Alignment-Driven Non-Redundancy in Early Warning Signals: Monte Carlo Evidence for a Geometric Indicator in Fold Transitions
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
Classical early warning signals (EWS) for critical transitions—most prominently variance and lag-1 autocorrelation (AR(1))—are derived from the covariance structure of observed processes and therefore share a common informational substrate. While effective under additive Gaussian noise, this covariance-based foundation can degrade under non-Gaussian or state-dependent noise, where variance may lose monotonic trend significance in the pre-transition regime. This work introduces V(θ) = Φ·A*, a geometric early warning indicator for fold-type bifurcations that combines a compression functional Φ (normalized spectral entropy of the delay-embedding covariance matrix) with an alignment functional A (cosine similarity between the dominant eigenvector of that covariance and the local deterministic drift gradient). The formulation is motivated by information geometry: near a fold bifurcation, degeneration of the statistical manifold involves directional structure that is not captured by scalar covariance summaries. A Monte Carlo study (n = 50 independent realizations per regime) of the canonical fold system dx = (θ − x²)dt + σ(x)dW is conducted under both additive Gaussian noise and multiplicative state-dependent noise. Across all seeds and both regimes, correlations between V*(θ) and variance are near zero (Gaussian: μ = −0.052; multiplicative: μ = −0.093; p < 0.001), with consistent near-zero rank correlations, supporting statistical non-redundancy beyond linear association. Component analysis identifies the alignment functional A as the primary source of non-redundancy, while Φ exhibits partial and variable redundancy with covariance-based indicators. Under multiplicative noise, variance loses monotonic trend significance (Kendall τ not significant), whereas V*(θ) retains a robust monotonic trend toward the transition (p < 0.001). In addition, V*(θ) yields substantially lower false alarm rates (~0.02) compared to AR(1) (~0.51) in the pre-transition regime. However, V*(θ) does not provide earlier detection than AR(1), lagging by approximately 0.10 θ units across both regimes. The contribution of V*(θ) is therefore one of specificity rather than sensitivity: it provides geometrically grounded, statistically non-redundant information complementary to covariance-based EWS, particularly under multiplicative noise conditions where classical indicators degrade. Claims are restricted to fold-type transitions in one-dimensional driven stochastic systems; no universality beyond this setting is asserted. early warning signals, fold bifurcation, geometric indicators, information geometry, Fisher information, alignment functional, spectral entropy, delay embedding, multiplicative noise, non-Gaussian noise, covariance non-redundancy, eigenvector geometry, directional information, Kendall tau, false alarm rate, Monte Carlo simulation
Fields of Study & Subjects
Ecosystem dynamics and resilience
Chaos control and synchronization
Mathematical and Theoretical Epidemiology and Ecology Models
Publication details
- Journal / source
- Zenodo (CERN European Organization for Nuclear Research)
- Publication date
- 31 March 2026
- Language
- EN
- Document type
- Preprint