English

A Frobenius-Optimal Projection for Enforcing Linear Conservation in Learned Dynamical Models

Dynamical Systems 2025-12-29 v1 Machine Learning Numerical Analysis Numerical Analysis

Abstract

We consider the problem of restoring linear conservation laws in data-driven linear dynamical models. Given a learned operator A^\widehat{A} and a full-rank constraint matrix CC encoding one or more invariants, we show that the matrix closest to A^\widehat{A} in the Frobenius norm and satisfying CA=0C^\top A = 0 is the orthogonal projection A=A^C(CC)1CA^A^\star = \widehat{A} - C(C^\top C)^{-1}C^\top \widehat{A}. This correction is uniquely defined, low rank and fully determined by the violation CA^C^\top \widehat{A}. In the single-invariant case it reduces to a rank-one update. We prove that AA^\star enforces exact conservation while minimally perturbing the dynamics, and we verify these properties numerically on a Markov-type example. The projection provides an elementary and general mechanism for embedding exact invariants into any learned linear model.

Cite

@article{arxiv.2512.22084,
  title  = {A Frobenius-Optimal Projection for Enforcing Linear Conservation in Learned Dynamical Models},
  author = {John M. Mango and Ronald Katende},
  journal= {arXiv preprint arXiv:2512.22084},
  year   = {2025}
}
R2 v1 2026-07-01T08:41:40.799Z