GPU Accelerated Nuclear Cross Section Calculation in the Glauber Model

Information on the nuclear radii and other parameters of nucleon distributions has been obtained from nucleus-nucleus collisions at energies of about 1 GeV. The experimental data on nucleus-nucleus cross-section is usually analyzed using the Glauber theory. It is possible to obtain values for relevant cross-sections without any simplifications (optical or rigid target approximations) using Monte Carlo (MC) integration techniques. To increase accuracy of numerical MC integration and decrease computational time, calculations are usually done using Message Passing Interface (MPI) together with a programming language of choice on a CPU computational cluster. In contrast to MPI, executing the calculation on a single modern consumer GPU using the CUDA C++nets a factor of roughly 1000x speedup in execution over a single CPU thread. This means a single consumer GPU (we use an RTX 2070, far from the highest performance available) can outperform even a modestly sized CPU cluster and deliver higher accuracy MC results in less time. We discuss the developed CUDA MC routine, demonstrate key differences between GPU-based and MPI-based algorithms for MC integration, present nuclear cross-section calculations and extract parameters of nuclear densities for various stable and unstable isotopes.