[gmx-users] A special case of cutoffs related to mdrun with -rerun option

[Ioannis Beis]

Dear GROMACS users,

I am interested in using mdrun of a special version of GROMACS 4.0.2  
made by Ollila et. al with -rerun option, made so that it can allow  
calculations of properties related to local pressure in lipid bilayer  
systems. I have used GROMOS54a7 with Poger parameters, but among other  
slight modifications I have used PME instead of reaction field with  
rcoulomb=0.9 and maintained the twin range cutoff used in the original  
study for van der Waals, using rlist=0.9 (instead of the original 0.8)  
and rvdw=1.4. The inner sphere neighborlist is updated every step,  
whereas the outer sector neighborlist every 5 steps.

The mdrun with -rerun flag of this package does not support PME,  
therefore the tpr file has to be reassembled without it. Instead, much  
larger plain cutoff than 1.8 alone has been shown by Sonne et al. to  
give close enough results to short cutoffs used together with PME, at  
least in a qualitative manner.

I can thereby use rcoulomb=2.0 for the purposes of the -rerun, but  
this no longer allows the twin range cutoff scheme. I tried to check  
if there could be some trick to be done with rlistlong (although I was  
almost sure there wouldn't be anything and that rlistlong just serves  
for switch/shift functions), but as far as my searching is concerned  
it didn't even exist as an option in GROMACS 4.0.2 (I assume that twin  
range cutoff was still supported but rlistlong would be automatically  
placed in the max(rcoulomb,rvdw).

Therefore my current problem is about the reliability of -rerun with  
plain vdw cutoff, while in the original simulation twin range cutoff  
was used. Is there any study comparing the results of the two  
different schemes mentioned above for GROMOS simulations? What would  
the consequenses in the physics be and how would the choice of the  
cutoff radius and number of steps affect the results? If someone  
really wants to obtain those calculations, what would be the best  
compromise?

If anyone has another recommendation about obtaining lateral presure  
profiles and associated properties by lipid bilayer simulations, such  
that would be compatible with my algorithm choices, please let me know.

I am grateful in advance.

Best regards,
Ioannis