Iron spin(转载）

greencalm

问：Dear users

I am using CASTEP to carry out electronic structure calculations on interfaces
in steel. I am trying to carry out spin optimised calculations on Fe supercells
(containing 50 atoms). As I understand  I have to specify an initial spin value
equivalent to the high spin value of the system (ie approx 200+) as it is not
possible to reach the high spin state from the low spin state. With these
values I get the following error:

Error in parameters file: SPIN-> |spin| too big

I would be very grateful if somebody could explain what I am doing wrong and how
the initial spin is related to the system being used. How important is this
initial value? The above error does not occur when I decrease the initial spin
value or increase the number of empty bands. However often the band structure
does not converge. Is there a general reason why this would happen? I would
also appreciate it if somebody could explain what I need to do in terms of
increasing the number of empty bands and how this is related as well.

Thanks
Charlotte

greencalm

The main difficulty with a calculation that has spin as a degree of
freedom is that is is a multi-minimum problem for the electronic
structure. It will (should!) find a minimum but there is no guarantee that
the minimum energy electronic structure will be the spin state with the
lowest energy. Hence it's always good to start with a spin state that is
"close" to the final state - that will help. By default castep will start
with a spin of zero/one for an even/odd electron system run as spin
polarised. This is quite far from the spin of 200 you expect (50 atoms, 4
unpaired d electrons per atom).

As for the error you get, that occurs when you attempt a spin state for
which there are not enough electrons in the system do do it (eg. you can't
get a spin of 200 with less than 200 electrons). Alternatively you do not
have enough bands in which to put the electrons. This implies that you
have a mistake in your input files (obviously!). So let me ask some stupid
questions: are you sure you have 50 atoms? Are you sure your
pseudopotential is correct? Are you sure you are asking for a spin of 200.
Do you have enough bands (Fe uspcc has 8 electrons, so you have 400
electrons in the system - for a spin of 200, you will need more than 300
up bands and 100 down bands). As a quick test, set the spin ot zero and in
the header of castep it will tell you the answer to each of these
questions.

The number of empty bands is important too. When doing a metallic system
we need empty bands as part of the minimisation procedure so that we can
find the Fermi level. In spin polarised calculations we quite often need
quite a few more empty bands than in the non-spinpolarised case because
the up and down electrons are not equally populated. If you do not have
enough empty bands then you might not be able to get to the spin state
required and it also makes the task of the minimiser more difficult.

In a band structure calculation, it occationally says that it is
unconverged. The default tolerances in the band structure calculation are
quite tight so you may need to increase the number of iterations.

Best Regards,
Stewart.

阿欢

好文章！顶