Émilie Charlier 1 Department of Mathematics, University of Liège, Liège, Belgium Narad Rampersad.

Similar documents
RECOGNIZABLE SETS OF INTEGERS

(ABSTRACT) NUMERATION SYSTEMS

Abstract Numeration Systems or Decimation of Languages

On the cost and complexity of the successor function

Games derived from a generalized Thue-Morse word

Abstract β-expansions and ultimately periodic representations

Automata for arithmetic Meyer sets

On multiplicatively dependent linear numeration systems, and periodic points

Automata and Numeration Systems

An Efficient Algorithm to Decide Periodicity of b-recognisable Sets Using MSDF Convention

State complexity of the multiples of the Thue-Morse set

Numeration systems: a link between number theory and formal language theory

On the concrete complexity of the successor function

SYNTACTICAL AND AUTOMATIC PROPERTIES OF SETS OF POLYNOMIALS OVER FINITE FIELDS

MULTIDIMENSIONAL GENERALIZED AUTOMATIC SEQUENCES AND SHAPE-SYMMETRIC MORPHIC WORDS

On the Recognizability of Self-Generating Sets

Remarks on Separating Words

On negative bases.

A Generalization of Cobham s Theorem to Automata over Real Numbers

On the characterization of canonical number systems. Author(s) Scheicher, Klaus; Thuswaldner, Jorg M. Osaka Journal of Mathematics. 41(2) P.327-P.

Büchi Automata and their closure properties. - Ajith S and Ankit Kumar

Combinatorial Interpretations of a Generalization of the Genocchi Numbers

An extension of the Pascal triangle and the Sierpiński gasket to finite words. Joint work with Julien Leroy and Michel Rigo (ULiège)

The Formal Inverse of the Period-Doubling Sequence

From combinatorial games to shape-symmetric morphisms (3)

On the negative base greedy and lazy representations

2. Syntactic Congruences and Monoids

A Generalization of Cobham s Theorem to Automata over Real Numbers 1

arxiv: v1 [cs.fl] 19 Mar 2015

Finite Automata, Palindromes, Patterns, and Borders

The Separating Words Problem

Automata Theory. Lecture on Discussion Course of CS120. Runzhe SJTU ACM CLASS

The Separating Words Problem

Concerning Kurosaki s Squarefree Word

Mix-Automatic Sequences

Open Problems in Automata Theory: An Idiosyncratic View

Note On Parikh slender context-free languages

arxiv: v3 [cs.fl] 2 Jul 2018

Finite Automata, Palindromes, Powers, and Patterns

ON THE DECIMAL EXPANSION OF ALGEBRAIC NUMBERS

Automatic Sequences and Transcendence of Real Numbers

COBHAM S THEOREM AND SUBSTITUTION SUBSHIFTS

Further discussion of Turing machines

Hierarchy among Automata on Linear Orderings

arxiv:math/ v3 [math.nt] 23 May 2008

Aperiodic languages and generalizations

Decision issues on functions realized by finite automata. May 7, 1999

Unary Automatic Graphs: An Algorithmic Perspective 1

#A31 INTEGERS 11 (2011) COMPLETELY MULTIPLICATIVE AUTOMATIC FUNCTIONS

Mergible States in Large NFA

languages by semifilter-congruences

Closure Under Reversal of Languages over Infinite Alphabets

DM17. Beregnelighed. Jacob Aae Mikkelsen

3515ICT: Theory of Computation. Regular languages

Definition of Büchi Automata

FORMAL LANGUAGES, AUTOMATA AND COMPUTATION

ALMOST EVERY NUMBER HAS A CONTINUUM. INTRODUCTION. Let β belong to (1, 2) and write Σ =

Arturo Carpi 1 and Cristiano Maggi 2

NOTES ON AUTOMATA. Date: April 29,

Uncountable Automatic Classes and Learning

Automata and Number Theory

Regular Sequences. Eric Rowland. School of Computer Science University of Waterloo, Ontario, Canada. September 5 & 8, 2011

Minimal Paths and Cycles in Set Systems

Paul Glendinning and Nikita Sidorov

Computational Models - Lecture 4

BALA RAVIKUMAR Department of Computer Science, Sonoma State University Rohnert Park, CA 94928, USA

The Logical Approach to Automatic Sequences

1 More finite deterministic automata

On Minimal Words With Given Subword Complexity

CS243, Logic and Computation Nondeterministic finite automata

IN CUBIC SALEM BASE IN F q ((X 1 )) Faïza Mahjoub

FORMAL LANGUAGES, AUTOMATA AND COMPUTABILITY

Finite repetition threshold for large alphabets

CS 154 Introduction to Automata and Complexity Theory

Avoiding Approximate Squares

ON PATTERNS OCCURRING IN BINARY ALGEBRAIC NUMBERS

Theory of Computation

Lecture 3: Nondeterministic Finite Automata

CSC173 Workshop: 13 Sept. Notes

The Target Discounted-Sum Problem

Finiteness properties for Pisot S-adic tilings

HKN CS/ECE 374 Midterm 1 Review. Nathan Bleier and Mahir Morshed

arxiv: v2 [math.co] 24 Oct 2012

What we have done so far

Vertex colorings of graphs without short odd cycles

Notes on generating functions in automata theory

Great Theoretical Ideas in Computer Science. Lecture 4: Deterministic Finite Automaton (DFA), Part 2

On the sequentiality of the successor function. Christiane Frougny. Universite Paris 8. and. 4 place Jussieu, Paris Cedex 05, France

Decidable and Expressive Classes of Probabilistic Automata

Definition 6.1 (p.277) A positive integer n is prime when n > 1 and the only positive divisors are 1 and n. Alternatively

V Honors Theory of Computation

arxiv: v2 [cs.fl] 29 Nov 2013

Extensions naturelles des. et pavages

About Duval Extensions

Approximation by polynomials with bounded coefficients

The Distribution of Generalized Sum-of-Digits Functions in Residue Classes

On the Sequentiality of the Successor Function

CERNY CONJECTURE FOR DFA ACCEPTING STAR-FREE LANGUAGES

Expansions in non-integer bases

CS 530: Theory of Computation Based on Sipser (second edition): Notes on regular languages(version 1.1)

Transcription:

#A4 INTEGERS B () THE MINIMAL AUTOMATON RECOGNIZING m N IN A LINEAR NUMERATION SYSTEM Émilie Charlier Department of Mathematics, University of Liège, Liège, Belgium echarlier@uwaterloo.ca Narad Rampersad Department of Mathematics, University of Liège, Liège, Belgium nrampersad@ulg.ac.be Michel Rigo Department of Mathematics, University of Liège, Liège, Belgium M.Rigo@ulg.ac.be Laurent Waxweiler Department of Mathematics, University of Liège, Liège, Belgium Received: 7/9/, Revised: //, Accepted: 7/5/, Published: // Abstract We study the structure of automata accepting the greedy representations of N in a wide class of numeration systems. We describe the conditions under which such automata can have more than one strongly connected component and the form of any such additional components. Our characterization applies, in particular, to any automaton arising from a Bertrand numeration system. Furthermore, we show that for any automaton A arising from a system with a dominant root β >, there is a morphism mapping A onto the automaton arising from the Bertrand system associated with the number β. Under some mild assumptions, we also study the state complexity of the trim minimal automaton accepting the greedy representations of the multiples of m for a wide class of linear numeration systems. As an example, the number of states of the trim minimal automaton accepting the greedy representations of m N in the Fibonacci system is exactly m. This author is currently a post-doctoral fellow at the David R. Cheriton School of Computer Science of the Faculty of Mathematics of the University of Waterloo.

INTEGERS: B (). Introduction Cobham [] showed that ultimately periodic sets of non-negative integers are the only sets that are recognized by a finite automaton in every integer base numeration system. The ultimately periodic sets are also exactly the sets definable by first order formulas in the Presburger arithmetic N, +. In the context of a non-standard numeration system U, if N is U-recognizable, then U is easily seen to be a linear numeration system, that is, U satisfies a linear recurrence with integer coefficients [4]. For linear numeration systems, ultimately periodic sets are all recognized by finite automata if and only if N is (see Theorem below). Conditions on a linear numeration system U for N to be U-recognizable are considered in [6, ]. From the point of view of the Chomsky hierarchy, a U-recognizable set X of integers can be considered as having a low computational complexity: the greedy representations of the elements in X in the numeration system U have simple syntactical properties recognized by some finite automaton, i.e., rep U (X) is a regular language. Since the seminal work of Alan Cobham [] many properties of U-recognizable sets have been investigated, e.g., algebraic, logical or automatic characterizations of U-recognizable sets for integer base numeration systems [7], extensions of these characterizations to systems based on a Pisot number [6], study of the normalization map [4], introduction of abstract numeration systems [9],... Among linear numeration systems for which N is U-recognizable, the class of systems whose characteristic polynomial is the minimal polynomial of a Pisot number has been widely studied [6]. An example of such a system is given by the Fibonacci numeration system (see Example ). In particular, the automata accepting these numeration languages are well-known. Another well-known class of numeration languages, which has given rise to many successful applications concerning β-numerations, consists of the languages arising from Bertrand systems associated with a Parry number (see Section ) [5, 5]. Currently little is known about the automata accepting other kind of numeration languages. In the first part of this paper we study the structure of these automata for a wide class of numeration systems. In Section we review the needed background concerning numeration systems. Then in Section 3 we provide several examples in order to illustrate the different types of automata that can arise from these numeration systems. In Section 4 we describe the conditions under which such automata can have more than one strongly connected component and the form of any such additional strongly connected component. In the case where the numeration system has a dominant root β > (see the next section for the definition), we are able to provide a more specific description of the structure. For instance, we show that for any automaton A arising from a numeration system with a dominant root β >, there is a morphism mapping A onto the automaton arising from the Bertrand system associated with the number β. Our primary motivation is to understand the state complexity of languages of the

INTEGERS: B () 3 form rep U (mn), that is, the language of the representations of the multiples of m in a given numeration system U (see [, 8]), in connection with the following decidability problem. Let U be a linear numeration system and X be a U-recognizable set of non-negative integers given by some deterministic finite automaton recognizing the greedy representations of elements of X. For integer base systems, Honkala proved that one can decide whether or not X is ultimately periodic [7]. Another, shorter proof of this result can be found in []. For a wide class of linear numeration systems containing the Fibonacci numeration system, the same decidability question is answered positively in [, 3]. For all the above mentioned reasons ultimately periodic sets of integers and, in particular, the recognizability of a given divisibility criterion by finite automata deserve special interest. Lecomte and Rigo [9] showed the following: given a regular language L = {w < w < } genealogically ordered, extracting from L words whose indices belong to an ultimately periodic set I N is a regularity-preserving operation defining a language L I. Krieger et al. [8] considered the state complexity of this operation. If the minimal automaton of L has n states, it is natural to give bounds or try to estimate the number of states of the minimal automaton of L I as a function of n, the preperiod and period of I. Such results could be useful in solving the decidability question mentioned in the last paragraph. For example, Alexeev [] recently gave the following formula for the number of states of the minimal automaton of the language rep b (mn), that is, the set of b-ary representations of the multiples of m. The GCD of two integers a and b is denoted by (a, b). Let N, M be such that b N < m b N+ and (m, ) < (m, b) < < (m, b M ) = (m, b M+ ) = (m, b M+ ) =. The minimal automaton of rep b (mn) has exactly inf{n,m } m (m, b N+ ) + b t (m, b t ) t= states. In the second part of this paper, we study the state complexity for the divisibility criterion by m in the framework of linear numeration systems. Under some mild assumptions, Theorem 34 gives the number of states of the trim minimal automaton of rep U (mn) from which infinitely many words are accepted. As a corollary, we show that, for a certain class of numeration systems, we can give the precise number of states of this automaton. For instance, for the Fibonacci numeration system, the corresponding number of states is m, see Corollary 39. Finally we are able to give a lower bound for the state complexity of rep U (m N) for any numeration system. Note that the study of state complexity could possibly be related to the length of the formulas describing such sets in a given numeration system. It is noteworthy that for linear numeration systems whose characteristic polynomial is the minimal polynomial of a Pisot number, U-recognizable sets can be characterized by first order formulas of a convenient extension of N, +, see [6]. ()

INTEGERS: B () 4 This paper is a combined and expanded version of [8, 9].. Background on Numeration Systems In this paper, when we write x = x n x where x is a word, we mean that x i is a letter for all i {,..., n }. An increasing sequence U = (U n ) n of integers is a numeration system, or a numeration basis, if U = and C U := sup n Un+ U n < +. We let A U be the alphabet {,..., C U }. A greedy representation of a non-negative integer n is a word w = w l w over A U satisfying l w i U i = n and j {,..., l}, i= j w i U i < U j. i= We denote the greedy representation of n > satisfying w l by rep U (n). By convention, rep U () is the empty word ε. The language rep U (N) is called the numeration language. A set X of integers is U-recognizable if rep U (X) is regular, i.e., accepted by a finite automaton. If N is U-recognizable, then we let A U = (Q U, q U,, F U, A U, δ U ) denote the trim minimal automaton of the language rep U (N) having #A U states. The numerical value map val U : A U N maps any word d l d over A U to l i= d iu i. For example, if (U, U, U ) = (,, 3) and A U = {, }, then val U () = 3 and val U (3) = {, }. Definition. A numeration system U = (U n ) n is said to be linear, if there exist k and a,..., a k Z such that n N, U n+k = a k U n+k + + a U n. () We say that k is the length of the recurrence relation. Theorem. [4, Proposition 3..9] Let p, r. If U = (U n ) n is a linear numeration system, then val U (p N + r) = {w A U val U (w) p N + r} is accepted by a deterministic finite automaton that can be effectively constructed. In particular, if N is U-recognizable, then any eventually periodic set is U-recognizable. Let u, v be two finite words of the same length (resp. two infinite words) over an alphabet A N. We say that u is lexicographically less than v and we write u < v, if there exist p A, a, b A with a < b and words u, v over A such that u = pau, v = pbv. If u and v are two finite words (not necessarily of the same length), then we say that u is genealogically less than v if either u < v, or

INTEGERS: B () 5 u = v and u < v (with respect to the lexicographic order). We also write u < v to denote the genealogical order. Note that if U is a numeration system, then for all m, n N, we have m < n if and only if rep U (m) is genealogically less than rep U (n). Observe that if uv is a greedy representation, then so is v. However, if u is a greedy representation, there is no reason for u to still be greedy. As an example, if U =, U = 3 and U = 5, then is a greedy representation but is not. Definition 3. A numeration system U = (U n ) n is a Bertrand numeration system if, for all w A + U, w rep U(N) w rep U (N). Let us recall the theorems of Bertrand [5] (also see [, Thm. 7.3.8]) and Parry [3] (also see [, Thm. 7..9]). Let β > be a real number. The β-expansion of a real number x [, ] is the sequence d β (x) = (x i ) i N ω satisfying x = + i= x i β i and which is the maximal element in N ω having this property with respect to the lexicographic order over N. Note that the β-expansion is also obtained by using the greedy algorithm and that it only contains letters in the canonical alphabet A β = {,..., β }. Also observe that, for all x, y [, ], we have x < y d β (x) < d β (y). The set Fact(D β ) is the set of factors occurring in the β-expansions of the real numbers in [, ). If d β () = t t m ω, with t,..., t m A β and t m, then we say that d β () is finite and we set d β () = (t t m (t m )) ω. Otherwise, we set d β () = d β(). If d β () is ultimately periodic, then β is said to be a Parry number. The following lemma is not difficult to prove. It will be used in the proof of Theorem. Lemma 4. Let x = x k x be a word over N. We have l {,..., k}, x l x ω { < d β() l {,..., k}, l { < x i β i l. Theorem 5 (Bertrand [5]). Let U = (U n ) n be a numeration system. There exists a real number β > such that rep U (N) = Fact(D β ) if and only if U is a Bertrand numeration system. In that case, if d β () = (t i) i, then i= U n = t U n + + t n U +. (3) Note that if β is a Parry number, then (3) defines a linear recurrence sequence and β is a root of its characteristic polynomial. Theorem 6 (Parry [3]). A sequence s = (s i ) i over N is the β-expansion of a real number in [, ) if and only if (s n+i ) i is lexicographically less than d β () for all n N.

INTEGERS: B () 6 As a consequence of the previous two theorems, with any Parry number β is canonically associated a deterministic finite automaton A β = (Q β, q β,, F β, A β, δ β ) accepting the language Fact(D β ). Let d β () = t t i (t i+ t i+p ) ω where i and p are the minimal preperiod and period respectively. The set of states of A β is Q β = {q β,,..., q β,i+p }. All states are final. For every j {,..., i + p}, we have t j edges q β,j q β, labeled by,..., t j and, for j < i + p, one edge q β,j q β,j labeled by t j. There is also an edge q β,i+p q β,i labeled by t i+p. See, for instance, [3, 5, ]. Note that in [, Thm. 7..3], A β is shown to be the trim minimal automaton of Fact(D β ). A deterministic finite automaton is trim if it is accessible and coaccessible, i.e., any state can be reached from the initial state and from any state, a final state can be reached. Example 7. Let β be the dominant root of the polynomial X 3 X. We have d β () = ω and d β () = ()ω. The automaton A β is depicted in Figure., 3 Figure : The automaton A β for d β () = ()ω. Definition 8. Let U be a linear numeration system. If lim n + U n+ /U n = β for some real β >, then U is said to satisfy the dominant root condition and β is called the dominant root of the recurrence. Remark 9. If U is a linear numeration system satisfying the dominant root condition and if rep U (N) is regular, then the dominant root β is a Parry number [6]. In the case where U has a dominant root β >, some connections between A U and A β have been previously explored by several authors [5,, ]. Our aim in this paper is to provide a more comprehensive analysis of the relationship between these two automata. Recall [] that the states of the minimal automaton of an arbitrary language L over an alphabet A are given by the equivalence classes of the Myhill-Nerode congruence L, which is defined by w, z A, w L z if and only if {x A wx L} = {x A zx L}. Equivalently, the states of the minimal automaton of L correspond to the sets w L = {x A wx L}. In this paper the symbol will be used to denote Myhill-Nerode congruences. Remark. In Theorem we will describe a map between a restriction of A U and A β. Note that similar observations have been considered in other contexts

INTEGERS: B () 7 [3, 6]. For example, if U is the Bertrand numeration system associated with a Pisot number β, then for any U-recognizable set X of integers, there exist an automaton recognizing X and a morphism mapping this automaton onto A U = A β [6]. 3. Examples of Automata A U The first two examples present the well-known Fibonacci numeration system and its generalization to an l-order recurrence relation. Note that in the first four examples, Examples to 4, the automaton A U is exactly an automaton of the kind A β. Example (Fibonacci numeration system). With U n+ = U n+ +U n and U =, U =, we get the usual Fibonacci numeration system associated with the Golden Ratio. The dominant root is β = ( + 5)/. For this system, A U = {, } and A U accepts all words over A U except those containing the factor. Moreover, we have d β () = ω and d β () = ()ω. Figure : The automaton A U for the Fibonacci numeration system. Example (l-bonacci numeration system). Let l. Consider the linear recurrence sequence defined by l n N, U n+l = i= U n+i and for i {,..., l }, U i = i. For this system, A U = {, } and A U accepts all words over A U except those containing the factor l. We have d β () = l ω and d β () = (l ) ω. Figure 3: The automaton A U for the 4-bonacci numeration system. The third example is also classical. Compared to the previous examples where the β-expansions of the real numbers in [, ) avoid a single factor, here the β- expansions avoid factors in an infinite regular language.

INTEGERS: B () 8 Example 3 (Square of the Golden Ratio). With U n+ = 3U n+ U n, U = and U = 3, we get the Bertrand numeration system associated with β = (3 + 5)/ (the square of the Golden Ratio). We have A U = {,, } and is the set of minimal forbidden factors. Moreover d β () = d β () = ω., Figure 4: The automaton A U for the Bertrand system associated with (3 + 5)/. The recurrence involved in the following example will show some interesting properties and is related to Example 3. Example 4. With U n+ = U n+ + U n, U =, U = 3, we have the Bertrand numeration system (U n ) n =, 3, 7, 7, 4, 99, 39,... associated with β = +. We have d β () = ω and d β () = ()ω. The corresponding automaton A U is depicted in Figure 5., q q Figure 5: The automaton A U for the Bertrand system associated with +. The next example reveals some interesting properties and should be compared with the usual Fibonacci system. Observe that we have the same strongly connected component as for the Fibonacci system but the automaton in Figure 6 has one more state, from which only finitely many words may be accepted. Example 5 (Modified Fibonacci system). Consider the sequence U = (U n ) n defined by the recurrence U n+ = U n+ +U n of Example but with the initial conditions U =, U = 3. We get a numeration system (U n ) n =, 3, 4, 7,, 8, 9, 47,... which is no longer Bertrand. Indeed, is a greedy representation but is not because rep U (val U ()) =. For this system, A U = {,, } and A U is depicted in Figure 6. The following example illustrates the case where β is an integer. Example 6. Consider the numeration system U = (U n ) n defined by U n+ = 3U n + and U =. We have A U = {,,, 3, 4}. This system is linear and has the dominant root β = 3. We have d β () = 3 ω and d β () = ω. The automaton A U is depicted in Figure 7.

INTEGERS: B () 9 3 Figure 6: The automaton A U for the modified Fibonacci system.,, 3 3 4 Figure 7: The automaton A U for U n+ = 3U n + and U =. As a prelude to Theorem 9, the next example shows that when the initial conditions are changed, the automaton A U may have the same transition graph as the canonical automaton A β, but the set of final states may change. Example 7. Consider the recurrence relation U n+3 = U n+ + U n. If we choose (U, U, U ) = (, 3, 7), we get the Bertrand numeration system U such that A U is exactly the automaton A β from Example depicted in Figure. If (U, U, U ) = (,, 4), we get the same graph but only state is final. If (U, U, U ) = (,, 5), we get the same graph but only states and 3 are final. Finally, with (U, U, U ) = (, 3, 6), states and are final. 4. Structure of the Automaton A U In this section we give a precise description of the automaton A U when U is a linear numeration system satisfying the dominant root condition and such that rep U (N) is regular. Definition 8. A directed graph is strongly connected if for all pairs of vertices (s, t), there is a directed path from s to t. A strongly connected component of a directed graph is a maximal strongly connected subgraph. Such a component is said to be non-trivial if it does not consist of a single vertex with no loop. For instance, state 3 in Figure 6 is not a non-trivial strongly connected component and state in Figure 7 is a non-trivial strongly connected component. Theorem 9. Let U be a linear numeration system such that rep U (N) is regular. (i) The automaton A U has a non-trivial strongly connected component C U containing the initial state.

INTEGERS: B () (ii) If p is a state in C U, then there exists N N such that δ U (p, n ) = q U, for all n N. In particular, if q (resp. r) is a state in C U (resp. not in C U ) and if δ U (q, σ) = r, then σ. (iii) If C U is the only non-trivial strongly connected component of A U, then we have lim U n+ U n = +. n + (iv) If lim n + U n+ U n = +, then the state δ U (q U,, ) belongs to C U. Proof. (i) The initial state q U, has a loop with label and therefore A U has a non-trivial strongly connected component C U containing q U,. (ii) Let p be a state in C U. There exist u, v A U such that δ U (q U,, u) = p and δ U (p, v) = q U,. We have x A U, uvx rep U (N) u v x rep U (N). Indeed, if uvx is a greedy representation, so is u v x. Furthermore, if u v x is a greedy representation, so is x, which must be accepted from q U, = δ U (q U,, uv). Hence, uvx is a greedy representation. In other words, uv rep U (N) u v and δ U (p, v ) = q U,. Since q U, has a loop labeled by, we obtain the desired result. (iii) Assume that A U has only one non-trivial strongly connected component C U. Since n is a greedy representation for all n, infinitely many words are accepted from δ U (q U,, ), and so δ U (q U,, ) belongs to C U. From (ii), there exists a minimal t N such that δ U (q U,, t ) = q U,. Observe that U n is the number of words of length n in rep U (N). For each word x (resp. y) in rep U (N) of length n (resp. n t), the word x (resp. t y) has length n + and belongs to rep U (N). Therefore, we obtain U n+ U n + U n t for all n t. (iv) Assume that lim U n+ U n = +. It is enough to show that there exists n + l such that δ U (q U,, l ) = q U,. That is, we have to show that there exists l N, x A U, l x rep U (N) if and only if x rep U (N). Since we can always distinguish two states by a word of length at most g = (#A U ), it is equivalent to show that there exists l N, x A g U, l x rep U (N) if and only if x rep U (N), where A g U denotes the set of the words of length at most g over A U. Since U n+ U n tends to +, there exists l such that for all n l, we have U n+ U n > U g, which shows that l x is a greedy representation for any greedy representation x of length less than or equal to g. The other direction is immediate. Theorem. Let U be a linear numeration system, having a dominant root β >, such that rep U (N) is regular. Let x be a word over A U such that infinitely many

INTEGERS: B () words are accepted from δ U (q U,, x). Then y ω d β () for all suffixes y of x. Furthermore, the state δ U (q U,, x) belongs to C U if and only if y ω < d β () for all suffixes y of x. In particular, in this case, the word x only contains letters in {,..., β }. Remark. Let q be a state of A U distinct from q U,. Since A U is minimal, there exists a word w q that distinguishes q U, and q: that is, either w q is accepted from q U, and not from q, or w q is accepted from q and not from q U,. Let us show that in the setting of numeration languages the second situation never occurs. Let x be such that δ U (q U,, x) = q. Assume that xw q is accepted by A U. Then w q is a greedy representation which must be accepted from q U,. Proof of Theorem. To prove the result we use Lemma 4. Let x = x k x be a word over A U such that infinitely many words are accepted from δ U (q U,, x). Due to the greediness of the representations, there exist infinitely many n such that x n is a greedy representation. We obtain l {,..., k}, l x i U i+n < U l+n for infinitely many n. Dividing by U l+n and letting n tend to infinity, we get l {,..., k}, i= l x i β i l. Now assume that δ U (q U,, x) belongs to C U. From (ii) and (iv) of Theorem 9, there exist m, N N such that for all n N, we have δ U (q U,, x m n ) = q U,, which is a final state. By the same reasoning as before, we obtain that This implies that l {,..., k}, i= l x i β i l + β l m. i= l {,..., k}, l x i β i l <. To show the other direction, now assume that δ U (q U,, x) does not belong to C U. For all n N, we have δ U (q U,, x n ) q U,. Therefore, by Remark, for all n N, there exists a greedy representation w (n) of length at most (#A U ) such that x n w (n) is not a greedy representation. Hence, by the pigeonhole principle, there exists a greedy representation w of length at most (#A U ) such that for infinitely many n, the word x n w is not a greedy representation. Therefore l {,..., k}, i= l x i U i+n+ w + val U (w) U l+n+ w i=

INTEGERS: B () for infinitely many n. We conclude that l {,..., k}, l x i β i l. i= Using Lemma 4, we obtain the desired result. Theorem. Let U be a linear numeration system, having a dominant root β >, such that rep U (N) is regular. There exists a map Φ: C U Q β such that Φ(q U, ) = q β,, and for all states q and all letters σ such that q and δ U (q, σ) are states in C U, we have Φ(δ U (q, σ)) = δ β (Φ(q), σ). Furthermore, if q is a state in C U and σ is the maximal letter that can be read from Φ(q) in A β, then for any letter α in A U, the state δ U (q, α) is in C U if and only if α σ. Proof. Consider the automaton whose transition diagram is the subgraph induced by C U and where all states are assumed to be final. From Theorems 5, 6 and, the language accepted by this automaton is exactly the same as the one accepted by A β. Note that A β is a trim minimal automaton [, Theorem 7..3]. From a classical result in automata theory (see, for instance, [, Chap. 3, Thm. 5.]), such a map Φ exists. Example 3. Consider the same recurrence relation as in Example 7 but with (U, U, U ) = (, 5, 6). In Example 7 (see also Example 7), the automaton A β with d β () = ω and A U had the same transition graph. Here we get a more complex situation described in Figure 8. The non-trivial strongly connected component C U consists of the states Q U \ {g}. The map Φ is the map that sends the states a, b, c onto ; the states d, e onto ; and the states f onto 3; where {,, 3} is the set of states of the automaton A β given in Figure. 3, 4 a d f g b c e Figure 8: The automaton A U for (U, U, U ) = (, 5, 6).

INTEGERS: B () 3 Theorem 4. Let U be a linear numeration system, having a dominant root β >, such that rep U (N) is regular. If there exists a non-trivial strongly connected component distinct from C U, then d β () is finite. In this case, if s denotes the longest prefix of d β () which does not end with, then δ U (q U,, u) C U for all proper prefixes u of s and δ U (q U,, s) / C U. In addition, if x is a word over A U such that δ U (q U,, x) is a state not in C U leading to such a component, then there exists a word y over {,..., β } such that δ U (q U,, y) Φ (q β, ) and x = ys n for some n. In particular, the number of non-trivial strongly connected components distinct from C U is bounded by #Φ (q β, s ). Proof. Assume that there exists a non-trivial strongly connected component distinct from C U. Consider a state q not in C U leading to such a component and a word u over A U such that δ U (q U,, u) = q. Take the longest prefix x of u such that δ U (q U,, x) C U. Hence from Theorem x A β and if σ A U and v A U are such that u = xσv, then δ U (q U,, xσ) / C U. Using Theorem, there exists a suffix z of x such that d β () = zσ ω, and so d β () is finite. The longest prefix of d β () which does not end with is s = zσ. Furthermore, by Theorem again, we see that v belongs to. We still have to show that if x = yz, then δ U (q U,, y) belongs to Φ (q β, ), or equivalently in view of Theorem, δ β (q β,, y) = q β,. This is immediate by the definitions of A β and d β (). Example 5. We give an illustration of the fact that if A U contains more than one strongly connected component, then all components other than C U consist of cycles labeled by. Here we are able to build a cycle with label t for all t N. Consider the sequence defined by U =, U tn+ = U tn + and U tn+r = U tn+r, for < r t. This is a linear recurrence sequence and we get rep U (N) = {, } {, } ( t ). Theorem 6. Let U be a linear numeration system, having a dominant root β >, such that rep U (N) is regular. If U n+ /U n β as n tends to infinity, then the only non-trivial strongly connected component is C U. Proof. Suppose that U n+ /U n β but A U has more than one non-trivial strongly connected component. Let x = x k x be a word such that δ(q U,, x) is not in C U and such that there exists an infinite sequence j < j < such that for all n, the word x jn is a greedy representation. Thus l {,..., k}, n, l U i+jn x i <. (4) U l+jn i=

INTEGERS: B () 4 Since U n+ /U n β and by Theorem, we see that l U i+jn x i U l+jn i= ( l ) + x i β i l = + i= as n +, which is not possible in view of (4). Theorem 7. Let U be a linear numeration system, having a dominant root β >, such that rep U (N) is regular. If the following conditions hold: () U n+ /U n β +, as n tends to infinity, () there exists infinitely many n such that U n+ /U n β, and (3) d β () is finite, then A U has more than one non-trivial strongly connected component. Note that, if β / N, then () holds true. Proof. From (3), we may assume that d β () = s ω, where s = s k s is a word over A β. In view of Theorem 4, to show that there is a second strongly connected component, it suffices to show that for infinitely many n the words s n are greedy representations. Equivalently, it suffices to show that for infinitely many n, we have Let l {,..., k}. We have β l k l l k s i β i l = s i β i k = i= i= l U i+n l {,..., k}, s i <. (5) U l+n i= i= k s i β i k i=l k s i β i k = i=l s i β i k β l k since d β () is obtained by using the greedy algorithm. Applying the hypotheses () and (), we obtain (5), as required. Example 8. The numeration systems of Example 5 satisfy the hypotheses of the previous theorem and we have already shown that the corresponding automata have more than one non-trivial strongly connected component. 5. State Complexity for Divisibility Criterion We now turn to second issue of this paper. Namely we will study the state complexity of rep U (mn).

INTEGERS: B () 5 Definition 9. Let U = (U n ) n be a numeration system and m be an integer. The sequence (U n mod m) n satisfies a linear recurrence relation of minimal length. This integer is denoted by k U,m or simply by k if the context is clear. This quantity is given by the largest t such that U U U t U U U t det H t (mod m), where H t =....... U t U t U t Example 3. Let m = and consider the sequence introduced in Example 4. The sequence (U n mod ) n is constant and trivially satisfies the recurrence relation U n+ = U n with U =. Therefore, we get k U, =. For m = 4, one can check that k U,4 =. Definition 3. Let U = (U n ) n be a numeration system and m be an integer. Let k = k U,m. Consider the system of linear equations H k x b (mod m) where H k is the k k matrix given in Definition 9. We let S U,m denote the number of k-tuples b in {,..., m } k such that the system H k x b (mod m) has at least one solution x. Example 3. Again take the same recurrence relation as in Example 4 and m = 4. Consider the system { x + 3 x b (mod 4) 3 x + 7 x b (mod 4) We have x b b (mod 4). Hence for each value of b in {,..., 3}, b can take at most values. One can therefore check that S U,4 = 8. Remark 33. Let l k = k U,m. Then the number of l-tuples b in {,..., m } l such that the system H l x b (mod m) has at least one solution equals S U,m. Let us show this assertion for l = k +. Let H l denote the l k matrix obtained by deleting the last column of H l and let x denote the k-tuple obtained by deleting the last element of x. Observe that the l-th column of H l is a linear combination of the other columns of H l. It follows that if b = (b,..., b k, b) T {,..., m } l is an l-tuple for which the system H l x b (mod m) has a solution, then b = (b,..., b k ) T {,..., m } k is a k-tuple for which the system H k x b (mod m) also has a solution. Furthermore, the l-th row of H l is a linear combination of the other rows of H l, so for every such b, there is exactly one b such that H l x b (mod m) has a solution. This establishes the claim.

INTEGERS: B () 6 We define two properties that A U may satisfy in order to get our results: (H.) A U has a single strongly connected component denoted by C U, (H.) for all states p, q in C U, with p q, there exists a word x pq such that δ U (p, x pq ) C U and δ U (q, x pq ) C U, or, δ U (p, x pq ) C U and δ U (q, x pq ) C U. Theorem 34. Let m be an integer. Let U = (U n ) n be a linear numeration system satisfying the recurrence relation () such that (a) N is U-recognizable and A U satisfies the assumptions (H.) and (H.), (b) (U n mod m) n is purely periodic. Then the number of states of the trim minimal automaton A U,m of the language rep U (mn) from which infinitely many words are accepted is (#C U )S U,m. From now on we fix an integer m and a numeration system U = (U n ) n satisfying the recurrence relation () and such that N is U-recognizable. Let k = k U,m. Definition 35. We define a relation U,m over A U. For all u, v A U, { u u U,m v rep U (N) v and i {,..., k }, val U (u i ) val U (v i ) (mod m) where rep U (N) is the Myhill-Nerode equivalence for the language rep U (N) accepted by A U. Lemma 36. Let u, v, x A U. If u U,m v and ux, vx rep U (N), then ux U,m vx and in particular, val U (ux) val U (vx) (mod m). Proof. By assumption, for all i {,..., k }, val U (u i ) val U (v i ) (mod m). Hence, for all i {,..., k }, a i val U (u i ) a i val U (v i ) (mod m) where the a i s are the coefficients in (). Assume that u = u l u. Note that k l a i val U (u i ) = i= j= l a i U j+i = u j U j+k = val U (u k ). u j k i= Therefore, we can conclude that val U (u k ) val U (v k ) (mod m). Iterating this argument, we have j= n, val U (u n ) val U (v n ) (mod m). (6)

INTEGERS: B () 7 Since the Myhill-Nerode relation is a right congruence, we have that ux rep U (N) vx. Let i {,..., k }. From (6), we deduce that val U (u x +i ) + val U (x i ) val U (v x +i ) + val U (x i ) (mod m) and therefore val U (ux i ) val U (vx i ) (mod m). Proposition 37. Assume that the numeration system U satisfies the assumptions of Theorem 34. Let u, v A U be such that δ U(q U,, u) and δ U (q U,, v) belong to C U. We have u U,m v if and only if u rep U (mn) v. Proof. From (b) the sequence (U n mod m) n is purely periodic, say of period p. Assume that u U,m v. Our aim is to show that there exists a word y A U that distinguishes u and v in the minimal automaton of rep U (mn), i.e., either uy rep U (mn) and vy rep U (mn), or uy rep U (mn) and vy rep U (mn). As a first case, assume u rep U (N) v. Since δ U (q U,, u) and δ U (q U,, v) both belong to C U, this means that δ U (q U,, u) and δ U (q U,, v) are two different states in C U. By (H.), without loss of generality, we may assume that there exists a word x such that δ U (q U,, ux) C U and δ U (q U,, vx) C U. Since by (H.) A U contains only one strongly connected component, only finitely many words may be accepted from δ U (q U,, vx). Let T be the length of the longest word accepted from δ U (q U,, vx). Let i {,..., m} be such that val U (ux) + i (mod m). Using properties (ii) (iv) from Theorem 9 i times and the fact that δ U (q U,, ) is final, there exist r,..., r i > such that the word y = x( rp )( rp ) ( rip ) has a length larger than T + x and is such that uy is a greedy representation. Moreover, due to the periodicity of (U n mod m) n, we have val U (uy) (mod m) and therefore uy belongs to rep U (mn). Hence, the word y distinguishes u and v for the language rep U (mn). Now assume that u rep U (N) v and there exists j {,..., k } such that val U (u j ) val U (v j ) (mod m). There exists i < m such that val U (u j ) + i (mod m) and val U (v j )+i (mod m). As in the first case there exist s,..., s i > such that the word distinguishes u and v. y = ( sp )( sp ) ( sip )

INTEGERS: B () 8 Consider the other implication and assume that u U,m v. Let x be a word such that ux rep U (mn). From Lemma 36, we only have to show that vx is a greedy representation, which is true since u rep U (N) v. Hence the conclusion follows. Proof of Theorem 34. If u is a word such that δ U (q U,, u) belongs to C U, then with the same reasoning as in the proof of Proposition 37, there exist infinitely many words x such that ux rep U (mn). On the other hand, by (H.), if v is a word such that δ U (q U,, v) does not belong to C U, there exist finitely many words x such that vx rep U (mn). Therefore, the number of states of the trim minimal automaton of the language rep U (mn) from which infinitely many words are accepted is the number of sets u rep U (m N) where u is a word over A U such that δ U (q U,, u) belongs to C U. Hence, as a consequence of Proposition 37, this number is also the number of equivalence classes [u] U,m with u being such that δ U (q U,, u) C U. What we have to do to conclude the proof is therefore to count the number of such equivalence classes. First we show that there are at most #C U S U,m such classes. By definition, if u, v A U are such that δ U (q U,, u) δ U (q U,, v), then u U,m v. Otherwise, u U,m v if and only if there exists l < k such that val U (u l ) val U (v l ) (mod m). Let u = u r u A U. We let b u denote the k-tuple (b,..., b k ) T {,..., m } k defined by s {,..., k }, val U (u s ) b s (mod m). (7) Using the fact that the sequence (U n ) n satisfies (), there exist α,..., α k such that r k s {,..., k }, val U (u s ) = u i U i+s = α i U i+s. (8) Using (7) and (8), we see that the system H k x b u (mod m) has a solution x = (α,..., α k ) T. If u, v A U are such that δ U(q U,, u) = δ U (q U,, v) but u U,m v, then b u b v. From the previous paragraph the systems H k x b u (mod m) and H k x b v (mod m) both have a solution. Therefore, there are at most #C U S U,m infinite equivalence classes. Second we show that there are at least #C U S U,m such classes. Let c = (c,..., c k ) T {,..., m } k be such that the system H k x c (mod m) has a solution x c = (α,..., α k ) T. Let q be any state in C U. Our aim is to build a word y over A U such that δ U (q U,, y) = q and, for all s {,..., k }, val U (y s ) c s i= i= (mod m). Since A U is accessible, there exists a word u A U such that δ U (q U,, u) = q. With this word u is associated a unique b u = (b,..., b k ) T {,..., m } k given by (7). The system H k x b u (mod m) has a solution denoted by x u.

INTEGERS: B () 9 Define γ,..., γ k {,..., m } by x c x u (γ,..., γ k ) T (mod m). Thus H k (x c x u ) c b u (mod m). (9) Using properties (ii) (iv) from Theorem 9 from the initial state q U,, there exist t,,..., t,γ such that the word w = ( pt, ) ( pt,γ ) satisfies δ U (q U,, w ) C U F U and val U (w ) γ U (mod m). We can iterate this construction. For j {,..., k}, there exist t j,,..., t j,γj such that the word w j = w j ( ptj, j j ) ( ptj,γ j j j ) satisfies δ U (q U,, w j ) C U F U and val U (w j ) val U (w j ) + γ j U j (mod m). Consequently, we have val U (w k ) γ k U k + + γ U (mod m). Now take r and r large enough such that δ U (q U,, w k rp ) = q U, and r p u. Such an r exists by (ii) in Theorem 9. The word y = w k (r+r )p u u is such that δ U (q U,, y) = δ U (q U,, u) = q and taking into account the periodicity of (U n mod m) n, we get val U (y) val U (w k ) + val U (u) (mod m). In view of (9), we obtain k s {,..., k }, val U (y s ) γ i U i+s + b s c s b s + b s = c s i= (mod m). Corollary 38. Assume that the numeration system U satisfies the assumptions of Theorem 34. Assume moreover that A U is strongly connected (i.e., A U = C U ). Then the number of states of the trim minimal automaton of the language rep U (mn) is (#C U )S U,m. Proof. We use the same argument as in the beginning of the proof of Theorem 34. Since A U = C U, all of the sets u rep U (m N) are infinite. Hence, infinitely many words are accepted from any state of A U,m.

INTEGERS: B () Corollary 39. Let l. For the l-bonacci numeration system U = (U n ) n defined by U n+l = U n+l + + U n and U i = i for all i < l, the number of states of the trim minimal automaton of the language rep U (mn) is lm l. Proof. First note that the trim minimal automaton of rep U (N) consists of a unique strongly connected component made of l states (see Figure ) and A U satisfies all the required assumptions. The matrix H l has a determinant equal to ±. Therefore, for all b {,..., m } l, the system H l x b (mod m) has a solution. There are m l such vectors b. We conclude by using Corollary 38. Remark 4. Compared to Alexeev s result () the previous formula is much simpler. This can be explained by the fact that the last coefficient in the recurrence relation defining the l-bonacci numeration system is equal to, which is invertible modulo m for all m. To build the minimal automaton of rep U (mn), one can use Theorem to first have an automaton accepting the reversal of the words over A U whose numerical value is divisible by m. We consider the reversal representations, that is least significant digit first, to be able to handle the period of (U n mod m) n. Such an automaton has m times the length of the period of (U n mod m) n states. Then minimizing the intersection of the reversal of this automaton with the automaton A U, we get the expected minimal automaton of rep U (mn). Taking advantage of Proposition 37, we get an automatic procedure to obtain directly the minimal automaton A U,m of rep U (mn). States of A U,m are given by (k + )-tuples. The state reached by reading w has as first component the state of A U reached when reading w and the other components are val U (w) mod m,..., val U (w k ) mod m. Example 4. Consider the Fibonacci numeration system and m = 3. The states of A U depicted in Figure are denoted by q and q. The states of A U,3 are r,..., r 7. The transition function of A U,3 is denoted by τ and is described in Table. All the systems presented in Examples, and 4 are Bertrand numeration systems. As a consequence of Parry s theorem [3], the canonical automaton A β associated with β-expansions is a trim minimal automaton (therefore, any two distinct states are distinguished) which is moreover strongly connected. The following result is therefore obvious. Proposition 4. Let U be the Bertrand numeration system associated with a noninteger Parry number β >. The set N is U-recognizable and the trim minimal automaton A U of rep U (N) fulfills properties (H.) and (H.). Another option is to consider a non-deterministic finite automaton reading most significant digits first.

INTEGERS: B () w r = (δ U (q, w), val U (w), val U (w)) τ(r, ) τ(r, ) ε,, 3 r = (q,, ) r r r = (q,, ) r, r = (q,, ) r 3 r 4 r 3 = (q,, ) r 5 r 6 r 4 = (q,, ) r 7, () 3 r 5 = (q,, ) r 8 r 9 r 6 = (q,, ) r, () r 7 = (q,, ) r r 4, 4 r 8 = (q,, ) r r 3 3 r 9 = (q,, ) r, 7 r = (q,, ) r 7 r 4 r = (q,, ) r 5 5 r = (q,, ) r 5 r 6 4 r 3 = (q,, ) r 8 r 4 = (q,, ) r 6 r 5 = (q,, ) r r 7 5 r 6 = (q,, ) r 5 6 r 7 = (q,, ) r 3 Table : The transition function of A U,3. We can therefore apply Theorem 34 to the class of Bertrand numeration systems. Finally, we give a lower bound when the numeration system satisfies weaker hypotheses than those of Theorem 34. Proposition 43. Let U be any numeration system (not necessarily linear). The number of states of A U,m is at least rep U (m). Proof. Let n = rep U (m). For each i {,..., n}, we define p i (resp. s i ) to be the prefix (resp. suffix) of length i (resp. n i) of rep U (m). We are going to prove that for all i, j {,..., n}, we have p i rep U (m N) p j. Let i, j {,..., n}. We may assume that i < j. Obviously, the word p j s j belongs to rep U (m N). On the other hand, observe that p i s j {,..., n }. Therefore the word p i s j does not belong to rep U (m N) since it cannot simultaneously be greedy and satisfy val U (p i s j ) (mod m). Hence, the word s j distinguishes p i and p j. 6. Perspectives and Conjectures We use the same notation as in Theorem 9. In the case where the numeration system U has a dominant root β >, if d β () is finite, then d β () = (t t m (t m )) ω where t m and then we clearly have

INTEGERS: B () #Φ (q β,i ) #Φ (q β,i+ ) for all i {,..., m }. We conjecture that, in this case, #Φ (q β,m ) =. In other words, we conjecture that, in this case, A U has at most two non-trivial strongly connected components. When the numeration system U does not satisfy the dominant root condition, we have not provided a precise description of A U. In this case, new kinds of phenomena may appear. For instance, in the following two examples, there exist more than one non-trivial strongly connected components containing transitions not labeled by. Furthermore, thanks to the first example, we see that A U may have more than two non-trivial strongly connected components. Example 44. Consider the numeration system (U n ) n defined by U n+3 = 4U n and (U, U, U ) = (,, 6). The corresponding trim minimal automaton is depicted in Figure 9. States in the same strongly connected component have the same label:, and 3, respectively.,,, 3, 3,,, 3 3 3 3, 3,,,, Figure 9: An automaton A U for a numeration system U = (U n ) n not satisfying the dominant root condition. Example 45. Consider the numeration system (U n ) n defined by U n+4 = 3U n+ +U n and (U, U, U, U 3 ) = (,, 3, 7). The corresponding trim minimal automaton is depicted in Figure. Again, states in the same strongly connected component have the same label: and, respectively. Even though the sequence U n+ /U n does not converge, we have lim n + U n+ /U n = lim n + U n+3 /U n+ = (3 + 3)/. Note that the latter observation is consistent with Hollander s conjecture [6]. With the same assumptions as in Theorem 34, can we count the number of states from which only finitely many words are accepted? Can we weaken the assumptions of Theorem 34? If X is a finite union of arithmetic progressions, can we give bounds for the number of states of the trim minimal automaton accepting rep U (X)?

INTEGERS: B () 3, Figure : An automaton A U for a numeration system U = (U n ) n not satisfying the dominant root condition. References [] B. Alexeev, Minimal DFA for testing divisibility, J. Comput. Syst. Sci. 69 (4), 35 43. [] J.-P. Allouche, N. Rampersad, J. Shallit, Periodicity, repetitions, and orbits of an automatic sequence, Theoret. Comput. Sci. 4 (9), 795 83. [3] J. P. Bell, E. Charlier, A. S. Fraenkel, M. Rigo, A decision problem for ultimately periodic sets in non-standard numeration systems, Int. J. Algebra and Computation 9 (9), 89 839. [4] V. Berthé, M. Rigo, Eds., Combinatorics, Automata and Number Theory, Encyclopedia of Math. and its Applications, vol. 35, Cambridge University Press (). [5] A. Bertrand, Comment écrire les nombres entiers dans une base qui n est pas entière, Acta Math. Hungar. 54 (989), 37 4. [6] V. Bruyère, G. Hansel, Bertrand numeration systems and recognizability, Theoret. Comput. Sci. 8 (997), 7 43. [7] V. Bruyère, G. Hansel, C. Michaux, R. Villemaire, Logic and p-recognizable sets of integers, Bull. Belg. Math. Soc. (994), 9 38. [8] E. Charlier, N. Rampersad, M. Rigo, L. Waxweiler, State complexity of testing divisibility, Proceedings of the th Annual Workshop on Descriptional Complexity of Formal Systems, McQuillan, I., Pighizzini, G. (Eds), 48 57 (). [9] E. Charlier, N. Rampersad, M. Rigo, L. Waxweiler, Structure of the minimal automaton of a numeration language and applications to state complexity, Proceedings of the 3th Mons Theoretical Computer Science Days, Amiens, 6 September. [] E. Charlier, M. Rigo, A decision problem for ultimately periodic sets in non-standard numeration systems, Lect. Notes in Comput. Sci. 56 (8), Mathematical Foundations of Computer Science 8, 4 5. [] A. Cobham, On the base-dependence of sets of numbers recognizable by finite automata, Math. Systems Theory 3 (969) 86 9. [] S. Eilenberg, Automata, languages, and machines, Vol. A, Pure and Applied Mathematics, Vol. 58, Academic Press, New York (974). [3] S. Fabre, Substitutions et β-systèmes de numération, Theoret. Comput. Sci. 37 (995), 9 36. [4] C. Frougny, Representations of numbers and finite automata, Math. Systems Theory 5 (99), 37 6.

INTEGERS: B () 4 [5] C. Frougny, B. Solomyak, On representation of integers in linear numeration systems, in Ergodic theory of Z d actions (Warwick, 993 994), 345 368, London Math. Soc. Lecture Note Ser. 8, Cambridge Univ. Press, Cambridge (996). [6] M. Hollander, Greedy numeration systems and regularity, Theory Comput. Systems 3 (998), 33. [7] J. Honkala, A decision method for the recognizability of sets defined by number systems, Theor. Inform. Appl. (986), 395 43. [8] D. Krieger, A. Miller, N. Rampersad, B. Ravikumar, J. Shallit, Decimations of languages and state complexity, Theoret. Comput. Sci. 4 (9), 4 49. [9] P. Lecomte, M. Rigo, Numerations systems on a regular language, Theory Comput. Syst. 34 (), 7 44. [] P. Lecomte, M. Rigo, Real numbers having ultimately periodic representations in abstract numeration systems, Inform. and Comput. 9 (4), 57 83. [] N. Loraud, β-shift, systèmes de numération et automates. J. Théor. Nombres Bordeaux 7 (995), 473 498. [] M. Lothaire, Algebraic Combinatorics on Words, Encyclopedia of Math. and its Applications, vol. 9, Cambridge University Press (). [3] W. Parry, On the β-expansions of real numbers, Acta Math. Acad. Sci. Hungar. (96), 4 46. [4] J. Shallit, Numeration systems, linear recurrences, and regular sets, Inform. and Comput. 3 (994), 33 347.