[Pw_forum] a question about LDA+U
rjxiao at blem.ac.cn
Sat Oct 23 18:07:32 CEST 2004
Dear Stefano Fabris,
Thank you very much for your instructive advice.I did some calculation as what you have told me.The cell with 2 iron atoms was choose because it can be finished quickly.I did a series of LDA+U by setting U=0.000000001(since the LDA+U can not be done under U=0 in this software,I use this small value to instead U=0),U=4.29,5.29,6.29,7.29,8.29,9.29.For both irons, the eigenvalues of one spin direction are close to 1, while eigenvalues of the other spin direction are fractional.The occupations show the same result.But I also notice that three of the fractional eigenvalues approach to 0 along with the increase of U and the other two approach to 1.For example, when U=0,they are 0.3051347 0.3051347 0.4077667 0.4469277 0.4469277,for U=9.29, they are 0.0544810 0.1226590 0.1226590 0.7087178 0.7087178. It seems that the Hubbard+U does work. But the result is still a conducting solution even if U is as high as 9.29. It is strange to me.
Another phenomena puzzled me is the negative occupation.How can the occupation be such as -0.18?It is difficult to understand for me.
The problem of convergency in model with 8 iron atoms without U is solved.But the way to slove it is unexpected to me. I just change one position of Fe1 and Fe2,though I think that the two models are equivalent for the FeO with antiferromagnetic between neighour  plane.Maybe the model after change is more beautiful:-) I don't know why there is such great differece between them. Also, the model with U=4.29 is not convergency, and the occupations and eigenvalues are fractional.
Institute of Physics,
Chinese Academy of Sciences
More information about the users