Computational Electrochemistry

Modeling electrochemical interfaces with finite field molecular dynamics

A realistic representation of an electrochemical interface requires treating electronic, structural, and dynamic properties on an equal footing. The density functional theory-based molecular dynamics (DFTMD) method is perhaps the only approach that can provide a consistent atomistic description. However, the challenge for DFTMD modeling of material’s interfacial dielectrics is the slow convergence of the polarization P, where P is a central quantity to connect all the dielectric properties of an electrochemical interface.

Our contribution in this area is to develop finite-field molecular dynamics simulation techniques for modeling electrochemical systems [1, 2]. Constant D Hamiltonian, originally designed for treating spontaneous polarization in ground-state ferroelectric systems, is a new statistical mechanics ensemble. We show that the advantage of constant D simulations in computational electrochemistry is three-fold: a) It significantly speeds up simulations for both polar liquids and insulating oxides; b) It eliminates the finite size effect for modelling the electric double layer due to the periodic boundary condition; c) It controls the surface charge density in metallic electrodes.

Simulating ion transport in liquid and polymer electrolytes

Lithium-ion batteries are electrochemical devices that involve multiple time-scale and length-scale to achieve optimal performance and safety requirements. In terms of the electrolyte which serves as the ionic conductor, a molecular-level understanding of the corresponding transport phenomenon is crucial for the rational design.

Currently, we are working on molecular dynamics simulations of ion transport in different types of electrolytes from liquid electrolytes to polymer electrolytes (with Daniel Brandell) which are relevant to battery applications. Our contribution in this area is to elucidate the reference-frame dependence in the computation of the Onsager coefficients and transference numbers [3]. This step is crucial for maximizing the power of MD simulation and connecting experiment and molecular simulation.

Developing atomistic machine learning for electrochemical systems

Machine learning is becoming increasingly important in materials modelling and discovery. Atomic neural networks (ANN), which constitute a class of ML methods, have been very successful in predicting both physicochemical properties and approximating potential energy surfaces.

Recently, we have taken the initiative and developed an open-source Python library named PiNN (https://github.com/Teoroo-CMC/PiNN/), allowing researchers to easily train state-of-the-art ANN architectures and implement new ML models for making chemical predictions (e.g. the response function). In particular, we have designed and implemented an interpretable and high-performing graph convolutional neural network architecture PiNet, and demonstrate how the chemical insight “learned” by such a network can be extracted [4]. This allows us to carry out the atomistic simulation of electrochemical systems powered by ML models, as demonstrated by the PiNNwall interface [5].

References

1. Knijff, J. Mei and C. Zhang, in Encyclopaedia of Solid-Liquid Interfaces., 2024, 567 DOI: 10.1016/B978-0-323-85669-0.00012-X
2. Andersson, and C. Zhang, Curr. Opin. Electrochem., 2023, 42: 101407 DOI: doi.org/10.1016/j.coelec.2023.101407
3. Shao, H. Gudla, J. Mindemark, D. Brandell and C. Zhang, Acc. Chem. Res., 2024, 57: 1123 DOI: 10.1021/acs.accounts.3c00791
4. Shao, L. Knijff, F. M. Dietrich, K. Hermansson and C. Zhang, Batter. Supercaps, 2021, 4: 585 DOI: 10.1002/batt.202000262
5. Dufils, L. Knijff, Y. Shao and C. Zhang, J. Chem. Theory Comput., 2023, 19: 5199 DOI: 10.1021/acs.jctc.3c00359