Professional Documents
Culture Documents
05 Sensitivity Analysis
05 Sensitivity Analysis
05 Sensitivity Analysis
1
Agenda
Wyndor Glass Co. Revisited
Perturbing input parameters: Solver Sensitivity Report
Shadow prices: Economic interpretation
Range of validity of shadow prices
Simultaneous perturbation of multiple parameters: The 100% Rule
Reduced costs
Effect of perturbing the coefficients of the objective function
Templates: Your completed model for the Wyndor Glass Company problem.
2
Recap: Wyndor Glass Company
The Wyndor Glass Company produces high-quality glass products,
including windows and glass doors that feature handcrafting and the finest
workmanship.
Although the products are expensive, they fill a market niche by providing
the highest quality available in the industry for the most discriminating
buyer.
Adapted from Frederick S. Hillier and Mark S. Hillier, Introduction to Management Science: A Modeling
and Case Studies Approach with Spreadsheets, 5/e, p 23-25.
3
Recap: Wyndor Glass Company
Because of declining sales for certain products, top management has
decided to revamp the company’s product line.
4
Recap: Wyndor Glass Company
It is estimated that each door will require one hour of production time in
Plant 1 and three hours in Plant 3.
For each window, about two hours will be needed in Plant 2 and two hours
in Plant 3.
The production facilities in Plant 1 needed for the 8-foot glass doors will
be available approximately 40 hours per week. (The rest of the time Plant 1
will continue with current products.)
5
Recap: Wyndor Glass Company
6
Recap: Mathematical Formulation
90
180
7
Recap: An Optimal Solution 8
8
Generating Answer Report
Answer Report
OFV: 300×30+500×45
10
Wyndor Co.: Effect of Changing the RHS
• Let’s make one-at-a-time changes to the available hours and make three
more Solver runs (one for each change).
11
Generating Sensitivity Report
12
Sensitivity Report
13
Sensitivity Report: Constraint Status
• Constraint R.H. Side refers to the original constant in the RHS of the
constraint.
• Final Value refers to the total value of the LHS when the variables take the
optimal values.
• Note: To have a clean interpretation of the Sensitivity Report, it is critical
to have only constants in the RHS and all the variables in the LHS.
14
Sensitivity Report: Shadow Prices
Changing the level of a resource
Shadow prices: They reflect the marginal instantaneous increment
(decrement) in the objective value Z if the corresponding resource is
marginally increased (decreased).
15
Shadow Price
Let’s make a change by more than one unit (30 units) for Plant 2.
What’s wrong?
16
Allowable Increase and Decrease
The shadow price is valid only
– for its allowable increase and allowable decrease, AND
– for changing only one constraint at a time
17
Sensitivity Report: Changing a Resource Level
Changing the level of a resource: Binding constraint
Example: For Plant 2 Time RHS, the reference value is 90. From here,
• Increase: RHS could be increased up to 90+90=180.
• Decrease: RHS could be decreased down to 90−30=60.
18
Sensitivity Report: Changing a Resource Level
Changing the level of a resource: Non-binding constraint
Interpretation for the range: By how much we could perturb the RHS
without changing the optimal corner point
Example: For Plant 1 Time RHS, the reference value is 40. From here,
• Increase: RHS could be increased up to 40+∞=∞
• Decrease: : RHS could be decreased down to 40−10=30
19
A Note about Shadow Prices
• The shadow price anticipates the effect of changing the RHS of a
constraint in the OFV but not in the decision variables.
– For instance, increasing the RHS of Plant 2 from 90 to 180 will imply
an increase in the OFV of 90×150=13,500, from 31,500 to 45,000
– In order to get the new optimal solution, we need to resolve
20
Shadow Prices: Decreasing Marginal Returns
• Scarce resource (i.e., constraint is binding): below the “Allowable
Decrease”, the resource is more valuable than the original shadow price
– In Wyndor, Plant 2 Time is binding. Reducing the RHS below 90-
30=60 will increase (in fact, non decrease) the corresponding shadow
price. For instance, when RHS=59, the sensitivity report gives:
• Analogously, for a scarce resource, when the availability goes above the
“Allowable Increase”, the resource is less valuable than the original
shadow price
21
Simultaneous Changes in the RHS
• When using the shadow price in calculations we assume that only one
constraint RHS is changed.
• If we change two or more constraint RHS then we must use the
“100% rule”.
• The shadow prices are valid (i.e., tell the actual OFV change) as long as
the percentage increase/decrease used up does not exceed 100% of the
allowable ranges.
22
Simultaneous Changes in the RHS
• When we simultaneously increase & decrease the RHS of several
constraints, the total cumulative change (in absolute terms) has to
be calculated.
- The 100% rule applies on this grand total
23
Simultaneous Changes in the RHS
• 100% Rule: The shadow prices are valid (i.e., tell the actual
change) as long as the percentage increase/decrease used up does
not exceed 100%.
• Question: What if we go beyond 100%?
• When the 100% rule does not hold, we cannot anticipate with certainty the
impact of the changes (i.e., shadow prices may or may not be valid). We need
to resolve!
24
Other Uses of Shadow Prices
Suppose that Wyndor Glass Co. evaluates adding a new Product #3 to its
mix:
25
Other Uses of Shadow Prices
New model setup:
26
Other Uses of Shadow Prices
• The production of each unit of Product 3 would imply the
consumption of some of the resources devoted to the other
products:
– The consumption of 1 hour at Plant 2 would reduce profits by $150
– The consumption of 2 hours at Plant 3 would reduce profits by
2×$100=$200
• In total, the profit would be reduced by $350.
28
Sensitivity Report: Reduced Costs
29
Reduced Cost: Interpretations
One common way to interpret a reduced cost:
– When there are bound(s) on the variables, and the bounds are
binding at optimality, the reduced cost of a variable is the shadow
price on the binding bound constraint of that variable.
30
Reduced Cost: Bounds on the Dec. Vbles.
• New Constraint: Assume that we require that no more than 15 doors and 45
windows can be sold in a week.
31
Reduced Cost: Bounds on the Dec. Vbles.
• New Constraint: Assume that we require that no more than 15 doors and 45
windows can be sold in a week.
When there are bounds on the variables, the reduced cost is the shadow
price on the binding bound constraints.
32
“Anomalies” in the Sensitivity Report
Degeneracy
1. The shadow prices and their ranges may still be interpreted in the usual
way, but they may not be unique.
A different set of shadow prices and ranges may also apply to the problem
(even if the optimal solution is unique)
33
Degenerate Solution: Example
Max Z = 300 x1 + 500 x2
Subject to:
1 x1 40
2 x2 90
3 x1 + 2 x2 180
3 x1 + 3 x2 225 New constraint
x1, x2 0.
Note: Constraint is
binding, but shadow price
is zero
34
Objective Coefficient Sensitivity
• Solver provides useful economic information about objective
coefficients.
35
Sensitivity Report: Objective Function
Changing the coefficients in the objective function
38
“Anomalies” in the Sensitivity Report
Alternate Optimal Solutions
When the allowable increase or decrease of the objective function coefficient
for one or more variables is zero and the solution is non-degenerate, there exists
an alternate optimal solution.
Allowable increase = 0
Non-degenerate
39
Alternate Solutions: How to Find Them?
Procedure
1. Add a constraint to the model that holds the objective function at the
current optimal value.
Max Z = x2
Subject to:
1 x1 40
2 x2 90
3 x1 + 2 x2 180
300 x1 + 200 x2 = 18000 New constraint
x1, x2 0.
40
Alternate Solutions: How to Find Them?
Procedure
41
Summary
• The sensitivity analysis corresponds to the impact of perturbing some of the
input parameters of the model
‒ Focus is on the RHS of the constraints, and on the coefficients of the
objective function
‒ Considers one-at-a-time perturbations
‒ Extension with the “100% rule” for simultaneous changes
• Anomalies
‒ Degeneracy & alternate optimal solutions
42
An (Artificial) Problem 43