I haven’t been able to find an implementation of an iterative solver (invertible matrix, no further specialization) general enough to work on matrices of matrices, i.e. A = [B C D…; E F G…; H I J…; …], b = [K; L; M; …], x = [N; O; P; …], A is block sparse, b, x is block dense. I would like to avoid concatenation in my use-case.
Shouldn’t this be generally possible using GMRES? Perhaps there’s an implementation out there that I’ve missed, or a way to use IterativeSolvers.jl/KrylovMethods.jl more appropriately than I have tried?