[Pw_forum] the warning in scf

Stefano de Gironcoli degironc at sissa.it
Mon Jan 3 09:50:02 CET 2005


It's a warning saying that the current threshold in the iterative
diagonalization was too
tight to allows convergence in a small number of iterations for some,
usually the last,
eigenvectors.

If this happens in one or a few intermediate iterations toward scf you
should not worry.
If this happens in the last iteration when scf is achieved this may, in
principle, affect the final results.
If your system is a metal (or even if it is not) you can try to increase
the number of bands and/or change
a little the mixing parameter and see if this solve the problem.

Stefano de Gironcoli

Shaorui Sun wrote:

>Dear all:
>       when I calculated the HgS, there is some warning in one scf iteration as follows:
>------------------------------------------------------------------------------
>      iteration #  6     ecut=    20.00 ryd     beta=0.70
>     Davidson diagonalization (with overlap)
>   WARNING:     1 eigenvalues not converged
> warning :   1 eigenvectors not converged after   1 attemps
>   WARNING:     1 eigenvalues not converged
> warning :   1 eigenvectors not converged after   1 attemps
>   WARNING:     1 eigenvalues not converged
> warning :   1 eigenvectors not converged after   1 attemps
>   WARNING:     1 eigenvalues not converged
> warning :   1 eigenvectors not converged after   1 attemps
>     ethr =  5.94E-06,  avg # of iterations =  5.8
>
>     total energy              =  -291.67066368 ryd
>     estimated scf accuracy    <     0.03218677 ryd
>
>     total cpu time spent up to now is   1832.44 secs
>--------------------------------------------------------------------------
>Is it a fatal warning?
>
>
>Best Regards!
>                        S. R. Sun
>---------------------------------------------
>Beijing Synchrotron Radiation Facility
>Institute of High Energy Physics
>Chinese Academy of Sciences
>P. O. Box 918, 100049 Beijing
>P. R. China
>Tel: 0086+10 88236710
>email:sunsr at ihep.ac.cn
>
>_______________________________________________
>Pw_forum mailing list
>Pw_forum at pwscf.org
>http://www.democritos.it/mailman/listinfo/pw_forum
>  
>






More information about the Pw_forum mailing list