Disentangling solvation and structural response in quantitative calculations of spectroscopic parameters

Abstract

Understanding processes on the molecular level is the central element of chemistry. In addition to experimental techniques such as spectroscopy, this understanding is increasingly being gained through the use of theoretical methods. Historically, quantum chemical calculations or molecular dynamics (MD) simulations have been the methods of choice for gaining deeper insights into chemical processes. In recent years technological advances in handling large amounts of data have also brought forth artificial intelligence methods, which are now being applied in chemistry as well. In this work, all these methods were combined to gain a deeper understanding of chemistry at the atomic level. A particular focus was placed on understanding solvation conditions and their changes under extreme conditions such as high hydrostatic pressure. On the experimental side, changes in nuclear magnetic resonance (NMR) spectroscopy parameters under high hydrostatic pressure were investigated. These were used to optimize theoretical methods in order to accurately model behavior at ambient conditions and under high hydrostatic pressure. This involved a combination of quantum chemical calculations and ab initio (AI), classical force field, and machine learning (ML)-based MD simulations. In the quantum chemical calculations, the Embedded Cluster Reference Interaction Site Model (EC-RISM) was used together with explicit solvation yielding highly accurate NMR and EPR parameters. The calculated EPR parameters were used to analyze and quantify the reason for the pH dependency of an EPR spin probe. Although a significant change in the solvent environment of the EPR-active nitroxide could be observed it was found that about 75% of the change in EPR parameters is due to an altered electronic structure caused by the additional proton. The remaining 25% could be attributed to the changed solvent environment. Lastly, an ML potential (MLP) was developed that was trained on quantum chemical EC-RISM calculations. The MLP is capable of making predictions for most neutral, organic molecules. The accuracy of the potential was tested for predicting tautomeric equilibria, where an additional error of less than 1 kcal/mol compared to the direct reference calculation was observed. Predictions based on the MLP, however, are several orders of magnitude faster. Through the close integration of experimental spectroscopic measurements and a wide range of theoretical methods, this work not only determined further structural changes of molecules under high-pressure conditions or pH variations but also devoted a substantial part to evaluating theoretical methods in terms of their accuracy and efficiency, thereby providing recommendations for potential further applications.