The paper studies a probability question about random walks built from a special kind of random sequence. Start with n random numbers, none of which are zero. The sequence has two symmetry properties: first, shuffling the numbers in any order does not change the overall distribution (exchangeability); second, flipping the sign of any subset of them also leaves the distribution unchanged (sign-invariance). From these numbers, form the running sums: the first number alone, then the first two added together, then the first three, and so on. The central question is: how likely is it that all of these running sums stay positive, or at least non-negative? These likelihoods are called the strong and weak persistence probabilities, respectively.
Earlier work had already pinned down some of the extreme values these probabilities could take, but two of the four bounds (the maximum of the weak version and the minimum of the strong version) were still open. This paper closes those gaps, establishing a clean chain of inequalities involving familiar quantities from combinatorics, specifically binomial coefficients. The four bounds form a neat ordered sequence, which the authors prove is tight, meaning there exist specific distributions that actually achieve each extreme.
A notable consequence is that both persistence probabilities, regardless of the specific distribution chosen (as long as it satisfies the two symmetry conditions), always behave like one over the square root of n for large n. This gives a universal scaling law: no matter how you construct such a symmetric random sequence, the chance that all running sums remain on one side of zero shrinks at this predictable rate. The paper also extends the results to a couple of related settings, including one where some of the random numbers are allowed to equal zero.