[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: FEFF: Broadening



Hi,

Thanks for all the advice, including the corrections broadening term makes my simulated ELNES peaks resemble experimental data more accurately.

Alex, when l talk about convergence i do not mean the scf calculation. My mistake, sorry! As you know it is possible to alter the radius of the cluster of atoms. I have calculated the ELNES for increasing numbers of complete shells of atoms to see what the effect will be. I would normally expect the shape of the ELNES to 'converge' or keep constant when enough atoms are included in the calculation. This does not happen with BaTiO3, even for very big calculations. (Over 550atoms, out to R~11A) Basically l was trying to find out if there was a way to manipulate my calculations in order to accurately simulate a bulk environment around the atom under excitation and get the ELNES to stop changing.

Best Wishes,
Paula



Hi feffusers,

> I am investigating the system, BaTiO3. Specifically l am looking at the
> elnes generated by the xanes card. I would like to introduce broadening
> into my calculation as it does not seem to converge, even for very big
> clusters. If l increase Eimag for LDOS , (module two) will this be carried
  Do you mean that self-consistency loop does not converge?
  If that is the case, then use SCF  4.0   card. In may experience
  this typically leads to good potential, while rarely has problem
  with convergence.  Of course for LDOS and XANES(ELNES) calculation
  you might want to use larger radius (e.g. FMS 8.0)
> through to the xanes calculation or is there another way to introduce
> broadening?
  No. The Im part in LDOS card is not carried over to XANES calculations.
  Use EXCHANGE card to do that (EXCHANGE  0  0  1.0 to add 1 eV Lorentzian
 braoadening). This is the best way to introduce broadening, but
 you have to rerun starting from the second module. Bruce Ravel
suggested to use CORRECTION  0 1.0. This is faster, since it also makes
Lorentzian broadening but require to rerun only the last module (ff2chi
or CINTROL  0 0 0 0 0 1 ).
> Also, if this is the correct way to introduce broadening and get the
> calculation to converge with smaller clusters, then what would be an
> appropriate value for Eimag?
  What convergence are we talking about? SCF loop or convergence
with cluster size?

Matt Newville pointed out that convolution with Voigt function
might be prefereable. If someone has  tested FORTRAN routines for
such convolution, I can easily include them. For example,
we can use (CORRECTION 0  1.0  type_conv ), where type_conv=0
for Lorentsian, 1- for Gaussian, 2-for Voigt, etc..

Best wishes
Alex Ankudinov