New separation theorems and sub-exponential time algorithms for packing and piercing of fat objects

Abstract: For $\cal C$ a collection of $n$ objects in $R^d$, let the packing and piercing numbers of $\cal C$, denoted by $Pack({\cal C})$, and $Pierce({\cal C})$, respectively, be the largest number of pairwise disjoint objects in ${\cal C}$, and the smallest number of points in $R^d$ that are common to all elements of ${\cal C}$, respectively. When elements of $\cal C$ are fat objects of arbitrary sizes, we derive sub-exponential time algorithms for the NP-hard problems of computing ${Pack}({\cal C})$ and $Pierce({\cal C})$, respectively, that run in $n^{O_d({{Pack}({\cal C})}^{d-1\over d})}$ and $n^{O_d({{Pierce}({\cal C})}^{d-1\over d})}$ time, respectively, and $O(n\log n)$ storage. Our main tool which is interesting in its own way, is a new separation theorem. The algorithms readily give rise to polynomial time approximation schemes (PTAS) that run in $n^{O({({1\over\epsilon})}^{d-1})}$ time and $O(n\log n)$ storage. The results favorably compare with many related best known results. Specifically, our separation theorem significantly improves the splitting ratio of the previous result of Chan, whereas, the sub-exponential time algorithms significantly improve upon the running times of very recent algorithms of Fox and Pach for packing of spheres.