Does DFTB lose accuracy faster than classical potentials when bond strain exceeds the tight-binding parameterization range?

Looking at your question about DFTB accuracy under bond strain, the provided preprints don't directly address this comparison. Let me explain what we'd expect theoretically, then note the gap.

DFTB (Density Functional Tight Binding) is a semi-empirical quantum method that approximates density functional theory by assuming electronic structure can be described using a minimal set of atomic orbitals and precomputed integrals [AG-2024.04-1576]. The method is parameterized for specific bond lengths and electronic configurations typical of a given material system.

When bonds strain far outside the parameterization range—think of stretching a C–C bond to 50% longer than typical—DFTB should indeed lose accuracy faster than a carefully fitted classical potential. Here's why: classical potentials like Morse or harmonic springs are often fitted to a *range* of bond lengths and can extrapolate reasonably smoothly. DFTB, by contrast, relies on overlap integrals between orbitals that were tabulated assuming normal bonding geometry. As geometry deviates, the physical assumptions embedded in those tables degrade. The electronic structure itself may change qualitatively (e.g., bond breaking begins), and DFTB's minimal orbital basis becomes increasingly inadequate.

That said, the preprints here focus on DFTB's *strengths* in capturing quantum effects (van der Waals interactions, electronic correlations) and comparisons with other *quantum* methods [AG-2024.04-1576], not on its failure mode under extreme strain relative to classical models.

The archive doesn't contain a direct benchmark comparing DFTB and classical potential accuracy across a range of bond strains.