Modélisation conjointe du silicium, de l’oxygène et de la silice

Abstract

This thesis focuses on the development of interatomic potentials for silicon and silica, followed by a numerical study of amorphous silicon, using said potential. Silicon and silica are fundamental constituents of many constructions materials, such as clays, cement materials, and alkaliactivated polymers. This work proposes a joint interatomic potential description of the silica compound (SiO2) and its elemental components, silicon (Si) and oxygen (O), implicitly handling charge transfer. This combined description and the complexities associated with charge transfer present challenges for traditional methods, which often encounter spatiotemporal limitations and transferability issues. Herein, silica and its elemental components—silicon and oxygen—were used as prototypes to assess the ability of the Moment Tensor Potential (MTP) to simultaneously and accurately describe disjoint configuration spaces while integrating charge transfers. The MTP is based on machine learning. It has shown promise in the non-joint description of materials such as semiconductors, water, and alloys. Initially, a comprehensive database for silica, silicon, and oxygen was generated from thousands of calculations based on density functional theory (DFT). Subsequently, the Moment Tensor Potential was parameterized using this extensive dataset, adopting a unified approach for the disjoint system Si-SiO2-O. The resulting potential was then employed for static calculations and molecular dynamics simulations, with results compared to DFT data, experimental outcomes, and semi-empirical models. The potential was further tested by coupling it with the Activation Relaxation Technique nouveau (ARTn) to model amorphous silicon. The predictions of the unified potential align well with experimental data, demonstrating competitiveness with semi-empirical results for certain properties while surpassing them for others. Furthermore, reaction coordinates are sufficient to describe charge transfers, while a careful selection of the number of model parameters enables the effective simultaneous description of disjoint regions of the configurational space. Following these results, a database was developed and generated for water, as well as for molecules and structures related to silica gel. The water phase serves as a starting point for modeling more complex systems associated with cementitious and clay materials. The computational cost associated with the modeling in this thesis remains modest due to the size of the generated database and the simulations conducted. There exists further potential for cost optimization by developing a concise work plan and refining the codes involved in the potential’s development.

Date	28 Apr 2025
Original language	French
Awarding Institution	École de technologie supérieure
Supervisor	Claudiane Ouellet-Plamondon (Supervisor) & Laurent Karim Beland (Co-supervisor)