English

Uniqueness of Markov random fields with higher-order dependencies

Probability 2023-07-13 v3 Mathematical Physics math.MP

Abstract

Markov random fields on a countable set V\sf V are studied. They are canonically set by a specification γ\gamma, for which the dependence structure is defined by a pre-modification (he)eE(h_e)_{e\in {\sf E}} -- a consistent family of functions he:Se[0,+)h_e : S^e\to [0,+\infty), where SS is a standard Borel space and E\sf E is an infinite collection of finite eVe\subset {\sf V}. Different ee may contain distinct number of elements, which, in particular, means that the dependence graph H=(V,E){\sf H}=({\sf V}, {\sf E}) is a hypergraph. Given eEe\in {\sf E}, let δ(e)\delta (e) be the logarithmic oscillation of heh_e. The result of this work is the assertion that the set of all fields G(γ)\mathcal{G}(\gamma) is a singleton whenever δ(e)\delta(e) satisfies a condition, a particular version of which can be δ(e)ϰg(nL(e))\delta(e) \leq \varkappa g(n_{\sf L}(e)), holding for all ee and some H\sf H-specific ϰ(0,1)\varkappa\in (0,1). Here gg is an increasing function, e.g., g(n)=a+logng(n) = a+\log n, and nL(e)n_{\sf L}(e) is the degree of ee in the line-graph L(H){\sf L}({\sf H}), which may grow ad infinitum. This uniqueness condition is essentially less restrictive than those based on classical Dobrushin's methods, according to which either of e|e|, nL(e)n_{\sf L}(e) and δ(e)\delta(e) should be globally bounded. We also prove that its fulfilment implies that the unique element of G(γ)\mathcal{G}(\gamma) is globally Markov.

Keywords

Cite

@article{arxiv.2304.11369,
  title  = {Uniqueness of Markov random fields with higher-order dependencies},
  author = {Dorota Kepa-Maksymowicz and Yuri Kozitsky},
  journal= {arXiv preprint arXiv:2304.11369},
  year   = {2023}
}
R2 v1 2026-06-28T10:14:27.470Z