Internal Noise Interference to Warnings of Tipping Points in Generic Multidimensional Dynamical Systems

Andreas Morr, Niklas Boers, Peter Ashwin

TL;DR

This paper reveals a critical flaw in how we detect impending "tipping points" in complex systems. While standard methods rely on "critical slowing down" (CSD) indicators like variance and autocorrelation, this research shows that noise in other dimensions of a multi-dimensional system can mask or even reverse these expected signals, leading to false alarms or missed warnings. This is crucial because accurately predicting tipping points is vital for managing risks in climate, ecosystems, and financial markets.

What Problem Does This Paper Address?

Many natural and engineered systems can abruptly shift from one stable state to another – these are known as "tipping points." Predicting these shifts is crucial for avoiding catastrophic consequences. A common approach to detecting an impending tipping point is to look for "critical slowing down" (CSD). This phenomenon, observed in systems approaching a generic "fold-type" bifurcation, manifests as slower responses to disturbances, leading to increased variance and autocorrelation in time-series data. These are often called "early warning signals" (EWS).

The problem arises when we try to apply these EWS to multi-dimensional systems. While theory suggests that the dynamics near a fold-type bifurcation can be reduced to a single "critical" dimension (the center manifold), this simplification often overlooks the impact of noise. Specifically, if noise affects multiple dimensions simultaneously, it can interfere with the expected CSD signals in observable quantities, even if the underlying critical dimension is behaving as predicted. This paper investigates how this "internal noise interference" can lead to misleading EWS in generic multi-dimensional systems.

Methods

The authors employ a combination of theoretical analysis and numerical simulations.

  • Center Manifold Theory: They start by invoking center manifold theory, a standard tool in dynamical systems, to explain why multi-dimensional systems near a fold-type bifurcation can often be reduced to a one-dimensional normal form. This reduction is the theoretical basis for applying CSD indicators.
  • Stochastic Differential Equations (SDEs): The core of their analysis involves modeling the system dynamics using SDEs. They extend the one-dimensional Ornstein-Uhlenbeck process (a standard model for CSD) to a two-dimensional setting, incorporating a non-diagonal (anisotropic) noise coupling matrix. This allows them to explicitly model how noise in one dimension can influence another.
  • Linearized Dynamics: To derive analytical expressions for the statistical properties of observables, they analyze the linearized version of the SDE around the equilibrium point. This simplifies the mathematical treatment while still capturing the essential dynamics near the bifurcation.
  • Covariance and Autocorrelation Analysis: They derive closed-form solutions for the covariance matrix and time-correlation functions of linear observables. This allows them to precisely quantify how variance and AC(1) behave as the system approaches the bifurcation.
  • Numerical Integration: To validate their theoretical findings and demonstrate the phenomenon in a non-linear setting, they numerically integrate the full non-linear SDE using the Euler-Mayurama scheme.

Datasets

Not discussed in the paper. The study focuses on theoretical models and synthetic data generated from these models.

Key Insights and Ideas

  • Noise Interference is Ubiquitous: The paper demonstrates that even with uncoupled deterministic dynamics and stationary noise, cross-dimensional noise coupling can lead to misleading EWS. This phenomenon is not an edge case but a generic feature of multi-dimensional systems.
  • "Deceitful" Trends: The most striking finding is that observables can exhibit trends in variance and AC(1) that are opposite to what CSD would predict. Instead of increasing as a bifurcation approaches, these indicators can decrease, creating a false sense of stability or even suggesting a system is moving away from a tipping point when it is actually approaching one.
  • Observable Choice Matters: The "optimality" of an observable (i.e., its ability to reliably signal CSD) is highly dependent on its definition. While the "critical" dimension itself might show clear CSD, a linear combination of dimensions (a common scenario when direct access to the critical dimension is not possible) can obscure these signals.
  • False Alarms and Missed Warnings: This noise interference can lead to both false alarms (suggesting a tipping point is near when it's not) and missed warnings (failing to detect an impending tipping point).
  • Genericity of the Problem: The authors provide a theoretical result (Theorem A.1) suggesting that this phenomenon is generic for a wide range of parameters, meaning it's not confined to very specific system configurations.

Key Results

  • The paper analytically derives expressions for the variance and lag-1 autocorrelation (AC(1)) of linear observables in a generic two-dimensional fold-type bifurcation setting with anisotropic noise.
  • They show that for specific choices of observables and noise parameters, both variance and AC(1) can decrease as the system approaches a bifurcation (i.e., as a destabilizing parameter approaches zero), directly contradicting the expected CSD behavior.
  • Numerical simulations of a non-linear system confirm these theoretical predictions, demonstrating the practical relevance of the phenomenon.
  • The study highlights that while CSD indicators might still show an increasing trend in some observables, the presence of noise interference means that these indicators might not be as robust or reliable as previously assumed, especially in complex, multi-dimensional systems.

Limitations and Future Work

  • Focus on Linear Observables: The analytical derivations are primarily for linear observables. While the authors suggest the phenomenon is not limited to linear cases, further investigation into non-linear observables would be beneficial.
  • Two-Dimensional System: The detailed analytical treatment is for a two-dimensional system. While the principles are argued to be generalizable, specific quantitative results for higher-dimensional systems are not presented.
  • Stationary Noise: The analysis assumes stationary noise. The paper acknowledges that non-stationary or non-white noise can also complicate EWS, and future work could explore the interplay of these factors with multi-dimensional noise interference.
  • Noise-Induced Tipping: The paper touches upon noise-induced tipping but doesn't delve deeply into its prediction or mitigation in this multi-dimensional context.
  • Practical Application: While the paper identifies the problem, providing a general guideline for which observables are safe or how to mitigate the effect in real-world applications remains a challenge due to the dependence on specific parameter values. The authors provide a Python script to help assess the prevalence of the effect for different parameterizations, suggesting that further computational tools and empirical studies are needed.

The authors suggest that whenever possible, a thorough physical understanding and model validation are crucial to ensure that the conventional assumption of observing a directly noisy dynamical system along its center manifold is warranted. If this assumption holds, conventional EWS are likely still expressive. However, when it doesn't, the challenge of identifying reliable EWS in multi-dimensional systems becomes significantly more complex.

Summary updated

Abstract

A deterministic dynamical system that slowly passes through a generic fold-type (saddle-node) bifurcation can be reduced to one-dimensional dynamics close to the bifurcation because of the centre manifold theorem. It is often tacitly assumed that the same is true in the presence of stochasticity or noise so that, for example, critical slowing down (CSD) indicators can be applied as if the system were one-dimensional. In this work, we show that this is only true when given suitable system observables; specifically, we demonstrate that noise in other dimensions may interfere with indicators of CSD, also referred to as early warning signals (EWS). We point out a generic mechanism by which both variance and lag-1 autocorrelation [AC(1)], as well as other EWS, can fail to signal an approaching bifurcation. This can, in principle, occur whenever one noise source drives multiple system components simultaneously. Even under the favourable assumptions of uncoupled deterministic dynamics and stationary noise, some system observables can then exhibit false negative or false positive CSD indications. We isolate this phenomenon in an example that represents a generic two-dimensional fold-type bifurcation setting.
BibTeX
@article{morr2024internal,
  author = {Andreas Morr and Niklas Boers and Peter Ashwin},
  title = {{Internal Noise Interference to Warnings of Tipping Points in Generic Multidimensional Dynamical Systems}},
  journal = {SIAM J. Appl. Dyn. Syst.},
  year = {2024},
  doi = {10.1137/24m1669104},
}
Figure