Institute for Data Analysis and Visualization

We study piecewise linear approximation of quadratic functions defined on Rn. Invariance properties and canonical Caley/Klein metrics that help in understanding this problem can be handled in arbitrary dimensions. However, the problem of optimal approximants in the sense that their linear pieces are of maximal size by keeping a given error tolerance, is a difficult one. We present a detailed discussion of the case n=2, where we can partially use results from convex geometry and discrete geomerty. The case n=3 is considerably harder, and thus just a few results can be formulated so far.