Beruflich Dokumente
Kultur Dokumente
Barbara A. Shipman
To cite this article: Barbara A. Shipman (2012) A Comparative Study of Definitions on Limit and
Continuity of Functions, PRIMUS, 22:8, 609-633, DOI: 10.1080/10511970.2011.630714
Barbara A. Shipman
1. INTRODUCTION
While the differences in these definitions may appear small, they lead to conse-
quences that are profound. For example, in standard calculus courses, students
learn that a function has a limit at a point if and only if it has both a left-hand
limit and a right-hand limit there and these limits are equal. Calculus textbooks
often state this as a theorem, as in [16] and [17]. The textbook [16] emphasizes
in bold-face that limx→a f (x) refers to a two-sided limit, and [17] highlights in
italics and in bold-face that to have a limit at c, a function must be defined on
both sides of c and that “ordinary limits” are called two-sided. √ [17] gives an
example in which the function H : [−2, 2] → R, with H(x) = 4 − x2 , “does
not have ordinary two-sided limits at either −2 or 2.” [10] also emphasizes in
italics that the point at which a limit is taken is approached from both sides.
Using the definitions of mathematical analysis, however, this theorem of calcu-
lus is no longer true. A function can have a limit at a point without having both
a right-hand limit and a left-hand limit there. For example, by the definitions
of analysis, H has a limit at −2 but not a left-hand limit there.
As another example, consider the function β : Q → R where β(x) = x +
4, defined only on the rational numbers. By the definitions in standard calculus
textbooks, β does not have a limit anywhere, but by the definitions of analysis,
β has a limit at every real number.
The reverse situation also occurs, where facts of analysis no longer hold
by the definitions taught in standard calculus courses. In analysis, continuity of
a function f at a non-isolated point p of its domain is defined by the three-fold
requirement that f have a limit at p, that f be defined at p, and that these values
be equal. By this definition, each of the functions H and β defined above is
continuous at every point in its domain. However, if one uses the definitions
from standard calculus courses, this classic characterization of continuity at
a point, which embodies so elegantly the spirit of the concept, is no longer
true. In calculus, β is not continuous anywhere, and the continuity of H at a
point in its domain is judged by three different criteria, depending on where
in this interval the point lies (whether it is a left endpoint, a right endpoint, or
an interior point). In standard calculus courses, H may said to be continuous,
continuous on [−2, 2], or both, depending on the textbook, but, in all of these
cases, the definition of limit in calculus precludes H from being continuous by
the classic criterion of analysis.
The subtlety of having to make different definitions of continuity at an
endpoint of a domain [a, b] in calculus is not only inconvenient but can lead to
occasional slips of language. For example, [15] defines continuity of f at a to
mean limx→a f (x) = f (a) (with the limit being two-sided). The textbook then
Comparative Study of Definitions on Limit and Continuity 611
This paper addresses these apparent contradictions through five guided class-
room discoveries ideal for students in a first rigorous course in real analysis,
an honors calculus course, or a course on transitions to higher-level mathe-
matical thinking. In resolving these issues, students encounter, perhaps for the
first time, philosophical questions that arise in creating mathematical defini-
tions. In formulating a definition, what is the behavior that one is trying to
capture? For what objects is it relevant to ask whether this behavior is sat-
isfied or not? Once a definition correctly captures the concept in all cases
where the idea is relevant, should one then extend the definition to objects
for which the concept is vacuous or irrelevant? What effect does extending a
definition to a more general context have on mathematical consequences of the
definition?
In the historical development of what is now modern analysis, mathemati-
cians were not initially concerned with specifying the class of functions that
their definitions considered. In the time of Cauchy (1789–1857), the notion of
a function was not yet fully developed or rigorous. Even simple step functions
defined on the real line, such as those that Fourier obtained in his investiga-
tions of infinite series, were not recognized as functions by mathematicians at
the time (see [4], for example, which studies the ideas and definitions of anal-
ysis in their historical setting). Early definitions on limit and continuity sought
to capture the expected behavior of a function without specifying what func-
tions they were considering or what exactly constituted a function. One sees
this, for example, in Cauchy’s frequently quoted description of convergence,
as translated in [5, p. 6]:
Comparative Study of Definitions on Limit and Continuity 613
vote on the answer. To improve response rates if needed, I ask the students to
indicate whether they are, say, 70% sure of the answer being considered. I then
give them a few minutes to work out their differences with each other. This
second round of peer-to-peer discussion, where students defend and question
their arguments, is helpful in improving students’ learning, according to a study
of calculus instruction [12]. Throughout the deliberations and voting process, I
remain neutral to the answer, even challenging the class with seemingly good
arguments for the incorrect answer. An open class discussion then resolves the
question.
● Three birds:
– the cardinal
– the house sparrow
– the red-tailed hawk
● Three functions:
– the function f : R − {3} → R, where f (x) = (x2 − 9)/(x − 3)
– the function g : R − {3} → R, where g(x) =√ 1/(x − 3)2
– the function h : [−3, 3] → R, where h(x) = 9 − x2
Solution: f is like the house sparrow; it satisfies the conditions on domain and
behavior. g is like the red-tailed hawk; it satisfies the restriction on the domain
but not the behavioral condition at 3. h is like the cardinal. While h fails the
condition imposed on the domain, it behaves correctly near 3, within its domain
for p = 3.
Comparative Study of Definitions on Limit and Continuity 617
Students will recall the notation limx→p f (x) = L to denote this limit.
To write this definition rigorously, I work with the class to translate it piece by
piece into the standard ε - δ form. Taking time to work this out is well worth
the effort. Section 6.1 in [13] provides an activity to guide the class in rewriting
Definition L2 as follows:
Students will recall from calculus that the derivative of a function F at p is the
limit of the difference quotient Q(x) = (F(x) − F(p))/(x − p) at p. (For exam-
ple, the limit at 3 of f (x) = (x2 − 9)/(x − 3) in Guided Discoveries 1 and 2 is
the derivative of x2 at 3.) Since the derivative is itself a limit, any restriction
on the domain in defining a limit applies to the domain of the difference quo-
tient in defining the derivative. Thus, a typical course in calculus can define the
derivative of F only at a point p where F is defined on an open interval around
p, except possibly at p.
It is interesting to note that most undergraduate analysis textbooks that
use Definition L3 for a limit place additional restrictions on the domain in
defining the derivative, even when the limit of the difference quotient may exist
by Definition L3 without the additional restriction. For example, [1], [3], and
[11] consider the limit at a cluster point of the domain but define the derivative
only at points in an (arbitrary) interval contained in the domain. Thus, taking
β : Q → R, where β(x) = x + 4 and using the definitions in [1], [3], and [11],
the difference quotient (β(x) − β(3))/(x − 3) has a limit at 3, but β does not
have a derivative at 3 (see Part III of Guided Discovery 2).
Of the references cited in this paper, only [8] defines both the limit and
the derivative at a cluster point of the domain. [3] notes that the derivative of
a function at p may be defined on more general domains since p need only be
in the domain and a cluster point of the domain. However, [3] then states that
because the significance of derivatives is “most naturally apparent for functions
defined on intervals,” its definition of derivative restricts attention to functions
on intervals.
Below are three functions and two questions about the functions.
An instructive exercise for the students is to answer the two questions about
each function, first using the definitions from their calculus textbook and then
using the definitions in their analysis textbook.
The three functions:
● S : R → R, where S(x) = x2
● H : [3, ∞) → R, where H(x) = x
● β : Q → R, where β(x) = x + 4
In answering these questions, students will notice that the domain of the
difference quotient is the domain of F with 3 removed, where the removal of 3
from the domain has no influence on the existence or value of a limit there. This
622 Shipman
activity will likely have surprising outcomes; it makes a good preparation for a
class discussion and a good topic for a mathematical essay or group homework
project.
The answers to these questions are not the same by all textbooks that may
be used in undergraduate analysis courses. For example, in [1], [2], [3], [8],
and [11], the difference quotient has a limit at 3 for all three functions, but it
is only in [8] that all three functions have derivatives at 3; by the definitions in
[1], [3], and [11], only S and H have derivatives at 3, and in [2], only S has a
derivative at 3. By [9], the difference quotient has a limit at 3 for S and H (but
not β) while only S has a derivative at 3, and by the definitions in [18], neither
the limit of the difference quotient nor the derivative exists for H and β, while
both exist for S.
In the author’s view, since both the limit of the difference quotient and the
derivative make sense conceptually for all three functions, the definitions of
limit and derivative in analysis should allow both of these to exist for all three
functions, as in [8]. Accordingly, in defining a derivative of f at p in an analysis
course, my preference is to place the least restrictive condition on the domain
(that p be an element of the domain and a cluster point of the domain) so that a
derivative may be spoken of in all cases where it is meaningful and interesting.
3. CONTINUITY
Solution: f̂ and h are continuous at 3, but f and g̃ are not since f is not defined
at 3 and g̃ does not have a limit at 3.
(b) Which of the functions in (a) is continuous at 3 by Definition C1, using
Definition L1 of a limit, from calculus?
Solution: Only f̂ is continuous at 3 by these definitions; h is not.
The students may now check what their calculus books say about continuity
at endpoints of a domain. According to the book’s definitions, is h continuous
at every point of its domain? How does this outcome compare with the con-
clusions in (a) and (b)? Students will find that the definitions in calculus yield
answers to these questions that differ from the answers in analysis. A careful
study of the differences, possibly considering other functions as well, makes a
good topic for a mathematical essay, a team project, or student presentations.
624 Shipman
3. On checking this against Definition C3, students then realize that α is con-
tinuous at 3 by this definition; some have used the word “vacuous” to describe
the situation.
A philosophical question about definitions now arises. In making a defini-
tion, about the behavior of functions, for example, should one allow functions
to satisfy the definition that may be described as “vacuous,” “irrelevant,” or
“extraneous” to the concept being defined? While both choices in answering
this preserve the meaning of the concept in all cases where it is interesting,
the convention among mathematicians is to write definitions to include the
extraneous cases as well.
Part II illustrates the difference in perspective between the two choices.
The questions in (a) and (b) are directed to the students, allowing them to con-
sider which perspective they might prefer, before explaining which choice is
the convention.
Part II: Should we allow extraneous cases?
1. These ants do not have wings. They will not add anything interesting to the
study so we will exclude them.
2. Allowing the ants into the study does not cause any trouble. It is not
necessary to exclude them so we will let them in.
Solution: The project will have the same outcome in both cases on the insects
for which it is relevant. The point here is to recognize the difference and to
think about which choice one might make and why.
(b) In defining continuity of a function at a point, suppose someone wants to
allow functions to satisfy the definition for which the meaning of continuity
is irrelevant. Which choice would you make?
Solution: Only α fails for the first reason. g, g̃, J, J̃, I, and k fail for the second
reason, and f, T and γ fail for the third reason. f̃ fails for the fourth reason,
and h, T̃, and β are continuous at 3.
(c) Is the language of Definition C2 as helpful as C1 in distinguishing why a
function may fail to be continuous at a point?
Solution: C1 lists the three criteria for continuity explicitly, making it easy to
determine distinct reasons for why continuity at p may fail. The language of
C2 is not as helpful here.
Solution: Only the last two. If f has a limit at p, then (re)- defining f at p to be
equal to this limit will make the function continuous at p. If f does not have a
limit at p, then changing f only at p cannot create a limit there since the status
of f at p has nothing to do with a limit at p. In this case, one would need to
change f at infinitely many points to make the repair.
628 Shipman
Solution: At any cluster point of the domain, the first condition is equivalent
to the third and the second is equivalent to the fourth. The first situation is a
minor problem that is easy to fix; we call this a removable discontinuity. The
second situation is a serious breach of continuity; we call this type essential.
This is summarized in Definitions D1 and D2.
Definition D1: Let p be a cluster point of D, and suppose f : D → R is not
continuous at p. To say that f has a removable discontinuity at p means that
the function can be made continuous at p by changing it only at p. Otherwise,
f is said to have an essential discontinuity at p.
Definition D2: Let p be a cluster point of D, and suppose f : D → R is not
continuous at p. f is said to have a removable discontinuity at p in the case that
f has a limit at p. f is said to have an essential discontinuity at p in the case
that f does not have a limit at p.
(c) Among the functions in Part I(a), which have removable discontinuities at
3? Which have essential discontinuities at 3?
Solution: f , f̃ , T, and γ have removable discontinuities at 3, and g, g̃, J, J̃, I,
and k have essential discontinuities at 3. For α, these terms are not defined
since 3 is not a cluster point of the domain.
Solution: All of these functions are continuous (no matter which of Definitions
C1, C2, and C3 is used for continuity at a point).
Before carefully considering definitions on continuity, students in a first
undergraduate course in analysis have almost unanimously believed that none
of these functions, except for ln, and possibly h, is continuous. The next three
questions investigate this further.
(c) Does any of the eight continuous functions in (b) have a removable or
essential discontinuity by Definition D2?
Solution: f and T have removable discontinuities at 3; g, J, and k have essen-
tial discontinuities at 3; ln has an essential discontinuity at zero, and γ has a
removable discontinuity at every rational number.
(d) Consider α : N → R where α(x) = −x. By Definitions C and C3, is α
continuous?
Solution: Yes; by these definitions, α is continuous.
630 Shipman
(e) Is there anything strange about the conclusions that you have made in this
activity? Is there anything you would change in the definitions?
Solution: It may seem strange that continuous functions can have discon-
tinuities and that graphs of continuous functions can have jumps, holes,
asymptotes, or isolated points. These conclusions, however, are correct and
sound by the well-crafted definitions of analysis.
4. OUTLOOK
In view of the conflicts that arise among definitions on limit and continuity
as treated in calculus courses and in analysis, one may wonder why textbooks
on calculus consistently require that the domain of a function contain an open
interval around a point in defining limit and continuity there. The standard
approach is that since the concept of a cluster point is reasonably considered
too complex for the general student of calculus, one may start with functions
defined on intervals and then, in future courses, extend the ideas to functions
on more general domains. This is a good plan, but it should be handled in a way
that avoids images and outcomes that conflict starkly with those of analysis.
The definitions in Section 4.1 resolve the major problems highlighted in the
Introduction. The belief that a function should have a limit at a point only if it
has equal left-hand and right-hand limits there (which is so difficult to amend
when students enter a course in analysis) is no longer true in this framework.
With the definitions of Section 4.1, students will see in calculus that familiar
√ √
functions such as x and 4 − x2 have limits at every point of their domains,
consistent with analysis.
Furthermore, continuity of a function at a point is now defined by the
same
√ simple requirement used in analysis. In this framework, functions such
as 4 − x2 are continuous, not by three different criteria to judge points at dif-
ferent locations in the domain (as in traditional calculus textbooks), but for the
same reason that they are continuous in analysis.
With definitions L4, C4, and C, students will also see, in their calculus
courses, the variety of graphs that continuous functions can have, which may
include breaks, jumps, and other discontinuities at points not in the domain.
Within this broader picture of continuity, one may then discuss images and
consequences of continuity for functions restricted to intervals. Students will
632 Shipman
ACKNOWLEDGEMENT
REFERENCES
BIOGRAPHICAL SKETCH