6
$\begingroup$

Let $\{a_n\}_{1\leq n\geq N}$, $a_n\in \mathbb{C}$. Let $F(s) = \sum_{n=1}^N a_n n^{-s}$. By a mean-value theorem (Montgomery-Vaughan, 1973), $$\int_0^T |F(i t)|^2 = \sum_{n=1}^N |a_n|^2 (T + O(n)).$$

Now, suppose that $\{a_n\}$ has prime support (i.e., $a_n=0$ for all composite $n$). Can we obtain a better bound? (Can we save a factor of roughly $\log n$ on the term $O(n)$, say?)

(The motivation here is an (imperfect) analogy with the large sieve for sequences with prime support.)


An interesting case would be given by $a_p = (\log p)^2/p$ with $a_n=0$ for $n$ composite. Then $\sum_{n=1}^N |a_n|^2 n$ is in the order of $(\log N)^4$. Can we go down to $O((\log N)^3)$?

$\endgroup$
6
  • 3
    $\begingroup$ Lemma 1 here: arxiv.org/pdf/1706.03755.pdf might be what you want. $\endgroup$ Commented Oct 4, 2022 at 17:44
  • $\begingroup$ Ah, nice! Has any work been done on improving the constant? (I suppose one can use a nicely optimized function in the style of Selberg, Vaaler, etc.?) $\endgroup$ Commented Oct 4, 2022 at 18:53
  • $\begingroup$ (Things get a bit messy in the range $T\leq n\leq T^{1+\epsilon}$, but, in my application, that may not matter much.) $\endgroup$ Commented Oct 4, 2022 at 18:54
  • 1
    $\begingroup$ @HAHelfgott For (at least part of) the story of B-S, see Vaaler’s very nice Bulletin of the AMS paper on extremal Fourier analysis. Also, shouldn’t you expect to save log N instead of N? $\endgroup$
    – 2734364041
    Commented Oct 5, 2022 at 16:43
  • 1
    $\begingroup$ Yes, typo! Thanks! $\endgroup$ Commented Oct 5, 2022 at 19:10

0

You must log in to answer this question.