Double-jump phase transition for the reverse Littlewood--Offord problem

Erd\H{o}s conjectured in 1945 that for any unit vectors $v_1, \dotsc, v_n$ in $\mathbb{R}^2$ and signs $\varepsilon_1, \dotsc, \varepsilon_n$ taken independently and uniformly in $\{-1,1\}$, the random Rademacher sum $\sigma = \varepsilon_1 v_1 + \dotsb + \varepsilon_n v_n$ satisfies $\|\sigma\|_2 \leq 1$ with probability $\Omega(1/n)$. While this conjecture is false for even $n$, Beck has proved that $\|\sigma\|_2 \leq \sqrt{2}$ always holds with probability $\Omega(1/n)$. Recently, He, Ju\v{s}kevi\v{c}ius, Narayanan, and Spiro conjectured that the Erd\H{o}s' conjecture holds when $n$ is odd. We disprove this conjecture by exhibiting vectors $v_1, \dotsc, v_n$ for which $\|\sigma\|_2 \leq 1$ occurs with probability $O(1/n^{3/2})$. On the other hand, an approximated version of their conjecture holds: we show that we always have $\|\sigma\|_2 \leq 1 + \delta$ with probability $\Omega_\delta(1/n)$, for all $\delta > 0$. This shows that when $n$ is odd, the minimum probability that $\|\sigma\|_2 \leq r$ exhibits a double-jump phase transition at $r = 1$, as we can also show that $\|\sigma\|_2 \leq 1$ occurs with probability at least $\Omega((1/2+\mu)^n)$ for some $\mu > 0$. Additionally, and using a different construction, we give a negative answer to a question of Beck and two other questions of He, Ju\v{s}kevi\v{c}ius, Narayanan, and Spiro, concerning the optimal constructions minimising the probability that $\|\sigma\|_2 \leq \sqrt{2}$. We also make some progress on the higher dimensional versions of these questions.