Séminaire IMAGE : « Dynamical models in computational chemistry: theory and numerical inference », Hadrien Vroylandt

Nous aurons le plaisir d’écouter Hadrien Vroylandt, post-doctorant au CERMICS (École des Ponts) et à MATHERIALS (INRIA Paris).
Il donnera un séminaire IMAGE le mercredi 16 octobre à 14h00 en salle de séminaire F-200.

Titre : « Dynamical models in computational chemistry: theory and numerical inference »

Résumé :
Kinetics plays a crucial role in chemistry by providing insights into the rates and mechanisms of chemical reactions, which are key to understanding reaction dynamics. In this presentation, I will discuss the process that links atomistic numerical simulations of chemical reactions to the determination of kinetic properties. Several machine learning approaches are being actively developed to facilitate this connection, and I will showcase recent advancements in this area.
Numerical simulations enable the time evolution of high-dimensional systems, and to capture the essence of the reaction behavior, it is common to define collective variables—simplified representations of the complex processes. I will emphasize the identification of optimal collective variables, which involves determining the most relevant directions in high-dimensional space to model system transformations.
Using these collective variables, we can then construct dynamical models to describe the time evolution of the system. Various types of models are explored, particularly those incorporating inertia and memory effects (i.e. autoregressive models). I will present how these dynamical models can be inferred directly from the time series data generated by the simulations, while also addressing the remaining challenges in the inference process.