On the Fine-Grained Complexity of Approximating k-Center in Sparse Graphs
Abstract
We study the fine-grained complexity of the famous $k$-center problem in the metric induced by a graph with $n$ vertices and $m$ edges. The problem is NP-hard to approximate within a factor strictly better than $2$, and several $2$-approximation algorithms are known.
Two of the most well-known approaches for the $2$-approximation are (1) finding a maximal distance $r$-independent set (where the minimum pairwise distance is greater than $r$) and (2) Gonzalez's algorithm that iteratively adds the center farthest from the currently chosen centers.
For the approach based on distance-$r$ independent sets, Thorup [SIAM J. Comput. '05] already gave a nearly linear time algorithm. While Thorup's algorithm is not complicated, it still requires tools such as an approximate oracle for neighborhood size by Cohen [J. Comput. Syst. Sci. '97]. Our main result is a nearly straightforward algorithm that improves the running time by an $O(\log n$) factor. It results in an $(2+\eps)$-approximation for $k$-center in $O((m + n \log n)\log n \log(n/\eps))$ time.
For Gonzalez's algorithm [Theor. Comput. Sci. 85], we show that the simple $\widetilde{O}(mk)$-time implementation is nearly optimal if we insist the {\em exact} implementation. On the other hand, we show that an $(1+\eps)$-approximate version of the algorithm is efficiently implementable, leading to an $(2+\eps)$-approximation algorithm in running time $O((m + n \log n)\log^2 n / \eps)$. We also show that, unlike in the distance $r$-independent set-based algorithm, the dependency of $1/\eps$ in the running time is essentially optimal for $(1 + \eps)$-approximate Gonzalez's.
Two of the most well-known approaches for the $2$-approximation are (1) finding a maximal distance $r$-independent set (where the minimum pairwise distance is greater than $r$) and (2) Gonzalez's algorithm that iteratively adds the center farthest from the currently chosen centers.
For the approach based on distance-$r$ independent sets, Thorup [SIAM J. Comput. '05] already gave a nearly linear time algorithm. While Thorup's algorithm is not complicated, it still requires tools such as an approximate oracle for neighborhood size by Cohen [J. Comput. Syst. Sci. '97]. Our main result is a nearly straightforward algorithm that improves the running time by an $O(\log n$) factor. It results in an $(2+\eps)$-approximation for $k$-center in $O((m + n \log n)\log n \log(n/\eps))$ time.
For Gonzalez's algorithm [Theor. Comput. Sci. 85], we show that the simple $\widetilde{O}(mk)$-time implementation is nearly optimal if we insist the {\em exact} implementation. On the other hand, we show that an $(1+\eps)$-approximate version of the algorithm is efficiently implementable, leading to an $(2+\eps)$-approximation algorithm in running time $O((m + n \log n)\log^2 n / \eps)$. We also show that, unlike in the distance $r$-independent set-based algorithm, the dependency of $1/\eps$ in the running time is essentially optimal for $(1 + \eps)$-approximate Gonzalez's.
Research Areas
