Strong arithmetic properties of the integral solutions of X³ + DY³ + D²Z³ - 3DXYZ = 1, where D = M³ ± 1, M ∈ ℤ* (Q4260433)

This paper is concerned with prime divisors of linear recurrence sequences. Let \(X= \{x_i\}_{i=0}^\infty\) be a sequence of integers that satisfies a linear recurrence, that is there exists an integer \(k\) and \(a_1,\dots, a_k\), such that \(x_{n+k}= a_1x_{n+k-1}+ a_2x_{n+k-2}+\cdots+ a_kx_n\) for every \(n\geq 0\). The smallest \(k\) that satisfies the above requirements is said to be the order of the recurrence. A prime \(p\) is said to divide \(X\), if it divides at least, one term of the sequence. The prime \(p\) is said to be a maximal divisor of a \(k\)th order recurrence if there are \(k\) consecutive terms in the sequence such that \(p\) divides the first \(k-1\) terms, but not the \(k\)th one. Laxton has defined an equivalence relation on linear recurrences of order two and a composition yielding a group. All recurrences in the same equivalence class have the same density of prime divisors. If a recurrence is in an equivalence class of finite order in this group, it is said to be torsion. The Lucas sequences, for example, are torsion. Hasse discovered an unconditional method to compute the density of primes dividing a second order torsion recurrence having a reducible (over \(\mathbb{Q}\)) characteristic polynomial. Lagarias showed the method to apply to some second order torsion recurrences having irreducible characteristic polynomial. In a recent preprint, extending earlier work by the reviewer and \textit{P. Stevenhagen} [Acta Arith. 82, 403-410 (1997; Zbl 0913.11048)], Stevenhagen sealed this line of research by showing unconditionally that every second order torsion recursion has a positive rational density of prime divisors. Thus presently the prime division theory of second order recurrences is relatively well understood, but unfortunately does not generalize well to higher order recurrences (cf. MI preprint 2000-03 by \textit{H. Roskam}, \url{http://www.math.leidenuniv.nl/reports/2000-03.shtml}, see also J. Number Theory 81, No. 1, 93-109 (2000). However, after putting together bits of work of Lucas, Hasse, Laxton and Lagarias concerning the prime division theory of recurrences of order two, the author constructed a group structure for higher order recurrences through which much of the second order theory can be retained [Mem. Am. Math. Soc. 551 (1995; Zbl 0827.11006)]. In this generalization, one replaces prime division by prime maximal division. In this same memoir, the author shows that some recurrences that are torsion in this group, which he calls Laxton-Ballot group, have a density of maximal prime divisors that can be computed unconditionally. But he does so only for recurrences having rational roots. In the paper under review the author studies certain third order recurrences (which he calls Williams-Ballot sequences) linked to the pure cubic field \(\mathbb{Q} (\root 3\of {D})\). These sequences were shown by H. Williams to retain many properties that ordinary Lucas sequences have. The author shows that for discriminants \(D= M^3+e\), where \(e=\pm 1\), these sequences have more to offer: besides having irrational roots, they are torsion in the Laxton-Ballot group and have computable densities of maximal divisors. I seize the opportunity of this review to point out an error first noticed by Roskam and add some comments. As often in a density computation for a torsion sequence, certain degrees of Kummerian extensions have to be computed. The generic value of these degrees is one thing to compute, but usually there are a few exceptional cases where certain coincidences occur that result in a degree that is less than the generic value. Various papers in this area contain some errors due to mistakes in these degree computations. Thus, for example the main result of \textit{P. J. Stephens} [J. Number Theory 8, 313-332 (1976; Zbl 0334.10018)] on the density of prime divisors of sequences of the form \(\{a^n- b\}_{n=0}^\infty\) with \(a\) and \(b\) integers is false due to such a mistake. For a correction and generalization of his result see the recent preprint `A two variable Artin conjecture' by the reviewer and P. Stevenhagen (MI preprint 1999-30). In the present paper some degree computations are contained in Lemma 4.1 which are not correct in all cases. In case \(M=1\) or \(M=19\) and \(e=1\), then for \(j>k\) the degree is lower by a factor three. This, by a minor modification of the author's proofs, then leads to different densities; namely \(d(D_3)= 1/78\) rather than \(d(D_3)= 3/26\) (in Theorem 4.6) and \(d(D_1)= 1/8\) rather than \(d(D_1)= 3/8\) (in Theorem 4.4). However, the reviewer would not be surprised if there are only finitely many \(M\) for which the densities computed in the paper are wrong. In the above, natural density is meant, when the word density is used. The author uses the more restrictive notion of Dirichlet density. Actually it is not difficult to prove his results for natural density and even with an error term, following arguments due to \textit{R. W. K. Odoni} [J. Number Theory 13, 303-319 (1981; Zbl 0471.10031)]. As in his (well-written) memoir (loc. cit.), the author brushes over the fact that if \(\{S_i\}_{i=1}^\infty\) is an infinite family of disjoint sets each having a Dirichlet density, it might not be true that their union has Dirichlet density \(\sum_{i=1}^\infty d(S_i)\) (for example when \(S_i\) is the set containing only the \(i\)th rational prime).