commons-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Luc Maisonobe (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (MATH-1100) QR factorization fails in revealing rank-deficient matrix
Date Thu, 20 Feb 2014 16:53:20 GMT

    [ https://issues.apache.org/jira/browse/MATH-1100?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13907168#comment-13907168
] 

Luc Maisonobe commented on MATH-1100:
-------------------------------------

In your code, you don't set the singularity threshold as you call new QRDecomposition(M2)
with only the matrix as an argument.
This ends up with using an exact 0 as the threshold.

If you use new QRDecomposition(M2, 2.2e-14), the matrix is corrctly identified as singular.
There seem to be two very small values after decomposition on the diagonal of the triangula
matrix. One is about 2.05e-14 the other is about 2.2e-14, so depending on the threshold, you
should get a rank of 379, 380, or 381.

Do you agree with this analysis?

> QR factorization fails in revealing rank-deficient matrix
> ---------------------------------------------------------
>
>                 Key: MATH-1100
>                 URL: https://issues.apache.org/jira/browse/MATH-1100
>             Project: Commons Math
>          Issue Type: Bug
>    Affects Versions: 3.2
>         Environment: Java HotSpot(TM) Client VM (build 19.1-b02, mixed mode, sharing)
an windows 7 professional
>            Reporter: alberto trivellato
>         Attachments: QRtest.zip
>
>
> given a matrix that has not full rank, the method getSolver().isNonSingular() of the
class QRDecomposition returns true.



--
This message was sent by Atlassian JIRA
(v6.1.5#6160)

Mime
View raw message