Lattice-free gradient polyhedra are optimality certificates for mixed integer convex minimization models. We consider how to construct these polyhedra for unconstrained models with two integer variables. A classic result of Bell, Doignon, and Scarf states that a lattice-free gradient polyhedron exists with at most four facets. We show how to construct a sequence of (not necessarily lattice-free) gradient polyhedra, each of which has at most four facets, that finitely converges in a lattice-free gradient polyhedron. Each update requires constantly many gradient evaluations ⋆ . This update procedure imitates the gradient descent algorithm, and consequently, it yields a gradient descent type of algorithm for problems with two integer variables. An open question is to improve the convergence rates to obtain a minimizer or a lattice-free set.