POLONICI MATHEMATICI LXIII.1 (1996)
Laplace ultradistributions on a half line and a strong quasi-analyticity principle
by Grzegorz Lysik (Warszawa)
Abstract. Several representations of the space of Laplace ultradistributions sup- ported by a half line are given. A strong version of the quasi-analyticity principle of Phragm´ en–Lindel¨ of type is derived.
The theory of ultradistributions was founded by Buerling and Roumieu in the sixties as a generalization of the theory of Schwartz distributions.
Since then it was extensively studied by many authors: Bj¨ ork, Braun, Ko- matsu, Meise, Pilipovi´ c, Taylor , . . . , to mention but a few. The most system- atic treatment was presented by Komatsu [2], [3]. He derived, in particular, the boundary value representation of the space D
(Mp)0(Ω) of ultradistri- butions on an open set Ω ⊂ R
n, structure theorems for D
(Mp)0(Ω) and described the image of the space D
(MK p)0of ultradistributions with com- pact support in K under the Fourier–Laplace transformation. Following his approach Pilipovi´ c [9] recently introduced and investigated the space S
(Mp)0(R) of tempered ultradistributions. On the other hand, in the study of the Laplace transformation it is convenient to consider the space L
0(ω)(Γ ) of Laplace distributions of type ω ∈ R supported by a half line Γ . Since in the logarithmic variables the Laplace transformation is the Mellin transfor- mation we refer here to the book of Szmydt and Ziemian [11], where the latter transformation was systematically studied following the approach of Zemanian [12].
The aim of the present paper is to unify the theory of ultradistributions with that of Laplace distributions. We present it in the case of the space L
(M(ω)p)0(Γ ) of Laplace ultradistributions of Buerling type. Our theory is based on the Seeley type extension theorems for ultradifferentiable functions re-
1991 Mathematics Subject Classification: Primary 46F12, 44A10, 30D15.
Key words and phrases: ultradistributions, boundary values, quasi-analyticity.
Supported by KBN grant No. 2104591091.
[13]
cently proved by Langenbruch [4] and Meise and Taylor [7]. We describe the image of the space L
(M(ω)p)0(Γ ) under the Laplace, Taylor and (modified) Cauchy transformations. In the latter case we follow the method of Mo- rimoto [8]. As an application of our theory we give, in the final section, a version of the quasi-analyticity principle of Phragm´ en–Lindel¨ of type. It says that a function holomorphic and of exponential type in the half plane {Re z > 0} vanishes if it satisfies some growth conditions along vertical lines and decreases superexponentially along a ray in {Re z > 0}.
0. Notation. Let t > 0. We denote by e B(t) the universal covering of the punctured disc B(t) \ {0} and by e C that of C \ {0}. We treat e B(t) and C as Riemann manifolds. Recall that any point x ∈ e e B(t) can be written in the form x = |x| exp i arg x with |x| < t.
We denote by µ : C → e C the biholomorphism µ(z) = e
−zfor z ∈ C,
i.e. µ(z) = x ∈ e C with |x| = e
− Re z, arg x = − Im z. Then the inverse mapping µ
−1: e C → C is given by
µ
−1(x) = − ln x for x ∈ e C.
Let v ∈ R. We set
Γ
v= [v, ∞) and I
v= (0, e
−v].
Observe that I
v= µ(Γ
v). In the following we omit the subscript v as long as it is fixed. For z ∈ C we define the function exp
z: R → C by
exp
zy = e
yz, y ∈ R.
For A ⊂ C we set
A
ε= {z ∈ C : dist(z, A) < ε}, ε > 0.
We write D for the differential operator d/dx.
Let {P
τ}
τ ∈Tbe a family of multivalued vector spaces. Then lim −→
τ ∈TT
τ(resp. lim ←−
τ ∈TP
τ) denotes the inductive limit (resp. projective limit) of P
τ, τ ∈ T .
O(W ) denotes the set of holomorphic functions on an open subset W of some Riemann manifold. The value of a functional S on a test function ϕ is denoted by S[ϕ].
1. Laplace ultradistributions on a half line. Let (M
p)
p∈N0be a
sequence of positive numbers. Throughout the paper we assume that (M
p)
satisfies the following conditions:
(M.1) (Logarithmic convexity )
M
p2≤ M
p−1M
p+1for p ∈ N;
(M.2) (Stability under ultradifferential operators) There are constants A, H such that
M
p≤ AH
pmin
0≤q≤p
M
qM
p−qfor p ∈ N
0;
(M.3) (Strong non-quasi-analyticity) There is a constant A such that
∞
X
q=p+1
M
q−1M
q≤ Ap M
pM
p+1for p ∈ N.
Some results remain valid when (M.2), (M.3) are replaced by the follow- ing weaker conditions:
(M.2
0) (Stability under differential operators) There are constants A, H such that
M
p+1≤ AH
pM
pfor p ∈ N
0; (M.3
0) (Non-quasi-analyticity)
∞
X
p=1
M
p−1M
p< ∞.
Define
m
p= M
p/M
p−1for p ∈ N.
Then (M.1) is equivalent to saying that the sequence m
pis non-decreasing, and by (M.3
0) it follows that m
p→ ∞.
Note that the condition (M.3
0) implies the following: for every h > 0 there exists δ > 0 such that
(1) M
ph
p≥ δ for p ∈ N
0,
which is equivalent to the finiteness of the associated function M defined by
(2) M (%) = sup
p∈N0
ln %
pM
0M
pfor % > 0.
If M
p/p! satisfies (1) the growth function M
∗is defined by
(3) M
∗(%) = sup
p∈N0
ln %
pp!M
0M
pfor % > 0.
Example 1. The Gevrey sequence of order s > 1 is defined by M
p=
(p!)
s, p ∈ N
0. It satisfies all conditions (M.1)–(M.3) and M (%) ∼ %
1/s,
M
∗(%) ∼ %
1/s−1as % → ∞.
R e m a r k 1. It follows from Lemma 4.1 of [2] that if M
psatisfies (M.1) and (M.3
0) then the associated function M is sublinear, i.e. M (%)/% → 0 as
% → ∞.
R e m a r k 2. If M
psatisfies (M.1) and (M.3
0) then
(4) lim
p→∞
(M
p/p!)
1/p= ∞.
P r o o f. Take any l < ∞. Then by (M.1) and (M.3
0) there exists p
l∈ N such that M
p≥ lpM
p−1for p ≥ p
l. Hence
M
p≥ C
l· l
p· p! for p ≥ p
l, where C
l= M
pl−1M
0(p
l− 1)! l
1−pl, and we easily get (4).
Definition. Let Γ = [v, ∞) with v ∈ R. The space D
(Mp)0(Γ ) of ultra- distributions on Γ of class (M
p) is defined as the dual space of
D
(Mp)(Γ ) = lim −→
K⊂Γ
lim ←−
h>0
D
(MK,hp)(Γ ), where for any compact set K ⊂ Γ and h > 0,
D
(MK,hp)(Γ )
=
ϕ ∈ C
∞(Γ ) : supp ϕ ⊂ K and kϕk
(MK,hp)= sup
y∈K
sup
α∈N0
|D
αϕ(y)|
h
αM
α< ∞
. By ϕ ∈ C
∞(Γ ) we mean a restriction to Γ of some function ϕ ∈ C e
∞(R).
Definition. Let ω ∈ R∪{∞}. We define the space L
(M(ω)p)0(Γ ) of Laplace ultradistributions as the dual space of
L
(M(ω)p)(Γ ) = lim −→
a<ω
L
(Ma p)(Γ ), where for any a ∈ R,
L
(Ma p)(Γ ) = lim ←−
h>0
L
(Ma,hp)(Γ ),
L
(Ma,hp)(Γ ) =
ϕ ∈ C
∞(Γ ) : kϕk
(Ma,hp)= sup
y∈Γ
sup
α∈N0
|e
−ayD
αϕ(y)|
h
αM
α< ∞
. Lemma 1. Assume that (M
p) satisfies (M.1) and (M.3
0). Then D
(Mp)(Γ ) is a dense subspace of L
(M(ω)p)(Γ ). Thus, L
(M(ω)p)0(Γ ) is a subspace of the space of ultradistributions D
(Mp)0(Γ ).
P r o o f. Making a translation if necessary we can assume that Γ = R
+.
Let ϕ ∈ L
(M(ω)p)(Γ ). Then there exist a < b < ω such that ϕ ∈ L
(Ma p)(Γ ) ⊂
L
(Mb p)(Γ ). By the Denjoy–Carleman–Mandelbrojt theorem (cf. [2], [6]) there
exists a function ψ ∈ D
(Mp)(Γ ) such that 0 ≤ ψ(y) ≤ 1 for y ∈ Γ , ψ(y) = 1 for 0 ≤ y ≤ 1 and ψ(y) = 0 for y ≥ 2. Put ϕ
ν(y) = ψ(y/ν)ϕ(y) for y ∈ Γ , ν ∈ N. Then ϕ
ν∈ D
(Mp)(Γ ) and we shall show that ϕ
ν→ ϕ in L
(Mb p)(Γ ) as ν → ∞. To this end take any h > 0. Noting that (M.1) implies M
qM
p−q≤ M
0M
pfor 0 ≤ q ≤ p, by the Leibniz formula we get
kϕ
ν− ϕk
(Mb,hp)= sup
y∈Γ
sup
α∈N0
|e
−byD
α(ϕ(y)(ψ(y/ν) − 1))|
h
αM
α≤ sup
y∈Γ
sup
α∈N0
e
−ay|D
αϕ(y)|
h
αM
αe
(a−b)y|ψ(y/ν) − 1|
+ sup
y∈Γ
sup
α∈N
X
0≤β<α
α β
e
−ay|D
βϕ(y)|
h
βM
β· e
(a−b)y|D
α−β(ψ(y/ν) − 1)|M
0h
α−βM
α−β. Since ψ(y/ν) = 1 for 0 ≤ y ≤ ν the first summand tends to zero as ν → ∞.
Put K = [1, 2]. Then for β < α and any h
1> 0 we have
|D
α−β(ψ(y/ν) − 1)| = |ν
−(α−β)ψ
(α−β)(y/ν)| ≤ ν
−1kψk
(MK,hp)1
h
α−β1M
α−β. We also have for any h
2> 0 and β ≥ 0, e
−ay|D
βϕ(y)| ≤ kϕk
(Ma,hp)2
h
β2M
β. So the second summand is bounded by
M
0e
(a−b)νν sup
α∈N
X
β<α
α β
kϕk
(Ma,hp)2
h
2h
βkψk
(MK,hp)1
h
1h
α−β≤ M
0ν kϕk
(Ma,hp)2
kψk
(MK,hp)1
if h
2+ h
1≤ h and thus tends to zero as ν → ∞, proving the lemma.
Example 2. Let (M
p) satisfy (1). Then the function Γ 3 y → exp
zy = e
yzbelongs to L
(M(ω)p)(Γ ) if and only if Re z < ω. Furthermore, in this case for any a < ω and h > 0 we have
k exp
zk
(Ma,hp)= M
0−1exp{(Re z − a)v + M (|z|/h)}.
Let (M
p) satisfy (M.1) and (1), and let z ∈ C. Then the operation of multiplication
exp
z: L
(M(ω)p)(Γ ) → L
(M(ω+Re z)p)(Γ ) is continuous. Thus the formula
exp
zS[ϕ] = S[exp
zϕ] for ϕ ∈ L
(M(ω+Re z)p)(Γ ), S ∈ L
(M(ω)p)0(Γ ) defines a continuous operation
exp
z: L
(M(ω)p)0(Γ ) → L
(M(ω−Re z)p)0(Γ ).
Let (M
p) satisfy (M.2). An ultradifferential operator P (D) of class (M
p) is defined by
P (D) =
∞
X
α=0
a
αD
α,
where a
α∈ C satisfy the following condition: there are constants K < ∞ and C < ∞ such that
(5) |a
α| ≤ C K
αM
αfor α ∈ N
0.
The entire function C 3 z → P (z) is called a symbol of class (M
p). As in the proof of Theorem 2.12 of [2] one can show that an ultradifferential operator of class (M
p) defines linear continuous mappings
P (D) : L
(M(ω)p)(Γ ) → L
(M(ω)p)(Γ ), P (D) : L
(M(ω)p)0(Γ ) → L
(M(ω)p)0(Γ ), where for S ∈ L
(M(ω)p)0(Γ ) and ϕ ∈ L
(M(ω)p)(Γ ),
P (D)S[ϕ] = S[P
∗(D)ϕ] with P
∗(D) =
∞
X
α=0
(−1)
αa
αD
α. For a ∈ R and ω ∈ R ∪ {∞} we define
(6) Y
a= span{exp
c}
c≤a, Y
(ω)= [
a<ω
Y
a.
Proposition 1. Let b < a. Then L
(Mb p)(Γ ) is contained in the closure of Y
ain L
(Ma p)(Γ ). Thus Y
(ω)is dense in L
(M(ω)p)(Γ ).
P r o o f. Since the multiplication by exp
−ais a continuous isomorphism of L
(Mc p)(Γ ) onto L
(Mc−ap)(Γ ) and Y
conto Y
c−a, where c ∈ R, it is sufficient to assume that a = 0. Let ϕ ∈ L
(Mb p)(Γ ). It is enough to show that for every ε > 0 and h > 0 there exists ψ ∈ Y
0such that kϕ − ψk
(M0,hp)< ε. To this end fix ε > 0 and h > 0. By the proof of Lemma 1 there exists e ψ ∈ D
(Mp)(Γ ) such that kϕ − e ψk
(M0,hp)< ε/2. Put η(x) = ψ ◦ µ
−1(x) for x ∈ I. Then η has compact support in I = µ(Γ ) and by the Roumieu theorem ([10], Th´ eor` eme 13), η ∈ D
(Mp)(I). By the Weierstrass type theorem ([2], Theorem 7.3) for any δ > 0 and h
1> 0 there exists a polynomial p = P
Nν=0
c
νx
νsuch that
(7) kη − pk
(M¯ p)I,h1
< δ.
Put ψ(y) = p ◦ µ(y) = P
Nν=0
c
νe
−νyfor y ∈ Γ . Then ψ ∈ Y
0and we shall
show that for a suitable choice of δ and h
1, k e ψ − ψk
(M0,hp)< ε/2. To this end
put f = η − p. Following the proof of Th´ eor` eme 13 of [10] one can show that the derivatives D
αf ◦ µ are estimated by
(8) kf k
(M¯ p)I,h1
·
α
X
β=1
e
−vβM
ββ! h
β1(α − 1)!α!
(α − β)!(β − 1)! . For γ ∈ N
0put
(9) H
γ= sup
β≥γ
(β!/M
β)
1/β.
Then by Remark 2, H
γ→ 0 as γ → ∞. Hence we can find γ ∈ N
0and h
1> 0 such that
(10) ( p
e
−vh
1+ pH
γ)
2· H < h,
where H is the constant in (M.2). Since by (M.2), M
p+q≤ AH
p+qM
pM
qfor p, q ∈ N
0and by (M.1), M
βM
α−β≤ M
1M
αfor 0 ≤ β ≤ α, we get for α ∈ N
0, 0 ≤ β ≤ α,
(11) M
βM
α−β+γ≤ M
βAH
α−β+γM
α−βM
γ≤ C
γH
αM
α,
where C
γ= AM
1M
γmax(H
γ, 1).
Observe also that
α < β(α − β + γ) for 1 ≤ β ≤ α, α ∈ N, and
α
X
β=1
α − 1 β − 1
2x
β−1y
α−β≤ ( √ x + √
y)
2α−2for α ∈ N, x ≥ 0, y ≥ 0.
Hence using (8), (9) and (11) we derive for α ∈ N, y ∈ Γ ,
|D
αf ◦ µ(y)|
≤ kf k
(M¯ p)I,h1
·
α
X
β=1
e
−vβh
β1H
γα−β+γ(α − 1)!α!
(α − β)!(α − β + γ)!β!(β − 1)! M
βM
α−β+γ≤ kf k
(M¯ p)I,h1
· C
γH
γγe
−vh
1 αX
β=1
(e
−vh
1)
β−1H
γα−β(α − 1)!
(α − β)!(β − 1)!
2H
αM
α≤ kf k
(M¯ p)I,h1
· C
γH
γγe
−vh
1( p
e
−vh
1+ pH
γ)
−2(( p
e
−vh
1+ pH
γ)
2H)
αM
α≤ e C
γM
αL
αkf k
(M¯ p)I,h1
, where L = ( p
e
−vh
1+ pH
γ)
2H.
Finally, choosing δ < ε/(2 e C
γ) in (7) we get by (10),
k e ψ − ψk
(M0,hp)= sup
y∈Γ
sup
α∈N0
|D
αf ◦ µ(y)|
h
αM
α≤ sup
α∈N0
C e
γL
αh
αkf k
(M¯ p)I,h1
< ε 2 . To end this section we quote the following version of
Seeley extension theorem ([4], [7]). Let Γ = [v, ∞), Γ
1= [v
1, ∞) with v
1< v and a ∈ R. Then there exists a linear continuous extension operator
(12) E : L
(Ma p)(Γ ) → L
(Ma p)(Γ
1) such that for every ϕ ∈ L
(Ma p)(Γ ), supp E ϕ ⊂ (v
1, ∞).
Corollary 1. Let S be a linear functional on L
(M(ω)p)(Γ ). Then S ∈ L
(M(ω)p)0(Γ ) if and only if for every a < ω there exist h > 0 and C < ∞ such that
(13) |S[ϕ]| ≤ Ckϕk
(Ma,hp)for ϕ ∈ L
(Ma p)(Γ ).
2. The Paley–Wiener type theorem for Laplace ultradistribu- tions. We assume the conditions (M.1), (M.2) and (M.3). Let Γ = [v, ∞) with v ∈ R. By Example 2 the function exp
zbelongs to L
(M(ω)p)(Γ ) if and only if Re z < ω. Hence we can define the Laplace transform of S ∈ L
(M(ω)p)0(Γ ) by
LS(z) = S[exp
z] for Re z < ω.
Since the mapping
{Re z < ω} 3 z → exp
z∈ L
(M(ω)p)(Γ ) is holomorphic, LS is a holomorphic function on {Re z < ω}.
Define
(14) O
v(Mp)(Re z < ω)
= {F ∈ O(Re z < ω) :
for every a < ω there exist h > 0 and C < ∞ such that
|F (z)| ≤ C exp{v Re z + M (|z|/h)} for Re z ≤ a}.
Applying Corollary 1 with ϕ = exp
zand Re z ≤ a, by Example 2 we get Theorem 1. Let S ∈ L
(M(ω)p)0(Γ ) and F (z) = LS(z) for Re z < ω. Then F ∈ O
v(Mp)(Re z < ω).
Theorem 2. Let ω
1≤ ω
2, S
1∈ L
(M(ωp)01)
(Γ ) and S
2∈ L
(M(ωp)02)
(Γ ). If (15) LS
1(z) = LS
2(z) for Re z < ω
1then S
1= S
2in L
(M(ωp)01)
(Γ ).
P r o o f. We have to prove that for arbitrary ϕ ∈ L
(M(ωp)1)
(Γ ), S
1[ϕ] = S
2[ϕ].
To this end fix ϕ ∈ L
(M(ωp)1)
(Γ ), choose b < ω
1such that ϕ ∈ L
(Mb p)(Γ ) and take b < a < ω
1. Since by (15), S
1[exp
c] = S
2[exp
c] for c ≤ a the proof follows from Proposition 1.
To prove the converse of Theorem 1 we need two lemmas. The first one is a restatement of Lemma 9.1 of [11] (cf. also [12]).
Lemma 2. Let a ∈ R. Suppose that G is holomorphic on the set {Re z ≤ b} and satisfies there the estimate
|G(z)| ≤ C
hzi
2e
v Re zwith some C < ∞, v ∈ R. Put
g(y) = 1 2πi
R
c+iR
G(z)e
−zydz for y ∈ R.
Then g does not depend on the choice of c ≤ b; it is a continuous function on R with support in Γ = [v, ∞); the function Γ 3 y → e
byg(y) is bounded ; g ∈ L
(M(b)p)0(Γ ) and G(z) = Lg(z) for Re z < b.
Lemma 3. Let ω ∈ R and k > 0. Then there exists a symbol P of class e (M
p) not vanishing on {Re z < ω + 1} such that e
(16) exp M (k|z|)
P (z) ≤ 1
hzi
2for Re z ≤ ω. e
P r o o f. Since m
p→ ∞ as p → ∞ (by (M.3
0)) we can find p
0∈ N such that m
p> 2k| ω| + k and |m e
p− kz| ≥ k|z| for p ≥ p
0and Re z ≤ ω. Put e
P (z) = (z − e ω − 1)
p0+1∞
Y
p=p0
1 − kz
m
pfor z ∈ C.
Then P does not vanish on {Re z < e ω} and by the Hadamard factorization theorem (cf. [2], Propositions 4.5 and 4.6) it is a symbol of class (M
p). On the other hand, if Re z ≤ ω we estimate from below: e
∞
Y
p=p0
1 − kz
m
p≥
∞
Y
p=p0
1 − k| ω| e m
psup
q≥p0
q
Y
p=p0
1 − kz m
p≥
∞
Y
p=p0
1 − k| ω| e m
psup
q≥p0
q
Y
p=p0
k|z|
m
p= C|z|
−p0+1exp M (k|z|),
where
C = M
p0−1M
0∞
Y
p=p0
1 − k| ω| e m
p> 0.
Hence, possibly multiplying P by a suitable constant, we get (16).
Theorem 3. Let ω ∈ R∪{∞} and let F ∈ O
v(Mp)(Re z < ω). Then there exists a Laplace ultradistribution S ∈ L
(M(ω)p)0(Γ ) such that
(17) F (z) = LS(z) for Re z < ω.
P r o o f. Fix a < ω. Choose ω ∈ R such that a < e ω < ω and assume e (14). By Lemma 3 we can find a symbol P of class (M
p), not vanishing on {Re z < e ω + 1} and satisfying (16). Next we apply Lemma 2 to the function
G(z) = F (z)
P (z) , Re z ≤ e ω.
We get a continuous function g which belongs to L
(M(a)p)0(Γ ) and satisfies Lg(z) = G(z) for Re z < a. Put S = P (−D)g. Then S ∈ L
(M(a)p)0(Γ ) and LS(z) = P (z)Lg(z) = F (z) for Re z < a.
Thus for every a < ω we can find S
a∈ L
(M(a)p)0(Γ ) such that LS
a(z) = F (z) for Re z < a. By Theorem 2 the definition S = S
aon L
(M(a)p)(Γ ), a < ω, defines correctly a functional S ∈ L
(M(ω)p)0(Γ ) which satisfies (17).
It follows from the proof of Theorem 3 that Laplace ultradistributions can be characterized as follows.
Theorem 4 (Structure theorem). An ultradistribution S ∈ D
(Mp)0(R) is in L
(M(ω)p)0(Γ ) if and only if for every a < ω there exist an ultradifferential operator P
aof class (M
p) and a function g
acontinuous on R with support in Γ such that
|g
a(y)| ≤ Ce
−ayfor y ∈ Γ,
|Lg
a(z)| ≤ C
hzi
2e
v Re zfor Re z < a and
(18) S = P
a(D)g
ain L
(M(a)p)0(Γ ).
3. Boundary value representation. In this section we use the follow- ing version of the Phragm´ en–Lindel¨ of theorem.
3-line theorem ([1]). Let R > 0 and F ∈ O(Γ
R) ∩ C
0(Γ
R). Suppose that for some k > 0 the function
Γ
R3 z → exp{−k|z|}F (z)
is bounded. If F is bounded on the boundary of Γ
Rthen it is also bounded on Γ
R.
Definition. Let R > 0, k > 0 and a ∈ R. We define the spaces L e
a(Γ
R) = {ϕ ∈ O(Γ
R) ∩ C
0(Γ
R) : kϕk
a,ΓR= sup
z∈ ¯ΓR
|ϕ(z)e
az| < ∞},
L e
(Ma,kp)(Γ
R\ Γ ) = {ϕ ∈ O(Γ
R\ Γ ) : ϕ · exp{−M
∗(k/|Im z|)} ∈ C
0(Γ
R), kϕk
(Ma,k,Γp)R
= sup
z∈ ¯ΓR
|ϕ(z) exp{az − M
∗(k/|Im z|)}| < ∞}.
By the 3-line theorem e L
a(Γ
R) is a closed subspace of the Banach space L e
(Ma,kp)(Γ
R\ Γ ) and we can define
H
a,k(Mp)(Γ
R, Γ ) = e L
(Ma,kp)(Γ
R\ Γ )/e L
a(Γ
R).
Further, we define L e
a(C) = lim ←−
R→∞
L e
a(Γ
R), L e
(Ma,kp)(C \ Γ ) = lim ←−
R→∞
L e
(Ma,kp)(Γ
R\ Γ ), L e
(Ma p)(C \ Γ ) = lim −→
k→∞
L e
(Ma,kp)(C \ Γ ), H e
a(Mp)(C, Γ ) = e L
(Ma p)(C \ Γ )/e L
a(C),
e
H
a(Mp)(Γ ) = lim −→
R→0 k→∞
H
a,k(Mp)(Γ
R, Γ ).
Let a < b. Then the natural mappings H e
b(Mp)(C, Γ ) → e H
a(Mp)(C, Γ ),
e
H
b(Mp)(Γ ) → e
H
a(Mp)(Γ )
are well defined and by the 3-line theorem they are injections. Thus, for ω ∈ R ∪ {∞}, we can define
H e
(ω)(Mp)(C, Γ ) = lim ←−
a<ω
H e
a(Mp)(C, Γ ), e
H
(ω)(Mp)(Γ ) = lim ←−
a<ω
e
H
a(Mp)(Γ ).
An element f ∈ e H
(ω)(Mp)(C, Γ ) is given by a set {F
a}
a<ωof functions such that for a < ω, F
a∈ e L
(Ma p)(C \ Γ ) and for a < b < ω, F
a− F
b∈ e L
a(C).
On the other hand, an element g ∈ e
H
(ω)(Mp)(Γ ) is given by a set {G
a}
a<ωof functions such that for a < ω there exist R
a> 0 and k
a< ∞ such that G
a∈ e L
(Ma,kp)a
(Γ
Ra\ Γ ) and for a < b < ω, G
a− G
b∈ e L
a(Γ
Ra∩ Γ
Rb, Γ ).
The natural mapping
(19) i : e H
(ω)(Mp)(C, Γ ) → e
H
(ω)(Mp)(Γ )
is defined by retaining the same set of defining functions. Obviously it is an
injection.
Lemma 4. Let S ∈ L
(M(ω)p)0(Γ ) and a < ω. Put C
aS(z) = 1
2πi S e
a(·−z)z − ·
for z ∈ C \ Γ.
Then C
aS ∈ e L
(Ma p)(C \ Γ ). Furthermore, if a < b < ω then C
aS − C
bS ∈ L e
a(C).
P r o o f. Take a < c < ω. By Theorem 4 we can find an ultradifferential operator P
cof class (M
p) and a continuous function g
cwith support in Γ satisfying |g
c(y)| ≤ Ce
−cyfor y ∈ Γ and S = P
c(D)g
cin L
(M(c)p)0(Γ ). Since for fixed z ∈ C \ Γ the function
Γ 3 y → e
a(y−z)z − y belongs to L
(M(c)p)(Γ ) we have
C
a(z) = 1
2πi e
−azR
Γ
g
c(y)P
c∗(D
y)
e
ayz − y
dy.
Let P
c∗(D) = P
∞α=0
(−1)
αa
αD
αwith a
αsatisfying (5) and let R > 0. Then for z ∈ Γ
R\ Γ we estimate
P
c∗(D)
e
ayz − y
≤
∞
X
α=0
|a
α|
D
αe
ayz − y
≤ e
ay∞
X
β=0
|a|
ββ!
∞
X
α=β
|a
α|α!
|Im z|
α−β+1≤ e
ay∞
X
β=0
|a|
βR
ββ!
∞
X
α=0
|a
α|α!
|Im z|
α+1≤ e
ay+|a|R· AC
HK exp M
∗2HK
|Im z|
since by (5) and (M.2
0) we have
∞
X
α=0
|a
α|α!
|Im z|
α+1≤ C
∞
X
α=0
K
αα!
|Im z|
α+1M
α≤ 2C sup
α∈N0
(2K)
αα!
|Im z|
α+1M
α≤ 2AC 2HK sup
α∈N0
(2HK)
α+1α!
|Im z|
α+1M
α+1≤ AC
HK exp M
∗2HK
|Im z|
. Put k = 2HK. Then for every R > 0 there exists C < ∞ such that
|C
aS(z)| ≤ C exp{− a Re z + M
∗(k/|Im z|)} for z ∈ Γ
R\ Γ.
Thus, C
aS ∈ e L
(Ma p)(C \ Γ ). If a < b < ω we take c < ω such that b < c and
note that for z ∈ C the function
Γ 3 y → e
a(y−z)− e
b(y−z)z − y
belongs to L
(M(c)p)(Γ ). Thus, the holomorphic extension of C
aS − C
bS is given by
(C
aS − C
bS)(z) = 1 2πi
R
Γ
g
c(y)P
c∗(D) e
a(y−z)− e
b(y−z)z − y
dy
and we easily find that C
a− C
b∈ e L
a(C).
Definition. Let S ∈ L
(M(ω)p)0(Γ ). Then by Lemma 4 the set {C
aS}
a<ωof functions defines an element f ∈ e H
(ω)(Mp)(Γ ). We call f the Cauchy transform of S and write f = CS.
Proposition 2. Let F
a∈ e L
(Ma,kp)(Γ
R\Γ ) with a ≤ 0. Then there exist an ultradifferential operator P
a(D) of class (M
p) and functions H
a±∈ O(Γ
R∩ {± Im z > 0 or Re z < v}) such that
1
◦P
a(D)H
a±= F
a;
2
◦For every 0 < R
0< R and a
0< a there exists C < ∞ such that
|H
a±(z)| ≤ C exp{−a
0Re z} for z ∈ Γ
R0∩ {± Im z > 0};
3
◦H
a±(x + iy) converges uniformly as y → 0+ to a function h
±acontin- uous on (v − R, ∞) and analytic on (v − R, v) satisfying
|h
±a(x)| ≤ C exp{−a
0x} for x ∈ (v − R
0, ∞).
Furthermore, if we put
S
a= F
a+(x + i0) − F
a−(x − i0), where F
a±(x ± i0) = P
a(D)h
±a, then S
aextends to a Laplace ultradistribution e S
a∈ L
(M(a)p)0(Γ ) defined by (20) S e
a[ϕ] =
∞
R
v−R
(h
+a(x) − h
−a(x))P
a∗(D)E ϕ(x) dx for ϕ ∈ L
(M(a)p)(Γ ).
In (20), E is a linear continuous extension mapping E : L
(M(a)p)(Γ ) → L
(M(a)p)([v − R, ∞)), which exists by the Seeley extension theorem.
P r o o f. Put
P (ζ) = (1 + ζ)
2∞
Y
p=1
1 + kζ
m
pfor ζ ∈ C.
Then P is a symbol of class (M
p). Define the Green kernel for P by G(z) = 1
2πi
∞
R
0
e
zζP (ζ) dζ for Re z < 0.
Then by Lemma 11.4 of [2], G can be holomorphically continued to the Riemann domain {z : −π/2 < arg z < 5π/2}, on which we have
P (D)G(z) = − 1 2πi
1 z . Furthermore, since for any 0 ≤ ψ < π/2,
|z|
|1 − z| ≤ 1
cos ψ and 1
|1 − z| ≤ 1 cos ψ
for z ∈ C with |arg z| ≤ ψ + π/2, following the proof of the above-mentioned lemma we conclude that on the domain {−ψ ≤ arg z ≤ 2π + ψ}, G is bounded by C/cos ψ with C < ∞ not depending on ψ. We also have
|g(x)| ≤ A √
x exp{−M
∗(k/x)} for x > 0, where
g(z) = G
+(z) − G
−(e
2πiz) for Re z > 0,
G
+being the branch of G on {−π/2 < arg z < π/2} and G
−that on {3π/2 < arg z < 5π/2}. Put
H
a±(z) = ±i R
γ±
G(±i(z − w))F
a(w) dw,
where γ
±is a closed curve encircling z once, in the anticlockwise direction, such that −π/2 < arg(±i(z − w)) < 5π/2 for w ∈ γ
±. We choose a starting point z
◦±of γ
±in such a way that |arg{±i(z − z
◦±)}| < π/2. Then H
a±is a holomorphic function on Γ
R∩ {± Im z > 0 or Re z < v} and does not depend on the choice of γ
±with a fixed starting point z
◦±. For a fixed γ and z changing in a compact set in the domain bounded by γ we have
P (D)H
a±(z) = ±i R
γ
P (D
z)G(±i(z − w))F
a(w) dw
= −1 2πi
R
γ
F
a(w)
z − w dw = F
a(z).
Let 0 < R
0< R and z ∈ Γ
R0∩{Im z > 0}. Fix z
◦∈ Γ
R∩{Im z > r
0}∩{Re z <
v − R
0} and take γ
+= γ
1∪ γ
2∪ γ
3∪ γ
4, where γ
1= [x
◦+ iy
◦, x + iy
◦],
γ
2= [x + iy
◦, x + iy], γ
3= [x + iy, x + iy
◦] and γ
4= [x + iy
◦, x
◦+ iy
◦]. Since
0 ≤ arg(i(z − w)) ≤ ψ for w ∈ γ
1and 2π ≤ arg(i(z − w)) ≤ 2π + ψ for
w ∈ γ
4, where 0 ≤ ψ < π/2 is such that tan ψ = (x − x
◦)/(y
◦− y), by the
boundedness of G on {0 ≤ arg ψ ≤ 2π + ψ} we have |G(i(z − w))| ≤ Cx for w ∈ γ
1∪ γ
4, where C does not depend on x. So
R
γ1∪γ4
G(i(z − w))F
a(w) dw
≤ Cx
2e
−axfor z ∈ Γ
R0∩ {Im z > 0}.
On the other hand,
R
γ2∪γ3
G(i(z − w))F
a(w) dw =
− i
y◦−y
R
0
g(t)F (x + i(y + t)) dt
≤ AC
y◦−y
R
0
√ t exp
M
∗k y + t
− M
∗k t
− ax
dt
≤ AC(y
◦− y)
3/2e
−axfor z ∈ Γ
R0∩ {Im z > 0}.
Thus, for any a
0< a one can find C < ∞ such that
|H
a+(z)| ≤ Ce
−a0xfor z ∈ Γ
R0∩ {Im z > 0}.
The estimate of H
a−is obtained in an analogous way.
The assertion 3
◦is clear from the above estimates.
Let ψ ∈ D
(Mp)((v − R, ∞)). By 1
◦and 2
◦we derive F
a(x ± i0)[ψ] = P
a(D)h
±a[ψ] =
∞
R
v−R
h
±a(x)P
a∗(D)ψ(x) dx
= lim
y→0+
∞
R
v−R
H
a(x ± iy)P
a∗(D)ψ(x) dx
= lim
y→0+
∞
R
v−R
F
a(x ± iy)ψ(x) dx.
Since for ψ ∈ D
(Mp)((v − R, v)), S
a[ψ] = lim
y→0+
v
R
v−R
P
a(D)(H
a+(x + iy) − H
a−(x − iy))ψ(x) dx = 0, S
ahas support in Γ and we can define the extension of S
aby (20).
Let f ∈ e
H
a(Mp)(Γ ). Then there exist R > 0, k < ∞ and a function
F
a∈ e L
(Ma,kp)(Γ
R\ Γ ) such that f = [F
a]. If a ≤ 0 we can apply Proposition 2
to F
a. If a > 0 then we apply Proposition 2 to F
a∗= e
azF
ainstead of F
a.
In this case denote by e S
a∗the element of L
(M(0)p)0(Γ ) given by (20) and define
S e
a= e
−axS e
a∗. In both cases e S
a∈ L
(M(a)p)0(Γ ) does not depend on the choice
of a defining function F
afor f . Thus, the assignment f → e S
adefines a
mapping
(21) b :
e
H
a(Mp)(Γ ) → L
(M(a)p)0(Γ ).
Since (21) holds for every a < ω we have b :
e
H
(ω)(Mp)→ lim ←−
a<ω
L
(M(a)p)0(Γ ) ' ( lim −→
a<ω
L
(M(a)p)(Γ ))
0= L
(M(ω)p)0(Γ ), where the isomorphism ' follows by the formula (1.2) of [2].
Theorem 5. The mapping
C : L
(M(ω)p)0(Γ ) → e H
(ω)(Mp)(C, Γ ) is a topological isomorphism with inverse b ◦ i.
P r o o f. Let S ∈ L
(M(ω)p)0(Γ ) and CS = f ∈ e H
(ω)(Mp)(Γ ). Then f = [{F
a}
a<ω] with F
a(z) = 1
2πi S e
a(·−z)z − ·
for z ∈ C \ Γ.
Treat f as an element of e
H
(ω)(Mp)(Γ ) and put e S = b(f ) ∈ L
(M(ω)p)0(Γ ). Fix a < ω. Then for ϕ ∈ Y
(a)we have
S[ϕ] = − e R
∂Γε
F
a(z)ϕ(z) dz = S
− 1 2πi
R
∂Γε
e
a(·−z)z − · ϕ(z) dz
= S[ϕ],
by the Cauchy integral formula. Since Y
(a)is dense in L
(M(a)p)(Γ ) and a < ω is arbitrary, we have e S[ϕ] = S[ϕ] for ϕ ∈ L
(M(ω)p)(Γ ). Thus b ◦ i ◦ C = id.
Let f ∈ e H
(ω)(Mp)(C, Γ ), f = [{F
a}
a<ω] with F
a∈ e L
(Ma p)(C \ Γ ) and F
a− F
b∈ e L
a(C) for a < b < ω. Put
e
f = i(f ) and fix a < ω. Then e
f = [F
a] in e
H
a(Mp)(Γ ) and by (21), e S
a= b(
e
f ) ∈ L
(M(a)p)0(Γ ). Observe that for ε > 0 we have
S e
a[ϕ] = − R
∂Γε
F
a(z)ϕ(z) dz for ϕ ∈ Y
(a). On the other hand, by the part of the theorem just proved,
S e
a[ϕ] = − R
∂Γε
1 2πi S e
ae
a(·−z)z − ·
ϕ(z) dz for ϕ ∈ Y
(a).
So for ϕ ∈ Y
(a), (22) R
∂Γε
ψ
a(z)ϕ(z) dz = 0, where ψ
a(z) = 1 2πi S e
ae
a(·−z)z − ·
− F
a(z).
Then ψ
a∈ e L
(Ma,kp)(C \ Γ ) and we shall show that ψ
aextends holomorphically to a function e ψ
a∈ e L
a(C), which proves that C ◦b◦i = id. To this end observe that (22) holds also for ϕ ∈ L
(M(a)p)(Γ ) ∩ O(Γ
ε) and put for any b < a, R > ε,
G
b(z) = R
∂ΓR
ψ
a(ζ) e
b(ζ−z)z − ζ dζ for z ∈ Γ
R.
Then |G
b(z)| ≤ C exp{−b Re z} for z ∈ Γ
R0with R
0< R. Using (22) with ϕ(ζ) = exp{b(ζ − z)}/(z − ζ), z ∈ Γ
R\ Γ
ε, we get G
b(z) = ψ
a(z) for z ∈ Γ
R\ Γ
ε. Put
ψ e
a(z) = ψ
a(z) for z ∈ Γ
R\ Γ , G
b(z) for z ∈ Γ
R.
Then e ψ
a∈ O(Γ
R) and by the 3-line theorem e ψ
a∈ e L
a(Γ
R). Since R > ε was arbitrary we have e ψ ∈ e L
a(C).
4. Mellin ultradistributions
Definition. Let ω ∈ R ∪ {∞}, v ∈ R and I = (0, e
−v]. We define the space M
(ω)(Mp)0(I) of Mellin ultradistributions as the dual space of
M
(ω)(Mp)(I) = lim −→
a<ω
lim ←−
h>0
M
a,h(Mp)(I), where for any a ∈ R and h > 0,
M
a,h(Mp)(I) =
ψ ∈ C
∞(I) : %
(Ma,hp)(ψ) = sup
x∈I
sup
α∈N0
|x
a+1(Dx)
αψ(x)|
h
αM
α< ∞
. Lemma 5. Let a ∈ R, h > 0, ψ ∈ M
a,h(Mp)(I) and ϕ = µ · ψ ◦ µ. Then ϕ ∈ L
(Ma,hp)(Γ ) and kϕk
(Ma,hp)= %
(Ma,hp)(ψ). Thus, the mapping
M
(ω)(Mp)(I) 3 ψ → µ · ψ ◦ µ ∈ L
(M(ω)p)(Γ ) is a continuous isomorphism with inverse
L
(M(ω)p)(Γ ) 3 ϕ → exp
1◦µ
−1· ϕ ◦ µ
−1∈ M
(ω)(Mp)(I).
P r o o f. The proof follows easily from the formula
D
αy(µ(y)ψ ◦ µ(y)) = (−1)
αx(D
xx)
αψ(x), for α ∈ N
0, x = µ(y), which can be proved by induction.
Let S ∈ L
(M(ω)p)0(Γ ). Put
S ◦ µ
−1[ψ] = S[µ · ψ ◦ µ] for ψ ∈ M
(ω)(Mp)(I).
Then by Lemma 5, S ◦ µ
−1is a well defined element of M
(ω)(Mp)0(I) and the mapping
L
(M(ω)p)0(Γ ) 3 S → S ◦ µ
−1∈ M
(ω)(Mp)0(I) is continuous.
Observe that the function
I 3 x → x
−z−1= exp
z+1◦µ
−1(x)
belongs to M
(ω)(Mp)(I) if and only if Re z < ω. Thus, we can define the Mellin transform of T ∈ M
(ω)(Mp)0(I) by
MT (z) = T [exp
z+1◦µ
−1] for Re z < ω.
Let S ∈ L
(M(ω)p)0(Γ ) and T = S ◦ µ
−1. Then for Re z < ω we have MT (z) = S ◦ µ
−1[exp
z+1◦µ
−1] = S[exp
z] = LS(z).
5. Strong quasi-analyticity principle
Definition. Let S ∈ L
(M(ω)p)0(Γ ). We define the Taylor transform of S by
T S(x) = LS(ln x) for x ∈ e B(e
ω).
We also define O
(Mv p)( e B(e
ω))
= {u ∈ O( e B(e
ω)) :
for every t < e
ωthere exist k < ∞ and C < ∞ such that
|u(x)| ≤ C exp{M (k(ω − ln |x| + |arg x|))} · |x|
vfor |x| ≤ t}.
By Theorems 1 and 3 we get
Theorem 6. The Taylor transformation is an isomorphism of L
(M(ω)p)0(Γ ) onto O
(Mv p)( e B(e
ω)).
Let u ∈ O
(Mv p)( e B(e
ω)). Then for any t < e
ω, u
|(0,t]∈ M
(v)(Mp)0((0, t]) and M
tu(z) =
t
R
0
u(x)x
−z−1dx for Re z < v.
By Theorem 6, u(x) = S[x
·] for x ∈ e B(e
ω) with S = T
−1u ∈ L
(M(ω)p)0(Γ ), Γ = [v, ∞). For Re z < v we derive
M
tu(z) = S h Rt
0
x
·−z−1dx i
= S t
·−z· − z
= −2πiC
ln tS(z).
Thus, by Lemma 4, M
tu extends holomorphically to a function M
tu ∈ L e
(Mln tp)(C \ Γ ) and the set of functions {M
tu}
t<eωdefines an element of H e
(ω)(Mp)(C, Γ ), which will be denoted by Mu and called the Mellin transform of u.
We can summarize Theorems 1, 3, 5 and 6 as follows:
Corollary 2. We have the following diagram of linear topological iso- morphisms:
M
(ω)(Mp)0(I) O
v(Mp)(Re z < ω)
L
(M(ω)p)0(Γ ) O
v(Mp)( e B(e
ω))
e
H
(ω)(Mp)(Γ ) H e
(ω)(Mp)(C, Γ ).
M
//
◦µ
◦−µ−1
T//
L
ooo ooo ooo oo77
OOOOO
COOOOO ''
◦µ−1
OO
M bOO
oo
iFollowing [13] we call the elements of O
v(Mp)(Re z < ω) generalized ana- lytic functions determined by L
(M(ω)p)0(Γ ). Generalized analytic functions have the following quasi-analyticity property:
Theorem 7. Let u ∈ O
v(Mp)( e B(e
ω)). Suppose that for some t < e
ωand every m ∈ N there exist C
msuch that
|u(x)| ≤ C
mx
mfor 0 < x ≤ t.
Then u ≡ 0.
P r o o f. By Theorem 6, u(x) = T S(x) for x ∈ e B(e
ω) with some S ∈ L
(M(ω)p)0(Γ ). The assumption that u is flat of arbitrary order m ∈ N on (0, t) implies that M
tu ∈ O(C). Since for every R > 0, L
(Ma,kp)(Γ
R)∩O(Γ
R) = L
a(Γ
R), Mu defines the zero element in e H
(ω)(Mp)(Γ ). Thus S = 0 and u ≡ 0.
Theorem 8 (Strong quasi-analyticity principle). Let −π/2 < θ < π/2, l
θ= {z = re
iθ: r > 0} and F ∈ O(Re z > 0). Suppose that for some v ∈ R and every κ > 0 there exist k < ∞ and C < ∞ such that
(23) |F (z)| ≤ C exp{v Re z + M (k|z|)} for Re z ≥ κ.
If for some τ > 0 and every m ∈ N there exists C
m< ∞ such that (24) |F (z)| ≤ C
me
−m Re zfor z ∈ l
θ, Re z ≥ τ,
then F ≡ 0.
P r o o f. Put u(x) = F ◦ µ
−1(x) for x ∈ e B(1). Then u ∈ O
v(Mp)( e B(1)).
Set t = e
−τ, let γ
t,θbe the set of x ∈ e B(1) that satisfy
x = t exp{−ir sin θ} for 0 ≤ r ≤ τ /cos θ, exp{−r(cos θ + i sin θ)} for r ≥ τ /cos θ, and observe that
(25) M
tu(z) = R
γt,θ
u(x)x
−z−1for z ∈ Ω
v,θ,
where Ω
v,θ= {z ∈ C : Re z < v and sin θ Im z > cos θ(Re z − v)}. Using (24) we infer that the right hand side of (25) is defined for z ∈ C. Thus, M
tu ∈ O(C). As in the proof of Theorem 7 this implies that u ≡ 0 and hence F ≡ 0.
R e m a r k 3. The conclusion of Theorem 8 does not hold if instead of (23) we assume that for every ε > 0 and κ > 0 there exists C
ε,κsuch that
|F (z)| ≤ C
ε,κexp{v Re z + ε|z|} for Re z ≥ κ.
In this case the function u = F ◦ µ
−1is the Taylor transform of an analytic functional with carrier at {∞} and need not be equal to zero.
R e m a r k 4. The results of the paper can be easily extended to the n-dimensional case if Γ is a cone of product type. The case of an arbi- trary convex, proper cone in R
nis more difficult and will be studied in a subsequent paper.
References
[1] E. H i l l e, Analytic Function Theory , Vol. 2, Chelsea, New York, 1962.
[2] H. K o m a t s u, Ultradistributions, I. Structure theorems and a characterization, J.
Fac. Sci. Univ. Tokyo 20 (1973), 25–105.
[3] —, Ultradistributions, II. The kernel theorem and ultradistributions with support in a submanifold , ibid. 24 (1977), 607–628.
[4] M. L a n g e n b r u c h, Bases in spaces of ultradifferentiable functions with compact support , Math. Ann. 281 (1988), 31–42.
[5] G. L y s i k, Generalized analytic functions and a strong quasi-analyticity principle, Dissertationes Math. 340 (1995), 195–200.
[6] S. M a n d e l b r o j t, S´ eries adh´ erentes, r´ egularisation de suites, applications, Gau- thier-Villars, Paris, 1952.
[7] R. M e i s e and A. T a y l o r, Linear extension operators for ultradifferentiable func- tions of Beurling type on compact sets, Amer. J. Math. 111 (1989), 309–337.
[8] M. M o r i m o t o, Analytic functionals with non-compact carrier , Tokyo J. Math. 1 (1978), 77–103.
[9] S. P i l i p o v i ´ c, Tempered ultradistributions, Boll. Un. Mat. Ital. B (7) 2 (1988),
235–251.
[10] C. R o u m i e u, Ultra-distributions d´ efinies sur R
net sur certaines classes de vari´ et´ es diff´ erentiables, J. Anal. Math. 10 (1962-63), 153–192.
[11] Z. S z m y d t and B. Z i e m i a n, The Mellin Transformation and Fuchsian Type Par- tial Differential Equations, Kluwer, Dordrecht, 1992.
[12] A. H. Z e m a n i a n, Generalized Integral Transformations, Interscience, 1969.
[13] B. Z i e m i a n, Generalized analytic functions, Dissertationes Math., to appear.
INSTITUTE OF MATHEMATICS POLISH ACADEMY OF SCIENCES P.O. BOX 137
SNIADECKICH 8´
00-950 WARSZAWA, POLAND E-mail: LYSIK@IMPAN.GOV.PL