Contents

3 Variational and nonvariational methods for elliptic PDE 2

3.1 Minimization and Euler equation . . . . . . . . . . . . . . . . 2
3.1.1 Minimization without constraints . . . . . . . . . . . . 2
3.1.2 Minimization with constraint . . . . . . . . . . . . . . 12
3.2 Nonvariational method . . . . . . . . . . . . . . . . . . . . . . 17
3.2.1 Method of sub- and supersolution . . . . . . . . . . . . 17
3.2.2 Fixed point Method . . . . . . . . . . . . . . . . . . . 23

1
Chapter 3
Variational and nonvariational
methods for elliptic PDE

For the construction of this chapter, we refer the books of Evans [1] and
the books of Ambrosetti et al. [2, 3]. In this chapter, we study methods to
solve elliptic PDEs. More precisely we search to prove the existence of weak
solutions of general Euler PDEs as follow:
in Ω

−∆u = f (u) + h
(E)
u=0 on ∂Ω
where Ω is a smooth bounded domain, f and h are real continuous functions.
The problem (E) is called semilinear as f is not linear or constant, for instance
f (t) = t2 , f (t) = et or f (t) = t|t|p−1 for some p > 1. In this chapter, we
study semilinear elliptic PDEs in general involving the Dirichlet boundary
condition u = 0.

3.1 Minimization and Euler equation

3.1.1 Minimization without constraints

In this section, we treat the question of existence of the weak solutions of

equations whose the form is as (E) by minimization without constraint. We
also consider equations which belong to a larger class of Euler equation.
Indeed, we consider the equations involving a general functional L : (ξ, s, x) ∈
RN × R × Ω → R and the nonhomogenous Dirichlet boundary condition

2
dened as:
in Ω

− div(∇ξ L(∇u, u, x)) + ∂s L(∇u, u, x) = 0
(P)
u=g on ∂Ω

Motivation

To dene a weak solution of (P), we need some growth conditions on L and

its derivatives (in the same way on f and h for equation (E)). We assume
the function L satises for any (ξ, s, x) ∈ RN × R × Ω:
|L(ξ, s, x)| ≤ C(1 + |ξ|p + |s|p ) (3.1)
and
|∇ξ L(ξ, s, x)| + |∂s L(ξ, s, x)| ≤ C(1 + |ξ|p−1 + |s|p−1 ) (3.2)
for some constant C > 0.
With theses conditions, we say u ∈ W 1,p (Ω) whose the trace of u is g on ∂Ω,
is a weak solution of problem (P) for the Euler equation provided, for any
ϕ ∈ W01,p (Ω)
 
∇ξ L(∇u, u, x).∇ϕ dx + ∂s L(∇u, u, x)ϕ dx = 0. (3.3)
Ω Ω

Remark 1. The growth conditions allow to give a sense for the previous
formulation. Indeed, (3.2) implies that for any v ∈ W 1,p (Ω), ∇ξ L(∇v, v, x)
and ∂s L(∇v, v, x) belong to the dual space of Lp .
These conditions give conditions examples but they are not necessary opti-
mal.
In the same way, we say a weak solution u of the Euler equation (E) is a
function belonging to a suitable functional space (depending on the properties
of f or g ) as for instance H01 (Ω) such that for any ϕ belonging to another
suitable set:  
∇u∇ϕ dx = (f (u) + h)ϕ dx.
Ω Ω
The functional space is chosen obviously to give a sense for each integral in
the previous denition.
Rewriting the previous equality as follows:
 
∇u∇ϕ dx − (f (u) + h)ϕ dx = 0,
Ω Ω

3
the solution u of (E) can be seen as a critical point of a energy function J
dened on a suitable functional space by
 
1 2
J(v) = |∇v| dx − F (v) + hv dx
2 Ω Ω

where F is an antiderivative function of f . Indeed, if we assume that J is

dierentiable in u the solution of (E), we have for any ϕ:
 
0 = du J(ϕ) = ∇u∇ϕ dx − (f (u) + h)ϕ dx.
Ω Ω

Hence to obtain a weak solution of (E), we search critical points of a suitable

energy function. More precisely, in this section we are going to determine a
global minimizer of the energy function i.e. J(u) = min J(v).
Returning to the case of the Euler problem (P), a weak solution can be
interpreted as a critical point of the following energy functional dened by

I(v) = L(∇v, v, x) dx
Ω

on an admissible set A = {v ∈ W 1,p (Ω) | v = g on ∂Ω in the trace sense}.

Remark 2. We remark that if L(ξ, s, x) = |ξ|2 −F (s)−h(x)s and g = 0 thus
I(v) = J(v) dened on H01 . Namely, the Euler equation (E) is a particular
case of (P). However, we continue to study the both equations to highlight
the diculties of the computations and to be able to extend the method for
other PDEs.

Existence of minimizers

In this part, we give some conditions on the energy functions I and J which
ensure the existence of a minimizer.
We begin with a classical result dealing with functional J : E → R (where
E is a Banach space) which are coercive i.e.

lim J (u) = +∞
kuk→∞

and weakly lower semi-continuous (for short w.l.s.c.) i.e. for any sequence
(un ) ⊂ E such that un * u, then J (u) ≤ lim inf J (un ). Thus we have

4
Theorem 3.1. Let E be a reexive Banach space and let J : E → R be
coercive and w.l.s.c.. Then J has a global minimum on E namely there
exists u0 ∈ E such that J (u0 ) = minu∈E J(u).
Application:
We dene J : H01 (Ω) → R as follows
 
2
J(v) = |∇u| dx − F (x, u) dx
Ω Ω

where Ω is a smooth bounded domain of RN ,

 t
F (x, t) = f (x, s) ds
0

and f : Ω × R → R is a Carathéodory function, more precisely f satises

ˆ s → f (x, s) is continuous for almost every x ∈ Ω;

ˆ x → f (x, s) is measurable for all s ∈ R.

Moreover, we assume that there exists a constant c > 0 and α ∈ (0, 1) such
that
|f (x, t)| ≤ c(1 + |t|α ), for any (x, t) ∈ Ω × R. (3.4)
Thus, we have the following result


Proposition 3.2. The mapping H01 (Ω) 3 u → Ω
F (x, u) dx is C 1 .

Proof. (Proof in video)

We dene φ : H01 (Ω) → R as follows

φ(u) = F (x, u(x)) dx.
Ω

Step 1: The mapping u → f (., u) is continuous from L (Ω) to L N +2 (Ω).

2N 2N
N −2

Let (un ) ⊂ L N −2 (Ω) such that un → u in L N −2 . Then, up to a subsequence,

2N 2N

un → u a.e. in Ω and there exists h ∈ L N −2 (Ω) such that |un | ≤ h in Ω.

2N

Using (3.4) and since α ∈ (0, 1), we have

2N
|f (x, un )| ≤ c(1 + |un |α ) ≤ c(1 + hα ) ∈ L N +2 (Ω)

5
and we deduce from Dominated convergence theorem the continuity in L N −2 .
2N

Step 2: The mapping u → F (., u) is continuous from L (Ω) to L1 (Ω).

2N
N −2

We note, using Young inequality: ab ≤ p1 ap + p10 b where p0 = p−1

p0 p
is the
conjugate exponent of p, that for any (x, t) ∈ Ω × R, there exist constants c̃1
and c̃2 such that
(3.5)
2N
|F (x, t)| ≤ c̃1 (1 + |t|1+α ) ≤ c̃2 (1 + |t| N −2 ).

Hence, since H01 (Ω) ⊂ L N −2 (Ω) by Sobolev embedding, the mapping x →

2N

F (x, u(x)) belongs to L1 (Ω) and φ is well-dened.

Using the same idea as Step 1 and the previous inequalities, we deduce Step
2. We obtain also the continuity of F from L2 to L1 .
Step 3: φ is continuous in H01 (Ω).

The function φ can be obtained by composition according the following dia-

gram:
Id 2N F
H01 (Ω) → L N −2 (Ω) → L1 (Ω) →
R
u → u → F (., u) → Ω F (x, u(x)) dx

where Id stands for the embedding of H01 into L N −2 .

2N

Step 4: φ is dierentiable on H01 (Ω).

By Sobolev embedding, H01 (Ω) ⊂ L N −2 (Ω), by Step 1 for any u, v ∈ H01 (Ω),
2N

f (., u)v ∈ L1 (Ω). Then we introduce N : H01 (Ω) → H −1 the mapping dened
for any u ∈ H01 (Ω) by,

N (u).v = f (x, u(x))v(x) dx for v ∈ H01 (Ω).
Ω

First, we prove F : u → F (., u) is Fréchet-dierentiable from L N −2 (Ω) to

2N

L1 (Ω) and
dF (., u).v = f (., u)v.
We dene for any u, v ∈ L
2N
N −2 (Ω)

R(u, v) = kF (., u + v) − F (., u) − f (., u)vkL1 .

6
By Taylor expansion and Hölder inequality, we get:
R(u, v) ≤ kvk 2N kwk 2N
L N −2 L N +2

where  1
w(x) = |f (x, u + θv) − f (x, u)| dθ.
0

Then, with r = 2N
N +2
, we apply Hölder inequality and Fubini Theorem:
  1
kwkrLr ≤ |f (x, u + θv) − f (x, u)|r dθdx
Ω 0
 1
= |f (x, u + θv) − f (x, u)|r dxdθ
0 Ω
 1
= kf (x, u + θv) − f (x, u)krLr dθ.
0

Finally, by Step 1 the mapping u → f (., u) is continuous from L N −2 to

2N

L N +2 , we deduce that kwkLr goes to 0 as kvk N2N−2 → 0 and hence R(u, v) =

2N

L
o(kvkL1 ).
From the derivative of composite mappings, it follows that φ is dierentiable
and 
dφ(u).v = N (u).v = f (x, u(x))v(x) dx.
Ω

It remains to prove that N is continuous on H01 (Ω). We have


kN (u) − N (v)k = sup (f (x, u(x) − f (x, v(x))w dx
kwkH 1 ≤1 Ω
0

≤ sup kf (., u) − f (., v)k 2N kwk 2N

kwkH 1 ≤1 L N +2 L N −2
0

≤ kf (., u) − f (., v)k 2N

L N +2

Hence, if v → u in H01 (Ω), by Sobolev embedding v → u in L N −2 (Ω), we

2N

conclude by Step 1, kN (u) − N (v)k → 0.

Now, we come back to the study of the energy function J (computations in
video). From the previous proposition J is a C 1 function on H01 (Ω) and we

7
have: for any ϕ ∈ H01 (Ω)
 
dJ(u).ϕ = ∇u∇ϕ dx − f (x, u)ϕ dx.
Ω Ω

We arm, with the conditions on f , J is coercive and w.l.s.c.. Indeed, we

have, for any v ∈ H01 (Ω)

1 1
J(v) = kvk2H 1 − F (x, v) dx ≥ kvk2H 1 − c1 kvkL1 − c2 kvkα+1
Lα+1
2 0
Ω 2 0

1
≥ kvk2H 1 − c1 kvkH01 − c3 kvkα+1
H01
2 0

where we use the Sobolev embedding H01 ⊂ Lα+1 in the last inequality.
Hence, since α + 1 < 2, we deduce that
lim J(u) = +∞.
kvkH 1 →∞
0

Moreover, let (un ) ⊂ H01 (Ω) such that un * u in H01 . Taking into account
that H01 is compactly embedded in L2 , we deduce that un → u in L2 (Ω).
From (3.5) and by the same computations of Step 1, we
can deduce φ is
continuous on L2 (Ω). Finally, we have by convexity of v → Ω |∇v|2 dx:
 
1
J(un ) = |∇un |2 − F (x, un ) dx
2 Ω
 Ω  (3.6)
1 2
≥ |∇u| dx + ∇u.∇(un − u) dx − F (x, un ) dx
2 Ω Ω Ω

and passing to the limit, we have

  
∇u.∇(un − u) dx → 0 and F (x, un ) dx → F (x, u) dx.
Ω Ω Ω

In the rst limit, we have used un * u in H01 and in the second limit, we use
the continuity of F and un → u in L2 . Hence we get lim inf J(un ) ≥ J(u).
We apply Theorem 3.1 which gives the existence of global minimizer u0 of
J on H01 (Ω). Since J is dierentiable, u0 is a critical point of J and this
implies dJ(u0 ) = 0. Hence we prove that there exists a weak solution u0 of
the following Euler equation:
in Ω

−∆u = f (u)
u=0 on ∂Ω

8
 in sense that for any ϕ ∈ H01 (Ω), u0 satises
 
∇u∇ϕ dx = f (x, u)ϕ dx.
Ω Ω

We use the convexity in (3.6) to obtain the weak lower semicontinuity of

J . Indeed this geometric property is suitable in the study of global mini-
mizer. For instance, we have the following theorem:
Theorem 3.3. Let E a Banach space and let J : E → R be convex function.
Assume J is Gâteaux dierentiable on E . Then J is w.l.s.c. in E .
Proof. Since J is convex and Gâteaux dierentiable on E , for any u, v ∈ E ,
we have:
J (v) ≥ J (u) + J 0 (u).(v − u)
where J 0 (u) is dierential function of J at u in sense of Gâteaux.
Let (un ) ⊂ E such that un * u in E , Then, we have
J (un ) ≥ J (u) + J 0 (u)(un − u).
Since J 0 (u) ∈ E 0 , it is sucient to pass to the limit in the previous inequality
to conclude.
Theorem 3.3 can be released considering the energy function I
Theorem 3.4. Assume that the mapping L is smooth, bounded from below
and convex with respect to the rst variable i.e. RN 3 ξ → L(ξ, t, x) is convex
for each t ∈ R, x ∈ Ω.
Then, I is w.l.s.c. on W 1,p (Ω).
Proof. For the proof, see the proof of Theorem 1 Paragraph 8.2 of [1].
Hence we can extend Theorem 3.1 considering the functional I .
Theorem 3.5. Assume that there exist nonnegative constants a and b such
that for any ξ ∈ RN , t ∈ R and x ∈ Ω:
L(ξ, t, x) ≥ a|ξ|p − b. (3.7)
Assume in addition L is convex with respect to the rst variable and A =
6 ∅.
Then, there exists at least u0 ∈ A such that
I(u0 ) = min I(v).
v∈A

9
Remark 3. The condition (3.7) does not give the coercivity in the sense
whom we have dened early. Indeed, the condition (3.7) implies I(v) ≥
αk∇vkpLp − β for some α, β i.e. I(v) → +∞ as k∇vkLp → ∞.

Proof. If g = 0, Poincaré inequality implies k∇vkLp is a norm on A =

W 1,p (Ω) and hence we apply Theorem 3.1.
For the general case, we dene m = inf v∈A inf I(v). If m = +∞, we are
done. Hence assume m < +∞ and we consider (un ) a minimizing sequence
i.e. I(un ) → m as n → +∞.
Without losing generalities, we can assume b = 0 in the inequality (3.7).
Thus we get 
I(un ) ≥ |∇un |p dx
Ω

and we deduce the sequence (k∇un kLp ) is bounded. Let v ∈ A xed, then
for any n, un − v ∈ W01,p (Ω). Hence by Poincaré inequality:
kun kLp ≤ kun − vkLp + kvkLp
≤ Ck∇(un − v)kLp + kvkLp
≤ C̃

where C does not depend on n.

Hence (un ) is bounded in W 1,p (Ω). Consequently there exist a subsequence
(un0 ) and u ∈ W 1,p (Ω) such that un0 * u in W 1,p (Ω).
It remains to prove that u ∈ A. Since W01,p (Ω) is a closed, linear subspace
of W 1,p (Ω). Thus W01,p (Ω) is weakly closed (here we use that any C convex
and closed in the strong topology then C is weakly closed).
We deduce u − v ∈ W01,p (Ω) and the trace of u is equal to g on ∂Ω.
Finally, by Theorem 3.4, I is w.l.s.c. then
I(u) ≤ lim
0
inf I(un0 ) = m ≤ I(u).
n →+∞

Since u ∈ A, we conclude the result of Theorem 3.5.

Theorem 3.6. Assume L satises the growth conditions (3.1)-(3.2) and u0 ∈
A veries I(u0 ) = minv∈A I(v).
Then u0 is a weak solution of (P) in sense that u0 satises (3.3).

10
Proof. (Proof in video)
Let v ∈ W01 (Ω). Then for any t ∈ R, w = u0 + tv ∈ A. We dene I(t) =
I(w) = I(u0 + tv), for any t ∈ R. Then, we have

I(t) − I(0) 1
= L(∇w, w, x) − L(∇u0 , u0 , x) dx. (3.8)
t t Ω

As t → 0, we get almost everywhere x ∈ Ω:

1
(L(∇w, w, x) − L(∇u0 , u0 , x)) → ∇ξ L(∇u0 , u0 , x).∇v + ∂s L(∇u0 , u0 , x)v
t
and Taylor expansion yields
 t
d
L(∇w, w, x) − L(∇u0 , u0 , x) = L(∇(u0 + θv), u0 + θv, x) dθ
0 dθ
 t
= ∇ξ L(∇(u0 + θv), u0 + θv, x).∇v dθ
0
 t
+ ∂s L(∇(u0 + θv), u0 + θv, x)v dθ.
0

Consider the second term in the right hand-side, the next estimate also holds
for the rst term in the right hand-side. Without losing generalities we
assume t > 0.
Using, rst the Young inequality ab ≤ p1 ap + p10 bp and (3.2), we get for some
0

constants C
 t
∂s L(∇(u0 + θv), u0 + θv, x)v dθ
0
  t 
p p0
≤ C t|v| + |L(∇(u0 + θv), u0 + θ| dθ
0
  t 
p p−1

0
p−1 p
≤ C t|v| + 1 + |∇(u0 + θv)| + |u0 + θv| dθ
0
  t 
p p p
≤ C t|v| + 1 + |∇(u0 + θv)| + |u0 + θv| dθ
0
≤ Ct 1 + |v|p + |∇u0 |p + |u0 |p + tp (|∇v|p + |v|p ) .

Finally, for t small enough: t < 1, we obtain

 t
1
∂s L(∇(u0 + θv), u0 + θv, x)v dθ ≤ C(1 + |v|p + |∇u0 |p + |u0 |p + |∇v|p )
t 0

11
where the terms in right-hand side belong to L1 (Ω).
Hence, applying the Dominated Convergence Theorem in (3.8), we conclude
by that
 
0
I (0) = ∇ξ L(∇u0 , u0 , x).∇ϕ dx + ∂s L(∇u0 , u0 , x)ϕ dx.
Ω Ω

Since I admits a minimum for t = 0, we have I 0 (0) = 0 and thus u0 is a

weak solution of (P).
Remark 4.
ˆ In the previous proof, we do not need the dierentiability of L to obtain
a weak solution. In general, the Gâteaux dierentiability is sucient.
ˆ This minimization method works in case there exists a global mini-
mizer. However, the method do not detect the local minimizers or the
saddle-points. For these other points, the Mountain Pass Theorem (see
Theorem 2 paragraph 8.5 in [1]) is a suitable tool but it asks more
conditions.

3.1.2 Minimization with constraint

In this section, we consider application of the calculus of variations to min-

imization problems with constraints and we discuss the importance of La-
grange multipliers in the corresponding Euler equation. More precisely, we
study in this section Euler equations (E) with h = 0.
Let begin with an example, consider the following problem for N ≥ 3:
in Ω

−∆u = |u|p−1 u
u=0 on ∂Ω
where p ∈ (1, NN −2
+2
). We dene the energy function as
 
1 2 1
J(v) = |∇v| dx − |u|p+1 dx.
2 Ω p+1 Ω

Since p + 1 < N2N−2 = 2∗ (where 2∗ is the critical exponent in the Sobolev

embedding H 1 ⊂ Lq for q ≤ 2∗ ), the functional J is well-dened on H01 (Ω)
and we have by Sobolev embedding and Poincaré inequality:
1
J(v) ≥ k∇uk2L2 − Ck∇u|p+1
L2 .
2
12
Since p > 1, we can not conclude the coercivity of J . Hence the previous
minimization method fails.
We come back the general problem (E). An approach to prove the existence
of a weak solution is to look at the problem of minimizing the function:

˜ =1
J(v) |∇u|2 dw
2 Ω

over all functions v ∈ H01 (Ω) with integral constraint, namemly under an
suitable condition: 
C(v) = F (v) dx = 0,
Ω
where F : R → R is a smooth function such that f = F 0 . Here we assume f
satises
∀t ∈ R |f (t)| ≤ C(1 + |t|) (3.9)
for some constant C > 0. Hence we introduce the suitable admissible set
A = {v ∈ H01 (Ω) | C(v) = 0}. Then we have the following result:
Theorem 3.7. Under the previous conditions on F and assume A =
6 ∅.
Then, there exists u0 ∈H01 (Ω) such that J(u0 ) = minv∈A J(v).
˜ ˜
Proof. (Proof in video)
Dene m = inf v∈A J(v)
˜ ≥ 0 and a minimizing sequence (un ) belonging to A
with J(u
˜ n ) → m as n → +∞.
Obviously, (un ) is bounded in H01 (Ω). Hence we can extract a subsequence
(un0 ) such that un0 * u in H01 (Ω). Moreover, J˜ is w.l.s.c. and we deduce
˜
J(u) ≤ m.
It remains to prove that C(u) = 0.
We have H01 (Ω) ⊂ L2 (Ω) with compact embedding, hence we deduce un0 → u
in L2 (Ω).
Since un ∈ A, we can write:

|C(u)| = |C(u) − C(un0 )| ≤ |F (u) − F (un0 )| dx
Ω
  1
≤ |u − un0 | |f (u + θ(un0 − u))| dθdx
Ω 0

≤C |u − un0 |(1 + |u| + |un0 |) dx

Ω
≤ Cku − un0 kL2 (1 + kun0 kL2 + kukL2 )
≤ Cku − un0 kL2 → 0 as n0 → +∞

13
Thus, u ∈ A.
The next theorem gives the existence of a weak solution of the corre-
sponding Euler equation coming from the minimization problem. For this
result, we recall another well-known analysis theorem:
Theorem 3.8 (Implicit function theorem). Let X be a vectorial space, Y be
a Banach space and O ⊂ X × Y an open set. Consider F : O → Y a C m −
function in O. Assume there exists (x0 , y0 ) ∈ O such that F (x0 , y0 ) = 0 and
dy F (x0 , y0 ) is invertible. Then, there exist open neighbourhoods U ⊂ X of x0
and V ⊂ Y of y0 and a C m − function Ψ such that
(x, y) ∈ U × V and F (x, y) = 0 ⇔ y = Ψ(x).

In particular, for any x ∈ U , F (x, Ψ(x)) = 0.

Hence we have
Theorem 3.9. Under the previous conditions on F , let u0 ∈ A satisfy
J(u0 ) = minv∈A J(v).
˜ ˜
Then, there exists λ ∈ R (Lagrange multiplier) such that for any ϕ ∈ H01 (Ω):
 
∇u0 .∇ϕ dx = λ f (u0 )ϕ dx. (3.10)
Ω Ω

Proof. (Proof in video)

Consider two cases f (u0 ) ≡ 0 or f (u0 ) 6≡ 0 a.e. in Ω. In the rst case,
from Ω is connected we deduce that F (u0 ) = Cst in Ω. Indeed, ∇F (u) =
f (u)∇u = 0 . a.e. in Ω (approximating f by bounded functions and using
the derivative
of composite mappings in Sobolev space). Since u belongs to
A, C(u0 ) = Ω F (u0 ) dx = 0, we deduce F (u0 ) = 0 and furthermore u0 = 0
on ∂Ω implies C(0) = 0 and then 0 ∈ A.
Finally, the rst case implies u0 ≡ 0 otherwise J(u ˜ 0 ) > J(0)
˜ = 0. Hence,
since f (u0 ) = 0, (3.10) is veried for any λ ∈ R.
For the second case, since f (u0 ) 6≡ 0, we choose w ∈ H01 (Ω) such that

f (u0 )w dx 6= 0.
Ω

14
Dene a : R2 → R as a(s, t) = C(u0 + sϕ + tw) for any function ϕ ∈ H01 (Ω).
Then, a ∈ C 1 (R2 ) and

∂a
(s, t) = f (u0 + sϕ + tw)ϕ dx
∂s
Ω (3.11)
∂a
(s, t) = f (u0 + sϕ + tw)w dx
∂t Ω

Note that a(0, 0) = C(u0 ) = 0 and


∂a
(0, 0) = f (u0 )w dx 6= 0.
∂t Ω

Hence applying the Implicit function Theorem, there exist Ψ : R → R a C 1 -

function and s0 small enough such that Ψ(0) = 0 and for any s ∈ (−s0 , s0 ),
a(s, Ψ(s)) = 0. Moreover, we have

∂a ∂a
(s, Ψ(s)) + Ψ0 (s) (s, Ψ(s)) = 0
∂s ∂t
and thus 
f (u0 )ϕ dx
Ψ (0) = −  Ω
0
.
f (u0 )w dx
Ω

Now we dene b(s) = J(u

˜ 0 + sϕ + ψ(s)w). Then, for any s ∈ (−s0 , s0 ),
a(s, Ψ(s)) = C(u0 + sϕ + Ψ(s)w) = 0, we deduce sϕ + Ψ(s)w ∈ A and b
admits a minimum at 0 on (−s0 , s0 ). Thus, we get

0
b (s) = ∇(u0 + sϕ + Ψ(s)w).∇(ϕ + Ψ0 (s)w) dx
Ω

and hence

0
0 = b (0) = ∇u0 .∇(ϕ + Ψ0 (0)w) dx
Ω 
 f (u0 )ϕ dx 
= ∇u0 .∇ϕ dx − Ω ∇u0 .∇w dx.
Ω Ω
f (u0 )w dx
Ω

15
Set 
∇u0 .∇w dx
λ= Ω
f (u0 )w dx
Ω

where w is xed. We deduce, for any ϕ ∈ H01 (Ω),

 
∇u0 .∇ϕ dx = λ f (u0 )ϕ dx.
Ω Ω

16
3.2 Nonvariational method

3.2.1 Method of sub- and supersolution

We focus now on a basic property of elliptic PDE: the Maximum principle.

The dierent resulting comparison arguments are suitable tools in the reso-
lution of certain semilinear problems. The idea is to use ordering properties
for solutions. Indeed, if we have a sub- and a supersolution of a particular
equation such that the subsolution is smaller than the supersolution, then
we can nd a solution which is between the sub- and the supersolution.
In this section, we consider the equation (E) with h ≡ 0, namely
in Ω

−∆u = f (u)
(S)
u=0 on ∂Ω
where we assume that f : R → R is globally Lipschitz on R i.e. there exists
Λ > 0 such that

|f (s) − f (t)| ≤ Λ|u − v| ∀s, t ∈ R.

Remark 5. A direct consequence of this assumption is that the mapping

t → f (t) + λt is nondecreasing for any λ ≥ Λ. Indeed, let t, s ∈ R such that
t ≤ s, we have

f (s) − f (t) + λ(s − t) ≥ −Λ|s − t| + λ(s − t) ≥ (λ − Λ)(s − t) ≥ 0.

We need to dene the notions of sub- and supersolutions.

Denition 3.10. Let u, u ∈ H 1 (Ω). We say that u is a weak subsolution of
(S ) if  
∇u.∇ϕ dx ≤ f (u)ϕ dx
Ω Ω

for any ϕ ∈ H01 (Ω), ϕ ≥ 0 a.e. in Ω and u ≤ 0 on ∂Ω in the trace sense. In

the same way, we say that u is a weak supersolution of (S ) if
 
∇u.∇ϕ dx ≥ f (u)ϕ dx
Ω Ω

for any ϕ ∈ H01 (Ω), ϕ ≥ 0 a.e. in Ω and u ≥ 0 on ∂Ω in the trace sense.

17
Remark 6. If u ∈ C 2 (Ω) ∩ C(Ω) (respectively u), then u is a classical
subsolution (respectively supersolution) of (S ) if
−∆u ≤ f (u) in Ω −∆u ≥ f (u) in Ω

 
resp.
u≤0 on ∂Ω u≥0 on ∂Ω
With these notions, we have the following existence theorem:
Theorem 3.11. Assume there exist u, u a weak sub- and supersolution of
(S ) such that
u ≤ u a.e. in Ω.
Then, there exists a weak solution u ∈ H01 (Ω) of (S ) such that
u ≤ u ≤ u a.e. in Ω.

Proof. (Proof in video)

Let λ ≥ Λ where we recall that Λ is the Lipschitz constant of f .
We dene inductively the sequence (un ) ⊂ H01 (Ω) by u0 = u and un+1 is the
weak solution of the following problem:
in Ω

−∆un+1 + λun+1 = f (un ) + λun
(Sn )
un+1 = 0 on ∂Ω.
We set fλ = f + λId to simplify the computations. By remark 5, fλ is
nondecreasing on R. By Exercise 1 Series 2 Chapter 3, the sequence (un ) is
well-dened.
Step 1: for any n ∈ N, u ≤ un ≤ u a.e. in Ω.
First, we are going to prove u0 = u ≤ u1 ≤ u
Since u0 = u is a weak subsolution and u1 is a weak solution of (Sn ) with
n = 0, we have
  
∇u1 .∇ϕ + λu1 ϕ dx = fλ (u0 )ϕ dx ≥ ∇u0 .∇ϕ + λu0 ϕ dx
Ω Ω Ω

for any ϕ ∈ H01 (Ω), ϕ ≥ 0. Hence, for ϕ = (u0 − u1 )+ = max(0, u0 − u1 ) ∈

H01 (Ω), we get:

∇(u0 − u1 ).∇(u0 − u1 )+ + λ(u0 − u1 )(u0 − u1 )+ dx ≤ 0.
Ω

18
Moreover, we know that for any u ∈ H 1 (Ω), thus u+ ∈ H01 (Ω) and
if u(x) ≥ 0;

+ ∇u(x)
∇u (x) =
0 if u(x) < 0.
Hence we deduce, with the notation {u0 ≥ u1 } = {x ∈ Ω | u0 (x) ≥ u1 (x)}:

0≤ |∇(u0 − u1 )|2 + λ|u0 − u1 |2 dx ≤ 0
{u0 ≥u1 }

and then u = u0 ≤ u1 a.e. in Ω.

In the same way, considering the supersolution u we get for any ϕ ∈ H01 (Ω),
ϕ≥0
  
∇u1 .∇ϕ + λu1 ϕ dx = fλ (u0 )ϕ dx ≤ fλ (u)ϕ dx
Ω Ω
 Ω

≤ ∇u.∇ϕ + λuϕ dx
Ω

Hence, for ϕ = (u1 − u)+ ∈ H01 (Ω), this yields:


0≤ |∇(u1 − u)2 + λ(u1 − u)2 dx ≤ 0
{u1 ≥u}

and we deduce u1 ≤ u.
Let n ≥ 2, assume now that u ≤ un−1 ≤ u a.e. in Ω.
We want to prove that u ≤ un ≤ u a.e. in Ω.
By the induction hypothesis and since fλ is nondecreasing, we have for any
ϕ ∈ H01 (Ω), ϕ ≥ 0:
  
∇un .∇ϕ + λun ϕ dx = fλ (un−1 )ϕ dx ≥ fλ (u)ϕ dx
Ω Ω Ω (3.12)
≥ ∇u.∇ϕ + λuϕ dx
Ω

and
  
∇un .∇ϕ + λun ϕ dx = fλ (un−1 )ϕ dx ≤ fλ (u)ϕ dx
Ω Ω Ω (3.13)
≤ ∇u.∇ϕ + λuϕ dx.
Ω

19
Taking ϕ = (u − un )+ in (3.12) and ϕ = (un − u)+ in (3.13), we obtain

0≤ |∇(u − un )|2 + λ|u − un |2 dx ≤ 0
{u≥un }

and 
0≤ |∇(un − u)2 + λ(un − u)2 dx ≤ 0.
{un ≥u}

From the both last inequality, we deduce u ≤ un ≤ u a.e. in Ω.

Step 2: (un ) is nondecreasing.
We have already proved u0 = u ≤ u1 . Assume now un−1 ≤ un and we are
going to prove un ≤ un+1 .
Since un (resp. un+1 ) is a weak solution of (Sn ) with n (resp. with n + 1),
we have  
∇un .∇ϕ + λun ϕ dx = fλ (un−1 )ϕ dx
Ω Ω
and respectively
 
∇un+1 .∇ϕ + λun+1 ϕ dx = fλ (un )ϕ dx. (3.14)
Ω Ω

for any ϕ ∈ H01 (Ω).

Subtracting the two previous equalities and taking ϕ = (un − un+1 )+ ∈
H01 (Ω), we obtain as previously

0≤ |∇(un − un+1 )|2 + λ|un − un+1 |2 dx
{un ≥un+1 }

= (fλ (un−1 ) − fλ (un ))(un − un+1 )+ dx ≤ 0.
Ω

The last inequality is obtained using fλ nondecreasing and by induction hy-

pothesis un−1 ≤ un a.e. in Ω. Hence we conclude un ≤ un+1 a.e. in Ω.
From Step 1 and Step 2, we have:
u ≤ ... ≤ un ≤ un+1 ≤ ... ≤ u. (3.15)
Step 3: Existence of a weak solution of (S )
Dene u(x) = limn→∞ un (x). Since (3.15) holds , the Dominated convergence

20
Theorem implies the sequence (un ) converges to u in L2 (Ω). Moreover we
have,
 
|f (un ) − f (u)| dx ≤ Λ 2 2
|un − u|2 dx → 0 as n → ∞
Ω Ω

and thus f (un ) → f (u) in L (Ω). 2

Taking ϕ = un+1 in (3.14), Hölder inequality implies,

 
2
|∇un+1 | dx ≤ |∇un+1 |2 + λ|un+1 |2 dx
Ω Ω
= f (un )un+1 + λun un+1 dx
Ω


≤ kun+1 kL2 kf (un )kL2 + kun kL2
≤ C.
Hence (un ) is bounded in H01 (Ω) and then there exists a subsequence (unk )
such that unk * u in H01 (Ω).
Now, we claim that u is a weak solution of (S ). Indeed, we have for any
ϕ ∈ H01 (Ω) and for any k :
  
∇unk .∇ϕ dx + λunk ϕ dx = f (unk −1 )ϕ + λunk −1 ϕ dx.
Ω Ω Ω

As k → ∞, the weak convergence in H01 implies that the rst term in the
left hand-side goes to 
∇u.∇ϕ dx
Ω
while the strong convergence in L (Ω) of a subsequence produces the limits:
2
  
unk ϕ dx, unk −1 ϕ dx → uϕ dx
Ω Ω Ω

and  
f (unk −1 )ϕ dx → f (u)ϕ dx.
Ω Ω
Finally, we obtain
 
∇u.∇ϕ + λuϕ dx = f (u)ϕ + λuϕ dx
Ω Ω

and removing the term involving λ, we conclude u is a weak solution of

(S ).

21
 The condition on f can be weakened. Indeed, we just need that f is
Lipschitz on [u, u]. Hence considering classical sub- and supersolution, we
have the following corollary
Corollary 3.12. Let f : R → R be a function locally Lipschitz. Assume
that there exist u, u ∈ C 2,α (Ω) a sub- and supersolution of (S ) in sense of
Remark 6 such that
u≤u in Ω.
Then, there exists a classical solution u ∈ C 2,α (Ω) of (S ) such that
u ≤ u ≤ u a.e. in Ω.

Proof. (Proof in video)

The proof is close to the previous proof.
Let λ ≥ Λ̄ where Λ̄ is the Lipschitz constant of f on I = [minΩ u, maxΩ u].
Dening again fλ = f + λId, then Remark 5 holds in I and hence fλ is
nondecreasing on I .
As previously, we dene the sequence (un ) ⊂ H01 (Ω) by u0 = u and un+1 is
the weak solution of the following problem:
in Ω

−∆un+1 + λun+1 = f (un ) + λun
(Sn )
un+1 = 0 on ∂Ω.
The sequence (un ) is well-dened by Exercise 1 Series 2 Chapter 3.
Step 1 and Step 2 of proof of Theorem 3.11 hold, hence we get (3.15).
As previously, we dene u(x) = limn→∞ un (x). Step 3 is still valid and we
deduce that u is a weak solution of (S ).
From (3.15), we deduce that the sequence (fλ (un )) is bounded in L∞ (Ω) i.e.
there exists a constant C such that for any n ∈ N,
kfλ (un )kL∞ = kf (un ) + λun kL∞ ≤ C

hence by elliptic regularity, for any p ≥ 1, there exists a constant Cp inde-

pendent of n such that
kun kW 2,p ≤ Cp .
For p large enough, Sobolev embedding gives W 2,p (Ω) ⊂ C 1,β (Ω) for some
β = β(p). We deduce for any β ∈ (0, 1), there exists a constantCβ indepen-
dent of n such that
kun kC 1,β ≤ Cβ

22
and hence we get for any n ∈ N
kf (un ) + λun kC β ≤ C

For some constant C independent of n. Using the elliptic regularity in the

Hölder spaces, we deduce that (un ) is bounded in C 2,β (Ω). By Ascoli-Arzela
Theorem, C 2,β (Ω) ⊂ C 2,γ (Ω) with compact embedding if γ ∈ (0, β).
So there exists a subsequence (un0 ) such that un0 converges to u in C 2,γ (Ω).
Thus we conclude that, for any β ∈ (0, 1), u belongs to C 2,γ (Ω) for any
γ ∈ (0, β). Then, u belongs to C 2,α (Ω) and u is a classical solution of
(S ).

3.2.2 Fixed point Method

In this part, we recall rst dierent xed point Theorems and we are going to
give applications of these theorems. There is not a general theorem to solve a
PDE using xed point method but the important point is the construction of
the mapping which will admit the xed point. We begin by the most famous:
the Banach xed point Theorem:
Theorem 3.13. Let X a Banach space and A : X → X a mapping satisfying
for any x, y ∈ X
kA(x) − A(y)kX ≤ γkx − ykX
for some γ ∈ (0, 1) independent of x and y.
Then, there exists an unique x0 ∈ X such that A(x0 ) = x0 .
Application (Computations in video)
Consider the Euler equation:
in Ω

−∆u = u3 + h
u=0 on ∂Ω
where Ω ⊂ R3 is a smooth and bounded domain and h 6≡ 0 belongs to L2 (Ω).
We dene the nonlinear operator A : L6 (Ω) → L6 (Ω) as follows for any
v ∈ L6 (Ω), u = Av is the unique solution of the following problem:

in Ω

−∆u = v 3 + h
u=0 on ∂Ω

23
The existence of u comes from Lax-Milgram Theorem (or Riesz representa-
tion) with

a(U, V ) = ∇U.∇V dx dened on H01 (Ω) × H01 (Ω)
Ω

and the linear and continuous mapping on H01 (Ω):


L(ϕ) = (v 3 + h)ϕ dx.
Ω

For N=3, H01 (Ω) ⊂ L6 (Ω) and hence u = Av ∈ L6 (Ω).

First approach: we wish to apply the Banach xed point Theorem, for that
it is sucient to verify the contraction property for A i.e. for any v1 , v2
belonging to L6 (Ω), u1 = Av1 and u2 = Av2 ,
ku1 − u2 kL6 ≤ γkv1 − v2 kL6

for some γ ∈ (0, 1). We have, for any ϕ ∈ H01 (Ω),

 
∇u1 .∇ϕ dx = (v13 + h)ϕ dx
Ω Ω

and  
∇u2 .∇ϕ dx = (v23 + h)ϕ dx.
Ω Ω

Taking ϕ = u1 − u2 ∈ H01 (Ω) in the last both equations, and subtracting

them we get 
ku1 − u2 k2H 1 = (v13 − v23 )(u1 − u2 ) dx
0
Ω
We apply Hölder and Poincaré inequalities on the both terms in the right
hand-side :
ku1 − u2 k2H 1 ≤ kv13 − v23 kL2 ku1 − u2 kL2
0

≤ C1 kv13 − v23 kL2 ku1 − u2 k2H 1

0

where C1 is the best constant in the Poincaré inequality

kukL2 ≤ C1 k∇ukL2 = C1 kukH01 .

24
Hence we obtain
kAv1 − Av2 kL6 = ku1 − u2 kL6 ≤ C2 ku1 − u2 kH01 ≤ C1 C2 kv13 − v23 kL2 (3.16)

where C2 is the best constant in the Sobolev embedding H01 ⊂ L6 (since the
kukL6
embedding is continuous, we have C2 = sup ).
u∈H01 (Ω) kukH01
Problem! The mapping x → x3 is not globally Lipschitz. Hence we can not
nd γ ∈ (0, 1) without imposing restrictions on v1 , v2 and/or on the function
h such that the mapping A is γ−Lipschitz in L6 (Ω).
For this reason, we have to contemplate a new approach. More precisely, we
consider only the functions belonging to a closed ball BR = B(0, R) of L6 (Ω)
for a certain R > 0 that we will dene later. First, we have to prove the
stability i.e. A(BR ) ⊂ BR .
For this, consider v ∈ L6 and u = Av thus we have:
 
kuk2H 1 = |∇u| dx = 2
(v 3 + h)u dx
0
Ω Ω
≤ kvk3L6 kukL2 + khkL2 kukL2
≤ C1 kukH01 kvk3L6 + khkL2 .

Considering now v ∈ BR and the Sobolev embedding H01 ⊂ L6 , we get

1
kukL6 ≤ kukH01 ≤ C1 (R3 + khkL2 ).
C2
Hence, choosing khkL2 < 2 32 where α0 = C1 C2 , there exist R1 , R2 > 0
(3α0 )
such that R1 → 0 as khkL2 → 0 and for any
R ∈ [R1 , R2 ], we have
C1 C2 (R3 + khkL2 ) ≤ R.
With theses choices, we deduce A(BR ) ⊂ BR .
Now, we come back to the estimate (3.16) considering v1 and v2 belonging
to BR :
 
|v13 − v23 |2 dx = |v1 − v2 |2 |v12 + v1 v2 + v22 |2 dx
Ω Ω
≤ kv1 − v2 k2L6 kv12 + v1 v2 + v22 k2L3

2
≤ kv1 − v2 k2L6 kv1 k2L6 + kv2 k2L6 + kv1 v2 kL3
≤ 9R4 kv1 − v2 k2L6 .

25
Hence, the previous estimate yields in (3.16)
3 2
kAv1 − Av2 kL6 ≤ 3C1 C2 R2 kv1 − v2 kL6 = R kv1 − v2 kL6 .
α0
Hence, if R < √3α
1
0
, A is a contraction mapping.
Finally, plugging the restrictions on h and R, we apply the Banach xed
point in BR and we deduce under some restrictions the existence of a weak
solution of
−∆u = u3 + h in Ω,


u=0 on ∂Ω.
Remark 7. The Banach xed point Theorem gives the uniqueness in BR
but another solution ũ can exist such that kũkL6 > R.
The following xed point Theorem is an extension of the Brouwer's xed
point Theorem: Schauder's xed point theorem
Theorem 3.14. Let X be a Banach space and K ⊂ X be a compact and
convex set of X . let A : K → K be a continuous function. Then, A has a
xed point.
The last theorem is an alternative of the Schauder's xed point which is
usually more suitable to solve nonlinear PDEs. First we need to extend the
notion of compact operator for the nonlinear operators.
Denition 3.15. Let X be an Banach space. A mapping L : X → X is
said compact provided for any bounded sequence (un ) of X , the sequence
(L(un ))) admits a subsequence which converges.

With this denition, we have the Schaefer's xed point theorem:

Theorem 3.16. Let X be a Banach space and A : X → X be a continuous
and compact mapping. In addition, assume that the set {x ∈ X | x =
λA(x) for some λ ∈ [0, 1]} is bounded in X . Then, A has a xed point.

Application (Computations in video)

Consider the Euler equation:
in Ω

−∆u + a(∇u) + λu = 0
(SP F )
u=0 on ∂Ω

26
where Ω ⊂ RN is a smooth and bounded domain and a : RN → R is a smooth
Lipschitz continuous function such that there exists a constant C > 0, for
any ξ ∈ RN ,
|a(ξ)| ≤ C(1 + |ξ|). (3.17)
Dene A : H01 (Ω) → H01 (Ω) as for v ∈ H01 (Ω), u = Av is the unique solution
of
−∆u + λu = −a(∇v) in Ω


u=0 on ∂Ω
By the growth condition on a, for any v ∈ H01 (Ω), a(∇v) belongs to L2 (Ω).
Hence the existence and the uniqueness of u arise from Lax-Milgram Theo-
rem. Moreover, by the regularity theory, we have u ∈ H 2 (Ω) and
kukH 2 ≤ Cka(∇v)k ≤ C(1 + kvkH01 ) (3.18)
where the constants C are independent of u.
We arm that A : H01 (Ω) → H01 (Ω) is continuous.
Let vn → v in H01 . Then, by (3.18), we have with un = Avn , for any n
sup kun kH 2 ≤ C < ∞ (3.19)
n

where C is independent of n. Since H 2 (Ω) ⊂ H 1 (Ω) with compact embed-

ding, there exist a subsequence un0 and u ∈ H01 (Ω) such that un0 → u in
H01 (Ω).
Hence from the following equality, for any ϕ ∈ H01 (Ω)
  
∇un0 ∇ϕ dx + λ un0 ϕ dx = − a(∇vn0 )ϕ dx
Ω Ω Ω

we can pass to the limit as n0 → +∞ and we get:

  
∇u∇ϕ dx + λ uϕ dx = − a(∇v)ϕ dx.
Ω Ω Ω

This implies u = Av and thus if vn → v in H01 then Avn → Av in H01 .

By the same way, we arm that A : H01 (Ω) → H01 (Ω) is compact in the sense
of Denition 3.15. Indeed, consider (vn ) a bounded sequence of H01 (Ω), then
with un = Avn , (3.19) holds and we deduce that(un ) bounded in H 2 (Ω) which
is compactly embedded in H 1 (Ω), admits a subsequence which converges
strongly in H01 (Ω).

27
It remains to prove that the set {v ∈ H01 (Ω) | v = µAv for µ ∈ [0, 1]} is
bounded in H01 (Ω).
Let v ∈ H01 (Ω) such that v = µAv for some µ ∈ (0, 1].
Then, we have for any ϕ ∈ H01 (Ω)
  
1 λ
∇v∇ϕ dx + vϕ dx = − a(∇v)ϕ dx.
µ Ω µ Ω Ω

In particular, for ϕ = v , using the growth condition (3.17) and Young in-
equality
  
2 2
|∇v| dx + λ v dx = −µ a(∇v)v dx
Ω Ω  Ω

≤C (|∇v| + 1)v dx
Ω 
1 2
≤ |∇v| dx + C v 2 + 1 dx.
2 Ω Ω

Hence, choosing λ large enough, we get

 
1 2
|∇v| dx ≤ (C − λ) v 2 dx + C̃ ≤ C̃
2 Ω Ω

and we deduce kvkH01 ≤ C where the constant is independent of µ.

Finally, we apply the Schaefer's xe point theorem with A dened in H01 (Ω)
which implies the existence of a solution of Euler problem (SP F ).

28
Bibliography

[1] Partial Dierential Equations, L. C. Evans

[2] A primer of Nonlinear Analysis, A. Ambrosetti and G. Prodi.
[3] Nonlinear Analysis and semilinear Elliptic Problemsn A. Am-
brosetti and A. Malchiodi.

29