Lucas Sequences Whose 12th or 9th Term Is A Square: A. Bremner N. Tzanakis
Lucas Sequences Whose 12th or 9th Term Is A Square: A. Bremner N. Tzanakis
Lucas Sequences Whose 12th or 9th Term Is A Square: A. Bremner N. Tzanakis
N. Tzanakis
Introduction
Let P and Q be non-zero relatively prime integers. The Lucas sequence {Un (P, Q)}
is defined by
U0 = 0,
U1 = 1,
Un = P Un1 QUn2
(n 2).
(1)
The sequence {Un (1, 1)} is the familiar Fibonacci sequence, and it was proved
by Cohn [12] in 1964 that the only perfect square greater than 1 in this sequence
is U12 = 144. The question arises, for which parameters P , Q, can Un (P, Q)
be a perfect square? This has been studied by several authors: see for example
Cohn [13] [14] [15], Ljunggren [22], and Robbins [25]. Using Bakers method on
linear forms in logarithms, work of Shorey & Tijdeman [26] implies that there
can only be finitely many squares in the sequence {Un (P, Q)}. Ribenboim and
McDaniel [23] with only elementary methods show that when P and Q are odd,
and P 2 4Q > 0, then Un can be square only for n = 0, 1, 2, 3, 6 or 12; and that
there are at most two indices greater than 1 for which Un can be square. They
characterize fully the instances when Un = , for n = 2, 3, 6, and observe that
U12 = if and only if there is a solution to the Diophantine system
P = , P 2 Q = 2, P 2 2Q = 3, P 2 3Q = , (P 2 2Q)2 3Q2 = 6. (2)
When P is even, a later paper of Ribenboim and McDaniel [24] proves that if
Q 1 (mod 4), then Un (P, Q) = for n > 0 only if n is a square or twice a
square, and all prime factors of n divide P 2 4Q. Further, if p2t |n for a prime p,
then Up2u is square for u = 1, . . . , t. In addition, if n is even, then Un = only
if P is a square or twice a square. A remark is made that no example is known
Department of Mathematics, Arizona State University, Tempe AZ, USA, e-mail: bremner@asu.edu , http://math.la.asu.edu/andrew/bremner.html
of an integer pair P , Q, and an odd prime p, such that Up2 = (note none can
exist for P , Q odd, P 2 4Q > 0).
In this paper, we complete the results of Ribenboim and MacDaniel [23] by
determining all Lucas sequences {Un (P, Q)} with U12 = (in fact, the result
is extended, because we do not need the restrictions that P , Q be odd, and
P 2 4Q > 0): it turns out that the Fibonacci sequence provides the only example. Moreover, we also determine all Lucas sequences {Un (P, Q)} with U9 = ,
subject only to the restriction that (P, Q) = 1. Throughout this paper the symbol
means square of a non-zero rational number.
Theorem 1. Let (P, Q) be any pair of relatively prime non-zero integers. Then,
U12 (P, Q) = iff (P, Q) = (1, 1) (corresponding to the Fibonacci sequence).
U9 (P, Q) = iff (P, Q) = (2, 1) (corresponding to the sequences Un = n
and Un = (1)n+1 n).
The remainder of the paper is devoted mainly to the proof of this theorem.
Theorems 3 and 6 of [23] combined with the first statement of Theorem 1 imply
the following.
Theorem 2. Let P, Q be relatively prime odd integers, such that P 2 4Q > 0.
Then the nth term, n > 1, of the Lucas sequence Un = Un (P, Q) can be a square
only if n = 2, 3, 6 or 12. More precisely1 :
U2 = iff P = a2 .
U3 = iff P = a, Q = a2 b2 .
U6 = iff P = 3a2 b2 , Q =
a8 + 12a4 b4 9b8
.
2
U12 = iff (P, Q) = (1, 1). Moreover, this result is also valid even if we
remove all restrictions on P, Q except for gcd(P, Q) = 1.
The proof of Theorem 1 hinges, in both cases, upon finding all rational points
on a curve of genus 2. When the rank of the Jacobian of such a curve is less than
2, then methods of Chabauty [11], as expounded subsequently by Coleman [16],
Cassels and Flynn [10] and Flynn [17] may be used to determine the (finitely
many) rational points on the curve. When the rank of the Jacobian is at least 2,
as is the case here, a direct application of these methods fails. In order to deal
with such situations, very interesting methods have been developed recently by a
number of authors; see chapter 1 of Wetherells Ph.D. thesis [31], Bruin [2, 3, 4],
Bruin and Flynn [5, 6], Flynn [18], and Flynn & Wetherell [19, 20]. For the
purpose of this paper, the method of [18] or [19] is sufficient.
1
Below it is understood that parameters a, b are in every case chosen so that P, Q are odd,
relatively prime and P 2 4Q > 0.
2
2.1
(3)
(P 2 2Q)(P 4 4P 2 Q + Q2 ) = ,
(1 2x)(1 4x + x2 ) = ,
and of these four elliptic curves, only the curve with = 2 has positive rational
rank. Torsion points on the three other curves do not provide any solutions for
P , Q. We are thus reduced to considering the equations
P (P 2 3Q)(P 2 Q) = 2,
P (P 2 3Q) = 2, P 2 Q = ,
(P 2 2Q)(P 4 4P 2 Q + Q2 ) = 2.
or P (P 2 3Q) = , P 2 Q = 2.
P = P 2 3Q = 2
P = P 2 3Q = 2
P = 2 P 2 3Q =
P = 2 P 2 3Q =
where = 1, 3.
The latter case implies one of
P2 Q =
P 2 Q =
P2 Q =
P 2 Q =
P = P 2 3Q = P 2 Q = 2
P = P 2 3Q = P 2 Q = 2
(4)
(5)
where = 1, 3.
Solvability in R or elementary congruences shows impossibility of the above equations (4), (5), except in the following instances:
P = ,
P = 3,
P = 6,
P = ,
P = 6,
P = ,
P = 3,
P 2 3Q = 2,
P 2 3Q = 6,
P 2 3Q = 3,
P 2 3Q = 2,
P 2 3Q = 3,
P 2 3Q = ,
P 2 3Q = 3,
3
P2 Q =
P2 Q =
P2 Q =
P 2 Q =
P 2 Q =
P 2 Q = 2
P 2 Q = 2.
(6)
(7)
2
6
3
2
3
6
Now the rational ranks of the following elliptic curves are 0:
(x+1)(x2 4x+1) = 2,
(3x+1)(x2 4x+1) = 3,
(x+1)(x2 4x+1) = 2,
and consequently the rational points on the curves corresponding to the first
three rows of the above table are straightforward to determine: they are (P, Q) =
(1, 1), (0, 1), and (1, 1) respectively. These lead to degenerate Lucas sequences
with U12 = 0.
It remains only to find all rational points on the following curve:
P = , P 2 3Q = , P 2 Q = 2, P 2 2Q = 3, P 4 4P 2 Q + Q2 = 6 ,
satisfying (P, Q) = 1. Note that this is the curve (2), though we have removed
the restriction that P and Q be odd, and P 2 4Q > 0.
Put Q/P 2 = 1 2u2 , so that
3u2 1 = 2,
4u2 1 = 3,
2u4 + 2u2 1 = 3.
(8)
The equations (8) define a curve of genus 9, with certainly only finitely many
points. We restrict attention to the curve of genus 2 defined by
4u2 1 = 3,
2u4 + 2u2 1 = 3.
4
Define
K
=
Q(
3),
with
ring
of
integers
O
=
Z[
3], and fundamental unit
K
u2 1 + u2 3 = 3 2 ,
u2 (4u2 1)(u2 (1 + 3) 1) = 3 3V 2 , V K.
E1 : y 2 = x(x (3 + 3))(x 4 3)
(9)
2.2
The case U9
(10)
So
P 2 Q = ,
P 6 6P 4 Q + 9P 2 Q2 Q3 = ,
For = 1, the first curve has rational rank 0, and torsion points do not lead
to non-zero solutions for P , Q. For = 3, both curves at (12) have rational
rank 1, so that the rank of the Jacobian of (11) equals 2. To solve the equation
U9 (P, Q) = , it is necessary to determine all integer points on the two curves
and
P 6 9P 4 R2 + 18P 2 R4 + 9R6 = ,
P 6 9P 4 R2 18P 2R4 + 9R6 = .
(13)
(14)
3
2.2.1
Equation (13) may be written in the form
NormL/Q (P 2 + (5 + + 2 )R2 ) = S 2 , say,
and it follows that
P 2 + (5 + + 2 )R2 = U 2 ,
(15)
(16)
Suppose P is a first degree prime ideal of OL dividing (). Then for the norm of
to be a square, must also be divisible by one of the conjugate prime ideals of
P. It follows that P, or one of its conjugates, divides both and . Then this
prime will divide ((5 + + 2 ) (5 2 + 2 )) = (3) = (1 )3 . So P has
to be (1 ), with (1 )2 dividing , contradicting squarefree. If the residual
degree of P is 3, then the norm of cannot be square. Finally, the residual degree
of P cannot be 2, otherwise = 5 2 m (mod P), for some rational
integer m, so that = 1 2 + 31 2 is congruent to a rational integer modulo P,
impossible. In consequence, is forced to be a unit, of norm +1. Without loss
of generality, the only possibilities are = 1, 1 , 2 , 1 2 . However, specializing
the left hand side of (15) at the root 0 = 1.8793852415... of x3 3x1 = 0 shows
that P 2 + 0.4114..R2 = (0 )U(0 )2 , so that (0 ) > 0, giving unsolvability of
(15) for = 2 , 1 2 . There remain the two cases = 1, with solution
(P, R, U) = (1, 0, 1), and = 1 , with solution (P, R, U) = (0, 1, 4 2 ). From
(15) and (16) we now have
P 2 (P 2 + (5 + + 2 )R2 )(P 2 + (5 2 + 2 )R2 ) = W 2 ,
with = = 1 or 1 + 2 . Accordingly, X = P 2/R2 gives a point on the
elliptic curve:
X(X + (5 + + 2 ))(X + (5 2 + 2 )) = Y 2 .
(17)
2
P2
W
When = 1 + 2 , then (x, y) = ( (1)
2 R2 , (1)3 R3 ) is a point on the elliptic
curve E2 over Q(), where
E2 : y 2 = x(x + (2 + 2 ))(x + (1 + + 2 )),
2
(18)
x = (4 + 22 )x Q2 .
satisfying (1)
2.2.2
Equation (14) may be written in the form
NormL/Q (P 2 + (5 + + 2 )R2 ) = S 2 , say,
so that
P 2 + (5 + + 2 )R2 = U 2 ,
(19)
(1+2 ) P 2
(1)2 R2
P 2 + (5 2 + 2 )R2 = (2 + 2 )V 2 ,
(20)
(1)
2
2
satisfying (1+
2 ) x = (4 + 2 )x Q . However, a straightforward calculation shows that the Q()-rank of (20) is equal to 0, with torsion group the
obvious group of order 4. There are no corresponding solutions for P , Q.
Here we justify our assertions about the elliptic curves E1 at (9) and E2 at (18).
These curves are defined over fields F (where F =K or L, respectively) with
unique factorization, and have F -rational two-torsion. So standard two-descents
over F work analogously to the standard two-descent over Q for an elliptic curve
with rational two-torsion; see for example Silverman [28], Chapter 10.4. It is thus
straightforward to determine generators for Ei (F )/2Ei (F ), i = 1, 2 (and in fact
software packages such as that of Simon [30] written for Pari-GP, and ALGAE [7]
of Bruin written for KASH, with m-ALGAE [8] for MAGMA, also
perform
this
calculation effortlessly). Such generators are the classes of P1 = ( 3, 3 3) for
the curve E1 , and P2 = (1, ) for the
curve E2 . In fact P1 and P2 are generators
for the Mordell-Weil groups E1 (Q( 3)) and E2 (Q()) respectively. To show this
necessitates detailed height calculations over the appropriate number field, with
is the canonical
careful estimates for the difference h(Q)
21 h(Q) where h(Q)
height of the point Q, and h(Q) the logarithmic height. The KASH/TECC
package of Kida [21] was useful here in confirming calculations. The standard
Silverman bounds [29] are numerically too crude for our purposes, so recourse
was made to the refinements of Siksek [27]. Full details of the argument are given
in Section 3 of [1]. Actually, determination of the full F -rational Mordell-Weil
7
4.1
The problems to which we were led in section 2 are of the following shape.
Problem: Let
E : y 2 + a1 xy + a3 y = x3 + a2 x2 + a4 x + a6 ,
(21)
There is a group isomorphism between E and the subgroup of E(Qp ()) consisting of those points Q whose reduction modp is the zero point of the reduced
modp curve, defined by
z 7 Q, where Q = (z/w(z) , 1/w(z)) if z 6= 0 and Q = O if z = 0,
with w(z) a p-adically convergent power series, that can be explicitly calculated
up to any p-adic precision. The inverse map is given by z(O) = 0 and for
Q E(Qp ()) different from O whose reduction modp is zero, z(Q) = x(Q)
.
y(Q)
The remarkable property relating the functions z and F is that, for any points
Q1 , Q2 as Q above,
z(Q1 + Q2 ) = F (z(Q1 ), z(Q2 )) .
(22)
With respect to E, a logarithmic function log is defined on pZp [] and an exponential function exp is defined on pr Zp [], where r = 1 if p > 2 and r = 2 if p = 2.
These functions are mutually inverse and can be explicitly calculated as p-adic
power series up to any precision. Moreover, if r is as above and z1 , z2 pr Zp [],
then
log F (z1 , z2 ) = log z1 + log z2 and exp(z1 + z2 ) = F (exp z1 , exp z2 ).
(23)
Suppose now we know a point such that z(Q) pr Zp [] and assume further that,
for a certain specifically known point P Q()E(Zp []), or for P = O, we want
to find all n Z for which x(P +nQ) is a rational number (or, more particularly,
a square of a rational). According to whether P is a finite point or P = O, we
express x(P + nQ) or 1/x(nQ) first as an element of Zp [][[z(nQ)]] and then,
using properties (22) and (23), as a sum 0 (n) + 1 (n) + + d1 (n)d1 , where
each series i (n) is a p-adically convergent power series in n with coefficients in
Zp , which can be explicitly calculated up to any desired p-adic precision. In order
that this sum be a rational number we must have i (n) = 0 for i = 1, . . . , d 1.
At this point we use Strassmans Theorem3 , which restricts the number of padic integer solutions n. If the maximum number of solutions implied by this
theorem is equal to the number of solutions that we actually know, then we have
explicitly all solutions. Sometimes, as in some instances of the sections below,
even Strassmans Theorem is not necessary.
In the following two sections we apply the above method to equations (9) and
(18). At the suggestion of the referee, we give only a few computational details;
most of our computational results, including the explicit form of the functions
F (z1 , z2 ), w(z), log and exp with the required precision, can be found in section
5.1 of [1].
4.2
Equation (9)
4.3
Equation (18)
2 , 0) , (0, 0) , (2 + 2 , 0)}. For the point Q = 4P1 we have z(Q) 4Z2 [] and
we write any point on E(Q()) in the form n1 P1 + T = (4n + r)P1 + T = nQ + P ,
with P = rP1 + T , r {1, 0, 1, 2} and T a torsion point as above. Therefore,
there are 16 possibilities for P , one of which is P = O.
Working as in section 4.2 we check that the only solutions (x, y) such that
x Q2 are (x, y) = (0, 0), 2P1 + (0, 0), 2P1 + (0, 0). To give an idea of how
we apply Strassmans Theorem, let us consider the instance when P = 2P1 + T
with T = (0, 0). We compute 1 (n) = 26 7n + 26 3n2 + 0 n3 + 0 n4 +
0 n5 + 27 n6 ( ). By Strassmans Theorem, 1 (n) = 0 can have at most two
solutions in 2-adic integers n. On the other hand, a straightforward computation
shows that x(P + 0 Q) = 4 = x(P Q), which implies, in particular, that
1 (0) = 0 = 1 (1). Hence, n = 0, 1 are the only solutions obtained for the
above specific value of P , leading to the points (x, y) = P + 0 Q = 2P1 + (0, 0)
and (x, y) = P + (1)Q = 2P1 + (0, 0), both having x = 34 (1 2 ) and x = 4.
Conclusion. The only points on (18) satisfying x Q2 are those with
x = 0 (leading to P = 0), and x = 34 (1 2 ) giving successively (in the notation
2
of section 2.2.1) PR2 = 4 and (P, Q) = (2, 1), corresponding to degenerate Lucas
sequences.
Acknowledgement
We are grateful to the referee for suggestions which have greatly improved the
presentation of this paper.
References
[1] A. Bremner and N. Tzanakis, Lucas sequences whose 12th or 9th term
is a square. Extended version.
http://www.math.uoc.gr/~tzanakis/Papers/Fibonacci 9 12.pdf.
[2] N. Bruin, Chabauty methods and covering techniques applied to generalized
Fermat equations, CWI Tract, vol. 133, Stichting Mathematisch Centrum
voor Wiskunde en Informatica, Amsterdam, 2002, Dissertation, University
of Leiden, Leiden 1999.
[3] N. Bruin, The diophantine equations x2 y 4 = z 6 and x2 + y 8 = z 3 ,
Compositio Math. 118 (1999), 305-321.
[4] N. Bruin, Chabauty methods using elliptic curves, J. reine angew. Math.
562 (2003), 27-49.
[5] N. Bruin and E.V. Flynn, Towers of 2-covers of hyperelliptic curves,
Pacific Institute Math. Sci., preprint PIMS-01-12 (2001).
11
[6] N. Bruin and E.V. Flynn, N-covers of hyperelliptic curves, Math. Proc.
Camb. Phil. Soc. 134 (2003), 397-405.
[7] N. Bruin, http://www.cecm.sfu.ca/~bruin/ell.shar
[8] N. Bruin, http://www.cecm.sfu.ca/~bruin/malgae.tgz
[9] J.W.S. Cassels, Local Fields, LMS Student Texts 3, Cambridge University Press, Cambridge and London 1986.
[10] J.W.S. Cassels and E.V. Flynn, Prolegomena to a Middlebrow Arithmetic of Curves of Genus 2 LMS Lecture Note Series 230, Cambridge
University Press, Cambridge 1996.
[11] C. Chabauty, Sur les points rationnels des courbes algebriques de genre
superieur `a lunite, C. R. Acad. Sci. Paris 212 (1941), 882-885.
[12] J.H.E. Cohn, On square Fibonacci numbers, J. London Math. Soc. 39
(1964), 537-541.
[13] J.H.E. Cohn, Eight Diophantine equations, Proc. London Math. Soc. 16
(1966), 153-166.
[14] J.H.E. Cohn, Five Diophantine equations, Math. Scand. 21 (1967), 61-70.
[15] J.H.E. Cohn, Squares in some recurrent sequences, Pacific J. Math. 41
(1972), 631-646.
[16] R.F. Coleman, Effective Chabauty, Duke Math. J. 52 (1985), 765-770 .
[17] E.V. Flynn, A flexible method for applying Chabautys Theorem, Compositio Math. 105 (1997), 79-94.
[18] E.V. Flynn, On Q-Derived Polynomials, Proc. Edinburgh Math. Soc. 44
(2001), 103-110.
[19] E.V. Flynn and J.L. Wetherell, Finding rational points on bielliptic
genus 2 curves, Manuscripta Math. 100 (1999), 519-533.
[20] E.V. Flynn and J.L. Wetherell, Covering Collections and a Challenge
Problem of Serre, Acta Arithm. 98 (2001), 197-205.
[21] M. Kida, TECC manual version 2.4, The University of ElectroCommunications, September 2000.
[22] W. Ljunggren, New propositions about the indeterminate equation
(xn 1)/(x 1) = y q , Norske Mat. Tidsskrift, 25, 1943, 17-20.
12
[23] P. Ribenboim and W.L. McDaniel, The square terms in Lucas sequences, J. Number Theory, 58, 1996, 104-123.
[24] P. Ribenboim and W.L. McDaniel, Squares in Lucas sequnces having
an even first parameter, Colloq. Math., 78, 1998, 29-34.
[25] N. Robbins, On Pell numbers of the form P X 2 , where P is prime, Fibonacci Quart. (4), 1984, 340-348.
[26] T.N. Shorey and R. Tijdeman, Exponential Diophantine Equations,
Cambridge Univ. Press, Cambridge, 1986.
[27] S. Siksek, Infinite descent on elliptic curves, Rocky Mountain J. Math.,
Vol. 25, No. 4, 1995, 1501-1538.
[28] J.H. Silverman, The Arithmetic of Elliptic Curves, Graduate Texts in
Math., vol. 106, Springer-Verlag, Berlin and New York, 1986.
[29] J.H. Silverman, The difference between the Weil height and the canonical
height on elliptic curves, Math. Comp. 55 (1990), no. 192, 723-743.
[30] D. Simon, http://www.math.unicaen.fr/~simon/ell.gp
[31] J.L. Wetherell, Bounding the number of rational points on certain
curves of high rank, Ph.D. Dissertation, University of California at Berkeley, 1997.
13