We address the large-scale regularity theory for the stationary Navier–Stokes
equations in highly oscillating bumpy John domains. These domains are
very rough, possibly with fractals or cusps, at the microscopic scale,
but are amenable to the mathematical analysis of the Navier–Stokes
equations. We prove a large-scale Calderón–Zygmund estimate, a large-scale
Lipschitz estimate, and large-scale higher-order regularity estimates, namely,
and
estimates. These nice regularity results are inherited only at mesoscopic scales, and
clearly fail in general at the microscopic scales. We emphasize that the large-scale
regularity
is obtained by using first-order boundary layers constructed via a new argument. The
large-scale
regularity relies on the construction of second-order boundary layers, which allows for
certain boundary data with linear growth at spatial infinity. To the best of our
knowledge, our work is the first to carry out such an analysis. In the wake of many
works in quantitative homogenization, our results strongly advocate in favor of
considering the boundary regularity of the solutions to fluid equations as a
multiscale problem, with improved regularity at or above a certain scale.