simulating vibrational spectra in the solid state, help me!

Florian Schiffmann flosch... at gmail.com
Wed May 15 12:16:54 UTC 2013


Hi Daria,

I attached the NLCC pseudos. As said before they are only optimized for PBE 
and there is no warranty that they work for anything else (e.g. PBEsol, 
PBE0,...). 
Previously I forgot to mention two other option to overcome that problem:
1) Use SMOOTING (splines) in the XC_GRID section. Personally I wouldn't 
recommend that option. It solves the riplle problem but has a hugh effect 
on the total energy and converges slower to the PW basis set limit. 
Furthermore I am not 100% whether the error due to smoothing is systematic 
or not.
2) Use the USE_FINER_GRID option in XC_GRID. This will be expensive for 
calculations dominated by FFT anyway. This uses a grid with 4*CUTOFF for 
the xc part. I am about to commit the option to specify the finer grid by 
the user but it isn't quite ready. With this option (PBE case) you should 
get away with a CUTOFF of 400.

In my oppinion the best options are NLCC's or the finer grid. NLCC's have 
the problem that they are hard to fit and there is no hope that there will 
be an extended set in the forseeable future. Therefore the FINER GRID 
provides an option for all calculations and has several advatages to the 
other options:
1) It really improves accuracy unlike smoothing
2) forces and 2nd derivatives are better compared to an increase of the 
overall cutoff. This can be understood as the energy ripples are mainly due 
to the xc bit and will be accurate to the limit of the Finer grid. The 
potential neverthelees gets mapped back on the total grid and thus the 
ripples appear on a larger grid spacing than in case of high total cutoffs. 
If you take now the numerical derivative of the grid 'ripples' the larger 
spacing significantly reduces the force contribution.

Hope that helps
Flo
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20130515/f06d2790/attachment.htm>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: NLCC.dat
Type: application/octet-stream
Size: 2684 bytes
Desc: not available
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20130515/f06d2790/attachment.obj>


More information about the CP2K-user mailing list