6
$\begingroup$

Let $\{a_n\}_{n\geq 1}$, $|a_n|\leq 1$, be such that $F(s) = \sum_{n=1}^\infty a_n n^{-s}$ can be continued analytically to a neighborhood of the line $\Re s = 1$. (For instance, let $a_n = \mu(n)$.) The truncated Perron formulae I know (e.g., Corollary 5.3 in Montgomery-Vaughan, or Thm 2.3 in Chapter II.2 of Tenenbaum (3rd ed)) give $$\sum_{n\leq x} a_n = \int_{1-iT}^{1+i T} F(s) x^s \frac{ds}{s} + O\left(\frac{x \log x}{T}\right),$$ and, assuming that $F(s) = 0$, $$\sum_{n\leq x} \frac{a_n}{n} = \int_{-iT}^{i T} F(1+s) x^s \frac{ds}{s} + O\left(\frac{ \log x}{T}\right).$$

Question:

Can one give a truncated Perron formula with an error term of about $O(x/T)$ or $O(1/T)$, respectively (or $O(x (\log T)/T)$ or $O((\log T)/T)$, say), at least under some reasonable circumstances?

Bonus points if the formula is explicit

$\endgroup$

3 Answers 3

4
$\begingroup$

The mentioned Theorem 1.2 of https://ramare-olivier.github.io/Maths/TruncatedPerron-5.pdf is not quite enough: it leads to $\mathcal{O}((x\log x)/T^2)$ for a large set of $T$'s near your wanted value. But if you keep reading, you'll reach Theorem 5.3. It is not exactly usuable as you want, as $T^*$ depends on $T$, but the proof will apply: select $\delta = 1$ (hence $\kappa = 2$), split the integral at $1/T$, $1/T^{1-\ell/k}$, $\infty$ with $k$ large and $\ell \le k/2$. One gets the contributions:

-- $\mathcal{O}( x / T )$ for the small $u$'s by $\sum\limits_{\cdots \le ux} \ldots\ll ux$.

-- $\mathcal{O}( x (\log x) T^{(k-1)(k-\ell)/k} / T^k)$ with a constant independant of $k$ and $\ell$. The exponent of $T$ reads $1 + (k-1) \ell /k$.

So one can take $\ell = 1+[(2\log x)/\log T ]$ and $k = 2 \ell$. This gives what you want up to the fact that the set of admissible $T^*$ depends on inital $T$ and $x$. This should be harmless. I just had my morning tea, so caution applies :)

$\endgroup$
2
  • $\begingroup$ Why take $\kappa=2$? Doesn't that give us a term $x^2/T^k$, which is much too large? It would make more sense to take $kappa = 1 + 1/log x$. And do we really need to split the integral? I don't see the need. On the other hand, while I agree the value of $k$ you choose is good, I do not see why one couldn't take a much larger $k$, and thereby get a much better error term. Is there really no condition on how large $k$ can be? $\endgroup$ Sep 5, 2021 at 1:49
  • 1
    $\begingroup$ Ok, I was too fast, but in case anyone looks for it, a solution along similar lines is here: ramare-olivier.github.io/Maths/… $\endgroup$ Sep 5, 2021 at 15:27
1
$\begingroup$

Ah, I get it - the factor of $\log x$ is really there because the truncation is sharp. If the truncation is continuous (and of bounded variation), then, not unexpectedly, the factor disappears. (See, e.g., https://arxiv.org/abs/1703.00261) Then one can start from the bound for the case of continuous truncation and deduce that there is a $T^*$ for which the sharp truncation at $T^*$ also has a good bound.

$\endgroup$
0
$\begingroup$

Look at this article, it seems that what you need, if you take in $\delta$ greater value than $\frac{1}{\log\log T}$ (see Theorem 1.2) https://scholar.google.com/citations?view_op=view_citation&hl=en&user=c8ViUvsAAAAJ&cstart=20&pagesize=80&citation_for_view=c8ViUvsAAAAJ:ns9cj8rnVeAC

$\endgroup$
1
  • $\begingroup$ In order to avoid a factor of $\log T$, wouldn't we need to set $\delta$ equal to a (possibly small) constant? Also, I don't see how the formula avoids the factor of $\log x$: we would need to set $\kappa$ no larger than $O(1/\log x)$ to avoid gaining a factor like that, and that would make the sum $\sum_n |a_n|/x^\kappa$ be of size about $\log x$ after all. $\endgroup$ Sep 3, 2021 at 22:29

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.