Extreme biases in prime number races with many contestants (Q2423423)

Let \(q\in\mathbb{N}\) and let \(a_1,\ldots,a_n\) be coprime to \(q\) and represent distinct congruence classes modulo \(q\). Assuming the Generalized Riemann hypothesis and a linear independence condition (LI) on the zeros of \(L\)-functions, \textit{M. Rubinstein} and \textit{P. Sarnak} [Exp. Math. 3, No. 3, 173--197 (1994; Zbl 0823.11050)] showed that the set of \(x\) for which \[\pi(x;q,a_1)>\ldots>\pi(x;q,a_n)\] has a logarithmic density, which we denote by \(\delta(q;a_1,\ldots,a_n)\). These densities can be different from \(1/n!\), which is there average value, but Rubinstein and Sarnak showed that \[\lim_{q\to\infty}\max_{a_1,\ldots,a_n}|\delta(q;a_1,\ldots,a_n)-1/n!| =0\;\;\;(*)\] for any fixed \(n\ge 2\). The present paper investigates deviations from the expected value of \(1/n!\) when \(n\) is allowed to grow with \(q\). An unpublished conjecture of Ford and Lamzouri says that one should still have the property (*) when \(n\le(\log q)^{1-\varepsilon}\), for any fixed \(\varepsilon>0\), but that if \(n\ge(\log q)^{1+\varepsilon}\) one should be able to find sets of residue classes with \(n!\delta(q;a_1,\ldots,a_n)\) arbitrarily small, and sets where it is arbitrarily large. The first half of the conjecture, for \(n\le(\log q)^{1-\varepsilon}\), has been proved by \textit{A. J. Harper} and \textit{Y. Lamzouri} [Probab. Theory Relat. Fields 170, No. 3--4, 961--1010 (2018; Zbl 1445.11096)], and the current paper establishes the second half. More precisely, assuming GRH and LI, it is shown that there is an absolute constant \(C>0\) such that, whenever \(1\ll n\le\phi(q)\), one can find classes with \[\delta(q;a_1,\ldots,a_n)\le \exp\left(-\frac{\min\{n,\phi(q)^{1/50}\}}{C\log q}\right)\frac{1}{n!},\] and also classes with \[\delta(q;b_1,\ldots,b_n)\ge \exp\left(\frac{\min\{n,\phi(q)^{1/50}\}}{C\log q}\right)\frac{1}{n!}.\] Rubinstein and Sarnak gave an expression for the densities \(\delta(q;a_1,\ldots,a_n)\) in terms of a probability measure, and the proof investigates its properties, with little further number-theoretic input.