[gmx-users] mdrun has different results with similar files

[Mark Abraham]

Mathieu, Jennifer J. wrote:
> I've searched through your database and found many helpful suggestions, 
> but I am not sure why mdrun successfully runs on some files but not 
> others.  The protein files are pretty but I get the following results 
> when running mdrun.  (Yesterday I tried attaching the files but they 
> made the email too big and it got returned)
>  
> ABPa1OUT.pdb fails (Number of grid cells is zero. Probably the system 
> and box collapsed.)
> QBPq1OUT.pdb succeeds with errors (some bonds rotated more than 30 degrees)
> GBPg1OUT.pdb succeeds
>  
> Also, the topology files for ABP and GBP are made using 1 file, while 
> the QBP topology file is made using 2 files.  (when 2 files are used, I 
> edit the topol.top file to have the posre_A.itp and posre_B.itp files -- 
> #include "posre_A.itp")
>  
> Here are the commands I'm running:
> pdb2gmx -f file_name.pdb -o -p < input.txt
> editconf -bt octahedron -f -o -c -d 0.7
> genbox -cp conf.gro -cs -o -p
> grompp -f -c out.gro -p -o
> mdrun -s -o -c -g -e

This looks like a classic question for this mailing list. MD simulations 
will not "just work" because there are many things that have to be just 
right and it is prohibitively difficult to ensure that for infinitely 
many possible starting structures. If you have a look at some 
literature, you will see that nobody reports a strategy like the above. 
The reasons are varied. Often, either before or after you solvate a 
system, there are atoms that are too close together or far apart to 
resemble a real physical system. These need to be relaxed with an energy 
minimization stage, or you risk having the dynamics fail because things 
are too rough. Next, the velocities that are generated by the solvation 
routine aren't magically going to be realistic all of the time. The 
generation is a stochastic process and simply won't generate a coherent 
self-consistent ensemble. That only happens after you allow the system 
to equilibrate for a while. You should check the progress and results of 
these processes visually - gromacs comes with a simple visualization 
routine n_gmx, and I also recommend VMD. Only once you think it's 
equilibrated & thus its behaviour might be similar to a real physical 
system, should you start "observing" the system.

Mark