So, seems like matrix exponentiation is faster than computing eigenvalues / eigenvectors in practice? Take a look: http://mathoverflow.net/questions/62904/complexity-of-eigenv...