Electronic International Standard Serial Number (EISSN)
1573-1375
abstract
We study Hamiltonian Monte Carlo (HMC) samplers based on splitting the Hamiltonian H as H0(θ,p)+U1(θ), where H0 is quadratic and U1 small. We show that, in general, such samplers suffer from stepsize stability restrictions similar to those of algorithms based on the standard leapfrog integrator. The restrictions may be circumvented by preconditioning the dynamics. Numerical experiments show that, when the H0(θ,p)+U1(θ) splitting is combined with preconditioning, it is possible to construct samplers far more efficient than standard leapfrog HMC.
Classification
subjects
Mathematics
Physics
Robotics and Industrial Informatics
keywords
bayesian analysis; hamiltonian dynamics; markov chain monte carlo; splitting integrators