[gmx-users] Relaxation / precision

[Andrew Horsfield]

> >(I cannot get a double precision code to work with PME, which is why I
> >have not tested this out.)
> Why not?

1) Version 3.0.5 drops out with the error

   Conjugate gradients:
      Tolerance         =  5.00000e+02
      Starting Energy   = -8.666447784387536e+04
      F-Norm            =          nan
 
   Fatal error: ci = -2147483648 should be in 0 .. 1715 [FILE nsgrid.c, LINE 214]

2) I was kindly sent a copy of the development code by Gerrit Groenhof, 
   and this gave a new error:

   Fatal error: Routine gmx_sumd called in network.c, 331

   Looking at the source code I found that the relevant logic was:

   congrad   calls  f_norm
   f_norm    calls  gmx_sumd
   gmx_sumd  fails  because I am not using MPI:
 
    #ifndef USE_MPI
      MYFATAL("gmx_sumd");
    #else

3) I have MPICH, so I reconfigured with --enable-mpi, and recompiled. I
   get another strange error:

   Fatal error: relax.tpr.tpb does not exist
   [0] MPI Abort by user Aborting program !
   [0] Aborting program!

   I am not sure what causes this error. (The mpich tests work.)

Cheers,

Andrew