Optical Photon Emission in Extended Airshowers -- Hybrid computing in the context of CORSIKA 8

TL;DR

This paper discusses the redevelopment of the CORSIKA air shower simulation in C++, integrating hardware acceleration for optical photon propagation using deep learning hardware like Nvidia GPUs to enhance simulation speed and efficiency.

Contribution

It introduces a new C++ version of CORSIKA with hardware-accelerated photon propagation leveraging deep learning hardware, advancing simulation performance.

Findings

Successful implementation of GPU-accelerated photon propagation.

Significant speed-up in optical photon simulations.

Compatibility with existing experimental data

Abstract

With the motivation to improve experimental gains and precision, established astroparticle experiments are currently undergoing massive upgrades. In addition, several new experiments are being built or planned. With the resulting gain in observational quality, the amount and accuracy of simulated data required for the analysis is also rising. In order to meet the increasing requirements and complexity due to the experiments' growth and to provide a unified software ecosystem, it was decided to re-develop the de facto standard extensive air shower simulation CORSIKA completely in C++ based on the original Fortran code. Since one of the largest runtime consumers is the propagation of millions of optical Cherenkov and fluorescence photons, and many experiments are starting to use them for measurements, it was decided to develop hardware-accelerated code to speed up the simulation. Specific methods have been developed to propagate photons on deep learning acceleration hardware similar to classical GPUs to take additional advantage of the current and future growth of the deep learning sector. In particular, Nvidia accelerators were tested.