333

CHAPTER FOURTEEN
14. Linear Programming
Linear Programming involves problems where a variable has to be maximised or
minimised subject to certain conditions. The conditions are called CONSTRAINTS and
the variable, the OBJECTIVE FUNCTION. They are represented in linear form.
It is best illustrated by example.
14.1

Example 1

A manufacturer makes products P1 and P2 by using machines M1 and M2. The time in
hours required to make P1 and P2 on each machine is shown in the table below.
P1
M1

M2

P2
1
2

The profits on P1 and P2 are $30 and $15 each respectively. The manufacturer operates
for 420 hours a month.
How many of each product P1 and P2 should be made per month to maximise profit?
The problem may be set up as follows.
Let x and y represent the number of P1 and P2 made respectively per month.
The profit = $30x + 15y is called the OBJECTIVE FUNCTION.
Time on machine M1 per month cannot exceed 420 hours.
1
! 4x + 1 y ! 420
2

Similarly the time condition of machine M2 leads to

3x + 2y ! 420
Clearly x ! 0 and y ! 0.
The problem may be summarised.
Maximise 30x + 15y OBJECTIVE FUNCTION

334

Subject to

x !0

(1)

y !0

(2)

1
4x + 1 y ! 420 (3)
2

3x + 2y ! 420

CONSTRAINTS

(4)

There are many ways of solving linear programming problems like this but for simple
two variables cases the simplest and most intuitive are by graphical methods. The
constraint may be drawn as follows.

(4)
B

(1)
(3)
O

(2)

The shaded area OABC represents the intersection of the four constraints (1), (2), (3), (4).
We call OABC the FEASIBLE REGION. This means that any point (with integer coordinates) inside OABc is a feasible solution to the problem not necessarily the
optimum solution. For example, (10,20) is a possible solution, i.e. we could produce 10
or P1 and 20 of P2. Clearly however we can do better by making more of each.
The objective function is represented by 30x + 15y = M where M is the profit.
30x +15y = M is a line with slope -2 whose position is determined by the value of M.

335

M = 4000

M = 3000

M = 2000

The graph shows profit lines for different values of M. What we need to do is to find a
point in the feasible region lying on a profit line whose profit (M) is greatest.
Superimposing the profit lines on the feasible region produces the fact that point B lies on
a profit line which is farthest out from the origin, i.e. having the maximum value for M,
i.e. maximum profit.
y
C

30x + 15y = 3600

B

336

1
Point B has co-ordinates (60,120) (i.e. the intersection of 4x + 1 y = 420 and 3x + 2y =
2

420) and hence the manufacturer does best by producing 60 of product P1a nd 120 of
product P2 for a total profit of $3600.
Example 2
Two workers, A and B, ear $6 per hour and $8 per hour respectively. A makes 6
products of P1 and 4 products of P2 per hour. B makes 10 products of P1 and 2 products of
P2 per hour. It is necessary to manufacture 148 of P1 and 52 of P2 to complete an order.
How many hours should each work to fill the order at minimum labour cost? Find the
minimum labour cost.
Solution
Let x,y, be hours A and B work respectively.
Then, since there has to be at least 148 of P1 produced,
6x + 10y 148 (1)
Similarly with respect to P2,
4x + 2y 52 (2)
Also x 0 and y 0.
Therefore the feasible region is as shaded below, i.e. x axis DEF y axis.
y
F
(2)
E
(1)
D

337

The cost lines are 6x + 8y = C where C is the cost.

C = 280

C = 180
C = 60

We need to find the point in the feasible region lying on a cost line which has minimum
value for C.
A visual superimposing of the cost lines upon the feasible region shows us that
point E ist he point required. E has co-ordinates (8,10) the intersection of 6x + 10y =
148 and 4x + 2y = 52.
i.e. Worker A works for 8 hours, B for 10 hours with a total cost of
8 x $6 + 10 x $8, i.e. $128
A problem involving more constraints follows.
Example 3
Two warehouses W1 and W2 have 12 cartons and 8 cartons respectively of a
product which has to be shipped to three stores S1, S2, S3 which need 8, 6 and 6 cartons
respectively, Shipping costs vary according to the table below.
To

From

S1

S2

S3

W1

$7

$3

$2

W2

$3

$1

$2

What is the most economical way to ship the cartons?

338

Solution
Let x be the number of cartons shipped from W1 to S1.
Let y be the number of cartons shipped from W1 to S2
Then the table showing the shipping of cartons is as below.

From

S1

S2

S3

W1

12 x y

W2

8x

6y

x+y-6

Clearly each of these values is not negative hence

x 0, y 0, 12 x y 0
8 x 0, 6 y 0, x + y 6 0
i.e. 0 x 8, 0 y 6,

6 x + y 12.

Constraints

The most is $(7x + 3y + 2(12 x y) + 3(8 x) + 1(6 y) + 2(x + y 6))

i.e. cost is $4x + 2y + 42

Objective Function

cost lines
D

339

Drawing the constraints produces the feasible region ABCDE. The cost lines
superimposed indicate that E is the optimum solution, i.e. at (0,6).
i.e. the optimum solution occurs when x = 0 and y = 6.
i.e. the shipping table is
S1

S2

S3

W1

W2

The total cost is $54.

From the examples quoted it appears that a vertex of a feasible region is an
optimum solution and in fact this is true in all cases. The proof is beyond the scope of this
text but note that the optimum solution is not necessarily unique. For example, the
objective function may be parallel to a boundary of the feasible region leading to the
result that all points on the boundary are optimum solutions.
From the examples quoted it might also appear true that the optimum solution is
the intersection point of the two boundary lines of the feasible region having slopes one,
the smallest of the slopes greater than the objective function and the other, the largest of
the slopes less than the objective function. This is not true and the following example
shows the advantage of drawing a diagram rather than relying on the algebraic data.
Example 4
A man buys 100 pills to satisfy his vitamin requirements of 700 units of B1, 600
units of B2 and 260 units of B6.
Pills 1, 2 and 3 contain units of B1, B2 and B6 as in the table below.
B1

B2

B6

Cost

Pill 1

10

$0.05 each

Pill 2

10

$0.06 each

Pill 3

$0.04 each

340

Find the number of each pill he should buy to satisfy his vitamin requirement at least
cost.
Solution
Let x be the number bought of Pill 1.
Let y be the number bought of Pill 2.
Then 100 x y is the number bought of Pill 3.
The cost function is 5x + 6y + 4(100 x y)
i.e. Cost = x 2y + 400
Constraints
x 0, y 0, 100 x y 0.
B1 requirement constraints 10x + 10y + 6(100 x y) 700 (1)
B2 requirement constraints 5x + 2y + 7(100 x y) 600 (2)
B6 requirement constraints 3x + 8y + 2(100 x y) 360 (3)
Simplified, the problem becomes
Minimise x + 2y + 400
Subject to x 0, y 0, x + y 100.
4x + 4y 100 (1)
2x + 5y 100 (2)
x + 6y 60
y
25

(3)

Feasible Region

20
C
10

A
25

50

60

341

Cost Lines

C = 450
C = 430 C = 440
C = 420
x

The cost lines shown above when superimposed upon the feasible region lead to the
result that B is the vertex of the feasible region lying on a cost line which has smallest
cost.
B has co-ordinates (18,7) the intersection of 4x + 4y = 100 and x + 6y = 60.
i.e. the man should buy 18 of Pill 1, 7 of Pill 2 and hence 75 of Pill 3 for a total
cost of 432, i.e. $4.32.
Exercise 14.1
1. A manufacturer is to make an unknown number of two models A and B. The
models require machine work by three machines M1, M2 and M3 as indicated in
the table.
Time in hours
M1

M2

M3

Model A

1.6

Model B

1.6

No machine may work more than 48 hours per week and profit on Model A is $4 and
on Model B is $3. How many should be produced per week of each to maximise
profit?

342

2. A linear programming problem leads to the constraints x 0, y 0, x + 2y 18,

2x + y 12. Find the point (x,y) such that k is maximised in the cases below.
i)

y=k

ii)

y= -

iii)

y=

iv)

y=-x+k

v)

y = - 2x + k

vi)

y = -3 + k

1
x+k
4

1
x+k
2

3. Use the same details as in 14.1, Example 3 except the cost table is: S1

S2

S3

W1

$2

$1

$3

W2

$5

$3

$4

W1 and W2 have 12 cartons each and S1, S2, S3 need 8 cartons each.
4. Two mines A and B produce three grades of ore: - High, Medium and Low.
Together they have to produce 12 unites of High, 8 unites of Medium and 24 units
of Low to complete an order.
The production table per day is:
High

Medium

Low

12

The running costs per day are $2,000 for A and $1,600 for B. Find the optimum
working arrangement of the mines so that the total cost is minimised when
completely the order.
5. A man has a diet whereby he has to have a minimum weekly requirement of 32
units of protein, a minimum of 25 units of carbohydrates and a maximum of 61
units of fats. He can choose between three types of food A, B or C. The prices
and contents of the foods, in units are in the table on the next page.

343

Food
A

Protein

Carbohydrates

Fats

Cost per pound

$3.00

$1.50

$2.00

He is to buy a weekly total of exactly 20 pounds. How many pounds of each food
should he buy each week to minimise his cost? What is that minimum cost?
6. Maximise 4x + y
Subject to x + y 7, x 0, y 0, 2x y 3, 2x + 4y 24, 4x y 8.
7. A manufacturer produces two types of products A and B using three machines
lathe, grinder and drill. The machine requirements in minutes for manufacturing
each product is set in the table below.
A

Lathe

Grinder

Drill

The lathe can work at most 400 mins per day. The grinder can work at most 450 mins
per day. The drill can work at most 480 mins per day.
The profits on products A and B are $34 and $27 respectively.
i)

Find the optimum production plan.

ii)

The manufacturer has enough money to buy one more machine and by
so doing he will increase the capacity of that type of machine by 40
minutes per day. Which machine should he buy?

8. A dietician wishes to mix two types of food so that the vitamin content of the
mixture contains at least 9 units of vitamin A, 7 units of vitamin B, 10 units of
vitamin C and 12 units of vitamin D. Foods 1 and 2 contain vitamins in units per
pound as shown in the table.

344

Vitamin
A

Food 1

Food 2

Food 1 costs $5 per pound and Food 2 costs $7 per pound. find the minimum cost of a
mixture satisfying the vitamin requirements.

Excerise 14.1 Answers

1. 18 of Model A.

12 of Model B

Profit = $108.

2. i) (0,9) ii) (0,9) iii) {(0,9), (2,8)} iv) (2,8)

v) {(2,8), (3,6), (4,4), (5,2), (6,0)}

vi) (6,0)
3.
S1

S2

S3

W1

W2

Minimum cost = $64

.
4. A opens for 1 day, B for 3 days.
5. 5 pounds of A, 13 pounds of B, 2 pounds of C, at a total costs of $38.50.
6. x = 3, y = 4 Maximum value of 16.
7. i) Manufacture 40 of A and 80 of B, for total profit of $3520.
ii) Drill (New optimum production is 50 of A and 75 of B with profit of $3725.)
8. 4 pounds of Food 1 and 3 pounds of Food 2, for total cost of $41.
10.2 Linear Programming With More Than 2 Variables.
Where more than two variables is involved, a graphical approach to solutions of
linear programming problems is not recommended since, with two variables for example,
it requires graphs of planes in R3 with three dimensional feasible regions of irregular
shapes with planes as boundaries. Furthermore, the objective function is a plane: all
rather difficult to visualise and draw.

345

However, the result that a vertex is still the optimum solution still holds true and a
crude but effective method of solution is to find the vertices of the feasible region and
find the optimum vertex by direct substitution into the objective function.
While this may be time-consuming to do by hand it is, of course, an easy task for
the computer which is ideally suited for solving linear programming problems.
Example
A manufacturer makes three types of cabinets A,B and C. These products use
three kinds of wood oak, teak and plywood. He has available 18- feet of oak, 180 feet
of teak and 250 feet of plywood. Cabinet styles A, B and C require wood as in the table
set out below.
Styles
A

Oak

Teak

Plywood

Profits on Cabinets A, B, C are $30, $40, $50 respectively. How many cabinets of each
type should he make to maximise his profit?
Solution
Let z1, z2, z3 be the number of cabinets made of styles A, B, C respectively.
Then z1 0, z2 0 z3 0.
The oak constraint is 2z1 + 3z3 180.
The teak constraint is 5z2 + 2z3 180
The plywood constraint is 3z1 + 4z2 + 2z3 250.
Objective Function Maximise P = 30z1 + 40z2 + 50z3.
We need to find the vertices of the feasible region. This is done by noting that a vertex of
the feasible region is the intersection of three planes satisfying the other three
inequalities. There will be many points of intersection (20 in fact for this problem) but

346

many of these will not lie in the feasible region because they will not satisfy the other
inequalities. For example, the intersection of z1= 0, z2 = 0 and 3z1 + 4z2 + 2z3 = 250 is
(0,0,125). But (0,0,125) is not a vertex of the feasible region because (0,0,125) does not
satisfy 2z1 + 3z3 180.
As noted, the process of finding the vertices of the feasible region is time-consuming.
Students with knowledge of computer languages and techniques could use this
knowledge to facilitate the finding of the vertices.
The vertices of the feasible region with the corresponding profits are listed below.
O

Vertices (0,0,0) (0,0,60) (0,36,0)

1
(83 ,0,0)
3

Profit

2500

3000

1440

(0,12,60) (78,0,8)

1
( 35 ,36,0)
3

(30,20,40)

3480

2500

3700

2740

Point G (30,20,40) is hence the optimum vertex providing the maximum profit of $3,700.
The best production plan is to make 30 of cabinet A, 20 of Cabinet B and 40 of
Cabinet C.
The fact of non-integer co-ordinates for some vertices of the feasible region is a
minor difficulty overcome by rounding off to the nearest valid integer co-ordinate.
More sophisticated techniques will be developed later for investigating problems
with three or more variables.
The following problem will be analysed later very thoroughly but we show now
the vertex substitutions methods.
Example
Minimise 12x1 + 5x2 + 2x3.
Subject to constraints
x1, x2, x3 0.
x1 + x2 + x3 5.
x1 x2 + x3 3.
x1 2x2 x3 4.

347

1 1
2 1
The vertices of the feasible region are (5,0,0) ( 4 ,0, ) and (4 , ,0) . The feasible
2 2
3 3

region is an unbounded one. the corresponding values of the objection function are 60,
2
55, 57 respectively. The minimum value of the objective function is hence obtained at
3
1 1
the vertex ( 4 ,0, ) .
2 2
1
1
i.e. Solution is x1 = 4 , x2 = 0, x3 = yielding a minimum of 55. We will refer back to
2
2

this problem later on.

Example
A manufacturer uses one or more of four production processes involving labour in
man-hours and tons of raw material. The manufacturer wishes to determine the optimum
daily production schedule with the information as below.
Process 1

Process 2

Process 3

Process 4

Man hours

Tons of raw material

Profit per unit

$60

$40

$70

$50

There is available a total of 100 man-hours daily and 500 tons of raw material are
available daily.
Solution
Let z1, z2, z3, z4 be the number of units produced using processes 1,2,3,4
respectively.
Then z1, z2, z3, z4 0
2z1 + z2 + 2z3 + z4 100
2z1 + 3z2 + 5z3 + 6z4 500
Maximise 60z1 + 40z2 + 70z3 + 50z4

348

The vertices of the feasible region and corresponding profits are:

O

Vertices (0,0,0,0) (0,0,0,83) (0,0,50,0) (0,0,14,71) (0,100,0,0) (0,34,0,66) (50,0,0,0) (10,0,0,80)

Profit

4150

3500

4530

4000

4660

3000

4600

Some of the co-ordinates have been rounded off. The optimum vertex is point E
(0,34,0,66) with a profit of $4,660. The optimum production schedule is 34 units using
process 2 and 66 units using process 4.
Certain problems involve finding a linear functional relationship between two variables,
for which data are available, which best approximates the true relationship.
Example

MINIMAX

We know from an experiment that when x = 1, y = 4.5, when x = y, y = 6.5, and

x = 3, y = 9.5. Find the best linear relationship between x and y using the MINIMAX
method. Minimax means minimising the maximum difference between the true value and
the value estimated by the linear relationship.

9
6

349

i.e. the dotted line is the line of best fit where the maximum vertical distance between any
of the three points and the dotted line is minimised.
Solution
Let y = mx + b be the line of best fit A visual inspection of the data leas us to
deduce that m 0 and b 0 (not completely necessary, but desirable).
Let d = max {vertical distance}
= max {|m + b 4.5|, |2m + b 6.5|, |3m + b 9.5|}
We wish to minimise d
subject to the following constraints.
|m + b 4.5| d
|2m + b 6.5| d
|3m + b 9.5| d
m, b, d 0.
But |z| d is equivalent to z d and z d.
i.e. our constraints are:
m + b 4.5 d
4.5 m b d
2m + b 6.5 d
6.5 2m b d
3m + b 9.5 d
9.5 3m b d
m 0, b 0, d 0 .
Rearranging these constraints yields:
m + b d 4.5
m + b + d 4.5
2m + b d 6.5
2m + b + d 6.5
3m + b d 9.5
3m + b + d 9.5
m 0, b 0, d 0

Minimise d.

350

1
1
The vertices (m,b,d) of the feasible region are (0,0,9.5), (0,7,2.5), (2 , 0,2), (3 , 0,1),
2
2

(2,3,

1
1
1
3 1
), (3,1, ), (2 , 1 , ).
2
2
2
4 4

1
3 1
It therefore follows that the vertex having the minimum value for d is (2 , 1 , ).
2
4 4

i.e. the optimum solution occurs when m = 2

difference of

1
3
and b = 1 yielding the maximum
2
4

1
.
4

1
3
i.e. y = 2 x + 1 is the line of best fit
2
4
3
This means that we could extrapolate the conjecture that when x = 4, y = 11 .
4

Example
Matrix Games
Players A and B play a game where A and B simultaneously choose a playing
card Jack, Queen or King and compare their choices. Player A then pays player B in
dollars according to the following chart.
B

Jack

Queen

King

Jack

-2

Queen

-1

King

-3

This chart means, for example that if A chooses the Queen and B chooses the Jack then A
pays B $2. The negative entries indicate that B has to pay the corresponding amount, e.g.
if A chooses the King and B the Jack then B plays A $3.
A wishes to determine his optimum strategy.
A assigns probabilities p1, p2, 1 p1 p2 to his choices Jack, Queen, King
respectively and then tries to determine p1 and p2 so that the amount he pays B is a
minimum (of course, he would like it to be a negative payment if possible).

351

If B always chooses the Jack, then, IN THE LONG RUN, A expects to pay B,
0p1 + 2p2 3(1 p1 p2) dollars, i.e. $( 3p1 + 5p2 3) each time.
If B always chooses the Queen, then in the long run, A expects to pay B,
-2p1 + 0p2 1(1 p1 p2) dollars, i.e. $( -3p1 p2 + 1) each time.
If B always chooses the King, then in the long run, A expects to pay B,
3p1 -1p2 + 0(1 p1 p2) dollars, i.e. $( 3p1 - 5p2 ) each time.
Since A wishes to minimise the amount he expects to pay B each time regardless
of Bs strategy, then the problem can be set up as follows.
Let M be the maximum amount A expects to pay B each time in the long run.
Then A wishes to: Minimise M
subject to
3p1 + 5p2 3 M
-3p1 p2 + 1 M
3p1 - 5p2 M
p1 0, p2 0, 1 p1 p2 0
These constraints can be simplified and rearranged below.
3p1 + 5p2 - M 3
3p1 + p2 + M 1
3p1 - 5p2 - M 0
p1 0
p2 0
p1 + p2 1
The vertices (p1, p2, M) of the feasible region are
1 1
2 1
1 1
1 1
( , ,0) , (0, , ) , ( , ,0) , (0,1,2), ( ,0, ) , (0,0,1) and (1,0,3).
6 2
3 3
2 2
6 2

352

1 1
i.e. the optimum vertex is ( , ,0) since this has the smallest value for M, i.e. zero. This
6 2

means that As optimum strategy for the long run is to choose the Jack,
the Queen,

1
th of the time,
6

1
1
of the time and hence to choose the King, of the time.
2
3

This result seems a little counter-intuitive when one re-examines the payment
chart. It seems at first glance that A should select the King more frequently the reason
he does not is that in fact Bs strategy is also to choose the Jack
Queen,

1
th of the time, the
6

1
1
of the time and hence to choose the King, of the time and hence A does not
2
3

collect $3 from B very often and pays out $1 far more often when (A) selects the King.
14.2 ii) Gaussian Elimination
To find solutions to equations in three or more variables a method is called
GAUSSIAN ELIMINATION is readily applicable and is illustrated by the following
example:
Solve -

x + 2y + 3z = 9

(1)

4x + 3y + 2z = 11

(2)

5x + 5y + z = 9

(3)

We eliminate x from equations (2) and (3).

i.e. multiply (1) by 5 and subtract (3). Replace (3) with this resulting equation.
Multiply (1) by 4 and subtract (2). Replace (2) with this resulting equation.
i.e.
x + 2y + 3z = 9 (1)
5y + 10z = 25 (2)
5x + 4y + z = 9 (3)
Simplify (2) and (3), then multiply (2) by 3 and subtract (3). Replace (3) with the
result.
i.e. x + 2y + 3z = 9 (1)
y + 2z = 5 (2)

353

-z = 3 (3)
We now solve the resulting system of equations by back substitution, i.e. z = 3 from (3).
Substituting in (2) yields y = - 1 and substituting in (1) yields x = 2.
i.e. (2, -1, 3) is the solution.
What has happened is that the original equation
&1 2 3 # & x # & 9 #
$4 3 2! $ y ! = $11!
$
! $ ! $ !
$%5 4 1 !" $% z !" $% 9 !"
&1 2 3 #
$0 1 2 !
$
!
$%0 0 '1!"

&x #
$ y! =
$ !
$% z !"

&1 0 0# & x #
$0 1 0 ! $ y ! =
$
!$ !
$%0 0 1!" $% z !"

&9#
$5!
$ !
$%' 3!"

&2#
$' 1!
$ !
$% 3 !"

i.e. the system of equations is solvable providing the co-efficients matrix is

diagonalisable.
Difficulties may arise when leading co-efficient of the first equation at the first stage is
zero of when the lead co-efficient of the second equation at the second stage is zero and
so on. Interchanging the equation with a later one will overcome this difficulty if the
system of equations has a solution.
The Gaussian Elimination method may be generalised to four or more variables and is
recommended in those cases.

354

Exercise 14.2
1. i) Solve

2x y + z = 4

by Gaussian Elimination.

x 2y z = -1
x + y 2z = 1
ii) repeat for:
2x y + z = 4
x 2y z = -1
x + y 2z = 1
2. Solve:
w + x + 2y + z = 5
2w + 2x y + z = 4
2w+ 3x + y z = 5
3w + 2x + y 2z = 4
3. Maximise 2x + 2y + 6z
subject to 6x + 6y + 10z 25
3x + 8y + 10z 20
x 0, y 0, z 0.
4. Maximise 2x + 3y + 4z
subject to 2x + y + z 13
x + 2y + 2z 17
x 0, y 0, z 0.
5. A manufacturer can produce four types of cabinets A, B, C or D using oak, teak and
plywood. He has available 2000 feet of oak, 2000 feet of teak and 10,000 feet of
plywood. Cabinet require wood as in the table.
A

Available Supply

Oak

2000

Teak

2000

Plywood

12

10000

Profit

$45

$15

$30

$15

Find the optimum production plan.

355

6. See previous example on matrix games.

Find the optimum strategy for player A where the payment chart is as below:
B

Jack

Queen

King

Jack

-5

Queen

-4

King

-3

7. Find the optimum strategy for both player A and B in the following matrix game:
B

Jack

Queen

King

Ace

Jack

Queen

King

Find the amount that A has to pay B each time, in the long run.
8. Using the set of data below, estimate the value of y when x = 4.5. Use the
minimax method to find the line of best fit.
x

6.5

Exercise 14. 2 Answers

1. i) (2,1,1) ii)

2. (1,1,1,1)

5 3
3. ( ,0, ) .
3 2

4. (3,0,7) or (0,0, . Maximum value is 34.

5. 250 of A, none of B, 1500 of C, 500 of D.

1
5
of the time, Queen
of the time and King
4
12

Profit is $63,750.

6. A chooses Jack

1
of the time.
3

7. Strategy for A is Jack,

King never. Strategy for B is Jack,

Maximum value is 14.

2
1
rds of time, Queen rd of the time,
3
3

2
1
rds of time, Queen rd of the time, King never.
3
3

8. y = 1.5x + 1.5. When x = 4.5, y = 8.25

356

14. 3 (i) Duality

Linear programming problems are such that for a minimising problem there is a
corresponding maximising problem (and vice versa) such that the objective functions
have the same optimum value. The problems are called DUALS of each other. Although
the underlying proof and theory is beyond the scope of this book, the result is remarkable.
Note on previous example is
Minimise 12x1 + 5x2 + 2x3
subject to
x1 + x2 + x3 5
x1 x2 + x3 3
x1 2x2 x3 4
x 0, x2 0, x3 0.
This could be written in the matrix form
& x1 #
Minimise [12 5 2] $$ x 2 !!
$% x 3 !"

Subject to
1#
&1 1
$1 ' 1 1 !
$
!
$%1 ' 2 ' 1!"

& x1 #
&5 #
$ x ! $ 3!
$ 2!
$ !
$% x 3 !"
$%4!"

and
& x 1 # &0 #
$ x ! $0 !
$ 2! $ !
$% x 3 !" $%0!"

If we label the matrices as follows

& x1 #
X = $$ x 2 !!
$% x 3 !"

1#
&1 1
$
A = $1 ' 1 1 !!
$%1 ' 2 ' 1!"

We have Minimise CX subject to

&5 #
R = $$3!!
$%4!"

and C = [12 5 2]

357

AX R and X 0
The corresponding dual problem is set up using matrices A, R and C and a new matrix Z
where Z is a corresponding one row matrix with the correct number of elements to make
the following matrix multiplication possible.
i.e. Maximise ZR
subject to ZA C
Here Z = [z1

z2

and z 0.

z3 ]

i.e. the corresponding maximising problem is: Maximise [z1

Subject to [z1
and [z1

z2

z2

&5 #
z 3 ] $$3!!
$%4!"

z2

1#
&1 1
$
z 3 ] $1 ' 1 1 !! [12 5 2]
$%1 ' 2 ' 1!"

z 3 ] [0 0 0]

i.e. Maximise 5z1+ 3z2 + 4z3

subject to z1 0, z2 0, z3 0
and
z1+ z2 + z3 12
z1 z2 2z3 5
z1 + z2 z3 2
Solving the dual maximizing problem using techniques developed earlier in the chapter
gives us a feasible region whose vertices are as listed below.
Vertex

(0,7,5) (7,0,5) (0,0,12)

(0,2,0) (2,0,0) (0,0,0)

Values of objection Fn

41

55

48

10

From this we deduce that the vertex (7,0,5) yields the maximum value 55 for the
objective function. Note that 55 is also the optimum value of the objective function in the
original minimizing problem.

358

A urther remarkable aspect of this duality relationship is that if AX R is compared

with z 0 and X 0 is compared with ZA C, then, if then problems have solutions,
taking inequalities in corresponding pairs, at least one must be an equality for the
relevant solution point. This may be summarised as:
AX R Z 0
X 0 ZA C
Minimise CX Maximise ZR
For our example
Minimising Problem

Maximising Problem

1 1
Solution point (4 ,0, )
2 2

Solution point (7,0,5)

E x1 + x2+ x3 5

z1 0
I

I x1 + x2+ x3 3

z2 0
E

E x1 - 2x2 x3 4

z3 0
I

I x1 0

z1 + z2 + z3 12
E

E x2 0

z1 z2 z3 5

I x3 0

z1 + z2 z3 2

Objective Function Value for both problems

12x1 + 5x2 + 2x3

= 55 =

5z1 + 3z2 + 4z3

I and E refer to the fact of whether the relevant constraint is an inequality of Equality for
the respective solution point. Note that for the solutions, comparing corresponding
constraints yields exactly one is an inequality and the other is an equality. This is true for
the majority of linear programming problems and gives us an alternative approach to
solutions for these types of problems. In problems where the solution is not unique, e.g.

359

where any point on a whole line segment may represent the solution then in the dual
problem the solution may or may not be unique and comparing corresponding constraints
may produce the result that both are equalities.
Nevertheless it is always true that at least one constraint is an equality.
In a previous example we learned that the problem could be summarized.
z1 0, z2 0, z3 0

2z1

+ 3z3 180
5z2 + 2z3 180

3z1 + 4z2 + 2z3 250

Maximise 30z1 + 40z2 + 50z3
With a solution (30,20,40).
In matrix form this is

[z1
[z1

z2
z2

Maximise [z1

z 3 ] [0 0 0]
z3 ]

& 2 0 3#
$0 5 4! [180 180 250]
$
!
$%3 2 2!"

z2

&30 #
z 3 ] $$40!!
$%50 !"

The dual minimising problem is therefore

& 2 0 3# & x 1 #
&30 #
$0 5 4! $ x ! $40!
$
! $ 2!
$ !
$%3 2 2!" $% x 3 !"
$%50 !"
& x1 #
&0 #
$ x ! $0 !
$ 2!
$ !
$% x 3 !"
$%0!"

360

& x1 #
Minimise [180 180 250] $$ x 2 !!
$% x 3 !"

i.e. problems are:

z1 0

2x1

+ 3x3 30

z2 0

+ 5x2 + 5x3 40

z3 0

3x1 + 2x2 + 2x3 50

2z1

+ 3z3 180

x1 0

+ 5z2 + 3z3 180

x2 0

3z1 + 4z2 + 2z3 250

x3 0

Maximise 30z1 + 40z2 + 50z3

Solution (30,20,40)
Maximal value 3700

Minimise 180x1 + 180x2 + 250x3

Solution (

450 240 110

,
,
)
41 41 41

Minimal value 3700

The solution of the minimising problem was obtained by checking the vertices of the
feasible region and is merely quoted here without derivation for simplicitys sake.
Note: i) The two objective functions have a common value of 3700
ii) The constraints are matched
I ---- E
I ---- E
I ---- E
E ---- I
E ---- I
E ---- I
Example
Consider a minimising problem which leads to the constraints

361

2x1 + 4x2 + 4x3 + 18x4 105

5x1 + 3x2 + 3x3 + 14x4 140
x1 0, x2 0, x3 0, x4 0
Minimise 90x1 + 110x2 + 160x3 + 630x4
Clearly a graphical approach is not possible but its dual problem is considerably more
simple, viz:
z1 0 z2 0 (1) (2)
2z1+ 5z2 90

(3)

4z1+ 3z2 110 (4)

4z1+ 3z2 160 (5)
18z1 + 14z2 630 (6)
Maximise 105z1 + 140z2 = P
Graphically this is
z2
18

B(2,0,10)

27.5

z1

This feasible region is easy to arrive at because it becomes readily apparent that in fact
constraints (5) and (6) are irrelevant. The objective function lines are marked as

hence the optimum point is B, (20,10) the intersection of 2z1 + 5z2 = 90 and 4z1 + 3z2 =
110.
The maximum value of the objective function is 105 x 20 + 140 x 10 i.e. 3500.
Cross-referencing the constraints in the two problems and using the at least one is an
equality condition leads to the result that

362

i)

Minimum value of 90x1 + 130x2 + 160x3 + 630x4 is 3500

ii)

x3 = 0 and x4 = 0 (because (5) and (6) are inequalities)

iii)

x1 = 17

1
1
and x2 = 17 (solving 2x1 + 4x2 + 0 + 0 = 105 and
2
2

5x1 + 3x2 + 0 + 0 = 140)

1
1
i.e. solution to the original minimising problem is (17 ,17 ,0,0).
2
2

363

14.3 (ii) Simplex Method

The simplex method for solving linear programming problems, introduced by C.
Dantzig in 1947, is a long and cumbersome one by hand, but in the age of the high speed
computer it is readily programmable for solving problems with any number of variables.
There are many revisions of the basic simplex method for dealing with non-standard
problems but this text will attempt to deal with the standard maximising problem only.
Students interested in further linear programming should refer to the wealth of text
available on specialised linear programming problems.
It should be remembered that any standard minimising problem has its dual standard
maximising problem so that the method shown here will suffice for any standard
problem.
Example 1
Maximise 4x1 + x2
subject to

x1 + x2 7
2x1 x2 3
2x1 + 4x2 24
4x1 x2 8

x1 0, x2 0
The data are set out in a table with constraint co-efficients in the main body of the table
and the constraint terms entered in the right hand column.
The bottom row constraints the negatives of the co-efficients of the objective function.
1

-1

24

-1

-4

-1

(1)

(2)

364

The (1) and (2) are labels referring to x1 and x2. This initial table refers to the situation at
(0,0) wit the corresponding value of the objective function, i.e. zero entered in the bottom
right hand corner.
The simplex method involves moving from the vertex to the vertex of the feasible
region in a productive way, improving the value of the objective function each time, until
no further improvement can be made. A new table is set up each time corresponding to
the new vertex.
Method
Step 1 Locate the column having the least value in the bottom row, i.e. 4. Call this the
pivotal column
Step 2 For each positive entry in the pivotal column find the quotient of the entry in the
right hand column divided by the entry in the pivotal column. The pivot is the entry in
the pivotal column corresponding to the smallest quotient, i.e. 2. (Since 3 2 is the
smallest of 7 1, 3 2, 24 2, 8 4) Call the pivot p.
Step 3 A new table is then produced using the following rules.
i) Each entry in the pivotal column is divided by p
ii) Each entry in the pivotal row is divided by + p
iii) p is replaced by

1
p

iv) All other entries are transformed by the diagonal rule.

i.e. p c
.
.
.
b..a
a is replaced by a -

bxc
where b and c are entries forming the two other corners of
p

rectangle formed by a and p.

365

Step 4 The label of the pivotal column is reassigned to the pivotal row.
The new table becomes

1
2

1
2

3
2
!

1
2

11
2
3
2

-1

21

-2

-3

(1)

(2)
3
3
This table represents the situation at ( , 0) (note (1) is next to ) where the value of the
2
2

objective function is 6. Steps 1 to 4 are then repeated until the bottom row contains no
negative entries.
The readjustment of the table at each stage is, in fact, the result of a Gaussian
Elimination to arrive at the new vertex.
the new pivot is 1 (row 4 column 2) and then rearranged table is
5
2

3
2

5
2

1
2

1
2

5
2

-5

11

-2

-4

12

(1)

(2)

5
This last table represents the situation at ( , 2) (note (1) and (2) labels) with an objective
2

functional value of 12.

366

The -4 in the bottom rows tells us that this can be improved upon and the new pivot is
5
(row 1 column 1)
2

The next (and final) table is

2
5

1
5
_

3
5

1
5

18
5

2
5

4
5

1
5

12

-4

(1)

(2)

This is the final table because each entry in the bottom row is positive. The solution to the
problem occurs when x1 = 3 and x2 = 4 (see (1) and (2) labels) yielding a maximum
objective function of 16. Graphically what has happened is shown below.
E(0,6)
E

D(2,5)

C(3,4)
1
B (2 ,2)
2

(0,0)

1
A (1 ,0)
2

367

The feasible region is OABCDE. The initial table represents the situation at O.
Subsequent tables reflect the situations at A, B, C in order.
Example
Minimise 12x1 + 3x2 + 7x3
subject to
6x1 2x2 + 5x3 3
2x1 + 3x2 4x3 -2
3x1 + 9x2 + x3 8
x1 0, x2 0, x3 0.
Normally we could not deal with this problem by Simplex because it is both a minimising
problem and has a negative constant term in one of the constraints.
However, the dual of this problem is
z1 0, z2 0, z3 0
6z1 + 2z2 + 3z3 12
-2z1 + 3z2 + 9z3 3
5z1 4z2 + z3 7
Maxmimise 3z1 2z2 + 8z3
which we can handle using a standard simplex method.
Table 1
6

12

-2

-4

-3

-8

(1)

(2)

(3)

The initial pivot is 9

368

Table 2

2
3

2
9

1
3

2
9

-4

1
3

43
9

42
9

(1)

(2)

60
47

307
47

2
47

7
47

1
3

11

1
9

1
3

(3)

The new pivot is 5

1
9

8
9

24
9

27
141

117
47

5
47

29
47

(3)

1
47

60
47

(1)

37
47

412
47

2
9

2
3

Table 3
-

47
9
43
47

39
47
33
47

(2)
This is the final table. The label (2) left unmoved in the final row indicates z2 = 0.
The optimum solution for the dual is therefore (
value of

60
29
,0,
) with an objective function
47
47

412
.
47

Comparing this with the original minimising problem and relying on the idea in the
duality section that at least one of the two corresponding constraints must be an Equality,
we deduce that the optimum vertex for the minimising problem occurs at the intersection
of x1 = 0, 6x1 2x2 + 5x3 = 3 and 3x1 + 9x2 + x3 = 8.
i.e. optimum vertex is (0,

37 43
,
)
47 47

369

Incidentally it is not a coincidence that

37
43
and
appear in the bottom row of the final
47
47

simplex table. They are obtained as follows.

Table 1
9

,, refer to x1, x2, x3 in the original problem.

(1), (2), (3) refer to z1, z2, z3 in the dual problem.

(1)

(2)

(3)

Since 9 is the pivot then (3) and interchange giving

Table 2

9
5

2
9

(1)

(2)

(3)

In Table 2, 5

2
is the new pivot since labels (1) and interchange.
9

Table 3
x

117
47

29
47

(3)

60
47

(1)

43
47

33
47

37
47

412
47

(2)

370

Since remains unmoved in the final column, x1 = 0 but x2 =

indicated in the final row to yield the optimum value of

37
43
and x3 =
as
47
47

412
. In other words, Simplex
47

solves the two problems simultaneously.

For further reading, students are recommended G.B. Dantzig, Linear Programming and
its Extensions (Princeton University Press) 1963, and W.A. Spivey, Linear Programming
An Introduction. (Macmillan New York) 1968.
(Authors note)
A recent (1980) mathematical paper by a Russian mathematician L. G. Khachian,
suggest that a much more efficient method for solving linear programming problems will
soon be possible.
At present the Simplex method is said to be exponential time one. This means that the
computer time required to solve a problem increases exponentially as the number of
variables increases. Khachians method has a polynomial time solution. The dime
difference between the two is illustrated by the fact that a polynomial time algorithm
requiring x3 steps takes about

1
second of computer time when x = 60 whereas an
5

exponential time algorithm requiring 3x steps takes billions of centuries of computer time
when x = 60.
The Simplex method involves a polygon in multi-dimensional space whose
boundaries are determined by the constraints in the problem and whose vertices are
possible solutions.
Khachians method involves the construction of a sequence of ellipsoids in multidimensional space that close in automatically on the optimum solution.

371

Exercise 14.3
1. Find the dual problem of
Minimise 2x1 + 3x2 + 4x3
subject to 5x1 + 6x2 + 7x3 8
9x1 + 10x2 + 11x3 12
x1 0, x2 0, x3 0
and solve by any method.
2. Find the dual problem of
Maximise 5z1 + 3z2
subject to z1 + 2z2 3, 4z1 + 5z2 6,
7z1 + 8z2 9, 10z1 + 11z2 12
z1 0, z2 0
3. A manufacturer has 240, 360, and 420 pounds of wood, plastic and steel
respectively. Products of A, B, C require quantities of these materials as outlined
in the table below.
A

Oak

Teak

Steel

Profits on products A, B, C are $3, $4, $5 respectively. Find the optimum production
plan.
The manufacturer is able to obtain free, from the supplier, 60 pounds of wood, plastic
or steel. Which should he choose?
4. Use Simplex Method to solve.
Minimise 5x1 + 12x2 + 10x3
subject to x1 + x2 + 2x3 4
x1 - 2x2 + 3x3 7
x1 + 2x2 x3 5

372

-x1 + 4x2 + 4x3 -1

x1 0, x2 0, x3 0

373

Exercise 14.3 (contd)

5. Use the Simplex method to find the optimum strategy for A in the follow matrix
game. What is the expected payout reach time in the long run?
B

Jack

Queen

King

Jack

-3

Queen

-3

-5

King

-5

6. Determine the straight line y = mx + b which best fits the following data in the
minimax sense.
x

6.2

6.8

7.0

7.8

8.5

9.4

7. Maximise z1 + z2 + z3 + z4
subject to z1 z2 + z3 z4 2
z1 + z2 z3 + z4 4
-z1 + z2 + z3 z4 = 0
z1 0 , z2 0, z3 0, z4 0.
(Hint: z = 0 z 0 and z 0)

374

Exercise 14.3 Answers

1. Maximise 8z1 + 12z2
Subject to 5z1 + 9z2 2
6z1 + 10z2 3
7z1 + 11z2 4
z1 0
z2 0
8
16
Solution for minimising problem is ( ,0,0) with minimum value of
.
5
5

2. Minimise 3x1 + 6x2 + 9x3 + 12x4.

Subject to x1 + 4x2 + 7x3 + 10x4 5
2x1 + 5x2 + 8x3 + 11x4 5
x1 0, x2 0, x3 0, x4 0
6
Solution to maximizing problem is ( , 0) with maximum value of 6.
5

3. 96 of A, none of B, 72 of C.
60 pounds of wood.
4. (

47 5 7
335
, , ) Minimum value is
. note solution is obtained by solving the dual
9 18 9
9

problem by simplex.
5. i) A chooses Jack
ii)

5
8
1
of the time, Queen
of the time and King
of the time.
14
14
14

10
7

6. y = 0.64x + 6.18 (max. d is 0.02)

1
1
7. z1 = 0, z2 = 1 , z3 = 0, z4 = 1 .Max. value is 3.
2
2