The paper studies a fundamental geometric question: given a large collection of random points scattered uniformly across some smooth surface or region, how well can those points be used to "pack" the underlying space? Specifically, it looks at maximal packings, which are arrangements where you place as many non-overlapping balls as possible on the surface, with each ball centered at one of your random sample points. The key quantity of interest is the smallest gap between any sample point and the center of the best possible packing configuration. Understanding how this quantity behaves as you gather more and more data tells you something precise about the geometry of the underlying space from which the data came.
The main theoretical contribution is proving that these minimal distances follow a specific statistical pattern called a Gumbel distribution in the limit of large samples. The Gumbel distribution is one of the classical "extreme value" distributions, the kind that naturally governs the behavior of maxima and minima of many random quantities. What makes this result particularly useful is that the authors work out explicitly how to rescale the distances to get the Gumbel limit, and these rescaling constants depend in a transparent way on the geometric properties of the surface, such as its dimension and curvature. This explicitness is important because it means you can actually use the result in practice without having to estimate unknown constants from the data.
The practical payoff is that the result provides confidence bounds for the unknown underlying space from which your data was sampled, a problem that matters in topological data analysis and manifold learning. If you observe a cloud of points and want to know what surface they came from, this work gives you principled statistical guarantees. The authors illustrate everything with concrete examples from directional statistics, including data on circles, spheres, and tori, and back up the theory with simulations. The mathematical tools they use are notably diverse, blending geometric probability, combinatorics, and a special function from complex analysis called the Lambert W function.