Indr 460

Operations Research
Applications

Sensitivity and Post-optimality

Analysis

Koc University – Spring 2019

1
Post-optimality Analysis
p When the output of a model is studied, often
many what-if questions arise:
n What if estimates of the parameters in the model
are inaccurate?
n How do the conclusions change if different
assumptions are made about the problem?
n What happens when certain managerial options are
pursued that are not incorporated in the current
model?

2
Why Post-optimality Analysis?
1. Typically, many of the input parameters are only
estimates and need to be refined if the model output
is “sensitive” to small changes in these parameters.

2. Possible future changes in a dynamic problem

environment can often be analyzed easily without
resolving the model.

3. When certain parameters in the model represent

managerial policy decisions, post-optimality analysis
provides guidance to management about the impact
of altering these policies.

3
Sensitivity Analysis
p Sensitivity analysis is used to determine how
the optimal solution is affected by changes in
model inputs

p After an optimal solution is obtained to the

model, the range of values for
n the objective function coefficients
n the right-hand side (RHS) of constraints
that would keep the current results valid are
investigated

4
Graphical Sensitivity Analysis
p For linear programming problems with two
decision variables, graphical methods can be
used to perform sensitivity analysis on
n the objective function coefficients, and
n the right-hand-side values for the constraints.

5
Changes in Objective Function
Coefficients
p Suppose an LP has been solved to optimality
with specified values of the input data and
the coefficient of a variable in the objective
function takes a different value.
1. Would the feasible region be affected?
2. Would the optimal solution be affected?
3. Without solving the new LP, can we conclude if
the optimal solution will be different?

6
Objective Function Coefficients
p The range of optimality for each coefficient
provides the range of values over which the
current solution will remain optimal.

p We should focus on those objective

coefficients that have a narrow range of
optimality and coefficients near the endpoints
of the range.

7
Example 1
p LP Formulation

Max z = 5x1 + 7x2

s.t. x1 < 6 (1)

2x1 + 3x2 < 19 (2)
x1 + x2 < 8 (3)

x1, x2 > 0

8
Example 1 Graphical Solution
x2
8
x1 + x2 < 8
Max 5x1 + 7x2
7

6
x1 < 6
5
Optimal:
4
x1 = 5, x2 = 3, z = 46
3

2
2x1 + 3x2 < 19
1

1 2 3 4 5 6 7 8 9 10 x1
9
Example 1 Changing the Slope of the
Objective Function
x2 8

5 5
4

3
Feasible
4
2
Region 3
1

1 2
1 2 3 4 5 6 7 8 9 10 x1
10
Range of Optimality
p Changing a coefficient in the objective function
changes the slope of the objective function line.

p Note that when the objective function is c1x1 + c2x2,

the objective function line c1x1 + c2x2 = z has slope
= -c1/c2 ).
p However, as long as the slope of the objective
function line lies between the slopes of the binding
constraints at the optimal point, the optimal solution
remains the same.

11
Calculating Range of Optimality
1. Determine which constraints are binding at
the current optimal solution.
2. Find the slopes of the binding constraints.
Let s1 and s2 be the slopes, where s1 £ s2.
3. To guarantee the optimality of the current
solution the following must hold:
s1 £ slope of objective function line £ s2

12
Calculating Range of Optimality

For the coefficient of x1:

n Replace the coefficient of x1 in the objective by a
parameter, c1 and write the slope of the objective
function line.
n Write s1 £ slope of objective function line £ s2, and
solve for c1.
For the coefficient of x2:
n Replace the coefficient of x2 in the objective by a
parameter, c2 and write the slope of the objective
function line
n Write s1 £ slope of objective function line £ s2,
and solve for c2.
13
Example 1 Range of optimality
x2
8
x1 + x2 < 8 Max 5x1 + 7x2
7

5
x1 < 6
4

1 2x1 + 3x2 < 19

1 2 3 4 5 6 7 8 9 10 x1
14
Example 1
p Slope of binding constraints:
2
!" = −1, !' = −
3
p Range of optimality for coefficient of x1:
Find the range of values for c1 (with c2 staying at
7) such that the objective function line slope lies
between that of the two binding constraints:
−*"
≥ −1 ⟹ *" ≤ 7
7
−*" 2 14
≤ − ⟹ *" ≥
7 3 3
14
≤ *" ≤ 7
3 15
Class Exercise
p Range of Optimality for coefficient of x2:
Find the range of values for c2 (with c1
staying at 5) such that the objective function
line slope lies between that of the two binding
constraints:

!"
#$
≥ −1 ⟹ )* ≥ 5
−5 2 15
≤ − ⟹ )* ≤
)* 3 2
15
≤ )* ≤ 5 16

2
Changes in Right Hand-Side Coefficients

p Suppose an LP has been solved to

optimality with specified values of the
input data and the RHS coefficient of a
constraint takes a different value.

1. Would the feasible region be affected?

2. Would the optimal solution be affected?
3. Without solving the new LP, can we measure
the effect on the optimal solution and its
objective value?

17
Example 1 RHS changes
x2
8
x1 + x2 < 8
7

5
x1 < 6
Optimal:
4
x1 = 5, x2 = 3,
3 z = 46
2
Max 5x1 + 7x2
1 2x1 + 3x2 < 19

1 2 3 4 5 6 7 8 9 10 x1
18
Right-Hand Sides
p As the RHS of a (non-redundant) constraint
changes, the feasible region will expand or
contract.
n If the constraint is in the form
a1x1 + a2x2 £ b,
then increasing b would relax the constraint, so the
feasible region expands.
n If the constraint is in the form
a1x1 + a2x2 ³ b,
then increasing b would restrict the constraint, so
the feasible region contracts.
19
Right-Hand Sides
p The change in the objective function value of
the optimal solution per unit increase in the
right-hand side is called the shadow price.

p The range of feasibility is the range over

which the shadow price is applicable.

p If the change in RHS is within the range of

feasibility, then we can use the shadow price
information.
p Outside the range, other constraints will
become binding and the shadow price is no
more meaningful.
20
 Shadow Price Calculation

1. Shadow price of a non-binding constraint is 0.

2. To determine the shadow price for a binding

constraint:
n add 1 to the right hand side of the constraint
n take the two binding constraints and resolve for the
optimal solution values
n shadow price = (optimal objective function value of
the new problem) – (optimal objective function
value of the original problem).

21
Class Exercise
p Shadow Prices?
n Constraint 1: x1 < 6
n Constraint 2: 2x1 + 3x2 < 19
n Constraint 3: x1 + x2 < 8

x2
8
x1 + x2 < 8
7

6
x1 < 6
5
Optimal:
4
x1 = 5, x2 = 3,
3 z = 46
2
Max 5x1 + 7x2
1 2x1 + 3x2 < 19

1 2 3 4 5 6 7 8 9 10 x1 22
Range of Feasibility
p The range of feasibility for a change in the
right hand side value is the range of values
for this coefficient such that the shadow price
remains constant.

p Graphically, the range of feasibility is

determined by finding the values of a right
hand side coefficient such that the same two
lines that determined the original optimal
solution no longer determine the optimal
solution for the problem.
23
Class Exercise
p Range of feasibility?
n Constraint 1: x1 < 6
n Constraint 2: 2x1 + 3x2 < 19
n Constraint 3: x1 + x2 < 8

x2
8
x1 + x2 < 8
7

6
x1 < 6
5
Optimal:
4
x1 = 5, x2 = 3,
3 z = 46
2
Max 5x1 + 7x2
1 2x1 + 3x2 < 19

1 2 3 4 5 6 7 8 9 10 x1 24
Standard Computer Output
Software packages provide the following LP
information:
1. Information about the objective function:
n its optimal value
n coefficient ranges (ranges of optimality)
2. Information about the decision variables:
n their optimal values
n their reduced costs
3. Information about the constraints:
n the amount of slack or surplus
n right-hand side ranges (ranges of feasibility)
n the shadow prices 25
Example 2: Olympic Bike Co.

Olympic Bike is introducing two new

lightweight bicycle frames, the Deluxe and
the Professional, to be made from special
aluminum and steel alloys.

p The anticipated unit profits are $10 for the

Deluxe and $15 for the Professional.

p A supplier delivers 100 pounds of the

aluminum alloy and 80 pounds of the steel
alloy weekly.
26
Example 2: Olympic Bike Co.

The number of pounds of each alloy needed

per frame is summarized below:
Aluminum Alloy Steel Alloy
Deluxe 2 3
Professional 4 2

How many Deluxe and Professional frames

should Olympic produce each week?

27
Example 2: Model Formulation
p Decision Variables
x1 = Number of Deluxe frames per week

x2 = Number of Professional frames per week

(Total Weekly Profit) Max 10x1 + 15x2

s.t.
(Aluminum Available) 2x1 + 4x2 < 100
(Steel Available) 3x1 + 2x2 < 80
x1, x2 > 0

28
 Example 2: Olympic Bike Co.
p Sensitivity Report
Adjustable Cells
Final Reduced Objective Allowable Allowable
Cell Name Value Cost Coefficient Increase Decrease
$C$11 Bikes made Deluxe 15 0 10 12.5 2.5
$D$11 Bikes made Professional 17.5 0 15 5 8.333333333

Constraints
Final Shadow Constraint Allowable Allowable
Cell Name Value Price R.H. Side Increase Decrease
$C$16 Aluminum Amount Used 100 3.125 100 60 46.66666667
$C$17 Steel Amount Used 80 1.25 80 70 30

29
Example 2: Olympic Bike Co.
Adjustable Cells
Final Reduced Objective Allowable Allowable
Cell Name Value Cost Coefficient Increase Decrease
$C$11 Bikes made Deluxe 15 0 10 12.5 2.5
$D$11 Bikes made Professional 17.5 0 15 5 8.333333333

Constraints
Final Shadow Constraint Allowable Allowable
Cell Name Value Price R.H. Side Increase Decrease
$C$16 Aluminum Amount Used 100 3.125 100 60 46.66666667
$C$17 Steel Amount Used 80 1.25 80 70 30

p Range of Optimality
Question 1: Suppose the profit on deluxe frames is
increased to $20. Is the above solution still optimal?
What is the value of the objective function when this unit
profit is increased to $20?

30
 Example 2: Olympic Bike Co.
Adjustable Cells
Final Reduced Objective Allowable Allowable
Cell Name Value Cost Coefficient Increase Decrease
$C$11 Bikes made Deluxe 15 0 10 12.5 2.5
$D$11 Bikes made Professional 17.5 0 15 5 8.333333333

Constraints
Final Shadow Constraint Allowable Allowable
Cell Name Value Price R.H. Side Increase Decrease
$C$16 Aluminum Amount Used 100 3.125 100 60 46.66666667
$C$17 Steel Amount Used 80 1.25 80 70 30

p Range of Optimality
Question 2: If the unit profit on deluxe
frames were $6 instead of $10, would the
optimal solution change? 31
 Example 2: Olympic Bike Co.
Adjustable Cells
Final Reduced Objective Allowable Allowable
Cell Name Value Cost Coefficient Increase Decrease
$C$11 Bikes made Deluxe 15 0 10 12.5 2.5
$D$11 Bikes made Professional 17.5 0 15 5 8.333333333

Constraints
Final Shadow Constraint Allowable Allowable
Cell Name Value Price R.H. Side Increase Decrease
$C$16 Aluminum Amount Used 100 3.125 100 60 46.66666667
$C$17 Steel Amount Used 80 1.25 80 70 30

p Shadow Price and Range of Feasibility

Question 3: The supplier has capacity to deliver 50 extra
pounds of steel alloy. What is the maximum amount the
company should pay for each extra pound of the alloy?

What if 100 extra pounds are available? 32

 Example 2: Olympic Bike Co.
Adjustable Cells
Final Reduced Objective Allowable Allowable
Cell Name Value Cost Coefficient Increase Decrease
$C$11 Bikes made Deluxe 15 0 10 12.5 2.5
$D$11 Bikes made Professional 17.5 0 15 5 8.333333333

Constraints
Final Shadow Constraint Allowable Allowable
Cell Name Value Price R.H. Side Increase Decrease
$C$16 Aluminum Amount Used 100 3.125 100 60 46.66666667
$C$17 Steel Amount Used 80 1.25 80 70 30

p Shadow Price and Range of Feasibility

Question 4: Another manufacturer, EZ-Bikes, wants to
purchase 40 pounds of steel alloy for its own weekly
production and is willing to pay Olympic Bike $35 for it.

Should Olympic Bike take this offer and sell EZ-Bikes 40 lbs
of the 80 lbs of steel alloy it receives each week? 33
 Example 2: Olympic Bike Co.
Adjustable Cells
Final Reduced Objective Allowable Allowable
Cell Name Value Cost Coefficient Increase Decrease
$C$11 Bikes made Deluxe 15 0 10 12.5 2.5
$D$11 Bikes made Professional 17.5 0 15 5 8.333333333

Constraints
Final Shadow Constraint Allowable Allowable
Cell Name Value Price R.H. Side Increase Decrease
$C$16 Aluminum Amount Used 100 3.125 100 60 46.66666667
$C$17 Steel Amount Used 80 1.25 80 70 30
p Modifying the model:
Question 5: Olympic Bike’s Marketing Department warns that the
Deluxe model received more demand than the Professional model
in their market research. The management is considering to set
up a policy to produce at least as many Deluxe bikes as
Professional bikes.
With this policy,
n Will the optimal solution be different?
n How will the total profit be affected? 34
Example 2: Olympic Bike Co.
p Modifying the Model (Sunk Cost vs. Relevant
Cost)
Question 6: Olympic makes an agreement
with the supplier so that it will pay the
supplier only for the amount of alloy it uses
(up to the available amounts).

If the supplier charges $2 per pound of

aluminum alloy, and $1 per pound of steel
alloy, how would you change your model?
35
Reduced Cost
Given an optimal solution,

p if a decision variable has value 0, then its

reduced cost is the amount the variable's
objective function coefficient would have to
improve (increase for maximization problems,
decrease for minimization problems) before
this variable could assume a positive value.

p if a decision variable has a positive value,

then its reduced cost is 0.
36
Example 3: Make vs. Buy Decisions
p Electro-Poly, a leading manufacturer of
slip-rings, received an order for three of
its models. Model 1 Model 2 Model 3
Number ordered 3,000 2,000 900
Hours of wiring/unit 2 1.5 3
Hours of harnessing/unit 1 2 1
Cost to Make $50 $83 $130
Cost to Buy $61 $97 $145
• The company has 10,000 hours in wiring capacity and
5,000 hours of harnessing capacity
37
Example 3: The Decision Variables

Mi = Number of model i slip rings to make in-

house
Bi = Number of model i slip rings to buy from
competitor
for i = 1,2,3

38
Example 3: Formulation

Min 50 M 1 + 61B1 + 83M 2 + 97 B2 + 130M 3 + 145B3

s.t. M 1 + B1 ³ 3,000
M 2 + B2 ³ 2,000
M 3 + B3 ³ 900
2M 1 + 1.5M 2 + 3M 3 £ 10,000
M 1 + 2M 2 + M 3 £ 5,000
M 1 , B1 , M 2 , B2 , M 3 , B3 ³ 0
39
Example 3: Sensitivity Analysis Report

p At what cost would the company be willing to

buy any of Model 1?
p At what cost would the company make fewer
units of Model 1?
40
Example 3: Make vs. Buy (contd.)

p If Electro-Poly receives an order of one more

unit of Model 1, what is the additional cost to
Electro-Poly? 41
Example 3: Make vs. Buy (contd.)

p What if the company had only

n 9600 hours of wiring available?
n 9000 hours of wiring available?
42
 Example 3: Make vs. Buy (contd.)

p What would the optimal objective value be if the company had

n 100 extra hours of harnessing?
n 1000 extra hours of harnessing?
n 100 less hours of harnessing?
43
n 1200 less hours of harnessing?
 Example 3: Make vs. Buy (contd.)

p The company can increase harnessing time by scheduling overtime at a

cost of $10 per hour.
If overtime is available up to 200 hours, how much overtime would the
company use?
p What if the cost of overtime is $6 per hour?
44
Modify your formulation to find how many hours of overtime will be used.
Notes
1. The feasible region of an LP is determined
only by the constraints in the LP
formulation.
2. Changes in the objective function
coefficients cannot affect the feasible
region. However, a new corner point in the
feasible region may become optimal.

1. The feasible region of an LP may be

modified if
n a constraint in the formulation is modified,
n a new constraint is added to the formulation, and
n a constraint is removed from the formulation. 45