• DOI:10.20382/jocg.v7i2a4
• Corpus ID: 9667940

Approximability of the discrete Fréchet distance

@inproceedings{Bringmann2015ApproximabilityOT,
  title={Approximability of the discrete Fr{\'e}chet distance},
  author={Karl Bringmann and Wolfgang Mulzer},
  booktitle={Journal of Computational Geometry},
  year={2015}
}

• K. Bringmann, Wolfgang Mulzer
• Published in
  Journal of Computational…
  11 June 2015
• Computer Science, Mathematics

The Frechet distance is a popular and widespread distance measure for point sequences and for curves. About two years ago, Agarwal et al. [SIAM J. Comput. 2014] presented a new (mildly) subquadratic algorithm for the discrete version of the problem. This spawned a flurry of activity that has led to several new algorithms and lower bounds. In this paper, we study the approximability of the discrete Frechet distance. Building on a recent result by Bringmann [FOCS 2014], we present a new conditional lower bound showing that strongly subquadratic algorithms for the discrete Frechet distance are unlikely to exist, even in the one-dimensional case and even if the solution may be approximated up to a factor of 1.399. This raises the question of how well we can approximate the Frechet distance (of two given $d$-dimensional point sequences of length $n$) in strongly subquadratic time. Previously, no general results were known. We present the first such algorithm by analysing the approximation ratio of a simple, linear-time greedy algorithm to be $2^{\Theta(n)}$. Moreover, we design an $\alpha$-approximation algorithm that runs in time $O(n\log n + n^2/\alpha)$, for any $\alpha\in [1, n]$. Hence, an $n^\varepsilon$-approximation of the Frechet distance can be computed in strongly subquadratic time, for any $\varepsilon > 0$.