(538e) Developing a Fingerprinting and Machine Learning Framework Linking Structural Disorder to Oxidation Behavior in Metal Grain Boundaries for CO2 Electrocatalysis

The electrochemical reduction of CO₂ to yield CO and hydrocarbon fuels can be applied to carbon capture utilization and renewable energy storage. However, currently available electrocatalysts are affected by several shortcomings, such as poor product selectivity, low catalytic activity, high overpotentials, and the inability to tune catalyst properties to produce different products. Metal interfaces are able to stabilize adsorbed intermediates for electrochemical reaction on their catalytically active, undercoordinated, and oxidizable sites. Though different metal surfaces generally favor single products with variable product selectivity, Cu surfaces can uniquely produce multiple products selectively by tuning surface orientation, albeit with high overpotentials. Relative to surfaces, metal grain boundaries (GBs) have more active catalytic sites and lower electrochemical reduction overpotentials. Additionally, structural disorder distinguishes different metal GBs from one another and reaction conditions can affect the relative favorability of GBs. Therefore, given that differently oriented interfacial structures can selectively yield different reaction products, GB metal catalysts could simultaneously demonstrate high product selectivity, high catalytic activity, low overpotential, and product yields tunable by structure and reaction conditions.

The structural disorder distinguishing different GBs is affected by metal composition and structure, as well as metal-oxygen interfacial interactions resulting from deriving GBs from metal oxides. Thus, this work aspires to understand how GB metal properties and oxidation impact GB structural disorder, as well as linking structural disorder to relative GB energetic favorability. This aspiration is pursued by developing a fingerprinting method categorizing GBs by their radial distribution functions (RDFs), then applying those RDFs to a Markov chain Monte Carlo (MCMC) machine learning (ML) analysis generalizing the link between GB structural disorder and energetics. Firstly, an initial screening of metal GB candidates will be performed via molecular mechanics, indicating which structural properties best distinguish GB systems and which best correlate with energetics. Subsequently, density functional theory (DFT) calculations will be completed on screened structures and their oxidized analogues. A MCMC ML analysis will then discern the impact of oxidation on GB structural disorder and energetic favorability. By generalizing these structure-energy relationships, a fundamental understanding of how oxidation can tune GB structure to modify electrochemical product selectivity can be achieved.