OR Chap 3
OR Chap 3
Duality is a property of the simplex method that adds further versatility to the general LP model. The
term dual indicates that there are two ways of looking at each problem. Associated with every LPP,
there is another intimately related LPP. The original LPP is called the primal problem while the
corresponding intimately related problem is called its dual problem. In fact either of the problems can
be considered original as both originate from the same data; consequently the other becomes its dual.
The format of the simplex method is such that when the primal is solved, its associated dual is also
solved simultaneously.
Consider the following general primal problem which is given in the form:
Max. Z = C1X1 + C2X2 + … + CnXn n
Max. Z=∑ C j X j
j=1
Subject to: Subject to:
a11X1 + a12X2 + … + a1nXn ≤ b1 n
j=1
Subject to: Subject to:
a11y1 + a21y2 + … + am1ym≥C1 m
∑ aij y i ≥C j, for j = 1, 2, …, n
i=1
a12y1 + a22y2 + … + am2ym≥C2 yi ≥ 0, for i = 1, 2, …, m
. . . .
. . . .
a1ny1 + a2ny2 + … + amnym≥Cn
y1, y2,…, ym ≥ 0
Where y1, y2,…,ym are the dual decision variables.
3.2. Economic interpretation of duality
The primal problem: with a given unit of value of each output (C j) and a given upper limit for the availability
of each input (bi), how much of each output (Xj) should be produced in order to maximize the value of the total
output.
The dual problem: with a given availability of each input (b i) and a given lower limit of unit value for each
output (Cj), what unit values should be assigned to each input in order to minimize the value of total input. The
optimal values of the dual variables are also called shadow prices, marginal value or opportunity cost.
E.g: Solve both the primal and dual of the following LPP using simplex method and observe their relation:
1
Max. Z = 20X1 + 50X2 Its dual problem is: Min. Z = 20y1 + 15y2
Subject to: Subject to:
5X1 + 4X2 ≤ 20 5y1 + 3y2 ≥ 20
3X1 + 5X2 ≤ 15 4y1 + 5y2 ≥ 50
X1, X2 ≥ 0 y1, y2 ≥ 0
2
Solution of primal problem:
Since all the Cj-Zj values are less than or equal to zero, the current solution is optimal and is given by X 1 = 0
and X2 = 3 and maximum Z is 150.
Cj 20 15 0 0 M M
CB BV y1 y2 S1 S2 A1 A2 Xb Min. ratio
M A1 5 3 -1 0 1 0 20 4
M A2 4 5 0 -1 0 1 50 12.5
Zj 9M 8M -M -M M M
Cj-Zj 20-9M 15- 8M M M 0 0
R1’ =1/5R1 and R2’= R2- 4R1’
Cj 20 15 0 0 M
CB BV y1 y2 S1 S2 A2 Xb Min. ratio
20 y1 1 3/5 -1/5 0 0 4 20/3
M A2 0 13/5 4/5 -1 1 34 170/13
Zj 20 12+13/5M -4 +4/5M -M M
Cj-Zj 0 -13/5M+3 -4/5M+4 M 0
Cj 20 15 0 0 M
CB BV y1 y2 S1 S2 A2 Xb Min. ratio
15 y2 5/3 1 -1/3 0 0 20/3 -
M A2 -13/3 0 5/3 -1 1 50/3 10
Zj 25-13/3M 15 -5+5/3M -M M
Cj-Zj -5+13/3M 0 5-5/3M M 0
3
R2”’=3/5R2” and R1”’=R1+1/3R2’”
Cj 20 15 0 0
CB BV y1 y2 S1 S2 Xb Min. ratio
15 y2 4/5 1 0 -1/5 10 15
0 S1 -13/5 0 1 -3/5 10 0
Zj 12 15 0 -3 150
Cj-Zj 8 0 0 3
Since all the Cj-Zj values are greater than or equal to zero, the current solution is optimal and is given
by y1 = 0 and y2 = 10 and minimum Z* is 150.
Primal Dual
Max. Z = 6X1 + 4X2 Min. Z* = 12y1 + 20y2 + 24y3
Subject to: Subject to:
4X1 + X2 ≤ 12 4y1 + 9y2 + 3y3 ≥ 6
9X1 + 2X2 ≤ 20 y1 + 2y2 + 5y3 ≥ 4
3X1 + 5X2 ≤ 24 y1, y2, y3 ≥ 0
X1, X2 ≥ 0
Here we should note some of the symmetrical features as follow:
1. The primal is maximization 1. The dual is minimization
2. In primal we write objective function as Z 2. In dual we write objective function as Z*
3. In primal the variables are X1 and X2 3. In dual we have new set of variables y1, y2 and y3
4. Primal has two variables X1 and X2 4. The dual, therefore, has two constraints
5. The primal has three constraints 5. The dual, therefore, has three variables
6. In primal’s objective function, 6 and 4 are 6. In dual 6 and 4 become constants of constraints on
the coefficients the right hand side.
7. In primal the coefficients of constraints 7. In dual each column takes the position row wise
() ( )
41 493
column wise are 9 2 125
35
8. In primal the signs of constraints are ≤ 8. In dual the signs of constraints are ≥
9.The non-negativity constraints are as many as 9. The non-negativity constraints are as many as the
4
the variables in the primal, i.e. 2 variables in the dual, i.e. 3
Cj - Zj 0 0 -1/2 -6/10
5
the technical coefficients (per unit resource consumption or production of each decision variables) are constants
and known with certainty. However, in realty these coefficients are subject to change with time or error.
If such changes will be occurred, there should be a means to check for how long the present optimal solution
continues as optimal. The method of evaluating the degree to which the present optimal solution is continued as
optimal is called sensitivity analysis.Sensitivity analysis is concerned with the study of ‘sensitivity’ of the
optimal solution of a LPP with changes in parameters. In this case, we are going to determine the range (both
lower and upper) over which the linear programming model parameters can change without affecting the current
optimal solution.
A. Changes in Cj of variables
One of the important parameters of LPP is the coefficient of objective function. The test of sensitivity
of the objective function involves finding the range of values with in which each C j can lie without
changing the current optimal solution.
i. Change in coefficient Cj of non-basic variable Xj
Among the variables included in the LPP some are non-basic variables. If C j is the coefficient of non-
basic variable Xj.
It doesn’t affect any of Cj values of basic variables.
It doesn’t affect any of Zj values.
It affects ∆j=(Cj-Zj) values.
If optimality is to be maintained for maximization LPP, all coefficients in the index row should be
non-positive.
Example: Max. Z = 2X1 + 3X2
Subject to:
4X1 + 3X2 ≤ 18
5X1 + 2X2 ≤ 19
X1, X2 ≥ 0
Cj 2 3 0 0
CB BV X1 X2 S1 S2 Xb Min. ratio
0 S1 4 3 1 0 18 6
0 S2 5 2 0 1 19 9.5
Zj 0 0 0 0
∆ j(Cj-Zj) 2 3 0 0
' 1
R1= R 1and R'2=R 2−2 R '1
3
Cj 2 3 0 0
CB BV X1 X2 S1 S2 Xb
3 X2 4/3 1 1/3 0 6
0 S2 7/3 0 -2/3 1 7
Zj 4 3 1 0 18
∆j(Cj-Zj) -2 0 -1 0
In the final simplex table, X 1 and S1 are non-basic variables where as X 2 and S2 are basic variables. C1 = 2 is the
'
coefficient of non-basic variable X 1, and assume that C1 is subject to change by ∆C 1. Where∆ C 1=C 1−C 1.As
you remember to maintain the optimality condition all the elements in the index row (i.e. ∆j) should be all non-
positive.
In this case:
' '
∆j(C j−Zj )≤ 0 and C 1=C1 + ∆ C1
¿ ¿) ≤ 0
6
C 1+ ∆ C1−Z 1≤ 0
2+ ∆C 1−4 ≤ 0
∆ C1≤ 2
¿
'
C 1=C1 + ∆ C1 (2+2=4)
Therefore, the range over which the parameter (C1) can change without affecting the current optimal solution X 1
= 0 and X2 = 6 is (−∞ , 4 ¿. )Simply for non-basic variableXk, the coefficient (Cj) can change without affecting
the optimal solution if Cj - Zj≤ ∆ C k .
Change in the coefficients of basic variable can affect Z j, ∆j and the value of objective function. For the above
table X2 is basic variable, and suppose C2 is changed by ∆C2i.e.C '2=3+∆ C 2.
Cj 2 3 + ∆C2 0 0
CB BV X1 X2 S1 S2 Xb
3+∆C2 X2 4/3 1 1/3 0 6
0 S2 7/3 0 -2/3 1 7
Zj 4 3 + ∆C2 1 0 18 + 6∆C2
4 + ∆ C2 1+ ∆ C 2
3 3
∆j(Cj-Zj) 4 0 1 0
−2− ∆ C2 -1− ∆ C2
3 3
'
The current solution will be optimal if ∆j ((C j−Zj) ≤ 0) values are non-positive. Therefore
4 1
−2− ∆ C2 ≤ 0∨−1− ∆ C 2 ≤ 0
3 3
3
∆ C 2 ≥− ∨∆ C 2 ≥−3
2
3 −3 3
Their intersection is ∆ C 2 ≥− or¿ .C2’=3 + ∆C2= 3+( )= . Therefore, the range over which the
2 2 2
parameter C2 can change without affecting the current optimal solution X1 = 0 and X2 = 6 is ¿ .
Exercise: For the following LPP, find the range in which the coefficients of basic variables can
change or take possible values without affecting the current optimal solution.
Max. Z = 4X1 + 3X2
Subject to:
X1 + X2 ≤ 40
5X1 + X2 ≤ 60
X1, X2 ≥ 0
If we carry out the simplex method, the optimal solution would be as follows and continue from this
on.
Cj 4 3 0 0
CB BV X1 X2 S1 S2 Xb
3 X2 0 1 2 -1 20
4 X1 1 0 -1 1 20
Zj 4 3 2 1 140
(Cj-Zj) 0 0 -2 -1
7
In LPP bi’s represent capacity or availability of resources which are critical to the selection of optimal
alternative solution.
i. Sensitivity of bi’s of fully utilized resources
If resources are fully utilized, a non-zero shadow price will appear in the ∆j row with these constants
(corresponding to slack or surplus variables). The negative of that number indicates that by how much
the value of objective function will change for one additional unit of the corresponding resource.
Based on the previous example, the optimal simplex table is:
Example: Max. Z = 2X1 + 3X2
Subject to:
4X1 + 3X2 ≤ 18
5X1 + 2X2 ≤ 19
X1, X2 ≥ 0
Cj 2 3 0 0
CB BV X1 X2 S1 S2 Xb
3 X2 4/3 1 1/3 0 6
0 S2 7/3 0 -2/3 1 7
Zj 4 3 1 0 18
∆j -2 0 -1 0
The first resource is fully utilized. Therefore, its shadow price is –(-1). That means, if we add one
additional unit of the first resource the value of Z will increase by 1 unit. Suppose the first resource
increase by one unit and become 19 (i.e., b1 = 19). The new solution would be:
( )( )
1 19
( ) ()
X2
S2
=
6
7
+1
3
=
−2 19
3
3 3
Cj 2 3 0 0
CB BV X1 X2 S1 S2 Xb
3 X2 4/3 1 1/3 0 6
0 S2 7/3 0 -2/3 1 7
Zj 4 3 1 0 18
∆j -2 0 -1 0
Therefore, the new solution would be X 1 = 0, X2 = 19/3 and Z = 19. When there is a change in the
value of bi’s of fully utilized resources, the current optimal solution will change. Therefore, there is no
need to run sensitivity analysis for fully utilized resources.
If resources are not fully utilized a zero shadow price will appear in the ∆j row with these constants
(corresponding to slack or surplus variables). The value indicates that the value of objective function
will not change for any additional unit of that resource. Here let us assume that the second resource
availability changes (i.e., becomes ∆b2).
'
b 2=b2 + ∆ b2
In order to get a feasible solution bi must be greater than or equal to zero (bi≥ 0).
8
( ) () ()
X2
S2
=
6 ∆ b 0 ≥0
7
+ 2
1
( ) ( )( )
X2
S2
=
6
7
+
∆
0 ≥0
b2
(7+6∆ b ) ≥0
2
7+∆b2 ≥ 0
∆b2 ≥ -7;∆b2 ≥[-7, ∞ ¿
'
b 2=b2 + ∆ b2=19+ (−7 )=12
Therefore, the range over which the parameter (b 2) can change without affecting the current optimal
solution (0, 6) is [ 12 , ∞ ) .
Non-Linear Programming is a mathematical technique for determining the optimal solution to many
business problems. Knowledge of differential calculus is essential to do computational work in
solving the problems. In linear programming problems, we use to deal with linear objective functions
and constraints to find the optimal solution. The constraints we have used in linear programming
technique are of ≤ or ≥ type or a combination of these two. It is also assumed in linear programming
that the cost of production or unit profit contribution or problem constraints do not vary for the
planning period and also at different levels of production. But it is only an assumption to simplify the
matter. But in real world problem the profit, requirement of resources by competing candidate all will
vary at different levels of production. Also due to many economic behaviors of demand, cost etc. the
objective function tends to be non-linear many a time. Therefore it is often we need to deal directly
with nonlinear programming.
4.2. General non-linear programming problem
If each of the firm’s products has a similar profit function, say, Pj(xj) for producing and
selling
n xj units of product j, then the overall objective function is
f(x)=∑ pi (x ¿¿ j¿)¿ ¿, a sum of non linear functions.
j=1
The application of transportation problem is to determine an optimal plan for shipping goods from
various sources to various destinations, given supply and demand constraints to minimize the
shipping cost. The assumption in the transportation problem is that the cost per unit shipped from a
given source to a given destination is fixed, regardless of the amount shipped. In actuality, the
shipping costs may not be fixed. Volume discounts sometimes are available for large shipments, so
that the marginal cost of shipping one more unit might follow a pattern like the following figure
which cause a piecewise linear cost function.
10
4.3. Graphical Illustration of Nonlinear Programming Problems
The optimal solution is no longer a corner point feasible (CPF) solution anymore. (Sometimes, it is;
sometimes, it isn’t). But, it still lies on the boundary of the feasible region. We no longer have the
remarkable simplification used in LP of limiting the search for an optimal solution to just the CPF
solutions.
What if the constraints are linear; but the objective function is not?
11
The optimal solution lies inside the feasible region (3, 3). Therefore, a general algorithm
for solving similar problems needs to consider all solutions in the feasible region, not
just those on the boundary. Another complication that arises in nonlinear programming
is that a local maximum need not be a global maximum.
The above figure is a function with several local maxima (X=0,2,4), but only X = 4 is
global maximum.
Non-linear programming algorithms generally are unable to distinguish between a local
optimal and a global optimal (except by finding another better local maximum). Therefore, it
is desired a condition under which any local optimal is guaranteed to be a global optimal.
If a non-linear programming problem has no constraint, the objective
function being concave (convex) guarantees the local maximum
(minimum) is a global maximum (minimum).
2
∂ f ( x 1 , x 2) ≥0 ≤0
2
∂x 1
2
∂ f ( x 1 , x 2) ≥0 ≤0
2
∂ x2
How to judge a multi-variables function is convex or concave?
The sum of convex functions is a convex function, and the sum of concave functions is
a concave function.
If there are constraints, then one more condition will provide the guarantee, namely,
that the feasible region is a convex set.
Convex set: is a collection of points such that, for each pair of points in the collection,
the entire line segment joining these points is also in the collection. In general, the
feasible region for a non linear programming problem is a convex set when ever all the
gi(x)(constraints)≤ bi is convex.
Condition for local maximum=global maximum (with gi(x)≥bi constraints).
To guarantee that a local maximum is a global maximum for a non linear
13
programming problem with constraint gi(x) ≤bi and xi > 0, the objective function
f(x) must be a concave function and each gi(x) must be a convex function. Such a
problem is called a convex programming problem.
4.4. Types of non-linear programming problems
Where,
14
Example: Solve Max f(x)=12x–3x4–2x6 using the bisection method.
Because of the second derivative is non-positive everywhere f(x) is concave function, so the
bisection method can be safely applied to find its global maximum (assuming global maximum
exits). A quick inspection of the equation indicates that f(x) is positive for small values of x but it
is negative for x < 0 or x > 2. Therefore, x = 0 and x can be used as the initial bounds, with their
midpoint, x’ = 1, as the initial trial solution. Let ∈=0.01 be the error tolerance for x * in the
stopping rule, so the final ( x – x) ≤ 0.02 with the final x’ at the midpoint. Applying the bisection
method then yields the sequence of the following results;
Having fixed xi at the beginning of iteration i, note that f(x i), f’(xi) and f’’(xi) are also fixed constants in
this approximating function on the right. Thus, this approximating function is just a quadratic function
of xi+1. Furthermore, this quadratic function is such a good approximation of f(x i+1) in the neighborhood
of xi that their values and their first and second derivatives are exactly the same when x i+1 = xi.
16
The above table shows the results from applying Newton’s to this example. Just after 4 iterations, this
method has converged to x = 0.83762 as the optimal solution with a very high degree of precision. A
comparison of this table with the previous table illustrates how much more rapidly Newton’s method
converges than the bisection method. Nearly 20 iterations would be required for the bisection method
to converge with the same degree of precision that Newton’s method achieved after only four
iterations. Although this rapid convergence is fairly typical of Newton’s method, its performance does
vary from problem to problem. Since the method is based on using a quadratic approximation of f(x),
its performance is affected by the degree of accuracy of the approximation.
A natural approach is to use the values of the partial derivatives to select the specific
∂f ∂f ∂f
direction in which to move. The gradient at point x=x’ is ∇ f(x)= , …... ∂ xn
∂ x1 ∂ x2
.
The direction of the gradient is interpreted as the direction of the directed line segment from
∂f ∂f ∂f
the origin to the point ( , ,… ) which is the direction of changing x that will
∂ x1 ∂ x2 ∂ x1
maximize f(x) change rate. However, normally it would not be practical to change x
continuously in the direction of ∇ f(x) because this series require continuously reevaluating
∂f
the and the changing direction of the path.
∂x
A better approach is to keep moving in a fixed direction from the current trial solution,
not stopping until f(x) stops increasing. The stopping point would be the next trial
solution and reevaluate gradient. The gradient would be recalculated to determine the
new direction in which to move.
17
E.g. Maxf(x) = 2x1x2 + 2x2- x12- 2x22
∂f ∂f
= 2x2 – 2x1, = 2x1 + 2- 4x2
∂ x1 ∂ x2
We can verify thatf(x)is concave. To begin the gradient search procedure, after choosing a suitable
small value of ∈(normally well under 0.1) suppose that x’ = (0,0) as the initial trial solution the
respective partial derivatives at this point are 0 and 2, therefore the gradient is; ∇ f(0,0) = (0, 2).With ∈
< 2, the stopping rule says to perform iteration.
Iteration 1: with values of 0 and 2 for the respective partial derivatives, the 1 st iteration starts by
setting.
1
This completes the first iteration. For this new trial solution, the gradient is ∇ f(0, ) = (1, 0). With ∈<
2
1 the stopping rule says to perform another iteration.
18
Iteration 2:To be the second iteration, use the value of 1 and 0 for the respective partial derivatives to
1 1
set x’ = (0, ) + t(1,0) = (t, ).
2 2
This completes the second iteration. With a typically small value of ∈, the procedure now would
continue on to several more iterations in a similar fashion. The above procedure can be organized in
the table as showed below;
The second column gives the current trial solution and the rightmost column shows the eventual new
trial solution, which then is carried down in to the second column for the next iteration. The fourth
column the expressions for xj in terms of t that need to be substituted into f(x) to give the fifth column.
1 3 3 3 3 7 7 7
By continuing this fashion, the subsequent trial solutions would be ( , ), ( , ), ( , ), ( , ),….as
2 4 4 4 4 8 8 8
shown in the following graph. Finally these points are converging to x* = (1,1) this solution is the
optimal solution, as verified by the fact that ∇ f(1,1) = (0,0)
The optimal solution to the Lagrange function is determined by taking partial derivatives of the
function L with respect to each variable (including Lagrange multipliers and setting each partial
derivative to zero and finding the values that make the partial derivatives zero. Then the solution will
turn out to be the solution to the original problem.
E.g.
The constraints can be modified to the form h(x 1, x2,…xn) ≤ 0 by introducing a function h (x1, x2,…xn)
= g (x1, x2,…xn) – c
Maximize Z = f (x)
Subject to
h (x) ≤ 0 and x ≥ 0 where, x∈Rn.
This problem can be slightly modified by introducing a new variable S. Define S r= – h(x) or h(x) +
S2 = 0, S can be interpreted as slack variable. It appears as its square in the constraint equation so as to
ensure its being non-negative. The problem can be restated as;
The slack variable was introduced to convert the unequal constraints to an equal one, so it may be
discarded as S2≥ 0, equation 2 gives:
h (x) ≤ 0 …….5
Whenever h(x), 0 from equation 4, we get λ = 0, whenever λ > 0 h (x) = 0. λ is unrestricted in sign
whenever h (x) ≤ 0 and the problem reduces to the problem of equation constraint. The necessary
conditions for the point x to be a point of maximum are stated as:
fj – λhj = 0 (j = 1, 2, 3, …n)
λh = 0 maximum f
h ≤ 0 subject to the constraint λ ≥ 0 and h ≤ 0.
General case of the constrained optimization of nonlinear function in n variables under m (< n)
inequality constraint.Consider NLPP;
Maximize Z = f (x). x∈Rn
Subject to
gi (x) ≤ ci where i = 1, 2, ….m and x ≥ 0.
Introducing the function hi(x) = gi(x) – ci for all i = 1, 2, ….m the inequality constraint can be written
as;
Chapter 5
5.1. Risk
Risk is situation where the exact outcome is not known, but the probabilities associated with possible
outcomes are known. Probabilities are estimated ‘objectively’ based on data collected over time.
Uncertainty is a situation where the probabilities of the outcome are established ‘subjectively.’
22
Depending on the risk attitude of the individual farmer, there are three possible types: risk-averse, risk-
neutral, and risk-taker.
A risk-averse farmer is described as one who is a cautious individual with preference for less risky
sources of income. Risk-averse individuals will sacrifice some amount of expected income to reduce
the chances of low or adverse outcomes.
A risk-neutral individual makes his decisions without regard to the risks involved. A farm of this type
doesn’t care about risks involved in farming and makes his decisions as if the risks don’t exist or are
unimportant.
Risk-takers tend on the other hand to be adventuresome with their liking for risky farm ventures and
investments which are risky and which have less than a fair chance of succeeding. They are sometimes
called gamblers. Risk attitudes vary from individual to individual. It is widely believed that most
farmers are risk-averse, even though the degree of risk version varies from individual to individual.
The most common source of risk can be described in the following categories.
1. Production & technical risk: In manufacturing firms, input & output can be known with
certainty. This is not the case with most agricultural production activities. Because of its
nature, crop & livestock performance depends upon biological processes, which are affected by
weather, soil types, weeds, pests & diseases, infertile breeding livestock & other factors which
make the yield not to be accurately predicted & cause yield variability. This creates uncertainty
about the output which will be received from any input level.Technical risk also contributes to
the problem of determining the proper input level. This creates uncertainty about input level &
corresponding output levels.
2. Marketing or price risk: A major source of risk in agriculture is price variability. That means
the prices of agricultural commodities vary seasonally within a year as well as changing from
year to year. Many forces cause price to fluctuate. For example, the supply of inputs & outputs
are affected by a combination of production decision made by many farmers & the resulting
weather, which result price variability. In general, all of the factors leading to unpredictable
shift to supply of & demand for inputs & outputs are sources of price uncertainty. Because of
the time lag in agricultural production, the price received for the output may be greatly
different from the price at the time the production decisions were made.
3. Financial risk:It occurs when money is borrowed to finance the operation of the farm. i.e. it is
the risk of losing equity due to decline in income as borrowing & the debt/equity ratio
increases. Yield & price uncertainty combine to generate financial risk or uncertainty about the
23
farm’s ability to repay debt.In other words, a combination of lower than expected prices and
yields can make debt repayment difficult, put a strain on the farm’s cash flow & possibly
reduce equity. A series of such events can result in bankruptcy.
4. Institutional risk: It refers to irregularities in the provision of services such as the supply of
credit, purchased inputs & information, from both traditional & modern institutions.The
uncertainty of government policies, programs, rules & regulations that are subject to change
create another source of uncertainty for farmers. For example, government policies regarding
to production subsidies, regulation in food quality & chemical use, rules for animal waste
disposal, the level of price or income support payments, target prices& loan rates, etc. often
change from year to year can have a major impact on the farm activities. In addition, policies
regarding inflation, unemployment, money supply, & foreign trade influence price levels &
other factors contributing to price uncertainty.
5. Technological Risk: It arises from the development & adoption of new techniques or methods
of production.In fact, new crop varieties, chemicals, feed combinations, models of machines, &
the like are continually being developed by researchers. While these new developments are
usually based on approved experimental procedures, the result realized may be different on a
given farm from those expected.It also arises from the rapidity of technical change. A new
method may be adopted, but still a better method may follow close behind, making the first
investment obsolete. For example, the first mechanical cotton pickers were soon made obsolete
by improved models. In such case, a substantial portion of the value of the machine disappears
as soon as the new model comes on the market.
6. Casual Risk:It is associated with property loss due to fire, flood, windstorm, theft, etc, which
result losses in farm business. Casual losses can generally be covered by insurance. However,
income may still be reduced by the interruption of normal business activity that often follows a
major loss.
7. Human or personal Risk: Individuals & their changeable nature also generate some
uncertainty. Farmers must deal with spouses, neighbors, bankers, suppliers, dealers, & land
lords, any of whom can change their attitude, policy, or business relationship. Accidents,
illness & death, for example often threaten & disrupt farm performance.
Decision-making under risk describes a situation in which each strategy results in more than one
outcome or payoffs and the manager attaches a probability measure to these payoffs. This model
covers the case when the manager projects two or more outcomes for each strategy and he or she
knows, or is willing to assume, the relevant probability distribution of the outcomes. The optimal
24
strategy in decision making under risk is identified by the strategy with highest expected utility (or
highest expected value).
The existence of risk & uncertainty adds complexity to many problems of a farm and to the decision-
making process. Yet decision must be made with consideration of risk. Some measures of assessing
risk are described below even though they are not exactly precise methods;
I. Forming Expectation
The manager must form expectation about the input & output price & somehow arrive at an expected
value of use in the decision making process. Several methods can be used to form an expectation about
the future prices, yields, & other values which are not known with certainty. Some of the methods used
to form expectation are:
A. Simple average: - A series of actual past prices or yields can be used to find the simple
average over some specified time period. Relatively simple method to use if the past data are
available. The primary problem is selecting the length of the data series to use in calculating
simple average (i.e. 3, 5, or 10 years). The choice is usually depend on the subjective estimate
of the decision maker.
B. Weighted average:-this method usually weights the more recent values heavier than the older
once using some predetermined weighing system. Two problems are inherent in this system:
i. How many years should be used?
ii. What weighting system is best?
The experiences & preferences of the decision maker along with knowledge of the data provided the
only guidelines that can be easily applied.
3. Mathematical expectation: - when either the true or subjective probabilities of the expected
outcomes are available, it is possible to calculate the mathematical expectation. The mathematical
expectation is the average outcome of conducting the experiment or repeating the event many times.
b. The mathematical expectation is 29.4 qt/ha and less than the most likely yield of 32qt because the
probabilities are not systematical, being skewed toward the lower yields.
II. Variability
It represents dispersion of possible outcomes around expected value. If two alternatives have the same
expected values, most managers would choose the one whose potential outcomes have the least
variability. Different measures of variability are:
i. Range:-is the difference between the lowest & the highest possible outcomes. Alternatives
with a smaller range are preferred over those with a wider range provided that their
expected values are the same. It is not a good measure of variability because it does not
consider the probabilities associated with the extreme values in the distribution of
outcomes.
ii. Standard Deviation: is a common statistical measure of variability which can be
calculated from a continuous probability distribution or from a random sample of values
drawn from the total population of some random variable.
26
Population standard deviation= √∑ (x−μ) 2
or
N
n−1
A larger standard deviation indicates a greater dispersion of possible outcomes and, therefore, a
greater possibility that the actual outcome will be further from the mean or expected values.
3. Coefficient of variation: -It measures variability relative to the expected value or mean of the
probability distribution.
Standard devation
Coefficient of variation=
Expected value∨mean
Smaller coefficients of variation mean the distribution has less variability in relation to its expected
value than other distributions.
27
Strategy: is a rule or plan of action for playing a game of a specific course of action with clearly
defined values for the policy variables.
Strategic interaction can involve many players and many strategies. For simplicity let us limit our
discussion to two-person games with finite strategies. This enables us to present the game using a
payoff matrix. The payoff matrix is a table showing the payoffs to each player given her/his decision
and the decision of the opponents. The payoff matrix of a firm is a table showing the payoffs accruing
to his firm as a result of each possible combination of strategies adopted by it and by its rival(s).
When every player has a dominant strategy, we call the outcome of the game equilibrium in
dominant strategieswhich means outcome of a game in which each player is doing the best he can,
regardless of what his opponents are doing.
E.g. suppose that two persons A and B are playing a simple game. Person A writes one of the two
words on a piece of paper “top” or “bottom”. Simultaneously, person B will write independently
“left” or “right” on a piece of paper. When the paper is examined each player gets the
payoffs/benefits depicted below.
Player B
Left Right
Player A Top 1,2 0,1
Bottom 2,1 1,0
28
Person A has 2 strategies: he can choose top or bottom. For person A it is always better to choose
“bottom” since his payoff from that choice (2 or 1) is greater than the corresponding entries in the
top (1 or 0).
Person B has 2 strategies: he can choose left or right. It is always better for B to choose “left” since 2
and 1 dominate1 and 0. We would expect that the equilibrium strategy is for A to play bottom and
for B to play left. Therefore in this case an equilibrium outcome is the one in which A plays “ bottom”
receiving an equilibrium payoff of 2 and B plays “left,” receiving an equilibrium payoff of 1.
But not every game has a dominant strategy for each player. As a result, we need a more general
equilibrium concept. The concept of Nash equilibrium is widely applicable and intuitively appealing in
the absence of dominant strategies.
Nash Equilibrium
Nash equilibrium refers to a situation where each firm is doing the best it can, given what its
competitors are doing. Dominant strategy equilibrium is a special case of Nash equilibrium.
Player B
Left Right
Player A Top 2,1 0,0
Bottom 0,0 1,2
Here, if B chooses left, A wants to choose top, if B chooses right, A chooses bottom. If A chooses top,
B chooses left, and if A chooses bottom B chooses right. This implies A's optimal choice depends on
what it thinks B chooses, and what is optimal for B will depend on A's choice. It means that rather
than requiring that A's choice be optimal for all choices of B, we just require that it be optimal for
optimal choices of B. A pair of strategy is a Nash equilibrium if A's choice is optimal given B's choice,
and B's choice is optimal given A's choice. It can also be interpreted as a pair of expectations about
each person's choice such that, when the other person's choice is revealed, neither individual wants
to change his behavior. In the above example (Top, Left) and (Bottom, Right) are Nash equilibriums.
Limitations to the concept of Nash equilibrium;
A game may have more than one Nash equilibrium
There are games that have no Nash equilibrium.
Mixed Strategies:
If each agent is making a decision and sticks to it, this situation is called pure strategy. We can also
think of a case where agents may randomly select their strategies, i.e. assign a probability to each
choice and play their choices according to those probabilities. This kind of strategy is called a mixed
strategy. Example:consider the following payoff matrix
29
Player B
Left Right
Player A Top 0,0 0,-1
Bottom 1,0 -1,3
Let A choose to play top 50% of the time and bottom 50% of the time. Player B chooses to play left
50% of the time and right 50% of the time. This means they will have a probability of ¼ of ending up
in each of the 4 cells of the payoff matrix. Thus,
Expected payoff of A = (1/4 X 0) + (1/4 X 0) + (1/4 X1) + [1/4 X (-1)] = 1/4-1/4 = 0
Expected payoff of B = (1/4X0) + (1/4X0) + [1/4X (-1)] + (1/4X3) = (-1/4) + ¾ = 2/4 = ½
Nash equilibrium in mixed strategies refers to an equilibrium in which each agent chooses the
optimal probability with which to play his strategies given the probability choices of the other agent.
Nash equilibrium in mixed strategies always exists for the above type of game.
The Prisoners’ Dilemma
A classic example in game theory, called the prisoners’ dilemma, illustrates the problem faced by
oligopolistic firms. It goes as follows: Two prisoners have been accused of collaborating in a crime.
They are in separate jail cells and can’t communicate with each other.
Each has been asked to confess.
If both prisoners confess, each will receive a prison term of five years.
If neither confesses, the prisoners’ case will be difficult to make, so the prisoners can expect
to plea bargain and receive terms of two years.
On the other hand, if one prisoner confesses and the other does not, the one who confesses
will receive a term of only one year while the other will go to prison for 10 years.
As stated above, the prisoners face a dilemma. If they could both agree not to confess, then each
would go to jail for only 2 years. If prisoner-A does not confess, he risks being taken advantage of
by his former accomplice. After all, no matter what prisoner A does, prisoner B comes out ahead
by confessing. Likewise prisoner A always comes out ahead by confessing, so prisoner B must
worry that by not confessing, he will be taken advantage of. So both prisoners will probably
confess and go to jail for five years.
The payoff matrix given by the table below summarizes the possible outcomes;
Player B
Confess Don’t confess
Player A Confess -5, -5 -1, -10
Don’t confess -10, -1 -2, -2
The ideal outcome is one in which neither prisoner confesses, so that they both get 2 years in prison.
However, confessing is a dominant strategy for each prisoner, since it yields a high payoff regardless
of the strategy of the other prisoner. The strategy (Don’t confess, Don’t confess) is Pareto-efficient
30
since there is no other strategy choice which makes both players better off while the strategy
(Confess, Confess) is Pareto-inefficient.
Parametric Methods: Parametric methods are based on certain functional form assumptions for the
efficient frontier. Parametric approaches are divided into deterministic and stochastic models. In
deterministic models, all observations by frontier and existing technology are enveloped as technical
inefficiency by determining the difference between observed production and maximum production.
The most widely used method in the parametric approach is Stochastic Frontier Analysis.
31
CHAPTER 6
6.1. TRANSPORTATION PROBLEM AND ITS APPLICATION
Transportation problem is a special type of linear programming which is applied in the area of physical
distribution (transportation) of goods from several supply centers (origins) to several demand centers
(destinations). The transportation problem, received this name because many of its applications involve
determining how to optimally transport goods.
It is easy to express a transportation problem mathematically in terms of an LP model, which can be solved by
the simplex method. However, since it involves a large number of variables and constraints, it takes a long time
to solve it. Therefore, other methods (transportation algorithms) have been developed for this purpose. The
structure of transportation problem involves a large number of shipping routes from several supply origins to
several demand destinations.
The objective of transportation problem is to determine the amount of commodities which should be
transported from several sources to different destinations, at the minimum transportation cost and/or time. The
places where goods originate from (like plants, warehouse, etc.) are called the sources or the origins and places
where they are to be shipped are called the destinations.
The transportation algorithm applies to minimize the total cost of transporting a homogenous commodity
(product) from supply origins to demand destinations. However, it can also be applied to the maximization of
some total value or utility, in such a way that the profit is maximized.
A transportation problem model, which has ‘m’ sending locations (origins) and ‘n’ receiving locations
(destinations) provides a framework for presenting all relevant data. These are:
32
1. Quantity supply of each origin(SSi)
2. Quantity demand of each destination(DDi)
3. Unit transportation cost from each origin to each destination(Cij)
Destination
To D1 D2 … Dn Total
From Supply
S1 X11 C11 X12 C12 … X1n C1n SS1
Source (origins)
: : : : : :
Sm Xm1 Cm1 Xm2 Cm2 … Xmn Cmn SSm
a. All goods are homogeneous, so that any origin is capable of supplying to any destination.
b. Transportation costs are a linear function of (or directly proportional to) the quantity
shipped over any route.
c. Each source has a fixed supply of units, where this entire supply must be distributed to the
destinations. Similarly, each destination has a fixed demand for units, where this entire
demand must be received from the sources.
33
As a remark we have to know that before applying the transportation algorithm to solve a specific problem it is
necessary to satisfy the following conditions:
E.g. A Company has three production facilities S1, S2 and S3 with production capacity of 7, 9 and 18 units per
week of a product respectively. These units are to be shipped to four warehouses D1, D 2, D3 and D4 with
requirements of 5, 8, 7 and 14 respectively. The transportation cost (in Birr) per unit between the warehouses is
given in the table below.
Destination
To D1 D2 D3 D4 Total
Sourc
From Supply
34
S1 X11 19 X12 30 X13 50 X14 10 7
e
S3 X31 40 X32 8 X33 70 X34 20 18
Total 5 8 7 14 34
Demand 34
In this method we do not consider the unit cost of transportation. The North West Corner cell X 11 is chosen for
allocation. The origin S1 has 7unitsand destination D 1requires only 5 units. Hence it is enough to allot 5 items
from S1 to D1.
The origin S1 which is alive with 2more units can supply to the destination to the right of D 1 namely D2 whose
requirement is 8. So, we supply 2 units to D 2 thereby the origin S 1 is exhausted. D2 requires 6 units more. Now
consider the origin S2 that has 9 units to spare. We allot 6 unitsto the cell (X 22) so that D2 is completed,the origin
S2 is left with 3 items can supply to the destination to the right of D 2 namely D3(X23) whose requirement is 7 and
S2 is exhausted.
Now D3 requires 4 more units to satisfy its requirement 7 units therefore, move to origin S 3 which has 18 units
to supply and supply 4more units to the destination D 3. Now the requirement of the destination D 3 is complete
and S3 is left with 14 items and the same can be allotted to the destination D 4. Now the origin S3 is emptied and
the requirement at the destination D3 is also complete. This completes the initial solution to the problem.
Destination
To D1 D2 D3 D4 Total
From Supply
S1 5 19 2 30 50 10 7
Source
S2 70 6 30 3 40 60 9
S3 40 8 4 70 14 20 18
Total 5 8 7 14 34
Demand 34
The total cost of transportation by this method will be:(5 × 19)+ (2× 30)+(6× 30)+ (3× 40)+ (4× 70)+ (14 ×20)=
95 + 60 + 180 + 120 + 280 + 280 = Birr1015.
The least-cost method finds a better starting solution by concentrating on the cheapest routes. The method
assigns as much as possible to the cell with the smallest unit cost (ties are broken arbitrarily). Next, the satisfied
row or column is crossed out and the amounts of supply and demand are adjusted accordingly. If both a row and
35
a column are satisfied simultaneously, only one is crossed out, the same as in the northwest-corner method.
Next, look for the uncrossed-out cell with the smallest unit cost and repeat the process until exactly one row or
column is left uncrossed out.
Destination
To D1 D2 D3 D4 Total
From Supply
S1 19 30 50 10 7
Source
S2 70 30 40 60 9
S3 40 8 70 20 18
Total 5 8 7 14 34
Demand 34
We examine the rows S1, S2 and S3, 8 is the least cost element in the cell (S 3, D2). The origin S3 supply 8units to
D2 and alive with 10 units and D 2 is completed so shade column D 2, we observe that 10 is the least element in
the cell (S1, D4) and examine the supply at S1 and demand at D4. S1 has 7 unitsso, we supply 7 units to D 4
thereby the origin S1 is exhausted. Hence, shade the row S1.
Shading S1, we observe that 20 is the least element in the cell (S 3, D4).The destination D4receive 7 units from S1
above so it left with 7 more requirements to be equal14units and this requirement can be satisfied from S 3 so
that the column D4 is completed so shade it. Next, 40 is the least cost element in the cell (S 3, D1) and (S2, D3)
and the tie can be broken arbitrarily. Select (S 2, D3). We supply 7 units from S 2 to D3 so D3 is completed and
shad it. The next least cost element is 40 (S 3, D1), D1 requires 5 units and it can receive 3 units from S 3 and 2
unitsfrom S2. Now all demands are completed and all supplies are exhausted.
Destination
To D1 D2 D3 D4 Total
From Supply
S1 19 30 50 7 10 7
Source
S2 2 70 30 7 40 60 9
S3 3 40 8 8 70 7 20 18
Total 5 8 7 14 34
Demand 34
The cost of the allocation by the least cost method is (7 x 10) + (2 x 70) + (7 x 40) + (3 x 40) + (8 x 8) + (7 x
20) = 70 + 140+ 280 + 120 + 64 + 140 = 814. The quality of the least-cost starting solution is better than that of
the north-west corner method because it yields a smaller value oftransportation cost.
36
(iii) Vogel Approximation Method
In this method, we use concept of opportunity cost. Opportunity cost is the penalty for nottaking correct
decision. To find the row/column opportunity cost in the given matrix deduct thesmallest element in the
row/column from the next highest element. Write row opportunity costs of each row just by the side of
availability constraintand similarly write the column opportunity cost of each column just below the
requirementconstraints. These are known as penalty column and penalty row.
The rationale in deducting the smallest element form the next highest element is:Let us say the smallest element
is 3 and the next highest element is 6. If we transport one unitthrough the cell having cost Br3, the cost of
transportation per unit will be Br 3.Instead we transport through the cell having cost of Br6, then the cost of
transportationwill be Br 6 per unit. That is for not taking correct decision; we are spending Br3 more(Br.6 –
Br.3 = Br.3). This is the penalty for not taking correct decision and hence theopportunity cost. This is the
lowest opportunity cost in that particular row or columns as we are deducting the smallest element form the
next highest element.
Note: If the rowor column has two elements of the samemagnitude as the smallest element then the
opportunity cost of that row or column iszero.
Steps in VAM:
Step 1: For each row (column), determine a penalty measure by subtracting the smallest unit cost element in the
row (column) from the next smallest unit cost element in the same row (column).
Step 2: Identify the row or column with the largest penalty. If there is a tie (equal penalty) it can be broken by
selecting the cell where maximum allocation can be made. Allocate as much as possible to the variable with the
least unit cost in the selected row or column. Adjust the supply and demand, and cross out the satisfied row or
column. If a row and a column are satisfied simultaneously, only one of the two is crossed out, and the
remaining row (column) is assigned zero supply (demand).
Step 3: (a) If exactly one row or column with zero supply or demand remains uncrossed out, stop.
(b) If one row (column) with positive supply (demand) remains uncrossed out, determine the basic
variables in the row (column) by the least-cost method. Stop.
(c) If all the uncrossed out rows and columns have (remaining) zero supply and demand, determine the
zero basic variables by the least-cost method. Stop.
(d) Otherwise, go to step 1.
Step 4: Make sure that all the rim conditions are satisfied and (m+n−1) cells are allocated.
Let us apply Vogel's Approximation Method to the above example,
Destination
To D1 D2 D3 D4 Total
From Supply
S1 19 30 50 10 7
Source
S2 70 30 40 60 9
S3 40 8 70 20 18
Total 5 8 7 14 34
Demand 34
Calculate thedifference betweenthesmallest element in each row and in each columnfrom the next highest
element in each row and in each column. We choose the maximum from among the differences. The first
37
individual allocation will be to the smallest cost of a row or column with the largest difference. So we select the
column D2(penalty = 22) for the first individual allocation, and allocate to (S 3, D2) since this cell has the least
cost location. Thus 8 units from S1are allocated to D2 and D2 is completed and S3 is alive with 10 more units.
The next step is to shad out row D 2 (as it is completed) and observe the next largest difference corresponds to
the row D1(21). This leads to an allocation in the corresponding minimum cost location cell (S 1, D1). We can
allocate5 units from S1 to D1 and D1 is completed andS1 is alive with 2 more units.
The next step is to shad out row D 1 (as it is completed) and observe the next largest difference corresponds to
the row S3 (50). S3 can supply 10 units to D4 at (S3, D4) and exhausted so shade it and observe the maximum
difference again which corresponds to column D 4 (40) which can receive 2 units from S 1at (S1, D4)(least cost in
the column). Now S1 is exhausted and shaded. Now, we have only one row S2and two columns D3 and
D4indicating that the entire available amount from S2 has to be moved to D3 and D4as per their requirements.
S2 70 30 7 40 2 60 9 10 20 20 20
S3 40 8 8 70 10 20 18 12 20 50 -
(3rd)
Total 5 8 7 14 34
Demand 34
Column 21 22(1st) 10 10
penalty
21(2nd) - 10 10
- - 10 10
- - 10 50(4th)
The cost of allocation (i.e., the associated objective value) by Vogel's Approximation Method will be: (19 × 5)
+ (2× 10) + (7× 40) + (2× 60) + (8× 8) + (10× 20) = 95 + 20+280 + 120+ 64 + 200= 799.
Now let us compare the three methods of getting basic feasible solution:
North – west corner method. least cost method Vogel’s Approximation Method
1. The allocation is made from the The allocations are madedepending The allocations are made
left hand side topcorner irrespective on the cost of the cell. Lowest cost is depending on the opportunity cost
of the costof the cell. first selected and then next highest of the cell.
etc.
2. As no consideration isgiven to the As the cost of the cell is considered As the allocations are made
cost of the cell,naturally the while making allocations, the total depending on the opportunity cost
totaltransportation cost will behigher cost of transportation will be of the cell, the basic feasible
than the othermethods. comparatively less. solution obtained will be very
nearer to optimal solution.
3. It takes less time. Thismethod is The basic feasible solution, we get will It takes more time for gettingbasic
38
suitable to getbasic feasible solution be very nearer tooptimal solution. It feasible solution. But thesolution
quickly. takes moretime than northwest we get will be verynearer to
coronermethod. optimal solution.
4. When basic feasiblesolution alone When optimal solution is asked, better VAM and MODI is the best option
is asked, it isbetter to go for to go for LC method for basic feasible to get optimal solution.
northwestcorner method. solution and MODI for optimal
solution.
In the above problem given, the total cost of transportation for NWC is Birr1015, the total cost of transportation
for LCM is Birr 814 and that of VAM is Birr 799. The total cost got by VAM appears to be less. That of
Northwest coroner method is highest. Now let us discuss the method of getting optimal solution or methods of
giving optimality test for basic feasible solution.
Once, we get the basic feasible solution for a transportation problem, the next duty is to test its optimality.
AnOptimal solution is a solution which is the best among all feasible solution to achieve the objective function.
It is the one where there is no other set of transportation routes (allocations) will further reduced the total
transportation cost.
This can be done by two methods by Stepping Stone Method, and by Modified Distribution Method, or MODI
method.
Let us take the basic feasible solution we got by VAM in the previous e.g. and give optimality test to it by
stepping stone method.
Destination
To D1 D2 D3 D4 Total
From Supply
S1 5 19 30 50 2 10 7
Source
S2 70 30 7 40 2 60 9
S3 40 8 8 70 10 20 18
Total 5 8 7 14 34
Demand 34
S.no Empty cell Loop formation Change in cost (improvement index)
Destination
40
To D1 D2 D3 D4 Total
From Supply
S1 5 19 30 50 2 10 7
Source
S2 70 + 30 7 40 - 2 60 9
S3 40 - 8 8 70 + 10 20 18
Total 5 8 7 14 34
Demand 34
To select the leaving variable observe the minus sign in the loop and take the variable with the lowest shipment
or allocation which is S2, D4 for current case. The next is to generate the new solution. Add the shipment
amount of the leaving variable (2 in this case) to each plus sign and subtract this amount from each minus sign
in the loop. Thus new solution is;
Destination
To D1 D2 D3 D4 Total
From Supply
S1 5 19 30 50 2 10 7
Source
S2 70 2 30 7 40 60 9
S3 40 6 8 70 12 20 18
Total 5 8 7 14 34
Demand 34
Then check the optimality of the current solution;
Empty cell Loop formation Change in cost (improvement index)
Now the solution is optimal (all values of improvement index are positive).
Therefore the improved solution using stepping stone method is (19 × 5) + (2× 10) + (2× 30) + (7× 40) + (6×
8) + (12× 20) = 95 + 20+60 + 280+ 48 + 240= 743
One of the drawbacks of stepping stone method is that we have to write a loop for every emptycell. Hence it is
tedious and time consuming. Hence, for optimality test we use MODI method ratherthan the stepping stone
method.
41
A basic feasible solution is optimal if and only if C ij −ui−v j ≥0 for every (i, j/row, column/) such that X ij is
non-basic (empty cell). Thus, the only work required by the optimality test is the derivation of the values of ui
and v jfor the current basic feasible solution and then the calculation of these C ij −ui−v jcan be described. Since
C ij −ui−v jis required to be zero, if X ij is a basic variable, uiand v jsatisfy the set of equations C ij −ui−v jfor
each (i, j) such that X ij is basic. There are m−n−1 basic variables (occupied cell).Since the number of
unknowns (the ui and v j) is m−n, one of these variables can be assigned a value arbitrarily without violating the
equations. The choice of this one variable and its value does not affect the value of any C ij −ui−v j, even when
X ij is non-basic, so the only (minor) difference it makes is in the ease of solving these equations. A convenient
choice for this purpose is to select the ui ,that has the largest number of allocationsin its row(break any tie
arbitrarily) and to assign to it the value zero. Because of the simple structure of these equations, it is then very
simple to solve for the remaining variables algebraically. If the values C ij −ui−v j is negative we conclude that
the current basic feasible solution is not optimal.
Let us take the basic feasible solution we got by VAM in the previous e.g. again and give optimality test to it by
MODI method.
Destination
To D1 D2 D3 D4 Total
From Supply
S1 5 19 30 50 2 10 7 u1
Source
S2 70 30 7 40 2 60 9 u2
S3 40 8 8 70 10 20 18 u3
Total 5 8 7 14 34
Demand 34
v1 v2 v3 V4
a. Basic variables
Assign any one of the ui’s or vj’s the value zeroarbitrarily. Here u1 is assigned zero.
C 11−u1 −v 1=0 , 19−u1−v 1 =0, let u1 = 0, v1 = 19
C 14−u1−v 4 =0, 10−0−v 4=0, v 4= 10
C 34−u3−v 4 =0, 20−u3−10=0, u3 = 10
C 32−u3−v 2=0, 8−10−v 2=0, v 2= -2
C 24−u2−v 4 =0, 60−u2−10=0, u2= 50
C 23−u2−v 3=0, 40−50−v 3=0, v 3 = -10
b. Non-basic variables
42
C 21−u2−v 1=0, 70−50−19=0, = 1
C 22−u2−v 2=0, 30−50−(−2 ) =¿-18
C 31−u3−v 1=0, 40−10−19=¿ 11
C 33−u3−v 3=0, 70−10−(−10) = 70
According to the optimality criterion for cost minimizing transportation problem, the current solution is not
optimal, since the opportunity costs of the unoccupied cells are not all zero or positive. The value -18 is
indicating that the total transportation cost can be reduced by shifting an allocation to this cell. If we do have
more than one negative, we select the one with the larger in absolute terms.
The next step is tracing a closed loop from this cell and to check the optimality.
Destination
To D1 D2 D3 D4 Total
From Supply
S1 5 19 30 50 2 10 7
Source
S2 70 + 30 7 40 - 2 60 9
S3 40 - 8 8 70 + 10 20 18
Total 5 8 7 14 34
Demand 34
Destination
To D1 D2 D3 D4 Total
From Supply
S1 5 19 30 50 2 10 7
Source
S2 70 2 30 7 40 60 9
S3 40 6 8 70 12 20 18
Total 5 8 7 14 34
Demand 34
c. Basic variables
Assign any one of the ui’s or vj’s the value zero. Here v1 is assigned zero.
Therefore the improved solution using MODI method is (19 × 5) + (2× 10) + (2× 30) + (7 × 40) + (6× 8) + (12×
20) = 95 + 20 +60 + 280 + 48 + 240 = 743 which is the same as stepping stone method. And the opportunity
costs of empty cells are also the same for both. Hence the combination of VAM and MODI can be conveniently
used to solve the transportation problem when optimal solution is asked to minimize a tedious work in stepping
stone method.
1.1.3. Special cases in transportation problem
1. Unbalanced supply and demand
For a feasible solution to exist, it is necessary that the total supply must equal to demand. That is, total supply =
total demand. However there may be a situation in which a total supply not equal to total demand. The
following are the two cases of this situation;
a. If total supply exceeds total demand (SS > DD), and to change this unbalanced case into its balanced
form we should introduce dummy receiver. These shipments indicate unused supply capacity.
Customers
To D1 D2 D3 D4 D5 Total Supply
From
S1 X11 3 X12 6 X13 8 X14 5 X15 4 10
Total 20 20 15 10 5 75
Demand 70
Customers
To D1 D2 D3 D4 D5 R Total
D6
From Supply
S1 X11 3 X12 6 X13 8 X14 5 X15 4 10
Total 20 20 15 10 5 5 75
Demand
75
b. If total demand exceeds total supply (SS < DD), and to change this unbalanced case into its
balanced form we should introduce dummy supplier. These shows unmet demand.
Customers
To D1 D2 D3 D4 D5 Total
From Supply
S1 X11 3 X12 6 X1 8 X14 5 X15 4 10
Source
44
S3 X31 7 X32 8 X3 3 X34 9 X35 3 5
3
Total 20 20 15 10 5 65
Demand 70
S 5
S5
Total 20 20 15 10 5 70
Demand 70
2. Degeneracy
This condition occurs when the No of occupied cells in any solutions less than the N o of rows plus the No of
columns minus 1 in a transportation table.
i.e. Noof occupied calls <m+n -1 .....................Degeneracy
In such cases, the current solution cannot be improved because it is not possible to draw a closed path for every
occupied cell. In addition the values U i and Vj which are used to test the optimality cannot be computed. Thus,
we need to remove the detergency to test and improve the given solution.
The degeneracy in the transportation problems may occur at two stages:When obtaining an initial solution or
during improvement (at any stage while moving towards optimal solution).
The purpose of epsilon/delta is to enable evaluation of the remaining empty cells. The choice of location for the
epsilon/delta can be somewhat tricky: some empty cells may be unsuitable if they do not enable evaluations of
remaining empty cells. Not all choices would be acceptable.Actually, the No of epsilon/deltas needed will equal
the difference between the No of completed cells and m+n-1.Howerver; you will only be exposed to the most
common case in which one more completed cell is needed.
45
The epsilon/delta cannot be placed in a cell which later turns out to be in a negative position of a cell path
involved in reallocation because epsilon/delta will be the “smallest quantity a negative position “ and shifting
that minute quantity around the cell path will leave the solution virtually unchanged. Consequently, a certain
amount of trial and error may be necessary before a satisfactory location can be identified for epsilon/delta.
To resolve degeneracy which occurs during optimality test, the quantity() may be allocated to one or more
cells which have become unoccupied recently to have m+n-1 number of occupied cells in the new solution.
80
demand 50 30
80
Solution:
Using NWCM and MODI, the initial solution is:
1 2 Supply
3 3
1 50
50
4 6
2 30
30
Demand 50 30 80 80
Note: m=2 and n=2==>2+2-1=3==>Occupied cells=2< 3 (Degeneracy).
1 2 Supply Ui
3 3
1 50 U1=0
50
4 6
2 30 U2=3
30
Demand 50 30 80 80
Vj V1=3 V2=3
Test of optimality
1 2 Supply
3- 3 +
1 50
50
4+ 6 46
2 30
30 -
Demand 50 30 80 80
The new allocation is
1 2 Supply Ui
3 3
1 20 30 50 U1=0
4 6
2 30 30 U2=1
Demand 50 30 80
Vj V1=3 V2=3
Test of optimality
The existence of alternative optimal solution can be determined by an inspection of the opportunity costs (net
contribution of empty cells), ∆ ij for the unoccupied cells. If theunoccupied cell in the optimal solution has
opportunity cost of zero, an alternative optimal solution can be formed with another set of allocations without
increasing the total transportation cost.
Destination
To D1 D2 D3 D4 Total
From Supply
S1 4 76 8 8 0 76 u1=-16
Source
S2 16 21 24 41 16 20 0 82 u2=0
S3 72 8 5 16 24 0 77 u3=-8
Total 72 102 41 20 235
Demand 235
v1=16 v2=24 v3=16 V4=0
47
C 21−u2−v 1=16−0−16=0
C 33−u3−v 3=24−(−8 )−16=16
C 34−u3−v 4 =0− (−8 )−0=8
Thenet contributions of all empty cells are positive except for the cell C 21 which has a zero. This means if cell
C 21entered in to a basis, no change in the transportation cost would occur. To determine this alternative
solution, form a closed path for cell C 21as shown in the table below.
Destination
To D1 D2 D3 D4 Total
From Supply
S1 4 76 8 8 0 76 u1=-16
S2 16 21 24 41 16 20 0 82 u2=0
Source
+ -
S3 72 8 5 16 24 0 77 u3=-8
- +
Total 72 102 41 20 235
Demand 235
v1=16 v2=24 v3=16 V4=0
The maximum quantity which can be allocated to cell C 21is 21. After this change the new solution is:
Destination
To D1 D2 D3 D4 Total
From Supply
S1 4 76 8 8 0 76 u1=-16
S2 16 24 41 16 20 0 82 u2=0
Source
21
S3 51 8 26 16 24 0 77 u3=-8
Total 72 102 41 20 34
Demand 235
v1=16 v2=24 v3=16 V4=0
Since all ∆ ij values are positive or zero, the solution given in the table is optimal with a minimum total
transportation cost of 2424 which is the same as in the previous solution.
4. Maximization transportation problem
In general, transportation model is used for cost minimization problems. However, it is also used to solve
problems in which the objective is to maximize total value or benefit. That is instead of unit cost C ij, the unit
48
profit or payoff Pij associated with each route, (i, j) is given. The objective is to maximize the total profit for
which the profit matrix is given.
m n
MaxZ =∑ ∑ PijXij
i=1 j=1
For this we have to convert the maximization problem in to minimization problem (profit matrix in to loss
matrix) by subtracting all the elements (profit) from the highest element (profit) in the given transportation
table. This modified minimization problem can be solved in the usual manner. i.e. the algorithm for solving
modified problem is the same as for the minimization problem.
E.g. Solve the following transportation problem to maximize the profit.
Destination
To D1 D2 D3 D4 Total Supply
From
S1 15 51 42 33 23
Source
S2 80 42 26 81 44
S3 90 40 66 60 33
Total 23 31 16 30 100
Demand 100
Solution: subtracting all the elements (profit) from the highest element (profit) 90 in this casethen the modified
minimization transportation problem becomes,
Destination
To D1 D2 D3 D4 Total Supply
From
S1 75 39 48 57 23
Source
S2 10 48 64 9 44
S3 0 50 24 30 33
Total 23 31 16 30 100
Demand 100
Then using VAM method the solution becomes,
Destination
To D1 D2 D3 D4 Total Supply
From
S1 75 23 39 48 57 23
Source
S2 6 10 8 48 64 30 9 44
S3 17 0 50 16 24 30 33
Total 23 31 16 30 100
Demand 100
And check the optimality
Let u2=0, then v1 = 10, v2 = 48, v3 = 34, v4 = 9, u1 = -9, u3 = -10
Finally check the opportunity cost of unoccupied cells
C11-u1-v1=75-(-9) – 10 = 74, C13-u1-v3=48-(-9) – 34 = 23, C14-u1-v4=57-(-9) – 9 = 57
C23-u2 –v3 = 64- 0– 34 = 30, C32-u3-v2=50 –(-10) – 48 = 12 and C34-u3-v4=30-(-10) – 9 = 31
49
AllCij – ui– vjvalues are positive indicating the solution is optimal. Then to find the maximization problem
solution put the allocated value of the modified minimization problem solutions to the original maximization
transportation problem as given below.
Destination
To D1 D2 D3 D4 Total Supply
From
S1 15 23 51 42 33 23
Source
S2 6 80 8 42 26 30 81 44
S3 17 90 40 16 66 60 33
Total 23 31 16 30 100
Demand 100
50
Total profit = 23x51 + 6x80 + 8x42 + 30x81 + 17x90 + 16x66 =Br7,005
51