Possible errors in CoREAS parallel (MPI) runs compared to serial-run showers
We have produced an ensemble of 140 showers, and found some strange-looking radio footprints. That is, some footprints of showers with very similar Xmax values are way too small. See an example here:
This also shows up in an Xmax reconstruction by fluence fitting (this is how we discovered this). For comparison we have simulated the same 140 showers, with as much as possible the same inputs, in a serial run on our own cluster.
The outliers can also be found in a footprint size vs Xmax plot, where we take the radius of maximum fluence as a proxy for footprint size:
So, we conclude that roughly 15% of showers look odd, whereas the others seem to be OK.
Version numbers were: serial run @RU-Nijmegen: 7.7410
parallel run @KIT: 7.7420
Other options were Sibyll 2.3d, URQMD 1.3cr and the options PYTHIADIR PARALLEL CURVED MPIRUNNER CHARM THIN PARALLELIB COREAS SLANT UPWARD
(for the serial run, no MPIRUNNER and PARALLELIB)
We are currently running re-tests by running showers again, also trying different numbers of cores etc.
The input steering files, including a GDAS atmosphere file, are found here: 800052_steering.tar.bz2
I hope you can shed some light on this and hopefully find out what is wrong, either with our inputs or with the parallel running.