On a distribution property of the residual order of \(a\pmod p\). II (Q5893873)

(For notation, citations [R1--R4] and comments on Theorem 1.1 see the review of part I. (Zbl 1045.11066) above). Let \(a>0\) be an integer that is not the \(h\)th power of an integer with \(h\geq 2\). In part I the authors proved, under GRH, the existence of the density \(\delta_a(k)\) of primes \(p\) such that the order of \(a(\text{mod~}p)\) is congruent to \(k(\text{mod~}4)\). The case where \(k\) is even has been well studied. So let us restrict ourselves in the sequel to the case where \(k\) is odd. Then the authors' expression for \(\delta_a(k)\) is a sum of two four fold sums involving Galois theoretic coefficients. In part I the parameter \(a\) was further restricted, which allowed the authors to establish that \(\delta_a(1)=\delta_a(3)\) in that case, without a need for explicitly evaluating their four fold sums. In the present paper the Galois theoretic coefficients are evaluated and \(\delta_a(k)\) is explicitly computed (Theorem 1.2). This is quite a tour de force and not surprisingly does not make for easy reading. It turns out that, despite the arithmetic complexity of these densities, they always satisfy the inequality \(\delta_a(1)\leq \delta_a(3)\). The authors find that \(\delta_a(k)=b_1+b_2C\), with \(b_1\) and \(b_2\) rational numbers (depending possibly on \(a\)) and \(C\) a constant which can be written as an Euler product (singular series) and give the `rough approximative value' \(C\approx 0.64365\). In [R4] it is shown that actually \(C=0.64365067966\dots\) and that \(C\) can be expressed in terms of special values of Dirichlet \(L\)-series (allowing \(C\) to be evaluated with high decimal precision). From Theorem 1.2 one easily infers, on GRH, that \(\delta_a(k)=1/6\) for almost all \(a\). It turns out that \(1/6\) is the \(a\)-average of \(\delta_a(k)\). Another interpretation of the \(1/6\) is that it is the average density of elements of order congruent to \(k(\text{mod~}4)\) in a finite field of prime cardinality. If the modulus \(4\) is replaced by an arbitrary one, analogous results turn out to be true [R3]. Since this subject is so tied up with the arithmetic of Kummer extensions, quadratic fields and cyclotomic fields, it is puzzling that the authors do not work with the notion of the discriminant of a quadratic field. This would lead to a considerable shortening of statements of several of the results, cf. the formulation of Corollary 1 of [R1] with that of Theorem 1.2. Such a shortening was first pointed out to the authors by Don Zagier. It is a 19th century fact that the conductor of a quadratic field equals the absolute value of its discriminant. For this the authors, surprisingly enough, refer to a preprint of the reviewer. A preprint that, moreover, has appeared long ago: the reviewer [Acta Arith. 89, 9--21 (1999; Zbl 0926.11003)]. In [R1] the reviewer expressed \(\delta_a(k)\) for arbitrary rational numbers \(a\) as a sum of two single summand (as opposed to four fold) sums. This leads, as can be imagined, to a rather more compact treatment of the problem considered by the authors (and greater generality). Moreover, in [R2] the reviewer showed that if \(4\) is replaced by an arbitrary natural number, under GRH, the corresponding density exists and can be expressed as a double sum involving Galois theoretic coefficients. The arithmetic behaviour of \(Q_a(x;k,l)\) is quite complicated, so it is to the merit of the authors that they did not shy away from this thorny problem. Indeed, in various Japanese publications dating back to 2000 they advocated the study of this problem and reported on their progress leading up to the present two papers.