Introduction to Optimization, DIKU Monday 19 November David Pisinger. Duality, motivation

Similar documents
LP in Standard and Slack Forms

1 The Primal and Dual of an Optimization Problem

LINEAR PROGRAMMING. Introduction. Prototype example. Formulation of the LP problem. Excel Solution. Graphical solution

Linear Programming and the Simplex Method

Integer Linear Programming

Optimization Methods MIT 2.098/6.255/ Final exam

The Simplex algorithm: Introductory example. The Simplex algorithm: Introductory example (2)

Lecture 19. Curve fitting I. 1 Introduction. 2 Fitting a constant to measured data

15.081J/6.251J Introduction to Mathematical Programming. Lecture 21: Primal Barrier Interior Point Algorithm

Research Article Extension of Wolfe Method for Solving Quadratic Programming with Interval Coefficients

Linear Programming! References! Introduction to Algorithms.! Dasgupta, Papadimitriou, Vazirani. Algorithms.! Cormen, Leiserson, Rivest, and Stein.

The Hypergeometric Coupon Collection Problem and its Dual

Keywords: duality, saddle point, complementary slackness, Karush-Kuhn-Tucker conditions, perturbation function, supporting functions.

: Transforms and Partial Differential Equations

Summary: CORRELATION & LINEAR REGRESSION. GC. Students are advised to refer to lecture notes for the GC operations to obtain scatter diagram.

Probabilistic Analysis of Rectilinear Steiner Trees

Some remarks on the paper Some elementary inequalities of G. Bennett

Definitions and Theorems. where x are the decision variables. c, b, and a are constant coefficients.

Chapter 2. Asymptotic Notation

Optimization Methods: Linear Programming Applications Assignment Problem 1. Module 4 Lecture Notes 3. Assignment Problem

Statistics and Data Analysis in MATLAB Kendrick Kay, February 28, Lecture 4: Model fitting

School of Mechanical Engineering Purdue University. ME375 Transfer Functions - 1

subject to A 1 x + A 2 y b x j 0, j = 1,,n 1 y j = 0 or 1, j = 1,,n 2

CSI 2101 Discrete Structures Winter Homework Assignment #4 (100 points, weight 5%) Due: Thursday, April 5, at 1:00pm (in lecture)

X. Perturbation Theory

1 Duality revisited. AM 221: Advanced Optimization Spring 2016

Supplementary Material

ENGI Series Page 6-01

Name Period ALGEBRA II Chapter 1B and 2A Notes Solving Inequalities and Absolute Value / Numbers and Functions

Time-Domain Representations of LTI Systems

Zeros of Polynomials

and then substitute this into the second equation to get 5(11 4 y) 3y

Optimally Sparse SVMs

Online Learning & Game Theory

Complex Analysis Spring 2001 Homework I Solution

LC Oscillations. di Q. Kirchoff s loop rule /27/2018 1

10-701/ Machine Learning Mid-term Exam Solution

Brief summary of linear programming and duality: Consider the linear program in standard form. (P ) min z = cx. x 0. (D) max yb. z = c B x B + c N x N

Some Examples on Gibbs Sampling and Metropolis-Hastings methods

Weight Moving Average = n

Markov Decision Processes

We are mainly going to be concerned with power series in x, such as. (x)} converges - that is, lims N n

Introduction to Optimization Techniques. How to Solve Equations

Inverse Matrix. A meaning that matrix B is an inverse of matrix A.

5.6 Binomial Multi-section Matching Transformer

Data Movement in Flash Memories

Transfer Function Analysis

Recursive Algorithms. Recurrences. Recursive Algorithms Analysis

MASSACHUSETTS INSTITUTE OF TECHNOLOGY

Section 1 of Unit 03 (Pure Mathematics 3) Algebra

September 2012 C1 Note. C1 Notes (Edexcel) Copyright - For AS, A2 notes and IGCSE / GCSE worksheets 1

Overview of course. Introduction to Optimization, DIKU Monday 12 November David Pisinger

(s)h(s) = K( s + 8 ) = 5 and one finite zero is located at z 1

w (1) ˆx w (1) x (1) /ρ and w (2) ˆx w (2) x (2) /ρ.

Support vector machine revisited

CALCULUS BASIC SUMMER REVIEW

Scheduling under Uncertainty using MILP Sensitivity Analysis

International Journal of Mathematical Archive-4(9), 2013, 1-5 Available online through ISSN

Recurrence Relations

Today in Physics 217: separation of variables IV

( ) = p and P( i = b) = q.

A Solution Procedure to Solve Multi objective Fractional Transportation Problem

Frequency Domain Filtering

Ma/CS 6a Class 22: Power Series

Boosting. Professor Ameet Talwalkar. Professor Ameet Talwalkar CS260 Machine Learning Algorithms March 1, / 32

x c the remainder is Pc ().

Summary and Discussion on Simultaneous Analysis of Lasso and Dantzig Selector

P.3 Polynomials and Special products

3.2 Properties of Division 3.3 Zeros of Polynomials 3.4 Complex and Rational Zeros of Polynomials

Solutions for the Exam 9 January 2012

19.1 The dictionary problem

Lebesgue Constant Minimizing Bivariate Barycentric Rational Interpolation

( ) (( ) ) ANSWERS TO EXERCISES IN APPENDIX B. Section B.1 VECTORS AND SETS. Exercise B.1-1: Convex sets. are convex, , hence. and. (a) Let.

FUZZY TRANSPORTATION PROBLEM WITH ADDITIONAL RESTRICTIONS

We have also learned that, thanks to the Central Limit Theorem and the Law of Large Numbers,

Bertrand s postulate Chapter 2

Topic 1 2: Sequences and Series. A sequence is an ordered list of numbers, e.g. 1, 2, 4, 8, 16, or

Algorithms and Data Structures 2014 Exercises and Solutions Week 13

The Binomial Multi-Section Transformer

MEI Casio Tasks for Further Pure

Answer Key, Problem Set 1, Written

Lecture 11. Solution of Nonlinear Equations - III

5.6 Binomial Multi-section Matching Transformer

Define a Markov chain on {1,..., 6} with transition probability matrix P =

Summer MA Lesson 13 Section 1.6, Section 1.7 (part 1)

A.1 Algebra Review: Polynomials/Rationals. Definitions:

Lecture 10: Bounded Linear Operators and Orthogonality in Hilbert Spaces

SNAP Centre Workshop. Basic Algebraic Manipulation

Game-Theoretic Analysis of the Nodal Pricing Mechanism for Electricity Markets

Differentiable Convex Functions

GRAPHING LINEAR EQUATIONS. Linear Equations ( 3,1 ) _x-axis. Origin ( 0, 0 ) Slope = change in y change in x. Equation for l 1.

Orthogonal Functions

Analysis of Algorithms. Introduction. Contents

NUMERICAL METHODS FOR SOLVING EQUATIONS

Note that the argument inside the second square root is always positive since R L > Z 0. The series reactance can be found as

MIDTERM 3 CALCULUS 2. Monday, December 3, :15 PM to 6:45 PM. Name PRACTICE EXAM SOLUTIONS

NICK DUFRESNE. 1 1 p(x). To determine some formulas for the generating function of the Schröder numbers, r(x) = a(x) =

Chapter 7: The z-transform. Chih-Wei Liu

Q-BINOMIALS AND THE GREATEST COMMON DIVISOR. Keith R. Slavin 8474 SW Chevy Place, Beaverton, Oregon 97008, USA.

Math 475, Problem Set #12: Answers

Transcription:

Itroductio to Optiizatio, DIKU 007-08 Moday 9 Noveber David Pisiger Lecture, Duality ad sesitivity aalysis Duality, shadow prices, sesitivity aalysis, post-optial aalysis, copleetary slackess, KKT optiality coditio, CPLEX sesitivity Taha sectio.6.,.,.,.,.5 all exaples ca be read briefly Duality, otivatio Lookig for a upper boud o LP: axiize x + x subject to x + x x + x 5 x,x 0 Multiply secod costrait by two gives upper boud 0 x + x 0 Add first ad secod costrait gives upper boud 9 x + x 9 Liear cobiatio of costraits rewritte to Dual proble y (x + x )+y (x + x ) y + 5y (y + y )x +(y + y )x y + 5y iiize y + 5y subject to y + y y + y y,y 0 Duality Prial proble Dual proble ax j= c j x j s.t. j= a i j x j b i i =,..., x j 0 j =,..., i i= b i y i s.t. i= a i j y i c j j =,..., y i 0 i =,..., Strog duality For every bouded, feasible LP with prial solutio x, there exists dual feasible y such that j= c j x j = Proof: Show at VA course. i= b i y i Weak duality For every prial feasible x, dual feasible y Proof j= c j x j j=( i= j= c j x j a i j y i )x j = i= b i y i i=( j= a i j x j )y i i= b i y i If x prial feasible, y dual feasible, j= c j x j = i= b i y i the x is prial optial, ad y is dual optial.

Dualizatio (Taha table.) Prial-Dual Relatio Maxiizatio proble Miiizatio Proble Costraits Variables 0 0 = urestricted Variables Costraits 0 0 urestricted = Exaple ax x + x + 5x s.t. 5x + x + x = 8 (y ) x + x + 8x (y ) 6x + 7x + x (y ) x (y ) x 0 x,x R Dual Optial Ifeasible Ubouded Optial Possible Ipossible Ipossible Prial Ifeasible Ipossible Possible Possible Ubouded Ipossible Possible Ipossible i 8y + y + y + y s.t. 5y + y + 6y + y = (x ) y + y + 7y = (x ) y + 8y + y 5 (x ) y 0 y 0 y 0 y R 5 6 Ecooic iterpretatio of Dual variables (Taha..) y i, dual variables, shadow prices, argial prices If costrait i express a liit o resource i the dual variable y i represets the worth per uit of resource i If prial proble ax j= c j x j s.t. j= a i j x j b i i =,..., x j 0 j =,..., has optial solutio value z. The ax j= c j x j s.t. j= a i j x j b i + ε i i =,..., x j 0 j =,..., Reddy Mikks Prial Dual ax z = 5x + x s.t. 6x + x (resource M) x + x 6 (resource M) x + x (arket) x (dead) x,x 0 i w = y + 6y + y + y s.t. 6y + y y 5 y + y + y + y y,y,y,y 0 has optial solutio value z + i= y i ε i The chages eed to be sufficiet sall, such that curret solutio reais optial Weak duality theore: su of profits su of worth of resources x = (, ) y = (,,0,0) z = 7 8

Reddy Mikks Copleetary slackess (Core probl. 9-) profit of icreasig M oe uit z G z C 6 = 0 = prial (P) ax cx s.t. Ax b x 0 dual (D) i yb s.t. ya c y 0 Assue x feasible to (P) ad y feasible to (D) Necessary ad sufficiet coditios for x,y to be optial: ( j= a i j x j = b i or y i = 0) for i =,..., ad ( i= a i j y i = c j or x j = 0) for j =,..., Ecooical iterpretatio profit of icreasig M oe uit z C z B 6 9 = 0 = j= a i j x j < b i y i = 0 If resource b i is ot fully used, the o additioal profit ca be obtaied by icreasig the resource (i.e. y i = 0). i= a i j y i > c j x j = 0 If costs of resources for producig ite j exceed sellig cost, the we should ot produce ite j (i.e. x j = 0). 0 KKT optiality coditio Karush-Kuh-Tucker optiality coditio for geeral o-liear costraied iiizatio proble For LP-probles ax cx s.t. Ax b x 0 the KKT optiality coditio becoes Ax b, x 0 (prial feasible) ya c, y 0 (dual feasible) i yb s.t. ya c y 0 y(ax b) = 0, x(c ya) = 0 (copleetary slack) Fidig the dual variables if the ith costrait is a iequality the the y i will appear i the colu associated with the correspodig slack variable. if the ith costrait is a iequality the the absolute value of y i will appear i the colu associated with the correspodig surplus variable. Exaple, Reddy Mikks Iteratio 0: Iteratio : y = (,,0,0) basic z x x x x x 5 x 6 solutio z 5 0 0 0 0 0 x 0 6 0 0 0 x 0 0 0 0 6 x 5 0 0 0 0 x 6 0 0 0 0 0 basic z x x x x x 5 x 6 solutio z 0 0 0 0 x 0 0 0 0 x 0 0 0 8 0 0 x 5 0 0 0 8 5 5 0 x 6 0 0 8 0

Whe Siplex teriates Fidig the dual variables i geeral Assue that proble is bouded ad feasible. Whe siplex teriates we have solutio value objective fuctio reduced costs basis equatios z = c B A B b z+(c B A B A N c N )x N = (c B A B )b c = c B A B A N c N 0 x B + A B A Nx N = A B b oegativity of all variables x 0 (P) ax cx s.t. Ax = b x 0 (D) i yb s.t. ya c Assue that x is a optial solutio. The dual variables ca be foud as y = c B A B Let y = c B A B, verify strog duality theore: yb = c B A B b = z ya = c B A B A = c B A B (A B A N = ( c B c B A B A ) N (c B c N ) = c sae solutio dual feasible Uderstadig Siplex Tableau (Taha..) Optial Dual Solutio (Taha..) Method ( Optial ) value ( ) Optial prial z-coefficiet of of dual variable y i = startig variable x i + origial objective coefficiet x i z+(c B A B A N c N )x N + (c B A B I d) s = (c BA B )b ax cx+ds s.t. Ax+Is = b x,s 0 basic solutio i Siplex A B x B + A N x N + Is = b x B + A B A N A B s = A B b objective fuctio z+(c B A B A N c N )x N +(c B A B I d)s = (c BA B )b Method ( ) Optial values of dual = variables Row vector of origial objective coefficiets of optial prial basic variables y = c B A B ( ) Optial prial iverse 5 6

Exaple Sesitivity Aalysis (Taha.6) Prial ax z = 5x + x + x MR s.t. x + x + x + x = 0 x x + x + R = 8 x,x,x,x,r 0 Dual i w = 0y + 8y s.t. y + y 5 y y y + y y 0 y M Fial Siplex table basic x x x x R solutio 9 z 0 0 5 5 5 + M 5 5 x 0 5 5 5 5 7 6 x 0 5 5 5 5 Method y = 9 5 + 0 = 9 5, y = 5 + M +( M) = 5 Method (y,y ) = (,5)( 5 5 5 5 ) = ( 9 5, 5 ) Sesitivity Aalysis: fid liits of paraeters such that curret solutio reais optial Post-optial Aalysis: deterie ew optiu resultig fro targeted chages i iput data 7 8 Chages i the right-had side (Taha.6.) Let D i be chage to costrait i Add D i to right-had side Write up feasibility coditios for curret solutio to reai optial Deterie liits of D i for which curret solutio reais optial TOYCO odel ax z = x + x + 5x s.t. x + x + x 0 x + x 60 x + x 0 x,x,x 0 Fial Siplex table basic x x x x x 5 x 6 solutio z 0 0 0 50 x 0 0 00 x 0 0 0 0 x 6 0 0 0 Slack variables x,x 5,x 6, their coefficiets are dual variables y =,y =,y = 0 Feasibility Rage ax z = x + x + 5x s.t. x + x + x 0+D x + x 60+D x + x 0+D x,x,x 0 i atrix for after addig slack variables basic solutio i Siplex optial table Ax = b+d A B x B + A N x N = b+d x B + A B A Nx N = A B (b+d) x = 00+ D D x = 0+ D x 6 = 0 D + D + D Curret solutio reais feasible as log as x,x,x 6 0 9 0

Feasibility Rage Chages i the objective fuctio (Taha.6.) Chagig operatio tie fro 0 to 0+D x = 00+ D 0 x = 0 0 x 6 = 0 D 0 Iplyig 00 D 0 Chagig operatio tie fro 60 to 60+D Iplyig 0 D 00 Chagig operatio tie fro 0 to 0+D Iplyig 0 D Checkig feasibility of chage Checkig whether D = 0, D =, D = 0 feasible x = 00+ D D = 88 0 x = 0+ D = 0 x 6 = 0 D + D + D = 78 0 Let d i be chage to coefficiet c i Add d i to respective coefficiet Write up feasibility coditios for curret solutio to reai optial Deterie liits of d i for which curret solutio reais optial TOYCO odel ax z = x + x + 5x s.t. x + x + x 0 x + x 60 x + x 0 x,x,x 0 Fial Siplex table basic x x x x x 5 x 6 solutio z 0 0 0 50 x 0 0 00 x 0 0 0 0 x 6 0 0 0 Objective Fuctio ax z = (+d )x + (+d )x + (5+d )x s.t. x + x + x 0 x + x 60 x + x 0 x,x,x 0 Siplex algorith, fids objective fuctio z+((c B + d B )A B A N (c N + d N )x N = ((c B + d B )A B )b Curret solutio reais optial as log as reduced costs are oegative for all obasis variables c N = (c B + d B )A B A N (c N + d N ) 0 reduced costs ca be rewritte c N = (c B+ d B )A B A N (c N + d N ) = c N + d B A B A N d N = c N + d B A N d N where A N is the cotets of the fial siplex table TOYCO odel Fial siplex table with d-values at appropriate places d d d 0 0 0 basic x x x x x 5 x 6 solutio z 0 0 0 50 d x 0 0 00 d x 0 0 0 0 0 x 6 0 0 0 curret solutio reais optial as log as reduced costs c = d + d d 0 c = + d 0 c 5 = d + d 0 axiize z = (+d )x + x + 5x coditio d 0 0 0 givig < d axiize z = x +(+d )x + 5x givig < d 8 axiize z = x + x +(5+d )x givig 8 < d

Post-optiality aalysis (Taha.5) Post-optiality aalysis Periodic recalculatio of the optiu solutio. Post-optiality aalysis deteries the ew solutio i a efficiet way. Coditio after paraeters chage Curret solutio reais optial ad feasible Curret solutio becoes ifeasible Curret solutio becoes ooptial Curret solutio becoes ooptial ad ifeasible Recoeded actio No further actio ecessary Use dual siplex to recover feasibility Use prial siplex to recover feasibility Use geeralized siplex to obtai ew solutio TOYCO Exaple.5- ax z = x + x + 5x s.t. x + x + x 0 x + x 60 x + x 0 x,x,x 0 Solutio x = 0, x = 00, x = 0. Add costrait x + x + x 500 Redudat Add costrait x + x + x 500 Ru dual siplex 5 6 Dual Siplex Dual siplex starts fro (better tha) optial ifeasible basis solutio. Searches for feasible solutio. leavig variable: x r is the basic variable havig ost egative value. optiality criteria: if all basic variables are oegative eterig variable: obasic variable with a r j < 0 { } c i,a r j < 0 Exaple Siplex table a r j i z = x + x + x s.t. x x x x x x 6 x + x + x x,x,x 0 basic x x x x x 5 x 6 solutio z 0 0 0 0 x 0 0 x 5 0 0 6 x 6 0 0 7 All reduced costs c j 0. Ifeasible Leavig variable x 5 as 6 sallest. { } c i,a r j < 0 = i {, }, = a r j Eterig variable x Iteratio : basic x x x x x 5 x 6 solutio z 5 0 0 0 x 0 0 x 0 0 x 6 0 0 Leavig variable x Eterig variable { 5 i, }, = Iteratio : basic x x x x x 5 x 6 solutio z 0 0 9 0 0 x 0 x 0 0 0 0 x 6 0 0 0 0 Optial ad feasible 8

Sesitivity with CPLEX Sesitivity with CPLEX TOYCO exaple axiize x + x + 5 x subject to x + x + x <= 0 x + x <= 60 x + x <= 0 bouds x >= 0 x >= 0 x >= 0 ed Solutio Dual siplex - Optial: Objective =.500000000e+0 Solutio tie = 0.00 sec. Iteratios = () CPLEX> dis se obj - OBJ Sesitivity Rages Variable Nae Reduced Cost Dow Curret Up x -.0000 -ifiity.0000 7.0000 x zero zero.0000 0.0000 x zero. 5.0000 +ifiity CPLEX> dis se rhs - RHS Sesitivity Rages Costrait Nae Dual Price Dow Curret Up c.0000 0.0000 0.0000 0.0000 c.0000 0.0000 60.0000 860.0000 c zero 00.0000 0.0000 +ifiity CPLEX> dis sol var - Variable Nae Solutio Value x 00.000000 x 0.000000 All other variables i the rage - are zero. CPLEX> dis sol dua - Costrait Nae Dual Price c.000000 c.000000 All other dual prices i the rage - are zero. 9 0