DOCSLIB.ORG

A Tutorial on Convex Optimization II: Duality and Interior Point Methods

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
A Tutorial on Convex Optimization II: Duality and Interior Point Methods A Tutorial on Convex Optimization II: Duality and Interior Point Methods Haitham Hindi Palo Alto Research Center (PARC), Palo Alto, California 94304 email: [email protected] Abstract— In recent years, convex optimization has become a and concepts. For detailed examples and applications, the computational tool of central importance in engineering, thanks reader is refered to [8], [2], [6], [5], [7], [10], [12], [17], to its ability to solve very large, practical engineering problems [9], [25], [16], [31], and the references therein. reliably and efficiently. The goal of this tutorial is to continue the overview of modern convex optimization from where our We now briefly outline the paper. There are two main ACC2004 Tutorial on Convex Optimization left off, to cover sections after this one. Section II is on duality, where we important topics that were omitted there due to lack of space summarize the key ideas the general theory, illustrating and time, and highlight the intimate connections between them. the four main practical applications of duality with simple The topics of duality and interior point algorithms will be our examples. Section III is on interior point algorithms, where focus, along with simple examples. The material in this tutorial is excerpted from the recent book on convex optimization, by the focus is on barrier methods, which can be implemented Boyd and Vandenberghe, who have made available a large easily using only a few key technical components, and yet amount of free course material and freely available software. are highly effective both in theory and in practice. All of the These can be downloaded and used immediately by the reader theory we cover can be readily extended to general conic both for self-study and to solve real problems. programs, such as second order cone programs (SOCP) and semidefinite programs (SDP); see [8] for details. I. INTRODUCTION Notation Our notation is standard [8]. For example, we will The objectives are to continue the overview of modern use dom to denote the domain of a function, int denotes the convex optimization from where our ACC2004 Tutorial on interior of a set, relint denotes the interior of a set relative Convex Optimization [18] left off. Specifically, we review to the smallest affine set containing it (its affine hull), and the role of duality and demonstrate both its practical and () to denote componentwise inequality when applied to theoretical value in convex optimization; describe interior vectors, and semidefinite ordering, when applied to matrices. n point methods for solving convex optimization problems; and S+ denotes the set of symmetric n × n positive semidefinite highlight the intimate connections between duality and the matrices. solution methods. We aim to give an overview of the essential ideas mainly II. DUALITY defining concepts and listing properties without proof. With In this section, we present the basic ideas of duality the exception of minimal narrative comments by the present theory, illustrating along the way its four main practical author, all of the material in this tutorial is excerpted from uses: bounding nonconvex problems; stopping criteria in chapters 5, 9, 10 and 11 of the book Convex Optimization algorithms; decomposition of large problems; and sensitivity by Boyd and Vandenberghe [8], where complete details analysis. with lucid explanations can be found. This will be our main reference in this tutorial. I am deeply indebted to the A. The Lagrange dual function authors, for generously allowing me to use their material in preparing this tutorial. The authors have also made available 1) The Lagrangian: We consider an optimization problem on the internet a large amount of free course material and in the standard form: software [14], [22]. minimize f0(x) The reader is assumed to have a working knowledge of subject to fi(x) ≤ 0; i = 1; : : : ; m (1) linear algebra and basic vector calculus, and some (minimal) hi(x) = 0; i = 1; : : : ; p; exposure to optimization. However, due to its different focus, this paper can be read quite independently of our Part I with variable x 2 Rn, and where the functions from Rn ! R paper [18]. The following references also cover the topics f0, f1; : : : ; fm, and h1; : : : ; hp are the objective, inequality of optimization [26], [24], [1], [3], [4], convex analysis [23], constraints and equality constraints, respectively. We assume m dom p dom [28], [30], [19], [15], and numerical computation [29], [13], its domain D = i=0 fi \ i=1 hi is nonempty, [11], [20]. and denote the optimal value of (1) by p?. For now, we do T T Also, in order keep the paper quite general, we have not assume the problem (1) is convex. tried to not to bias our presentation toward any particular The basic idea in Lagrangian duality is to take the con- audience. Hence, the examples used in the paper are very straints in (1) into account by augmenting the objective simple and intended merely to clarify the optimization ideas function with a weighted sum of the constraint functions. We n m p define the Lagrangian L : R × R × R ! R associated which has inequality constraint functions fi(x) = −xi, i = with the problem (1) as 1; : : : ; n. To form the Lagrangian we introduce multipliers λi for the n inequality constraints and multipliers νi for the equality m p constraints, and obtain L(x; λ, ν) = f0(x) + λifi(x) + νihi(x); T n T L(x; λ, ν) = c x − λixi + ν (Ax − b) i=1 i=1 i=1 X X = −bT ν + (c + AT ν − λ)T x: m p with dom L = D × R × R . We refer to λi as the La- P The dual function is grange multiplier associated with the ith inequality constraint T T T fi(x) ≤ 0; similarly we refer to νi as the Lagrange multiplier g(λ, ν) = inf L(x; λ, ν) = −b ν + inf(c + A ν − λ) x; x x associated with the ith equality constraint hi(x) = 0. The vectors λ and ν are called the dual variables or Lagrange which is easily determined analytically, since a linear function is bounded below only when it is identically zero. Thus, g(λ, ν) = multiplier vectors associated with the problem (1). −∞ except when c + AT ν − λ = 0, in which case it is −bT ν: 2) The Lagrange dual function: We define the Lagrange m p −bT ν AT ν − λ + c = 0 dual function (or just dual function) g : R × R ! R as g(λ, ν) = m −∞ otherwise. the minimum value of the Lagrangian over x: for λ 2 R , ν 2 Rp, Note that the dual function g is finite only on a proper affine m p m p subset of R × R . We will see that this is a common occurrence. The lower bound property (2) is nontrivial only g(λ, ν) = inf f0(x) + λifi(x) + νihi(x) : T x2D when λ and ν satisfy λ 0 and A ν − λ + c = 0. When i=1 i=1 ! T X X this occurs, −b ν is a lower bound on the optimal value of the When the Lagrangian is unbounded below in x, the dual LP (3). function takes on the value −∞. Since the dual function is the pointwise infimum of a family of affine functions B. The Lagrange dual problem of (λ, ν), it is concave, even when the problem (1) is not For each pair (λ, ν) with λ 0, the Lagrange dual convex. function gives us a lower bound on the optimal value p? of 3) Lower bounds on optimal value: It is easy to show [8] the optimization problem (1). Thus we have a lower bound that the dual function yields lower bounds on the optimal that depends on some parameters λ, ν. A natural question value p? of the problem (1): For any λ 0 and any ν we is: What is the best lower bound that can be obtained from have the Lagrange dual function? g(λ, ν) ≤ p?: (2) This leads to the optimization problem maximize g(λ, ν) This important property is easily verified. Suppose x~ is a (4) feasible point for the problem (1), i.e., fi(x~) ≤ 0 and subject to λ 0: hi(x~) = 0, and λ 0. Then we have This problem is called the Lagrange dual problem associated m p with the problem (1). In this context the original problem (1) λifi(x~) + νihi(x~) ≤ 0; is sometimes called the primal problem. The term dual i=1 i=1 feasible, to describe a pair (λ, ν) with λ 0 and g(λ, ν) > X X since each term in the first sum is nonpositive, and each term −∞, now makes sense. It means, as the name implies, that in the second sum is zero, and therefore (λ, ν) is feasible for the dual problem (4). We refer to (λ?; ν?) as dual optimal or optimal Lagrange multipliers if m p L(x;~ λ, ν) = f (x~) + λ f (x~) + ν h (x~) ≤ f (x~): they are optimal for the problem (4). 0 i i i i 0 The Lagrange dual problem (4) is a convex optimization i=1 i=1 X X problem, since the objective to be maximized is concave and Hence the constraint is convex. This is the case whether or not the primal problem (1) is convex. g(λ, ν) = inf L(x; λ, ν) ≤ L(x;~ λ, ν) ≤ f0(x~): x2D 1) Making dual constraints explicit: The example above Since g(λ, ν) ≤ f0(x~) holds for every feasible point x~, the shows that it can happen (and often does) that the domain inequality (2) follows.
Load more

Recommended publications
  • Boosting Algorithms for Maximizing the Soft Margin
    Boosting Algorithms for Maximizing the Soft Margin Manfred K. Warmuth∗ Karen Glocer Gunnar Ratsch¨ Dept. of Engineering Dept. of Engineering Friedrich Miescher Laboratory University of California University of California Max Planck Society Santa Cruz, CA, U.S.A. Santa Cruz, CA, U.S.A. T¨ubingen, Germany Abstract We present a novel boosting algorithm, called SoftBoost, designed for sets of bi- nary labeled examples that are not necessarily separable by convex combinations of base hypotheses. Our algorithm achieves robustness by capping the distribu- tions on the examples. Our update of the distribution is motivated by minimizing a relative entropy subject to the capping constraints and constraints on the edges of the obtained base hypotheses. The capping constraints imply a soft margin in the dual optimization problem. Our algorithm produces a convex combination of hypotheses whose soft margin is within δ of its maximum. We employ relative en- ln N tropy projection methods to prove an O( δ2 ) iteration bound for our algorithm, where N is number of examples. We compare our algorithm with other approaches including LPBoost, Brown- Boost, and SmoothBoost. We show that there exist cases where the numberof iter- ations required by LPBoost grows linearly in N instead of the logarithmic growth for SoftBoost. In simulation studies we show that our algorithm converges about as fast as LPBoost, faster than BrownBoost, and much faster than SmoothBoost. In a benchmark comparison we illustrate the competitiveness of our approach. 1 Introduction Boosting methods have been used with great success in many applications like OCR, text classifi- cation, natural language processing, drug discovery, and computational biology [13].
    [Show full text]
  • On the Dual Formulation of Boosting Algorithms
    On the Dual Formulation of Boosting Algorithms Chunhua Shen, and Hanxi Li Abstract—We study boosting algorithms from a new perspective. We show that the Lagrange dual problems of ℓ1 norm regularized AdaBoost, LogitBoost and soft-margin LPBoost with generalized hinge loss are all entropy maximization problems. By looking at the dual problems of these boosting algorithms, we show that the success of boosting algorithms can be understood in terms of maintaining a better margin distribution by maximizing margins and at the same time controlling the margin variance. We also theoretically prove that, approximately, ℓ1 norm regularized AdaBoost maximizes the average margin, instead of the minimum margin. The duality formulation also enables us to develop column generation based optimization algorithms, which are totally corrective. We show that they exhibit almost identical classification results to that of standard stage-wise additive boosting algorithms but with much faster convergence rates. Therefore fewer weak classifiers are needed to build the ensemble using our proposed optimization technique. Index Terms—AdaBoost, LogitBoost, LPBoost, Lagrange duality, linear programming, entropy maximization. ✦ 1 INTRODUCTION that the hard-margin LPBoost does not perform well in OOSTING has attracted a lot of research interests since most cases although it usually produces larger minimum B the first practical boosting algorithm, AdaBoost, was margins. More often LPBoost has worse generalization introduced by Freund and Schapire [1]. The machine learn- performance. In other words, a higher minimum margin ing community has spent much effort on understanding would not necessarily imply a lower test error. Breiman [11] how the algorithm works [2], [3], [4].
    [Show full text]
  • Lagrangian Duality in Convex Optimization
    Lagrangian Duality in Convex Optimization LI, Xing A Thesis Submitted in Partial Fulfillment of the Requirements for the Degree of Master of Philosophy in Mathematics The Chinese University of Hong Kong July 2009 ^'^'^xLIBRARr SYSTEf^N^J Thesis/Assessment Committee Professor LEUNG, Chi Wai (Chair) Professor NG, Kung Fu (Thesis Supervisor) Professor LUK Hing Sun (Committee Member) Professor HUANG, Li Ren (External Examiner) Abstract of thesis entitled: Lagrangian Duality in Convex Optimization Submitted by: LI, Xing for the degree of Master of Philosophy in Mathematics at the Chinese University of Hong Kong in July, 2009. Abstract In convex optimization, a strong duality theory which states that the optimal values of the primal problem and the Lagrangian dual problem are equal and the dual problem attains its maximum plays an important role. With easily visualized physical meanings, the strong duality theories have also been wildly used in the study of economical phenomenon and operation research. In this thesis, we reserve the first chapter for an introduction on the basic principles and some preliminary results on convex analysis ； in the second chapter, we will study various regularity conditions that could lead to strong duality, from the classical ones to recent developments; in the third chapter, we will present stable Lagrangian duality results for cone-convex optimization problems under continuous linear perturbations of the objective function . In the last chapter, Lagrange multiplier conditions without constraint qualifications will be discussed. 摘要 拉格朗日對偶理論主要探討原問題與拉格朗日對偶問題的 最優值之間“零對偶間隙”成立的條件以及對偶問題存在 最優解的條件，其在解決凸規劃問題中扮演著重要角色， 並在經濟學運籌學領域有著廣泛的應用。 本文將系統地介紹凸錐規劃中的拉格朗日對偶理論，包括 基本規範條件，閉凸錐規範條件等，亦會涉及無規範條件 的序列拉格朗日乘子。 ACKNOWLEDGMENTS I wish to express my gratitude to my supervisor Professor Kung Fu Ng and also to Professor Li Ren Huang, Professor Chi Wai Leung, and Professor Hing Sun Luk for their guidance and valuable suggestions.
    [Show full text]
  • Applications of LP Duality
    Lecture 6 Duality of LPs and Applications∗ Last lecture we introduced duality of linear programs. We saw how to form duals, and proved both the weak and strong duality theorems. In this lecture we will see a few more theoretical results and then begin discussion of applications of duality. 6.1 More Duality Results 6.1.1 A Quick Review Last time we saw that if the primal (P) is max c>x s:t: Ax ≤ b then the dual (D) is min b>y s:t: A>y = c y ≥ 0: This is just one form of the primal and dual and we saw that the transformation from one to the other is completely mechanical. The duality theorem tells us that if (P) and (D) are a primal-dual pair then we have one of the three possibilities 1. Both (P) and (D) are infeasible. 2. One is infeasible and the other is unbounded. 3. Both are feasible and if x∗ and y∗ are optimal solutions to (P) and (D) respectively, then c>x∗ = b>y∗. *Lecturer: Anupam Gupta. Scribe: Deepak Bal. 1 LECTURE 6. DUALITY OF LPS AND APPLICATIONS 2 6.1.2 A Comment about Complexity Note that the duality theorem (and equivalently, the Farkas Lemma) puts several problems related to LP feasibility and solvability in NP \ co-NP. E.g., Consider the question of whether the equational form LP Ax = b; x ≥ 0 is feasible. If the program is feasible, we may efficiently verify this by checking that a \certificate” point satisfies the equations. By taking this point to be a vertex and appealing to Hwk1 (Problem 4), we see that we may represent this certificate point in size polynomial in the size of the input.
    [Show full text]
  • Duality Theory and Sensitivity Analysis
    4 Duality Theory and Sensitivity Analysis The notion of duality is one of the most important concepts in linear programming. Basically, associated with each linear programming problem (we may call it the primal problem), defined by the constraint matrix A, the right-hand-side vector b, and the cost vector c, there is a corresponding linear programming problem (called the dual problem) which is constructed by the same set of data A, b, and c. A pair of primal and dual problems are closely related. The interesting relationship between the primal and dual reveals important insights into solving linear programming problems. To begin this chapter, we introduce a dual problem for the standard-form linear programming problem. Then we study the fundamental relationship between the primal and dual problems. Both the "strong" and "weak" duality theorems will be presented. An economic interpretation of the dual variables and dual problem further exploits the concepts in duality theory. These concepts are then used to derive two important simplex algorithms, namely the dual simplex algorithm and the primal dual algorithm, for solving linear programming problems. We conclude this chapter with the sensitivity analysis, which is the study of the effects of changes in the parameters (A, b, and c) of a linear programming problem on its optimal solution. In particular, we study different methods of changing the cost vector, changing the right-hand-side vector, adding and removing a variable, and adding and removing a constraint in linear programming. 55 56 Duality Theory and Sensitivity Analysis Chap. 4 4.1 DUAL LINEAR PROGRAM Consider a linear programming problem in its standard form: Minimize cT x (4.1a) subject to Ax = b (4.1b) x:::O (4.1c) where c and x are n-dimensional column vectors, A an m x n matrix, and b an m­ dimensional column vector.
    [Show full text]
  • Duality Gap Estimation Via a Refined Shapley--Folkman Lemma | SIAM
    SIAM J. OPTIM. \bigcircc 2020 Society for Industrial and Applied Mathematics Vol. 30, No. 2, pp. 1094{1118 DUALITY GAP ESTIMATION VIA A REFINED SHAPLEY{FOLKMAN LEMMA \ast YINGJIE BIy AND AO TANG z Abstract. Based on concepts like the kth convex hull and finer characterization of noncon- vexity of a function, we propose a refinement of the Shapley{Folkman lemma and derive anew estimate for the duality gap of nonconvex optimization problems with separable objective functions. We apply our result to the network utility maximization problem in networking and the dynamic spectrum management problem in communication as examples to demonstrate that the new bound can be qualitatively tighter than the existing ones. The idea is also applicable to cases with general nonconvex constraints. Key words. nonconvex optimization, duality gap, convex relaxation, network resource alloca- tion AMS subject classifications. 90C26, 90C46 DOI. 10.1137/18M1174805 1. Introduction. The Shapley{Folkman lemma (Theorem 1.1) was stated and used to establish the existence of approximate equilibria in economy with nonconvex preferences [13]. It roughly says that the sum of a large number of sets is close to convex and thus can be used to generalize results on convex objects to nonconvex ones. n m P Theorem 1.1. Let S1;S2;:::;Sn be subsets of R . For each z 2 conv i=1 Si = Pn conv S , there exist points zi 2 conv S such that z = Pn zi and zi 2 S except i=1 i i i=1 i for at most m values of i. Remark 1.2.
    [Show full text]
  • On the Ekeland Variational Principle with Applications and Detours
    Lectures on The Ekeland Variational Principle with Applications and Detours By D. G. De Figueiredo Tata Institute of Fundamental Research, Bombay 1989 Author D. G. De Figueiredo Departmento de Mathematica Universidade de Brasilia 70.910 – Brasilia-DF BRAZIL c Tata Institute of Fundamental Research, 1989 ISBN 3-540- 51179-2-Springer-Verlag, Berlin, Heidelberg. New York. Tokyo ISBN 0-387- 51179-2-Springer-Verlag, New York. Heidelberg. Berlin. Tokyo No part of this book may be reproduced in any form by print, microfilm or any other means with- out written permission from the Tata Institute of Fundamental Research, Colaba, Bombay 400 005 Printed by INSDOC Regional Centre, Indian Institute of Science Campus, Bangalore 560012 and published by H. Goetze, Springer-Verlag, Heidelberg, West Germany PRINTED IN INDIA Preface Since its appearance in 1972 the variational principle of Ekeland has found many applications in different fields in Analysis. The best refer- ences for those are by Ekeland himself: his survey article [23] and his book with J.-P. Aubin [2]. Not all material presented here appears in those places. Some are scattered around and there lies my motivation in writing these notes. Since they are intended to students I included a lot of related material. Those are the detours. A chapter on Nemyt- skii mappings may sound strange. However I believe it is useful, since their properties so often used are seldom proved. We always say to the students: go and look in Krasnoselskii or Vainberg! I think some of the proofs presented here are more straightforward. There are two chapters on applications to PDE.
    [Show full text]
  • Subdifferentiability and the Duality
    Subdifferentiability and the Duality Gap Neil E. Gretsky ([email protected]) ∗ Department of Mathematics, University of California, Riverside Joseph M. Ostroy ([email protected]) y Department of Economics, University of California, Los Angeles William R. Zame ([email protected]) z Department of Economics, University of California, Los Angeles Abstract. We point out a connection between sensitivity analysis and the funda- mental theorem of linear programming by characterizing when a linear programming problem has no duality gap. The main result is that the value function is subd- ifferentiable at the primal constraint if and only if there exists an optimal dual solution and there is no duality gap. To illustrate the subtlety of the condition, we extend Kretschmer's gap example to construct (as the value function of a linear programming problem) a convex function which is subdifferentiable at a point but is not continuous there. We also apply the theorem to the continuum version of the assignment model. Keywords: duality gap, value function, subdifferentiability, assignment model AMS codes: 90C48,46N10 1. Introduction The purpose of this note is to point out a connection between sensi- tivity analysis and the fundamental theorem of linear programming. The subject has received considerable attention and the connection we find is remarkably simple. In fact, our observation in the context of convex programming follows as an application of conjugate duality [11, Theorem 16]. Nevertheless, it is useful to give a separate proof since the conclusion is more readily established and its import for linear programming is more clearly seen. The main result (Theorem 1) is that in a linear programming prob- lem there exists an optimal dual solution and there is no duality gap if and only if the value function is subdifferentiable at the primal constraint.
    [Show full text]
  • Lagrangian Duality
    Lagrangian Duality Richard Lusby Department of Management Engineering Technical University of Denmark Today's Topics (jg Lagrange Multipliers Lagrangian Relaxation Lagrangian Duality R Lusby (42111) Lagrangian Duality 2/30 Example: Economic Order Quantity Model (jg Parameters I Demand rate: d I Order cost: K I Unit cost: c I Holding cost: h Decision Variable: The optimal order quantity Q Objective Function: dK hQ minimize + dc + Q 2 Optimal Order Quantity: r 2dK Q∗ = h R Lusby (42111) Lagrangian Duality 3/30 EOQ Model Consider the following extension(jg Suppose we have several items with a space constraint q The problem P dj Kj hQj minimize + dj cj + j Qj 2 P subject to: Qj q j ≤ We have the following possibilities 1 The constraint is non-binding/slack, i.e r X 2dj Kj < q h j 2 The constraint is binding/active R Lusby (42111) Lagrangian Duality 4/30 Lagrange Multiplier (jg The problem P dj Kj hQj minimize T (Q ; Q ;:::; Qn) = + dj cj + 1 2 j Qj 2 P subject to: j Qj = q Lagrange multiplier λ X minimize T (Q ; Q ;:::; Qn) + λ( Qj q) 1 2 − j Differentiate with respect to Qj : @L dj Kj h = 2 + + λ = 0 j @Qj − Qj 2 8 Solving for Qj r 2dj Kj Qj = j h + 2λ 8 R Lusby (42111) Lagrangian Duality 5/30 Lagrange multiplier Continued(jg Substituting this into the constraint we have r X 2dj Kj = q h + 2λ j Solving for λ and Qj gives: P 2 p2dj Kj j h q − λ = λ∗ = 2 r 2dj Kj Q∗ = j j h + 2λ∗ 8 R Lusby (42111) Lagrangian Duality 6/30 Interpretation of λ (jg In linear programming a dual variable is a shadow price: ∗ ∗ @Z yi = @bi Similarly, in the EOQ
    [Show full text]
  • Arxiv:2011.09194V1 [Math.OC]
    Noname manuscript No. (will be inserted by the editor) Lagrangian duality for nonconvex optimization problems with abstract convex functions Ewa M. Bednarczuk · Monika Syga Received: date / Accepted: date Abstract We investigate Lagrangian duality for nonconvex optimization prob- lems. To this aim we use the Φ-convexity theory and minimax theorem for Φ-convex functions. We provide conditions for zero duality gap and strong duality. Among the classes of functions, to which our duality results can be applied, are prox-bounded functions, DC functions, weakly convex functions and paraconvex functions. Keywords Abstract convexity · Minimax theorem · Lagrangian duality · Nonconvex optimization · Zero duality gap · Weak duality · Strong duality · Prox-regular functions · Paraconvex and weakly convex functions 1 Introduction Lagrangian and conjugate dualities have far reaching consequences for solution methods and theory in convex optimization in finite and infinite dimensional spaces. For recent state-of the-art of the topic of convex conjugate duality we refer the reader to the monograph by Radu Bot¸[5]. There exist numerous attempts to construct pairs of dual problems in non- convex optimization e.g., for DC functions [19], [34], for composite functions [8], DC and composite functions [30], [31] and for prox-bounded functions [15]. In the present paper we investigate Lagrange duality for general optimiza- tion problems within the framework of abstract convexity, namely, within the theory of Φ-convexity. The class Φ-convex functions encompasses convex l.s.c. Ewa M. Bednarczuk Systems Research Institute, Polish Academy of Sciences, Newelska 6, 01–447 Warsaw Warsaw University of Technology, Faculty of Mathematics and Information Science, ul.
    [Show full text]
  • Lagrangian Duality and Perturbational Duality I ∗
    Lagrangian duality and perturbational duality I ∗ Erik J. Balder Our approach to the Karush-Kuhn-Tucker theorem in [OSC] was entirely based on subdifferential calculus (essentially, it was an outgrowth of the two subdifferential calculus rules contained in the Fenchel-Moreau and Dubovitskii-Milyutin theorems, i.e., Theorems 2.9 and 2.17 of [OSC]). On the other hand, Proposition B.4(v) in [OSC] gives an intimate connection between the subdifferential of a function and the Fenchel conjugate of that function. In the present set of lecture notes this connection forms the central analytical tool by which one can study the connections between an optimization problem and its so-called dual optimization problem (such connections are commonly known as duality relations). We shall first study duality for the convex optimization problem that figured in our Karush-Kuhn-Tucker results. In this simple form such duality is known as Lagrangian duality. Next, in section 2 this is followed by a far-reaching extension of duality to abstract optimization problems, which leads to duality-stability relationships. Then, in section 3 we specialize duality to optimization problems with cone-type constraints, which includes Fenchel duality for semidefinite programming problems. 1 Lagrangian duality An interesting and useful interpretation of the KKT theorem can be obtained in terms of the so-called duality principle (or relationships) for convex optimization. Recall our standard convex minimization problem as we had it in [OSC]: (P ) inf ff(x): g1(x) ≤ 0; ··· ; gm(x) ≤ 0; Ax − b = 0g x2S n and recall that we allow the functions f; g1; : : : ; gm on R to have values in (−∞; +1].
    [Show full text]
  • The Idea Behind Duality Lec12p1, ORF363/COS323
    Lec12p1, ORF363/COS323 This lecture Instructor: Amir Ali Ahmadi Fall 2014 Linear programming duality + robust linear programming Intuition behind the derivation of the dual TAs: Y. Chen, • G. Hall, • Weak and strong duality theorems J. Ye ○ Max-flow=Min-cut • Primal/dual possibilities • Interpreting the dual • An application: robust linear programming The idea behind duality • For any linear program (LP), there is a closely related LP called the dual. The feasible and optimal solutions of the dual provide very useful information about the original (aka primal) LP. • In particular, if the primal LP is a maximization problem, the dual can be used to find upper bounds on its optimal value. (Similarly, if the primal is a minimization problem, the dual gives lower bounds.) Note that this is useful in certifying optimality of a candidate solution to the primal. One does not get information of this type from feasible solutions to the primal alone. Let's understand this concept through an example first. Consider this LP (from [DPV08]): max. s.t. • Somebody comes to you and claims with objective value 1900 is optimal. • How can we check his claim? Well, one thing we can do is take combinations of the constraints to produce new "valid inequalities" that upper bound the objective function when evaluated on the feasible set. For example, • This means that it is impossible for the objective function to take value larger than 2000 when evaluated at a feasible point (why?). Can we bring down this upper bound further? Lec12 Page 1 Lec12p2, ORF363/COS323 • What if we try This indeed shows that is optimal.
    [Show full text]