> >>it is normal that as you increase your system to include many >> >equivalent (or almost) atoms > > I do not understand why many equivalent atoms will affect the convergence.
Hi, I explained it already: >> because of electrons flopping between nearly degenerate states. In other words, the states just under the Fermi energy can be occupied in different way, and the electrons are just shifted around, without reaching the convergence. >>To get an idea, look at the results (Mulliken charges; DOS) >> >underway, after some number of iterations. > > Could you give me some ditails about checking the results of Mulliken > charges or DOS after some number of iterations? You print out Mulliken charges, there is a switch for it. You look at them and think whether they are as you expect (i.e., the atoms which ought to be magnetic are magnetic, and in a right way). Check if you find anything useful in the following lecture: http://www.home.uni-osnabrueck.de/apostnik/Lectures/APostnikov-Magnets.pdf > Usually, I just check the following information of the scf. >--------------------------------------------------------------------------- > siesta: iscf Eharris(eV) E_KS(eV) FreeEng(eV) dDmax Ef_up > Ef_dn(eV) > ...... > siesta: 799 -32180.0924 -32180.0919 -32180.0919 0.0011 -3.0670 > -3.0670 > siesta: 800 -32180.0924 -32180.0919 -32180.0919 0.0011 -3.0670 > -3.0670 ... > siesta: 826 -32180.0929 -32180.0925 -32180.0925 0.0010 -3.0669 > -3.0669 > siesta: 827 -32180.0931 -32180.0927 -32180.0927 0.0010 -3.0668 > -3.0668 > ...... This is good but does not tell you anything but that your calculation is not quite converged. In fact it doesn't look like a divergence, and the Fermi level(s) are not a priori unreasonable. So, before doing anything else, try to make sense out of your calculation (position of bands, magnetic moments, etc.) ---------------------------------------------------------------------------- > > I'd like to increase the DM.MixingWeight to 0.3 to accelerate the > calculation. Is it correct? If it smoothly converges to a right solution, that's fine. But I'd rather suggest, with system as large as yours, to be more prudent. 0.3 is a huge mixing weight. > Since you said I need a small mixing > parameter, > how to understand the mixing parameter? I read some notes about > DM.MxingWeight, but there is still something puzzled me. > > alpha has to be small (0.1-0.3) for insulator and semiconductors, > tipically much smaller for metals. Yes; what is the problem? In metals you need a good sampling of states near the Fermi level; imagine you have a narrow band which shifts across the E_f; then you'd like to damp the resulting fluctuations. In semiconductors the situation is (usually) less critical because bands are either occupied or empty. > BTW, whether my setting of the antiferromagnetism is correct. I don't know the structure, sorry. Apart from this - yes; some Fe atoms marked "+" and others "-", as it should be. Best regards Andrei Postnikov
