Linnéa Andersson: Advancing Molecular Simulation Methods for Electrified Metal/Electrolyte Interfaces

Abstract

Metal/electrolyte interfaces play a key role in electrochemistry because it is where the electronic conductor and the ionic conductor meet. Across the metal/electrolyte interface there often exists a strong electric field, which governs the orientation of solvent molecules, the distribution of charges and the rate of (electro)chemical reactions. Therefore, an atomic scale understanding of their structure and dynamics is required for the knowledge-based development of electrochemical processes. To that aim, this thesis focuses on improving the descriptions of electrified metal/electrolyte interfaces in molecular dynamics (MD) simulations. The different approaches for controlling electrical bias in molecular simulations of metal/electrolyte interfaces are first reviewed.

To include important interfacial processes, such as chemisorption of water and ions, using finite-field DFTMD simulations for describing metal/electrolyte interfaces was explored. In particular, the charged Au(111)/NaCl(aq) and Au(100)/NaCl(aq) interfaces were modeled while keeping the electrode charge and ionic charge as separate degrees of freedom. Cl is known to adsorb extremely rapidly on Au, which was also found in the simulations. At the positive Au(100) surface this leads to a charge inversion as inferred from the reorientation of water molecules into an H-down configuration. The finite field method can also be used to calculate the image plane of a metal surface from the response charge. The image plane position is known to correlate with the negative capacitance response of metals. A simple capacitor-inseries relationship between the image plane position and increased double layer capacitance was established through the synergy with semi-classical models. Therefore, this explains the high capacitance found at the Au/NaCl(aq) interfaces and provides a way to also improve semiclassical models from finite-field DFTMD simulations.

MD simulations based on semi-classical models are often used to describe the electronic response of polarizable electrodes. Here the domain of applicability of semi-classical models was extended with a machine learning (ML) approach. In the PiNet-χ models the so-called charge response kernel (CRK) is learned from polarizability data at the density functional theory (DFT) level. The construction of the CRK was made possible by using atomistic predictions from the GCNN (graph convolutional neural network) PiNet. Several of the PiNet-χ models have been shown to be capable of predicting the charge response of both conducting and insulating systems, which enables their use for heterogeneous electrodes and electrodes of varying metallicity. The performance of PiNet-χ has been further improved with the use of the equivariant GCNN PiNet2.Finally, finite-field methods were combined with machine learning molecular dynamics (MLMD) to extend the accessible time and length scales. This approach was applied to fielddriven proton transport in water, which provides the foundation to a further development for MLMD simulations of solid-liquid interfaces at constant electric displacement.