Analysis of solid/liquid interface and solution reactions with molecular simulations and machine learning
Embargo Date
2024-02-29
OA Version
Citation
Abstract
Understanding the behavior of electrolytes at the solid-liquid interface is important to numerous applications, such as in batteries and fuel cells. The molecular organization of a liquid at the interface is usually distinct from the bulk, and frequency measurements from experimental techniques such as vibrational sum frequency generation can be used to provide a characterization of local electrostatics at the interface. However, to infer molecular features of the interface based on these measurements, detailed atomistic simulations are required.
Our approach to computationally studying solid-liquid interfaces involves a combination of classical molecular dynamics simulations with vibrational frequency calculations using semiempirical frequency maps. The first part of this work investigates the effect of anion size on structure and electrostatics at the nitrile-functionalized gold-ionic liquid interface. We observed that the intercalation of smaller ions into the nitrile layer leads to higher electric fields and experimentally, larger redshifts. In the second part, we aim to understand solvatochromic frequency shifts at the nitrile-functionalized gold-water interface in the presence of other ligands. The electrostatic environment, in this case, is highly heterogeneous and subject to ligand placement, surface density, and the chemical nature of the ligands. Despite the heterogeneity, water access to nitrile probes turned out to make an essential contribution to the trend in vibrational frequencies.
The second part of the thesis aims to understand chemical reactions in the condensed phase, which often requires sampling multiple degrees of freedom. Constructing multidimensional free energy surfaces with reliable accuracy requires extensive conformational sampling, which can be prohibitively expensive. To address this challenge, first, we compute a multidimensional free energy surface using QM/MM reinforced dynamics simulations with inexpensive but less accurate DFTB3 as the QM method. In the next step, we improve the accuracy of the free energy surface using delta machine learning trained on previously accessed conformations. Finally, we apply this scheme to study biologically relevant phosphoryl transfer and phosphate hydrolysis reactions to gain novel mechanistic insights.
Description
License
Attribution 4.0 International