Unexpected fringe field behaviour in 1.7.5

[Lode Vanhecke]

At two spots after a sbend element, these is a certain fraction of elelctrons seemingly being scattered at a near constant angle. In the appendix I put a zip with the files + distribution file needed to observe this difference in behaviour between 1.6 and 1.7. I run "bdsim --file=perle-2.1.250.gmad --output=none".

[Laurie Nevay]

Hello,

Thanks - I think we discussed this a little offline before. My colleague Will had a look and we think we understand why it’s happening. Can you confirm which geant4 version you’re using please?

Thanks,
Laurie

[Lode Vanhecke]

We did discuss it shortly yes. At the moment I am using version 10.7.3.

[Laurie Nevay]

Agreed it is in an issue.

[Laurie Nevay]

We discussed and Will and I understood the problem. For thin things (e.g. fringes, rf fringes, multipole kicks, thin hkicker/vkickers), the relative tolerance on integration error as compared to the step length was really too strict. This was never a problem in the past until Geant4 fixed their implementation recently that never really worked well.

In BDSIM, Will implemented a fix where we have a different numerical tolerance for thin things. Also, for dipole fringes the error was reported wrongly from the integrator which was also fixed. This is fixed as of BDSIM v1.7.7.