Reaction-Network Reasoning with Frontier Models for Experimentally Confirmed Catalyst-Selectivity Hypotheses

The paper presents a human-AI co-thinking framework that constrains frontier language models to reason over explicit reaction networks rather than static descriptors. Applied to electrochemical CO2 reduction, it identifies ketene desorption and hydroxide capture as part of an acetate-forming pathway, and proposes a distinct adsorbed-CO/CH2 coupling route to ketene. The resulting control levers include local alkalinity, controlled iron incorporation, and restricted access to interfacial proton donors. These hypotheses guided synthesis of a copper-iron oxide catalyst that reportedly achieved a threefold increase in acetate selectivity over matched Cu-rich baselines. The abstract does not provide absolute selectivity values or full experimental details.