arXiv:1310.2396v1 [cs.AI] 9 Oct 2013 e fdfial ocpsi eea eeaie og e mo set rough generalized several math in the investigated concepts [22] definable Pei of investigated sets. set Li rough and based Ge covering [10], familie of In f set aspects. de definable mainly algebraic the the studies from studied of these sets [33] Xu structure sets, and the rough Yang sets. Pawlak Since rough in topic. clear this and on simple done been have ies arbitrar and [26] [17,35,36,40]. sets sets rough similarit based to relation extended tolerance been [27], has [6,7,25, theory development set new rough some Pawlak hand, gained theory thi set on rough researches based base deep covering done have establish [2,3,5,23,24,42] to scholars coverings used bee has have [38] sets Zakowski rough appli Pawlak many of extensions for restrictive meaningful several too are partitions and relations [8,13,31 topo sets [12,18,28,29], fuzzy matroids with [32], theory connected hyperstructure inf been in Th have knowledge sets mining. vague rough data and ory, uncertain in with data dealing of for types useful various analyze and organize Introduction 1 eesr n ufiin odto o w eain to relations two for condition sufficient and necessary A ⋆ orsodn uhr -al [email protected] E-mail: author. Corresponding enbesti udmna ocp nvrostpso ro of types various in concept fundamental a is set definable A partit or relations equivalence on built is theory set Rough tool a as [20,21] Pawlak by proposed been has theory set Rough Abstract. Keywords. relatio two family. for set condition clo definable the sufficient of se and concept necessary definable the a on same present based the paper, this induce ba In w relations discussed. relation research been in two much issue condition been fundamental what a has der as There However, structure complex. topic. the this more sets, on bit rough a generalizing is in families but clear, and ple nuetesm enbestfamily set definable same the induce nPwa og es h tutr ftedfial e famil set definable the of structure the sets, rough Pawlak In eain enbest lsr frelations. of Closure set; Definable Relation; innNra nvriy hnzo 600 China 363000, Zhangzhou University, Normal Minnan a fGaua Computing, Granular of Lab u a,WlimZhu William Yao, Hua ⋆ oy[41,1,ads on. so and [14,15,41], logy WlimZhu) (William ain.T drs hsissue, this address To cations. iayrlto ae rough based relation binary y hoy eety covering Recently, theory. s st nuetesame the induce to ns eainbsdruhsets rough based relation y enbest ftntypes ten of sets definable rpsd noehand, one On proposed. n rainsses nthe- In systems. ormation og e hoy Many theory. set rough d ueo eain,we relations, of sure ftedfial set definable the of mtclsrcueo the of structure ematical 3] atcs[4,9,16,30], lattices ,36], frlto ae rough based relation of s e og es un- sets, rough sed 43,9.O h other the On 34,37,39]. csdo generalizing on ocused smto sespecially is method is esaddsusdthe discussed and dels nbestfmle is families set finable aiyhsnot has family t os u equivalence but ions, g es aystud- Many sets. ugh r focusing ork oconceptualize, to e ssim- is ies relationship between different rough set models. In [1], Ali et al. investigated the topo- logical structures associated with definable sets in the generalized approximation space (X,Y,T ). Liu and Zhu [17] presented the necessary and sufficient condition for defin- able set families to be nonempty and extended the concept of definable set. However, as a fundamental issue in relation based rough sets, under what condition two relations induce the same definable set family has not been discussed. In this paper, based on the conceptof the closure of relations, we present a necessary and sufficient condition for two relations to induce the same definable set family. First, we introduce some definitions and results of relation closures. Secondly, we study fur- ther some properties of definable sets. Thirdly, we simplify the expression of equivalent closures under certain conditions and prove that serial relations satisfy these conditions. Finally, we prove that a serial relation and its equivalent closure induce the same defin- able set family, and based on this, we present a necessary and sufficient condition for two relations to induce the same definable set family. The remainder of this paper is organized as follows. In Section 2, we review the relevant concepts and introduce some existing results, which include relations, relation based rough sets and relation closures. In Section 3, we investigate some fundamen- tal properties of definable sets. In Section 4, we simplify the expression of equivalent closures under certain conditions. In Section 5, we present a necessary and sufficient condition for two relations to induce the same definable set family. Section 6 concludes this paper.

2 Preliminaries

In this section, we recall some basic concepts of relations, relation based rough sets and closures of relations. In this paper, we denote ∪X∈SX by ∪S, where S is a set family. The fact that A ⊆ B and A 6= B is denoted by A ⊂ B. We denote the set of positive integers by N +.

2.1 Relation and relation based rough sets

Relations, especially binary relations, are a basic concept in set theory. They play an important role in rough set theory as well.

Definition 1. (Relation) Let U be a set. Any R ⊆ U × U is called a binary relation on U. If (x, y) ∈ R, we say x has relation R with y, and denote this relationship as xRy.

Throughout this paper, a binary relation is simply called a relation. On the basis of relations, we introduce the concepts of successor neighborhood and predecessor neigh- borhood.

Definition 2. (Successor neighborhood and predecessor neighborhood) Let R be a re- lation on U and x ∈ U. The successor neighborhood and predecessor neighborhood of x are defined as SR(x)= {y|xRy} and PR(x)= {y|yRx}, respectively.

It is obvious y ∈ SR(x) ⇔ x ∈ PR(y). Below we introduce some special relations. Definition 3. (Reflective, symmetric, transitive and serial relation) Let R be a relation on U. If for any x ∈ U, xRx, we say R is reflective. If for any x, y ∈ U, xRy implies yRx, we say R is symmetric. If for any x,y,z ∈ U, xRy and yRz imply xRz, we say R is transitive. If for any x ∈ U, there exists some y ∈ U such that xRy, we say R is serial.

If R is serial, we have that for any x ∈ U, it follows that SR(x) 6= ∅. Among various types of relations, there is an important type of relations called equivalent relations.

Definition 4. (Equivalence relation [11]) Let R ⊆ A × A and A 6= ∅. If R is reflective, symmetric and transitive, we say R is an equivalence relation on A.

Based on the concept of equivalence relation, we introduce the concept of equiva- lence class.

Definition 5. (Equivalence class [11]) Let R be an equivalence relation on a nonempty set A. We denote SR(x) as [x]R and call it the equivalence class of x with respect of R.

Given an equivalence relation, we can define a set family named quotient set.

Definition 6. (Quotient set [11]) Let R be an equivalence relation on a nonempty set A. We define the quotient set of A with respect of R as A/R = {[x]R|x ∈ A}.

There exists a concept called partition, which is closely related to the concept of equivalence relation.

Definition 7. (Partition [11]) Let A be a nonempty set and P be a family of subsets of A. P is called a partition on A if the following conditions hold: (1) ∅ ∈/ P ; (2) ∪P = U; (3) for any K,L ∈ P , if K 6= L, K ∩ L = ∅.

The following theorem presents the relationship between partitions and equivalence relations.

Theorem 1. ([11]) Let A be a nonempty set. Then (1) If R is an equivalence relation on A, A/R is a partition on A; (2) If P is a partition on A, {(x, y)|∃K ∈ P ({x, y}⊆ K)} is an equivalence relation on A.

This theorem indicates that there is an one-to-one mapping between all the equiva- lence relations on a nonempty set A and all the partitions on A. Pawlak rough sets have been extended to various types of generalizing rough sets. This paper studies relation based rough sets.

Definition 8. (Rough set based on a relation [35]) Suppose R is a relation on a uni- verse U. A pair of approximation operators, R, R: P (U) → P (U), are defined by

R(X)= {x|SR(x) ⊆ X} and R(X)= {x|SR(x) ∩ X 6= ∅}. They are called the lower approximation operation and the upper approximation oper- ation of X, respectively. It is obvious R(X) = {x|∀y(xRy → y ∈ X)} and R(X) = {x|∃y(xRy ∧ y ∈ X)}. When R is an equivalence relation on U, the above two approximation operations are called Pawlak approximation operations. The lower approximation operators and the upper approximation operators have the following properties.

Proposition 1. ([35]) (1) R(X)= −R(−X)}; (2) R(X)= −R(−X)}.

When R is serial, the lower approximation operators and the upper approximation operators have the following property.

Proposition 2. ([35]) Let R be a serial relation on U. For any X ⊆ U, it follows that R(X) ⊆ R(X).

As an importantconceptin rough set theory,definablesets have been studied widely. Below we introduce its definition.

Definition 9. (Definable set family [22]) Let R be a relation on U and X ⊆ U. If R(X) = X, we call X an inner definable set. If R(X) = X, we call X an outer definable set. If X is both inner and outer definable set, we call X a definable set. We denote the family of all the inner definable sets, outer definable sets and definable sets of U induced by R as I(U, R), O(U, R) and D(U, R), respectively.

It is obvious D(U, R)= I(U, R) ∩ O(U, R).

2.2 The closure of relations For presenting the computational formulae of closure operators, we need to intro- duce some concepts and notations.

Definition 10. Let A be a set. We denote {(x, x)|x ∈ A} as IA.

For any reflective relation R on A, it is obvious IA ⊆ R. Below we introduce the converse of a relation.

Definition 11. Let R be a relation. We definethe converse of R as R−1 = {(x, y)|(y, x) ∈ R}.

For any symmetric relation R on A, it is obvious R = R−1. Conversely, if R = R−1, R is symmetric. Below we introduce the compound of two relations.

Definition 12. ([11]) Let F, G be two relations. We define F ◦G as F ◦G = {(x, y)|∃z((x, z) ∈ G ∧ (z,y) ∈ F )}.

The compound of relations satisfies the associative law.

Proposition 3. ([11]) Let R1, R2, R3 be three relations. Then (R1 ◦ R2) ◦ R3 = R1 ◦ (R2 ◦ R3). Since the associative law of the compound of relations holds, we can define the power of a relation. Definition 13. ([11]) Let R ⊆ A × A and n be a natural number. We denote the nth n n n power of R as R , where (1) R0 = IA; (2) R +1 = R ◦ R.

In orderto turnan arbitraryrelation to a reflectiveor symmetric or transitive relation, we introduce the concept of the closure of a relation.

Definition 14. (Reflective (symmetric or transitive) closure [11]) Let A 6= ∅ and R ⊆ A × A. R′ is called the reflective (symmetric or transitive) closure of R iff R′ satisfies the following three conditions: (1) R′ is reflective (symmetric or transitive); (2) R ⊆ R′; (3) For any reflective (symmetric or transitive) relation R′′ on A, if R ⊆ R′′, R′ ⊆ R′′. We denote the reflective, symmetric and transitive closure of R as r(R), s(R) and t(R), respectively.

The following theorem presents the computational formulae of the above three clo- sures.

Theorem 2. ([11]) Let R ⊆ A × A and A 6= ∅. Then − (1) r(R)= R ∪ IA; (2) s(R)= R ∪ R 1; (3) t(R)= R ∪ R2 ∪ · · · .

The following proposition indicates that mixing the above three closure operations, we can obtain only two possibly different relations.

Proposition 4. ([11]) Let A 6= ∅ and R ⊆ A × A. Then rts(R) = trs(R) = tsr(R) and rst(R)= str(R)= srt(R).

The following example indicates that rst(R) may not be an equivalence relation.

Example 1. Let A = {1, 2, 3, 4} and R = {(1, 2), (2, 3), (1, 4)}. Then rst(R)= IA ∪ {(1, 2), (2, 3), (1, 3), (1, 4), (2, 1), (3, 2), (3, 1), (4, 1)}.It is obvious IA∪{(1, 2), (2, 3), (1, 3), (1, 4), (2, 1), (3, 2), (3, 1), (4, 1)} is not transitive. Thus rst(R) is not an equiv- alence relation.

However, rts(R) has the following properties.

Proposition 5. ([11]) Let A 6= ∅ and R ⊆ A × A. Then rts(R) satisfies the following properties: (1) rts(R) is an equivalence relation; (2) R ⊆ rts(R); (3) For any equivalence relation R′′ on A, if R ⊆ R′′, rts(R) ⊆ R′′.

Based on the above proposition, we introduce the concept of equivalent closure.

Definition 15. (Equivalent closure [11]) Let A 6= ∅ and R ⊆ A × A. We define the equivalent closure of R as e(R)= rts(R). 3 The fundamental properties of definable sets

In this section, on the basis of some existing results, we investigate further some fundamental properties of definable sets. Liu and Zhu [17] gave the following proposi- tion.

Theorem 3. ([17]) Let R be a relation on U. Then D(U, R) 6= ∅ iff R is serial.

Based on the above theorem, in order to study D(U, R), in most cases we first assume R is serial. In fact, we can still obtain the following proposition.

Proposition 6. Let R be a relation on U. If I(U, R)= O(U, R), R is serial.

Proof. We use the proof by contradiction. Suppose R is not serial. It is obvious U ∈ I(U, R) − O(U, R). Thus I(U, R) 6= O(U, R). It is contradictory. 

For the simplicity of the description of the following propositions, we define several new notations.

Definition 16. Let R be a relation on U and X ⊆ U. We define RR(X), PR(X) and VR(X) as SR(X) = ∪x∈X SR(x), PR(X) = ∪x∈X PR(x) and VR(X) = RR(X) ∪ PR(X), respectively.

In [33], Yang and Xu gave the following proposition.

Proposition 7. ([33]) Let R be a relation on U and X ⊆ U. Then X ⊆ R(X) iff SR(X) ⊆ X; R(X) ⊆ X iff PR(X) ⊆ X; R(X) ⊆ X ⊆ R(X) iff VR(X) ⊆ X.

Based on Propositions 2 and 7, we have the following proposition.

Proposition 8. Let R be a serial relation on U and X ⊆ U. Then X ∈ D(U, R) iff VR(X) ⊆ X.

Proof. (⇒): By X ∈ D(U, R), we know that R(X) = X = R(X). Thus R(X) ⊆ X ⊆ R(X). By Proposition 7, we know that VR(X) ⊆ X. (⇐): By Proposition 2, we know that R(X) ⊆ R(X). By Proposition 7, we know that R(X) ⊆ R(X). Thus R(X) = R(X). Again by Proposition 7, we have that R(X)= X = R(X). Then X ∈ D(U, R). 

In fact, under the condition that R is serial, we have the following proposition.

Proposition 9. Let R be a serial relation on U. Then for any X ⊆ U, it follows that VR(X)= X iff VR(X) ⊆ X.

Proof. (⇒): It is straightforward. (⇐): We use the proof by contradiction.Suppose VR(X) 6= X. Then X−VR(X) 6= ∅. Without loss of generality, suppose a ∈ X − VR(X). Since R is serial, there exists some b ∈ U such that b ∈ SR(a). Since SR(a) ⊆ X, b ∈ X. It is obvious a ∈ PR(b). Since PR(b) ⊆ VR(X), a ∈ VR(X). It is contradictory.  The following example indicates that the converse of the above proposition is not true.

Example 2. Let U = {1, 2, 3} and R = {(1, 2), (3, 1)}. Since SR(2) = ∅, R is not serial, but {X ⊆ U|VR(X) ⊆ X} = {∅, {1, 2, 3}} = {X ⊆ U|VR(X)= X}.

Based on Propositions 9 and 8, we have the following proposition.

Proposition 10. Let R be a serial relation on U and X ⊆ U. Then X ∈ D(U, R) iff VR(X)= X.

The following proposition presents a relationship between inner and outer definable set families.

Proposition 11. X ∈ I(U, R) ⇔−X ∈ O(U, R).

Proof. By Definition 9 and Proposition 1, we have that X ∈ I(U, R) ⇔ R(X)= X ⇔ −R(−X)= X ⇔ R(−X)= −X ⇔−X ∈ O(U, R). 

According to the above proposition, we have the following proposition.

Proposition 12. X ∈ D(U, R) ⇔−X ∈ D(U, R).

Proof. X ∈ D(U, R) ⇔ (X ∈ I(U, R) ∧ X ∈ O(U, R)) ⇔ (−X ∈ O(U, R) ∧−X ∈ I(U, R)) ⇔−X ∈ D(U, R). 

Based on some above propositions, we present a sufficient condition for I(U, R)= O(U, R).

Proposition 13. Let R be a serial relation on U. If for any X ⊆ U, SR(X) ⊆ X implies PR(X) ⊆ X, I(U, R)= O(U, R).

Proof. For any X ∈ I(U, R), we know that R(X) = X. Thus X ⊆ R(X). By Proposition 7, we know that SR(X) ⊆ X. Hence PR(X) ⊆ X. Then VR(X) ⊆ X. By Proposition 8, we know that X ∈ D(U, R) ⊆ O(U, R). Therefore I(U, R) ⊆ O(U, R). For any Y ∈ O(U, R), by Proposition 11, we know that −Y ∈ I(U, R). Thus R(−Y ) = −Y . Hence −Y ⊆ R(−Y ). By Proposition 7, we know that SR(−Y ) ⊆ −Y . Then PR(−Y ) ⊆ −Y . Thus VR(−Y ) ⊆ −Y . By Proposition 8, we know that −Y ∈ D(U, R). By Proposition 12, we know that Y ∈ D(U, R) ⊆ I(U, R). Therefore O(U, R) ⊆ I(U, R). Then I(U, R)= O(U, R). 

The converse of the above proposition is not true. To illustrate this, let us see an example.

Example 3. Let U = {1, 2, 3} and R = {(1, 1), (1, 2), (2, 1), (3, 2)}. Then SR({1, 2}) = {1, 2}⊆{1, 2} and PR({1, 2})= {1, 2, 3} * {1, 2}, but I(U, R)= {∅, {1, 2, 3}} = O(U, R).

For applying Proposition 13, we first present the following lemma. Lemma 1. Let R be a symmetric relation on U. Then for any X ⊆ U, SR(X) = PR(X).

Proof. For any a ∈ U and any b ∈ SR(a), we know that aRb. Since R is symmetric, bRa. Thus b ∈ PR(a). Hence SR(a) ⊆ PR(a). Similarly, PR(a) ⊆ SR(a). Then SR(X)= ∪x∈X SR(x)= ∪x∈X PR(x)= PR(X). 

As a application of Proposition 13, we have the following proposition.

Proposition 14. Let R be a serial and symmetric relation on U. Then I(U, R) = O(U, R).

Proof. It follows from Lemma 1 and Proposition 13. 

4 Simplification of equivalent closures under certain conditions

In this section, we simplify the expression of equivalent closures under certain con- ditions and prove that serial relations satisfy these conditions. By Theorem 2, we obtain the computational formula of equivalent closures.

− − Corollary 1. Let R ⊆ A×A and A 6= ∅. Then e(R)= IA ∪(R∪R 1)∪(R∪R 1)2 ∪ ··· .

For dealing with the power of relations, we present the following proposition.

Proposition 15. Let R ⊆ A×A, A 6= ∅, k ∈ N + and k ≥ 2. Then (x, y) ∈ Rk iff there exist x1, x2, ··· , xk−1 ∈ A such that (x, x1), (x1, x2), ··· , (xk−2, xk−1), (xk−1,y) ∈ R.

Proof. (⇒): We prove this assertion using induction on k. If k = 2, this assertion fol- lows easily from Definitions 12 and 13. Assume this assertion is true for k ≤ t−1. Now assume k = t. By Definition 13, we have that (x, y) ∈ Rt−1 ◦ R. By Definition 12, t−1 we know that there exists some x1 such that (x, x1) ∈ R and (x1,y) ∈ R . By the assumption of the induction, we know that there exist x2, x3, ··· , xt−1 ∈ A such that (x1, x2), (x2, x3), ··· , (xt−2, xt−1), (xt−1,y) ∈ R. Thus (x, x1), (x1, x2), ··· , (xt−2, xt−1), (xt−1,y) ∈ R. (⇐): We prove this assertion using induction on k. If k =2, this assertion follows easily from Definitions 12 and 13. Assume this assertion is true for k ≤ t − 1. Now t−1 assume k = t. By the assumption of the induction, we know that (x1,y) ∈ R . Again t−1 t by (x, x1) ∈ R, Definitions 12 and 13, we have that (x, y) ∈ R ◦ R = R . 

The following two lemmas present two properties of symmetric relations.

Lemma 2. Let A 6= ∅, R ⊆ A × A, R be symmetric, k ∈ N + and x ∈ A. Then (x, x) ∈ Rk iff (x, x) ∈ R2. Proof. (⇒): If k = 1, this assertion is obviously true. Below we suppose k ≥ 2. By Proposition 15, we know that there exist x1, x2, ··· , xk−1 ∈ A such that (x, x1), (x1, x2), ··· , (xk−2, xk−1), (xk−1, x) ∈ R. Since R is symmetric, (x1, x) ∈ R. Thus (x, x) ∈ R2. (⇐): It is straightforward.  Lemma 3. Let A 6= ∅, R ⊆ A × A, R be symmetric and x ∈ A. Then (x, x) ∈ R2 iff SR(x) 6= ∅. Proof. (⇒): It is obvious there exists some y ∈ A such that (x, y) ∈ R. Thus y ∈ SR(x) 6= ∅. (⇐): Without loss of generality, suppose z ∈ SR(x). Thus (x, z) ∈ R. Since R is symmetric, (z, x) ∈ R. Hence (x, x) ∈ R2.  The following proposition presents a simple property of successor neighborhoods and predecessor neighborhoods.

Proposition 16. Let R1, R2 be two relations on U and x ∈ U. If R1 ⊆ R2, SR1 (x) ⊆

SR1 (x) and PR1 (x) ⊆ PR1 (x).

Proof. For any y ∈ SR1 (x), we have that (x, y) ∈ R1. Thus (x, y) ∈ R2. Hence y ∈ SR2 (x). Therefore SR1 (x) ⊆ SR2 (x). In the same way, we have that PR1 (x) ⊆

PR2 (x).  By the above proposition, we present an expression of the successor neighborhoods of the union of two relations.

Lemma 4. Let A 6= ∅, R1 and R2 be two relations on A and x ∈ A. Then SR1∪R2 (x)=

SR1 (x) ∪ SR2 (x).

Proof. (⊆): For any z ∈ SR1∪R2 (x), we have that (x, z) ∈ R1 ∪ R2. If (x, z) ∈ R1, we have that z ∈ SR1 (x). If (x, z) ∈ R2, we have that z ∈ SR1 (x). Hence z ∈

SR1 (x) ∪ SR2 (x). Therefore SR1∪R2 (x) ⊆ SR1 (x) ∪ SR2 (x). (⊇): It follows from Proposition 16.  Based on some above results, we present a necessary and sufficient condition for e(R)= t(s(R)). Theorem 4. Let R ⊆ A × A and A 6= ∅. Then e(R) = ts(R) iff for any x ∈ A, it follows that SR(x) ∪ SR−1 (x) 6= ∅. Proof. By Theorem 2, Corollary 1, Lemmas 2, 3 and 4, we have that e(R)= ts(R) ⇔ − − − − − IA ∪(R∪R 1)∪(R∪R 1)2∪· · · = (R∪R 1)∪(R∪R 1)2 ∪···⇔ IA ⊆ (R∪R 1)∪ (R ∪ R−1)2 ∪···⇔∀x ∈ A∃k ∈ N +((x, x) ∈ (R ∪ R−1)k) ⇔ ∀x ∈ A((x, x) ∈ −1 2 (R ∪ R ) ) ⇔ ∀x ∈ A(SR∪R−1 (x) 6= ∅) ⇔ ∀x ∈ A(SR(x) ∪ SR−1 (x) 6= ∅).  Applying the above theorem to a serial relation, we have the following corollary. Corollary 2. Let R ⊆ A × A and A 6= ∅. If R is serial, e(R)= ts(R) = (R ∪ R−1) ∪ (R ∪ R−1)2 ∪ · · · .

Proof. Since R is serial, for any x ∈ A, it follows that SR(x) 6= ∅. Thus SR(x) ∪ −1 SR−1 (x) 6= ∅. By Theorems 4 and 2, we have that e(R)= ts(R) = (R ∪ R ) ∪ (R ∪ R−1)2 ∪ · · · . 5 A necessary and sufficient condition for two relations to induce the same definable set family

In this section, based on the above sections, we present a necessary and sufficient condition for two relations to induce the same definable set family. The following lemma presents a property of definable set families.

Lemma 5. Let R be a serial relation on U, x ∈ X ⊆ U and (x, y) ∈ R ∪ R−1. If X ∈ D(U, R), y ∈ X.

− Proof. It is obvious (x, y) ∈ R or (x, y) ∈ R 1. Then y ∈ SR(x) or y ∈ PR(x). Thus y ∈ VR(X). By Proposition 8, we know that VR(X) ⊆ X. Hence y ∈ X. 

The following lemma is an extension of the above lemma.

Lemma 6. Let R be a serial relation on U, x ∈ X ⊆ U, k be a nonnegative integer −1 and (x, x1), (x1, x2), ··· , (xk−1, xk), (xk,y) ∈ R ∪ R . If X ∈ D(U, R), y ∈ X. Proof. We prove this assertion using induction on k. The assertion under the condition of k = 0 has been proved in Lemma 5. Assume this assertion is true for k ≤ t − 1. Now assume k = t. By the assumption of the induction, we know that xk ∈ X. Thus by Lemma 5, we have that y ∈ X. 

Now we can prove one of the main results in this paper, which indicates that a serial relation and its equivalent closure induce the same definable set family.

Theorem 5. Let R be a serial relation on U. Then D(U, R)= D(U,ts(R)).

Proof. (⊆): Let x ∈ X ∈ D(U, R). By Corollary 2, we know that for any y ∈ −1 −1 2 Sts(R)(x), it follows that (x, y) ∈ (R ∪ R ) ∪ (R ∪ R ) ∪ · · · . Thus (x, y) ∈ R ∪ R−1 or (x, y) ∈ (R ∪ R−1)k, where k ∈ N + ∧ k ≥ 2. If (x, y) ∈ R ∪ R−1, by Lemma 5, we have that y ∈ X. If (x, y) ∈ (R ∪ R−1)k, where k ∈ N + ∧ k ≥ 2, by Proposition 15, we know that there exist x1, x2, ··· , xk−1 ∈ X such that −1 (x, x1), (x1, x2), ··· , (xk−2, xk−1), (xk−1,y) ∈ R∪R . By Lemma 6, we knowthat y ∈ X. Therefore Sts(R)(x) ⊆ X. Since Pts(R)(x)= Sts(R)(x), Pts(R)(x) ⊆ X. Thus Vts(R)(X) ⊆ X. By Proposition 8, we know that X ∈ D(U,ts(R)). (⊇): By Proposition 8, we know that for any X ∈ D(U,ts(R)) and any x ∈ X, it follows that Sts(R)(x) = Pts(R)(x) ⊆ X. By (2) of Definition 15 and Proposition 16, we have that SR(x) ⊆ Sts(R)(x) and PR(x) ⊆ Pts(R)(x). Thus SR(x) ⊆ X and PR(x) ⊆ X. Hence VR(X) ⊆ X. By Proposition 8, we know that X ∈ D(U, R). 

The following example indicates that if R is not serial, D(U, R)= D(U,ts(R)) is incorrect.

Example 4. Let U = {1, 2, 3} and R = {(1, 2), (3, 3)}.Then ts(R)= {(1, 1), (2, 2), (3, 3), (1, 2), (2, 1)} and D(U, R)= ∅ 6= {∅, {3}, {1, 2}, {1, 2, 3}} = D(U,ts(R)).

For presenting a necessary and sufficient condition for two relations to induce the same definable set family, we need to prove the following proposition. Proposition 17. Let R1, R2 be two equivalence relations on U. Then D(U, R1) = D(U, R2) iff R1 = R2.

Proof. (⇒): We use the proof by contradiction. Suppose R1 6= R2. By Theorem 1, we have that U/R1 6= U/R2. Thus U/R1 − U/R2 6= ∅. Without loss of generality, suppose K ∈ U/R1 − U/R2. Then there exists some L ∈ U/R2 such that K ⊂ L or for any J ∈ U/R2, it follows that K * J. If K ⊂ L, we have that R2(K) = ∅. Thus K ∈ D(U, R1) − D(U, R2). Hence D(U, R1) 6= D(U, R2). If for any J ∈ U/R2, it follows that K * J, there exists some A ⊆ U/R2 such that K ⊆ ∪A, |A| ≥ 2 and for any B ⊂ A, it follows that K * ∪B. For any I ∈ A, it is obvious K ∩ I 6= ∅ and K − I 6= ∅. Thus K ⊆ R1(I). Hence ∅ 6= K − I ⊆ R1(I) − I. Then R1(I) 6= I. Therefore I ∈ D(U, R2) − D(U, R1). Then D(U, R1) 6= D(U, R2). (⇐): It is straightforward. 

Finally, based on some aboveresults, we present a necessary and sufficient condition for two relations to induce the same definable set family.

Theorem 6. D(U, R1)= D(U, R2) iff both R1 and R2 are not serial, or both R1 and R2 are serial and ts(R1)= ts(R2).

Proof. (⇒): It is obvious D(U, R1)= D(U, R2)= ∅ or D(U, R1)= D(U, R2) 6= ∅. If D(U, R1) = D(U, R2) = ∅, by Theorem 3, we have that both R1 and R2 are not serial. If D(U, R1)= D(U, R2) 6= ∅, by Theorem 3, we have that both R1 and R2 are serial. Thus by Theorem 5, we have that D(U,ts(R1)) = D(U, R1) = D(U, R2) = D(U,ts(R2)). By Proposition 17 and Corollary 2, we have that ts(R1)= ts(R2). (⇐): If both R1 and R2 are not serial, by Theorem 3, we have that D(U, R1) = ∅ = D(U, R2). If both R1 and R2 are serial and ts(R1) = ts(R2), by Theorem 5, we have that D(U, R1)= D(U,ts(R1)) = D(U,ts(R2)) = D(U, R2). 

6 Conclusions

In this paper, we studied under what condition two relations induce the same defin- able set family. First, we introduced some definitions and results of relation closures. Secondly, we investigated some fundamental properties of definable sets. Thirdly, we simplified the expression of equivalent closures under certain conditions. Finally, based on the research of equivalent closures, we presented a necessary and sufficient condi- tion for two relations to induce the same definable set family. Relation based rough sets are only one type of generalizing rough sets. There are some issues related to definable sets unsolved in various types of generalizing rough sets, which will be investigated in our future works.

Acknowledgments

This work is in part supported by the National Natural Science Foundation of China under Grant Nos. 61170128, 61379049 and 61379089, the Natural Science Foundation of Fujian Province, China under Grant No. 2012J01294, the Fujian Province Founda- tion of Higher Education under Grant No. JK2012028, and the Postgraduate Education Innovation Base for Computer Application Technology, Signal and Information Pro- cessing of Fujian Province (No. [2008]114, High Education of Fujian).

References

1. M. I. Ali, B. Davvaz, M. Shabir, Some properties of generalized rough sets, Information Sciences 224 (2013) 170–179. 2. Z. Bonikowski, E. Bryniarski, U. Wybraniec-Skardowska, Extensions and intentions in the rough set theory, Information Sciences 107 (1998) 149–167. 3. E. Bryniarski, A calculus of rough sets of the first order, Bulletin of the Polish Academy of Sciences 37(1-6) (1989) 71–78. 4. J. Dai, Logic for rough sets with rough double stone algebraic semantics, in: Rough Sets, Fuzzy Sets, Data Mining, and Granular Computing, vol. 3641 of LNCS, 2005. 5. D. Chen, W. Zhang, D. Yeung, E. Tsang, Rough approximations on a complete completely distributive lattice with applications to generalized rough sets, Information Sciences 176 (13) (2006) 1829–1848. 6. M. Diker, A. A. Ugur, Textures and covering based rough sets, Information Sciences 184 (2012) 44–63. 7. Y. Du, Q. Hu, P. Zhu, P. Ma, Rule learning for classification based on neighborhood covering reduction, Information Sciences 181 (24) (2011) 5457–5467. 8. D. Dubois, H. Prade, Rough fuzzy sets and fuzzy rough sets, International Journal of General Systems 17 (2–3) (1990) 191–209. 9. A. A. Estaji, M. R. Hooshmandasl, B. Davvaz, Rough set theory applied to lattice theory, Information Sciences 200 (2012) 108–122. 10. X. Ge, Z. Li, Definable subsets in covering approximation spaces, Int J Comput Math Sci 5 (1) (2011) 31–34. 11. S. Geng, W. Qu, H. Wang, Discrete mathematics course, Peiking University Press, 2002. 12. A. Huang, W. Zhu, Geometric lattice structure of covering-based rough sets through ma- troids, Journal of Applied Mathematics 2012 (2012) Article ID 236307, 25 pages. 13. O. Kazanci, S. Yamak, B. Davvaz, The lower and upper approximations in a quotient hyper- module with respect to fuzzy sets, Information Sciences 178 (10) (2008) 2349–2359. 14. M. Kondo, On the structure of generalized rough sets, Information Sciences 176 (5) (2005) 589–600. 15. E. Lashin, A. Kozae, A. A. Khadra, T. Medhat, Rough set theory for topological spaces, International Journal of Approximate Reasoning 40 (1-2) (2005) 35–43. 16. G. Liu, Generalized rough sets over fuzzy lattices, Information Sciences 178 (6) (2008) 1651–1662. 17. G. Liu, W. Zhu, The algebraic structures of generalized rough set theory, Information Sci- ences 178 (21) (2008) 4105–4113. 18. Y. Liu, W. Zhu, Matroidal structure of rough sets based on serial and transitive relations, Journal of Applied Mathematics 2012 (2012) Article ID 429737, 16 pages. 19. D. Miao, D. Li, Rough Sets Theory Algorithms and Applications, Tsinghua University Press, 2008. 20. Z. Pawlak, Rough sets, International Journal of Computer and Information Sciences 11 (1982) 341–356. 21. Z. Pawlak, Rough sets: theoretical aspects of reasoning about data, Kluwer Academic Pub- lishers, Boston, 1991. 22. D. Pei, On definable concepts of rough set models, Information Sciences 177 (19) (2007) 4230–4239. 23. J. A. Pomykala, Approximation operations in approximation space, Bulletin of the Polish Academy of Sciences, Mathematics 35 (9-10) (1987) 653–662. 24. J. A. Pomykala, On definability in the nondeterministic information system, Bulletin of the Polish Academy of Sciences, Mathematics 36 (1988) 193–210. 25. Z. Shi, Z. Gong, The further investigation of covering-based rough sets: Uncertainty char- acterization, similarity measure and generalized models, Information Sciences 180 (2010) 3745–3763. 26. A. Skowron, J. Stepaniuk, Tolerance approximation spaces, Fundamenta Informaticae 27 (1996) 245–253. 27. R. Slowinski, D. Vanderpooten, A generalized definition of rough approximations based on similarity, IEEE Transactions on Knowledge and Data Engineering 12 (2) (2000) 331–336. 28. J. Tang, K. She, W. Zhu, Matroidal structure of rough sets from the viewpoint of graph theory, Journal of Applied Mathematics 2012 (2012) Article ID 973920, 27 pages. 29. S. Wang, Q. Zhu, W. Zhu, F. Min, Matroidal structure of rough sets and its characterization to attribute reduction, Knowledge-Based Systems 35 (2012) 155–161. 30. S. Wang, Q. Zhu, W. Zhu, F. Min, Quantitative analysis for covering-based rough sets through the upper approximation number, Information Sciences 220 (2013) 483–491. 31. W. Wu, Y. Leung, J. Mi, On characterizations of (I, T) -fuzzy rough approximation operators, Fuzzy Sets and Systems 154 (1) (2005) 76–102. 32. S. Yamak, O. Kazanci, B. Davvaz, Soft hyperstructure, Computers and Mathematics with Applications 62 (2011) 797–803. 33. L. Yang, L. Xu, Algebraic aspects of generalized approximation spaces, International journal of approximate reasoning 51 (1) (2009) 151–161. 34. T. Yang, Q. Li, B. Zhou, Related family: A new method for attribute reduction of covering information systems, Information Sciences 228 (2013) 175–191. 35. Y. Y. Yao, Constructive and algebraic methods of theory of rough sets, Information Sciences 109 (1998) 21–47. 36. Y. Y. Yao, On generalizing pawlak approximation operators, in: Lecture Notes in Artificial Intelligence, vol. 1424, 1998. 37. Y. Y. Yao, B. Yao, Covering based rough set approximations, Information Sciences 200 (2012) 91–107. 38. W. Zakowski, Approximations in the space (u, π), Demonstratio Mathematica 16 (1983) 761–769. 39. Y. Zhang, M. Luo, Relationships between covering-based rough sets and relation-based rough sets, Information Sciences 225 (2013) 55–71. 40. W. Zhu, Generalized rough sets based on relations, Information Sciences 177 (22) (2007) 4997–5011. 41. W. Zhu, Topological approaches to covering rough sets, Information Sciences 177 (6) (2007) 1499–1508. 42. W. Zhu, F. Wang, Reduction and axiomization of covering generalized rough sets, Informa- tion Sciences 152 (1) (2003) 217–230.