The paper studies Gaussian processes, which are mathematical objects used to model random phenomena that vary continuously over space or time. Specifically, it looks at what happens when the minimum value of such a process is unusually large (a "high minimum"), which is a rare event. The central question is: given that something unlikely has happened, what can we say about how unlikely it was and where the minimum occurred?
The main finding concerns the "overshoot," which measures by how much the minimum exceeds a high threshold. The paper proves that when you scale this overshoot appropriately, it follows an exponential distribution in the limit as the threshold grows. The key parameter controlling this distribution is a quantity called the minimum covariance energy, which captures a geometric property of the process related to how its values at different points are correlated. This result generalizes earlier work that required the process to be smooth, extending it to a broader class of Gaussian processes.
The paper also addresses where the minimum is likely to occur, given that it is unusually high. It shows that the location of the minimum tends to concentrate around certain special points determined by the same covariance energy minimization problem. If there is a unique such special point or distribution of points, the location of the minimum converges to it in a precise probabilistic sense. The authors demonstrate these results concretely using several well-known examples, including stationary Gaussian processes and fractional Brownian motion, a popular model in physics and finance that generalizes ordinary Brownian motion to allow for long-range correlations.