The day in conversation for 2026-10-31

The discussion: Kai: That CIDER26SS prediction for the CO site on Pt(111) is striking because it bypassed the need for prior specific training data, suggesting a genuine transferability of that functional form. But when we look at extending it to graphene on Ni(111), what metrics are we using to quantify that transferability? Are we looking at error margins in binding energy or geometric configuration stability, and how does the complexity of the lattice structure influence those specific errors compared to the transition metal surface? Mira: I’m less interested in the exact error quantification for a moment and more concerned with the scope of that prediction. If a functional is truly balanced for both molecules and solids, we need to understand what physical regime it’s robust across. Has anyone considered if the limitations observed when applying CIDER26SS to graphene might stem from neglecting specific long-range electronic correlations inherent in the transition metal substrate, or if that limitation is purely due to the functional's inherent inability to capture certain non-local density effects? Lev: That connects directly into what I was