L 1 Approximation and the Analysis of Data

Abstract

When estimating the true form of experimental data that contain some very inaccurate points, the assumption that L1 approximation is superior to L2 approximation can easily be justified theoretically. This is an old idea, but since best L1 approximations have only recently been readily obtained, we present here some numerical evidence with which to judge the extent of the improvement over orthodox least‐squares approximations.