Markov Random Fields: Structural Properties, Phase Transition, and Response Function Analysis

TL;DR

This paper surveys Markov random fields (MRFs) for discrete spatial data, with emphasis on binary observations and latent variables, to clarify structural properties, phase transitions, and response-function analysis. It formalizes the SEF-MRF as $q(\mathbf{y}|\boldsymbol{\xi})=\frac{1}{Z(\boldsymbol{\xi})}\exp(\boldsymbol{\xi}'\mathbf{T}(\mathbf{y}))$ and compares Physics-Ising, Autologistic, Ising, and Potts formulations. A central contribution is the response-function tool for prior analysis, linking parameter changes to marginal/joint behavior and diagnosing phase transition through Monte Carlo estimates of $T_1$ and $T_2$. The paper also discusses parameterizations with covariates, hierarchical external fields, extensions to flexible Potts models and CRFs, and practical fitting methods, providing guidance for practitioners modeling spatially dependent discrete data.

Abstract

This paper presents a focused review of Markov random fields (MRFs)--commonly used probabilistic representations of spatial dependence in discrete spatial domains--for categorical data, with an emphasis on models for binary-valued observations or latent variables. We examine core structural properties of these models, including clique factorization, conditional independence, and the role of neighborhood structures. We also discuss the phenomenon of phase transition and its implications for statistical model specification and inference. A central contribution of this review is the use of response functions, a unifying tool we introduce for prior analysis that provides insight into how different formulations of MRFs influence implied marginal and joint distributions. We illustrate these concepts through a case study of direct-data MRF models with covariates, highlighting how different formulations encode dependence. While our focus is on binary fields, the principles outlined here extend naturally to more complex categorical MRFs and we draw connections to these higher-dimensional modeling scenarios. This review provides both theoretical grounding and practical tools for interpreting and extending MRF-based models.

Figures (4)

• Figure 1: Example of a regular lattice and the NUGs that can be used to represent dependence between the areal units of the lattice.
• Figure 2: Monte Carlo estimates of the response functions for an MRF with an Ising formulation and no external field: the mean and variance of the distribution of the proportion of black units (left column) and the proportion of matches (right column). The dashed line marks the known critical value of $\psi\approx 0.88$.
• Figure 3: Plot of the covariate used in the response function and prior predictive response function simulation studies for binary MRF formulations with an external field parameterized as a linear function of the covariate.
• Figure 4: Plotted are the estimated prior predictive response functions for the autologistic, centered autologistic, and Ising models with $\beta_1$ and $\beta_0$ marginalized out over the prior $\mathrm{N}(0,1)$. The covariates are a planar gradient such that $x_i=(r(i)+c(i)-n_r-1)/(n_r-1)$. The top four plots show the estimated response functions corresponding to the mean and standard deviation of the proportion of black units and the proportion of matching neighbors. The bottom row shows the expected dominant color and expected misclassification rate.

Theorems & Definitions (7)

• Definition 2.1
• Definition 2.2
• Definition 2.3
• Definition 2.4
• Definition 2.5
• Definition 3.1
• Definition 3.2