10
$\begingroup$

The exact question I am interested in is the following.

Fix a small $\varepsilon\in(0,1)$ and an integer $q\ge 2$ (you may assume that $q$ is prime if it helps though I believe it shouldn't matter too much). For a large prime $P$ and an integer $a\in\mathbb Z$, define $G(a,P)=\{aq^m\mod P: m=0,1,2,\dots\}$ where the remainders are taken in the range $(-P/2, P/2)$ (i.e., with the minimal possible absolute value).

Is it true that for all primes $P$ outside of a set of density at most $\varepsilon$ (in any sense of the word "density" that is subadditive), $G(a,P)$ contains a remainder in the range $(-\varepsilon P,\varepsilon P)$ for every choice of $a\in \mathbb Z$?

However I'll be also interested in any nontrivial results in the same direction even if they fall somewhat short of a complete answer (be it affirmative or negative).

$\endgroup$
13
  • 2
    $\begingroup$ $a$ is chosen after $P$? $\endgroup$
    – markvs
    Commented Jan 14, 2021 at 18:16
  • 7
    $\begingroup$ If I understand correctly, you're almost asking if the cosets of multiplicative subgroups of the residues mod $p$ are equidistributed. This is known as long as the size of the subgroup is at most $p^{\delta}$ for some fixed $\delta$. Of course, there's still an issue of how q relates to the size of the induced multiplicative subgroup. $\endgroup$
    – Mark Lewko
    Commented Jan 14, 2021 at 18:33
  • 5
    $\begingroup$ For almost all primes $P$, one will have that $G(a,P)$ is a coset of a subgroup of the multiplicative group mod P with size at least $P^{1/3}$. It seems an affirmative answer will follow from the result of Bourgain--Glibichuk--Konyagin appearing as Theorem 7 here, along with the Erdos-Turan inequality (Lemma 6): insa.nic.in/writereaddata/UpLoadedFiles/IJPAM/20019e3c_15.pdf $\endgroup$ Commented Jan 14, 2021 at 18:34
  • 2
    $\begingroup$ @MarkLewko Yep, for almost all primes $P$ and $a\ne 0$ we shall have $|G(a,P)|\ge P^{\frac 12-\varepsilon}$ by an elementary counting argument, so it seems to answer the question in the affirmative. That's great. Let me see if I can understand the proof in Don's reference :-) $\endgroup$
    – fedja
    Commented Jan 14, 2021 at 18:56
  • 2
    $\begingroup$ @katago No, what I stated was exactly what would be sufficient to figure out what I wanted to figure out. I suspected that a stronger result might have been known (as it turned out to be the case) but I tried to request the minimum I would be happy with :-) $\endgroup$
    – fedja
    Commented Jan 14, 2021 at 22:01

1 Answer 1

3
$\begingroup$

$\newcommand{\F}{\mathbb F}$ $\newcommand{\eps}{\varepsilon}$ (As reqested by the OP, and to address @Mark Lewko's comments, here is the argument showing that the statement is true for the primes satisfying a certain condition; the missing counterpart is to prove that almost all primes satisfy the condition in question.)

Claim. Suppose that $p$ is a prime, $H<\F_p^\times$, and $a\in\F_p^\times$. If $|H|>(2\eps)^{-1}\sqrt p\log p$, then the coset $aH$ has a non-empty intersection with the interval $I:=(-\eps p,\eps p)\subset\F_p$.

Proof. Let $H^\perp<\widehat{\F_p^\times}$; that is, $H^\perp$ is the subgroup of those multiplicative characters of $\F_p$ with $H$ in their kernel. Assuming for a contradiction that $aH\cap I=\varnothing$, we have $$ \sum_{z\in\F_p^\times} \Big(\sum_{\chi\in H^\perp}\chi(a^{-1}z)\Big)\ \Big(\sum_{g\in I} \sum_{\psi\in\widehat{\F_p}} \psi(z-g)\Big) = 0 $$ where $\psi$ runs over all additive characters of $\F_p$. The contribution of the principal character $\psi=1$ is $$ |I| \sum_{z\in\F_p^\times} \sum_{\chi\in H^\perp}\chi(a^{-1}z) = |I|p; $$ therefore, changing the order of summation and separating the terms with $\psi=1$, we get $$ |I|p \le \sum_{\psi\ne 1} \Big| \sum_{g\in I} \psi(-g)\Big| \sum_{\chi\in H^\perp} \Big|\sum_{z\in\F_p^\times} \chi(a^{-1}z)\chi(z) \Big|. $$ The sum over $z$ is a Gauss sum; as such, it does not exceed $\sqrt p$ in the absolute value. This gives $$ |I|p \le |H^\perp|\,\sqrt p \sum_{\psi\ne 1} \Big| \sum_{g\in I} \psi(-g)\Big|. $$ The outer sum in the right-hand side can be written explicitly as $$ \sum_{u=1}^{p-1} \Big| \sum_{g=-\eps p}^{\eps p} e^{2\pi i ug/p}\Big|, $$ which easily yields the (well-known) upper bound $p\log p$ for the whole sum. As a result, $$ |I|p \le |H^\perp| p^{3/2}\log p $$ and the assertion follows in view of $|H^\perp|=(p-1)/|H|$.

$\endgroup$
3
  • $\begingroup$ Thanks! It looks nice, indeed. Thanks to @MarkLewko too for pointing out how to improve the elementary counting argument (though "elementary" stops being "trivial" in that improvement) :-) Anyway, the question has been completely answered, so I'll push the "accept" button. However, if someone still wants to add something in comments or another answer, that will be always welcome. $\endgroup$
    – fedja
    Commented Jan 18, 2021 at 2:16
  • $\begingroup$ I believe that it should be $\chi(a^{-1}z){\color{red}\psi(z)}$ in the rearranged sum and it looks like that if we use the Fejer kernel instead of the Dirichlet one, we can kill the $\log p$ factor. Am I right? $\endgroup$
    – fedja
    Commented Jan 18, 2021 at 17:28
  • $\begingroup$ @fedja: For the first remark, the answer is "yes"; any other typos, to fix them all together? For the second remark - yes, as far as I can see, using Fejer's kernel saves the logarithm. $\endgroup$
    – Seva
    Commented Jan 19, 2021 at 7:00

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .