Normed and Banach spaces

In this chapter we introduce the basic setting of functional analysis, in the form of normed
spaces and bounded linear operators. We are particularly interested in complete, i.e. Banach,
spaces and the process of completion of a normed space to a Banach space.

1. Vector spaces

Note that for us the ‘scalars’ are either the real numbers or the complex numbers – usually
the latter. To be neutral we denote by K either R or C, but of course consistently. Then our set
V – the set of vectors with which we will deal, comes with two ‘laws’. These are maps
(1.1) + : V × V −→ V, · : K × V −→ V.

which we denote not by +(v,w) and ·(s,v) but by v+w and sv. Then we impose the axioms of a
vector space – see Section (14) below! These are commutative group axioms for +, axioms for
the action of K and the distributive law linking the two.
The basic examples:
• The field K which is either R or C is a vector space over itself.
The vector spaces Kn consisting of ordered n-tuples of elements of K. Addition is by
•
components and the action of K is by multiplication on all components. You should
be reasonably familiar with these spaces and other finite dimensional vector spaces.
• Seriously non-trivial examples such as C([0,1]) the space of continuous functions on
[0,1] (say with complex values).
In these and many other examples we will encounter below the ‘component addition’
corresponds to the addition of functions.
9

Lemma 1. If X is a set then the spaces of all functions

(1.2) F(X;R) = {u : X −→ R}, F(X;C) = {u : X −→ C}

are vector spaces over R and C respectively.

The main point to make sure one understand is that, because we do know how to add and
multiply in either R and C, we can add functions and multiply them by constants (we can
multiply functions by each other but that is not part of the definition of a vector space so we
ignore it for the moment since many of the spaces of functions we consider below are not
multiplicative in this sense):-
(1.3) (c1f1 + c2f2)(x) = c1f1(x) + c2f2(x) defines the function c1f1 + c2f2

if c1, c2 ∈ K and f1, f2 ∈ F(X;K).

 Although you are probably most comfortable with finite-dimensional vector spaces it is
the infinite-dimensional case that is most important here. The notion of dimension is based
on the concept of the linear independence of a subset of a vector space. Thus a subset E ⊂ V
is said to be linearly independent if for any finite collection of elements vi ∈ E, i = 1,...,N, and
any collection of ‘constants’ ai ∈ K, i = 1,...,N we have the following implication
N

(1.4) Xaivi = 0 =⇒ ai = 0 ∀ i.
i=1

That is, it is a set in which there are ‘no non-trivial finite linear dependence relations between
the elements’. A vector space is finite-dimensional if every linearly independent subset is
finite. It follows in this case that there is a finite and maximal linearly independent subset –
a basis – where maximal means that if any new element is added to the set E then it is no
longer linearly independent. A basic result is that any two such ‘bases’ in a finite dimensional
vector space have the same number of elements – an outline of the finite-dimensional theory
can be found in Problem 1.

Definition 1. A vector space is infinite-dimensional if it is not finite dimensional, i.e. for

any N ∈ N there exist N elements with no, non-trivial, linear dependence relation between
them.
As is quite typical the idea of an infinite-dimensional space, which you may be quite keen to
understand, appears just as the non-existence of something. That is, it is the ‘residual’ case,
where there is no finite basis. This means that it is ‘big’.
2. NORMED SPACES

So, finite-dimensional vector spaces have finite bases, infinite-dimensional vector spaces
do not. The notion of a basis in an infinite-dimensional vector spaces needs to be modified to
be useful analytically. Convince yourself that the vector space in Lemma 1 is infinite
dimensional if and only if X is infinite.
2. Normed spaces

In order to deal with infinite-dimensional vector spaces we need the control given by a
metric (or more generally a non-metric topology, but we will not quite get that far). A norm
on a vector space leads to a metric which is ‘compatible’ with the linear structure.
Definition 2. A norm on a vector space is a function, traditionally denoted
(1.5) k · k : V −→ [0,∞),
with the following properties
(Definiteness)
(1.6) v ∈ V, kvk = 0 =⇒ v = 0.
(Absolute homogeneity) For any λ ∈ K and v ∈ V,
(1.7) kλvk = |λ|kvk.
(Triangle Inequality) The triangle inequality holds, in the sense that for any two elements v,
w∈V
(1.8) kv + wk ≤ kvk + kwk.
Note that (1.7) implies that k0k = 0. Thus (1.6) means that kvk = 0 is equivalent to v = 0.
This definition is based on the same properties holding for the standard norm(s), |z|, on R
and C. You should make sure you understand that
( x if x ≥ 0
R 3 x −→ |x| = ∈ [0,∞) is a norm as is
(1.9) −x if x ≤ 0
.
Situations do arise in which we do not have (1.6):-
Definition 3. A function (1.5) which satisfes (1.7) and (1.8) but possibly not (1.6) is called
a seminorm.
A metric, or distance function, on a set is a map
(1.10) d : X × X −→ [0,∞)
satisfying three standard conditions
(1.11) d(x,y) = 0 ⇐⇒ x = y,
(1.12) d(x,y) = d(y,x) ∀ x,y ∈ X and
(1.13) d(x,y) ≤ d(x,z) + d(z,y) ∀ x,y,z ∈ X.
The point of course is Proposition 1. If k · k is a norm on V then
(1.14) d(v,w) = kv − wk is a distance on V turning it into a metric
space.
Proof. Clearly (1.11) corresponds to (1.6), (1.12) arises from the special case λ = −1 of
(1.7) and (1.13) arises from (1.8).
We will not use any special notation for the metric, nor usually mention it explicitly – we
just subsume all of metric space theory from now on. So kv − wk is the distance between two
points in a normed space.
Now, we need to talk about a few examples; there are more in Section 7. The most basic
ones are the usual finite-dimensional spaces Rn and Cn with their
Euclidean norms

(1.15)
where it is at first confusing that we just use single bars for the norm, just as for R and C, but
you just need to get used to that.
 There are other norms on Cn (I will mostly talk about the complex case, but the real case
is essentially the same). The two most obvious ones are |x|∞ = max|xi|, x = (x1,...,xn) ∈ Cn,

(1.16) |x|1 = X|xi|

i

but as you will see (if you do the problems) there are also the norms

(1.17) .
In fact, for p = 1, (1.17) reduces to the second norm in (1.16) and in a certain sense the case
p = ∞ is consistent with the first norm there.
In lectures I usually do not discuss the notion of equivalence of norms straight away.
However, two norms on the one vector space – which we can denote k · k(1) and k · k(2) are
equivalent if there exist constants C1 and C2 such that
(1.18) kvk(1) ≤ C1kvk(2), kvk(2) ≤ C2kvk(1) ∀ v ∈ V.
The equivalence of the norms implies that the metrics define the same open sets – the
topologies induced are the same. You might like to check that the reverse is also true, if two
norms induced the same topologies (just meaning the same collection of open sets) through
their associated metrics, then they are equivalent in the sense of (1.18) (there are more
efficient ways of doing this if you wait a little).

One important class of normed spaces consists of the spaces of bounded continuous
functions on a metric space X :
(1.19) C∞(X) = C∞(X;C) = {u : X −→ C, continuous and bounded}.
3. BANACH SPACES

That this is a linear space follows from the (obvious) result that a linear combination of
bounded functions is bounded and the (less obvious) result that a linear combination of
continuous functions is continuous; this we know. The norm is the best bound
(1.20) kuk∞ = sup |u(x)|.
x∈X

That this is a norm is straightforward to check. Absolute homogeneity is clear, kλuk∞ =

|λ|kuk∞ and kuk∞ = 0 means that u(x) = 0 for all x ∈ X which is exactly what it means for a
function to vanish. The triangle inequality ‘is inherited from C’ since for any two functions
and any point,
(1.21) |(u + v)(x)| ≤ |u(x)| + |v(x)| ≤ kuk∞ + kvk∞
by the definition of the norms, and taking the supremum of the left gives
ku + vk∞ ≤ kuk∞ + kvk∞.
Of course the norm (1.20) is defined even for bounded, not necessarily continuous
functions on X. Note that convergence of a sequence un ∈ C∞(X) (remember this means with
respect to the distance induced by the norm) is precisely uniform convergence
(1.22) kun − vk∞ → 0 ⇐⇒ un(x) → v(x) uniformly on X.
Other examples of infinite-dimensional normed spaces are the spaces lp, 1 ≤ p ≤ ∞
discussed in the problems below. Of these l2 is the most important for us. It is in fact one form
of Hilbert space, with which we are primarily concerned:-

(1.23) l2 = {a : N −→ C;X|a(j)|2 < ∞}.

j

It is not immediately obvious that this is a linear space, nor that

(1.24)
is a norm. It is. From now on we will generally use sequential notation and think of a map
from N to C as a sequence, so setting a(j) = aj. Thus the ‘Hilbert space’ l2 consists of the square
summable sequences.
3. Banach spaces

Definition 4. A normed space which is complete with respect to the induced metric is a
Banach space.
Lemma 2. The space C∞(X), defined in (1.19) for any metric space X, is a Banach space.
Proof. This is a standard result from metric space theory – basically that the uniform limit
of a sequence of (bounded) continuous functions on a metric space is continuous. However,
it is worth recalling how one proves completeness at least in outline. Suppose un is a Cauchy
sequence in C∞(X). This means that given δ > 0 there exists N such that
(1.25) n,m > N =⇒ kun − umk∞ = sup|un(x) − um(x)| < δ.
X

Fixing x ∈ X this implies that the sequence un(x) is Cauchy in C. We know that this space is
complete, so each sequence un(x) must converge (we say the sequence of functions converges
pointwise). Since the limit of un(x) can only depend on x, we define u(x) = limn un(x) in C for
each x ∈ X and so define a function u : X −→ C. Now, we need to show that this is bounded and
continuous and is the limit of un with respect to the norm. Any Cauchy sequence is bounded
in norm – take δ = 1 in (1.25) and it follows from the triangle inequality that
(1.26) kumk∞ ≤ kuN+1k∞ + 1, m > N
and the finite set kunk∞ for n ≤ N is certainly bounded. Thus kunk∞ ≤ C, but this means |un(x)|
≤ C for all x ∈ X and hence |u(x)| ≤ C by properties of convergence in C and thus kuk∞ ≤ C.
The uniform convergence of un to u now follows from (1.25) since we may pass to the
limit in the inequality to find
n > N =⇒ |un(x) − u(x)| = lim |un(x) − um(x)| ≤ δ
(1.27) m→∞
 =⇒ kun − uk|infty ≤ δ.
The continuity of u at x ∈ X follows from the triangle inequality in the form

|u(y) − u(x)| ≤ |u(y) − un(y)| + |un(y) − un(x)| + |un(x) − un(x)|

≤ 2ku − unk∞ + |un(x) − un(y)|.
Give δ > 0 the first term on the far right can be make less than δ/2 by choosing n large using
(1.27) and then the second term can be made less than δ/2 by choosing d(x,y) small enough.

There is a space of sequences which is really an example of this Lemma. Consider the
space c0 consisting of all the sequences {aj} (valued in C) such that limj→∞ aj = 0. As remarked
above, sequences are just functions N −→ C. If we make {aj} into a function α : D =
{1,1/2,1/3,...} −→ C by setting α(1/j) = aj then we get a function on the metric space D. Add 0
to D to get
D = D ∪ {0} ⊂ [0,1] ⊂ R; clearly 0 is a limit point of D and D is, as the notation dangerously
indicates, the closure of D in R. Now, you will easily check (it is really the definition) that α :
D −→ C corresponding to a sequence, extends to a

continuous function on D vanishing at 0 if and only if limj→∞ aj = 0, which is to say, {aj} ∈ c0.
Thus it follows, with a little thought which you should give it, that c0 is a Banach space with
the norm
(1.28) kak∞ = supkajk.
j

3. BANACH SPACES

What is an example of a non-complete normed space, a normed space which is not a

Banach space? These are legion of course. The simplest way to get one is to ‘put the wrong
norm’ on a space, one which does not correspond to the definition. Consider for instance the
linear space T of sequences N −→ C which ‘terminate’, i.e. each element {aj} ∈ T has aj = 0 for
j > J, where of course the J may depend on the particular sequence. Then T ⊂ c0, the norm on
c0 defines a norm on T but it cannot be complete, since the closure of T is easily seen to be all
of c0 – so there are Cauchy sequences in T without limit in T . Make sure you are not lost here
– you need to get used to the fact that we often need to discuss the ‘convergence of sequences
of convergent sequences’ as here.
One result we will exploit below, and I give it now just as preparation, concerns absolutely
summable series. Recall that a series is just a sequence where we ‘think’ about adding the
terms. Thus if vn is a sequence in some vector space V then there N

is the corresponding sequence of partial sums wN = P vi. I will say that {vn} is a
i=1
series if I am thinking about summing it.
So a sequence {vn} with partial sums {wN} is said to be absolutely summable if

(1.29), i.e. X kwN − wN−1kV < ∞.

N>1

Proposition 2. The sequence of partial sums of any absolutely summable series in a normed
space is Cauchy and a normed space is complete if and only if every absolutely summable series
in it converges, meaning that the sequence of partial sums converges.
Proof. The sequence of partial sums is

(1.30) .
Thus, if m ≥ n then

(1.31) .
It follows from the triangle inequality that

(1.32) .
So if the series is absolutely summable then

and lim .
Thus {wn} is Cauchy if {vj} is absolutely summable. Hence if V is complete then every
absolutely summable series is summable, i.e. the sequence of partial sums converges.
Conversely, suppose that every absolutely summable series converges in this sense. Then
we need to show that every Cauchy sequence in V converges. Let un be a Cauchy sequence. It
suffices to show that this has a subsequence which converges, since a Cauchy sequence with
a convergent subsequence is convergent.
To do so we just proceed inductively. Using the Cauchy condition we can for every k find an
integer Nk such that

(1.33) n,m > Nk =⇒ kun − umk < 2−k.

Now choose an increasing sequence nk where nk > Nk and nk > nk−1 to make it increasing. It
follows that

(1.34) kunk − unk−1k ≤ 2−k+1.

Denoting this subsequence as u0k = unk it follows from (1.34) and the triangle
inequality that
∞

(1.35) X 0n − u0n−1k ≤ 4
ku n=1
so the sequence , is absolutely summable. Its sequence of
partial sums is so the assumption is that this converges, hence the original Cauchy
sequence converges and V is complete.

Notice the idea here, of ‘speeding up the convergence’ of the Cauchy sequence by
dropping a lot of terms. We will use this idea of absolutely summable series heavily in the
discussion of Lebesgue integration.

4. Operators and functionals

The vector spaces we are most interested in are, as already remarked, spaces of functions
(or something a little more general). The elements of these are the objects of primary interest
but they are related by linear maps. A map between two vector spaces (over the same field,
for us either R or C) is linear if it takes linear combinations to linear combinations:-

(1.36) T : V −→ W, T(a1v1+a2v2) = a1T(v1)+a2T(v2), ∀ v1, v2 ∈ V, a1,a2 ∈ K.

The sort of examples we have in mind are differential, or more especially, integral operators.
For instance if u ∈ C([0,1]) then its indefinite Riemann integral

(1.37) (
is continuous in x ∈ [0,1] and so this defines a map

(1.38) T : C([0,1]) −→ C([0,1]).

This is a linear map, with linearity being one of the standard properties of the Riemann
integral.
In the finite-dimensional case linearity is enough to allow maps to be studied. However
in the case of infinite-dimensional normed spaces we need to impose continuity. Of course it
makes perfectly good sense to say, demand or conclude, that a map as in (1.36) is continuous
if V and W are normed spaces since they are then
4. OPERATORS AND FUNCTIONALS

metric spaces. Recall that for metric spaces there are several different equivalent conditions
that ensure a map, T : V −→ W, is continuous:
(1.39) vn → v in V =⇒ Tvn → Tv in W
(1.40) O ⊂ W open =⇒ T−1(O) ⊂ V open
(1.41) C ⊂ W closed =⇒ T−1(C) ⊂ V closed.
For a linear map between normed spaces there is a direct characterization of continuity
in terms of the norm.
Proposition 3. A linear map (1.36) between normed spaces is continuous if and only if it is
bounded in the sense that there exists a constant C such that
(1.42) kTvkW ≤ CkvkV ∀ v ∈ V.
Of course bounded for a function on a metric space already has a meaning and this is not it!
The usual sense would be kTvk ≤ C but this would imply kT(av)k = |a|kTvk ≤ C so Tv = 0.
Hence it is not so dangerous to use the term ‘bounded’ for (1.42) – it is really ‘relatively
bounded’, i.e. takes bounded sets into bounded sets. From now on, bounded for a linear map
means (1.42).
Proof. If (1.42) holds then if vn → v in V it follows that kTv − Tvnk = kT(v − vn)k ≤ Ckv − vnk
→ 0 as n → ∞ so Tvn → Tv and continuity follows.
For the reverse implication we use the second characterization of continuity above.
Denote the ball around v ∈ V of radius > 0 by
.
Thus if T is continuous then the inverse image of the the unit ball around the origin,
T−1(BW(0,1)) = {v ∈ V ;kTvkW < 1}, contains the origin in V and so, being open, must contain
some . This means that
(1.43) .
Now proceed by scaling. If 0 6= v ∈ V then where shows that kTv0k
≤ 1 but this implies (1.42) with – it is trivially true if v = 0.
As a general rule we drop the distinguishing subscript for norms, since which norm we
are using can be determined by what it is being applied to.
So, if T : V −→ W is continous and linear between normed spaces, or from now on
‘bounded’, then
(1.44) kTk = sup kTvk < ∞.
kvk=1

Lemma 3. The bounded linear maps between normed spaces V and W form a linear space
B(V,W) on which kTk defined by (1.44) or equivalently
(1.45) kTk = inf{C; (1.42) holds} is a norm.
Proof. First check that (1.44) is equivalent to (1.45). Define kTk by (1.44). Then for any v
∈ V, v 6= 0,

(1.46)
since as always this is trivially true for v = 0. Thus C = kTk is a constant for which (1.42) holds.
Conversely, from the definition of 0 then there exists v ∈ V with kvk = 1 such that
for any C for which (1.42) holds. Since
> 0 is arbitrary, kTk ≤ C and hence kTk is given by (1.45).
From the definition of kTk, kTk = 0 implies Tv = 0 for all v ∈ V and for λ 6= 0,
(1.47) kλTk = sup kλTvk = |λ|kTk
kvk=1

and this is also obvious for λ = 0. This only leaves the triangle inequality to check and for any
T, S ∈ B(V,W), and v ∈ V with kvk = 1
(1.48) k(T + S)vkW = kTv + SvkW ≤ kTvkW + kSvkW ≤ kTk + kSk
so taking the supremum, kT + Sk ≤ kTk + kSk.

Thus we see the very satisfying fact that the space of bounded linear maps between two
normed spaces is itself a normed space, with the norm being the best constant in the estimate
(1.42). Make sure you absorb this! Such bounded linear maps between normed spaces are
often called ‘operators’ because we are thinking of the normed spaces as being like function
spaces.
You might like to check boundedness for the example of a linear operator in (1.38),
namely that in terms of the supremum norm on C([0,1]), kTk ≤ 1.
One particularly important case is when W = K is the field, for us usually C. Then a simpler
notation is handy and one sets V 0 = B(V,C) – this is called the dual space of V (also sometimes
denoted V ∗.)
Proposition 4. If W is a Banach space then B(V,W), with the norm (1.44), is a Banach space.
Proof. We simply need to show that if W is a Banach space then every Cauchy sequence
in B(V,W) is convergent. The first thing to do is to find the limit. To say that Tn ∈ B(V,W) is
Cauchy, is just to say that given > 0 there exists N such that n, m > N implies By
the definition of the norm, if v ∈ V
then kTnv − TmvkW ≤ kTn − TmkkvkV so Tnv is Cauchy in W for each v ∈ V. By assumption, W is
complete, so
(1.49) Tnv −→ w in W.
However, the limit can only depend on v so we can define a map T : V −→ W by
Tv = w = limn→∞ Tnv as in (1.49).
This map defined from the limits is linear, since Tn(λv) = λTnv −→ λTv and
Tn(v1 +v2) = Tnv1 +Tnv2 −→ Tv2 +Tv2 = T(v1 +v2). Moreover, |kTnk−kTmk| ≤ kTn − Tmk so kTnk is
Cauchy in [0,∞) and hence converges, with limit S, and
(1.50) kTvk = lim kTnvk ≤ Skvk n→∞
so kTk ≤ S shows that T is bounded.
Returning to the Cauchy condition above and passing to the limit in kTnv −
shows that and hence Tn → T in
B(V,W) which is therefore complete.

Note that this proof is structurally the same as that of Lemma 2. One simple
consequence of this is:-
5. SUBSPACES AND QUOTIENTS

Corollary 1. The dual space of a normed space is always a Banach space.

However we should be a little suspicious here since we have not shown that the dual
space V 0 is non-trivial, meaning we have not eliminated the possibility that V 0 = {0} even
when V 6= {0}. The Hahn-Banach Theorem, discussed below, takes care of this.
 One game you can play is ‘what is the dual of that space’. Of course, the dual is the dual,
but you may well be able to identify the dual space of V with some other Banach space by
finding a linear bijection between V 0 and the other space, W, which identifies the norms as
well. We will play this game a bit later.
5. Subspaces and quotients

The notion of a linear subspace of a vector space is natural enough, and you are likely
quite familiar with it. Namely W ⊂ V where V is a vector space is a (linear) subspace if any
linear combinations λ1w1 + λ2w2 ∈ W if λ1, λ2 ∈ K and w1, w2 ∈ W. Thus W ‘inherits’ its linear
structure from V. Since we also have a topology from the metric we will be especially
interested in closed subspaces. Check that you understand the (elementary) proof of
Lemma 4. A subspace of a Banach space is a Banach space in terms of the restriction of the
norm if and only if it is closed.
There is a second very important way to construct new linear spaces from old. Namely
we want to make a linear space out of ‘the rest’ of V, given that W is a linear subspace. In finite
dimensions one way to do this is to give V an inner product and then take the subspace
orthogonal to W. One problem with this is that the result depends, although not in an
essential way, on the inner product. Instead we adopt the usual ‘myopia’ approach and take
an equivalence relation on V which identifies points which differ by an element of W. The
equivalence classes are then ‘planes parallel to W’. I am going through this construction
quickly here under the assumption that it is familiar to most of you, if not you should think
about it carefully since we need to do it several times later.
So, if W ⊂ V is a linear subspace of V we define a relation on V – remember this is just a
subset of V × V with certain properties – by
(1.51) v ∼W v0 ⇐⇒ v − v0 ∈ W ⇐⇒ ∃ w ∈ W s.t. v = v0 + w.
This satisfies the three conditions for an equivalence relation:
(1) v ∼W v
(2) v ∼W v0 ⇐⇒ v0 ∼W v
(3) v ∼W v0, v0 ∼W w00 =⇒ v ∼W v00
which means that we can regard it as a ‘coarser notion of equality.’
Then V/W is the set of equivalence classes with respect to ∼W . You can think of the
elements of V/W as being of the form v + W – a particular element of V plus an arbitrary
element of W. Then of course v0 ∈ v+W if and only if v0 −v ∈ W meaning v ∼W v0.
The crucial point here is that
(1.52) V/W is a vector space.
You should check the details – see Problem 1. Note that the ‘is’ in (1.52) should really be
expanded to ‘is in a natural way’ since as usual the linear structure is inherited from V :
(1.53) λ(v + W) = λv + W, (v1 + W) + (v2 + W) = (v1 + v2) + W.
The subspace W appears as the origin in V/W.
 Now, two cases of this are of special interest to us.
Proposition 5. If k · k is a seminorm on V then
(1.54) E = {v ∈ V ;kvk = 0} ⊂ V
is a linear subspace and
(1.55) kv + EkV/E = kvk
defines a norm on V/E.
Proof. That E is linear follows from the properties of a seminorm, since kλvk = |λ|kvk
shows that λv ∈ E if v ∈ E and λ ∈ K. Similarly the triangle inequality shows that v1 + v2 ∈ E if
v1, v2 ∈ E.
To check that (1.55) defines a norm, first we need to check that it makes sense as a
function k · kV/E −→ [0,∞). This amounts to the statement that kv0k is the same for all elements
v0 = v + e ∈ v + E for a fixed v. This however follows from the triangle inequality applied twice:
(1.56) kv0k ≤ kvk + kek = kvk ≤ kv0k + k − ek = kv0k.

Now, I leave you the exercise of checking that k·kV/E is a norm, see Problem 1.

The second application is more serious, but in fact we will not use it for some time so I
usually do not do this in lectures at this stage.
Proposition 6. If W ⊂ V is a closed subspace of a normed space then
(1.57)
defines a norm on V/W; if V is a Banach space then so is V/W.
For the proof see Problems 1 and 1.

6. Completion

A normed space not being complete, not being a Banach space, is considered to be a defect
which we might, indeed will, wish to rectify.
Let V be a normed space with norm k · kV . A completion of V is a Banach space B with the
following properties:-
(1) There is an injective (i.e. 1-1) linear map I : V −→ B
(2) The norms satisfy
(1.58) kI(v)kB = kvkV ∀ v ∈ V.
(3) The range I(V ) ⊂ B is dense in B.
Notice that if V is itself a Banach space then we can take B = V with I the identity map.
So, the main result is:
Theorem 1. Each normed space has a completion.
There are several ways to prove this, we will come across a more sophisticated one (using
the Hahn-Banach Theorem) later. In the meantime I will give two proofs. In the first the fact
that any metric space has a completion in a similar sense is recalled and then it is shown that
the linear structure extends to the completion. A second, ‘hands-on’, proof is also outlined
with the idea of motivating the construction of the Lebesgue integral – which is in our near
future.
Proof 1. One of the neater proofs that any metric space has a completion is to use Lemma
2. Pick a point in the metric space of interest, p ∈ M, and then define a map
(1.59) M 3 q 7−→ fq ∈ C∞(M), fq(x) = d(x,q) − d(x,p) ∀ x ∈ M.
That fq ∈ C∞(M) is straightforward to check. It is bounded (because of the second term) by
the reverse triangle inequality
|fq(x)| = |d(x,q) − d(x,p)| ≤ d(p,q)
and is continuous, as the difference of two continuous functions. Moreover the distance
between two functions in the image is
(1.60) sup |fq(x) − fq0(x)| = sup |d(x,q) − d(x,q0)| = d(q,q0)
x∈M x∈M using the reverse triangle inequality (and evaluating at x = q). Thus

the map (1.59) is well-defined, injective and even distance-preserving. Since C∞0 (M) is
complete, the closure of the image of (1.59) is a complete metric space, X, in which M can be
identified as a dense subset.
Now, in case that M = V is a normed space this all goes through. The disconcerting thing
is that the map q −→ fq is not linear. Nevertheless, we can give X a linear structure so that it
becomes a Banach space in which V is a dense linear subspace. Namely for any two elements
fi ∈ X, i = 1,2, define

(1.61) λ1f1 + λ2f2 = lim fλ1p +λ q n→∞

n 2 n

where pn and qn are sequences in V such that fpn → f1 and fqn → f2. Such sequences exist by the
construction of X and the result does not depend on the choice of sequence – since if p0n is
another choice in place of pn then fp0n − fpn → 0 in X (and similarly for qn). So the element of the
left in (1.61) is well-defined. All of the properties of a linear space and normed space now
follow by continuity from V ⊂ X and it also follows that X is a Banach space (since a closed
subset of a complete space is complete). Unfortunately there are quite a few annoying details
to check!
‘Proof 2’ (the last bit is left to you). Let V be a normed space. First we introduce the rather
large space

(1.62) and
the elements of which, if you recall, are said to be absolutely summable. Notice that the
elements of Ve are sequences, valued in V so two sequences are equal, are the same, only when
each entry in one is equal to the corresponding entry in the other – no shifting around or
anything is permitted as far as equality is concerned. We think of these as series (remember
this means nothing except changing the name, a series is a sequence and a sequence is a
series), the only difference is that we ‘think’ of taking the limit of a sequence but we ‘think’
of summing the elements of a series, whether we can do so or not being a different matter.
Now, each element of Ve is a Cauchy sequence – meaning the corresponding N

sequence of partial sums vN = P uk is Cauchy if {uk} is absolutely summable. As

k=1

noted earlier, this is simply because if M ≥ N then

M M

(1.63) kvM − vNk = k X ujk ≤ X kujk ≤ X kujk

j=N+1 j=N+1 j≥N+1

gets small with N by the assumption that Pkujk < ∞.

j

Moreover, Ve is a linear space, where we add sequences, and multiply by constants, by

doing the operations on each component:-
(1.64) .
This always gives an absolutely summable series by the triangle inequality:

(1.65) .

Within Ve consider the linear subspace

(1.66)
of those which sum to 0. As discussed in Section 5 above, we can form the quotient

(1.67) B = V /Se

the elements of which are the ‘cosets’ of the form {uk} + S ⊂ Ve where {uk} ∈ V .e This is our
completion, we proceed to check the following properties of this B.
(1) A norm on B (via a seminorm on V˜) is defined by

(1.68)
(2) The original space V is imbedded in B by
(1.69) V 3 v 7−→ I(v) = {uk} + S, u1 = v, uk = 0 ∀ k > 1
and the norm satisfies (1.58).
(3) I(V ) ⊂ B is dense.
(4) B is a Banach space with the norm (1.68).
 So, first that (1.68) is a norm. The limit on the right does exist since the limit of the norm
of a Cauchy sequence always exists – namely the sequence of norms is itself Cauchy but now
in R. Moreover, adding an element of S to {uk} does not change the norm of the sequence of
partial sums, since the additional term tends to zero in norm. Thus kbkB is well-defined for
each element b ∈ B and kbkB = 0 means exactly that the sequence {uk} used to define it tends
to 0 in norm, hence is in S hence b = 0 in B. The other two properties of norm are reasonably
clear, since
6. COMPLETION

if b, b0 ∈ B are represented by {uk}, {u0k} in Ve then tb and b + b0 are represented by {tuk} and
{uk + u0k} and
(1.70)

for s.t.

Now the norm of the element I(v) = v,0,0,··· , is the limit of the norms of the sequence of partial
sums and hence is kvkV so kI(v)kB = kvkV and I(v) = 0 therefore implies v = 0 and hence I is
also injective.
We need to check that B is complete, and also that I(V ) is dense. Here is an extended
discussion of the difficulty – of course maybe you can see it directly yourself (or have a better
scheme). Note that I suggest that you to write out your own version of it carefully in Problem
1.
Okay, what does it mean for B to be a Banach space, as discussed above it means that
every absolutely summable series in B is convergent. Such a series {bn} is given by
where and the summability condition is that
N

(1.71) ∞ > XkbnkB = X lim kXu(kn)kV .

N→∞
n n k=1

So, we want to show that Pbn = b converges, and to do so we need to find the
n

limit b. It is supposed to be given by an absolutely summable series. The ‘problem’ is that this
series should look like PPu(kn) in some sense – because it is supposed
n k

to represent the sum of the bn’s. Now, it would be very nice if we had the estimate
(1.72) XXku(kn)kV < ∞
n k

since this should allow us to break up the double sum in some nice way so as to get an
absolutely summable series out of the whole thing. The trouble is that (1.72) need not hold.
We know that each of the sums over k – for given n – converges, but not the sum of the sums.
All we know here is that the sum of the ‘limits of the norms’ in (1.71) converges.
So, that is the problem! One way to see the solution is to note that we do not have to
choose the original to ‘represent’ bn – we can add to it any element of S. One idea is to
rearrange the – I am thinking here of fixed n – so that it ‘converges even faster.’ I will not
go through this in full detail but rather do it later when we need the argument for the
completeness of the space of Lebesgue integrable functions. Given > 0 we can choose p1 so
that for all p ≥ p1,

(1.73)
Then in fact we can choose successive pj > pj−1 (remember that little n is fixed here) so that
(1.74)

p1 j

Now, ‘resum the series’ defining instead v1(n) = P u(kn), vj(n) = P u(kn) and
k=1 k=pj−1+1

do this setting for the nth series. Check that now

(1.75) .
Of course, you should also check that so that these new summable series work
just as well as the old ones.
After this fiddling you can now try to find a limit for the sequence as

(1.76) b = {wk} + S, wk = X vl(p) ∈ V.

l+p=k+1

So, you need to check that this {wk} is absolutely summable in V and that bn → b as n → ∞.
Finally then there is the question of showing that I(V ) is dense in B. You can do this using
the same idea as above – in fact it might be better to do it first. Given an element b ∈ B we
need to find elements in V, vk such that kI(vk) − bkB → 0 as
Nj

k → ∞. Take an absolutely summable series uk representing b and take vj = P uk

k=1

where the pj’s are constructed as above and check that I(vj) → b by computing

(1.77) kI(vj) − bkB = lim k X ukkV ≤ X kukkV .

→∞ k>pj k>pj
7. More examples

• c0 the space of convergent sequences in C with supremum norm, a Banach space.

• lp one space for each real number 1 ≤ p < ∞; the space of p-summable series with
corresponding norm; all Banach spaces. The most important of these for us is the
case p = 2, which is (a) Hilbert space.
• l∞ the space of bounded sequences with supremum norm, a Banach space with c0 ⊂
l∞ as a closed subspace with the same norm.
• C([a,b]) or more generally C(M) for any compact metric space M – the Banach space
of continuous functions with supremum norm.
• C∞(R), or more generally C∞(M) for any metric space M – the Banach space of
bounded continuous functions with supremum norm.
7. MORE EXAMPLES

• C0(R), or more generally C0(M) for any metric space M – the Banach space of
continuous functions which ‘vanish at infinity’ (see Problem 1) with supremum
norm. A closed subspace, with the same norm, in C∞0 (M).
• Ck([a,b]) the space of k times continuously differentiable (so k ∈ N) functions on [a,b]
with norm the sum of the supremum norms on the function and its derivatives. Each
is a Banach space – see Problem 1.
• The space C([0,1]) with norm

(1.78)
given by the Riemann integral of the absolute value. A normed space, but not a
Banach space. We will construct the concrete completion, L1([0,1]) of Lebesgue
integrable ‘functions’.
• The space R([a,b]) of Riemann integrable functions on [a,b] with kuk defined by
(1.78). This is only a seminorm, since there are Riemann integrable functions (note
that u Riemann integrable does imply that |u| is Riemann integrable) with |u| having
vanishing Riemann integral but which are not identically zero. This cannot happen
for continuous functions. So the quotient is a normed space, but it is not complete.
• The same spaces – either of continuous or of Riemann integrable functions but with
the (semi- in the second case) norm

(1.79) .
Not complete in either case even after passing to the quotient to get a norm for
Riemann integrable functions. We can, and indeed will, define Lp(a,b) as the
completion of C([a,b]) with respect to the Lp norm. However we will get a concrete
realization of it soon.
• Suppose 0 < α < 1 and consider the subspace of C([a,b]) consisting of the ‘H¨older
continuous functions’ with exponent α, that is those u : [a,b] −→ C which satisfy
 (1.80) |u(x) − u(y)| ≤ C|x − y|α for some C ≥ 0.

Note that this already implies the continuity of u. As norm one can take the sum of
the supremum norm and the ‘best constant’ which is the same as

(1.81) ;

it is a Banach space usually denoted Cα([a,b]).

• Note the previous example works for α = 1 as well, then it is not denoted C1([a,b]),
since that is the space of once continuously differentiable functions; this is the space
of Lipschitz functions – again it is a Banach space.
• We will also talk about Sobolev spaces later. These are functions with ‘Lebesgue
integrable derivatives’. It is perhaps not easy to see how to
define these, but if one takes the norm on C1([a,b])

(1.82)
and completes it, one gets the Sobolev space H1([a,b]) – it is a Banach space (and a Hilbert space). In
fact it is a subspace of C([a,b]) = C0([a,b]).
Here is an example to see that the space of continuous functions on [0,1] with norm (1.78) is not complete;
things are even worse than this example indicates! It is a bit harder to show that the quotient of the Riemann
integrable functions is not complete, feel free to give it a try.
Take a simple non-negative continuous function on R for instance
(
1 − |x| if |x| ≤ 1
(1.83) f(x) =
0 if |x| > 1.

Then . Now scale it up and in by setting

(1.84) fN(x) = Nf(N3x) = 0 if |x| > N−3.

So it vanishes outside [−N−3,N−3] and has . It follows that the sequence {fN} is absolutely
summable with respect to the integral norm in (1.78) on [−1,1]. The pointwise series PfN(x) converges
everywhere except at x = 0 – N
since at each point x 6= 0, fN(x) = 0 if N3|x| > 1. The resulting function, even if we ignore the problem at x = 0, is
not Riemann integrable because it is not bounded.
 Linear spaces and the Hahn Banach Theorem

Many objects in mathematics — particularly in analysis — are, or may be described in

terms of, linear spaces (also called vector spaces). For example:

(1) C(M) = space of continuous functions (R or C valued) on a manifold M.

(2) A(U) = space of analytic functions in a domain U ⊂ C.
(3) Lp(µ) = {p integrable functions on a measure space M,µ}.
The key features here are the axioms of linear algebra,
Definition 1.1. A linear space X over a field F (in this course F = R or C) is a set on

which we have defined

(1) addition: x,y ∈ X 7→ x + y ∈ X
and
(2) scalar multiplication: k ∈ F, x ∈ X 7→ kx ∈ X

with the following properties

(1) (X,+) is an abelian group (+ is commutative and associative and ∃ identity and
inverses.)
• identity is called 0 (“zero”)
• inverse of x is denoted −x
(2) scalar multiplication is
• associative: a(bx) = (ab)x,
• distributive: a(x + y) = ax + by and (a + b)x = ax + bx, and satisfies 1x =
x.
Remark. It follows from the axioms that 0x = 0 and −x = (−1)x.
Recall from linear algebra that a set of vectors S ⊂ X is linearly independent if

= 0 with x1,...,xn ∈ S =⇒ a1 = ··· = an = 0

and that the dimension of X is the cardinality of a maximal linearly independent set in X. The
dimension is also the cardinality of a minimal spanning set, where the span of a set S

is the set
)
spanand x1,...,xn ∈ S
,
and S is spanning, or spans X, if spanS = X.

More or less, functional analysis is linear algebra done on spaces with infinite dimension.
Stated this way it may seem odd that functional analysis is part of analysis. For finite

dimensional spaces the axioms of linear algebra are very rigid: there is essentially only
1-1
1-2 1. LINEAR SPACES AND THE HAHN BANACH THEOREM

one interesting topology on a finite dimensional space and up to isomorphism there is only
one linear space of each finite dimension. In infinite dimensions we shall see that topology
matters a great deal, and the topologies of interest are related to the sort of analysis that one
is trying to do.
That explains the second word in the name ”functional analysis.” Regarding “functional,”
this is an archaic term for a function defined on a domain of functions. Since most of the
spaces we study are function spaces, like C(M), the functions defined on them are
“functionals.” Thus “functional analysis.” In particular, we define a linear functional to be a
linear map ` : X → F, which means
`(x + y) = `(x) + `(y) and `(ax) = a`(x) for all x,y ∈ X and a ∈ F.
Often, one is able to define a linear functional at first only for a limited set of vectors Y ⊂
X. For example, one may define the Riemann integral on Y = C[0,1], say, which is a subset of
the space B[0,1] of all bounded functions on [0,1]. In most cases, as in the example, the set Y
is a subspace:

Definition 1.2. A subset Y ⊂ X of a linear space is a linear subspace if it is closed

under addition and scalar multiplication: y1,y2 ∈ Y and a ∈ F =⇒ y1 + ay2 ∈ Y .

For functionals defined, at first, on a subspace of a linear space of R we have

Theorem 1.1 (Hahn (1927), Banach (1929)). Let X be a linear space over R and p a real
valued function on X with the properties
(1) p(ax) = ap(x) for all x ∈ X and a > 0 (Positive homogeneity)
(2) p(x + y) ≤ p(x) + p(y) for all x,y ∈ X (subadditivity).
If ` is a linear functional defined on a linear subspace of Y and dominated by p, that is `(y) ≤ p(y)
for all y ∈ Y , then ` can be extended to all of X as a linear functional dominated by p, so `(x) ≤
p(x) for all x ∈ X.
Example. Let X = B[0,1] and Y = C[0,1]. On Y , let (Riemann integral). Let
p : B → R be p(f) = sup{|f(x)| : x ∈ [0,1]}. Then p satisfies (1) and (2) and `(f) ≤ p(f). Thus we
can extend ` to all of B[0,1]. We will return to this example and see that we can extend ` so
that `(f) ≥ 0 whenever f ≥ 0. This defines a finitely additive
set function on all(!) subset of [0,1] via µ(S) = `(χS). For Borel measurable sets it turns out the
result is Lebesgue measure. That does not follow from Hahn-Banach however.
The proof of Hahn-Banach is not constructive, but relies on the following result
equivalent to the axiom of choice:
Theorem 1.2 (Zorn’s Lemma). Let S be a partially ordered set such that every totally ordered
subset has an upper bound. Then S has a maximal element.
To understand the statement, we need
Definition 1.3. A partially ordered set S is a set on which an order relation a ≤ b is

defined for some (but not necessarily all) pairs a,b ∈ S with the following properties
(1) transitivity: if a ≤ b and b ≤ c then a ≤ c
(2) reflexivity: if a ≤ a for all a ∈ S.
(Note that (1) asserts two things: that a and c are comparable and that a ≤ c.) A subset T of S
is totally ordered if x,y ∈ T =⇒ x ≤ y or y ≤ x. An element u ∈ S is an upper bound

for T ⊂ S if x ∈ T =⇒ x ≤ u. A maximal element m ∈ S satisfies m ≤ b =⇒ m = b.

1. LINEAR SPACES AND THE HAHN BANACH THEOREM 1-3

Proof of Hahn-Banach. To apply Zorn’s Lemma, we need a poset, S =

{extensions of ` dominated by p}.
That is S consists of pairs (`0,Y 0) with `0 a linear functional defined on a subspace Y 0 ⊃ Y so
that
`0(y) = `(y), y ∈ Y and `0(y) ≤ p(y), y ∈ Y 0. Order S as
follows
(`1,Y1) ≤ (`2,Y2) ⇐⇒ Y1 ⊂ Y2 and `2|Y1 = `1.

If T is a totally ordered subset of S, let (`,Y ) be

Y = [{Y 0 : (`0,Y 0) ∈ T}
and

`(y) = `0(y) for y ∈ Y 0.

Since T is totally ordered the definition of ` is unambiguous. Clearly (`,Y ) is an upper bound
for T. Thus by Zorn’s Lemma there exists a maximal element (`+,Y +).
To finish, we need to see that Y + = X. It suffices to show that (`0,Y 0) has an extension
whenever Y 0 6= X. Indeed, let x0 ∈ X. We want `00 on Y 00 = {ax0 + y : y ∈ Y, a ∈ R}. By
linearity we need only define `00(x0). The constraint is that we need
a`00(x0) + `0(y) ≤ p(ax0 + y)
for all a,y. Dividing through by |a|, since Y 0 is a subspace, we need only show
± `00(x0) ≤ p(y ± x0) − `0(y) (1.1)
for all y ∈ Y 0. We can find `00(x0) as long as
`0(y0) − p(y0 − x0) ≤ p(x0 + y) − `0(y) for all y,y0 ∈ Y 0, (1.2)
or equivalently
`0(y0 + y) ≤ p(x0 + y) + p(y0 − x0) for all y,y0 ∈ Y 0. (1.3)
Since
`0(y0 + y) ≤ p(y0 + y) = p(y0 − x0 + y + x0) ≤ p(x0 + y) + p(y0 − x0),
(1.3), and thus (1.2), holds. So we can satisfy (1.1).
In finite dimensions, one can give a constructive proof involving only finitely many
choices. In infinite dimensions the situation is a quite a bit different, and the approach via
Zorn’s lemma typically involves uncountably many “choices.”

Geometric Hahn-Banach Theorems

We may use Hahn-Banach to understand something of the geometry of linear spaces. We

want to understand if the following picture holds in infinite dimension:

Figure 2.1. Separating a point from a convex set by a line hyperplane

Definition 2.1. A set S ⊂ X is convex if for all x,y ∈ S and t ∈ [0,1] we have tx + (1 − t)y ∈ S.

Definition 2.2. A point x ∈ S ⊂ X is an interior point of S if for all y ∈ X ∃ε > 0 s.t.

|t| < ε =⇒ x + ty ∈ S.

Remark. We can a define a topology using this notion, letting U ⊂ X be open ⇐⇒ all x ∈ U
are interior. From the standpoint of abstract linear algebra this seems to be a “natural”
topology on X. In practice, however, it has way too many open sets and we work with weaker
topologies that are relevant to the analysis under considerations. Much of functional analysis
centers around the interplay of different topologies.
We are aiming at the following

2-1
2-2 2. GEOMETRIC HAHN-BANACH THEOREMS

Theorem 2.1. Let K be a non-empty convex subset of X, a linear space over R, and suppose K
has at least one interior point. If y 6∈ K then ∃ a linear functional ` : X → R s.t.

`(x) ≤ `(y) for all x ∈ K, (2.1)

with strict inequality for all interior points x of K.

This is the “hyperplane separation theorem,” essentially validates the picture drawn
above. A set of the form {`(x) = c} with ` a linear functional is a “hyperplane” and the sets
{`(x) < c} are “half spaces.”
To accomplish the proof we will use Hahn-Banach. We need a dominating function p.

Definition 2.3. Let K ⊂ X be convex and suppose 0 is an interior point. The gauge

of K (with respect to the origin) is the function pK : X → R defined as

pK(x) = inf{a : a > 0 and .

(Note that pK(x) < ∞ for all x since 0 is interior.)

Lemma 2.2. pK is positive homogeneous and sub-additive.

Proof. Positive homogeneity is clear (even if K is not convex). To prove sub-additivity we

use convexity. Consider pK(x + y). Let a,b be such that x/a,y/b ∈ K. Then

,
so

Thus pK(x + y) ≤ a + b, and optimizing over a,b we obtain the result.

Proof of hyperplane separation thm. Suffices to assume 0 ∈ K is interior and c = 1. Let pK be

the gauge of K. Note that pK(x) ≤ 1 for x ∈ K and that pK(x) < 1 if x is interior, as then (1 + t)x ∈
K for small t > 0. Conversely if pK(x) < 1 then x is an interior point (why?), so
 pK(x) < 1 ⇐⇒ x ∈ K.
Now define `(y) = 1, so `(ay) = a for a ∈ R. Since y 6∈ K it is not an interior point and so
pK(y) ≥ 1. Thus pK(ay) ≥ a for a ≥ 0 and also, trivially, for a < 0 (since pK ≥ 0). Thus

pK(ay) ≥ `(ay)

for all a ∈ R. By Hahn-Banach, with Y the one dimensional space {ay}, ` may be extended to
all of x so that pK(x) ≥ `(x) which implies (2.1).

An extension of this is the following

Theorem 2.3. Let H,M be disjoint convex subsets of X, at least one of which has an interior
point. Then H and M can be separate by a hyperplane `(x) = c: there is a linear functional ` and
c ∈ R such that

`(u) ≤ c ≤ `(v)∀u ∈ H, v ∈ M.
2. GEOMETRIC HAHN-BANACH THEOREMS 2-3

Proof. The proof rests on a trick of applying the hyperplane separation theorem with the
set
K = H − M = {u − v : u ∈ H and v ∈ M}
and the point y = 0. Note that 0 6∈ K since H ∪ M = ∅. Since K has an interior point (why?), we
see that there is a linear functional such that `(x) ≤ 0 for all x ∈ K. But then
`(u) ≤ `(v) for all u ∈ H, v ∈ M.

In many applications, one wants to consider a vector space X over C. Of course, then X is
also a vector space over R so the real Hahn-Banach theorem applies. Using this one can show
the following
Theorem 2.4 (Complex Hahn-Banach, Bohenblust and Sobczyk (1938) and Soukhomlinoff
(1938)). Let X be a linear space over C and p : X → [0,∞) such that
(1) p(ax) = |a|p(x)∀a ∈ C,x ∈ X.
(2) p(x + y) ≤ p(x) + p(y) (sub-additivity).
Let Y be a C linear subspace of X and ` : Y → C a linear functional such that
|`(y)| ≤ p(y) (2.2)
for all y ∈ Y . Then ` can be extended to all of X so that (2.2) holds for all y ∈ X.
Remark. A function p that satisfies (1) and (2) is called a semi-norm. It is a norm if p(x) =
0 =⇒ x = 0.
Proof. Let `1(y) = Re`(y), the real part of `. Then `1 is a real linear functional and −`1(iy) =
−Rei`(y) = Im`(y), the imaginary part of `. Thus
`(y) = `1(y) − i`1(iy). (2.3)
 Clearly |`1(y)| ≤ p(y) so by the real Hahn-Banach theorem we can extend `1 to all of X so
that `1(y) ≤ p(y) for all y ∈ X. Since −`1(y) = `1(−y) ≤ p(−y) = p(y), we have |`1(y)| ≤ p(y) for all y
∈ X. Now define the extension of ` via (2.3). Given y ∈ X let θ = argln`(y). Thus `(y) = eiθ`1(e−iθy)
(why?). So,
|`(y)| = |`1(e−iθy)| ≤ p(y).

Lax gives another beautiful extension of Hahn-Banach, due to Agnew and Morse, which
involves a family of commuting linear maps. We will cover a simplified version of this next
time.
Applications of Hahn-Banach

To get an idea what one can do with the Hahn-Banach theorem let’s consider a concrete
application on the linear space X = B(S) of all real valued bounded functions on a set S. B(S)
has a natural partial order, namely x ≤ y if x(s) ≤ y(s) for all s ∈ S. If 0 ≤ x then x is nonnegative.
On B(S) a positive linear functional ` satisfies `(y) ≥ 0 for all y ≥ 0.

Theorem 3.1. Let Y be a linear subspace of B(S) that contains y0 ≥ 1, so y0(s) ≥ 1 for all s ∈ S.
If ` is a positive linear functional on Y then ` can be extended to all of B as a positive linear
functional.
This theorem can be formulated in an abstract context as follows. The nonnegative
functions form a cone, where

Definition 3.1. A subset P ⊂ X of a linear space over R is a cone if tx + sy ∈ P whenever x,y

∈ P and t,s ≥ 0. A linear functional on X is P-nonnegative if `(x) ≥ 0 for all x ∈ P.
Theorem 3.2. Let P ⊂ X be a cone with an interior point x0. If Y is a subspace containing x0
on which is defined a P ∩Y -positive linear functional `, then ` has an extension to X which is P-
positive.
Proof. Define a dominating function p as follows
p(x) = inf{`(y) : y − x ∈ P, y ∈ Y }.
Note that y0 − tx ∈ P for some t > 0 (since y0 is interior to P), so 1t y0 − x ∈ P. This shows that
p(x) is well defined. It is clear that p is positive homegeneous. To see that it is sub-additive,
let x1,x2 ∈ X and let y1,y2 be so that yj −xj ∈ P. Then y1+y2−(x1+x2) ∈ P, so
p(x1 + x2) ≤ `(y1) + `(y2).
Minimizing over y1,2 gives sub-additivity.
Since `(x) = `(x−y)+`(y) ≤ `(y) if x ∈ Y and y −x ∈ P we conclude that `(x) ≤ p(x) for all x ∈ Y
. By Hahn-Banach we may extend ` to all of X so that `(x) ≤ p(x) for all x.
Now let x ∈ P. Then p(−x) ≤ 0 (why?), so −`(x) = `(−x) ≤ 0 which shows that ` is P-positive.
 The theorem on B(S) follows from the Theorem 3.2 once we observe that the condition
y0 ≥ 1 implies that y0 is an interior point of the cone of positive functions. The linear functional
that one constructs in this way is monotone:
x≤y =⇒ `(x) ≤ `(y),
which is in fact equivalent to positivity. This can allow us to do a little bit of analysis, even
though we haven’t introduced notions of topology or convergence.
3-1
3-2 3. APPLICATIONS OF HAHN-BANACH

To see how this could work, let’s apply the result to the Riemann integral on C[0,1].
We conclude the existence of a positive linear functional ` : B[0,1] → R that gives `(f) =
. This gives an “integral” of arbitrary bounded functions, without a measurability
condition. Since the integral is a linear functional, it is finitely additive. Of course it is not
countably additive since we know from real analysis that such a thing doesn’t exist.
Furthermore, the integral is not uniquely defined.
Nonetheless, the extension is also not arbitrary, and the constraint of positivity actually
pins down `(f) for many functions. For example, consider χ(a,b) (the characteristic function of
an open interval). By positivity we know that
Z1 Z1
f ≤ χ(a,b) ≤ g, f,g ∈ C[0,1] =⇒ f(x)dx ≤ `(χ(a,b)) ≤ g(x)dx.
0 0

Taking the sup over f and inf over g, using properties only of the Riemann integral, we see
that `(χ(a,b)) = b − a (hardly a surprising result). Likewise, we can see that `(χU) = |U| for any
open set, and by finite additivity `(χF ) = 1 − `(Fc) = 1 − |Fc| = |F| for any closed set
(| · | is Lebesgue measure). Finally if S ⊂ [0,1] and
sup{`(F) : F closed and F ⊂ S} = inf {`(U) : U open and U ⊃ S}
then `(χS) must be equal to these two numbers. We see that `(χS) = |S| for any Lebesgue
measurable set. We have just painlessly constructed Lebesgue measure from the Riemann
integral without using any measure theory!
The rigidity of the extension is a bit surprising if we compare with what happens in finite
dimensions. For instance, consider the linear functional `(x,0) = x defined on Y = {(x,0)} ⊂ R2.
Let P be the cone {(x,y) : |y| ≤ αx}. So ` is P ∩ Y -positive. To extend ` to all of R2 we need to
define `(0,1). To keep the extension positive we need only require
y`(0,1) + 1 ≥ 0
if |y| ≤ α. Thus we must have |`(0,1)| ≤ 1/α, and any choice in this interval will work. Only
when α = ∞ and the cone degenerates to a half space does the condition pin `(0,1) down. So
some interesting things happen in ∞ dimensions.
A second example application assigning a limiting value to sequences
a = (a1,a2,...).
Let B denote the space of all bounded R-valued sequences and let L denote the subspace of
sequences with a limit. We quickly conclude that there is an positive linear extension of the
positive linear functional lim to all of B. We would like to conclude a little more, however.
After all, for convergent sequences,
liman+k = liman
for any k.
To formalize this property we define a linear map on B by

T(a1,a2,...) = (a2,a3,...).
(T is the “backwards shift” operator or “left translation.”) Here,
Definition 3.2. Let X1,X2 be linear spaces over a field F. A linear map T : X1 → X2

is a function such that

T(x + ay) = T(x) + aT(y) ∀x,y ∈ X1 and a ∈ F.
Normed and Banach Spaces

The Hahn-Banach theorem made use of a dominating function p(x). When this function is
non-negative, it can be understood roughly as a kind of “distance” from a point x to the origin.
For that to work, we should have p(x) > 0 whenever x 6= 0. Such a function is called a norm:

Definition 4.1. Let X be a linear space over F = R or C. A norm on X is a function k·k : X →

[0,∞) such that
(1) kxk = 0 ⇐⇒ x = 0.
(2) kx + yk ≤ kxk + kyk (subadditivity)
(3) kaxk = |a|kxk for all a ∈ F and x ∈ X (homogeneity).
A normed space is a linear space X with a norm k·k.
The norm on a normed space gives a metric topology if we define the distance between
two points via
d(x,y) = kx − yk.
Condition 1 guarantees that two distinct points are a finite distance apart. Sub-additivity
gives the triangle inequality. The metric is
(1) translation invariant: d(x + z,y + z) = d(x,y)
and
(2) homogeneous d(ax,ay) = |a|d(x,y).

Thus any normed space is a metric space and we have the following notions:
(1) a sequence xn converges to x, xn → x, if d(xn,x) = kxn − xk → 0.
(2) a set U ⊂ X is open if for every x ∈ U there is a ball .
 (3) a set K ⊂ X is closed if X \ K is open.
(4) a set K ⊂ X is compact if every open cover of K has a finite sub-cover.
The norm defines the topology but not the other way around. Indeed two norms k·k1 and
k·k2 on X are equivalent if there is c > 0 such that

ckxk1 ≤ kxk2 ≤ c−1 kxk2 ∀x ∈ X.

Equivalent norms define the same topology. (Why?)

Recall from real analysis that a metric space X is complete if every Cauchy sequence xn

converges in X. In a normed space, a Cauchy sequence xn is one such that

such that
A complete normed space is called a Banach space.

4-1
4-2 4. NORMED AND BANACH SPACES

Not every normed space is complete. For example C[0,1] with the norm

Z1
kfk1 = |f(x)|dx
0

fails to be complete. (It is, however, a complete space in the uniform norm, kfku = supx∈[0,1]
|f(x)|.) However, every normed space X has a completion, defined abstractly as

a set of equivalence classes of Cauchy sequences in X. This space, denoted X is a Banach space.

Examples of Normed and Banach spaces

(1) For each p ∈ [1,∞) let
∞

`p = {p summable sequences} = {(a1,a2,...) | X|aj|p < ∞}.

j=1

Define a norm on `p via

.
 Then `p is a Banach space.
(2) Let
`∞ = {bounded sequences} = B(N),
with norm
kak∞ = sup|aj|. (?)
j

Then `∞ is a Banach space.

(3) Let

c0 = {sequences converging to 0} = {(a1,a2,...) | lim aj = 0},

j→∞

with norm (?). Then c0 is a Banach space.

(4) Let

F = {sequences with finitely many non-zero terms}

= {(a1,a2,...) | ∃N ∈ Nsuch that n ≥ N =⇒ an = 0}.

Then for any p ≥ 1, Fp = (F,k·kp) is a normed space which is not complete. The
completion of Fp is isomorphic to `p.
(5) Let D ⊂ Rd be a domain and let p ∈ [1,∞).
(a) Let X = Cc(D) be the space of continuous functions with compact support in D,
with the norm

.
EXAMPLES OF NORMED AND BANACH SPACES 4-3

Then X is a normed space, which is not complete. Its completion is denoted

Lp(D) and may be identified with the set of equivalence classes of measurable
functions f : D → C such that
Z
|f(x)|pdx < ∞ (Lebesgue measure),
D

with two functions f,g called equivalent if f(x) = g(x) for almost every x.
(b) Let X denote the set of C1 functions on D such that
Z Z
|f(x)|pdx < ∞ and |∂jf(x)|pdx < ∞, j = 1,...,n.
D D

Put the following norm on X,

 .
Then X is a normed space which is not complete. Its completion is denoted
W1,p(D) and is called a Sobolev space and may be identified with the subspace of Lp(D)
consisting of (equivalence classes) of functions all of whose first derivatives are in Lp(D)
in the sense of distributions. (We’ll come back to this.) Note that a ∈ `1 is summable,
X aj ≤ kak1 .
j

Theorem 4.1 (Ho¨lder’s Inqequality). Let 1 < p < ∞ and let q be such that

.
If a ∈ `p and b ∈ `q then ab = (a1b1,a2b2,...) ∈ `1
and

.
Proof. First note that for two non-negative numbers a,b it holds that

.
For p = q = 2 this is the familiar “arithmetic-geometric mean” inequality which follows since
(a−b)2 ≥ 0. For general a,b this may be seen as follows. The function x 7→ exp(x) is convex:
exp(tx + (1 − t)y) ≤ texp(x) + (1 − t)exp(y). (Recall from calculus that a C2 function f is convex
if f00 ≥ 0.) Thus,

.
Applying this bound co-ordinate wise and summing up we find that

.
4-4 4. NORMED AND BANACH SPACES

The result follows from this bound by “homogenization:” we have

kakp ,kbkq =⇒ kabk ≤ 1,

so a b

from which the desired estimate follows by homogeneity of the norm.

Similarly, we have
Theorem 4.2 (H¨older’s Inequality). Let 1 < p < ∞ and let q be the conjugate exponent. If f
∈ Lp(D)
and g ∈ Lq(D) then fg ∈ L1(D) and Z
|f(x)g(x)|dx ≤ kfkp kgkq .D
 Noncompactness of the Ball and Uniform Convexity

First a few more definitions:

Definition 5.1. A normed space X over F = R or C is called separable if it has a

countable, dense subset.

Most spaces we consider are separable, with a few notable exceptions.
(1) The space M of all signed (or complex) measures µ on [0,1], say, with norm
Z 1 kµk =
|µ|(dx).
0

Here |µ| denotes the total variation of µ,

|µ|(A) = sup .
Partitions

This space is a Banach space. Since the point mass

is an element of A and kδx − δyk = 2 if x 6= y, we have an uncountable family of

elements of M all at a fixed distance of one another. Thus there can be no countable
dense subset. (Why?)
(2) `∞ is also not separable. To see this, note that to each subset of A ⊂ N we may
associate the sequence χA, and

kχA − χBk∞ = 1
if A 6= B.
(3) `p is separable for 1 ≤ p < ∞.
(4) Lp(D) is separable for 1 ≤ p < ∞.
(5) L∞(D) is not separable.

Noncompactness of the Unit Ball

Theorem 5.1 (F. Riesz). Let X be a normed linear space. Then the closed unit ball B1(0) = {x
: kXk ≤ 1} is compact if and only if X is finite dimensional.

Proof. The fact that the unit ball is compact if X is finite dimensional is the HeineBorel
Theorem from Real Analysis. To see the converse, we use the following
5-1
5-2 5. NONCOMPACTNESS OF THE BALL AND UNIFORM CONVEXITY

Lemma 5.2. Let Y be a closed proper subspace of a normed space X. Then there is a unit
vector z ∈ X, kzk = 1, such that

Proof of Lemma. Since Y is proper, ∃x ∈ X \ Y . Then

.
(This is a property of closed sets in a metric space.) We do not know the existence of a
minimizing y, but we can certainly find y0 such that
0 < kx − y0k < 2d.
Let . Then

Returning to the proof of the Theorem, we will use the fact that every sequence in a
compact metric space has a convergent subsequence. Thus it suffices to show that if X is
infinite dimensional then there is a sequence in B1(0) with no convergent subsequence.
Let y1 be any unit vector and construct a sequence of unit vectors, by induction, so that

span{y1,...,yk−1}.
(Note that span{y1,...,yk−1} is finite dimensional, hence complete, and thus a closed subspace
of X.) Since X is finite dimensional the process never stops. No subsequence of yj can be
Cauchy, much less convergent.

Uniform convexity
The following theorem may be easily shown using compactness:
Theorem 5.3. Let X be a finite dimensional linear normed space. Let K be a closed convex
subset of X and z any point of X. Then there is a unique point of K closer to z than any other point
of K. That is there is a unique solution y0 ∈ K to the minimization problem

. (?)
Try to prove this theorem. (The existence of a minimizer follows from compactness; the
uniqueness follows from convexity.)
The conclusion of theorem does not hold in a general infinite dimensional space.
Nonetheless there is a property which allows for the conclusion, even though compactness
fails!
Definition 5.2. A normed linear space X is uniformly convex if there is a function
 ), such that
is increasing.

) for all x,y ∈ B1(0), the unit ball of X.

UNIFORM CONVEXITY 5-3

Theorem 5.4 (Clarkson 1936). Let X be a uniformly convex Banach space, K a closed convex
subset of X, and z any point of X. Then the minimization problem (?) has a unique solution y0 ∈
K.
Proof. If z ∈ K then y0 = z is the solution, and is clearly unique. When z 6∈ K,
we may assume z = 0 (translating z and K if necessary). Let
s = inf kyk.
y∈K

So s > 0. Now let yn ∈ K be a minimizing sequence, so

kynk → s.
Now let xn = yn/kynk, and consider

for suitable t. So tyn + (1 − t)ym ∈ K and

Thus
.

We conclude that xn is a Cauchy sequence, from which it follows that yn is Cauchy. The limit
y0 ∈ limn yn exists in K since X is complete and K is closed. Clearly ky0k = s.
Warning: Not every Banach space is uniformly convex. For example, the space C(D) of
continuous functions on a compact set D is not uniformly convex. It may even happen that

for unit vectors f and g. (They need only have disjoint support.) Lax gives an example of a
closed convex set in C[−1,1] in which the minimization problem (?) has no solution.
It can also happen that a solution exists but is not unique. For example, in C[−1,1] let K =
{functions that vanish on [−1,0]}. and let f = 1 on [−1,1]. Clearly

and the distance 1 is attained for any g ∈ K that satisfies 0 ≤ g(x) ≤ 1.

 Linear Functionals on a Banach Space

Definition 6.1. A linear functional ` : X → F on a normed space X over F = R or C is bounded

if there is c < ∞ such that
|`(x)| ≤ ckxk ∀x ∈ X.
The inf over all such c is the norm k`k of `,

. (?)
Theorem 6.1. A linear functional ` on a normed space is bounded if and only if it is
continuous.
Proof. It is useful to recall that

Theorem 6.2. Let X,Y be metric spaces. Then f : X → Y is continuous if and only if f(xn) is a
convergent sequence in Y whenever xn is convergent in X.

Remark. Continuity =⇒ the sequence condition in any topological space. That the
sequence condition =⇒ continuity follows from the fact that the topology has a countable
basis at a point. (In a metric space, B2−n(x), say.) Specifically, suppose the function is not
continuous. Then there is an open set U ⊂ Y such that f−1(U) is not open. So there is x ∈ f−1(U)
such that for all n B2−n(x) 6⊂ f−1(U). Now let xn be a sequence such that
(1) xn ∈ B2−n(x) \ f−1(U) for n odd.
(2) xn = x for n even.
Clearly xn → x. However, f(xn) cannot converge since limk f(x2k) = f(x) and any limit point of
f(x2k+1) lies in the closed set Uc containing all the points f(x2k+1).
In the normed space context, note that
|`(xn) − `(x)| ≤ k`kkxn − xk,
so boundedness certainly implies continuity.
Conversely, if ` is unbounded then we can find vectors xn so that `(xn) ≥√nkxnk. Since this
inequality is homogeneous under scaling, we may suppose that kxnk = 1/ n, say. Thus xn → 0
and `(xn) → ∞, so ` is not continuous.

The set X0 of all bounded linear functionals on X is called the dual of X. It is a linear space,
and in fact a normed space under the norm (?). (It is straightforward to show that (?) defines
a norm.)

Theorem 6.3. The dual X0 of a normed space X is a Banach space.

Proof. We need to show X0 is complete. Suppose `m is a Cauchy Sequence. Then for each x
∈ X we have
 |`n(x) − `m(x)| ≤ k`n − `mkkxk,
so `n(x) is a Cauchy sequence of scalars. Let `(x) = lim `n(x) ∀x
∈ X.
n→∞

It is easy to see that ` is linear. Let us show that it is bounded. Since |k`nk − k`mk| ≤ k`n −
`mk (this follows from sub-additivity), we see that the sequence k`nk is Cauchy, and thus
bounded. So,

and ` is bounded. Similarly,

|`n(x) − `(x)| ≤ sup k`n − `mkkxk,
m≥n

so
k`n − `k ≤ sup k`n − `mk
m≥n

and it follows that `n → `.

Of course, all of this could be vacuous. How do we know that there are any bounded linear
functionals? Here the Hahn-Banach theorem provides the answer.
Theorem 6.4. Let y1,...,yN be N linearly independent vectors in a normed space X and α1,...,αN
arbitrary scalars. Then there is a bounded linear functional ` ∈ X0 such that `(yj) = αj, j = 1,...,N.
Proof. Let Y = span{y1,...,yN} and define ` on Y by
N N

`(Xbj) = Xbjαj.
j=1 j=1

(We use linear independence here to guarantee that ` is well defined.) Clearly, ` is linear.
Furthermore, since Y is finite dimensional ` is bounded.
(Explicitly, since any two norms on a finite dimensional space are equivalent, we can find
c such that
.

Thus,
Thus ` is a linear functional on Y dominated by the norm k·k. By the Hahn-Banach
theorem, it has an extension to X that is also dominated by k·k, i.e., that is bounded.

A closed subspace Y of a normed space X is itself a normed space. If X is a Banach space,

so is Y . A linear functional ` ∈ X0 on X can be restricted to Y and is still bounded. That is there
is a restriction map R : X0 → Y 0 such that
R(`)(y) = `(y) ∀y ∈ Y.
It is clear that R is a linear map and that
kR(`)k ≤ k`k.
6. LINEAR FUNCTIONALS ON A BANACH SPACE 6-3

The Hahn-Banach Theorem shows that R is surjective. On the other hand, unless Y = X, the
kernel of R is certainly non-trivial. To see this, let x be a vector in X \ Y and define ` on span{x}
∪ Y by
`(ax + y) = a ∀a ∈ F and y ∈ Y.
Then ` is bounded on the closed subspace span{x} ∪ Y and by Hahn-Banach has a closed
extension. Clearly R(`) = 0. The kernel of R is denoted Y ⊥, so
Y ⊥ = {` ∈ X0 : `(y) = 0 ∀y ∈ Y },
and is a Banach space.
Now the quotient space X/Y is defined to be the set of “cosets of Y,”
{Y + x : x ∈ X}.
The coset Y + x is denoted [x]. The choice of label x is, of course, not unique as x + y with y ∈
Y would do just as well. It is a standard fact that
[x1] + [x2] = [x1 + x2]
gives a well defined addition on Y/X so that it is a linear space.
Lemma 6.5. If Y is a closed subspace of a normed space then
k[x]k = inf kx + yk
y∈Y

is a norm on X/Y . If X is a Banach space, so is X/Y .

Proof. Exercise.

A bounded linear functional that vanishes on Y , that is an element ` ∈ Y ⊥, can be

understood as a linear functional on X/Y since the definition
`([x]) = `(x)
is unambiguous. Thus we have a map J : Y ⊥ → (X/Y )0 defined by J(`)([x]) = `(x). Conversely,
there is a bounded linear map Π : X → X/Y given by Π(x) = [x] and any linear functional ` ∈
(X/Y )0 pulls back to a bounded linear functional ` ◦ Π in Y ⊥. Clearly J(` ◦ Π) = `. Thus we
have, loosely, that
(X/Y )0 = Y ⊥.
Isometries of a Banach Space

Definition 7.1. Let X,Y be normed spaces. A linear map T : X → Y is bounded if there is c >
0 such that kT(x)k ≤ ckxk. The norm of T is the smallest such c, that is

.
 Theorem 7.1. A linear map T : X → Y between normed spaces X and Y is continuous if and
only if it is bounded.
Remark. The proof is a simple extension of the corresponding result for linear functionals.
An isometry of normed spaces X and Y is a map M : X → Y such that
(1) M is surjective.
(2) kM(x) − M(y)k = kx − yk.
Clearly translations Tu : X → X, Tu(x) = x + u are isometries of a normed linear space. A linear
map T : X → Y is an isometry if T is surjective and
kT(x)k = kxk ∀x ∈ X.
A map M : X → Y is affine if M(x) − M(0) is linear. So, M is affine if it is the composition of
a linear map and a translation.
Theorem 7.2 (Mazur and Ulam 1932). Let X and Y be normed spaces over R. Any isometry
M : X → Y is an affine map.
Remark. The theorem conclusion does not hold for normed spaces over C. In that context
any isometry is a real -affine map (M(x) − M(0) is real linear), but not necessarily

a complex-affine map. For example on C([0,1],C) the map f 7→ f (complex conjugation) is an

isometry and is not complex linear.
Proof. It suffices to show M(0) = 0 =⇒ M is linear. To prove linearity it suffices to show

(Why?)
Let x and y be points in X and ). Note that

,
7-2 7. ISOMETRIES OF A BANACH SPACE

so z is “half-way between x and y.” Let

x0 = M(x), y0 = M(y), z0 = M(z). We need
to show 2z0 = x0 + y0. Since M is an isometry,

,
and all of these are equal to . So z0 is “half-way between x0 and y0.” It may happen
that ) is the unique point of Y with this property (in which case we are done). this
happens, for instance, if the norm in Y is strictly sub-additive, meaning
βx0 6= αy0 =⇒ kx0 + y0k < kx0k + ky0k.
In general, however, there may be a number of points “half-way between x0 and y0.” So,
let

and
.
Since M is an isometry, we have . Let d1 denote the diameter of A1,
d1 = sup ku − vk.
u,v∈A1

This is also the diameter of . Now, let

,
the set of “centers of A1.” Note that z ∈ A2 since if u ∈ A1 then 2z − u ∈ A1:
kx − (2z − u)k = ku − yk = kx − uk = ky − (2z − u)k.
Similarly, let

.
Again, since M is an isometry we have . In a similar way, define decreasing
sequences of sets, Aj and A0j, inductively by

diam(Aj−1)},
and

diam( .
Again M(Aj) = A0j and z ∈ Aj since Aj−1 is invariant under inversion around z: u ∈ Aj−1 =⇒ 2z −
u ∈ Aj−1. Since diam(Aj) ≤ 21−jd1 we conclude that
∞
\
= {z}, and.
j=1

Since z0 ∈ A0j for all j, (?) follows.