LU decomposition problem

General CULA Dense (LAPACK & BLAS) support and troubleshooting. Use this forum if you are having a general problem or have encountered a bug.

LU decomposition problem

Postby syrka » Fri Feb 11, 2011 2:39 am

Can anyone tell me why CULA have problem with this matrix?


Matrix S and matrix L*D*L^T should be the same.

when I take different matrix, it's works correctly
Attachments
wydruk2.JPG
wydruk2.JPG (91.35 KiB) Viewed 3766 times
wydruk.JPG
wydruk.JPG (93.91 KiB) Viewed 3766 times
syrka
 
Posts: 2
Joined: Thu Feb 10, 2011 1:29 pm

Re: LU decomposition problem

Postby kyle » Fri Feb 11, 2011 6:44 am

LU factorization is not LDL^T factorization. Mainly the pivots will ruin your symmetric input matrix. Try xPOTRF if your matrix is positive definite.
kyle
Administrator
 
Posts: 301
Joined: Fri Jun 12, 2009 7:47 pm

Re: LU decomposition problem

Postby syrka » Fri Feb 11, 2011 8:18 am

I also use xPOTRF but I want have two options in my work. I read in some source that I can transform LU to LDL^T if my matrix is symetric. There is a version xGETRF without pivot? Because as I wrote the Doolittle LU method on CPU without piwot it works
syrka
 
Posts: 2
Joined: Thu Feb 10, 2011 1:29 pm

Re: LU decomposition problem

Postby john » Fri Feb 11, 2011 9:43 am

LDLT is the SYTRF routine, which is not (yet) in CULA. It's similar to a cross between POTRF and GETRF, because it maintains the symmetric nature of the matrix, but also features pivoting. LU is rarely implemented without a pivot because it will very likely lead to unstable solutions.
john
Administrator
 
Posts: 587
Joined: Thu Jul 23, 2009 2:31 pm


Return to CULA Dense Support

Who is online

Users browsing this forum: No registered users and 1 guest

cron