culaSyvex ifail and abstol

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

culaSyvex ifail and abstol

Postby asempey » Wed Nov 03, 2010 8:49 am

Hi admin,

according to the cula reference, the culaSyvex(double or single) api get ifail and abstol 2 special parametres, the abstol value is recommended to 2x machine epsilon which for double is 2.2204e-016 and for single is 1.1921e-007 in my PC. the ifail parametre indicate the indice of eigenvalue , where there is a culaNoError returned or not, 0 means no error and the Eigenvector converge.

my question is i tried many matrix and never get 0 in the ifail. i tried abstol=0 or 2.2204e-016, each one never got 0 in ifail, so what does the values in the ifail mean exactely. need some explanations plz.
asempey
 
Posts: 11
Joined: Tue Sep 21, 2010 1:52 am

Re: culaSyvex ifail and abstol

Postby kyle » Wed Nov 03, 2010 8:18 pm

IFAIL is an output INTEGER array of dimension N. That contains the index of the EIGENVECTORS that failed for converge. When returning a data error, the info will tell you how many eigenvectors failed to converge. So, for example, if the data error was 3, you should look at the first 3 elements of IFAIL to see their indexes.

This matches the behavior of the standard LAPACK implementation.

On another note, don't worry about the ABSTOL value. It's not related to the eigenvalues converging.
kyle
Administrator
 
Posts: 301
Joined: Fri Jun 12, 2009 7:47 pm

Re: culaSyvex ifail and abstol

Postby asempey » Thu Nov 04, 2010 3:28 am

kyle wrote:IFAIL is an output INTEGER array of dimension N. That contains the index of the EIGENVECTORS that failed for converge. When returning a data error, the info will tell you how many eigenvectors failed to converge. So, for example, if the data error was 3, you should look at the first 3 elements of IFAIL to see their indexes.

This matches the behavior of the standard LAPACK implementation.

On another note, don't worry about the ABSTOL value. It's not related to the eigenvalues converging.


thx for ur reply, appreciate it, however i give u an example of matrix [2 1;1 2] the result of culaDSyvex gives:

Eigenvalue: 1.000000 3.000000
ifail[0]= -532545
ifail[1]= -407961618

the value of ifail are both negative and they are surely not the index of eigenvector who cant converge. so maybe i got some wrong settings of parametre of culaDSyvex? but the Eigenvalue and eigenvector are right.
asempey
 
Posts: 11
Joined: Tue Sep 21, 2010 1:52 am

Re: culaSyvex ifail and abstol

Postby kyle » Thu Nov 04, 2010 5:42 am

The value of IFAIL when there is a no culaDataError is undetermined. Hence the gibberish.
kyle
Administrator
 
Posts: 301
Joined: Fri Jun 12, 2009 7:47 pm

Re: culaSyvex ifail and abstol

Postby asempey » Thu Nov 04, 2010 6:05 am

Ok got it, thx for ur reply admin, i appreciate it.
asempey
 
Posts: 11
Joined: Tue Sep 21, 2010 1:52 am


Return to CULA Dense Support

Who is online

Users browsing this forum: No registered users and 0 guests

cron