Understanding Structure Thermodynamics And Dynamics Of Silica Liquids And Glasses Using Atomistic Simulations And Machine Learning

Download or read book Understanding Structure, Thermodynamics, and Dynamics of Silica Liquids and Glasses Using Atomistic Simulations and Machine Learning written by Zheng Yu and published by . This book was released on 2022 with total page 0 pages. Available in PDF, EPUB and Kindle. Book excerpt: Glass materials are found all around us, but fundamental questions about the nature of their amorphous structures and formation processes remain unsolved after decades of research. In this dissertation, silica, an archetype glass former, is taken as an example glass for investigations into glass structures, thermodynamics, dynamics, two-level systems (TLS), and atomic interactions based on molecular dynamics simulations and machine learning methods. Specifically, we investigate the structure-thermodynamic stability relationship using a library of silica inherent structures generated from melt-quench and replica exchange molecular dynamics simulations. Based on machine learning, we find that short-range and medium-range features play very different roles on the glass stability across the liquid and glass regions. We then revisit an interesting dynamical transition in silica liquid, the fragile-to-strong transition (FTS), from the perspective of microscopic dynamics. By machine learning to classify atomic rearrangements, the FTS is found to originate from the two types of energy barriers in silica, representing a fast and a slow microscopic dynamics channel. The fast channel controlled by the short-range defects closes rapidly with decreasing temperature, causing the fragility crossover. A similar approach is also applied to investigate TLS. We predict TLS densities in a large number of inherent structures with a variety of glass stability using machine learning and verify them using molecular dynamics simulations. We find a decrease in the TLS density with the fictive temperature, which can be described by a quadratic function as suggested by the random first-order transition theory. Lastly, we introduce a linear machine learning force matching approach that can directly extract pair atomic interactions from ab initio calculations in amorphous materials. This approach is applied to silica to understand the atomic interactions within its structure and develop a new classical force field. Through the comprehensive fundamental investigations on the nature of silica glass and liquid, I hope the understandings and methods presented in this dissertation can be transferred to study other glass-forming systems.