Probabilistic graphical models typically assume Gaussian distributions, which limits their use on real-world data with heavy tails or other non-Gaussian structure. A new arXiv paper aims to relax that assumption by bringing copulas into the framework of undirected graphical models.

The paper establishes a characterization of the pairwise Markov property—a core condition that determines which variables are conditionally independent—in terms of copulas. This provides a theoretical foundation for building Markov networks that can model dependencies without relying on Gaussianity.

The title indicates a focus on Student copulas, which are commonly used for tail dependence, though the abstract does not specify the exact family or provide experimental results. The paper is available on arXiv under identifier 2610.19135.