|
Article information
2004 , Volume 9, ¹ 4, p.66-76
Mikheev S.E.
Convex quadratic approximation
If a sum of quadratic deviations of parameters of a convex quadratic approximation (CQA) from the parameters of an unconditional quadratic approximation (UCA) is selected as a quality criteria of CQA, then the best CQA, according to this criterion, may be constructed in two finite stages. At first, one finds the best UQA, according to its own criteria. Then, the finite algorithm finds (in the convex cone of positively semi definite matrix) the element that is closest to the matrix of quadratic form in UQA, which is a quadratic part of the best CQA. The linear part of the best CQA coincides with the linear part of the best UQA. Validation of the algorithm is presented. Dependence of the node choice on the best CQA is investigated.
[full text] Classificator Msc2000:- *90C20 Quadratic programming
- 90C25 Convex programming
Keywords: discrepancy, unitary matrix, minimizer
Author(s):
Mikheev Serguei Eugenievich
Dr. , Professor
Position: Associate Professor
Office: St. Petersburg State University
Address: 198504, Russia, St-Petersburg, SPb, Universitetskiy pr., 35
Phone Office: (812) 428 42 91
E-mail: him2@mail.ru
Bibliography link:
Mikheev S.E. Convex quadratic approximation // Computational technologies. 2004. V. 9. ¹ 4. P. 66-76
|
