Title:Strong equivalence between metrics of Wasserstein type

Abstract:The sliced Wasserstein and more recently max-sliced Wasserstein metrics $\mW_p$ have attracted abundant attention in data sciences and machine learning due to its advantages to tackle the curse of dimensionality. A question of particular importance is the strong equivalence between these projected Wasserstein metrics and the (classical) Wasserstein metric $\Wc_p$. Recently, Paty and Cuturi have proved the strong equivalence of $\mW_2$ and $\Wc_2$. We show that the strong equivalence also holds for $p=1$, while we show that the sliced Wasserstein metric does not share this nice property.