Home On bivariate fractal interpolation for countable data and associated nonlinear fractal operator
Article Open Access

On bivariate fractal interpolation for countable data and associated nonlinear fractal operator

  • Kshitij Kumar Pandey , Nicolae Adrian Secelean and Puthan Veedu Viswanathan EMAIL logo
Published/Copyright: October 19, 2024
Become an author with De Gruyter Brill

Abstract

Fractal interpolation has been conventionally treated as a method to construct a univariate continuous function interpolating a given finite data set with the distinguishing property that the graph of the interpolating function is the attractor of a suitable iterated function system. On the one hand, attempts have been made to extend the univariate fractal interpolation from a finite data set to a countably infinite set. On the other hand, fractal interpolation in higher dimensions, particularly the theory of fractal interpolation surfaces (FISs), has received increasing attention for more than a quarter century. This article targets a two-fold extension of the notion of fractal interpolation by providing a general framework to construct FISs for a prescribed set consisting of countably infinite data on a rectangular grid. By using this as a crucial tool, we obtain a parameterized family of bivariate fractal functions simultaneously interpolating and approximating a prescribed bivariate continuous function. Some elementary properties of the associated nonlinear (not necessarily linear) fractal operators are established, thereby allowing the interaction of the notion of fractal interpolation with the theory of nonlinear operators.

MSC 2010: 28A80; 41A05; 47H14; 47H09

1 Introduction

Over the past three decades, the subject of fractal interpolation has been one of the major research themes among the fractals community. The concept of fractal interpolation function (FIF) was introduced by Barnsley in his fundamental and pioneering work [1]. He proved the existence of a univariate continuous function interpolating a given finite data set, with its novelty lying in the fact that the graph of the constructed interpolant is a fractal, that is, the attractor of a suitable iterated function system (IFS) [2,3]. We do not attempt here to provide a complete list of references on fractal interpolation, as the field is quite large. Instead, we refer the reader to a few recent research on fractal interpolation [411] and the references therein. For a compendium of fractal interpolation and related topics, we refer the reader to the monograph [12].

As is well known, interpolation and approximation are two intimately related concepts. However, the interplay between these two theories is more subtle in the fractal setting. It is our opinion that the notion of α -fractal function [13,14] is mostly responsible for establishing interesting interconnections between interpolation and approximation theories of univariate fractal functions. In fact, α -fractal function provides a parameterized family of fractal functions that simultaneously interpolate and approximate a given univariate continuous function. The parameters can be adjusted so that the fractal functions share or modify the properties of the original function, for instance, smoothness and shape properties [15,16]. The analytical properties of the fractal operator that maps a function to its fractal counterpart, and new approximation classes of self-referential functions obtained by “fractalizing” various approximation classes of (non-fractal) functions (such as polynomials, rational functions, and trigonometric polynomials) have received considerable attention in the literature on univariate fractal approximation theory [13,14,1719]. We stress here that the studies on the aforementioned univariate fractal operator are mostly confined to the realm of bounded linear operators on various function spaces.

As with any such salient idea, numerous questions and results based on fractal interpolation were spawned. One such question is related to its extension to the multivariate case. In this direction, several works have been done to interpolate a given bivariate data using bivariate FIFs (fractal surfaces); see, for instance, [2028] and a few research works deal with the multivariate FIFs [29,30]. Among various constructions available in the literature, the general framework to construct fractal surfaces given in [31] aroused our interest, partially due to the fact that the construction thereat is amenable to obtain a bivariate analogue of the α -fractal function, which is a natural entry point to delve into the theory of bivariate fractal approximation, see also [32,33].

Much of the existing literature dealing with univariate and multivariate FIFs concentrates primarily on finite data sets. This is no happenstance, as fractal interpolation hinges on the Hutchinson’s fundamental result about the existence of an invariant set (attractor) for a set-valued map induced by a finite number of contractions [2]. Attempts were made in literature to define these concepts in the setting of countably infinite number of maps. For instance, Secelean adapted the Hutchinson approach so as to handle a countable number of contractions [34,35] and deduced the existence of a univariate fractal function interpolating a data set consisting of countably infinite points [35,36]. Bivariate FIF for an infinite sequence of data is not studied hitherto, and naturally, we want to generalize the construction of fractal surfaces so as to accommodate countably infinite data sets. Let us note that this is practically important. For instance, in the theory of sampling and reconstruction, often one works with infinite sequence of data points, and a general approach focuses to seek an approximate rather than the perfect reconstruction on some restricted class of signals.

One of the objectives of this article is to unveil a method for constructing fractal interpolation surface (FIS) for a countably infinite data set over rectangular grids. We are inspired by the finite case treated in [31], and we will adapt this construction to the setting of countable data. This part of the current findings may be viewed also as a sequel to [36], where the univariate fractal interpolation for a countable data set has been studied. The countable bivariate scenario explored in this study necessitates certain extra considerations and assumptions, making the analysis more difficult. As hinted earlier, among various constructions of FISs, our choice of [31] is guided by the fact that it offers an efficient platform to obtain a parameterized family of fractal functions corresponding to a prescribed bivariate continuous function. Crucial to further development will be the aforementioned parameterized family consisting of approximate fractal reconstructions of the original function (seed function) defined on a rectangular domain. This family of fractal functions is obtained by sampling the seed function at infinite number of grid points in the domain and applying the countable FIS scheme developed in the first part of this article.

We define a fractal operator that sends each (non-fractal) seed function to its fractal approximant, and study some analytical properties of this operator, which, in general, is nonlinear. Though the bounded linear fractal operator has its origin in the theory of univariate α -fractal function (see, for example, [14]), our focus will be to its intriguing links with the perturbation theory of operators (not necessarily linear or bounded). Thus, the research reported here could open the door for an intense and fruitful interaction of two fields – fractal interpolation and the theory of nonlinear operators, and the potential applications lie, for instance, in the field of sampling and reconstruction. The present article is a corrected and improved version of the note available in [37].

2 Preliminaries

We list pertinent definitions and notation from the theory of countable IFS [34,36] and nonlinear operators for use in the subsequent parts of this article. The terminologies on perturbation theory of nonlinear operators assembled here can be found in [38], and many of these are obvious modifications to the notions present in the well known treatise for perturbation theory of linear operators by Kato [39].

2.1 Countable IFS

Definition 2.1

Let ( X , d ) be a compact metric space and ( ω i ) i N be a sequence of continuous self-maps on X . The system { X , ( ω i ) i N } is said to be a countable iterated function system (CIFS). We say that the CIFS is hyperbolic if the functions ω i , i N , are contractions with respective contractivity factors r i such that sup i N r i < 1 .

Definition 2.2

A non-empty set A X is said to be a set fixed point of the CIFS { X , ( ω i ) i N } if

A = i N ω i ( A ) ¯ ,

where the notation bar denotes the closure of the respective set. A non-empty set A X is said to be the attractor of the CIFS if A is compact, and it is the unique set fixed point of the CIFS.

Definition 2.3

Let ( X ) denote the set of all non-empty closed (and hence compact) subsets of X endowed with the Hausdorff-Pompeiu metric [3]. The set-valued operator W : ( X ) ( X ) defined by

(2.1) W ( B ) = i N ω i ( B ) ¯ B ( X )

is said to be the Hutchinson operator associated with the CIFS { X , ( ω i ) i N } .

Theorem 2.4

[34] Suppose that the CIFS { X , ( ω i ) i N } is hyperbolic. Then the corresponding Hutchinson operator is a contraction on the complete metric space ( X ) , and consequently, by the Banach fixed point theorem, it has an attractor. That is, there exists a unique non-empty compact set A X such that

A = W ( A ) = i N ω i ( A ) ¯ .

Moreover,

  1. for every nonempty closed set B X , the sequence ( W n ( B ) ) n N , where W n denotes the n-fold autocomposition of W , converges to A.

  2. the attractor A can be approximated with respect to the Hausdorff-Pompeiu metric by the attractors of the partial IFSs { X , ( ω i ) i = 1 n } n N .

2.2 A few basics of linear and nonlinear operator theory

In this subsection, let X and Y be two normed linear spaces over the same field K , where K is R or C .

Definition 2.5

An operator A : D ( A ) X Y is said to be closed if its graph G { ( x , A ( x ) ) : x D ( A ) } is a closed subset of X × Y . That is, for any sequence ( x n ) n N in D ( A ) , x n x and A ( x n ) y Y implies x D ( A ) and A ( x ) = y .

Definition 2.6

An operator A : D ( A ) X Y is said to be closable if it has a closed extension. That is, there exists a set X 0 , D ( A ) X 0 X and a closed operator A ˜ : X 0 X Y such that A ˜ ( x ) = A ( x ) for all x D ( A ) .

Definition 2.7

Let A : D ( A ) X Y and B : D ( B ) X Y be two operators such that D ( B ) D ( A ) . If for every sequence ( x n ) n N in D ( B ) with x n x , B ( x n ) y and A ( x n ) z imply x D ( A ) and A ( x ) = z , then A is said to be B -closed.

Definition 2.8

Let A : D ( A ) X Y and B : D ( B ) X Y be two operators such that D ( B ) D ( A ) . Assume that, for every pair of convergent sequences ( x n ) n N , ( x n ) n N in D ( B ) having a same limit x X , the sequences ( B ( x n ) ) n N and ( B ( x n ) ) n N also converge to a same limit. If for every pair of sequences ( x n ) n N and ( x n ) n N having the same limit x , we have

A ( x n ) z , A ( x n ) z x D ( A ) and z = z ,

then A is said to be B -closable.

Definition 2.9

Let X and Y be normed linear spaces over the field K , where K = R or C , with norms · X and · Y respectively. Let A : D ( A ) X Y be an operator. Define p ( A ) by

p ( A ) = max sup x D ( A ) , x 0 A ( x ) Y x X , A ( 0 ) Y .

Definition 2.10

If p ( A ) < , then the operator A is said to be norm-bounded operator, and the quantity p ( A ) is called the norm of A .

Definition 2.11

An operator A : D ( A ) X Y is said to be topologically bounded if it maps bounded sets to bounded sets.

Remark 2.12

Let D ( A ) = X . In contrast to the case of linear operators, here, the two notions – norm-boundedness and topologically boundedness – are not equivalent.

Definition 2.13

Let A : D ( A ) X Y and B : D ( B ) X Y be two operators such that D ( B ) D ( A ) . Then we say that A is relatively (norm) bounded with respect to B or simply B -bounded if for some non-negative constants a and b , the following inequality holds:

(2.2) A ( x ) Y a x X + b B ( x ) Y x D ( B ) .

The infimum b 0 of all values of b for which the above inequality is satisfied is called the relative bound of A with respect to B or simply the B -bound of A .

Remark 2.14

If b is chosen very close to b 0 , then the other constant a will, in general, have to be chosen very large; thus it is, in general, impossible to set b = b 0 in the inequality (2.2) in the aforementioned definition; see [39].

Definition 2.15

An operator A : D ( A ) X Y is said to be Lipschitz if there exists a constant M > 0 such that

A ( x ) A ( y ) Y M x y X x , y D ( A ) .

For a Lipschitz operator A : D ( A ) X Y , the Lipschitz constant is defined by

A sup x y A ( x ) A ( y ) Y x y X .

Definition 2.16

Let A : D ( A ) X Y and B : D ( B ) X Y be two operators such that D ( B ) D ( A ) . We say that A is relatively Lipschitz with respect to B or simply B -Lipschitz if the following inequality holds for some non-negative constants M 1 and M 2 .

(2.3) A ( x ) A ( y ) Y M 1 x y X + M 2 B ( x ) B ( y ) Y x , y D ( A ) .

The infimum of all such values of M 2 is called the B -Lipschitz constant of A .

Remark 2.17

Similar to Remark 2.14, let us note that the inequality (2.3) may not hold when M 2 is replaced with the infimum of all values of M 2 .

Definition 2.18

Let X , Y be normed linear spaces, X * , Y * be the corresponding dual spaces, and T : X Y be a bounded linear operator. The adjoint or dual T * of T is the unique map T * : Y * X * defined by

T * ( ψ ) ψ T for all ψ Y * .

Definition 2.19

Given a Banach space X , the annihilator (or pre-annihilator) of a subset S of X * is the subspace defined as follows:

S { x X : ψ ( x ) = 0 ψ S } .

Further, given a Banach space X and a bounded linear map T : X X , we call a subspace Y of X invariant under T or T -invariant if T ( Y ) Y . Further, if Y , Y X , then Y is called a nontrivial invariant subspace.

Lemma 2.20

Let X be a non-separable Banach space and T : X X be a bounded linear operator. Then, for every non-zero element x X , the subspace span { x , T ( x ) , T 2 ( x ) , } ¯ is a non-trivial closed invariant subspace. Here, the span denotes the linear span and bar denotes the closure.

Lemma 2.21

Let X be a Banach space and T : X X be a bounded linear operator. If Y is a closed invariant subspace of the operator T * : X * X * , then Y is a closed invariant subspace of T.

3 Construction of countable bivariate FIS

Denote N 0 N { 0 } and N n { 1 , 2 , , n } for n N . Let us first recall that corresponding to each double sequence s : N 0 × N 0 R denoted by s ( i , j ) = s i , j , there can be considered three important limits, namely,

lim i , j s i , j , lim i ( lim j s i , j ) and lim j ( lim i s i , j ) .

Further, it is worth to recall that the existence of lim i , j s i , j does not ensure, in general, the existence of the limits:

  1. lim i s i , j for each fixed j N 0 .

  2. lim j s i , j for each fixed i N 0 .

  3. iterated limits: lim i ( lim j s i , j ) and lim j ( lim i s i , j ) .

Definition 3.1

A set D = { ( x i , y j , z i , j ) : i , j N 0 } R 3 is said to be a bivariate countable system data (CSD) if

  1. the sequences ( x i ) i N 0 and ( y j ) j N 0 are strictly increasing and bounded,

  2. the double sequence ( z i , j ) is convergent in the sense that lim i , j z i , j exists and it is finite,

  3. lim j z i , j < for each fixed i N 0 , and lim i z i , j < for each fixed j N 0 .

Notation 3.2

For a bivariate CSD, let M z lim i , j z i , j . Let a x 0 , b x lim i x i , c y 0 and d y lim j y j . Set I [ a , b ] and J [ c , d ] . Assume that K is a sufficiently large compact interval containing the set { z i , j : i , j = 0 , 1 , 2 , } and X I × J × K .

3.1 Construction of FIS for countable data

Extensions of Barnsley’s framework [1] in two ways – by considering a countably infinite one-dimensional data set [36] and a bivariate finite data set [31] – have been one of the directions of further studies in fractal interpolation. With an intent to generalize these two approaches and inspired by the two publications [31,36], we ask:

Question 3.3

Can we construct a continuous function g : I × J R interpolating the bivariate CSD, that is, g ( x i , y j ) = z i , j for all i , j N 0 , satisfying the following properties?

  1. g is the fixed point of a suitable Read-Bajraktarevíc type operator,

  2. the graph of g can be realized as the attractor of an appropriate CIFS.

In what follows, we answer the aforementioned question in the affirmative.

For i , j N , let I i [ x i 1 , x i ] and J j [ y j 1 , y j ] . For i N 0 , let

s i 1 + ( 1 ) i 2 ,

(i.e., s i = 0 if i is odd, and s i = 1 if i is even). Define τ : N × { 0 , } N by

(3.1) τ ( i , 0 ) i 1 + s i and τ ( i , ) i s i .

For i , j N , let u i : I I i and v j : J J j be contractive homeomorphisms satisfying

(3.2) u i ( a ) = s i 1 x i 1 + s i x i , u i ( b ) = s i x i 1 + s i + 1 x i ,

(3.3) u i ( x ) u i ( x ) a i x x x , x I ,

(3.4) v j ( c ) = s j 1 y j 1 + s j y j , v j ( d ) = s j y j 1 + s j + 1 y j ,

(3.5) v j ( y ) v j ( y ) b j y y y , y J ,

where a i , b j are positive constants such that sup i a i < 1 and sup j b j < 1 . Let

c i , j max { a i , b j } .

Clearly sup i , j c i , j < 1 . By using (3.2) and (3.4), one can easily observe that for all i , j N ,

(3.6) u i 1 ( x i ) = s i a + s i + 1 b = u i + 1 1 ( x i ) and v j 1 ( y j ) = s j c + s j + 1 d = v j + 1 1 ( y j ) .

For every ( i , j ) N × N , we consider the constants θ i , λ j , α i , j in ( 0 , ) such that the following assertions hold:

(3.7) lim i θ i = lim j λ j = 0 ;

(3.8) lim j α i , j = 0 for all i N , lim i α i , j = 0 for all j N , and lim i , j α i , j = 0 ;

(3.9) α = sup i , j α i , j < 1 .

For ( i , j ) N × N , let us now define the functions F i , j : X K such that for every pair of points ( x , y , z ) and ( x , y , z ) in X = I × J × K ,

(3.10) F i , j ( x , y , z ) F i , j ( x , y , z ) θ i x x + λ j y y ,

(3.11) F i , j ( x , y , z ) F i , j ( x , y , z ) α i , j z z ,

(3.12) F i , j ( x k , y l , z k l ) = z τ ( i , k ) , τ ( j , l ) k , l { 0 , } .

Remark 3.4

From (3.10), (3.11) and the following inequality

F i , j ( x , y , z ) F i , j ( x , y , z ) F i , j ( x , y , z ) F i , j ( x , y , z ) + F i , j ( x , y , z ) F i , j ( x , y , z ) ,

we deduce that, for each i , j N , F i , j is a Lipschitz continuous function.

We note that with some mild conditions on the constants involved thereat, (3.7)–(3.11) imply that the maps F i , j , i , j N , are equi-Lipschitz.

Proposition 3.5

Suppose that sup i θ i and sup j λ j are real numbers. Denote

β 2 max { sup i θ i , sup j λ j } , β 3 max { sup i θ i , sup j λ j , sup i , j α i , j } .

Then

F i , j ( x , y , z ) F i , j ( x , y , z ) β ( x , y ) ( x , y ) , ( x , y ) , ( x , y ) I × J , z K ,

and

F i , j ( x , y , z ) F i , j ( x , y , z ) β ( x , y , z ) ( x , y , z ) , ( x , y , z ) , ( x , y , z ) X ,

where · denotes the Euclidean norm on R 2 or R 3 .

Proof

We use the elementary inequality

x x + y y 2 ( x x ) 2 + ( y y ) 2

and (3.10) to obtain

F i , j ( x , y , z ) F i , j ( x , y , z ) β ( x , y ) ( x , y ) , ( x , y ) , ( x , y ) I × J , z K .

The inequality

x x + y y + z z 3 ( x x ) 2 + ( y y ) 2 + ( z z ) 2 ,

and (3.11) provide

F i , j ( x , y , z ) F i , j ( x , y , z ) β ( x , y , z ) ( x , y , z ) , ( x , y , z ) , ( x , y , z ) X ,

completing the proof.□

Recall that X = I × J × K . Having defined the maps u i , v j and F i , j , for each ( i , j ) N × N , we define W i , j : X X by

(3.13) W i , j ( x , y , z ) ( u i ( x ) , v j ( y ) , F i , j ( x , y , z ) ) .

Then { X , ( W i , j ) ( i , j ) N × N } is a CIFS. By definition, we have

(3.14) W i , j ( x k , y l , z k l ) = ( x τ ( i , k ) , y τ ( j , l ) , z τ ( i , k ) τ ( k , l ) ) for all ( k , l ) { 0 , } .

For ( x , y , z ) , ( x , y , z ) R 3 and δ > 0 , define the metric d δ as follows:

(3.15) d δ ( ( x , y , z ) , ( x , y , z ) ) ( x , y ) ( x , y ) + δ z z .

It can be easily verified that the metric d δ defined earlier is equivalent to the Euclidean metric on R 3 for all δ > 0 . This ensures that the metric space ( X , δ X ) is compact.

Proposition 3.6

The countable IFS { X , W i , j : ( i , j ) N × N } is hyperbolic with respect to the metric d δ defined in (3.15), with

δ inf i , j 1 c i , j 2 β ,

where β is as in the previous proposition. Hence, it possesses an attractor, that is, there exists a unique non-empty compact set A X such that A = W ( A ) = i , j 1 W i , j ( A ) ¯ .

Proof

For ( x , y , z ) , ( x , y , z ) X , we have

d δ ( W i , j ( x , y , z ) , W i , j ( x , y , z ) ) = ( u i ( x ) , v j ( y ) ) ( u i ( x ) , v j ( y ) ) + δ F i , j ( x , y , z ) F i , j ( x , y , z ) ( u i ( x ) , v j ( y ) ) ( u i ( x ) , v j ( y ) ) + δ [ F i , j ( x , y , z ) F i , j ( x , y , z ) + F i , j ( x , y , z ) F i , j ( x , y , z ) ] c i , j ( x , y ) ( x , y ) + δ [ θ i x x + λ j y y + α z z ] c i , j ( x , y ) ( x , y ) + δ [ β 2 x x + β 2 y y + α z z ] ( c i , j + β δ ) ( x , y ) ( x , y ) + δ α z z max { sup i , j ( c i , j + β δ ) , α } [ ( x , y ) ( x , y ) + δ z z ] .

By the choice of δ , we have sup i , j ( c i , j + β δ ) < 1 . Hence, for each ( i , j ) N × N , W i , j is a contraction with respect to the metric d δ . Consequently, the Hutchinson operator W : ( X ) ( X ) is a contraction map (see also Theorem 2.4). The rest of the proof follows from the Banach fixed point theorem.□

Assume that for each ( i , j ) N × N , the function F i , j : X K further satisfies the following, the so-called matching conditions:

  1. for all i N and x * = u i 1 ( x i ) = u i + 1 1 ( x i ) ,

    (3.16) F i , j ( x * , y , z ) = F i + 1 , j ( x * , y , z ) y J , z K , and

  2. for all j N and y * = v j 1 ( y j ) = v j + 1 1 ( y j ) ,

    (3.17) F i , j ( x , y * , z ) = F i , j + 1 ( x , y * , z ) x I , z K .

Furthermore, we assume that the following limits exist for each fixed ( x , y , z ) X .

(3.18) lim i F i , j ( x , y , z ) , j N and lim j F i , j ( x , y , z ) , i N .

Let us denote by C ( I × J ) the set of all real-valued continuous functions defined on I × J . Consider the set

C * ( I × J ) { h C ( I × J ) : h ( x k , y l ) = z k l for all k , l { 0 , } } ,

endowed with the uniform metric. It is plain to see that:

Lemma 3.7

The metric subspace C * ( I × J ) is closed in the complete space C ( I × J ) , and hence, it is complete.

For each h C * ( I × J ) , we define T ( h ) : I × J R as follows:

(3.19) T ( h ) ( x , y ) F i , j ( u i 1 ( x ) , v j 1 ( y ) , h ( u i 1 ( x ) , v j 1 ( y ) ) ) : if ( x , y ) I i × J j for some ( i , j ) N × N , lim j F i , j ( u i 1 ( x ) , d , h ( u i 1 ( x ) , d ) ) : if x I i for some i N , and y = d , lim i F i , j ( b , v j 1 ( y ) , h ( b , v j 1 ( y ) ) ) : if x = b , and y J j for some j N , z : if x = b and y = d .

In what follows, we intend to show that T : C * ( I × J ) C * ( I × J ) defined by h T ( h ) is well defined. We begin with a few comments on the definition of T ( h ) above.

Remark 3.8

The existence of limits in the definition of T ( h ) is guaranteed by (3.18). Furthermore, note that x i I i + 1 I i , it appears that T ( h ) ( x i , d ) receives two expressions, namely,

T ( h ) ( x i , d ) = lim j F i , j ( u i 1 ( x i ) , d , h ( u i 1 ( x i ) , d ) ) ,

T ( h ) ( x i , d ) = lim j F i + 1 , j ( u i + 1 1 ( x i ) , d , h ( u i + 1 1 ( x i ) , d ) ) .

However, condition (3.16) enables to determine T ( h ) ( x i , d ) uniquely. Similarly, T ( h ) ( b , y j ) is defined univocally for each j N .

Next we state an elementary result from analysis [40], crucial for the continuity arguments in the upcoming lemma, but omit the proof.

Lemma 3.9

Let f : ( a , b ) × ( c , d ) R and λ ( c , d ) . Then the following hold.

  1. lim ( x , y ) ( b , λ ) f ( x , y ) = L if and only if for any non-decreasing sequence ( r n ) in ( a , b ) , where r n b , and for any sequence ( t n ) in ( c , d ) , where t n λ , the sequence ( f ( r n , t n ) ) converges to L .

  2. lim ( x , y ) ( b , λ ) f ( x , y ) = L if and only if for any non-decreasing sequence ( r n ) in ( a , b ) , where r n b , and for any monotone sequence ( t n ) in ( c , d ) , where t n λ , the sequence ( f ( r n , t n ) ) converges to L .

  3. lim ( x , y ) ( b , d ) f ( x , y ) = L if and only if for every pair of non-decreasing sequences ( r n ) in ( a , b ) and ( t n ) in ( c , d ) such that r n b , t n d , the sequence ( f ( r n , t n ) ) converges to L .

Lemma 3.10

For each h C * ( I × J ) , the function T ( h ) : I × J R is well defined. Moreover, T ( h ) C * ( I × J ) whenever h C * ( I × J ) . That is, the operator T : C * ( I × J ) C * ( I × J ) is well defined.

Proof

It follows from (3.16) and (3.17) that T ( h ) is well defined on the boundary of I i × J j for all ( i , j ) N × N . For instance, let us note the following. Set y j 1 y y j . Then ( x i 1 , y ) ( I i 1 × J j ) ( I i × J j ) . Treating ( x i 1 , y ) as an element in I i 1 × J j , we have

(3.20) T ( h ) ( x i 1 , y ) = F i 1 , j ( u i 1 1 ( x i 1 ) , v j 1 ( y ) , h ( u i 1 1 ( x i 1 ) , v j 1 ( y ) ) ) .

On the other hand, treating ( x i 1 , y ) as an element in I i × J j , we obtain

(3.21) T ( h ) ( x i 1 , y ) = F i , j ( u i 1 ( x i 1 ) , v j 1 ( y ) , h ( u i 1 ( x i 1 ) , v j 1 ( y ) ) ) .

From (3.6), one has u i 1 1 ( x i 1 ) = s i 1 a + s i b = u i 1 ( x i 1 ) . Hence, using (3.16), it follows, according to (3.20) and (3.21), that T ( h ) ( x i 1 , y ) is uniquely determined.

It is obvious that T ( h ) is continuous on [ a , b ) × [ c , d ) . Next we will prove the continuity of T ( h ) at ( b , y ) , where y J . For this purpose, let us consider a sequence ( ( r n , t n ) ) in [ a , b ) × [ c , d ) such that ( r n ) is monotonically increasing and ( r n , t n ) ( b , y ) as n . We have to show that T ( h ) ( r n , t n ) T ( h ) ( b , y ) . Two cases can occur.

Case 1

Let y < d . Then y J j 0 [ c , d ) for some j 0 N .

For each n = 1 , 2 , , let i n , j n N be such that ( r n , t n ) I i n × J j n .

Subcase 1.1

Assume that y ( y j 0 1 , y j 0 ) . Since t n y , one can find N N such that t n J j 0 for all n N . Let n N . Then

(3.22) T ( h ) ( r n , t n ) = F i n , j 0 ( u i n 1 ( r n ) , v j 0 1 ( t n ) , h ( u i n 1 ( r n ) , v j 0 1 ( t n ) ) ) .

On the other hand,

(3.23) T ( h ) ( b , y ) = lim n F i n , j 0 ( b , v j 0 1 ( y ) , h ( b , v j 0 1 ( y ) ) ) .

In view of (3.18), equations (3.22) and (3.23) with some basic algebra of limits provide

(3.24) lim n T ( h ) ( r n , t n ) T ( h ) ( b , y ) = lim n F i n , j 0 ( u i n 1 ( r n ) , v j 0 1 ( t n ) , h ( u i n 1 ( r n ) , v j 0 1 ( t n ) ) ) F i n , j 0 ( b , v j 0 1 ( y ) , h ( b , v j 0 1 ( y ) ) ) lim n [ F i n , j 0 ( u i n 1 ( r n ) , v j 0 1 ( t n ) , h ( u i n 1 ( r n ) , v j 0 1 ( t n ) ) ) F i n , j 0 ( u i n 1 ( r n ) , v j 0 1 ( t n ) , h ( b , v j 0 1 ( y ) ) ) + F i n , j 0 ( u i n 1 ( r n ) , v j 0 1 ( t n ) , h ( b , v j 0 1 ( y ) ) ) F i n , j 0 ( b , v j 0 1 ( y ) , h ( b , v j 0 1 ( y ) ) ) ] lim n [ α i n , j 0 h ( u i n 1 ( r n ) , v j 0 1 ( t n ) ) h ( b , v j 0 1 ( y ) ) + θ i n u i n 1 ( r n ) b + λ j 0 v j 0 1 ( t n ) v j 0 1 ( y ) ] lim n [ 2 h α i n , j 0 + ( b a ) θ i n + λ j 0 v j 0 1 ( t n ) v j 0 1 ( y ) ] ,

where in the second step, we used the triangle inequality, the third step is consequent upon (3.10) and (3.11) and the last step is plain. Now using (3.7), (3.8), t n y as n , and the continuity of v j 0 1 , we have

lim n α i n , j 0 = 0 , lim n θ i n = 0 , lim n v j 0 1 ( t n ) v j 0 1 ( y ) = 0 .

Consequently, (3.24) yields lim n T ( h ) ( r n , t n ) T ( h ) ( b , y ) = 0 , and hence we infer that T ( h ) is continuous at ( b , y ) for y ( y j 0 1 , y j 0 ) .

Subcase 1.2

Let y = y j 0 1 . Let us suppose that ( t n ) is monotonically increasing and ( r n , t n ) ( b , y ) . Then one can find N 0 N such that t n ( y j 0 2 , y j 0 1 ] J j 0 1 for all n N 0 . Then computations similar to that in (3.24) provide

lim n T ( h ) ( r n , t n ) T ( h ) ( b , y ) = lim n F i n , j 0 1 ( u i n 1 ( r n ) , v j 0 1 1 ( t n ) , h ( u i n 1 ( r n ) , v j 0 1 1 ( t n ) ) ) F i n , j 0 1 ( b , v j 0 1 1 ( y ) , h ( b , v j 0 1 1 ( y ) ) ) lim n [ F i n , j 0 1 ( u i n 1 ( r n ) , v j 0 1 1 ( t n ) , h ( u i n 1 ( r n ) , v j 0 1 1 ( t n ) ) ) F i n , j 0 1 ( u i n 1 ( r n ) , v j 0 1 1 ( t n ) , h ( b , v j 0 1 1 ( y ) ) ) + F i n , j 0 1 ( u i n 1 ( r n ) , v j 0 1 1 ( t n ) , h ( b , v j 0 1 1 ( y ) ) ) F i n , j 0 1 ( b , v j 0 1 1 ( y ) , h ( b , v j 0 1 1 ( y ) ) ) ] lim n [ α i n , j 0 1 h ( u i n 1 ( r n ) , v j 0 1 1 ( t n ) ) h ( b , v j 0 1 1 ( y ) ) + θ i n u i n 1 ( r n ) b + λ j 0 1 v j 0 1 1 ( t n ) v j 0 1 1 ( y ) ] lim n [ 2 h α i n , j 0 1 + ( b a ) θ i n + λ j 0 1 v j 0 1 1 ( t n ) v j 0 1 1 ( y ) ] = 0 .

Similarly, if ( t n ) n N is monotonically decreasing, then there exists N 0 N such that t n [ y j 0 1 , y j 0 ) J j 0 for all n N 0 . By using a similar argument as earlier, we can prove that

lim n T ( h ) ( r n , t n ) T ( h ) ( b , y ) = 0 .

By combining the two aforementioned subcases, we conclude that T ( h ) is continuous at ( b , y ) for any y [ c , d ) .

Case 2

Let y = d . In this case, due to the previous lemma, we can assume that both ( r n ) and ( t n ) are monotonically increasing. For each n = 1 , 2 , , let i n , j n N be such that r n I i n , t n J j n . By using (3.7), (3.11), and (3.12), we have as mentioned earlier

lim n T ( h ) ( r n , t n ) T ( h ) ( b , d ) lim n [ F i n , j n ( u i n 1 ( r n ) , v j n 1 ( t n ) , h ( u i n 1 ( r n ) , v j n 1 ( t n ) ) ) F i n , j n ( b , d , h ( u i n 1 ( r n ) , v j n 1 ( t n ) ) ) + F i n , j n ( b , d , h ( u i n 1 ( r n ) , v j n 1 ( t n ) ) ) F i n , j n ( b , d , h ( b , d ) ) + F i n , j n ( b , d , h ( b , d ) ) z ] lim n [ θ i n u i n 1 ( r n ) b + λ j n v j n 1 ( t n ) d + α i n , j n h ( u i n 1 ( r n ) , v j n 1 ( t n ) ) h ( b , d ) ] + lim n F i n , j n ( b , d , z ) z lim n [ θ i n ( b a ) + λ j n ( d c ) + 2 α i n , j n h ] + lim n z τ ( i n , ) , τ ( j n , ) z = 0 .

Similarly, we can prove that T ( h ) is continuous at ( x , d ) for all x [ a , b ) . We next prove that T ( h ) interpolates the given bivariate CSD. To this end, let us note that for every ( i , j ) N × N , by the condition on τ given in (3.1), we can choose ( k , l ) { 0 , } × { 0 , } such that i = τ ( i , k ) and j = τ ( j , l ) . By (3.2), it follows that x k = u i 1 ( x i ) and y l = v j 1 ( y j ) . Using (3.12) and (3.19),

T ( h ) ( x i , y j ) = F i , j ( x k , y l , h ( x k , y l ) ) = F i , j ( x k , y l , z k l ) = z τ ( i , k ) , τ ( j , l ) = z i , j .

The continuity of T ( h ) further implies T ( h ) ( x k , y l ) = z k l for k , l { 0 , } . Consequently, T maps C * ( I × J ) into itself. This completes the proof.□

Theorem 3.11

Let { X , W i , j : ( i , j ) N × N } be the CIFS defined through (3.1)–(3.13)and (3.16)–(3.18). Then there exists a unique continuous function g : I × J R such that g ( x i , y j ) = z i , j for all i , j N 0 × N 0 , and the graph of g, namely,

G = { ( x , y , g ( x , y ) ) : ( x , y ) I × J } ,

is the attractor of the CIFS defined earlier.

Proof

By the aforementioned lemma, we have T : C * ( I × J ) C * ( I × J ) is well defined. We shall prove that it is, in fact, a contraction map. To this end, let h 1 , h 2 C * ( I × J ) . For ( x , y ) [ a , b ) × [ c , d ) , choose ( i , j ) N × N such that ( x , y ) I i × J j

T ( h 1 ) ( x , y ) T ( h 2 ) ( x , y ) = F i , j ( u i 1 ( x ) , v j 1 ( y ) , h 1 ( u i 1 ( x ) , v j 1 ( y ) ) ) F i , j ( u i 1 ( x ) , v j 1 ( y ) , h 2 ( u i 1 ( x ) , v j 1 ( y ) ) ) α i , j h 1 ( u i 1 ( x ) , v j 1 ( y ) ) h 2 ( u i 1 ( x ) , v j 1 ( y ) ) α h 1 h 2 .

By the continuity of T ( h 1 ) and T ( h 2 ) , the aforementioned inequality holds for all other ( x , y ) I × J as well. Therefore,

T ( h 1 ) T ( h 2 ) α h 1 h 2 .

That is, T is a contraction map. In view of Lemma 3.7, the Banach fixed point theorem now ensures that there exists a unique g C * ( I × J ) satisfying

(3.25) T ( g ) = g .

Now, let G = { ( x , y , g ( x , y ) ) : ( x , y ) I × J } be the graph of g and G ˜ { ( x , y , g ( x , y ) ) : ( x , y ) [ a , b ) × [ c , d ) } G . For ( x , y ) [ a , b ) × [ c , d ) , there exist i 0 , j 0 N × N such that ( x , y ) I i 0 × J j 0 . Thus, by (3.25),

( x , y , g ( x , y ) ) = ( x , y , F i 0 j 0 ( u i 0 1 ( x ) , v j 0 1 ( y ) , g ( u i 0 1 ( x ) , v j 0 1 ( y ) ) ) ) = W i 0 j 0 ( u i 0 1 ( x ) , v j 0 1 ( y ) , g ( u i 0 1 ( x ) , v j 0 1 ( y ) ) ) W i 0 j 0 ( G ) .

Hence, G ˜ i , j 1 W i , j ( G ) . Taking closure on both sides, we obtain G i , j 1 W i , j ( G ) ¯ .

Conversely, for ( x , y ) I × J and ( i , j ) N × N , we have u i ( x ) I i and v j ( y ) J j . Therefore

W i , j ( x , y , g ( x , y ) ) = ( u i ( x ) , v j ( y ) , F i , j ( x , y , g ( x , y ) ) ) = ( u i ( x ) , v j ( y ) , g ( u i ( x ) , v j ( y ) ) ) G .

Since G is closed, one has

i , j 1 W i , j ( G ) ¯ G ,

and so,

G = i , j 1 W i , j ( G ) ¯ .

This concludes the proof.□

Remark 3.12

As mentioned in Section 1, the motivation for the aforementioned construction of FIS corresponding to a countably infinite data set comes from the construction of FIS in the case of finite data set given by Ruan and Xu [31]. The interested reader is encouraged to compare the above construction with that in [31, Theorem 3.1], in particular, the conditions imposed on the maps F i , j and the definition of the Read-Bajraktarević operator T . In the setting of a countable bivariate data set, establishing the continuity of T is a more subtle matter, for which relatively stronger conditions that we imposed on F i , j came to our aid. On the other hand, for the construction of a parameterized family of fractal functions, which forms the subject matter of the second part of this article, these assumptions on F i , j are easier to come by.

3.2 Approximation of the attractor of countable IFS

Let ( X , d ) be a compact metric space. For ( i , j ) N × N , let ω i , j : X X be contraction maps with contractivity factor r i , j such that sup i , j r i , j < 1 . Consider the CIFS { X , ( ω i , j ) i , j N } . Assume that m , n N . Let us refer to { X , ( ω i , j ) ( i , j ) N m × N n } as the partial IFS associated with the CIFS { X , ( ω i , j ) i , j N } . We will denote the attractor of the aforesaid partial IFS by A m , n . Our aim is to prove that the double sequence ( A m , n ) m , n N provides an approximation of the attractor of the CIFS in the space ( X ) endowed with the Hausdorff-Pompeiu metric.

Let us begin with the following theorem:

Theorem 3.13

Let ( X , d ) be a complete metric space and ( K m , n ) m , n N be a double sequence of compact subsets of X.

  1. Assume that K m , n K m + 1 , n and K m , n K m , n + 1 for all m , n N . If the set K m , n 1 K m , n is relatively compact, then

    K ¯ = m , n 1 K m , n ¯ = lim m , n K m , n .

  2. If K m + 1 , n K m , n and K m , n + 1 K m , n for all m , n N , then

    C m , n 1 K m , n = lim m , n K m , n .

Proof

It can be proved similar to [34, Lemma 2.4].□

Lemma 3.14

For every m , n N , we have A m , n A m + 1 , n , A m , n A m , n + 1 and A m , n A m + 1 , n + 1 .

Proof

Choose i 1 , i 2 , , i p { 1 , 2 , , m } , j 1 , j 2 , , j q { 1 , 2 , , n } . Define the functions ω i 1 i 2 i p , j 1 j 2 j q : X X by

ω i 1 i 2 i p , j 1 j 2 j q ( x ) = ω i 1 j 1 ω i p j q ( x ) .

Obviously, ω i 1 i 2 i p , j 1 j 2 j q is a contraction map with contractivity factor at most k = 1 p l = 1 q r i k j l < 1 . Let a i 1 i 2 i p , j 1 j 2 j q be its unique fixed point, then following [2, § 3.1 (3)], we have

A m , n = { a i 1 i 2 i p , j 1 j 2 j q : i 1 , i 2 , , i p N m , j 1 , j 2 , , j q N n } ¯ .

The proof follows immediately from the aforementioned observation.□

Lemma 3.15

[34, Lemma  2.2] If ( E λ ) λ Λ is a family of subsets of a topological space, then

λ Λ E λ ¯ ¯ = λ Λ E λ ¯ .

The following theorem is analogous to [34, Prop. 3.6].

Theorem 3.16

The set A = m , n 1 A m , n ¯ in ( X ) is the attractor of the CIFS { X , ( ω i , j ) i , j N } .

Proof

For every m , n N , we have

(3.26) i , j 1 ω i , j ( A m , n ) ¯ = i = 1 m j = 1 n ω i , j ( A m , n ) ¯ i = m + 1 i = n + 1 ω i , j ( A m , n ) ¯ = A m , n i = m + 1 i = n + 1 ω i , j ( A m , n ) ¯ .

By Lemma 3.15, we have

(3.27) i , j 1 m , n 1 ω i , j ( A m , n ) ¯ = m , n 1 i , j 1 ω i , j ( A m , n ) ¯ = m , n 1 i , j 1 ω i , j ( A m , n ) ¯ ¯ = m , n 1 ( A m , n i m + 1 j n + 1 ω i , j ( A m , n ) ¯ ) ¯ = m , n 1 A m , n ¯ m , n 1 i m + 1 j n + 1 ω i , j ( A m , n ) ¯ .

By Lemma 3.14, we obtain

(3.28) m , n 1 i m + 1 j n + 1 ω i , j ( A m , n ) ¯ m , n 1 A m , n ¯ .

By combining (3.27) and (3.28), we obtain

(3.29) i , j 1 m , n 1 ω i , j ( A m , n ) ¯ = m , n 1 A m , n ¯ .

Now, using the continuity of ω i , j for each i , j N , (3.29) and Lemma 3.15, we have

(3.30) i , j 1 ω i , j ( m , n 1 A m , n ¯ ) ¯ i , j 1 ω i , j ( m , n 1 A m , n ) ¯ ¯ = i , j 1 ω i , j ( m , n 1 A m , n ) ¯ = i , j 1 m , n 1 ω i , j ( A m , n ) ¯ = m , n 1 A m , n ¯ .

Conversely, by using (3.29), we have

(3.31) m , n 1 A m , n ¯ = i , j 1 m , n 1 ω i , j ( A m , n ) ¯ = i , j 1 ω i , j ( m , n 1 A m , n ) ¯ i , j 1 ω i , j ( m , n 1 A m , n ¯ ) ¯ .

Equations (3.30) and (3.31) together yield the following:

i , j 1 ω i , j ( m , n 1 A m , n ¯ ) ¯ = m , n 1 A m , n ¯ .

Since X is compact, and A = m , n 1 A m , n ¯ X is closed, it follows that A is the attractor.□

The aforementioned theorem in conjunction with Theorem 3.13 and Lemma 3.14 provides the promised approximation of the attractor of CIFS by the attractors of the partial IFSs. To be more precise, we have the following result.

Theorem 3.17

Let A be the attractor of the CIFS { X , ( ω i , j ) i , j N } and ( A m , n ) be the double sequence of attractors of the associated partial IFSs { X , ( ω i , j ) ( i , j ) N m × N n } . Then

A = lim m , n A m , n ,

where the limit is taken with respect to the Hausdorff-Pompeiu metric.

As a special case of the previous theorem, we have the following.

Corollary 3.18

Let us consider the CIFS { X , ( W i , j ) i , j N } defined in the construction of our countable FIS (see (3.13)) and denote by G its attractor obtained in Proposition 3.6. For m , n N , let G m , n be the attractor of the partial IFS { X , ( W i , j ) ( i , j ) N m × N n } . Then

lim m , n G m , n = G ,

where the limit is taken with respect to the Hausdorff-Pompeiu metric.

4 A parameterized family of bivariate fractal functions and associated fractal operator

As mentioned in the introduction, in order to explore some approximation theoretic aspects, we consider here a special case of the countable bivariate FIF constructed in the previous section. This is influenced by the notion of α -fractal function [13], an offspring of the univariate FIF for a finite data set.

Definition 4.1

Let I × J = [ a , b ] × [ c , d ] R 2 . We say that Δ = { x i : i = 0 , 1 , 2 , } × { y j : j = 0 , 1 , 2 , } I × J is a partition of I × J if

  1. the sequences ( x i ) i N 0 and ( y j ) i N 0 are strictly increasing such that x 0 = a and y 0 = c ,

  2. lim i x i = b x and lim j y j = d y .

Let Lip ( I × J ) C ( I × J ) denote the set of all Lipschitz continuous real-valued functions defined on I × J . That is, f Lip ( I × J ) if there exists a constant l f > 0 such that

f ( x , y ) f ( s , t ) l f ( x , y ) ( s , t ) ( x , y ) , ( s , t ) I × J .

Definition 4.2

Let α Lip ( I × J ) and Δ be a partition of I × J . For each ( i , j ) N × N , define I i = [ x i 1 , x i ] , J j = [ y j 1 , y j ] and set

α i , j = α , I i × J j sup ( x , y ) I i × J j α ( x , y ) .

We say that α is a scaling function if the following conditions are satisfied:

  1. lim i , j α i , j = lim i ( sup j α i , j ) = lim j ( sup i α i , j ) = 0 ;

  2. α sup ( x , y ) I × J α ( x , y ) = sup i , j α i , j < 1 .

Remark 4.3

The aforementioned conditions imposed on the scaling function α coincide with those in (3.8).

Fix f Lip ( I × J ) . We refer to f as the germ function or seed function. Consider the countable data set

D = { ( x i , y j , f ( x i , y j ) ) : i , j N 0 } ,

where Δ = { ( x i , y j ) : i , j N 0 } is a partition of I × J . We construct a class of bivariate FIFs corresponding to the countable data D = { ( x i , y j , f ( x i , y j ) ) : i , j N 0 } by choosing appropriate maps u i , v j and F i , j that constitute the countable IFS defined in the previous section.

For i N { 0 } , as mentioned earlier, let us denote s i 1 + ( 1 ) i 2 .

  1. Consider the affine function u i : I I i satisfying (3.2). That is,

    u i ( a ) = s i 1 x i 1 + s i x i , u i ( b ) = s i x i 1 + s i + 1 x i , u i ( x ) = a i x + c i x I .

  2. For j N , we consider the affine function v j : J J j satisfying (3.3), that is,

    v j ( c ) = s j 1 y j 1 + s j y j , v j ( d ) = s j y j 1 + s j + 1 y j , v j ( y ) = b j y + d j y J .

  3. Assume that L : Lip ( I × J ) Lip ( I × J ) is an operator satisfying the boundary conditions

    (4.1) L ( f ) ( x k , y l ) = f ( x k , y l ) for all k , l { 0 , } .

Let K be a sufficiently large compact interval containing the set { f ( x i , y j ) : i , j = 0 , 1 , 2 , } and X = I × J × K . Choose k > 0 such that

z k z K .

For i , j N , define F i , j : X K by

(4.2) F i , j ( x , y , z ) α ( u i ( x ) , v j ( y ) ) z + f ( u i ( x ) , v j ( y ) ) α ( u i ( x ) , v j ( y ) ) L ( f ) ( x , y ) .

Let us consider the CIFS { X , W i , j : ( i , j ) N × N } , where

W i , j ( x , y , z ) = ( u i ( x ) , v j ( y ) , F i , j ( x , y , z ) )

as in the previous section. In what follows, we will denote by l g the Lipschitz constant of a function g Lip ( I × J ) .

Theorem 4.4

Assume that the partition Δ , scaling function α and operator L are fixed. Then corresponding to each f Lip ( I × J ) , there exists a unique continuous function f * : I × J R such that

  1. f * interpolates f at the points in Δ , that is, f * ( x i , y j ) = f ( x i , y j ) for all ( x i , y j ) Δ ,

  2. the graph of f * is the attractor of the CIFS { X , W i , j : ( i , j ) N × N } defined earlier.

Furthermore, f * satisfies

(4.3) f * ( x , y ) = f ( x , y ) + α ( x , y ) ( f * L ( f ) ) ( u i 1 ( x ) , v j 1 ( y ) ) if ( x , y ) I i × J j for s o m e ( i , j ) N × N , f ( x , d ) + α ( x , d ) ( f * L ( f ) ) ( u i 1 ( x ) , d ) if x I i for s o m e i N , and y = d , f ( b , y ) + α ( b , y ) ( f * L ( f ) ) ( b , v j 1 ( y ) ) , if x = b , and y J j for s o m e j N , f ( b , d ) , if x = b and y = d . .

Proof

Consider the countable data set

D = { ( x i , y j , f ( x i , y j ) ) : i , j N 0 } .

We shall first prove that the functions F i , j satisfy (3.10)–(3.11), and the matching conditions (3.16)–(3.17) prescribed in Theorem 3.11. To this end, let ( x , y , z ) , ( x , y , z ) X = I × J × K , we have

F i , j ( x , y , z ) F i , j ( x , y , z ) = [ α ( u i ( x ) , v j ( y ) ) z + f ( u i ( x ) , v j ( y ) ) α ( u i ( x ) , v j ( y ) ) L ( f ) ( x , y ) ] [ α ( u i ( x ) , v j ( y ) ) z + f ( u i ( x ) , v j ( y ) ) α ( u i ( x ) , v j ( y ) ) L ( f ) ( x , y ) ] 2 k l α [ a i x x + b j y y ] + 2 l f [ a i x x + b j y y ] + 2 α i , j l L ( f ) [ x x + y y ] + 2 L ( f ) l α [ a i x x + b j y y ] [ 2 a i ( k l α + l f + l α L ( f ) ) + 2 l L ( f ) sup j α i , j ] x x + [ 2 b j ( k l α + l f + l α L ( f ) ) + 2 l L ( f ) sup i α i , j ] y y .

Thus, the map F i , j given in (4.2) satisfies:

  1. condition (3.10) with

    θ i = 2 a i ( k l α + l f + l α L ( f ) ) + 2 l L ( f ) sup j α i , j , λ j = 2 b j ( k l α + l f + l α L ( f ) ) + 2 l L ( f ) sup i α i , j ;

  2. condition (3.11) with

    α i , j = α , I i × J j sup ( x , y ) I i × J j α ( x , y ) ;

  3. conditions required in (3.7), since by the definition of scaling function

    lim i , j α i , j = sup i lim j α i , j = sup j lim i α i , j = 0 ;

    and since a i = x i x i 1 b a , b j = y j y j 1 d c provide

    lim i a i = lim j b j = 0 ;

  4. interpolation condition at the four vertices of the rectangle I × J , given in (3.12), with

    F i , j ( x k , y l , f ( x k , y l ) ) = f ( x τ ( i , k ) , y τ ( j , l ) ) k , l { 0 , } ;

  5. matching conditions (3.16)–(3.17);

  6. the existence of limits in (3.18).

Consequently, by Theorem 3.11, there exists a continuous function f * interpolating the data set D , and the graph of f * is the attractor of the CIFS { X , ( W i , j ) i , j N } . The self-referential equation for f * also follows from Theorem 3.11.□

Definition 4.5

Emphasizing its dependence on the function α , the function f * is referred to as the (countable bivariate) α -fractal function associated to the germ function f , with respect to the parameters α , Δ and L . To indicate its dependence with parameters α , Δ and L , we shall denote f * by f Δ , L α .

Remark 4.6

In fact, we obtain a family of fractal functions { f Δ , L α } corresponding to each germ function f , obtained for different choices of parameters α , Δ , and L . Note that each member of the family { f Δ , L α } interpolates the germ function f at points in Δ , that is,

f Δ , L α ( x i , y j ) = f ( x i , y j ) i , j N 0 , and f Δ , L α ( x k , y l ) = f ( x k , y l ) for k , l { 0 , } .

Remark 4.7

The fundamental impetus for the definition of parameterized family of fractal functions reported above was Navascues’s construction of the α -fractal function, which is widely recognized in the approximation theory of univariate fractal functions (see, for instance, [13,14]). The difference here is that the germ function is bivariate and that the sampling of the germ function is done at a countably infinite number of points in the domain. Also, we do not make a standing assumption that L is a bounded linear operator.

As in the context of univariate α -fractal function, we consider the self-referential function f * = f Δ , L α as the image of a given (non-fractal) function f under an operator Δ , L α , thus associating with f its self-referential counterpart f Δ , L α . More precisely, we have

Definition 4.8

Let α , Δ , and L be fixed. The operator

Δ , L α : Lip ( I × J ) C ( I × J ) C ( I × J ) , Δ , L α ( f ) f Δ , L α ,

which assigns to each f Lip ( I × J ) its self-referential counterpart f Δ , L α is called the α -fractal operator on Lip ( I × J ) .

The following results, which point to the error committed in “fractalizing” the germ function f , is well known in the univariate setting [13].

Proposition 4.9

Let f Lip ( I × J ) be the germ function and f Δ , L α be the α -fractal function associated with f corresponding to the partition Δ , scale function α , and parameter map L. Then we have the following inequality:

f Δ , L α f α f Δ , L α L ( f ) .

Proof

Choose ( x , y ) I i × J j for some ( i , j ) N × N . From the self-referential equation (4.3), we have

f Δ , L α ( x , y ) f ( x , y ) = α ( x , y ) ( f Δ , L α L ( f ) ) ( u i 1 ( x ) , v j 1 ( y ) ) α f Δ , L α L ( f ) .

If x I i for some i N and y = d , then

f Δ , L α ( x , d ) f ( x , d ) = α ( x , d ) ( f Δ , L α L ( f ) ) ( u i 1 ( x ) , d ) α f Δ , L α L ( f ) .

Similarly, if x = b and y J j for some j N , one has

f Δ , L α ( b , y ) f ( b , y ) α f L ( f ) .

Also, f Δ , L α ( b , d ) = f ( b , d ) . Consequently,

f Δ , L α f α f Δ , L α L ( f ) ,

completing the proof.□

Corollary 4.10

Let f Lip ( I × J ) be the germ function and f Δ , L α be the α -fractal function associated with f with respect to the parameters α , Δ , and L . Then we have the following inequality:

f Δ , L α L ( f ) 1 1 α f L ( f ) .

In particular, if L = I d , the identity operator on Lip ( I × J ) , then Δ , L α = I d .

Proof

We have

f Δ , L α L ( f ) = f Δ , L α f + f L ( f ) f Δ , L α f + f L ( f ) α f Δ , L α L ( f ) + f L ( f ) .

Therefore,

f Δ , L α L ( f ) 1 1 α f L ( f ) ,

completing the proof.□

Corollary 4.11

Let f Lip ( I × J ) be the germ function and f Δ , L α be the α -fractal function associated with f corresponding to α , Δ , and L. Then we have the following inequality:

f Δ , L α f α 1 α f L ( f ) .

Proof

From Proposition 4.9 and the triangle inequality

f Δ , L α f α f Δ , L α L ( f ) α ( f Δ , L α f + f L ( f ) )

proving the claim.□

As an immediate consequence of the previous corollary, we obtain sequences of fractal functions converging uniformly to a prescribed bivariate Lipschitz continuous function, as specified in the upcoming result.

Corollary 4.12

Let f Lip ( I × J ) be the germ function.

  1. Assume that the partition Δ and scale function α are fixed. Let ( L n ) n N be a sequence of operators on Lip( I × J ) such that L n ( f ) ( x k , y l ) = f ( x k , y l ) for k , l { 0 , } and for each n N . We further assume that L n ( f ) f uniformly. Then the sequence ( f Δ , L n α ) n N of α -fractal functions associated to f converges uniformly to f.

  2. Assume that the partition Δ and the operator L : Lip ( I × J ) Lip ( I × J ) are fixed. Let ( α n ) n N be a sequence of scale functions such that α n 0 as n . Then the sequence ( f Δ , L α n ) n N converges uniformly to f.

The proof of the following proposition is similar to its univariate counterpart in [13], but included here at the referee’s behest.

Proposition 4.13

If L : Lip ( I × J ) Lip ( I × J ) is a linear operator, then Δ , L α is a linear operator.

Proof

Let f , g Lip ( I × J ) and c 1 , c 2 R . By using the functional equation for f Δ , L α and g Δ , L α , we have

f Δ , L α ( x , y ) = f ( x , y ) + α ( x , y ) ( f Δ , L α L ( f ) ) ( u i 1 ( x ) , v j 1 ( y ) ) ,

g Δ , L α ( x , y ) = g ( x , y ) + α ( x , y ) ( g Δ , L α L ( f ) ) ( u i 1 ( x ) , v j 1 ( y ) ) ,

for all ( x , y ) I i × J j . On multiplying the aforementioned expressions by c 1 and c 2 , respectively, and adding, we obtain

( c 1 f Δ , L α + c 2 g Δ , L α ) ( x , y ) = ( c 1 f + c 2 g ) ( x , y ) + α ( x , y ) ( c 1 f Δ , L α + c 2 g Δ , L α L ( c 1 f + c 2 g ) ) ( u i 1 ( x ) , v j 1 ( y ) ) ,

for all ( x , y ) I i × J j , where ( i , j ) N × N . Similar expressions for the other points ( x , y ) I × J . This shows that ( c 1 f Δ , L α + c 2 g Δ , L α ) is the fixed point of the RB-operator associated with the construction of ( c 1 f + c 2 g ) Δ , L α . Hence, by the uniqueness of the fixed point of the RB-operator, we obtain

( c 1 f Δ , L α + c 2 g Δ , L α ) = ( c 1 f + c 2 g ) Δ , L α ,

as desired.□

Remark 4.14

As noted earlier, when the germ function is univariate and Δ is finite, the notion of α -fractal function and associated fractal operator are well studied. Further, in the literature, the fractal operator is studied with the standing assumption that the parameter map L is a bounded linear operator [13,14,18,19,33]. Consequently, the fractal operator is widely investigated only within the confines of the theory of bounded linear operators. Here, we do not assume L to be linear or bounded, thereby enhancing the scope of the fractal operator.

Proposition 4.15

Let Lip ( I × J ) C ( I × J ) be endowed with the uniform norm. If L : Lip ( I × J ) Lip ( I × J ) is a continuous operator (not necessarily linear), then so is the operator Δ , L α .

Proof

Let ( f n ) n N be a sequence of Lipschitz functions such that f n f in Lip ( I × J ) . By using the self-referential equation and by routine computations, we have

Δ , L α ( f n ) Δ , L α ( f ) 1 1 α f n f + α 1 α L ( f n ) L ( f ) .

The aforementioned inequality in conjunction with the convergence of ( f n ) n N and the continuity of L establishes that Δ , L α ( f n ) Δ , L α ( f ) . This guarantees the continuity of Δ , L α .□

Proposition 4.16

The α -fractal operator Δ , L α : Lip ( I × J ) C ( I × J ) is an L-bounded operator with the L-bound not exceeding α 1 α .

Proof

According to Corollary 4.11, we have

Δ , L α ( f ) f α 1 α ( f + L ( f ) ) .

Thus,

Δ , L α ( f ) 1 1 α f + α 1 α L ( f ) ,

proving the claim.□

Corollary 4.17

If L is a norm-bounded nonlinear operator, that is, p ( L ) < , then Δ , L α is also norm-bounded.

Proof

From the previous proposition, we have

Δ , L α ( f ) 1 1 α f + α 1 α L ( f ) .

Since L is norm-bounded, we have

sup f Lip ( I × J ) , f 0 L ( f ) f < p ( L ) < .

Consequently,

sup f Lip ( I × J ) , f 0 Δ , L α ( f ) f 1 1 α + α 1 α p ( L ) .

Hence, we have

p ( Δ , L α ) max sup f Lip ( I × J ) , f 0 Δ , L α ( f ) f , Δ , L α ( 0 ) < ,

as required.□

Similarly, one can prove the following.

Corollary 4.18

If L is a topologically bounded operator, then so is the fractal operator Δ , L α .

Proposition 4.19

The operator Δ , L α is L-Lipschitz with the L-Lipschitz constant less than or equal to α 1 α .

Proof

By using similar computations as in Proposition 4.15, one obtains

(4.4) Δ , L α ( f ) Δ , L α ( g ) 1 1 α f g + α 1 α L ( f ) L ( g ) ,

proving the claim.□

Corollary 4.20

If L : Lip ( I × J ) C ( I × J ) Lip ( I × J ) is a Lipschitz operator, then so is the fractal operator Δ , L α . Furthermore,

Δ , L α 1 + α L 1 α .

Proof

The assertion follows clearly from the aforementioned computations.□

Proposition 4.21

If L : Lip ( I × J ) C ( I × J ) Lip ( I × J ) is a Cauchy-continuous operator (that is, L maps Cauchy sequences to Cauchy sequences), then Δ , L α is not a closed operator.

Proof

Let us assume on the contrary that the fractal operator Δ , L α is closed. Choose a function f C ( I × J ) \ Lip ( I × J ) , and consider a sequence ( f n ) n N in Lip ( I × J ) such that f n f as n . Since L is a Cauchy-continuous operator and ( f n ) n N is a Cauchy sequence in Lip ( I × J ) , it follows that ( L ( f n ) ) n N is a Cauchy sequence. By using (4.4) with f = f m and g = f n , one can infer that ( Δ , L α ( f n ) ) n N is a Cauchy sequence in the Banach space C ( I × J ) . Assume that Δ , L α ( f n ) g . By the fact that Δ , L α is closed, it follows that f Lip ( I × J ) and g = Δ , L α ( f ) , which contradicts the choice of f .□

Proposition 4.22

If L : Lip ( I × J ) Lip ( I × J ) is a closed operator, then the operator Δ , L α is L-closed.

Proof

Let ( f n ) n N be a sequence of Lipschitz functions such that f n f , L ( f n ) g and Δ , L α ( f n ) h as n . Since L is a closed operator, f n f , L ( f n ) g together imply f Lip ( I × J ) and g = L ( f ) . By (4.4), we have

Δ , L α ( f n ) Δ , L α ( f ) 1 1 α f n f + α 1 α L ( f n ) L ( f ) .

Consequently, Δ , L α ( f n ) Δ , L α ( f ) as n . By the uniqueness of the limit, h = Δ , L α ( f ) and hence, the assertion.□

Similarly, one can prove the following result.

Proposition 4.23

If L : Lip ( I × J ) Lip ( I × J ) is a closable operator, then the operator Δ , L α is L-closable.

5 Extension of fractal operator and some properties

In this short section, we extend the fractal operator Δ , L α to the whole of C ( I × J ) , and we shall refer to this extension operator as the α -fractal operator on C ( I × J ) . The following lemma is a standard result [41].

Lemma 5.1

Let A : D ( A ) X Y be a Lipschitz operator with Lipschitz constant A , where X , Y are metric spaces, the second one being complete. Then there exists a Lipschitz extension A ˜ : D ( A ) ¯ X Y of A such that A ˜ = A .

The operators L : Lip ( I × J ) C ( I × J ) and Δ , L α : Lip ( I × J ) C ( I × J ) are densely defined. Therefore, if L is a Lipschitz operator, then, by the aforementioned lemma and Corollary 4.20, we have Lipschitz extensions L ˜ : C ( I × J ) C ( I × J ) of L and ˜ Δ , L α : C ( I × J ) C ( I × J ) of Δ , L α preserving their respective Lipschitz constants. By a slight abuse of notation, we denote this extension of the fractal operator Δ , L α also by Δ , L α . This observation is formally recorded in the following proposition.

Proposition 5.2

If L : Lip ( I × J ) C ( I × J ) Lip ( I × J ) , is a Lipschitz operator, then the α -fractal operator Δ , L α : Lip ( I × J ) C ( I × J ) C ( I × J ) has a Lipschitz extension Δ , L α : C ( I × J ) C ( I × J ) with Lipschitz constant

Δ , L α 1 + α L 1 α .

Definition 5.3

For a prescribed germ function f C ( I × J ) , the function f Δ , L α = Δ , L α ( f ) , where Δ , L α is the Lipschitz extension in the previous proposition, is called the (countable bivariate) α -fractal function associated to the germ function f with respect to the parameters α , Δ , and L .

Remark 5.4

If L is a bounded linear operator, then Δ , L α : C ( I × J ) C ( I × J ) is also a bounded linear operator. In the setting of univariate functions, the bounded linear fractal operator Δ , L α : C ( I × J ) C ( I × J ) is well studied [13,14,16,19].

The notion of invariant subspace is fundamental in operator theory. We next provide a class of proper closed invariant subspaces for the bounded linear fractal operator Δ , L α : C ( I × J ) C ( I × J ) . Recall that the dual of C ( [ a , b ] × [ c , d ] ) is isomorphic to the space of all Borel measures equipped with total variation norm, and it is a non-separable Banach space; see [42, §IV.6.3, Theorem 3] for details. Consider the non-zero linear functional ψ a , c ( C ( [ a , b ] × [ c , d ] ) ) * given by ψ ( a , c ) ( f ) = f ( a , c ) for all f C ( [ a , b ] × [ c , d ] ) .

Theorem 5.5

The subspace W ( x i , y j ) { f C ( [ x 0 , x ] × [ y 0 , y ] ) : f ( x i , y j ) = 0 } is invariant for the fractal operator Δ , L α for all i , j N 0 { } , for permissible choices of α , Δ and bounded linear operator L.

Proof

For any partition Δ , scale vector α , and bounded linear operator L and any f C ( [ x 0 , x ] × [ y 0 , y ] ) , we have

( Δ , L α ) * ( ψ ( x i , y j ) ) ( f ) = ψ ( x i , y j ) Δ , L α ( f ) = ψ ( x i , y j ) ( f Δ , L α ) = f Δ , L α ( x i , y j ) = f ( x i , y j ) = ψ ( x i , y j ) ( f ) .

Consequently,

( ( Δ , L α ) * ) m ( ψ ( x i , y j ) ) = ψ ( x i , y j ) m N ,

and hence, by Lemma 2.20,

span { ψ ( x i , y j ) , ( Δ , L α ) * ( ψ ( x i , y j ) ) , ( ( Δ , L α ) * ) 2 ( ψ ( x i , y j ) ) , } ¯ = span { ψ ( x i , y j ) }

is a non-trivial closed invariant subspace of ( C ( [ x 0 , x ] × [ y 0 , y ] ) ) * . Now, from Lemma 2.21, it follows that

( span { ψ ( x i , y j ) } ) = W ( x i , y j )

is a non-trivial closed invariant subspace for Δ , L α .□

Remark 5.6

Along with the proof of the previous theorem, it is perhaps worth recalling that the closed invariant subspace Y in Lemma 2.21 can be trivial. For example, consider the space X = 1 and take Y = c 0 , the space of all real sequences convergent to zero. Then, Y is a closed invariant subspace of the identity operator I d : . However, the preannihilator of Y is zero.

Theorem 5.7

Let Δ = { x i : i N 0 } × { y j : j N 0 } be a partition of the rectangle [ a , b ] × [ c , d ] , and L be a bounded linear operator. Then

W { f C ( [ a , b ] × [ c , d ] ) : f ( x i , y j ) = 0 , ( i , j ) N 0 × N 0 , f ( a , c ) = f ( a , d ) = f ( b , c ) = f ( b , d ) = 0 }

is a non-trivial closed invariant subspace for the α - fractal operator Δ , L α .

Proof

By the above theorem, we have W ( x k , y l ) are non-trivial closed invariant subspaces of Δ , L α for k , l { 0 , } . The result follows immediately by taking the intersection of these subspaces.□

6 Concluding remarks

Fractal interpolation is one of the few methods of interpolation that can produce both smooth and nonsmooth functions interpolating a prescribed finite set of univariate data. The apparent intricacy of the definition of fractal interpolation belies the remarkable richness of its structure so much so that the theory of fractal interpolation has developed at a rapid pace and continues to flourish. Extensions of Barnsley’s original framework of FIF in two ways – considering a countably infinite one-dimensional data set and considering a finite bivariate data set – have been one of the topics of further research studies in fractal interpolation. On the other hand, another area of study in the theory of fractal interpolation is the concept of α -fractal function, which is an offspring of the FIF. The α -fractal formalism paved the way for fruitful interactions between fractal interpolation and other branches of mathematics. By combining these three lines of research, in this article, we investigated fractal interpolation for a bivariate countably infinite data set for the first time in the literature and studied the associated notion of α -fractal function. The concept of fractal operator that emerges quite naturally with the notion of α -fractal function is studied in a more general setting of nonlinear operators. In contrast to the countable univariate and finite bivariate counterparts, the countable bivariate situation considered in this article required some additional considerations and assumptions, making the analysis less straightforward. Overall, the present article makes some modest contributions to the field of fractal approximation theory by studying certain generalizations of fractal interpolation. As in the case of univariate fractal interpolation and fractal surfaces, it is felt that the countable bivariate fractal interpolants developed in the present work can find applications in the approximation theory, the theory of sampling and reconstruction, and in the interface of fractal interpolation and operator theory.

Acknowledgements

We would like to thank the anonymous reviewers for several constructive and competent criticism on the earlier version of this article. As always, any remaining errors are our own.

  1. Funding information: The second author is financially supported by Lucian Blaga University of Sibiu & Hasso Plattner Foundation research grants LBUS-IRG-2021-07. The third author expresses gratitude for the support provided by project CRG/2020/002309 from the Science and Engineering Research Board (SERB), Government of India.

  2. Author contributions: All authors have accepted responsibility for the entire content of this manuscript and approved its submission.

  3. Conflict of interest: The authors state no conflict of interest.

  4. Data availability statement: Data sharing is not applicable to this article as no datasets were generated or analyzed during the current study.

References

[1] M. F. Barnsley, Fractal functions and interpolation, Constr. Approx. 2 (1986), 303–329. 10.1007/BF01893434Search in Google Scholar

[2] J. Hutchinson, Fractals and self-similarity, Indiana Univ. Math. J. 30 (1981), 713–747. 10.1512/iumj.1981.30.30055Search in Google Scholar

[3] M. F. Barnsley, Fractals Everywhere, Academic Press, Orlando, Florida, 1988. Search in Google Scholar

[4] A. Deniz and Y. Özdemir, Graph-directed fractal interpolation functions, Turkish J. Math. 41 (2017), 829–840. 10.3906/mat-1604-39Search in Google Scholar

[5] D.-C. Luor, Fractal interpolation functions with partial self similarity, J. Math. Anal. Appl. 464 (2018), no. 1, 911–923. 10.1016/j.jmaa.2018.04.041Search in Google Scholar

[6] P. R. Massopust, Non-stationary fractal interpolation, Mathematics 7 (2019), no. 8, 14 pp. 10.3390/math7080666Search in Google Scholar

[7] P. R. Massopust, Fractal interpolation over nonlinear partitions, Chaos Solitons Fractals 162 (2022), 112503.10.1016/j.chaos.2022.112503Search in Google Scholar

[8] R. Miculescu, A. Mihail, and C. M. Pacurar, A fractal interpolation scheme for a possible sizeable set of data, J. Fractal Geom. 9 (2022), no. 3/4, DOI: https://doi.org/10.4171/JFG/117. 10.4171/jfg/117Search in Google Scholar

[9] R. Miculescu, A. Mihail, and C. M. Pacurar, Interpolation type iterated function systems, J. Math. Anal. Appl. 519 (2023), no. 1, 126747. 10.1016/j.jmaa.2022.126747Search in Google Scholar

[10] S. Ri, A new nonlinear fractal interpolation function, Fractals 25 (2017), no. 6, 1750063. 10.1142/S0218348X17500633Search in Google Scholar

[11] H-Y. Wang and J-S. Yu, Fractal interpolation functions with variable parameters and their analytical properties, J. Approx. Theory 175 (2013), 1–18. 10.1016/j.jat.2013.07.008Search in Google Scholar

[12] P. R. Massopust, Fractal Functions, Fractal Surfaces, and Wavelets, Academic Press, London, 2nd ed., 2016. 10.1016/B978-0-12-804408-7.00008-4Search in Google Scholar

[13] M. A. Navascués, Fractal polynomial interpolation, Z. Anal. Anwend. 25 (2005), no. 2, 401–418. 10.4171/zaa/1248Search in Google Scholar

[14] M. A. Navascués, Fractal approximation, Complex Anal. Oper. Theory 4 (2010), no. 4, 953–974. 10.1007/s11785-009-0033-1Search in Google Scholar

[15] S. Chen, The non-differentiability of a class of fractal interpolation functions, Acta Math. Sci. 19 (1999), no. 4, 425–430. 10.1016/S0252-9602(17)30526-XSearch in Google Scholar

[16] P. Viswanathan, A. K. B. Chand, and M. A. Navascués, Fractal perturbation preserving fundamental shapes: Bounds on the scale factors, J. Math. Anal. Appl. 419 (2014), 804–817. 10.1016/j.jmaa.2014.05.019Search in Google Scholar

[17] M. A. Navascués, Fractal trigonometric approximation, Electron. Trans. Numer. Anal. 20 (2005), 64–74. Search in Google Scholar

[18] P. Viswanathan and A. K. B. Chand, Fractal rational functions and their approximation properties, J. Approx. Theory 185 (2014), 31–50. 10.1016/j.jat.2014.05.013Search in Google Scholar

[19] P. Viswanathan and M. A. Navascués, A fractal operator on some standard spaces of functions, Proc. Edinb. Math. Soc. 60 (2017), 771–786. 10.1017/S0013091516000316Search in Google Scholar

[20] A. K. B. Chand and G. P. Kapoor, Hidden variable bivariate fractal interpolation surfaces, Fractals 11 (2003), 277–288. 10.1142/S0218348X03002129Search in Google Scholar

[21] L. Dalla, Bivariate fractal interpolation functions on grids, Fractals 10 (2002), 53–58. 10.1142/S0218348X02000951Search in Google Scholar

[22] Z. Feng, Variation and Minkowski dimension of fractal interpolation surfaces, J. Math. Anal. Appl. 176 (1993), 561–586. Search in Google Scholar

[23] J. S. Geronimo and D. Hardin, Fractal interpolation surfaces and a related 2D multiresolution analysis, J. Math. Anal. Appl. 176 (1993), 561–586. 10.1006/jmaa.1993.1232Search in Google Scholar

[24] R. Malysz, The Minkowski dimension of the bivariate fractal interpolation surfaces, Chaos Solitons Fractals 27 (2006), 27–50. 10.1016/j.chaos.2005.05.007Search in Google Scholar

[25] P. R. Massopust, Fractal surfaces, J. Math. Anal. Appl. 151 (1990), 275–290. 10.1016/0022-247X(90)90257-GSearch in Google Scholar

[26] W. Metzer and C. H. Yun, Construction of fractal interpolation surfaces on rectangular grids, Internat. J. Bifur. Chaos 20 (2010), 4079–4086. 10.1142/S0218127410027933Search in Google Scholar

[27] H. Xie and H. Sun, The study on bivariate fractal interpolation functions and creation of fractal interpolated surfaces, Fractals 5 (1997), 625–634. 10.1142/S0218348X97000504Search in Google Scholar

[28] N. Zhao, Construction and application of fractal interpolation surfaces, Vis. Comput. 12 (1996), 132–146. 10.1007/s003710050053Search in Google Scholar

[29] P. Bouboulis and L. Dalla, A general construction of fractal interpolation functions on grids of Rn, European J. Appl. Math. 18 (2007), no. 4, 449–476. 10.1017/S0956792507007024Search in Google Scholar

[30] D. Hardin and P. R. Massopust, Fractal interpolation functions from Rn into Rm and their projections, Z. Anal. Anwend. 12 (1993), 535–548. 10.4171/zaa/549Search in Google Scholar

[31] H.-J. Ruan and Q. Xu, Fractal interpolation surfaces on rectangular grids, Bull. Aust. Math. Soc. 91 (2015), 435–446. 10.1017/S0004972715000064Search in Google Scholar

[32] S. Jha, A. K. B. Chand, M. A. Navascués, and A. Sahu, Approximation properties of bivariate α-fractal functions and dimension results, Appl. Anal. 100 (2020), 3426–3444, DOI: https://doi.org/10.1080/00036811.2020.1721472. 10.1080/00036811.2020.1721472Search in Google Scholar

[33] S. Verma and P. Viswanathan, A fractal operator associated with bivariate fractal interpolation functions on rectangular grids, Results Math. 75 (2020), no. 28, 26 pp. 10.1007/s00025-019-1152-2Search in Google Scholar

[34] N. A. Secelean, The existence of the attractor of countable iterated function systems, Mediterr. J. Math. 9 (2012), no. 1, 61–79. 10.1007/s00009-011-0116-xSearch in Google Scholar

[35] N. A. Secelean, Countable Iterated Function Systems, LAP Lambert Academic Publishing, Saarbrücken, Germany, 2013. Search in Google Scholar

[36] N. A. Secelean, The fractal interpolation for countable systems of data, Univ. Beograd. Publ. Elektrotehn. Fak. Ser. Mat. 14 (2003), 11–19. 10.2298/PETF0314011SSearch in Google Scholar

[37] K. K. Pandey and P. Viswanathan, On bivariate fractal interpolation for countable data and associated nonlinear fractal operator, 2020, arXiv:2010.0546. Search in Google Scholar

[38] H. M. Riedl and G. F. Webb, Relative boundedness conditions and the perturbations of nonlinear operators, Czechoslovak Math. J. 24 (1974), 584–597. 10.21136/CMJ.1974.101277Search in Google Scholar

[39] T. Kato, Perturbation Theory for Linear Operators, Springer-Verlag, New York, 1980. Search in Google Scholar

[40] B. S. Thomson, J. B. Bruckner, and A. M. Bruckner, Elementary Real Analysis, Prentice-Hall, New Jersey, USA, 2001. Search in Google Scholar

[41] P. Viswanathan, Fractal approximation of a function from a countable sample set and associated fractal operator, Rev. R. Acad. Cienc. Exactas Fís. Nat. Ser. A Mat. RACSAM 114 (2020), no. 1, 32. 10.1007/s13398-019-00772-8Search in Google Scholar

[42] N. Dunford and J. T. Schwartz, Linear Operators, Part 1: General Theory, John Wiley & Sons, New York, 1988. Search in Google Scholar

Received: 2022-12-09
Revised: 2023-11-23
Accepted: 2024-04-26
Published Online: 2024-10-19

© 2024 the author(s), published by De Gruyter

This work is licensed under the Creative Commons Attribution 4.0 International License.

Articles in the same Issue

  1. Regular Articles
  2. On the p-fractional Schrödinger-Kirchhoff equations with electromagnetic fields and the Hardy-Littlewood-Sobolev nonlinearity
  3. L-Fuzzy fixed point results in -metric spaces with applications
  4. Solutions of a coupled system of hybrid boundary value problems with Riesz-Caputo derivative
  5. Nonparametric methods of statistical inference for double-censored data with applications
  6. LADM procedure to find the analytical solutions of the nonlinear fractional dynamics of partial integro-differential equations
  7. Existence of projected solutions for quasi-variational hemivariational inequality
  8. Spectral collocation method for convection-diffusion equation
  9. New local fractional Hermite-Hadamard-type and Ostrowski-type inequalities with generalized Mittag-Leffler kernel for generalized h-preinvex functions
  10. On the asymptotics of eigenvalues for a Sturm-Liouville problem with symmetric single-well potential
  11. On exact rate of convergence of row sequences of multipoint Hermite-Padé approximants
  12. The essential norm of bounded diagonal infinite matrices acting on Banach sequence spaces
  13. Decay rate of the solutions to the Cauchy problem of the Lord Shulman thermoelastic Timoshenko model with distributed delay
  14. Enhancing the accuracy and efficiency of two uniformly convergent numerical solvers for singularly perturbed parabolic convection–diffusion–reaction problems with two small parameters
  15. An inertial shrinking projection self-adaptive algorithm for solving split variational inclusion problems and fixed point problems in Banach spaces
  16. An equation for complex fractional diffusion created by the Struve function with a T-symmetric univalent solution
  17. On the existence and Ulam-Hyers stability for implicit fractional differential equation via fractional integral-type boundary conditions
  18. Some properties of a class of holomorphic functions associated with tangent function
  19. The existence of multiple solutions for a class of upper critical Choquard equation in a bounded domain
  20. On the continuity in q of the family of the limit q-Durrmeyer operators
  21. Results on solutions of several systems of the product type complex partial differential difference equations
  22. On Berezin norm and Berezin number inequalities for sum of operators
  23. Geometric invariants properties of osculating curves under conformal transformation in Euclidean space ℝ3
  24. On a generalization of the Opial inequality
  25. A novel numerical approach to solutions of fractional Bagley-Torvik equation fitted with a fractional integral boundary condition
  26. Holomorphic curves into projective spaces with some special hypersurfaces
  27. On Periodic solutions for implicit nonlinear Caputo tempered fractional differential problems
  28. Approximation of complex q-Beta-Baskakov-Szász-Stancu operators in compact disk
  29. Existence and regularity of solutions for non-autonomous integrodifferential evolution equations involving nonlocal conditions
  30. Jordan left derivations in infinite matrix rings
  31. Nonlinear nonlocal elliptic problems in ℝ3: existence results and qualitative properties
  32. Invariant means and lacunary sequence spaces of order (α, β)
  33. Novel results for two families of multivalued dominated mappings satisfying generalized nonlinear contractive inequalities and applications
  34. Global in time well-posedness of a three-dimensional periodic regularized Boussinesq system
  35. Existence of solutions for nonlinear problems involving mixed fractional derivatives with p(x)-Laplacian operator
  36. Some applications and maximum principles for multi-term time-space fractional parabolic Monge-Ampère equation
  37. On three-dimensional q-Riordan arrays
  38. Some aspects of normal curve on smooth surface under isometry
  39. Mittag-Leffler-Hyers-Ulam stability for a first- and second-order nonlinear differential equations using Fourier transform
  40. Topological structure of the solution sets to non-autonomous evolution inclusions driven by measures on the half-line
  41. Remark on the Daugavet property for complex Banach spaces
  42. Decreasing and complete monotonicity of functions defined by derivatives of completely monotonic function involving trigamma function
  43. Uniqueness of meromorphic functions concerning small functions and derivatives-differences
  44. Asymptotic approximations of Apostol-Frobenius-Euler polynomials of order α in terms of hyperbolic functions
  45. Hyers-Ulam stability of Davison functional equation on restricted domains
  46. Involvement of three successive fractional derivatives in a system of pantograph equations and studying the existence solution and MLU stability
  47. Composition of some positive linear integral operators
  48. On bivariate fractal interpolation for countable data and associated nonlinear fractal operator
  49. Generalized result on the global existence of positive solutions for a parabolic reaction-diffusion model with an m × m diffusion matrix
  50. Online makespan minimization for MapReduce scheduling on multiple parallel machines
  51. The sequential Henstock-Kurzweil delta integral on time scales
  52. On a discrete version of Fejér inequality for α-convex sequences without symmetry condition
  53. Existence of three solutions for two quasilinear Laplacian systems on graphs
  54. Embeddings of anisotropic Sobolev spaces into spaces of anisotropic Hölder-continuous functions
  55. Nilpotent perturbations of m-isometric and m-symmetric tensor products of commuting d-tuples of operators
  56. Characterizations of transcendental entire solutions of trinomial partial differential-difference equations in ℂ2#
  57. Fractional Sturm-Liouville operators on compact star graphs
  58. Exact controllability for nonlinear thermoviscoelastic plate problem
  59. Improved modified gradient-based iterative algorithm and its relaxed version for the complex conjugate and transpose Sylvester matrix equations
  60. Superposition operator problems of Hölder-Lipschitz spaces
  61. A note on λ-analogue of Lah numbers and λ-analogue of r-Lah numbers
  62. Ground state solutions and multiple positive solutions for nonhomogeneous Kirchhoff equation with Berestycki-Lions type conditions
  63. A note on 1-semi-greedy bases in p-Banach spaces with 0 < p ≤ 1
  64. Fixed point results for generalized convex orbital Lipschitz operators
  65. Asymptotic model for the propagation of surface waves on a rotating magnetoelastic half-space
  66. Multiplicity of k-convex solutions for a singular k-Hessian system
  67. Poisson C*-algebra derivations in Poisson C*-algebras
  68. Signal recovery and polynomiographic visualization of modified Noor iteration of operators with property (E)
  69. Approximations to precisely localized supports of solutions for non-linear parabolic p-Laplacian problems
  70. Solving nonlinear fractional differential equations by common fixed point results for a pair of (α, Θ)-type contractions in metric spaces
  71. Pseudo compact almost automorphic solutions to a family of delay differential equations
  72. Periodic measures of fractional stochastic discrete wave equations with nonlinear noise
  73. Asymptotic study of a nonlinear elliptic boundary Steklov problem on a nanostructure
  74. Cramer's rule for a class of coupled Sylvester commutative quaternion matrix equations
  75. Quantitative estimates for perturbed sampling Kantorovich operators in Orlicz spaces
  76. Review Articles
  77. Penalty method for unilateral contact problem with Coulomb's friction in time-fractional derivatives
  78. Differential sandwich theorems for p-valent analytic functions associated with a generalization of the integral operator
  79. Special Issue on Development of Fuzzy Sets and Their Extensions - Part II
  80. Higher-order circular intuitionistic fuzzy time series forecasting methodology: Application of stock change index
  81. Binary relations applied to the fuzzy substructures of quantales under rough environment
  82. Algorithm selection model based on fuzzy multi-criteria decision in big data information mining
  83. A new machine learning approach based on spatial fuzzy data correlation for recognizing sports activities
  84. Benchmarking the efficiency of distribution warehouses using a four-phase integrated PCA-DEA-improved fuzzy SWARA-CoCoSo model for sustainable distribution
  85. Special Issue on Application of Fractional Calculus: Mathematical Modeling and Control - Part II
  86. A study on a type of degenerate poly-Dedekind sums
  87. Efficient scheme for a category of variable-order optimal control problems based on the sixth-kind Chebyshev polynomials
  88. Special Issue on Mathematics for Artificial intelligence and Artificial intelligence for Mathematics
  89. Toward automated hail disaster weather recognition based on spatio-temporal sequence of radar images
  90. The shortest-path and bee colony optimization algorithms for traffic control at single intersection with NetworkX application
  91. Neural network quaternion-based controller for port-Hamiltonian system
  92. Matching ontologies with kernel principle component analysis and evolutionary algorithm
  93. Survey on machine vision-based intelligent water quality monitoring techniques in water treatment plant: Fish activity behavior recognition-based schemes and applications
  94. Artificial intelligence-driven tone recognition of Guzheng: A linear prediction approach
  95. Transformer learning-based neural network algorithms for identification and detection of electronic bullying in social media
  96. Squirrel search algorithm-support vector machine: Assessing civil engineering budgeting course using an SSA-optimized SVM model
  97. Special Issue on International E-Conference on Mathematical and Statistical Sciences - Part I
  98. Some fixed point results on ultrametric spaces endowed with a graph
  99. On the generalized Mellin integral operators
  100. On existence and multiplicity of solutions for a biharmonic problem with weights via Ricceri's theorem
  101. Approximation process of a positive linear operator of hypergeometric type
  102. On Kantorovich variant of Brass-Stancu operators
  103. A higher-dimensional categorical perspective on 2-crossed modules
  104. Special Issue on Some Integral Inequalities, Integral Equations, and Applications - Part I
  105. On parameterized inequalities for fractional multiplicative integrals
  106. On inverse source term for heat equation with memory term
  107. On Fejér-type inequalities for generalized trigonometrically and hyperbolic k-convex functions
  108. New extensions related to Fejér-type inequalities for GA-convex functions
  109. Derivation of Hermite-Hadamard-Jensen-Mercer conticrete inequalities for Atangana-Baleanu fractional integrals by means of majorization
  110. Some Hardy's inequalities on conformable fractional calculus
  111. The novel quadratic phase Fourier S-transform and associated uncertainty principles in the quaternion setting
  112. Special Issue on Recent Developments in Fixed-Point Theory and Applications - Part I
  113. A novel iterative process for numerical reckoning of fixed points via generalized nonlinear mappings with qualitative study
  114. Some new fixed point theorems of α-partially nonexpansive mappings
  115. Generalized Yosida inclusion problem involving multi-valued operator with XOR operation
  116. Periodic and fixed points for mappings in extended b-gauge spaces equipped with a graph
  117. Convergence of Peaceman-Rachford splitting method with Bregman distance for three-block nonconvex nonseparable optimization
  118. Topological structure of the solution sets to neutral evolution inclusions driven by measures
  119. (α, F)-Geraghty-type generalized F-contractions on non-Archimedean fuzzy metric-unlike spaces
  120. Solvability of infinite system of integral equations of Hammerstein type in three variables in tempering sequence spaces c 0 β and 1 β
  121. Special Issue on Nonlinear Evolution Equations and Their Applications - Part I
  122. Fuzzy fractional delay integro-differential equation with the generalized Atangana-Baleanu fractional derivative
  123. Klein-Gordon potential in characteristic coordinates
  124. Asymptotic analysis of Leray solution for the incompressible NSE with damping
  125. Special Issue on Blow-up Phenomena in Nonlinear Equations of Mathematical Physics - Part I
  126. Long time decay of incompressible convective Brinkman-Forchheimer in L2(ℝ3)
  127. Numerical solution of general order Emden-Fowler-type Pantograph delay differential equations
  128. Global smooth solution to the n-dimensional liquid crystal equations with fractional dissipation
  129. Spectral properties for a system of Dirac equations with nonlinear dependence on the spectral parameter
  130. A memory-type thermoelastic laminated beam with structural damping and microtemperature effects: Well-posedness and general decay
  131. The asymptotic behavior for the Navier-Stokes-Voigt-Brinkman-Forchheimer equations with memory and Tresca friction in a thin domain
  132. Absence of global solutions to wave equations with structural damping and nonlinear memory
  133. Special Issue on Differential Equations and Numerical Analysis - Part I
  134. Vanishing viscosity limit for a one-dimensional viscous conservation law in the presence of two noninteracting shocks
  135. Limiting dynamics for stochastic complex Ginzburg-Landau systems with time-varying delays on unbounded thin domains
  136. A comparison of two nonconforming finite element methods for linear three-field poroelasticity
Downloaded on 12.9.2025 from https://www.degruyterbrill.com/document/doi/10.1515/dema-2024-0014/html
Scroll to top button