The pseudo-spectral gap of a Markov chain controls how quickly it mixes and how strongly its empirical averages concentrate. Estimating this quantity from data is a natural but delicate problem, especially when only one trajectory is available. A new arXiv preprint tackles this in the minimax framework, deriving optimal rates for continuous state spaces.

The work builds directly on earlier finite-state results by Wolfer and Kontorovich (2024), generalising their approach to chains with continuous state spaces. The abstract indicates that the authors obtain minimax bounds for estimating the pseudo-spectral gap, though the specific rates and conditions are not detailed in the available text.

Beyond estimation itself, the authors highlight applications to empirical concentration bounds. These would let practitioners construct data-dependent guarantees for averages of functions of a Markov chain, using only the observed path. The abstract is truncated, so the full scope of the results remains to be seen, but the extension to continuous spaces is a notable step forward.