Sensitivity analysis

and Duality of LP
problems
Ha Thi Xuan Chi, PhD

1
 Sensitivity Analysis in LP
How do changes in an LP’s parameters (objective function coefficients, right-
hand sides, and technological coefficients) change the optimal solution?

Types of Sensitivity Analysis

■ Changes in the objective function coefficients

■ Changes in Right-hand-side values (RHS)
■ Changes in the constraint coefficients
■ Addition of a new constraint
■ Addition of a new decision variable
■ Often involves a series of what-if question

2
NOTATIONS

S*=B-1
A*= B-1. A=S*.A
y*=cB B-1

z*=cB B-1 A = y*.A

z*-c=cB B-1 A-c= y*.A-c

Z*= cB B-1 b= y*.b

b*= B-1 b=S*.b

3
 Consider the following example:

Maximize Profit = $50 X1 + $120 X2

Subject to:
2 X1 + 4 X2  80
60
3 X1 + X2  60
X1, X2  0
• Optimal solution at point A
• X1 = 0; X2 = 20; Profit = 2400
• Basic variables: X2=20 , S2 = 40
A 20
• Non-basic: X1= S1=0. 50 X1 + 120 X2 =2400

20 40 50

4
Changing in the objective function coefficients, c

5
Changing in the objective function coefficients, c (cont.)
Final simplex table:

50 120 0 0
Cj Basic Eq. Z X1 X2 S1 S2 RHS
Z (0) 1 10 0 30 0 2400
120 X2 (1) 0 0.5 1 0.25 0 20
0 S2 (2) 0 2.5 0 -0.25 1 40

a. Change coefficients of Nonbasic variables: x1

z *1 c1  z *1 c1  (c1  c1)
2
 y *. A  c1  30 0    50  c1
3 
 10  c1  0
 c1  10
 c1  60
The optimal does not change if c1  60
6
■ Changing in coefficient of the basic variable

50 C(X2) 0 0
Basic Eq. Z X1 X2 S1 S2 RHS
Z (0) 1 10 0 30 0 2400
C(X2) X2 (1) 0 0.5 1 0.25 0 20
0 S2 (2) 0 2.5 0 -0.25 1 40

The optimality will be unchanged if and only if Zj – Cj ≥ 0.

• With X1: 0.5 × C2 + 0 × 2.5 – 50 ≥ 0 => C2 ≥ 100.

• With S1: 0.25 × C2+ 0 × (-0.25) – 0 ≥ 0 => C2 ≥ 0.

• Combine both conditions, we have 100 ≤ C2

7
 Changing in coefficient of the basic variable
50 C(X2) 0 0
Basic Eq. Z X1 X2 S1 S2 RHS
Z (0) 1 10 0 30 0 2400
C(X2) X2 (1) 0 0.5 1 0.25 0 20
0 S2 (2) 0 2.5 0 -0.25 1 40
Consider the optimal simplex tableau
Row 0 : 10 0 30 0 
Change c2  120 to c2  120  c2
New row 0 : 10 0-c2 30 0 
For x 2 to stay a basic z 2 *-c2 must be equal to 0.
Gaussian elimination:
Row 0: 10 0-c2 30 0  
c2  Row 1: 0.5c2 c2 0.25c2 0  =
10  0.5c2 0 0.25c2  30 0 
10  0.5c2  0 c2  20
The current Basic Variable     c2  20  c2  100
0.25c2  30  0 c2  120
remain optimal if
8
Changing the right hand side, bi(cont.)

9
 Changing the right hand side, bi (cont.)

■ The only revision in the model is the changes in RHS.

■ RHS of final row 0: z*= y*b
■ RHS at final rows 1,...,m = S*b
■
Basic Eq. Z X1 X2 S1 S2 RHS
Z (0) 1 10 0 30 0 2400
C(X2) X2 (1) 0 0.5 1 0.25 0 20
0 S2 (2) 0 2.5 0 -0.25 1 40

b1 *   20   0.25 0   b1 

b *  B b  B b  B (b  b )   40    0.25 1   b 
1 1 1

 2       2
if b1 change to b1  b1  b1
b1 *   20   0.25 0   b1   20  0.25b1 
b *   40    0.25 1      40  0.25b 
 2      0   1
10
Changing the right hand side, bi (cont.)

0  b1  240

11
The duality in LP

■ Two of the most important topics in linear programming are

sensitivity analysis and duality.

■ Every linear programming problem has associated with it

another linear programming problem called the dual.

12
 The duality in LP

Primal problem Dual problem

n
Maximize Z=  c j x j
m
Minimize W=  b j y j
j 1
i 1
subject to : subject to :
n

a x
m

j 1
ij j  b, for i  1, 2,..., m a
i 1
ij yi  c, for j  1, 2,..., n

and x j  0, for j  1, 2,..., n and yi  0, for i  1, 2,..., m

13
WYNDOR GLASS CO.
PROBLEM

 Glass products : windows and glass doors.

 Plant 1: Aluminum frames and hardware : Product 1
 Plant 2: Wood frame: Product 2
 Plant 3: The glass and assembles the products: Product 1 & 2.
 Product 1: An 8-foot glass door with aluminum framing
 Product 2: A 4x6 foot double-hung wood-framed window
 WYNDOR GLASS CO problem: Determine what the production rates
should be for the two products in order to maximize their total profit

14
 WYNDOR GLASS CO Data

Tuesday,
September 15
 WYNDOR GLASS CO Data

Primal problem Dual problem

Maximize Z = 3x1 + 5x2 Minimize W = 4y1 + 12y2 +18y3
Subject to Subject to
x1 ≤4 y1 +3y3 ≥ 3
2x2 ≤ 12 2y2 +2y3 ≥ 5
3x1 + 2x2 ≤ 18 y1 ≥0
x1 ≥0 y2 ≥0
x2 ≥ 0. y3 ≥0

16
 WYNDOR GLASS CO Data

Row 0 and corresponding dual solution for each iteration for the
Wyndor Glass Co. example

17
The Duality in LP

■ The Dual
– An alternate formulation of a linear programming problem as
either the original problem or its mirror image, the dual, which
can be solved to obtain the optimal solution.
– Its variables have a different economic interpretation than the
original formulation of the linear programming problem (the
primal).
– It can be easily used to determine if the addition of another
variable to a problem will change the optimal.

18
The Duality in LP

■ Dual
– The number of decision variables in the primal is equal to the
number of constraints in the dual.
– The number of decision variables in the dual is equal to the
number of constraints in the primal.
– Since it is computationally easier to solve problems with less
constraints in comparison to solving problems with less
variables, the dual gives us the flexibility to choose which
problem to solve.

19
The Duality in LP

Primal (Dual) Dual (Primal)

Maximization Profit Minimization Opportunity Cost

Constraints type ≤ Constraints type ≥
Constraints: resources Constraints: Product profits
Max. Profit = $50 X1 + $120 X2 Min. Opportunity Cost = 80 U1 + 60
Subject to: U2
Subject to:
2 X1 + 4 X2  80 ( U1) 2 U1 + 3 U2 ≥ 50 ( X1)
3 X1 + X2  60 ( U2) 4 U1 + 1 U2 ≥ 120 ( X2)
X1, X2  0 U1, U2  0

20
The Duality in LP

■ Steps to form a Dual:

– If the primal is maximization, the dual is a minimization, and
vice versa
– The RHS values of the primal (dual) constraints become the
dual’s (primal’s) objective function coefficients.
– The transpose of the primal constraint coefficients become the
dual constraint coefficients.
– Constraints inequality signs are reversed.

21
 The Duality in LP

Primal Standard primal Dual

Max Z = 5x1 + 12 x2 + 4x3 Max Z = 5x1 + 12 x2 + 4x3 + 0x4 Min Y = 10 U1+8U2
St. St. St.
x1 + 2x2 + x3 ≤ 10 x1 + 2x2 + x3 + x4 = 10 (U1) 1U1 + 2U2 ≥ 5
2x1 – 1x2 + 3x3 = 8 2x1 – 1x2 + 3x3 + 0x4 = 8 (U2) 2U1 - 1U2 ≥ 12
x1,x2, x3 ≥ 0 x1, x2, x3, x4 ≥ 0 1U1 + 3U2 ≥ 4
U1 ≥0
U2 unrestricted

22
 Solving Dual problem
Primal (Dual) Dual (Primal)
Max. Profit = $50 X1 + $120 X2 Min. Opportunity Cost = 80 U1 + 60 U2
Subject to: Subject to:
2 X1 + 4 X2  80 2 U1 + 3 U2 ≥ 50
3 X1 + X2  60 4 U1 + 1 U2 ≥ 120
Primal optimal solution Dual optimal solution
Cj Solut X1 X2 S1 S2 Quan- Cj Solut U1 U2 S1 S2 A1 A2 Quan
ion tity ion -tity
120 X2 1/2 1 1/4 0 20 80 U1 1 1/4 0 -1/4 0 1/2 30
0 S2 5/2 0 -1/4 1 40 0 S1 0 -5/2 1 -1/2 -1 ½ 10
Zj 60 120 30 0 2400 Zj 80 20 0 -20 0 40 2400
Cj-Zj -10 0 -30 0 Cj-Zj 0 40 0 20 M M-40
X1=0; X2= 20; S1=0; S2 = 40 U1= 30; U2= 0; S1= 10; S2= 0

- Absolute values of number in the (Cj-Zj) row under slack variables are the solutions of the
dual (Ui’s). These are shadow prices.
- Optimal value of objective functions of both problems are equal. Always we have Obj. value
of max. problem ≤ Obj. value of min.problem.

23
The Duality in LP

24