[Pw_forum] SCF convergency

Paolo Giannozzi giannozz at nest.sns.it
Mon Nov 7 17:34:09 CET 2005


On Sunday 30 October 2005 14:55, Jun-jun Yang wrote:

> I have tried other two minimization algorithms (david and diis)
>  and it seems that the result with cg is the best. 

in PWscf these are "diagonalization" algorithms. The "minimization"
algorithm uses charge densiity mixing to achieve self-consistency 
(i.e. the minimum).

If you perform the diagonalization with sufficient accuracy, the
convergence rate to self-consistency will depend only on the
mixing algorithm, not on the type of diagonalization. In practise
the various diagonalization algorithms will give small differences
in convergence. The difference in execution time is another story:
Davidson is usually faster, CG slower but requires less memory,
DIIS is fast when it works, i.e. seldom.

Your system is a nasty one: it contains transition metals, it is 
magnetic, it has a rather large unit cell, it may not behave in 
DFT as you expect ...I haven't noticed anything obviously wrong 
in the data you sent (those complete with atomic positions). I did 
notice, though, that it doesn't converge easily. Before starting 
large calculations, you should experiment a little bit with more 
manageable systems. Gaining some expertise with simpler 
systems (maybe metallic Fe and Cr, or simpler oxides of Fe and 
of Cr, if  any) should help in determining optimal parameters 
for heavier calculations. 

> I have sent this mail three times with a graph attached [...] 
> The third one should be below 40KB, but I haven't seen its 
> appearing on the list yet. 

posts that are "held for moderator's approval", for whatever 
reason (size, e-mail not registered, wrong phase of the moon)
are automatically deleted after a while. Unfortunately it is not 
possible to replace the default automatic message with a more 
realistic one.

Once again: if a message doesn't show up on the list, it has 
to be resent (once the reason for rejection is corrected)

Paolo

-- 
Paolo Giannozzi             e-mail:  giannozz at nest.sns.it
Scuola Normale Superiore    Phone:   +39/050-509876, Fax:-563513 
Piazza dei Cavalieri 7      I-56126 Pisa, Italy



More information about the users mailing list