A highly-parallel Monte-Carlo-Simulation of X-Ray-Scattering using a Particle-Mesh-Code on GPUs

In this thesis a software solution was developed that simulates the scattering of X-rays in matter using a Monte Carlo approach.

The application of small-angle X-ray scattering in the studies of the complex processes occurring during the interaction of short intense laser pulses in solid matter provides the motivation for this work. Therefore this technique is described and it is shown how Fourier transformation can be used for approximating the scattering results. It is shown how they can be efficiently implemented in computers using the fast Fourier transform (FFT) and why this approach has limitations when describing scattering processes.
To circumvent these, a model was developed that uses photon-like particles to describe the X-rays.
Billions of such particles are required to provide a good approximation of the physical processes involved, which is why the simulation algorithm described in this work was designed from the ground up to support the massively parallel structure of modern graphic processing units (GPUs) allowing to simulate many particles at once. The implemented algorithm is described focusing on the special methods required to make the most use out of GPUs.

As the choice of the appropriate data types is vital for the correctness and precision of the algorithm a comprehensive analysis and test of the numerical accuracy was deployed. It is shown that even reduced precision provides results that are accurate enough for a wide range of applications. Therefore, smaller data types can be used allowing to simulate much larger experiments on a given hardware.
Finally the correctness and good scalability of the parallel algorithm are demonstrated.