Abstract
In this paper we consider some new algorithms for computing the Kalman-Bucy gain in stationary systems requiring a number of equations of ordern (rather thann 2) whenever the ordern of the system is much larger than the dimension of the output. These equations were independently obtained by Kailath and Lindquist in continuous and discrete time respectively. We briefly discuss the relations with some recent related results due to Casti, Kalaba & Murthy and Rissanen. Some of the reasons for these reductions are inherent in the properties of general stationary processes, and therefore a considerable portion of the paper is devoted to exploring the connections with some previous work by Levinson, Whittle and Wiggins & Robinson, and also with the Szegö theory of polynomials orthogonal on the unit circle and some continuous analogs due to Krein. We demonstrate that the Bellman-Krein formula is the fundamental relation in continuous time, the trick being to introduce a “reversed time” counterpart of the weighting function (Fredholm resolvent). This is suggested by the “forward and backward innovation” approach in a previous paper by the author, the essential relations of which we reformulate in terms of Fredholm integral equations (in continuous time) and Toeplitz equations (in discrete time). Therefore we also derive the discrete-time Bellman-Krein formulas of which there are actually two—one corresponding to the one-step predictor and one to the pure filter. In this way we shall be able to pin down the reasons for the striking discrepancies between the continuous-time and the discrete-time cases. Finally we clarify the relations between Levinson's equations and Chandrasekhar'sX- andY-functions.
Similar content being viewed by others
References
N. I. Akiezer,The Classical Moment Problem, Hafner Publishing Company, New York, 1965.
R. Bellman, Functional equations in the theory of dynamic programming—VII: A partial differential equation for the Fredholm resolvent,Proc. Amer. Math. Soc.,8, 1957, pp. 435–440.
R. S. Bucy andP. D. Joseph,Filtering for Stochastic Processes with Applications to Guidance, Interscience Publishers, New York, 1968.
J. Casti, R. Kalaba andV. K. Murthy, A new initial value method for on-line filtering and estimation,IEEE Trans. on Information Theory, July 1972, pp. 515–517.
J. Casti andR. Kalaba,Imbedding Methods in Applied Mathematics, Addison-Wesley, Reading, Mass., 1973.
J. Casti andE. Tse, Optimal linear filtering theory and radiative transfer: Comparisons and interconnections,J. Math. Anal. and Appl., vol. 40, October 1972, pp. 45–54.
S. Chandrasekhar,Radiative Transfer, Oxford University Press, 1950.
L. Ya. Geronimus,Orthogonal Polynomials, Consultant Bureau, New York, 1961.
T. Kailath, Some Chandrasekhar-type algorithms for quadratic regulators,Proc. IEEE Decision and Control Conference, Dec. 1972.
T. Kailath, Some new algorithms for recursive estimation in constant linear systems,IEEE Trans. Information Theory, Vol. IT-19, November 1973, pp. 750–760.
T. Kailath, M. Morf andS. Sidhu, Some new algorithms for recursive estimation in constant discrete-time linear systems,Proc. Seventh Princeton Symp. on Information and System Science, March 1973.
M. Krein, On a new method for solving linear integral equations of the first and second kinds,Dokl. Akad. NAUK. SSSR,100, 1955, pp. 413–416.
M. Krein, Continuous analogs of propositions concerning polynomials orthogonal on the unit circle,Dokl. Akad. NAUK. SSSR,105, 1955, pp. 637–640.
N. Levinson, The Wiener RMS (root mean square) error in filter design and prediction, Appendix B inN. Wiener,Extrapolation, Interpolation and Smoothing of Stationary Time Series, M.I.T. Press, Cambridge, Mass., 1942.
A. Lindquist, A new algorithm for optimal filtering of discrete-time stationary processes,SIAM J. Control,4, 1974.
A. Lindquist, Optimal filtering of continuous-time stationary processes by means of the backward innovation process,SIAM J. Control,4, 1974.
A. Lindquist, Optimal control of linear stochastic systems with applications to time lag systems,Information Sci., Vol. 5 (Jan. 1973), pp. 81–126.
J. Rissanen, Algorithms for triangular decomposition of block Hankel and Toeplitz matrices with applications to factoring positive matrix polynomials,Mathematics of Computation,27 (1973), pp. 147–154.
J. Rissanen, A fast algorithm for optimum linear prediction,IEEE Transactions Automatic Control, vol. AC-18, p. 555, Oct. 1973.
R. A. Wiggins andE. A. Robinson, Recursive solutions to the multichannel filtering problem,J. Geophys. Research,70, pp. 1885–1891 (1965).
P. Whittle, On the fitting of multivariate autoregressions and the approximate canonical factorization of a spectral density matrix,Biometrica, vol. 50, 1963, pp. 129–134.
T. Kailath, Application of a resolvent identity to a linear smoothing problem,SIAM J. Control, vol. 7, February 1969, pp. 68–74.
Author information
Authors and Affiliations
Additional information
Communicated by W. Fleming
This research was carried out while the author held a visiting position at the Division of Applied Mathematics, Brown University, Providence, Rhode Island, and was supported by the National Science Foundation under grant NSF-46614.
Rights and permissions
About this article
Cite this article
Lindquist, A. On Fredholm integral equations, Toeplitz equations and Kalman-Bucy filtering. Appl Math Optim 1, 355–373 (1975). https://doi.org/10.1007/BF01447958
Received:
Issue Date:
DOI: https://doi.org/10.1007/BF01447958