Many industrial chemical processes are homogeneous, in that the catalyst (often a transition metal complex), the feedstocks and products are all dissolved in solution. Understanding how such processes work, i.e. the mechanism of reaction, is important for the future design of new, more efficient and selective catalysts. One effective way to provide such insight is to use computational modelling. However, standard modelling approaches usually ignore the role played by solvent, or only treat this in a very approximate fashion. This project set out to design a computational approach that would properly include solvent molecules in the modelling of catalytic processes.
In order to incorporate solvent effects on reactivity, large numbers (hundreds) of solvent molecules need to be included in the calculation. We therefore employed hybrid methods, where the catalyst and reacting molecules are dealt with at a high level of theory (density functional theory, DFT) but the solvent molecules are treated at a lower level of theory based on molecular mechanics (MM). Another challenging issue is the conformational flexibility of these large models, i.e. the number of different arrangements in space that the different molecules can adopt. To tackle this we employed molecular dynamics (MD) simulations in order to ensure a proper coverage of the conformational space. The combination of DFT/MM calculations and MD calculations meant that our initial targeted system, the methoxycarbonylation of simple alkenes at Pd catalysts, represented too great a technical challenge for the exploration of our approach. Instead we identified a number of well-defined chemical reactions occurring at transition metal centres for which experimental data on the energetic barriers to reaction were available. These were then used to benchmark our approach. In particular these reactions involved the cleavage of ionic bonds and hence the involvement of charged species. Modelling these processes in the gas-phase is completely unrealistic and the inclusion of solvent is essential. Given the many important processes that involve the participation of charged species, our work has the potential to have a significant impact on the modelling of real chemical processes in solution.
To date the results from our studies are very promising. With a simple (but difficult to model) chemical process such as H2O/Cl- substitution at [Pt(C2H4)Cl3]- we correctly reproduce the selectivity for the trans position and also obtain good absolute values for the activation energies (trans: 19 kcal/mol; cis: 3 kcal/mol). This is extended to the successive H2O/Cl- substitution at cis-[Pt(Cl)2(NH3)2] (with activation barriers of 19 kcal/mol and 23 kcal/mol respectively). We have also considered the Monsanto Reaction, an important industrial process for the production of acetic acid. Here we correctly reproduce the expected "SN2" mechanism, unlike recent computational reports where this has been discounted.
A further benefit of our approach is the ability to describe the interaction between solvent and solute molecules correctly. Such interactions have a strong contribution from dispersion effects that are poorly modelled with standard DFT calculations. Thus the phosphine dissociation energy from trans-[Pd(Cl)(Ph)(PPh3)2] is correctly modelled when both the reactant and the dissociated fragments are incorporated into a box containing several hundred benzene molecules.
The PDRA employed (at Heriot Watt) on this project (at Her brought the necessary expertise to the project, but required training in QM methods for gas-phase calculations and resulted two papers. We acted as advisors on experimental aspects of the project as originally envisaged.