PiNN: Equivariant Neural Network Suite for Modeling Electrochemical Systems

Area of Science:

• Materials Science
• Computational Chemistry
• Electrochemistry

Background:

• Electrochemical energy storage and conversion are crucial for global electrification and sustainable development.
• Understanding and designing electrochemical materials at the atomistic level is a key challenge.
• Molecular modeling, powered by machine learning (ML), is essential for this endeavor.

Purpose of the Study:

• To upgrade the PiNN (pairwise interaction neural network) Python package for enhanced molecular modeling of electrochemical systems.
• To introduce equivariant features into the PiNet2 architecture for improved potential energy surface fitting, dipole/charge predictions, and charge response kernel generation.
• To establish PiNN as a versatile, high-performance ML-accelerated platform for electrochemical research.

Main Methods:

• Developed PiNet2 architecture with equivariant features for potential energy surface fitting.
• Integrated PiNet2-dipole for accurate dipole and charge predictions.
• Introduced PiNet2-χ for generating atom-condensed charge response kernels.
• Utilized plug-ins like PiNNAcLe for adaptive ML potential generation and PiNNwall for modeling electrodes under bias.

Main Results:

• The equivariant PiNet2 demonstrated significant performance improvements over the original PiNet architecture.
• Benchmarking on diverse datasets (small molecules, crystals, electrolytes) confirmed state-of-the-art overall performance.
• The enhanced PiNN package effectively predicts key properties for electrochemical materials.

Conclusions:

• The upgraded PiNN package, featuring equivariant PiNet2, offers a powerful and versatile platform for ML-accelerated molecular modeling in electrochemistry.
• This advancement facilitates atomistic precision in understanding, controlling, and designing next-generation electrochemical energy materials.
• The PiNN platform is poised to accelerate research and development in sustainable energy solutions.