Certified Passive Compression of Mittag–Leffler Memory: Whole-Line Error, Spectral Stiffness, and Exact Discrete Energy
Authors/Creators
Description
Abstract. Uniformly accurate positive finite-state compression of Mittag–Leffler memory is possible on the entire half-line while preserving passivity, but every whole-line positive approximation must pay a quantifiable spectral-stiffness price. We prove both statements. For every normalized completely monotone kernel, a mass-preserving logarithmic-centroid rule gives the second-order certificate sup t≥0 |K(t) − Km(t)| ≤ Etail + κ2 2 X j wj Varj (log r), κ2 = 0.3090047859 . . . . For the Mittag–Leffler relaxation used by Atangana–Baleanu memory, a closed spectral distribution and quantile yield an explicit positive selector with m = Oα,η(ε −1/2 log(1/ε)) factors, independent of the scale parameter. Conversely, a general cusp–tail theorem shows that wholeline error ε forces a rate at least of order ε −(1−α)/α and a rate at most of order ε 1/α log(1/ε); accuracy therefore necessarily widens the relaxation spectrum. Exact exponential factor updates preserve a discrete energy identity, while an end-to-end estimate separates kernel compression from input projection without a stiffness-dependent time-step term. Reproducible computations verify the certificates, reproduce the energy balance to roundoff, and exhibit linear-state versus quadratic-history cost separation in one declared implementation. The positive spectral representation and diffusive-state framework are inherited. The contribution is the combined whole-line log-centroid certificate, closed Atangana–Baleanu quantile selector, positive-sum stiffness barrier, and exact discrete passivity/error-transfer theorem package. The results claim neither factor-count optimality nor universal superiority over rational, Gaussian-quadrature, FFT, or fast convolution-quadrature methods.