I just re-ran gen-system.py from the paper's SI on 32 MPI ranks.
I used the Ubuntu 18.4 docker container from
That's the image with which the Scafacos integration for Espresso is tested continually.
I am not able to reproduce the error you reported.
Before you spend time on getting
this resolved, I'd suggest the following;
* develop a small enough version of your simulation model so you are able to use DipolarDirectsumCpu. The paper from 2012 was done with that (although at that point still using TCL as scripting language).
* Once that works, and if it becomes clear you need a larger system, come back to P2NFFt and Scafacos.
As a first step, run the scripts from the SI with 8 mpi cores. Using more than that is often not useful with Espresso.
The issue you report might be a bug (unrelated to magnetostatic interactions which are not active in gen-system.py) in Espresso with a very low probability of ocurring. With Espresso 4.0.2 it is not yet possible to make a simulation fully deterministic, including the pseudo-random numbers used by the thermostat.
This will be possible with Espresso 4.1. Once that is close to release, I'll update the P2NFFT rrelated run-scripts at