SOME (FUZZY) TOPOLOGIES ON GENERAL FUZZY AUTOMATA

Transcription

1 Iranian Journal of Fuzzy Systems Vol. 10, No. 6, (2013) pp SOME (FUZZY) TOPOLOGIES ON GENERAL FUZZY AUTOMATA M. HORRY AND M. M. ZAHEDI Abstract. In this paper, by presenting some notions and theorems, we obtain different types of fuzzy topologies. In fact, we obtain some Lowen-type and Chang-type fuzzy topologies on general fuzzy automata. To this end, first we define a Kuratowski fuzzy interior operator which induces a Lowen-type fuzzy topology on the set of states of a max- min general fuzzy automaton. Also by proving some theorems, we can define two fuzzy closure (two fuzzy interior) operators on the certain sets related to a general fuzzy automaton and then according to these notions we give some theorems and obtain some different Chang-type fuzzy topologies. 1. Introduction and preliminaries The theory of fuzzy sets was introduced by Zadeh [22]. Wee [20] introduced the idea of fuzzy automata. Automata have a long history both in theory and application [1, 2]. Automata are the prime example of general computational systems over discrete spaces [8]. Among the conventional spectrum of automata (i.e. deterministic finite-state automata, non-deterministic finite-state automata, probabilistic automata and fuzzy finite-state automata), deterministic finite-state automata have been the most applied automata to different areas [3, 14, 15]. See [18] for more applications. Fuzzy automata not only provide a systematic approach to handle uncertainty in such systems, but also are able to handle continuous spaces [19]. In general, fuzzy automata provide an attractive systematic way for generalizing discrete applications [5]. Moreover, fuzzy automata are able to create capabilities which are hardly achievable by other tools [21]. Let X be a set. A word on X is the product of a finite sequence of elements in X, Λ is empty word and X is the set of all words on X. In fact, X is the free monoid on X. For a nonempty set X, P(X) denots the set of all fuzzy subsets on X and P (X) denots the set of all subsets on X. A deterministic finite-state automaton is a five-tuple denoted as A = (Q, X, f, T, s), where Q is a finite set of states, X is a finite set of input symbols, the function f from Q X into Q is the state transition, T is a subset of Q of accepting states and s Q is the initial state. A word x = x 1 x 2... x n X is said to be accepted by A if there exist states q 0, q 1,..., q n satisfying Received: July 2011; Revised: October 2011; Accepted: April 2013 Keywords and phrases: (General) Fuzzy automata, (Lowen-type, Chang-type) Fuzzy topology, Closure operator, Topology, Fuzzy closure operator, Fuzzy interior operator.

2 74 M. Horry and M. M. Zahedi (1) q 0 = s (2) f(q i 1, x i ) = q i for i = 1, 2,..., n, (3) q n T. The empty word is accepted by A if and only if s T. A nondeterministic finite-state automaton is a five-tuple denoted as A = (Q, X, f, T, s), where Q is a finite set of states, X is a finite set of input symbols, the function f from Q X into P (Q) is the state transition, T is a subset of Q of accepting states and s Q is the initial state. A fuzzy finite-state automaton (FFA) is a six-tuple denoted as F= (Q, Σ, R, Z, δ, ω), where Q is a finite set of states, Σ is a finite set of input symbols, R is the start state of F, Z is a finite set of output symbols, δ : Q Σ Q [0, 1] is the fuzzy transition function which is used to map a state (current state) into another state (next state) upon an input symbol, attributing a value in the interval [0,1] and ω : Q Z is the output function. In an FFA, as can be seen, associated with each fuzzy transition, there is a membership value in [0,1]. We call this membership value the weight of the transition. The transition from state q i (current state) to state q j (next state) upon input a k is denoted as δ(q i, a k, q j ). We use this notation to refer both to a transition and its weight. Whenever δ(q i, a k, q j ) is used as a value, it refers to the weight of the transition, otherwise, it specifies the transition itself. Also, the set of all transitions of F is denoted as. The above definition is generally accepted as a formal definition for FFA [13, 16, 17]. There is an important problem which should be clarified in the definition of FFA. It is the assignment of membership values to the next states. There are two issues within state membership assignment. The first one is how to assign a membership value to a next state upon the completion of a transition. Secondly, how should we deal with the cases where a state is forced to take several membership values simultaneously via overlapping transition? In 2004, M. Doostfatemeh and S.C. Kremer extended the notion of fuzzy automata and gave the notion of general fuzzy automata [7]. Then we followed it in [23] and now we follow [7,23] and give some new notions and results as mentioned in the abstract. Definition 1.1. [7] A general fuzzy automaton (GFA) F is an eight-tuple machine denoted as F = (Q, Σ, R, Z, ω, δ, F 1, F 2 ), where (i) Q is a finite set of states, Q = {q 1, q 2,..., q n }, (ii) Σ is a finite set of input symbols, Σ = {a 1, a 2,..., a m }, (iii) R is the set of fuzzy start states, R P (Q), (iv) Z is a finite set of output symbols, Z = {b 1, b 2,..., b k }, (v) ω : Q Z is the output function, (vi) δ : (Q [0, 1]) Σ Q [0, 1] is the augmented transition function, (vii) F 1 : [0, 1] [0, 1] [0, 1] is called membership assignment function. Function F 1 (µ, δ) as is seen, is motivated by two parameters µ and δ, where µ is the membership value of a predecessor and δ is the weight of a transition.

3 Some (Fuzzy) Topologies on General Fuzzy Automata 75 In this definition, the process that takes place upon the transition from state q i to q j on input a k is represented as: µ t+1 (q j ) = δ((q i, µ t (q i )), a k, q j ) = F 1 (µ t (q i ), δ(q i, a k, q j )), which means that the membership value (mv) of the state q j at time t + 1 is computed by function F 1 by using both the membership value of q i at time t and the weight of the transition. There are many options which can be used for the function F 1 (µ, δ), for example max{µ, δ}, min{µ, δ} or (µ + δ)/2. (viii) F 2 : [0, 1] [0, 1] is called multi-membership resolution function. The multi-membership resolution function resolves the multi-membership active states and assigns a single membership value to them. Let Q act (t i ) be the set of all active states at time t i, i 0. We have Q act (t 0 ) = R, Q act (t i ) = {(q, µ ti (q)) : q Q act (t i 1 ), a Σ, δ(q, a, q) }, i 1. Since Q act (t i ) is a fuzzy set, in order to show that a state q belongs to Q act (t i ) and T is a subset of Q act (t i ), we should write: q Domain(Q act (t i )) and T Domain(Q act (t i )). Hereafter, we simply denote them as: q Q act (t i ) and T Q act (t i ). The combination of the operations of functions F 1 and F 2 on a multi-membership state q j will lead to the multi-membership resolution algorithm. Algorithm 1.2. [7] (Multi-membership resolution) If there are several simultaneous transitions to the active state q j at time t + 1, the following algorithm will assign a unified membership value to that: (1) Each transition weight δ(q i, a k, q j ) together with µ t (q i ), will be processed by the membership assignment function F 1, and will produce a membership value. Call this v i, v i = δ((q i, µ t (q i )), a k, q j ) = F 1 (µ t (q i ), δ(q i, a k, q j )). (2) These membership values are not necessarily equal. Hence, they will be processed by another function F 2, called the multi-membership resolution function. (3) The result produced by F 2 will be assigned as the instantaneous membership value of the active state q j, Where µ t+1 (q j ) = F n 2 [v i ] = F n 2 [F 1 (µ t (q i ), δ(q i, a k, q j ))]. n : is the number of simultaneous transitions to the active state q j at time t + 1. δ(q i, a k, q j ) : is the weight of the transition from q i to q j upon input a k. µ t (q i ) : is the membership value of q i at time t. µ t+1 (q j ) : is the final membership value of q j at time t + 1. Example 1.3. [7] Consider the GFA in Fig.1 with several transition overlaps. It is specified as: F = (Q, Σ, R, Z, ω, δ, F1, F 2 ), where Q = {q 0, q 1, q 2, q 3, q 4 } is the set of states, Σ = {a, b} is the set of input symbols, Z = and ω is not applicable, R = Qact (t 0 ) = {(q 0, µ t0 (q 0 ))} = {(q 0, 1)}, v i = δ((q i, µ t (q i )), a k, q j ) = F 1 (µ t (q i ), δ(q i, a k, q j )) and

4 76 M. Horry and M. M. Zahedi Figure 1. The GFA of Example n n F 1 (µ, δ) = δ, [v i ] = µ t+1 (q m) = (F 1 (µ t (q i ), δ(q 1, a k, q m))), F 2 2 n n F 1 (µ, δ) = min(µ, δ), F 2 [v i ] = µ t+1 (q m) = (F 1 (µ t (q i ), δ(q 1, a k, q m))), 3 n n F 1 (µ, δ) = min(µ, δ), F 2 [v i ] = µ t+1 (q m) = (F 1 (µ t (q i ), δ(q 1, a k, q m))), 4 n n F 1 (µ, δ) = max(µ, δ), F 2 [v i ] = µ t+1 (q m) = (F 1 (µ t (q i ), δ(q 1, a k, q m))), 5 n n F 1 (µ, δ) = max(µ, δ), F 2 [v i ] = µ t+1 (q m) = (F 1 (µ t (q i ), δ(q 1, a k, q m))), 6 n n F 1 (µ, δ) = min(µ, δ), F 2 [v i ] = µ i+1 (q m) = F 1 (µ t (q i ), δ(q 1, a k, q m))/n, 7 F 1 (µ, δ) = µ + δ n n, F 2 [v i ] = µ t+1 (q m) = (F 1 (µ t (q i ), δ(q 1, a k, q m))), 2 Where n is the number of simultaneous transitions to the active state q m at time t + 1. If we choose 1 n n F 1 (µ, δ) = δ, F 2 [v i ] = µ t+1 (q m) = (F 1 (µ t (q i ), δ(q 1, a k, q m))), then we have : µ t 0 (q 0 ) = 1, µ t 1 (q 1 ) = F 1 (µ t 0 (q 0 ), δ(q 0, a, q 1 )) = F 1 (1, 0.4) = 0.4, µ t 1 (q 4 ) = F 1 (µ t 0 (q 0 ), δ(q 0, a, q 4 )) = F 1 (1, 0.5) = 0.5, µ t 2 (q 1 ) = F 1 (µ t 1 (q 4 ), δ(q 4, a, q 1 )) = F 1 (0.5, 0.4) = 0.4, µ t 2 (q 2 ) = F 1 (µ t 1 (q 1 ), δ(q 1, a, q 2 )) = F 1 (0.4, 0.8) = 0.8, µ t 2 (q 4 ) = F 1 (µ t 1 (q 1 ), δ(q 1, a, q 4 )) = F 1 (0.4, 0.35) = 0.35, µ t 3 (q 2 ) = F 1 (µ t 2 (q 4 ), δ(q 4, b, q 2 )) F 1 (µ t 2 (q 2 ), δ(q 2, b, q 2 )) = F 1 (0.4, 0.1) F 1 (0.8, 0.6) = = 0.1, µ t 3 (q 3 ) = F 1 (µ t 2 (q 1 ), δ(q 1, b, q 3 )) F 1 (µ t 2 (q 2 ), δ(q 2, b, q 3 )) F 1 (µ t 2 (q 4 ), δ(q 4, b, q 3 )) = F 1 (0.4, 0.3) F 1 (0.8, 0.45) F 1 (0.35, 0.7) = = 0.3,

5 Some (Fuzzy) Topologies on General Fuzzy Automata 77 which there are two simultaneous transitions to the active state q 2 at time t 3 and there are three simultaneous transitions to the active state q 3 at time t 3. So we can draw the Table 1. The operation of this fuzzy automaton upon input string a 2 b 2 is shown in Table 1 for different membership assignment functions and multi-membership resolution strategies. In this table, we have considered different cases for combining functions F 1 and F 2. time t 0 t 1 t 2 t 3 t 4 input Λ a a b b Q act(t i) q 0 q 1 q 4 q 1 q 2 q 4 q 2 q 3 q 2 q 3 q 4 mv mv mv mv mv mv mv Table 1. Active States and Their Membership Values (mv) at Different Times in Example 1.3 Definition 1.4. [23] Let F = (Q, Σ, R, Z, ω, δ, F 1, F 2 ) be a general fuzzy automaton, which is defined in Definition 1.1. We defined max-min general fuzzy automata of the form: F = (Q, Σ, R, Z, ω, δ, F 1, F 2 ) such that : δ : Q act Σ Q [0, 1] where Q act = {Q act (t 0 ), Q act (t 1 ), Q act (t 2 ),... } and let for every i, i 0 δ ((q, µ ti (q)), Λ, p) = { 1, q = p 0, otherwise, and for every i, i 1 δ ((q, µ ti 1 (q)), u i, p) = δ((q, µ ti 1 (q)), u i, p), δ ((q, µ ti 1 (q)), u i u i+1, p) = ( δ((q, µ ti 1 (q)), u i, q ) δ((q, µ ti (q )), u i+1, p)), q Q act(t i) and recursively δ ((q, µ t0 (q)), u 1 u 2... u n, p) = { δ((q, µ t0 (q)), u 1, p 1 ) δ((p 1, µ t1 (p 1 )), u 2, p 2 ) δ((p n 1, µ tn 1 (p n 1 )), u n, p) p 1 Q act (t 1 ), p 2 Q act (t 2 ),..., p n 1 Q act (t n 1 )}, in which u i Σ, 1 i n and assuming that the entered input at time t i be u i, 1 i n 1.

6 78 M. Horry and M. M. Zahedi Definition 1.5. [23] Let F be a max-min general fuzzy automaton. The response function r F : Σ Q [0, 1] of F, for any x Σ, q Q, is defined by r F (x, q) = δ ((q, µ t0 (q )), x, q). q Q act(t 0) Definition 1.6. [23] Let q Q and 0 c < 1. Then q is called an accessible state of F with threshold c if there exists x Σ such that r F (x, q) > c. Definition 1.7. [23] Let A Q. Then F is said to be connected with threshold c on A, if A = Q c, where Q c is the set of all accessible states with threshold c. Definition 1.8. [12] Let X be an arbitrary set. The function ψ : P (X) P (X) is called a Kuratowski fuzzy interior operator if for any two elements λ and γ of P (X), we have (i) ψ(k) = k, k constant, (ii)ψ(λ) λ, (iii)ψ(λ γ) = ψ(λ) ψ(γ), (iv) ψ(ψ(λ)) = ψ(λ). Definition 1.9. [12] A subset τ of P (X) is called a Lowen -type fuzzy topology on X if (i) k constant, k τ, (ii) If λ 1, λ 2 τ, then λ 1 λ2 τ, (iii) If λ i τ, i I, then i I λ i τ. Remark [12] Let τ be a Lowen-type fuzzy topology on a set X. Then for any α [0, 1), l α (τ) = {λ 1 (α, 1] : λ τ} is a topology on X, referred to as the α-level topology of τ. Remark [12] Let ψ be a Kuratowski fuzzy interior operator.then ψ induces a Lowen-type fuzzy topology of the form τ(ψ) = {λ : ψ(λ) = λ}. Definition [11] Let X be an arbitrary set. The function ψ : P (X) P (X) is called a closure operator on X, if for any two elements A and B of P (X), we have (i) ψ( ) =, (ii) A ψ(a), (iii) ψ(a B) = ψ(a) ψ(b), (iv) ψ(ψ(a)) = ψ(a). Definition [11] Let X be an arbitrary set. The function ψ : P (X) P (X) is called a fuzzy closure operator if for any two elements λ and γ of P (X), we have (i) ψ(0) = 0, (ii) ψ(λ) λ, (iii) ψ(λ γ) = ψ(λ) ψ(γ), (iv) ψ(ψ(λ)) = ψ(λ). Also, a fuzzy closure operator ψ is called saturation if for any family {λ i } i I of elements of P (X), we have ψ( i I λ i) = i I ψ(λ i).

7 Some (Fuzzy) Topologies on General Fuzzy Automata 79 Definition [11] Let X be an arbitrary set. The function ψ : P (X) P (X) is called a fuzzy interior operator if for any two elements λ and γ of P (X), we have (i) ψ(1) = 1, (ii) ψ(λ) λ, (iii) ψ(λ γ) = ψ(λ) ψ(γ), (iv) ψ(ψ(λ)) = ψ(λ). Definition [4] A subset τ of P (X) is called a Chang-type fuzzy topology on X if (i) 0, 1 τ, (ii) If λ 1, λ 2 τ, then λ 1 λ2 τ, (iii) If λ i τ, i I, then i I λ i τ. Remark [12] Let ψ be a fuzzy closure operator. Then ψ induces a Changtype fuzzy topology of the form τ(ψ) = {λ C : ψ(λ) = λ}, where λ C = 1 λ. Let ψ be a fuzzy interior operator. Then ψ induces a Chang-type fuzzy topology of the form τ(ψ) = {λ : ψ(λ) = λ}. 2. Some Lowen-type Fuzzy Topologies on General Fuzzy Automata Theorem 2.1. Let F = (Q, Σ, R, Z, ω, δ, F 1, F 2 ) be a max-min general fuzzy automaton, λ be a fuzzy subset on Q and for any p Q, consider D(λ)(p) = {λ(p) r F (x, p) : x Σ }. Define S λ : P (Q) P (Q) A S λ (A) where S λ (q) = {p Q : D(λ)(p) = D(λ)(q)}, S λ (A) = S λ (q). Then S λ is a closure operator on Q. Proof. (i) S λ ( ) = is obvious. (ii) Let q A. Since q S λ (q) and S λ (q) S λ (A), we get that A S λ (A). (iii) S λ (A B) = q A B S λ (q) = ( q A S λ (q)) ( q B S λ (q)) = S λ (A) Sλ (B). (iv) By (ii), we have S λ (A) S λ ( S λ (A)). Conversely, let q S λ ( S λ (A)). Then there exists q S λ (A) such that q S λ (q ). Thus q S λ (q ), for some q A. Consequently, we have D(λ)(q) = D(λ)(q ), D(λ)(q ) = D(λ)(q ). So D(λ)(q) = D(λ)(q ). Hence, q S λ (q ) S λ (A). Therefore S λ ( S λ (A)) S λ (A). Theorem 2.2. Let F = (Q, Σ, R, Z, ω, δ, F 1, F 2 ) be a max-min general fuzzy automaton, λ be a fuzzy subset on Q and S λ (A) = S q A λ (q). Then T (Q) = {A C : A Q, S λ (A) = A} is a topology on Q. q A

8 80 M. Horry and M. M. Zahedi Proof. (i) Since S λ ( ) =, then Q = ( ) C T (Q). (ii) Since S λ is a closure operator on Q, we have Q S λ (Q). On the other hand, since S λ (Q) P (Q), we have S λ (Q) Q. Thus, Sλ (Q) = Q. Therefore we conclude that = (Q) C T (Q). (iii) Let A C 1 and A C 2 belong to T (Q). Then S λ (A 1 ) = A 1 and S λ (A 2 ) = A 2. Thus, we have S λ (A 1 A2 ) = S λ (A 1 ) Sλ (A 2 ) = A 1 A1. That is, A C 1 A C 2 = (A 1 A2 ) C T (Q). (iv) Let A C i T (Q), i I. Then S λ (A i ) = A i, i I. Since S λ is a closure operator on Q, we have i I A i S λ ( i I A i). On the other hand, since A i ( i I A i) = A i, we get that S λ (A i ) ( S λ ( i I A i)) = S λ (A i ). Then S λ ( i I A i) S λ (A i ) = A i. Thus S λ ( i I A i) i I A i. Hence, S λ ( i I A i) = i I A i. That is, i I AC i = ( i I A i) C T (Q). Theorem 2.3. Let F = (Q, Σ, R, Z, ω, δ, F 1, F 2 ) be a max-min general fuzzy automaton. Define where C : P (Q) P (Q) λ C(λ) C(λ)(p) = {λ(q) : q S λ (p)}. Then C is a Kuratowski fuzzy interior operator. Proof. (i) C(k) = k, k constant. (ii) Since p S λ (p) = {q Q : D(λ)(q) = D(λ)(p)}, we have C(λ)(p) = {λ(q) : q S λ (p)} λ(p). (iii) C(λ 1 λ2 )(p) = {λ 1 (q) λ 2 (q) : q S λ (p)} = ( {λ 1 (q) : q S λ (p)}) ( {λ 2 (q) : q S λ (p)}) = C(λ 1 (p)) C(λ 2 )(p). (iv) By (ii), we have C(C(λ))(p) C(λ)(p). For the reverse inequality, we have C(C(λ))(p) = {C(λ)(q) : q S λ (p)} = { {λ(r) : r S λ (q)} : q S λ (p)}. Since for any r S λ (q) and q S λ (p), we have D(λ)(r) = D(λ)(q) and D(λ)(q) = D(λ)(p). Then D(λ)(r) = D(λ)(p). Thus r S λ (p). Therefore we have C(C(λ))(p) {λ(r) : r S λ (p)} = C(λ)(p). Consequently, C(C(λ)) = C(λ). Now, by Remark 1.11, we conclude that : Corollary 2.4. Let F = (Q, Σ, R, Z, ω, δ, F 1, F 2 ) be a max-min general fuzzy automata and τ(c) = {λ P (Q) : C(λ) = λ}. Then τ(c) is a Lowen-type fuzzy topology on Q. Theorem 2.5. Let F = (Q, Σ, R, Z, ω, δ, F 1, F 2 ) be a max-min general fuzzy automaton. Then l α (τ(c)) = {A : A Q, A C T (Q)}, α [0, 1).

9 Some (Fuzzy) Topologies on General Fuzzy Automata 81 Proof. Let α [0, 1). By Remark 1.10, we have : l α (τ(c)) = {λ 1 (α, 1] : λ τ(c)} = {λ 1 (α, 1] : λ P (Q) : C(λ) = λ}. Let A = λ 1 (α, 1] l α (τ(c)). Then C(λ) = λ. Since S λ is a closure operator on Q, thus A S λ (A). On the other hand, let p S λ (A). Then p S λ (q), for some q A. Thus D(λ)(p) = D(λ)(q) and λ(q) > α. So q S λ (p). Now, we have λ(p) = C(λ)(p) = {λ(q) : q S λ (p)} > α. Hence, p A. Therefore S λ (A) A. So we get that A C T (Q). Consequently, l α (τ(c)) {A : A Q, A C T (Q)}. Conversely, let A C T (Q). Then S λ (A) = A. To prove A l α (τ(c)), it is enough to show that (i) A = 1 1 A (α, 1], (ii) C(1 A ) = 1 A. (i) is true. To show part (ii), let p Q. If C(1 A )(p) = 1, then (ii) follows obviously. If C(1 A )(p) < 1, then there exists q S λ (p) such that 1 A (q) < 1. Thus q A = S λ (A). Since q S λ (p), then p A. Hence 1 A (p) = 0. Therefore C(1 A )(p) = {1 A (q) : q S λ (p)} 1 A (p). On the other hand, since C is a Kuratowski fuzzy interior operator, we have C(1 A )(p) 1 A (p). Consequently, C(1 A ) = 1 A. Thus l α (τ(c)) {A : A Q, A C T (Q)}. Definition 2.6. Let F be a max-min general fuzzy automaton, p Q, q Q act (t i ), i 0 and 0 c < 1. Then p is called a successor of q with threshold c if there exists x Σ such that δ ((q, µ ti (q)), x, p) > c. Definition 2.7. Let F be a max-min general fuzzy automaton, q Q act (t i ), i 0 and 0 c < 1. Also let S c (q) denote the set of all successors of q with threshold c. If T Q, then S c (T ) the set of all successors of T with threshold c is defined by S c (T ) = {S c (q) : q T }. Theorem 2.8. Let F be a max-min general fuzzy automaton and 0 c < 1. Then (i) q S c (q), q Q. (ii) If r S c (p), p S c (q), then r S c (q). Proof. (i) Since for all q Q, there exists i 0 such that q Q act (t i ) and δ ((q, µ ti (q)), Λ, q) = 1 > c, then q S c (q). (ii) Since p S c (q), then q Q act (t i ) and there exists x Σ such that δ ((q, µ ti (q)), x, p) > c. Also, r S c (p) implies p Q act (t j ) and there exists y Σ such that δ ((p, µ tj (p)), y, r) > c. Thus, we have δ ((q, µ ti (q)), xy, r) = q Q act(t j) [ δ ((q, µ ti (q)), x, q ) δ ((q, µ tj (q )), y, r)] δ ((q, µ ti (q)), x, p) δ ((p, µ tj (p)), y, r) > c. So we get that r S c (q). Example 2.9. Let F be a max-min general fuzzy automaton in Example 1.3. If we choose 5 n n F 1 (µ, δ) = max(µ, δ), [v i ] = µ t+1 (q m ) = (F 1 (µ t (q i ), δ(q 1, a k, q m ))), F 2

10 82 M. Horry and M. M. Zahedi then we have : δ ((q 0, µ t0 (q 0 )), a 2, q 1 ) = [ δ ((q 0, µ t0 (q 0 )), a, q ) q Q act(t 1) δ ((q, µ t1 (q )), a, q 1 )] = [ δ((q 0, µ t0 (q 0 )), a, q 1 ) δ((q1, µ t1 (q 1 )), a, q 1 )] [ δ((q 0, µ t0 (q 0 )), a, q 4 ) δ((q4, µ t1 (q 4 )), a, q 1 )] = [F 1 (µ t0 (q 0 ), δ(q 0, a, q 1 )) F1 (µ t1 (q 1 ), δ(q 1, a, q 1 ))] [F 1 (µ t0 (q 0 ), δ(q 0, a, q 4 )) F1 (µ t1 (q 4 ), δ(q 4, a, q 1 ))] = [F 1 (1, 0.4) F 1 (1, 0)] [F 1 (1, 0.5) F 1 (1, 0.4)] = [1 1] [1 1] = 1 1 = 1. Thus δ ((q 0, µ t0 (q 0 )), a 2, q 1 ) = 1. Similarly, we have : δ ((q 0, µ t0 (q 0 )), a 2, q 2 ) = 1, δ ((q 0, µ t0 (q 0 )), a 2, q 3 ) = 1, δ ((q 0, µ t0 (q 0 )), a 2, q 4 ) = 1. Also, we have δ ((q 0, µ t0 (q 0 )), Λ, q 0 ) = 1. Thus S c (q 0 ) = Q, c, 0 c < 1. Theorem Let F be a max-min general fuzzy automaton, 0 c < 1 and Then S c is a closure operator on Q. S c : P (Q) P (Q) A S c (A). Proof. (i) S c ( ) =. (ii) Let q A Q. By Theorem 2.8, q S c (q). Then q S c (q) S c (A), Thus, A S c (A). (iii)s c (A B) = q A B S c(q) = ( q A S c(q)) ( q B S c(q)) = S c (A) S c (B). (iv) By (ii), we have S c (A) S c (S c (A)). Conversely, let p S c (S c (A)). Then there exists q S c (A) such that p S c (q ). Thus q S c (q ), for some q A. Consequently, by Theorem 2.8, p S c (q ). Hence, p S c (A). Therefore S c (S c (A)) S c (A). Theorem Let F be a max-min general fuzzy automaton. Then τ = {A C : A Q, S c (A) = A} is a topology on Q. Proof. The proof is similar to that of Theorem 2.2, by using suitable modification. Definition Let F be a max-min general fuzzy automaton and 0 c < 1. Then we say that F is good with threshold c, if q Q, q Q act (t 0 ) : q S c (q ). Example Let F be the max-min general fuzzy automaton in Example 2. 9 and 0 c < 1. Since S c (q 0 ) = Q, then F is good with threshold c.

11 Some (Fuzzy) Topologies on General Fuzzy Automata 83 Theorem Let F be a max-min general fuzzy automaton and 0 c < 1. Then F is good with threshold c if and only if F is connected with threshold c on Q. Proof. Let F be good with threshold c and q Q. Then there exists q Q act (t 0 ) such that q S c (q ). Thus, there exists x Σ such that δ ((q, µ t0 (q )), x, q) > c. So r F (x, q) = δ ((q, µ t0 (q )), x, q) > c. q Q act(t 0) Consequently, by Definitions 1.6 and 1.7, F is connected with threshold c on Q. Conversely, let F be connected with threshold c on Q and q Q. Then there exists x Σ such that r F (x, q) = δ ((q, µ t0 (q )), x, q) > c. q Q act(t 0) Hence, there exists q Q act (t 0 ) such that δ ((q, µ t0 (q )), x, q) > c. So q S c (q ). Therefore F is good with threshold c. Theorem Let F be a max-min general fuzzy automaton, 0 c < 1 and suppose that pr c q p S c (q), q S c (p). Then R c is an equivalence relation on Q. Proof. By Theorem 2.8 the proof is obvious. Theorem Let F be a max-min general fuzzy automaton, 0 c < 1, A Q, B c (q) = {p Q : pr c q} and B c (A) = q A B c(q). Define Then B c is a closure operator on Q. Proof. (i) B c ( ) =. B c : P (Q) P (Q) A B c (A). (ii) Let q A. Since qr c q, then q B c (q) B c (A). Thus, A B c (A). (iii)b c (A D) = q A D B c(q) = ( q A B c(q)) ( q D B c(q)) = B c (A) B c (D). (iv) By (ii), we have B c (A) B c (B c (A)). Conversely, let q B c (B c (A)). Then there exists q B c (A) such that q B c (q ). Thus q B c (q ), for some q A. Consequently, qr c q and q R c q. By Theorem 2.15, qr c q. Thus q B c (q ) B c (A). Therefore B c (B c (A)) B c (A). Theorem Let F be a max-min general fuzzy automaton, 0 c < 1 and p, q Q. Then p B c (q) if and only if B c (p) = B c (q). Proof. Let p B c (q). If q B c (q), then q R c p and pr c q. Thus, q R c q. Therefore we get that q B c (q). So B c (p) B c (q). On the other hand, if q B c (q), then q R c q and pr c q. Thus, q R c p. So we get that q B c (q). Consequently, B c (q) B c (p). Conversely, let B c (p) = B c (q). Since pr c p, then p B c (p). Thus, p B c (q). 3. Some Chang-type Fuzzy Topologies on General Fuzzy Automata Theorem 3.1. Let F = (Q, Σ, R, Z, ω, δ, F 1, F 2 ) be a max-min general fuzzy automaton and λ be a fuzzy subset on Q. Define

12 84 M. Horry and M. M. Zahedi where and D : P (Q act (t 0 )) P (Q act (t 0 )) λ D(λ) D(λ)(p) = {λ(p) r F (x, p) : x Σ } r F (x, p) = q Q act(t 0) δ ((q, µ t0 (q )), x, p). Then D is a saturation fuzzy closure operator. Proof. (i) D(0) = 0 is obvious. (ii) D(λ)(p) = {λ(p) r F (x, p) : x Σ } λ(p) r F (Λ, p) = λ(p) 1 = λ(p). (iii) D( i I λ i)(p) = {( i I λ i(p)) r F (x, p) : x Σ } = ( {λ i (p) r F (x, p) : x Σ }) = i I i I D(λ i )(p). (iv) By (ii), we have D( D(λ))(p) D(λ)(p). For the reverse inequality, let x, x be fixed and x x. Then we have λ(p) r F (x, p) r F (x, p) λ(p) r F (x, p) {λ(p) r F (z, p) : z Σ } = D(λ)(p). Then D( D(λ))(p) = { D(λ)(p) r F (x, p) : x Σ } = {λ(p) r F (x, p) r F (x, p)} D(λ)(p). x Σ x Σ Thus, D( D(λ)) = D(λ). Now, by Remark 1.16, we conclude that: Corollary 3.2. In Theorem 3.1, let T 1 = {λ P (Q act (t 0 )) : D(λ) = λ}. Then τ( D) = {λ C : λ T 1 } is a Chang-type fuzzy topology on Q act (t 0 ). Theorem 3.3. Let F = (Q, Σ, R, Z, ω, δ, F 1, F 2 ) be a max-min general fuzzy automaton, λ be a fuzzy subset on Q and 0 c < 1. Then D(λ) > c if and only if F is connected with threshold c on Q act (t 0 ), and λ(p) > c, p Q act (t 0 ). Proof. Let F be connected with threshold c on Q act (t 0 ) and λ(p) > c, p Q act (t 0 ). Then for any p Q act (t 0 ), there exists x Σ such that r F (x, p) > c. Thus λ(p) r F (x, p) > c which implies that D(λ)(p) = {λ(p) r F (x, p) : x Σ } > c. Conversely, let D(λ) > c. Then we have D(λ)(p) = {λ(p) r F (x, p) : x Σ } > c. Thus, for p Q act (t 0 ), there exists x Σ such that r F (x, p) > c and λ(p) > c. So F is connected with threshold c on Q act (t 0 ) and λ(p) > c, p Q act (t 0 ).

13 Some (Fuzzy) Topologies on General Fuzzy Automata 85 Definition 3.4. Let F = (Q, Σ, R, Z, ω, δ, F 1, F 2 ) be a max-min general fuzzy automaton, λ be a fuzzy subset on Q and A Q. Then we say that λ is normal on A if λ(p) r F (x, p), p A, x Σ. Theorem 3.5. Let F = (Q, Σ, R, Z, ω, δ, F 1, F 2 ) be a max-min general fuzzy automaton, λ be a fuzzy subset on Q. If λ is normal on Q act (t 0 ), then λ C τ( D). Proof. Let λ be normal on Q act (t 0 ). Then we have Therefore λ C τ( D). λ(p) r F (x, p), p Q act (t 0 ), x Σ λ(p) r F (x, p) = λ(p), x Σ D(λ)(p) = {λ(p) r F (x, p) : x Σ } = λ(p) D(λ) = λ. Theorem 3.6. Let F = (Q, Σ, R, Z, ω, δ, F 1, F 2 ) be a max-min general fuzzy automaton, λ be a fuzzy subset on Q and Q = Q Q act (t 0 ). Suppose where and D : P (Q ) P (Q ) λ D(λ) D(λ)(p) = {λ(p) R F (x, p) : x Σ } R F (x, p) = Then D is a fuzzy interior operator. q Q act(t 0) δ ((q, µ t0 (q)), x, p). Proof. (i) D(1) = 1 is obvious. (ii) D(λ)(p) = {λ(p) R F (x, p) : x Σ } λ(p) R F (Λ, p) = λ(p) 0 = λ(p). (iii)d(λ 1 λ 2 )(p) = {(λ 1 (p) λ 2 (p)) R F (x, p) : x Σ } = ( {(λ 1 (p) R F (x, p) : x Σ }) ( {(λ 2 (p) R F (x, p) : x Σ }) = D(λ 1 )(p) D(λ 2 )(p). (iv) By (ii), we have D(D(λ))(p) D(λ)(p). For the reverse inequality, let x, x be fixed and x x. Then we have λ(p) R F (x, p) R F (x, p) λ(p) R F (x, p) Then {λ(p) R F (z, p) : z Σ } = D(λ)(p). D(D(λ))(p) = {D(λ)(p) R F (x, p) : x Σ } = x Σ x Σ {λ(p) R F (x, p) R F (x, p)} D(λ)(p). Thus, D(D(λ)) = D(λ). Now, by Remark 1.16, we conclude that:

14 86 M. Horry and M. M. Zahedi Corollary 3.7. In Theorem 3.6, let τ(d) = {λ P (Q ) : D(λ) = λ}. Then τ(d) is a Chang-type fuzzy topology on Q. Definition 3.8. Let F = (Q, Σ, R, Z, ω, δ, F 1, F 2 ) be a max-min general fuzzy automaton, λ be a fuzzy subset on Q and A Q. Then we say that λ is subnormal on A if λ(p) R F (x, p), p A, x Σ. Theorem 3.9. Let F = (Q, Σ, R, Z, ω, δ, F 1, F 2 ) be a max-min general fuzzy automaton, λ be a fuzzy subset on Q and Q = Q Q act (t 0 ). If λ is subnormal on Q, then λ τ(d). Proof. Let λ be subnormal on Q. Then we have Therefore λ τ(d). λ(p) R F (x, p), p Q, x Σ λ(p) R F (x, p) = λ(p), x Σ D(λ)(p) = {λ(p) R F (x, p) : x Σ } = λ(p) D(λ) = λ. Theorem Let F = (Q, Σ, R, Z, ω, δ, F 1, F 2 ) be a max-min general fuzzy automaton, λ be a fuzzy subset on Σ and Σ = {x Σ : r F (x, p) = 1, p Q act (t 0 )}. Consider B : P (Σ ) P (Σ ) λ B(λ) where B(λ)(x) = {λ(x) r F (x, p) : p Q} and r F (x, p) = q Q act(t 0) δ ((q, µ t0 (q )), x, p). Then B is a saturation fuzzy closure operator. Proof. (i) B(0) = 0 is obvious. (ii) B(λ)(x) = {λ(x) r F (x, p) : p Q} λ(x) r F (Λ, p 0 ) = λ(x) 1 = λ(x), where p 0 Q act (t 0 ). (iii) B( i I λ i )(x) = {( i I λ i (x)) r F (x, p) : p Q} = i I ( {λ i (x) r F (x, p) : p Q}) = i I B(λi )(x). (iv) By (ii), we have B( B(λ))(x) B(λ)(x). For the reverse inequality, let p, q Q be fixed and p q. Then we have λ(x) r F (x, q) r F (x, p) λ(x) r F (x, p) Then {λ(x) r F (x, r) : r Q} = B(λ)(x). B( B(λ))(x) = { B(λ)(x) r F (x, p) : p Q} = p Q q Q {λ(x) r F (x, q) r F (x, p)} B(λ)(x).

15 Some (Fuzzy) Topologies on General Fuzzy Automata 87 Thus, B( B(λ)) = B(λ). Now, by Remark 1.16, we conclude that: Corollary In Theorem 3.10, let T 2 = {λ P (Σ ) : B(λ) = λ}. Then τ( B) = {λ C : λ T 2 } is a Chang-type fuzzy topology on Σ. Theorem Let F = (Q, Σ, R, Z, ω, δ, F 1, F 2 ) be a max-min general fuzzy automaton, λ be a fuzzy subset on Σ, Q = Q Q act (t 0 ) and Σ = {x Σ : R F (x, p) = 0, p Q }. Define where and Then B is a fuzzy interior operator. Proof. (i) B(1) = 1 is obvious. B : P (Σ ) P (Σ ) λ B(λ) B(λ)(x) = {λ(x) R F (x, p) : p Q} R F (x, p) = q Qact (t 0 ) δ ((q, µ t 0 (q)), x, p). (ii) B(λ)(x) = {λ(x) R F (x, p) : p Q} λ(x) R F (x, p 0 ) = λ(x) 0 = λ(x), where p 0 Q. (iii)b(λ 1 λ 2 )(x) = {(λ 1 (x) λ 2 (x)) R F (x, p) : p Q} = ( {(λ 1 (x) R F (x, p) : p Q}) ( {(λ 2 (x) R F (x, p) : p Q}) = B(λ 1 )(x) B(λ 2 )(x). (iv) By (ii), we have B(B(λ))(p) B(λ)(p). For the reverse inequality, let p, q Q be fixed and p q. Then we have Then Thus, B(B(λ)) = B(λ). λ(x) R F (x, p) R F (x, q) λ(x) R F (x, p) {λ(x) R F (x, r) : r Q} = B(λ)(x). B(B(λ))(x) = {B(λ)(x) R F (x, p) : p Q} = p Q q Q {λ(x) R F (x, p) R F (x, q)} B(λ)(x). Now, by Remark 1.16, we conclude that Corollary In Theorem 3.12, let τ(b) = {λ P (Σ ) : B(λ) = λ}. Then τ(b) is a Chang-type fuzzy topology on Σ. 4. Conclusions As it is shown, in this manuscript by considering (in fact proving) some fuzzy interior (closure) operators, we have introduced some Lowen-type and Chang-type fuzzy topology structures on general fuzzy automata. So these results open a way to study deeply the properties of these structures in future, for example to discuss open sets, closed sets, Hausdorff spaces and compact spaces. Acknowledgements. The authors would like to thank the reviewers and editorial board for their valuable comments and suggestions.

16 88 M. Horry and M. M. Zahedi References [1] M. A. Arbib, From automata theory to brain theory, International Journal of Man-Machin Studies, 7(3) (1975), [2] W. R. Ashby, Design for a brain, Chapman and Hall, London, [3] D. Ashlock, A. Wittrock and T. Wen, Training finite state machines to improve PCR primer design, In: Proceedings of the 2002 Congress on Evolutionary Computation (CEC), 20 (2002). [4] C. Cattaneo, P. Flocchini, G. Mauri, C. Q. Vogliotti and N. Santoro, Cellular automata in fuzzy backgrounds, Physica D, 105 (1997), [5] C. L. Chang, Fuzzy topological spaces, J. Math. Anal. Appl., 24 (1968), [6] P. Das, A fuzzy topology associated with a fuzzy finite state machine, Fuzzy Sets and Systems, 105 (1999), [7] M. Doostfatemeh and S.C. Kremer, New directions in fuzzy automata, International Journal of Approximate Reasoning, 38 (2005), [8] B. R. Gaines and L. J. Kohout, The logic of automata, International Journal of General Systems, 2 (1976), [9] M. Horry and M. M. Zahedi, On general fuzzy recognizers, Iranian Journal of Fuzzy Systems, 8(3) (2011), [10] M. Horry and M. M. Zahedi, Hypergroups and general fuzzy automata, Iranian Journal of Fuzzy Systems, 6(2) (2009), [11] K. Kuratowski, Topology, Academic Presss, [12] R. Lowen, Fuzzy topological spaces and fuzzy compactness, Journal of Mathematical Analysis and Applications, 56 (1976), [13] R. Maclin and J. Shavlik, Refing domain theories expressed as finite-state automata, In: L.B.G. Collins (ed.), Proceedings of the 8th International Workshop on Machine Learning (ML 91), Morgan Kaufmann, San Mateo CA, [14] R. Maclin and J. Shavlik, Refing algorithm with knowledge-based neural networks: improving the choufasma algorithm for protein folding, In: S. Hanson, G. Drastal, R. Rivest (eds.), Computational Learning Theory and Natural Learning Systems, MIT Press, Cambridge, MA, [15] J. N. Mordeson and D. S. Malik, Fuzzy Automata and Languages, Theory and Applications, Chapman and Hall/CRC, London/Boca Raton, FL, [16] W. Omlin, K. K. Giles and K. K. Thornber, Equivalence in knowledge representation: automata, rnns, and dynamical fuzzy systems, Proceeding of IEEE, 87(9) (1999), [17] W. Omlin, K. K. Thornber and K. K. Giles, Fuzzy finite-state automata can be deterministically encoded into recurrent neural networks, IEEE Transactions on Fuzzy Systems, 5(1) (1998), [18] B. Tucker (ed.), The computer science and engineering handbook, CRC Press, Boca Raton, FL, [19] J. Virant and N. Zimic, Fuzzy automata with fuzzy relief, IEEE Transactions on Fuzzy Systems, 3(1) (1995), [20] W. G. Wee, On generalization of adaptive algorithm and application of the fuzzy sets concept to pattern classification, Ph.D. dissertation, Purdue University, Lafayette, IN, [21] M. Ying, A formal model of computing with words, IEEE Transactions on Fuzzy Systems 10(5) (2002), [22] L. A. Zadeh, Fuzzy sets, Information and Control, 8 (1965), [23] M. M. Zahedi, M. Horry and K. Abolpor, Bifuzzy (General) topology on max-min general fuzzy automata, Advanced in Fuzzy Mathematics, 3(1) (2008),

17 Some (Fuzzy) Topologies on General Fuzzy Automata 89 M. Horry, Shahid Chamran University of Kerman, Kerman, Iran address: Chamran edu.ir M. M. Zahedi, Department of Mathematics, Shahid Bahonar University of Kerman, Kerman, Iran address: zahedi *Corresponding author