Chapitre 5

A Brief Introduction to Sobolev

spaces and applications

5.1 Derivatives in L2
In this first paragraph we define the Sobolev spaces of L2 -functions whose derivatives in
the sense of distributions are also in L2 .

5.1.1 Definition
We begin with the one dimensional case.
Définition 5.1. We denote by H 1 pRq the set of functions u P L2 pRq whose derivative in the
sense of distributions is in L2 pRq.
We recall that the derivative of u P L2 pRq in the sense of distributions is said to be in
L pRq if there exists v P L2 pRq such that Tu1 “ Tv . In other words,
2

ż ż
@φ P C08 pRq, ´ uφ1 dx “ vφ dx. (5.1)
R R

In this case v is unique and it is denoted by u1 .

Remark 5.2. If f P C 1 pRq is compactly supported then it belongs to H 1 pRq. In general, even
if f is of class C 1 , f and f 1 are well defined as functions but they are not necessarily in L2 pRq.
In this case f is not in H 1 pRq. On the other hand, a function can be in H 1 pRq even if it is
not of class C 1 .
Example 5.3. — For x P R we set
#
1 ´ |x| if |x| ď 1,
f pxq “ (5.2)
0 if |x| ą 1.

In the sense of distributions we have

$
&1
’ if x Ps ´ 1, 0r,
f pxq “ ´1 if x Ps0, 1r,
1
’
%
0 if |x| ą 1.

Thus f and f 1 are in L2 pRq, so f P H 1 pRq.

— Let H be the Heaviside function defined by (4.11). In the sense of distributions we
have H 1 “ δ. But δ is not the distribution given by a L2 function on R (see Proposition
4.19), so H is not in H 1 pRq.
The above definition can be extended to L2 functions in any dimension d P N˚ and we
can consider any order k P N˚ of derivatives.

93
M1 ESR - Distributions - Fourier

Définition 5.4. For k P N we set

� (
H k pRd q “ u P L2 pRd q : B α u P L2 pRd q for all α P Nd such that |α| ď k ,

where B α u is the derivative of u in the sense of distributions. In other words, a function

u P L2 pRd q belongs to H k pRd q if for all α P Nd with |α| ď k there exists vα P L2 pRd q such
that ż ż
@φ P C08 pRd q, p´1q|α| u B α φ dx “ vα φ dx.
Rd Rd

In this case vα is unique (up to equality almost everywhere) and we set B α u “ vα .

Remark 5.5. By the Riesz Theorem and by density of C08 pRd q in L2 pRd q, a function u P
L1loc pRd q belongs to H k pRd q is and only if for all α P Nd with |α| ď k there exists Cα ą 0
such that ˇż ˇ
8
ˇ ˇ
d
@φ P C0 pR q, ˇ ˇ uB φ dxˇˇ ď Cα }φ}L2 pRd q .
α
d R

Example 5.6. The function f defined by (5.2) is in H 1 pRq but not in H 2 pRq.
Example 5.7. Let α Ps´8, d´1r. We have seen in Paragraph 4.3.3 that the derivatives in the
´α ´α´2
sense of distributions of f : x ÞÑ |x| are given by ∇f pxq “ ´α |x| x. Let χ P C08 pRd q.
1
Then if α ă 2 ´ 1 we have χf P H pR q with
d d

∇pχf q “ f ∇χ ` χ∇f P L2 pRd q.

More generally, we can check that if α ă d

2 ´ k for some k P N then we have χf P H k pRd q.
We can similarly define the Sobolev spaces H k pΩq on any open subset Ω of Rd . We also
define the Sobolev spaces W k,p pΩq of Lp functions on Ω with all derivatives up to order k
in Lp , but we do not discuss these issues in this brief introduction. The discussion of the
following paragraph is only valid when p “ 2 and Ω “ Rd .

5.1.2 Characterisation via the Fourier transform

We can use the Fourier transform to give a simple characterisation of H k pRd q. We begin
with the following lemma.
Lemma 5.8. Let k P N. There exist C1 , C2 ą 0 such that
` 2 ˘k
ÿ ` 2 ˘k
@ξ P Rd , C1 1 ` |ξ| ď ξ 2α ď C2 1 ` |ξ| .
|α|ďk

Proof. Let ξ P Rd . For j P �0, k� we have

2j
ÿ
|ξ| “ pξ12 ` ¨ ¨ ¨ ` ξd2 qj “ ξi21 . . . ξi2j ď j d sup ξ 2α ,
1ďi1 ,...,ij ďd |α|ďk

so ˜ ¸
2 k
ÿ
k
2j
ÿ
k ÿ
p1 ` |ξ| q “ Ckj |ξ| ď Ckj j d ξ 2α .
j“0 j“0 |α|ďk

The first inequality follows with some C1 ą 0 independant of ξ. Now for α P Nd with |α| ď k
we have `
2|α| 2 ˘k
ξ 2α ď |ξ| ď 1 ` |ξ| ,
which gives the second inequality.

A function u belongs to H k pRd q if its derivatives are in L2 pRd q. After a Fourier transform,
this conditions turns into a condition about û multiplied by some polynomial.

Proposition 5.9. Let k P N˚ and u P L2 pRd q. Then u P H k pRd q if and only if

ż
` 2 ˘k
1 ` |ξ| |ûpξq|2 dξ ă `8. (5.3)
Rd

94 J. Royer - Université Toulouse 3

 A Brief Introduction to Sobolev spaces and applications

Proof. We have û P L2 pRd q. Let α P Nd with |α| ď k. If we identify functions with the
corresponding distributions we have by Proposition 4.94
FpB α uq “ piξqα û.
Thus B α u belongs to L2 pRd q if and only if the map ξ ÞÑ ξ α ûpξq does. Then u P H k pRd q if
and only if ż ÿ 2
ξ 2α |ûpξq| dξ ă `8.
Rd |α|ďk

By Lemma 5.8, this is equivalent to (5.3).

Remark 5.10. If u P L2 pRd q is such that Δu belongs to L2 pRd q, then u belongs to H 2 pRd q.
Exercise 5.11. Let u P L2 pRd q such that ΔpΔuq ` 2Δu ´ u P L2 pRd q. Prove that u P H 4 pRd q.

5.1.3 Regularity of functions in Sobolev spaces

It is not clear that being in some Sobolev space is a regularity property for a function
u. However, if u has enough weak derivatives in L2 , we recover some regularity in the usual
sense.
Proposition 5.12. Let k ą d2 and u P H k pRd q. Then u is continous and goes to 0 at infinity
(in the sense that u has a representative which satisfies these properties). In particular it is
bounded. More generaly, if k ą n ` d2 for some n P N then u is of class C n .
Proof. ‚ By the Cauchy-Schwarz inequality we have
ż ˆż ˙ 12 ´ ¯ 12
2 ` 2 ˘´k ` 2 ˘k 2
|ûpξq| dξ ď 1 ` |ξ| dξ 1 ` |ξ| |ûpξq| dξ ă `8,
Rd Rd

so û P L1 pRd q. This implies that u is continuous and goes to 0 at infinity. If k ą n ` d2 then

for all α P Nd with |α| ď n we have B α u P H k´n pRd q, so B α u is a continuous function. This
implies that u is of class C n .

5.2 Topology on the Sobolev spaces

In this section we define the norms on the Sobolev spaces we have just defined, and we
give some properties of these new functional spaces.

5.2.1 Hilbert structure

Theorem 5.1. Let k P N. For u, v P H k pRd q we set
ÿ
�u, v�H k pRd q “ �B α u, B α v�L2 pRd q . (5.4)
|α|ďk

This defines an inner product for which H k pRd q is a Hilbert space. Moreover, the correspon-
ding norm is equivalent to the norm defined by
ż
2 ` 2 ˘k 2
}u} “ 1 ` |ξ| |ûpξq| dξ. (5.5)
Rd

Proof. ‚ The fact that (5.4) defines an inner product on H k pRd q is left as an exercice. We
prove that H k pRd q is complete for the corresponding norm, given by
2
ÿ 2
}u}H k pRd q “ }B α u}L2 pRd q .
|α|ďk

Let pun qnPN be a Cauchy sequence. Then the sequences pB α un qnPN for |α| ď k are Cauchy
sequences in L2 pRd q. Since L2 pRd q is complete, there exist vα P L2 pRd q for |α| ď k such that
B α un goes to vα . For |α| ď k and φ P C08 pRd q we have
ż ż ż ż
p´1q|α| v0 B α φ dx “ p´1q|α| lim un B α φ dx “ lim B α un φ dx “ vα φ dx.
Rd nÑ`8 Rd nÑ`8 Rd Rd

Année 2020-2021 95
M1 ESR - Distributions - Fourier

This proves that in the sense of distributions we have B α v0 “ vα . Then v0 P H k pRd q and

}un ´ v0 }H k pRd q ÝÝÝÝÝÑ 0.

nÑ`8

Thus the sequence pun qnPN has a limit H k pRd q. This proves that H k pRd q is complete.
‚ By the Parseval identity we have for u P H k pRd q
ÿ ÿ ż ÿ
2 2 2 2
}u}H k pRd q “ }B α u}L2 pRd q “ }ξ α û}L2 pRd q “ ξ 2α |ûpξq| dξ.
|α|ďk |α|ďk Rd |α|ďk

As in the proof of Proposition 5.9, we conclude that this norm is equivalent to (5.5) with
Lemma 5.8.
Remark 5.13. If u P H k pRd q for all k P N then we have u P C 8 pRd q.

5.2.2 Density of smooth functions

In this paragraph we prove the density of smooth functions in the Sobolev spaces.
Theorem 5.2. Let k P N. Then C08 pRd q is dense in H k pRd q.
Proof. Let u P H k pRd q and ε ą 0. Let χ P C08 pRd q be supported in Bp0, 2q and equal to 1 on
Bp0, 1q. For m P N˚ and x P Rd we set χm pxq “ χp m x
q. Let α P Nd with |α| ď k. Let m P N˚ .
By the Leibniz rule we have χm u P H pR q and
k d

ÿ ˆα˙
B α pχm uq ´ χm B α u “ B α´β χm B β u.
β
0ďβďα
β‰α

Since p1´χm q and B α´β χm vanish on Bp0, mq for all β, we have by the dominated convergence
theorem

}B α pχm uq ´ B α u}L2 pRd q

` ˘
“ pχm ´ 1qB α u ` B α pχm uq ´ χm B α u
ż ÿ ˆα˙ › ›
ż
ˇ β ˇ
ď }χm ´ 1}L8 pRd q
2
|B α upxq| dx ` ›B α´β χm › 8 d ˇB upxqˇ2 dx
β L pR q
|x|ěm 0ďβďα |x|ěm
β‰α
ÝÝÝÝÝÑ 0.
mÑ`8

Therefore there exists m P N˚ such that

ε
}u ´ χm u}H k pRd q ď .
2
We set v “ χm u. Now let pρn qnPN be an approximation of the identity with ρn P C08 pRd q for
all n P N. For n P N the function un “ pρn ˚ vq is smooth because ρn is, and it is compactly
supported because v and ρn are. Let α P Nd with |α| ď k. We have B α un “ pρn ˚ B α vq so

}B α un ´ B α v}L2 pRd q ÝÝÝÝÝÑ 0.

nÑ`8

Thus there exists n P N such that

ε
}un ´ v}H k pRd q ď .
2
The conclusion follows.
With this density result we can extend to H k pRd q many results known for functions of
class C k . We give for instance the Green formula in Rd .
Proposition 5.14 (Green Formula on Rd ). Let u, v P H 1 pRd q and j P �1, d�. We have
ż ż
Bj u v dx “ ´ u Bj v dx.
Rd Rd

96 J. Royer - Université Toulouse 3

 A Brief Introduction to Sobolev spaces and applications

Proof. Let pun qnPN and pvn qnPN be sequences in C08 pRd q which go to u and v in H 1 pRd q. An
integration by parts gives, for all n P N,
ż ż
Bj un vn dx “ ´ un Bj vn dx.
Rd Rd

Taking the limit n Ñ `8 gives the result.

5.3 Examples of applications for partial differential equa-

tions
5.3.1 The Helmholtz equation
In section 2.6 we have discussed on Rd the equation

´Δu ` u “ f,

where f P L2 pRd q. Using the Fourier transform we saw that for any f P SpRd q this problem
has a unique solution u P SpRd q and that the map R : f ÞÑ u extends to a continuous map
on L2 pRd q.
The Sobolev spaces are the relevent context to discuss this kind of equation. We first
observe that the operator p´Δ ` Idq defines a continuous map from H 2 pRd q to L2 pRd q. In
fact, it defines a bijection with continuous inverse.
Proposition 5.15. Let f P L2 pRd q. Then there exists a unique u P H 2 pRd q such that

´Δu ` u “ f

in the sense of distributions, and there exists C ą 0 independant of f such that

}u}H 2 pRd q ď C }f }L2 pRd q .

If moreover f P H k pRd q for some k P N then u P H k`2 pRd q.

Proof. If we consider on H 2 pRd q the norm given by (5.5) we see that for all u P H 2 pRd q we
have
}p´Δ ` Idqu}L2 pRd q “ }u}H 2 pRd q ,
so p´Δ ` Idq defines a continuous and injective map from H 2 pRd q to L2 pRd q. Moreover,
its inverse, defined on Ranp´Δ ` Idq, is continuous. It remains to prove that p´Δ ` Idq
is surjective (or equivalently that Ranp´Δ ` Idq “ L2 pRd q). For this we use the Fourier
transform as in Section 2.6. Given f P L2 pRd q we consider the function u P L2 pRd q such that

fˆpξq
ûpξq “ .
ξ2 ` 1

Then u P H 2 pRd q by Proposition 5.9, and it belongs to H k`2 pRd q if f P H k pRd q. Taking the
inverse Fourier transform in the equality pξ 2 ` 1qûpξq “ fˆpξq we obtain p´Δ ` Idqu “ f . This
completes the proof.

5.3.2 The Heat equation

Given u0 P L2 pRd q we consider the heat equation
duptq
@t ą 0, “ Δuptq (5.6)
dt
with the initial condition
up0q “ u0 . (5.7)
It was already discussed in Paragraph 4.5.4 with the help of the convolution product.
Here we use the Fourier transform to give a solution of the heat equation. We begin with a
result of uniqueness.

Année 2020-2021 97
M1 ESR - Distributions - Fourier

Proposition 5.16. Let u P C 0 pR` , H 2 pRd qq X C 1 pR˚` , L2 pRd qq be a solution of (5.6)-(5.7)

with u0 “ 0. Then uptq “ 0 for all t ě 0.
2
Proof. The map t ÞÑ }uptq}L2 pRd q takes non-negative values, it is continuous on r0, `8r and
it is differentiable on s0, `8r. For t ą 0 we have by the Green formula
d 2 2
}uptq}L2 pRd q “ 2 Re �uptq, Δuptq�L2 pRd q “ ´2 }∇uptq}L2 pRd q ď 0.
dt
2 2
Since }up0q}L2 pRd q “ 0, we deduce that }uptq}L2 pRd q “ 0 for all t ą 0.

Proposition 5.17. Let k P N and u0 P H k pRd q. For t ě 0 we consider the unique uptq P
L2 pRd q such that for all t ě 0 and ξ P R we have
2
ûpt, ξq “ e´tξ u
x0 pξq.
Then u is continuous from r0, `8r to H k pRd q and differentiable from s0, `8r to H N pRd q
for any N P N, and it solves the heat equation (5.6)-(5.7).
Proof. For t0 ě 0 and t ě 0 we have
ż
2 ˇ ´t|ξ|2 2 ˇ2 ` 2 ˘k 2
}uptq ´ upt0 q}H k pRd q “ ˇe ´ e´t0 |ξ| ˇ 1 ` |ξ| |x
u0 pξq| dξ.
Rd

For all t ě 0 and ξ P Rd we have

ˇ ´t|ξ|2 2 ˇ2 ` 2 ˘k 2 ` 2 ˘k 2
ˇe ´ e´t0 |ξ| ˇ 1 ` |ξ| u0 pξq| ď 1 ` |ξ|
|x |x
u0 pξq| ,
and the right-hand side defines an integrable function on R d , so by the dominated convergence
theorem we have
2
}uptq ´ upt0 q}H k pRd q ÝÝÝÑ 0.
tÑt0

Now let N P N and t0 ą 0. For t ą 0, t ‰ t0 , we have

› ›2 ż ˇ ˇ2
˘N ˇˇ e´t|ξ| ´ e´t0 |ξ| ˇ
2 2
› uptq ´ upt0 q › ` 2 2 2ˇ 2
›
› t ´ t0 ´ Δupt 0 q›
› N d “ 1 ` |ξ| ˇ ` |ξ| e ´t0 |ξ|
ˇ |ûpξq| dξ.
H pR q Rd ˇ t ´ t 0 ˇ

If |t ´ t0 | ď t0
2 and ξ P R we have
ˇ ˇ2
ˇ e´t|ξ|2 ´ e´t0 |ξ|2 ˇ t0 `
2 Nˇ 2 ´t0 |ξ|2 ˇ 2 2 ˘N 4 t0 |ξ|2
2
p1 ` ξ q ˇ ` |ξ| e ˇ |ûpξq| ď 1 ` |ξ| |ξ| e´ 2 |ûpξq| .
ˇ t ´ t0 ˇ 4

Then we can apply the dominated convergence theorem, and we obtain that the map t ÞÑ
uptq P H N pRd q is differentiable at t0 with
u1 pt0 q “ Δupt0 q.
The conclusion follows.

5.3.3 The Wave equation

Let c ą 0. We consider on Rd the wave equation
Bt2 u ´ c2 Δu “ 0, (5.8)
with initial conditions
up0q “ u0 , Bt up0q “ u1 , (5.9)
2 1
for some u0 P H pR q and u1 P H pR q.
d d

We recall that for u0 P C 2 pRq and u1 P C 1 pRq the problem (5.8)-(5.9) has a unique
solution u P C 2 pR2 q, given by
ż
u0 px ` ctq ` u0 px ´ ctq 1 x`ct
upt, xq “ ` u1 psq ds.
2 2c x´ct

98 J. Royer - Université Toulouse 3

 A Brief Introduction to Sobolev spaces and applications

Theorem 5.3. Let u0 P H 2 pRd q and u1 P H 1 pRd q. Then the problem (5.8)-(5.9) has a unique
solution u P C 0 pR, H 2 pRd qq X C 1 pR, H 1 pRd qq X C 2 pR, L2 pRd qq.
Assume that u is a solution. Taking the Fourier transform û of u with respect to x we
obtain for t P R and ξ P Rd
B2
ûpt, ξq ` c2 ξ 2 ûpt, ξq “ 0.
Bt2
x0 pξq and Bt ûp0, ξq “ u
Moreover ûp0, ξq “ u x1 pξq. For each ξ P Rd we solve this second order
equation with respect to t. This gives

ûpt, ξq “ cospct |ξ|qx

u0 pξq ` t sincpct |ξ|qx
u1 pξq, (5.10)

where, for θ P R, #
sinpθq
if θ ‰ 0,
sincpθq “ θ
1 if θ “ 0.
Conversely, for all t P R we define uptq as the inverse Fourier transform of (5.10) with respect
to ξ. Then we check that u is indeed a solution of (5.8)-(5.9).

The uniqueness is given by the linearity of the problem and the following result about
the conservation of the energy.
Proposition 5.18. Let u P C 0 pR, H 2 pRd qq X C 1 pR, H 1 pRd qq X C 2 pR, L2 pRd qq be a solution
of (5.8). For t P R we set
2 2
Eptq “ }Bt uptq}L2 pRd q ` }∇uptq}L2 pRd q .

Then for all t P R we have Eptq “ Ep0q.

Année 2020-2021 99