Gaussian concentration implies modified logarithmic Sobolev inequalities under Ricci curvature bounds

About 10 years old · traced to

Let (X,Q,π)(\mathcal{X},Q,\pi) be a Markov chain with invariant measure π\pi, and suppose that a concentration property with respect to the distance dWd_{\mathcal{W}} holds with profile

α(r)=Me−ρr2.\alpha(r)=Me^{-\rho r^2}.

Assume that Ric⁡(X,Q,π)≥−κ\operatorname{Ric}(\mathcal{X},Q,\pi)\geq-\kappa for some κ>0\kappa>0. Gaussian-concentration modified logarithmic Sobolev conjecture. There exists a constant τ(M)\tau(M) and a constant λ(κ,M,ρ)\lambda(\kappa,M,\rho) such that, if

κρ<τ(M),\frac{\kappa}{\rho}<\tau(M),

then MLSI⁡(λ(κ,M,ρ))\operatorname{MLSI}\bigl(\lambda(\kappa,M,\rho)\bigr) holds. If moreover Ric⁡(X,Q,π)≥0\operatorname{Ric}(\mathcal{X},Q,\pi)\geq0, then MLSI⁡(cMρ)\operatorname{MLSI}(cM\rho) holds for some universal constant cc. The open aspect identified in the source is the value of the constant.

References

Primary source

Matthias Erbar and Max Fathi, “Poincaré, modified logarithmic Sobolev and isoperimetric inequalities for Markov chains with non-negative Ricci curvature”, arXiv:1612.00514 (2016).

Progress summary

Never refreshed

Nothing recorded yet. Refresh searches the literature and the public web for attempts on this problem, and writes the first summary here.

Solutions 0

No solutions have been posted yet.