necessary framework for phonon dispersion analysis in the harmonic approximation. However, the extraordinary thermal conductivity of BAs is governed by unique phonon-phonon scattering processes that the standard harmonic approximation cannot fully describe. To address this, this work explores the integration of a Machine Learning Potential (MLP), specifically trained to capture the complex, high-order anharmonicity of the potential energy surface. The reliability and transferability of the MLP were validated through Molecular Dynamics (MD) simulations across NVE, NVT, and NPT ensembles. Structural and thermodynamic properties, including the Radial Distribution Function (RDF) and bulk modulus, were analyzed to ensure consistency with experimental and ab initio benchmarks. The core research applies the Stochastic Self-Consistent Harmonic Approximation (SSCHA) to investigate temperature-dependent phonon renormalization. Simulations were conducted across a broad thermal range from 200 K to 1200 K, covering both cryogenic and extreme operating conditions. By analyzing the Auxiliary Force Constants ($\boldsymbol{\Phi}$) and atomic centroids, we demonstrate the structural stability of the BAs lattice under significant thermal fluctuations.Furthermore, by applying the Single Relaxation Time Approximation (SRTA) within the Boltzmann Transport Equation (BTE), phonon lifetimes and thermal conductivity were evaluated. The comparison between SSCHA and standard perturbative methods highlights the necessity of non-perturbative treatments for materials with strong intrinsic anharmonicity. These findings validate the developed Machine Learning Potential and confirm BAs as a superior platform for advanced passive cooling, providing a robust workflow for the design of high-conductivity materials.