Values of Dirichlet \(L\)-functions, Gauss sums and trigonometric sums (Q409025)

Motivated by theta-function identities from Ramanujan's notebooks, \textit{B. C. Berndt} and \textit{A. Zaharescu} [Math. Ann. 330, No. 3, 551--575 (2004; Zbl 1099.11039)] discovered several families of trigonometric identities involving class numbers of imaginary quadratic fields. The ``mother of these identities,'' going back further to Dirichlet, Lebesgue, and Schemmel is \[ \sum_{ n=1 }^{ k-1 } \chi(n) \cot {{ \pi n } \over { k }} = 2 \sqrt k \, h(-k) \, , \] where \(\chi\) is an odd, real, primitive character modulo \(k\) (which is odd and \(\geq 7\)), and \(h(-k)\) denotes the class number of \({\mathbb Q}(\sqrt{-k})\). Berndt and Zaharescu's identities have been extended in various directions, and the paper under review continues this line of study by establishing trigonometric identities involving evaluations of Dirichlet \(L\)-functions \(L(s,\chi) := \sum_{ n \geq 1 } {{ \chi(n) } \over { n^s }}\) associated to a character \(\chi\), evaluated at some positive integer. A sample identity is \[ \begin{multlined} {{ \pi^7 } \over { k^7 }} \sum_{ m=1 }^{ k-1 } \chi(m) \left({{ 31 \cos ({{ \pi m } \over { k }}) } \over { 45 \sin^3 ({{ \pi m } \over { k }}) }} - {{ 3 \sin ({{ 2 \pi m } \over { k }}) } \over { \sin^4 ({{ \pi m } \over { k }}) }} - {{ 13 \cos ({{ 3 \pi m } \over { k }}) } \over { 3 \sin^5 ({{ \pi m } \over { k }}) }} + {{ 5 \sin ({{ 4\pi m } \over { k }}) } \over { 2 \sin^6 ({{ \pi m } \over { k }}) }} + {{ \cos ({{ 5 \pi m } \over { k }}) } \over { 2 \sin^7 ({{ \pi m } \over { k }}) }} \right) \\ = L(7,\chi) \, ,\end{multlined} \] where \(\chi\) is any odd Dirichlet character modulo \(k\). The proofs of this identity and its companions start with the following theorem: Let \(\nu \in \{ 0, 1 \}\) and \(\chi\) be a Dirichlet character modulo \(k\) such that \(\chi(-1) = (-1)^\nu\). Denote the \(k\)th Bernoulli number by \(B_k\), and let \[ S(m, \chi) := \sum_{ j = 1 }^k \left( { j \over k } \right)^m \;\sum_{ n = 0 }^{ k-1 } \chi(n) \, e^{ { 2 \pi i n j } \over k }. \] If \(\nu\) and \(r \geq 2\) are integers of opposite parity then \[ \sum_{ q=0 }^{ 2 \lfloor { r \over 2 } \rfloor } \left( { r \atop q } \right) B_q \, S(r-q, \chi) = 0 \, , \] and if \(\nu\) and \(r \geq 1\) have the same parity then \[ \sum_{ q=0 }^{ 2 \lfloor { r \over 2 } \rfloor } \left( { r \atop q } \right) B_q \, S(r-q, \chi) = {{ (-1)^{ \nu + 1 } k r! } \over { i^r 2^{r-1} \pi^r }} \, L(r, \chi) \, . \] Among many other things, one can use this theorem to go full circles by noting that, with \(L(1, \chi) = { \pi \over { \sqrt k }} \, h(-k)\), it implies the above-mentioned classic cotangent identity involving \(h(-k)\).