search

SixTrack / sixtracklib

Library for single charged particle simulations in accelerators
GNU Lesser General Public License v2.1
12 stars 16 forks source link

Bug with apertures and `track_line` in CUDA? #128

Closed aoeftiger closed 4 years ago

aoeftiger commented 4 years ago

When tracking only part of a beam line -- step-by-step with cudatrackjob.track_line -- and losing particles in an aperture, they will be "revived" after starting the next turn: their state variable was 0 after the end of the turn, but a next track_line will set them back to state=1.

Perhaps I'm not using the SixTrackLib python API correctly, but the attached example should demonstrate just what I mean:

reproducing.ipynb.zip reproducing.py.zip

martinschwinzerl commented 4 years ago

Thank you for reporting this issue and for providing a minimal example for reproducing the problem! The problem was a combination of a missing initial-state check in the track_line method and an overly enthusiastic global aperture check, happily re-labeling already lost particles if they meet the global aperture criteria.

Note, that the problem is most likely also affecting other backends (i.e. CPU and OpenCL track jobs). However, since the main use-case for track_line is currently the deep integeration with other cuda libraries, this has not been found earlier.

aoeftiger commented 4 years ago

Excellent job, @martinschwinzerl ! https://github.com/SixTrack/sixtracklib/pull/129 solves the mystery!