[MATH-342] SVD crashes when applied to a strongly rectangular matrix (typical case of least-squares problem)

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 2.0
• Fix Version/s: 2.1
• Assignee: Dimitri Pourbaix
• Reporter: Dimitri Pourbaix
• Created: 21/Feb/10 21:41
• Updated: 03/Apr/10 03:45
• Resolved: 21/Feb/10 21:50

Description

When SVD is applied to a strongly rectangular matrix (number of rows way larger than number of columns, typical case of least-squares problem), finite precision arithmetics shows up:

• in EigenDecompositionImpl.isSymmetric: a by-definition symmetric matrix returns false;
• in EigenDecompositionImpl.findEigenVectors: too many iterations exception