[gmx-users] Domain decomposition error while running coarse grained simulations on cluster

[Avijeet Kulshrestha]

Hi all,
I am running martini coarse-grained simulation with 15 fs of time step in
gromacs 2018.6. I have 25859 number of atoms and my box size is:
12.00000  14.00000  18.00000
Where I have Protein, membrane (DPPC) and ions.
I have minimized energy with 16 processor and -rdd option as 2.5. It worked
fine but later in NVT simulation, this started giving errors.
*I am getting the following error without -rdd option when I used 1 node
and 8 processor per node:*
Fatal error:
There is no domain decomposition for 8 ranks that is compatible with the
given
box and a minimum cell size of 7.69783 nm
Change the number of ranks or mdrun option -rdd or -dds

*--> *I tried using several other numbers of processors but it didn't work.
It works fine with single processor per node.

I have used -rdd  2.5 with 16 processors then simulation runs but at some
point, it gives me the different kind of error,
Software inconsistency error:
Some interactions seem to be assigned multiple times
In few other simulations, I also get this kind of errors when using -rdd
option:
Fatal error:
30 of the 1378 bonded interactions could not be calculated because some
atoms involved moved further apart than the multi-body cut-off distance
(2.5 nm) or the two-body cut-off distance (2.5 nm), see option -rdd, for
pairs and tabulated bonds also see option -ddcheck

*Can someone tell me the proper protocol to find an optimum number of
processor required?*
*How to decide box size?*

*If -rdd option is required then what value should I choose?*


*I kept on getting these kinds of errors and I don't want trial and error
method so,*
*Can you suggest me some reading to get technical insight on this?*