[Pw_forum] MgFeO LDA+U convergence problem
Matteo Cococcioni
matteo at umn.edu
Thu May 17 16:42:58 CEST 2007
Dear Sasha,
I'm not sure you got my reply (the message seemed to be too big for some
reasons...) so I will try again.
I think you can get a good idea on how to choose the starting
occupations looking at the eigenvectors of the occupation matrix (from the
first iteration). another thing you can do is decrease the initial
magnetization on atoms 1 and 2 (that seem the most problematic ones):
try to use
for example 0.4 and -0.3.
hope this helps.
Matteo
sasha wrote:
> Greetings Matteo,
> Thanks for the advice, your suggestion in fact works very well. It
> turns out that I should have more closely examined my input. I have
> another quick question regarding the starting_ns_eigenvalue, in the
> previous example (below) from the eigenvalues at the first iteration of
> the "standard" lda+u calculation it is clear how to set the
> ns_eigenvalues however what does one do if the starting eigenvalues are
> as follows:
>
> atom 1 spin 1 0.9985560 0.9998587 1.0007556 1.0013322 1.0057882
> atom 1 spin 2 0.0484071 0.0493657 0.0578860 0.1216687 0.1308162
>
> atom 2 spin 1 0.0535213 0.0592929 0.1256650 0.1321563 0.1439048
> atom 2 spin 2 0.9979128 0.9993943 1.0004671 1.0010265 1.0058542
>
> atom 3 spin 1 0.9947113 0.9953027 0.9992115 0.9997585 1.0044052
> atom 3 spin 2 0.0693133 0.1280564 0.1459512 0.1547219 0.4858838
>
> atom 4 spin 1 0.0637767 0.0681158 0.1421888 0.1436934 0.2272403
> atom 4 spin 2 0.9965028 0.9979222 0.9991654 1.0005289 1.0054916
>
> where only some of the eigenvalues for certain atoms and spins are
> degenerate? Once again thank you in advance for any advice or
> suggestions.
> regards,
> Sasha
> On Mon, 2007-05-14 at 09:48 -0500, Matteo Cococcioni wrote:
>
>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: matteo.vcf
Type: text/x-vcard
Size: 294 bytes
Desc: not available
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20070517/a5c9dcab/attachment.vcf>
More information about the users
mailing list