Computation and theory of Mordell-Tornheim-Witten sums. II. (Q2346300)

The multidimensional Mordell-Tornheim-Witten (MTW) zeta function \[ \omega(s_1,\dots,s_{K+1})=\sum_{m_1,\dots,m_K>0}\frac{1}{m_1^{s_1}\cdots m_K^{s_K}(m_1+\cdots+m_K)^{s_{K+1}}} \] has many connections with combinatorics, number theory and even mathematical physics. The previous results concentrate mainly on the evaluation of this function but not on the derivatives. In this paper, the authors push the question forward and ask for relations for the derivative of the generalized MTW zeta function \[ \omega(s_1,\dots,s_M\mid t_1,\dots,t_N)=\sum_{m_1,\dots,m_M,n_1,\dots,n_N>0}\prod_{j=1}^M\frac{1}{m_j^{s_j}}\prod_{k=1}^N\frac{1}{n_k^{t_k}}, \] where for the summation indices the additional restriction \(\sum_{j=1}^M m_j=\sum_{k=1}^N n_k\) is taken. It is proven that in some cases these derivatives can be reduced to lower depth and, ultimately, to multiple zeta values. Also, some numerical relations among the different derivative values are discovered. One of the main tools is the PSLQ algorithm. The underlying arithmetic in this algorithm often requires to find numerical approximations to several hundred digits or even more. The authors carefully write down how such a precision can be attained for all the quantities they needed during their research: Riemann zeta values and its derivatives at positive integer arguments, derivatives of polylogarithms, the Gamma function, etc. It is interesting to note that the authors had to develop some new computer routines to perform their calculations as the most well known commercial software packages were unable to offer the sufficient numerical precision. The paper closes with some challenges to be overcome in the future. For Part I, see the first author et al. [Math. Comput. 83, No. 288, 1795--1821 (2014; Zbl 1296.33004)].