[Pw_forum] the procedure of calculating the Raman tensor using the finite electric field method

[Serge Nakhmanson]

li niu wrote:
> Dear all.
> Thanks for your advice!
> I'm sure that the positions from CPMD are correctly passed to the scf 
> code because
> I have checked the input geometry using xcrysden. The atomic positions 
> of *.xyz file from
> CPMD are the same with positions in the case of xcysden-display-unit of 
> repetition-translational
> asummetric unit.

 > Why are the forces and stress so big?

As you mentioned yourself, CPMD and PW use different functionals. The
minimum-energy structural configurations that these functionals find
for your model will not be the same. However, they should be reasonably
close to each other if no mistakes were made during the CPMD quench.
It is up to you to decide what is "reasonable," but IMHO the stress
on your cell is a bit large. This might suggest that the CPMD quench
has not been done carefully enough, and your PW calculation shows it.
But my previous comment was that, in general, it is quite hard to
converge to a well relaxed configuration with the "quench-from-the-melt"
procedure (this requires a long time and careful control of temperature),
and it does not matter what particular approach (CPMD, PW, TB, etc) you
are using.


> And the microstructures,electric structures and optical properties of 
> this semiconductor model
> have been analyed and published. 

And this is not a proof that the model that you are using is an
accurate one (people can publish anything they want these days).
In any case, PLZ provide a reference.

> I should use vc-relax (pw.x code) to compute the equilibrium geometry
> before running finite electric field method to calculate Raman tensor. 
> Is it right?

And how can you compute phonons if your system is not at equilibrium?
However, you have plenty of problems (as indicated in your previous
message) getting to equilibrium. It is bad (but well expected) that
your cell becomes non-cubic during stress relaxation. You should
either force it to stay cubic by manipulating the stress-tensor
components or work at some reasonable fixed volume.

Then, since you will re-relax this model from its minimum-energy
configuration found with CPMD, its structural characteristics will
change. So, before you do anything else with it, you will need to
check that you model is still structurally sound. This includes
computing RDF (should match experiment), bond/dihedral angle
distributions (the former should be narrow enough), EDOS (should
have a gap) and VDOS (should also match experiment). Atom coordination
analysis is a plus too, since a good model should be 99.9% (or
whatever it is experimentally for ta-C) 4-fold coordinated. Only after
you convince yourself that the properties of your model are in reasonable
agreement with experiment, you can proceed and compute the Raman tensor
for it, so good luck with all that.

In general, creating models for amorphous tetrahedral solids is
an *art* and the level of questions that you are asking shows that you
do not understand how this process works. There are many well-known
(both correct and incorrect) papers on that and you should read them
before you start blindly burning your CPU time. Read the WWW PRL
paper I mentioned in my previous e-mail and then read some more
recent papers that are quoting it. Then do the same with Car-Parrinello
PRL (1988).


-- 
*********************************************************
  Serge M. Nakhmanson               phone: (630) 252-5205
  Assistant Scientist                 fax: (630) 252-4798
  MSD-212, Rm. C-224
  Argonne National Laboratory
  9700 S. Cass Ave.
  Argonne, IL 60439
*********************************************************