At the beginning of my career, I focused on studying the catalytic activity of a ruthenium surface covered with oxygen adatoms to oxidize CO using hot electrons generated by an IR femtosecond laser impinging on the metal surface. My research included examining several coverages [10.1021/acs.jpcc.1c01618], conducting ab initio molecular dynamics simulations to confirm the predicted minimum energy path [10.1021/acs.jpclett.2c02327], and later investigating the influence of surface deformation [10.1021/acs.jpcc.3c01192][10.3389/fchem.2023.1235176]. This research led to i) a deep understanding of oxidative mechanisms and ii) the creation of a database used to develop a machine learning potential [10.1021/jacsau.4c00197]. This MLP allowed precise and long molecular dynamics simulations for the first time, elucidating the behavior observed in femtosecond experiments.

The very exciting field of the generation of machine learning interatomic potentials rapidly trapped me in its net. I became obsessed with creating an accurate potential to model molecular dynamics under various conditions, including solid state and solid/liquid interfaces. This led to numerous questions about the accuracy of describing configurational space and the need for sparse, accurate, and non-redundant sampling. As a result, I became interested in both statistical methods and enhanced sampling techniques.

When starting a career with the first principle method, one learns to be patient and to use small systems to describe larger ones. The need to accurately describe a large system, to perform long-time molecular dynamics, and to explore  a vast configurational space is now supported by machine learning potentials. However, I remain captivated by existing alternatives, such as semi-empirical methods (tight-binding [10.1039/D4SC02355B]) and multi-method simulations (QM/MM). While the ultimate goal is to model macro-systems, I am also interested in microkinetic models. It is exciting to envision creating multiscale modeling that spans from Angstrom-level details using DFT to microreactor simulations utilizing statistical physics, integrating tight-binding methods, MLIP, and QM/MM.