[robo]

I'm working on an external flow that I'm running in simpleFoam at the moment trying to get a steady state solution. The mesh is around 2 million cells; developed using snappyHexMesh. It passes all mesh checks, and I've run it in serial for around 100 steps with no problems. When I decompose it using scotch decomposition in decomposePar, if I use more then 2 processors, it will crash on the first step, claiming that the U matrix is singular. If I use a simple decomposition, it runs fine.

I would like to be using more processors then I can easily develop a simple decomposition for, however clearly something is wrong with scotch/decomposePar and I am not sure how to debug it. Has anyone run into similar issues? Any advice on using the scotch decomposition? Is metis more reliable? I'm using OF 2.3.0 if that's relevant.