Abstract
This research generalizes the classical Fisher information metric—a cornerstone of information geometry—into $L^p$-metrics (Finsler metrics) on infinite-dimensional spaces of differential forms and diffeomorphism groups. By leveraging this new framework, we derive novel results across several mathematical domains, including a real-line counterpart to Teichmüller theory, the geometry of symplectic forms, and hydrodynamics on Orlicz spaces. Key contributions include the derivation of explicit geodesic equations, the generalization of Gelfand-Fuchs cocycles to higher dimensions, and the resolution of open questions regarding $\alpha$-geometries in Orlicz spaces.
Introduction
In classical information geometry, a family of probability distributions is treated as a Riemannian manifold equipped with the Fisher information metric. This metric, fundamentally based on an $L^2$ inner product, has traditionally been applied to finite-dimensional statistical models. Recent advancements have extended this to the infinite-dimensional space of all probability densities on a compact manifold, establishing deep links between information geometry, fluid dynamics, and diffeomorphism groups.
However, the classical $L^2$ framework faces three primary limitations: it does not naturally extend to $L^p$ norms (which would necessitate a transition from Riemannian to Finsler geometry), it struggles with non-compact spaces like the real line $\mathbb{R}$ where standard integrals often diverge, and it is primarily focused on top-dimensional volume forms rather than lower-dimensional forms like symplectic structures. This paper addresses these gaps by constructing a rigorous mathematical framework for $L^p$-geometry on these complex spaces.
Methodological Framework
The study employs techniques from infinite-dimensional differential geometry, specifically utilizing Fréchet manifolds and Inverse Limit Hilbert (ILH) structures. To navigate the complexities of these spaces, several key methods are utilized:
- Moser’s Trick: This foundational technique is used to relate the space of probability densities to the quotient space of diffeomorphism groups, specifically $\operatorname{Diff}(M)/\operatorname{Diff}_\mu(M)$.
- Finsler Geometry: The author defines $L^p$-Fisher-Rao metrics as Finsler metrics, calculating their geodesic equations through the analysis of energy functionals.
- Isometric Embeddings: To simplify the highly non-linear geodesic equations, the author constructs mappings that embed these curved spaces of densities isometrically into flat function spaces or $L^p$-spheres.
- Luxemburg Norms: The framework is extended to Orlicz spaces, allowing for the study of generalized hydrodynamics equations beyond the standard $L^p$ constraints.
Key Insights and Geometric Applications
Geodesics via Flat Space Isometry
A significant challenge in $L^p$-geometry is that geodesic equations for the $L^p$-Fisher-Rao metric on the space of positive densities are notoriously difficult to solve. The author demonstrates that by mapping this space isometrically to a flat function space, the geodesics are transformed into straight lines. This allows for the derivation of explicit, closed-form solutions that were previously inaccessible.
Teichmüller Theory on the Real Line
By investigating diffeomorphism groups on the real line with specific decay conditions (denoted as $\operatorname{Diff}_{-\infty}(\mathbb{R})$), the research develops a "real counterpart" to classical Teichmüller theory, which is traditionally situated in the realm of complex Riemann surfaces. This includes the definition of an $L^p$-approximation of the Schwarzian derivative and the establishment of a real-line version of the Bers embedding. Notably, the author proves that the mapping $\Phi_\infty$ is an isometric diffeomorphism, implying that the group $\operatorname{Diff}_{-\infty}(\mathbb{R})$ is contractible.
Geometry of Symplectic Forms
The space of all symplectic forms with constant volume on a manifold forms an infinite-dimensional bundle. The research provides a Čencov-type theorem for this space, proving that the $L^p$-metric uniquely measures the "horizontal" tangent bundle in a manner invariant under all diffeomorphisms. Furthermore, the classical Gelfand-Fuchs 2-cocycle—essential to string theory—is generalized to an $(n+1)$-cocycle on the space of probability densities for any $n$-dimensional closed manifold.
Solving Gibilisco's Problem
The paper successfully defines a $C^1$-Finsler structure on Orlicz spaces, providing an answer to an open question posed by P. Gibilisco regarding $\alpha$-geometries. This generalization allows for the derivation of the corresponding Euler-Arnold equation for the diffeomorphism group of the circle, extending the reach of information geometry into generalized functional analysis.
Limitations and Future Directions
While the research provides explicit geodesics for positive densities, finding closed-form solutions for probability densities remains an open challenge. The isometry in the latter case maps densities to an $L^p$-sphere rather than a flat space, complicating the geodesic equations. Additionally, the Finsler metric defined on Orlicz spaces currently lacks higher-order smoothness ($C^1$ only), rendering some hydrodynamic calculations formal in nature.
Future work will focus on determining the exact image of the real Bers embedding—which requires solving a complex real Riccati equation—and investigating the existence of nontrivial, diffeomorphism-invariant metrics on the vertical fibers of the symplectic form space.
Summary updated