Sharp upper bounds for the deviations from the mean of the sum of independent Rademacher random variables

TL;DR

This paper establishes sharp upper bounds on the probability that the sum of independent Rademacher variables deviates from its mean by a certain multiple of the standard deviation, extending previous results and improving bounds.

Contribution

It provides precise, sharp lower bounds for deviation probabilities of sums of Rademacher variables, extending known results and solving a conjecture in the equal coefficient case.

Findings

Lower bounds are sharp and significantly better than Chebyshev bounds.

Results have applications in probability theory and random walk analysis.

The bounds improve upon previous solutions for specific deviation levels.

Abstract

For a fixed unit vector a=(a_1,a_2,...,a_n) in S^{n-1}, i.e. sum_{i=1}^n a_i^2=1, we consider the 2^n sign vectors epsilon=(epsilon_1,epsilon_2,...,epsilon_n) in {-1,1}^n and the corresponding scalar products a.epsilon=sum_{i=1}^n a_i epsilon_i. Holtzman and Kleitman formulated the following conjecture. It states that among the 2^n sums of the form sum +/- a_i there are not more with |sum_{i=1}^n +/- a_i|>1 than there are with |sum_{i=1}^n +/- a_i| <= 1. The result is of interest in itself, but has also an appealing reformulation in probability theory and in geometry. In this paper we will solve an extension of this problem in the uniform case where all the a's are equal. More precisely, for S_n being a sum of n independent Rademacher random variables, we will give, for several values of xi, precise lower bounds for the probabilities P_n:=P{-xi sqrt{n} <= S_n <= xi sqrt{n}}. There is an obvious relationship with the binomial distribution with parameters n and p=1/2. The obtained lower bounds are sharp and much better than for instance the bound that can be obtained from application of the Chebishev inequality. In case xi=1 Van Zuijlen solved this problem. We remark that our bound will have nice applications in probability theory and especially in random walk theory.