Chapter 6

Riesz Representation Theorems

6.1 Dual Spaces

Definition 6.1.1. Let V and W be vector spaces over R. We let

L(V, W ) = {T : V → W | T is linear}.

The space L(V, R) is denoted by V ] and elements of V ] are called linear functionals.

Example 6.1.2. 1) Let V = Rn . Then we can identify R] with R as follows:

For each a = (a1 , a2 , . . . , an ) define φa : Rn → R by
n
X
φa ((x1 , x2 , . . . , xn )) = x · a = xi ai .
i=1

2) Let (X, d) be a compact metric space. Let x0 ∈ X. Define φx0 : C(X) → R by

φx0 (f ) = f (x0 ).

Then φx0 ∈ C(X)] .

Definition 6.1.3. Let (V, k · kV ) and (W, k · kW ) be normed linear spaces. Let T : V → W be linear. We
say that T is bounded is
sup {k T (x) kW } < ∞.
kxkV ≤1

In this case, we write

k T k= sup {k T (x) kW }.
kxkV ≤1

Otherwise, we say that T is unbounded.

The next result establishes the fundamental criterion for when a linear map between normed linear spaces
is continuous. It’s proof is left as an exercise.

Theorem 6.1.4. Let (V, k · kV ) and (W, k · kW ) be normed linear spaces. Let T : V → W be linear. Then
the following are equivalent.
1) T is continuous.
2) T is continuous at 0.

91
 2) T is bounded.
Proof. 1) ⇒ 2) This is immediate.

2) ⇒ 3) Assume that T is continuous at 0. Let δ be such that if k x kV ≤ δ, then k T (x) kW . It follows

easily that k T k≤ 1δ .

3) ⇒ 1) Note that we may assume that k T k> 0 otherwise T = 0 and hence is obviously continuous.
Let x0 ∈ V and let  > 0. Let δ = kT k . Then if k x − x0 kV < δ, we have

k T (x) − T (x0 ) kW =k T (x − x0 ) kW ≤k T k · k x − x0 kV < .

Remark 6.1.5. 1) Let (V, k · kV ) and (W, k · kW ) be normed linear spaces. Let T : V → W be linear.
Then we can easily deduce from the previous theorem that . if T is bounded, then T is uniformly
continuous.
2) Let
B(V, W ) = {T : X → Y |T is linear and T is bounded}.
Let T1 and T2 be in B(V, W ). Then if x ∈ V , we have

k T1 + T2 (x) kW = k T1 (x) + T2 (x) kW

≤ k T1 (x) kW + k T2 (x) kW
≤ k T1 kk x kV + k T1 kk x kV
= (k T1 k + k T2 k) k x kV .

As such T1 + T2 ∈ B(V, W ) and in particular

k T1 + T2 k≤k T1 k + k T1 k

It follows that (B(V, W ), k · k) is also a normed linear space.

Theorem 6.1.6. Assume that (W, k · kW ) be a Banach space. Then so is (B(V, W ), k · k).
Proof. Assume that {Tn } is Cauchy. Let x ∈ V . Since

k Tn (x) − Tm (x) kW ≤k Tn − T2 kk x kV

it follows easily that {Tn (x)} is also Cauchy in W . As such we can define T0 by

T0 (x) = lim Tn (x).

n→∞

To see that T0 is linear observe that

T0 (αx + βy) = lim Tn (αx + βy)

n→∞
= lim αTn (x) + βTn (y)
n→∞
= αT0 (x) + βT0 (y)

To see that T0 is bounded first observe that being Cauchy, {Tn } is bounded. Hence we can find an M > 0
such that kTn k ≤ M for each n ∈ N. Moreover, since kT0 (x)kW = lim kTn (x)k ≤ M kxkV , we have that
n→∞
kT0 k ≤ M .

92
 Now let  > 0 and choose an N ∈ N so that if n, m ≥ N , then

kTn − Tm k < .

Let x ∈ V with kxkV ≤ 1. Then since kTn (x) − Tm (x)kW <  for each m ≥ N , we have

kTn (x) − T0 (x)kW = lim kTn (x) − Tm (x)kW ≤ .

m→∞

In particular
T0 = lim Tn
n→∞

in B(X, Y ).

Definition 6.1.7. Let (V, k · k) be a normed linear space. The space B(V, R) is called the dual space of V
and is denoted by V ∗ .

Example 6.1.8. 1) Let V = Rn with the usual norm k · k2 . For each a = (a1 , a2 , . . . , an ) we defined
φa : Rn → R by
n
X
φa ((x1 , x2 , . . . , xn )) = x · a = xi ai .
i=1

∗
Then in fact φa ∈ Rn and
kφa k = kak2 .

2) Let (X, d) be a compact metric space. Again, if x0 ∈ X and we define φx0 : (C(X), k · k∞ ) → R by

φx0 (f ) = f (x0 ),

then φx0 ∈ C(X)∗ . In this case kφx0 k.

3) Let (X, A, µ) be a measure space and let 1 ≤ p ≤ ∞ with p1 + 1q = 1. Hölder’s Inequality allows us to
define for each g ∈ Lq (X, A, µ) and element φg ∈ Lp (X, A, µ)] by
Z
φg (f ) = f g dµ.

Moreover, Hölder’s Inequality also shows that φg ∈ Lp (X, A, µ)∗ with

kφg k ≤ kgkp .

Note that φg has the additional property that if g ≥ 0 µ-a.e., then φg (f ) ≥ 0 whenever f ∈ Lp (X, A, µ)
and f ≥ 0 µ-a.e.
4) Let (X, d) be a compact measure space and let µ be a finite regular signed measure on B(X). Define
φµ ∈ C(X)] by Z
φµ (f ) = f dµ.

Since Z
|φµ (f )| ≤ |f | d|µ| ≤ kf k∞ kµkmeas

we see that in fact φµ ∈ C(X)∗ and kφµ k ≤ kµkmeas .

93
 We note again that φµ has the additional property that if µ is a positive measure on B(X), then
φµ (f ) ≥ 0 whenever f ∈ C(X) and f ≥ 0. Furthermore, in this case since
Z
φµ (1) = 1 dµ = µ(X) = kµkmeas ,

if µ is a positve measure we have

kφµ k = kµkmeas .

Problem 6.1.9. In Examples 3) and 4) above we have shown respectively that every element in Lq (X, A, µ)
determines a continuous functional on Lp (X, A, µ) and that if (X, d) is a compact metric space, then every
finite regular signed measure on B(X) determines a continuous linear functional on C(X). It is natural to
ask:
Do all continuous linear functionals on Lp (X, A, µ) and C(X) arise in this fashion?

6.2 Riesz Representation Theorem for Lp (X, A, µ)

In this section we will focus on the following problem:

Problem 6.2.1. What is Lp (X, A, µ)∗ ?

We have already established most of the following result:

Lemma 6.2.2. If (X, A, µ) is a measure space and if 1 ≤ p ≤ ∞ with p1 + 1q = 1, then for every g ∈ Lq (X, µ)
the map Γg : Lp (X, µ) → R defined by Γg (f ) = X f g dµ is a continuous linear functional on Lp (X, µ).
R

Further, kΓg k ≤ kgkq and if 1 < p ≤ ∞ then kΓg k = kgkq .

Proof. Assignment.
If (X, µ) is σ-finite, then equality holds for p = 1 as well.

Lemma 6.2.3. Let (X, A, µ) be a finite measure space and if 1 ≤ p < ∞. Let g be an integrable function
such that there exists a constant M with | gϕ dµ ≤ M kϕkp for all simple functions ϕ. Then g ∈ Lq (X, µ),
R

where p1 + 1q = 1.
1
Proof. Assume that p > 1. Let ψn be a sequence of simple functions with ψn % |g|q . Let ϕn = (ψn ) p sgn(g).
R 1 1
Then ϕn is also simple and kϕn kp = ( ψn dµ) p . Since |ϕn g| ≥ |ϕn ||ψn | q = |ψn |, we have
Z Z Z  p1
ψn dµ ≤ ϕn g dµ ≤ M kϕn kp = M ψn dµ

Therefore ψn dµ ≤ M q . By the Monotone Convergence Theorem we get that kgkq ≤ M , so g ∈ Lq (X, µ).
R

If p = 1, then we need to show that g is bounded almost everywhere. Let E = {x ∈ X| |g(x)| > M }. Let
1
f = µ(E) χE sgn(g). Then f is a simple function and kf k1 = 1. This is a contradiction.

Lemma 6.2.4. Let 1 ≤ p < ∞. Let {En } be aPsequence of disjoint sets. Let {fn } ⊆ Lp (X,Pµ) be such that
∞ ∞
fn (x) = 0 if x ∈
/ En Pfor each n ≥ 1. Let f = n=1 fn . Then f ∈ Lp (X, µ) if and only if n=1 kf kpp < ∞.
∞
In this case, kf kpp = n=1 kf kpp .
Proof. Exercise.

94
Theorem 6.2.5 [Riesz Representation Theorem, I]. Let Γ ∈ Lp (X, µ)∗ , where 1 ≤ p < ∞ and µ is
σ-finite. Then if p1 + 1q = 1, there exists a unique g ∈ Lq (X, µ)∗ such that
Z
Γ(f ) = f g dµ = φg (f )
X

Moreover, kΓk = kgkq .

Proof. Assume that µ is finite. Then every bounded measurable function isSin Lp (X, µ). Define λ : A →
∞
R : E 7→ PΓ(χE ). Let {En } ⊆ A be a sequence of disjoint sets, and let E = n=1 E Pn∞. Let αn = sgnΓ(χEn )
∞ p
P∞
and f = n=1 αn χEn . Then f ∈ L (X, µ) and Γ(f ) = n=1 |λ(En )| < ∞ and so n=1 |λ(En )| = Γ(χE ) =
λ(E). Therefore λ is a finite signed measure. Clearly, if µ(E) = 0 then χE = 0 almost everywhere, so
λ(E) − Γ(0) = 0. RTherefore |λ|  µ. By the Radon-Nikodym Theorem, R there is an integrable function g
such that λ(E) = E g dµ for all E ∈ A. If ϕ is simple, then Γ(ϕ) = ϕg dµ by linearity of the integral. But
|Γ(ϕ)| ≤ kΓkkϕkp for all simple functions ϕ, so g ∈ Lq (X, µ) by the lemma above. Now Γ − φg ∈ Lp (X, µ)∗
and Γ − φg = 0 on the space of simple functions. Since the simple functions are dense in Lp (X, µ), Γ − φg = 0
on LP (X, µ), so Γ = φg . We have that kΓk = kφg k = kgk Sq∞as before.
Now asume that µ is σ-finite. We can write X = n=1 Xn , where µ(Xn ) < ∞ and Xn ⊆ Xn+1 for
all n ≥R 1. For each n ≥ 1, the proof above gives us gn ∈ Lq (X, µ), vanishing outside Xn , such that
Γ(f ) = f g dµ for all f ∈ Lp (X, µ) vanishing off of Xn . Moreover, kgn kq ≤ kΓk. By the uniqueness of the
gn ’s, we can assume that gn+1 = gn on Xn . Let g(x) = limn→∞ gn (x). We have that |gn | % |g|. By the
Monotone Convergence Theorem
Z Z
q
|g| dµ = lim |gn |q dµ ≤ kΓkq
n→∞

Hence g ∈ Lq (X, µ). Let f ∈ Lp (X, µ) and fn = f χXn . Then fn → f pointwise and fn ∈ Lp (X, µ) for
all n ≥ 1. Since |f g| ∈ L1 (X, µ) and fn g| ≤ |f g|, the Lebesque Dominated Convergence Theorem shows
Z Z Z
f g dµ = lim fn g dµ = lim fn gn dµ = lim Γ(fn ) = Γ(f )
n→∞ n→∞ n→∞

If p = 1, then we cannot drop the assumption of σ-finiteness.

Theorem 6.2.6 [Riesz Representation Theorem, II]. Let Γ ∈ Lp (X, µ)∗ , where 1 < p < ∞. Then if
1 1 q
p + q = 1, there exists a unique g ∈ L (X, µ) such that
Z
Γ(f ) = f g dµ

for all f ∈ Lp (X, µ). Moreover, kΓk = kgkq .

q
R E ⊆ X be σ-finite. then there exists a unique gE ∈ L (X, µ), vanishing outside of E, such that
Proof. Let
Γ(f ) = f gE dµ for all g ∈ Lp (X, µ) vanishing outsideRof E. Moreover, if A ⊆ E, then gA = gE almost
everywhere on A. For each σ-finite set E let λ(E) = |gE |q dµ. If A ⊆ E, then λ(A) ≤ λ(E) ≤ kΓkq .
Let M =Ssup{λ(E)|E is σ-finite}. Let {En } be a sequence of σ-finite sets such that limn→∞ λ(En ) = M .
∞
If H = n=1 En then H is σ-finite and λ(H) = M . If E is σ-finite with H ⊆ E, then gE = gH almost
everywhere on H. But
Z Z
|gE |q dµ = λ(E) ≤ λ(H) = |gH |q dµ

so gE = 0 almost everywhere on E\H. Let g = gHχH . Then g ∈ Lq (X, µ) and if E is σ-finite with H ⊆ E
then gE = g almost everywhere. If f ∈ Lp (X, µ), then let E = {x ∈ X|f (x) 6= 0}. E is σ-finite and hence
E1 = E ∪ H is σ-finite. Hence

95
 Z Z
Γ(f ) = f gE1 dµ = f g dµ = φg (f )

Therefore Γ = φg and as before kΓk = kgkq .

We have shown that if 1 < p < ∞ and p1 + 1q = 1, then for any measure space (X, A, µ), Lp (X, µ)∗ = ∼
∗ ∼ ∞
L (X, µ). If µ is σ-finite, then L (X, µ) = L (X, µ). What happens when p = ∞? L (X, µ) ,→ L (X, µ)∗ ,
q 1 1 ∞

but this embedding is not usually surjective. There exists a compact Hausdorff space Ω such that L∞ (X, µ) ∼ =
C(Ω). What is C(Ω)?
Let ϕ : [a, b] → R be defined by ϕ(f ) = f (x0 ). Then ϕ ∈ C[a, b]∗ , and kϕk = 1. Let µx0 be the measure
Rb
on [a, b] of the point mass x0 . If g ∈ L1 ([a, b], m), then ϕg (f ) = a f g dm is a linear functional on C[a, b],
and kϕg k R≤ kgk1 . g is the Radon-Nikodym derivative R of an absolutely continuous measure µ on [a, b], and
ϕg (f ) = f dµ. If µ ∈ Meas[a, b], then ϕn (f ) = f dµ is a bounded linear functional on C[a, b], with
kϕµ k ≤ kµkMeas .

6.3 Riesz Representation Theorem for C([a,b])

Theorem 6.3.1. [Jordan Decomposition Theorem]
Let Γ ∈ C([a, b])∗ . Then there exist positive linear functionals Γ+ , Γ− ∈ C([a, b])∗ such that

Γ = Γ + − Γ−

and
k Γ k= Γ+ (1) + Γ− (1).
Proof. Assume that f ≥ 0. Define
Γ+ (f ) = sup Γ(ϕ).
0≤φ≤f

Then Γ (f ) ≥ 0 and Γ (f ) ≥ Γ(f ). It is also easy to see that if c ≥ 0, then Γ+ (cf ) = cΓ+ (f ).
+ +

Let f, g ≥ 0. If 0 ≤ φ ≤ f and 0 ≤ ψ ≤ g, then 0 ≤ φ + ψ ≤ f + g so

Γ(φ) + Γ(ψ) ≤ Γ+ (f + g)

and hence,
Γ+ (f ) + Γ+ (g) ≤ Γ+ (f + g).
If 0 ≤ ψ ≤ f + g, then let ϕ = inf {f, ψ} and ξ = ψ − ϕ. Then 0 ≤ ϕ ≤ f and 0 ≤ ξ ≤ g. It follows that

Γ(ψ) = Γ(ϕ) + Γ(ξ) ≤ Γ+ (f ) + Γ+ (g).

This shows that

Γ+ (f + g) ≤ Γ+ (f ) + Γ+ (g)
Therefore,
Γ+ (f + g) = Γ+ (f ) + Γ+ (g)
Let f ∈ C[a, b]. Let α, β be such that f + α1 ≥ 0 and f + β1 ≥ 0. Then

Γ+ (f + α1 + β1) = Γ+ (f + α1) + Γ+ (β1)

= Γ+ (f + β1) + Γ+ (α1)

This shows that

Γ+ (f + α1) − Γ+ (α1) = Γ+ (f + β1) − Γ+ (β1)
As such , if we let
Γ+ (f ) = Γ+ (f + α1) − Γ+ (α1),

96
then Γ+ is well defined.
Let f, g ∈ C[a, b]. Let α, β be chosen so that f + α1 ≥ 0 and g + β1 ≥ 0. Then f + g + (α + β)1 ≥ 0 so

Γ+ (f + g) = Γ+ (f + g + (α + β)1) − Γ+ ((α + β)1)

= Γ+ (f + α1) + Γ+ (g + β1) − Γ+ ((α + β)1)
= Γ+ (f + α1) − Γ+ (α1) + Γ+ (g + β1) − Γ+ (β)1)
= Γ+ (f ) + Γ+ (g).

That is Γ+ is additive.
It is also clear that Γ+ (cf ) = cΓ+ (f ) when c ≥ 0. But since Γ+ (−f ) + Γ+ (f ) = Γ+ (0) = 0, we get that

Γ+ (−f ) = −Γ+ (f )

so Γ+ is linear.
Let
Γ− = Γ+ − Γ
Since it is clear that Γ+ (f ) ≥ Γ(f ) if f ≥ 0, Γ− is also positive.
We know that
k Γ k≤k Γ+ k + k Γ− = Γ+ (1) + Γ− (1)
Let 0 ≤ ψ ≤ 1. Then k 2ψ − 1 k∞ ≤ 1. As such

k Γ k≥ Γ(2ψ − 1) = 2Γ(ψ) − Γ(1)

and therefore

kΓk ≥ 2Γ+ (1) − Γ(1)

= Γ+ (1) + Γ− (1)

Hence
k Γ k= Γ+ (1) + Γ− (1).

Theorem 6.3.2. [Riesz Representation Theorem for C([a, b])]

Let Γ ∈ C([a, b])∗ . Then there exists a unique finite signed measure µ on the Borel subsets of [a, b] such
that Z
Γ(f ) = f dµ
[a,b]

for each f ∈ C([a, b]). Moreover, k Γ k=| µ | ([a, b]).

Proof. First, we will assume that Γ is positive.

For a ≤ t < b and for n large enough so that t + n1 ≤ b, let

 1 if x ∈ [a, t]
ϕt,n (x) = 1 − n(x − t) if x ∈ (t, t + n1 ]
0 if x ∈ (t + n1 , b]


97
Note that if n ≤ m, then
0 ≤ ϕt,m ≤ ϕt,n ≤ 1

It follows that {Γ(ϕt,n )} is decreasing and bounded below by 0. Therefore, we can define

 0 if t < a
g(t) = lim Γ(ϕt,n ) if t ∈ [a, b)
 n→∞
Γ(1) if t ≥ b

Moreover, if t1 > t, we have

ϕt,m ≤ ϕt1 ,n .

98
 Since Γ is positive, g(t) is monotonically increasing.
It is clear that g(t) is right continuous if t < a or if t ≥ b. Assume that t ∈ [a, b). Let  > 0 and choose
n large enough so that
kΓk
n > max(2, )

and
g(t) ≤ Γ(ϕt,n ) ≤ g(t) + .
Let
if x ∈ [a, t + n12 ]

 1 2
n 1
ψn (x) = 1− n−2 (x −t− n2 ) if x ∈ (t + n12 , t + n1 − 1
n2 ]
if x ∈ (t + n1 − n12 , b]

0
Then
1
k ψn − ϕt,n k∞ ≤
n
.

Therefore,
1
Γ(ψn ) ≤ Γ(ϕt,n ) + k Γ k≤ g(t) + 2.
n

99
This means that
1
g(t) ≤ g(t + ) ≤ g(t) + 2.
n2
However, as g(t) is increasing, this is sufficient to show that g(t) is right continuous.
The Hahn Extension Theorem gives a Borel measure µ such that µ((α, β]) = g(β) − g(α). In particular,
if a ≤ c ≤ b, then
µ([a, c]) = µ((a − 1, c]) = g(c).
Let f ∈ C([a, b]) and let  > 0. Let δ be such that if | x − y |< δ and x, y ∈ [a, b], then

| f (x) − f (y) |< .

Let P = {a = t0 , t1 , . . . , tm = b} be a partition with sup(tk − tk−1 ) < 2δ . Then choose n large enough so that
2
n < inf(tk − tk−1 ) and
(*) g(tk ) ≤ Γ(ϕt,n ) ≤ g(tk ) + mkf k∞ .
Next, we let
m
X
f1 (x) = f (t1 )ϕt1 ,n + f (tk )(ϕtk ,n − ϕtk−1 ,n )
k=2

and
m
X
f2 (x) = f (t1 )χ[t0 ,t1 ] + f (tk )χ[tk−1 ,tk ] )
k=2

Note that f1 is continuous and piecewise linear. f2 is a step function. It is also true that both f1 and f2
agree with f (x) at each point tk for k ≥ 1. Moreover, the function f1 takes on values between f (tk−1 and
f (tk ) on the interval [tk−1 , tk ]. As such
k f1 − f k∞ ≤ 
and
sup{| f2 (x) − f (x) || x ∈ [a, b]} ≤ .
From this we conclude that
| Γ(f ) − Γ(f1 ) |≤  k Γ k .
We use (*) to see that for 2 ≤ k ≤ m

| Γ(ϕtk ,n − ϕtk−1 ,n ) − (g(tk ) − g(tk−1 )) |≤
m k f k∞

Next, we apply Γ to f1 and integrate f2 with respect to µ to get

Z
| Γ(f1 ) − f2 dµ |≤ 
[a,b]

We also have that Z Z

f2 dµ − f dµ |≤ µ([a, b]).
[a,b] [a,b]

Therefore, Z
| Γ(f ) − f dµ |≤ (2 k Γ k +µ([a, b]).
[a,b]

Since  is arbitrary, Z
Γ(f ) = f dµ
[a,b]

for each f ∈ C[a, b]. Moreover, k Γ k= Γ(1) =| µ | ([a, b]).

The general result follows from the previous theorem.

100
6.4 Riesz Representation Theorem for C(Ω)
In this section we will briefly discuss how to extend the Riesz Representation to C(Ω) when (Ω, d) is a
compact metric space. In fact we can state this extension in greater generality:

Theorem 6.4.1. [Riesz Representation Theorem for C(Ω)] Let (Ω, τ ) be a compact Hausdorff space. Let
Γ ∈ C(Ω)∗ . Then there exists a unique finite regular signed measure µ on the Borel subsets of Ω such that
Z
Γ(f ) = f dµ
Ω

for each f ∈ C(Ω). Moreover, k Γ k=| µ | (Ω).

Remark 6.4.2. Let µ ∈ M eas(Ω, B(Ω)). If Γµ is defined by

Z
Γµ (f ) = f dµ (∗)
Ω

for each f ∈ C(Ω), then Γµ ∈ C(Ω)∗ and

k Γµ k=| µ | (Ω) = kµkmeas .

Problem 6.4.3. For the converse how do we construct the measure µ?

Sketch: We will sketch a solution in the special case where (Ω, d) is a compact metric space.
By the Jordan Decomposition Theorem, we may again assume that Γ is positive.
Key Observation: Let K ⊆ Ω be compact. Assume that {ϕn } is a sequence of continuous functions
such that
0 ≤ ϕn+1 (t) ≤ ϕn (t) ≤ 1
for every t ∈ Ω with
lim ϕn = χK
n→∞

pointwise. Then
lim Γ(ϕn )
n→∞

exists. Moreover, if µ is a measure satisfying (∗), then the Lebesgue Dominated Convergence Theorem shows
that Z Z
µ(K) = χK dµ = lim ϕn dµ = lim Γ(ϕn ).
Ω n→∞ Ω n→∞

From here, let K be compact. For each n ∈ N let

[ 1
Un = B(x, )
n
x∈K

and let Fn = Ω \ Un . Then define

dist(x, Fn )
ϕn (x) =
dist(x, Fn ) + dist(x, K)

where dist(x, A) = inf {d(x, y) | y ∈ A}.Then ϕn (x) = 1 if x ∈ K and ϕn (x) = 0 if x ∈ Fn . Hence ϕn → χK

pointwise.
Moreover since {dist(x, Fn )} is decreasing, we get

0 ≤ ϕn+1 (t) ≤ ϕn (t) ≤ 1.

101