Partial actions and automata

We use the notion of a partial action of a monoid to introduce a generalization of automata, which we call ``a preautomaton''. We study properties of preautomata and of languages recognized by preautomata.

Gespeichert in:
Bibliographische Detailangaben
Datum:2011
Hauptverfasser: Dokuchaev, M., Novikov, B., Zholtkevych, G.
Format: Artikel
Sprache:English
Veröffentlicht: Інститут прикладної математики і механіки НАН України 2011
Schriftenreihe:Algebra and Discrete Mathematics
Online Zugang:https://nasplib.isofts.kiev.ua/handle/123456789/154801
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Назва журналу:Digital Library of Periodicals of National Academy of Sciences of Ukraine
Zitieren:Partial actions and automata / M. Dokuchaev, B. Novikov, G. Zholtkevych // Algebra and Discrete Mathematics. — 2011. — Vol. 11, № 2. — С. 51–63. — Бібліогр.: 7 назв. — англ.

Institution

Digital Library of Periodicals of National Academy of Sciences of Ukraine
id nasplib_isofts_kiev_ua-123456789-154801
record_format dspace
spelling nasplib_isofts_kiev_ua-123456789-1548012025-02-09T10:07:23Z Partial actions and automata Dokuchaev, M. Novikov, B. Zholtkevych, G. We use the notion of a partial action of a monoid to introduce a generalization of automata, which we call ``a preautomaton''. We study properties of preautomata and of languages recognized by preautomata. This paper was partially supported by CNPq and FAPESP (Brazil). 2011 Article Partial actions and automata / M. Dokuchaev, B. Novikov, G. Zholtkevych // Algebra and Discrete Mathematics. — 2011. — Vol. 11, № 2. — С. 51–63. — Бібліогр.: 7 назв. — англ. 1726-3255 2000 Mathematics Subject Classification:20M30, 20M35, 68Q70 https://nasplib.isofts.kiev.ua/handle/123456789/154801 en Algebra and Discrete Mathematics application/pdf Інститут прикладної математики і механіки НАН України
institution Digital Library of Periodicals of National Academy of Sciences of Ukraine
collection DSpace DC
language English
description We use the notion of a partial action of a monoid to introduce a generalization of automata, which we call ``a preautomaton''. We study properties of preautomata and of languages recognized by preautomata.
format Article
author Dokuchaev, M.
Novikov, B.
Zholtkevych, G.
spellingShingle Dokuchaev, M.
Novikov, B.
Zholtkevych, G.
Partial actions and automata
Algebra and Discrete Mathematics
author_facet Dokuchaev, M.
Novikov, B.
Zholtkevych, G.
author_sort Dokuchaev, M.
title Partial actions and automata
title_short Partial actions and automata
title_full Partial actions and automata
title_fullStr Partial actions and automata
title_full_unstemmed Partial actions and automata
title_sort partial actions and automata
publisher Інститут прикладної математики і механіки НАН України
publishDate 2011
url https://nasplib.isofts.kiev.ua/handle/123456789/154801
citation_txt Partial actions and automata / M. Dokuchaev, B. Novikov, G. Zholtkevych // Algebra and Discrete Mathematics. — 2011. — Vol. 11, № 2. — С. 51–63. — Бібліогр.: 7 назв. — англ.
series Algebra and Discrete Mathematics
work_keys_str_mv AT dokuchaevm partialactionsandautomata
AT novikovb partialactionsandautomata
AT zholtkevychg partialactionsandautomata
first_indexed 2025-11-25T16:07:04Z
last_indexed 2025-11-25T16:07:04Z
_version_ 1849779106931015680
fulltext Algebra and Discrete Mathematics RESEARCH ARTICLE Volume 11 (2011). Number 2. pp. 51 – 63 c© Journal “Algebra and Discrete Mathematics” Partial actions and automata M. Dokuchaev, B. Novikov, G. Zholtkevych Abstract. We use the notion of a partial action of a monoid to introduce a generalization of automata, which we call “a preau- tomaton”. We study properties of preautomata and of languages recognized by preautomata. Introduction The concept of the partial action has been introduced in [2] for groups and extended to monoids in [7]. Therefore it is natural to investigate the influence in Automata Theory of the replacement of the full action of a free monoid by a partial action. Our article is devoted to the study of this topic. We propose the term “preaction” for a notion which is called “partial action” in [7] and “strong partial action” in [4]. A generalization of the notion of an automaton which appears here, will be called “a preautomaton”. This change of the terminology is caused by the fact that the term “partial automaton” is widely used in Automata Theory in a different sense (see, e.g., [3]). We shall use the following facts from Automata Theory: the Kleene theorem on regular languages, the Myhill–Nerode theorem on languages and congruences, the Eilenberg theorem on prefix decomposition. All of them can be found in [1, 3, 5] etc. This paper was partially supported by CNPq and FAPESP (Brazil). 2000 Mathematics Subject Classification: 20M30, 20M35, 68Q70. Key words and phrases: Partial action, monoid, automaton, recognizable lan- guage, prefix code. 52 Partial actions and automata The paper begins with a preliminary section on preactions. Then we define a preautomaton (Section 2) and describe its globalization. These notions are illustrated by an example in Section 3. Next we pass to languages which are recognized by preautomata. The Eilenberg theorem is generalized and examples are given which show that the Kleene theorem does not hold for preautomata (Section 6). In the last section we compare the preautomata with other classes of machines. We use the notation ϕ : A 99K B for a partial mapping from A to B (unlike to a full mapping A → B). If ϕ(a) is not defined for a ∈ A we write ϕ(a) = ∅. The free semigroup on the alphabet Σ is denoted by Σ+, the free monoid on Σ by Σ∗ and its identity element by ε. We use right preactions, as it is customary in Automata Theory to deal with right actions. 1. Preactions We shall need some known information about partial actions (preactions) of monoids which we recall in this section. Definition 1. [7] Let M be a monoid with the identity e, X a set. A preaction M on X is a partial mapping X × M 99K X : (x, a) 7→ xa, such that ∀x ∈ X xe = x, ∀a, b ∈ M ∀x ∈ X xa 6= ∅ & (xa)b 6= ∅ =⇒ x(ab) 6= ∅ & (xa)b = x(ab), ∀a, b ∈ M ∀x ∈ X xa 6= ∅ & x(ab) 6= ∅ =⇒ (xa)b 6= ∅ & (xa)b = x(ab). The preactions of the given monoid M form a category PAM : its objects are sets with preactions given on them, and morphisms are such mappings ϕ : X → Y that ∀a ∈ M ∀x ∈ X xa 6= ∅ =⇒ ϕ(x)a 6= ∅ & ϕ(xa) = ϕ(x)a. Preactions appear in the following situation. Suppose that a full action of a monoid M is given on a set Y and X ⊂ Y . Then the restriction of the action to X is a preaction. Conversely, let X ×M 99K X be a preaction and Y ⊃ X. Definition 2. A full action Y ×M → Y is called a globalization if its restriction on X coincides with the given preaction. The following construction gives a globalization for any preaction X ×M 99K X. Define on the set X ×M a relation ⊢: (x, ab) ⊢ (xa, b) ⇐⇒ xa 6= ∅. M. Dokuchaev, B. Novikov, G. Zholtkevych 53 Let ≃ is the equivalence generated by ⊢ and Y = (X×M)/ ≃. Denote by [x, a] the ≃-class containing the pair (x, a). Set [x, a]b = [x, ab] for [x, a] ∈ Y, b ∈ M . This defines a full action on Y . Theorem 1. [7] The above defined action Y ×M → Y is a globalization of the preaction X ×M 99K X; the mapping α : X → Y : x 7→ [x, e] is an injective morphism, and any morphism of the preaction M on X in a full action of M can be factored through α. 2. Preautomata We will use the definition of an automaton in the following form (in this section the condition of finiteness is ignored): Definition 3. Let X be a set and Σ∗ the free monoid over an alphabet Σ. An automaton is a triple (Σ, X, δ∗) where δ∗ : X × Σ∗ → X is a mapping such that ∀x ∈ X δ∗(x, ε) = x, (1) ∀u, v ∈ Σ∗ ∀x ∈ X δ∗(x, uv) = δ∗(δ∗(x, u, ), v). (2) The main object of this article is a more general concept: Definition 4. A preautomaton is such a triple (Σ, X, δ∗), where δ∗ : X × Σ∗ 99K X is a partial mapping, that a) the above condition (1) is satisfied; b) if δ∗(x, u) 6= ∅ and δ∗(δ∗(x, u), v) 6= ∅, then δ∗(x, uv) 6= ∅ and δ∗(x, uv) = δ∗(δ∗(x, u), v); (3) c) if δ∗(x, u) 6= ∅ and δ∗(x, uv) 6= ∅, then δ∗(δ∗(x, u), v) 6= ∅ and also the equation (3) is fulfilled. Clearly, preautomata correspond to preactions of the free monoid Σ∗. It will be convenient to omit the symbol δ∗, i. e. to write xu instead of δ∗(x, u), and denote the preautomaton by (Σ, X). Theorem 1 enables us to associate to the preautomaton M = (Σ, X) an automaton Mgl = (Σ, Y ), where Y = (X × Σ∗)/ ≃ is the set constructed in Section 1. We call Mgl a globalization of M. Using the fact that the monoid Σ∗ is free, the description of ≃ can be simplified: 54 Partial actions and automata Theorem 2. Define the following relation ≈ on the set X × Σ∗: (x, a) ≈ (y, b) ⇐⇒ ∃ a′, b′, p ∈ Σ∗ : a = a′p, b = b′p, xa′ = yb′ 6= ∅. Then ≈ coincides with ≃. Proof. If (x, a) ≈ (y, b) then (x, a′p) ⊢ (xa′, p) = (yb′, p) and (y, b′p) ⊢ (yb′, p), whence ≈⊆≃. Let us prove the converse inclusion. Obviously, ≈ is reflexive and symmetric. Let us check its transitivity. Let (x, a) ≈ (y, b) ≈ (z, d), a = a′p, b = b′p = c′q, d = d′q and xa′ = yb′ 6= ∅, yc′ = zd′ 6= ∅. Since Σ∗ is free, either p divides q or vice versa. Let, say, p = uq. Then c′ = b′u. Since yb′ 6= ∅, y(b′u) 6= ∅ then (xa′)u = (yb′)u = y(b′u) 6= ∅. Hence (xa′)u = x(a′u), as xa′ 6= ∅. On the other hand a′p = (a′u)q, and we obtain (x, a′p) = (x, (a′u)q) ≈ (z, d′q), since (xa′)u = (yb′)u = y(b′u) = yc′ = zd′. Thus, ≈ is an equivalence, and ≈⊃≃ as clearly ≈⊃⊢. Corollary 1. Every ≃-class can be uniquely written either in the form [x,w], where x ∈ X, w = a1 . . . an ∈ Σ∗ and x(a1 . . . ai) = ∅ for any i, 1 ≤ i ≤ n, or in the form [x, ε] (n = 0). Theorems 1 and 2 allow to give the following interpretation of a finite preautomaton. Suppose that a system with a possibly infinite set of states is given, and only a finite subset of its states is accessible to our observation. Then the preautomaton, obtained by restriction of the system to the set of observable states, can be considered as a model of the observable part of the system. 3. Example Consider the infinite automaton N = (Σ,Z) for which Σ = {a, b} is a two-lettered alphabet, Z is the set of integers and the transition function is defined by the equalities n · a = n+1, n · b = n− 1. Observe that this is the natural action of Σ∗ on its factor-monoid by the congruence generated by the relation ab = ε, and this factor-monoid is isomorphic to the infinite cyclic group. Set X = {0, 1} ⊂ Z. Then the restriction of the action of Σ∗ on X gives a preautomaton M = (Σ, X). In order to describe the corresponding preaction, we denote the length of a word w by |w|, the number of entries of the letter a in w by |w|a, and set ‖w‖ = |w|a − |w|b. Then 0 · w =    0 if ‖w‖ = 0, 1 if ‖w‖ = 1, ∅ otherwise, 1 · w =    0 if ‖w‖ = −1, 1 if ‖w‖ = 0, ∅ otherwise. M. Dokuchaev, B. Novikov, G. Zholtkevych 55 Corollary 1 allows to describe the globalization Mgl = (Σ, Y ). A word w = aα1bβ1 . . . aαnbβn will be called 1-simple if ‖aα1bβ1 . . . aαibβi‖ > 0 for all i ≥ 1 (this implies, in particular, that α1 > 0). The word ε will be considered 1-simple too. Similarly, a word w = bβ1aα1 . . . bβnaαn (and also the word ε) is 0-simple if ‖bβ1aα1 . . . bβiaαi‖ < 0 for all i ≥ 1. It is easy to see that every ≃-class has form [0, w] or [1, w], where the word w is 0-simple or 1-simple, respectively. Note that the preautomaton M = (Σ, X) cannot be considered as a restriction of a finite automaton. Indeed, let P = (Σ,W ) be such an automaton with a transition function δ : W × Σ → W and X ⊂ W . Since W is finite, δ(0, bk) = δ(0, bm) for some distinct k,m. Suppose k < m. As ‖bkak‖ = 0 then δ(δ(0, bk), ak) = δ(0, bkak) = 0 · (bkak) = 0. Further, δ(0, bmak) = δ(δ(0, bm), ak) = δ(δ(0, bk), ak) = 0, and since our preautomaton is a restriction of P, then 0 · (bmak) = 0. This contradicts the inequality ‖bmak‖ < 0. In addition, it is possible to obtain from Theorem 2 the description of the semigroup of Mgl = (Σ, Y ). We recall that the semigroup of an automaton is the factor-monoid obtained from Σ∗ by identification of the words equally acting of the states. Proposition 1. The semigroup of Mgl = (Σ, Y ) coincides with Σ∗ = {a, b}∗. Proof. Suppose that the words u, v ∈ Σ∗, u 6= v, act equally on all states of Mgl = (Σ, Y ). In particular, [0, bβ ]u = [0, bβ ]v for all β > 0. As bβu = bβv, then by Theorem 2 u = u′w, v = v′w and 0 · bβu′ = 0 · bβv′ 6= ∅. Hence, ‖u′‖ = ‖v′‖ = { β β + 1 ≥ β. But ‖u′‖ = |u′|a − |u′|b ≤ |u′| ≤ |u| is a bounded quantity in contrary with arbitrariness of β. Remark. It is easy to see that the semigroup of the original automaton N = (Σ, Z) is isomorphic to Z. 4. Recognizability In what follows we will consider only finite preautomata, i. e. preautomata M = (Σ, X) such that |X| < ∞. A preautomaton M = (Σ, X), in which an initial state x0 ∈ X and a terminal subset T ⊂ X are chosen, will be called a preacceptor and denoted by M = (Σ, X, x0, T ). 56 Partial actions and automata As well as in the classical situation, we call a language L ⊂ Σ∗ recognizable if there is a preacceptor M = (Σ, X, x0, T ) for which L = {w ∈ Σ∗ | ∅ 6= x0w ∈ T}. In what follows recognizability will be understood in this sense. The following assertion generalizes the Myhill–Nerode theorem and gives an algebraic characterization of recognizable languages: Theorem 3. A language L ⊂ Σ∗ is recognized by a preacceptor if and only if L is the union of a finite number of classes of some right congruence on Σ∗. Proof. Suppose that L is recognized by a (finite) preacceptor M = (Σ, X, x0, T ). Consider its globalization Mgl = (Σ, Y ). It follows from Corollary 1 that L is recognized by the acceptor N = (Σ, [x0, ε]Σ ∗, [x0, ε], [T, ε]) where [T, ε] = {[t, ε] | t ∈ T}, [x0, ε]Σ ∗ ⊆ Y. The relation ρ = {(u, v) ∈ Σ∗ × Σ∗ | [x0, ε]u = [x0, ε]v} is a right congruence, L is a union of ρ-classes and the number of these classes does not exceed |[T, ε]| = |T | < ∞. Conversely, let ρ be a right congruence on Σ∗ and L be the union of a finite number (say, n) of ρ-classes C1, . . . , Cn. In the (infinite) automaton K = (Σ,Σ∗/ρ) we choose the ρ-class E containing ε as the initial state, and T = {C1, . . . , Cn} ⊂ Σ∗/ρ as the terminal subset. Then (Σ, T ∪ {E}, E, T ) is a finite preacceptor which is a restriction of K and recognizes L. Corollary 2. Given languages L,M ⊂ Σ∗ which are recognized by preac- ceptors, then L ∩M is also recognizable. Proof. Let L and M are finite unions of classes of right congruences λ and µ respectively. Then L ∩M is a union of a finite number of classes of the congruence λ ∩ µ. Corollary 3. For the single-letter alphabet Σ = {a} a language L ⊂ Σ∗ is recognizable by a preacceptor exactly when L is recognizable by an acceptor. Proof. The semigroup Σ+ = {a}+ is the unique infinite monogenic semi- group up to isomorphism [6]. As it is commutative, all its right congruences are two-sided ones. Factor-semigroups by these congruences (except the M. Dokuchaev, B. Novikov, G. Zholtkevych 57 trivial one) are finite. Therefore if a language L is recognized by a preac- ceptor, but not by an acceptor, it should be a union of a finite number of classes of the trivial congruence, i. e. L is a finite subset in Σ+, hence L is regular. Example 1. The language L = {anbn | n > 0} over the alphabet Σ = {a, b} is a class of the right syntactic congruence [6]. Therefore it is recognized by a preacceptor. As it is well-known [6], L is not recognized by any finite acceptor. Example 2. For an arbitrary finite Σ each ideal of the monoid Σ∗ is recognizable, since it is an element of the Rees factor-semigroup. 5. Minimization of a preacceptor The study of recognizability of languages by preacceptors leads to the notion of the syntactic equivalence of preacceptors. Definition 5. Preacceptors M1 and M2 over the same alphabet Σ are called syntactically equivalent if they recognize the same language. As in the theory of acceptors, a question arises how to find in a class of syntactically equivalent preacceptors a preacceptor whose set of states has minimal cardinality. The first step is given by the following lemma. Lemma 1. Let M = (Σ, X, x0, T ) be a preacceptor, Y = {x0} ∪ T , M0 = (Σ, Y, x0, T ) the restriction of M on Y . Then M0 is syntactically equivalent to M. Proof. is obvious. Thanks to Lemma 1, in the study of language recognition by preaccep- tors we can restrict the set of states including only the initial and terminal states. We recall from [6] that the right syntactic congruence on Σ∗ of a language L ⊂ Σ∗ is the relation ≡L defined by w1 ≡L w2 ⇐⇒ ∀u ∈ Σ∗ (w1u ∈ L ⇔ w2u ∈ L) The language L is the union of some classes of this relation; moreover, any right congruence, such that L is a union of its classes, is contained in ≡L. This property allows us to reformulate Theorem 3 as follows: Lemma 2. A language L ⊂ Σ∗ is recognized by a (finite) preacceptor if and only if it is the union of a finite number of classes of its right syntactic congruence. 58 Partial actions and automata We will denote the ≡L-class containing u ∈ Σ∗ by [u]≡L . Fix a set of representatives u1, . . . , uk ∈ L of the ≡L-classes of L. For the set X≡L = {[ε]≡L , [u1]≡L , . . . , [uk]≡L } ⊂ Σ∗/ ≡L we define the partially defined map δ∗ : X≡L × Σ∗ 99K X≡L as follows: ∀ [u]≡L ∈ X≡L δ∗([u]≡L , ε) = [u]≡L , (4) ∀ [u]≡L ∈ X≡L ∀w ∈ Σ+ uw /∈ L ⇒ δ∗([u]≡L , w) = ∅, (5) ∀ [u]≡L ∈ X≡L ∀w ∈ Σ+ uw ∈ L ⇒ δ∗([u]≡L , w) = [uw]≡L . (6) Theorem 4. Let L ⊂ Σ∗ is a finite union of ≡L-classes. Then the preacceptor M≡L = (Σ, X≡L , [ε]≡L , {[u]≡L | u ∈ L}) with the partial mapping δ∗, defined by (4)–(6), recognizes L. Moreover, M≡L has the minimal set of states among all finite preac- ceptors recognizing L. Proof. First of all we check that δ∗ defines a preaction. Indeed, condition a) of Definition 4 is fulfilled by virtue of (4), and conditions b) and c) follow from (5) and (6). Thus, M≡L is a preacceptor which recognizes L. Now, let M = (Σ, X, x0, T ) be a preacceptor recognizing L. Define a mapping f : X≡L → X by the formulas f([ε]≡L ) = x0, f([ui]≡L ) = x0 · ui, i = 1, . . . , k. As ui ∈ L, x0 · ui ∈ T . We wish to show that f is injective. Assume that x0 · ui = x0 · uj for some 1 ≤ i 6= j ≤ k. Suppose that uiw ∈ L, so ∅ 6= x0 · (uiw) ∈ T . Since ui ∈ L, then ∅ 6= (x0 · ui) · w = x0 · (uiw) ∈ T by the definition of a preautomaton. The assumption x0 · ui = x0 · uj implies ∅ 6= x0 · (ujw) = (x0 ·uj) ·w ∈ T , i. e. ujw ∈ L. Thus uiw ∈ L yields ujw ∈ L, and similarly ujw ∈ L =⇒ uiw ∈ L. Hence ui ≡L uj , contradicting i 6= j. Thus f is injective and |X≡L | ≤ |X|. 6. Prefix decomposition In this section Theorem VI.4.1 from [1] is generalized to finite preautomata. Lemma 3. Suppose that L is a language over an alphabet Σ which is recognized by a preacceptor M = (Σ, X, x0, T ) and let Ly (y ∈ T ) be languages such that Ly is recognized by a preacceptor My = (Σ, X, x0, Ty = {y}). Then L = ∐ y∈T Ly (disjoint union). M. Dokuchaev, B. Novikov, G. Zholtkevych 59 Proof. w ∈ L if and only if x0w = y for some y ∈ T , i. e. w ∈ Ly. We remind that a (possibly, empty) language L ⊂ Σ∗ is called a prefix code [6] if u, uv ∈ L implies v = ε. The language {ε} is also considered as a prefix code (note that if a prefix code contains ε then it is equal {ε}). It is well-known and easily verified that the monoid S = L∗, generated by a prefix code L, is free. Such a monoid is called unitary and is characterized by the property: u, uv ∈ S =⇒ v ∈ S. (7) The following lemma generalizes (7): Lemma 4. Let H and C be prefix codes over an alphabet Σ. Then u, uv ∈ HC∗ =⇒ v ∈ C∗. Proof. Suppose that u = hc1 . . . cm, uv = h′c′1 . . . c ′ n (h, h′ ∈ H, ci, c ′ j ∈ C). Then hc1 . . . cmv = h′c′1 . . . c ′ n. Since H is a prefix code, then h = h′, and hence c1 . . . cmv = c′1 . . . c ′ n. This implies ci = c′i (1 ≤ i ≤ m ≤ n) and v = c′m+1 . . . c ′ n ∈ C∗. It is interesting to note that the property of prefix codes to generate free monoids can be generalized as follows: Proposition 2. 1 Let P1, . . . , Pn (n ≥ 1) be prefix codes over Σ and w ∈ P1 · . . . · Pn. Then the decomposition w = w1 . . . wn, where wi ∈ Pi for i = 1, . . . , n, is unique. Proof. We use induction on n. For n = 1 the statement is evident. Suppose that it is true for k < n. Let w = w′ 1 . . . w ′ n = w′′ 1 . . . w ′′ n be two representa- tions of a word w ∈ P1 · . . . ·Pn where w′ i, w ′′ i ∈ Pi for i = 1, . . . , n. Denote u′ = w′ 2 . . . w ′ n, u′′ = w′′ 2 . . . w ′′ n and assume that |u′| 6= |u′′|, for example, |u′| < |u′′|. Then |w′ 1| > |w′′ 1 |, so w′′ 1 is a prefix of w′ 1, which is impossible for a prefix code. Hence, |u′| = |u′′|. Then u′ = u′′ as suffixes of w of the same length, and w′ 1 = w′′ 1 as prefixes. By induction the equality u′ = u′′ implies w′ 2 = w′′ 2 , . . . , w ′ n = w′′ n. Definition 6. A language L ∈ Σ∗, such that L = HC∗ where H and C are prefix codes over Σ, will be called a p-language. 1Proposition 2 will not be used in this article. 60 Partial actions and automata Remark. Notice that if H = ∅ then HC∗ = ∅. Theorem 5. A language L over an alphabet Σ is recognized by some preacceptor M = (Σ, X, x0, {y}) with a single terminal state if and only if L is a p-language. Proof. Let L be a p-language. We consider the possible cases. 1) L1 = ∅. Then L1 is recognized by the preacceptor M = (Σ, {x0, y}, {x0}, {y}) where the action is trivial: x0w = x0, yw = y. 2) L2 = C∗. Then L2 is recognized by the preacceptor M = (Σ, {x0}, {x0}, {x0}) where preaction It is given by formulas: x0w = { x0 if w ∈ C∗, ∅ if w /∈ C∗. The fact that the above is a preaction, is verified using (7). 3) L3 = HC∗ with H 6= {ε}. Then L3 is recognized by the preacceptor M = (Σ, {x0, y}, {x0}, {y}) with the preaction: x0w =    y, if w ∈ HC∗, x0, if w = ε, ∅, if w 6∈ HC∗ ∪ {ε}, yw = { y if w ∈ C∗, ∅ if w 6∈ C∗. We need to show that the above formulas define a preautomaton. For the action of an input word on the state y the fulfilment of the conditions of Definition 4 follows from (7). As to the action on x0, the condition b) of Definition 4 is immediate, and in order to see c) suppose that x0u 6= ∅ and x0(uv) 6= ∅. If u = ε, the condition c) is obvious. Otherwise u, uv ∈ HC∗ and by Lemma 4, v ∈ C∗. Now we prove that if a language L is recognized by a preacceptor of the form M = (Σ, X, x0, {y}) then it is a p-language. For every word w ∈ Σ∗ denote by Px(w) the set of all proper prefixes of w. For K ⊂ Σ∗ write Px(K) = {Px(w) | w ∈ K}. Set H = {w ∈ Σ∗ | x0w = y & ∀u ∈ Px(w) x0u 6= y}. M. Dokuchaev, B. Novikov, G. Zholtkevych 61 and similarly, C = {w ∈ Σ∗ | yw = y & ∀u ∈ Px(w) yu 6= y}. By construction H and C are prefix codes and, moreover, L ⊃ HC∗ since x0w = y for any word w ∈ HC∗. It remains to show that L ⊂ HC∗. Let w ∈ L, i. e. x0w = y. First assume that x0 = y. Write w = uv, where u is the least prefix of w for which x0u = x0 (and therefore, u ∈ H = C). By definition of a preaction ∅ 6= (x0u)v = x0w, whence x0v = x0, and consequently v ∈ C∗. Thus we obtain w = uv ∈ HC∗ = C∗. If x0 6= y, then write w = uv, where u is the prefix of the least length, such that x0u = y. It follows that u ∈ H and yv = y. As above, v ∈ C∗, whence w ∈ HC∗. Theorem 5 allows us to transfer to preautomata the concept of a prefix decomposition ([1], Theorem VI.4.1): Corollary 4. If a language L over the alphabet Σ is recognized by some preacceptor then L decomposes into a disjoint union L = ∐ i HiC ∗ i , where Hi, Ci are prefix codes over Σ. Proof. The corollary follows from Theorem 5 and Lemma 3. The converse statement does not hold: Example 3. It is directly seen that the language H1 = {a}+ is recognized (even by a finite acceptor), and the language H2 = {anbn | n ≥ 1} is a prefix code. Take C1 = C2 = {ε}. The union H = H1∪H2 = H1C ∗ 1∪H2C ∗ 2 is not recognized by any preacceptor. Indeed, let ≡H be the right syntactic congruence of H, and take x ∈ Σ∗. Since anx ∈ H ⇐⇒ x = bn or x ∈ H1 ∪ {ε}, the words an form one-element ≡H -classes, and by Theorem 3 the language H is non-recognizable. The given example shows also that the set of languages, recognized by finite preacceptors, is not union-closed. Moreover, it is not closed with respect to the other Kleene’s operations, i.e. the product and the iteration: 62 Partial actions and automata Example 4. Let H1 and H2 be the same as in Example 3. By Theorem 5 the language H∗ 2 is recognizable. Consider the product K = H∗ 2H1. Then an ∈ K for n > 0 and {x ∈ Σ∗ | anx ∈ K} ∩ {b}+ = {bn}. Therefore all words of the form an are pairwise non-equivalent with respect to the right congruence ≡K . As in Example 3, it follows that K is non- recognized. Example 5. The language L = H2 ∪ {a} is recognized by a preacceptor, since it consists of two ≡L-classes H2 and {a}. Consider the iteration M = L∗. It is easy to see that ambn ∈ M iff m ≥ n. Hence, {x ∈ Σ∗ | amx ∈ M} ∩ {b}+ = {bk | k ≤ m}. As above, the words am are pairwise non-equivalent with respect to ≡M . Therefore, M = L∗ is not recognized by any preacceptor. Conclusion The obtained results allow to clarify relations between the class of finite preautomata FPA and other types of machines. Obviously, FPA includes the class of finite automata FA. This inclusion is strict since the language from Example 1 is not recognized by any finite automaton. On the other hand, it is known [6, § 9.1] that the prefix language {anbncn | n ≥ 1} is not context-free. According to Theorem 5 it means that FPA is not contained in the class of automata with stack memory FSA. At the same time, FPA does not include the class of Turing machines TM. Indeed, the language {an 2 | n ≥ 1} is not recognized by a finite automaton [5, ex. 4.1.2], and by Corollary 3 it is not recognized also by a preautomaton. On the other hand, the cardinal of the class of all prefix codes over some finite alphabet equals to the cardinal of continuum, whereas the class of a recursively enumerable languages over a finite alphabet is countable. It follows that TM does not contain FPA. References [1] S. Eilenberg, Automata, Languages, and Machines, vol. B., Academic Press, 1976. [2] R. Exel, Partial actions of groups and actions of semigroups, Proc. Amer. Math. Soc., 126(1998), pp.3481-3494. [3] W.M. L. Holcombe, Algebraic Automata Theory, Cambridge Univ. Press, 1982. M. Dokuchaev, B. Novikov, G. Zholtkevych 63 [4] C. Hollings, Partial actions of monoids, Semigroup Forum, 75(2007), pp.293-316. [5] J. E. Hopcroft, R. Motwani, J. D. Ullman, Introduction to Automata Theory, Lan- guages and Computation, 2nd edition, Addison Wesley, 2001. [6] G. Lallement, Semigroups and Combinatorial Applications, Pure and Applied Mathematics Series, Wiley, New York, 1979. [7] M. Megrelishvili, L. Schröder, Globalization of confluent partial actions on topolog- ical and metric spaces, Topology and its Appl., 145(2004), pp.119-145. Contact information M. Dokuchaev Instituto de Matemática e Estat́ıstica Universi- dade de São Paulo, Rua do Matão, 1010, CEP 05508-090, São Paulo, SP, Brazil E-Mail: dokucha@ime.usp.br B. Novikov Kharkov National University, Svobody sq., 4, 61077, Kharkov, Ukraine E-Mail: novikov@univer.kharkov.ua G. Zholtkevych Kharkov National University, Svobody sq., 4, 61077, Kharkov, Ukraine E-Mail: g.zholtkevych@gmail.com Received by the editors: 13.04.2011 and in final form 05.05.2011.