Talagrand's discrete-cube anti-concentration conjecture

Let n1n\geq 1 and t0t\geq 0, let μt=1et2δ1+1+et2δ1\mu_t=\frac{1-e^{-t}}{2}\delta_{-1}+\frac{1+e^{-t}}{2}\delta_1, and let μ=μ\mu=\mu_\infty be the uniform probability measure on {1,1}\{-1,1\}. Write μn\mu^n for the product measure and let Ttf=fμtnT_t f=f*\mu_t^n on functions f:{1,1}nRf:\{-1,1\}^n\to\mathbb R. For non-negative ff, write f1\|f\|_1 for its L1(μn)L^1(\mu^n) norm. Talagrand's conjecture. For every t>0t>0, there exists a function φt:[1,)[1,)\varphi_t:[1,\infty)\to[1,\infty) with

limαφt(α)=\lim_{\alpha\to\infty}\varphi_t(\alpha)=\infty

such that for every f:{1,1}nR+f:\{-1,1\}^n\to\mathbb R_+ and every α>1\alpha>1,

μn({x{1,1}n:Ttf(x)>αf1})1αφt(α).\mu^n\left(\{x\in\{-1,1\}^n:T_tf(x)>\alpha\|f\|_1\}\right)\leq\frac{1}{\alpha\varphi_t(\alpha)}.

This is presented as a discrete-cube generalization of the Gaussian conjecture and is proved by approximately embedding Gaussian space into discrete cubes of growing dimension via the central limit theorem.

Sources & referencesView supporting material

Primary source

Ronen Eldan and James R. Lee, “Regularization under diffusion and anti-concentration of the information content”, arXiv:1410.3887 (2017).

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.