Multifractal Analysis of the Sinkhorn Algorithm: Unveiling the Intricate Structure of Optimal Transport Maps

Abstract

The Sinkhorn algorithm has emerged as a powerful tool for solving optimal transport problems, finding applications in various domains such as machine learning, image processing, and computational biology. Despite its widespread use, the intricate structure and scaling properties of the coupling matrices generated by the Sinkhorn algorithm remain largely unexplored. In this paper, we delve into the multifractal properties of these coupling matrices, aiming to unravel their complex behavior and shed light on the underlying dynamics of the Sinkhorn algorithm. We prove the existence of the multifractal spectrum and the singularity spectrum for the Sinkhorn coupling matrices. Furthermore, we derive bounds on the generalized dimensions, providing a comprehensive characterization of their scaling properties. Our findings not only deepen our understanding of the Sinkhorn algorithm but also pave the way for novel applications and algorithmic improvements in the realm of optimal transport.