Computational Complexity of the $α$-Ham-Sandwich Problem
The classic Ham-Sandwich theorem states that for any $d$ measurable sets in $\mathbb{R}^d$, there is a hyperplane that bisects them simultaneously. An extension by Bárány, Hubard, and Jerónimo [DCG 2008] states that if the sets are convex and \emph{well-separated}, then for any given $α_1, \dots, α_d \in [0, 1]$, there is a unique oriented hyperplane that cuts off a respective fraction $α_1, \dots, α_d$ from each set. Steiger and Zhao [DCG 2010] proved a discrete analogue of this theorem, which we call the \emph{$α$-Ham-Sandwich theorem}. They gave an algorithm to find the hyperplane in time $O(n (\log n)^{d-3})$, where $n$ is the total number of input points. The computational complexity of this search problem in high dimensions is open, quite unlike the complexity of the Ham-Sandwich problem, which is now known to be PPA-complete (Filos-Ratsikas and Goldberg [STOC 2019]). Recently, Fearley, Gordon, Mehta, and Savani [ICALP 2019] introduced a new sub-class of CLS (Continuous Local Search) called \emph{Unique End-of-Potential Line} (UEOPL). This class captures problems in CLS that have unique solutions. We show that for the $α$-Ham-Sandwich theorem, the search problem of finding the dividing hyperplane lies in UEOPL. This gives the first non-trivial containment of the problem in a complexity class and places it in the company of classic search problems such as finding the fixed point of a contraction map, the unique sink orientation problem and the $P$-matrix linear complementarity problem.