Skip to main content
Log in

A remark on robustness of linear best estimates

  • Veröffentlichungen
  • Published:
Metrika Aims and scope Submit manuscript

Summary

Let ξ(t) be a random process with correlation functionB 0, defining a norm ‖ ‖0, and letη 0 (ξ) be the best linear estimate of some unknown value ξ of the process ξ(t). SupposeB 0 is not known and a “pseudobest” estimateη 1 (ξ) is obtained under some hypothetic correlationB 1 giving the norm ‖ ‖1. Then, denoting by

$$\begin{gathered} \sigma _0 = \parallel \xi - \eta _0 (\xi )\parallel _0 , \hfill \\ \sigma (B_1 ) = \parallel \xi - \eta _1 (\xi )\parallel _0 , \hfill \\ \sigma _1 (B_1 ) = \parallel \xi - \eta _1 (\xi )\parallel _1 , \hfill \\ \end{gathered} $$

the corresponding least square errors, the following robustness property holds true under a rather general assumption:

$$\mathop {\lim }\limits_{h \to 0} \mathop { \sup }\limits_{|B_1 - B_0 | \leqslant h} \sigma _1 (B_1 ) = \mathop {\lim }\limits_{h \to 0} \mathop { \sup }\limits_{|B_1 - B_0 | \leqslant h} \sigma (B_1 ) = \sigma _0 ,$$

whereh→0 means pointwise convergence.

Zusammenfassung

ξ(t) sei ein stochastischer Prozess mit der KorrelationsfunktionB 0, die eine Norm ‖ ‖0 definiert, und es seiη 0 (ξ) der beste lineare Schätzer eines unbekannten Wertes ξ des Prozesses ξ (t). Angenommen,B 0 ist nicht bekannt, und ein „pseudobester” Schätzerη 1 (ξ) wird bestimmt unter einer hypothetischen KorrelationB 1, die die Norm ‖ ‖1, definiert. Wenn dann

$$\begin{gathered} \sigma _0 = \parallel \xi - \eta _0 (\xi )\parallel _0 , \hfill \\ \sigma (B_1 ) = \parallel \xi - \eta _1 (\xi )\parallel _0 , \hfill \\ \sigma _1 (B_1 ) = \parallel \xi - \eta _1 (\xi )\parallel _1 \hfill \\ \end{gathered} $$

die entsprechenden kleinsten Fehler bezeichnen, so gilt unter einer recht allgemeinen Voraussetzung die folgende Robustheitsaussage:

$$\mathop {\lim }\limits_{h \to 0} \mathop { \sup }\limits_{|B_1 - B_0 | \leqslant h} \sigma _1 (B_1 ) = \mathop {\lim }\limits_{h \to 0} \mathop { \sup }\limits_{|B_1 - B_0 | \leqslant h} \sigma (B_1 ) = \sigma _0 ,$$

wobeih→0 die punktweise Konvergenz bedeutet.

This is a preview of subscription content, log in via an institution to check access.

Access this article

Price excludes VAT (USA)
Tax calculation will be finalised during checkout.

Instant access to the full article PDF.

Similar content being viewed by others

References

  • Grenander, U., andM. Rosenblatt: Statistical Analysis of Stationary Time Series, New York 1966.

  • Rozanov, Yu.A.: Stationary Random Processes, San Francisco 1967.

  • Rozanov, Yu.A.: On the stability of solutions of linear problems for stationary processes, Theory of Probability and its Applications, Vol.IX (3), 1964.

Download references

Author information

Authors and Affiliations

Authors

Rights and permissions

Reprints and permissions

About this article

Cite this article

Rehder, W. A remark on robustness of linear best estimates. Metrika 23, 1–6 (1976). https://doi.org/10.1007/BF01902844

Download citation

  • Received:

  • Revised:

  • Issue Date:

  • DOI: https://doi.org/10.1007/BF01902844

Keywords

Navigation