[CPMD-list] monoclinic cell

[Axel Kohlmeyer]

On Tue, 10 Jul 2007, Samir H. Mushrif wrote:

SM> Hello Axel,

hello samir,

SM>             Thanks for your reply. I also got a reply from Juerg Hutter 
SM> which states that
SM> **************************
SM> please note that
SM> POINT GROUP MOLECULE
SM> 
SM> only works for single (isolated) molecules and needs
SM> a special standard orientation of the molecule.
SM> It should not be used for molecular crystals.
SM> **************************

since juerg is the 'one-who-knows' and i am just trying
to get along, you should trust his advice more than mine.

[...]

SM> Though I did not get the point group symmetry, the next part of the 
SM> output shows POINT GROUP as 2/m(c2h) and that is what we are aiming to 
SM> get. I am not clear exactly what is happening here though.

as juerg and i tried to point out, that main reason for having the 
point group symmetry support is to reduce the computational effort
for (symmetric) isolated molecules, e.g. when optimizing the geometry
or even more so for computing vibrational spectra. as juerg wrote, 
for that to work, the molecule has to be oriented in a specific way,
which you have no to option to do with a crystal structure.

SM> Your other suggestion was:
SM> 
SM> turn off symmetry, set MAXSTEP to 1 and add RHOOUT.
SM> this should give you the coordinates and the density
SM> of the initial guess in a DENSITY file. do you really
SM> want to enforce that symmetry? most calculations will run fine without.
SM> 
SM> Following your suggestion, I did the calculations by turning off the 
SM> symmetry and then looked at the structure with multiple cells, using 
SM> VMD. What I observed was that the molecules looked same in both the 
SM> cases and there was no difference in the arrangement of atoms (and 
SM> molecules) in both the cases due to POINT GROUP SYMMETRY.

they should look the same. the question is, does the crystal lattice
look correct, i.e. as you expect it? you can also add an isosurface
representation in VMD to check, whether the atom positions match
the electron density maxima.

SM> My aim here is to do geometry and cell optimization (using TSDC option 
SM> so as to keep the same shape). I would like to know if having a POINT 
SM> GROUP SYMMETRY will have some difference in the results. The geometry 

for the geometry optimization the point group symmetry could have 
some effect, as it turns on symmetrization of the forces, which may
not be that symmetric if your input is not perfectly symmetric or
match the fft grid.

optimizing the cell is a totally different issue, since variable
cell calculations with plane wave basis sets are subject to pulay
stress. your effective basis set size will change with the cell
volume, since you are not exactly running a constant cutoff calculation,
but a constant number of plane wave calculation, which in turn
will translate into an increase of the effective cutoff with a smaller 
cell. to avoid bogus results, you have to converge the stress tensor
with respect to the plane wave cutoff significantly beyond the minimum,
and the stress tensor itself converges much slower than the forces.

in many, many cases just running a series of single point calculations
(with a 'normal' cutoff) with different lattice constant and then 
fitting the resulting energies to an equation of state is an easier 
way to find the optimal lattice constant. this topic is also frequently
discussed on the quantum espresso (formerly PWScf) mailing list.
you may want to want search their archives, since that code also uses
plane waves with pseudopotentials and thus has the same issues.
generally, variable cell calculations for both codes are most useful
for looking at, e.g. high-pressure induced structural changes etc.

if i remember correctly, there was somebody noting a while ago,
that the ISOTROPIC CELL option is currently not working.

one more thing to consider: from your explanation it looks as if 
you have a crystal with individual molecules, i.e. the major
interaction between molecules and thus between the cell images
would be dispersion interactions, which are, unlike bonds, not
so well represented with current DFT functionals and people have 
tried to find several ways to compensate for it (systematically
and empirically).

SM> of the molecule that I have put in the cell is already optimized as a 
SM> single isolated molecule and I am expecting small change in the 
SM> geometry after putting it in the cell with periodic boundary 
SM> conditions. With this in mind I don't know if the inclusion of POINT 
SM> GROUP SYMMETRY should cause any difference in my final results. May I 
SM> know your opinion on this?

in most cases, i would expect the forces to be symmetric enough
when the input structure is symmetric, too. however, since CPMD
evaluates functionals on a grid and also computes forces from 
gridded data, the tightness of that grid, governed by wavefunction
and density cutoff, may have some impact (ripples effect).

since what you plan to do, has not been done a lot with CPMD,
(most people i know do CP-dynamics MD calculations in NVT or
NVE ensemble), it is advisable to be extra careful and approach 
your final goal in steps where you can double check the validity 
of each step before you start the following, more expensive 
calculations. since you have been very careful until now, i 
don't expect that this will be big a problem. :)

cheers,
   axel.


SM> 
SM> Thanks a ton!
SM> 
SM> Best Regards
SM> Samir
SM> 

[...]

-- 
=======================================================================
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.