Evaluation of double \(L\)-values. (Q1429812)

The double \(L\)-values in question are defined by \[ L(\chi,\psi;a,b)=\lim_{N\to\infty} L_N(\chi,\psi;a,b)= \lim_{N\to\infty}\sum_{0<m<n\leq N} {\chi(m)\psi(n)\over m^an^b} \] where \(\chi,\psi\) are Dirichlet characters of conductor \(D,E\) respectively and \(a,b\) are positive integers with \(a+b\geq 3\). Set \(m=\text{ lcm}(D,E,\phi(D),\phi(E)), F=\text{ lcm}(D,E)\) and suppose that either \(b>1\) or \(\psi\neq1\) and that \(\chi\psi(-1)=(-1)^{a+b-1}\). The evaluation of the double L-value is a finite sum \(\sum a_jl_jm_j\) where each \(l_j,m_j\) is either an \(L\)-series value or a value of a polylogarithm at an \(F\)th root of unity and each \(a_j\) is in \({\mathbb Q}(\zeta_m)\). The idea of the derviation is to generate \({\mathbb Z}\)-linear combinations of \(L_N\) and related double sums which are equal to linear combinations of products of the polylogarithms \(Li_F(\zeta_m)\) plus error terms. The relations come from the double shuffle \[ L_N(\chi,\psi;a,b)+L_N(\psi,\chi;b,a)+\sum_{n=1}^N {\chi(n)\psi(n)\over n^{a+b}} =L_N(\chi,a)L_N(\psi,b) =\sum_{m,n=1}^N{\chi(m)\psi(n)\over m^an^b} \] and repeated use of the partial fraction decomposition \[ {1\over m^a(m-n)^b}=\sum_{c+d=a+b} \left((-1)^b{d-1\choose b-1}{1\over m^cn^d} + (-1)^{a+c}{c-1\choose a-1}{1\over n^c(m-n)^d}\right). \] The approach yields rapidly converging series for double \(L\)-values which led to the discovery of many formulae which were subsequently derived from the evaluation theorem. As just one example, with \(\chi(n)=\left({-3\over n}\right)\), \[ L(\chi,1;1,3)={1\over2}\log3L_\chi(3)+{13\over9}L_\chi(1)\zeta(3) -{1\over3}L_\chi(2)\zeta(2)-L_\chi(4). \]