2 General Notation

0.1 Pointwise Convergence

Let {fk }k∈N be a sequence of functions on a set X, either complex-valued or

extended real-valued. We say that fk converges pointwise to a function f if for
each individual element x ∈ X, the scalar fk (x) converges to f (x) as k → ∞.
We state this explicitly as a definition.

Definition 0.5 (Pointwise Convergence). Let fk , f be extended real-

valued or complex-valued functions defined on a set X.
(a) We say that fk converges pointwise to f , and write fk → f pointwise, if

∀ x ∈ X, lim fk (x) = f (x).

k→∞

(b) If fk → f pointwise and the fk are extended real-valued functions that

are monotone increasing, i.e.,

f1 (x) ≤ f2 (x) ≤ · · · , all x ∈ X,

then we say that fk increases pointwise to f. We denote this with the

notation fk ր f, and make a similar definition for monotone decreasing
sequences. ♦

If the functions fk are complex-valued then we can rewrite the definition

of pointwise convergence in a form that will allow a better comparison to
the coming discussions of uniform convergence and convergence in measure.
Specifically, if fk (x) is a finite scalar (real or complex) for each x, then fk
converges pointwise to f if and only if

∀ x ∈ X, lim |f (x) − fk (x)| = 0. (0.1)

k→∞

It is instructive for the reader to write this out completely in terms of epsilons.

Example 0.6. As an example, consider fk = χ(0, k1 ) , the characteristic function

of the open interval (0, k1 ). For each individual x ∈ R, we have

lim |χ(0, k1 ) (x) − 0| = 0.

k→∞

Hence χ(0, k1 ) converges pointwise to the zero function (in fact, χ(0, k1 ) ց 0).
However, if 0 < x < 1 then the smaller that x is, the longer we have to wait
(more precisely, the larger that k must become) in order for χ(0, k1 ) (x) to get
close to 0. ♦

A loose summary of pointwise convergence is that fk (x) and f (x) must be

close together for all k large enough, but how long we have to wait for fk (x)
and f (x) to get close can depend on which element x that we are looking at.
 0.2 Uniform Convergence 3

Additional Problems

0.5. Let Fb (R) be the vector space of all bounded, complex-valued functions
on R. Show that there is no norm k · k on Fb (R) such that

fk → f pointwise ⇐⇒ lim kf − fk k = 0.
k→∞

0.2 Uniform Convergence

Again let {fk }k∈N be a sequence of functions on a set X. Uniform convergence

requires a “simultaneity” of convergence that is not required for pointwise
convergence. In essence, we require that fk (x) and f (x) be simultaneously
close for all x ∈ X when k is large enough, or we might describe this by saying
that the maximum difference between fk (x) and f (x) must be small for all
large enough k (technically, it would be better to speak of the suprema of
the differences instead of the maximum difference, but the latter terminology
seems much more descriptive).
Note that in order to for the “maximum difference” between functions to be
small, the function values cannot be ±∞. Therefore, when speaking of uniform
convergence we do not allow our functions take extended real values. Since
real-valued functions are just a special case of complex-valued functions, this
means that we can state the definition for complex-valued functions. There
are several equivalent ways to precisely formulate the definition of uniform
convergence; we choose to do this as follows.

Definition 0.7 (Uniform Convergence). Let fk , f be complex-valued

functions defined on a set X. We say that fk converges uniformly to f , and
write fk → f uniformly, if
 
lim sup |f (x) − fk (x)| = 0. ♦ (0.2)
k→∞ x∈X

In precise terms, the “maximum difference” between fk and f that we

have been loosely referring to is the number supx∈X |f (x) − fk (x)|. Because
this quantity is so important, we introduce the following terminology.

Definition 0.8 (Uniform Norm). Let X be a set.

(a) The uniform norm of a function f : X → C is the supremum of |f | over X,
and is denoted by
kf ku = sup |f (x)|.
x∈X

(b) If f and g are two functions on X, then kf − gku is called the distance
between f and g in the uniform norm. ♦
4 General Notation

In this terminology, uniform convergence simply means that the distance

between fk and f shrinks to zero when measured by the uniform norm:

fk → f uniformly ⇐⇒ lim kf − fk ku = 0. (0.3)

k→∞

We can literally think of kf − fk ku as being the distance from fk to f. If this

distance converges to zero as k increases, then fk converges to f uniformly.
Uniform convergence implies pointwise convergence, but the next example
shows that the converse fails in general.

Example 0.9. We saw in Example 0.6 that the functions χ(0, k1 ) converge point-
wise to the zero function. However, since

sup |χ(0, k1 ) (x) − 0| = 1

x∈R

for every k, the maximum difference between χ(0, k1 ) and the zero function
does not converge to zero. Hence χ(0, k1 ) does not converge uniformly to the
zero function. ♦

The next exercise gives four properties that are satisfied by the uniform
norm. In general, a norm is a function k · k on a vector space V that satisfies
the analogues of these four properties on the space V. The fact that there
is a norm lying behind the notion of uniform convergence makes this type
of convergence very easy to deal with in certain ways. In contrast, pointwise
convergence cannot be characterized in terms of a single norm (Problem 0.5).

Exercise 0.10. Let Fb (X) (b for “bounded”) be the set of all bounded,
complex-valued functions on X, i.e.,

Fb (X) = f : X → C : f is bounded .

This is a vector space over the complex field. Show that uniform norm satisfies
the following properties on Fb (X):
(a) 0 ≤ kf ku ≤ ∞ for all f ∈ Fb (X),
(b) kf ku = 0 if and only if f = 0 (the zero function),
(c) kcf ku = |c| kf ku for all f ∈ Fb (X) and scalars c ∈ C,
(d) kf + gku ≤ kf ku + kgku for all f, g ∈ Fb (X). ♦

Property (b) in the list above is a uniqueness requirement: Only the zero
vector can have zero norm. Property (c) is sometimes called the homogeneity
or positive homogeneity property of the uniform norm, and Property (d) is
usually referred to as the Triangle Inequality for the uniform norm. If we
replace Fb (X) by a different vector space V, then properties (a)–(d) above are
exactly what are required in order for a function on V to be called a norm.
As we stated in equation (0.3), uniform convergence means that the dis-
tance between fk and f shrinks as k increases. Consequently, fk is close to f
 0.2 Uniform Convergence 5

when k is large enough, say within ε in distance for all k > N. Since the
Triangle Inequality implies that

kfj − fk ku ≤ kfj − f k + kf − fk ku , (0.4)

it follows that fj must be close to fk for all j, k large enough (specifically,

the distance will be at most 2ε for all j, k > N ). Therefore, a sequence that
converges uniformly is Cauchy in the following sense.

Definition 0.11 (Uniformly Cauchy). Let X be a set, and let {fk }k∈N be
a sequence of funtions in Fb (X). We say that {fk }k∈N is uniformly Cauchy or
Cauchy in the uniform norm if

∀ ε > 0, ∃ N > 0, ∀ j, k ≥ N, kfj − fk ku < ε. ♦ (0.5)

Sometimes we write limj,k→∞ kfj − fk ku = 0 as an abbreviation for equa-

tion (0.5), but it should be understood that this is not a limit in the usual
sense. Instead, it is only a shorthand for equation (0.5).
It follows from equation (0.4) that every uniformly convergent sequence
in Fb (X) is uniformly Cauchy. The converse question is extremely important.
Indeed, this is a fundamental question whenever we have a norm on a vector
space: If {fk } is a Cauchy sequence, must it converge to some vector in the
space? If the answer is yes, then we say that the space is complete. For example,
Rd and Cd are complete with respect to the usual Euclidean norm. We show
next that Fb (X) is complete with respect to the uniform norm.

Theorem 0.12. If X is a set, then the space Fb (X) is complete with respect
to the uniform norm. That is, if {fk }k∈N is a uniformly Cauchy sequence in
Fb (X), then there exists a function f ∈ Fb (X) such that fk → f uniformly.

Proof. Suppose that {fk }k∈N is a uniformly Cauchy sequence in Fb (X). If we

fix an element x ∈ X, then it follows from the definition of the uniform norm
that
|fj (x) − fk (x)| ≤ kfj − fk ku , all j, k ∈ N.
Consequently {fk (x)}k∈N is a Cauchy sequence of complex scalars. Since C is
complete with respect to the Euclidean norm, this sequence of scalars must
converge to some complex number. Therefore we can define

f (x) = lim fk (x).

k→∞

Doing this for each x ∈ X gives us a function f such that fk → f pointwise.

We still have to show that f belongs to the space Fb (X), and further we have
to show that fk → f uniformly.
Fix an ε > 0. By definition of Cauchy sequence, there must exist an N
such that kfj − fk ku < ε for all j, k > N. If we fix an integer k > N, then for
every x ∈ X we have
6 General Notation

|f (x) − fk (x)| = lim |fj (x) − fk (x)| ≤ kfj − fk ku ≤ ε.

j→∞

Taking the supremum over all x ∈ X, we see that kf − fk ku ≤ ε for all k > N.
This precisely says that fk converges uniformly to f. Finally, by the Triangle
Inequality, for any k we have

kf ku ≤ kf − fk ku + kfk ku < ∞,

so f is bounded and hence belongs to Fb (X). ⊓

⊔

A complete normed vector space is also called a Banach space (see Defi-
nition 0.4), so we have shown that Fb (X) is a Banach space with respect to
the uniform norm.

Remark 0.13. The reader should be aware that the word “complete” is heavily
overused in mathematics. We have already assigned one meaning to this word
when we defined a complete measure, and now we have used the same word
to describe the entirely distinct notion of a complete space. As long as we are
careful to observe the context in which the word “complete” is being used,
this will not be a problem. ♦