Stability of solutions to convex problems of optimization (Q911031)

These lecture notes are concerned with two classes of convex optimization problems: mathematical programming in finite dimensional spaces and optimal control problems. It is assumed that the data of the problems depend on a finite dimensional vector parameter and that the set of admissible parameters is convex and open. The book consists of two parts. Part I (``Global stability'') contains the formulations of the optimization problems and several theorems concerning the Lipschitz continuity of the solutions and the associated Lagrange multipliers with respect to parameters. These theorems are derived for: (a) a standard convex programming problem, (b) an abstract optimization problem being a model for a class of convex optimal control problems with pointwise control constraints and (c) a convex optimal control problem with an integral cost functional and poinwise state and control constraints. Part II (``Differential stability'') is mainly devoted to the question of one-sided directional differentiability of the solutions and the associated Lagrange multipliers. The analysis presented here is constructive in the sense that the directional derivatives can be calculated as solutions to some auxiliary quadratic problems of optimization. Besides problems (a)-(c) of Part I, this analysis is done for some discrete-time optimal control problem and for some boundary control problem for a system described by a partial differential equation of parabolic type. The results presented in this book are either due to the author or taken from the literature. For both cases, detailed proofs of all the statements are included. The assumptions regarding the optimization problems considered are essentially the same for Parts I and II. They include twice continuous differentiability of the functions appearing in the problem, strong convexity of the cost functional and the linear independence of the gradients of all active constraint functions. These assumptions imply the existence and uniqueness of both the solutions and the associated Lagrange multipliers. The strict complementarity condition is generally not required. In my opinion, this book is a valuable addition to earlier monographs devoted to stability theory for optimization problems.