An n-dimensional Rosenbrock Distribution for MCMC Testing

03/22/2019

∙

by   Filippo Pagani, et al.

∙

0

∙

share

The Rosenbrock function is an ubiquitous benchmark problem for numerical optimisation, and variants have been proposed to test the performance of Markov Chain Monte Carlo algorithms. In this work we discuss the two-dimensional Rosenbrock density, its current n-dimensional extensions, and their advantages and limitations. We then propose our own extension to arbitrary dimensions, which is engineered to preserve the key features of the density -- such as the curved correlation structure -- and is analytically tractable. We conclude with numerical experiments that show how a naively tuned Random Walk fails to to give a representative sample in reasonable time, and how even a Simplified Manifold MALA algorithm struggles on this target.