PEOPLE’S DEMOCRATIC REPUBLIC OF

ALGERIA
MINISTRY OF HIGHER EDUCATION AND
SCIENTIFIC RESERACH

Mohamed Boudiaf University of M’sila

Faculty of Mathematics and Informatics
Departement of Mathematics

Domain : Mathematics and Informatics

Specialty: Mathematics
Option : Partial Differential Equations and Applications

Lessons and exercises on

Variational methods for nonlinear elliptic problems

By :
Abdelhak MOKHTARI

08/09/2023.

University years 2023/2024

Contents

1 Operators on Banach spaces 3

1.1 Some notions in functional analysis . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1.1 Dual space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1.2 Some notions of convergence . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.1.3 Reflexive space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.1.4 Continuity, coercivity and convexity of operators . . . . . . . . . . . . . . 6
1.1.5 Lower and Upper Semi-contiwuity . . . . . . . . . . . . . . . . . . . . . . 7
1.2 Functional spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.2.1 Lp Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.2.2 Some Results about Integration . . . . . . . . . . . . . . . . . . . . . . . 9
1.2.3 W 1.p (Ω) Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.2.4 Sobolev Injections and Inequalities . . . . . . . . . . . . . . . . . . . . . 10
1.2.5 W01.p (Ω) Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.3 Differentiability of Functionals and Critical Points . . . . . . . . . . . . . . . . . 11
1.3.1 Gâteaux differentiability . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.3.2 Fréchet differentiability . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.3.3 Critical points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

2
Chapter 1
Operators on Banach spaces

1.1 Some notions in functional analysis

1.1.1 Dual space
Let (E, k · kE ), (F, k · kF ) be two Banach spaces.

Definition 1.1 Let A : E → F be a linear operator. One says that A is bounded (or continu-
ous) if there is a constant c ≥ 0 such that:

kAukF ≤ ckukE , ∀u ∈ E.

Notation 1.1 • We denote by L(E, F ) the space of continuous (bounded) linear operators
from E into F equipped with the norm

kT kL(E,F ) = sup kT (x)kF

x∈E,kxkE ≤1

• If E = F , we write L(E) instead of L(E, E).

• Let E be a vector space over R. We recall that a functional is a function defined on E,

or on some subspace of E, with values in R. We denote by E 0 the dual space of E, that
is, the space of all continuous linear functionals on E; the dual norm on E (the norm in
E 0 ) is defined by
kf kE 0 = sup |f (x)| = sup f (x).
x∈E,kxkE ≤1 x∈E,kxkE ≤1

• When there is no confusion we shall also write kf k instead of kf kE 0 .

• Given f ∈ E 0 and x ∈ E we shall often write hf, xi instead of f (x); we say that h·, ·i is
the scalar product for the duality E 0 , E. So, we have the following inequality

∀f ∈ E 0 , ∀x ∈ E : |hf, xi| ≤ kf kE 0 kxkE . (1.1)

• It is well known that E 0 is a Banach space, i.e., E 0 is complete (even if E is not); this
follows from the fact that R is complete.

3
Variational methods for nonlinear elliptic problems
• If E is a Hilbert space, then, according to Riesz’s theorem (see [1]), the spaces E and its
dual E 0 can be identified, we write
E 0 = E. (1.2)

The equality (1.2) does not mean equal elements, but rather in the sense that there is a
canonical isometry from E onto E 0 .
 0  0
Example 1.1 Let 1 < p < ∞, we have Lp (Ω) = Lq (Ω) where p1 + 1q = 1. L1 (Ω) = L∞ (Ω).

1.1.2 Some notions of convergence

Definition 1.2 (Strong convergence) Convergence of a sequence (xn )n∈N in a normed vec-
tor space (E, k · kE ) to an element x ∈ E, defined in the following way:

∀ε > 0, ∃n0 ∈ N, ∀n ∈ N : n ≥ n0 ⇒ kxn − xkE ≤ ε.

Or,
lim kxn − xkE = 0.
n→+∞

We denote by xn → +x as n → +∞.

Definition 1.3 We say that the sequence (xn )n∈N ⊂ E converges weakly to x ∈ E if and only
if
hf, xn i → hf, xi, ∀f ∈ E 0 . (1.3)

for n → +∞. This weak convergence is written as xn * x,

Remark 1.1 • Note that the convergence in (1.3) is convergence for a sequence of real
numbers.

• If xn * x, then (xn )n∈N is bounded in E; i.e.

∃M > 0 : kxn k ≤ M.

Example 1.2 1. Let {en }n∈N be an Hilbertian base for a Hilbert space (H, h·, ·i). Then, for
each x ∈ H can be decomposed as
∞
X
x= hen , xien
n=1

Or
m
X
kx − hem , xiem k → 0,
n=1

as m → +∞. From this it follows that hen , xi → 0 = h0, xi for any x ∈ H 0 = H, this
means that en * 0 in H.
On the other hand, we have ken − 0k = ken k = 1, so {en } does not converge. We deduce
that {en }n is weakly convergent but it is not strongly convergent.

4 Abdelhak MOKHTARI
 Variational methods for nonlinear elliptic problems
2. Let Ω ⊂ RN an open set, let 1 < p < ∞. A sequence un * u in Lp (Ω) means that
Z Z
un v dx → uv dx, ∀ ∈ v ∈ Lq (Ω),
Ω Ω
1 1
where p
+ q
= 1.

3. Let {fn }n∈N be the sequence of functions defined by


 √1 , if x ∈ [n, 2n];
n
fn (x) =
 0, if not.

We can verify that {fn }n∈N converges weakly to 0 in L2 ([0; +∞[) but does not converge
strongly in L2 ([0; +∞[).

1.1.3 Reflexive space

Definition 1.4 Let (E, k · k) be a Banach space and let E 0 be the dual space. The bidual E 00 is
the dual of E 0 with norm

kf kE 00 = sup |hf, gi|, ∀f ∈ E 00 .

g∈E 0 ,kgkE 0 ≤1

There is a canonical injection i : E → E 00 defined as follows:

i : E → E 00
x 7→ ix,

where ix is defined as:

ix : E 0 → R
f 7→ hix, f iE 00 ,E 0 = hf, xiE 0 ,E ,

It is clear that i is linear and that i is an isometry, that is,

kixkE 00 = kxkE . (exercice)

Definition 1.5 (Reflexive space) The space E is said to be reflexive if the canonical injec-
tion i : E → E 00 is surjective, i.e. i(E) = E 00 .

When E is reflexive, E 00 is usually identified with E; (E 00 = E).

Example 1.3 1. finite dimensional spaces are reflexive (since dim E = dim E 0 = dim E 00 ).

2. Lp spaces are reflexive for 1 < p < ∞. L1 and L∞ are not reflexive.

3. Hilbert spaces are reflexive.

The following result shows an important compactness property in reflexive spaces (see [1])

Theorem 1.2 Assume that E is a reflexive Banach space and let(xn ) be a bounded sequence
in E. Then there exists a subsequence (xnk ) that converges weakly to some x in E.

Abdelhak MOKHTARI 5
Variational methods for nonlinear elliptic problems
1.1.4 Continuity, coercivity and convexity of operators
In the following, we state some definitions concerning continuity properties.

Definition 1.6 The operator T : E → F is called continuous at the point x0 ∈ E if for any
sequence (xn ) ⊂ E

∀(xn ) ⊂ E : xn → x0 in E ⇒ T (xn ) → T (x0 ) in F

 
kxn − x0 kE → 0 ⇒ kT (xn ) − T (x0 )kF → 0

This definition is equivalent to:

∀ε > 0, ∃δ > 0, ∀x ∈ E : kx − x0 kE ≤ δ ⇒ kT (x) − T (x0 )kF ≤ ε.

Example 1.4 Let (E, k · k) be a Banach space. We consider N : E → R; N (x) = kxk. N is a

continuous functional. Indeed, for any (xn )n ⊂ E such that xn → x in E, i.e. (kxn − xk → 0).
Since
N (xn ) − N (x) = kxn k − kxk ≤ kxn − xk,

then we have N (xn ) → N (x) in R.

There are some other definitions that are related to the concept of weak convergence.

Definition 1.7 Let T : E → F be an operator. T is called weakly continuous at x0 ∈ E if for

every sequence (xn ) ⊂ E, we have

xn * x0 in E ⇒ T (xn ) * T (x0 ).

Remark 1.2 For functionals defined on E, the concepts of convergence and weak convergence
coincide. So the functional T :→ R is called weakly continuous if for every sequence (xn ) ⊂ E,
we have
xn * x0 in E ⇒ T (xn ) → T (x0 ) inR.

Note that the concept of weak continuity is stronger than the concept of continuity because the
convergence implies the weak convergence.

Definition 1.8 (Compact operator) An operator T : E → F is called compact if it is

continuous and T (B) is relatively compact set in F (i.e. T (B) is compact set in F ); here
B = {x ∈ E : kxkE < 1}.

There is an equivalent definition using sequences.

Definition 1.9 An operator T : E → F is called compact if it is continuous


and if every
bounded sequence (xn )n∈N ⊂ E has a subsequence (xnk ) for which T (xnk ) converges.

Remark 1.3 If T is a linear operator then the requirement that T is continuous on E, which
then equivalent that T is bounded operator, is superfluous

6 Abdelhak MOKHTARI
 Variational methods for nonlinear elliptic problems
The compact operators defined on reflexive spaces have an important property that appears in
the following theorem.

Theorem 1.3 Let E be a reflexive Banach space and F a Banach space. If T : E → F is

linear and compact. Then we have

xn * x0 in E ⇒ T (xn ) → T (x0 ) in F

Example 1.5 The application norm E 3 x 7→ kxk is a compact operator.

In the search for the global minimum, which we will discuss in Chapter 2, a related concept is
convexity and coercivity.

Definition 1.10 1. We say that a part K of E is convex if On dit qu’une partie K de X

est convexe si:
∀x, y ∈ K, ∀θ ∈ [0, 1], θx + (1 − θ)y ∈ K.

2. When K is convex and J : K −→ R is a functional. We say that J is convex if:

∀x, y ∈ K, ∀θ ∈ [0, 1], J(θx + (1 − θ)y) ≤ θJ(x) + (1 − θ)J(y).

We say that J is strictly convex if:

∀x, y ∈ K, avec x 6= y ∀θ ∈]0, 1[, J(θx + (1 − θ)y) < θJ(x) + (1 − θ)J(y).

3. We say that J is (strictly) concave if −J is (strictly) convex.

Definition 1.11 A functional J : E → R on a Banach space E is called coercive if, for every
sequence (un )n∈N ⊂ E,
lim J(un ) = +∞.
kun k→+∞

1.1.5 Lower and Upper Semi-contiwuity

The following concepts are often found in minimization theory of functionals.

Definition 1.12 1. A functional J : ER is called lower semi-continuous at a point x0 ∈ E

if for every sequence (xn ) ⊂ E, we have:

xn → x0 ⇒ f (x0 ) ≤ lim inf f (xn ).

n→+∞

2. J : ER is called weakly lower semi-continuous (w.l.s.c) at a point x0 ∈ E if for every

sequence (xn ) ⊂ E, we have:

xn * x0 ⇒ f (x0 ) ≤ lim inf f (xn ).

n→+∞

3. J : ER is called upper semi-continuous at a point x0 ∈ E if for every sequence (xn ) ⊂ E,

we have:
xn → x0 ⇒ f (x0 ) ≥ lim sup f (xn ).
n→+∞

Abdelhak MOKHTARI 7
Variational methods for nonlinear elliptic problems
4. J : ER is called weakly upper semi-continuous (w.u.s.c) at a point x0 ∈ E if for every
sequence (xn ) ⊂ E, we have:

xn * x0 ⇒ f (x0 ) ≥ lim sup f (xn ).

n→+∞

Exampleq
1.6 Let (H, h·, ·i be a Hilbert space. The functional N : E 3 x 7→ N (x) = kxk, where
k · k = h·, ·i. We have N is w.l.s.c. Indeed; let (xn ) ⊂ H and x0 ∈ H such that xn * x0 , then

hxn , yi → hx, yi, ∀y ∈ H 0 = H,

in particular, taking y = x0 ∈ H, we get

hxn , x0 i → hx0 , x0 i = kx0 k2 . (1.4)

By a direct calculation, we get

kxn − x0 k2 = hxn − x0 , xn − x0 i
= kxn k2 − 2hxn , x0 i + kx0 k2 .

Since kxn − x0 k2 ≥ 0, we deduce that

kxn k2 ≥ 2hxn , x0 i − kx0 k2 . (1.5)

Passing to the limit in (1.5), and using (1.4), we get

lim inf kxn k2 ≥ 2kx0 k2 − kx0 k2 = kx0 k2 .

n→+∞

which implies that N is w.l.s.c.

1.2 Functional spaces

In this chapter, we briefly mention some spaces and their most important properties and theo-
ries, which we will need in the following lessons. For more details see e.g [2].

1.2.1 Lp Spaces
Definition 1.13 Let p ∈ R with 1 ≤ p < ∞, we set
 Z 
Lp (Ω) = f : Ω → RN ; f is measurable and |f (x)|p dx < ∞ .
Ω

with Z 1
p
kf kLp = kf kp = |f (x)|p dx .
Ω

Definition 1.14 We set

n o
L∞ (Ω) = f : Ω → RN ; f is measurable and |f (x)| ≤ c a.e.on Ω ,

with
kf kL∞ = kf k∞ = inf {c : |f (x)| ≤ c a.e.on Ω} ,
with c is a constant.
If f ∈ L∞ (Ω) then we have |f (x)| ≤ kf k∞ a.e. in Ω.

8 Abdelhak MOKHTARI
 Variational methods for nonlinear elliptic problems
Theorem 1.4 Lp (Ω) is a Banach space for any 1 ≤ p ≤ ∞, and reflexive for 1 ≤ p < ∞, and
separable for 1 ≤ p < ∞.

Notation 1.5 Let 1 ≤ p ≤ ∞, we denote by p0 the conjugate exponent

1 1
+ 0 = 1.
p p

Theorem 1.6 (Hölder’s inequality) Assume that f ∈ Lp and g ∈ Lp with 1 ≤ p ≤ ∞, then

f g ∈ L1 and Z
|f g| ≤ kf kLp kgkLp0 .

1.2.2 Some Results about Integration

Theorem 1.7 (Lebesgue’s Dominated Convergence Theorem) Let (fn ) be a sequence of func-
tions of L1 . We suppose that

1. fn (x) −→ f (x) a.e.on Ω,

2. there exists a function g ∈ L1 such that for every n, |fn (x)| ≤ g(x) a.e.in Ω.

We have f ∈ L1 (Ω) and kfn − f kL1 −→ 0.

Theorem 1.8 (Lebesgue’s Dominated Convergence Inverse Theorem) Let (fn ) be a sequence
in Lp and let f ∈ Lp be such that kfn − f kLp −→ 0.
Then, there exist a subsequence (fnk ) and a function h ∈ Lp such that

1. fnk (x) −→ f (x) a.e.on Ω,

2. |fnk (x)| ≤ h(x) ∀k, a.e.in Ω.

Lemma 1.1 (Fatou Lemma) Let (fn ) be a sequence of functions in L1 that satisfy

1. for all n ∈ N, fn ≥ 0 a.e.

2. supn fn < ∞.
R

For almost x ∈ Ω we set f (x) = lim inf n→+∞ fn (x). Then f ∈ L1 and
Z Z
f ≤ lim inf fn .
n→+∞

1.2.3 W 1.p (Ω) Spaces

Let Ω ∈ RN be an open set, and let p ∈ R with 1 ≤ p ≤ ∞.

Definition 1.15 We denote by D(Ω) the set of function of class C ∞ (Ω) with support compact
include in Ω. The Sobolev space W 1.p (Ω) is defined by
( )
1.p p p
Z
∂ϕ Z
W (Ω) = u ∈ L (Ω), ∃gi ∈ L (Ω)such that: u dx = − gi ϕdx, ∀ϕ ∈ D(Ω), ∀i = 1, 2, ..., N .
Ω ∂xi Ω

Abdelhak MOKHTARI 9
Variational methods for nonlinear elliptic problems
We set
H 1 (Ω) = W 1.2 (Ω).
∂u
For u ∈ W 1.p (Ω) we set ∂xi
= gi , and we write
∂u ∂u ∂u
∇u = grad u = ( , , ..., ).
∂x1 ∂x2 ∂xN
The space W 1.p is equipped with the norm kukW 1.P = kukLP + k∇ukLP , or sometimes with the
1
equivalent norm kukW 1.P = (kukpLp + k∇ukpLp ) p if (1 ≤ p < ∞).
The space H 1 (Ω) is equipped with the scalar product
Z Z
uvdx + ∇u∇vdx.
Ω Ω
1
The associated norm kukH 1 = (kuk2L2 + k∇uk2L2 ) 2 is equivalent to the W 1.2 norm.

Proposition 1.1 [?].W 1.p (Ω) is a Banach space for 1 ≤ p ≤ ∞, reflexive for 1 < p < ∞, and
separable for 1 ≤ p < ∞.
In particular H 1 (Ω) is reflexive, separable and Hilbert space.

1.2.4 Sobolev Injections and Inequalities

Let (E, k.kE ), (F, k.kF ) Banach spaces
Notation 1.9

1. E is injected continuously into F , means that the canonical injection j : E → F is

continuous i.e, ∃c > 0, ∀x ∈ E : kxkF ≤ c kxkE , and we denote by E ,→ F.

2. E is injected in compact into F means that the canonical injection j : E → F is compact

i.e for all sequence bounded un in E we can extract subsequence unk convergent in F ,
and we denote by E ,→c F.
If 1 ≤ p < ∞, the Sobolev exponent of p defined by p∗ = NN−pp
or p1∗ = p1 − N1 .

Theorem 1.10 Let 1 ≤ p ≤ ∞, we suppose that Ω is on open set of class C 1 a bounded fron-
tier, and we take Ω = RN
+

1. W 1.p (Ω) ,→ Lq (Ω) ∀q ∈ [1, p∗ [ if p < N.

2. W 1.p (Ω) ,→ Lq (Ω) ∀q ∈ [p, ∞[ if p = N.

3. W 1.p (Ω) ,→ L∞ (Ω) if p > N.

Theorem 1.11 (Rellich-Kondrachon) Let Ω ⊂ RN be a bounded domain of class c1

1. W 1.p (Ω) ,→c Lq (Ω) ∀q ∈ [1.p∗ [ if p < N.

2. W 1.p (Ω) ,→c Lq (Ω) ∀q ∈ [p.∞[ if p = N.

 
3. W 1.p (Ω) ⊂ C Ω if p > N.

10 Abdelhak MOKHTARI
 Variational methods for nonlinear elliptic problems

1.2.5 W01.p (Ω) Space

Definition 1.16 .Let 1 ≤ p ≤ ∞, W01.p means the closing of D(Ω) in W 1.p , we notice

W 1.p
W01.p (Ω) = D(Ω)
n o
= u ∈ W 1.P (Ω) : u = 0 sur ∂Ω ,

and
H01 (Ω) = W01.2 (Ω).

The space W01.p (Ω) provided with norm induced by W 1.p , H01 is a Hilbert space for the scalar
product of H 1 we put kukH 1 = kuk.
0

Remark 1.4 When Ω = RN , we know that D(RN ) is dense in W 1.p (RN ), and there for

W01.p (RN ) = W 1.p (RN ).

Proposition 1.2 (Poincaré’s inequality) Let Ω ⊂ RN on open set, Then there exists a
constant c > 0 such that

kukW 1.p (Ω) ≤ c k∇ukLp (Ω) ∀u ∈ W01.p (Ω).

In other words, on W01.p , the quantity k∇ukLp (Ω) is a norm equivalent to the W 1.p norm.

1 1
Theorem 1.12 (Young’s inequality) . For a, b ≥ 0 and p, q ≥ 1 such that p
+ q
= 1 we
have
1 1
ab ≤ ap + bq .
p q

1.3 Differentiability of Functionals and Critical Points

In what follows, we present the two principal definitions of differentiability and their main
properties, we start with the directional derivative.

1.3.1 Gâteaux differentiability

Definition 1.17 Let E be a Banach space, Ω ⊂ E an open set, and let I : Ω −→ R be a
functional, we say that I is Gâteau differentiable (G-differentiable) at u ∈ Ω, if there exists
A ∈ E 0 (A linear and continuous), denoted by IG0 (u) such that, for all v ∈ E,

I(u + tv) − I(u)

lim = hA, vi , (1.6)
t→0 t
If I is Gâteaux differentiable at u, there exists only one linear functional A ∈ E 0 satisfying
(1.6). It is called the Gâteaux differential of I at u and is denoted by IG0 (u).

Abdelhak MOKHTARI 11
Variational methods for nonlinear elliptic problems
Definition 1.18 If the functional I is Gâteaux differentiable at every u of an open set U ⊂ E,
we say that I is Gâteaux differentiable on U . The map

IG0 : E → E 0
u 7→ IG0 (u)

is called the Gâteaux derivative of I.

Example 1.7 Let 1 < p < ∞, the functional

I : Lp (Ω) → R
Z
u 7→ I(u) = |u|p dx
Ω

is Gâteaux differentiable and we have

Z
hIG0 (u), vi =p |u|p−2 uvdx.
Ω

Indeed; let v ∈ Lp (Ω), we have

I(u + tv) − I(u) |u + tv|p dx − |u|p dx

R R
Ω Ω
lim = lim .
t→0 t t→0 t
We define

gu,v : [0, t] → R
s = gu.v (s) = |u + sv|p .

since gu,v is continuous on [0, t] and differentiable on ]0, t[, then according to the mean value
theorem, there exists a real ct ∈]0, t[ such that

0
gu,v (t) − gu,v (0) = gu,v (ct )t

or,
|u + tv|p − |u|p = p |u + ct v|p−2 (u + ct v)vt,

when t → 0, we have ct → 0, so

|u + tv|p − |u|p
lim = lim p|u + ct v|p−2 (u + ct v)v
t→0 t ct −→0

= p|u|p−2 uv. (1.7)

Let us recall by the following preliminary inequality,

(a + b)p ≤ 2p−1 (ap + bp ), ∀p > 1, ; ∀a, b > 0. (1.8)

Using (1.8), we get

|u + ct v|p−2 (u + ct v)v ≤ |u + ct v|p−1 |v| ≤ 2p−1 ≤ C(|u|p−1 |v| + |v|p ).

12 Abdelhak MOKHTARI
 Variational methods for nonlinear elliptic problems
We verify that |u|p−1 |v| + |v|p ) ∈ L1 (Ω); that is, by Hölder inequality we get
Z
|u|p−1 |v| dx ≤ kukp−1
Lp kvkLp < ∞
Ω

then |u|p−1 |v| + |v|p ) ∈ L1 (Ω). Using this and (1.7), applying Dominated Convergence Theorem,
we get
p p
Ω |u + tv| dx − Ω |u| dx
R R Z
lim = p |u|p−2 uvdx.
t→0 t Ω
Now, we define

A : Lp → R
Z
v 7→ A(v) = p |u|p−2 uvdx.
Ω
p 0
we prove that A ∈ (L (Ω)) , i.e. A is linear and continuous.
A is linear, that is, let v1 , v2 ∈ Lp (Ω) and let α, β ∈ R
Z
A(αv1 + βv2 ) = p |u|p−2 u(αv1 + βv2 )dx
ZΩ Z
p−2
= p[ |u| uαv1 dx + |u|p−2 uβv2 dx]
Ω
Z ΩZ

= αp |u|p−2 uv1 dx + βp |u|p−2 uv2 dx

Ω Ω
= αA(v1 ) + βA(v2 ).

then A is linear. For continuity of A, let u, v ∈ Lp (Ω)

Z Z
p−2
|A(v)| = |p |u| uvdx| ≤ p |u|p−1 |v|dx
Ω ZΩ Z
0 1 1
≤ p( |u|(p−1)p dx) p0 ( |v|p dx) p
ZΩ p
Ω
p−1
Z
1
(p−1)( p−1 )
≤ p( |u| dx) p ( |v|p dx) p
Ω Ω
= kukp−1
Lp kvkLp .

Then A is a continuous, so the functional I is G-differentiable and hIG0 (u), vi = p |u|p−2 uvdx.
R
Ω

Example 1.8 In the previous example 1.7, if we set p = 2, the functional

I : L2 (Ω) → R
Z
u 7→ I(u) = |u|2 dx = kuk2L2
Ω

is Gâteaux differentiable and we have

Z
hIG0 (u), vi =2 uv dx.
Ω

Example 1.9 The functional defined by

J : W01,p (Ω) → R
Z
u 7→ J(u) = |∇u|p dx = kuk2W 1,p
Ω 0

is Gâteaux differentiable and we have

Z
hJG0 (u), vi =p |∇u|p−2 ∇u∇v dx.
Ω

Abdelhak MOKHTARI 13
Variational methods for nonlinear elliptic problems
1.3.2 Fréchet differentiability
Definition 1.19 Let E be a Banach space, Ω ⊂ E an open set and let I : Ω −→ R be a
functional, we say that I is Fréchet differentiable at u ∈ Ω, if there exists Au ∈ E 0 such that
I(u + v) − I(v) − Au v
lim =0 (1.9)
kvk→0 kvk
Thus, for a Fréchet differentiable functional I, we have

I(u + v) − I(u) = Au (v) + o(kvk)

as kvk → 0.

If a functional I is differentiable at u, then Au is a unique.

Definition 1.20 Let I : Ω → R be differentiable at u ∈ Ω. The unique element Au such that

(1.9) holds is called the Fréchet differential of I at u, and is denoted by I 0 (u). Then we can
write
I(u + v) − I(u) = hI 0 (u), vi + o(kvk)
as kvk → 0.

Definition 1.21 Let Ω ⊂ E be an open set.

1. If the functional I is differentiable at every u ∈ Ω, we say that I is differentiable on Ω.

2. The map I 0 : Ω → E 0 that sends u ∈ Ω to I 0 (u) ∈ E 0 is called the Fréchet derivative of I.

Note that I 0 is in general a nonlinear map.

3. If the derivative I 0 is continuous from Ω to E 0 we say that I is of class C 1 on Ω and we

write I ∈ C 1 (Ω).

Remark 1.5 • Notice that I 0 (u) is defined on E, even if I is defined only in Ω.

• If I is Fréchet differentiable at u ∈ Ω then I is continuous at u.

• If I is Fréchet differentiable at u ∈ Ω then I is Gâteaux differentiable at u ∈ Ω and

I 0 (u) = IG0 (u).

The converse of the third point of Remark 1.5 is not always true but we have the following
result:

Proposition 1.3 Suppose that Ω ⊆ E is on open set, such that I G-differentiable in Ω and
that IG0 : Ω ∈ E → E 0 is continuous, then I is also Fréchet differentiable at u, and we have
IG0 = I 0 (u).

The proof of Proposition 1.3 is given in [3].

Remark 1.6 The importance of Proposition 1.3 reside in the fact that calculate the Gâteau
derivative and then to prove that it is continuous is often technically easier to directly prove the
Fréchet differentiability. Also we deduce that the functional is of class C 1 .

14 Abdelhak MOKHTARI
 Variational methods for nonlinear elliptic problems
Example 1.10 We prove that the functional J : Lp (]0, 1[) → R (p > 2) defined by
Z 1
u → J(u) = |u|p dx.
0
p
is a Fréchet differentiable on L (]0, 1[). Indeed; we already prove that J is G-differentiable, with
Z 1
hJG0 (u), vi =p |u|p−1 uv dx, ∀u, v ∈ Lp (]0, 1[). (1.10)
0
it remains to prove that

JG0 : Lp (Ω) → (Lp (Ω))0

u 7→ JG0 (u)

where JG0 (u) is defined as in (1.10).

Let (un ) ⊂ Lp (Ω) such that
un → u in Lp (Ω),
we prove that JG0 (un ) → JG0 (u) in (Lp (Ω))0 . By definition we have,

kJG0 (un ) − JG0 (u)k(Lp (Ω))0 = sup |hJG0 (un ) − JG0 (u), vi|
kvkLp =1
p
let v ∈ L (Ω), such that kvkLp = 1, we have

| hJG0 (un ) − JG0 (u), vi | = | hJG0 (un ), vi − hJG0 (u), vi |

Z 1 Z 1
= p| |un |p−2 un vdx − |u|p−2 uv dx|
0 0
Z 1 Z 1
p−2
≤ p| |un | un vdx − |u|p−2 uvdx|
0 0
Z 1 p
Z 1
1
p−2 p−2
≤ p[ ||un | un − |u| u| p−1 dx]( |v|p dx) p
0 0
Z 1 p
≤ p ||un |p−2 un − |u|p−2 u| p−1 dx.
0
p
We set wn = ||un |p−2 un − |u|p−2 u| p−1 , we know that un → u in Lp (0, 1), there is a subsequence
unk = un , such that un (x) → u(x) a.e. in ]0, 1[, so we have

wn (x) → 0a.e. in]0, 1[. (1.11)

Moreover there exists g ∈ Lp such that

|un (x)| ≤ g(x) a.e. in ]0, 1[.

Thus,
p
|un (x)|p−1 ≤ |g(x)|p−1 ∈ L p−1 (0, 1) ⊂ L1 (0, 1). (1.12)
According (1.11) and (1.12), by the dominated convergence theorem we have wn → 0 in
L1 (Ω); that is Z 1 p
||un |p−2 un − |u|p−2 u| p−1 dx → 0,
0
as n → +∞. Consequently
| hJG0 (un ) − JG0 (u), vi | → 0.
 0
as n → +∞, which means that JG0 (un ) → JG0 (u) in Lp (0, 1) , so JG0 is continuous functional.
In conclusion, the functional J is Fréchet differentiable and it is of class C 1 .

Abdelhak MOKHTARI 15
Variational methods for nonlinear elliptic problems
1.3.3 Critical points
Definition 1.22 Let Ω be an open set of Banach space E, assume that I : Ω → R is differen-
tiable. We say that u ∈ Ω is a critical point of I, if

I 0 (u) = 0.

Remark 1.7 1. As I 0 (u) is an element of the dual space E 0 , then I 0 (u) = 0 means that
hI 0 (u), vi for all v ∈ E.

2. If u is not critical point, then we say that u is regular point of I.

3. If I(u) = c and I 0 (u) = 0 we say that u is a critical point for I at level c and we say that
c is a critical level for I. If c is not a critical level, then c is called a regular level for I.

4. The equation I 0 (u) = 0 is called the Euler, or Euler-Lagrange equation associated to the
functional I.

16 Abdelhak MOKHTARI
Bibliography

[1] H. Brezis, Analyse fonctionnelle, théorie et applications, Masson, Paris, (1983).

[2] R.A. Adams, Sobolev spaces, Academic Press, (1975)

[3] E.W.C. Van Groesen, Variational methods for nonlinear operator equations, Mathema-
tisch Centrum, Amsterdam (1979).

17