ebook img

A refinement of a classic theorem on continued fractions PDF

0.11 MB·English
Save to my drive
Quick download
Download
Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.

Preview A refinement of a classic theorem on continued fractions

A refinement of a classic theorem on continued fractions Paloma Bengoechea 3 January 28, 2013 1 0 2 n Abstract a J 5 We express the set of transformations occuring in two different continued 2 fraction algorithms as subspaces of PGL(2,Z) defined by certain simplelinear inequalities. As a consequence, we improve a Hurwitz’s classic theorem on ] T continued fractions giving, for γ PGL(2,Z), a bound depending only on γ N ∈ for the index of the term from which the continued fractions of two irrational . h numbers related by γ start being identical. t a m Different algorithms of Diophantine approximation give different expansions of [ a real number x in continued fraction. In the first section we express the set of 1 transformations appearing in the algorithm of classic positive continued fraction for v an irrational number x as a subspace of PGL(2,Z) defined by linear inequalities 4 4 on the action on and x. In the second section we give a similar description for 9 ∞ the transformations appearing in a different algorithm of continued fraction playing 5 . an important role in [B]. This proves a special case of a more general reduction 1 0 conjecture formulated by Zagier. 3 1 In the third section, we improve Hurwitz’s classic theorem which says that the : classic continued fractions of two irrational numbers are the same after a finite v i number of steps if and only if the numbers are PGL(2,Z)-equivalent. In Hurwitz’s X theorem the index from which the continued fractions start being identical depends r a on the irrational numbers themselves and hence it is not bounded. We give, for γ PGL(2,Z), a bound depending only on γ for the index of the term from which ∈ the continued fractions of two irrational numbers on the orbit of γ coincide. 1 Classic continued fraction We denote by Γ the group PGL(2,Z) andby ε and T the transformations that corre- a b ax+b spond to the inversion and the translation for the usual action x := (cid:18)c d(cid:19) cx+d 1 on the projective line: 0 1 1 1 ε = , T = . (cid:18)1 0(cid:19) (cid:18)0 1(cid:19) The positive continued fraction of a real number x, 1 x = n + (n Z, n 1 i 1), 0 1 0 ∈ i ≥ ∀ ≥ n + 1 1 n + 2 ... also denoted by x = [n ,n ,...], is given by the algorithm 0 1 1 x = x, n = [x ], x = = εT−ni(x ) (i 0). (1.1) 0 i i i+1 i x n ≥ i i − Clearly each x is the image of x by a matrix γ = γ Γ, given explicitly by i i i,x ∈ 0 1 0 1 γ = γ := Id, γ = γ := (i 1) (1.2) 0 0,x i i,x (cid:18)1 n (cid:19)···(cid:18)1 n (cid:19) ≥ i−1 0 − − and recursively by γ = Id, γ = εT−niγ (i 0). (1.3) 0 i+1 i ≥ Akeyideaistoreplacethesequence (γ ,γ ,γ ,...)ofelementsinΓbytheunordered 1 2 3 set Γ(x) = γ ,γ ,γ ,... Γ. 1 2 3 { } ⊂ p i The i-th convergent of x is denoted by = [n ,...,n ]. The integers p and q 0 i i i q i satisfy the recurrence p = 0 p = 1, p = n p +p (i 0), −2 −1 i i i−1 i−2 ≥ q = 1, q = 0, q = n q +q (i 0), −2 −1 i i i−1 i−2 ≥ the equation p q p q = ( 1)i (1.4) i+1 i i i+1 − − and the inequalities (1) q q 0 for all i 0 and q > q > 0 for all i 2, i i−1 i i−1 ≥ ≥ ≥ ≥ (2) p p for all i 2 and p > p for all i 3, i i−1 i i−1 | | ≥ | | ≥ | | | | ≥ 1 1+√5 i (3) q for all i 0, i ≥ √5(cid:16) 2 (cid:17) ≥ 1 1+√5 i (4) p for all i 0. i | | ≥ √5(cid:16) 2 (cid:17) ≥ 2 It is well known that any rational number p/q satisfying p 1 x < (cid:12)q − (cid:12) 2q2 (cid:12) (cid:12) (cid:12) (cid:12) (cid:12) (cid:12) is a convergent of x. The numbers δ (i 1) defined by i ≥ − δ = ( 1)i(p q x) (1.5) i i−1 i−1 − − satisfy the recurrence δ i−1 δ = x, δ = 1, δ = n δ +δ with n = −1 0 i+1 i i i−1 i − (cid:20) δ (cid:21) i and the inequalities 1 = δ > δ > ... 0. If x is rational, then x = p /q for some 0 1 i i i ≥ i and the recurrence stops with δ = 0; if x is irrational, the δ are all positive and i+1 i converge to 0 with exponential rapidity. With these notations, one has p p x δ γ−1 = i−1 i−2 , γ = i−1 . (1.6) i (cid:18)q q (cid:19) i(cid:18)1(cid:19) (cid:18) δ (cid:19) i−1 i−2 i The algorithm (1.1) is a reduction algorithm in the sense that the norms of the x vectors γ decrease to zero. i(cid:18)1(cid:19) r s r s Lemma 1.1 If and are two rational numbers such that t,u > 0, x t u t ≤ ≤ u r s and ru st = 1, then or is a convergent of x. − ± t u r s Proof. Let x be an element in [ , ]. If one of the inequalities t u r 1 s 1 x < or x < (cid:12)t − (cid:12) 2t2 (cid:12)u − (cid:12) 2u2 (cid:12) (cid:12) (cid:12) (cid:12) r s (cid:12) (cid:12) (cid:12) (cid:12) is satisfied, then or is a convergent of x. If not, then t u 1 r s r s 1 1 = = x + x + , tu t − u t − u − ≥ 2t2 2u2 (cid:12) (cid:12) (cid:12) (cid:12) (cid:12) (cid:12) (cid:12) (cid:12) (cid:12) (cid:12) (cid:12) (cid:12) (cid:12) (cid:12) (cid:12) (cid:12) (cid:12) (cid:12) r s which can only happen if t = u = 1. Then s = r+1, so [ , ] = [r,r+1] and either t u s p r p 0 0 = x = or = [x] = . u q t q 0 0 (cid:3) 3 Theorem 1.2 For all x R irrational, the set Γ(x) equals W (W W ), where 1 2 ∈ − ∪ W = γ Γ 1 γ( ) 0, γ(x) > 1 { ∈ | − ≤ ∞ ≤ } and W = γ W γ( ) = 0, det(γ) = 1 , W = γ W γ( ) = 1, det(γ) = 1 . 1 2 { ∈ | ∞ } { ∈ | ∞ − − } The sets W and W have respectively exactly one and at most one element. 1 2 Proof. One easily checks that 1 1+n 0 1 − 0 if n 2, W = − , W =  (cid:26)(cid:18) 1 n (cid:19)(cid:27) 1 ≥ 1 (cid:26)(cid:18)1 1 n (cid:19)(cid:27) 2 − 0 − − 0  if n = 1. 1 ∅  This proves the second statement. It is also easy to see that Γ(x) W (W W ). 1 2 ⊆ − ∪ δ q i−1 i−2 Indeed γ (x) = > 1 and γ ( ) = for i 1 with inequalities (1) imply i i δ ∞ −q ≥ i i−1 γ W (W W ). i 1 2 ∈ − ∪ We therefore have to show that W Γ(x) W W . Let γ Γ satisfy 1 2 ⊆ ∪ ∪ ∈ 1 γ( ) < 0, γ(x) > 1. − ≤ ∞ Denote r s u s γ−1 = , γ = − . (cid:18) t u (cid:19) (cid:18) t r (cid:19) − The conditions γ( ) < 0 and γ(x) > 0 imply that u and t have the same sign, as ∞ r s r s r s r s s r well as ux s and tx+r, so x , ( , denotes [ , ] if or [ , ] − − ∈ |t u| |t u| t u t ≤ u u t r s otherwise). By Lemma 1.1, or is a convergent of x. t u r If is a convergent of x, then γ and γ−1 are t p p +kp q +kq p kp γ−1 = i i−1 i , γ = i−1 i − i−1 − i (1.7) (cid:18)q q +kq (cid:19) (cid:18) q p (cid:19) i i−1 i i i − or p p kp q kq p +kp γ−1 = i − i−1 − i , γ = − i−1 − i i−1 i (1.8) (cid:18)q q kq (cid:19) (cid:18) q p (cid:19) i i−1 i i i − − − q with k Z and i 0. In the case (1.7), 1 γ( ) = i−1 k < 0 if and ∈ ≥ − ≤ ∞ − q − i only if k = 0 and i 1, or k = 1 and i = 0, because q q . Moreover i−1 i ≥ ≤ δ δ i i γ(x) = k > 1 if and only if k n 1 because = n . Hence i+1 i+1 δ − ≤ − (cid:20)δ (cid:21) i+1 i+1 4 1 γ( ) < 0 and γ(x) > 1 if and only if k = 0 and i 1 or n 2, k = 1 and 1 − ≤ ∞ ≥ ≥ i = 0. In the first case q p γ = i−1 − i−1 = γ (i 1); (cid:18) q p (cid:19) i+1 ≥ i i − in the second case 1 1+n γ = − 0 W . (cid:18) 1 n (cid:19) ∈ 2 0 − q i−1 In the case (1.8), if 1 γ( ) = + k < 0, then k < 0 and γ(x) = − ≤ ∞ q i δ i +k < 0, so γ W. −δ 6∈ i+1 s If is a convergent of x, then γ and γ−1 are: u p +kp p q p γ−1 = i−1 i i , γ = i − i (1.9) (cid:18)q +kq q (cid:19) (cid:18) q kq p +kp (cid:19) i−1 i i i−1 i i−1 i − − or p kp p q p γ−1 = − i−1 − i i , γ = i − i (1.10) (cid:18) q kq q (cid:19) (cid:18)q +kq p kp (cid:19) i−1 i i i−1 i i−1 i − − − − δ with k Z and i 0. In the case (1.9), γ(x) = i+1 > 1 if and only if ∈ ≥ δ kδ i i+1 − δ q i i k = n because = n . If k = n , then 1 γ( ) = < 0. So i+1 i+1 i+1 (cid:20)δ (cid:21) − ≤ ∞ −q i+1 i+1 q p γ = i − i = γ (i 0). (cid:18) q p (cid:19) i+2 ≥ i+1 i+1 − q i In the case (1.10), if 1 γ( ) = < 0, then k < 0 and γ(x) = − ≤ ∞ q +kq i−1 i δ i+1 < 0, so γ W. δ +kδ 6∈ i i+1 − 0 1 Let γ = be the unique element of W that satisfies γ( ) = 0 and (cid:18)1 u(cid:19) ∞ 1 det(γ) = 1. We have 1 < γ(x) = if and only if 0 < x + u < 1, namely − x+u u = [x], and γ = γ Γ(x). 1 − ∈ (cid:3) Remark 1.3 For x Q, Theorem 1.2 is still true if the value is allowed for γ(x), ∈ ∞ with γ W. ∈ 5 2 The slow continued fraction We consider the slower version of the classic algorithm of reduction (1.1) x +1 = T(x ) if x 0, i i i ≤   1 x = x, x =  1 = T−1ε(x ) if 0 < x 1, (2.1) 0 i+1  i i  x − ≤  i  xi −1 = T−1(xi) if xi > 1.   With this algorithm the expansion of x in continued fraction is 1 x = 1 1+ ± ±···± 1 1+ +1+ |n0| ··· 1 | {z } n1 1+···+1+... | {z } n2 | {z } where n ,n ,n ,..., are given in (1.1) and each equals the sign of n . 0 1 2 0 ± | | Each x is the image of x by a matrix γ′ = γ′ Γ given recursively by i i i,x ∈ Tγ′ if x 0, i i ≤  γ′ = Id, γ′ =  T−1εγ′ if 0 < x 1, (2.2) 0 i+1  i i ≤  T−1γ′ if x > 1.  i i    Inasimilarwayasintheprevioussection, wewillreplacethesequence(γ′,γ′,γ′,...) 1 2 3 of elements in Γ by the unordered set Γ(x)′ = γ′,γ′,γ′,... Γ, given explicitly { 1 2 3 } ⊂ by q +kq p kp Γ(x)′ = i−2 i−1 − i−2 − i−1 , 1 k n . (2.3) (cid:26)(cid:18) q p (cid:19) ≤ ≤ i(cid:27) − i−1 i−1 i≥1 Theorem 2.1 For all x R irrational, the set Γ(x)′ equals W′ W′, where ∈ − 1 W′ = γ Γ γ( ) 1, γ(x) > 0 { ∈ | ∞ ≤ − } and W′ = γ W′ γ( ) = 1, det(γ) = 1 . 1 { ∈ | ∞ − } The set W′ has exactly one element. 1 Proof. One easily proves the second statement checking that 1 n W′ = − 0 . 1 (cid:26)(cid:18) 1 n +1(cid:19)(cid:27) 0 − 6 δ It is also easy to see that Γ(x)′ W′ W′. Indeed, γ′(x) = i−1 k > 0 because ⊆ − 1 i δ − i δ q i−1 = n and γ′( ) = i−2 k and inequalities (1) imply γ′ W′ W′. (cid:20) δ (cid:21) i i ∞ −q − i ∈ − 1 i i−1 We therefore have to see that W′ Γ(x)′ W′. Let γ Γ satisfy ⊆ ∪ 1 ∈ γ( ) 1, γ(x) > 0. ∞ ≤ − As for Theorem 1.2, the conditions γ( ) < 0 and γ(x) > 0 imply that γ−1( ) or ∞ ∞ γ−1(0) is a convergent of x. Hence γ is equal to q +kq p kp q kq p +kp i−1 i − i−1 − i or − i−1 − i i−1 i , (cid:18) q p (cid:19) (cid:18) q p (cid:19) i i i i − − or q p q p i − i or i − i (cid:18) q kq p +kp (cid:19) (cid:18)q +kq p kp (cid:19) i−1 i i−1 i i−1 i i−1 i − − − − with k Z and i 0. ∈ ≥ q i−1 In the first case we have, when n 2, γ( ) = k 1 if and only 1 ≥ ∞ − q − ≤ − i if k 1; when n = 1, γ( ) 1 if and only if k 0 and k = 0 for i = 1, 1 ≥ ∞ ≤ − ≥ 6 6 δ i because q q . Moreover, γ(x) = k > 0 if and only if k n because i−1 i i+1 ≤ δ − ≤ i+1 δ i = n . So γ( ) 1 and γ(x) > 0 if and only if i+1 (cid:20)δ (cid:21) ∞ ≤ − i+1 q +kq p kp γ = i−1 i − i−1 − i (1 k n ) (cid:18) q p (cid:19) ≤ ≤ i+1 i i − or, when n = 1, 1 1 n γ = − 0 W′. (cid:18) 1 n +1(cid:19) ∈ 1 0 − q δ i−1 i In the second case, if γ( ) = + k 1, then k < 0 and γ(x) = + ∞ q ≤ − −δ i i+1 k < 0, so γ W′. 6∈ q q i i−1 In the third case we have γ( ) = 1 if and only if 0 < + ∞ −q +kq ≤ − q i−1 i i k 1, if and only if k = 0 and i 1, or k = 1 and i = 0. Moreover, γ(x) = ≤ ≥ δ i+1 > 0 if and only if k n . So γ( ) 1 and γ(x) > 0 if and only if i+1 δ kδ ≤ ∞ ≤ − i i+1 − q p γ = i − i (i 1), (cid:18) q p (cid:19) ≥ i−1 i−1 − or 1 n γ = − 0 W′. (cid:18) 1 1+n (cid:19) ∈ 1 0 − 7 q δ i i+1 Inthelastcase, ifγ( ) = 1, thenk < 0andγ(x) = < 0, ∞ q +kq ≤ − δ +kδ i−1 i i i+1 so γ W′. − 6∈ (cid:3) Remark 2.2 Theorem 2.1 proves a special case of a more general reduction con- jecture formulated by Zagier. For a Fuchsian group G, consider a partition of P1(R) into a finite set of intervals I , where α is a generator of G, together with a map α ρ : P1(R) P1(R) such that ρ = α. The map ρ can be viewed as the map → |Iα x ρ¯ : R2 R2 defined by (0,0) (0,0) and (x,y) = (0,0) α if (x : y) I . → 7→ 6 7→ (cid:18)y(cid:19) ∈ α Such an algorithm is a reduction algorithm if ρ¯(x,y) (x,y) for all (x,y) R2 k k ≤ k k ∈ with equality only if (x,y) = (0,0) or α is parabolic and (x : y) ∂I . Zagier α ∈ conjectured that, if ∆ is the diagonal x = y of R2, the map ρ¯ possesses a global attractor set D = ∞ ρ¯r(R2 ∆) with a finite rectangular structure, on which ∩r=0 − ρ¯ is essentially bijective and such that every point (x,y) of R2 ∆ is mapped to − D after finitely many iterations of ρ¯. Katok and Ugarcovici ([KU]) proved it for the special two-parameter family of maps ρ (suggested by Zagier) defined by the a,b transformations of SL(2,Z) x+1 if x < a 1 ρ (x) =  if a x < b a,b  −x ≤ x 1 if x b. − ≥   3 Main theorem For the classic continued fraction defined in Section 1, the following theorem is well known: Theorem 3.1 (Hurwitz) Twoirrationalnumbers x = [n ,n ,...]andy = [m ,m ,...] 0 1 0 1 are Γ-equivalent if and only if there exists s,t 0 such that n = m for all i 0. s+i t+i ≥ ≥ The proof of this classic theorem can be found in [HW]. As we said in the introduction, if we consider x and y belonging to a fixed irrational orbit of Γ, the indexes s and t in Theorem 3.1 are not bounded. In the next theorem, we show that there is a bound for s and t in terms of the matrix in Γ relating x and y. A Definition 3.2 For α = Q with B > 0 and (A,B) = 1, we define B ∈ log(√5min( A ,B)) M(α) := | | +2r +3, log((1+√5)/2) where r is the number of convergents of α. 8 Theorem 3.3 Let γ be an element of Γ. There exists N = N(γ) such that for all x R Q, if we write x = [n ,n ,...] and γ(x) = [m ,m ,...], then there exists 0 1 0 1 ∈ − 0 s,t N with n = m for all i 0. We can take N = 3 if γ( ) = and s+i t+i ≤ ≤ ≥ ∞ ∞ N = max(M(γ( )),M(γ−1( ))) otherwise. ∞ ∞ Proof. Denote by (x ) and (x′) the respective series defined in (1.1) for x i i≥0 i i≥0 and γ(x). Because x = γ (x) and x′ = γ (γ(x)), if there exists s,t 0 such s s,x t t,γ(x) ≥ that γ = γ γ, then x = x′. In this case, n = m and so γ = γ γ s,x t,γ(x) s t s t s+1,x t+1,γ(x) (by (1.3)). Again x = x′ and n = m . By induction we would have s+1 t+1 s+1 t+1 γ = γ γ and n = m for all i 0. s+i,x t+i,γ(x) s+i t+i ≥ To prove that such s,t smaller than N exist, we may bound by N the cardinal number of the sets γ Γ(γ(x))γ and γ Γ(x)γ−1 . { i,x 6∈ }i≥0 i,γ(x) 6∈ i≥0 (cid:8) (cid:9) 1 b If γ( ) = , then γ = ± with b Z. One easily sees from Theorem 1.2 ∞ ∞ (cid:18) 0 1(cid:19) ∈ that 1 0 Γ(x+b)Tb = Γ(x), γ Γ( x) − if γ ( ) 0,1 , i,x ∈ − (cid:18) 0 1(cid:19) i,x ∞ 6∈ { } which is the case for all i 3. So ≥ γ Γ(γ(x))γ < 3, γ Γ(x)γ−1 < 3. i,x i≥0 i,γ(x) i≥0 |{ 6∈ }| | 6∈ | (cid:8) (cid:9) Suppose γ( ) = ; we will prove the inequality ∞ 6 ∞ γ γ−1 Γ(γ(x)) M(γ−1( )). i,x 6∈ i≥0 ≤ ∞ (cid:12)(cid:8) (cid:9)(cid:12) Applying this inequali(cid:12)ty to γ−1 instead of(cid:12)γ and γ(x) instead of x, we will obtain γ γ Γ(x) M(γ( )) i,γ(x) 6∈ i≥0 ≤ ∞ (cid:12)(cid:8) (cid:9)(cid:12) and thus the theorem wi(cid:12)ll be proved. (cid:12) p Suppose there exists i 0 such that γ γ−1 Γ(γ(x)). We put γ−1( ) = i,x ≥ 6∈ ∞ q with (p,q) = 1. The description of Γ(γ(x)) given by Theorem 1.2 and the fact that γ γ−1(γ(x)) = γ (x) > 1 reduce the supposition above to one of the following i,x i,x conditions: p (i) γ 0, i,x q ≥ (cid:16) (cid:17) p (ii) γ 1. i,x q ≤ − (cid:16) (cid:17) For the case (i) we have p q p p q i−2 i−2 γ = − 0. i,x q q p+p q ≥ (cid:16) (cid:17) i−1 i−1 − 9 p p p p p i−2 i−1 i−2 i−1 This implies , and hence (Lemma 1.1) or is a convergent q ∈ |q q | q q i−2 i−1 i−2 i−1 p of . q For the case (ii) we have p q p p q i−2 i−2 γ = − 1. i,x q q p+p q ≤ − (cid:16) (cid:17) i−1 i−1 − p p p p p p p i−1 i−2 i−1 i−1 i−2 i−1 This implies − , and hence (Lemma 1.1) − or is q ∈ |q q q | q q q i−1 i−2 i−1 i−1 i−2 i−1 p − − a convergent of . q p p p i−1 i−2 If − is a convergent of , then we deduce from inequalities (1) and (2) q q q i−1 i−2 − and the recurrence satisfied by the convergents that p (n 1)p +p = p p p (i 4) i−3 i−1 i−2 i−3 i−1 i−2 | | ≤ | − | | − | ≤ | | ≥ q (n 1)q +q = q q q (i 2). i−3 i−1 i−2 i−3 i−1 i−2 ≤ − − ≤ ≥ We deduce from inequalities (3) and (4) log(√5 p ) log(√5q) i | | +3, i +3 (i 4). ≤ log((1+√5)/2) ≤ log((1+√5)/2) ≥ Finally we obtain γ γ−1 Γ(γ(x)) M(γ−1( )). i,x 6∈ i≥0 ≤ ∞ (cid:12)(cid:8) (cid:9)(cid:12) (cid:12) (cid:12) (cid:3) References [B] Bengoechea, P.: From continued fractions and quadratic functions to modular forms, In preparation. [HW] Hardy, G.H.; Wright, E.M.: An Introduction to the Theory of Numbers, 4th ed., Oxford: Clarendon Press (1965). [KU] Katok, S.; Ugarcovici, I.: Structure of attractors for (a,b)-continued fraction transformations, Journal of Modern Dynamics, 4 (2010), 637 - 691. 10

See more

The list of books you might like

Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.