The investigation of phonon lifetime and thermal transport mechanisms from Molecular Dynamics

Phonon-phonon interactions give many materials extended properties that they
otherwise would not have had without them. Phonons are quasiparticles of
discrete, quantised vibrations. They are carriers of momentum, which makes
them relevant to the transport of energy, particularly heat. For example, many
thermoelectric and superconducting properties are due to electron-phonon and
phonon-phonon scattering. Understanding these interactions requires complex
theoretical frameworks, which are computationally expensive.
The aim of this thesis is to test Python-based tools for direct analysis of phonon
lifetimes in a range of materials from Molecular Dynamics simulations, and
to improve the efficiency by using Machine Learning (GAP) to accelerate the
Molecular Dynamics to directly calculate phonon lifetimes.
In this thesis, current ab initio methods establish the theoretical framework
behind the calculation of phonon lifetimes, which are traditionally performed
through perturbative methods. New implemented methods for extracting
phonon lifetimes and phonons at finite temperatures through Molecular Dynamics using several pre-existing methodologies have been used to extract
phonon lifetimes. However, at high or low T they become disadvantageous
and have been circumvented by combining some of these methodologies.
Through the combined use of Hybrid-MD and GAP, this thesis explores various
methodologies and establishes a workflow for developing GAP potentials
tailored for phonon calculations in MD. We also demonstrate that the GAP
potential framework inherently misses crucial physics such as LO/TO splitting
when applied to strongly ionic systems at finite T. Having extracted the phonon
band structure at a range of finite T we have shown it is possible to get the
lifetime.