Abstract
We introduce Reflective Hamiltonian Monte Carlo (ReHMC), an HMC-basedalgorithm, to sample from a log-concave distribution restricted to a convexpolytope. We prove that, starting from a warm start, it mixes in $\widetildeO(\kappa d^2 \ell^2 \log (1 / \varepsilon))$ steps for a well-rounded polytope,ignoring logarithmic factors where $\kappa$ is the condition number of thenegative log-density, $d$ is the dimension, $\ell$ is an upper bound on thenumber of reflections, and $\varepsilon$ is the accuracy parameter. We alsodeveloped an open source implementation of ReHMC and we performed anexperimental study on various high-dimensional data-sets. Experiments suggestthat ReHMC outperfroms Hit-and-Run and Coordinate-Hit-and-Run regarding thetime it needs to produce an independent sample.