Chapter 13

Linear Optimization

Linear Optimization
Optimization is the process of selecting values of
decision variables that minimize or maximize
some quantity of interest.
Optimization models have wide applicability in
operations and supply chains, finance, marketing,
and other disciplines.
This chapter focuses only on linear optimization
models.

Building Linear Optimization Models

1.

2.

3.

4.

Identify the decision variables the unknown values

that the model seeks to determine.
Identify the objective function the quantity we seek
to minimize or maximize.
Identify all appropriate constraints limitations,
requirements, or other restrictions that are imposed on
any solution, either from practical or technological
considerations or by management policy.
Write the objective function and constraints as
mathematical expressions.

Example 13.1 Sklenka Ski Company:

Identifying Model Components

SSC sells two snow ski models - Jordanelle & Deercrest

Manufacturing requires fabrication and finishing.
The fabrication department has 12 skilled workers, each of
whom works 7 hours per day. The finishing department has 3
workers, who also work a 7-hour shift.
Each pair of Jordanelle skis requires 3.5 labor-hours in the
fabricating department and 1 labor-hour in finishing.
The Deercrest model requires 4 labor-hours in fabricating and
1.5 labor-hours in finishing.
The company operates 5 days per week.
SSC makes a net profit of $50 on the Jordanelle model and
$65 on the Deercrest model.

Example 13.1 Continued

Step 1: Identify the decision variables
The company wants to determine how many of
each model should be produced on a daily basis
to maximize net profit.
Define

Jordanelle = number of pairs of Jordanelle skis

produced/day
Deercrest = number of pairs of Deercrest skis
produced/day

Clearly specify the dimensions of the variables!

Example 13.1 Continued

Step 2: Identify the objective function

SSC wishes to maximize net profit, and we are given the
net profit figures for each type of ski.
SSC makes a net profit of $50 on the Jordanelle model and $65
on the Deercrest model.

Example 13.1 Continued

Step 3: Identify the constraints

Look for clues in the problem statement that describe limited resources
that are available, requirements that must be met, or other restrictions.

Both the fabrication and finishing departments have limited numbers

of workers, who work only 7 hours each day; this limits the amount
of production time available in each department:
Fabrication: Total labor hours used in fabrication cannot exceed the amount of
labor hours available.
Finishing: Total labor hours used in finishing cannot exceed the amount of labor
hours available.

The problem also notes that the company anticipates selling at least
twice as many Deercrest models as Jordanelle models:
Number of pairs of Deercrest skis must be at leasttwice the number of parts of
Jordanelle skis.

Negative values of the decision variables cannot occur

(nonnegativity constraints)

Translating Model Information into

Mathematical Expressions

Represent decision variables by descriptive

names, abbreviations, or subscripted letters
(X1, X2, etc.)
For mathematical formulations involving many
variables, subscripted letters are often more
convenient.
In spreadsheet models, we recommend using
more descriptive names to make the models and
solutions easier to understand.

Example 13.2: SSC Modeling the

Objective Function
Profit

per pair of skis sold:

$50 for Jordanelle skis, $65 for Deercrest skis

Objective

Function:
Maximize total profit
= 50 Jordanelle + 65 Deercrest
Note how the dimensions verify that the
expression is correct:
($/pair of skis)(number of pairs of skis) = $.

Translating Constraints
Mathematically

Constraints are expressed as algebraic inequalities or

equations, with all variables on the left side and constant
terms on the right.
Look for key words in word statements of constraints:
Cannot exceed translates mathematically as
At least, would translate as
Must contain exactly, would specify an = relationship.

All constraints in optimization models must be one of

these three forms.

Constraint Functions

A constraint function is the left-hand side of a

constraint.
E.g.: Total labor-hours used in fabrication cannot
exceed the amount of labor hours available.

Example 13.3: SSC Modeling the

Constraints

Fabrication constraint
Available fabrication labor hours: (12 workers)(7 hours/day) = 84
hours/day
Required fabrication labor hours per ski pair: 3.5 hours for
Jordanelle, 4 hours for Deercrest
Fabrication constraint: 3.5 Jordanelle + 4 Deercrest 84

Finishing constraint
Available finishing labor hours: (3 workers)(7 hours/day) = 21
hours/day
Required finishing labor hours per ski pair: 1 hour for Jordanelle;
1.5 hours for Deercrest
Finishing constraint: 1 Jordanelle + 1.5 Deercrest 21

Example 13.3 Continued

Market mixture constraint

The number of pairs of Deercrest skis must be at least
twice the number of Jordanelle skis.
Deercrest 2 Jordanelle,
or 2 Jordanelle + 1 Deercrest 0

Nonnegativity constraints:
Jordanelle 0
Deercrest 0

SSC Optimization Model

Maximize total profit = 50 Jordanelle + 65 Deercrest
3.5 Jordanelle + 4 Deercrest 84
1 Jordanelle + 1.5 Deercrest 21
2 Jordanelle + 1 Deercrest
0
Jordanelle 0
Deercrest 0
The highlighted portions are the constraint functions

More About Constraints

Some examples:
The amount of money spent on research and
development projects cannot exceed the assigned
budget of $300,000.
Amount spent on research and development 300,000

Contractual requirements specify that at least 500 units

of product must be produced.
Number of units of product produced 500

A mixture of fertilizer must contain exactly 30% nitrogen.

Amount of nitrogen in mixture/total amount in mixture = 0.30

Example 13.4: Modeling a Mixture

Constraint

A fertilizer mixture is made of two ingredients and must contain exactly

30% nitrogen. Ingredient X contains 20% nitrogen. Ingredient Y
contains 33% nitrogen.
Define x = the number of pounds of X in the mixture and y = the
number of pounds of Y in the mixture
Amount of nitrogen in mixture = 0.20x + 0.33y
Total amount of mixture = x + y
Fraction of nitrogen in mix = (0.20x + 0.33y)/(x + y)

Since the fraction of nitrogen must be 0.30, the constraint

would be
(0.20x + 0.33y)/(x + y) = 0.30, or simplified as -0.1x - 0.03y = 0
Note that the first version is not linear; however the simplified
constraint is linear.

Characteristics of Linear Optimization

Models
A linear optimization model (often called a
linear program, or LP) has two basic properties.
1. The objective function and all constraints are
linear functions of the decision variables.

This means that each function is simply a sum of terms,

each of which is some constant multiplied by a decision
variable.
2.

All variables are continuous

This means that they may assume any real value
(typically, nonnegative).

Implementing Linear Optimization

Models on Spreadsheets

Put the objective function coefficients, constraint coefficients, and

right-hand values in a logical format in the spreadsheet.
For example, you might assign the decision variables to columns and the
constraints to rows

Define a set of cells (either rows or columns) for the values of the
decision variables.
The names of the decision variables should be listed directly above the
decision variable cells.
Use shading or other formatting to distinguish these cells.

Define separate cells for the objective function and each constraint
function (the left-hand side of a constraint).
Use descriptive labels directly above these cells.

Example 13.5: A Spreadsheet Model for

Sklenka Skis

Decision variables Objective function Constraint functions

Correspondence Between the Model and

the Spreadsheet
Maximize Jordanelle + 65 Deercrest
3.5 Jordanelle + 4 Deercrest 84
1 Jordanelle + 1.5 Deercrest 21
2 Jordanelle + 1 Deercrest 0
Jordanelle 0
Deercrest 0

Maximize D22 = B9* B14 + C9* C14

D15 = B6* B14 + C6* C14 D6
D16 = B7* B14 + C7* C14 D7
D19 = C14 - 2* B14 0
B14 0
C14 0

Using the SUMPRODUCT Function

In Excel, the pairwise sum of products of terms

can easily be computed using the SUMPRODUCT
function.
B9* B14 + C9*C14 = SUMPRODUCT(B9:C9,B14:C14)

This often simplifies the model-building process,

particularly when many variables are involved.

Excel Functions to Avoid in Linear

Optimization

Several common functions in Excel can cause difficulties

when attempting to solve linear programs using Solver
because they are discontinuous (or nonsmooth) and do
not satisfy the conditions of a linear model.
These include:

IF
MAX
INT
ROUND
COUNT

Solving Linear Optimization Models

A feasible solution to an optimization problem is
any solution that satisfies all of the constraints.
An optimal solution is the best of all the feasible
solutions.
Software for determining optimal solutions

Solver (standard Solver) is a free add-in packaged with

Excel for solving optimization problems.
Premium Solver, which is a part of Analytic Solver
Platform has better functionality, accuracy, reporting, and
interface.

Using the Standard Solver

Data > Analysis > Solver in the Excel ribbon
Use the Solver Parameters dialog to define the
objective, decision variables, and constraints from
your spreadsheet model.

Example 13.6: Using Standard Solver for

the SSC Problem
Solver Parameters dialog
Objective function cell
Decision variables cells
Constraints

to enter click Add and fill in the Add

Constraint dialog:

Check box for Nonnegativity

Always select Simplex LP

Solver Results Dialog

Three reports: Answer,

Sensitivity, and Limits
To add them to your Excel
workbook, click on the
ones you want and then
click OK.

Do not check the box

Outline Reports; this is
an Excel feature that
produces the reports in
"outlined format."

Optimal Solution to SSC Problem

Using Premium Solver

After installing Analytic Solver Platform, Premium
Solver will be found under the Add-Ins tab in the
Excel ribbon.
Premium Solver has a different user interface than
the standard Solver.

Example 13.7: Using Premium Solver for

the SSC Model

Solver Parameters
dialog
First, click on Objective
and then click the Add
button. The Add
Objective dialog
appears, prompting you
for the cell reference for
the objective function
and the type of objective
(min or max).

Example 13.7 Continued

Next, highlight Normal

under the Variables list
and click Add; this will
bring up an Add
Variable Cells dialog.
Enter the range of the
decisions variables in
the Cell Reference field.

Example 13.7 Continued

Next, highlight Normal

under the Constraints
list and click the Add
button; this brings up
the Add Constraint
dialog, just like in the
standard version.

Example 13.7 Continued

Check this box

Select Standard
LP/Quadratic for the
solving method

Example 13.7 Continued

Completed
Premium Solver
dialog

Solver Answer Report

The Solver Answer Report provides basic information about the

solution, including the values of the original and optimal objective
function (in the Objective Cell section) and decision variables (in the
Decision Variable Cells section).
In the Constraints section, Cell Value refers to the value of the
constraint function using the optimal values of the decision
variables.
A binding constraint is one for which the Cell Value is equal to the
right-hand side of the value of the constraint.
The Status column tells whether each constraint is binding or not
binding.
Slack refers to the difference between the left- and right-hand sides
of the constraints for the optimal solution.

Example 13.8: Interpreting the SSC Answer

Report

Example 13.8 Continued

Understanding slack values

Maximize profit = 50 Jordanelle + 65 Deercrest
3.5 Jordanelle + 4 Deercrest 84 (fabrication)
1 Jordanelle + 1.5 Deercrest 21 (finishing)
2 Jordanelle + 1 Deercrest 0
(market
mix)
Jordanelle
0
Deercrest 0
Optimal solution: Jordanelle = 5.25; Deercrest = 10.5

Fabrication constraint: 3.5(5.25) + 4(10.5) = 60.375 84

23.625 excess fabrication hours
Finishing constraint: 1(5.25) + 1.5(10.5) = 21 21
No excess finishing hours
Market mix constraint: 2(5.25) + 1(10.5) = 0 0
Exactly twice the number of Deercrest skis as Jordanelle skis

Graphical Interpretation of Linear

Optimization

The set of feasible solutions is called the feasible region.

For a problem with only two decision variables, x1 and x2, we can draw the
feasible region on a two-dimensional coordinate system by plotting the
equations corresponding to each constraint.
Nonnegativity constraints:

Example 13.9: Graphing the Constraints in the SSC

Problem

Fabrication constraint: 3.5 Jordanelle + 4 Deercrest 84

Plot the equation: 3.5 Jordanelle + 4 Deercrest = 84
Set Jordanelle = 0; Deercrest = 21
Set Deercrest = 0; Jordanelle = 24

Example 13.9 Continued

Finishing constraint: 1 Jordanelle + 1.5 Deercrest 21

Plot the equation: 1 Jordanelle + 1.5 Deercrest = 21
Set Jordanelle = 0; Deercrest = 14
Set Deercrest = 0; Jordanelle = 21

Example 13.9 Continued

Market mix constraint: -2 Jordanelle + 1 Deercrest 0

Plot the equation: -2 Jordanelle + 1 Deercrest = 0
Set Jordanelle = 5; Deercrest = 10
Set Deercrest = 0; Jordanelle = 0

Example 13.10 Identifying the Feasible Region and

Optimal Solution

Feasible region

Corner Points

The points at which the constraint lines intersect along

the feasible region are called corner points.
If an optimal solution exists, then it will occur at a corner
point.

Example 13.10 Continued

Because our objective is

to maximize profit, we
seek a corner point that
has the largest value of
the objective function Total
Profit = 50 Jordanelle + 65
Deercrest.
Graph the profit line and
move in an improving
direction until it passes
through the last corner
point of the feasible
region.
Solve the two intersecting
equations simultaneously
to find the optimal
solution.

How Solver Works

Solver uses a mathematical algorithm called the simplex

method, which was developed in 1947 by the late Dr.
George Dantzig.
The simplex method characterizes feasible solutions algebraically
by solving systems of linear equations.
It moves systematically from one corner point to another to
improve the objective function until an optimal solution is found
(or until the problem is deemed infeasible or unbounded).
It is quick and efficient.

Example 13.11: Crebo Manufacturing

Crebo Manufacturing produces 4 types of structural

support fittings.
Machining centers have a capacity of 280,000 minutes
per year.
Gross margin/unit and machining requirements:

How many units of each product type should be

produced to maximize gross profit margin?

Example 13.11 Continued

Define X1, X2, X3, and X4 as the number of plugs, rails,

rivets, and clips to produce.
Objective:
Maximize gross profit margin = 0.3 X1 + 1.3 X2 + 0.75X3 + 1.2X4

Constraints:
1X1 + 2.5 X2 + 1.5X3 + 2X4 280,000
X1, X2, X3, X4 0

How the Simplex Method Works

The simplex method evaluates the impact of constraints

in terms of their contribution to the objective function for
each variable.
For the simple case of only one constraint, the optimal
(maximum) solution is found by simply choosing the
variable with the highest ratio of the objective coefficient
to the constraint coefficient.

Example 13.12: Solving the Crebo

Manufacturing Model

Clips have the highest marginal profit per unit of resource consumed.

Maximum possible production of clips

= 280,000 minutes minutes/unit
= 280,000 2 = 140,000
Profit for maximum production of clips
= gross margin/unit * max possible production
= $1.20 * 140,000 = $168,000

How Solver Creates Names in Reports

Solver assigns names to:

Target cells
Changing cells
Constraint function cells

Names are formed by concatenating the first cell

containing text to the:
Left of the cell and
Above the cell

SSC Example

Name assigned to objective function

Cell D22: Profit Contribution + Total Profit

Names assigned to decision

variables:
Cell B14: Quantity Produced + Jordanelle
Cell B15: Quantity Produced + Deercrest

Names assigned to constraints:

Cell D15: Fabrication + Hours Used
Cell D16: Finishing + Hours Used
Cell D19: Market mixture + Excess
Deercrest

Solver Outcomes

Unique optimal solution

there is exactly one solution that will result in the maximum (or
minimum) objective.

Alternative (multiple) optimal solutions

the objective is maximized (or minimized) by more than one
combination of decision variables, all of which have the same
objective function value.

Unbounded solution
the objective can be increased or decreased without bound (i.e.,
to infinity for a maximization problem or negative infinity for a
minimization problem)

Infeasibility
no feasible solution exists

Example 3.13: A Model with Alternative

Optimal Solutions

New objective function in the SSC problem:

Max 50 Jordanelle + 75 Deercrest

13-

Example 13.14: A Model with an Unbounded

Solution

Remove the finishing and fabrication constraints from the Sklenka

Ski problem.

Solver message:

Example 13.15: An Infeasible Model

Suppose, by mistake, the modeler in the Sklenka Ski problem used

a sign in the fabrication constraint (instead of ):

Using Optimization Models for Prediction

and Insight

Models should be used to provide insight for

making better decisions.
What might happen should the model assumptions
change or when the data used in the model are
uncertain?
With Solver, answers to such questions can easily be
found by simply changing the data and re-solving the
model.

Example 13.16: Using Solver for What-If

Analysis
Four questions are posed by the managers of
Sklenka Ski company:
1. If the Jordanelle skis profit increased $10/pair,
how would the optimal solution change?
2. If the Jordanelle skis profit decreased $10/pair,
how would the optimal solution change?
3. If 10 additional finishing hours were available,
how would manufacturing plans be affected?
4. If 2 fewer finishing hours were available, how
would manufacturing plans be affected?

Example 13.16 Continued

Summary of What-If scenarios:

Solver Sensitivity Report

The Sensitivity Report allows us to understand how

the optimal objective value and optimal decision variables
are affected by changes in the objective function
coefficients,
the impact of forced changes in certain decision variables,
or
the impact of changes in the constraint resource limitations
or requirements.

The Sensitivity Report information applies to changes in

only one of the model parameters at a time; all others
are assumed to remain at their original values.

Sensitivity Report for the SSC

Problem

Example 13.17: Interpreting Sensitivity

Information for Decision Variables

Reduced Cost: How much the objective function coefficient needs to be

reduced for a nonnegative variable that is zero in the optimal solution to
become positive. If a variable is positive in the optimal solution, its reduced
cost is zero.
If the objective coefficient of any one variable that has positive value in the
current solution changes but stays within the range specified by the Allowable
Increase and Allowable Decrease, the optimal decision variables will stay the
same; however, the objective function value will change.

Example 13.18: Understanding Nonzero

Reduced Costs

Sensitivity report for the Sklenka Ski model after

changing the profit on Jordanelle skis from $50 to $40

Example 13.19: Interpreting Sensitivity

Information for Constraints

Shadow Price - how much the objective function will change as the
right hand side of a constraint is increased by 1.
Whenever a constraint has positive slack, the shadow price is zero.
When a constraint involves a limited resource, the shadow price
represents the economic value of having an additional unit of that
resource.

Limits Report

Shows the upper and lower limits that each

decision variable can assume while satisfying all
constraints and holding the other variables
constant.

Using the Sensitivity Report

If a change in an objective function coefficient remains

within the Allowable Increase and Allowable Decrease
ranges in the Decision Variable Cells section of the
report, then the optimal values of the decision variables
will not change. However, you must recalculate the value
of the objective function using the new value of the
coefficient.
If a change in an objective function coefficient exceeds
the Allowable Increase or Allowable Decrease limits in
the Decision Variable Cells section of the report, then
you must re-solve the model to find the new optimal
values.

Using the Sensitivity Report

(Continued)

If a change in the right-hand side of a constraint remains

within the Allowable Increase and Allowable Decrease ranges
in the Constraints section of the report, then the shadow price
allows you to predict how the objective function value will
change. Multiply the change in the right-hand side (positive if
an increase, negative if a decrease) by the value of the
shadow price. However, you must re-solve the model to find
the new values of the decision variables.
If a change in the right-hand side of a constraint exceeds the
Allowable Increase or Allowable Decrease limits in the
Constraints section of the report, then you cannot predict how
the objective function value will change using the shadow
price. You must re-solve the problem to find the new solution.

Example 13.20: Using Sensitivity

Information to Evaluate Scenarios

Suppose that the unit profit on Jordanelle skis is increased by $10.

How will the optimal solution change? What is the best product mix?

Is the increase in the objective function coefficient is within the range of

the Allowable Increase and Allowable Decrease in the Decision Variable
Cells portion of the report?
Because $10 is less than the Allowable Increase of infinity, we can safely
conclude that the optimal quantities of the decision variables will not
change.
However, because the objective function changed, we need to compute
the new value of the total profit: 5.25($60) + 10.5($65) = $997.50.

Example 13.20 Continued

Suppose that the unit profit on Jordanelle skis is

decreased by $10 because of higher material costs. How
will the optimal solution change? What is the best
product mix?

The change in the unit profit exceeds the Allowable Decrease

($6.67). We can conclude that the optimal values of the decision
variables will change, although we must re-solve the problem to
determine what the new values would be.

Example 13.20 Continued

Suppose that 10 additional finishing hours become

available through overtime. How will manufacturing
plans be affected?

Check if the change in the right-hand-side value is within the

range of the Allowable Increase and Allowable Decrease in the
Constraints section of the report.
Ten additional finishing hours exceeds the Allowable Increase.
Therefore, we must re-solve the problem to determine the new
solution.

Example 13.20 Continued

What if the number of finishing hours available is

decreased by 2 hours because of planned equipment
maintenance? How will manufacturing plans be
affected?

A decrease of 2 hours in finishing capacity is within the Allowable

Decrease. Total profit will decrease by the value of the shadow
price for each hour that finishing capacity is decreased.
Therefore, we can predict that the total profit will decrease by 2
$45 = $90 to $855. However, we must re-solve the model in
order to determine the new values of the decision variables.

Parameter Analysis in Analytic Solver

Platform

Solver can be used to perform sensitivity analysis

by either:
Examining the sensitivity reports or
Changing data in the model and re-solving it

Analytic Solver Platform offers an alternative

approach to sensitivity analysis called parameter
analysis, which allows you to run multiple
optimizations while varying model parameters
within predefined ranges.

Example 13.21: Single Parameter

Analysis for the SSC Problem

Investigate the impact of

changing finishing hour
capacity over a range from
10 to 60.
Choose an empty cell in the
spreadsheet (e.g., F3)
From Analytic Solver Platform
ribbon, click Parameters button
and choose Optimization.
Define range in Function
Arguments dialog.
Replace the value in cell D7 by
=F3

Example 13.21 Continued

From the Reports button in the

Analysis group in the Analytic
Solver Platform ribbon, select
Optimization Reports and then
Parameter Analysis
Move decision variable and
objective function cells to the right
Result Cells pane, and move the
parameter cell F3 to the right
Parameters pane.
In the drop-down box, select Vary
All Selected Parameters
Simultaneously.
Set Major Axis Points to number
of parameter values to test.

Example 13.21 Continued

Parameter analysis results

Reformat the results to make them easier to understand.

For example, name the columns with descriptive labels
instead of cell references; use charts to visualize the
results.

Example 13.22: Multiple Parameter

Analysis for the SSC Problem
Suppose that we wish to examine the effect on the
optimal profit of changing both the Fabrication and
Finishing hour limitations, similar to a two-way
data table.
Follow the procedure in Example 13.21 to define
the parameter for the Finishing limitation.

In the Function Arguments dialog, set the range for the

Fabrication limitation between 50 and 100.

Example 13.22 Continued

In the Multiple Optimizations

Report dialog, choose both
parameter cells F2 and F3;
however, we can only choose
one result cell. In this case,
choose $D$22, which represents
the objective function value.
In the drop-down box, select Vary
Two Selected Parameters
Independently.

Example 13.22 Continued

Results