Unweighted events with setSecBiasing

[Sebastien_Procureur]

Hi, I have a question concerning bias I simulated a basic setup consisting of electrons hitting a tungsten target in vacuum, and a gas chamber nearby detecting the resulting X-rays through a Be window (e.g. X-ray tube + detector).
I used the command /process/em/setSecBiasing eBrem world 100 30 keV to enhance the statistics through splitting (I also used my region “anode” instead of world, without any difference). It basically works, I have weighted photons with w=0.01. But it keeps simulating unweighted photons (w=1) which largely spoils the acceleration factor of this biasing technique. More precisely, if I write N1 (resp. N2) the number of unweighted (resp. weighted) photons in my detector, and f=100, the relative error is given by sqrt(N1+N2/f²)/(N1+N2/f). Of course if N1=0 (no unweighted photons) the error goes with 1/sqrt(N2) which makes this biasing option very interesting, but if N1 is constant with f (which I observe), the biasing gain quickly saturates (in my case, around f=10 only!).
Do you please have any idea of what I’m doing wrong?
Thanks in advance and kind regards,
Sébastien

Geant4 Version: 4-11-02
Operating System: Ubuntu 23.10
Compiler/Version:
_CMake Version:_3.27.4

---

[ctwhite]

These weight = 1.0 particles usually occur when the product from a non-split interaction scatters off something in the geometry and ends up traveling in the same direction as the split particles. Your splitting target might be too small, or the geometry not shielding adequately.

[Sebastien_Procureur]

Hi, you’re right, I had some “leaks” in the setup, now it’s fine. Thanks for your help!