A GENERALIZATION OF THE BANACH-STEINHAUS

THEOREM FOR FINITE PART LIMITS

RICARDO ESTRADA AND JASSON VINDAS

Abstract. It is well known, as follows from the Banach-Steinhaus

∞
theorem, that if a sequence {yn }n=1 of linear continuous function-
als in a Fréchet space converges pointwise to a linear functional Y,
Y (x) = limn→∞ hyn , xi for all x, then Y is actually continuous. In
this article we prove that in a Fréchet space the continuity of Y still
holds if Y is the finite part of the limit of hyn , xi as n → ∞. We also
show that the continuity of finite part limits holds for other classes
of topological vector spaces, such as LF-spaces, DFS-spaces, and
DFS ∗ -spaces, and give examples where it does not hold.

1. Introduction
Let X be a topological vector space over K, K being R or C. We
denote as X 0 the dual space, that is, the space of continuous linear
functionals on X; the evaluation of y ∈ X 0 on x ∈ X will be denoted
as hy, xi or as y (x) ; we shall denote as Xal0 the algebraic dual of X,
but if z ∈ Xal0 we denote evalutions as z (x) only.
Let {yn }∞ 0
n=1 be a sequence of elements of X and suppose that

(1.1) lim hyn , xi = Y (x) ,

n→∞

exists for each x ∈ X, thus defining a function Y : X → K. It is clear

that Y is linear, an element of Xal0 , and simple examples show that Y
does not have to be continuous, that is, maybe Y ∈ / X 0 . However, it is
well known [9, 12] that if X is barreled, in particular if X is a Fréchet
space or an LF space, then one must have that Y ∈ X 0 ; this result is
quite important in the theory of distributions since the usual spaces
of test functions are barreled and thus (1.1) provides a method, rather
frequently employed, to construct new distributions as limits.

2010 Mathematics Subject Classification. 46A04, 46A13.

Key words and phrases. Finite part limits, Hadamard finite part, Banach-
Steinhaus theorem.
R. Estrada gratefully acknowledges support from NSF, through grant number
0968448.
1
2 R. ESTRADA AND J. VINDAS

Our aim is to consider the continuity of Y in case the standard1 finite

part of the limit
(1.2) F.p. lim hyn , xi = Y (x) ,
n→∞

exists for each x ∈ X. We will show that Y ∈ X 0 in case X is a Fréchet

space or in case it is an inductive limit of Fréchet spaces. Naturally
several distributions are defined as finite parts, so such a result would
be very useful.
The meaning of (1.2) is that for each x ∈ X there is k = kx ∈ N0 =
N ∪ {0}, exponents α1 > · · · > αk > 0, scalars Rα1 (x) , . . . , Rαk (x) ∈
K \ {0} , and zn (x) ∈ K for n ≥ 0 such that
(1.3) hyn , xi = nα1 Rα1 (x) + · · · + nαk Rαk (x) + zn (x) ,
where
(1.4) lim zn (x) = Y (x) .
n→∞

Observe that kx could be 0, meaning that hyn , xi = zn (x) converges to

Y (x) .
We call nα1 Rα1 (x) + · · · + nαk Rαk (x) the infinite part of hyn , xi as
n → ∞ and zn (x) the finite part. Clearly the infinite and finite part, if
they exist, are uniquely determined, so that the finite part of the limit,
if it exists, is likewise uniquely determined.
It is important to point out that maybe sup {kx : x ∈ X} = ∞ and
that the set of exponents,
[
(1.5) {αj : 1 ≤ j ≤ kx } ,
x∈X

does not have to be finite. We shall show that when X is a Fréchet

space then sup {kx : x ∈ X} < ∞ and actually the set (1.5) is finite,
but give examples in other types of spaces where these results do not
hold.
We shall also show that the Rα admit extensions as elements of
the algebraic dual Xal0 , and show that while in general they are not
continuous, they must belong to X 0 when X is a Fréchet space or an
inductive limit of Fréchet spaces.
The plan of the article is as follows. In Section 2 we give some basic
facts about finite parts that hold in any topological vector space. The
central part of the article is Section 3, where we study finite parts in
a Fréchet space. Extensions to more general finite parts and to more
general topological vector spaces are considered in Sections 4 and 5,
respectively. Finally we present several illustrations in Section 6.
1We shall consider more complicated finite part limits later on.
 A GENERALIZATION OF THE BANACH-STEINHAUS THEOREM 3

2. General results
We shall first consider several results that hold in any topological
vector space.
Thus let {yn }∞ 0
n=1 be a sequence of elements of the dual space X of
the topological vector space X, and suppose that for each x ∈ X the
finite part of the limit
(2.1) Y (x) = F.p. lim hyn , xi ,
n→∞

exists, or, in other words, that the evaluation hyn , xi can be decomposed
as
(2.2) hyn , xi = un (x) + zn (x) ,
with the infinite part of the form
(2.3) un (x) = nα1 Rα1 (x) + · · · + nαk Rαk (x) ,
where α1 > · · · > αk > 0, and Rα1 (x) , . . . , Rαk (x) ∈ K \ {0} , and
with the finite part, zn (x) , such that the limit
(2.4) Y (x) = lim zn (x)
n→∞
exists.
The following result is very easy to prove, but it is also very impor-
tant.
Lemma 2.1. The decomposition (2.2) in finite and infinite parts is
unique.
It is convenient to define Rα (x) for all α > 0 and all x ∈ X. We
just put Rα (x) = 0 if α is not one of the exponents α1 , . . . , αk in the
expression of the infinite part of hyn , xi . This allows us to rewrite (2.3)
as
X
(2.5) un (x) = nα Rα (x) ,
α>0

since only a finite number of terms of the uncountable sum do not

vanish.
Lemma 2.2. If Y (x) = F.p. limn→∞ hyn , xi exists for all x ∈ X then
Y is linear: Y ∈ Xal0 .
Proof. Indeed, if x1 , x2 ∈ X and c ∈ K, then hyn , x1 + cx2 i admits the
decomposition
hyn , x1 + cx2 i = hyn , x1 i + c hyn , x2 i
= (un (x1 ) + cun (x2 )) + (zn (x1 ) + czn (x2 )) .
4 R. ESTRADA AND J. VINDAS

Since
X
(2.6) un (x1 ) + cun (x2 ) = nα (Rα (x1 ) + cRα (x2 )) ,
α>0

has the form of an infinite part, the Lemma 2.1 yields

un (x1 + cx2 ) = un (x1 )+cun (x2 ) , zn (x1 + cx2 ) = zn (x1 )+czn (x2 ) ,
and consequently Y (x1 + cx2 ) equals
lim zn (x1 + cx2 ) = lim (zn (x1 ) + czn (x2 )) ,
n→∞ n→∞

that is, Y (x1 ) + cY (x2 ) . 

If we now use the fact that Rα (x) = F.p. limn→∞ n−α hyn , xi , or
employ (2.6), we immediately obtain the ensuing result.
Lemma 2.3. For any α > 0 the function Rα is linear, Rα ∈ Xal0 .
In general Y nor all Rα will not be continuous, as the Example 6.5
shows.

3. Finite parts in a Fréchet space

We shall now consider the continuity and structure of finite parts in
a Fréchet space. We start with some useful preliminary results.
Lemma 3.1. Let X be a Fréchet space and let {yn }∞
n=0 be a sequence of
0
non zero elements of X . Then there exists x ∈ X such that hyn , xi =
6 0
∀n ∈ N0 .
Proof. Indeed, since yn 6= 0 the kernel of yn , Fn = {x ∈ X : hyn , xi = 0}
is a closed proper subspace of X and thus of first category. Hence
∪∞n=0 Fn 6= X. 
Observe that this result fails in spaces that are not Fréchet. Consider,
for example, the sequence {δ (t − n)}∞ 0
n=0 in the space D (R) .
Recall that a function f : W → V, where W and V are topological
spaces, is called a Baire function of the first class if there exists a
sequence of continuous functions from W to V, {fn }∞ n=0 , such that
f (w) = limn→∞ fn (w) for all elements w ∈ W.
Lemma 3.2. Let X be a Fréchet space and let A : X → R be a function
that satisfies the following three properties:
1. A is a Baire function of the first class;
2. A (x − y) ≤ max {A (x) , A (y)} ;
3. A (cx) = A (x) if c 6= 0.
Then A is bounded above in X and it actually attains its maximum.
 A GENERALIZATION OF THE BANACH-STEINHAUS THEOREM 5

Proof. We shall first show that A is bounded above. If F is a subset of

X, denote by MF = sup {A (x) : x ∈ F } . If U is a neighborhood of 0,
then 3 yields that MX = MU . Let now V be any set with non empty
interior; then V − V is a neighborhood of 0 and thus 2 yields that
MX = MV −V ≤ MV ≤ MX so that MX = MV .
Let αn be a sequence of continuousSfunctions from X to R that con-
verges to A everywhere. Then X = ∞ k=0 {x ∈ X : αn (x) ≤ k , ∀n} so
that there exists k ∈ N such that the set Vk = {x ∈ X : αn (x) ≤ k ∀n}
has non empty interior. This yields that MX = MVk ≤ k, so that A is
bounded above by k in the whole space X.
We should now show that there exists x e ∈ X such that A (ex) = MX .
If not, the function B (x) = 1/ (MX − A (x)) satisfies the same three
conditions as A, and from what we have already proved, B must
be bounded above by some constant λ > 0; but this means that
A (x) ≤ MX − 1/λ, for all x ∈ X, and consequently MX ≤ MX − 1/λ,
a contradiction. 

We now apply the Lemma 3.2 to the study of finite parts. Indeed, let
{yn }∞ 0
n=1 be a sequence of elements of the dual space X of the Fréchet
space X, and suppose that for each x ∈ X the finite part of the limit
Y (x) = F.p. limn→∞ hyn , xi exists. If the infinite part of hyn , xi has the
expression as a finite sum,
X
(3.1) un (x) = nα1 Rα1 (x) + · · · + nαk Rαk (x) = nα Rα (x) ,
α>0

where α1 > · · · > αk > 0, and Rα1 (x) , . . . , Rαk (x) ∈ K \ {0} , define
A (x) = 0 if un (x) = 0 and as
(3.2) A (x) = α1 = max {α > 0 : Rα (x) 6= 0} ,
otherwise.
Lemma 3.3. The function A is bounded above and attains its maxi-
mum in X.
Proof. It is enough to prove that A satisfies the three conditions of the
Lemma 3.2. However, condition 1 follows from the limit formula
ln [|hyn , xi| + 1]
(3.3) A (x) = lim ,
n→∞ ln n
while 2 and 3 are obvious. 

The Lemma 3.3 not only means that if α e = max {A (x) : x ∈ X}

e, but it also means that Rαe 6= 0. The linear
then Rα (x) = 0 if α > α
6 R. ESTRADA AND J. VINDAS

form Rαe is actually continuous as follows from the Banach-Steinhaus

theorem since
(3.4) Rαe (x) = lim n−eα1 hyn , xi ,
n→∞

for each x ∈ X.
We can then replace yn by yn − nαe Rαe and apply the same ideas as
above. Therefore, for some integers k we obtain exponents α e1 = α
e>
··· > α ek > 0 such that Rα (x) = 0 if α > α ek , α 6= αej , 1 ≤ j < k,
while Rαej is continuous and Rαej 6= 0 for 1 ≤ j ≤ k. In principle one
could think that this is possible for each k ≥ 0, but if it were then we
would obtain an infinite sequence of non zero continuous functionals
∞
Rαej j=0 and the Lemma 3.1 would give us the existence of x∗ ∈ X


such that Rαej (x∗ ) 6= 0 for all j, a contradiction, since for any x ∈ X the
set {α > 0 : Rα (x) 6= 0} is finite. Summarizing, we have the following
result.
Theorem 3.4. Let {yn }∞ n=1 be a sequence of elements of the dual space
X 0 of the Fréchet space X, and suppose that for each x ∈ X the finite
part of the limit Y (x) = F.p. limn→∞ hyn , xi exists. Then there exists
k ∈ N, exponents α e1 > · · · > α
ek > 0, and continuous non zero linear
 k
functionals Rαej j=1 such that for all x ∈ X

(3.5) hyn , xi = nαe1 Rαe1 (x) + · · · + nαek Rαek (x) + zn (x) ,

where the finite part zn is continuous for all n and where
(3.6) Y (x) = lim zn (x) ,
n→∞

is also a continuous linear functional on X.

Proof. The only thing left to prove is the continuity of the zn ’s and the
continuity of Y. But the continuity of the Rαej ’s yields the continuity of
the zn ’s because of (3.5) while the continuity of Y follows from (3.6)
and the Banach-Steinhaus theorem. 

4. More general finite parts

One can consider a general finite part limit process as follows. Let
Λ ∪ {λ0 } be a topological space where λ0 ∈ Λ \ Λ, and let (E, ≺) be a
totally ordered set. Let B = {ρα }α∈E be the “basic infinite functions,”
that is, a family of functions with the following properties:
(1) For each α ∈ E, ρα : Λ → (0, ∞) , and limλ→λ0 ρα (λ) = ∞;
(2) If α ≺ β then ρα (λ) = o (ρβ (λ)) as λ → λ0 .
 A GENERALIZATION OF THE BANACH-STEINHAUS THEOREM 7

Let yλ ∈ K, where K is R or C, for each λ ∈ Λ. If we can write

(4.1) yλ = uλ + zλ ,
where the “infinite part” has the form
k
X
(4.2) uλ = Rαj ραj (λ) ,
j=1

where α1  · · ·  αk , and Rα1 , . . . , Rαk ∈ K \ {0} , and where the

“finite part,” zλ , satisfies that the limit
(4.3) Y = lim zλ
λ→λ0

exists, then we say that the finite part of the limit of yλ as λ → λ0

with respect to B exists and equals Y, and write2
(4.4) Y = F.p.B lim yλ .
λ→λ0

We have considered the standard system B = {ρα }α>0 where Λ = N,

λ0 = ∞, and ρα (λ) = λα . Naturally one can consider the same standard
system for functions defined in any unbounded set Λ ⊂ (0, ∞), in
particular for Λ = (0, ∞).
We can also consider Hadamard finite part limits3, where the infinite
basic functions are products of powers and powers of logarithms. Ex-
plicitly, let E = [0, ∞)2 \ {(0, 0)} , with the order given by (α1 , β1 ) ≺
(α2 , β2 ) if α1 < α2 or if α1 = α2 and β1 < β2 . Here Λ ⊂ (1, ∞), and
the basic infinite functions B = ρ(α,β) (α,β)∈E are given as ρ(α,β) (λ) =
λα lnβ λ.
We can also take a set  ⊂ (0, 1) and consider limits as ε ∈ 
tends to 0. For standard finite part limits E = (0, ∞) and the basic
infinite functions are ρα (ε) = ε−α ; for Hadamard finite part limits,
E = [0, ∞)2 \ {(0, 0)} and ρ(α,β) (ε) = ε−α |ln ε|β .
The continuity of the finite part of the limit in Fréchet spaces, The-
orem 3.4, will also hold for these more general systems of basic infinite
functions. This is of course the case for standard finite limits. For
Hadamard finite parts the proof can be modified as follows. Indeed,
let X be a Fréchet space, Λ ⊂ (1, ∞) is an unbounded set and yλ ∈ X 0
2Let Pk
V be the vector space of all functions of the form j=1 cj ραj + µ with
limλ→λ0 µ(λ) = 0. The triple N = (Λ, K, V ) forms a neutrix in the sense of van
der Corput [13]. In his terminology, the finite part limit (4.4) coincides with the
neutrix value yN .
3Hadamard was probably the first to use finite parts; in his 1923 work [7], he
employs them to find fundamental solutions of partial differential equations.
8 R. ESTRADA AND J. VINDAS

for each λ ∈ Λ. Suppose that for each x ∈ X the evaluation hyλ , xi can
be written as
(4.5) hyλ , xi = uλ (x) + zλ (x) ,
where the infinite part has the ensuing form for some k = kx ,
(4.6) uλ (x) = λα1 lnβ1 λ R(α1 ,β1 ) (x) + · · · + λαk lnβk λ R(αk ,βk ) (x) ,
where the exponents (αj , βj ) ∈ E satisfy (α1 , β1 )  · · ·  (αk , βk ) ,
where R(α1 ,β1 ) (x) , . . . , R(αk ,βk ) (x) ∈ K \{0} , and where the finite part,
zλ (x) , satisfies that the limit
(4.7) Y (x) = lim zλ (x) ,
λ→∞,λ∈Λ

exists. As before, we set R(α,β) (x) = 0 if (α, β) 6= (αj , βj ) for 1 ≤ j ≤

kx . Then we have the following generalization of the Lemma 3.3.
Lemma 4.1. Let

(4.8) E (x) = (A (x) , B (x)) = max (α, β) ∈ E : R(α,β) (x) 6= 0 .
Then E attains its maximum, (α∗ , β ∗ ) , and R(α∗ ,β ∗ ) is continuous and
not zero.
Proof. The proof of the Lemma 3.3 applies to A, so there exists α∗ =
maxx∈X A (x) . For this exponent α∗ we consider the function given by
B ∗ (x) = 0 if R(α∗ ,β) (x) = 0 for all β and otherwise by
B ∗ (x) = max β : R(α∗ ,β) (x) 6= 0 .

(4.9)
The Lemma 3.2 yields the existence of β ∗ = maxx∈X B ∗ (x) because
 −α∗ 
ln λ |hy λ , xi| + 1
(4.10) B ∗ (x) = lim .
λ→∞ ln ln λ
Since
∗ ∗
(4.11) R(α∗ ,β ∗ ) (x) = lim (λ−α ln−β λ) hyλ , xi ,
λ→∞

the continuity of R(α∗ ,β ∗ ) follows. 

Therefore we obtain that the Theorem 3.4 holds for Hadamard finite
parts.
These ideas can be further generalized. Let us take m positive func-
tions F1 , F2 , F3 , . . . , Fm defined on an unbounded set Λ ⊂ (1, ∞), each
of them tending to ∞ as λ → ∞ and such that Fj+1 (λ) = o(Fjα (λ)) as
λ → ∞ for all α > 0. We now consider E = [0, ∞)m \ {(0, 0, . . . , 0)}
with the lexicographical order ≺ . Set F = (F1 , F2 , . . . , Fm ) and if
~ ∈ E write Fα~ = F1α1 F2α2 · · · Fmαm , where α
α ~ =(α1 , α2 , . . . , αm ). We
now choose the basic infinite functions as B = Fα~ α~ ∈E . If yλ ∈ X 0
 A GENERALIZATION OF THE BANACH-STEINHAUS THEOREM 9

for each λ ∈ Λ, where X is again a Fréchet space, and for each x ∈ X

the evaluation hyλ , xi can be decomposed as in (4.5) where the infinite
part is now taken of the form (for some k = kx ,)

(4.12) uλ (x) = Fα~ 1 (λ)Rα~ 1 (x) + · · · + Fα~ k (λ)Rα~ 1 (x) ,

with α~1  · · ·  α~ k , and Rα~ 1 (x) , . . . , Rα~ k (x) ∈ K \ {0} , and where
zλ (x) satisfies (4.7), we can then define the finite part limit of yλ (x) as
Y (x). Defining Rα~ (x) = 0 if α ~ does not occur in (4.12), the proof of
Lemma 4.1 can be readily adapted to show that the function A(x) =
max {~α ∈ E : Rα~ (x) 6= 0} also attains its maximum, α ~ ∗ , and that Rα~ ∗ ∈
0
X \ {0}. This leads to a general version of Theorem 3.4 for finite part
limits with respect to the system of infinite functions B = Fα~ α~ ∈E .
Naturally, the Hadamard finite part corresponds to the choices m = 2,
F1 (λ) = λ, and F2 (λ) = log λ.

5. Other types of topological vector spaces

The continuity of finite part limits holds not only in Fréchet spaces,
but in other types of spaces, those that carry a final locally convex
topology given by a family of Fréchet spaces [9]. Indeed, let (Xi , ui )I
be a family of Fréchet spaces and linear mappings ui : Xi → X for
each i ∈ I. If X is provided with the finest locally convex topology
that makes all mappings ui continuous, the continuity of the finite part
limits and the Rα~ ’s follows at once from the fact that y ∈ X 0 if and
only if y ◦ ui ∈ Xi0 , ∀i ∈ I. In particular, the result holds for any
inductive limit of an inductive system of Fréchet spaces. Important
instances of such inductive limits are those that can be written as
countable inductive unions of Fréchet spaces, such as the LF -spaces,
the DFS-spaces, and the DFS ∗ -spaces [11].
In these more general spaces, however, the set of exponents for which
Rα~ (x) 6= 0 does not have to be finite, not bounded, in general (see
Examples 6.2, 6.3, and 6.4 below).

Theorem 5.1. Let X be a locally convex space that is the inductive

limit of a system of Fréchet spaces. Let Λ be an unbounded subset of
(1, ∞) and for each λ ∈ Λ let yλ ∈ X 0 . Suppose that for each x ∈ X
the finite part of the limit Y (x) = F.p. limλ→∞ hyλ , xi exists. Then
Y ∈ X 0.
Likewise, Rα~ is continuous for each α ~ , but while for each x ∈ X the
set {~
α : Rα~ (x) 6= 0} is finite, the set {~
α : Rα~ 6= 0} could be infinite and
not bounded above.
10 R. ESTRADA AND J. VINDAS

6. Examples
In order to better understand our results, it is useful to look at several
examples.
Example 6.1. The best known example of finite parts are the dis-
tributions constructed as the finite part of divergent integrals [4, 10].
Suppose G is a homogenous continuous function in Rd \ {0} , homoge-
nous of degree λ ∈
R. Then G gives a well defined distribution of the
space D0 Rd \ {0} , which R without loss of generality we cand still de-

note by G, as hG, φi = Rd \{0} G (x) φ (x) dx, for φ ∈ D R \ {0} .
R
When λ < −d then the
integral Rd G (x) φ (x) dx, would be divergent
in general if φ ∈ D Rd and
thus there is no canonical distribution cor-
0 d
responding to G in D R . One can, however, define the distribution
F.p. (G) , the radial finite part4 of G by setting for φ ∈ D Rd


Z
(6.1) hF.p. (G) , φi = F.p. lim G (x) φ (x) dx ,
n→∞ |x|≥1/n

a standard finite part if −λ ∈/ N, and a Hadamard finite part if λ is an

integer. Similar ideas are needed to construct thick distributions from
locally integrable functions [15] .
Example 6.2. Let D = {z ∈ C : |z| < 1} be the unit disc in C. Let
Hk be the Banach space of functions continuous in D \ {0} , analytic
in D \ {0} , and that have a pole at z = 0 of order k; the norm being
kf k = max|z|≤1 |z|k |f (z)| . Let H be the inductive limit of the Hk as
k → ∞. Consider the functionals yn ∈ H 0 given as
(6.2) hyn , f i = f (1/n) ,
that is, yn = δ (z − 1/n) . For each f ∈ H the finite part of the limit
F.p. limn→∞ hyn , f i = Y (f ) exists, and equals the finite part of f at
z = 0; in fact, if f (z) = kj=1 aj z −j + g (z) , where g is analytic at 0,
P
then
k
X
(6.3) hyn , f i = aj nj + g (1/n) .
j=1

Observe that the infinite part is kj=1 aj nj , which has arbitrary large
P

exponents; here the set (1.5) is infinite. Also Y (f ) = g (0) , the usual
4If
instead of removing balls of small radius, solids of other shapes are removed
one obtains a different finite part distribution [5, 8, 14], an important fact in the
numerical solution of integral equations [5]. The known formulas for the distribu-
tional derivatives of inverse power fields [6] and the corresponding finite parts [2, 3]
hold for radial finite parts.
 A GENERALIZATION OF THE BANACH-STEINHAUS THEOREM 11

finite part of the analytic function at the pole. Our results will yield
the continuity of Y, but one can I prove this directly, for example, by
observing that Y (f ) = (2πi)−1 z −1 f (z) dz for any r ∈ (0, 1].
|z|=r
Interestingly, if X is the space of all analytic function in D \ {0} with
its standard topology, then H is dense in X and the yn ’s and Y admit
continuous extensions to X 0 , but the extension of Y is not the finite
part of the limit of the extensions of the yn ’s.
Example 6.3. Let us consider the distributions
∞
X
k 2
(6.4) fn (x) = n1/k + 1 δ (x − k) ,
k=1
0
of the space D (R) . If φ ∈ D (R) satisfies supp φ ⊂ (−∞, k + 1) then
the infinite part of hfn , φi is the sum of k 2 − 1 terms, corresponding to
the exponents αj = j/k for 1 ≤ j ≤ k 2 . Hence the finite part of the
limit is the Dirac comb
X∞
(6.5) F.p. lim fn (x) = δ (x − k) ,
n→∞
k=1

while Rα (x) 6= 0 precisely when α is a positive rational number; and

actually if k is the smallest integer for which α = j/k and j ≤ k 2 , then
∞  2 2
X q k
(6.6) Rj/k (x) = δ (x − qk) .
q=1
qj

One
P could represent the infinite part of fn (x) as the infinite sum
α∈Q+ Rα (x) ; the set of exponents α for which Rα 6= 0 is the infi-
nite unbounded set Q+ , but upon evaluation on a test function the
sum becomes finite since hRα , φi =
6 0 for only a finite set of exponents.
In the previous example X = D (R) is an LF space, and it is not
hard to see that in an LF space the set of exponents α for which Rα 6= 0
is countable at the most. We can easily construct an example where
this set of exponents is the whole (0, ∞) .
Example 6.4. Let X be the space of functions f : (0, ∞) → R such
that the set {α ∈ (0, ∞) : f (α) 6= 0} is
finite. We give X the inductive
limit topology of the system RF , iF , where F is a finite subset of
(0, ∞) , F %, and if f ∈ RF , then iF (f ) = fF ∈ X is given by
fF (α) = f (α) if α ∈ F and fF (α) = 0 if α ∈ / F. Theorem 5.1 applies
in X.
12 R. ESTRADA AND J. VINDAS

Let yn ∈ X 0 be given as
X
(6.7) yn (x) = nα δ (x − α) ,
α∈(0,∞)
α
P
that is hyn , f i is the finite sum α∈(0,∞) n f (α) . Then Rα (x) =
δ (x − α) for all α ∈ (0, ∞) , so that the set (1.5) is the whole (0, ∞) .
Notice also that F.p. limn→∞ yn (x) = 0.
Example 6.5. Consider the space X whose elements are the continu-
ous functions in [0, 1] , with the topology of pointwise convergence on
[0, 1] . If 0 < β < 1, let us consider the functional fn ∈ X 0 given by
n−1  
X k+β
(6.8) fn (x) = δ x− ,
k=0
n
that is,
n−1  
X k+β
(6.9) hfn , φi = φ ,
k=0
n
for φ ∈ X. Then the Euler-Maclaurin formula [1] yields
Z 1
(6.10) hfn , φi = n φ (x) dx + B1 (β) (φ (1) − φ (0)) ,
0

where B1 (x) = x − 1/2 is the Bernoulli polynomial of order 1. The

finite part is
(6.11) F.p. lim fn (x) = B1 (β) (δ (x − 1) − δ (x)) ,
n→∞

which is actually continuous in X, but R1 , the coefficient of n in the

infinite
R1 part of hfn , φi is not continuous since it is given by φ
0
φ (x) dx, which belongs to Xal0 but not to X 0 .
Example 6.6. Let Ω be a complex region and let ξ ∈ Ω. Suppose
yω ∈ X 0 is weakly–∗ analytic in ω ∈ Ω \ {ξ} , that is, for each x ∈ X
the function hyω , xi is analytic in Ω \ {ξ} . Suppose also that hyω , xi has
a pole at ω = ξ for each x.
If X is a Fréchet space then there exists a fixed number N such that
the order of the pole is N at the most for all x, and the finite part
(6.12) yξ∗ (x) = F.p. lim yξ+1/n , x ,
n→∞

is an element of X 0 .
If X is an inductive limit of Fréchet spaces, then yξ∗ is still continuous,
but the order of the pole of hyω , xi at ξ does not have to be bounded,
 A GENERALIZATION OF THE BANACH-STEINHAUS THEOREM 13

that is, maybe Rk (x) = F.p. limn→∞ n−k yξ+1/n , x does not vanish in
X for an infinite number of values of k.
References
[1] Estrada, R., On the Euler-Maclaurin asymptotic formula, Bol. Soc. Mat. Mex.
3 (1997), 117–133.
[2] Estrada, R. and Kanwal, R. P., Regularization and distributional derivatives

n/2
of x21 + · · · x2p in D0 (Rp ) , Proc. Roy. Soc. London A 401 (1985), 281–297.
[3] Estrada, R. and Kanwal, R. P., Regularization, pseudofunction, and Hadamard
finite part, J. Math. Anal. Appl. 141 (1989), 195–207.
[4] Estrada, R. and Kanwal, R.P., A distributional approach to asymptotics. The-
ory and applications, Second Edition, Birkhäuser, Boston, 2002.
[5] Farassat, F., Introduction to generalized functions with applications in aerody-
namics and aeroacoustics, NASA Technical Paper 3248 (Hampton, VA: NASA
Langley Research Center) (1996); http://ntrs.nasa.gov.
[6] Frahm, C.P., Some novel delta-function identities, Am. J. Phys. 51 (1983),
826–829.
[7] Hadamard, J., Lectures on Cauchy’s Problem in Linear Partial Differential
Equations, Yale University Press, New Haven, 1923, (reprinted by Dover, New
York, 1953).
[8] Hnizdo, V., Generalized second-order partial derivatives of 1/r, Eur. J. Phys.
32 (2011), 287–297.
[9] Horváth, J., Topological Vector Spaces and Distributions, vol. I., Addison-
Wesley, Reading, Massachusetts, 1966.
[10] Kanwal, R.P., Generalized Functions: Theory and Technique, Third Edition,
Birkhäuser, Boston, 2004.
[11] Komatsu, H., Projective and injective limits of weakly compact sequences of
locally convex spaces, J. Math. Soc. Japan 19 (1967), 366–383.
[12] Trèves, F., Topological Vector Spaces, Distributions, and Kernels, Academic
Press, New York, 1967.
[13] van der Corput, J. G., Neutrices, J. Soc. Indust. Appl. Math., 7 (1959), 253–
279.
[14] Yang, Y. and Estrada, R., Regularization using different surfaces and the sec-
ond order derivatives of 1/r, Appl. Anal. 92 (2013), 246–258.
[15] Yang, Y. and Estrada, R., Distributions in spaces with thick points, J. Math.
Anal. Appl. 401 (2013), 821–835.

R. Estrada, Department of Mathematics, Louisiana State Univer-

sity, Baton Rouge, LA 70803, U.S.A.
E-mail address: restrada@math.lsu.edu

J. Vindas, Department of Mathematics, Ghent University, Krijgslaan

281 Gebouw S22, B 9000 Gent, Belgium
E-mail address: jvindas@cage.UGent.be