Model Error Resonance: The Geometric Nature of Error Dynamics

This paper introduces a geometric theory of model error, treating true and model dynamics as geodesic flows generated by distinct affine connections on a smooth manifold. When these connections differ, the resulting trajectory discrepancy–termed the Latent Error Dynamic Response (LEDR)–acquires an intrinsic dynamical structure governed by curvature. We show that the LEDR satisfies a Jacobi-type equation, where curvature mismatch acts as an explicit forcing term. In the important case of a flat model connection, the LEDR reduces to a classical Jacobi field on the true manifold, causing Model Error Resonance (MER) to emerge under positive sectional curvature. The theory is extended to a discrete-time analogue, establishing that this geometric structure and its resonant behavior persist in sampled systems. A closed-form analysis of a sphere–plane example demonstrates that curvature can be inferred directly from the LEDR evolution. This framework provides a unified geometric interpretation of structured error dynamics and offers foundational tools for curvature-informed model validation.

💡 Research Summary

The paper proposes a geometric framework for understanding model error by treating both the true physical system and its computational model as geodesic flows generated by distinct affine connections on a smooth manifold. The discrepancy between the two trajectories, called the Latent Error Dynamic Response (LEDR), is not a static residual but a vector field that evolves according to a Jacobi‑type differential equation whose coefficients are determined by the curvature of the true connection and by the mismatch between the two connections.

In continuous time, the authors derive the LEDR equation
 D²ξ/dt² + Rᵗ(Tᵐ, ξ) Tᵐ = F_{ΔΓ}(Tᵐ, ξ) + higher‑order terms,
where D denotes covariant differentiation with respect to the true connection ∇ᵗ, Rᵗ is its Riemann curvature tensor, Tᵐ is the model velocity, and ΔΓ = Γᵗ – Γᵐ captures the connection difference. When the model connection is flat (Γᵐ ≡ 0) the forcing term vanishes and the equation reduces to the classical Jacobi equation D²ξ/dt² + Rᵗ(Tᵐ, ξ) Tᵐ = 0. If the sectional curvature K(t) of the plane spanned by Tᵐ and ξ is positive and constant, the scalar form ξ̈ + K ξ = 0 yields harmonic oscillations with natural frequency ω = √K. The authors name this phenomenon Model Error Resonance (MER), interpreting it as a curvature‑induced restoring force that continually pulls the model trajectory toward the true geodesic.

The theory is discretized by approximating covariant derivatives with central differences, leading to a second‑order recurrence
 ξ_{k+1} = 2 ξ_k – ξ_{k−1} – h² Rᵗ(Tᵐ_k, ξ_k) Tᵐ_k + h² F_{ΔΓ,k} + O(h³).
For a flat model connection the recurrence simplifies to ξ_{k+1} = 2 ξ_k – ξ_{k−1} – h² K ξ_k. The characteristic equation shows that oscillatory (MER) behavior occurs when 0 < λ = h²K < 4, i.e., for sufficiently small sampling steps any positive curvature produces discrete MER. Conversely, negative curvature leads to exponential divergence, and zero curvature yields linear drift.

A key contribution is the derivation of lower bounds on the LEDR norm when the true and model curvatures differ. The authors prove that if |Kᵗ(t) – Kᵐ(t)| ≥ κ₀ > 0 over an interval, then ∥ξ(t)∥ cannot decay to zero and satisfies a bound of the form ∥ξ(t)∥ ≥ C∫|Kᵗ–Kᵐ| dt – ε(t). This establishes that curvature mismatch imposes a fundamental limit on error reduction, independent of controller or estimator design.

The framework is illustrated with a closed‑form sphere‑plane example: the true system evolves on a sphere of radius R (sectional curvature K = 1/R²) while the model assumes a flat Euclidean plane. The LEDR reduces to a harmonic oscillator with ω = 1/R, and the authors show that curvature can be recovered directly from measured LEDR data using the discrete estimator K ≈ (2 ξ_k – ξ_{k+1} – ξ_{k−1})/(h² ξ_k). This demonstrates that curvature is an observable quantity embedded in error dynamics, even without explicit knowledge of the true connection.

Overall, the paper provides a unified geometric interpretation of structured model error, linking curvature, connection mismatch, and error dynamics. It offers theoretical tools for curvature‑informed model validation, suggests practical curvature estimation from residuals, and highlights the inevitability of resonant error behavior when the underlying manifold has positive curvature. Potential applications span system identification, model‑based control, robotics navigation, and the reliability analysis of data‑driven predictors.