Neural Rendering of Relativistic Spacetime: A Physics-Informed Neural Network Approach to the Starbow Effect
Year: 2026
Type: preprint
Abstract: Traditional ray-tracing engines for simulating relativistic spacetime and gravitational lensing rely on computationally expensive, pixel-by-pixel trigonometric calculations. This computational overhead bottlenecks real-time virtual astrophysics and large-scale cosmological light-cone generation. In this paper, we present a novel methodology utilizing a Physics-Informed Neural Network (PINN) to bypass classical rendering mechanics. By training a deep multi-layer perceptron to minimize a loss function governed strictly by the Lorentz transformations and Einstein's deflection equations, the artificial intelligence autonomously learns the non-linear continuous geometry of spacetime. We demonstrate that the trained network can instantaneously collapse a 4D cosmological timeline of gravitational lensing into a 2D relativistic viewport for an observer traveling at 0.99c, successfully rendering both the Starbow effect and Einstein rings. This approach presents a highly efficient alternative for modeling emergent gravity and relativistic aberration.
Author: Gurvir Singh
Language: English
Subfield: Astronomy and Astrophysics
Field: Physics and Astronomy
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7225429743