OPERATIONS RESEARCH

LECTURE NINE
Duality and Sensitivity Analysis
Lecturer: Dr. Emily Roche

INTRODUCTION
This lecture will focus on solving minimization problems in linear programming
together with sensitivity analysis.

Intended learning outcomes

At the end of this lecture, you will be able to define the dual of a LPP, Perform
sensitivity analysis on LPP and apply sensitivity analysis in interpretation of business
problems.

References
These lecture notes should be supplemented with relevant topics from the books listed
in the Bibliography at the end of the lecture

Dual and Minimization

A solution of a standard maximum problem produces the solution of an associated
standard minimum problem and vice-versa. These associated problems are called the
dual of each other.

Solving minimization problems with Dual

A standard minimization problem is of the form
Minimize 𝑦0 = 𝑐1 𝑦1 + 𝑐2 𝑦2 + 𝑐3 𝑦3 + ⋯ + 𝑐𝑛 𝑦𝑛
 𝑎11 𝑦1 + 𝑎12 𝑦2 + 𝑎13 𝑦3 + ⋯ + 𝑎1𝑛 𝑦𝑛 ≥ 𝑏1
𝑎21 𝑦1 + 𝑎22 𝑦2 + 𝑎23 𝑦3 + ⋯ + 𝑎2𝑛 𝑦𝑛 ≥ 𝑏2
𝑎 𝑦 + 𝑎32 𝑦2 + 𝑎33 𝑦3 + ⋯ + 𝑎3𝑛 𝑦𝑛 ≥ 𝑏3
Subject to 31 1
…………………………………………………
𝑎𝑚1 𝑦1 + 𝑎𝑚2 𝑦2 + 𝑎𝑚3 𝑦3 + ⋯ + 𝑎𝑚𝑛 𝑦𝑛 ≥ 𝑏𝑚
𝑦1 , 𝑦2 , 𝑦3 , … 𝑦𝑛 ≥ 0

The following steps are then followed to solve the minimization problem:

1. Find the dual standard maximum problem. This is obtained through

 Form the augmented matrix for the given system of linear inequalities, with the top
row consisting of the coefficients of the objective function. That is:
𝑐1 𝑐2 𝑐3 ⋯ 𝑐𝑛 ⋮ 0
⋯ ⋯ ⋯ ⋯ ⋯ ⋯ ⋯
𝑎11 𝑎12 𝑎13 ⋯ 𝑎1𝑛 ⋮ 𝑏1
𝑎21 𝑎22 𝑎23 ⋯ 𝑎2𝑛 ⋮ 𝑏2
𝑎31 𝑎32 𝑎33 ⋯ 𝑎3𝑛 ⋮ 𝑏3
⋮ ⋮ ⋮ ⋯ ⋮ ⋮ ⋮
[𝑎𝑚1 𝑎𝑚2 𝑎𝑚3 ⋯ 𝑎𝑚𝑛 ⋮ 𝑏𝑚 ]
 Form the transpose of this matrix
𝑐1 ⋮ 𝑎11 𝑎21 𝑎31 ⋮ 𝑎𝑚1
𝑐2 ⋮ 𝑎12 𝑎22 𝑎32 ⋯ 𝑎𝑚2
𝑐3 ⋮ 𝑎13 𝑎23 𝑎33 ⋮ 𝑎𝑚3
⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮
𝑐𝑛 ⋮ 𝑎1𝑛 𝑎2𝑛 𝑎3𝑛 ⋮ 𝑎𝑚𝑛
⋯ ⋯ ⋯ ⋯ ⋯ ⋯ ⋯
[0 ⋮ 𝑏1 𝑏2 𝑏3 ⋮ 𝑏𝑚 ]
With the last row being the coefficients of the objective function.
 Develop the dual maximization problem using the transposed matrix

Maximize 𝑥0 = 𝑏1 𝑥1 + 𝑏2 𝑥2 + 𝑏3 𝑥3 + ⋯ + 𝑏𝑚 𝑥𝑚
𝑎11 𝑥1 + 𝑎21 𝑥2 + 𝑎31 𝑥3 + ⋯ + 𝑎𝑚1 𝑥𝑚 ≤ 𝑐1
𝑎12 𝑥1 + 𝑎22 𝑥2 + 𝑎32 𝑥3 + ⋯ + 𝑎𝑚2 𝑥𝑚 ≤ 𝑐2
𝑎 𝑥 + 𝑎23 𝑥2 + 𝑎33 𝑥3 + ⋯ + 𝑎𝑚3 𝑥𝑚 ≤ 𝑐3
Subject to …13…1… … … ……………………………………
𝑎1𝑛 𝑥1 + 𝑎2𝑛 𝑥2 + 𝑎3𝑛 𝑥3 + ⋯ + 𝑎𝑚𝑛 𝑥𝑚 ≤ 𝑐𝑛
𝑥1 , 𝑥2 , 𝑥3 , … 𝑥𝑚 ≥ 0
2. Solve the dual maximization problem using the simplex method.
3. The maximum value of the dual objective function (𝑥0 ), will be the same as the
minimum value of the primal objective function (𝑦0 ).
4. The optimum value of the primal decision variables will be the entries in the indicator
row of the columns corresponding to the slack variables.

Example
Use dual theory to solve the LPP given below

Minimize 𝑤 = 2𝑥1 + 10𝑥2 + 8𝑥3

𝑥1 + 𝑥2 + 𝑥3 ≥ 6
𝑥2 + 2𝑥3 ≥ 8
Subject to
−𝑥1 + 2𝑥2 + 2𝑥3 ≥ 4
𝑥1 , 𝑥2 , 𝑥3 ≥ 0
Solution
2 10 8 0
1 1 1 6
The augmented matrix of the system is [ ]
0 1 2 8
−1 2 2 4

2 1 0 −1
10 1 1 2
The transpose then becomes [ ]
8 1 2 2
0 6 8 4

This then gives the dual problem as

Maximize 𝑧 = 6𝑦1 + 8𝑦2 + 4𝑦3
𝑦1 − 𝑦3 ≤ 2
𝑦1 + 𝑦2 + 2𝑦3 ≤ 10
Subject to
𝑦1 + 2𝑦2 + 2𝑦3 ≤ 8
𝑦1 , 𝑦2 , 𝑦3 ≥ 0
In standard form:
Maximize 𝑧 − 6𝑦1 − 8𝑦2 − 4𝑦3 = 0
𝑦1 − 𝑦3 + 𝑠1 = 2
𝑦 + 𝑦2 + 2𝑦3 + 𝑠2 = 10
Subject to 1
𝑦1 + 2𝑦2 + 2𝑦3 + 𝑠3 = 8
𝑦1 , 𝑦2 , 𝑦3 , 𝑠1 , 𝑠2 , 𝑠3 ≥ 0
The initial tableau in using the simplex procedure is
Basic 𝑧 𝑦1 𝑦2 𝑦3 𝑠1 𝑠2 𝑠3 RHS
𝑧 1 −6 −8 −4 0 0 0 0
𝑠1 0 1 0 −1 1 0 0 2
𝑠2 0 1 1 2 0 1 0 10
𝑠3 0 1 2 2 0 0 1 4

The final simplex tableau is

Basic 𝑧 𝑦1 𝑦2 𝑦3 𝑠1 𝑠2 𝑠3 RHS
𝑧 1 0 0 2 2 0 4 36
𝑦1 0 1 0 −1 1 0 0 2
𝑠2 0 0 0 3/2 −1/2 1 −1/2 5
𝑦2 0 0 1 3/2 −1/2 0 1/2 3

The optimal primal solution from the dual final simplex tableau is:
Minimum 𝑤 = Maximum 𝑧 = 36

𝑥1 = 2, 𝑥2 = 0, 𝑥3 = 4 these are from the coefficients of the slack variables

in the indicator row.
Note
1. Slack variables correspond to the primal basic variables in the optimal solution.
2. Values in the indicator row corresponding to the columns of the slack variables
gives the optimum values of the basic primal variables.
 3. Values of the slack variables of the primal are given by the indicator row under
the non-basic variables of the dual.
4. The value of the objective function is the same for both the primal and the dual
problems.

Interpretation of the Simplex solution.

Example
A company can produce three products 𝑃, 𝑄, and 𝑅.
The products yield a contribution of 𝑆ℎ𝑠 8, 𝑆ℎ𝑠 5 and 𝑆ℎ𝑠 10 respectively.
The products use a machine which has 400 hours capacity in the next period. Each unit
of the products uses 2, 3 and 1 hour respectively of the machine’s capacity.
There are only 150 units available in the period of a special component which is used
singly in products 𝑃 and 𝑅.
200𝐾𝑔𝑠 only of a special alloy is available in the period. Product 𝑃 uses 2𝐾𝑔𝑠 per unit
and product 𝑅 uses 4𝐾𝑔𝑠 per unit.
There is an agreement to produce no more than 50 units of product 𝑄 in the period.
Formulate and solve the LOM that would enable the company determine the optimal
production plan.

Solution
Let 𝑥𝑖 𝑖 = 1, 2, 3 be the number of units of product 𝑃, 𝑄, 𝑅
Maximize 𝑧 = 8𝑥1 + 5𝑥2 + 10𝑥3
2𝑥1 + 3𝑥2 + 𝑥3 ≤ 400 (𝑀𝑎𝑐ℎ𝑖𝑛𝑒 ℎ𝑜𝑢𝑟𝑠 𝑐𝑜𝑛𝑠𝑡𝑟𝑎𝑖𝑛𝑡)
𝑥1 + 𝑥3 ≤ 150 (𝐶𝑜𝑚𝑝𝑜𝑛𝑒𝑛𝑡 𝑐𝑜𝑛𝑠𝑡𝑟𝑎𝑖𝑛𝑡)
Subject to 𝑥1 + 4𝑥3 ≤ 200 (𝐴𝑙𝑙𝑜𝑦 𝑐𝑜𝑛𝑠𝑡𝑟𝑎𝑖𝑛𝑡)
𝑥2 ≤ 50 (𝑆𝑎𝑙𝑒𝑠 𝑐𝑜𝑛𝑠𝑡𝑟𝑎𝑖𝑛𝑡)
𝑥1 , 𝑥2 , 𝑥3 ≥ 0
In standard form:
Maximize 𝑧 − 8𝑥1 − 5𝑥2 − 10𝑥3 = 0
2𝑥1 + 3𝑥2 + 𝑥3 + 𝑠1 = 400 (𝑀𝑎𝑐ℎ𝑖𝑛𝑒 ℎ𝑜𝑢𝑟𝑠 𝑐𝑜𝑛𝑠𝑡𝑟𝑎𝑖𝑛𝑡)
𝑥1 + 𝑥3 + 𝑠2 = 150 (𝐶𝑜𝑚𝑝𝑜𝑛𝑒𝑛𝑡 𝑐𝑜𝑛𝑠𝑡𝑟𝑎𝑖𝑛𝑡)
Subject to 𝑥1 + 4𝑥3 + 𝑠3 = 200 (𝐴𝑙𝑙𝑜𝑦 𝑐𝑜𝑛𝑠𝑡𝑟𝑎𝑖𝑛𝑡)
𝑥2 + 𝑠4 = 50 (𝑆𝑎𝑙𝑒𝑠 𝑐𝑜𝑛𝑠𝑡𝑟𝑎𝑖𝑛𝑡)
𝑥1 , 𝑥2 , 𝑥3 , 𝑠1 , 𝑠2 , 𝑠3 , 𝑠4 ≥ 0
The slack variables represent the spare capacity in the limitations.

From the simplex method the final tableau is:

Basic 𝑥0 𝑥1 𝑥2 𝑥3 𝑠1 𝑠2 𝑠3 𝑠4 RHS
𝑧 1 0 0 6 0 0 4 5 1050
𝑠1 0 0 0 −3 1 0 −1 −3 50
𝑠2 0 0 0 −1 0 1 −1/2 0 50
𝑥1 0 1 0 2 0 0 −1/2 𝑜 100
𝑥2 0 0 1 0 0 0 0 1 50

Information from the tableau can be interpreted as

1. Reading from the RHS of the final tableau:
 Optimum product mix 𝑥1 = 100 and 𝑥2 = 50 means that the company should
produce 100 of product 𝑃, and 50 of product 𝑄.
 𝑠1 = 50 and 𝑠2 = 50. Note that 𝑠1 corresponds to the machine hours constraint, thus,
50 machine hours remain unused at optimum. Similarly 50 components are unused
at optimum.
 The other two slack variables do not have values on the RHS of the final simplex
tableau meaning 𝑠3 = 𝑠4 = 0. This means that alloy and sales constraints have no
unused capacity at optimum implying the constraints they represent are fully
utilized.
 The maximum contribution at optimum (value of the objective function) is 1050.
 𝑥3 = 0 implies that product 𝑅 was not in the optimum production plan.
 2. Reading from the indicator row
 The value 6 under the 𝑥3 column means that if any unit of product 𝑅 was produced,
then the overall contribution would fall by 𝑠ℎ 6.
 The values for the slack variables in the indicator row are valuations of resources
and are known as shadow prices.
The shadow price of a binding constraint provides valuable guidance to the
management as it indicates the magnitude of change in the overall contribution that
would result from altering the amount of the available scarce resource. Therefore:
 𝑠1 = 𝑠2 = 0 means the no value would gained by increasing both the
machine hours and number of components.
 𝑠3 = 4 this means that for every kilogram of alloy available, 𝑠ℎ 4 extra
overall contribution will be gained.
 𝑠4 = 5 means the every extra unit of 𝑄 that was allowed to be produced, the
overall contribution would increase by 𝑠ℎ 5.
Note – only the constraints 𝑠3 and 𝑠4 are binding constraints (fully utilized) have non-
zero shadow prices.
The shadow prices of the binding constraints can be used to confirm the overall
contribution i.e.
𝐴𝑙𝑙𝑜𝑦 𝑎𝑣𝑎𝑖𝑙𝑎𝑏𝑖𝑙𝑖𝑡𝑦 = 200𝑘𝑔 × 4 = 𝑠ℎ 1800
𝑆𝑎𝑙𝑒𝑠 𝑐𝑜𝑛𝑠𝑡𝑟𝑎𝑖𝑛𝑡 = 50 × 5 = 𝑠ℎ 250
𝑇𝑜𝑡𝑎𝑙 𝑐𝑜𝑛𝑡𝑟𝑖𝑏𝑢𝑡𝑖𝑜𝑛 = 𝑠ℎ 1050

Sensitivity analysis (Post-optimality analysis)

This is usually undertaken to explore the effect of changes in the LP parameters on the
optimum solution. It is concerned with the extent of sensitivity of the optimum solution
on change in one or more of:
i. The profit or cost coefficients of the objective function.
ii. The LHS coefficients of the variables in the constraints.
iii. The RHS quantities of the constraints.
This analysis can be of the following types.
Type I
To determine the range within which each of the profit (cost) coefficients can be varied
without altering the basic variables though the optimum value would actually vary.

Type II
To indicate the extent of increase of the RHS quantities without altering the existing
basic variables.

Type III
To investigate the effect of adding r deleting constraints and/or variables on the
optimum product mix.

Type IV
To determine the sequence of the basic solutions that become optimum as the changes
in the LP parameters are extended further and further.

Bibliography
Lucey, T. (2002). Quantitative Techniques (6th ed.). Cengage Learning.

Taha, H. A. (2017). Operation Research An introduction (10th ed.). Prentice-Hall, Inc.