Onsager-Machlup Functional for SDE with Time-Varying Fractional Noise
Yanbin Zhu, Xiaomeng Jiang, Yong Li
TL;DR
This work derives the Onsager–Machlup functional for SDEs driven by time-varying fractional noise with H ∈ (1/4,1), expressing J(φ,φ̇) in closed form across singular, regular, and standard regimes via a Girsanov transformation and fractional calculus. The analysis hinges on small-ball probability estimates, the fractional Girsanov transform, and measurability of norms, with the inverse operator (K_H^σ)^{-1} playing a central role and the admissible Hölder norms depending on H. The results generalize previous OM formulations to nonstationary fractional inputs and time-dependent coefficients, and are corroborated through numerical experiments on a double-wwell potential showing how time-varying fractional noise modulates metastable transitions. The framework provides a variational lens to identify the most probable transition paths and their sensitivity to noise regularity and temporal structure, with potential applications in physics and chemistry where nonlocal, memory-bearing noise is present.
Abstract
In this paper, we derive the Onsager-Machlup functional for stochastic differential equations driven by time-varying fractional noise of the form X(t) = x0 + integral from 0 to t b_s(X(s)) ds + integral from 0 to t sigma_s dB^H(s), where B^H denotes fractional Brownian motion with Hurst parameter H. Our main results are established for H in (1/4, 1) by extending small ball probability estimates and the Girsanov theorem for fractional Brownian motion to the setting with time-dependent coefficients. Regarding the choice of norms, for 1/4 < H < 1/2 the analysis is valid under the supremum norm and Holder norms of order 0 < beta < H - 1/4. For 1/2 < H < 1 the analysis applies to Holder norms of order beta satisfying H - 1/2 < beta < H - 1/4. In the case H = 1/2, the admissible norms depend on the spatial regularity of the drift coefficient b: specifically, if b is n-times continuously differentiable, then Holder norms of order 0 < beta < 1/2 - 1/(2n) are permissible. To validate our theoretical findings, we perform numerical simulations for a classical double-well potential system, illustrating how time-varying fractional noise influences transition dynamics between metastable states.