(209a) A Data-Driven Approach to Understanding and Predicting the Early Formation of the Solid-Liquid Electrolyte Interphase | AIChE

(209a) A Data-Driven Approach to Understanding and Predicting the Early Formation of the Solid-Liquid Electrolyte Interphase

Authors 

Persson, K. - Presenter, Lawrence Berkeley Lab
A remaining challenge in the pursuit of rational design of novel and optimized electrolytes is understanding and ultimately predicting the reaction cascade responsible for the creation of a functional liquid-anode solid-electrolyte interfaces (SEI). We here present an overview of the challenge, in different energy storage systems and advocate that a data-driven approach coupled with a high-throughput quantum chemistry computational infrastructure1 can address the complexity of such reaction cascade. To this end, we have constructed a chemical reaction network containing over 50,000 elementary reactions which allow for bond breaking/formation as well as oxidation/reduction reactions. The reaction energetics are obtained from our computational framework which automate geometry optimization and vibrational frequency calculations for molecules, including radical, charged, metal-coordinated, and solvated species. Our framework, using select levels of theory and a suite of on-the-fly error handlers, is able to successfully converge to a minimum energy structure and calculate the molecular thermodynamic properties in over 97% of cases. Individual calculations can then be combined to perform complex workflows in a fully automated fashion. To date, we have applied this framework to over 25,000 unique molecules relevant to the formation of solid-electrolyte interphases in Li-ion and Mg-ion batteries. To analyze possible reaction pathways, we apply a graph representation to our reaction network, leveraging an iterative reweighting strategy that solves for all prerequisite branching such that traditional pathfinding algorithms correctly capture coordinated pathways. Machine-learning algorithms operating on bond-formation and breaking energetics aid in rapid evaluation of highly reactive processes. We show that without any apriori chemical intuition, our automated framework recovers the most favorable reaction paths to form key SEI components, which were carefully identified over the past two decades. Thus, our data-driven approach and infrastructure show promise to accelerate the understanding of chemical reactivity in complex environments, in the aid of novel electrolyte design.