UMAP Module 780

Modules in
Undergraduate Applications of
Mathematics
and Its
Sequences and
Applications Limits in Calculus
Published in Yves Nievergelt
cooperation with

The Society for

Industrial and
Applied Mathematics, 2 x 10–12

1.5 x 10–12
The Mathematical
1 x 10–12
Association of America,
5 x 10–13
U0 X W0
The National Council 0.00001
2 x 10–6 4 x 10–6 6 x 10–6 8 x 10–6
of Teachers of –5 x 10–13

Mathematics, –1 x 10–12
F
–1.5 x 10–12
The American
Mathematical
Association of
Two-Year Colleges,

The Institute for

Operations Research
and the Management
Sciences, and

The American
Statistical Association.

Applications of Precalculus,
Calculus, and Analysis
to Chemistry and Finance

COMAP, Inc., Suite 210, 57 Bedford Street, Lexington, MA 02420 (781) 862–7878
110 Tools for Teaching 1999

INTERMODULAR DESCRIPTION SHEET: UMAP Unit 780

TITLE: Applications of Sequences and Limits in Calculus

AUTHOR: Yves Nievergelt

Dept. of Mathematics, MS 32
Eastern Washington University
526 5th Street
Cheney, WA 99004–2431
ynievergelt@mail.ewu.edu

MATHEMATICAL FIELD: Precalculus, calculus, and analysis

APPLICATION FIELD: Chemistry, finance

TARGET AUDIENCE: Students in precalculus, calculus, or analysis.

ABSTRACT: This Module introduces applications of the mathemat-

ical concepts of “sequence” and “limit” by way of the
method of bisection. In precalculus, bisection provides
an example of an algorithm, together with approxima-
tions of solutions, estimates of accuracy, sequences, and
mathematical induction. In beginning calculus, bisec-
tion provides a motivation for studying in detail the
convergence of sequences to limits. In advanced calcu-
lus and real analysis, bisection provides a motivation
for precise definitions of the real numbers, for instance,
as Dedekind cuts, or as equivalence classes of Cauchy
sequences of rational numbers.

PREREQUISITES: A working knowledge of rational arithmetic, the con-

cepts of mathematical functions, and mathematical in-
duction.

RELATED UNITS: Unit 263: Horner’s Scheme and Related Algorithms, by

Werner C. Rheinboldt. Tools for Teaching 1980, 497–
520. Boston, MA: Birkhäuser, 1981. Reprinted in The
UMAP Library. CD-ROM. Lexington, MA: COMAP,
1996.
Unit 264: Algorithms for Finding Zeros of Functions, by
Werner C. Rheinboldt. Reprinted in The UMAP Jour-
nal 2 (1): 43–72. Reprinted in Tools for Teaching 1981,
471–500. Boston, MA: Birkhäuser, 1982. Reprinted
in The UMAP Library. CD-ROM. Lexington, MA:
COMAP, 1996.

Tools for Teaching 1999, 109–134. Reprinted from The UMAP Journal 20 (1) (1999) 139–164.

Copyright
c 1999, 2000 by COMAP, Inc. All rights reserved.

Permission to make digital or hard copies of part or all of this work for personal or classroom use
is granted without fee provided that copies are not made or distributed for profit or commercial
advantage and that copies bear this notice. Abstracting with credit is permitted, but copyrights
for components of this work owned by others than COMAP must be honored. To copy otherwise,
to republish, to post on servers, or to redistribute to lists requires prior permission from COMAP.

COMAP, Inc., Suite 210, 57 Bedford Street, Lexington, MA 02420

(800) 77-COMAP = (800) 772-6627, or (781) 862-7878; http://www.comap.com
 Applications of Sequences and Limits 111

Applications of Sequences and

Limits in Calculus
Yves Nievergelt
Dept. of Mathematics, MS 32
Eastern Washington University
526 5th Street
Cheney, WA 99004–2431
ynievergelt@mail.ewu.edu

Table of Contents
1. INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

2. REASONS FOR SOLVING EQUATIONS . . . . . . . . . . . . . . . . . . . 2

3. SOLVING EQUATIONS THROUGH BISECTION . . . . . . . . . . . . . . . 3

4. FURTHER ISSUES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
5. SOLVING EQUATIONS THROUGH Regula Falsi . . . . . . . . . . . . . 12
6. CONCLUSIONS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

7. SOLUTIONS TO THE ODD-NUMBERED EXERCISES . . . . . . . . . . . . 15

REFERENCES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

ABOUT THE AUTHOR . . . . . . . . . . . . . . . . . . . . . . . . . . 22

112 Tools for Teaching 1999

MODULES AND MONOGRAPHS IN UNDERGRADUATE

MATHEMATICS AND ITS APPLICATIONS (UMAP) PROJECT

The goal of UMAP is to develop, through a community of users and devel-

opers, a system of instructional modules in undergraduate mathematics and
its applications, to be used to supplement existing courses and from which
complete courses may eventually be built.
The Project was guided by a National Advisory Board of mathematicians,
scientists, and educators. UMAP was funded by a grant from the National
Science Foundation and now is supported by the Consortium for Mathemat-
ics and Its Applications (COMAP), Inc., a nonprofit corporation engaged in
research and development in mathematics education.

Paul J. Campbell Editor

Solomon Garfunkel Executive Director, COMAP
 Applications of Sequences and Limits 113

1. Introduction
The mathematical concepts of “sequence” and “limit” are introduced in
calculus texts of many genres as a technicality necessary for the presentation
of such mathematical concepts as “continuity,” “derivative,” “integral,” and
“series” [Etgen 1999; Ostebee and Zorn 1997, 1998; Spivak 1980; Stewart 1995;
Strang 1991]. Consequently, if the corresponding calculus course fails to pro-
vide extensive and intensive practice with limits, then students can earn high
grades in calculus without having a clue about the nature and significance of
the concept of a limit. As evidence of students’ weak command of the concept
of limit, an informal poll of students in second term calculus and linear algebra
showed that over 93% of students do not solve the following exercise correctly:
1. Indicate whether the following statement is true or whether it is false:
“For each function f defined in an open interval containing a number T ,
if the value of f (t) gets closer and closer to a number L
as the value of t gets closer and closer to the number T ,
then limt→T f (t) = L. ”

Students’ command of limits appears to strengthen with additional prac-

tice and demonstrations that the mathematical concept of limit is not a mere
technicality. Instead, the mathematical concept of limit is a major mathematical and
philosophical accomplishment. Indeed, since Archimedes’ algorithm to compute
π and the Babylonian and Chinese algorithms to extract square roots and cube
roots [Neugebauer 1969, van der Waerden 1983], it took humanity over two
millenia before Karl Weierstrass was able to state a precise definition (inde-
pendent of any infinitesimal deus ex machina ) of the mathematical concept of
limit [Edwards 1979, 333]. The mathematical concept of limit seems to be the
only concept from calculus that has such a major historical and philosophical
status, and yet its detailed study is omitted in many courses. Mathematicians
appear thus to form the only profession that has intentionally and successfully
removed its major achievement from all of its introductory freshman courses,
as evidenced by [Mac Lane 1997].
Of course, despite all the merits attributed to the use of history in pedagogy,
dropping such famous names as Archimedes, Weierstrass, and the Chinese
Han dynasty in such pompous sentences as the preceding ones will also fail to
strengthen students’ command of limits!
Therefore, these notes introduce the mathematical concepts of sequence and
limit through a practical use that already contains the germs of the abstract idea:
the method of bisection to solve equations with one unknown. These notes can
be used before derivatives and integrals in the first term of calculus; or before
series in subsequent terms of calculus, depending on the syllabus; or as a fast
motivation or review for students in beginning real analysis.

1
114 Tools for Teaching 1999

2. Reasons for Solving Equations

The need to solve equations arises because some numbers are defined im-
plicitly (“abstractly”) as the solutions of equations. For some equations, there
exists no formula to calculate the solutions, or the existing formula merely gives
a name to the solution but does not provide a method to calculate it. In either
case, the computation of the solutions can involve iterative methods, as is the
case for the bisection method presented here. In general, iterative methods
produce infinite sequences of numbers, which can approximate solutions of
equations to any specified accuracy. Moreover, well-designed iterative meth-
ods can prove more effective than exact but ill-suited formulae, especially with
the finite arithmetic of digital computers.

Example 1. The equation 3x = 2 defines the number x = 2/3 . Yet the

formula 2/3 merely gives a name to the solution x of 3x = 2 , without
providing its decimal expansion. Nevertheless, the iterative method of
long division produces an infinite sequence of numbers 0.6 , 0.66 , 0.666 ,
. . . , which yields the decimal expansion 2/3 = 0.666 666 . . . . (However,
the formula 2/3 does contain the expansion of x in base 3 : x = 2/3 =
2 ∗ 3−1 = 0.2three .)

Example 2. Consider the problem of computing the length d of the

diagonal of a square with all sides of length 1 (see Figure 1).

d
1

Figure 1. d2 = 12 + 12 .

The only information about d provided by geometry comes from

the Pythagorean theorem, which states that d2 = 12 + 12 but does not
produce digits for the length d of the diagonal. The bisection method,
among many other iterative methods, will produce the decimal expansion
of d .

Exercises
2. Write √down an equation involving only integer arithmetic (polynomial)
with 3 as one of its solutions.

2
 Applications of Sequences and Limits 115

3. Write √
down an equation involving only integer arithmetic (polynomial)
3
with 2 as one of its solutions.

3. Solving Equations through Bisection

The method of bisection is perhaps the simplest, most reliable, and most
general algorithm to solve equations. Though the method of bisection does
not instantly produce the solution exactly, it refines the accuracy at each step
and can achieve any specified accuracy. Moreover, the method of bisection
is a simple prototype for iterative algorithms, which form the basis of most of
scientific computing.
To solve an equation of the form f (x) = 0, the method of bisection requires
only two initial estimates u0 ≤ w0 with
f (u0 ) ≤ 0 ≤ f (w0 ),
as in Figure 2, from which successive bisections will produce a solution in the
closed (including both endpoints) interval [u0 , w0 ] .

fHw 0 L > 0

u0

w 0

fHu 0 L < 0

Figure 2. Bisection yields a solution in [u0 , w0 ] if f (u0 ) ≤ 0 ≤ f (w0 ) .

To narrow the interval, the method tests the value of f at the midpoint
u0 + w0
v0 := ,
2
where two cases may arise. (The symbol := defines the yet undefined left-hand
side in terms of the already defined right-hand side.)
• (+) If f (v0 ) ≥ 0 , then
f (u0 ) ≤ 0 ≤ f (v0 ), f (w0 ),

3
116 Tools for Teaching 1999

and the method narrows the interval to the left-hand half [u1 , w1 ] := [u0 , v0 ] .
• (−) If f (v0 ) < 0 , then

f (u0 ), f (v0 ) ≤ 0 ≤ f (w0 ),

and the method narrows the interval to the right-hand half [u1 , w1 ] :=
[v0 , w0 ] .

In either case, f (u1 ) ≤ 0 ≤ f (w1 ) with a narrower interval [u1 , w1 ] .

To solve a problem through the method of bisection as just described, a
restatement of the problem into an equation of the form f (x) = 0 may first be
necessary.

Example 3. To compute the length d of the diagonal of a square with

sides of length 1 , first rewrite d as the positive solution of the equation
x2 = 2 , and then rewrite the equation as x2 −2 = 0 so that f (x) = x2 −2 .
Then select two initial estimates — perhaps after some trial and error—
for instance, as in Figure 3:

u0 := 1, f (u0 ) = f (1) = 12 − 2 = −1 < 0,

w0 := 2, f (w0 ) = f (2) = 22 − 2 = 2 > 0.

fHxL = x 2 - 2

2 fHw 0 L > 0

u0 w 0
1 2

-1 fHu 0 L<0

Figure 3. Initial estimates [u0 , w0 ] = [1, 2] to solve x2 − 2 = 0 .

4
 Applications of Sequences and Limits 117

Step 1. The method of bisection then tests f at the midpoint v0 :

u0 + w0 1+2
v0 := = = 3/2,
2 2
f (v0 ) = f ( 3/2) = ( 3/2)2 − 2 = 9/4 − 2 = 1/4 > 0.

Consequently,
f (u0 ) < 0 < f (v0 ), f (w0 ),
so the shorter interval [u0 , v0 ] = [1, 3/2] may replace the longer initial
interval [u0 , w0 ] = [1, 2] ; set

u1 := u0 = 1, w1 := v0 = 3/2.

Thus, the first step yields the result [u1 , w1 ] = [1, 3/2] .
Step 2. The method of bisection then proceeds by testing f at the new
midpoint v1 :

u1 + w1 1 + 3/2
v1 := = = 5/4,
2 2
f (v1 ) = f ( 5/4) = ( 5/4)2 − 2 = 25/16 − 2 = − 7/16 < 0.

Consequently,
f (u1 ), f (v1 ) < 0 < f (w1 ),
so the shorter interval [v1 , w1 ] = [ 5/4, 3/2] may replace the longer initial
interval [u1 , w1 ] = [1, 3/2] ; set

u2 := v1 = 5/4, w2 := w1 = 3/2.

Thus, the second step yields the result [u2 , w2 ] = [ 5/4, 3/2] .
Similarly, the next few steps produce the results of Table 1.

Table 1.
Steps in the calculation, by means of bisection, of the length of the diagonal of a unit square.

n un vn wn

0 1 1.5 2
1 1 1.25 1.5
2 1.25 1.375 1.5
3 1.375 1.4375 1.5
4 1.375 1.40625 1.4375
5 1.40625 1.4140625 1.4375
6 1.4140625 1.41796875 1.421875
7 1.4140625 1.416015625 1.41796875
8 1.4140625 1.4150390625 1.416015625
9 1.4140625 1.41455078125 1.4150390625
10 1.4140625 1.414306640625 1.41455078125
.. .. .. ..
. . . .

5
118 Tools for Teaching 1999

The last line displayed shows that 1.4140 < d < 1.4146 ; thus, the bi-
section method has produced the first four decimal digits: d = 1.414 . . . .

In practice, the question arises how to determine in advance how many

steps, how many arithmetic operations, or how much computing time will
deliver a specified accuracy. The following considerations explain how to de-
termine in advance how many steps of the bisection method will deliver a
specified accuracy. In general, each single step of the method of bisection starts
from two initial estimates un ≤ wn with

f (un ) ≤ 0 ≤ f (wn ).

To narrow the interval, the method tests the value of f at the midpoint
un + wn
vn := ,
2
where two cases may arise.

• (+) If f (vn ) ≥ 0 , then

f (un ) ≤ 0 ≤ f (vn ), f (wn ),

and the interval narrows to the left-hand half [un+1 , wn+1 ] := [un , vn ] .
• (−) If f (vn ) < 0 , then

f (un ), f (vn ) ≤ 0 ≤ f (wn ),

and the interval narrows to the right-hand half [un+1 , wn+1 ] := [vn , wn ] .

In either case, f (un+1 ) ≤ 0 ≤ f (wn+1 ) with a narrower interval [un+1 , wn+1 ] .

The method of bisection thus produces sequences of numbers with features
described in the following proposition, the proof of which illustrates the concept
of mathematical induction at the level of precalculus, for instance, at the level
of Hungerford and Mercer [1985, 517–526].

Proposition 1. For each function f defined on a closed interval [u0 , w0 ] where

u0 ≤ w0 and
f (u0 ) ≤ 0 ≤ f (w0 ),
the method of bisection produces sequences of numbers (un ) and (wn ) such
that
u0 ≤ · · · ≤ un ≤ un+1 ≤ · · · ≤ wn+1 ≤ wn ≤ · · · ≤ w0 ,
f (un ) ≤ 0 ≤ f (wn ),
|w0 − u0 |
|wn − un | = .
2n

6
 Applications of Sequences and Limits 119

Proof: Because bisection proceeds inductively, the proof proceeds by induction.

By hypothesis,

u0 ≤ w0 ,
f (u0 ) ≤ 0 ≤ f (w0 ),
w0 − u0 = (w0 − u0 )/20 .

Hence, assume that for some nonnegative integer n ,

un ≤ wn ,
f (un ) ≤ 0 ≤ f (wn ),
wn − un = (w0 − u0 )/2n .

Step n . Then un ≤ vn ≤ wn , because

un + un un + wn un + wn wn + wn
un = ≤ = vn = ≤ = wn .
2 2 2 2
Two cases may arise:
• If f (vn ) ≥ 0 , then wn+1 := vn and the preceding inequalities give

un+1 = un ≤ vn = wn+1 ,

whence

u0 ≤ · · · ≤ un ≤ un+1 ≤ vn ≤ wn+1 ≤ wn ≤ · · · ≤ w0 ,

f (un+1 ) = f (un ) ≤ 0 ≤ f (vn ) = f (wn+1 ),

wn+1 − un+1 = vn − un
un + wn wn − un (w0 − u0 )/2n w0 − u0
= − un = = = .
2 2 2 2n+1

• Whereas if f (vn ) < 0 , then un+1 := vn and

un+1 = vn ≤ wn = wn+1 ,

whence

u0 ≤ · · · ≤ un ≤ un+1 ≤ vn ≤ wn+1 ≤ wn ≤ · · · ≤ w0 ,

f (un+1 ) = f (vn ) ≤ 0 ≤ f (wn ) = f (wn+1 ),

wn+1 − un+1 = wn − vn
un + wn wn − un (w0 − u0 )/2n w0 − u0
= wn − = = = .
2 2 2 2n+1

7
120 Tools for Teaching 1999

In either case,

u0 ≤ · · · ≤ un ≤ un+1 ≤ vn ≤ wn+1 ≤ wn ≤ · · · ≤ w0 ,

f (un+1 ) ≤ 0 ≤ f (wn+1 ),
w0 − u0
wn+1 − un+1 = .
2n+1
Example 4. Consider the problem of determining in advance the number
n of steps sufficient (though perhaps not necessary) to compute to twelve
significant digits the solution d of the equation x2 − 2 = 0 from u0 := 1
and w0 := 2 . Before starting the bisection algorithm, we know that the
formula
w0 − u0 2−1 1
wn − un = n
= n = n
2 2 2
expresses the accuracy that will be achieved at the n th step. Thus, the
problem amounts to solving for n the equation

( 1/2) ∗ 10−12 ∗ d ≥ wn − un = 2−n .

Because d remains unknown, but 1 < d < 2 , it suffices to solve

( 1/2) ∗ 10−12 ∗ d > ( 1/2) ∗ 10−12 ∗ 1 ≥ wn − un = 2−n .

Hence
2n−1 ≥ 1012 .
Logarithms to base ten would give the solution in the form n − 1 ≥
12 log(10)/ log(2) = 12/ log(2) = 39.863 . . . , so that n ≥ 41 suffices.
However, computing a logarithm can demand more computational re-
sources than computing d . One way to find such a value of n without
logarithms is to compute powers of 2 until one of them exceeds 1012 , for
instance, 2 , 22 , 24 , 28 = 256 , 216 , 232 > 4∗109 , 232 ∗28 > 4∗109 ∗256 >
1012 , whence n − 1 = 32 + 8 = 40 and n = 41 suffices. A faster but
less conservative method is to use the inequality 24 > 10 to simplify the
problem 2n−1 ≥ 1012 into the form

2n−1 ≥ (24 )(n−1)/4 > 10(n−1)/4 ≥ 1012

whence (n − 1)/4 ≥ 12 and finally n ≥ 49 . This faster method in-

volves neither logarithms nor powers of two, but it leads to 49 − 41 = 8
additional unnecessary steps.

8
 Applications of Sequences and Limits 121

Exercises
The following four exercises provide practice using the method of bisection.

√
4. Apply the method of bisection to compute 3.
√
5. Apply the method of bisection to compute 3 2 .
√
6. Determine a number n of steps sufficient to compute 3 by the method
of bisection to twelve significant digits.
√
3
7. Determine a number n of steps sufficient to compute 2 by the method
of bisection to twelve significant digits.

The following two exercises show results from the method of bisection ap-
plied to situations that do not satisfy the conditions stated for the method of
bisection.
8. Apply the method of bisection to

x4 + x2 − 6
f (x) = 4
x − 5x2 + 6
with u0 = 1 and w0 = 2 .
9. Apply the method of bisection to

5x4 − 3x3 − 3x2 − 3x − 8

g(x) =
x4 − 5x2 + 6
with u0 = 1 and w0 = 2 .

The following four exercises demonstrate the solution of equations in ap-

plications.

10. Apply the method of bisection to compute the positive solution v∗ of the
equation

10,193.75v 4 + 193.75v 3 + 193.75v 2 + 193.75v − 9,987.50 = 0.

This equation gives the internal rate of return r∗ = 2[(1/v∗ ) − 1] of the

two-year note issue by the U.S. Treasury on 31 August 1993, in terms of
the purchase price $9,987.50 on the day of issue, the semi-annual amount
of interest, $193.75 , and the value redeemed at maturity, $10,000.00 [Wall
Street Journal 1993].

9
 122 Tools for Teaching 1999

11. Apply the method of bisection to compute the positive solution X of the
equation

X 3 + 0.1000269X 2 − 8.070 000 100 7 × 10−7 X − 2.70883 × 10−19 = 0.

In chemistry, X represents the equilibrium concentration of hydrogen ions

[H+ ] in a water solution of monoprotic acid, H(CH3 CO2 ) , found in vinegar,
and sodium acetate, Na(CH3 CO2 ) , the “salt” obtained by reducing the acid.
The quantity pH = − log10 ([H+ ]) = − log10 X represents the acidity of the
solution [MacLeod 1984].
12. Determine a number n of steps sufficient to compute the solution v∗ of

10,193.75v 4 + 193.75v 3 + 193.75v 2 + 193.75v − 9,987.50 = 0

by the method of bisection to twelve significant digits. Your answer may

depend on your initial values u0 and w0 .
13. Determine a number n of steps sufficient to compute the solution X of

X 3 + 0.1000269X 2 − 8.070 000 100 7 × 10−7 X − 2.70883 × 10−19 = 0

by the method of bisection to twelve significant digits. Your answer may

depend on your initial values U0 and W0 .

The following two exercises demonstrate issues arising in the design of

algorithms.

14. Investigate the consequences of modifying the test so that either f (vn ) > 0
or f (vn ) ≤ 0 , instead of f (vn ) ≥ 0 or f (vn ) < 0 .
15. Investigate the consequences of requiring that f (u0 ) < 0 < f (w0 ) instead
of f (u0 ) ≤ 0 ≤ f (w0 ) .

4. Further Issues
The preceding considerations raise a first and fundamental issue, about the
precise nature of the “number” d . First, d is not a rational number.
Proposition 2. There is no rational number d such that d2 = 2 .

Proof: The main resource for this proof is the Fundamental Theorem of Arith-
metic and the associated unique prime factorization, borrowed from algebra
[Birkhoff and Mac Lane 1991, 23]. If d2 = 2 and d = p/q is rational, with
relatively prime integers p and q , then 2 = (p/q)2 , whence 2q 2 = p2 . The
unique prime factorizations of p = pr11 · · · prkk and q = q1s1 · · · qs then would
give 2q12s1 · · · q2s = 2q 2 = p2 = p2r
1 · · · pk , where the left-hand side has an
1 2rk

10
 Applications of Sequences and Limits 123

odd power of 2 while the right-hand side has an even power of 2 , so that they
cannot equal each other.
Therefore, discussing the “number” d requires numbers different from the
rational numbers. What are such numbers? Essentially, the answer lies right
in the numerical experiments from the preceding examples. All a user ever
sees from d are sequences of rational numbers such as (un ) and (wn ) . Thus,
d corresponds to all the sequences produced by the bisection method (or any
√ of x − 2 = 0 ; the
2
other algorithm) in computing the nonnegative solution
collection
√ of all such sequences could be denoted by 2 . This description of
2 in terms of a collection of sequences is similar to the description of a rational
number 2/3 in terms of such equivalent fractions as 4/6 or -14/-21 . Such ways
to define real numbers are called equivalent Cauchy sequences and are outlined
in Spivak [1980, 565].
Alternatively, because all that a user ever sees from d are either rationals p/q
√
such that (p/q)2 < 2 or rationals p/q such that 2 < (p/q)2 , the “number” 2
can be defined as the set of all positive rationals
√ p/q such that 2 < (p/q)2 . All
that a digital computer ever produces for 2 is either a rational number from
that set, with 2 < (p/q)2 , or a positive rational number from its complement,
with (p/q)2 < 2 . Such ways to define real numbers are called Dedekind cuts
and are described in detail in the original source [Dedekind 1963] and in the
calculus text [Spivak 1980, 555].
Thus, whether the results from the method of bisection “converge” toward
a solution of the equation f (x) = 0 depends on both the function f and the
type of “numbers” allowed by the problem under consideration. For example,
with f (x) = x2 − 2 and real numbers in√ [u0 , w0 ] = [1, 2] , the sequences (un )
and (wn ) “converge” to the solution 2 . In contrast, working with only
rational numbers produces
√ the same sequences (un ) and (wn ) —but they fail
to converge, because 2 is not a rational number.
The mathematical concept of “convergence” of a sequence does not mean
that the terms of the sequence get closer and closer to the limit, but, rather, that
they eventually get and remain closer than any specified tolerance.
Definition 1. A sequence (vn ) converges to a limit v∗ if and only if
for every ε > 0 (the “tolerance”)
there exists an integer m such that
|vn − v∗ | < ε for every integer n > m .

Example 5. In computing the positive solution d of the equation x2 −

2 = 0 from the initial estimates u0 := 1 and w0 := 2 , the method of
bisection produces three sequences,
√ (un ) , (vn ) , and (wn ) , which all
three converge to the limit 2 in the real numbers. Indeed,
for every ε > 0 ,
there exists an integer m such that ε > 2−m > 0 ,
whence for every n > m , we have
|un − v∗ | , |vn − v∗ | , |wn − v∗ | ≤ wn − un < 2−n < ε .

11
 124 Tools for Teaching 1999

The justification of the statement that “for every ε > 0 there exists an
integer m such that ε > 2−m > 0 ” depends on the definition of the real
numbers adopted in the course.

Another issue pertains to multiple zeroes. If a function f has more than one
zero inside an interval [u0 , w0 ] , then the method of bisection converges to some
zero of f , which zero depends on f , as seen in Benjamin [1987]. Moreover,
the rate at which the method of bisection converges does not depend on the
function f but depends only on the binary expansion of the selected zero
[Nievergelt 1995].
Yet another issue relates to simultaneous systems of equations. Systems of
equations with more than one real unknown require more sophisticated meth-
ods specialized to individual systems. For one complex polynomial equation
with one complex unknown, which constitutes a special case of a system of
two equations with two real unknowns, an algorithm of Herman Weyl’s works
similarly to the method of bisection, as explained by Victor Pan [1997].
Still another issue relates to the type of equation that defines a number.
Some numbers cannot be defined by polynomial equations and require equa-
tions involving nonalgebraic functions. One such number is the ratio π of the
circumference to the diameter of a circle [Lang 1965, 493–494].
A different issue relates to the accuracy obtainable in practice. In practice,
the accuracy of the results depends on the accuracy of the computations of
the values f (vn ) , in particular, whether the computer carries enough digits to
determine which of f (vn ) ≥ 0 or f (vn ) < 0 holds. Eventually, a computer’s
rounding can yield the wrong sign.

Exercises
16. Define the positive solution of x3 = 2 in terms of Dedekind cuts.
17. This exercise pertains to the issue of computability.
a) Prove that computations carried out to exactly ten decimal digits (round-
ing all intermediate arithmetic results to ten significant decimal digits)
erroneously indicate that the following quadratic equation has no real
solutions:

0.3 124 999 999 ∗ Z 2 − 0.707 106 781 1 ∗ Z + 0.4 = 0.

b) Prove that the following quadratic equation has two distinct real solu-
tions:
0.3 124 999 999 ∗ Z 2 − 0.707 106 781 1 ∗ Z + 0.4 = 0.

5. Solving Equations through Regula Falsi

The purpose of this section is to allow students to study on their own an
algorithm not presented in class, as may happen after they leave school. To

12
 Applications of Sequences and Limits 125

this end, this section outlines the algorithm and leaves the investigations to the
exercises.
The method of false position, also called by its Latin name regula falsi, proceeds
as does the method of bisection, except for a different choice of the intermediate
points vn , which need no longer lie at the midpoint of [un , wn ] . To solve an
equation of the form
f (x) = 0,
regula falsi requires only two initial estimates u0 < w0 with

f (u0 ) < 0 < f (w0 ).

For a more convenient notation, define

p0 := f (u0 ),
q0 := f (w0 ),

so that the graph of f passes through the two points

(u0 , p0 ), (w0 , q0 ),

with p0 < 0 < q0 . Therefore, the straight line through (u0 , p0 ) and (w0 , q0 )
crosses the horizontal axis at some point (v0 , 0) with u0 < v0 < w0 . Regula
falsi selects this point v0 .
Then, exactly as for the method of bisection, at least two cases may arise.
• (+) If f (v0 ) > 0 , then

f (u0 ) ≤ 0 ≤ f (v0 ), f (w0 ),

and the interval narrows to the left-hand half [u1 , w1 ] := [u0 , v0 ] .

• (−) If f (v0 ) < 0 , then

f (u0 ), f (v0 ) ≤ 0 ≤ f (w0 ),

and the interval narrows to the right-hand half [u1 , w1 ] := [v0 , w0 ] .

In either case, f (u1 ) ≤ 0 ≤ f (w1 ) with a narrower interval [u1 , w1 ] . Regula
falsi then iterates the same steps, finding a point v1 between u1 and w1 , and
so on.

Exercises
18. Design a way to specify what the algorithm does for the yet unexamined
cases f (u0 ) = 0 , f (v0 ) = 0 , and f (w0 ) = 0 .
19. Write an equation y = mv + b for the straight line through two points
(u0 , p0 ) and (w0 , q0 ) , and solve for the value v0 where the line crosses the
horizontal axis: this gives a formula for v0 in terms of u0 , p0 , w0 , q0 .

13
 126 Tools for Teaching 1999

20. Imagine that you are a newly hired junior member of a scientific computing
team and that you have just received the following assignment, with results
to be reported to your boss: Locate a source in the library or elsewhere
that proves that regula falsi converges to a solution of f (x) = 0 . You may
look in the index of books in the category called “numerical analysis” under
Library of Congress Classification Numbers (“call numbers”) QA297 . . . ,
where a similar (but not quite as good) method called “secant method”
may also appear. You need not reproduce the content of the proof here.
Just name the source and report the page numbers, the authors’ full names,
the full book title, publisher, year, and the location in the library (such as
Dewey Decimal Classification Number or Library of Congress Classification
Number), so that on the sole basis of your report your boss can find this
proof very quickly.
√
21. Apply a few steps of regula falsi to compute 3 .
22. Prove that each of the sequences (un ) and (wn ) produced by regula falsi
converges to some limit in the real numbers.
23. Apply the method of regula falsi to compute the nonnegative solution X of

X 3 + 0.1000269X 2 − 8.070 000 100 7 × 10−7 X − 2.70883 × 10−19 = 0.

(See also Exercise 11.)

24. If you have access to a calculator or a computer with a built-in program or
software to solve equations, try it to compute the nonnegative solution X
of

X 3 + 0.1000269X 2 − 8.070 000 100 7 × 10−7 X − 2.70883 × 10−19 = 0.

(See also Exercise 11.)

6. Conclusions
The mathematical concepts of sequence and limit are not only technicali-
ties necessary for other concepts; they are also concepts that arise in scientific
computing, and they form the basis for all the other concepts from calculus,
including derivatives, integrals, inverse functions, and infinite series. More-
over, the mathematical concept of limit appears to be the major philosophical
achievement of mathematics presented, distorted, or withheld, in calculus.

14
 Applications of Sequences and Limits 127

7. Solutions to the Odd-Numbered

Exercises
1. The following statement is false: “For each function f defined in an open
interval containing a number T , if the value of f (t) gets closer and closer to
a number L as the value of t gets closer and closer to the number T , then
limt→T f (t) = L. ” Indeed, consider the function f defined by f (t) = 1−t2 ,
for instance, with T = 0 and L = 3 . Then f (t) gets closer and closer to 3
as t gets closer and closer to 0 , but the limit is 1 , not 3 .
The concept of limit does not merely require that f (t) get closer and
closer to L ; it requires that for each positive specification s > 0 the value
of f (t) gets within s of the limit L for every value of t within some
positive tolerance r > 0 of the point T :
for each s > 0 ,
there exists some r > 0 (which may depend on f , T , and s ) such that
|f (t) − L| < s
for every t with 0 < |t − T | < r .
For the particular example with f (t) = 1 − t2 and T = 0 , the value of f (t)
never gets within s = 1 of 3 ; indeed, while f (t) gets closer and closer to
3 , it always remains at least 2 away.
3. x3 = 2 .
5. To solve the equation x3 − 2 = 0 with g(x) := x3 − 2 through the method
of bisection, for instance, starting with u0 := 1 and w0 := 2 , the first few
steps produce the results of Table 2.

Table 2.
Steps in the calculation, by means of bisection, of the real cube root of 2.

n un vn wn

0 1 1.5 2
1 1 1.25 1.5
2 1.25 1.375 1.5
3 1.25 1.3125 1.375
4 1.25 1.28125 1.3125
5 1.25 1.265625 1.28125
6 1.25 1.2578125 1.265625
7 1.2578125 1.26171875 1.265625
8 1.2578125 1.259765625 1.26171875
9 1.259765625 1.2607421875 1.26171875
10 1.259765625 1.26025390625 1.2607421875
.. .. .. ..
. . . .

√
3
The last line of the table shows that 1.259765625 < 2 < 1.2607421875 .

15
 128 Tools for Teaching 1999

7. Proceed as in Example 4, with the same result n ≥ 41 .

√
9. From u0 = 1 and w0 = 2 , the method of bisection converges to 2 , which
is not a solution of g(x) = 0 but instead marks a vertical asymptote of g :

5x4 − 3x3 − 3x2 − 3x − 8 5(x − 1.6)(x + 1)(x2 + 1)

g(x) = = .
x4 − 5x2 + 6 (x2 − 2)(x2 − 3)

11.
-12
2×10
-12
1.5×10
-12
1×10
-13
5×10
U0 X W 0
2×10
-6
4×10
-6
6×10
-6
8×10
-6 0.00001
-13
-5×10
-12
-1×10
-12 F
-1.5×10

Figure 4. F (X) = X 3 + 0.1000269X 2 − 8.070 000 100 7 × 10−7 X − 2.70883 × 10−19 .

See Figure 4. Substitutions show that F (0) = −2.70883 ∗ 10−19 < 0 <
1 − 2.70883 ∗ 10−19 < F (1) . Consequently, U0 := 0 and W0 := 1 satisfy the
requirements, but from such initial values the bisection requires dozens of
steps to produce at least one exact significant digit, showing that 8.0∗10−6 <
X < 8.1∗10−6 . Additional steps give the approximation X ≈ 8.067∗10−6 .
See also Figure 4.
13. Because 8.0 ∗ 10−6 < X < 8.1 ∗ 10−6 , it follows that for an accuracy of
twelve significant digits it suffices that

( 1/2) ∗ 10−12 ∗ X > ( 1/2) ∗ 10−12 ∗ 8.0 ∗ 10−6 = 4 ∗ 10−18 ≥ (W0 − U0 )/2n .

With U0 := 0 and W0 := 1 , these inequalities lead to 2n+2 = 2n ∗ 4 ≥ 1018 ,

whence n ≥ 58 suffices.

15. With f (u0 ) < 0 < f (w0 ) , it can happen that f (vn ) = 0 at some later stage
n > 0 . This does not cause any difficulty, but it requires a special case
in the bisection algorithm, for instance, stopping the algorithm, or setting
uk := wk := vn for every k > n .

16
 Applications of Sequences and Limits 129

17. a) For the quadratic equation

0.3 124 999 999 ∗ Z 2 − 0.707 106 781 1 ∗ Z + 0.4 = 0,

hand calculations rounding every result to ten significant decimal digits

or a ten-digit decimal calculator (HP-15C) computes the discriminant as
follows, where float(r) denotes the result of computing r to exactly
ten significant decimal digits.

float(b2 ) = float([−0.707 106 781 1]2 )

= 0.499 999 999 9;
float(a ∗ c) = float(0.3 124 999 999 ∗ 0.4)
= 0.125 000 000 0;
float(4 ∗ a ∗ c) = float(4 ∗ 0.3 124 999 999 ∗ 0.4)
= 0.500 000 000 0;
float(b2 − 4 ∗ a ∗ c) = float([−0.707 106 781 1]2 − 4 ∗ 0.3 124 999 999 ∗ 0.4)
= 0.499 999 999 9 − 0.500 000 000 0
= −0.000 000 000 1
< 0.

This negative value of the computed discriminant would indicate that

the proposed quadratic equation has no real solution.
b) Yet exact calculations reveal that

b2 = [−0.707 106 781 1]2

> 0.499 999 999 877;
a ∗ c = 0.3 124 999 999 ∗ 0.4
= 0.124 999 999 96;
4 ∗ a ∗ c = 4 ∗ 0.3 124 999 999 ∗ 0.4
= 0.499 999 999 84;
b − 4 ∗ a ∗ c = [−0.707 106 781 1]2 − 4 ∗ 0.3 124 999 999 ∗ 0.4
2

> 0.499 999 999 877 − 0.499 999 999 84

= +0.000 000 000 037
> 0.

Therefore, the proposed quadratic equation has two distinct real roots.
19. Write the two-point equation of the line, and solve for v0 where y = 0 :
y2 − y1
y − y1 = (x − x1 ),
x2 − x1
q 0 − p0
y − p0 = (x − u0 ),
w0 − u0

17
 130 Tools for Teaching 1999

q 0 − p0
0 − p0 = (v0 − u0 ),
w0 − u0
w0 − u 0
−p0 ∗ = (v0 − u0 ),
q 0 − p0
w0 − u 0
u 0 − p0 ∗ = v0 .
q 0 − p0

21. For instance, choose u0 = 1 and w0 = 2 :

u0 = 1, p0 = f (u0 ) = f (1) = 12 − 3 = −2 < 0 : (u0 , p0 ) = (1, −2),

w0 = 2, q0 = f (w0 ) = f (2) = 22 − 3 = 1 > 0 : (w0 , q0 ) = (2, 1),
w0 − u0 2−1 1 5
v0 = u0 − p0 ∗ = 1 − (−2) ∗ = 1 + 2 ∗ = = 1.666 . . . .
q 0 − p0 1 − (−2) 3 3
f (v0 ) = f ( 5/3) = ( 5/3)2 − 3 = 25/
9 −3= 25/
9 − 27/
9 = − 2/9 < 0.

Thus, u1 = v0 = 5/3 , and the new interval is [u1 , w1 ] = [ 5/3, 2] . Repeat the
same process:

u1 = 5/3, p1 = f (u1 ) = f ( 5/3) = − 2/9 < 0 : (u1 , p1 ) = ( 5/3, − 2/9),

w1 = w0 = 2, q1 = f (w1 ) = f (2) = 22 − 3 = 1 > 0 : (w1 , q1 ) = (2, 1),
w1 − u1 2 − 5/3 1/
3
v1 = u1 − p1 ∗ = 5/3 − (− 2/9) ∗ = 5/ + ( 2/ ) ∗
3 9
q 1 − p1 1 − (− /9)
2 11/9
57
= 5/3 + 2/33 = = 1.72 72 72 . . . .
33
f (v1 ) = f ( 57/33) = ( 57/33)2 − 3 = − 2/121 < 0.

Thus, u2 = v1 = 57/33 , and the new interval is [u1 , w1 ] = [ 57/33, 2] . Repeat

the same process:

u2 = 57/33, p2 = f (u2 ) = f ( 57/33) = − 2/121 < 0 : (u2 , p2 ) = ( 57/33, − 2/121),

w2 = w1 = 2, q2 = f (w2 ) = f (2) = 22 − 3 = 1 > 0 : (w2 , q2 ) = (2, 1),
w2 − u2 2 − 57/33 9/
33
v2 = u2 −p2 ∗ = 57/33−(− 3/121)∗ = 57/ +( 2/ )∗
33 121
q 2 − p2 1 − (− /121)
2 123/121
2,343
= 57/33 + 6/1,353 = = 1.731 707 317 . . . .
1,353
f (v2 ) = f ( 2,343/1,353) = ( 2,343/1,353)2 − 3 < 0.

Thus, u3 = v2 = 2,343/1,353 , w3 = w2 = 2 and the new interval is [u2 , w2 ] =

[ 2,343/1,353, 2] = [1.731 . . . , 2 . . . ] .

18
 Applications of Sequences and Limits 131

23. See also Figure 4.

X 3 + 0.1000269X 2 − 8.070 000 100 7 × 10−7 X − 2.70883 × 10−19 = 0.
At X = U0 = 0 the left-hand side is negative, while at X = W0 = 10−5 it
is positive. the method of false position leads to X ≈ 8.067 ∗ 10−6 , which
corresponds to pH = − log10 (8.067 ∗ 10−6 ) ≈ 5.093 . . . as corroborated in
MacLeod [1984]. See item (4) below for greater accuracy, starting from the
following computational detail to locate the positive zero of
F (X) = X 3 + 0.1000269X 2 − 8.070 000 100 7 × 10−7 X − 2.70883 × 10−19 .

1. Because the calculator produced the estimate X̃ ≈ 8.067 ∗ 10−6 , it

may be prudent to verify the nearest powers of ten, 10−6 and 10−5 ,
to verify that 10−6 < X < 10−5 :
F (10−6 ) = 10−18 + 0.1000269 ∗ 10−12 − 8.070 000 100 7 × 10−7 ∗ 10−6
− 2.70883 × 10−19
< 10−18 + 1.0003 ∗ 10−13 − 8.0 ∗ 10−13 − 2.7 × 10−19
= 10 ∗ 10−19 + 1.0003 ∗ 10−13 − 8.0 ∗ 10−13 − 2.7 × 10−19
= 7.3 ∗ 10−19 − 7.0007 ∗ 10−13 < −6.007 ∗ 10−13
< 0;
F (10−5 ) = 10−15 + 0.1000269 ∗ 10−10 − 8.070 000 100 7 × 10−7 ∗ 10−5
− 2.70883 × 10−19
> 10−15 + 1.0 ∗ 10−11 − 10.0 ∗ 10−12 − 2.7 × 10−19
= 10−15 + 0 − 2.7 × 10−19
> 0.
2. It may also be convenient and more accurate to factor out a power of
ten. To this end, let x := 106 X , so that 1 < x < 10 , and factor out
10−13 from F , so that
F (X) = X 3 + 0.1000269X 2 − 8.070 000 100 7 × 10−7 X
− 2.70883 × 10−19
= (10−6 x)3 + 0.1000269(10−6 x)2 − 8.070 000 100 7 × 10−7 (10−6 x)
− 2.70883 × 10−19

= 10−13 ∗ 10−5 ∗ x3 + 1.000269 ∗ x2 − 8.070 000 100 7 ∗ x

− 883 × 10−6
= 10−13 ∗ f (x),
f (x) := 10−5 ∗ x3 + 1.000269 ∗ x2 − 8.070 000 100 7 ∗ x
− 2.70883 × 10−6 .

19
132 Tools for Teaching 1999

3. This confirms that x lies in the narrower range 8 < x < 9 , so that
8 ∗ 10−6 < X < 9 ∗ 10−6 :

f (9) := 10−5 ∗ 93 + 1.000269 ∗ 92 − 8.070 000 100 7 ∗ 9

− 2.70883 × 10−6
= 0.00729 + 1.000269 ∗ 81 − 8.070 000 100 7 ∗ 9
− 2.70883 × 10−6
> 0.00729 + 1.0 ∗ 81 − 8.1 ∗ 9 − 3.0 × 10−6
= 0.00729 + 81 − 72.9 − 3.0 × 10−6
> 81 − 72.9 = 8.2 > 0;
f (8) := 10−5 ∗ 83 + 1.000269 ∗ 82 − 8.070 000 100 7 ∗ 8
− 2.70883 × 10−6
= 0.00512 + 1.000269 ∗ 64 − 8.070 000 100 7 ∗ 8
− 2.70883 × 10−6
< 0.00512 + 1.001 ∗ 64 − 8.07 ∗ 8
− 0 × 10−6
= 0.00512 + 64.064 − 64.56
< 0.00512 − 0.504 < −0.498 < 0.

4. From u := 8 and w := 9 , regula falsi applied to f produces v :=

8.060 163 699 51 and f (v) < 0 . The method stabilizes and stays at
v9 := 8.067 179 571 27 , so that X̃ = 8.067 179 571 27 ∗ 10−6 .
5. Try x := 8.1 to verify that 8 < x < 8.1 , so that 8 ∗ 10−6 < X <
8.1 ∗ 10−6 :

f (8.1) := 10−5 ∗ (8.1)3 + 1.000269 ∗ (8.1)2 − 8.070 000 100 7 ∗ (8.1)

− 2.70883 × 10−6
> 0.00524961 + 1.0 ∗ 64.81 − 8.1 ∗ 8.1 − 2.70883 × 10−6
= 0.00524961 + 64.81 − 64.81 − 2.70883 × 10−6
> 0.00524961 + 0 − 3.0 × 10−6 = 0.005 246 61
> 0.

6. Thus, X = 8.05 ∗ 10−6 ± 0.05 ∗ 10−6 , with a relative error less than
1% :
|X − X̃| 0.05 ∗ 10−6
< = 0.00625 = 0.625%.
|X| 8 ∗ 10−6
(The number of digits listed has nothing to do with accuracy: anyone
who insists on writing either 8.0 or 8.1 doubles the error.)

20
 Applications of Sequences and Limits 133

References
Benjamin, Arthur. 1987. The bisection method. Which root? American Mathe-
matical Monthly 94 (9): 861–863.
Birkhoff, Garrett, and Saunders Mac Lane. 1977. A Survey of Modern Algebra.
4th ed. 50th anniversary 1991 printing. New York: Macmillan.
Dedekind, Richard. 1963. Essays on the Theory of Numbers. New York: Dover.
1963. (Republication of the English translation from the German published
by the Open Court Publishing Company in 1901.)
Edwards, Charles Henry. 1979. The Historical Development of the Calculus. New
York: Springer-Verlag.
Etgen, Garret, reviser. 1999. Salas and Hille’s Calculus: One and Several Variables.
8th ed. New York: Wiley.
Hungerford, Thomas W., and Richard Mercer. 1985. Precalculus Mathematics.
2nd ed. Philadelphia, PA: Saunders College Publishing.
Lang, Serge. 1965. Algebra. Reading, MA: Addison-Wesley.
Mac Lane, Saunders. 1997. On the Harvard Consortium Calculus. Notices of
the American Mathematical Society 44 (8): 893.
MacLeod, Allan J. 1984. The calculation of pH. International Journal of Mathe-
matics Education in Science and Technology 15 (6): 691–696.
Neugebauer, O. 1969. Vorlesungen über die Geschichte der antiken mathematischen
Wissenschaften. Erster Band: Vorgriechische Mathematik. 2nd ed. Berlin:
Springer-Verlag.
Nievergelt, Yves. 1995. Bisection hardly ever converges linearly. Numerische
Mathematik 70 (1995): 111–118. Reviewed in Mathematical Reviews #95k:65055.
Ostebee, Arnold, and Paul Zorn. 1997. Calculus from Graphical, Numerical, and
Symbolic Points of View. 2 vols. Fort Worth, TX: Saunders College Publishing.
. 1998. Multivariable Calculus from Graphical, Numerical, and Symbolic
Points of View. Fort Worth, TX: Saunders College Publishing.
Pan, Victor Y. 1997. Solving a polynomial equation: Some history and recent
progress. SIAM Review 39 (2): 187–220.
Pulskamp, Richard J., and James A. Delaney. 1991. Computer and Calculator
Computation of Elementary Functions. UMAP Modules in Undergraduate
Mathematics and Its Applications: Module 708. Lexington, MA: COMAP,
1991. Reprinted in The UMAP Journal 12 (1991): 315–348. Reprinted in
UMAP Modules: Tools for Teaching 1991, edited by Paul J. Campbell, 1–34.
Lexington, MA: COMAP, 1992.
Schelin, C.W. 1983. Calculator function approximation. American Mathematical
Monthly 90 (5): 317–325.

21
134 Tools for Teaching 1999

Spivak, Michael. 1980. Calculus. 2nd ed. Wilmington, DE: Publish or Perish.
Stewart, James. 1995. Calculus. 3rd ed. Pacific Grove, CA: Brooks/Cole.
Strang, Gilbert. 1991. Calculus. Wellesley, MA: Wellesley-Cambridge Press.
Wall Street Journal. 1993. Western Ed. 79 (25 August 1993): C21.
van der Waerden, Bartel Leenert. 1983. Geometry and Algebra in Ancient Civi-
lizations. New York: Springer-Verlag.

About the Author

Yves Nievergelt graduated in mathematics from the École Polytechnique
Fédérale de Lausanne (Switzerland) in 1976, with concentrations in functional
and numerical analysis of PDEs. He obtained a Ph.D. from the University of
Washington in 1984, with a dissertation in several complex variables under the
guidance of James R. King. He now teaches complex and numerical analysis
at Eastern Washington University.
Prof. Nievergelt is an associate editor of The UMAP Journal. He is the au-
thor of several UMAP Modules, a bibliography of case studies of applications
of lower-division mathematics (The UMAP Journal 6 (2) (1985): 37–56), and
Mathematics in Business Administration (Irwin, 1989). His new book is Wavelets
Made Easy (Birkhäuser, 1999).

22