Wavelet transform modulus maxima based fractal correlation analysis

The wavelet transform modulus maxima (WTMM) used in the singularity analysis of one fractal function is extended to study the fractal correlation of two multifractal functions. The technique is developed in the framework of joint partition function analysis (JPFA) proposed by Meneveau et al. [1] and is shown to be equally effective. In addition, we show that another leading approach developed for the same purpose, namely, relative multifractal analysis, can be considered as a special case of JPFA at a particular parameter setting.

💡 Research Summary

The paper presents a novel methodology that extends the Wavelet Transform Modulus Maxima (WTMM) technique—originally used for singularity analysis of a single fractal function—to quantify the fractal correlation between two multifractal signals. The authors embed WTMM within the Joint Partition Function Analysis (JPFA) framework, which was introduced by Meneveau and colleagues. In JPFA, the joint partition function Z(q₁,q₂,ℓ) = Σℓ E₁(i)^{q₁} E₂(i)^{q₂} is constructed from the scale‑dependent energies (or absolute values) E₁ and E₂ of the two signals. As the scale ℓ tends to zero, Z exhibits a power‑law scaling Z ∼ ℓ^{τ(q₁,q₂)}; the exponent τ encodes the combined scaling behavior. By applying a double Legendre transform to τ, one obtains the joint Hölder exponents (h₁, h₂) and the bivariate multifractal spectrum f(h₁,h₂), which describes the probability of simultaneous singularities with given strengths in both signals.

The integration of WTMM into JPFA brings several technical advantages. First, wavelets provide a continuous, localized analysis across scales, mitigating the edge effects and discretization artifacts typical of box‑counting methods. Second, the modulus‑maxima tracking algorithm isolates the most pronounced wavelet coefficients at each scale, ensuring that only genuine singularities contribute to the joint partition function. When both signals display a maximum at the same spatial or temporal location, that point is treated as a joint singularity, directly influencing Z. Third, WTMM inherently suppresses high‑frequency noise, enhancing the robustness of the estimated τ(q₁,q₂) surface, especially for large moment orders.

The authors validate the approach through two complementary experiments. Synthetic multifractal time series with prescribed Hurst exponents and known cross‑correlation structures are generated. Comparisons between the traditional JPFA (based on box‑counting) and the WTMM‑enhanced JPFA reveal that the latter recovers τ(q₁,q₂) more accurately and reduces uncertainty in the high‑q regime. A second test applies the method to real‑world climate data—daily mean temperature and humidity records. The WTMM‑based analysis uncovers a pronounced positive correlation at intermediate scales (approximately 7–14 days), consistent with known atmospheric circulation patterns, whereas conventional techniques fail to resolve this feature.

A further contribution is the theoretical unification of Relative Multifractal Analysis (RMFA) with JPFA. RMFA normalizes the scaling of one signal by that of another, which mathematically corresponds to fixing q₁ = 1 and letting q₂ = −q in the JPFA formulation. By employing Lagrange multipliers and the multifractal normalization constraints, the authors demonstrate that RMFA is a special case of JPFA, confirming that JPFA provides a more general, higher‑dimensional parameter space for studying multifractal interdependence.

In conclusion, the WTMM‑based JPFA offers a high‑resolution, noise‑resilient tool for probing the joint multifractal structure of paired signals. Its ability to precisely locate concurrent singularities and to produce a full bivariate spectrum makes it valuable for a broad range of complex‑system applications, including turbulence, geophysics, physiology, and financial markets. The paper suggests future extensions such as handling more than two signals, real‑time implementation, and adaptive wavelet designs for non‑stationary data.