Hi everyone,
I’m building a generalized FEM framework for linear stability (eigenvalue) analysis of incompressible Navier–Stokes flows.
My test case is the classic 2D flow around a cylinder (Taylor–Hood, P2–P1). I impose Dirichlet BCs at the inlet (input velocity) and cylinder (no-slip), and have tried both Neumann and Robin at the outlet. Top and bottom are set as free-flow (homogeneous Neumann).
The problem:
After assembling and solving the eigenvalue problem (SLEPc, shift-invert, LU preconditioned), all computed eigenmodes are highly localized at the outlet or at the bottom right corner.
The baseflow is physically reasonable and matches literature. Mesh refinements at the outlet and switching between Neumann and Robin have not helped.
Has anyone seen similar eigenmodes concentrating at an outlet/corner, or thinks that the issue sounds familiar?
Any suggestions would be very helpful.
Framework and case setup are at /ferdean/lsa-fw
in GitHub if anyone wants to dig in further (ferdean-solver-fix
branch). Still in development (documentation not updated)
Thanks for any ideas or pointers.
— Ferran