A matrix is a concatenation of columns yes, but how does that help us multiply matrices by using a sparse linear-system solver? Perhaps it's trivial but I don't see it.