[gmx-users] Problem with Rescaling of the COM

[kai.exner at alumni.uni-ulm.de]

Dear Collagues,

We are having a question relating to our MD simulations, in which we  
connect a gold nanoparticle (Au NP) with a drug-binding peptide and  
the corresponding drug molecule. The peptide is bound via a cysteine  
residue to the Au NP and here the corresponding bonded interactions,  
angles and dihedrals between Au and S have been assigned in the  
ffbonded.itp file. In the same way, we proceeded with the non-bonded  
interactions between the Au atoms in the ffnonbonded.itp file. When we  
started to heat our system, the Au NP was disintegrating, i.e. the  
non-bonded interactions between the Au atoms are not sufficient to  
keep the NP together. Therefore, we considered position restraints for  
all Au atoms that we maintained during heating, relaxation and  
production phase (NPT). Since pressure coupling with absolute position  
restraints requires refcoord-scaling, we applied the  
refcoord-scaling=com option in the mdp-file.
The simulations performed are stable, run without any further problems  
relating to disintegration of the Au NP and we obtain reasonable  
results. However, there is one problem, namely as soon as we want to  
extract information from the production phase relating to the Au atoms  
between two subsequent MD runs (simply extending the time under the  
same conditions in the second MD run), we observe an artificial jump  
in certain quantities, for instance when calculating the RMSD, the  
radius of gyration or the solvent accessible surface area with respect  
to the Au atoms with position restraints. As long as we calculate  
these properties in a single MD run, these quantities are stable (of  
course fluctuating, but without any artificial jumps).
We ascribe the observed artificial jump that is always occurring at  
the very beginning of the extended MD run to the rescaling of the com.  
We tried different approaches to get rid of this physically not  
meaningful jump by combining the trajectories of subsequent MD runs  
(gmx trjcat) and then calculate these properties with respect to the  
Au atoms or by calculating the center of mass (com) for the Au atoms  
in the NP for both simulations and then translating the ?new com?  
(first frame in the extended MD simulations) to the ?old com? (last  
frame in the first MD run). Both approaches were not successful, still  
the artificial jump is occurring. We also tried to adjust the  
reference (for ?-s? in the commands we applied various tpr- and  
gro-files), but this is also not the solution of the problem.
Does anyone have had the same or a similar problem or even better does  
someone know how to get rid of this issue without virtually shifting  
the output data of the subsequent MD run towards the output data of  
the first MD run?
Any help is appreciated. Thanks a lot for your efforts in advance.
With my best regards,
Kai

P.S.: We also outlined that the observed jump is indeed artificial,  
namely by taking an extended MD simulation and running it from the  
original starting point directly for the double time. In that case, no  
artificial jump occurs.