GitLab

Papers dealing with solvers for saddlepoint problems (like the incompressible Navier-Stokes equations, example) usually define the block matrix as

M = \begin{pmatrix}
  A & B^T\\
  B & C
\end{pmatrix}

where A is the block for the derivatives of the momentum balance equation residuals w.r.t velocity ("velocity block") and C is is the block for the derivatives of the mass balance equation residuals w.r.t pressure ("pressure block").

For incompressible fluids, C is zero.

However, the multitype blockmatrix in Dumux for staggered problems looks like this:

M = \begin{pmatrix}
  C & B^T\\
  B & A
\end{pmatrix}

which causes confusion and requires special care, e.g, for the implementation of the Uzawa solver !1827 (closed)

I suggest we adapt the matrix structure such that the velocity block is on the upper left. Are there any concerns / objections?