CHAPTER IV

Linear Programming Matrix Simplex

1. INTRODUCTION

In the previous chapter, we discussed solving linear programs using the graph

method. This method can only solve linear programs that have two decision variables.

The graph method cannot solve management problems that have a large number of

decision variables. How to practically solve LP problems that have a large number of

decision variables? The answer is, you can use the QSB (Quantitative System For

Business) computer program or use the simplex method. In reality, the use of computers

is more efficient, but the basic method used in computer operation is still the simplex

method.

Manual solution of linear programs using the simplex method still requires us to

develop LP formulation skills. By studying the mechanism of the simplex method, the

information obtained is not only the optimal solution, but also the economic

interpretation and information for conducting sensitivity analysis.

2. GENERAL FORM OF SIMPLEX TABLE

The simplex method is a development of the algebraic method that only examines

part of the number of base solutions in tabular form. The simplex table only describes

the linear program problem in the form of its coefficients, both the objective function

coefficient and the coefficient of each constraint. Because linear program problems can

be described in various forms such as maximum or minimum with constraints can also
be in the form of smaller equal to (≤) , greater equal to (≥) or equal to (¿), then a

standard form is needed that is commonly used to solve linear program problems.

The commonly used standard form is the "standard form" which has the following

characteristics:

(1) The value of the right-hand side of each constraint is nonnegative. If the right-hand

side of the constraint is negative, it can be converted to positive by multiplying it by

minus one. (−1). For example 2 x1 +3 x 2 ≤−30, it can be changed to −2 x 1−3 x 2 ≤30

(2) All constraints are in the form of equations, except nonnegativity constraints. If the

constraint is smaller than (≤) can be converted into an equation by adding a "slack

variable". Suppose the constraint is in the form: 2 x1 +3 x 2 ≤ 60, the equation can be

changed to: 2 x1 +3 x 2 +S 1=60. If the constraint is greater than (≥) can be converted

into an equation by subtracting it with "surplus variable". Suppose the constraint is

in the form: 2 x1 +3 x 2 ≥ 30, the equation can be changed to 2 x1 +3 x 2−S1=30 .

(3) All decision variables are nonnegative.

(4) The objective function is in maximum form. If the objective function is in

minimum form, it can be converted to maximum form by multiplying it by minus

one. (−1). For example, Z min=40 x 1+ 30 x 2 , the maximum can be changed to

−Z mak=−40 x 1−30 x2.

Consider the following general (standard) form of LP formulation.

z mak=c 1 x 1 +c 2 x 2 +c 3 x 3 +…+ c n x n

d.k

(1) a 11 x 1 +a 12 x 2+ a13 x 3 +…+ a1 j x j +S 1=b 1

 (2) a 21 x 1+ a22 x 2 +a23 x3 + …+a2 j x j + S2 =b2

(3) a 31 x 1+ a32 x 2+ a33 x 3 +…+ a3 j x j + S3 =b3

......

......

(i) a i1 x 1+ ai 2 x 2 +ai 3 x3 + …+aij x j +S i=b i

If the standard form is put into a table, the general form of the simplex table will be

obtained as shown in table 4.1 below.

 Table 4.1. General Form of Simplex Table

Vrb Cj C1 C2 C3 … Cj
CB
base Bj a1 a2 a3 … a3

CB 1 S1 b1 a 11 a 12 a 13 … a1 j

CB 2 S2 b2 a 21 a 22 a 23 … a2 j

CB 3 S3 b3 a 31 a 32 a 33 … a3 j

: : : : : : : :

CB j Sj bj a j1 a j2 a j3 … a ij

Z Z1 −C1 Z 2−C2 Z3 −C3 … Z j−C j

Table Description:

1. CB describes the objective function coefficients for the variables in the base.

2. The variable column in the base contains slack variables that will be replaced by

decision variables.

3. Column b j contains the constant right-hand side of each constraint.

4. Line C j contains the objective function coefficients of each decision variable.

5. Line a j row contains the number of subtraction results Z j−C j which will provide

information on whether the table is optimized or not.

3. STEPS FOR USING THE SIMPLEX METHOD

(1) Convert the linear programming problem into standard form (see characteristics of

standard form above).

(2) Check if each constraint has a "basis variable". If not, add one artificial (pseudo)

variable that acts as a basis variable, e.g. Q1 or Q2 the number of which

corresponds to the need.

The base variables are:

A variable that has a coefficient of one, while the other constraints have a value of

zero.

(3) Enter all constraint function values into the simplex table.

(4) Enter the objective function coefficient values in the row Z j−C j , with the formula:

Z j−C j =CB yj−C j .

This formula is only used at the beginning of the simplex table.

(5) Determine the key column, which is the column that has the largest negative value

in the row. Z j−C j . If there are two equal largest values, one can be chosen.

(6) Determine the key row, which is the value that has the smallest index number and is

not negative, using the formula:

nilai pada kolom bl X bl

min , atau min , ,Y ≥0
nilai pada kolom kunci Y lk lk

(7) Find the new number in the key row by dividing all the numbers in the key row by

the key number. The key number is the number contained in the crossing of the key

row with the key column.

(8) Find the new number in another row with the formula:

New number = Value in the old row minus the multiplication of the coefficients

in the new row

key column with the new number of the key row.

(9) If the optimal solution has not been found, go back to the fifth step above, so that

the value contained in row Z j−C j ≥ 0.

Figure 4.1 shows a summary of the steps for using the simplex method.
SOLVING THE SIMPLEX METHOD MAXIMUM PRODUCT MIX PROBLEM
In this section, we will solve PT Umsini's product mix problem using the simplex

method.

For simplicity, let's rewrite the problem as follows:

Z mak. = 40 x 1 +30 x 2

d.k

(1) 2 x1 +3 x 2 ≤ 60

(2) 2 x 2 ≤ 60

(3) 2 x1 +1 x 2 ≤ 40

(4) x 1 ≥ 0(nonnegativity)

(5) x 2 ≥ 0(nonnegativity)

Where x 1 = number of astir production units, and x 2 = number of cosmos produced per

week.

Step 1.

Transform the LP formulation into a standard form by adding slack variables,

except nonnegative constraints. The standard form of the PT Umsini problem is:

(1) 1 x1 +3 x 2+ S 1+ 0 S 2+ 0 S 3=60

(2) 2 x 2+0 S1 + S2 +0 S3 =30

(3) 2 x1 +1 x 2+ 0 S 1+ 0 S 2+ S 3=40

The objective function can be written as follows.

Z mak. = 40 x 1 +30 x 2+0 S1 +0 S2 +0 S3

Step 2.
 Check if all constraints have base variables, as a basis for creating the initial

simplex table. The first constraint has a base variable that is S1 , second constraint S2 ,

And the third constraint S3. Thus every constraint of smaller form equal to (≤), if

converted into standard form must have a base variable. So the initial simplex table can

be constructed directly.

The initial simplex table of PT Umsini's problem can be seen in Table 4.2, by

entering all values contained in the constraints and objective functions into the table.

Specifically to fill in the numbers contained in the row Z j−C j formula is used:

CBY j−C j . This formula only applies to fill the rows Z j−C j in the initial table.

Table 4.2 Initial Simplex Table of PT Umsini Problem

Vrb. in C1 40 30 0 0 0
CB Index
base C2 x1 x2 S1 S2 S3

0 S1 60 2 3 1 0 0 60 /2=30

0 S2 30 0 2 0 1 0 30/0=∝

0 S3 40 2 1 0 0 1 40/2=20

Z j−C j 0 -40 -30 0 0 0

Line 1. This first line contains the objective function coefficients of each

variable, namely:

C 1=40 , C 2=30 ,C 3=0 ,C 4 =0 ,C 5=0 .

Line 2. Contains the variable that corresponds to the first line, namely:

x 1 , x 2 , S1 , S 2 dan S3
Rows 3,4,5 contain the coefficients of the three constraints. In Table 4.2, it can be

seen below

S1 , S 2 ,and S3 are basis variables with the respective values

60, 30, and 40. These values are the constants of the right-hand side of

each constraint.

To fill the value in the sixth or new row

Z j−C j , filled in using the formula above.

[]
60
1. Z=( 0 , 0 , 0 ) 30 −0=0
40

[]
2
2. Z1 =( 0 , 0 ,0 ) 0 −40=−40
2

[]
3
3. Z 2=( 0 , 0 , 0 ) 2 −30=−30
1

[]
1
4. Z3 =( 0 , 0 , 0 ) 0 −0=−0
0

[]
1
5. Z 4=( 0 , 0 , 0 ) 0 −0=0
1
 []
0
6. Z5 =( 0 , 0 , 0 ) 0 −0=−0
1

A value of Z=0, indicating that in the initial table the values of x 1and

x 2 are equal to zero (not yet in production). If entered in the objective

function Z = 40 (0) + 0 (60)+0 (30)+0 (40)=0

Step 3.

Is the initial table optimal? The answer is definitely no. The table is

optimal if the value in the row Z j−C j ≥ 0.

Step 4.

Seek a better solution through literacy.

1. Determine the key column, which is the column that has the largest negative Z -C jj

value, in this case column x1 . Thus x1 will be included in the base.

2. Determines the key row, which is the row that has a number. The smallest and

non-negative index, in this case S3 . Thus S3 will come out of the base and its

place will be replaced by the variable x1 .

3. Determine the key number, which is the number contained in the crossing of the

key column with the key row, in this case the key number = 2.

4. Find the new number in the key row, by dividing all the numbers in the key row

by the key number.

New numbers = 40/2, 2/2, 1/2, 0/2, 0/2. 1/2.

 Or = 20,1, 1/2, 0,0,1/2.

5. Find the new number on the other row i.e:

Line S1 .

Old number = [ 60 2 3 1 0 0 ]

New key row number= [ 20 1 1/2 0 0 1/2 ] (2)

New number = [ 20 0 2 1 0 -1 ]

Line S2 .

Old number = [ 30 0 2 0 1 0 ]

New key row number= [ 20 1 1/2 0 0 1/2 ] (0)

New number = [ 30 0 2 0 1 0 ]

Line Z -Cjj .

Old number = [ 0 -40 -30 0 0 0 ]

New key row number= [ 20 1 1/2 0 0 1/2 ] (-40)

New number = [ 800 0 -10 0 0 20 ]

The results of the above calculations will appear in the new simplex table, Table

4.3, which is the result of the first iteration.

 Table 4.3. Iteration Table 1

Vrb. Cj 40 30 0 0 0
CB Index
in line bj x1 x2 S1 S2 S3

0 S1 20 0 2 1 0 -1 20/2=10

0 S2 30 0 2 0 1 0 30/2=15

40 x1 20 1 1/2 0 0 1/2 20/0.5=40

Z -Cjj 800 0 -10 0 0 20

Is the iteration 1 table optimal? The answer is no. By repeating the steps above, the

results of iteration 2 will appear as in Table 4.4 below.

Table 4.4. Iteration 2 Table (Optimum Table)

Vrb. Cj 40 30 0 0 0
CB Index
in line bj x1 x2 S1 S2 S3

30 x2 10 0 1 1/2 0 -1/2

0 S2 10 0 0 0 -1 1

40 x1 15 1 0 0 -1/4 3/4

Z -Cjj 900 0 0 0 5 15
Optimal solution Table 4.4 shows that the total value of Z = 900 with the value of each

decision variable x1 = 15 and x2 = 10.

In Table 4.5 (S1 = S3 = 0 is a non-base variable).

Table 4.5

Coefficient
Variable value
Base variable objective x
base
function

x2 30 x 10 = 300

S2 0 x 10 = 0

x1 40 x 15 = 600

_____

Total = 900

Conclusion:

1. In iteration 2 table is the final simplex table, with the optimal solution is

x1 (astro) = 15 units

x2 (cosmos) = 10 units

Z (profit) = Rp.900,00

2. The second constraint (raw material B) still remains as much as 10 kg which is

shown by the value of S2 = 10, in the optimal table.

3. Constraints 1 and 3 have no remainder (full capacity), which is indicated by the

value S1 = S3 = 0 (non-base variable). This can also be proven by inserting the

values of x1 and x2 into constraints 1 and 3.

Constraint 1: 2x1 + 3x2 = 60

2(15) + 3(10) = 60

60= 60

raw materials used = available

Constraint 3: 2x1 + 1x2 = 40

2(15) + 1(10) = 40

40= 40

Working hours used = available

5. SOME OTHER FORMS

In section 4 of this chapter, we have illustrated the simplex method used to solve

maximization problems, and all constraints are of the form smaller than (s). In this

section, we will discuss another form of LP, namely the minimization problem and

constraints of the form greater than equal to (s) or equal to (=).

Minimization Problem

The minimization problem is very likely to be encountered in the LP formulation.

How to solve an LP problem if the objective function is in the form of minimization?

Suppose the objective function is Z min = 40x1 + 30x2

To deal with this problem, there are two methods that can be done, namely:
Method 1

Change the minimum objective function to the maximum. The trick is to multiply

the minimum objective function by minus one. For example, the objective function Z

min = 40x1 + 25x2 , the maximum becomes -Z* mak. = Z = -40x1 - 25x2 .

If this is done, the following conditions apply.

[1] The simplex table ends (optimal) if the value contained in the row Zj - Cj ≤ 0.

[2] In the initial table, the value in the row Z j - Cj that corresponds to the decision

variable is positive.

[3] The key column is selected from the largest positive value.

[4] The key row still follows the minimum comparison rule and is non-negative.

[5] The next iteration process is the same as the previous one.

Example 4.1 Product Mix Minimization Problem

An LP problem that has been formulated is as follows:

Minimum Z = 40x1 + 25x2

d.k [1] 3x1 + 2x2 ≤ 150

[2] 8x1 + 2x2 ≤ 200

[3] x1 ≥0

[4] x2 ≥0

The LP formulation above can be converted into standard form with the objective

function converted into maximum form.

Maximum - Z* = Z = -40x1 - 25x2 - 0S1 + 0S2

d.k [1] 3x1 + 2x2 + S1 + 0S2 = 150

[2] 8x1 + 2x2 + 0S1 + S2 = 200

u.h x1 , x2 , S1 , S2 ≥0

By following the steps of the simplex method, the solution to the product mix

minimization problem is as follows.

Table 4.6. Initial Table of Minimization Problem SImplex

Vrb. Cj - 40 -25 0 0
CB Index
in line bj x1 x2 S1 S2

0 S1 150 3 2 1 0 150/3=50

0 S2 200 8 2 0 1 200/8=25

Z -Cjj 0 40 25 0 0

Table 4.7. Iteration Table 1

Vrb. Cj - 40 -25 0 0
CB Index
in line bj x1 x2 S1 S2

0 S1 75 0 1,25 1 -3/8 75/1,25=60

-40 x1 25 1 0,25 0 1/8 25/0,25=100

Z -Cjj -1.000 0 15 0 -5

Table 4.8. Iteration 2 (Optimum) Table

Vrb. Cj - 40 -25 0 0
CB Index
in line bj x1 x2 S1 S2

-25 x1 60 0 1 0,8 -0,3

 -40 x2 10 1 0 -0,2 0,2

Z -Cjj -1.000 0 1 -12 -0,5

The results of Table 4.8 of the 2nd iteration, show that the optimum solution is: x 1 = 10,

and x2 = 60.

Minimum Z = - Z* = - 40(10) - 25(60)

= - (-1.900) = 1.900.

Method 2

In this method, we do not change the form of the objective function, but the minimum

objective function is directly included in the table (it remains as the original).

If this is done, the following conditions apply:

[1] The simplex table ends (optimal) if the value contained in the row Zj - Cj ≥ 0.

[2] Since the objective function is a minimum, the key column is chosen to have the

smallest negative value in the row Zj - Cj .

[3] The key row still follows the minimum comparison rule and is non-negative.

[4] The next iteration process is the same as the previous one.

Refer back to Example 4.1 above. The standard form of the product mix minimization

problem is as follows:

Minimum Z = 40x1 + 25x2 + 0S1 + 0S2

d.k [1] 3x1 + 2x2 + S1 + 0S2 = 150

[2] 8x1 + 2x2 + 0S1 + S2 = 200

u.h x1 , x2 , S1 , S2 ≥0
If the standard form is solved according to method 2, the result is as follows:

Table 4.9. Initial Simplex Table of Minimization Problem

Vrb. Cj - 40 -25 0 0
CB Index
in line bj x1 x2 S1 S2

0 S1 150 3 2 1 0 150/2=75

0 S2 200 8 2 0 1 200/2=100

Z -Cjj 0 -40 -25 0 0

Table 4.10 Iteration Table 1

Vrb. Cj - 40 -25 0 0
CB Index
in line bj x1 x2 S1 S2

0 x2 75 1,5 1 0,5 0 75/1,5= 50

0 S2 50 5 0 -1 1 50/5 = 10

Z -Cjj 1.875 -5/2 0 12,5 0

Table 4.11 Iteration 2 (Optimum) Table

Vrb. Cj - 40 -25 0 0
CB Index
in line bj x1 x2 S1 S2

25 x2 60 0 1 0,8 -0,3

40 x1 10 1 0 -0,2 0,2

Z -Cjj 1.900 0 1 12 0,5

The optimum table of the two methods shows the same results, namely x1 = 10 units and

x2 = 60 units with a total value of Z = Rp. 1,900.00.

Constraints of the Form Greater Equals (≥)

The solution of LP problem with simplex method, requires an initial basic feasible

solution, i.e. each constraint values the base variable. If we have an LP that has

constraints greater than equal to (≥), for example 2x +3x 12 ≥ 30, it can be converted into

an equation by subtracting the surplus variable to the right of the constraint as follows:

2x1 + 3x2 - S = 30, where S ≥ 0. The constraint equation has no base variable.

Therefore, the initial table cannot be created yet.

To form an initial table of flexible simplex, one artificial variable, Q, can be

added, which acts as a base variable. So the constraint changes to: 2x 1 + 3x2 - S + Q =

30, where Q is an artificial variable. The use of pseudo variables is a very systematic

way to form the initial simplex table.

Constraints of the Form Equal to (=)

Constraints of the form equal to (=) also have no basis variable. Therefore, add

one pseudo base variable, so that the initial simplex table can be formed. For example,

2x +4x12 = 20, can be changed to 2x1 + 4x2 + Q = 20, where Q is a pseudo base variable.

Although all constraints have base variables, the addition of pseudo-variables is

not a flexible solution to the original problem. The pseudo-variables must be reduced to

zero. There are two methods that can be used to zeroize all variables, namely:

[1] Large M method.

[2] Two-phase method (two stages).

Big M Method

In this method, the objective function coefficients for the pseudo-variables are

given very large values of negative M or - M for the maximum objective function and

positive M or +M for the minimum objective function.

Example 4.2. Pseudo Variable Problem

The formulated LP model is as follows:

Maximum Z = 50x1 + 80x2

d.k [1] x1 ≤ 40

[2] x2 ≥ 20

[3] x1 + x2 = 50

[4] x1 , x2 ≥0

The LP model above is converted into standard form with the addition of pseudo-

variables as follows:

Maximum Z = 50x1 + 80x2 + 0S1 - 0S2 - MQ1 - MQ2

d.k [1] x1 + S1 + 0S2 + 0Q1 + 0Q2 = 40

[2] x2 + 0S1 - S2 + Q1 + 0Q2 = 20

[3] x1 + x2 + 0S1 + 0S2 + 0Q1 + Q2 = 50

[4] x1 , x2 , S1 , S2 , Q1 , Q2 ≥ 0

The initial simplex table can be made as follows:

Table 4.12. Initial Table of Large M Method Simplex

 Vrb. Cj 50 80 0 0 -M -M
CB Index
in line bj x1 x2 S1 S2 Q1 Q2

0 S1 40 1 0 0 0 0 0 40/0 = ~

-M Q1 20 0 1 0 -1 1 0 20/1 = 20

-M Q2 50 1 1 0 0 0 1 50/1 = 50

Z -Cjj -70M -M-50 -2M-80 0 M 0 0

The value contained in row Z -Cjj Table 4.12 is filled in using the following method.

[]
40
Z=[0 ,−M ,−M ] 20 −0=0−20 M −50 M =−70 M
50

[]
1
Z1 =[ 0 ,−M ,−M ] 0 −50=0−M −50=−M −50
1

[]
0
Z 2=[ 0 ,−M ,−M ] 1 −80=0−M −M −80=−2 M −80
1

[]
0
Z 4=[ 0 ,−M ,−M ] −1 −0=0+ M +0=M
1

And so on.
By following the previous steps of the simplex method, the solution of Example 4.2 can

be seen below.

Table 4.13. Iteration Table 1

Vrb. Cj 50 80 0 0 -M -M
CB Index
in line bj x1 x2 S1 S2 Q1 Q2

0 S1 40 1 0 0 0 0 0 40/1 = 40

80 x2 20 0 1 1 -1 1 0 20/0 = ∞

-M Q2 30 1 1 0 0 -1 1 30/1 = 30

Z -Cjj -30M -M 0 0 M 2M 0

1.600 -50 -80 +80

Table 4.14. Iteration 2 (Optimum) Table

Vrb. Cj 50 80 0 0 -M -M
CB Index
in line bj x1 x2 S1 S2 Q1 Q2

0 S1 40 1 0 1 0 0 0

80 x2 50 1 1 0 0 1 1

0 S2 30 1 0 0 1 -1 1

Z -Cjj 4.000 30 0 0 0 M M

+ 80

The optimum solution is reached when x1 = 0 and x2 = 50, with a value of Z = 4,000.
Two-Phase Method

In the two-phase method, the solution is separated into two stages. Each stage

uses a simplex table and the work process still uses the steps of the simplex method.

Phase 1.

The first stage aims to zeroize/eliminate pseudo-variables, by creating a pseudo

objective function. The pseudo objective function has the same number of variables as

the number of pseudo variables. Then the objective function of all is maximized with

the simplex table. The objective function coefficient for all variables is given a value of

minus one or (-1) if the objective function is maximum and plus one or (+1) if the

objective function is minimum. Phase one ends when the pseudo objective function has

zero value. The process continues to the second phase.

Look back at Example 4.2 above. There are two pseudo-variables Q 1 and Q2 .

Therefore, the pseudo objective function is Maximum Z = -Q 1 - Q2 . We maximize this

pseudo objective function, so the first phase solution appears as follows.

Table 4.15. First Phase Starting Table

Vrb. Cj 0 0 0 0 -1 -1
CB Index
in line bj x1 x2 S1 S2 Q1 Q2

0 S1 40 1 0 1 0 0 0 40/0 = ~

-1 Q1 20 0 1 0 -1 1 0 20/1 = 20

-1 Q2 50 1 1 0 0 -0 1 50/1 = 50

Z -Cjj -70 -1 -2 0 1 0 0
 Table 4.16. Iteration Table 1 First Phase

Vrb. Cj 0 0 0 0 -1 -1
CB Index
in line bj x1 x2 S1 S2 Q1 Q2

0 S1 40 1 0 1 0 0 0 40/0 = ~

0 x2 20 0 1 0 -1 1 0 20/-1 = -20

-1 Q2 30 1 0 0 1 -1 1 30/1 = 30

Z -Cjj -30 -1 0 0 -1 2 0

Table 4.17. Iteration Table 2 First Phase (Optimum)

Vrb. Cj 0 0 0 0 -1 -1
CB Index
in line bj x1 x2 S1 S2 Q1 Q2

0 S1 40 1 0 1 0 0 0

0 x2 50 1 1 0 0 0 1

0 S2 30 1 0 0 1 -1 1

Z -Cjj 30 1 0 0 0 1 1

In Table 4.17 (optimum) the pseudo objective function has been optimized, and the

pseudo variables Q1 and Q2 are out of the base. The process can proceed to the second

phase.

If the pseudo variable is still in the base with a positive value, then the problem is

not feasible. It may be an error in the calculation process or an error in the LP

formulation. There is no need to continue with the second stage of the process.

Phase 2.
 The end table of the first phase is the start table of the second phase. It is then

optimized by entering the original objective function. Because in the first phase we have

zeroed out the pseudo-variables, in the second phase the pseudo-variables do not need to

be included in the table anymore (omitted). See the initial table of the second phase

below.

Table 4.18. Second Phase Starting Table

Vrb. Cj 0 0 0 0
CB Index
in line bj x1 x2 S1 S2

0 S1 40 1 0 1 0

80 x2 50 1 1 0 0

0 S2 30 1 0 0 1

Z -Cjj 4.000 30 0 0 0

After the original objective function coefficients are entered into the initial table

of the second phase, the initial table immediately shows the optimum table. Because the

value contained in the row Z -Cjj ≥ 0. The optimum solution is x1 = 0 and x2 = 50.

Comparing the large M method with the two-phase method, it can be concluded

that both methods use pseudo variables. The difference lies in the stages of completion.

In addition, the large M method is more complicated. What needs to be considered in

using the large M and two-phase methods are:

[1] Pseudo-variables are only added to get an initial solution that is flexible. If we use

a computer program such as QSB (Quantitative System for Business), then the
 addition of pseudo variables is not necessary, because QSB has been programmed

in such a way as to deal with various forms of constraints.

[2] If the pseudo-variable has left the base, then in the next table the pseudo-variable

does not need to reappear.

[3] In the optimum table all pseudo-variables must come out of the base. If the

pseudo-variables are still in the base with positive values, then the problem is not

feasible.

6. SOME SPECIAL PROBLEMS IN THE SIMPLEX METHOD

In chapter 3 section 4, it was mentioned that there are three special problems in

linear programming, namely: [1] multi optimum solutions, [2] np feasible solutions, and

[3] unbounded solutions. Each problem is illustrated in a graphical example. In this

section we will discuss these three special problems using the simplex method.

Multi Optimum Solutions

Multi optimum solutions (alternative optimum), occurs when a non-base variable

in the optimum table has a value of zero in row Z -C jj . if this variable is included in the

base, it will change the value of the decision variable. But it will not change the optimal

value of the objective function Z (the optimal value of the objective function is the

same). See the following optimum table.

Table 4.19. Optimum Table

Vrb. Cj 0 0 0 0 0
CB Index
base bj x1 x2 S1 S2 S3

0 S1 58 0 0 1 -0,2 1,6
 2 x2 12 0 1 0 0,2 -0,6

3 x1 42 1 0 0 1,2 0,4

Z -Cjj 150 0 0 0 1 0

If we look at the optimum table above, there are two non-base variables, namely

S2 and S3 . The variable S3 has zero value in the row Z -Cjj , this means there are

multiple optimum solutions. In other words, there is another combination between x 1

and x2 , which produces the same Z value. How to find the alternative optimum? Put the

variable S3 into the base and with the minimm comparison rule, the variable S 1 will

come out of the base (see the following table).

Table 4.20. Optimum Substitute Table

Vrb. Cj 0 0 0 0 0
CB Index
Base bj x1 x2 S1 S2 S3

0 S3 36,25 0 0 5/8 -1/8 1

2 x2 33,75 0 1 3/8 -1/8 0

3 x1 27,50 1 0 1/4 1/4 0

Z -Cjj 150 0 0 0 1 0

Thus it can be concluded that the optimum solution is:

alternative 1:

x1 = 42 and x2 = 12

Z = 3(42) + 2(12) = 126 + 24 = 150

Alternative 2:

x1 = 27.50 and x2 = 33.75

Z = 3(27,50) + 2(33,75) = 82,50 + 67,50 = 150

Unbounded Solutions

Another problem that may arise from the minimum comparison rule is if the base

variable that should be excluded cannot be determined, because none of the variable

coefficients have positive values. Thus, the minimum comparison rule cannot be

performed even though the table is not yet optimum. Such cases are called unbounded

solutions (see the following example.

Table 4.21. Initial Simplex Table

Vrb. Cj 0 0 0 0
CB Index
Base bj x1 x2 S1 S2

0 S1 60 3 -3 1 0

0 S2 120 -9 3 0 1 120/3 = 40

Z -Cjj 0 -6 -9 0 0
 Table 4.22. Iteration Table (Unbounded Solution)

Vrb. Cj 0 0 0 0
CB Index
Base bj x1 x2 S1 S2

0 S1 180 -6 0 1 0

9 x2 40 -3 1 0 1/3

Z -Cjj 360 -33 0 0 3

In Table 4.22 of the first iteration, x1 should be in the base, because the value in

row Z -Cjj is still negative 33. But the variable coefficients contained in column x 1 are

not positive. Therefore, the key row cannot be determined.

No Feasible Solutions

If one or more artificial variables in the optimum table are still in the base with

positive values, the linear program is not feasible (no feasible solution). See Example

4.3 below.

Maximum Z = 2x1 + 5x 2

d.k [1] x1 + x2 ≤ 10

[2] 2x1 + 3x2 ≥ 48

[3] x1 , x2 ≥ 0

The linear program model is converted into standard form into:

Maximum Z = 2x1 + 5x2 + 0S1 + 0S2 - MQ 1

d.k [1] x1 + x2 + S1 + 0S2 + 0Q1 = 10

[2] 2x1 + 3x2 + 0S1 - S2 + Q1 = 48

 [3] x1 , x2 , S1 , S2 , Q1 ≥ 0

Table 4.23. Initial Simplex Table Example 4.3

Vrb. Cj 2 5 0 0 -M
CB Index
in line bj x1 x2 S1 S2 Q1

0 S1 10 1 1 1 0 0 10/1 = 10

-M Q1 48 2 3 0 -1 1 48/3 = 16

Z -Cjj -48M -2M -3M 0 M 0

-2 -5

Table 4.24. Iteration Table 1 Example 4.3

Vrb. Cj 2 5 0 0 -M
CB Index
in line bj x1 x2 S1 S2 Q1

5 x2 10 1 1 1 0 0

-M Q1 18 -1 0 -3 -1 1

Z -Cjj 50 M 0 3M M 0

-18M +3 +5

Table 4.24 of the first iteration above is an optimum table, because the value in

the row Z -Cjj ≥ 0. But the pseudo variable Q 1 is still in the base with a positive value of

18. Therefore Example 4.3 is called no feasible solutions. Errors may occur when

formulating the LP model.

Degeneracy
 The degeneracy problem arises because the decision variable contained in the base

is zero (see example 4.4) below.

Maximum Z = 4x1 + 1x2 + 3x3 + 0S1 + 0S2 + 0S 3

d.k [1] 2x1 - 2x2 + S1 + 0S2 + 0S3 = 60

[2] 4x1 + 2x3 + 0S1 + S2 + S3 = 120

[3] 2x1 + 2x2 + 2x3 + 0S1 + 0S2 + S3 = 90

[4] x1 , x2 , x3 , S1 , S2 , S3 ≥ 0

The process of solving Example 4.4 will appear as follows:

Table 4.25. Initial Simplex Table for Example 4.4

Vrb. Cj 4 1 3 0 0 0
CB Index
base bj x1 x2 x3 S1 S2 S3

0 S1 60 2 -2 0 1 0 0 60/2 = 30

0 S2 120 0 0 2 0 1 0 120/4 = 30

0 S3 90 0 2 2 0 0 1 90/2 = 45

Z -Cjj 0 0 -1 -3 0 0 0

Table 4.26. Iteration Table 1 for Example 4.4

Vrb. Cj 4 1 3 0 0 0
CB Index
base bj x1 x2 x3 S1 S2 S3

4 x1 30 1 -1 0 1/2 0 0 30/-1 = -30

0 S2 0 0 4 2 -2 1 0 0/4 = 0

0 S3 30 0 4 2 -1 0 0 30/4 = 7,5

Z -Cjj 120 0 -5 -3 2 0 0
 Table 4.27. Iteration Table 2 for Example 4.4

Vrb. Cj 4 1 3 0 0 0
CB Index
base bj x1 x2 x3 S1 S2 S3

4 x1 30 1 0 1/2 0 1/4 0 30/0,5 = 60

1 x2 0 0 1 1/2 -1/2 1/4 0 0/0,5 = 0

0 S3 30 0 0 0 1 -1 1 30/0 = ~

Z -Cjj 120 0 0 -1/2 -1/2 1,25 0

Table 4.28. Iteration Table 3 for Example 4.4

Vrb. Cj 4 1 3 0 0 0
CB Index
Base bj x1 x2 x3 S1 S2 S3

4 x1 30 1 -1 0 0,5 0 0 30/0,5 = 60

3 x3 0 0 2 1 -1 0,5 0

0 S3 30 0 0 0 1 -1 1 30/1 = 30

Z -Cjj 120 0 1 0 -1 1,5 0

Table 4.29. Iteration Table 4 for Example 4.4

Vrb. Cj 4 1 3 0 0 0
CB Index
Base bj x1 x2 x3 S1 S2 S3

4 x1 15 1 -1 0 0 1/2 -1/2

3 x3 30 0 2 1 0 -1/2 1

0 S1 30 0 0 0 1 -1 1

Z -Cjj 150 0 1 0 0 1/2 1

 In Table 4.27 and Table 4.28 the values of the decision variables x 2 and x3 are

zero. Although such cases may be rare, Example 4.4 above illustrates the possibility of a

table not changing the value of Z. This results in the simplex table being less efficient to

use. This makes the simplex table less efficient to use.

Negative Signed Decision Variable

One of the assumptions in the LP formulation is that all decision variables cannot

be negative. Similarly, in the standard form, all decision variables must be nonnegative.

In some cases it is possible or even desirable to have negative decision variables.

Maximum Z = 6x1 + 3x 2

d.k [1] 6x1 + 8x2 ≥ 48

[2] x1 + x2 ≤ 12

[3]-x1 + x2 ≤ 4

[4] 2x1 + x2 ≤ 2

u.h x1 ≥ -2

x2 ≥ 0

In the above example, the decision variable x 1 can be negative as indicated by x1 ≥-2. To

fulfill the assumptions in the LP formulation and the standard form, where the decision

variable must be nonnegative, it is necessary to make changes to the LP formulation

above. The way that can be done is to add a new variable to replace the decision

variable x1. The new variable is x 1 = x1 + 2, thus x1 = x1 ' = 2. The variable x 1 ' is a

replacement variable that will determine the magnitude of the variable x 1 .

The changes to the LP formulation above are as follows:

Maximum Z = 6(x1 ' - 2) + 3x 2

d.k [1] 6(x1 ' - 2) + 8x2 ≥ 48

[2] 1(x1 ' - 2) + x2 ≤ 12

[3] -1(x1 ' - 2) + x2 ≤ 4

[4] 2(x1 ' - 2) + x2 ≤ 2

u.h x1 ' ≥ 0

x2 ≥ 0

or,

Maximum Z = 6x1 ' + 3x2 - 12

d.k [1] 6x1 ' + 8x2 ≥ 60

[2] x1 ' + x2 ≤ 14

[3]-x1 ' + x2 ≤ 2

[4] 2x1 ' + x2 ≤ 6

u.h x1 ' ≥ 0

x2 ≥ 0

Infinite Decision Variables

If the decision variable is infinite, it is changed by adding two different

nonnegative variables in place of the infinite decision variable (see example below).

Maximum Z = x1 - 2x2 + 3x3

d.k [1] x1 + x2 + x3 ≤ 7

[2] x1 - x2 + x3 ≥ 2

[3] 3x1 + x2 + 2x3 = 5

[4] x1 ≥ 0

[5] x2 ≥ 0
 [6] x3 ≥ ~

In the above example, the decision variable x 3 is infinite. The way that can be done is to

change the variable x3 with him variables of different signs, namely x 4 - x5 where x4 ≥ 0,

and x5 ≥ 0. The value of x3 is positive or negative depending on the two variables. The

changes to the example above are as follows

Maximum Z = x1 - 2x2 + 3(x4 - x )5

d.k [1] x1 + x2 + 1(x4 - x5 ) ≤ 7

[2] x1 - x2 + 1(x4 - x5 ) ≥ 2

[3] 3x1 + x2 + 2(x4 - x5 ) = 5

[4] x1 ≥ 0

[5] x2 ≥ 0

[6] x4 ≥ 0

[7] x5 ≥ 0

or,

Maximum Z = x1 - 2x2 + 3x4 - 3x5

d.k [1] x1 + x2 + x4 - x5 ≤ 7

[2] x1 - x2 + x4 - x5 ≥ 2

[3] 3x1 + x2 + 2x4 - 2x5 ) = 5

[4] x1 ≥ 0

[5] x2 ≥ 0

[6] x4 ≥ 0

[7] x5 ≥ 0

Practice Questions

Problem 1
Solve the linear program below by simplex method
Mak Z = x1 + 3x2
d.k. (1) x 1 ≤5
(2) x1 + 2x2 ≤ 10
(3) x2 ≤4
(4) x1 ; x2 ≥0
Problem 2
Solve the linear program below by simplex method:
Min Z = 3x1 + x2 + x3 + x4
d.k. (1) -2x1 + 2x2 + x 3 =4
(2) 3x1 + x + 2 x4 =6
(3) x1 ; x2 ; x3 ; x4 ≥0
Check if there is an alternative solution, if there is look for an alternative solution.

Problem 3

The formulated linear program model is as follows:

Mak Z = 50x1 + 40x2
d.k (1) x1 + x2 ≥ 5
(2) 10x1 + 15x2 ≤ 150
(3) 20x1 + 10x2 ≤ 180
(4) 30x1 + 10x2 ≥ 135
(5) x1 - 3x2 ≤ 0
(6) x1 ; x2 ≥ 0

Solve the above linear program by simplex method.