A remark on Artin's conjecture (Q800409)

Artin's conjecture states that if \(a\) is an integer other than \(-1\) or a square, then \(a\) is a primitive root for infinitely many primes. Until now the only significant result on the problem was obtained by \textit{C. Hooley} [J. Reine Angew. Math. 225, 209--220 (1967; Zbl 0221.10048)]. He showed that Artin's conjecture would follow from the Riemann Hypothesis for a suitable set of Dedekind zeta-functions. The present paper makes a very important advance in a different direction, namely it is shown that there exists some \(a\) which is a primitive root for infinitely many primes. More precisely let \(q,r,s\) be any three distinct primes and let \[ S=\{qs^2, q^3r^2, q^2r, r^3s^2, r^2s, q^2s^3, qr^3, q^3rs^2, rs^3, q^2r^3s, q^3s, qr^2s^3, qrs\}. \] Then for any sufficiently large \(x\) there is an \(a\in S\) which is a primitive root for at least \(\delta x(\log x)^{-2}\) primes \(p\leq x\). Here \(\delta\) is a numerical constant. (It is stated that there exists an \(a\) which works for all \(x\), but this appears not to follow from the argument given.) Unfortunately there is no way of telling which value of \(a\) one should take. It may be noted that one would expect at least \(\delta x(log x)^{-1}\) primes \(p\leq x\) to have \(a\) as a primitive root. The proof is not difficult, the main input being a familiar sieve result. Specifically one uses the fact that there are at least \(\delta x(log x)^{-2}\) primes \(p\leq x\) for which \(\frac{1}{2} (p-1)\) is composed only of odd prime factors greater than \(x^{\alpha}\). Various admissible constants \(\alpha >0\) have been obtained, and any of these would suffice for a qualitative form of the result. To get as small a set \(S\) as possible one uses a value \(\alpha =1/4+\varepsilon\). This follows from the work of \textit{E. Fouvry} and \textit{H. Iwaniec} [Acta Arith. 42, 197--218 (1983; Zbl 0517.10045)], where a form of the Bombieri-Vinogradov prime number theorem is proved which allows a ``level of distribution'' greater than \(\frac{1}{2}\) to be used in the sieve. For ``almost all'' of the primes obtained by this sieve result, the multiplicative group \(\pmod p\) is shown to be generated by \(q,r\) and \(s\). There is then a subsidiary argument to deduce that this group is even generated by one of the elements of \(S\). It seems to the reviewer that the result can be improved in two ways. Firstly one can replace the set \(S\) by its subset \[ S'=\{qs^2, q^3r^2, q^2r, r^3s^2, r^2s, q^2s^3, qrs\}. \] Secondly one can prove the same result for a set \(S''\) consisting of any 9 distinct primes. It follows that Artin's conjecture holds whenever \(a\) is prime, save for at most 8 exceptional values of \(a\). These results may be capable of further improvement by working harder on the sieve aspect of the proof.