Introduction
The Generalized Gaussian (GG) distribution family serves as a versatile framework in engineering and statistics, capable of modeling a wide spectrum of symmetric data distributions. By adjusting a single shape parameter $p$, the GG family encompasses the Laplace distribution ($p=1$), the standard Gaussian distribution ($p=2$), and approaches the Uniform distribution as $p \to \infty$. This flexibility allows for the modeling of both "heavy-tailed" data, characterized by frequent outliers, and "light-tailed" data.
Despite its widespread use in fields such as signal processing, communications, and economics, the deep analytical properties of the GG family have historically been fragmented across various literatures or left unexplored. This paper provides a systematic characterization of the GG family's moments, transforms, and decomposition properties, establishing a rigorous theoretical foundation for its application.
The Critical $p=2$ Threshold
A central finding in the analysis of GG distributions is the behavioral divergence that occurs at the shape parameter $p=2$. For $p \le 2$, the probability density function (pdf) behaves as a "positive definite function" and can be represented as a Gaussian mixture. This property is lost when $p > 2$, where the characteristic function begins to exhibit zero-crossings, indicating a fundamental shift in the distribution's underlying structure.
Analytical Properties and Decomposition
The Moment Problem
A significant contribution of this work is the resolution of the "moment problem" for GG distributions. It is demonstrated that for $p \ge 1$, a GG distribution is uniquely determined by its moments. However, for $p < 1$, the distribution becomes "indeterminate." In this regime, distinct distributions can share the same set of integer moments, which serves as a critical caveat for researchers relying on the method of moments for parameter estimation.
Product Decomposition
The paper establishes that any GG random variable with $p \le 2$ can be decomposed into the product of a standard Gaussian variable and an independent positive random variable. This representation as a "Gaussian Mixture" significantly simplifies complex engineering problems by allowing researchers to leverage the well-understood properties of the normal distribution.
Stochastic Ordering
The GG family is shown to be naturally ordered by second-order stochastic dominance. As the shape parameter $p$ increases (while maintaining constant variance), the distribution becomes mathematically "less risky" or less spread out.
Key Results in Distribution Theory
Infinite Divisibility
The study provides a complete characterization of infinite divisibility within the GG family. A GG distribution is infinitely divisible—meaning it can be represented as the sum of $n$ independent, identically distributed variables for any $n$—if and only if $p \in (0, 1] \cup \{2\}$.
Characteristic Function and Tail Behavior
The analysis of the characteristic function reveals that for $p \le 2$, the function remains strictly positive with no zeros. Conversely, for $p > 2$, at least one zero-crossing must occur. Furthermore, the paper derives asymptotically tight bounds for the characteristic function, describing its decay as the frequency $t \to \infty$. It is also shown that the characteristic function is "real analytic" for $p > 1$, but loses this property when $p < 1$.
Conclusion and Future Directions
While this work fills significant gaps in the univariate theory of Generalized Gaussian distributions, several avenues for future research remain. The existence of alternative decompositions for the light-tailed regime ($p > 2$) is currently unknown. Additionally, extending these analytical results to multivariate GG vectors and developing numerically stable algorithms for evaluating the pdf of mixing variables are essential steps for broader practical application.
Summary updated