Renewal processes are mathematical models for counting how often random events repeat over time, like machine failures or bus arrivals, where the gaps between events follow some probability distribution. When those gaps have a "heavy tail," meaning very large gaps are more likely than in normal distributions, the long-run behavior of the process becomes harder to analyze. This paper focuses on a specific family of heavy-tailed distributions where the tail decays like a power law, and asks: beyond the leading-order approximation of how many renewals have occurred by time t, what does the next correction term look like?
The main finding is that the correction has a two-layer structure. One part, called the local correction, is controlled by something called the equilibrium tail, which captures how the distribution behaves near the current time. A second, larger-scale part is controlled by an integrated version of that same quantity and reflects how the process deviates from its long-run average over a broad window of time. The paper covers a range of power-law exponents between 2 and 3, meaning distributions that have a finite variance but a heavy enough tail to behave unusually. A special focus is the boundary case where the exponent equals exactly 2, which is a critical threshold where the usual power-law hierarchy breaks down and the correction is instead governed by a slowly varying function, though the local correction mechanism still survives in a recognizable form.
The mathematical toolkit relies heavily on Laplace transforms, which convert time-domain questions into algebraic ones, and on Tauberian theorems, which are classical results that translate behavior in the transform domain back into precise statements about the original time-domain quantities. The results unify the second-order picture across the entire range of exponents considered and suggest natural extensions to related problems, such as branching processes where individuals reproduce at random times drawn from similar heavy-tailed distributions.