Experimental test of Monte Carlo proton transport at grazing incidence in GEANT4, FLUKA and MCNPX.

The ability of the Monte Carlo (MC) particle transport codes GEANT4.8.1 and GEANT4.8.2, FLUKA2006 and MCNPX2.4.0 to model proton transport at grazing incidence onto tungsten blocks has been tested and compared to experimental measurements. The test geometry consisted of a narrow proton beam of two energies, 98 MeV and 180 MeV, impinging on a thick tungsten alloy block at grazing incidence. The distribution of forward out-scatter from the tungsten alloy block was measured with a fluorescent screen viewed with a CCD camera via a mirror. In the MC simulations, the experimental setup was modelled and the dose deposited to the fluorescent screen material was scored. Simulations and measurements were made for four different incidence angles (3.5, 5.0, 7.5 and 10 degrees ). Several different sets of calculations were performed, studying the impact of different user-defined settings in the different MC packages. The study of different parameters settings in the GEANT4.8.1 simulation showed a strong dependence of the calculated out-scatter probability on the maximum allowed step length. For the largest incidence angle an increase of 60% in the out-scatter probability was found when restricting the maximum allowed step length to 0.05 cm. We also observed that the stepping algorithm of GEANT4.8.1 and 4.8.2 introduces a small non-physical directional and positional asymmetry at the exit boundary of the tungsten alloy block. The shape of the energy spectrum of protons being out-scattered agreed between the codes. The dose-weighted forward out-scatter probability, i.e. the ratio between the total signal from the unscattered beam and the out-scattered beam, showed a qualitative agreement of simulations compared to measurements. Quantitatively, the deviation of the simulations reached as high as 37%, while the experimental uncertainty was 14%. The mean emission angle of the simulations was within 16% of the measurement for all incidence angles with a measurement uncertainty of 8%.