[CPMD-list] force values and constants with constraint & restraint

[Axel Kohlmeyer]

On Tue, 26 Feb 2008, Daria Khvostichenko wrote:

DK> Dear Axel,

dear daria,

DK> Thank you for your reply. I actually looked into the source code, and 
DK> found most of the answers I was looking for. Apparently, the 
DK> potential energy functions for various distortions were not defined 
DK> exactly in the way I thought they were, and, consequently, the 
DK> behavior of derivatives (forces) seemed erratic to me. For example, 

yep. happens to everybody once in a while. there is nothing that
beats a little experimentation and having a look at the source
code to make sure that you get what you think your get or to find
out why you don't get it... ;-)

[...]

DK> than simply through angles. It would be great if the manual listed 
DK> equations for restraints rather than text descriptions. Anyway, they 
DK> were relatively easy to locate in the source code, which, thankfully, 
DK> did have these equations explicitly written in the comments.

since you already looked at it, how about you take the manual.tex
file, add what you found out or change it the way it would be 
more obvious and send the changes to me or cpmd at cpmd.org for 
inclusion into the next release. 

[...]

DK> Thanks for the warning about the improper behavior of restraints in 
DK> certain types of calculations. I've also found earlier posts about 
DK> double-counting of restraints in BOMD calculations, and about missing 
DK> conversion of units for some restraints.

the conversions should be all done by now. please let me know
if you encounter a restraint where parameters are not properly
converted from angstroms to atomic units. in any case, using 
atomic units throughout and not using the ANGSTROM flag is 
always the safe option.

DK> As for collection of restraint energies, at least in geometry 
DK> optimizations those energies do not seem to be added to the total 
DK> energy of the system, and there doesn't seem to be a separate entry 
DK> for the restraint energy in the output file.

yep, i'm working on that and related issues right now. so hopefully
in the next release this will be sorted out.

DK> You mentioned that the worst case for restraints is geometry 
DK> optimization with restraints in absence of constraints. Why is that?

constraint energy and and restraint energy are added into the 
same variable before the total energy is computed. this variable
is set to zero when using a constraint (and occasionally at different
places in the "drivers") and then restraints are added to it.
when using geometry optimization or BO-MD (with default wavefunction
optimization and not LANCZOS or DAVIDSON) then the restraints are
added in two places. so your energies can be very wrong and the
forces doubled. with plain CP-MD it should be more or less ok
(the total energy may be a little off or drifting).

the whole thing is somewhat interesting from an "archeological 
perspective" because you can almost reconstruct how the CPMD code has 
evolved over the years and how functionality was added by copying and 
modifying files as well as splitting files (and subroutines) into 
multiple parts when they became too large or too convoluted. ;-)

cheers,
   axel.

DK> Regards,
DK> 
DK> Daria
DK> 
DK> 
DK> _______________________________________________
DK> CPMD-list mailing list
DK> CPMD-list at cpmd.org
DK> http://cpmd.org/mailman/listinfo/cpmd-list
DK> 

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