Applying freezing and restrains to speed up MD simulation

User discussions
, ,
1

GROMACS version:
GROMACS modification: Yes/No
Here post your question

Dear GROMACS community

I am using MD simulation to simulate a system of 12 chains that has ~700 K atoms including solvent molecules and ions. Since the size of the system is huge, the simulation will be computationally expensive. On the other hand, my focus is only on a small region of the system, particularly, on a small region of 3 chains out of the 12 chains. To save computational time and resources, I would like to divide the system into three regions:

§ Region A - a small region that I am focused on, where atoms will be allowed to move “ freely “ during the simulation.

§ Region B - a region that is surrounding the small region A (within a certain distance) where atoms will be “ restrained ” during the simulation.

§ Region C - a region that represents the rest of the system outside region A and B where atoms will be “ frozen ” during the simulation.

My questions are:

  1.   How do I apply those restrains and freezing on the regions described above to the topology file?

  2.   If I can do so, would that impact the reliability of the results of the MD simulation? FYI, we believe that the atoms outside the small region of interest are not impacting the functionality of the region of interest.

  3.   Would that save time on the simulation since we are not calculating any forces on the frozen atoms in region C?

Thank you so much (^_^)
Amnah

2

In older versions of GROMACS this setup might have saved you a bit of time, but with the Verlet cut-off scheme we no longer support the energy group exclusions feature that made it possible to skip interactions. And even with that, one would still need to compute the PME-mesh contribution of all charges, so the gain would be limited.

3

So if we use the older version (is it 2018 or 2019?) and define freeze groups along with energy group exclusions, what else would you recommend to speed up the simulation? Is there a better cut-off scheme we should use other than Verlet? And is there a way to avoid computing the PME-mesh contribution of all charges?