On the covering radius of lattice zonotopes and its relation to view-obstructions and the lonely runner conjecture (Q525086)

It is known that following three geometric problems are equivalent: a) billiard ball motion inside a cube avoiding an inner cube; b) lines in a multidimensional torus avoiding an inner cube; c) view unobstructed by a lattice arrangement of cubes. In this setting consider the quantity \(\varepsilon(n,m)\) that equals to the supremum of \(\varepsilon\) such that the line \(\{u_0+t\alpha: t\in\mathbb{R}\}/\mathbb{Z}^m\) in \(\mathbb{T}^m\) intersects \([\varepsilon ;1-\varepsilon ]^m\) for every \(u_0\in\mathbb{R}^m\) and every \(\alpha\in(\mathbb{R}^m\setminus \{0\})^m\) with \(\dim_{\mathbb{Q}}\alpha\geq m-n\). Here, \(\dim_{\mathbb{Q}}\alpha\) is the dimension of \(\mathbb{Q}\)-vector space generated by the coordinates of \(\alpha\). Obviously, \(\varepsilon(n,m)\) also can be defined in the terms of two other geometric problems. Further, let \(\overline{\varepsilon}(n,m)\) be an analogue of \(\varepsilon(n,m)\) with the following supplementary condition: every \(\dim_{\mathbb{Q}}\alpha\) coordinates of \(\alpha\) are linearly independent over \(\mathbb{Q}\). In the paper, a new approach to studying of \(\varepsilon(n,m)\) and \(\overline{\varepsilon}(n,m)\) is introduced. This approach is based on the reformulation of the problem in the terms of covering radius of some special lattice zonotope. It is proved that \(\overline{\varepsilon}(n,m)\leq \frac{1}{2(n+1)}\) and \(\frac{m-O(n\log n)}{2(m-n+1)}\leq \varepsilon(n,m)\leq \frac{m-n}{2(m-n+1)}\). The most nontrival part of this result is a bound from below. Its proof is based on one Khinchin's result from the diophantine approximation theory. Now, let \(\varepsilon_0(n,m)\) and \(\overline{\varepsilon}_0(n,m)\) be analogues of \(\varepsilon(n,m)\) and \(\overline{\varepsilon}(n,m)\) with the supplementary condition \(u_0=0\). Then, some nontrivial bounds for \(\varepsilon_0(n,m)\) and \(\overline{\varepsilon}_0(n,m)\) are established. Note that the equations \(\varepsilon_0(m-1,m)=\overline{\varepsilon}_0(m-1,m)= \frac{1}{m+1}\) are equivalent to the famous lonely runner conjecture. The methods of the paper do not give new estimates for this conjecture but give its new reformulation in the terms of covering radii of some zonotopes.