Abstract
We show that any distribution function on $\mathbb{R}^d$ with nonnegative, nonzero and integrable marginal distributions can be characterized by a norm on $\mathbb{R}^{d+1}$, called F-norm. We characterize the set of F-norms and prove that pointwise convergence of a sequence of F-norms to an F-norm is equivalent to convergence of the pertaining distribution functions in the Wasserstein metric. On the statistical side, an F-norm can easily be estimated by an empirical F-norm, whose consistency and weak convergence we establish. The concept of F-norms can be extended to arbitrary random vectors under suitable integrability conditions fulfilled by, for instance, normal distributions. The set of F-norms is endowed with a semigroup operation which, in this context, corresponds to ordinary convolution of the underlying distributions. Limiting results such as the central limit theorem can then be formulated in terms of pointwise convergence of products of F-norms. We conclude by showing how, using the geometry of F-norms, we may characterize nonnegative integrable distributions in $\mathbb{R}^d$ by simple compact sets in $\mathbb{R}^{d+1}$. We then relate convergence of those distributions in the Wasserstein metric to convergence of these characteristic sets with respect to Hausdorff distances.
1. Introduction
In probability and statistics, describing the relationship between multiple variables in multivariate distributions is a complex task. Traditional tools like characteristic functions (Fourier transforms) primarily track weak convergence, which often fails to capture the full geometric structure or the "transport cost" between distributions.
This paper introduces F-norms, a novel class of mathematical norms that uniquely characterize multivariate probability distributions with nonnegative, integrable components. By mapping complex distributions to geometric objects, the authors demonstrate that convergence in the Wasserstein metric—a powerful measure of distance between distributions—is equivalent to simple pointwise convergence of these norms.
2. Methodology: The F-norm Framework
The authors define the F-norm of a random vector $\mathbf{X}$ as: $$\|\mathbf{x}\|_F = E[\max(|x_0|, |x_1|X_1, \dots, |x_d|X_d)]$$ where $\mathbf{x} \in \mathbb{R}^{d+1}$. This definition builds upon "D-norms" used in extreme value theory but addresses their lack of uniqueness; Theorem 2.2 establishes that the mapping between a distribution and its F-norm is one-to-one. If you know the norm, you know the distribution.
Algebraic Structure
The set of F-norms is endowed with a multiplication operation ($*$). If the F-norms of two independent random variables are multiplied, the result is the F-norm of their component-wise product. This turns the set of F-norms into a commutative monoid, with the sup-norm acting as the identity element. To accommodate distributions that can be negative, such as the Multivariate Normal distribution, the authors apply an exponential transformation to create a sub-class known as log F-norms.
3. Key Insights and Theoretical Results
Convergence and Wasserstein Metrics
The most significant theoretical contribution is the proof that pointwise convergence of a sequence of F-norms to a limit F-norm is equivalent to the convergence of the underlying distributions in the Wasserstein metric. This provides a functional analysis bridge to optimal transport theory, which is often computationally difficult to handle directly.
Geometric Representation
Every multivariate distribution can be visualized as a specific compact set—the unit sphere of its F-norm in the positive quadrant. This allows researchers to study probability through the lens of Hausdorff distance, measuring the distance between distributions as the distance between their characteristic shapes in $\mathbb{R}^{d+1}$.
Statistical Inference
The authors introduce the empirical F-norm, allowing for direct estimation from data samples via a simple averaging process. Unlike D-norms, which require sophisticated tail-estimation, F-norms are accessible for standard statistical inference. Theorem 3.3 proves that the empirical F-norm process converges to a Gaussian process, enabling the construction of confidence intervals and goodness-of-fit tests for multivariate data.
4. Conclusion and Future Directions
The F-norm framework provides a robust mathematical "skeleton" for multivariate distributions. While currently requiring marginal distributions to have finite first moments (integrability), it offers a powerful new way to formulate limiting results, such as the Central Limit Theorem, in terms of pointwise convergence of norm products.
Future research is expected to explore high-dimensional characterizations (beyond the two-dimensional case solved in Corollary 2.6) and the behavior of F-norms under dependent data structures, such as stationary mixing sequences common in time-series analysis.
Summary updated