Why do density-functional corrections capture electron correlation better than fixed empirical potentials at bond-breaking geometries?

Your question asks about a specific comparison—density-functional corrections versus fixed empirical potentials at geometries where bonds break—but the provided preprints don't directly address that contrast.

However, I can point to one result that's closely relevant. At bond-breaking geometries (R > 0.5 Å), the analytical density-functional expression for correlation energy in H₂ recovers >99.5% of the nearly exact SCAN functional and correctly dissociates the H–H bond with errors of only 0.002 Å and 0.19 eV [AG-2024.05-2255]. The reason this works is subtle: the correlation energy is encoded not as a fixed empirical number, but as a *functional* — a rule that depends on the electron density ρ itself. As bond length changes, the density ρ reshapes, and the functional expression adapts automatically. A fixed empirical potential, by contrast, doesn't know about this density evolution and therefore fails at extreme geometries like dissociation.

More broadly, the archive emphasizes that static and dynamic electron correlation require *flexible* representations. Multi-configurational and explicit-correlation approaches can encode correlation directly in the wavefunction ansatz, allowing them to capture the reshuffling of electrons during bond-breaking in ways fixed potentials cannot [AG-2024.04-2360].

The preprints don't contain a head-to-head comparison of density-functional corrections against fixed empirical potentials specifically at bond-dissociation curves.