Interested in the following function:

$$ \Psi(s)=\sum_{n=2}^\infty \frac{1}{\pi(n)^s}=\sum_{n=1}^\infty \frac{\lambda_n}{n^s}, $$

where $\pi(n)$ is the prime counting function.

Was thinking about:

$$ \bar L(s,\chi)=\sum_{n=1}^\infty\frac{|\chi(n)|^2}{n^s}, $$


$$ L_k(s,\chi)=\sum\limits_{n=1}^\infty \frac{|\chi(n)|^{2k}}{n^s}. $$

However, I don't think these can yield non-periodic integer sequences in the numerator because $\exists k\in\Bbb Z^+: \chi(n)=\chi(n+k)\,\forall n. $

So to achieve non periodicity I settled on modifying the denominator, specifically changing $n$ to $\pi(n)$:

$$ \Psi(s)=\sum_{n=2}^\infty \frac{1}{\pi(n)^s}=1+\frac{1}{2^s}+\frac{1}{2^s}+\frac{1}{3^s}+\frac{1}{3^s}+\frac{1}{4^s}+\frac{1}{4^s}+... $$

and combining like terms:

$$ \Psi(s)=\sum_{n=1}^\infty \frac{\lambda_n}{n^s}=1+\frac{2}{2^s}+\frac{2}{3^s}+\frac{4}{4^s}+\frac{2}{5^s}+\frac{4}{6^s}+\frac{2}{7^s}+... $$

So far I've computed the non-periodic sequence, $\lambda_n$, to $34$ terms:


Just found that $\lambda_n$ is sequence A001223 in the oeis; Prime gaps: differences between consecutive primes.


Does this sum converge for all $Re(s)>1$?

Is there a closed form for the sums?

Can $\Psi(s)$ be analytically continued? If so, how?

Where does $\Psi(s)=0$?

Is there a Euler product for $\Psi(s)?$

  • 5
    $\begingroup$ It definitely converges for $Re(s)>1$ since $\pi(n)\geq\frac12\cdot\frac{n}{\log n}$ for all $n$ sufficiently large. In particular, the summand is $\ll \frac{(\log n)^s}{n^s}\leq\frac{1}{n^{s-\varepsilon}}$. $\endgroup$ – Clayton Apr 25 at 21:16
  • $\begingroup$ oh okay awesome $\endgroup$ – Ultradark Apr 25 at 21:17
  • $\begingroup$ Almost surely the other three questions do not have tidy answers. Although it definitely has a singularity at $s=1$ by Landau's theorem. $\endgroup$ – Greg Martin Apr 25 at 21:25
  • $\begingroup$ Yeah I figured they would be difficult questions, but definitely attainable $\endgroup$ – Ultradark Apr 25 at 21:26

$$F(s)=\sum_{m=2}^\infty \pi(m)^{-s}=\sum_{n=1}^\infty n^{-s} (p_{n+1}-p_n)= \sum_{n=1}^\infty p_{n+1} (n^{-s}-(n+1)^{-s})\\ =\sum_{n=1}^\infty n \ln n (1+o(1)) (s n^{-s-1}+O(s(s+1)n^{-s-2})$$ so it converges and it is analytic for $\Re(s) > 1$ and as $s \to 1$, $F(s) \sim -s\zeta'(s) \sim \frac{1}{(s-1)^2}$.

For the analytic continuation under the RH $n=\pi(p_n) = Li(p_n)+O(p_n^{1/2+\epsilon})$ thus $p_n = Li^{-1}(n+O(n^{1/2+\epsilon}))=Li^{-1}(n)+O(n^{1/2+\epsilon})$ and $F(s)-s\sum_{n=1}^\infty n^{-s-1} Li^{-1}(n)$ is analytic for $\Re(s) > 1/2$.

So it is now a problem about the asymptotic expansion of $Li^{-1}$

  • $\begingroup$ Thank you, could you elaborate on the last remark? $\endgroup$ – Ultradark Apr 25 at 22:14
  • 1
    $\begingroup$ @Ultradark You can start with a 1,2 or 3 terms asymptotic for $Li(x)$ to deduce a 3 terms asymptotic for its inverse (and show us what you get). The question is if the remainder is $O(x^a)$ with $a < 1$ or not $\endgroup$ – reuns Apr 25 at 22:20
  • $\begingroup$ why is the question if the remainder is $O(x^a)$ with $a<1$ $\endgroup$ – Ultradark Apr 28 at 0:21

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.