Uniform ergodicity of the iterated conditional SMC and geometric ergodicity of particle Gibbs samplers (Q2405204)

The paper considers the Markov chain obtained by iterating the conditional sequential Monte Carlo algorithm for a fixed target distribution. The approach relies on certain assumptions for uniform convergence that are proved to be necessary and sufficient, and leads to quantitative bounds on performance measures in terms of the number of particles involved. The sufficient conditions for uniform ergodicity of the iterated conditional sequential Markov chain are transferred into sufficient conditions for geometric ergodicity of the associated particle Gibbs Markov chain. Under some conditions the particle Gibbs sampler is shown to be asymptotically as efficient as the Gibbs sampler as the number of particles increases