1
$\begingroup$

In Tao's blog we can find this exercise: Let $[M,M+N]$ be an interval for some $M \in {\bf R}$ and $N > 0$, and let $\xi_1,\dots,\xi_J \in {\bf R}$ be $\delta$-separated. For any complex numbers $c_1,\dots,c_J$, show that $$ \int_M^{M+N} |\sum_{j=1}^J c_j e( \xi_j t )|^2\ dt \ll (N + \frac{1}{\delta}) \sum_{j=1}^J |c_j|^2.$$

Since it is well-known that the discrete version of the large sieve inequality holds with $N-1+\frac{1}{\delta}$, I was wondering if the previous inequality can be proved with $N-1+\frac{1}{\delta}$ instead of $N+\frac{1}{\delta}$. I tried to follows the same proof for the case $N+\frac{1}{\delta}$ but I'm not able to obtain the wanted term.

$\endgroup$

1 Answer 1

3
$\begingroup$

In the continuous version of the problem, it is perfectly possible to have $N < \frac{1}{2}$ and $\delta >2$ so that $N + \frac{1}{\delta}<1$ but the left-hand side is positive, so the $N + \frac{1}{\delta}-1$ version cannot be true.

As to whether the inequality is true with some reasonable assumptions on $\delta$, the inequality doesn't pass dimensional analysis. If we multiply all the $M,N$ by a constant $C$ and divide $\xi_j$ by $C$ for all $j$, we multiply the left hand side by $C$ will multiplying $N$ and $\delta^{-1}$ by $C$ but not multiplying $1$ by $C$. In particular, if $C$ is small, doing this makes the inequality stronger. So if the bound you state holds for $\delta\leq \frac{1}{2}$, say, then taking $C = 1/(2\delta)$ we deduce the stronger inequality $N + \delta^{-1} - \frac{\delta^{-1}}{2}$, but I think this version is not true (or if it is one should just state $N+ \delta^{-1}/2$ rather than the $-1$ version.)

If you are interested for some reason in $\xi_j$ that are $\delta$-separated in the $\mathbf R/\mathbf Z$ sense, then one can easily reduce to the discrete case: For any function $f$ we have

$$\int_{M}^{M+N} f(t)dt = \int_0^1 \sum_{\substack{n \in \mathbb Z\\ n+x \in [M,M+N]}} f(n+x) dx$$ and for $f=|\sum_{j=1}^J c_j e( \xi_j t )|^2$ you can apply the large sieve inequality with the $-1$ to the sum over $n$ and then deduce the large sive inequality with the $-1$ for the integral.

$\endgroup$
5
  • $\begingroup$ Thank you but, following Tao's blog, it seems that the delta separation is connected to the distance of a number to the nearest integer, so delta is a number between 0 and 1/2. Am I wrong? $\endgroup$ Commented Mar 21 at 13:15
  • 2
    $\begingroup$ @Yep When you ask a question referencing a specific source like Tao's blog, you should always mention the specific blog post, and link to it, in case it helps people to answer the question. This is particularly the case if you want to reference a specific definition. $\endgroup$ Commented Mar 21 at 13:27
  • 1
    $\begingroup$ You're referencing the definition of $\delta$-separated for $\xi_j \in \mathbf R/\mathbf Z$. The definition of $\delta$-separated for $\xi_j\in \mathbf R$ would use the distances in $\mathbf R$. In general $\delta$-separated makes sense for any metric space and uses the metric on that space. $\endgroup$ Commented Mar 21 at 13:32
  • $\begingroup$ You are right, I should have included the link. However, I wonder if the inequality is plausible with reasonable assumptions about delta. $\endgroup$ Commented Mar 21 at 15:17
  • $\begingroup$ @Yep See additional remarks I added to the answer. $\endgroup$ Commented Mar 21 at 15:29

You must log in to answer this question.

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.