You are on page 1of 20

Chapter 2

Continuity
2.1 Definitions
The definition of continuity that appears in freshman calculus textbooks is typ-
ically a pointwise definition. Two alternatives appear in common use.
2.1.1 Definition (-δ Continuity)
A function f is said to be continuous at a point a of its domain if for every > 0
there is a δ > 0 such that |f[x] −f[a]| < whenever |x −a| < δ.
2.1.2 Definition (Limit Continuity)
A function f is said to be continuous at a point a if
1. f[a] is defined;
2. lim
x→a
f[x] exists; and
3. lim
x→a
f[x] = f[a].
Mathematically speaking, it is a matter of taste which definition one chooses,
because the two are equivalent. Indeed, if f is continuous at a in the sense of
Definition 2.1.1, then a lies in the domain of f, so that condition 1 of Definition
2.1.2 is satisfied. Moreover, from Definition 2.1.1 and the definition of limit,
we conclude that lim
x→a
f[x] exists and equals f[a]. On the other hand, if f
is continuous in the sense of Definition 2.1.2, then from the condition that f[a]
is defined we conclude that a lies in the domain of f. From the condition that
lim
x→a
f[x] = f[a], we conclude that no matter what > 0 may be given, we
can find δ > 0 such that |f[x] − f[a]| < whenever 0 < |x − a| < δ. Noting,
27
28 CHAPTER 2. CONTINUITY
that the limit is actually f[a], we may drop the first of the inequalities in the
condition 0 < |x −a| < δ to obtain the condition of Definition 2.1.1.
Pedagogically speaking, the second of the definitions above is probably prefer-
able in a freshman class. The ’s and δ’s come hard to freshmen, and the fewer
of them, the better. Moreover, Definition 2.1.2 is close in spirit to the reason for
introducing continuity in the first place: Continuous functions are those that
behave well in relation to limits.
2.1.3 Theorem (Continuity and Limits)
Suppose that lim
x→a
g[x] = L and that f is a function continuous at L. Then
lim
x→a
f[g[x]] = f
_
lim
x→a
g[x]
_
. (2.1)
Proof: The function f is given continuous at L, so the quantity on the right-
hand side of (2.1), which is f[L], is defined. We must show that lim
x→a
f[g[x]] =
f[L]. To this end, let > 0 be given. We must find δ > 0 with the property
that 0 < |x − a| < δ guarantees |f[g[x]] − f[L]| < . But by the continuity of
f, we know that, corresponding to our given > 0, there is an η > 0 such that
0 < |u − L| < η implies that |f[u] − f[L]| < . And we have also been given
that lim
x→a
g[x] = L, so we can find δ > 0 such that |g[x] − L| < η whenever
0 < |x − a| < δ. But then it follows that for 0 < |x − a| < δ we must have
|f[g[x]] −f[L]| < , and so lim
x→a
f[g[x]] = f[L].•
Thus, continuous functions are those which commute with the limit operation.
And compositions of continuous functions are continuous.
2.1.4 Corollary
If g is a function continuous at a, with g[a] = b, and if f is a function continuous
at b, then the composition f ◦ g is continuous at a.•
2.2 Continuity on Intervals and Other Sets
2.2.1 Definition (Continuity on Open Intervals)
If a and b are real numbers with a < b, and if f is a function, we say that f is
continuous on (a, b) when f is continuous at each point of (a, b).
Thus, a function is continuous on an open interval if it is continuous at each
point of the interval. There is a subtle point here that will arise again in the
2.3. THE CONTINUITY THEOREMS 29
future. If f is continuous on (a, b) and if x
1
and x
2
are points of (a, b), then f
must be continuous at x
1
and f must be continuous at x
2
. Accordingly, if > 0
be given, there must be δ
1
> 0 such that |f[x] −f[x
1
]| < whenever |x−x
1
| <
and there must be δ
2
> 0 such that |f[x] − f[x
2
]| < whenever |x − x
2
| < δ
2
.
But it need not be the case that |f[x] −f[x
2
]| < when |x −x
2
| < δ
1
unless it
happens that δ
1
< δ
2
, and it need not be the case that |f[x] −f[x
1
]| < when
|x −x
1
| ≤ δ
2
unless it happens that δ
2
≤ δ
1
. We will return to these matters.
Another problem arises when we want to talk about the continuity of a function
on an interval that contains one or both of its endpoints. Typically, we require
“one-sided” continuity at such endpoints:
2.2.2 Definition (One-Sided Continuity)
The function f is continuous from the right at a if
1. f[a] exists;
2. lim
x→a
+
f[x] exists; and
3. lim
x→a
+
f[x] = f[a].
f is continuous from the left at a if the corresponding statements are true with
x →a
+
replaced by x →a

.
2.2.3 Definition (Continuity on a Closed Interval)
The function f is continuous on [a, b] if f is continuous on (a, b), f is continuous
from right at a, and f is continuous from the left at b.
It is often desirable to talk about continuity of functions on sets that are not
intervals. Here is the definition, which subsumes all of the previous definitions.
2.2.4 Definition (Continuity on a Set)
Let D be a subset of the domain of a function f. f is said to be continuous on
D if for every d ∈ D and for every > 0 is is possible to find δ > 0 such that
|f[x] −f[d]| < whenever x ∈ D is such that |x −d| < δ.
2.3 The Continuity Theorems
2.3.1 Definition (Algebra of Functions)
If f and g are functions having a common domain and p, q = 0 are integers,
then the functions f +g, f −g, f · g, f
p/q
and f/g are, respectively, the functions
30 CHAPTER 2. CONTINUITY
defined by the equations
(f +g)[x] = f[x] +g[x]; (2.2)
(f −g)[x] = f[x] −g[x]; (2.3)
(f · g)[x] = f[x] · g[x]; (2.4)
(f
p/q
)[x] = (f[x])
p/q
; (2.5)
(f/g)[x] =
f[x]
g[x]
. (2.6)
It is common to suppress the “·” in f · g; we commonly write (fg)[x] in place of
(f · g)[x].
2.3.2 Theorem (Algebra of Continuity)
Let f and g be functions continuous on an interval (a, b). Then the functions
f + g, f − g, and f · g are all continuous on the interval (a, b). If 0 < f[x] for
all x ∈ (a, b), then the function f
p/q
is continuous on (a, b). The function f/g
is continuous at those points x of (a, b) for which g[x] does not vanish.
Proof: Immediate from the definitions and the Addition, Multiplication, Con-
stant, Extraction, and Division Limit Theorems of the previous chapter.•
2.3.3 Theorem (Continuity of Rational Functions)
The functions x → c (where c is a fixed real number) and x → x are both
continuous on R. In consequence, every polynomial function is continuous on R
and every rational function is continuous on its domain.
Proof: Continuity of constant functions is immediate from the Constant Limit
Theorem, and continuity of the function x →x is immediate from the Identity
Function Limit Theorem. The rest of the theorem now follows from Theorem
2.3.2.•
2.3.4 Discontinuities
Not many authors take the effort to give a formal definition for the term discon-
tinuity; the term is used informally. Generally speaking, we would like to say
that a function f has a discontinuity at any point a where it is not continuous,
but we should probably be more careful. Following the usual convention that
the domain of a function is the largest set of real numbers for which the defi-
nition of the function is meaningful, let us put f[x] =

x
2
−1. The domain of
f is then the set (−∞, −1] ∪ [1, ∞). Most authors would hesitate to say that f
had a discontinuity at the origin—though f is certainly not continuous at that
point. On the other hand it is not uncommon to find authors writing of the
2.4. SEQUENCES & CONTINUITY 31
discontinuity of the function x →x
−1
at the origin.
One kind of discontinuity frequently does receive a formal definition. If f is a
function that fails of continuity at a point a because it violates condition 1 or
condition 3 (but not condition 2) of Definition 2.2.2, then f is said to have a
removable discontinuity (or a removable singularity) at a. As the term suggests,
a function f which has a removable discontinuity at a possesses an extension F
that agrees with f away from a but is continuous at a. One defines F by
F[x] =
_
_
_
f[x]; x = a;
lim
x→a
f[x]; x = a.
(2.7)
Alternatively, we often use the existence of such an extension in order to evaluate
a limit. This is, in fact, another point of view of the Equality Theorem (Corollary
1.5.9).
2.4 Sequences & Continuity
2.4.1 Definition (Sequence; Limit of a Sequence)
A sequence is a function x : N → R. When x is a sequence, we usually write
x
n
instead of x[n] for the value of the function x and the natural number n.
We usually write {x
n
}

n=1
or { x
n
: n ∈ N} for the sequence x. If L is a
real number, we write lim
n→∞
x
n
= L if, whatever > 0 may be given, it is
always possible to find N ∈ N such that |x
n
−L| < whenever n ∈ N and n ≥ N.
It is not at all difficult to prove analogs of the theorems of §1.5 for limits of
sequences. We leave formulation of the statements, as well as proofs, as exercises
for the reader.
2.4.2 Theorem (Sequential Continuity)
Let L ∈ R, and let f be a function whose domain contains L. Then f is con-
tinuous at L if and only if lim
k→∞
f[x
k
] = f[L] for every sequence {x
k
}

k=1
for
which lim
k→∞
x
k
= L.
Proof: Suppose that f is continuous at L, and let {x
n
}

n=1
be any sequence with
lim
n→∞
x
n
= L. Let > 0 be given. Choose δ > 0 such that |f[x] − f[L]| <
whenever |x −L| < δ. Then, because lim
n→∞
x
n
= L, we can find N ∈ N such
that |x
n
− L| < δ whenever n ∈ N and n ≥ N. But then |f[x
n
] − f[L]| <
whenever n ∈ N and n ≥ N, and this means that lim
n→∞
f[x
n
] = f[L].
If, on the other hand, f is not continuous at L, then for some > 0 it is
possible to find x
n
, for each n ∈ N, so that |x
n
− L| < 1/n on the one hand,
32 CHAPTER 2. CONTINUITY
but |f[x
n
] − f[L]| ≥ on the other. It is clear that lim
n→∞
x
n
= L but
lim
n→∞
f[x
n
] = f[L].•
2.5 Some Topological Considerations
In our discussion of continuous, real-valued functions of a real variable, it will be
helpful to adopt a somewhat more advanced point of view than we ask freshmen
to adopt. To this end, we introduce some topological terminology and derive
some facts that lie beyond the scope of typical freshman calculus courses.
2.5.1 Definition (Open Set)
A subset U of the real numbers is said to be open if every point u ∈ U is the
center of an interval (u − δ, u + δ) (where, of course, δ > 0) which is itself a
subset of U. That is, a subset U of the real numbers is open if for every u ∈ U
there is a δ > 0 such that (u −δ, u + δ) ⊆ U.
An open interval, as we have been using the term, is a special case of an open
subset of the real numbers. It can be shown (though we shall not do so here
because we have no use for the fact in what follows) that a subset of the real
numbers is open if, and only if, it is a union of countably many pairwise disjoint
open intervals.
Note further that the set of all real numbers is an open set, as is the empty set.
(The latter is vacuously empty: Every member of the empty set is the center of
an interval contained in the empty set—because there are no such members.)
2.5.2 Definition (Upper Bound; Least Upper Bound)
Let S ⊆ R. A real number m is said to be an upper bound for S if s ≤ m for
every s ∈ S. A real number m is said to be a least upper bound (or a supremum)
for S if m is an upper bound for S and no upper bound for S is smaller than
m.
We leave it to the reader to formulate analogous definitions for the terms lower
bound and greatest lower bound (alternatively, infimum).
2.5.3 Definition (Open Cover; Sub-Cover)
Let S be a set of real numbers. An open cover of S is any family U = { U
α
: α ∈
A} of open sets whose union contains S. If U is an open cover of S and V ⊆ U
such that the union of the sets in V also contains S, then V is called a sub-cover
for U. If V is a sub-cover for an open cover U and if V consists of finitely many
of the open sets from U, then V is called a finite sub-cover for U.
2.5. SOME TOPOLOGICAL CONSIDERATIONS 33
2.5.4 Theorem (Completeness Conditions)
If the set R of real numbers has any one of the following properties, then it has
all of them:
1. The Least Upper Bound Property: Every non-empty set of real num-
bers that has an upper bound also has a least upper bound.
2. The Heine-Borel Property: If a, b ∈ R with a < b, then every open
cover of [a, b] admits a finite sub-cover.
3. The Nested Interval Property: If a family of intervals [a
n
, b
n
], n ∈ N,
is such that (1) lim
n→∞
(b
n
−a
n
) = 0, and (2) for every n ∈ N, [a
n
, b
n
] ⊇
[a
n+1
, b
n+1
], then there is exactly one real number common to all of the
intervals [a
n
, b
n
], n ∈ N.
Proof: Suppose first that R has the Least Upper Bound Property. Suppose
also that we are given an interval [a, b] and an open cover U = { U
α
: α ∈ A}
of [a, b]. Let us call a point x of the interval [a, b] accessible if there are finitely
many indices α
1
, . . . , α
n
∈ A such that [a, x] ⊆

n
k=1
U
α
k
. Let S denote the set
of all accessible points of [a, b]. It is clear that a is accessible, so S is nonempty.
If b is accessible, then we are done; if b is not accessible then b is an upper
bound for S and—because we have assumed that R has the Least Upper Bound
Property—we may conclude that S has a least upper bound M. Clearly M ≤ b.
From the facts that M ∈ [a, b] and that U is an open cover of [a, b], we conclude
that there is an α
0
∈ A such that M ∈ U
α0
. Because U
α0
is open, there is a
δ > 0 such that (M − δ, M + δ) ⊆ U
α0
. But M is the least upper bound of
the set of accessible points in [a, b], so the interval (M − δ, M] must contain
an accessible point x
0
. We let U
α1
, U
α2
, . . . , U
αn
be members of U such that
[a, x
0
] ⊆

n
k=1
U
α
k
, and then we observe that [a, M] ⊆

n
k=0
U
α
k
. Thus, M is
accessible. But then the points of (M, M + δ) are also accessible (because they
all lie in U
α0
), and this contradicts our choice of M as the least upper bound of
the set of accessible points in [a, b] unless M = b. This establishes that if R has
the Least Upper Bound Property then it also has the Heine-Borel Property.
Suppose now that we are given a family of intervals [a
n
, b
n
], n ∈ N, such that
lim
n→∞
(b
n
−a
n
) = 0 and for every n ∈ N, [a
n
, b
n
] ⊇ [a
n+1
, b
n+1
]. Suppose also
that


k=1
[a
k
, b
k
] = Ø. For each n ∈ N, let U
n
= (−∞, a
n
) ∪ (b
n
, ∞). From the
fact that


k=1
[a
k
, b
k
] = Ø, we see that U = { U
n
: n ∈ N} is an open cover for
[a
1
, b
1
]. (If it were not, then some x
0
∈ [a
1
, b
1
] would not lie in any of the sets
U
n
, n ∈ N; such a point would necessarily lie in every one of the intervals [a
n
, b
n
],
n ∈ N—and we have assumed that there is no such point.) But it is clear that U
admits no finite subcover, because the nesting condition [a
n
, b
n
] ⊇ [a
n+1
, b
n+1
]
guarantees that no finite sub-family of U can cover [a
1
, b
1
]. If we suppose that
R has the Heine-Borel Property, then it must follow that


k=1
[a
k
, b
k
] = Ø is
impossible. From the condition lim
n→∞
(b
n
− a
n
) = 0, we see that it is not
possible for the intersection to contain more than one point. (If there were two
34 CHAPTER 2. CONTINUITY
points x
1
< x
2
in the intersection, we would have to have a
n
≤ x
1
< x
2
≤ b
n
for
all n, and this would make lim
n→∞
(b
n
−a
n
) = 0 impossible.) This shows that
if R has the Heine-Borel Property then it also has the Nested Interval Property.
Now let S be a non-empty set of real numbers, and suppose that m is an upper
bound for S. Choose a
1
∈ S and put b
1
= m. If a
1
is a least upper bound for S,
then we have shown that S has a least upper bound, so suppose that a
1
is not an
upper bound for S. Let c
1
=
1
2
(a
1
+b
1
). There are two possibilities: Either c
1
is
an upper bound for S or it is not. In the first case we put a
2
= a
1
and b
2
= c
1
.
In the second case, we put a
2
= c
1
, b
2
= b
1
. In either case, we have obtained
an interval [a
2
, b
2
] whose left-hand end-point is not an upper bound for S and
whose right-hand end-point is an upper bound for S. We can continue this con-
struction inductively to obtain a sequence of intervals { [a
k
, b
k
] : k ∈ N} such
that [a
n
, b
n
] ⊇ [a
n+1
, b
n+1
] for every n ∈ N and (b
n
−a
n
) = 2
(1−n)
(b
1
−a
1
) →0
as n →∞. If R has the Nested Interval Property, there is then a single number
M that lies in every one of the intervals [a
n
, b
n
].
Suppose that there is an s ∈ S such that M < s. Then there is an n
0
∈ N such
that 2
(1−n0)
(b
1
−a
1
)+M < s. But M lies in the interval [a
n0
, b
n0
], and the length
of this interval is 2
(1−n0)
(b
1
−a
1
). This means that b
n0
< s, which violates the
fact that b
n0
is an upper bound for S. Consequently, M is an upper bound for S.
Suppose now that T is any number such that T < M. Then there is n
1
∈ N
such that T < M − 2
(1−n1)
(b
1
− a
1
), which means that T < a
n1
, by reasoning
similar to that of the preceding paragraph. But a
n1
is not an upper bound for
S, and this means that T cannot be either. Hence no upper bound for S can be
smaller than M, and M must be a least upper bound for S. Thus, we have seen
that if R has the Nested Interval Property it must also have the Least Upper
Bound Property.•
The method we have used in the last part of this proof, where we showed that
the Nested Interval Property implies the Least Upper Bound Property is called
“the Method of Bisection,” and we will see it again.
Note that we have not proved that R has any one of the three conditions listed in
Theorem 2.5.4; we have only proved that if R has any one of them, then it has all
of them. If our purpose here were to construct the set of real numbers, we would
be careful to build one of these properties into the construction. However, this
is not the place for such a construction and we will adopt a different strategy.
2.5.5 Axiom (Completeness of the Reals)
Axiom: The real numbers have the Least Upper Bound Property.
2.6. MORE ON CONTINUITY 35
2.6 More on Continuity
2.6.1 Definition (Inverse Image)
Let f be a function carrying some set of real numbers into the real numbers,
and suppose that A is collection of real numbers. By the inverse image of A
under f (written f
−1
[A]), we mean the set { x : f[x] ∈ A}.
We have introduced these terms because continuity has a very simple description
using them: A function is continuous if and only if inverse images of open sets
are open sets:
2.6.2 Theorem (Continuity Characterization)
Let f be a function defined on a subset D of R. Then f is continuous on D iff
for every open subset U of R, there is another open subset V of R such that
f
−1
[U] = D ∩ V .
Proof: First, suppose that f is continuous on its domain D, and let U be an
open subset of R, the real numbers. If x
0
∈ f
−1
[U], then f[x
0
] ∈ U. Con-
sequently, because U is open, there is an > 0 such that the open interval
(f[x
0
] −, f[x
0
] +) ⊆ U. But f is continuous at x
0
, so there is a corresponding
δ
x0
> 0 such that |f[x] −f[x
0
]| < whenever x ∈ D and |x−x
0
| < δ
x0
. Putting
I
x0
= { x : |x−x
0
| < δ
x0
}, we let V =

x∈f
−1
[U]
I
x
. Then f
−1
[U] = D∩V and
V is open, for if v ∈ V , then v ∈ I
x
⊆ V for some x ∈ f
−1
[U].
Now let us suppose that whenever U ⊆ R is open, it is possible to find an open
subset V of R such that f
−1
[U] = D ∩ V . Let x
0
∈ D, and suppose that > 0
has been given. We note that U = { y : |y − f[x
0
]| < } is an open subset of
R, and so there is an open subset V of R such that f
−1
[U] = D ∩ V . Because
x
0
∈ f
−1
[U] ⊆ V , we know that x
0
∈ V . Consequently, there is a positive δ such
that the open interval (x
0
−δ, x
0
+ δ) ⊆ V . But this means that if |x −x
0
| < δ
and x ∈ D, then x ∈ f
−1
[U]—or, in other words, f[x] ∈ U. The latter is
equivalent to |f[x] −f[x
0
]| < , so we have now seen that whatever x
0
∈ D we
begin with, and whatever > 0 may be given, it is always possible to find δ > 0
such that whenever x ∈ D and |x − x
0
| < δ we will have |f[x] − f[x
0
]| < .
Consequently f is continuous on D.•
We are now ready to give proofs of a number of theorems whose statements,
but not whose proofs, are standard parts of freshman calculus courses. We
begin with the boundedness of a continuous function whose domain is a closed
bounded interval.
36 CHAPTER 2. CONTINUITY
2.6.3 Theorem (Boundedness)
Let f be a continuous function defined on a closed, bounded interval [a, b]. There
is a real number M such that |f[x]| < M for all x ∈ [a, b].
Proof: For each n ∈ N, the set V
n
= (−n, n) is an open set of real numbers,
and so by Theorem 2.6.2 there is, for each n ∈ N, an open set U
n
such that
f
−1
[V
n
] = [a, b] ∩U
n
. Now


k=1
V
n
= R, and therefore [a, b] ⊆ f
−1
[


k=1
V
k
] =


k=1
f
−1
[V
k
] ⊆


k=1
U
k
. The Heine-Borel Property assures us that there are
finitely many n
1
< n
2
< · · · < n
m
∈ N such that [a, b] ⊆ U
n1
∪ U
n2
∪ · · · ∪ U
nm
.
But then if x ∈ [a, b] we must have f[x] ∈ V
n
k
for some k ∈ {1, 2, . . . , m}.
Because it is automatic that V
n
k
⊆ V
nm
it follows that f[x] ∈ V
nm
, which is the
same as −n
m
< f[x] < n
m
. Putting M = n
m
, we are done.•
2.6.4 Theorem: (Extreme Value Theorem)
Let f be a continuous function defined on a closed, bounded interval [a, b]. There
are points η, ξ ∈ [a, b] such that f[η] ≤ f[x] ≤ f[ξ] for every x ∈ [a, b].
Proof: By Theorem 2.6.3, the non-empty set S = { f[x] : x ∈ [a, b] } of real
numbers is bounded above; the Least Upper Bound Property assures us that S
has a least upper bound, α. If f[x] = α for every x ∈ [a, b], then the function F
given by
F[x] =
1
α −f[x]
(2.8)
is continuous on [a, b]. Applying Theorem 2.6.3 once again, we find M > 0 such
that |F[x]| < M for all x ∈ [a, b].
Take = 1/M. Then > 0, and because α is a least upper bound for S, there
must be an element s of S in the interval (α−, α). But if s ∈ S, then s = f[x
0
]
for some x
0
∈ [a, b]. For this x
0
, we have α − < f[x
0
] < α, and from this it
follows that α −f[x
0
] < . This means that M = 1/ < 1/(α −f[x
0
]) = F[x
0
],
which is a contradiction. This establishes the existence of ξ ∈ [a, b] such that
f[x] ≤ f[ξ] for all x ∈ [a, b].
The existence of η ∈ [a, b] such that f[η] ≤ f[x] for all x ∈ [a, b] now follows by
applying what we have already established to the function g[x] = −f[x].•
2.6.5 Theorem (Local Invariance of Sign)
If f is continuous on some open interval that contains the number ξ and if
f[ξ] > 0, (respectively, f[ξ] < 0) then there is a δ > 0 such that f[x] > 0 (resp.,
f[x] < 0) for all x ∈ (ξ −δ, ξ + δ).
2.6. MORE ON CONTINUITY 37
Proof: Take =
1
2
f[ξ] > 0, and find δ > 0 so that |f[x] − f[ξ]| < whenever
|x − ξ| < δ. Then for x ∈ (ξ − δ, ξ + δ) we have − < f[x] − f[ξ] < , which
implies that 0 <
1
2
f[ξ] = f[ξ] − < f[x].
The proof is the same for f[ξ] < 0.•
2.6.6 Theorem (Intermediate Value Theorem)
Let f be a continuous function defined on a closed, bounded interval [a, b]. If
f[a] > 0 and f[b] < 0, then there is a number ξ ∈ (a, b) such that f[ξ] = 0.
Proof: We use the Method of Bisection. Let a
1
= a, b
1
= b, and let c
1
=
1
2
(a
1
+b
1
). If f[c
1
] = 0 we put ξ = c
1
and we are through. Otherwise, there are
two cases: f[c
1
] > 0 or f[c
1
] < 0. In the first case, we put a
2
= c
1
and b
2
= b
1
; in
the second, a
2
= a
1
and b
2
= c
1
. The interval [a
2
, b
2
] then meets the conditions
that f[a
2
] > 0 and f[b
2
] < 0 and is half the length of the interval [a
1
, b
1
]. It is
clear how to proceed inductively to obtain either ξ ∈ [a, b] for which f[ξ] = 0 or
a sequence of closed intervals [a
1
, b
1
] ⊇ [a
2
, b
2
] ⊇ · · · ⊇ [a
k
, b
k
] ⊇ · · · all meeting
the conditions f[a
k
] > 0 and f[b
k
] < 0, as well as those of the Nested Interval
Property.
In the latter case, we obtain a unique number ξ common to all of the intervals
[a
k
, b
k
]. Suppose that f[ξ] > 0. By Local Invariance of Sign (Theorem 2.6.5),
there is a δ > 0 such that f[x] > 0 whenever x ∈ [a, b] and |ξ − x| < δ.
By our construction of the intervals [a
k
, b
k
], we can find N ∈ N so large that
b
N
− a
N
< δ. Now ξ ∈ [a
N
, b
N
], so |ξ − b
N
| < δ. Hence f[b
N
] > 0. But,
by construction, f[b
N
] < 0, and the contradiction shows that f[ξ] > 0 is not
possible.
Suppose, on the other hand, that f[ξ] < 0. Again by Local Invariance of Sign,
there is a δ > 0 such that f[x] < 0 for all x ∈ [a, b] such that |ξ − x| < δ.
Once again, we can find N ∈ N such that b
N
− a
N
< δ. Then ξ ∈ [a
N
, b
N
],
so |ξ − a
N
| < δ. Hence f[a
N
] < 0. But, by construction, f[a
N
] > 0, and the
contradiction shows that f[ξ] < 0 is not possible.
Having found that both f[ξ] < 0 and f[ξ] > 0 are impossible, we conclude that
f[ξ] = 0.•
2.6.7 Corollary
Let f be continuous on an interval (which may be open or closed or neither)
which contains the points α and β. If η is any real number lying between f[α]
and f[β], then there is a number γ lying between α and β and such that f[γ] = η.
Proof: We may assume that α < β (if not, we simply interchange the names)
and that f[α] > f[β] (if not, we work with −f). All that remains is to apply
38 CHAPTER 2. CONTINUITY
the Theorem to the function F[x] = f[x] −η.•
2.7 Uniform Continuity
Uniform continuity is rarely presented in freshman calculus courses. It is not at
all clear that this is wise. The difference between pointwise continuity (which is
what we presented in the first section of this chapter) and uniform continuity is
subtle, and it is clear that we don’t need to talk about the difference between the
two in the presence of freshmen. But experience suggests that freshmen initially
interpret pointwise continuity on an interval as uniform continuity. The mistake
seems to be natural; Cauchy himself fell into the trap of confusing pointwise
concepts with the uniform ones.
2.7.1 Definition (Uniform Continuity)
A function f defined on a set D of real numbers is said to be uniformly contin-
uous on D if for every > 0 there is a corresponding δ > 0 such that whenever
u ∈ D and v ∈ D are such that |v −u| < δ it follows that |f[v] −f[u]| < .
In order to understand the difference between pointwise continuity on a set D
and uniform continuity on D, if is helpful to render the definitions formally.
2.7.2 Pointwise Continuity on D:
The statement f is continuous on D means
(∀u ∈ D)(∀ > 0)(∃δ > 0)(∀v ∈ D)[(|v −u| < δ) ⇒ (|f[v] −f[u]| < )]. (2.9)
2.7.3 Uniform Continuity on D:
The statement f is uniformly continuous on D means
(∀ > 0)(∃δ > 0)(∀u ∈ D)(∀v ∈ D)[(|v −u| < δ) ⇒ (|f[v] −f[u]| < )]. (2.10)
Compare (2.9) and (2.10) with a view toward proving that a function f has one
of the properties. In the first instance we must first select a point u ∈ D and an
> 0. Then we must find a δ > 0—which may depend upon both u and —in
such a way that every v ∈ D with a certain property satisfies a certain inequal-
ity. Note the emphasis above, which we repeat here: In the case of pointwise
continuity δ may depend upon both u and .
In the second instance, however, we choose , and only, first of all. Then we
must be able to find δ > 0 in such a way that every pair of numbers u, v that
are within δ of each other satisfy a certain inequality. In the case of uniform
2.7. UNIFORM CONTINUITY 39
continuity, δ may depend upon only, and not upon both of the numbers u and .
Thus, to demonstrate pointwise continuity on D, we may choose different δ’s
in different parts on D, even though has not changed. But to demonstrate
uniform continuity of D, once has been specified, we must find a single δ and
show that it works everywhere in D. We therefore expect that it will be harder
to show that a function is uniformly continuous than to show that it is pointwise
continuous. This is best seen through an example.
2.7.4 Example (x
2
)
Let f[x] = x
2
when x ∈ R. That f is continuous on R is immediate from what
we have already seen (see Theorem 2.3.3), but we will prove it directly. Then we
will show that f is not uniformly continuous on R. We hope that this will help
to illuminate the difference between pointwise continuity and uniform continuity.
Let u ∈ R, and let > 0 be given. Let us take
δ = min
_

1 + 2|u|
, 1
_
. (2.11)
With this choice of δ, if |u −v| < δ, we may write:
|f[v] −f[u]| = |v
2
−u
2
| (2.12)
= |v +u| · |v −u| (2.13)
= |(v −u) + 2u| · |v −u| (2.14)
≤ (|v −u| + 2|u|) · |v −u| (2.15)
< (1 + 2|u|) · |v −u|, (2.16)
where the last of these inequalities is correct because |v − u| < δ ≤ 1. But we
also have δ ≤ (1 + 2|u|)
−1
, and this means that the quantity on the right side
of (2.16) cannot exceed . This shows that f is continuous at u for each u ∈ R.
The reader should note how δ is chosen to depend upon u.
In order to establish that f is not uniformly continuous on R, we must show
that condition (2.10) fails, or, equivalently, that
(∃ > 0)(∀δ > 0)(∃u ∈ R)(∃v ∈ R)[(|u −v| < δ) ∧ (|f[u] −f[v]| ≥ )]. (2.17)
We will show that if = 1, then whatever δ > 0 may be, it is possible to find
u and v such that |u − v| < δ, but |u
2
− v
2
| > . For let δ > 0 be given. Take
u = δ
−1
, and take v = u +
1
2
δ. Then |v −u| =
1
2
δ < δ, but
|v
2
−u
2
| =
¸
¸
¸
¸
¸
_
1
δ
+
1
2
δ
_
2

_
1
δ
_
2
¸
¸
¸
¸
¸
(2.18)
40 CHAPTER 2. CONTINUITY
=
¸
¸
¸
¸
1
δ
2
+ 1 +
δ
2
4

1
δ
2
¸
¸
¸
¸
(2.19)
= 1 +
δ
2
4
, (2.20)
and this is larger than 1 = . Thus, f is not uniformly continuous on R.
Although the preceding argument shows that we will not be able to prove that
f is uniformly continuous on R, it is instructive to see where an attempt to con-
struct a proof fails. In order to give such a proof, we would have to let > 0 be
given. We would then like to find δ > 0 such that (|u−v| < δ) ⇒(|u
2
−v
2
| < ).
The conclusion here, |u
2
− v
2
| < , is equivalent to |u + v| · |u − v| < , and
this is where things go wrong. We cannot control the magnitude of the product
(u + v)(u − v) simply by controlling the magnitude of the difference (u − v)—
which we are allowed to do. We must also be able to control the magnitude
of the sum (u + v), and we cannot do this unless we can control either u or
v—which we are forbidden to do—as well.
It is not possible to find an example of a function that is uniformly continuous
on a set D but not pointwise continuous on D.
2.7.5 Theorem (Uniform Continuity Implies Continuity)
If f is a function uniformly continuous on a set D, then f is continuous at each
point of D.
Proof: Let d ∈ D, and let > 0 be given. Find δ > 0 so that |f[v] −f[u]| <
whenever u and v are points of D for which |v − u| < δ. Then it follows that
|f[v] − f[d]| < whenever v ∈ D is such that |v − d| < δ, and this means that
f is continuous at d.•
One of the reasons why we might want to consider replacing pointwise continu-
ity with uniform continuity in the elementary calculus sequence is that the bulk
of what we do in that sequence deals with functions that are (pointwise) contin-
uous on a closed interval of finite length. Under those circumstances, uniform
continuity is an automatic consequence of the assumed pointwise continuity.
2.7.6 Theorem (Continuity on Intervals [a, b])
If the function f is (pointwise) continuous on the interval [a, b], then f is also
uniformly continuous on [a, b].
Proof: Let > 0 be given. For each u ∈ [a, b], find δ
u
so that |f[v] −f[u]| <
1
2

whenever v ∈ [a, b] is such that |u −v| < δ
u
. (We know that we con accomplish
this because f is continuous at each point of [a, b].) By the Heine-Borel Property
of the real numbers, we can find u
1
, u
2
, . . . , u
n
, all in [a, b], such that [a, b] ⊆
2.7. UNIFORM CONTINUITY 41

n
k=1
(u
k
− δ
u
k
, u
k
+ δ
u
k
). Let ∆ be the set of all end-points of the intervals
(u
k
−δ
u
k
, u
k

u
k
) that lie in [a, b] together with the numbers a and b themselves.
Thus, sorting and renaming, we have ∆ = {x
0
, x
1
, . . . , x
m
}, with a = x
0
< x
1
<
· · · < x
m−1
< x
m
= b. We put
δ = min{ x
k
−x
k−1
: k = 1, 2, . . . , m}. (2.21)
If u and v are now points of [a, b] with |v − u| < δ, then at most one of the
numbers x
0
, x
1
, . . . , x
m
can lie in the closed interval whose endpoints are u and
v themselves. This means that u and v must both lie in some one of the intervals
(u
k0
−δ
u
k
0
, u
k0
+ δ
u
k
0
), and this guarantees in turn that
|f[v] −f[u]| = |f[v] −f[u
k0
] +f[u
k0
] −f[u]| (2.22)
≤ |f[v] −f[u
k0
]| + |f[u
k0
] −f[u]| (2.23)
<

2
+

2
= .• (2.24)
We end this section with a theorem that will be of use later on and a final
example. We need a definition first.
2.7.7 Definition (Dense Subset)
Let I be an interval, and let D ⊆ I. D is said to be dense in I if for every x ∈ I
and every δ > 0 the interval (x −δ, x + δ) always contains a point of D.
2.7.8 Theorem (Extension)
Suppose that D is a dense subset of an interval [a, b], and let f be a function
which is defined and uniformly continuous on D. Then there is a unique func-
tion F, defined on [a, b] and uniformly continuous there, such that F[x] = f[x]
for every x ∈ D.
Proof: Let us show first that there can be at most one such function. Suppose
that we have two functions, F and G, with common domain [a, b], and such
that F[x] = f[x] = G[x] for all x ∈ D, and that F and G are both uniformly
continuous on [a, b]. Let x
0
be any point of [a, b] that does not lie in D. Because
F and G are both uniformly continuous on [a, b], they are both continuous at
x
0
, and so lim
x→x0
F[x] = F[x
0
] and lim
x→x0
G[x] = G[x
0
].
Let > 0 be given, and choose δ > 0 so that |x − x
0
| < δ implies that both of
the inequalities |F[x] −F[x
0
]| < /2 and |G[x] −G[x
0
]| < /2 hold. Because D
is dense in [a, b], we can find x
1
∈ D such that |x
1
− x
0
| < δ. For this x
1
, we
have
|F[x
0
] −G[x
0
]| = |F[x
0
] −f[x
1
] +f[x
1
] −G[x
0
]| (2.25)
42 CHAPTER 2. CONTINUITY
= |F[x
0
] −F[x
1
] +G[x
1
] −G[x
0
]| (2.26)
≤ |F[x
0
] −F[x
1
]| + |G[x
1
] −G[x
0
]| (2.27)
<

2
+

2
= . (2.28)
But was an arbitrary positive number, and so it follows that F[x
0
] = G[x
0
],
from whence we conclude that F and G are the same function.
We must now show how to define F at points of [a, b] that do not lie in D. To
this end, we choose, once and for all, positive numbers η
n
, n ∈ N, such that
|f[v]−f[u]| < (2n)
−1
whenever u and v are elements of D for which |v−u| < η
n
.
We can accomplish this in such a way that η
n+1
< η
n
for each n ∈ N.
If x
0
∈ [a, b], but x
0
/ ∈ D, then for each n ∈ N, we put
S
x0,n
=
_
f[u] : u ∈ D and x
0

η
n
2
< u < x
0
+
η
n
2
_
. (2.29)
S
x0,n
is not empty because D is dense in [a, b]. In fact, let us choose u
x0,n

D ∩ (x
0
−η
n
/2, x
0
+ η
n
/2) arbitrarily. Suppose that y ∈ S
x0,n
. Then y = f[x]
for some x ∈ D∩(x
0
−η
n
/2, x
0

n
/2), and because |u
x0,n
−x| < η
n
, we must
have |f[u
x0,n
] − f[x]| < (2n)
−1
. Consequently, f[u
x0,n
] + (2n)
−1
is an upper
bound for the non-empty set S
x0,n
and f[u
x0,n
] − (2n)
−1
is a lower bound for
the same set. It follows that S
x0,n
has a greatest lower bound, a
x0,n
, and a least
upper bound b
x0,n
, and that b
x0,n
−a
x0,n
< n
−1
. Noting that for each n ∈ N we
have S
x0,n+1
⊆ S
x0,n
, we see that a
x0,n+1
≥ a
x0,n
and b
x0,n+1
≤ b
x0,n
for every
n ∈ N. Appealing to the Nested Interval Property, we conclude that there is
one and only one number ξ common to all of the intervals [a
x0,n
, b
x0,n
], n ∈ N.
We put F[x
0
] = ξ. If we put F[x] = f[x] whenever x ∈ D, this gives a function
F defined on all of [a, b].
Let us note that if > 0 is given, it is possible to find a δ > 0 such that whenever
x
0
lies in [a, b] and x ∈ D is such that |x
0
−x| < δ, then |F[x
0
] −f[x]| < . Be-
cause we know that we can do this when x
0
∈ D, we may suppose that x
0
/ ∈ D.
We need only choose n
0
∈ N so that n
−1
0
< and take δ =
1
2
η
n0
. Then F[x
0
] ∈
[a
x0,n0
, b
x0,n0
] and if |x
0
− x| < δ, we will have f[x] ∈ S
x0,n0
⊆ [a
x0,n0
, b
x0,n0
],
where b
x0,n0
−a
x0,n0
< n
−1
0
. From this it follows that |F[x
0
] −f[x]| < .
It remains to show that F is uniformly continuous on [a, b]. So let > 0 be
given. Choose δ
1
> 0 so that whenever x

and x

lie in D with |x

− x

| < δ
1
we must have |f[x

] − f[x

]| <
1
3
and whenever w ∈ [a, b] and x ∈ D are such
that |w −x| < δ
1
we must have |F[w] −f[x]| <
1
3
. Put δ =
1
3
δ
1
. Suppose that
u and v lie in [a, b] with |v − u| < δ. Find x

∈ D with |u − x

| < δ, so that
|F[u]−f[x

]| <
1
3
, and find x

∈ D with |v−x

| < δ, so that |F[v]−f[x

]| <
1
3
.
Then
|x

−x

| = |x

−v +v −u +u −x

| (2.30)
2.8. CONTINUITY & INVERSE FUNCTIONS 43
≤ |x

−v| + |v −u| + |u −x

| (2.31)
< δ + δ + δ = δ
1
, (2.32)
and, in consequence, |f[x

] −f[x

]| <
1
3
. Thus,
|F[v] −F[u]| = |F[v] −F[x

] +F[x

] −F[x

] +F[x

] −F[u]| (2.33)
≤ |F[v] −f[x

]| + |f[x

] −f[x

]| + |f[x

] −F[u]| (2.34)
<

3
+

3
+

3
= .• (2.35)
2.7.9 Example (sin x
−1
Again)
Let f be the function given by
f[x] = sin
1
x
, (2.36)
and let a = 0, b = 1, D = (0, 1). Then f is continuous on D (as we will show
later). However, lim
x→0
+ f[x] does not exist, as can be seen by the argument
given in Example 1.3.3. Consequently, there can be no function F, defined and
(uniformly) continuous on [0, 1], such that F[x] = f[x] for all x ∈ D.
This example shows us two things. First, the function f cannot be uniformly
continuous on D, for if it were, a function such as that described in the Extension
Theorem (Theorem 2.7.8) would exist. Consequently, the condition that the
interval in Theorem 2.7.6 be closed cannot be relaxed. Second, the uniform
continuity hypothesis in the Extension Theorem cannot be relaxed to pointwise
continuity.
2.8 Continuity & Inverse Functions
2.8.1 Definition (Monotonicity)
A function f defined on an interval I is said to be increasing on I if from
u, v ∈ I with u < v it follows that f[u] ≤ f[v]. f is said to be strictly increasing
on I if from u, v ∈ I with u < v it follows that f[u] < f[v]. We say that f
is decreasing on I when the conditions u, v ∈ I and u < v imply f[u] ≥ f[v].
If the same conditions imply that f[u] > f[v], we call f strictly decreasing on
I. We will use the term (strictly) monotonic of a function to mean that that
function is (strictly) increasing on I or that it is (strictly) decreasing on I.
There is considerable variation in terminology and definition here, and this is
something we will return to later in our discussion of differential criteria for
determining when a function has one of these properties. Some authors make
those criteria into the definitions. (We think that, in so doing, those authors
display poor judgement.) For now, we will content ourselves with pointing out
that other authors use the terms non-decreasing and non-increasing instead of
increasing and decreasing.
44 CHAPTER 2. CONTINUITY
2.8.2 Theorem (Extension of Monotonicity)
Let F be a continuous function on an interval [a, b], and suppose that f is strictly
increasing on (a, b). Then F is strictly increasing on [a, b].
Proof: We must show that if a ≤ x
1
< x
2
≤ b, then f[x
1
] < f[x
2
], and we
need only worry about the possibilities that a = x
1
, b = x
2
, or both—for the
conclusion that f[x
1
] < f[x
2
] is given otherwise. Let us begin with the case
a = x
1
< x
2
< b. We suppose, by way of contradiction, that f[a] ≥ f[x
2
]. By
Corollary 2.6.7, we can find x
3
∈ (a, x
2
) such that f[x
3
] =
1
2
(f[a] +f[x
2
]). But
this yields a < x
3
< x
2
, with f[x
3
] ≥ f[x
2
], which is not possible. We conclude
that f[x
1
] < f[x
2
].
Similar reasoning shows that if a < x
1
< x
2
= b, then f[x
1
] < f[x
2
].
Finally, we note that if a = x
1
< x
2
= b, we can choose any x
3
∈ (a, b) and note
that we must have f[x
1
] < f[x
3
] < f[x
2
] by what we have already shown.•
Entirely similar reasoning shows that we can remove both instances of the word
“strictly” from Theorem 2.8.2 without affecting the truth of the statement. We
leave the details to the reader.
We now add a useful tool to our collection of limit theorems.
2.8.3 Theorem (Bounded Monotonic Convergence)
Let f be a function defined and increasing on an interval (a, b). If there is a
number M such that f[x] ≤ M for every x ∈ (a, b), then lim
x→b
− f[x] exists
and is no greater than M.
Proof: Consider the set of numbers S = { f[x] : a < x < b }. By hypothesis,
S is bounded above (by M), and so by Axiom 2.5.5 S has a least upper bound L.
Let > 0 be given. We must be able to find a number x
0
in (a, b) such that
L − < f[x
0
] ≤ L, for if there were not such a number then L − would be an
upper bound for S that is smaller than L, and this is impossible. Let us take
δ = b − x
0
. Note that δ is positive because x
0
< b. Moreover, if b − δ < x < b
we have L− < f[x
0
] ≤ f[x] by the monotonicity of f and f[x] ≤ L because of
the way L is defined. It follows from this that lim
x→b
− f[x] = L ≤ M.•
We leave it to the reader to establish a similar theorem regarding the limit as
x →a
+
of an increasing function which is bounded above below on (a, b).
The following definitions are standard; we give them for the sake of completeness
and to avoid difficulties that may result from slight technical variations. The
2.8. CONTINUITY & INVERSE FUNCTIONS 45
reader should note that an occasional author uses injective for one-to-one and
surjective for onto; this is execrable judgement.
2.8.4 Definition (Onto)
A function f, defined on some domain D ⊆ R, and taking its values in some set
E ⊆ R is said to be onto E if it is the case that for every e ∈ E there is a d ∈ D
such that f[d] = e.
2.8.5 Definition (One-To-One)
A function f defined on some set D ⊆ R is said to be one-to-one on D if it
is the case that whenever d
1
, d
2
∈ D are such that d
1
= d
2
we always have
f[d
1
] = f[d
2
].
2.8.6 Definition (Inverse Function)
Let D and E be sets of real numbers, and let f : D → R and g : E → R be
functions. If f[g[e]] = e for every e ∈ E and g[f[d]] = d for every d ∈ D, then
we say that f and g are inverse to each other. In this case we may write f
−1
in
place of g and we may write g
−1
in place of f.
It is a standard theorem of pre-calculus, which we do not prove here, that a
function f : D →R has an inverse whose domain is E ⊆ R iff f carries D onto
E in one-to-one fashion.
2.8.7 Theorem (Continuous Inverse)
Suppose that f : (a, b) →(α, β) is strictly monotonic and onto. Then f is con-
tinuous on (a, b) and has an inverse f
−1
: (α, β) → (a, b) which is also strictly
monotonic, onto, and continuous.
Proof: We will assume, without loss of generality, that f is strictly increasing;
the proof is very similar for strictly decreasing f. Choose x
0
∈ (a, b), and let
> 0 be given. We may assume that is so small that f[x
0
] ± ∈ (α, β), and
thus that α < f[x
0
] − < f[x
0
] < f[x
0
] + < β. Because f is onto, we can
find x

0
and x
+
0
, both in (a, b) so that f[x

0
] = f[x
0
] − and f[x
+
0
] = f[x
0
] + .
We put δ = min{x
0
−x

0
, x
+
0
−x
0
}. Now if x is chosen so that |x −x
0
| < δ, it
follows that x

0
< x < x
+
0
, so that, by monotonicity, f[x
0
] − = f[x

0
] < f[x] <
f[x
+
0
] = f[x
0
] + , or, equivalently, |f[x] −f[x
0
]| < . Hence f is continuous on
(a, b).
That f
−1
exists is immediate from the strict monotonicity of f (which guaran-
tees that f is one-to-one); that the domain of f
−1
is (α, β) follows from the fact
that f maps (a, b) onto (α, β). Moreover, if a < x < b, then f[x] ∈ (α, β) so
46 CHAPTER 2. CONTINUITY
that x = f
−1
[f[x]]. This means that f
−1
maps (α, β) onto (a, b).
Suppose that α < y
1
< y
2
< β. We have f
_
f
−1
[y
1
]
¸
= y
1
< y
2
= f
_
f
−1
[y
2
]
¸
,
and from the monotonicity of f, we find that f
−1
[y
1
] ≥ f
−1
[y
2
] is not possible.
Hence f
−1
is a strictly monotonic function mapping (α, β) onto (a, b), and it
follows from what we have already established that f
−1
is continuous.•
2.8.8 Theorem (Continuous and One-To-One)
Suppose that f is a continuous one-to-one function from an interval (a, b) into
R. If there are points x
1
, x
2
∈ (a, b) with x
1
< x
2
and f[x
1
] < f[x
2
], then f
is strictly increasing on (a, b). If there are no such points, then f is strictly
decreasing on (a, b).
Proof: Suppose that u ∈ (x
1
, x
2
), v ∈ (x
1
, x
2
) and u < v. If f[x
2
] < f[u],
then from f[x
1
] < f[x
2
] < f[u] and Corollary 2.6.7, it follows that we can find
ξ ∈ (x
1
, u) such that f[ξ] = f[x
2
], and this violates the condition that f be
one-to-one on (a, b), for it is clear that ξ < u < x
2
. Thus f[x
2
] ≤ f[u] is not
possible. A similar argument rules out the possibility that f[u] ≤ f[x
1
], and we
conclude that f[x
1
] < f[u] < f[x
2
].
Since we now know that f[u] < f[x
2
] and v ∈ (u, x
2
), the same argument es-
tablishes that f[u] < f[v] < f[x
2
]. Thus f is strictly increasing on [x
1
, x
2
].
Now suppose that a < u < v < x
1
and f[x
1
] < f[u]. Let y
0
be the smaller of
f[u] and f[x
2
], and choose η ∈ (f[x
1
], y
0
). Apply Corollary 2.6.7 to conclude
that there is a ξ
1
∈ (u, x
1
) such that f[ξ
1
] = η. Apply Corollary 2.6.7 another
time to conclude that there is also a ξ
2
∈ (x
1
, x
2
) with f[ξ
2
] = η. But then
f[ξ
1
] = f[ξ
2
] with ξ
1
< x
1
< ξ
2
, and this violates the one-to-one condition on
f. We conclude that f[u] < f[x
1
]. But then v lies between u and x
1
, where
f[u] < f[x
1
], and by the first part of this proof, it follows that f[u] < f[v].
Hence, f is increasing on (a, x
1
].
Careful examination of the arguments above reveals that we have shown that if
a < u < x
1
, then f[u] < f[x
1
] and if x
1
< v < x
2
, then f[x
1
] < f[v] < f[x
2
].
From this and the fact that f is increasing in each of the intervals (a, x
1
] and
[x
1
, x
2
], we see that f is increasing on (a, x
2
]. The remainder of the argument,
extending these results to (a, b) is similar, and we omit it.
If every pair of numbers x
1
< x
2
satisfies f[x
1
] > f[x
2
] (f[x
1
] = f[x
2
] is not
possible, because f is given one-to-one and x
1
= x
2
) we can repeat the above
arguments, with appropriate, and obvious, changes, to see that f is strictly de-
creasing on (a, b).•
Note that we allow either or both of a and α to be −∞ and either or both of b
and β to be +∞ in Theorems 2.8.7 and 2.8.8.

28

CHAPTER 2. CONTINUITY

that the limit is actually f [a], we may drop the first of the inequalities in the condition 0 < |x − a| < δ to obtain the condition of Definition 2.1.1. Pedagogically speaking, the second of the definitions above is probably preferable in a freshman class. The ’s and δ’s come hard to freshmen, and the fewer of them, the better. Moreover, Definition 2.1.2 is close in spirit to the reason for introducing continuity in the first place: Continuous functions are those that behave well in relation to limits.

2.1.3

Theorem (Continuity and Limits)

Suppose that limx→a g[x] = L and that f is a function continuous at L. Then
x→a

lim f [g[x]] =

f lim g[x] .
x→a

(2.1)

Proof: The function f is given continuous at L, so the quantity on the righthand side of (2.1), which is f [L], is defined. We must show that limx→a f [g[x]] = f [L]. To this end, let > 0 be given. We must find δ > 0 with the property that 0 < |x − a| < δ guarantees |f [g[x]] − f [L]| < . But by the continuity of f , we know that, corresponding to our given > 0, there is an η > 0 such that 0 < |u − L| < η implies that |f [u] − f [L]| < . And we have also been given that limx→a g[x] = L, so we can find δ > 0 such that |g[x] − L| < η whenever 0 < |x − a| < δ. But then it follows that for 0 < |x − a| < δ we must have |f [g[x]] − f [L]| < , and so limx→a f [g[x]] = f [L].• Thus, continuous functions are those which commute with the limit operation. And compositions of continuous functions are continuous.

2.1.4

Corollary

If g is a function continuous at a, with g[a] = b, and if f is a function continuous at b, then the composition f ◦ g is continuous at a.•

2.2
2.2.1

Continuity on Intervals and Other Sets
Definition (Continuity on Open Intervals)

If a and b are real numbers with a < b, and if f is a function, we say that f is continuous on (a, b) when f is continuous at each point of (a, b). Thus, a function is continuous on an open interval if it is continuous at each point of the interval. There is a subtle point here that will arise again in the

2. respectively.3 Definition (Continuity on a Closed Interval) The function f is continuous on [a. Here is the definition. b] if f is continuous on (a. we require “one-sided” continuity at such endpoints: 2. It is often desirable to talk about continuity of functions on sets that are not intervals. f ·g. which subsumes all of the previous definitions. if > 0 be given.3 2. b).3. f is continuous from right at a. x→a f is continuous from the left at a if the corresponding statements are true with x → a+ replaced by x → a− . and f is continuous from the left at b. f p/q and f /g are.2. Another problem arises when we want to talk about the continuity of a function on an interval that contains one or both of its endpoints. and x→a+ 3. f [a] exists. f is said to be continuous on D if for every d ∈ D and for every > 0 is is possible to find δ > 0 such that |f [x] − f [d]| < whenever x ∈ D is such that |x − d| < δ. 2.3. there must be δ1 > 0 such that |f [x] − f [x1 ]| < whenever |x − x1 | < and there must be δ2 > 0 such that |f [x] − f [x2 ]| < whenever |x − x2 | < δ2 . We will return to these matters. q = 0 are integers. then f must be continuous at x1 and f must be continuous at x2 . Accordingly. then the functions f +g. 2. But it need not be the case that |f [x] − f [x2 ]| < when |x − x2 | < δ1 unless it happens that δ1 < δ2 . lim f [x] exists. lim+ f [x] = f [a]. b). THE CONTINUITY THEOREMS 29 future. the functions .2.2 Definition (One-Sided Continuity) The function f is continuous from the right at a if 1. and it need not be the case that |f [x] − f [x1 ]| < when |x − x1 | ≤ δ2 unless it happens that δ2 ≤ δ1 . f −g.2. Typically.2. If f is continuous on (a.1 The Continuity Theorems Definition (Algebra of Functions) If f and g are functions having a common domain and p.4 Definition (Continuity on a Set) Let D be a subset of the domain of a function f . 2. b) and if x1 and x2 are points of (a.

5) (2. (f [x])p/q .3. b).3 Theorem (Continuity of Rational Functions) The functions x → c (where c is a fixed real number) and x → x are both continuous on R. Proof: Continuity of constant functions is immediate from the Constant Limit Theorem. and f · g are all continuous on the interval (a.6) It is common to suppress the “·” in f · g. we would like to say that a function f has a discontinuity at any point a where it is not continuous. Extraction. b). b) for which g[x] does not vanish. f − g. The domain of f is then the set (−∞. ∞). b). Constant. every polynomial function is continuous on R and every rational function is continuous on its domain. = g[x] (2.3) (2. Then the functions f + g. and Division Limit Theorems of the previous chapter. On the other hand it is not uncommon to find authors writing of the . The function f /g is continuous at those points x of (a.• 2. CONTINUITY f [x] + g[x]. Proof: Immediate from the definitions and the Addition. f [x] − g[x].• 2. and continuity of the function x → x is immediate from the Identity Function Limit Theorem.3.3.2. the term is used informally.30 defined by the equations (f + g)[x] (f − g)[x] (f · g)[x] (f p/q )[x] (f /g)[x] = = = = CHAPTER 2. The rest of the theorem now follows from Theorem 2. −1] ∪ [1. f [x] . f [x] · g[x]. In consequence. Following the usual convention that the domain of a function is the largest set of real numbers for which the defi√ nition of the function is meaningful. If 0 < f [x] for all x ∈ (a. b).2 Theorem (Algebra of Continuity) Let f and g be functions continuous on an interval (a. we commonly write (f g)[x] in place of (f · g)[x]. then the function f p/q is continuous on (a. Most authors would hesitate to say that f had a discontinuity at the origin—though f is certainly not continuous at that point. let us put f [x] = x2 − 1. 2. but we should probably be more careful.4) (2. Multiplication.3.4 Discontinuities Not many authors take the effort to give a formal definition for the term discontinuity.2) (2. Generally speaking.

If. It is not at all difficult to prove analogs of the theorems of §1. and this means that limn→∞ f [xn ] = f [L]. Then f is continuous at L if and only if limk→∞ f [xk ] = f [L] for every sequence {xk }∞ for k=1 which limk→∞ xk = L. Limit of a Sequence) A sequence is a function x : N → R. 31 One kind of discontinuity frequently does receive a formal definition. Proof: Suppose that f is continuous at L. we often use the existence of such an extension in order to evaluate a limit. on the other hand. a function f which has a removable discontinuity at a possesses an extension F that agrees with f away from a but is continuous at a. We leave formulation of the statements. we usually write xn instead of x[n] for the value of the function x and the natural number n. This is.4. Choose δ > 0 such that |f [x] − f [L]| < whenever |x − L| < δ.7)  lim f [x]. x = a. Then.4.2. f is not continuous at L. x = a. in fact. If L is a n=1 real number.5 for limits of sequences.4. x→a Alternatively. whatever > 0 may be given. As the term suggests. we write limn→∞ xn = L if. then f is said to have a removable discontinuity (or a removable singularity) at a. .1 Sequences & Continuity Definition (Sequence. and let f be a function whose domain contains L. One defines F by  f [x]. for each n ∈ N.2. and let {xn }∞ be any sequence with n=1 limn→∞ xn = L.2 Theorem (Sequential Continuity) Let L ∈ R. When x is a sequence. We usually write {xn }∞ or { xn : n ∈ N } for the sequence x.2.  F [x] = (2. But then |f [xn ] − f [L]| < whenever n ∈ N and n ≥ N . SEQUENCES & CONTINUITY discontinuity of the function x → x−1 at the origin.4 2. because limn→∞ xn = L. as well as proofs. so that |xn − L| < 1/n on the one hand. Let > 0 be given. it is always possible to find N ∈ N such that |xn −L| < whenever n ∈ N and n ≥ N . as exercises for the reader. If f is a function that fails of continuity at a point a because it violates condition 1 or condition 3 (but not condition 2) of Definition 2. 2.9). we can find N ∈ N such that |xn − L| < δ whenever n ∈ N and n ≥ N . another point of view of the Equality Theorem (Corollary 1.5. then for some > 0 it is possible to find xn . 2.

We leave it to the reader to formulate analogous definitions for the terms lower bound and greatest lower bound (alternatively.32 CHAPTER 2.5 Some Topological Considerations In our discussion of continuous. (The latter is vacuously empty: Every member of the empty set is the center of an interval contained in the empty set—because there are no such members. To this end. That is. If U is an open cover of S and V ⊆ U such that the union of the sets in V also contains S. then V is called a finite sub-cover for U. δ > 0) which is itself a subset of U . A real number m is said to be a least upper bound (or a supremum) for S if m is an upper bound for S and no upper bound for S is smaller than m. is a special case of an open subset of the real numbers. infimum).5.2 Definition (Upper Bound. An open cover of S is any family U = { Uα : α ∈ A } of open sets whose union contains S.3 Definition (Open Cover. Note further that the set of all real numbers is an open set.5.• 2. Sub-Cover) Let S be a set of real numbers. An open interval. If V is a sub-cover for an open cover U and if V consists of finitely many of the open sets from U. of course. as is the empty set. .) 2. Least Upper Bound) Let S ⊆ R. it will be helpful to adopt a somewhat more advanced point of view than we ask freshmen to adopt.5. we introduce some topological terminology and derive some facts that lie beyond the scope of typical freshman calculus courses. as we have been using the term. a subset U of the real numbers is open if for every u ∈ U there is a δ > 0 such that (u − δ. 2.1 Definition (Open Set) A subset U of the real numbers is said to be open if every point u ∈ U is the center of an interval (u − δ. and only if. u + δ) ⊆ U . It can be shown (though we shall not do so here because we have no use for the fact in what follows) that a subset of the real numbers is open if. A real number m is said to be an upper bound for S if s ≤ m for every s ∈ S. it is a union of countably many pairwise disjoint open intervals. It is clear that limn→∞ xn = L but limn→∞ f [xn ] = f [L]. 2. u + δ) (where. real-valued functions of a real variable. CONTINUITY but |f [xn ] − f [L]| ≥ on the other. then V is called a sub-cover for U.

and this contradicts our choice of M as the least upper bound of the set of accessible points in [a. such that limn→∞ (bn − an ) = 0 and for every n ∈ N. But then the points of (M. (If it were not. . Suppose also ∞ that k=1 [ak . b1 ].5. b]. If b is accessible. Uα2 . Suppose now that we are given a family of intervals [an . is such that (1) limn→∞ (bn − an ) = 0. Let us call a point x of the interval [a. bn ]. n ∈ N. b] and that U is an open cover of [a. an ) ∪ (bn . bn ].5. From the condition limn→∞ (bn − an ) = 0. [an . Thus. 2. SOME TOPOLOGICAL CONSIDERATIONS 33 2. Let S denote the set of all accessible points of [a. such a point would necessarily lie in every one of the intervals [an . n ∈ N—and we have assumed that there is no such point. b1 ]. then it has all of them: 1. From the ∞ fact that k=1 [ak . M + δ) are also accessible (because they all lie in Uα0 ). [an . b1 ] would not lie in any of the sets Un . 3. and then we observe that [a. bn ]. M ] must contain an accessible point x0 . . bk ] = Ø is impossible. bk ] = Ø. . The Least Upper Bound Property: Every non-empty set of real numbers that has an upper bound also has a least upper bound. so S is nonempty.) But it is clear that U admits no finite subcover. M ] ⊆ k=0 Uαk .2. b] and an open cover U = { Uα : α ∈ A } of [a. let Un = (−∞. . .4 Theorem (Completeness Conditions) If the set R of real numbers has any one of the following properties. But M is the least upper bound of the set of accessible points in [a. Suppose also that we are given an interval [a. n ∈ N. Uαn be members of U such that n n [a. bn ] ⊇ [an+1 . x] ⊆ k=1 Uαk . bn+1 ]. b ∈ R with a < b. We let Uα1 . . x0 ] ⊆ k=1 Uαk . For each n ∈ N. Clearly M ≤ b. Proof: Suppose first that R has the Least Upper Bound Property. Because Uα0 is open. there is a δ > 0 such that (M − δ. then it must follow that k=1 [ak . bk ] = Ø. then every open cover of [a. b]. b]. then we are done. The Nested Interval Property: If a family of intervals [an . bn+1 ]. if b is not accessible then b is an upper bound for S and—because we have assumed that R has the Least Upper Bound Property—we may conclude that S has a least upper bound M . If we suppose that ∞ R has the Heine-Borel Property. bn ] ⊇ [an+1 . It is clear that a is accessible. bn ]. n ∈ N. bn+1 ] guarantees that no finite sub-family of U can cover [a1 . The Heine-Borel Property: If a. bn ] ⊇ [an+1 . ∞). b] admits a finite sub-cover. . we see that it is not possible for the intersection to contain more than one point. . M + δ) ⊆ Uα0 . so the interval (M − δ. because the nesting condition [an . From the facts that M ∈ [a. M is accessible. then there is exactly one real number common to all of the intervals [an . (If there were two . and (2) for every n ∈ N. we conclude that there is an α0 ∈ A such that M ∈ Uα0 . b]. b] unless M = b. we see that U = { Un : n ∈ N } is an open cover for [a1 . n ∈ N. then some x0 ∈ [a1 . b] accessible if there are finitely n many indices α1 . αn ∈ A such that [a. This establishes that if R has the Least Upper Bound Property then it also has the Heine-Borel Property.

There are two possibilities: Either c1 is an upper bound for S or it is not. Choose a1 ∈ S and put b1 = m. But an1 is not an upper bound for S. We can continue this construction inductively to obtain a sequence of intervals { [ak .” and we will see it again. we have obtained an interval [a2 . and the length of this interval is 2(1−n0 ) (b1 − a1 ). Then there is an n0 ∈ N such that 2(1−n0 ) (b1 −a1 )+M < s.5 Axiom (Completeness of the Reals) Axiom: The real numbers have the Least Upper Bound Property. Consequently. we have seen that if R has the Nested Interval Property it must also have the Least Upper Bound Property. bk ] : k ∈ N } such that [an . this is not the place for such a construction and we will adopt a different strategy. which means that T < an1 . which violates the fact that bn0 is an upper bound for S. so suppose that a1 is not an 1 upper bound for S. b2 ] whose left-hand end-point is not an upper bound for S and whose right-hand end-point is an upper bound for S. we would have to have an ≤ x1 < x2 ≤ bn for all n.34 CHAPTER 2. bn ] ⊇ [an+1 . If a1 is a least upper bound for S. then we have shown that S has a least upper bound. and M must be a least upper bound for S. bn+1 ] for every n ∈ N and (bn − an ) = 2(1−n) (b1 − a1 ) → 0 as n → ∞.• The method we have used in the last part of this proof. In the second case. 2.5. bn ]. where we showed that the Nested Interval Property implies the Least Upper Bound Property is called “the Method of Bisection. bn0 ]. there is then a single number M that lies in every one of the intervals [an . then it has all of them. and this would make limn→∞ (bn − an ) = 0 impossible.5. Thus.) This shows that if R has the Heine-Borel Property then it also has the Nested Interval Property. Let c1 = 2 (a1 + b1 ). However. we put a2 = c1 . we would be careful to build one of these properties into the construction. Suppose that there is an s ∈ S such that M < s. CONTINUITY points x1 < x2 in the intersection. Note that we have not proved that R has any one of the three conditions listed in Theorem 2. we have only proved that if R has any one of them. Then there is n1 ∈ N such that T < M − 2(1−n1 ) (b1 − a1 ). by reasoning similar to that of the preceding paragraph. Suppose now that T is any number such that T < M . But M lies in the interval [an0 . and suppose that m is an upper bound for S. This means that bn0 < s. In the first case we put a2 = a1 and b2 = c1 . and this means that T cannot be either. . Now let S be a non-empty set of real numbers. M is an upper bound for S. Hence no upper bound for S can be smaller than M . b2 = b1 . If R has the Nested Interval Property.4. In either case. If our purpose here were to construct the set of real numbers.

then x ∈ f −1 [U ]—or.6. f [x0 ] + ) ⊆ U . for if v ∈ V . the real numbers. Consequently. . f [x] ∈ U . and whatever > 0 may be given. Proof: First. Now let us suppose that whenever U ⊆ R is open. x0 + δ) ⊆ V .• We are now ready to give proofs of a number of theorems whose statements. If x0 ∈ f −1 [U ]. so there is a corresponding δx0 > 0 such that |f [x] − f [x0 ]| < whenever x ∈ D and |x − x0 | < δx0 .6. We note that U = { y : |y − f [x0 ]| < } is an open subset of R. Then f is continuous on D iff for every open subset U of R. We have introduced these terms because continuity has a very simple description using them: A function is continuous if and only if inverse images of open sets are open sets: 2.1 More on Continuity Definition (Inverse Image) Let f be a function carrying some set of real numbers into the real numbers. because U is open. Consequently. and let U be an open subset of R. suppose that f is continuous on its domain D. so we have now seen that whatever x0 ∈ D we begin with. Let x0 ∈ D. and suppose that A is collection of real numbers. it is possible to find an open subset V of R such that f −1 [U ] = D ∩ V . but not whose proofs.2 Theorem (Continuity Characterization) Let f be a function defined on a subset D of R. there is a positive δ such that the open interval (x0 − δ. But f is continuous at x0 . then f [x0 ] ∈ U .6 2. Putting Ix0 = { x : |x − x0 | < δx0 }. Because x0 ∈ f −1 [U ] ⊆ V . But this means that if |x − x0 | < δ and x ∈ D. then v ∈ Ix ⊆ V for some x ∈ f −1 [U ]. in other words. there is another open subset V of R such that f −1 [U ] = D ∩ V . The latter is equivalent to |f [x] − f [x0 ]| < .6. there is an > 0 such that the open interval (f [x0 ] − . We begin with the boundedness of a continuous function whose domain is a closed bounded interval. Consequently f is continuous on D. we let V = x∈f −1 [U ] Ix . we know that x0 ∈ V . Then f −1 [U ] = D ∩ V and V is open. MORE ON CONTINUITY 35 2. we mean the set { x : f [x] ∈ A }.2. By the inverse image of A under f (written f −1 [A]). and suppose that > 0 has been given. it is always possible to find δ > 0 such that whenever x ∈ D and |x − x0 | < δ we will have |f [x] − f [x0 ]| < . and so there is an open subset V of R such that f −1 [U ] = D ∩ V . are standard parts of freshman calculus courses.

If f [x] = α for every x ∈ [a. an open set Un such that ∞ ∞ f −1 [Vn ] = [a. b]. This means that M = 1/ < 1/(α − f [x0 ]) = F [x0 ]. the non-empty set S = { f [x] : x ∈ [a. b] ⊆ Un1 ∪ Un2 ∪ · · · ∪ Unm . . we have α − < f [x0 ] < α.5 Theorem (Local Invariance of Sign) If f is continuous on some open interval that contains the number ξ and if f [ξ] > 0.. and therefore [a. ξ + δ). the set Vn = (−n.6. b] such that f [x] ≤ f [ξ] for all x ∈ [a. b]. There are points η. ξ ∈ [a. α. b]. b]. and so by Theorem 2.3. b] ∩ Un . b] such that f [η] ≤ f [x] for all x ∈ [a. b]. . we are done. n) is an open set of real numbers. Putting M = nm .6. m}. .• 2.3 Theorem (Boundedness) Let f be a continuous function defined on a closed. and from this it follows that α − f [x0 ] < .8) is continuous on [a.6. then s = f [x0 ] for some x0 ∈ [a. For this x0 . The existence of η ∈ [a. there must be an element s of S in the interval (α − . b]. But if s ∈ S. CONTINUITY 2. Then > 0.6.6. But then if x ∈ [a. Take = 1/M . f [ξ] < 0) then there is a δ > 0 such that f [x] > 0 (resp. for each n ∈ N. b] such that f [η] ≤ f [x] ≤ f [ξ] for every x ∈ [a. b] now follows by applying what we have already established to the function g[x] = −f [x].3 once again. b]. then the function F given by F [x] = 1 α − f [x] (2. bounded interval [a. the Least Upper Bound Property assures us that S has a least upper bound. . bounded interval [a. Proof: For each n ∈ N.4 Theorem: (Extreme Value Theorem) Let f be a continuous function defined on a closed. α). . Applying Theorem 2. (respectively. b]. 2. b] ⊆ f −1 [ k=1 Vk ] = ∞ ∞ −1 [Vk ] ⊆ k=1 Uk .• 2. we find M > 0 such that |F [x]| < M for all x ∈ [a. This establishes the existence of ξ ∈ [a. The Heine-Borel Property assures us that there are k=1 f finitely many n1 < n2 < · · · < nm ∈ N such that [a.6. f [x] < 0) for all x ∈ (ξ − δ.36 CHAPTER 2. There is a real number M such that |f [x]| < M for all x ∈ [a. Proof: By Theorem 2. and because α is a least upper bound for S. which is the same as −nm < f [x] < nm . which is a contradiction. b]. b] } of real numbers is bounded above. Now k=1 Vn = R. b] we must have f [x] ∈ Vnk for some k ∈ {1. Because it is automatic that Vnk ⊆ Vnm it follows that f [x] ∈ Vnm .2 there is.

b] for which f [ξ] = 0 or a sequence of closed intervals [a1 . bk ]. The proof is the same for f [ξ] < 0. so |ξ − aN | < δ. that f [ξ] < 0. Proof: We use the Method of Bisection. bounded interval [a. b1 ] ⊇ [a2 . Hence f [bN ] > 0. f [bN ] < 0. b2 ] ⊇ · · · ⊇ [ak . If η is any real number lying between f [α] and f [β]. b) such that f [ξ] = 0. we can find N ∈ N such that bN − aN < δ. by construction. Suppose. Otherwise. Let a1 = a. so |ξ − bN | < δ. which 1 implies that 0 < 2 f [ξ] = f [ξ] − < f [x].6 Theorem (Intermediate Value Theorem) Let f be a continuous function defined on a closed. But. bk ]. then there is a number γ lying between α and β and such that f [γ] = η. Then ξ ∈ [aN . b2 ] then meets the conditions that f [a2 ] > 0 and f [b2 ] < 0 and is half the length of the interval [a1 . then there is a number ξ ∈ (a. in the second.6. b]. Now ξ ∈ [aN . b1 = b.2. there is a δ > 0 such that f [x] > 0 whenever x ∈ [a.• 2. Suppose that f [ξ] > 0. Proof: We may assume that α < β (if not. It is clear how to proceed inductively to obtain either ξ ∈ [a. In the latter case. If f [c1 ] = 0 we put ξ = c1 and we are through. there is a δ > 0 such that f [x] < 0 for all x ∈ [a. we conclude that f [ξ] = 0. Again by Local Invariance of Sign. on the other hand. b] and |ξ − x| < δ. bN ]. Then for x ∈ (ξ − δ. we put a2 = c1 and b2 = b1 . The interval [a2 . we obtain a unique number ξ common to all of the intervals [ak .• 2. as well as those of the Nested Interval Property. All that remains is to apply . MORE ON CONTINUITY 37 Proof: Take = 1 f [ξ] > 0.5). If f [a] > 0 and f [b] < 0. By our construction of the intervals [ak . By Local Invariance of Sign (Theorem 2. bk ] ⊇ · · · all meeting the conditions f [ak ] > 0 and f [bk ] < 0. Hence f [aN ] < 0. a2 = a1 and b2 = c1 .6.6. there are two cases: f [c1 ] > 0 or f [c1 ] < 0.7 Corollary Let f be continuous on an interval (which may be open or closed or neither) which contains the points α and β. Once again. Having found that both f [ξ] < 0 and f [ξ] > 0 are impossible. we simply interchange the names) and that f [α] > f [β] (if not. But. and the contradiction shows that f [ξ] > 0 is not possible. b1 ]. ξ + δ) we have − < f [x] − f [ξ] < . and let c1 = 1 2 (a1 + b1 ). we work with −f ). In the first case.6. and the contradiction shows that f [ξ] < 0 is not possible. we can find N ∈ N so large that bN − aN < δ. and find δ > 0 so that |f [x] − f [ξ]| < whenever 2 |x − ξ| < δ. by construction. bN ]. f [aN ] > 0. b] such that |ξ − x| < δ.

• CHAPTER 2. But experience suggests that freshmen initially interpret pointwise continuity on an interval as uniform continuity. and it is clear that we don’t need to talk about the difference between the two in the presence of freshmen. In the second instance. Then we must be able to find δ > 0 in such a way that every pair of numbers u.9) and (2. 2. 2.1 Definition (Uniform Continuity) A function f defined on a set D of real numbers is said to be uniformly continuous on D if for every > 0 there is a corresponding δ > 0 such that whenever u ∈ D and v ∈ D are such that |v − u| < δ it follows that |f [v] − f [u]| < . however. which we repeat here: In the case of pointwise continuity δ may depend upon both u and . In order to understand the difference between pointwise continuity on a set D and uniform continuity on D. first of all.10) The statement f is uniformly continuous on D means (∀ > 0)(∃δ > 0)(∀u ∈ D)(∀v ∈ D)[(|v − u| < δ) ⇒ (|f [v] − f [u]| < )]. if is helpful to render the definitions formally.7.7. In the first instance we must first select a point u ∈ D and an > 0.38 the Theorem to the function F [x] = f [x] − η.9) The statement f is continuous on D means (∀u ∈ D)(∀ > 0)(∃δ > 0)(∀v ∈ D)[(|v − u| < δ) ⇒ (|f [v] − f [u]| < )]. Note the emphasis above. Then we must find a δ > 0—which may depend upon both u and —in such a way that every v ∈ D with a certain property satisfies a certain inequality. v that are within δ of each other satisfy a certain inequality.10) with a view toward proving that a function f has one of the properties.7. It is not at all clear that this is wise.7 Uniform Continuity Uniform continuity is rarely presented in freshman calculus courses.3 Uniform Continuity on D: (2. Compare (2. 2. The mistake seems to be natural. In the case of uniform . Cauchy himself fell into the trap of confusing pointwise concepts with the uniform ones. and only. The difference between pointwise continuity (which is what we presented in the first section of this chapter) and uniform continuity is subtle.2 Pointwise Continuity on D: (2. we choose . CONTINUITY 2.

4 Example (x2 ) Let f [x] = x2 when x ∈ R. But we also have δ ≤ (1 + 2|u|)−1 . we may write: |f [v] − f [u]| = = = ≤ < |v 2 − u2 | |v + u| · |v − u| |(v − u) + 2u| · |v − u| (|v − u| + 2|u|) · |v − u| (1 + 2|u|) · |v − u|. we may choose different δ’s in different parts on D. and take v = u + 1 δ. For let δ > 0 be given. The reader should note how δ is chosen to depend upon u. to demonstrate pointwise continuity on D. Let us take δ = min 1 + 2|u| . but |u2 − v 2 | > . This shows that f is continuous at u for each u ∈ R. We therefore expect that it will be harder to show that a function is uniformly continuous than to show that it is pointwise continuous. once has been specified. but we will prove it directly.12) (2.7. Then |v − u| = 2 δ < δ. Let u ∈ R. or. Thus.3).2. we must find a single δ and show that it works everywhere in D.16) where the last of these inequalities is correct because |v − u| < δ ≤ 1. then whatever δ > 0 may be. (2. that (∃ > 0)(∀δ > 0)(∃u ∈ R)(∃v ∈ R)[(|u − v| < δ) ∧ (|f [u] − f [v]| ≥ )]. δ may depend upon only.3. we must show that condition (2. it is possible to find u and v such that |u − v| < δ. equivalently.1 . (2.15) (2.7. We hope that this will help to illuminate the difference between pointwise continuity and uniform continuity. UNIFORM CONTINUITY 39 continuity. and not upon both of the numbers u and . In order to establish that f is not uniformly continuous on R. (2. and this means that the quantity on the right side of (2.18) .13) (2. This is best seen through an example. 2.10) fails.14) (2. even though has not changed. but 2 |v 2 − u2 | = 1 1 + δ δ 2 2 − 1 δ 2 (2. Take 1 u = δ −1 . and let > 0 be given. That f is continuous on R is immediate from what we have already seen (see Theorem 2. But to demonstrate uniform continuity of D. Then we will show that f is not uniformly continuous on R.17) We will show that if = 1.16) cannot exceed . if |u − v| < δ.11) With this choice of δ.

all in [a. Although the preceding argument shows that we will not be able to prove that f is uniformly continuous on R. .19) (2. is equivalent to |u + v| · |u − v| < . we would have to let > 0 be given. Find δ > 0 so that |f [v] − f [u]| < whenever u and v are points of D for which |v − u| < δ. Then it follows that |f [v] − f [d]| < whenever v ∈ D is such that |v − d| < δ. Under those circumstances. and let > 0 be given. such that [a. CONTINUITY δ2 1 1 +1+ − 2 2 δ 4 δ 2 δ = 1+ . and this means that f is continuous at d. b]. The conclusion here.7. . Thus. un .40 = CHAPTER 2. . find δu so that |f [v] − f [u]| < 1 2 whenever v ∈ [a. (We know that we con accomplish this because f is continuous at each point of [a. it is instructive to see where an attempt to construct a proof fails. 2.20) and this is larger than 1 = . b].5 Theorem (Uniform Continuity Implies Continuity) If f is a function uniformly continuous on a set D. f is not uniformly continuous on R. 2. We would then like to find δ > 0 such that (|u−v| < δ) ⇒ (|u2 −v 2 | < ).• One of the reasons why we might want to consider replacing pointwise continuity with uniform continuity in the elementary calculus sequence is that the bulk of what we do in that sequence deals with functions that are (pointwise) continuous on a closed interval of finite length. |u2 − v 2 | < . . Proof: Let d ∈ D. In order to give such a proof. u2 .7. b]. It is not possible to find an example of a function that is uniformly continuous on a set D but not pointwise continuous on D. uniform continuity is an automatic consequence of the assumed pointwise continuity. b].) By the Heine-Borel Property of the real numbers. and this is where things go wrong. then f is continuous at each point of D. Proof: Let > 0 be given. we can find u1 . b]) If the function f is (pointwise) continuous on the interval [a. We must also be able to control the magnitude of the sum (u + v). and we cannot do this unless we can control either u or v—which we are forbidden to do—as well. b] is such that |u − v| < δu . b]. For each u ∈ [a. b] ⊆ . 4 (2. then f is also uniformly continuous on [a. We cannot control the magnitude of the product (u + v)(u − v) simply by controlling the magnitude of the difference (u − v)— which we are allowed to do.6 Theorem (Continuity on Intervals [a.

with common domain [a. . b] with |v − u| < δ. and such that F [x] = f [x] = G[x] for all x ∈ D. .7 Definition (Dense Subset) Let I be an interval. x1 . uk + δuk ). they are both continuous at x0 . x + δ) always contains a point of D. m }. UNIFORM CONTINUITY n k=1 (uk − δuk . b]. Because F and G are both uniformly continuous on [a. We put δ = min{ xk − xk−1 : k = 1.25) . . b].7. Then there is a unique function F . Because D is dense in [a. uk0 + δuk0 ). Proof: Let us show first that there can be at most one such function. we can find x1 ∈ D such that |x1 − x0 | < δ. We need a definition first. (2. For this x1 . we have ∆ = {x0 .7. Let x0 be any point of [a. 2. b]. b] together 41 set of all end-points of the intervals with the numbers a and b themselves. b]. b] that does not lie in D. uk +δuk ) that lie in [a. such that F [x] = f [x] for every x ∈ D. and this guarantees in turn that |f [v] − f [u]| = |f [v] − f [uk0 ] + f [uk0 ] − f [u]| ≤ |f [v] − f [uk0 ]| + |f [uk0 ] − f [u]| < 2 + 2 = . D is said to be dense in I if for every x ∈ I and every δ > 0 the interval (x − δ. . and so limx→x0 F [x] = F [x0 ] and limx→x0 G[x] = G[x0 ]. Suppose that we have two functions. This means that u and v must both lie in some one of the intervals (uk0 − δuk0 . . xm can lie in the closed interval whose endpoints are u and v themselves. Let > 0 be given. sorting and renaming. b].• (2. and let f be a function which is defined and uniformly continuous on D. F and G. . . and that F and G are both uniformly continuous on [a. .23) (2. . 2. . . defined on [a. and let D ⊆ I.2. xm }.24) We end this section with a theorem that will be of use later on and a final example. 2. and choose δ > 0 so that |x − x0 | < δ implies that both of the inequalities |F [x] − F [x0 ]| < /2 and |G[x] − G[x0 ]| < /2 hold. then at most one of the numbers x0 .8 Theorem (Extension) Suppose that D is a dense subset of an interval [a. we have |F [x0 ] − G[x0 ]| = |F [x0 ] − f [x1 ] + f [x1 ] − G[x0 ]| (2. Thus.7.21) If u and v are now points of [a. b] and uniformly continuous there. x1 .22) (2. . with a = x0 < x1 < · · · < xm−1 < xm = b. Let ∆ be the (uk −δuk .

n0 .30) .n . If x0 ∈ [a. from whence we conclude that F and G are the same function.26) (2.n ∈ D ∩ (x0 − ηn /2. ax0 . it is possible to find a δ > 0 such that whenever x0 lies in [a. we will have f [x] ∈ Sx0 . / We need only choose n0 ∈ N so that n−1 < and take δ = 1 ηn0 .n − x| < ηn . If we put F [x] = f [x] whenever x ∈ D.n . we may suppose that x0 ∈ D. we see that ax0 .n = f [u] : u ∈ D and x0 − ηn ηn < u < x0 + 2 2 . where bx0 . and that bx0 .n and bx0 . we conclude that there is one and only one number ξ common to all of the intervals [ax0 . once and for all. Appealing to the Nested Interval Property. Noting that for each n ∈ N we have Sx0 . let us choose ux0 .42 CHAPTER 2. we put / Sx0 .28) But was an arbitrary positive number. Suppose that y ∈ Sx0 . b] and x ∈ D is such that |x0 − x| < δ. Because we know that we can do this when x0 ∈ D.n+1 ≥ ax0 .n ] + (2n)−1 is an upper bound for the non-empty set Sx0 . It follows that Sx0 . then |F [x0 ] − f [x]| < . We can accomplish this in such a way that ηn+1 < ηn for each n ∈ N. we choose.n . 0 It remains to show that F is uniformly continuous on [a. Suppose that 3 u and v lie in [a. positive numbers ηn . and a least upper bound bx0 . and so it follows that F [x0 ] = G[x0 ].n0 − ax0 .n ] − f [x]| < (2n)−1 . and because |ux0 .27) (2. we must have |f [ux0 . b] with |v − u| < δ.n . f [ux0 . In fact. this gives a function F defined on all of [a. b]. n ∈ N. (2. bx0 .n+1 ≤ bx0 .n − ax0 . Choose δ1 > 0 so that whenever x and x lie in D with |x − x | < δ1 we must have |f [x ] − f [x ]| < 1 and whenever w ∈ [a. So let > 0 be given. From this it follows that |F [x0 ] − f [x]| < . then for each n ∈ N.n0 ⊆ [ax0 .n+1 ⊆ Sx0 . such that |f [v]−f [u]| < (2n)−1 whenever u and v are elements of D for which |v −u| < ηn . Then y = f [x] for some x ∈ D ∩ (x0 − ηn /2.n has a greatest lower bound. and find x ∈ D with |v−x | < δ. We put F [x0 ] = ξ. b]. b] that do not lie in D.n for every n ∈ N.n0 ]. To this end.n ]. Put δ = 3 δ1 . x0 + ηn /2). x0 + ηn /2) arbitrarily.n is not empty because D is dense in [a. Let us note that if > 0 is given. We must now show how to define F at points of [a. 3 Then |x − x | = |x − v + v − u + u − x | (2.n ] − (2n)−1 is a lower bound for the same set. so that 1 |F [u]−f [x ]| < 1 . so that |F [v]−f [x ]| < 3 . bx0 . Find x ∈ D with |u − x | < δ. CONTINUITY = |F [x0 ] − F [x1 ] + G[x1 ] − G[x0 ]| ≤ |F [x0 ] − F [x1 ]| + |G[x1 ] − G[x0 ]| < 2 + 2 = . b]. Consequently.n .n0 . b].n < n−1 .n and f [ux0 . b] and x ∈ D are such 3 1 that |w − x| < δ1 we must have |F [w] − f [x]| < 1 .n0 < n−1 . Then F [x0 ] ∈ 0 2 [ax0 . (2.n0 ] and if |x0 − x| < δ. bx0 . but x0 ∈ D.29) Sx0 . n ∈ N.

2. D = (0. defined and (uniformly) continuous on [0. Then f is continuous on D (as we will show later). f is said to be strictly increasing on I if from u.31) (2. If the same conditions imply that f [u] > f [v]. However.34) < 3 + 3 + 3 = . |f [x ] − f [x ]| < 1 3 43 (2. 1].7. in consequence. Some authors make those criteria into the definitions. and this is something we will return to later in our discussion of differential criteria for determining when a function has one of these properties. Consequently. in so doing. (2.1 Continuity & Inverse Functions Definition (Monotonicity) A function f defined on an interval I is said to be increasing on I if from u. b = 1.8.36) x and let a = 0. Consequently. v ∈ I with u < v it follows that f [u] ≤ f [v].7. there can be no function F . 1). Second. for if it were. We will use the term (strictly) monotonic of a function to mean that that function is (strictly) increasing on I or that it is (strictly) decreasing on I. and.32) .35) 2. |F [v] − F [u]| = |F [v] − F [x ] + F [x ] − F [x ] + F [x ] − F [u]| (2. those authors display poor judgement. . First. Thus. v ∈ I and u < v imply f [u] ≥ f [v]. we will content ourselves with pointing out that other authors use the terms non-decreasing and non-increasing instead of increasing and decreasing. v ∈ I with u < v it follows that f [u] < f [v]. limx→0+ f [x] does not exist. the uniform continuity hypothesis in the Extension Theorem cannot be relaxed to pointwise continuity.8) would exist. CONTINUITY & INVERSE FUNCTIONS ≤ |x − v| + |v − u| + |u − x | < δ + δ + δ = δ1 . we call f strictly decreasing on I. the condition that the interval in Theorem 2. as can be seen by the argument given in Example 1. (We think that.33) ≤ |F [v] − f [x ]| + |f [x ] − f [x ]| + |f [x ] − F [u]| (2. the function f cannot be uniformly continuous on D.3.3.8. a function such as that described in the Extension Theorem (Theorem 2.• (2. We say that f is decreasing on I when the conditions u.7.9 Example (sin x−1 Again) Let f be the function given by 1 f [x] = sin . There is considerable variation in terminology and definition here. such that F [x] = f [x] for all x ∈ D.) For now.8 2.2.6 be closed cannot be relaxed. This example shows us two things.

We suppose.5 S has a least upper bound L. We now add a useful tool to our collection of limit theorems. The . we can find x3 ∈ (a. S is bounded above (by M ). By hypothesis. and so by Axiom 2.2 Theorem (Extension of Monotonicity) Let F be a continuous function on an interval [a. by way of contradiction. then f [x1 ] < f [x2 ].2 without affecting the truth of the statement. Similar reasoning shows that if a < x1 < x2 = b.6. 2. b]. We must be able to find a number x0 in (a. CONTINUITY 2. if b − δ < x < b we have L − < f [x0 ] ≤ f [x] by the monotonicity of f and f [x] ≤ L because of the way L is defined. b). which is not possible. b]. b) and note that we must have f [x1 ] < f [x3 ] < f [x2 ] by what we have already shown. Note that δ is positive because x0 < b. It follows from this that limx→b− f [x] = L ≤ M . with f [x3 ] ≥ f [x2 ]. b) such that L − < f [x0 ] ≤ L.5. By Corollary 2. then f [x1 ] < f [x2 ]. that f [a] ≥ f [x2 ]. Proof: Consider the set of numbers S = { f [x] : a < x < b }. The following definitions are standard.3 Theorem (Bounded Monotonic Convergence) Let f be a function defined and increasing on an interval (a. and we need only worry about the possibilities that a = x1 . Moreover. and suppose that f is strictly increasing on (a. Proof: We must show that if a ≤ x1 < x2 ≤ b. for if there were not such a number then L − would be an upper bound for S that is smaller than L. Let us begin with the case a = x1 < x2 < b.• We leave it to the reader to establish a similar theorem regarding the limit as x → a+ of an increasing function which is bounded above below on (a. Then F is strictly increasing on [a. We conclude that f [x1 ] < f [x2 ].8. and this is impossible. or both—for the conclusion that f [x1 ] < f [x2 ] is given otherwise. If there is a number M such that f [x] ≤ M for every x ∈ (a. then limx→b− f [x] exists and is no greater than M . we can choose any x3 ∈ (a. b). Let us take δ = b − x0 . We leave the details to the reader. But 2 this yields a < x3 < x2 . b).• Entirely similar reasoning shows that we can remove both instances of the word “strictly” from Theorem 2.8. we note that if a = x1 < x2 = b. Finally. we give them for the sake of completeness and to avoid difficulties that may result from slight technical variations.7. b).44 CHAPTER 2. b = x2 . Let > 0 be given.8. x2 ) such that f [x3 ] = 1 (f [a] + f [x2 ]).

In this case we may write f −1 in place of g and we may write g −1 in place of f .8.8. or.6 Definition (Inverse Function) Let D and E be sets of real numbers. β).5 Definition (One-To-One) A function f defined on some set D ⊆ R is said to be one-to-one on D if it is the case that whenever d1 . and taking its values in some set E ⊆ R is said to be onto E if it is the case that for every e ∈ E there is a d ∈ D such that f [d] = e. b). that a function f : D → R has an inverse whose domain is E ⊆ R iff f carries D onto E in one-to-one fashion. onto.8. by monotonicity. which we do not prove here. and thus that α < f [x0 ] − < f [x0 ] < f [x0 ] + < β. this is execrable judgement. Moreover. b) and has an inverse f −1 : (α.2. b) → (α. 2. defined on some domain D ⊆ R.7 Theorem (Continuous Inverse) Suppose that f : (a. CONTINUITY & INVERSE FUNCTIONS 45 reader should note that an occasional author uses injective for one-to-one and surjective for onto. without loss of generality. |f [x] − f [x0 ]| < . Proof: We will assume.8. 2. then f [x] ∈ (α. 0 0 0 0 We put δ = min{x0 − x− . it 0 0 follows that x− < x < x+ . We may assume that is so small that f [x0 ] ± ∈ (α. x+ − x0 }. we can find x− and x+ . Because f is onto. β) so . Choose x0 ∈ (a. b). d2 ∈ D are such that d1 = d2 we always have f [d1 ] = f [d2 ]. b) which is also strictly monotonic. b) so that f [x− ] = f [x0 ] − and f [x+ ] = f [x0 ] + . equivalently. that f is strictly increasing. β) follows from the fact that f maps (a. β). β) → (a. That f −1 exists is immediate from the strict monotonicity of f (which guarantees that f is one-to-one). the proof is very similar for strictly decreasing f . Hence f is continuous on 0 (a. It is a standard theorem of pre-calculus. If f [g[e]] = e for every e ∈ E and g[f [d]] = d for every d ∈ D. that the domain of f −1 is (α. 2.8. if a < x < b. and let f : D → R and g : E → R be functions. β) is strictly monotonic and onto. b) onto (α. 2. both in (a. Then f is continuous on (a.4 Definition (Onto) A function f . and let > 0 be given. Now if x is chosen so that |x − x0 | < δ. and continuous. then we say that f and g are inverse to each other. f [x0 ] − = f [x− ] < f [x] < 0 0 0 f [x+ ] = f [x0 ] + . so that.

we see that f is increasing on (a. b).• 2.6. and by the first part of this proof. and we omit it. changes. b) into R.8. it follows that f [u] < f [v]. with appropriate. and from the monotonicity of f . x2 ]. extending these results to (a. because f is given one-to-one and x1 = x2 ) we can repeat the above arguments. x1 ] and [x1 . v ∈ (x1 . Suppose that α < y1 < y2 < β. Hence f −1 is a strictly monotonic function mapping (α. f is increasing on (a. . If there are no such points. Now suppose that a < u < v < x1 and f [x1 ] < f [u].6. and obvious. x2 ]. x2 ∈ (a. then f is strictly decreasing on (a. This means that f −1 maps (α. Let y0 be the smaller of f [u] and f [x2 ].• Note that we allow either or both of a and α to be −∞ and either or both of b and β to be +∞ in Theorems 2. then from f [x1 ] < f [x2 ] < f [u] and Corollary 2. and we conclude that f [x1 ] < f [u] < f [x2 ]. But then f [ξ1 ] = f [ξ2 ] with ξ1 < x1 < ξ2 . b). Hence. and it follows from what we have already established that f −1 is continuous. Apply Corollary 2. b) is similar. We conclude that f [u] < f [x1 ]. β) onto (a. then f [x1 ] < f [v] < f [x2 ]. b).8.46 CHAPTER 2. y0 ). the same argument establishes that f [u] < f [v] < f [x2 ]. x1 ]. where f [u] < f [x1 ]. it follows that we can find ξ ∈ (x1 . and this violates the condition that f be one-to-one on (a.7. We have f f −1 [y1 ] = y1 < y2 = f f −1 [y2 ] . x2 ).7 to conclude that there is a ξ1 ∈ (u.6. and this violates the one-to-one condition on f . x2 ) and u < v.8. x1 ) such that f [ξ1 ] = η. Apply Corollary 2. and choose η ∈ (f [x1 ]. β) onto (a. Since we now know that f [u] < f [x2 ] and v ∈ (u. Thus f is strictly increasing on [x1 . b). to see that f is strictly decreasing on (a. b) with x1 < x2 and f [x1 ] < f [x2 ]. From this and the fact that f is increasing in each of the intervals (a. then f [u] < f [x1 ] and if x1 < v < x2 . b). b). Careful examination of the arguments above reveals that we have shown that if a < u < x1 . x2 ). If f [x2 ] < f [u]. we find that f −1 [y1 ] ≥ f −1 [y2 ] is not possible. for it is clear that ξ < u < x2 . But then v lies between u and x1 . CONTINUITY that x = f −1 [f [x]]. If every pair of numbers x1 < x2 satisfies f [x1 ] > f [x2 ] (f [x1 ] = f [x2 ] is not possible. The remainder of the argument. x2 ) with f [ξ2 ] = η.7 and 2. Thus f [x2 ] ≤ f [u] is not possible.8 Theorem (Continuous and One-To-One) Suppose that f is a continuous one-to-one function from an interval (a. x2 ]. A similar argument rules out the possibility that f [u] ≤ f [x1 ]. then f is strictly increasing on (a. Proof: Suppose that u ∈ (x1 .7 another time to conclude that there is also a ξ2 ∈ (x1 . If there are points x1 . u) such that f [ξ] = f [x2 ].8.