Hamiltonian Monte Carlo with explicit, reversible, and volume-preserving adaptive step size control

Hamiltonian Monte Carlo is a Markov chain Monte Carlo method that uses Hamiltonian dynamics to efficiently produce distant samples. It employs geometric numerical integration to simulate Hamiltonian dynamics, which is a key of its high performance. We present a Hamiltonian Monte Carlo method with adaptive step size control to further enhance the efficiency. We propose a new explicit, reversible, and volume-preserving integration method to adaptively set the step sizes, which does not violate the detailed balance condition or require a large increase in computational time.