Exceptional zeros, sieve parity, Goldbach (Q6044271)

This paper explores the connection between ``exceptional real zeros'' of Dirichlet \(L\)-functions, and the ``parity phenomenon'' for sieves. In particular it examines the binary additive problem \[ G(N):=\sum_{l+m=N}\Lambda(l)\Lambda(m) \] and its generalization \[ F(N):=\sum_{l+m=N}a(l)b(m), \] where \(a(l)\) and \(b(m)\) are supported on almost-primes. One motivating result, due to \textit{J. B. Friedlander} et al. [J. Number Theory 233, 78--86 (2022; Zbl 1493.11140)], states that if a weak form \[ \delta\mathfrak{S}(n)n<G(n)<(2-\delta)\mathfrak{S}(n)n \] of the expected asymptotic formula holds for all large even \(n\), then there are no exceptional zeros. The question then becomes, what happens with sums like \(F(n)\)? For suitable \(a(l)\) and \(b(m)\) the paper gives an asymptotic evaluation of the sum \[ S(N,q)=\sum_{n\equiv 0(\mathrm{mod}\, q)}F(n)e^{-n/N} \] when \(N\) is suitably large compared with \(q\), under the assumption of an exceptional zero for the modulus \(q\). One can then compare this with the asymptotic behaviour predicted by the naive heuristics. The outcome is that the formulae are inconsistent whenever \(a(l)\) and \(b(m)\) are supported on integers with a fixed parity for the number of prime divisors. One can see this as a substantial generalization of the above result for \(G(n)\). On the other hand, if one takes \(a(l)=\Lambda_a(n)\) and \(b(m)=\Lambda_b(m)\), where \(\Lambda_k=\mu\, *\,\log^k\), one finds that there is no inconsistency unless \(a=b=1\). The situation here is strongly reminiscent of Bombieri's asymptotic sieve [\textit{E. Bombieri}, Rend., Accad. Naz. XL, V. Ser. 1--2, 243--269 (1976; Zbl 0422.10042)].