The 3-D time-dependent Navier–Stokes equations on multi-connected domains with inhomogeneous boundary conditions and spectral hyperviscosity

Abstract

AbstractWe consider the time-dependent 3-D Navier–Stokes equations (NSE) on a multi-connected bounded domain $$\Omega \subset \mathbb {R^{\textrm{3}}}$$ Ω ⊂ R 3 with inhomogeneous boundary data $$\beta \in H^{1/2}(\Gamma )$$ β ∈ H 1 / 2 ( Γ ) on $$\partial \Omega =\Gamma $$ ∂ Ω = Γ , where $$\Gamma $$ Γ is a union of Lipschitz continuous surfaces $$\Gamma _{0},\Gamma _{1},\dots ,\Gamma _{l}$$ Γ 0 , Γ 1 , ⋯ , Γ l . This assumption includes the particular case when the $$\Gamma _{i}$$ Γ i are disjoint, the stationary version of which is classically known as Leray’s problem. Existence results for Leray’s problem have either assumed flux conditions beyond the general flux condition necessitated by compatibility constraints, or required size restrictions on the data. Here we incorporate a spectral hyperviscosity term in the time-dependent case and obtain existence and foundational results, assuming only the general flux condition and without imposing size restrictions on the boundary data $$\beta $$ β . For any such $$\beta \in H^{1/2}(\Gamma )$$ β ∈ H 1 / 2 ( Γ ) we establish global existence and uniqueness of mild solutions. Then on any interval [0, T] on which these solutions and the corresponding NSE solution share a common $$H^{1}$$ H 1 -bound (as is present on local intervals of existence of strong solutions, in certain special cases, or as is commonly assumed in achieving strong convergence results in numerical studies) we show for slightly more regular $$\beta $$ β that the spectrally-hyperviscous solutions converge strongly and uniformly in $$H^{1}(\Omega )$$ H 1 ( Ω ) to the NSE solution as the spectral hyperviscosity term vanishes in the limit of key parameters. To achieve this robust sense of approximation of the NSE system, an involved setup and specially-adapted semigroup techniques assume essential roles, and the exposition is new for the case $$\beta =0$$ β = 0 as well. Our final results adapt the NSE reformulation in Liu et al. (J Comput Phys 229(9):3428–3453, 2010) to recast our approximating system in a form potentially more adaptable to computation.