convergence problem :: Hubbard-I

[gmm05126]

Dear all,

I am experiencing convergence problem for 4f system with spin-orbit coupling within Hubbard I approximation. I am observing a periodic pattern in quantities (to which I am checking convergence) as the total energy (:TOT) and charge distance (:DIS) in case.scf file. For LDA+DMFT I have adapted Ce-gamma.py script [http://ipht.cea.fr/triqs/doc/user_manual/wien2k/Ce-gamma_script.html]
according to advice in
[https://github.com/TRIQS/triqs/issues/140].

I have first performed self-consistent GGA (should one strictly use LSDA?) and after I have continued with Hubbard-I. In attached figure
https://dl.dropboxusercontent.com/u/83195899/Hub-I_convergence.png
you can see plot of total energy (top panel) and charge distance (bottom panel) as a function of iterations. From beginning system after 25 iterations total energy started to oscillate while the charge density blows up. After 65 iterations I have changed case.py file on the fly (not interrupting self-consistent calculations) and have increased number of Loop to 4 and decreased Mix to 0.3. A periodic pattern in total energy and charge distance appeared and further playing around with the parameters and mixing scheme does not really help.

Could you please comment on this convergence issue or give some advices how to improve it?
Thanks in advance,
Martin Gmitra
Uni Regensburg

[leopo]

I have not tested it with GGA. Do you experience the same problem with LDA?

The problem may come from the initial spin-polarization you have when you run GGA with Wien-2k. In Hubbard-I the Ce 4f levels are atomic, i.e. sharp, and even very small changes in the GGA exchange field is sufficient to change significantly their occupancy, in particular, at low T. Hence, the resulting spin-polarization, which is then used to generate the new Kohn-Sham potential, changes as well.

You may try to regenerate GGA Wein-2k input files starting from *.inst file with no initial spin-polarization. Then after your GGA run you should get no spin-polarization at all (check *.scf file) . After that you may try to run GGA+DMFT starting from it.

[gmm05126]

Thanks for your hints. I have to apologize for being not so specific. Let me put more details. The system I am doing is hexaboride SmB6, and I would like to reproduce results from the paper Phys. Rev. B 79, 165104 (2009) [http://link.aps.org/doi/10.1103/PhysRevB.79.165104].
In self-consistent Wien2k calculations I am using for xc LDA (option: 5) and the overall job was executed with runsp_c_lapw to fix magnetic moment to zero value on the level of LDA. The LDA+DMFT (Hubbart-I) introduced a tiny change in magnetic moment (:MMTOT) of about 0.00002 in the present (oscillatory) calculations.

The problem with the oscillations is actually with LDA. I am newbie and I do not have feeling what convergence one can expect/achieve with LDA+DMFT. Should the :ENE and :TOT goes zero as one observes in LDA self-consistent calculations?

Thank.
Martin