Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

On the Cyclostationary Linear Inverse Models: A Mathematical Insight and Implication

Justin Lien, Yan-Ning Kuo, Hiroyasu Ando

TL;DR

This work advances the analysis of cyclostationary stochastic systems by introducing two variants of cyclostationary linear inverse models, $e$-CS-LIM$ and $l$-CS-LIM, built on interval-wise and pointwise linear Markov approximations, coupled with a periodic fluctuation-dissipation framework. It formalizes how the time-varying dynamics are encoded in the correlation structure via $\partial_s\mathbf{K}(s,t)|_{s=0}$ and establishes the theoretical linkages to the classical LIM. Through 1D and higher-dimensional numerical experiments and an ENSO-focused real-world case using Niño 3.4 SST data, the authors show that $e$-CS-LIM provides robust, phase-accurate dynamics estimation, while $l$-CS-LIM excels in diffusion estimation albeit with noise that can be mitigated by filtering. The results demonstrate that the CS-LIM family extends the LIM toolkit for climate and other cyclostationary systems, enabling faithful reconstruction of temporal structure and extreme-event likelihoods from data.

Abstract

Cyclostationary linear inverse models (CS-LIMs), generalized versions of the classical (stationary) LIM, are advanced data-driven techniques for extracting the first-order time-dependent dynamics and random forcing relevant information from complex non-linear stochastic processes. Though CS-LIMs lead to a breakthrough in climate sciences, their mathematical background and properties are worth further exploration. This study focuses on the mathematical perspective of CS-LIMs and introduces two variants: e-CS-LIM and l-CS-LIM. The former refines the original CS-LIM using the interval-wise linear Markov approximation, while the latter serves as an analytic inverse model for the linear periodic stochastic systems. Although relying on approximation, e-CS-LIM converges to l-CS-LIM under specific conditions and shows noise-robust performance. Numerical experiments demonstrate that each CS-LIM reveals the temporal structure of the system. The e-CS-LIM optimizes the original model for better dynamics performance, while l-CS-LIM excels in diffusion estimation due to reduced approximation reliance. Moreover, CS-LIMs are applied to real-world ENSO data, yielding a consistent result aligning with observations and current ENSO understanding.

On the Cyclostationary Linear Inverse Models: A Mathematical Insight and Implication

TL;DR

This work advances the analysis of cyclostationary stochastic systems by introducing two variants of cyclostationary linear inverse models, -CS-LIMl\partial_s\mathbf{K}(s,t)|_{s=0}el$-CS-LIM excels in diffusion estimation albeit with noise that can be mitigated by filtering. The results demonstrate that the CS-LIM family extends the LIM toolkit for climate and other cyclostationary systems, enabling faithful reconstruction of temporal structure and extreme-event likelihoods from data.

Abstract

Cyclostationary linear inverse models (CS-LIMs), generalized versions of the classical (stationary) LIM, are advanced data-driven techniques for extracting the first-order time-dependent dynamics and random forcing relevant information from complex non-linear stochastic processes. Though CS-LIMs lead to a breakthrough in climate sciences, their mathematical background and properties are worth further exploration. This study focuses on the mathematical perspective of CS-LIMs and introduces two variants: e-CS-LIM and l-CS-LIM. The former refines the original CS-LIM using the interval-wise linear Markov approximation, while the latter serves as an analytic inverse model for the linear periodic stochastic systems. Although relying on approximation, e-CS-LIM converges to l-CS-LIM under specific conditions and shows noise-robust performance. Numerical experiments demonstrate that each CS-LIM reveals the temporal structure of the system. The e-CS-LIM optimizes the original model for better dynamics performance, while l-CS-LIM excels in diffusion estimation due to reduced approximation reliance. Moreover, CS-LIMs are applied to real-world ENSO data, yielding a consistent result aligning with observations and current ENSO understanding.
Paper Structure (14 sections, 5 theorems, 25 equations, 6 figures, 2 algorithms)

This paper contains 14 sections, 5 theorems, 25 equations, 6 figures, 2 algorithms.

Table of Contents

  1. Introduction
  2. e-CS-LIM
  3. The classical LIM
  4. The periodic fluctuation-dissipation relation
  5. Algorithm
  6. l-CS-LIM
  7. Mathematical background
  8. Algorithm
  9. The relationship among LIMs
  10. Numerical Experiments
  11. An ideal Lg-dimensional case study
  12. A higher-dimensional case study
  13. A real-world example: Niño 3.4 SST Index
  14. Conclusion

Key Result

Theorem 2.1

With the notation as above, the probability distribution $P(x,t) = \langle \delta({\bf x}(t)-x)\rangle$ of the stochastic process ${\bf x}$ satisfies where $\textbf{L}_{\textit{FP}}(t)$ is the Fokker-Planck operator at $t$.

Figures (6)

  • Figure 1: The stencil configurations. The blue dots on the top indicate the sampling time and the yellow and purple in the bottom are the time coordinates of the covariance function and the other three variables, respectively. \newlabelFig:Stencil0
  • Figure 1: The ground truth and the LIM results for a sample path of \ref{['Eq:Periodic-Diffusion-Process']} with $T_f = 5000$. For the results of $l$-CS-LIM, the dashed lines are ${A_{\textit{l}}}$ and ${Q_{\textit{l}}}$; the solid lines are $A_\text{MA}$ and $Q_\text{MA}$ sampled on ${T_{\textit{e}}}$.
  • Figure 2: The distribution of the relative errors of sine wave fitting for each model.
  • Figure 3: The distribution of $\phi$ for each model. The CS-LIM, $e$-CS-LIM, and $l$-CS-LIM from the top to the bottom are represented by yellow, orange, and blue, respectively.
  • Figure 4: The reconstruction by $l$-CS-LIM and the application of filters.
  • ...and 1 more figures

Theorems & Definitions (7)

  • Theorem 2.1: Fokker-Plank equation
  • Corollary 2.2: The classical fluctuation-dissipation relation
  • Theorem 2.3
  • Theorem 2.4
  • Proof 1
  • Theorem 3.1
  • Proof 2