Learning to See: Applying Inverse Recurrent Inference Machines to See through Refractive Scattering

Abstract

The Event Horizon Telescope (EHT) has produced horizon-resolving images of Sagittarius A* (Sgr A*). Scattering in the turbulent plasma of the interstellar medium distorts the appearance of Sgr A* on scales only marginally smaller than the fiducial resolution of EHT. The scattering process both diffractively blurs and adds stochastic refractive substructures that limits the practical angular resolution of EHT images of Sgr A*. We explore the ability of a novel recurrent neural network machine learning framework to mitigate these scattering effects, after training on sample data that are agnostic to general relativistic magnetohydrodynamics (GRMHD). We demonstrate that if instrumental limitations are negligible, it is possible to nearly completely mitigate interstellar scattering at a wavelength of 1.3 mm. We validate and quantify the fidelity of this scattering mitigation scheme with physically relevant GRMHD simulations. We find that we can accurately reconstruct resolved structures at the scale of 3 μ as, well below the nominal instrumental resolution of EHT, 24 μ as.

Keywords

• Supermassive black holes
• Galactic center
• Interstellar scattering
• Convolutional neural networks
• Neural networks