In version 8.2 (but not in 7.1), sometimes an electron is stuck on the surface between two volumes (one of them vacuum (world, G4Box), the other is steel (G4Cons)). There it makes ~10^5 (!) steps, each infinitesimally small [0 or 10^-(20-30) mm)], "bouncing" between the two volumes, until finally somehow escapes and tracking continues. This happenes 3-4 times in every 100 events. FYI, I use a global magnetic field, but the bug occurs with both zero and non-zero fields. See the URL for the application in which the bug occurs. This application already contains a workaround for the bug (see hbarStepAction.cc). However, to compile, you need a slightly modified GEANT4, so contact me for further information.
We are tracking a couple of similar problems. A candidate fix is in preparation for the case of geometries with overlaps and a magnetic field. We will continue to investigate, and try to extend to the non-field case also.
The problem should be fixed in the most recent patched versions of Geant4. If still persisting for the specific use-case, please open a new problem report with all the details. Thanks.
