An elementary proof of anti-concentration for degree two non-negative Gaussian polynomials (original) (raw)
Abstract:A classic result by Carbery and Wright states that a polynomial of Gaussian random variables exhibits anti-concentration in the following sense: for any degree ddd polynomial fff, one has the estimate P(∣f(x)∣leqvarepsiloncdotE∣f(x)∣)leqO(1)cdotdvarepsilon1/dP( |f(x)| \leq \varepsilon \cdot E|f(x)| ) \leq O(1) \cdot d \varepsilon^{1/d}P(∣f(x)∣leqvarepsiloncdotE∣f(x)∣)leqO(1)cdotdvarepsilon1/d, where the probability is over xxx drawn from an isotropic Gaussian distribution. In this note, we give an elementary proof of this result for the special case when fff is a degree two non-negative polynomial.
Submission history
From: Stephen Tu [view email]
[v1] Sun, 15 Jan 2023 00:06:37 UTC (6 KB)