On the number of strong primes (Q313427)

The author calls a prime ``strong'' if neither \(p-1\) nor \(p+1\) is smooth. (It is suggested that such primes are appropriate for constructing an RSA modulus, since Pollard's \(p-1\) method and its variants are less likely to detect them.) Let \(N(x,B)\) denote the number of primes \(p\leq x\) such that both \(p-1\) and \(p+1\) have a prime factor of size at least \(B\). Then it is shown that \[ N(x,\sqrt{x})\geq(\tfrac18+o(1))\pi(x). \] For the proof one uses a simple inclusion-exclusion argument to relate \(N(x,B)\) to \[ M_{\pm}(x,B):=\sum_{r<B}\Lambda(r)\pi(x;r,\pm 1) \] and \[ M(x,B):=\sum_{r,s<B}\Lambda(r)\Lambda(s)| \{p\leq x:\, r\mid p-1, s\mid p+1\}| . \] This follows the method of \textit{M. Goldfeld} [Mathematika 16, No. 1, 23--27 (1969; Zbl 0201.05301)], which was itself based on a classical idea of Chebyshev. One requires upper bounds for \(M_{\pm}(x,B)\) and a lower bound for \(M(x,B)\). For the latter it suffices to restrict attention to the case \(rs\leq x^{1/2}(\log x)^{-A}\), where one can use the Bombieri-Vinogradov theorem. It seems to the reviewer that one can obtain a bound \[ N(x,x^{\theta})\gg\pi(x) \] with an exponent \(\theta\) strictly greater than a half, by using the Brun-Titchmarsh theorem to estimate the terms in \(M_{\pm}(x,B)\) with \(r> x^{1/2}(\log x)^{-A}\), as in Goldfeld's work.