4
$\begingroup$

Is there by any chance anything resembling a closed expression for, say, the integral $$I = \frac{1}{2 \pi} \int_{-\infty}^\infty \frac{dt}{|\zeta(1+i t)|^2 t^2} ?$$ It is easy to show (by Plancherel) that $$I = \int_0^\infty |m(x)|^2 \frac{d x}{x}$$ for $m(x):=\sum_{n\leq x} \mu(n)/n$, but I'd like to know whether there is something "more closed" than that.

$\endgroup$
10
  • $\begingroup$ Note that this has been crossposted to MSE: math.stackexchange.com/q/4234446/9754 $\endgroup$ Commented Aug 27, 2021 at 16:11
  • $\begingroup$ (Should I have stated that, and, if so, how?) $\endgroup$ Commented Aug 27, 2021 at 16:21
  • $\begingroup$ Generally I'd recommend choosing the site that seems like the best fit and just posting it there. (In this case, that seems like MO). If that doesn't yield answers after a bit, then perhaps ask somewhere else and link between them. This is basically what's recommended here. Some people are really sensitive to people posting the same question on different forums, thinking that they're quickly trying to get an answer and not worrying that this might encourage duplication of effort. $\endgroup$ Commented Aug 27, 2021 at 16:32
  • 5
    $\begingroup$ Given that the finiteness of this integral implies the prime number theorem, it is unlikely that there is going to be a closed form for this expression that is manifestly finite without the assistance of this theorem, unless you are willing to allow expressions that depend on the location of the zeroes of the zeta function as a "closed form". $\endgroup$
    – Terry Tao
    Commented Aug 27, 2021 at 22:11
  • 3
    $\begingroup$ Sure, but they are pretty disgusting: you've already basically expressed your integral in terms of the (logarithmically weighted) Mertens function, and that function can be written (assuming simple zeroes) as a sum over zeroes (but with coefficients that are reciprocals of products of differences of zeroes, so not super pleasant to deal with). After interchanging some sums you should be able to write your expression as a sum over pairs of zeroes of some nasty product over zeroes. Can't say that this will be too enlightening, though. $\endgroup$
    – Terry Tao
    Commented Aug 27, 2021 at 22:15

1 Answer 1

1
$\begingroup$

This is not properly an answer, after the comments of Tao it is difficult to give an answer. Only an explanation of my comment above. I still think that my series and the integral are equal.

We can write the integral $I$ in the form $$I=\frac{1}{2\pi}\int_{-\infty}^\infty\frac{1}{\zeta(1+it)\zeta(1-it)}\frac{dt}{t^2}.$$ Hence I consider the function $$u(\sigma)=\frac{1}{2\pi}\int_{-\infty}^\infty\frac{1}{\zeta(\sigma+it)\zeta(\sigma-it)}\frac{dt}{(\sigma-1)^2+t^2},\qquad \sigma>1.$$ I expect to have $\lim_{\sigma\to1^+}u(\sigma)=I$. For $\sigma>1$ we may write $$u(\sigma)=\frac{1}{2\pi}\sum_{a=1}^\infty\sum_{b=1}^\infty\frac{\mu(a)\mu(b)}{(ab)^\sigma}\int_{-\infty}^\infty(b/a)^{it}\frac{dt}{(\sigma-1)^2+t^2}.$$ So that $$u(\sigma)=\frac{1}{2\pi}\sum_{a=1}^\infty\sum_{b=1}^\infty\frac{\mu(a)\mu(b)}{(ab)^\sigma}\frac{\pi e^{-(\sigma-1)|\log(b/a)|}}{\sigma-1},$$ $$u(\sigma)-\frac{1}{2\zeta(\sigma)^2(\sigma-1)}= \frac{1}{2}\sum_{a=1}^\infty\sum_{b=1}^\infty\frac{\mu(a)\mu(b)}{(ab)^\sigma}\frac{e^{-(\sigma-1)|\log(b/a)|}-1}{\sigma-1},$$ and $$u(\sigma)-\frac{1}{2\zeta(\sigma)^2(\sigma-1)}=- \frac{1}{2}\sum_{a=1}^\infty\sum_{b=1}^\infty\frac{\mu(a)\mu(b)}{(ab)^\sigma} |\log(b/a)|\int_0^1 e^{-(\sigma-1)|\log(b/a)|x}\,dx.$$ It is not easy here to justify to take limit for $\sigma\to1^+$ term by term, but if correct we will get $$I=-\frac{1}{2}\sum_{a=1}^\infty\sum_{b=1}^\infty\frac{\mu(a)\mu(b)}{ab} |\log(b/a)|=\sum_{n=1}^\infty\frac{f(n)}{n}.$$ With $$f(n):=-\frac12\sum_{ab=n}\mu(a)\mu(b)|\log(b/a)|.$$ The function $f(n)=0$ except if $n=mk^2$ with $|\mu(mk)|=1$. In this case $f(n)=f(m)$. For $m$ squarefree $f(m)$ have the sign of $-\mu(m)$ multiplied by a logarithm of a number with prime divisors dividing $m$. But this number depends of the relative size of the divisors of $m$. For example with $p<q<r$ primes $f(pqr)$ can be equal to $2\log (pqr)$ or $3\log r$ according to $pq>r$ or $pq<r$.

$\endgroup$
5
  • $\begingroup$ Interesting! Does this check out numerically? $\endgroup$ Commented Aug 28, 2021 at 11:57
  • $\begingroup$ Your expression for $I$ can also be written as $\sum_n \frac{\mu(n)}{n} \check{m}(a)$, where $\check{m}(a) = \sum_{m\leq a} \frac{\mu(m)}{m} \log \frac{a}{m}$. Since $\check{m}(a) = 1 + O((\log a)^{-C})$, it is clear that the sum converges. $\endgroup$ Commented Aug 28, 2021 at 12:14
  • $\begingroup$ @Helfgott I computed the integral aproximately with Mathematica for $2\times 10^{-6}<|t|<2\times 10^6$ obtaining $0.81492827$ and then my sum oscillates, but the mean of the sums to 300000 give me the value $0.814346$. This is not too good, it is what I can say. $\endgroup$
    – juan
    Commented Aug 28, 2021 at 12:17
  • $\begingroup$ @Helfgott When I say the sum oscillates, it oscillates around 0.8 say, so that I took the mean of the sums to get an idea of the possible limit. $\endgroup$
    – juan
    Commented Aug 28, 2021 at 12:24
  • $\begingroup$ I'd expect $\sum_{n\leq x} (\mu(n)/n) \widehat{m}(a)$ to converge more quickly, but not necessarily more quickly than $O(1/\sqrt{x})$. $\endgroup$ Commented Aug 28, 2021 at 12:41

You must log in to answer this question.

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