[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Help-gsl] test release for GSL 2.2

From: Peter Johansson
Subject: Re: [Help-gsl] test release for GSL 2.2
Date: Tue, 30 Aug 2016 11:42:35 +1000
User-agent: Mozilla/5.0 (X11; Linux x86_64; rv:45.0) Gecko/20100101 Thunderbird/45.2

Thanks. That's very useful.


On 08/30/2016 11:30 AM, Patrick Alken wrote:

Ah yes that helps narrow it down. The behavior of gsl_linalg_cholesky_decomp has changed slightly. Before (2.1 and prior) the work was done using the lower triangle and then the lower triangle was copied to the upper triangle to have a fully symmetric matrix.

However, in order to use the new condition number estimation, gsl_linalg_cholesky_rcond, it needs to compute the norm of the original matrix. Therefore, now the cholesky factor is stored in the lower triangle and the upper triangle contains the original matrix.

This is more in line with the lapack storage scheme, although I realize it will break a few codes, such as yours.

One quick fix would be to add a call to gsl_matrix_transpose_tricpy after your cholesky decomp call to copy the cholesky factor to the upper triangle.

Long term it would be better to only reference the lower triangle.

Sorry for the trouble :/

On Aug 29, 2016 5:32 PM, Peter Johansson <address@hidden> wrote:

    Sorry I missed to answer this question.

    On 08/30/2016 09:27 AM, Peter Johansson wrote:
    > Does your code work with GSL 2.1?

    Yes, I get different behaviour when I link with GSL 2.1 and GSL
    That's why I'm concerned that something has changed in GSL.


reply via email to

[Prev in Thread] Current Thread [Next in Thread]