2007/11/9, Marcos Verissimo Alves <[EMAIL PROTECTED]>: > > Hi Vasilii, > > Vous avez écrit / You have written / Lei ha scritto / Você escreveu... > Vasilii Artyukhov > > Hi everybody, > > > > I have a somewhat awkward question. I'm relaxing a structure that is > > supposed to be magnetic. This means, a SCF run with a broken initial > spin > > symmetry gives a nonzero magnetic moment. But then, after some > relaxation > > steps (unfortunately, SIESTA only outputs the magnetic moment at the > final > > step - or am I missing some input flag?), the band structure becomes > > completely spin-degenerate. > > If you set the mulliken charges to be output at the end of each scf step, > you can see the mulliken charges for up and down spins. At the end of each > spin listing, siesta gives you the spin charges. Then you just have to > subtract one from the other.
I have a big unit cell, so this would seriously harm my log file :) > > > So after the relaxation I have a final structure with a zero magnetic > > moment. > > Check the mulliken charges on each atom on the last scf step: you might > have some magnetic ordering that gives you an overall zero net spin. Checked the DOS and BS, 100.00% coincidence for alpha and beta spin. > But when I restart the calculation building the DM from scratch, > > the SCF gives a nonzero magnetic moment, and a nonzero force. > > I will suppose that you are using the XV file as an input to this restart. > If this is so, it strikes me as a bit weird, but there could be small > numerical errors that might lead to what you are seeing. No, the restart for other structures gives exactly the same force etc. (set aside negligible differences from an extra SCF step). However, at the > end of this new relaxation, does the structure return to a zero net spin? > Or does it acquire a new, appreciable net spin? I ran a second relaxation, and it gave an extremely weak magnetic ordering. Have to look at the BS in more detail to figure it all out. So, the problem is solved by now, but the interesting part would be to hear some physical reasoning for that. > The questions are: > > > > - what physical conclusions could be made from this behavior (AFM, > etc.)?, > > and > > In principle, none. Magnetic systems can be very sensitive to the choice > of parameters, namely the mesh cutoff and the k-point sampling but, maybe > even more importantly, to the electronic temperature (that is, the > smearing). When I did my calculations for vacancies in graphene, I ran a > calculation for a supercell that became non-magnetic at some point of the > CG run. I simply stopped the calculation, lowered the electronic > temperature (which I had raised because I was having problems in making > the initial SCF cycles to converge) and re-started using the XV, but not > the DM. It went to the magnetic state that I was expecting. My experience > is that some systems can become non-magnetic because of this, especially > if the magnetism is not too robust with respect to stresses. Yes, that's exactly why I performed all the calculations with a zero electronic temperature. As I said in the first message, the mesh cutoff and k-point sampling have been thoroughly tested. Btw, I'm also doing vacancies in graphene, so it would be interesting if you could drop me a couple of lines about what you were doing off the list :) > - how do I make SIESTA build a new initial density matrix at every > > relaxation step? > > Why would you want to do this? It would waste your time. The best thing to > do is to relax your system until the forces are reasonably low (say of the > order of 0.1 eV/Ang) and then to restart using the XV or STRUCT file, but > re-building the density matrix. That's what I did (see above), and it seems to have worked out. But rebuilding the DM at every step, while wasting my time, would have saved me another run. Another thing that might help is to > diminish the max displacement and set PreconditionVariableCell to a value > higher than the default (say 6.5-7 ang) in order to limit the changes in > the cell vectors. Oops, haven't seen this flag in the manual. What does it do? Anyway, I'm doing fixed cell calculations since the manual says that cell vector constraints are not yet implemented (also, it's somewhat simpler this way in my case). Thanks a lot for your comments! Best regards, Vasilii
