You are on page 1of 13

MATHEMATICS fOR PHYSICAL SCIENCES: Fourier

TRANSfORMATION-II

Pr BOUETOU BOUETOU THOMAS


UNIVERSITY Of YAOUNDé I
summary

0.1 the Fourier integral . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2


0.2 Equivalent forms of statements of the Fourier integral theorem . . . . . . . . . . . . . . . . . . . . . 2
0.3 Fourier transformation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
0.3.1 Fourier transform in cosine and sine. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
0.4 Parseval identity for Fourier integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
0.5 Convolution for Fourier integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
0.6 Properties of Fourier transforms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
0.7 Example of a solution using the Fourier transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
0.8 Separation of variables (Fourier method) method for the wave equation. . . . . . . . . . . . . . . . . 5
0.8.1 Problem at the limits of the vibrating wire equation . . . . . . . . . . . . . . . . . . . . . . . 5
0.9 Method of separation of variables for the resolution of the heat propagation equation. . . . . . . . . 8
0.10 Transformation de Fourier en dimension n. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
0.11 Reminder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
0.11.1 Euclidean vector space : . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
0.11.2 Hilbert Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
0.11.3 Normal space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

1
In the studies on transformations in Fourier trigonometric series, the theory and application of the development
of a function f (x) of period 2l in Fourier series is known. Now the question is, what happens as l tends to infinity

0.1 the Fourier integral


In what follows, we will show that the Fourier series becomes the Fourier integral.
Theorem 1. Suppose that f satisfies the following conditions :
1 f (x) and f 0 (x)
R ∞ are piecewise(path by path) continuous in any finite range.
2 the integral −∞ f (x) dx converges, i.e. is absolutely integrable in ] − ∞, ∞[ or else f (x) ∈ L1 ,
then Z ∞
f (x) = {A(α) cos αx + B(α) sin αx} dx (1)
0
where Z ∞ Z ∞
1 1
A(α) = f (x) cos αx dx, B(α) = f (x) sin αx dx (2)
π −∞ π −∞

f (x+0)+f (x−0)
The result (Eq. 1) is true if x is a point of continuity of f (x). Otherwise, we have to replace f (x) by 2
as we did in the case of Fourier series

Remark 1. The above conditions are sufficient but not necessary. The similarity between (Eq. 1) and (Eq. 2) and
the corresponding results in the case of Fourier series is obvious and the right side of (Eq. 1) is sometimes called
the Fourier integral expansion of f (x).

0.2 Equivalent forms of statements of the Fourier integral theorem


Theorem 2. The Fourier integral theorem can be stated as follows :

1 ∞
Z Z ∞
f (x) = f (u) cos α(x − u) du dα (3)
π α=0 u=−∞
Z ∞Z ∞
1
f (x) = f (u)eiα(x−u) du dα (4)
2π −∞ −∞
Z ∞ Z ∞
1
f (x) = e iαx
dα f (u)e−iαu du (5)
2π −∞ −∞

where we imply that when f (x) is not continuous at the point x we replace f (x) by f (x+0)+f
2
(x−0)
.
These results can be simplified if f (x) is an even or odd function and it comes
Z ∞ Z ∞
1
f (x) = sin αx dx f (u) sin αu dx si f est odd (6)
2π −∞ −∞
Z ∞ Z ∞
1
f (x) = cos αx dx f (u) cos αu dx si f est even (7)
2π −∞ −∞

0.3 Fourier transformation


R∞
The integral F(α) = −∞ f (u)e−iαu du is called Fourier transform or Fourier transformation and the integral
1 ∞
F(α)eiαu dα is called inverse Fourier transform or inverse Fourier transformation.
R
f (u) = 2π −∞

2
0.3.1 Fourier transform in cosine and sine.
If f (x) is an odd function, then the derivative transform
Z ∞
Fs (α) = f (x) sin αx dx is called the Fourier transform in sine and (8)
0

Z ∞
f (x) = Fs (α) sin αx dα is called the inverse Fourier transform sine of Fs (α) (9)
0

If f (x) is an even function, then the transform is as follows :


Z ∞
Fc (α) = f (u) cos αu du is called the Fourier transform in cosine and (10)
0

Z ∞
f (x) = Fc (α) sin αu dα is called the inverse Fourier transform sine Fc (α) (11)
0

0.4 Parseval identity for Fourier integrals


If F(α) and G(α) are the Fourier transformations of the functions f (x) and g(x) then the integral :
Z ∞ Z ∞
1 ¯ dα
f (x)g(x) dx = F(α)G(α) (12)
∞ 2π ∞

makes sense. The bar indicates the complex conjugate term


If f (x) = g(x) ⇒ F(α) = G(α), and the integral becomes
Z ∞ Z ∞
2 1
|f (x)| dx = |F(α)|2 dα. (13)
−∞ 2π −∞
These two identities are called PARSEVAL identities for Fourier integrals.
Exercise
Establish the PARSEVAL identities for the cases of Fourier transformations in sine and cosine.

0.5 Convolution for Fourier integrals


The convolution of two functions f and g is defined by :
Z ∞
f ∗g = f (u)g(x − u) du (14)
−∞

We show that

F(f ∗ g) = F(f ).F(g), f ∗ g = g ∗ f, f ∗ (g ∗ h) = (f ∗ g) ∗ h, f ∗ (g + h) = f ∗ g + g ∗ h. (15)

Example Find the Fourier transform of



1 |x| < a
f (x) = (16)
0 |x| > a
∞ a
e−iαu eiαa − e−iαa
Z
−iαu sin αa
F(α) = f (u)e du = = =2 , α 6= 0. (17)
−∞ −iα −a iα a
If α ≈ 0 It comes that F(α) ≈ 2α.

3
0.6 Properties of Fourier transforms

∂v ∂2v ∂v ∂F(v)
(F( ) = iαF(v), F( 2 ) = −α2 F(v), F( ) = ( ), (18)
∂x ∂x ∂t ∂t
Z ∞ Z ∞
∂v ∂v −iαx ∞
F( ) = e dx = e−iαx v −∞ + iα ve−iαx dx, (19)
∂x −∞ ∂x −∞

∂2v
F( ) = −α2 F(α), (20)
∂x2
∂w ∂2w ∂w
v= , F( ) = iαF( ) = (iα)2 F(w). (21)
∂x ∂x ∂x
We can generalize
∂nv
 
F = (iα)n F(v).
∂xn
Z ∞ Z ∞
∂v ∂v −iαx ∂ ∂
F( )= e dx = ve−iαx dx = F(v) (22)
∂t −∞ ∂t ∂t −∞ ∂t

0.7 Example of a solution using the Fourier transform


1) Consider the heat propagation equation with initial conditions

∂u ∂2u
= k 2, u(x, 0) = f (x), |u(x, t)| < M, où − ∞ < x < ∞. (23)
∂t ∂x
2) Give the physical interpretation
Solution 1) By taking the transform part and others of this equality and using (Eq. 20) and (Eq. 22), it comes :
∂û
= −kα2 û. (24)
∂t
2
By solving this equation, it comes : û = Ce−kα t ,
2
û = F(u(x, t)) = Ce−kα t . (25)

for t = 0, on a : F(u(x, 0)) = C = F(f (x)).


Hence the solution 2
û = f (u(x, t)) = F(f (x))e−kα t . (26)
On the other hand
(r )
−kα2 t 1 −(x2 /4kt)
e =F e (27)
4πkt
(r )
1 −(x2 /4kt)
⇒ ū = F(f (x))F e (28)
4πkt
By applying the convolution theorem, it comes :
r
1
u(x, t) = f (x) ∗ e − (x2 /4kt) (29)
4πkt

Z r
1 −((x−u)2 /4kt)
⇒ u(x, t) = f (u) e du (30)
−∞ 4πkt

4
(x−u)2 x−u
If we operate a change of variable u = f (z) defined by 4kt = z 2 or √
2 kt
= z, the expression (Eq. 30) becomes

1
Z ∞
2 √
u(x, t) = √ e−z f (x − 2z kt) dz. (31)
π −∞

2)The physical interpretation of the problem comes down to determining the temperature of a thin and infinite
bar whose surface is isolated and the initial temperature is f (x).

0.8 Separation of variables (Fourier method) method for the wave


equation.
A popular method in solving boundary problems of partial differential equations is the Fourier method of
separating variables. We will approach this method using two examples.

0.8.1 Problem at the limits of the vibrating wire equation


Let the equation be given
∂2u ∂2u
= , x ∈ (0, l), t > 0, (32)
∂t2 ∂x2
with homogeneous limit data(Because the second part of the equality is equal to zero)

u(t, 0) = u(t, l) = 0, (33)

and the non-homogeneous initial conditions(Because the second part of the equality is not equal to zero)

∂u
u(0, x) = ϕ(x), = ψ(x), 0 ≤ x ≤ l. (34)
∂t t=0

Equation (Eq. 32) is linear and homogeneous, which is why the sum of the particular solutions will always be a
solution of this equation.
Let us look for a non-trivial solution, that is to say non-zero, written as the product of two functions :

u(x, t) = T (t)X(x) (35)


where T depends on t and X of x.
On the other hand, we will seek this solution such that it can verify the limiting condition (4.2). Let us substitute
the right part of (4.4) instead of u(t, x) in (Eq. 32) we have :

X 00 T = T 00 X, (36)

or after division by XT
X 00 (x) T 00 (t)
= . (37)
X(x) T (t)
The left part of this equality does not depend on t and the right part does not depend on x. Thus, we thus obtain
the equality of two functions of different variables :

X 00 (x) T 00 (t)
= = −λ. (38)
X(x) T (t)
(In the following we will show that this constant is real).
From the previous relation it comes that :

X 00 (x) + λX(x) = 0, (39)


T 00 (t) + λT (t) = 0. (40)

5
To define the functions X(x) and T (t), the boundary conditions (4.2) give

u(t, 0) = X(0)T (t) = 0,


u(t, l) = X(l)T (t) = 0.

It happens that the function X(x) should check the boundary conditions

X(0) = 0, X(l) = 0. (41)

If not, we had T (t) = 0 that is to say u(t, x) = 0 and we are looking for the non-trivial solution. So we arrived
at the problem of eigenvalues.
We need to look for the number λ called clean solution and we need to find that solution. The problem posed
is that of STURM-LIOUVILLE.
Here we will consider three cases :
λ < 0, λ = 0, λ > 0. (42)
1) If λ < 0, The problem does not admit a non-trivial solution ; indeed, the general solution is
√ √
−λx
X(x) = C1 e + C2 e− −λx
(43)

and the boundary conditions give √ √


−λl
C1 + C2 = 0, C1 e + C 2 e− −λl
(44)
i.e. √ √
−λl
C1 = −C2 , C1 ( −e− −λl
) = 0 =⇒ C1 = 0, C2 = 0, because eα − e−α 6= 0, ∀α ∈ R (45)
Thus,
X(x) ≡ 0, (46)
2) If λ > 0, the general solution contains imaginary exponents ; that’s why we can write it
√ √
X(x) = C1 cos λx + C2 sin λx (47)

√ the condition X(0) = 0, it is necessary that C1 = 0. Then the condition X(l) causes that C2 sin
To check λl = 0,
or sin λl = 0 (because
√ if C2 = 0, we get a √ trivial solution).
THe equation sin λl = 0 is verified if l λ = nπ i.e.
n2 π 2
λ= , avec n 6= 0, since we have supposed that λ 6= 0. (48)
l2
For the negative n, λ takes the same values as the positive n. This is why all the values λ for which equation
(Eq. 39) admits a non-trivial solution satisfying the limit condition (Eq. 41) gives the formula

n2 π 2
λn = , n = 1, 2, 3, ... (49)
l2
Thus we had found the eigenvalues posed for the problem. To these eigenvalues correspond the eigenfunctions

Xn (x) = C2 sin
x, (50)
l
seen by the equation (Eq. 39), the eigenfunctions are defined up to a positive constant on we can set equal to 1,
or choose it such that the eigenfunction Xn (x) satisfies an additional condition called condition of normality. It is
necessary to introduce the standard such as
Z l
Xn2 (x) dx = 1. (51)
0
This shows that r
2
C2 = . (52)
l

6
Indeed we have
l
1 − cos 2nπ
Z l Z l
l x
Z
nπ 2 l
Xn2 (x) dx = x dx = C22
sin dx = C22 = 1. (53)
0 0 l 0 2 2
In the following, in order not to write the standard, we set C2 = 1
Now, let’s deal with the equation posed at the start (Eq. 32) - (Eq. 34). Let us substitute in equation (Eq. 40)
(for the function T (t)) the value λ found. He comes :
n2 π 2
Tn00 + Tn = 0. (54)
l2
After resolution, we have :
nπ nπ
Tn (t) = An cos t + Bn sin t, (55)
l l
where An et Bn are the coefficients.
Let’s set
nπ nπ nπ
un (t, x) = Xn (x)T (t) = (An cos t + Bn sin t) sin x (56)
l l l
which verifies the equation (Eq. 32) and the boundary conditions in (Eq. 33) for all An and P∞Bn constants. By
virtue of the linearity and homogeneity of the equation, the sum of the particular solutions n=1 un (t, x) is also
the solution of this equation.
Let’s try to define the constants An , Bn such that the infinite series
∞ ∞
X X nπ nπ nπ
u(t, x) = un (t, x) = (An cos t + Bn sin t) sin x (57)
n=1 n=1
l l l
also verifies the equation (Eq. 32) and the boundary conditions (Eq. 33) and the initial conditions (Eq. 34), let’s
start with the initial conditions.

X nπ
u(0, x) = An sin x = ϕ(x) (58)
n=1
l
on the other hand, if we could differentiate the series (Eq. 57) member by member, we would have :

∂u(t, x) X nπ nπ
= Bn sin x = ψ(x). (59)
∂t
t=0 n=1
l l

Suppose that the functions ϕ and ψ can be uniformly and absolutely decomposed into convergent series in
sin nπ
l x in the segment [0, l]. In the theory of trigonometric series we know that this is only possible if the functions
ϕ(x) and ψ(x) are continuous together with their first derivatives and that ϕ(0) = ϕ(l) = 0, ψ(0) = ψ(l) = 0, that
is to say at the end of the segment, these functions become null. Suppose the previous conditions are true then we
will have :

2 l
Z
X nπ nπ
ϕ(x) = ϕn sin x, ϕn = ϕ(ζ) sin ζ dζ, (60)
n=1
l l 0 l

2 l
Z
X nπ nπ
ψ(x) = ψn sin x, ψn = ψ(ζ) sin ζ dζ. (61)
n=1
l l 0 l
The series for ϕ(x) and ψ(x) are absolutely and uniformly convergent. If we compare these series with the
formulas (Eq. 58), (Eq. 59) and (Eq. 57), we can differentiate in members by t, thus so that the initial conditions
are satisfied, we must set
l
An = ϕn , Bn = ψn . (62)

thus the solution of the equation (Eq. 32)-(Eq. 34) will be :

X nπ l nπ nπ
u(t, x) = (ϕn cos t+ ψn sin t) sin x. (63)
n=1
l nπ l l

7
0.9 Method of separation of variables for the resolution of the heat
propagation equation.
For the resolution of the problem of heat propagation through a limited bar, we can also use the method of
separation of variables that we have already studied in the case of the vibrating wire.
Let us find the solution of the heat propagation equation.

∂u ∂2u
= a2 2 , où u(x, t) ∈ Q = {x, t : 0 < x < l, 0 < t < T } (64)
∂t ∂x
with initial condition
u(0, x) = ϕ, (65)
and boundary condition
u(t, 0) = u(t, l) = 0 (66)
Using the variable separation method we get :

1 T 0 (t) X 00 (x)
u(t, x) = T (t)X(x), = = −λ.
a2 T (t) X(x)

X 00 (x) + λX(x) = 0, T 0 (t) + a2 λT (t) = 0. (67)


We get once again the eigenvalue problem X 00 + λX = 0 with X(0), X(l) = 0, which we had already solved.
 nπ 2 nπ
λ= , Xn (x) = sin x. (68)
l l
Values of λn correspond to the solutions of the second equation of (Eq. 67) :
2
Tn (t) = Cn e−a λn t
. (69)

Thus we obtain the particular solution of the heat propagation equation with the initial data of the form
2 nπ
un (t, x) = Cn e−a λn t
sin x. (70)
l
To check the initial condition, let’s establish the series :
∞ ∞
X X 2 nπ
u(t, x) = un (t, x) = Cn e−a λn t
sin x. (71)
n=1 n=1
l

From the initial condition, determine the coefficients Cn from the Fourier coefficients of the initial function ϕ(x).
We have :

X nπ
u(0, x) = Cn sin x, (72)
n=1
l
that is, Cn is the Fourier coefficient of the function ϕ by the Fourier series decomposition into sines.
Z l
2 nπ
Cn = ϕn = ϕ(ζ) sin ζ dζ. (73)
l 0 l

to complete the resolution of the problem we must show that the series in (Eq. 71) with Cn defined by the
formula (Eq. 73) converges uniformly with the second derivative with respect to x and to the first derivative with
respect to to t (this is so that the series (Eq. 71) verifies the initial equation (Eq. 64).

8
If ϕ is continuous and piecewise differentiable and satisfies the condition ϕ(0) = ϕ(l) = 0, then the series (Eq.
326) converges absolutely and uniformly in Q̄ = {x, t such that 0 ≤ x ≤ l and 0 ≤ t ≤ T } i.e. define a function
continuous in Q̄.
In the following, it remains to show that the series obtained can be derivative member by member, that is to
say that it verifies the heat propagation equation. Let us show that the series (Eq. 71) for t ≥ t0 > 0 can be derived
member by member with respect to t and x several times for any t0 > 0. In addition, all the series obtained are
convergent (uniform convergence) so in this case the properties of the solution of the heat propagation equation
differ from that of the vibrating wire. This quite simply because in the solution in S, the exponential is present and
raised to the negative power (for t ≥ t0 > 0). This is why the majoring series

X ∞
X   nπ 2 
αn = Mnq exp −a 2
t0 (74)
n=1 n=1
l

always converges ; what we can verify by the D’Alembert criterion


 q − π22 a2 (n2 +2n+1)t0  q
αn+1 n + 1 e l 1 π2 2
lim = lim 2 2 2 = lim 1 + e− l2 a (2n+1)t0 (75)
n→∞ αn n→∞ n π
e− l2 a n t0 n→∞ n

Thus, we have shown that if ϕ(x) ∈ C 1 (Q̄) où Q̄ = {x, t : 0 ≤ x ≤ l, t ≥ 0}, ϕ̄0 piecewise continue ϕ(0) = ϕ(l) = 0,
then there is a unique solution to the problem

∂u ∂2u
= a2 2 , (x, t) ∈ Q, u(0, x) = ϕ(x), u(t, 0) = u(t, l) = 0 (76)
∂t ∂x
continuous in the domain Q̄ and satisfying the given equation inside Q. This solution is presented as a series

nπ 2 2 nπ
C n e −( )
X
a t
u(t, x) = l sin x. (77)
n=1
l

The solution (Eq. 328) is infinitely differentiable with respect to t and x for t > 0.

0.10 Transformation de Fourier en dimension n.


We will denote by R2 , Euclidean space of dimension n at each point x we denote these Cartesian coordinates by
(x1 , x2 , x3 , ..., xn ). We will denote by Dm (R) which are n times differentiable and take the value 0 in a certain ball.
Let α = (α1 , ..., αn ) with components αi > 0, ∀i through Dα f (x) the derivative of f (x) such that |α| = α1 + ... + α2 ,

∂ |α| f (x1 , ..., xn )


Dα f (x) = (78)
∂xα α2
1 ∂x2 ...∂xn
1 αn

The Fourier transform of the function f (x) where x is taken in Rn is defined by


Z
fˆ(ζ) = e−i(x,ζ) f (x) dx (79)
Rn

with (x, ζ) = hx, ζi the scalar product.


The function fˆ(ζ) may not be absolutely convergent in Rn , which is why it is necessary to explain in which
sense we must understand the integral ( Eq. 79).

9
0.11 Reminder
0.11.1 Euclidean vector space :
Definition 1. :
Let V be a K − vectorspace. The real function denoted (x, y) or < x, y > or < x|y > for any couple (x, y)
belonging to V is called a scalar product if

(x, y) = (y, x) (80)


(x1 + x2 , y) = (x1 , y) + (x2 , y) ∀x1 , x2 , y ∈ V (81)
(λx, y) = λ(x, y) ∀x, y ∈ V and∀λ ∈ K (82)
(x, y) ≥ 0, si (x, x) = 0 =⇒ x = 0 (83)

Definition 2. :
Any vector space in which the scalar product is defined is called Euclidean vector space. In a Euclidean vector
space, we can define the norm or the distance associated with this space
p
||x|| = (x, x) (84)

d(x, y) = ||x − y||. (85)


In a Euclidean vector space, the Cauchy-Schwarz inequality is verified

|(x, y)| ≤ ||x||.||y||. (86)

In a Euclidean vector space, if we have two vectors x and y which satisfy the condition (x, y) = 0, then we say
that the vectors x and y are orthogonal.
If in a Euclidean vector space we have a vector system (xα )α∈N which is such that

(xα , xβ ) = 0 si α 6= β, (87)

then we say that the system (xα )α∈N is orthogonal


Exemple 1. :
We know that ((Rn , +, .)) is a R − vector − space on the field R. To show that this is a Euclidean vector space,
let’s construct the following scalar product :

x ∈ Rn =⇒ x = (x1 , ..., xn ) (88)

y = (y1 , ..., yn ) (89)


n
X
(x, y) = xi yi (90)
i=1

all that remains is to demonstrate


Pn the (4) properties.
In (Cn , +, .), (x, y) = i=1 xi y i .
In the sequence n X o
l2 = x = (x1 , x2 , ..., xn ) x2i < ∞ , (91)

we have ! 21
n
X
||x|| = x2i (92)
i=1

10
n
X
(x, y) = xi yi . (93)
i=1
In the sequence n X p o
lp = x = (x1 , x2 , ..., xn )/ xi < ∞ (94)
we have ! p1
n
X
||x||p = xpi (95)
i=1
The space C 2 ([a, b]) is the set of functions which are differentiable twice in the interval [a, b].
Z b
∀f, g ∈ C 2 ([a, b]), (f, g) = f (t)g(t) dt (96)
a
Exercise 1 : Show that  
1 2πt
, cos n (97)
2 b−a n∈N
forms an orthogonal system in C 2 ([a, b]).
exercise 2 : Do the same for  
1 2πt
, sin n (98)
2 b−a n∈N
exercise 3 : Do the same for  
2πt 2πt
cos n , sin n (99)
b−a b−a n∈N

0.11.2 Hilbert Space


Definition 3. We call Hilbert space any Euclidean vector space which is infinite and complete in the sense of the
metric defined in this space.
We say that any space is complete if any Cauchy sequence in this space converges.
Theorem 3. for a normal vector space V to be Euclidean, it is necessary and sufficient that for each two elements
f and g the following equality is satisfied :
||f + g||2 + ||f − g||2 = 2(||f ||2 + ||g||2 ) (100)
Exemple 2. Let’s onsider the space
n X p o
lp = x = (x1 , x2 , ..., xn )/ xi < ∞ (101)

n
! p1
X
||x||p = xpi . (102)
i=1
We see that with this norm thus defined, the space will not always be a Euclidean space. Let’s try to see for which
p it will be :
We set f (1, 1, 0, ..., 0) and g(1, −1, 0, ..., 0). Let’s try to calculate the norm of each vector.
1 1
||f ||p = 2 p , ||g||p = 2 p , ||f + g||p = 2, ||f − g||p = 2, (103)

||f + g||2p + ||f − g||2p = 2(||f ||2p + ||g||2p ) (104)


2 2
=⇒ 4 + 4 = 2(2 + 2 ) p p (105)
2
=⇒ 2 = 2 p (106)
=⇒ p = 2 (107)
Hence the space is Euclidean for p = 2.

11
0.11.3 Normal space
One of the important classes among normed spaces is the class of integrable functions in the sense of Lebesgue.
These are the spaces L1 and L2 . Let X be a measurable space with the measure µ. The space denoted by L1 (X, µ)
and which is made up of all the functions f which are absolutely integrable along X, is defined by
 Z 
L1 (X, µ) = f, |f | dµ < ∞ . (108)
X

The norm is Z
||f || = |f (µ)| dµ (109)
X
This space is infinite and complete.
The space L2 is the space denotedR L2 (X, µ) made up of all the functions which are therefore integrable by
quadrature following X if the integral f 2 (x) dµ < ∞ and exists.
The set Rof all these functions will be noted L2 (X, µ).
(f, g) = f (µ)g(µ) dµ is the scalar product ins L2 .
Definition 4. Banach space
Any full normal vector space is called Banach space.
Definition 5. Schwarz space

S = u(x), u(x) ⊂∞ (R), ∃ Cαβ : (1 + |x|β )|Dα u| ≤ Cαβ ∀x ∈ R, α, β > 0.



(110)

In other words, the space S consists of the functions u(x) ∈ C ∞ (R) which together decrease with its derivatives
faster than a function
1
∀β ∈ R. (111)
1 + |x|β
2
Exercise : Show that u(x) = e−x ∈ S.

12

You might also like