Thermodynamics and dynamics of non-compact prismatic dislocation loops simulated using a machine-learning model

Abstract

We explore how the thermodynamic properties and dynamics of a self-interstitial prismatic dislocation loop are affected by microscopic-scale variations in its geometric configuration, an aspect that rarely received attention in literature. We develop a machine-learning (ML) model to predict the formation energy of an arbitrary geometrically complex configuration of a self-interstitial atom dislocation loop. Trained on atomistic simulation data, the ML model achieves high predictive accuracy across a broad range of configurations, with a typical error in the 1% range. From the ML model, we evaluate the density of configurational microstates as a function of loop's formation energy and derive analytical expressions valid in tractable limiting cases. Using statistical mechanics, we derive the configurational free energy, the average energy, and the thermodynamic entropy of a dislocation loop as a function of temperature. We also simulate the dynamics of self-climb of dislocation loops with various geometries and evaluate their diffusion coefficients and effective activation energy for diffusion. Analysis shows that there is a single universal parameter describing the morphological irregularity of the loop configuration in its ground state. This parameter determines the thermodynamic properties of a loop as well as its dynamics, and simulations illustrate how the properties and mobility of a configurationally complex loop vary as functions of the irregularity parameter.