A piecewise deterministic Monte Carlo method for diffusion bridges

Abstract We introduce the use of Zig-Zag sampler to problem sampling conditional diffusion processes (diffusion bridges). The is a rejection-free scheme based on non-reversible continuous piecewise deterministic Markov process. Similar Lévy–Ciesielski construction Brownian motion, we expand path in truncated Faber–Schauder basis. coefficients within basis are sampled using sampler. A key innovation fully local algorithm for that allows exploit sparsity structure implied by dependency graph and subsampling technique reduce complexity algorithm. illustrate performance proposed methods number examples.