Abstract
Permutation entropy is widely used to quantify complexity in nonlinear and physiological systems, yet its value can also vary systematically in purely linear Gaussian processes. Here we construct a linear-Gaussian reference for permutation entropy in ARMA models. Ordinal events are formulated as Gaussian region probabilities in increment space, showing that the ordinal distribution of a stationary Gaussian linear process is determined by the covariance matrix of the embedded vector. For embedding dimension $m=3$ and delay $\tau=1$, permutation entropy reduces to an exact function of the correlation between consecutive increments, yielding closed-form expressions for AR(1), MA(1), and ARMA(1,1) processes in terms of reciprocal-root parameters. Numerical simulations extend the framework to higher embedding dimensions and second-order root configurations, showing structured entropy landscapes induced by real, repeated, and complex-conjugate roots. As a proof of concept, we reanalyzed previously published intracranial pressure recordings from a porcine model of intracranial hypertension. With $m=5$, the empirical entropy of ICP increments decreased from 0.3462 to 0.2939 between low- and high-pressure windows, while the ARMA reference decreased from 0.3592 to 0.2978. Overall, 97.0\% of empirical entropies fell within the empirical 2.5--97.5\% ARMA surrogate interval. These results show that dominant empirical entropy changes can be compatible with local linear covariance structure, providing a baseline before assigning nonlinear or physiological complexity interpretations.