Rootin' Tootin' Efficient Ray Shootin': Creating Microlensing Magnification Maps with GPUs

TL;DR

This paper introduces a highly efficient GPU-based code for generating gravitational microlensing magnification maps, essential for analyzing upcoming large-scale surveys of lensed quasars and supernovae.

Contribution

It presents the fastest publicly available microlensing map generation code utilizing GPUs, fast multipole method, and inverse polygon mapping for improved efficiency.

Findings

Significantly faster map generation compared to previous methods

Utilization of GPU parallelism for microlensing simulations

Reduction in computational complexity with advanced algorithms

Abstract

The impending discovery and monitoring of hundreds of new gravitationally lensed quasars and supernovae from upcoming ground and space based large area surveys such as LSST, \textit{Euclid}, and \textit{Roman} necessitates the development of improved numerical methods for studying gravitational microlensing. We present in this work the fastest microlensing map generation code currently publicly available. We utilize graphics processing units to take advantage of the inherent parallelizable nature of creating magnification maps, in addition to using 1) the fast multipole method to reduce the runtime dependence on the number of microlenses and 2) inverse polygon mapping to reduce the number of rays required. The code is available at https://github.com/weisluke/microlensing/.