Continued fraction expansions and permutative 9 0 representations of the Cuntz algebra 0 O∞ 2 n Katsunori Kawamura∗, Yoshiki Hayashi† a College of Science and Engineering Ritsumeikan University, J 1-1-1 Noji Higashi, Kusatsu, Shiga 525-8577, Japan, 5 1 and Dan Lascu ‡ ] A Mircea cel Batran Naval Academy, 1 Fulgerului, 900218 Constanta, Romania O . h t a Abstract m [ We show a correspondence between simple continued fraction ex- pansionsofirrationalnumbersandirreduciblepermutativerepresenta- 1 tionsoftheCuntzalgebra ∞. Withrespecttothecorrespondence,it v O isshownthatthe equivalenceofrealnumberswithrespecttomodular 6 8 transformationsis equivalentto the unitary equivalence of representa- 1 tions. Furthermore, we show that quadratic irrationals are related to 2 irreducible permutative representations of ∞ with a cycle. . O 1 0 Mathematics Subject Classifications (2000). 11A55, 46K10 9 0 Key words. Cuntz algebra, permutative representation, continued frac- : v tion expansion. i X r a 1 Introduction The purpose of this paper is to show a new relation between number theory and the representation theory of operator algebras. At the beginning, we show our motivation. Explicit mathematical statements will be given after 1.2. The main theorems will be shown in 1.4. § § e-mail: [email protected]. ∗ †e-mail: [email protected]. ‡e-mail: [email protected]. 1 1.1 Motivation Continued fractions furnish important tools in number theory [9, 11, 16], and continued fraction transformations induce typical dynamical systems [3, 4, 10, 17]. It is well known that the dynamical system of the simple continued fraction transformation on the set of irrational numbers in the interval [0,1] is conjugate with the one-sided full shift on the set N ∞ ≡ (n ) :n N for all i [15]byusingcontinuedfractionexpansionswhere i i 1 i { ≥ ∈ } N 1,2,3,... . ≡ { } On the other hand, such dynamical systems induce representations of Cuntz algebras and their relations were studied [12, 14]. For example, the one-sidedfullshiftonN inducestheshiftrepresentationof ,whichacts ∞ O∞ on the representation space l (N ) [2]. The shift representation is decom- 2 ∞ posed into the direct sum of irreducibles unique up to unitary equivalence and the decomposition is multiplicity free. Every irreducible component in the decomposition is a permutative representation, and any irreducible permutative representation appears in the decomposition. From these facts, we are interested in a relation between continued fraction expansions of irrationals and such representations of by the in- O∞ termediary of theone-sided fullshifton N . We roughly illustrate relations ∞ among theories as follows: One-sided full shift on N ∞ continued fraction → shift representation expansion → ← ← ? Irrationals in [0,1] Representations of O∞ ←→ The position in the above question mark is the content of this study. 1.2 Continued fraction expansion map and continued frac- tion transformation on the set of irrationals We review the continued fraction expansion map and the continued fraction transformation according to [9]. Let [0,1] denote the closed interval from 0 to 1, and let Ω denote the set of all irrationals in [0,1], that is, Ω= [0,1] Q. (1.1) \ Remark that we consider only Ω but not the whole of [0,1] in this paper. Any x Ω has a unique infinite continued fraction expansion ([9], Theorem ∈ 2 170), that is, there exists a unique infinite sequence (a (x)) of positive i i 1 ≥ integers such that 1 x= . (1.2) 1 a (x)+ 1 1 a (x)+ 2 1 a (x)+ 3 ... From this, we define the map CFE from Ω to N by ∞ CFE(x) (a (x)) . (1.3) i i 1 ≡ ≥ It is known that the map CFE is bijective ([9], Theorem 161, 166, 169). We call CFE the continued fraction expansion map. If x Ω is a solution of ∈ a quadratic equation with integral coefficients, then we call x a quadratic irrational. Fact 1.1 ([9], Theorem 177) If x Ω is a quadratic irrational, then there ∈ exists a finite sequence (m ,...,m ) Nl and a nonperiodic finite 1 l ∈ ∪ {∅} sequence (n ,...,n ) Nk such that 1 k ∈ CFE(x) = (m ,...,m ,n ,...,n ,n ,...,n ,n ,...,n ,...). (1.4) 1 l 1 k 1 k 1 k In Fact 1.1, we call (n ,...,n ) the repeating block of CFE(x) ([16], Chap. 1 k 3) where we suppose that there is no shorter such repeating block and that the initial block does not end with a copy of the repeating block. Hence the repeating block of CFE(x) is uniquely defined for each quadratic irrational x in Ω. The converse of Fact 1.1 is also true ([9], Theorem 176). 1.3 Permutative representations of O∞ We review permutative representations of the Cuntz algebra in this O∞ subsection. 1.3.1 O∞ Let denote the Cuntz algebra [6], that is, a C -algebra which is univer- ∗ O∞ sally generated by s :i N satisfying i { ∈ } k s s = δ I (i,j N), s s I, (k N), (1.5) ∗i j ij ∈ i ∗i ≤ ∈ Xi=1 3 where I denotes the unit of . O∞ A -representationof isapair( ,π)suchthatπisa -homomorphism ∗ O∞ H ∗ from to the C -algebra ( ) of all bounded linear operators on a com- ∗ O∞ L H plex Hilbert space [1]. We call -representation as representation for the H ∗ simplicity of description. For two representations ( ,π ) and ( ,π ) of 1 1 2 2 H H , ( ,π ) and ( ,π ) are unitarily equivalent if there exists a unitary 1 1 2 2 O∞ H H u from onto such that uπ (x)u = π (x) for each x . We state 1 2 1 ∗ 2 H H ∈ O∞ that a representation ( ,π) of is irreducible if there is no invariant H O∞ closed subspace of except 0 and ; ( ,π) is multiplicity free if any two H { } H H subrepresentations of ( ,π) are not unitarily equivalent. A representation H ( ,π) is irreducible if and only if the commutant X ( ) : Xπ(A) = H { ∈ L H π(A)X for all A of π( ) equals to CI. ∈ O∞} O∞ Since is simple, that is, there is no nontrivial closed two-sided O∞ ideal, any representation of is injective. If t : i N are bounded i O∞ { ∈ } operators on a Hilbert space such that t : i N satisfy (1.5), then i H { ∈ } the correspondence s t for i N is uniquely extended to a unital - i i 7→ ∈ ∗ representation of on from the uniqueness of . Therefore we call O∞ H O∞ such a correspondence among generators by a representation of on . O∞ H Assume that s : i N are realized as operators on a Hilbert space . i { ∈ } H According to (1.5), is decomposed intoorthogonal subspacesas i Nsi . H ⊕∈ H Since s is an isometry, s has the same dimension as . From this, we i i H H see that there is no finite dimensional representation of which preserves O∞ the unit. The following illustration is helpful in understanding s :i N : i { ∈ } H s i ↓ s s 1 i H ··· H ··· The algebra appears in quantum field theory [13] and metrical O∞ number theory [14]. 1.3.2 Permutative representations We review permutative representations in this subsubsection. Definition 1.2 [2,7,8,13]Let s :i N denote thecanonicalgenerators i { ∈ } of . O∞ 4 (i) A representation ( ,π) of is permutative if there exists an or- H O∞ thonormal basis ( ) of such that π(s ) for each i N. i E ⊂ H H E ⊂ E ∈ (ii) For J = (j )k Nk with 1 k < , let P(J) denote the class l l=1 ∈ ≤ ∞ of representations ( ,π) of with a cyclic unit vector v such that π(s )v = v anHd π(s O∞s )v k is an orthonormal ∈famHily in J { jl··· jk }l=1 where s s s . H J ≡ j1··· jk (iii) For J = (j ) N , let P(J) denote the class of representations l l 1 ∞ ≥ ∈ ( ,π) of with a cyclic unit vector v such that π(s ) v : H O∞ ∈ H { J(n) ∗ n N is an orthonormal family in where J (j ,...,j ). (n) 1 n ∈ } H ≡ The vector v in both (ii) and (iii) is called the GP vector of ( ,π). H We recall properties of these classes as follows: For any J, P(J) in Definition 1.2(ii) and (iii) always exists and it is a class of permutative representations, which contains only one unitary equivalence class. From this, we can always identify P(J) with a representative of P(J) [2, 7, 8]. A representation ( ,π) of is a permutative representation with a cycle (chain) if there exHists J ON∞k for 1 k < (resp. J N ) such that ∞ ∈ ≤ ∞ ∈ ( ,π) is P(J). Details will be explained in 2.2. H § 1.4 Main theorems We show our main theorems in this subsection. For this purpose, we con- structtwo representations of as follows. For anonempty setA, let l (A) 2 O∞ denote the complex Hilbert space with an orthonormal basis e : a A . a { ∈ } We call e : a A the standard basis of l (A). a 2 { ∈ } Definition 1.3 Let Ω be as in (1.1) and let s :i N denote the canon- i { ∈ } ical generators of . O∞ (i) For i N, define the map α from Ω to Ω by i ∈ 1 α (x) (x Ω). (1.6) i ≡ x+i ∈ Define the representation π of on l (Ω) by α 2 O∞ π (s )e e (x Ω, i N). (1.7) α i x ≡ αi(x) ∈ ∈ (ii) For i N, define the map β from N to N by i ∞ ∞ ∈ β (n ,n ,...) (i,n ,n ,...) ((n ,n ,...) N , i N). (1.8) i 1 2 1 2 1 2 ∞ ≡ ∈ ∈ 5 Define the representation π of on l (N ) by β 2 ∞ O∞ π (s )e e (a N , i N). (1.9) β i a ≡ βi(a) ∈ ∞ ∈ The representation (l (N ),π ) is called the shift representation [2]. 2 ∞ β Then the following holds. Theorem 1.4 Two representations (l (Ω),π ) and (l (N ),π ) are uni- 2 α 2 ∞ β tarily equivalent. From Theorem 1.4, we can compare irreducible components of (l (Ω),π ) 2 α with those of (l (N ),π ) and consider how an irreducible component in 2 ∞ β (l (Ω),π )isrealized. Forthispurpose,weintroduceanequivalencerelation 2 α of real numbers as follows. Definition 1.5 [5, 9, 16] If x and y are two real numbers such that ay+b x = (a,b,c,d Z) (1.10) cy+d ∈ where ad bc = 1, then x is said to be equivalent to y. In this case, we − ± write x y. The transformation (1.10) is called a modular transformation ∼ in a broad sense. Remark that the transformation (1.10) for a,b,c,d with ad bc = 1 is also − called a modular transformation in a narrow sense. According to the unitary equivalence in Theorem 1.4, the following holds. Theorem 1.6 For x Ω, let [x] denote the equivalence class of x in Ω with ∈ respect to modular transformations, that is, [x] = y Ω : y x . Let { ∈ ∼ } (l (Ω),π ) be as in Definition 1.3(i). 2 α (i) The following irreducible decomposition holds: l (Ω)= (1.11) 2 [x] H M [x] Ω/ ∈ ∼ where denotes the closed subspace of l (Ω) generated by the subset [x] 2 H e : y [x] . y { ∈ } (ii) For x Ω, let η denote the subrepresentation of π associated with [x] α ∈ the subspace in (1.11), that is, [x] H η π . (1.12) [x] ≡ α|H[x] Then η and η are unitarily equivalent if and only if x y. Espe- [x] [y] ∼ cially, (1.11) is multiplicity free. 6 (iii) Let CFE be as in (1.3) and let Ω(2) denote the set of all quadratic irrationals in Ω. (a) If x Ω Ω(2), then η is P(CFE(x)). [x] ∈ \ (b) If x Ω(2), then η is P(CFE (x)) where CFE (x) denotes the [x] 0 0 ∈ repeating block of CFE(x). (iv) Any irreducible permutative representation of is unitarily equiva- O∞ lent to η for some [x] Ω/ . [x] ∈ ∼ In consequence, Theorem 1.6 shows that the set Ω/ of all equiva- ∼ lence classes of irrationals in [0,1] is one-to-one correspondence in the set IPR( )/ of all unitary equivalence classes of irreducible permutative O∞ ∼ representations of : O∞ η Ω/ = IPR( )/ ; [x] η (1.13) ∼ ∼ O∞ ∼ 7→ [x] where we identify η with the unitary equivalence class of η for conve- [x] [x] nience. Especially, the following equivalence holds as the restriction of η on the subset Ω(2)/ of Ω/ : ∼ ∼ η Ω(2)/ = IPR ( )/ (1.14) ∼ ∼ cycle O∞ ∼ where IPR ( )/ denotes the set of all unitary equivalence classes of cycle O∞ ∼ irreducible permutative representations of with a cycle. O∞ Remark 1.7 The equivalence of numbers by modular transformations is well known in number theory, such that the discriminant of an irrational is invariant with respect to modular transformations [5, 18]. On the other hand, the unitary equivalence of representations is basic in the representa- tion theory of -algebras. Since these two equivalence relations are indepen- ∗ dently introduced in different mathematical areas, the equivalence of two equivalence relations in Theorem 1.6(ii) is nontrivial. From Theorem 1.6, the following natural questions are thought up. Problem 1.8 (i) Show the meaningof thediscriminant for the represen- tation associated with a quadratic irrational. What are the discrimi- nant and the class number ([18], p. 59) in the representation theory of ? O∞ (ii) Findsimilarrelations betweentheCuntzalgebra with2 N < N O ≤ ∞ [6] and real numbers. 7 (iii) From Theorem 1.6, we suspect that there exists a relation between the group of all modular transformations and . Make clear this O∞ relation. In 2, we prove Theorem 1.4 and 1.6. In 3, we show examples of § § Theorem 1.6. 2 Proofs of theorems In this section, we prove main theorems. 2.1 Continued fraction expansion and one-sided full shift We review relations between continued fraction expansions and the one- sided full shift on N in this subsection. Let [a (x),a (x),...] denote (1.2) ∞ 1 2 for simplicity of description. Fact 2.1 ([9], Theorem 175) For the equivalence in Definition 1.5, two ir- rational numbers x and y are equivalent if and only if there exist positive integers a ,...,a , b ,...,b such that 1 m 1 n x= [a ,...,a ,c ,c ,...], y = [b ,...,b ,c ,c ,...]. (2.1) 1 m 1 2 1 n 1 2 Definethesimplecontinuedfractiontransformation (ortheGaussmap) τ from Ω to Ω by 1 1 τ(x) (x Ω), (2.2) ≡ x −(cid:22)x(cid:23) ∈ where denotes the floor (entire) function [10, 17]. Then we see that ⌊·⌋ τ([a ,a ,...]) = [a ,a ,...] for the continued fraction [a ,a ,...]. 1 2 2 3 1 2 Define the map σ from N to N by ∞ ∞ σ(n ,n ,...) (n ,n ,...). (2.3) 1 2 2 3 ≡ Thedynamicalsystem (N ,σ)is called theone-sided full shiftonN ([15], ∞ ∞ 7.2). Then we see that CFE in (1.3) satisfies the following equation ([10], § (1.1.2)): CFE(τ(x)) = σ(CFE(x)) (x Ω). (2.4) ∈ From (2.4), two dynamical systems (Ω,τ) and (N ,σ) are conjugate. ∞ Definition 2.2 For a,b N , let a b denote when there exist p,q 1 ∞ ∈ ∼ ≥ such that σp(a) = σq(b). 8 We call the tail equivalence in N ([2], Chap. 2). Then the following ∞ ∼ holds by definition. Fact 2.3 For x,y Ω, x y if and only if CFE(x) CFE(y). ∈ ∼ ∼ For α : i N and β : i N in Definition 1.3, the following holds: i i { ∈ } { ∈ } CFE(α (x)) = β (CFE(x)) (x Ω, i N). (2.5) i i ∈ ∈ Remark that they are closely related to τ and σ as follows: (τ α )(x) = x, (σ β )(a) = a (x Ω, a N , i N). (2.6) i i ∞ ◦ ◦ ∈ ∈ ∈ 2.2 Properties of representations of O∞ In this subsection, we recall properties of permutative representations and theshiftrepresentationof . DefineN Nk. ForJ = (j )m ,K = O∞ ∗ ≡ 1≤k<∞ l l=1 (kl)ml=′1 ∈ N∗, we write J ∼ K if m = mS′ and K = pJ where pJ = (j ,...,j ) for any cyclic permutation p Z . For J N , we call J p(1) p(m) m ∗ ∈ ∈ nonperiodic if pJ = J for any cyclic permutation p = id. If J N is non- ∗ 6 6 ∈ periodic, then we see that ( ,π) is P(J) if and only if there exists a cyclic H vector v such that π(s )v = v. For J N , we call J nonperiodic if J J ∞ ∈ H ∈ has no repeating block. Proposition 2.4 Let P(J) be as in Definition 1.2. (i) Any permutative representation of is decomposed into the direct O∞ sumof cyclic permutative representations uniquelyup to unitary equiv- alence. (ii) Any cyclic permutative representation is either one of the following two cases: (a) P(J) for J N . ∗ ∈ (b) P(J) for J N . ∞ ∈ (iii) For two representations π and π of , let π π denote the 1 2 1 2 O∞ ∼ unitary equivalence between π and π . If J N and K N , then 1 2 ∗ ∞ ∈ ∈ P(J) P(K). 6∼ (iv) For J,K N N , then P(J) P(K) if and only if J K where ∗ ∞ ∈ ∪ ∼ ∼ we define J K when J N and K N . ∗ ∞ 6∼ ∈ ∈ (v) For J N N , P(J) is irreducible if and only if J is nonperiodic. ∗ ∞ ∈ ∪ 9 Proof. See Appendix A.1. Next, we show properties of the shift representation of (see also O∞ [2], Chap. 6). Proposition 2.5 Let (l (N ),π ) be as in Definition 1.3(ii). Define [a] 2 ∞ β ≡ b N :b a where is as in Definition 2.2. ∞ { ∈ ∼ } ∼ (i) The following irreducible decomposition holds: l (N ) = (2.7) 2 ∞ [a] K [a]MN / ∈ ∞ ∼ where denotes the closed subspace of l (N ) generated by the set [a] 2 ∞ K e :b [a] . b { ∈ } (ii) For a N , let θ denote the subrepresentation of π associated with ∞ [a] β ∈ the subspace , that is, [a] K θ π . (2.8) [a] ≡ β|K[a] Then θ and θ are unitarily equivalent if and only if a b. Espe- [a] [b] ∼ cially, (2.7) is multiplicity free. (iii) (a) If a N has no repeating block, then θ is P(a). ∞ [a] ∈ (b) If a N∞ has the repeating block a′, then θ[a] is P(a′). ∈ (iv) Any irreducible permutative representation of is unitarily equiva- O∞ lent to θ for some [a] N / . [a] ∞ ∈ ∼ Proof. See Appendix A.2. 2.3 Proofs of Theorem 1.4 and 1.6 We prove Theorem 1.4 and 1.6 in this subsection. Proof of Theorem 1.4. Let CFE be as in (1.3). Define the unitary U from l (Ω) to l (N ) by 2 2 ∞ Ue e′ (x Ω) (2.9) x ≡ CFE(x) ∈ where we write {ex : x ∈ Ω} and {e′a : a ∈ N∞} as standard basis of l2(Ω) and l (N ), respectively. From (2.5), we can verify that Uπ (s )U = 2 ∞ α i ∗ 10