Graph-Guided C–C Coupling Pathways from Reaction Networks to Excited-State Dynamics in CO2-to-C2H4 Photoelectroreduction

Efficient photoelectrocatalytic CO2 reduction to multi-carbon products remains challenging due to branched reaction networks, multiple proton-coupled electron transfer (PCET) steps, and competing C–C coupling pathways. Here, we establish a multiscale framework integrating graph theory, physics-informed graph neural networks (CatalGNN), constant-potential ab initio molecular dynamics (cp-AIMD), nonadiabatic molecular dynamics (NAMD), and real-time time-dependent density functional theory (rt-TDDFT) to resolve CO2-to-C2H4 conversion on a Cu-anchored polymeric carbon nitride (Cu@PCN) catalyst. First, a heterogeneous directed bipartite graph reaction network systematically enumerates all complete pathways within the predefined reaction network. These pathways are encoded into CatalGNN, where potential-dependent edge features are updated via the computational hydrogen electrode relation using the pathway energetic span as the activity descriptor, allowing GNNExplainer to identify the performance-controlling elementary steps. The thermodynamic screening flags *CO-*CO dimerization as the most favorable coupling pathway. Crucially, cp-AIMD and NAMD simulations reveal that the pre-coupling CuN4@*CO-*CO precursor contains CO-derived 2π* antibonding states above the conduction-band minimum, serving as hot electron acceptors under photoexcitation. Hot-electron relaxation into these states occurs within 50 fs, providing a rapid electronic pre-activation channel. After C–C coupling, the *COCO intermediate introduces localized trap states that reshape carrier recombination dynamics. Notably, the spin-down channel exhibits a carrier lifetime of 16.4 ps, nearly three times longer than that of the spin-up channel, thereby sustaining localized charge at the reaction center for subsequent PCET steps. Finally, to bypass the thermodynamic bottleneck of *CO hydrogenation, rt-TDDFT simulations uncover a photoinduced radical mechanism where femtosecond laser excitation drives direct *CO hydrogenation and subsequent trapping of the *CHO radical at the CuN4 site through robust Cu–C bonding. This multiscale paradigm shows that efficient photoelectrocatalytic CO2-to-C2H4 conversion is governed by the synergy of pathway thermodynamics, potential-dependent electronic structures, spin-selective carrier lifetimes, and photoinduced radical chemistry, offering generalizable guidance for designing advanced multi-carbon products.