Abstract
This research establishes Gaussian analogues for two fundamental results in discrete probability: Talagrand’s correlation inequality and the Benjamini–Kalai–Schramm (BKS) noise sensitivity theorem. By employing a novel "geometric influence" based on lower Minkowski content, the authors bridge the gap between discrete product spaces and continuous Gaussian measures.
Introduction
In the study of Boolean functions and probability, the concept of "influence"—how much a single input coordinate can change the output of a function—is a cornerstone. Traditionally, these tools were perfected for the discrete cube (like a series of coin flips). This paper, published in Annales de l’Institut Henri Poincaré, extends these powerful discrete tools into the continuous Gaussian realm and beyond.
Geometric Influence and Methodology
The core innovation lies in the definition of influence for continuous spaces. Instead of the discrete "bit-flip" definition, the authors use lower Minkowski content. For a set $A$, the influence of the $i$-th coordinate is defined by looking at "fibers" (lines) through the set in that direction and measuring the boundary of the set along those lines.
The paper relies on three primary pillars to prove its results:
- Discrete-to-Continuous Limits: Approximating Gaussian functions using sequences of functions defined on discrete cubes of increasingly high dimensions and applying the Central Limit Theorem.
- Ornstein–Uhlenbeck Semigroup: Using this mathematical tool to describe how functions evolve under Gaussian noise.
- Reverse Hypercontractivity: Combining the Ornstein–Uhlenbeck operator with Borell’s reverse hypercontractive inequality for direct Gaussian proofs.
Main Results
The authors establish two primary Gaussian analogues that mirror key results in the discrete world:
- Gaussian Talagrand Bound: The authors established a quantitative version of the Gaussian FKG inequality. They proved that for two increasing sets $A$ and $B$, the correlation is bounded below by a function of the sum of their geometric influences.
- Quantitative BKS Theorem: They proved that a sequence of sets is "noise sensitive" (meaning small input perturbations lead to uncorrelated outputs) if the sum of the squares of their geometric influences vanishes.
These bounds are shown to be tight (up to a constant factor) for "threshold sets"—sets defined by whether the average of the inputs exceeds a certain value.
Discussion and Future Work
The paper reinforces that geometric influence is the natural continuous counterpart to discrete influence, allowing for a seamless transition between the two worlds. It provides a more natural scaling for biased discrete spaces than previous definitions.
Future work remains in understanding influences in spaces where coordinates are dependent (non-product measures). Additionally, the authors suggest that future research might combine hypercontractivity and reverse hypercontractivity into a single, even more powerful inequality, with potential applications in percolation theory and isoperimetric inequalities.
Summary updated