{"id":"arxiv/2605.08362","url":"https://arxiv.org/abs/2605.08362","tags":["math.NA"],"score":0,"title":"Kernel-based linear system identification using augmented Krylov subspaces","domain":"math.NA","source":"arxiv","authors":["Fabio Matti","Martin Skovgaard Andersen","Tianshi Chen","Daniel Kressner"],"reviews":0,"subject":"math.NA","summary":"We propose a novel Krylov subspace method for estimating the finite impulse response (FIR) of a one-dimensional linear time-invariant systems. The method approximates the system's FIR using a kernel-based formulation combined with hyperparameter selection based on maximum likelihood estimation (MLE), which requires repeated evaluation of two terms: The data fit $\\boldsymbol{y}^{\\top} (λ\\boldsymbol{I} + \\boldsymbol{A})^{-1} \\boldsymbol{y}$ and the model complexity $\\log(\\det (λ\\boldsymbol{I} + \\boldsymbol{A}))$, where $\\boldsymbol{A}$ is a certain positive semidefinite matrix that admits fast…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-05-08","reactions":{}}