Why does density functional tight binding with dispersion corrections still fail where classical potentials do for certain crack geometries?

# Why DFTB+MBD Still Struggles With Certain Crack Geometries

The short answer: DFTB with many-body dispersion corrections captures quantum effects classical potentials miss, but both still struggle because crack mechanics involve competing length scales — electronic, atomic, and continuum — that no single model naturally spans.

What DFTB+MBD Does Get Right

The DFTB+MBD framework [AG-2024.04-1576] combines Density Functional Tight Binding (a computationally efficient quantum method) with many-body dispersion (MBD) to describe van der Waals interactions more accurately than classical potentials. This matters because van der Waals forces, though weak individually, can dominate the adhesive energy holding crack surfaces together in certain materials. Where classical Lennard-Jones potentials assume pairwise interactions, MBD captures how three or more atoms modify each other's dispersion forces — a genuinely quantum effect.

The paper demonstrates that "classical simplified approximations at those scales are insufficient, and quantum-based modeling is required" [AG-2024.04-1576] for accurate mechanical predictions. So DFTB+MBD should, in principle, handle crack tips better than classical approaches.

Why Both Still Fail at Certain Geometries

The preprints suggest several failure modes:

1. Electronic structure fidelity at high strain. DFTB is itself an approximation. While it includes quantum effects, it truncates the density functional at a tight-binding level and may lose accuracy when bonds stretch or break sharply near a crack tip — especially for crack geometries where electronic structure changes non-monotonically. The papers on many-body wave function interpolation [AG-2024.02-1771] and comparisons between DFT variants [AG-2024.02-1831] hint that different methods rank differently depending on the geometry; DFTB may simply be poorly suited to some crack configurations.

2. Many-body dispersion becomes ambiguous at interfaces. MBD works well for isolated molecules or weakly interacting systems, but at a crack surface — where you have a disordered, strained interface — the many-body correction itself becomes geometry-dependent and harder to compute reliably. The entanglement structure of dispersion forces [AG-2024.07-1701] shows that "entanglement monogamy determines whether many-body corrections to the pair potential are attractive, repulsive, or zero," meaning the sign and magnitude of the correction depends sensitively on the atomic arrangement. A small shift in crack geometry can flip the balance.

3. The continuum-quantum bridge remains broken. Neither DFTB+MBD nor classical potentials naturally couple to continuum fracture mechanics (stress intensity factors, crack-growth rates). A crack geometry that triggers mode-switching (e.g., from mode I to mode II loading) or plasticity near the tip involves length scales from nanometers to millimeters — a span no single atomistic method handles well.

The Bottom Line

DFTB+MBD is a genuine upgrade over classical potentials for including quantum effects. But it inherits the fundamental limitation that all atom-by-atom models face: the crack tip is a singularity where electronic structure, crystal defects, and continuum stress all collide. Some geometries (say, symmetric mode-I cracks in simple crystals) may fall in a "sweet spot" where DFTB+MBD is accurate. Others — asymmetric, mixed-mode, or involving defect accumulation — push the method into regimes (high strain, charge redistribution, complex many-body entanglement) where it loses fidelity just as classical potentials do, albeit for different reasons.

The preprints provided don't directly address why DFTB+MBD fails at specific crack geometries — that comparison would need materials-mechanics papers with explicit failure-case studies.