[gmx-users] loss of precision in *.gro

Mehmet Suezen suzen at theochem.tu-muenchen.de
Wed Sep 11 09:01:01 CEST 2002


As far as you compiled gromacs in double precision, your trajectory 
records are double.(for example in trr). Basicly your analysis tools
(trajectory averages) worked without loss of precision. Anyway for 
conventional classical molecular dynamics f8.3 starting structure is not
that bad.(if one assume equilibrium structure will reached shortly)
Do you think classical MD can predict better then 0.01 A?
But if one goes to quantum mechanical regime, it may affect. 
Or single structure analysis would absolutely affected as well. 
As you said, cutting out from trr would be a solution in your case.
I'd guess Idea of optimized geometry is to obtaine plausable starting 
structure for QM input,As you said, cutting out from trr would be a
in this case.


sekirin at iop.kiev.ua wrote:
>      Dear colleagues,
>      Could you please explain me, why the others do not feel the
> following problem and how it could be treated:
>      After optimization of geometry, we easily obtain (e.g., with
> emstep=0.001, emtol=1.0) maximal force ~ 4.5e-1 . However,
> the coordinates (in nm) are written into the file confout.gro with
> the use of format f8.3 . As a result, precision is only ca. 0.01
> Angstrom ,
> and we have drastic loss of precision. Naturally, at consequent
> normal mode analysis the program reports that the maximal force
> is ~ 5.1e2, so the results can be inadequate, etc.
>      One can easily demonstrate (by considering diatomic molecule
> with variable equilibrium interatomic distance) that just the round-up
> is the cause of this effect.
>      Evidently this effect _can not_ be avoided by the use of double
> precision (as far as format f8.3 is a part of specification of *.gro).
> Probably one could surgically cut the coordinates from *.trr (or use
> user's own additional output) and insert them into *.tpr, but it would
> be
> desirabre to have some documented method.
>      Thank you              I.V.Sekirin
> _______________________________________________
> gmx-users mailing list
> gmx-users at gromacs.org
> http://www.gromacs.org/mailman/listinfo/gmx-users
> Please don't post (un)subscribe requests to the list. Use the
> www interface or send it to gmx-users-request at gromacs.org.

More information about the gromacs.org_gmx-users mailing list