Multi-GPU Accelerated High-Fidelity Simulations of Beam-Beam Effects in Particle Colliders

Numerical simulation of beam-beam effects in particle colliders are crucial in understanding and the design of future machines such as electron-ion colliders (JLEIC), linac-ring machines (eRHIC) or LHeC. These simulations model the non-linear collision dynamics of two counter rotating beams in particle colliders for millions of turns. In particular, at each turn, the algorithm simulates the collision of two directed beams propagating at different speeds with different number of bunches each. This leads to non-pair-wise collisions of beams with different number of bunches that results in an increase in the computational load proportional to the number of bunches in the beams. Simulating these collisions for millions of turns using traditional CPUs is challenging due to the complexity in modeling non-linear dynamics of the beams and the need to simulate collision of every bunch in a reasonable amount of time. In this Thesis, we present a high-performance scalable implementation to simulate the beam-beam effects in electron-ion colliders using a cluster of NVIDIA GPUs. The parallel implementation is optimized to minimize the communication overhead and the performance scales near linearly with number of GPUs. Further, the new code enables tracking and collision of the beams for millions of turns, thereby making the previously inaccessible long-term simulations tractable. As of now, there is no other code in existence that can accurately model the single particle non-linear dynamics and the beam-beam effects at the same time for a large enough number of turns required to verify the long-term stability of a collider