Perturbation Analysis of Optimization Problems J. Frédéric Bonnans 1 and Alexander Shapiro 2 1 INRIA-Rocquencourt, Domaine de Voluceau, B.P. 105, 78153 Rocquencourt, France, and Ecole Polytechnique, France 2 School of Industrial and Systems Engineering, Georgia Institute of Technology, Atlanta, Georgia 30332-0205, USA
2
Contents 1 Introduction 3 2 Background Material 11 2.1 Basic Functional Analysis... 11 2.1.1 Topological Vector Spaces... 11 2.1.2 The Hahn-Banach Theorem... 20 2.1.3 Banach Spaces... 25 2.1.4 Cones, Duality and Recession Cones... 34 2.2 Directional Differentiability and Tangent Cones... 37 2.2.1 First Order Directional Derivatives... 37 2.2.2 Second Order Derivatives... 41 2.2.3 Directional Epiderivatives of Extended Real Valued Functions 43 2.2.4 Tangent Cones... 48 2.3 Elements of Multifunctions Theory... 59 2.3.1 The Generalized Open Mapping Theorem... 59 2.3.2 Openness, Stability and Metric Regularity... 62 2.3.3 Stability of Nonlinear Constrained Systems... 65 2.3.4 Constraint Qualification Conditions... 72 2.3.5 Convex Mappings... 77 2.4 Convex Functions... 79 2.4.1 Continuity... 79 2.4.2 Conjugacy... 83 2.4.3 Subdifferentiability... 87 2.4.4 Chain Rules... 98 2.5 Duality Theory... 101 2.5.1 Conjugate Duality... 101 2.5.2 Lagrangian duality... 111 2.5.3 Examples and Applications of Duality Schemes... 114 2.5.4 Applications to Subdifferential Calculus... 121 2.5.5 Minimization of a Maximum over a Compact Set... 125 2.5.6 Conic Linear Problems... 133 2.5.7 Generalized Linear Programming and Polyhedral Multifunctions... 141 3
4 CONTENTS 3 Optimality Conditions 155 3.1 First Order Optimality Conditions... 156 3.1.1 Lagrange Multipliers... 156 3.1.2 Generalized Lagrange Multipliers... 162 3.1.3 Ekeland s Variational Principle... 166 3.1.4 First Order Sufficient Conditions... 168 3.2 Second Order Necessary Conditions... 172 3.2.1 Second Order Tangent Sets... 172 3.2.2 General Form of Second Order Necessary Conditions... 183 3.2.3 Extended Polyhedricity... 190 3.3 Second Order Sufficient Conditions... 196 3.3.1 General Form of Second Order Sufficient Conditions... 196 3.3.2 Quadratic and Extended Legendre Forms... 203 3.3.3 Second Order Regularity of Sets and No Gap Second Order Optimality Conditions... 209 3.3.4 Second Order Regularity of Functions... 219 3.3.5 Second Order Subderivatives... 223 3.4 Specific Structures... 229 3.4.1 Composite Optimization... 229 3.4.2 Exact Penalty Functions and Augmented Duality... 234 3.4.3 Linear Constraints and Quadratic Programming... 241 3.4.4 A Reduction Approach... 253 3.5 Nonisolated Minima... 258 3.5.1 Necessary Conditions for Quadratic Growth... 258 3.5.2 Sufficient Conditions... 262 3.5.3 Sufficient Conditions Based on General Critical Directions 270 4 Stability and Sensitivity Analysis 275 4.1 Stability of the Optimal Value and Optimal Solutions... 276 4.2 Directional Regularity... 281 4.3 First Order Differentiability Analysis of the Optimal Value Function287 4.3.1 The Case of Fixed Feasible Set... 287 4.3.2 Directional Differentiability of the Optimal Value Function Under Abstract Constraints... 293 4.4 Quantitative Stability of Optimal Solutions and Lagrange Multipliers303 4.4.1 Lipschitzian Stability in the Case of a Fixed Feasible Set. 303 4.4.2 Hölder Stability Under Abstract Constraints... 307 4.4.3 Quantitative Stability of Lagrange Multipliers... 311 4.4.4 Lipschitzian Stability of Optimal Solutions and Lagrange Multipliers... 316 4.5 Directional Stability of Optimal Solutions... 319 4.5.1 Hölder Directional Stability... 320 4.5.2 Lipschitzian Directional Stability... 321
CONTENTS 5 4.6 Quantitative Stability Analysis by a Reduction Approach... 331 4.6.1 Nondegeneracy and Strict Complementarity... 332 4.6.2 Stability Analysis... 337 4.7 Second Order Analysis in Lipschitz Stable Cases... 340 4.7.1 Upper Second Order Estimates of the Optimal Value Function341 4.7.2 Lower Estimates Without the Sigma Term... 350 4.7.3 The Second Order Regular Case... 355 4.7.4 Composite Optimization Problems... 359 4.8 Second Order Analysis in Hölder Stable Cases... 366 4.8.1 Upper Second Order Estimates of the Optimal Value Function366 4.8.2 Lower Estimates and Expansions of Optimal Solutions.. 375 4.8.3 Empty Sets of Lagrange Multipliers... 377 4.8.4 Hölder Expansions for Second Order Regular Problems.. 383 4.9 Additional Results... 385 4.9.1 Equality Constrained Problems... 385 4.9.2 Uniform Approximations of the Optimal Value and Optimal Solutions... 389 4.9.3 Second Order Analysis for Nonisolated Optima... 399 4.10 Second Order Analysis in Functional Spaces... 406 4.10.1 Second Order Tangent Sets in Functional Spaces of Continuous Functions... 407 4.10.2 Second Order Derivatives of Optimal Value Functions... 412 4.10.3 Second Order Expansions in Functional Spaces... 415 5 Additional Material and Applications 423 5.1 Variational Inequalities... 423 5.1.1 Standard Variational Inequalities... 423 5.1.2 Generalized Equations... 430 5.1.3 Strong Regularity... 435 5.1.4 Strong Regularity and Second Order Optimality Conditions 446 5.1.5 Strong Stability... 451 5.1.6 Some Examples and Applications... 453 5.2 Nonlinear Programming... 460 5.2.1 Finite Dimensional Linear Programs... 460 5.2.2 Optimality Conditions for Nonlinear Programs... 465 5.2.3 Lipschitz Expansions of Optimal Solutions... 470 5.2.4 Hölder Expansion of Optimal Solutions... 478 5.2.5 High Order Expansions of Optimal Solutions and Lagrange Multipliers... 484 5.2.6 Electrical Networks... 487 5.2.7 The Chain Problem... 491 5.3 Semi-definite Programming... 496 5.3.1 Geometry of the Cones of Negative Semidefinite Matrices. 497
CONTENTS 1 5.3.2 Matrix Convexity... 503 5.3.3 Duality... 505 5.3.4 First Order Optimality Conditions... 509 5.3.5 Second Order Optimality Conditions... 513 5.3.6 Stability and Sensitivity Analysis... 518 5.4 Semi-infinite Programming... 522 5.4.1 Duality... 524 5.4.2 First Order Optimality Conditions... 533 5.4.3 Second Order Optimality Conditions... 542 5.4.4 Perturbation Analysis... 549 6 Optimal Control 555 6.1 Introduction... 555 6.2 Linear and Semilinear Elliptic Equations... 555 6.2.1 The Dirichlet Problem... 556 6.2.2 Semilinear Elliptic Equations... 562 6.2.3 Strong Solutions... 565 6.3 Optimal Control of a Semilinear Elliptic Equation... 567 6.3.1 Existence of Solutions, First Order Optimality System.. 567 6.3.2 Second Order Necessary or Sufficient Conditions... 571 6.3.3 Some Specific Control Constraints... 576 6.3.4 Sensitivity Analysis... 578 6.3.5 State Constrained Optimal Control Problem... 581 6.3.6 Optimal Control of an Ill-Posed System... 583 6.4 The Obstacle Problem... 587 6.4.1 Presentation of the Problem... 587 6.4.2 Polyhedricity... 588 6.4.3 Basic Capacity Theory... 589 6.4.4 Sensitivity Analysis and Optimal Control... 595 7 Bibliographical Notes 599 7.1 Background Material... 599 7.2 Optimality Conditions... 601 7.3 Stability and Sensitivity Analysis... 603 7.4 Applications... 607 7.4.1 Variational Inequalities... 607 7.4.2 Nonlinear Programming... 609 7.4.3 Semi-definite Programming... 609 7.4.4 Semi-infinite Programming... 610 7.5 Optimal Control... 610