Transferability and Accuracy of Ionic Liquid Simulations with Equivariant Machine Learning Interatomic Potentials.

Area of Science:

• Computational Chemistry
• Materials Science
• Electrochemistry

Background:

• Ionic liquids (ILs) are versatile electrolytes with tunable properties, acting as "designer solvents" in energy storage and beyond.
• The application of machine learning interatomic potentials (MLIPs) for simulating ILs remains an underexplored area.

Purpose of the Study:

• To assess the viability and transformative potential of MLIPs for simulating ionic liquids.
• To determine if MLIPs can be compositionally transferable for IL mixtures and additives.

Main Methods:

• Training a MLIP on a limited dataset of density functional theory (DFT) calculations for ionic liquids.
• Evaluating the MLIP's accuracy for predicting properties of IL mixtures not included in the training set.
• Synthesizing and experimentally characterizing a novel ionic liquid.

Main Results:

• Demonstrated compositional transferability of the MLIP, enabling accurate simulations of IL mixtures.
• The developed MLIP showed reasonable agreement with experimental data and DFT calculations for a novel IL.
• MLIPs trained on approximately 200 DFT frames proved effective.

Conclusions:

• MLIPs show significant promise for advancing the simulation and design of ionic liquids.
• Compositional transferability is a key feature of MLIPs for handling complex IL electrolyte systems.
• MLIPs offer a computationally efficient and accurate approach for studying novel ILs.