Higher-dimensional Dedekind sums and their bounds arising from the discrete diagonal of the \(n\)-cube (Q2490030)

The authors consider higher-dimensional Dedekind sums as a generalization of a one-dimensional probability model of Dilcher and Girstmair to a \(d\)-dimensional cube. More precisely, they define \[ S_d(a; b_1,b_2,\dots, b_d)={1\over a} \sum^{a-1}_{m=0} \Biggl[{mb_1\over a}\Biggr]\cdots\Biggl[{mb_d\over a}\Biggr] \] for positive integers \(a,b_1,\dots, b_d\), where \([x]\) denotes the integral part of a real number \(x\). This sum is easily shown to be equal to \[ {1\over a} \sum^{b_1- 1}_{k_d= 0} k_1\cdots k_d f_{a;b_1,\dots, b_d}(k_1,\dots, k_d), \] where \(f_{a;b_1,\dots, b_d}(k_1,\dots, k_d)\) is the number of diagonal lattice points along the main diagonal of the cube which is constructed by dividing the \(d\)-dimensional cube \([0,a)^d\) into \(b_1\cdots b_d\) equal boxes and labelled by \((k_1,\dots,k_d)\) with \(0\leq k_j\leq b_j- 1\) for each \(j\). For the case \(d= 1\), they also consider the \(k\)th moment \[ M_k(a; b)= {1\over a} \sum^{a-1}_{m=0} \Biggl[{mb\over a}\Biggr]^k, \] which gives \(M_k(a;b)= S_k(a; b,\dots, b)\). For \(k= 2\), this is related to the classical Dedekind sum \(s(a, b)\) by \[ M_2(a; b)= {(b^2+ 1)(a- 1)(2a- 1)\over 6a^2}- {(a- 1)/b\over 2a}- {2b\over a} s(b,a). \] They obtain new formulae for \(M_2(a; b)\) in some cases and provide several types of lower and upper bounds. Also by interpreting \(S_d(a;{\mathbf b})\) as the expected value of the product of random variables \(X^{(a)}_i ({\mathbf m};{\mathbf b})\) with respect to some probability function on a \(d\)-dimensional discrete sample space, they give the upper bounds for \(S_d(a;{\mathbf b})\) in terms of \(M_r\)'s via the Cauchy-Schwarz inequality, and this makes them to consider the ratio \(R_d(a;{\mathbf b})\) of \(S_d(a;{\mathbf b})\) to its upper bound. For the case \(d= 2\), the ratio is defined by \[ R_2(a; b_1,b_2)= {S_2(a; b_1, b_2)\over (M_2(a; b_1)M_2(a; b_2))^{{1\over 2}}}, \] and from empirical evaluations they make a conjecture which is \(R_2(a; b, c)\geq R_2(5; 2,3)= 2\) for all \(a,b,c\geq 3\).