A General Theory of Prediction: Monotone Distinguishability, Recognition Cascades, and the Transport Law of Learning

Summarise with VIRAT
Authors: Duggan Matthews
Cited by: 0

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

We develop a mathematical framework in which predictive learning, thermodynamic irreversibility, and spacetime geometry arise from a common structural principle: monotone distinguishability under admissible transformations. Divergence contraction induces Fisher information geometry, and steepest descent yields natural gradient learning dynamics. Predictive systems generically undergo recognition cascades — sequences of codimension-1 structural bifurcations governed by spectral degeneracy. Near each elementary recognition event the reduced potential admits a universal quartic normal form, implying a quartic recognition law. The central result is the transport law of learning: the total continuous Fisher dissipation plus total discrete recognition expenditure equals the initial predictive mismatch. This is a conservation law for intelligence — the mismatch budget is exactly conserved and spent through two channels, smooth iteration and sudden recognition, with nothing lost and nothing created. The discrete recognition expenditure defines a gauge-invariant intrinsic scalar along a predictive cascade. Extending the framework to nested operator algebras associated with causal diamonds, we demonstrate that the unique local renormalisation of nested distinguishability is proportional to boundary area. Under a predictive equilibrium condition this yields generalised entropy stationarity and reproduces the Einstein equation in the small-diamond limit — establishing that the Einstein field equation, diffeomorphism invariance, the equivalence principle, and the metric structure are derived from monotone distinguishability, not assumed. Newton's constant is determined by self-consistency of the boundary entanglement structure. The semiclassical regime is not a limitation but a prediction — the framework derives both the equation and its domain of validity. The structural chain runs from one inequality to the geometry of spacetime: monotone distinguishability → Fisher geometry → natural gradient dynamics → recognition cascades → quartic universality → transport law → nested entropy → boundary renormalisation → Einstein equation. One principle generates the entire chain. The results suggest that physical laws may be understood as consistency conditions governing distinguishability across scales.

Fields of Study & Subjects

Relativity and Gravitational Theory Black Holes and Theoretical Physics Complex Systems and Dynamics

Publication details

Journal / source
OSF Preprints (OSF Preprints)
Publication date
20 March 2026
Language
EN
Document type
Preprint