Constrained and Unconstrained Optimization

Prof. Adrijit Goswami

Department of Mathematics
Indian Institute of Technology, Kharagpur

Lecture – 03
Geometry of LPP

So, in the last class we have started the formulation of LPP. We have just told how LPP
should be written.

(Refer Slide Time: 00:32)

Now, let see what the assumptions of LPP are, assumptions of LPP. One is
proportionality. This proportionality means you have the decision variable or .
Whatever you say the decision variable is associated with the cost function is . In
the objective function the contribution of the decision variable is . Whereas for
the ith constraint contribution is . So, basically a decision variable is connected
with the objective function with the variables . And in the ith constraint with the
coefficient . So, whenever value of changes there will be effect on the decision
variable as well as on the coefficient or the constraint; that means, if one increase other
will also decrease. Or in other sense we can say that we there is no need of giving extra
proportionality or extra weightage to this particular variable .
So, here we want to say is that proportionality means if I have variables
then all these variables will be given the equal weightage, no one will be given any
extra weightage. The meaning is each variable will be given the equal proportionality or
equal weightage. The next one is additivity, means you have the total cost; basically the
total cost is nothing but the sum of the individual costs of the components.

Because total cost is nothing but . So, the total cost is the sum of individual
component cost and on the same way the total contribution of the ith restriction if you
think about the ith constraint or ith constraint or restriction. In that case, the total
contribution of the ith constraint is equal to the sum of the individual contribution of the
activities . So, basically the total cost is nothing but the sum of individual costs, and
the ith constraint the contribution of ith constraint is equal to the sum of the ith sum of
the activities of the ith constraint. And this means that no need of any substitution or
introduction effect will be required.

So, for these activities the meaning is that no substitution or interaction effect will be
required. This is additivity, the next one is divisibility. Divisibility means whenever you
are taking about the variable division variable , I if I want I can reduce the value of
these variable to some fractional level also to some fractional level; that means they
can take non-integer values also. So, in other sense divisibility means I can change I can
the decision variable can be divided into some fractional level value also, or in other
sense can take the value as non-integer also not only integer, but non-integer. And the
last one is deterministic.

You are using the coefficients in the development or formulation of the

LPP. All these coefficients all these should be deterministic only. So, if
you have any probabilistic coefficient. In that case the probabilistic coefficient has to be
transformed into deterministic one at first and then only we can try to solve the problem.

So, we can say that these 4 properties are assumptions which are used to obtain the
solution of linear programming problems, one is proportionality; that means each and
every decision variable will be given the equal weightage. Additivity the total cost is sum
of the individual costs. As well as the ith contribution of ith constraint is equals to the
sum of the constraints sum of the activities sum of the constraints activities contribution
of the activities. And divisibility that means can be divided into very smaller fraction
or in other sense can take non-integer values also. And similarly other one is the
coefficients whatever we are considering in the formulation of the LPP that is
all of them have to be deterministic. And if they are not deterministic then
we have to convert it first into deterministic and then only we have to solve the problem.

(Refer Slide Time: 07:09)

Now, the problem manipulation that is whenever you have certain different type of
constraints then how to manipulate it to bring it into the standard form; suppose you have
an activity like this, . We will discuss about this later, but these
inequality greater than equals to inequality if we try to convert it into equality format, in
that case we can write it as . Since it is greater than equals So, I have to

subtract some variable. So, I am writing where; obviously the

variable that is there must be positive.

So, greater than equals to inequality whenever it is there by subtracting one variable we
can make it equality constraint. Similarly if you have some inequality constraint like this
for this case this case be converted into equality constraint by adding
one variable say , , where . So, for the less than
equals I have to add. So, that; that means, equal. So, greater than equals and less than
equals inequalities can be converted into equality constraint by this way.

And these variables have some name specific names that we will discuss whenever we
are coming to the solution of the LPP. Even if you have some equality constraint that is
these can be converted into 2 inequality constraints like this.
and .

So, if I have to convert the equality constraint into 2 inequality constraints, then I can
convert it like this. and . That is another

concept you are using the decision variable and we are saying that this is these always
should be greater than equals 0 that is nonnegative. But suppose if I have a variable like
this if is unrestricted in sign; that means we do not know that sign of this it may be
positive it may be negative. If I have a decision variable like this, which is unrestricted
in sign in that case we cannot find the solution. So, I have to rewrite in some other
decision variables such that those decision variables are greater than equals 0, or in other
sense I can write down where .

So, please note these one that, whenever I have the unrestricted sign variable in that case
I can convert it into some other variables , where both . So, these
are the basic formulations in formulations if we have different type of constraints or
some problem in the decision variable, we can convert it into the standard form and we
can try to find out the solution. So, if you see the solution of an LPP is nothing but I have
to find out the optimum value of the LPP objective function, which will satisfy the
constraints you have the constraints.

So, first that value of the decision variables should satisfy the constraints, then we have
to check whether that value gives the optimum value of the objective function or not.
Various kind of solutions you can obtain the solutions can be classified as follows.
(Refer Slide Time: 12:13)

So, I can tell where the solution of LPP. The solution of LPP can be classified as
something like this. It can be feasible it can be infeasible. A solution which satisfies all
the solutions sorry, a solution which satisfies all the constraints is known as feasible
solution. A solution which satisfies all the constraints is known as feasible solution;
whereas, a solution which does not satisfy the constraints is known as infeasible solution.

Similarly, you are having optimal solution, and non-optimal solution. Your optimal
solution is one is the feasible solution for which we get the optimum value of the
objective function. That means you may have many number of many feasible solutions
out of those feasible solutions all may not be optimal solution. Optimal solution will be
that one will be that feasible solution for which we obtain the optimum value of the
objective function. Whereas, the non-optimal solution will be the other feasible solutions
for which we are not obtaining the optimum solution or optimum value of the objective
function. Other one is your solution of an LPP may be unique it may be multiple also it
may be multiple also.

So, if there exists only one optimum solution in that case we say that the solution is
unique. If we obtain only one optimum solution we call it as unique whereas, if the more
than one optimum solution we obtain for some feasible solutions then we can obtain the
multiple solutions. Whenever we are going through the graphical solution of LPP in that
case we will discuss these things again.

(Refer Slide Time: 14:57)

Now, let us see how to formulate one LPP from a problem. Let us consider this problem
I will come to this later.

So, basically the problems states that roll of papers having a fixed length and width of
one 8 centimeter are being manufactured by a paper mill. The rolls have to be cut to
satisfy the following demand. So, the roll is cut the roll we cut it the width can be of 3
types 8 centimeter 45 centimeter 27 centimeter. And the number of rolls we require is
200 120 and 130. So, we want to discuss the linear programming formulation of the
problem to determine the cutting paper. So, that the demand is satisfied and wastage of
the paper is minimum. Please note that the; I have to satisfy this demand for 8 centimeter
I must have 200 rolls for 45 centimeter 120 rolls and 27 centimeter 130 rolls. Whereas,
whenever I am cutting the papers into rolls then there will be certain wastage. So, I have
to satisfy the demand as well as I have to minimize the wastage of the paper.

Now, this last column if you see this table. This table says how we are cutting feasible
pattern of cutting is given here. Number of rolls cut that also you has given
. Then wastage per roll wherever I am cutting a roll per roll what is the
wastage, and how many rolls obtained each from each roll and each mother roll of width
height 0 center 45 centimeter and 27 centimeter. So, this is our problem. We have to
formulate the problem because we have told first I have to identify the problem. And
next whenever the problem is given to us that problem has to be transformed into some
mathematical model. So, that we can analyze it and third part come the solution of the
model.

So, the problem is given to you now we have to see how we can formulate the
mathematical model of this particular problem. Or in other sense how to find out or how
to formulate the LPP for a particular problem. So, here for this case if you see your aim
is to find out the minimization of the wastage of paper. Whenever you are cutting into
rolls some wastage of papers are there and I have to minimize the wastages. This table
tells you how many number of cuts how many that is and the wastage is 20. Similarly
for the wastage is to number of rolls I made the cut is 80 plus 80 and wastage is
20.

So, therefore, what I have to do I have to formulate the model like this, if you see here
for this problem. I have to optimize or minimize the wastage of paper. So, your
formulation of the problem will be a minimize z equals.

(Refer Slide Time: 18:39)

Now, from here how many rolls we got? I got roll in the first roll if you see the roll is
number of rolls is , and what is the wastage per roll wastage per roll is 20. Therefore,
for rolls your wastage will be .

So, in your formulation here it will come as plus again on the similar way in the
next row number of rolls cut was and the wastage per roll is 10. So, total wastage per
these will be plus and like this way for each roll I have to find out what is the total
wastage and I have to make the sum. So, I am writing directly . Here if
you see for the wastage is 0. So, that we are not including because it will be 0.
So, directly we are writing again here if you see here it is for number of
roll the wastage is 0. So, for there will be no component last component will be
.

So, objective function .

I am just finding what is the total wastage whenever we are preparing the number of
rolls. Because in the first row the number of rolls we produced is , and the wastage per
roll was 20. So, I made plus the next one was like this way it was going on.
What are your next criteria of the problem? Next criteria of the problem were. So, that
the demand is satisfied and we know for 80 centimeter number of rolls I require 200 for
45 centimeter I need the roll of 120 and for 27 centimeter I need the roll of 130.

Here number of rolls are obtained that we have written here for 80 centimeter. I got 2
rolls of , one roll of , one roll of and one roll of . So, for 80 centimeter total
how many rolls I got 2 into x 1 plus x 2 plus x 3 plus x 4. So, subject to what will
happen? From here . This value , this is 80
centimeter rolls we are getting from here number of rolls cut. And this value I want it
must satisfy the demand of 80 centimeter and for 80 centimeter I need 200 rolls. So, the
value should be equals to 200.

In the same way for 45 centimeter number of rolls cut is given in this column. And from
here I can also formulate like this way and for 45 centimeter total number
should be equals to 120. So, the using the second column of this 45 centimeter the
second constraint also we can write down like
. And then the same way for the 27 centimeter
roll these are the numbers which I will obtain. So, using the total number of rolls cut I
have to find out the third constraint and that value should be equals to 130.

So, from this third column I can write down the third constraint your third constraint will
be equals to . So, these are the constraints.
So, if you see the problem we were told that whenever you are cutting it the demand
should be satisfied, and wastage of paper should become minimum and we are doing that
thing.

So, the total wastage I am finding out and I have to minimize this subject to what is the
total demand I am generating that should be equals to the specified demand. And since it
is number of rolls therefore, the values of these variables . So, like this
way we can formulate the problem into or we can convert the problem into a
corresponding LPP problem and then we may try to find out the solution of the problem.

(Refer Slide Time: 24:59)

So, this is one thing let us take one more example. So, that it is clear to you, the second
problem is if you see jet airways is adding more flights to and fro it is hub airport. And
so, it needs to hire additional customer service agents.
So, the following table shows the number of agents required for different time period as
well as daily cost per agent in different shifts. So, different timings are given here. And
each column my time period covered in each shift 6 am to 2 pm or what is required that
is given by this. And daily cost per agent is given over here and minimum number of
agents required in 6 am to 8 am is 48 or 8 am to 10 am 79 something like this way. It is
given now what is the problem. The problem is to determine how many agents should be
assigned 2 respective seats each day to minimize total personal cost per agent.

So, you see the problem is very clear you have to minimize the total personal cost. Costs
are given here and minimum number of agents required that is also given here. So,
therefore, if I consider that for each shifts we are having 1 2 3 4 5. So, if I consider 5
variables here corresponding to each of them. That is this is this is this is this is
and this is . Suppose number of agents required in 6 am to 2 pm like this way
others also. So, total cost or wages for them total personal cost will be
.

So, your problem will be now to minimize the function, because you want to minimize
total personal cost. So, from here you are writing like this way which
you can calculate easily.

(Refer Slide Time: 27:22)

So, it should be . Now in the first page that
is 6 am to 8 to am, in this period you require how many personal? You require 48
personal. And there coming from where only all of them during this time period 6 am to
2 pm. So, corresponding variable is . So, value of should be here we have told
minimum number of agents required.

So, the corresponding value would be subject to it should be greater than equals 48.
We have given greater than equals 48 because this is the minimum number of required. It
may be more also. Similarly, for this row it should be for the third row on
the same fashion it should be . And I think you can write down the rest also
like this. Next one would be . The next is .
. . . that is the last one.
This one is .

So, like this way these are the constraint which has to be satisfied. And since
are minimum number of agents required. So, therefore, must be greater
than equals 0 for . So now, I think the it is quite clear to you that whenever a
problem is specified how to formulate the LPP, or I have to check whether it is
minimization problem, or maximization problem, subject to what are the constraints and
also what is the nature of the decision variables, that is whether the decision variables
greater than equals 0 or unrestricted in sign. So, in the next class I think we will start
how to find out the solution graphically of an LPP.