Sensitivity Analysis

The LP model assumes that all parameters are known and constant. That
assumption is made for modeling purposes but is not always true. In fact,
the parameters of the LP model are frequently difficult to estimate and may
vary with time. It is therefore necessary to perform sensitivity analysis to
study the effects of changes of the parameters on the optimum solution. It
can be seen that the final simplex tableau contains information that is useful
for the sensitivity analysis. That is, from the final simplex tableau we have
the following relations:

Consider the standard LP problem:

Maximize Z = cjxj
subject to C>, aijxj 5 b;, for i = 1,2, . . ,m
xj 0
In this section, we will consider the following cases:
Then from equations 1-5 we can write

f o r k = 1 , 2 , - . - , mand j = 1 , 2 , . . - , n .

0.1 General.Procedure for Sensitivity Analysis

Procedure
Step 1 - Calculate Ay;, Ab;, A(z; - cj) and Aa;,. Add them (if nonzero)
to the corresponding entries in the final simplex tableau.
Step 2 - Convert the modified tableau to the proper form (if necessary)
to identify or evaluate the current basic solution. Perform iterations by the
simplex procedure (if necessary).
Step 3 - Test the feasibility of the current solution.
Step 4 - Test the optimality of the current solution (if feasible).
Step 5 - If the solution fails either test, perform iterations with the current
tableau as the initial one.

0.2 CASE 1 - Changes in the bi

Procedure
Step l - Calculate A yg (Eq. 6) and Ab; (Eq. 7) for k = 1,2, . . . ,m
Step 2 - (not applicable)
Step 3 - Test the solution for feasibility (bf 2 O?).
Step 4 - Test the solution for optimality (in this case, if the solution is
feasible it is optimal).
Step 5 - Perform iterations (if necessary).
 Example 0.2.1 Solve the following problem by the simplex method.
b 18
→ bi : 14
Maximize Z = 3x1 +5x2

3xtzxzi-4o-b.bz
=

ibisbibi-bz.tl 418 X1 L 4
3x1 +2x2 5 18 .bz →
Xl,X2 2 0
Suppose that the second constraint is replaced by 3x1+2x2 5 14. Use the sen-
sitivity analysis procedure to determine whether the original optimal solution
is still optimal.

Solution:

Cj 3 5 0 0
X; C; A' A2 A3 A4 b; a
xi,

53 0 1 0 1 0 4 *
5 4 0 3 2 0 1 18 9
-4

囌
?

y 。

。
We have Ab2 = 14 - 18 = -4. Then

暴警
"… "

o o

Hence
+
1
(yg)' = y,' Ay: = 35
~
(b;)' = b; + A b ; = 4
(b;)' = b ; + A b ; = 7
Because all the (h:)' are nonnegative, the new solution is feasible and optimal.
That is (X;)' = 7, (X;)' = 4 and ty,*)'= 35.
0.3 CASE 2a - Changes in the coefficients of a non-
basic variables
Assume that X; is a nonbasic variable in the optimal solution. The changes
of the coefficients cj, alj, azj, . . ,amj of this variable will be considered in
this case.
Procedure
Step 1- Calculate A($-cj) (Eq. 8) and Aaij (Eq. 9) for k = 1,2,. ,m. + .

Add the resulting changes to the corresponding entries in the simplex table.
Step 2 - (not applicable)
Step 3 - (not applicable) I

Step 4 - Test the solution for optimality (cj - zj' 5 O?) l

Step 5 - Perform iterations (if necessary).

-_-
Example 0.3.1 Solve the L P problem in Example 0.2.1 by substituting the

⼀
+ +
constraint 3x1 2x2 5 l8 by the new constraint xl 2x2 F 18.
。
Solution:
_
From the optimal solution obtained in Example 0.2.1 it is noted that xl
is a nonbasic variable. Then

1
*

Hence
( c-2 ) = 112
= 1
(a;,)' = 112
 The new simplex tables appear below.

Newti
Shtn

0.4 CASE 2b - Introduction of a new variable

Procedure
In this case we consider the new variable xj as it has been in the original
model with all its coefficients equal to zero, and it was a nonbasic variable in
the current basic feasible solution. Therefore, if we change these zero coeffi-
cients to their actual values for the new variable, the computation procedure
for this case becomes identical to that for Case 2a.

Example 0.4.1 Using the optimal solution obtained in Example 0.2.1 solve
the following LP problem.

Maximize Z = 3x1 +5x2 $725

51 +x5 1 4
3x1 +2x2 + 2 ~ 5 < l8
2 1 2 0 X220 3 5 2 0
 Solution:
The variable x5 can be considered as the new variable of the problem in
Example 0.2.1. The final simplex tableau becomes:

We have:

Hence

Substitute the new values of (c5- z;)', (a;,)',(a;,)' into the simplex
tableau:
0.5 CASE 3 - Changes in the coefficients of a basic
variable
Procedure
All the five steps of the general procedure are required for case 3 (we
use step 2 in this case because the current basic solution may not have a
coefficient of 1 in its row of the simplex tableau and a coefficient of zero
in every other row. Therefore, it may be necessary to restore the current
simplex tableau t o its proper form. This could make the current solution
either infeasible or nonoptimal).

Example 0.5.1 Solve the following LP problem using the optimal solution
obtained in Example 0.2.1.

Maximize Z = 3xl -t($2

X1 < 4
3x1 +&c2 5 l8
x 1 , ~ 2 > 0

Solution:
Following the optimal solution obtained in Example 0.2.1, X;! is a basic
variable. We have

Hence
(c2- .G)' = -7
(a;,)' = 0
= 2
 X.basicwriable-n nf80torml.io
The new simplex tableau is

lnp-fonrik-Becausethastr.in
0.6 CASE 4 - Introduction of a new constraint
Procedure
Step l - Introduce the new constraint into the final simplex tableau with
the slack (or artificial) variable as the basic variable for the new row.
Step 2 - Convert the modified tableau to the proper form (the coefficients
of other basic variables in the new row must be zero).
Step 3 - Test the feasibility of the current solution.
Step 4 - Test the optimality of the current solution (if feasible).
Step 5 - If the solution fails either test, perform iterations (using primal
or dual simplex techniques).

Example 0.6.1 Solve the following LP problem using the optimal solution
obtained in Example 0.2.1.

wasoohhdiruitidy !
xztxs-6-Becausenewonsideet.vn
Maximize Z = 3x1 +5x2

(new constraint)
X1 5 4
3x1 +2x2 5 18
22 5 6
X1122 L 0

hasaisensincethefmdatinAULPm.de
 N_n
fonn
眺-
{
-

c-n ewonswitwmrf-g8ofo-noteasibuPudf.no
8
8

3 - zj
C' -912 0 0 -512 0 Y,"
Ratio 3 * * 5 *

Metbd