Title: On the strong concavity of the dual function of an optimization problem

Abstract: Strong concavity of the dual function of convex optimization problems with linear constraints was shown recently in [3]. We provide three new proofs of the strong concavity of the dual function for such problems. For convex problems with both linear and nonlinear constraints, the local strong concavity of the dual function was also shown in [3]. For such problems, we show that this strong concavity property holds if we weaken the strong convexity assumption of the objective used in [3] to strict convexity and we also show that this assumption as well as the assumption that the gradients of all constraints at the optimal solution are linearly independent cannot be further weakened. Finally, we illustrate our results with several examples.
[3] V. Guigues, Inexact Stochastic Mirror Descent for two-stage nonlinear stochastic programs, Mathematical Programming, to appear.