Requirement-Induced Predictive Geometry for Finite-Resource Prediction in Dynamical Systems
Abstract
In nonlinear prediction, two state representations with the same local uncertainty volume can have radically different predictive value. We consider differentiable finite-time dynamics together with a quadratic terminal requirement that specifies which terminal state differences matter. Pulling this requirement back through the tangent map defines a requirement-induced predictive geometry on the present tangent space. This geometry is fixed by the dynamics and the requirement; under the fixed local uncertainty-volume constraint used here, the capacity parameter fixes only the scale of the representation that realizes it. We show that the unique optimal covariance has the inverse predictive shape and, equivalently, that optimality is characterized by isotropization of propagated uncertainty in the terminal requirement geometry. The gain over a baseline-isotropic representation is the arithmetic-to-geometric mean ratio of the predictive spectrum, making it a coordinate-invariant measure of predictive anisotropy. In Euclidean coordinates this yields a general finite-time spectral law, and for two-dimensional area-preserving dynamics an exact identity between predictive gain and the condition number of the optimal geometry. Stronger directional stretching therefore increases potential predictive gain only together with the anisotropy required to realize it; bounding that anisotropy imposes a finite gain ceiling. Standard-map and double-pendulum calculations demonstrate these relations in nonlinear Hamiltonian systems. The framework provides a basis for adaptive sensing, state estimation, and model-predictive computation in which representation geometry is matched to predicted task consequences.
Related
- Dynamical resource theory of time-reversal symmetry breaking
- Initial-State Precision as a Predictive Resource: From Tori to Strange Nonchaotic Attractors and Chaos
- Short-term solar eclipse predictions using Euclidean geometry and dynamical systems
- Basin Geometry and Reliable Recall of Dynamical Memories in Reservoir Computing
- From Predictive Dynamics to Spectral Geometry
BibTeX
@misc{kim2026requirementinduced,
title = {{Requirement-Induced Predictive Geometry for Finite-Resource Prediction in Dynamical Systems}},
author = {Song-Ju Kim},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2609.16351},
eprint = {2609.16351},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2609.16351},
}
RIS
TY - UNPB
TI - Requirement-Induced Predictive Geometry for Finite-Resource Prediction in Dynamical Systems
AU - Song-Ju Kim
PY - 2026
DA - 2026/09/14
PB - arXiv
DO - 10.48550/arxiv.2609.16351
UR - https://arxiv.org/abs/2609.16351
AB - In nonlinear prediction, two state representations with the same local uncertainty volume can have radically different predictive value. We consider differentiable finite-time dynamics together with a quadratic terminal requirement that specifies which terminal state differences matter. Pulling this requirement back through the tangent map defines a requirement-induced predictive geometry on the present tangent space. This geometry is fixed by the dynamics and the requirement; under the fixed local uncertainty-volume constraint used here, the capacity parameter fixes only the scale of the representation that realizes it. We show that the unique optimal covariance has the inverse predictive shape and, equivalently, that optimality is characterized by isotropization of propagated uncertainty in the terminal requirement geometry. The gain over a baseline-isotropic representation is the arithmetic-to-geometric mean ratio of the predictive spectrum, making it a coordinate-invariant measure of predictive anisotropy. In Euclidean coordinates this yields a general finite-time spectral law, and for two-dimensional area-preserving dynamics an exact identity between predictive gain and the condition number of the optimal geometry. Stronger directional stretching therefore increases potential predictive gain only together with the anisotropy required to realize it; bounding that anisotropy imposes a finite gain ceiling. Standard-map and double-pendulum calculations demonstrate these relations in nonlinear Hamiltonian systems. The framework provides a basis for adaptive sensing, state estimation, and model-predictive computation in which representation geometry is matched to predicted task consequences.
ER -