《1998 On the projected subgradient method for nonsmooth convex optimization in a Hilbert space》.pdf

Mathematical Programming 81 (1998) 23 35 On the projected subgradient method for nonsmooth convex optimization in a Hilbert space Y a . I . A l b e r 1, A . N . I u s e m *,2, M . V . S o l o d o v 2 Instituto de Matemdtica Pura e Aplicada, Estrada Dona Castorina 110, Jardirn Bothnico, Rio de Janeiro, R J, CEP 22460-320, Brazil Received 18 January 1996; revised manuscript received 1 January 1997 Abstract We consider the method for constrained convex optimization in a Hilbert space, consisting of a step in the direction opposite to an ek-subgradient of the objective at a current iterate, fol- lowed by an orthogonal projection onto the feasible set. The normalized stepsizes ek are exog- enously given, satisfying ~ = 0 c~k ec, ~ = 0 c~ ec, and ek is chosen so that ek ~/~k for some # 0. We prove that the sequence generated in this way is weakly convergent to a minimizer if the problem has solutions, and is unbounded otherwise. Among the features of our conver- gence analysis, we mention that it covers the nonsmooth case, in the sense that we make no assumption of differentiability o f f , and much less of Lipschitz continuity of its gradient. Also, we prove weak convergence of the whole sequence, rather than just boundedness of the se- quence and optimality of its weak accumulation points, thus improving over all previously known convergence results. We present also convergence rate results. © 1998 The Mathema- tical Programming