Outline relationship among topics secrets LP with upper

  • Slides: 42
Download presentation
Outline è è è relationship among topics secrets LP with upper bounds è by

Outline è è è relationship among topics secrets LP with upper bounds è by Simplex method è basic è feasible solution (BFS) by Simplex method for bounded variables è extended è optimality conditions for bounded variables è è basic feasible solution (EBFS) ideas of the proof examples Example 1 for ideas but inexact è Example 2 for the exact procedure è 1

A Depot for Multiple Products è multi-product by a fleet of trucks Possible Formulation:

A Depot for Multiple Products è multi-product by a fleet of trucks Possible Formulation: objective function common constraints, e. g. , trucks, DC capacity, etc. network constraints for type 1 product . . depot network constraints for type 1 product non negativity constraints 2

A General Type of Optimization Problems è structure of many problems: è network constraints:

A General Type of Optimization Problems è structure of many problems: è network constraints: easy è other constraints: hard objective function network constraints hard constraints non negativity constraints making use of the easy constraints to solve the problems è solution methods: large-scale optimization è è column generation, Lagrangian relaxation, Dantzig-Wolfe decomposition … è basis: linear programming, network optimization (and also non-linear optimization, integer optimization, combinatorial optimization) 3

Relationship of Solution Techniques è two directions of theoretical development for network programming linear

Relationship of Solution Techniques è two directions of theoretical development for network programming linear prog. èfrom special structures of networks network prog. èfrom linear programming è ideal: understanding development in both directions non linear prog. dynamic prog. … int. prog. 4

Relationship of Solution Techniques minimum cost flow network algorithms shortest path algorithms column generation,

Relationship of Solution Techniques minimum cost flow network algorithms shortest path algorithms column generation, Dantzig Wolfe decomposition network simplex revised simplex method linear algebra Lagrangian relaxation non linear optimization 5

Our Topics è simplex method for bounded variables linkage between LP and network simplex

Our Topics è simplex method for bounded variables linkage between LP and network simplex è optimality conditions for minimum cost flow networks è è minimum cost algorithms standard, and successive shortest path è equivalence among network and LP optimality conditions è è è revised simplex column generation Dantzig-Wolfe decomposition Lagrangian relaxation It takes more than one semester to cover these topics in detail! We will only cover the ideas. 6

Secrets 7

Secrets 7

The Most Beautiful … 8

The Most Beautiful … 8

Maybe the Most Beautiful of All… è linear algebra geometric properties algebraic properties matrix

Maybe the Most Beautiful of All… è linear algebra geometric properties algebraic properties matrix properties 9

LP with Upper Bounds 10

LP with Upper Bounds 10

LP with Upper Bounds è upper bounds: common in network problems, e. g. ,

LP with Upper Bounds è upper bounds: common in network problems, e. g. , an arc with finite capacity è quite some theory of network optimization being from LP 11

To Solve LP with Upper Bounds è incorporate the upper bound constraints into the

To Solve LP with Upper Bounds è incorporate the upper bound constraints into the set of functional constraints and solve accordingly 12

To Solve LP with Upper Bounds è In the simplex method the lower bound

To Solve LP with Upper Bounds è In the simplex method the lower bound constraints 0 x do not appear in A. è Is it possible to work only with A even with upper bound constraints? è Yes. 13

BFS for Standard LP è Am n, m n, of rank m è basic

BFS for Standard LP è Am n, m n, of rank m è basic feasible solution (BFS) x of LP, i. e. , èfeasible: Ax b, 0 x èbasic ènon basic variables: (at least) n m variables = 0 èbasic variables: m non negative variables with linearly independent columns 14

Extended Basic Feasible Solution of LP with Bounded Variables è Am n, m n,

Extended Basic Feasible Solution of LP with Bounded Variables è Am n, m n, of rank m è extended basic feasible solution ( EBFS ) x of LP with bounded variables, i. e. , èfeasible: èbasic Ax b, 0 x u solution ènon basic variables: (at least) n m variables = 0, or = their upper bounds variables: m variables of the form 0 xi ui, with linearly independent columns èBasic 15

Optimality Conditions of Standard LP è Maximum è Conditions: BFS x is maximal if

Optimality Conditions of Standard LP è Maximum è Conditions: BFS x is maximal if 0 for all non basic variable xj = 0 è Minimum Conditions: BFS è x is minimal if 0 for all non basic variable xj = 0 è intuition : increase of the objective function by unit increase in xj è maximum condition: no good to increase non basic xj è minimum condition: no good to decrease non basic xj è 16

Optimality Conditions of LP with Bounded Variables è Maximum Conditions: EBFS x is maximal

Optimality Conditions of LP with Bounded Variables è Maximum Conditions: EBFS x is maximal if è 0 for all non basic variable xj = 0, and è 0 for all non basic variable xj = uj è Minimum Conditions: EBFS x is minimal if è 0 for all non basic variable xj = 0, and è 0 for all non basic variable xj = uj 17

How to Prove? 18

How to Prove? 18

General Idea è optimality conditions of the EBFS èfrom duality theory and complementary slackness

General Idea è optimality conditions of the EBFS èfrom duality theory and complementary slackness conditions 19

Complementary Slackness Conditions è primal dual pair è Theorem 1 (Complementary Slackness Conditions) èif

Complementary Slackness Conditions è primal dual pair è Theorem 1 (Complementary Slackness Conditions) èif x primal feasible and y dual feasible èthen x primal optimal and y dual optimal iff xj(y. TA j cj) = 0 for all j, and yi(bi Ai x) = 0 for all i 20

Complementary Slackness Conditions è primal dual pair è Theorem 2 (Necessary and Sufficient Condition)

Complementary Slackness Conditions è primal dual pair è Theorem 2 (Necessary and Sufficient Condition) èif x primal feasible èthen x primal optimal iff there exists dual feasible y such that x and y satisfy the Complementary Slackness Conditions 21

Complementary Slackness Conditions for LP with Bounded Variables è by Theorem 2, primal feasible

Complementary Slackness Conditions for LP with Bounded Variables è by Theorem 2, primal feasible x and dual feasible (y. T, T) are optimal iff èxj(y. TA j + j cj ) = 0, j èyi(bi Ai x) = 0, i è j(uj xj ) = 0, j 22

General Idea of the Proof è optimality conditions of the EBFS è from duality

General Idea of the Proof è optimality conditions of the EBFS è from duality theory and complementary slackness conditions è ideas of the proof è given an EBFS x satisfying the upper bound optimality conditions è then possible to find dual feasible variables (y. T, T)T such that x and (y. T, T)T satisfy the complementary slackness conditions 23

Example 1. Upper-Bound Constraints as Functional Constraints è max 2 x + 5 y,

Example 1. Upper-Bound Constraints as Functional Constraints è max 2 x + 5 y, min 2 x 5 y, x + 2 y 20, è s. t. è è è 2 x + y 16, 0 x 2, 0 y 8. 24

Examples of LP with Bounded Variables 25

Examples of LP with Bounded Variables 25

Example 1. Upper-Bound Constraints as Functional Constraints è min 2 x 5 y, è

Example 1. Upper-Bound Constraints as Functional Constraints è min 2 x 5 y, è s. t. è x + 2 y 20, è 2 x + y 16, è 0 x 2, 0 y 8. è max. value = 44 è x* = 2 and y* = 8 26

The following procedure is not exactly the Simplex Method for Bounded Variables. It primarily

The following procedure is not exactly the Simplex Method for Bounded Variables. It primarily brings out the ideas of the exact method. 27

Example 1. Upper-Bound Constraints by Optimality Conditions of Bounded Variables 5 èy as the

Example 1. Upper-Bound Constraints by Optimality Conditions of Bounded Variables 5 èy as the entering variable è 2 y + s 1 = 20 è y + s 2 = 16 èy 8 min 2 x 5 y, s. t. x + 2 y 20, 2 x + y 16, 0 x 2, 0 y 8. 28

Example 1. Upper-Bound Constraints by Optimality Conditions of Bounded Variables mark the non basic

Example 1. Upper-Bound Constraints by Optimality Conditions of Bounded Variables mark the non basic variable y at its upper bound è for y = 8 è fun. : 2 x – 5 y – z = 0 2 x z = 40 è eqt. (1): x + 2 y + s 1 = 20 x + s 1 = 4 è eqt. (2): 2 x + y + s 2 = 16 2 x + s 2 = 8 è obj. 29

Example 1. Upper-Bound Constraints by Optimality Conditions of Bounded Variables èx as the entering

Example 1. Upper-Bound Constraints by Optimality Conditions of Bounded Variables èx as the entering variable èx + s 1 = 4 è 2 x èx + s 2 = 8 2 min 2 x 5 y, s. t. x + 2 y 20, 2 x + y 16, 0 x 2, 0 y 8. 30

Example 1. Upper-Bound Constraints by Optimality Conditions of Bounded Variables è for x at

Example 1. Upper-Bound Constraints by Optimality Conditions of Bounded Variables è for x at its upper bound 2, mark x, and è obj. fun. : 2 x – z = 40 z = 44 è eqt. (1): x + s 1 = 4 s 1 = 2 è eqt. (2): 2 x + s 2 = 8 s 2 = 4 min 2 x 5 y, s. t. x + 2 y 20, 2 x + y 16, 0 x 2, 0 y 8. 31

Example 1. Upper-Bound Constraints by Optimality Conditions of Bounded Variables è satisfying variables è

Example 1. Upper-Bound Constraints by Optimality Conditions of Bounded Variables è satisfying variables è è è z* the optimality condition for bounded 0 for all non basic variable xj = 0, and 0 for all non basic variable xj = uj = 44, with x* = 2 and y* = 8 32

Example 1 Being Too Specific è in general, variables swapping among all sorts of

Example 1 Being Too Specific è in general, variables swapping among all sorts of status è non-basic at 0 è basic between 0 and upper bound è basic at upper bound è non-basic at upper bound è Simplex method for bounded variables: a special algorithm to record all possibilities 33

The following example follows the exact procedure of the Simplex Method for Bounded Variables.

The following example follows the exact procedure of the Simplex Method for Bounded Variables. 34

Example 2 è max 3 x 1 + 5 x 2 + 2 x

Example 2 è max 3 x 1 + 5 x 2 + 2 x 3 min 3 x 1 5 x 2 2 x 3, è s. t. è x 1 + x 2 + 2 x 3 7, è 2 x 1 + 4 x 2 + 3 x 3 15, è 0 x 1 4, 0 x 2 3, 0 x 3 3. 35

Example 2 by Simplex Method for Bounded Variables è potential entering variable: x 2

Example 2 by Simplex Method for Bounded Variables è potential entering variable: x 2 è bounded by upper bound 3 è define = u 2 x 2 = 3 x 2 min 3 x 1 5 x 2 2 x 3, s. t. x 1 + x 2 + 2 x 3 7, 2 x 1 + 4 x 2 + 3 x 3 15, 0 x 1 4, 0 x 2 3, 0 x 3 3. 36

Example 2 by Simplex Method for Bounded Variables 37

Example 2 by Simplex Method for Bounded Variables 37

Example 2 by Simplex Method for Bounded Variables è x 1 as the (potential)

Example 2 by Simplex Method for Bounded Variables è x 1 as the (potential) entering variable è s 2 as the leaving variable èa min 3 x 1 5 x 2 2 x 3, s. t. x 1 + x 2 + 2 x 3 7, 2 x 1 + 4 x 2 + 3 x 3 15, 0 x 1 4, 0 x 2 3, 0 x 3 3. pivot operation as in standard Simplex Method 38

Example 2 by Simplex Method for Bounded Variables è which can be an entering

Example 2 by Simplex Method for Bounded Variables è which can be an entering variable? è can s 1 be a leaving variable? Yes è can x 1 be a leaving variable? Yes min 3 x 1 5 x 2 2 x 3, s. t. x 1 + x 2 + 2 x 3 7, 2 x 1 + 4 x 2 + 3 x 3 15, 0 x 1 4, 0 x 2 3, 0 x 3 3. 39

Example 2 by Simplex Method for Bounded Variables è when = 1. 25, x

Example 2 by Simplex Method for Bounded Variables è when = 1. 25, x 1 reaches its upper bound 4 è replace x 1 by and is a basic variable = 0 min 3 x 5 x 2 x , è result s. t. 1 2 3 x 1 + x 2 + 2 x 3 7, 2 x 1 + 4 x 2 + 3 x 3 15, 0 x 1 4, 0 x 2 3, 0 x 3 3. 40

Example 2 by Simplex Method for Bounded Variables è. èa min 3 x 1

Example 2 by Simplex Method for Bounded Variables è. èa min 3 x 1 5 x 2 2 x 3, s. t. x 1 + x 2 + 2 x 3 7, 2 x 1 + 4 x 2 + 3 x 3 15, 0 x 1 4, 0 x 2 3, 0 x 3 3. “normal” pivot operation with aij < 0 41

Example 2 by Simplex Method for Bounded Variables è minimum è z* min 3

Example 2 by Simplex Method for Bounded Variables è minimum è z* min 3 x 1 5 x 2 2 x 3, s. t. x 1 + x 2 + 2 x 3 7, 2 x 1 + 4 x 2 + 3 x 3 15, 0 x 1 4, 0 x 2 3, 0 x 3 3. = 20. 75, x 1* = 4, x 2* = 1. 75, x 3* = 0 42