On a distribution property of the residual order of \(a\pmod p\). I (Q597126)

Denote by \(D_a(p)\) the order (or residual order) of a number \(a\) modulo a prime \(p\), and let \(I_a(p)\) denote the corresponding index, so that \(I_a(p)D_a(p)=p-1\). Let \(N_a(x,n)\) be the set of primes \(p\leq x\) with \(I_a(p)=n\). \textit{C. Hooley} [J. Reine Angew. Math. 225, 209--220 (1967; Zbl 0221.10048)] estimated \(\bigl| N_ a(x,1)\bigr| \) subject to the Generalised Riemann Hypothesis (GRH). \textit{L. Murata} [Arch. Math. 57, 555--565 (1991; Zbl 0755.11029)] considered \(\bigl| N_a(x,n)\bigr| \) under a similar hypothesis when \(a\) is squarefree and \(a\not\equiv 1(\text{mod~}4)\), and \textit{S. S. Wagstaff} [Acta Arith. 41, 141--150 (1982; Zbl 0496.10001)] considered this problem in greater generality. In this paper the central object is \(Q_a(x;k,l)\), the set of primes \(p\leq x\) with \(D_a(p)\equiv l(\text{mod~}k)\). The authors restrict themselves to the case where \(k=4\) (the arithmetic complexity of the subject more or less forces one to make a restriction on \(k\), certainly if one strives for explicit results). Meanwhile, for arbitrary rational \(a\), the reviewer considered \(Q_a(x;k,l)\) as well; for \(k=3,4\) in ``On the distribution of the order and index of \(g(\text{mod~}p)\) over residue classes. I'', preprint, arXiv.org/abs/math.NT/0211259], J. Number Theory (to appear). When \(k\) is a prime power he did this in `` On the distribution of the order and index of \(g(\text{mod~}p)\) over residue classes. II'', preprint, arXiv.org/abs/math.NT/0404339]. In `` On the distribution of the order and index of \(g(\text{mod~}p)\) over residue classes. III'', preprint, arXiv.org/abs/math.NT/0405527] he studied the general case. Finally he considered the problem `on average' in [Finite Fields Appl. 10, 438--463 (2004; Zbl 1061.11050)]. These four works will be denoted by [R1],[R2],[R3] and [R4], respectively. The quantity \(Q_a(x;k,l)\) in the case \(k| 2l\) has been well studied, in particular by \textit{K. Wiertelak} [Funct. Approximatio Comment. Math. 28, 237--241 (2000; Zbl 1009.11056)]. Then the related density is rational and this can be established unconditionally. However, the authors seem to have been the first to address the more general and much harder case. It is thus natural that the authors consider the cases \(l=0,2\) and \(l=1,3\) (for \(k=4\)) in two separate theorems (Theorem 1.1 and Theorem 1.2). They restrict themselves to the case where \(a\geq 3\) is a squarefree positive integer such that \(a\equiv 1(\text{mod~}4)\). Theorem 1.1, however, is not new. It is a weak corollary of a more general (\(a>0\) and rational) and complicated result due to Wiertelak [ibid], who also established it with a much sharper error term. In Theorem 1.2 it is shown that \(\bigl| Q_a(x;4,l)\bigr| /\pi(x)=1/6+O(1/\log \log x)\), as \(x\) tends to infinity, assuming GRH. For an integer \(a>0\) that cannot be written as an \(h\)th power with \(h\geq 2\) they first establish a similar result. Instead of \(1/6\) they then find associated densities which can be expressed as a sum of two fourfold sums. In the more restrictive case under consideration these densities are easily seen to be equal, which then yields the result. The method of proof involves an initial transfer of considerations about \(D_a(p)\) to corresponding ones about \(I_a(p)\), which are more amenable to be dealt with by tools from (algebraic) analytic number theory and by Chebotarev's density theorem in particular. The method employed by the authors goes back to the paper Murata [loc. cit.] concerning \(N_a(x,n)\). The results of that paper, however, can be rather more easily derived by employing Hooley's original method whilst keeping track of the \(n\) dependency of the error term for \(N_a(x,n)\) (cf. [R1]). Indeed, by a variation of Hooley's method whilst keeping track of dependencies of error terms on various quantities, in [R2] the existence of the limit \(\bigl| Q_a(x;k,l)\bigr| /\pi(x)\) for arbitrary rational \(a\), \(k\) and \(l\) is established on GRH and with a rather better error term than given in Theorem 1.2. This limit is expressed as a double sum (rather than a four fold sum) and involves Galois theoretic coefficients, and the main problem is to get a sufficient explicit handle on them so as to derive properties of this limit. In the case \(k=4\) this double sum can be even simplified to a single variable sum [R1]. Thus in the reviewer's opinion the method employed by the authors is far from optimal.