[gmx-users] Normal Mode Analysis in Double-Precision

[Ran Friedman]

Inon Sharony wrote:
>
>  Hi Ran, thanks for the reply
>
> I ran "make tests" after compiling with double precision, and it came
> out fine.
>
> Could it be that this, second, compilation might have caused some
> problems -- I had GROMACS compiled in single-precision, and it worked
> fine -- could compiling in double-precision after the initial
> single-precision overrun some files? I'm not sure anymore...
>
> Also, the grompp in double-precision works fine (no warnings, etc.)
> but I get the segfault from mdrun (also in double-precision).
> Sometimes this happens in the first minimization step, and sometimes
> in the second. I was told that this might hint that the input file/s
> (e.g. the *.gro file) were input in single- and not double-precision.
> However, since the files have coordinates written in nanometers to
> four significant digits, I can't see how the input numbers could be in
> single or double precision. It was suggested that I insert a "print to
> screen" line in the source code immediately after the *.gro file is
> read, to see if the printed data is in single-precision or
> double-precision format. What do you think?
>
The tpr should contain the data in double precision. You can check this
with gmxdump.

Other things that you may want to consider:

1. Do you get step*.pdb files after running mdrun? This may point out to
something wrong in the input.
2. Are you using a solvated system? Normally NMA is done in vacuo, maybe
with some scaling of the dielectric.
3. Did you examine the system that you use in the input? Does it make
sense to you?

Ran.