The paper studies two types of partial differential equations (elliptic equations) where a term in the equation becomes extremely large near a single point, the origin. These so-called singular perturbations involve coefficients that blow up like the inverse square of the distance to the origin. The central question is: under what conditions on the parameter lambda do solutions to these equations have enough smoothness to belong to a function class called H2, meaning their second derivatives are square-integrable? The authors work in five or more spatial dimensions, where the geometry and the competition between different mathematical effects make the problem especially subtle.
The main achievement is identifying sharp numerical thresholds for lambda that precisely separate the regime where H2 regularity holds from the regime where it fails. For the first equation, which involves a gradient term weighted by the inverse square distance, the authors prove that solutions are H2-smooth exactly when lambda exceeds a specific critical value involving the dimension N, and they show by counterexample that smoothness genuinely breaks down at and below that threshold. This is a significant improvement over a classical approach using a tool called Kato perturbation theory, which only guaranteed H2 regularity over a narrower range of lambda. The second equation, involving the solution itself divided by the square distance, is handled with similar precision.
A key technical ingredient the authors develop along the way is a new family of sharp Hardy-Rellich inequalities tailored to the specific differential operators appearing in the two equations. Hardy and Rellich inequalities are fundamental analytic tools that control how singular a function or its derivatives can be near a point. By sharpening these inequalities to match the operators at hand, the authors are able to push the regularity analysis exactly to the boundary of what is possible, leaving no gap between where they can prove smoothness and where they demonstrate it fails.