Let $N(T,\chi)$ denote the number of zeros of $L(s,\chi)$ with imaginary part between $0$ and $T$, with any zero with imaginary part equal to $T$ or to $0$ (not that the latter kind really exists) counting as half a zero. Here I am following the convention in Montgomery-Vaughan, rather than that in part of the literature, where $N(T,\chi)$ means what I would call $N(T,\chi) + N(T,\overline{\chi})$.

The explicit literature generally (McCurley, Trudgian, Bennett-Martin-O'Bryant-Rechnitzer...) generally bounds $N(T,\chi) + N(T,\overline{\chi})$. The question is: what kind of explicit bounds we can extract from their proofs for $N(T,\chi)$?

The first step is easy: we can express $N(T,\chi)$ as $\text{main term} + S(T,\chi)-S(0,\chi)$, as in Montgomery-Vaughan, Thm. 14.5, where $S(T,\chi) = \frac{1}{\pi} \arg L(1/2+iT,\chi)$. One would then decompose $$S(T,\chi)-S(0,\chi) = \frac{1}{\pi} \left(\arg L(\sigma+i T,\chi)|_{\sigma=\sigma_0}^{1/2} + \arg L(\sigma_0+i t,\chi)|_{t=0}^T + \arg L(\sigma,\chi)|_{\sigma=1/2}^{\sigma_0}\right)$$ for some $\sigma_0>1$ of our choice. The literature gives the bound $2 \log \zeta(\sigma_0)$ on $\left|\arg L(\sigma_0+it)|_{t=-T}^T\right|$. The reason is a mystery to me -- it is obvious that $2 \sum_p \arcsin p^{-\sigma}$ is a tighter upper bound on $\left|\arg L(\sigma_0+it)|_{t=-T}^T\right|$ (and it is easy to compute). I do not know how to do better than $2 \sum_p \arcsin p^{-\sigma}$ as an upper bound on $\left|\arg L(\sigma_0+it)|_{t=0}^T\right|$, and suspect one cannot, in general, as $t$ and $\chi$ could conspire.

The bulk of the explicit literature deals with bounding $\arg L(\sigma+i T,\chi)|_{\sigma=\sigma_0}^{1/2}$. Is there a better bound on $\arg L(\sigma,\chi)|_{\sigma=1/2}^{\sigma_0}$ than what one would get just by setting $T=0$?

  • $\begingroup$ Important self-correction: I should have said $\arcsin$, not $\arctan$. You still get a tighter upper bound. $\endgroup$ Dec 6, 2021 at 7:48
  • $\begingroup$ And yes, by the linear independence of $\pi$ and $\log 2, \log 3,\dotsc,\log p$ over $\mathbb{Q}$, $t$ and $\chi$ can conspire, and so the bound is tight, for $t$ and $\chi$ unbounded. (As @juan points out below - use Kronecker's theorem.) $\endgroup$ Dec 6, 2021 at 7:50

1 Answer 1


For the first question about $2\log\zeta(\sigma_0)|$, I think the reasoning is this:

For $\sigma>2$ we have $$|L(s,\chi)-1|\le \sum_{n=2}^\infty\frac{1}{n^\sigma}=\zeta(\sigma)-1<1.$$ Hence $\log L(s,\chi)$ can be defined by the power series $$\log L(s,\chi)=-\sum_{k=1}^\infty \frac{1}{k}(1-L(s,\chi))^k$$ In particular $|\Im \log L(2+it,\chi)|<\pi/2$ and coincide with $\arg(L(2+it,\chi))$. Also this is equal to $$\log L(2+it)=\sum_{n=2}^\infty \frac{\Lambda(n)}{\log n}\frac{\chi(n)}{n^{2+it}}$$ It follows that $$|\arg L(2+it)|\le |\log L(2+it)|\le \sum_{n=2}^\infty \frac{\Lambda(n)}{\log n}\frac{1}{n^{2}}=\log\zeta(2).$$ Now it follows that $$|\arg L(2+it,\chi)-\arg L(2,\chi)|\le 2\log\zeta(2).$$

  • $\begingroup$ Oh, I know how to derive that - I was just pointing out that (a) one can easily do better, (b) I don't know how to get rid of the factor of $2$ (and one most likely can't). $\endgroup$ Dec 4, 2021 at 17:48
  • $\begingroup$ The mystery is why people don't give a better bound :). $\endgroup$ Dec 4, 2021 at 17:49
  • 1
    $\begingroup$ @Helfgott I misunderstood you, I was surprised that you were unaware of this. I think the factor 2 can not be eliminated, perhaps using Kronecker's theorem we can prove this. $\endgroup$
    – juan
    Dec 4, 2021 at 18:09
  • $\begingroup$ That's also my impression. $\endgroup$ Dec 5, 2021 at 1:07

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.