Comment by bee_rider

The algebraic manipulation that you use to break the dependency seems to be specific to the matrix being Poisson on a finite-difference grid (is that right?). In that case wouldn’t it make more sense to compare against specialized Poisson solvers?