Question about doses varying between using original source and derived phasespace as source of particles

[nexustowers]

I am using egs_chamber from EGSnrc 2025 to create a simple model of a linac head with a phantom underneath. The head model consists of a target, primary collimator etc etc and the primary source of particles is an electron beam hitting the target creating bremstrahlung. The surrounding medium is air.

The head assembly is surrounded by a spherical shell of air which I am using to create a IAEA phase space of particles exiting the shell using the from-regions / to-regions method . I am interested in head leakage, hence the spherical phase space geometry instead of a plane under the jaws. I am tracking all particle types.

The phantom is underneath the jaws and outside the phase space shell. I am calculating the dose to a dosebox on the cax inside the phantom at some depth.

I run the simulation with ncase, say, 1e8 electrons, and this creates a phasespace with original particles close to (but not exactly) 1e8 and, say, 1e6 total particles in the phasespace. I note the dose to the dosebox.

I run the simulation again, this time with the phasespace as particle source and set ncase in my .egsinp file to 1e6. I understand this should use all the particles in the phasespace file once and normalize the results to the number of original particles used to create the phase space file ie 1e8. I am not doing any recycling or variance reduction.

Since using the phase space is essentially recreating the original scenario and the normalization should be consistent, I am expecting the dose reported to the dosebox from the phasespace simulation to be the same as the first simulation (within uncertainties, which in the case I have calculated is about 15%) but it is not - it is about an order of magnitude different with the phase space simulation dose being the higher. Can anyone please explain any conceptual reasons why they're not the same?

[rtownson]

Are you sure that you're actually capturing the entire source in the phase-space that you're scoring? If some of the particles managed to get around the phase-space scoring "to regions" and "from regions" pairs, then the normalization will end up different.

Just to be sure, I've tested this myself. An enclosed sphere for me scored to a phase-space, which is then used as the source, gives the same dose. However, if I use a source only partially covered by the phase-space scoring regions, the dose ends up different.