Positive-Allocation Companion Predictors for Nonlinear Dynamics and Their Finite-Difference Diagnostics (Q7284249)

We develop a positive-allocation companion predictor for nonlinear dynamics which constructs a low-dimensional linear model from an input snapshot by representing it as a nonnegative combination of earlier training snapshots. We establish that the spectrum of the resulting companion matrix is contained in the closed unit disk and includes 1, which we leverage to derive finite-difference diagnostics for model behavior without needing diagonalization.