0% found this document useful (0 votes)
125 views55 pagesCh8.Quantitative Methods For Business David R. Anderson
Chapter 8 discusses sensitivity analysis in linear programming, focusing on how changes in coefficients affect optimal solutions. It highlights the importance of this analysis for decision-making in dynamic environments, using Chevron's refinery operations as a case study to illustrate practical applications. The chapter also covers the use of tools like Excel Solver for conducting sensitivity analysis without needing to reformulate the entire linear programming problem.
Uploaded by
Sanwicklith MkalingaCopyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF or read online on Scribd
0% found this document useful (0 votes)
125 views55 pagesCh8.Quantitative Methods For Business David R. Anderson
Chapter 8 discusses sensitivity analysis in linear programming, focusing on how changes in coefficients affect optimal solutions. It highlights the importance of this analysis for decision-making in dynamic environments, using Chevron's refinery operations as a case study to illustrate practical applications. The chapter also covers the use of tools like Excel Solver for conducting sensitivity analysis without needing to reformulate the entire linear programming problem.
Uploaded by
Sanwicklith MkalingaCopyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF or read online on Scribd
/ 55
CHAPTER 8
Linear Program
ming:
Sensitivity Analysis and
Interpretation of Solution
CONTENTS
&1 INTRODUCTION TO
SENSITIVITY ANALYSIS
82 OBJECTIVE FUNCTION
(COEFFICIENTS
83. RIGHT-HAND SIDES
(Cautionary Note on the
Interpretation of Shadow
Prices
84 LIMITATIONS OF CLASSICAL,
SENSTIVITY ANALYSIS
Simultaneous Changes
Changes in Constraint
Coefficients
Nonintuitive Shadow Prices
85 MORE THAN TWO DECISION
VARIABLES
Modified RMC Problem
Bluegrass Farms Problem
86 ELECTRONIC
COMMUNICATIONS
PROBLEM
Problem Formulation
Solution and Interpretation
APPENDIX 8.1
‘SENSITIVITY ANALYSIS
WITH EXCEL SOLVER
APPENDIX 8.2
‘SENSITIVITY ANALYSIS.
WITH LINGO Chapter 8. Linear Programming: Sensitivity Analysis end Interpretation of Solution 309
Sensitivity analysis is the study of ow changes inthe coefficients of a linear programming
problem affect the optimal solution. Using sensitivity analysis, we can answer questions
such as the following:
1. How will a change in an objective function coefficient affect the optimal solution?
2. How will a change in 4 right-hand-side value for a constraint affect the optimal
sohation?
Because sensitivity analysis fs concerned with how these changes affect the optimal solu-
tion, sensitivity analysis does not begin until the optimal solution to the original linear
programming problem has been obtained. Forthis reason, sensitivity analysis i sometimes
referred to as postoprimality analysis.
‘Our approach to sensitivity analysis parallels the approach used te introduce linear pro-
‘gramming in Chapter 7. We introduce sensitivity analysis by using the graphical method for
a linear programming problem with two decision variables, Then, we show how Excel Solver
cean be used to provide more complete sensitivity analysis information. Finally, we extend the
discussion of problem formulation started in Chapter 7 by formulating and solving three larger
‘near programming problems. In discussing the solution for each ofthese problems, we focus
‘on manageria interpretation of the optimal solution and sensitivity analysis information.
Sensitivity analysisand the interpretation of the optima solution are important aspects
of applying linear programming. The QM. in Action, Optimizing Refinery Operations
at Chevron Using Linear Programming, explains how Chevron uses linear programing
based tools and sensitivity analysis to improve thei il refinery operations. Later in the
chapter other Q.M. in Action features illustrate how Performance Analysis Corporation
uses sensitivity analysis as part of an evaluation model for « chain of fast-food outlets,
how GE Plastics (now part of the Saudi Basic Industries Corporation) uses a lineat pro-
‘gramming model involving thousands of variables and constraints to determine optimal
production quantities, how Kimpton Hotels uses linear program to set prices and room,
availability on Priceline, and how Duncan Industries Limited’ linear programsning model
for tea distribution convinced management of the benefits of using quantitative analysis,
techniques to support the decision-making process.
OPTIMIZING REFINERY OPERATIONS AT CHEVRON USING LINEAR PROGRAMMING —
‘Chevron is worldwite eneray company with operations
‘in more than 180 counties, Chevron has business units
‘hat operate in simost al aspects of energy exploration
and production, including oll end gas, geothermal, soar,
‘wind, and others. ChevTon’s ol and as bsiness requires
‘refineries to fake input crude oil and transform the crag
«il into products such as gasoline and lubricants that are
sold to downstream operations. The processes required
1 transform crude oil to useable prosiuts ace complex
Boren Kat, Movs rN Keaton Strom ood
1M Coma, “Opialng Choon Rnrax” es 4,9.)
Uonwanebar 2014), 39-54
‘Chevron has developed a linear programming-based
tool to assist with many of is tactical and strategic
sions related to its refining operations. The decisions that
are made based on the linear programming tool output
include which erude oils to buy for input, which prod-
Luts to manufactore as output, tnd how to best operate
the refinery. There are many cettplicating factors in these
‘decisions, such asthe fact that rude ols purchased from
different areas all have slightly different characterises
(continued) 310
that require differing refining operations, which therefore
‘makes them best suited for different output products, Fur-
ther, prices for output products re often volatile and mast
be considered when deciding which products to refine,
‘Chevron frst stated using linear programs to as-
‘ist with its decision making in refinery operations in
the 1950s, Trough the following decades, the company
‘continued to update and expand its linear prograrn-based
‘twols to include additional complexities and exploit ad-
‘vances in computing power. Curent, Chevron uses a
linear progran-based tool at each of ts seven refineries,
‘The tool includes a spreadsheet interface tallow for ease
‘of use in a familiar software environment,
One of the biggest benefits to the linear program
‘based tools is that they allow Chevron to consider a
Chapler 8 Linear Programming: Sensitivity Analysis and lnerprtaton of Solution
mukitude of possible scenarios through sensitivity analy
sis. Because prices fr gasoline, lubricants, and other output
products change offen, Chevron can use sensitivity analy-
sis to evaluate the effects of changes on output product
prices to its refinery decisions on which input crude oil to
‘purchase and which outpt product to proces,
‘Chevron believes thatthe results ofits quantitative
‘methods-based tools, including the inear program mod-
‘ls, add more than $1 billion per year in overt benefits.
‘In particular, Chevron estimates that asing their tools to
‘optimize their choice of crude ol inputs and choosing the
‘best set of output predacts to refine generates more than
‘$600 million per year in adltional earnings and nother
‘$400 million per year of benefits through more efficient
use of capital.
(1)
(81) Introduction to Sensitivity Analysis
Sensitivity analysis is important to decision makers because real-world problems exist in a
‘changing environment. Prices of raw materials change, product demands change, produc-
tion capacities change, stock prices change, and so on. Ifa linear programming model has
been used in such an environment, we can expect some of the coefficients in the model
to change over time. As a result, we will want to determine how these changes affect the
‘optimal solution. Sensitivity analysis provides information needed to respond 0 such
changes without requiring a complete solution of a revised linear program.
Recall the RMC problem introduced in Chapter 7. RMC wanted to determine the num-
ber of tons of fuel additive (F) and the number of tons of solvent base (S) to produce in
order to maximize the total profit contribution for the two products. Three raw material
constraints limit the amounts of the two products that can be produced. The RMC linear
programming model is restated here:
Max 40F + 305
st
O4F +05 < 20 Material 1
02 5 Material 2
0.6F + 038 < 21 Material 3
F.S=0
‘The optimal solution, F = 25 tons and $= 20 tons, provided a maximum profit contribu:
tion of $1600.
‘The optimal solution was based on profit contributions of $40 per ton for he fuel add-
tive and $30 per ton forthe solvent base. However, suppose that we late learn that a price
reduction causes the profit contribution fr the fuel additive to fll from $40 to $30 per ton.
Sensitivity analysis can be used to determine whether producing 25 tons of fue alive and
20 tons of solvent basis still best. If its, solVing a modified linear programming problem
with 30F + 308 as the new objective function is not necessary. 8.2 Objective Function Coaficients aun
Sensitivity analysis can also be used to determine which coefficients in a linear pro-
gramming model are crucial. For example, suppose that management believes that the
$830 per ton profit contribution forthe solvent base is only a rough estimate ofthe profit contri-
bation that will actually be obtained. If sensitivity analysis shows that 5 tons of fuel additive
‘and 20 ons of solvent base willbe the optimal solution as long as the profit contribution forthe
solvent base is between $20 and $50, management should feel comfortable with the $30 per ton
estimate and the recommended production quantities. However. if sensitivity analysis shows
that 25 tons of fuel additive an 20 tons of solvent base wil be the optimal solution only ifthe
profit contribution forthe solvent tase is between $29.90 and $30.20 per ton, management
‘may want to review the accuracy ofthe $20 per ton estimate
‘Another aspect of sensitivity analysis concems changes in the right-hand-side values
of the constraints. Recall that in the RMC problem the optimal solution used all available
‘material 1 and material 3. What would happen tothe optimal solution and total profit con-
tribution if RMC could obtain additional quantities of either of these resources? Sensitivity
analysis can help determine how much each added ton of material is worth and how many
tons can be added before diminishing retums set in
Objective Function Coefficients
Let us begin sensitivity analysis by using the graphical solution procedure to demonstrate
how a change in an objective function coefficient can affect the optimal solution to a linear
‘programming problem. We begin with the graphical solution tothe original RMC problem
shown in Figure 8.1. The feasible region is shaded. The objective function 40F + 30S takes
on its maximum value at the extreme point F = 25 and § = 20, Thus, F = 25 and $ = 20
is the optimal solution and 40(25) + 30(20) = 1600 is the value of the optimal solution
FIGURE 8.1. OPTIMAL SOLUTION TO THE ORIGINAL RMC PROBLEM
o a ee ee nc
“Tons of Fuel Additive 312
Chapter 8 Linear Programming: Sensitivity Analysis and Interpretation of Solution
Now suppose RMC learns that a price reduction in the fuel additive has reduced its
profit contribution to $30 per ton. With this reduction, RMC’s management may question
the desirability of maintaining the original optimal solution of F = 25 tons and $ = 20 tons.
Perhaps different solution is now optimal. The RMC linear program with the revised
objective function is as follows:
Max 30F + 30S
st
O4F + 0.55520 Material 1
0.25< 5 Material 2
O.6F +0.3521 Material 3
FS=0
‘Note that only the objective function has changed. Because the constraints have not
‘changed, the feasible region for the revised RMC problem remains the same as the original
problem. The graphical solution tothe RMC problem with the objective function 30F + 305
is shown in Figure 8.2. Note that the extreme point providing the optimal solution is still
F = 25 and S = 20. Thus, although the total profit contribution decreased to 30(25) +
30(20) = 1350, the decrease in the profit contribution for the fuel additive from $40 per ton
to $30 per ton does not change the optimal solution F = 25 and $ = 20.
Now let us suppose that a further price reduction causes the profit contribution for the
fuel additive to be reduced to $20 per ton, Is F = 25 and S = 20 still the optimal solution?
Figure 8.3 shows the graphical solution to the RMC problem with the objective function
revised to 20F + 30S, The extreme point providing the optimal solution is now F = 18.75
FIGURE 8.2 REVISED OPTIMAL SOLUTION WITH THE RMC OBJECTIVE FUNCTION
30F + 308
«
i
Gx
: Q ima Stton
3 Pans sa20
P
10
9 i 0 3 «0 50
‘Tons of Fuel Additive 8.2 Objective Function Coefficients 313
FIGURE 8.3 REVISED OPTIMAL SOLUTION WITH THE RMC OBJECTIVE FUNCTION.
20F +308
° a a a ar)
“Tons of Fuel Adlitive
The graphical solution is
used here to help the reader
Visualize how changes 10
‘an objective function
coaficien may or may
rot change the optimal
lution.
Computer solutions
peal provide sensitivity
analysis information.
and $ = 25. The total profit contribution decreased to 20(18.75) + 30(25) = 1125. How-
cever, inthis case, we see that decreasing the profit contribution for the fuel additive to
{$20 per ton changes the optimal solution. The solution F = 25 tons and § = 20 tons is no
longer optimal. The solution F = 18.75 and $ = 25 now provides the optimal production
quantities for RMC.
‘What do we learn from the graphical solutions in Figures 8.1, 8.2, and 8.3? Chang-
ing one objective function coefficient changes the slope of the objective function line but
leaves the feasible region unchanged. Ifthe change in the objective function coefficient is
small the extreme point that provided the optimal solution tothe original problem may still
provide the optimal solution. However, if the change in the objective function coefficient is
large enough, a different extreme point will provide a new optimal solution,
Fortunately, computer packages such as Excel can easily provide sensitivity analysis
information about the objective function coefficients forthe original RMC linear program-
‘ming problem. You do not have to reformulate and re-solve the linear programming prob-
Jem to obtain the sensitivity analysis information. Appendix 8.1 explains how to generate a
‘Sensitivity Report using Excel Solver. A sensitivity report similar to the output provided by
Excel for the original RMC linear programming problem is shown in Figure 8.4. In addition
toall of the Excel Solver Sensitivity Repor information, the report in Figure 8.4 also shows
the variables we used in our model. This allows you to easily link the sensitivity report to the
‘model under discussion. We shall use this style of sensitivity report throughout this chapter.
In the Variable Cells section of the sensitivity repor, the column labeled Final Value
contains the optimal values of the decision variables. For the RMC problem the optimal
solution is to produce 25 tons of fuel additive and 20 tons of solvent base, Associated with
each decision variable is a reduced cost, We will discuss reduced costs in more detail after
{introducing the concept of shadow prices later in this chapter. 34
a file]
CChopler 8. Linear Programming: Sensitviy Anclysis and inerpreation of Solin
FIGURE 8.4. SENSITIVITY REPORT FOR THE RMC PROBLEM
Variable
“Model ———~*~*~<—~*~*:*~*”:SSCSnad~—«sRedduced Od jective Allowable Allowable
Variable Name Value Cost _Coefficient Increase Decrease
'F__ Tons Produced Fuel Additive 25.000 0.000 40.000 20.000 16.000.
‘S__ Tons Produced Solvent Base 20.000 0.000 30.000 20.000 10.000,
CConstesits
‘Constraint Final Shadow — Constraint Al
Number __Name Value Price _R.H. Side
1 Material | Amount Used 20.000 33.333 20.000 500 6.000,
2 Material? Amount Used 4.000 0.000.000 TE+30 1.000
‘3 Material 3 Amount Used 21.000 44.444 21.000 9.000 2.250
To the right of the Reduced Cost column in Figure 8.4, we find three columns labeled
Objective Coefficient, Allowable Increase, and Allowable Decrease. For example, the ob-
jective function coefficient for the fuel additive is $40, with an allowable increase of $20
and an allowable decrease of $16. Therefore, as long as the profit contribution associated
with fuel additive is between $40 + $20 = $60 and $40 — $16 = $24, the optimal solution
‘of 25 tons of fuel additive and 20 tons of solvent base will not change. The value of $20 is
often referred to asthe objective function coefficient allowable increase, and the value of
$16 isthe objective function coefficient allowable decrease. The range between $24 and
‘S60 is referred to asthe objective coefficient rangeor range of optimality fr the fuel addi-
tive variable. If the profit contribution forthe fuel additive is outside this range, a different
extreme point and a different solution will become optimal
‘The objective function coefficient for the solvent base variable is $30. The allowable
decrease of $10 and allowable increase of $20 for the solvent base variable show that the
‘optimal solution will not change so long as the profit contribution for solvent base is be-
tween $30 + $20 = $50 and $30 ~ $10 = $20.
Burkey
1, The sensitivity analysis information provided for ofthe original problem remain unchanged. Ths,
the objective function coefficients is based onthe an objective coefficient range is oly applicable
assumption that only one objective function coef-__ for changes toa single objective coefficient. We
{cient changesata time andthatallother aspects examine this isue in more dep in Section 84
(#3) Right-Hand Sides
Let us expand the discussion of sensitivity analysis by considering how a change in the
‘right-hand side of a constraint affects the feasible region and the optimal solution to linear
programming problem. As with sensitivity analysis forthe objective function coefficients,
‘we consider what happens when we make one change at a time. For example, suppose that
in the RMC problem an additional 4.5 tons of material 3 becomes available. In this case, Sense analysis for
right-hand sides is based
‘on the assumption hat
‘only one righthand side
changes ara tine All ther
‘spect ofthe problem are
‘sumed to bea tated ix
the original problem.
Shadow prices often
‘provide the economic
Information tha helps make
decisions about acauring
‘aational resource,
8.3. RightHond Sides a5
the right-hand side of the third constraint increases from 21 tons to 25.5 tons. The revised
RMC linear programming model is as follows:
Max 40F + 305
st
O4F +0520 Material 1
028s 5 Material 2
0.6F + 035255 Material 3
FS=0
‘The graphical solution to this problem is shown in Figure 8.5. Note how the feasible
region expands because ofthe additional 4.5 tons of material 3. Application ofthe graphical
solution procedure shows thatthe extreme point F = 37.5 tons and $= 10 tons is the new
‘optimal solution. The value of the optimal solution is 40(37.5) + 30(10) = $1800. Recall
thatthe optimal solution tothe original RMC problem was F = 25 tons and $ = 20 tons and
the value of the optimal solution was $1600. Thus, the additional 4.5 tons of material 3 in
the revised problem provides a new optimal solution and increases the value of the optimal
solution by $1800 ~ $1600 = $200. On a per-ton basis, the additional 4.5 tons of material
3 increases the value of the optimal solution a te rate of $200/4.5 = $44.44 per ton.
The shadow priceis the change in the optimal objective function value per unit increase
the right-hand side of a constraint. Hence, the shadow price for the material 3 constraint
is $44.44 per ton. In other words, if we increase che right-hand side of the material 3 con-
straint by 1 ton, the value ofthe optimal solution will increase by $44.44, Conversely, if we
decrease the right-hand side of the material 3 constraint by 1 ton, the value ofthe optimal
solution will decrease by $44.4
FIGURE 8.5 GRAPHICAL SOLUTION TO THE RMC PROBLEM WITH MATERIAL
3 CONSTRAINT 0.6F + 0.58 = 24.5
s
so 4
we
40
g
a [New Feasible Region
i” / Moeludes This Area
3
i”
Original
Feasible
ie Region
F
o a a a)
‘Tons of Fuel Adaltive 316
Computer solutions
‘npally provide the
‘shadow price for each
CChopler 8 Lineor Programming: Sensitivity Analysis end Interpretation of Solution
Fortunately, the sensitivity report for the original linear programming problem provides
the shadow prices for all the constrains. You do not have to reformulate and re-solve the
linear programming problem to obtain the shadow price information. The sensitivity report
for the original RMC linear programming problem is shown in Figure 84
Examine the Constraints section of the sensitivity report. The entries in the Final Value
column indiate the number of tons of each material used in the optimal solution. Thus,
RMC will use 20 tons of material 1, 4 tons of material 2, and 21 tons of material 3 in order
to produce the optimal solution af 25 tons of fuel additive and 20 tons of solvent base.
‘The Values in the Constraint R.H. Side column are the right-hand sides of the constraints
for the RMC problem. The differences between the entries in the Constraint RLH. Side
column and the Final Value column provide the values of the slack variables for the RMC
problem. Thus, there are 20 — 20 = 0 tons of stack for material 1, 5 — 4= 1 ton of slack
for material 2, and 21 ~ 21 = 0 tons of slack for material 3
The column labeled Shadow Price provides the following information:
Constraint Shadow Price
Material | Amount Used $33.33
Material 2 Amount Used $ 0.00
Material 3 Amount Used S444
[Note thatthe shadow price for material 3, $44.44 per ton, agrees with the calculations we
made using the graphical solution procedure. We also observe that the shadow price for
the material 1 constant indicates thatthe value of the optim solution will nerease at the
rate of $33.33 per ton of material 1. Finally, note that the shadow price forthe material 2
constraint is $0.00. The optimal solution to the RMC problem shows that material 2 has a
slack of I ton. Thus, a the optimal solution, 1 ton sf material 2 unused. The shadow price
‘of $0.00 tells us that additional tons of material 2 will simply add to the amount of slack for
constraint 2 and will not change the value of the optimal solution.
‘We caution here that the value of a shadow price may be applicable only for small
increases in the right-hand side. As more and more resources are obtained and as the right-
hhand side continues to increase, other constraints will become binding and limit the change
inthe value of the optimal solution. At some point, the shadow price ean no longer be used
to determine the improvement in the value of the optimal solution,
Now that we have introduced the concept of shadow prices, we can define the reduced
cost associated with each variable. The reduced cost asociated with a variable is equal
to the shadow price for the nonnegativity constraint associated with the variable.’ From
Figure 84 we see thatthe reduced cost for both variables are zero, This makes sense. Con-
sider fuel additive. The nonegativity constraint associated with the fuel additive variable,
F is F = 0, so changing the nonnegatvity constraint to F = 1 has no effect on the optimal
solution value. Because increasing the right-hand side by one unit has no effect on the opt-
‘al objective function value, the shadow price (ie, reduced cost) of this nonnegativity con-
straint is zero. A similar argument applic othe solveat base variable, 5. Later we introduce
4 modified RMC problem that has a nonzero reduced costo beter explain tis concept
‘The last two columns in the Constraints section ofthe sensitivity report contain the
right-hand side allowable increase and allowable deerease for e2ch constraint. For
example, consider the material 1 constraint with an allowable increase value of 1.5 and an
allowable decrease value of 6. The values in the Allowable Increase and Allowable Decrease
"We lo noe ht the value of varolein en optimal oon sequal othe ver bound ofthe vail, han the
redveed ot lb the shadow price of his upper ound conse. The range of feasbiry is
also somexies referred
twas the right-hand-side
Similar to sensitivity
‘analysis for objective
Function coefficients, the
seit analysis for
Fighthand sider of
that only one consaint
righthand side changes
8.3. RightHand Sides 317
columns indicate that the shadow price of $33.33 is applicable for increases up to 20+ 1.5 =
21.5 tons and decreases down to 20 ~ 6 = 14 tons. The values between 14 tons and
21.5 tons are often referred to as the range of feasibility for the material 1 constraint.
‘Note that unlike the objective function coefficient ranges, iis not tre hat the optimal
solution will not change if you stay within the range of feasibility. The range of feasibility
only implies thatthe same set of binding constraints will remain binding and hence that
the shadow price will accurately predict what will happen to the optimal objective function
value as the right-hand side is changed.
In summary, the range of feasibility information provides the limits where the shadow
prices are applicable. For changes outside the range, the problem must be re-solved to find
the new shadow price. Note thatthe sensitivity analysis information for right-hand sides of
constraints is only applicable for changes toa single right-hand side. If two or more right-
hand sides of constraints change atthe same time, itis easiest to re-solve the problem to see
the effect of these changes. This issue is discussed in more detail in Section 8.4.
‘The QM. in Action, Evaluating Eificiency at Performance Analysis Corporation,
illustrates the use of shadow prices as part of an evaluation model for a chain of fast-food
outlets. This type of model will be studied in more detail in the next chapter when we
diseuss an application referred to as data envelopment analysis.
EVALUATING EFFICIENCY AT PERFORMANCE ANALYSIS CORPORATZON*
Performance Analysis Corporation specializes in the
“se of quantitative models 1 design more efficient and
effective operations fora wide variety of chain stores.
‘One such application uses liseee programming. meth-
‘dology to provide an evalvation model for a chain of
fast-food outlets.
‘According to the concept of Pareto optimality, @
restaurant in a given chain is relatively inefficient if
other restaurants in the same chain exhibit the following
characteristic:
J. Operate in the same or worse environment
2, Produce at least the same level ofall outputs
3. Utilize no more of any resource and less of at
‘east one of the resources
‘To determine which of the restaurants are Pareto inefti-
cient, Performance Analysis Corporation developed and
solved @ near programming model. Model constraints
involve requirements concerning the minimam accept-
able levels of output and conditions imposed by uncon-
‘ollable elements in ce environment, and the objective
unetioe calls for the minimization of the resources
__ Fed on iefomation provided by Richard C. Mena of Frferonce
Andis Cerpation
necessary to produce the output. Solving the model pro-
aces the following output foreach restaurants
1. A score that assesses the level of so-called rela-
tive technical efficiency achieved by the partic-
‘lar restautant over the time period in question,
2. The reduction in controllable resourees or the
‘increase of outputs over the time period in ques-
tion nceded for an inefficient restaurant 19 be
‘ted as efficient,
3. A peee group of other restaurants with which
‘each restaurant can be compared in the futare,
Sensitivity analysis provides important managerial ie
formation, For example, for each constrainteoncering 2
‘minimum acceptable output vet, the shadow price tells
the manager how much one more unit of output would
increase te efficiency measur.
‘The analysis typically identifies 40% to SO% of the
restaurants as underperforming, given the previously
stated conditions conceming the inputs available and ou
puts produced. Performance Analysis Corporation finds
that if all the relative inefficiencies identified are elimi-
‘ated simultaneously, corporat profits typically increase
approximately 5% to 10%. This increase is truly substan-
tial given the large scale of operations involved. 318
CChopler 8 Linear Progremming: Sensitivity Analysis and Interpretation of Solution
Reiurks nnd
1. Some texts and computer programs use the term,
dual value or dual price instead of shadow price.
Often the meaning of these terms
to the definition given here for shadow price.
However, you must be careful to understand
‘identical exactly what is meant by the term being used.
(Only relevant cost should
be included in the objective
funcion.
Cautionary Note on the Interpretation of Shadow Prices
AAs stated previously, the shadow price is the change in the value of the optimal sotation
‘er unit increase inthe right-hand side of a constraint. When the right-hand side of the con-
straint represents the amount ofa resource available, the shadow price is often interpreted as
the maximum amount one should be willing to pay for one additional unit ofthe resoures.
However, such an interpretation is not always cortect. To see why, we need to understand
the difference between sunk and relevant costs. A stink cost is one that is not affected by
the decision made, It willbe incurred no matter What values the decision variables assume.
A relevant cost is one that depends on the decision made. The amount ofa relevant cost
will vary depending on the values ofthe decision variables.
Letus reconsider the RMC problem, The amount of material 1 available is 20 tons. The
cost of material 1 is a sunk cost if it must be paid regardless of the number of tons of fuel
Additive and solvent base produced. It would be a relevant cost if RMC only had to pay for
the number of tons of material 1 actually used to produce fuel additive and solvent base. All
relevant costs should be included in the objective function of a linear program. Sunk costs
should not be included in the objective function. For RMC we have been assuming thatthe
company has already paid for materials 1,2, and 3. Therefore, the cost of the raw materials
for RMC is a sunk cost and has not been inciuded inthe objective function.
‘When the cost of a resource is sunk, the shadow price can be interpreted as the maxi-
‘mum amount the company should be willing to pay for one additional unit of the resource.
When the cost of a resource used is relevant, the shadow price can be interpreted a the
amount by Which the value of the resouree exceeds its cost. Ths, when the resource cast
is relevant, the shadow price can be interpreted as the maximum premium over the normal
cost that the company should be willing to pay for one unit ofthe resource.
Prurkccncaed
1, Most computer software packages for solving
linear programs provide the optimal solution,
shadow price inforniation, the objective co-
efficient ranges, and the ranges of feasibility,
‘The labels used for these ranges may vary, but
the meaning is usually the same as what wehave
described here.
2. We defined the shadow price as the change in
the optimal objective function value per unit in-
crease in a right-hand side of a constraint. The
negative of the shadow price gives the change
in the optimal objective function value per unit
sectease inthe right-hand side
3, Whenever one of the right-hand sides is at an
endpoint of its range. the shadow price only
provides one-sided information. In this case,
the shadow price only predicts the change in
the optimal value of the objective function for
changes toward the interior of the range.
4, A condition called degeneracy can cause a subtle
difference in ha we interpret changes inthe ob-
{Jective Function coefficients heyond the endpoints
‘of te objective coeficient range. Degeneracy 0c-
ccurs when the shadow price equals zero for one
of the binding constraints. Degeneracy does not
affect the interpretation of changes toward the 8.4. Limitations of Classical Sensitivity Analysis
interior ofthe objective coefficient range. How-
‘ever, when degeneracy is present, changes beyond
the endpoints ofthe range do not necessarily mean
different soution will be optimal. From a practi-
‘al point of view, changes beyond the endpoints of
the range necessitate resolving the problem.
319
the new technology is developed, or purchased,
Jnorder to conserve resources. The shadow price
‘can be helpful in such cases because it cau be
used to determine the savings attributable to the
‘new technology by showing the savings per unit
of resource eouserved,
‘5. Managers are frequently called on to provide an
‘economic justification fornew technology. Often
Limitations of Classical Sensitivity Analysis
‘As we have seen, classical sensitivity analysis can provide useful information on the sen-
sitivity of the solution to changes in the model input data. However, classical sensitivity
analysis does have its imitations, In this section we discuss three such limitations: simul-
taneous changes in input data, changes in constraint coefficients, and nonintuitive shadow
prices. We give examples ofthese three cases and discuss how to deal effectively with these
through re-solving the model with changes. In fact, in our experience, its rarely the case
that one solves a model once and makes a recommendation. More often than not, a series
cof models are solved using a variety of input datasets before a final plan is adapted. With
{improved algorithms and more powerful computers, solving multiple runs of a model is
extremely cost- and time-effective.
Simultaneous Changes
‘Classical sensitivity analysis is based on the assumption that only one coefficient changes;
itis assumed that all other coefficients will remain as stated in the original problem, Thus,
the range analysis forthe objective function coefficients and the constraint right-hand sides
is only applicable for changes in a single coefficient. In many cases, however, we are inter-
ested in what would happen if two or more coefficients are changed simultaneously. The
easiest way to examine the effect of simultaneous changes is to rerun the model. Computer
solution methods such as Excel Solver make rerunning the model easy and fast for many
applications,
Consider again the original RMC problem, Suppose RMC’s accounting department
reviews both the price and cost data for the two products. Asa result, the profit contribution
for the fuel additive is increased to $48 per ton and the profit contribution for the solvent
base is decreased to $27 per ton, Figure 8.6 shows the answer report for this revised prob-
lem. The total profit has increased to $48(25) + $27(20) = $1740, but the optimal solution
‘of 25 tons of fuel additive and 20 tons of solvent base has not changed.
Now suppose thatthe profit contribution for fuel additive is increased again to $55 per
ton and the profit contribution for the solvent base remains at $27 per ton. If RMC produces
25 tons of fuel additive and 20 tons of solvent base, this will generate a profit of $55(25) +
$2720) = $1915. However, the answer report in Figure 8.7 shows that if we re-solve thi
problem wit the new profit contribution values, the optimal solution changes. The optimal
solution is to produce 35 tons of fuel additive and zero tons of solvent base. This optimal
solution results ina profit of $55(35) + $27(0) = $1925.
‘Sensitivity analysis for the right-hand side of constraints has a similar limitation. The
right-hand-side sensitivity analysis information is based on the assumption that only one
right-hand side changes at atime. If two or more right-hand sides change simultaneously,
the easiest way to observe the effect of these changes is to re-solve the model 320 Chopler 8 Linear Programming: Sensitivity Analysis and Interpretation of Solution
FIGURE 8.6 ANSWER REPORT FOR RMC PROBLEM WITH CHANGE IN PROFITS PER TON FOR FUEL
ADDITIVE TO $48 AND SOLVENT BASE TO $27
‘Original Value Final Value
Maximize Total Profit 0.000 1740.00 7
Variable Cells
“Model Variable Name. ‘Original Value Final Value Integer
A “Tons Produced Fuel Adaitive 0.000 725.000 ‘Contin
B “Tons Produced Solvent Base 2.000 720.000 Contin
‘Constraints
Constraint Number ‘Name Call Value Status Slack
1 Material | Amount Used. "2000 Binding 0.000
2 ‘Material 2 Amount Used 4.000) ‘Not Binding 1.000
3 ‘Material 3 Amount Used 2,000. Binding 0.000)
FIGURE 8.7 ANSWER REPORT FOR THE RMC PROBLEM WITH ADDITIONAL INCREASE IN PROFIT
PER TON FOR FUEL ADDITIVE TO $55
Objective Cell Max
Nane Origa Value Final Vanie =
Msinie Tol Poi 00D 1925.000
Variable cos
“Model Variable Nome Original Value Final Value Tnteger_
A Tons Produced Fuel Aine con 35000 Contin
B “Tons Produced Solvent Base ‘2000 0.000 Conin
CConsnins
Constraint Number Tame Calvan Sites Siac
H Maal Amount Used 14.000, Not Binding 6.000
2 Matra Amount Used 2.000 Not Binding 5.000
3 Mai} Amoont Used 7.00, Binding 00
Changes in Constraint Coefficients
Classical sensitivity analysis provides no information about changes resulting from a
change in the coefficient of a variable in a constraint. We return to the RMC problem to
‘ustrate this idea
Suppose RMC is considering a different blending formula such that a ton of fuel ad-
itive uses 0.5 tons of material I instead of 0.4 tons. The constraint for material 1 would
then change to
OSF + 0.58520
Even though this is a single change in a coefficiem in the model, there is no way to ‘ell from,
classical sensitivity analysis what impact the change in the coefficient of F will have on the ~ FIGURE 8.8 ANSWER.
8.4 Limitations of Classical Sensitivity Analysis. 321
REPORT FOR THE RMC PROBLEM WITH CHANGES TO CONSTRAINT
(COEFFICIENTS
Objective Cell Max) nal =!
‘Name ‘Oviginal Value Final Value
Maximize Total Profit 0.000 150.000 a
Variable Cells
‘Model Variable ___Name ‘Original Value Final Value Tntener
A Tons Produced Fuel Additive 0.000 30.000 ‘Contin
B “Tons Produced Solvent Base 0.000 10.000, ‘Contin
Constraints
Constraint Number Name Call Value Status ‘Slack
1 Material | Amount Used 20000 Binding 0.000
2 ‘Material 2 Amount Used 2.000 ‘Not Binding 3.000
3 ‘Material 3 Amount Used 21.000 Binding 0.000
solution. Instead, we must simply change the coefficient and rerun the model. The answer report
appears in Figure 8.8. Note tha itis optimal to prxiuce 30 tons of fel additive and 10 tons of
solvent base, The optimal profit has also changed from $1600 to $40(30) + $3020) = $1500.
Changing to this new blending formula will ost RMC $1600 ~ $2500 = $100.
Nonintuitive Shadow Prices
Constraints with variables naturally on both the Iet-hand and right-hand sides often lead
to shadow prices that have @nonintuitve explanation. To illustrate such a case and how we
‘may deal with it, let us again reconsider the RMC problem
‘Suppose that after reviewing the solution to the original RMC problem (sensitv-
ity report shown in Figure 8.4), management decides that itis concerned with solutions
requiring greater production of fuel additive than solvent base. Management believes that
the profit generated per ton of fuel additive could decrease in the future, so it is more
comfortable producing a greater amount of solvent base than fuel additive. If management
wants to specify that RMC produces atleast as much solvent base as fuel additive, then we
must add the constraint
S=P
‘This new constraint will require RMC to produce atleast as many tons of solvent base as
fuel additive. The sensitivity report generated from resolving this problem with the new
constraint is shown in Figure 8.9. This shows that itis optimal to produce 22.222 tons of
fuel additive and 22.222 tons of solvent base. The total profit using this optimal solution is.
$40(22.222) + $30(22.222) = $1556.
Let us consider the shadow price for the Min Solvent Base Required constraint, The
shadow price of ~8.89 indicates that a one-unit increase inthe right-hand side of the Min
Solvent Base Required constraint will lower profits by $8.89. Thus, what the shadow price
is really telling us is what will happen to the value of the optimal solution ifthe constraint
is changed to
S=F+1 322
CChopter 8 Linear Programming: Sensitivity Analysis and Interpretation of Solution
FIGURE 8.9 SENSITIVITY REPORT FOR RMC PROBLEM WITH ADDITIONAL CONSTRAINT FOR
MINIMUM SOLVENT BASE PRODUCTION REQUIRED
‘Variable Cells
‘Model Final Reduced’ Objective Allowable Allowable
Variable Name Value ___Cost__Coefficient__Increase_ Decrease
F___ Teas Produced Fuel Additive 22.222 0.000. 40,000 1E+30_ 16.000
'S____ Tons Produced Solvent Base 22.222 0.000. 30,000 20.000 70.000
Constraints
Final Shadow Constraint Allowable Allowable
Number Name Value Price R.H.Side__Inerease __Deerease
Material 1 Amount Used 20000 77.778 20.000 1.000 20.000
2 ‘Materia? Amount Used ada 0.000 5.000, TE=30 0556
3 ‘Material 3 Amount Used 20.009 0.000 21.000 130 1,000
4 ‘Min Solvent Base Required 22223 BARD 0.000) 6250) 3.000)
‘The interpretation for this shadow price of ~8.89 is correctly stated as follows: If we are
forced to produce | ton more of solvent base over and above the amount of fuel additive
produced, total profits will decrease by $8.89. Conversely, if we relax the requirement by
1 ton ($= F ~ 1), total profits will increase by $8.89,
‘We might instead be more interested in what happens if we change the coefficient on
F For instance, what if management reqsired RMC to produce an amount of solvent base
that is at least 110% of the amount of fuel additive produce? In other words, the constraint
would change to
Se LiF
‘The shadow price does not tellus what will happen in this case. Because we have changed
the coefficient of F from 1,0 10 1.1, tis isthe same as the case discussed in the previous
section: a change in the constraint coefficient, Since there is no way to get this information
from classical sensitivity analysis, we need to resolve the problem using the constraint S=
LAF. To test the sensitivity ofthe solution to changes in the minimum required percentage
of solvent base required, we can resolve the model replacing the coefficient of F with any
Percentage of interest
To get a feel for how the required percentage impacts total profit, we solved ver-
sions ofthis model varying the percentage from 100% to 200% in increments of 10%.
nother words, we varied the coefficient of F from 1.0 to 2.0 in increments of 0.1. The
impact of changing this percentage is shown in Figure 8.10, andthe results are shown
in Table 8.1
‘What have we learned from this analysis? Notice from Figure 8.10 thatthe slope
of the graph becomes steeper for values larger than 130%. This indicates that there is
a shift in the rate of deterioration in profit starting at 130%. Table 8.1 shows why this
is the case. For all percentages larger than 130%, we produce 25 tons of solvent base
We are unable to produce additional solvent base due to the material 2 constraint. This,
is because the left-hand side of the material 2 constraint 0.2(25) = 5, which is equal
to the right-hand side. Thus, the material 2 constraint is binding whenever we produce
25 tons of material 2. 8.5. More Than Two Decision Variables 123
FIGURE 8.10 PROFIT FOR VARIOUS VALUES OF REQUIRED SOLVENT BASE AS A.
PERCENTAGE OF FUEL ADDITIVE PRODUCED
1600.0
$1550.
1500.0]
_ $1450.00
E 1400.00]
* sraso00
$1300.00
125000)
1200.00
100% Tho Ta0% 160% 180% 200%
Solvent Base Required as a Percentage of Fuel Additive Produced
TABLE 8.1 SOLUTIONS FOR VARIOUS VALUES OF MINIMUM REQUIRED PRODUCTION
OF SOLVENT BASE AS A PERCENTAGE OF FUEL ADDITIVE PRODUCED
Percent Profit Fuel Additive Solvent Base
100% $1556.00 22.222 22.222
110% $1537.00, 21.053 23.158,
120% $1520.00, 20.000 24.000
130% $1505.00, 19.048 24.762
140% $1464.00 17857 25.000
150% $1417.00 16.667 25.000
160% $1375.00 15.625 25.000
170% $1338.00 14.706 25.000
180% $1306.00 13,889 25.000
190% $1275.00 13.158 25.000
200% $1250.00 12.500 25.000
Management now knows that minimum perceatage requirements between 100% and
130% result in modest profit losses. Minimum percentage requirements greater than 130%
result in greater profit losses. Greater minimum percentage requirements will result in
‘more significant profit losses because we are unable to produce more than 25 tons of sol
vent base due to the materisi 2 constraint,
(88) More Than Two Decision Variables
‘The graphical solution procedure is useful only for linear programs involving two deci-
sion variables. In practic, the problems solved using linear programming usually involve
large numbers of variables and constraints. Fo instance, the Q.M. in Action, Determining Chapter 8 Linear Programming: Sensitivity Analysis and Interpretation of Solution
Optimal Production Quantities at GE Plastics, describes how a linear programming modet
with 3100 variables and 1100 constraints was solved in less than 10 seconds to determine
{he optimal production quantities at GE Plastics. In this section we discuss the formulation
and computer solution for two linear programs with three decision variables. In doing so,
‘we will show how (o interpret the reduced-cost portion of the computer output and will
also illustrate the interpretation of shadow prives for constraints that involve percentages.
Modified RMC Problem
‘The RMC linear programming problem was introduced in Section 7.1. The original prob-
Jem formulation is restated here:
Max
st
40F + 305
OF + 0.5820. Material 1
0.25 <= 5 Material 2
0.6F + 0.35 <21 Material 3
s=0
‘Suppose that management also is considering producing a carpet cleaning fluid. Estimates
are that each ton of carpet cleaning fluid will require 0.6 tons of material 1, 0.1 tons of
‘material 2, sd 0.3 tons of material 3. Because ofthe unique capabilities of the new prod-
‘uct, RMC’s management believes that the company will realize a profit contribution of,
‘$50 for each ton of carpet cleaning fluid produced during the current production perio.
Let us consider the modifications in the original linear programming model that
are needed to incorporate the effect of this additional decision variable, We let C denote
the number of tons of carpet cleaning fluid produced. After adding C to the objective
DETERMINING OPTIMAL PRODUCTION QUANTITIES AT GE PLASTICS”
GE Plastics (GEP) isa $5 billion global materials sup:
plier of plastics and raw materials to many industries
(€4, automotive, compuier, and medical equipment).
GEP has plants ait over the globe. In the past, GEP fol-
“Towed a pole-centric manufacturing approach wherein
‘each product was manufactured inthe geographic area
(Americas, Europe, or Pacific) where it was to be de-
livered. When many of GEP’s customers started shift.
ing their manufacturing operations to the Pacific,
‘geographic imbalance was created between GEP's ca-
pacity and demand in the form of overcapacity in the
‘Americas and undercapacity in the Pacific,
Recognizing that a pole-centric appmach was no
longer effective. GEP adopted « global approach to its
“Boxed an R yogPKalch, ond K Albay, “GE Paws Optinins
the Twocholon Global Flliant Netvrk at Is Highferlomorce
Poynrs Dison,” herfoces Sopember/Octber 200: 359-266.
‘manufacturing operations. Initial work foetsed of the
high performance polymers (HPP) division. Using anear
programming model, GEP was ale to determine the opti-
‘mal production quantities at each HPP plant wo maximize
the total contiation margin for the division, The model
included demand constsnts, manulaturng capacity con-
straints, and constants that modeled the low of materials
‘produce at resin plants to the finishing plants and on to
‘warehouses in three geographical regions (Americas, Eu-
‘ope and Pacific). The mathematical model or aone- year
problema has 3100 variables and 1100 constants and can
‘be solved ines than 10 seconds, The new system proved
succesful atthe HPP division, and oer GE Pastis divi-
sons are adapting it fr ther supply chan planning.
In 2007, GEP was acquired by the Saudi Basic In-
Aysties Corporation (SABIC), the langst company’ in
the Middle East and one of the Jargest companies in the
worl 8
8.5. More Than Two Decision Variables
FIGURE 8.11 SENSITIVITY REPORT FOR MODIFIED RMC PROBLEM
‘Variable Cells
“Model Final Reduced Objective Allowable Allowable
Variable Name Value___Cost Coefficient Increase _ Decrease
WEB, 'F Tons Produced Fuel Additive 275006060 40.000 60.000 6.667
TS Tons Produced Solvent Base. (0.000 —12.590 30.000 _12.500___1B +30
Modtiearc-—_ C Tons Produced Carpe: Cleaning uid 15.000 0.000 50,000 10.000 16.667
Constraints
Constraint Final Shadow Constraint Allowable Allowable
‘Number Name Value Price __R.NLSide Increase Decrease
L ‘Matera 1 Amount Used 20.000 75.000 20.900 14.000 6,000,
2 “Material 2 Amount Used 1.500 0.000 5.009__1B+30 3.500
3 Material 3 Amount Used 21.000 16.667 21.000 9.000 11.000
function and to each of the three constraints, we obtain the linear program for the modi-
fied problem:
Max 40F + 308 +500
st
O4F +0.5S-+0.6C $20 Material 1
025 +0.1C = 5 Material 2
0.6 +038 +0.3C =21 Material 3
FS,CzO
Figure 8.11 shows the sensitivity report for this solution tothe modified RMC problem. The
cptimal soliton calls fr the production of 27.5 tons of fuel additive, 0 tons of solvent base,
and 15 cons of carpet cleaning fluid, The value ofthe optimal solution is $40(27.5) + $30(0) +
$50(15) = $1850.
Note the information contained in the Reduced Costs column of the Variable Cells
section, Recall that reduced costs are the shadow prices of the corresponding nonnegativity
constraints. As Figure 8.11 shows, the reduced costs for fuel additive and carpet cleaning
fluid variables are zero because increasing the right-hand side of these nonnegativity con-
straints would not change the optimal objective function value. However, the reduced cost
for the solvent base decision variable is —12.50. This means thatthe shadow price for the
‘nonnegativity constraint associated with the solvent base decision variable is ~12.50. The
interpretation for this value is that if the nonnegatvity constraint, S = 0, was changed to S
= 1, the optimal objective function value would decrease by $12.50. In other words, if we
forced the production of atleast I ton of solvent base, the profit for the optimal solution
‘would decrease by $12.50
Figure 8.11 also shows thatthe shadow prices for material 1 amount used and material
3 amount used are 75.000 and 16.667, respectively, indicating that these two constraints
are binding inthe optimal solution. Thus, each additional ton of material | would increase
the value of the optimal solution by $75 and each additional ton of material 3 would
increase the valve of the optimal solution by $16.67,
®ancthar iserproin fe tii that we “reduce the cot of he objotve function ain fr sont base by
=12.50 fe. change he profit contbution fom selves baw to $30 ~ [$12.50] = $42.50], than heron opal
sali where we produce nonare amour of cent bose. Chapter 8 Linear Programming: Sensitivity Analysis and Interpretation of Solution
Bluegrass Farms Problem
‘To provide additional practice in formulating and interpreting the computer solution for
linear programs involving more than two decision variables, we consider a minimization
problem involving three decision variables. Bluegrass Farms, located in Lexington, Ken-
tucky, has been experimenting with a special diet fr its racehorses. The feed components
available for the diet are @ standard horse feed product, an enriched oat product, and a
new vitamin and mineral feed additive. The nutritional values in units per pound and the
costs forthe three feed components are summarized in Table 8.2; for example, each pound
ofthe standard feed component contains 0.8 units of ingredient A, 1 unit of ingredient B,
and 0.1 units of ingredient C. The minimum daily diet requirements foreach horse are
3 units of ingredient A, 6 units of ingredient B, and 4 units of ingredient C. Tn addition,
to control the weight of the horses, the total daily feed for a horse should not exceed 6
pounds. Bluegrass Farms would like to determine the minimum-cost mix that will satis
the daily diet requirements
‘To formulate a linear programming model forthe Bluegrass Farms problem, we intro-
dace three decision variables:
‘S = number of pounds of the standard horse feed product
E = number of pounds of the enriched oat product
‘A= number of pounds of the vitamin and mineral feed additive
Using the data in Table 8.2, the objective function that will minimize the toal cost associ-
ated with the daily feed can be written as follows:
Min 0.255 + 0.5E +34
Because the minimum daily requirement for ingredient A is 3 units, we obtain the constraint
O88 +028 = 3
‘The constraint for ingredient B is
1.08 + LSE +304 = 6
and the constraint for ingredient C is
OS + 0.68 + 2.08 = 4
Finally, the constraint that restricts the mix to at most 6 pounds is
S+E+AS6
TABLE 8.2 NUTRITIONAL VALUE AND COST DATA FOR THE BLUEGRASS
FARMS PROBLEM
Feed Component Standard Enriched Oat Additive
Ingredient A os. 02 00
Ingredient B 10 15 30
Ingredient © 01 06 20
Cost per pound $0.25 $0.50 $3.00 8.5 More Than Two Decision Variables 327
‘Combining all the constraints with the nonnegatvity requirements enables us to write the
‘complete linear programming model for the Bluegrass Farms problem as follows
Min 0.255 +0.50E+ 34
st.
088+ 028 =3° Ingredient A
105+ 15€+30A=6 Ingredient B
O.1S + O.6E +2.0A =4 Ingredient C
S+ E+ A<6 Weight
S.EA=0
The sensitivity eport for the Bluegrass Farms problem is shown in Figure 8.12. After
rounding, we see thatthe optimal solution calls fora daily diet consisting of 3.51 pounds of
the standard horse feed product, 0.95 pounds of the enriched oat product, and 1.54 pounds.
of the vitamin and mineral feed additive. Thus, with feed component cost of $0.25, $0.50,
and $3.00, the total cost of the optimal det is
3.51 pounds @ $0.25 per pound = $0.88
0.95 pound @ $0.50 per pound = $0.47
1.54 pounds @ $3.00 per pound = $4.62
Total cost = $5.97
Looking at the Constraints section of the sensitivity report, we see that the final value
for the Ingredieat B LHS constraint is 9.554 and the right-hand side of this constraint is
6, Because dhis constraint is a greater-than-or-equalto constraint, 3.554 is the surplus; the
‘optimai solution exceeds the minimum daily diet requirement for ingredient B (6 units) by
4.554 units. Because the final values for the ingredient A and ingredient C constraints are
FIGURE 8.12 SENSITIVITY REPORT FOR THE BLUEGRASS FARMS PROBLEM.
Variable Ceti
“Model Final Reduced Objective Allowable Allowable
Variable Name Value Cost Coefficient _Increase__Decrease
‘S__ Number of Pounds Standard 3.514 0.000 _0250_—‘1B¥30 0.683,
E Number of Pounds Enriched Oat 0.946 0.000 0.500 0425 1E-+30
‘A___ Number of Pounds Additive 15410000 3.000 1E430__*1478
Constraints
Constraint Final Shadow — Constraint Allowable Allowable
Number Name Value Price __RM.Side Increase __ Decrease
1 Tgredient A 3.000 1216 3.000 0.368. 1857
2 Tngredient B 9554 0.00 6.000 3.554 1E+30
3 “Ingredient 400 195940000875 1.900
4 Weight 6000-0919 60002478 (0.437 328
Chapter 8 Lineor Programming: Sensitivity Analysis and Interpretation of Soliton
‘equal to the right-hand sides for these constraints, we see thatthe optimal diet just meets
‘the minimum requirements for ingredients A and C. The final value for the weight con-
straintis equal to the right-hand side (6 pounds). This tells us that this constraint has a slack
value of zero, which means that the optimal solution provides a total daly feed weight of
6 pounds
‘The shadow price (after rounding) for ingredient A is 1.22. Thus, increasing the
right-hand side of the ingredient A constraint by one unit will cause the solution value to
increase by $1.22, Conversely, i is also correct to conclude that a decrease of one unit
in the right-hand side of the ingredient A constraint will decrease the total cost by $1.22.
Looking at the Allowable Increase and Allowable Decrease columns in the Constraints
section of the sensitivity report, we sce that these interpretations are correct as long as
the right-hand side of the ingredient A constraint is between 3 ~ 1.857 = 1.143 and
3 + 0,368 = 3.368
‘Suppose thatthe Bluegrass management is willing to reconsider its position regarding
the maximum weight ofthe daly diet. The shadow price of 0,92 (aftr rounding) for the
‘weight constraint shows that a one-unit increase in the right-hand side of constraint 4 will
reduce total cost by $0.92. The Allowable Increase column in the sensitivity report shows
that this interpretation is correct for increases in the right-hand side up to a maximum of
6 + 2.478 = 8.478 pounds. Thus, the effect of increasing the right-hand side of the weight
‘constraint from 6 to 8 pounds is a decrease in the total daily cost of 2 X $0.92, or $1.84.
Keep in mind that if this change were made, the feasible region would change, and we
would obtain a new optimal solution.
Next we look at the Variable Cells section of the sensitivity report. The Allowable
Decrease column shows a lower limit of 0.25 ~ 0.683 = —0.393 for the objective func-
tion coetficient associated withthe standard horse feed product variable, S. Clearly, in a
real problem, the objective function coefficient of 5 (the cost ofthe standard horse feed
product) cannot take on a negative value. So, from a practical point of view, we can
think of the lower limit for the objective function coefficient vf 5 as being zero. We can
thus conclude that no matter how much the cost of the standard mix were to decrease,
the optimal solution would not change. Even if Bluegrass Farms could obtain the stan-
dard horse feed product for free, the optimal solution would still specify a daily diet of
3.51 pounds of the standard horse feed product, 0.95 pounds of the enriched oat
product, and 1.54 pounds of the vitamin and mineral feed additive. However, any
decrease in the per-unit cost of the standard feed would result in decrease inthe total
cost for the optimal daily diet.
Note from Figure 8.12 that the allowable increases for the objective function coef-
ficients associated with standard horse feed product and additive are shown as 1E+30.
This is the same notation as is used in Excel's Sensitivity Report; it means that these
objective function coefficients have no upper limit. Even if the cost of additive were to
increase, for example, from $3,00 to $13.00 per pound, the optimal solution would not
change; the total cost of the solution, however, Would increase by $10 (the amount of
the increase) X 1.541, or $15.41. You must always keep in mind that the interpretations
‘we make using classical sensitivity analysis information are only appropriate if all other
coefficients in the problem do not change. To conser simultaneous changes, we must
re-solve the problem.
‘Linear programming has been successfully applied to a variety of applications. The
QM. in Action, Kimpton Hotels Uses Optimization for Setting Prices on Priceine,
provides an example from the hotel industry. The Q.M. in Action discusses how Kimpton
Hotels uses linear programming and, specifically, the value of shadow price, to set prices
‘of rooms sold through Priceline. 86 Electronic Communicotions Problem
KIMPTON HOTELS USES OPTIMIZATION FOR SETTING PRICES ON PRICELINE®
ow to price rooms to maximize revenue is a problem
faced by all hotels, If prices are set to0 low, demand
‘will be higher, but total revenue may be lower than what
‘would have been generated ifthe customer's willingness
to pay was known. Ithe price is too high, demand may
