Vous êtes sur la page 1sur 104

LINEAR PROGRAMMING 1

LINEAR PROGRAMMING

Vassilis Kostoglou
E-mail: vkostogl@it.teithe.gr
URL: www.it.teithe.gr/~vkostogl/en







LINEAR PROGRAMMING 2

The problem of limited resources

Every business activity requires resources (means)

There are many types of resources
Capital
Human resources
Equipment
Available time
Raw materials
Existing constraints in every type of resources

The problem: the optimal allocation of limited resources



LINEAR PROGRAMMING 3

What is Mathematical Programming?

Mathematical programming is used to find the best or optimal solution to a problem that
requires a decision or set of decisions about how best to use a set of limited resources
to achieve a state goal of objectives.

Steps involved in Mathematical Programming

- Conversion of stated problem into a mathematical model that abstracts all the
essential elements of the problem.
- Exploration of different solutions of the problem.
- Finding out the most suitable or optimum solution.

Linear programming requires that all the mathematical functions in the model be
linear functions.


LINEAR PROGRAMMING 4

What is Linear Programming (LP)?

A mathematical technique to help plan and make decisions relative to the trade-offs
necessary to allocate resources.

Will find the minimum or maximum value of the objective.

Guarantees the optimal solution to the model formulated.

Linear programming is a widely used mathematical modeling technique to determine
the optimum allocation of scarce resources among competing demands. Resources
typically include raw materials, manpower, machinery, time, money and space.







LINEAR PROGRAMMING 5

The technique is very powerful and found especially useful because of its application to
many different types of real business problems in areas like finance, production, sales
and distribution, personnel, marketing and many more areas of management.

As its name implies, the linear programming model consists of linear objectives and
linear constraints, which means that the variables in a model have a proportionate
relationship.













LINEAR PROGRAMMING 6

Historical perspective

- 1928 John von Neumann published related central theorem of game theory.

- 1944 Von Neumann and Morgenstern published Theory of Games and Economic
Behavior.

- 1936 W.W. Leontief published "Quantitative Input and Output Relations in the
Economic Systems of the US" which was a linear model without objective function.

- 1939 Kantoravich (Russia) actually formulated and solved a LP problem.

- 1941 Hitchcock poses transportation problem (special case of LP).

- WW II Allied forces formulate and solve several LP problems related to military.



LINEAR PROGRAMMING 7

A breakthrough occurred in 1947...

- US Air Force wanted to investigate the feasibility of applying mathematical
techniques to military budgeting and planning.

- George Dantzig had proposed that interrelations between activities of a large
organization can be viewed as a LP model and that the optimal program (solution)
can be obtained by minimizing a (single) linear objective function.

- Air Force initiated project SCOOP (Scientific Computing of Optimum Programs)
SCOOP began in June 1947 and at the end of the same summer, Dantzig and
associates had developed:
1) An initial mathematical model of the general linear programming problem.
2) A general method of solution called the Simplex method.





LINEAR PROGRAMMING 8

Essentials of a Linear Programming model

For a given problem situation, there are certain essential conditions that need to be
solved by using Linear Programming.

1. Limited resources: limited number of labour, material equipment and finance.

2. Objective: refers to the aim to optimize (maximize the profits or minimize the costs).

3. Linearity: increase in labour input will have a proportionate increase in output.

4. Homogeneity: products, workers, efficiency, and machines are assumed to be
identical.

5. Divisibility: it is assumed that resources and products can be divided into fractions
(in case the fractions are not possible, like production of one-third of a computer - a
modification of linear programming called integer programming can be used).



LINEAR PROGRAMMING 9

Some characteristic LP applications

1. Scheduling school buses to minimize total distance traveled.

2. Allocating police patrol units to high crime areas in order to minimize
response time to 911 calls.

3. Scheduling tellers at banks so that needs are met during each hour of the day while
minimizing the total cost of labour.

4. Selecting the product mix in a factory to make best use of machine- and labor-
hours available while maximizing the firms profit.

5. Picking blends of raw materials in feed mills to produce finished feed
combinations at minimum costs.



LINEAR PROGRAMMING 10

6. Determining the distribution system that will minimize total shipping cost.

7. Developing a production schedule that will satisfy future demands for a firms
product and at the same time minimize total production and inventory costs.

8. Allocating space for a tenant mix in a new shopping mall so as to maximize
revenues to the leasing company.


LINEAR PROGRAMMING 11

Requirements of an LP problem

1. LP problems seek to maximize or minimize some quantity (usually profit or cost)
expressed as an objective function.

2. The presence of restrictions, or constraints, limits the degree to which we can
pursue our objective.

3. There must be alternative courses of action to choose from.

4. The objective and constraints in linear programming problems must be expressed in
terms of linear equations or inequalities.







LINEAR PROGRAMMING 12

A Linear Programming model seeks to maximize or minimize a linear function, subject
to a set of linear constraints.

The linear model consists of the following components:
- A set of decision variables
- An objective function
- A set of constraints



LINEAR PROGRAMMING 13

The importance of Linear Programming

Many real world problems lend themselves to linear programming modeling.
Many real world problems can be approximated by linear models.

There are well-known successful applications in:
Manufacturing
Marketing
Finance (investment)
Advertising
Agriculture

There are efficient solution techniques that solve linear programming models.
The output generated from linear programming packages provides useful what if
analysis (sensitivity analysis).



LINEAR PROGRAMMING 14

Assumptions of the linear programming model

- The parameter values are known with certainty.

- The objective function and constraints exhibit constant returns to scale.

- he additivity assumption: There are no interactions between the decision variables.

- The continuity assumption: Variables can take any value within a given feasible
range.


LINEAR PROGRAMMING 15

Integer and mixed-integer problems

A linear programming problem in which all the decision variables must have integer
values is called an integer programming problem.

A problem in which only some of the decision variables must have integer values is
called a mixed-integer programming problem.

Sometimes, some (or all) of the decision variables must have the value of either 0 or 1.
Such problems are then called zero-one mixed-integer programming problems.

Simplex method cannot be used to such problems. Advanced methods are available for
this purpose.



LINEAR PROGRAMMING 16

Generally about LP

- Problems of resources optimal allocation.

- Expressed with a linear first degree function.

- Optimization of a function which is subjected to constraints.

- For non linear functions non linear programming is used.

- Other cases: Mathematical programming, Integer programming, Programming of
separable variables.






LINEAR PROGRAMMING 17

Model example

- A chemical factory produces three independent products A, B and C.

- Each product is processed sequentially by the machines , Y and as following:
- Product : processing by the machines and Y
- Product : processing by the machines , Y and
- Product C: processing by the machines Y and

Graphical representation of the production process



LINEAR PROGRAMMING 18

- Each product liter requires one unit of machine capacity

- Daily machine capacity
- : 100 units
- : 200 units
- : 400 units

- Net profits for each liter of , and C in proportion 3:4:2

- Unlimited demand for products and C

- Maximum daily demand of : 80 liters







LINEAR PROGRAMMING 19

According to current managements decision, the following amounts are produced
during one day:
80 liters of product
20 liters of product
120 liters of product C

Is this the most appropriate policy regarding profit maximization?







LINEAR PROGRAMMING 20

Configuration of the mathematical model

max F = 3*x
1
+4*x
2
+2*x
3


with structure constraints:
x
1
+x
2
s 100 (for )
x
2
+x
3
s 200 (for Y)
x
1
+x
2
+x
3
s 400 (for )
x
2
s 80

and non-negativity constraints
x
1
, x
2
, x
3
> 0


LINEAR PROGRAMMING 21

The standard LP model


with structure constraints:





and non-negativity constraints


n n
x c x c x c f + + = ... ) ( max
2 1 1
X
m n mn m m
n n
n n
b x a x a x a
b x a x a x a
b x a x a x a
s + + +
s + + +
s + + +
...
.... .......... .......... .......... ..........
...
...
2 2 1 1
2 2 2 22 1 21
1 1 2 12 1 11
0 , ... , 0 , 0
2 1
> > >
n
x x x


LINEAR PROGRAMMING 22

Model variations

- Minimization of the objective function

- Some constraints may have > or >= or < or =

- Some of the decision variables are not necessarily non-negative

- In every case, a mathematical representation equivalent to the standard LP model
can be identified


LINEAR PROGRAMMING 23

Alternative LP model formulation


with constraints

where




] ... [ 2 1 n c c c = C

X C X * ) ( max = f
0
P X A s * 0 > X
(
(
(
(
(
(
(

=
nn n n
n
n
a a a
a a a
a a a
...
.
.
.
.
.
.
.
.
.
.
.
.
...
...
2 1
2 22 21
1 12 11
A
(
(
(
(
(
(
(

=
n
x
x
x
.
.
.
2
1
X
(
(
(
(
(
(
(

=
m
b
b
b
.
.
.
2
1
0
P


LINEAR PROGRAMMING 24

LP terminology

- Restrictive line: A straight line corresponding to a constraint of the model.

- Vertex or extreme point: A point at which two restrictive lines are intersecting.

- Solution: Each combination of the decision variables values.

- Feasible solution: A solution satisfying all the constraints.

- Non feasible solution: A solution not satisfying at least one of the constraints.

- Extreme point feasible solution: Is one of the vertexes of the feasible region.



LINEAR PROGRAMMING 25

- Neighboring feasible solutions: They are connected with an edge (boundary) of
the feasible region.

- Feasible region: The (curved) region of feasible solutions formed by the
restrictive lines.

- Basic solution (extreme point solution): A solution corresponding to a vertex
(has non-zero variables equal to the number of the constraints)

- Non basic solution: A solution that is not on a vertex of the feasible region and
can be feasible or non feasible

- Basic feasible solution: A basic solution that corresponds to a vertex of the
feasible region and effectively all its variables are non-negative



LINEAR PROGRAMMING 26

- Slack value: Any excess of an available resource (constraint symbol < or <).

- Surplus value: Any "surpassing" of a requirement (constraint symbol > or >).

- Auxiliary variables: Variables that correspond to slack values and surplus values.

- Basic (non basic) variable: A non-zero (zero) variable in a solution that contains
decision and auxiliary variables

- Mandatory or active constraint: When the resource is completely consumed or
there is no surplus (zero slack or surplus respectively)

- Non mandatory constraint : When a resource is not exhausted (<) or a
requirement is being surpassed (>) (non-zero slack or surplus respectively)



LINEAR PROGRAMMING 27

- Optimal solution: The feasible solution of an extreme point that gives to the
objective function the optimal value (maximum or minimum).
The optimal solution can be only one, but there are cases with unlimited optimal
solutions, no optimal solution, or the value of the objective function tends to infinity.
In every case the amount of feasible solutions of the extreme point is finite. When a
feasible solution of an extreme point is better than all its neighboring then this
solution is the optimal.

- Optimal value: The value of the objective function that corresponds to the optimal
solution.



LINEAR PROGRAMMING 28

Example 1

A manufacturing industrial unit produces three conventional television models
(standard, deluxe and super) and a high-tech model (high-tech). The management is
certain that all production can be purchased in the market. Each device passes in the
production process from all three relevant departments: the workshop, the assembly,
and the final control.

The number of working hours required for each device type in each section is listed in
the following table.






LINEAR PROGRAMMING 29

Departments /TV types Standard Deluxe Super High-tech
Workshop 12 15 15 25
Assembly 10 12 13 20
Control 1/2 3/5 2 2

The total capacity of the unit and the capacities of the three individual departments do
not allow more than 2500, 3000 and 240 man-hours per day in the workshop, assembly
and control respectively. Moreover, due to existing signed contracts at least 50
standard and 50 deluxe devices must be produced daily. The net profit (selling price -
total unit cost) from the sale of one unit of each device is:
Standard Deluxe Super Hightech
Profit () 25 30 40 100



LINEAR PROGRAMMING 30

Required:

1. Formulate a linear programming model to determine the optimal production mix.

2. Investigate the possibility of reducing the dimensions of the Simplex tableau by
removing any redundant constraints.













LINEAR PROGRAMMING 31

Problem analysis

- Decision variables:
The production quantities of the four TV types: x
1
, x
2
, x
3
, x
4

- Objective function:
Maximization of the total profit - F = 25*x
1
+ 30*x
2
+ 40*x
3
+ 100*x
4

- Constraints for each department

- Constraints for the production quantities






LINEAR PROGRAMMING 32

Model formulation



with constraints






and




50
50
240 2 2 ) 5 / 3 ( ) 2 / 1 (
3000 20 13 12 10
2500 25 15 15 12
2
1
4 3 2 1
4 3 2 1
4 3 2 1
>
>
s + + +
s + + +
s + + +
x
x
x x x x
x x x x
x x x x
0 , , ,
4 3 2 1
> x x x x
4 3 2 1
100 40 30 25 max x x x x F + + + =


LINEAR PROGRAMMING 33

Observing the first two structure constraints it can be seen that:


i.e. the left side of the second constraint is obviously less than the right side.
So, the second constraint is surplus and can be removed.

The problem can be simplified more if the following transformation of variables is
carried out:



3000 2500 25 15 15 12 20 13 12 10
4 3 2 1 4 3 2 1
s s + + + s + + + x x x x x x x x
0 0 50 50
0 0 50 50
2 2 2
1 1 1
>
'
> >
>
'
> >
x x x
x x x


LINEAR PROGRAMMING 34

Substituting in the constraints

and

indicates that



Dividing the first constraint by ~ 6.22 (= 1150/185) indicates that


This relation 'covers' the second constraint of the problem
50
1 1
+
'
= x x
185 2 2 ) 5 / 3 ( ) 2 / 1 (
1150 25 15 15 12
4 3 2 1
4 3 2 1
s + +
'
+
'
s + +
'
+
'
x x x x
x x x x
185 4 4 . 2 4 . 2 93 . 1
4 3 2 1
s + +
'
+
'
x x x x
50
2 2
+
'
= x x


LINEAR PROGRAMMING 35

Final LP model of the problem



with just one structure constraint



and


where

4 3 2 1
100 40 30 25 max x x x x F + +
'
+
'
=
'
1150 25 15 15 12
4 3 2 1
s + +
'
+
'
x x x x
0 , , ,
4 3 2 1
> ' ' x x x x
50
1 1
= ' x x 50
2 2
= ' x x
2750 + =
'
F F


LINEAR PROGRAMMING 36

Example 2

A housewife faces the problem of purchasing quantities of meat, potatoes and
vegetables, which constitute her main personal diet. She has already been informed by
a women's magazine dietitian that her full weekly diet nutrition should contain as a
minimum requirement 8 units of carbohydrates, 15 units of protein and 6 units of
vitamins.

The unit numbers of these three constituents contained in each unit of weight (kg) of
the three foods below and the food unit costs are shown in the following table.








LINEAR PROGRAMMING 37

Foods
Nutrients
Meat Potatoes Vegetables
Carbohydrates 3 1 1
Proteins 4 3 4
Vitamins 1 3 1
Cost (/kg) 10 1 2

The housewife would like to know the quantities of each food she should purchase, so
that to meet the minimum requirements of these basic ingredients and minimize the
total purchasing cost.





LINEAR PROGRAMMING 38

After some time a new nutrient requirement consisting of 7 units of calcium was added
to housewifes dietary. Each weight unit of meat, potatoes and vegetables contain 2, 1
and 4 units of calcium, respectively. How this additional requirement of the new
ingredient affects the quantities that must be purchased with a minimum cost?












LINEAR PROGRAMMING 39

Problem analysis

- Decision variables:
The quantities (kg) of meat, potatoes and vegetables - x
1
, x
2
, x
3

- Objective function:
Minimization of total cost:
F = 10*x
1
+ 1*x
2
+ 2*x
3

- Constraints of nutrients (minimum required quantities)


LINEAR PROGRAMMING 40

Model formulation



with constraints






and


3 2 1
2 10 min x x x F + + =
6 1 3 1
15 4 3 4
8 1 1 3
3 2 1
3 2 1
3 2 1
> + +
> + +
> + +
x x x
x x x
x x x
0 , ,
3 2 1
> x x x


LINEAR PROGRAMMING 41

For the new additional nutrient (calcium) it is require the addition of a new constraint



If the problem with the new constraint is solved the influence of the new nutrient to the
cost and composition of the diet can be found.

- What is sensitivity analysis
7 4 1 2
3 2 1
> + + x x x


LINEAR PROGRAMMING 42

Example 3

A hydraulic components production small company manufactures two types of safety
valves. For this purpose the company buys matrices (molds) from an external supplier,
and then processes, holes and smoothes them. The respective production rates -
expressed in number of valves per hour - of the above three phases are presented in
the following table.

Valves production rates per phase
Valve type
Phase
Type Type
Processing 30 40
Holing 28 35
Smoothing 30 25


LINEAR PROGRAMMING 43

Each matrix for valves of type A costs 2, while that for type B costs 3. The ready
valves of type A and B are sold 5 and 6 respectively. The use of the three basic
production machines (lathe, drill, and grinder) implies a current operating cost, which
equals 20, 14 and 20 per working hour respectively. It is also considered that
there is no problem in the promotion to the market of any combination of the valves
types.

Develop an appropriate mathematical model to determine the mix of production, which
maximizes the net profit.







LINEAR PROGRAMMING 44

Problem analysis

- Decision variables:
The number of valves of type and produced per hour - x
1
, x
2


- Constraints:
There are not obvious constraints originating neither from the production process
nor from the capacities of the 3 phases.
The only constraint is time, essentially the hourly rates of valves production.
Lathe (phase 1):
In one hour produces 30 valves or 40 valves
one unit of type is produced in 1/30 of the hour, while one unit of type in 1/40 of
the hour x
1
units of type in x
1
/30 hours and x
2
units of type in x
2
/40 hours
total required time: x
1
/30 + x
2
/40
The time constraint of lathe is: x
1
/30 + x
2
/40 s 1 (hour)



LINEAR PROGRAMMING 45

Drill (phase 2):
In one hour produces 28 valves of type or 35 valves of type
one unit of type is produced in 1/28 of the hour and the 1 in 1/35
x
1
units of A in x
1
/28 and x
2
units of B in x
2
/35
total time: x
1
/28 + x
2
/35
The time constraint of lathe is:
x
1
/28 + x
2
/35 s 1 (hour)

Grinder (phase 3):
In one hour produces 30 valves or 25 valves
one unit of type in 1/30 of the hour, while one unit of type in 1/25
x
1
units of A in x
1
/30 and x
2
units of B in x
2
/25
total time: x
1
/30 + x
2
/25
the constraint coming out is: x
1
/30 + x
2
/25 s 1 (hour)


LINEAR PROGRAMMING 46

Cost of each valve of type :
If in 1 hour the lathe produces 30 units and the hourly operating cost is 20 ,
this means that the cost of each unit is 20/30.
In the drill (with the same reasoning) the cost is 14/28.
In the grinder the production cost of each unit is 20/30.

Total cost of producing a unit of type A: 20/30 +14/28 +20/30 = 11/6

Cost of each valve of type B:
With the same analysis and summing the individual costs on the 3 machines,
the total production cost of each unit is: 20/30 +14 / 28 +20 / 30 = 11/6






LINEAR PROGRAMMING 47

Net profit from the sale of each unit:

Profit from the sale of each unit of type A:
5 - (2 + 11/6 ) = 7/6

Profit from the sale of each unit of type B:
6 - (3 + 17/10 ) = 13/10

Total hourly production profit:
(7/6)*x
1
+ (13/10)*x
2



LINEAR PROGRAMMING 48

Final LP model



with constraints




and

2 1
10
13
6
7
max x x F + =
1
25 30
1
35 28
1
40 30
2 1
2 1
2 1
s +
s +
s +
x x
x x
x x
0 ,
2 1
> x x


LINEAR PROGRAMMING 49

Example 4

A company manufactures two new models of ultramodern refrigerators with
corresponding names "no frost'' and" freezer ". The production area is divided in three
departments: formatting (F) where the final surfaces are formed, assembly (A), and
painting (P) where the final finishing and painting of the refrigerators are made. Each
refrigerator type needs a different number of working hours in each department. So,
each batch of 12 refrigerators "no frost" needs 60, 80 and 20 working hours in
departments F, A and P respectively, while the respective batch of 12 refrigerators
"freezer" 70, 85 and 10 working hours in each of the three departments respectively.
The staffing of the departments is such, that the working hours available to each of
them are 2400, 3000 and 600 respectively during a working month.

Design a monthly production schedule minimizing the total idle time of the staff.





LINEAR PROGRAMMING 50

Problem analysis

- Decision variables:
Number of batches of the 2 refrigerators types - x
NF
, x
F


- Constraints in the available working hours in each department:
Formatting (F): 60x
NF
+ 70x
F
s 2400
Assembly (A): 80x
NF
+ 85x
F
s 3000
Painting (P): 20x
NF
+ 10x
F
s 600

The difficulty of the problem lies in the creation of the objective function while there
are no economic data, the aim being to minimize the unused time.





LINEAR PROGRAMMING 51

Observing the composition of the three constraints referred to man-hours, it comes
up that the unused time in each department is the difference between the right
(available working hours) and the left (actual working hours) side of the
corresponding constraint.

Thus we can define new variables x
1
, x
2
, and x
3
, symbolizing the unused (idle) times
of departments F, A and P respectively.

Idle times of each department:

F: 60x
NF
+ 70x
F
s 2400 60x
NF
+ 70x
F
+ x
1
= 2400
A: 80x
NF
+ 85x
F
s 3000 80x
NF
+ 85x
F
+ x
2
= 3000
P: 20x
NF
+ 10x
F
s 600 20x
NF
+ 10x
F
+ x
3
= 600




LINEAR PROGRAMMING 52

LP model formulation



with constraints



and

F NF
x x F 165 160 6000 min =
600 10 20
3000 85 80
2400 70 60
s +
s +
s +
F NF
F NF
F NF
x x
x x
x x
0 , >
F NF
x x


LINEAR PROGRAMMING 53

Example 5

Nori & Leets Co., one of the largest steel producers in the world, is located next to
Steeltown and is nearly the sole employer of its inhabitants. Steeltown grew and
progressed rapidly along with the company, which now employs nearly 50,000 of its
inhabitants. For this reason the mentality of its residents for many years was: "whatever
is good for Nori & Leets is good for the city too." However this has now changed,
because uncontrolled air pollution from factory furnaces has altered the city
environment and endangered the health of the population.

A recent uprising of the shareholders led to the election of a new enlarged managing
board. New management is determined to pursue a responsible social policy and has
already discussed with officials of the municipality and representatives of the citizens
about what should be done concerning the problem of air pollution. They have mutually
concluded to strict specifications of air quality for citys atmosphere.


LINEAR PROGRAMMING 54

The three main types of pollution factors in the atmosphere of the region are
particulates, sulfuric oxides and hydrocarbons. The new measures require that the
company must reduce its annual emissions of these pollution factors by the following
amounts (expressed in thousands of tons).

Required annual reductions of contaminants emissions

Pollution factor Required reduction of annual
emission rate (thousand tons)
Particulates 60
Sulfuric oxides 150
Hydrocarbons 125





LINEAR PROGRAMMING 55

The managing board instructed companys engineers to determine how the required
decreases can be achieved in the most economical manner.

The operation of the factory itself implies two main pollution sources: a) the blast
furnaces used to produce cast iron, b) the open hearth furnaces used to convert iron
into steel. In both cases, engineers suggested that the most effective ways of mitigating
pollution are: 1) increasing the height of the furnaces, 2) using filters (including gas
traps) in the furnaces, and 3) improving the quality of fuels (using cleaner high-quality
ingredients) used in the furnaces.

All these methods have some specific technological limits on the amount of the
emission of the pollution factors that can wipe out as shown in the table below.
Nevertheless, these methods can be used in any fraction (percentage) of their potential
of pollution mitigation. Moreover, as the function of the three methods is independent,
the emission reduction achieved by each of them is not significantly affected by
simultaneous use of anyone of the others.


LINEAR PROGRAMMING 56

Reductions of emission rates by the full use of each method
of pollution mitigation (in thousand tons per year)

Higher furnaces Filter use Improved fuels
Pollution factors Blast
furnaces
Open hearth
furnaces
Blast
furnaces
Open hearth
furnaces
Blast
furnaces
Open hearth
furnaces
Particulates 12 9 25 20 17 13
Sulfate oxides 35 42 18 31 56 49
Hydrocarbons 37 53 28 24 29 20








LINEAR PROGRAMMING 57

After the collection of the above data it became obvious that none of the three methods
was sufficient itself to achieve all necessary reductions in environmental pollution.
However, the combination of all methods in full capacity on the one hand would be
more than enough, and on the other hand the total cost would be prohibitive for the
company if, as natural, the prices of its products should continue to be competitive.

For these reasons the company's engineers concluded that combination of the
methods should be used, probably with fractions (percentages) of their full capacity of
pollution mitigation based on the corresponding costs. Moreover, because of the
differences between the blast furnaces and the open hearth furnaces, probably should
not be used the same combination of methods in these two types of furnaces.







LINEAR PROGRAMMING 58

It was also specified that the cost of a methods reduced use is directly proportional to
the percentage of its capacity. Thus, for any given fraction used, the corresponding
annual cost is equal to the appropriate percentage of the full cost. The company then
conducted detailed cost accounting to estimate the total annual cost for the
implementation of each of the three pollution mitigating methods. Apart of the increased
operating and maintenance costs, it was also given emphasis in both the fixed cost of
each method (which was converted to its equivalent on annual basis) and at any
intermediate loss in the efficiency of the production process. This analysis resulted in
the following estimates of the annual cost for using each method to full capacity.









LINEAR PROGRAMMING 59

Annual cost of using each pollution mitigation method (in millions of )

Mitigation method Blast furnaces Open hearth furnaces
Higher furnaces 8 10
Filter use 7 6
Improved fuels 11 9

The final stage of the preliminary study was the development of a general frame for the
overall plan of the company on pollution mitigation. The project consists of determining
the methods which will be used for the reduction of the pollution and the possible
fractions of their use.

Due to the combined nature of the problem of determining the plan that meets all the
requirements at minimum cost, an Operational Research group was formed to solve it.
The group adopted an approach using Linear Programming. Which was it?


LINEAR PROGRAMMING 60

Problem analysis

- Decision variables:
x
1
: heighten the blast furnaces
x
2
: heighten the open hearth furnaces
x
3
: filter use in the blast furnaces
x
4
: filter use in the open hearth furnaces
x
5
: fuels improvement in the blast furnaces
x
6
: fuels improvement in the open hearth furnaces

where
x
i
(i = 1, 2, 6) the fraction used of the corresponding method

- Total cost:
8x
1
+10x
2
+7x
3
+6x
4
+11x
5
+9x
6



LINEAR PROGRAMMING 61

- Constraints from the reductions of the pollution factors:

For the particulates:
12x
1
+9x
2
+25x
3
+20x
4
+17x
5
+13x
6
> 60

For the sulfate oxides:
35x
1
+42x
2
+18x
3
+31x
4
+56x
5
+49x
6
> 150

For the hydrocarbons:
37x
1
+53x
2
+28x
3
+24x
4
+29x
5
+20x
6
> 125


LINEAR PROGRAMMING 62

The LP model


with nine structure constraints




and

6 5 4 3 2 1
9 11 6 7 10 8 min x x x x x x F + + + + + =
1 , , , , ,
125 20 29 24 28 53 37
150 49 56 31 18 42 35
60 13 17 20 25 9 12
6 5 4 3 2 1
6 5 4 3 2 1
6 5 4 3 2 1
6 5 4 3 2 1
s
> + + + + +
> + + + + +
> + + + + +
x x x x x x
x x x x x x
x x x x x x
x x x x x x
0 , , , , ,
6 5 4 3 2 1
> x x x x x x


LINEAR PROGRAMMING 63

Example 6

A paper industry produces among other products blotting paper, which is wrapped in
two meters long pieces in standardized cylinders of one meter height. The blotting
paper is cut into a number of smaller and probably different lengths, with fixed height of
one meter, in order to meet all customers orders.
The paper industry received last week the following orders. The lengths are expressed
in palms (tenths of a meter, i.e. 10 cm), which is the usual measure of blotting paper.

Weekly orders
Length (palms) Number of items
9 30
7 150
5.5 65


LINEAR PROGRAMMING 64

The industry would like to cut the fixed length of 20 palms blotting papers so that to
satisfy all the above orders, and to minimize the total wastage (loss).

Design an appropriate mathematical model for solving the problem of the paper
industry with the use of Linear Programming.













LINEAR PROGRAMMING 65

Problem analysis

It is not possible to cut more than three pieces of blotting paper from one cylinder
(e.g. 3 * 5.5 = 16.5 and 3.5 palms as wastage - this way can be symbolized as:
5.5, 5.5, 5.5 wastage 3.5).

Each cylinder can be cut to one, two or three pieces by the following 13 possible ways
(lengths of pieces - wastage):
(9, 0, 0 - 11) - (9, 9, 0 - 2) - (9, 7, 0 - 4) - (9, 5.5, 0 - 5.5) - (9, 5.5, 5.5 - 0) -
(7, 0, 0 - 13) - (7, 7, 0 - 6) - (7, 7, 5.5 - 0.5) - (7, 5.5, 0 - 7.5) - (7, 5.5, 5.5 - 2) -
(5.5, 0, 0 14.5) - (5.5, 5.5, 0 - 9) and (5.5, 5.5, 5.5 - 3.5).

Every possible way of cutting the blotting paper can be represented by a variable x
i
.

Consequently 13 variables are defined:
x
1,
x
2
, x
13
(i = 1, 2,.. 13): number of cylinders that are cut by the i
th
way


LINEAR PROGRAMMING 66

Total loss:

Minimization of the function
F = 11*x
1
+ 2*x
2
+ 4*x
3
+ 5.5*x
4
+ 0*x
5
+ 13*x
6
+ 6*x
7
+ 0.5*x
8
+ 7.5*x
9
+
2*x
10
+ 14.5*x
11
+ 9*x
12
+ 3.5*x
13


Constraints:

For the 30 pieces of 9 palms: 1x
1
+ 2x
2
+ 1x
3
+ 1x
4
+ 1x
5
= 30
For the 150 pieces of 7 palms: 1x
3
+ 1x
6
+ 2x
7
+ 2x
8
+ 1x
9
+ 1x
10
= 150
For the 65 pieces of 5.5 palms: 1x
4
+ 2x
5
+ 1x
8
+ 2x
9
+ 2x
10
+ 1x
11
+ 2x
12
+ 3x
13
= 65

Also, for all variables non-negativity constraints apply.





LINEAR PROGRAMMING 67

Organization of the mathematical model

1) Conversion of the objective function and constraints in the model format.
2) Change the inequality direction of the required constraints, so that the right hand
sides of all constraints are non-negative numbers. Consequently, if the right hand
side of a constraint is negative, a change of the sign and the inequality direction is
needed (multiplying both sides by -1).
3) Conversion of the structure constraints which are inequalities to equivalent
equalities. This is achieved by the accession (addition if the direction of the
inequality is < or or deduction if the direction is of the format > or ) of new non-
negative variables, which are called dummy variables (slack or surplus).

4) Construction in the left hand side of the structure constraints of a complete unary
table by adding the appropriate number of new variables, called artificial variables.


LINEAR PROGRAMMING 68

Methodology of graphical solution

1. Each constraint is initially considered as equality and is plotted by a line (or a level
at three-dimensional problems).

2. The part of the level (or of the space in three-dimensional problems), whichs all
points verify a specific constraint is identified for every constraint.
[The fastest way is to check whether the intersection of the axes (coordinate 0,0)
verifies the constraint (what is valid for this coordinate will also be valid for all points
belonging to the same semi-level)]

3. The intersection of all semi-levels is defined, which verifies all the constraints.
Consequently all vertices of the intersection are being defined.

4. The objective function f(x) is plotted, consisting of a whole family of lines with a
standard direction coefficient. The different values of the objective function are
represented by corresponding parallel lines.


LINEAR PROGRAMMING 69

5. The function line moves in parallel with herself to the direction where her value
increases (or decreases if the aim is its minimization), until it meets with the part of
the level which is the intersection of all constraints. From the family of lines meeting
the intersection of the constraints, only the lines crossing the intersection vertices
are considered.

6. The values of the lines crossing the intersection vertices are calculated. The vertex,
from which passes the line with the highest (smallest in minimization problems)
value is the optimum solution of the problem. The values of the variables consisting
the optimum solution are the coordinates of this vertex.



LINEAR PROGRAMMING 70

Graphical solution of maximization problems

max f(x) = x
1
+ 2x
2


with structure constraints
2x
1
+ x
2
s 20
x
1
+ x
2
s 12
x
1
+ 3x
2
s 15

and non-negativity constraints
x
1
, x
2
> 0


LINEAR PROGRAMMING 71



Graphical solution of maximization problem


LINEAR PROGRAMMING 72

Vertex Coordinates f(x) value
(0,0) 0
(0,5) 10
C (9,2) 13
D (10,0) 10

Thus, the maximum value of f(x) is 13, the corresponding line meets vertex C, and the
values of the variables are x
1
= 9 and x
2
= 2.



LINEAR PROGRAMMING 73

Graphical solution of minimization problems

min z(x) = 24x
1
+ 18x
2


with structure constraints
120x
1
+ 60x
2
> 480
56x
1
+ 112x
2
> 440
103x
1
+ 120x
2
> 720

and non-negativity constraints
x
1
, x
2
> 0



LINEAR PROGRAMMING 74


Graphical solution of minimization problem


LINEAR PROGRAMMING 75

Solving the system of the two equations results in:
x
1
= 1.752 and x
2
= 4.496
Thus, the minimum value of the objective function is:
Min = 24*1.752 + 18*4,496 = 122.98




LINEAR PROGRAMMING 76

The Simplex method

A large variety of Simplex-based algorithms exist to solve LP problems.

Other (polynomial type) algorithms have been developed for solving LP problems:
- Khachian algorithm (1979)
- Kamarkar algorithm (AT&T Bell Labs, mid 80s)

None of these algorithms have been able to beat Simplex in actual practical
applications; hence Simplex (in its various forms) is and will most likely remain the most
dominant LP algorithm for at least the near future



LINEAR PROGRAMMING 77

Fundamental theorem

Extreme point (or Simplex filter) theorem:
If the maximum or minimum value of a linear function defined over a polygonal convex
region exists, then it is to be found at the boundary of the region.

A finite number of extreme points imply a finite number of solutions.
Hence, search is reduced to a finite set of points.
However, a finite set can still be too large for practical purposes.

Simplex method provides an efficient systematic search guaranteed to converge in a
finite number of steps.



LINEAR PROGRAMMING 78

General form of the initial Simplex tableau

Base x
1
x
2
x
n
x
n+1
x
n+2
... x
n+m
R.H.
x
n+1
x
n+2
...
x
n+m

a
11
a
12
... a
1n
1 0 ... 0
a
21
a
22
... a
2n
0 1 ... 0
.............................................................
a
m1
a
m2
... a
mn
0 0 ... 1
b
1
b
2
...
b
m

-f c
1
c
2
... c
n
0 0 ... 0 0



LINEAR PROGRAMMING 79

Previous Simplex tableau

Base x
1
... x
m
x
m+1
... X
s
... x
n+m
R.H.
x
1
1 ... 0 a
1,m+1
... a
1,s
... a
1,n
b
1

... ... ... ... ... ... ...
... 0 0 ... ... ... ...
... ... ... ... ... ... ...
x
r
0 0 a
r,m+1
... a
r,s
a
r,n
b
r

... ... ... ... ... ... ...
... 0 0 ... ... ... ...
... ... ... ... ... ... ...
x
m
0 ... 1 a
r,m+1
... a
m,s
... a
m,n
b
m

... 0 ... 0 c
m+1
... c
s
... c
n
-f
0



LINEAR PROGRAMMING 80

New (current) Simplex tableau

Base x
1
.. x
r
.. .. x
m
.... x
m+1
......... x
s
............ x
n+m
R.H.
x
1


...

x
r

...
...
x
m

1 0 - a
1s
a
rr
0.. ...a
1,m+1
-a
1,s
a
r,m+1
0 a
1,m+1
-a
1,s
a
r,m+1

0 ... ... ... ...
... ... ... ... ... ...
1 a
r,m+1
a
r,n

0 a
rr
= 0 a
r,m+1
= 1 a
rn
=
a
rs
a
rs
a
rs

... ...
... ... ... ... ... ...
... ... ... ... ... ...
0 -a
ms
a
rr
1 a
m,n+1
-a
m,s
a
r,m+1
0 a
m,n
-a
m,s
a
r,n

b
1
-a
1,s
b
r


b
r
= b
r

a
rs




b
m
-a
ms
b
r

-f 0 .... c
s
a
rr
..... 0 c
m+1
-c
s
a
r,m+1
.... 0 .... c
n
-c
s
a
r,n
-f
0
-c
s
a
r,n



LINEAR PROGRAMMING 81

Basic steps of Simplex

1. Begin the search at an extreme point (i.e., a basic feasible solution).

2. Determine if the movement to an adjacent extreme point can improve on the
optimization of the objective function.
If not, the current solution is optimal.
If, however, improvement is possible, then proceed to the next step.

3. Move to the adjacent extreme point which offers (or, perhaps, appears to offer) the
most improvement in the objective function.

4. Continue steps 2 and 3 until the optimal solution is found or it can be shown that
the problem is either unbounded or infeasible.


LINEAR PROGRAMMING 82

Synopsis of Simplex method steps
(after the determination of the initial basic feasible solution)

Step 1: The incoming non-basic variable is identified.

Step 2: The outgoing non-basic variable is identified.

Step 3: By elementary operations between the lines of the current tableau next
Simplex tableau is indicated, in which the changes that practically lead to the
next vertex of the feasible region are reflected.

Step 4: The new (current) basic feasible solution is being tested as to its perfection
If this solution is optimal then the procedure stops, otherwise:
Move to step 1.




LINEAR PROGRAMMING 83

Steps of LP problem solution - Maximization problems

1. Introduction to the base of the variable x
s
with the largest positive coefficient c
s
(lowest negative in minimization problems).

2. Removal from the base of the variable x
r
with the smallest positive quotient b
r
/a
rs
,
where a
rs
> 0.

3. Formulation of a new Simplex tableau replacing into the base variable x
r
with x
s
.

4. Calculation of the elements of the new line of variable x
s
dividing the corresponding
elements of the previous row of variable x
r
with the pivot element a
rs
.






LINEAR PROGRAMMING 84

5. Replacement of all remaining elements according to the equation:

New element = respective previous element
[(respective element of previous column of x
s
)*(respective element of new row of x
r
)]

6. Back to step 1

The above procedure is repeated continuously until one of the followings occurs:
a) All cost coefficients become negative or zero.
In this case the optimal solution has been found.
or
b) All the a
is
coefficients become negative or zero.
In this case the problem is unbounded and there is no finite solution.




LINEAR PROGRAMMING 85

Solution of the model example

max F = 3*x
1
+4*x
2
+2*x
3

with structure constraints
x
1
+x
2
s 100
x
2
+x
3
s 200
x
1
+x
2
+x
3
s 400
x
2
s 80
and non-negativity constraints
x
1
, x
2
, x
3
> 0





LINEAR PROGRAMMING 86

Adding the appropriate number of slack variables (artificial variables are not needed as
all the structure constraints are of the type s) the model example is formulated as:

maxf = 3x
1
+ 4x
2
+ 2x
3


x
1
+ x
2
+ x
4
= 100
x
2
+ x
3
+ x
5
= 200
x
1
+ x
2
+ x
3
+ x
6
= 400
x
2
+ x
7
= 80


LINEAR PROGRAMMING 87


Sequential Simplex tableaux (maximization problem)



Base x
1
x
2
x
3
x
4
x
5
x
6
x
7
R.H.
x
4
1 1 0 1 0 0 0 100
x
5
0 1 1 0 1 0 0 200
x
6
1 1 1 0 0 1 0 400
x
7
0 1* 0 0 0 0 1 80
-f 3 4

2 0 0 0 0 0

x
4
1* 0 0 1 0 0 -1 20
x
5
0 0 1 0 1 0 -1 120
x
6
1 0 1 0 0 1 -1 320
x
2
0 1 0 0 0 0 1 80
-f 3 0 2 0 0 0 -4

-320



LINEAR PROGRAMMING 88


x
1
1 0 0 1 0 0 -1 20
x
5
0 0 1* 0 1 0 -1 120
x
6
0 0 1 -1 0 1 0 300
x
2
0 1 0 0 0 0 1 80
-f 0 0 2

-3 0 0 -1 -380

x
1
1 0 0 1 0 0 -1 20
x
3
0 0 1 0 1 0 -1 120
x
6
0 0 0 -1 -1 1 1 180
x
2
0 1 0 0 0 0 1* 20
-f 0 0 0 -3 -2 0 1 -620

x
1
1 1 0 1 0 0 0 100
x
3
0 1 1 0 1 0 0 200
x
6
0 -1 0 -1 -1 1 0 100
x
7
0 1 0 0 0 0 1 80
-f 0 -1 0 -3 -2 0 0 -700

Optimal solution:


LINEAR PROGRAMMING 89

x
1
=100 x
2
=0 x
3
=200 x
4
=0 x
5
=0 x
6
=100 x
7
=80 max f=700
Steps of LP problem solution Minimization problems

1. Formulation of the artificial function F = x
i
, where x
i
the artificial variables expressed
as functions of the other variables of the corresponding constraint limitations.

2. Construction of the first Simplex tableau with the coefficients of F under those of the
initial function f.

3. Minimization of F with the known procedure of Simplex method (steps 1 to 6 of the
algorithm), inserting each time to the base, the variable with the smallest negative
coefficient c
i
.

4. Repetition of the minimization procedure for function F, until it takes the zero value
(all the cost coefficients will become negative or zero).





LINEAR PROGRAMMING 90


5. Removal of the coefficient row of F and of the columns of all artificial variables.

6. Minimization of the initial function f, inserting each time in the base the variable with
the smallest negative coefficient.

7. Repetition of the minimization procedure for function f, until all cost coefficients
become positive or zero.



LINEAR PROGRAMMING 91

Solution of a minimization problem

The following minimization problem is given:

minf = x
1
+ 2x
2


with structure constraints
2x
1
+ x
2
s 6
x
1
+ x
2
= 4
x
1
+ x
2
> 8
and non-negativity constraints
x
1
> 0, x
2
> 0


LINEAR PROGRAMMING 92

The existing inequalities are transformed to equalities (by adding slack variables x
3
and
x
4
) and the unary table is created (by adding the artificial variables x
5
and x
6
). Thus, the
problem is formulated as following:
minf = x
1
+ 2x
2
with constraints
2x
1
+ x
2
+ x
3
= 6
x
1
+ x
2
+ x
5
= 4
x
1
+ 3x
2
- x
4
+ x
6
= 8
and
x
1
,

x
2
, x
3,
x
4
, x
5
, x
6
> 0




LINEAR PROGRAMMING 93

According to the minimization algorithm, at first the artificial function F has to be
created.

F = x
5
+ x
6
= (4 - x
1
- x
2
) + (8 - x
1
- 3x
2
+ x
4
) F = 12 - 2x
1
- 4x
2
+ x
4

Simplex tableaux are following, up to the determination of the optimal solution,
minimizing successively functions F and f.





Sequential tableaux of Simplex method (minimization problem)


LINEAR PROGRAMMING 94

Base x
1
x
2
x
3
x
4
x
5
x
6
R.H.
x
3
2 1 1 0 0 0 6
x
5
1 1 0 0 1 0 4
x
6
1* 3 0 -1 0 1 8
-f 1 2 0 0 0 0 0
-F -2 -4 0 1 0 0 -12

x
3
5/3 0 1 1/3 0 -1/3 10/3
x
5
2/3* 0 0 1/3 1 -1/3 4/3
x
2
1/3 1 0 -1/3 0 1/3 8/3
-f 1/3 0 0 2/3 0 -2/3 -16/3
-F -2/3 0 0 -1/3 0 4/3 -4/3
x
3
0 0 1 -1/2 -5/2 1/2 0
x
1
1 0 0 1/2 3/2 -1/2 2
x
2
0 1 0 -1/2 -1/2 1/2 2
-f 0 0 0 1/2 -1/2 -1/2 -6
-F 0 0 0 0 1 1 0
Final Simplex tableau


LINEAR PROGRAMMING 95


Base x
1
x
2
x
3
x
4
R.H.
x
3

x
1

x
2

0 0 1 -
1 0 0
0 1 0 -
0
2
2
-f 0 0 0 -6

Optimal solution: x
1
=2, x
2
=2, x
3
=x
4
=x
5
=0 min f = 6


LINEAR PROGRAMMING 96

Models without unique optimal solutions

Infeasibility:
Occurs when a model has no feasible point.

Unboundness:
Occurs when the objective can become infinitely large (max) or infinitely small (min).

Alternate solution:
Occurs when more than one point optimizes the objective function


LINEAR PROGRAMMING 97

LP application

A product mix problem

A manufacturer has fixed amounts of different resources such as raw material, labour,
and equipment.

These resources can be combined to produce any one of several different products.

The quantity of the i
th
resource required to produce one unit of the j
th
product is known.

The decision maker wishes to produce the combination of products that will maximize
total income.

The Giga company produces two large products: I and II.



LINEAR PROGRAMMING 98

The raw material requirements, space needed for storage, production rates, and selling
prices for these products are given in the following table.

Production data for Giga company

Product
I II
Storage space (m
2
/unit) 4 5
Raw material (kg/unit) 5 3
Production rate (units/hr) 60 30
Selling price ($/unit) 13 11


The total amount of raw material available per day for both products is 1575 kg.

The total storage space for all products is 1500 m
2
, and a maximum of 7 hours per day
can be used for production.


LINEAR PROGRAMMING 99

Graphical solution




LINEAR PROGRAMMING 100

When the decision variables are more than two, it is always advisable to use Simplex
method to avoid lengthy graphical procedure.

The Simplex method is not used to examine all the feasible solutions.
It deals only with a small and unique set of feasible solutions, the set of vertex points
(i.e. extreme points) of the convex feasible space that contains the optimal solution.

The Simplex solution yielded the following optimum production programme for Giga:

The company can maximize its sale income to 4335 by producing 270 units of
product I and 75 units of product II.
There will be no surplus of raw materials or production time, but there will be 45 units of
unused storage space.

The managers are interested to know whether it is worthwhile to increase its production
by purchasing additional units of raw materials and by either expanding its production
facilities or working overtime.


LINEAR PROGRAMMING 101

The critical questions are:

Which is the income value (or marginal value) of each additional unit of each type of
resources?

Which is the maximum cost (or marginal cost) that they should be willing to pay for
each additional unit of resources?

Answers to these questions can be obtained from the objective function in the last
tableau of the Simplex solution:

4335 $
7
16
7
15
3 2
= + + S S Z

that is,



LINEAR PROGRAMMING 102

3 2
7
16
7
15
4335 $ S S Z =


Because S
1
, S
2
and S
3
represent surplus resources, the negatives of these variables
(i.e., -S
1
, -S
2
, -S
3
) represent additional units of these resources that can be made
available. The income values (or marginal values of additional units of these resources
can be obtained by taking the partial derivatives of Z with respect to -S
1
, -S
2
and -S
3
.

Therefore, the marginal value of one additional unit of each resource is:

0 $
) (
1
=
c
c
=
S
Z
space Storage

7
15
$
)
2
(
=
c
c
=
S
Z
materials Raw

7
16
$
) (
Pr
3
=
c
c
=
S
Z
time oduction



LINEAR PROGRAMMING 103


Thus, the marginal values of additional units of resources can be obtained directly from
the coefficients of the objective function in the last tableau of a Simplex solution.
The Giga Company should be willing to pay up to $ 15/7 for an additional unit of raw
materials and $ 16/7 for an additional unit of production time.

If the actual cost of an additional unit (i.e. marginal cost) of these resources is smaller
than the marginal value, the company should be able to increase its income by
increasing production.

The marginal values above are valid, however, only as long as there is surplus storage
space available.



LINEAR PROGRAMMING 104

Sensitivity analysis

This analysis helps to test the sensitivity of the optimum solution with respect to
changes of the objective function coefficients, the coefficients in the constraints
inequalities, or the constant terms (right hand sides) of the constraints. For example in
the above case study:
- The actual selling prices (or market values) of the two products may vary from time
to time. Over what ranges can these prices change without affecting the optimality
of the present solution?
- Will the present solution remain the optimum solution if the amount of raw
materials, production time, or storage space is suddenly changed because of
shortages, machine failures, or other events?
- The amount of each resource type needed to produce one unit of product type can
be either increased or decreased. Will such charges affect the optimal solution?

Vous aimerez peut-être aussi