I think that's a very sharp observation. I tried it, but in order to exponentiate the vectors (instead of floating point numbers) I end up with the original matrix algorithm (pre-eigenvalues)...