[Pw_forum] Projection, convergence and imaginary phonon frequencies

Axel Kohlmeyer akohlmey at cmm.chem.upenn.edu
Mon Jan 12 18:27:03 CET 2009


On Mon, 12 Jan 2009, John Kendrick wrote:

JK> Hi,

hi,

JK> Yes Happy New Year.
JK> 
JK> I only gave the first few soft modes for each case.  It is these soft 
JK> modes I am interested in.  There are 174 modes (58 atoms) altogether.  I 

hmmmm... since the topic of treating dispersion interactions came up
the other day, this would be a case where it may be worth investigating
how much a dispersion correction, even an empirical one, would impact 
those modes. i know of somebody who did this successfully on an alanine 
crystal with some other plane wave pseudopotential code (i can give you 
references, if needed).

JK> listed 13 in one case as there was more of a match with the intensities 
JK> from the other projection methods.
JK> 
JK> The comment about cutoff is interesting as I had found that the 
JK> optimiser (BFGS) was having problems.  This would also be consistent 
JK> with  getting more consistent results with the USP potentials.

depending on the settings. with ultra-soft psps it is important
that you have a high enough density cutoff. e.g. with a wfc cutoff
at 30ry, you may need 300ry for rho. if you had been mix-and-matching
potentials or kept the 80ry, then you would have been in the right
ball park (80ry*4 = 320ry which would be the default rho cutoff).

cheers,
   axel.

JK> 
JK> Thanks
JK> 
JK> John
JK> 
JK> Axel Kohlmeyer wrote:
JK> > On Mon, 12 Jan 2009, Nicola Marzari wrote:
JK> > 
JK> > NM> Dear John,
JK> > 
JK> > dear nicola,
JK> > 
JK> > happy new year from cold philly.
JK> > 
JK> > NM> if it is a crystal, the most critical parameter should be k-point
JK> > NM> sampling. Increase that, maybe using shifted meshes of the form
JK> > NM> 2n 2n 3n 1 1 1 , with n=1,2,3...
JK> > 
JK> > but it is a _molecular_ crystal, so i guess there are some 
JK> > areas with comparatively low electron density and quite weak 
JK> > interactions, as seen by the long list of rather soft modes.
JK> > i would suspect that will need to be converged very well wrt 
JK> > to the basis set to not see any ripples on the potential surface.
JK> > 
JK> > for TM oxygen pseudos i would go to 120ry or higher to get
JK> > tight convergence. for MD 80-85ry is usually sufficient 
JK> > (the more traditional 70ry has been shown to be very sloppy),
JK> > but if you want a tight geometry convergence you will have
JK> > to be more careful. BFGS can be quite sensitive and will 
JK> > essentially refuse to converge if you have "ripples" due
JK> > to a very flat potential surface unless you crank up the
JK> > cutoff significantly.
JK> > 
JK> > NM> Norm-conserving and ultrasoft should matter little (unless
JK> > NM> there is a problem with one pseudo), while GGAs should be a bit
JK> > NM> tougher to converge than LDA.
JK> > NM> 
JK> > NM> Do "crystal" and "simple" really lead to a different number of
JK> > NM> modes (10 vs 13) ??
JK> > 
JK> > those are obviously only the lowest few modes. 
JK> > it is a 9x9x6 angstrom cell!
JK> > 
JK> > cheers,
JK> >    axel.
JK> > 
JK> > NM> 
JK> > NM> 			nicola
JK> > NM> 
JK> > NM> 
JK> > NM> John Kendrick wrote:
JK> > NM> > Hi,
JK> > NM> > 
JK> > NM> > I have been doing some calculations recently on molecular crystals. 
JK> > NM> > Because of our interest in the lowest phonon frequencies I have been 
JK> > NM> > concerned that translational invariance is maintained as far as 
JK> > NM> > possible.  To this end I have increased the accuracy of the optimisation 
JK> > NM> > and the scf convergence, to a point where I cannot see how to converge 
JK> > NM> > the calculation any tighter.  I noticed that the norm conserving 
JK> > NM> > pseudopotentials were having greater difficulties than the others.  In 
JK> > NM> > particular for one molecule I have seen significant differences between 
JK> > NM> > the different projection schemes that ensure translational invariance. 
JK> > NM> > Particularly the intensities in the infrared a very different.  Below is 
JK> > NM> > some output from a representative calculation.
JK> > NM> 
JK> > NM> ---------------------------------------------------------------------
JK> > NM> Prof Nicola Marzari   Department of Materials Science and Engineering
JK> > NM> 13-5066   MIT   77 Massachusetts Avenue   Cambridge MA 02139-4307 USA
JK> > NM> tel 617.4522758 fax 2586534 marzari at mit.edu http://quasiamore.mit.edu
JK> > NM> _______________________________________________
JK> > NM> Pw_forum mailing list
JK> > NM> Pw_forum at pwscf.org
JK> > NM> http://www.democritos.it/mailman/listinfo/pw_forum
JK> > NM> 
JK> > 
JK> _______________________________________________
JK> Pw_forum mailing list
JK> Pw_forum at pwscf.org
JK> http://www.democritos.it/mailman/listinfo/pw_forum
JK> 

-- 
=======================================================================
Axel Kohlmeyer   akohlmey at cmm.chem.upenn.edu   http://www.cmm.upenn.edu
   Center for Molecular Modeling   --   University of Pennsylvania
Department of Chemistry, 231 S.34th Street, Philadelphia, PA 19104-6323
tel: 1-215-898-1582,  fax: 1-215-573-6233,  office-tel: 1-215-898-5425
=======================================================================
If you make something idiot-proof, the universe creates a better idiot.



More information about the users mailing list