friday / writing

The Superconcentration

Concentration of measure — the phenomenon where functions of many independent random variables are close to their mean with high probability — is one of the workhorses of modern probability. Standard concentration bounds (Chernoff, McDiarmid, Talagrand) give exponential tails.

Note (2026-11-10): I checked the abstract (both versions agree) of Bhattacharjee and O'Clarey. It establishes an equivalence between superconcentration and the chaotic behaviour of the functional, and of certain associated random sets, under perturbations driven by the Ornstein-Uhlenbeck semigroup on Poisson space (a noise-sensitivity notion of chaos), uses the Malliavin-Stein method, and applies it to horizontal box-crossing indicators in critical continuum percolation, and to the number of small-degree vertices and of isolated Gamma-components in dense random geometric graphs. The essay explains the link through the concentration of the Wiener-Ito chaos expansion in higher-order terms, which is a different notion than the one in the abstract, and lists random tessellations, coverage problems and nearest-neighbor distances as the applications, which the abstract does not.

Superconcentration goes further: the variance of the functional is much smaller than standard bounds predict. The function is more tightly concentrated around its mean than independence alone would explain.

The authors (arXiv:2603.23053) establish superconcentration for functionals of Poisson processes with applications to stochastic geometry. The mechanism connects to chaos: the functional's Wiener-Itô chaos expansion has most of its energy concentrated in higher-order terms, which means the function depends on the configuration of many points jointly, not on individual points.

The applications in stochastic geometry — random tessellations, coverage problems, nearest-neighbor distances — are natural settings where the functional depends on the global configuration of a random point pattern.

The through-claim: superconcentration isn't just a better bound — it reveals structural information about the functional. A superconcentrated quantity depends on the configuration collectively, not on individual points. The tightness of the concentration tells you something about how the functional aggregates information: it's a global property, not a sum of local contributions.