Paperlane

no. 16 · Tuesday 17 November 2026

Denoising-Induced Recovery of Effective Dynamics

Narcicegi Kiran

math.DS

Abstract

Recovering dynamical equations from data is a central problem in nonlinear science. In practice, measurements are corrupted by noise and often contain fast oscillations that are not part of the slow dynamics of interest. For this reason, data are typically pre-processed before model recovery or classsification. Here, we reveal the effects of filtering, a specific denoising technique, on recovery. Our analysis show that equation discovery performed on filtered data recovers, up to a fixed precision, the normal-form (effective) equations rather than the original system.

fields
Dynamical Systems
pdf ↗ page arXiv:2611.19516 ↗