Dual simplex method

Any LPP for which it is possible to find infeasible but better than optimal initial basic solution can be
solved by using dual simplex method. Such a situation can be recognised by first expressing the
constraints in <= form and the objective function in the maximisation form.

After adding slack variables, if any RHS element is negative and the optimality condition is satisfied
then the problem can be solved by dual simplex method.

Negative element on RHS system that the corresponding slack variables are negative.

This means that the problem starts with optimal but infeasible basic solution and we proceed
towards its feasibility.

The dual simplex method is similar to the standard simplex method except that in the latter the
starting initial basic solution is feasible but not optimum while in the former it is infeasible but
optimum or better than optimum.

The dual simplex method works towards feasibility while simplex method works towards optimality.

Computational procedure of dual simplex method

The iterative procedure is as follows:

1. First convert the min. LPP into max form (if Min.)
2. Convert the >= type inequalities of given LPP, if any into those of <= type by multiplying the
corresponding constraint by -1.
3. Introduce slack variables in the constraints of the given problem and obtain an initial basic
solution.
4. Test the nature of ∆j in the starting table
a) If all ∆j and XB are non-negative, then an optimum basic feasible solution has been
attained.
b) If all (∆j) are non-negative and at least one basic variable (XB) is negative then go to step
5.
c) If at least (∆j) one is negative, the method is not appropriate
5. Select the most negative (XB)
- The corresponding basis vector then leaves the basis set B.
- Let Xr be the most negative basic variable.
6. Test the nature of Xr
- If all Xr are non-negative, then there does not exist any feasible solution to the given
problem.
- If at least one Xr is negative, then compute max ∆j/ Xr and determine the least negative
for incoming vector.
7. Test the new iterated dual simplex table for optimality.
8. Repeat the entire procedure until either one optimum feasible solution has been attained in
finite number of steps.
Min Z = 2X1 + X2

S.t. 3X1 + X2>= 3

4X1 + X2>= 6

X1 + 2X2>= 3

SOLUTION:

1. Max Z’ = -2X1 – X2

S.t. -3X1 – X2<= -3

-4X1 – 3X2<= - 6

-X1 – 2X2 <= -3

X1 , X2 >= 0

2. Max Z’ = -2X1 – X2

S.t. -3X1- X2 + S1 = -3

-4X1 – 3X2 + S2 = -6

-X1 – 2X2 + S3 = -3

Xj , Sj >= 0

Cj -2 -1
Basis X1 X2 S1 S2 S3 RHS
S1 -3 -1 1 0 0 -3
S2 -4 -3 0 1 0 -6
S3 -1 -2 0 0 1 -3
Z 2 1 0 0 0 0
2/-4 1/-3 aij < 0

Basis X1 X2 S1 S2 S3 RHS
S1 -5/3 0 1 -1/3 0 -1
X2 4/3 1 0 -1/3 0 2
S3 5/3 0 0 -2/3 1 1
Z 2/3 0 0 1/3 0 -2
Ratio -2/5 0 0 -1 0

Basis X1 X2 S1 S2 S3 RHS
X1 1 0 -3/5 1/5 0 3/5
X2 0 1 4/5 -3/5 0 6/5
S3 0 0 1 -1 1
Z 0 0 2/5 1/5 0 -12/5
∆j >= 0 and XB >=0 therefore optimal

Min Z = 3 X1 + X2

S.t. X1 + X2 >= 1

2X1 + 3X2 >= 2

X1 and X2 >= 0
Artificial starting solution

LPS in which all the constraints are <= with non-negative RHS offer q convenient all slack starting
basic feasible solution. Models involving (=) &/or (>=) constraint do not

Artificial var obj coeff

-M, in max problem

+M, in min problem

e.g. Min Z = 4X1 + X2

S.t. 3X1 + X2 = 3

4X1 + 3X2 >= 6

X1 + 2X2 <= 4

X1, X2 >= 0

Min Z = 4X1 + X2

S.t. 3X1 + X2 = 3

4X1 + 3X2 – X3 = 6

X1 + 3X2 – X3 = 6

X1, X2, X3, X4 >= 0

Min Z = 4X1 + X2 + MR1 + MR2

S.t. 3X1 + X2 + R1 =3

4X1 + 3X2 – X3 + R2 =6

X1 + 2X2 + X4 =4

X1, X2, X3, X4, R1, R2 >= 0

R1 = 3

R2 = 6 M = 100

X4 = 4

Basis X1 X2 X3 X4 R1 R2 RHS
R1 3 1 0 0 1 0 3
R2 4 3 -1 0 0 1 6
X4 1 2 0 1 0 0 4
Z -4 -1 0 0 -100 -100 0
In the above table

X1 = X2 = X3 = 0 , <= yield

Z = 100 * 3 + 100 * 6 = 900

(instead of ‘0’ as the RHS of Z-row)

This inconsistency shows from the fact that R1 and R2 have non – zero coeff (-100, -100) in the Z-
row.

This inconsistency can be eliminated by

New Z- row = old Z- row + 100 * R1 row + 100 * R2 row

Modified table

Basis X1 X2 X3 X4 R1 R2 RHS Ratio

R1 3 1 0 0 1 0 3 1
R2 4 3 -1 0 0 1 6 1.5
X4 1 2 0 1 0 0 4 4
Z 696 399 -100 0 0 0 900

Z’ = old Z- row + 100 (3, 1, 0, 0, 1, 0, 3) + 100 (4, 3, -1, 0, 0, 1, 6)

= (-4, -1, 0, 0, -100, -100, 0) + 100 (3, 1, 0, 0, 1, 0, 3) + 100 (4, 3, -1, 0, 0, 1, 6)

= 696, 399, -100, 0, 0, 0, 900

Entering variable – (Max positive in Z – row for min)

Notice that Z = 900 which is consistent now with the values of the starting basic feasible solution

R1 = 3, R2 = 6, X4 = 4

Basis X1 X2 X3 X4 R1 R2 Solution Ratio

X1 1 1/3 0 0 1/3 0 1 3
R2 0 5/3 -1 0 -4/3 1 2 6/5
X4 0 5/3 0 1 -1/3 0 3 9/5
Z 0 167 -100 0 -232 0 204

Basis X1 X2 X3 X4 R1 R2 RHS Ratio

X1 1 0 1/5 0 3/5 -1/5 3/5
X2 0 1 -3/5 0 -4/5 3/5 6/5
X4 0 0 1 1 1 -1 1
Z 0 0 1/5 -232 -492/5 -501/5 18/5
R1’ = (1, 1/3, 0, 0, 1/3, 0, 1) – 1/3 (0, 1, -3/5, 0, -4/5, 3/5, 6/5)

R3’ = (0, 5/3, 0, 1, -1/3, 0, 3) – 5/3 (0, 1, -3/5, 0, -4/5, 3/5, 6/5)

RZ’ = (0, 167, -100, 0, -232, 0, 204) – 167 (0, 1, -3/5, 0, -4/5, 3/5, 6/5)

Basis X1 X2 X3 R1 R2 X4 Solution
X1 1 0 0 -2/5 0 -1/5 2/5
X2 0 1 0 -1/5 0 -3/5 9/5
X3 0 0 1 1 -1 1 1
Z 0 0 0 -493/5 -100 -1/5 17/5

Min Z = 17/5 (2/5, 9/5)

Unbounded solution

And unbounded solution is one in which the objective function value can be arbitrarily increased
without violating the constraints.

Ex.

Max Z = X1 + X 2

S.t. X1 >= 5

X2 <= 10

X1, X2 >= 0

Just by inspection one would arrive at the solution

X1 = ∞ , X2 = 10, And Z = ∞ (unbounded)

In LP problems, the constraints may be such that the feasible region may be unbounded at least in
one direction.

Sensitivity analysis :

Sensitivity analysis is an exercise of obtaining the new solution corresponding to a change in the data
of the original LP problem, given the original problem and the final simplex table, without solving
afresh the new problem with changed data. The starting point is the final simplex table of the
original problem, followed by an investigation of the extent to which the current optimal solution
will be affected because of a change in the data of the original problem. The new solution is
determined by further iterations. In many cases, the number of these iterations will be far less than
any those regarding to solve the problem afresh from its new starting solution.

A change in data of the original problem, may affect optimality or feasibility of the current solution.
Three cases are possible:
 - Changes that can affect only optimality- this can occur because of a change in the
coefficients in the objective function.
- Changes that can affect feasibility- this can occur because of a change in the right hand
side value.
- Changes that can affect both optimality and feasibility- this can occur because of a
simultaneous change in the coefficients in the objective function and in the right hand
side values.

Sensitivity analysis:

In LP, the parameters (input data) of the model can change within certain limits without causing the
optimal solution to change. This is referred to as sensitivity analysis.

Whereas the post optimal analysis which deals with determining the new optimum solution resulting
from making targeted changes in the input data.

Two cases will be considered:

1. Sensitivity analysis of the optimal solution to changes in the availability of the resources
(RHS of the constrained)
2. Sensitivity of the optimum solution to changes in unit profit or unit cost (coefficient of the
objective function)

E.g.

Max Z = 30X1 + 20X2

S.t. 2X1 + X2 <= 8

X1 + 3X2 <= 8

X1, X2 >= 0

A company produces 2 products on 2 machines. A unit of product 1 requires 2 hours on machine 1

and 1 hour on machine 2 for product 2. Unit 2 requires 1 hour on machine1 and 3 hours on machine
2. The revenues/unit of products 1 and 2 are $30 and $20 respectively. The total daily processing
time available for each machine is 8 hours.

Change in optimal solution when changes are made in the __________ of Machine 1 from 8 hours to
9 hours.

Rate of revenue change resulting

from increasing machine 1 capacity = ZG – ZC = 142 - 128 = $14.00/ hr
Capacity change 9-8
by 1 hour