On the Error of Random Sampling: Uniformly Distributed Random Points on Parametric Curves

Abstract

Given a parametric polynomial curve $\gamma :[a,b]\to {\mathbb{R}}^n$, how can we sample a random point $\mathfrak{x}\in \mathrm{im}(\gamma )$ in such a way that it is distributed uniformly with respect to the arc-length? Unfortunately, we cannot sample exactly such a point-even assuming we can perform exact arithmetic operations. So we end up with the following question: how does the method we choose affect the quality of the approximate sample we obtain? In practice, there are many answers. However, in theory, there are still gaps in our understanding. In this paper, we address this question from the point of view of complexity theory, providing bounds in terms of the size of the desired error.

10 pages, 5 figures, 1 table. 2nd version: New title, major changes

References in corpus (2)

• Log-concave sampling: Metropolis-Hastings algorithms are fast
• Learning Algebraic Varieties from Samples