The law of the iterated logarithm is a classical result in probability theory that describes how far a random process wanders from its average over a long time. When the random variables involved have long-range dependence, meaning that values far apart in time are still meaningfully correlated, the usual version of this law breaks down and must be replaced by a more complex one. This paper focuses on a specific technical setting called "Hermite rank two," which describes a particular way that the long-range dependence influences the limiting behavior. The key object of study is a constant that appears in this law, essentially a scaling factor that tells you how large the fluctuations are. The authors show that this constant can be identified as the largest eigenvalue of a certain mathematical operator, connecting a probabilistic question to a problem in functional analysis.
The paper then explores how this constant behaves under two different limiting operations: concentrating weight into a smaller and smaller region, and approaching a critical boundary where the strength of memory in the system is just barely enough to change the nature of the fluctuations. Both limits turn out to be well-behaved, and remarkably, the order in which you take the two limits does not matter, which is a non-obvious and mathematically pleasing result. Near the memory boundary, the constant shrinks at a rate proportional to the square root of the distance to that boundary, and the authors pin down the coefficient governing this decay with extraordinary precision.
The numerical work in the paper is particularly striking. The leading coefficient is computed to 25 decimal places, and the authors provide certified error bounds, meaning they can guarantee the answer is correct within a proven interval rather than just reporting a numerical approximation and hoping it is accurate. This level of rigor is unusual and makes the results reliable benchmarks for future theoretical and computational work. Overall, the paper advances understanding of how randomness with long memory behaves at a fundamental level, combining tools from probability, spectral theory, and careful numerical analysis.