This CAREER project was focused on studying local solvation environments to better understand how to carry out energetically efficient hydrogenations.  The research project to date has resulted in 17 scientific publications, helped support three PhD students at the University of Pittsburgh at different stages of their academic careers, and helped establish the PI as a leader in computational chemistry modeling applications and method development. The project also allowed the PI to explore quantitative ways to measure diversity (in different definitions) to assess how diversity impacts group project performance that were reported in ASEE conference proceedings.  

The first activities of this project were to understand the extent that explicit and discrete solvation interactions can impact chemical reaction mechanisms in homogeneous solution and how to best capture those interactions. We investigated the mechanism for the Morita-Baylis-Hillman reaction (doi: 10.1039/C8SC01424H) and proposed an automatable scheme to introduce explicit solvation interactions by building solvent clusters around reacting molecules and then placing these so-called microsolvated species in a continuum solvent model. We learned that when continuum solvation models are used alone in such a complex reaction mechanism, large and unphysical errors can ensue. On the other hand, when explicit interactions are accounted for, microsolvating clusters can sometimes erroneously favor unphysical structures that in turn can cause significant errors. The key finding was that using mixed continuum/explicit models requires significant care to ensure that the correct explicit interactions are accounted for in the system. Overall, we find that key interactions can be identified through a) having the correct chemical intuition and adding a minimal number of explicit interactions in the correct locations, b) systematically generating microsolvating clusters that eventually become large enough to capture the correct explicit interactions, or c) using computationally expensive explicit solvation models that sample all the correct locations. In general, the computational complexity increases from a) to b) to c), but we find that careful treatments using procedure b) are moderately costly and potentially not worth the effort unless procedure c) is not possible due to a lack of forcefields or other issues related to computational expense.

We also found that by using similar schemes where we systematically increased microsolvating environments, we could quantitatively predict solvation energies of ions in solution (doi: 10.1021/acs.jctc.9b00605). In that work, we found a scheme where single ion solvation energies would gradually converge with respect to the local solvation environment converging to a similar atomic scale structure as determined using a SOAP sketchmap analysis.  Interestingly, the solvation energies that were being predicted quantitatively were reflecting so-called "absolute" solvation energies that contained the energetic contribution known as the surface potential. The distinction between "absolute" solvation energies and "real" solvation energies that do not contain the surface potential contribution has not always been well-appreciated in the literature, and our work helped clarify a long-standing question about why different computational schemes for predicting solvation energies would provide different results that were individually consistent with one of two different sets of experimental data.  

Our awareness of relevant treatments of explicit interactions helped us write perspectives and reviews on the topic of solvation environments (doi: 10.1002/wcms.1446, 10.1063/1.5143207, 10.1021/acs.jpcc.0c11345), and it helped us collaborate with other researchers to explore thermodynamic energies of biomimetic organic hydride donors (doi: 10.1021/jacs.7b13526), mechanisms for glutathione oxidation on graphene (doi: 10.1021/acsami.0c11539), and investigations into palladium alloys that are catalysts for CO2 reduction while being resistant to CO poisoning (10.1021/acscatal.9b00330.

Finally, our last work studying the role of local solvation environments explored how different classes of solvation models impacted mechanisms for hydride transfers in aqueous solution with and without the presence of other ions in solution (10.1021/acs.jpca.0c08961). The main findings of that paper showed that explicit solvation interactions were critical, continuum solvation models of many different forms were challenged by capturing reaction energetics correctly, and this led us to appreciate that promising paths forward include the development of better explicit solvation models, likely using machine learning. This project supported the PI on a sabbatical stay at the University of Luxembourg, where the PI worked in Prof. Alex Tkatchenko's research group and initiated the development of efficiently trained machine learning models for solvents (doi: 10.1039/D3DD00011G) while preparing a widely read review article about how to combine computational quantum chemistry and machine learning for predicting chemical insights (doi: 10.1021/acs.chemrev.1c00107). This project also helped support the PI in exploring the viability of computational quantum alchemy procedures for catalysis applications (dois: 10.1002/adts.201800142, 10.1002/qua.26380, 10.1063/5.0079483, 10.1063/5.0079487, as well as develop new semiempirical quantum chemistry methods that have promise as next generation computational tools for reaction mechanism analyses (doi: 10.1021/acs.jctc.2c00334).

Last Modified: 07/03/2024
Modified by: John A Keith