show/hide this revision's text 2 deleted 111 characters in body

Let $a_1,\ldots,a_n$ be real numbers such that $\sum_i a_i^2 =1$ and let $X_1,\ldots,X_n$ be independent, uniformly distributed, Bernoulli $\pm 1$ random variables. Define the random variable

$S:= \sum_i X_i a_i $

Is it true (with no further assumption on $a_1,\ldots,a_n$) that $S$ is unlikely to be small and that, specifically,

Are there are positive absolute constants $c_1, c_2, c_3$ \epsilon >0$ and $\delta<1$ such that for every $a_1,\ldots,a_n$

${\mathbb P} [|S| \leq \epsilon ] \leq c_1\cdot \epsilon + c_2 \cdot n^{-c_3}$ delta$ ?
show/hide this revision's text 1

Anti-concentration of bernoulli sums

Let $a_1,\ldots,a_n$ be real numbers such that $\sum_i a_i^2 =1$ and let $X_1,\ldots,X_n$ be independent, uniformly distributed, Bernoulli $\pm 1$ random variables. Define the random variable

$S:= \sum_i X_i a_i $

Is it true (with no further assumption on $a_1,\ldots,a_n$) that $S$ is unlikely to be small and that, specifically, there are positive constants $c_1, c_2, c_3$ such that

${\mathbb P} [|S| \leq \epsilon ] \leq c_1\cdot \epsilon + c_2 \cdot n^{-c_3}$ ?