Dice Question Streamline Icon: https://streamlinehq.com

Fast approximation algorithms for 1D EMDuT

Determine whether there exist constant‑factor or (1+ε)‑approximation algorithms for Earth Mover’s Distance under Translation (EMDuT) on the real line that run in strongly subquadratic time O(n^{2−δ}) for some constant δ>0 or in near‑linear time, and specifically whether a (1+ε)‑approximation can be computed in time \widetilde{O}(n^{2−δ}/poly(ε)) for some constant δ>0 (independent of n and ε) or even in time \widetilde{O}(n/poly(ε)).

Information Square Streamline Icon: https://streamlinehq.com

Background

The paper gives near‑quadratic algorithms and OVH‑based lower bounds for asymmetric EMDuT in one dimension, ruling out (1+ε)‑approximations in O(n{2−δ}/ε{o(1)}) time for any constant δ>0 when |B|=Ω(n). However, the existence of faster approximation algorithms in 1D remains unresolved.

This question targets whether one can beat the quadratic barrier with either constant‑factor approximations or (1+ε)‑approximations, potentially even achieving near‑linear time with controlled ε‑dependence.

References

Over \mathbb{R}1, we leave open whether there are fast approximation algorithms: Can a constant-factor approximation be computed in time O(n{2-\delta}) for some constant \delta > 0? Or even in time \widetilde{O}(n)? Can a (1+\varepsilon)-approximation be computed in time \widetilde{O}(n{2-\delta} / \textup{poly}(\varepsilon)) for some constant \delta > 0 (independent of n and \varepsilon)? Or even in time \widetilde{O}(n / \textup{poly}(\varepsilon))?

Fine-Grained Complexity of Earth Mover's Distance under Translation (2403.04356 - Bringmann et al., 7 Mar 2024) in Section Open problems, subparagraph “EMDuT in 1D”