Some informational inequalities involving generalized trigonometric functions and a new class of generalized moments
David Puertas-Centeno, Steeve Zozor
TL;DR
This work develops a unified framework for analyzing probability densities through Generalized Trigonometric Densities (GTDs), built from generalized trigonometric functions, and a novel two-parameter family of generalized cumulative moments that regulate tail weight. The authors show that GTDs minimize Rényi entropy power for a fixed generalized Fisher information and saturate a class of sharp information inequalities; they also establish a three-parameter Stam-type bound with GTDs as extremizers. A key contribution is the introduction of cumulative moments, which extend standard moments to heavy-tailed densities and enable density reconstruction via a differential-escort transform, linking to MaxEnt with cumulative constraints. The results provide a rigorous way to characterize heavy-tailed and stretched Gaussian families and offer a pathway to multivariate generalizations, with potential implications for modeling and inference under tail uncertainty.
Abstract
In this work, we define a family of probability densities involving the generalized trigonometric functions defined by Drábek and Manásevich [1], which we name Generalized Trigonometric Densities. We show their relationship with the generalized stretched Gaussians and other types of laws such as logistic, hyperbolic secant, and raised cosine probability densities. We prove that, for a fixed generalized Fisher information, this family of densities is of minimal Rényi entropy. Moreover, we introduce generalized moments via the mean of the power of a deformed cumulative distribution. The latter is defined as a cumulative of the power of the probability density function, this second parameter tuning the tail weight of the deformed cumulative distribution. These generalized moments coincide with the usual moments of a deformed probability distribution with a regularized tail. We show that, for any bounded probability density, there exists a critical value for this second parameter below which the whole subfamily of generalized moments is finite for any positive value of the first parameter (power of the moment). In addition, we show that such generalized moments satisfy remarkable properties like order relation w.r.t. the first parameter, or adequate scaling behavior. Then we highlight that, if we constrain such a generalized moment, both the Rényi entropy and generalized Fisher information achieve respectively their maximum and minimum for the generalized trigonometric densities. Finally, we emphasis that GTDs and cumulative moments can be used to formally characterize heavy-tailed distributions, including the whole family of stretched Gaussian densities.