INSTITUTE OF MATHEMATICS POLISH ACADEMY OF SCIENCES
WARSZAWA 2000
AN ALGEBRAIC DERIVATIVE ASSOCIATED TO THE OPERATOR Dδ
V. M. A L M E I D A, N. C A S T R O and J. R O D R´ I G U E Z Departamento de An´ alisis Matem´ atico, Universidad de La Laguna
38271 La Laguna (Tenerife), Canary Islands, Spain E-mail: valmeida@ull.es
Abstract. In this paper we get an algebraic derivative relative to the convolution
(f ∗ g)(t) =
Z
t 0f (t − ψ)g(ψ)dψ
associated to the operator D
δ, which is used, together with the corresponding operational calcu- lus, to solve an integral-differential equation. Moreover we show a certain convolution property for the solution of that equation.
1. Introduction. W. Kierat and K. Sk´ ornik [2], using the Mikusi´ nski operational calculus, have solved the differential equation
t d
2x
dt
2+ (c − t) dx
dt − ax = 0 (c, a ∈ C)
which for c = 1 reduces to the Laguerre differential equation and one of its solutions is x
a(t) =
∞
X
k=0
−a k
(−1)
kt
kΓ(k + 1) satisfying the convolutional property
d
dt (x
a∗ x
b)(t) = x
a+b(t) where ∗ represents the Mikusi´ nski convolution.
We define the algebraic derivative Df (t) = −I
δ−1δ tf (t), for the convolution (f ∗ g)(t) = Z
t0
f (t − ψ)g(ψ)dψ where I
αf (t) =
Γ(α)1R
t0
(t − τ )
α−1f (τ )dτ represents the Riemann-Liouville fractional in- tegral operator.
2000 Mathematics Subject Classification: 44A40, 26A33, 33A20.
The paper is in final form and no version of it will be published elsewhere.
[71]
The convolution ∗ is defined on the set C
δ= n
f (t) =
∞
X
k=1
a
kt
kδ−1uniformly convergent on compact subsets of [0, ∞) o introduced by Alamo and Rodr´ıguez in [1].
Using a similar technique as in[2] and the appropriate operational calculus for ∗ we can get a solution of the following integral-diferential equation
−D(D
δ)
2x + (1 + D)D
δx − ax = 0 (a ∈ C) (δ > 1) which we denote by x
a(t), satisfying
D
δ[x
a∗ x
b](t) = x
a+b(t).
2. An operational calculus for D
δ. The algebraic derivative D. Let δ > 1 be a fixed real number (when δ = 1, it reduces to Kierat and Sk´ ornik’s case). As Alamo and Rodr´ıguez [1] did, we define the set of positive real variable functions with complex values
C
δ= n f (t) =
∞
X
k=1
a
kt
kδ−1uniformly convergent on compact subsets of [0, ∞) o . They proved that (C
δ, +, ·
C) is a vector space.
Unlike these authors, we will consider in C
δthe Mikusi´ nski convolution given by (f ∗ g)(t) = R
t0
f (t − ψ)g(ψ)dψ.
From the definition of ∗ we get immediately the following propositions.
Proposition 1. 1. t
kδ−1∗ t
mδ−1= B(kδ, mδ)t
(k+m)δ−1. 2. (f ∗ g)(t) = P
∞k=2
{ P
k−1j=1
a
jb
k−jB[jδ, (k − j)δ]}t
kδ−1. Here B(u, v) = R
10
(1 − t)
u−1t
v−1dt represents the beta function, f (t) = P
∞k=1
a
kt
kδ−1and g(t) = P
∞k=1
b
kt
kδ−1.
This proposition shows us that ∗ is a closed operation on C
δ, so we can conclude that (C
δ, +, ∗) is a subring of (C, +, ∗). Here C represents the set of continuous complex func- tions of a positive real variable. Mikusi´ nski [3] and Yosida [5] showed that the convolution
∗ has no zero divisors and there is no unit element on the set C, thus we can state the next proposition.
Proposition 2. (C
δ, +, ∗) is a commutative non-unitary ring without zero divisors.
Remark. It can be proved in a direct way that (C
δ, +, ∗) is a ring.
Therefore, C
δcan be extended to its field of fractions M
δ= C
δ×(C
δ−{0})/ ∼, where the equivalence relation ∼ is defined, as usual, by (f
1, g
1) ∼ (f
2, g
2) ⇔ f
1∗ g
2= g
1∗ f
2; actually M
δis a subfield of the Mikusi´ nski field. The elements of M
δwill be called operators, and from now on we denote by
fgthe equivalence class of the pair (f, g).
The operations of sum, multiplication and product by a scalar can be defined on M
δthrough
• f
1g
1+ f
2g
2= f
1∗ g
2+ g
1∗ f
2g
1∗ g
2• f
1g
1· f
2g
2= f
1∗ f
2g
1∗ g
2• α f g = αf
g
Alamo and Rodr´ıguez [1] showed that the operator D
δis an endomorphism on C
δand proved that for all f (t) = P
∞k=1
a
kt
kδ−1in C
δD
δI
δf (t) = f (t),
I
δD
δf (t) = f (t) − a
1t
δ−1= f (t) − [t
1−δf (t)]
t=0t
δ−1, (2.1) (I
δ)
m(D
δ)
mf (t) = f (t) −
m
X
j=1
a
jt
jδ−1. (2.2)
These identities will be useful for our development.
The next proposition allows us to identify the operator I
δand its positive integer powers with certain functions in C
δ.
Proposition 3. Let f (t) ∈ C
δand k ∈ N, then we have 1.
tΓ(δ)δ−1∗ f (t) = I
δf (t).
2.
tΓ(kδ)kδ−1∗ f (t) = (I
δ)
kf (t) = I
kδf (t).
Proof. The first asertion is a consequence of the definition of the convolution ∗, and using induction method we can get the second one.
Following Mikusi´ nski [3], we denote by l
δ=
tΓ(δ)δ−1≡ I
δ. So when we write l
δf (t) we will understand I
δf (t).
Now we remark that we can consider C
δ⊂ M
δsince C
δis isomorphic to a subring of M
δthrough the map f ;
lδlδf. In a similar way the field C of complex numbers can be embedded into M
δby associating with every α ∈ C the so called numerical operator [α] =
αttδ−1δ−1. The following basic properties of these numerical operators are immediate.
Proposition 4. 1. [α] + [β] = [α + β].
2. [α] · [β] = [αβ].
From now on we denote the numerical operators [α] by α when it leads to no confusion.
Proposition 5. Let v
δ∈ M
δbe the algebraic inverse of l
δ. For any function f (t) = P
∞k=1
a
kt
kδ−1,
v
δf (t) = D
δf (t) + Γ(δ)a
1(2.3) v
mδf (t) = (D
δ)
mf (t) +
m
X
j=1
a
jΓ(jδ)v
m−jδ(2.4)
Proof. To see (2.3), having the identity (2.1) we act on both sides by the operator v
δand take into account that a
1t
δ−1is identified with
lδa1ltδ−1δ
∈ M
δ, so v
δa
1t
δ−1=
a1tlδ−1δ
=
[Γ(δ)a
1] = Γ(δ)a
1. For (2.4) it is analogous, acting on both sides of (2.2) by v
δm.
The next step is to define an operator over C
δwhich will be an algebraic derivative.
Definition 1. Let f ∈ C
δ. We define the operator D as follows:
Df (t) = − I
δ−1δ tf (t).
We need to know how D acts on any member of C
δ. Proposition 6. Df (t) ∈ C
δfor all f (t) ∈ C
δ. Proof. It is not dificult to show that if f (t) = P
∞k=1
a
kt
kδ−1, then
− I
δ−1δ tf (t) =
∞
X
k=1
b
kt
(k+1)δ−1where b
k= −a
kΓ[(k+1)δ]kΓ(kδ). An equivalent and more manageable expression is
− I
δ−1δ tf (t) = (−t
δ−1) ∗ h X
∞k=1
c
kt
kδ−1i where c
k=
Γ(δ)kak.
Now we establish a proposition which shows that D is an algebraic derivative on C
δ. Proposition 7. For any functions f and g in C
δ, we have:
1. D[f (t) + g(t)] = Df (t) + Dg(t).
2. D(f ∗ g)(t) = ([Df ] ∗ g)(t) + (f ∗ [Dg])(t).
Proof. 1. It immediately follows by taking into account that
−Iδ−1
δ
t is a linear operator.
2. Let f (t) = P
∞k=1
a
kt
kδ−1and g(t) = P
∞k=1
b
kt
kδ−1, then we have:
D(f ∗ g)(t) = D
∞
X
k=2
n
k−1X
j=1
a
jb
k−jB[jδ, (k − j)δ] o t
kδ−1. If we denote c
k= P
k−1j=1
a
jb
k−jB[jδ, (k − j)δ], by using the result obtained in the proof of proposition 6 and the second identity of proposition 1, we can get
D(f ∗ g)(t) = (−t
δ−1) ∗
∞
X
k=2
k
Γ(δ) c
kt
kδ−1. In a similar way, it can be proved that
([Df ] ∗ g)(t) = (−t
δ−1) ∗
∞X
k=2
k−2X
j=1
j
Γ(δ) a
jb
k−jB[jδ, (k − j)δ]
t
kδ−1and
(f ∗ [Dg])(t) = (−t
δ−1) ∗
∞X
k=2
k−2X
j=1
(k − j)
Γ(δ) a
jb
k−jB[jδ, (k − j)δ]
t
kδ−1and using the last three identities the proof is concluded.
Now we can extend the definition of D to the field M
δ, as usual, by:
D f
g = [Df ] ∗ g − f ∗ [Dg]
g ∗ g (f ∈ C
δ, g ∈ (C
δ− {0})), D p
q = [Dp] · q − p · [Dq]
q
2(p ∈ M
δ, q ∈ (M
δ− {0})).
The next proposition shows the behavior of the algebraic derivative over some partic- ular members of M
δand will be used to solve an integral-differential equation.
Proposition 8. Let 1 =
tδ−1
tδ−1
the unit of M
δ, 0 =
tδ−10, v
δ=
l1δ
the algebraic inverse of l
δin M
δand n ∈ N. Then:
1. D1 = 0.
2. Dα = 0 (α being a numerical operator).
3. D(αp) = αDp (for any p ∈ M
δ).
4. Dl
δn= −nl
n+1δ. 5. Dv
δn= nv
δn−1.
6. D(1 − αl
δ)
n= nαl
2δ(1 − αl
δ)
n−1. 7. D(v
δ− α)
n= n(v
δ− α)
n−1.
Proof. (1) and (2) follow by a simple calculation. (3) is a direct consequence of (2).
In (4) we will use induction. Since in our case Dl
δ= D t
δ−1Γ(δ) = − t
2δ−1Γ(2δ) = −l
2δ, if we suppose that (4) is true for n = k, then
Dl
δk+1= D(l
δ· l
kδ) = [Dl
δ] · l
kδ+ l
δ· [Dl
δk] = −(k + 1)l
k+2δ. For (5) we consider the fact that v
δ=
l1δ
, so it is not difficult to see that Dv
δ= 1 using (1) and (4), afterwards we can use induction again. Finally, to get (6) and (7),
D(1 − αl
δ)
n= D
nX
k=1
n k
(−αl
δ)
n−k= −
n
X
k=1
n k
(−α)
n−k(n − k)l
n−k+1δ= −
n
X
k=1
n n − 1 k
(−α)l
2δ(−αl
δ)
n−k−1= nαl
δ2(1 − αl
δ)
n−1however (v
δ− α)
n=
(1−αllnδ)nδ
, using (6), (4) and the definition of D on M
δthe proof can be concluded.
Remark. The last proposition holds for n ∈ Z since p
−n=
p1nfor any p ∈ M
δ. The second identity of the last proposition tell us that the algebraic derivative of the numerical operators is zero, but furthermore we can establish the inverse result.
Proposition 9. Given p ∈ M
δ, if Dp = 0 then p is a numerical operator.
Proof. Let p =
fgand Dp = 0. Since
Dp = ([Df ] ∗ g)(t) − (f ∗ [Dg])(t) (g ∗ g)(t)
it follows that:
([Df ] ∗ g)(t) − (f ∗ [Dg])(t) = 0. (2.5) If we denote f (t) = P
∞k=1
a
kt
kδ−1and g(t) = P
∞k=1
b
kt
kδ−1, then we have ([Df ] ∗ g)(t) = (−t
δ−1) ∗
∞X
k=2
k−1X
j=1
j
Γ(δ) a
jb
k−jB(jδ, (k − j)δ)
t
kδ−1and
(f ∗ [Dg])(t) = (−t
δ−1) ∗
∞X
k=2
k−1X
j=1
(k − j)
Γ(δ) a
jb
k−jB(jδ, (k − j)δ)
t
kδ−1so, (2.5) implies that
k−1
X
j=1
(2j − k)a
jb
k−jB[jδ, (k − j)δ] = 0 (∀k ≥ 2). (2.6)
Now let us suppose b
16= 0. If we take in (2.6) k = 3 and k = 4 we can get respectively a
1b
2= a
2b
1and a
1b
3= a
3b
1;
next it is easy to prove that a
mb
n= a
nb
mwhen a
1b
n= a
nb
1and a
1b
m= a
mb
1.
Finally, in order to get that a
1b
k= a
kb
1for any k ≥ 2 we take into account the following identities
k−1
X
j=1
(2j − k)a
jb
k−jB[jδ, (k − j)δ]
=
r−1
X
j=1
(2j − 2r)(a
jb
2r−j− a
2r−jb
j)B[jδ, (2r − j)δ] (k = 2r),
k−1
X
j=1
(2j − k)a
jb
k−jB[jδ, (k − j)δ]
=
r
X
j=1
[2j − (2r + 1)](a
jb
2r+1−j− a
2r+1−jb
j)B[jδ, (2r + 1 − j)δ] (k = 2r + 1).
Therefore if b
16= 0 we can establish that a
k=
ab11
b
kfor any k ≥ 1, in other words f
g = αg
g = αt
δ−1t
δ−1= [α]
α = a
1b
1.
To conclude the proof we remark that, however b
1= 0 and a
16= 0 allow us to prove that b
k= 0 for any k in opposition to the fact that g(t) ∈ C
δ− {0}, b
1= 0 implies a
1= 0 so we can start with b
26= 0 and so on.
3. The use of D to solve an integral-differential equation. As an application of the results obtained in the preceding section, we will solve the integral-differential equation
−D(D
δ)
2x(t) + (1 + D)D
δx(t) − ax(t) = 0 (x(t) ∈ C
δ) (a ∈ C)
[t
1−δx(t)]
t=0= 0. (3.1)
Making use of (2.1), (2.2), (2.3), (2.4) and proposition 8, the equation (3.1) becomes Dx(t)
x(t) = a − 1 v
δ− a
v
δ− 1 = l
δ[l
δ(1 − a) − 1]
1 − l
δ. (3.2)
Several facts are immediately deduced from this expression.
Proposition 10. 1. x
a= l
δ(1 − l
δ)
−a∈ M
δis a solution of (3.2).
2. x
a(t) = t
δ−1Γ(a)
1Ψ
1
(a, 1);
t
δ(δ, δ);
is a solution of (3.1).
3. D
δZ
t0
(t − τ )
δ−1Γ(a)
1Ψ
1
(a, 1);
(t − τ )
δ(δ, δ);
τ
δ−1Γ(b)
1Ψ
1
(b, 1);
τ
δ(δ, δ);
dτ
= t
δ−1Γ(a + b)
1Ψ
1
(a + b, 1);
t
δ(δ, δ);
where
1Ψ
1represents the Wright generalized hypergeometric functions (cf. [4]).
Proof. 1. We have
D(1 − l
δ)
−a= D
1 +
∞
X
k=1
−a k
(−1)
kt
kδ−1Γ(kδ)
=
∞
X
k=1
(−a) −a − 1 k − 1
(−1)
k−1t
(k+1)δ−1Γ[(k + 1)δ] = (−a)l
2δ(1 − l
δ)
−a−1thus
D[l
δ(1 − l
δ)
−a]
l
δ(1 − l
δ)
−a= l
δ[l
δ(1 − a) − 1]
1 − l
δ.
2. The solution x
a= l
δ(1 − l
δ)
−aadmits a representation of the form (cf. [3, p. 171]) x
a= l
δ(1 − l
δ)
−a=
∞
X
k=0
−a k
(−1)
kl
δk+1= t
δ−1∞
X
k=0
(a)
kΓ(k + 1)
t
kδΓ[(k + 1)δ]
= t
δ−1Γ(a)
∞
X
k=0
Γ(k + a) Γ(kδ + δ)
t
kδΓ(k + 1) thus (cf. [4, p. 50]),
x
a(t) = t
δ−1Γ(a)
1Ψ
1
(a, 1);
t
δ(δ, δ);
. 3. It is consequence of the preceding items.
Remark. If −a ∈ N the series which appears in the proof of the last proposition becomes a polynomial of fractional degree.
References