Least-Squares Data Adjustment with Rank-Deficient Data Covariance Matrices
- The University of Arizona, Tucson, AZ 85721-0119 (United States)
A derivation of the linear least-squares adjustment formulae is required that avoids the assumption that the covariance matrix of prior parameters can be inverted. Possible proofs are of several kinds, including: (i) extension of standard results for the linear regression formulae, and (ii) minimization by differentiation of a quadratic form of the deviations in parameters and responses. In this paper, the least-squares adjustment equations are derived in both these ways, while explicitly assuming that the covariance matrix of prior parameters is singular. It will be proved that the solutions are unique and that, contrary to statements that have appeared in the literature, the least-squares adjustment problem is not ill-posed. No modification is required to the adjustment formulae that have been used in the past in the case of a singular covariance matrix for the priors. In conclusion: The linear least-squares adjustment formula that has been used in the past is valid in the case of a singular covariance matrix for the covariance matrix of prior parameters. Furthermore, it provides a unique solution. Statements in the literature, to the effect that the problem is ill-posed are wrong. No regularization of the problem is required. This has been proved in the present paper by two methods, while explicitly assuming that the covariance matrix of prior parameters is singular: i) extension of standard results for the linear regression formulae, and (ii) minimization by differentiation of a quadratic form of the deviations in parameters and responses. No modification is needed to the adjustment formulae that have been used in the past. (author)
- Research Organization:
- American Society for Testing and Materials - ASTM International, 100 Barr Harbor Drive, PO Box C700, West Conshohocken, PA, 19428-2959 (United States); European Working Group on Reactor Dosimetry - EWGRD, SCK.CEN, Mol (Belgium)
- OSTI ID:
- 22086976
- Report Number(s):
- INIS-US-13-ISRD-14-P1-09; TRN: US13V0003045423
- Resource Relation:
- Conference: ISRD-14: 14. International Symposium on Reactor Dosimetry, Bretton Woods, NH (United States), 22-27 May 2011; Other Information: Country of input: France; 6 refs.
- Country of Publication:
- United States
- Language:
- English
Similar Records
Regularization of feedwater flow rate inference
Inverse determination of boundary conditions in steady heat conduction with heat generation