GROMACS version: 2024
GROMACS modification: Yes/No
Hello all
I am trying to run a free energy calculation procedure changing only the vdw-lambda values. All the rest of the lambdas are constant at zero.
For vdw-lambda values of 0 0,05 and 0.1 the simulation runs with no problem.
However, when I got to lambda value of 0.15 the simulation crashes immediately when I try to run in on multiple nodes claiming that “There is no compatible domain decomposition for 98 ranks that is compatible with the given box and a minimum cell size of 2.7426 nm).”
When I change the integrator from sd to md the simulation goes through with no issues.
Obviously, the Langevin dynamics is more appropriate, so does anyone knows what could be the problem and how to solve it?
I can’t think of any specific reason why the sd integrator would crash like this. Could you post the mdp files - the one with the sd integrator and the one with the md integrator?
Hi
Thank you for the response attached is the MDP file, and I only change the type of integrator from sd to md.
Since then, even with the md the system crashes (though much later than with the sd integrator) and the error is
"
WARNING: Listed nonbonded interaction between particles 27447 and 27448
at distance 17288.656 which is larger than the table limit 11.600 nm."
which indicates that the system is not equilibrated - but since I took the FE input structure as the output structure from a 200ns simulation I doubt this is the case, and a distance of 17288nm (for a system of size ~15nm) suggests that the algorithm hit a division by zero singularity somewhere.
Thank you for any help you can provide.
sincerely
Yinon
Regarding the issue with the sd integrator (the domain size and the number of ranks), how many nodes are you running on? 98 ranks is quite a lot. I’d suggest trying one per gpu or cpu (not cpu core) depending what is highest. Small system cannot be divided into small enough domains to run on many ranks. But that should be the same with the md integrator.
Regarding the crash you reported with the md integrator, that looks like an ordinary explosion. Did you start this run from the end of lambda 0.1 or directly from the equilibration at lambda 0? Regarding equilibration end stability it is usually best to chain the runs, but It might not be the optimal for parallelisation.
Thank you so much for the info.
the system is quite large (~200000 atoms including water and lipids).
yes, chaining the simulations sounds like a good idea - it takes less than an hour per ns, so running 20 chained simulations should not be a problem.
It does seem that the system suffer a zero point catastrophe which maybe comes from a smaller vdw repulsion resulting in atomic overlap.
Do you have any insight as to the Coloumbic\Vdw lambda values and in what order they should be arranged (there are several mdp files floating out there with different arrangements).
Your help is immensely appreciated.
Cheers
Yinon
When having a closer look at your mdp file I see that you are keeping the Coulomb interactions turned on while you are turning off the van der Waals interactions. That will usually lead to problems at some point (unless the decoupled molecule does not have any partial charges).