Thrust 1: machine-learned classical DFT and solvation models

Accurate first-principles electrochemistry requires beyond-DFT methods to correctly capture the charge states of molecules on electrode surfaces, and accurate electrolyte solvation to determine the electrode charge at the specified potential within a grand-canonical electronic structure simulation. The expense of beyond-DFT methods such as RPA necessitate the avoidance of liquid/electrolyte thermodynamic sampling using molecular dynamics, which is possible using the framework of joint density-functional theory (JDFT). Advancing grand-canonical JDFT for first-principles electrochemistry requires techniques that can capture atomic-scale electrolyte structure in charged electrochemical interfaces.

Thrust 1 will develop the next generation of liquid density models that capture the electrochemical double layer structure for accurate first-principles electrochemical calculation using grand-canonical JDFT methods. Specifically, we will develop free energy functionals for classical DFT and bridge functionals for RISM treatment of arbitrary electrolytes, as well as electrode-electrolyte coupling functionals, all using a new machine learned classical DFT approach. We will train these models exclusively to ​ab initio data, using large molecular dynamics simulations using ML potentials trained to small AIMD simulations, making it possible to automatically develop ​ab initio​ electrochemical solvation for ​any electrolyte.