Professional Documents
Culture Documents
TYNSKAAI2018
TYNSKAAI2018
net/publication/349110457
CITATION READS
1 5,814
3 authors, including:
4 PUBLICATIONS 3 CITATIONS
University of Barisal
9 PUBLICATIONS 7 CITATIONS
SEE PROFILE
SEE PROFILE
All content following this page was uploaded by Shek Ahmed on 08 February 2021.
Abstract— The purpose of this paper is to implement Astronomy, Physics, Structural and Molecular
the concept of Sensitivity Analysis (SA) of Linear Biology, Engineering, Computer Science and
Programming Problems (LPPs) in real life. By using Medicine. LPP is a mathematical method for
this post optimality analysis one can decide how to determining a way to achieve the best outcome (such
change in the cost coefficients, changes in the right side as maximum profit or lowest cost) in a
of the constraints, changes in the coefficients of the given mathematical model for some list of
constraints, addition of a new variable, addition of a requirements represented as linear relationships. More
new constraints and deletion of a variable. We formally, Linear programming is a technique for
demonstrate our post optimality test by changing the optimization of a linear objective function subject
various parameters on the optimal solution of a LPP. If to linear equality and inequality constraints.
the number of decision variables and constraints for It’s feasible region is a convex polyhedron, which is a
LPP is large then it will be very difficult and time set defined as the intersection of finitely many half
consuming to discuss the Sensitivity Analysis spaces, each of which is defined by a linear inequality.
manually. For this purpose we have also developed Its objective function is a real-valued affine
computer code using LINDO to describe the test. In function defined on this polyhedron. A linear
this paper we consider a production planning problem programming algorithm finds a point in the polyhedron
of a company to discuss the analysis. where this function has the smallest (or largest) value
if such a point exists.
LINDO (Linear, Interactive, and Discrete Optimizer) is
Index Terms— Feasible Solution, Linear a software package for linear programming, integer
Programming Problem, LINDO, Optimal Solution, programming, nonlinear programming, stochastic
Sensitivity Analysis. programming and global optimization. LINDO also
creates "What's Best!" which is an add-in for linear,
I. INTRODUCTION integer and nonlinear optimization. First released for
S
ENSITIVITY analysis plays a crucial role to Lotus 1-2-3 and later also for Microsoft Excel.
discuss the variation of different parameters in Objective of this paper is to discuss the effect on
Linear Programming Problem. It gives a direct optimality by changing the various parameters of a
test of optimality of a particular problem due to LPP.
the change of a component of cost or requirement
vector. It works as a decisive step in the model building II. PRELIMINARIES
and result communication process. Through SA we In this section we discuss some basic definitions and
gain essential insights on the model behavior, on its techniques relevant to our work.
structure and on its response to changes in the model
inputs. There is a very wide range of uses to which A. Linear Programming Problem (LPP) [4]
Sensitivity Analysis is introduced. There are four main The Linear Programming Problem (LPP) is to find the
categories to discuss Sensitivity Analysis such as
decision variables 𝑥𝑗 , (𝑗 = 1,2, … …. …. , 𝑛) which
decision making or improvement of recommendations
is optimizing (minimizing or maximizing) the
for decision makers, communication, increase
objective function.
understanding or quantification of the system, and
model development. The general mathematical form of a (LP) problem is as
follows:
Linear Programming, sometimes known as linear Optimize 𝑍 = 𝑐1 𝑥1 + 𝑐2 𝑥2 + . … . + 𝑐𝑛 𝑥𝑛
optimization, is the problem of maximizing or
minimizing a linear function over a convex polyhedron Subject to: 𝑎11 𝑥1 + … . + 𝑎1𝑛 𝑥𝑛 (≤, =, ≥) 𝑏1
specified by linear and non-negativity constraints. LPP 𝑎21 𝑥1 + … . + 𝑎2𝑛 𝑥𝑛 (≤, =, ≥) 𝑏2
is the optimization of an outcome based on some set of … … … … … …
constraints using a linear mathematical model. … … … … … …
Optimization problems arise in all branches of 𝑎𝑚1 𝑥1 +. .. + 𝑎𝑚𝑛 𝑥𝑛 (≤, =, ≥) 𝑏𝑚
Economics, Finance, Chemistry, Materials Science,
𝑥𝑗 ≥ 0 , 𝑤ℎ𝑒𝑟𝑒 𝑗 = 1,2, … … , 𝑛
Where one and only one of the signs ≤ , = , ≥ holds for Case (i):
each constraints and the sign may vary from one
At the optimal table all 𝐸𝑗 − 𝑐𝑗 ≥ 0 (For maximization
constraint to another. The coefficients 𝑐𝑗 , (𝑗 =
problem).
1,2, …. , 𝑛) are called profit (or cost) coefficients.
The constants 𝑏𝑖 , (𝑖 = 1,2 …. , 𝑚) in the constraints Let ∆𝑐𝑗 be the amount to be added to 𝑐𝑗 corresponding
conditions are called the stipulations and the constants to a non-basic variable 𝑥𝑗 . For discrete change of the
𝑎𝑖𝑗 , (𝑖 = 1,2 …. , 𝑚, 𝑗 = 1,2, … …. …. , 𝑛) are cost component 𝑐𝑗 the simplex table remains optimal
called the structural coefficients. proved,
In matrix form 𝐸𝑗 − (𝑐𝑗 + ∆𝑐𝑗 ) ≥ 0
Optimize 𝑍 = 𝑐𝑇 𝑥 ⇒ 𝐸𝑗 − 𝑐𝑗 ≥ ∆ 𝑐𝑗 … … … (𝑖)
Subject to 𝐴𝑥 = 𝑏 From (i), we can say that there is no lower bound of ∆𝑐𝑗
and 𝑥 ≥ 0. and any change of 𝑐𝑗 does not change the value of the
objective function since 𝑥𝑗 = 0 (non-basic variable).
B. Sensitivity Analysis [18]
Case (ii):
The study of the effect of discrete changes in
parameters on the optimal solution is called sensitivity Let 𝑥𝑘 = 𝑥𝐵𝑟 be the 𝑟-th component of basic variable
analysis or post-optimality analysis, while that of (𝑟 = 1,2, … … … , 𝑚) and the cost component 𝑐𝑘 = 𝑐𝐵𝑟
continuous changes in parameters is called parametric corresponding to 𝑥𝑘 changes to 𝑐𝐵𝑟 + ∆𝑐𝐵𝑟 .
programming. Now the change of 𝑐𝑘 effects all 𝐸𝑗 , for all 𝑗 in the basis.
Sensitivity analysis plays a main job to discuss the Now due to the changes of 𝑐𝑘 ,
nature in linear programming problem. It gives a direct 𝐸𝑗 − 𝑐𝑗 = 𝑐𝐵 𝑦𝑗 − 𝑐𝑗 [𝑓𝑜𝑟 𝑎𝑙𝑙 𝑗 ∉ 𝐵] = ∑𝑚 𝑖=1 𝑐𝐵𝑖 𝑦𝑖𝑗 −
test of sensitivity of a particular problem due to the 𝑖≠𝑟
variation of a component of cost or requirement vector 𝑐𝑗 + (𝑐𝐵𝑟 + ∆𝑐𝐵𝑟 )𝑦𝑟𝑗 = ∑𝑚
𝑖=1 𝑐𝐵𝑖 𝑦𝑖𝑗 − 𝑐𝑗 + 𝑐𝐵𝑟 𝑦𝑟𝑗
etc. Mainly sensitivity analysis is in general a post- = 𝐸𝑗 − 𝑐𝑗 + ∆𝑐𝐵𝑟 𝑦𝑟𝑗
optimality test. It determines the range of a component for optimal solution, we must have all 𝐸𝑗 − 𝑐𝑗 ≥ 0 for
of the cost or requirement vector when all other all 𝑗 ∉ 𝐵.
component remain unchanged, so that the optimal
solution remains unaffected. Such type of change is Thus 𝐸𝑗 − 𝑐𝑗 + ∆𝑐𝐵𝑟 𝑦𝑟𝑗 ≥ 0, If 𝑦𝑟𝑗 = 0 for some 𝑗,
called a discrete change. then 𝐸𝑗 − 𝑐𝑗 + ∆𝑐𝐵𝑟 𝑦𝑟𝑗 = 𝐸𝑗 − 𝑐𝑗 ≥ 0
−(𝐸𝑗 −𝑐𝑗 )
For 𝑦𝑟𝑗 ≤ 0, ∆𝑐𝐵𝑟 ≤ … … … (𝑖𝑖) also for
𝑦𝑟𝑗
III. MEASUREMENT OF SENSITIVITY ANALYSIS
−(𝐸𝑗 −𝑐𝑗 )
[17] 𝑦𝑟𝑗 > 0, ∆𝑐𝐵𝑟 ≥ … … … (𝑖𝑖𝑖)
𝑦𝑟𝑗
To clarify the Sensitivity Analysis, we will discuss the Combining these two conditions in (𝑖𝑖) and (𝑖𝑖𝑖) we
following modifications: can say that the optimal basis remains unaffected if
∆𝑐𝑘 = ∆𝑐𝐵𝑟 lies in the interval given below,
Changes in the cost coefficients 𝑐𝑗 .
−(𝐸𝑗 − 𝑐𝑗 )
Changes in the right hand side of the max { , 𝑦𝑟𝑗 > 0} ≤ ∆𝑐𝐵𝑟 = ∆𝑐𝑘
𝑗 𝑦𝑟𝑗
constraints 𝑏𝑖 . −(𝐸𝑗 − 𝑐𝑗 )
Change in the coefficients of the constraints ≤ min{ , 𝑦𝑟𝑗 < 0}
𝑗 𝑦𝑟𝑗
𝑎𝑖𝑗 . −(𝐸𝑗 −𝑐𝑗 )
or, max { , 𝑦𝑟𝑗 > 0} ≤ ∆𝑐𝑘 ≤
Addition of a new variable. 𝑗 𝑦𝑟𝑗
−(𝐸𝑗 −𝑐𝑗 )
Deletion of a variable. min{ , 𝑦𝑟𝑗 < 0} … … … (𝑖𝑣)
𝑗 𝑦𝑟𝑗
Addition of a new constraint for all 𝑗 ∉ 𝐵. If no 𝑦𝑟𝑗 > 0, there is no upper bound of
Deletion of a constraint. ∆𝑐𝑘 and if no 𝑦𝑟𝑗 < 0, there is no lower bound of ∆𝑐𝑘 .
But here the value of the objective function changes by
∆𝑐𝐵𝑟 𝑥𝐵𝑟 = ∆𝑐𝑘 𝑥𝑘 .
A. Changes in the Cost Coefficients 𝑐𝑗 :
There are two different cases to discuss: B. Changes in the Right Hand Side of the Constraints
𝑏𝑖 :
𝑐𝑗 is the cost component corresponding to
a non-basis variable. We assume that an optimum basic feasible solution 𝑥𝐵
𝑐𝑗 is the cost component corresponding to with basis matrix 𝐵 for the linear programming
problem,
an optimal basic variable.
Maximize 𝑧 = 𝑐𝑥,
Combining (ii) and (iii), we can say that the optimality 𝑥𝑛+𝑚+1 = −𝑎𝐵 𝐴−1
𝐵 𝐴𝑗 + 𝑎𝑗 𝑎𝐵 𝐴−1
𝐵 𝑏
remains unaffected if ∆𝑏𝑘 lies in the interval, −𝛽
−𝑥𝐵𝑖 𝑧 = 𝑐𝑗 − 𝑐𝐵 𝐴−1 𝑐𝐵 𝐴−1
Max( , 𝑏𝑖𝑘 > 0) ≤ ∆𝑏𝑘 𝐵 𝐴𝑗 𝐵 𝑏
𝑖 𝑏𝑖𝑘
−𝑥𝐵𝑖
≤ Min ( , 𝑏𝑖𝑘 < 0) … … … (𝑖𝑖) Note that the bottom row is unchanged because
𝑖 𝑏𝑖𝑘
If no 𝑏𝑖𝑘 > 0, there is no upper bound and if no 𝑏𝑖𝑘 < 𝐴𝑗 𝐴−1 0 𝐴𝑗
𝑐𝑗 − 𝑐𝐵 𝐴̃−1
𝐵 [ 𝑎 ] = 𝑐𝑗 − [𝑐𝐵 0] [ 𝐵 −1 ][ ]
0, there is no lower bound. 𝑗 𝑎𝐵 𝐴𝐵 −1 𝑎𝑗
= 𝑐𝑗 − 𝑐𝐵 𝐴−1
𝐵 𝐴𝑗
C. Change in the Coefficients of the Constraints 𝑎𝑖𝑗 : Since the bottom row of the new tableau is
nonnegative, the corresponding point is dual feasible.
When chances took place in the constraint coefficients Either the value of 𝑥𝑙+1 is nonnegative and the tableau
of a non-basic variable in a current optimal solution, is optimal, or it is negative. In the latter case, dual
feasibility of the solution is not affected. The only simplex pivots can be performed until an optimal
effect, if any, may be on the optimality of the solution. solution is found or it is determined that the problem is
However, if the constraint coefficients of a basic primal infeasible. This is an instance when the current
variable get changed, things become more complicated basis has to be changed.
since the feasibility of the current optimal solution may G. Deletion of a Constraint:
also be affected (lost). The basis matrix is affected,
which, in turn, may affect all the quantities given in the The constraints to be deleted may be either binding or
current optimal table. Under such circumstances, it unbinding on the optimal solution. The deletion of a
may be better to solve the problem over again. new unbinding constraint can only enlarge the feasible
region but will not affect the optimal solution.
D. Addition of a New Variable:
Suppose we have solved a problem with an optimal IV. SENSITIVITY ANALYSIS ON A REAL LIFE
basis and we desire to add an extra variable with PROBLEM
constraint matrix column 𝑎 = ℝ𝑚 and objective A. Formulation of the Problem
coefficients 𝑑 = ℝ, that is we now have,
In this section we apply the idea of sensitivity analysis
Maximize 𝑧 = 𝑐𝑥 + 𝑑𝑦,
on a real life problem by collecting data from a
Subject to 𝐴𝑥 + 𝑎𝑦 = 𝑏, 𝑥, 𝑦 ≥ 0 business organization of Bangladesh.
To check whether adding this column effect the basis, “Amrita Consumer Food Products Ltd” is one of the
we just calculate the corresponding reduced entry. If renowned quality food producer in the greater southern
the entry is negative that is, 𝑑 − 𝑐𝐴−1 𝑎 < 0 , then the region of Bangladesh. Their products have lots of
basis is changed, otherwise it remains optimal. demand into market. Buyer especially middle class
family like their products due to cheap rate and
healthful for all aged of people. Day by day, they are That means to achieve maximum profit the company
gaining more and more trust and confidence of the has to manufacture 1 dozen of Super chanachur and 2
customers and also of associated in its field. Amrita dozen of Hot Chanachur without manufacturing the
Chanachur, Amrita snacks item, Amrita Biscuit, Bar-B-Q Chanachur.
Amrita Oil, Amrita Dal vaja are the most popular daily Now we discuss the following cases:
products among them. They supply their products all
over in Bangladesh. They want to optimize their profit B.2. Analysis on the Change of Cost Coefficient of Non-
by reorganizing manpower, material cost, Basic Variable
transportation cost etc. “Amrita Food Products Ltd”
wants to manufacture three types of Chanachur like (i) Find the ranges of the profit of non-basic variable
Super Chanachur, Hot Chanachur and Bar-B-Q Bar-B-Q Chanachur (c3) such that the current optimal
product mix remains optimal.
Chanachur. The profits per dozen on these products are
Tk.4, Tk. 6, Tk. 2 respectively. These products require (ii) What happens if the profit per dozen of Bar-B-Q
two types of resources- man-power and material. The Chanachur (c3) is increased to Tk. 12?
following Linear Programming Problem (LPP) is
formulated for determining the optimal product mix: Result and discussion on (i)
Maximize 𝑍 = 4𝑥1 + 6𝑥2 + 2𝑥3 , For table 1 to remain optimal,
Subject to 𝑥1 + 𝑥2 + 𝑥3 ≤ 3, (man-power), −1
𝑐̅3 ≤ 0 Or, 𝑐3 − (4,6) [ ] ≤ 0, Or, 𝑐3 − (−4 +
𝑥1 + 4𝑥2 + 7𝑥3 ≤ 9, (material), 2
12) ≤ 0, Or, 𝑐3 ≤ 8
𝑥1 , 𝑥2 , 𝑥3 ≥ 0 This means that as long as the profit per dozen of Bar-
Where 𝑥1 , 𝑥2 , 𝑥3 are the number of dozen of Super B-Q Chanachur is less than or equal to Tk. 8, it is not
Chanachur, Hot Chanachur and Bar-B-Q Chanachur. profitable to manufacture. We also show this result
using LINDO.
New optimal product mix is 𝑥1 = 2, 𝑥2 = 0, 𝑥3 = cost coefficient of the other basic variable Hot
1 𝑎𝑛𝑑 𝑧𝑚𝑎𝑥 = 𝑇𝑘. (4 × 2 + 6 × 0 + 12 × 1) = Chanachur.
𝑇𝑘. 20.
That means to achieve maximum profit the company B.3. Analysis on the Change of Right Hand Side of
has to manufacture 2 dozen of Super chanachur and 1 Constraint
dozen of Bar-B-Q Chanachur without manufacturing Find the range of changes of manpower and material
the Hot Chanachur. such that the current optimal solution remains optimal.
B.3. Analysis on the Change of Cost Coefficient of Result and discussion
Basic Variable We calculate the ranges of manpower and material
(i) Find the ranges of the profit of basic variable Super restriction using LINDO which is given bellow:
Chanachur (𝑐1 ) such that the current optimal solution
remains optimal. Right hand side ranges
Row Current Allowable Allowable
(ii) What happens if the profit per dozen of Super
RHS Increase Decrease
Chanachur (𝑐1 )is increased to Tk. 8?
2 3.000000 6.000000 0.750000
Result and discussion on (i) 3 9.000000 3.000000 6.000000
We calculate the range of c1 using LINDO
We observe from the above output of LINDO that
the range of change of manpower lies between 2.25 to
Range in which the basis is unchanged 9 and the range of change of material lies between 3 to
OBJ Coefficient Ranges 12.
Variable Current Allowable Allowable
B.4. Analysis on the Change in the Coefficients of the
Coefficient Increase Decrease
Constraints aij.
𝑥1 4.000000 2.000000 2.500000
(i) What is the effect on optimal solution if we change
𝑥2 6.000000 9.999999 2.000000 1 2
the column [ ] to [ ] in the expenditure of non-basic
7 2
𝑥3 2.000000 6.000000 Infinity variable Bar-B-Q Chanachur?
(ii) What is the effect on optimal solution if we change
We observe from the output of LINDO that the 1 3
the column [ ] to [ ] in the expenditure of basic
Range on 𝑐1 for the optimal product mix to remain 1 1
optimal is 1.5 ≤ 𝑐1 ≤ 6. variable Super Chanachur?
(iii) What is the effect on optimal solution if we change
Result and discussion on (ii) 1 1
𝑐 1 the column [ ] to [ ] in the expenditure of basic
When 𝑐1 = 8, 𝑐̅5 = 1 − 2 = ∗ 8 − 2 = 2/3 4 2
3 3
variable Hot Chanachur?
As 𝑐̅5 becomes positive, the solution given in Table 1
no longer remains optimal. Slack variable 𝑥5 enters the
solution. This is shown in Table 3: Result and discussion on (i)
Table 3 LP Optimum 2
𝑐𝑗 4 6 2 0 0 found at step
Objective
𝑐𝐵 Basis 𝑥1 𝑥2 𝑥3 𝑥4 𝑥5 b Function Value
8 𝑥1 1 1 1 1 0 1 1) 16.00000
0 𝑥2 0 3 6 -1 1 6 Variable Value Reduced Cost
𝐸𝑗 8 8 8 8 0 𝑥1 1.000000 0.000000
= ∑𝑐𝐵 𝑎𝑖𝑗 𝑥2 2.000000 0.000000
𝑐̅𝑗 = 𝑐𝑗 − 𝐸𝑗 0 -2 -6 -8 0
𝑥3 0.000000 6.000000
Thus the new optimal product mix is 𝑥1 = 3 dozen We observe from the above output of LINDO that if we
with 𝑧𝑚𝑎𝑥 = 𝑇𝑘. 24. change the column for non-basic variable the
That means to achieve maximum profit the company feasibility of the solution is not affected. The only
has to manufacture 3 dozen of Super Chanachur effect, if any, may be on the optimality of the solution.
without manufacturing Bar-B-Q Chanachur and Hot
Chanachur. Similarly we can analyze the change of