A positive integer is called "r-free" if it is not divisible by the r-th power of any prime. For r=2, these are the familiar squarefree numbers, integers not divisible by any perfect square greater than 1. About 61% of all integers are squarefree, so gaps where many consecutive integers all fail to be squarefree are rare. This paper precisely measures how rare: it finds the exact asymptotic density of long consecutive runs of non-r-free integers, pinning down not just the leading term but also the next two corrections. The result settles a longstanding question about "Mirsky's constant," which governs the spacing between consecutive squarefree numbers, improving the best prior estimate by a factor of roughly log(h) / log(log(h)).
The proof strategy is elegant. For the upper bound, the authors observe that small primes, whose r-th powers are too large to fit many times in a window of length k, are forced to cover specific positions with no room to maneuver. This means many positions in the window must each be covered by the r-th power of a distinct large prime, and counting the ways to arrange these coverings leads to a combinatorial sum that can be evaluated using a saddle-point approximation from complex analysis. For the lower bound, the authors use the Chung-Erdos inequality from probability theory, which is powerful enough to recover the full value of that same sum rather than just its largest term. This two-sided precision is what makes the linear coefficient in the density formula explicit and computable.
A surprising consequence connects this to covering systems, which are collections of arithmetic progressions that together cover every integer. The paper shows that any covering system of length k must have a modulus (the product governing the progressions) at least exponentially large, and that the standard construction using the Chinese Remainder Theorem falls short by a factor of two compared to what the true density implies. This means a famous 1951 result of Erdos about gaps between squarefree numbers is secretly equivalent to saying that the first long run of non-squarefree integers appears far earlier than the scale set by any covering system, at the so-called first-moment threshold rather than at the minimal modulus scale.