A unified framework for divergences, free energies, and Fokker-Planck equations
Anna L. F. Lucchi, Jean H. Y. Passos, Max Jauregui, Renio S. Mendes · 2025
Read A unified framework for divergences, free energies, and Fokker-Planck equations by Anna L. F. Lucchi, Jean H. Y. Passos, Max Jauregui, Renio S. Mendes (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2510-16690.txt.
Introduction
Anna L. F. Lucchi and three co-authors build a single framework that ties together divergences, generalized free energies, generalized Fokker-Planck equations, and the H-theorem. These tools describe how systems settle over time, including systems that defy standard statistical mechanics. The Fokker-Planck equation tracks how a probability distribution drifts and spreads.
Starting from a free energy that includes potential energy gives a direct link between that energy and the final steady state. Starting from a divergence as the free energy removes the potential from the equation, so an extra relation must be added to restore it. The authors apply both routes across several cases, showing old and new results from one viewpoint.
Catalog record
| CREATOR | Anna L. F. Lucchi, Jean H. Y. Passos, Max Jauregui, Renio S. Mendes |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 66 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2510-16690
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
A unified framework for divergences, free energies, and Fokker-Planck equations Anna L. F. Lucchi, Jean H. Y. Passos, Max Jauregui, Renio S. Mendes Abstract Many efforts have been made to explore systems that show significant deviations from predictions related to the standard statistical mechanics. The present work introduces a unified formalism that connects divergences, generalized free energies, generalized Fokker-Planck equations, and H H -theorem. This framework is applied here in a range of scenarios, illustrating both established and novel results. In many cases, the approach begins with a free energy functional that explicitly includes a potential energy term, leading to a direct relation between this energy and the stationary solution. Conversely, when a divergence is used as free energy, the associated Fokker-Planck-like equation lacks any explicit dependence on the potential energy, depending instead on the stationary solution. To restore a potential-based interpretation, an additional relation between the stationary solution and the potential energy must be imposed. This duality underlines the flexibility of the formalism and its capacity to adapt to systems where the potential energy is unknown or unnecessary. 1 Introduction In the last four decades, there has been a growing interest in investigating situations related to deviations from the formalism of Boltzmann-Gibbs-Shannon statistical mechanics. To deal with these cases, some proposals, such as the Tsallis [ 1 ] and Kaniadakis [ 2 ] entropies, have been considered. In the Tsallis context, the current investigations include dynamics of chaotic maps [ 3 ] , tumor edge detection [ 4 ] , Raman scattering [ 5 ] , high-energy physics [ 6 ] , and chemical reactions [ 7 ] , among a myriad of others [ 8 ] . Concerning the Kaniadakis approach, recent advances are related, for instance, to dark energy [ 9 ] , electron-acoustic waves [ 10 ] , nuclear reactors [ 11 ] , and relativity [ 12 ] . A comprehensive list of investigations involving the Kaniadakis entropy can be found in [ 13 ] . Moreover, even a formulation of statistical mechanics based on the Rényi entropy has also been investigated [ 14 ] . In the Tsallis and Kaniadakis contexts, time-dependent situations involving (nonlinear) Fokker-Planck equations were also investigated [ 15 , 16 , 17 ] . Their stationary solutions correspond to the canonical distributions and there are H -theorems related to them. A few further but representative developments based on (nonlinear) Fokker-Planck equations and their H -theorems begin with [ 18 , 19 , 20 , 21 ] , while extensions of these studies include inhomogeneous cases [ 22 , 23 ] , fractional derivatives [ 24 , 25 ] , coupled nonlinear equations [ 26 , 27 ] , and measures of complexity [ 28 ] . In addition, several applications and methodological approaches to nonlinear Fokker-Planck equations can be found in [ 29 ] . From a broader perspective and a formal point of view, we highlight that some general aspects are independent of the particular thermostatistics considered. For instance, this is observed in the universality of the Legendre transform structure, regardless of the entropy employed [ 30 , 31 ] , and in the classification of entropies [ 32 ] . Commonly, entropies are associated with divergences which quantify how two probability distributions differ from each other [ 33 ] . Terms such as relative entropy, information divergence, information gain, and even distance are also used to refer to divergences [ 34 ] . A divergence between two probability density functions, \rho and \rho_{0} , is here denoted by \mathcal{D}(\rho,\rho_{0}) and satisfies the inequality \mathcal{D}(\rho,\rho_{0})\geq 0 with \mathcal{D}(\rho,\rho_{0})=0 if and only if \rho=\rho_{0} [ 34 ] . The Shannon entropy is closely related to the well-known Kullback-Leibler divergence [ 35 ] . Applications of this divergence are vast, including image retrieval and classification [ 36 , 37 , 38 , 39 ] , text characterization [ 40 , 41 ] , seismic information extraction [ 42 ] , decision-making systems [ 43 ] , optimization problems [ 44 ] , and machine learning [ 45 ] , among many others. Analogously, there are the Rényi entropy and a related divergence [ 46 ] , which have a wide range of applications [ 47 , 48 , 49 , 50 , 51 ] . Similarly, the Tsallis entropy [ 1 ] and an associated divergence [ 52 ] have been employed in quantum physics [ 53 , 54 ] , variational inference [ 55 ] , radioactive source detection [ 56 ] , and Markov decision processes [ 57 ] . Likewise, the Kaniadakis entropy [ 2 ] and a corresponding divergence [ 58 ] have also been studied. All these entropies and divergences recover the Shannon entropy and the Kullback-Leibler divergence as limiting cases. More general divergences are the f -divergence [ 46 ] (also called Csiszár [ 59 ] ), the Bregman divergence [ 60 ] and the Burbea-Rao divergence [ 61 ] . As should be clear from the previous discussion, depending on the problem and the data, some aspects of an analysis may be more relevant than others; therefore, a triad composed of entropy, divergence, and the Fokker-Planck equation may be more appropriate than another one. By presenting a unified formalism, this work connects divergence, (generalized) free energy, the (generalized) Fokker-Planck equation, and the H -theorem, and demonstrates its versatility through both new and known applications. Our approach is consistent with previous works that connect the difference between two free energies with divergence in Shannon, Tsallis and Kaniadakis contexts [ 62 , 63 , 64 , 65 , 17 ] . The present work is organized as follows. In Sec. 2 , several entropies and their corresponding divergences are reported. We begin with the Shannon and Kullback-Leibler pair. The Tsallis and Kaniadakis scenarios are addressed through the introduction of the q -logarithm and \kappa -logarithm functions, respectively. Likewise, the Rényi case is considered. These scenarios are also extended to incorporate generalized logarithms. In addition, the Bregman and Burbea-Rao divergences are discussed. In Sec. 3 , we introduce a unified approach that identifies each divergence with the difference between two generalized free energies. This integrated perspective is achieved via a Fokker-Planck-like equation and its H -theorem. Given a free energy, a Fokker-Planck equation is obtained and its stationary solution is connected with the minimum of this free energy. In Sec. 4 , for all divergences presented in Sec. 2 , the corresponding free energies are exhibited as well as their Fokker-Planck equations and stationary solutions. We show that some divergences (free energies) are connected with drift forces and, therefore, with their potential energies. When a stationary solution \rho_{0} is not associated with a potential energy V , we consider the possibility of Fokker-Planck-like equations that explicitly depend on \rho_{0} , without directly involving V . The last section presents our concluding remarks. 2 Entropies and divergences The Shannon entropy is a measure of the amount of information contained in a discrete probability distribution. This measure attains its maximum value, corresponding to the smallest quantity of information, when applied to the uniform distribution. For an absolutely continuous distribution with density \rho , the Shannon entropy is defined by