Chapter 5 Henstock-Kurzweil Laplace Transform 5.1 Introduction The Laplace transform of a function f : [0, ∞) → R, at s ∈ C, is defined as the integral [15] Z ∞ 0 e -st f (t) dt. (5.1) Many authors, in some previous decades, showed their interest in studying the Laplace transform in a classical sense; by classical meaning Lebesgue integral and/or Riemann integral on the real line. In this chapter, we focus on an integral transform that generalizes the clas- sical Laplace transform. Here the generalization is given by the replacement, in the definition of Laplace transform, i.e., in (5.1), of Lebesgue integral with Henstock-Kurzweil (HK) integral. The Henstock-Kurzweil integral is defined in terms of Riemann sums but with slight modification, yet it includes the Riemann, improper Riemann, and Lebesgue integrals as special cases. This integral is equivalent to the Denjoy and Perron integrals. 117
35
Embed
Chapter 5 Henstock-Kurzweil Laplace Transformshodhganga.inflibnet.ac.in/bitstream/10603/25466/11/11_chapter_05.pdf · The rst question arises is that whether (or in which conditions)
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Chapter 5
Henstock-Kurzweil Laplace
Transform
5.1 Introduction
The Laplace transform of a function f : [0,∞) → R, at s ∈ C, is defined as
the integral [15] ∫ ∞0
e−stf(t) dt. (5.1)
Many authors, in some previous decades, showed their interest in studying
the Laplace transform in a classical sense; by classical meaning Lebesgue
integral and/or Riemann integral on the real line.
In this chapter, we focus on an integral transform that generalizes the clas-
sical Laplace transform. Here the generalization is given by the replacement,
in the definition of Laplace transform, i.e., in (5.1), of Lebesgue integral with
Henstock-Kurzweil (HK) integral. The Henstock-Kurzweil integral is defined
in terms of Riemann sums but with slight modification, yet it includes the
Riemann, improper Riemann, and Lebesgue integrals as special cases. This
integral is equivalent to the Denjoy and Perron integrals.
117
The first question arises is that whether (or in which conditions) the
Laplace transform when treated as HK integral make sense. The sufficient
condition for the existence of Laplace transform is that the function f is
locally integrable on [0,∞), because the product of locally integrable func-
tion and bounded measurable function is locally integrable. Since Henstock-
Kurzweil integral allows nonabsolute convergence, it makes an ideal setting
for the Laplace transform. It was proved [9] that, on a compact interval, the
product of nonabsolute integrable function and a bounded variation function
is still nonabsolute integrable and this remains valid in case of an unbounded
interval [1]. We observed in (5.1) that the kernel function e−st, s ∈ C, is
not bounded variation function on [0,∞) hence we cannot claim (as in the
classical sense) that the HK Laplace transform exists for any HK integrable
function.
The convolution plays an important role in pure and applied mathemat-
ics, approximation theory, differential and integral equations and many other
areas. Laplace transform has the property that it can interact with con-
volutions so we obtain various results on convolution. We give necessary
and sufficient conditions so that the convolution operator as HK integral is
continuous. Finally we solve the problem of inversion by using generalized
differentiation.
The results in this chapter are appeared in [2, 14].
5.2 Existential Conditions
In this section, we tackle the problem of existence of Laplace transform as
HK integral.
According to Zayed [16], the classical sufficient condition for the existence
118
of Laplace transform is that the function f is locally integrable on [0,∞), i.e.,
f ∈ Lloc[0,∞). This is because the multiplier for the Lebesgue integrable
functions is bounded measurable function and the function e−st, s ∈ C, is
bounded measurable function. We know that the multipliers for HK inte-
grable functions are precisely the functions of bounded variation [1]. Note
that in (5.1), the function e−st, s ∈ C, is not of bounded variation on [0,∞)
but it is of bounded variation on any compact interval J ⊂ R. Hence we
can not consider (5.1) as HK integral directly. So we cannot get any specific
condition for the existence of Laplace transform as HK integral. However we
do have the following different conditions.
Theorem 5.2.1. Let f : [0,∞) → R be any continuous function such that
F (x) =∫ x
0f , 0 ≤ x <∞, is bounded on [0,∞). Then the Laplace transform
L{f}(s), i.e., L{f}(s) =∫∞
0e−stf(t) dt, Re.s > 0 exists.
Proof. Let 0 < x <∞, s = σ + iω, σ, ω ∈ R.
We have (F (t) e−st)′= −s F (t) e−st + f(t) e−st.
Since e−st is of bounded variation on any compact interval, say [a, b] and F (t)
is bounded on [0,∞). Then F (t) e−st is HK integrable on [a, b].
Therefore −s F (t) e−st is integrable over I = [a, b].
Now ∫ b
a
∣∣∣(e−st)′∣∣∣ dt =
∫ b
a
∣∣−s e−st∣∣ dt=
∫ b
a
√σ2 + ω2 e−σt dt
=
√σ2 + ω2
σ[e−σa − e−σb].
Without loss of generality let us take a = 0. Therefore∫ b
0
∣∣∣(e−st)′∣∣∣ dt =
√σ2 + ω2
σ[1− e−σb] <∞.
119
Hence (e−st)′
is absolutely integrable over [a, b].
Now by Hake’s theorem [1], we can write
limb→∞
∫ b
0
∣∣∣(e−st)′∣∣∣ dt = limb→∞
√σ2 + ω2
σ[1− e−σb]
=
√σ2 + ω2
σ, σ > 0
=
√σ2 + ω2
σ, Re.s > 0.
This shows that (e−st)′
is absolutely integrable on [0,∞).
Again limt→∞
e−st = 0, Re.s > 0.
Hence the function e−st : [0,∞)→ R is continuous on [0,∞) with limt→∞
e−st =
0 and (e−st)′
is absolutely integrable over [0,∞) and we have assumed that
f : [0,∞)→ R is continuous function such that F (x) =∫ x
0f , 0 ≤ x <∞, is
bounded on [0,∞).
Therefore by Dedekind’s test [10], we can say that the integral∫ ∞0
e−stf(t) dt exists forRe.s > 0.
Remark 5.2.2. The above result can also be proved by using Du-Bois Rey-
mond’s test [1] as follows:
Take φ(t) = e−st, s = σ + iω, J = [0,∞).
Then clearly φ(t) is differentiable and φ′ ∈ L1(J).
Since F is bounded, we have limt→∞
F (t) e−st = 0, Re.S > 0.
Hence F (x) =∫ x
0f , 0 ≤ x <∞ is bounded on J , e−st is differentiable on J ,
(e−st)′ ∈ L1(J) and F (t) e−st → 0 as t→∞, Re.s > 0.
Therefore by Du-Bois Reymond’s test, f(t)e−st ∈ HK(J).
That is, the Laplace transform of f(t),∫∞
0e−stf(t) dt exists, Re.s > 0.
120
Theorem 5.2.3. If the Laplace transform of f(t),∫∞
0e−stf(t) dt exists for
Re.s > 0, then the function f ∈ HKloc.
Proof. Suppose the integral∫∞
0e−stf(t) dt exists for Re.s > 0.
Note that the function e−st, s ∈ C, as a function of real variable t, is not
of bounded variation on [0,∞) but is of bounded variation on any compact
interval I = [a, b].
By Hake’s theorem [1], we can write∫ ∞0
e−stf(t) dt = lima→0b→∞
∫ b
a
e−stf(t) dt.
Since the limits on R.H.S. exist, we can say that the HK integral∫ bae−stf(t)dt
exists, Re.s > 0 for some compact interval [a, b].
Hence the function f(t) is HK integrable necessarily on the compact interval
I = [a, b], that is, f ∈ HKloc.
Theorem 5.2.4. Let f ∈ HK([0,∞)). Define F (x) =∫∞xf(t) dt. Then
L{f}(s) exists if and only if∫∞
0e−stF (t) dt exists at s ∈ C.
Proof. Let T > 0. The function e−st, s ∈ C, is of bounded variation on a
compact interval [0, T ] and the function f is HK integrable on [0,∞).
Hence e−stf(t), s ∈ C, is HK integrable on [0, T ] and by integrating by parts,
we get ∫ T
0
e−stf(t) dt = e−sTF (T )− F (0) + s
∫ T
0
e−stF (t) dt.
Since the function f is HK integrable on [0,∞), F (0) is finite, say A and by
[9.12(a), [5]], F is continuous function such that limT→∞
F (T ) = 0.
Therefore by Hake’s theorem [1],
limT→∞
∫ T
0
e−stf(t) dt = limT→∞
{e−sTF (T )− F (0) + s
∫ T
0
e−stF (t) dt
}.
121
Hence ∫ ∞0
e−stf(t) dt = s
∫ ∞0
e−stF (t) dt− A.
This shows that L{f}(s) exists if and only if∫∞
0e−stF (t) dt exists at s ∈ C,
Re.s > 0.
Theorem 5.2.5. (Uniqueness theorem) Let L{f}(s) =∫∞
0e−stf(t) dt
and L{g}(s) =∫∞
0e−stg(t) dt. If f(t) = g(t) a.e. on [0,∞), then L{f}(s) =
L{g}(s).
5.3 Basic Properties
The usual elementary properties such as linearity, dilation, modulation, trans-
lation are remains same for the HK Laplace transform, the proofs of which
are quite similar to that of classical ones. Also there are some other re-
sults that are analogous to the classical one with some modification in the
hypothesis which we discuss below.
Theorem 5.3.1. (HK Laplace transform of derivative) Let f : [0,∞)→
R be a function which is in ACGδ(R) such that limt→∞
f(t)est
= 0. Then for Re.s >
0 both L{f}(s) and L{f ′}(s) fail to exist or L{f ′}(s) = s L{f}(s)− f(0).
Proof. Let T > 0. Consider the integral J =∫ T
0e−stf
′(t) dt.
This integral is valid since the function e−st, s ∈ C, is bounded variation on
any compact interval [0, T ] and the function f′
is HK integrable on R.
By integrating by parts, we get
J = e−sTf(T )− f(0) + s
∫ T
0
e−stf(t) dt.
By Hake’s theorem [1],
limT→∞
∫ T
0
e−stf′(t) dt = lim
T→∞e−sTf(T )− f(0) + lim
T→∞s
∫ T
0
e−stf(t) dt
122
= s
∫ ∞0
e−stf(t) dt− f(0).
Therefore ∫ ∞0
e−stf′(t) dt = s
∫ ∞0
e−stf(t) dt− f(0).
That is, L{f ′}(s) = sL{f}(s)− f(0).
Remark 5.3.2. The above result can be generalize by imposing the condi-
tions on f as:
For k = 0, 1, 2, . . . , (n− 1), f (k) are in ACGδ(R) and limt→∞
f (k)
est= 0, Re.s > 0.
Then L{f (n)}(s) and L{f}(s) fails to exist or
L{f (n)}(s) = sn L{f}(s)− sn−1 f(0)− sn−2 f′(0)− . . .− f (n−1)(0).
Theorem 5.3.3. (Differentiation of HK Laplace transform) Let f :
[0,∞) → R be a function such that L{f}(s) exists on a compact interval,
say J = [α, β], α, β ∈ R. Define g(t) = t f(t) and assume that g ∈ HK(R+).
Then L{g}(s) exists and L{g}(s) = −(L{f}(s))′ a.e. Here s ∈ R.
Proof. Suppose
L{f}(s) =
∫ ∞0
e−stf(t) dt exists on [α, β], α, β ∈ R.
Define g(t) = t f(t).
We know the necessary and sufficient condition for differentiating under the
HK integral is that∫ ∞t=0
∫ b
s=a
e−st tf(t) ds dt =
∫ b
s=a
∫ ∞t=0
e−st tf(t) dt ds (5.2)
for all [a, b] ⊂ [α, β] [12].
Now ∫ ∞t=0
∫ b
s=a
e−st tf(t) ds dt =
∫ ∞t=0
(e−at − e−bt)f(t) dt
123
= L{f}(a)− L{f}(b)
which exists.
Therefore the double HK integral∫∞t=0
∫ bs=a
e−st tf(t) ds dt exists.
Hence by [lemma 25(a), [13]], the equality in (5.2) holds and we can differ-
entiate under the HK integral. By (5.2), we have∫ b
s=a
∫ ∞t=0
e−st tf(t) dt ds = L{f}(a)− L{f}(b)
= −∫ b
a
(L{f}(s))′ds.
Therefore ∫ b
s=a
∫ ∞t=0
e−st tf(t) dt ds = −∫ b
a
(L{f}(s))′ds
for all [a, b] ⊂ [α, β]. Which implies that∫ ∞0
e−st tf(t) dt = −(L{f}(s))′ a.e. on [α, β].
That is, L{t f(t)}(s) = −(L{f}(s))′ a.e. on [α, β].
Hence the derivative of the HK Laplace transform of f(t) exists a.e. on
[α, β].
Remark 5.3.4. Let G(s) = L{t f(t)}(s) and H(s) = −L{f(t)}(s).
Then (H(s))′
= G(s) a.e. on [α, β]. And, we know that [5], A function
f : [a, b] → R is HK integrable on [a, b] if and only if there exists an ACGδfunction F on [a, b] such that F
′= f a.e. on [a, b].
Again, by fundamental theorem of calculus [10], (H(s))′
is HK integrable,
i.e., every derivative is HK integrable. Hence G(s) is also HK integrable.
Consequently, the function H(s) is ACGδ on [α, β].
Continuing in this way n-times, we can say that if tnf(t) is HK integrable,
then any order of derivative of L{f}(s) exists a.e. So L{f}(s) is analytic a.e.
124
Theorem 5.3.5. (Multiplication by 1t) Let f : [0,∞)→ R be a function
such that L{f}(s) exists and assume that f(t)t∈ HK(R+). Then L
{f(t)t
}(s)
exists and
L
{f(t)
t
}(s) =
∫ ∞s
L{f}(u) du.
Here s ∈ R.
Proof. Assume that f(t)t∈ HK(R+).
Since the function e−st is of bounded variation on any compact interval I =
[a, b], we have e−st f(t)t∈ HK([a, b]). That is,
∫ bae−st f(t)
tdt exists.
Without loss of generality, let us take a = 0 so that∫ b
0e−st f(t)
tdt exists.
By Hake’s theorem [1], limb→∞
∫ b0e−st f(t)
tdt exists.
Hence L{f(t)t
}(s) exists.
Now consider the integrals
I1 =
∫ ∞0
∫ ∞s
e−ut f(t) du dt, I2 =
∫ ∞s
∫ ∞0
e−ut f(t) dt du.
I1 =
∫ ∞0
[e−st − 0]f(t)
tdt
=
∫ ∞0
e−stf(t)
tdt which exists.
Therefore I1 exists on R× R.
Also, note that the function e−st is of bounded variation on any compact
interval J ⊂ R. Therefore∫R VJ(e−st) dt < MJ , for some constant MJ > 0.
Hence by [lemma 25(a), [13]], I2 exists on R× R and I1 = I2 on R× R. So∫ ∞0
e−stf(t)
tdt =
∫ ∞s
∫ ∞0
e−ut f(t) dt du
=
∫ ∞s
L{f}(u) du.
That is, L{f(t)t
}(s) =
∫∞s
L{f}(u) du.
125
Theorem 5.3.6. (HK Laplace transform of integral) Let f : [0,∞)→ R
be HK integrable function such that L{f}(s) exists. Then L{∫ t
0f(u) du
}(s)
exists and L{∫ t
0f(u) du
}(s) = 1
sL{f}(s).
Proof. Define F (t) =∫ t
0f(u) du, t ∈ R+. Then F
′(t) = f(t) a.e. on R+
[9.12(b), [5]].
Since f ∈ HK(R+), the integral∫ ∞0
f(u) du exists on R+.
That is,
limt→∞
∫ t
0
f(u) du exists on R+.
Hence limt→∞
F (t) is bounded on R+.
Now for T > 0, consider∫ T
0
e−st f(t) dt =
∫ T
0
e−st F′(t) dt.
This integral exists since e−st is of bounded variation on [0, T ] and the func-
tion f(t) is HK integrable on R+.
By integrating by parts, we get∫ T
0
e−st f(t) dt = e−sT F (T ) + s
∫ T
0
e−st F (t) dt.
Since limT→∞
F (T ) is bounded on R, set limT→∞
F (T ) = A.
By Hake’s theorem [1], we can write∫ ∞0
e−st f(t) dt = limT→∞
∫ T
0
e−st f(t) dt
= limT→∞
e−sT F (T ) + s limT→∞
∫ T
0
e−st F (t) dt
= s
∫ ∞0
e−st F (t) dt Re.s > 0.
126
Therefore ∫ ∞0
e−st F (t) dt =1
s
∫ ∞0
e−st f(t) dt, Re.s > 0.
That is, ∫ ∞0
e−st∫ t
0
f(u) du dt =1
sL{f}(s), Re.s > 0.
Hence L{∫ t
0f(u) du
}= 1
sL{f}(s), Re.s > 0.
5.4 Convolution
In this section we shall discuss some classical results of Laplace convolution
in the HK integral setting, called HK-convolution. First we give two results
about the existence of HK-convolution.
Theorem 5.4.1. Let f ∈ HK(R+) and g ∈ BV(R+). Then we have f ∗ g
exists on R+.
Proof. By definition [4],
f ∗ g(t) =
∫ t
0
f(t− τ) g(τ) dτ
which exists as HK-integral by multiplier theorem [1].
By integrating by parts, we get
|f ∗ g(t)| ≤ A‖f‖{
infR+|g|+ V[0,t] g(τ)
}.
Hence f ∗ g exists on R+.
Theorem 5.4.2. Let f ∈ HKloc and g ∈ BV such that the support of g is
in the compact interval I = [a, b]. Then
|f ∗ g(t)| ≤ A‖f‖[t−b,t−a]
{inf[a,b]|g|+ V[a,b] g(τ)
}.
127
Proof. Suppose supp(g) ⊂ [a, b]. Then we have
f ∗ g(t) =
∫ b
a
f(t− τ)g(τ) dτ.
Integrate by parts, we get
|f ∗ g(t)| ≤∣∣∣∣g(τ)
∫ b
a
f(t− τ) dτ
∣∣∣∣+
∣∣∣∣∫ b
a
∫ b
a
f(t− τ) dg(τ)
∣∣∣∣≤ inf
τ∈[a,b]|g(τ)| ·
∣∣∣∣∫ b
a
f(t− τ) dτ
∣∣∣∣+
∣∣∣∣∫ b
a
f(t− τ) dτ
∣∣∣∣ · V[a,b] g(τ)
≤{
infτ∈[a,b]
|g(τ)|+ V[a,b] g(τ)
}sup
[t−b,t−a]
∣∣∣∣∫ t−a
t−bf(u) du
∣∣∣∣ .Therefore
|f ∗ g(t)| ≤ A‖f‖[t−b,t−a]
{inf[a,b]|g|+ V[a,b] g(τ)
}.
The following result shows that the HK-convolution is bounded.
Theorem 5.4.3. Let f ∈ HK(R+) and g ∈ L1(R+) ∩ BV(R+). Then we
have A‖f ∗ g‖ ≤ A‖f‖ ‖g‖1.
Proof. For 0 ≤ a < b ≤ ∞, we have∫ b
a
f ∗ g(t) dt =
∫ b
a
∫ t
0
f(t− τ)g(τ) dτ dt.
Let
I1 =
∫ b
a
∫ t
0
f(t− τ)g(τ) dτ dt and I2 =
∫ t
0
∫ b
a
f(t− τ)g(τ) dt dτ.
Consider
I2 =
∫ t
0
∫ b
a
f(t− τ)g(τ) dt dτ.
Since f ∈ HK(R+), we have∫ baf(t− τ) dt exists and finite.
Let ∫ t−b
t−af(τ) dτ = M.
128
Therefore
|I2| ≤M
∫ t
0
|g(τ)| dτ.
Since g ∈ L1(R+), we have for some finite K,∫ t
0
|g(τ)| dτ = K.
So |I2| ≤MK and hence I2 exists on R+ × R+.
Now as g ∈ L1(R+), we have ‖g‖1 ≤ K1 and also∫R+ V[0,t] g ≤M1.
Therefore by [lemma 25, [13]], I1 = I2 on R+ × R+.
Hence ∣∣∣∣∫ b
a
f ∗ g(t) dt
∣∣∣∣ ≤ sup[t−a,t−b]
∣∣∣∣∫ t−b
t−af(u) du
∣∣∣∣ · ∫ t
0
|g(τ)| dτ
≤ A‖f‖[t−a,t−b] · ‖g‖1.
Therefore
sup[a,b]
∣∣∣∣∫ b
a
f ∗ g(t) dt
∣∣∣∣ ≤ A‖f‖[t−a,t−b] · ‖g‖1.
That is, A‖f ∗ g‖ ≤ A‖f‖ ‖g‖1.
Now we give some basic elementary properties of HK-convolution.
Theorem 5.4.4. Suppose f ∈ HK(R+) and g ∈ BV(R+). Then the follow-
ing holds:
I) f ∗ g(t) = g ∗ f(t) for all t ∈ R+.
II) (λf) ∗ g = f ∗ (λg) = λ(f ∗ g) for some λ ∈ C.
III) (f ∗ g)x(t) = fx ∗ g(t) = f ∗ gx(t).
IV) h ∗ (f + g)(t) = h ∗ f(t) + h ∗ g(t) for some h ∈ HK(R+).
Proof. The proof is similar to that of the classical case.
129
The associative property of HK-convolution is given in the next theorem.
Theorem 5.4.5. If f ∈ HK(R+), g ∈ BV(R+) and h ∈ L1(R+), then
(f ∗ g) ∗ h(t) = f ∗ (g ∗ h)(t), ∀ t ∈ R+.
Proof. Consider
(f ∗ g) ∗ h(t) =
∫ t
0
∫ τ
0
f(ξ) g(τ − ξ) h(t− τ) dξ dτ.
By changing the order of integration, we get
(f ∗ g) ∗ h(t) =
∫ t
ξ=0
f(ξ)
∫ t−ξ
0
g(u)h(t− u− ξ) du dξ
=
∫ t
ξ=0
f(ξ) g ∗ h(t− ξ) dξ
= f ∗ (g ∗ h)(t).
Now Consider
I1 =
∫ t
τ=ξ
∫ t
ξ=0
f(ξ) g(τ − ξ)h(t− τ) dξ dτ
and
I2 =
∫ t
ξ=0
f(ξ)
∫ t
τ=ξ
g(τ − ξ)h(t− τ) dτ dξ.
Let
J =
∫ t
τ=ξ
g(τ − ξ)h(t− τ) dτ.
Integrate by parts, we get
|J | ≤∣∣∣∣g(t− ξ)
∫ t
τ=ξ
h(t− τ) dτ
∣∣∣∣+
∣∣∣∣∫ t
τ=ξ
∫ u
τ=ξ
h(t− τ) dτ dg(τ − ξ)∣∣∣∣
≤ inf[ξ,t]|g| ·
∣∣∣∣∫ t
τ=ξ
h(t− τ) dτ
∣∣∣∣+
∣∣∣∣∫ t
τ=ξ
h(t− τ) dτ
∣∣∣∣ · V[ξ,t] g(τ − ξ)
≤ inf[ξ,t]|g| ·
∫ t
τ=ξ
|h(t− τ)| dτ +
∫ t
τ=ξ
|h(t− τ)| dτ · V[ξ,t] g(τ − ξ)
≤ inf[ξ,t]|g| · ‖h‖1 + ‖h‖1 · V[ξ,t] g(τ − ξ)
130
≤ ‖h‖1
{inf[ξ,t]|g|+ V[ξ,t] g(τ − ξ)
}.
Therefore
I2 ≤ supR+
∣∣∣∣∫ t
0
f(ξ) dξ
∣∣∣∣ ·{‖h‖1
{inf[ξ,t]|g|+ V[ξ,t] g(τ − ξ)
}}≤ A‖f‖ · ‖h‖1
{inf[ξ,t]|g|+ V[ξ,t] g(τ − ξ)
}.
Hence I2 exists on R+ × R+. And, by [lemma 25, [13]] we are through.
Now we show that the HK-Laplace transform of convolution of two func-
tions is the product of their HK-Laplace transforms.
Theorem 5.4.6. Let f1 ∈ HK(R+), f2 ∈ L1(R+) ∩ BV(R+). Then f ∗ g
exists and if L {f1(t)} and L {f2(t)} exist at s ∈ C, then L {f1 ∗ f2(t)} exists
at s ∈ C and L {f1 ∗ f2(t)} (s) = L {f1(t)} (s) L {f2(t)} (s).
Proof. Consider
L {f1 ∗ f2(t)} (s) =
∫ ∞0
e−stf1 ∗ f2(t) dt
=
∫ ∞0
∫ t
0
e−sτf1(τ)e−s(t−τ)f2(t− τ) dτ dt.
By interchanging the iterated integrals, we have
L {f1 ∗ f2(t)} (s) =
∫ ∞τ=0
e−sτf1(τ)
∫ ∞t=τ
e−s(t−τ)f2(t− τ) dt dτ
=
∫ ∞0
e−sτf1(τ) dτ
∫ ∞0
e−suf2(u) du.
But by hypothesis both the integrals∫ ∞0
e−sτ f1(τ) dτ and
∫ ∞0
e−sτ f2(τ) dτ
exist at s ∈ C.
Therefore L {f1 ∗ f2(t)} exists at s ∈ C and
131
L {f1 ∗ f2(t)} (s) = L {f1(t)} (s) L {f2(t)} (s).
Now it remains to show the justification for interchanging the iterated inte-
grals.
Since the function f2 ∈ L1(R+), ∃ KI such that ‖f2‖1 < KI , where I =
[a, b] ⊂ R+ is any compact interval.
Clearly, V[a,b] e−s(t−τ) f2(t− τ) ≤ e−σu V[t−b,t−a] f2(τ) + |f2(τ)| 2 e−σr for some
reals u and r and so∫ ∞0
V[t−b,t−a]e−sτ f2(τ) dτ ≤ e−σu
∫ ∞0
V[t−b,t−a]f2(τ) dτ+2e−σr∫ ∞
0
|f2(τ)| dτ.
Therefore∫ ∞0
V[t−b,t−a] e−sτ f2(τ) dτ ≤ e−σu
∫ ∞0
V[t−b,t−a] f2(τ) dτ + 2KI e−σr.
Since f2 ∈ BV(R+), there exists a constant MI such that∫ ∞0
V[t−b,t−a]f2(τ) dτ < MI .
Then ∫ ∞0
V[t−b,t−a]e−sτf2(τ) dτ ≤ e−σuMI + 2KI e
−σr.
Thus for each compact interval I = [a, b] ⊂ R+, the integral∫ ∞0
V[t−b,t−a]e−sτf2(τ) dτ
is finite and ‖f2‖1 ≤ KI .
Hence the integral ∫ ∞τ=0
∫ ∞t=τ
e−stf1(τ) f2(t− τ) dt dτ
exists on R+×R+. And, by [lemma 25, [13]], we can interchange the iterated
integrals.
132
Theorem 5.4.7. In addition to the hypothesis of the above theorem if f1 ∗
f2(t) is in HK(R+) and f2 is continuous, then f1 ∗ f2(t) is continuous with
respect to the Alexiewicz norm.
Proof. Choose any δ > 0 such that 0 < δ ≤ 1 (The case for negative δ is
analogous).
Define D(t, δ) = f1 ∗ f2(t+ δ)− f1 ∗ f2(t) = I1 + I2.
Where
I1 =
∫ t
0
f1(τ) [f2(t+ δ − τ)− f2(t− τ)] dτ, I2 =
∫ t+δ
t
f1(τ)f2(t+δ−τ) dτ.
Since f2 ∈ BV(R+), f2 is bounded on R+.
So there exists M2 such that |f2(x)| ≤M2, for all x ∈ R+. Then
|I2| ≤M2
∣∣∣∣∫ t+δ
t
f1(τ) dτ
∣∣∣∣ .But f1 ∈ HK(R+) implies that
limδ→0
∣∣∣∣∫ t+δ
t
f1(τ) dτ
∣∣∣∣ −→ 0.
Therefore |I2| → 0 as δ → 0.
Now consider
I1 =
∫ t
0
f1(τ) [f2(t+ δ − τ)− f2(t− τ)] dτ.
By integrating by parts, we get
|I1| ≤∣∣∣∣[f2(t+ δ − τ)− f2(t− τ)]
∫ t
0
f1(τ) dτ
∣∣∣∣+
∣∣∣∣∫ t
0
∫ u
0
f1(τ) dτ d(f2(u+ δ − τ)− f2(u− τ))
∣∣∣∣≤ |f2(t+ δ − τ)− f2(t− τ)| · sup
t∈R+
∣∣∣∣∫ t
0
f1(τ) dτ
∣∣∣∣+ sup
u∈R+
∣∣∣∣∫ u
0
f1(τ) dτ
∣∣∣∣ · ∣∣∣∣∫ t
0
d(f2(u+ δ − τ)− f2(u− τ))
∣∣∣∣133
= |f2(t+ δ − τ)− f2(t− τ)| · supt∈R+
∣∣∣∣∫ t
0
f1(τ) dτ
∣∣∣∣+ sup
u∈R+
∣∣∣∣∫ u
0
f1(τ) dτ
∣∣∣∣ · |f2(u+ δ − τ)− f2(u− τ)|.
Therefore
|I1| ≤ |f2(t+ δ − τ)− f2(t− τ)| · supt∈R+
∣∣∣∣∫ t
0
f1(τ) dτ
∣∣∣∣+ sup
u∈R+
∣∣∣∣∫ u
0
f1(τ) dτ
∣∣∣∣ · |f2(u+ δ − τ)− f2(u− τ)| .
Since f2 is continuous on R+, for given ε > 0 there exists η > 0 such that
|I1| ≤ ε, ∀ δ < η.
Therefore |D(t, δ)| ≤ |I1|+ |I2| < ε, ∀ δ < η.
That is, for given ε > 0 there exists η > 0 such that |D(t, δ)| < ε, ∀ δ < η.
Hence f1 ∗ f2 is continuous at t ∈ R+.
Now, let α, β ∈ R+.
Consider∫ β
α
[f1 ∗ f2(t+ δ)− f1 ∗ f2(t)] dt =
∫ α+δ
α
f1 ∗ f2(t) dt−∫ β+δ
β
f1 ∗ f2(t) dt.
Write F1,2(x) =∫ x
0f1 ∗ f2(t) dt. Then
supα, β∈R+
∣∣∣∣∫ β
α
[f1 ∗ f2(t+ δ)− f1 ∗ f2(t)] dt
∣∣∣∣ ≤ supα∈R+
|F1,2(α + δ)− F1,2(α)|
+ supβ∈R+
|F1,2(β + δ)− F1,2(β)| .
Since F1,2 is an indefinite HK integral of f1 ∗ f2 on R+, it is continuous on
R+ [5]. Therefore for given ε > 0 there exists η > 0 such that
|F1,2(ξ + δ)− F1,2(ξ)| < ε
2, ∀ δ < η and for all ξ ∈ R+.
Hence we have
supα, β∈R+
∣∣∣∣∫ β
α
[f1 ∗ f2(t+ δ)− f1 ∗ f2(t)] dt
∣∣∣∣ < ε, ∀ δ < η.
134
Thus for given ε > 0 there exists η > 0 such that A‖f1∗f2(t+δ)−f1∗f2(t)‖ <
ε, ∀ δ < η.
This shows that f1 ∗ f2(t) is continuous at t ∈ R+ with respect to the Alex-
iewicz norm.
The following result concerns with necessary and sufficient conditions
on fn and gn : [0, t] → R so that the HK-convolution as an operator is
continuous. It involves either uniform boundedness or uniform convergence
of the indefinite HK integral of fn.
Theorem 5.4.8. Let {fn} be a sequence of HK-integrable functions, fn :
[0, t] → R, ∀ n such that∫ t
0fn =
∫ t0f as n → ∞ for some HK-integrable
function f : [0, t]→ R. Define Fn (x) =∫ x
0fn, F (x) =
∫ x0f .
Let {gn} be a sequence of uniform bounded variation functions, gn : [0, t]→
R, ∀ n, such that gn → g pointwise on [0, t], where g : [0, t] → R. Then
fn ∗ gn(t)→ f ∗ g(t) as n→∞ if and only if
i) Fn → F uniformly on [0, t] as n→∞.
ii) Fn → F pointwise on [0, t], {Fn} is uniformly bounded and V (gn − g)→
0.
Proof. i) Suppose Fn → F uniformly on [0, t] as n→∞.
By assumption and multiplier theorem [1], fn ∗ gn(t) exists at t ∈ R+ and
fn ∗ gn(t) =
∫ t
0
fn(u) gn(t− u) du and f ∗ g(t) =
∫ t
0
f(u) g(t− u) du.
Consider fn ∗ gn(t)− f ∗ g(t) = I1 + I2,
where
I1 =
∫ t
0
[fn(u)− f(u)] gn(t− u) du, I2 =
∫ t
0
[gn(t− u)− g(t− u)] f(u) du.
135
By integrating by parts, we get
I1 = gn(t− u)
∫ t
0
[fn(u)− f(u)] du−∫ t
0
∫ y
0
[fn(u)− f(u)] dgn
= gn(t− u)[Fn(t)− F (t)]−∫ t
0
[Fn(u)− F (u)] dgn.
Then
|I1| ≤Max0≤u≤t
|Fn(u)− F (u)|∫ t
0
dgn +M |Fn(t)− F (t)|
−→ 0 as n→∞ (Since Fn → F uniformly).
Hence I1 → 0 as n→∞.
And by integrating by parts, we get
I2 = {gn(t− u)− g(t− u)}∫ t
0
f −∫ t
0
F dgn +
∫ t
0
F dg.
By assumption the first term tends to 0 and since F is continuous and each
gn is of bounded variation with gn → g as n → ∞, so we have∫ t
0F dgn →∫ t
0F dg. Hence I2 → 0 as n→∞. Therefore fn ∗ gn(t)→ f ∗ g(t) as n→∞.
Now for the converse part, suppose that either Fn 9 F on [0, t] or Fn−F → 0
not uniformly on [0, t].
Then there is a sequence {xi}i∈Λ in [0, t] on which Fn 9 F uniformly. So,
by Bolzano-Weierstrass theorem [7], we can find a subsequence {yi}i∈Γ of
{xi}i∈Λ, Γ ⊂ Λ, such that Fn(yi) 9 F (yi) for all i and yi → y as i→∞, i.e.,
Fn(yi)− F (yi) 9 0.
Let us assume without loss of generality that 0 < yn ≤ y ≤ t. Let H be
Heaviside step function.
Define
gn(t− u) =
H(u− yn), n ∈ Γ
H(u− y), otherwise
136
and g(t− u) = H(u− y).
Then we have, for n ∈ Γ, fn∗gn(t) = Fn(t)−Fn(yn) and f∗g(t) = F (t)−F (y).
Since Fn(yn) 9 F (yn) as n → ∞, yn → y as n → ∞ and F is continuous,
we have Fn(yn) 9 F (y).
But Fn(t) → F (t) as n → ∞. Therefore fn ∗ gn(t) 9 f ∗ g(t) as n → ∞
which is contradiction. Hence we must have Fn → F uniformly on [0, t].
ii) Suppose Fn → F pointwise on [0, t], Fn is uniformly bounded and V (gn−