Information-theoretical analysis of statistical measures for multiscale dynamics

Naoki Asuke,Tomoki Yamagami,Takatomo Mihana,André Röhm,Ryoichi Horisaki,Makoto Naruse

DOI: https://doi.org/10.1063/5.0141099

2023-01-05

Abstract:Multiscale entropy (MSE) has been widely used to examine nonlinear systems involving multiple time scales, such as biological and economic systems. Conversely, Allan variance has been used to evaluate the stability of oscillators, such as clocks and lasers, ranging from short to long time scales. Although these two statistical measures were developed independently for different purposes in different fields in the literature, their interest is to examine multiscale temporal structures of physical phenomena under study. We show that, from an information-theoretical perspective, they share some foundations and exhibit similar tendencies. We experimentally confirmed that similar properties of the MSE and Allan variance can be observed in low-frequency fluctuations (LFF) in chaotic lasers and physiological heartbeat data. Furthermore, we calculated the condition under which this consistency between the MSE and Allan variance exists, which is related to certain conditional probabilities. Heuristically, physical systems in nature including the aforementioned LFF and heartbeat data mostly satisfy this condition, and hence the MSE and Allan variance demonstrate similar properties. As a counterexample, an artificially constructed random sequence is demonstrated, for which the MSE and Allan variance exhibit different trends.

Chaotic Dynamics

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to explore the relationship between multiscale entropy (MSE) and Allan variance, as well as their consistency and differences in analyzing physical phenomena at different time scales. Specifically:

1. **Multiscale Entropy (MSE)** and **Allan Variance** are two statistics used to analyze dynamic systems containing multiple time scales. MSE is mainly used to evaluate nonlinear systems, such as those in biology and economics; while Allan variance is mainly used to evaluate the stability of oscillators (such as atomic clocks or lasers).

2. **The core of the problem**: Although these two statistics have been independently developed in different fields, they both aim to characterize dynamic systems containing multiple time scales. The paper attempts to explore the similarities and differences between these two statistics from the perspective of information theory and reveal the underlying mechanisms.

3. **Experimental verification**: The author experimentally verified that in the chaotic laser in low - frequency fluctuations (LFF) and physiological heartbeat data, MSE and Allan variance exhibit similar properties.

4. **Conditional probability**: The paper further explored the mechanism of the consistency between MSE and Allan variance under certain conditions. These conditions are mainly related to certain conditional probabilities. The author also constructed an artificial random sequence that violates these conditions to show the different trends of MSE and Allan variance in this case.

5. **Application background**: Studying the statistical characteristics of multiscale dynamic systems is crucial for understanding physics - based computing and communication technologies (such as reservoir computing, secure communication based on laser chaos, and laser networks for solving reinforcement learning problems).

In summary, the main objective of this paper is to reveal the similarities and differences between MSE and Allan variance in analyzing multiscale dynamic systems from the perspective of information theory, and verify these findings through experimental and theoretical analysis.