A general fixed-interval smoother with exact initial conditions

In this work we derive a relationship between tbe exact fixed-interval smoothed moments and those obtained from an arbitrarily initialized smoother. Combining this result witbh a conventional smoother we obtain a new algoritbm with exact initial conditions, that can be applied to stationary, nonstationary or partially nonstationary systems, with deterministic and/or stochastic inputs. Besides an easy analytical derivation, other advantages of this smoother are its computational efficiency and numerical stability.