AR(1) covariance appended-row innovation is supported only at the new state
ProvedAppliedSDE.SarkkaSolin.ar1_covariance_row_innovationapplied-sdear1covarianceinnovationkalman-filter
For the stationary scalar AR(1) covariance K_ij=V a^|i−j|, subtracting a times row n from row n+1 gives zero at all earlier observation coordinates i<=n and gives V(1−a²) at the newly appended diagonal i=n+1. This exact sparse innovation identity captures the Markov structure needed for Schur-complement induction in the open general Gaussian-process versus sequential Kalman theorem.
Preamble
import Mathlib
Formal statement
theorem AppliedSDE.SarkkaSolin.ar1_covariance_row_innovation
(n : ℕ) (a V : ℝ) (i : Fin (n + 2)) :
V * a ^ Nat.dist (n + 1) i.val -
a * (V * a ^ Nat.dist n i.val) =
if i.val = n + 1 then V * (1 - a ^ 2) else 0 := by
sorrySource
Särkkä and Solin, Applied Stochastic Differential Equations (2019), Chapters 10 and 12, AR(1) Gaussian state-space kernel.