Cartesian cut cell meshes generate grids by intersecting a regular Cartesian grid with an irregular domain boundary. The intersection creates arbitrarily small cells near the boundary, and these small cells force infeasible CFL restrictions on explicit time stepping — the time step must shrink in proportion to the smallest cell, which can be arbitrarily small.
Note (2026-11-12): checked against the paper's abstract and full text (Birke, Engwer, Giesselmann and May). The result is stated correctly as far as it goes: a consistency result for the Domain of Dependence stabilization for an arbitrary polynomial degree, given an exact solution of sufficient regularity. The essay says more than that: the paper states that its goal is not a complete error estimate but a generalized consistency result for high-order schemes (for linear PDEs), noting that earlier error analysis was thorough only for degree zero and that a two-dimensional result for linear advection existed only for a first-order method; error analysis typically hinges on such a consistency result, which this one could open the way toward. So 'the gap between it works and we can prove it works' is not closed by the paper, and 'the proof confirms what was empirically established' overstates a consistency result as a proof of accuracy.
The Domain of Dependence (DoD) stabilization fixes this: it allows time steps based on the underlying Cartesian mesh size, ignoring the small cut cells. For polynomial degree zero, this was proven to maintain accuracy. For higher orders, numerical experiments showed it worked, but the theory was missing.
The theory now exists (arXiv:2603.10754): DoD stabilization is consistent for arbitrary polynomial degree with sufficiently regular exact solutions. The gap between “it works” and “we can prove it works” spanned the entire range of practical polynomial degrees.
The structural observation: numerical evidence outran theory by the full width of the method's applicability. Every practitioner knew the stabilization worked at high order because they used it daily. The proof confirms what was already empirically established, but it does more — it opens the path to refined error analysis that numerical experiments alone cannot provide. The proof's value is not in confirming the result but in enabling further theory that requires the result as a foundation.