Research Topic:

SENSITIVITY ANALYSIS

Presented by:
Mozombite Taipe, Stephanie
- Sierra Berdiales, Víctor
- Trinidad Rojas, Geraldyne
- Vega Bernuy, Gabriela
Zarzosa Sánchez, Andeleine

Professor:
Mg. Sánchez Guzmán, Jorge

National University José Faustino Sánchez Carrión

Faculty of Chemical Engineering and Metallurgy
School of Chemical Engineering

Huacho - 2016
Content
INTRODUCTION TO SENSITIVITY ANALYSIS
SENSITIVITY ANALYSIS.......................................................................................................................... 4
GRAPHICAL SENSITIVITY ANALYSIS................................................................................................... 4
ALGEBAIC SENSITIVITY ANALYSIS. CHANGES ON THE RIGHT SIDE ...................................... 6
ALGEBRIC SENSITIVITY ANALYSIS. OBJECTIVE FUNCTION ........................................................ 12
SENSITIVITY ANALYSIS WITH TORA, SOLVER, AND AMP................................................... 16
BIBLIOGRAPHY.................................................................................................................................... 19

SENSITIVITY ANALYSIS 2
 INTRODUCTION TO SENSITIVITY ANALYSIS
The work of the operations research team only begins when it has been applied
successfully the simplex method to identify an optimal solution. An assumption of
Linear programming is that all the parameters of the model (aij, bij, and cij) are constants.
known. In reality, the values of the parameters used in this model are only
estimates based on a prediction of future conditions. The data obtained for
developing these estimates is often quite imperfect or non-existent, it is because
this reason that the parameters of the original formulation may represent little more than
some small rules provided by the line staff who perhaps felt
pressured to give their opinion. The data may even represent optimistic estimates
the pessimists who protect the interests of the estimators.
For all this, a reasonable manager and the operations research staff will maintain
certain skepticism regarding the original values provided by the computer and, in the
In many cases, they will be considered only as a starting point for subsequent analysis.
of the problem. An 'optimal' solution is optimal only with respect to the model
specific that is being used to represent the real problem, and such a solution does not become
in a reliable guide for action until it is verified that your behavior is good
for other reasonable representations of the problem. Even more, sometimes the parameters
The model (in particular bii) is established as a result of policy decisions.
managerial decisions, and these decisions must be reviewed after detecting their consequences.
These are the reasons why it is important to carry out an analysis of
sensitivity, to investigate the effect it would have on the optimal solution provided by
the simplex method the fact that the parameters would take other possible values. In general,
there will be some parameters to which any reasonable value can be assigned without
affect the optimality of the solution. However, there will also be parameters with values
probable that will lead us to a new optimal solution. This situation is particularly
worrying, if the original solution takes on substantially lower values in the function
goal, or maybe not feasible!
Therefore, the fundamental objective of sensitivity analysis is to identify the parameters
sensitive, (for example, the parameters whose values cannot change without changing the
optimal solution). For certain parameters that are not classified as sensitive, also
it can be very useful to determine the range of values of the parameter for which
the optimal solution does not change. (This range of values is known as permissible range
to remain optimal). In some cases, changing the value of a parameter can affect
the feasibility of the optimal BF solution. For such parameters, it is useful to determine the interval
of values for which the optimal BF solution (with the adjusted values of the variables
basic) will continue to be feasible. (This interval is called the permissible interval for
remain feasible).
Information of this kind is invaluable in two ways. First, it identifies the parameters
most important, with which special care can be taken in making their estimates and
when selecting a solution that performs well for most of the values
possible. Second, identify the parameters that will need to be closely monitored when the
study is put into practice. If it is discovered that the actual value of a parameter is found

SENSITIVITY ANALYSIS 3
outside of its permissible value range, this is a sign that it is necessary to change the
solution.
In essence, the fundamental idea immediately reveals the way in which changes to the model
they would alter the numbers in the final simplex table (if it is assumed that it is doubled the
same sequence of algebraic operations that the simplex method performed the first time.
Therefore, after doing a few calculations to update this simplex table, we
you can easily check if the original optimal BF solution is now non-optimal (or not
feasible). If so, this solution will be used as the initial basic solution to start from.
new the simplex method (or the dual simplex) to find a new optimal solution, if it
desire. If the changes made to the model are not major changes, only minor adjustments will be required.
a few iterations to obtain the new optimal solution from this basic solution
advanced initial.

SENSITIVITY ANALYSIS
In PL, the parameters (input data) of the model can change within certain limits.
without changing the optimal solution. This is known as sensitivity analysis and will be the
theme of this section. Later, in chapter 4 we will study the post-optimal analysis,
which is related to the determination of the new optimal solution when changes are made
certain input data.
The presentation explains the basic ideas of sensitivity analysis through the solution
graph, and then they extend to the general problem of LP based on the results that
they appear in the simplex table.

GRAPHICAL SENSITIVITY ANALYSIS

This section demonstrates the general idea of sensitivity analysis. Two cases will be considered:
1. The sensitivity of the optimal solution to changes in the availability of the
resources (right side of the constraints).
2. The sensitivity of the optimal solution to changes in unit utility or cost
unitary (coefficients of the objective function).
Example: (Changes on the right side).
JOBCO manufactures two products on two machines. One unit of product 1 requires 2 hours.
in machine 1, and 1 hour in machine 2. One unit of product 2 requires 1 hour in the
machine 1, and 3 hours on machine 2. The income per unit of products 1 and 2 is
$30 and $20, respectively. The total available daily processing time in each
the machine is for 8 hours.
Yes 1y 2they are the daily quantities of units of products 1 and 2, respectively, the
The PL model is given as:
Maximizez = 30x1+ 20x2
s.a:
2 1+ 2≤ 8 (Machine 1)
1+ 3x2<= 8 (Machine 2)
1 , 2greater than or equal to 0

SENSITIVITY ANALYSIS 4
The figure below illustrates the change in the optimal solution when the changes
capacity of machine 1. If the daily capacity increases from 8 to 9 hours, the new
optimal will move to point G. The exchange rate in the optimal range as a consequence of the change
the capacity of machine 1 from 8 to 9 hours is calculated as:

Graphical sensitivity of the optimal solution to changes in resource availability (side

right of restrictions.

SENSITIVITY ANALYSIS 5
The calculated rate provides a direct link between the input data to the model
(resources) and their results (total income). It is said that a unitary increase (reduction) in
the capacity of machine 1 will increase (decrease) the income by $14.00.
The unit value name of a resource is an appropriate description of the exchange rate.
from the objective function by unit change of a resource. However, the first
developments of the PL coined the abstract name dual price (or shadow), and now this
name is a standard in all PL literature and in software packages.
The presentation in this book conforms to this standard.
In figure 3.12 we can see that the dual price of $14/hour remains valid for changes.
(increments or reductions) in the capacity of machine 1 that move its constraint
parallel to itself to any point on the line segment BF. We calculate the
machine 1 capabilities at points B and F as follows:
Minimum capacity of machine 1 [inB = (0.267)] =2× 0 +1× 20.67 =2.67h
Maximum capacity of machine 1 [inF = (8,0)] = 2× 8 +1× 0 = 16h

The conclusion is that the dual price of$ 14 ⁄ remains valid in the interval
ℎ
2.67h ≤ capacity of the machine1≤ 16h
Changes outside this interval lead to a different dual price (value per unit).
By making similar calculations, we can verify that the dual price for the capacity of the
machine 2 is from$ 2.00 ⁄ , and which does not change when its capacity remains within
ℎ
line segment DE. Now:

Minimum capacity of the machine 2 [inD = (4,0)] = 1× 4 + 3 × 0 = 4h

Maximum capacity of machine 2 [inE = (8,0)] =1× 0 + 3 × 8 = 24h
Therefore, the dual price of$ 200 ⁄ for machine 2 does not change within the interval
ℎ
4h ≤ machine capacity2≤ 24h
The calculated limits for machines 1 and 2 are known as feasibility intervals.
All software packages provide information about dual prices and their
feasibility intervals. Section 3.6.4 shows how this information is generated by AMPL,
Solver and TORA.

ALgebraic SENSITIVITY ANALYSIS. CHANGES ON THE SIDE

RIGHT
The sensitivity analysis or post-optimal for Linear Programming models aims to
Objective to identify the impact that results from the outcomes of the original problem after
certain variations in the parameters, variables, or constraints of the model, without
I had to go through solving the problem again.
That is, whether we solve our model graphically or using the Simplex Method,
what is sought is that these variations or sensitivities make use of the optimal solution and value
actually, without having to solve a new problem for each variation. In particular

SENSITIVITY ANALYSIS 6
nos concentraremos en el análisis de sensibilidad o postoptimal que hace uso de la tabla final
of the Simplex Method.

What is it?

Following the notation used in the section dedicated to theSimplex Methodon our site,
this operates for Linear Programming models in a standard format.
My cTx
s.a Ax = b
x >= 0
Where the final table of the Method maintains the following structure:

Where:
I: Identity Matrix
Reduced costs associated with basic variables
B: Matrix of basic variables
D: Matrix of non-basic variables
Right side
Cb: Coefficients in the objective function associated with the basic variables

Cd: Coefficients in the objective function associated with non-basic variables

1. Change in the 'right side' of the restrictions:What is sought to identify is whether the
The actual basic variables remain after the modification of one or more parameters.
associated with the "right side" of the model. If we calculate:

and it is fulfilled The same basic variables are also of the new one.
optimal solution, calculated with the new one. If the above is not met, the can be applied.
Dual Simplex Method.
EXAMPLE: Without resolving the problem again, it is desired to know if the current variables
optimal basics of the problem are also those of the same problem, where the right sides
corresponds to the vector b = (20,30). (Observation: X4 and X5 are slack variables of the
restriction 1 and 2 respectively

SENSITIVITY ANALYSIS 7
Max 2x1 + 7x2 - 3x3
sa: x1 + 3x2 + 4x3 <= 30
x1 + 4x2 - x3 <= 10
x1, x2, x3 >= 0

X1 X2 X3 X4 X5

0 -1 5 1 -1 20

1 4 -1 0 1 10

0 1 1 0 2 201

To analyze this scenario we must calculate the basic variable vector and verify if
All its components are positive definite. Note that for this we need matrix B.
inverse, which we can easily retrieve by identifying the parameters associated with X4 and
X5 (slack variables of constraint 1 and 2 respectively) in the final table of the Method
Simplex:

Then, since at least one of the coefficients of the new right side has a value
negative, change the current optimal base. It is worth noting that in this situation it is not necessary
resolve the new scenario starting from scratch, but what needs to be done is to use the table
final of the simplex of the base scenario, updating the right side and value of the function
objective.

X1 X2 X3 X4 X5

0 -1 5 1 -1 -10

1 4 -1 0 1 30

0 1 1 0 2 60

Subsequently, iteration continues using the Dual Simplex Method. (See reference
to the right).
[Link] of a new variable:We must evaluate if the new variable is a contribution.
significant to the results of the original model. Then, to say if the current solution
the basic is optimal for the new problem, we calculate the reduced cost of the new variable
as:

, where k is the index of the new variable and Ak its respective column.
in the coefficient matrix. If it holds that rk>=0, the current optimal solution is preserved.

SENSITIVITY ANALYSIS 8
Otherwise, one can continue with the Simplex by adding a new column to the table.
with inputs B-1Ak and rk and taking as the entering variable to the new base that we just finished.
to introduce to the problem.
EXAMPLE: The possibility of developing a new product with benefit is desired to be studied.
net equals 8 and requires 4, 2, and 5 units of the resources associated with each constraint.
Without solving the problem again, is it worth producing the product?
Max 9x1 + 12x2
sa: 4x1 + 3x2 <= 180
2x1 + 3x2 <= 150
4x1 + 2x2 <= 160
x1, x2 >= 0

X1 X2 X3 X4 X5

1 0 1/2 -1/2 0 15

0 1 -1/3 2/3 0 40

0 0 -4/3 2/3 1 20

0 0 1/2 7/2 0 615

It is necessary to evaluate rk and determine if it is >=0.

In this example rk=1>=0, therefore it is not advisable to incorporate this new variable.
model, that is, even when it is incorporated we will not obtain an optimal value that exceeds the
actual V (P)=615. Anyway, we will show how it is included in the final table of
This modification is simple so that the reader can understand its incorporation when it is
necessary

SENSITIVITY ANALYSIS 9
 X1 X2 X3 X4 X5 XNew

-
1 0 1/2 0 1 15
1/2

-
0 1 2/3 0 0 40
1/3

-
0 0 2/3 1 1 20
4/3

0 0 1/2 7/2 0 1 615

If the reduced cost of this new variable had been zero, then the new scenario
it would have infinite solutions.

3. Change in the Coefficients Objective Function: The aim is to identify what happens with the current
optimal solution of the base scenario if one or more of the coefficients that define it are changed
the objective function. The current optimal solution will also be for the new scenario
as long as the new reduced costs are greater than or equal to zero (note that also
change the value of the objective function in the current optimal solution). That is, it must be fulfilled
what:

Otherwise, the Simplex is applied from the final table of the original model, with the
new reduced costs and new value of the current basic solution.
EXAMPLE: Without resolving the problem again, we want to know what happens if it is modified.
the parameters of the objective function, remaining as follows: Z = x1 + 5x2 -
2x3. (X4 and X5 are the slack variables of constraint 1 and 2 respectively).
Max 2x1 + 7x2 - 3x3
sa: x1 + 3x2 + 4x3 <= 30
x1 + 4x2 - x3 <= 10
x1, x2, x3 >= 0

SENSITIVITY ANALYSIS 10
 X1 X2 X3 X4 X5

0 -1 5 1 -1 20

1 4 -1 0 1 10

0 1 1 0 2 20

Because changes in the parameters of the objective function occur in more than
we will consider the following formula:

Because at least one of the reduced costs of the non-basic variables has become
negative, then change the current solution and optimal value of the problem. To incorporate
This modification in the final table of the Simplex Method updates the reduced costs.
associated with the non-basic variables, in addition to the optimal value, resulting as follows:

X1 X2 X3 X4 X5

0 -1 5 1 -1 20

1 4 -1 0 1 10

0 -1 1 0 1 10

4. Inclusion of a new restriction:To know if the current solution and optimal value are
will be maintained after incorporating a new constraint to the problem, the solution must be evaluated.
current and verify if it meets the new restriction. If so, the current solution
it will also be about the problem with the new restriction, otherwise the new one will be incorporated
restriction to the final table of the Simplex of the base scenario.

EXAMPLE: Without resolving the problem again, we want to know what happens if it is considered
a new constraint of the form: 3x1 + 2x2 + 3x3 <= 25. (Note: Consider the same
model and final table of the previous example

SENSITIVITY ANALYSIS 11
The current solution is evaluated against the constraint: 3*(10) + 2*(0) + 3*(0) <= 25. It does not satisfy.
Thus, this new constraint is added as a row to the final Simplex table.
Additionally, X6 is added as a slack variable associated with this new constraint:

X1 X2 X3 X4 X5 X6

0 -1 5 1 -1 0 20

1 4 -1 0 1 0 10

3 2 3 0 0 1 25

0 1 1 0 2 0 20

An alternative to find the optimum through this table is to form the identity.
(we must make the parameter associated with X1 in the third row zero) by multiplying row 2 by
-3 and adding that result to row 3. This way it is obtained:

X1 X2 X3 X4 X5 X6

0 -1 5 1 -1 0 20

1 4 -1 0 1 0 10

0 -10 6 0 -3 1 -5

0 1 1 0 2 0 20

Finally, we obtain X4, X1, and X6 as basic variables. This is the result of the transformation.
the right side remains negative and in this case we can proceed using the
Dual Simplex Method.

ALGEBRAIC SENSITIVITY ANALYSIS. OBJECTIVE FUNCTION

Previously, we used graphical sensitivity analysis to determine the conditions.
that will maintain the optimality of the solution of a linear programming problem with two variables. Now
We will extend these ideas to the general linear programming problem.
DEFINITION OF REDUCED COST
To facilitate the explanation of the sensitivity analysis of the objective function, first
we need to define the reduced costs. In the TOYCO model, the objective equation z
what appears in the optimal table can be written as

= − − −

SENSITIVITY ANALYSIS 12
The optimal solution does not produce toy trains.x1= 0The reason becomes apparent
in the equation z, where a unit increase in x1 (over its current zero value) reduces
a z in $4, that is, z = 1350 - 4 * (1) - 1 * (0) - 2 * (0) = $1346.
We can consider the coefficient of x1 in the equation z (5 4) as a unit cost because
It reduces income z. But where does this 'cost' come from? We know that the income per unit
X1 is $3 (according to the original model). We also know that the production of trains of
A toy incurs a cost because it consumes resources (operational time).
consequently, from the point of view of optimization, the "attractiveness" of x1 depends on the
cost of the resources consumed in relation to income. This relationship defines the so-called
reduced cost and is formalized in the literature of PL as

To appreciate the importance of this definition, in the original TOYCO model, the income
for toy trucks (= $2) is less than that of toy trains (= $3). No
despite the optimal solution recommending the production of toy trucks = 100 units) and
nothing of trains = 0). The reason is that the cost of the resources consumed by a truck
toy (that is, operation time) is less than its unit price; on the contrary
what happens in the case of toy trains.
With the given definition of reduced cost, we can see that an unprofitable variable (such as
) can become profitable in two ways:
Increasing unit income.
2. Reducing the unit cost of consumed resources.
In most situations, market conditions dictate the price per unit and
it can be difficult to increase it at will. On the other hand, a more viable option is to reduce
the consumption of resources because the manufacturer can reduce costs by making the process of
make production more efficient.
DETERMINATION OF OPTIMALITY INTERVALS
Now we focus on determining the conditions that will maintain an optimal
solution. The development is based on the definition of reduced cost. In the TOYCO model,
sean d1, d2 and d3 the changes in the unit revenues of trucks, trains, and cars,
respectively. The objective function is then written as
Maximize z = (3 +d1 ) x1+ (2 +d2 ) x2+ (5 +d3 ) x3
First, we consider the general situation in which all the objective coefficients change.
at the same time.
With the simultaneous changes, row z in the home table appears as:

SENSITIVITY ANALYSIS 13
When we generate the simplex table with the same sequence of the input variables and
outputs used in the original model (before the changes are made tod ), la
optimal iteration will appear as follows (make sure this is the case by performing the
simplex row operations

The new optimal table is the same as the original optimal table, except for the reduced costs.
(coefficients of the equation z). This means that changes in the coefficients of the
Objective function can only affect the optimality of the problem.
In fact, you do not have to perform the simplex row operation to calculate the new costs.
reduced. An examination of the new row z shows that the coefficients ofd are taken
directly from the coefficients of the constraints of the optimal table. One
A convenient way to calculate the new reduced cost is to add a new top row and
a new column further to the left of the optimal table, as shown by the areas
shadowed in the following illustration.

The entries in the top row are the changes associated with the variable.d In the column
to the far left, the top element is 1 in row z followed by the change of di the
basic variablex1 Please note thatd = 0 for the slack variablex .

SENSITIVITY ANALYSIS 14
To calculate the new reduced cost for any variable (or the value of z), multiply
the elements of your column by the corresponding elements that appear in the column
To the extreme left, add them and subtract the element in the top row of the sum. For example,
forx1 , we have

Reduced cost ofx1= [4 * 1 + (- -1) * d2+ 3* d3+ 2 * 0] -d1

4 2

4–1d2+ 3d3- d1
4 2

The current solution remains optimal as long as the reduced costs (coefficients of the
equation z) remain non-negative (maximization case). Therefore, we have the
simultaneous optimality conditions corresponding to x1, x4, and x5 do not
basic:

Remember that the reduced cost of a basic variable is always zero, as shown by the
modified optimal table. To illustrate the use of these conditions, suppose that the function
The objective of TOYCO changes from z = 3x1 + 2x2 + 5x3 to z = 5 - 2x1 + x2 + 6x3. Then, d1 = 2
- 3 = -$1, d2 = 1-2 = - $1 and d3 = 6 -5 = $1. The substitution in the given conditions presents
the result

The results show that the proposed changes will maintain the current solution (x1 = 0,
x2 = 100, x3 = 230) optimal (with a new value of z = 1350 + 100d2 + 230d3 = 1350 + 100
- 1 + 230 * 1 = $1480. If any condition is not met, a new one must be established.
solution
The previous topic addressed the case of maximization. The only difference in the case of
minimization is that the reduced costs (coefficients of the z equation) must be ≤ 0 for
maintain optimality.
The intervals of optimality that have to do with the changes ofd one at a time they can
develop based on simultaneous optimality conditions. For example, suppose
that the objective coefficient of x2 only changes to 2 1 d2; that is, d1 5 d3 5 0. The
simultaneous optimality conditions therefore reduce to

SENSITIVITY ANALYSIS 15
In the same way, you can verify that the individual changes (3 + d3) and (5 + d3) for x1 and
x3 and the optimality intervals d1 < 4 and d3 ≥ -8$ 2, respectively.
3

The given individual conditions can be translated into unit income intervals.
totals. For example, for the toy trucks (variable x2), the total unit income is 2
+ d2, and its optimal interval -2 ≤ d2 ≤ 8 translates to $0 ≤ (unit income of the truck
(toy) ≤ $10
It is supposed that the unit revenues from toy trains and cars remain fixed at $3.
$5, respectively.
It is important to note that the changes d1, d2, and d3 may be within their intervals.
permissible individuals without meeting simultaneous conditions and vice versa. For example,
consider z=6x1+8x2+3x3. In this case d1 = 6 - 3 = $3, d2 = 8 - 2 = $6 and d3 = 3 - 5 = -$2,
which fall within the permissible individual intervals (-∞, d1 ≤ 4, -2 d2 ≤ 8, and
- 8d3 <∞) However, the corresponding simultaneous conditions yield
3

SENSITIVITY ANALYSIS WITH TORA, SOLVER, AND AMP

We now have all the tools to decipher the results provided by
the PL software, particularly regarding sensitivity analysis. We will use the
example of TOYCO to demonstrate the results obtained with TORA, Solver and AMPL. The report of
The PL results obtained with TORA provide the data for the sensitivity analysis.
automatically as shown in figure 1.4.1 (file tora [Link]).

SENSITIVITY ANALYSIS 16
 FIGURE 1.4.1 Sensitivity analysis, conducted with TORA for the TOYCO model

The results include reduced costs and dual prices as well as the intervals of
permissible optimality and feasibility. Figure 1.4.2 shows the TOYCO model
analyzed with Solver (file [Link]) and its sensitivity analysis report.

FIGURE 1.4.2 Sensitivity analysis report conducted with Excel Solver for the
TOYCO model
After clicking on the Solver option in the Solver Parameters dialog box, you can
request the sensitivity analysis report in the new Solver Results dialog box.
Then click on the Sensitivity Report 1 tab to view the results. The report is
similar to that of TORA, with three exceptions:

SENSITIVITY ANALYSIS 17
The reduced cost has an opposite sign.

(2) Use the term shadow price instead of dual price.

The optimality intervals are for the changes dj and Dj and not for the coefficients.
total objectives and the right-hand sides of the constraints. The differences are minimal, and the
interpretation of the results no change.

In AMPL, the sensitivity analysis report is obtained immediately. The file

[Link] provides the necessary code to determine the obtained results
with the sensitivity analysis. Additional instructions are required (the report is sent to
file [Link]) following:

CPLEX option instructions are required to obtain the analysis report.

sensibilidad estándar. En el modelo de TOYCO, las variables y restricciones con subíndices
they use the root names x and oper., respectively. Using these names, the suffixes
The .down, .current, and .up allusions in the instructions automatically generate the report of
sensitivity analysis formatted that appears in figure 1.4.3 The suffixes .dual and .rc
they provide the dual price and the reduced cost.

FIGURE 1.4.3 Sensitivity analysis report obtained with AMPL for the model of
TOYCO

SENSITIVITY ANALYSIS 18
BIBLIOGRAPHY
[Link]
in-linear-programming-using-the-final-table-of-the-simplex-method/
[Link]
[Link]
Taha, H. A. (2012). Operations Research. MEXICO: PEARSON
EDUCATION.

SENSITIVITY ANALYSIS 19