Question about atom types

Hi to all,

I am planning to run an NPT simulation of a hydrocarbon system. For certain reasons related to Lammps restrictions, and the fact that I have to exclude several 1-5, 1-6 pairwise interactions, I have to use a separate atom type for each atom.
The system consists of 24000 atoms (8000 carbon atoms and 16000 hydrogen atoms), i.e., 24000 atom types in total. My system representation makes easy the production of a data file through several do-loops.

The forcefield includes two types of bonded interactions, three types of angle interactions, one type of dihedral interaction and Lennard-Jones pairwise interactions (with tail correction). The pair coefficients will be given for all possible combinations.
Crude way, but it’s OK for the one or two testing simulations I am interested.

In another comment I read that Lammps is able to deal with 2 billions atom types but memory or speed problems can exist, so I wonder if finally I would obtain results in a reasonable time.

Is there any idea about the following :

(a) How much memory (approximately) is required?

(b) How much slower will be the executable (approximately, for example 10/100 times) from the one that would use just two atom types, with the same force field.

© Should I expect any parallelization problems?

Thank you very much.

LAMMPS stores some 2d arrays that are Ntype by Ntype,
so that’s 24000^2, on every proc, which is a huge amount of memory.

I imagine it will run slower if it runs at all. You can try it and see.

I suggest thinking of another way to represent your model.


1 Like