The algorithmic Fried Potato Problem in two dimensions
Francisco Criado, Francisco Santos
TL;DR
This work addresses Conway's Fried Potato Problem in two dimensions, seeking a division of a convex polygon into $n$ parts via $n-1$ cuts that minimizes the largest inradius. Building on Bezdek–Bezdek's result that the optimum is realized by $n-1$ parallel cuts, the authors reduce the problem to finding a distance parameter $\rho$ and a direction $v$ such that $\operatorname{width}(P^{\rho})=2n\rho$, and develop a quasi-linear algorithm to compute these in $O(m \log^4 m)$ time for polygons with $m$ edges. A key innovation is the 3D dome $\widetilde{P}$ construction and its efficient preprocessing via a Dobkin–Kirkpatrick hierarchy, enabling fast linear-programming queries on the dome and its slices. This yields a substantial improvement over prior quadratic-time approaches and provides a practical framework for efficiently solving the Baker's Potato problem in the plane.
Abstract
Conway's Fried Potato Problem seeks to determine the best way to cut a convex body in $n$ parts by $n-1$ hyperplane cuts (with the restriction that the $i$-th cut only divides in two one of the parts obtained so far), in a way as to minimize the maxuimum of the inradii of the parts. It was shown by Bezdek and Bezdek that the solution is always attained by $n-1$ parallel cuts. But the algorithmic problem of finding the best direction for these parallel cuts remains. In this note we show that for a convex $m$-gon $P$, this direction (and hence the cuts themselves) can be found in time $O(m \log^4 m)$, which improves on a quadratic algorithm proposed by Cañete-Fernández-Márquez (DMD 2022). Our algorithm first preprocesses what we call the dome (closely related to the medial axis) of $P$ using a variant of the Dobkin-Kirkpatrick hierarchy, so that linear programs in the dome and in slices of it can be solved in polylogarithmic time.
