BPDN l0-norm approximation

Why must the problem be convex? Don’t assume it, prove it.