The paper studies a natural optimization problem in number theory: given a large number x and a spacing requirement y, how do you choose a set of positive integers up to x, with no two chosen numbers too close together, to maximize the total count of divisors across all chosen numbers? Some integers have many more divisors than others (highly composite numbers like 120 or 360), so the challenge is picking the richest integers while respecting the spacing rule. The authors focus on a specific range of spacing values that grow like a power of the logarithm of x, and they want to know how much the spacing constraint hurts you compared to having no constraint at all.
The main finding is that there is a sharp threshold around a critical spacing size. When the spacing is below this threshold (corresponding to a parameter lambda less than 2), the constraint costs essentially nothing: you can still achieve the same order of magnitude for the total divisor sum as if there were no restriction at all. Above the threshold (lambda greater than 2), the constraint genuinely hurts, and the maximum achievable sum drops by a power of the logarithm of x. The authors even identify precisely what that power is in terms of lambda, giving an exact asymptotic formula in parts of the range. This kind of threshold behavior, where a combinatorial constraint suddenly starts mattering at a critical scale, is a recurring theme across mathematics and physics.
The proof draws on a wide toolkit. On the analytic side, the authors use classical results about how many prime factors a typical integer has (the Sathe-Selberg theory) and generating function techniques involving Cauchy's integral formula to estimate sums over smooth squarefree numbers, which turn out to be the integers contributing most to the divisor sum. On the combinatorial side, they model the problem as finding a large independent set in a graph where edges connect integers that are too close together, and they apply two clever tools: the Caro-Wei bound, which gives a guaranteed independent set size based on vertex degrees, and a concentration inequality for random permutations, which shows that a random arrangement of the best candidate integers will almost surely yield a well-spaced selection.