On finite-precision representations of geometric objects (Q584309)

Motivated by the geometric problem of the approximations of a hyperplane \(a_ 1x_ 1+...+a_ nx_ n+a_{n+1}=0\) by such a hyperplane with integral coefficients of absolute value \(\leq Q\) (given number), the author discusses the following problem: Given n reals \(w_ 1,...,w_ n\) and a positive integer Q, find \(n+1\) integers \(p_ 1,...,p_ n\) and q such that \(| q| \leq Q\) and the error \(\max_ i | w_ i-P_ i/q|\) is small. An experimental study of several algorithms for the solution of this problem is presented. One algorithm is based on continued fraction expansion and one on basis reduction as proposed by Lenstra-Lenstra-Lovasz.