Abstract:This article studies the convergence properties of trans-dimensional MCMC algorithms when the total number of models is finite. It is shown that, for reversible and some non-reversible trans-dimensional Markov chains, under mild conditions, geometric convergence is guaranteed if the Markov chains associated with the within-model moves are geometrically ergodic. This result is proved in an $L^2$ framework using the technique of Markov chain decomposition. While the technique was previously developed for reversible chains, this work extends it to the point that it can be applied to some commonly used non-reversible chains. The theory herein is applied to reversible jump algorithms for three Bayesian models: a probit regression with variable selection, a Gaussian mixture model with unknown number of components, and an autoregression with Laplace errors and unknown model order.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is about the convergence properties of the Trans - dimensional Markov Chain Monte Carlo (MCMC) algorithm. Specifically, the author studies the geometric convergence of the trans - dimensional MCMC algorithm when the total number of models is finite. The main contribution of the paper is to prove that under certain conditions, if the Markov chain related to the internal movement of the model is geometrically ergodic, then the reversible and some irreversible trans - dimensional Markov chains can also ensure geometric convergence. This result is proved in the $L^2$ framework using the Markov chain decomposition technique and extends the application range of this technique, enabling it to be applied to some commonly used irreversible chains. ### Main Problems 1. **Geometric Ergodicity Conditions**: The paper provides verifiable sufficient conditions to ensure the geometric convergence of the trans - dimensional MCMC algorithm in the $L^2$ distance. These conditions mainly involve the geometric ergodicity of the internal movement of the model and the irreducibility of the movement between models. 2. **Theoretical Applications**: The author applies these theories to the reversible jump algorithms (Reversible Jump MCMC) of three Bayesian models, including: - The Probit regression model with variable selection - The Gaussian mixture model with an unknown number of components - The Laplace - error autoregressive model with an unknown model order ### Key Results - **Theorem 1**: Suppose that the Markov chain within each model is geometrically ergodic and the movement between models is irreducible, then the trans - dimensional chain geometrically converges under $L^2(\Pi)$. - **Theorem 7**: Provides a quantitative bound on the convergence rate, further verifying the conditions of geometric ergodicity. ### Methods - **Markov Chain Decomposition**: Analyze the properties respectively by decomposing the dynamics of the trans - dimensional Markov chain into the internal movement of the model and the movement between models. - **Geometric Ergodicity**: Use the known geometric ergodicity conditions (such as drift and minorization conditions) to derive the convergence of the trans - dimensional chain. ### Practical Applications - **Variable Selection**: In Bayesian Probit regression, perform variable selection through the reversible jump MCMC algorithm. - **Mixture Model**: In the Gaussian mixture model, determine the unknown number of components. - **Autoregressive Model**: In the autoregressive model with Laplace error, determine the unknown model order. ### Conclusion Through strict mathematical proofs and practical applications, the paper shows how to ensure the geometric convergence of the trans - dimensional MCMC algorithm, providing a theoretical basis for parameter estimation and model selection in complex statistical models.

Geometric ergodicity of trans-dimensional Markov chain Monte Carlo algorithms

Geometric allocation approaches in Markov chain Monte Carlo

Geometric Ergodicity and the Spectral Gap of Non-Reversible Markov Chains

Equivalences of Geometric Ergodicity of Markov Chains

Convergence Rates of Attractive-Repulsive MCMC Algorithms

Convergence Analysis of MCMC Algorithms for Bayesian Multivariate Linear Regression with Non‐Gaussian Errors

MCMC-Based Inference in the Era of Big Data: A Fundamental Analysis of the Convergence Complexity of High-Dimensional Chains

Markov chain Monte Carlo Methods For Lattice Gaussian Sampling:Convergence Analysis and Enhancement

Generalization Performance of ERM Algorithm with Geometrically Ergodic Markov Chain Samples

Geometric ergodicity of Rao and Teh's algorithm for Markov jump processes

Information Geometry of Reversible Markov Chains

Geometry and Duality of Alternating Markov Chains

Quantifying Uncertainty in Transdimensional Markov Chain Monte Carlo Using Discrete Markov Models

On the Geometric Ergodicity of Metropolis-Hastings Algorithms for Lattice Gaussian Sampling

Fluctuations of Interacting Markov Chain Monte Carlo Methods

Non-asymptotic Estimates for Markov Transition Matrices with Rigorous Error Bounds

Information-geometric Markov Chain Monte Carlo methods using Diffusions

Lower bounds on the rate of convergence for accept-reject-based Markov chains in Wasserstein and total variation distances

Geometry and Dynamics for Markov Chain Monte Carlo

Generalization Bounds of Erm Algorithm with V-Geometrically Ergodic Markov Chains

Asymptotic bias of inexact Markov Chain Monte Carlo methods in high dimension