OPTIMIZATION TECHNIQUES

MODULE - 3

1
 OPTIMIZATION TECHNIQUES

Module 3: Advanced Linear Programming

• Revised Simplex Method

• Duality in LP
• Primal-Dual Relations
• Dual Simplex Method
• Sensitivity Analysis

2
 REVISED SIMPLEX METHOD

What is Revised Simplex Method?

• Revised Simplex Method is an optimized version of the Simplex Method for solving Linear Programming
(LP) problems.
• Uses matrix-based computations to improve efficiency compared to the standard simplex method.
• Instead of maintaining the entire tableau, it stores and updates only essential information.
• Reduces computational burden, especially for large-scale LP problems.
• More efficient in terms of memory and time complexity than the standard simplex method.

3
 REVISED SIMPLEX METHOD

Uses of Revised Simplex Method?

• Efficiency: The revised simplex method is more efficient in terms of computation because it focuses on
updating only the necessary parts of the tableau, rather than the entire tableau.
• Numerical Stability: By reducing the amount of data processed at each iteration, the revised simplex
method improves numerical stability and reduces the chances of computational errors.
• Memory Utilization: It requires less memory compared to the classical simplex method, making it more
suitable for large-scale linear programming problems.
• Simplicity in Implementation: The revised simplex method simplifies the implementation process by
allowing the use of matrix operations, which are well-supported by modern computer software and
hardware.

4
 REVISED SIMPLEX METHOD

Revised Simplex Method Algorithm

Step 1:
Introduce Slack or Surplus and Artificial variables, if necessary, express the given LPP in standard form
after converting it in to Maximization/Minimization type.
Step 2:
Obtain an initial basic feasible solution with an initial basis 𝑩 = 𝑰𝒎 and form the auxiliary matrix 𝐵෠ and

𝑩 𝟎 𝑩−𝟏 𝟎
𝐵෠ −1 , such that 𝑩
෡= ෡
and 𝑩 =
−𝟏
𝑪𝑩 𝟏 𝑪𝑩 𝑩−𝟏 𝟏

where
I – Identity Matrix
m – No. of Slack or Surplus Variables
𝐶𝐵 – Cost of Basic Variable

5
 REVISED SIMPLEX METHOD

Step 3 :
State the objective function 𝒁 = 𝑪𝑿
Subject to the constraints 𝑨𝑿 = 𝒃
መ 𝑏෠ such that
and form 𝐴,

෡= 𝑨 ෡= 𝒃 .
𝑨 and 𝒃
−𝑪 𝟎
where
𝐶 – Cost of the Decision Variables
𝐴 – Coefficient Matrix of Constraint Equations
Step 4 :
Compute the net evaluations

𝑨
𝒁𝒋 − 𝑪𝒋 = 𝑪𝑩 𝑩−𝟏 𝟏
−𝑪
Case (i) : If all 𝒁𝒋 − 𝑪𝒋 ≥ 𝟎, the current basic feasible solution is optimal.

6
 REVISED SIMPLEX METHOD

Case (ii) : If at least one 𝒁𝒋 − 𝑪𝒋 < 𝟎, the current basic feasible solution is not optimal.
If 𝒁𝒋 − 𝑪𝒋 is the most negative, then the variable 𝒙𝒌 enters the basis, go to step (5).

Step 5
෠ 𝑐𝑢𝑟𝑟𝑒𝑛𝑡
Compute 𝑥ො k = 𝐵 −1 ෡ 𝒄𝒖𝒓𝒓𝒆𝒏𝒕 is the current basis matrix
𝑎ො k , where 𝑩
Case (i) : If all 𝑥𝑖𝑘 ≤ 0, then there exists an unbounded solution to the given LPP.

Case (ii) : If at least one 𝑥𝑖𝑘 > 0, consider the current 𝑋𝐵 and determine the leaving variable, go to
step (6).
where 𝒙𝒊𝒌 denotes the entries of 𝒙
ෝk

Step 6 :
Write down the results obtained from step (2) to step (5) in the revised simplex table.

7
 REVISED SIMPLEX METHOD

Step 7 :
Convert the leading element to unity and all other elements of the entering column to zero and update the
current basic feasible solution.

Step 8 :
Go to step (4) and repeat the procedure until an optimum basic feasible solution is obtained or there is an
indication of an unbounded solution.

8
 REVISED SIMPLEX METHOD

REVISED SIMPLEX METHOD PROCEDURE:

Maximum 𝒁 = 𝑪𝑿
Subject to 𝑨𝑿 = 𝒃, 𝑿 ≥ 𝟎

෡= 𝑨 ෡= 𝒃
• 𝑨 , 𝒃
−𝑪 𝟎
• 𝑩 = Basis

෡ −𝟏 𝑩−𝟏 𝟎
• 𝑩 𝒄𝒖𝒓𝒓𝒆𝒏𝒕 =
𝑪𝑩 𝑩−𝟏 𝟏

𝑨 ෡ −𝟏
• 𝒁𝒋 − 𝑪𝒋 = 𝑪𝑩 𝑩−𝟏 𝟏 , where 𝑪𝑩 𝑩−𝟏 𝟏 is the last row of 𝑩 𝒄𝒖𝒓𝒓𝒆𝒏𝒕
−𝑪
෡𝒌 = 𝑩
• 𝑿 ෡ −𝟏 ෝ𝒌
𝒄𝒖𝒓𝒓𝒆𝒏𝒕 𝒂

෡𝑩 = 𝑩
• 𝑿 ෡ −𝟏 ෡
𝒄𝒖𝒓𝒓𝒆𝒏𝒕 𝒃 (Solution)

9
 REVISED SIMPLEX METHOD

Example 1
Use Revised Simplex Method to solve the LPP.
𝑀𝑎𝑥𝑖𝑚𝑖𝑧𝑒 𝑍 = 𝑥1 + 𝑥2
𝑆𝑢𝑏𝑗𝑒𝑐𝑡 𝑡𝑜
3𝑥1 + 2𝑥2 ≤ 6 ,
𝑥1 +4𝑥2 ≤ 4,
𝑥1 , 𝑥2 ≥ 0
Solution:
By introducing the non-negative slack variables 𝑠1 𝑎𝑛𝑑 𝑠2 , the standard form of LPP becomes
𝑀𝑎𝑥𝑖𝑚𝑖𝑧𝑒 𝑍 = 𝑥1 + 𝑥2 + 0𝑠1 + 0𝑠2
S𝑢𝑏𝑗𝑒𝑐𝑡 𝑡𝑜
3𝑥1 + 2𝑥2 + 𝑠1 + 0𝑠2 = 6,
𝑥1 +4𝑥2 + 0𝑠1 + 𝑠2 = 4,
𝑥1 , 𝑥2 , 𝑠1 , 𝑠2 ≥ 0

10
 REVISED SIMPLEX METHOD

𝑥1
𝑥2
𝑀𝑎𝑥 𝑍 = 1 1 0 0 𝑠1 = CX
𝑠2
Subject to

𝑥1
3 2 1 0 𝑥2 6
𝑠1 =
1 4 0 1 4
𝑠2
A X b
𝑎𝑛𝑑 𝑥1 , 𝑥2 , 𝑠1 , 𝑠2 ≥ 0

where 𝐶 = 1 1 0 0 , 𝐶𝐵 = 0 0
6
3 2 1 0 6 𝑏
𝐴= , 𝑏= ∴ 𝑏෠ = = 4
1 4 0 1 4 0
0

11
 REVISED SIMPLEX METHOD

The initial basic feasible solution is 𝑠1 = 6, 𝑠2 = 4

1 0 1 0
With 𝐼𝑚 (𝑚 = 2) as the initial basis, 𝐼2 = 𝐵 = = 𝐵 −1 and 𝐶𝐵 𝐵−1 = 0 0 = 0 0
0 1 0 1

𝑎ො1 𝑎ො2 𝑎ො3 𝑎ො4

𝐴
Now 𝐴መ = = 3 2 1 0
−𝐶 1 4 0 1
−1 −1 0 0

1 0 0
𝐵 −1 0
𝐵෠𝑐𝑢𝑟𝑟𝑒𝑛𝑡
−1
= = 0 1 0
𝐶𝐵 𝐵−1 1
0 0 1
The net evaluation are given by

𝐴
𝑍𝑗 − 𝐶𝑗 = 𝐶𝐵 𝐵−1 1
−𝐶
3 2 1 0
= 0 0 1 1 4 0 1
−1 −1 0 0
= −1 −1 0 0
Since 𝑍𝑗 − 𝐶𝑗 < 0 the most negative, 𝑥ො1 enters the basis.
12
 REVISED SIMPLEX METHOD

1 0 0 3
Now 𝑥ො1 =𝐵෠𝑐𝑢𝑟𝑟𝑒𝑛𝑡
−1
𝑎ො1 = 0 1 0 1
0 0 1 −1
𝟑
= 𝟏
−𝟏
1 0 0 6
And 𝑋෠𝐵 = 𝐵෠𝑐𝑢𝑟𝑟𝑒𝑛𝑡 𝑏෠ = 0 1 0
−1
4
0 0 1 0
𝟔
= 𝟒
𝟎
Initial Simplex Table:

𝑌𝐵 𝐵෠𝑐𝑢𝑟𝑟𝑒𝑛𝑡 𝑥ො1 𝑋෠𝐵 Ratio Min

𝑋෠ 𝐵
𝑥ො1

𝑅1 𝑠1 1 0 0 (3) 6 6
=2 Now 𝒔𝟏 leaves the basis.
3
𝑅2 𝑠2 0 1 0 1 4 4
=4
1
𝑅3 0 0 1 -1 0
13
 REVISED SIMPLEX METHOD

𝑅1 𝑁𝑒𝑤 = 𝑅1 𝑂𝑙𝑑 ÷ 3
𝑅1 𝑁𝑒𝑤 = 1 0 0 3 6 ÷3
𝟏
=𝟑 𝟎 𝟎 1 2

𝑅2 𝑁𝑒𝑤 = 𝑅2 𝑂𝑙𝑑 − 𝑅1 (𝑁𝑒𝑤)

𝑅2 𝑁𝑒𝑤 = 0 1 0 1 4
−1/3 0 0 − 1 − 2
−𝟏/𝟑 𝟏 𝟎 0 2

𝑅3 𝑁𝑒𝑤 = 𝑅3 𝑂𝑙𝑑 + 𝑅1 (𝑁𝑒𝑤)

𝑅3 𝑁𝑒𝑤 = 0 0 1 − 1 0
1/3 0 0 1 2
𝟏/𝟑 𝟎 𝟏 0 2

14
 REVISED SIMPLEX METHOD

First Iteration:
Introduce 𝑥ො1 and drop 𝑠1 .
1
0 0
3
−1
𝐵෠𝑛𝑒𝑥𝑡
−1
= 3
1 0
𝟏
𝟎 1
𝟑

𝐴
𝑍𝑗 − 𝐶𝑗 = 𝐶𝐵 𝐵 −1 1
−𝐶
3 2 1 0
𝟏
= 𝟎 1 1 4 0 1
𝟑
−1 −1 0 0
−1 1
= 0 3 3
0
−1
Since 𝑍𝑗 − 𝐶𝑗 = ෝ𝟐 enters basis
is most negative, 𝒙
3

15
 REVISED SIMPLEX METHOD

1 𝟐
0 0
3 𝟑
−1 2 𝟏𝟎
∴ 𝑥ො2 = 𝐵෠𝑛𝑒𝑥𝑡
−1
𝑎ො2 = 1 0 4 =
3 −1 𝟑
1 −𝟏
0 1
3 𝟑
1
0 0
3 6 𝟐
−1
and ∴ 𝑋෠𝐵 = 𝐵෠𝑛𝑒𝑥𝑡
−1
𝑏෠ = 3
1 0 4 = 𝟐
1 0 𝟐
3
0 1

𝑌𝐵 𝐵෠𝑛𝑒𝑥𝑡 𝑥ො2 𝑋෠𝐵 Ratio

𝑋෠ 𝐵
Min 𝑥ො2

𝑅1 𝑥1 1/3 0 0 2/3 2 6
=3
2 Now 𝒔𝟐 leaves the basis.
𝑅2 𝑠2 -1/3 1 0 (10/3) 2 6
= 0.6
10
𝑅3 1/3 0 1 -1/3 2
16
 REVISED SIMPLEX METHOD

𝑅2 𝑁𝑒𝑤 = 𝑅2 𝑂𝑙𝑑 × 3/10

𝑅2 𝑁𝑒𝑤 = −1/3 1 0 10/3 2 × 3/10
= −𝟏/𝟏𝟎 𝟑/𝟏𝟎 𝟎 1 3/5

𝑅1 𝑁𝑒𝑤 = 𝑅1 𝑂𝑙𝑑 − (2/3)𝑅2 (𝑁𝑒𝑤)

𝑅1 𝑁𝑒𝑤 = 1/3 0 0 2/3 2
2/30 −6/30 0 − 2/3 − 2/5
2/5 −𝟏/𝟓 𝟎 0 8/5

𝑅3 𝑁𝑒𝑤 = 𝑅3 𝑂𝑙𝑑 + (1/3)𝑅2 (𝑁𝑒𝑤)

𝑅3 𝑁𝑒𝑤 = 1/3 0 1 − 1/3 2
−1/30 3/30 0 1/3 1/5
𝟑/𝟏𝟎 𝟏/𝟏𝟎 𝟏 0 11/5

17
 REVISED SIMPLEX METHOD

Second Iteration:
Introduce 𝑥2 and drop 𝑠2 .
2 −1
0
5 5
−1 3
𝐵෠𝑐𝑢𝑟𝑟
−1
= 10 10
0
𝟑 𝟏
1
𝟏𝟎 𝟏𝟎
𝐴
𝑍𝑗 − 𝐶𝑗 = 𝐶𝐵 𝐵 −1 1
−𝐶
𝟑 𝟏 3 2 1 0
= 1 1 4 0 1
𝟏𝟎 𝟏𝟎 −1 −1 0 0
3 1
= 0 0
10 10
Since all 𝒁𝒋 − 𝑪𝒋 ≥ 𝟎 , the current basic feasible solution is optimal.

18
 REVISED SIMPLEX METHOD

2 −1 8
0
5 5 5
−1 3 6 3
𝑋෠𝐵 = 𝐵෠𝑐𝑢𝑟𝑟
−1 𝑏෠ = 0 4 =
10 10 0 5
3 1 11
1
10 10 5

8 3
⟹ 𝑥1 = , 𝑥2 =
5 5
Max 𝑍 = 𝑥1 + 𝑥2
8 3
=5+5
11
= 5

Hence the optimal solution is,

11 8 3
Max 𝑍 = , 𝑥1 = , 𝑥2 =
5 5 5

19
 REVISED SIMPLEX METHOD

Example 2
Use revised simplex method to solve the LPP.
𝑀𝑖𝑛𝑖𝑚𝑖𝑧𝑒 𝑍 = 𝑥1 + 2𝑥2
𝑆𝑢𝑏𝑗𝑒𝑐𝑡 𝑡𝑜
2𝑥1 + 5𝑥2 ≥ 6 ,
𝑥1 + 𝑥2 ≥ 2,
𝑥1 , 𝑥2 ≥ 0
Solution:
By introducing the non-negative surplus variables 𝑠1 , 𝑠2 and Artificial Variables 𝐴1 , 𝐴2 the standard form of LPP
becomes
𝑀𝑖𝑛𝑖𝑚𝑖𝑧𝑒 𝑍 = 𝑥1 + 2𝑥2 + 0𝑠1 + 0𝑠2 + 𝑀𝐴1 + 𝑀𝐴2

S𝑢𝑏𝑗𝑒𝑐𝑡 𝑡𝑜
2𝑥1 + 5𝑥2 − 𝑠1 + 0𝑠2 + 𝐴1 + 0𝐴2 = 6,
𝑥1 +𝑥2 + 0𝑠1 − 𝑠2 + 0𝐴1 + 𝐴2 = 4,
𝑥1 , 𝑥2 , 𝑠1 , 𝑠2 , 𝐴1 , 𝐴2 ≥ 0

20
 REVISED SIMPLEX METHOD

REVISED SIMPLEX METHOD:

Maximize 𝑍 ∗ = −Minimize 𝑍 = −(𝑥1 + 2𝑥2 + 0𝑠1 + 0𝑠2 + 𝑀𝐴1 + 𝑀𝐴2 )

𝑴𝒂𝒙𝒊𝒎𝒊𝒛𝒆 𝒁∗ = −𝒙𝟏 − 𝟐𝒙𝟐 + 𝟎𝒔𝟏 + 𝟎𝒔𝟐 − 𝑴𝑨𝟏 − 𝑴𝑨𝟐

𝑥1
𝑥2
𝑠1
𝑀𝑎𝑥 𝑍 ∗ = −1 −2 0 0 −𝑀 −𝑀 𝑠2 = CX
𝐴1
𝐴2
Subject to
𝑥1
𝑥2
2 5 −1 0 1 0 𝑠1 6
𝑠2 =
1 1 0 −1 0 1 2
𝐴1
𝐴2
A X b
𝑥1 , 𝑥2 , 𝑠1 , 𝑠2 , 𝐴1 , 𝐴2 ≥ 0 21
 REVISED SIMPLEX METHOD

where
𝐶 = −1 −2 0 0 −𝑀 −𝑀 ,

𝐶𝐵 = −𝑀 −𝑀 ,

2 5 −1 0 1 0 ,
𝐴=
1 1 0 −1 0 1

6
𝑏= ,
2

6
𝑏
𝑏෠ = = 2
0
0

22
 REVISED SIMPLEX METHOD

The initial basic feasible solution is 𝐴1 = 6, 𝐴2 = 2

1 0 1 0
With 𝐼𝑚 (𝑚 = 2) as the initial basis, 𝐼2 = 𝐵 = = 𝐵 −1 and 𝐶𝐵 𝐵−1 = −𝑀 − 𝑀 = −𝑀 − 𝑀
0 1 0 1

𝑎ො1 𝑎ො2 𝑎ො3 𝑎ො4 𝑎ො5 𝑎ො6

𝐴
Now 𝐴መ = = 2 5 −1 0 1 0
−𝐶 1 1 0 −1 0 1
1 2 0 0 𝑀 𝑀

1 0 0
𝐵 −1 0
𝐵෠𝑐𝑢𝑟𝑟𝑒𝑛𝑡
−1
= = 0 1 0
𝐶𝐵 𝐵−1 1
−𝑀 −𝑀 1
The net evaluation are given by

𝐴
𝑍𝑗∗ − 𝐶𝑗 = 𝐶𝐵 𝐵−1 1
−𝐶
2 5 −1 0 1 0
= −𝑀 −𝑀 1 1 1 0 −1 0 1
1 2 0 0 𝑀 𝑀
= −3𝑀 + 1 −𝟔𝑴 + 𝟐 𝑀 𝑀 0 0
Since 𝒁∗𝒋 − 𝑪𝒋 < 𝟎 the most negative, 𝒙
ෝ𝟐 enters the basis.
23
 REVISED SIMPLEX METHOD

1 0 0 5
Now 𝑥ො2 = 𝐵෠𝑐𝑢𝑟𝑟𝑒𝑛𝑡
−1
𝑎ො2 = 0 1 0 1
−𝑀 −𝑀 1 2
𝟓
= 𝟏
−𝟔𝑴 + 𝟐
1 0 0 6
And 𝑋෠𝐵 = 𝐵෠𝑐𝑢𝑟𝑟𝑒𝑛𝑡
−1
𝑏෠ = 0 1 0 2
−𝑀 −𝑀 1 0
𝟔
= 𝟐
−𝟖𝑴
Initial Simplex Table:

𝑌𝐵 𝐵෠𝑐𝑢𝑟𝑟𝑒𝑛𝑡 𝑥ො2 𝑋෠𝐵 Ratio Min

𝑋෠ 𝐵
𝑥ො1

𝑅1 𝐴1 1 0 0 (5) 6 6
= 1.2 Now 𝑨𝟏 leaves the basis.
5
𝑅2 𝐴2 0 1 0 1 2 2
=2
1
𝑅3 -M -M 1 -6M+2 -8M
24
 REVISED SIMPLEX METHOD

𝑅1 𝑁𝑒𝑤 = 𝑅1 𝑂𝑙𝑑 ÷ 5
𝑅1 𝑁𝑒𝑤 = 1 0 0 5 6 ÷5
𝟏 6
=𝟓 𝟎 𝟎 1 5

𝑅2 𝑁𝑒𝑤 = 𝑅2 𝑂𝑙𝑑 − 𝑅1 (𝑁𝑒𝑤)

𝑅2 𝑁𝑒𝑤 = 0 1 0 1 2
−1/5 0 0 − 1 − 6/5
−𝟏/𝟓 𝟏 𝟎 0 4/5

𝑅3 𝑁𝑒𝑤 = 𝑅3 𝑂𝑙𝑑 − (−𝟔𝑴 + 𝟐)𝑅1 (𝑁𝑒𝑤)

𝑅3 𝑁𝑒𝑤 = −𝑀 −𝑀 1 − 6𝑀 + 2 − 8𝑀
( 6M − 2)/5 0 0 6𝑀 − 2 (36𝑀 − 12)/5
(M-2)/5 −𝑴 𝟏 0 (−4𝑀 − 12)/5

25
 REVISED SIMPLEX METHOD

First Iteration:
Introduce 𝑥ො2 and drop 𝐴1 .
1
0 0
5
−1
𝐵෠𝑛𝑒𝑥𝑡
−1
= 5
1 0
𝑴−𝟐
−𝑴 1
𝟓

𝐴
𝑍𝑗∗ − 𝐶𝑗 = 𝐶𝐵 𝐵−1 1
−𝐶
2 −1 0 1
𝑴−𝟐
= −𝑴 1 1 0 −1 0
𝟓
1 0 0 𝑀
−3𝑀+1 −𝑀+2 6𝑀−2
= 𝑀
5 5 5

Since 𝑍𝑗∗ − 𝐶𝑗 is most negative, 𝒙

ෝ𝟏 enters basis

26
 REVISED SIMPLEX METHOD
1 𝟐
0 0
5 𝟓
−1 2 𝟑
∴ 𝑥ො1 = 𝐵෠𝑛𝑒𝑥𝑡
−1
𝑎ො1 = 1 0 1 =
5 1 𝟓
𝑀−2 −𝟑𝑴 + 𝟏
−𝑀 1
5 𝟓
1 𝟔
0 0
5 𝟓
−1 6 𝟒
∴ 𝑋෠𝐵 = 𝐵෠𝑛𝑒𝑥𝑡
−1
𝑏෠ = 1 0 2 =
5 0 𝟓
𝑀−2 −𝟒𝑴 − 𝟏𝟐
−𝑀 1
5 𝟓
𝑌𝐵 𝐵෠𝑛𝑒𝑥𝑡 𝑥ො1 𝑋෠𝐵 Ratio
𝑋෠ 𝐵
Min 𝑥ො2

𝑅1 𝑥2 1/5 0 0 2/5 6/5 6

=3
2 Now 𝑨𝟐 leaves the basis.
𝑅2 𝐴2 -1/5 1 0 (3/5) 4/5 4
= 1.33
3
𝑅3 𝑀−2 −𝟑𝑴 + 𝟏 −𝟒𝑴 − 𝟏𝟐
-M 1
5
𝟓 𝟓 27
 REVISED SIMPLEX METHOD

𝑅2 𝑁𝑒𝑤 = 𝑅2 𝑂𝑙𝑑 × 5/3

𝑅2 𝑁𝑒𝑤 = −1/5 1 0 3/5 4/5 × 5/3
= −𝟏/𝟑 𝟓/𝟑 𝟎 1 4/3

𝑅1 𝑁𝑒𝑤 = 𝑅1 𝑂𝑙𝑑 − (2/5)𝑅2 (𝑁𝑒𝑤)

𝑅1 𝑁𝑒𝑤 = 1/5 0 0 2/5 6/5
2/15 −10/15 0 − 2/5 − 8/15
3/15 −𝟏𝟎/𝟏𝟓 𝟎 0 10/15

𝑅3 𝑁𝑒𝑤 = 𝑅3 𝑂𝑙𝑑 − (−𝟑𝑴 + 𝟏/𝟓)𝑅2 (𝑁𝑒𝑤)

𝑀−2
𝑅3 𝑁𝑒𝑤 = −𝑀 1 (−3𝑀 + 1)/5 (−4𝑀 − 12)/5
5
(−3M + 1)/15 (3𝑀 − 1)/3 0 (3𝑀 − 1)/5 (12𝑀 − 4)/15
𝟏 𝟏
−𝟑 −𝟑 𝟏 0 − 40/15
28
 REVISED SIMPLEX METHOD

Second Iteration:
Introduce 𝑥1 and drop 𝐴2 .
1 −2
0
3 3
−1 5
𝐵෠𝑐𝑢𝑟𝑟
−1
= 3 3
0
−𝟏 −𝟏
1
𝟑 𝟑

𝐴
𝑍𝑗∗ − 𝐶𝑗 = 𝐶𝐵 𝐵−1 1
−𝐶

−𝟏 −𝟏 −1 0 1 0
= 1 0 −1 0 1
𝟑 𝟑 0 0 𝑀 𝑀

3 1
= 0 0
10 10

Since all 𝒁∗𝒋 − 𝑪𝒋 ≥ 𝟎 , the current basic feasible solution is optimal.

29
 REVISED SIMPLEX METHOD

1 −2 2
0
3 3 3
−1 5 6 4
∴ 𝑋෠𝐵 = 𝐵෠𝑐𝑢𝑟𝑟
−1
𝑏෠ = 0 2 =
3 3 0 3
−1 −1 −8
1
3 3 3
4 2
⟹ 𝑥1 = , 𝑥2 =
3 3
−4 2 2 −8
𝑀𝑎𝑥𝑖𝑚𝑖𝑧𝑒 𝑍 ∗ = −𝑥1 − 2𝑥2 = − =
3 3 3
Maximize 𝑍 ∗ = −Minimize 𝑍
−8/3 = −Minimize 𝑍

The optimal solution is

8 4 2
𝑀𝑖𝑛 𝑍 = 3 , 𝑥1 = 3 , 𝑥2 = 3

30
 DUALITY IN LP

What is Duality in LP?

• Every Linear Programming Problem always associated with another Linear Programming Problem which is
based upon same data and having the same solution.

• This property of Linear Programming Problem is termed as Duality in Linear Programming Problem.

• The original problem is called as Primal Problem and associated problem is called as Dual problem.

• Any of these Linear Programming Problem can be taken as primal and other as dual therefore these problems
are simultaneously called as Primal-Dual Pair

31
 DUALITY IN LP

General Steps to Formulate the Dual Problem

Identify the Primal Problem:

• If the primal is a maximization problem with constraints in the form of “ ≤ ”, then the dual will be a
minimization problem with “ ≥ ” constraints.

• If the primal is a minimization problem with constraints in the form of “ ≥ ”, then the dual will be a
maximization problem with “ ≤ ” constraints.

Define the Dual Variables:

• Each constraint in the primal corresponds to a variable in the dual.

• Each variable in the primal corresponds to a constraint in the dual.

32
 DUALITY IN LP

Construct the Dual Objective Function:

• If the primal is maximization the dual objective function will be a minimization problem, and vice versa.

• The coefficients of the objective function in the primal become the right-hand side constants in the dual.

Construct the Dual Constraints:

• The right-hand side values of the primal constraints become the coefficients of the dual’s objective function.

• The coefficients of variables in the primal constraints become the coefficients of constraints in the dual.

Sign Restrictions:

• If the primal has a ≤ constraint, then the corresponding dual variable is ≥ 0.

• If the primal has a ≥ constraint, then the corresponding dual variable is ≤ 𝟎.

• If the primal variable is unrestricted, the corresponding dual constraint is an equality.

33
 DUALITY IN LP

Let the Primal Problem be The corresponding Dual Problem is defined as

Max 𝒁 = 𝒄𝟏 𝒙𝟏 + 𝒄𝟐 𝒙𝟐 + ⋯ + 𝒄𝒏 𝒙𝒏 Min 𝒁 = 𝒃𝟏 𝒘𝟏 + 𝒃𝟐 𝒘𝟐 + ⋯ + 𝒃𝒎 𝒘𝒎
Subject to Subject to
𝒂𝟏𝟏 𝒙𝟏 + 𝒂𝟏𝟐 𝒙𝟐 + ⋯ + 𝒂𝟏𝒏 𝒙𝒏 ≤ 𝒃𝟏 𝒂𝟏𝟏 𝒘𝟏 + 𝒂𝟐𝟏 𝒘𝟐 + ⋯ + 𝒂𝒎𝟏 𝒘𝒎 ≥ 𝒄𝟏
𝒂𝟐𝟏 𝒙𝟏 + 𝒂𝟐𝟐 𝒙𝟐 + ⋯ + 𝒂𝟐𝒏 𝒙𝒏 ≤ 𝒃𝟐 𝒂𝟏𝟐 𝒘𝟏 + 𝒂𝟐𝟐 𝒘𝟐 + ⋯ + 𝒂𝒎𝟐 𝒘𝒎 ≥ 𝒄𝟐
⋮ ⋮
𝒂𝒎𝟏 𝒙𝟏 + 𝒂𝒎𝟐 𝒙𝟐 + ⋯ + 𝒂𝒎𝒏 𝒙𝒏 ≤ 𝒃𝒏 𝒂𝟏𝒏 𝒘𝟏 + 𝒂𝟐𝒏 𝒘𝟐 + ⋯ + 𝒂𝒎𝒏 𝒘𝒎 ≥ 𝒄𝒏
And 𝒙𝟏 , 𝒙𝟐 , … , 𝒙𝒏 ≥ 𝟎 And 𝒘𝟏 , 𝒘𝟐 , … , 𝒘𝒏 ≥ 𝟎

34
 DUALITY IN LP

Important Characteristics of Duality

• Dual of Dual is Primal.

• If any of the two problems has an infeasible solution, then the value of the objective function of the other is
unbounded.

• The value of the objective function for any feasible solution of the primal is less than the value of the objective
function for any feasible solution of the dual.

• If either the primal or dual has an unbounded solution, then the solution to the other problem is infeasible.

• If the primal has a feasible solution, but the dual does not have then the primal will not have a finite optimum
solution and vice versa.
35
 DUALITY IN LP

Example 1:
Write the dual of the following problem
𝑀𝑖𝑛 𝑍 = 20𝑥1 + 40𝑥2
𝑆𝑢𝑏𝑗𝑒𝑐𝑡 𝑡𝑜
36𝑥1 + 6𝑥2 ≥ 108
20𝑥1 + 10𝑥2 ≥ 36
And 𝑥1 , 𝑥2 ≥ 0

Solution:
Dual of the give problem is:
𝑀𝑎𝑥 𝑍 = 108𝑤1 + 36𝑤2
𝑆𝑢𝑏𝑗𝑒𝑐𝑡 𝑡𝑜
36𝑤1 + 20𝑤2 ≤ 20
6𝑤1 + 10𝑤2 ≤ 40
And 𝑤1 , 𝑤2 ≥ 0

36
 DUAL SIMPLEX METHOD

Procedure for Dual Simplex Method:

Step 1:
Convert the problem to maximization form if it is initially in the minimization form.

Step 2:
Convert (≥) type constraints, if any to (≤) type by multiplying both sides by -1.

Step 3:
• Convert the inequality constraints into equalities by introducing slack variables.

• Obtain the initial basic solution and express this information in the simplex table.

37
 DUAL SIMPLEX METHOD

Step 4:
Test the nature of (𝒁𝒋 − 𝑪𝒋 ) and 𝑿𝑩𝒊 .
Case (i) : If all (𝒁𝒋 − 𝑪𝒋 ) ≥ 𝟎 and all 𝑿𝑩𝒊 ≥ 𝟎, then the current solution is an optimum feasible solution.

Case (ii) : If all (𝒁𝒋 − 𝑪𝒋 ) ≥ 𝟎 and at least one 𝑿𝑩𝒊 < 𝟎, then the current solution is not an optimum basic
feasible solution and go to the step 5.

Case (iii) : If any (𝒁𝒋 − 𝑪𝒋 ) < 𝟎, then this method fails.

Step 5:
I. Leaving Variable:
The leaving variable is the basic variable corresponding to the most negative value of 𝑿𝑩𝒊 . Let 𝑥𝑘 be the
leaving variable.

38
 DUAL SIMPLEX METHOD

II. Entering Variable:

(𝒁𝒋 −𝑪𝒋 )
Compute 𝜽 = 𝒎𝒂𝒙 , 𝒂𝒊𝒌 < 𝟎 .
𝒂𝒊𝒌

• The entering variable is the non-basic variable corresponding to the maximum ratio 𝜃.

• If there is no such ratio with -ve denominator, then the problem does not have a feasible
solution

Step 6:
• Carry out the row operations as in the regular simplex method.

• Repeat the procedure until either an optimum feasible solution is obtained or there is an indication of non-
existence of a feasible solution.

39
 DUAL SIMPLEX METHOD

Example 1:
Use Dual Simplex Method to solve the following LPP
Max 𝒁 = −𝟐𝒙𝟏 − 𝒙𝟑
Subject to
𝒙𝟏 + 𝒙𝟐 − 𝒙𝟑 ≥ 𝟓
𝒙𝟏 − 𝟐𝒙𝟐 + 𝟒𝒙𝟑 ≥ 𝟖
and 𝒙𝟏 , 𝒙𝟐 , 𝒙𝟑 ≥ 𝟎
Solution:
The given problem can be written as:
Max 𝑍 = −2𝑥1 − 𝑥3
Subject to
−𝒙𝟏 − 𝒙𝟐 + 𝒙𝟑 ≤ −𝟓
−𝒙𝟏 + 𝟐𝒙𝟐 − 𝟒𝒙𝟑 ≤ −𝟖
and 𝒙𝟏 , 𝒙𝟐 , 𝒙𝟑 ≥ 𝟎
40
 DUAL SIMPLEX METHOD

Adding the Slack Variables 𝒔𝟏 and 𝒔𝟐 , we get

Max 𝑍 = −2𝑥1 − 𝑥3 + 0𝑠1 + 0𝑠2
Subject to
−𝑥1 − 𝑥2 + 𝑥3 + 𝑠1 = −5
−𝑥1 + 2𝑥2 − 4𝑥3 + 𝑠2 = −8
and 𝑥1 , 𝑥2 , 𝑥3 , 𝑠1 , 𝑠2 ≥ 0

The non-basic variables are 𝑥1 , 𝑥2 and 𝑥3 .

Hence the non-basic solutions are 𝑥1 = 0, 𝑥2 = 0 and 𝑥3 = 0.

The basic variables are 𝑠1 and 𝑠2

The Initial Basic Feasible Solution (IBFS) is given by,
𝒔𝟏 = −𝟓, 𝒔𝟐 = −𝟖

41
 DUAL SIMPLEX METHOD

Initial Iteration:

𝑪𝒋 −𝟐 𝟎 −𝟏 𝟎 𝟎
𝑪𝑩 𝑩 𝑿𝑩 𝒙𝟏 𝒙𝟐 𝒙𝟑 𝒔𝟏 𝒔𝟐
𝑹𝟏 0 𝑠1 −5 −1 −1 1 1 0

𝑹𝟐 0 𝑠2 −8 −1 2 [−𝟒] 0 1

𝒁𝒋 0 0 0 0 0 0
𝒁𝒋 − 𝑪𝒋 2 0 1 0 0

Since all 𝒁𝒋 − 𝑪𝒋 ≥ 𝟎 and all 𝑿𝑩𝒊 < 𝟎. The solution is not To find the entering variable:
optimal. (𝒁𝒋 −𝑪𝒋 ) 2 1
𝑴𝒂𝒙 , 𝒂𝒊𝒌 < 𝟎 = 𝑀𝑎𝑥 , − , ,−,−
𝒂𝒊𝒌 −1 −4

= 𝑀𝑎𝑥 −2, −, −0.25, − , −

𝑿𝑩𝟐 = −𝟖 is the most negative, the basic variable 𝒔𝟐
= −0.25
leaves the basis. ∴ 𝒙𝟑 enters the basis. 42
 DUAL SIMPLEX METHOD

𝑅2 𝑁𝑒𝑤 = 𝑅2 𝑂𝑙𝑑 ÷ −4

𝑅2 𝑁𝑒𝑤 = −8 −1 2 −4 0 1 ÷ −4
1 1 1
= 2 −2 1 0 −4
4

𝑅1 𝑁𝑒𝑤 = 𝑅1 𝑂𝑙𝑑 − 𝑅2 (𝑁𝑒𝑤)

𝑅1 𝑁𝑒𝑤 = −5 −1 −1 1 1 0
1 1 1
= −2 −4 −1 0
2 4

5 1 1
−7 −4 −2 0 1 4

43
 DUAL SIMPLEX METHOD

First Iteration:
Drop 𝒔𝟐 and Introduce 𝒙𝟑

𝑪𝒋 −𝟐 𝟎 −𝟏 𝟎 𝟎
𝑪𝑩 𝑩 𝑿𝑩 𝒙𝟏 𝒙𝟐 𝒙𝟑 𝒔𝟏 𝒔𝟐
𝑹𝟏 0 𝑠1 −7 −5/4 [−𝟏/𝟐] 0 1 1/4

𝑹𝟐 −1 𝑥3 2 1/4 −1/2 1 0 −1/4

𝒁𝒋 −2 −1/4 1/2 −1 0 1/4

𝒁𝒋 − 𝑪𝒋 7/4 1/2 0 0 1/4

To find the entering variable:

Since all 𝒁𝒋 − 𝑪𝒋 ≥ 𝟎 and one 𝑿𝑩𝒊 < 𝟎. The (𝒁𝒋 −𝑪𝒋 ) 7/4 1/2
𝑴𝒂𝒙 , 𝒂𝒊𝒌 < 𝟎 = 𝑀𝑎𝑥 , , −, − , −
𝒂𝒊𝒌 −5/4 −1/2
solution is not optimal.
−7
𝑿𝑩𝟏 = −𝟕 is the most negative, the basic = 𝑀𝑎𝑥 , −1, − ,− ,−
5

variable 𝒔𝟏 leaves the basis. = 𝑀𝑎𝑥 −1.4, −1, − , − , −

∴ 𝒙𝟐 enters the basis. = −1
44
 DUAL SIMPLEX METHOD

𝑅1 𝑁𝑒𝑤 = 𝑅1 𝑂𝑙𝑑 × −2

𝑅1 𝑁𝑒𝑤 = −7 − 5/4 − 1/2 0 1 1/4 × −2

= 14 5/2 1 0 −2 − 1/2

𝑅2 (𝑁𝑒𝑤) = 𝑅2 (𝑂𝑙𝑑) + (1/2)𝑅2 (𝑁𝑒𝑤)

𝑅2 𝑁𝑒𝑤 = 2 1/4 − 1/2 1 0 − 1/4

=7 5/4 1/2 0 −1 − 1/4

9 3/2 0 1 −1 − 1/2

45
 DUAL SIMPLEX METHOD

Second Iteration:
Drop 𝒔𝟏 and Introduce 𝒙𝟐
𝑪𝒋 −𝟐 𝟎 −𝟏 𝟎 𝟎
𝑪𝑩 𝑩 𝑿𝑩 𝒙𝟏 𝒙𝟐 𝒙𝟑 𝒔𝟏 𝒔𝟐
𝑹𝟏 0 𝑥2 14 5/2 1 0 −2 −1/2

𝑹𝟐 −1 𝑥3 9 3/2 0 1 −1 −1/2

𝒁𝒋 −9 −3/2 0 −1 1 1/2
𝒁𝒋 − 𝑪𝒋 1/2 0 0 1 1/2

Since all 𝒁𝒋 − 𝑪𝒋 ≥ 𝟎 and all 𝑿𝑩𝒊 ≥ 𝟎. The current feasible solution is optimum.
The optimal solution at 𝒙𝟏 = 𝟎, 𝒙𝟐 = 𝟏𝟒 𝒂𝒏𝒅 𝒙𝟑 = 𝟗
Max 𝒁 = −𝟐𝒙𝟏 − 𝒙𝟑
= −𝟐 𝟎 − 𝟗
= −𝟗
46
 DUAL SIMPLEX METHOD

Example 2:
Use Dual Simplex Method to solve the following LPP
Min 𝒁 = 𝟓𝒙𝟏 + 𝟔𝒙𝟐
Subject to
𝒙𝟏 + 𝒙𝟐 ≥ 𝟐
𝟒𝒙𝟏 + 𝒙𝟐 ≥ 𝟒
and 𝒙𝟏 , 𝒙𝟐 ≥ 𝟎
Solution:
The given problem can be written as:
Max 𝒁∗ = −𝟓𝒙𝟏 − 𝟔𝒙𝟐
Subject to
−𝒙𝟏 − 𝒙𝟐 ≤ −𝟐
−𝟒𝒙𝟏 − 𝒙𝟐 ≤ −𝟒
and 𝒙𝟏 , 𝒙𝟐 ≥ 𝟎
47
 DUAL SIMPLEX METHOD

Adding the Slack Variables 𝒔𝟏 and 𝒔𝟐 , we get

Max 𝑍 ∗ = −5𝑥1 − 6𝑥2 + 0𝑠1 + 0𝑠2
Subject to
−𝑥1 − 𝑥2 + 𝑠1 = −2
−4𝑥1 − 𝑥2 + 𝑠2 = −4
and 𝑥1 , 𝑥2 , 𝑠1 , 𝑠2 ≥ 0

The non-basic variables are 𝑥1 and 𝑥2 .

Hence the non-basic solutions are 𝑥1 = 0 and 𝑥2 = 0.

The basic variables are 𝑠1 and 𝑠2

The Initial Basic Feasible Solution (IBFS) is given by,
𝒔𝟏 = −𝟐, 𝒔𝟐 = −𝟒

48
 DUAL SIMPLEX METHOD

Initial Iteration:

𝑪𝒋 −𝟓 −𝟔 𝟎 𝟎
𝑪𝑩 𝑩 𝑿𝑩 𝒙𝟏 𝒙𝟐 𝒔𝟏 𝒔𝟐
𝑹𝟏 0 𝑠1 −2 −1 −1 1 0

𝑹𝟐 0 𝑠2 −4 [−𝟒] −1 0 1

𝒁∗𝒋 0 0 0 0 0
𝒁∗𝒋 − 𝑪𝒋 5 6 0 0

To find the entering variable:

Since all 𝒁∗𝒋 − 𝑪𝒋 ≥ 𝟎 and all 𝑿𝑩𝒊 < 𝟎 . The
(𝒁∗𝒋 −𝑪𝒋 ) 5 6
𝑴𝒂𝒙 , 𝒂𝒊𝒌 < 𝟎 = 𝑀𝑎𝑥 , ,− ,−
solution is not optimal. 𝒂𝒊𝒌 −4 −1

= 𝑀𝑎𝑥 −1.25, −6, − , −

𝑿𝑩𝟐 = −𝟒 is the most negative, the basic = −1.25
variable 𝒔𝟐 leaves the basis. ∴ 𝒙𝟏 enters the basis.
49
 DUAL SIMPLEX METHOD

𝑅2 𝑁𝑒𝑤 = 𝑅2 𝑂𝑙𝑑 ÷ −4

𝑅2 𝑁𝑒𝑤 = (−4 −4 −1 0 1) ÷ −4
= 1 1 1/4 0 − 1/4

𝑅1 𝑁𝑒𝑤 = 𝑅1 𝑂𝑙𝑑 + 𝑅2 (𝑁𝑒𝑤)

𝑅1 𝑁𝑒𝑤 = −2 −1 −1 1 0
1 1 1/4 0 − 1/4
−1 0 − 3/4 1 − 1/4

50
 DUAL SIMPLEX METHOD

First Iteration:
Drop 𝒔𝟐 and Introduce 𝒙𝟏
𝑪𝒋 −𝟓 −𝟔 𝟎 𝟎
𝑪𝑩 𝑩 𝑿𝑩 𝒙𝟏 𝒙𝟐 𝒔𝟏 𝒔𝟐
𝑹𝟏 0 𝑠1 −1 0 −3/4 1 [−𝟏/𝟒]

𝑹𝟐 −5 𝑥1 1 1 1/4 0 −1/4

𝒁∗𝒋 −5 −5 −5/4 0 5/4

𝒁∗𝒋 − 𝑪𝒋 0 19/4 0 5/4

Since all 𝒁∗𝒋 − 𝑪𝒋 ≥ 𝟎 and one 𝑿𝑩𝒊 < 𝟎. The solution is To find the entering variable:

not optimal. 𝑴𝒂𝒙

(𝒁∗𝒋 −𝑪𝒋 )
, 𝒂𝒊𝒌 < 𝟎 = 𝑀𝑎𝑥 −,
19/4
, −,
5/4
𝒂𝒊𝒌 −3/4 −1/4

−19
= 𝑀𝑎𝑥 −, , − , −5
𝑿𝑩𝟏 = −𝟏 is the most negative, the basic variable 𝒔𝟏 3

= 𝑀𝑎𝑥 −, −6.33, −, −5
leaves the basis. = −1
∴ 𝒔𝟐 enters the basis.
51
 DUAL SIMPLEX METHOD

𝑅1 𝑁𝑒𝑤 = 𝑅1 𝑂𝑙𝑑 × −4

𝑅1 𝑁𝑒𝑤 = −1 0 − 3/4 1 − 1/4 × −4

= 4 0 3 −4 1

𝑅2 (𝑁𝑒𝑤) = 𝑅2 (𝑂𝑙𝑑) + (1/4)𝑅2 (𝑁𝑒𝑤)

𝑅2 𝑁𝑒𝑤 = 1 1 1/4 0 − 1/4

=1 0 3/4 −1 1/4

2 1 1 −1 0

52
 DUAL SIMPLEX METHOD

Second Iteration:
Drop 𝒔𝟏 and Introduce 𝒔𝟐

𝑪𝒋 −𝟓 −𝟔 𝟎 𝟎
𝑪𝑩 𝑩 𝑿𝑩 𝒙𝟏 𝒙𝟐 𝒔𝟏 𝒔𝟐
𝑹𝟏 0 𝑠2 4 0 3 −4 1

𝑹𝟐 −5 𝑥1 2 1 1 −1 0

𝒁∗𝒋 −10 −5 −5 5 0
𝒁∗𝒋 − 𝑪𝒋 0 1 5 0

Since all 𝒁∗𝒋 − 𝑪𝒋 ≥ 𝟎 and all 𝑿𝑩𝒊 ≥ 𝟎. The current feasible solution is optimum.
The optimal solution at 𝒙𝟏 = 𝟐, 𝒙𝟐 = 𝟎
Max 𝒁∗ = −𝟓𝒙𝟏 − 𝟔𝒙𝟐 Min 𝒁 = − Max 𝒁∗
= −𝟓 𝟐 − 𝟔(𝟎) Min 𝒁 = −(−𝟏𝟎)
= −𝟏𝟎 Min 𝒁 = 𝟏𝟎
53
 SENSITIVITY ANALYSIS

What is Sensitivity Analysis?

• Sensitivity Analysis is a technique used in optimization to evaluate how changes in input parameters affect the
optimal solution.

• It helps determine the robustness of the solution and provides insights into how small variations in
constraints, objective function coefficients, or decision variables impact the optimal outcome.

54
 SENSITIVITY ANALYSIS

Why is Sensitivity Analysis Important?

1. Stability of the Optimal Solution

• Determines how much an input parameter (such as cost, resource availability, or constraint limit) can
change before the optimal solution is affected.

• Ensures that small changes in data do not lead to large fluctuations in decision-making.

2. Better Decision-Making Under Uncertainty

• Real-world problems often involve incomplete or changing data (e.g., market fluctuations, demand
changes).

• Sensitivity analysis helps prepare for different scenarios and reduces risks.
55
 SENSITIVITY ANALYSIS

3. Resource Allocation Optimization

• Helps businesses and organizations decide how much of a resource is valuable by analyzing shadow
prices (dual values).

4. Cost and Profit Optimization

• Helps companies adjust pricing, production levels, and investment decisions based on varying profit
margins or cost structures.

56
 SENSITIVITY ANALYSIS

Types of Parameter Changes:

1. Changes in the cost coefficients 𝒄𝒋

• Examines how modifications in the cost or profit coefficients affect the optimal solution.

• Determines the range within which these coefficients can change without altering the optimal basis.

2. Changes in the right-hand-side constants 𝒃𝒊

• Studies how variations in resource availability or demand affect the optimal solution.

• Uses shadow prices (dual values) to measure the rate of change in the objective function due to small
changes in RHS.

3. Changes in the coefficients of the constraints 𝒂𝒊𝒋

• Investigates how modifications in the constraint matrix affect feasibility and optimality.

• Important in applications like resource allocation and production planning.

57
 SENSITIVITY ANALYSIS

4. Addition of new variables

• Evaluates the impact of introducing a new decision variable, such as a new product in a manufacturing
process.

• Checks whether it improves the objective function and integrates into the optimal solution.

5. Addition of new constraints

• Determines whether adding a new constraint makes the current solution infeasible or changes the
optimal value.

• Useful for considering additional limitations or market conditions.

58
 SENSITIVITY ANALYSIS

Ranges comparision of 𝑪𝒌 , 𝒂𝒌 and 𝒃𝒌

𝐶𝑘 𝑎𝐾 𝑏𝐾
Case 1: Case 2: Case 1: Case 2: Case 1:
𝑪𝑲 ∉ 𝑪𝑩 𝑪𝑲 ∈ 𝑪𝑩 𝒂𝒌 ∉ 𝑩 𝒂𝒌 ∈ 𝑩
𝒁𝒌 − 𝑪𝒌 ≥ 𝚫𝐂𝐤 Max
−(𝒁𝒋 −𝑪𝒋 )
, 𝒙𝒌𝒋 > 𝟎 ≤ Max
−(𝒁𝒌 −𝑪𝒌 )
, 𝑪𝑩 𝜷𝒌 > 𝟎 ≤ Feasibility Condition: Max
−𝑿𝑩𝒊
, 𝜷𝒊𝒌 > 𝟎 ≤ 𝜟𝒃𝒌 ≤ Min
𝒙𝒌𝒋 𝑪𝑩 𝜷𝒌 𝜷𝒊𝒌
−(𝒁𝒌 −𝑪𝒌 ) −𝑿𝑩𝒊
(or) 𝜟𝑪𝒌 ≤ Min
−(𝒁𝒋 −𝑪𝒋 )
, 𝒙𝒌𝒋 < 𝟎 𝜟𝒂𝒓𝒌 ≤ Min , 𝑪𝑩 𝜷𝒌 < 𝟎 −𝑿𝑩𝒊 , 𝜷𝒊𝒌 < 𝟎
𝒙𝒌𝒋
𝑪𝑩 𝜷𝒌 Max , 𝑸𝒊 > 𝟎 ≤ 𝜟𝒂𝒓𝒌 ≤ 𝜷𝒊𝒌
𝑸𝒊
−𝑿𝑩𝒊
𝜟𝑪𝒌 ≤ 𝒁𝒌 − 𝑪𝒌 Min , 𝑸𝒊 < 𝟎
𝑸𝒊

Where
𝑸𝒊 = 𝑿𝑩𝒊 𝜷𝒌𝒌 − 𝜷𝒊𝒌 𝑿𝑩𝒌

Optimality Condition:

−(𝒁𝒋 −𝑪𝒋 )
Max , 𝑸𝒋 > 𝟎 ≤ 𝜟𝒂𝒓𝒌 ≤
𝑸𝒋
−(𝒁𝒋 −𝑪𝒋 )
Min , 𝑸𝒋 < 𝟎
𝑸𝒋

Where
𝑸𝒋 = 𝒁𝒋 − 𝑪𝒋 𝜷𝒌𝒌 − 𝑿𝒌𝒋 𝑪𝑩 𝜷𝒌

𝒋′ 𝒔 are not in the basis.

59
 SENSITIVITY ANALYSIS

1. Changes in the cost coefficients 𝒄𝒋

• Consider the LP problem be

Max 𝒁 = 𝑪𝑿
Subject to
𝑨𝑿 = 𝒃
and 𝑿 ≥ 𝟎

• Let 𝑿𝑩 be the optimum basic feasible solution of the LPP, 𝑪𝑩 be the associated cost vector of the optimum basic
feasible solution 𝑋𝐵 .
𝑿𝑩 = 𝑩−𝟏 𝒃
𝑪 = [𝑪𝟏 , 𝑪𝟐 , … , 𝑪𝒏 ]

60
 SENSITIVITY ANALYSIS

• If the change is observed in the coefficient 𝑪𝒌 (say) of the cost vector 𝐶.

• Let 𝚫𝑪𝒌 be the amount added to 𝐶𝑘 , then the new value of 𝐶𝑘 becomes
𝑪∗𝒌 = 𝑪𝒌 + 𝚫𝑪𝒌

• Even if we change the coefficient 𝑪𝒌 of the cost vector 𝐶, the optimum solution 𝑿𝑩 will not change.

• The optimality condition 𝒁𝒋 − 𝑪𝒋 ≥ 𝟎 ∀ 𝒋 is satisfied for the optimum solution 𝑿𝑩 before changes in
𝑪𝒌 , need not necessarily be satisfied when 𝑪𝒌 changed as 𝑪∗𝒌 .

• Hence the variations in the vector 𝑪 can be studied in the following two cases.

61
 SENSITIVITY ANALYSIS

• Case 1:
𝑪𝑲 ∉ 𝑪𝑩

The net evaluation corresponding to 𝑪∗𝒌 is given by

𝒁∗𝒌 − 𝑪∗𝒌 = 𝒁𝒌 − (𝑪𝒌 + 𝚫𝑪𝒌 )

The current optimum solution 𝑿𝑩 remains optimum for the new LPP if

𝑍𝑘∗ − 𝐶𝑘∗ ≥ 0

⟹ 𝑍𝑘 − 𝐶𝑘 + Δ𝐶𝑘 ≥ 0

⟹ 𝒁𝒌 − 𝑪𝒌 ≥ 𝚫𝐂𝐤 (or) 𝚫𝑪𝒌 ≤ 𝒁𝒌 − 𝑪𝒌

62
 SENSITIVITY ANALYSIS

i.e., If 𝒁𝒌 − 𝑪𝒌 ≥ 𝚫𝐂𝐤 , then the optimum solution as well as optimum value of the objective function will not
be affected.

• Case 2:
𝑪𝑲 ∈ 𝑪𝑩

The net evaluations before changes in the coefficient 𝐶𝑘

𝑍𝑗 − 𝐶𝑗 = 𝐶𝐵 𝑥𝑗 − 𝐶𝑗 , 𝑗 = 1,2, … , 𝑛

𝒁𝒋 − 𝑪𝒋 = σ𝒎
𝒊=𝟏 𝑪𝑩𝒊 𝒙𝒊𝒋 − 𝑪𝒋 , 𝒋 = 𝟏, 𝟐, … , 𝒏

63
 SENSITIVITY ANALYSIS

The net evaluations after changes in the coefficient 𝐶𝑘 with 𝐶𝑘∗ ,

𝒁∗𝒋 − 𝑪∗𝒋 = 𝒁𝒋 − 𝑪𝒋 + 𝚫𝑪𝒌 𝒙𝒌𝒋

The current basic feasible solution 𝑿𝑩 is to remain optimum solution even after the changes in
𝑪𝒌 , the Δ𝐶𝑘 must satisfy the optimality condition

𝑍𝑗∗ − 𝐶𝑗∗ ≥ 0

⟹ 𝑍𝑗 − 𝐶𝑗 + Δ𝐶𝑘 𝑥𝑘𝑗 ≥ 0

⟹ Δ𝐶𝑘 𝑥𝑘𝑗 ≥ −(𝑍𝑗 − 𝐶𝑗 )

64
 SENSITIVITY ANALYSIS

(𝑍𝑗 −𝐶𝑗 )
⟹ Δ𝐶𝑘 ≥ − 𝑥𝑘𝑗

(𝒁𝒋 −𝑪𝒋 )
⟹ 𝜟𝑪𝒌 ≥ − if 𝒙𝒌𝒋 > 𝟎
𝒙𝒌𝒋

(𝒁𝒋 −𝑪𝒋 )
⟹ 𝜟𝑪𝒌 ≤ − if 𝒙𝒌𝒋 < 𝟎
𝒙𝒌𝒋

−(𝒁𝒋 −𝑪𝒋 ) −(𝒁𝒋 −𝑪𝒋 )

i.e., Max , 𝒙𝒌𝒋 > 𝟎 ≤ 𝜟𝑪𝒌 ≤ Min , 𝒙𝒌𝒋 < 𝟎
𝒙𝒌𝒋 𝒙𝒌𝒋

The new value of the objective function becomes,

𝒁∗ = 𝒁 + 𝜟𝑪𝒌 𝑿𝒌
65
 SENSITIVITY ANALYSIS

Ranges Comparison Table

Case 1: 𝑪𝑲 ∉ 𝑪𝑩 Case 2: 𝑪𝑲 ∈ 𝑪𝑩

𝒁𝒌 − 𝑪𝒌 ≥ 𝚫𝐂𝐤 (or) 𝜟𝑪𝒌 ≤ 𝒁𝒌 − 𝑪𝒌 Max

−(𝒁𝒋 −𝑪𝒋 )
, 𝒙𝒌𝒋 > 𝟎 ≤ 𝜟𝑪𝒌 ≤ Min
−(𝒁𝒋 −𝑪𝒋 )
, 𝒙𝒌𝒋 < 𝟎
𝒙𝒌𝒋 𝒙𝒌𝒋

66
 SENSITIVITY ANALYSIS

Example 1:
Consider the LPP
Max 𝒁 = 𝟒𝒙𝟏 + 𝟔𝒙𝟐 + 𝟐𝒙𝟑
Subject to
𝒙𝟏 + 𝒙𝟐 + 𝒙𝟑 ≤ 𝟑
𝒙𝟏 + 𝟒𝒙𝟐 + 𝟕𝒙𝟑 ≤ 𝟗
And 𝒙𝟏 , 𝒙𝟐 , 𝒙 𝟑 ≥ 𝟎
Obtain the variations in 𝑪𝒋 , 𝒋 = 𝟏, 𝟐, 𝟑 which are permitted without changing the optimum solution.
Solution:

The given problem is converted to standard form by adding non-negative slack variables 𝑠1 and 𝑠2 .
The standard form of the LP problem is

67
 SENSITIVITY ANALYSIS

Maz 𝑍 = 4𝑥1 + 6𝑥2 + 2𝑥3 + 0𝑠1 + 0𝑠2 ⟹ Max 𝑍 = 𝐶𝑋

Subject to 𝑥1 + 𝑥2 + 𝑥3 + 𝑠1 + 0𝑠2 = 3 Subject to
𝑥1 + 4𝑥2 + 7𝑥3 + 0𝑠1 + 𝑠2 = 9 𝐴𝑋 = 𝑏
And 𝑥1 , 𝑥2 , 𝑥3 , 𝑠1 , 𝑠2 ≥ 0 𝑋>0
Where
𝐶 = 𝐶𝑗 = 𝐶1 , 𝐶2 , 𝐶3 , 𝐶4 , 𝐶5 = 4,6,2,0,0
The non-basic variables are 𝑥1 , 𝑥2 and 𝑥3 .
Hence the non-basic solutions are 𝑥1 = 0, 𝑥2 = 0 and 𝑥3 = 0. 𝑎11 𝑎12 𝑎13 1 1 1
𝐴= 𝑎 𝑎22 𝑎23 =
21 1 4 7
The basic variables are 𝑠1 and 𝑠2
∴ Initial Basic Feasible Solution (IBFS) are 𝑏1 3
𝑏= =
𝑏2 9
𝑠1 = 3 and 𝑠2 = 9

68
 SENSITIVITY ANALYSIS

Initial Iteration:

𝑪𝒋 𝟒 𝟔 𝟐 𝟎 𝟎 Min Ratio
𝑪𝑩 𝑩 𝑿𝑩 𝒙𝟏 𝒙𝟐 𝒙𝟑 𝒔𝟏 𝒔𝟐
𝑹𝟏 0 𝑠1 3 1 1 1 1 0 3/1=3

𝑹𝟐 0 𝑠2 9 1 [𝟒] 7 0 1 9/4=2.25

𝒁𝒋 0 0 0 0 0 0
𝒁𝒋 − 𝑪𝒋 −4 −6 −2 0 0

Since some 𝑍𝑗 − 𝐶𝑗 < 0. The current basic feasible solution is not optimal.
Negative minimum 𝑍𝑗 − 𝐶𝑗 is −6 and its column index is 2. So, the entering variable is 𝒙𝟐 .
Minimum ratio is 2.25 and its row index is 2. So, the leaving basic variable is 𝒔𝟐 .
∴ The pivot element is 4.
𝐸𝑛𝑡𝑒𝑟𝑖𝑛𝑔 = 𝑥2 , 𝐷𝑒𝑝𝑎𝑟𝑡𝑖𝑛𝑔 = 𝑠2 , 𝑃𝑖𝑣𝑜𝑡 𝐸𝑙𝑒𝑚𝑒𝑛𝑡 = 4
69
 SENSITIVITY ANALYSIS

𝑅2 𝑁𝑒𝑤 = 𝑅2 𝑂𝑙𝑑 ÷ 4

𝑅2 𝑁𝑒𝑤 = 9 1 4 7 0 1 ÷4

= 9/4 1/4 1 7/4 0 1/4

𝑅1 𝑁𝑒𝑤 = 𝑅1 𝑂𝑙𝑑 − 𝑅2 (𝑁𝑒𝑤)

= 3 1 1 1 1 0

−9/4 − 1/4 1 − 7/4 0 1/4

3/4 3/4 0 − 3/4 1 − 1/4

70
 SENSITIVITY ANALYSIS

First Iteration:

𝑪𝒋 𝟒 𝟔 𝟐 𝟎 𝟎 Min Ratio
𝑪𝑩 𝑩 𝑿𝑩 𝒙𝟏 𝒙𝟐 𝒙𝟑 𝒔𝟏 𝒔𝟐
𝑹𝟏 0 𝑠1 3/4 [𝟑/𝟒] 0 −3/4 1 −1/4 1

𝑹𝟐 6 𝑥2 9/4 1/4 1 7/4 0 1/4 9

𝒁𝒋 54/4 6/4 6 63/4 0 6/4

𝒁𝒋 − 𝑪𝒋 −10/4 0 55/4 0 6/4

Since one 𝑍𝑗 − 𝐶𝑗 < 0. The current basic feasible solution is not optimal.
Negative minimum 𝑍𝑗 − 𝐶𝑗 is −10/4 and its column index is 1. So, the entering variable is 𝒙𝟏 .
Minimum ratio is 1 and its row index is 1. So, the leaving basic variable is 𝒔𝟏 .
∴ The pivot element is 3/4.
𝐸𝑛𝑡𝑒𝑟𝑖𝑛𝑔 = 𝑥1 , 𝐷𝑒𝑝𝑎𝑟𝑡𝑖𝑛𝑔 = 𝑠1 , 𝑃𝑖𝑣𝑜𝑡 𝐸𝑙𝑒𝑚𝑒𝑛𝑡 = 3/4
71
 SENSITIVITY ANALYSIS

𝑅1 𝑁𝑒𝑤 = 𝑅1 𝑂𝑙𝑑 × 4/3

𝑅1 𝑁𝑒𝑤 = [3/4 3/4 0 − 3/4 1 − 1/4] × 4/3

= 1 1 0 −1 4/3 − 1/3

𝑅2 𝑁𝑒𝑤 = 𝑅2 𝑂𝑙𝑑 − (1/4)𝑅1 (𝑁𝑒𝑤)

= 9/4 1/4 1 7/4 0 1/4

−1/4 − 1/4 0 1/4 − 1/3 1/12

2 0 1 2 − 1/3 1/3

72
 SENSITIVITY ANALYSIS

𝑪𝒋 𝟒 𝟔 𝟐 𝟎 𝟎
𝑪𝑩 𝑩 𝑿𝑩 𝒙𝟏 𝒙𝟐 𝒙𝟑 𝒔𝟏 𝒔𝟐
𝑹𝟏 4 𝑥1 1 1 0 −1 4/3 −1/3

𝑹𝟐 6 𝑥2 2 0 1 2 −1/3 1/3

𝒁𝒋 16 4 6 8 10/3 2/3
𝒁𝒋 − 𝑪𝒋 0 0 6 10/3 2/3

Since all 𝑍𝑗 − 𝐶𝑗 ≥ 0. The current basic feasible

𝑪𝟏 = 𝟒 ∈ 𝑪𝑩 ⟹ Case 2
solution is optimal.
In the given problem,
𝑪𝟐 = 𝟔 ∈ 𝑪𝑩 ⟹ Case 2
𝑪 = 𝑪𝟏 , 𝑪𝟐 , 𝑪𝟑 = 𝟒, 𝟔, 𝟐
𝐶𝐵 = 4, 6
𝑪𝟏 = 𝟐 ∉ 𝑪𝑩 ⟹ Case 1
73
 SENSITIVITY ANALYSIS

Variation in 𝑪𝟏 :
Since 𝑪𝟏 ∈ 𝑪𝑩 , the change of 𝚫𝐂𝟏 is obtained so that the optimum solution remains same, is
given by

−(𝒁𝒋 −𝑪𝒋 ) −(𝒁𝒋 −𝑪𝒋 )

Max , 𝒙𝒌𝒋 > 𝟎 ≤ 𝜟𝑪𝒌 ≤ Min , 𝒙𝒌𝒋 < 𝟎
𝒙𝒌𝒋 𝒙𝒌𝒋

Let 𝒋 = 𝟏, 𝟐, 𝟑, 𝟒, 𝟓 and 𝒌 = 𝟏

−(𝑍1 −𝐶1 ) − 𝑍2 −𝐶2 − 𝑍3 −𝐶3 − 𝑍4 −𝐶4 − 𝑍5 −𝐶5

Max , 𝑥 , , , , 𝑥11 , 𝑥12 , 𝑥13 , 𝑥14 , 𝑥15 > 0 ≤ 𝛥𝐶1 ≤
𝑥11 12 𝑥13 𝑥14 𝑥15

−(𝑍1 −𝐶1 ) − 𝑍2 −𝐶2 − 𝑍3 −𝐶3 − 𝑍4 −𝐶4 − 𝑍5 −𝐶5

Min , , , , , 𝑥11 , 𝑥12 , 𝑥13 , 𝑥14 , 𝑥15 < 0
𝑥11 𝑥12 𝑥13 𝑥14 𝑥15

74
 SENSITIVITY ANALYSIS

10 2
−0 − −6 −
⟹ Max , −, −, 4
3
,− ≤ 𝛥𝐶1 ≤ Min −, −, −1 , −, 31
1 −
3 3

−10
⟹ Max 0, −, −, ,− ≤ 𝛥𝐶1 ≤ Min −, −, 6, −, 2
4

⟹ 𝟎 ≤ 𝜟𝑪𝟏 ≤ 𝟐
We know that 𝑪𝟏 ≤ 𝑪𝟏 + 𝚫𝑪𝟏

Adding 𝑪𝟏 on both sides, we get

4+0 ≤ 𝐶1 ≤ 4 + 2

𝑪𝟏 + 𝟎 ≤ 𝑪𝟏 + 𝜟𝑪𝟏 ≤ 𝑪𝟏 + 𝟐
⟹ 𝟒 ≤ 𝑪𝟏 ≤ 𝟔

Since 𝐶 = 𝐶1 , 𝐶2 , 𝐶3 = [4, 6, 2] ⟹ 𝑪𝟏 = 𝟒

75
 SENSITIVITY ANALYSIS

Variation in 𝑪𝟐 :
Since 𝑪𝟐 ∈ 𝑪𝑩 , the change of 𝚫𝐂𝟐 is obtained so that the optimum solution remains same, is
given by

−(𝒁𝒋 −𝑪𝒋 ) −(𝒁𝒋 −𝑪𝒋 )

Max , 𝒙𝒌𝒋 > 𝟎 ≤ 𝜟𝑪𝒌 ≤ Min , 𝒙𝒌𝒋 < 𝟎
𝒙𝒌𝒋 𝒙𝒌𝒋

Let 𝒋 = 𝟏, 𝟐, 𝟑, 𝟒, 𝟓 and 𝒌 = 𝟐

−(𝑍1 −𝐶1 ) − 𝑍2 −𝐶2 − 𝑍3 −𝐶3 − 𝑍4 −𝐶4 − 𝑍5 −𝐶5

Max , 𝑥 , , , , 𝑥11 , 𝑥12 , 𝑥13 , 𝑥14 , 𝑥15 > 0 ≤ 𝛥𝐶2 ≤
𝑥11 12 𝑥13 𝑥14 𝑥15

−(𝑍1 −𝐶1 ) − 𝑍2 −𝐶2 − 𝑍3 −𝐶3 − 𝑍4 −𝐶4 − 𝑍5 −𝐶5

Min , , , , , 𝑥11 , 𝑥12 , 𝑥13 , 𝑥14 , 𝑥15 < 0
𝑥11 𝑥12 𝑥13 𝑥14 𝑥15

76
 SENSITIVITY ANALYSIS

2 10
0 6 − −
⟹ Max −, 1 , 2 , −, 13 ≤ 𝛥𝐶2 ≤ Min −, −, −, 3
1 ,−
3
−3

⟹ Max −, 0,3, −, −2 ≤ 𝛥𝐶2 ≤ Min −, −, −, 10, −

⟹ 𝟑 ≤ 𝜟𝑪𝟐 ≤ 𝟏𝟎
We know that 𝑪𝟐 ≤ 𝑪𝟐 + 𝚫𝑪𝟐
Adding 𝑪𝟐 on both sides, we get
6 + 3 ≤ 𝐶2 ≤ 6 + 10
𝑪𝟐 + 𝟑 ≤ 𝑪𝟐 + 𝜟𝑪𝟐 ≤ 𝑪𝟐 + 𝟏𝟎
⟹ 9≤ 𝑪𝟐 ≤ 𝟏𝟔
Since 𝐶 = 𝐶1 , 𝐶2 , 𝐶3 = [4, 6, 2] ⟹ 𝑪𝟐 = 𝟔

77
 SENSITIVITY ANALYSIS

Variation in 𝑪𝟑 :
Since 𝑪𝟑 ∉ 𝑪𝑩 , the change of 𝚫𝐂𝟑 is obtained so that the optimum solution remains same, is
given by

𝛥𝐶3 ≤ 𝑍3 − 𝐶3
𝛥𝐶3 ≤ 6

𝐶3 ≤ 𝐶3 + 𝛥𝐶3
𝐶3 ≤ 2 + 6
𝑪𝟑 ≤ 𝟖

78
 SENSITIVITY ANALYSIS

2. Changes in the right-hand-side constants 𝒃𝒊

• Consider the LP problem be

Max 𝒁 = 𝑪𝑿
Subject to
𝑨𝑿 = 𝒃
and 𝑿 ≥ 𝟎

• 𝑿𝑩 = 𝑩−𝟏 𝒃 is an optimum basic feasible solution of LPP,

Where
𝑩−𝟏 is the inverse of basic solution matrix (Slack Variables),
𝒃𝟏
𝒃
𝒃= 𝟐
:
𝒃𝒎
79
 SENSITIVITY ANALYSIS

• Any changes in the requirement vector 𝒃, the optimality condition 𝒁𝒋 − 𝑪𝒋 (𝑗 = 1,2, . . , 𝑛) does
not affect.

• On the other hand, any changes in the vector b, the optimum solution 𝑿𝑩 = 𝑩−𝟏 𝒃 will be
affected.

• If one of the right-hand side constant say 𝒃𝒌 is changed to 𝒃𝒌 + 𝜟𝒃𝒌 , then new vector 𝒃∗
becomes,
𝒃𝟏
𝒃𝟐
:
𝒃∗ =
𝒃𝒌 + 𝜟𝒃𝒌
:
𝒃𝒎

80
 SENSITIVITY ANALYSIS

• New basic feasible solution is given by

𝑿∗𝑩 = 𝑩−𝟏 𝒃∗
𝛽11 𝛽12 ⋯ 𝛽1𝑘 ⋯ 𝛽1𝑚 𝑏1
𝛽21 𝛽22 ⋯ 𝛽2𝑘 ⋯ 𝛽2𝑚 𝑏2
⋮ ⋮ ⋯ ⋮ ⋯ ⋮ :
=
𝛽𝑖1 𝛽𝑖2 ⋯ 𝛽𝑖𝑘 ⋯ 𝛽𝑖𝑚 𝒃𝒌 + 𝜟𝒃𝒌
⋮ ⋮ ⋯ ⋮ ⋯ ⋮ :
𝛽𝑚1 𝛽𝑚2 ⋯ 𝛽𝑚𝑘 ⋯ 𝛽𝑚𝑚 𝑏𝑚

• Now we separate 𝑏 ∗ as,

𝑏1 0 0
𝑏2 0 0
: : :
𝑏∗ = + =𝒃+
𝒃𝒌 𝜟𝒃𝒌 𝜟𝒃𝒌
: : :
𝑏𝑚 0 0

81
 SENSITIVITY ANALYSIS

• Then 𝑿∗𝑩 becomes

0
0
:
𝑋𝐵∗ = 𝐵−1 𝑏+
𝛥𝑏𝑘
:
0

0
0
:
= 𝐵−1 𝑏 + 𝐵−1
𝛥𝑏𝑘
:
0

𝑿∗𝑩 = 𝑿𝑩 + 𝜷𝒌 𝜟𝒃𝒌 , where 𝜷𝒌 is 𝒌𝐭𝐡 column vector.

𝑿∗𝑩𝒊 = 𝑿𝑩𝒊 + 𝜷𝒊𝒌 𝜟𝒃𝒌 , where 𝜷𝒊𝒌 is the 𝒊, 𝒌 𝐭𝐡 element of 𝑩−𝟏 and 𝒌𝐭𝐡 element of 𝜷𝒌

82
 SENSITIVITY ANALYSIS

• 𝑿∗𝑩 must be non-negative. Because feasibility of the solution.

∴ We must have 𝑿∗𝑩𝒊 ≥ 𝟎, 𝒊 = 𝟏, 𝟐, . . , 𝒎.

⟹ 𝑋𝐵𝑖 + 𝛽𝑖𝑘 𝛥𝑏𝑘 ≥ 0

⟹ 𝛽𝑖𝑘 𝛥𝑏𝑘 ≥ −𝑋𝐵𝑖

−𝑋𝐵𝑖
⟹ 𝛥𝑏𝑘 ≥
𝛽𝑖𝑘

−𝑿𝑩𝒊 −𝑿𝑩𝒊
⟹ 𝜟𝒃𝒌 ≥ for 𝜷𝒊𝒌 > 𝟎 and 𝜟𝒃𝒌 ≤ for 𝜷𝒊𝒌 < 𝟎
𝜷𝒊𝒌 𝜷𝒊𝒌

83
 SENSITIVITY ANALYSIS

• The range of 𝜟𝒃𝒌 to maintain the feasibility of the current solution would be

−𝑿𝑩𝒊 −𝑿𝑩𝒊
Max , 𝜷𝒊𝒌 > 𝟎 ≤ 𝜟𝒃𝒌 ≤ Min , 𝜷𝒊𝒌 < 𝟎
𝜷𝒊𝒌 𝜷𝒊𝒌

84
 SENSITIVITY ANALYSIS

Example 2:
Consider the LPP
Max 𝒁 = 𝟒𝒙𝟏 + 𝟔𝒙𝟐 + 𝟐𝒙𝟑
Subject to
𝒙𝟏 + 𝒙𝟐 + 𝒙𝟑 ≤ 𝟑
𝒙𝟏 + 𝟒𝒙𝟐 + 𝟕𝒙𝟑 ≤ 𝟗
And 𝒙𝟏 , 𝒙𝟐 , 𝒙 𝟑 ≥ 𝟎
Determine the ranges for discrete changes in the components 𝒃𝟏 and 𝒃𝟐 of the requirement
vector so as to maintain the feasibility of the current optimum solution.
Solution:
Similar step of Example 1 solution
The given problem is converted to standard form by adding non-negative slack variables 𝑠1 and 𝑠2 .
The standard form of the LP problem is

85
 SENSITIVITY ANALYSIS

Maz 𝑍 = 4𝑥1 + 6𝑥2 + 2𝑥3 + 0𝑠1 + 0𝑠2 ⟹ Max 𝑍 = 𝐶𝑋

Subject to 𝑥1 + 𝑥2 + 𝑥3 + 𝑠1 + 0𝑠2 = 3 Subject to
𝑥1 + 4𝑥2 + 7𝑥3 + 0𝑠1 + 𝑠2 = 9 𝐴𝑋 = 𝑏
And 𝑥1 , 𝑥2 , 𝑥3 , 𝑠1 , 𝑠2 ≥ 0 𝑋>0
Where
𝐶 = 𝐶𝑗 = 𝐶1 , 𝐶2 , 𝐶3 , 𝐶4 , 𝐶5 = 4,6,2,0,0
The non-basic variables are 𝑥1 , 𝑥2 and 𝑥3 .
Hence the non-basic solutions are 𝑥1 = 0, 𝑥2 = 0 and 𝑥3 = 0. 𝑎11 𝑎12 𝑎13 1 1 1
𝐴= 𝑎 𝑎22 𝑎23 =
21 1 4 7
The basic variables are 𝑠1 and 𝑠2
∴ Initial Basic Feasible Solution (IBFS) are 𝑏1 3
𝑏= =
𝑏2 9
𝑠1 = 3 and 𝑠2 = 9

86
 SENSITIVITY ANALYSIS

Initial Iteration:

𝑪𝒋 𝟒 𝟔 𝟐 𝟎 𝟎 Min Ratio
𝑪𝑩 𝑩 𝑿𝑩 𝒙𝟏 𝒙𝟐 𝒙𝟑 𝒔𝟏 𝒔𝟐
𝑹𝟏 0 𝑠1 3 1 1 1 1 0 3/1=3

𝑹𝟐 0 𝑠2 9 1 [𝟒] 7 0 1 9/4=2.25

𝒁𝒋 0 0 0 0 0 0
𝒁𝒋 − 𝑪𝒋 −4 −6 −2 0 0

Since some 𝑍𝑗 − 𝐶𝑗 < 0. The current basic feasible solution is not optimal.
Negative minimum 𝑍𝑗 − 𝐶𝑗 is −6 and its column index is 2. So, the entering variable is 𝒙𝟐 .
Minimum ratio is 2.25 and its row index is 2. So, the leaving basic variable is 𝒔𝟐 .
∴ The pivot element is 4.
𝐸𝑛𝑡𝑒𝑟𝑖𝑛𝑔 = 𝑥2 , 𝐷𝑒𝑝𝑎𝑟𝑡𝑖𝑛𝑔 = 𝑠2 , 𝑃𝑖𝑣𝑜𝑡 𝐸𝑙𝑒𝑚𝑒𝑛𝑡 = 4
87
 SENSITIVITY ANALYSIS

𝑅2 𝑁𝑒𝑤 = 𝑅2 𝑂𝑙𝑑 ÷ 4

𝑅2 𝑁𝑒𝑤 = 9 1 4 7 0 1 ÷4

= 9/4 1/4 1 7/4 0 1/4

𝑅1 𝑁𝑒𝑤 = 𝑅1 𝑂𝑙𝑑 − 𝑅2 (𝑁𝑒𝑤)

= 3 1 1 1 1 0

−9/4 − 1/4 1 − 7/4 0 1/4

3/4 3/4 0 − 3/4 1 − 1/4

88
 SENSITIVITY ANALYSIS

First Iteration:

𝑪𝒋 𝟒 𝟔 𝟐 𝟎 𝟎 Min Ratio
𝑪𝑩 𝑩 𝑿𝑩 𝒙𝟏 𝒙𝟐 𝒙𝟑 𝒔𝟏 𝒔𝟐
𝑹𝟏 0 𝑠1 3/4 [𝟑/𝟒] 0 −3/4 1 −1/4 1

𝑹𝟐 6 𝑥2 9/4 1/4 1 7/4 0 1/4 9

𝒁𝒋 54/4 6/4 6 63/4 0 6/4

𝒁𝒋 − 𝑪𝒋 −10/4 0 55/4 0 6/4

Since one 𝑍𝑗 − 𝐶𝑗 < 0. The current basic feasible solution is not optimal.
Negative minimum 𝑍𝑗 − 𝐶𝑗 is −10/4 and its column index is 1. So, the entering variable is 𝒙𝟏 .
Minimum ratio is 1 and its row index is 1. So, the leaving basic variable is 𝒔𝟏 .
∴ The pivot element is 3/4.
𝐸𝑛𝑡𝑒𝑟𝑖𝑛𝑔 = 𝑥1 , 𝐷𝑒𝑝𝑎𝑟𝑡𝑖𝑛𝑔 = 𝑠1 , 𝑃𝑖𝑣𝑜𝑡 𝐸𝑙𝑒𝑚𝑒𝑛𝑡 = 3/4
89
 SENSITIVITY ANALYSIS

𝑅1 𝑁𝑒𝑤 = 𝑅1 𝑂𝑙𝑑 × 4/3

𝑅1 𝑁𝑒𝑤 = [3/4 3/4 0 − 3/4 1 − 1/4] × 4/3

= 1 1 0 −1 4/3 − 1/3

𝑅2 𝑁𝑒𝑤 = 𝑅2 𝑂𝑙𝑑 − (1/4)𝑅1 (𝑁𝑒𝑤)

= 9/4 1/4 1 7/4 0 1/4

−1/4 − 1/4 0 1/4 − 1/3 1/12

2 0 1 2 − 1/3 1/3

90
 SENSITIVITY ANALYSIS

Second Iteration:

𝑪𝒋 𝟒 𝟔 𝟐 𝟎 𝟎
𝑪𝑩 𝑩 𝑿𝑩 𝒙𝟏 𝒙𝟐 𝒙𝟑 𝒔𝟏 𝒔𝟐
𝑹𝟏 4 𝑥1 1 1 0 −1 𝟒/𝟑 −𝟏/𝟑

𝑹𝟐 6 𝑥2 2 0 1 2 −𝟏/𝟑 𝟏/𝟑

𝒁𝒋 16 4 6 8 10/3 2/3
𝒁𝒋 − 𝑪𝒋 0 0 6 10/3 2/3

Since all 𝑍𝑗 − 𝐶𝑗 ≥ 0. The current basic feasible solution is optimal.

Hence, optimal solution at 𝑥1 = 1, 𝑥2 = 2, 𝑥3 = 0 is
Max 𝒁 = 𝟒𝒙𝟏 + 𝟔𝒙𝟐 + 𝟐𝒙𝟑
= 4 1 + 6 2 + 2(0)
= 4 + 12
= 16 91
 SENSITIVITY ANALYSIS

In the given problem,

𝑏1 3
𝑏= =
𝑏2 9

From the above final optimum simplex table,

𝑋𝐵1 1
𝑋𝐵 = =
𝑋𝐵2 2

𝛽11 𝛽12 𝟒/𝟑 −𝟏/𝟑

𝐵−1 = 𝛽1 𝛽2 = =
𝛽21 𝛽22 −𝟏/𝟑 𝟏/𝟑

92
 SENSITIVITY ANALYSIS

Variation in 𝒃𝟏 :
Let 𝜟𝒃𝟏 be the change in 𝑏1 .
−𝑿𝑩𝒊 −𝑿𝑩𝒊
Max , 𝜷𝒊𝒌 > 𝟎 ≤ 𝜟𝒃𝒌 ≤ Min , 𝜷𝒊𝒌 < 𝟎
𝜷𝒊𝒌 𝜷𝒊𝒌

Let 𝒊 = 𝟏, 𝟐 and 𝒌 = 𝟏. i.e., consider the first column of 𝑩−𝟏 .

−𝑿𝑩𝟏 −𝑿𝑩𝟐 −𝑿𝑩𝒊 −𝑿𝑩𝒊

⟹ Max , 𝜷 , 𝜷𝟏𝟏 , 𝜷𝟐𝟏 > 𝟎 ≤ 𝜟𝒃𝟏 ≤ Min , , 𝜷𝟏𝟏 , 𝜷𝟐𝟏 < 𝟎
𝜷𝟏𝟏 𝟐𝟏 𝜷𝟏𝟏 𝜷𝟐𝟏

−𝟏 −𝟐
⟹ Max , − ≤ 𝜟𝒃𝟏 ≤ Min −, −𝟏/𝟑
𝟒/𝟑

𝟑
⟹ − 𝟒 ≤ 𝜟𝒃𝟏 ≤ 𝟔

93
 SENSITIVITY ANALYSIS

Adding 𝒃𝟏 on both sides, we get

⟹ 𝒃𝟏 − 𝟑/𝟒 ≤ 𝒃𝟏 + 𝜟𝒃𝟏 ≤ 𝒃𝟏 + 𝟔

We know that,
𝒃𝟏 ≤ 𝒃𝟏 + 𝜟𝒃𝟏

𝑏1 3
Since 𝑏 = = ⟹ 𝒃𝟏 = 𝟑
𝑏2 9
∴ The range of 𝒃𝟏 to maintain the feasibility of the LPP is given by,

𝟑
𝟑 − 𝟒 ≤ 𝒃𝟏 ≤ 𝟑 + 𝟔

𝟗
⟹ ≤ 𝒃𝟏 ≤ 𝟗
𝟒

94
 SENSITIVITY ANALYSIS

Variation in 𝒃𝟐 :
Let 𝜟𝒃𝟐 be the change in 𝑏1 .
−𝑿𝑩𝒊 −𝑿𝑩𝒊
Max , 𝜷𝒊𝒌 > 𝟎 ≤ 𝜟𝒃𝒌 ≤ Min , 𝜷𝒊𝒌 < 𝟎
𝜷𝒊𝒌 𝜷𝒊𝒌

Let 𝒊 = 𝟏, 𝟐 and 𝒌 = 𝟐. i.e., consider the second column of 𝑩−𝟏 .

−𝑿𝑩𝟏 −𝑿𝑩𝟐 −𝑿𝑩𝒊𝟏 −𝑿𝑩𝟐

⟹ Max , 𝜷 , 𝜷𝟏𝟐 , 𝜷𝟏𝟐 > 𝟎 ≤ 𝜟𝒃𝟐 ≤ Min , 𝜷 , 𝜷𝟏𝟐 , 𝜷𝟏𝟐 < 𝟎
𝜷𝟏𝟐 𝟐𝟐 𝜷𝟏𝟐 𝟐𝟐

−𝟐 −𝟏
⟹ Max −, 𝟏/𝟑 ≤ 𝜟𝒃𝟐 ≤ Min ,−
−𝟏/𝟑

⟹ −𝟔 ≤ 𝜟𝒃𝟐 ≤ 𝟑

95
 SENSITIVITY ANALYSIS

Adding 𝒃𝟐 on both sides, we get

⟹ 𝒃𝟐 − 𝟔 ≤ 𝒃𝟐 + 𝜟𝒃𝟐 ≤ 𝒃𝟐 + 𝟑

We know that,
𝒃𝟐 ≤ 𝒃𝟐 + 𝜟𝒃𝟐

𝑏1 3
Since 𝑏 = = ⟹ 𝒃𝟐 = 𝟗
𝑏2 9
∴ The range of 𝒃𝟐 to maintain the feasibility of the LPP is given by,

𝟗 − 𝟔 ≤ 𝒃𝟐 ≤ 𝟗 + 𝟑

⟹ 𝟑 ≤ 𝒃𝟐 ≤ 𝟏𝟐
96
 SENSITIVITY ANALYSIS

3. Changes in the coefficients of the constraints 𝒂𝒊𝒋

• Consider the LP problem be

Max 𝒁 = 𝑪𝑿
Subject to
𝑨𝑿 = 𝒃
and 𝑿 ≥ 𝟎
where
𝑨 is an 𝒎 × 𝒏 matrix,

• If any coefficient 𝒂𝒊𝒋 (𝑖 = 1,2, … . , 𝑚, 𝑗 = 1,2, . . , 𝑛) is changed in the matrix 𝐴, let us suppose that some 𝒂𝒓𝒌 is
changed to 𝒂𝒓𝒌 + 𝚫𝐚𝐫𝐤 , consider it as
𝒂𝒓𝒌 = 𝒂𝒓𝒌 + 𝚫𝒂𝐫𝐤

97
 SENSITIVITY ANALYSIS

• This element will be in the 𝒌𝒕𝒉 column of 𝑨, let 𝑎𝑘∗

𝒂𝟏𝒌
𝒂𝟐𝒌
⋮
∴ 𝒂∗𝒌 = 𝒂 + 𝚫𝒂
𝒓𝒌 𝒓𝒌
⋮
𝒂𝒎𝒌

• This variation may occur in the following two possible cases

1. 𝒂𝒌 ∉ 𝑩 i.e., the element 𝒂𝒓𝒌 = 𝒂𝒓𝒌 + 𝚫𝒂𝐫𝐤 in 𝒂𝒌 does not belongs to the basis matrix 𝑩.

2. 𝒂𝒌 ∈ 𝑩 i.e., the element 𝒂𝒓𝒌 = 𝒂𝒓𝒌 + 𝚫𝒂𝐫𝐤 in 𝒂𝒌 belongs to the basis matrix 𝑩.

98
 SENSITIVITY ANALYSIS

• Case 1:
𝒂𝒌 ∉ 𝑩

• The post optimal changes do not affect on the feasibility of the current optimum solution 𝑿𝑩 , but
only the optimality conditions 𝒁𝒋 − 𝑪𝒋 will be affected.

• For maintaining the optimality conditions of the LPP, we must have

𝒁∗𝒋 − 𝑪∗𝒋 ≥ 𝟎

where
𝑍𝑗∗ = 𝐶𝐵 𝑥𝑘∗
= 𝐶𝐵 𝐵−1 𝑎𝑘∗

99
 SENSITIVITY ANALYSIS

𝒂𝟏𝒌
𝒂𝟐𝒌
⋮
𝑍𝑘∗ = 𝐶𝐵 𝐵−1 𝒂𝒓𝒌 + 𝜟𝒂𝒓𝒌
⋮
𝒂𝒎𝒌

𝒂𝟏𝒌 𝟎
𝒂𝟐𝒌 𝟎
⋮ ⋮
= 𝐶𝐵 𝐵−1 𝒂𝒓𝒌 + 𝐶𝐵 𝑩−𝟏
𝜟𝒂𝒓𝒌
⋮ ⋮
𝒂𝒎𝒌 𝟎

0
0
⋮
= 𝐶𝐵 𝐵−1 𝑎𝑘 + 𝐶𝐵 [𝜷𝟏 𝜷𝟐 … 𝜷𝒌 … 𝜷𝒎 ]
Δ𝑎𝑟𝑘
⋮
0
100
 SENSITIVITY ANALYSIS

𝑍𝑘∗ = 𝐶𝐵 𝐵−1 𝑎𝑘 + 𝐶𝐵 𝜷𝒌 𝚫𝒂𝒓𝒌

𝑍𝑘∗ = 𝒁𝒌 + 𝐶𝐵 𝛽𝑘 Δ𝑎𝑟𝑘

𝑍𝑘∗ − 𝐶𝑘∗ = 𝑍𝑘 + 𝐶𝐵 𝛽𝑘 Δ𝑎𝑟𝑘 − 𝐶𝑘

Hence for maintaining the optimality condition,

𝑍𝑘∗ − 𝐶𝑘∗ ≥ 0

⟹ 𝑍𝑘 + 𝐶𝐵 𝛽𝑘 Δ𝑎𝑟𝑘 − 𝐶𝑘 ≥ 0

⟹ 𝐶𝐵 𝛽𝑘 Δ𝑎𝑟𝑘 ≥ −(𝑍𝑘 − 𝐶𝑘 )
101
 SENSITIVITY ANALYSIS

−(𝑍𝑘 −𝐶𝑘 )
⟹ Δ𝑎𝑟𝑘 ≥ 𝐶𝐵 𝛽𝑘

−(𝒁𝒌 −𝑪𝒌 )
⟹ 𝜟𝒂𝒓𝒌 ≥ for 𝑪𝑩 𝜷𝒌 > 𝟎
𝑪𝑩 𝜷𝒌

−(𝒁𝒌 −𝑪𝒌 )
⟹ 𝜟𝒂𝒓𝒌 ≤ for 𝑪𝑩 𝜷𝒌 < 𝟎
𝑪𝑩 𝜷𝒌

⟹ 𝜟𝒂𝒓𝒌 is unrestricted if 𝑪𝑩 𝜷𝒌 = 𝟎

Hence the range of 𝜟𝒂𝒓𝒌 in the coefficient matrix 𝐴 for maintaining the optimality condition is given by,

−(𝒁𝒌 −𝑪𝒌 ) −(𝒁𝒌 −𝑪𝒌 )

Max , 𝑪𝑩 𝜷𝒌 > 𝟎 ≤ 𝜟𝒂𝒓𝒌 ≤ Min , 𝑪𝑩 𝜷𝒌 <𝟎
𝑪𝑩 𝜷𝒌 𝑪𝑩 𝜷𝒌

102
 SENSITIVITY ANALYSIS

• Case 2:
𝒂𝒌 ∈ 𝑩

• The current optimum solution 𝑿𝑩 and the optimality conditions 𝒁𝒋 − 𝑪𝒋 both are affected.

• We need to find a specified range for the discrete change Δ𝑎𝑟𝑘 in 𝑎𝑟𝑘 so that feasibility and optimality are
to be maintained.

• Let 𝒂𝒓𝒌 is changed to 𝒂𝒓𝒌 + 𝚫𝒂𝒓𝒌

Conditions for maintaining feasibility of the solution:

Let 𝑩 = 𝒃𝟏 𝒃𝟐 … 𝒃𝒌 … 𝒃𝒎
where, 𝒃𝒋 ∈ 𝑨, 𝒋 = 𝟏, 𝟐, … , 𝒎.
103
 SENSITIVITY ANALYSIS

𝒃𝟏 , 𝒃𝟐 , … , 𝒃𝒎 are column vectors of basis matrix 𝑩.

𝑩−𝟏 = [𝜷𝟏 𝜷𝟐 … 𝜷𝒌 … 𝜷𝒎 ]

Where, 𝜷𝟏 𝜷𝟐 … 𝜷𝒌 … 𝜷𝒎 are column vectors of 𝑩−𝟏 .

• Since 𝒂𝒌 ∈ 𝑩, then any change of 𝑎𝑟𝑘 will affect the 𝑘𝑡ℎ column of 𝐵. (𝑖. 𝑒. , 𝑏𝑘 )

• Therefore, after post optimal changes, the change in 𝑩 by indicating 𝑩∗ , we get

𝑩∗ = 𝒃𝟏 𝒃𝟐 … 𝒃∗𝒌 … 𝒃𝒎
𝒃𝟏𝒌
𝒃𝟐𝒌
⋮
Where 𝒃∗𝒌 =
𝒃𝒓𝒌 + 𝚫𝒂𝒓𝒌
⋮
𝒃𝒎𝒌
104
 SENSITIVITY ANALYSIS

Now 𝒃∗𝒌 can be expressed as a linear combination of the basis vectors of 𝑩.

𝒃∗𝒌 = 𝝀𝟏 𝒃𝟏 + 𝝀𝟐 𝒃𝟐 + ⋯ + 𝝀𝒌 𝒃𝒌 + ⋯ + 𝝀𝒎 𝒃𝒎

𝒃∗𝒌 = 𝑩𝝀,

⟹ 𝝀 = 𝑩−𝟏 𝒃∗𝒌

𝒃𝟏𝒌
𝒃𝟐𝒌
⋮
⟹ 𝝀 = 𝑩−𝟏
𝒃𝒓𝒌 + 𝚫𝒂𝒓𝒌
⋮
𝒃𝒎𝒌

105
 SENSITIVITY ANALYSIS

𝒃𝟏𝒌 𝟎
𝒃𝟐𝒌 𝟎
⋮ ⋮
⟹ 𝝀 = 𝑩−𝟏 + 𝑩−𝟏
𝒃𝒓𝒌 𝚫𝒂𝒓𝒌
⋮ ⋮
𝒃𝒎𝒌 𝟎

𝟎
𝟎
⋮
⟹ 𝝀 = 𝑩−𝟏 𝒃𝒌 + 𝑩−𝟏
𝚫𝒂𝒓𝒌
⋮
𝟎

𝟎
𝟎
⋮
⟹ 𝝀 = 𝑩−𝟏 𝒃𝒌 + [𝜷𝟏 𝜷𝟐 … 𝜷𝒌 … 𝜷𝒎 ]
𝚫𝒂𝒓𝒌
⋮
𝟎 106
 SENSITIVITY ANALYSIS

⟹ 𝝀 = 𝑩−𝟏 𝒃𝒌 + 𝜷𝒌 𝚫𝒂𝒓𝒌

⟹ 𝝀 = 𝒆𝒌 + 𝜷𝒌 𝚫𝒂𝒓𝒌

where 𝑒𝑘 = 𝐵 −1 𝑏𝑘 = σ𝑚
𝑖=1 𝛽𝑖 𝑏𝑖𝑘

𝜆1 0 𝛽1𝑘
𝜆2 0 𝛽2𝑘
⋮ ⋮
= + ⋮ Δ𝑎𝑟𝑘
𝜆𝑘 1 𝛽 𝑘𝑘
⋮ ⋮ ⋮
𝜆𝑚 𝛽
0 𝑚𝑘

𝝀𝒊 = 𝜷𝒊𝒌 𝚫𝒂𝒓𝒌 for 𝒊 ≠ 𝒌

𝝀𝒌 = 𝟏 + 𝜷𝒌𝒌 𝚫𝒂𝒓𝒌 for 𝒊 = 𝒌

107
 SENSITIVITY ANALYSIS

• For maintaining feasibility of the new solution, we should have

𝑿∗𝑩 = 𝑩∗ −𝟏 𝒃 ≥ 𝟎

where 𝑩∗ −𝟏 exists only if 𝐵 ∗ is non-singular.

• So we must have 𝜆𝑘 ≠ 0

• 𝑿∗𝑩 can be expressed as

∗
𝑿𝑩 = 𝑩∗ −𝟏 𝒃 = 𝑬𝑩−𝟏 𝒃 (∵ 𝐵 ∗ −1 = 𝐸𝐵 −1 )

• Now consider

𝒃∗𝒌 = 𝝀𝟏 𝒃𝟏 + 𝝀𝟐 𝒃𝟐 + ⋯ + 𝝀𝒌 𝒃𝒌 + ⋯ + 𝝀𝒎 𝒃𝒎

𝝀𝒌 𝒃𝒌 = −𝝀𝟏 𝒃𝟏 − 𝝀𝟐 𝒃𝟐 − ⋯ − 𝒃∗𝒌 − ⋯ − 𝝀𝒎 𝒃𝒎

108
 SENSITIVITY ANALYSIS

Both sides divide with 𝝀𝒌 we get

𝜆 𝑏 𝜆2 𝑏2 𝑏𝑘∗ 𝜆𝑚 𝑏𝑚
𝑏𝑘 = − 𝜆1 1 − 𝜆𝑘
− ⋯+ 𝜆𝑘
− ⋯− 𝜆𝑘
𝑘

The matrix 𝐸 differs from the identity in the 𝑘 𝑡ℎ column only.

𝜆
1 0 ⋯ − 𝜆1 ⋯ 0
𝑘
𝜆
0 1 ⋯ − 𝜆2 ⋯ 0
𝑘

𝐸= ⋮ ⋮ ⋯
1
⋮ ⋯ ⋮ , 𝜆𝑘 ≠ 0
0 0 ⋯ ⋯ 0
𝜆𝑘
⋮ ⋮ ⋯ ⋮ ⋯ ⋮
𝜆
0 0 ⋯ − 𝜆𝑚 ⋯ 1
𝑘

109
 SENSITIVITY ANALYSIS

The optimum solution after the changes 𝑋𝐵∗ is obtained by

∗
𝑿𝑩 = 𝑬𝑩−𝟏 𝒃

∗
𝑿𝑩 = 𝑬𝑿𝑩

𝜆
1 0 ⋯ − 𝜆1 ⋯ 0
∗ 𝑘
𝑋𝐵1 𝜆
𝑋𝐵1
∗
𝑋𝐵2 0 1 ⋯ − 𝜆2 ⋯ 0 𝑋𝐵2
𝑘
⋮ ⋮
∗ = ⋮ ⋮ ⋯ ⋮ ⋯ ⋮
𝑋𝐵𝑘 0 0 ⋯
1
⋯ 0 𝑋𝐵𝑘
⋮ 𝜆𝑘 ⋮
∗
𝑋𝐵𝑚 ⋮ ⋮ ⋯ ⋮ ⋯ ⋮ 𝑋𝐵𝑚
𝜆𝑚
0 0 ⋯ − ⋯ 1
𝜆𝑘

110
 SENSITIVITY ANALYSIS

𝜆
∗
𝑋𝐵1 − 𝜆1 × 𝑋𝐵𝑘
𝑋𝐵1 𝑘
∗ 𝜆
𝑋𝐵2 𝑋𝐵2 − 𝜆2 × 𝑋𝐵𝑘
𝑘
⋮ ⋮
∗ =
𝑋𝐵𝑘 𝑋𝐵𝑘 /𝜆𝑘
⋮ ⋮
∗
𝑋𝐵𝑚 𝜆
𝑋𝐵𝑚 − 𝜆𝑚 × 𝑋𝐵𝑘
𝑘

𝝀
𝑋𝐵𝑖 − 𝝀 𝒊 × 𝑋𝐵𝑘 , 𝑖 ≠ 𝑘
∗ 𝒌
∴ 𝑋𝐵𝑖 = ൞𝑋
𝐵𝑘
, 𝑖=𝑘
𝝀𝒌

Substitute 𝝀𝒊 , 𝝀𝒌 values, we get

𝜷 𝚫𝒂𝒓𝒌
𝑋𝐵𝑖 − 𝟏+𝜷𝒊𝒌 × 𝑋𝐵𝑘 , 𝑖 ≠ 𝑘
∗ 𝒌𝒌 𝚫𝒂𝒓𝒌
𝑋𝐵𝑖 =൞ 𝑋𝐵𝑘
, 𝑖=𝑘
𝟏+𝜷𝒌𝒌 𝚫𝒂𝒓𝒌

111
 SENSITIVITY ANALYSIS

Therefore, for maintaining the feasibility of the solution, we must have

∗ ∗
𝑿𝑩 ≥ 𝟎 for which 𝑿𝑩𝒊 ≥𝟎

𝜷𝒊𝒌 𝚫𝒂𝒓𝒌 𝑋𝐵𝑘

⟹ 𝑋𝐵𝑖 − ≥𝟎
𝟏+𝜷𝒌𝒌 𝚫𝒂𝒓𝒌

𝛽𝑖𝑘 𝛥𝑎𝑟𝑘 𝑋𝐵𝑘

⟹ 𝑋𝐵𝑖 ≥ 1+𝛽𝑘𝑘 𝛥𝑎𝑟𝑘

⟹ 𝑋𝐵𝑖 (1 + 𝛽𝑘𝑘 𝛥𝑎𝑟𝑘 ) ≥ 𝛽𝑖𝑘 𝛥𝑎𝑟𝑘 𝑋𝐵𝑘

⟹ 𝑋𝐵𝑖 + 𝑋𝐵𝑖 𝛽𝑘𝑘 𝛥𝑎𝑟𝑘 ≥ 𝛽𝑖𝑘 𝛥𝑎𝑟𝑘 𝑋𝐵𝑘

⟹ 𝑋𝐵𝑖 + 𝑋𝐵𝑖 𝛽𝑘𝑘 𝛥𝑎𝑟𝑘 − 𝛽𝑖𝑘 𝛥𝑎𝑟𝑘 𝑋𝐵𝑘 ≥ 0

112
 SENSITIVITY ANALYSIS

⟹ 𝛥𝑎𝑟𝑘 (𝑋𝐵𝑖 𝛽𝑘𝑘 − 𝛽𝑖𝑘 𝑋𝐵𝑘 ) ≥ −𝑋𝐵𝑖

−𝒙𝑩𝒊
⟹ 𝜟𝒂𝒓𝒌 ≥ 𝑿𝜷 for 𝑿𝑩𝒊 𝜷𝒌𝒌 − 𝜷𝒊𝒌 𝑿𝑩𝒌 > 𝟎
𝒌𝒌 −𝜷𝒊𝒌 𝑿𝑩𝒌

−𝑿𝑩𝒊
⟹ 𝜟𝒂𝒓𝒌 ≤ 𝑿 for𝑿𝑩𝒊 𝜷𝒌𝒌 − 𝜷𝒊𝒌 𝑿𝑩𝒌 < 𝟎
𝑩𝒊 𝜷𝒌𝒌 −𝜷𝒊𝒌 𝑿𝑩𝒌

Let 𝑸𝒊 = 𝑿𝑩𝒊 𝜷𝒌𝒌 − 𝜷𝒊𝒌 𝑿𝑩𝒌

The range of 𝜟𝒂𝒓𝒌 for maintaining the feasibility of the solution is given by

−𝑿𝑩𝒊 −𝑿𝑩𝒊
Max , 𝑸𝒊 > 𝟎 ≤ 𝜟𝒂𝒓𝒌 ≤ Min , 𝑸𝒊 <𝟎
𝑸𝒊 𝑸𝒊

Conditions for maintaining the optimality of the solution:

−(𝒁𝒋 −𝑪𝒋 ) −(𝒁𝒋 −𝑪𝒋 )

Max , 𝑸𝒋 > 𝟎 ≤ 𝜟𝒂𝒓𝒌 ≤ Min , 𝑸𝒋 < 𝟎 Where 𝑸𝒋 = 𝒁𝒋 − 𝑪𝒋 𝜷𝒌𝒌 − 𝑿𝒌𝒋 𝑪𝑩 𝜷𝒌 𝒋′ 𝒔 are not in the
𝑸𝒋 𝑸𝒋

basis.
113
 SENSITIVITY ANALYSIS

Ranges Comparison Table

Case 1: 𝒂𝒌 ∉ 𝑩 Case 2: 𝒂𝒌 ∈ 𝑩

Max
−(𝒁𝒌 −𝑪𝒌 )
, 𝑪𝑩 𝜷𝒌 > 𝟎 ≤ 𝜟𝒂𝒓𝒌 ≤ Min
−(𝒁𝒌 −𝑪𝒌 )
, 𝑪𝑩 𝜷𝒌 <𝟎 Feasibility Condition:
𝑪𝑩 𝜷 𝒌 𝑪𝑩 𝜷 𝒌
−𝑿𝑩𝒊 −𝑿𝑩𝒊
Max , 𝑸𝒊 > 𝟎 ≤ 𝜟𝒂𝒓𝒌 ≤ Min , 𝑸𝒊 <𝟎
𝑸𝒊 𝑸𝒊

Where
𝑸𝒊 = 𝑿𝑩𝒊 𝜷𝒌𝒌 − 𝜷𝒊𝒌 𝑿𝑩𝒌

Optimality Condition:

−(𝒁𝒋 −𝑪𝒋 ) −(𝒁𝒋 −𝑪𝒋 )

Max , 𝑸𝒋 > 𝟎 ≤ 𝜟𝒂𝒓𝒌 ≤ Min , 𝑸𝒋 < 𝟎
𝑸𝒋 𝑸𝒋

Where
𝑸𝒋 = 𝒁𝒋 − 𝑪𝒋 𝜷𝒌𝒌 − 𝑿𝒌𝒋 𝑪𝑩 𝜷𝒌

𝒋′ 𝒔 are not in the basis.

114
 SENSITIVITY ANALYSIS

Example 3:
Consider the LPP
Max 𝒁 = 𝟒𝒙𝟏 + 𝟔𝒙𝟐 + 𝟐𝒙𝟑
Subject to
𝒙𝟏 + 𝒙𝟐 + 𝒙𝟑 ≤ 𝟑
𝒙𝟏 + 𝟒𝒙𝟐 + 𝟕𝒙𝟑 ≤ 𝟗
And 𝒙𝟏 , 𝒙𝟐 , 𝒙 𝟑 ≥ 𝟎
Find the ranges of 𝒂𝟏𝟏 and 𝒂𝟐𝟑 so that the new solution remains optimal feasible solution.
Solution:
Similar step of Example 1 solution
The given problem is converted to standard form by adding non-negative slack variables 𝑠1 and 𝑠2 .
The standard form of the LP problem is

115
 SENSITIVITY ANALYSIS

Maz 𝑍 = 4𝑥1 + 6𝑥2 + 2𝑥3 + 0𝑠1 + 0𝑠2 ⟹ Max 𝑍 = 𝐶𝑋

Subject to 𝑥1 + 𝑥2 + 𝑥3 + 𝑠1 + 0𝑠2 = 3 Subject to
𝑥1 + 4𝑥2 + 7𝑥3 + 0𝑠1 + 𝑠2 = 9 𝐴𝑋 = 𝑏
And 𝑥1 , 𝑥2 , 𝑥3 , 𝑠1 , 𝑠2 ≥ 0 𝑋>0
Where
𝐶 = 𝐶𝑗 = 𝐶1 , 𝐶2 , 𝐶3 , 𝐶4 , 𝐶5 = 4,6,2,0,0
The non-basic variables are 𝑥1 , 𝑥2 and 𝑥3 .
Hence the non-basic solutions are 𝑥1 = 0, 𝑥2 = 0 and 𝑥3 = 0. 𝑎11 𝑎12 𝑎13 1 1 1
𝐴= 𝑎 𝑎22 𝑎23 =
21 1 4 7
The basic variables are 𝑠1 and 𝑠2
∴ Initial Basic Feasible Solution (IBFS) are 𝑏1 3
𝑏= =
𝑏2 9
𝑠1 = 3 and 𝑠2 = 9

116
 SENSITIVITY ANALYSIS

Initial Iteration:

𝑪𝒋 𝟒 𝟔 𝟐 𝟎 𝟎 Min Ratio
𝑪𝑩 𝑩 𝑿𝑩 𝒙𝟏 𝒙𝟐 𝒙𝟑 𝒔𝟏 𝒔𝟐
𝑹𝟏 0 𝑠1 3 1 1 1 1 0 3/1=3

𝑹𝟐 0 𝑠2 9 1 [𝟒] 7 0 1 9/4=2.25

𝒁𝒋 0 0 0 0 0 0
𝒁𝒋 − 𝑪𝒋 −4 −6 −2 0 0

Since some 𝑍𝑗 − 𝐶𝑗 < 0. The current basic feasible solution is not optimal.
Negative minimum 𝑍𝑗 − 𝐶𝑗 is −6 and its column index is 2. So, the entering variable is 𝒙𝟐 .
Minimum ratio is 2.25 and its row index is 2. So, the leaving basic variable is 𝒔𝟐 .
∴ The pivot element is 4.
𝐸𝑛𝑡𝑒𝑟𝑖𝑛𝑔 = 𝑥2 , 𝐷𝑒𝑝𝑎𝑟𝑡𝑖𝑛𝑔 = 𝑠2 , 𝑃𝑖𝑣𝑜𝑡 𝐸𝑙𝑒𝑚𝑒𝑛𝑡 = 4
117
 SENSITIVITY ANALYSIS

𝑅2 𝑁𝑒𝑤 = 𝑅2 𝑂𝑙𝑑 ÷ 4

𝑅2 𝑁𝑒𝑤 = 9 1 4 7 0 1 ÷4

= 9/4 1/4 1 7/4 0 1/4

𝑅1 𝑁𝑒𝑤 = 𝑅1 𝑂𝑙𝑑 − 𝑅2 (𝑁𝑒𝑤)

= 3 1 1 1 1 0

−9/4 − 1/4 1 − 7/4 0 1/4

3/4 3/4 0 − 3/4 1 − 1/4

118
 SENSITIVITY ANALYSIS

First Iteration:

𝑪𝒋 𝟒 𝟔 𝟐 𝟎 𝟎 Min Ratio
𝑪𝑩 𝑩 𝑿𝑩 𝒙𝟏 𝒙𝟐 𝒙𝟑 𝒔𝟏 𝒔𝟐
𝑹𝟏 0 𝑠1 3/4 [𝟑/𝟒] 0 −3/4 1 −1/4 1

𝑹𝟐 6 𝑥2 9/4 1/4 1 7/4 0 1/4 9

𝒁𝒋 54/4 6/4 6 63/4 0 6/4

𝒁𝒋 − 𝑪𝒋 −10/4 0 55/4 0 6/4

Since one 𝑍𝑗 − 𝐶𝑗 < 0. The current basic feasible solution is not optimal.
Negative minimum 𝑍𝑗 − 𝐶𝑗 is −10/4 and its column index is 1. So, the entering variable is 𝒙𝟏 .
Minimum ratio is 1 and its row index is 1. So, the leaving basic variable is 𝒔𝟏 .
∴ The pivot element is 3/4.
𝐸𝑛𝑡𝑒𝑟𝑖𝑛𝑔 = 𝑥1 , 𝐷𝑒𝑝𝑎𝑟𝑡𝑖𝑛𝑔 = 𝑠1 , 𝑃𝑖𝑣𝑜𝑡 𝐸𝑙𝑒𝑚𝑒𝑛𝑡 = 3/4
119
 SENSITIVITY ANALYSIS

𝑅1 𝑁𝑒𝑤 = 𝑅1 𝑂𝑙𝑑 × 4/3

𝑅1 𝑁𝑒𝑤 = [3/4 3/4 0 − 3/4 1 − 1/4] × 4/3

= 1 1 0 −1 4/3 − 1/3

𝑅2 𝑁𝑒𝑤 = 𝑅2 𝑂𝑙𝑑 − (1/4)𝑅1 (𝑁𝑒𝑤)

= 9/4 1/4 1 7/4 0 1/4

−1/4 − 1/4 0 1/4 − 1/3 1/12

2 0 1 2 − 1/3 1/3

120
 SENSITIVITY ANALYSIS

𝑪𝒋 𝟒 𝟔 𝟐 𝟎 𝟎
𝑪𝑩 𝑩 𝑿𝑩 𝒙𝟏 𝒙𝟐 𝒙𝟑 𝒔𝟏 𝒔𝟐
𝑹𝟏 4 𝑥1 1 1 0 −1 𝟒/𝟑 −𝟏/𝟑

𝑹𝟐 6 𝑥2 2 0 1 2 −𝟏/𝟑 𝟏/𝟑

𝒁𝒋 16 4 6 8 10/3 2/3
𝒁𝒋 − 𝑪𝒋 0 0 6 10/3 2/3
Since all 𝑍𝑗 − 𝐶𝑗 ≥ 0. The current basic feasible
solution is optimal.
𝐶𝐵 = 4 6
The optimum solution of the given LPP is 𝑍 = 16 and

𝑋𝐵1 𝑥1 1
𝑋𝐵 = = 𝑎11 𝑎12 𝑎13 1 1 1
𝑋𝐵2 𝑥2 2 𝐴= 𝑎 =
21 𝑎22 𝑎23 1 4 7
𝟒/𝟑 −𝟏/𝟑
𝐵 −1 = 𝛽1 𝛽2 =
−𝟏/𝟑 𝟏/𝟑
121
 SENSITIVITY ANALYSIS

Range of 𝒂𝟏𝟏 :
Since 𝑎11 ∈ 𝐵, the basis
𝑟 = 1, 𝑘 = 1, 𝑖 = 1,2

4/3 16 6 10
𝐶𝐵 𝛽1 = 4 6 = 3 −3= 3
−1/3

−1/3 −4 6 2
𝐶𝐵 𝛽2 = 4 6 = + =
1/3 3 3 3

Range of 𝒂𝟏𝟏 for maintaining the feasibility condition:

The change of the coefficient 𝑎11 expressed as

𝒂∗𝟏𝟏 = 𝒂𝟏𝟏 + 𝜟𝒂𝟏𝟏

122
 SENSITIVITY ANALYSIS

The range of 𝜟𝒂𝟏𝟏 for maintaining the feasibility condition is given by

−𝑿𝑩𝒊 −𝑿𝑩𝒊
Max , 𝑸𝒊 > 𝟎 ≤ 𝜟𝒂𝒓𝒌 ≤ Min , 𝑸𝒊 <𝟎
𝑸𝒊 𝑸𝒊

Where 𝑸𝒊 = 𝑿𝑩𝒊 𝜷𝒌𝒌 − 𝜷𝒊𝒌 𝑿𝑩𝒌 𝒊 = 𝟏, 𝟐 𝒌=𝟏 𝒓=𝟏

For 𝒊 = 𝟏, 𝒌 = 𝟏 𝑸𝟏 = 𝑿𝑩𝟏 𝜷𝟏𝟏 − 𝜷𝟏𝟏 𝑿𝑩𝟏 = 𝟎

𝟒 𝟏 𝟖 𝟏 𝟗
For 𝒊 = 𝟐, 𝒌 = 𝟏 𝑸𝟐 = 𝑿𝑩𝟐 𝜷𝟏𝟏 − 𝜷𝟐𝟏 𝑿𝑩𝟏 = 𝟐 × 𝟑 − − 𝟑 × 𝟏 = 𝟑 + 𝟑 = 𝟑 = 𝟑

−𝑋𝐵1 −𝑋𝐵2 −𝑋𝐵1 −𝑋𝐵2

Max , , 𝑄1 , 𝑄2 > 0 ≤ 𝛥𝑎11 ≤ Min , , 𝑄1 , 𝑄2 < 0
𝑄1 𝑄2 𝑄1 𝑄2

−2
Max −, ≤ Δ𝑎11 ≤ Min −, −
3

𝟐
− 𝟑 ≤ 𝚫𝒂𝟏𝟏 < ∞

123
 SENSITIVITY ANALYSIS

Adding 𝒂𝟏𝟏 on all sides

𝟐
𝒂𝟏𝟏 − 𝟑 ≤ 𝒂𝟏𝟏 + 𝚫𝒂𝟏𝟏 < 𝒂𝟏𝟏 + ∞

𝑎11 𝑎12 𝑎13 1 1 1

Since 𝐴 = 𝑎 𝑎22 𝑎23 = also we know that 𝒂𝟏𝟏 ≤ 𝒂𝟏𝟏 + 𝜟𝒂𝟏𝟏
21 1 4 7

𝟐
𝟏 − 𝟑 ≤ 𝒂𝟏𝟏 < 𝟏 + ∞

𝟏
⟹ ≤ 𝒂𝟏𝟏 < ∞
𝟑

124
 SENSITIVITY ANALYSIS

Range of 𝒂𝟏𝟏 for maintaining the optimality condition:

The range of 𝜟𝒂𝟏𝟏 for maintaining the optimality condition is given by

−(𝒁𝒋 −𝑪𝒋 ) −(𝒁𝒋 −𝑪𝒋 )

Max , 𝑸𝒋 > 𝟎 ≤ 𝜟𝒂𝒓𝒌 ≤ Min , 𝑸𝒋 < 𝟎
𝑸𝒋 𝑸𝒋

Where 𝑸𝒋 = 𝒁𝒋 − 𝑪𝒋 𝜷𝒌𝒌 − 𝑿𝒌𝒋 𝑪𝑩 𝜷𝒌 𝒌=𝟏 𝒓=𝟏

𝒋 = 𝟑, 𝟒, 𝟓 (Which are not in the basis)

𝟒 −𝟏 ×𝟏𝟎 𝟏𝟎 𝟑𝟒
For j= 𝟑, 𝒌 = 𝟏 𝑸𝟑 = 𝒁𝟑 − 𝑪𝟑 𝜷𝟏𝟏 − 𝑿𝟏𝟑 𝑪𝑩 𝜷𝟏 = 𝟔 × 𝟑 − =𝟖+ =
𝟑 𝟑 𝟑

𝟏𝟎 𝟒 𝟒 𝟏𝟎 𝟒𝟎 𝟒𝟎
For j= 𝟒, 𝒌 = 𝟏 𝑸𝟒 = 𝒁𝟒 − 𝑪𝟒 𝜷𝟏𝟏 − 𝑿𝟏𝟒 𝑪𝑩 𝜷𝟏 = ×𝟑−𝟑× = − =𝟎
𝟑 𝟑 𝟗 𝟗

𝟐 𝟒 −𝟏 𝟏𝟎 𝟖 𝟏𝟎 𝟏𝟖
For j= 𝟓, 𝒌 = 𝟏 𝑸𝟓 = 𝒁𝟓 − 𝑪𝟓 𝜷𝟏𝟏 − 𝑿𝟏𝟓 𝑪𝑩 𝜷𝟏 = × − × = + = =𝟐
𝟑 𝟑 𝟑 𝟑 𝟗 𝟗 𝟗 125
 SENSITIVITY ANALYSIS

−(𝒁𝟑 −𝑪𝟑 ) −(𝒁𝟒 −𝑪𝟒 ) − 𝒁𝟓 −𝑪𝟓 −(𝒁𝟑 −𝑪𝟑 ) −(𝒁𝟒 −𝑪𝟒 ) − 𝒁𝟓 −𝑪𝟓
⟹ Max , 𝑸 , 𝑸 , 𝑸𝟑 , 𝑸𝟒 , 𝑸𝟓 > 𝟎 ≤ 𝜟𝒂𝟏𝟏 ≤ Min , 𝑸 , 𝑸 , 𝑸𝟑 , 𝑸𝟒 , 𝑸𝟓 < 𝟎
𝑸𝟑 𝟒 𝟓 𝑸𝟑 𝟒 𝟓

−6 −2/3
⟹ Max , −, ≤ 𝛥𝑎11 ≤ Min −, −, −
34/3 2

−18 −2
⟹ Max , −, ≤ 𝛥𝑎11 < ∞
34 6

−9 −1
⟹ Max , −, ≤ 𝛥𝑎11 < ∞
17 3

𝟗
⟹ − ≤ 𝜟𝒂𝟏𝟏 < ∞
𝟏𝟕

126
 SENSITIVITY ANALYSIS

Adding 𝒂𝟏𝟏 on all sides

𝟗
𝒂𝟏𝟏 − 𝟏𝟕 ≤ 𝒂𝟏𝟏 + 𝚫𝒂𝟏𝟏 < 𝒂𝟏𝟏 + ∞

𝑎11 𝑎12 𝑎13 1 1 1

Since 𝐴 = 𝑎 𝑎22 𝑎23 = also we know that 𝒂𝟏𝟏 ≤ 𝒂𝟏𝟏 + 𝜟𝒂𝟏𝟏
21 1 4 7

𝟗
𝟏 − 𝟏𝟕 ≤ 𝒂𝟏𝟏 < 𝟏 + ∞

𝟖
⟹ ≤ 𝒂𝟏𝟏 < ∞
𝟏𝟕

127
 SENSITIVITY ANALYSIS

Range of 𝒂𝟐𝟑 :
𝒂𝟐𝟑 ∉ 𝑩

The change of coefficient 𝑎23 is expressed as

𝒂∗𝟐𝟑 = 𝒂𝟐𝟑 + 𝚫𝒂𝟐𝟑

The range of Δ𝑎23 for maintaining the optimality is given by

−(𝒁𝒌 −𝑪𝒌 ) −(𝒁𝒌 −𝑪𝒌 )
Max , 𝑪𝑩 𝜷𝒌 > 𝟎 ≤ 𝜟𝒂𝒓𝒌 ≤ Min , 𝑪𝑩 𝜷𝒌 <𝟎
𝑪𝑩 𝜷𝒌 𝑪𝑩 𝜷𝒌

𝟐
𝑪𝑩 𝜷𝟐 = , 𝒁𝟑 − 𝑪𝟑 = 𝟔 𝒓 = 𝟐, 𝒌 = 𝟑
𝟑

−𝟔
⟹ ≤ 𝜟𝒂𝟐𝟑 < ∞
𝟐/𝟑

128
 SENSITIVITY ANALYSIS

−𝟏𝟖
⟹ ≤ 𝜟𝒂𝟐𝟑 < ∞
𝟐

⟹ −𝟗 ≤ 𝜟𝒂𝟐𝟑 < ∞

Adding 𝒂𝟐𝟑 on all sides

𝒂𝟐𝟑 − 𝟗 ≤ 𝒂𝟐𝟑 + 𝚫𝒂𝟐𝟑 < 𝒂𝟐𝟑 + ∞

𝑎11 𝑎12 𝑎13 1 1 1

Since 𝐴 = 𝑎 𝑎22 𝑎23 = also we know that 𝒂𝟐𝟑 ≤ 𝒂𝟐𝟑 + 𝜟𝒂𝟐𝟑
21 1 4 7

𝟕 − 𝟗 ≤ 𝒂𝟐𝟑 < 𝟕 + ∞

⟹ −𝟐 ≤ 𝒂𝟐𝟑 < ∞

129
 SENSITIVITY ANALYSIS

4. Addition of a new variable

Consider the LP problem be
Max 𝒁 = 𝑪𝑿
Subject to
𝑨𝑿 = 𝒃
and 𝑿 ≥ 𝟎
where
𝑨 is an 𝒎 × 𝒏 matrix,
𝑪 = 𝑪𝟏 𝑪𝟐 𝑪𝟑 … 𝑪𝒏 is a cost vector,

𝒙𝟏
𝒙𝟐
𝑿= ⋮
𝒙𝒏

130
 SENSITIVITY ANALYSIS

• Let a non-negative variable 𝒙𝒏+𝟏 with the cost coefficient 𝑪𝒏+𝟏 be added to the LPP after the optimum
solution is obtained.

• Then there will be an extra column 𝑨𝒏+𝟏 corresponding to the coefficients of new variable 𝒙𝒏+𝟏 .

• To study the sensitivity analysis, it requires the following additional computations.

𝒙𝒏+𝟏 = 𝑩−𝟏 𝑨𝒏+𝟏

𝒁𝒏+𝟏 − 𝑪𝒏+𝟏 = 𝑪𝑩 𝒙𝒏+𝟏 − 𝑪𝒏+𝟏

Now there are two possibilities as

131
 SENSITIVITY ANALYSIS

• Case 1:

If 𝒁𝒏+𝟏 − 𝑪𝒏+𝟏 ≥ 𝟎, then the current optimum solution remains optimum.

• Case 2:

• If 𝒁𝒏+𝟏 − 𝑪𝒏+𝟏 < 0, then the current optimum solution can be improved by introducing 𝒙𝒏+𝟏 into the
basis.

• The usual simplex procedure can be applied to obtain the new optimum solution.

132
 SENSITIVITY ANALYSIS
Example 4:
Consider the LPP
Max 𝒁 = 𝟒𝒙𝟏 + 𝟔𝒙𝟐 + 𝟐𝒙𝟑
Subject to
𝒙𝟏 + 𝒙𝟐 + 𝒙𝟑 ≤ 𝟑
𝒙𝟏 + 𝟒𝒙𝟐 + 𝟕𝒙𝟑 ≤ 𝟗
And 𝒙𝟏 , 𝒙𝟐 , 𝒙 𝟑 ≥ 𝟎

𝟏
a) If a new variable 𝒙𝟔 ≥ 𝟎 is introduced with cost 3 and 𝑨𝟔 = , then discuss the effect of the
𝟐
addition of new variable and obtain the revised solution if any.

−𝟐
b) If a new variable 𝒙𝟔 ≥ 𝟎 is introduced with cost 9 and 𝑨𝟔 = , then discuss the effect of the
𝟒
addition of new variable and obtain the revised solution if any.
Solution:
Similar step of Example 1 solution
133
 SENSITIVITY ANALYSIS

The given problem is converted to standard form by adding

⟹ Max 𝑍 = 𝐶𝑋
non-negative slack variables 𝑠1 and 𝑠2 .
Subject to
The standard form of the LP problem is
𝐴𝑋 = 𝑏
Maz 𝑍 = 4𝑥1 + 6𝑥2 + 2𝑥3 + 0𝑠1 + 0𝑠2
𝑋>0
Subject to 𝑥1 + 𝑥2 + 𝑥3 + 𝑠1 + 0𝑠2 = 3
Where
𝑥1 + 4𝑥2 + 7𝑥3 + 0𝑠1 + 𝑠2 = 9
𝐶 = 𝐶𝑗 = 𝐶1 , 𝐶2 , 𝐶3 , 𝐶4 , 𝐶5 = 4,6,2,0,0
And 𝑥1 , 𝑥2 , 𝑥3 , 𝑠1 , 𝑠2 ≥ 0

𝑎11 𝑎12 𝑎13 1 1 1

The non-basic variables are 𝑥1 , 𝑥2 and 𝑥3 . 𝐴= 𝑎
21 𝑎22 𝑎23 =
1 4 7
Hence the non-basic solutions are 𝑥1 = 0, 𝑥2 = 0 and 𝑥3 = 0.

𝑏1 3
𝑏= =
The basic variables are 𝑠1 and 𝑠2 𝑏2 9
∴ Initial Basic Feasible Solution (IBFS) are
𝑠1 = 3 and 𝑠2 = 9
134
 SENSITIVITY ANALYSIS

Initial Iteration:

𝑪𝒋 𝟒 𝟔 𝟐 𝟎 𝟎 Min Ratio
𝑪𝑩 𝑩 𝑿𝑩 𝒙𝟏 𝒙𝟐 𝒙𝟑 𝒔𝟏 𝒔𝟐
𝑹𝟏 0 𝑠1 3 1 1 1 1 0 3/1=3

𝑹𝟐 0 𝑠2 9 1 [𝟒] 7 0 1 9/4=2.25

𝒁𝒋 0 0 0 0 0 0
𝒁𝒋 − 𝑪𝒋 −4 −6 −2 0 0

Since some 𝑍𝑗 − 𝐶𝑗 < 0. The current basic feasible solution is not optimal.
Negative minimum 𝑍𝑗 − 𝐶𝑗 is −6 and its column index is 2. So, the entering variable is 𝒙𝟐 .
Minimum ratio is 2.25 and its row index is 2. So, the leaving basic variable is 𝒔𝟐 .
∴ The pivot element is 4.
𝐸𝑛𝑡𝑒𝑟𝑖𝑛𝑔 = 𝑥2 , 𝐷𝑒𝑝𝑎𝑟𝑡𝑖𝑛𝑔 = 𝑠2 , 𝑃𝑖𝑣𝑜𝑡 𝐸𝑙𝑒𝑚𝑒𝑛𝑡 = 4
135
 SENSITIVITY ANALYSIS

𝑅2 𝑁𝑒𝑤 = 𝑅2 𝑂𝑙𝑑 ÷ 4

𝑅2 𝑁𝑒𝑤 = 9 1 4 7 0 1 ÷4

= 9/4 1/4 1 7/4 0 1/4

𝑅1 𝑁𝑒𝑤 = 𝑅1 𝑂𝑙𝑑 − 𝑅2 (𝑁𝑒𝑤)

= 3 1 1 1 1 0

−9/4 − 1/4 1 − 7/4 0 1/4

3/4 3/4 0 − 3/4 1 − 1/4

136
 SENSITIVITY ANALYSIS

First Iteration:

𝑪𝒋 𝟒 𝟔 𝟐 𝟎 𝟎 Min Ratio
𝑪𝑩 𝑩 𝑿𝑩 𝒙𝟏 𝒙𝟐 𝒙𝟑 𝒔𝟏 𝒔𝟐
𝑹𝟏 0 𝑠1 3/4 [𝟑/𝟒] 0 −3/4 1 −1/4 1

𝑹𝟐 6 𝑥2 9/4 1/4 1 7/4 0 1/4 9

𝒁𝒋 54/4 6/4 6 63/4 0 6/4

𝒁𝒋 − 𝑪𝒋 −10/4 0 55/4 0 6/4

Since one 𝑍𝑗 − 𝐶𝑗 < 0. The current basic feasible solution is not optimal.
Negative minimum 𝑍𝑗 − 𝐶𝑗 is −10/4 and its column index is 1. So, the entering variable is 𝒙𝟏 .
Minimum ratio is 1 and its row index is 1. So, the leaving basic variable is 𝒔𝟏 .
∴ The pivot element is 3/4.
𝐸𝑛𝑡𝑒𝑟𝑖𝑛𝑔 = 𝑥1 , 𝐷𝑒𝑝𝑎𝑟𝑡𝑖𝑛𝑔 = 𝑠1 , 𝑃𝑖𝑣𝑜𝑡 𝐸𝑙𝑒𝑚𝑒𝑛𝑡 = 3/4
137
 SENSITIVITY ANALYSIS

𝑅1 𝑁𝑒𝑤 = 𝑅1 𝑂𝑙𝑑 × 4/3

𝑅1 𝑁𝑒𝑤 = [3/4 3/4 0 − 3/4 1 − 1/4] × 4/3

= 1 1 0 −1 4/3 − 1/3

𝑅2 𝑁𝑒𝑤 = 𝑅2 𝑂𝑙𝑑 − (1/4)𝑅1 (𝑁𝑒𝑤)

= 9/4 1/4 1 7/4 0 1/4

−1/4 − 1/4 0 1/4 − 1/3 1/12

2 0 1 2 − 1/3 1/3

138
 SENSITIVITY ANALYSIS

𝑪𝒋 𝟒 𝟔 𝟐 𝟎 𝟎
𝑪𝑩 𝑩 𝑿𝑩 𝒙𝟏 𝒙𝟐 𝒙𝟑 𝒔𝟏 𝒔𝟐
𝑹𝟏 4 𝑥1 𝟏 1 0 −1 𝟒/𝟑 −𝟏/𝟑

𝑹𝟐 6 𝑥2 𝟐 0 1 2 −𝟏/𝟑 𝟏/𝟑

𝒁𝒋 16 4 6 8 10/3 2/3
𝒁𝒋 − 𝑪𝒋 0 0 6 10/3 2/3
Since all 𝑍𝑗 − 𝐶𝑗 ≥ 0. The current basic feasible
solution is optimal.
1 1
The optimum solution of the given LPP is 𝑍 = 16 and 𝐵 = 𝑥1 𝑥2 =
1 4
𝑋𝐵1 𝑥1 𝟏
𝑋𝐵 = =
𝑋𝐵2 𝑥2 𝟐 𝐶𝐵 = [4 6]
𝟒/𝟑 −𝟏/𝟑
𝐵 −1 = 𝛽1 𝛽2 =
−𝟏/𝟑 𝟏/𝟑
139
 SENSITIVITY ANALYSIS

1
a) New variable 𝑥6 is introduced with 𝐴6 = , cost 𝐶6 = 3, then first we calculate
2

4 2
4/3 −1/3 1 2/3 −3
3
𝒙𝟔 = 𝑩−𝟏 𝑨𝟔 = = 1 2 =
−1/3 1/3 2 −3+ 3 1/3

2
8 6 14 5
𝒁𝟔 − 𝑪𝟔 = 𝐶𝐵 𝑥6 − 𝐶6 = 4 6 3 − 3 = + − 3 = −3= >0
1 3 3 3 3
3

Hence 𝒁𝟔 − 𝑪𝟔 ≥ 𝟎, the current optimum solution remains optimum.

140
 SENSITIVITY ANALYSIS

−2
b) New variable 𝑥6 is introduced with 𝐴6 = , cost 𝐶6 = 9, then first we calculate
4

8 4
4/3 −1/3 −2 −3 − 3 −12/3 −4
𝒙𝟔 = 𝑩−𝟏 𝑨𝟔 = = 2 4 = =
−1/3 1/3 4 +3 6/3 2
3

−4
𝒁𝟔 − 𝑪𝟔 = 𝐶𝐵 𝑥6 − 𝐶6 = 4 6 − 9 = −16 + 12 − 9 = −13 < 0
2

Hence 𝒁𝟔 − 𝑪𝟔 < 𝟎, then adding 𝑥6 into the basis and calculate the optimum solution by usual
simplex table.

141
 SENSITIVITY ANALYSIS

New Simplex table (1):

𝑪𝒋 𝟒 𝟔 𝟐 𝟎 𝟎 𝟗 Min Ratio
𝑪𝑩 𝑩 𝑿𝑩 𝒙𝟏 𝒙𝟐 𝒙𝟑 𝒔𝟏 𝒔𝟐 𝒙𝟔
𝑹𝟏 4 𝑥1 1 1 0 −1 4/3 −1/3 −4 −

𝑹𝟐 6 𝑥2 2 0 1 2 −1/3 1/3 [𝟐] 2

=1
2

𝒁𝒋 16 4 6 8 10/3 2/3 −4
𝒁𝒋 − 𝑪𝒋 0 0 6 10/3 2/3 −9

Since one 𝑍𝑗 − 𝐶𝑗 < 0. The current basic feasible solution is not optimal.
Negative minimum 𝑍𝑗 − 𝐶𝑗 is −9 and its column index is 6. So, the entering variable is 𝒙𝟔 .
Minimum ratio is 1 and its row index is 2. So, the leaving basic variable is 𝒙𝟐 .
∴ The pivot element is 2.
𝐸𝑛𝑡𝑒𝑟𝑖𝑛𝑔 = 𝑥6 , 𝐷𝑒𝑝𝑎𝑟𝑡𝑖𝑛𝑔 = 𝑥2 , 𝑃𝑖𝑣𝑜𝑡 𝐸𝑙𝑒𝑚𝑒𝑛𝑡 = 2
142
 SENSITIVITY ANALYSIS

𝑅2 𝑁𝑒𝑤 = 𝑅2 𝑂𝑙𝑑 ÷ 2

𝑅2 𝑁𝑒𝑤 = (2 0 1 2 − 1/3 1/3 2) ÷ 2

=1 0 1/2 1 − 1/6 1/6 1

𝑅1 𝑁𝑒𝑤 = 𝑅1 𝑂𝑙𝑑 + 4𝑅2 (𝑁𝑒𝑤)

= 1 1 0 −1 4/3 − 1/3 −4

4 0 2 4 − 2/3 2/3 4

5 1 2 3 2/3 1/3 0

143
 SENSITIVITY ANALYSIS

New Simplex table (2):

𝑪𝒋 𝟒 𝟔 𝟐 𝟎 𝟎 𝟗
𝑪𝑩 𝑩 𝑿𝑩 𝒙𝟏 𝒙𝟐 𝒙𝟑 𝒔𝟏 𝒔𝟐 𝒙𝟔
𝑹𝟏 4 𝑥1 5 1 2 3 2/3 1/3 0

𝑹𝟐 9 𝑥6 1 0 1/2 1 −1/6 1/6 1

𝒁𝒋 𝟐𝟗 4 25/2 21 7/6 17/6 9

𝒁𝒋 − 𝑪𝒋 0 13/2 19 7/6 17/6 0

Since all 𝑍𝑗 − 𝐶𝑗 ≥ 0. The current basic feasible solution is optimal.

Hence, optimal solution at 𝑥1 = 5, 𝑥2 = 0, 𝑥3 = 0, x6 = 9 is
Max 𝑍 = 4𝑥1 + 6𝑥2 + 2𝑥3 + 9𝑥6
=4 5 +9 1
= 𝟐𝟗
144
 SENSITIVITY ANALYSIS

5. Addition of new constraint

• Consider the LP problem be
Max 𝒁 = 𝑪𝑿
Subject to
𝑨𝑿 = 𝒃
and 𝑿 ≥ 𝟎

• Let 𝑿𝑩 be the current optimum basic feasible solution.

• If any new constraint is added to the LPP, then the following two cases may be raised.

• Case 1:
If the new constraint is satisfied by the current optimum solution 𝑿𝑩 , then it has no effect on
the current optimum solution. i.e., 𝑿𝑩 remains same.
145
 SENSITIVITY ANALYSIS

• Case 2:
• If the new constraint is satisfied by the current optimum solution 𝑿𝑩 , then the current
optimum solution becomes infeasible.

• Apply dual simplex method to find an optimum solution to the new LPP.

146
 SENSITIVITY ANALYSIS
Example 5:
Consider the LPP
Max 𝒁 = 𝟑𝒙𝟏 + 𝟐𝒙𝟐
Subject to
𝟐𝒙𝟏 + 𝒙𝟐 ≤ 𝟒𝟎
𝒙𝟏 + 𝒙𝟐 ≤ 𝟐𝟒
𝟐𝒙𝟏 + 𝟑𝒙𝟐 ≤ 𝟔𝟎
And 𝒙𝟏 , 𝒙𝟐 ≥ 𝟎
a) If a constraint 𝒙𝟏 ≤ 𝟏𝟓 is added to the LPP, then discuss the post-optimality analysis of the problem.
b) If a constraint 𝒙𝟏 ≤ 𝟐𝟎 is added to the LPP, then discuss the post-optimality analysis of the problem.
Solution:
The given problem is converted to standard form by adding non-negative slack variables 𝑠1 and 𝑠2 .
The standard form of the LP problem is

147
 SENSITIVITY ANALYSIS

Maz 𝑍 = 3𝑥1 + 2𝑥2 + 0𝑠1 + 0𝑠2 + 0𝑠3

Subject to 2𝑥1 + 𝑥2 + 𝑠1 = 40
𝑥1 + 𝑥2 + 𝑠2 = 24
2𝑥1 + 3𝑥2 + 𝑠3 = 60
And 𝑥1 , 𝑥2 , 𝑥3 , 𝑠1 , 𝑠2 , 𝑠3 ≥ 0

The non-basic variables are 𝑥1 , 𝑥2 .

Hence the non-basic solutions are 𝑥1 = 0, and 𝑥2 = 0.

The basic variables are 𝑠1 , 𝑠2 and 𝑠3

∴ Initial Basic Feasible Solution (IBFS) are
𝑠1 = 40, 𝑠2 = 24 and 𝑠3 = 60

148
 SENSITIVITY ANALYSIS

The optimum simplex table is

𝑪𝒋 𝟑 𝟐 𝟎 𝟎 𝟎
𝑪𝑩 𝑩 𝑿𝑩 𝒙𝟏 𝒙𝟐 𝒔𝟏 𝒔𝟐 𝒔𝟑
𝑹𝟏 3 𝑥1 16 1 0 1 −1 0

𝑹𝟐 2 𝑥2 8 0 1 −1 2 0

𝑹𝟑 0 𝑠3 4 0 0 1 −4 1

𝒁𝒋 64 3 2 1 1 0
𝒁𝒋 − 𝑪𝒋 0 0 1 1 1
Since all 𝑍𝑗 − 𝐶𝑗 ≥ 0. The current basic feasible solution is
optimal.
The optimum solution of the given LPP is
Max 𝑍 = 64 𝑥1 = 16, 𝑥2 = 8
149
 SENSITIVITY ANALYSIS

a) The constraint 𝒙𝟏 ≤ 𝟏𝟓 is added to the given LPP.

• The optimum solutions are 𝒙𝟏 = 𝟏𝟔, 𝒙𝟐 = 𝟖

∴ The new constraint is not satisfied the optimum solution.

• Hence the optimum solution is infeasible.

• For the constraint 𝒙𝟏 ≤ 𝟏𝟓, add slack variable 𝒔𝟒 ≥ 𝟎 we get

𝑥1 + 𝑠4 = 15

• The optimum solution of the LPP is now changed for the inclusion of the new constraint.
150
 SENSITIVITY ANALYSIS

The modified optimum simplex table with these changes is expressed below:

𝑪𝒋 𝟑 𝟐 𝟎 𝟎 𝟎 𝟎
𝑪𝑩 𝑩 𝑿𝑩 𝒙𝟏 𝒙𝟐 𝒔𝟏 𝒔𝟐 𝒔𝟑 𝒔𝟒
𝑹𝟏 3 𝑥1 16 1 0 1 −1 0 0

𝑹𝟐 2 𝑥2 8 0 1 −1 2 0 0

𝑹𝟑 0 𝑠3 4 0 0 1 −4 1 0

𝑹𝟒 0 𝑠4 15 1 0 0 0 0 1
𝒁𝒋 64 3 2 1 1 0 0
𝒁𝒋 − 𝑪𝒋 0 0 1 1 0 0

151
 SENSITIVITY ANALYSIS

Since the vector 𝒙𝟏 is in the basis, the corresponding coefficient in the additional constraint must be
converted to zero by using relevant row operations. Then using Dual Simplex Method

𝑅4 = 𝑅4 − 𝑅1

= 15 1 0 0 0 0 1
−16 −1 0 −1 1 0 0
−1 0 0 −1 1 0 1

152
 SENSITIVITY ANALYSIS

𝑪𝒋 𝟑 𝟐 𝟎 𝟎 𝟎 𝟎
𝑪𝑩 𝑩 𝑿𝑩 𝒙𝟏 𝒙𝟐 𝒔𝟏 𝒔𝟐 𝒔𝟑 𝒔𝟒
𝑹𝟏 3 𝑥1 16 1 0 1 −1 0 0

𝑹𝟐 2 𝑥2 8 0 1 −1 2 0 0

𝑹𝟑 0 𝑠3 4 0 0 1 −4 1 0

𝑹𝟒 0 𝑠4 −1 0 0 [−𝟏] 1 0 1
𝒁𝒋 64 3 2 1 1 0 0
𝒁𝒋 − 𝑪𝒋 0 0 1 1 0 0

Since all 𝒁𝒋 − 𝑪𝒋 ≥ 𝟎 and one 𝑿𝑩𝒊 < 𝟎. The solution is not

To find the entering variable:
optimal. (𝒁𝒋 − 𝑪𝒋 ) 1
𝑴𝒂𝒙 , 𝒂𝒊𝒌 < 𝟎 = 𝑀𝑎𝑥 −, − , , −, −, −
𝒂𝒊𝒌 −1

𝑿𝑩𝟑 = −𝟖 is the most negative, the basic variable 𝒔𝟑 = 𝑀𝑎𝑥 −, −, −1, −, − , −

leaves the basis. = −1

153
 SENSITIVITY ANALYSIS

𝑅4 𝑁𝑒𝑤 = 𝑅4 𝑂𝑙𝑑 × −1 𝑅2 𝑁𝑒𝑤 = 𝑅2 𝑂𝑙𝑑 + 𝑅4 (𝑁𝑒𝑤)

𝑅4 𝑁𝑒𝑤 = −1 0 0 −1 1 0 1 × −1 = 8 0 1 −1 2 0 0
= 1 0 0 1 −1 0 −1 1 0 0 1 −1 0 −1

9 0 1 0 1 0 −1
𝑅1 𝑁𝑒𝑤 = 𝑅1 𝑂𝑙𝑑 − 𝑅4 (𝑁𝑒𝑤)

= 16 1 0 1 −1 0 0 𝑅3 𝑁𝑒𝑤 = 𝑅3 𝑂𝑙𝑑 − 𝑅4 (𝑁𝑒𝑤)

−1 0 0 −1 1 0 1 = 4 0 0 1 −4 1 0
15 1 0 0 0 0 1 −1 0 0 −1 1 0 1
3 0 0 0 −3 1 1

154
 SENSITIVITY ANALYSIS

𝑪𝒋 𝟑 𝟐 𝟎 𝟎 𝟎 𝟎
𝑪𝑩 𝑩 𝑿𝑩 𝒙𝟏 𝒙𝟐 𝒔𝟏 𝒔𝟐 𝒔𝟑 𝒔𝟒
𝑹𝟏 3 𝑥1 15 1 0 0 0 0 1

𝑹𝟐 2 𝑥2 9 0 1 0 1 0 −1

𝑹𝟑 0 𝑠3 3 0 0 0 −3 1 1

𝑹𝟒 0 𝑠4 1 0 0 1 −1 0 −1
𝒁𝒋 𝟔𝟑 3 2 0 2 0 1
𝒁𝒋 − 𝑪𝒋 0 0 0 2 0 1

Since all 𝒁𝒋 − 𝑪𝒋 ≥ 𝟎 and all 𝑿𝑩𝒊 ≥ 𝟎. The solution is optimal. Hence the new optimal solution is,
Max 𝑍 = 𝟑𝒙𝟏 + 𝟐𝒙𝟐
= 3 15 + 2(9)
= 45 + 18
= 𝟔𝟑
155
 SENSITIVITY ANALYSIS

b) The constraint 𝒙𝟏 ≤ 𝟐𝟎 is added to the given LPP.

• The optimum solutions are 𝒙𝟏 = 𝟏𝟔, 𝒙𝟐 = 𝟖

∴ The new constraint is satisfied the optimum solution.

• Hence the optimum solution remains same.

i.e., the new constraint not affects the optimum solution.

156