Banach spaces: definitions and examples

P. B. Kronheimer, for Math 114

October 27, 2010

1. Normed spaces and Banach spaces

Let F denote either the field R or C. A normed vector space is a vector space X over F
together with a function (the norm),

k · k : X → R≥

satisfying

• kxk = 0 if and only if x = 0;

• kλ xk = |λ|kxk for all x ∈ X and λ ∈ F;

• kx + yk ≤ kxk + kyk.

These conditions mean in particular that we can define a distance function d(x , y) =
kx − yk and obtain a metric space (X , d). So in a normed space, we can talk about
all the things that go with metric spaces: open and closed sets, Cauchy sequences and
convergence, and so on. In particular, a set S ⊂ X is open if for all x ∈ X there exists a
δ > 0 such that the ball B(x ; δ ) is entirely contained in S. Here

B(x ; δ ) = { x | kx − yk ≤ δ }.

And a sequence (x n )n∈N is a Cauchy sequence if for all  > 0, there exists n(  ) such that
for all m , n ≥ n(  ) we have
kx m − x n k ≤ .
2 2. A criterion for completeness

As with any metric space, we say that a normed space is complete if every Cauchy se-
quence (x n )n∈N converges: that is, there exists x ∈ X with

kx n − xk → 0, as n → ∞.

We write x n → x as n → ∞.

Definition 1.1. A Banach space (over R or C) is a complete normed vector space.

If X is a vector space with two different norms, k · k and k · k0 , then the norms are said
to be equivalent if there are positive constants C1 and C2 such that

C1 kxk ≤ kxk0 ≤ C2 kxk

for all x ∈ X . If two norms are equivalent, then Cauchy sequences for one norm are
Cauchy sequences for the other. The same goes for convergent sequences. So one of
the norms makes X complete if and only if the other does.

2. A criterion for completeness

The following criterion for completeness of a normed space is often useful. (We used it
first in proving that L 1 is complete.) If X is a normed space and vn , n ∈ N, are elements
of X , then we say that the series is absolutely summable if the series
X
kvn k
n

is convergent. We say that the series is convergent if the sequence of partial sums converges
P
in X : that is, if sn = m≤n vm , then we ask that there exists s ∈ X with sn → s as n → ∞.
Then we have the following:

Lemma 2.1. A normed space X is complete if and only if every absolutely summable series is
convergent.

Proof. We do the “if ” direction (which is the more interesting one). Suppose every
absolutely summable series in X converges. Let x n , n ∈ N be a Cauchy sequence. To
show that X is complete, we must show that the sequence x n converges. For each j ≥ 1,
we can find n j such that
kx n − x n0 k ≤ 2− j
 3

for all n, n 0 ≥ n j . Define

v 1 = x n1
and define
v j = x n j − x n j −1
for j ≥ 2. These definitions achieve the following. We have
kv j k ≤ 2− j
P
so that the series v j is absolutely summable, and hence convergent, by our hypothesis.
We also have arranged that the partial sum
X
sj = vi
i< j
P
is equal to x n j . Because the series v j is convergent, we have
s j → s,
as j → ∞. In other words, x n j → s as j → ∞. This is saying that a subsequence of the
original Cauchy sequence {x n }n∈N converges. But it follows easily that the original
sequence converges. Indeed, given  > 0, we can find n∗ so that
kx n − x n0 k ≤  /2
for all n, n 0 ≥ n∗ ; and we can find j so that n j ≥ n∗
kx n j − sk ≤  /2.
From these and the triangle inequality, it follows that kx n − sk ≤  for all n ≥ n∗ .

3. Some examples

We give some examples of Banach spaces. In some of these examples, there are things
we need to verify to check that we have a Banach space: the triangle inequality for the
norm, and completeness. We omit these proofs. For the interesting case of L p ( Rd ), we
will give proofs in a separate handout. We again use F to denote

The space c0 . This is the space of all sequence a = (an )n∈N with an ∈ F such that an → 0
as n → ∞. The norm is
kak∞ = sup |an |.
n
4 3. Some examples

The space `∞ . This is the space of all bounded sequences a = (an )n∈N with an ∈ F. The
norm is
kak = sup |an |.
n

The space ` p . For 1 ≤ p < ∞, this is the space of all sequences a = (an )n∈N , with an ∈ F,
such that
∞
X
|an | p
n =1

is convergent. The norm is

∞
!1/p
X
kak p = |an | p .
n =1

The triangle inequality for ` p is a version of the Minkowski inequality.

All these sequence spaces can be considered with sequences indexed by any countable
set in place of N. For example, we will write ` p ( Z ) to denote the space of sequences
(an ) indexed by n ∈ Z such that ∞ p
P
−∞ |an | is finite.

The space C(Ω). If Ω is a compact metric space, C( Ω ) is the space of continuous func-
tions on Ω with values in F. The norm is the sup norm:

k f k0 = sup | f ( ω )|.
ω∈Ω

The space C0 (Rd ). This is the space of continuous functions f : Rd → F satisfying

| f (x)| → 0 as |x| → ∞. The norm is the sup norm, sup x ∈Rd | f (x)|.

The space L p (Rd ). For 1 < p < ∞, the Lebesgue space L p ( Rd ) is the space of equivalence
classes of measurable functions f : Rd → R such that | f | p is integrable. The norm is
Z 1/p
p
k f kp = |f| .
 5

The space L ∞ (Rd ). A measurable function f is essentially bounded if there exists M such
that | f | ≤ M almost everywhere. The space of such essentially bounded functions is a
Banach space, with the norm being

k f k∞ = inf{ M : | f | ≤ M a.e. }.

The space F. The field F = R or C is itself a Banach space, with norm equal to the
absolute value or complex modulus |x|.

The space Fn . The space Rn or Cn is a Banach space, and can be given a variety of
norms. For example, the norms from the sequence spaces ` p for 1 ≤ p ≤ ∞ all give
rise to norms of a similar sort on Fn . Any two norms on a finite-dimensional vector
space are equivalent. (This is outlined in a problem set.) A norm on any vector space is
completely determined by describing the set of vectors of length less than or equal to 1:
the closed unit ball about the origin. In the case of Rn , the unit ball for any norm has
to be a compact, convex subset of Rn which is invariant under the symmetry x 7→ − x.
Any such convex set corresponds to a norm.

4. Bounded linear operators.

Let X and Y be normed spaces. A linear map T : X → Y is bounded if there exists an

M ≥ 0 such that
kT xkY ≤ M kxkX
for all x ∈ X . The smallest such M is the norm or operator norm of T , and is written kT k.
As long as X is non-zero, the norm kT k can be described variously as

kT xk
kT k = sup
x 6=0 kxk

or as
kT k = sup kT xk.
kxk=1

We write L(X , Y ) for the set of all bounded linear operators from X to Y .

Lemma 4.1. Equipped with the operator norm, L(X , Y ) is a normed linear space.
6 4. Bounded linear operators.

Proof. We must check that if T1 and T2 are bounded linear operators, then so are λT1
and T1 + T2 . This is straightforward, and the verification shows

kT1 + T2 k ≤ kT1 k + kT2 k

and
kλT1 k = |λ|kT1 k.
For example, for the first of these, we just use

k(T1 + T2 )xk = kT1 x + T2 xk

≤ kT1 xk + kT2 xk
≤ kT1 kkxk + kT2 kkxk
= kT1 k + kT2 k kxk.

Finally we must note that kT k = 0 only if T = 0, which is clear from the definition.

So far, we have not assumed that either X or Y is complete. We now look at complete-
ness:
Lemma 4.2. If Y is complete, then so is the normed space L(X , Y ) with the operator norm. That
is, L(X , Y ) is a Banach space.

Proof. Suppose Y is complete (a Banach space). Let Tn , n ∈ N, be a Cauchy sequence in

L(X , Y ). This means that for all  > 0 we can find n(  ) such that

kTn − Tm k ≤ 

for all n , m ≥ n(  ). We will show that Tn converges to some T ∈ L(X , Y ) in operator

norm.

For each fixed x ∈ X , we can consider the sequence Tn x. This is a Cauchy sequence in
Y , because
kTn x − Tm xk ≤  kxk (1)
for all n , m ≥ n(  ). So for each x ∈ X there exists a unique y in Y such that

Tn x → y .

Define T : X → Y by T x = y. We omit the easy verification that T is a linear map. We

must check that T is bounded. To do this, we take the limit as n → ∞ in (1) to get

kT x − Tm xk ≤  kxk (2)
 7

for m ≥ n(  ), and then we use the triangle inequality to deduce

kT xk ≤ kTm kkxk +  kxk.

This shows that T is bounded, with kT k ≤ kTm k +  , for any m ≥ n(  ). The inequality (2)
also shows that kT − Tm k ≤  for m ≥ n(  ), so Tm → T in operator norm as m → ∞.

If there are mutually inverse bounded linear maps T : X → Y and S : Y → X , then the
Banach spaces X and Y are equivalent. If T (and hence S also) preserves the norm, so
that kxk = kT xk for all x, then X and Y are isometrically isomorphic.

5. Dual spaces

If X is a normed space over F = R or C, then we can consider the space L(X , F ). By the
previous lemma, this is a Banach space. It is called the dual space of X . We write it X ∗ .
Here are some examples, without proof. Some proofs will appear in another handout.

• The dual space of c 0 is (isometrically isomorphic to) `1 . The isomorphism is given

as follows. For a = (an )in`1 , we define an element α ∈ (c 0 )∗ by
X
α (b) = an b n .
n

• The dual space of `1 is (isometrically isomorphic to) `∞ .

• The dual space of ` p for 1 < p < ∞ is isometrically isomorphic to `q , where q

satisfies 1/p + 1/q = 1 (the “dual exponent”). In particular `2 is its own dual.

• The dual space of C0 ( Rd ) is isometrically isomorphic to L 1 ( Rd ). The isomorphism

is as follows. Given a in L 1 ( Rd ), we define α ∈ C0 ( Rd )∗ by
Z
α( f ) = f
Rd

for f ∈ C0 ( Rd ).

• The dual space of L 1 ( Rd ) is isometrically isomorphic to L ∞ ( Rd ).

• For 1 < p < ∞, the dual space of L p ( Rd ) is L q ( Rd ), where q is the dual exponent.
8 6. Dense subsets

• The dual space of L ∞ ( Rd ) is something rather large and unmanageable. An el-

ement α in the dual space of L ∞ ([0, 1]) is completely determined by specifying
its value on the χ E for all measurable subsets E of [0, 1]. These values n E ∈ R or
C are subject only to the constraint that they be additive under finite unions of
disjoint sets: n E1 + n E2 = n E1 ∪E2 if E 1 ∩ E 2 is null.

6. Dense subsets

A subset S ⊂ X in normed space X is dense if every point of X is a limit point of points

of X . An infinite-dimensional Banach space may have dense linear subspaces. Here are
some examples:
Proposition 6.1. In the Lebesgue space L p ( Rd ), for 1 ≤ p < ∞, the following subspaces are dense.

• The functions f of bounded support: those that vanish outside a bounded set {|x| ≤ R}.
• The bounded functions of bounded support.
• The simple functions.
• The step functions.
• The continuous functions of bounded support.
• The functions of Schwartz class.

In proving that subspaces such as these are dense in L p , the following is useful:
Proposition 6.2 (Dominated convergence theorem, formulated for L p .). Let fn be a sequence in
L p ( Rd ) with fn → f almost everywhere. Suppose that there exists g in L p ( Rd ), with | fn | ≤ g
almost everywhere for all n. Then f is in L p and
k f − fn k p → 0
as n → ∞.

A Banach space is separable if it contains a countable dense subset. The Lebesgue space
L p is separable for 1 ≤ p < ∞, because a dense subset is the set of step functions
N
X
g= an χ R n
n =1
 9

where the coefficients an are rational numbers and the rectangles Rn have rational coor-
dinates for their vertices. This does not work for L ∞ . You will show in the problem sets
that L ∞ is not separable.