Professional Documents
Culture Documents
Ma691 ch1 PDF
Ma691 ch1 PDF
The theory of distributions has been introduced by L. Schwartz (1915-2002) ca. 1944-1950 as ”ob-
jects” which generalize functions [Schwartz, 1945, Schwartz, 1966]. They extend the notion of derivative
to all integrable functions and are now widely used to formulate generalized solutions of partial differ-
ential equations. The Lebesgue Lp spaces contain non regular and non continuous functions for which
derivatives are not defined in the classical sense. Nonetheless, the classical derivatives generally exist
almost everywhere. It seems then reasonable to generalize the notion of derivative to be independent of
zero-measure subsets. This leads to the concept of weak derivative introduced by J. Leray (1906-1998)
and S.L. Sobolev (1908-1989). The main concept underlying the notion of weak derivative is the concept
of distributions.
Contents
1.1 Infinitely smooth functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2 The concept of distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
According to [Schwartz, 1966], O. Heaviside (1850-1925) introduced in 1894 a function denoted H and
called the unit step function. It corresponds to the characteristic function of the interval [0, +∞[ of R and
is a discontinuous function whose value is zero for negative argument and one for positive argument (cf.
Figure 1.1). This function has no derivative at the origin. Nonetheless, Heaviside defined the derivative
H ! of H at any point in R, and called it the unit impulse function, as:
!
! 0 x "= 0
H (x) = (1.1)
+∞ x=0
This function H cannot be considered as a usual function. Since H ! vanishes almost everywhere, it should
be assigned integral zero: " a
H ! (x) dx = 0 , ∀a ∈]0, +∞[ . (1.2)
−a
3
4 Chapter 1. Distributions on open sets of Rd
Furthermore, we have:
" a
H ! (x) dx = H(a) − H(−a) = 1 , ∀a ∈]0, +∞[ . (1.3)
−a
Actually, the function H ! has been reintroduced by P. Dirac (1902-1984) in 1926 and is widely used in the
context of quantum physics, where it is called the Dirac delta function, denoted δ. It has the following
property: " +∞
f (x)δ(x) dx = f (0) ,
−∞
for all continuous functions f . However, there is no function verifying this property.
The purpose of the distribution theory is to eradicate all these contradictions by giving a rigorous
definition of generalized functions, called distributions. Distributions have the remarkable property of
being all indefinitely differentiable (in a certain sense). Furthermore, an noteworthy property is that the
derivation is a continuous operation in the distribution space.
H(x − a) f (x)
1 h
a x 0 a−ε a a+ε x
Figure 1.1: The Heaviside step function (left) and the impulse function (right).
1
This presentation has been inspired by various lecture notes, in particular those of J.-M. Bony, Analyse, Ecole Polytech-
nique, Palaiseau, of M. Tucsnak, Distributions et équations fondamentales de la physique, Institut Elie Cartan, Université
Nancy I and of J.-Y. Chemin, Analyse réelle, UPMC, Paris.
1.1. Infinitely smooth functions 5
Proposition 1.1 Suppose I is an open interval of R and f is continuous in I and differentiable except
at point x0 ∈ I. If x ∈ I and lim f ! (x) = a ∈ R, then f ! (x0 ) exists and we have f ! (x0 ) = a.
x→x0
Definition 1.2 Consider an open subset Ω ⊂ Rd and k ∈ N. A function f ∈ C k (Ω) if one of the
following conditions holds:
(i) k = 0 and f is continuous on Ω,
∂f
(ii) k ≥ 1 and f has partial derivatives , j = 1, . . . , d that are in C k−1 (Ω).
∂xj
Moreover, f ∈ C ∞ (Ω) if f ∈ C k (Ω) for all k ∈ N.
Lemma 1.2 (Schwarz) If f ∈ C 2 (Ω) then
∂2f ∂2f
= , ∀i, j ∈ {1, . . . , d} .
∂xi ∂xj ∂xj ∂xi
Notice that if f ∈ C k (Ω, this lemma refers to the property of interchanging the order of taking partial
derivatives of a function. The matrix of second-order partial derivatives of f is called the Hessian matrix.
In most cases, this matrix is a symmetric matrix.
Theorem 1.1 (Taylor’s formula) Let Ω be an open subset of Rd , m ∈ N and f ∈ C m+1 (Ω). Consider
x, y ∈ Ω such that x + ty ∈ Ω, for all t ∈ [0, 1]. Then, we have:
# yα # yγ " 1
f (x + y) = ∂ f (x) + (m + 1)
α
(1 − t)m ∂ γ f (x + ty) dt . (1.4)
α! γ! 0
|α|≤m |γ|=m+1
Corollary 1.1 Let Ω be an open subset of Rd , f ∈ C m+1 (Ω) and K a compact subset of Ω. Consider
x, y ∈ K such that [x, x + y] ⊂ K. Then, there exists a constant C = C(K, m, f ) such that
$ $
$ $
$ # α
y α $
$f (x + y) − ∂ f (x)$$ ≤ C'y'm+1 . (1.5)
$ α!
$ |α|≤m $
Corollary 1.2 If f ∈ C k (B) where B = {x ∈ Rd , 'x' < 1}, then there exists f1 , . . . , fd ∈ C k−1 (B) such
that:
d
#
∂ α ∂j f (0)
∂ α fj (0) = , sup |∂ α fj | ≤ sup |∂ α ∂j f | , ∀j = 1, . . . , d f (x) − f (0) = xj fj (x) .
1 + |α| x∈B x∈B j=1
6 Chapter 1. Distributions on open sets of Rd
In this case,
( x Supp(f ) = [0, ∞[ and 0 ≤ f (x) ≤ 1, for all x ∈ R. We introduce the functions g(x) = f (x)f (1 − x) and
G(x) = 0 g(t) dt. We notice that 0 ≤ g(x) ≤ 1, for all x ∈ R and that Supp(g) ⊂ [0, 1] and g "≡ 0. The function
G(x)
θ= ∈ C ∞ (R) is an increasing function and is such that
G(1)
!
0 x≤0
θ(x) =
1 x≥1
Proposition 1.2 Consider four real numbers a, b, c and d such that a < c < d < b. Then, there exists
ρ ∈ D(R) such that:
Proposition 1.3 Consider a compact set K ⊂ Rd . Then, for every ε > 0, there exists u ∈ D(K2ε ) such
that: !
u(x) = 1 ∀x ∈ Kε
0 ≤ u(x) ≤ 1 ∀x ∈ Rd
p
) & '
5ε
B xj , ⊂ K2ε ,
j=1
3
we conclude that ũ ∈ D(K2ε ). Taking u = θ(ũ(x)) where θ ∈ C ∞ (R) is the function introduced in the previous
lemma, yields the result. !
8 Chapter 1. Distributions on open sets of Rd
Corollary 1.3 Let K1 , K2 be two disjoined compact subsets of the open set Ω ⊂ Rd . Then, there exists
a function u ∈ D(Ω) such that
!
1 ∀x ∈ K1
u(x) =
−1 ∀x ∈ K2
K1 ⊂ U1 , K2 ⊂ U2 , U1 ∩ U2 = ∅ .
From the previous proposition, we know that there exists u1 , u2 ∈ D(Ω) such that:
The function u defined as: u(x) = u1 (x) − u2 (x), ∀x ∈ Ω satisfies the desired properties. !
Proposition 1.4 Let K be a compact subset of Rd and let consider an open cover (Uj )j=1,...,N of K.
Then, there exists compacts sets (Kj )j=1,...,N such that Kj ⊂ Uj , for all j = 1, . . . , N and
N
)
K= Kj . (1.6)
j=1
% )
Proof. For every x ∈ K, consider rx > 0 such that B(x, rx ) ⊂ Uj . Hence, we have K ⊂ B(x, rx ) and
x∈Uj x∈K
M
)
thus there exists x1 , . . . , xM ∈ K such that K ⊂ B(xi , rxi ). We pose
i=1
)
Kj = K ∩ B(xi , rxi ) .
B(xi ,rxi )⊂Uj
Then by definition, Kj is a compact set included in K such that Kj ⊂ Uj . Let consider x ∈ K. There exists
i ∈ {1, . . . , M } such that x ∈ B(xi , rxi ). Moreover, there exists j0 ∈ {1, . . . , N } such that xi ∈ Uj0 , thus
B(xi , rxi ) ⊂ Uj0 . We conclude that x ∈ Kj0 . !
Theorem 1.3 (Partition of unity) Consider a compact subset K of Rd and open sets Uj of Rd . Sup-
N
)
pose we have K ⊂ Uj , then there exists a collection (uj )j=1,...,N such that uj ∈ D(Uj ), 0 ≤ uj ≤ 1 for
j=1
N
#
all j = 1, . . . , N and such that uj = 1 in the neighborhood of K.
j=1
1.1. Infinitely smooth functions 9
Proof. We know that there exists compact sets (Kj )j∈{1,...,N } such that Kj ⊂ Uj , for every j = 1, . . . , N and
N
)
such that K = Kj . Moreover, from the previous proposition, we deduce that for every j = 1, . . . , N , there
j=1
exists vj ∈ D(Uj ) such that vj (x) ∈ [0, 1], for all x ∈ Rd and vj (x) = 1 if x ∈ Kj . Consider the open set
N
) #N
V = x∈ Uj , vj (x) > 0 .
j=1 j=1
We have then K ⊂ V . Hence, there exists η ∈ D(V ) such that η(x) ∈ [0, 1] for x ∈ Rd and η ≡ 1 on an open set
W such that K ⊂ W ⊂ V . Considering the function
vj
ui = *N ,
(1 − η) + k=1 vk
we notice that ui ∈ D(Uj ) as the denominator is strictly positive on V and is equal to 1 outside V . Since η ≡ 1 on
N
#
the set W , the relation implies that ui ≡ 1 on W . !
j=1
Definition 1.5 Under the hypothesis of Theorem 1.3, the collection (uj )j=1,...,N is called a partition of
unity subordinate to the open cover (Uj )j=1,...,N of K.
is a Banach space. If p ≥ 1, we denote by Lploc (Ω) the space of locally integrable functions f such that
f ∈ Lp (K), for every compact K ∈ Ω.
Lemma 1.5 If f ∈ Lploc (Ω) then, there exists a largest open set V in Ω such that f |V = 0 almost
everywhere in V .
Definition 1.6 Consider Ω ⊂ Rd and f ∈ L1loc (Ω). We call essential support of f the closed set Ω\V
where V is the largest open V ⊂ Ω such that f |V = 0 almost everywhere in V .
Theorem 1.4 (Lebesgue dominated convergence) Let (fn )n∈N be a sequence of functions of L1 (Ω),
with Ω an open set of Rd . Suppose that
(ii) there exists a function g ∈ L1 (Ω) such that for every n, |fn (x)| ≤ g(x) almost everywhere in Ω.
(i) the function fx : Ω → R defined by fx (λ) = f (x, λ) belongs to C k (Ω) for every fixed x ∈ Rp ,
10 Chapter 1. Distributions on open sets of Rd
where wα ∈ L1 (Rp ).
"
Then, the function g(λ) = f (x, λ) dx is a function of C k (Ω) and for every λ ∈ R we have:
Rp
"
∂ α g(λ) = ∂λα f (x, λ) dx , ∀α ∈ Nn |α| ≤ k .
Rp
Consider Ω1 ⊂ Rn1 , Ω2 ⊂ Rn2 two open sets and a measurable function F : Ω1 × Ω2 → C. We have
two classical results.
"
Theorem 1.6 (Tonelli) Suppose that |F (x, y)| dy < ∞ for almost every x ∈ Ω1 and that
Ω2
" "
dx |F (x, y)| dy < ∞ .
Ω1 Ω2
Then F ∈ L1 (Ω1 × Ω2 ).
Theorem 1.7 " (Fubini) Suppose F ∈ L (Ω1 × Ω2 ). Then, we have F (x, y) ∈ L (Ω2 ) for almost every
1 1
x ∈ Ω1 and F (x, y) dy ∈ L1 (Ω1 ). Similarly, we have F (x, y) ∈ L1 (Ω1 ) for almost every y ∈ Ω2 and
" Ω 2
Theorem 1.8 (i) If p ≥ 1, then the space C00 (Ω) is dense in Lp (Ω).
Definition 1.7 We call distribution on Ω any linear form T on D(Ω) such that the following continuity
property holds: for every compact set K ⊂ Ω, there exists an integer N and a constant CK such that:
|)T, u+| ≤ CK sup '∂ α u'L∞ , ∀u ∈ DK (Ω) (1.8)
|α|≤N
Hence,
( we can show that f = 0 almost everywhere in Ω̃ and thus almost everywhere in Ω. Therefore,
Ω f (x)u(x) dx = 0 for every function u ∈ D(Ω). This is in contradiction with he assumption (1.9)
when u(a) "= 0.
The continuity property of a distribution can be characterized using sequences, as for the linear operators
in normed spaces. We introduce first the notion of convergence in D(Ω).
Definition 1.8 Let consider an open set Ω ⊂ Rd . The sequence (un )n∈N of elements of D(Ω) converges
toward u ∈ D(Ω) if
1. there exists a compact set K in Ω such that Supp (un ) ⊂ K, for all n ∈ N.
2. for every α ∈ Nn , we have lim '∂ α (un − u)'∞ = 0.
n→∞
Definition 1.9 Let consider a sequence (Tn )n∈N of elements in D! (Ω) and a distribution T on Ω. The
sequence (Tn )n∈N is said to converge toward T if and only if:
∀u ∈ D(Ω) , lim )Tn , u+ = )T, u+ .
n→∞
Theorem 1.9 A linear form T : D(Ω) → C is a distribution on Ω if and only if, for any sequence
(un )n∈N ⊂ D(Ω) tending toward 0 in D(Ω), we have:
lim )T, un + = 0 .
n→∞
12 Chapter 1. Distributions on open sets of Rd
Proof. Suppose T ∈ D$ (Ω) and consider (un )n∈N ⊂ D(Ω) tending toward 0 in D(Ω). From the previous definition
of the convergence in D(Ω), we know that there exists a compact subset K in Ω such that Supp (un ) ⊂ K for each
n ∈ N. Hence, there exist constants CK and NK such that
#
|)T, un +| ≤ CK '∂ α un '∞ , ∀n ∈ N .
|α|≤NK
Since limn→∞ '∂ un '∞ = 0 for each given α, we deduce the result for every sequence (un )n∈N ⊂ D(Ω) tending
α
Take uN = uN,N for every N ∈ N . It is obvious that uN "= 0, hence assume that the following equality holds
∗
1
sup '∂ α uN '∞ = .
|α|≤N N
|)T, uN +| > 1 , ∀N ∈ N .
lim uN = 0 , in D(Ω) ,
N →∞
Remark 1.1 We will assume now that any function f in L1loc (Ω) is identical to the distribution
"
u /→ f (x)u(x) dx .
Ω
When considering that the distribution T is a function, this will mean that there exists a function f ∈
L1loc (Ω) such that "
∀u ∈ D(Ω) , )T, u+ = f (x)u(x) dx .
Ω
The following proposition establishes that the Dirac mass is the limit of the approximations of identity.
Proposition 1.5 Consider a function χ in D(Rd ) of integral 1 and (εn )n∈N a sequence of real numbers
tending toward 0. We pose & '
def −d x
χεn (x) = εn χ .
εn
Then, the sequence (χεn )n∈N converges toward δ0 in the distributional sense, where δ0 is the linear form
defined by δ0 : D(Rd ) → C, u /→ u(0).
1.2. The concept of distribution 13
Remark 1.2 This result has a physical meaning. Indeed, if the functions εn are considered as mass
densities, the limit (in a certain sense) must be seen as a ponctual mass associated with the origin.
1
Definition 1.10 We call principal value of the function 1/x and we denote it pv the distribution
x
defined by: " ∞
1 1 u(x) − u(−x)
)pv , u+ = dx , ∀u ∈ D(R) .
x 2 −∞ x
Notice that this formula defines a distribution since we have, for every test function u ∈ D[−R,R] ,
$ $
$)pv 1 , u+$ ≤ R'u! 'L∞ .
$ $
$ x $
Proof. Since T is a distribution, for every compact K in Ω, there exists a constant C and an integer N such
that
∀v ∈ DK , |)T, v+| ≤ C sup '∂ α v'L∞ .
|α|≤N
Definition 1.12 Suppose T ∈ D! (Ω). Then, the distribution T (j) is called the partial derivative of T
with respect to the variable xj and is denoted ∂xj T and is such that:
Notice that if f is a differentiable function, the derivative of the distribution associated with f coincide
with the usual derivative of f .
Example 1.2 (i) The Heaviside step function H is the characteristic function of R+ . We have ∂x H =
δ0 in D! (R). Indeed, for any test function u ∈ D(R), we have:
6 7 " +∞
dH !
)∂x H, u+ = , u = −)H, u + = − u! (x)H(x) dx
dx −∞
" +∞
(1.11)
=− u! (x) dx = u(0) since u(+∞) = 0
0
= )δ0 , u+ .
Hence, the derivative of H in the sens of distributions is the Dirac mass defined hereabove.
(ii) Since the function f (x) = log |x| is in the space L1loc (R), it defines an element of D! (R). Indeed,
1
the derivative of f in the distributional sense is the distribution pv introduced previously:
x
df 1
= pv .
dx x
According to the definition of the derivative in D! (R), we have for all u ∈ D(R):
6 7 "
df
, u = − log(|x|)u! (x) dx .
dx R
We observe that:
" "
log(|x|)u! (x) dx = [u(−ε) − u(ε)] log ε − u! (x) dx ,
|x|>ε |x|>ε
The next results establishes a relation between the derivative and the primitive, in the distributional
sense.
Theorem 1.11 Consider a distribution T on an open interval I ⊂ R. If its derivative, in the distribu-
tional sense, T ! is a function of L1loc (I), then T is a continuous function and we have, for all a ∈ I:
" x
T (x) = T ! (x)dx + C .
a
1. the distributions on I for which the identity T ! = 0 holds, are the constant functions.
This implies the existence of a unique function v ∈ D(I) such that v $ = η. Hence, there exists a unique function
v ∈ D(I) such that &" '
v$ = u − u(x) dx θ . (1.12)
I
&" '
)T, u+ = u(x) dx )T, θ+ + )T, v $ + .
I
The term )T, v $ + = −)T $ , v+ is equal to zero and we obtain by denoting C the constant term )T, θ+:
"
)T, u+ = C u(x) dx .
I
Lemma 1.7 Consider T ∈ D! (Rd ) and suppose that ∂xj T = 0 for all i = 1, . . . , d. Then, T is a constant
(function).
16 Chapter 1. Distributions on open sets of Rd
Definition 1.13 The product of the distribution T ∈ D! (Ω) by the function f ∈ C ∞ (Ω) is the distribution
defined by:
)f T, u+ = )T, f u+ , ∀u ∈ D(Ω) .
Example 1.3 (i) For f ∈ C ∞ (Ω) and a ∈ Ω, we have f δa = f (a)δa . Indeed, we can observe that
In particular, xδ0 = 0.
(ii) We can find all solutions T ∈ D! (Ω) of the equation xT = 0. Let consider χ ∈ D(R) such that
χ(0) = 1. For every function u ∈ D(R), the function u − u(0)χ vanishes at the origin, hence there
exists a function v ∈ C ∞ (R) such that
u = u(0)χ + xv .
Since u and χ are functions with compact support, the relation means that v ∈ D(R). Thus, we
have
)T, u+ = u(0))T, χ+ + )T, xv+ .
By assumption, we have )T, xv+ = )xT, v+ = 0. Denoting C the constant )T, χ+, we obtain that
)T, u+ = Cu(0), hence T = Cδ0 .
∂f
∂xi (f T ) = T + f ∂xi T , ∀i ∈ {1, . . . , n} .
∂xi