Microcaustics: Fast GPU-Accelerated Dynamical Microlensing of Time-Dependent Extended Sources
Abstract
Gravitational microlensing probes unresolved quasar accretion disc structure and compact objects in lens galaxies and is an important systematic in strong-lens time-delay cosmography. Most analyses use static microlensing map simulations because evolving individual microlenses and caustics is computationally prohibitive. We present Microcaustics, a GPU-accelerated Python package built on PyTorch for dynamical microlensing of time-dependent extended sources in strong gravitational lenses. It combines a local-exact/far-field deflection solver, source-directed lens-plane scouting with temporal reuse, and biquadratic inverse polygon mapping (IPM). GPU batching and fused Triton rasterization accelerate the calculation, while binary caustic labels track source-centred crossings. Across three simulated quadruply lensed quasar systems, our package generates a 10-year, 147-frame multiband light curve in an amortized median of 0.68 s (0.88 s with labels) on a consumer GPU with a median light-curve RMSE of 0.35 mmag, while recovering every reference source-centred caustic crossing. At comparable map accuracy, the batched throughput is 780 times higher than that of a recent CUDA IPM implementation and extrapolates to a speedup of millions relative to classic GPU-D. We demonstrate that our IPM converges as Nray−0.84, approaching the idealized Nray−1 convergence rate of piecewise-affine IPM and improving upon the Nray−1/2 inverse ray-shooting rate. We also introduce a GPU-resident general relativistic and reverberating quasar source model and demonstrate our method with an evolving supernova source. These advances make fully dynamical stellar fields practical for Bayesian and simulation-based inference.
Related
- Neglecting Dynamical Microlensing Can Bias Accretion Disk Size Measurements at the Population Level
- Revisiting the Einstein Cross with X-shooter spectra: New estimates of accretion disk parameters and broad emission line profiles from microlensing
- lenscape: Fast, High-resolution Δχ2 Maps of Microlensing Events
- Same Events, Weaker Bound: Model Dependence in Magellanic Microlensing Limits on Primordial Black Holes
- Gaia Science Alerts microlensing events with mass estimates verifiable in Gaia Data Release 4
BibTeX
@misc{fagin2026microcaustics,
title = {{Microcaustics: Fast GPU-Accelerated Dynamical Microlensing of Time-Dependent Extended Sources}},
author = {Joshua Fagin and Connor Stone and James Hung-Hsu Chan and Sophia Miskiewicz and Henry Best and Paras Sharma and Matthew O'Dowd and Laurence Perreault-Levasseur and Yashar Hezaveh and Simon Birrer},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.19703},
eprint = {2610.19703},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.19703},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Microcaustics: Fast GPU-Accelerated Dynamical Microlensing of Time-Dependent Extended Sources
AU - Joshua Fagin
AU - Connor Stone
AU - James Hung-Hsu Chan
AU - Sophia Miskiewicz
AU - Henry Best
AU - Paras Sharma
AU - Matthew O'Dowd
AU - Laurence Perreault-Levasseur
AU - Yashar Hezaveh
AU - Simon Birrer
PY - 2026
DA - 2026/10/14
PB - arXiv
DO - 10.48550/arxiv.2610.19703
UR - https://arxiv.org/abs/2610.19703
AB - Gravitational microlensing probes unresolved quasar accretion disc structure and compact objects in lens galaxies and is an important systematic in strong-lens time-delay cosmography. Most analyses use static microlensing map simulations because evolving individual microlenses and caustics is computationally prohibitive. We present Microcaustics, a GPU-accelerated Python package built on PyTorch for dynamical microlensing of time-dependent extended sources in strong gravitational lenses. It combines a local-exact/far-field deflection solver, source-directed lens-plane scouting with temporal reuse, and biquadratic inverse polygon mapping (IPM). GPU batching and fused Triton rasterization accelerate the calculation, while binary caustic labels track source-centred crossings. Across three simulated quadruply lensed quasar systems, our package generates a 10-year, 147-frame multiband light curve in an amortized median of 0.68 s (0.88 s with labels) on a consumer GPU with a median light-curve RMSE of 0.35 mmag, while recovering every reference source-centred caustic crossing. At comparable map accuracy, the batched throughput is 780 times higher than that of a recent CUDA IPM implementation and extrapolates to a speedup of millions relative to classic GPU-D. We demonstrate that our IPM converges as N_ray^−0.84, approaching the idealized N_ray^−1 convergence rate of piecewise-affine IPM and improving upon the N_ray^−1/2 inverse ray-shooting rate. We also introduce a GPU-resident general relativistic and reverberating quasar source model and demonstrate our method with an evolving supernova source. These advances make fully dynamical stellar fields practical for Bayesian and simulation-based inference.
ER -