Abstract
We propose an extension of a result by Repetowicz et al. regarding Wick's theorem and its applications. We demonstrate that Wick's theorem can be extended first to the uniform distribution on the sphere and subsequently to the entire class of elliptical distributions. Furthermore, we detail this result for distributions that are Gaussian scale mixtures. Finally, we show that these results allow for a simplified recovery of Folland’s theorem concerning the integration of polynomials over the sphere.
Introduction
In fields ranging from quantum field theory to multivariate statistics and financial modeling, the calculation of "mixed moments"—the expected value of a product of random variables, such as $E[X_1 X_2 X_3 X_4]$—is a fundamental task. For Gaussian (normal) distributions, this task is simplified by Wick’s theorem (also known as Isserlis' theorem). The theorem provides a combinatorial recipe: the mixed moment of a multivariate Gaussian vector is equal to the sum of the products of the covariances of all possible pairings of the variables.
However, empirical data often exhibit non-Gaussian characteristics, such as heavy tails or spherical symmetries that do not align with the bell curve. Traditionally, calculating moments for these distributions requires grueling integration. This research addresses this gap by extending the "Wick-like" shortcut to a much broader family of distributions, including spherical, elliptical, and Gaussian scale mixtures.
Theoretical Framework: Stochastic Representation
The core methodology relies on stochastic representation. Rather than tackling complex integrals directly, the authors represent non-Gaussian random vectors as transformations of simpler, well-understood vectors.
1. Spherical Distributions
The derivation begins with a vector $U$ uniformly distributed on a unit sphere $S_{n-1}$. By representing $U$ as a normalized Gaussian vector ($U = N/\|N\|$), the authors exploit the independence between a Gaussian vector's direction and its magnitude. This allows them to derive a scaling factor composed of Gamma functions that adjusts the standard Gaussian Wick formula for the spherical case.
2. Elliptical Distributions
Elliptical distributions are essentially affine transformations (stretching or compressing) of spherical distributions. By utilizing the linear properties of the expectation operator, the authors extend the spherical result to the entire elliptical class, maintaining the combinatorial pairing structure.
3. Gaussian Scale Mixtures
The framework is further generalized to Gaussian scale mixtures, where a Gaussian vector is scaled by an independent random variable (e.g., $X = N/\sqrt{a}$). This generalization is particularly significant as it covers widely used distributions such as the Student-t and Cauchy distributions.
Key Insights
The Persistence of Pairings
The most striking insight of this work is that the fundamental logic of Wick’s theorem—summing over all possible pairings ($\sigma \in \Pi_{2m}$)—remains valid beyond the Gaussian domain. The combinatorial structure of the problem does not change; only the coefficients do.
The Correction Factor
The difference between a Gaussian moment and a spherical or elliptical moment is reduced to a single multiplicative constant. This "volume correction" (a ratio of expectations or Gamma functions) means that researchers do not need an entirely new theorem for non-Gaussian cases; they simply need to apply a specific scaling factor to the existing Wick formula.
Unification of Folland’s Theorem
The paper demonstrates that a well-known result by Folland regarding the integration of polynomials over a sphere is a special case of this generalized Wick’s theorem. By framing Folland's result within this new context, the authors simplify a complex integration problem into a straightforward application of combinatorial pairings.
Key Results and Formulas
Spherical Wick Formula: The paper provides a concrete formula for the moments of a uniform distribution on a sphere $S_n$: $$E \prod_{i=1}^{2m} \mathcal{L}_i (U) = \frac{\Gamma(\frac{n}{2})}{2^m \Gamma(m + \frac{n}{2})} \sum_{\sigma} \prod l_i^t l_{\sigma(i)}$$
Elliptical Extension: For any elliptical vector $X$, the moment is shown to be the standard Gaussian Wick result multiplied by the ratio $\frac{E a^{2m}}{E r^{2m}}$, where $a$ represents the scaling factor and $r$ is the radius of the underlying spherical distribution.
Broad Applicability: The extension successfully covers Gaussian scale mixtures, provided the scaling variable $a$ possesses finite moments of the required order ($E a^{-m} < \infty$).
Limitations and Conclusion
While this extension significantly broadens the utility of Wick's theorem, there are important constraints:
- Existence of Moments: For extremely heavy-tailed distributions, high-order moments may not exist. The theorem is only applicable if the scaling variable $a$ has a finite moment of order $-m$.
- Zero Mean Assumption: The current derivations assume vectors have a mean of zero. While standard for theoretical proofs, applying these results to non-zero mean elliptical distributions would require the inclusion of additional terms to account for non-central moments.
In conclusion, this work provides a unified mathematical framework that simplifies the calculation of mixed moments for a wide variety of distributions. By showing that the combinatorial heart of Wick's theorem is a property of symmetry rather than just the Gaussian density, the authors provide a powerful tool for theoretical physics and multivariate statistics.
Summary updated