(532at) Breaking Adsorption-Energy Scaling Limitations of Electrocatalytic Nitrate Reduction Via Machine Learned Insights

The electrochemical reduction of nitrate (NO3RR) plays an important role to rebalance the current nitrogen cycle by sustainably converting excess nitrate to nitrogen or ammonia. However, one of the challenges has been to find active and selective electrocatalysts to perform this chemistry. Experimentally, it is known that Cu(100) is active towards the formation of NH3 under alkaline conditions; however, large overpotentials are required and the hydrogen evolution reaction reduces selectivity. While various Cu based alloys have been synthesized to tackle these issues, the lack of clear design principles and mechanistic understandings hinders the search for viable electrocatalysts.

We used density functional theory (DFT) calculations to elucidate the NO3RR reaction pathways at (111) and (100) Cu facets under alkaline conditions. The improved activity and selectivity of the (100) facet is understood as coming from stronger nitrate and nitrite binding. Subsequently, a descriptor-based approach was used to explain the NO3RR activity trend for pure metals via the *NO₃ and *N binding energies as reactivity descriptors. However, the scaling between these descriptors makes it difficult to find electrocatalysts toward the top of the activity volcano. Further interrogation of this scaling via an interpretable machine learning framework shows that simple d-band center tuning of the active site does not break the adsorption-energy scaling. However, tailoring the adsorbate-substrate coupling can potentially lead to a weakening of the *N binding while leaving *NO₃ unaffected. This occurs specifically in crystal structures with strong orbital overlap between the subsurface atom and the *N. DFT Screening of intermetallic systems which satisfy this physical criteria reveals several potentially active and stable electrocatalysts. This approach is verified by experimental synthesis and electrochemical testing of CuPd and CuAu nanocubes. This study shows that interpretable machine learning can be used to guide electrocatalyst design while providing important insights into breaking scaling relations.