Big-M Method [Introduction]

 
In order to obtain an initial basic feasible solution, it is necessary to convert the given LPP into
its standard form; in order to obtain the standard form; a non-negative variable is added to the
left side of each of the equation that lacks the much needed starting basic variables. So added
variable is called an artificial variable and it places the role in providing initial basic feasible
solution.
 
The starting point of the big-M method is a basic solution that is feasible to the artificial
problem. This solution allows us to start the Simplex algorithm expeditiously, but it is not a
feasible solution to the original problem. Our goal is to iterate toward solutions that are inside the
original feasible set, assuming that it is not empty.
 
The artificial variables have no physical meaning from the view of the original problem; the
method will be valid only if it is possible to force these variables at zero level when the optimum
solution is attained.

To accomplish this goal, we designed an artificial objective function that is an aggregation of

two functional parts, of which one is a copy of the original objective function and the other is a
“penalty function” associated with the artificial variables.
 
The magnitude of the penalty, M, needs to be chosen to ensure that the contribution to this
aggregate objective function from the second part, whenever it is positive, always outweighs that
from the first part, for any solution.
 

Artificial Variable

 
 
If there is a constraint which is of > = or equality type then it is needed to write the constraint in
standard form because it is not possible to get unit column vector. The additional variable, which
is added in order to get unit column vector, is called an artificial variable.
 
Two methods are generally employed for the solution of linear programming problems having
artificial variables:
 
1.  Two-Phase Method.
 
2.  Big-M Method (or) Method of Penalties. We will have discussion only on Big-M Method
here.
 

The Big M Method Procedure

 
 
If an LP has any > or = constraints, the Big M method or the two-phase simplex method may be
used to solve the problem.
 
The Big M method is a version of the Simplex Algorithm that first finds a best feasible solution
by adding “artificial” variables to the problem. The objective function of the original LP must, of
course, be modified to ensure that the artificial variables are all equal to 0 at the conclusion of
the simplex algorithm. The iterative procedure of the algorithm is given below.
 
Step-1
 
 
Modify the constraints so that the RHS of each constraint is non-negative (This requires that
each constraint with a negative RHS be multiplied by -1. Remember that if any negative number
multiplies an inequality, the direction of the inequality is reversed). After modification, identify
each constraint as a <, >, or = constraint.
 
Step-2
 
 
Convert each inequality constraint to standard form (If a constraint is a ≤ constraint, then add a
slack variable Xi; and if any constraint is a ≥ constraint, then subtract an excess variable Xi ,
known as surplus variable).
 
Step-3
 
 
Add an artificial variable a1 to the constraints identified as ‘≥’ or with ‘=’ constraints at the end
of Step2. Also add the sign restriction ai ≥ 0.
 
 
Step-4
 
 
Let M denote a very large positive number. If the LP is a minimization problem, add (for each
artificial variable) Mai to the objective function. If the LP is a maximization problem, add (for
each artificial variable) -Mai to the objective function.
 
Step-5
 
 
Since each artificial variable will be in the starting basis, all artificial variables must be
eliminated from row 0 before beginning the simplex.
 
Now we solve the transformed problem by the simplex (In choosing the entering variable,
remember that M is a very large positive number). If all artificial variables are equal to zero in
the optimal solution, we have found the optimal solution to the original problem. If any artificial
variables are positive in the optimal solution, the original problem is infeasible!!!

 
1. If at least one artificial variable is present in the basis with zero value, in such a case the
current optimum basic feasible solution is a degenerate solution.
 
2. If at least one artificial variable is present in the basis with a positive value, then in such a
case, the given LPP does not have an optimum basic feasible solution. The given problem is said
to have a pseudo-optimum basic feasible solution.
 
Example-1
 
 
Solve the following LPP by using Big -M Method
 
Maximize Z = 6X1+4X2
 
Subject to constraints:
 
2X1+3X2<=30
 
3X1+2X2<=24
 
X1+X2>=3
 
 
Solution
 
 
Introducing slack variables S1>=0, S2>=0 to the first and second equations in order to convert
<= type to equality and add surplus variable to the third equation S3 >=0 to convert >= type to
equality.
 
Then the standard form of LPP is
 
MAX Z=6X1+4X2+0S1+0S2+0S3 Subject to constraints
 
2X1+3X2+S1=30
 
3X1+2X2+S2=24
 
X1+X2-S3=3
 
Clearly there is no initial basic feasible solution. So an artificial variable A1>=0 is added in the
third equation. Now the standard form will be
 
MAX Z=6X1+4X2+0S1+0S2+0S3+A1
 
Subject to constraints
 
2X1+3X2+S1=30
 
3X1+2X2+S2=24
 
X1+X2-S3+A1=3
 
Now the first simplex table is placed below with following additional information.
 
It clearly shows the net evaluations for each column variable; from these calculations, it is clear
that X1 is the entering variable and A1, the artificial variable leaves the basis. Introduce Y1 and
drop Y6

Then, in the usual row operations, we modify this table and the new table is arrived.
 
Table-2
Introduce Y5 and drop Y4 from the basis; the new modified table-3 is given below;
Since all Zj –Cj >=0, an optimum solution has reached. Thus an optimum basic feasible solution
to the LPP is
 
X1=8
 
X2=0
 
MAX Z =48
 
 
Example-2
 
 
Use big M method to solve a given LPP. Minimize Z = 5X1-6X2-7X3
 
Subject to constraints
 
X1+5X2-3X3>=15
 
5X1-6X2+10X3<=20
 
X1+X2+X3=5
 
X1, X2 & X3 >= 0
 
Solution
 
 
Introducing slack variables X4>=0 to the first equations in order to convert <= type to equality
and add surplus variable X5>=0 to the second equation in order to convert >= type to equality.
 
Then the standard form of LPP is
 
MIN Z=5X1-6X2-7X3+X4-X5 Subject to constraints
 
5X1-6X2+10X3+X4=20
 
X1+5X2-3X3-X5=15
 
X1+X2+X3=5
 
Clearly there is no initial basic feasible solution. So two artificial variables A1>=0 and A2>=0
are added in the second and third equation. Now the standard form will be
 
MIN Z=5X1-6X2-7X3+X4-X5+A1+A2 Subject to constraints
 
5X1-6X2+10X3+X4=20
 
X1+5X2-3X3-X5+A1=15
 
X1+X2+X3+A2=5

From the calculations related to entering column and leaving variable, which is summarized in
the above table, it is clear that introduces X2 and drop X6 from the basis. The new simplex table
is given below;
The new modified table- 3 is given below
 
-1/5 +32/5*1/5*5/8
Since all Zj –Cj <=0, an optimum solution has reached. Thus an optimum basic feasible solution
to the LPP is
 
X3=5/4
 
X2=15/4
 
MIN Z= (5*0) + (-6*15/4) + (-7* 5/4)
 
=   (-90/4) + (-35/4)
 
=   -125/4
 Tags : Operations Management - Introduction to Operations Research

Last 30 days 3760 views

1. Simplex method (BigM method) example (

Enter your problem )

( Enter your problem )

1. Simplex method Other related methods

Algorithm
2. BigM method 1. Simplex method
Algorithm (BigM method)
3. Maximization 2. Two-Phase method
example 3. Graphical method
4. Minimization 4. Primal to dual
example conversion
5. Degeneracy 5. Dual simplex
example-1 (Tie for method
leaving basic 6. Integer simplex
variable) method
6. Degeneracy 7. Branch and Bound
example-2 (Tie - method
first Artificial 8. 0-1 Integer
variable removed) programming
7. Unrestricted problem
variable example 9. Revised Simplex
8. Multiple optimal method
solution example
9. Unbounded
solution example
10. Infeasible solution
example

3.
Maximization 5. Degeneracy example-1 (Tie
example for leaving basic variable)
(Previous (Next example)
example)
4. Minimization example

Find solution using Simplex(BigM) method

MIN Z = x1 + x2
subject to
2x1 + 4x2 >= 4
x1 + 7x2 >= 7
and x1,x2 >= 0

Solution:
Problem is

Min Z = x1 + x2

subject to

2 x1 + 4 x2 ≥ 4

x1 + 7 x2 ≥ 7

and x1,x2≥0;

The problem is converted to canonical form

by adding slack, surplus and artificial
variables as appropiate

1. As the constraint-1 is of type '≥' we

should subtract surplus variable S1 and
add artificial variable A1

2. As the constraint-2 is of type '≥' we

should subtract surplus variable S2 and
add artificial variable A2

After introducing surplus,artificial

variables
Min Z = x1 + x2 + 0 S1 + 0 S2 + M A1 + M A2

subject to

2 x1 + 4 x2 - S1 + A1 =4

x1 + 7 x2 - S2 + A2 = 7

and x1,x2,S1,S2,A1,A2≥0

Ite
rat
Cj 1 1 0 0 M M
ion
-1

M
i
n
R
 x2  a
Ente t
B XB x1 ring S1 S2 A1 A2 i
vari o
able
X
B
x
2

4
4
A1 4 2 4 -1 0 1 0
=
1

 A 7 1  (7 0 -1 0 1 7
2  )  7
Lea (piv =
vin ot 1
g elem →
 var
iab ent)
le

   
 - M  M 
 -
M  M= M=
 3 M 
 11 - M M
  M  -
 Z M 1 M=M × ×
Zj  3M= M=M
=0  1M=M ×( 1+ 0+
Zj M× ×0
0= ×4+ - M M
= 2+M +M
Zj= M×7 1)+ × ×
∑ ×1 ×(
∑C Zj= M× 0 1
CB Zj= -1)
BXB ∑CB 0 Zj Zj
xj ∑C Zj=
x2 Zj= = =
Bx1 ∑C
∑C ∑ ∑
BS2
BS1 CB CB
A1 A2

 -  -
3M+ 11M+    
 M   M 
Cj 1  1 - 0  0 
M=0 M=0
- - 11M+ 0= 0=
-(- -(-
Zj 3M+ 1=1- M- M-
M) M)
1=1 11M M M
-3M ↑

Negative minimum Cj-Zj is -11M+1 and its

column index is 2. So, the entering
variable is x2.

Minimum ratio is 1 and its row index is 2.

So, the leaving basis variable is A2.

∴ The pivot element is 7.

Entering =x2, Departing =A2, Key Element

=7
R2(new)=R2(old)÷7

R1(new)=R1(old)-4R2(new)

Ite
rat
Cj 1 1 0 0 M
ion
-2

M
i
n
R
a
 x1 
t
Enteri
i
B XB ng x2 S1 S2 A1
o
variab
le
X
B
x
1

 A  0   0   -  47   1 0

1 
Lea
0=4   1
-
( 0=4 1 
- -
47=0- 1=1 1
4× - 0
vin
g
4×
)
1 07  
4× 1=(
1 -
( -
4× 7
0 =
var R1( 107=2- R1( 1)- R1( 0
iab new 4×17 new 4× new →
le )=R (pivot
1(o elemen
)=R 0
1(o R1(
)
17 )=R
R1(ne 1(o
ld) t) ld) new w)=R1 ld)
- R1(new - )=R (old) -
4R2 )=R1(o 4R2 1(o - 4R2
(ne ld)- (ne ld) 4R2(n (ne
w) 4R2(ne w) - ew) w)
w) 4R2
 (ne
w)

 1   1  0   0 

 -17 
1=7 1=7 0=0 0=0
 17  -
÷7 ÷7 ÷7 ÷7 1
17=1÷ 17=(-
R2( R2( R2( R2( 1
7 1)÷7
x2 new new new new 7
R2(new R2(ne
)=R )=R )=R )=R =
)=R2(o w)=R2
2(o 2(o 2(o 2(o 7
ld)÷7 (old)
ld) ld) ld) ld)
÷7
÷7 ÷7 ÷7 ÷7

 4M7
 - -17 
M  4M7-
 10M7  1  - 17=M  M 
 Z
+17 1 1=M M=M ×47+ M=M
=1   Z
0M7+17 ×0 ×( 1× ×1
1=1 j 
=M×10 +1 - +1
×1 Zj=
Zj=
∑C
∑C
Bxj
7+1×1
7
×1 1)+
Zj= 1× ( -
×0
Zj=
Zj=∑C ∑C 0 ∑C
BXB
Bx1 Bx2 Zj=
∑C
17 ) BA1

BS1 Zj=∑
CBS2

 -  -
10M7+6 4M7+1
7 - 7 -
10M7+6 4M7+1
 M 
7=1-  0  7=0-  0 
Cj- M=0
0=1 0=M
Zj
(10M -1
-(-
M) (
4M -M
7+17 7-

) ↑ 17)
Negative minimum Cj-Zj is -10M7+67 and its
column index is 1. So, the entering
variable is x1.

Minimum ratio is 0 and its row index is 1.

So, the leaving basis variable is A1.

∴ The pivot element is 107.

Entering =x1, Departing =A1, Key Element

=107

R1(new)=R1(old)×710

R2(new)=R2(old)-17R1(new)

Ite
rat
Cj 1 1 0 0
ion
-3

M
i
n
R
a
 S2 E
t
ntering
B XB x1 x2 S1 i
variabl
o
e
X
B
S
2

 x  0   1   0   - 0

1 
Lea
0=0 1=10 0=0 710 
×71 7×7 ×71 -
  ( 25
2
5
 0 10 0
710=(
-
)  25
vin R1(n R1(n R1(n =47×71
1)×7 0
g ew)= ew)= ew)= =
10 (pivot 0
var R1(o R1(o R1(o
R1(ne element →
iab ld) ld) ld)
w)=R1 )
le ×71 ×71 ×71
(old) R1(new)
0 0 0
×710 =R1(old
)×710

 110 
110=0
 -15 
 0  -
 1   1  17× -
0=17
1=1- 1=1-
17×
0
-
17×
17×
0 (-
15= - (
1
x2
R2(n
ew)=
R2(n
R2(n
ew)=
710
17 - -)-

R2(o
ld)-
ew)=
R2(o
R2(o
ld)-
) 17×25 -
R2(new)
ld)- R2(ne
17R1 17R1 =R2(old
17R1 w)=R2
(new (new )-
(new (old)
) ) 17R1(ne
) -
w)
17R1(
new)

 Z  Zj  1   1   -35   15 1

=1  Zj= 1=1 1=1 - 5=1×25
1=1 ∑CB ×1+ ×0+ 35=1
×0
+1
xj 1×0
Zj=
1×1
Zj=
× (+1× (
×1 ∑CB ∑CB
Zj=
∑C
x1 x2 -
710
-15 )
Zj=∑CB
BXB
)
+1
S2

×110
 Zj=∑
CBS1

 35   -15 
35=0- -15=0-
 0   0 
Cj-
Zj
0=1- 0=1-
1 1
( ( - 15

) )
35 ↑

Negative minimum Cj-Zj is -15 and its

column index is 4. So, the entering
variable is S2.

Minimum ratio is 0 and its row index is 1.

So, the leaving basis variable is x1.

∴ The pivot element is 25.

Entering =S2, Departing =x1, Key Element

=25

R1(new)=R1(old)×52

R2(new)=R2(old)+15R1(new)

It
er
at
Cj 1 1 0 0
io
n-
4

B XB x1 x2 S1 S2 M
i
n
 R
a
t
i
o

 -74 
-
 0   52   0   1 
0=0×
52
52=1
×52 52
(
0=0× 74= - 1=25
×52
S2 R1(ne
w)=R1
R1(ne
w)=R1
R1(ne
w)=R1 710) R1(ne
w)=R1
(old) (old) (old) ×52 (old)
×52 ×52 ×52 R1(new) ×52
=R1(old
)×52

 0 
 -14 

 1 
 12 
 1 
-
14=110+ 0= (
12=0+ -
1=1+1
15×5
1=1+1
5×0 15× (
5×0
R2(ne
2
R2(ne )15
x2
w)=R2
(old)
R2(ne
w)=R2
w)=R2
(old)
-74 ) +15×
1
(old)
+15R1 +15R1 R2(new) R2(ne
+15R1
(new) (new) =R2(old w)=R2
(new) )
(old)
+15R1(n
+15R1
ew)
(new)

   Zj   12   1   -14   0 

Z= Zj=∑ 12=0 1=0× - 0=0×
1  CBxj ×52+ 0+1× 14=0× 1+1×
1= 1×12 1 0

×
1 Zj=∑
CBx1
Zj=∑
CBx2
( -
Zj=∑
CBS2
 1
74) +1

(
Zj
= × -
∑
CB
XB )
14
Zj=∑CB
S1

 12   14 1
12=1- 4=0-

Cj-Zj (  0 
12 0=1-1 ( -
 0 
0=0-0

) )
14

Since all Cj-Zj≥0

Hence, optimal solution is arrived with

value of variables as :
x1=0,x2=1

Min Z=1

LESSON 2. Big M Method - Problems

ARTIFICIAL  VARIABLE TECHNIQUES
You may recall that while introducing the slack and surplus
variables, we had assigned a zero cost to them in the objective
function. Moreover, the slack variables readily provided the
initial basic feasible solution. There are, however, many linear
programming problems where slack variables cannot provide
such a solution. To solve such linear programming problems,
there are two (closely related) methods, viz.,


The "big M-method" or the "method of

penalties" due to A. Charnes, and



"two phase method" due to Dantzig, Orden and

Wolfe.


SAMPLE   PROBLEMS
1. Use penalty (or Big  'M') method to
       Minimize  z  = 4xi + 3x2
      subject to the constraints :
                            2x1+ x2  ≥ 10,   -3x1, +  2x2  ≤ 6
                            x1 + x2 ≥ 6,   x1 ≥ 0 and x2  ≥ 0.
 Solution. Introducing surplus (negative slack) variables x3  ≥ 0,
x5  ≥ 0 and slack variable x4  ≥ 0 in the constraint inequations,
the problem becomes
 Maximize  z*  =  - 4x1 - 3x2 + 0.x3 + 0.x4 + 0.x5
subject to the constraints :
                                    2x1 + x2 - x3 = 10,  - 3x1 + 2x2 + x4 =  6
                                      xl + x2 – x5 =  6,  xj ≥ 0 Q(j = 1,2, 3, 4, 5)
 Clearly, we do not have a ready basic feasible solution. The
surplus variables carry negative coefficients (-1). We introduce
two new variables A1 ≥ 0 and A2 ≥ 0 in the first and third
equations respectively. These extraneous variables, commonly
termed as artificial variables, play the same role as that of slack
variables in providing a starting basic feasible solution.
 We assign a very high penalty cost (say -M, M ≥ 0) to these
variables in the objective function so that they may be driven to
zero while reaching optimality.
 Now the following initial basic feasible solution is available :
                                         Al = 10, x4 = 6   and A2 =  6
 with B = (a6 , a4, a7) as the basis matrix. The cost matrix
corresponding to basic feasible solution is cB = ( -M, 0, -M )
 Now, corresponding to the basic variables A1, x4 and A2. the
matrix Y = B-1A and the net evaluations zj - cj (j = 1, 2, .... 7) are
computed. The initial basic feasible solution is displayed in the
following simplex table :
We observe that the most negative zj - cj is  4 - 3M (= zl – c1).
The corresponding column vector y1, therefore, enters the basis.
Further, since min.  = 5;  the element y11 (=2) becomes the
leading element for the first iteration
First Iteration:   Introduce y2 and drop y7.

 In the above table, we omitted all entries of column vector y6,
because the artificial variables Al has left the basis and we
would not like it to re-enter in any subsequent iterations.
Now since the most negative (zj—cj) is  z2-c2; the non-basic
vector y2 enters the basis. Further, since min   is 2 which occurs
for the  element y32 ( = 1/2), the corresponding basis vector y7
leaves the basis and the element  y32 becomes the leading
element for the next iteration.
 Final Iteration: Optimum Solution,

 It is clear from the table that all zj - cj are positive. Therefore an
optimum basic feasible solution has been attained which is given
by
        x1= 4, x2 = 2,  maximum  z = 22.
 2. Maximize z = 3x1, + 2x2   subject to the constraints :
            2x1 + x2  ≤ 2,  3x1 + 4x2  ≥  12, x1, x2  ≥ 0.
 Solution:  
Introducing slack variable x3 ≥ 0,  surplus variable x5 ≥ 0 and an
artificial variable A1 ≥ 0, the reformulated L.P.P. can be written
as :
Maximize z  = 3x1 + 2x2 + 0.x3 + 0.x4 – MA1
subject to the constraints :
2x1 + x2 + x3 = 2,
3x1 + 4x2 - x4 + A1 = 12
x1, x2, x3, x4 ≥ 0 and A1≥ 0.
An obvious   starting basic feasible solution is :
x3 = 2   and   A1 = 12.
The iterative simplex tables are :
Initial Iteration:  Introduce y2 and drop y3.

F
inal Iteration.   No solution.
 Here the coefficient of M in each zj - cj is non-negative and an
artificial vector appears in the basis, not at the zero level. Thus
the given L.P.P. does not possess any feasible solution.