The study relies on the simulation of reactive species in explicit solvent, hence the initial choice of DFT molecular dynamics. Moreover, the latest development in machine learning interatomic potentials (MLIPs) shows promising results in efficiently and effectively simulating all-atom systems in solution. Therefore, we chose to start with the development of an accurate MLIP for our selected organomagnesium reagents.
The training of an accurate MLIP needs a large amount of data. To obtain such data efficiently, we adopted an active learning procedure. It consisted of training a crude model on a small set of molecules and then leveraging the speed of the MLIP to obtain new data points. The relevant fraction of these new structures that was seen to be useful to augment our model was added to the training set. This procedure was repeated for several cycles until the model was able to describe the phase space of our solutions accurately. Comparison with literature results confirmed the validity of the obtained MLIP.
The resulting MLIP was used to simulate molecular dynamics trajectories of four reagents containing Mg of general formula RMgCl (R=Me, Et, iPr, tBu) to study their structure and reactivity in solution. We looked at three main aspects: association dynamics, solvation structure, and, finally, the energetics of a key process in the chemistry of these molecules, the Schlenk equilibrium. Results from this work shed light on the aspects that govern the four reagents in solution, highlighting the differences among them. In particular, the relationship of the reactivity with the bulk of the reagent nicely agrees with and explains experimental observations found in the literature.
On another front, the distribution, structure, and reactivity of one of the reagents, iPrMgCl, combined with a suitable substrate, acetophenone (AcPh), was studied in an organic/DES mixture with molecular dynamics simulations. The chosen deep eutectic solvent was a solution of choline chloride and glycerol, which was shown to improve the reactivity when used in combination with iPrMgCl in a tetrahydrofuran (THF) solution. We performed simulations of AcPh in bulk glycerol and DES, showing the key role of the added choline chloride salt. We also simulated the THF/DES mixture containing both iPrMgCl and AcPh; the analysis of the distribution of the species in the mixed phase shows that the organic substrate, initially located in the DES, slowly transfers to the organic solvent, whereas the Mg reagent tends to accumulate at the interface due to its amphipathic nature. The increased concentration of iPrMgCl at the interface, combined with the flow of AcPh from the DES to the organic solvent, creates the favourable conditions for the reaction to occur.
Finally, we modelled with a QM/MM approach the reaction of an iPrMgCl with AcPh in the mixed THF/DES phase. In this reaction, the formation of a new C-C bond results in the functionalization of AcPh, opening the road to many synthetic campaigns. Experimental results showed a net increase in reaction kinetics and a noticeable decrease in byproduct presence when conducted in mixed THF/DES phases. We focused on two main aspects that can affect the reactivity: the aggregation of iPrMgCl and the charge of the overall reacting complex. For each of the investigated systems, we also studied the energetics of the main competing reaction that causes unwanted byproducts. Analysis of the trajectories and of the energetics of the reaction showed how a synergistic combination of reagent aggregation, charge and localization at the DES interface helps to promote the main reaction pathway, suppressing any unwanted side products.
