9
$\begingroup$

I have seen justifications for various conjectures in analytic number theory, e.g. the (generalized) Riemann hypothesis, the Chowla conjectures, etc. justified by a heuristics in which the relevant arithmetic functions are modeled by random variables.

In the setting of primes in arithmetic progression, these heuristics say that the error term $E(x;q,a)$ for primes in particular primitive residue class should be approximately on the order of magnitude of $\frac{\sqrt{x}}{\varphi(q)}$, which is approximately GRH. If one sums this for $q$ up to $\sqrt{x}$, then the accumulated error is approximately on the order of magnitude of $x$, and the Bombieri-Vinogradov theorem actually establishes a bound of this nature.

The Elliott-Halberstam conjecture asserts that we actually get an error bound of $\frac{x}{(\log x)^A}$ for any fixed $A$ if we sum over moduli all the way up to $x^{1-\epsilon}$. My question is whether there is a convincing heuristic model that this should be the case, since the standard one that suggests GRH doesn't seem to accommodate such large moduli.

Improve this question
$\endgroup$
1
  • 4
    $\begingroup$ I am not sure your initial analysis is correct. From GRH one gets an error term in $\psi(x;q,a)$ of size $\sqrt x$ (ignoring logs), and summing over the characters gives the same size bound for $E(x;q,a)$, not $\sqrt x/\phi(q)$ as you say. But one expects cancellation over this sum, so the "proper" size is expected more like $\sqrt{x/\phi(q)}$, with logs and maybe $x^\epsilon$. Plugging this in gives Elliott-Halberstam (for the reader, it might be convenient to actually give the E-H sum: $\sum_{q\le Q} \max |E(x;q,a)|\ll x/\log^\infty x$). $\endgroup$
    – Conder
    Apr 29, 2014 at 0:13

1 Answer 1

Reset to default
2
$\begingroup$

The Elliott-Halberstam conjecture is an immediate consequence of Montgomery's (modified) conjecture, which proposes that if $\epsilon>0$, $\gcd(a,q)=1$ and $x>q^{1+\epsilon}$, then

$\Big|\displaystyle\sum_{\substack{n\leq x \\ n\equiv a\pmod{q}}}\log p -\frac{x}{\varphi(q)} \Big|\ll_{\epsilon} x^{1/2+\epsilon}q^{-1/2+\epsilon}$.

For a very nice heuristic for why one might believe this, see Heuristic for Montgomery's conjecture.

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.