Is it possible to simplify $\psi^{(2)}(\frac18)$, where $\psi$ denotes the polygamma function?
Or more generalized, $\psi^{(2)}(\frac pq)$ and $\psi^{(2n)}(\frac pq)$?

Noticing there is a formula for $\psi(\frac pq)$, where $p,q\in\mathbb{N}$ and the following formulae:
$\psi^{(2)}(\frac12)=-14\zeta(3)$, $\psi^{(2)}(\frac13)=-26\zeta (3)-\frac{4\pi^3}{3\sqrt{3}}$, $\psi^{(2)}(\frac14)=-2\pi^3-56\zeta(3)$ and $\psi^{(2)}(\frac16)=-182\zeta (3)-4\sqrt{3}\pi^3$, it is natural to ask if there are general formulae.
$$\psi^{(2)}\left(\frac pq\right)=\int_0^1\frac{\ln^2 t}{t-1}t^{p/q-1}dt\\=q^3\int_0^1\frac{\ln^2 u}{u^q-1}u^{p-1}du$$ I tried to use contour integration to solve this integral but failed.

  • 1
    $\begingroup$ Removing the syntactic sugar, the OP is asking for a closed form for $$\sum_{n\geq 0}\frac{1}{(qn+p)^3}.$$ $\endgroup$ – Jack D'Aurizio Oct 12 '18 at 19:51

This is an answer showing that why $\psi^{(2n)}\left(\frac pq\right)$ has a closed-form when $q=2,3,4,6$ and conjecturing it does not have a general closed-form when $p=5$ or $p\ge7$. (Assuming $L(s,\chi)$ is not a closed-form)
1.$L(2n+1,\chi)$ has a closed-form when $\chi(-1)=-1$
(Where $L$ denotes the $L$-series)
First, denote $f(x,\chi)$ the analytic continuation of $\displaystyle\sum_{n=0}^\infty \chi(n)x^n$. In this question, I showed that $f(x)=-\chi(-1)f\left(\frac1x\right)$.
Noticing $$\int_0^1 x^{n-1}(-\ln x)^{s-1}\frac{dx}{\Gamma(s)}=n^{-s},$$ one can prove that it is valid to change the position of $\sum$ and $\int$: $$L(2s+1,\chi)=\int_0^1\sum_{n=0}^\infty\chi(n)x^{n-1}(-\ln x)^{2s}\frac{dx}{\Gamma(s)}\\ =\frac{1}{2\Gamma(s)}\int_0^\infty\frac{f(x)}{x}(-\ln x)^{2s}dx$$ Since $f(x)$ can be expressed in form $\frac{P(x)}{Q(x)}$, where $P$ and $Q$ are polynomial funtions, the integral has a closed-form according to Residue Theorem.
2.$L(n,\chi)$ has a closed-form when $\chi$ is a principal character
If $n$ is a prime, it is trivial.
If $n=p_1^{\alpha_1}\cdots p_{\tau}^{\alpha_\tau}$, it is always possible to add and subtract some terms to make the series $\zeta(n)$. Furthermore, $L(n,\chi)=\frac pq\zeta(n)$
3.It is possible to deduce $\psi^{(2n)}\left(\frac pq\right)$ from the values of $L$-series when $q=2,3,4,6$.
For simpleness, I will discuss $\zeta\left(n,\frac pq\right)$ instead of $\psi$ here.
$3^nL(n,\chi_{3,1})=\zeta(n,\frac13)+\zeta(n,\frac23)$, $3^nL(n,\chi_{3,2})=\zeta(n,\frac13)-\zeta(n,\frac23)$
$4^nL(n,\chi_{4,1})=\zeta(n,\frac14)+\zeta(n,\frac34)$, $4^nL(n,\chi_{4,2})=\zeta(n,\frac14)-\zeta(n,\frac34)$
$6^nL(n,\chi_{6,1})=\zeta(n,\frac16)+\zeta(n,\frac56)$, $6^nL(n,\chi_{6,2})=\zeta(n,\frac16)-\zeta(n,\frac56)$
Solving this simultaneous equation gives the values of $\zeta\left(n,\frac pq\right)$. (The values of $L$-series in LHS is known when $n$ is odd)
4.Conjecturing it does not have a general closed-form when $p=5$ or $p\ge7$
This post shows that $\varphi(n)>2$ when $n>6$. It is also obvious that $\varphi(5)>2$.
There is $\frac12\varphi(k)+1$ values known in $L(2s+1,\chi)$, and we have $\varphi(k)$ variables. Since $\varphi(k)>2$ for $k=5$ and $k>6$, $\varphi(k)>\frac12\varphi(k)+1$. It is not possible to deduce $\zeta(2s+1,\frac pq)$ via $L$-series.


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.