On a distribution property of the residual order of \(a\pmod p\). II (Q5920510): Difference between revisions

The authors consider the multiplicative order of \(a(\text{ mod~}p)\), \(D_a(p)\), and are interested in the set of primes \(p\) such that \(D_a(p)\equiv j(\text{ mod~}k)\). In this note they announce various results and sketch some of the proofs. The work presented is a continuation of [\textit{K. Chinen} and \textit{L. Murata}, J. Number Theory 105, No. 1, 60--81 (2004; Zbl 1045.11066) and 82--100 (2004; Zbl 1045.11067)]. Under the assumption that \(a\geq 2\) is not of the form \(a_0^h\) with \(a_0\) an integer and \(h\geq 2\) they announce, under GRH (the Generalized Riemann Hypothesis), various interesting results. (For the rest of the review I assume GRH.) They announce that the latter set has a density \(\Delta_a(k,j)\) and that the values of \(\Delta_a(k,j)\) are effectively computable. Moreover, one should have that \(\Delta_a(q^r,j)=\Delta_a(q^{r-1},j)/q\) if \(q\) is an odd prime and \(r\geq 2\). They also give an explicit expression for \(\Delta_a(q,j)\) (Theorem 3). Furthermore, they explicitly work out \(\Delta_5(12,j)\) in Theorem 5. Except for the last example, these results have been established independently and in greater generality by the reviewer in [``On the distribution of the order and index of \(g\pmod p\) over residue classes. II'', Preprint, \url{arXiv:0404339}, J. Number Theory (to appear), see \url{doi:10.1016/j.jnt.2005.06.006}]. (This preprint was posted April 19, 2004 at the ArXiv, whereas the present note bears the communication date Nov. 12, 2004.) The authors express the density \(\Delta_a(k,j)\) as a six fold sum, the reviewer, however, expressed it as a two fold sum. The reviewer showed more generally, that for any rational number \(a\) the set of primes \(p\) such that \(D_a(p)\equiv j(\text{ mod~}k)\) and \(p\equiv b(\text{ mod~}f)\) has a density that can be expressed as a two fold sum involving Galois theoretic intersection coefficients. Moreover, he showed that if \(q| d_1\) is an odd prime, then \(\Delta_a(qd_1,j)=\Delta_a(d_1,j)/q\). Unfortunately, Theorem 3 and Theorem 5 are inconsistent with each other (a priori one should have that \(\sum_{j=0}^{3}\Delta_5(12,1+3j)=\Delta_5(3,1)\), an equality that is not satisfied if one believes both Theorem 3 and Theorem 5). In fact, it turns out that the densities in Theorem 5 for \(j=4\) and \(j=8\) are incorrect. Let \(a\) satisfy the conditions assumed by the authors and \(D\) denote the discriminant of the quadratic number field obtained by adjoining \(\sqrt{a}\) to the rationals. If \(D\) and \(6\) are coprime it can be shown that the intersection coefficients are all equal to one. This then implies that \(\Delta_a(12,b)=\Delta_a(12,b+6)\), a property not satisfied by the densities the authors propose. In fact, a computation shows that \(\Delta_5(12,4)=5/48+C109/1880\), rather than \(1/6-5C/376\) and that \(\Delta_5(12,8)=5/48-C109/1880\), rather than \(1/24+5C/376\). Here \(C\) is a constant given by an infinite Euler product, which on direct computation gives \(C\approx 0.86989\) (as stated in this note). By expressing this constant as product of special values of \(L\)-series one can calculate it up to higher accuracy resulting in \(C\approx 0.86988561214817\ldots\), see [\textit{P. Moree}, Finite Fields Appl. 10, No.3, 438--463 (2004; Zbl 1061.11050)]. The latter densities are numerically rather close to the ones proposed by the authors, but distinctly in better accord with the numerical data. The outline of the proofs given are too sketchy for the reviewer to be able to find the mistake in the purported proof of Theorem 5. Part I, cf. ibid. Proc. Japan Acad., Ser. A 79, No. 2, 28--32 (2003; Zbl 1071.11054).