This is a preview. Log in through your library . Abstract Under what circumstances can finite difference approximations serve as substitutes for analytical derivatives in unconstrained optimization?