Ok, I’ve taken a step in a different (hopefully simpler) direction to try to work this out. I’m trying to write a (crude) Moore-Penrose numerical continuation scheme that uses SNESSolve. The crux of this problem leads to a bordered matrix of the form:
A = [[J, Jl],[yq',yl]];
where again J is a sparse matrix, Jl and yq are vectors, and yl is a scalar. (all are Mat objects).
Since this is only a 2x2 block system, I wanted to solve this directly using the Schur factorization. However, I am having trouble telling PETSc how to handle the scalar yl in the fieldsplit. Any suggestions?
continueSNES_MWE.edp (6.1 KB)