Solving Pell s equation using the nearest square continued fraction

Transcription

1 Solving Pell s equation using the nearest square continued fraction Keith Matthews 12th June 2008

2 Abstract This talk is about the nearest square continued fraction of A.A.K. Ayyangar (1941) and its use in finding the smallest positive solution of Pell s equation x 2 Dy 2 = ±1. Contrary to the 1944 review by D.H. Lehmer, its slight blemishes are indeed compensated by the its period length being about 70% of that of the regular continued fraction of D and also having a 3-case mid-point criterion for solving Pell s equation.

3 Hugh Williams and Peter Buhr (1979) gave a 6-case midpoint criterion in terms of Hurwitz continued fraction of the first kind. Their paper was rather complicated. It was after studying their paper that Jim White and John Robertson came up with a 3-case midpoint criterion using the nearest square continued fraction.

4 Euler (1765) gave a two-case midpoint criterion for solving Pell s equation x 2 Dy 2 = ±1 using the regular continued fraction (RCF) expansion of D. 1 D = a0 + 1 a 1 + a 2 + where a 0 = D and a i 1 for all i. We write D = [a0, a 1, a 2,...].

5 The n-th convergent is defined by A n /B n = a 0 + a a n a n A n and B n can be computed recursively: A 0 = a 0, B 0 = 1, A 1 = a 0 a 1 + 1, B 1 = a 1, for i 1. A i+1 = a i+1 A i + A i 1 B i+1 = a i+1 B i + B i 1,

6 The RCF for D is periodic with period-length k: D = { [a0, a 1,..., a h 1, a h 1,..., a 1, 2a 0 ] if k = 2h 1, [a 0, a 1,..., a h 1, a h, a h 1,..., a 1, 2a 0 ] if k = 2h. The smallest positive integer solution of x 2 Dy 2 = ±1 is given by (x, y) = (A k 1, B k 1 ), where A n /B n is the n-th convergent to D.

7 Euler observed that if k = 2h 1, while if k = 2h, A k 1 = A h 1 B h 1 + A h 2 B h 2 B k 1 = B 2 h 1 + B2 h 2, A k 1 = B h 1 (A h + A h 2 ) + ( 1) h B k 1 = B h 1 (B h + B h 2 ).

8 Also if ξ n = (P n + D)/Q n = [a n, a n+1,...] is the n-th complete quotient of the RCF expansion of D, then the equations Q h = Q h 1 if k = 2h 1, P h = P h+1 if k = 2h, enable us to determine h, as Q v = Q v+1 (k odd) and P v = P v+1, (k even), 1 v < k imply k = h.

9 The nearest square continued fraction (NSCF) of a quadratic surd ξ 0 was introduced by A.A.K. Ayyangar (AAK) in 1940 and based on the cyclic method of solving Pell s equation due to Bhaskara in The NSCF is a half-regular continued fraction. ie. ξ 0 = a 0 + where the a i are integers and a 1 + ɛ 1 ɛ 2 a 2 + a i 1, ɛ i = ±1 and a i + ɛ i+1 1 if i 1. We write the continued fraction as ξ 0 = a 0 + ɛ 1 a 1 + ɛ 2 a 2 +

10 Let ξ 0 = P+ D Q be a quadratic surd in standard form. ie. D is a non-square positive integer and P, Q 0, D P2 Q integers, having no common factor other than 1. Then with a = ξ 0, the integer part of ξ 0, we can represent ξ 0 in one of two forms (positive or negative representations) where P + D Q P + D Q = a + > 1 and P + D Q Q P + D = a + 1 Q P + D, > 1 are also standard surds. are

11 These equations imply (1) P = aq P; (2) P = (a + 1)Q P, (3) P 2 = R QQ ; (4) P 2 = R + QQ. These in turn imply (5) P P = Q and (6) P + P = Q + Q. We get P and Q from (1) and (3), then P and Q from (5) and (6), respectively.

12 AAK chose the partial denominator a 0 and numerator ɛ 1 of the new continued fraction development as follows: (a) a 0 = a if Q < Q, or Q = Q and Q < 0, (b) a 0 = a + 1 if Q > Q, or Q = Q and Q > 0. Also ɛ 1 = 1 and ξ 1 = P + D Q in case (a), ɛ 1 = 1 and ξ 1 = P + D Q in case (b).

13 Then ξ 0 = a 0 + ɛ 1 ξ 1 and ɛ 1 = ±1, a 0 is an integer and ξ 1 = P 1+ D Q 1 > 1. We proceed similarly with ξ 1 and so on: and ξ n = a n + ɛ n+1 ξ n+1 ξ 0 = a 0 + ɛ 1 a 1 + ɛ 2 a 2 + ξ n+1 is called the successor of ξ n.

14 Relations analogous to those for regular continued fractions hold for P n, Q n and a n, n 0: P n+1 + P n = a n Q n P 2 n+1 + ɛ n+1 Q n Q n+1 = D. The Q n successively diminish as long as Q n > D and so eventually, we have Q n < D. When this stage is reached, 0 < P i < 2 D and 0 < Q i < D for i n + 1. This implies eventual periodicity of the complete quotients and hence the partial quotients.

15 AAK defines ξ v to be a special surd if Q 2 v Q2 v D, Q 2 v Q2 v 1 D. A semi-reduced surd is the successor of a special surd. A reduced surd to the successor of a semi-reduced surd. Properties: 1. A semi-reduced surd is a special surd. 2. A quadratic surd has a purely periodic NSCF expansion if and only if it is reduced. 3. If ξ v is reduced, then P v > 0, Q v > 0 and a v 2. Examples: (i) p+q+ p 2 +q 2 p, p > 2q > 0, gcd(p, q) = 1, (ii) the successor of D.

16 The NSCF development of D has the form D = a0 + ɛ 1 a 1 + ɛ k, (1) 2a 0 where the asterisks denote a period of length k and ξ p = ξ p+k, ɛ p = ɛ p+k, a p = a p+k for p 1. Note: a 0 is the nearest integer to D. There are two types of NSCF expansions of D:

17 (I) No complete quotient of the cycle has the form p+q+ p 2 +q 2 p, where p > 2q > 0, gcd(p, q) = 1. This type possesses the classical symmetries of the regular continued fraction if k > 1: a v = a k v (1 v k 1) Q v = Q k v (1 v k 1) ɛ v = ɛ k+1 v (1 v k) P v = P k+1 v (1 v k). Note: If k = 2h + 1, then Q h = Q h+1. Conversely Q v = Q v+1, 1 v < k implies v = h. If k = 2h, then P h = P h+1. Conversely P v = P v+1, 1 v < k implies v = h.

18 Examples. 73 = (odd period) 19 = (even period) n = n + 1 2n (n 1), n 2 1 = n 1 2n (n > 1).

19 (II) There is one complete quotient ξ h in the cycle of the form p+q+ p 2 +q 2 p, where p > 2q > 0, gcd(p, q) = 1. In this case k 4 is even and h = k/2. This type also possesses the symmetries of Type I, apart from a central set of three unsymmetrical terms: a k 2 = 2, ɛ k 2 = 1, ɛ k 2 +1 = 1, a k 2 1 = a k D = a0 + ɛ 1 ɛ k a 1 a k a k ɛ k. 2a 0 For example 29 = Other examples are 53, 58, 85, 97.

20 E n is the number of D < 10 n of Type I with even period. O n is the number of D < 10 n of Type I with odd period. F n is the number of D < 10 n of Type II. N n is the number of D < 10 n. n E n O n F n N n

21 Note: P v P v+1, 1 v < k. ɛ h = 1, Q h 1 is even and P h = Q h Q h 1 (observed by John Robertson and Jim White). Conversely if ɛ v = 1, Q v 1 is even and P v = Q v Q v 1, 1 v < k, then D is of Type II and v = h.

22 For both types I and II, we have Q k = 1. For Hence P 1 = a 0. Then D = a0 + ɛ 1Q 1 P 1 + D = a 0 + ɛ 1Q 1 (P 1 D) P 2 1 D = a 0 P 1 + D. P 1 = P k (symmetry) P 1 = P k+1 (periodicity) 2a 0 = 2P 1 = P k + P k+1 = a k Q k = 2a 0 Q k. Note: ξ k = P k+ D Q k = a 0 + D.

23 A classical result for a half-regular expansion of ξ 0 = P 0+ D Q 0 A 2 n DB 2 n = ( 1) n+1 (ɛ 1 ɛ 2 ɛ n+1 )Q n+1 Q 0. In the special case ξ 0 = D, where Q 0 = 1 = Q k, we have A 2 k 1 DB2 k 1 = ( 1)k ɛ 1 ɛ 2 ɛ k. Also by periodicity, Q n = 1 if k divides n. is

24 Conversely, suppose Q n = 1, n 1. Then ξ n = P n + D. We prove P n = [ D] = a 0, the nearest integer to D. Then ξ n = a 0 + D = ξ k and k divides n.

25 We start with P 2 n + ɛ n Q n 1 Q n = D, noting that Q n 1 > 0, Q n > 0. Case 1. P n > D. Then ɛ n = 1. Hence P n = [ D]. Pn 2 D = Q n 1 < D (ξ n is reduced) 0 < P n D D D < P n + D < 2 D = 1 2.

26 Case 2. P n < D. Then ɛ n = 1. Q 2 n Q2 n D = P 2 n + Q n 1 (ξ n is reduced) (Q n )2 P 2 n Q n P n Q n 1 P n D Pn 2 = Q n 1 P n 0 < P n D P n < P n = 1 D + Pn 2P n 2. Again P n = [ D].

27 The convergents A kt 1 /B kt 1, t 1, in fact give all positive integer solutions of Pell s equation x 2 Dy 2 = ±1. For if x 2 Dy 2 = ±1, x > 0, y > 0, we can prove that x/y is an NSCF convergent to D, as follows. It is certainly an RCF convergent. We now introduce a transformation T 1 of Perron, which converts a half-regular continued fraction to an RCF:

28 To get the RCF partial quotients: Before a negative partial numerator, insert the term Replace each a n, n 0 by: (a) a n if ɛ n = +1, ɛ n+1 = +1, (b) a n 1 if ɛ n = +1, ɛ n+1 = -1, or ɛ n = -1, ɛ n+1 = +1, (c) a n 2 if ɛ n = -1, ɛ n+1 = -1. Here ɛ 0 = 1. Note: If ξ v and ξ v+1 are NSCF reduced quadratic surds and ɛ v = 1 and ɛ v+1 = 1, then a v 3. Hence T 1 produces a genuine RCF, ie. with no zero partial quotients.

29 For n 0, (i) ɛ n+1 = 1 gives rise to RCF convergents A m 1/B m 1 = (A n A n 1 )/(B n B n 1 ), A m/b m = A n /B n and RCF complete quotients P m + D Q m = ξ n+1 /(ξ n+1 1), P m+1 + D Q m+1 = ξ n+1 1. (ii) ɛ n+1 = 1 gives rise to RCF convergent A n /B n and RCF complete quotient ξ n+1.

30 It is not difficult to show that x/y does not have the form (A n A n 1 )/(B n B n 1 ) and hence x/y must also be an NSCF convergent. Remark. Arguing along these lines shows that the period length of the RCF expansion of D is k + r, where r is the number of ɛ n = 1 occurring in the period partial numerators ɛ 1,..., ɛ k of the NSCF expansion of D.

31 Example. D = 97. The NSCF expansion of 97 is of type II, with period-length 6. There are five ɛ i = 1 in the period range 1 i 6 and the period-length of the RCF expansion is 11. j i ξ i ξ j ɛ i a i a j A i /B i A j /B j /1 9/ / /7 59/ / /20 128/ / /33 325/ /86 522/ / / / / / / / / /

32 Exactly one of the following P, Q and PQ tests will apply for any D > 0, not a square: P-test: For some h, 1 h < k, P h = P h+1, in which case k = 2h and In this case A 2 k 1 DB2 k 1 = 1. A k 1 = A h B h 1 + ɛ h A h 1 B h 2 B k 1 = B h 1 (B h + ɛ h B h 2 ).

33 Q-test: For some h, 0 h < k, Q h = Q h+1, in which case k = 2h + 1 and A k 1 = A h B h + ɛ h+1 A h 1 B h 1 B k 1 = B 2 h + ɛ h+1b 2 h 1. In this case A 2 k 1 DB2 k 1 = ɛ h+1.

34 PQ-test: For some h, 1 h < k, Q h 1 is even, P h = Q h Q h 1 and ɛ h = 1, in which case k = 2h and A k 1 = A h B h 1 B h 2 (A h 1 A h 2 ) B k 1 = 2B 2 h 1 B hb h 2. In this case A 2 k 1 DB2 k 1 = 1.

35 The formulae for A k 1 and B k 1 depend on the following conservation identities which are proved using downward induction on t: (i) Let k = 2h + 1. Then for Type I and 0 t h, we have A 2h = A h+t B h t + ɛ h+1+t A h+t 1 B h t 1 B 2h = B h+t B h t + ɛ h+1+t B h+t 1 B h t 1 (ii) Let k = 2h. Then for Type I and 0 t h, or Type II with h 2 and 2 t h, we have A 2h 1 = A h+t 1 B h t + ɛ h+t A h+t 2 B h t 1 B 2h 1 = B h+t 1 B h t + ɛ h+t B h+t 2 B h t 1

36 Let π(d) and p(d) respectively denote the periods of the NSCF and RCF expansions of D, where D is not a perfect square and let Π(n) = π(d), P(n) = p(d). D n D n n Π(n) p(n) Π(n)/P(n)

37 There are grounds for believing that Π(n)/P(n) 1+ log ( 5 2 ) = log 2 For D with a long RCF period, we expect π(d)/p(d) to be near this value. For example, D = , Daniel Shanks (1974) and quoted by William Adams (1979). p(d) = , π(d) = , π(d)/p(d) = This D obeys the PQ-test.

38 AAK s paper and a L A TEX version are available at fractions.html BCMATH versions of NSCF and some other continued fraction algorithms are available at