Real Analysis Exchange

Vol. 26(2), 2000/2001 , pp. 669–686

Pavel Kostyrko and Tibor Šalát, Department of Algebra and Number

Theory, Comenius University, Mlynská dolina, SL-842 15 Bratislava,
Slovakia. e-mail: [email protected]
Wladyslaw Wilczyński, Department of Real Functions, Faculty of
Mathematics, University of Lódź, Banacha 22, PL-90-238 Lódź, Poland.
e-mail: [email protected]

I-CONVERGENCE∗

Abstract
In this paper we introduce and study the concept of I-convergence
of sequences in metric spaces, where I is an ideal of subsets of the
set N of positive integers. We extend this concept to I-convergence of
sequence of real functions defined on a metric space and prove some
basic properties of these concepts.

1 Introduction
This paper was inspired by [14], where the concept of I-convergence of se-
quences of real numbers is introduced. We will often quote some results from
[14] that can be easily transferred to sequences of points in a metric space. In
[14] it is shown that our I-convergence is, in a sense, equivalent to µ-statistical
convergence of J. Connor ([8]).
The concept of statistical convergence is introduced in [10] and [29] and
developed in [6], [7], [8], [9], [10], [12], [26] and [28]. Some applications of
statistical convergence in number theory and mathematical analysis can be
found in [4], [5] and [19]. The concept of I-convergence is a generalization of
statistical convergence and it is based on the notion of the ideal I of subsets
of the set N of positive integers.
This paper consists of five sections with the new results in sections 3–5.
In the Section 3 the concept of the I-convergence of sequences in a metric
space is introduced and its fundamental properties are studied. In Section
4 we introduce and study the concept of an I-cluster point and an I-limit
Key Words: statistical convergence, ideals of sets, Baire classification of functions
Mathematical Reviews subject classification: Primary 40A30; Secondary 40A99 40C15
Received by the editors July 28, 2000
∗ The work on this paper was partially supported by GRANT VEGA 1/7173/20.

669
670 Pavel Kostyrko, Tibor Šalát and Wladyslaw Wilczyński

point of a sequence in a metric space. In Section 5 we extend the concept

of I-convergence to sequences of real functions defined on a metric space and
we will discuss some questions concerning the limit functions of I-convergent
sequences, specially in the case when all functions in the sequence are contin-
uous.

2 Definitions and Notation

Throughout the paper N will denote the set of positive integers χA – the
characteristic function of A ⊂ N, R – the set of all real numbers, (X, ρ) –
a metric space, B(ξ, ) – an open ball in X with center ξ ∈ X and radius
 > 0. The topological terminology in taken from [20]. We recall the concept
of asymptotic and logarithmic density of a set A ⊂ N (see [22], pp. 71, 95-96
and [27]).
Pn Pn χ
Let A ⊂ N. Put dn (A) = n1 k=1 χA (k) and δn (A) = s1n k=1 Ak(k)
Pn 1
for n ∈ N, where sn = k=1 k . The numbers d(A) = lim inf n→∞ dn (A)
and d(A) = lim supn→∞ dn (A) are called the lower and upper asymptotic
density of A, respectively. Similarly, the numbers δ(A) = lim inf n→∞ δn (A)
and δ(A) = lim supn→∞ δn (A) are called the lower and upper logarithmic
density of A, respectively. If d(A) = d(A) (δ(A) = δ(A)), then d(A) = d(A)
is called the asymptotic density of A (δ(A) = δ(A) is called the logarithmic
density of A, respectively). It is well known that for each A ⊂ N d(A) ≤
δ(A) ≤ δ(A) ≤ d(A) (see [22], 70-75, 95-96). Hence if d(A) exists, then also
δ(A) exists and d(A) = δ(A). The converse is not true. Obviously all numbers
d(A), d(A), δ(A), δ(A) (and so d(A), δ(A)) belong to [0, 1]. Since
n
X 1 1
= ln n + γ + O( ) (1)
k n
k=1
Pn χ
where γ is an Euler constant, if we put δn∗ (A) = ln1n k=1 Ak(k) for n ∈ N,
then δ(A) = lim inf n→∞ δn∗ (A), δ(A) = lim supn→∞ δn∗ (A) and if δ(A) exists,
then δ(A) = limn→∞ δn∗ (A).
Now recall the concept of statistical convergence of real sequences (see [9],
[28]).
Definition A. A sequence x = {xn }n∈N of real numbers is said to be statis-
tically convergent to ξ ∈ R provided that for each  > 0 we have d(A()) = 0,
where A() = {n ∈ N : |xn − ξ| ≥ }.
Recall that if X is a non-empty set then a family of sets I ⊂ 2X is an
ideal if and only if for each A, B ∈ I we have A ∪ B ∈ I and for each A ∈ I
I-Convergence 671

and each B ⊂ A we have B ∈ I. A non-empty family of sets F ⊂ 2X is

a filter on X if and only if φ ∈ / F, for each A, B ∈ F we have A ∩ B ∈ F
and for each A ∈ F and each B ⊃ A we have B ∈ F. An ideal I is called
non-trivial if I 6= ∅ and X ∈/ I. I ⊂ 2X is a non-trivial ideal if and only if
F = F(I) = {X \ A : A ∈ I} is a filter on X. A non-trivial ideal I ⊂ 2X is
called admissible if and only if I ⊃ {{x} : x ∈ X}.

3 I-Convergence of Sequences of Elements of a Metric

Space
3.1 I-Convergence, Examples and Properties
In what follows (X, ρ) is a fixed metric space and I denotes a non-trivial ideal
of subsets of N.

Definition 3.1. A sequence {xn }n∈N of elements of X is said to be I-converg-

ent to ξ ∈ X (ξ = I-limn→∞ xn ) if and only if for each  > 0 the set A() =
{n ∈ N : ρ(xn , ξ) ≥ } belongs to I. The element ξ is called the I-limit of the
sequence x = {xn }n∈N .

Example 3.1. (a) Take for I the class If of all finite subsets of N. Then If
is a non-trivial admissible ideal and If -convergence coincides with the usual
convergence with respect to the metric ρ in X.
(b) Denote by Id (Iδ ) the class of all A ⊂ N with d(A) = 0 (δ(A) = 0, re-
spectively). Then Id and Iδ are non-trivial admissible ideals, Id -convergence
coincides with the statistical convergence. Iδ -convergence is said to be loga-
rithmic statistical convergence.
(c) The uniform density of a set A ⊂ N is defined as follows.For integers
t ≥ 0 and s ≥ 1 let A(t + 1, t + s) = card{n ∈ A : t + 1 ≤ n ≤ t + s}. Put
βs = lim inf t→∞ A(t+1, t+s), β s = lim supt→∞ A(t+1, t+s). It can be shown
s
(see [3]) that the following limits exist: u(A) = lims→∞ βss , u(A) = lims→∞ βs .
If u(A) = u(A), then u(A) = u(A) is called the uniform density of the set A.
Put Iu = {A ⊂ N : u(A) = 0}. Then Iu is a non-trivial ideal and Iu -
convergence is said to be the uniform statistical convergence.
(d) A wide class of I-convergences can be obtained as follows. Let T =
{tn,k }n,k∈N be a regular non negative matrix (see [23], p. 8). For A ⊂ N we
(n) P∞ (n)
put dT (A) = k=1 tn,k · χA (k) for n ∈ N. If limn→∞ dT (A) = dT (A) exists,
then dT (A) is called P a T -density of A (see [18]). From the regularity of T it
∞
follows that limn→∞ k=1 tn,k = 1 and from this we see that dT (A) ∈ [0, 1]
(if it exists). Put IdT = {A ⊂ N : dT (A) = 0}. Then IdT is a non-trivial
ideal and IdT contains both Id - and Iδ -convergence as special cases. Indeed,
672 Pavel Kostyrko, Tibor Šalát and Wladyslaw Wilczyński

1
Id -convergence can be obtained by choosing tn,k = n for k ≤ n, tn,k = 0
1
for k > n and Iδ -convergence by choosing tn,k = sn for k ≤ n, tn,k = 0 for
k

Pn
k > 0, where sn = j=1 1j for n ∈ N. Choosing tn,k = φ(k) n for k ≤ n, k|n
and tn,k = 0 for k ≤ n k - n and tn,k = 0 for k > n we get φ-convergence of
Schoenberg (see [29]), where φ is the Euler function.
P∞case of IdT -convergence is the following. Take an arbitrary
Another special
divergent series n=1 cn , where cn > 0 for n ∈ N and put tn,k = scnk for k ≤ n,
Pn
where sn = j=1 cj , and tn,k = 0 for k > n (see [1]).
(e) Let ν be an arbitrary finitely additive normed measure defined on a
field U ⊂ 2N . Suppose that U contains all singletons {n}, n ∈ N. Then
Iν = {A ⊂ N : ν(A) = 0} is a non-trivial ideal in N which generates the
Iν -convergence.
(f) Suppose that µm : 2N → [0, 1] is a finitely additive normed measure
for m ∈ N. If for A ⊂ N there exists µ(A) = limm→∞ µm (A), then the set
A is said to be measurable and µ(A) is called the measure of A. Obviously
µ is a finitely additive measure on some field S ⊂ 2N (see [16]). The family
Iµ = {A ⊂ N : µ(A) = 0} is a non-trivial ideal which generates the Iµ -
convergence.
For µm we canS∞take dm or δm (compare the definition).
(g) Let N = j=1 ∆j be a decomposition of N such that each ∆j is infinite
and obviously ∆i ∩ ∆j = ∅ for i 6= j. Denote by E the class of all A ⊂ N that
intersect only a finite number of ∆0j s. Then E is a non-trivial ideal.
(h) Put Ic = {AP⊂ N : a∈A a−1 < ∞} (see [23]). Then Ic is a non-
P
trivial ideal. Since a∈A a−1 < ∞ implies d(A) = 0 (see [24]), we see that
Ic -convergence implies statistical convergence.
Remark 3.1. Note that if I is an admissible ideal, then the usual convergence
in X implies I-convergence in X.
We shall now investigate which axioms of convergence are satisfied by I-
convergence. The following properties are the most familiar axioms of conver-
gence (see [17]):
(S) Every constant sequence {ξ, ξ, . . . , ξ, . . . } converges to ξ.
(H) The limit of any convergent sequence is uniquely determined.
(F ) If a sequence x = {xn }n∈N has the limit ξ, then each of its subsequences
has the same limit.
(U ) If each subsequence of the sequence x = {xn }n∈N has a subsequence
which converges to ξ, then x converges to ξ.
 I-Convergence 673

Proposition 3.1. Suppose that X has at least two points. Let I ⊂ 2X be an

admissible ideal.
(i) The I-convergence satisfies (S), (H) and (U ).
(ii) If I contains an infinite set, then I-convergence does not satisfy (F ).

Proof. (i) (S) is obviously fulfilled. To prove (H) it is sufficient to observe

that for any A1 , A2 ∈ I we have (N \ A1 ) ∩ (N \ A2 ) 6= ∅ since the last two sets
belong to the filter associated with I. If there are two limits ξ, η ∈ X, ξ 6= η,
choose  such that
1
0 <  < ρ(ξ, η) (2)
2
and put A1 = {n ∈ N : ρ(xn , ξ) ≥ }, A2 = {n ∈ N : ρ(xn , η) ≥ }.
Suppose now that (U ) does not hold. Then there exists 0 > 0 such that

A(0 ) = {n ∈ N : ρ(xn , ξ) ≥ 0 } ∈
/ I.

But then A(0 ) is an infinite set since I is admissible. Let A(0 ) = {n1 < n2 <
· · · < nk < . . . }. Put yk = xnk for k ∈ N. Then y = {yk }k∈N is a subsequence
of x without a subsequence I-convergent to ξ.
(ii) Suppose that A ∈ I is an infinite set, A = {n1 < n2 < · · · < nk < . . . }.
B = N \ A = {m1 < m2 < · · · < mk < . . . }. The set B is also infinite since
I is non-trivial ideal. Define x = {xn }n∈N by choosing ξ1 , ξ2 ∈ X, ξ1 6= ξ2
and put xnk = ξ1 , xmk = ξ2 for k ∈ N. Obviously I-limk→∞ xk = ξ2 , but the
subsequence yk = xnk , k ∈ N, I-converges to ξ1 .

Remark 3.2. If I is an admissible ideal which does not contains any infinite
set, then I-convergence coincides with the usual convergence and obviously
fulfills (F ).

3.2 I-Convergence and I ∗ -Convergence

The following result is well known in the theory of statistical convergence. A
sequence {xn }n∈N of real numbers is statistically convergent to ξ if and only
if there exists a set M = {m1 < m2 < . . . mk < . . . } ⊂ N such that d(M ) = 1
and limk→∞ xmk = ξ (see [10], [11], [18], [26]).
This result suggests the introduction of the following concept of conver-
gence (which we shall call I ∗ -convergence) closely related to I-convergence.

Definition 3.2. A sequence x = {xn }n∈N of elements of X is said to be I ∗ -

convergent to ξ ∈ X if and only if there exists a set M ∈ F(I) (i.e. N\M ∈ I),
M = {m1 < m2 < · · · < mk < . . . } such that limk→∞ ρ(xmk , ξ) = 0.
 674 Pavel Kostyrko, Tibor Šalát and Wladyslaw Wilczyński

Proposition 3.2. Let I be an admissible ideal. If I ∗ -limn→∞ xn = ξ, then

I-limn→∞ xn = ξ.

Proof. By assumption there exists a set H ∈ I such that for M = N \ H =

{m1 < m2 < · · · < mk < . . . } we have

lim xmk = ξ. (3)

k→∞

Let  > 0. By virtue of (3) there exists k0 ∈ N such that ρ(xmk , ξ) <  for
each k > k0 . Then obviously

A() = {n ∈ N : ρ(xn , ξ) ≥ } ⊂ H ∪ {m1 < m2 < · · · < mk0 }. (4)

The set on the right-hand side of (4) belongs to I (since I is admissible). So

A() ∈ I.
The converse implication between I- and I ∗ -convergence depends essen-
tially on the structure of the metric space (X, ρ).

Theorem 3.1. Let (X, ρ) be a metric space.

(i) If X has no accumulation point, then I- and I ∗ -convergence coincide for
each admissible ideal I ⊂ 2N .
(ii) If X has an accumulation point ξ, then there exists an admissible ideal
I ⊂ 2N and a sequence {yn }n∈N of elements of X such that I-limn→∞ yn = ξ
but I ∗ -lim yn does not exist.

Proof. (i) Let ξ ∈ X and I-limn→∞ xn = ξ. By virtue of Proposition 1.2. it

suffices to prove that I ∗ -limn→∞ xn = ξ. Since X has no accumulation points,
there exists δ > 0 such that B(ξ, δ) = {x ∈ X : ρ(x, ξ) < δ} = {ξ}. From the
assumption we have {n ∈ N : ρ(xn , ξ) ≥ δ} ∈ I. Hence

{n ∈ N : ρ(xn , ξ) < δ} = {n ∈ N : xn = ξ} ∈ F(I)

and obviously I ∗ -limn→∞ xn = ξ.

(ii) Since ξ is an accumulation point of X, there exists a sequence {xn }n∈N
of elements of X such that ξ = limn→∞ xn and the sequence {ρ(xn , ξ)}n∈N is
decreasing to 0. Put n = ρ(xn , ξ) for n ∈ N. For I we take the ideal E from
Example 1.1. (g).
Define the sequence {yn }n∈N by yn = xj if n ∈ ∆j . Let η > 0. Choose
ν ∈ N such that ν < η. Then A(η) = {n : ρ(yn , ξ) ≥ η} ⊂ ∆1 ∪ · · · ∪ ∆ν .
Hence A(η) ∈ E and E-limn→∞ yn = ξ.
 I-Convergence 675

Suppose that E ∗ -limn→∞ yn = ξ. Then there exists a set H ∈ E such that

for M = N\H = {m1 < m2 < · · · < mk < . . . } we have limk→∞ ρ(ymk , ξ) = 0.
By definition of E there exists l ∈ N such that H ⊂ ∆1 ∪ · · · ∪ ∆l . But then
∆l+1 ⊂ N \ H = M , so for infinitely many k’s (each ∆i is an infinite set) we
have ρ(ymk , ξ) = l+1 > 0, which contradicts ymk → ξ. Also the assumption
E ∗ -limn→∞ yn = y for y 6= ξ leads to the contradiction.
Now we shall formulate a necessary and sufficient condition (for the ideal
I) under which I- and I ∗ -convergence are equivalent. This condition (AP ) is
similar to the condition (AP O) used in [7] and [10].
Definition 3.3. An admissible ideal I ⊂ 2N is said to satisfy the condition
(AP ) if for every countable family of mutually disjoint sets {A1 , A2 , . . . } be-
longing to I there exists a countable family
S∞ of sets {B1 , B2 , . . . } such that
Aj 4Bj is a finite set for j ∈ N and B = j=1 Bj ∈ I.
Remark 3.3. Observe that also Bj ∈ I for j ∈ N.
Theorem 3.2. Let I ⊂ 2N be an admissible ideal.
(i) If the ideal I has property (AP ) and (X, ρ) is an arbitrary metric space, then
for arbitrary sequence {xn }n∈N of elements of X I-limn→∞ xn = ξ implies I ∗ -
limn→∞ xn = ξ.
(ii) If (X, ρ) has at least one accumulation point and for arbitrary sequence
{xn }n∈N of elements of X and for each ξ ∈ X I-limn→∞ xn = ξ implies
I ∗ -limn→∞ xn = ξ, then I has property (AP ).
Proof. (i) Suppose that I satisfies condition (AP ). Let I-limn→∞ xn = ξ.
Then A() = {n ∈ N : ρ(xn , ξ) ≥ } ∈ I for  > 0. Put A1 = {n ∈ N :
ρ(xn , ξ) ≥ 1} and An = {n ∈ N : n1 ≤ ρ(xn , ξ) < n−1 1
} for n ≥ 2, n ∈ N.
Obviously Ai ∩ Aj = ∅ for i 6= j. By condition (AP ) there exists S
a sequence of
∞
sets {Bn }n∈N such that Aj 4Bj are finite sets for j ∈ N and B = j=1 Bj ∈ I.
It is sufficient to prove that for M = N \ B we have

lim xn = ξ.
n→∞
(5)
n∈M

1
Let η > 0. Choose k ∈ N such that k+1 < η. Then {n ∈ N : ρ(xn , ξ) ≥ η} ⊂
Sk+1
j=1 Aj . Since Aj 4Bj , j = 1, 2, . . . , k + 1 are finite sets, there exists n0 ∈ N
such that
 k+1
[   k+1
[

Bj ∩ {n ∈ N : n > n0 } = Aj ∩ {n ∈ N : n > n0 }. (6)
j=1 j=1
676 Pavel Kostyrko, Tibor Šalát and Wladyslaw Wilczyński

Sk+1 Sk+1
If n > n0 and n ∈ / B, then n ∈ / j=1 Bj and, by (6), n ∈ / j=1 Aj . But then
1
ρ(xn , ξ) < n+1 < η; so (5) holds.
(ii) Suppose that ξ ∈ X is an accumulation point of X. There exists a
sequence {xn }n∈N of elements of X such that ξ = limn→∞ xn and the sequence
{ρ(xn , ξ)}n∈N is decreasing to 0. For n ∈ N let n = ρ(xn , ξ). Let {An }n∈N
be a disjoint family of non-empty sets from I. Define a sequence {yn }n∈N
by yn = xj if n ∈ Aj . Let η > 0. Choose m ∈ N such that m < η. Then
A(η) = {n ∈ N : ρ(yn , ξ) ≥ η} ⊂ A1 ∪ · · · ∪ Am . Hence A(η) ∈ I and I-
limn→∞ yn = ξ. By virtue of our assumption we have also I ∗ -limn→∞ yn = ξ.
Hence there exists a set B ∈ I such that if M = N \ B = {m1 < m2 < . . . },
then
lim ymk = ξ. (7)
k→∞
S∞
Put Bj = Aj ∩ B for j ∈ N. Then Bj ∈ I for each n. Further j=1 Bj =
S∞ S∞
B ∩ j=1 Aj ⊂ B. Hence j=1 Bj ∈ I. Fix j ∈ N. From (7) it is clear
that Aj has only a finite numbers of elements common with the set M . Thus
there exists k0 ∈ N such that Aj ⊂ (Aj ∩ B) ∪ {m1 , m2 , . . . , mk0 }. Hence
Aj 4Bj = Aj \ Bj ⊂ {m1 , m2 , . . . , mk0 }; so Aj 4Bj is a finite set. From the
arbitrariness of j ∈ N it follows that I has property (AP ).
In [18] it is proved that IdT - and Id∗T -convergence are equivalent P
(in R) pro-
n
vided that T = {tn,k }n,k∈N is a non-negative triangular matrix with k=1 tnk =
1 for n ∈ N. From this we get that Id -, Iδ -convergence (Example 1.1. (b))
and Iφ -convergence (Example 1.1. (d)) coincide, respectively, with Id∗ -, Iδ∗ -
and Iφ∗ -convergence.

3.3 Functions Preserving I-Convergence

Let (X, ρ) be a metric space and I ⊂ 2N -an admissible ideal. As in [2] we say
that a function g : X → X preserves I-convergence in X if I-limn→∞ xn = ξ
implies I-limn→∞ g(xn ) = g(ξ) for each sequence {xn }n∈N of elements of X
and each ξ ∈ X. As is not difficult to predict, we have the following.
Proposition 3.3. A function g : X → X preserves I-convergence in X (for
an arbitrary admissible ideal I) if and only if g is continuous on X.
Proof. 1. Let I-limn→∞ xn = ξ. If g is continuous, then for each η > 0
there exists δ > 0 such that if x ∈ B(ξ, δ), then g(x) ∈ B(g(ξ), η). But then
we have

C(δ) = {n ∈ N : ρ(xn , ξ) < δ} ⊂ {n ∈ N : ρ(g(xn ), g(ξ)) < η} = D(η)

and D(η) ∈ F(I), since C(δ) ∈ F(I). Hence I-limn→∞ g(xn ) = g(ξ).
I-Convergence 677

2. If g is not continuous at some ξ ∈ X, then there exists a number

η > 0 and a sequence {xn }n∈N of elements of X such that limn→∞ xn = ξ and
ρ(g(xn ), g(ξ)) ≥ η for n ∈ N. Hence g does not preserve I-convergence for any
ideal I.

3.4 Relationship between Id - and Iδ -Convergence and Cesaro Summ-

ability
Recall that a sequence {xn }n∈N of real numbers is said toPbe (C, 1)-summable
n
(or (C, 1)-convergent) to ξ ∈ R if and only if limn→∞ n1 i=1 xi = ξ (see [23],
p. 3) (abreviated (C, 1)-limn→∞ xn = ξ)Pand is said to be strongly (C, 1)-
n
summable to ξ if and only if limn→∞ n1 i=1 |xi − ξ| = 0 (see [23], p. 5,
[7]).
If {xn }n∈N ∈ `∞ , then Id -limn→∞ xn = ξ implies (C, 1)-limn→∞ xn = ξ
(see [7], [10], [26]). The converse is obviously not true (e.g. {0, 1, 0, 1, . . . }).
However, in `∞ the Id -convergence to some number is equivalent to strong
Cesaro summability to the same number. For Iδ -convergence the situation is
different.

Proposition 3.4. In `∞ there is a sequence {xn }n∈N such that Iδ -limn→∞ xn =

0 and (C, 1)-limn→∞ xn does not exist.
S∞ 2 2 2
Proof. Put A = k=2 Ak , where Ak = {k k +1, k k +2, . . . , k k +1 } for k ∈ N,
k ≥ 2. If A(n) = dn (A) for n ∈ N (compare section 2), then
2 2 2
A(k k +1 ) k k +1 − k k
d(A) ≥ lim sup 2 ≥ lim sup = 1.
k→∞ k k +1 k→∞ k k2 +1

Hence
d(A) = 1. (8)
1 1
P
Simultaneously by (1) we have j∈An j = ln k + O( kk2 ) for k ∈ N, k ≥ 2.
From this by a simple calculation we get
Pn
k=1 ln k + O(1) n ln n + O(1)
δ(A) ≤ lim Pnn2 +1 1 ≤ lim = 0.
n→∞ n→∞ (n2 + 1) ln n + O(1)
j=1 j

So we have δ(A) = 0 and consequently

d(A) = 0. (9)

So d(A) does not exist.

 678 Pavel Kostyrko, Tibor Šalát and Wladyslaw Wilczyński

Define x = {xn }n∈N by

(
0 if n ∈ N \ A
xn =
1 if n ∈ A.

Since δ(A) = 0 we have Iδ -limn→∞ xn = 0. But (C, 1)-limn→∞ xn does not

Pn
exist because n1 i=1 xi = A(n)
n for n ∈ N (compare (8), (9)).

4 I-Limit Points and I-Cluster Points

Recall that a number ξ ∈ R is said to be a statistical limit point of a sequence
{xn }n∈N of real numbers provided that there exists a set M = {m1 < m2 <
. . . } ⊂ N such that d(M ) > 0 and limk→∞ xmk = ξ. A number ξ ∈ R is said
to be a statistical cluster point of x = {xn }n∈N provided that for each  > 0
we have d{n ∈ N : |xn − ξ| < } > 0 (see [9], [12], [13]).
We can extend these concepts to I-convergence in the following way.

Definition 4.1. Let (X, ρ) be a metric space, x = {xn }n∈N a sequence of

elements of X.
a) An element ξ ∈ X is said to be an I-limit point of x provided that there is
a set M = {m1 < m2 < . . . } ⊂ N such that M ∈/ I and limk→∞ xmk = ξ.
b) An element ξ ∈ X is said to be an I-cluster point of x if and only if for each
 > 0 we have {n ∈ N : ρ(xn , ξ) < } ∈
/ I.

Denote by I(Λx ) and I(Γx ) the set of all I-limit and I-cluster points of
x, respectively.

Proposition 4.1. Let I be an admissible ideal. Then for each sequence x =

{xn }n∈N of elements of X we have I(Λx ) ⊂ I(Γx ).

Proof. Let ξ ∈ I(Λx ). Then there exists a set M = {m1 < m2 < . . . } ∈
/I
such that
lim ρ(xmk , ξ) = 0. (10)
k→∞

Take δ > 0. According to (10) there exists k0 ∈ N such that for k > k0 we
have ρ(xmk , ξ) < δ. Hence {n ∈ N : ρ(xn , ξ) < δ} ⊃ M \ {m1 , . . . , mk0 } and
so {n ∈ N : ρ(xn , ξ) < δ} ∈
/ I, which means that ξ ∈ I(Γx ).

Theorem 4.1. Let I be an admissible ideal.

(i) The set I(Γx ) is closed in X for each sequence x = {xn }n∈N of elements of
X.
 I-Convergence 679

(ii) Suppose that (X, ρ) is a separable metric space. Suppose that there exists a
disjoint sequence of sets {Mn }n∈N such that Mn ⊂ N and Mn ∈ / I for n ∈ N.
Then for each closed set F ⊂ X there exists a sequence x = {xn }n∈N of
elements of X such that F = I(Γx ).

Proof. (i) Let y ∈ I(Γx ). Take  > 0. There exists ξ0 ∈ I(Γx ) ∩ B(y, ).
Choose δ > 0 such that B(ξ0 , δ) ⊂ B(y, ). We obviously have

{n ∈ N : ρ(y, xn ) < } ⊃ {n ∈ N : ρ(ξ0 , xn ) < δ}.

Hence {n ∈ N : ρ(y, xn ) < } ∈ / I and y ∈ I(Γx ).

(ii) Let A = {a1 , a2 , . . . } ⊂ F be a countable set dense in F . For n ∈ Mi we
put xn = ai . Obviously we have I(Γx ) ⊂ F . To prove the converse inclusion
take z ∈ F and  > 0. There exists i0 ∈ N such that ρ(ai0 , z) < . Since
xn = ai0 for each n ∈ Mi0 , we obtain {n ∈ N : ρ(xn , z) < } ⊃ Mi0 ; so
{n ∈ N : ρ(xn , z) < } ∈/ I and z ∈ I(Γx ).

In [13] the following result has been established for sequences of real num-
bers.

Theorem A.
(i) For each sequence x = {xn }n∈N of real numbers the set Id (Λx ) is of type
Fσ .
(ii) If F ⊂ R is a set of type Fσ , then there exists a sequence x = {xn }n∈N of
real numbers such that F = Id (Λx ).

A detailed analysis of the proof of Theorem 4 in [13] shows that in this

theorem Id can be replaced by Iδ . It would be desirable to extend Theorem
4 for more general I-convergence.
It is not difficult to observe that I-convergence cannot in general be metriz-
able.

Proposition 4.2. Suppose that X has at least two elements and I ⊂ 2N is

an admissible ideal containing an infinite set M ⊂ N. Then I-convergence
cannot be metrizable.

Proof. Suppose that there exists a metric σ on X such that I-limn→∞ xn = ξ

if and only if limn→∞ σ(xn , ξ) = 0. Take ξ1 , ξ2 ∈ X, ξ1 6= ξ2 and put xn = ξ1
if n ∈ M , xn = ξ2 if n ∈ / M . Obviously we have I-limn→∞ xn = ξ2 and
limn→∞ σ(xn , ξ2 ) = 0, which implies σ(ξ1 , ξ2 ) = 0 since N \ M is an infinite
set. This contradicts ξ1 6= ξ2 .
 680 Pavel Kostyrko, Tibor Šalát and Wladyslaw Wilczyński

In connection with the above mentioned results about Id (Λx ) (compare

[14]) we can conjecture that (at least under some reasonable conditions) the
set I(Λx ) is of type Fσ . To show that the separability of (X, ρ) is essential,
we prove the following.

Theorem 4.2. Suppose that (X, ρ) is not separable and I ⊂ 2N is an admis-

sible ideal.
(i) There exists a closed set F ⊂ X such that for each sequence x = {xn }n∈N
of elements of X we have I(Γx ) 6= F.
(ii) There exists an open set G ⊂ X (so also G is of type Fσ ) such that for
each sequence x = {xn }n∈N of elements of X we have I(Λx ) 6= G.

In the proof we shall need the following lemma (see [21]).

Lemma 4.1. Suppose that (X, ρ) is not separable. Then there exists 1 > 0
and an uncountable set D = {d0 , d1 , . . . , dα , . . . } ⊂ X, α < Ω such that for
α, β < Ω, α 6= β we have ρ(dα , dβ ) ≥ 1 .

Proof of Theorem 4.2 (i) Let D be the (closed) set from Lemma 4.1. We
shall prove that D 6= I(Γx ) for each sequence x = {xn }n∈N of elements of X.
Consider the uncountable family {B(dα , 21 ) : α < Ω} of disjoint balls. There
S that {n ∈ N
exists α0 < Ω such : xn ∈ B(dα0 , 21 )} = ∅. Hence dα0 ∈
/ I(Γx ).
(ii) Put G = α<Ω B(dα , 2 ). The proof is similar to that of part (i).
1

5 I-Convergence of Sequences of Functions

In a natural manner we can extend the notion of I-convergence of sequences
in X to I-convergence of sequences of functions.

Definition 5.1. Let X be a non-empty set and let (Y, τ ) be a metric space.
Let I ⊂ 2N be an admissible ideal. The sequence of functions {fn }n∈N trans-
forming X into Y is said to I-converge to a function f : X → Y provided that
for each x ∈ X we have I-limn→∞ fn (x) = f (x).

The function f is called the I-limit function of the sequence {fn }n∈N and
we write f = I-limn→∞ fn .

Remark 5.1. (a) From Proposition 3.1 it follows that the I-limit function is
uniquely determined.
(b) From Proposition 4.2 it follows that the I-convergence of sequences of
functions is not metrizable.
I-Convergence 681

If we additionally assume that (X, ρ) is a metric space and (Y, τ ) is R

equipped with the natural metric, the question arises whether the I-limit
function of a sequence of continuous functions belongs to B1 (Baire class one)
and a similar question arises for higher Baire classes. Taking into account
that Id -convergence of bounded real sequences implies (C, 1)-summability one
can answer positively for questions concerning all Baire classes. However, for
ideals different from Id the situation is a little bit more complicated. Under
some conditions (concerning the ideal I) we are able to give the answer to the
question concerning continuous functions.
We shall suppose that µ is a finitely additive, normed measure defined on
some class of subsets of N (as in Example 3.1 (f)), µ(A) = limm→∞ µm (A),
where µm : 2N → [0, 1] is a finitely additive measure on 2N . Further, we shall
assume that each µm satisfies the following condition.
For each m ∈ N and for each A ⊂ N µm (A)
(11)
= lim µm (A ∩ {n ∈ N : n ≤ k}).
k→∞

Theorem 5.1. Suppose that (X, ρ) is a complete metric space and fn : X → R

are continuous functions for n ∈ N. If Iµ -lim fn = f , then f ∈ B1 .
Proof. Suppose that f ∈ / B1 . Then from [25] (see also [15], p. 39), we can
conclude that there exists a perfect set F ⊂ X and two numbers a, b, a < b
such that each of the sets T1 = {x ∈ F : f (x) < a}, T2 = {x ∈ F : f (x) > b}
is dense in F. Choose x1 ∈ T1 ∩ F . Then Iµ -limn→∞ fn (x1 ) = f (x1 ) < a.
Put 1 = 21 (a − f (x1 )) and A(1 ) = {n ∈ N : fn (x1 ) < f (x1 ) + 1 } = {n ∈
N : fn (x1 ) < 12 (a + f (x1 )}. Then µ(A(1 )) = 1; so limm→∞ µm (A(1 )) = 1.
Therefore there exists m1 ∈ N such that µm1 (A(1 )) > 12 . By (11) we conclude
that there exists k1 ∈ N such that µm1 (A(1 ) ∩ {n ∈ N : n ≤ k1 }) > 21 .
Since all functions fj for j ≤ k1 , j ∈ A(1 ) are continuous, there exists
δ1 > 0 such that for each x ∈ B(x1 , δ1 ) and each n ≤ k1 , n ∈ A(1 ) we have
fn (x) ≤ a and so
1
µm1 ({n ∈ N : n ≤ k1 , n ∈ A(1 ) and fn (x) ≤ a for each x ∈ B(x1 , δ1 )}) > .
2
Choose arbitrary x2 ∈ T2 ∩F ∩B(x1 , δ1 ). Then Iµ -limn→∞ fn (x2 ) = f (x2 ) > b.
Take 2 = 21 (f (x2 ) − b) and put
1
A(2 ) = {n ∈ N : fn (x2 ) > f (x2 ) − 2 } = {n ∈ N : fn (x2 ) > (f (x2 ) + b}.
2
We have µ(A(2 )) = 1, so as before we can find m2 ∈ N and k2 ∈ N (moreover,
k2 > k1 ) such that µm2 (A(2 ) ∩ {n ∈ N : n ≤ k2 }) > 21 . Again from the
682 Pavel Kostyrko, Tibor Šalát and Wladyslaw Wilczyński

continuity of all functions fj , j ≤ k2 , j ∈ A(2 ) it follows that there exists

δ2 > 0 such that δ2 < 12 δ1 , B(x2 , δ2 ) ⊂ B(x1 , δ1 ) and for each x ∈ B(x2 , δ2 )
and each n ≤ k1 , n ∈ A(2 ) we have fn (x) ≥ b and so

1
µm2 ({n ∈ N : n ≤ k2 , n ∈ A(2 ) and fn (x) ≥ b for each x ∈ B(x2 , δ2 )}) > .
2
In this way (by induction) we construct a sequence {xn }n∈N of points of F ,
a sequence {δn }n∈N decreasing to 0, a sequence {n }n∈N of positive numbers,
a sequence A(n ) of subsets of N and a descending sequence {B(xn , δn )}n∈N
of closed sets with diameters tending to 0. Simultaneously we obtain two
increasing sequences of natural numbers {mi }i∈N and {ki }i∈N such that

µm2i−1 ({n ∈ N :n ≤ k2i−1 , n ∈ A(2i−1 ) and fn (x) ≤ a

1 (12)
for each x ∈ B(x2i−1 , δ2i−1 )}) >
2

µm2i ({n ∈ N :n ≤ k2i , n ∈ A(2i )and fn (x) ≥ b

1 (13)
for each x ∈ B(x2i , δ2i )}) >
2
T∞
Let x0 ∈ k=1 B(xk , δk ). By the monotonicity of µm , m ∈ N from (12) and
(13) we obtain
1
µm2i−1 ({n ∈ N : fn (x0 ) ≤ a}) > (14)
2
1
µm2i ({n ∈ N : fn (x0 ) ≥ b}) > (15)
2
for i ∈ N. Suppose that Iµ -limn→∞ fn (x0 ) = f (x0 ). If f (x0 ) ≤ a, we obtain a
contradiction to (15), if f (x0 ) ≥ b; a contradiction to (14), if a < f (x0 ) < b; a
contradiction to both (14) and (15).

Remark 5.2. According to Theorem 5.1 I-limit function of a sequence of

continuous real functions belongs to Baire class one if I = Id , Iδ , IdT , Iφ .
Using similar technique one can prove that the same holds also for the ideal
from Example 3.1 (g).

Now we shall show that there are also admissible ideals I ⊂ 2N and se-
quences {fn }n∈N of continuous real functions defined on [0, 1] such that f = I-
limn→∞ fn does not belong to B1 . Using Zorn’s lemma one can show that in
the family of all admissible ideals I ⊂ 2N there exists a maximal ideal (with
respect to inclusion). We shall need the following properties of maximal ideals.
I-Convergence 683

Lemma 5.1. If I ⊂ 2N is a maximal admissible ideal, then for each A ⊂ N

we have either A ∈ I, or N \ A ∈ I.

Lemma 5.2. Let I ⊂ 2N be a maximal admissible ideal. Then each bounded

sequence x = {xn }n∈N of real numbers is I-convergent; i.e., there exists ξ ∈ R
such that ξ = I-limn→∞ xn .

Proof. See Theorem 4.1. in [14].

Theorem 5.2. Let I ⊂ 2N be a maximal admissible ideal. There exists an I-

convergent sequence {fn }n∈N of continuous functions transforming [0, 1] onto
[0, 1] such that f = I-limn→∞ fn does not belong to B1 .

Proof. Let {ri }i∈N be a sequence containing each rational number from [0, 1]
exactly once. For n ∈ N, let fn : [0, 1] → [0, 1] be a continuous function such
that fn (ri ) = 1 for i ∈ {1, . . . , n} and λ({x ∈ [0, 1] : fn (x) > 0}) < 2−n , where
λ is a Lebesgue measure. Then we have lim supn→∞ fn (x) = 0 a.e. on [0, 1]
and limn→∞ fn (ri ) = 1 for i ∈ N. Since obviously I-limn→∞ fn (x) = f (x)
exists everywhere by Lemma 5.2 and I-limn→∞ fn (x) ≤ lim supn→∞ fn (x),
we have I-limn→∞ fn (x) = 0 a.e.; so f is 1 on a set dense in [0, 1] and is 0 on
another set dense in [0, 1] and consequently f ∈ / B1 .
Observe that in the above proof f is Lebesgue measurable. We can con-
struct another sequence {fk }k∈N of continuous real functions defined on [0, 1]
such that I-limk→∞ fk is non-measurable for each admissible maximal ideal
I ⊂ 2N .
A function f is a function of accumulation of {fk }k∈N if and only if for
each  > 0 and for each finite set {x1 , x2 , . . . , xm } (included in the domain)
there exists an infinite subset K = {k1 , k2 , . . . , kp , . . . } ⊂ N such that for each
k ∈ K and for each i ∈ {1, . . . , m} we have |fk (xi ) − f (xi )| < .
Sierpiński in [30] found a sequence {fk }k∈N of continuous functions trans-
forming [0, 1] onto [0, 1] such that each function of accumulation of this se-
quence is non-measurable.
Observe that if f = I-limn fn , then f is a function of accumulation of {fn }.
Indeed, take  > 0 and a finite set {x1 , x2 , . . . , xm } ⊂ [0, 1]. Then for each
Sm
i ∈ {1, . . . , m} Ei () = {n
Sm: |f n (x i ) − f (x i )| ≥ } ∈ I; so i=1 Ei () ∈ I. But
then (for our ideals) N \ i=1 Ei () is an infinite Sm set (It does
Tmnot belong to I.)
and for each i ∈ {1, . . . , m} and each k ∈ N \ i=1 Ei () = i=1 (N \ Ei ()) we
have |fk (xi )−f (xi )| < . Now if I is a maximal ideal and {fn }n∈N is a sequence
constructed by Sierpiński, then there exists a function f = I-limn fn . Hence
f is also a function of accumulation and therefore f is non-measurable.
684 Pavel Kostyrko, Tibor Šalát and Wladyslaw Wilczyński

References
[1] R. Alexander, Density and multiplicative structure of sets of integers,
Acta Arithm., 12 (1967), 321–332.

[2] R. G. Bartle, J. I. Joichi, The preservation of convergence of measurable

functions under composition, Proc. Amer. Math. Soc., 12 (1968), 122–
126.

[3] T. C. Brown, A. R. Freedman, The uniform density of sets of integers

and Fermat’s last Theorem, Compt. Rendus Math. L’Acad. Sci., 12
(1990), 1–6.

[4] R. C. Buck, The measure theoretic approach to density , Amer. J. Math.,

68 (1946), 560–580.

[5] R. C. Buck, Generalized asymptotic density , Amer. J. Math., 75 (1953),

335–346.

[6] J. Červeňanský, Statistical convergence and statistical continuity ,

Zbornik Vedeckých Prác Mt FSTU, 6 (1998), 207–212.

[7] J. S. Connor, The statistical and strong p-Cesaro convergence of se-

quences , Analysis 8 (1988), 47–63.

[8] J. S. Connor, Two valued measures and summability, Analysis, 10

(1990), 373–385.

[9] J. S. Connor, J. Kline, On statistical limit points and consistency of

statistical convergence , J. Math. Appl., 197 (1996), 392–399.

[10] H. Fast, Sur la convergence statistique, Coll. Math., 2 (1951), 241–244.

[11] J. A. Fridy, On statistical convergence, Analysis, 5 (1985), 301–313.

[12] J. A. Fridy, Statistical limit points, Proc. Amer. Math. Soc., 118 (1993),
1187–1192.

[13] P. Kostyrko, M. Mačaj, T. Šalát and O. Strauch, On statistical limit

points, Proc. Amer. Math. Soc., to appear.

[14] P. Kostyrko, M. Mačaj, T. Šalát, Statistical convergence and I-

convergence, Real Analysis Exch., submitted.

[15] K. Kuratowski, Topology, Academic Press, Warszawa, 1966.

I-Convergence 685

[16] D. Maharam, Finitely additive measures on the integers , Sankhya Indian

J. Stat., 38A (1976), 44–59.
[17] P. Mikusiński, Axiomatic theory of convergence, (Polish) Uniw. Śla̧ski.
Prace Nauk. Prace Matem., 12 (1982), 13–21.
[18] H. I. Miller, A measure theoretic subsequence characterization of statis-
tical convergence, Trans. Amer. Math. Soc., 347 (1945), 1811–1819.
[19] D. S. Mitrinović, J. Sandor, B. Crstici, Handbook of Number Theory,
Kluwer Acad. Publ., Dordrecht-Boston-London, 1996.
[20] Jun-iti Nagata, Modern General Topology, North - Holland Publ. Comp.,
Amsterdam-London, 1974.
[21] T. Neubrunn, J. Smital, T, Šalát, On certain properties characterizing
locally separable metric spaces, Čas. pešt. mat., 92 (1967), 157–161.
[22] H. H. Ostmann, Additive Zahlentheorie I , Springer-Verlag, Berlin-
Göttingen-Heidelberg, 1956.
[23] G. M. Petersen, Regular Matrix Transformations, Mc Graw-Hill,
London-New York-Toronto-Sydney, 1966.
[24] B. J. Powell, T. Šalát, Convergence of subseries of the harmonic series
and asymptotic densities of sets of positive integers, Publ. Inst. Math.
(Beograd), 50 (1964), 60–70.
[25] D. Preiss, Approximate derivatives and Baire classes, Czechoslovak
Math. J. Z1, 96 (1971), 373–382.
[26] T. Šalát, R. Tijdeman, On statistically convergent sequences of real num-
bers, Math. Slov., 30 (1980), 139–150.
[27] T. Šalát, R. Tijdeman, Asymptotic densities of sets of positive integers,
Math. Slov., 33 (1983), 199–207.
[28] T. Šalát, R. Tijdeman, On density measures of sets of positive integers,
Coll. Math. Soc. J. Bolyai 34. Topics in Classical Number Theory, Bu-
dapest 1981, 1445–1457.
[29] I. J. Schoenberg, The integrability of certain functions and related
summability methods , Amer. Math. Monthly , 66 (1959), 361–375.
[30] W. Sierpiński, Sur un suite infinie de fonctions continues dont tout fonc-
tion d’accumulation est non measurable, Publ. Inst. Math. Beograd, 1
(1947), 5–10
686 Pavel Kostyrko, Tibor Šalát and Wladyslaw Wilczyński