INDR 262 Optimization Models and Mathematical Programming

DUAL SIMPLEX METHOD

In dual simplex method, the LP starts with an optimum (or better) objective function
value which is infeasible. Iterations are designed to move toward feasibility without violating
optimality. At the iteration when feasibility is restored, the algorithm ends.
In order to maintain optimality and move toward feasibility at each iteration, the
following two conditions are employed.

Dual Feasibility: The leaving variable xr is the basic variable having the most negative value.
If all the basic variables are nonnegative, the algorithm ends.

Dual Optimality: The entering variable is determined from among the nonbasic variables as
the one corresponding to
z c
min j j
, rj < 0
xj
rj

where rj is the constraint coefficient of the tableau associated with the row of leaving
variable xr and the column of the entering variable xj.

Example:
Primal:max 3x1+6x2+3x3 Dual: min 3y1+2y2
s.t. 3x1+4x2+x3 3 s.t. 3y1+ y2 3
x1+3x2+x3 2 4y1+3y2 6
x1, x2 0 y1+ y2 3
x3 0 y1, y2 0

min W
W-3y1-2y2 =0
3y1+ y2 y3 =3
4y1+3y2 - y4 =6
y1+y2 + y5 = 3
y1, y2, y3, y4, y5 0

1 Metin Turkay
 INDR 262 Optimization Models and Mathematical Programming

Initial Dual Simplex Tableau:

y1 y2 y3 y4 y5 RHS
-3 -2 0 0 0 0
y3 -3 -1 1 0 0 -3 Leaving variable

y4 -4 -3 0 1 0 -6
y5 1 1 0 0 1 3

Variable y1 y2 y3 y4 y5
row 0 (zj-cj) -3 -2 0 0 0
row 2 (4j) -4 -3 0 1 0

z j c j
ratio 3/4 2/3 - - -
4 j
Entering variable

Iteration 1)
y1 y2 y3 y4 y5 RHS
-1/3 0 0 -2/3 0 4
Leaving variable
y3 -5/3 0 1 -1/3 0 -1
y2 4/3 1 0 -1/3 0 2
y5 -1/3 0 0 1/3 1 1
Ratio 1/5 - - 2 -

Iteration 2)
y1 y2 y3 y4 y5 RHS
0 0 -1/5 -3/5 0 21/5
y1 1 0 -3/5 1/5 0 3/5
y2 0 1 4/5 -3/5 0 6/5
y5 0 0 -1/5 2/5 1 6/5

=> x1 = 1/5, x2 = 3/5

2 Metin Turkay
 INDR 262 Optimization Models and Mathematical Programming

The Simplex Method with Bounds

It is common in linear programming problems to have bounds on some of the variables.

These bounds can be either lower bounds, upper bounds or both. A variable with bounds is
represented as follows:
x Lj x j xUj

It is possible to treat these bounds as constraints and obtain standard notation by defining
slack, ecxess and artificial varaibles before applying the traditional simplex method.
However, this approach is not very efficient since it would lead to a dramatic increase in the
total number of variables and constraints. Therefeore, it is necessary to modeify the simplex
algorithm to handle the bounds on variables more effectively.

Lower Bounds:
The lower bounds can be handled with a minor modification in the model by change of
variables.
since x j x Lj , replace x j by xj = x j + x Lj 0

Therefore, the simplex method can directly be applied to the linear programming problem
without any modification after this new definition.

Upper Bounds:
The upper bounds can be replaced by,
x j + y j = xUj

when xj=0, then y j = xUj . Here, yj is referred to as the complementary variable. At any point
in the iterations of the simplex method, it os possible to,
1. use xj when 0 x j xUj or
2. replace xj by xUj y j when 0 y j xUj

This choice is arbitrary, since the variables xj and yj are complementary to each other (i.e.,
when the value for xj is known, it is possible to determine the value of yj and vice versa).

The upper bound technique uses the following rulet o maket his choice:
Rule: Begin with choice 1 (i.e., use xj). Whenever xj=0, use xj as the nonbasic variable.
Whenever x j = xUj , use yj as the nonbasic variable.

Example 1:
max z= 2x1 + x2 + 2x3
s.t. 4x1 + x2 =12
-2x1 + x3 = 4
0x14, 0x214, 0x36
x1 = 4, x 2U = 14, x 3U = 6
U

The simplex tableau is constructed as follows:

3 Metin Turkay
 INDR 262 Optimization Models and Mathematical Programming

B.V. Eqn. Z x1 x2 x3 RHS Ratio

z 0 1 -2 -1 -2 0
x2 1 0 4 1 0 12
x3 2 0 -2 0 1 4
z 0 1 -2 0 0 20 x3U = 6 Negative elements are eligible
x2 1 0 4 1 0 12 12/4=3 pivot elements indicating that the
levels of basic variables can
x3 2 0 -2 0 1 4 (6-4)/2=1 increase. The minimum ratio test
must be modified to limit the
z 0 1 0 0 1 22 increase to the upper bound.
x2 1 0 0 1 2 8
x1 2 0 1 0 1/2 1

replace x3=6-y3:
-2x1 + x3 = 4 becomes
-2x1 +6 - y3 = 4 rearranging
-2x1 - y3 = -2 converting to proper form
x1 +1/2 y3 = 1 can be inserted back into the simplex tableau

The optimal solution is:

z*=22, x1=1, x2=8, x3=6

4 Metin Turkay