2
$\begingroup$

I am interested in the proof of the following result: Suppose that $A > 1$, $\lambda \in \mathbb{R}$, and for $0 < Z \leq 1$, let $U(Z)$ be the number of integer solutions $v$ of \begin{eqnarray} |v| < ZA \ \ \ \text{ and } \ \ \ \| \lambda v \| < Z A^{-1}. \end{eqnarray} Then, if $0 < Z_1 < Z_2 \leq 1$, we have $$ U(Z_1) \gg (Z_1/Z_2) \ U(Z_2). $$

I would greatly appreciate any comments or hints on this! Thank you very much!

PS Here $\|x\|$ denotes the distance the closest integer.

$\endgroup$
4
  • $\begingroup$ What does $\|x\|$ mean? $\endgroup$ Commented Jun 20, 2016 at 2:49
  • $\begingroup$ The implied constant in the $\gg$ – an absolute constant? or, depends on $A$? on $\lambda$? $\endgroup$ Commented Jun 20, 2016 at 5:26
  • 1
    $\begingroup$ Where does it come from? Why are you sure that it is true? $\endgroup$ Commented Jun 20, 2016 at 7:13
  • $\begingroup$ @GerryMyerson I think it is an absolute constant $\endgroup$
    – Johnny T.
    Commented Jun 20, 2016 at 13:56

3 Answers 3

2
$\begingroup$

If you don't mind, I'll reformulate your problem slightly. Let $X = ZA$, $B = A^2$. Then $XB^{-1} = ZA^{-1}$. We would like to know the number of integer solutions $U'(X)$ to the system of inequalities

$$ \begin{cases}|v| < X;\\\|\lambda v\|<XB^{-1}.\end{cases} $$

Let $\lfloor x \rfloor$ denote the largest integer not exceeding $x$, and put $\{x\} = x - \lfloor x \rfloor$. Then

$$ \|\lambda v\| = \|(\lfloor \lambda \rfloor + \{\lambda\})v\| = \|\lfloor \lambda\rfloor v + \{\lambda\}v\| = \|\{\lambda\}v\|. $$

Since $0 \leq \{\lambda\} < 1$, without loss of generality we may assume that $\lambda$ satisfies $0 \leq \lambda < 1$.

Suppose that $XB^{-1} > 1/2$. Then $0 \leq \|\lambda v\| \leq 1/2 < XB^{-1}$ is true for any choice of $\lambda$ or $v$, which means that $U(X)$ is equal to $2\lfloor X\rfloor + 1$. The same observation appplies to the case when $\lambda = 0$.

Suppose that $XB^{-1} \leq 1/2$ and $0 < \lambda < 1$. Denote the largest integer in the interval $(-\lambda X-1, \lambda X+1)$ by $k$. The fact that $\|\lambda v\| < XB^{-1}$ simply means that there exists some integer $n \in \{-k, -k+1, \ldots, k-1, k\}$ such that

$$ \left|n - \lambda v\right| < XB^{-1}. $$

Verify that there are at least $\lambda X/2$ and at most $2 \lambda X +3$ possible values of $n$.

Further, each $\lambda v$ is contained in exactly one interval of the form $(n - XB^{-1},n + XB^{-1})$, whose length is $2XB^{-1}$. Verify that this interval contains at least $X(\lambda B)^{-1}$ and at most $2X(\lambda B)^{-1} + 1$ numbers of the form $\lambda v$, where $v$ is an integer.

We conclude that, when $X/B \leq 1/2$ and $0 < \lambda < 1$,

$$ \frac{\lambda X}{2}\cdot \frac{X}{\lambda B} \leq U'(X) \leq (2 \lambda X + 3)(2 X(\lambda B)^{-1} + 1). $$

Otherwise,

$$ U'(X) = 2\lfloor X\rfloor + 1. $$

Finally, note that $U(Z) = U'(ZA)$. Plug in $B = A^2$ and $X=ZA$ inside the above expressions and deduce $U(Z_1) \gg (Z_1/Z_2)U(Z_2)$.

$\endgroup$
2
  • $\begingroup$ I don't understand your conclusion. The factor $- 2 \lambda X$ in your final estimate is usually larger than the main term. $\endgroup$
    – js21
    Commented Jun 22, 2016 at 8:35
  • $\begingroup$ Could you possible explain the part when you wrote: "Verify that there are at least $\lambda X/2$" by any chance? Thanks! $\endgroup$
    – Johnny T.
    Commented Dec 31, 2016 at 22:05
1
$\begingroup$

Is it a reference you want? Check chapter 12 (if I remember correctly) in Davenport's book on diophantine equations and inequalities.

$\endgroup$
1
$\begingroup$

I can show the (exact) inequality $$(*) \ \ \ \ \ \ V(Z_1) \geq \left(\frac{Z_1}{Z_2} \right)^2 V(Z_2) \ \ \quad (\frac{2}{A} \leq Z_1 \leq Z_2 \leq \frac{A}{2}), $$ for a smoothed version of $U$ defined by $$ V(Z) = \sum_{\nu \in \mathbb{Z}} \mathrm{sinc}^2\left( \frac{\nu}{2ZA} \right) \left( 1 - \frac{A ||\lambda \nu||}{Z} \right)_{+}. $$ Indeed, the Fourier transform of the function $f(x) =\mathrm{sinc}^2\left( \frac{x}{2} \right)$ is given by the formula $\hat{f}(x)= (1- |x|)_+$, so that we have $$V(Z) = \sum_{\nu \in \mathbb{Z}} \ f \left( \frac{\nu}{ZA} \right) \hat{f}\left( \frac{A ||\lambda \nu||}{Z} \right).$$ Now, if $Z^{-1} A \geq 2$ and $ZA \geq 2$, then Poisson summation formula yields $$ \hat{f}\left( \frac{A ||\lambda \nu||}{Z} \right) = \sum_{n \in \mathbb{Z}} \hat{f}\left( \frac{A (\lambda \nu + n )}{Z} \right) = \frac{Z}{A} \sum_{\mu \in \mathbb{Z}} f \left( \frac{Z \mu}{A} \right) e(\lambda \mu \nu), \\ \text{and similarly} \ \ \ \sum_{\nu \in \mathbb{Z}} f \left( \frac{\nu}{ZA} \right) e(\lambda \mu \nu) = ZA \ \hat{f}\left( ZA ||\lambda \nu||\right), $$ where $e(x) = e^{2 i \pi x}$. We thus get $$V(Z) = \frac{Z}{A} \sum_{\nu \in \mathbb{Z}} \ f \left( \frac{\nu}{ZA} \right) \left( \sum_{\mu \in \mathbb{Z}} f \left( \frac{Z \mu}{A} \right) e(\lambda \mu \nu)\right) \\ = \frac{Z}{A} \sum_{\mu \in \mathbb{Z}} f \left( \frac{Z \mu}{A} \right) \left( \sum_{\nu \in \mathbb{Z}} \ f \left( \frac{\nu}{ZA} \right) e(\lambda \mu \nu)\right) \\ =Z^2 \sum_{\mu \in \mathbb{Z}} f \left( \frac{Z \mu}{A} \right) \hat{f}\left( ZA ||\lambda \nu||\right) \\ = Z^2 V(Z^{-1}). $$ Since $V(Z^{-1})$ is a decreasing function of $Z$, this yields $(*)$.

Note that $(*)$ implies an inequality of the form $$ U'(Z_1) \gg \left(\frac{Z_1}{Z_2} \right)^2 U(Z_2),$$ with $$ U'(Z_1) = \sum_{\substack{\nu \in \mathbb{Z} \\ ||\lambda \nu|| < Z_1 A^{-1}}} \min \left( 1, \left(\frac{Z_1 A}{|\nu|}\right)^2 \right). $$

$\endgroup$
2
  • 1
    $\begingroup$ I think, inequality $U(Z_1)\gg (Z_1/Z_2)^2 U(Z_2)$ is less or more clear: partition the arc $\|z\|<Z_2A^{-1}$ onto arcs $\delta_1,\dots,\delta_k$ of length $2Z_1A^{-1}$, and also partition integers between 1 and $Z_2A$ onto segments $\Delta_1,\dots,\Delta_k$ of length $Z_1A$, $k$ is about $Z_2/Z_1$. By pigeonhole principle there exist indices $u,v$ such that at least $U(Z_2)/k^2$ elements of $\Delta_u$ belong to $\delta_v$, this implies $U(Z_1)\geqslant U(Z_2)/k^2$. $\endgroup$ Commented Jun 22, 2016 at 11:03
  • $\begingroup$ I was wondering would a similar argument work if I change the definition of $U(Z)$ to be the number of $v$ with $|v|< ZA$ and $\|\lambda (v - t_0)\| <ZA^{-1}$, where $t_0$ is some integer? Thank you very much! $\endgroup$
    – Johnny T.
    Commented Dec 31, 2016 at 22:35

You must log in to answer this question.

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