First-principles and machine learning investigations into the atomic and mechanical properties of cement hydrates

Concrete is the second most used substance in the world after water, with more than 35 billion tonnes produced annually. Yet, understanding the atomic and mechanical properties of its principal component, calcium-silicate-hydrate (C-S-H)—the complex binder phase of concrete—remains a considerable challenge owing to its structural complexity and disordered nature. This work aims to develop and validate a machine learning force field (MLFF) capable of accurately reproducing the structural and mechanical behaviour of C-S-H. To this end, density functional theory (DFT) calculations were employed to study the electronic structure, bonding characteristics, and elastic response of C-S-H at the atomic level. Subsequently, an on-the-fly MLFF was trained using ab initio molecular dynamics (AIMD) simulations at 400 K. After a thorough refitting and validation process, the resulting MLFFs were assessed through structure relaxation and molecular dynamics simulations, yielding energy, force, and stress tensor predictions comparable to accurate DFT results. The MLFFs were then used to compute key thermodynamic and mechanical properties, including the equation of state (EOS) and bulk modulus, obtaining values between 55–58 GPa, in favourable agreement with available experimental data. Furthermore, the transferability of the MLFF was evaluated and confirmed across a temperature range of 200–400 K, maintaining physical consistency throughout.

• 1. Verfasser: Balarezo Balarezo, Juan Gabriel (author)
• Format: bachelorThesis
• Veröffentlicht: 2025
• Schlagworte: Teoría del Funcional de Densidad; Aprendizaje automático; Campos de fuerza; Density Functional Theory; Machine learning; Force fields
• Online Zugang: https://repositorio.yachaytech.edu.ec/handle/123456789/1025