Counting partitions on the abacus (Q2269012)

The partition function was famously studied by Hardy and Ramanujan using the circle method; it has even an exact formula. A weaker asymptotic formula proved by them asserts \(\log p(n) \sim 2 \sqrt{n\pi^2/6}\) as \(n \rightarrow \infty\). The method is essentially analytic. Erdös showed the inequality \(\log p(n) \leq 2 \sqrt{n\pi^2/6}\) for all \(n\) by elementary arguments. Using such elementary arguments, Erdös was also able to prove the asymptotic formula \(p(n) \sim e^{2 \sqrt{n\pi^2/6}}/bn\) for some real number \(b>0\). In the paper under review, the author gives beautiful combinatorial arguments to prove this result as well as the following inequality: \textit{For each \(\epsilon>0\), there is a constant \(A(\epsilon)\) such that \(\log p(n) \leq A(\epsilon) n^{1/2+ \epsilon}\) \(\forall n\).} The idea is to use the representation theory of the symmetric group. In particular, the author represents partitions using the so-called abacus of James and Kerber. He obtains combinatorially a bijection for partitions. This establishes that \(p(n) = \sum_r t((n-r(r+1)/2)/2)\) where \(t(n) = \sum_m p(m)p(n-m)\) where the sum on the left side runs over those non-negative integers \(r\) for which \(n-r(r+1)/2\) is a non-negative even integer.