Let $\alpha$ be irrational. A famous theorem of Vinogradov says that $\{ \alpha p\}$ is equidistributed in $[0,1]$ as $p$ runs over all primes.

Let $a,q$ be natural numbers with $\gcd(a,q) = 1$. Then is the sequence $\{ \alpha p\}$ equidistributed in $[0,1]$, as $p$ runs over primes with $p \equiv a \bmod q$?

Almost certainly this must be known. So I'm looking for a precise reference in the literature as I need it in a paper. Ideally, it would be nice to have an effective version which makes explicit the speed of convergence (via the Erdős-Turán inequality, say).


I think the sought result follows from Vinogradov's theorem. By Weyl's criterion and the orthogonality of Dirichlet characters, the sought result can be reformulated as follows. For every nonzero integer $k$, and for every Dirichlet character $\chi$ modulo $q$, we have $$\sum_{p<x}\chi(p)e(k\alpha p)=o(\pi(x))\qquad\text{as}\qquad x\to\infty.$$ The function $n\mapsto\chi(n)$ can be written as a linear combination of additive characters $n\mapsto e((a/q)n)$ with $a\in\mathbb{Z}$, hence it suffices to show that $$\sum_{p<x}e((a/q+k\alpha)p)=o(\pi(x))\qquad\text{as}\qquad x\to\infty.$$ This in turn follows from the equidistribution of $\{(a/q+k\alpha)p\}$, because $a/q+k\alpha$ is an irrational number. QED

The error terms will be similar as in Vinogradov's theorem, but with additional constants depending on $q$. The rate of convergence will depend on how well $\alpha$ can be approximated by rational numbers. See Chapter XI and the subsequent notes in Vinogradov's "The method of trigonometrical sums in the theory of numbers". I am sure there is a modern reference, but I am no expert in this subject.

  • 2
    $\begingroup$ Great answer, thanks. $\endgroup$ Sep 2 '20 at 9:30
  • $\begingroup$ @DanielLoughran: Thank you. Actually, we don't even need to use Dirichlet characters. The condition $p\equiv a\pmod{q}$ can be detected directly by additive characters modulo $q$, so one arrives at the last display (with different $a$'s) directly. $\endgroup$
    – GH from MO
    Sep 2 '20 at 19:28
  • 1
    $\begingroup$ Yes I had realised this as well. It's a nice trick. $\endgroup$ Sep 2 '20 at 20:07
  • 1
    $\begingroup$ A more interesting problem seems to be the following. Let $K/\mathbb{Q}$ be a number field. Then is $\{\alpha p\}$ equidistributed as $p$ runs over all primes completely splti in $K$? (Or more general Chebotarev sets.) This doesn't seem to be immediately reducible to Vinogradov's result. $\endgroup$ Sep 3 '20 at 8:24
  • 2
    $\begingroup$ It is not difficult to extend Vinogradov's approach to $K/\mathbb Q$. Vaughan's identity is is a decomposition of $-\zeta'/\zeta$ and here you only have to replace zeta by the Dedekind zeta. $\endgroup$
    – Dr. Pi
    Sep 14 '20 at 16:53

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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