Continuity properties of the inf-sup constant for the divergence
For researchers computing numerical approximations of the LBB constant on complex domains, this work clarifies when such approximations are reliable and when they may fail.
The paper establishes conditions under which the inf-sup constant for the divergence (LBB constant) is upper semi-continuous or continuous with respect to domain perturbations, and provides sufficient conditions for convergent numerical approximation of this constant. Numerical examples demonstrate that these conditions are nearly optimal.
The inf-sup constant for the divergence, or LBB constant, is explicitly known for only few domains. For other domains, upper and lower estimates are known. If more precise values are required, one can try to compute a numerical approximation. This involves, in general, approximation of the domain and then the computation of a discrete LBB constant that can be obtained from the numerical solution of an eigenvalue problem for the Stokes system. This eigenvalue problem does not fall into a class for which standard results about numerical approximations can be applied. Indeed, many reasonable finite element methods do not yield a convergent approximation. In this article, we show that under fairly weak conditions on the approximation of the domain, the LBB constant is an upper semi-continuous shape functional, and we give more restrictive sufficient conditions for its continuity with respect to the domain. For numerical approximations based on variational formulations of the Stokes eigenvalue problem, we also show upper semi-continuity under weak approximation properties, and we give stronger conditions that are sufficient for convergence of the discrete LBB constant towards the continuous LBB constant. Numerical examples show that our conditions are, while not quite optimal, not very far from necessary.