Duality in Linear Programming

For every LPP there is a related unique LPP involving the same data that also describes
the original problem. The given original problem is called a primal problem. This can be solved
by transposing the rows and columns of the algebraic statements of the problem. Inverting the
problem in this way results in a dual problem. Duality is an extremely important concept in LPP.
The various useful aspects of this are:

1. If the primal problem contains a large number of constraints and a small number of
variables, the computational procedure can be reduced by converting it into dual and
then solve it.
2. It gives additional information as to how the optimal solution changes as a result of
the changes in the coefficients and the formulation of the problem. This is termed as
post optimality or sensitivity analysis.
3. Calculation of the dual checks the accuracy of the primal solution.

Dual problem when the primal is in the canonical form.

The general LPP in canonical form is written as

Maximize z = c1x1 + c2x2 + …….. + cnxn,

Subject to
a11x1 + a12x2 + …. + a1nxn ≤ b1 ,
a21x1 + a22x2 + …. + a2nxn ≤ b2 , ……….. (1)

am1x1 + am2x2 + …. + amnxn ≤ bm ,

x1 , x2 , …….. , xn ≥ 0.

If the above is referred as primal, then its associated dual will be

Minimize w = b1y1 + b2y2 + …….. + bmym,

Subject to
a11y1 + a21y2 + …. + am1ym ≥ c1 ,
a12y1 + a22y2 + …. + am2ym ≥ c2 , ………. (2)

a1ny1 + a2ny2 + …. + amnym ≥ cn ,

where the associated dual variables y1 , y2 , …. , ym ≥ 0.

Using the matrix notation, the primal can be written as

Maximise z = cTx, where cT ∈ Rn , x ∈ Rn.

Subject to
Ax ≤ b ,
x ≥ 0 where A is an m×n matrix and bT ∈ Rm
The dual problem can be written as

Minimise y = bTw

Subject to
ATw ≥ c ,
w ≥ 0.

Note

1. If the primal contains n variables and m constraints, then the dual will contain m
variables and n constraints.
2. The maximization in the primal becomes the minimization in the dual and vice versa.
3. The maximization problem has ( ≤ ) constraints while the minimization problem has
( ≥ ) constraints.
4. The constants c1 , c2 , … cn in the objective function of the primal appear in the
constraints in the dual.
5. The constants b1 , b2, …, bm in the constraints of the primal appear in the objective
function of the dual.
6. The variables in both problems are non-negative.

Problem 1

Construct the dual of the problem

Minimize z = 3x1 – 2x2 + 4x3,

Subject to constraints

3x1 + 5x2 + 4x3 ≥ 7

6x1 + x2 + 3x3 ≥ 4
7x1 – 2x2 – x3 ≤ 10
x1 – 2x2 + 5x3 ≥ 3
4x1 + 7x2 – 2x3 ≥ 2
x1, x2, x3 ≥ 0

Remark

1. If the primal problem is in standard form, then the dual variables will always be
unrestricted in sign (irrespective of the sign of the primal ones). This condition
however will be satisfied only for those primal constraints that were initially in
equation.
2. If some primal variables are unrestricted in sign, then those dual constraints will be
equations that correspond to the said primal variables.
Problem 2

Construct the dual of

Maximize z = 3x1 + 10x2 + 2x3,

Subject to
2x1 + 3x2 + 2x3 ≤ 3
3x1 – 2x2 + 4x3 = 3
x1, x2, x3 ≥ 0.

Problem 3

Obtain the dual of

Maximize z = x1 + 3x2 – 2x3 + 5x4,

Subject to
3x1 – x2 + x3 – 4x4 = 2
5x1 + 3x2 – x3 – 2x4 = 3
x1, x2 ≥ 0
and x3, x4 are unrestricted in sign.

Properties of Primal and Dual optimal solutions.

The final simplex table giving optimal solution to the primal problem also contains the
optimal solution to the dual and vice versa.

1. The optimal value of the objective function of the primal is equal to the optimal
value of the objective function of the dual.

Zmax.= Wmin. or Zmin.= Wmax.

2. If the primal (dual) variable corresponding to a slack starting variable in the dual
(primal) problem, its optimum value is directly read off from the net evaluation
row of the optimum dual (primal) simplex table, as the net evaluation
corresponding to this slack variable.
3. If the primal (dual) variable corresponds to an artificial starting variable in the
dual (primal) problem, its optimum value is directly read off from the net
evaluation row of the optimum dual (primal) simplex table, as the net evaluation
corresponding to this artificial variable after deleting the constant M.
Problem

Using duality solve the following LPP

Maximize z = 3x1 – 2x2,

Subject to
x1 ≤ 4
x2 ≤ 6
x1 + x2 ≤ 5
–x2 ≤ –1
x1, x2 ≥ 0.

Dual Simplex Algorithm

1. Convert the minimization LPP into that of maximization, if it is in the minimization

form. Convert the ≥ type in equations, representing the constraints of the given LPP,
if any, into ≤ type.
2. Introduce the slack variables in the constraints of the given problem and obtain an
initial basic solution. Put this solution in the starting dual simplex table.
3. Test the nature of zj – cj in the simplex table.
(i) If all zj – cj and xBi are non-negative, for all i and j, then an optimum basic
feasible solution has been obtained.
(ii) If all zj – cj are non-negative and at least one of xBi < 0 , then go to
step 4.
(iii) If at least one zj – cj < 0, the method is not applicable to the given
problem.
4. Select the most negative xBi. The corresponding basis vector then leaves the basis YB. B

Let xBk be the most negative basic variable so that yk leaves the basis YB. B

5. Test the nature of ykj, j = 1,2, …, n.

(i) If all ykj ≥ 0, there does not exist any feasible solution to the given
problem.
(ii) If at least one ykj < 0, compute the replacement ratio {(zj – cj) / ykj, for ykj
< 0}, j = 1, 2, …., n. and choose the maximum of these. The
corresponding column vector, say, yr, then enters the basis yB. B

6. Test the new iterated dual simplex table for optimality.

Repeat the procedure until either an optimum feasible solution has been obtained (in a
finite number of steps) or there is an indication of the non-existence of a feasible solution.

Problem

Use dual simplex method to

Minimize z = x1 + 2x2 + 3x3

Subject to
x1 – x2 + x3 ≥ 4
x1 + x2 + 2x3 ≤ 8
x2 – x3 ≥ 2
x1, x2 , x3 ≥ 0.
Economic Interpretations

The shadow/dual price equivalency states that the optimal value of the ith dual variable is the
amount by which the optimal value of the primal objective function will change per allowable
unit increase in bi, all other parameters held as constant. For a maximization primal where the ≤
constraints represent limited resources, the corresponding dual price is measure of the marginal
value of additional units of a resource.

Base on these economic interpretations, the canonical dual problem is often called a
pricing problem. Its optimal solution yields the minimum prices that management should accept
for selling or liquidating resources; the optimal value of the objective function is the minimum
acceptable payment for all these resources.

One of the important properties of the duality theory is, for any pair of feasible primal
and dual solutions, (objective value in the maximization problem) ≤ (objective value in the
minimization problem). At the optimum, the relationship holds as a strict equation.
By this property, we have

n m
z = ∑ c j x j ≤ ∑ bi y i = w
j =1 i =1

The strict equality holds when both primal and dual solutions are optimal. The primal problem
represents a resource allocation model and z as profit dollars.

The dual variables yi `s represents the worth per unit of resource i and if z < w. That is,
(profit) < (worth of resources), the corresponding primal and dual solutions cannot be optimal.

Fundamental theorem of Duality

If both the primal and dual problems have feasible solutions, then both have optimal solutions and
max z = min w.

Remark

The dual of the dual is the primal problem.