Erdős-Littlewood-Offord problem with arbitrary probabilities (Q2166270)

Based on author's abstract: Consider the random variable \(X = a_1 \xi_1 + \cdots + a_n \xi_n\), where \(a_i\) are fixed non-zero real numbers and \(\xi_i\) are i.i.d. Bernoulli random variables with parameter \(1/2\). The classical Erdős-Littlewood-Offord theorem states that the maximum possible concentration probability \[ \max_{x \in \mathbb{R}} \mathbb{P} (X=x) \leq \binom{n}{\lfloor n/2 \rfloor} / 2^n, \] and the equality holds if all the \(a_i\) are \(1\). This paper considers the general case where \(\xi_i\) are i.i.d. Bernoulli random variables with parameter \(0 < p <1\). Using purely combinatorial techniques, The author shows that the exact maximum concentration probability is achieved for some choice of the coefficients \(a_i\) with \(a_i \in\{-1,1\}\) for all \(i\). Furthermore, by using of Fourier-analytic techniques, the optimal numbers of \(1\)s and \(-1\)s is found that it can be far from equal in some cases.