Approximating the Simplicial Depth

Abstract

Let $P$ be a set of $n$ points in $d$-dimensions. The simplicial depth, $\sigma_P(q)$ of a point $q$ is the number of $d$-simplices with vertices in $P$ that contain $q$ in their convex hulls. The simplicial depth is a notion of data depth with many applications in robust statistics and computational geometry. Computing the simplicial depth of a point is known to be a challenging problem. The trivial solution requires $O(n^{d+1})$ time whereas it is generally believed that one cannot do better than $O(n^{d-1})$. In this paper, we consider approximation algorithms for computing the simplicial depth of a point. For $d=2$, we present a new data structure that can approximate the simplicial depth in polylogarithmic time, using polylogarithmic query time. In 3D, we can approximate the simplicial depth of a given point in near-linear time, which is clearly optimal up to polylogarithmic factors. For higher dimensions, we consider two approximation algorithms with different worst-case scenarios. By combining these approaches, we compute a $(1+\varepsilon)$-approximation of the simplicial depth in time $\tilde{O}(n^{d/2 + 1})$ ignoring polylogarithmic factor. All of these algorithms are Monte Carlo algorithms. Furthermore, we present a simple strategy to compute the simplicial depth exactly in $O(n^d \log n)$ time, which provides the first improvement over the trivial $O(n^{d+1})$ time algorithm for $d>4$. Finally, we show that computing the simplicial depth exactly is #P-complete and W[1]-hard if the dimension is part of the input.