The MD group in Groningen has developed a DPD integrator with nice properties, which we would like to include in the Gromacs distribution.
There is a basic version, but some work needs to be done to make it work with MPI, (OpenMP) threads, the Verlet cut-off scheme and constraints.
I am not up to date on the details of the integrator. Nicu, could you post a link to the manuscript describing the integrator?
I assume the integrator should randomly pick a fraction of the pairs to act on. Then we need an mdp parameter to set this and then loop over the complete nbnxn pairlist (multi-threaded) and select pairs using a random number generator. All our other stochastic integrators now produce reproducible dynamics, even when running on different numbers of ranks and/or threads. It is going to be hard to achieve that for DPD. The only easy way I can see to achieve this is to use a random number for each particle pair, which is going to be very expensive when the probability of taking a pair is set low.
#1 Updated by Nicolae goga almost 5 years ago
Here is the link to the article:
Constraints had a problem. Therefore a second article was written and published quite recentely. Attached here as a file.
Keep in touch