A tauberian theorem for a class of function spaces

Miguel A. Jimenez
Benemerita Universidad Aut onoma de Puebla. Puebla, Mexico.
[email protected]
Monografas de la Academia de Ciencias de Zaragoza. 20: 7785, (2002).
Abstract
Let (E, .
E
) and (F, .
F
) , F E, be Banach spaces. Assume that .
F
:=
.
E
+ (.), where is a seminorm. It is proved that sequences in F that converge
in .
E
and whose elements satisfy certain equicontinuous behavior, also converge in
.
F
to the same limit points. Quantitative estimates of the degree of convergence
are obtained. Examples of applications to dierent function spaces are presented.
Mathematics Subject Classication: 41A65
Keywords and phrases: tauberian theorem, H older approximation, Lipschitz
function, total variation, absolutely continuous function, equilipschitzian set, H older
space, Besov space, Bernstein polynomial.
1 Introduction
Let X be either the real interval [0,1] or the multiplicative group T = {z C : |z| = 1} .
Let Lip

X (Lip

for short) , 0 < < 1, be the Holder space of continuous real (or
complex) functions f C (X), which satisfy the H older (also called Lipschitz) condition

(f) := sup
>0

(f, ) < , (1)

where

(f, ) := sup {|f (x) f (y)| /d (x, y)

: 0 < d (x, y) } . (2)

Here d (x, y) := |x y| if X = [0, 1] or equal to the length of the shortest arc which
joins x and y if X = T. In the last case, if functions on T are identied with 2-periodic
functions on R, d should be the semidistance between elements of R, given by
d (x + 2j, y + 2k) := min {|x y| , 2 |x y| : x, y [0, 2[; j, k Z} . (3)

This paper has been partially supported by CONACyT Project 32181-E, Mexico and University of
Jaen, Spain.
77
Setting
f
,
:= f

(f) (4)
or another equivalent norm, the linear space Lip

becomes a Banach space. Further,

denote by lip

X (lip

for short) , 0 < < 1, the Banach subspace of those functions

f Lip

, for which

(f, ) 0 as 0. (5)
Basic results on Holder spaces can be found in [4] and [5]. A recent survey of approx-
imation in these spaces is given in [3].
From (4) , a sequence that converges in Lip

also converges in the sup-norm .

,
with the same limit. The converse is false, of course. However, there is a certain tauberian
condition (*) which lets us to prove the following assertion:
(f
n
) lip

p
, f
n
f
p
0 and (*) =f
n
f
,p
0. (6)
In fact, in 1985, Leindler, Meir and Totik proved a rst result of type (6) for X
being the group T and (f
n
) dened by a convolution process K
n
f, f lip

(see
[8]). They also estimated the degree of convergence. Later, Bustamante-Jimenez [2]
introduced the following tauberian condition: A sequence (f
n
) lip

X, 0 < < 1, is
called equilipschitzian if (5) holds uniformly in n, i.e. if
sup {

(f
n
, ) : n N} 0 as 0. (7)
The main theorem in [2] states that any equilipschitzian sequence (f
n
) in lip

p
converges
in this space whenever it converges in the sup-norm, i.e. (6). Since sequences dened by
convolution processes (K
n
f) , f lip

p
(T) and (K
n
) bounded in L
1
(T) , are equilips-
chitzian, we get another view of the qualitative part of paper [8].
When 1 p < , one denes Lip

p
and lip

p
in L
p
, through standard procedures.
Leindler, Meir and Totik announced the possibility of extending their results to lip

p
(T).
Further, in [7], Jimenez-Martnez extended most of results in [2] to these spaces.
With these antecedents at hand, one should expect a more general theorem that covers
and unies these particular results. In fact, in the next section, using a concept similar to
(7) , we establish and prove such a theorem. Estimates of the degree of convergence will
also be obtained. The last section is devoted to applications in dierent function spaces.
78
2 Denitions and results
In order to follow the ideas of this section, let us keep our mind on the examples given
by lip

p
.
Set R
+
:= {t R: t 0} , R

+
:= {t R: t > 0} and denote by I the real open
interval ]0,b[ (or semi-open ]0,b]) where I = R

+
is possible. Let E be a real or complex
linear space and
: E I R
+
{} , (8)
a family (., ) , I, of quasi-seminorms on E, i.e. the subadditivity of usual semi-
norms is substituted by the most general assertion that there exists a constant C 1
(that here we assume is independent of ), such that for every pair of elements f, g E,
one has (f + g, ) C ( (f, ) + (g, )). Without loss of generality it is also assumed
that for every xed f E, (f, .) is an increasing function (in the large sense) of . Set
(f) := sup { (f, ) : I} . (9)
Consider
F : = {f E : (f) < } (10)
F : = {f F : (f, ) 0 as 0} (11)
Then, F and F are linear subspaces of E, that are quasi-seminormed by (9) and that,
eventually, could coincide .
We remark that F is a closed subspace of (F, ). In fact, let (f
n
) F be a sequence
that converges to f F. Fix > 0. First, take n such that (f
n
f) and then
0
> 0
such that (f
n
, ) , for every
0
Thus (f, ) C ( (f
n
f) + (f
n
, )) 2C .
Denition 1 A set G F is called 0-equicontinuous if
(G, ) := sup { (g, ) : g G} 0 as 0. (12)
A sequence (f
n
) is called 0-equicontinuous if the set {f
n
: n N} is. In that case we
simplify the notation by writing
((f
n
) , ) : = ({f
n
: n N} , ) .
Of course, equilipschitzian sets in our introductory section not only are examples of
0-equicontinuous sets but also the starting point of the present denition.
79
Proposition 2 Let (f
n
) be a convergent sequence in the quasi-seminormed space (F, ).
Then such a sequence is 0-equicontinuous.
Proof: Suppose (f
n
f) 0 for some f F. Fix > 0 and choose N such that
(f
n
f) whenever n > N. Also choose
0
I, such that (f,
0
) . Then, for
any 0<
0
and n > N,
(f
n
, ) C ( (f
n
f, ) + (f, )) C ( (f
n
f) + (f,
0
)) 2C .
For i = 1, 2, ..., N, choose
i
such that (f
i
,
i
) . Set := min {
i
: 0 i N} . Thus
sup { (f
n
, ) : n N} 2C .
In the remainder of this section we assume E to be a topological vector space whose
topology is dened by a distance d
E
, which is complete and translation invariant. We
dene another distance or quasi-distance in F by setting
d
F
(f, g) := d
E
(f, g) + (f g) . (13)
Write d

(f) instead of d

(f, 0), where could be either E or F. Then, d

(f g) =
d

(f, g) .
From (13) , a sequence that converges in (F, d
F
) also converges in (E, d
E
) and to the
same limit. The converse assertion is false in general. However, as we have already
pointed out, we shall prove a certain converse result. In order to establish it we need a
link between d
E
and .
Denition 3 The family of quasi-seminorms (., ) , I, dened above, is said to be
admissible with respect to the distance d
E
if the following conditions are satised:
i) (F, d
F
) is complete
ii) There exists a constant K > 0 and a function : I R
+
R
+
such that for
each I,
lim
t0
(, t) = (, 0) := 0
and for every f F,
(f) K (f, ) + (, d
E
(f)) . (14)
With respect to condition i), since F is a closed subspace of (F, ), it follows from (13)
that F is also a closed subspace of (F, d
F
). Then, if (F, d
F
) is complete, so is (F, d
F
) .
80
Theorem 4 (tauberian) Suppose that (F, d
F
) has been dened by a family of admissible
quasi-seminorms (., ) , I, on (E, d
E
). Let (f
n
) F be a convergent sequence in
(E, d
E
) , to an element f. If (f
n
) is 0-equicontinuous , then f F and (f
n
) converges to
f in (F, d
F
) . Moreover, if for each I, (, .) is continuous in R
+
, then
(f
n
f) 2C K ((f
n
) , ) + (, d
E
(f
n
f)) . (15)
Proof: Assume we have already proved that ( (f
n
)) is a real Cauchy sequence. Since
the hypothesis of the theorem include that (f
n
) is a Cauchy sequence in E, it would
follow from (13) that (f
n
) is a Cauchy sequence in (F, d
F
). But F is a complete metric
space, then there exits g F such that d
F
(f
n
g) 0 as n . Also by (13),
d
E
(f
n
g) d
F
(f
n
g) , then d
E
(f
n
g) 0. But d
E
(f
n
f) 0 as n . That
forces f = g. In order to prove that ( (f
n
)) is a Cauchy sequence, x > 0. For every
I, we use (14) to obtain,
(f
n
f
m
) K (f
n
f
m
, ) + (, d
E
(f
n
f
m
)) . (16)
Take such that ((f
n
) , ) . Further, take N such that for every n > N and
m > N, (, d
E
(f
n
f
m
)) . By substituting into (16),
(f
n
f
m
) (2CK + 1) .
The qualitative part of the theorem has been proved. In particular (f
n
f
m
)
(f
n
f) as m . Then, using (16) and the continuity of (, .) we deduce (15) .
Equivalent distances to (13) are given by
d
F
(f) := (d
E
(f)
p
+ (f)
p
)
1/p
, 1 < p < , (17)
d
F
(f) := max {d
E
(f) , (f) } , p = . (18)
In those cases, using (15) , we remark that
d
F
(f
n
f) (d
E
(f
n
f)
p
+ [2CK ((f
n
) , ) + (, d
E
(f
n
f))]
p
)
1/p
, (19)
if 1 p < ; or
d
F
(f
n
f) max { d
E
(f
n
f) , 2C K ((f
n
) , ) + (, d
E
(f
n
f))} , (20)
if p = .
Also we remark that formula (15) is a general one. Therefore its accuracy could be
improved in particular problems. In the same way, optimal values for depend on the
problem on hand.
81
Theorem 5 Suppose that (F, d
F
) has been dened from (E, d
E
) by a family of admissible
quasi-seminorms (., ) , I. Then a set A F is compact with respect to the topology
induced by d
F
if and only if A is compact in (E, d
E
) and 0-equicontinuous.
Proof. Let (f
n
) A. If A is a compact set of (E, d
E
) , there exists a subsequence
(f
n
k
) that converges to an element f A with respect to d
E
. If A is a 0-equicontinuous
set, then (f
n
k
) converges to f with respect to d
F
. Reciprocally, if A is a compact set of
(F, d
F
) , there exists a subsequence (f
n
k
) that converges to an element f A with respect
to d
F
Then (f
n
k
) converges to the same limit with respect to d
E
.
3 Examples and Applications
In this section we show that well known function spaces are included in the class of spaces
dened above . Of course, it is impossible to examine here the great variety of important
function spaces not even to examine only a few of them in their general setting (see Triebel
[9], for instance). Thus the particular examples below are conceived just to conform an
illustrative sample of applications.
Example 6 Set E := C (X). Taking (f, ) :=

(f, ), dened in (2) , we obtain F =

Lip

and F = lip

. Set K := 1. Thus , with (, t) := 2t/

, the family of seminorms

is admissible. An application of (15) leads to
f
n
f
,
(1+2/

) f
n
f

+ 2 ((f
n
) , ) (21)
The qualitative part of this application is the main theorem in Bustamante-Jimenez
[2]. In particular, the sequence of Bernstein polynomials (B
n
f) , f lip

([0, 1]) is
0-equicontinuous . In fact, Bustamante-Jimenez proved that (B
n
f) converges to f in
lip

p
[0, 1], i.e. in the norm (4) which implies convergence in the seminorm (1). Then
Proposition 2 asserts that (B
n
f) is 0-equicontinuous. On the other hand, theorem 5
characterizes the compact sets in lip

in the same way that it was done in [2].

Example 7 In the last example, take X := T and change (2) by
(f, ) : = sup { (f, t) : 0 < t } ,
(f, t) : = sup {|f (x + t) f (x)| /(t) : x T} ,
where : R

+
R

+
, is an increasing function. For f F, dene the sequence f
n
:= K
n

f, where K
n
L
1
(T) and M := sup {K
n

1
: n N} < . Then (f
n
) is 0-equicontinuous
with ((f
n
) , ) M (f, ) . Assume that f
n
f in uniform norm. Set K := 1 and
(, t) := 2t/() . In this situation (21) is transformed into
f
n
f
F
(1+2/()) f
n
f

+ 2 M (f, ) .
82
This is the estimate given by Leindler, Meir and Totik, from which several implications
to Fourier series follow ([8])
Example 8 We can use modulus of smoothness of higher order r. For instance, set
X := T and E := L
p
(T), 1 p < . Dene
(f, ) : = sup { (f, t) : 0 < t } ,
(f, t) : =
_
_
1
2
_
2
0
|
r
t
f (x)|
p
d (x)
_
1/p
/t

_
.
Here
t
f :=
1
t
f := f (. + t) f,
r
t
f :=
t
_

r1
t
f
_
. Set K := 1. Then, with the
function (, t) := 2
r
t/

, the family of semi-norms is admissible in denition 3.

Example 9 Set E := L
p
(T), 1 p < . In [6] , the author has dened homogeneous
H older spaces B

p
, > 0, which are equivalent in norm to certain Besov spaces. A function
f L
p
(T) is in B

p
, if F

(x, y) := (f (x) f (y)) /d (x, y)

L
p
(T
2
) . A crucial point
here is that d is given by (3) and then F

has period 2 in each variable. Set

(f) :=
_
1
4
2
_
2
0
_ _
2
0
|F

(x, y)|
p
dx
_
dy
_
1/p
.
Then B

p
becomes a homogeneous Banach (Hilbert if p = 2) space under the norm
f
,p
:=
_
f
p
L
p
(T)
+F

p
L
p
(T
2
)
_
1/p
.
Taking
(f, ) :=
_
1
2
2
_

0
_ _
2
0
|
t
f (x) /t

|
p
dx
_
dt
_
1/p
,
we can show that (f) = (f, ) .
Thus F = F = B

p
. Set K := 1. Therefore, with the function
(, t) :=
_
2

dx
x
p
_
1/p
t,
the family of seminorms is admissible.
With the following two examples, we show the connection of section 2 with the theory
of Measure and Integration and also the convenience of considering the general scope in
which the tauberian theorem above has been established.
Example 10 Let E be the complex linear space of all bounded complex functions f on R
that are continuous to the right and such that f (x) 0 as x . For all > 0, set
(22) (f, ) := sup
_

1im
|f (y
i
) f (x
i
)| : x
1
< y
1
x
2
< ... < y
m
;
m = 1, 2, ..;

1im
y
i
x
i

_
.
83
Thus (f) stands for the total variation of f in R; (F, ) is dened to be the Banach space
of functions of bounded variation and F is its closed subspace of absolutely continuous
functions.
We remark that for a given function f, it could happen that (f, ) 0 as 0,
but (f) = . For instance, f (x) := sin (x) /x. However such a function is not in F by
[10] and [11] .
On the other hand, since (22) is equal to
sup
__
A
|f

(x)| d (x) : meas (A) =
_
, f F,
this example is connected with the next one, for which the theoretical background can
be found in chapter 4 of [1]. However, to avoid technical diculties that are not any
objective at present, we restrict ourself to a set of nite measure.
Example 11 Let E be the complex linear space of all measurable complex functions f
on [0, 1] . We identify functions that are equal Lebesgue almost everywhere and consider
any complete and translation invariant distance d
E
which characterize the convergence in
measure. .
For any f E, 0 < p < and 0 < 1, dene
(f, ) := sup
_
__
A
|f|
p
d (x)
_
1/p
: meas (A) =
_
.
Then F = F = L
p
[0, 1] . A sequence (f
n
) is 0-equicontinuous if and only if it is equi-
integrable and it is known that convergence of (f
n
) in L
p
[0, 1], occurs if and only if (f
n
)
is a Cauchy sequence in measure and equi-integrable. In this example, the function
depends on the particular distance d
E
at hands. In fact, for a given function f F and
0 < 1, x a measurable set A, with meas(A) = , such that for any pair x A and
y A
c
, f (y) f (x) . Using typical procedures, we obtain
(f) =
__
1
0
|f|
p
d (x)
_
1/p
K
_
(f, ) +
__
A
c
|f|
p
d (x)
_
1/p
_
,
with K := C = 1 if 1 p < or K := C = 2
1/p
if 0 < p < 1. Then, in terms of the
sequence (f
n
) and its limit in measure f,
(f
n
f) 2K
2
((f
n
) , ) + K
n
(1 )
1/p
,
where the sequence
n
, that converges to 0 when n , can be expressed in terms of
d
E
(f
n
f).
84
References
[1] Bourbaki, N. ,

Elements de Mathematique, Livre VI Integration, Chapitres 1, 2, 3 et
4. Hermann, Paris (1975).
[2] Bustamante, J. ; Jimenez, M. A., Chebyshev and H older approximation, Aportaciones
Matem aticas, Serie Comunicaciones, 27 (2000), 23-31.
[3] Bustamante, J. ; Jimenez, M. A., Trends in H older approximation, Approximation,
Optimization and Mathematical Economics, Physica-Verlag (2001), 81-95.
[4] Butzer P. L. ; Berens, H., Semi-Groups of Operators and Approximation, Springer-
Verlag (1967).
[5] DeVore, R. A. ; Lorentz, G. G., Constructive Approximation, Grundlehren der math-
ematischen Wissenschaften 303, Springer-Verlag, (1993).
[6] Jimenez, M. A., A new approach to Lipschitz spaces of periodic integrable functions,
Aportaciones Matem aticas, Serie Comunicaciones, 25 (1999), 153-157.
[7] Jimenez, M. A.; Mart

inez, G., Equilipschitzian sets of Holder integrable functions,

Aportaciones Matem aticas, Serie Comunicaciones 29 (2001), 55-60.
[8] Leindler, L., Meir A., and Totik, V., On approximation of continuous functions in
Lipschitz norms, Acta Math. Hung., 45 (3-4) (1985), 441-443.
[9] Triebel, H. Theory of Function spaces, Birkha user Verlag, Basel, Vol. 1 (1983), Vol. 2
(1992).
85