spinor-talk — For people to stay in touch

[Spinor-talk] Re: spinor

[Gerhard T. <the...@sa...>]

Dear Sushil,

1) The [FixedSpinMoment] option can only be used in combination with
spin eigen states, i.e. not with general spinor wave functions. Then,
for spin polarized calculations the FixedSpinMoment settings offer the
option to enforce a certain spin polarization either for the entire
calculation of for a certain number of iterations before this constraint
is dropped.

2) [SpinSplit] controls the output of energy differences between bands.
It has no affect on the calculation!

3) [B] offers the option to apply an external magnetic field for a
certain number of iterations. It is used to force or direct a
calculation towards a certain spin alignment.

4) When you use general spinor states and GLSDA you can in principle
obtain non-collinear spin arrangements. Since no atomic sphere
approximation is made the non-collinearities can occur on the atomic
scale. Keep in mind however that the local density approximation does
not include any angle terms! Partially occupied spin-orbit bands can
also lead to small non-collinearities on the atomic scale.

Auluck Sushil wrote:
> 
>         dear dr theurich,
>         this is wishing you a great 2002.
>         could you please throw some light on these points...
>         1.....FixedSpinMoment
>         2.....SpinSplit
>         3.....B
>         4.....non collinear magnetism

[Spinor-talk] Re: spinor

[Gerhard T. <the...@sa...>]

Dear Jaime,

First off, I am really sorry for how long it took me to get back to you.
There has been a lot going on at work this past week that needed my
immediate attention and everything else had to wait.

O.k. then let me answer your questions. Best, I think, will be to answer
between the lines, but before I have a question. Did you download the
code from the source forge web site or did you get it some other way?
The reason I ask is for me to know which version you are using.

> 
> 1. I do not know how to generate the ssd files, and if they have any
> explicit rules. In particular, I need the ssd files for bcc and fcc.

I assume you mean the *.ssg files, which stands for spin-space-group.
You can find a description of the file format on the web page. As long
as you are not doing complicated spin arrangements including spin-orbit
coupling it is very easy to set these files up by hand. I always copy an
existing *.ssg file which gives me all 48 cubic symmetry operations for
the first block and then I edit the remaining blocks. Now for bcc and
fcc you can use the Oh.ssg file (Oh obviously stands for the
corresponding point group) which you can get out of the 'misc' package
on spinors download page which I assume you used before.

> 2. How feasible is to use your code for molecules and clusters? say
> Fe_3 or Mn_5?

I never tried it since I was always working on bulk materials. But a
definite downside of Spinor is its speed. It is very slow! The reason
for this is that only the direct energy minimization with conjugate
gradients is correctly implemented. The much faster charge density
mixing method is somehow broken and I never had time to fix it. Also,
the code does not offer any level of parallelism. These are all things
that would need to be worked on to make the code more efficient.
However, if you are only planing to use the code in order to evaluate
some other results it should be worth a try even for molecules and
clusters. 


> 3. How could I generate the pseudopotentials? I have not enough
> information to use the uniPPlib. I was not able to get the phi2upf tool

Yes, I noticed that I did not add the fhi2upf tool to the download page.
I will attach the source file to this email. For instructions check the
web page. 

The answer to your question really depends on whether you want to
include spin-orbit coupling in the calculation or not. The FHI98PP code
is only a scalar-relativistic code. Can Siesta read in the FHI
potentials? If yes then this would be a good way to go. If not then I
have a couple of patches to the pseudopotential code that comes with the
Siesta package (maintained by Alberto Garcia I guess) which allows to
make upf files at the same time that you make Siesta style
pseudopotentials.


> 4. Which xc functionals does the code handle? I was able to find PZ81, but
> nothing else.

Yes, that's all I implemented. Yet another area where one could add some
more features to the code...

> 
> Now, I've got a few questions regarding the param file:
> 
> 1. [Init] What is the meaning of the parameters in this block?

This block is a bit complicated. The first number is simply the number
of initialization lines to follow. Then each initialization line
contains 4 columns. The first column holds the number of bands that are
affected by this line, the second entry specifies which initialization
mode is to be used for these bands, either the initialization considers
the kinetic energy of each state in the plane wave expansion or it sets
up the initial states using the atomic pseudo wave functions that come
with the pseudopotentials. The meaning of the third entry really depends
on whether you are using general spinor wave functions or not. In the
latter case the third entry controls whether these initialized bands are
to be considered spin up or down bands. In the case of general spinor
wave functions it only influences the initial wave functions but each
band can still change character during the course of the calculation.
Finally, the fourth entry in each line is either the x in the weighing
factor (e_kin^x) for each plane wave component if kinetic initialization
is chosen for these bands or, for the pseudo wave function choice, it
uses random linear combinations when this parameter is -1 and otherwise
(>-1) it sets the wavefunctions of this line equal to the corresponding
pseudo orbital. This is really confusing at first but it does make
sense.

> 2. [Grid] Which criterium do you use to have integrals converged?

The grid is chosen as to match the required cutoff energy. Since only
LDA is implemented it is sufficient to choose a grid which is 2*G_max in
each direction where G_max is the largest plane wave component still
under the cutoff. The determination of the cutoff needs to be done by
hand for each pseudopotential by repeatedly increasing the cutoff until
convergence is achieved for the total energy.

> 3. [Subspace] I do not know waht this is.

This is of importance for metallic systems with fractional occupation
numbers. When this is switched on then after each minimization a
diagonalization of the Hamiltonian is carried out in the subspace of the
current Kohn-Sham states obtaining better trial states for the next
step.

> 4. [Symmetry] symiter ?

In order to speed up the direct conjugate gradient steps it is often
convenient to NOT symmetrize each charge density that is calculated on
the way. However in certain cases (few k-points) this will lead to
unphysical charge densities especially in the beginning of the
calculation. symiter controls how many iterations shall include
symmetrization of all occurring charge densities. After this number of
iterations is reached symmetrization will no longer be carried out
during the cg step.

> 
> Finally, concerning output files:
> 1. Which information is stored in the wave-functions cwf info and
> dat files? Likewise occup files, and lpot.dat.

There is a cwf.info and cwf.dat file for each k-point. The info file
stores the number of plane waves at this k-point and the number of
bands. It also holds the table mapping the expansion coefficients to the
FFT grid. The dat files contain the complex coefficients for each band.
There are always two components per line (up and down for spinor
states). 

The occup files store occupation of each band in the first line and spin
information for each state in the second line. Again there are as many
occup files as there are k points.

The lpot.dat file contains the real space local part of the total
potential. This information is only produced for post-processing
purposes.

> 2. edens.dat probably is the electronic density, bndstrct the band
> structure, and dos the density of states, but do not understand the
> format.

So the edens.dat is a long vector through the FFT grid with the fastest
moving index being ngz, then ngy and then ngx. Each entry contains four
components (n, m_x, m_y, m_z) where n is the electron density, and (m_x,
m_y, m_z) is the spin polarization vector. The dos  bndstrct.dat file
contains as many lines as there are k-points and has besides the k-point
index as many columns as there are bands. It can be plotted with
'xmgrace -nxy bndstrct.dat' (or xmgr). The dos.dat file contains a
legend in the first line which should help. Besides the total and spin
up/down DOS it also contains projected partial density of states.

> 
> I will be grateful with all the information you can provide to me.

Hope this will help you getting started with Spinor. Let me know how
things are going. 

Take care,
Gerhard

P.S. I just want to point out that Spinor is under the GNU General
Public License. This means that you have full access to the sources and
you can modify and redistribute the sources (as long as it stays GPL).
The reason I say this is that I am currently not working much on Spinor
but if someone else would like to pick it up or even just use parts of
it it is perfectly fine. Especially the fact that it is written in ANSI
C should be attractive for new students. 

P.P.S. Jaime, I cc'ed this email onto the spinor-talk list so that other
people might benefit from this conversation as well. You can subscribe
via Spinor's web page on sourceforge.net

[Spinor-talk] Update: Spinor Project moved to Sourceforge.net...

[Gerhard J. T. <the...@sa...>]

Dear all,

This is to inform you that the official Spinor web page now has the
following new URL:

http://spinor.sourceforge.net/

The move to sourceforge provides some additional benefits:

1) We now have a mailing list called 'spinor-talk' which I encourage you
to sign up for to stay informed about the Spinor project. Instructions
are given on the 'Latest News ...' page.

2) The way you download files (sources and input files) has changed and
now is handled through the sourceforge file release system. Just follow
the link under 'Spinor @ Sourceforge.net'

3) Don't hesitate to sign up as a developer!


That's it. Please let me hear any feedback you might have,

Gerhard


P.S.: THIS IS THE LAST EMAIL YOU WILL RECEIVE ABOUT THE SPINOR PROJECT
UNLESS YOU SIGN UP FOR THE spinor-talk LIST AS MENTIONED ABOVE!

[Spinor-talk] test the spinor-talk list

[Gerhard J. T. <the...@sa...>]

test