“When an Ornstein–Uhlenbeck Process Meets a Decreasing Random Boundary”

Giovedì 29 Ottobre 2026, ore 11:30 - Aula 2AB45 - Jonatha Anselmi (Inria)

Abstract

We investigate a first-passage-time problem in which an Ornstein–Uhlenbeck process hits a decreasing random boundary that evolves in response to the process itself. This problem arises naturally in the analysis of autoscaling systems, where computing resources are dynamically activated and deactivated in response to fluctuating workloads.

Starting from a stochastic model of such a system, we establish a diffusion approximation and prove convergence of the associated first-passage time. The resulting problem can be expressed as the hitting time of a nonstationary Gaussian process to zero. However, the process is neither Markovian nor differentiable, preventing the direct application of classical results based on the Kac–Rice formula.

To address this difficulty, we develop a Rice-inspired heuristic that replaces the classical crossing-rate term with a conditional infinitesimal drift, yielding an explicit approximation for the expected first-passage time. We establish a rigorous asymptotic characterization of the mean hitting time, showing logarithmic growth with the initial separation and a bounded additive error for the proposed approximation. Numerical experiments demonstrate its accuracy in the parameter regimes motivated by the autoscaling application.



Research group on Probability and Finance