Rules Describing CO₂ Activation on Single-AtomAlloys from DFT-Meta-GGA Calculations and Artificial Intelligence

Single-atom alloys (SAAs) arise as a promising concept for the design of improved CO₂ hydrogenation catalysts. However, from the immense number of possible SAA compositions and structures, only a few might display the properties required to be useful catalysts. Thus, the direct, high-throughput screening of materials is inefficient. Here, we use artificial intelligence to derive rules describing surface sites of SAAs that provide an effective CO₂ activation, a crucial initial step to convert the molecule into valuable products. We start by modeling the CO₂ interaction with 780 sites of flat and stepped surfaces of SAAs composed by Cu, Zn, and Pd hosts via high-quality DFT-mBEEF calculations. Then, we apply subgroup discovery to determine constraints on key physicochemical properties, out of 24 offered candidate descriptive parameters, characterizing subgroups (SGs) of surface sites where chemisorbed CO₂ displays large elongations of its C–O bonds. The key identified parameters are free-atom properties of the elements constituting the surface sites, such as their electron affinity, electronegativity, and radii of the d-orbitals. Additionally, the generalized coordination number is selected as a key geometrical parameter. The SG rules are applied to identify promising surface sites from a candidate space of over 1500 possible ones in different single-atom and dual-atom alloys. Some of the promising alloys predicted by the SG rules were explicitly tested by additional DFT-mBEEF calculations and confirmed to provide a significant CO₂ activation.