Speaker
Description
Multilayer radiation shielding is only effective if it is optimised for the specific radiation environment in which it is used. Even though the parameter space of possible multilayer radiation shielding configurations is infinite, it can be systematically explored using Monte Carlo particle transport codes like the ESA software Geant4 Radiation Analysis for Space (GRAS) and the MUlti LAyer Shielding SImulation Software (MULASSIS).
GRAS allows the use of the Geometry Description Markup Language (GDML) to generate large numbers of shielding and detector volumes procedurally. This places the complexity into the geometry file, while the simulation setup and analysis are simple.
The main drawback of this approach is that the computational overhead introduced by the simultaneous scoring of many detector volumes grows faster than the number of simulated configurations and becomes prohibitive for large simulations exceeding ~2500 configurations in the same simulation.
MULASSIS is a lightweight 1D Monte Carlo particle transport code based on the same Geant4 physics engine but optimised for multilayer optimisation. While it is strictly single-threaded and supports only one multilayer shielding configuration per run, a large number of MULASSIS instances can be deployed in parallel with different input parameters to scan the parameter space. Due to each instance being fully independent of the other instances, the number of simulated particles can be dynamically adjusted and the computational cost scales strictly linearly with the number of configurations. This approach shifts the complexity away from the geometry definition towards setting up large numbers of independent MULASSIS instances, which, for large parameter sweeps, cannot be done manually.
For this purpose, we use Python to procedurally generate large numbers of configuration files for independent MULASSIS instances. A dynamic scheduler is used to read preliminary results and to schedule additional runs to reduce statistical uncertainty. The scheduler also allows increasing precision in interesting regions while the simulation is running, by, for example, topping up configurations around a forming minimum in parameter space while configurations that are falling behind are stopped early. Additional efficiency gains are obtained through spectrum truncation, spitting, biasing, and mono-energetic dose-response sweeps that separate expensive particle transport calculations from inexpensive spectral weighting.
The resulting workflow produces high-resolution maps of multilayer shielding performance, supporting searches for optimal layer structures for space radiation shielding applications.