GPU-accelerated Monte Carlo Transport with the Einstein Toolkit

Simulating particle transport, such as that of neutrinos or radiation, is a critical component for high-fidelity models of relativistic astrophysical phenomena, including core-collapse supernovae, binary neutron star mergers, and accretion disks. However, Monte Carlo (MC) methods, while highly accurate and inherently parallelizable, are computationally expensive and often represent the primary performance bottleneck in large-scale simulations.

We present a new, GPU-accelerated Monte Carlo transport (MCT) module for the Einstein Toolkit, a community-driven, open-source software platform for numerical relativity. This module is designed to seamlessly integrate with the Toolkit's existing infrastructure, namely CarpetX, the AMReX-based adaptive mesh refinement (AMR) driver.

By offloading the computationally intensive particle advection and interaction kernels to Graphics Processing Units (GPUs), our implementation leverages their massive parallel architecture to accelerate the calculations by orders of magnitude. We describe the design of this new thorn, its coupling to the spacetime and fluid variables, and its performance on modern GPU architectures. We validate the module against a suite of standard radiation and transport benchmarks. This work enables more realistic and computationally feasible microphysics in demanding astrophysical simulations, paving the way for next-generation studies of GRMHD with radiation and neutrino-driven phenomena.