You are here: MIMS > EPrints
MIMS EPrints

2008.49: A Newton Algorithm for the Nearest Correlation Matrix

2008.49: Rudiger Borsdorf (2007) A Newton Algorithm for the Nearest Correlation Matrix. Masters thesis, University of Manchester.

Full text available as:

PDF - Requires a PDF viewer such as GSview, Xpdf or Adobe Acrobat Reader
719 Kb

Abstract

Firstly, we describe and investigate the algorithm of Qi and Sun which solves the problem of finding the nearest correlation matrix to a symmetric matrix. This algorithm claims a quadratic convergence. We discuss improving this algorithm's efficiency and reliability and detect a problem when we are aiming at a nearest correlation matrix with a high accuracy, using small error tolerences. As a consequence, we suggest a modified version, based on the algorithm of Qi and Sun, which is also a quadratically convergent algorithm, has improved efficiency and is modified so that the algorithm can return the nearest correlation matrix to high accuracy showing a robust and reliable behaviour.

Secondly, we investigate the general alternating projections method and also Higham's alternating projections method for the nearest correlation matrix. We discuss variations of the latter and include a further projection which allows more constraints to be added to the problem. We introduce a new algorithm and compare its convergence behaviour with Higham's alternating projections method.

Item Type:Thesis (Masters)
Additional Information:

Uncontrolled Keywords:correlation matrix, positive semidefinite matrix, Newton's method, preconditioning, rounding error, Armijo line search conditions, alternating projections method
Subjects:MSC 2000 > 15 Linear and multilinear algebra; matrix theory
MSC 2000 > 65 Numerical analysis
MIMS number:2008.49
Deposited By:Nick Higham
Deposited On:21 April 2008

Download Statistics: last 4 weeks
Repository Staff Only: edit this item