Sei sulla pagina 1di 21

Lecture 9

Duality and Post-Optimal Analysis


Overview

Dual of an LP Problem (4.1)

Primal-Dual Relationship (4.2.1)

Simplex Table Computations (4.2.24.2.4)

Operations Research 1
LP Dual
The original LP is referred to as primal
The dual is another LP associated with the given LP
It is defined systematically
An LP and its dual are closely related:
an optimal solution of one provides information about an optimal
solution of the other one
Rules of the dual LP formulation: consider LP in a standard form
To each equation assign a variable
The objective coefficients in the dual are given by the right-hand side
of the primal (system of equations): if primal objective is max
the dual objective is min (and v.v)
To each primal variable assign a constraint
The column coefficients of a primal variable define the row coefficients
of a dual constraint, while the objective cost of the variable defines
the right-hand side of the constraint in the dual

Operations Research 2
Rules for Standard Form: MAXIMIZATION
Standard LP (primal) problem
maximize z = c1x1 + c2x2 + + c n xn
subject to a11x1 + a12x2 + + a1nxn = b1 y1
a21x1 + a22x2 + + a2nxn = b2 y2
.. .. .. .. .. ..
am1x1 + am2x2 + + amnxn = bm ym
xj 0, j = 1, . . . , n
Dual problem
minimize v = b1y1 + b2y2 + + bmym
subject to a11y1 + a21y2 + + am1ym c1 (x1)
a12y1 + a22y2 + + am2ym c2 (x2)
.. .. .. .. ..
a1ny1 + a2nx2 + + amnym cn (xn)

Sign is (the same as the sign of primal variables)


The dual variables are unrestricted
Operations Reseaech 3
Rules for Standard Form: MINIMIZATION
Standard LP (primal) problem
minimize z = c1x1 + c2x2 + + c n xn
subject to a11x1 + a12x2 + + a1nxn = b1 y1
a21x1 + a22x2 + + a2nxn = b2 y2
.. .. .. .. .. ..
am1x1 + am2x2 + + amnxn = bm ym
xj 0, j = 1, . . . , n
Dual problem
maximize v = b1y1 + b2y2 + + bmym
subject to a11y1 + a21y2 + + am1ym c1
a12y1 + a22y2 + + am2ym c2
.. .. .. .. ..
a1ny1 + a2nx2 + + amnym cn
Sign is (the opposite of the sign of primal variables)
The dual variables are unrestricted

Operations Reseaech 4
Example

Primal Problem

maximize z = 5x1 + 12x2 + 4x3


subject to x1 + 2x2 + x3 10
2x1 x2 + 3x3 = 8
x1 , x 2 , x 3 0

Bring it to its standard form:

maximize z = 5x1 + 12x2 + 4x3


subject to x1 + 2x2 + x3 + x4 = 10 y1
2x1 x2 + 3x3 = 8 y2
x1 , x 2 , x 3 , x 4 0

Apply the rules and determine its dual


Operations Reseaech 5
Its dual

minimize v = 10y1 + 8y2


subject to y1 + 2y2 5
2y1 y2 12
y1 + 3y2 4
y1 + 0y2 0

Last constraint is equivalent to y1 0

Operations Reseaech 6
Primal Problem Another Example

minimize z = 15x1 + 12x2


subject to x1 + 2x2 3
2x1 4x2 5
x1 x2 = 3
x1 , x 2 0

Bring it to its standard form:

minimize z = 15x1 + 12x2


subject to x1 + 2x2 x3 = 3 y1
2x1 4x2 + x4 = 5 y2
x1 x2 = 3 y3
x1 , x 2 , x 3 , x 4 0

Apply the rules and determine its dual


Operations Reseaech 7
Its dual

maximize v = 3y1 + 5y2 + 3y3


subject to y1 + 2y2 + y3 15
2y1 4y2 y3 12
y1 0
y2 0

There is an alternative way to write the dual from general rules (primal does
not have to be in the standard form)

Operations Reseaech 8
General rules (the book, page 155)

Maximization Minimization
Constraints Variables
0
0
= unrestricted

Variables Constraints
0
0
unrestricted =

No designation primal or dual, applies to the type of problem.

Operations Reseaech 9
Primal-Dual Properties

Dual of the dual is primal


The value of a feasible point in the maximization LP problem is less
than or equal to a value of a feasible point in the minimization problem

Operations Reseaech 10
If one problem is unbounded, the other one is infeasible
If one problem has an optimal solution, so does the other, and the
optimal values are the same
Having an optimal solution for one of them, we can determine a dual
optimal solution for the other
There are two ways to do this
Both use the data from the optimal simplex table: the original problem,
initial basis, and the optimal table
Valid only for the problems in standard form.

Operations Reseaech 11
Determining Dual Optimal Solution from Optimal Simplex Table

Operations Reseaech 12
First Method: Uses Dual Problem and Optimal Simplex Table
Primal
maximize 5x1 + 12x2 + 4x3
subject to x1 + 2x2 + x3 10
2x1 x2 + 3x3 8
x1 , x 2 , x 3 0

Its Optimal Simplex Table

Basic x1 x2 x3 x4 x5 Solution
z 0 0 3/5 29/5 2/5 54&4/5
x2 0 1 1/5 2/5 1/5 12/5
x1 1 0 7/5 1/5 2/5 26/5

How can we find an optimal dual? - Look at the optimal basis and the dual
problem
Operations Reseaech 13
minimize 10y1 + 8y2
maximize 5x1 + 12x2 + 4x3 subject to y1 + 2y2 5 (x1)
subject to x1 + 2x2 + x3 + x4 = 10 2y1 y2 12 (x2)
2x1 x2 + 3x3 + x5 = 8 y1 + 3y2 4 (x3)
x1 , x 2 , x 3 , x 4 , x 5 0 y1 0 (x4)
y2 0 (x5)
Optimal basic variables: x1 and x2
In the dual: take the inequalities corresponding to x1 and x2 and solve
the system in dual variables corresponding to these constraints as equalities:

y1 + 2y2 = 5, 2y1 y2 = 12.

Solution to the system: y1 = 29/5 and y2 = 2/5.


This is an optimal dual solution.

Operations Reseaech 14
Second Method: Initial Simplex Table and Optimal Table

Its Optimal Simplex Table

Basic x1 x2 x3 x4 x5 Solution
z 0 0 3/5 29/5 2/5 54&4/5
x2 0 1 1/5 2/5 1/5 12/5
x1 1 0 7/5 1/5 2/5 26/5

Initial basic variables: x4 and x5


Optimal Inverse Look at the columns corresponding to the initial basis
and write down the resulting matrix:
" #
2/5 1/5
1/5 2/5

Operations Reseaech 15
Take the dual variables corresponding to x4-row and x5-row: y1 and y2
Take the original objective coefficients of the variables in the basis
(respecting the order): 12 and 5
Set up the system:
optimal dual = objective coeff. of the basis optimal inverse
" #
h i h i 2/5 1/5
y1 y2 = 12 5
1/5 2/5

Solve the system to obtain an optimal dual solution

y1 = 29/5, y2 = 2/5

Operations Reseaech 16
Relations in Simplex Table
Given a simplex table at any iteration i, we have
Column Rule
xj -column data = inverse at iteration i original xj -column
Here, column data does not include the z -row values. The rule applies
to rhs-column as well.

Reduced Cost Rule (z -row data)


reduced cost of xj = l.h.s. of j -th dual constraint r.h.s. of j -th
dual constraint
The left hand-side of dual constraint is evaluated at the shadow price

Operations Reseaech 17
Relations in Simplex Table

Operations Reseaech 18
Relations in Simplex Table

Operations Reseaech 18
Relations in Simplex Table

Operations Reseaech
19

Potrebbero piacerti anche