Created attachment 660 [details] G4RegularNavigation.cc The G4RegularNavigation is producing large amounts of Exceptions indicating that the track is stuck when small voxels (~100um edge length) are used. This issue was introduced with the fix for issue 2196. https://bugzilla-geant4.kek.jp/show_bug.cgi?id=2196 I had posted a possible solution over there, but I guess it was not seen as the problem is already marked as fixed. Please find attached my approach for the bugfix. I have used this is the last months with several hundred hours of runtime without issues, but I am not sure about the Geant4 naming conventions.
The difference between your solution and mine is the following: You: when there are more than 10000 small steps (< kCarTolerance) you abort the event) Me: I loop 10000 steps, so there can never be more than those (if the step number reaches 10000, I abort the event) Each step (after a few initial steps for which I push kCarTolerance) is pushed std::min(101*kCarTolerance*pow(10,fNumberZeroSteps-2),0.1); I think that my solution is better. But every time there is a small step, I send a warning message, while you sent none. This is the difference that causes the big output I will protect this message with a fWarnPush, which will only be active if the user has compiled with G4VERBOSE=ON I consider the bug resolved, tell me if you do not agree
I finally decided to protect the excessive warnings under @ifdef G4DEBUG_NAVIGATION. It is a safer choice to avoid you or other users having the screen filled with those messages
I am currently testing your solution. I think, it is also important to reset the fNumberZeroSteps to 0 when a correct step was made. Otherwise I got several tracks killed which got stuck a few times but could be recovered after each first consecutive push. Additionally, I get a lot of the following Exceptions (2800 exceptions in 100000 events): -------- WWWW ------- G4Exception-START -------- WWWW ------- *** G4Exception : GeomNav1002 issued by : G4PhantomParameterisation::GetReplicaNo() Corrected the copy number! It was negative or too big LocalPoint: (10.4,-11.8785,2.6816) LocalDir: (-0.695189,-0.625715,-0.353825) Voxel container size: 10.4 49.6 19 LocalPoint - wall: -1.77636e-15 -61.4785 -16.3184 *** This is just a warning message. *** -------- WWWW -------- G4Exception-END --------- WWWW ------- This happens with both my and your solution in 10.7 but not in 10.5.
I do not understand your point: fNumberOfZeroSteps is set to 0 at the beginning of G4RegularNavigation::ComputeStepSkippingEqualMaterials About the second point, it is just a warning that a precision problem has been corrected (see your LocalPoint - wall: -1.77636e-15 ). You can deactivate the warning: surround it by #ifdef G4DEBUG_NAVIGATION Probably it is a good idea that I do the same for the release
I am not sure but I assume, this happened in an area of my voxel model where the material does not change over a longer path (which is quite typical for some medical voxel models), so the navigator stays in the same function call. Regarding the message: If this is not an issue, it would be good to deactivate it for the next release. As a user, a flooded error log is hard to judge as normal and real errors are getting lost between all those debug messages.
*** Problem 2359 has been marked as a duplicate of this problem. ***
I have added if( fabs(localPoint.x()-fContainerWallX) > kCarTolerance && fabs(localPoint.y()-fContainerWallY) > kCarTolerance && fabs(localPoint.z()-fContainerWallZ) > kCarTolerance ){ // only if difference is big This I think would be enough. I will put it in the next release
