Asymtotic Behaior

September 1, 2011 10:53 World Scientic Book - 9in x 6in maj16
Chapter I
Asymptotic Behavior of Generalized
Functions
0 Preliminaries
0.1. We denote by R and N the sets of real and natural numbers;
N
0
= N0. The following notation will be used. If x = (x
1
, . . . , x
n
), y =
(y
1
, . . . , y
n
) R
n
, then x y = x
1
y
1
+ + x
n
y
n
; |x|
2
= x
2
1
+ +
x
2
n
; x 0 x
i
0, i = 1, . . . , n; x x
i
, i =
1, . . . , n; R
n
+
= x R
n
; x > 0. If x = (x
1
, . . . , x
n
) R
n
and
k = (k
1
, . . . , k
n
) N
n
0
, then [k[= k
1
+ + k
n
, k! = k
1
! . . . k
n
! , x
k
=
x
k
1
1
. . . x
k
n
n
; D
k
=
k
1
/x
k
1
1
. . .
k
n
/x
k
n
n
, f
(k)
= D
k
f (f
(0)
= f);
z = (z
1
, . . . , z
n
) C
n
; D(a, r) denotes the polydisk z C
n
; [z
i
a
i
[<
r
i
, i = 1, . . . , n, where [z
j
[
2
= x
2
j
+ y
2
j
, z
j
= x
j
+ iy
j
C. B(a, r) denotes
the open ball x R
n
, |a x|< r and H is the Heaviside function:
H(t) = 0, t 0; H(t) = 1, t > 0.
0.2. A cone with vertex at zero in R
n
is a non-empty set such that
x and k > 0 imply kx . The cone is called solid if int ,= . The
conjugate cone (dual cone)
to the cone is the set R

n
; x 0
for each x . It is obvious that
is also a cone which is convex and

closed. The cone is called acute if
is a solid cone.
0.3. A function : (a, ) R, a R
+
, is called regularly varying at
innity [76] if it is positive, measurable, and if there exists a real number
such that for each x > 0
lim
k
(kx)
(k)
= x
. (0.1)
The number is called index of regular variation. If = 0, then is called
slowly varying at innity and for such a function the letter L will be used.
We then have that any regularly varying function can be written as (x) =
1
2 Asymptotic Behavior of Generalized Functions
x
L(x), x > a, where L is slowly varying. It is known that the convergence

of (0.1) is uniform on every xed compact interval [a
, b
], a < a
< b
< ,
and that is necessarily bounded (hence integrable) on it [9].
Let L be a slowly varying function at innity. Then, for each > 0,
(i) there exist constants C
1
, C
2
> 0 and X > a such that
C
1
x
L(x) C
2
x
, x X ; (0.2)
(ii) lim
x
x
L(x) = +, lim
x
x
L(x) = 0.
We know ([9], p. 16) that if L
2
(x) , x , and L
1
, L
2
are
slowly varying, then L
1
L
2
= L
1
(L
2
) is slowly varying, as well. Hence, for
x > ,
lim
h
L(x +h)
L(h)
= lim
u
L(log ut)
L(log u)
= 1 . (0.3)
The denition of a regular varying function at zero is similar. For the
denition of regularly generalized functions see [156] and [162].
0.4. The class of distributions f
, R, belonging to o
+
(see 0.5.1.)
is dened in the following way:
f
(t) =
_
_
H(t)t
1
/(), > 0,
f
(m)
+m
(t), 0, +m > 0,
where H is the Heaviside function, and the derivative f
(m)
is taken to be in
the distributional sense (see [192], Chapter I, 1). We therefore have f
0
= ,
the Dirac delta distribution; f
m
=
(m)
, m N; and f
p
f
q
= f
p+q
. We
also use the notations t
+
= ( + 1)f
+1
(t) and t
= (t)
+
, / N.
Let g o
+
. We denote g
()
= f
g, R ( the is convolution
symbol).
The sequence
m
mN
C
(R
n
) is called a -sequence if:
a) supp
m
[
m
,
m
],
m
0, m ; b)
m
0, m N;
c)
_
R
n
m
(t)dt = 1, m N.
If T, then
m
, m in T, hence
m
; m N is a
bounded set in T.
0.5. We will repeat denitions and some basic properties of generalized
functions dened as elements of dual spaces.
0. Preliminaries 3
0.5.1. The Schwartz spaces of test functions and distributions on R
n
are
denoted by T = T(R
n
) and T
= T
(R
n
), respectively (R
n
will be omitted
wherever n is not xed). The space T is a locally convex, barrelled, Montel
and complete space. If a lter with a countable basis is weakly convergent
in T
, then it is convergent in T
with the strong topology, as well (cf.

[146]).
o is the space of rapidly decreasing functions and its dual o
is the space
of tempered distributions [146]. For a closed cone R
n
, o
= f
o
; supp f . In the one-dimensional case o
+
= f o
(R); supp f
[0, ). Recall ([189]):
o() = C
(); ||
p
< , p N ,
where
||
p
= sup
x, ||p
(1 +|x|
2
)
p/2
[
()
(x)[ .
By o
p
() is denoted the completion of the set o() with respect to the
norm ||
p
. Note o() =
pN
0
o
p
() and o
() =
pN
0
o
p
(), where the
intersection and the union have topological meaning. A sequence f
n
nN
in o
() converges to f o
() if and only if it belongs to some o
q
() and
converges to f o
q
() in the norm of o
q
(). The space o
is isomorphic
to o
() if is closed convex solid cone ([189], [192]).

c
the space of distributions with compact support; it is isomorphic to

the dual space of c = C
(R
n
) (cf. [146]).
T
L
p, 1 p , is the space of smooth functions with all derivatives
belonging to L
p
([146]), T
L
p T
L
q if p < q.
.
B
is a subspace of B = T
L
, dened as follows:
.
B
if and only if
[
()
(x)[ 0 as |x| for every N
n
0
.
T
L
p, 1 < p is the dual space of T
L
q , 1 q < ,
1
p
+
1
q
= 1. T
L
1
is the dual of
B
and T
L
is denoted by B
.
O
c
is the space of distributions with fast descent:
O
c
= T T
; (1 +|x|
2
)
m
T B
, for every m N.
/
p
, p 1, is the spaces of functions C
with the property:
m
() = sup
xR
n
,||m
exp(m|x|
p
)[D
(x)[< , m = 1, 2, . . .
The elements of /
p
are called rapidly exponentially decreasing functions.
Then /
p
is the dual of /
p
.
The convolution in T
is dened as follows. Let

m
mN
be a sequence
in T(R
2n
) such that for every compact set K R
2n
there exists m
0
(K)
such that
m
(x) = 1, x K, m m
0
(K) and
sup
xR
2n
[
()
m
(x)[< C
, N
n
0
.
The convolution of T, S T
is dened by
T S, = lim
m
T(x)S(y),
m
(x, y)(x +y), T,
if this limit exists for every
m
m
(then, it does not depend on
m
m
).
By the BanachSteinhaus theorem we know that T S T
.
The spaces T
() and o
() with the operation are associative and

commutative algebras. The convolution in this case is separately continu-
ous.
We refer also to [63] and [3] for the theory of distributions.
0.5.2 Ultradistribution spaces
We follow the notation and denitions from [79], [81] and [86]. By M
p
p
is denoted a sequence of positive numbers, M
0
= M
1
= 1, satisfying some
of the following conditions:
(M.1) M
2
p
M
p1
M
p+1
, p N;
(M.2) M
p
/(M
q
M
pq
) AB
p
, 0 q p, p N;
(M.2)
M
p+1
AB
p
M
p
, p N
0
, N
0
= N 0;
(M.3)
q=p+1
M
q1
/M
q
A
p
M
p
/M
p+1
, p N;
(M.3)
p=1
M
p1
/M
p
< ,
where A and B are constants independent of p.
In the sequel, we will always assume (M.1), (M.2)
and (M.3)
.
Let h be a positive number and let h
p
p
be a real positive sequence
increasing to . We denote
H
p
=
_
h
p
, for the ultradierentiable functions of class (M
p
)
h
1
. . . h
p
, for the ultradierentiable functions of class M
p
.
(0.4)
0. Preliminaries 5
Let K be a compact set in an open set of R
n
.
We denote by c
H
p
M
p
K
the space of smooth functions c such that
sup
xK,N
n
0
[
()
(x)[/H
||
M
||
< . (0.5)
We set T
H
p
M
p
K
= c
H
p
M
p
K
; supp K, it is a Banach space with
norm
q
H
p
M
p
(f) = sup
xK,N
n
0
[
()
(x)[
H
||
M
||
. (0.6)
Then, the basic spaces are dened by
T
(M
p
)
K
= proj lim
h0
T
h
p
M
p
K
, T
{M
p
}
K
= ind lim
{H
p
=h
1
...h
p
}
T
H
p
M
p
K
T
() T
= ind lim
K
T
K
,
where denotes either (M
p
) or M
p
.
The spaces with the upper index (M
p
) are the Beurling-type spaces of ul-
tradierentiable functions and with the upper index M
p
are the Roumieu-
type spaces of ultradierentiable functions. Their strong duals are spaces
of Beurling and Roumieu-type ultradistributions, respectively.
The space c
= c
() is the dual of
c
(M
p
)
= proj lim
K
proj lim
h0
c
h
p
M
p
K
;
c
{M
p
}
= proj lim
K
ind limc
H
p
M
p
K
.
Weighted ultradistribution spaces are dened by
T
(M
)
L
1
(R
n
) = T
(M
)
L
1
= proj lim
h
T
M
L
1
,h
, T
{M
}
L
1
= ind lim
h0
T
M
L
1
,h
,
where T
M
L
1
,h
, h > 0, is the Banach space of smooth functions on R
n
with
nite norm
||
L
1
,h
= sup
N
n
0
h
||
M
||
|
()
|
L
1 .
T
= T
(R
n
) is dense in T
L
1
, and the inclusion mapping is continuous.
The strong dual of T
L
1
is denoted by B
.
Spaces of tempered ultradistributions are dened as the strong duals of
the following testing function spaces:
o
(M
)
(R
n
) = o
(M
)
= proj lim
h
o
M
h
,
o
{M
}
(R
n
) o
{M
}
= ind lim
h0
o
M
h
,
where o
M
h
, h > 0, is the Banach space of smooth functions on R with
nite norm
h
() = sup
,N
n
0
h
||+||
M
||
M
||
|(1 +[x[
2
)
||/2
()
|
L
.
The Fourier transform is an isomorphism of o
onto o
; T
= T
(R
n
) is
dense in o
, o
is dense in T
L
1
and the inclusion mappings are continuous.
The strong dual of o
, o
, is the space of tempered ultradistributions

(of Beurling and Roumieu types). There holds T
, where
means that the left space is dense in the right one and that the inclusion
mapping is continuous. Thus, c
. We denote o
+
= f
o
(R); supp f [0, ).

Let
o
[0,)
= C
[0, ); = [
[0,)
for some o
with the induced convergence structure from o
; its strong dual is in fact

o
+
.
An operator of the form: P(D) =
||=0
a
, a
C, N
n
0
, is
called ultradierential operator of class (M
p
) (of class M
p
) if there are
constants L > 0 and C > 0 (for every L > 0 there is C > 0) such that
[a
[ CL
||
/M
||
, [[ N
0
.
0.5.3. Fourier hyperfunctions. There are many equivalent denitions
of hyperfunctions, Laplace hyperfunctions and Fourier hyperfunctions (cf.
[71], [80], [144], [145], [204]), but we will use denitions and results collected
in [75]. Let I be a convex neighborhood of zero in R
n
and let be a non-
negative constant. A function F, holomorphic on R
n
+iI, is said to decrease
exponentially with type (), 0, if for every compact subset K I
and every > 0, there exists C
K,
> 0 such that
[F(z)[ C
K,
exp(( )[Re z[), z R
n
+iK . (0.7)
The set of all such functions is denoted by

O
(D
n
+iI), (

O(D
n
+iI) for
= 0), where D
n
denotes the directional compactication of R
n
: D
n
=
R
n
S
n1
(S
n1
consists of all points at innity in all direction). Space T
is dened by
T
= ind lim
I0
ind lim
0
(D
n
+iI) . (0.8)
0. Preliminaries 7
The dual space of T
is the space Q(D

n
) of the Fourier hyperfunctions.
Q(D
n
) is a space of FS-type. We can give a representation of elements
from Q(D
n
). Let O be the sheaf of holomorphic functions. Denote
U
j
= (D
n
+iI) Imz
j
,= 0, j = 1, . . . , n; (D
n
+iI)#D
n
= U
1
U
n
,
(D
n
+iI)#
j
D
n
= U
1
U
j1
U
j+1
U
n
.
Then
Q(D
n
) =

O((D
n
+iI)#D
n
)/
n
j=1
O((D
n
+iI)#
j
D
n
) . (0.9)
Thus, f Q(D
n
) is dened as the class [F], where F

O((D
n
+iI)#D
n
).
F is called a dening function of f and it is represented by 2
n
functions
F
, F = F
, where F

O(D
n
+ iI
); D
n
+iI
= D
n
+i(I
) is an
innitesimal wedge of type R
n
+i
0,
are open -th orthants in R

n
.
An f L
1
loc
(R
n
) is called a function of infra-exponential type if for
every > 0, there exists C
> 0 such that [f(x)[ C
exp([x[), x R
n
.
Then by f is denoted the hyperfunction dened by f.
We denote by the set of n-th variations of 1, 1.
The boundary-value representation of f Q(D
n
) is:
f = [F] :=
(x + i
0) . (0.10)
F
(x +i
0) denotes the element of the quotient space given in (0.9); it is

determined by F
.
The dual pairing between T
and g = [G] Q(D

n
) is given by
g, =
_
R
n
g(x)(x)dx =
_
Imw=v
(w)(w)dw, w = u +iv,
where v
.
Similarly, Q
(D
n
), > 0, is dened using

O
instead of

O (cf.
Denition 8.2.5 in [75]).
An innite-order dierential operator J(D) of the form
J(D) =
|a|0
b
a
D
a
with lim
|a|
|a|
_
[b
a
[a! = 0, b
a
C,
is called a local operator. J(D) maps continuously O(U) into O(U), U
being an open set in C
n
, and also Q(D
n
) into Q(D
n
).
The Fourier transform on Q(D
n
) is dened by using the functions
1
(z
1
) . . .
n
(z
n
), where
k
= 1, k = 1, . . . , n, = (
1
, . . . ,
n
)
and
+
(t) = e
t
/(1 +e
t
),
(t) = 1/(1 +e
t
), t R. Let
u(x)

=
(x +i
0) =

(

U
)(x +i
0) ,
where

U
Q(D
n
+ iI
), and

U
decreasing exponentially
along the real axis outside the closed -th orthant. The Fourier transform
of u is dened by
T(u)()

=

T(

U
)( i

0)
=

_
Imz=y
e
iz
(

U
)(z)dz, y
, = +i,
where T(

U
)

O(D
n
iI

) and T(

U
)(z) = O(e
|x|
) for a suitable
> 0 along the real axis outside the closed -orthant. T is an automor-
phism of Q(D
n
).
0.6. Let E be a locally convex topological vector space. An absolute
convex closed absorbent subset of E is called a barrel. If every barrel is
a neighborhood of zero in E, then E is called barreled. Throughout this
book T
g
stands for a locally convex barrelled complete Hausdor space of
smooth functions (the subscript g stands for general), T
g
c = c(R
n
),
and T
g
stands for the strong dual space of T
g
; observe c
g
. If T T
g
and T
g
, then T, is the dual pairing between T and . We write T
0
if all elements of T
g
are compactly supported; T
0
denotes the dual space of
T
0
; observe that a notion of support for elements of T
0
can be dened in
the usual way. In T
g
, a weakly bounded set is also a strongly bounded one
(MackeyBanachSteinhaus theorem). The spaces of distributions, ultra-
distributions, Fourier hyperfunctions, . . . , are of this kind. Furthermore, we
shall always use the notation T = T
g
if / T, where / = T, T
, or T
;
in such case T
= T
, T
, or Q(D
n
), respectively, and we say that
T
is a distribution, ultradistribution, or Fourier hyperfunction space, re-

spectively. We set T
= T T
; supp T .
We suppose that the following operations are well dened on T
g
:
Dierentiation: We assume that /
x
i
, T
g
T
g
, are continuous oper-
ators. Let k N
n
0
. Then,
_

k
k
1
x
1
. . .
k
n
x
n
T(x), (x)
_
=
_
T(x), (1)
|k|

k
k
1
x
1
. . .
k
n
x
n
(x)
_
, T
g
.
0. Preliminaries 9
Change of variables: If T T
g
and k > 0, then by denition
T(kx), (x) =
_
T(x),
1
k
n

_
x
k
__
, T
g
.
If T T
g
and h R
n
,
T(x +h), (x) = T(x), (x h), T
g
.
Furthermore, it is always assumed that k (/k), R
+
T
g
, and h
( + h), R
n
T
g
, are continuous. Consequently, by the mean value
theorem, one readily veries that both maps are indeed C
.
Let c, if is a continuous mapping from T
g
into T
g
, then we
say that is a multiplier of T
g
. Then T is by denition T, = T, .
The set of multipliers of T
g
is denoted by M
()
.
We shall say that c is a convolutor of T
g
if the mapping

, T
g
T
g
, is well dened and continuous, where

(t) = (t). We denote
by M
()
the set of convolutors of T
g
. If M
()
, then for T T
g
, (T )
is dened by T , = T,

, T
g
.
mN
, a sequence in M
()
T
g
, is called a -sequence in T
g
if
m

0, m N, and for every T
g
,
m
in T
g
, m .
We shall say that the convolution with compactly supported elements is
well dened in T
g
if a notion of support makes sense in T
g
and the following
denition applies: given S, T T
g
, where supp S = K is compact in R
n
,
their convolution is dened by
S T, = S(x) T(y), (x)(x +y) = S
x
T
y
, (x)(x +y) ,
where the function T
g
has compact support, supp =

K, so that the
set K int

K and (x) = 1, x K. For T
it coincides with the usual

denition of convolution.
We assume that T
g
contains regular elements f L
1
loc
. They are iden-
tied with f itself:
f : f, =
_
R
n
f(x)(x)dx, T
g
,
if f L
1
for every T
g
and if
m
0 in T implies f,
m
0.
There is also another situation in which we shall identify locally in-
tegrable functions with elements T T
. Let / = T, T
, or T
and
suppose / T. We identify T with f L
1
loc
if T, = f, ,
for all /. In such a case we simply write T = f. For example,
T(x) = xe
x
2
sin(e
x
2
) o
(R) is dened in this way, and not as a regular

element of o
(R) in the sense described above.

1 S-asymptotics in F
g
1.1 Denition
Denition 1.1. Let be a cone with vertex at zero and let c be a positive
real-valued function dened on . It is said that T T
g
has S-asymptotic
behavior related to c with limit U if T(x +h)/c(h) converges weakly in T
g
to U when h , |h|, i.e. (w. lim = weak limit):
w. lim
h,h
T(x +h)/c(h) = U in T
g
(1.1)
or
lim
h,h
T(x +h)/c(h), (x) = U, , T
g
. (1.2)
If (1.1) is satised, it is also said that T has S-asymptotics and we write in
short: T(x +h)
s
c(h)U(x), h .
Remarks. 1) If is a convex cone we could use another limit in .
Let h
1
, h
2
. We say that h
1
h
2
if and only if h
1
h
2
+ ; is now
partially ordered.
For a real-valued function dened on , we write
lim
h, h
(h) = A R
if for any > 0 there exists h() such that (h) (A , A+) when
h h(), h .
If is a convex cone, then the S-asymptotics with respect to this limit
might be dened as:
lim
h, h
T(x +h)/c(h), (t) = u, , T
g
. (1.3)
In case n = 1, the limits (1.2) and (1.3) coincide. We will mostly use
Denition 1.1 in this book, for S-asymptotics dened by (1.3) see also [135].
2) If T
g
is a Montel space, then the strong and the weak topologies in
T
g
are equivalent on a bounded set. If B is a lter with a countable basis
and if w. lim
hB
T(h) = U in T
g
, then this limit exists in the sense of the
strong topology. Hence, (1.1) is equivalent to
s. lim
h,h
T(x +h)/c(h) = U in T
g
.
1. S-asymptotics in F
g
11
(Furthermore, from now on, we will omit the symbol s. for the strong
convergence).
For the rst ideas of the S-asymptotics see [3] and [146]. The starting
point of the theory is [127].
1.2 Characterization of comparison functions and limits
Proposition 1.1. Let be a convex cone. Suppose T T
g
has the S-
asymptotics T(x +h)
s
c(h)U(x), h . If U ,= 0, then:
a) There exists a function d on such that
lim
h, h
c(h +h
0
)/c(h) = d(h
0
) , for every h
0
. (1.4)
b) The limit U satises the equation
U( +h) = d(h)U, h .
Proof. a) Since U ,= 0, there exists a T
g
such that U, ,= 0. For
this and a xed h
0

lim
h, h
c(h +h
0
)
c(h)
_
T(x + (h +h
0
))
c(h +h
0
)
, (x)
_
= lim
h, h
_
T((x +h
0
) +h)
c(h)
, (x)
_
.
(1.5)
Hence, for every h
0

lim
h, h
c(h +h
0
)
c(h)
=
U(x +h
0
), (x)
U,
= d(h
0
) .
b) Now we can take in (1.5) any function T
g
instead of . Then we
have
d(h
0
)U, = U(x +h
0
), , T
g
which proves b).
We now restrict the space of generalized functions to a space of distri-
bution, ultradistribution, or Fourier hyperfunction type. So, we have the
following explicit characterization of the comparison function and limit.
Proposition 1.2. Let be a convex cone with int ,= (int is the in-
terior of ). Let T T
have S-asymptotics T(x + h)

s
c(h)U(x), h ,
where U ,= 0 and c is a positive function dened on R
n
. Then:
a) For every h
0
R
n
there exists
lim
h(h
0
+),h
c(h +h
0
)/c(h) =

d(h
0
) .
b) There exists R
n
such that

d(x) = exp( x), x R
n
.
c) There exists C R such that U(x) = C exp( x).
Proof. a) Let a int . Then there exists r > 0 such that B(a, r) .
Consequently, for every > 0, B(a, r) , as well.
We shall prove that for every h
0
R
n
and every R > 0 the set (h
0
+
) x R
n
; |x|> R is not empty. The rst step is to prove that
(h
0
+ ) is not empty.
Suppose that y B(a, r/2) . Then, for every
0
> 2|h
0
|/r > 0,
|a (h
0
+y)| |a y|+|h
0
| r ,
hence h
0
+y B(a, r) .
For a xed R > 0, we can choose such that |h
0
+ y|> R. Then
h
0
+y is a common element for (h
0
+ ), and x R
n
; |x|> R.
Now we can use the limit (1.5) when h (h
0
+) , and in the same
way as in the proof of Proposition 1.1 a), we obtain
lim
h(h
0
+), h
c(h +h
0
)
c(h)
=

d(h
0
) =
U(t +h
0
),
U,
.
From the existence of this limit, it follows:
1)

d extends d to the whole R
n
;
2)

d(0) = 1;

d C
(R
n
) and

d satises
d(h +h
0
) =

d(h)d(h
0
), h, h
0
R
n
. (1.6)
We can take h
0
= (0, . . . 0, t
i
, 0, . . . , ) R
n
in (1.6); then the limit
lim
t
i
0
d(h +h
0
)

d(h)
t
i
=

d(h) lim
t
i
0
d(h
0
)

d(0)
t
i
exists and gives
h
i
d(h) =
_

h
i
d(h)
_
h=0
d(h), for i = 1, . . . , n.
g
13
We introduce the function V given by
d(h) = e
(h)
V (h),
i
=
_

h
i
d(h)
_
h=0
, i = 1, . . . , n.
Then

h
i
V (h) = 0 for every i = 1, . . . , n. Consequently, V (h) = 1, h R
n
.
For the proof of c), we now have U(x +h) = exp( h)U(x). Dierenti-
ating with respect to h and then setting h = 0, we obtain that U satises
the dierential equations

x
j
U =
j
, 1 j n. Proceeding as in the
proof of b), we obtain U(x) = C exp( x), for some C R.
Remarks.
1. We only assumed that c is a positive function. But if we know that
there exist T T
g
and U ,= 0 such that T(x +h)
s
c(h)U(t), h , then
we can nd a function c C
and with the property

lim
h,h
c(h)/c(h) = 1.
This function c can be dened as follows: c(h) = T(x + h), (x)/U, ,
where is chosen so that U, ,= 0. In this sense we can suppose, whenever
needed, that c C
, and we do not loose generality.

Similarly, we have (see also Lemma 1.4 in 1.12) that
lim
h
,h
c(h)/ c(h +x) = exp( x) in c
if is a convex cone, int ,= ,
is compact int ).
2. If in Proposition 1.2 we replace T
for the general space T
g
, then
a) and b) still hold. On the other hand, c) will not be longer true, in
general. We will show this fact in Remark 5 below by constructing explicit
counterexamples.
3. In the one-dimensional case the cone can be only R, R
+
or R
.
In all these three cases int ,= . Consequently,

d from Proposition 1.2 has
always the form

d(x) = exp(x), where R.
Let us write c(x) = L(e
x
) exp(x), x R. We will show that L is a
slowly varying function. Proposition 1.2 a) gives us the existence of the
limit
lim
h,h
L(exp(h +h
0
))/L(exp(h)) = 1, h
0
R.
If = R
+
, then
lim
xR
+
,x
L(xp)/L(x) = 1, p R
+
and this denes a slowly varying function (cf. 0.3). Thus if T T
g
(R)
and T(x + h)
s
c(h)U(x), in R
+
, with U ,= 0, then it follows that c has
the form c(x) = exp(x)L(exp(x)), x a > 0, where L is a slowly varying
function at innity. Similarly, if = R
, then L is slowly varying at the

origin, while if = R, then L is slowly varying at both innity and the
origin.
4. The explicit form of the function c given in 3. is not known in
the n-dimensional case, n 2. This problem is related to the extension
of the denition of a regularly varying function to the multi-dimensional
case ([135], [162]) and with certain q-admissible and q-strictly admissible
functions ([192]).
5. As mentioned before, c) in Proposition 1.2 does not have to hold for
a general space T
g
. From the proof of Proposition 1.2, we can still obtain
the weaker conclusion U(x + h) = exp( h)U(x), h R
n
, which in turn
implies the dierential equations (/x
j
)U =
j
. For distribution, ultra-
distribution, and Fourier hyperfunction spaces, these dierential equations
imply that U must have the form c) of Proposition 1.2. However, the latter
fact is not true in general. We provide two related counterexamples below.
Let
/
0
(R) = C
(R); lim
x
(m)
(x) exists and is nite, m N
0
,
it is a Frechet space with seminorms:
k
() = sup
x(,k],mk
[
(m)
(x)[ .
Let us rst observe that the denition of /
0
(R) does not tell all the true
about its elements. Notice that if /
0
(R), then for m = 1, 2, . . . , we
have lim
x
(m)
(x) = 0, while lim
x
(x) may not be zero. Indeed,
the proof is easy, it is enough for m = 1, if lim
x
(x) = M, then
(x) = (0) +
_
x
0
(t)dt (0) +Mx, x ,

but since has limit at , then M = 0. So, we have
/
0
(R) = C
(R); lim
x
(x) exists and lim
x
(m)
(x) = 0, m N .
g
15
Its dual space /
0
(R) contains a generalized function concentrated at
which contradicts c) of Proposition 1.2. Dene the Dirac delta concentrated
at by
, := lim
x
(x), /
0
(R) .
Notice that the constant multiples of
are the only elements of /
0
(R)
satisfying the dierential equation U
= 0. This generalized function is

translation invariant, i.e.,
( + h) =
, for all h R; in particular,

it has the S-asymptotics
(x +h)
s

(x), h = R.
Therefore, we have found an example of a non-constant limit for S-
asymptotics related to the constant function c(h) = 1.
We can go beyond the previous example and give a counterexample
for the failure of conclusion c) of Proposition 1.2 with a general c and S-
asymptotics in T
g
. Let c(h) = exp(h)L(exp h), where L is slowly varying
at innity and R. We assume that c is C
and, for all m N,

c
(m)
(h)
m
c(h), h (otherwise replace c by c given in Remark 1).
Next, we dene
/
c
(R) = C
(R); lim
x
(m)
(x)
c(x)
exists and is nite, m N
0
.
It is a Frechet space with seminorms:
k,c
() = sup
x(,k],mk
[
(m)
(x)[
c(x)
.
Note that /
c
(R) = c(x) /
0
(R), consequently,
(x) C
c(x), x ,
where C
= lim
x
(x)/c(x). An inductive argument shows that
for all m,
(m)
(x) (1)
m
m
C
c(x), x . Set now g

c,
=
(1/c(x))
c
(R), a generalized function concentrated at innity
and given by
g
c,
, =
_
(x),
(x)
c(x)
_
= lim
x
(x)
c(x)
, /
c
(R) .
It is easy to show that the constant multiples of g
c,
are the only element
of /
c
(R) satisfying the functional equation U(x +h) = exp(h)U(x) (and
hence the dierential equation U
= U); in particular,
g
c,
(x +h)
s
e
h
g
c,
(x), h = R.
In addition, there is an innite number of elements of /
c
(R) having S-
asymptotics in the cone = R
+
related to c(h) with limit of the form
Cg
c,
(x), C R. In fact, consider
(m)
, the derivatives of the Dirac
delta concentrated at the origin. We have that
(m)
(x +h)
s
c(h)
m
g
c,
(x), h = R
+
,
since
lim
h
(m)
(x +h), (x)
_
c(h)
= lim
h
(1)
m
(m)
(h)
c(h)
=
m
g
c,
(x), (x) .
1.3 Equivalent denitions of the S-asymptotics in F
Theorem 1.1. Let T T
and let int ,= . The following assertions are

equivalent:
a)
w. lim
h,h
T(x +h)
c(h)
= U(x) = M exp(x) in T
, M ,= 0 . (1.7)
b) For a -sequence
m
m
(cf. 0.6) there exists a sequence M
m
m
in
R, such that M
m
M ,= 0, m , and
w. lim
h,h
(T
m
)(x +h)
c(h)
= M
m
exp(x), in T
, uniformly in m N.
(1.8)
c) For a -sequence
m
m
, (cf. 0.6),
lim
h,h
(T
m
)(h)
c(h)
= p
m
, m N, (1.9)
where p
m
,= 0 for some m, and for every T,
sup
h,h0
(T )(h)
c(h)
< . (1.10)
If T
= T
or T
= T
the following assertion is also equivalent to a).

d) (T( + h)/c(h)) converges to (U )(h) in c(R
n
), for each
T ( T
).
g
17
Proof. a) b). Let
m
m
be a -sequence. For any T,
m
;
m N is a compact set in T. We have by the properties of the convolution
(cf. 0.6) and the BanachSteinhaus theorem
lim
h,h
_
(T
m
)(x +h)
c(h)
, (x)
_
= lim
h,h
(T
m
)(x), (x h)
c(h)
= lim
h,h
T, (
m
)( h)
c(h)
= lim
h,h
_
T(x +h)
c(h)
, (
m
)(x)
_
= Me
x
, (
m
)(x) = M
m
e
x
, (x) , (1.11)
uniformly in m. Now (1.11) implies (1.8) and b).
b) a). Let T and
a
m,h
=
_
(T
m
)(x +h)
c(h)
, (x)
_
=
(T (
m
))(h)
c(h)
, m N, h .
We have a
m,h
a
m
, h , |h|, uniformly for m N, where
a
m
= M
m
exp( x), (x), m N,
a
m
a = Mexp( x), (x), m .
Also a
m,h
a
h
, m , where
a
h
=
_
T(x +h)
c(h)
, (x)
_
, h .
This implies a
h
a, h , |h|, what is in fact a).
a) c). From (1.7) it follows that (T )(h)/c(h) converges for every
T, when h , |h| . Hence, (T )(h)/c(h) is bounded, h
, |h| 0; (1.9) follows directly from (1.7).
c) a). First, we shall prove that the set G =
m
( +x), m N, x
R
n
is dense in T. Suppose that T T
and that
T,
m
( +x) = 0, m N, x R
n
.
It follows that (T

m
)(x) = 0, m N, x R
n
. Then, for any
T, T
m
, = 0, m N, and consequently
T, = lim
m
T,
m
= lim
m
T
m
, = 0 .
This implies that T = 0 and hence, the set G is dense in T by the Hahn-
Banach theorem. Thus, by (1.10) and the BanachSteinhaus theorem, c)
implies a).
a) d). Note that (1.1) implies the strong convergence of T( +h)/c(h)
to U in T
and T
respectively. Since the convolution in T
and T
is
hypocontinuous, it follows the following equality in the sense of convergence
in c(R
n
)
lim
h,h
_
T( +h)
c(h)

_
=
__
lim
h,h
T( +h)
c(h)
_

_
= U
in both cases ( T, respectively T
).
The paper which can be consulted for related results is [128].
1.4 Basic properties of the S-asymptotics
g
.
a) If T(x+h)
s
c(h)U(x), h , then for every k N
n
0
, T
(k)
(x+h)
s
c(h)U
(k)
(x), h .
b) Assume additionally that T
g
is a Montel space. Let g M
()
(set
of multipliers of T
g
(cf. 0.6)); let c, c
1
be positive functions. If for every
T
g
, (g(x + h)/c
1
(h))(x) converges to G(x)(x) in T
g
when h
, |h| and if T(x + h)
s
c(h)U(x), h , then g(x + h)T(x + h)
s
c
1
(h)c(h)G(x)U(x), h .
c) If T T
0
and supp T is compact, then T(x + h)
s
c(h) 0, h ,
for every positive function c.
d) Suppose that T
g
is a Montel spaces in which the convolution with
compactly supported elements is well dened and hypocontinuous (cf. 0.6).
Let S T
g
, supp S being compact. If T(x + h)
s
c(h)U(x), h , then
(S T)(x +h)
s
c(h)(S U)(x), h .
e) Let T T
= T
and T(x+h)
s
c(h) U(x), h , in T
. Assume
that (M.2) holds. Let P(D) be an ultradierential operator of class . Then
(P(D)T) (x +h)
s
c(h) (P(D)U)(x), h , in T
.
f ) Let T T
= Q(D
n
). Let P(D) be a local operator and let T(x +
h)
s
c(h) U(x), h , |h| in Q(D
n
), then (P(D)T)(x + h)
s
c(h) (P(D)U)(x), h , |h| in Q(D

n
), as well.
g
19
Proof. a) The assertion is a consequence of the denition of the deriva-
tive of a generalized function. Namely,
lim
h,h
_
T
(k)
(x +h)
c(h)
, (x)
_
= lim
h,h
_
T(x +h)
c(h)
, (1)
k
(k)
(x)
_
= (1)
k
U(x),
(k)
(x)
= U
(k)
, , T
g
.
b) Since T
g
is a Montel space, we have that lim
h,h
T(x +h)
c(h)
= U(x)
in T
g
with respect to strong topology. As (g(x+h)/c
1
(h) G(x))(x), h
, |h| 0, is a bounded set in T
g
, it follows:
lim
h,h
g(x +h)T(x +h)/(c
1
(h)c(h)), (x)
= lim
h,h
_
T(x +h)/c(h),
_
g(x +h)
c
1
(h)
G(x)
_
(x)
_
+ lim
h,h
T(x +h)/c(h), G(x)(x)
=
_
U(x), lim
h,h
_
g(x +h)
c
1
(h)
G(x)
_
(x)
_
+U(x), G(x)(x)
= 0 +U(x)G(x), (x) = UG, , T
g
.
c) For each T
0
there exists r
> 0 such that supp B(0, r
) =
x R
n
; |x|< r
. The support of T(x + h) is (supp T h). Thus, by

our assumption, there exists
r
such that for all h , |h|>

r
the set
(supp T h) B(0, r
) is empty and consequently T(x+h), (x) = 0, h

, |h|
r
.
d) By denition of the convolution
(S T)(x +h), (x) = (S T)(x), (x h)
= S
t
T
y
, (t)(t +y h)
= S
t
T
y
( +h), (t)(t +y) .
(1.12)
Hence, (S T)(x +h) = (S T( +h))(x).
Since in a Montel space the weak and the strong convergence are equiv-
alent, we have by (1.12)
lim
h,h
(ST)(x+h)/c(h) = lim
h,h
_
S
T( + h)
c(h)
_
(x) = (SU)(x) .
For the proof of e) and f), we note that an ultradierential operator
P(D) maps continuously T
into T
and a local operator maps continu-

ously Q(D
n
) into Q(D
n
).
Remark. From assertion a) of Theorem 1.2, a natural question arises
for spaces T
: The limit U can be a constant generalized function, hence

U
= 0. Is there a positive function c such that T
has S-asymptotics related

to this c, but with a limit dierent from zero?
In general the answer is negative as shown by the following example: Let
T be dened by x
2
+sin(exp x
2
), x R. Then, T(x +h)
s
h
2
1, h R
+
.
But T
(x) = 2x(1 + exp(x

2
) cos(exp x
2
)). The same situation is obtained
with the distribution f(x) = x
2
+xsin x.
We can now formulate an open problem: Suppose that T(x +
h)
s
c(h)U(x), h . If U(x) is a constant generalized function, then
the problem is to nd some additional conditions on T which guarantee the
existence of a function c such that T
has the S-asymptotics in related to

c.
More generally, let S T
and T = (/x
k
)S. If T(x + h)
s
c(h)U(x), h . The question is what we can say about the S-asymptotics

of S.
Recall the well known result: Let h be a real-valued function which
has the rst derivative h
(x) ,= 0, x x
0
and h(x) , x .
If a function F has its rst derivative on (x
0
, ) such that there exists
lim
x
F
(x)/h
(x) = A, then there exists lim

x
F(x)/h(x) = A, as well.
We know that an opposite assertion does not hold, and this is at the
basis of the open problem quoted above.
We give a theorem to illustrate the relation between the S-asymptotics
of a distribution and the S-asymptotics of its primitive. We refer to [114]
for the proof.
Theorem 1.3. 1) Let f, g T
(R) and for some m N, g

(m)
= f.
g
21
a) If f(x + h)
s
h
L(h) 1, h R
+
, where > 1, then g(x + h)
s
h
+m
L(h) 1, h R
+
.
b) If f(x +h)
s
exp(h)L(exp h) exp(x), h R
+
, R, and
x
_
0
exp(h)L(exp h)dh , when x , then
g(x+h)
s
_
_
h
_
0
h
m1
_
0
. . .
h
1
_
0
exp(t)L(exp t)dtdh
1
. . . dh
m1
_
_
exp(x), h R
+
.
2) Let
0
T
(R) such that

_

0
(t)dt = 1. If
lim
h
_
g
(i)
(x +h)
exp(h)L(exp h)
,
0
(x)
_
=
i
exp(x),
0
(x), i = 0, 1, . . . , m1
and
f(x +h)
s
exp(h)L(exp h)
m
exp(x), h R
+
,
then
g(x +h)
s
+
.
3) Suppose that T T
, = x R
n
; x = (0, . . . , x
k
, 0, . . . , 0) and
T = (/x
k
)S. If T(x+h)
s
c(h)U(x), h and c(h) is locally integrable
in h
k
such that
c
1
(h
k
) =
h
k
_
h
0
k
c(v)dv
k
as h
k
, h
0
k
0,
then S(x +h)
s
c
1
(h)U(x), h .
4) Suppose that S T
and that for an m 1, 2, . . . , n,

(D
t
m
S)(x +h)
s
c(h) U(x), h .
Let V T
, D
t
m
V = U and
0
T(R),
_
R
0
()d = 1. Let
lim
h,h
S(x +h)/c(h),
0
(x
m
)
m
( x) = V,
0
m
,
where x = (x
1
, . . . , x
m1
, x
m+1
, . . . , x
n
) and
m
( x) =
_
R
(x
1
, . . . , x
m
, . . . , x
n
)dx
m
, T.
Then S(x +h)
s
c(h)V (x), h .
The following theorem asserts that the S-asymptotics is a local property
if the elements of T
g
are compactly supported.
Theorem 1.4. Let T
1
, T
2
T
0
. Let the open set R
n
have the following
property: for every r > 0 there exists a
r
> 0 such that the ball B(0, r) =
x R
n
; |x|< r is in h; h , |h|
r
. If T
1
= T
2
on and
T
1
(x +h)
s
c(h)U(x), h , then T
2
(x +h)
s
c(h)U(x), h , as well.
Proof. Let T
0
with supp B(0, r). We shall prove that
lim
h,h
_
T
1
(x +h) T
2
(x +h)
c(h)
, (x)
_
= 0 . (1.13)
The complement of the set supp(T
1
(x + h) T
2
(x + h)) contains the set
h, h . By our supposition the number
r
is xed in such a way
that the sets h; h , |h|
r
contain B(0, r) and consequently
supp . Since T
1
has the S-asymptotics related to c and with the limit U,
(1.13) implies
lim
h,h
T
2
(x +h), (x)
c(h)
= lim
h,h
T
1
(x +h), (x)
c(h)
= U, .
Remark. The open set , by its property, has to contain a set

0
x
R
n
; |x|> R where
0
is an open acute cone such that
0
and R is a
positive number.
1.5 S-asymptotic behavior of some special classes of
generalized functions
1.5.1 Examples with regular distributions
1. exp(a (x +h))
s
exp(a h) exp(a x), h R
n
.
2. exp(
_
(x + h)
2
+ (x +h))
s
exphexp
_
x +
1
2
_
, h R
+
.
3. Let w S
n1
= x R
n
; |x|= 1, p = (p
1
, . . . , p
n
) R
n
+
and
= qw; q R
+
. Denote, J = k 1, . . . , n; w
k
,= 0 and =
iJ
p
i
.
Then
(x +h)
p
s
q
iJ
w
p
i
i
1, h ,
g
23
and
(1 +x +h)
p
s
q
iJ
w
p
i
i
1, h ,
((x +h)
p
= (x
1
+h
1
)
p
1
. . . (x
n
+h
n
)
p
n
) .
4. For a slowly varying function L(t), t > 0 we have
L(t +h)
s
L(h) 1, h R
+
.
Namely,
lim
h
L(t +h)/L(h), (t) = lim
h
r
_
r
(t)L(t +h)/L(h)dt
= lim
q
e
r
_
e
r
(log y)L(log(yq))/L(log q)
dy
y
=
_
R
(t)dt, T(R) .
We used above that L(log y) is also a slowly varying function (cf. 0.3)
and that L(log(uh))/L(log h) converges uniformly to 1 as h if u stays
in compact intervals [
1
,
2
], 0 <
1
<
2
< .
5. Let g L
1
(R). A distribution dened by g has the S-asymptotic behavior
related to c = 1 and with limit U = 0. To show this, let R and
T(t) =
t
_
g(x)dx, t R.
Then, T(t + h)
s
1
g(x)dx, h R
+
. By Theorem 1.2 a), g(t + h)
s
1 0, h R
+
.
1.5.2 Examples with distributions in subspaces of T
6. Let f
, R, be given as in 0.4. It denes a distribution in T
[0,)
.
For = k, k = 0, 1, . . . , f
k
=
(k)
. Since
(k)
is supported by 0,
it has the S-asymptotics zero related to every c > 0 (cf. Theorem 1.2 c)).
For > 0,
f
(x +h)
s
(1/())h
1
1, h R
+
.
In the case < 0, ,= 1, 2, . . . , f
(x) = (1/()) Pf(H(x)x

1
),
where Pf is the nite part or partie nie (see [146], pp. 4143).
Note, supp (x h) = supp + h, T(R). Thus for T(R), we
can nd h
0
such that supp (x h) (a, ), a > 0, h h
0
. Then
lim
h
f
(x)/h
1
, (x h) = lim
h
_
a
1
()
_
x
h
_
1
(x h)dx
= lim
h
_
R
1
()
_
u +h
h
_
1
(u)du =
_
R
1
()
(u)du .
Hence, f
(x +h)
s
h
1
1
()
, h R
+
, < 0, ,= 1, 2, . . .
7. If T o
, then there exists a real number k

0
such that T has S-
asymptotic behavior related to c(h)|h|
k
0
, where c(h) tends to innity as
[h[ , h R
n
and with limit U = 0.
By Theorem VI, chapter VII in [146], there exists a number k
0
such
that the set of distributions T( +h)/(1 +|h|
2
)
k
0
/2
; h R
n
is bounded
in T
. Hence, this set is weakly bounded and

_
T(x +h)/(c(h)|h|
k
0
, (x)
_
=
(1 +|h|
2
)
k
0
/2
c(h)|h|
k
0
_
T(x +h)
(1 +|h|
2
)
k
0
/2
, (x)
_
tends to zero as |h|.
8. Let T T
(R) have the following property:

For a -sequence
m
m
there is a sequence p
m
m
in R, such that
p
m
p ,= 0, m , and
lim
h
(T
m
)(h)
c(h)
= p
m
, m N, ()
where the limit is uniform for m N.
Then T has S-asymptotics related to c.
We will prove that b) in Theorem 1.1 is satised, which is equivalent to
a) in the same theorem. For every compact set K R
(T
m
)(x +h)
c(h)
=
(T
m
)(x +h)
c(x +h)
c(x +h)
c(h)
p
m
exp(x), h , |h|,
uniformly for x K, because of
c(x +h)
c(h)
exp(x), |h| , uniformly
for x K (cf. Remarks after Proposition 1.2). Then, T has the S-
asymptotics related to c.
g
25
Property (*) is not equivalent to the existence of the S-asymptotics.
The next example illustrates that this condition is not necessary.
Assume that S C(R) L
1
(R) but not being bounded on R. This
function has S-asymptotics equal to zero related to c = 1 (see 5.). For T
we take 1 +S(x). Then, T(x +h)
s
1 1, h R
+
and
lim
h
[(1 +S)
m
](h) = lim
h
1 +S(x +h),

m
(x) = 1,
m
(x) = p
m
.
This limit is not uniform in m N because lim
m
[(1 +S)
m
](h) does not
necessarily exist.
9. Every distribution in T
L
p, 1 p < has S-asymptotic behavior related
to c = 1, and = R
n
, with limit U = 0. Let us show this.
By Theorem XXV, Chapter VI in [146] it follows that (T ) L
p
(R
n
)
for every T. Every derivative of (T ),

h
k
(T )(h) = (T
x
k
)(h), h R
n
, is also in L
p
(R
n
). Hence (T ) T
L
p. We know
that every element of T
L
p, 1 p < is bounded over R
n
and tends to
zero when |h| ([146], p. 199).
It has been examined in [153] how slowly the function (T) T
L
p
,
T, tends to zero as |h|.
In fact, the question is:
Let p 1. Is it possible to nd a positive function c such that c(x)
0, |x| and [(x)/c(x)[ C
, |x| R
, for every T
L
p? C
and
R
are positive constants depending on .

The answer is negative (see [153]).
A similar question can be asked for T T
L
p, 1 p < , but related to
the S-asymptotics. Precisely, whether there exists c(h) > 0, c(h) 0, |h|
, such that
[T(x +h)/c(h), (x)[ M
, |h|>
T,
where
and M
are positive constants depending on .

The answer to this question is also negative (cf. [153]).
10. Let T /
p
. Then there exists k
0
N
0
such that T has S-asymptotic
behavior with limit U = 0 related to c(h) exp(k
0
|h|
p
), where c(h) tends to
innity as |h|.
First, we prove that there exists a positive integer k, such that the set
T( +h) exp(k|h|
p
), h R
n
is bounded in T
.
We start by giving a bound for the seminorms
k
(( h)), /
p
:
k
[( h)] = sup
xR
n
,|a|k
exp(k|x|
p
)[D
a
(x h)[
= sup
xR
n
,|a|k
exp(k|x +h|
p
)[D
a
(x)[
exp(2
p
k|h|
p
) sup
xR
n
,|a|2pk
exp(2
p
k|x|
p
)[D
a
(x)[
exp(2
p
k|h|
p
)
2
p
k
().
By assumption, T is a continuous linear functional on /
p
. Note, the
sequence of norms
k
k
is increasing. Thus, there exist > 0 and k
0
N
0
such that
[T, [ 1 for /
p
,
k
0
() .
This inequality holds for all k k
0
. Hence
[T, [
1
k
(), k k
0
for every /
p
.
We know that T /
p
and that the inclusion is continuous. Let T,
Then
[exp(2
p
k|h|
p
)T(x +h), (x)[= [T(x), exp(2
p
k|h|
p
)(x h)[

1
exp(2
p
k|h|
p
)
k
[(x h)]
1
2
p
k
(), k > k
0
.
We can choose k
0
2
p
k. The set exp(k
0
|h|
p
)T(x + h); h R
n
is
bounded in T
(and weakly bounded in T
, as well).
Now, for every T :
lim
h
exp(k
0
|h|
p
)T(x +h)/c(h), (x)
= lim
h
1
c(h)
exp(k
0
|h|
p
)T(x +h), (x) = 0.
g
27
1.5.3 S-asymptotics of ultradistributions and Fourier hyperfunc-
tions Comparisons with the S-asymptotics of distributions
11. If a distribution has S-asymptotics in T
(see Denition 1.1), it has the

same S-asymptotics in T
, as well. But the opposite is not true. This is

illustrated by the following example:
T(x) = 1 +
n=1
(n)
(x n)/M
n
, x R,
has S-asymptotics in T
(M
p
)
(R), but it does not have S-asymptotics in
T
(R). We will show this. Let T

(M
p
)
(R). Then,
_

n=1
(n)
(x+hn)/M
n
, (x)
_
=
n=1
(1)
n
(n)
(nh)/M
n
0, h .
This is a consequence of the property
sup
xR
[
(n)
(x)[/k
n
M
n
0, n for every k > 0 .
Suppose that there exists a function c such that for every T(R)
n=1
(1)
n
(n)
(n h)/M
n
c(h) ,
converges, as h . Taking h = n this implies that
(n)
(0)/M
n
c(n)
converges to zero, as n , for every T(R). However, such a func-
tion c does not exist (Borels theorem). Namely, for every given sequence
M
n
c(n), n N, there exists C
0
(R) such that
(n)
(0)/M
n
c(n) ,
as n .
This example is the motivation for the following assertion.
12. Let T T
(R) be such that lim

h
T(x + h)/c(h) exists in T
(R).
Assume that for some s N and T
(R) with the property

_
R
(t)dt = 1,
_
R
t
j
(t)dt = 0, j = 1, . . . , s,
the following limit
lim
h
_
T(t +h)
c(h)
,
t
s
p
s+1
_
t
p
__
, p (0, 1],
exists uniformly in p. Then, lim
h
T(x +h)/c(h) exists in T
(R) as well.
Proof. Let T(R), h h
0
> 0. The function F
h
(t) = T(x + h +
t), (x), t R, is smooth and by the Taylor formula, we have:
F
h
(t) = F
h
(0) +t
d
dt
F
h
(0) + +
t
s1
(s 1)!
d
s1
dt
s1
F
h
(0)
+
t
s
(s 1)!
1
_
0
(1 p)
s1
F
(s)
h
(pt)dp, s 1, 0 < p 1 .
This implies
T(x +h), (x) = T(x +h +t), (x) T(x +h),
(x)

(1)
s1
t
s1
(s 1)!
T(x +h),
(s1)
(x)
(1)
s
t
s
(s 1)!
1
_
0
(1 p)
s1
T(x +h +pt),
(s)
(x)dp,
h h
0
, t R.
Multiplying both sides of the last equality by (t), integrating with respect
to t, and using Fubinnis theorem, we obtain
T(x +h), (x) = T(x +h +t), (t), (x)
(1)
s
(s 1)!
1
_
0
(1 p)
s1
T(x +h +pt), t
s
(t),
(s)
(x)dp .
Set
G(x, h, p) = T(x +h +pt), t
s
(t), x R, h h
0
, p [0, 1] .
We have G(x, h, 0) = 0, x R, h h
0
and
lim
h
G(x, h, p)/c(h) = Ce
ax
e
apt
, t
s
(t), x R, p (0, 1] ,
for some a R (cf. Proposition 1.2 a)), where the limit is uniform in
p (0, 1] and x supp . The limit function is continuous in x R and
p [0, 1], because of e
apt
, t
s
(t) 0, as p 0. Because of that, we
obtain
lim
h
1
c(h)
T(x +h +pt), t
s
(t),
(s)
(x)
=
_
lim
h
1
c(h)
T(x +h +pt), t
s
(t),
(s)
(x)
_
= Ce
ax+apt
, t
s
(t),
(s)
(x) = Ce
ax
,
(s)
(x)e
apt
, t
s
(t) .
g
29
This implies
lim
h
_
1
c(h)
T(x +h), (x)
_
=
__
lim
h
T(x +h +t)
c(h)
, (t)
_
, (x)
_

(1)
s
(s 1)!
1
_
0
(1 p)
s1
e
apt
, t
s
(t)dpCe
ax
,
(s)
(x) ,
which proves the assertion.
Such an assertion can be proved in the multi-dimensional case by ad-
justing the previous argument. But an open problem is to nd necessary
and sucient conditions for a distribution T, which has S-asymptotics in
T
, to have the S-asymptotics in T
as well.
13. We shall construct an ultradistribution out of the space of Schwartz
distributions and having S-asymptotics.
Assume (M.2) holds. Let P(D) be an ultradierential operator of class
of innite order (a
,= 0 for innitely ) (see 0.5.2). Then, P(D) is an

element of T
which is not a distribution and which has S-asymptotics in

T
equal to zero related to any c.

If T T
and T(x + h)
s
1 1, h in T
, the ultradistribution
T +P(D) is not a distribution, but (T +P(D))(x +h)
s
1 1, h in
T
:
lim
h,h
((T(x +h), (x) +P(D)(x +h), (x))
= 1, + lim
h,h
(x +h),
(1)
m
a
m
(m)
(x) = 1, , T
.
14. Let P(D) be a local operator
||0
b
, b
,= 0, [[ 0 (see
0.5.3.). The Fourier hyperfunction f = 1 + P(D) has S-asymptotics
related to c = 1 in any cone and with limit U = 1, but f is not a
distribution. For the S-asymptotics of f it is enough to prove that
lim
h,h
P(D)(x +h), (x) = 0, T
.
Since P(D) maps T
into T
,
P(D)(x +h), (x) = (x +h), P(D)(x) = (h) ,
where = P(D). By the property of the elements of T
(see 0.5.3), we
have lim
h,h
(h) = 0, for every cone .
A hyperfunction g supported by the origin is uniquely expressible as
g =

P(D), where

P(D) is a local operator. In such a way, the above
proof implies that every Fourier hyperfunction with support at 0 has
S-asymptotics with limit equal zero.
Since P(D) =
||0
b
is a distribution if and only if b
,= 0 for a
nite number of , we note that 1 +P(D) is not a distribution, but it has
the S-asymptotics related to c = 1.
We can also nd coecients b
of a local operator P(D) such that

f = 1+P(D) is not dened by an ultradistribution belonging to the Gevrey
class T
(s)
or T
{s}
, s > 1 (see [81]). For the sake of simplicity, we shall

consider the one-dimensional case. Choose P(D) such that the coecients
of P(D) are: b
n
= (n! )
(1+c
n
)
, n N, where c
n
= (log log n)
1
. With
these coecients, P(D) is a local operator. Namely,
lim
n
n
_
b
n
n! = lim
n
(n! )
1
nlog log n
= 0 .
Also, any ultradistribution in the Gevrey class s > 1, supported by 0,
is of the form
J(D) =
n=0
a
n
D
n
, necessarily with [a
n
[ Ck
n
/(n! )
s
for some constants k and C (Beurlings type) or for any k > 0 with a con-
stant C (Roumieus type). But the coecients b
n
= (n! )
(1+c
n
)
do not
satisfy these conditions, therefore P(D) cannot represent an ultradistri-
bution. Namely, since c
n
0 when n , for any s > 1, there exists n
0
such that 1 +c
n
< s, n n
0
. Thus,
(n! )
(1+c
n
)
> Ck
n
/(n! )
s
, n n
0
, k > 0 .
Consequently, P(D) does not represent an ultradistribution of Gevrey
type.
On the other hand, if we suppose that g = P(D) is an ultradistribu-
tion with support 0 in the Gevrey class s > 1, then, we would have an
ultradierential operator J
1
(D) such that
g = J
1
(D) =
n=0
e
n
D
n
, [e
n
[ Ck
n
/(n! )
s
.
g
31
But in this case J
1
(D) is also a local operator and J
1
(D) ,= P(D). This
contradicts the fact that a hyperfunction with support at 0 is given by a
unique local operator.
1.6 S-asymptotics and the asymptotics of a function
We suppose in this subsection that elements of T are compactly supported
and, as usual, that the topology in T is stronger than the topology in c.
Recall, we use the notation T
0
in this case.
Every locally integrable function f denes an element of T
0
(regular
generalized functions).
We shall compare the asymptotic behavior of a locally integrable func-
tion f and the S-asymptotic behavior of the generalized function generated
by it.
A function f has asymptotics at innity if there exists a positive function
c such that lim
x
f(x)/c(x) = A ,= 0, (in short f(x) Ac(x), x ) .
1. The following example points out that a continuous and L
1
-integrable
function can have S-asymptotics as a distribution without having an ordi-
nary asymptotics. Suppose that g L
1
(R) C(R) has the property that
g(n) = n, n N and that it is equal to zero outside suitable small intervals
I
n
n, n N. Denote by f(t) = e
t
t
_
0
g(x)dx, t R. It is easy to see that
f(t +h)
s
e
h
e
t
_
0
g(x)dx, h R
+
.
By Theorem 1.2 a) f
(t) has S-asymptotics related to e

h
and with the same
limit. But, in view of the properties of g, f
(t) = f(t) + e
t
g(t) has not the
same asymptotics (in the ordinary sense). Moreover, g can be chosen so
that f
has no asymptotics at all.

2. The following example shows that a function f can have asymptotic
behavior without having S-asymptotics with limit U dierent from zero. An
example is x exp(x
2
), x R. Suppose that exp(x
2
) has S-asymptotics
related to a c(h) > 0, h R
+
with a limit U dierent from zero. By
Proposition 1.2 c), U has the form U(x) = C exp(ax), C > 0. Then, for
every T
0
such that > 0 we have
lim
h
1
c(h)
_
exp[(x +h +h
0
)
2
](x)dx = e
ah
0
Ce
ax
, (x) .
Therefore,
e
ah
0
U, = exp(h
2
0
) lim
h
1
c(h)
_
e
(x+h)
2
e
2h
0
(x+h)
(x)dx
exp(h
2
0
)U, , for every h
0
> 0 .
But this inequality is absurd. Consequently, exp(x
2
) cannot have such an
S-asymptotic behavior.
One can prove a more general assertion.
Proposition 1.3. Let f L
1
loc
(R) T
0
(R) have one of the four properties
for > 1, > 0, x x
0
, h > 0, M > 0 and N > 0 :
a) f(x +h) M exp(h
)f(x) 0,
a) f(x +h) M exp(h
)f(x) 0,
b) 0 f(x +h) N exp(h
)f(x),
b) 0 f(x +h) N exp(h
)f(x).
Then f cannot have S-asymptotics with limit U ,= 0, but the function f can
have asymptotics.
For the proof see ([135], p. 89).
It is easy to show that for some classes of real functions f on R the
asymptotic behavior at innity implies the S-asymptotics.
Proposition 1.4. a) Let c be a positive function and let T L
1
loc
(R
n
).
Suppose that there exist locally integrable functions U(x) and V (x), x R
n
,
such that for every compact set K R
n
[T(x +h)/c(h)[ V (x), x K, |h|> r
K
,
lim
h,h
T(x + h)/c(h) = U(x), x K .
Then, T(x +h)
s
c(h)U(x), h in T
0
.
b) Let T L
1
loc
(R) have the ordinary asymptotic behavior
T(x) exp(x)L(exp x), x , R,
where L is a slowly varying function. Then,
T(x +h)
s
+
, in T
0
(R) .
g
33
Proof. a) For every T
0
lim
h,h
_
T(x +h)
c(h)
, (x)
_
= lim
h,h
_
R
n
T(x +h)
c(h)
(x)dx. (1.14)
Since supp K R
n
and T has all the listed properties, Lebesgues
theorem implies the result.
b) It is enough to use in (1.14) that L(yt)/L(t) 1, t uniformly
in y, when y stays in compact interval contained in R
+
.
A more general result is the following one ([135], pp. 8990).
Proposition 1.5. Let be a cone and let R
n
be an open set such that
for every r > 0 there exists a
r
such that B(0, r) h; h , |h|
r
. Suppose that G L
1
loc
() and it has the following properties: There
exist locally integrable functions U and V in R
n
such that for every r > 0
we have
[G(x +h)/c(h)[ V (x), x B(0, r), h , |h|
r
;
lim
h,h
G(x +h)/c(h) = U(x), x B(0, r).
If G
0
T
0
coincides with G on , then
G
0
(x +h)
s
c(h)U(x), h .
Proof. By Theorem 1.4, it is enough to proof that
lim
h,h
_
G(x +h)
c(h)
(x)dx =
_
R
n
U(x)(x)dx;
but as in the proof of Proposition 1.4 a), the exchange of the limit and the
integral sign is justied by our assumptions and Lebesgues theorem.
The following proposition gives a sucient condition under which the
S-asymptotics of f L
1
loc
(R), in T
(R), implies the ordinary asymptotic

behavior of f.
1
loc
(R), c(h) = h
L(h), where > 1

and L be a slowly varying function. If for some m N, x
m
f(x), x >
0, is monotonous and f(x + h)
s
c(h) 1, h R
+
,, in T
(R), then
lim
h
f(h)/c(h) = 1. If we suppose that L is monotonous, then we can omit
the hypothesis > 1.
For the proof see [115].
1.7 Characterization of the support of T F
0
We suppose in this subsection that the topology in T
0
is dened in such
a way that a sequence
n
in T
0
converges if and only if there exists a
compact set K R
n
and
(k)
m
converges to
(k)
uniformly on K for every
k N
n
0
as m .
We already proved in Theorem 1.4 a relation between the support of a
distribution and its S-asymptotics. Now, we shall complete this result.
We need a property of the S-asymptotics given in the next lemma.
Lemma 1.1. Let be a cone and let

be a convex cone (it is partially
ordered). A necessary and sucient condition that for every c(h) > 0,
h
a) w. lim
h,h
T(x +h)/c(h) = 0 in T
0
,
b) w.lim
h,h
T(x +h)/c(h) = 0 in T
0
,
is that for every T
0
the following holds:
In case a): There exists () > 0 such that
T(x +h), (x) = 0, |h| (), h . (1.15)
In case b): There exists h

such that
T(x + h), (x) = 0, h h
, h

. (1.16)
Proof. We have to prove only that the condition is necessary in both
cases. It is obvious that the condition is sucient. Let us suppose the
opposite, that is, the condition is not necessary. Then, we could nd a
sequence a
m
m
in such that in case a) |a
m
| in , and in case b)
a
m
in

, (m ) and that
T(x +a
m
), (x) =
m
,= 0, m N.
Let nd c such that c(h) =
m
for h = a
m
, m N. Clearly, for such
a c(h) the function h T(x + h)/c(h), (x) cannot converge to zero as
|h| , h , or h , h

. This contradicts our supposition that
the S-asymptotics equals zero in both cases.
g
35
Theorem 1.5. Let be a cone and let T T
0
. A necessary and sucient
condition that for every r > 0 there exists
r
such that the sets
supp T B(h, r), h , |h|
r
are empty
is that T(x +h)
s
c(h) 0, h for every positive function c on .
Proof. Theorem 1.2 c) and Theorem 1.4 assert that the condition in
Theorem 1.5 is necessary. We have to prove only that this condition is also
sucient.
Let us suppose that
w. lim
h,h
T(x +h)/c(h) = 0 in T
0
.
Let T
0
. By Lemma 1.1 a), we know that there exists
0
() =
inf (), where () is such that (1.15) holds. We shall prove that the
set
0
(); T
0
, supp K is bounded for every compact set K
R
n
. Let us suppose the opposite. Then we could nd a sequence
k
k
in
T
0
, supp
k
K, k N, and a sequence h
k
k
in , |h
k
| such that
T(t +h
k
),
p
(t) = A
k,p
=
_
a
k
,= 0, p = k
0, p < k .
We give the construction of sequences
k
and h
k
. Let
k

T
0
, supp
k
K, k N, be such that
0
(
k
)
k
is a strictly increasing sequence which
tends to innity. Then, there exist h
k
k
in and
k
> 0, k N, such that
0
(
k1
) +
k
|h
k
|
0
(
k
)
k
. Now, we shall construct the sequence
p
(t)
p
in T
0
, supp
p
K, p N, for which we have
T(t +h
k
),
p
(t) =
_
0, p ,= k
a
k
, p = k .
()
Put
p
(t) =
p
(t)
p
1
1
(t)
p
p1
p1
(t), p > 1, t R
n
.
The sequence
p
i
i
will be determined in such a way that
p
(t) satises (*)
the sought property, p N.
It is easy to see that T(t + h
k
),
k
(t) = a
k
and T(t + h
k
),
p
(t) =
0, k > p. For a xed p and k < p we can nd
p
i
, i = 1, . . . , p 1, such that
for k = 1, . . . , p 1,
0 = T(t +h
k
),
p
(t) = A
k,p

p
1
A
k,1

p
p1
A
k,p1
.
Hence
p
1
A
k,1
+ +
p
p1
A
k,p1
= A
k,p
, k = 1, . . . , p 1, p > 1 .
Since A
k,k
,= 0 for every k, the system always has a solution.
We introduce a sequence of numbers b
k
k
,
b
k
= sup2
k
[
(i)
k
(t)[; i k, k N.
Then, the function
(t) =
p=1
p
(t)/b
p
, t R
n
, is in T
0
and supp K .
Thus, we have
T(t +h
k
), (t) =
p=1
T(t +h
k
),
p
(t)/b
p
= a
k
/b
k
.
Now, if we choose c(h) such that c(h
k
) = a
k
/b
k
, k N, then
_
T(t +h), (t)
c(h)
_
does not converge to zero when |h|, h .
This proves that for every compact set K there exists a
0
(K) such that
T(t +h), (t) = 0, |h|
0
(K), h , T
0
, supp K.
It follows that T(t +h) = 0 over B(0, r), |h| (r), h and T(t) = 0
over B(h, r), |h| (r), h .
Remark. The condition of Theorem 1.5 implies that the support of
T has the following property: The distance from supp T to a point h
, d(supp T, h), tends to innity when |h|, h .
The next proposition shows that if in Denition 1.1 we take the limit
(1.3) instead of the limit (1.2), then a more precise result is obtained.
0
and

be an acute, open and convex cone
(partially ordered, see Remark 1 after Denition 1.1). A necessary and
sucient condition for
supp T C
R
n(a +

) for some a

(1.17)
is that
lim
h
,h
T(x +h)/c(h) = 0 in T
0
for every c(h) (1.18)
(C
R
nA = R
n
A).
g
37
Proof. If (1.17) holds, then for any ball B(0, 1), there exists a h
r

such that B(h, r) (a +

) for h h
r
, h

. This implies (1.18), and by
Lemma 1.1, (1.17) follows.
Let us suppose now that (1.16) and consequently (1.18) hold, but for
any a

, suppT , C
R
n(a +

). We x such an a = a
0
> 0. There exists
an a
1
(a
0
+

) supp T. Since supp T , C
R
n(2a
1
+

), there exists an
a
2
(2a
1
+

supp T). In such a way, we construct a sequence a
k
k
in

such that a
k
(ka
k1
+

) supp T and a
k
ka
0
, k N.
Since a
k
supp T, k N, it follows that there exists a sequence
k
k
in T
0
such that supp
k
B(0, 1), k N, and
T(x +a
k
),
k
(x) ,= 0, k N.
We put now:
c
k,i
= T(x +a
i
),
k
(x), k, i N;
b
k
= supp2
k
[
(j)
(x)[; j k, x R
n
, k N.
We have to prove that there exists a sequence c
k
k
such that c
k

1, k N and
k1
c
k,i
/(b
k
c
k
) ,= 0, i N. (1.19)
First, we notice that c
k,k
,= 0. If
k=1
c
k,1
/b
k
= 0, then we take c
1
> 1,
and if this series is dierent from 0, then we put c
1
= 1. Let i = 2. If
c
1,2
/(b
1
c
1
) +
k2
c
k,2
/b
k
= 0 (,= 0), we take c
2
> 1 (c
2
= 1)
such that
c
1,1
/(c
1
b
1
) +c
2,1
/(c
2
b
2
) +
k3
c
k,1
/b
k
,= 0 .
Let i = 3,
c
1,3
/(b
1
c
1
) +c
2,3
/(b
2
c
2
) +
k=3
c
k,3
/b
k
= 0 or ,= 0 .
Then we take c
3
> 1 and c
3
= 1 respectively such that
c
1,1
/(b
1
c
1
) +c
2,1
/(b
2
c
2
) +c
3,1
/(b
3
c
3
) +
k=4
c
k,1
/b
k
,= 0,
c
1,2
/(b
1
c
1
) +c
2,2
/(b
2
c
2
) +c
3,2
/(b
3
c
3
) +
k=4
c
k,2
/b
k
,= 0.
Continuing in this way, we construct a sequence c
k
k
for which (1.19)
holds.
Let us put
k
(x) =
k
(x)/(b
k
c
k
), k N and =
ki
k
.
From the properties of sequences b
k
k
and c
k
k
, we can easily show
that
N
k=1
k
in T
0
when N .
Relation (1.18) implies
T(x +a
i
), (x) =
k=1
c
k,1
/(b
k
c
k
) ,= 0, i N.
We obtain that (1.18) does not hold for . This completes the proof.
In Theorem 1.6 the support of T can be just C
R
n(a +

). The question
is: Is it possible to obtain a similar proposition for the S-asymptotics given
by Denition 1.1? This question is analyzed in the next examples. We take
T
0
= T in which the S-asymptotics by the weak and strong convergence
are equivalent.
Examples. i) Let T(x, y) =
m1
m(xm, y). The given series converges
in T
(R
2
). Since for a T(R
2
), supp B(0, r), we have
lim
n
m=1
m(x m, y), (x, y) =
1mr
m(m, 0).
It follows that T is a distribution on R
2
(see Theorem XIII, Chapter 3
in [146]).
Let us remark that the support of T lies on the half line (p, 0)
R
2
; p > 0. We can take for the cone R
2
+
(, ) R
2
; > 0, >
0. It is a convex, open and acute cone in R
2
. We shall show that the limit
lim
h,h
T(u +h), (u)
g
39
does not exist. To do this, it is enough to take the limit over the half line
(0,
0
) + ,
0
> 0, which belongs to . If we choose such that
> 0 and (0,
0
) = 1, then for h = (p,
0
)
T(u +h), (u) =
m1
m(mp,
0
) p.
Note, if |h| , then p , as well. Consequently, the answer to the
posed question is negative.
ii) The following example shows that if
lim
h ,h
T(x +h)/c(h) = 0 in T
for every positive c(h), and every = pw, p > 0, w , then this does
not imply that
lim
h,h
T(x +h)/c(h) = 0 in T
.
Let us remark that both limits on a , when h or |h|, are equal.
Let T be given by T(x, y) =
m1
m(xm, y
1
m
) on R
2
. The support
of T lies on the curve (x, 1/x); x > 0. Let = R
2
+
, w = (cos , sin ), 0 <
<

2
and pw; p > 0. Then, for a T(R
2
)
T(x+h
1
, y+h
2
), (x, y) =
m1
m
_
mh
1
,
1
m
h
2
_
0, |h|, h .
In order to show that
lim
h,h
T(x +h
1
, y +h
2
), (x, y)
does not exist, take h to belong to = (x, a); x > 0 for a xed a > 0, as
we did in example i).
Remarks. As a consequence of Theorem 1.5 and Theorem 1.6, we have
some results concerning the convolution product (see [135], p. 102).
Let
G
1
f C
; supp f C
R
nh ; |h|
f
,
G
2
f C
; supp f C
R
nh

; h h
f
.
Corollary 1.1. For a xed T T
, the convolution T maps T into G

1
if and only if the support of T has the property given in Theorem 1.5.
Corollary 1.2. For a xed T T
and for a convex, open and partially

ordered cone

the convolution T maps T into G
2
if and only if supp T
C
R
n(a +

) for some a

.
1.8 Characterization of some generalized function spaces
Theorem 1.7. A necessary and sucient condition for a distribution T to
belong to:
a) c
is that T(x +h)

s
c(h) 0, h R
n
for every positive c.
b) O
c
is that T has S-asymptotic behavior related to every c(h) =
|h|
, R
+
and with limit U = 0.
c) B
is that T has S-asymptotic behavior related to every positive

c, c(h) , |h|, and with limit U = 0.
Proof. a) It is a direct consequence of Theorem 1.5.
b) It is enough to apply Theorem IX Chapter VII in [146] which states:
A necessary and sucient condition that a distribution T belongs to O
c
is that for every T the function (T )(h) is continuous and of fast
descent at innity. (see 0.5.1).
c) By Theorem XXV, Chapter VI in [146] a distribution T B
if and
only if T L
(R
n
) for every T. Suppose that T B
. Then for
every T and c(h) , |h|
lim
h,h
_
T(x +h)
c(h)
, (x)
_
= lim
h,h
(T )(h)
c(h)
= 0 .
Suppose that (T )(h)/c(h) 0, |h| , for every T and for
every c, c(h) as |h| . We will show that (T )(h) L
(R
n
)
for every T. Then, by the same theorem, it follows that T B
.
Let us assume the contrary, i.e., that (T
o
)(h) is not bounded for
a
o
T. Then, we could nd two sequences h
m
m
in R
n
and c
m
m
in R such that [c
m
[ as m , |h
m
| m and (T
o
)(h
m
) = c
m
.
Now, for c
o
(h) such that c
o
(h
m
) =
_
[c
m
[, m N, the limit < T(x +
h)/c
o
(h),
o
(x) > would not exist, as |h| . This is in a contradiction
with our assumption that T has the S-asymptotics related to every c(h)
which tends to innity as |h|.
Proposition 1.7. a) If for every rapidly decreasing function c, T has the
S-asymptotic behavior related to c
1
and with limit U
c
(U
c
= 0 is included),
then T o
.
g
41
b) If for every rapidly exponentially decreasing function c (for every
k > 0, c(h) exp(k|h|) 0, |h|) a distribution T has the S-asymptotic
behavior related to c
1
with limit U
c
(U
c
= 0 is included), then T /
1
.
Proof. a) Let c be given. There exists
0
such that for every h, |h|
0
,
and for every T
[< T(x +h) c(h), (x) > [ [< U
c
, > [+
.
Therefore, the set T(x + h) c(h); h
0
is weakly bounded and thus
bounded in T
. By Theorem VI 4
o
, Chapter VII in [146] if c(h)T(x +
h); |h|>
0
is bounded in T
, for every c of fast descent, then T o
.
b) The proof is similar to that of a), if we use the following theorem
proved which will be showed below (cf. Theorem 1.15):
Let T T
. If for every rapidly exponentially decreasing function r on

R
n
the set r(h)T(x +h); h R
n
is bounded in T
, then T /
1
.
1.9 Structural theorems for S-asymptotics in F
In the analysis of the S-asymptotics and its applications, it is useful to

know analytical expression for generalized functions having S-asymptotics,
especially if it is given via continuous functions and their derivatives. The
next theorems are of this kind. These types of theorems are usually referred
as structural theorems.
Our rst result concerns the one-dimensional case, a renement will be
obtained in Theorem 1.9 below.
Theorem 1.8. Let R and let L be a slowly varying function. Suppose
that f T
(R). If f(x +h)

s
c(h) e
x
, h R
+
, then there is an m
0
N
such that for every m m
0
the following holds:
(1) Let ,= 0 and c(h) = e
h
L(e
h
), h > 0. Then there are g
m,i

C(1, ), i = 0, 1, . . . , m such that
f(x) =
m
i=0
g
(i)
m,i
(x), x (1, ),
and
g
m,i
(x) C
i
x
m
exp(x)L(exp x), x , i = 0, . . . , m,
where C
i
are suitable constants.
(2) Let = 0 and c(h) = h
L(h), h > 0. Then,

a) If > 1, then there is F
m
C(1, ) such that f = F
(m)
m
and
F
m
(x) x
m+
L(x), x ;
b) If 1, then there are f
m,i
C(1, ) and A
m,i
,= 0, i =
0, 1, . . . , m, such that
f
m,i
(x) A
m,i
x
m+i
L(x), i = 0, 1, . . . , m
and
f(x) =
m
i=0
f
(mi)
m,i
(x), x (1, ) .
Proof. Case ,= 0. By the remark 1. after Proposition 1.2, there exists
a function c C
(R) such that c(h)/ c(x + h) exp(x), h in R

and c
(i)
(x +h)/c(h)
i
exp(x), i N
0
.
Since f(x + h)/c(h) exp(x) with strong convergence and the set
c(h)(x)/ c(x +h); h A is bounded in T(R), we can then use Theorem
XI, Chapter III in [146] to obtain
lim
h
f(x +h)
c(x +h)
, (x) = lim
h
f(x +h)
c(h)
,
c(h)
c(c +h)
(x)
= 1, (x), T(R) .
Let C
, (x) = 0 for x < 0 and (x) = 1 for x > 1. We have

(x +h)f(x +h)
c(x +h)
1 in T
(R) as h .
Thus, (x+h)f(x+h)/ c(x+h); h > 0 is a bounded subset of T
(R). This
implies that this set is bounded in o
(R) as well (cf. Theorem XXV, Chap-

ter VI in [146]). Since T(R) is dense in o(R), by the BanachSteinhaus
theorem, we obtain
(x +h)f(x +h)
c(x +h)
1 in o
(R) as h , i.e.,
lim
h
_
(f/ c)(x +h)
d(h)
, (x)
_
= 1, , for every o(R),
where d(h) = 1, h > A. We shall now borrow two theorems about quasi-
asymptotics (see Denition 2.2) which will be shown later. Since the S-
asymptotics in o
+
with > 1 implies the quasi-asymptotics of f/ c
g
43
(see Theorem 2.47), the structural theorem for the quasi-asymptotics (see
Theorem 2.2) implies that there is m
0
N such that for every m > m
0
there is F
m
C(R) such that
(f/ c)(x) = F
(m)
m
(x), x R,
and F
m
(x) x
m
as x . Thus, we obtain
f(x) = c(x)F
(m)
m
(x), x (1, ).
The Leibnitz formula implies
f(x) =
m
i=0
_
m
i
_
(1)
i
( c
(i)
(x)F
m
(x))
(mi)
, x (1, ).
Since
c
(i)
(x +h)
c(h)

i
e
x
, h , x R,
we obtain
c
(i)
(h)
i
c(h), h .
This implies the result when ,= 0 and c(h) = e
h
L(e
h
), h > 0.
In case = 0 and c(h) = h
L(h), h > 0, > 1, the S-asymptotics of

f related to c(h) = h
L(h), h > 0, > 1, implies the quasi-asymptotics

of f related to the same c(h) (see Theorem 2.47). The assertion follows
now from Theorem 2.2.
If 1, then we take k > 0 such that k + > 1. With as in the
preceding proof and by Theorem 1.2 b), we have
(1 + (x +h)
2
)
k/2
(x +h)f(x +h)
s
h
k+
L(h) 1, h R
+
.
By the same arguments as in the preceding proof, we have that there is
m
0
N such that for every m > m
0
there is an F
m
C(R), supp F
m

(0, )
F
m
(x) x
+k+m
L(x), x
(1 +x
2
)
k/2
(x)f(x) = F
(m)
m
(x), x R.
Thus, for x (1, ),
f(x) =
m
i=0
(1)
i
_
m
i
___
1
(1 +x
2
)
k/2
_
(i)
F
m
(x)
_
(mi)
.
The result now follows from the fact
_
1
(1 +x
2
)
k/2
_
(i)
C
i
x
ki
, x ,
where C
i
,= 0 are suitable constants, i = 0, . . . , m.
We can show a more general result which describes the precise structure
of a distribution having S-asymptotics.
Theorem 1.9. If T T
has S-asymptotics related to the open cone and

the continuous and positive function c(h), h , then for the ball B(0, r)
there exist continuous functions F
i
, [i[ m, such that F
i
(x + h)/c(h) con-
verges uniformly for x B(0, r) when h , |h| , and the restriction
of the distribution T on B(0, r)+ can be given in the form T =
|i|m
D
i
F
i
.
We need the following lemma for the proof of the theorem:
Lemma 1.2. Let T T
, T(x + h)
s
c(h) U(x), h . Then, for an
open ball B(0, r) and a relatively compact open neighborhood of zero in
R
n
, there exists an m N
0
such that for every , T
m
the function
(T)(x) is continuous for x B(0, r)+. Moreover, the set of functions
(T
h
)(x); h converges uniformly for x B(0, r) to (U)(x),
as h , |h|; T
h
= T(x + h)/c(h).
Proof. Suppose that T has S-asymptotics related to c(h). Then, the set
T(x + h)/c(h) T
h
; h is weakly bounded in T
and consequently,
bounded in T
. A necessary and sucient condition that a set B
is
bounded in T
is: for every T the set of functions T ; T B
is bounded on every compact set K R

n
(see 7, Chapter VI in [146]).
Moreover T ; T B
denes a bounded set of regular distributions.

Let Cl() = K (Cl() is the closure of ); K is a compact set. For a
xed C
0
, supp K, the linear mappings (T
h
) , h ,
are continuous mappings of T
K
into c because of the separate continuity
of the convolution. Since the set T
h
; h is a bounded set in T
, for
every ball B(0, r) the set of mappings (T
h
) ; h is the set
of equicontinuous mappings of T
K
into L
B
, where B = B(0, r). Now there
exists an m N
0
such that the linear mappings (, ) T
h
which
map T
K
T
K
into L
B
can be extended to T
m
T
m
in such a way that

(, ) T
h
, h , are equicontinuous mappings of T
m
T
m
into
L
B
(see for example the proof of Theorem XXII, Chapter VI in [146]).
g
45
We proved that for every , T
m
and every h the functions

x (T
h
)(x) are continuous in x B(0, r). From the relation
(T
h
)(x) = (T )(x + h)/c(h) and from the properties of c it
follows that y (T )(y) is a continuous function for y B(0, r) +
and , T
m
.
It remains to prove that T
h
converges to U as |h|, h
, in L
B
for , D
m
. We know that T is a dense subset of T

m
, m 0.
We can construct a subset A of T
K
to be dense in T
m
. The set of functions

T
h
converges in L
B
for , A, when h , |h|. Taking care of
the equicontinuity of the mappings T
m
T
m
into L
B
, dened by T
h
,
we can use the BanachSteinhaus theorem to prove that T
h
converges
in L
B
when h , |h|.
Proof of the Theorem. We shall use (VI, 6; 23) from [146].
2k
(E E T) 2
k
(E T) + ( T) = T , (1.20)
where E is a solution of the iterated Laplace equation
k
E = ; , T
.
We have only to choose the natural number k large enough so that E belongs
to T
m
. Now, it is possible to take F

1
= E E T, F
2
= E T and
F
3
= T. All of these functions are of the form F
i
= T
i
,
i
, ,
T
m
, i = 1, 2, 3.
The following holds:
F
i
(x +h)/c(h) = (F
i
(x)/c(h))
h
= (T
i

i
, )(x)
h
/c(h)
= ((T(x)
h
)/c(h) (
i

i
) = (T
h

i

i
)(x), x B(0, r), h .
Hence, by Lemma 1.2 it follows that F
i
(x + h)/c(h) converges uniformly
for x B(0, r) when h , |h|.
Consequences of Theorem 1.9
a) If the functions F
i
, [i[ m, have the property given in Theorem 1.9
and if = R
n
, then the regular distributions dened by the functions F
i
/c
have the S-asymptotic behavior related to c
1
(h) 1.
b) If T o
, then all functions F

i
, [i[ m, are continuous for x
B(0, r) + , and of slow growth (F
i
= (1 + r
2
)
q
f
i
, [i[ m, q R, where
r = |x| and f
i
, [i[ m, are continuous and bounded functions).
c) The converse of Theorem 1.9 is also true. Therefore, it completely
characterizes those distributions having S-asymptotics.
Proof. a) Suppose that = R
n
. Then by the properties of F
i
, [i[ m,
the functions F
i
/c are continuous and F
i
(x)/c(x) converge to the numbers
C
i
as |x|. Now, for [i[ m we have
lim
h
F
i
(x +h)/c(x +h), (x)
= lim
h
_
R
n
(F
i
(x +h)/c(x +h))(x)dx = C
i
, , T.
b) If T o
, then there exists a q R such that the set of distributions

T(x + h)/(1 + |h|
2
)
q
; h R
n
= W is bounded in T
(Theorem VI,
Chapter VII in [146]). We can now repeat the rst part of the proof of
Lemma 1.2 but with c(h) = (1 + |h|
2
)
q
, h = R
n
. In this may we
obtain that there exists a p N
0
such that for , T
p
and x R
n
the
function x (T)(x) is continuous and (T)(x)/(1+|x|
2
)
q
, x R
is bounded. It remains only to choose the number k in (1.20) large enough
so that E T
max(m,p)
.
c) It follows directly from Theorem 1.2 a) and Proposition 1.5.
We can also characterize the structure of ultradistributions having S-
asymptotics.
Theorem 1.10. ([132]). Let T T
. Suppose: (M.1), (M.2) and (M.3)

are satised by M
p
; =
1
+
1
, where
1
R
n
is open set and
1
R
n
is
convex cone; is a subcone of
1
. Then T has the S-asymptotics related to c
and if and only if for a given open and relatively compact set A( A ),
there exist an ultradierential operator P(D) of class and continuous
functions f
1
and f
2
on A + such that
lim
h,h
f
i
(x +h)/c(h), i = 1, 2,
exist uniformly in x A, i = 1, 2, and T = P(D)f
1
+f
2
on A + .
Proof. One can easily prove that the condition is sucient.
The condition is necessary. Suppose that T(x +h)
s
c(h) U(x), h
in T
. Denote by T
h
=
h
T/c(h) and by (CB)
A
the space of continuous
and bounded functions on A. By Theorem 6.10 in [79], F
h
: T
h
, h
, are continuous mappings: T
B
r+
(CB)
A
. Consequently, F
h
are the
continuous mappings: T
K
(CB)
A
, where K = B
r
. The set T
h
; h
, |h| , > 0, is bounded in T
because of the S-asymptotics of T.

g
47
Thus, for a xed T
K
, the set F
h
(); h , |h| is bounded in
(CB)
A
. Since T
K
is a barrelled space, by the Banach theorem it follows that
the family of functions F
h
= F
h
; h , |h| is equicontinuous.
Therefore, there exists H
p
, p N of the form (0.1) and > 0 such that
any function in F
h
maps the neighborhood of zero
V
=
_
T
K
; sup
xK,||N
0
[
()
(x)[/H
||
M
||
<
_
(1.21)
into the unit ball B(0, 1) in (CB)
A
. Denote by

T
H
p
M
p
K
the completion of
T
K
under the norm q
H
p
M
p
(see (0.5)).
Using the extension of a function through its continuity (see [12]), we
shall show that the family F
h
can be extended on

T
H
p
M
p
K
, keeping the
uniform continuity; let us denote it by F
h
.
Let

T
H
p
M
p
K
and let
j
j
be a sequence in T
K
which converges to
being a Cauchy sequence in the norm q
H
p
M
p
. We shall prove that T
h

j
converges, as j , in (CB)
A
uniformly in h , |h| . It is enough
to prove that T
h

j
j
is a Cauchy sequence in (CB)
A
.
Every neighborhood of zero W in (CB)
A
contains the ball B(0, ) for
some > 0. The neighborhood of zero V
, given by (1.21), satises T

h

V
W, when h , |h| .
Let j
0
N
0
be such that
i
j
V
if i, j j
0
. Then,
T
h

i
T
h

j
= T
h
(
i
j
) W, h , |h| .
This proves the existence of the family F
h
and that lim
j
T
h

j
=
g
h
(CB)
A
, |h| . We shall prove that g
h
= T
h
. The sequence
j
j
converges to also in c
B
r+
, as well. So, T
h

j
converges to T
h
as
j , in T
, where = B
r+
(see [79], p. 73). Thus, (T
h
)[
A
must be g
h
.
It remains to prove that for

T
H
p
M
p
K
, T
h
converges in (CB)
A
as
h , |h| . Since F
h
is an equicontinuous family of functions, for
every

T
H
p
M
p
K
the set T
h
; h , |h| is bounded in (CB)
A
.
By the BanachSteinhaus theorem T
h
converges in (CB)
A
.
Now we will give the analytic form of T. Let T
Q
such that Q is
a compact subset of the interior of K and q
H
p
M
p
() < . One can easily
prove that

T
H
p
M
p
K
. Now, by Theorem 2.11 in [81], T = P(D)( T)
wT, where we can take that

T
H
p
M
p
K
and w T
K
. Let us denote T
by f
1
and w T by f
2
. The properties of these two functions follow by the
previous part of the proof.
We pass now to the case of Fourier hyperfunctions.
Theorem 1.11. Let f = [F
] Q(D
n
), F

O((D
n
+iI#D
n
). If f has the
S-asymptotics related to c and , then there exist an elliptic local operator
J(D) and functions q
s
C
(R
n
), s , is the set of n-vectors with entry
1, 1), of infra exponential type such that:
1. q
s
(z)

O(D
n
+iI
s
), where D
n
+iI
s
is an innitesimal wedge of type
T
n
+i
s
0, s .
2. f = J(D)
s
q
s
(x).
3. For every s there exists
0
> 0 such that q
s
(x +is)/c(x), s
converge as |x| , x , for every xed , 0 < <
0
.
Proof. Let
f

=
(x +i
), F
= sgnF
and
T(f) = [R]

=

T(
)( i

0) .
Then there exists a monotone increasing continuous positive valued function
(r), r 0, which satises (0) = 1, (r) , r and such that
[R()[ C
k
exp([[/([[),
1
k
[Im
j
[ 1 ,
where j = 1, . . . , n and k N (cf. [74], p. 652).
By Lemma 1.2 in [73], we can choose an entire function J of infra
exponential type on C
n
which satises the estimate:
[J()[ C exp([[/([[)), [Im[ 1 .
Then J
2
()

O(D
n
+i[[< 1) and
[R()/J
2
()[ C
k
exp([[/([[)) C
k
exp([[
) ,
where
1
k
[Im
j
[ 1, j = 1, . . . , n, and 0 < < 1.
g
49
Denote by g = T
1
(1/J
2
). By Theorem 8.2.6 in [75], g Q
1
(D
n
). By
Proposition 8.4.3. in [75]
f g =
_
R
f( t)g(t)dt Q(D
n
)
and
T(f g) = T(f)T(g) .
By the Corollaries of Proposition 2 in [165]
f = J
0
(D)(g f), J
0
= J
2
. (1.22)
We can always assume that there exists R
+
such that I = (, )
n
.
Then we denote by I
= I
, .
By Proposition 8.3.2 in [75] the following assertions are true:
F

O(x + iI
) and it decreases exponentially outside any cone

containing

as a proper subcone.
T(F

)

O(x iI
) and it decreases exponentially outside any cone

containing

T(F

)J
2
has the same cited properties as T(F

).
T(F

)J
2
s

O(xiI
s
) and it decreases exponentially outside any
cone containing
and
s
proper subcone.
T
1
(T(F

)J
2
s
)

O(x+i(I
I
s
)) and it decreases exponentially
outside any cone containing
s
Consider now the Fourier hyperfunction f g given in (1.22):
f g = T
1
(T(f)T(g))
=
1
(2)
n

_
R
n
e
iz
T(

F
)(

)/J
2
(

)d ,
where

I

and z
R
n
+iI
.
Let be xed. Then for and z
R
n
+iI
, we have
S
,
(z
) =
1
(2)
n
_
R
n
e
iz

T(

F
)(

)/J
2
(

)d ;
[S
,
(z
)[
1
(2)
n
_
R
n
e
x

y
T(

F
)(

)/J
2
(

)d .
One can see that S
,
(z
), z
R
n
+ I
are continuable to the real

axis. The corresponding functions x S
,
(x) are continuous and of infra
exponential type on R
n
. By Lemma 8.4.7 in [75], for x R
n
,
S
,
(x)

= S

(x +i
0), ,
(f g)(x) =

S
,
(x), x R
n
.
(1.23)
Functions S
,
can be written in the form
S
,
(z
) =
1
(2)
n
_
R
n
e
iz

T(
)(

)/J
2
(

)
s
(

)d, z
R
n
+I
.
Let s . Then functions
S
, ,s
(z
) =
1
(2)
n
_
R
n
e
iz

T(

F
)(

)/J
2
(

)
s
(

)d ,
z
R
n
+iI
, , , s , are also continuable to the real axis. The corre-

sponding functions x S
, ,s
(x) are continuous and of infra exponential
type on R
n
. Moreover, on R
n
,
S
, ,s
(x)

= S
, ,s
(x +i
0) ,
S
,
(x) =
s
S
, ,s
(x) .
(1.24)
Let us analyze functions
I
s
() = J
2
()e
s
s
(), (R
n
+i[[< 1) ,
where s and > 0. These functions are elements of T
because of
[I
s,
()[ = [J
2
()[exp
_
i=1
s
i
i
_
n
i=1
[
s
i
(
i
)[
[J
2
()[
n
i=1
[
s
i
(
i
) exp(s
i
i
)
C exp(
n
i=1
[
i
[), [[< 1.
Therefore, I
s,

O
(D
n
+i[[< 1), s . Since the Fourier transform
maps T
onto T
, there exists
s,
T
such that T(
s,
) = I
s,
. By
Proposition 8.2.2 in [75],
s,

O
1
(D
n
+i[y[< ), s .
g
51
Denote by
q
s
(x) =

S
, ,s
(x)

T
1
(T(F

)J
2

)(x +i(

s
)0), x R
n
. (1.25)
We prove that functions q
s
, s , have the properties cited in Theorem
1.11.
Property 1 follows from (1.24) and (1.25).
By (1.22) and (1.23) property 2 is satised.
It remains to prove property 3.
If f Q(D
n
) and T
, then f

O(D
n
+ iI
), where I
is
an interval containing zero. We shall use this fact and the properties of
functions I
s,
, already analyzed.
For a xed s there exists
0
> 0 such that s belongs to all in-
nitesimal wedges of the form D
n
+i(
s
)0 which appear in (1.25). For
(0,
0
], we have
q
s
(x +is) =

1
(2)
n
_
R
n
e
i(x+is)

T(F

)(

)J
2
(

)
si
(

)d
=

1
(2)
n
_
R
n
e
ix

T(F

)(

)T(
s,
)(

)d
=

((F

)
s,
)(x)
=
__

_

s,
_
(x) = (f
s,
)(x).
Now, for every xed (0,
0
], and s ,
lim
x,x
q
s
(x +is)/c(x) = lim
x,x
(f
s,
)(x)/c(x)
= lim
x,x
f(t +x)/c(x),
s,
(t).
We cite some papers related to this problem: ([131], [132], [155],[151], [123]).
1.10 S-asymptotic expansions in F
g
A sequence
n
n
of positive real-valued functions
n
(t), t (t
0
, ), t
0
0
(dened on (0, t
0
), t
0
> 0) is said to be asymptotic if and only if
n+1
(t) =
o(
n
(t)), t (t 0). The formal series
n1
u
n
(t) is an asymptotic
expansion of the function u related to the asymptotic sequence
p
(t) if
u(t)
k
n=1
u
n
(t) = o(
k
(t)), t (t 0) (1.26)
for every k N. We write in this case:
u(t)
n=1
u
n
(t)[
n
(t), t (t 0). (1.27)
If u
n
(t) = c
n
n
(t), for every n N, where c
n
are complex numbers, then
expansion (1.27) is unique. Indeed, the numbers c
n
can be unambiguously
computed from (1.26). In this case, we omit from the notation
n
(t) in
(1.27). A series which is an asymptotic expansion of a function f can be
also convergent. However, the series is divergent in general; nevertheless,
several terms of it can give valuable information, and very often its good
approximation properties come actually from the fact that the series is
divergent. Sometimes, if we take more terms from the asymptotic series,
we obtain a worse approximation; consequently, the determination of the
optimal number of terms for a good approximation depends on a careful
analysis of the problem under consideration.
An asymptotic expansion does not determine only one function. The
following example illustrates this fact:
exp
_
1
x
_
k=0
x
k
k!
and exp
_
1
x
_
+ exp(x
2
)
k=0
x
k
k!
, x .
In many problems of applied mathematics one is led to the use of asymp-
totic series. (See [44], [10], [15], [14], [192], [69]). A clear exposition of the
theory and the use of asymptotic series of functions and distributions can
be found in ([50][56]).
We shall discuss in this section the S-asymptotic expansion of general-
ized functions belonging to T
g
.
g
53
1.10.1 General denitions and assertions
In this section will be a convex cone with the vertex at zero belonging to
R
d
and () the set of all real-valued and positive functions c(h), h .
We shall consider the asymptotic expansion when |h|, h .
Denition 1.2. A distribution T T
g
has a S-asymptotic expansion re-
lated to the asymptotic sequence c
n
(h)
n
(), if for every T
T(t +h), (t)
n=1
U
n
(t, h), (t)[c
n
(h), |h|, h , (1.28)
where U
n
(t, h) T
g
(with respect to t) for n N and h . We write in
short:
T(t +h)
s
n=1
U
n
(t, h)[c
n
(h), |h|, h . (1.29)
Remarks. 1) In the special case U
n
(t, h) = u
n
(t)c
n
(h), u
n
T
g
, n
N, we simply write
T(t +h)
s
n=1
u
n
(t)c
n
(h), |h|, h . (1.30)
In this case the given S-asymptotic expansion is unique.
2) Brychkovs general denition is in o
(R) and slightly dierent from

ours ([15], [16] and [20]); his idea reformulated in T
g
(R) gives the following
denition. Suppose that f T
g
(R) and that the function exp(ixh), where
h is a real parameter, is a multiplier in T
g
(R).
Denition 1.3. Suppose that f T
g
(R). It is said that f(x)e
ixh
has an
asymptotic expansion related to the asymptotic sequence
n
(h)
n
if for
every T
g
(R)
f(x)e
ixh
, (x)
n=1
C
n
(x, h), (x)[
n
(h), h ,
where C
n
(x, h) T
g
(R) (with respect to x), n N, h h
0
. We write in
short:
f(x)e
ixh
n=1
C
n
(x, h)[
n
(h), h .
To obtain an equivalent denition of this asymptotic expansion,
Brychkov has supposed that T
g
(R) = o
(R) and T
g
(R) = o(R). Then, by
putting g =

f = T(f), c
n
(, t) = T(C
n
(, t)) and = T(), Denition 1.3
reduces to:
Denition 1.4. A distribution g o
(R) has an asymptotic expansion

related to the sequence
n
(h)
n
if for every o(R)
g(h t), (t)
n=1
c
n
(t, h), (t)[
n
(h), h ,
where c
n
(, h) o
(R), n N and h h
0
.
Denition 1.4 is, of course, a particular case of Denition 1.2 if we use
T(t) = g(t) and the cone = R
.
In [15] and [20] authors studied asymptotic expansions of tempered dis-
tributions given by Denition 1.3. In [16] Brychkov extended Denition 1.4
to the n-dimensional case, but only on a ray y; > 0 for a xed y R
n
.
We study in this section the asymptotic expansion not only in o
(R)
and not only on a ray, but on a cone in R
n
. The next remarks state some
motivations for such investigations.
Remarks. 1) A distribution in o
(R) can have an S-asymptotic expan-

sion in T
(R) without having the same S-asymptotic expansion in o
(R).
Such an example is the regular distribution f dened by
f(t) = H(t) exp(1/(1 +t
2
)) exp(t), t R,
where H is the Heaviside function.
It is easy to prove that for h R
+
f(t +h)
s
n=1
1
(n 1)!
(1+(t +h)
2
)
1n
exp(t h)[e
h
h
2(1n)
, h ,
while
t U
n
(t, h) = (1 + (t +h)
2
)
1n
exp(t h), n N, h > 0
do not belong to o
(R).
2) The regular distribution g dened by
g(t) = exp(1/(1 +t
2
)) exp(t), t R
g
55
belongs to T
(R) but not to o
(R). It has the S-asymptotic expansion in

T
(R) :
g(t +h)
s
n=1
1
(n 1)!
(1 + (t +h)
2
)
1n
exp(t +h)[e
h
h
2(1n)
, h ,
where = R
+
.
3) We can distinguish two cases of S-asymptotic expansions. If in
Denition 1.2,
U
n
(t, h) = u
n
(t)c
n
(h), n N,
then the S-asymptotic expansion is called of second type. (See Remark
after Denition 1.2). If U
n
(t, h) = u
n
(t +h), n N, then the S-asymptotic
expansion is of rst type.
The following example illustrates the dierence between these two types
of S-asymptotic expansions.
Let f(x) =
x
2
+x, x > 0 and f(x) = 0, x 0. A S-asymptotic
expansion of the rst type for this distribution is
f(x +h)
s
n=1
_
1/2
n 1
_
(x +h)
2n
[h
2n
, h ,
but the sequence u
n
(x) =
_
1/2
n1
_
x
2n
cannot give a S-asymptotic expansion
of the second type for f.
S-asymptotic expansions have similar properties as those of S-
asymptotics.
g
and
T(t +h)
s
n=1
U
n
(t, h)[c
n
(h), |h|, h .
Then:
a) T
(k)
(t +h)
s
n=1
U
(k)
n
(t, h)[c
n
(h), |h|, h .
b) Let the open set have the property: for every r > 0 there exists a
0
such that the closed ball B(0, r) = x R
n
, |x| r is in h, h
, |h|
0
. If T, T
1
T
0
and T
1
= T over , then
T
1
(t +h)
s
n=1
U
n
(t, h)[c
n
(h), |h|, h ,
as well.
c) Assume additionally that T
g
is a Montel space and that in T
g
the
convolution is well dened (see 0.6) and hypocontinuous. Let S T
g
,
supp S being compact. Then
(S T)(t +h)
s
n=1
(S U
n
)(t, h)[c
n
(h), |h|, h .
Proof. We prove only a). The proofs of b) and c) are the same as in the
proof of Theorem 1.2. We have
lim
h,h
_
T
(k)
(t +h)
m
n=1
U
(k)
n
(t, h), (t)
_
c
m
(h)
= lim
h,h
_
T(t +h)
m
n=1
U
n
(t, h), (1)
|k|
(k)
(t)
_
c
m
(h)
= 0 .
A relation between the asymptotic expansion of a locally integrable func-
tion f and its S-asymptotic expansion when seen as a regular generalized
function is provided in the following proposition (see [152]).
Proposition 1.8. Let f(t), U
n
(t, h) and V
n
(t), t R
n
, n N, h ,
be locally integrable functions such that for every compact set K R
n
the
following ordinary asymptotic expansion holds,
f(t +h)
n=1
U
n
(t, h)[c
n
(h), |h|, h , t K
and for every k N
1
c
k
(h)
f(t +h)
k
n=1
U
n
(t, h)
V
k
(t), t K, h , |h| r(k, K) .
Then for f T
0
, we have
f(t +h)
s
n=1
U
n
(t, h)[c
n
(h), |h|, h .
g
57
Proposition 1.9. Suppose that has the nonempty interior. Let T T
and
T(t +h)
s
n=1
u
n
(t)c
n
(h), |h|, h .
If u
m
,= 0, m N, then u
m
has the form
u
m
(t) =
m
k=1
P
m
k
(t) exp(a
k
t), t R
n
, m N,
where a
k
= (a
k
1
, . . . , a
k
n
) R
n
and P
m
k
are polynomials with degrees less
than k at every t
i
, i = 1, . . . , n.
Proof. Denition 1.2 and the given asymptotics implies
lim
h,h
T(t +h)/c
1
(h) = u
1
(t) ,= 0 in T
.
Now Proposition 1.2 implies the explicit form of u
1
.
The following limit gives u
2
:
lim
h,h
T(t +h), (t) u
1
(t), (t)c
1
(h)
c
2
(h)
= u
2
, , T .
Note,
lim
h,h
(D
t
i
a
1
i
)T(t +h), (t)
c
2
(h)
= (D
t
i
a
1
i
)u
2
(t), (t), T .
Two cases are possible.
a) If (D
t
i
a
1
i
)u
2
= 0, i = 1, . . . , n, then u
2
(t) = C
2
exp(a
1
t).
b) If (D
t
i
a
1
i
)u
2
,= 0 for some i, then by Proposition 1.2, (D
t
i

a
1
i
)u
2
(t) = c exp(a
2
t) and u
2
has the form
C
2
exp(a
1
t) +P
2
2
(t
1
, . . . , t
n
) exp(a
2
t) ,
where P
2
2
is a polynomial of the degree less than 2 with respect to each
t
i
, i = 1, . . . , n.
In the same way, we prove the assertion for every u
m
.
We will give an example of a function which has S-asymptotic expansion
of the rst type but does not have the asymptotic expansion as a function:
Let (t) = 1, t (n 2
n
, n + 2
n
), n N, and (t) = 0 outside of
these intervals. Let
(x) = e
x
_
x
0
(t)dt, x R, R.
Since
_
x
0
(t)dt 2 as x , we have that
(x) 2e
x
, x but
(x) does not have an ordinary asymptotic behavior (see Example 5. in

1.5.1).
Let
j
j
be a strictly decreasing sequence of positive numbers. Let
be a function, C
, 1 for x > 1, 0 for x < 1/2 and f(x) =
i=1
i
(x)(x i), x R. We have
f(x)
i=1
i
(x)[2 exp(
i
x), x .
This implies that f has the S-asymptotic expansion of the rst type:
f(x +h)
s
i=1
i
(x +h)[2e
i
h
, h
and
F(x +h) = f
(x +h)
s
i=1
i
(x +h)[2e
i
h
, h
but F does not have the ordinary asymptotic expansion.
For S-asymptotic expansions see also ([108], [152]).
1.10.2 S-asymptotic Taylor expansion
Estrada and Kanwal have introduced in [53] and [54] the asymptotic Taylor
expansion of distributions (see also [56]). We generalize it to T
g
.
Denition 1.5. Let f T
g
and D
k
f be the k-th partial derivative of f.
A formal series
|k|=0
D
k
f(x)y
k
k!

|k|
, where y R
n
is xed,
is the asymptotic Taylor expansion of f, as 0.
It means that for any test function T
g
f(x +y), (x)
N
|k|=0
D
k
f(x), (x)y
k
k!

|k|
= O(
N+1
), as 0 .
g
59
We write in short
f(x +y)
|k|=0
(D
k
f(x)y
k
)
|k|
k!
, 0 . (1.31)
In fact, the asymptotic Taylor expansion is the S-asymptotic expansion
related to the asymptotic sequence c
n
(h) = h
n
; n N, where h 0,
with U
n
(x, h) = u
n
(x)c
n
(h) and = y; > 0, where y R
n
is xed
(compare with Denition 1.2). We illustrate Denition 1.5 by two examples:
1. (x +y)
|N|=0
1
N!
(D
N
(x)y
N
)
N
, as 0,
where
N
=
N
1
++N
n
.
2. If , Z, then the distribution x
+
T
(R) has the asymptotic Taylor

expansion
(x )
k=0
_
k
_
(1)
k
x
k
+

k
, 0 ,
This means that if T(R), then
F.p.
_
0
(x )
(x)dx
k=0
__
k
_
(1)
k
F.p.
_
0
x
k
(x)dx
_
k
, 0 ,
where F.p. stands for the nite part (cf. [146] , [56]).
We now discuss the problem of the convergence of an asymptotic Taylor
expansion.
If a regular distribution is dened by a real analytic function f, then
the asymptotic Taylor expansion for f is a convergent power series in an
appropriate domain. But if f is only smooth, then the asymptotic Taylor
expansion for f does not necessarily have a domain of convergence. The
asymptotic Taylor expansion of distribution, given in Example 1, is not
convergent in T
(R); we know that the series a

k
(k)
diverges unless a
k
=
0, |k| n
0
N.
One could attempt to solve this problem by trying to interpret the
asymptotic Taylor expansion for a distribution in a wider space T
g
of
generalized functions asking the question: Find a necessary and sucient
condition for the asymptotic Taylor expansion of a generalized function to
be convergent in T
g
, i.e., so that it becomes its Taylor series.
In the following theorems we give the answer to this question if T
g
is the
space of distributions (T
g
= T
), the space of ultradistributions in which

(M.1), (M.2) and (M.3) are satised (see 0.5.2) (T
g
= T
) and the space

of Fourier hyperfunctions (T
g
= Q).
Theorem 1.13. ([163]). Let f T
(f T
) and y = (y
1
, . . . , y
n
) y
i
,=
0, i = 1, . . . , n. The asymptotic Taylor expansion for f, on the straight line
y; R, is a convergent Taylor series in T
(in T
) if and only if
there exists an r = (r
1
, . . . , r
n
), r
i
> 0, i = 1, . . . , n, such that f is a real
analytic function on R
n
which can be extended as a holomorphic function
on z C
n
; [Imz
i
[< r
i
, i = 1, . . . , n.
Proof. Suppose that f T
(f T
) and that w T (w T
). Then
f(x +y), w(x) = (f w)(y), y R
n
, R (1.32)
and
(f w)(y +y
0
) = (f(t) w(t y
0
))(y), y, y
0
R
n
, (1.33)
where is the sign of convolution and w(t y
0
) = w(t +y
0
). This shows
that the expansion in a polydisc around y
0
can be transferred to the ex-
pansion around 0.
In the sequel we shall use the following results proved in ([146], Chap.
VI, Theorem XXIV and [161], Theorem 1):
The generalized function f T
(f T
) is a real analytic function if

and only if f w is real analytic for every w T (w T
).
Suppose that there exists r = (r
1
, . . . , r
n
), r
i
> 0, i = 1, . . . , n such
that f is holomorphic in x C
n
; [Imz
i
[< r
i
, i = 1, . . . , n. Then its
Taylor series converges in D(a, r) = D(a
1
, r
1
) D(a
n
, r
n
) for every
a R
n
. Therefore for every r
= (r
1
, . . . , r
n
), 0 < r
i
< r
i
, i = 1, . . . , n
and every compact set K R
n
, by the characterization of a real analytic
function,
[f
(k)
(x)[ C(r
)
k
k! , k N
n
0
, x K .
Let us prove that
|k|=0
f
(k)
(x)
k!
(r
)
k
(1.34)
g
61
converges in T
(in T
) for any r
, 0 < r
i
< r
i
, i = 1, . . . , n. It is enough
to prove that for every w T (w T
) and every p N
lim
N
_
N+p
|k|=N
f
(k)
(x)
k!
(r
)
k
, w(x)
_
= lim
N
N+p
|k|=N
_
R
n
(f
(k)
(x)/(k! ))w(x)dx(r
)
k
= 0 .
Since for every w T (w T
_
R
n
(f
(k)
(x)/(k! ))w(x)dx
C(r
)
k
_
R
n
[w(x)[dx, [k[ n
0
(K) ,
the series (1.34) converges for every r
, 0 < r
i
< r
i
. Consequently, there
exists ,= 0 such that 0 < [y
i
[< r
i
and that series (1.31) converges in T
(in T
).
Suppose now that the series (1.31) converges in T
(in T
) for a
xed y R
n
and a xed > 0. Then for every w T (w T
|k|=0
D
k
f(x), w(x)
k!
(y)
k
is a convergent series in the polydisc D(0, r), where r
i
= y
i
, i = 1, . . . , n.
By (1.33), (f w)(z) is a holomorphic function on z C
n
; [Imz
i
[< r
i
, i =
1, . . . , n. By the cited theorem, f is a real analytic function. We now
continue the proof dividing it into two cases.
Case f T
.
By (VI, 6; 22) in [146]
f =
k
(gE f) (v f) , (1.35)
where is the Laplacian, v T, and k is large enough so that for a xed
m, gE T
m
V
, where V is a relatively compact neighborhood of zero.
By the property of f w we just proved, it follows that for every w T
and for every y
0
R
n
there exists M > 0 such that
(r
)
k
[(f
(k)
)(y
0
)[ M k! ,
where k N
n
0
and 0 < r
< r. By Theorem XXII Chap. VI in [146], there

exists m 0 such that this inequality is also true if w = w T
m
V
, where V
is a relatively compact neighborhood of zero. Consequently, it follows that
f w is a holomorphic function in B(y
0
, r). By (1.35), f is also holomorphic
in B(y
0
, r) for every y
0
R
n
.
Case f T
. In the proof of this case we use the following Theorem

([81]):
Let the sequence M
p
satisfy conditions (M1), (M2), and (M3). For
a given H
p
and a compact neighborhood Q of zero in R
n
there exist an
ultradierential operator P(D) of class and functions C
, and w
T
Q
such that
P(D) = +w
supp Q, sup
xQ
[
(k)
(x)[/H
|k|
M
|k|
0, [k[ .
Following the proof of Theorem 1 in [161], one can conclude that if
[(D
k
f w)(y
0
)[= [(f w)
(k)
(y
0
)[ Ck! (r
)
k
,
where w T
and k N
n
0
, then there exists H
p
such that the same
inequality holds for w = w

T
H
p
M
p
K
, where

T
H
p
M
p
K
is the completion of
T
K
under the norm g
H
p
M
p
. Therefore f is a holomorphic function in
B(y
0
, r).
By the cited theorem
f = P(D)( f) +f w, (1.36)
where f and f w are holomorphic in a ball around any point of R
n
.
So the proof will be nished if we prove that for a real analytic function
and an ultradierential operator P(D) of (*)-class, P(D) is real analytic.
Therefore, we will prove rst the next assertion.
Lemma 1.3. Let P(D) be an ultradierential operator of (*)-class and
be real analytic in a neighborhood of x
0
R
n
. Then P(D) is real analytic
in a neighborhood of x
0
.
Proof. We will prove the assertion in the case n = 1 and P(D) being of
M
p
-class which is equivalent to P(D) = a
k
D
k
, where a
k
C, k N
0
and there exist C > 0 and h > 0 such that
[a
k
[
Ch
k
M
k
, k N
0
.
g
63
We have to prove that there exists H > 0 such that in a ball B
r
= x; [x
x
0
[ r
sup
N
n
0
,xB
r
H
[(P(D))
()
(x)[
M
< .
We have (x B
r
, N
0
)
1
M
k=0
a
k
D
k
(x)
k=0
a
k
D
+k
(x)
_
CC
1
H
k=0
h
k
M
k
C
k+
1
(k +)!
_
,
where we have used [D
k
(x)[ C
k+1
1
k! , x B
r
which holds for some
C
1
> 0. Since (k +)! e
k+
k! ! , we continue
CC
1
k=0
e
+k
H
h
k
C
k+
1
k!
M
M
k

C sup
(C
1
e)
! H
k=0
(C
1
h)
k
k!
M
k
.
The right side is bounded for every H > 0. This proves the lemma and
consequently the Theorem.
Theorem 1.14. ([164]) Let q Q, = (
1
, . . . ,
n
),
i
> 0, i = 1, . . . , n
and > 0. The asymptotic Taylor expansion (1.31) for q on the straight
line h; h R is a convergent Taylor series in the topology of Q if and
only if there exists an r = (r
1
, . . . , r
n
), r
i
> 0, i = 1, . . . , n, such that q is
given by a real analytic function which can be extended as a holomorphic
function on z C
n
; [Imz
i
[< r
i
, i = 1, . . . , n.
Proof. In the rst part of the proof, we suppose that series (1.31)
converges in Q for a xed y = = (
1
, . . . ,
n
),
i
> 0, i = 1, . . . , n, and a
xed > 0. Since Q is an FS-space it is equivalent to suppose that series
(1.31) converges weakly in Q.
By Theorem 1.3 and Remark 1.4 in [73], there exists an elliptic local
operator J
1
(D) and an innitely dierentiable function g rapidly decreasing
([g(x)[ C exp(|x|), x R
n
for some > 0) such that
= J
1
(D)g ( is the delta distribution) .
Also (cf. Theorem in [74]) for every Fourier hyperfunction q Q, we
can nd an elliptic local operator J
2
(D) and an innitely dierentiable
function f of infra-exponential growth such that q = J
2
(D)f. Then, by the
properties of the convolution:
q = (J
2
(D)f) (J
1
(D)g) = J
2
(D)J
1
(D)((f) (g)) , (1.37)
where J(D) = J
2
(D)J
1
(D) is also an elliptic operator. (f denotes the
hyperfunction dened by the function f).
From the proof of two mentioned theorems and by Theorem 8.2.6 in
[75] it follows that there exist two sets of functions F
; and
G

; such that:
a) F

O(D
n
+iI
) and G

O
(D
n
+iI

), > 0, , ;
b) the functions F
and G

can be extended to the real axis R
n
as
innitely dierentiable functions F
(x) and G

(x), respectively. F
(x) are
infra-exponential and G

(x) are rapidly decreasing, , .
c) f(x) =
(x) and g(x) =

G

(x), x R
n
; g is analytic
outside 0.
By Carlemanns theorem, we have
f(x) :=
(x +i
0), g(x) :=

G

(x +i

0), x R
n
(cf. [75] Lemma 8.4.7, and [145], 7).
Our rst step is to prove that the Fourier hyperfunction (f) (g) is
dened by the infra-exponential function
_
R
n
f(x u)g(u)du, x R
n
.
By the denition of the convolution, we have
(lf) (lg)(u) :=

_
Imw=v

F
(z w)G

(w)du, w = u +iv, v

I

,
(1.38)
where z w R
n
+ i
0. Hence z can move inside R

n
+ i(
+

)0.
We can shift the integral path in the last integral to the real axis R
n
. This
change of the path is justied by Cauchys integral formula and the growth
rate of the functions f and g at innity. Then we have
(f) (g)(x) :=

_
R
n
F
(z u)G

(u)du .
g
65
Every integral in the last sum is a function into

O(D
n
+i(
+

)0),
respectively, and can be extended to the real axis as a slowly increasing
continuous function. Therefore by the same Carlemanns theorem
(f) (g)(x) =
_

_
R
n
F
(x u)G

(u)du
_
=
_
_
R
n
f(x u)g(u)du
_
. (1.39)
Hence, the rst step is proved.
In the second step, we shall prove that (lf) (lg)(x) is a real analytic
function in a neighborhood of zero, that is it can be extended on a complex
neighborhood of zero as a holomorphic function.
By denition of an innitesimal wedge U
1
of type D
n
+
1
0 for every
proper subcone
1
there exists w > 0 such that U
1
D
n
+i(
[y[< w). Put

G(u) = G(u +iw/2). Then

G

O
(D
n
+ i[v[< w/2)
and

G T
, as well.
Let us consider an addend in (1.38) and the corresponding integral in
(1.39). We have
(F
G

)(x) :=
_
R
n
F
_
z
_
u +
1
2
iv

__
G

(u +
1
2
iv

)du, x R
n
,
where v

[y[<
1
2
w. Thus, for x R
n
,
(F
G

)
_
x +
1
2
iv

_
=
_
R
n
F
(x u)

G

(u)du
=
_
R
n
F
(x +u)

G

(u)du
=
_
R
n
F
_
x +u +
1
2
iv

_
G

_
u +
1
2
iv

_
du .(1.40)
By (1.38)(1.40), we have
((f) (g))
_
x +
1
2
iv

_
=

f(x +u),

G

(u),

G

T
.
Let > 0 be xed. The assumption of the theorem, implies that
q(x + u), G

(u), G

T
, as a function in x R
n
, can be extended
on B(0, ) C
n
as a holomorphic function. Then
q(x +u), G

(u) = J
2
(D)f(x +u), G

(u)
= J
2
(D)f(x +u), G

(u), x R
n
.
Since J
2
(D) is an elliptic local operator, it follows that for every
, f( + u), G

(u) is also real analytic. Consequently, ((f) (g)) is
holomorphic in B(0, ) for ,
i
> 0, i = 1, . . . , n.
Now it is easy to prove that ((f) (g))(z) is holomorphic on R
n
+
B(0, ).
In order to nish the rst part of the proof, we have only to remark that
from (1.37) and from the property of an elliptic local operator it follows that
q is a real analytic function which can also be extended as a holomorphic
function on R
n
+B(0, ).
For the second part of the proof suppose that q is given by a real analytic
function which can be extended on R
n
+ B(0, ),
i
> 0, i = 1, . . . , n, as
a holomorphic function. Then for > 0 and R
n
,
q( + x) := q( +x +iI
0
), x R
n
and, for y
0
I
0
,
q( +x), (x) =
_
Imz=y
0
q( +x +iy
0
)(x +iy
0
)dx, T
.
By the same arguments which we have used to transform (1.38) into
(1.39), we have
q( +x), (x) =
_
R
n
q( +x)(x)dx, T
.
Since q(z) is holomorphic on R
n
+ B(0, ), the same holds for q( +
x), (x).
For asymptotic Taylor expansions, see also [163], [1], [50], [56].
1.11 S-asymptotics in subspaces of distributions
We discuss in this section the following problem: Let /
be a subspace of
T
. If T /
and if T has the S-asymptotics in T
, is it true that T has the

S-asymptotics in /
, as well? The answer is not simple. We shall analyze

two cases, /
= o
and /
= /
1
. First, we shall illustrate the problem by
an example.
g
67
The distribution H(t) exp(t) (H is Heavisides function) belongs to
o
(R). It has S-asymptotics in T
(R) related to c(h) = exp(h) with limit

U = exp(t). We know that exp(t) does not belong to o
(R) and T(R)

is dense in o(R). Hence H(t) exp(t) cannot have the S-asymptotics in
o
(R).
To answer our question, we need to introduce some additional conditions
over c(h).
Although we dened the S-asymptotics of distributions in a general cone
, we shall restrict ourselves to the cone = R
n
+
for the sake of simplicity.
Recall, for a, b R
n
, a b means a
i
b
i
, i = 1, . . . , n.
The set of real-valued functions h c(h), h R
n
, dened on R
n
,
dierent from zero for h R
n
+
, is denoted by (R
n
). We assume, without
losing generality, that function c in (R
n
) are positive and equal to 1 in
R
n
+
(R
n
+
+a), where a R
n
+
depends on c and R
n
+
+a = x+a; x R
n
+
.
By
e
(R
n
) is denoted the subset of (R
n
) such that c
e
(R
n
) if and
only if for some C > 0, d > 0, k > 0 and h
0
= (h
0,1
, . . . , h
0,n
) R
n
+
(E.1) c(h +r) Cc(h) exp(k|r|), h > h
0
, r R
n
.
(E.2) c(h) exp(k|h|) d, h > h
0
.
By
p
(R
n
) is denoted another subset of (R
n
) dened as follows,
c
p
(R
n
) if and only if for some C > 0, d > 0, k > 0 and h
0
=
(h
0,1
, . . . , h
0,n
) R
n
+
:
(P.1) c(h +r) Cc(h)(1 +|r|)
k
, h > h
0
, r R
n
.
(P.2) c(h)(1 +|h|)
k
d, h > h
0
.
Obviously,
p
(R
n
)
e
(R
n
).
We begin with some properties of the sets
e
(R
n
) and
p
(R
n
), we
restrict our attention to n = 1 in order to analyze explicit representations
for their elements.
The set
p
(R).
By Remark 3 after Proposition 1.2, a function c from Denition 1.1 has
the form c(h) = exp(h)L(exp(h)), h > h
0
> 0, where R and L is a
slowly varying function, and the limit distribution U(x) = C exp(x). For
the S-asymptotics in o
, has to be zero: Then c(h) = L(exp h), h > h

0
.
Proposition 1.10. Suppose that c (R).
a) If there exists T o
(R) such that T(x + h)

s
c(h)U(x), h R
+
with U ,= 0, then c(h) = L(exp h) where L is a slowly varying function, and
so c satises (P.2).
b) If c = h
L(h), h > h
1
and c(h) = 1, h h
1
, where R, h
1
> 0 and
L is a slowly varying and monotonous function for h > h
1
, then c satises
(P.1).
c) If c is of the same form as in b), where L is only slowly varying, then
c satises (P.1) but for r R
+
.
For the proof cf. [107].
The set
e
(R).
Proposition 1.11. Suppose that c (R).
a) If there exists T T
(R) such that T(x+h)

s
c(h)U(x), U ,= 0, then
(E.2) holds for c.
b) If c(h) = exp([h[)h
L(h), h > h
1
, c(h) = 1, h h
1
, where ,
R, h
1
> 0 and L is a monotonous slowly varying function for h > h
1
, then
(E.1) holds for c.
c) If c is of the same form as in b), where L is only a slowly varying
function, then c satises (E.1) but for r R
+
.
For the proof cf. [107].
We prove now that if T /
1
(R) (T o
(R)) has the S-asymptotics

in the cone R
+
related to some c(h)
e
(R) (c(h)
p
(R)) with limit
U in the space T
(R), then the limit

lim
h
T(x +h)/c(h) = U(x)
also exists in /
1
(R) (in o
(R)). If n > 1, we have to assume some addi-

tional conditions which imply the same assertion for the multidimensional
case.
First, we shall prove a theorem which extends Proposition 1.7 b) and
which will be used later.
. If for every rapidly exponentially decreas-

ing function r (for every k > 0, r(x) exp(k|x|) 0 as |x| ) the set
r(h)T(x +h); h R
n
is bounded in T
, then T /
1
.
g
69
Proof. Let K be an arbitrary compact set in R
n
. For every T
K
T,
h r(h)T(x +h), (x), h R
n
,
is a bounded function. This implies that for some k
1
= k
1
() > 0 and some
C = C() > 0
[T(x +h), (x)[ C exp(k
1
|h|), h R
n
. (1.41)
If h R
n
is xed, then
log
+
[T(x +h), (x) [ /(1 +|h|), T
K
,
a denes a continuous function on T
K
. It follows from (1.41) that
log
+
[< T(x + h), (x) > [/(1 + |h|); h R
n
is a bounded family of
continuous functions on T
K
; from the classical theorem of Baire, it follows
the existence of some k > 0, which does not depend on T
K
, such that
the set T(x+h) exp(k|h|), (x); h R
n
is bounded for each T
K
,
and hence, for each T. This implies (see Theorem XXII, Chapter VI
in [146]), that for a given open bounded set R
n
, 0 , there exists a
compact neighborhood of zero K and m N
0
such that for every T
m
K
,
x (T(t +h) (t))(x)/exp(k|h|); h R
n
is a bounded family of continuous bounded functions on .

Since (T(t +h) (t))(x) = (T )(x+h), setting x = 0, we obtain that
h (T )(h)/exp(k|h|), h R
n
, is a bounded function on R
n
for any
T
m
K
. Now, by (VI, 6; 22) in [146], we obtain
T =
N
(E T) T , (1.42)
where E is the fundamental solution of
N
E = ( is the Laplacian),
T
K
, 1 in a neighborhood of 0 and T
K
. If N is suciently
large, E T
m
K
. Thus, E T and T are in /
1
, and this completes the
proof.
Theorem 1.16. Let T /
1
(R) (T o
(R)) and c

e
(R) (c
p
(R)). a) If the set T(x + h)/c(h); h > a is bounded in T
(R), then
this set is bounded in /
1
(R) (in o
(R)) as well.
b) If there exists the limit
lim
h
T(x +h)/c(h), (x) = S, , T(R),
then this limit exists for every /
1
(R) (for every o(R)). In partic-
ular, S /
1
(R)(S o
(R)).
Proof. We prove the theorem for T /
1
(R), because for T o
(R) it
can be done in a similar way. (We have to replace exp(k||) by (1 +||)
k
).
a) Using the last part of the proof of Theorem 1.15, we obtain that for
some m
1
N
0
and some compact neighborhood of zero, K
1
, h (T
)(h)/c(h), h > 0 is a bounded function for every T
m
1
K
1
(R). Since
T /
1
(R), it holds that for some k > 0, m
2
N
0
and some compact
neighborhood of zero K
2
, h (T )(h)/exp(k[h[), h R, is a bounded
function for every T
m
2
K
2
(R). Thus, by taking N in (1.42) suciently
large ( = d
2
/dx
2
) and K = K
1
K
2
, we obtain that for some m N
0
T =
m
i=0
F
(i)
i
,
where F
i
, i = 0, . . . , m, are continuous functions on R such that for some
M
1
> 0, M
2
> 0 and k > 0
sup[F
i
(x)/c(x)[; x > 0, i = 0, . . . , m M
1
, (1.43)
sup[F
i
(x)[exp(k[x[); x R, i = 0, . . . , m M
2
,
Let /
1
(R) and h > h
0
be xed. We put
I
i
(h, ) I
i
=
([
(i)
(x)|F
i
(x +h)[/c(h))dx
=
_
h
_
_
h
_
([
(i)
(x)|F
i
(x +h)[/c(h))dx
= I
i
(, h) +I
i
(h, ), i = 0, . . . , m, h > h
0
.
If x (, h), then [x +h[= [x[h and by (E.2), we obtain
I
i
(, h) d
1
h
_
[
(i)
(x)[exp(2k)[x[)[F
i
(x +h)/exp(k[x +h[)[dx
Md
1
[
(i)
(x)[exp(k[x[)dx.
From the denition of the space /
1
(R) it follows that the last integral
is nite.
g
71
Because of (E.1) and (1.43), we obtain that for some k
1
> 0,
I
i
(h, )
_
h
[
(i)
(x)|F
i
(x +h)/c(x +h)[exp(k
1
[x[)dx
M
1
[
(i)
(x)[exp(k
1
[x[)dx < .
Since
[T(x +h)/c(h), (x)[
m
i=0
I
i
(h, ),
from the preceding inequalities, we obtain that for some A > 0 which does
not depend on h > h
0
and for /
1
,
[T(x +h)/c(h), (x)[ A, /
1
, h > h
0
.
Thus the proof of a) is complete.
b) If the limit given in b) exists, then for an a R, the set T(x +
h)/c(h); h a is bounded in /
(R) (in o
(R)). Since T(R) is dense

in /
1
(R) (in o(R)), it is now enough to use the assertion of a) and the
BanachSteinhaus theorem.
If we assume more on a distribution T, then we can assume less on the
function c.
(R) and supp T [0, ). Suppose that c

(R) and satises (E.2) and (E.1) with r R
+
(satises (P.2) and (P.1)
with r R
+
).
a) If the set T(x + h)/c(h); h > a is bounded in T
(R), then this set

is bounded in /
1
(R) (in o
(R)), as well.
b) If there exists the limit
lim
h
T(x +h)/c(h), (x) = S, , T(R),
then this limit exists for every /
1
(R) (for every o(R)). In partic-
ular, S /
1
(R) (S o
(R)).
We give other versions of Theorem 1.17 because we want to empha-
size that the space o
(R) is natural for the distributions having the

S-asymptotics related to c(h) = h
L(h), h > 0, R, while the space

/
1
(R) is natural for those distributions having the S-asymptotics re-
lated to c(h) = exp(ah)L(exp(h)), h > 0, a R, where L is a slowly
varying function.
(R), supp T (, ), > 0 and T(x +h)

s
e
ah
L(e
h
) Ce
ax
, h R
+
in T
(R) (and T(x + h)

s
h
L(h) C, h R
+
in T
(R)), then T /
1
(R) (T o
(R)) and T has the S-asymptotics in

/
1
(R) (in o
(R)) related to the same c and with the same limit.

Proof. By Theorem 1.9 and Lemma 1.2 there exist functions F
i
, i =
0, . . . , m, continuous on (, ), such that
T =
m
i=0
D
i
F
i
on (, ) ,
where
[F
i
(x +h)/c(h)[ M
i
, h R
+
, x (, ), i = 0, . . . , m,
supp F
i
( , ), i = 0, . . . , m, > 0.
For a slowly varying function L there exists a slowly varying function
L
(c,)
(R), > 0, such that L
(x)/L(x) 1, x . We refer
to the Remark after Proposition 1.2 for the construction of L
. Therefore,
we can suppose, without any restriction, that L, which appeared in c, is
in C
(,)
(R), > 0. Let us denote by M
i
(h) = F
i
(h)/c(h), h R; M
i
is a continuous and bounded function on R and F
i
= cM
i
. This im-
plies that T /
1
. The next step is to prove that for h
0
R the set
T(x + h)/c(h); h h
0
is weakly bounded in /
1
which is equivalent to
the strong boundedness in /
1
.
We need the following inequality:
c(x +h)/c(h) = e
x
L(e
x
e
h
)/L(e
h
) Ae
ax+|x|
, x, h R,
where > 0 (see [9], p. 25). Let /
1
, then
[T(x +h)/c(h), (x)[
m
i=0
_
R
[F
i
(x +h)/c(h)|
(i)
(x)[dx
g
73
i=0
_
R
[M
i
(x +h)c(x +h)/c(h)|
(i)
(x)[dx
i=0
_
R
e
ax+|x|
[
(i)
(x)[dx.
This proves that T(x+h)/c(h); h h
0
1
. Since T
is dense in /
1
, the BanachSteinhaus theorem implies the S-asymptotics
in /
1
related to c.
In order to extend Theorem 1.17 to the multidimensional case, we have
to introduce the following notation.
We denote by the set of all nth class variations of elements 1, 1.
If (a
1
, . . . , a
n
) , then we put
(a
1
, . . . , a
n
) = h R
n
;
n
i=1
sgn(a
i
h
i
) = n.
(This means if a
i
= 1, (a
i
= 1), then h
i
> 0(h
i
< 0)). For example
(1, . . . , 1) = R
n
+
and (1, . . . , 1) = R
n
. Let
c(h) = c
1
(h
1
) . . . c
n
(h
n
), h
i
R, i = 1, . . . , n, (1.44)
where
c
i

e
(R) (c
i

p
(R)), i = 1, . . . , n.
Obviously, c
e
(R
n
) (c
p
(R
n
)). Let (a
1
, . . . , a
n
) be given. We
denote by j
i
, i = 1, . . . , r the components of (a
1
, . . . , a
n
) which are equal to
1, and those which are equal to 1 by s
i
, i = 1, . . . , m (r +m = n).
Let h (a
1
, . . . , a
n
) and k > 0. We put
c
k
(a
1
,...,a
n
)
(h) = c
j
(h
j
) . . . c
j
r
(h
j
r
) exp(k([h
s
1
[+ +[h
s
m
[))
Theorem 1.19. Let T /
1
(T o
) and c be of the form (1.44).

a) If there exists k > 0 such that for every (a
1
, . . . , a
n
)
T(x +h)/c
k
(a
1
,...,a
n
)
(h); h (a
1
, . . . , a
n
)
is bounded in T
, then T(x +h)/c(h); h > 0 is bounded in /
1
(in o
).
b) If
lim
h
T(x +h)/c(h), (x) = S(x), (x), T,
and if for some k > 0 and every (a
1
, . . . , a
n
) (1, . . . , 1), (1, . . . , 1)
the sets
T(x +h)/c
k
(a
1
, . . . , a
n
)(h); h (a
1
, . . . , a
n
),
are bounded in T
, then T(x + h)/c(h) converges to S in /
1
(in o
), as
h . In particular, S /
1
(S o
).
Instead of analyzing the two special subspaces o and /
1
of T we can
consider a general subspace / of T.
Let be a convex cone. We denote
q
() a subset of () such that
c
q
() if and only if there exist C > 0 and a positive locally integrable
function p such that
c(h +x) Cc(h)p(x), h, x B(0, r) . (1.45)
In the sequel, put G = x R
n
(B(0, r)). We denote by / a barrelled
vector space of smooth functions such that T is dense in / with its topology
ner than the topology induced by /; /
is the dual space of /, /
.
We will suppose that the elements of / satisfy the following condition:
for every y B(0, r), p(x)(x +y) L
1
.
Theorem 1.20. Suppose that T /
and c
q
().
a) If the sets:
Q
1
= T(x +h)/c(h); h
Q
2
= T(x +k +h)/(c(h)p(k)); h , k G
are weakly bounded in T
, then the set Q

1
as well.
b) If T has S-asymptotics in T
related to c(h) with the limit U and if

the set Q
2
is weakly bounded in T
, then T has the S-asymptotics in /
related to c(h) and with the limit U.

We remark that the well-known basic spaces as /
1
, o, T
L
p(1 p < )
and
B
satisfy our conditions assumed in Theorem 1.20 for the space /. The
space B = T
L
is an example of one which does not satisfy them, due to
the fact that T is not dense in B.
g
75
If a distribution f has the S-asymptotic expansion in T
, with adequate
additional conditions, it can have the S-asymptotic expansion in a subspace
of T
. The following theorem gives such conditions.

Theorem 1.21. Suppose that f /
1
(R) (f o
(R)) and u
i
/
1
(R)
(u
i
o
(R)). If
f(x +h)
s
i=1
u
i
(x +h)[c
i
(h), h in T
(R)
and c
i
(h) = exp(
i
h)L
i
(exp h) (resp. c
i
= h
i
L
i
(h)), h > h
0
), where
L
i
, i N, are monotonous slowly varying functions, then
f(x +h)
s
i=1
u
i
(x +h)[c
i
(h), h , in /
1
(R) (in o
(R)),
as well.
Proof. The proof is based on Proposition 1.11 and Theorem 1.18, we
leave the details to the reader.
The next two propositions give the relation between the asymptotic
behavior of functions and the S-asymptotics of distributions dened by
these functions.
1
loc
(R) be such that f denes a regular distri-
bution in /
1
(R), i.e., f L
1
(R) for every /
1
(R). Further, assume
f(x) L(exp x) exp(x) as x ,
where L is a slowly varying monotonous function dened on (a, ). If
c(h) = L(exp h) exp(h), h > A, and g(x) = exp(x), x R, then
f(x +h)
s
c(h)g(x), h R
+
in /
1
.
Proof. Let T and supp [a, b]. Since (L(h)/L(h)) 1, h ,
uniformly on any compact interval contained in (0, ), we obtain
f(x +h)
exp(h)L(exp h)
(x)dx =
b
_
a
f(x +h)
exp((x +h))L(exp(x +h))
exp(x)
L(exp x exp h)
L(exp h)
(x)dx
b
_
a
exp(x)(x)dx, h .
The assertion follows now from Proposition 1.11 and Theorem 1.16.
Proposition 1.13. Suppose that the functions f and u
i
, i N, are locally
integrable and dene regular distributions belonging to /
1
(R), i.e., f
and u
i
are in L
1
(R) for every /
1
(R). Let c
i
(h) = exp(
i
h)L
i
(exp h),
where L
i
are monotonous slowly varying functions and
i
R, i N. If
f(x)
i=1
U
i
(x)[ c
i
(x), x ,
then
f(x +h)
s
i=1
u
i
(x +h)[c
i
(h), h .
Proof. The proof is similar to that of Proposition 1.13.
We shall prove a structural theorem for a distribution T B
having
the S-asymptotics in a cone with the nonempty interior.
Theorem 1.22. Suppose T
0
B
and T
0
(x + h)
s
1 U(x), h in T
,
then
a) U = C.
b) T
0
=
2
i=0
ik
F
i
, where F
i
are continuous functions belonging to L
;
c) For every 0 i 2 functions F
i
(x+h) of part b) converge uniformly
to a constant when x belongs to a compact set K and h , |h|.
d) T
0
has the S-asymptotics in B
, related to c = 1 and with the limit

U = C in the cone .
Proof. a) U has to be a constant because c = 1 (Proposition 1.2).
b) From the fact that T
0
B
it follows that (T
0
) = T
0
(x+), (x)
L
for every T (see [146], VI, 8) and the set of distributions Q =

T
h
T
0
(x +h); h R
n
is weakly bounded and bounded in T
.
We will construct another bounded set of distributions. Denote S =
T; ||
L
1 1. We have seen that for a xed T, (T
0

) L
.
Now, for every S :
[T
0

, [= [T
0

, [= [
_
R
(T
0

)(t)(t)dt[ |T
0

|
L
||
L
1.
g
77
Hence, the set of regular distributions, dened by the set of continuous
functions U
T
0

; S is weakly bounded and therefore bounded
in T
.
A set W
is bounded if and only if for every T the set of

functions T ; T W
is bounded on every compact set M belonging

to R
n
(see [146], VI, 7 ). Hence T ; T W
denes a bounded set of

regular distributions. In such a way T
h
; T
h
Q and U
; U
H
are bounded sets of regular distributions. Now, for these two sets we can
repeat twice a part of the proof of Theorem XXII, Chapter VI in [146].
We denote by an open neighborhood of zero in R
n
which is relatively
compact in R
n
, namely,

= K is a compact set. Then, by the mentioned
part of the proof, there exist m
1
0 and m
2
0, such that the mappings
(, ) U
( ) or (, ) T
h
( ), h R
n
, are equicontinuous
and map T
m
1
T
m
1
and T
m
2
T
m
2
into L
B
respectively; B is the ball
B(0, r), where r is a positive number. Hence, for every x B and h R
n
the function (T
h
)(x) = (T
0
)(x +h) is continuous.
Let Z(0, ) be a ball in L
B
. Then there exists a neighborhood V
1
(m
1
,
1
,
K
1
) in T
m
1
, such that U
() Z(0, ) for , V
1
(m
1
,
1
, K
1
), U

H and a neighborhood V
2
(m
2
,
2
, K
2
) T
m
2
, such that T
h
() Z(0, )
for , V
2
(m
2
,
2
, K
2
), T
h
Q. Let K
0
= K
1
K
2
,
0
= min(
1
,
2
) and
m = max(m
1
, m
2
). We shall use (VI, 6; 23) in [146].
T
0
=
2k
(E E T
0
) 2
k
(E T
0
) + ( T
0
) ,
where E is a solution of the iterated Laplace equation;
k
E = ; ,
T
, supp and supp belonging to K

0
= K
1
K
2
. We have only to
choose the number k large enough so that E T
m
. Now, we can take:

F
2
= E E T
0
; F
1
= E T
0
and F
0
= T
0
. All of these
functions are of the form: F
i
= T
0

i

i
;
i
,
i
V (m,
0
, K
0
),
0
> 0.
We have to prove that F
i
, i = 0, 1, 2, have the properties given in
Theorem 1.22. For
i
,
i
V (m,
0
, K
0
) and S
[(T
0

i

i
), [= [[T
0

) (
i
i
)](0)[ (
0
/
0
)
2
M .
Now let ,= 0 be any element in L
1
. Then /||
L
1 S and [(T
0
(
i
i
)), [ M||
L
1 which proves that T
0
(
i

i
), i = 0, 1, 2, belong to
L
. Since F
i
= T
0
(
i

i
),
i
,
i
V (m,
0
, K
0
), F
i
, i = 0, 1, 2, are
continuous and belong to L
.
c) We shall continue the investigations of the properties of F
i
. By the
properties of the convolution, we have
F
i
(x +h) = F
i

h
= T
0
(
i

i
)
h
= T
h
(
i

i
),
i
,
i
T
m
,
where
h
is the translation operator.
We have proved that the mappings (, ) T
h
, T
h
Q, are
equicontinuous and map T
m
T
m
into L
B
. T is a dense subset of T
m
, m
0. We can construct a subset of T
K
, cl = K, which is dense in T
m
. Since
T
h
( ) C for T
, then T
h

i

i
converges to
C
i

i
, as well (see [146], VI, 7),when |h|, h , i = 0, 1, 2.
d) It remains to prove the last part of Theorem 1.22. For T
L
1 and
T B, we have:
[T
0
(x +h), (x)[
2
i=0
_
R
n
[F
i
(x +h)
ik
(x)[dx
2
i=0
M
i
_
R
n
[
ik
(x)[dx,
where M
i
= sup[F
i
(x)[, x R
n
. Hence the set T
0
(x + h), h R
n
, is
weakly bounded in B
. Since T is dense in T
L
1 , by the BanachSteinhaus
theorem the limit:
lim
h,h
T
0
(x +h), (x), T
L
1 ,
exists, as well, and it is equal to C, .
1.12 Generalized S-asymptotics
Denition 1.1 corresponds to some subsets of distributions of a limited
growth (see Theorem 1.16). This is the reason for introducing a general-
ization of the S-asymptotics ([119]).
Denition 1.6. Suppose that e is a function on R
n
such that e
1
M
()
(see 0.6). Then, it is said that T T
g
has generalized S-asymptotics in
the cone , related to e, if there exists
w. lim
h,h
T(x +h)/e(x +h) = 1 in T
g
.
We write in short T(x)
gs
e(x), |x|, x .
If we compare Denition 1.1 and Denition 1.6, we obtain.
g
79
Proposition 1.14. Let e be a function on R
n
such that e
1
M
()
. Then,
T T
g
has generalized S-asymptotics in , related to e if and only if T/e
has the S-asymptotics in related to c = 1 and with limit U = 1.
In such a way, dividing T T
g
by e, we may apply our results on
the S-asymptotics to this new context. So, all the assertions on the S-
asymptotics can be transferred for the generalized S-asymptotics by simply
using Proposition 1.14. For this reason, we underline only some of the key
properties of generalized S-asymptotics.
The generalized S-asymptotics is a local property, as well (see Theorem
1.4). From Theorems 1.2,d) and 1.3,4) it follows:
Proposition 1.15. a) Let S c
and T T
. If T(x)
gs
e(x), |x|
, x , then (S (T/e))(c +h)
s
1 (S 1)(x), h .
As consequence of T(x)
gs
e(x), |x| , x , it follows that for any
partial derivative D
x
i
(D
x
i
(T/e))(x +h)
s
1 0, |h|, h .
In order to compare the existence of the generalized S-asymptotics with
that of the S-asymptotics of a T T
, we need the following notation and

lemma.
Let be a cone with the vertex at zero. We employ the notation pr
for the intersection of and the unit sphere in R
n
.
Lemma 1.4. Let T T
and let c () where is a convex cone with

the nonempty interior, (
_
,= ). Let
be a closed cone,
int .
Suppose that T(x + h)
s
c(h) U(x), h with U ,= 0. Then, there exist
e C
, positive on R
n
, and a R
n
such that
lim
h
,h
c(h)/e(x +h) = exp((a x)) in c . (1.46)
Proof. We know (see Proposition 1.2) that U(x) = C exp(a x), a R
n
,
C ,= 0. Let
0
T such that U,
0
,= 0. We introduce functions
e
1
, e
2
, e
3
, e
4
and e in the following way: e
1
is a smooth function
e
1
(y) = T(x +y),
0
(x)/U,
0
, y R
n
.
Since
lim
h,h
T(x +h)/c(h),
0
(x) = U,
0
,
there exists
0
> 0 such that e
1
(y) > 0, y z
, |z|
0
.
Functions e
2
, e
3
and e
4
are dened as follows: e
2
(x) = maxe
1
(x), 0,
x R
n
, e
3
is the characteristic function of the set C
R
nx
, |x|
0
,
e
4
= e
2
+e
3
. Thus e
4
is positive and locally integrable.
Let w be a non-negative function belonging to (
such that w(x) = 1

for x B(0, 1), w(x) = 0 for |x|> 2, and
_
R
n
exp((a x))w(x)dx = 1.
Now, we can construct the sought function e (
: e = (e
4
w)(x),
x R
n
. The function e(x) is positive because of
e(x) =
_
R
n
e
4
(x t)w(t)dt
_
B(0,1)
e
4
(x t)dt > 0, x R
n
.
It only remains to prove (1.45). The number m := d(pr
, C
R
n int ) is
positive because pr
is a compact set in R
n
and C
R
n int is a closed one.
Every ball B(b, m
), 0 < m
< min (m, 1), b pr
is contained in int
and for every > 0 the ball B(b, m
), 0 < m
< min(m, 1), is contained

in int , as well.
For every compact set K R
n
, K B(0, ) and every
0
R
+
there
exists > 0 such that x + h y
; |y|
0
for x K and h y
, y|> . Since h
, h can be written as h = b, b pr
, > 0.
Thus B(b, m
) for every > 0. If > /m
, then x+h B(b, m
)
for x K B(0, ) and h = b because |x + h b|= |x| < m
.
Now,
|x +h| [|x||h|[ (1 m
).
Then, we can take = max/m
,
0
/(1 m
).
For a compact set K
0
R
n
the set K = K
0
B(0, 2) is also a compact
set and belongs to a ball B(0, ). Let us suppose that we have found
which corresponds to K and
0
as above. Then, by the denition of e
4
we
have e
4
(x t +h) = e
1
(x t +h) for x t K; h
, |h| . Hence
lim
h
,h
e
4
(x t +h)/(c(h)) = lim
h
,h
_
T(y +x t +h)
c(h)U,
0
,
0
(y)
_
= exp(a (x t))
g
81
and this limit is uniform for xt K. Now, for x K
0
the following limit
lim
h
,h
e(x +h)
c(h)
= lim
h
,h
_
e
4
(t +h)
c(h)
w(t)
_
(x)
= (exp(a t) w(t))(x) = exp(a x)
is uniform, as well.
Proposition 1.16. Let be a convex cone with nonempty interior,
_
,=
, and let
be a closed cone such that

B(0, 1) int . Let c ()
and T T
. If for U(x) = C exp( x), x R, C R, C ,= 0,

T(x +h)
s
c(h) U(x), h , U ,= 0,
then there exists e (
such that e(x) ,= 0, x R

n
, and T(x)
gs
Ce(x), |x|, x
.
Proof. Let e be as in Lemma 1.4. It is enough to apply Proposition 1.2
and Theorem 1.2,b) to
T(x +h)/e(x +h), (x) =
_
c(h)
e(x +h)
T(x +h)
c(h)
, (x)
_
, T.
The next proposition also gives a relation between the S-asymptotics
and the generalized S-asymptotics.
Proposition 1.17. Suppose that T T
, = x R
n
, x = (0, . . . , x
k
,
0, . . . , 0), T/c = D
x
k
S and T(x)
gs
ac(x), |x| , x . Then S(x+h)
s
h
k
1, h .
Proof. By LHospitals rule with Stolzs improvement, we have (with
h = (0, . . . , h
k
, 0, . . . , 0) and S (h) = S(x +h), (x))
lim
h
k
S(x +h)/h
k
, (x) = lim
h
k
(S

(h))
h
k
= lim
h
k
_
T
c

_
(h)
= lim
h
k
_
T(x +h)
c(x +h)
, (x)
_
= a, (x).
Finally, let us point out that if we take the limit
lim
h,h
T(x +h)/(c(h)e(x +h)),
in Denition 1.6., then nothing new is obtained (cf. Denition 1.6) for the
following three cases: the one-dimensional case, when R
n
is a ray, and
when has nonempty interior.
2 Quasi-asymptotics in F
2.1 Denition of quasi-asymptotics at innity over a cone

Quasi-asymptotics has been originally dened and studied for tempered dis-
tributions. The motivation for such a choice can be found in Theorem 2.3
below. The rst paper dealing with the analysis of the quasi-asymptotics
was written by Zavyalov [200]. Thereafter, many results concerning the
theory and applications of this notion have appeared. The main features of
the theory (published until the year 1986) have been collected in the mono-
graph [192]. We start with the most general denition of quasi-asymptotics
of tempered distributions ([192]).
Let be a closed convex acute solid cone (cf. 0.2) in R
n
and let
U
k
; k I R be a family of linear nonsingular transforms of R
n
which
leave the cone invariant (automorphisms of ). We assume that J
k
=
det U
k
> 0 and that I has as a limit point. Furthermore, let (k) be
a positive function dened on I. Denote by o
the space of tempered

distributions with supports in the cone . It is important to mention that
o
is isomorphic to the space o
() (see 0.5).
Denition 2.1. Let T o
. It is said that T has the quasi-asymptotics

in the cone over the family U
k
; k I related to the positive measurable
function (k) if there exists a tempered distribution g ,= 0 such that
lim
kI,k
1
(k)
T(U
k
x) = g(x) in o
.
We will discuss the quasi-asymptotics in a particular case, namely, if
T T
, U
k
= k and I = (0, ), where is a closed, convex and acute
cone in R
n
(see [33]). For results related to Denition 2.1, see [192].
Denition 2.2. Let T T
. It is said that T has the quasi-asymptotics in

the cone related to a positive measurable function if there exists g ,= 0
such that
w.lim
k
1
(k)
T(kx) = g(x) in T
.
We write in short T(kx)
q
(k)g(x), k in T
; we will omit the

space from the notation whenever it is clear from the context.
2. Quasi-asymptotics in F
83
The denition itself forces and g to have very specic forms. The
following proposition states precisely to which classes of functions and dis-
tributions and g must belong.
Proposition 2.1. Let T T
. If T(kx)
q
(k)g(x), k , then:
a) is a regularly varying function of the form (k) = k
L(k), k k
0
,
(see 0.3) where R and L is a slowly varying function.
b) g is a homogeneous element of T
with degree , that is, g(kx) =

k
g(x), k > 0.
Proof. We take T
such that T, ,= 0. Let K (0, ) be any

compact set. Then, by the assumption, we have
lim
k
_
T(ktx)
(kt)
, (x)
_
= g(x), (x) (2.1)
uniformly in t, provided that t belongs to the compact set K. On the other
hand, we have uniformly in t K
lim
k
_
T(kx)
(k)
,
(x/t)
t
_
=
_
g(x),
(x/t)
t
_
. (2.2)
Combining relations (2.1) and (2.2), we obtain for t > 0,
lim
k
(kt)
(k)
= lim
k
T(kx),
(x/t)
t

T(ktx), (x)
=
g(x),
(x/t)
t

g(x), (x)
= C(t)
uniformly in t K. This is just the denition of a regularly varying function
(see 0.3). We thus have C(t) = t
, t > 0, for some R; therefore, has

the desired form.
In order to prove that g is homogeneous of degree , we have to use
(2.1):
g(tx), (x) = lim
k
_
T(ktx)
(k)
, (x)
_
= lim
k
(kt)
(k)
_
T(ktx)
(kt)
, (x)
_
= t
g(x), (x), T
, t > 0 .
Remark. It is easy to prove that in case T
= o
R
+
o
+
, then
g = Cf
+1
, where f
is the homogeneous tempered distribution of degree

1 given in 0.4.
We emphasize that that all homogeneous distributions on the real line
are explicitly known. Recall ([56]), they are of the form
g(x) = C
+C
+
x
+
if , N,
g(x) =
(k1)
(x) + x
k
if = k N
(2.3)
where C
, C
+
, C and the distribution x
k
stands for the standard
regularization of the corresponding function [56, 63], i.e., x
1
= (log [x[)
,
k x
k1
= (x
k
)
. We refer to ([40, 56, 63]) for the explicit form of

multidimensional homogeneous distributions.
An more general analysis of homogeneous generalized functions can be
found in [85], but only in one dimension. We comment only part of it.
Let W
(I) be an abstract locally convex function space whose elements

are dened on I = (, ) or on I = (0, ) and for which
()
_
y
_
, y > 0 ,
is a continuous mapping from W
(I) to W
(I). If an element g W
(I),
where W
(I) is the dual space of W
(I), satises the equation

g(y) = y
g(), y > 0, R,
in the sense that g(yx), (x) = y
g(x), (x) for all W
(I), then g
is called a homogeneous generalized function from W
(I) of degree .
Let us suppose that W
(I) is given by the subspace of (
(I 0) for
which all the seminorms
p
n,
() =
_
I
[x[
x
n
(n)
(x)
dx, n N
0
are nite. In particular p
0,
is a norm.
One can easily prove that for a given W
(I) the sequence

n
nN
,
where
n
is an even positive function in T(I) such that
n
(x) = 0 for
x [0, 1/2n] [n + 1, ) and
n
(x) = 1 for x [1/n, n], n N, converges
to in W
(I) as n . This implies that the space T(I 0) is dense

in W
(I). Thus, all homogeneous generalized functions of order / N in

W
(I) are of the form

A
1
x
+
+A
2
x
resp. A
1
x
+
if I = R resp. I = (0, ) . (2.4)
85
Let |
(I), R, be a subspace of W
(I) with a D EI Qq3]TJ true2 D4933(hh.8 12492.8 cmBI/IM BPC 1ID G0 gq16e742(j)6.68793(1)-2.41true/W 1/H 14371/BPC 1ID EIq3]TJ tr Qq 6 0 0 -4r)/R12dG0 gq16e742(j)6.68ri1461/; ej 627758 Tf6.84 2.88 Td[(;)Tjj/R27 9d[(W)4.4904176 0 -31/BPC 15.70074.76 0 51 .1581()99(w)0214.802 9.96264 .15997(o)-5.833[(W)4.4904141.66.802 9.96264 0 086326 -1.44 Td[()-6.87I
IeFj a84471]TJ/RR12m T-0 0 0 -6 8663.8 1214.]TJ/R149 5.97758 Tf41/2.88 Td[(0)3.64889]TJ/R16 7.97011 Tf215.52 -2.88 Td[(8)4.0324(5)4.0324]TJ/R12 9.//R25 9.9622q 6 0 0 -4.4q 6 0 0 -424028]TJ/rueA850129 G0 gq1628]TJ/true 1ID9748510 0 0 -6 86f31369Tf9.36 -1.485 Tf6.2eID EI.64 0 Td39.8 1090.8 r j j0
0 0 12546 8589 92(e742(j)6.68r)/R125666izr Q 8582(=.105(D49.)-54633742(j)6.68H 9 9 QTJ2BI/IM042186(0)89]TJe742(j)45563]TJ2BI/IM )6.68H 1(o)-36.11638140 - 091/BPC6241 EI Qq 400 0 0 -6 865.8 1214.8 cmBI/IM true/W 1/H 1/BPC 1ID EI Qq 6r [(I)5.370 0 12546 8589 92(49.)-54633742(j)6.68H 4W92I QTJ2BI/IM042186(0)89]TJ968r8./2(=.10512546 85841/6 0 02(d5.72 435(o)-36.1748140 - 508(oPC6241 EI Qq 400 0 0 -6 865.8 1214.8 cmBI/IM true/W 1/H 1/BPC 1ID EI Qq 6r)/R1256/H 1 3203369d5.72 435(49.)-54613W2 6 4902(e742(j)6.68r)/R125666izr Q 8582(=.105(D49.)-546e742(j)105.38143]TJj)6.68H 15540r Q414(a)-331.1056.68rTLT*(a)-331.1056rTLT*(a)-331.105T*(a)-331.1051/e742(j)6.68H -)3PC 482.88 T49.6 0 068 G0 g33742(j)6.68H 4W.38143]TJ=e/W56Tf989]TJ/R16 7.97011 Tf1T28]49.6 0 068 G0 g83 481.38(c)4I 00BI4IMr2/W 1i)-2.23675(n)0.24i Wn
convex solid acute cone is denoted by
1,
. The function
K
C
(z) =
_
e
izt
dt, z R
n
+iC
is called the Cauchy-Szego kernel of the tube domain R
n
+iC. The closed
convex acute solid cone is regular if K
C
(z) is a divisor of unity in the
Vladimirov algebra 1(C) = /(o
()), where / denotes the Laplace trans-

form (cf. [189], 12]). In case n = 1, 2, 3 it is well-known (see Chapter 1,
2.7 in [192]) that all (closed convex acute solid) cones are regular. For
regular cones the distribution
,
is given by
_
e
izt
,
(t)dt = K
C
(z), z R
n
+iC, R.
This distribution has many interesting properties (see [33]):
1) supp
,
and
,
o
;
2)
,
(kt) = k
n(1)
,
(t), t , k > 0;
3)
,

,
=
+,
;
4) For every m 0 there exists n
0
such that
,
C
m
(R
n
), > n
0
;
5) supp
,
(e t) |t|< R, where R > 0 do not depend on
and e pr = |x|= 1.
6) For any p N
0
, there exists N
0
> 0 such that for some A and q > 0
and for every n N
0
, |
n,
(x t)|
p,
A(1 + |x|)
q
, x (cf. 0.5 for
| |
p,
).
Operations of fractional derivatives (for 1) and fractional integrals
(for > 1) can be dened on o
via
,
: T
,
T ,
the latter denes a continuous linear operator on o
. We will use the

notation T
()
for
,
T. By the properties of the convolution, we have
T
()
(t) = T(),
,
(t ) ,
if is suciently large.
Denition 2.3. Suppose that f L
loc
(). It is said that f has an asymp-
totic behavior in the cone related to the positive function (k), k (0, )
if there exists a function g ,= 0 such that
lim
k
f(kx)/(k) = g(x), x , (2.5)
87
[f(kx)/(k)[ h(x), x , k k
0
0 , (2.6)
and (x)h(x) L
1
(), for every T.
Proposition 2.3. If a locally integrable function f has the asymptotic be-
havior in the cone related to (Denition 2.3), and if f denes a regular
element of T
, then f(kx)
q
(k)g(x), k .
Proof. We have
lim
k
f(kx)/(k), (x) = lim
k
_
f(kx)(x)/(k)dx, T.
The assumption over f allows one to exchange the limit with the integral,
by the Lebesgue theorem.
Proposition 2.4. Let T have the property that
_
t
k
_
; k k
0
> 0
is bounded in T for each T. Suppose that S, T T
, and (k) =
k
L(k), > n, or = n and L(k) as k .

a) If S has compact support, then
w.lim
k
S(kx)/(k) = 0, in T
.
b) If T = S on |x|> R for some R > 0, and S(kx)
q
(k)g(x), k , then T(kx)

q
(k)g(x), k , as well.
Proof. a) For every T
S(kx)/(k), (x) =
1
k
n
(k)
_
S(x),
_
x
k
__
.
The set (x/k); k k
0
> 0 is a bounded set in T. Consequently, the set
f(x), (x/k); k k
0
> 0 is bounded in R, too. Since k
n
(k) , k
, the assertion in a) follows.
b) By assumption, T S has a compact support. Then, by a), for each
T
lim
k
T(kx)/(k), (x) = lim
k
(T S)(kx)/(k), (x)
+ lim
k
S(kx)/(k), (x)
= lim
k
S(kx)/(k), (x)
which proves b).
Remark. Proposition 2.4 b) asserts that the quasi-asymptotics related
to (k) = k
L(k), > n, is a local property. If < n, then the quasi-

asymptotics has no longer this property. The next example, in o
(R),
illustrates this fact: If f =
(m)
+ (x
+
), m N
0
, / N, then f has
quasi-asymptotics related to (k) = k
q
, where q = max(m 1, ) (see
Examples 4 and 7 in 2.3). On the other hand, if = 1, then this
property depends on L. See also Example 5 of 2.3 in relation to Proposition
2.4 b).
Proposition 2.5. Let m = (m
1
, . . . , m
n
) (N
0
)
n
and T T
. If
T(kx)
q
(k)g(x), k and x
m
M
()
then
(kx)
m
T(kx)
q
k
|m|
(k)x
m
g(x), k .
Proof. We have
lim
k
_
(kx)
m
T(kx)
k
|m|
(k)
, (x)
_
= lim
k
_
T(kx)
(k)
, x
m
(x)
_
= g(x), x
m
(x) = x
m
g(x), (x), T.
Remark. If g is a distribution with support 0, then t
m
g(t) may be
equal zero.
One of the most useful and frequently used theorems that characterize
the quasi-asymptotic behavior is the following one.
Theorem 2.1. Let T o
and be a regular cone. Then, T has quasi-

asymptotics in related to if and only if there exists R
+
such that
T
()
has the quasi-asymptotics in related to k
n
(k).
Proof. By the mentioned properties of
,
,
1
k
n
(k)
T
()
(kx), (x) =
k
n(+1)
(k)
T
()
(x), (x/k)
=
k
n(+1)
(k)
T(x),
,
(), ((x +)/k)
=
1
(k)
T(kx),
,
(), (x +),
89
where o(). Since the cone is regular, the mapping: (x)
,
(t),
(x + t) is an automorphism of the space o(). In order to complete the
proof, it is enough to take limit in the last equality.
The same theorem can be proved in the same way for o
if (M.1),
(M.2) and (M.3) are satised.
We now give a structural theorem for quasi-asymptotics on cones.
Theorem 2.2. Suppose that is a regular cone. Then T o
has quasi-
asymptotics in related to if and only if there exists an integer N such
that T
(N)
is a continuous function and has the asymptotic behavior in
related to k
nN
(k) (see Denition 2.3).
Proof. The suciency follows from Theorem 2.1. Let us prove the
necessity.
Since T(kx)/(k) converges to g in o
(in o
()), there exists p N

0
such that it converges in o
p
(), as well (see 0.5.1). Also, by properties 4)
and 5) of
,
there exists N
0
N such that for every t R
n
and N N
0
the function
N,
(t x) o
p
(). Then T
(N)
is continuous and
lim
k
1
k
nN
(k)
T
(N)
(kx) = lim
k
1
k
nN
(k)
T(t),
N,
(kx t)
= lim
k
T(kt)/(k),
N,
(x t)
= g(t),
N,
(x t) .
By property 6) of
,
, there exist A and q > 0 such that
1
k
nN
(k)
T
(N)
(kx)
C|
N,
(x t)|
p,
A(1 +|x|)
q
, x . (2.7)
In this way we proved that T
(N)
has the asymptotic behavior in related
to k
nN
(k) (cf. Denition 2.3) .
We shall quote a remark given in [192]:
In many cases it is important to know exactly which primitive of a dis-
tribution f, having quasi-asymptotics, already has an asymptotics. There
is no simple or universal criterion. For instance, it would be natural to
suppose that the condition f o
p
() guarantees the existence of such an
N (depending perhaps on p, on a family U
k
, k I and on a function
(k), k I) for which the primitive f
(N)
has the asymptotics.
The following example shows that it is not the case, in general:
Let = R
+
and let f
n
() = H() exp(i
1/n
); n = 1, 2, . . . be a family
of functions in o
+
. Each function f
n
has the quasi-asymptotics i
n
n! ()
over the family of transforms U
k
= k; k > 0 related to the function
(k) = 1/k, k > 0 (cf. Denition 2.1). For every n N the function f
(N)
n
has the asymptotics over the family of transforms U
k
= k, k > 0 with
respect to the function k
N1
, k > 0 if and only if N 1 > ((n 1)/n)N;
that is, when N > n and, hence, N , as n . On the other hand,
all the functions f
n
, n = 1, 2, . . . are innitely dierentiable for > 0 and
uniformly bounded (cf. [192]).
We prove now that the space of tempered distributions is naturally
related to the concept of quasi-asymptotic behavior at innity.
, supp T , where is regular. If

lim
k
_
T(kx)
(k)
, (x)
_
= g, , T (2.8)
for some regularly varying function and g ,= 0, then T o
, and it has
quasi-asymptotic behavior at innity related to .
Proof. By property 5) of
,
there exists a suciently large R > 0
such that
m,
(e x) = 0 for e pr , x and |x|> R. Let T be
such that (x) = 1 for |x| R. Then,
_
(x)
T(kx)
(k)
_
, k N, is convergent
in o
(). Hence there exists a p N such that it converges in o
p
() (see
0.5.1). By properties 4) and 5) of
,
, we can nd a suciently large
m N such that
lim
k
1
(k)
(x)T(kx),
m,
(e x) = g
m
(e),
where
m,
(ex) is in o
p
(). Now the left-hand side can be written without
since
m,
(e x) = 0 for [x[ R, and this implies that
lim
k
1
(k)
1
k
mn
T
(m)
(ke) = lim
k
1
(k)
T(kx),
m,
(e x) = g
m
(e) (2.9)
and the last limit exists.
As in Theorem 2.2, it follows that T
(N)
L
1
loc
and that the limit in
(2.5) for f = T
(N)
and with k
Nm
(k) instead (k) exists. Also (2.6) is
satised.
91
Consequently, T
(N)
is a locally integrable function having the asymp-
totic behavior related to k
nN
(k) in the cone . Now, we can use Theorem
2.2 to complete the proof.
We now discuss some results in the context of ultradistributions.
Theorem 2.4. Suppose that M
p
satises (M.1), (M.2) and (M.3). If f
c
(R) and supp f [0, ), then there exists a p N

0
such that
lim
k
k
p+1
f(kx), (x) = C, o
[0,)
,
where C can be zero, too.
Proof. By Theorem 10.3 in [79], there exist an ultradierential operator
P(D) =
n=0
a
n
D
n
of class and a compactly supported continuous function
G, supp G = K suppf, such that
f =
n=0
a
n
D
n
G.
Suppose that a
p
,= 0 and a
n
= 0, n < p. We have to analyze
k
p+1
f(kx), (x) =
_
k
p+1
_

n=0
a
n
D
n
G
_
(kx), (x)
_
= k
p
n=p
(1)
n
a
n
k
n
_
G(x),
(n)
_
x
k
__
= (1)
p
a
p
_
K
G(x)
(p)
_
x
k
_
dx
+
n=p+1
(1)
n
a
n
k
n+p
_
K
G(x)
(n)
_
x
k
_
dx, o
[0,)
.
Since
_
K
G(x)
(p)
_
x
k
_
dx
_
K
G(x)dx
(p)
(0), k ,
it remains to prove that
n=p+1
(1)
n
a
n
k
n+p
_
K
G(x)
(n)
_
x
k
_
dx
n=p+1
k
n+p
_
K
[G(x)[
L
n
M
n
dx
C
1
n=p+1
k
n+p
_
K
[G(x)[dx 0, k ,
in both cases, (M
p
) and M
p
.
Theorem 2.5. Suppose that (M.1), (M.2) and (M.3) are satised. Let
T T
(R), supp T [0, ) and c(k) = k
L(k), k > 0, > 1. If

T(x + h)
s
c(h)U, then T o
+
and T has the quasi-asymptotics related
to c, as well.
Proof. Without loss of generality, we can suppose that c is continuous
in (0, ) (cf. Remarks after Proposition 1.2).
Denote by a function belonging to T
such that (x) = 1, x

[0, h
0
], h
0
> 0. Then, T = T + (1 )T. The support of T is compact,
therefore T c
. By Theorem 2.4
lim
k
_
(kt)T(kt)
c(k)
, (t)
_
= 0, o
[0,)
.
The S-asymptotics is a local property if > 1 (Theorem 1.2, c)).
Therefore (1 )T has the same S-asymptotics as T and supp(1 )T
[h
0
, ).
We shall use Theorem 1.10 with the assumptions that the set A of
that theorem satises A + R
+
[h
0
, ), h
0
> 0, ( = R
+
) and that
supp f
i
(h
0
, ), h
0
> 0, i = 1, 2.
Denote by g
i
(h) = f
i
(h)/c(h), h (h
0
, ). Then f
i
(h) =
c(h)g
i
(h), h h
0
> 0, i = 1, 2, and lim
h
g
i
(h) = C
i
, i = 1, 2. Therefore,
((1 )T)(x) = P(D)c(x)g
1
(x) +c(x)g
2
(x), x > h
0
> 0 .
It follows that (1 )T belongs to o
. Consequently, T o
. Suppose
now that o
[0,)
. Then
lim
k
_
(1 (kt))T(kt)
c(k)
, (t)
_
= lim
k
__
c(kt)g
1
(kt)
c(k)
,
i=1
(1)
i
k
i
a
i
_
d
dt
_
i
(t)
_
+
_
c(kt)
c(k)
(a
0
g
1
(kt) +g
2
(kt)), (t)
__
. (2.10)
93
By the properties of g
1
and g
2
, the last summand in (2.10) has a limit for
every o
[0,)
. More precisely,
lim
k
_
c(kt)
c(k)
(a
0
g
1
(kt) +g
2
(kt)), (t)
_
=
_
(a
0
C
1
+C
2
)x
, (x)
_
.
Since
_
c(k)g
1
(k)
c(k)
; k 1
_
is a bounded set in o
[0,)
and since
i=1
_
1
k
_
i
a
i
_
d
dx
_
i
(x)
tends to zero in o
[0,)
as k , the rst summand in (2.10) tends to zero
as k .
From the existence of the S-asymptotics of T related to c, it follows that
a
0
C
1
+C
2
,= 0.
Thus, we have proved that T o
+
and that T has the quasi-
asymptotics related to c.
Let o
[0,)
and let f
, R, be the function dened in 0.4.

Then (f
)() = (), 0 belongs to o
[0,)
. One can prove that the
mapping f
is an automorphism of o
[0,)
. The method for proving
this property of f
is the same as for the space o[0, ).

Let f o
[0,)
. Recall, we denote by f
(m)
an element belonging to
o
[0,)
dened by
f
()
(x), (x) = f(x), (

f
)(x), o
[0,)
.
Proposition 2.6. Suppose that f o
[0,)
and that is a real number.
Then, f has quasi-asymptotics related to if and only if f
()
has the
quasi-asymptotics related to k
(k).
Proof. By the denition of f
()
,
f
()
, = f(), (

f
)()
and this exists for every o
[0,)
. Therefore, for every o
[0,)
and
k > 0, we have
_
f
()
(k), ()
_
=
1
k
_
f
()
(),
_
k
__
=
_
k
k
f(), (

f
)
_
k
__
= k
f(k), (

f
)().
Since the mapping:

f
is an automorphisms of o
[0,)
, this
completes the proof.
Theorem 2.6. Suppose that (M.1), (M.2) and (M.3) are satised and that
c(h) = h
L(h), h > h
0
, 1. If T T
(R), T(x + h)
s
c(h)U, h
R
+
and supp T [0, ), then T o
+
.
If = 1 and

L(x) =
x
_
h
1
t
1
L(t)dt , as x , then T has
quasi-asymptotics related to t
1
L(t). In the other cases, T has the quasi-

asymptotics related to t
p
, p N, but the limit may be zero.
Proof. Let w be the function used in Theorem 2.5. Then T = wT +
(1w)T. Since wT has a compact support, by Theorem 2.4, wT has quasi-
asymptotics related to c(k) = k
n
for an n N.
The support of (1 w)T belongs to [h
0
, ), h
0
> 0. Since the S-
asymptotics is a local property, (1 w)T has the S-asymptotics related to
c, as well, and
((1 w)T)(t) =
i=0
a
i
D
i
(t
L(t)E
1
(t)) + (t
L(t)E
2
(t)), t > h
0
, (2.11)
where supp E
i
(h
1
, ), 0 < h
1
h
0
and lim
t
E
i
(t) = C
i
, i = 1, 2. We
know that a
0
C
1
+C
2
,= 0 because (1 w)T has the S-asymptotics related
to h
L(h).
We can choose h
0
and h
1
in such a way that a
0
E
1
(t) + E
2
(t) does not
change the sign when t (h
1
, ). Using (2.11) and the convolution with
f
1
(cf. 0.4) it follows
((1 w)T)
(1)
= f
1
(t
L(t)(a
0
E
1
(t) +E
2
(t)))
+
i=1
a
i
D
i
(f
1
(t
L(t)E
1
(t))).
We have to analyze the function F = f
1
(t
L(t)(a
0
E
1
(t) + E
2
(t))).
This function is equal to zero in (0, h
1
) and
F(x) =
x
_
h
1
t
L(t)(a
0
E
1
(t) +E
2
(t))dt, x h
1
.
95
Case 1. Let < 1 or = 1 and
_
h
1
x
1
L(x)dx < . Then,
F(x)
_
h
1
t
L(t)(a
0
E
1
(t) +E
2
(t))dt, as x .
Consequently, F has S-asymptotics related to 1. Since
_
h
1
t
1
L(t)E
1
(t)dt
< , ((1 w)T)
(1)
has S-asymptotics related to 1, too. By Theorem 2.5,
((1 w)T)
(1)
has quasi-asymptotics related to 1, and by Proposition 2.6,
(1 wT) has quasi-asymptotics related to h
1
.
Case 2. Let = 1 and
_
h
1
t
1
L(t) , as x . Then, F(x) =
L(x), where

L is also a slowly varying function (see Proposition 1.5.9 in
[9]); f has S-asymptotics related to

L(x). We have the same situation with
_
h
1
t
1
L(t)E
1
(t)dt =

L
1
(x). By Theorem 2.5, ((1 w)T)
(1)
has quasi-
asymptotics related to L
0
= max(
L,
L
1
) and by Proposition 2.6, (1 w)T
has quasi-asymptotics related to k
1
L
0
(k).
Taking care of the quasi-asymptotics of wT and (1 w)T, we have
established the assertion of Theorem 2.6.
The following example illustrates dierent possibilities in case < 1.
We use three functions
f(x) =
_
0, 0 x 1
x
1
, 1 < x, 1 < < 2,
and F(x) = H(x)H(1 x), x R; H is the Heaviside function.
The quasi-asymptotics are given by:
f(kx), (x) = k
1
a
_
1/k
x
1
(x)dx
=
1
_
1
k
_
1
k

1
1
1
_
1
k
_
1
k
2

1
1
1
1
k
1+
a
_
1/k
x
1
(x)dx, k > 0;
kF(kx), (x) = k
1/k
_
0
(x)dx =
1
_
0
_
t
k
_
dt (0), k ;
k
2
(DF)(kx), (x) = kF(kx),
(x)
(0), k , o
+
.
Take now the distribution T = f +C
1
F+C
2
DF. For appropriated constants
C
1
and C
2
, T can have the quasi-asymptotics related to k
1
, k
2
, k
1
.
The following problem is discussed in [38]: Let f o
+
and let
k
kN
be a sequence in o. Assume that the following limit exists
lim
k
1
(k)
f(kt),
k
(t) = c ,
where is a regularly varying function.
The question is to nd conditions under which the limit
lim
k
1
(k)
f(kt), (t) = c
exists for all o, i.e., that f o
+
has the quasi-asymptotics related to
.
The authors proved that
k
(t)
kN
cannot belong to o. But they con-
structed a new space o
a,M
b,N,
of test functions and the space (o
a,M
b,N,
)
of dis-
tributions in which the asked question has a positive answer. (cf. Theorem
7 in [38]).
The relation between the spaces o
+
and o
a,M
b,N,
is given by the following
projective limits:
M,MZ
+
o
a,M
b,N,
= o
a,
b,
o
a
b
,
a,bR
o
a
b
= o
+
.
The results of Theorem 7 in [38] can be applied in the analysis of
quasi-asymptotics of solutions to dierential equations, and to Abelian and
Tauberian type theorems for integral transforms of distributions.
2.3 Quasi-asymptotic behavior at innity of some
generalized functions
We examine rst the case when a regular distribution T is dened by a
locally integrable function F on R and has support in an interval [a, ), a >
0. We write T = H(x a)F, where H is the Heaviside function.
1. Let T = H(x a)F(x), x R, for a > 0, where F is a locally
integrable function satisfying
_
a
[F(x)[dx < . Then T has the quasi-
asymptotics related to k
1
.
97
It follows at once from
lim
k
kT(kx), (x) = lim
k
_
a
F(x)
_
x
k
_
dx = C, ,
where C =
_
a
F(x)dx.
We obtain special cases if either F(x) x
L(x) as x and < 1

or F(x) L(x)/x, x , provided that
_
a
[L(x)/x[dx < .
2. Let a > 0 and T(x) = H(x a)F(x), x R, where F is a locally
integrable function such that F(x) L(x)/x as x . If
L
(x) :=
x
_
a
L(t)
t
dt, x > a
diverges to innity as x , then T has the quasi-asymptotics related to
L
(k)/k, k .
L
is also slowly varying at innity (see 0.3 and [9]). Let

G(x) := (H T)(x) =
x
_
a
F(t)dt, x R.
Since
lim
x
G(x)
L
(x)
= lim
x
F(x)
L(x)/x
= 1
and
d
dx
G = T, the Structural Theorem 2.2 implies the claim.
3. Let a > 0 and T(x) = H(x a)F(x), x R, where F is a locally
integrable function such that F(x) x
L(x) as x for > 1. Then

T has the quasi-asymptotics related to (k) = k
L(k), k .
It is obvious, since G = (H T) is a continuous function on R such that
G(x)
_
x
+1
/( + 1)
_
L(x), as x .
Now we give the quasi-asymptotics of some distributions.
4. Denote by (x a), a 0, the delta distribution with support in a.
Then
(m)
(kx a)
q
k
m1
(m)
(x), k .
Indeed, observe that
(kx a), (x) =
1
k
_
a
k
_
=
(0) +O(1/k)
k

(0)
k
, k , o;
therefore, (kx a)
q
k
1
(x), k . The result now follows by dier-
entiating m-times the last quasi-asymptotics (cf. Theorem 2.1).
5. For every S c
+
there exists a natural number p such that S
has the quasi-asymptotics related to k
p
, k .
For a given S c
+
there exists m N
0
and a continuous function
G on R with supp G [0, ) such that S = D
m
G (D =
d
dx
). If supp S
[0, a], a 0, then we have that G is equal to some polynomial of the order
m 1 on the interval (a, ). Thus for some 0 q m 1 and some
C ,= 0
G(x) Cx
q
as x .
This implies that G has the quasi-asymptotics related to k
q
, k . The
Structural Theorem 2.2 implies that S has the quasi-asymptotics related to
k
qm
, k ; in fact S(kx)
q
Ck
qm
(mq1)
(x).
6. Let F be a locally integrable function on R0 equal to zero outside
of some interval [0, a], a > 0, such that
F(x) x
L(x) as x 0
+
,
where 1 and L is a slowly varying function at zero.
This function can be identied with a distribution S dened by
S, :=
a
_
0
F(x)((x) (0)
x
m1
(m1)!
(m1)
(0))dx (2.12)
if (m+ 1) < m, m N and o(R) (see [[138], p. 13]).
The distribution S dened by (2.12) has quasi-asymptotics related to
1
k
m+1
, k . Let us prove this. For o(R), we have
k
m+1
S(kx), (x)=k
m
a
_
0
F(x)
_
_
x
k
_
(0)
x
m1
k
m1
(m1)!
(m1)
(0)
_
dx
= k
m
a
_
0
F(x)
1
m!
_
x
k
_
m
(m)
_
x
k
_
dx, 0 <
x
< x,
hence
lim
k
k
m+1
S(kx), (x) =
(1)
m
m!

(m)
,
a
_
0
x
m
F(x)dx, as x .
99
Let us remark that if > 1, then the distribution S given by F is
regular and S(kx)
q
Ck
1
, k , with C =
_
a
0
F(x)dx.
7. Let L be slowly varying at zero and at innity.
The distribution R(x) = (x
L(x))
+
(see [138]) has the quasi-
asymptotics related to (x) = x
L(x), x , if , Z
= 1, 2, . . .
and related to
1
(x) = x
(x) if Z
, where L
(x) =
x
_
a
L(t)t
1
dt, x
a.
8. Let f denote a measurable function on R with support in [0, )
satisfying
f(x) x
L
1
(x) as x 0
+
, (2.13)
where L
1
is a slowly varying function at zero and m 1 < m. We
suppose that f satises the following additional condition
x
m
f(x) is integrable on (a, ), a > 0 . (2.14)
We denote by

f the following distribution in o
(R) dened by f :
f, =
_
0
f(x)((x) (0)
x
m1
(m1)!
(m1)
(0))dx, o(R) ,
(2.15)
if m1 < < m, and for some a > 0, and if = m,
f, =
_
0
f(x)((x) (0)
x
m2
(m2)!
(m2)
(0)
x
m1
(m1)!
(m1)
(0)H(a x))dx, o(R) . (2.16)
The distribution

f from (2.15), respectively (2.16), has quasi-
asymptotics related to k
m1
, resp. k
m
, provided that both (2.13) and
(2.14) hold.
9. Let f satisfy (2.13) for m1 < < m and
f(x) x
L
2
(x) as x ( < m),
where L
2
is slowly varying at innity. Then

f dened by (2.15) (m1 <
< m), respectively by (2.16) ( = m), has quasi-asymptotics related
to k
m1
and k
m
respectively.
Observe that this quasi-asymptotics does not depend on the functions
L
1
and L
2
.
10. Let f satisfy (2.13) for m1 < < m and
f(x) x
L
2
(x) as x ( > 0) , (2.17)
where L
2
is slowly varying at innity.
Then

f dened by (2.15) has the quasi-asymptotics related to x
L
2
(x).
We suppose in the next example, as usual, that M
p
satises (M.1),
(M.2) and (M.3).
The quasi-asymptotics of ultradistributions is a natural extension of the
same notion for distributions. Suppose that f o
[0,)
; it denes an ul-
tradistribution f o
[0,)
. If it has the quasi-asymptotics as a distribution
related to , then it has also the quasi-asymptotics as an ultradistribution
related to , and with the same limit.
On the other hand, the next examples show that there exist elements
of o
[0,)
which are not in o
[0,)
and have the quasi-asymptotics as ultra-
distributions.
We deal with the Beurling ultradistributions. In the case of Roumieus
ultradistributions the treatment is the same.
11. Let P(D) =
n=0
a
n
D
n
be an ultradierential operator of class (M
p
)
and let a
0
,= 0 and a
i
,= 0 for innitely many i. Then P(D) is an element
of o
(M
p
)
[0,)
which is not a distribution. We shall show that (P(D)) (kx)
q
a
0
k
1
(x).
By Denition 2.2, we have to consider the following limit:
lim
k
k(P(D))(kx), (x)
= lim
k
(y),
n=0
(1)
n
a
n
k
n
(n)
(y/k)
= a
0
(0) + lim
k
n=1
(1)
n
a
n
k
n
(n)
(0), o
(M
p
)
[0,)
.
It remains to show that the last limit equals zero. Since for every L > 0
sup
n
L
n
/M
n
sup
x[0,)
[
(n)
(x)[< C
1
,
101
it follows
n=1
(1)
n
a
n
k
n
(n)
(0)[ C
1
n=1
k
n
L
n
/M
n
[
(n)
(0)
C
2
k
1
0, k .
Consequently, (P(D))(kx)
q
(a
0
/k)(x) as k in o
(M
p
)
[0,)
.
12. There are also distributions which have no quasi-asymptotics as
distributions, but they have the quasi-asymptotics as ultradistributions. One
of such distributions is the following one
f =
n=0
(n)
(x e
n
)/M
n
.
Suppose that there exist and g o
[0,)
such that f(kx)
q
(k)g(x)
in o
[0,)
. Then, we would have
lim
k
1
(k)
_

n=0
(n)
(kx e
n
)/M
n
, (x)
_
= g,
for every o
[0,)
.
By Borels theorem, there exists
0
o
[0,)
such that supp
0

[e
1
, e], and the set exp(p
2
p)/((e
p
)M
p
)[
(p)
0
(1)[; p N is not
bounded in R. For this
0
and for the subset e
p
; p N, we have
1
(e
p
)
f(e
p
x),
0
(x)
1
(e
p
)
n=0
(1)
n
M
n
exp(np +p)
(n)
0
(e
np
)
=
exp(p
2
p)
(e
p
)M
p
[
(p)
0
(1)[ , p .
This is in contradiction to our assumption f(kx)
q
(k)g(x) in o
[0,)
.
This distribution has the quasi-asymptotics as ultradistribution; the
proof is the same as for P(D).
2.4 Equivalent denitions of quasi-asymptotics at innity
Let T o
+
. We denote by /T the Laplace transform of T: /(T)(z) =
T(t), e
izt
, z R + iR
+
(cf. [190], Ch. II, Part 9, [189] and [192], Ch.
I, Part 2). We collect in the following theorem some equivalent ways to
dene quasi-asymptotics over the cone R
+
. The part c) is a prototype of a
Tauberian characterization for quasi-asymptotics; Tauberian theorems for
various integral transforms will be the main subject of 4.2.
+
and (k) = k
L(k), k k
0
. The following
statements are equivalent:
a) lim
k
T(k )
(k)
= Cf
+1
, in o
(R), C ,= 0.
b) lim
k
T(k +b)
(k)
= Cf
+1
, in T
(R), C ,= 0, b R.
c) A) lim
y0
+
y
(1/y)
/T(iy) = M ,= 0;
B) there exist D
1
> 0, m N
0
, and r
0
> 0 such that
r
(1/r)
/T(re
i
)
D
1
sin
m
, 0 < r < r
0
, 0 < < .
d) For every T(R),
lim
k
(T )
(k)
(k ) = M
f
+1
, in T
(R), where M
,= 0.
e) There exists
0
T(R) with the property /
0
(0) ,= 0 such that
lim
k
(T
0
)(k )
(k)
= M
0
f
+1
, in T
(R), M
0
,= 0 .
f ) For a -sequence
n
n
(cf. 0.4) there is a C ,= 0 such that
lim
k
(T
n
)(k )
(k)
= Cf
+1
, in T
(R), and uniformly for n N.

Proof. a) b). We start with the relation
_
T(kx +b)
(k)
, (x)
_
=
_
T(kx)
(k)
,
_
x
b
k
__
, T(R) .
The set (
b
k
); k 1 is bounded in T
(R). By using the equivalence

of the weak and strong sequential convergence in T
(R) and the fact that

( b/k) , k , in T(R), we obtain
lim
k
_
T(kx +b)
(k)
, (x)
_
= lim
k
_
T(kx)
(k)
,
_
x
b
k
__
= Cf
+1
, , T(R).
103
By Theorem 2.3 the last limit holds in o
(R), as well.
b) a) Take b = 0 and apply Theorem 2.3.
a) c) and c) a) is proved in [192].
c) d) For a T o
+
, we have T o
[a,)
, a R. We shall show
that T satises c), it would imply that T satises a) and then T
would satisfy d). We have
lim
y0
+
y
(1/y)
/T(yi)/(iy) = lim
y0
+
y
(1/y)
/T(yi)/(0) = M
0
.
Moreover, there exist D
2
> 0, m N
0
and r
0
> 0 such that
r
(1/r)
/T(re
i
)/(re
i
)
r
(1/r)
/T(re
i
)
max
0<rr
0
0<
[/(re
i
)[
D
2
sin
m
.
d) e) It is obvious.
e) a) The assumptions in e) imply a) and consequently c) for T
0
.
Now, we have
lim
y0
+
y
(1/y)
/T(iy)/
0
(iy) = lim
y0
+
y
(1/y)
/T(iy)/
0
(i0) = M ,= 0.
Taking care of the property that /
0
(0) ,= 0, we have
r
(1/r)
/T(re
i
)
sin
m
,
where 0 < r r
0
, and 0 < for appropriate r
0
, D
and m
. This gives
c) for T and consequently a).
a) f) Let T(R) and n N. Then
_
(T
n
)(kx)
(k)
, (x)
_
=
1
k(k)
_
(T
n
)(x),
_
x
k
__
=
1
k(k)
_
T(x),
_
n

_
k
__
(x)
_
=
1
(k)
_
T(kx),
_
R
n
(t)
_
x
t
k
_
dt
_
.
The set
_
_
R
n
(t)
_
x
t
k
_
dt; n N, k (0, )
_
is bounded in T(R) because of the properties of
n
n
(cf. 0.4). Therefore,
we have
_
T(kx)
(k)
,
_
R
n
(t)
_
x
t
k
_
dt
_
Cf
+1
,
__
R
n
(y)dy
_
, k ,
uniformly for n N. Thus, for any > 0 there is a k
0
() such that
_
(T
n
)(kx)
(k)
, (x)
_
_
Cf
+1
,
_
T(kx)
(k)
,
_
R
n
(t)
_
x
t
k
_
dt
_
Cf
+1
(x),
_
R
n
(t)
_
x
t
k
_
dt
Cf
+1
(x),
_
R
n
(t)
_
x
t
k
_
dt Cf
+1
,
< , k > k
0
();
which proves a) f).
f) e) Since
n
(x) = 1, n N, we have /
n
(0) ,= 0. This implies
that e) holds.
2.5 Quasi-asymptotics as an extension of the classical
asymptotics
We have seen in Proposition 2.3 that if a locally integrable function f
has asymptotic behavior related to a function , then

f has the quasi-
asymptotics related to . The following theorem also goes in this direction.
Theorem 2.8. Let be a closed convex acute solid cone in R
n
, f o
()
L
1
loc
( R
n
; [[> R) for some R > 0 and let (k) = k
L(k), > n.
If, for any e pr (pr = x R
n
; x , |x|= 1), the limits
lim
||
1
([[)
f([[e) = g(e) ,= 0
105
exist, and for some g
1
L
1
(pr) the estimate
[f()[
([[)
g
1
_

[[
_
, [[> R, ,
is satised, then f has quasi-asymptotics related to .
The proof is similar to that of Proposition 2.3 (see Theorem 2, Chapter
I, 3.3. in [192]).
Remark. The condition > n in Theorem 2.8 is essential. This fact
is illustrated by the following two examples.
Let f(x) = H(x 1)x
1
, x R, > 0. Then f(x) x
1
, x
, but by Example 1 in 2.3 the regular distribution f has the quasi-
asymptotics related to (k) = k
1
for any > 0.
Let T(x) = H(x2)(xlog
2
x)
1
, x R. Then T(x) (xlog
2
x)
1
, x
. But by the same Example 1 in 2.3, T has the quasi-asymptotics related
to (k) = k
1
.
A more dicult question is to nd conditions under which the quasi-
asymptotics implies the ordinary asymptotics of f. A partial answer is given
in the ensuing theorem.
+
be equal to a locally integrable function f
in some interval [b, +), b > 0, with quasi-asymptotic behavior related to
(k) = k
L(k), > 1. If for some m N the function x

m
f(x), x b, is
monotonous, then f has the asymptotic behavior at innity related to the
same regularly varying function .
Proof. By Proposition 2.5, the function g(x) = x
m
f(x), x b, has
the quasi-asymptotic behavior at innity related to
1
(k) = k
+m
L(k).
The monotonicity of g implies that its distributional derivative Dg can be
written as Dg = B + , where B c
+
and is a positive measure with
support in [b, ). By Theorem 2.1, Dg has the quasi-asymptotics related
to k
+m1
L(k). From Proposition 2.4 a), we see that Dg and have the
same quasi-asymptotic behavior at innity, related to k
+m1
L(k), i.e.,
lim
k
1
k
+m1
L(k)
(kx), (x) = Cf
+m
, , o(R) . (2.18)
Choosing
and
in T(R) with the properties
(x) = 1, for [x[ 1

2
and
(x) = 0 for [x[ 1,

(x) = 1, for [x[ 1 and
(x) = 0 for [x[ 1 +

2
(0 < < 1), we obtain
1
k
+m1
L(k)
(kx),
(x)
1
k
+m1
L(x)
(kx), H(1 x)
1
k
+m1
L(k)
(kx)
(x) , (2.19)
since
(x) H(1 x)
(x) for x > 0 .

Using (2.18), we see that, both, the left and the right hand side of (2.19)
tend to numbers which do not dier more than . Thus the expression in
the middle of (2.18) tends to some limit independent of . Since (H ) is
equal to g(x) on (b
1
, ), b
1
b and
1
k
+m1
L(k)
(kx), H(1 x) =
1
k
+m
L(k)
(H )(k),
we conclude that g has ordinary asymptotic behavior at innity related to
1
, and this implies the statement.
Remark. It is easy to nd a function which has quasi-asymptotics but
not an asymptotic behavior. An example is given by f(x) = H(x) sin x, x
R; f(kx)
q
k
1
(x), k .
In such a way, the quasi-asymptotics extends the notion of the asymp-
totic behavior of a locally integrable function.
2.6 Relations between quasi-asymptotics in D
(R)
and S
(R)
In this section we make a preliminary investigation of some questions raised
by Theorem 2.3:
As in Theorem 2.3, suppose that T T
satises (2.8), with g T
, but we
now remove the assumptions over the support of T (we allow the support to
be any subset of R
n
). It then is natural to ask: Does T o
? Furthermore,
does the limit exist in o
? Let us observe that Theorem 2.3 does not give

an answer to such a question because the cone is acute and it cannot be
the whole R
n
.
107
The following theorem and its corollary give a partial answer to this
question. We will postpone the complete answer (in one dimension) for
2.10.
Theorem 2.10. Let f T
(R) and suppose that there exists, in the sense

of convergence in T
(R), the limit

lim
k
f(kx)/(k) = g(x) ,= 0 , (2.20)
where (k), k (0, ), is a positive continuous function on (0, ). Then,
(i) (x) = x
v
L(x), x (0, ), for some v R and some slowly varying
function L, and g is a homogeneous distribution with degree of homogeneity
v.
(ii) f o
(R).
(iii) If v > 1, then the limit (2.20) exists in the sense of convergence
in o
(R).
(iv) If v = 1 and 1/L(x), x (a, ), is bounded, then the limit (2.20)
exists in the sense of convergence in o
(R), as well.
It should be noticed that the proof is quite dierent from that of Theo-
rem 2.3 where we used the fact that T
and o
are convolution algebras.

The absence of the convolution algebra structure makes the argument more
complex.
Proof. (i) Let T(R) such that g, ,= 0. We have
lim
k
f(kmx)/(k), (x) = g(mx), (x);
lim
k
((mk)/(k))f(kmx)/(km), (x)
= g(x), (x) lim
k
((mk)/(k)), m > 0 .
This implies that, for every m > 0, we have the existence of the limit
lim
k
((mk)/(k)) = d(m) .
By [[148], p. 17], we obtain that for some v R and some slowly varying
function L, (k) and d(k) are of the form
(k) = k
v
L(k), d(k) = k
v
, k (a, ) .
Since g(mx), (x) = m
v
g(x), (x), m > 0, we obtain that g is homoge-
neous of degree v.
(ii) The set f(kx)/(k
v
L(k)); k > 0 is a bounded subset of T
(R).
From Theorem XXII, Chapter VI in [146], it follows that for a given open
bounded neighborhood of zero there exists a compact neighborhood of
zero K and a non-negative integer m such that for any T
m
K
(R)
x ((f(kt)/(k
v
L(k))) (t))(x), k (0, ) , (2.21)
is a family of functions which are continuous and uniformly bounded on
. Let = (2, 2) and K = [, ]. Since the weakly bounded family
(2.21) is strongly bounded in T
m
K
(R), we obtain that for every bounded set
A T
m
[,]
, the set of functions
x ((f(kt)/(k
v
L(k))) (t))(x); k > 0, A
is a bounded family of continuous functions on . Let T
m
[,]
and let
k
(x) = (kx)/k
m
, x R, k 1 .
Since supp
k
(x) x; [x[ /k [, ], we have that A =
k
(x); k
1 is a bounded family in T
m
[,]
and that
((f(kt)/(k
v
L(k)))
r
(t))(x); k > 0, r 1
is a bounded family of continuous functions on . Taking r = k, we obtain
that for some M > 0
[ ((f(kt)/(k
v
L(k))) ((kt)/k
m
))(x) [ M, x (2, 2), k 1 .
From
(f(kt) (kt))(x) = f(kt), (k(x t)) = k
1
f(t), (kx t)
= k
1
(f )(kx),
we obtain that
[ (f )(kx)/(k
v+m+1
L(k)) [< M for x (2, 2), k 1 .
Taking x = 1 and x = 1 it follows that for any T
m
[,]
there exists
M
> 0 such that

[ (f )(x) [ M
(1 +[x[
v+m+1
L([x[)), x R.
By (VI, 6; 22) in [146], we obtain
f =
d
2s
dx
2s
(E f) f , (2.22)
109
where E is the fundamental solution of d
2s
E/dx
2s
= , T
[,]
, 1
in a neighborhood of zero and T
[,]
. If s is large enough, then E
T
m
[,]
. Thus (2.22) implies that f o
(R).
(iii) We can rewrite (2.22) in the form f = d
2s
f
1
/dx
2s
+f
2
, where f
1
is
a continuous and f
2
a smooth function such that
sup[f
1
(x)[, [f
2
(x)[ M(1 +[x[
v+m+1
L([x[)), x R,
for some M > 0.
Since
f(kx)/(k
v
L(k)) =
__
d
2s
dx
2s
f
1
_
(kx) +f
2
(kx)
_
/(k
v
L(k))
=
d
2s
dx
2s
(f
1
(kx))/(k
v+2s
L(k)) +f
2
(kx)/(k
v
L(k)),
we obtain, for s large enough that for any T(R) :
lim
k
f(kx)/(k
v
L(k)), (x)
= lim
k
f
2
(kx)/(k
v
L(k)), (x) = g(x), (x) . (2.23)
Let us set
f
2+
(x) =
_
f
2
(x), x > 0
0, x 0
, f
2
(x) =
_
f
2
(x), x < 0
0, x 0.
Clearly, for any T(0, ) (supp (0, )),
lim
k
f
2
(kx)/(k
v
L(k)), (x) = lim
k
f
2+
(kx)/(k
v
L(k)), (x) .
If T(0, ), then (t) = (e
t
)e
t
, t R, is an element in T(R).
Moreover, the mapping dened above is a bijection. Since for
T(0, ),
_
0
f
2+
(kx)(x)dx =
f
2+
(ke
t
)(e
t
)e
t
dt,
by putting k = e
r
, r R, we obtain that for a function F
1,2
(t) =
f
2+
(e
t
), t R, there exists the limit
lim
r
F
2,1
(t +r)/(e
vr
L(e
r
)), (t)
for any T(R). Using again Theorem XXII, Chapter VI in [146], we
obtain that for any open set 0 there exists a compact neighborhood of
0 and a non-negative integer m such that for any T
m
K
(R),
x (F
2,1
(t +r) (t))(x)/(e
vr
L(e
r
)), r 0,
is a bounded family of continuous functions on . Since (F
2,1
(t + r)
(t))(x) = (F
2,1
(t) (t))(x + r), x R taking x = 0 and using (VI;
6; 22) in [146], we obtain
F
2,1
=
d
2l
dx
2l
H
2,1
+G
2,1
,
where H
2,1
is a continuous function and G
2,1
is a smooth function on R
such that
sup[H
2,1
(t)[, [G
2,1
(t)[ < Me
vt
L(e
t
), t > 0 . (2.24)
If t (, 0), then e
t
(0, 1) and since F
2,1
(t) is bounded on (, 0),
we obtain that H
2,1
and G
2,1
are bounded on (, 0). Namely, both func-
tions are equal to the convolution of F
2,1
with suitable functions with com-
pact supports.
From
f
2+
(e
t
) =
d
2l
dx
2l
H
2,1
(t) +G
2,1
(t), t (, ),
we obtain
f
2+
(x) =
p=1
a
p
x
p
d
p
dx
p
(H
2,1
(log x)) +G
2,1
(log x), x > 0,
where a
p
are suitable constants.
Set now
G
2,1
(x) =
_
G
2,1
(log x), x > 0
0, x 0,
H
2,1
(x) =
_
H
2,1
(log x), x > 0
0, x 0.
The distributions f
2+
T
(, ) and
2l
p=1
a
p
x
p d
p
dx
p
H
2,1
+

G
2,1
are equal
to each other. For every T(R), we have
_
0
f
2+
(kx)(x)dx = f
2+
(kx), (x)
=
2l
p=1
a
p
(kx)
p
1
k
p
(

H
2,1
(kx))
(p)
, (x) +
G
2,1
(kx), (x)
111
=
2l
p=1
a
p
(1)
p
H
2,1
(kx), (x
p
(x))
(p)
+
G
2,1
(kx), (x) .
The functions H
2,1
and G
2,1
are bounded on (, 0). Thus, by (2.24),
we have that for some M > 0
sup[
G
2,1
(kx)[, [

H
2,1
(kx)[ M(kx)
v
L(kx), x >
1
k
, (2.25)
sup[
G
2,1
(kx)[, [

H
2,1
(kx)[ M, 0 < x
1
k
, (2.26)
Since v > 1, these inequalities imply that for any T(R),
_
0
f
2+
(kx)(x)dx/(k
v
L(k)) < . (2.27)
In a similar way as in the proof of Theorem 2.3, one can prove that
(2.25) holds for every o(R). If we put F
2,2
(t) = f
2
(e
t
), t > 0, then,
by the same arguments as above, one can prove that for every o(R),
0
_
f
2
(kx)(x)dx/(k
v
L(k)) < . (2.28)
By the BanachSteinhaus Theorem, it follows from (2.23), (2.27) and
(2.28) that
lim
k
f(kx)/k
v
L(k) = g(x)
in the sense of convergence in o
(R).
(iv) The proof is the same as (iii). Namely, in this case the estimates
(2.25) and (2.26) imply the claim as well.
We can extend (ii) in Theorem 2.10 to the case v / N. First, we shall
recall two assertions (see Lemma 1 and Lemma 2, 7.4 in [192]).
Lemma 2.1. Let f o
(R) and let be a regularly varying function of

the degree ,= 1, 2, . . .
a) If there exists the limit
f(kt)/(k) g ,= 0, k , in o
(R) ,
where f = f
+
+ f
(f
+
o
+
and f
), then there exists N N

0
such that for f
+,N
= t
N
f
+
and f
,N
= t
N
f
,
lim
k
f
,N
(kx)/(k
N
(k)) = C
f
+N+1
(t) ,
where (C
+
, C
) ,= (0, 0).
b) If f o
+
and if, for N N
0
, f
N
= t
N
f has quasi-asymptotics at
related to k
N
(k), then:
1
0
If > 1, then f has quasi-asymptotics at related to (k), k > 0.
2
0
If < 1, then there exist a
j
R, j = 0, 1, . . . , p, such that
g(t) = f(t) +
p
j=0
a
j
(j)
(t)
has the quasi-asymptotics related to (k).
Corollary of Theorem 2.10. Let f T
(R). Suppose the following

limit exists, in the sense of convergence in T
(R),
lim
k
f(kx)/(k
v
L(k)) = g(x) ,= 0,
where v R 1, 2, 3, . . . . Then this limit exists in the sense of
convergence in o
(R) as well.
Proof. Theorem 2.10. implies that f o
(R).
Let n N be such that v +n > 1. Clearly, for f
1
(x) = x
n
f(x), x R,
there holds
lim
k
f
1
(kx)/(k
v
L(k)), (x) = x
n
g(x), (x), T(R) .
There exist distributions f
+
(x) o
+
and f
(supp f

(, 0]) such that f = f
+
+f
. This decomposition of f implies the decom-

position of f
1
: f
1
(x) = x
n
f
+
(x) +x
n
f
(x), x R, where x
n
f
+
(x) o
+
and x
n
f
(x) o
.
From Lemma 2.1 a), it follows that for some m N,
(t
m+n
f
(t))(kx)/(k
m+n+v
L(k)) C
f
v+m+n+1
(x) in o
(R) as k ,
where (C
+
, C
) ,= (0, 0).
Now, Lemma 2.1 b) implies that for some constants a
, = 0, 1, . . . , p,
and b
, = 0, 1, . . . , r,
(f
+
(kt) +
p
=0
a
()
(kt))/(k
v
L(k)) C
1
f
v+1
(t) in o
(R) as k ,
(f
(kt) +
r
=0
b
()
(kt))/(k
v
L(k)) C
2
f
v+1
(t) in o
(R) as k ,
113
where (C
1
, C
2
) ,= (0, 0).
Let us notice that in Lemma 2.1 b) it is assumed that the limit distri-
bution g is dierent from 0. But this assertion also holds for g = 0.
Thus for suitable constants

C
, = 0, 1, . . . , s, s = maxp, r
(f(kt) +
s
=0
()
(kt))/(k
v
L(k)) C
1
f
v+1
(t) +C
2
f
v+1
(t),
as k in o
(R). The fact that v R N implies that

1
k
v
L(k)
s
=0
()
(kt) 0, k ,
and this completes the proof.
2.7 Quasi-asymptotics at
In 2.12.5 the essential assumption was that the generalized functions were
supported by an acute cone . In this section, we relax this restriction over
the support in the one-dimensional case. In fact, we already faced this sit-
uation in 2.6 and studied some problems occurring when the distributions
are not supported by an acute cone. We will focus in the case of distri-
bution spaces. Let us observe that some of the results below can be also
generalized to T
(R
n
).
Denition 2.4. It is said that f T
(R) has quasi-asymptotics at

related to some positive measurable function c(k), k (a, ), a > 0, if for
some g T
(R)
lim
k
f(kx)/c(k), (x) = g(x), (x), T(R).
We write in short: f(kx)
q
c(k)g(x), k in T
(R), or simply f
q
g at
related to c(k).
The results of 2.7 give us already some important properties of quasi-
asymptotics at , they are stated in the following remark.
Remark. If g ,= 0 and T
(R) = T
(R) in Denition 2.4, then, by

Theorem 2.10, f o
(R), c(x) = x
v
L(x), x (0, ), for some v R and
some slowly varying function L, and g is a homogeneous distribution with
degree of homogeneity v.
Several properties of the quasi-asymptotics at are listed in the
following theorems.
(R) and f
q
g at related to k
L(k). Then:
(i) f
(m)
q
g
(m)
at related to k
m
L(k), k > a, m N;
(ii) if m N then x
m
f
q
x
m
g at related to k
+m
L(k);
(iii) if c(R) and c
1
is a measurable positive function on some
interval (a, ), a > 0, such that
(kx)/c
1
(k)
0
(x) in c(R), k , x R,
then f
q
g
0
at related to c
1
(k)k
L(k).
Proof. Properties (i) and (ii) follow easily from the denition. For (iii),
observe that if T(R), then (x)(kx)/c
1
(k) (x)
0
(x) in T(R). By
the equivalence between weak and strong sequential convergence in T
, we
have
lim
k
_
f(kx)(kx)
c
1
(k)k
L(k)
, (x)
_
= lim
k
_
f(kx)
k
L(k)
, (x)
(kx)
c
1
(k)
_
= g
0
, .
Theorem 2.12. Let f c
(R) and f
q
g at related to k
L(k), g ,= 0.
Then L(k) = 1, k > a , N, and g(x) = C
(1)
(x), for some
constant C. Moreover, the limit in Denition 2.4 can be extended on o(R).
Proof. It is the same as that of Example 5 in 2.3.
Theorem 2.13. Let F be a locally integrable function on R and R, >
1, such that
lim
x
F(x)
[x[
L([x[)
= C
,
where L is slowly varying at . Then F
q
g at related to k
L(k),
where
g(x) = C
+
x
+
+C
.
Proof. Let us put F
+
(x) = H(x)F(x) and F
(x) = H(x)F(x), x R,
we can now apply Example 3 in 2.3 to each F
(see also Theorem 2.8).

115
We have the following structural theorem, it will be extended in 2.10.
(R) and f
q
g at related to k
L(k), where
g ,= 0 and R(N). There are m N
0
and a locally integrable function
F such that
f = F
(m)
and lim
x
F(x)
[x[
+m
L([x[)
= C
,
where (C
+
, C
) ,= (0, 0).
Proof. Since f o
(R) (Theorem 2.10), let f = f

+
+f
, where f
+
o
+
and f
(supp f
(, 0]). The Corollary of Theorem 2.10 implies

that for every o(R),
_
f(kx)
k
L(k)
, (x)
_
g(x), (x) as k .
As in the proof of the Corollary of Theorem 2.10, one may choose f
satisfying the additional requirement

_
f
(kx)
k
L(k)
, (x)
_

f
+1
(x), (x) as k .
By Theorem 2.2, there exist locally integrable functions F
1
and F
2
with
supp F
1
[0, ), supp F
2
(, 0], and m N
0
such that
f
+
(x) = F
(m)
1
(x), f
(x) = F
(m)
2
(x), x R,
and
lim
x
F
1
(x)
x
+m
L(x)
= C
+
, lim
x
F
2
(x)
[x[
+m
L([x[)
= C
.
This completes the proof.
The proof of the next result can be found in [111].
Theorem 2.15. Let f o
(R) and
0
T(R) such that
_

0
(t)dt = 1.
Let f
q
g at related to k
v
L(k), v R, g ,= 0, and
_
f(kx)
k
v+1
L(k)
,
0
(x)
_
g
0
(x),
0
(x) ,
where g
0
o
(R) and g
0
= g. Then f
q
g
0
at related to k
v+1
L(k).
Recall that the Fourier transform of a tempered distribution f is denoted
by T(f) or

f. We shall analyze in 2.8 the quasi-asymptotics of a distribution
at zero (cf. Denition 2.5). In the following theorem, which is very useful,
we already compare these two notions via the Fourier transform.
(R) and v R (N). If

f
q
g at related to k
v
L(k) , (2.29)
with g ,= 0, then
lim
k
f(x/k)
(1/k)
v1
L
1
(1/k)
= g(x) in o
(R) , (2.30)
where L
1
() = L(1/) is slowly varying at the origin.
Conversely, if f o
(R) and (2.30) holds with v R, then (2.29) holds,

as well.
Proof. Let o(R). By the Corollary of Theorem 2.10, f o
(R) and
_
f(kx)
k
v
L(k)
,
(x)
_
=
_

f(x/k)
(1/k)
v1
L
1
(1/k)
, (x)
_
, k > 0 .
This implies the assertion.
Theorem 2.17. Let T c
(R) and T
q
g
1
at related to k
v
, v N,
g
1
,= 0. Let f T
(R) and f
q
g at related to k
L(k), R (N),
g ,= 0. Then T f
q
g
1
g at related to k
+v+1
L(k).
Proof. Let o(R). Using the properties of the Fourier transform, we
have
_
(T f)(kx)
k
+v+1
L(k)
,
(x)
_
=
_

T(x/k)

f(x/k)
k
+v+2
L(k)
, (x)
_
=
_

f(x/k)
(1/k)
1
L(1/k)
,
T(x/k)
(1/k)
v1
(x)
_
. (2.31)
Since

T is an entire function of polynomial growth when [x[ , it
must be of the form

T(x) = x
v1
T
1
(x), x R, where T
1
is an entire
function of polynomial growth such that T
1
(0) = C ,= 0. All the derivatives
of

T are of polynomial growth when [x[ . So, the same holds for T
1
.
This implies that for any o(R)
1
k
v+1
(x/k)
v1
T
1
(x/k)(x) = x
v1
T
1
(x/k)(x) x
v1
T
1
(0)(x) ,
117
k , in the sense of convergence in o(R). Let us note that g
1
(x) =
x
v1
T
1
(0), x R.
In the spaces o(R) and o
(R) the strong and weak sequential conver-

gence are equivalent. This implies that
_

f(x/k)
(1/k)
1
L
1
(1/k)
,
T(x/k)
(1/k)
v1
(x)
_
g(x), g
1
(x)(x) = (g
1
g)(x),

(x) k .
By (2.31), the claim follows from Theorem 2.16.
(R) and f(kx)/k
L(k); k > a, R
(N), be a bounded subset of T
(R). Let T c
(R) and T
q
g
1
at
related to k
1
, g
1
,= 0. If T f
q
g
2
at related to k
L(k), g
2
,= 0, then
f
q
g at related to k
L(k) and g
1
g = g
2
. (Note, g
1
= C).
Proof. The same arguments, as in the proof of Theorem 2.10 yield that
f o
(R) and that

f(k)/(k
L(k)); k > a
is a bounded subset of o(R). With the same arguments as above, we have
( o(R))
_

f(x/k)
(1/k)
1
L
1
(1/k)
, (x)
_
1
T(x/k)
T(0)
__
0 as k .
This implies the assertion.
2.8 Quasi-asymptotics at the origin
Denition 2.5. Let f T
(R) and c(x), x (0, a), a > 0, be a positive

measurable function. It is said that f has quasi-asymptotics at 0 in T
(R)
related to c(1/k) if there is g T
(R) such that

lim
k
_
f(x/k)
c(1/k)
, (x)
_
= g(x), (x), T(R).
We write in short: f
q
g at 0 related to c(1/k) in T
(R).
Remark. The limit in Denition 2.5 may be formulated as
lim
+0
_
f(x)
c()
, (x)
_
= g(x), (x), T(R).
Therefore, we often denote also quasi-asymptotics at 0 by
f(x)
q
c()g(x), 0
+
in T
(R),
or simply by f
q
g at 0 related to c().
If c = 1 then this Denition 2.5 is a slight generalization of the well-
known Lojasiewicz denition of a value at 0 of a distribution (see [93]
and [105]), and it leads to a notion of jump behavior for distributions [173],
[176], [178], [181] and [57]. Note that in the Lojasiewicz denition 0
from both sides.
We list some properties of the quasi-asymptotics at 0 for distributions,
we omit the proofs if they are similar to those of the corresponding prop-
erties of the quasi-asymptotics at innity. For details see [112].
The next proposition shows that we may always assume c is regularly
varying at the origin (cf. 0.3). The proof is essentially the same as that of
(i) in Theorem 2.10.
Proposition 2.7. Let f T
(R) and c satisfy the conditions of Denition

2.5. Assume further that g ,= 0. Then, for some real number v and some
slowly varying function L at 0
+
c(x) = x
v
L(x), x (0, a) .
Moreover, g is homogeneous with the degree of homogeneity v.
Some obvious properties of the quasi-asymptotics at 0 in o
(R) are given

in the following proposition.
Proposition 2.8. Let f o
(R) and f
q
g at 0 related to
v
L() in o
(R).
Then:
(i) f
(m)
q
g
(m)
at 0 related to
vm
L() in o
(R), m N;
(ii) x
m
f(x)
q
x
m
g(x) at 0 related to
v+m
L() in o
(R), m N.
The same assertions hold for the quasi-asymptotics at 0 in T
(R).
119
We have seen (see remark after Proposition 2.4) that the quasi-
asymptotics at innity is not in general a local property. The next propo-
sition asserts that the quasi-asymptotics at 0 is a local property.
(R) and f
q
g at 0 related to
v
L() in
T
(R), and let f

1
T
(R) be such that f = f

1
in some neighborhood of
zero. Then f
1
q
g at 0 related to
v
L(), as well.
Proof. The assertion follows from f(x), (x/) = f
1
(x), (x/)
which holds for any T(R) if <
0
().
Remark. The same assertion holds for the quasi-asymptotics at 0 in
o
(R). This was proved in [34], Lemma 1.6. This claim also follows directly
if we combine Proposition 2.9 with Theorem 2.35 from 2.11.1.
Proposition 2.10. (Theorem 3, Chapter I, 3.3 in [192]) Let f o
(R) be
a locally integrable function in (a, a), a > 0. Let c() =
L(), > 1,
where as usual L is slowly varying at 0
+
. If
lim
x0
f(x)/c([x[) = C
,
then f has the quasi-asymptotics at zero in o
(R) related to c and

lim
0
+
f(x)/c() = C
+
x
+
+C
in o
(R) .
Proof. According to the remark after Proposition 2.9, it is sucient to
prove that for any o(R) the limits of the following expressions exist,
when 0
+
,
a
_
0
f(x)/c()(x)dx and
0
_
a
f(x)/c()(x)dx.
Let us consider the rst one:
lim
k
a
_
0
f(x)/c()(x)dx = lim
k
a
_
0
c(x)
c()
f(x)
c(x)
(x)dx.
But since the term in the second integral is dominated by an L
1
function
(c(x)/c() = O(
), where > 1, [148]), it is possible to pass the

limit under the integral. The second expression can be considered in the
same way.
Proposition 2.10 provides a relation between the asymptotic behavior
at zero of a locally integrable function and the quasi-asymptotics at zero of
the distribution dened by it.
The next Proposition is a direct consequence of Theorem 2.15 and
Theorem 2.16.
(R) such that xf

q
g at 0 related to
v+1
L(), v R (N) in o
(R). Let
0
T(R) such that
_
R
0
(x)dx = 1
and
_
f(x)
v
L()
,
0
(x)
_
g
0
(x),
0
(x) as 0
+
such that g
0
o
(R) and xg
0
(x) = g(x), x R. Then, f
q
g
0
at 0
related to
v
L() in o
(R) (g and g
0
are homogeneous of order v + 1 and
v, respectively).
Proposition 2.10 and (i) of Proposition 2.8 directly yield
(R) and f = F
(m)
in some neighborhood of
0, where m N
0
and F is a locally integrable function such that for some
v > 1 and some slowly varying function L,
lim
x0
F(x)
[x[
v
L([x[)
= C
.
Then f
q
g at 0 related to
v
L() in o
(R), where g = (C
+
x
v
+
+C
x
v
)
(m)
.
We now discuss an structural theorem. A more complete result will be
the subject of 2.10.32.10.5.
(R) have quasi-asymptotics at 0 in T
(R) re-
lated to
v
L(). If v > 0 or if v > 1 but L is bounded on some interval
(0, a), a > 0, then there exist a continuous function F, dened on (1, 1),
an integer m, and constants C
+
, C
, such that
f = F
(m)
and lim
x0
F(x)
[x[
v+m
L([x[)
= C
. (2.32)
Proof. The proof of this theorem is similar to the proof of the
Lojasiewicz structural theorem for a distribution having a value at 0, given
121
in [102], pp. 4952. However, it is necessary to make several non-trivial
renements in the quoted argument.
Let I = (2, 2). Since f(x)/(
v
L()) g(x) in T
(R), 0
+
, there
is a family of continuous functions F
, (0,
0
], dened on I, an m R
0
such that
F
(m)
(x) =
f(x)
v
L()
, x I, (0,
0
]
and F
(x) g
1
(x) uniformly on I when 0
+
, where
g
1
(x) = (C
+
x
v+m
+
+C
x
v+m
)/m! .
With no loss of generality, we assume that
0
1. Let us put
(x) = F
(x)
v
L(), x I, 0 < 1 .
From

F
(m)
1
= f it follows (
m
F
1
(x))
(m)
= f(x), x I, (0, 1].
So, because

F
(m)
(x) = f(x), we obtain

(

F
1
(x)
m

F
(x))
(m)
= 0, x I, (0, 1] .
This implies that there is a polynomial which depends on such that
(x) =
m
(

F
1
(x) +b
0
() +b
1
()x + +b
m1
()(x)
m1
) ,
x I, (0, 1]. We have
(x)
v
L()
g
1
(x) =
1
m+v
L()
[

F
1
(x) g
1
(x)L([x[)
+ b
0
() + +b
m1
()(x)
m1
]
+ g
1
(x)
_
L([x[)
L()
1
_
, x I, (0, 1].
We obtain that for any x I 0
1
m+v
L()
(G(x) +b
0
()(x) + +b
m1
()(x)
m1
) 0
as 0
+
, where we put G(x) =

F
1
(x) g
1
(x)L([x[).
Note that the last limit is not uniform in general because we have only
the following property of a slowly varying function:
L([x[)
L()
1, 0
+
, uniformly for [x[ [a, b], b > a > 0 .
Let us x m points x
1
, . . . , x
m
I such that x
i
,= 0, i = 1, . . . , m, and let
d =
1
2
min[x
i
[; i = 1, . . . , m, J = x; [x[> d I .
Because of the quoted property of L, we have
1
m+v
L()
(G(x) +b
0
() +b
1
()x + +b
m1
()(x)
m1
) 0, 0
+
,
uniformly on J. This implies that for some monotonously increasing func-
tion (), > 0, () 0 as 0
+
, there holds
[G(x) +b
0
() +b
1
()x + +b
m1
()
m1
x
m1
[<
m+v
L()(),
x J, (0, 1] . (2.33)
Let > 0 and < 2. If we put in (2.33) instead of and x
i
/
instead of x, i = 1, . . . , m), (note x
i
/ J, i = 1, . . . , m), we obtain
[G(x
i
) +b
0
() +b
1
()x
i
+ +
m1
b
m1
()x
m1
i
[<
m+v
L()() .
From (2.33) and the last inequality it follows
[b
0
() b
0
() +(b
1
() b
1
())x
i
+ +
m1
(b
m1
() b
m1
())x
m1
i
[< 2
m+v
L()().
Now, in the same way as in [102], p. 51 one can prove
[b
i
() b
i
()[< 2
i+1
K()L()
m+vi
,
< 2, i = 0, . . . , m1 , (2.34)
where K is a suitable constant. Let < < 1/2. Take r N
0
such that
/2
r+1
< /2
r
. (2.34) implies
[b
i
(/2
j1
) b
i
(/2
j
)[< 2
i+1
K(/2
j1
)L(/2
j1
)(/2
j1
)
m+vi
,
(2.35)
i = 0, . . . , m1; j = 1, . . . , r, and
[b
i
(/2
r
) b
i
()[< 2
i+1
K(/2
r
)L(/2
r
)(/2
r
)
m+vi
, (2.36)
i = 0, . . . , m1.
Let v > 0 and C = suptL(t); t (0, ). (2.35) and (2.36) imply
(0 < < < 1/2)
[b
i
() b
i
()[ 2
i+1
K()C
_
r+1
j=1
(/2
j1
)
_
m+vi1
2
i+v+m+2
K()C
m+vi1
, i = 0, 1, . . . , m1 . (2.37)
123
Note that the assumption v > 0 is essential in the above inequality.
From (2.37) it follows b
i
= lim
0
+
b
i
() < , i = 0, 1, . . . , m1, and
[b
i
b
i
()[< K()
m+vi1
, i = 0, . . . , m1,
where K is a suitable constant.
From (2.33) and the last inequality it follows
[G(x) +b
0
+b
1
x + +
m1
b
m1
x
m1
[< K
1
m+v
L()() , (2.38)
x J, 0 < 1 (with suitable K
1
).
Let us show that the function
F(x) = G(x) +b
0
+b
1
x + +b
m1
x
m1
+g
1
(x)L([x[), x (2, 2) ,
satises the conditions of the Theorem. Clearly, F
(m)
= f. Put in (2.38)
x = 1. We obtain
[F() g()L()[< K
1
m+
L()(), 0 < 1 .
Let v > 1 and L(x) < C, x (0, a). Then from (2.35) and (2.36) it
follows (for 0 < < , < a, < 1/2)
[b
i
() b
i
()[ 2
i+1
KC()
r+1
j=1
(/2
j1
)
m+vi
2
i+1
KC()
m+vi
r+1
j=1
(1/2
j1
)
m+vi
, i = 0, . . . , m1.
Now, the proof follows as in the previous case. The proof is complete.
The assertion of Theorem 2.19 also holds for v < 0. This situation is
analyzed in the following theorem. The proof of it has been given in [116].
We omit the proof because more general results will be given in 2.10.3.
(R) and let f have the quasi-asymptotic behav-

ior at 0 in o
(R) relate to
v
L(), where v (, 0), v ,= 1, 2, . . . and
L is bounded on some interval (0, a), a > 0. Then there exist a continuous
function F dened on (1, 1), an integer m and (C
+
, C
) ,= (0, 0) such
that (2.32) holds.
Remarks. 1. If v = 0 and L 1, then Theorem 2.19 generalizes the
well-known Lojasiewicz structural theorem ([93]) for a distribution which
has a value at 0.
2. Let f o
(R) and f
q
g at 0 related to c() in T
(R). The question

is: Does the same hold in o
(R)? We shall prove in this section that for

c() =
v
L(), < a, the answer to the question is armative if v > 0 or
if 0 v > 1 and L is bounded in some interval (0, ), > 0. Otherwise,
this question remained open for quite long time. A complete armative
answer has been recently obtained in [186], it will be presented below in
2.11 (Theorem 2.35).
3. Theorems 2.19 and 2.20 describes the structure of quasi-asymptotics
only under restrictions over and L. The complete structural theorems
will be discussed in 2.10.
Proposition 2.12 and Theorem 2.19 directly imply
(R) satisfy conditions of Theorem 2.19. Then

f
q
g at 0 related to
v
L() in o
(R).
Let us denote by : the space of Fourier transforms of elements from
T(R) supplied by the convergence structure transferred from T(R)(: =
T(T)). Let f o
(R). As usual, we write f

q
g at 0 related to
v
L() in
:
if g :
, and
lim
0
+
_
f(x)
v
L()
, (x)
_
= g(x), (x), : .
Using the Fourier transform and Theorem 2.10 (ii) one can easily obtain
that g o
(R).
For v < 0, we have the ensuing related result. above ([112]).
(R) and f
q
g at 0 related to
v
L() in :
,
where v < 0, v , N, and g ,= 0. Then f
q
g at 0 related to
v
L() in
o
(R).
Proof. Apply Fourier transform, (iii) in Theorem 2.10, and then
Theorem 2.16.
125
2.9 Quasi-asymptotic expansions
Chapter III, 10 in [192] and 4 in [32] were devoted to the quasi-asymptotic
expansion of tempered distributions with support in [0, ). Two kind of
expansions were dened therein. We will recall these denitions but in T
+
.
We also survey other denitions appearing in the literature.
Let f
; R be the family of tempered distributions belonging to

o
+
which is dened in 0.4.
We denote by
(by
0
) the set of all positive slowly varying functions
at (at 0
+
).
Let R and L
(L
0
). We introduce another family of
tempered distributions:
(f
L
)
+1
(t) =
_
H(t)L(t)t
/( + 1), > 1 ,
D
n
(f
L
)
+n+1
(t), 1, +n > 1 ,
(2.39)
where n is the smallest natural number such that +n > 1.
Obviously, (f
L
)
+1
q
f
+1
at (at 0
+
) related to k
L(k) (to
(1/k)
L(1/k)).
Denition 2.6. ([192]) A distribution g o
+
is said to have a closed
quasi-asymptotic expansion of order and of length , 0 if there
exist N N,
j
R and c
j
C, j = 1, . . . , N, such that
lim
k
1
k
_
g(kt)
N
j=1
c
j
f
j
+1
(kt)
_
= 0, in o
+
.
Denition 2.7. ([192]) A distribution g o
+
is said to have an open
quasi-asymptotic expansion of and of length , 0 < if for every
1
< , g has closed quasi-asymptotic expansion of order and of length
1
.
In [101, 117, 122] these two denitions were slightly altered and ex-
tended by using the family (f
L
)
; R instead of the family f
;
R (cf. (2.39)). In the next denition we assume that (f
L
)
; R
T
+
.
Denition 2.8. We say that an f T
+
has a closed quasi-asymptotic
expansion at (at 0
+
) of order (, L) R
(of order (, L)
R
0
) and of length , 0 < , related to k
L
0
(k) (related to
(1/k)
+
L
0
(1/k)) if f has quasi-asymptotics at (at 0
+
) related to
k
L(k) ((1/k)
L(1/k)) and if there exist

i
R, L
i

(L
i

0
), and
c
i
C, i = 1, . . . , N, N N,
1

2

N
(
1

2
. . .
N
),
and such that f is of the form
f(t) =
N
i=1
c
i
(f
L
i
)
i+1
(t) +h(t) , (2.40)
where for every T(R)
lim
k
_
h(kt)
k
L
0
(k)
, (t)
_
= 0,
_
lim
k
_
h(t/k)
(1/k)
+
L
0
(1/k)
, (t)
_
= 0
_
.
We write in short:
f
q.e.
i=1
c
i
(f
L
i
)
i
+1
at (at 0
+
) of order (, L).
Obviously, we shall assume that c
i
,= 0 and that
N
(
N
+).
We shall always assume that in representation (2.40),
1
>
2
> >
N
(
1
<
2
< <
N
), because, (f
L
j
)
+1
+ (f
L
k
)
+1
(f
L
j
+L
k
)
+1
.
Observe (f
L
j
)
1
+1
and (f
L
k
)
2
+1
have the same quasi-asymptotics at
(0
+
) if and only if
1
=
2
and L
j
L
k
. So, we have:
+
satisfy the conditions of Denition 2.8
and assume that there are two representations of f, with the same length l,
f(t) =
N
i=1
c
i
(f
L
i
)
i
+1
+h(t), f(t) =
M
i=1
c
i
(f
L
i
)

i
+1
+

h(t)
for which all the assumptions given above hold. Then, M = N,
1
=

1
, . . . ,
N
=
N
, L
1

L
1
, . . . , L
N

L
N
.
Examples: All examples are in o
+
(R).
1. We have that
r=1
H(x 1)
r! x
r
, x R, converges uniformly to H(x
1)e
1/x
but
H(x 1)e
1/x
q.e.
H(x) + ((log x)
+
)
at of order (0, L 1)
127
related to k
1
log k and
H(x 1)e
1/x
q.e.
H(x) + ((log x)
+
)
+
_
1 +
r=2
1
r!
1
r 1
_
(x) at
of order (0, L 1) related to k
1
.
2. H(t 1)/t
3
q.e.
(
)/2 at of order (1, L 1) related to

k
3
log k. Moreover, let n > 2; then for j n 2
H(t 1)/t
n
q.e.
1
(n 1)

1
(n 2)1!
+ +
(1)
j1
(j1)
(n 1)(j 1)!
at of order (1, L 1) related to k
j
;
H(t 1)/t
n
q.e.
1
(n 1)
+
1
(n 2)1!
+ +
(1)
n2
(n 2)!

(n2)
at of order (1, L 1) related to k
n
log k.
Following [32], we dene the extended open quasi-asymptotic expansion.
Denition 2.9. An f T
+
has open quasi-asymptotic expansion at
(at 0
+
) of order (, L) R
((, L) R
0
) and of the
length s, 0 < s , if and only if for every , 0 < s, f has
closed quasi-asymptotic expansion of order (, L) and of length , related
to k
(k) ((1/k)
+
L
(1/k)).
By the same arguments as in Proposition 2.13 one can prove the follo-
wing proposition:
+
have open quasi-asymptotic expansion at
(at 0
+
) of order (, L) and of length s and let 0
1
<
2
< s. Suppose
that
f
q.e.
i=1
a
i
(f
L
i
)
i
+1
at ( at 0
+
)
related to k
1
L
1
(k)((1/k)
+
1
L
1
(1/k)), and
f
q.e.
i=1
b
i
(f
L
i
)
i
+1
at ( at 0
+
)
related to k
2
L
2
(k)((1/k)
+
2
L
2
(1/k)).
Then, M N and a
i
= b
i
,
i
=
i
, L
i

L
i
, i = 1, . . . , N.
Let us note that if f has the closed quasi-asymptotic expansion at
of order (, L) related to k
L(k), then for any s , f has the open

quasi-asymptotic expansion at of order (, L) and of length s. A similar
conclusion holds for the point 0
+
as well.
We will actually use the notation
f
q.e.
i=1
c
i
(f
L
i
)
i
+1
[k
i
L
i
(k) (
k
i
L
i
())
and instead of the open or closed quasi-asymptotic expansion of order
(, L) and length s, we will just say that f has the quasi-asymptotic
expansion at or with respect to the given scale.
Let us redene the notion of quasi-asymptotic expansion. We state
our denition at 0
+
, but one can do the same for the quasi-asymptotic
expansion at .
We denote by the set N or a nite set of the form 1, 2, . . . , N, N
N. In the second case, we shall sometimes use the symbol
N
. In the
following denition, for
i
R and L
i

0
, i , we assume that if i, j
, i < j, then
i

j
and if
i
=
j
, then L
j
()/L
i
() 0, 0
+
.
Denition 2.10. An f T
(R) has quasi-asymptotic expansion at 0

+
related to
k
L
k
(); k if there are complex numbers A
k
,= 0, k ,
so that for any m
w.lim
0
+
_
f
m
k=1
A
k
(f
L
k
)
k
+1
_
(x)
m
L
m
()
= 0 in T
(R) .
We write in short:
f
q.e.
k
A
k
(f
L
k
)
k
+1
[
k
L
k
(),
or simply f
q.e.
k
A
k
(f
L
k
)
k
+1
related to
k
L
k
(); k . One can
easily prove that in this case
f
q.e.
k
A
k
(f
L
k
)
k
+1
[
k
L
k
().
(R). If
f
q.e.
k
A
k
(f
L
k
)
k
+1
[
k
L
k
()
129
and
f
q.e.
A
k
(f
L
k
)

k
+1
[

k
L
k
() ,
then
k
=
k
, A
k
=

A
k
and L
k
()

L
k
(), 0
+
, k .
Proof. Since 1 , by the properties of the quasi-asymptotics at
0
+
,
1
=
1
, A
1
=

A
1
and L
1
()

L
1
(), 0
+
. If
A
1
(f
L
1
)
1
+1
+A
2
(f
L
2
)
2
+1
+R
2
= A
1
(f
L
1
)

1
+1
+

A
2
(f
L
2
)

2
+1
+

R
2
,
where
w.lim
0
+
R
2
(x)
2
L
2
()
= 0 and w.lim
0
+
R
2
(x)

2
L
2
()
= 0 in T
(R) .
Since
2
>
1
and
2
>
1
, the assumption
2
,=
2
gives a contradiction.
In the same way L
2
()

L
2
(), 0
+
and thus A
2
=

A
2
. The rest follows
by induction.
We give a structural proposition in o
+
.
Proposition 2.16. If f o
+
,
f
q.e.
k
A
k
f
k
+1
[
k
L
k
() ,
then for each m there is a p
m
N
0
and a continuous function F
m
with
supp F
m
[0, ) such that
f =
_
F
m
+
m
k=1
A
k
f
k
+p
m
+1
_
(p
m
)
and lim
x0
+
F
m
(x)
x
m
+p
m
L
m
(x)
= 0 .
Proof. Observe that a version of Theorem 2.2 holds at 0
+
. Now, let
s m. By Denition 2.10
_
f(x)
m
k=1
A
k
f
k
+1
(x)
m
L
m
()
, (x)
_
0, as 0
+
, o(R) ,
and the claim follows.
Other denitions and results of the quasi-asymptotic expansions of dis-
tributions are given in [52], [53], [55], [102], [118], [194], and [205].
We end this section with an important example of quasi-asymptotic
expansion at .
Example 3. The quasi-asymptotic expansion at
f(kx)
q
j=0
(1)
j
j
j! k
j+1

(j)
(x) ,
is called the Estrada-Kanwal moment asymptotic expansion.
Estrada and Kanwal have extensively studied this expansion in several
distribution spaces as well as its numerous applications. For example, it
holds if f has compact support, and actually not just in the space T
but
in c
. The constants
j
are the moments of f, namely, they are determined
by
j
=
f(x), x
j
_
. Observe that the moment asymptotic expansion gives
a clear explanation of why the quasi-asymptotics at innity is not a local
property. For theory and applications of this interesting and important
asymptotic expansion, we refer to [47], [52], [53], and [54].
2.10 The structure of quasi-asymptotics. Up-to-date
results in one dimension
In this section we present results related to quasi-asymptotics in T
(R) and
o
(R). We give complete answers to some questions started in the previous

sections (cf. 2.6, 2.7 and 2.8). We shall follow the exposition from [171],
[172], [173] and [186].
Our rst aim is to describe the structure of (one-dimensional) quasi-
asymptotics by means of complete structural theorems; this will be done in
2.10.3 and 2.10.5. The key tool for obtaining such results is the concepts
of asymptotically and associate asymptotically homogeneous functions; we
discuss these classes of functions in detail in 2.10.2 and 2.10.4.
We should employ a new notation for quasi-asymptotics at 0 and ,
which is more convenient for the purposes of this section. In order to
emphasize the role of the slowly varying function, we will use the following
notation for quasi-asymptotics,
f(x) = L()g(x) +o(
L()) in T
(R) , (2.41)
where the parameter is taken to either zero or innity. We call the
degree of the quasi-asymptotic behavior. Observe that g may be identically
zero, and all the results presented in this section are applicable to this
situation as well. Recall that if (2.41) is satised and T
= T
, then g is
automatically homogeneous of degree and it therefore has the form (2.3),
depending on whether R (N) or not.
131
Let g
j
T
(R) and let

j
be arbitrary measurable functions, j =
1, 2, . . . , n, we write
f(x) =
n
j=1
j
()g
j
(x) +o(
n
()) in T
(R) ,
if
f(x), (x) =
n
j=1
j
() g
j
(x), (x) +o(
n
()), T(R).
2.10.1 Remarks on slowly varying functions
In this section we collect some results about slowly varying functions to be
used in the future.
Let us assume that L is a slowly varying function at the origin (cf. 0.3).
Similar considerations are applicable for slowly varying functions at innity.
Our rst obvious observation is that only the behavior of L near 0 plays
a role in (2.41), and so we may impose to L any behavior we want in
intervals of the form [A, ). Moreover, if

L is any slowly varying function
which satises
lim
x0
+
L(x)
L(x)
= 1,
we may replace L by

L in any statement about quasi-asymptotics without
loosing generality.
One of the most basic (and most important) results in the theory of
slowly varying functions is the representation formula (see rst two pages
of [148]). Furthermore, the representation formula completely characterizes
all the slowly varying functions; L is slowly varying at the origin if and only
if there exist measurable functions u and w dened on some interval (0, B],
u being bounded and having a nite limit at 0 and w being continuous on
[0, B] with w(0) = 0, such that
L(x) = exp
_
u(x) +
_
B
x
w(t)
t
dt
_
, x (0, B].
This formula is important because it enables us to obtain some estimates
on L. Since we are looking for suitable modications of L, our rst remark
is that we can always assume that L is dened in the whole (0, ) and L is
everywhere positive. This is shown by extending u and w to (0, ) in any
way we want.
Given any xed > 0, then, by modifying u and w, we can assume,
when it is convenient, that B = 1, u is bounded on all over (0, ) and
[w(x)[ < , x (0, ). In particular, we obtain the estimate
M min
_
x
, x
_
<
L(x)
L()
< M max
_
x
, x
_
, x, (0, ) , (2.42)
for some positive constants M and

M. This result is known as Potters
estimate [9], p. 25, and will be of vital importance in our investigations of
the structural properties for quasi-asymptotics. Under the assumption of
the last estimate we can use Lebesgues dominated convergence theorem in
_

0
_
L(x)
L()
1
_
(x)dx,
for o(R), to deduce that
L(x)H(x) = L()H(x) +o (L()) , in o
(R) . (2.43)
The reader should keep in mind (2.42) and (2.43), since from now on they
will be implicitly used without any further reference, especially for dieren-
tiating asymptotic expressions in the future sections. We nally comment
a well-known fact [9], [148]: As soon as L(ax) L(x) holds for each a > 0,
it automatically holds uniformly for a in compact subsets of (0, ) .
2.10.2 Asymptotically homogeneous functions
We study some properties of asymptotically homogeneous functions which
will be applied later to the structural study of quasi-asymptotics. Let us
proceed to dene this class of functions.
Denition 2.11. A function b is said to be asymptotically homogeneous of
degree at the origin (resp. at innity) with respect to the slowly varying
function L, if it is measurable and dened in some interval (0, A) (resp. on
(A, )), A > 0, and for each a > 0,
b(ax) = a
b(x) +o(L(x)), x 0
+
(resp. x ) . (2.44)
Obviously, asymptotically homogeneous functions at the origin and at
innity are connected by the change of variables x x
1
; therefore, most
133
of the properties of the class of asymptotically homogeneous functions at
innity can be obtained from those of the corresponding class at the origin.
Observe that no uniformity with respect to a is assumed in Denition
2.11; however, the denition itself forces (2.44) to hold uniformly for a
on compact subsets. Indeed, we will show this fact by using a classical
argument of J. Korevaar, T. van Aardenne Ehrenfest and N. G. de Bruijn
[83], [9], [148] and [186].
Lemma 2.2. Let b be asymptotically homogeneous of degree with respect
to L. Then, the relation
b(ax) = a
b(x) +o(L(x)),
holds uniformly for a in compact subsets of (0, ).
Proof. We show the assertion at the origin, the case at innity can
be obtained by the change of variables x x
1
. So assume that b is
asymptotically homogeneous function of degree at the origin with respect
to L. We may assume that b is dened on (0, 1]. We rather work with the
functions c(x) = e
x
b (e
x
) and s(x) = L(e
x
), hence c and s are dened
in [0, ). By using a linear transformation between an arbitrary compact
subinterval of [0, ) and [0, 1], it is enough to show that
c(h +x) c(x) = o(e
x
s(x)), x , (2.45)
uniformly for h [0, 1]. Suppose that (2.45) is false. Then, there exist
0 < < 1, a sequence h
m
m=1
[0, 1]
N
and an increasing sequence of real
numbers x
m
m=1
, x
m
, m , such that
[c(h
m
+x
m
) c(x
m
)[ e
x
m
s(x
m
), m N. (2.46)
Dene, for n N,
A
n
=
_
h [0, 2]; [c (h +x
m
) c(x
m
)[ <

3
e
x
m
s (x
m
) , m n
_
,
B
n
=
_
h [0, 2]; [c(h+x
m
+h
m
)c(h
m
+x
m
)[<

3
e
x
m
s(x
m
+h
m
), m n
_
.
Note that
[0, 2] =
_
nN
A
n
=
_
nN
B
n
,
so we can select N such that (A
n
), (B
n
) >
3
2
(here () stands for
Lebesgue measure), for all n N. For each n N, put C
n
= h
n
+ B
n
.
Then, we have (C
n
) >
3
2
, n N, and C
n
, A
n
[0, 3]. It follows that
A
n
C
n
,= , n > N. For each n N, select u
n
A
n
C
n
. In particular,
we have u
n
h
n
B
n
, and hence,
[c (u
n
+x
n
) c (x
n
)[ <

3
e
x
n
s (x
n
) ,
[c (u
n
+x
n
) c (x
n
+ h
n
)[ <

3
e
x
n
s (x
n
+h
n
)
which implies that for all n N,
[c (x
n
+h
n
) c (x
n
)[ <

3
e
x
n
(s (x
n
) +s (x
n
+h
n
)) .
Using that s(x +h) s(x) = o(s(x)), x , uniformly for h on compact
subsets of (0, ), we have that for all n suciently large, s (x
n
+h
n
)
2s (x
n
), which implies that for n big enough
[c (x
n
+h
n
) c (x
n
)[ < e
x
n
s (x
n
) ,
in contradiction to (2.46), Therefore, (2.45) must hold uniformly for h
[0, 1].
Corollary 2.1. If b is asymptotically homogeneous at the origin (resp. at
innity) with respect to a slowly varying function, then b is locally bounded
in some interval of the form (0, B) (resp. (B, )).
Proof. It follows directly from Lemma 2.2.
We now obtain the behavior of asymptotically homogeneous functions
when the degree is not a negative integer.
Theorem 2.23. Suppose that b is asymptotically homogeneous at the origin
(resp. at innity) with respect to the slowly varying function L. Assume that
its degree is not a negative integer. Then
(i) If > 0 (resp. < 0 for the case at innity), then
b(x) = o(L(x)), x 0
+
(resp. x ) . (2.47)
(ii) If < 0 (resp. > 0), then there exists a constant such that
b(x) = x
+o (L(x)) , x 0
+
(resp. x ) . (2.48)
135
Proof. We show only the assertion at the origin, the case at innity
follows again from a change of variables.
Let us rst show i). Assume that > 0. Let 0 < be any arbitrary
number. We keep < 2
1. Let x
0
> 0 such that
b
_
x
2
_
2
b(x)
L(x) and [L(2x) L(x)[ L(x), 0 < x < x

0
,
(2.49)
and M = sup
_
[b(x)[ /L(x);
1
2
x
0
x x
0
_
< . Let x [x
0
/2, x
0
]. We
obtain from (2.49),
b(x/2
n
)
L(x/2
n
)
2
n
[b(x)[
L(x/2
n
)
+
n1
j=0
2
(n1j)
L(x/2
j
)
L(x/2
n
)
.
Thus, with t = x/2
n
, and t [x
0
/2
n+1
, x
0
/2
n
],
b(t)
L(t)
2
n
M
L(2
n
t)
L(t)
+
n1
j=0
2
j
L(2
j+1
t)
L(t)
.
By this and
L(2
j+1
t)/L(2
j
t) (1 +), j = 0, . . . , n 1 ,
we have that if t
_
2
(n+1)
x
0
, 2
n
x
0
, then
b(t)
L(t)
M
_
1 +
2
_
n
+(1 +)
j=0
_
1 +
2
_
j
= M
_
1 +
2
_
n
+(1 +)
2
1
.
Let us prove that for every > 0 there exists a positive such that
[b(t)/L(t)[< , t (0, ). First, we have to take , small enough, such that
(1 +)
2
1
<

2
and n
0
N such that
M
_
1 +
2
_
n
<

2
, n n
0
.
Then, it follows that [b(t)/L(t)[< , t (0, ), if we take = x
0
/2
n
0
. This
completes the rst part of the proof.
We now show (ii). Assume that < 0. We rather work with c(x) =
e
x
b(e
x
) and s(x) = L(e
x
). Then c satises
c(h +x) c(x) = o (e
x
s(x)) , x ,
uniformly for h [0, 1]. Given > 0, we can nd x
0
> 0 such that for all
x > x
0
and h [0, 1],
[c(x +h) c(x)[ e
x
s(x) and [s(h +x) s(x)[
_
e
2
1
_
s(x).
So we have that
[c(h +n +x) c(x)[ [c(h +n +x) c(n +x)[ +[c(n +x) c(x)[
e
(n+x)
s(n + x) +
n1
j=0
[c(j + 1 +x) c(j +x)[
e
x
n
j=0
e
j
s(j +x)
e
x
s(x)
1
1 e
2
,
where the last estimate follows from s(x + j) s(x)e
j/2
. Since s(x) =
o (e
x
) as x , it shows that there exists R such that
lim
x
c(x) = .
Moreover, the estimate shows that
c(x) = +o (e
x
s(x)) , x ,
thus, changing the variables back, we have obtained,
b(x) = x
+o (L(x)) , x 0
+
.
We remark that (2.47) and (2.48) trivially imply that b is asymptotically
homogeneous of degree with respect to L.
Asymptotically homogeneous functions of degree zero have a more com-
plex asymptotic behavior. For example if L 1, any asymptotically homo-
geneous function is the logarithm of a slowly varying function. Instead of
attempting to nd their behavior in the classical sense, we will study their
distributional behavior.
137
Lemma 2.3. Let b be asymptotically homogeneous of degree 0 at the origin
(respectively at innity) with respect to the slowly varying function L. If
< 0 (resp. > 0) then,
b(x) = o (x
) , x 0
+
(resp. x ).
In particular, b(x) (L(x))
1
is integrable near the origin (resp. locally inte-
grable near ).
Proof. We know that L(x) = o (x
). Then for each a > 0, b(ax) =

b(x) +o (x
) and this implies that x
b(x) is asymptotically homogeneous

of degree with respect to the constant function 1. From (i) of Theorem
2.23, it follows that b(x) = o (x
).
We now describe the behavior of asymptotically homogeneous functions
of degree zero at the origin. The next theorem will be very important in
the next section.
Theorem 2.24. Let b be asymptotically homogeneous of degree zero at the
origin with respect to the slowly varying function L. Suppose that b is
integrable on (0, A]. Then
b(x)(H(x) H(x A)) = b()H(x) +o(L()) as 0
+
in T
(R) ,
(2.50)
where H is the Heaviside function.
Proof. Let T(R). Find B such that supp [B, B], then there
exists
< 1 such that

b(x), (x) =
_ A
0
b(x)(x)dx =
_
B
0
b(x)(x)dx, <
. (2.51)
Replacing (x) by B(Bx) and
by B
, we may assume that B = 1.

Our aim is to show that for some
0
< 1,
b(x) b()
L()
, x (0, 1], <
0
,
is dominated by an integrable function in (0, 1] for the use of the Lebesgue
theorem. For this goal, we assume that L satises the following estimate,
L(x)
L()
Mx
1
2
, x (0, 1], (0,
) . (2.52)
By Lemma 2.2, there exists 0 <
0
<
such that
[b(x) b()[ < L(), x [1/2, 2], <
0
.
We keep <
0
and x
_
2
n1
, 2
n
. Then
[b(x) b()[ [b(2x) b((2x)/2)[ +[b(2x) b()[
L(2x) +[b(2x) b()[
n
i=1
L
_
2
i
x
_
+L()
i=1
(2
i
x)
1/2
L() +L(),
where the last inequality follows from (2.52). Then, if <
0
and x 1,
then
b(x) b()
L()
M
1
x
1
2
+ 1,
where M
1
= M(
2 + 1). Therefore we can apply Lebesgues dominated

convergence theorem to deduce (2.50).
We also have a similar result at innity, this fact is stated in the next
theorem. Since its a corollary of Theorem 2.27 , we omit its proof and refer
the reader to 2.10.4.
Theorem 2.25. Let b be asymptotically homogeneous of degree zero at in-
nity with respect to the slowly varying function L. Suppose that b is locally
integrable on [A, ). Then
b(x)H(x A) = b()H(x) +o(L()) as in o
(R).
2.10.3 Relation between asymptotically homogeneous functions
and quasi-asymptotics
We introduced asymptotically homogeneous functions in order to study
the structure of the quasi-asymptotics for Schwartz distributions. We now
derive structural theorems for quasi-asymptotics in some cases from the
fundamental properties of this class of functions (Theorems 2.23, 2.24 and
2.25).
The technique to be employed here is based on the analysis of the
parametric coecients resulting after performing several integrations of the
quasi-asymptotic behavior, these coecients are naturally connected with
the class of asymptotically homogeneous functions. The technique of in-
tegration of distributional asymptotic relations goes back to the classical
139
work of Lojasiewicz [93, 116] (cf. Theorem 2.19 in 2.8). Later on, the
properties of the parametric coecients were singled out and recognized
as asymptotically homogeneous functions in [45], [171], [172], [173], [176],
[186].
The next proposition provides the intrinsic link between quasi-asympto-
tics and asymptotically homogeneous functions.
(R) have quasi-asymptotic behavior in

T
(R)
f (x) = L()g(x) +o (
L()) as (resp. 0
+
) , (2.53)
where L is a slowly varying function and g is a homogeneous distribution
of degree R. Let n N. Suppose that g admits a primitive of order n,
that is, G
n
T
(R) and G
(n)
n
= g, which is homogeneous of degree n + .
Then, for any given F
n
, an n-primitive of f in T
(R), there exist functions

b
0
, . . . , b
n1
, continuous on (0, ), such that
F
n
(x) = L()G
n
(x)+
n1
j=0
+n
b
j
()
x
n1j
(n 1 j)!
+o
_
+n
L()
_
(2.54)
as (resp. 0
+
) in T
(R), where each b

j
is asymptotically
homogeneous of degree j 1.
Proof. Recall that any T(R) is of the form
= C
0
+
, where C
=
_

(t)dt, T(R) , (2.55)

and
0
T(R) is chosen so that
_
0
(t)dt = 1. The evaluations of
primitives F
1
of f and G
1
of g on are given by
F
1
, = C
F
1
,
0
f, and G
1
, = C
G
1
,
0
g, .
This implies
_
F
1
(x)
+1
L()
, (x)
_
= C
_
F
1
(x)
+1
L()
,
0
(x)
_
_
f(x)
L()
, (x)
_
, (2.56)
and
_
G
1
(x)
+1
L()
, (x)
_
= C
_
G
1
(x)
+1
L()
,
0
(x)
_
_
g(x)
L()
, (x)
_
. (2.57)
With c
0
() = (F
1
G
1
)(x),
0
(x), (0, ), from (2.53), it follows
F
1
(x) = L()G
1
(x) +c
0
() +o
_
+1
L()
_
in T
(R) . (2.58)
So relation (2.54) follows by induction from (2.58) and (2.53).
We shall now concentrate in showing the property of the b
j
s. We set
F
m
= F
(nm)
n
and G
m
= G
(nm)
n
, m 1, . . . , n. By dierentiating rela-
tion (2.54) (n m)-times, it follows that
F
m
(x) = L()G
m
(x) +
m1
j=0
+m
b
j
()
x
m1j
(m1 j)!
+o
_
+m
L()
_
(2.59)
in T
(R). Choose T(R) such that

_
(x)x
j
dx = 0 for j = 1, . . . , m
1, and
_
(x)dx = 1. Then evaluating (2.59) at , we have that

(a)
+m
b
m1
(a) +L(a) G
m
(ax), (x) +o
_
+m
L()
_
= F
m
(ax), (x) =
1
a
_
F
m
(x),
_
x
a
__
=
+m
b
m1
() +L() G
m
(ax), (x) +o
_
+m
L()
_
,
and so, with j = m1 0, . . . , n 1, for each a > 0,
b
j
(a) = a
j1
b
j
() + o (L()) .
With the aid of asymptotically homogeneous functions, we can now
obtain our rst structural theorem. Observe that for the case at we
recover Theorem 2.14, while at 0 we actually extend Theorem 2.19 and
Theorem 2.20 (cf. Remark 2 in 2.8).
(R) have quasi-asymptotic behavior at

(resp. at the origin) in T
(R),
f(x) = C
L()
(x)
( + 1)
+C
+
L()
(x)
+
( + 1)
+o (
L()) . (2.60)
If / 1, 2, . . . , then there exist a non-negative integer m > 1
and an m-primitive F of f such that F is continuous (resp. continuous
near 0) and
lim
x
( +m + 1)F(x)
x
m
[x[
L([x[)
= C
(resp. lim
x0
) . (2.61)
Conversely, if these conditions hold, then (by dierentiation) (2.60) follows.
141
Proof. On combining Proposition 2.17 and Theorem 2.23, one obtains
that for each n N and F
n
, an n-primitive of f, there exist constants
0
, . . . ,
n1
such that in the sense of convergence in T
(R),
F
n
(x) =
n1
j=0
j
(x)
j
j!
+C
(1)
n
L()(x)
+n
( +n + 1)
+C
+
L()(x)
+n
+
( +n + 1)
+o
_
+n
L()
_
. (2.62)
It follows from the convergence T
(R) that there is m N, suciently

large, such that any m-primitive of f is continuous and (2.62) holds (with
n = m) uniformly for x [1, 1]. Pick a specic m-primitive of f, say F
m
,
then from (2.62) there is a polynomial p of degree at most m1 such that
F
m
(x) = p(x) +C
L()
(1)
m
(x)
+m
( +m+ 1)
+C
+
L()
(x)
+m
+
( +m+ 1)
+o
_
+m
L()
_
,
uniformly for x [1, 1]. Then setting F = F
m
p, x = 1, 1 and
replacing by x, relation (2.61) follows at once. The converse follows by
dierentiation and the properties of regularly varying functions.
We now start to analyze quasi-asymptotics of negative integral de-
grees. In this section we will only study the quasi-asymptotics f(x)
q
L()(x). We postpone the general case for 2.10.5, after the introduc-
tion of associate asymptotically homogeneous function in 2.10.4.
(R) have quasi-asymptotic behavior at

(at the origin) in T
(R),
f(x) = L()(x) +o
_
1
L()
_
as (resp. 0
+
) . (2.63)
Then, there exist m N, a function b, being asymptotically homogeneous
of degree 0 with respect to L, and an (m+1)-primitive F of f such that F
is continuous (resp. continuous near 0) and
F(x) = L([x[)
x
m
2m!
sgnx +b ([x[)
x
m
m!
+o ([x[
m
L([x[)) . (2.64)
Conversely, if (2.64) holds, then (2.63) follows by dierentiation.
Proof. The existence of m, b, and F satisfying (2.64) follows from the
weak convergence of (2.63), Proposition 2.17 and Theorem 2.23, as in the
proof of Theorem 2.26. The converse is shown by applying Theorem 2.25
(resp. Theorem 2.24) and dierentiating (m+ 1)-times.
2.10.4 Associate asymptotically homogeneous functions
We now introduce the main tool for the study of structural properties of
quasi-asymptotics of negative integral degree. Associate asymptotically
homogeneous functions are a generalization of asymptotically homogeneous
functions. Let us dene this class of functions.
Denition 2.12. A function b is said to be associate asymptotically ho-
mogeneous of degree 0 at the origin (resp. at innity) with respect to the
slowly varying function L, if it is measurable and dened in some interval
(A, ), A > 0, and there exists a constant such that for each a > 0,
b(ax) = b(x) +L(x) log a +o(L(x)) , x 0
+
(resp. x ) . (2.65)
Remark. Associate asymptotically homogeneous functions are also
known as de Haan functions (cf. [9] and [11]).
We may use the same argument employed in the proof of Lemma 2.2 to
show uniform convergence.
Lemma 2.4. Relation (2.65) holds uniformly in compact subsets of (0, ).
We shall study the distributional asymptotic properties of this class of
functions in detail. We rst roughly estimate the behavior of associate
asymptotically homogeneous functions of degree 0.
Lemma 2.5. Let b be associate asymptotically homogeneous of degree 0 at
the origin (resp. at innity) with respect to L, then for each < 0 (resp.
> 0),
b(x) = o(x
) , x 0
+
(resp. x ) . (2.66)
Hence, b is integrable near the origin (resp. locally integrable near innity).
Proof. We know that L(x) = o(x
), for each > 0 [148]. Hence b(ax) =

b(x) +o(x
) and thus x
b(x) is asymptotically homogeneous of degree

with respect to L 1, so (2.66) follows from Theorem 2.23.
The next two theorems will be crucial in 2.10.5. They generalize
Theorems 2.24 and 2.25. We only give the proof at innity, the proof
at the origin is similar to that of Theorem 2.24.
143
Theorem 2.27. Let b be a locally integrable associate asymptotically homo-
geneous function of degree zero at innity with respect to the slowly varying
function L. Suppose that b is dened on [A, ). Then
b(x)H(xA) = b()H(x)+L()H(x) log x+o(L()) in o
(R) , (2.67)
, where H is the Heaviside function.
Proof. Let
0
be any positive number. The function b can be de-
composed as b = b
1
+ b
2
, where b
1
L
1
(R) has compact support and
b
2
(x) = b(x)H(x
0
) is associate asymptotically homogeneous function
of degree zero at innity. Since b
1
satises the moment asymptotic expan-
sion (cf. Example 3 in 2.9), it follows that b
1
(x) = O(
1
) = o(L()) as
in o
(R). Therefore, we can always assume that A =

0
, where
0
is selected at our convenience.
Our aim is to show that there is some
0
> 1 such that
J(x, ) := (x)
b(x) b() L() log x
L()
H(x
0
)
is dominated by an integrable function, whenever o(R), for the use of
the Lebesgue dominated convergence theorem. For this goal, we can always
assume that L is positive everywhere and satises the following estimate
(cf. 2.10.1)
L(x)
L()
M max
_
x
1
4
, x
1
4
_
, x, (0, ) , (2.68)
for some positive constant M. Because of the uniformity of (2.65) on com-
pact sets, there exists a
0
> 1 such that
[b(x) b() L() log x[ < L(), x [1, 2],
0
< .
Let n be a positive integer. We keep
0
< and x
_
2
n
, 2
n+1
. Then
[b(x) b() L() log x[ [b(x) b()[ +[[ L() log x
[[L() log x +[b(2(x/2)) b(x/2) L(x/2) log 2[
+[[L(x/2) log 2 +[b(x/2) b()[
[[L() log x + (1 +[[ log 2) L(x/2) +[b(x/2) b()[
(1 +[[log 2)
n
j=1
L(2
j
x) +[[L() log 2x +L()
_
Mx
1
4
(1 +[[ log 2)
n
j=1
(1/2)
j
4
+[[ log 2x + 1
_
L() ,
where the last inequality follows from (2.68). So if
0
< and 1 < x, then
b(x) b() L() log x

L()
M
1
x
1
4
,
for some M
1
> 0. Now if
0
/ < x < 1, we have that
b(x) b() L() log x

L()
_
1 +
L(x)
L()
_
[ log x[ +
b() b(x) L(x) log x

1
L()
_
1 +Mx
1
4
_
[ log x[ +
L(x)
L()
b(x(x
1
)) b(x) L(x) log x
1
L(x)
_
1 +Mx
1
4
_
[ log x[ + MM
1
x
1
2
.
Therefore J(x, ) is dominated by an integrable function for >
0
, so we
apply Lebesgue dominated convergence theorem to deduce that
lim
_

0
J(x, )dx = 0 .
Finally,
b(x)H(x
0
), (x) b()
_

0
(x)dx L()
_

0
log x (x)dx
=
_

0
/
b(x)(x)dx b()
_

0
(x)dx L()
_

0
log x (x)dx
= L()
_

0
J(x, )dx +L()O
_
log
_
+O
_
b()
_
= o(L()) +L()O
_
b()
L()
_
= o(L()) , ,
where in the last equality we have used Lemma 2.5 and the fact that slowly
varying functions are o(
) for any > 0. This completes the proof of

(2.67).
Theorem 2.28. Let b be a locally integrable associate asymptotically ho-
mogeneous function of degree zero at the origin with respect to the slowly
varying function L. Suppose that b is dened on (0, A]. Then
b(x)(H(x) H(xA)) = b()H(x) +L()H(x) log x+o(L()) , (2.69)
as 0
+
in T
(R).
145
Corollary 2.2. Let b be an associate asymptotically homogeneous function
of degree 0 with respect to L. Then, there exists an associate asymptotically
homogeneous function c C
[0, ) such that b(x) = c(x) +o(L(x)).

Proof. We may assume that L C
[0, ) [148], Section 1.4. Find

B such that b is locally bounded in [B, ), this can be done because
of Lemma 2.4. Take T(R) such that
_
0
(t)dt = 1 and set
c(x) =
_
B/x
b(xt)(t)dt L(x)
_
0
(t) log tdt, the corollary now follows
from Theorem 2.27 (resp. Theorem 2.28).
We may also use Corollary 2.2 to obtain a representation formula for
associate asymptotically homogeneous functions, this is the analog to [148],
Theorem 1.2 for slowly varying functions.
Theorem 2.29. The function b is associate asymptotically homogeneous
of degree 0 at satisfying (2.65) if and only if there is a positive number
A such that
b(x) = (x) +
_
x
A
(t)
t
dt , x A, (2.70)
where is a locally bounded measurable function on [A, ) such that (x) =
M +o(L(x)) as x , for some number M, and is a (
-function such
that (x) L(x) as x .
Proof. Assume rst that b
1
is (
, dened on [0, ) and satises that

hypothesis of the theorem. We can nd L
1
L which is (
and satises
xL
1
(x) = o(L(x)) as x [148], p. 7. Let and c as in the proof of
Corollary 2.2 corresponding to b
1
and L
1
, additionally assume that supp
(0, ). From Theorem 2.27, we have that
b
1
(x) =
b
1
()
(x) +
L()
Pf
_
H(x)
x
_
+o
_
L()
_
as
in o
(R), where Pf(H(x)/x) = (H(x) log [x[)
, since distributional asymp-

totics can be dierentiated. Then, for x positive
xc
(x) = x
_

0
b
1
(xt)t(t)dt xL
1
(x)
_

0
(t) log t dt
= x
_

0
b
1
(xt)t(t)dt +o(L(x))
= b
1
(x) 0 +L(x)
_

0
(t)dt +o(L(x))
= L(x) +o(L(x)) as x .
Set (x) = xc
(x). Find A > 0 such that L is locally integrable on [A, ),

one has that b
1
(x) = c(A) +
_
x
A
((t)/t)dt +o(L(x)).
In the general case, let A be a number such that b and L are locally
bounded on [A, ) and let b
1
the function from Corollary 2.2 such that
b(x) = b
1
(x) + o(L(x)), then we can apply the previous argument to b
1
to
nd as before, so we obtain (2.70) with (x) = b(x)
_
x
A
((t)/t) dt =
c(A) +o(L(x)).
A change of variables x x
1
in Theorem 2.29 implies the analog
result at 0.
Theorem 2.30. The function b is associate asymptotically homogeneous
of degree 0 at the origin satisfying (2.65) if and only if there is a positive
number A such that
b(x) = (x) +
_
A
x
(t)
t
dt , x A, (2.71)
where is a locally bounded measurable function on [A, ) such that (x) =
M +o(L(x)) as x 0
+
, for some number M, and is a (
-function such
that (x) L(x) as x 0
+
.
A slightly dierent representation formula is given in [148] , but, except
for the smoothness of , both are equivalent.
2.10.5 Structural theorems for negative integral degrees. The
general case
This section is dedicated to the study of structural properties of quasi-
asymptotics with negative integral degree, solving the question posed in
Remark 3 of 2.8.
The next lemma reduces the analysis of negative integral degrees to the
case of degree 1.
Lemma 2.6. Let f T
(R) and k be a positive integer. Then f has the

quasi-asymptotic behavior in T
(R),
f(x) =
k
L()
(k1)
(x) +(1)
k1
(k 1)! L()(x)
k
+o
_
k
L()
_
if and only if there exists a (k 1)-primitive g of f satisfying
g(x) =
1
L() (x) +L()(x)
1
+ o
_
1
L()
_
in T
(R).
Proof. Apply Proposition 2.17.
147
We should introduce some notation that will be needed. In the following
for all n N we denote by l
n
the primitive of log [x[ with the property that
l
n
(0) = 0 and l
n
= l
n1
. We have an explicit formula for them:
l
n
(x) =
x
n
n!
log [x[
x
n
n!
n
j=1
1
j
, x R,
which can be easily veried by direct dierentiation. They satisfy
l
n
(ax) = a
n
l
n
(x) +
(ax)
n
n!
log a, a > 0 . (2.72)
(R) have quasi-asymptotics at (resp. at

the origin) of the form
f(x) =
1
L()(x)+
1
L()x
1
+o
_
1
L()
_
in T
(R) . (2.73)
Then, there exist an associate asymptotically homogeneous function b
satisfying
b(ax) = b(x) + log aL(x) +o(L(x), x (resp. x 0
+
) , (2.74)
an integer m, and a continuous (resp. continuous near 0) (m+1)-primitive
F of f such that
F(x) = b ([x[)
x
m
m!
+
x
m
2n!
L([x[) sgnx L([x[)
x
m
m!
m
j=1
1
j
+o ([x[
m
L([x[))
(2.75)
as x (resp. x 0), in the ordinary sense. Conversely, relation
(2.75) implies (2.73).
Proof. We will show the assertion only at innity, the proof at the origin
is exactly the same. We shall study, as we have been doing, the coecients
of the integration of (2.73). For each n N, choose an n primitive F
n
of
f satisfying F
n
= F
n1
. We now proceed to integrate (2.73) once, so we
obtain
F
1
(x) = b() +

2
L()sgnx +L() log [x[ +o(L()) in T
(R) . (2.76)
Now, using the standard trick of evaluating at T(R) with the property
_
(x)dx = 1, one obtains that

b(a) +

2
L(a)
_

sgnx (x)dx +L(a)

_

log[x[(x)dx +o(L())
= F
1
(ax), (x) =
1
a
_
F
1
(x),
_
x
a
__
= b() +

2
L()
_

sgnx(x)dx +L()
_

log[ax[(x)dx +o(L()) ,
, for each a > 0. So, we see that b satises (2.74) for each a > 0.
Further integration of (2.76) gives,
F
n+1
(x) = b()
(x)
n
n!
+
n
j=1
n
b
j
()
x
nj
(n j)!
+L()sgnx
(x)
n
2n!
+L()
n
l
n
(x) +o (
n
L()) as in T
(R) .
As in the proof of Proposition 2.17, one shows that the b
j
s are asymptot-
ically homogeneous functions of degree j with respect to L. Hence if we
apply Theorem 2.23 to the b
j
s, we obtain that
F
m+1
(x) = b()
(x)
m
m!
+L()
(x)
m
2m!
sgnx+L()
m
l
m
(x) +o (
m
L())
(2.77)
in the sense of convergence in T
(R). Moreover, it follows from the deni-

tion of convergence in T
(R) there exists m

0
N such that for all m m
0
the distribution F
m+1
is a continuous function and (2.77) holds uniformly
for x [1, 1]. Relation (2.75) is shown by making x = 1 in (2.77) and
then changing x.
Conversely, since only the behavior of b at innity plays a roll in (2.75),
we may assume that b is locally integrable, so the converse is obtained after
application of Theorem 2.27 and then (m + 1) dierentiations.
Theorem 2.31 is a structural theorem, but we shall give a version free
of b.
(R). Then f has quasi-asymptotics at

(resp. at the origin) of the form (2.73) if and only if there exists an (m+1)-
primitive F of f, continuous (resp. continuous near 0), such that for each
a > 0,
lim
x
m! (a
m
F(ax) (1)
m
F(x))
x
m
L(x)
= + log a
_
resp. lim
x0
+
_
. (2.78)
Proof. The limit (2.78) follows from (2.75), (2.74) and (2.72) by direct
computation. For the converse, rewrite (2.78) as
a
m
F(ax) (1)
m
F(x) = ( + log a)
x
m
m!
L(x) +o (x
m
L(x)) ,
for each a > 0. Set
b(x) = m! x
m
F(x)
_
_
2

m
j=1
1
j
_
_
L(x) , x > 0 .
149
By setting a = 1 in (2.78), one sees that for x < 0,
F(x) = b ([x[)
x
m
m!
+L([x[)
x
m
2m!
sgnxL([x[)
x
m
m!
m
j=1
1
j
+o ([x[
m
L([x[)) .
Since
a
m
F(ax) F(x) = log a
x
m
m!
L(x) +o (x
m
L(x)) ,
it is clear that for each a > 0,
b(ax) = b(x) + log aL(x) +o(L(x)) .
It is remarkable that, initially, no uniform condition on a is assumed
in (2.78). However, the proof of Theorem 2.32 forces this relation to hold
uniformly for a in compact subsets.
We are now ready to state the general structural theorem for negative
integral degrees which now follows directly from Lemma 2.6, Theorem 2.31
and Theorem 2.32.
(R) and let k be a positive integer. Then f

has the quasi-asymptotic behavior in T
(R) at (resp. at the origin),

f(x) =
k
L()
(k1)
(x) +(1)
k1
(k 1)! L()(x)
k
+o
_
k
L()
_
if and only if there exist m N, m k, an associate asymptotically
homogeneous function b of degree 0 at innity (resp. at the origin) with
respect to L satisfying
b(ax) = b(x) + log aL(x) +o(L(x)) , x (resp. x 0
+
),
for each a > 0, and an m-primitive F of f which is continuous (resp.
continuous near 0) and satises
F(x) = b ([x[)
x
mk
(mk)!
+L([x[)
x
mk
2(mk)!
sgnx
L([x[)
x
mk
(mk)!
mk
j=1
1
j
+o([x[
mk
L([x[))
as x (resp. x 0), in the ordinary sense. The last property is
equivalent to
lim
x
(mk)!
_
a
km
F(ax) (1)
mk
F(x)
_
x
mk
L(x)
= + log a
_
resp. lim
x0
+
_
,
(2.79)
for each a > 0.
It should be noticed that in (2.79) is not absolutely necessary to assume
that the limit is of the form + log a. Indeed, we have the following
corollary.
(R). Then f has quasi-asymptotics at innity

(resp. at the origin) of degree k, k N, if and only if there exists a
continuous (resp. continuous near 0) m-primitive F of f, m k, such that
the following limit exists for each a > 0,
lim
x
_
a
km
F(ax) (1)
mk
F(x)
_
x
mk
L(x)
= I(a)
_
resp. lim
x0
+
_
.
Proof. We show that I(a) must be of the form I(1) + log a, for some
constant . We easily see that I is measurable and satises
I(ab) = I(a) +I(b) I(1) ,
setting h(x) = e
I(x)I(1)
, one has that h is positive, measurable and satises
h(ab) = h(a)h(b), from where it follows [148] that h(x) = x
, for some ,
and so I has the desired form.
2.11 Quasi-asymptotic extension
We analyze some problems about the extensions of distributions to other
spaces together with their quasi-asymptotic properties, we name this prob-
lem quasi-asymptotic extension problem.
Let f T
have quasi-asymptotic behavior in T
, that is,
f(x), (x)
L() g(x), (x) , T . (2.80)

Suppose that f belongs also to another space |
such that T | (not

necessarily densely contained). For various spaces, we investigate in this
section the possibility of extending (2.80) to |
, in the sense of obtaining

the asymptotic behavior of f(x), (x) , for each |. Sometimes the
statement f |
is also part of the problem. The results of the present

section were obtained in [171], [172], [173], [186].
2.11.1 Quasi-asymptotics at the origin in T
(R) and o
(R)
In this subsection we conclude the discussion initiated in Remark 2 of 2.8
(cf. Theorem 2.21).
151
(R). If f has quasi-asymptotic behavior at 0 in

T
(R), then f has the same quasi-asymptotic behavior at 0 in in the space

o
(R).
Proof. Let be the degree of the quasi-asymptotic behavior. We shall
divide the proof into three cases:
/ 1, 2, 3, . . . ,
= 1,
= 2, 3, . . .
Suppose its degree is / 1, 2, 3, . . . and
f(x) = C
L()
(x)
( + 1)
+C
+
L()
(x)
+
( + 1)
+o (
L())
as 0
+
in T
(R). Then, by using Theorem 2.26 and the fact f o
(R),
we conclude the existence of an integer m, a real number such that
m > , > m +, and a continuous m-primitive F of f such that
F(x) =
[x[
m+
(m + + 1)
L([x[) ((1)
m
C
H(x) +C
+
H(x))
+ o( [x[
m+
L([x[) ), x 0
+
,
and
F(x) = O( [x[
), [x[ . (2.81)
We make the usual assumptions over L. Assume (cf. 2.10.1) that L is
positive, dened in (0, ) and there exists M
1
> 0 such that
L(x)
L()
M
1
max
_
x
1
2
, x
1
2
_
, , x (0, ) . (2.82)
Let o(R), then we can decompose =
1
+
2
+
3
, where supp
1

(, 1], supp
2
is compact and supp
3
[1, ). Observe that since
2
T(R) we have that
f(x),
2
(x) = (1)
m
C
L()
_
x
( + 1)
,
2
(x)
_
+ C
+
L()
_
x
+
( + 1)
,
2
(x)
_
+o (
L()) , 0
+
.
(2.83)
So, if we want to show (2.83) for , it is enough to show it for
3
placed
instead of
2
in the relation because by symmetry it would follow for
1
and hence for . Set
G(x) =
F(x)
x
+m
L(x)
, x > 0.
Then
lim
x0
+
G(x) =
C
+
( +m+ 1)
. (2.84)
On combining (2.81), (2.82) and (2.84), we nd a constant M
2
> 0 such
that
[G(x)[ < M
2
(1 +x
+
1
2
m
), x > 0 . (2.85)
Relation (2.85) together with (2.82) show that for 1,
G(x)
L(x)
L()
x
+m
(m)
3
(x)
2M
1
M
2
x
+1
(m)
3
(x)
H(x 1) .
The right hand side of the last estimate belongs to L
1
(R) and thus we can
use the Lebesgue dominated convergence theorem to obtain,
lim
0
+
1
L()
f(x),
3
(x)
= lim
0
+
(1)
m
_

0
G(x)
L(x)
L()
x
+m
(m)
3
(x)dx
= (1)
m
C
+
( +m + 1)
_

0
x
+m
(m)
3
(x)dx
= C
+
_
x
+
( + 1)
,
3
(x)
_
.
This shows the result in the case / 1, 2, 3, . . . .
We now consider the case = 1. Assume that
f(x) =
1
L()(x) +
1
L()x
1
+o
_
1
L()
_
as 0
+
in T
(R).
As in the last case, it suces to assume that o(R), supp [1, )
and show that
lim
0
+
L()
f(x), (x) =
_

1
(x)
x
dx.
We may proceed as in the previous case to apply the structural theorem,
but we rather reduce it to the previous situation. So, set g(x) = xf(x),
then
g(x) = L() +o(L()) as 0
+
in T
(R) . (2.86)
153
But g o
(R), then since the degree of the quasi-asymptotics is 0, the rst

case implies that (2.86) is valid in o
(R). Therefore
lim
0
+
L()
f(x), (x) = lim
0
+
1
L()
_
g(x),
(x)
x
_
=
_

1
(x)
x
dx.
This shows the case = 1.
It remains to show the theorem when 2, 3, . . . . Suppose the
degree is k, k 2, 3, . . . . It is easy to see that any primitive of degree
(k 1) of f has quasi-asymptotics of degree 1 at the origin with respect
to L (in fact this is the content of Proposition 2.17 when combined with
Theorem 2.23). The (k 1)-primitives of f are in o
(R), so we can apply

the case = 1 to them, and then, by dierentiation, it follows that f has
quasi-asymptotics at the origin in o
(R).
This completes the proof of Theorem 2.35.
2.11.2 Quasi-asymptotic extension problem in T
(0, )
The purpose of this subsection is to study the following problem. Suppose
that a distribution f T
(R) with support in [0, ) has quasi-asymptotics

of degree in the space T
(0, ), that is, for each T(0, )

f(x), (x)
L() g(x), (x) . (2.87)

What can we say about the quasi-asymptotic properties of f in T
(R)?
We can apply the techniques from 2.10.3 and 2.10.5 to give a complete
answer to this question. The answer depends on . We start with the
quasi-asymptotic behavior at innity, the same arguments are applicable
to quasi-asymptotics at the origin.
Let us start with the case > 1. It is not dicult to show that g
must be of the form Cx
+
/( + 1), for some constant C. Next, Proposi-
tion 2.17 still holds replacing the space T
(R) by T
(0, ) (actually this

holds without the restriction > 1). Hence, the same argument given
in Theorem 2.26 applies here, but this time we only require the uniform
convergence on [1/2, 2], and hence we can still conclude the existence of
the integer such that (2.61) holds with the limit taken only as x .
Actually, because > 1, relation (2.61) holds for any m-primitive of f.
Let F be the m-primitive of f supported on the interval [0, ), then we
have that
F(x)
Cx
+m
L(x)
( +m+ 1)
, x ,
so we have that F(x) = CL()(x)
+m
+
/( + m + 1) +o(
+m
L()) in
the space o
(R), dierentiating m-times, we obtain the following result.

(R) be supported on [0, ). If f has quasi-

asymptotic behavior of degree > 1 in T
(0, ), then it is a tempered

distribution and has the same quasi-asymptotic behavior in the space o
(R).
Suppose now that < 1 and is not a negative integer. This case
diers from the last one essentially in one point, we cannot conclude (2.61)
for every m-primitive of f but only for some of them. In any case, denoting
again by F the m-primitive (we keep m > 1) of f supported on [0, ),
we have that there exists a polynomial of degree at most m1 such that
F(x) p(x)
Cx
+m
L(x)
( +m+ 1)
, x ;
therefore,
F(x) =
CL()(x)
+m
+
( +m+ 1)
+
m1
j=0
a
j
(x)
j
+
+o(
+m
L()) as ,
in the space o
(R), for some constants a

0
, . . . , a
m1
. Thus, our arguments
immediately yield the next theorem.
(R) be supported on [0, ). Suppose that

f(x) = CL()
(x)
+
( + 1)
+o(
L()) as in T
(0, ).
If < 1 and is not a negative integer, then f is a tempered distribution.
Moreover, there exist constants a
0
, a
1
, . . . , a
n
(n < 1) such that
f(x) = CL()
(x)
+
( + 1)
+
n
j=0
a
j
(j)
(x)
j+1
+o(
L()) as in o
(R).
When = k, k being a positive integer, the distribution g in (2.87)
must have the form Cx
k
T
(0, ), for some constant C; these distri-

butions are homogeneous as elements of T
(0, ), but they do not have

homogeneous extensions to T(R). The behavior of f(x) as in
o
(R) is described in the next theorem.

We denote by Pf(H(x)/x
k
) the distribution
_
Pf
_
H(x)
x
k
_
, (x)
_
= F.p.
_

0
(x)
x
k
dx, T(R),
155
where F.p. stands for the Hadamard nite part of the divergent integral
[56].
(R) be supported on [0, ). Suppose that

f(x) = CL()
H(x)
(x)
k
+o
_
L()
k
_
as in T
(0, ) ,
where k is a positive integer. Then f is a tempered distribution and there
exist an associate asymptotically homogeneous function b satisfying
b(ax) = b(x) +
(1)
k1
(k 1)!
CL(x) log a +o(L(x)), x , (2.88)
for each a > 0, and constants a
0
, a
1
, . . . , a
k2
such that
f(x) = C
L()
k
Pf
_
H(x)
x
k
_
+
b()
k

(k1)
(x) +
k2
j=0
a
j
(j)
(x)
j+1
+o
_
L()
k
_
(2.89)
as in o
(R).
Proof. For each n N, let F
n
denote the n-primitive of f with support
in [0, ). Set C
1
= (1)
k1
C/(k 1)!. Adapting the arguments of 2.10.5
and reasoning as in the previous two cases, we obtain the existence of a
positive integer m > k such that F
m
is continuous and
F
m
(x) = b
1
(x)
x
mk
(mk)!
C
1
L(x)
x
mk
(mk)!
mk
j=1
1
j
+p
m1
(x)+o(x
mk
L(x)) ,
x , where b
1
is a locally integrable associate asymptotically homo-
geneous function satisfying (2.88) and p
m1
is a polynomial of degree at
most m 1. Throwing away the irrelevant terms of the polynomial p
m1
and using Theorem 2.27, we obtain the following asymptotic expansion as
in the space o
(R),
F
m
(x) = b
1
()
(x)
mk
+
(mk)!
+C
1
mk
L()l
mk
(x)H(x)
+
k2
j=0
a
j
(x)
mj1
+
(mj 1)!
+o(
mk
L()) .
Dierentiating (mk)-times this expansion, we have that
F
k
(x) = b
1
()H(x) +C
1
L()H(x) log x +
k2
j=0
a
j
(x)
kj1
+
(k j 1)!
+o(L()) .
(2.90)
The well known formulas [56], p. 68,
d
dx
(H(x) log x) = Pf
_
H(x)
x
_
and
d
dx
_
Pf
_
H(x)
x
n
__
= nPf
_
H(x)
x
n+1
_
+
(1)
n
(n)
(x)
n!
imply that
d
k1
dx
k1
_
Pf
_
H(x)
x
__
= (1)
k1
(k 1)! Pf
_
H(x)
x
k
_
(k1)
(x)
k1
j=1
1
j
.
Hence, dierentiating (2.90) k-times, one has (2.89) with
b(x) = b
1
(x) +
(1)
k
C
(k 1)!
_
_
k1
j=1
1
j
_
_
L(x).
The corresponding result at the origin is stated in the next theorem.
Theorem 2.39. Let f
0
T
(0, ). Let L be slowly varying at the origin

and R. Suppose that
f
0
(x) =
L()g
0
(x) +o (
L()) as 0
+
in T
(0, ) , (2.91)
g
0
T
(0, ). Then f
0
admits extensions to [0, ). Let f be any of such
extensions. Then f has the following asymptotic properties at the origin:
(i) If / N, then there exist constants C, a
0
, . . . , a
n
such that
f(x) =
n
j=0
a
j
(j)
(x) +C
L()x
+
+o(
L()) ,
as 0
+
in T
(R).
(ii) If = k, k N, then there are constants C, a
0
, . . . , a
n
and
an associate asymptotically homogeneous function of degree 0 with
respect to L satisfying
b(ax) = b(x) +
(1)
k1
(k 1)!
CL(x) log a +o(L(x)) , (2.92)
such that
f(x) =
n
j=0
a
j
(j)
(x) +b()
(k1)
(x) +C
k
L()Pf
_
H(x)
x
k
_
+o
_
k
L()
_
, (2.93)
as 0
+
in T
(R).
157
2.11.3 Quasi-asymptotics at innity and spaces 1
(R)
Sometimes is very useful to have the right of evaluating (2.41) in more test
functions than in o(R), this section is dedicated to give some conditions
under the test function which guarantee one can do this for the quasi-
asymptotic behavior at . We shall now improve Theorem 2.10 and its
corollary (cf. 2.6).
We need the following denition.
Denition 2.13. Let c(R) and R. We say that
(x) = O([x[
) strongly as [x[ , (2.94)

if for each m 0, 1, 2, . . .
(m)
(x) = O([x[
m
) as [x[ . (2.95)
The set of satisfying Denition 2.13 for a particular forms the space
1
(R) which we topologize in the obvious way [56]. These spaces and their
dual spaces are very important in the theory of asymptotic expansions of
distributions [56]. In fact, if we set 1(R) =
(R) (the union having

a topological meaning), we have that 1
(R) is the space of distributional

small distributions at innity [47], [56], they satisfy the moment asymptotic
expansion at innity (cf. Example 3 in 2.9). We point out that in [56] these
spaces are denoted by = 1
= /
and 1
= /
.
The next theorem shows that if f has quasi-asymptotics at , then
the distributional evaluation of f at 1
(R) makes sense under some

conditions on , specically, we show that f has extensions to some of the
spaces 1
(R).
(R) have quasi-asymptotic behavior of degree

at with respect to the slowly varying function L. If + < 1, then
f admits extensions to 1
(R).
Proof. Let > 0 such that + + < 1, then from Theorem 2.26,
Theorem 2.33 and Lemma 2.5 we deduce that there exist m N and a
continuous m-primitive of f, say F, such that
F(x) = O([x[
m++
) as [x[ . (2.96)
Notice that here we have used that L(x) = O(x
) as x [148]. So it is
evident that the extension of f to 1
(R) is given by
f(x), (x) = (1)
m
_

F(x)
(m)
(x)dx, 1
(R) , (2.97)
which in view of (2.95) and (2.96) is well-dened and denes an element of
1
(R).
We now show that the quasi-asymptotic behavior remains valid in 1
(R)
for one extension of f, with the assumption under imposed in Theorem
2.40.
(R) have quasi-asymptotic behavior at of

degree with respect to a slowly varying function L, then f admits an ex-
tension to 1
(R) which has the same quasi-asymptotics in 1
(R), provided
that + < 1.
Proof. The proof is similar to that of Theorem 2.35 with some modica-
tions in the estimates. We use one of the extensions of f found in Theorem
2.40; denote the extension by

f. We shall divide the proof into two cases:
/ 1, 2, 3, . . . and 1, 2, 3, . . . .
Suppose its degree is / 1, 2, 3, . . . and
f(x) = C
L()
(x)
( + 1)
+C
+
L()
(x)
+
( + 1)
+o (
L()) as ,
in T
(R). Find > 0 such that + + < 1. Then from Theorem 2.26,
there are an m such that m + > 0 and a continuous m-primitive F of f
such that
F(x) =
x
m
[x[
(m+ + 1)
L([x[) (C
H(x) +C
+
H(x)) +o
_
[x[
m+
L([x[)
_
,
x . We recall that H denotes the Heaviside function. We make the
usual assumptions over L (cf. 2.10.1), assume that L is positive, dened
and continuous in (0, ) and there exists M
1
> 0 such that
L(x)
L()
M
1
max
_
x
, x
_
, 1, x (0, ) . (2.98)
Let 1
(R). As in the proof of Theorem 2.35, we may assume that

supp [1, ), and the proof would be complete after we show
_
f(x), (x)
_
C
+
L()
_
x
+
( + 1)
, (x)
_
, (2.99)
159
as .
Set
G(x) =
F(x)
x
+m
L(x)
for x 1 , (2.100)
then
lim
x
G(x) =
C
+
( +m+ 1)
. (2.101)
So, we can nd a constant M
2
> 0 such that
[G(x)[ < M
2
, globally. (2.102)
Relation (2.102) together with (2.98) show that for 1,
G(x)
L(x)
L()
x
+m
(m)
(x)
M
1
M
2
x
+m+
(m)
(x)
H(x 1) .
Since 1
(R), the right hand side of the last estimate belongs to

L
1
(R) and thus we can use the Lebesgue dominated convergence theorem
to obtain,
lim
f(x), (x)
_
L()
= lim
(1)
m
_

0
G(x)
L(x)
L()
x
+m
(m)
(x)dx
= (1)
m
C
+
( +m + 1)
_

0
x
+m
(m)
(x)dx
= C
+
_
x
+
( + 1)
, (x)
_
.
This shows the result in the case / 1, 2, 3, . . . .
We now consider the case = k, k N. Assume that
f(x) =
k
L()
(k1)
(x) +
k
L()x
k
+o
_
k
L()
_
,
as in T
(R). As in the last case, it suces to assume that

1
(R), supp [1, ) and show that

lim
k
L()
_
f(x), (x)
_
=
_

1
(x)
x
k
dx.
So, set g(x) = x
k
f(x), then
g(x) = L() +o(L()) as in T
(R). (2.103)
But 1
(R) implies (x)/x

k
1
k
(R) then since the degree of the
quasi-asymptotic behavior of g is 0, last case implies that (2.103) is valid
in 1
k
(R) for a suitable extension g because k < 1, therefore
lim
k
L()
_
f(x), (x)
_
= lim
1
L()
_
g(x),
(x)
x
k
_
=
_

1
(x)
x
k
dx.
This completes the proof of Theorem 2.41.
The importance Theorem 2.41 lies in the fact that we can relax the
growth restrictions over the test functions, this permits to apply quasi-
asymptotics to obtain ordinary asymptotics in many interesting situations,
for example for certain integral transforms or for solutions to partial dier-
ential equations. We discuss a simple example.
Example. Let f T
(R) have quasi-asymptotic behavior at innity

of degree < 1,
f(x) =
L()g(x) +o (
L()) as in T
(R) .
Consider the Poisson kernel,
P(t) =
1
(t
2
+ 1)
.
Clearly P 1
2
(R). By Theorem 2.41, f has an extension

f such that the
evaluation of

f at P is well dened and

f preserves the quasi-asymptotic
properties of f. Thus
U(z) = U(x +yi) =
_
f(t),
1
y
P
_
x t
y
__
is a solution of the boundary value problem
2
U
x
2
+

2
U
y
2
= 0, U(x +i0
+
) = f(x) (in T
(R)) .
Using Theorem 2.41, we can nd the asymptotic behavior of U at innity
over cones. Indeed, let 0 < < /2, then Theorem 2.41 implies that as
r
U(re
i
) sin
()C
L(r), uniformly for [, ] ,

where C
= g P(cot ).
2.12 Quasi-asymptotic boundedness
This section is intended to study the structure of the distributional relation
f(x) = O(()) ,
where here or 0
+
and is a regularly varying function. Our
approach to the problem follows the exposition from [172]. Distributions
satisfying this relation will be called quasi-asymptotically bounded distribu-
tions, we make this more precise in the following denition.
161
Denition 2.14. Let L be a slowly varying function at innity (respec-
tively at the origin). We say that f T
is quasi-asymptotically bounded
at innity (at the origin) in T
(R) with respect to
L(), R, if
f(x), (x) = O(
L()) as T(R), (2.104)

(respectively 0
+
). If (2.104) holds, it is also said that f is quasi-
asymptotically bounded of degree at innity (at the origin) with respect
to the slowly varying function L. We express (2.104) by
f(x) = O(
L()) as in T
(R) , (2.105)
(respectively 0
+
).
Note that in analogy to the quasi-asymptotic behavior of distributions,
we may talk about (2.105) in other spaces of distributions. In the case
at innity, It will follow from our structural theorem that f o
(R) and
actually the relation (2.105) holds in o
(R). The case at the origin is related

to the problem of extension of distributions from R 0 to R. Indeed, if
f T
(R 0) and (2.104) holds for all T(R 0), we will see later
that f admits an extension to R.
We now proceed to obtain the structure of quasi-asymptotically
bounded distributions. For this aim, the program established in 2.10 will
be followed. We will integrate the relation (2.105) and the coecients of
this integration will satisfy the properties of the following denition.
Denition 2.15. A function b is said to be asymptotically homogeneously
bounded of degree at innity with respect to the slowly varying function
L if it is measurable and dened in some interval (A, ), A > 0, and for
each a > 0
b(ax) = a
b(x) +O(L(x)), x . (2.106)

Similarly, one denes asymptotically homogeneously bounded functions
at the origin. Our rst goal is to study the asymptotic properties of this
class of functions. Proceeding as in Lemma 2.2, or using the results of [148],
Section 2.4, one has that (2.106) must hold uniformly in compact subsets of
(0, ). Most of the proofs of the following results are the analog to those
for asymptotically homogeneous functions by replacing the o symbol by
the O symbol and making obvious modications to the estimates, therefore
they will be omitted.
Proposition 2.19. Let b be asymptotically homogeneously bounded at in-
nity (at the origin) with respect to the slowly varying function L. If the
degree is negative (respectively positive), then b(x) = O(L(x)), as x
(x 0
+
).
Proposition 2.20. Let b be asymptotically homogeneously bounded at in-
nity (at the origin) with respect to the slowly varying function L. If the
degree is positive (respectively negative), then there exits a constant
such that b(x) = x
+O(L(x)), as x (x 0
+
).
Note that for the case at innity since L(x) = O(x
) as x , for
any > 0, then any asymptotically homogeneously bounded function of
degree 0 at innity satises that b(x)/x
is asymptotically homogeneously
bounded of degree with respect to the trivial slowly varying function
L 1 and hence by Proposition 2.19 it satises b(x) = O(x
) as x ,
hence for large argument it is a regular tempered distribution. Similarly,
any asymptotically homogeneously bounded function of degree 0 at the
origin satises b(x) = O(x
) as x 0
+
, for any > 0, consequently it
is a distribution for small argument. The proof of the next proposition is
totally analogous to those of Theorems 2.24 and 2.27, and therefore it will
be omitted.
Proposition 2.21. Let b be asymptotically homogeneously bounded of de-
gree zero at the innity (at the origin) with respect to the slowly vary-
ing function L. Suppose that b is locally integrable on [A, ) (respectively
(0, A]). Then
b(x)H(x A) = b()H(x) +O(L()) as in o
(R), (2.107)
(resp. b(x)(H(x) H(x A)) = b()H(x) + O(L()) as 0
+
in
T
(R)).
Corollary 2.3. Let b be an asymptotically homogeneously bounded function
of degree 0 at innity (at the origin) with respect to L. Then, there exists
c C
[0, ), being asymptotically homogeneously bounded of degree 0,

such that b(x) = c(x) +O(L(x)) as x (resp. as x 0
+
).
Proof. We only show the assertion at innity, the case at the origin is sim-
ilar. Find B such that b is locally bounded in [B, ). Take T(R) sup-
ported in (0, ) such that
_
0
(t)dt = 1 and set c(x) =
_
B/x
b(xt)(t)dt,
the corollary now follows from Proposition 2.21.
163
The main connection between quasi-asymptotically bounded distribu-
tions and asymptotically homogeneously bounded functions is given in the
next proposition, again the proof will be omitted since it is analogous to
that of Proposition 2.17.
(R) be quasi-asymptotically bounded of de-

gree at innity (at the origin) with respect to the slowly varying function
L. Let m N. Then, for any given F
m
, an m-primitive of f in T
(R),
there exist functions b
0
, . . . , b
m1
, continuous on (0, ), such that
F
m
(x) =
m1
j=0
+m
b
j
()
x
m1j
(m1 j)!
+O
_
+m
L()
_
in T
(R),
(2.108)
as (respectively 0
+
), where each b
j
is asymptotically homoge-
neously bounded of degree j 1 with respect to L.
Thus we obtain from Propositions 2.192.22 our rst structural theorem.
(R) and / N. Then f is quasi-

asymptotically bounded of degree at innity (at the origin) with respect to
the slowly varying function L if and only if there exist m N, m+ > 0,
and a continuous (continuous near 0) m-primitive F of f such that
F(x) = O
_
[x[
m+
L([x[)
_
, (2.109)
as [x[ (respectively x 0) in the ordinary sense. Moreover, in the
case at innity, f belongs to o
(R) and is quasi-asymptotically bounded of

degree with respect to L in o
(R).
Proof. We only discuss the case at innity, the proof of the assertion at
the origin is similar to this case. It follows from Proposition 2.22, Proposi-
tion 2.19 and Proposition 2.20 that given m N and an m-primitive F
m
,
there is a polynomial p
m1
of degree at most m1 such that
F
m
(x) = p
m1
(x) +O(
+m
L()) as in T
(R), (2.110)
from the denition of boundedness in T
(R) it follows that there is an

m > such that (2.110) holds uniformly for x [1, 1]. We let F =
F
m
p
m1
, so by taking x = 1, x = 1 and replacing by x in (2.110) we
obtain (2.109). The converse follows by observing that (2.109) implies that
F(x) = O(
+m
L()) in o
(R) which gives the result after dierentiating

m-times.
We now analyze the case of negative integral degree.
(R) and let k be a positive integer. Then f is

quasi-asymptotically bounded of degree k at innity (at the origin) with
respect to L if and only if there exist m > k, an asymptotically homoge-
neously bounded function b of degree 0 at innity (at the origin) with respect
to L and a continuous (continuous near 0) m-primitive F of f such that
F(x) = b ([x[) x
mk
+O
_
[x[
mk
L([x[)
_
, (2.111)
as [x[ (x 0). Moreover (2.111) is equivalent to have
a
km
F(ax) (1)
mk
F(x) = O
_
x
mk
L(x)
_
, (2.112)
as x (x 0
+
), for each a > 0. In the case at innity, it follows that
f is quasi-asymptotically bounded of degree k with respect to L in o
(R).
Proof. Again we only give the proof of the assertion at innity, the
case at the origin is similar. If f(x) = O(
k
L()) in T
(R), then af-

ter k 1 integrations Proposition 2.22 and Proposition 2.20 provide us of a
(k1)-primitive of f which is quasi-asymptotically bounded of degree 1 at
innity with respect to L, hence we may assume that k = 1. Next, Propo-
sition 2.22, Proposition 2.19 and the denition of boundedness in T
(R)
give to us the existence of an m > 1, an asymptotically homogeneously
bounded function of degree 1 with respect to L and an m-primitive F of
f such that F(x) is continuous for x [1, 1] (hence F is continuous on
R) and F(x) =
m1
b()x
m1
+ O(
m1
L()) as uniformly for
x [1, 1]. By taking x = 1, x = 1 and replacing by x one gets (2.111).
Assume now (2.111), by using Corollary 2.3, we may assume that b is lo-
cally integrable on [0, ). This allows the application of Proposition 2.21
to deduce that F(x) =
m1
b()x
m1
+O(
m1
L()) as in o
(R)
and hence the converse follows by dierentiating m-times. That (2.111) im-
plies (2.112) is a simple calculation; conversely, setting b(x) = x
km
F(x)
for x > 0, one obtains (2.111).
Remark I.1. Even if not assumed initially, the proof of Theorem 2.43
forces (2.112) to hold uniformly on compact subsets of (0, ).
165
We remark that the results from 2.11 are also true in the context of
quasi-asymptotic boundedness. Indeed, if one proceeds as in 2.11 but now
using the structural theorems of the present section, then one obtains the
proofs for the following theorems.
(R). If f is quasi-asymptotically bounded at

0, with respect to a slowly varying function L, in T
(R), then f is quasi-

asymptotically bounded at 0 of the same degree with respect to L in the space
o
(R).
(R) satisfy f(x) = O(
L()) as in
the space T
(R). If + < 1, then f admits an extension to 1
(R)
which is equally quasi-asymptotic bounded in the space 1
(R).
Theorem 2.46. Let f
0
T
(0, ). Let L be slowly varying at the origin

and R. Suppose that
f
0
(x) = O(
L()) as 0
+
in T
(0, ). (2.113)
Then f
0
admits extensions to [0, ). Let f be any of such extensions. Then
f has the following asymptotic properties at the origin:
(i) If / N, then there exist constants a
0
, . . . , a
n
such that
f(x) =
n
j=0
a
j
(j)
(x) +O(
L()),
as 0
+
in T
(R).
(ii) If = k, k N, then there are constants a
0
, . . . , a
n
and an
associate asymptotically homogeneously bounded function of degree
0 with respect to L,
b(ax) = b(x) +O(L(x)), (2.114)
such that
f(x) =
n
j=0
a
j
(j)
(x) +b()
(k1)
(x) +O
_
k
L()
_
, (2.115)
as 0
+
in T
(R).
2.13 Relation between the S-asymptotics and
quasi-asymptotics at
It is not easy to nd conditions for a T T
which imply that T has both

the S-asymptotics and quasi-asymptotics. The S-asymptotics is a local
property (Theorem 1.4), whereas the quasi-asymptotics has in general a
global character (Proposition 2.4, Example 4 in 2.3, and Example 3 in
2.9). Also, we have to nd a subspace of T
in which we can compare these

two denitions of the asymptotic behavior and choose the class of functions
which will measure the asymptotic behavior following these denitions.
We have seen that the space of tempered distributions is a natural
one for the quasi-asymptotics (Theorem 2.3, Theorem 2.10 and 2.11.2),
while for the S-asymptotics the space /
1
has this role (Theorem 1.17).
The following examples illustrate the problem of comparison of these two
types of asymptotic behavior in o
(R).
i) The regular distribution T(t) = H(t)e
iat
, t R, a ,= 0, has the
quasi-asymptotics
i
a
in o
(R) related to c(k) = k

1
(cf. [32]):
kH(kt)e
ikat
, (t) = k
_
0
e
ikat
(t)dt =
1
ia
_
0
_
x
k
_
d(e
iax
)
=
1
ia
(0)
1
k
_
0
e
iax
_
x
k
_
dx
i
a
(0), k .
But the distribution T has no S-asymptotics related to h
with a U ,= 0
for any R. We start with
H(t +h)e
ia(t+h)
, (t) = e
iah
_
h
e
iat
(t)dt
e
iah
e
iat
(t)dt, h R
+
.
This distribution has the S-asymptotics, but related to the oscillatory func-
tion c(h) = e
iah
.
167
ii) The regular distribution T(t) = H(t) sint, t R, has the quasi-
asymptotics related to c(k) = k
1
, but it has no S-asymptotics at all:
kH(kt) sin kt, (t) =
_
0
sin u
_
u
k
_
du = (0) +
1
k
_
0
_
u
k
_
cos u du .
For the S-asymptotics we have, h ,
H(t+h) sin(t+h), (t) = cos h
sin t(t)dt+sin h
cos t(t)dt+o(1) .
iii) For the regular distribution T = H(t) sin
t, t R, we cannot nd
an R and a distribution U
,= 0 such that
lim
k
_
0
k
sin
kt (t)dt = U
, , o(R) .
Suppose on the contrary that such and U
do exist. We choose o
+
such that, (t) = e
pt
, t > 0, where Rep > 0. Then, we have
lim
k
k
_
0
sin
kte
pt
dt = U
(t), e
pt
.
The value of the last integral is
k/
_
4p
3
exp(k/4p) and U
(y), e
pt
is the Laplace transform of U
. Thus the last relation says that the Laplace

transform of U
equals zero for Rep > 0, hence U
= 0.
If we change the basic space, the conclusion can be quite dierent. Sup-
pose that the basic space is /
1
(R).
The function H(x) cosh x =
H(x)
2
(e
x
+e
x
), x R denes an element
in /
1
and exp(px
2
); p > 0 is in /
1
.
In order to nd the quasi-asymptotics at innity in /
1
of H(x)cosh(x),
x R, we use
_
0
cosh(kx)e
px
2
dx =
1
2
_
0
cosh(k
x)e
px
dx
x
=
4p
exp(k
2
/4p) .
This shows that there exists no function c(k) of the form e
ak
r
k
b
L(k),
where a, b R, 0 r < 2, L(k) is a slowly varying function, such that
H(x) cosh x has the quasi-asymptotics related to c(k) with a limit U ,= 0.
For the S-asymptotics, we have
e
h
_
h
cosh(x +h)(x)dx =
1
2
_
h
e
x
(x)dx +
1
2
e
2h
_
h
(x)dx
1
2
e
x
(x)dx, k , /
1
.
Therefore, H(x) cosh x, x R, has the S-asymptotics related to c(h) =
e
h
with the limit U =
1
2
e
x
.
The common space in which we can compare the two types of asymp-
totic behavior is the space of tempered distributions and the common class
of functions for the comparison is the class of regularly varying functions
h
L(h); R, where L is a slowly varying function (cf. 0.3).

The rst result of this kind is the following one.
Proposition 2.23. ([32]). Let g o
+
, > 1, and let
0
o(R) be
such that (T
0
)(x) = 1 on a neighborhood of zero. If there exists
lim
h
g(x +h)/h
,
0
(x) = C
_

0
(x)dx, C ,= 0 ,
then g(kx)
q
k
Cf
+1
(x), k , in o
(R). (For the function f
see
0.4).
Later on, this result has been improved (see Theorem 6, Chapter I, 3.3
in [192]) in such a way that, instead of h
, h
L(h) was used, where L is a

slowly varying function. In case 1, the situation is quite dierent. In
[120], one can nd precise results for = 1, but for < 1 the obtained
results must include the quasi-asymptotic behavior when the limit equals
zero, as well.
We quote some results of this kind proved in [159]. Denote

L(x) =
x
_
a
L(t)
t
dt, x > a. If

L(x) and L is slowly varying, then

L is slowly
varying, as well.
Theorem 2.47. Suppose that T o
+
has S-asymptotics related to
h
L(h). Then T has also the quasi-asymptotics related to h
(h), where
and L
are determined as follows: If > 1, then = , L
= L; if
169
= 1 and

L(x) , x , then = 1, L
=

L; if < 1 or
= 1 and

L(x) < , then = 1, L
= const., but the limit can be zero.

Finally, as an illustration of the case < 1 in Theorem 2.47, we
can use the distribution T = C
1
(r)
+C
2
(x
L(x))
+
; the quasi-asymptotics
depends not only on but also on r.
We can compare the quasi-asymptotics with the S-asymptotics not only
for the elements belonging to o
+
but also when they belong to T
. In this
case, we shall use Denition 2.4 instead of Denition 2.2. We know that
the general form of the function c, related to which we can measure the
S-asymptotics, is c(h) = exp(h)L(exp h), R.
Theorem 2.48. If T T
(R) has S-asymptotics related to c(h) = exp(h)

L(exph) with ,= 0 and with non-zero limit, then T cannot have quasi-
asymptotics.
At the end of this part dealing with the quasi-asymptotics, we shall
cite additionally the following papers: [8], [46], [50], [55], [113], [121], [124],
[159], [174], [199] and [201].

Asymtotic Behaior

Hochgeladen von

Dokumentinformationen

Originalbeschreibung:

Originaltitel

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Asymtotic Behaior

Hochgeladen von

Copyright:

Verfügbare Formate

September 1, 2011 10:53 World Scientic Book - 9in x 6in maj16

to the cone is the set R

is also a cone which is convex and

L(x), x > a, where L is slowly varying. It is known that the convergence

with the strong topology, as well (cf.

; supp f . In the one-dimensional case o

() if and only if it belongs to some o

() if is closed convex solid cone ([189], [192]).

the space of distributions with compact support; it is isomorphic to

with the property:

is dened as follows. Let

() with the operation are associative and

, is the space of tempered ultradistributions

(R); supp f [0, ).

with the induced convergence structure from o

; its strong dual is in fact

consists of all points at innity in all direction). Space T

is the space Q(D

are open -th orthants in R

> 0 such that [f(x)[ C

0) denotes the element of the quotient space given in (0.9); it is

and g = [G] Q(D

is a distribution, ultradistribution, or Fourier hyperfunction space, re-

it coincides with the usual

(R) is dened in this way, and not as a regular

(R) in the sense described above.

have S-asymptotics T(x + h)

and with the property

, and we do not loose generality.

for the general space T

, then L is slowly varying at the

(t)dt (0) +Mx, x ,

are the only elements of /

= 0. This generalized function is

, for all h R; in particular,

and, for all m N,

c(x), x . Set now g

Theorem 1.1. Let T T

and let int ,= . The following assertions are

the following assertion is also equivalent to a).

respectively. Since the convolution in T

c(h) (P(D)U)(x), h , |h| in Q(D

> 0 such that supp B(0, r

. The support of T(x + h) is (supp T h). Thus, by

such that for all h , |h|>

) is empty and consequently T(x+h), (x) = 0, h

and a local operator maps continu-

: The limit U can be a constant generalized function, hence

= 0. Is there a positive function c such that T

has S-asymptotics related

(x) = 2x(1 + exp(x

has the S-asymptotics in related to

c(h)U(x), h . The question is what we can say about the S-asymptotics

(x) = A, then there exists lim

(R) and for some m N, g

(R) such that

and that for an m 1, 2, . . . , n,

Remark. The open set , by its property, has to contain a set

, R, be given as in 0.4. It denes a distribution in T

(x) = (1/()) Pf(H(x)x

, then there exists a real number k

. Hence, this set is weakly bounded and