11
$\begingroup$

Let $n \ge 2$. Let $p_1,\dots,p_m$ be distinct primes $\equiv 1 \pmod{n}$. Let $N=p_1 p_2 \cdots p_m$. If $\gcd(a,N)=1$ and the equation $x^n \equiv a \pmod{N}$ has a solution then it has $n^m$ solutions. This question concerns how these $n$-th roots are distributed as $a$ varies, particularly with $n$ fixed and $m$ large.

Question. Do there exist $\alpha$ and $\beta$ with $0 \le \alpha \le \beta <1$ and $\beta-\alpha$ small, such that for every $a$ coprime to $N$ that is an $n$-th power modulo $N$, there is $b$ satisfying $\alpha N \le b \le \beta N$ and $b^n \equiv a \pmod{N}$? Can we take $\beta-\alpha=O(n^{-m})$?

$\endgroup$
1
  • 2
    $\begingroup$ You could look at my paper with Sitar: math.ubc.ca/~gerg/index.shtml?abstract=ETMTERRQDQ ... in Section 3 we treat the case $n=2$, but looking over all moduli at once rather than those with the factorization you have. The techniques might well generalize to your setting. $\endgroup$ Feb 9, 2014 at 2:46

2 Answers 2

3
$\begingroup$

This is a partial answer. The idea is from my recent preprint http://arxiv.org/abs/1509.03768

Additionally assuming that $p_1, \cdots, p_m$ are least among the congruence $1$ mod $n$,

By repeated application of Linnik's theorem, there is an absolute constant $L$ such that $N< n^{L^m}$.

We use the following result on exponential sums over $\mathbb{Z}_N^{*}$ by J. Bourgain:

Let $N\geq 1$. For any $\epsilon>0$, there exist a constant $\delta_0=\delta_0(\epsilon)>0$ such that for any subgroup $H$ of $\mathbb{Z}_N^{*}$ with $|H|>N^{\epsilon}$, $$ \max_{(m,N)=1}\left|\sum_{a\in H}e^{2\pi i m\frac {a}N}\right|<N^{-\delta_0}|H|.$$

Note that we can use the above for any cosets of $H$.

The set of solutions of the congruence $x^n \equiv a$ mod $N$ forms a coset of the subgroup $H$ of $\mathbb{Z}_N^{*}$ where $$ H=\{y\in\mathbb{Z}_N^{*} | y^n \equiv 1 \textrm{ mod }N\}.$$

Taking $\epsilon = \frac{m}{L^m}$, we have $N^{\epsilon}<|H|$.

Then by Bourgain's exponential sum result applied on the coset, and Erdős-Turán inequality, there exists $\delta(m)>0$ such that $$ \left||\{ 0<x<N : x^n\equiv a (\textrm{ mod }N), \frac xN \in (\alpha,\beta) \textrm{ mod $1$} \}| - (\beta-\alpha)n^m\right|\leq n^m N^{-\delta(m)}$$ Thus, $$ |\{ 0<x<N : x^n\equiv a (\textrm{ mod }N), \frac xN \in (\alpha,\beta) \textrm{ mod $1$} \}| = (\beta-\alpha)n^m +O(n^m N^{-\delta(m)}).$$ Therefore, if $N^{-\delta(m)}=o(\beta-\alpha)$, then we are guaranteed the existence of solution to $x^n\equiv a$ mod $N$ in the interval $(\alpha N, \beta N)$.

Here's the reference:

J. Bourgain, Exponential sum estimates over subgroups of $\mathbb{Z}_q^{*}$, $q$ arbitrary , Journal d'Analyse Mathematique, December 2005, Volume 97, Issue 1, pp 317-355.

$\endgroup$
2
$\begingroup$

Partial answer.

For $n=2$ finding the smallest solution is NP-complete via reduction from SAT.

NP-Complete Decision Problems for Binary Quadratics p. 3

The problem of solving $ x^2 \equiv a \pmod{N}$ with natural solution satisfying $ 0 \le x \le \gamma$ is NP-complete.

Since this is NP-complete and $a$ depends on the SAT instance I don't expect very sharp bounds, might be wrong on this.

$\endgroup$

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.