Mathematical Programming

Introduction
There are many activities in an organization which involve the allocation of resources. These
resources include labor, raw materials, machinery and money. The allocation of these resources
is sometimes referred to as programming. Problems arise because the resources are usually in
limited supply. If a company makes several products, using the same machine and labor force,
management must decide how many of each product to produce. The decision will be made so
that management’s objective is satisfied. Management may wish to plan production in such a
way as to maximize the total contribution made each month, or to maximize the utilization of the
machinery each week, or to minimize the cost of labor each week. The decision variable in this
case is the amount of each product to be made in a given time period.
Similarly, if the company has an amount of capital to invest in a number of projects, the money
allocated to each project will be governed by some objective. It may be necessary to minimize
risk, or to maximize capital growth. The decision variable in this case is the amount of money
allocated to each project.
In general, the objective is to determine the most efficient method of allocating these resources
to the variables so that some measure of performance is optimized. Modelling methods can often
be used to aid in this allocation process.
Mathematical programming is the use of mathematical models and techniques to solve
programming problems.
Linear programming is a suitable method for modelling an allocation problem if the objective and
the constraints on the resources can all be expressed as linear relationships of the variables. The
techniques have a number of distinct steps.
The linear programme must first be formulated mathematically. This means that the variables over
which we have control and the objective must be identified. The objective and the constraints on
the resources are then written down as linear relationships in terms of the variables.
Once the linear programme is complete, all of the feasible combinations of the variables are
identified. The combination which optimizes the objective may then be selected. If only two
variables are involved, a graphical solution is possible. If, however, we have a multivariable
problem, we must resort to an algebraic method for which a computer package will be used.
Once the optimum solution has been identified, it must be evaluated. This will include a sensitivity
analysis.
As with any other mathematical aid to decision making, the solution from the linear programme is
just one piece of management information which contributes to the final decision.
Definition of key terms
1. Linear programming - A branch of mathematics that uses linear inequalities to solve decision-
making problems involving maximums and minimums; or, a mathematical procedure
for minimizing or maximizing a linear function of several variables, subject to a finite number of
linear restrictions.
2. Slack variable - In linear programming a slack variable is a variable which is added to a
constraint to turn the inequality into an equation. This is required to turn an inequality into an
equality where a linear combination of variables is less than or equal to a given constant in the
former.
3. Surplus variable - In linear programming a surplus variable is a variable which is subtracted
from a constraint to turn the inequality into an equation.
This is required to turn an inequality into an equality where a linear combination of variables is
greater than or equal to a given constant in the former.
4. Basic variables – These are variables associated with unit columns of the Simplex Tableau.
All other variables are called non-basic variables.
5. Feasible region - The set of all values which satisfy all constraints; otherwise, it is an infeasible
region
6. Sensitivity analysis - A technique of assessing the extent to which changes in assumptions
or input variables will affect the ranking of alternatives.
7. Shadow price - is the maximum price that management is willing to pay for an extra unit of a
given limited resource.
The steps in formulating a linear-programming model are:
Step 1: Identify the decision variables
Step 2: Identify the linear objective function and the constraints
Step 3: Express the objective function in terms of the variables
Step 4: Express the constraints in terms of the variables
To formulate a two variable linear programme
Example 1
A small family firm produces two old-fashioned non-alcoholic drinks. ‘Pink fizz’ and ‘Mint pop’.
The firm can sell all that it produces but production is limited by the supply of a major ingredient
and by the amount of the machine capacity available. The production of 1 liter of ‘Pink fizz’
requires 0.02 hours of machine time, whereas the production of 1 liter of ‘Mint pop’ requires 0.04
hours of machine time. 0.01 kg of a special ingredient is required for 1 liter of ‘Pink fizz’. ‘Mint
pop’ requires 0.04 kg of this ingredient per liter. Each day the firm has 24 machine hours available
and 16kg of the special ingredient. The contribution is £0.10 on 1 liter of ‘Pink fizz’ and £0.30 on
1 liter of ‘Mint pop’. How much of each product should be made each day, if the firm wishes to
maximize the daily contribution?
Solution
Step 1: Identify the variables within the limits of the constraints. The firm can decide how much
of each type of drink to make. Let p be the number of litres of ‘Pink Fizz’ produced per day. Let
m be the number of litres of ‘mint pop’ produced per day.
Step 2: Identify the objective and the constraints. The objective is to maximize the daily
contribution Let £P per day be the contribution. This is maximized within constraints on the
amounts of machine time and the special ingredients available.
Step 3: Express the objectives in terms of the variables:
P = 0.01p + 0.30m (£/day)
This is the objective function – the quantity we wish to optimise.
Step 4: Express the constraints in terms of the variables. The contribution is maximised subject
to the following constraints on production:
Machine time: to produce p litres of ‘Pint fizz’ and m litres of ‘Mint pop’ requires (0.02p + 0.04m)
hours of machine time each day. There is a maximum of 24 machine hours available each day,
therefore the production must be such that the number of machine hours required is less than or
equal to 24 hours per day. Therefore:
0.02p + 0.04 m < 16 24 hours/day
Special ingredient: to produce p litres of ‘pink fizz’ and m litres of ‘Mint pop’ requires (0.01p +
0.04m) kg of the ingredient each day. There is a maximum of 16 kg available each day, therefore
the production must be such that the amount of the special ingredient required is at most 16kg
per day. Therefore:
0.01p + 0.04m < 16 kg/day
There are no further constraints on the production, but it is sensible to assume that the firm
cannot make negative amounts of drink, therefore.
Non negativity:
P >0, m >0
The complete linear programme is as follows.
Maximize:
P = 0.10p + 0.30 m (£/day)
Subject to:
Machine time: 0.02 p + 0.04m < 24 hours/day
Special ingredient: 0.01p + 0.04m < 16 kg/day
p, m >0
Example 2
A manufacturer of high precision machined components produces two different types, X and
Y. In any given week, there are 4,000 man-hours of skilled labour available. Each component
X requires one man-hour for its production and each component Y requires 2 man-hours. The
manufacturing plant has the capacity to produce a maximum of 2,250 components of type X each
week as well as 1,750 components of type Y. Each component X requires 2kg of plate. Each week
there are 10,000 kg each of rod and plate available. The company supplies a car manufacturer with
the unions that at least 1,500 components will be produced each week in total. If the unit
contribution for component X is £30 and for component Y is £40, how many of each type should
be made in order to maximize the total contribution per week?
Solution:
First the linear programme must be formulated.
Step 1: Choose the variables: produce X components of type X per week and Y components of
type Y per week.
Step 2: What is the objective? What are the constraints on the production? The objective is to
maximise the total weekly contribution. The production is constrained by the amount of:
a) Labour – maximum available is 4000 hours per week
b) Machine capacity – there is a separate limit for each product. The machines can make
at most 2,250 components of type X each week and at most 1750 components of type Y each week.
c) Rod – maximum available is 10,000kg per week
d) Plate – maximum available is 10,000kg per week
In addition, minimum amounts of each product are required:
a) Regular orders – the number of component X made must be at least enough to satisfy the regular
orders
b) Union agreement – the total number of components (x +y) must at least satisfy the agreement.
Step 3: The objective function. Let £P be the total contribution per week, where:
P = 30 x + 40 y (V/week)
Step 4: The constraints on production. For each resource, the amount of resource required each
week to produce X components of type X and Y components of type Y is given below, together
with the maximum amount of the resource available.
Labour required: 1 x + 2y ≤ 4000 hours/week
Machine capacity required: x ≤ 2250 components/week
Y ≤1750 components /week
Rod required: 2x + 5y ≤ 10,000 kg/week
Plate required: 5x + 2y ≤ 10,000 kg/week
In addition
Regular orders: x ≥ 600 components/week
Union agreement: x + y ≥1500 components/week
Non-negativity: x, y ≥ 0
The complete linear programme is:
Produce x components of type X and Y components of type Y each week. Maximise:
P = 30 x + 40 y (£ /week)
Subject to the constraints:
Labour: 1x + 2y ≤ 4000 hours/week
Machine capacity: x ≤ 2250 components/week
Y ≤1750 components/week
Rod: 2x + 5y ≤ 10,000 kg/week
Plate: 5x + 2y ≤ 10,000 kg/week
Regular orders: x ≥ 600 components/week
Union agreement: x + y ≥1500 components/week
Non-negativity: x, y ≥ 0
Example 3
Electra Plc produces personal computers and word processors. Currently four models are being
produced.
The Jupiter – 512K memory, single disk drive;
The Venus – 512K memory, double disk drive;
The Mars – 640K memory, double disk drive;
The Saturn – 640K memory, hard disk
Each computer passes through three departments in the factory – sub, assembly, and test. The
details of the times required in each department for each model are given in Table 1 together with
the maximum capacities of the departments. The marketing department has assessed the demand
of each model. The maximum forecast demands are also given in the table, together with the unit
contribution for each model:

Construct the linear programme for this product mix problem, if the objective is to maximise the
total contribution per month.
Solution
Step 1: Choose the variables produce:
Let J units of the Jupiter per month
V units of the Venus per month
M units of the Mars per month and
S units of the Saturn per month
Step 2: What is the objective? What are the constraints on the production? The objective is to
maximise the total contribution each month. The production is constrained by the number of the
labour hours available in the three departments and by the number of each model which can be
sold.
Step 3: The objective function. Let £p be the total contribution per month, where:
P = 15j + 30v + 120m + 130s (£/month)
Step 4: The constraints on production. For each department, the amount of time required to
produce j, v, m and s units of the respective models is linked to the maximum time available.
Sub – assembly: 5j + 8v+ 20m +24s ≤ 800 hours/month
Assembly: 2j + 3v + 8m + 14s ≤ 420 hours/month
Test: 0.1 j + 0.2v + 2m + 4s ≤ 150 hours/month
Demand for Jupiter: j ≤ 100 units/ month
Demand for Venus: v ≤ 45 units/month
Demand for Mars: m ≤ 25 units/month
Demand for Saturn: s ≤ 20 units/month
Non-negativity: j, v, m, s ≥0
The complete linear programme is each month produce j, v, m and s models, respectively of the
Jupiter, Venus, Mars and Saturn. Maximize:
P= 15 j + 30 v + 120m + 130 s (£/month)
Subject to the constraints given above.
Example 5:
A portfolio manager wishes to invest up to £100,000 to maximise the total annual interest income.
She has narrowed her choices to four possible investments, A, B, C and D. Investment A yields
6% per annum, investment B yields 8% per annum, investment C 10% per annum and investment
D 9% per annum. The four investments have varying risks and conditions attached to them. For
liquidity, at least 25% of the funds must be placed in investment D. Volatile government policies
indicate that no more than 20% of the investments should be in C, whereas tax considerations
require at least 30% of the funds to be placed in A. Formulate a linear programming model for this
investment problem.
2 3 8 quantitative techniques
STUD Y TEX T
Solution:
Invest £a in investments, A, £b in investment B, £c in investment C and £d in investment D. The
objective is to maximise the total interest income per year. The investment is constrained by the
considerations of safety, liquidity, government policy and taxation. Let £R be the total interest
income per year, where:
R = 0.06a + 0.08b + 0.10 c+ 0.09d (£/year)
Maximise subject to:
Total investment: a +b + c + d ≤ 100,000 £ available
Safety: a +b ≥ 0.5 (a +b + c +d) £
Liquidity: d ≥ 0.25 (a + b + c + d) £
Government policy: c≤ 0.2 (a + b +c +d) £
Tax: a≥ 0.3 (a + b + c + d) £
Non-negativity: a, b, c, d ≥ 0
To solve a linear programme, it is conventional to arrange the constraints so that the variables
appear only on the left-hand side of the inequalities. This is done below. The complete linear
programme is:
Invest £a in investment A,
£b in investment B
£c in investment C
And £d in investment D.
Maximise the total interest income per year, where:
R = 0.06a + 0.08 b + 0.10 c + 0.09 d (£ /year)
Subject to the constraints:
Total investment: a + b + c + d ≤ 100.000 £
Safety: 0.5 a + 0.5 b – 0.5c + 0.5 d ≥ 0 £
Liquidity: - 0.25 a – 0.25b – 0.5 c – 0.5 d ≥ 0 £
Government policy: - 0.2 a – 0.2 b + 0.8 c – 0.2 d ≤ 0 £
Tax: 0.7 a – 0.3 b – 0.3 c- 0.3 d ≥0 £
Non-negativity: a, b, c, d ≥ 0
The objectives in the four examples above have all required a quantity to be maximized. The
procedure is identical at this stage if the objective is to minimize some measure.
Solving the linear programme
We are now at the stage of considering how to find the values of the variables which will satisfy
all of the constraints simultaneously and optimize the objective. We are more accustomed to
dealing with equations rather than inequalities. It is a simple matter to change the inequalities into
equations. We add an additional variable to the left-hand side of the inequality.
To demonstrate this procedure, let us refer to example 2 which concerned the manufacturer of
machine components X and Y. We will include an additional variable in each constraint to produce
a set of equations. This variable is denoted by s hence s1 is included in the first constraint, s2 in
the second constraint and so on. We will also impose the condition that the values of these variables
cannot be negative i.e., s1 ≥ 0 all ≥. This means that the variable is added to the left-hand side for
all ≤ constraints and subtracted from all ≥ constraints. The linear programme becomes: produce x
components of type x and y components of type Y each week. The objective is to maximise the
total weekly contribution.
Maximise
P = 30x + 40y (£/week)
Subject to the constraints:
Labour: 1x + 2y + s1 = 4000 hours/week
Machine capacity: x + s2 = 2250 components/week
Y + s 3 = 1750 components/week
R od: 2x + 5y + s4 = 10,000 kg/week
Plate: 5x + 2y + s5 = 10,000 kg/week
Regular orders: x - s6 = 600 components/week
Union agreement: x + y –s7 = 1500 components/week
Non-negativity: x, y ≥ 0
These additional variables are called Slack variables. In the ≤ constraints. They represent the
amount of the resource not used, that is, the difference between the resource used and the maximum
available. For example, look at the labour constraint above. Suppose 1,000 components of hours.
Since 4,000 hours are available, the spare capacity, or slack is (4,000- 3,000) = 1,000 hours. For
this combination, the negative slack variables are referred to as Surplus variables since they
represent the amount of resource being used over and above the minimum requirement. For
example, look at the ‘regular orders’ constraint, when 1,000 components of type X are being
produced. The minimum number of type X required by this constraint is 600, hence a production
level of 1,000 gives a surplus of 400 components above the minimum. Therefore, s6 takes the
value 400.
We now have a set of simultaneous equations. However, the number of variables is greater than
the number of equations. A unique set of solutions will arise only if the number of variables and
the number of equations is the same. The best we can do is to identify a set of feasible solutions
to the equations. This set of feasible solutions gives all combinations of the variables which satisfy
all of the constraints. We will then select from this set the particular solution or solutions
which optimize the objective.
How do we set about identifying the set of feasible solutions? This can be done graphically if
the problem involves only two variables. However, we must resort to an algebraic method if the
problem is multivariate.
The algebraic solution of a linear programme
A linear equation represents a set of points which lie on a straight line. A linear inequality
represents an area of a graph. For example, x ≤ 7 says that X takes a value which is less than 7 or
is equal to 7. The situation can be illustrated graphically as follows. Draw the line x = 7, see left
hand graph in figure 12.1. This divides the graph into three sets of points for which x = 7, the line
itself; those for which x < 7, the area to the left of the line; and those for which x > 7, the area to
the right of the line. We do not require this last set. It is usual to shade the area not required.

Suppose x + y ≤ 10, which area does this represent? The procedure is the same as in the previous
example. First of all, we draw the line x + y = 10. See the left-hand graph below. Again, the line
divides the graph into three sets of points: those for which x +y = 10, the line; those for which x +
y < 10, the area below the line; and those for which x + y > 10, the area above the line.
A useful technique for deciding which is the rejected area on the graph is to take any point on the
graph away from the line and substitute its values into the inequality. If the inequality still makes
sense, then that point is feasible solution. If the inequality is untrue, then the point is infeasible
and lies in the rejected region. The origin is a convenient point to use. Substitute x = y = 0 into
the inequality x +y ≤ 10, we have 0 + 0 ≤ 10 which is a true statement, therefore the origin is a
feasible solution and we should reject the other side of the line.

l Programming
Each constraint in the linear programme can be drawn in this way and the rejected area shaded.
If all of the constraints are drawn on the same graph, the area which remains unshaded is the set
of points which satisfies all of the constraints simultaneously. This area is called the feasible
region. For a linear programme, it does not matter which variable is plotted on which axis. The
origin should always be included on the graph. False zeros must not be used. Let us now apply
this procedure to the linear programme for example about the production of the two types of soft
drink. We can illustrate the constraints graphically.
Machine time; 0.02p + 0.04 m≤ 24 hours/day.
Plot the line 0.2p + 0.04 m = 24. An easy way of plotting the line is to find the points where the
line crosses the p and the m axes. Put p = 0 into the equation and calculate p, i.e., when m = 0, p =
1200. Plot these two points and join them to give the line. This method always works unless the
line passes through the origin. In that case revert to the alternative procedure of substituting any
other value of p and finding the equivalent value of m.
To find which side of the line to shade put p = 0 and m = 0 in the inequality: 0.02 x 0 + 0.04 x 0 <
24. This statement is true, so the origin is included in the feasible area
Special ingredient: 0.01p + 0.04m ≤ 16
Then you plot the line:
0.01p + 0.04 m = 16
Again, the origin is included in the feasible region so we shade out the area above the line. Shade
out negative values of each variable. Putting these four constraints together on one graph gives:

The area left unshaded by all of the constraints is the feasible region and this contains all of the
possible combinations of productions which will satisfy the given constraints. The co-ordination
of any point within the feasible region represents a possible combination of soft drink production
for this firm.
We must now consider how to choose the production which will maximise the firm’s daily
contribution. The objective function is:
P= 0.01p + 0.30m (£/day)
If we like p = 100 £ per day, then we can illustrate the objective function graphically. If we then
give p another value, the new line will be parallel to the one for p = 100 £ per day.
We can generate the entire family of possible contribution lines by drawing one in particular, then
moving across the feasible region parallel to it. The further from the origin we move, the larger is
the contribution.
If we draw a contribution line on the graph of the linear programme, as in figure we can move
parallel to this line across the feasible region in the direction of increasing contribution until we
reach the last feasible solution(s), before the line moves into the infeasible region.

We can see the point A is the last feasible solution. The co-ordinates of point A give the optimum
combination of production for the two drinks. The approximate co-ordinates of point A can be
read from the graph, but, for precision, the co-ordinates are calculated by solving simultaneously
the equations of the two constraints which form point A.
These two constraints are called the binding or limiting constraints. They are the resources which
are being used fully and therefore prevent the daily contributions from increasing further.
The optimum solution is the intersection of:
0.02p + 0.04 m = 24 (1)
0.01p + 0.04m = 16 (2)
Subtract (2) from (1)
0.01p = 8
Therefore:
P = 800 litres/day’
Substitute into (2) and find m:
x 800 + 0.04 m = 16
Therefore: m = 200 litres/day
To maximise the daily contribution, the firm should produce 800 litres of ‘Pink Fizz’ and 200 litres
of ‘Mint Pop’ each day. This will yield a maximum contribution of:
0.10 x 800 + 0.30 x 200 = £ 140/day
This combination utilises all of the machine time and special ingredient available each day. There
is no spare capacity or slack on either of the constraints.
This method of identifying the optimum corner depends on a suitable profit line being drawn.
The following is a practical note which will help to obtain a suitable profit line from which to
identify the optimum corner. Choose any inconvenient point near the middle of the feasible region.
Suppose in the above example the point m = 200, p = 200 is chosen. The daily contribution from
this product mix is:
P = 0.10 p + 0.30 m = 0.10 × 200 + 0.30 x 200 = £80 /day
The other entire product mix which gives daily contribution of £80 lie on the line:
80 = 0.10p + 0.30 m (£/day)
One point on this line is already known, i.e., m = 200, p = 200. A second point might be m =
0, hence, p = 800. This particular daily contribution line is now drawn on the graph and the
procedure described above is followed to identify the optimum solution (s). It is clear from the
procedure that the optimum will always be at a corner of the feasible region, or, if the objective
function is parallel to one of the constraints, at any point on the line joining two corners.
We have assumed that the variables in the linear programme are continuous or, if not, then fractions
are acceptable. It will often be the case that part units are allowed the time period of the problem.
For example, if two models of a car are being produced and the objective of the linear
programme is to maximise the machine usage per week. For such a product ‘work in progress’ is
allowable on weekly basis.
If, however, we are allocating workers to task, part workers are not acceptable. In this case the
optimum solution must produce integer values. The feasible solutions are all the points in the
feasible region where the variables are integers. The last point within the feasible region which
has integer coordinates is selected and this may no longer be at a corner of the feasible region.
For two variable linear programmes, it does not make much difference to the solution procedure
if the variables must be integers. The feasible region is placed by the set of feasible points within
the constraint boundaries. The typical objective’s function is moved through these points, rather
than through the feasible region as a whole. In the multivariable case, however, the method of
integer programming is used.
Refer to Example 2 which is concerned with the production of two machined components. We
wish to know the product mix which will achieve maximum total contribution per week.
Solution:
The feasible region for each constraint is as follows:
The feasible region, containing all of the possible product mixes for this problem, is shown
unshaded.
We wish to identify the optimum product mix which will maximise the weekly contribution. The
objective function is:
P = 30 x + 40 y (£/week)
To plot this function for a typical value of the weekly contribution, we select the point
x = 1000, y = 1000 which is in the feasible region. The weekly contribution for this product mix
is:
P = 30 x 1000 + 40 x 1000 = £ 70,000/week
We will use the contribution line:
70,000 = 30x + 40y (£/week) as the trial line. This line also passes through the point
x = 0, y = 1750. It is shown by the broken line on figure 5. In the direction of increasing
contribution leads us to point A as the last feasible solution.
The binding constraints are therefore:
Labour: x + 2y ≤ 4000 hours/week
Plate: 5x + 2y ≤ 10,000 kg/week
Solving the corresponding equations simultaneously gives
X + 2y = 4000 (1)
5x + 2y = 10,000 (2)
(2) – (1) 4x = 6,000
Therefore x = 1250, and y = 1500 by substitution.
The optimum product mix is 1,500 of component x and 1,250 of component y each week. The
maximum contribution per week will then be:
P max = 30 x 1500 + 40 x 1250 = £95,000/week
This product mix uses all of the labour hours available and all plate. These are the binding
constraints. However, there will be spare capacity on machine time for both components and
spare rod capacity. The production will also exceed the minimum required by the regular orders
and the minimum required by the union agreement.
We find the value of the slack variables in the machine time constraints are 750 of component x
and 500 of machine tool y, i.e.
1500 + s2 = 2250, therefore s2 = 750 components/week, and
1250 + s3= 1750, therefore s3 = 500 components/week
The slack in the rod constraint is:
2x 1500 + 5 x 1250 + s = 10,000
Therefore s = 750 kg/week
The surplus on the regular orders constraints is:
1500 – s6 = 600
Therefore: s6 = 900 components/week
Above the minimum needed for the regular orders. The surplus on the agreement is:
1500 + 1250 – s7 = 1500
Therefore s7 = 1250 components/week
Above the minimum required by the union agreement.
The optimum solution will normally be at a corner of the feasible region.
It is possible, therefore, once the graph has been drawn to identify the optimum corner by
evaluating the objective function at each corner of the feasible region. A basic solution is the
name given to the set of variable values at a corner of the feasible region. The basic variables
are those variables which have non-zero values at a particular corner.
Occasional problems arise when solving a linear programme. The problem may be infeasible.
In this case, there is no feasible region. No combination of the variables satisfies all of the
constraints simultaneously and the linear programme is unbounded. In this case, the solution
can be increased indefinitely without violating constraint. This usually means that the linear
programme is formulated incorrectly, with some constraints missing.
The issue of non-unique solutions was mentioned earlier. These arise when objective function is
parallel to a binding constraint. Any point on that constraint between the two optimum corners,
will give the optimum value of the objective function. Any one of these points forms an optimum
solution to the model. This can be useful situation since it gives the decision maker some
flexibility.
Sensitivity Analysis
In most decision-making activities, it is prudent to maximise the preferred course of action to
see what effect changes in the problem will have on the decision. Linear programming is no
exception. There are three aspects of the problem which we need to consider.
The effect of additional supplies of the limiting resources;
The effect of changes in the non-limiting resources;
The effect of changes in the coefficients of the objective function.
How do changes in the non-limiting resource affect the optimum solution?
In example 2 we considered the two limiting constraints of labour and plate. The other constraints
are not binding at the original optimum solution. These constraints are:
-Machine time to produce component X
-Machine time to produce component Y
-Rod
-Regular orders
-Union agreement
What happens as each of these constraints is charged?
The first three are less-than-or – equal – to constraints. Any of their availabilities will not affect
the optimum solution. However, any decrease in these three constraints can affect the solution.
The tightening of one of the non-limiting constraints will cause it to move towards the origin. At
first, the only change will be a reduction in the size of the feasible region. When, however, the
particular constraint passes through the original optimum corner. It will itself become limiting and
a new optimum solution will emerge.
It is useful to know what the lower limits are on these constraints. The machine capacity for
X can be reduced by 750 hours, from 2,250 to 1,500 hours, before it affects the solution. The
machine capacity for Y can be reduced by 500, from 1,750 to 1,250 hours. The supply of rod
can be reduced by 750 per week, from 10,000kg to 9,250kg. These reductions are the values of
the slack variables mentioned earlier. The greater- than- or-equal to constraints, for the regular
orders and the union agreement, act in the opposite way. Any reduction in these requirements
will increase the feasible region but will not affect the optimum solution.
Any increase in these constraints will first reduce the feasible region but will not affect the
optimum solution. If the regular orders for X increase by at least 90 to 1,500, the optimum will
begin to change. If the union’s agreement was increased by at least 1,250, to more than 2,750,
there would be no feasible region and no solution. These increases are the surplus referred to
earlier.
How do changes in the coefficients of the objective function affect the optimum
solution?
It is inevitable that the circumstances under which a linear programme is formulated will change.
Major changes will probably mean that the work will have to be done again but may be possible
to identify the effect of minor changes from the solution to the original problem. In this section,
we consider changes to the objective function. If the objective is to maximise weekly contribution,
a change in the cost of raw material will alter the coefficients in the objective function.
In an investment portfolio problem, if the objective is to maximise the annual return on the
investments, a change in the interest rate earned on one of the investments will change that
coefficient in the objective function.
We will consider situations when the coefficient change is one at a time.
Suppose:
P = ax + 4y (£/week) represents the objective function for a profit maximizing linear programme,
where £4 per unit is the profit on product Y and £a per unit is the profit on product X. The profit
on product X is liable to change. Suppose this linear programme has been graphed with X and
Y in the conventional directions. It is helpful to re-arrange the objective function so that y is the
subject:
Y = p/4 – (a/4) x
The profit line cuts the Y axis at p/4 and the slope of the profit line is – (a/4). The intercept on
the Y axis is independent of the value of a, but the slope of the line increases as an increase,
and decreases as a decrease. In other words, as the value of slope change, the profit lines rotate.
Small rotations in either direction will not usually alter the optimum corner. However, larger
rotations will result in different corners emerging as the optimum. It is useful to know the range
which ‘a’ can take before a particular corner cease to be the optimum. A similar argument would
apply if the coefficient of x was fixed and the coefficient of y was liable to change.

The simplex solution of multi-variable linear programmes.

An algebraic solution method is required if a linear programme contains more than two variables.
The basic principle of solution of multi-variable model is very simple. It is assumed that the
optimum solution is at one of the ‘corners’ of the feasible region. Therefore, we systematically
evaluate the objective function for each corner until we find the one which gives the optimum
value of the objective function. We employ the techniques of matrix algebra and an algorithm
for moving from corner to corner of the feasible region in such a way that a move is made only if
it improves the value of the objective function. If, at a particular basic solution, no further move
is recommended then we know that the optimum solution has been reached. This algorithm is
called the Simplex method. A detailed explanation of the simplex method is not necessary,
since multi-variable linear programming models are normally solved by using one of the many
computer packages which are readily available for this purpose. However, an understanding of
the basic principles of the method are helpful for fully interpreting and evaluating the solution to
a linear programme which has been obtained by computer package.
The basic simplex method assumes that the linear programming model is a maximizing one,
subject to a set of ≤ constraints. This means that the algorithm can take the origin as the initial
corner. The search for the optimum always starts from a zero value for the objective function.
The simplex method can be adopted for minimizing problems and for problems with ≥or =
constraints. This involves the introduction of artificial, as well as slack and surplus, variables.
The basic model with which we will work may be formally written as:
Maximise Z = c1 x1 + c2 x 2 + …… + c n x n
The C1 are constants. This is maximized subject to a set of m linear constraints:

There are n variables and m constraints. The double subscripts for the coefficients in the left
hand side of the constraints refer first to the constraints, then to the variable. For example, + a32
is in constraint 3 and is the coefficient of the variable x2. We will illustrate the use of the simplex
method by considering a simple two variable problem which we will first solve graphically. This
will enable us to compare the graphical and algebraic solutions.
Interpretation of computer-generated solution
Example
Maximise

Mathematica l Programming
Required
Interpret the data generated by the computer.
Solution.
Table 1:
Objective value, is the solution to objective function (e.g. the solution to this example is 71,436)
The four columns of table 1 are to be interpreted as follows;
•• Variable: these are the variables of the model. In our example we have x1 = Xtragrow,
x2 = Youngrow and x3 = Zupergrow
•• Value: this is value that the variables assume at optimal solution (to optimize the objective
function one needs to produce this amounts of the variables). In our example we are required to
produce 1,666.67 of x1 and 1,750 of x2 and none of x3
•• Objective coefficients: these are the coefficients of the objective function
•• Objective value contribution: this is the value contributed by each variable to the objective
function (for x1=25×1,666.67), the total of this is equal to our objective value (i.e.,
41,666.67+35,000=76,666.67).
The 3 columns of the second part of table1 can be interpreted as follows;
•• Constraints: this is constraints of the model representing the limited resources.
•• RHS: the Right-hand side value is the limiting value of the constraint. e.g., for the first constraint
the maximum amount of material A is 500 tons.
•• Slack/surplus: at optimal production not all the materials for some of the constraints will be fully
utilized, slack is the amount of material that is left over after production. For constraint 1 and 4 no
material remained; this also implies that these are the binding constraints i.e., their adjustment
directly affects the objective solution
To illustrate the use of the simplex method.
A firm manufactures two products, X and Y, subject to constraint on three raw materials, RM1,
RM2, and RM3. The objective of the firm is to select a product mix which will maximise weekly
profit.
The linear programme for the problem is:
Produce x units of product x per week and y units of product y per week.
Maximise the weekly profit, £P, where P = 2x =Y (£/week)
Maximise subject to:
RM1: 3x≤ 27 kg/week
RM2: 2y≤ 30 kg/week
RM3: x +y≤ 20 0kg/week
x, y ≥ 0
Determine the optimum product mix and the maximum value of the weekly profit. State the spare
capacity on each resource.
Solution
Simplex method arranges the coefficients in the left-hand side of the constraints equations in
matrix format. Label the columns with the name of the variables to which they refer. Put the right-
hand side values of the constraints in a separate column on the right of the matrix label the rows
with the names of the variables which are basic (have --- values) at the initial corner (the origin).
Finally, add the objective function as an addition row to the table. There are several slightly
different of the objective function to be entered as negative values. The resulting matrix is called
coefficients of the objective function to be entered as negative values. The resulting matrix is
called the first simplex tableau. The above procedure is step 1 in the logarithm

Step 2: Find the largest negative value in the objective function row (-2). The corresponding
column is called the pivot column x. Divide the right-hand side values (in the b column) by the
corresponding numbers in the pivotal column. This produces a set of ratios.

Step 3: Choose the smallest positive ratio, 9. The corresponding row, s, is the pivotal row. The
intersection of the pivotal column, x and the pivotal row s1, is the pivotal element 3, marked by
in table in step 2 above.
Step 4: Divide all of the elements in the pivotal row by the element, 3 replace the pivotal row by
this new row in Table below. Replace the row label, s1 by the label from the pivotal column, x.
the new row labels are the basic variables for the second basic solution.
Step 5: Using arithmetic operations on the rows (row operations in matrix algebra), reduce all of
the other elements in the pivotal column, x, to zero. These arithmetic operations must use only
the pivotal row as the basis.
R1 denotes the ith with the notation new R3 = Old R3 – New R1 means that the new row 3 is
obtained by subtracting the new pivotal row (row 1) from the old row 3. The operations used are
listed at the right-hand side of Table below.

2 5 2 quantitative techniques
Step 6; repeat steps 2 to 5 until all of the elements in the objective function row are zero or
positive.
All values in the objective function row are now positive or zero, therefore this tableau represents
the optimum solution.
To interpret the final simplex tableau, we will look at the values around the edge of the table
first.

l
Programming
The basic variables are those which are non-zero at the optimum corner. The values of the basic
variables are in the corresponding row of the B column. Therefore:
X = 9 units/week
Y = 11 units/week
And the slack for raw material 2 is 8kg per week.
All other variables are zero that is the slack on constraint 1, s1 and constraint 3, s3 are zero. This
means that these constraints are binding and the available raw material 1 and 3 are fully used. The
optimum value of the objective functions is in the ‘b’ column of the objective function row. The
maximum value of weekly profit is £ 29. This solution corresponds exactly with the graphical
solution. The figures in the objective function row and the slack variable columns
shown in Table above give the shadow prices. The shadow price on constraint 1, RMI, is £ 1/3
per kg and the shadow price for constraint 3 is £1 per kilograms of RM1 becomes available;
the weekly profit will increase by 33 pence (less any additional costs above the normal cost of
RM1). Similarly, if an extra kilogram of RM3 becomes available, the weekly profit will increase
by £ 1 (less any additional costs). The figures for the shadow prices may be checked from the
graphical solution.
We will look at constraint 1 only to illustrate the point.
Constraint 1, for raw material 1 is 3x = 27 kg per week, if this constraint is relaxed by one
kilogram, 3x = 28. The optimum corner will still be the intersection of constrains 1 and 3. Look
back at the graph to check this. The new optimum corner has the co-ordinates:
X = 28/3 = 9 1/3
And 28/3 + y = 20
Giving:
Y = 32/3 = 10 2/3
The new value of the maximum weekly profit is:
2 x (28/3) + (32/3) = 88/3 = £ 29.33/ week
This is again 33 pence for one kilogram
The shadow price for RM1 is 33 pence kilogram. The remaining values on the page on the edge
of the final tableau, are those in the objective function row and the variable columns. In this
example, the value in both the x and the y columns are zero. These numbers will be non- zero if
any of the variables is non-basic in the optimum solution. For example, if the optimum solution
had said that we should produce only product X, then Y would be non-basic, is y = 0. In that case,
the figure in the Y column of the objective function row tells us by how much the maximum value
of the objective function would decrease if we insisted on producing one unit of Y.
Suppose the following had been the final tableau for the current problem:

254q
In this solution, the optimum produces 9 units of x and none of y. If we feel that we must produce
some units of y, the value of the objective function will decrease by £0.5 for each unit of y which
is produced.
Shadow or dual prices
Definition: A shadow price or a dual price is the amount increase (or decrease) of the objective
function when one more (or one less) of the binding constraints is made available.
Maximise 4X1 + 3 X2
Subject to: 0.5X1 + 0.33X2 ≤ 12 (Machine hours)
0.5X1 + 0.5X2 ≤ 14 (labor hours)
Starting with machine hours; let’s assume that one more machine hour is available (with labour
hours remaining constant)
We get:
0.5X1 + 0.33X2 = 13
0.5X1 + 0.5X2 = 14
Solving this simultaneously we get the values of X1 and X2 as
0.17 X2 = 1
X2 = 5.88
X1 = 22.12
Thus, the contribution is
4(22.12) + 3(5.88) = Sh.106.12
Comparing this with its original contribution of Sh100.24 (see example 1) we see increasing
machine hours by one unit have increased contribution by Sh5.88, which is the shadow price per
machine hour.
Note: This figure is also arrived at if we assume that machine hours are reduced by 1-unit i.e.
12-1.
Similarly, assuming that one more labour hour is made available, then contribution change is:
0.5X1 + 0.33X2 = 12
0.5X1 + 0.5X2 = 15
Solving this simultaneously gives:
0.17 X2 =3
X2 = 17.65
X1 = 12.35
Which give a contribution of:
4(12.35) + 3(17.65) = Sh102.35
The contribution change is Sh2.11 which is the shadow price per labour hour.
Note:
The shadow prices apply in so far as the constraint is binding; for example, if more and more
labour hours are available, it will reach a point where labour hours are no longer scarce thus labour
hours cease to be a binding constraint and its shadow price becomes a zero. (All nonbinding
constraints have zero shadow price). Logically, it is senseless to pay more to increase a resource,
which is already abundant.
Interpretation of shadow prices
A shadow price of a binding constraint indicates to management how much extra contribution will
be gained by increasing a unit of the scarce resource.
In the example above, Sh2.11 is the shadow price for labour hours. This implies that management
is ready to pay up to Sh2.11 extra per hour for the extra hours i.e., say an employee is paid Sh5 per
hour and one day he works for two hours extra (overtime), the management is prepared to pay up
to Sh7.11 per hour for the two hours overtime worked.
Sensitivity analysis and the simplex method
The final tableau of the simplex algorithm can be used to carry out a sensitivity analysis of the
solution of a linear programming model. For limiting constraints, the values in the corresponding
slack variable column represent the change in the values of the basic variables if one additional
unit of the limiting resource is available.
Example
To illustrate the use of the final simplex tableau for a sensitivity analysis of the limiting constraints,
we will use the final simplex tableau of Example above to determine:
The effect on the optimum solution if one additional kilogramme of RM1 becomes available
The effect on the optimum solution if two additional kilogrammes of RM1 become available
The effect on the optimum solution if five additional kilogrammes of RM3 become available
The maximum number of additional kilogrammes of RM3 which can be used without spare
capacity being created
The effect on the optimum solution if two fewer kilogrammes of RM1 are available
Solution
The linear programme and the final simplex tableau are now produced
Maximise the weekly profit, £P, where P = 2x + y (£/week)
Subject to RM1: 3x ≤ 27 kg/week
RM2: 2y ≤ 30 kg/week
RM3: x +y ≤ 20 kg/week
x, y ≥ 0

If one additional kilogramme of RM1 is available, this limiting constraint is relaxed by one
kilogramme. The values in the s1 column are the changes in the basic variables which result from
this relaxation. The final tableau is re-written below with only the relevant values and calculations
shown.

One additional kilogramme of RM1 causes the value of x to increase by 1/3 of a unit, the slack
for RM2 to increase by 2/3 of a kilogramme, the value of y to decrease by 1/3 of a unit and
the maximum value of the weekly profit to increase by £1/3 i.e., by the shadow price. The new
optimum solution requires 9 1/3 of product x and 10 2/3 of product y to be produced each week.
The slack on constraint 2, the amount of raw materials not used, is 82/3 kg. The other variables
are zero. The slack on constraints 1 and 3 is zero; therefore, all RM1 and RM3 are used.
If two additional kilogrammes of RM1 are available, this limiting constraint is relaxed by two
kilogrammes. The values in the s1 column are multiplied by two. The resulting values are then
the changes in the values of the basic variables which arise from the additional two kilogrammes.
The final tableau is shown in table below with only the relevant values and calculations shown.
2 57
The new optimum solution requires 9 2/3 of product x and 10 1/3 of product y to be produced
each week. The slack on constraint 2 is 9 1/2kg. The other variables s2 and s3 are zero. This
means that these constraints are binding. The maximum value of weekly profit is £29.67. This
solution can be illustrated in the same way as 1 above.
If five additional kilogrammes of RM3 are available, this limiting constraint is relaxed by five
kilogrammes. The values in the s3 column multiplied by five kilogrammes are shown in the
modified final tableau below

Basic
A problem has now arisen. The value of the slack variable, s2 for raw materials 2, has become
negative. This is not allowed, since variables must always be positive or zero. If we consult the
graphical solution we can see at once what has happened. The RM3 constraint has been relaxed
so far that it is no longer limiting. The tableau gives a point outside the feasible region. We are
not able to use all the extra five kilogrammes of RM3. This problem is discussed further in section
4.
The RM3 constraint is represented by the s column of the final tableau. The only negative value
in the s column is the marginal value of s2 which is –2. As RM3 is relaxed, the value of s2
decreases by 2, but it cannot be negative. When s2 reaches zero, the limiting position for the
RM3 constraint will occur. Suppose this limiting position is reached when the RM3 constraint has
been relaxed by r kg, the value of s2 will then be zero, therefore:
8 + ( -2 xr) = 0
Which means that r = 4kg. The RM3 constraint may be relaxed by 4kg from 20 to 24kg before it
ceases to be binding. This takes it to intersection of the RM1 and RM2 constraints where x = 9,
y =15 and the RM3 constraint is x +y =24
If fewer kilogrammes of RM1 are available, this limiting constraint is tightened by two
kilogrammes.
The values in the s column are multiplied by two. The resulting values are deducted from the
values of the basic variables. The final tableau is re-written below.

s2
The new optimum solution requires 8 1/3 of product x and 11 2/3 of products y to be produced
each week. The slack on constraint 2 is 6 2/3kg while the slack on constraints 1 and 3 is zero.
This means that these constraints are binding. The maximum value of weekly profit is £ 28.33.
This analysis is tedious to do by hand using the simplex method, even with the simplest two
variable models. All commercial standard linear programming computer packages will provide
the information. This is how multivariable sensitivity analysis is done in practice. The principles
are exactly the same as those for the two variable models we have just completed.
The dual linear programming model
The dual linear programming is used to investigate a problem from a different perspective to the
one obtained from the usual primal model. The primal and dual models give the same solution
and the same sensitivity information. The only reason for using one rather than the other is that
computationally one may be easier to solve. With increasingly powerful computer packages the
need for the primal/dual switch is becoming less relevant. The variables in the dual model are
the shadow prices of the original, or primal, model. The structures of dual and primal models are
similar. If the primal model has been built, the corresponding dual model is derived from it. In
general, a linear programming problem can be described by:
The above linear programme maximises and has all ≤ constraints. Any linear programming model
can be put into this form and converted into its dual, as we show below. The dual model is:

There are m dual variables y, one for each of the m primal constraints, and n constraints, one for
each of the x variables in the primal. The coefficients c, in the primal objective function and the
right-hand side values, b, of the constraints in the primal constraints are interchanged row to
column on the dual. The dual variables, y are the shadow prices in the minimal problem and vice
versa. In this case the dual minimizes, the primal maximises. The primal has ≤ constraints and
the dual ≥ constraints.
To set and interpret the dual linear programme
A firm makes two products, R and Q both of which require two raw materials RM1 and RM2.
Each kilogramme of product R requires 2kg of RM1 and 3.5kg of RM2. Each kilogramme of
product Q requires 3kg of RM1 and 1.5kg of RM2. Each week 10kg of RM1 and 12kg of RM2
are available.
There is an unlimited supply of labour and machine time and the firm can sell all its production.
The unit profit on product R is £5 and on product Q is £8.
Required
1. Set up a profit maximizing linear programming model for this problem
2. Set up the dual linear programming model
3. Explain the relationship between the two models in 1 and 2
4. Find the optimum solution for the two models graphically.
Solution:
Produce x1 kg of product R and x2 kg of product Q each week. Maximise weekly profit, £p
where:
P = 5x1 + 8x2 (£/week)
Subject to; RM1: 2x1 + 3x2≤10
3.5x1+1.5x2≤12
Using the primal model, the dual model is:
minimize: G = 10 y1 + 12y2
subject to:
product R: 3y1 + 1.5 y2 ≥ 8 £/ unit
Primal model: The variables are the amount of each product to be produced each week. The
objective function is the total profit per week from the production of R and Q. Each constraint
refers to one raw material. The left-hand side of the constraint gives the total requirement for that
raw material. The left-hand side of the constraint gives the total requirement for that raw material
by both of the products. The right-hand side gives the total raw material available each week.
Dual model: the variables are the primal shadow price, that is, the amount which would be added
to the value of the objective function if one more unit of the raw material was available.
The shadow prices represent the value of one unit of the raw material. The objective function is
the total value per week of the raw materials used in the production of R and Q. each constraint
refers to one product. The left-hand side of the constraint gives the total value of both raw
materials used to make one kilogramme of that product. The right hand side gives the unit profit
generated by that product. Let us look at the dual model again and try to identify the individual
components
Minimize G:
10y1 + 12y2
Product R:
2y1 + 3.5 y2 ≥ £5 profit per R
Product Q:
3y1 + 1.5y2 ≥ £8 profit per Q
Each constraint says that the total value of the raw materials used in that product must be more
than or equal to the unit profit on that product. The solution of either the primal or the dual model
enables us to solve the other model.
The graphical solution for the primal problem is given below.

Primal model

To maximise profit, we should produce 31/2 kg of Q and zero kilogrammes of R. all raw material
1 will be used, but not all of RM2. The maximum profit is 31/2 x 8 = £ 26.67 per week.
The graphical solution for the dual is shown below.
Dual model
This is a minimization problem. We wish to make the objective function as small as possible,
hence, we move towards the origin, parallel to trial objective function. The last point in the feasible
region is given by Z. this is the optimum solution for the dual. Z is the intersection of the product
Q constraint and the y-axis i.e.
Y2 =0 and 3y1 + 1.5y2 = 8
Therefore y2 = 0 and y1 = 2 2/3
The minimum value for the dual problem is:
G = 10 x 2 2/3 + 12 x 0 = £26.67

This is the same value of the objective function as the one for the primal model. The two solutions
combined tell us that the maximum profit is £26.67 per week, when we produce 31/2 kg of Q and
none of R, the value of the raw materials, the shadow price, is £2.67 per kg of RM1 and zero for
RM2. This is the same as the information which we would derive from a full