Etudiant : WARDI YOUSSEF

Filière : Master Physique des Nouveaux Matériaux et Energies Renouvelables (PNOMER)

Encadrant : Pr. DERKAOUI ISSAM

Annèe : 2026

Résumé : The sustainable production of green hydrogen via water electrolysis requires the use of abundant, low-cost catalysts to replace scarce platinum, which poses a critical challenge to achieving clean energy and climate goals. This work presents a multi-scale framework accelerated by high-performance computing (HPC), combining high-throughput density functional theory (DFT) HSE06 and machine learning-based interatomic potentials (MLIPs) using graph neural networks (GNNs) to investigate ABO3 perovskites for the hydrogen evolution reaction (HER). More than 500 materials were processed via a scalable DFT+MLIP workflow, 150 fully stable structures were characterized using VASP on a high-performance computing (HPC) cluster. Additionally, the uma-m-1p1 graph neural network (GNN) based on the Hamiltonian was published as an alternative model for calculating hydrogen absorption energies (∆GH) at a cost representing only a fraction of that of DFT, enabling a screening that would have been impossible using DFT alone. The ASC (Activity–Stability–Cost) framework ranked all candidates and identified a shortlist of the top 10 materials. Following this, Phase 3 will train a supervised machine learning model on more than 500 surfaces, and Phase 4 will run kinetic Monte Carlo (KMC) simulations to derive real catalytic rates that bridge theory and experiment for the scalable production of green hydrogen.