1

Linear Programming
Linear programming (L.P.) is the mathematic technique designed to optimize the use of limited sources
linear programming is more applicable for military, industry, agriculture, transportation, economic
problems etc.
Example-1: Mr. Vishwas Sir started business of publishing Mathematics and physics Books. One
Mathematic book required 3 hours for printing and 1 hour for binding, one physics book required 2 hours
for printing and 2 hours for binding. Printing machine is available for 24 hours per week and binding
machine available for 12 hours per week. A market survey restricts maximum demand of physics books is
4 books per week. Demand of physics book cannot exceed that of mathematics book by more than 2
books per week. Vishwas Sir getting profit of Rs. 2.5 on mathematics book and Rs. 2 on physics book. Find
the mathematical model of the given problem. Also find the number of mathematics books and physics
books published, so that Vishwas Sir gets maximum profit per week.
Solution: Let X1 be the number mathematic books, and X2 be the number of physics books published.
Books  Mathematics Book (X1) Physics Book(X2) Number of hours
machine available
Machine 
Printing 3 2 24
Binding 1 2 12
Profit Rs.25 Rs.2
Profit due physics and mathematics books = 2.5 X1+ 2 X2
3 X1+ 2 X2≤ 24 (Printing)
1 X1 +2 X2≤12 (Binding)
1X2 -1 X1 can not>2 (Market survey)
1X2≤4 (Market survey)
X1≥o & X2≥o (Number of books non-negative)
Thus
Max Z=2.5 X1+ 2 X2 --------------------- (1) (Objective function)
3 X1+ 2 X2≤ 24 --------------------- (2) (Constraints)
1 X1 +2 X2≤12 --------------------- (2) (Constraints)
-1X1 +1 X2≤2 --------------------- (2) (Constraints)
1X2≤4 --------------------- (2) (Constraints)
X1≥o & X2≥o --------------------- (3) (Non-negative restriction)
This is the required mathematical model of the given linear programming problem.
1) X1& X2 are called decision variables
2) The function Z which is decided by the decision variables X1& X2 to the best optimization is called
objective function.
3) The conditions on the choice of decision variables are called constraints.
4) The decision variables normally non-negative real’s i.e. X1≥o & X2≥o, These are called non-negative
restriction. 5)
Any set of decision variables which satisfy the constraints of the given problem is called solution of the
given model.
6) If all the decision variables in a solution satisfy non-negative restriction then the solution is called
feasible solution. 7)
The region formed by the set of all feasible solutions is a feasible region.
8) Feasible solution that optimized the objective function is called optimal solution.
9) The value of the objective function at an optimal solution is called optimal value of given model.
10) Identify the corner point at which the value of the objective function is maximum. The co-ordinate of
this vertex is the optimal solution and the value of Z is the optimal value.
 2

Canonical form of linear programming problem:-

Z max=C1 X1+C2 X2+ C3 X3 +…………….+Cn Xn
a11X1+a12X2+a13X3+………………+ a1nXn ≤b1
a21X1+a22X2+a23X3+……………….+ a2nXn ≤b2
a31X1+a32X2+a33X3+………………..+ a3nXn ≤b3
am1X1+am2X2+am3X3+………………+ amn Xn ≤bm
Standard form of linear programming problem:-
Z max=C1 X1+C2 X2+ C3 X3 +…………….+Cn Xn +0s1+0s2+0s3+………..+0sm
a11X1+a12X2+a13X3+………………+ anXn+1S1 =b1
a21X1+a22X2+a23X3+……………….+ a2nXn+1 S 2 =b2
a31X1+a32X2+a33X3+………………..+ a3nXn +1S3=b3
am1X1+am2X2+am3X3+………………+ amn Xn +1 Sm =bm x1, x2,x3,……..xn ≥0
Procedure for Simplex Method: -
Step-1:- Express the problem in standard form. And Write the objective function as
Z − C1 X1 − C2 X2 − C3 X3 +0S1+0S2+0S3=0 (objective equation)
Express the all the constraints in the form of equality by adding slack variables, thus we have
a11X1+a12X2+a13X3+1S1+0S2+0S3=b1
a21X1+a22X2+a23X3+0S1+1S2+0S3=b2
a31X1+a32X2+a33X3+0S1+0S2+1S3=b3
this form is called standard form.
Step-2:- Find the initial basic solution
The slack variables S1, S2, &S3 which forms basis are called basic variables. The variables X1, X2, &X3 are
called non-basic variables. We start with initial basic solution.X1=0, X2=0, X3=0, S1= b1, S2 =b2, S3=b3
The matrix formed by co-efficient of X1, X2, &X3 is called the body matrix and the matrix formed by the
coefficient of S1, S2, &S3 is called unit matrix.
Co-efficient of R.H.S.(br) Ratio
Basic X1 X2 X3 S1 S2 S3 solution (Xbr /Xrk)
variables Xrk>0
Z − C1 − C2 − C3 0 0 0 0
S1 a11 a12 a13 1 0 0 b1 b1 / a12
S2 a21 a22 a23 0 1 0 b2 b2 / a22

S3 a31 a32 a33 0 0 1 b3 b3 / a32
↑
In the first row i.e. in the row of Z we find most negative value, marks this column by an arrow head (  )
at the bottom of that column. This column is called key column. Suppose − C2 is the most negative value.
br min  bi 
We find the value of =  , aik  0  and mark the row contain this ratio by the (  ) head on
xrk i  aik 
the right. This row is called the key row. The element common to both key row and key column is marked
with square and is called key element or pivot element. The element of the left of the key row is the
outgoing element. The element at the top of the key column is the incoming element. Now we replace S2
by X2 and divide all the element of key row by key element a22. Therefore key element becomes 1.we now
perform elementary row operations in such a way that all the other elements in key column are zero.
This operation is performed on the elements of the column of b’s but not on the elements of the last
column of the ratios.
Step-3:- The above procedure is repeated till the elements in the row above the dotted line i.e. all the co-
efficient in the objective equation are non-negative when this situation is obtained, the optimal is arrived.
The column of the basic variables states the variables in the optimal solutions and the column of solution
gives their values. The values in this column in the row of Z are the optimum value of Z
 3

Note:-if in any iteration all the values in the key column are negative then the problem has unbounded
solution.
Note: - If all the elements in the row of Z are positive and value of any one of the basic variable is
negative, then the given problem has infeasible solution.
Note: - If the coefficient of any one of the non-basic variable in the row of Z is zero then given problem
has alternative solution.
Note: - If the value of any one of the basic variable is zero, then the given problem has degenerate
solution
Big-M Method (or Penalty method):-
If anyone or some constraints are of greater than type then we have to subtract surplus variable, i.e. we
have to add -s1,-s2,-s3 ,……etc. to convert greater than or equal type inequality to equality. But we would
not get a unit matrix. To overcome this difficulty, we introduce in addition to surplus variable -s1,-s2,-s3
,…… artificial variable A1,A2,A3,…..with positive sign in these constraints. An artificial variable is introduced
even when constraints is of equality type. In the objective function we assign big penalty by subtracting
MA1, MA2, MA3, ..… If the objective function is of maximization type.
Consider the problem:-
Maximization Z max=C1 X1+C2 X2+ C3 X3
a11X1+a12X2+a13X3 ≥b1
a21X1+a22X2+a23X3 ≤b2
a31X1+a32X2+a33X3 ≤b3
X1, X2 , X3 ≥0
Z max = C1 X1+C2 X2+ C3 X3+0S1+0S2+0S3-MA1
i.e. Z- C1 X1-C2 X2- C3 X3-0S1-0S2-0S3+MA1=0 ……….(1)
a11X1+a12X2+a13X3-1S1 +0S2+0S3+1A1=b1 ………. (2)
a21X1+a22X2+a23X3 +0S1 +1S2+0S3+0A1= b2
a31X1+a32X2+a33X3 +01S1 +0S2+1S3+0A1= b3
We now write the objective function free from artificial variable by adding -M times the first constraints to
the objective function. i.e. in equation (1) we add –M times the equation (2)
Z- (C1+M a11) X1-(C2+M a12) X2- (C2+M a13) X3-MS1+0S2+0S3+0A1= -Mb1
Note: - In LPP, if variable x is un-restricted then we take x=xl –xll where xl & xll ≥ 0
Dual simplex Method:-
1) Convert the problem into minimization type
2) Convert all the constraints in less than or equal to type. If any constraints is greater than or equal to
Types multiply throughout by -1 and change the equality sign.
3) Convert all the constraints in to equality by adding slack variables
4) Put this information in a table
5) If all the coefficients in the row of Z are negative and all the right hand constants i.e. b’s are positive
Then the basic feasible solution is obtained. If all the coefficient in the row of Z are negative and
at least one b is negative go to the step (6)
Note: - If all the coefficient in the row of Z are positive the method fail
6) Select the row which contains the most negative number in the column of b’s. Denote it by an arrow
head like  .This is the key row and the corresponding variable on the left is the out going variable
7) Find the ratio of the elements in the row of Z to the corresponding elements in the key row i.e.
 Cj 
 
 xrj 

With Xrj <0 Write these ratios in another row below the table.
8) Select the column which contains the smallest ratio and denote it by arrow head like  . This is the
 4

Key column. The element where the key row & key column intersect is the key elements.
9) Construct another table as in the usual simplex method. Divide each element in the key row by the key
element, so that key element will be unity. Now make all other element in the key column zero by suitable
row operation
Determination of all possible basic solution: -
The LPP form includes m simultaneous linear equations in n unknowns or variable (m<n). we divide the n
variable into
1) n-m variables, to which we assign zero values.
2) The remaining variables, whose values are determined by solving the resulting m equations.
If the m equations yield a unique solution then associated variables are called basic variables and the
Remaining n-m variables are called non basic variables.

Note: - Maximum number of basic solutions for m equations and n variables is

ncm
Non-linear programming problem with equality constraints:-
Generalized lagrangian method of undetermined multipliers:-
Optimize Z=f(x1, x2, x3,………, xn) ………..(1)
With h1(x1, x2, x3,………, xn )=b1
h2(x1, x2, x3,………, xn )=b2
h3(x1, x2, x3,………, xn )=b3
……………………..
hm(x1, x2, x3,………, xn )=bm
x1, x2, x3,………, xn ≥0 and m < n
let
gi(x1, x2, x3,………, xn) = hi(x1, x2, x3,………, xn ) - bi i=1,2,3,…..m
We define a Lagrangian function as
m
L( x,  ) = f ( x) −  i gi ( x) m<n
i =1
Where λi (i=1, 2, 3…m) are called lagrangian multipliers
The necessary condition for a maximum or minimum of f(x) is the system of m+n equations

L f m
g
= −  i i = 0 , j=1, 2, 3,. . . , n and
L
= − gi = 0 , i= 1, 2, 3, . . m
x j x j i =1 x j i
Note:- For above lagrangian
m
L( x,  ) = f ( x) −  i gi ( x) m<n
i =1
Hessian Bordered matrix is denoted by HB and is given by

Omm Vmn 
HB =  T 
V
 nm H nn 
 5

 g1 g1 g1 

 x −−−−
x2 xn 
 1   2L 2L 2L 2L 
Where  g 2 g 2 g 2 
,
 x 2 X 1x2 X 1x3
.....
X 1xn 
 x −−−− 
xn 
1

x2  2L 2L 2L 2L 

= 
1
Vmn  .... 
 | | | |   X 2 x1 x2 2 X 2x3 X 2xn 
   2L  L
2
 L
2
 L 
2

 | | | |  H = .... 
 X 3x1 X 3x2 x32 X 3xn 
 g m g m g m   : : 
 −−−−   : : :
 x1 x2 xn  mn  :
 2
: : : : 

  L 2L 2L
....
2L 
 Xnx1 Xnx2 Xnx3 2 
xn  nn

Note:-The matrix HB is called Hessian Bordered matrix then the sufficient condition for Maxima
and Minima can be stated as follow.
Let (x* , λ*)be the stationary point for the Lagrangian function L(x, λ) and H B be the value of
corresponding Hessian Bordered matrix computed at this point then
1) x* is a maximum point if starting with principal minor (determinant) of order (2m+1), the last (n-
m)
Principal minor (determinant) of HB forms alternative sign pattern starting with (-1)m+1
2) x* is a minimum point if starting with principal minor (determinant) of order (2m+1), the last (n-
m)
Principal minor (determinant) of HB have same sign starting with (-1)m
Non-linear programming problem with inequality constraints:-
Kuhn tucker Method:-
Z =f(x), x=(x1, x2, x3… xn ) Subject to gi(x) ≤ bi therefore consider gi(x) +si2 = bi
We define Lgrangian function as
m
L ( x,  , s ) = f ( x ) −  i  gi ( x ) + si2 − bi 
i =1
For maximization:- Kuhn-tucker’s necessary &sufficient condition maximization of Z are

(1) L ( x ,  , s ) = 0 , j= 1,2,3,…..n (2) i  gi ( x ) − bi  = 0 , i=1,2,3,……m
x j
(3) gi(x) ≤ bi , i=1,2,3……m (4) λi ≥ 0 , i=1,2,3,….…m
For Minimization:- Kuhn-tucker’s necessary &sufficient condition minimization of Z are

(1)
L ( x ,  , s ) = 0 , j= 1,2,3,…..n (2) i  gi ( x ) − bi  = 0 , i=1,2,3,……m
x j
(3) gi(x) ≤ bi , i=1,2,3……m (4) λi ≤ 0 , i=1,2,3,….…m