Mixing time of a matrix random walk generated by elementary transvections

Abstract: We consider a Markov chain on invertible $n\times n$ matrices with entries in $\mathbb{Z}_2$ which moves by picking an ordered pair of distinct rows and add the first one to the other, modulo $2$. We establish a logarithmic Sobolev inequality with constant $n^2$, which yields an upper bound of $O(n^2\log n)$ on the mixing time. We also study another chain which picks a row at random and randomize it conditionally on the resulting matrix being invertible, and show that the mixing time is between $n\log n$ and $4n\log n$.