Towards routine modelling of condensed phases with the accuracy of quantum Diffusion Monte Carlo

Abstract

Molecular crystals are critical in diverse fields, including organic semiconductors, optoelectronics, medicine, and pharmaceuticals. A primary aim of computational studies in this area is to aid experimental structure determination and predict finite-temperature relative stabilities with quantitative accuracy. However, modelling molecular crystals remains highly challenging due to two key requirements: (i) an accurate description of the zero-temperature electronic potential energy surface (PES), and (ii) a fully anharmonic, quantum statistical treatment of ionic motion. Despite recent advancements addressing these challenges, routine and economically viable modelling of molecular crystals with quantitative agreement to experiments remains elusive. This thesis represents a significant effort toward overcoming these limitations. First, we demonstrate that quantum diffusion Monte Carlo calculations provide a high-accuracy description of a molecular crystal PES. These calculations establish: (i) a reference dataset to benchmark cost-effective electronic structure methods; (ii) insights into competing polymorphism of ice and the water phase diagram; and (iii) quantitative agreement with experiments across a wide range of energy and crystals. Next, we explore development and application of machine learning interatomic potentials, enabling efficient and accurate modelling of thermodynamic properties at finite temperatures. This includes: (i) resolving the long-debated phase behaviour of one-dimensional nano-confined water; and (ii) calculating sublimation enthalpies for organic molecular crystals and pharmaceutical systems, incorporating anharmonicity and nuclear quantum effects. In summary, this thesis presents a robust and comprehensive framework for accurately and efficiently modelling molecular crystals, marking an important step toward routine, reliable computational approaches for drug discovery and beyond.