Abstract:We show how a Cramér-Wold theorem for a family of multivariate probability distributions can be used to generate a similar theorem for mixtures (convex combinations) of distributions drawn from the same family. Using this abstract result, we establish a Cramér-Wold theorem for mixtures of multivariate Gaussian distributions. According to this theorem, two such mixtures can be distinguished by projecting them onto a certain predetermined finite set of lines, the number of lines depending only on the total number Gaussian distributions involved and on the ambient dimension. A similar result is also obtained for mixtures of multivariate $t$-distributions.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is about how to use the Cramér -Wold theorem to deal with the projection problem of mixed distributions (i.e., convex combinations of probability measures). Specifically, the authors hope to prove whether for certain types of mixed distributions (such as multivariate Gaussian distributions and multivariate t -distributions), these mixed distributions can be uniquely determined by a finite number of projections. ### Core problems of the paper 1. **Generalization of the Cramér -Wold theorem**: - The classical Cramér -Wold theorem states that a probability measure in Euclidean space can be uniquely determined by its one -dimensional projections in all directions. - The goal of this paper is to generalize this result to the case of mixed distributions, especially the mixtures of multivariate Gaussian distributions and multivariate t -distributions. 2. **Uniqueness of mixed distributions**: - For a given family of mixed distributions, how to distinguish different mixed distributions by a finite number of projections. - Specifically, the authors are concerned with how to choose appropriate projection lines so that if two mixed distributions are the same on this set of projection lines, then they themselves must also be the same. ### Main conclusions - **Cramér -Wold theorem for Gaussian mixture distributions**: - If two Gaussian mixture distributions have the same one -dimensional projections on a pre - determined finite set of lines, then these two mixed distributions are the same. The number of required projection lines only depends on how many Gaussian distributions there are in total and the ambient dimension $d$. \[ \text{The number of required projection lines}=\frac{1}{2}(\ell + m - 1)(d^{2}+d - 2)+1 \] - **Similar results for t -distribution mixtures**: - Similar results also apply to the mixtures of multivariate t -distributions. ### Methods and techniques - **Linear independence and identifiability**: - The authors first prove the linear independence (or equivalently, identifiability) of Gaussian distributions and t -distributions. This is achieved through complex mathematical analysis and the Borel theorem in complex analysis. - **Application of the abstract theorem**: - The authors establish an abstract theorem (Theorem 2.1), which shows how to derive the Cramér -Wold theorem for the mixed distributions of a family of probability measures from the Cramér -Wold theorem of this family of measures. ### Application prospects - These results have wide applications in statistics, especially in the fields of density estimation, machine learning, and clustering. In addition, these theoretical achievements may help simplify the model estimation process of high -dimensional data, especially when using the Markov chain Monte Carlo method in the Bayesian framework. In conclusion, this paper solves the problem of how to uniquely determine certain types of mixed distributions by introducing new theoretical tools and methods through a finite number of projections, which is of great significance for the fields of statistics and data analysis.

A Cramér-Wold theorem for mixtures

A formulation for continuous mixtures of multivariate normal distributions

Strong Laws of Large Numbers for Ρ˜-Mixing Random Variables

Universal Approximation on the Hypersphere

Complete convergence for ρ--mixing random fields with different distributions

Coag-Frag duality for a class of stable Poisson-Kingman mixtures

A Wasserstein-Type Distance for Gaussian Mixtures on Vector Bundles with Applications to Shape Analysis

Approximate independence of permutation mixtures

On the entropy and information of Gaussian mixtures

Wasserstein-p Bounds via Cumulant-Based Edgeworth Expansion for $α$-Mixing Random Fields

Multivariate Scale-Mixed Stable Distributions and Related Limit Theorems

A Probability Inequality for Convolutions of MTP2-Distribution Functions

The Aldous--Hoover Theorem in Categorical Probability

The Central Limit Theorem for uniformly strong mixing measures

Universality for Random Matrices

Universality of the Wigner-Gurau limit for random tensors

On mixtures of copulas and mixing coefficients

A generalized Budan-Fourier approach to generalized Gaussian and exponential mixtures

Universality laws for Gaussian mixtures in generalized linear models*

Yurinskii's Coupling for Martingales

Ratios and Cauchy Distribution