🌀 nonlinear sciences

Noise Effects on Ordinal Pattern Statistics via Majorization

This paper employs majorization theory to characterize and correct noise-induced distortions in ordinal pattern statistics, enabling robust dynamical classification and noise quantification without fitting procedures, as demonstrated by analyzing paleomagnetic records to distinguish between stochastic and chaotic Earth dipole evolution.

Original authors: Facundo Sapienza

Published 2026-08-25

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Facundo Sapienza

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Time series analysis is the art of reading the story hidden within a sequence of numbers. Whether tracking the daily temperature, the fluctuating price of a stock, or the rhythmic beating of a heart, scientists look for patterns that reveal whether a system is behaving randomly or following a hidden, complex rule. For decades, researchers have relied on a specific method called ordinal pattern analysis to make this distinction. Instead of worrying about the exact height of a wave or the precise value of a measurement, this method simply looks at the order of the numbers. It asks a simple question: did the value go up, then down, then up again? By reducing complex data to these sequences of rising and falling, scientists can distinguish between chaotic systems, which follow deterministic rules but appear unpredictable, and purely random noise. This approach has become a standard tool because it is robust and does not require knowing the underlying equations of the system.

However, real-world data is rarely perfect. Observations are often tainted by static, missing measurements, or irregular timing, much like trying to hear a conversation through a wall. This observational noise distorts the order of the numbers, blurring the line between a chaotic system and pure randomness. When scientists try to classify a noisy signal, the noise can push the data into the wrong category, leading to incorrect conclusions about how the system works. This is a particularly acute problem in fields like geophysics, where data is often sparse, irregular, and buried under a low signal-to-noise ratio. The challenge has been to find a way to see through the static to understand the true nature of the signal without needing to perfectly model the noise itself.

In a new study, researchers have developed a mathematical framework to solve this problem by using a concept known as majorization. Rather than trying to guess the exact amount of noise or fit a complex model to the data, the team demonstrated that noise has a predictable, one-way effect on the statistical order of a time series. They found that when noise is added to a signal, the resulting pattern of orders becomes more "mixed" or disordered than the original, noise-free signal. In mathematical terms, the noisy distribution is always "majorized" by the clean one. This means that if you were to line up the probabilities of all possible patterns from highest to lowest, the cumulative total for the noisy data would always fall below the cumulative total for the clean data. This relationship holds true for strictly stationary processes with absolutely continuous distributions, covering a wide range of practical noise models including independent noise and specific types of correlated noise, though it does not extend to all possible generic correlation structures.

The researchers proved this relationship theoretically for these specific conditions, showing that this majorization effect pushes the statistical signature of any time series toward a state of uniform randomness. In the visual language used by scientists to map these systems, known as the complexity-entropy causality plane, this means that noise pushes a data point in a specific, predictable direction. It moves the point toward the center of the map, where pure randomness lives, and away from the edges where distinct chaotic or stochastic behaviors are found. This discovery allows scientists to define a specific "admissible zone" on the map. If a noisy data point falls within this zone, it is consistent with being a noisy version of a specific clean system. If it falls outside, the system cannot be explained by that model, no matter how much noise is added.

To make this theory useful for real data, the team created a statistical test that can determine if a noisy dataset is consistent with a clean reference model. This test accounts for the fact that real-world data is always limited in size, ensuring that the conclusions are not just artifacts of small sample sizes. Furthermore, they developed a method to estimate the upper limit of the noise level. By gradually adding simulated noise to a reference model and checking when the majorization relationship breaks, they can pinpoint the maximum amount of noise that could plausibly be present in the observations. This provides a bound on the noise without requiring any complex fitting procedures or assumptions about the original signal.

The power of this approach was demonstrated by applying it to a long-standing mystery in geophysics: the evolution of Earth's magnetic dipole. Paleomagnetic records, which are essentially time series of the Earth's magnetic field intensity over millions of years, are notoriously difficult to analyze. They are sparse, irregularly sampled, and contain significant noise. The central question has been whether the fluctuations in the Earth's magnetic field are driven by a chaotic system, which follows deterministic rules, or by a stochastic process, which is driven by random chance. Both types of models have been proposed in the past, and both can produce data that looks similar to the geological record.

Using their new framework, the researchers illustrated the methodology by analyzing these ancient magnetic records to determine if the geological evolution of the Earth dipole is better described by a stochastic or chaotic system. By establishing the admissible zones for both types of systems and checking where the noisy geological data actually falls, they could determine which underlying dynamic was more likely. This work offers a new way to interpret imperfect data, allowing scientists to distinguish between chaos and randomness even when the signal is weak and the measurements are flawed. It suggests that the Earth's magnetic evolution might be better described by one type of system over the other, providing a clearer picture of the planet's deep history without the need for perfect data.

Technical Summary: Noise Effects on Ordinal Pattern Statistics via Majorization

Problem Statement
The Bandt-Pompe permutation entropy framework and the complexity-entropy causality plane are standard tools for characterizing the dynamical properties of time series, particularly for distinguishing between chaotic and stochastic dynamics. However, these methods are sensitive to observational noise and irregular sampling, which are common in real-world applications such as geophysics. Noise distorts ordinal pattern probability distributions, systematically shifting time series within the causality plane and potentially compromising dynamical classification. Existing literature on the robustness of ordinal patterns is often limited to scenarios with small noise levels or very long time series, leaving a gap in understanding how noise affects these statistics under realistic, sparse, and noisy measurement conditions.

Methodology
This work addresses the problem of characterizing ordinal patterns in the presence of observational noise using the mathematical formalism of majorization.

Ordinal Patterns and Noise: The study considers a time series $x_t$ corrupted by additive noise $\epsilon_t$ to produce an observation $y_t = x_t + \epsilon_t$ . The ordinal pattern distribution $p$ is derived from the noiseless series, while $\tilde{p}$ is derived from the noisy series.
Majorization Framework: The author utilizes the concept of majorization ( $\prec$ ), a partial order on probability vectors. A distribution $\tilde{p}$ is majorized by $p$ ( $\tilde{p} \prec p$ ) if the cumulative sums of the sorted components of $\tilde{p}$ are less than or equal to those of $p$ . Geometrically, this implies the Lorenz curve of the noisy distribution lies below (closer to the uniform distribution) the Lorenz curve of the noiseless distribution.
Theoretical Result (Theorem 1): The paper proves that under specific conditions—specifically for strictly stationary processes where the distribution of sorted gaps is independent of the permutation order, and where the noise conditional distribution is invariant under permutation of components—the noisy ordinal pattern distribution is majorized by the noiseless one ( $\tilde{p} \prec p$ ). This holds for independent identically distributed noise and certain equicorrelated Gaussian noise models.
Implications for Complexity-Entropy: Since permutation entropy and Jensen-Shannon divergence are Schur-concave functions, the majorization relation implies that noise increases entropy and alters complexity in a predictable direction. Specifically, the noisy point in the complexity-entropy causality plane lies to the right of and below the line connecting the origin to the noiseless point, defining a strict "admissible zone" for noisy observations.
Statistical Testing: To handle finite sample sizes, the author proposes a hypothesis test for majorization. The null hypothesis $H_0: p \succ q$ is tested against $H_1: p \not\succ q$ . The test employs a generalized moment selection (GMS) approach to handle composite null hypotheses defined by inequality constraints, using bootstrap resampling to calibrate the test statistic and avoid inflation from slack constraints.
Noise Quantification: The author introduces a method to estimate an upper bound for observational noise ( $\hat{\sigma}^*$ ). By incrementally adding noise to a reference model until the majorization test fails, they determine the maximum noise level compatible with the observed data without requiring explicit model fitting or parameter inference.

Key Contributions

Theoretical Proof: A formal proof establishing that observational noise generates ordinal pattern probabilities majorized by the noiseless distribution under a broad class of noise models, alongside numerical demonstrations for more generic cases.
Hypothesis Testing: A robust statistical test adapted for finite-sample regimes to determine if an observed distribution is consistent with a reference distribution under the majorization criterion.
Noise Quantification: A methodology to quantify the upper bound of observational noise without fitting procedures, relying solely on the ordinal pattern statistics and the majorization relation.
Application to Paleomagnetism: The framework is applied to paleomagnetic records to analyze the geological evolution of the Earth's magnetic dipole, specifically to determine if intensity fluctuations are better described by stochastic or chaotic dynamics.

Results

Synthetic Benchmarks: The methodology was validated on five synthetic benchmarks: the Hénon map, logistic map, Lorenz system, Ornstein–Uhlenbeck process, and fractional Brownian motion.
- In all cases, increasing noise levels pushed the Lorenz curves toward the uniform distribution, consistent with the majorization relation.
- The majorization relationship held even for colored (autoregressive) noise, which falls outside the strict conditions of Theorem 1.
- The hypothesis test correctly identified true majorization relationships (high p-values) and rejected false ones (low p-values) in finite-sample regimes.
- The noise upper bound estimation ( $\hat{\sigma}^*$ ) successfully recovered the true noise level ( $\sigma_0$ ) as the sample size increased, converging to the ground truth without parametric inference.
Paleomagnetic Application: The author applied the framework to paleomagnetic records to investigate the nature of Earth's dipole intensity fluctuations. The analysis aimed to distinguish between chaotic and stochastic reduced-order models, both of which are consistent with observed reversal rates and interval distributions. The study utilized the proposed methodology to characterize these records under realistic sparse and noisy conditions.

Significance and Claims
The paper claims that majorization offers a more localized and rigorous view of ordinal pattern distributions than standard information-theoretic metrics alone. By establishing a strict ordering between noiseless and noisy distributions, the work provides a theoretical justification for the behavior of ordinal patterns under noise and defines admissible zones in the complexity-entropy causality plane.

The author positions this work as a tool to restore discriminability in dynamical classification under realistic measurement conditions. They highlight the utility of their noise quantification method, which operates without the need for fitting procedures, making it particularly suitable for fields like geophysics where data is often sparse, irregularly sampled, and noisy. The application to paleomagnetism serves as a motivating demonstration of how ordinal patterns can be effectively used in fields characterized by imperfect observations to address fundamental questions about system dynamics.

Technical Summary: Noise Effects on Ordinal Pattern Statistics via Majorization

More like this