Using Interval Arithmetic for Determining the Structure of Convex Hulls

Abstract

We discuss the use of interval arithmetic in the computation of the convex hull of n points in D dimensions. Convex hull algorithms rely on simple geometric tests, such as “does some point p lie in a certain half-space or affine subspace?” to determine the structure of the hull. These tests in turn can be carried out by solving appropriate (not necessarily square) linear systems. If interval-based methods are used for the solution of these systems then the correct hull can be obtained at lower cost than with exact arithmetic. In addition, interval-based methods can determine the topological structure of the convex hull even if the position of the points is not known exactly. In the present paper we compare various interval linear solvers with respect to their ability to handle close-to-pathological situations. This property determines how often interval arithmetic cannot provide the required information and therefore some computations must be redone with exact arithmetic.