Hi Arun, The fact that you initialize the atomic spins does not guarantee the convergence to a state with spins of the same initial sign. Depending on the system, it could have many different (meta-stable) magnetic states, and for a given fixed geometry, it's SCF that would determine it, ultimately. It's up to you to find initial states that will lead to different final magnetic states, as far as I can see.
Maybe (but this is a wild guess) you could check if setting a very small mixing parameter would change the final spin state at, say the end of the first complete SCF cycle. If it doesn't, this means that the final state is robust and not just an artifact of a large mixing parameter. Usually you'll really have to start from many different spin configurations and compare the total energies of the eventually different final states to determine who is the candidate to a ground state. About Mulliken analysis, do not trust it blindly, as it is highly dependent on the basis set used. Check the list archives for that, but if I'm not mistaken water is the "classic" example. I'm not sure if results for spin states could be as unreliable as the individual charges on the atoms. Cheers, Marcos On Wed, Nov 4, 2009 at 5:13 AM, Arun Kumar Manna <[email protected]> wrote: > > Dear All, > > I am facing problem with spin-polarized AF calculation using of > DM.InitSpin block in siesta-2.0. > If I specify '+' or '-' in the block, according to siesta manual > it should start with maximum possible +ve or -ve spin polarization > on these two specified sites and after scf density matrix convergence > it should get into the right values for those site. So, there should > not be any spin inversion ('+' becomes '-') for any site. > > But in my calculation, I have observed that although explicitly I am > mentioning the sign of spin moment of a specified site, it gets reverted. > This is clear from Mulliken population analysis. For that specifiled site, > Q(up)-Q(down) becomes opposite in sign. > > Could anyone please tell me what is happening here? > > Regards, > Arun > -- > PhD Student > JNCASR, Bangalore-560064 > India >
