Note on a problem in second-order gradient methods