bibkey: “dambre2012capacity” authors: “Joni Dambre; David Verstraeten; Benjamin Schrauwen; Serge Massar” year: 2012 title: “Information Processing Capacity of Dynamical Systems” doi: “10.1038/srep00514” claim: “Information processing capacity is defined through optimal linear readouts of orthogonal input-history targets; the capacity sum is bounded by the rank of the feature space, with saturation under the paper’s stated dynamical and statistical hypotheses.” strata_touched: [] license: “citation-only” triage: “anchor”

Information Processing Capacity of Dynamical Systems

Verified locator

Scientific Reports 2, 514 (2012), DOI 10.1038/srep00514. Theorem 4 gives the capacity upper bound for orthogonal target functions; Theorem 7 gives saturation under fading memory, full-rank readout correlations, finite-moment, and infinite-data/initialization limits. Supplementary §§1.1–1.6 derive the Hilbert-space projection form and discuss rank-deficient features; Supplementary §2 distinguishes centered covariance from uncentered second-moment conventions.

The contextual-spacetime ML volume reuses only the projection/Bessel argument and keeps the paper’s dynamical equality hypotheses separate.