The study heavily relies on DFT calculations using pristine TM-HAB surfaces under vacuum and idealized conditions. While this provides useful trends, it ignores real-world complexities such as surface reconstructions, solvent interactions, and the formation of defects or dynamic restructuring under electrochemical conditions. Particularly for bifunctional catalysts operating across HER, OER, and ORR, surface oxidation or poisoning can significantly alter catalytic behavior. How would the inclusion of these effects, such as modeling explicit solvent layers, defect states, or operating potential cycling, affect the predicted activity, especially for the leading Rh-HAB and Co-HAB catalysts?
The study constructs the φOH and φH descriptors to predict catalytic activity using machine learning and intrinsic metal parameters. However, these descriptors are trained and validated within a narrow materials class, TM-HABs, with only a few elements. There’s no external benchmark, experimental validation, or cross-validation against unrelated catalyst systems. How can we be confident that these descriptors are not overfitted to this specific dataset? Would they hold predictive power for other 2D MOFs or non-MOF catalysts? Further, does their performance degrade when moving beyond mid-row transition metals?