The ABINIT Users Mailing List ( CLOSED )

[zhangting <zhangting@pku.edu.cn>]

Anglade Pierre-Matthieu，您好！

 

　　Thank you very much too!

 

======== 2006-06-06 11:26:00 您在来信中写道： ========

 

Hi,

It might have been useful to join you input file... Our comment would have been more accurate then.
Nevertheless here are a few basic recommandation:
- try changing the precondtioning shcheme your using. For very difficult system
"iprcel 49" may work best.
-try changing the mixing scheme: even simple mixing can sometimes out perform conjugate gradient. In abinit version 5.x.x you have "iscf 7" which is usualy very efficient
-try using "nline " and "nnsclo" they'll help improving the true convergence of your caclulation
-and yes tolvrs 1e-18 is almost mandatory if you wan't an nice phonon calculation.

regards

PMA

On 6/6/06, zhangtingPKU <zhangting@pku.edu.cn> wrote:

Dear ABINIT users:

    I'm calculating the (5,5) CNT's phonon spectrum with ABINIT code, with the input file like in the lesson_rf2. But I found that the first database, which is the initio SCF calculation, converges very slow. With the accuracy in the lesson, tolvrs=1E-18, the calculation has already run for about 600 SCF-steps, but has not reached the convergence condition yet. In fact it oscillates at about deltavrs = 1E-17 ~ 1E-16. My question is that in the first step in a phonon calculation, does it really necessary to set a converge condition as telvrs=1E-18 ? Or am I got something wrong within the input file, such that it can't reach the converge condition?(My input files is attached) Thanks!




致
礼！


zhangtingPKU
zhangting@pku.edu.cn
2006-06-06

--
Pierre-Matthieu Anglade

= = = = = = = = = = = = = = = = = = = = = =

　　　　　　　　致
礼！

 

　　　　　　　　　　　　　　zhangting

　　　　　　　　　　　　　　zhangting@pku.edu.cn

　　　　　　　　　　　　　　　　　2006-06-08