ENS Cachan Mathématiques

Preparation for the aggregation Year 2017/2018

Functional Analysis
Arthur Leclaire

References
[B] H. Bré[Link] Fonctionnelle. Dunod, 1999.
[CLF] A. Chambert-Loir, S. Fermigier, and V. Maillot. Math Exercises for the Aggregation. Masson, 1997.
[GT] S. Gonnord and N. Tosel. Topology and Functional Analysis. Ellipses, 1996.
[HL] F. Hirsch and G. Lacombe. Elements of Functional Analysis. Dunod, 1999.
[LM] G. Lacombe et P. [Link] Fonctionnelle, Exercices corrigés. Dunod, 1999.
[M] B. [Link] Fonctionnelle. Ellipses, 2004.
[S] L. [Link] Générale et Analyse Fonctionnelle. Hermann, 2008.
[Wa] C. [Link] et Analyse Fonctionnelle. Hermann, 2012.
[Wi] M. [Link] Fonctionnelle Élémentaire. Cassini, 2003.

The first chapters of the book [B] form a good summary of the major theorems of functional analysis.
(the English version contains corrected exercises). The book [HL] is also very clear and contains exercises.
pertinent points that are mostly corrected in [LM]. One can also refer to the very concise [Wi] which adopts a
a presentation that is a little less abstract (but be careful of non-standard notations). The books [Wa],[S] present the
grand theorems of functional analysis but in a more general framework than that addressed here. The books [LM],
[M], [GT], [CLF] contain many exercises (the last two being more difficult).

Table of contents
1 Some Classic Functional Spaces 2
1.1 Spaces with Semi-Norms 2
1.2 Regularizing Suites 3
1.3 Espaces de Fonctions de ClasseCk. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5. . .
1.4 L Spacesp. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

2 Continuous Linear Operators 15

2.1 General Information 15
2.2 The Banach-Steinhaus Theorem... 16
2.3 ... and its Spectacular Consequences. 17
2.4 Compact Operators 21
2.5 The Riesz-Fredholm Theorem 21

3 Duality 22
3.1 The Hahn-Banach Theorem 23
3.2 Dual of Classic Functional Spaces 24
3.3 Weak Topologies, Weak Convergence 27
3.4 Low Compactness 33

1
1 Some Classic Functional Spaces
1.1 Spaces with Semi-Norms
In this paragraph, it designates a vector space over K = R o U C.
We refer to [S] for the content of this paragraph.
Definition 1. A semi-norm on E is a mapping p: E → R + such as for all
vectors x, y, z ∈ E and any λ ∈ K,
p(0)=0,
p(λx)=|λ|p(x),
p(x+y) = 6p(x) + p(y).
A semi-norm is a semi-norm surE, six∈Eetr>0, we will note

Bp(x,r)={y∈E|p(x−y)<r}

open centered semi-sphere with radius r.

We consider a family(pi)i∈Iof semi-norms on E. We endow the topology O with
generated by the family of open semi-balls

{Bp(x,r);i∈I,x∈E,r>0}.
i

For this topology, we have the following properties.

The Intersections of open semi-balls form a basis of open sets.
A fundamental system of neighborhoods of x ∈ E is given by

Bp(x,rj);J⊂I nie,rj0.
\ j
j ∈ J

3. A sequence (xn elements of E converge towards x if and only if

For all i in I, pI(xn−x)−−−−→0.

n→∞

4. The topologyOest is separated if and only if(pi)i∈Iis separating, that is to say

For all x, 0, there exists i in I, p.i(x),0.

5. Let Fun be another vector space equipped with the topology defined by a family of seminorms.
norms(qk)k∈K . A linear application T:E→F is continuous if and only if for
For all k ∈ K, there exists J ⊂ I such that M > 0

For all x in E, qk(T(x))6Msuppj(x).

j∈J

6. If Iest ni and if(piis separating, then there exists a norm on Equi that induces the topology O.

2
 7. It is assumed here I=Net(pIseparating. The distance
1
d(x,y)= X minpk(x−y),1. (1)
2k
k is an element of the natural numbers
induces the topology O. Moreover, (xn is Cauchy if and only if

suppk(xm-xn → 0.
For all k in N,m,n >N N approaches infinity

Exercise 1. Show the above properties.

We will also note that the topology associated with a family of seminorms on Erend is continuous.
the addition (of E×E in E) as well as the scalar multiplication (of K×E in E). Furthermore, the
Intersections of open semicircles centered form a fundamental system.
of convex neighborhoods of dex. By using the gauge of a convex (cf. topology TD), we can
show that a topology making addition and scalar multiplication continuous and assigning
In every respect, a fundamental system of convex neighborhoods is necessarily defined by a
family of semi-norms [S].
Definition 2. A Fréchet space is a vector space equipped with a separating family.
countable of semi-norms (pk)k∈Nand which is complete for the distance defined by (1).

1.2 Regularizing Suites

Definition 3. A unit approximation is called a sequence (αn )n>1of measurable functions
positives on Rd , the integral 1 and such that

For all δ > 0,Z αn (x)dx−−−−→0.

n approaches infinity
|x|>δ
This is the definition given in [BP] limited to functions with positive values.
Proposition 1. If α: Rd→R+ is measurable with integral 1, then
αn (x)=nd α ( n x )
defines an approximation of the unit. In other words, we obtain an approximation of the unit by
renormalizing any probability density.
Exercise 2 .Soitf:Rd→This is (αn an approximation of unity.
When it makes sense, we will ask

αn∗f(x)= Z f(x−y)αn (y)dy.

Rd

1. It is assumed that the fest continues bounded on Rd . Show that αn∗well defined and thatαn∗f
tend uniformly on the compacts.
2. We are supposed to continue and there exists a compact such that Supp(αn For all n. Show
whatn∗party well defined and thatαn∗ftend uniformly on the compact sets.

3
Definition 4. A regularizing sequence is an approximation of the unit consisting of functions.
tionsC∞ to support compacts.

Exercise 3. There are regularizing sequences. [BP, Chap. 13]

It is about building α ∈ C∞ (Rd ,R+ compact support not identically zero.
1. Show that the function
ρ:t−→exp−1 t >0 1
t
is class C∞ surR.
2. Show that α: x → ρ(1 - |x|) answers2 the question and that moreover Supp(α) = B(0,1).
3. Show that φ: t → ρ(t)ρ(1−t) is of class C∞ support out[0,1]. Deduce that
α:x→φ(|x|) answers
2 the question. Also show that

ψ: R → R
u
1 .
u−→ Z φ(t)dt
Rφ 0

is a positive function C∞ such that {ψ=0} =]− ∞,0] and {ψ=1} = [1,+∞[.

Theorem 1. Sif:RdIt is locally integrable, and if α:RdRest of class Ckand to

support compact, so we can place

For all x in Rd ,α ∗f(x)= Z f(x−y)α(y)dy

Rd

what defines a function α∗ of class CksurRd Moreover, for any multi-index j of weight 6k,

∂ j(α∗f)=(∂j a ) * f .

In particular, if (αn ) is a regularizing sequence, then the functions αn∗fsontC∞ .

Exercise 4. Smooth version of Urysohn's lemma [BP, Chap. 13]
Let K be a compact subset of Rd and open ofRd such that K is a subset of omega.

[Link] why ρ := d(K, ωc)est>0.

ρ
2. Let η be fixed and2denote L = {u ∈ Rd|d(u,K)6η}.
We also introduce β∈C∞ (Rd ,R+ )de
supportB(0,η) and integral 1 (see Exercise 3). Verify that the function

χ(x)=1L∗β(x)= Z β(x−y)dy
L

estC∞ to compact support onRd , at values in [0,1], that it is 1 on Ket 0 on ωc .

4
1.3 Class C Function Spacesk
Refer to [ZQ] for this subsection.
In the previous TD sheet, we studied the space C(E) of continuous functions.
on a topological space E with values in K = R + C. We saw that C(E) can be endowed
of a metric inducing uniform convergence and making it complete. We also saw that
Cb(E) was a Banach space for the uniform norm; in particular, if K is a space
Compact topological space C(K) is a Banach space.
In this subsection, we will focus on spaces of regular functions on Rd On
will designate by an open set of Rd We will need the complete suite of compacts.

1
K jx is( an element of Ω such c
that the ) norm(j>1).
of x is less than or (2)
equal to d(x, Ω)
j

Thus, wej⊂Kj+1etΩ=K˚
j Starting from such a comprehensive sequence, one can construct a
suite(χj)of functionsC∞ surRdtoS values in [0,1] such that

χ j=1 surKj and χ j=0 surR\Kdj+1 ˚

(note that the second condition ensures that Supp(χj is included in Kj+1(but not in Kj+1 )). ˚
The existence of such plateau functions is given by Exercise 4.

Theorem 2. Let k ∈ N ∪ {∞}. The space Ck(Ω) of class C functions kequipped with the
countable family of semi-norms
p j,α ( f ) = s u p | ∂ α f(x)|(j>1,|α|≤k),
x is in Kj

which induces the topology of uniform convergence on the compact sets of Ω of the function and its
derivatives up to the order. It confers to Ck(Ω) a Fréchet space structure. Moreover, the sub-
spaceCck(Ω) of functions C k The support of the compact is dense in Ck(Ω). Better, C c∞ (Ω)
is dense in CkFinally, if K is a compact set included in Ω, then the subspace C Kk(Ω)des
functionsCkThe support included in is a closed set of C.k(Ω)(and therefore is complete).

Exercise 5. Demonstration of this theorem.

[Link] that(fn tend towards Ck(Ω) if and only if for all |α|6ket every com-
pactK⊂Ω,∂α fn→∂α uniformly on K.
2. Show that Ck(Ω)is complete.
[Link] that Cck(Ω)is dense in C k(Ω).
4. Show that Cc∞ (Ω) is dense in C k(Ω).
5. Show that CKk(Ω) is closed in C k(Ω).

5
Remark 1. The theorem states in particular that Ck(Ω) is complete for an associated metric
attached to the countable family of semi-norms (p j,α )j>1,|α |6kvia formula (1). It should be noted
that this definition assumes having chosen an enumeration of the family of semi-norms, and
so it is not canonical. In fact, the choice of such a metric does not really matter,
since all the metrics constructed in this way induce the same topology, the same
the notion of convergence, and the same notion of Cauchy sequence. However, in practice, you
must be able to write such a distance. As an exercise, we will practice doing so for
C(Ω),Ck (Ω)aveck<∞, etC∞ (Ω).
Theorem 3. For k<∞, the space Ck Functions that are 2π-periodic and of class CksurRest
a Banach space for the norm
k f k Ck(T) sup|f (j) (x)|.
j6k
x ∈ [0, 2π]

Exercise 6 .
1
[Link] C([0,1]) equipped with the norm kfk?1= | f ( x ) R
| d0 x i s c o m p l e t e ?
[Link] it Cc(R) endowed with the uniform norm is complete?
3. Is it CbIs (R) complete as a subspace of C(R)?
[Link] it C1 (]−1,1[) is complete as a subspace of C(]−1,1[)?
Note 2. It should be noted that Cck(Ω) is not complete as a subspace of C.k(Ω)
(because it is not closed).

Exercise 7*. Characterization of the compacts of C(Ω) [HL, Chap.2, §1, Exo 12]
We will see that A ⊂ C(Ω) is relatively compact in C(Ω) if and only if
Aest equicontinuous at every point of Ω,
(ii) for all x ∈ Ω, A(x) = {f(x), f ∈ A} is bounded.
[Link] that the conditions are necessary.
2. Let us suppose (i) and (ii). Let (fn a sequence of elements of A.
a. Using the Arzelà-Ascoli theorem on each Kj , show that(fn (admit a sub-
A sequence that converges uniformly on compact sets towards a function f ∈ C(Ω).
[Link].

Exercise 8*. IDEVJ Characterization of C compacts∞ (Ω)[ZQ]

It is said that A ⊂ C∞ (Ω) is bounded if for every neighborhood V of zero in C∞ (Ω), it
exists > 00 such that
For all t0greater than t, A is a subset of tV.
1. Show that A is bounded if and only if the derivatives of the elements of A are uniformly bounded.
bounded on every compact, that is to say that for all j > 1, and every multi-index α, there exists M > 0
as it is
For all f in A, supremum of the absolute value of the derivative. α f(x)|6M.
x is an element of Kj

6
2. We assume that A is compact. Show that A is bounded.
3. We assume that A is bounded. Let (fn a sequence of elements of A.
a. Using the characterization of the compacts of C(Ω) (see Exercise 7), show that one can
extract from(fn a subsequenceψ (n) such that for every multi-index α,(∂ α fψ (n) )n∈N converge
uniformly on the compact sets of Ω towards a function fα ∈C(Ω).
b. Conclude that (fψ (n) converge to a function in C∞ (Ω).
c. Conclude that A is compact.
4. Conclude that C∞ (Ω) is not normalizable.

Exercise [Link](Ω) is not normalizable [ZQ]

Let k ∈ N ∪ {∞}. Suppose there exists a norm k · k that induces the topology of C.k(Ω)(i.e.
the topology of locally uniform convergence of derivatives up to order k.
1. Show that B = {f ∈ Ck(Ω)|kfk61} contains a set of the form

V=f∈Ck (Ω)|supp j,α ( f ) < ε ,

|α|6l
wherej>1etl6k.
[Link] that there exists φ∈Ck(Ω) not identically zero such that ∀λ>0, λφ∈B.
3. Conclude.

Exercice 10. EspaceH(Ω), Théorème des familles normales [AM, §3.5]

Let Ω be an open set of C. We denote H(Ω) the space of holomorphic functions on Ω which is
a study of the topology of uniform convergence on compact sets.
1. Show that H(Ω) is a closed subspace of C(Ω).
2. It is said that A ⊂ H(Ω) is bounded if for every compact K,

sup sup|f(x)|<∞.
f belongs x
to Ai s an element of K

Show that A is compact in H(Ω) if and only if A is bounded.

Exercise 11 Space the Schwartz

We introduce the space S(Rd)functions of class C∞ surRdsuch as

For all p in N, Np( f ) : = s u p s u p | x β ∂α f(x) | < ∞.

|α|6p x ∈ R
d

|β|6p
This space is equipped with the countable family of semi-norms.p )p∈N .
1. Show that S(Rd)is complete for the distance
1
d ( f , g ) = X p min 1, Np( f − g ) .
2
p is an element of natural numbers

2. Show that the canonical injection of S(RdinC∞ (Rd)is ongoing.

7
1.4 L Spacesp
The reader will find in [BP] a very detailed presentation on the spaces.p . Le livre [Far]
provides a more concise and accessible presentation.

Let (E, A, µ) be a measurable space. We denote K = RouC.

Let Ω be an open set of Rdequipped with his Borealian tribe and the restriction of the measure of
Lebesgueλ.
Let p ∈ [1, ∞]. We denote q ∈ [1, ∞] as the conjugate exponent of p (which satisfies +q=1). 1 1
p

Definition 5. Let f:E→K be measurable. For p∈[1,∞[, we define

1/p
p etkfk∞ =infM>0|f|6Mµ−p.p.
k f k p= Z | f | d µ !
E

Theorem 4. Hölder's Inequality

For all measurable functions f, g: E → K,

Z | f g | d µ 6 k f k pkgkq.
E

Remark 3. Suppose 1 < p, q < ∞, kfkp<∞etkgkq<∞. So we have the case of equality in

Hölder if and only if there exists (α, β) ∈ R2{(0,0)}suchthatα|f= everywhere. p
| β|gµ-almost
| q
Do not forget the powers, which did not appear in the case of the Cauchy-Schwarz equality.
Refer to [BP] for more details on these equality cases.

Corollary 1. Minkowski inequality

For all measurable functions f, g: E → K,

k f + g k p6kfkp+kgkp .
Proposition 2. Definition of Lp
Notep(E,A,µ)(ouLp (E) or simply Lpwhen there is no ambiguity) the whole of
functions f:E→K measurable such that ||f|| p<∞, quotiented by equality µ-almost everywhere.
Thus, Lp is a vector space, and the semi-norm k · k ppass to the quotient and define a norm
onLp .
Theorem 5. Riesz-Fischer Theorem
Lp(E,A,µ) is a Banach space.

Corollary 2. L2 (E,A,µ) is a Hilbert space for the inner product

fg¯dµ.
hf,gi= Z
E

8
Exercise 12 Proof of the Riesz-Fischer theorem [R], [Far]
1. Handle the case when p=∞.

2. We now suppose p < ∞.

[Link]:E→Rmesurables positives. Montrer que
∞ ∞
X gn 6 X kgn kp.
n=0 p n=0

.Soit(fn a Cauchy sequence in Lp Show that we can extract(fnsuch as k

X k f nk−f
+ 1 k <∞.
n p k
k

∞
[Link] that
X ( f nk −f
+ 1 n)converge
k absolutely µ-almost everywhere.
k=0
On pose
∞
f(x) = fn(x)+
0 X fn(x)−f
k+1 n(x) k
k=0

(which exists for almost everything). Show that f ∈ Lpand whatn→fdansL

k
p.
[Link].
Proposition 3. Any Cauchy sequence of Lpadmit a subsequence that converges almost everywhere.
In particular, any convergent sequence of Lpadmit a subsequence that converges almost everywhere.
Exercise 13 Find counterexamples illustrating the following two statements.
[Link] everywhere convergence does not imply L convergence.p .
2. The convergence Lpdoes not imply convergence almost everywhere.

Theorem 6. It is assumed that p < ∞.

The set of integrable step functions is dense in L.p(E).
The set of linear combinations of indicator functions of cubes is dense in Lp(Ω).
[Link](Ω)is dense in L p(Ω).
Corollary 3. For p<∞, the space Lp(Ω) is separable.
Theorem 7. Let (αn an approximation of the unit and let f ∈ Lp(Rd).
It is assumed that p < ∞. Then

αn *f(x)= Z f(x−y)αn (y)dy

Rd

define a functionαnf belongs to Lp(Rd )et on a

k·k p
αn f n→∞

9
Note 4. One should be careful that the usual proofs of this result rely on
on the density of Cc Rd(see Exercise 20).

Theorem 8. If Sip < ∞, then Cc∞ (Ω) is dense in L p(Ω).

Proposition 4. Let us recall the definition of the exhaustive sequence of compact [Link] by
NoteLplocThe set of measurable functions f: Ω → K is invariant

p
j than 1, f1K∈L (Kj )
For all j greater

quotiented by the relation of almost everywhere equality. The countable family of semi-norms

q j( f ) = k f 1 Kkpj (j>1)

confer toLploc(Ω) a Fréchet space structure.

Proposition 5. Let us note LpThe set of measurable functions f: R → K verifies

f(x)=f(x+2π) a.e.

quotiented by the relation of equality almost everywhere. So Lp (T)est a Banach space for the
norm 1/p
2π
1 p
k f k L(T)
p = 2π Z
0
|f(x)|dx!

and the2 (T)est a Hilbert space for the inner product

2π
1
hf,gi= 2π Z 0 f(x)g(x)dx.

Exercise 14 .
Show that D={f∈C1 ([0,1])|f(1)=0} is dense in L1 ([0,1]).

Exercise 15 Links between theLp

1. It is assumed that µ(E) < ∞. Show that q > p ⇒ Lq ⊂Lp .
2. Give a counterexample in the case where µ(E) = ∞.

Exercise 16. Study ofp(N)[M]

For p ∈ [1, ∞[, we denote pthe space of sequences(xn such asp:= p <∞.
X |xn | 1/p

It is also noted∞ the space of sequences (xn )such as kxk∞ sup|xn |<∞.
[Link] thatpis a Banach space.
[Link] that for any sequence (xn and all them, we have∞ 6kxkq6kxkp6kxk1.

10
3. We suppose p < ∞.
[Link]∈N, we denote δn=m=n m∈N ) Show that for all x ∈ `p ,

∞
x= X xnδnwhere the series converges inp .
n=0

b. Deduce that the subspace of almost null sequences is dense inp .

c. Conclude thatpis separable.

Exercise 17.
1. Show that if sip<∞ or if µ is sigma-finite, then ∀f ∈ Lp ,kfkpsup Z fgdµ.
g∈L q
kgkq 61
2. Show that the result is no longer true when p = ∞ and µ is not sigma-finite (one can construct
a counter-example in the case E={a,b}, µ({a})=1, µ({b})=∞).

Exercise 18. [B, Lemma IV.2]

Let f ∈ Ll1ocsuch as

Z f(x)φ(x)dx=0.
For all φ in Cc∞ (Ω),
Ω

We are going to see almost everywhere.

1. Show that we can assume Ω bounded and f ∈ L1 (Ω). We will do it later.
2. On ⍞xeε>0. Show that there exists g ∈ Cc(Ω) such as kf - gk 1<ε.
3. Show that

K + ={x∈Ω|g(x)>ε} and K={x∈Ω|g(x)≤−ε}

−

are compact sets included in Ω. We will note K = K+ ∪K. −

4. Show that there exists φ ∈ C∞ (Ω)
c
such as quekφk ∞ 61, valant1surK+ et−1 on K.
−
4. c i c e E x eyscohr df.
n’,Urlemdmuelisseervsiloan utiliseoruprraO:n Indication

5. a. Show that gφ6ε.

RΩ
b. Show that |g|6ε+
RK RΩ\K |g|.
c. Conclude that
Ω R |g|6ε+2ε|Ω|.
d. Infer that fk162ε + 2ε |Ω| and conclude.

Note 5. Regarding the previous exercise, a much shorter proof can be given.
using an approximation of the unit. But in his book, Brézis uses the result of the ex-
nice to show the density of Cc(Ω)inL p(Ω),p<∞(and therefore cannot use the
result of the approximation of the unit that follows from it.

11
Exercise 19.L∞ (Ω) is not separable [B]
1. For all a ∈ Ω, we notea=sup{r>0|B(a,r)⊂Ω}, so that B(a,ra )⊂Ω.
On noteOa the ball of L∞ (Ω) centered on 1 B(a,r a) and of radius.21
Show that (Oa )a∈Ω is a family of non-empty open sets that are pairwise disjoint.
2. Show that L∞ (Ω) is not separable.

Exercise 20 Continuity of the translation operator on Lp(Rd)et applications

A part of the exercise can be found in [BP, Chap.13].
Pourh ∈ Rdetf∈Lp(R d) on noteτ hf(x)=f(x−h).
1. Show that τhf to τh made an isometry of Lp(Rd).
2. a. We assume that p < ∞. Show that for f ∈ L p (R d) , on a

1
limkτhf−fk p=0, and limkτhf−fk p=2kfkp. p
h→0
|h|→+∞

Do these results persist when p=∞?

3. (Proof of Theorem 7) Let (αn an approximation of the unit, and let f ∈ Lp(Rd). On
Suppose first that p < ∞.
a. Show that

αn*f(x)= Z f(x−y)αn (y)dy

Rd
define a function αn f belongs to Lp(Rd).
p p
[Link] that kαn*f - fkp6 Z kτyf−fk pαn (y)dy.
Rd
c. Deduce that
k·k p
αnf → f.n approaches infinity
Does this result persist for p=∞?

4. (Proof of Theorem 8) Show that Cc∞ (Ω)is dense in L p(Ω) for p < ∞.
5. Soit f ∈ L2 (R)
derivable on Ret such that f0∈L2 (R).
Show that there exists desfn∈C∞ (R) such as fntend towards L2ETFn0tend towards0in the2 .
6. [KLR, §8.3] Sif ∈ L1 (R), show that

ˆ
f(ξ):= −iξ x
Z f(x)e dxapproaches0.
R |ξ|→∞
.x d+ξ x( ξ i− e)x(f R Indication
)
−=)ξ(f,0,ξo
ˆ you can ask to the marquis: n
π

12
 The purpose of the two following exercises is the characterization of the compacts of L.p The technique
of proof is the same in the case of Rdand any open set Ω. But there are small difficulties
techniques appear in the second case, and that is why it is addressed in a separate exercise.

Exercise 21. IDEVJ Riesz-Fréchet-Kolmogorov Theorem on R d [HL]

Let p<∞ and let A ⊂ Lp(R d).
We will see that Aest is relatively compact in L.p(Rdif and only if
(i) Aest limited in Lp(Rd),

(ii) limit Z p
|f(x)|dx=0 uniformly with respect to f∈ A,
R→+∞
|x|>R
(iii)limτaf=fdansLp(Rd), uniformly with respect to f ∈ A.
a→0
1. Show that the relative compactness of A is equivalent to its precompactness.
2. Let's suppose A is relatively compact in Lp . Show the properties (i), (ii), (iii).
3. Let's assume (i), (ii), and (iii). Let's fix (ρn ) a regularizing sequence such that Supp(ρn )⊂B(0,n). 1

Let us give ε > 0, R > 0 and n ∈ N ∗ .

a. Show that

For all f in L p(Rd),kf−f*ρ nkp6supkf−τ yfkp .

|y|61/n

b. Note the closed ball of Rdof radius R.

Show that
(f*ρn )|Bf ∈ A
is relatively compact in C(B). Deduce that there exists f, ..., fk ∈1 Atelles that for
For every function f ∈ A, there exists j ∈ {1, ..., k} such that

For all x in B, |(f*ρn ) x ) - ( f j∗ρn )(x)|6el(B) −1/p .

[Link] by writing

| f − f j|6|f|1 B+|fj1c B+|f−f∗ρ n |+|f j−fj∗ρn |+|f∗ρn-fj∗ρn 1 B

c

and by adjusting carefully

13
Exercise 22*. IDEVJ Riesz-Fréchet-Kolmogorov Theorem on Ω[B]
Let p < ∞. We will denote ω ⊂⊂ Ω to say that ω is compact and included in Ω.
Let (ρn )a regularizing sequence such that Supp(ρn )⊂B(0,n ). 1
Let A ⊂ L(Ω). ˜
p For f ∈ L(Ω), wep denote by f the extension defined by zero on R. d

1. We assume that
(i) Aest limited in Lp(Ω),
(ii) For all ε>0, there exists ω contained in Ω,p such that sup kfkL(Ω\ω) 6e,
f belongs to A

(iii) For all ε > 0, for all ω ⊂⊂ Ω, there exists δ ∈ ]0, d(ω, Ω)c)[,sup supkτ hf−fk
p L(ω) <ε.

|h|<δf∈A
Fix ε > 0 and ω ⊂ Ω verifying (ii).
a. Show that for all n > 1, An={(f∗ρn ) |ω ;f ∈˜ A} is relatively compact
inC(ω).
b. Deduce that Aω ={f;f∈ A } can be covered by a finite number of balls
|ω
deLp(ω) of radius 2ε. Conclude that A is relatively compact in L.p(Ω).
2. Show the converse.
Note 6. If you wish to propose this theorem in development, you must know
that the result is used to show part of Sobolev injections. In this case, you
You are warmly invited to work on the following exercise, as well as Theorem VIII.7 from [B].
Exercise 23*. Sobolev space in dimension 1 [B],[CLF2]
Let I be an open interval of R for p ∈ [1, ∞]. We define W1,p (I) like the whole of
functions f ∈ Lp(It) tells that it exists g ∈ Lp(I) tell as

0
For all φ in Cc∞ (I)Z φ (x)f(x)dx = -φ(x)g(x)dx.
Z
I I
1. Show that such a function is necessarily unique (one can think of Exercise 18).
Mount that sif ∈ C1 (I)∩Lp(I), and if0 ∈Lp(I), then f ∈ W1,p (I)avecg=f0.
Sif ∈ W1,p(I), we can therefore note0the function appearing in the definition.
2. Suppose we are here I = ]0,1[. Show that f: x → |x| + x is in W.1,p(I) for all p∈[1,∞] and
calculate0Is it0 ∈W1,p (I)?
3. Show that W1,p(I) is a Banach space for the norm ||f|| = ||f||p+kf0 kp.
Show that sip<∞,C1 (I)∩Lpis dense in W1,p(I).
4. Let f ∈ W(I).1,pShow that there exists a unique function f ∈ C(I) that ˜ coincides
¯ almost
everywhere with such
y
¯ ˜ ˜
For all x, y in I, f(x) - f(y) = Z f 0 (t)dt.
x
5. We now suppose p to be bounded. We denote the conjugate exponent of p.
a. Show that if p > 1, the unit ball of W1,p(I) is relatively compact in C(I).
b. Show that if q < ∞, the unit ball of W1,1 (I) is relatively compact in Lq (I)
[Link]’Exdeer reslueltatutiliseoruprrao,nb our Indication

14
2 Continuous Linear Operators
For this part, we refer to [B] or [HL].

2.1 Generalities
Definition 6. Let E and F be two normed vector spaces.
Let note L(E,F) be the set of continuous linear applications from E to F. We will also denote
L(E)=L(E,E). We denote Isom(E,F) the set of bicontinuous isomorphisms from E to F. We
noteGL(E)=Isom(E,E), which is also the set of invertibles of L(E).
Proposition 6. Let L be a linear application from E to F. Then L ∈ L(E, F) if and only if

There exists C greater than 0, for all x in E, the norm of kL(x)k is less than or equal to C times the norm of kxk.
In this case, the best constant is called the operator norm of Let denoted

kLk=kLk L(E,F ) sup kL(x)k =supkL(x)k.

x,0
kxk kxk=1

Proposition 7. k · kL(E,F ) is a norm on L(E,F).

If E is a Banach space, then L(E,F) is also.
Exercise 24 .Forf∈L1 (Rd), the Fourier transform is defined as

d ˆ
For all ξ in R, f(ξ) = Z f(x)e−iξ .x dx.
Rd

1. Show that a continuous linear application F:L is defined like this.1 (Rd)→Cb(Rd).
1 d to ˆzero at infinity.
2. Show that for all f ∈ L(R), f tends

Exercise 25 .Pourx=(xn )n∈N∈`p(16p6∞), we define

Lx=(xn+1 )n>0etRx=(xn−1 1n>1 )n>0 .
The applications LetRs are respectively called shift to the left and shift to the right.
1. Show that L ∈ L(`p )etR∈L(`p and calculate their operator norms.
2. Show that L is surjective but not injective and that R is injective but not surjective.

Proposition 8. Let E and F be two Banach spaces.

Thus, Isom(E,F) is an open set of L(E,F). In particular, GL(E) is an open set of L(E).

Exercise 26 Demonstration of Proposition 8

1. Let H ∈ L(E) such that ||H|| < 1. Show that I - H ∈ GL(E) (explicit the inverse).
2. Show that Isom(E,F) is open in L(E,F).
3. Show that L → L−1is still in Isom(E,F) in Isom(F,E).

15
2.2 The Banach-Steinhaus Theorem...
Theorem 9. Banach-Steinhaus Theorem
Let E be a Banach space and F be a normed vector space. Let Ti)I∈Ia family (not necessarily-
countably many continuous linear applications from E to F. It is assumed that

For all x in E, sup kTi(x)k<+∞.

i is an element of I

So
supkTikL(E,F ) <+∞.
i is an element of I

More specifically, if
supkTikL(E,F ) = +∞,
i is an element of I

so, there exists aGδ -denseΩdeEtel que

For all x in Omega, supremum of kTi(x)k= +∞.

i∈I

Corollary 4. Let E be a Banach space and F a normed vector space.

Let (Tn A suite of continuous linear applications from E to F such that for every x ∈ E, the
suite(Tn (x) converges in F to a limit denoted T(x). Then T ∈ L(E, F) and

kTkL(E,F ) ≤lim infkT k .

n → +∞ n L(E,F )

Remark 7. The theorem is false if E is not a Banach space. Indeed, in the space E
real sequences that are null from a certain rank, equipped with the supremum norm, we obtain a counter-
example considering the operators Tn:E→R defined by
n

For every u in E, Tn u =X up.

p=0

Exercise 27. Demonstration of Theorem 9 [B]

1. We consider, for n ∈ N∗ , the sets An={x∈E,| ∀i∈I,kTi(x)k ≤ n.
∗ ◦
0 , asAn,∅.
Show that there exists n ∈ N 0

2. Deduce the existence of an open ball B(x, r), such

0 that

For all z in0B(x, r), kTi(z)k ≤n. 0

3. Conclude.

16
2.3 ... and its Spectacular Consequences
Theorem 10. There exists a Gδ dense Ω ⊂0 C(T) such that for all f ∈ Ω, the Fourier
0 series of
diverges to 0.

Exercise 28 .IDEVJDemonstration of Theorem 10 [R]

Questions 1 to 3 of this exercise aim to demonstrate the theorem. The end of the exercise
allows us to show an even stronger result, namely: there exists a Gδ dense of functions
continues for which the Fourier series diverge on Gδ -dense deR.
For f ∈ C(T), we define for n ∈ Z,
2π
cn ( f ) : = 1 Z −int
2π 0 f(t)e dt
and for n ∈ Netx ∈ R,

n
Snf(x):= ikx and Λnf=Snf(0).
X ck( f ) e
k = -n

Well, let's take notes.∗ ( f , x ) = s u p n∈N|Snf(x)|.

1 2π
[Link] that Λn is a continuous linear form on C(T) of norm kDnkL(T) 1 = 2π R0 |Dn |
whereDn ∈C(T)0is the Dirichlet kernel defined by

sin(n+)t 21
For all t in the interval (0, 2π), Dn (t)=
sin 2t

[Link] that kDnkL(T) −1−−−→∞.n→∞

3. Deduce that there exists a G.δ -denseΩ⊂C0 (T)such that ∀f∈Ω, s ∗ ( f , 0 ) = + ∞ .
0 0
4. Generalize this property to all x ∈R. We note ΩxleGδ -dense obtained.
5. So let X ⊂ ℝ^d be countable. Show that there exists a G.δ -denseΩX⊂C0 As such

For all f in Omega Xfor all x in X, s∗ ( f , x ) = + ∞

6. Show that if X is dense in R, then for all f ∈ ΩX,

A fx ∈ R, s∗ ( f , x ) = + ∞ }

is aGδ -dense in R.
7. Conclude.

The interest of the end of the exercise comes from the result demonstrated in the following exercise.

17
Exercise 29. [R]
We will show that in a complete metric space without isolated points, no set ...
countable is not a Gδ dense.
Let X=n∈NOn={xn { n ∈ N } u n G δ -countable dense.
T
1. Show that On {xn It is still a dense open.
[Link].

Exercise 30 [CLF3]
1
Let p ∈ [1, ∞] and q the conjugate exponentp1+ q=1). Let(a N
n )∈C as such

∞
For all (bn )∈`Xq , anbn converge.
n=0

Show that (an )∈`p .

Exercise 31. Continuity of bilinear applications [G-An], [Wa, §3.12]

1. Let E, F, G be Banach spaces and let a: E × F → G be a bilinear application. We
suppose that it continues separately, that is to say
y → a(x, y) is continuous.
For all x in E,
x→a(x,y) is continuous.
For all y in F,
Show that it is continuous.
1
2. On munitE=R[X] of the normekPk1= |P(x)|dx. Show that
0 R
a: E × E −→ R
1
(P,Q)−→
R0 P ( x ) Q ( x ) d x
is separate continues, but does not continue.

Before discussing the consequences of the Banach-Steinhaus theorem in operator theory

let's mention two other practical applications:
Non convergence of Lagrange interpolation, Runge's phenomenon (see TD of Inter-
Interpolation and Approximation or [CM] and [Wa, Exo 3.12.5].
Integration Numerical: Gaussian methods, Polya's theorem. [CM].

Theorem 11. Open Mapping Theorem

Let E and F be Banach spaces, T: E → F is a continuous linear operator.
If it is surjective, then there exists c > 0 such that

B F(0,c)⊂T(B)E(0,1)).

18
Theorem 12. Banach's Theorem
Let E and F be Banach spaces and T: E → F be a continuous linear operator.
If the test is bijective, then its inverse is continuous, and thus T ∈ Isom(E, F).

Note 8. An application is said to be open when the direct image of an open set is open.
One can show (exercise!) that the conclusion of Theorem 11 amounts to saying that Test a
open application, which explains the name of the theorem.
Note 9. If T: E → F is a continuous linear injective map, and if T−1Im(T) is continuously mapped to Eest,
so closed image test. This property is sometimes used to show that the inverse cannot be
continu.

Exercise 32 . On the assumptions of the Open Mapping Theorem.

1. Show that there exists a continuous linear application f: R → R that is not open.
[Link] that there exists a continuous surjective function f:R→R which is not open.
3. On note The set of real sequences (xn )n>1almost none equipped with the superior standard.
Show that there exists T ∈ L(E) bijective, whose inverse is not continuous.

Exercise 33 [B]
Let E be a normed vector space equipped with two norms k · k.1it · k2 , and complete for these
two standards. It is also assumed that

There exists k greater than 0, k multiplied by k.1≤kk · k2.

Show that the two norms are equivalent.

Exercise 34. [R]

Let us note (CZ the set of sequences(cn which tend to 0 when |n| → 0 endowed with the norm
0
Hi. It is reminded that the Fourier coefficients are defined in Exercise 28. By setting

For all f in L 1 (T),c(f):=(cn ( f ) ) ,n∈Z

we define a linear application c ∈ L(L1 (T),CZ ).
1. Show that it is valued in (CZ ). 0
2. Show that c is in L(L)1 (T),(C Z ) ).
0
3. Show that it is injective.
4. Let's suppose for the sake of contradiction that it is surjective.
a. Show that there exists δ > 0 such that,

For all f in L 1 (T),kc(f)k∞ >dkfk

1 .
L(T)

b. Obtain a contradiction by considering the functions [Link] in Exercise 28.

[Link].

19
Remark 10. We can explicitly construct a convergent trigonometric series that
is not the Fourier series of a function L1 (T)(see the Fourier Analysis TD or [ZQ]).
Theorem 13. Closed Graph Theorem. Let E and F be two Banach spaces and T: E → F
linear. The graph of Gest by definition the set
G ( T ) = { ( x , T x ) ; x ∈ E } ⊂ E × F.
If SiG(T) is closed in E×F, then T is continuous.
Note 11.
It will be noted that the converse is much easier. Indeed, as an exercise, we show...
The graph of a continuous application between two metric spaces is closed.
- Let's explain the practical importance of the closed graph theorem. To show that...
we generally have to show that

xn→x=⇒T(xn )→T(x).
According to the theorem, for a linear application T between two Banach spaces, it is sufficient to
verify that
xn→xetT(xn y = T(x)
in other words, we can assume a priori (T(xn )) also converges. By exploiting linearity
it is necessary to verify that

xn→0 etT(xn y = 0.
Exercise 35*. Closed subspaces of C([a,b]) formed by regular functions [GT].
Let Fun be a closed vector subspace of C([0,1]) equipped with uniform convergence.
1. It is assumed that all elements of F are in C.1 ([0,1]).
a. By using the closed graph theorem, show that there exists C > 0 such that

For all f in F, kf 0 k∞ 6Ckfk∞ .

b. Deduce that the closed unit ball BF deF (for the infinity norm) is compact.
c. What can we conclude from this?
2. We now suppose that the elements of F are Hölder functions, i.e.
For every f in F, there exists an α > 0, there exists a C > 0, αfor all x, y in [0, 1], |f(x) - f(y)| ≤ C|x - y|.
a. Using Baire's theorem, show that there exists n > 1 such that
For all f in F, for all x, y in [0, 1], the absolute value of f(x) minus f(y) is less than or equal to 6n times the absolute value of x minus y. 1/nk f k ∞ .
b. Conclude by using the method of the first question.

Exercise 36. [ZQ, Chapter VI, §2]

Let E be a Banach space and F a closed subspace of E.
Show that there exists a continuous linear projector from E to F if and only if there exists a
closed subspace GdeEtel that
E=F⊕G.

20
2.4 Compact Operators
Definition 7. Let E and F be two Banach spaces, and T: E → F be a linear operator.
It is said that an operator is compact if the image of the unit ball B...EdeEparTest relative-
compactly included in F, that is to say T(BE )is compact. We will note K(E,F) the set of
compact operators of EdansF, and K(E)=K(E,E).

Proposition 9. Let E, F, G be three Banach spaces, T ∈ L(E, F) and S ∈ L(F, G).

If S is compact, then S◦T is also.
Example 1. [HL, Chap.1, Prop. 3.4]
SiX,Y ⊂ Rdare compact and if K ∈ C(X×Y), then the operator T: C(X) → C(Y) defined by

T f(y)= Z K ( x , y ) f ( x ) d x
X

it is compact.

Exercise 37. [B]

1. Show that K(E, F) is a closed set of L(E, F).

2. Deduce that if we have Tn →TdansL(E,F)avecrg(Tn If <∞, then compact test.

3. Now suppose that F is a Hilbert space. Let T ∈ K(E, F) and ε > 0.
In recoveringK=T(BE by a certain number of balls of radius ε, and using a
Orthogonal projection, show that there exists a rank operator T.ε ∈L(E,F) such that

kTε − T k < 2 ε .
Note 12. On the other hand, outside the Hilbertian framework, a compact operator may not be
limit of rank operators is defined. We refer to the references given in [B].

Exercise 38.
Let (αn )∈`∞ . Show that T:(xn )∈`2→(αn xn )∈`2is compact ssiαntend to 0.

Other examples of compact operators are given by Hilbert-Schmidt operators.

see for example [CLF3].

2.5 The Riesz-Fredholm Theorem

Theorem 14. Fredholm alternative, light version
Let E be a normed vector space and T a compact operator of E. Then
1. Ker(Id−T) is of dimension one.
[Link](Id−T) is closed.
3. Ker(Id−T)={0} ⇐⇒ Im(Id−T)=E.

21
Note 13.
- Point 3 states that compact disturbances of identity are injective if and
only if they are surjective. Well-known property in infinite dimension (where an endomorphism-
a mapping is bijective if and only if it is injective or surjective), it is remarkable in dimension
inspire.
We refer to [B] for a detailed version of this result. The tricky point is to define the ad-
joint of an endomorphism in the general framework of Banach spaces. However, we will note
that Hilbert spaces form a relatively lightweight framework to state and prove the re-
specified result (because then we can define the adjoint using Riesz's representation theorem).
We will take care to note that the proof of point 3 given in [B] uses the notion of opera-
deputy adjunct. However, as noted in the M1 handout by L. Desvillettes, one can also
provide a direct proof. Indeed, suppose Im(Id−T)=E. If we had Ker(Id−T),0, then
Ker((Id−T)n )would be a strictly increasing sequence of closed subspaces of E. One could
so build a followingn as such
1
un∈Ker((Id−T)n )kunk=1, d un Ker(Id−T) n−1 >
2
and so we can show that
1
kTun− Yo u mk> 2
which contradicts the compactness of the operator T.
If you want to present this result in development, make sure to have an application.
convincing (you will find one in the M1 handout from Laurent Desvillettes).

3 Duality
In this part, it will designate a vector space over K=R or C.
In subsections 3.3 and 3.4, Esera will be equipped with a standard.
When it is standardized, we will denote it as [Link] closed unit ball of E.
Definition 8. The set of continuous linear forms on a normed vector space E is
dual topological space of E denoted by E0So we have0=L(E,K)etE0 is a Banach space
for the dual norm
k f k E0=sup|f(x)|. (3)
x is an element of E
kxk61

Exercise 39 [B], [M]

We will assume K=R. An affine hyperplane E is a set of the form

{x ∈ E, f(x) = α}

where is a linear form on E, not identically zero and where α ∈ R.

Show that a hyperplane of E is either closed or dense.

22
3.1 The Hahn-Banach Theorem
Theorem 15. Hahn-Banach Theorem, analytical version
Let there be a semi-norm on E, F a vector subspace of E and in n a linear form
surFvéri ant
For all x in F, the absolute value of f(x) is less than or equal to p(x).
So make a linear extension to E still noting verifying

For all x in E, |f(x)| is less than or equal to p(x).

Remark 14. Within the framework of real vector spaces, the theorem is still true without the
absolute values assuming only that it is a sub-norm, that is to say an application-
cationp:E→R which verifies

For all x in E, for all λ > 0, p(λx) = λp(x) (positive homogeneity)

For all x, y in E, p(x+y) ≤ p(x) + p(y) (sub-additivity).
It should be noted that the analytical version given in [B] is limited to the case of vector spaces.
surR(begins with the case of sub-norms). For the complex case, one can find a
proof in [S].
Corollary 5. Let E be a real normed vector space, and F a vector subspace.
Sif∈F0, so extend it g∈E0aslkdflk=jdkf.
Corollary 6. For every x ∈ E \\ {0}, there exists f ∈ E.0of such a norm

f(x)=kxk.

In other words, for all x ∈ E,

kxk=sup|f(x)|=max|f(x)|. (4)
f ∈ E0 f belongs to E0

kfk61 kfk61

Remark 15.
For all x ∈ E, we denote vxf → f(x) the application of evaluating linear forms with respect to x. We
notice queevx∈(E0 ) 0and whatxk61. The corollary states that we have precisely kevxk=1.
In other words, the application
E−→E 00
x→evx
is an isometric injection. When it is bijective, we will say that E is reflexive.
- It is remarkable that the supremum of the expression (4) is always reached. In contrast, there exist some
cases where the sup of Definition (3) is not reached.
There is not necessarily uniqueness of the element achieving the maximum of equation (4).
Theorem 16. Hahn-Banach Theorem, geometric version, real case
Note: It is assumed here that E is a vector space over R.
Let A and B be two disjoint convex sets of E.

23
 (i) We suppose A is open. Then there exists a closed hyperplane that separates A and B in the broad sense.
that is to say, there exists f∈E0such as

supf(a)6inff(b).
a is in A b is an element of B

(ii) Suppose A is compact and B is closed. Then there exists a closed hyperplane that separates A and B.
strict sense, meaning that there exists f ∈ E0as

supf(a)<inff(b).
a is an element ∈A
b of B

Remark 16. It is recalled that a hyperplane defined by the equation f=α strictly separates two
parts of It exists α<α<α such
1 that
2 one of the parts is included in the half-space
closed{f≤α}and1 the other in the of my-space{f≥α}. 2

Corollary 7. Let E be a normed vector space over R and Fun a vector subspace of E.
for all x ∈ E,
x belongs to F if and only if for all g in E0g=0 implies g(x)=0.
|F
Therefore, Fest is dense if and only if

For all f in |FE 0f=0 ⇒ f=0.

Remark 17. It should be noted that in a large number of practical cases, this density criterion
can be proved without using the Hahn-Banach theorem. As an exercise, it will be shown
in the case where E is a Hilbert space.

Exercise 40*. [GT]

For a >1, we definefa∈C([0,1])parfa(x)=
1 Show that if(an is a sequence of
x−a
real>1 which tends toward infinity, then

V=Vect{fa,n∈N}
n

is dense in C([0,1]).

Exercise 41*. [B, Th. III.23]

We assume that E is a Banach space such that E0 is separable.
Show that E is separable.

3.2 Classic Functional Space Duals

For this subsection, refer to [BP] or [R].
In this subsection, we denote (X, X, µ) a measured space.
It is recalled that it is called a σ-field if there exists an increasing sequence Xn ∈ X such that

X= [ Xn and For all n in N, µ(Xn )<+∞.

n∈N

24
Theorem 17. Dualityp -Lq
Suppose that µ is sigma-finite and that p ∈ [1, ∞[. Let q be the conjugate exponent of p.
Forg∈Lq on notegthe linear application on Lpdefined by

Φg ( f ) = Z fgdµ.
X

So the application
Lq −→(Lp ) 0
g−→Φ g
is a bijective isometry. In other words, every continuous linear form on Lpidentifies itself to a
function of Lq .

Exercise 42 . Particular case of Theorem 17 [M]

Show that sip<∞,p ) 0=`q .

Exercise 43.∞ ) 0,`1[M]

Construct using the Hahn-Banach theorem a continuous linear form on∞ who does not
may be represented by an element of1 .

Exercise 44*.
It is assumed that µ is a sigma-finite measure on X.
Soitg:X→R measurable such that for all f∈Lp , on aitfg∈L1.
We want to show that g ∈ Lq We resume the notation Φgof Theorem 17.
[Link] the case p=∞. In the following, we will assume p<∞.
2. We assume that Φg≡0. Show that g = 0 almost everywhere.
3. Let's note (Xn a increasing sequence of measurable sets of finite measures and union X. For n > 1,
on poseAn=Xn∩ {|g|6n} then

for all f in L p ,Φgn ( fZ) = fgdµ.

An

Show that Φgn∈(Lp ) 0, and that(Φgn converges simply towards Φg Deduce that Φg∈(Lp ) 0.
4. Conclude.

Exercise 45. Let X = {a, b} endowed with the discrete sigma-algebra and let µ be the measure on X assigning the
mass 1 to a and the mass +∞ to b. Does (L1 (X))0 =L∞ (X) in this case?

Remark 18. A very beautiful use of dualityp-qis given by the theorem of in-
Riesz-Thorin interpolation (for the curious, see [SW]).

25
 NB: Until the end of this subsection, we will assume that X is a metric space.
calmly compact (that is to say, such that every point admits a fundamental system of neighborhoods
compact) and separable. The space will be equipped with its Borel tribe.
Recall that C(X) denotes
0 the space of continuous functions on X that tend to 0 at
the infinity and that it is equipped with the uniform norm.

It is possible to define set measures on X with real or complex values.

Reference is made to [R, Chap.6] which details the properties of these real or complex measures. For
Real-valued measures, sometimes referred to as signed measures. In particular, we note
draw that a signed measure on X is written µ=µ+ −µ− where+ and− are positive measures
Sets exported by disjoint measurable sets and unions. It should be noted
that a crucial difference compared to positive measures is that one automatically
µ(X)<∞: this constraint comes from the property of countable additivity (which implies some
series commutatively and therefore absolutely convergent.

Theorem 18. Riesz Representation Theorem [R], [BP]

Let X be a locally compact and separable metric space.
It is a continuous linear form on (C(X), k · k)
0 ∞ ), so there exists a unique measure µ at
real or complex values such that

Z =fdµ.
For all0f in C(X), T(f)
X

In particular, if K is a compact metric space, then the topological dual of C(K) is exact.
the measure space on K (with real or complex values).
Remark 19.
In the previous theorem, we can replace (C(X), k · k∞0)par(Cc(X),k · k∞ because the
two dual spaces identify immediately. In fact, like Cc(X) is dense in C(X), 0
a continuous linear form on Cc(X) extends uniquely into a linear form. expanse
continue on C(X). 0
On the other hand, the result does not extend to the case of space(Cb(X),k · k∞ ).
There are other versions of the Riesz representation theorem concerning forms.
positive linear, that is to say which verify
If f > 0, then T f > 0.

This positivity constraint allows us to restrict ourselves to positive measures (finite or not),
and we branched out from the hypothesis of continuity. Precisely, we also have that a linear form
positiveTsurC(X)(resp.C
0 c(X) can be represented by a positive measure nie (resp.
by a positive measure defined on the compact sets). We refer to [BP], [HL], or [R] for a statement
summary of these results.

26
3.3 Weak Topologies, Weak Convergence
We will provide a brief overview of weak topologies here, and refer to [B] for
a much more detailed report on the subject.
In this subsection, it will designate a Banach space.
For all E0for all x in E, we note that hf, xi = f(x).

Definition 9. Weak topology σ(E, E)0 It is the initial topology defined by the elements.
ments deE0that is to say, it is the least coarse topology on Equi that makes the maps continuous
applications f ∈ E0.

In what follows, convergence in norm in E will be called strong convergence. Otherwise

this,(x n ) converge strongly in Esikxn-xk → 0.

Proposition 10. The weak σ topology (E, E)0 Check the following properties.
She has fewer open sets than the strong topology on E.
(ii) It is the topology defined by the family of seminormsf) f∈Egiven
0 by

p f(x)=|hf,xi|.

(iii) The weak topology σ(E, E)0 is separated.

(iv) The sequence(xn )converge versxpourσ(E,E0 if and only if

For all f in E 0,hf,xn napproaches

i−−−−→hf,xi.
infinity

If (xn strongly converges to x, then (xn )converge versxpourσ(E,E0 ).

(vi) If (xn )converge versxpourσ(E,E0 ), so(x n is narrow-minded and we have

kxk6lim infkxnk.

(vii) If (xn )converge versxpourσ(E,E0 )et si(fn converges strongly towards E0, then

hf n , x n i−−−−→hf,xi.
n approaches infinity

Remark 20.
We will take care to note that strong convergence implies weak convergence, but
On the other hand, a strong open (resp. closed) is not necessarily a weak open (resp. closed).
Note that the separation of the weak topology σ(E, E')0 calls upon Hahn's theorem
Banach.
In a dimension nie, one can show that the weak topology σ(E,E)0 coincides with the topology
strong.
- In infinite dimension, the weak topology σ(E,E)0 is never equal to the strong topology (cf.
Exercise 47).
In an infinite dimension, one can show that the weak topology σ(E,E)0 It is not measurable.
(we note that it is defined by a non-denumerable family of semi-norms).

27
Exercise 46. Show the properties of the topology σ(E, E)0 stated in Proposition 10.

Exercise 47*. Weak and strong topologies in infinite dimension [B]

SoitEun espace vectoriel normé de dimension in nie dont on noteS(resp.B) la sphère unité
(resp. the closed unit ball).
1. Show that the closure of S for the weak topology is included in B.
2. Let x belong to B, n belongs to N∗ and(f,
1 . . . ,fn Continuous linear form n-tuple. Show that it
there exists St such that for all i ∈ {1, ..., n}, f i(x)=fi(y). Deduce what is in the closure.
for weak topology.
3. What has been demonstrated?

Exercise 48*. Characterization of closed convex sets [B]

1. Show that every closed convex set in E is the intersection of the closed half-spaces that contain it.
2. Deduce that a convex set is strongly closed if and only if it is weakly closed.

Exercice 49. [B]

Let E and F be two Banach spaces and T a compact operator from E to F.
Show that if(xn ) weakly converge to x, then T xn strongly converge towards Tx.

Definition 10. We recall that for x ∈ E, we have denoted vxthe shape assessment application
linéaires enxdonnée par
For all f in E 0,evx( f ) = f ( x ) .
Weak-star topology 0, ratedσ(E0,E) is the initial topology defined by the fa-
thousandx)x ∈E that is to say, it is the least fine topology that makes the maps continuous
evaluation criteria.

One can see that the weak-star topology σ(E0E)surE0fewer open sets than the topology
weak σ(E0,E00 (which itself has fewer openings than the strong topology on E0).

Proposition 11. The weak-* topology on E0 verify the following properties.

It is the topology defined by the family of semi-norms. x|) x ∈E .
The weak-star topology is separated.
(iii) The sequence(fn )converge towards σ(E0if and only if

For all x in E, hfn , x in→∞

−−−−→hf,xi.

If (fn strongly converge to f, then (fn )converge for σ(E0,E).

(v) If (fn )converge to σ(E0,E), then(fn is narrow-minded and we have

k f k 6 l i m i n f k f nk.

28
 (vi) If (fn )converge toward σ(E0E)et if(xn converges strongly towards E, then

hf n , x n i−−−−→hf,xi.
n approaches infinity

Exercise 50. Show the properties of the topology σ(E)0statements in Proposition 11.
Remark 21. In the spaces Lpfor 1 < p < ∞, the weak and weak-star topologies coincide
cident.

Exercise 51. Questions of weak and weak-star convergence.

Let E be a Banach space.
a. Let D be a dense subset of E and (fn (a suite in)0Show that (fn converge towards
for the weak-star topology if and only if(fn ) it is closed and

For all x in D, hfn , x i − → h f , x i .

b. Provide a similar property for weak convergence in E.
2. Let H be a separable Hilbert space and (ep )p∈Na Hilbert basis of H.
[Link] whatn→ weakly in Hsi and only if (xn ) is narrow-minded and

For all p in N, hxn , e p i−−−−→hx,e p i.

n approaches infinity

b. What is the lower limit of the sequence(ep )?

Does the health condition persist if one removes (xn narrow-minded?
3. So φ ∈ Cc∞ (R).
[Link] that1[n,n+1]converges weakly to 0 in Lpfor 1 < p < ∞.
Is it true for p=1? Does it converge weakly-* in L?∞ ?
b. Show that x → nφ(n·) √ converges weakly in L2 (R).
c. Show that in L (0,1),x→einx converge weakly towards 0.
1

Exercise 52*. Separability and Weak-* Topology [B]

Let B be a Banach space.E the closed unit 0 ball of E, which inherits the
0 topology
0
induced by the weak-star topology on E. We will see that BEis measurable 0 for the topology
weak-star if and only if E is separable.
1. Suppose that E is separable. This implies that BEis also separable, and therefore we can
to continue a sequence(xn )dense inBE . For f, g ∈ BE, let's define
0

1
d(f,g)= X n |hf−g,x n i|.
n>1 2

[Link] that an open ball Bd(f,r) contains 0 a neighborhood defined for the
0 topology
weak star.
0
0 as ∈E for the weak-star topology.
b. Thus, let the neighborhood be defined

29
 Show that there exists ε > 0 and y, ...,1yk∈BEsuch as contains

V 0f is in BEFor0all i=1, ..., k, |hf−f,yi i|<ε. 0

Deduce that V contains an open ball Bd( f , r ) . 0

[Link].
[Link] there exists a distance on BEwhich induces 0 the weak-* topology.
a. Show that for all n > 1, there exists ε.n>0 and a part nieDn As such as the ball
1
openBd(0,ncontains the whole

f ∈ BEFor
0 all x in Dn ,|hf,xi|<ε n .

b. By using the density criterion stemming from the Hahn-Banach theorem, show that
D=∪Dnis dense in dance.
[Link].

Note 22. It is also true that the measurability of BEfor the weakσ topology (E, E)0 )est
equivalent to the separability of E0. IfE0is separable, by adapting the above exercise, one can
build a distance on BEwhich induces the weak topology. On the other hand, the reverse is more
difficult (we refer to the references given in [B]).

Exercise 53*.1the Schur property [GT]

In general, we have seen that weak convergence does not imply strong convergence. The object
The exercise is to show that1is a very particular case in which the two notions of
coincident convergence. Let (αn a suite in1 which converges to 0 for the topology
weak1 , ` ∞ ). We write αn =(αpn )p∈N .
1. Show that
For all p in N, αpn−n→∞
−−−→0.

2. The closed unit ball is furnished.∞ from the topology of simple convergence.
a. Recall why B equipped with this topology is a compact metric space.
[Link] let ε > 0. For n ∈ N, we set

FnFor all elements x in B, fork all k greater than n, the absolute value of the inner product of hx and αi is less than or equal to ε.

Show that there exists an integer such that Fnis empty interior.
[Link].

Remark 23. Thus, in1 , the weak and strong topologies are distinct, but curiously,
they have the same convergent sequences. This counterexample shows that two topologies having
The same convergent sequences are not necessarily equal.

30
 To conclude this paragraph, let's mention a very useful case of weak-* convergence in
probabilities. We have seen that the topological dual of C(Rd0)is the space M(Rd)signed measures
surRd . Thus, a sequence (µn M(dans)dconverges to the weak-* topology if and
only if
f n−−−−→n→∞ Z dµ.
For all0f in C(RZd),dµ f

In a more elementary way, a probability measure µ on Rdinduces linear forms

continues onCc (Rd),C(R0d), etc b(Rd) (equipped with the uniform standard). Therefore, we
there are three notions of weak-* convergence for a sequence of probability measures
surRd The Theorem 19 below shows that, for probability measures, the three notions
of convergence are equivalent. But before stating it, let's recall the

Definition 11. Let (µn ) measures on Rdto real or complex values.

It is said thatnnarrowly converge towards µsi

Z d),dµ
For all f in C b(R Z dµ.
f
f n−−−−→n approaches infinity

NotonsPXthe law of the random variable X on Rd(PXis therefore a measure of probability

surRd ). By definition, a sequence (Xn random variables taking values in Rdconverge in law
versXsi if and only if the sequencePXnarrowly
n
converge towards PXThat is why this notion
Narrow convergence plays a major role in probability.

Theorem 19. Let (µn probability measures on Rd .

Let Hun be a subspace of (Cb(Rd),k · k∞ whose adherence contains Cc(Rd).
Son converge closely towards µsi if and only if

For all f inZ H,dµ Z dµ.

f
f n −−−−→n approaches infinity

Pour la preuve de ce résultat et une discussion détaillée sur les liens entre les trois topologies
weak-star mentioned above, we refer to [Ouv, Chap.14]. A proof can also be found
more elementary aspect of this theorem in the integration and probability handout by J.F. Le Gall
(this proof can constitute a development).
Let us mention another practical characterization of the tight convergence of measures.
Probability on Over terms of distribution functions.

Theorem 20. Let (µn ) measures of probability on R.

The following(µn ) converge closely towards µssi for all xtel that µ({x})=0,

µn (]− ∞,x])−−−−→µ(]− ∞,x]).

n approaches infinity

31
Exercise 54.
Let (µn probability measures on Rd .
It is assumed that there exists a compact K ⊂ [Link]
Supp(µn )⊂Kfor alln.
Show thatnconverge closely towards µsi if and only if

f n−−−−→n→∞ Z dµ.
Z dµ
For all f in C(K), f

. ologtoiepdeTDcf. y s o h n, T i e d t z e e - U [Link] Indication

Exercise 55. Weyl's criterion for equidistribution

It is said that a sequence (xk)∈[0,1] is equidistributed
N∗ if for all 0 ≤ a < b ≤ 1,

1
Cardk ( ∈ {1, ..., n}xk[a, )b] −−
−−→b−a.
n n→∞

For all n>1, we introduce the probability measure

n
1
µn= δxk.
nX
k=1

1. Show that (xk It is evenly distributed if and only if µnconverge closely towards the measurement
Lebesgue on [0,1].
[Link] that(xkIt is equidistributed if and only if
n
1
for all p, 0, e 2ipπxk−
−−−→0.
nX
k=1
n approaches infinity

3. Let x, α ∈ R.
[Link] that(xk(x+αk) is uniformly distributed modulo 1 if and only if α < Q.
b. Deduce that if α ∈ R\Q, then αN + Z is dense in R.

32
3.4 Low Compactness
Theorem 21. Banach-Alaoglu Theorem [B]
The closed unit ball of E0is compact for the weak-* topology.
The proof of this result relies on Tychonoff's theorem on (non-countable) products.
d’espaces compacts et donc dépasse le programme de l’agrégation. On notera de plus que, tant
that we do not know that the unit ball of E0is measurable for the weak-* topology, we
does not have a sequential characterization of compactness and therefore we cannot use this theorem
remedy to extract subsequences, which is very bothersome in practice. For these two reasons, we
give now a simplified (and much more accessible) version of the result in the case where E
is separable.
Theorem 22. Banach-Alaoglu Theorem [HL]
We assume that E is a separable Banach space. Let (fn a suite in BESo it exists 0

a subsequence(fn k)etf∈E0 such as

For all x in E, hfk n, x i −k→∞

−−−→hf,xi.

Exercise [Link] of Theorem 22 [HL]

In xe a sequence (xp dense dance.
1. Using the diagonal extraction process, show that there exists a subsequence (fn)telle k
that for all p in N, (fn(xp )) k∈N
k
converge.
2. Let x ∈ E. Show that (fn(x) is Cauchy.
k
3. Conclude.

Exercise 57*. IDEVJ Existence of Invariant Measures [ZQ]

Let K be a compact metric space and T: K → K be continuous. Let E = C(K).
The objective of the exercise is to demonstrate the existence of a probability measure on K which is
T-invariant, that is to say such that

Z E,f◦T dµ=
For all f in Z fdµ.
K K

1. Show that if a sequence (µn )Probability measures on K converge closely to a

signed measure µ∈E0, then is a measure of probability.
2. Let's fix x ∈ K. For all n > 1, we consider
n
1
µn= δT k .
nX
k=0
(x )

Show that
For all f in E, h mu n f◦Ti − hµnn,fi−−−−→0.
approaches infinity
3. Deduce that there exists a T-invariant probability measure.

33
References
[AM] E. Amar et E. [Link] Complexe. Cassini, 2004.
[B] H. Bré[Link] Fonctionnelle. Dunod, 1999.
[BP] M. Briane and G. Pagès. Theory of Integration. Vuibert, 2007.
[CLF2] A. Chambert-Loir et S. [Link] de Mathématiques pour l’Agrégation : Analyse 2. Masson, 1995.
[CLF3] A. Chambert-Loir and S. Fermigier. Mathematics Exercises for the Aggregation: Analysis 3. Masson, 1996.
[CM] M. Crouzeix and A.L. Mignot. Exercises in Numerical Analysis of Differential Equations. Masson, 1982.
Far J. Faraut Integral Calculus. EDP Sciences, 2006.
[G-An] X. Gourdon. Mathematics in Mind, Analysis. Ellipses, 1994.
[GT] S. Gonnord et N. [Link] et Analyse Fonctionnelle. Ellipses, 1996.
[HL] F. Hirsch et G. [Link] d’Analyse Fonctionnelle. Dunod, 1999.
J.P. Kahane and P.G. Lemarié-Rieusset. Fourier Series and Wavelets, Cassini, 1998.
[LM] G. Lacombe and P. Massat. Functional Analysis, Corrected Exercises. Dunod, 1999.
[M] B. [Link] Fonctionnelle. Ellipses, 2004.
[Ouv] J.Y. [Link]és 2. Cassini, 2000.
[R] W. Rudin. Real and Complex Analysis. Masson, 1975.
[S] L. [Link] Générale et Analyse Fonctionnelle. Hermann, 2008.
[SW] E.M. Stein and G. Weiss. Introduction to Fourier Analysis on Euclidean Spaces. 1990.
[Wa] C. [Link] et Analyse Fonctionnelle. Hermann, 2012.
[Wi] M. [Link] Fonctionnelle Élémentaire. Cassini, 2003.
ZQ C. Zuily et H. Que é[Link] pour l’agrégation. Dunod, 2007.

34