Sensitivity analysis Methodology
When you use a mathematical model to describe reality you must make
approximations. The world is more complicated than the kinds of
optimization problems that we are able to solve. Linearity assumptions usually
are significant approximations.
Another important approximation comes because you cannot be sure of the
data that you put into the model. Your knowledge of the relevant technology
may be imprecise, forcing you to approximate values in A, b, or c. Moreover,
information may change. Sensitivity analysis is a systematic study of how
sensitive solutions are to (small) changes in the data. The basic idea is to be
able to give answers to questions of the form:
1. If the objective function changes, how does the solution change?
2. If resources available change, how does the solution change?
3. If a constraint is added to the problem, how does the solution change?
Sensitivity analysis (also called post-optimality analysis) is defined as study
of the effect of the optimal solution with respect to changes in the input
parameters of the original optimization problem.
Linear programing is a technique that aims at optimizing performance
regarding combinations of resources. LP can offer managers the capability of
building scenarios through its extensive “ what if ” analysis and sensitivity
analysis facilities. While most practical L.P. problems would require a very
long time to solve manually, computer software can be utilized to arrive at a
1
solution in a very short period. When we are dealing with sensitivity analysis,
we are initially looking into changes might happen to the parameters of L.P.
model. These possible changes would imply to investigate the changes in RHS
of the model constraints and the coefficient of the objective function .
Sensitivity analysis involves change in the following areas:
1. Changes in the R.H.S of the Constraints
2. Changes in coefficients of the Objective Function
3. Changes in coefficients of variables in constraints
4. Addition of anew variable
5. Addition of anew constraints
The change is made in several values according to the rules of sensitivity
analysis:
1. Change in Right Hand Side
Max Z=30x1 +50x2
Subject to : 2X1+X2 =< 16
X1+2 x2 =< 11
X1+3X2 =< 15
X1, X2 >= 0
B.V X1 X2 S1 S2 S3 R.H.S
Z -30 -50 0 0 0 0
S1 2 1 1 0 0 16
S2 1 2 0 1 0 11
S3 1 3 0 0 1 15
What if ……. these values are changed?
2
Does it affect the value of the Optimal Solution?
Optimal Solution
B.V X1 X2 S1 S2 S3 R.H.S
Z 0 0 10/3 70/3 0 310
S3 0 0 1/3 -5/3 1 2
X1 1 0 2/3 -1/3 0 7
X2 0 1 -1/3 2/3 0 2
The results of the last solution were as follows:
If we assume that the right side has changed as follows:
16 20
11 to
11
15 15
The first supplier has increased from 16 to 20
From the last table we take the matrix under the Slack Variables and multiply
this matrix by the new vertical vector to get:
XB=B-1b ≥ 0
XB = represent the basic variables in the table the optimal solution
B-1 = matrix under the Basic variables in the optimal solution except for the
objective function row
B = represent the values of R.H.S
3
S3 1/3 -5/3 1 20 20/3 -55/3 +15 10/3
X1 2/3 -1/3 0 11 40/3 -11/3 +0 29/3
= = =
X2 -1/3 2/3 0 15 -20/3 22/3 + 0 2/3
≥0
Objective function:
Max Z=30X1 +50X2 = 30*29/3 + 50*2/3= 290+ 33.33 =323.33
The value of the objective function increased and all the coefficients of the
right side of the equations were positive.
Work examples : 16 27
What if there is a change in the right hand side as ? 11 to 11
15 15
16 25
What if there is a change in the right hand side as 11 to 15
15 18
4
2.Changes in coefficients of the Objective Function
Changes in the parameters of decision variables directly affect the elements of
the solution matrix that lead to complexity in the calculations and can indicate
the right side of the corresponding problem constraints.
Max Z= 5x1+2x2+3x3
s.t x1 + 5x2 + 2x3 30
x1 - 5x2 - 5x3 40
x1,x2,x3 0
If we assume that the objective function in example has changed :
From Max Z= 5x1+2x2+3x3 To Max Z= 3x1+2x2+3x3
So the problem of linear programming will be as follows:
Max Z= 3x1+2x2+3x3
S.T. x1+5x2+2x3 30
x1-5x2-5x3 40
x1,x2,x3≥0
B.V X1 X2 X3 S1 S2 R.H.S
Z -3 -1 -3 0 0 0
S1 1 5 2 1 0 30
S2 1 -5 -5 0 1 40
5
Table Optimal solution after change
X1 X2 X3 S1 S2 R.H.S
B.V
Z 0 14 3 3 0 90
X1 1 5 2 1 0 30
S2 0 0 -7 -1 1 10
Here we need the Dual model to the current problem
The Dual form of the above problem is:
Min Z = 30y1 + 40y2
s.t y1 + y2 3
5y1 - 5y2 1
2y1 - 5y2 3
y1, y2 0
For the purpose of finding values( y1 y2) we calculate them according to the
equation:
Yb = Xb B-1
Whereas
Yb Basic variables in the optimal dual form (y1,y2 )
Xb The row of new objective function coefficients are sorted by the
resulting variables in the optimal solution table .
B-1 Matrix of slack variables as display in optimal solution
1 0
y1 y2 = 3 0 3 + 0 0+0
-1 1
y1 y2 =(3 0 )
As we have already explained that changing the coefficients of the objective
function leads to a change in the objective function row in the optimal
6
solution. So, it is required to calculate the coefficient of (X1, X2) in the
optimal solution table as follows :
It is equal to the difference between the left side of the each constraint and
the value of right side ,of the that constraint of the dual model.
Coefficient of Xi= (constraint)i - bi
Coefficient Xi = (constraint) i - bi )
Coefficient of X1 = (y1+y2) -3 = (3+0) -3= 0
Coefficient of X2 = (5y1-5y2) -1 = (15-0) -1= 14
Coefficient of X3 = (2y1-5y2) -3 = (6-0) -3= 3
y1 + y2 3
5y1 - 5y2 1 y1 y2 =( 3 0 )
2y1 - 5y2 3
New objective function : Max Z= 3X1+X2+3X3= 3(30)+0+0=90
B.V X1 X2 X3 S1 S2 R.H.S
Z 0 14 3 3 0 90
X1 1 5 2 1 0 30
S2 0 0 -8 -1 1 10
7
3. Addition of anew variable (variables)
IF we assumed that the dual model …..
Min Z = 16y1 + 11y2 + 15y3
s.t 2y1+y2+y3 35 of the primal model
y1+2y2+3y3 55
y1, y2, y3 0
Max Z= 35x1+55x2 Max Z= 35x1+55x2 +40X3
s.t 2x1 + x2 16 s.t 2x1 + x2 + 4X3 16
x1 + 2x2 11 x1 + 2x2 + 12X3 11
x1 + 3x2 15 x1 + 3x2 + 3X3 15
x1, x2 0 x1, x2 0
Min Z= 16y1+11y2 +15y3
s.t 2y1 + y2 + y3 ≥ 35
y1 + 2y2 + 3y3 ≥ 55
4y1 + 12y2 + 3y3 ≥ 40
y1, y2,y3 0
Let's add a new variable (X3) to the original problem:
The coefficients of (X3) in the three constraint are (3, 12, 4) respectively, and
the ….. coefficient of the objective function is 40
Then the new constraint in dual model 4y1+2y2+3y3 ≥ 40
The optimal solution for the previous model:
Y1=10/3, y2= 70/3, y3=0
B.V X1 X2 S1 S2 S3 R.H.S
Z 0 0 10/3 70/3 0 310
S3 0 0 1/3 -5/3 1 2
X1 1 0 2/3 -1/3 0 7
8
X2 0 1 -1/3 2/3 0 2
To compensate these values with the new constraint:
4y1+2y2+3y3 ≥ 40
4(10/3) + 2(70/3) + 3(0) ≥ 40
40/3 +140/3 +0 ≥ 40
60 ≥ 40
Therefore, the constraint that was added corresponds to the optimal solution.
For the purpose of calculating the coefficients of the new variable (X3) in the
last table, this is as follows :
1/3 -5/3 1 4 1
2/3 -1/3 0 2 2
X3 = =
-1/3 2/3 0 3
0
The results explained in the optimal solution table:
B.V X1 X2 X3 S1 S2 S3 R.H.S.
Z 0 0 20 10/3 70/3 0 310
S3 0 0 1 1/3 -5/3 1 2
X1 1 0 2 2/3 -1/3 0 7
X2 0 1 0 -1/3 2/3 0 2
The table still optimal solution because coefficient of X3 is positive
number in Z row . If we assume that the coefficient of X3 in
objective function is 65 the constraint in dual model is
9
4y1+2y2+3y3 65
So, compensation in the dual model the result will be as follows:
4 (10/3) + 2(70/3) + 3(0) 65
60 65
That is, the constraint will not met with the values of the optimal solution of
the model. The results in the last optimal table are shown as in the table:
B.V X1 X2 X3 S1 S2 S3 R.H.S.
Z 0 0 -5 10/3 70/3 0 310
S3 0 0 1 1/3 -5/3 1 2
X1 1 0 2 2/3 -1/3 0 7
X2 0 1 0 -1/3 2/3 0 2
Note that the solution does not optimal because the X3 coefficient in the Z row
is (-5), so to get optimal we must select X3 as entering variable and then
complete the solution.
4. Addition of new constraint (constraints)
In order to study the effect of adding a new constraint to the problem of linear
programming and in the case of achieving this constraint using the values of
the optimal solution. This constraint can be considered in this case a slack
does not affect the optimal solution.
If we assume that the constraint to be added to the model
Max Z= 30x1+50x2
s.t 2x1 + x2 16
x1 + 2x2 11
x1 + 3x2 15
x1, x2 0
10
X1+4X2 16
To substitute the values of X1e X2 in the optimal solution table:
B.V X1 X2 X3 S1 S2 S3 R.H.S.
Z 0 0 20 10/3 70/3 0 310
S3 0 0 1 1/3 -5/3 1 2
X1 1 0 2 2/3 -1/3 0 7
X2 0 1 0 -1/3 2/3 0 2
7 + 4(2) 16
15 16
Therefore, we find that the constraint is achieved and therefore this constraint
can be considered a slack that does not affect in the optimal solution. But if the
constraint is as follows:
X1+4X2 13
7+4(2) 13
15 13
Here we note that the above limitation will not be achieved:
In order to study the effect of this constraint on the problem, the new
constraint is added:
X1+ 4X2+ ….. +S4 =13
To the problem after extract the value of (x1,x2) and compensation
in the new constraint to get the value of S4 row :
X1+ (2/3) S1- (1/3) S2 = 7
11
From this we get the equation x1 in terms of the rest of the variables
X1= 7- (2/3) S1+ (1/3)S2
As well as for X2 where : X2 - (1/3)S1 + (2/3)S2 = 2
X2 = 2 + (1/3)S1 - (2/3)S2
And to compensate X1 X2 in the new constraint values we get:
[ 7- (2/3)S1 + (1/3)S2] + 4[2+(1/3)S1-(2/3)S2] + S4= 13
(2/3)S1-(7/3)S2+S4= -2
****************
12
