[gmx-users] problems with double precision gmx

[Michael Brunsteiner]

Hi everybody,

I keep having difficulties when trying to use the double 
precision version of Gromacs. 
Typically i get infinte forces at step one of a simulation 
with double precision while the same system behaves perfectly 
well with single precision.

Below i give an example of what i see when i try to minimize 
a short peptide in vacuum, the input files (pdb and mdp)
are attached.

I wonder if anybody else sees similar problems and has possibly
found a solution.

thanks in advance for any help or suggestions!
Michael


example:

I start with the speptide.pdb file from the gmx tutorial,
and proceed as follows...
(details about hardware etc below)

1) include as first line:
"CRYST1   50.000   50.000   50.000  90.00  90.00  90.00 P 1 "
into speptide.pdb.

2) make topology (use GROMOS96 43a1 force field)

echo "0" | pdb2gmx -f speptide.pdb

3) use grompp and mdrun, single precision to minimize the
   short peptide in vacuum for 100 steps.

  grompp   -f em.mdp -c conf.gro -p topol.top -o s.tpr

  mdrun -v -s s.tpr
  gives:

Step=    0, Dmax= 1.0e-02 nm, Epot=  2.94552e+02 Fmax= 6.05693e+03, atom= 120
Step=    1, Dmax= 1.0e-02 nm, Epot=  5.33718e+01 Fmax= 5.18432e+03, atom= 83
Step=    3, Dmax= 6.0e-03 nm, Epot= -1.56387e+02 Fmax= 2.95143e+03, atom= 120
Step=    5, Dmax= 3.6e-03 nm, Epot= -2.14290e+02 Fmax= 1.68914e+03, atom= 81
Step=    7, Dmax= 2.2e-03 nm, Epot= -2.49216e+02 Fmax= 1.55114e+03, atom= 120
etc...

  simulation finishes and gives expected result.

5) use grompp_d and mdrun_d, double precision.

   grompp_d -f em.mdp -c conf.gro -p topol.top -o d.tpr   

   mdrun_d -v -s d.tpr
   gives:

Step=    0, Dmax= 1.0e-02 nm, Epot= -2.18566e+03 Fmax=         inf, atom= 1
Step=   34, Dmax= 1.2e-12 nm, Epot= -2.18566e+03 Fmax=         inf, atom= 1
Stepsize too small, or no change in energy.
Converged to machine precision,
but not to the requested precision Fmax < 200

  ... simulation finishes and the coordinates in the confout,gro
  are EXACTLY the same as in the input (apart from the fact
  that the molecule is shifted by a boxlength, because the first
  atom of the input structure is outside the box)


more details:

*) the gmx double precision version was compiled exactly as
   suggested in the INSTALL file.

*) my soft/hardware: gmx-3.3.2 on a PC, AMD Turion 64, 
   linux-2.6.23, gcc-4.2.3
   ... though i see the same behaviour with the current
   (todays) cvs version of gmx on the same hardware,
   and with gmx-3.3.1 on a 32-bit PC

*) i see similar problems with other systems, i.e., in many
   (but not all) cases single precision works, double gives
   problems...




      ____________________________________________________________________________________
Never miss a thing.  Make Yahoo your home page. 
http://www.yahoo.com/r/hs
-------------- next part --------------
A non-text attachment was scrubbed...
Name: em.mdp
Type: application/octet-stream
Size: 385 bytes
Desc: 462865951-em.mdp
URL: <http://maillist.sys.kth.se/pipermail/gromacs.org_gmx-users/attachments/20080122/1b73e6a1/attachment.obj>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: speptide.pdb
Type: x-unknown/x-pdb
Size: 11890 bytes
Desc: 3672131110-speptide.pdb
URL: <http://maillist.sys.kth.se/pipermail/gromacs.org_gmx-users/attachments/20080122/1b73e6a1/attachment.bin>