1 Introduction

The notion of statistical convergence, which is an extension of the usual idea of convergence, was introduced by Fast [1] and also independently by Schoenberg [2] for real and complex sequences, but rapid developments were started after the papers of Šalát [3] and Fridy [4]. Nowadays it has become one of the most active area of research in the field of summability. Di Maio and Kočinac [5] introduced the concept of statistical convergence in topological spaces and statistical Cauchy condition in uniform spaces and established the topological nature of this convergence. Statistical convergence has several applications in different fields of mathematics: summability theory, number theory, trigonometric series, probability theory, measure theory, optimization, and approximation theory. Recently a lot of interesting developments have occurred in double statistical convergence and related topics (see [69] and [10]).

Before continuing with this paper we present some definitions and preliminaries.

By X we will denote an Abelian topological Hausdorff group, written additively, which satisfies the first axiom of countability. In [11], a sequence ( x k ) in X is called to be statistically convergent to an element L of X if, for each neighborhood U of 0,

lim n 1 n |{kn: x k LU}|=0,

where the vertical bars indicate the number of elements in the enclosed set and is called statistically Cauchy in X if for each neighborhood U of 0 there exists a positive integer n 0 (U), depending on the neighborhood U, such that

lim n 1 n |{kn: x k x n 0 ( U ) U}|=0.

The set of all statistically convergent sequences in X is denoted by st(X) and the set of all statistically Cauchy sequences in X is denoted by stC(X). It is well known that stC(X)=st(X) when X is complete.

By a lacunary sequence, we mean an increasing sequence θ=( k r ) of positive integers such that k 0 =0 and h r : k r k r 1 as r. Throughout this paper, the intervals determined by θ will be denoted by I r =( k r 1 , k r ], and the ratio ( k r ) ( k r 1 ) 1 will be abbreviated by q r .

In another direction, in [12], a new type of convergence called lacunary statistical convergence was introduced as follows: A sequence ( x k ) of real numbers is said to be lacunary statistically convergent to L (or S θ -convergent to L) if, for any ε>0,

lim r 1 h r | { k I r : | x k L | ε } |=0,

where |A| denotes the cardinality of AN. In [12] the relation between lacunary statistical convergence and statistical convergence was established among other things. In [13], Mursaleen and Mohiuddine extended the idea of lacunary statistical convergence with respect to the intuitionistic fuzzy normed space.

Çakalli [14] defined lacunary statistical convergence in topological groups as follows: A sequence (x(k)) is said to be S θ -convergent to L (or lacunary statistically convergent to L) if, for each neighborhood U of 0, lim r ( h r ) 1 |{k I r :x(k)LU}|=0. In this case, we write

S θ - lim k x(k)=Lorx(k)L( S θ )

and define

S θ (X)= { ( x ( k ) ) : for some  L , S θ - lim k x ( k ) = L } .

2 Definitions and notations

By the convergence of a double sequence we mean the convergence in Pringsheim’s sense (see [15]). A double sequence x=(x(k,l)) is said to be convergent in Pringsheim’s sense if for every ε>0 there exists NN such that |x(k,l)L|<ε whenever k,lN. L is called the Pringsheim limit of x. We shall describe such an x more briefly as ‘P-convergent’.

A double sequence x=(x(k,l)) is said to be Cauchy sequence if for every ε>0 there exists NN, where N is the set of natural numbers such that |x(p,q)x(k,l)|<ε for all pkN and qlN.

Let KN×N be a two-dimensional set of positive integers and let K m , n be the numbers of (i,j) in K such that in and jm. Then the two-dimensional analog of natural case density can be defined as follows: The lower asymptotic density of K is defined as

P- lim inf m , n K m , n m n = δ 2 (K).

In the case when the sequence { K m , n m n } m , n = 1 , 1 , has a limit we say that K has a natural density and is defined as

P- lim m , n K m , n m n = δ 2 (K).

For example, let K={( i 2 , j 2 ):i,jN×N}. Then

δ 2 (K)=P- lim m , n K m , n m n P- lim m , n m n m n =0

(i.e., the set K has double natural density zero), while the set {(i,2j):i,jN×N} has double natural density 1/2.

Recently the studies of double sequences have seen rapid growth. The concept of double statistical convergence, for the complex case, was introduced by Mursaleen and Edely [16] and others, while the idea of statistical convergence of single sequences was first studied by Fast [1]. Also the double lacunary statistical convergence was introduced by Patterson and Savaş [17].

Mursaleen and Edely have given the main definition.

Definition 2.1 ([16])

A double sequence x=(x(k,l)) is said to be P-statistically convergent to L provided that for each ε>0

P- lim m , n 1 m n { number of  ( k , l ) : k < m  and  l < n , | x ( k , l ) L | ε } =0.

In this case we write st 2 - lim k , l x(k,l)=L and we denote the set of all statistical convergent double sequences by st 2 .

It is clear that a convergent double sequence is also st 2 -convergent but the inverse is not true, in general. Also note that a st 2 -convergence does need not to be bounded. For example, the sequence x=( x k , l ) defined by

x(k,l)= { k l , if  k  and  l  are square , 1 , otherwise ,

is st 2 -convergent. Nevertheless it neither is convergent nor bounded.

It should be noted that in [16], the authors proved the following important theorem.

Theorem 2.2 The following statements are equivalent:

  1. (a)

    x is statistically convergent to L;

  2. (b)

    x is statistically Cauchy;

  3. (c)

    there exists a subsequence y of x such that

    lim j k y j k =L.

The double sequence θ={( k r , l s )} is called double lacunary if there exist two increasing sequences of integers such that

k 0 =0, h r = k r k k 1 as r

and

l 0 =0, h ¯ s = l s l s 1 as s.

Notations: k r , s = k r l s , h r , s = h r h ¯ s , θ is determined by I r ={(k): k r 1 <k k r }, I s ={(l): l s 1 <l l s }, I r , s ={(k,l): k r 1 <k k r  and  l s 1 <l l s }, q r = k r k r 1 , q ¯ s = l s l s 1 , and q r , s = q r q ¯ s . We will denote the set of all double lacunary sequences by N θ r , s .

Let KN×N have the double lacunary density δ 2 θ (K) if

P- lim r , s 1 h r , s | { ( k , l ) I r , s : ( k , l ) K } |

exists.

Example 1 Let θ={( 2 r 1, 3 s 1)} and K={(k,2l):k,lN×N}. Then δ 2 θ (K)=0. But it is obvious that δ 2 (K)=1/2.

In 2005, Patterson and Savaş [17] studied double lacunary statistically convergence by giving the definition for complex sequences.

Definition 2.3 Let θ be a double lacunary sequence; the double number sequence x is st θ 2 -convergent to L provided that, for every ε>0,

P- lim r , s 1 h r , s | { ( k , l ) I r , s : | x ( k , l ) L | ε } |=0.

In this case we write st θ 2 -limx=L or x(k,l)L( S θ 2 ).

More investigation in this direction and more applications of double lacunary and double sequences can be found in [1820] and [21].

In this presentation, our goal is to extend a few results known in the literature from ordinary (single) sequences to double sequences in topological groups and to give some important inclusion theorems.

Quite recently, Çakalli and Savaş [22] defined the statistical convergence of double sequences x=(x(k,l)) of points in a topological group as follows.

In a topological group X, a double sequence x=(x(k,l)) is called statistically convergent to a point L of X if for each neighborhood U of 0 the set

{ ( k , l ) , k n  and  l m : x ( k , l ) L U }

has double natural density zero. In this case we write S 2 - lim k , l x(k,l)=L and we denote the set of all statistically convergent double sequences by S 2 (X).

Now we are ready to give the definition of double lacunary statistical convergence in topological groups.

Definition 2.4 A sequence (x(k,l)) is said to be S θ 2 -convergent to L (or double lacunary statistically convergent to L) if, for each neighborhood U of 0,

P- lim r , s ( h r , s ) 1 | { ( k , l ) I r , s : x ( k , l ) L U } |=0.

In this case, we write

S θ 2 - lim k , l x(k,l)=Lorx(k,l)L ( S θ 2 )

and define

S θ 2 (X)= { ( x ( k , l ) ) : for some  L , S θ 2 - lim k , l x ( k , l ) = L }

and, in particular,

S θ 2 ( X ) 0 = { ( x ( k , l ) ) : S θ 2 - lim k , l x ( k , l ) = 0 } .

It is obvious that every double lacunary statistically convergent sequence has only one limit, that is, if a sequence is double lacunary statistically convergent to L 1 and L 2 then L 1 = L 2 .

3 Inclusion theorems

In this section, we prove some analogs for double sequences. For single sequences such results have been proved by Çakalli [14].

Theorem 3.1 For any double lacunary sequence θ={( k r , l s )}, S 2 (X) S θ 2 (X) if and only if lim inf q r >1 and lim inf q ¯ s >1.

Proof Sufficiency: Suppose that lim inf q r >1, and lim inf q ¯ s >1, lim inf q r = α 1 , and lim inf q ¯ s = α 2 , say. Write β 1 =( α 1 1)/2 and β 2 =( α 2 1)/2. Then there exist two positive integers r 0 and s 0 such that q r 1+ β 1 for r r 0 and q ¯ s 1+ β 2 for s s 0 . Thus, for r r 0 , and s s 0 ,

h r , s ( k r ) 1 ( l s ) 1 = ( 1 ( k r 1 ) ( k r ) 1 ) ( 1 ( l s 1 ) ( l s ) 1 ) = ( 1 ( q r ) 1 ) ( 1 ( q s ) 1 ) ( 1 ( 1 + β 1 ) 1 ) ( 1 ( 1 + β 2 ) 1 ) = ( β 1 ( 1 + β 1 ) 1 ) ( β 2 ( 1 + β 2 ) 1 ) .

Take any (x(k,l)) S 2 (X), and S 2 - lim ( k , l ) x(k,l)=L, say. We have S θ 2 - lim ( k , l ) x(k,l)=L. Let us take any neighborhood U of 0. Then, for r r 0 and s s 0 , we get

( k r ) 1 ( l s ) 1 | { k k r , l l s : x ( k , l ) L U } | ( k r ) 1 ( l s ) 1 | { ( k , l ) I r , s : x ( k , l ) L U } | = h r , s ( k r ) 1 ( l s ) 1 ( h r , s ) 1 | { ( k , l ) I r , s : x ( k l ) L U } | β 1 ( 1 + β 1 ) 1 β 2 ( 1 + β 2 ) 1 ( h r , s ) 1 | { ( k , l ) I r , s : x ( k , l ) L U } | .

Hence S θ 2 - lim ( k , l ) x(k,l)=L.

Necessity: Suppose that lim inf r q r =1 and lim inf s q ¯ s =1. Then we can choose a subsequence {( k r (j)),( l s (i))} of the lacunary sequence θ r , s ={ k r , l s } such that

k r ( j ) ( k r ( j ) 1 ) 1 <1+ j 1 , k r ( j ) 1 ( k r ( j 1 ) ) 1 >j

and

l s ( i ) ( l s ( i ) 1 ) 1 <1+ i 1 , l s ( i ) 1 ( l s ( i 1 ) ) 1 >i,

where r(j)>r(j1)+2 and s(i)>s(i1)+2. Take an element x of X different from 0. Define a sequence (x(k,l)) by x(k,l)=x if (k,l){ I r ( j ) , s ( i ) } for some j=1,2,,m, , i=1,2,,n, , and x(k,l)=0 otherwise. Then (x(k,l)) S 2 (X) (in fact (x(k,l)) S 0 2 (X)). To prove this take any neighborhood U of 0. Now we may choose a neighborhood W of 0 such that WU and xW. On the other hand, for each m and n we can find two positive numbers j m and i n such that k r ( j m ) <m k r ( j m ) + 1 and l s ( i n ) <n l s ( i n ) + 1 . Therefore

( m n ) 1 | { k m , l n : x ( k , l ) U } | k r ( j m ) 1 l s ( i n ) 1 | { k m , l n : x ( k , l ) W } | k r ( j m ) 1 l s ( i n ) 1 { | { k k r ( j m ) , l l s ( i n ) : x ( k , l ) W } | + | { k r ( j m ) < k m , l s ( i n ) < l n : x ( k , l ) W } | } k r ( j m ) 1 l s ( i n ) 1 | { k k r ( j m ) , l l s ( i n ) : x ( k , l ) W } | + k r ( j m ) 1 ( k r ( j m ) + 1 k r ( j m ) ) , l s ( i n ) 1 ( l s ( i n ) + 1 l s ( i n ) ) < ( ( j m + 1 ) 1 + 1 + j m 1 1 ) ( ( i n + 1 ) 1 + 1 + i n 1 1 ) < ( ( j m + 1 ) 1 + j m 1 ) ( ( i n + 1 ) 1 + i n 1 )

for each (m,n). Hence (x(k,l)) S 0 2 (X). Now let us observe that (x(k,l)) S 0 2 (X). Since X is a Hausdorff space, there exists a symmetric neighborhood V of 0 such that xV. Hence

lim j , i ( h r ( j ) , s ( i ) ) 1 | { k r ( j ) 1 < k k r ( j ) , l s ( i ) 1 < l l s ( i ) : x ( k , l ) V } | = lim j , i ( h r ( j ) ) 1 ( h s ( i ) ) 1 ( k r ( j ) k r ( j ) 1 ) ( l s ( i ) l s ( i ) 1 ) = lim j , i ( h r ( j ) ) 1 h r ( j ) ( h s ( i ) ) 1 ( h s ( i ) ) = 1

and

lim r , s r r ( j ) , s s ( i ) , j = i = 1 , 2 , h r , s 1 | { k r 1 < k k r , l s 1 < l l s : x ( k , l ) x V } |=10.

Therefore neither x nor 0 can be a double lacunary statistical limit of (x(k,l)). No other point of X can be a double lacunary statistical limit of the sequence as well. Thus (x(k,l)) S θ 2 (X). This completes the proof of this theorem. □

Theorem 3.2 For any lacunary sequence θ={( k r , l s )}, S θ 2 (X) S 2 (X) if and only if lim sup r q r < and lim sup s q ¯ s <.

Proof Sufficiency: If lim sup r q r < and lim sup s q ¯ s <, there exists an H>0 such that q r <H and q ¯ s <H for all (r,s). Let (x(k,l)) S θ 2 (X), S θ 2 - lim k , l x(k,l)=L, say. Take any neighborhood U of 0. Let ε>0. Write N r , s ={(k,l) I r , s :x(k,l)LU} by the definition of a double lacunary statistical convergence, there are positive integers r 0 and s 0 such that N r , s ( h r , s ) 1 <ε for all r> r 0 and s> s 0 . Let M=max{ N r , s :1r r 0  and 1s s 0 } and let n and m be such that k r 1 <m k r and l s 1 <n l s ; hence we have

( m n ) 1 | { k m , l n : x ( k , l ) L U } | r 0 s 0 M ( k r 1 ) 1 ( l s 1 ) 1 +ε H 2 q r , s .

Since lim r , s k r l s =, there exist two positive integers r 1 r 0 and s 1 s 0 such that

( k r 1 ) 1 ( l s 1 ) 1 < ( 4 r 0 s 0 M / ε ) 1 for r> r 1  and s> s 1 .

Thus for r> r 1 and s> s 1

( m n ) 1 | { k m , l n : x ( k , l ) L U } | ε 2 + ε 2 =ε.

Finally it follows that S 2 - lim k , l x(k,l)=L.

Necessity: We shall assume that lim sup r q r = and lim sup s q ¯ s =. Take an element x of X different from 0. Construct two subsequences ( k r ( j ) ) and ( l s ( i ) ) of the lacunary sequence θ r , s =( k r , l s ) such that k r ( j ) >j and l s ( i ) >i, k r ( j ) >j+3, and l s ( i ) >i+3 and define a sequence (x(k,l)) by x(k,l)=x if k r ( j ) 1 <k2 k r ( j ) 1 and l s ( i ) 1 <l2 l s ( i ) 1 for some i=j=1,2, , and x(k,l)=0 otherwise. Let U be a symmetric neighborhood of 0 that does not include x. Then, for j,i>1,

( h r ( j ) , s ( i ) ) 1 | { k k r ( j ) , l l s ( i ) : x ( k , l ) U } | < ( k r ( j ) 1 ) ( l s ( i ) 1 ) ( h r ( j ) , s ( i ) ) 1 < ( j 1 ) 1 ( i 1 ) 1 .

Hence (x(k,l)) S θ 2 (X). But (x(k,l)) S 2 (X). For

( 2 k r ( j ) 1 ) 1 ( 2 l s ( i ) 1 ) 1 | { k 2 k r ( j ) 1 , l 2 l s ( i ) 1 : x ( k , l ) U } | = ( 2 k r ( j ) 1 ) 1 ( 2 l s ( i ) 1 ) 1 [ k r ( 1 ) 1 + k r ( 2 ) 1 + + k r ( j ) 1 ] [ l s ( 1 ) 1 + l s ( 2 ) 1 + + l s ( i ) 1 ] > 1 / 4 ,

which implies that (x(k,l)) cannot be double statistically convergent. This completes the proof of the theorem. □

Corollary 3.3 Let θ={( k r , l s )} be a double lacunary sequence, then S 2 (X)= S θ 2 (X) iff

1<lim inf r q r lim sup r q r <

and

1<lim inf s q s lim sup s q s <.

In Section 2 we mentioned that the S θ 2 -limit is unique. However, it is possible for a sequence to have different S θ 2 -limits for different θ’s. The following theorem shows that this situation cannot occur if x S 2 (X).

Theorem 3.4 If (x(k,l)) belongs to both S 2 (X) and S θ 2 (X), then

S 2 - lim k , l x(k,l)= S θ 2 - lim k , l x(k,l).

Proof Take any (x(k,l)) S 2 (X) S θ 2 (X) and S 2 - lim k , l x(k,l)= L 1 , S θ 2 - lim k , l x(k,l)= L 2 , say. Assume that L 1 L 2 . Since X is a Hausdorff space, there exists a symmetric neighborhood U of 0 such that L 1 L 2 U. Then we may choose a symmetric neighborhood W of 0 such that W+WU. Then we obtain the following inequality:

( k m ) 1 ( l n ) 1 | { k k n , l l m : z ( k , l ) U } | ( k m ) 1 ( l n ) 1 | { k k m , l l n : x ( k , l ) L 1 W } | + ( k m ) 1 ( l n ) 1 | { k k m , l l m : L 2 x ( k , l ) W } | ,

where z(k,l)= L 2 L 1 for all k,lN×N. It follows from this inequality that

1 ( k m ) 1 ( l n ) 1 | { k k m , l l n : x ( k , l ) L 1 W } | + ( k m ) 1 ( l n ) 1 | { k k m , l l n : L 2 x ( k , l ) W } | .

The second term on the right side of this inequality approaches 0 as m,n. To observe this, write

( k m ) 1 ( l n ) 1 | { k k m , l l n : L 2 x ( k , l ) W } | = ( k m ) 1 ( l n ) 1 | { k , l r , s = 1 m , n I r , s : L 2 x ( k , l ) W } | = ( k m ) 1 ( l n ) 1 r , s = 1 m , n | { k , l I r : L 2 x ( k , l ) W } | = ( r , s = 1 m , n h r , s ) 1 ( r , s = 1 m , n h r , s t r , s ) ,

where t r , s = h r , s 1 |{k,l I r , s : L 2 x(k,l)W}| is a Pringsheim null sequence, since S θ 2 - lim k , l x(k,l)= L 2 . Hence the regular weighted mean transform of ( t r , s ) also tends to zero, that is,

P- lim m , n ( k m ) 1 ( l n ) 1 | { k k m , l l n : L 2 x ( k , l ) W } |=0.
(3.1)

On the other hand, since S- lim k , l x(k,l)= L 1 ,

P- lim m ( k m ) 1 ( l n ) 1 | { k k m , l l n : x ( k , l ) L 1 W } |=0.
(3.2)

By (3.1) and (3.2) it follows that

P- lim m , n ( k m ) 1 ( l n ) 1 | { k k m , l l n : z ( k , l ) U } |=0.

This contradiction completes the proof. □

Before presenting the next theorem, let us consider the following definition.

Definition 3.5 Let θ=( k r , l s ) be a double lacunary sequence; the sequence (x(k,l)) is said to be an S θ 2 -Cauchy sequence if there is a subsequence (x( k (r), l (s))) of (x(k,l)) such that k (r), l (s) I r , s , for each r, s, lim r , s x( k (r), l (s))=L and for each neighborhood U of 0,

P- lim r , s ( h r , s ) 1 | { ( k , l ) I r , s : x ( k , l ) x ( k ( r ) , l ( s ) ) U } |=0.

Finally we conclude this paper by presenting the multidimensional analog of Çakalli [14].

Theorem 3.6 A sequence (x(k,l)) is S θ 2 -convergent if and only if it is an S θ 2 -Cauchy sequence.

Proof Sufficiency: Assume that (x(k,l)) is an S θ 2 -Cauchy sequence. Let U be any neighborhood of 0. Then we may choose a neighborhood W of 0 such that W+WU. Therefore

( h r , s ) 1 | { ( k , l ) I r , s : x ( k , l ) L U } | ( h r , s ) 1 | { ( k , l ) I r , s : x ( k , l ) x ( k ( r ) , l ( s ) ) W } | + ( h r , s ) 1 | { ( k , l ) I r , s : x ( k ( r ) , l ( s ) ) L W } | .

Since lim r , s ( h r , s ) 1 |{(k,l) I r , s :x(k,l)x( k (r), l (s))U}|=0, in Pringsheim sense, and lim r , s x( k (r), l (s))=L, by the assumption, it follows from the last inequality that S θ 2 - lim k x(k,l)=L.

Necessity: Suppose that S θ 2 - lim k , l x(k,l)=L. Let ( U n , m ) be a nested base of neighborhoods of 0. Write K ( i , j ) ={(k,l)N×N:x(k,l)L U i , j } for each (i,j)N×N. Thus for each (i,j) we obtain the following: K ( i + 1 , j + 1 ) K ( i , j ) and lim r , s | K ( i , j ) I r , s | ( h r , s ) 1 =1. This implies that there exist m(1) and n(1) such that rm(1) and sn(1) and | K ( 1 , 1 ) I r , s | h r , s >0, that is, K ( 1 , 1 ) I r , s . We next choose m(2)>m(1) and n(2)>n(1) such that r>m(2) and s>n(2) implies that K ( 2 , 2 ) I r , s . Thus, for each (r,s) satisfying m(1)r<m(2) and n(1)s<n(2), we can choose ( k (r), l (s)) I r , s such that ( k (r), l (s)) K ( r , s ) I r , s , that is, x( k (r), l (s))L U 11 . In general, we choose m(p+1)>m(p), n(q+1)>n(q) such that r>m(p+1) and s>n(q+1) implies that I r , s K ( p + 1 , q + 1 ) . Then for all (r,s) satisfying m(p)r<m(p+1) and n(q)s<n(q+1) choose ( k (r), l (s)) I r , s K ( p , q ) , that is, x( k (r), l (s))L U p , q . Hence it follows that P- lim r , s x( k (r), l (s))=L. Let U be any neighborhood of 0. Then we may choose a symmetric neighborhood W of 0 such that W+WU. Now we write

( h r , s ) 1 | { ( k , l ) I r , s : x ( k , l ) x ( k ( r ) , l ( s ) ) U } | ( h r , s ) 1 | { ( k , l ) I r , s : x ( k , l ) L W } | + ( h r , s ) 1 | { ( k , l ) I r , s : L x ( k ( r ) , l ( s ) ) W } | .

Since S θ 2 - lim k , l x(k,l)=L and P- lim r , s x( k (r), l (s))=L, we have

lim r , s ( h r , s ) 1 | { ( k , l ) I r , s : x ( k , l ) x ( k ( r ) , l ( s ) ) U } |=0,

in Pringsheim’s sense. Thus the theorem is proven. □