Main group organometallic compounds are key components in a broad range of chemical reactions, from drug synthesis to fertilizer manufacturing. However, they are characterized by fast equilibria in solution involving the solvent which makes their experimental characterisation extremely challenging.
Computational methods are therefore best suited to unravel their mechanism of action. However, very accurate dynamical techniques are computationally very expensive and thus greatly limit the possibility to study a broad range of different compounds and environments. Machine learned potentials provide a promising solution to substitute DFT counterparts to decrease the computational demand without any appreciable loss in accuracy.
This project aims to develop a suitable machine learned force field for an archetypal selection of organomagnesium and organolithium reagents taking advantage of state-of-the-art GPU accelerated ML codes. Its application in MD simulations will allow to adequately sample the intricated free energy surfaces in which these reagent act and therefore shed light on their mechanism. This requires a significant amount of computational resources in the initial training part but will allow to achieve close to a 1000x speed-up in production simulation.
Universitetet i Oslo, Norway.