Observational Information Loss Under Non-Invertible Mappings
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
This note establishes a minimal, model-independent constraint linking non-invertibility of observation mappings to strict information loss. Let X be a random variable on a probability space and P: \mathcal{X} \to \mathcal{Y} a measurable mapping. We prove that P is injective almost everywhere if and only if the conditional entropy satisfies H(X \mid P(X)) = 0. Consequently, any failure of injectivity on a set of positive measure implies strictly positive conditional entropy and therefore strict reduction of mutual information, I(X; P(X)) < H(X). The result is purely measure-theoretic and does not depend on linearity, stationarity, noise models, or specific dynamical assumptions. It applies uniformly to discrete, continuous, and mixed distributions under standard regularity conditions. As a direct consequence, any dimension-reducing observation of a dynamical system induces a fundamental observability constraint: projected observations necessarily contain strictly less information about system parameters than the full state, and a positive-dimensional subspace of the state space remains unobservable regardless of data quantity, observation duration, or estimator choice. This establishes that information loss under non-invertible observation is not a statistical limitation but a structural property of the mapping itself. The result provides a unified theoretical basis for understanding irreducible limits of inference in complex systems and connects information-theoretic bounds with classical observability theory via rank deficiency of the observation operator. information theory, mutual information, conditional entropy, data processing inequality, non-invertible mappings, dimensionality reduction, observability, dynamical systems, Fisher information, entropy, complex systems, nonlinear systems, measure theory, rank deficiency, inference limits
Fields of Study & Subjects
Mathematical Dynamics and Fractals
Control Systems and Identification
Complex Systems and Dynamics
Publication details
- Journal / source
- Zenodo (CERN European Organization for Nuclear Research)
- Publication date
- 30 March 2026
- Language
- EN
- Document type
- Preprint