of Mathematics, University Berkeley, California 94720 Communicated
of Caljfornio,
by H. Zassenhaus
Received April 5, 1989
The following equation
theorem is proved: If n 2 1.365 x lo’, then the Diophantine 2 - (n - 1) xzy - (n + 2)
has only the “trivial”
.y? - y’ = + 1
(*I
solutions (k LO), (0, * I), ( + 1. f 1).
Moreover, we show that for 0 3. then (*) has only trivial solutions. ((6 1990 Academic Press. Inc.
Since the mid-1960’s, with the work of Baker, one is frequently able to find all the solutions to a given Diophantine equation (of Thue type), at least if the coefficients are reasonable in size (see [ 1,4, 10, 11, 13, 161 for examples). On the other hand, it seems quite difficult to find all the solutions to a parametrized family of equations, since the data shift from one equation to the next. In this paper we consider an infinite family of Diophantine equations (the norm-form equation for the “simplest cubic?-see Shanks [12]) and show that for a sufficiently large value of the parameter one can find all the solutions. The equation we consider is x3-(n-l)x2y-(n+2).xy2-y3=
+_l,
n >, 0.
(*)
Note that for all n this equation has the solutions (0, f I),
(h 1, Oh
(+I,
Tl);
235 0022-314X/90 Ml!34,‘2-X
$3.00
Copyright ( 1990 by Academic Press. Inc. All rights of reproduction m any form reserved.
236
EMERYTHOMAS
we call these the trivial solutions.
For n = 0, 1, and 3 it is known that (*) solutions; see Section 3. Our result is
has non-trivial THEOREM
1. Let n be an integer with n 2 1.365 x 10’. Then the equation x3-(n-l)x*y-(n+2)xy*-y3=
+l,
n 2 0,
has only the trivial solutions given above.
In this section we outline a proof of the Theorem, with some details put over to the next section. In Section 3 we find all solution to (*) with n between 0 and 103. The cases n = 0, 1 are already known [2,8, 151, and so we now assume n 3 2. Suppose, then, that (p, q) is a non-trivial solution to (*); since ( -p, -9) is also a solution, we may assume that q > 0. Since n 3 2, one readily verifies that (*) has no non-trivial solution with q = 1; hence, we now assume q 2 2. For an integer n, set F,(x)=x3-(n-1)x*-(n+2)x-1,
11)
and denote the roots of F,,(x) = 0 by &, > 2, > R,. Remark 1. We regard the roots as functions of n and so sometimes will write A,(n), i= 0, 1, 2. Also, we usually will write simply R ( = A(n)) for b(n).
By Lagrange (see Remark 2 at end of section) we have: If (p, q) is a non-trivial solution to (*), then p/q is a principal convergent for precisely one of the roots &.
(2)
In this case we say that (p, q) has type i (i=O, 1,2). The family (*) has the following striking property: If (p, q) is a solution of type i, then (q, - (p + q)) is a solution with type i+ 1 (with i+ 1 taken mod 3).
(3)
We prove this in the next section. Thus, to prove Theorem 1 it suffices to show PROPOSITION
1. If n > 1.365 x 107, then (*) has no non-trivial solution of
type 0.
Suppose that (p, q) is a non-trivial solution of type 0. We will show that then one must have n < 1.365 x 107.
CUBIC
DIOPHANTINE
237
EQUATIONS
Let K, = Q(n), where F,,(i) = 0. As is well known, (p, q) is a solution to (*) o p - q2 is a unit in K,. Set i=o,
Yi= P-d;,
1,2.
(4)
Thus, (p,q) is a solution to (*)~y~y,~~= f.1. The field K, is Galois with action I, -+O i., --)O i.,, where &=A,
A, = - l/(J. + l),
i,,=
-(I
+ l/3,).
(5)
As is easily shown (e.g., [14]), A,, and iI generate the group of units for the order Z[n] in K,. Thus there are integers A, B (with the choice of signs below A and B will be positive) such that yi= o’(PQf),
i=o,
1,2.
(6)
(We will say that the integers A and B are associated to the solution (p, q).) We will need some estimates for I, q, and the ‘J~s. For n B 2, (a)