Chapter

4 Linear Programming

B usinesses seek to maximize their profits while operating under budget, sup-
ply, labor, and space constraints. Determining which combination of variables
will result in the maximum profit may be done through the use of linear pro-
gramming. Although many factors affect a business’s profitability, linear pro-
gramming can help a business owner determine the “ideal” conditions for
business success.

4.1 Graphing Linear Inequalities

■ Graph linear inequalities
■ Set up and solve systems of linear
inequalities

4.2 Solving Linear Programming

Problems Graphically
■ Determine the feasible region of a
linear programming problem
■ Solve linear programming problems
in two variables

4.3 Solving Standard Maximization

Problems with the Simplex
Method
■ Apply the simplex method to solve
multivariable standard maximization
problems

4.4 Solving Standard Minimization

Problems with the Dual
■ Find the solution to a standard
minimization problem by solving the
dual problem with the simplex
method

4.5 Solving General Linear

Programming Problems with
the Simplex Method
■ Solve general linear programming
problems with the simplex method

171
172 CHAPTER 4 Linear Programming

4.1 Graphing Linear Inequalities

■ Graph linear inequalities

Many students work multiple part-time jobs to fi-
■ Set up and solve systems
of linear inequalities nance their education. Often the jobs pay different wages and offer varying hours.
Suppose that a student earns $10.50 per hour delivering pizza and $8.00 per hour
working in a campus computer lab. If the student has only 30 hours per week to work
and must earn $252 during that period, how many hours must he spend at each job
in order to meet his earnings goal? In this section, we will explain how linear in-
equalities may be used to answer this question. We will demonstrate how to graph
linear inequalities and show that the solution region of a system of linear inequali-
ties is the intersection of the graphs of the individual inequalities.

Linear Inequalities
In many real-life applications, we are interested in a range of possible solutions
instead of a single solution. For example, when you prepare to buy a house, a lender
will calculate the maximum amount of money it is willing to lend you; however,
the lender doesn’t require you to borrow the maximum amount. You may borrow
any amount of money up to the maximum. Recall that in mathematics, we use in-
equalities to represent the range of possible solutions that meet the given criteria.

INEQUALITY NOTATION
x # y is the set of all values of x less than or equal to y.
x $ y is the set of all values of x greater than or equal to y.
x , y is the set of all values of x less than but not equal to y.
x . y is the set of all values of x greater than but not equal to y.

The inequalities x , y and x . y are called strict inequalities because the

two variables cannot ever be equal. Although strict inequalities have many useful
applications, we will focus on the nonstrict inequalities in this chapter.
An easy way to keep track of the meaning of an inequality is to remember
that the inequality sign always points toward the smaller number. Consider these
everyday examples of inequalities:
You must be at least 16 years old to get a driver’s license. (16 # a) or
(a $ 16) .
You must be at least 21 years old to legally buy alcohol. (21 # a) or
(a $ 21) .
The maximum fine for littering is $200. (200 $ f ) or (f # 200) .
Your carry-on bag must be no more than 22 inches long. (22 $ l) or
(l # 22) .
 4.1 Graphing Linear Inequalities 173

A linear inequality looks like a linear equation with an inequality sign in the
place of the equal sign. Recall that linear inequalities may be manipulated alge-
braically in the same way as linear equations, with one major exception: When we
multiply or divide both sides of an inequality by a negative number, we must re-
verse the direction of the inequality sign. For example, if we multiply both sides
of 23x 1 2y # 10 by 21, we get 3x 2 2y $ 210, not 3x 2 2y # 210.

y Graphing Linear Inequalities

3 The graph of a linear inequality is a region bordered by a line called a boundary
line. The solution region of a linear inequality is the set of all points (including
2 the boundary line) that satisfy the inequality.
x + 2y = 4 Consider the inequality x 1 2y $ 4 (see Figure 4.1). The boundary line of
1 the solution region is x 1 2y 5 4, since the points satisfying this linear equation
are on the border of the solution region.
−1 1 2 3 4 5 6
x We need to find all points (x, y) that satisfy the inequality. We know that all
points on the line satisfy the inequality. Which points off the line satisfy the in-
−1
equality? Let’s pick a few points off the line (see Table 4.1) and test them to see
if they satisfy the inequality. In order to satisfy the inequality, x 1 2y must be at
FIGURE 4.1 least 4.

TABLE 4.1
In Solution
x y x 1 2y Region?
y
21 1 1 No
3 0 1 2 No
2 1 3 7 Yes
x + 2y = 4 2 2 6 Yes
1
3 0 3 No

−1 1 2 3 4 5 6
x 5 2 9 Yes
−1 6 1 8 Yes

FIGURE 4.2 Graphically speaking, what do the four points in the solution region have in com-
mon (see Figure 4.2)?
They are all on the same side of the boundary line! In fact, all points on or
y above this boundary line satisfy the inequality. We represent this notion by shad-
ing the region above the boundary line, as shown in Figure 4.3.
3
Although we checked multiple points in this problem, we need to check only
2 one point off the boundary line in order to determine which region to shade.
The linear inequality graphing process is summarized as follows.
1

x
−1 1 2 3 4 5 6
−1
x + 2y = 4

FIGURE 4.3
174 CHAPTER 4 Linear Programming

Linear Inequality Graphing Technique

To graph the solution region of the linear inequality ax 1 by # c (or

ax 1 by $ c), do the following:
1. Graph the boundary line ax 1 by 5 c.
2. Select a point on one side of the line. [If the line doesn’t pass through the
origin, (0, 0) is an excellent choice for easy computations.]
3. Substitute the point into the linear inequality and simplify. If the simpli-
fied statement is true, the selected point and all other points on the same
side of the line are in the solution region. If the simplified statement is
false, all points on the opposite side of the line are in the solution region.
4. Shade the solution region.

EXAMPLE 1 Graphing the Solution Region of a Linear Inequality

Graph the solution region of the linear inequality 2x 1 y # 4.

SOLUTION As shown in Chapter 1, the x-intercept is easily found by dividing

the constant term by the coefficient on the x term.
y 4
x5
5 2
4 (0, 4) 52

3 2x + y = 4 The point (2, 0) is the x-intercept.

The y-intercept is found by dividing the constant term by the coefficient on
2
the y term.
1
(2, 0) 4
y5
−1 1 2
x 1
−1 54

FIGURE 4.4 The point (0, 4) is the y-intercept. We graph the x- and y-intercepts and then
draw the line through the intercepts, as shown in Figure 4.4.
Next, we will pick the point (0, 0) to plug into the inequality.
2(0) 1 0 # 4
0#4
The statement is true, so all points on the same side of the line as the origin are
in the solution region. We shade the solution region (see Figure 4.5).
 4.1 Graphing Linear Inequalities 175

y
5

4 (0, 4)

3 2x + y = 4

1
(2, 0)
x
−1 1 2
−1

FIGURE 4.5

If you choose to convert lines from standard to slope-intercept form before

graphing them, the following properties will help you to quickly identify the so-
lution region without having to check a point.

SOLUTION REGION OF A LINEAR INEQUALITY

The solution region of a linear inequality y $ mx 1 b contains the line
y 5 mx 1 b and the shaded region above the line.
The solution region of a linear inequality y # mx 1 b contains the line
y 5 mx 1 b and the shaded region below the line.

Graphing Systems of Linear Inequalities

Just as we can graph systems of linear equations, we can graph systems of linear
inequalities. The solution region of a system of linear inequalities is the intersec-
tion of the solution regions of the individual inequalities. When we graph a solu-
tion region by hand, we will typically place arrows on the boundary lines to
indicate which side of the lines satisfies the given inequality. Once all of the lin-
ear inequality graphs have been drawn, we will shade the region that has arrows
pointing into the interior of the region from all sides.

EXAMPLE 2 Graphing the Solution Region of a System of

Linear Inequalities
Graph the solution region of the system of linear inequalities.
3x 1 2y # 5
x$0
y$0
176 CHAPTER 4 Linear Programming

y SOLUTION We first rewrite the linear inequality as a linear inequality in slope-

intercept form.
3
y = −1.5x + 2.5 3x 1 2y # 5
2 2y # 23x 1 5 Subtract 3x from both sides

1 3 5
y#2 x1 Divide both sides by 2
2 2
x y # 21.5x 1 2.5 Write as a decimal (optional)
−1 1
−1 The boundary line is a line with slope 21.5 and y-intercept (0, 2.5). Since y is
less than or equal to the expression 21.5x 1 2.5, we will shade the region below
the line, as shown in Figure 4.6.
FIGURE 4.6
The next two inequalities (x $ 0, y $ 0) limit the solution region to posi-
tive values of x and y. The line x 5 0 is the y axis. The line y 5 0 is the x axis.
Therefore, the solution region of the system of inequalities is the triangular region
to the right of the line x 5 0, above the line y 5 0, and below the line
y 5 21.5x 1 2.5 (see Figure 4.7).
y x=0

3
y = −1.5x + 2.5
2

1 y=0

x
−1 1
−1

FIGURE 4.7

If it is possible to draw a circle around the solution region, the solution re-
gion is bounded. If no circle can be drawn that will enclose the entire solution re-
gion, the solution region is unbounded. The solution region in Example 2 was
bounded. The solution region in Example 3 will be unbounded.

EXAMPLE 3 Graphing the Solution Region of a System of

Linear Inequalities
Graph the solution region of the system of linear inequalities.
4x 1 y $ 4
2x 1 y $ 1

SOLUTION The x-intercept of the boundary line 4x 1 y 5 4 is (1, 0) , and the

y-intercept is (0, 4) . Plugging in the point (0, 0) , we get
4(0) 1 (0) $ 4
0$4
Since the statement is false, we graph 4x 1 y 5 4 and place arrows on the side
of the line not containing the origin, as shown in Figure 4.8.
 4.1 Graphing Linear Inequalities 177

y The x-intercept of the boundary line 2x 1 y 5 1 is (21, 0) , and the y-in-

7 tercept is (0, 1) . Plugging in the point (0, 0) , we get
4x + y = 4 6
2(0) 1 (0) $ 1
5
4 0$1
3 Since the statement is false, we graph 2x 1 y 5 1 and place arrows on the side
−x + y = 1
2 of the line not containing the origin, as shown in Figure 4.8.
1 The solution region is not bounded above the line y 5 x 1 1 or above the
x
line y 5 24x 1 4. Consequently, the shaded solution region is unbounded.
−1 1
−1

FIGURE 4.8

EXAMPLE 4 Graphing a System of Linear Inequalities with an Empty

Solution Region
Graph the solution region of the system of linear inequalities.
22x 1 2y $ 6
2x 1 y # 1

SOLUTION We will graph the boundary lines by first rewriting them in slope-
intercept form.
Solving the first inequality for y, we get y $ x 1 3 and draw arrows point-
ing to the region above the line y 5 x 1 3. Solving the second inequality for y,
we get y # x 1 1 and draw arrows pointing to the region below the line
y 5 x 1 1 (see Figure 4.9).

y
11 y=x+3
10
9
8
7
6
5
4
3
2 y=x+1
1
0 x
0 1 2 3 4 5 6 7 8 9
FIGURE 4.9

Because the lines have the same slope (m 5 1) , they are parallel. Conse-
quently, the lines will never intersect. As seen in the graph, the two regions also
will never intersect; they will always be separated by the region between the two
lines. Therefore, this system of linear inequalities does not have a solution. That
is, there is no ordered pair (x, y) that can satisfy both inequalities simultaneously.
178 CHAPTER 4 Linear Programming

The corners of a solution region are called corner points. To find the coor-
dinates of each corner point, we solve the system of equations formed by the two
intersecting boundary lines that create the corner.

EXAMPLE 5 Finding the Corner Points of a Solution Region

Graph the solution region of the system of linear inequalities and determine the
coordinates of each corner point.
2x 1 y # 6
2x 1 y $ 0
x$0

SOLUTION We will graph the boundary line 2x 1 y 5 6 using its x- and

6
y-intercepts. The x-intercept is (3, 0) , since 5 3. The function has y-intercept
2
6
(0, 6) , since 5 6. Plugging in the point (0, 0) , we get
y 1
8 2(0) 1 (0) # 6
7 x=0
0#6
6
2x + y = 6
5 The statement is true, so we shade the side of the line containing the point (0, 0).
4 Solving the second inequality for y, we get y $ x, and we shade the region above
3 the line y 5 x. The third inequality restricts x to all positive values, so we shade
2 the region to the right of the y axis.
y=x
1
The triangular region shown in Figure 4.10 is the intersection of the three re-
x gions and is the solution region of the system of linear inequalities. From the
−1 − 1 1 2 3
−2
graph, it appears that the solution region has corner points at or near (0, 0) ,
(0, 6), and (2, 2). We will verify these results algebraically.
FIGURE 4.10 The coordinates of the first corner point may be found by solving the system
of equations formed by the boundary lines that make up the corner.
2x 1 y 5 0
x 50
Adding the first equation to the second equation yields y 5 0. Since the second
equation tells us that x 5 0, the coordinates of the first corner point are (0, 0) .
The coordinates of the second corner point may be found by solving the sys-
tem of equations formed by the boundary lines that make up the corner.
2x 1 y 5 6
x 50
The second equation tells us that x 5 0. Substituting this value of x into the first
equation yields
2x 1 y 5 6
2(0) 1 y 5 6 Substitute x 5 0
y56
The coordinates of the second corner point are (0, 6) .
 4.1 Graphing Linear Inequalities 179

The coordinates of the third corner point may be found by solving the sys-
tem of equations formed by the boundary lines that make up the corner.
2x 1 y 5 6
2x 1 y 5 0
The second equation may be rewritten as y 5 x. Substituting this result into the
first equation yields
2x 1 y 5 6
2x 1 (x) 5 6 Since y 5 x
3x 5 6
x52
Since x 5 y, the coordinates of the corner point are (2, 2) .

It may have seemed superfluous to calculate the coordinates of the corner

points algebraically in Example 5 when the coordinates were readily apparent
from the graph of the solution region. Despite the apparent redundancy of the pro-
cedure, it is a necessary step. Example 6 illustrates the hazards of relying solely
upon a graph for the coordinates of the corner points.

EXAMPLE 6 Finding the Corner Points of a Solution Region

Graph the solution region for the system of inequalities and determine the coor-
dinates of the corner points.
x1 y #5
25x 1 5y # 6
y$2

y SOLUTION The graph of the solution region is shown in Figure 4.11.

From the graph, it appears that the corner points of the region are at or near
x+y=5 (2, 3), (3, 2), and (0.75, 2) .
3 −5x + 5y = 6
The coordinates of the first corner point may be found by solving the system
of equations formed by the boundary lines that make up the corner.
2
x1 y 55
y=2
25x 1 5y 5 6
1
Adding five times the first equation to the second equation yields

0 0x 1 10y 5 31
x
0 1 2 3 4 5
y 5 3.1
FIGURE 4.11
Substituting this result back into the first equation yields
x1y55
x 1 (3.1) 5 5
x 5 1.9
The coordinates of the first corner point are (1.9, 3.1) . [From the graph, it looked
as if the corner point was (2, 3) .]
180 CHAPTER 4 Linear Programming

The coordinates of the second corner point may be found by solving the sys-
tem of equations formed by the boundary lines that make up the corner.
x1y 55
y52
Since the second equation tells us that y 5 2, we may substitute this value into
the first equation.
x1y55
x 1 (2) 5 5 Since y 5 2
x53
The coordinates of the second corner point are (3, 2) . Unlike the first corner
point, this result agrees with our graphical conclusion.
The coordinates of the third corner point may be found by solving the sys-
tem of equations formed by the boundary lines that make up the corner.
25x 1 5y 5 6
y52
Since the second equation tells us that y 5 2, we may substitute this value into
the first equation.
25x 1 5y 5 6
25x 1 5(2) 5 6 Since y 5 2
25x 1 10 5 6
25x 5 24
x 5 0.8
The coordinates of the third corner point are (0.8, 2) . [From the graph, it looked
as if the corner point was (0.75, 2) .]

Using Technology to Graph Linear Inequalities

Many graphing calculators can draw the graphs of linear inequalities, as detailed
in the following Technology Tip. Often, however, it is quicker to draw the graphs
by hand.

Graphing a System of Linear Inequalities

1. Enter the linear equations associated

with each inequality by using the Y
Editor. (We will use the system
y # 23x 1 6 and y $ 2x 1 4 for this
example.)

(Continued)
 4.1 Graphing Linear Inequalities 181

2. Move the cursor to the \ to the left of

Y1 and press ENTER repeatedly. This
will cycle through several graphing
options. We want to shade the region
below the line, so we will pick the lower
triangular option.

3. Move the cursor to the \ to the left of Y2

and press ENTER repeatedly. We want to
shade the region above the line, so we
will pick the upper triangular option.

4. Press GRAPH to draw each of the shaded

regions. The region with the crisscross
pattern is the solution region.

Linear Inequality Applications

Many real-life problems are subject to multiple constraints, such as budget,
staffing, resources, and so on. Yet even when subjected to these constraints, the
problems often have multiple solutions. As will be shown in Example 7, the lin-
ear inequality graphing techniques introduced earlier in this section may often be
used to find solutions to real-life problems.

EXAMPLE 7 Using a Linear System of Inequalities to Find the

Ideal Work Schedule
A student earns $8.00 per hour working in a campus computer lab and $10.50 per
hour delivering pizza. If he has only 30 hours per week to work and must earn at
least $252 during that period, how many hours can he spend at each job in order
to earn at least $252?

SOLUTION Let c be the number of hours the student works in the computer lab
and p be the number of hours he works delivering pizza. He can work at most
30 hours. This is represented by the inequality
c 1 p # 30 The maximum number of work hours is 30

The amount he earns working in the lab is 8.00c, and the amount of money he
earns delivering pizza is 10.50p. His total income must be at least $252. That is,
8c 1 10.5p $ 252 The minimum amount of income is $252
182 CHAPTER 4 Linear Programming

Solving the inequalities for p in terms of c, we get the following system of in-
equalities and its associated graph (Figure 4.12). (We add the restrictions
p $ 0 and c $ 0, since it doesn’t make sense to work a negative number of hours
at either job.)
p # 2c 1 30
16
p$2 c 1 24 (In decimal form, p $ 20.7619c 1 24 approximately)
21
p $ 0, c $ 0

p
(0, 30)
30
28
26
(0, 24)
24
22 p = −c + 30
20
18
16
14
12 p ≈ −0.7619c + 24
10
8
6 (25.2, 4.8)
4
2
0 c
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34
FIGURE 4.12

Every point of the solution region represents a combination of hours at the two
jobs that will result in earnings of at least $252.
The corner points of the solution region are (0, 24) , (0, 30) , and (25.2, 4.8) .
(The last point is found by calculating the intersection of the two boundary lines.)
Let’s calculate the student’s weekly earnings at the corner points and some of the
other points in the region (see Table 4.2).

TABLE 4.2
Weekly
Lab Hours Pizza Hours Earnings
Corner point 0 24 $252.00
Corner point 0 30 $315.00
Corner point 25.2 4.8 $252.00
Interior point 5 25 $302.50
Interior point 10 20 $290.00
Interior point 15 14 $267.00
 4.1 Graphing Linear Inequalities 183

The weekly earnings vary; however, in every case the number of work hours is
less than or equal to 30 hours and the earnings are greater than or equal to $252.

4.1 Summary

In this section, you learned how to graph linear inequalities. You discovered that
the solution to a system of linear inequalities is the intersection of the graphs of
the solution regions of the individual inequalities.

4.1 Exercises

In Exercises 1–10, graph the solution region of the 21. 2x 2 4y $ 16 22. 2x 2 2y $ 0

linear inequality. Then use the graph to determine if 9x 1 y # 24 3x 1 y # 4
point P is in the solution region. 23x 1 6y # 224 5x 2 y $ 5
1. 2x 1 y # 6; 2. 4x 1 y # 0; 23. 8x 2 y $ 3 24. 24x 1 y $ 2
P 5 (2, 4) P 5 (1, 1) x 1 2y # 11 22x 1 y $ 1
9x 1 y # 14 y$ 1
3. x 1 5y # 10; 4. 5x 1 6y # 30; x# 1
P 5 (0, 0) P 5 (0, 5)
25. 25x 1 y $ 0
5. 2 2x 1 4y $ 2 2; 6. x 2 y # 10; 2x 1 y # 4
P 5 (1, 2) P 5 (5, 25) y# 1
7. 5x 2 4y # 0; 8. 2 3x 2 3y # 9; x# 1
P 5 (1, 0) P 5 (2, 21)
In Exercises 26–30, set up the system of linear
9. 2x 2 y $ 8; 10. 7x 2 6y $ 12; inequalities that can be used to solve the problem. Then
P 5 (23, 2) P 5 (0, 21) graph the system of equations and solve the problem.

In Exercises 11–25, graph the solution region of the 26. Nutritional Content A 32-gram serving
system of linear equations. If there is no solution, of Skippy® Creamy Peanut Butter contains
explain why. 150 milligrams of sodium and 17 grams of fat. A
56-gram serving of Bumble Bee® Chunk Light
11. 24x 1 y $ 2 12. 25x 1 y $ 0 Tuna in Water contains 250 milligrams of sodium
22x 1 y $ 1 2x 1 y # 4 and 0.5 gram of fat. (Source: Product labeling.) Some
x#0 y$0 health professionals advise that a person on a 2500-
13. 22x 1 6y # 8 14. 10x 2 y $ 12 calorie diet should consume no more than 2400 mg
4x 2 12y # 26 9x 2 2y $ 2 of sodium and 80 grams of fat. Graph the region
showing all possible serving combinations of
15. 3x 2 2y # 4 16. 9x 2 6y # 0 peanut butter and tuna that a person could eat and
11x 2 20y $ 2 4x 1 5y # 23 still meet the dietary guidelines.
17. x 2 y # 25 18. 2x 1 5y # 2 27. Nutritional Content A Nature Valley®
9x 1 y # 25 3x 2 5y # 3 Strawberry Yogurt Chewy Granola Bar
19. 6x 1 2y # 10 20. x2 y$3 contains 130 milligrams of sodium and 3.5 grams
2x 2 2y $ 25 6x 1 7y # 44 of fat. A Nature’s Choice® Multigrain Strawberry
x$0 6x 2 7y # 16 Cereal Bar contains 65 milligrams of sodium and
y$0 1.5 grams of fat. (Source: Product labeling.) Some
184 CHAPTER 4 Linear Programming

health professionals advise that a person on a 2500- 2x 1 3y # 6

calorie diet should consume no more than 2400 mg 22x 1 4y $ 4
of sodium and 80 grams of fat. Graph the region
25x 1 y # 15
showing all possible serving combinations of
granola bars and cereal bars that a person could eat x# 5
and still meet the dietary guidelines. y$ 2
28. Student Wages A student earns $15.00 per hour 32. Graph the solution region of the system of linear
designing web pages and $9.00 per hour inequalities and identify the coordinates of the
supervising a campus tutoring center. She has at corner points.
most 30 hours per week to work, and she needs to 22x 1 y # 4
earn at least $300. Graph the region showing all
7x 1 2y $ 8
possible work-hour allocations that meet her time
and income requirements. x#0

29. Wages A salaried employee earns $900 per week 33. Graph the solution region of the system of linear
managing a copy center. He is required to work a inequalities and identify the coordinates of the
minimum of 35 hours but no more than 45 hours corner points.
weekly. As a side business, he earns $25 per hour 2x 1 y # 0
designing brochures for local business clients. In 2x 2 y $ 24
order to maintain his standard of living, he must
y$2
earn $1100 per week. In order to maintain his
quality of life, he limits his workload to 50 hours 34. Write a system of inequalities whose solution
per week. Given that he has no control over the region has the corner points (0, 0) , (1, 3) , (3, 5) ,
number of hours he has to work managing the and (2, 1) .
copy center, will he be able to consistently meet
35. Write a system of inequalities whose solution
his workload and income goals? Explain.
region has the corner points (1, 1) , (1, 3) , (5, 3) ,
30. Commodity Prices Today’s Market and (2, 1) .
Prices (www.todaymarket.com) is a daily
36. Write a system of inequalities whose unbounded
fruit and vegetable wholesale market price service.
solution region has the corner points (0, 5) ,
Produce retailers who subscribe to the service can
(2, 1) , and (5, 0) .
use the wholesale prices to aid them in setting
retail prices for the fruits and vegetables they sell. 37. Write a system of inequalities whose unbounded
A 25-pound carton of peaches holds 60 medium solution region has the corner points (0, 5) ,
peaches or 70 small peaches. In August 2002, the (4, 4) , and (5, 0) .
wholesale price for local peaches in Los Angeles
38. A student concludes that the corner points of a
was $9.00 per carton for medium peaches and
solution region defined by a system of linear
$10.00 per carton for small peaches. (Source:
inequalities are (0, 0) , (1, 1) , (0, 2) , and (2, 2) .
Today’s Market Prices.) A fruit vendor has budgeted
After looking at the graph of the region, the
up to $100 to spend on peaches. He estimates that
instructor immediately concludes that the student
weekly demand for peaches is at least 420 peaches
is incorrect. How did the instructor know?
but no more than 630 peaches. He wants to buy
enough peaches to meet the minimum estimated 39. Is it possible to have a bounded solution region
demand but no more than the maximum estimated with exactly one corner point? If so, give a system
demand. Graph the region showing which small- of inequalities whose solution region is bounded
and medium-size peach carton combinations meet and has exactly one corner point.
his demand and budget restrictions.
40. Is it possible to have an unbounded solution
region with exactly one corner point? If so, give a
Exercises 31– 40 are intended to challenge your
system of inequalities whose solution region is
understanding of the graphs of linear inequalities.
unbounded and has exactly one corner point.
31. Graph the solution region of the system of linear
inequalities and identify the coordinates of the
corner points.
 4.2 Solving Linear Programming Problems Graphically 185

4.2 Solving Linear Programming Problems Graphically

■ Determine the feasible

region of a linear Many products, such as printer ink, are sold to busi-
programming problem ness customers at a discount if large quantities are ordered. Profitable businesses
■ Solve linear and integer want to minimize their supply costs yet have sufficient ink on hand to fulfill their
programming problems in printing requirements. How much ink should they order? The process of acquiring,
two variables
producing, and distributing supplies can often be made more efficient by setting up
and solving systems of linear inequalities.
In this section, we show how a mathematical method called linear program-
ming can help businesses determine the most cost-effective way to manage their
resources. We will demonstrate how linear programming is used to optimize an ob-
jective function subject to a set of linear constraints. We will also reveal how to find
the whole-number solution of an integer programming problem. We will begin our
discussion with the following set of definitions.

LINEAR PROGRAMMING PROBLEM

A linear equation z 5 ax 1 by, called an objective function, may be maxi-
mized or minimized subject to a set of linear constraints of the form
cx 1 dy # f or cx 1 dy $ f
where x and y are variables (called decision variables) and a, b, c, d and f are
real numbers. A problem consisting of an objective function and a set of lin-
ear constraints is called a linear programming problem. The values of x and
y that optimize (maximize or minimize) the value of the objective function
are called the optimal solution. A linear programming problem with the ad-
ditional constraint that x and y are integers is called an integer program-
ming problem.

In Section 4.1, we gave the example of a student earning $10.50 an hour de-
livering pizza and $8.00 an hour working in a campus computer lab. He had only
30 hours per week to work, and he had to earn at least $252 during that period.
We let p be the number of hours he spent delivering pizza and c be the number of
hours he spent working in the computer lab. We had the constraints
c 1 p # 30 He can work at most 30 hours
8c 1 10.5p $ 252 He must earn at least $252
p $ 0, c $ 0 He must work a nonnegative number of hours at each job
which we rewrote as
p # 2c 1 30
16
p$2 c 1 24
21
p $ 0, c $ 0
186 CHAPTER 4 Linear Programming

From this scenario, we can set up linear programming problems to address

each of the following questions:
1. What is the largest amount of money he can earn?
Objective function: Maximize z 5 8c 1 10.5p Total amount earned
 p # 2c 1 30
 16
Subject to  p $ 2 21 c 1 24

 p $ 0, c $ 0
2. What is the least number of hours he can work?
Objective function: Minimize z5c1p Total hours worked

 p # 2c 1 30
16
Subject to  p $ 2 21c 1 24

 p $ 0, c $ 0
3. What is the maximum number of hours he can work in the computer lab?
Objective function: Maximize z5c Total computer lab hours
 p # 2c 1 30
 16
Subject to  p $ 2 21c 1 24

 p $ 0, c $ 0
Although each of the three objective functions has the same constraints, the
optimal solution to each linear programming problem will differ based upon the
objective function. However, all solutions will lie within the solution region of the
system of constraints. In the context of linear programming, we call the solution
region of the system of constraints the feasible region and the points within the
region feasible points. Which of all the feasible points will optimize each of the
objective functions? Testing all of the points in the feasible region would be an
impossible task! Fortunately, we don’t have to. The Fundamental Theorem of Lin-
ear Programming limits the number of points we have to test.

FUNDAMENTAL THEOREM OF LINEAR PROGRAMMING

1. If the solution to a linear programming problem exists, it will occur at a
corner point.
2. If two adjacent corner points are optimal solutions, then all points on the
line segment between them are also optimal solutions.
3. Linear programming problems with bounded feasible regions will always
have optimal solutions.
4. Linear programming problems with unbounded feasible regions may or
may not have optimal solutions.

Recall that the graph of the feasible region for the computer lab and pizza de-
livery problem had the corner points (0, 24) , (0, 30) , and (25.2, 4.8) , as shown
in Figure 4.13.
 4.2 Solving Linear Programming Problems Graphically 187

p
(0, 30)
30
28
26
(0, 24)
24
22 p = −c + 30
20
18
16
14
12 p ≈ −0.7619c + 24
10
8
6 (25.2, 4.8)
4
2
0 c
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34
FIGURE 4.13

The solution to each one of the linear programming problems will occur at one of
these points.
1. What is the largest amount of money he can earn?
Maximize z 5 8c 1 10.5p.
We evaluate the objective function at each corner point (see Table 4.3).

TABLE 4.3
Corner Point Objective Function
Total Earnings
Computer Lab Pizza Delivery (in dollars)
Hours (c) Hours ( p) (z 5 8c 1 10.5p)
0 24 252
0 30 315
25.2 4.8 252

The maximum value of z occurs at corner point (0, 30). The optimal solution
is c 5 0 and p 5 30. To maximize his earnings, he should work 0 hours in
the lab and 30 hours delivering pizza.
2. What is the least number of hours he can work?
Minimize z 5 c 1 p.
We evaluate the objective function at each corner point (see Table 4.4).
188 CHAPTER 4 Linear Programming

TABLE 4.4
Corner Point Objective Function
Computer Lab Pizza Delivery Total Work Hours
Hours (c) Hours (p) ( z 5 c 1 p)
0 24 24
0 30 30
25.2 4.8 30

The minimum value of z occurs at corner point (0, 24). The optimal solution
is c 5 0 and p 5 24. To minimize his work hours, he should work 0 hours
in the lab and 24 hours delivering pizza.
3. What is the maximum number of hours he can work in the computer lab?
Maximize z 5 c.
We evaluate the objective function at each corner point (see Table 4.5).

TABLE 4.5
Corner Point Objective Function
Computer Pizza Delivery Lab Work Hours
Lab Hours (c) Hours ( p) (z 5 c)
0 24 0
0 30 0
25.2 4.8 25.2

The maximum value of z occurs at corner point (25.2, 4.8). The optimal so-
lution is c 5 25.2 and p 5 4.8. The maximum amount of time he can work
in the computer lab is 25.2 hours. He will still have to work 4.8 hours deliv-
ering pizza to reach his earnings goal.
Why do we have to check only the corner points of the feasible region? We’ll
address this question by considering a “family” of objective functions. Suppose
we are asked to maximize the objective function z 5 3x 1 y subject to the fol-
lowing constraints:
4x 1 y # 12
2x 1 y # 8
x$0
y$0
The graph of the feasible region and the coordinates of the corner points are
shown in Figure 4.14.
 4.2 Solving Linear Programming Problems Graphically 189

12
4x + y = 12
10
(0, 8)
8

6
(2, 4)
4
2x + y = 8
2
(0, 0) (3, 0)
x
−1 1 2 3 4
−2

FIGURE 4.14

If we set the variable z in the objective function z 5 3x 1 y to a fixed value

c, then the graph of the line 3x 1 y 5 c will pass through all points (x, y) that
satisfy the equation 3x 1 y 5 c. If we repeat this for different values of c, we will
end up with a “family” of objective function lines. The lines will be parallel with
different y-intercepts. (In Figure 4.15, we set z equal to the following constant
values: 2, 4, 6, 8, 10, 12.)

12
z = 12
10
z = 10
8
z=8
6
z=6
4
z=4
2
z=2
x
−1 1 2 3 4
−2

FIGURE 4.15

Recall that we want to maximize the objective function subject to the con-
straints. Observe that although z 5 12 is the largest value of z shown in the fig-
ure, the line 3x 1 y 5 12 does not intersect the feasible region. Consequently, no
feasible solution lies on the line 3x 1 y 5 12.
The first line above the feasible region that intersects the feasible region is
the line 3x 1 y 5 10. This line intersects the feasible region at a single point: the
corner point (2, 4) . At this corner point, z 5 10. All other objective function lines
that cross the feasible region take on values of z less than 10. Therefore, the
objective function z 5 3x 1 y has an optimal solution at (2, 4) . At this point, the
190 CHAPTER 4 Linear Programming

objective function takes on its maximum value: z 5 10. This value of the objec-
tive function is referred to as the optimal value for the linear programming prob-
lem. Regardless of the objective function, the maximum (or minimum) value of
the objective function will occur at a corner point of the feasible region. An ar-
gument similar to that given here can be made for any objective function and any
feasible region.
The process of solving linear programming problems graphically is summa-
rized in the following box.

Graphical Method for Solving Linear

Programming (LP) Problems

1. Graph the feasible region determined by the constraints.

2. Find the corner points of the feasible region.
3. Find the value of the objective function at each of the corner points.
4. If the feasible region is bounded, the maximum or minimum value of the
objective function will occur at one of the corner points.
5. If the feasible region is an unbounded region in the first quadrant and the
coefficients of the objective function are positive, then the objective func-
tion has a minimum value at a corner point. The objective function will
not have a maximum value.

EXAMPLE 1 Solving a Linear Programming Problem Graphically

Solve the linear programming problem:
Maximize z 5 6x 1 2y
23x 1 y $ 2
 x 1 y # 10
Subject to 
x$0

 y $0
y
y = 3x + 2 SOLUTION We begin by solving each inequality for y.
12
(0, 10) 23x 1 y $ 2 x 1 y # 10
10
8 (2, 8) y $ 3x 1 2 y # 2x 1 10
6 Graphing the feasible region yields the graph shown in Figure 4.16.
4 y = −x + 10 The corner points occur where the boundary lines intersect. The corner points
of the feasible region are (0, 2), (0, 10), and (2, 8). Substituting each of these
2 (0, 2) points into the objective function, z 5 6x 1 2y, we get the results in Table 4.6.
x
−1 1 2 3
−2

FIGURE 4.16
 4.2 Solving Linear Programming Problems Graphically 191

TABLE 4.6

Corner Point Objective Function

x y z 5 6x 1 2y
0 2 4
0 10 20
2 8 28

The objective function z 5 6x 1 2y is maximized when x 5 2 and y 5 8. At that

point, z 5 28.

EXAMPLE 2 Solving an LP Problem with an Unbounded Feasible Region

Solve the linear programming problem:
Minimize z 5 2x 1 5y
 4x 1 y$ 4
2x 1 y$ 1
Subject to 
 x$ 0
 y$ 0

SOLUTION The objective function is

z 5 2x 1 5y
The constraints are
4x 1 y $ 4 x$0
2x 1 y $ 1 y$0
y We draw each boundary line and shade the feasible region, as shown in Fig-
7 ure 4.17.
6 The region is unbounded and has two corner points. Since the coefficients of
5 the objective function are both positive, the objective function will have a mini-
(0, 4) −x + y = 1 mum. The first corner point of the feasible region is (0, 4), the y-intercept of one
4
3
of the constraints. We determine the coordinates of the second corner point
by finding the intersection point of the boundary lines, y 5 x 1 1 and
2
(0.6, 1.6) y 5 24x 1 4.
1 4x + y = 4
24x 1 4 5 x 1 1 Solve the system of equations by the substitution method
x
−1 1 2 3 4 5 6
−1 25x 5 23

FIGURE 4.17
x 5 0.6
y 5 (0.6) 1 1 Substitute x 5 0.6 into y 5 x 1 1
y 5 1.6
The second corner point is (0.6, 1.6). Substituting each of the points into the
objective function, z 5 2x 1 5y, we obtain the values in Table 4.7.
192 CHAPTER 4 Linear Programming

TABLE 4.7
Corner Point Objective Function
x y z 5 2x 1 5y
0 4 20
0.6 1.6 9.2

The objective function has a minimum at (0.6, 1.6). At that point, z 5 9.2.

In Example 2, we had an unbounded feasible region. We were able to find a

solution because we were asked to minimize the objective function z 5 2x 1 5y.
If we had been asked to maximize the objective function, the problem would have
had no solution. For a linear programming problem with an unbounded feasible
region and an objective function with positive coefficients, no matter what “opti-
mal solution” we pick, we will always be able to find a point (x, y) in the feasi-
ble region that yields a greater “optimal value.”

EXAMPLE 3 Solving a Linear Programming Problem Graphically

Solve the linear programming problem:
Minimize z 5 25x 1 3y
 6x 1 y$ 6

 22x 1 y$ 1
Subject to  x# 2
 y$ 1

 x$ 0

SOLUTION After graphing each of the five boundary lines, we will use arrows to
show which side of the boundary line will be shaded, as shown in Figure 4.18. This
technique is especially helpful when working with a large number of constraints.

y
x=2
9
8
6x + y = 6
7
(0, 6)
6 −2x + y = 1
5 (2, 5)
4
x=0 3
(0.625, 2.25) y=1
2
1

x
−1 1 2 3
−1

FIGURE 4.18

We shade the region that has boundary lines with all arrows pointing to the inte-
rior of the region. Notice that the line y 5 1 does not form a boundary line of the
feasible region. This is okay so long as the feasible region is on the appropriate
 4.2 Solving Linear Programming Problems Graphically 193

side of y 5 1. Since the arrows on the line y 5 1 point toward the side that con-
tains the feasible region, the constraint y $ 1 is satisfied.
The first corner point (0, 6) is easily determined, since it is the y-intercept of
the boundary line 6x 1 y 5 6. The second corner point occurs at the intersection
of 6x 1 y 5 6 and 22x 1 y 5 1. We must solve the system of equations
6x 1 y 5 6
22x 1 y 5 1
Subtracting the second equation from the first equation yields
8x 5 5
x 5 0.625
To determine the value of y, we substitute the x value back into the equation
22x 1 y 5 1.
22x 1 y 5 1
22(0.625) 1 y 5 1 Since x 5 0.625
21.25 1 y 5 1
y 5 2.25
The second corner point is (0.625, 2.25) .
The third point occurs at the intersection of x 5 2 and 22x 1 y 5 1. The x
coordinate of the corner point is x 5 2. To determine the y coordinate, we sub-
stitute this result into 22x 1 y 5 1.
22x 1 y 5 1
22(2) 1 y 5 1 Since x 5 2
24 1 y 5 1
y55
The third corner point is (2, 5) .
With the corner points identified, we are ready to evaluate the objective func-
tion z 5 25x 1 3y at each corner point, as shown in Table 4.8.

TABLE 4.8
Corner Point Objective Function
x y z 5 25x 1 3y
0 6 18
0.625 2.25 3.625
2 5 5

Since we are looking for the minimum value of the objective function, the opti-
mal solution is (0.625, 2.25) and the optimal value is 3.625.

Real-Life Applications
As shown in Examples 4 and 5, many real-life problems can be analyzed using
the techniques of linear programming.
194 CHAPTER 4 Linear Programming

EXAMPLE 4 Using Linear Programming to Do Investment Analysis

Table 4.9 shows the average annual rate of return on two TIAA-CREF investment
accounts over a 10-year period.

TABLE 4.9
As of 6/30/04
CREF Variable
Annuity Accounts 10-Year Average
Bond Market 7.15%
Social Choice 10.31%
Source: www.tiaa-cref.com.

An investor wants to invest at least $3000 in the Bond Market and Social
Choice accounts. He assumes that he will be able to get a return equal to the
10-year average, and he wants the total return on his investment to be at least
9 percent. He assigns each share in the Bond Market account a risk rating of 2
and each share in the Social Choice account a risk rating of 4. The approximate
share price at the end of June 2004 was $74 per share for the Bond Market ac-
count and $102 per share for the Social Choice account. He will use these prices
in his analysis. How many shares of each account should he buy in order to mini-
mize his overall risk? [Note that fractions of shares may be purchased. Also, to
make computations easier, round each percentage to the nearest whole-number
percent (i.e., 10.31 percent 5 10 percent).]

SOLUTION Let x be the number of shares in the Bond Market account and y
be the number of shares in the Social Choice account. The rounded rate of return
for the Bond Market account is 7 percent, and that for the Social Choice account
is 10 percent. The share price for the Bond Market account is $74, and that for the
Social Choice account is $102. Each Bond Market share has a risk rating of 2, and
each Social Choice share has a risk rating of 4. We want to minimize the overall
risk. That is, we want to minimize z 5 2x 1 4y.
The amount of money invested in each account is equal to the share price
times the number of shares. The amount invested is 74x for the Bond Market ac-
count and 102y for the Social Choice account. The total amount invested is given
by the equation
74x 1 102y $ 3000 The total amount invested is at least $3000

The dollar amount of the return on the investment is equal to the product of
the rate and the amount invested. The return on the Bond Market account is
(0.07)(74x) 5 5.18x, and the return on the Social Choice account is
(0.10)(102y) 5 10.20y. Since we want to earn at least 9 percent on the total
amount of money invested (74x 1 102y) , the dollar amount of the minimum
combined return is (0.09)(74x 1 102y) 5 6.66x 1 9.18y. The combined return
is given by the equation
5.18x 1 10.20y $ 6.66x 1 9.18y The combined return is at least 9 percent of
the amount invested
21.48x 1 1.02y $ 0
 4.2 Solving Linear Programming Problems Graphically 195

Combining the objective function and each of the constraints yields the following
linear programming problem:
Minimize z 5 2x 1 4y

 74x 1 102y $ 3000 The total amount invested is at least
 $3000
 21.48x 1 1.02y $0 The combined return is at least 9 per-
Subject to 
 cent of the amount invested.
 x $0

 y $0
The graph of the feasible region is shown in Figure 4.19.

y
50
45 x=0
40
35 −1.48x + 1.02y = 0
(0, 29.412)
30
25
20 74x + 102y = 3000
15
10 (13.514, 19.608)
5 y=0
0 x
0 5 10 15 20 25 30 35 40
FIGURE 4.19

The first corner point is the y-intercept of the line 74x 1 102y 5 3000.
3000
Since < 29.412, the corner point is (0, 29.412) .
102
The second corner point is the point of intersection of 74x 1 102y 5 3000
and 21.48x 1 1.02y 5 0. We must solve the system of equations
74x 1 102y 5 3000
21.48x 1 1.02y 5 0
We will find the solution using matrices.

c d
74 102 3000
21.48 1.02 0

c d
74 102 3000
222 0 3000 R1 2 100R2

c d
0 306 6000 3R1 2 R2
222 0 3000

c d
0 1 19.608 1>306 R1
1 0 13.514 1>222 R2

c d
1 0 13.514 R2
0 1 19.608 R1

The second corner point is (13.514, 19.608) .

196 CHAPTER 4 Linear Programming

We evaluate the objective function z 5 2x 1 4y at each corner point, as

shown in Table 4.10.

TABLE 4.10
Corner Point Objective Function
x y z 5 2x 1 4y
0 29.412 117.648
13.514 19.608 105.460

The optimal solution is (13.514, 19.608) , and the optimal value is 105.460.
The risk is minimized when 13.514 shares of the Bond Market account and
19.608 shares of the Social Choice account are purchased.
Did the rounding of the percentages and share prices affect the solution? Yes.
However, since the 10-year average rate of return on each account was not guar-
anteed, the investor decided that rounding the percentages to whole-number per-
cents and rounding the share prices to whole-dollar amounts was good enough for
modeling purposes.

Integer Programming Problems

In many real-applications, we have the additional constraint that the objective
function input values must be whole numbers. Problems of this type are called in-
teger programming problems.

INTEGER PROGRAMMING PROBLEM

A linear programming problem with the additional constraint that the deci-
sion variables are integers is called an integer programming problem.

Integer programming problems containing many input variables and con-

straints can be very difficult to solve. Even in the two-decision-variable case,
solving these types of problems requires several extra steps. Although we won’t
require you to solve any of these types of problems in the exercises, we will work
an example here to illustrate the errors that may arise when the optimal solution
of a linear programming problem is rounded to whole-number values.

EXAMPLE 5 Using Linear Programming to Make Business Decisions

Marlborough Printer Supplies sells generic replacement ink cartridges for a vari-
ety of printers through its eBay store. Shipping is free when 10 or more cartridges
are ordered. In 2002, a single black ink cartridge for the Epson Color Stylus 660
printer cost $2.50, and a three-pack of black ink cartridges cost $6. (Source: eBay
online store.) The owner of a small business needs to purchase at least 20 ink car-
tridges and wants to minimize her overall cost. How many single cartridges and
how many three-packs should she buy?
 4.2 Solving Linear Programming Problems Graphically 197

t SOLUTION Let s be the number of single cartridges and t be the number of

8
three-packs. Since 10 or more cartridges will be ordered, shipping will be free.
7 (0, 6.67)
Consequently, the equation of the objective function (the cost function) is
6
5 s + 3t = 20 C 5 2.5s 1 6t
4
3 We have the constraints
2
(20, 0) s 1 3t $ 20 A minimum of 20 cartridges are ordered
1
s $ 0, t $ 0 The number of each type ordered is nonnegative
The feasible region is unbounded and has corner points A 0, 6 B and (20, 0) , as
s
−2 2 4 6 8 10 12 14 16 18 20 2
3
FIGURE 4.20 shown in Figure 4.20.
Evaluating the objective function C 5 2.5s 1 6t at the two corner points
yields the data in Table 4.11.

TABLE 4.11
Corner Point Objective Function
Single (s) Three-Pack (t) C 5 2.5s 1 6t
20 0 50
2
0 63 40

Since the smallest value of the objective function is 40, the optimal solution is
A 0, 6 23 B and the optimal value is 40. However, since we can’t order a fraction of a
three-pack, we must find the whole-number solution. Our natural tendency might
be to round the optimal solution to (0, 7) . However, doing so also alters the op-
timal value, as shown in Table 4.12.
t
8 Subproblem 2
7 TABLE 4.12
6 Rounded Optimal Solution Objective Function
5
4 Subproblem 1 Single (s) Three-Pack (t) C 5 2.5s 1 6t
3 0 7 42
2
1
s
Using whole-number values for s and t, is it possible to further reduce the cost?
−2 2 4 6 8 10 12 14 16 18 20 We will investigate this question by adding additional constraints. We create Sub-
problem 1 by adding the constraint t # 6 and Subproblem 2 by adding the con-
FIGURE 4.21 straint t $ 7, as shown in Figure 4.21. (These are the whole-number values on
2
either side of 6 .)
3
These new constraints split the feasible region into two separate regions. We
will solve Subproblem 1 first. The unbounded feasible region of Subproblem 1
has corner points (20, 0) and (2, 6).
198 CHAPTER 4 Linear Programming

TABLE 4.13
Corner Point of Subproblem 1 Objective Function
Single (s) Three-Pack (t) C 5 2.5s 1 6t
20 0 50
2 6 41

The optimal solution for Subproblem 1 is (2, 6), as shown in Table 4.13. Since
this is a whole-number solution, it makes sense in the context of the problem.
When two single cartridges and six three-packs are purchased, the total ink cost
is $41.
The feasible region of Subproblem 2 has the corner point (0, 7) (see
Table 4.14).

TABLE 4.14
Corner Point of Subproblem 2 Objective Function
Single (s) Three-Pack (t) C 5 2.5s 1 6t
0 7 42

The optimal solution for Subproblem 2 is (0, 7). Since this is also a whole-num-
ber solution, it makes sense in the context of the problem. When no single car-
tridges and seven three-packs are purchased, the total ink cost is $42.
The optimal whole-number solution for the entire linear programming prob-
lem will be the subproblem solution that yields the smallest value of the objective
function. The whole-number solution for Subproblem 1, (2, 6), had optimal value
$41. The whole-number solution for Subproblem 2, (0, 7), had optimal value $42.
Comparing the results of Subproblems 1 and 2, we conclude that the optimal
whole-number solution of the entire problem is (2, 6) . When two single car-
tridges and six three-packs are ordered, the overall cost is minimized. (It is im-
portant to note that if the business owner spent the extra dollar and ordered no
single cartridges and seven three-packs, she would get 21 cartridges instead of 20
cartridges. She may decide that the extra cartridge is worth the extra dollar.)

As shown in Example 5, we must be aware that rounding an optimal solution

to whole-number values does not guarantee that we have found the optimal
whole-number solution.

4.2 Summary

In this section, you learned how a mathematical method called linear program-
ming can help businesses determine the most cost-effective way to manage their
resources. You used linear programming to optimize an objective function subject
to a set of linear constraints.
 4.2 Solving Linear Programming Problems Graphically 199

4.2 Exercises

In Exercises 1–20, find the optimal solution and 8. Minimize z5x2y

optimal value of the linear programming problem. If a  24x 1 y $ 8
solution does not exist, explain why.  23x 1 y # 6
Subject to 
1. Minimize z 5 3x 1 7y  x$ 0
 4x 1 y $ 4  y$ 0
2x 1 y $ 1 9. Maximize z 5 22x 2 y
Subject to 
 x$0  24x 1 y $ 8
 y$0  23x 1 y # 6

2. Minimize z 5 6x 1 2y Subject to  x#4
 x$0
 6x 1 y $ 16 
22x 1 y $ 0  y$0
Subject to 
 x$0 10. Minimize z 5 5x 2 y
 y$0  23x 1 y $ 9
 22x 1 y # 6
3. Minimize z 5 9x 1 y 
Subject to  x# 3
 6x 1 y $ 16 
22x 1 y $ 0 
x$ 0
Subject to   y$ 0
 x$0
 y$0 11. Minimize z 5 22x 1 7y
4. Maximize z 5 9x 1 y  2x 1 4y $ 4
 2x 1 y $ 1
 6x 1 y # 16 Subject to 
22x 1 y # 0  x$0
Subject to   y$0
 x$0
 y$0 12. Minimize z 5 3x 1 5y
5. Maximize z 5 x 1 10y  6x 1 y $ 21

 6x 1 y # 16 Subject to  22x 1 y $ 1
22x 1 y # 0  x$3
Subject to   y$0
 x$0
 y$0 13. Minimize z 5 11x 1 9y
6. Minimize z 5 2x 2 5y  6x 1 y $ 16

 4x 1 y # 12 Subject to  22x 1 y $ 0
26x 1 2y # 24  x$0
Subject to   y$2
 x$0
 y$0 14. Maximize z 5 11x 1 9y
7. Maximize z 5 2x 2 5y  6x 1 y $ 16

 4x 1 y # 12 Subject to  22x 1 y $ 0
26x 1 2y # 24  x$0
Subject to   y$2
 x$0
 y$0
200 CHAPTER 4 Linear Programming

15. Maximize z 5 x 1 10y A 25-pound carton of peaches holds 60

 6x 1 y # 29 medium peaches or 70 small peaches. In August
 2002, the wholesale price for local peaches in Los
 22x 1 y # 23
Angeles was $9.00 per carton for medium peaches
Subject to  y#5
 and $10.00 per carton for small peaches. (Source:
 x$0
Today’s Market Prices.) A fruit vendor sells the
 y$0
medium peaches for $0.50 each and the small
16. Minimize z 5 20x 2 y peaches for $0.45 each. He estimates that weekly
 4x 1 y # 20 demand for peaches is at least 420 peaches but no
26x 1 2y # 40 more than 630 peaches. He wants to buy enough
Subject to  peaches to meet the minimum estimated demand,
 x$0
but no more than the maximum estimated
 y$0
demand. How many boxes of each size of peaches
17. Maximize z 5 20x 2 y should he buy if he wants to minimize his
 4x 1 y # 20 wholesale cost?
 26x 1 2y # 40
Subject to  22. Painkiller Costs An online drugstore
 x$0 sells Tylenol Extra Strength in a variety
 y$0 of bottle sizes. The 250-caplet bottle costs <
$15, and the 150-caplet bottle costs < $12.
18. Minimize z5x2y
(Source: www.drugstore.com.) A family wants to
 2x 1 y $ 0 order a supply of at least 750 caplets. How many
 3x 2 y # 0
Subject to  150-caplet bottles and how many 250-caplet
 x$1 bottles should the family order if it wants to
 y$1 minimize costs?
19. Maximize z 5 22x 1 y 23. Investment Choices The following
 24x 1 3y $ 0 table shows the average annual rate of
 return on a variety of TIAA-CREF investment
 23x 1 4y $ 21 accounts over a 10-year period.
Subject to  x#4

 x$0
As of 6/30/04
 y$2
CREF Variable
20. Minimize z5x1y Annuity Accounts 10-Year Average
 10x 1 2y $ 8
 Bond Market 7.15%
 220x 2 4y # 216
Subject to  x#4 Equity Index 11.34%

 x$0 Global Equities 7.24%
 y$0 Growth 8.56%
For Exercises 21–32, identify the objective function Money Market 4.22%
and constraints of the linear programming problem. Social Choice 10.31%
Then solve the problem and interpret the real-world
meaning of the results. Stock 9.97%
Source: www.tiaa-cref.com.
21. Minimum Commodity Cost Today’s
Market Prices (www.todaymarket.com) is An investor wants to invest at least $2000 in the
a daily fruit and vegetable wholesale market price Stock and Growth accounts. He assumes that he
service. Produce retailers who subscribe to the will be able to get a return equal to the 10-year
service can use wholesale prices to aid them in average, and he wants the total return on his
setting retail prices for the fruits and vegetables investment to be at least 9 percent. He assigns
they sell.
 4.2 Solving Linear Programming Problems Graphically 201

each share in the Stock account a risk rating of 6 The company assembles standard and deluxe
and each share in the Growth account a risk rating first aid kits for sale to consumers. A standard
of 7. The approximate share price at the end of first aid kit contains 40 plastic adhesive bandages
June 2004 was $174 per share for the Stock and 20 alcohol wipes. A deluxe first aid kit
account and $55 per share for the Growth contains 50 plastic adhesive bandages and
account. He will use these prices in his analysis. 28 alcohol wipes.
How many shares of each account should he buy The company makes a profit of $3 from each
in order to minimize his overall risk? [Note that standard first aid kit sold and $4 from each deluxe
fractions of shares may be purchased. Also, to first aid kit sold. Assuming that every kit
make computations easier, round each percentage produced will be sold, how many of each type of
to the nearest whole-number percent (i.e., for kit should the company assemble in order to
9.97 percent, use 10 percent).] maximize profit?
24. Investment Choices An investor wants 27. Furniture Production In June 2004, an
to invest at least $5000 in the Global online furniture retailer offered the
Equities and Equity Index accounts shown in following items at the indicated prices:
Exercise 23. She assumes that she will be able to Teak Double Rocker, $745
get a return equal to the 10-year average, and she Teak Tennis Bench, $124
wants the total return on her investment to be at (Source: www.outdoordecor.com.)
least 10 percent. She assigns each share in the Suppose that the number of hours required to
Global Equities account a risk rating of 6 and produce each item is as shown in the following
each share in the Equity Index account a risk table.
rating of 5. The approximate share price at the end
of June 2004 was $70 per share for the Global Cut Finish Package
Equities account and $72 per share for the Equity
Index account. She will use these prices in her Rocker 4 7 1
analysis. How many shares of each account Bench 1 3 1
should she buy in order to minimize her overall
risk? [Note that fractions of shares may be The company has a maximum of 360 labor hours
purchased. Also, to make computations easier, available in the Cutting Department, a maximum
round each percentage to the nearest whole- of 730 labor hours available in the Finishing
number percent (i.e., for 11.34 percent, use Department, and a maximum of 150 labor hours
11 percent).] available in the Packaging Department. Suppose
25. Pet Nutrition: Food Cost that the company makes a profit of $314 from the
PETsMART.com sold the following sale of each rocker and $57 from the sale of each
varieties of dog food in June 2003: bench. Assuming that all items produced are sold,
how many rockers and how many benches should
Nature’s Recipe Venison Meal & Rice Canine, the company produce in order to maximize profit?
20 percent protein, $21.99 per 20-pound bag
PETsMART Premier Oven Baked Lamb Recipe, 28. Furniture Production In June 2004, an
25 percent protein, $22.99 per 30-pound bag online furniture retailer offered the
(Source: www.petsmart.com.) following items at the indicated prices:
A dog breeder wants to make at least 300 Avalon Teak Armchair, $378
pounds of a mix containing at most 22 percent Teak Tennis Bench, $124
protein. How many bags of each dog food variety (Source: www.outdoordecor.com.)
should the breeder buy in order to minimize cost? Suppose that the number of hours required to
(Hint: Note that each bag is a different weight.) produce each item is as shown in the following table.
26. First Aid Kit Supplies Safetymax.com
sells first aid supplies to businesses. A Cut Finish Package
company that assembles first aid kits for Armchair 2 4 1
consumers purchases 3500 1s 3 3s plastic
Bench 1 3 1
adhesive bandages and 1800 alcohol wipes from
Safetymax.com.
202 CHAPTER 4 Linear Programming

The company has a maximum of 200 labor hours The school anticipates that 153 students will go
available in the Cutting Department, a maximum of on the trip. Each 10-passenger vehicle requires
480 labor hours available in the Finishing one chaperone, and each 57-passenger vehicle
Department, and a maximum of 150 labor hours requires six chaperones. (The chaperones will be
available in the Packaging Department. Suppose the traveling with the students.) At most 24
company makes a profit of $181 from the sale of chaperones are available to go on the trip. How
each armchair and $57 from the sale of each bench. many of each type of vehicle should the PTA
Assuming that all items produced are sold, how charter in order to minimize transportation costs?
many armchairs and how many benches should the
31. Television Advertising For the week of
company produce in order to maximize profit?
July 5-July 11, 2004, a national media
29. Transportation Costs A high school research company estimated that 14,834,000
PTA in southern Florida is planning an viewers watched CSI and 10,557,000 viewers
overnight trip to Orlando, Florida, for its watched Law and Order. (Source: www.nielsenmedia
graduating class. A Plus Transportation, a local .com.)
charter transportation company, offers the The amount of money a network can charge for
following rates (as of December 2003): advertising is based in part on the size of the
viewing audience. Suppose that a 30-second
Overnight Charter commercial running on CSI costs $3100 per spot
Rate (for 2 Days of and a 30-second commercial running on Law and
Vehicle Capacity Service) Order costs $2500 per spot. A beverage company
29 $1000 is willing to spend up to $87,000 for commercials
run during episodes of the two programs. The
49 $1800 company requires at least 10 spots to be run on
Source: www.buscharter.net. each program. How many spots on each program
should be purchased in order to maximize the
The school anticipates that 135 students will go number of viewers?
on the trip. Each 29-passenger vehicle requires
two chaperones, and each 49-passenger vehicle 32. Television Advertising For the week of
requires four chaperones. (The chaperones will be July 5-July 11, 2004, a national media
traveling with the students.) At most 16 research company estimated that 14,834,000
chaperones are available to go on the trip. How viewers watched CSI and 10,557,000 viewers
many of each type of vehicle should the PTA watched Law and Order. (Source: www
charter in order to minimize transportation costs? .nielsenmedia.com.)The amount of money a
network can charge for advertising is based in part
30. Transportation Costs A high school on the size of the viewing audience. Suppose that
PTA in southern Florida is planning an a 30-second commercial running on CSI costs
overnight trip to Orlando, Florida, for its $3100 per spot and a 30-second commercial
graduating class. A Plus Transportation, a local running on Law and Order costs $2500 per spot.
charter transportation company, offers the An athletic gear company is willing to spend up to
following rates (as of December 2003). $68,500 for commercials run during episodes of
the two programs. The company requires at least
Overnight Charter 10 spots to be run on CSI and at least 15 spots on
Rate (for 2 Days of Law and Order. How many spots on each
Vehicle Capacity Service) program should be purchased in order to
10 $625 maximize the number of viewers?
57 $1995
Source: www.buscharter.net.
 4.2 Solving Linear Programming Problems Graphically 203

For Exercises 33 –35, use the following data for maximizing total dividends. How many shares of
publicly traded recreational vehicle companies. The each company’s stock should the investor buy?
information was accurate as of July 16, 2004. (Assume that portions of shares may be
purchased.)
Earnings/ Dividend/
Share Price Share Share Exercises 36–40 are intended to challenge your
Company (dollars) (dollars) (dollars) understanding of linear programming.
Harley- 36. Consider the following linear programming
Davidson, problem.
Inc. (HDI) 62.70 2.56 0.40 Maximize P 5 3x 1 4y
Polaris  x1y #2

Industries Subject to  3x 1 5y #9
Inc. (PII) 50.50 2.50 0.92 
 x $ 0, y $ 0
Winnebago
Can a whole-number solution to a linear
Industries
programming problem be obtained by simply
Inc. (WGO) 33.42 1.80 0.20
rounding the noninteger solution to whole-number
Source: moneycentral.msn.com. values? Explain.

33. Investment Choices An investor has 37. Is it possible for a corner point (a, b) to
up to $4000 to invest in Harley- simultaneously minimize and maximize an
Davidson, Inc., and Polaris Industries, Inc. The objective function? If yes, give an example.
investor wants to earn at least $60 in dividends 38. Give an example of a linear programming
while maximizing total earnings. How many problem that does not have a solution.
shares of each company’s stock should the
investor buy? (Assume that portions of shares 39. Is it possible for the feasible region of a linear
may be purchased.) programming problem not to have any corner
points? If yes, give an example.
34. Investment Choices An investor has
up to $10,000 to invest in Harley- 40. Given the following linear programming problem,
Davidson, Inc., and Winnebago, Inc. The investor what is the maximum possible number of corner
wants to earn at least $50 in dividends while points the feasible region could have?
maximizing total earnings. How many shares of Maximize P5x1y
each company’s stock should the investor buy?  2ax 1 by # c
(Assume that portions of shares may be 
Subject to  2dx 1 fy # g
purchased.) 
 x $ 0, y $ 0
35. Investment Choices An investor has
Assume a, b, c, d, f, and g are positive.
up to $17,000 to invest in Polaris
Industries, Inc., and Winnebago, Inc. The investor
wants to have total earnings of at least $900 while
204 CHAPTER 4 Linear Programming

4.3 Solving Standard Maximization Problems

with the Simplex Method

■ Apply the simplex method

to solve multivariable Suppose you are a wholesale fruit buyer for a gro-
standard maximization cery store. You have been given the assignment to purchase five varieties of apples,
problems two varieties of peaches, and three varieties of pears. The grocery store has known
budget and space constraints. You are asked to maximize the number of pieces of
fruit purchased. This linear programming problem has 10 decision variables and can-
not be solved graphically. Although the graphical method works well for linear pro-
gramming problems with two decision variables, it does not work for problems
containing more than two decision variables. In this section, we will introduce an al-
ternative method that may be used to solve linear programming problems with any
number of decision variables.
In 1947, George B. Dantzig developed the simplex method to solve linear pro-
gramming problems. His method has been used to solve linear programming prob-
lems with hundreds of decision variables and hundreds of constraints. We will use
the method to solve standard maximization problems.

STANDARD MAXIMIZATION PROBLEM

A standard maximization problem is a linear programming problem with
an objective function that is to be maximized. The objective function is of
the form
P 5 ax 1 by 1 cz 1 # # #
where a, b, c, . . . are real numbers and x, y, z, . . . are decision variables.
The decision variables are constrained to nonnegative values. Additional
constraints are of the form
Ax 1 By 1 Cz 1 # # # # M
where A, B, C, . . . are real numbers and M is nonnegative.

Observe that substituting x 5 0, y 5 0, z 5 0, . . . into each constraint in-

equality Ax 1 By 1 Cz 1 # # # # M yields A(0) 1 B(0) 1 C(0) 1 # # # # M ,
which simplifies to 0 # M . Since the inequality 0 # M is valid for all non-
negative values of M, the origin (0, 0,. . . , 0) is contained in the solution region
of each constraint and, consequently, in the feasible region of the standard
maximization problem. Furthermore, if at least one of the constraints
Ax 1 By 1 Cz 1 # # # # M has all nonnegative coefficients, we are guaranteed
that the feasible region is bounded. For the vast majority of our real-life applica-
tions, the constraints will have nonnegative coefficients.
Graphically speaking, the simplex method starts at the origin and moves
from corner point to corner point of the feasible region, each time increasing the
 4.3 Solving Standard Maximization Problems with the Simplex Method 205

value of the objective function until it attains its maximum value. Remarkably, the
method doesn’t require that all corner points be tested, a fact that is appreciated
by those solving linear programming problems with hundreds of corner points.
Although the simplex method may be used for linear programming problems with
any number of decision variables, we will restrict our use to problems with four
variables or less. Larger problems are typically solved using computers.
We will introduce the simplex method with a two-variable example before
formally listing the steps of the method.
Suppose we are asked to use the simplex method to maximize P 5 6x 1 2y
subject to the constraints
y 2x 1 y # 10
12 x=0
11 x1y#8
10
9 (0, 8) x$0
8 2x + y = 10
7 y$0
6
5 (2, 6) Since the decision variables are nonnegative and additional constraints are of
4 x+y=8 the form Ax 1 By 1 Cz 1 # # # # M with M nonnegative, this linear program-
3 ming problem is a standard maximization problem. Consequently, we will be able
2 y=0
1 (0, 0) (5, 0)
to maximize the objective function using the simplex method. Although this is not
x part of the simplex method, we will draw the graph of the feasible region to give
−1 1 2 3 4 5 6 7 8 9
you a frame of reference as we work through the problem. The feasible region has
FIGURE 4.22 corner points (0, 0), (0, 8), (2, 6), and (5, 0) and is shown in Figure 4.22.
The value of the objective function, P 5 6x 1 2y, varies depending upon the
corner point. Substituting each corner point (x, y) into P 5 6x 1 2y and solving
for P yields Table 4.15.

TABLE 4.15
Objective Function
Corner Point Value
x y P 5 6x 1 2y
0 0 0
0 8 16
2 6 24
5 0 30

We see that the objective function P 5 6x 1 2y attains a maximum value of

30 at the corner point (5, 0) . Using the simplex method to solve the standard
maximization problem will yield the same result.
Recall that we are to solve the following standard maximization problem:
Maximize P 5 6x 1 2y
 2x 1 y# 10
 x1 y# 8
Subject to 
 x$ 0
 y$ 0
The first step of the simplex method requires us to convert each of the con-
straint inequalities into an equation. Since 2x 1 y # 10, there is some value
s $ 0 such that 2x 1 y 1 s 5 10. The variable s is called a slack variable
206 CHAPTER 4 Linear Programming

because it “takes up the slack.” That is, s adds in whatever value is necessary to
make the left-hand side of the equation equal 10. The value of s will vary de-
pending upon the value of x and y (see Table 4.16).

TABLE 4.16
Corner Point Slack Variable Value
x y s 5 10 2 2x 2 y
0 0 10
0 8 2
2 6 0
5 0 0

For the constraint inequality x 1 y # 8, we observe that there is some value

t such that x 1 y 1 t 5 8. Again, the value of t will vary depending upon the
value of x and y (see Table 4.17).

TABLE 4.17
Corner Point Slack Variable Value
x y t582x2y
0 0 8
0 8 0
2 6 0
5 0 3

We rewrite the objective function equation P 5 6x 1 2y by moving all terms

to the left-hand side of the equation, which yields the equation 26x 2 2y 1 P 5 0.
The system of equations representing the linear programming problem created by
combining the constraint equations and the objective function equation is
2x 1 y 1 s 5 10 Constraint 1
x1 y1 t58 Constraint 2
26x 2 2y 1 P 5 0 Objective function
This is a system of three equations and five unknowns (x, y, s, t, P). Since the
number of unknowns exceeds the number of equations, there will be more than
one solution to the system, if a solution exists.
We can represent the system of equations using matrix notation. The aug-
mented matrix is called the initial simplex tableau.
x y s t P
2 1 1 0 0 10 Constraint 1
£ 1 1 0 1 0 8§ Constraint 2
26 22 0 0 1 0 Objective function
The vertical and horizontal lines in the tableau are used to separate the bot-
tom row and rightmost column from the rest of the matrix. This separation will
become important when we calculate test ratios (as will be explained later).
 4.3 Solving Standard Maximization Problems with the Simplex Method 207

The variables of the columns of the simplex tableau that contain exactly one
nonzero entry are called active (basic) variables. The variables of the columns
that contain more than one nonzero entry are called inactive (nonbasic)
variables. For the initial tableau, the variables s, t, and P are the active variables,
since their corresponding columns in the tableau contain exactly one nonzero
value. The variables x and y are the inactive variables in the initial simplex
tableau, since their corresponding columns contain more than one nonzero value.
x y s t P
2 1 1 0 0 10 Constraint 1
£ 1 1 0 1 0 8§ Constraint 2
26 22 0 0 1 0 Objective function
The basic feasible solution of the tableau always corresponds with the origin.
The solution is obtained by substituting x 5 0 and y 5 0 into the equations that
generated the tableau.
2(0) 1 (0) 1 s 5 10 Constraint 1
(0) 1 ( 0) 1 t 5 8 Constraint 2
26(0) 2 2(0) 1 P 5 0 Objective function
Simplifying yields
s 5 10 Constraint 1
t58 Constraint 2
P50 Objective function
These results are summarized in Table 4.18.

TABLE 4.18
Objective
Corner Point Function Value Slack Variable Value
x y P 5 6x 1 2y s t
0 0 0 10 8

The objective function P 5 6x 1 2y has a value of 0 at the corner point (0, 0) .

An alternative and quicker approach to finding the solution may be obtained
from the initial tableau itself by crossing out each inactive variable column and read-
ing the result from the tableau. Crossing out an inactive variable column is equiva-
lent to setting the column variable to zero. For the initial simplex tableau, we have
x y s t P
2 1 1 0 0 10 Constraint 1
£ 1 1 0 1 0 8§ Constraint 2
26 22 0 0 1 0 Objective function
The first row of the resultant matrix is equivalent to
1s 1 0t 1 0P 5 10
s 5 10
208 CHAPTER 4 Linear Programming

The second row is equivalent to

0s 1 1t 1 0P 5 8
t58
The bottom row is equivalent to
0s 1 0t 1 1P 5 0
P50
These results are the same as those shown in Table 4.18. Clearly, the corner point
(0, 0) does not maximize the value of the objective function.
How can we make P as large as possible? The coefficients on the x and y
terms of the objective function P 5 6x 1 2y give us a clue. Since the coefficient
on the x term is 6, each 1-unit increase in x will result in a 6-unit increase in P.
Similarly, since the coefficient on the y term is 2, each 1-unit increase in y will
result in a 2-unit increase in P. Consequently, our initial strategy is to make the x
term as big as possible, since it contributes the most to the value of P. The pivot-
ing process detailed next will achieve this objective.
We will now use row operations to convert the initial simplex tableau into a
new tableau. We will first select a pivot column. Then, using row operations, we
will make the pivot column look like one of the columns of the 3 3 3 identity ma-
trix. To choose the pivot column, we find the negative number in the bottom row
that is furthest away from zero. The corresponding column is the pivot column. In
this tableau, the pivot column will be the x column, since 26 is further away from
0 than 22.
x y s t P
2 1 1 0 0 10
£ 1 1 0 1 0 8§
26 22 0 0 1 0
n
Pivot column

Why do we pick the column with the negative number that has the largest
magnitude? Recall that we are trying to maximize the objective function
P 5 6x 1 2y. As was pointed out earlier, the value of x contributes more to P
than the value of y because of the differences in their coefficients. Picking the col-
umn with the negative number of the largest magnitude guarantees that we are
making the variable that contributes the most to the value of P as large as possible.
The pivot is the nonzero entry in the pivot column that will remain after we
zero out the other column entries. It must be a positive number. If the pivot col-
umn contains more than one positive number, we will select the pivot by using a
test ratio. To calculate the test ratio, we divide the rightmost entry of each row
(excluding the bottom row) by the positive entry in the pivot column of that row.
x y s t P
2 1 1 0 0 10 10>2 5 5 d Test ratio for first row
£ 1 1 0 1 0 8§ 8>1 5 8 d Test ratio for second row
26 22 0 0 1 0
n
Pivot column
 4.3 Solving Standard Maximization Problems with the Simplex Method 209

Since the x column is the pivot column, these test ratios correspond geometrically
with the x-intercepts of the constraints (see Figure 4.23).

y
x=0
10
2x + y = 10
8

6 Corresponds with x+y=8

test ratio of 5

4
Corresponds with
test ratio of 8
2 y=0
(5, 0) (8, 0)
0 x
0 1 2 3 4 5 6 7 8 9
FIGURE 4.23

We select as the pivot the pivot-column entry that has the smallest test ratio.
That is, we pick the x-intercept on the graph that is closest to the origin to be the
pivot. This ensures that we remain in the feasible region. In this case, the first row
of the tableau had the smallest test ratio. Consequently, the pivot will be the 2 in
the first row and first column.
Pivot
mx y s t P
Pivot row k 2 1 1 0 0 10 10>2 5 5 d Test ratio for first row
£ 1 1 0 1 0 8§ 8>1 5 8 d Test ratio for second row
26 22 0 0 1 0
n
Pivot column

Why do we calculate test ratios only for positive column entries? When we
calculate test ratios for the x column, we are actually determining the values of
the x-intercepts of the constraints. If the column entry is a negative number, the
x-intercept of the constraint will be to the left of the origin. However, we are con-
strained to nonnegative values for both x and y. Therefore, determining the value
of the x-intercepts is necessary only for positive values. (If the column entry is
zero, the corresponding line is horizontal and doesn’t have an x-intercept.)
Recall that the pivot for this tableau is the 2 in the x column and the first row.
x y s t P
2 1 1 0 0 10
£ 1 1 0 1 0 8§
26 22 0 0 1 0
We will use the pivot to zero out the remaining entries in the column. However,
to ensure that all active variables remain nonnegative, row operations must be of
the form
aRc 6 bRP S Rc with a . 0 and b . 0
210 CHAPTER 4 Linear Programming

where Rc is the row to be changed and Rp is the row containing the pivot.
Using the row operations
2R2 2 R1 S R2
and
R3 1 3R1 S R3
we get the new tableau
x y s t P
2 1 1 0 0 10
£0 1 21 2 0 6§ 2R2 2 R1
0 1 3 0 1 30 R3 1 3R1

Columns x and t almost look like columns of the identity matrix; however, their
nonzero entry is a 2 instead of a 1. Using the row operations
1
R S R1
2 1
and
1
R S R2
2 2
we get
x y s t P
 1 1 
1 2 2
0 0 5 1>2 R1
 1 1 
0 22 1 0 3 1>2 R2
 2 
0 1 3 0 1 30 
The three columns that look like columns from the 3 3 3 identity matrix are x, t,
and P. Each of these columns has exactly one nonzero entry, so their correspon-
ding variables are the active variables. The remaining columns, y and s, have more
than one nonzero entry, so their corresponding variables are the inactive variables.
The system of equations represented by the tableau is
1 1
x1 y1 s55
2 2
1 1
y2 s1t53
2 2
1y 1 3s 1 P 5 30
Setting the inactive variables y and s to zero yields the system of equations
1 1
x1 (0) 1 (0) 5 5
2 2
1 1
(0) 2 (0) 1 t 5 3
2 2
1(0) 1 3(0) 1 P 5 30
which simplifies to x 5 5, t 5 3, and P 5 30. The inactive variables y and s are
both zero. As noted earlier, a quick way to see the values of the active variables
is to cross out the columns of the inactive variables and read the resultant values
 4.3 Solving Standard Maximization Problems with the Simplex Method 211

of the active variables directly from the tableau.

x y s t P
 1 1 
1 2 2
0 0 5
 1 1 
0 22 1 0 3
2
 
0 1 3 0 1 30 
Reading from the tableau, we have
x55
t53
P 5 30
The inactive variables y and s are both equal to zero. Since x 5 5 and y 5 0, this
solution corresponds with the corner point (5, 0) of the feasible region. At (5, 0),
P 5 30.
If there were negative entries in the bottom row of the tableau, we would select
a new pivot column and repeat the process. However, since the bottom row contains
only nonnegative entries, we are done. The objective function is maximized at
(5, 0). This corroborates our earlier conclusion, which was based on Table 4.15.
The simplex method is not intuitive and takes some effort to learn. We will
summarize the steps of the process and do several more examples to help you
master the method.

The Simplex Method

1. Verify that the linear programming problem is a standard maximization

problem.
2. Add slack variables to convert the constraints into linear equations.
3. Rewrite the objective function in the form ax 1 by 1 # # # 1 P 5 0,
where a, b, . . . are real numbers and x, y, . . . are decision variables. (Make
sure the coefficient on P is positive one.)
4. Set up the initial tableau from the system of equations generated in Steps
2 and 3. Be sure to put the objective function equation from Step 3 in the
bottom row of the tableau.
5. Select the pivot column by identifying the negative entry in the bottom
row with the largest magnitude.
6. Select the pivot by calculating test ratios for the positive entries in the
pivot column and choosing the entry with the smallest test ratio. The test
ratio of a row is calculated by dividing the last entry of the row by the
entry in the pivot column of the row.
7. Use the pivot to zero out the remaining entries in the pivot column. All
row operations must be of the form aRc 6 bRP S Rc with a . 0 and
b . 0, where Rc is the row to be changed and Rp is the pivot row. As
needed, multiply a row by a positive number to convert the pivot to a 1.
8. If the bottom row contains all nonnegative entries, cross out the columns
of the inactive variables and read the solution from the tableau. If the bot-
tom row contains negative entries, return to Step 5 and repeat the process
for the new tableau.
212 CHAPTER 4 Linear Programming

EXAMPLE 1 Using the Simplex Method to Solve a Standard

Maximization Problem
Maximize P 5 4x 1 3y subject to
3x 1 2y # 12
x1y#5
x $ 0, y $ 0

SOLUTION We confirm that the problem is a standard maximization problem.

Converting the linear programming problem into a system of linear equations,
we get
3x 1 2y 1 s 5 12
x1 y1 t55
24x 2 3y 1 P 5 0
and the initial simplex tableau
x y s t P
3 2 1 0 0 12
£ 1 1 0 1 0 5§
24 23 0 0 1 0
Since 24 is the negative number in the bottom row that has the largest magnitude,
we select the x column as the pivot column. Because the entries in the first and sec-
ond rows of the pivot column are positive, we must calculate the test ratios.
x y s t P
3 2 1 0 0 12 12>3 5 4
£ 1 1 0 1 0 5§ 5>1 5 5
24 23 0 0 1 0
n
Pivot column

Since the x column is the pivot column, the test ratios give us the x-intercepts
of the constraint equations, as shown in Figure 4.24.

y
x=0
6

5 3x + 2y = 12

4
Corresponds with
test ratio of 4
3 x+y=5

2 Corresponds with
test ratio of 5
1 y=0

(4, 0) (5, 0)
0 x
0 1 2 3 4 5
FIGURE 4.24
 4.3 Solving Standard Maximization Problems with the Simplex Method 213

By picking the column entry with the smallest test ratio, we guarantee that
our computations will yield a result that is in the feasible region. The pivot entry
is the 3 in the first row and the x column. We zero out the remaining entries in the
pivot column with the indicated operations. The resultant tableau is
x y s t P
3 2 1 0 0 12
£0 1 21 3 0 3§ 3R2 2 R1
0 21 4 0 3 48 3R3 1 4R1

Further simplifying the tableau yields

x y s t P
 2 1  1
1 3 3
0 0 4 R
3 1
 1 1  1
0 23 1 0 1 R
3 2
3
 0 21 16 
4 1
0 1 R
 3 3  3 3

We cross out the columns of the inactive variables and read the result:
x y s t P
 2 1 
1 3 3
0 0 4
 1 1 
0 23 1 0 1
3
0 16
1 4
23 0 1
 3 
x 5 4, t 5 1, and P 5 16. The inactive variables y and s are all zero. The corner
point (4, 0) yields an objective function value of P 5 16.
Observe that the fractional entries in the inactive variable columns of the
tableau were eliminated when we set the inactive variables to zero. To avoid
spending the time calculating fractional entries that will be eliminated anyway, an
alternative approach is to cross out the inactive variable columns before convert-
ing any active variable column entries to a 1. We then read the resultant equations
from the tableau and solve.
x y s t P
3 2 1 0 0 12
£0 1 21 3 0 3§
0 21 4 0 3 48
3x 5 12 3t 5 3 3P 5 48
x54 t51 P 5 16
Observe that this approach yields the same result and requires fewer computations.
The bottom row of the tableau is equivalent to the equation
2y 1 4s 1 3P 5 48
3P 5 y 2 4s 1 48
1 4
P5 y 2 s 1 16
3 3
214 CHAPTER 4 Linear Programming

Observe that since the coefficient on the y term is positive, increasing the value
of y while leaving the value of s unchanged will yield a value of P greater than
16. Consequently, the tableau
x y s t P
3 2 1 0 0 12
£0 1 21 3 0 3§
0 21 4 0 3 48
is not the final simplex tableau. The negative value in the bottom row indicates
that the value of P may be increased.
We will repeat the simplex method for this new tableau.
x y s t P
3 2 1 0 0 12
£0 1 21 3 0 3§
0 21 4 0 3 48
n
New pivot
column

The y column is the pivot column, since it is the column with a negative entry in
the bottom row. Because the entries in the first and second rows of the pivot col-
umn are positive, we must calculate the test ratios.
x y s t P
3 2 1 0 0 12 12>2 5 6
£0 1 21 3 0 3§ 3>1 5 3
0 21 4 0 3 48
The 1 in the second row is the pivot. We simplify the tableau using the indicated
row operations.
x y s t P
3 0 3 26 0 6 R1 2 2R2
£0 1 21 3 0 3§
0 0 3 3 3 51 R3 1 R2
3x 5 6 y53 3P 5 51
x52 P 5 17
Because the bottom row of the tableau is entirely nonnegative, we do not need to
repeat the pivoting process. The corner point (2, 3) yields a maximum objective
function value of P 5 17.
Observe that each new simplex tableau corresponds with a different corner
point. The initial simplex tableau corresponded with (0, 0) ; the second tableau,
with (4, 0) ; and the final tableau, with (2, 3) , as shown in Figure 4.25.
 4.3 Solving Standard Maximization Problems with the Simplex Method 215

y
x=0
6
3x + 2y = 12
5
Corner point for
final tableau
4
(2, 3)
3 x+y=5 Corner point for
second tableau
2 Corner point
for initial
1 simplex tableau y=0
(0, 0) (4, 0)
0 x
0 1 2 3 4 5
FIGURE 4.25

Graphically speaking, the simplex method moves from a corner point to an

adjacent corner point, each time increasing the value of the objective function.
We can verify the accuracy of our result by calculating the value of the ob-
jective function at each corner point (see Table 4.19).

TABLE 4.19
Corner Points Objective Function
x y P 5 4x 1 3y
0 0 0
4 0 16
2 3 17
0 5 15

EXAMPLE 2 Using the Simplex Method to Make Business Decisions

A store can buy at most 60 packages of three different brands of cheese. The first
brand costs $2 per package, the second brand costs $3 per package, and the third
brand costs $4 per package. The store can spend up to $120 on cheese. The first
brand resells for $5 per package, the second brand resells for $6 per package, and
the third brand resells for $8 per package. How many packages of each brand of
cheese should the store buy in order to maximize profit? (Note: Profit per
item 5 revenue per item 2 cost per item.)

SOLUTION Let x be the number of packages of the first brand, y be the num-
ber of packages of the second brand, and z be the number of packages of the third
brand. The total revenue from cheese sales is R 5 5x 1 6y 1 8z. The total cost
of the cheese is C 5 2x 1 3y 1 4z. Since profit is the difference between rev-
enue and cost,
P5R2C
5 (5x 1 6y 1 8z) 2 (2x 1 3y 1 4z)
5 3x 1 3y 1 4z
216 CHAPTER 4 Linear Programming

Since we want to maximize profit, we must maximize P 5 3x 1 3y 1 4z subject to

x 1 y 1 z # 60 There are at most 60 packages
2x 1 3y 1 4z # 120 At most, $120 is spent on cheese
x $ 0, y $ 0, z $ 0 Nonnegative number of each type of cheese is purchased
The problem is a standard maximization problem. Converting the linear program-
ming problem into a system of linear equations, we get
x 1 y 1 z 1 s 5 60 Package quantity constraint
2x 1 3y 1 4z 1 t 5 120 Budget constraint
23x 2 3y 2 4z 1 P 5 0 Objective function
and the initial simplex tableau
x y z s t P
1 1 1 1 0 0 60
£ 2 3 4 0 1 0 120 §
23 23 24 0 0 1 0
n
Pivot column

Since 24 is the negative number in the bottom row that has the largest magni-
tude, the z column is the pivot column. Because the entries in the first and second
rows of the pivot column are positive, we must calculate the test ratios.
x y z s t P
1 1 1 1 0 0 60 60>1 5 60
£ 2 3 4 0 1 0 120 § 120>4 5 30
23 23 24 0 0 1 0
The pivot entry is 4, since the second row had the smallest test ratio. We zero out
the remaining entries in the pivot column with the indicated row operations. The
resultant tableau is
x y z s t P
2 1 0 4 21 0 120 4R1 2 R2
£ 2 3 4 0 1 0 120 §
21 0 0 0 1 1 120 R3 1 R2

We could convert the nonzero entries in the active columns to a 1 by multiplying

1
Rows 1 and 2 by 4; however, that would result in messy fractional entries in the
tableau. We will hold off on the reduction until we have obtained the final sim-
plex tableau. Since there is still a negative value in the bottom row, we must again
select a pivot column and entry. The pivot column x has two positive entries, so
we must compute the test ratios.
x y z s t P
2 1 0 4 21 0 120 120>2 5 60
£ 2 3 4 0 1 0 120 § 120>2 5 60
21 0 0 0 1 1 120
n
Pivot column
 4.3 Solving Standard Maximization Problems with the Simplex Method 217

Since both entries have the same test ratio, we may pick either entry to be the
pivot. We select the entry in Row 1 as the pivot and zero out the remaining entries
in the pivot column with the indicated row operations. The resultant tableau is
x y z s t P
2 1 0 4 21 0 120
£0 2 4 24 2 0 0§ R2 2 R1
0 1 0 4 1 2 360 2R3 1 R1
Since the bottom row of the tableau is entirely nonnegative, it is unnecessary to
repeat the pivot selection process. Crossing out the inactive variable columns and
reading the result yields
x y z s t P
2 1 0 4 21 0 120
£0 2 4 24 2 0 0§
0 1 0 4 1 2 360
2x 5 120 4z 5 0 2P 5 360
x 5 60 z50 P 5 180
The corner point (x, y, z) that maximizes the objective function is (60, 0, 0) . It is
interesting to note that although z is an active variable, z 5 0. Active variables
may equal zero, but inactive variables must equal zero.
In the context of the scenario, the store should order 60 packages of the first
brand to earn the maximum profit of $180. Even though the profit per package
for the third brand ($4 per package) was higher than the profit per package for the
first brand ($3 per package), the higher wholesale cost of the third brand ($4 ver-
sus $2) limited the number of third-brand items available for sale. Ironically, the
lowest-priced brand of cheese resulted in the greatest profit.

In Examples 1 and 2, there was a unique solution to the standard maximiza-

tion problem; however, on occasion you’ll encounter a standard maximization
problem with infinitely many solutions. Example 3 is such a case.

EXAMPLE 3 Using the Simplex Method to Make Business Decisions

A store can purchase three brands of binders. The first brand, A, costs $2 per binder.
The second brand, B, costs $1 per binder. The third brand, C, costs $2 per binder.
The store can spend up to $100 and purchase up to 50 binders. It resells binder A
for $5 per binder, binder B for $3 per binder, and binder C for $5 per binder. Re-
call that profit per binder is revenue per binder minus cost per binder. How many
of each type of binder should the store buy if it wants to maximize its profit?

SOLUTION Let x be the number of binders of the first brand, y be the number
of binders of the second brand, and z be the number of binders of the third brand.
The total revenue from binder sales is R 5 5x 1 3y 1 5z. The total cost of the
binders is C 5 2x 1 1y 1 2z. Since profit is the difference between revenue
and cost,
P5R2C
5 (5x 1 3y 1 5z) 2 (2x 1 1y 1 2z)
5 3x 1 2y 1 3z
218 CHAPTER 4 Linear Programming

Since we want to maximize profit, we must maximize P 5 3x 1 2y 1 3z

subject to
x 1 y 1 z # 50 At most 50 binders are purchased
2x 1 y 1 2z # 100 At most $100 is spent on binders
x $ 0, y $ 0, z $ 0 A nonnegative number of each brand of
binder is purchased
The problem is a standard maximization problem. Converting the linear pro-
gramming problem into a system of linear equations, we get
x 1 y 1 z 1 s 5 50 Binder quantity constraint
2x 1 y 1 2z 1 t 5 100 Budget constraint
23x 2 2y 2 3z 1 P 5 0 Objective function
and the initial simplex tableau
x y z s t P
1 1 1 1 0 0 50
£ 2 1 2 0 1 0 100 §
23 22 23 0 0 1 0
n
Pivot column

Since 23 is the negative number in the bottom row that has the largest magni-
tude, we may select either column x or column z as the pivot column. We select
column z. Because the entries in the first and second rows of the pivot column are
positive, we must calculate the test ratios.
x y z s t P
1 1 1 1 0 0 50 50>1 5 50
£ 2 1 2 0 1 0 100 § 100>2 5 50
23 22 23 0 0 1 0
Since the entries have the same test ratio, we may select either one as the pivot.
We select the entry in the first row. We zero out the remaining entries in the pivot
column with the indicated row operations and obtain the tableau
x y z s t P
1 1 1 1 0 0 50
£0 21 0 22 1 0 0§ R2 2 2R1
0 1 0 3 0 1 150 R3 1 3R1

Crossing out the inactive variable columns and reading the resultant equations,
we get
x y z s t P
1 1 1 1 0 0 50
£0 21 0 22 1 0 0§
0 1 0 3 0 1 150
x 1 z 5 50 t50 P 5 150
z 5 2x 1 50
 4.3 Solving Standard Maximization Problems with the Simplex Method 219

There are infinitely many solutions (x, y, z) that maximize the objective function.
Each solution is of the form (n, 0, 2n 1 50) with 0 # n # 50.
In the context of the real-world scenario, the store may purchase any combi-
nation of 50 binders from brands A and C and still maximize profits. This is be-
cause both brands have the same wholesale cost and the same retail price. As a
result, the cost, revenue, and profit for each item are the same.

When a standard maximization problem has more than two constraints, ad-
ditional slack variables will be needed. This will result in additional columns in
the initial simplex tableau. We typically use the letters u, v, and so on for slack
variables in addition to the letters s and t. Even with additional rows or columns,
the simplex method is effective in finding the optimal solution to any standard
maximization problem.

EXAMPLE 4 Using the Simplex Method to Make Business Decisions

Clothing retailers typically mark up the price of apparel by 100 percent. That is,
the retail price is typically 200 percent of the wholesale cost. WholesaleFash-
ion.com claims to be the number one Internet business-to-business source for
fashion retailers of women’s apparel and shoes. It sells directly to businesses in-
stead of to consumers. In January 2003, WholesaleFashion.com offered a 3/4
Sleeves Stretch Top for $9.50, a Long Sleeve Turtle Neck Top for $12.50, a Heart
Neck Tank Top for $9.25, and an American Flag Tank Top for $9.75. (Source:
www.wholesalefashion.com.)
A women’s apparel store has up to $1000 to spend on the four items. It has
enough rack space for up to 60 tops. It anticipates the demand for tank tops to be
greater than the demand for the other tops, so it wants to order at least twice as
many tank tops as sleeved tops. The store intends to mark up all of the items by
100 percent. Assuming that all of the items that the store orders will sell, how
many of each item should it order if it wants to maximize revenue?

SOLUTION Let x be the number of 3/4 Sleeves Stretch Tops, y be the number
of Long Sleeve Turtle Neck Tops, z be the number of Heart Neck Tank Tops, and
w be the number of American Flag Tank Tops. The number of sleeved tops is
x 1 y, and the number of tank tops is z 1 w. Since the number of tank tops is to
be at least twice the number of sleeved tops, we have
z 1 w $ 2(x 1 y)
z 1 w $ 2x 1 2y
22x 2 2y 1 z 1 w $ 0
2x 1 2y 2 z 2 w # 0 Dividing by 21 reverses the inequality sign

We are asked to maximize the revenue from sales

R 5 19x 1 25y 1 18.50z 1 19.50w
subject to the constraints
x 1 y 1 z 1 w # 60 At most 60 tops are purchased
2x 1 2y 2 z 2 w # 0 Twice as many tank tops are purchased
9.5x 1 12.5y 1 9.25z 1 9.75w # 1000 At most $1000 is spent on tops
x $ 0, y $ 0, z $ 0, w $ 0 A nonnegative number of each type is purchased
220 CHAPTER 4 Linear Programming

Rewriting the objective function and adding the slack variables to the constraint
inequalities yields the system of equations
x1 y1 z1 w1s 5 60
2x 1 2y 2 z2 w 1t 50
9.5x 1 12.5y 1 9.25z 1 9.75w 1u 5 1000
219x 2 25y 2 18.50z 2 19.50w 1R50
and the associated initial simplex tableau
x y z w s t u R
1 1 1 1 1 0 0 0 60

≥ ¥
2 2 21 21 0 1 0 0 0
9.5 12.5 9.25 9.75 0 0 1 0 1000
219 225 218.5 219.5 0 0 0 1 0
n
Pivot column

The y column is the pivot column, since 225 is the negative entry in the bottom
row with the greatest magnitude. Since the first three entries in the y column are
positive, we must calculate three test ratios.
x y z w s t u R
1 1 1 1 1 0 0 0 60 60>1 5 60

≥ ¥
2 2 21 21 0 1 0 0 0 0>2 5 0
9.5 12.5 9.25 9.75 0 0 1 0 1000 1000>12.5 5 80
219 225 218.5 219.5 0 0 0 1 0
The pivot entry is the 2 in the second column and second row, since the second
row has the smallest test ratio. We zero out the remaining entries in the second
column with the indicated operations.
Because of the numerical complexity of the entries in the tableau, we will use
the Technology Tips following this example to do the row operations. The result-
ant tableau is
x y z w s t u R
0 0 3 3 2 21 0 0 120 2R1 2 R2

≥ ¥
2 2 21 21 0 1 0 0 0
26 0 31 32 0 212.5 2 0 2000 2R3 2 12.5R2
12 0 262 264 0 25 0 2 0 2R4 1 25R2
n
New pivot
column

The w column is the new pivot column, since 264 is the negative entry in the bot-
tom row that has the largest magnitude. Since the only positive entries in the w
column are in the first and third rows, we need to calculate only two test ratios.
x y z w s t u R
0 0 3 3 2 21 0 0 120 120>3 5 40

≥ ¥
2 2 21 21 0 1 0 0 0
26 0 31 32 0 212.5 2 0 2000 2000>32 5 62.5
12 0 262 264 0 25 0 2 0
 4.3 Solving Standard Maximization Problems with the Simplex Method 221

Since the entry in the second row of the pivot column was negative, we did not
calculate a test ratio for the second row. The pivot entry is the 3 in the w column
and first row. We will zero out the remaining entries in the w column with the in-
dicated operations. The resultant tableau is
x y z w s t u R
0 0 3 3 2 21 0 0 120

≥ ¥
6 6 0 0 2 2 0 0 120 3R2 1 R1
218 0 23 0 264 25.5 6 0 2160 3R3 2 32R1
36 0 6 0 128 11 0 6 7680 3R4 1 64R1
Since all entries in the bottom row are nonnegative, this is the final simplex
tableau. Crossing out the inactive variable columns and reading the resultant
equations, we get
x y z w s t u R
0 0 3 3 2 21 0 0 120

≥ ¥
6 6 0 0 2 2 0 0 120
218 0 23 0 264 25.5 6 0 2160
36 0 6 0 128 11 0 6 7680
x50 6y 5 120 z50 3w 5 120
y 5 20 w 5 40
s50 t50 6u 5 2160 6R 5 7680
u 5 360 R 5 1280
The maximum revenue that can be attained is $1280. This will be achieved when
20 Long Sleeve Turtle Neck Tops and 40 American Flag Tank Tops are purchased.
Recall that u was the slack variable associated with the equation for the total cost
of the tops. Since u 5 360, the amount spent by the store is $360 less than the
maximum amount allowed. That is, $640 is spent on the tops. The additional money
remains unspent. It was the space constraint (a maximum of 60 tops), not the budget
constraint (a maximum of $1000), that limited how many tops could be ordered.

For large tableaus or tableaus with noninteger entries, it is often helpful to do

row operations on the calculator. The next four Technology Tips detail how to
do this on the TI-83 Plus.

Interchanging Two Rows

1. Enter a matrix or simplex tableau using

the Matrix Editor. (Press 2nd x1 to
access the Matrix Editor.)

(Continued)
222 CHAPTER 4 Linear Programming

(Continued)
2. Close the Matrix Editor to store the
matrix, then reopen the Matrix Editor
and move the cursor to the MATH menu.

3. Scroll to C:rowSwap( and press

ENTER . This operation is used to
interchange one row with another row.

4. Type in the matrix name from the

Matrix Menu, the first row number, and
the second row number. For example,
rowSwap([A],1,2)is equivalent to
R1 4 R2.

5. Press ENTER to display the new matrix

or tableau.

Adding One Row to Another Row

1. Enter a matrix or simplex tableau using

the Matrix Editor. (Press 2nd x1 to
access the Matrix Editor.)

(Continued)
 4.3 Solving Standard Maximization Problems with the Simplex Method 223

2. Close the Matrix Editor to store the

matrix, then reopen the Matrix Editor
and move the cursor to the MATH menu.

3. Scroll to D:row+( and press ENTER .

This operation is used to add one row to
another row.

4. Type in the matrix name from the

Matrix Menu, the first row number, and
the second row number. The result will
be placed in the last row listed. For
example, row+([A],2,3) is equivalent
to R2 1 R3 S R3.

5. Press ENTER to display the new matrix

or tableau.

Multiplying a Row by a Nonzero Constant

1. Enter a matrix or simplex tableau using

the Matrix Editor. (Press 2nd x1 to
access the Matrix Editor.)

(Continued)
224 CHAPTER 4 Linear Programming

(Continued)
2. Close the Matrix Editor to store the
matrix, then reopen the Matrix Editor
and move the cursor to the MATH menu.

3. Scroll to E:*row( and press ENTER .

This operation is used to multiply a row
by a nonzero constant.

4. Type in the multiplier value, the matrix

name from the Matrix Menu, and the
row to be multiplied. The result will be
placed in the last row listed. For
example, *row(-2,[A],1) means
22R1 S R1.

5. Press ENTER to display the new matrix

or tableau.

Adding a Multiple of One Row to Another Row

1. Enter a matrix or simplex tableau using

the Matrix Editor. (Press 2nd x1 to
access the Matrix Editor.)

(Continued)
 4.3 Solving Standard Maximization Problems with the Simplex Method 225

2. Close the Matrix Editor to store the

matrix, then reopen the Matrix Editor
and move the cursor to the MATH menu.

3. Scroll to F:*row+( and press ENTER .

This operation is used to add a multiple
of one row to another row.

4. Type in the multiplier value, the matrix

name, the row to be multiplied, and the
row to be added. The result will be
placed in the last row listed. For
example, *row+(4,[A],2,3) means
4R2 1 R3 S R3.

5. Press ENTER to display the new matrix

or tableau.

Recall that simplex method row operations must be of the form

aRc 6 bRp S Rc, with both a and b positive. Alternatively, we may write this ex-
pression as 6bRp 1 aRc S Rc. The latter form is more consistent with the calcu-
lator functionality and may make using the calculator easier.

EXAMPLE 5 Using the Simplex Method to Make Business Decisions

In January 2003, WholesaleFashion.com offered a Long Halter Dress for $25.75,
a Long Open Back Dress for $21.75, a pair of Thai Silk Pump shoes for $17.75,
and a pair of Glitter Evening High Heels for $17.50.
(Source: www.wholesalefashion.com.)
A women’s apparel store has up to $1000 to spend on the four items. It will
mark up the dresses 100 percent but mark up the shoes only 50 percent. The store
anticipates that it will sell 80 percent of each type of dress at full price and sell
the remaining 20 percent of each type of dress at its cost. It expects to sell 90 per-
cent of each type of shoes at full price and the remaining 10 percent of each shoe
type at its cost. It expects to sell a pair of shoes with each dress it sells. (The store
will market a dress and shoes as a package.) Additionally, it may sell some shoes
separately. How many of each item should the store order to maximize its profit?
226 CHAPTER 4 Linear Programming

SOLUTION Let x be the number of Long Halter Dresses, y be the number of

Long Open Back Dresses, z be the number of pairs of Thai Silk Pump shoes, and
w be the number of pairs of Glitter Evening High Heels.
The revenue generated by selling 80 percent of each type of dress at full re-
tail price is given by
R 5 51.50(0.8x) 1 43.50(0.8y)
5 41.20x 1 34.80y
The revenue generated by selling the remaining 20 percent of each type of dress
at the wholesale cost is given by
R 5 25.75(0.2x) 1 21.75(0.2y)
5 5.15x 1 4.35y
The combined revenue from dress sales is
Dress revenue 5 (41.20x 1 34.80y) 1 (5.15x 1 4.35y)
5 46.35x 1 39.15y
The retail price of each shoe variety is 150 percent of the wholesale price.
Thai Silk Pumps 5 1.5(17.75)
5 26.625
< 26.63 Rounded to the nearest cent

Glitter Evening High Heels 5 1.5(17.50)

5 26.25
The revenue generated by selling 90 percent of each shoe variety at full retail
price is given by
R 5 26.63(0.9z) 1 26.25(0.9w)
< 23.97z 1 23.63w Rounded to the nearest cent

The revenue generated by selling 10 percent of each shoe variety at wholesale

cost is
R 5 17.75(0.1z) 1 17.50(0.1w)
< 1.78z 1 1.75w Rounded to the nearest cent

The combined revenue from shoe sales is

Shoe revenue 5 (23.97z 1 23.63w) 1 (1.78z 1 1.75w)
< 25.75z 1 25.38w
The combined revenue from dresses and shoes is the sum of the dress revenue and
the shoe revenue.
Combined revenue 5 dress revenue 1 shoe revenue
< (46.35x 1 39.15y) 1 (25.75z 1 25.38w)
< 46.35x 1 39.15y 1 25.75z 1 25.38w
The total cost of the dresses and shoes is
C 5 25.75x 1 21.75y 1 17.75z 1 17.5w
 4.3 Solving Standard Maximization Problems with the Simplex Method 227

Since profit is the difference in revenue and cost,

P5R2C
< (46.35x 1 39.15y 1 25.75z 1 25.38w)
2 (25.75x 1 21.75y 1 17.75z 1 17.50w)
< 20.60x 1 17.40y 1 8.00z 1 7.88w
We need to order at least as many pairs of shoes as dresses, so
Number of dresses # number of pairs of shoes
(x 1 y) # (z 1 w)
x1y2z2w#0
We are asked to maximize the profit P 5 20.6x 1 17.4y 1 8.00z 1 7.88w sub-
ject to the constraints
x1y2z2w#0 Quantity constraint
25.75x 1 21.75y 1 17.75z 1 17.5w # 1000 Budget constraint
x $ 0, y $ 0, z $ 0, w $ 0
Rewriting the objective function and adding the slack variables to the constraint
inequalities yields the following system of equations:
x1 y2 z2 w1s 50
25.75x 1 21.75y 1 17.75z 1 17.5w 1t 5 1000
220.60x 2 17.40y 2 8.00z 2 7.88w 1P50
and the associated initial simplex tableau
x y z w s t P
1 1 21 21 1 0 0 0
£ 25.75 21.75 17.75 17.50 0 1 0 1000 §
220.60 217.40 28.00 27.88 0 0 1 0
n
Pivot column

The x column is the pivot column, since 220.60 is the negative entry in the bot-
tom row that has the largest magnitude. We must calculate test ratios for the first
two rows.
x y z w s t P
1 1 21 21 1 0 0 0 0>1 5 0
£ 25.75 21.75 17.75 17.50 0 1 0 1000 § 1000>25.75 < 38.83
220.60 217.40 28.00 27.88 0 0 1 0
The pivot entry is the 1 in the x column and the first row. We will zero out the re-
maining entries in the first column with the indicated operations. Because of the
numerical complexity of the entries in the tableau, we will use the Technology
Tips described in this section to do the row operations.
228 CHAPTER 4 Linear Programming

The resulting tableau is

x y z w s t P
1 1 21 21 1 0 0 0
£0 24 43.5 43.25 225.75 1 0 1000 § 225.75R1 1 R2
0 3.2 228.60 228.48 20.60 0 1 0 20.60 R1 1 R3
n
New pivot
column

(Note: We wrote the row operations in the alternative form 6bRp 1 aRc S Rc.)
The z column is the new pivot column, since 2 28.60 is the negative entry in
the bottom row that has the largest magnitude. Since there is only one positive
entry in the z column above the bottom row, that entry is the pivot. The pivot entry
is the 43.5 in the z column and the second row. We will zero out the remaining en-
tries in the z column with the following row operations written in the alternative
form:
R2 1 43.5R1 S R1
28.6R2 1 43.5R3 S R3
To do this on the calculator requires a sequence of operations. To perform the op-
eration R2 1 43.5R1 S R1, we do the following to the tableau that is on the screen
of our calculator:
Step 1: *row(43.5, ans, 1) Multiplies R1 by 43.5
Step 2: ENTER Returns the modified matrix
Step 3: row 1 (ans, 2, 1) Adds the first two rows of the modified matrix
Step 4: ENTER Completes the operation

To perform the operation 28.6R2 1 43.5R3 S R3, we do the following to the

tableau that is on the screen of our calculator:
Step 1: *row(43.5, ans, 3) Multiplies R3 by 43.5
Step 2: ENTER Returns the modified matrix
Step 3: *row 1 (28.6, ans, 2, 3) Multiplies R2 by 28.6 and adds it to the
modified R3
Step 4: ENTER Completes the operation
The resultant tableau is
x y z w s t P
43.5 39.5 0 20.25 17.75 1 0 1000
£ 0 24 43.5 43.25 225.75 1 0 1000 §
0 24.8 0 21.93 159.65 28.60 43.5 28600
n
New pivot
column

The bottom row contains a negative entry in the w column, so we must repeat the
pivoting process. Since the only positive entry in the pivot column is 43.25, it is
the new pivot. The indicated row operations (in alternative form) will zero out the
additional terms in the pivot column.
0.25R2 1 43.25R1 S R1
1.93R2 1 43.25R3 S R3
 4.3 Solving Standard Maximization Problems with the Simplex Method 229

To perform the operation 0.25R2 1 43.25R1 S R1, we do the following to the

tableau that is on the screen of our calculator:
Step 1: *row(43.25, ans, 1) Multiplies R1 by 43.25
Step 2: ENTER Returns the modified matrix
Step 3: *row 1 (0.24, ans, 2, 1) Multiplies R2 by 0.25 and adds it to the
modifed R1
Step 4: ENTER Completes the operation

To perform the operation 1.93R2 1 43.25R3 S R3, we do the following to the

tableau that is on the screen of our calculator:
Step 1: *row(43.25, ans, 3) Multiplies R3 by 43.25
Step 2: ENTER Returns the modified matrix
Step 3: *row 1 (1.93, ans, 2, 3) Multiplies R2 by 1.93 and adds it to the
modified R3
Step 4: ENTER Completes the operation
The resultant simplex tableau is
x y z w s t P
1881.38 1707.38 10.88 0 761.25 43.5 0 43500
£ 0 24 43.5 43.25 225.75 1 0 1000 §
0 1064.88 83.96 0 6855.17 1238.88 1881.38 1238880
Crossing out the inactive variable columns and reading the resultant equations,
we get
x y z w s t P
1881.38 1707.38 10.88 0 761.25 43.5 0 43500
£ 0 24 43.5 43.25 225.75 1 0 1000 §
0 1064.88 83.96 0 6855.17 1238.88 1881.38 1238880
Reading from the tableau, we see that
1881.38x 5 43500 y50 z50 43.25w 5 1000
x 5 23.12 w 5 23.12
s50 t50 1881.38P 5 1238880
P 5 658.50
Since it doesn’t make sense to talk about a fraction of a dress or pair of shoes, we
will round the values to the nearest whole number. Rounding to whole numbers
doesn’t guarantee the maximum whole number solution. However, if the solution
isn’t the maximum whole number solution, it will be close to the optimal solu-
tion. We estimate that the maximum profit will be achieved when 23 Long Halter
Dresses and 23 pairs of Glitter Evening High Heels are purchased. Since t 5 0,
the amount spent by the store appears to be exactly $1000. (Actually, the cost is
$994.75, since we rounded the number of dresses and shoes.) The maximum
profit is expected to be $658.50. (Since we rounded down the number of shoes
and dresses, the exact profit is $655.04.)

Even though Example 5 required some rounding to make sense of the results,
the numerical analysis provided some valuable input to help us make a sound
business decision.
230 CHAPTER 4 Linear Programming

4.3 Summary

In this section, you learned the simplex method for solving standard maximiza-
tion problems. You learned that this is the method of choice in solving problems
with more than two decision variables.

4.3 Exercises

In Exercises 1–10, determine whether the problem is a 7. Maximize P 5 x 1 9y 1 8z

standard maximization problem. If it isn’t, explain why.  26x 2 y 1 4z # 224

1. Maximize P 5 9x 1 8y  29x 2 1.5y 1 6z # 36
Subject to  x$ 0
 2x 1 y # 10 
 2x 1 y#1  y$ 0
Subject to   z$ 0
 x$0
 y$0 8. Maximize P 5 4x 1 4y 1 9z
2. Maximize P 5 22x 2 5y  23x 2 2y 1 4z $ 222

 4x 1 9y # 10  5x 2 2.5y 1 6z # 36
 211x 1 y # 221 Subject to  x$ 0
Subject to   y$ 0
x$0 
  z$ 0
 y$0
9. Maximize P 5 28x 1 2y 2 z
3. Maximize P 5 6xy
 z#4
 6x 1 7y # 13 
 28x 2 4y # 12 Subject to 
x2y#3
Subject to   2x 2 z # 5
 x$0
x 1 y 1 z # 0
 y$0
10. Maximize P 5 5x 1 7z
4. Maximize P 5 4x 2 2y 1 z
 4x 1 2y # 16
 2x 1 y 1 4z # 24  4x 2
 2y # 3
 22x 1 3y 1 3z # 150 Subject to 
Subject to  x$0  x$ 0
  y$ 0
 y$0
 z$0
In Exercises 11–20, solve the standard maximization
5. Maximize P 5 21.2x 2 2.8y 1 4.3z problems by using the simplex method. Check your
answer by graphing the feasible region and
e
3.2x 1 1.5y 1 7.4z # 249.8
Subject to calculating the value of the objective function at each
22.7x 1 3.4y 1 3.9z # 190.1
of the corner points.
6. Maximize P 5 4x 1 4y 1 9z
11. Maximize P 5 9x 1 8y
 26x 2 y 1 4z # 224
  2x 1 y # 10
 29x 2 1.5y 1 6z # 36  2x 1 y # 1
Subject to  x$0 Subject to 
 y$0  x$0

 z$0  y$0
 4.3 Solving Standard Maximization Problems with the Simplex Method 231

12. Maximize P 5 22x 1 4y 20. Maximize P5x1y

 2x 1 y#6  5x 1 y # 20
 x1 
Subject to 
y#4  2x 1 2y # 29
 x$0 Subject to  10x 1 y # 30
 y$0  x$ 0

13. Maximize P 5 5x 2 2y  y$ 0

 23x 1 y#0 In Exercises 21–30, solve the standard maximization

 3x 1 y # 12
Subject to  problems by using the simplex method.
 x$0
 y$0 21. Maximize P 5 5x 1 6y 1 6z
 2x 1 3y 1 2z # 120
14. Maximize P 5 7x 1 10y 
 2x 1 y 1 z # 60
 x1 y#4 Subject to  x$0
 5x 1 y # 12 
Subject to   y$0
 x$0  z$0
 y$0
22. Maximize P 5 4x 1 6y 1 5z
15. Maximize P 5 4x 1 4y
 2x 1 3y 1 3z # 210
 4x 1 3y # 20 
 2x 1 2y # 6  x 1 y 1 z # 100
Subject to  Subject to  x$0
 x$0  y$0

 y$0  z$0
16. Maximize P 5 5x 2 2y 23. Maximize P 5 10x 1 6y 1 12z
 4x 1 5y # 28  8x 1 5y 1 9z # 360
 23x 1 2y # 2 
Subject to   x 1 y 1 z # 50
 x$0 Subject to  x$0
 y$0 
 y$0
17. Maximize P 5 6x 1 5y  z$0
 3x 1 2y # 12 24. Maximize P 5 x 2 2y 1 4z
 x1y#5  6x 2 5y 1 10z #
Subject to  300

 x$0  2x 1 5y 1 2z # 500
 y$0 Subject to  x$ 0
 y$
18. Maximize P 5 6x 1 2y  0
 z$ 0
 x1 y # 10
 25. Maximize P 5 5x 2 2y 1 4z
 2x 1 y # 16
Subject to  22x 1 y#0  8x 1 5y 1 9z # 360
 x$0 
  x 1 y 1 z # 50
 y$0 Subject to  x$0

19. Maximize P 5 7x 2 5y  y$0
 z$0
 6x 2 5y # 30
 26. Maximize P 5 4x 1 4y 2 10z
 2x 1 2y # 10
Subject to  25x 1 3y # 7  3x 1 3y 1 6z # 42
 x$0  2x 1 y 1 z # 8

 y$0  3x 1 2y 1 5z # 23
Subject to 
 x$0
 y$0
 z$0
232 CHAPTER 4 Linear Programming

27. Maximize P 5 2x 2 y 1 10z 32. Resource Allocation: Sandwiches A

 2x 1 2y 1 2z # 14 plain hamburger requires one ground
 4x 1 2y 1 2z # 16 beef patty and a bun. A cheeseburger requires one
 3x 1 2y 1 5z # 23 ground beef patty, one slice of cheese, and a bun.
Subject to  A double cheeseburger requires two ground beef
 x$0
patties, two slices of cheese, and a bun.
 y$0
 z$0 Frozen hamburger patties are typically sold in
packs of 12; hamburger buns, in packs of 8; and
28. Maximize P 5 x 1 2y 1 3z cheese slices, in packs of 24.
 x 1 z # 20 A family is in charge of providing burgers for a
 y 1 2z # 30 neighborhood block party. The family members
 x 1 2y 1 3z # 60 have purchased 13 packs of buns, 11 packs of
Subject to  hamburger patties, and 3 packs of cheese slices.
 x$0
 y$0 How many of each type of sandwich should the
 z$0 family prepare if its members want to maximize
the number of burgers with cheese?
29. Maximize P5x1y1z
33. Resource Allocation: Sandwiches
 x1z#0
 Repeat Exercise 32, except maximize the
y 1 2z # 5
 x 1 2y 1 3z # 8 number of hamburger patties used.
Subject to 
 x$0 34. Resource Allocation: Sandwiches
 y$0 Repeat Exercise 32. except maximize the
 z$0 number of double cheeseburgers.
30. Maximize P 5 x 1 2y 1 3z 35. Battery Sales AAA Alkaline Discount
 x 2 z # 20 Batteries sells low-cost batteries to
 y 1 z # 30 consumers. In August 2002, the company offered
 x 2 2y 1 z # 40 AA batteries at the following prices: 50-pack for
Subject to  $10.00, 100-pack for $18.00, and 600-pack for
 x$0
$96.00. (Source: www.aaaalkalinediscountbatteries
 y$0
.com.)
 z$0
An electronics store owner wants to buy at
In Exercises 31–36, set up and solve the standard most 1000 batteries and spend at most $175. She
maximization problem using the simplex method. expects that she’ll be able to resell all of the
batteries she orders for $4.00 per 4-pack. How
31. Resource Allocation: Beverages In the many packs (50-packs, 100-packs, or 600-packs)
mood for a new beverage recipe? should she order if she wants to maximize her
iDrink.com provides you with recipes for revenue? What is her maximum revenue?
alcoholic and nonalcoholic drinks based on the
ingredients you have on hand. 36. Battery Sales Repeat Exercise 35,
The Cranberry Cooler calls for 2 ounces of except maximize profit. Assume that her
lemon-lime soda and 4 ounces of cranberry juice only cost is the cost of the batteries. What is her
in addition to other ingredients. Nancy’s Party maximum profit?
Punch calls for 32 ounces of cranberry juice and
no lemon-lime soda. Jimmy Wallbanger calls for Exercises 37–38 are intended to expand your
6 ounces of lemon-lime soda and no cranberry understanding of the simplex method.
juice. (Source: www.idrink.com.) The Cranberry 37. Given the standard maximization problem
Cooler recipe yields 1 serving, the Nancy’s Party
Maximize z 5 4x 1 4y
Punch recipe yields 9 servings, and the Jimmy
Wallbanger recipe yields 1 serving.  2x 1 y # 4
 x 1 2y # 14
If you have 2 quarts (64 ounces) of lemon-lime Subject to 
soda and 1 gallon (128 ounces) of cranberry juice,  5x 1 2y # 30
at most how many drink servings can you make?  x $ 0, y $ 0
 4.4 Solving Standard Minimization Problems with the Dual 233

do the following: (d) Explain what the simplex method does in

(a) Graph the feasible region and find the terms of the graph of the feasible region.
coordinates of the corner points. 38. Repeat Exercise 37; however, this time pick a
(b) Solve the standard maximization problem different column to be the first pivot column.
using the simplex method. Compare and contrast your results with those of
(c) Find the feasible solution associated with Exercise 37.
each simplex tableau and label the point on
the graph that is associated with each feasible
solution.

4.4 Solving Standard Minimization Problems with the Dual

■ Find the solution to a

standard minimization A food distributor has two production facilities: one
problem by solving the in Portland, Oregon, and the other in Spokane, Washington. Subject to production
dual problem with the limitations and market demands, what shipment plan will minimize the distributor’s
simplex method shipment costs? Questions such as this may often be answered by solving a stan-
dard minimization problem.
In this section, we will introduce standard minimization problems and show
how finding the maximum solution for a dual problem leads us to the minimum
solution of the minimization problem.

STANDARD MINIMIZATION PROBLEM

A standard minimization problem is a linear programming problem with
an objective function that is to be minimized. The objective function is of
the form
P 5 ax 1 by 1 cz 1 # # #
where a, b, c, . . . are real numbers and x, y, z, . . . are decision variables.
The decision variables are constrained to nonnegative values. Additional
constraints are of the form
Ax 1 By 1 Cz 1 # # # $ M
where A, B, C,... are real numbers and M is nonnegative.

We are told that M is nonnegative. That is, M $ 0. Is the origin in the feasible
region? Substituting the origin x 5 0, y 5 0, z 5 0, . . . into each
constraint inequality Ax 1 By 1 Cz 1 # # # $ M yields the inequality
A(0) 1 B(0) 1 C(0) 1 # # # $ M which simplifies to 0 $ M . Since M is non-
negative, the inequality 0 $ M is satisfied if and only if M 5 0. That is, the ori-
gin is in the feasible region if and only if M 5 0 for every constraint. If M 2 0
for one or more constraints, the origin is not in the feasible region.
234 CHAPTER 4 Linear Programming

EXAMPLE 1 Determining If a Linear Programming Problem Is a Standard

Minimization Problem
Determine if the linear programming problem is a standard minimization problem.
Minimize P 5 2x 1 3y
3x 1 y $ 6
Subject to • 2x 2 y $ 4
x $ 0, y $ 0

SOLUTION The objective function is to be minimized, the constraints are of the

form Ax 1 By 1 Cz 1 # # # $ M with M nonnegative, and the decision variables
are nonnegative. Therefore, the problem is a standard minimization problem.

EXAMPLE 2 Determining If a Linear Programming Problem Is a Standard

Minimization Problem
Determine if the linear programming problem is a standard minimization problem.
Minimize P 5 4x 1 2y
23x 1 7y # 29
Subject to • 7x 1 5y $ 4
x $ 0, y $ 0

SOLUTION At first glance, the problem doesn’t look like a standard mini-
mization problem. The first constraint has a less than or equal sign, and the con-
stant term is negative. However, if we multiply the first constraint by 21, it
becomes 3x 2 7y $ 9. The problem may then be rewritten as
Minimize P 5 4x 1 2y
3x 2 7y $ 9
Subject to • 7x 1 5y $ 4
x $ 0, y $ 0
Therefore, the linear programming problem is a standard minimization problem.

When setting up a linear programming problem, if we make sure that the con-
stant term of each constraint is nonnegative, then we will be able to readily see if
the problem is a standard maximization or minimization problem.

The Dual
For a standard minimization problem whose objective function has nonnegative
coefficients, we may construct a standard maximization problem called the dual
problem. By solving the dual problem, we can find the solution to the standard
minimization problem. Consider the standard minimization problem,
Minimize P 5 6x 1 5y
x1y$2
Subject to • 2x 1 y $ 3
x $ 0, y $ 0
 4.4 Solving Standard Minimization Problems with the Dual 235

We first construct a matrix for the problem as shown.

1 1 2 x1y$2
A 5 £2 1 3§ 2x 1 y $ 3
6 5 1 6x 1 5y 5 P

The constraints are placed in the first two rows of the matrix, and the coefficients
of the objective function are placed in the last row of the matrix. The matrix for
the dual problem is found by transposing the matrix. The transpose of a matrix
A, which is written AT , is created by switching the rows and columns of A. In this
1
case, the first row of A is 31 1 24 , so the first column of AT is £ 1 § . Repeating
2
1 2 6
the process for the additional columns of AT yields AT 5 £ 1 1 5 § . We construct
2 3 1
the dual problem from AT . We first extract the constraints and the objective func-
tion from the matrix.
1 2 6 x 1 2y # 6
A 5 £1 1 5§ x1y#5
2 3 1 2x 1 3y 5 P

The dual problem is

Maximize P 5 2x 1 3y
x 1 2y # 6
Subject to • x 1 y # 5
x $ 0, y $ 0
How are the optimal values of the standard minimization problem and its dual
problem related? Let’s look at the graphs of the feasible regions (Figures 4.26
and 4.27).

y
y
5
3 (0, 3) x=0 x+y=5
2x + y = 3 4
2 (0, 3)
3
x+y=2
y=0 1 (1, 1) 2
x + 2y = 6
(2, 0) (4, 1)
x 1
−1 1 2 (5, 0)
−1 x
x=0 −1 1 2 3 4 5 6
−1 y=0
FIGURE 4.26 Standard
Minimization Problem FIGURE 4.27 Standard Maximization
Corner points: Problem
(0, 3), (1, 1), (2, 0) Corner Points: (0, 0), (0, 3), (4, 1), (5, 0)
236 CHAPTER 4 Linear Programming

For each problem, we calculate the value of the objective function at each
corner point (see Tables 4.20 and 4.21).

TABLE 4.20 Standard Minimization Problem TABLE 4.21 Standard Maximization Problem
x y P 5 6x 1 5y x y P 5 2x 1 3y
0 3 15 0 0 0
1 1 11 0 3 9
2 0 12 4 1 11
5 0 10

Observe that the minimum value of the minimization problem objective function
is 11. Similarly, the maximum value of the maximization problem objective func-
tion is 11. This leads us to the Fundamental Principle of Duality.

FUNDAMENTAL PRINCIPLE OF DUALITY

A standard minimization problem has a solution if and only if its dual prob-
lem has a solution. If a solution exists, the optimal value of the standard mini-
mization problem and the optimal value of the dual problem are equal.

Although we already know the optimal value of the dual problem, we will
demonstrate how using the simplex method to find the maximum value of the
dual problem helps us determine the minimum value of the standard minimiza-
tion problem. Recall that the dual problem is
Maximize P 5 2x 1 3y
x 1 2y # 6
Subject to • x 1 y # 5
x $ 0, y $ 0
Adding in the slack variables and rewriting the objective function yield the sys-
tem of equations
x 1 2y 1 s 56
x1 y 1t 55
22x 2 3y 1P50
and the corresponding initial simplex tableau
x y s t P
1 2 1 0 0 6
£ 1 1 0 1 0 5§
22 23 0 0 1 0
 4.4 Solving Standard Minimization Problems with the Dual 237

The y column is the pivot column, since it contains the negative entry in the bot-
tom row with the largest magnitude. We calculate the test ratios and identify
the pivot.
x y s t P
1 2 1 0 0 6 6>2 5 3
£ 1 1 0 1 0 5§ 5>1 5 5
22 23 0 0 1 0
The 2 in the first row of the y column is the pivot, since the first row has the
smallest test ratio. We zero out the remaining terms in the y column using the in-
dicated operations.
x y s t P
1 2 1 0 0 6
£ 1 0 21 2 0 4§ 2R2 2 R1
21 0 3 0 2 18 2R3 1 3R1
The x column is the new pivot column, since it is the only column with a nega-
tive entry in the bottom row. We calculate the test ratios and identify the pivot.
x y s t P
1 2 1 0 0 6 6>1 5 6
£ 1 0 21 2 0 4§ 4>1 5 4
21 0 3 0 2 18
The 1 in the second row of the x column is the pivot, since the second row has the
smallest test ratio. We zero out the remaining entries in the x column with the in-
dicated operations.
x y s t P
0 2 2 22 0 2 R1 2 R2
£1 0 21 2 0 4§
0 0 2 2 2 22 R3 1 R2

We must perform one additional step to make the nonzero term in each active col-
umn equal 1.
x y s t P
0 1 1 21 0 1 1>2 R1
£1 0 21 2 0 4§
0 0 1 1 1 11 1>2 R3

Reading from the tableau, we have

x54 y51 P 5 11
The corner point (4, 1) yields the maximum value of the objective function of the
dual problem P 5 2x 1 3y.
Ironically, the optimal solution of the standard minimization problem is also
nested in the final simplex tableau. With all active variable columns containing
only 1s and 0s, the values in the bottom row of the s and t columns correspond
with the x and y values, respectively, of the optimal solution of the standard min-
imization problem. That is, x 5 1 and y 5 1. The optimal value is the same as
that of the dual problem: P 5 11. We can verify that the corner point (1, 1) of
the standard minimization problem yields an objective function value of 11 by
238 CHAPTER 4 Linear Programming

substituting the point into the objective function of the standard minimization
problem P 5 6x 1 5y.
P 5 6x 1 5y
?
11 5 6(1) 1 5(1)
11 5 11
The relationship between the standard minimization problem and its dual prob-
lem is somewhat remarkable, and the problem-solving method is straightforward.

Using Duals to Solve Standard Minimization

Problems

1. Find the dual standard maximization problem.

2. Use the simplex method to solve the maximization problem.
3. The maximum value of the objective function of the dual problem is the
minimum value of the objective function of the minimization problem.
4. The optimum solution of the minimization problem is given in the entries
of the bottom row of the final tableau corresponding with the columns of
the slack variables (as long as the entry in the P column equals 1).

EXAMPLE 3 Solving a Standard Minimization Problem

Solve the standard minimization problem.
Minimize P 5 4x 1 5y 1 6z
x1y1z$3
Subject to • 2x 2 z $ 0
x $ 0, y $ 0, z $ 0

SOLUTION We begin by constructing a matrix from the constraints and objec-

tive function.
x y z
1 1 1 3 x1 y1 z$3
A 5 £2 0 21 0§ 2x 1 0y 2 z $ 0
4 5 6 1 4x 1 5y 1 6z 5 P
Next we find AT and extract the objective function and constraints. Note that the
constraints are of the form ax 1 by # M with M nonnegative.
1 2 4 x1 2y # 4

AT 5 ≥ ¥
1 0 5 x1 0y # 5
1 21 6 x2 y# 6
3 0 1 3x 1 0y 5 P
 4.4 Solving Standard Minimization Problems with the Dual 239

The dual problem is

Maximize P 5 3x
x 1 2y # 4

µ
x# 5
Subject to
x2y# 6
x$ 0, y $ 0, z $ 0
The dual is a standard maximization problem. We solve the problem using
the simplex method. We begin by adding slack variables to the constraints.
x 1 2y 1 s 5 4
x1t55
x2y1u56
x $ 0, y $ 0, z $ 0
Rewriting the objective function yields 23x 1 P 5 0. The initial simplex
tableau is
x y s t u P
1 2 1 0 0 0 4

≥ ¥
1 0 0 1 0 0 5
1 21 0 0 1 0 6
23 0 0 0 0 1 0
The x column is the pivot column, since it is the only column with a negative
entry in the bottom row. Mentally calculating the test ratios, we see that the entry
in the first row and the x column is the pivot.
The final tableau is obtained by using the indicated row operations.
x y s t u P
1 2 1 0 0 0 4

≥ ¥
0 22 21 1 0 0 1 R2 2 R1
0 23 21 0 1 0 2 R3 2 R1
0 6 3 0 0 1 12 R4 1 3R1
Reading from the tableau, we see that the objective function of the maximization
problem has a maximum value of 12. This occurs when x 5 4 and y 5 0. Ac-
cording to the fundamental principle of duality, the minimum value of the objec-
tive function of the minimization problem is also 12. The solution to the
minimization problem may be found by looking at the slack variable columns in
the bottom row of the final tableau. The s, t, and u columns of the final simplex
tableau are the x, y, and z values of the solution to the minimization problem. That
is, (3, 0, 0) yields an objective function value of 12. This can be verified by sub-
stituting x 5 3, y 5 0, and z 5 0 into P 5 4x 1 5y 1 6z.
P 5 4(3) 1 5(0) 1 6(0)
5 12

It is important to note that a standard minimization problem and its dual

problem may have different numbers of decision variables. In Example 3, the
standard minimization problem had three decision variables, x, y, and z, whereas
the dual problem had only two decision variables, x and y.
240 CHAPTER 4 Linear Programming

EXAMPLE 4 Solving a Standard Minimization Problem

Solve the standard minimization problem.
Minimize P 5 2x 1 y 1 2z
x 1 2y 1 z $ 6
Subject to • 3y 1 2z $ 12
x $ 0, y $ 0, z $ 0

SOLUTION We have
1 2 1 6 x 1 2y 1 z $ 6
A 5 £0 3 2 12 § 3y 1 2z $ 12
2 1 2 1 2x 1 y 1 2z 5 P
Therefore,
1 0 2 x# 2

AT 5 ≥ ¥
2 3 1 2x 1 3y # 1
and
1 2 2 x 1 2y # 2
6 12 1 6x 1 12y 5 P
The dual problem is
Maximize P 5 6x 1 12y
x# 2

Subject to µ
2x 1 3y # 1
x 1 2y # 2
x$ 0, y $ 0
Adding the slack variables and rewriting the objective function yields
x1s52
2x 1 3y 1 t 5 1
x 1 2y 1 u 5 2
26x 2 12y 1 P 5 0
The initial simplex tableau is
x y s t u P
1 0 1 0 0 0 2

≥ ¥
2 3 0 1 0 0 1
1 2 0 0 1 0 2
26 212 0 0 0 1 0
The y column is the pivot, since it contains the negative entry in the bottom row
with the largest magnitude. The pivot is the 3 in the second row and the y column
1
because the second row’s test ratio of 3 is smaller than the third row’s test ratio of
2
2
. Using the indicated row operations yields
 4.4 Solving Standard Minimization Problems with the Dual 241

x y s t u P
1 0 1 0 0 0 2

≥ ¥
2 3 0 1 0 0 1
21 0 0 22 3 0 4 3R3 2 2R2
2 0 0 4 0 1 4 R4 1 4R2

To obtain the final simplex tableau, we must ensure that the nonzero entry of each
active variable column is equal to 1.
x y s t u P
 1 0 1 0 0 0 2
 2 1 1 1
 3 1 0 0 0 3 R
3 2
3
 1 2 4 1
 23 0 0 23 1 0 3 R
3 3
 2 0 0 4 0 1 4
1
The solution to the dual problem is x 5 0 and y 5 3. The optimal value is 4.
From the bottom row of the final tableau, we read the solution of the minimization
problem from the slack variable columns: x 5 0, y 5 4, and z 5 0. The minimal
value of the objective function is 4. This can be verified by substituting the solu-
tion into the objective function equation of the standard minimization problem.
P 5 2x 1 y 1 2z
5 2(0) 1 (4) 1 2(0)
54
Substituting the point (0, 4, 0) back into the constraints of the standard mini-
mization problem yields
x 1 2y 1 z $ 6 3y 1 2z $ 12
(0) 1 2(4) 1 (0) $ 6 3(4) 1 2(0) $ 12
8$6 12 $ 12
We are further convinced that we have found the optimal solution, since the solu-
tion satisfies both constraints (as it should). Also, 3y 1 2z was made as small as
the constraint would allow.

EXAMPLE 5 Using a Standard Minimization Problem to

Minimize Shipping Costs
A northwest region canned food distributor has two production facilities: one in
Spokane, Washington, and one in Portland, Oregon. For two supermarkets, the
Portland facility produces a minimum of 700 cases of canned food per week and
the Spokane facility produces a minimum of 300 cases per week. A Seattle-area
supermarket requires at least 600 cases per week, and a supermarket in Ellens-
burg, Washington, requires at least 400 cases per week. Shipping costs (based on
a rate of $0.36 per mile) vary based on the point of origin and destination as
shown in Table 4.22.
242 CHAPTER 4 Linear Programming

TABLE 4.22
Destination
Seattle Ellensburg
Portland $0.62 per case $0.80 per case
Origin
Spokane $1.01 per case $0.63 per case

What shipping schedule will minimize the food distributor’s shipping costs?
SOLUTION
Let x be the number of cases shipped from Portland to Seattle.
Let y be the number of cases shipped from Spokane to Seattle.
Let z be the number of cases shipped from Portland to Ellensburg.
Let w be the number of cases shipped from Spokane to Ellensburg.
We need to minimize the shipment cost function,
C 5 0.62x 1 1.01y 1 0.80z 1 0.63w. We have the following constraints:
x 1 z $ 700 The Portland facility produces at least 700 cases
y 1 w $ 300 The Spokane facility produces at least 300 cases
x 1 y $ 600 The Seattle-area supermarket needs at least 600 cases
z 1 w $ 400 The Ellensburg supermarket needs at least 400 cases
x $ 0, y $ 0, z $ 0, w $ 0 The number of cases shipped is nonnegative
This is a standard minimization problem. We will solve it by using the dual and
the simplex method. We have
 1 0 1 0 700 
 
 0 1 0 1 300 
A 5 1 1 0 0 600 
 0 0 1 1 400 
 
 0.62 1.01 0.80 0.63 1
Thus
 1 0 1 0 0.62
 
 0 1 1 0 1.01
AT 5  1 0 0 1 0.80
 0 
1 0 1 0.63

700 300 600 400 1
The dual problem is
Maximize P 5 700x 1 300y 1 600z 1 400w
 x 1 z # 0.62

 y 1 z # 1.01
Subject to  x 1 w # 0.80

 y 1 w # 0.63
 x $ 0, y $ 0, z $ 0, w $ 0
 4.4 Solving Standard Minimization Problems with the Dual 243

The initial tableau is given by

x y z w s t u v P
 1 0 1 0 1 0 0 0 0 0.62
 
 0 1 1 0 0 1 0 0 0 1.01
 1 0 0 1 0 0 1 0 0 0.80
 0 1 0 1 0 0 0 1 0 0.63
 
2700 2300 2600 2400 0 0 0 0 1 0
The pivot is the 1 in the first row of the x column of the tableau. The second
tableau is generated with the indicated row operations. The 1 in the third row of
the w column is the new pivot.
x y z w s t u v P
1 0 1 0 1 0 0 0 0 0.62 
 
0 1 1 0 0 1 0 0 0 1.01 
0 0 21 1 21 0 1 0 0 0.18  R3 2 R1
0 1 0 1 0 0 0 1 0 0.63 
 
0 2300 100 2400 700 0 0 0 1 434  R5 1 700R1

The third tableau is generated with the indicated row operations. We may choose
either the y column or the z column to be the new pivot column. We pick the 1 in
the fourth row of the y column to be the new pivot.
x y z w s t u v P
1 0 1 0 1 0 0 0 0 0.62 
 
0 1 1 0 0 1 0 0 0 1.01 
0 0 21 1 21 0 1 0 0 0.18 
0 1 1 0 1 0 21 1 0 0.45  R4 2 R3
 
0 2300 2300 0 300 0 400 0 1 506  R5 1 400R3

The final simplex tableau is generated with the indicated row operations.
x y z w s t u v P
1 0 1 0 1 0 0 0 0 0.62
 
0 0 0 0 21 1 1 21 0 0.56 R2 2 R4
0 0 21 1 21 0 1 0 0 0.18
0 1 1 0 1 0 21 1 0 0.45
 
0 0 0 0 600 0 100 300 1 641 R5 1 300R4

The optimal solution of the dual problem is x 5 0.62, y 5 0.45, z 5 0, and

w 5 0.18 with optimal value P 5 641. We read the optimal solution for the min-
imization problem from the bottom row of the final tableau. When x 5 600,
y 5 0, z 5 100, and w 5 300, the optimal value of 641 is obtained. In the con-
text of the problem, the total shipping cost is minimized when 600 cases are
shipped from Portland to Seattle, no cases are shipped from Spokane to Seattle,
100 cases are shipped from Portland to Ellensburg, and 300 cases are shipped
from Spokane to Ellensburg. The total shipping cost is $641. (Note: This result
assumes that the minimum number of cases were produced at each facility.)
244 CHAPTER 4 Linear Programming

EXAMPLE 6 Using a Standard Minimization Problem to

Minimize Training Costs
Many companies with large government contracts use a competitive bidding
process to hire subcontractors to do much of the work, A company has a contract
to create at least 1000 hours of training and plans to hire three subcontractors
(Trainum, Teachum, and Schoolum) to help with the work. Each subcontractor re-
quires a contract for at least 200 hours of training development. Trainum charges
$250 per hour of training development, Teachum charges $300 per hour of train-
ing development, and Schoolum charges $250 per hour of training development.
How many hours should the company allocate to each subcontractor in order to
minimize costs?
SOLUTION
Let x be the number of hours allocated to Trainum.
Let y be the number hours allocated to Teachum.
Let z be the number of hours allocated to Schoolum.
We need to minimize the training development cost function,
C 5 250x 1 300y 1 250z, subject to the following constraints:
x 1 y 1 z $ 1000 At least 1000 training hours are produced
x $ 200 Trainum is allocated at least 200 hours
y $ 200 Teachum is allocated at least 200 hours
z $ 200 Schoolum is allocated at least 200 hours
This is a standard minimization problem. We will solve it by using the dual and
the simplex method.
 1 1 1 1000 
  1 1 0 0 250
 1 0 0 200 
AT 5 ≥ ¥
1 0 1 0 300
A 5 0 1 0 200  and
 0 200 
1 0 0 1 250
0 1
  1000 200 200 200 1
 250 300 250 1
The dual problem is
Maximize P 5 1000x 1 200y 1 200z 1 200w
 x 1 y # 250
 x 1 z # 300
Subject to 
 x 1 w # 250
 x $ 0, y $ 0, z $ 0, w $ 0
The initial simplex tableau is
x y z w s t u P
1 1 0 0 1 0 0 0 250

≥ ¥
1 0 1 0 0 1 0 0 300
1 0 0 1 0 0 1 0 250
21000 2200 2200 2200 0 0 0 1 0
The pivot column is the x column, since 21000 is the negative entry in the bot-
tom row with the largest magnitude. The test ratios of the first and third rows are
both equal to 250. Since this is smaller than the test ratio for the second row
 4.4 Solving Standard Minimization Problems with the Dual 245

(300), either the first or the third entry in the x column may be selected as the
pivot. We pick the first entry. Using the indicated row operations, we obtain the
second tableau.
x y z w s t u P
1 1 0 0 1 0 0 0 250

≥ ¥
0 21 1 0 21 1 0 0 50 R2 2 R1
0 21 0 1 21 0 1 0 0 R3 2 R1
0 800 2200 2200 1000 0 0 1 250000 R4 1 1000R1

Since the magnitude of the negative entries in the bottom row is the same, we may
pick either the z column or the w column to be the pivot column. We pick the w
column. The pivot is the 1 in the third row. Using the indicated row operation, we
get the third tableau. The new pivot is the entry in the second row of the z column.
x y z w s t u P
1 1 0 0 1 0 0 0 250

≥ ¥
0 21 1 0 21 1 0 0 50
0 21 0 1 21 0 1 0 0
0 600 2200 0 800 0 200 1 250000 R4 1 200R3

Using the indicated row operation, we obtain the final tableau.

x y z w s t u P
1 1 0 0 1 0 0 0 250

≥ ¥
0 21 1 0 21 1 0 0 50
0 21 0 1 21 0 1 0 0
0 400 0 0 600 200 200 1 260000 R4 1 200R2

The optimal solution for the dual problem is x 5 250, y 5 0, z 5 50, and w 5 0.
The optimal value is 260,000.
An optimal solution for the minimization problem is x 5 600, y 5 200, and
z 5 200 with optimal value 260,000. In the context of the problem, the training
development costs are minimized if Trainum develops 600 hours of training,
Teachum develops 200 hours of training, and Schoolum develops 200 hours of
training. The minimum training costs are $260,000. Since the hourly training de-
velopment costs for Trainum and Schoolum are the same, another optimal solu-
tion is (200, 200, 600) . In fact, any point of the form (t, 200, 800 2 t) for
0 # t # 800 will yield an optimal solution.

As shown in Example 6, a standard minimization problem may have more

than one optimal solution. Each solution, however, will generate the same opti-
mal value. The method of using the dual guarantees that we will find an optimal
solution, if there is one.

EXAMPLE 7 Using a Standard Minimization Problem to Minimize Food

Costs
PETsMART.com, an online retailer, allows pet owners to quickly compare nutri-
tion and pricing for different brands of pet food sold on its web site. A June 2003
query revealed the data given in Table 4.23.
246 CHAPTER 4 Linear Programming

TABLE 4.23
Price (dollars per
Brand Protein (percent) Fiber (percent) 20-lb bag)
Eukanuba Adult
Maintenance Formula 25 5 18.99
Nutro Natural Choice
Plus 27 3 17.99
Science Diet Active
Formula Canine
Maintenance 26.5 3.5 17.99
Source: www.petsmart.com.

A dog breeder wants to create at least 300 pounds of a dog food mix that is
at least 26 percent protein and 4 percent fiber while minimizing dog food cost.
How many 20-pound bags of each type of dog food should the breeder buy?
SOLUTION
Let x be the number of bags of the Eukanuba brand.
Let y be the number of bags of the Nutro brand.
Let z be the number of bags of the Science Diet brand.
To make computations easier, we will convert the percentages per bag to pounds
per bag by multiplying the percentage per bag by 20 pounds and dividing by 100,
as shown in Table 4.24.

TABLE 4.24
Brand Protein (pounds) Fiber (pounds)
Eukanuba Adult
Maintenance Formula 5.0 1.0
Nutro Natural Choice Plus 5.4 0.6
Science Diet Active Formula
Canine Maintenance 5.3 0.7

The total weight of the mix is 20x 1 20y 1 20z pounds.

The total amount of protein in the mix is 5.0x 1 5.4y 1 5.3z pounds. We want
the mix to be at least 26 percent protein. Thus we require that
0.26(20x 1 20y 1 20z) # 5.0x 1 5.4y 1 5.3z
5.2x 1 5.2y 1 5.2z # 5.0x 1 5.4y 1 5.3z
0.2x 2 0.2y 2 0.1z # 0
20.2x 1 0.2y 1 0.1z $ 0
 4.4 Solving Standard Minimization Problems with the Dual 247

The total amount of fiber in the mix is 1.0x 1 0.6y 1 0.7z pounds. We want the
mix to be at least 4 percent fiber. Thus we require that
0.04(20x 1 20y 1 20z) # 1.0x 1 0.6y 1 0.7z
0.8x 1 0.8y 1 0.8z # 1.0x 1 0.6y 1 0.7z
20.2x 1 0.2y 1 0.1z # 0
0.2x 2 0.2y 2 0.1z $ 0
For computational ease, we round the prices to the nearest dollar. We must mini-
mize C 5 19x 1 18y 1 18z subject to the following constraints:
20x 1 20y 1 20z $ 300 At least 300 pounds must be purchased
20.2x 1 0.2y 1 0.1z $ 0 The mix is at least 26 percent protein
0.2x 2 0.2y 2 0.1z $ 0 The mix is at least 4 percent fiber
x $ 0, y $ 0, z $ 0 A nonnegative number of bags are purchased
We have
20 20 20 300 20 20.2 0.2 19

A5 ≥ ¥ AT 5 ≥ ¥
20.2 0.2 0.1 0 20 0.2 20.2 18
and
0.2 20.2 20.1 0 20 0.1 20.1 18
19 18 18 1 300 0 0 1
The dual problem is
Maximize P 5 300x 1 0y 1 0z
20x 2 0.2y 1 0.2z # 19

µ
20x 1 0.2y 2 0.2z # 18
Subject to
20x 1 0.1y 2 0.1z # 18
x $ 0, y $ 0, z $ 0
with the corresponding initial tableau
x y z s t u P
20 20.2 0.2 1 0 0 0 19

≥ ¥
20 0.2 20.2 0 1 0 0 18
20 0.1 20.1 0 0 1 0 18
2300 0 0 0 0 0 1 0
The x column is the pivot column. Either the second or the third entry in the x col-
umn may be used as a pivot. Using the entry in the second row of the x column
as a pivot and the indicated row operations, we obtain the second tableau. The
entry in the third row of the z column is the new pivot, since it has the smallest
test ratio.
x y z s t u P
0 20.4 0.4 1 21 0 0 1 R1 2 R2

≥ ¥
20 0.2 20.2 0 1 0 0 18
0 20.1 0.1 0 21 1 0 0 R3 2 R2
0 3 23 0 15 0 1 270 R4 1 15R2
248 CHAPTER 4 Linear Programming

The indicated row operations give the next tableau. The new pivot is the 3 in the
first row of the t column.
x y z s t u P
0 0 0 1 3 24 0 1 R1 2 4R3

≥ ¥
20 0 0 0 21 2 0 18 R2 1 2R3
0 20.1 0.1 0 21 1 0 0
0 0 0 0 215 30 1 270 R4 1 30R3

The indicated row operations give the next tableau.

x y z s t u P
0 0 0 1 3 24 0 1

≥ ¥
60 0 0 1 0 2 0 55 3R2 1 R1
0 20.3 0.3 1 0 21 0 1 3R3 1 R1
0 0 0 5 0 10 1 275 R4 1 5R1

The final simplex tableau is obtained by converting the nonzero entry of each col-
umn with a single nonzero entry to a 1 by using the indicated row operations.
x y z s t u P
 1 4 1 1
0 0 0 1 23 0 R
3 3 3 1
1 1 1 11 1
0 0 0 0 R
 60 30 12 60 2
0 10 10 10 1
21 1 0 23 0 R
 3 3 0.3 3
0 0 0 5 0 10 1 275
Setting the inactive variable columns s and u to 0 yields
11 10 1
x5 2y 1 z 5 t5 P 5 275
12 3 3
10
z5y1
3

A 11 , r, r 1 3 B for nonnegative r maximizes the objec-

10
Any solution of the form 12
tive function of the dual problem. The optimal value is 275.
The minimization problem has the solution x 5 5, y 5 0, and z 5 10. When
5 bags of the Eukanuba brand and 10 bags of the Science Diet brand are pur-
chased, the total cost is $275. This is the minimum cost for a mix that meets the
breeder’s nutrition requirements.

4.4 Summary

In this section, you learned how to solve standard minimization problems using
the notion of the dual problem. You discovered that finding the maximum solu-
tion of a dual problem leads us to the minimum solution of the minimization
problem.
 4.4 Solving Standard Minimization Problems with the Dual 249

4.4 Exercises

In Exercises 1–10, determine if the problem is a 8. Minimize P 5 4x 1 4y 1 9z

standard minimization problem. If it isn’t, explain why. 23x 2 2y 1 4z # 222

1. Minimize P 5 9x 1 8y  5x 2 2.5y 1 6z $ 36
 2x 1 y $ 10 Subject to  x$ 0
  y$ 0
Subject to  2x 1 y $ 1 
 x$0  z$ 0
 y$0 9. Minimize P 5 28x 1 2y 2 z
2. Minimize P 5 22x 2 5y  z$4
 4x 1 9y $ 10  x 2 y $3
Subject to 
 2 $5
Subject to  211x 1 y $ 221  2x z
 x$0 x 1 y 1 z # 0
 y$0 10. Minimize P 5 5x 1 7z
3. Minimize P 5 6xy  4x 1 2y $ 16
 6x 1 
7y $ 13 Subject to  4x 2 2y $ 3

Subject to  28x 2 4y $ 12  x$0
 x$0  y$0
 y$0
In Exercises 11–20, find the transpose of the given
4. Minimize P 5 4x 2 2y 1 z matrix.
 2x 1 y 1 4z $ 24
 2 2 1 4 1 1 1 4
11. A 5 £ 5 3§ 12. A 5 £ 1 5§
 22x 1 3y 1 3z $ 150
9 7 1 1
Subject to  x$0
 8 1 0 1 4 5 1 1
y$0
 8 2 1
 z$0
13. A 5 c d 14. A 5 £ 2 7§
1 2 4
5
5. Minimize P 5 21.2x 2 2.8y 1 4.3z 6 8 1
1 7 1
Subject to e 15. A 5 c d
3.2x 1 1.5y 1 7.4z $ 249.8 9 3 21 4
22.7x 1 3.4y 1 3.9z $ 190.1 7 2 22 1
6. Minimize P 5 4x 1 4y 1 9z 1 22 1 4 11
 26x 2 y 1 4z # 224 16. A 5 £ 25 4 27 3 9§
 3 7 9 3 1
 29x 2 1.5y 1 6z $ 36
Subject to  x$0 9 2 18
 y$0
17. A 5 ≥ ¥
 23 22 6
 z$0 4 0 0
7. Minimize P 5 x 1 9y 1 8z 5 6 1
 26x 2 y 1 4z $ 224 6 5 4 3 2

 29x 2 1.5y 1 6z $ 36 18. A 5 £ 5 6 7 8 9§
Subject to  x$0 4 5 6 5 1
 y$0

 z$0
250 CHAPTER 4 Linear Programming

 0 2 1 27. Minimize P 5 3x 1 2y
 
25 4 7 x1y$ 5
19. A 5  8 1 0
Subject to µ
3x 1 2y $ 11
21 2 1
  4x 2 y $ 0
4 2 1 x$ 0, y $ 0
20. A 5 32 4 6 5 14 28. Minimize P 5 9x 1 7y
In Exercises 21–30, do the following: 5x 1 3y $ 8

Subject to µ
(i) Write the dual problem for the given standard 3x 1 5y $ 8
minimization problem. 8x 2 8y $ 0
(ii) Solve the dual problem using the simplex method. x $ 0, y $ 0
(iii) Use the final simplex tableau of the dual problem
to solve the standard minimization problem. 29. Minimize P 5 6x 1 5y
(iv) Check your answer by graphing the feasible x1y$4
region of the standard minimization problem and
Subject to µ
x2y$2
calculating the value of the objective function at
2x 2 y $ 6
each of the corner points.
x $ 0, y $ 0
21. Minimize P 5 x 1 2y
30. Minimize P 5 5x 1 7y
3x 1 y $ 6
Subject to • 2x 1 3y $ 11
x1y$4

Subject to µ
x2y$2
x $ 0, y $ 0
2x 2 y $ 6
22. Minimize P 5 4x 1 2y x $ 0, y $ 0
5x 1 7y $ 19
Subject to • 22x 1 3y $ 5 In Exercises 31–40, set up and solve the standard
minimization problem.
x $ 0, y $ 0
31. Pet Nutrition: Food Cost
23. Minimize P 5 9x 1 6y
PETsMART.com sold the following
5x 1 2y $ 30 varieties of dog food in June 2003. The price
Subject to • 10x 2 5y $ 15 shown is for an 8-pound bag.
x $ 0, y $ 0 Pro Plan Adult Chicken & Rice Formula,
25 percent protein, 3 percent fiber, $7.99
24. Minimize P 5 4x 1 4y Pro Plan Adult Lamb & Rice Formula,
3x 1 4y $ 18 28 percent protein, 3 percent fiber, $7.99
Subject to • 5x 1 3y $ 19 Pro Plan Adult Turkey & Barley Formula,
26 percent protein, 3 percent fiber, $8.49
x $ 0, y $ 0
(Source: www.petsmart.com.)
25. Minimize P 5 5x 1 20y A dog breeder wants to make at least 120
pounds of a mix containing at least 27 percent
25x 1 2y $ 10
Subject to •
protein and at least 3 percent fiber. How many
5x 1 y $ 25 8-pound bags of each dog food variety should the
x $ 0, y $ 0 breeder buy in order to minimize cost? (Round
26. Minimize P 5 2x 1 5y prices up to the nearest dollar.)

x1y$9
Subject to • 2x 2 2y $ 6
x $ 0, y $ 0
 4.4 Solving Standard Minimization Problems with the Dual 251

32. Pet Nutrition: Food Cost A dog breeder wants to make at least
PETsMART.com sold the following 210 pounds of a mix containing at least 25 percent
varieties of dog food in June 2003. protein. How many bags of each dog food variety
Authority Chicken Adult Formula, 32 percent should the breeder buy in order to minimize fat
protein, 3 percent fiber, $19.99 per 33-pound content?
bag 35. Food Distribution Cost Wal-Mart
Bil-Jac Select Dog Food, 27 percent protein, Stores, Inc., has food distribution centers
4 percent fiber, $18.99 per 18-pound bag in Monroe, Georgia, and Shelbyville, Tennessee,
Iams Minichuncks, 26 percent protein, 4 percent and Wal-Mart Supercenters in Birmingham,
fiber, $8.99 per 8-pound bag Alabama, and Scottsboro, Alabama. (Source:
(Source: www.petsmart.com.) www.walmart.com.) Suppose that the Monroe
A dog kennel wants to make at least 2178 distribution center must ship at least 600 cases of
pounds of a mix containing at least 29 percent peanut butter weekly and the Shelbyville
protein and at least 3.5 percent fiber. How many distribution center must ship at least 400 cases of
bags of each dog food variety should the kennel peanut butter weekly.* If the Birmingham store
buy in order to minimize cost? (Round prices up requires at least 700 cases of peanut butter weekly
to the nearest dollar.) and the Scottsboro store requires at least 300
33. Pet Nutrition: Food Cost cases of peanut butter weekly, what shipment plan
PETsMART.com sold the following will minimize the distribution cost?
varieties of dog food in June 2003. Estimated Distribution Cost per Case
Nature’s Recipe Venison Meal & Rice Canine,
Monroe Shelbyville
20 percent protein, 10 percent fat, $21.99 per
20-pound bag Birmingham $0.68 $0.57
Nutro Max Natural Dog Food, 27 percent Scottsboro $0.65 $0.32
protein, 16 percent fat, $12.99 per 17.5-pound
bag *Distribution costs and amounts are hypothetical.
PETsMART Premier Oven Baked Lamb
Recipe, 25 percent protein, 14 percent fat, 36. Food Distribution Cost Wal-Mart
$22.99 per 30-pound bag Stores, Inc., has food distribution centers
(Source: www.petsmart.com.) in Monroe, Georgia, and Shelbyville, Tennessee,
A dog breeder wants to make at least and Wal-Mart Supercenters in Birmingham,
175 pounds of a mix containing at least 25 Alabama, and Calhoun, Georgia. (Source:
percent protein and at least 14 percent fat. How www.walmart.com.) Suppose that the Monroe
many bags of each dog food variety should the distribution center must ship at least 600 cases
breeder buy in order to minimize cost? (Round of pickles weekly and the Shelbyville distribution
prices up to the nearest dollar.) center must ship at least 700 cases of pickles
34. Pet Nutrition: Fat Content
weekly.* If the Birmingham store requires at
PETsMART.com sold the following least 1000 cases of pickles weekly and the
varieties of dog food in June 2003. Calhoun store requires at least 300 cases of
pickles weekly, what shipment plan will minimize
Nature’s Recipe Venison Meal & Rice Canine, the distribution cost?
20 percent protein, 10 percent fat, $21.99 per
20-pound bag Estimated Distribution Cost per Case
Nutro Max Natural Dog Food, 27 percent
Monroe Shelbyville
protein, 16 percent fat, $12.99 per 17.5-pound
bag Birmingham $0.68 $0.57
PETsMART Premier Oven Baked Lamb Calhoun $0.39 $0.50
Recipe, 25 percent protein, 14 percent fat,
$22.99 per 30-pound bag *Distribution costs and amounts are hypothetical.
(Source: www.petsmart.com.)
252 CHAPTER 4 Linear Programming

37. Food Distribution Cost Wal-Mart 39. Marital Harmony It costs a couple 3 hours and
Stores, Inc., has food distribution centers $200 to go clothes shopping together. She gives
in Monroe, Georgia, and Shelbyville, Tennessee, shopping a fun rating of 10 points, while he gives
and Wal-Mart Supercenters in Scottsboro, it a fun rating of 1 point (a total of 11 fun points
Alabama, and Calhoun, Georgia. (Source: for the couple).
www.walmart.com.) Suppose that the Monroe It costs a couple 4 hours and $250 to go to a
distribution center must ship at least 400 cases of major league baseball playoff game. She gives the
potato chips weekly and the Shelbyville game a fun rating of 4 points, while he gives it a
distribution center must ship at least 200 cases of fun rating of 10 points (a total of 14 fun points for
potato chips weekly.* If the Scottsboro store the couple).
requires at least 300 cases of potato chips weekly The couple wants to spend at least 10 hours
and the Calhoun store requires at least 300 cases together (shopping and watching baseball) while
of potato chips weekly, what shipment plan will earning at least 36 fun points. She insists that they
minimize the distribution cost? go shopping at least twice. How shall they spend
their time if they want to minimize their financial
Estimated Distribution Cost per Case
costs?
Monroe Shelbyville 40. Marital Discord It costs a couple 3 hours to go
Calhoun $0.39 $0.50 clothes shopping together. He gives shopping with
her a fun rating of 6 points while she gives it a
Scottsboro $0.65 $0.32
fun rating of 1 point (a total of 7 fun points for
*Distribution costs and amounts are hypothetical. the couple).
It costs a couple 4 hours to go to a major
38. Food and Entertainment It costs $35 for a league baseball game. She gives going to a game
family to dine out and $10 for the family to eat at with him a fun rating of 7 points while he gives it
home. Eating out has a fun rating of 10 points, a fun rating of 2 points (a total of 9 fun points for
while eating at home has a fun rating of 2 points. the couple).
It costs $32 for a family to play a game of The couple wants to earn at least 30 fun
miniature golf and $30 to watch a movie. points. She refuses to go shopping with him
Miniature golf has a fun rating of 8 points, and unless he goes with her to at least one game. How
watching a movie has a fun rating of 6 points. shall they spend their time if they want to
The family must eat at least 21 meals weekly minimize their time together? (Things are not
and must go out for food or miniature golf at least going well in the relationship.)
five times weekly.The family wants to earn at least
82 fun points per week. How many times a week
should the family participate in each activity in
order to minimize food and entertainment costs?

4.5 Solving General Linear Programming Problems

with the Simplex Method

■ Solve general linear

programming problems An investor plans to invest at most $3000 in the
with the simplex method three publicly traded recreational vehicle companies shown in Table 4.25.
 4.5 Solving General Linear Programming Problems with the Simplex Method 253

TABLE 4.25
Company Share Price (dollars) Dividend/Share (dollars)
Harley- Davidson, Inc. (HDI) 62.70 0.40
Polaris Industries Inc. (PII) 50.50 0.92
Winnebago Industries Inc. (WGO) 33.42 0.20
Source: moneycentral.msn.com. (Accurate as of July 16, 2004.)

He wants to earn at least $50 in dividends while maximizing the number of shares
purchased. How many shares of each company should he purchase?
At first glance, this looks like a standard maximization problem. However, a
closer analysis reveals that it is not. While the investment constraint is of the form
Ax 1 By 1 Cz 1 c # M, the dividend constraint is of the form Ax 1 By 1 Cz 1
c $ M instead of Ax 1 By 1 Cz 1 c # M. This is a general linear programming
problem or a linear programming problem with mixed constraints.
In this section, we will show how to solve general linear programming problems.
We will demonstrate how minimization problems may be solved by maximizing the
negative of the objective function. We will return to the recreational vehicle invest-
ment problem in Example 3.

Recall that the feasible region of any standard maximization problem always
includes the origin. Graphically speaking, the simplex method starts at the origin
and moves from corner point to adjacent corner point, each time increasing the
value of the objective function until the maximum value is reached. The simplex
method works with standard maximization problems because we are guaranteed
that the origin is in the feasible region. But what if the origin is outside of the
feasible region? The feasible region of a linear programming problem with mixed
constraints often does not contain the origin.
A linear programming problem with mixed constraints has constraints in two
or more of the following forms: Ax 1 By 1 Cz 1 c # M , Ax 1 By 1 Cz 1
c $ M , or Ax 1 By 1 Cz 1 c 5 M . In all cases, M $ 0. Consider the fol-
lowing linear programming problem with mixed constraints.
Maximize P 5 3x 1 2y
2x 1 2y # 4

Subject to µ
2x 2 y # 4
x1y$5
x $ 0, y $ 0
Since the constraint x 1 y $ 5 contains a $ sign instead of a # sign, this is
not a standard maximization problem. It is a problem with mixed constraints. The
graph of the feasible region is shown in Figure 4.28.
254 CHAPTER 4 Linear Programming

y
x=0
2x − y = 4
5
−x + 2y = 4
4 (4, 4)

3 (2, 3)

2 (3, 2)
x+y=5
1
y=0

0 x
0 1 2 3 4 5
FIGURE 4.28

Notice that the origin is not contained within the feasible region. Our strat-
egy for solving linear programming problems with mixed constraints is a two-
stage process.
Stage 1: Get inside the feasible region.
Stage 2: Solve the problem with the simplex method.
Since our problem contains only two decision variables, we have the luxury
of observing how the process works graphically. This is not the case in problems
involving three or more decision variables.
We set up the problem for the simplex method as usual by adding in slack
variables to constraints of the form Ax 1 By 1 Cz 1 c # M .
2x 1 2y 1 s 5 4
2x 2 y 1 t 5 4
Can we add in slack variables to constraints of the form Ax 1 By 1 Cz 1
c $ M ? Let’s see. We have the constraint x 1 y $ 5. Recall that slack vari-
ables add in what is necessary to make the inequality an equality. By definition,
slack variables must be nonnegative. What must we do to the inequality
x 1 y $ 5 to make it an equality? Adding a slack variable to the inequality will
increase instead of decrease the value of the left-hand side. Since the left-hand
side of the inequality is greater than or equal to 5, we must subtract some non-
negative value u in order to make x 1 y equal to 5. That is,
x1y2u55
Since this variable takes away the surplus, it is referred to as a surplus variable.
We have
2x 1 2y 1 s 5 4
2x 2 y 1 t 5 4
x1y2u55
23x 2 2y 1 P 5 0
4.5 Solving General Linear Programming Problems with the Simplex Method 255

and the corresponding initial tableau

x y s t u P
21 2 1 0 0 0 4

≥ ¥
2 21 0 1 0 0 4
1 1 0 0 21 0 5
23 22 0 0 0 1 0
Setting the x and y variables equal to zero yields the solution to this tableau.
x50 y50 s54 t54 2u 5 5 P50
u 5 25
Recall that all decision, slack, and surplus variables are required to be nonnega-
tive. Since u , 0, this solution is not in the feasible region. This may be easily
seen graphically (Figure 4.29). The origin is not in the feasible region.

y
x=0
2x − y = 4
5
−x + 2y = 4
4 (4, 4)

3 (2, 3)

2 (3, 2)
Solution to x+y=5
1 tableau #1
y=0
(0, 0)
0 x
0 1 2 3 4 5
FIGURE 4.29

In the initial tableau, columns s, t, u, and P were the active variable columns.
Since the active variable u was negative, we want to modify the tableau in such a
way as to make u an inactive variable. We place a * on the left of the tableau to
highlight the rows above the bottom row that resulted in negative variable values.
In this case, the third row was the only one with a negative variable value.
x y s t u P
21 2 1 0 0 0 4

≥ ¥
2 21 0 1 0 0 4
* 1 1 0 0 21 0 5
23 22 0 0 0 1 0
Since all surplus variables in the initial tableau are negative, each row of the
initial tableau containing surplus variables will be starred. We select the column
with the largest positive entry in the starred row as the pivot column. Since the
positive values in the starred row are equal, we may select either the x column or
the y column to be the pivot column. We select the x column. To determine the
256 CHAPTER 4 Linear Programming

pivot, we calculate the test ratios and select the row with the smallest test ratio.
This will ensure that all nonnegative variables will remain nonnegative.
x y s t u P
21 2 1 0 0 0 4

≥ ¥
2 21 0 1 0 0 4 4>2 5 2
* 1 1 0 0 21 0 5 5>1 5 5
23 22 0 0 0 1 0
n
Pivot column

The 2 in the second row of the x column is the pivot. We modify the initial tableau
with the indicated operations and star any rows corresponding with negative ac-
tive variables.
x y s t u P
0 3 2 1 0 0 12 2R1 1 R2

≥ ¥
2 21 0 1 0 0 4
* 0 3 0 21 22 0 6 2R3 2 R2
0 27 0 3 0 2 12 2R4 1 3R2

2x 5 4 y50 2s 5 12 t50 22u 5 6 2P 5 12

x52 s56 u 5 23 P56
We starred the third row, since the value of u is negative. Although u is still neg-
ative, it is less negative than it was before. We will continue the process until u is
nonnegative. However, before continuing, let’s examine what is happening graph-
ically (Figure 4.30).

y
x=0
2x − y = 4
5
−x + 2y = 4
4 (4, 4)

3 (2, 3)

2 (3, 2)
Solution to x+y=5
1 tableau #2
y=0
(2, 0)
0 x
0 1 2 3 4 5
FIGURE 4.30

We have moved from the origin to the x-intercept of one of the constraints.
However, we still are outside of the feasible region. This is due to the fact that the
surplus variable u is still negative. We will pivot again and further increase the
value of u.
4.5 Solving General Linear Programming Problems with the Simplex Method 257

The largest value in the starred row is the 3 in the y column. The y column is
the new pivot column.
x y s t u P
0 3 2 1 0 0 12 12>3 5 4

≥ ¥
2 21 0 1 0 0 4
* 0 3 0 21 22 0 6 6>3 5 2
0 27 0 3 0 2 12
n
Pivot column

The pivot is the 3 in the third row of the y column, since that row has the small-
est test ratio. We zero out the remaining terms in the column with the indicated
operations and read the solution from the tableau.
x y s t u P
0 0 2 2 2 0 6 R1 2 R3

≥ ¥
6 0 0 2 22 0 18 3R2 1 R3
0 3 0 21 22 0 6
0 0 0 2 214 6 78 3R4 1 7R3

6x 5 18 3y 5 6 2s 5 6 t50 u50 6P 5 78
x53 y52 s53 P 5 13
Since none of the variables are negative, this solution is in the feasible region
(Figure 4.31). Since we are now in the feasible region, this completes Stage 1 of
the process.

y
x=0
2x − y = 4
5
−x + 2y = 4
4 (4, 4)

3 (2, 3)

2 Solution to (3, 2)
tableau #3 x+y=5
1
y=0

0 x
0 1 2 3 4 5
FIGURE 4.31

Stage 2 allows us to apply the simplex method to find the optimal solution.
The u column is the pivot column. The 2 in the first row of that column is the
pivot, since it is the only nonnegative value in the column.
x y s t u P
0 0 2 2 2 0 6

≥ ¥
6 0 0 2 22 0 18
0 3 0 21 22 0 6
0 0 0 2 214 6 78
n
Pivot column
258 CHAPTER 4 Linear Programming

We zero out the remaining terms in the u column using the indicated operations.
y
x=0
2x − y = 4 x y s t u P
5
Solution to 0 0 2 2 2 0 6
tableau #4
≥ ¥
4 (4, 4) 6 0 2 4 0 0 24 R2 1 R1

(2, 3)
0 3 2 1 0 0 12 R3 1 R1
3
0 0 14 16 0 6 120 R4 1 7R1
2 (3, 2)
6x 5 24 3y 5 12 s50 t50 2u 5 6 6P 5 120
−x + 2y = 4 x+y=5
1
y=0
x54 y54 u53 P 5 20

0 Since the bottom row of the tableau does not contain any negative values, this
x
0 1 2 3 4 5 is the final simplex tableau. The solution that maximizes the objective function
FIGURE 4.32 P 5 3x 1 2y is x 5 4 and y 5 4 (Figure 4.32). The maximum value of the ob-
jective function is P 5 20.
The following box details the steps used to solve the general linear program-
ming problem.

Solving General Linear Programming Problems

Stage 1: Get inside the feasible region

1. Star all rows that correspond with a negative value of a decision, slack, or
surplus variable.
2. Identify the largest positive entry in the starred row. The corresponding
column is the pivot column.
3. Calculate the test ratios for the positive entries above the bottom row of
the pivot column.
4. Pick the pivot column entry with the smallest test ratio as the pivot.
5. Row reduce the tableau using operations of the form aRc 6 bRp S Rc with
a and b positive. (Recall that Rc is the row we want to change and Rp is the
pivot row.)
6. From the tableau, calculate the value of all decision, slack, and surplus
variables. If any of the variables are negative, repeat Steps 1 through 5 for
the new tableau. Otherwise, go to Stage 2.
Stage 2: Solve the maximization problem with the simplex method.

EXAMPLE 1 Solving a Linear Programming Problem with No

Optimal Solution
Graph the feasible region associated with the given linear programming problem.
Then solve the problem using the two-stage method described previously, indi-
cating on the graph the solution that corresponds with each tableau.
Maximize P 5 3x 1 10y
4x 2 y $ 11

Subject to µ
x 1 2y $ 5
2x 1 y # 1
x $ 0, y $ 0
 4.5 Solving General Linear Programming Problems with the Simplex Method 259

y SOLUTION The graph of the feasible region is shown in Figure 4.33.

x=0
We can see from the graph that the feasible region is unbounded. Since the
5 (4, 5) objective function has positive coefficients and the region is unbounded, the ob-
−x + y = 1
4
jective function will not have a maximum value. Nevertheless, we will proceed
with the two-stage method to discover what indicators in the tableau let us know
x + 2y = 5
3 that there is no objective function maximum.
4x − y = 11
We add the slack and surplus variables and rewrite the objective function.
2 We have
(3, 1) 4x 2 y 2 s 5 11
1
y=0
(5, 0) x 1 2y 2 t 5 5
0 x
0 1 2 3 4 5 2x 1 y 1 u 5 1
FIGURE 4.33
23x 2 10y 1 P 5 0
and the corresponding initial tableau
x y s t u P
4 21 21 0 0 0 11

≥ ¥
1 2 0 21 0 0 5
21 1 0 0 1 0 1
23 210 0 0 0 1 0
Reading from the tableau, we have
x50 y50 2s 5 11 2t 5 5 u51 P50
s 5 211 t 5 25
Since both s and t are negative, we will star the corresponding rows.
x y s t u P
* 4 21 21 0 0 0 11

≥ ¥
* 1 2 0 21 0 0 5
21 1 0 0 1 0 1
23 210 0 0 0 1 0
We may choose either starred row to work with. We pick the row corresponding
with the variable t (second row). The largest value in the labeled columns of the
second row is the 2 in the y column. Consequently, the y column is the pivot col-
umn. We calculate the test ratios and locate the pivot.
x y s t u P
* 4 21 21 0 0 0 11

≥ ¥
* 1 2 0 21 0 0 5 5>2 5 2.5
21 1 0 0 1 0 1 1>1 5 1
23 210 0 0 0 1 0
n
Pivot column
260 CHAPTER 4 Linear Programming

We zero out the remaining terms in the pivot column using the indicated opera-
tions and star the rows that correspond with negative variable values.
x y s t u P
* 3 0 21 0 1 0 12 R1 1 R3

≥ ¥
* 3 0 0 21 22 0 3 R2 2 2R3
21 1 0 0 1 0 1
213 0 0 0 10 1 10 R4 1 10R3
x50 y51 2s 5 12 2t 5 3 u50 P 5 10
s 5 212 t 5 23
We observe that although t remains negative, it has become less negative. We will
again select the largest value in a labeled column of the second row. The 3 in the
x column is the largest value. Consequently, the x column is the pivot column.
x y s t u P
* 3 0 21 0 1 0 12 12>3 5 4

≥ ¥
* 3 0 0 21 22 0 3 3>3 5 1
21 1 0 0 1 0 1
213 0 0 0 10 1 10
n
Pivot column

The 3 in the second row of the x column is the pivot, since the second row has the
smallest test ratio. We zero out the remaining entries in the x column using the in-
dicated operations. We then star the row corresponding to the negative variable.
x y s t u P
* 0 0 21 1 3 0 9 R1 2 R2

≥ ¥
3 0 0 21 22 0 3
0 3 0 21 1 0 6 3R3 1 R2
0 0 0 213 4 3 69 3R4 1 13R2
3x 5 3 3y 5 6 2s 5 9 t50 u50 3P 5 69
x51 y52 s 5 29 P 5 23
Observe that the surplus variable t is now nonnegative. We need to continue
the process to make s nonnegative. The largest value in the labeled columns
of the starred row is the 3 in the u column. Consequently, the u column is the pivot
column.
x y s t u P
* 0 0 21 1 3 0 9 9>3 5 3

≥ ¥
3 0 0 21 22 0 3
0 3 0 21 1 0 6 6>1 5 6
0 0 0 213 4 3 69
n
Pivot column
 4.5 Solving General Linear Programming Problems with the Simplex Method 261

Using the 3 in the first row of the u column as a pivot, we zero out the remaining
column entries using the indicated operations.
y
x=0 x y s t u P
5 (4, 5) 0 0 21 1 3 0 9
−x + y = 1
≥ ¥
9 0 22 21 0 0 27 3R2 1 2R1
4
0 9 1 24 0 0 9 3R3 2 R1
x + 2y = 5
3 0 0 4 243 0 9 171 3R4 2 4R1
4x − y = 11 9x 5 27 9y 5 9 s50 t50 3u 5 9 9P 5 171
2
x53 y51 u53 P 5 19
(3, 1)
1
y=0
Since all decision, slack, and surplus variables are nonnegative, we are in the
(5, 0)
0 x feasible region. We can now proceed to Stage 2 of the process: applying the sim-
0 1 2 3 4 5
plex method. Before going on, let’s observe what has happened graphically (see
FIGURE 4.34 Figure 4.34).
The blue dots on the graph indicate the solutions to the various tableaus. We
started at (0, 0) , proceeded to (0, 1) , then to (1, 2) , and finally to (3, 1) .
Returning to the tableau, we apply the simplex method. The only negative
entry in the bottom row is in the t column, so the t column is our pivot column.
The only positive entry in the t column is the 1 in the first row, so that is our pivot.
x y s t u P
0 0 21 1 3 0 9

≥ ¥
9 0 22 21 0 0 27
0 9 1 24 0 0 9
0 0 4 243 0 9 171
n
Pivot column

We zero out the remaining entries in the t column with the indicated operations.
y
x=0
x y s t u P
5 (4, 5)
−x + y = 1 0 0 21 1 3 0 9

≥ ¥
4 9 0 23 0 3 0 36 R2 1 R1
x + 2y = 5 0 9 23 0 12 0 45 R3 1 4R1
3
0 0 239 0 129 9 558 R4 1 43R1
4x − y = 11
2 9x 5 36 9y 5 45 s50 t59 u50 9P 5 558
(3, 1) x54 y55 P 5 62
1
y=0

0
(5, 0) The s column is the new pivot column. However, we are unable to select a pivot
x
0 1 2 3 4 5 because every entry in the s column is negative. This signifies that the objective
FIGURE 4.35 function has no optimal solution. Figure 4.35 shows what has happened graphically.
We increased the value of the objective function by moving from the corner
point (3, 1) to the corner point (4, 5) . Out of all of the corner points, this is the
corner point that yields the largest value of the objective function. However, since
the region is unbounded, we can continue to increase the value of the objective
function indefinitely.
262 CHAPTER 4 Linear Programming

LINEAR PROGRAMMING PROBLEMS WITH NO

OPTIMAL SOLUTIONS
If none of the entries in the pivot column of a simplex tableau are positive,
the corresponding feasible region is unbounded and the objective function
has no optimal solution.

EXAMPLE 2 Solving a General Linear Programming Problem

Solve the general linear programming problem:
Maximize P 5 4x 1 3y 1 2z
2x 2 y 1 z $ 1

Subject to µ
22x 1 y 1 z $ 3
2x 1 y 1 z # 3
x $ 0, y $ 0, z $ 0

SOLUTION We add the slack and surplus variables and rewrite the objective
function.
2x 2 y 1 z 2 s 5 1
22x 1 y 1 z 2 t 5 3
2x 1 y 1 z 1 u 5 3
24x 2 3y 2 2z 1 P 5 0
The corresponding initial tableau is shown. We calculate the basic solution and
star the rows with negative variable values.
x y z s t u P
* 21 21 1 21 0 0 0 1

≥ ¥
* 22 1 1 0 21 0 0 3
21 1 1 0 0 1 0 3
24 23 22 0 0 0 1 0
x50 y50 z50 2s 5 1 2t 5 3 u53 P50
s 5 21 t 5 23
We may work with either starred row. We choose the second row. (This row cor-
responds to t 5 23.) The largest positive entries in the labeled columns of the
second row are the 1 in the y column and the 1 in the z column. We may pick ei-
ther column as the pivot column. We pick the y column. We calculate the test ra-
tios and locate the pivot. The test ratios are equal, so we may pick either the
second or the third row as the pivot row. We choose the second row.
x y z s t u P
21 21 1 21 0 0 0 1

≥ ¥
22 1 1 0 21 0 0 3 3>1 5 3
21 1 1 0 0 1 0 3 3>1 5 3
24 23 22 0 0 0 1 0
n
Pivot column
4.5 Solving General Linear Programming Problems with the Simplex Method 263

We zero out the y column using the indicated operations and star the row corre-
sponding with a negative variable value.
x y z s t u P
* 23 0 2 21 21 0 0 4 R1 1 R2

≥ ¥
22 1 1 0 21 0 0 3
1 0 0 0 1 1 0 0 R3 2 R2
210 0 1 0 23 0 1 9 R4 1 3R2
x50 y53 z50 2s 5 4 t50 u50 P59
s 5 24
The 2 in the z column is the largest positive value in the starred row. Conse-
quently, the z column is the pivot column. We calculate the test ratios and locate
the pivot.
x y z s t u P
23 0 2 21 21 0 0 4 4>2 5 2

≥ ¥
22 1 1 0 21 0 0 3 3>1 5 3
1 0 0 0 1 1 0 0
210 0 1 0 23 0 1 9
n
Pivot column

We zero out the z column with the indicated operations.

x y z s t u P
23 0 2 21 21 0 0 4

≥ ¥
21 2 0 1 21 0 0 2 2R2 2 R1
1 0 0 0 1 1 0 0
217 0 0 1 25 0 2 14 2R4 2 R1
x50 2y 5 2 2z 5 4 s50 t50 u50 2P 5 14
y51 z52 P57
Since all of the decision, slack, and surplus variables are positive, we are in
the feasible region. We may now move to Stage 2 and apply the simplex method.
Since the x column contains the negative value in the bottom row with the largest
magnitude, it is the pivot column. Since there is only one positive entry in the x
column, it is the pivot.
x y z s t u P
23 0 2 21 21 0 0 4

≥ ¥
21 2 0 1 21 0 0 2
1 0 0 0 1 1 0 0
217 0 0 1 25 0 2 14
n
Pivot column
264 CHAPTER 4 Linear Programming

We zero out the remaining terms in the x column using the indicated operations.
x y z s t u P
0 0 2 21 2 3 0 4 R1 1 3R3

≥ ¥
0 2 0 1 0 1 0 2 R2 1 R3
1 0 0 0 1 1 0 0
0 0 0 1 12 17 2 14 R4 1 17R3
x50 2y 5 2 2z 5 4 s50 t50 u50 2P 5 14
y51 z52 P57
Since all of the entries in the bottom row of the tableau are nonnegative, this so-
lution is the optimal solution. When x 5 0, y 5 1, and z 5 2, the objective func-
tion P 5 4x 1 3y 1 2z attains a maximum value of 7.
Figure 4.36 shows the graph of the constraints and the optimal solution. Each
constraint is a plane. The planes intersect at (0, 1, 2).

−2x + y + z = 3
− −y+z=1
−x

(0, 1, 2)

x
y

− +y+z=3
−x
FIGURE 4.36

EXAMPLE 3 Using Linear Programming to Make Investment Decisions

An investor plans to invest at most $3000 in the three publicly traded recreational
vehicle companies shown in Table 4.26. (Share prices are rounded to the nearest
dollar, and dividends per share are rounded to the nearest dime.)

TABLE 4.26
Company Share Price (dollars) Dividends/Share (dollars)
Harley-Davidson, Inc. (HDI) 63 0.40
Polaris Industries Inc. (PII) 51 0.90
Winnebago Industries Inc. (WGO) 33 0.20
Source: moneycentral.msn.com. (Accurate as of July 16, 2004.)
4.5 Solving General Linear Programming Problems with the Simplex Method 265

He wants to earn at least $50 in dividends while maximizing the number of shares
purchased. How many shares of each company should he purchase?

SOLUTION Let x be the number of shares of Harley-Davidson, y be the num-

ber of shares of Polaris Industries, and z be the number of shares of Winnebago
Industries. We must maximize P 5 x 1 y 1 z subject to
63x 1 51y 1 33z # 3000 The total amount invested is at most $3000
0.4x 1 0.9y 1 0.2z $ 50 Total dividends are at least $50
x $ 0, y $ 0, z $ 0
We add in the slack and surplus variables and rewrite the objective function.
63x 1 51y 1 33z 1 s 5 3000
0.4x 1 0.9y 1 0.2z 2 t 5 50
2x 2 y 2 z 1 P 5 0
We then write the initial tableau and star the row corresponding with a negative
variable value.
x y z s t P
63 51 33 1 0 0 3000
* £ 0.4 0.9 0.2 0 21 0 50 §
21 21 21 0 0 1 0
x50 y50 z50 s 5 3000 2t 5 50 P50
t 5 250
The largest entry in the starred row is the 0.9 in the y column; hence, the y col-
umn is the pivot column. We calculate the test ratios and locate the pivot.
x y z s t P
63 51 33 1 0 0 3000 3000>51 < 58.82
£ 0.4 0.9 0.2 0 21 0 50 § 50>0.9 < 55.56
21 21 21 0 0 1 0
We zero out the y column with the indicated operations and star the row corre-
sponding with a negative variable value.
266 CHAPTER 4 Linear Programming

x y z s t P
36.3 0 19.5 0.9 51 0 150 0.9R1 2 51R2
£ 0.4 0.9 0.2 0 21 0 50 §
20.5 0 20.7 0 21 0.9 50 0.9R3 1 R2
x50 0.9y 5 50 z50 0.9s 5 150 t50 0.9P 5 50
y < 55.56 s < 166.67 P < 55.56
Since all of the decision, slack, and surplus variables are nonnegative, we are in
the feasible region and may move to Stage 2.
The negative entry in the bottom row of a labeled column with the largest
magnitude is the 21 in the t column. Consequently, the t column is the pivot
column. Since 51 is the only positive entry in the t column, it is the pivot. We zero
out the t column with the indicated operations.
x y z s t P
36.3 0 19.5 0.9 51 0 150
£ 56.7 45.9 29.7 0.9 0 0 2700 § 51R2 1 R1
10.8 0 216.2 0.9 0 45.9 2700 51R3 1 R1
x50 45.9y 5 2700 z50 s50 51t 5 150 45.9P 5 2700
y < 58.82 t < 2.94 P < 58.82
The z column is the new pivot column, since it contains a negative entry in the
bottom row. We calculate the test ratios and identify the pivot.
x y z s t P
36.3 0 19.5 0.9 51 0 150 150>19.5 < 7.69
£ 56.7 45.9 29.7 0.9 0 0 2700 § 2700>29.7 < 90.91
10.8 0 216.2 0.9 0 45.9 2700
We zero out the z column with the indicated operations.
x y z s t P
36.3 0 19.5 0.9 51 0 150
£ 27.54 895.05 0 29.18 21514.7 0 48195 § 19.5R2 2 29.7R1
798.66 0 0 32.13 826.2 895.05 55080 19.5R3 1 16.2R1
x50 895.05y 5 48195 19.5z 5 150 s50 t50 895.05P 5 55080
y < 53.85 z < 7.69 P < 61.54
The investor should purchase 53.85 shares of Polaris Industries and 7.69 shares
of Winnebago Industries in order to maximize the number of shares while simul-
taneously earning dividends of at least $50.

Minimization Problems
Examples 1, 2, and 3 showed how to maximize an objective function of a linear
programming problem with mixed constraints. What if we want to minimize the
objective function? Fortunately, with one minor modification to the objective
function, the same procedure works. We begin by observing the relationship be-
tween a function f and the function 2 f 5 21 ? f shown in Figure 4.37.
 4.5 Solving General Linear Programming Problems with the Simplex Method 267

y Max of f

Max of − f

f a b
−f x

Min of f

Min of − f
FIGURE 4.37

Observe that the maximum of f and the minimum of 2f both occur at

x 5 a. Similarly, the minimum of f and the maximum of 2f both occur at
x 5 b. This relationship holds true for all functions. Consequently, if we want to
minimize an objective function P, we need only maximize the function 2P.

EXAMPLE 4 Solving a General Linear Programming Problem

Solve the general linear programming problem:
Minimize P 5 3x 1 5y 1 2z
x 1 2y 2 z # 0

Subject to µ
22x 1 4y 1 z $ 3
2x 1 y 1 z # 3
x $ 0, y $ 0, z $ 0

SOLUTION We first introduce a new function C 5 2P.

C 5 2P
5 2(3x 1 5y 1 2z)
5 23x 2 5y 2 2z
The solution that maximizes C will minimize P. Therefore, the linear program-
ming problem may be rewritten as
Maximize C 5 23x 2 5y 2 2z
x 1 2y 2 z # 0

Subject to µ
22x 1 4y 1 z $ 3
2x 1 y 1 z # 3
x $ 0, y $ 0, z $ 0
268 CHAPTER 4 Linear Programming

We add in the slack and surplus variables and rewrite the objective function. The
system of equations is
x 1 2y 2 z 1 s 5 0
22x 1 4y 1 z 2 t 5 3
2x 1 y 1 z 1 u 5 3
3x 1 5y 1 2z 1 C 5 0
and has the corresponding tableau
x y z s t u C
1 2 21 1 0 0 0 0

≥ ¥
22 4 1 0 21 0 0 3
21 1 1 0 0 1 0 3
3 5 2 0 0 0 1 0
x50 y50 z50 s50 2t 5 3 u53 C50
t 5 23
Since the surplus variable t is negative, we star the corresponding row. The y col-
umn is the pivot column because it contains the largest positive entry in the
starred row. Calculating the test ratios, we select the pivot.
x y z s t u C
1 2 21 1 0 0 0 0 0>2 5 0

≥ ¥
* 22 4 1 0 21 0 0 3 3>4 5 0.75
21 1 1 0 0 1 0 3 3>1 5 3
3 5 2 0 0 0 1 0
We zero out the remaining terms in the y column using the indicated operations
and star the row corresponding with the negative variable value.
x y z s t u C
1 2 21 1 0 0 0 0

≥ ¥
* 24 0 3 22 21 0 0 3 R2 2 2R1
23 0 3 21 0 2 0 6 2R3 2 R1
1 0 9 25 0 0 2 0 2R4 2 5R1
x50 2y 5 0 z50 s50 2t 5 3 2u 5 6 2C 5 0
y50 t 5 23 u53 C50
The largest positive value in the starred row is the 3 in the z column. Conse-
quently, the z column is the pivot column. We calculate the test ratios and select
the pivot.
x y z s t u C
1 2 21 1 0 0 0 0

≥ ¥
24 0 3 22 21 0 0 3 3>3 5 1
23 0 3 21 0 2 0 6 6>3 5 2
1 0 9 25 0 0 2 0
 4.5 Solving General Linear Programming Problems with the Simplex Method 269

We zero out the remaining terms in the z column with the indicated operations.
x y z s t u C
21 6 0 1 21 0 0 3 3R1 1 R2

≥ ¥
24 0 3 22 21 0 0 3
1 0 0 1 1 2 0 3 R3 2 R2
13 0 0 1 3 0 2 29 R4 2 3R2
x50 6y 5 3 3z 5 3 s50 t50 2u 5 3 2C 5 29
1 3 9
y5 z51 u5 C52
2 2 2
Since all decision, slack, and surplus variables are nonnegative, we are now in the
feasible region. (There is no restriction on the sign of the objective function. In
this case, C is negative.) We may now move to Stage 2 of the process. However,
since no entry in the bottom row of a labeled column is negative, this is the final
tableau. The objective function C 5 23x 2 5y 2 2z is maximized at A 0, 2, 1 B .
1

Consequently, the objective function P 5 3x 1 5y 1 2z is minimized at A 0, 2, 1 B .

1

9 9
The maximum value of C is 2 2. Since P 5 2C , the minimum value of P is 2.

So far, the general linear programming problems we have demonstrated have

included only linear inequalities as constraints. However, some linear program-
ming problems have linear equations as constraints. Any linear equation may be
written as a system of linear inequalities. For example, consider the linear equa-
tion 2x 1 3y 5 5. This equation is equivalent to the following system of linear
inequalities:
2x 1 3y # 5
2x 1 3y $ 5
When we are given a linear equation as a constraint, we will rewrite it as a sys-
tem of linear inequalities, as demonstrated in Example 5.

EXAMPLE 5 Using Linear Programming to Make Investment Decisions

An investor has $5000 to invest in three mutual funds: Bond, Index, and Growth.
The Bond fund is expected to earn 7 percent; the Index fund, 11 percent; and the
Growth fund, 9 percent. The investor wants to earn an annual return of at least 10
percent while minimizing the amount invested in the Index fund. How much
should she invest in each account?

SOLUTION Let x be the amount invested in the Bond fund, y be the amount in-
vested in the Index fund, and z be the amount invested in the Growth fund. We
must solve the following linear programming problem.
Minimize P5y
x 1 y 1 z 5 5000 $5000 is invested
Subject to • 0.07x 1 0.11y 1 0.09z $ 0.1(5000) The return is at least 10
x $ 0, y $ 0, z $ 0 of $5000
270 CHAPTER 4 Linear Programming

We rewrite the first constraint as two inequalities and simplify the second con-
straint. We then create a new objective function C 5 2P. Since P 5 y, C 5 2y.
The new linear programming problem is
Maximize C 5 2y
x 1 y 1 z # 5000

Subject to µ
x 1 y 1 z $ 5000
0.07x 1 0.11y 1 0.09z $ 5
x $ 0, y $ 0, z $ 0
We add the surplus and slack variables and rewrite the objective function.
x 1 y 1 z 1 s 5 5000
x 1 y 1 z 2 t 5 5000
0.07x 1 0.11y 1 0.09z 2 u 5 500
y1C50
We write the initial tableau and star the appropriate rows.
x y z s t u C
1 1 1 1 0 0 0 5000

≥ ¥
* 1 1 1 0 21 0 0 5000
* 0.07 0.11 0.09 0 0 21 0 500
0 1 0 0 0 0 1 0
x50 y50 z50 s 5 5000 2t 5 5000 2u 5 500 C50
t 5 25000 u 5 2500
Although we may use either starred row, we pick the second row of the tableau
because the entries are simpler. We may pick the x, y, or z column as the pivot col-
umn. We pick the x column. We calculate the test ratios and identify the pivot.
x y z s t u C
1 1 1 1 0 0 0 5000 5000>1 5 5000

≥ ¥
1 1 1 0 21 0 0 5000 5000>1 5 5000
0.07 0.11 0.09 0 0 21 0 500 500>0.07 < 7143
0 1 0 0 0 0 1 0
We may pick either the first or the second term in the x column to be the pivot.
We pick the first term and zero out the remaining entries with the indicated op-
erations. We then star the appropriate row.
x y z s t u C
1 1 1 1 0 0 0 5000

≥ ¥
0 0 0 21 21 0 0 0 R2 2 R1
* 0 0.04 0.02 20.07 0 21 0 150 R3 2 0.07R1
0 1 0 0 0 0 1 0
x 5 5000 y50 z50 s50 2t 5 0 2u 5 150 C50
t50 u 5 2150
 4.5 Solving General Linear Programming Problems with the Simplex Method 271

The y column is the pivot column, since the largest positive entry in the starred
row is the 0.04 in the y column. We calculate the test ratios and locate the pivot.
x y z s t u C
1 1 1 1 0 0 0 5000 5000>1 5 5000

≥ ¥
0 0 0 21 21 0 0 0
* 0 0.04 0.02 20.07 0 21 0 150 150>0.04 5 3750
0 1 0 0 0 0 1 0
We zero out the y column with the indicated operations.
x y z s t u C
0.04 0 0.02 0.11 0 1 0 50 0.04R1 2 R3

≥ ¥
0 0 0 21 21 0 0 0
0 0.04 0.02 20.07 0 21 0 150
0 0 20.02 0.07 0 1 0.04 2150 0.04R4 2 R3

0.04x 5 50 0.04y 5 150 z50 s50 2t 5 0 u50 0.04C 5 2150

x 5 1250 y 5 3750 t50 C 5 23750
Since all decision, slack, and surplus variables are nonnegative, we are in the fea-
sible region and may proceed to Stage 2.
Since the only negative entry in the bottom row of a labeled column is the
20.02 in the z column, the z column is the pivot column. We calculate the test ra-
tios and identify the pivot.
x y z s t u C
0.04 0 0.02 0.11 0 1 0 50 50>0.02 5 2500

≥ ¥
0 0 0 21 21 0 0 0
0 0.04 0.02 20.07 0 21 0 150 150>0.04 5 3750
0 0 20.02 0.07 0 1 0.04 2150
We zero out the remaining terms in the pivot column with the indicated operations.
x y z s t u C
0.04 0 0.02 0.11 0 1 0 50

≥ ¥
0 0 0 21 21 0 0 0
20.04 0.04 0 20.18 0 22 0 100 R3 2 R1
0.04 0 0 0.18 0 2 0.04 2100 R4 1 R1
x50 0.04y 5 100 0.02z 5 50 s50 2t 5 0 u50 0.04C 5 2100
y 5 2500 z 5 2500 t50 C 5 22500
Since all entries in the bottom row are positive, this is the final tableau. The ob-
jective function C reaches its maximum value when x 5 0, y 5 2500, and
z 5 2500. The maximum value is C 5 22500. The objective function P 5 2C
attains its minimum value of 2500 at the same point. In the context of the prob-
lem, a 10 percent return is earned when $0 is invested in the Bond fund, $2500 is
invested in the Index fund, and $2500 is invested in the Growth fund. This solu-
tion minimizes the amount of money invested in the Index fund.
272 CHAPTER 4 Linear Programming

4.5 Summary

In this section, you learned how to solve general linear programming problems.
You discovered that minimization problems may be solved by maximizing the
negative of the objective function.

4.5 Exercises

In Exercises 1–5, rewrite the general linear In Exercises 6–10, star the rows of the tableau that
programming problem as a system of equations with correspond with negative variable values.
slack and surplus variables and a rewritten objective
6. x y s t u P
function.
21 2 1 0 0 0 4
1. Maximize P 5 4x 1 2y 1 3z
≥ ¥
2 23 0 1 0 0 6
2x 1 y 1 z # 10 4 1 0 0 21 0 5

Subject to µ
x1z$3 26 22 0 0 0 1 0
x 2 2z $ 1 7. x y s t u P
x $ 0, y $ 0, z $ 0
9 4 21 0 0 0 10
2. Maximize P 5 2x 1 10y 1 z
≥ ¥
22 5 0 1 0 0 8
2x 1 3y # 2 1 1 0 0 21 0 12

Subject to µ
3x 1 z $ 4 5 4 0 0 0 1 0
5x 2 4z $ 1 8. x y s t u P
x $ 0, y $ 0, z $ 0
21 2 21 0 0 0 4
3. Maximize P 5 3x 2 y 1 9z
≥ ¥
2 0 22 1 0 0 6
212x 1 8y 1 2z # 15 4 0 1 0 25 0 5

Subject to µ
8x 1 5y 2 2z # 10 26 0 3 0 0 4 24
3x 2 2y 1 6z $ 5 9. x y s t u P
x $ 0, y $ 0, z $ 0
0 6 22 0 0 0 16
4. Maximize P5x2z
≥ ¥
5 1 0 1 0 0 1
2x 1 y 1 z # 12 0 3 0 4 24 0 6

Subject to µ
x 1 3y 2 5z $ 5 0 10 0 5 0 3 9
2x 1 y 2 2z # 1 10. x y s t u P
x $ 0, y $ 0, z $ 0
0 3 0 0 3 0 15
5. Maximize P 5 10x 1 9y 1 z
≥ ¥
0 0 1 1 2 0 1
4x 2 3y 2 2z # 12 2 0 0 22 24 0 4

Subject to µ
22x 1 4y 1 6z # 12 0 0 0 6 3 1 22
3x 2 2y 1 4z $ 12
x $ 0, y $ 0, z $ 0
 4.5 Solving General Linear Programming Problems with the Simplex Method 273

In Exercises 11–20, graph the feasible region of the 18. Minimize P 5 23x 1 8y
general linear programming problem. Then solve y# 6
the problem by using the methods demonstrated in the
Subject to µ
22x 1 y $ 0
section. For each tableau, indicate on the graph of
the feasible region the point that corresponds with the 2x 1 y $ 8
solution of the tableau. x$ 0, y $ 0
11. Maximize P 5 2x 2 3y 19. Minimize P 5 2x 1 y
2x 1 y # 10 x1y$ 1

Subject to µ
2x 2 3y $ 2
Subject to µ
x1y$3
3x 2 y $ 1 x 1 4y $ 4
x $ 0, y $ 0 x$ 0, y $ 0
12. Maximize P 5 4x 1 y 20. Maximize P 5 4x 1 2y
2x 1 y # 10 5x 1 2y $ 17

Subject to µ
x 2 2y # 1
Subject to µ
2x 1 3y $ 14
22x 1 y # 2 x1y$7
x $ 0, y $ 0 x $ 0, y $ 0
13. Maximize P 5 5x 2 8y In Exercises 21–35, solve the general linear
2x 1 y # 13 programming problem. If there is no solution, so state.

Subject to µ
9x 2 3y # 6 21. Maximize P 5 3x 1 2y 1 z
2x 1 2y $ 6
x1 y # 10
x $ 0, y $ 0
Subject to µ
x1 z#5
14. Maximize P 5 6x 1 3y y1 z$4
2x 1 y $ 0 x$ 0, y $ 0, z $ 0

Subject to µ
2x 1 y $ 9 22. Maximize P 5 5x 1 6y 1 2z
x 1 2y $ 6
x 1 2y # 10
x $ 0, y $ 0
Subject to µ
2x 1 z # 8
15. Maximize P 5 23x 1 4y y 1 2z $ 6
3x 1 12y $ 12 x $ 0, y $ 0, z $ 0

Subject to µ
2x 1 y # 8 23. Maximize P5x2y1z
x1y# 6
4x 1 3y 1 2z $ 12
x$ 0, y $ 0
Subject to µ
2x 1 2y 1 z # 8
16. Minimize P 5 3x 1 5y x 1 y 1 4z $ 4
25x 1 2y # 2 x $ 0, y $ 0, z $ 0

Subject to µ
2x 1 y # 4 24. Maximize P 5 6x 1 y
27x 1 4y $ 4
5x 1 2y 1 4z # 23
x $ 0, y $ 0
Subject to µ
2x 1 y 1 z $ 4
17. Minimize P 5 7x 1 4y x2y1z$4
3x 1 y $ 10 x $ 0, y $ 0, z $ 0

Subject to µ
x1y$6 25. Maximize P 5 2z
3x 1 2y # 17
x 1 2y 1 3z $ 12
x $ 0, y $ 0
Subject to µ
3x 1 y 1 2z # 12
2x 1 3y 1 z # 12
x $ 0, y $ 0, z $ 0
274 CHAPTER 4 Linear Programming

26. Maximize P 5 2x 1 y 1 z 34. Minimize P 5 x 1 2y 1 3z

2x 1 y 1 z # 5 x1y1z$6

Subject to µ Subject to µ
x2y1z#7 x2y1z$2
2x 2 y 1 z $ 3 x1z#4
x $ 0, y $ 0, z $ 0 x $ 0, y $ 0, z $ 0
27. Maximize P 5 6x 2 2y 1 6z 35. Maximize P 5 4x 1 3y 1 2z
2x 2 y 1 z # 1 x1y1z$8

Subject to µ Subject to µ
x1y1z#7 22x 1 z $ 1
2x 2 3y 1 z $ 12 3y 1 5z # 27
x $ 0, y $ 0, z $ 0 x $ 0, y $ 0, z $ 0
28. Minimize P 5 4y 1 9z
In Exercises 36–45, use the techniques demonstrated
22x 2 3y 1 z $ 0 in the section to set up and solve each problem.
Subject to µ
4x 1 y 1 z $ 10
36. Commodity Prices Today’s Market
24x 1 2y 1 z $ 3 Prices (www.todaymarket.com) is a daily
x $ 0, y $ 0, z $ 0 fruit and vegetable wholesale market price service.
29. Minimize P 5 2x 2 y 1 4z Produce retailers who subscribe to the service can
use wholesale prices to aid them in setting retail
22x 1 3y 1 z $ 4 prices for the fruits and vegetables they sell.
Subject to µ
2x 1 4y 1 z # 4 A 25-pound carton of peaches holds 60
24x 1 y 1 z # 12 medium peaches or 70 small peaches. In August
x $ 0, y $ 0, z $ 0 2002, the wholesale price for local peaches in Los
Angeles was $9.00 per carton for medium peaches
30. Minimize P 5 5x 1 4y 1 3z and $10.00 per carton for small peaches. (Source:
2x 2 5y 1 z $ 4 Today’s Market Prices.) A fruit vendor has budgeted
up to $100 to spend on peaches. He estimates that
Subject to µ
2x 1 4y 1 z # 13
z$3 weekly demand for peaches is no more than 660
peaches. He wants to buy at least four boxes of
x $ 0, y $ 0, z $ 0
each size of peach. Subject to these constraints,
31. Minimize P 5 9x 2 3y 2 3z how many boxes of each size of peach should he
23x 2 4y 1 z # 4 buy in order to maximize the number of peaches
available for sale?
Subject to µ
2x 2 5y 1 z # 14
3x 2 2y 1 z $ 8 37. Battery Sales AAA Alkaline Discount
x $ 0, y $ 0, z $ 0 Batteries sells low-cost batteries to
consumers. In August 2002, the firm offered AA
32. Minimize P 5 x 2 y 1 3z batteries at the following prices: 50-pack for
2x 1 2y 2 z # 4 $10.00, 100-pack for $18.00, and 600-pack for
$96.00. (Source: www.aaaalkalinediscountbatteries
Subject to µ
5x 2 4y 1 z $ 16
.com.)
x 2 4y 1 z $ 5
An electronics store owner wants to buy at least
x $ 0, y $ 0, z $ 0 900 batteries and spend at most $150. She expects
33. Minimize P 5 6x 2 2y 1 4z that she’ll be able to resell all of the batteries she
orders for $1.50 each. How many packs (50-
8x 1 y 2 z # 8
packs, 100-packs, or 600-packs) should she order
Subject to µ
x2y1z$1 if she wants to maximize her revenue? What is her
x 2 y 1 z # 10 maximum revenue? (Hint: At most one 600-pack
x $ 0, y $ 0, z $ 0 may be purchased without exceeding the $150
limit.)
 4.5 Solving General Linear Programming Problems with the Simplex Method 275

38. Resource Allocation: Food An ice bean and the manufacturer. A 1/2-cup serving of
cream parlor wants to make three Fred Meyer Pinto Beans contains 7 grams of fiber,
different types of ice cream: vanilla, strawberry, 1 gram of sugar, and 6 grams of protein. A 1/2-cup
and peach-cherry. The parlor has 120 cups of serving of Fred Meyer Kidney Beans contains
cream, 48 eggs, and 32 cups of sugar on hand. 11 grams of fiber, 1 gram of sugar, and 8 grams of
The vanilla ice cream recipe calls for 4 cups of protein. A 1/2-cup serving of Trader Joe’s Cuban
cream, 1 egg, and 0.75 cup of sugar. The Style Black Beans contains 2 grams of fiber,
strawberry ice cream recipe calls for 2 cups of 1 gram of sugar, and 6 grams of protein. (Source:
cream, 2 eggs, and 0.75 cup of sugar. The peach- Package labeling.) A bean dish is to be made using
cherry ice cream recipe calls for 4 cups of cream, the three bean varieties. The dish can contain at
1 egg, and 1.25 cups of sugar. Each recipe yields most 3 cups of beans but must include at least
1.5 quarts of ice cream. The parlor needs at least 39 grams of fiber and 42 grams of protein. How
18 quarts of vanilla and at least 6 quarts of each much of each type of bean should be included in
of the other varieties. The parlor wants to the dish in order to minimize the amount of kidney
maximize the amount of cream produced. How beans used? (Fractions of cups may be used.)
many batches of each variety of ice cream does
42. Nutrition The nutritional content of
the parlor need to produce?
canned vegetables varies based on the
39. Resource Allocation: Food A type of vegetable and the manufacturer. A 1/2-cup
delicatessen makes three types of serving of S&W Cut Blue Lake Green Beans
pudding: rice, tapioca, and vanilla. The deli has contains 2 grams of fiber, 2 grams of sugar, and
108 cups of milk, 150 cups of sugar, and 84 eggs 20 calories. A 1/2-cup serving of Safeway Golden
on hand. The rice pudding recipe requires 12 cups Sweet Whole Kernel Corn contains 2 grams of
of milk, 1.5 cups of sugar, and 9 eggs and yields fiber, 6 grams of sugar, and 80 calories. A 1/2-cup
24 servings. The tapioca pudding recipe requires serving of Pot O’Gold Sliced Carrots contains
12 cups of milk, 1.5 cups of sugar, and 9 eggs and 2 grams of fiber, 4 grams of sugar, and 30
yields 18 servings. The vanilla pudding recipe calories. (Source: Package labeling.) A dish is to be
requires 6 cups of milk, 1.5 cups of sugar, and made using the three types of vegetables. The dish
6 eggs and yields 12 servings. The deli requires can contain at most 5 cups of vegetables but must
that the sum of the number of batches of tapioca include at least 18 grams of fiber and at least
pudding and the number of batches of rice 32 grams of sugar (for flavor). How much of each
pudding be at least two. How many batches of type of vegetable should be included in the dish in
each recipe must the deli produce in order to order to minimize the amount of calories in the
maximize the number of servings? dish?
40. Advertising A large company advertises through 43. Resource Allocation: Schools A city has two
magazines, radio, and television. For every elementary schools. The first school has a
$10,000 spent on magazine advertising, the maximum enrollment of 500 students, and the
company estimates that it reaches 150,000 people. second school has a maximum enrollment of
For every $10,000 spent on radio advertising, the 420 students. The city is divided into two regions:
company estimates that it reaches 250,000 people. North and South. There are at least 400 students
For every $10,000 spent on television advertising, in the North region and at least 430 students in
the company estimates that it reaches 300,000 the South region. The annual transportation cost
people. The company has at most $2.5 million to varies by region as shown in the table.
spend on advertising. It requires that at least twice
Transportation Costs
as much be spent on radio as on television and
that the amount spent on magazines be at least School 1 School 2
$150,000 more than the amount spent on radio. (cost per (cost per
How much should the company spend on each student) student)
type of advertising in order to maximize the North 120 180
number of people reached?
South 100 150
41. Nutrition The nutritional content of
canned beans varies based on the type of
276 CHAPTER 4 Linear Programming

Based on these constraints, what is the Subject to these constraints, under what conditions
minimum possible transportation cost, and under are the company’s delivery costs minimized?
what conditions does it occur?
Exercises 46–50 are intended to challenge your
44. Resource Allocation: Schools A city has two
understanding of general linear programming problems.
high schools. The first school has a maximum
enrollment of 900 students, and the second school 46. Maximize P5x1y1z1w
has a maximum enrollment of 550 students. The
x1y1z#9
city is divided into two regions: Inner City and y 1 z 1 w # 8
Suburbs. There are at least 500 students in the 
Inner City region and at least 800 students in the Subject to  x 1 z 1 w $ 7
x 1 y 1 w $ 6
Suburbs region. The annual transportation cost 
varies by region as shown in the table.  x $ 0, y $ 0, z $ 0, w $ 0

Transportation Costs 47. Minimize P5x1y1z1w

 x1y1z#6
School 1 School 2 
(cost per (cost per y 1 z 1 w # 8
student) student) Subject to  x 1 z 1 w $ 9
x 1 y 1 w $ 7
Inner City 220 260 
 x $ 0, y $ 0, z $ 0, w $ 0
Suburbs 200 280
48. Create a two-variable linear programming
problem with three or more mixed constraints
Based on these constraints, what is the
with the property that the feasible region consists
minimum possible transportation cost, and under
of exactly one point. Provide evidence that the
what conditions does it occur?
feasible region contains exactly one point.
45. Transportation Costs A furniture
49. Create a three-variable linear programming
company has warehouses in Phoenix,
problem with three or more mixed constraints
Arizona, and Las Vegas, Nevada. The company has
with the property that the feasible region is empty.
customers in Kingman, Arizona, and Flagstaff,
Provide evidence that the feasible region is empty.
Arizona. Its Phoenix warehouse has 500 desks in
stock, and its Las Vegas warehouse has 200 desks 50. Explain what happens graphically in Stage 1 of
in stock. Its Kingman customer needs at least 250 the general linear programming problem-solving
desks, and its Flagstaff customer needs at least 400 process.
desks. Based on a rate of $0.01 per mile per item,
the cost of delivery per item is shown in the table.

Phoenix Las Vegas

Kingman $1.85 $1.05
Flagstaff $1.45 $2.51

Chapter 4 Review Exercises

Section 4.1 In Exercises 1–4, graph the solution

region of the linear inequality. Then use the graph to
determine if point P is a solution.
1. 4x 2 2y # 6; P 5 (2, 9) 3. 9x 2 8y # 12; P 5 (8, 9)
2. 3x 1 5y # 0; P 5 (1, 7) 4. 7x 1 6y # 42; P 5 (3, 4)
 Chapter 4 Review Exercises 277

In Exercises 5–8, graph the solution region of the 13. Maximize z 5 5x 2 y

system of linear equations. If there is no solution, 6x 1 2y # 16
explain why.
Subject to µ
23x 2 y # 10
5. 24x 1 3y $ 2 6. 210x 1 y $ 0 x$ 0
23x 1 2y $ 1 2x 1 y # 4 y$ 0
7. 2x 1 4y # 8 8. x1y#8
For Exercise 14, identify the objective function and
6x 2 2y # 26 2x 1 y # 0 constraints of the linear programming problem. Then
4x 2 2y # 8 solve the problem and interpret the real-world
meaning of your results.
In Exercise 9, set up and graphically solve the system 14. Family Food Storage A family wants
of linear inequalities. to purchase at least 75 pounds of beans.
A #10 can of pinto beans weighs 5.0 pounds and
9. Wages A salaried employee earns $800 per costs $2.75. A #10 can of white beans weighs
week managing a copy center. She is required to 5.3 pounds and costs $2.88. (Source: Kent
work a minimum of 40 hours but no more than Washington Cannery.) The family wants to buy
50 hours weekly. As a side business, she earns at least 25 pounds of pinto beans and at least
$30 per hour designing brochures for local 53 pounds of white beans. It has budgeted $50
business clients. In order to maintain her standard and, because of limited storage space, wants to
of living, she must earn $1000 per week. In order minimize the number of cans purchased. How
to maintain her quality of life, she limits her many cans of each type of beans should the
workload to 50 hours per week. Given that she has family purchase?
no control over the number of hours she will have
to work managing the copy center, will she be Section 4.3 In Exercises 15–16, determine if the
able to consistently meet her workload and problem is a standard maximization problem. If it isn’t,
income goals? Explain. explain why.
Section 4.2 In Exercises 10–13, find the optimal 15. Maximize P 5 22x 1 8y
solution to the linear programming problem, if it
Subject to e
2x 1 y # 1
exists. If a solution does not exist, explain why. 2x 1 2y # 220
10. Minimize z 5 5x 2 7y 16. Maximize P 5 29x 1 8y
4x 1 y $ 4 42x 1 19y # 10
Subject to µ
2x 1 y $ 1
Subject to µ
211x 1 19y # 21
x$0 x$ 0
y$0 y$ 0
11. Minimize z 5 9x 1 4y
In Exercises 17–18, solve the standard maximization
6x 1 y $ 16
problems by using the simplex method. Check your
Subject to µ
22x 1 y $ 0 answer by graphing the feasible region and
x$0 calculating the value of the objective function at each
y$0 of the corner points.
12. Maximize z 5 6x 1 10y 17. Maximize P 5 22x 1 10y
6x 1 y # 16 2x 1 y # 10

Subject to µ Subject to µ
22x 1 y # 0 2x 1 y # 1
x$ 0 x$ 0
y$ 0 y$ 0
278 CHAPTER 4 Linear Programming

18. Maximize P 5 5x 2 2y Section 4.4 In Exercises 23–25, determine if the

2x 1 y # 6 problem is a standard minimization problem. If it isn’t,
explain why.
Subject to µ
2x 1 2y # 8
x$ 0 23. Minimize P 5 11x 1 8y 1 2z
y$ 0  2x 1 y 1 z $ 10

2x 1 y 2 z $ 1
In Exercises 19–20, solve the standard maximization Subject to  x$0
problems by using the simplex method.  y$0

19. Maximize P 5 6x 1 4y 1 5z  z$0
2x 1 3y 1 2z # 120 24. Maximize P 5 22x 2 5y 1 7z

 x1y1z # 60  x 1 2y 1 z $ 11
Subject to  x $0 
  211x 1 y # 221
 y $0 Subject to  x$0
 z $0 

y$0
20. Maximize P 5 4x 2 y 1 z  z$0
2x 1 3y 1 3z # 210 25. Maximize P 5 4x 2 2y

 x1y1z # 100 6x 1 7y $ 13
Subject to  x $0
Subject to µ
 28x 2 4y $ 212
 y $0
x$ 0
 z $0
y$ 0
In Exercises 21–22, set up and solve the standard
maximization problem using the simplex method. In Exercises 26–28, find the transpose of the given
matrix.
21. Battery Sales AAA Alkaline Discount
Batteries sells low-cost batteries to 7 22 0 4
consumers. In August 2002, the company offered 26. A 5 £ 5 3 7 6§
AAA batteries at the following prices: 100-pack 21 1 2 1
for $18.00, 600-pack for $96.00, and 1200-pack
12 10 8
27. A 5 £ 14 10 §
for $180.00.
(Source: www.aaaalkalinediscountbatteries.com.)
12
An electronics store owner wants to buy at 16 14 12
most 1800 batteries and spend at most $300. She 1 0 1 0 2
28. A 5 £ 0 9§
expects that she’ll be able to resell all of the
1 0 1
batteries she orders for $1.00 each. How many
packs (100-packs, 600-packs, or 1200-packs) 1 1 1 1 11
should she order if she wants to maximize her
revenue? What is her maximum revenue? In Exercises 29–32, do the following:
(i) Write the dual problem for the given standard
22. Battery Sales Repeat Exercise 21 minimization problem.
except maximize profit. Assume that her
only cost is the cost of the batteries. What is her (ii) Solve the dual problem using the simplex method.
maximum profit? (iii) Use the final simplex tableau of the dual problem
to solve the standard minimization problem.
 Chapter 4 Review Exercises 279

(iv) Check your answer by graphing the feasible Nutro Max Natural Dog Food, 27 percent
region of the standard minimization problem and protein, 16 percent fat, $12.99 per 17.5-pound
calculating the value of the objective function at bag
each of the corner points. PETsMART Premier Oven Baked Lamb
Recipe, 25 percent protein, 14 percent fat,
29. Minimize P 5 4x 1 2y
$22.99 per 30-pound bag (Source:
4x 1 5y $ 20 www.petsmart.com.)
Subject to • 2x 1 3y $ 11 A dog breeder wants to make at least 300 pounds
x $ 0, y $ 0 of a mix containing at least 24 percent protein.
How many bags of each dog food variety should
30. Minimize P 5 x 1 6y the breeder buy in order to minimize fat content?
2x 1 9y $ 19
Subject to • 2x 1 10y $ 5 Section 4.5 In Exercises 35–36, rewrite the
x $ 0, y $ 0 general linear programming problem as a system of
equations with slack and surplus variables and a
31. Minimize P 5 2x 1 3y rewritten objective function.
5x 1 2y $ 26 35. Maximize P 5 2x 1 3y 1 z
Subject to • 10x 2 5y $ 25
2x 1 4y 1 z # 10
x $ 0, y $ 0
Subject to µ
x 1 2z $ 3
32. Minimize P 5 3x 1 5y 3x 2 2z $ 1
9x 1 4y $ 36 x $ 0, y $ 0, z $ 0
Subject to • 3x 1 2y $ 18 36. Maximize P 5 x 2 10y 1 6z
x $ 0, y $ 0
22x 1 2y 1 3z # 2

Subject to µ
In Exercises 33–34, set up and solve the standard 5x 1 2y 1 3z $ 4
minimization problem. 7x 2 3y 1 2z $ 1
x $ 0, y $ 0, z $ 0
33. Pet Nutrition: Food Cost
PETsMART.com sold the following
In Exercises 37–38, solve the general linear
varieties of dog food in June 2003.
programming problem. If there is no solution, so state.
Authority Chicken Adult Formula, 32 percent
protein, 3 percent fiber, $19.99 per 33-pound 37. Maximize P 5 4x 1 y 1 10z
bag x 1 3y 1 2z # 5
Bil-Jac Select Dog Food, 27 percent protein,
Subject to µ
2x 1 y 1 2x $ 4
4 percent fiber, $18.99 per 18-pound bag
2y 1 3z # 6
Iams Minichuncks, 26 percent protein, 4 percent
fiber, $8.99 per 8-pound bag (Source: x $ 0, y $ 0, z $ 0
www.petsmart.com.) 38. Minimize P 5 4x 1 y 1 10z
A dog kennel wants to make at least 330 pounds
x 1 3y 1 2z # 5
of a mix containing at least 30 percent protein and
Subject to µ
at least 3 percent fiber. How many bags of each 2x 1 y 1 2x $ 4
A K E I T
dog food variety should the kennel buy in order to 2y 1 3z # 6
minimize cost? (To make computations easier, x $ 0, y $ 0, z $ 0
round the price of each bag to the nearest dollar.) R E A L : P R O J E C T 4
34. Pet Nutrition: Fat Content
PETsMART.com sold the following
varieties of dog food in June 2003.
Nature’s Recipe Venison Meal & Rice Canine,
20 percent protein, 10 percent fat, $21.99 per
20-pound bag
 P R O J E C T 4

What to do
1. Find the wholesale and retail prices of three items of personal interest.
2. Estimate the number of cubic feet each item (or box of items) occupies.
3. Find a place in your home or workplace where you could store the items.
Calculate the number of cubic feet in the storage area.
4. Assume that you may spend up to 100 times the retail price of the most ex-
pensive item in purchasing a combination of the three items.
5. Assume that you will be able to sell at retail price all of the items you order
at wholesale. Determine how many of the items you should order if you
want to maximize your profit subject to the spending and storage area
constraints.

Where to Look for Prices

Wholesale Prices
Fruits: www.todaymarket.com
Vegetables: www.todaymarket.com
Flowers: www.flowersales.com
Computer parts: www.tcwo.com
Batteries: www.aaaalkalinediscountbatteries.com
Various items: Costco, Sam’s Club, etc.
Search the Internet at www.yahoo.com and type in the key word
“wholesale”.

Retail Prices
Local grocery stores, clothing stores, etc.
Advertisements from newspapers
Search the Internet and type in a key word for your product of interest.

280