Abstract:
In order to generate random orthogonal matrices, Hastings [Biometrika, 57 (1970), pp. 97–109] considered a Markov chain on the orthogonal group $SO(n)$ generated by random rotations on randomly selected coordinate planes. We investigate different ways to measure the convergence to equilibrium of this walk. To this end, we prove, up to a multiplicative constant, that the spectral gap of this walk is bounded below by $1/n^2$ and the entropy/entropy dissipation bound is bounded above by $n^3$.
Keywords:convergence to equilibrium, spectral gap.