7.

Dual problems and duality theorems

Let us start with an example.
Example 15. Let be given the following LP-programming problem with dual variables,
written aside to every main constraint:
max 2x1 - 5x2 dual variables
s.t. - 3x2 8 y1
x1 + 10x2 -5 y2
x0
We say that y1 is the dual variable, corresponding to the first main constarint and y2 the dual
variable, corresponding to the second main constarint. Then the dual problem,
corresponding to the initial problem, is
min 8y1 - 5y2
s.t. 0y1 + y2 2
-3y1 +10y2 -5
y0
Generally we can describe the formulation of the dual problem by the following
scheme. Let be given an initial problem with the dual variables, written aside to
corresponding main constraints of the initial problem:
max c1x1 + + cvxn dual var-s
s.t. a11x1++ a1nxn b1 y1
.
am1x1++ amnxn bm ym
x0
where c i , a ij , b i are constants. We denote

c = (c1, , cn),; A = (a ij ), i = 1,,m; j = 1,,n; b = (b1, , bm) T , x = (x1, , xm) T

Shortly we can formulate the initial problem as follows :
max cx
s.t. Ax b,
x 0
The corresponding dual problem we can formulate by the relations:
min b1y1 + + bmym
s.t. a11y1 + + am1ym c1
..
a1nyn + + amnym cn
y0
or shortly by the relations:
min yb
s.t. yA c
y0
where y = (y1,,ym).

If an initial problem is given by the relations

26
 min cx
s.t. Ax = b
x 0
then the corresponding dual problem is defined by the relations
max yb
s.t. yA c
We notice that if an initial problem has the equality constraints, then dual variables, according
them, have not any constraints in the corresponding dual problem. If in the initial problem a
variable is not constrained, then the corresponding constraint in the dual problem is an
equality.

Example 16. The dual problem for the problem

min x1+ 4x2
s.t. 3x1+ 2x2 = 3
x1+ 7x2 = -4
x0
is
max 3y1- 4y2
s.t. 3y1 + y2 1
2y1+ 7y2 4

Example 17. To formulate the dual problem for the problem

min 2x1 + 5x2
s.t. 3x1 + 4x2 = 11
x1 - 2x2 5
x1 0
We at first transform:the initial problem to the form
max -2x1- 5x2
s.t. 3x1+ 4x2 = 11
- x1 + 2x2 -5
x1 0
Now we can formulate the dual problem:
min 11y1 - 5y2
s.t. 3y1 - y2 -2
4y1 + 2y2 = -5,
y2 0

We present now some essential results about the duality relationship between LP-
programming problems.

Theorem 1. If x and y are feasible solutions of an initial and the corresponding dual problem
respectively, then cx yb.
Proof. We multiply the main constraints Ax b of an initial problem by y on the left:
y(Ax) yb
and the main constraints yA c of the corresponding dual problem by x on the right:

27
 (yA)x cx
As y(Ax) = (yA)x, then cx (yA)x yb.
Theorem 2. If x and y are such feasible solutions of an initial and the corresponding dual
problem respectively, for which objective functions. are equal (cx = yb), then x and y are the
optimal solutions of these problems.
Theorem 3. If x and y are optimal solutions of an initial and the corresponding dual problem
respectively, then cx = yb or the optimal values of the objective functions are equal.
Theorem 4. If the objective function of an initial problem is unbounded, then the
corresponding dual problem is contradictory. If the objective function of the dual problem is
unbounded, then an initial problem is contradictory.
The dual problem it is possible to solve with the help of the last simplex table of an initial
problem (an initial problem has been solved by ordinary or the dual simplex method). The
optimal values of the dual variables are equal to coefficients of the according additional
variables in zerothth row of the last simplex table of an initial problem.
Let us remember Example 5, where the initial problem (with the dual variables, corresponding
to the main constraints of the initial problem) was presented in the canonical form as follows:
max12 x1 + 9x2 dual variables
x1 + x3 = 1000 y1
x2 + x4 = 1500 y2
x1 + x2 + x5 = 1750 y3
4x1 + 2x2 + x6 = 4800 y4
x1,, x6 0
The last simplex table for this problem was:

x0 x1 x2 x3 x4 x5 x6 b
1 0 0 0 0 6 3/2 17 700
0 1 0 0 0 -1 1/2 650
0 0 0 0 1 -2 0 400
0 0 0 1 0 1 -1/2 350
0 0 1 0 0 2 -1/2 1100

As y1 correspond to x3; y2 to x4, y3 to x5 and y4 to x6, then the optimal solution of the
corresponding dual problem (which we advice to formulate independently by the reader) is:
y1* = 0 y2* = 0 y3* = 6 y4* = 3/2.
This solution we can interpretate so: if we add one handle, then the profit increases by 6$, if
we add one wheel, then we get 1,5$ additional profit, if we add one frame (larger or fewer),
then the profit does not change. Therefore these optimal solutions are called the shadow
prices.

Problems
13. Formulate the dual problems:
a ) min w = x1 + x2 b) max w = 7 x1 + 4 x2 c) max w = 2 x1 + 3 x2
x1 + 4 x2 = 1 4 x1 + 2 x2 1 x1 + 2 x2 = 1
2 x1 + x2 1 6 x1 + 2 x2 4 x1 =1
x1 , x2 0 x2 0 x1 0

28
 f ) min w = 2 x1 x2
d ) max w = 2 x1 x2 + 3 x3 e) min w = 2 x1 6 x2
2 x1 + x2 = 6
x1 + x2 + x3 = 12 x1 + x2 + x3 2
x1 + x2 = 5
2 x1 + x2 + 3 x3 6 2 x1 x2 + x3 1
x1 x2 = 2
x1 , x3 0
x1 0
.14. A company produces 2 different product and use for it 3 different materials. Datas about
the needness of these materials for producing 1 unit of both product are presented in the
following table:

Type of material I product II product

I 2 1
II 4 2
III 3 5

1) To find with the help of simplex method the profit maximizing producing plan, if each unit
of I product gives the profit 500 EU, each unit of II product 1000 EU and the availability of
first material is limited by 100, second material by 200 and third material by 300 unit.
2) To formulate the dual problem for initial problem.
3) To find the optimal solution for dual problem and to interpretate the answer.
15. A company produces sofas and armchairs. For producing these products are available
2100 m2 gobelin and 2400 kg poroloon; the resources of the other materials are not limited.
The datas about needness of materials and about the prices of products are given in the
following table:

Needness for materials for 1 unit

sofa armchair
Porolon 3 7
Gobelin 7 5
Costs for producing 1 unit 2400 1200
Price of one unit 2700 1400

1) To find the profit maximizing producing plan.

2) To formulate the dual problem for an initial problem.
3) To find the optimal solution for the dual problem and to interpretate the answer.

8. Transportation problem

Let be given m stores and n shops. We denote:

c ij - the price of transportation of the unit of article of a commercee from i-th store to j-th
shop; i = 1, , m; j = 1,, n;
a i - the goods reserve in i-th store,
b j - requirement of the goods in j-th shop.
All datas we can we can present in the following table:

29