Analysis of X-ray Polarization from Warped Accretion Disks: Machine Learning Aids Black Hole Astrophysics

This article introduces the open-source toolkit warped-disk-xray, maintained by juangjuang74-eng and hosted on GitHub (link: https://github.com/juangjuang74-eng/warped-disk-xray). It aims to analyze X-ray polarization signals from warped accretion disks around black holes using machine learning, providing data science solutions for studying physical processes in extreme gravitational fields near black holes. The toolkit integrates data processing, physical simulation, ML inference, and visualization functions, representing a cutting-edge exploration at the intersection of astrophysics and machine learning.

Black holes cannot be observed directly; they must be studied indirectly through accretion disks (high-temperature matter disks captured by black hole gravity). Accretion disks rotate due to conservation of angular momentum, and friction heats them to millions of degrees, emitting X-rays that carry key information such as black hole mass and rotation. The extreme gravitational field near black holes warps the accretion disk, altering X-ray polarization properties and making it a sensitive probe for detecting the black hole environment. Analyzing such complex polarization signals requires handling high-dimensional data and complex physical models, where machine learning can play an important role.

warped-disk-xray is an open-source data science and ML toolkit that provides a complete analysis pipeline:

1. Data Preprocessing: Reads, calibrates, and cleans X-ray polarization observation data (including photon energy, arrival time, polarization angle/degree, etc.), and performs instrument response correction;
2. Physical Model Simulation: Implements a radiative transfer model for warped accretion disks, calculating theoretical polarization signals under different parameters (black hole rotation, inclination angle, warping degree, etc.);
3. ML Inference: Uses neural networks to accelerate parameter inference (tens of thousands of times faster than traditional methods);
4. Visualization Tools: Interactively displays polarization data and model fitting results.

Polarization Generation Mechanisms:

• Compton Scattering: Thermal electrons scatter X-ray photons, and polarization properties depend on scattering geometry;
• Relativistic Beaming Effect: Accretion disk matter moves near the speed of light, and Doppler/aberration effects change polarization angle and degree;
• Gravitational Lensing Effect: The strong gravity of black holes bends light, altering the spatial distribution of polarization signals.

Characteristics of Warped Accretion Disks: The disk surface deviates from a single plane, driven by Lense-Thirring precession (frame-dragging by rotating black holes), radiation pressure instability, magnetic field effects, etc. Its polarization signal is significantly different from that of a flat disk, making it a powerful tool for probing three-dimensional geometry.

Limitations of Traditional Methods: High computational cost (radiative transfer simulations take minutes to hours), large parameter space (difficult to cover all free parameters), and non-uniqueness in parameter inference.

ML Advantages:

• Fast Inference: Trained models complete parameter inference in milliseconds;
• Handling High-Dimensional Data: Deep learning automatically extracts complex patterns;
• Uncertainty Quantification: Bayesian NNs and others provide parameter uncertainty estimates;
• Anomaly Detection: Identifies observations that do not fit standard models, suggesting new physical phenomena.

Data Pipeline: Supports IXPE telescope data formats, including instrument response modeling (energy/time response, polarization modulation factor), background estimation, and time-resolved analysis.

ML Models: Explores CNN (extracting spatial-energy features), RNN (processing time series), GNN (modeling energy bin relationships), generative models (VAE/GAN for data augmentation), and PINN (incorporating physical constraints).

Training and Validation: Synthetic data training (covering a wide parameter space), transfer learning, leave-one-out cross-validation, and physical consistency checks (energy conservation, causality, etc.).

Main Applications:

1. Black Hole Rotation Measurement: Infer the inner region geometry of the accretion disk through polarization signals to indirectly measure black hole rotation (validated with the iron Kα line method);
2. Accretion Disk Geometry Detection: Fit polarization features (e.g., angle variation with energy) to infer warping amplitude, radius, and precession rate;
3. Strong Gravitational Field Testing: Use polarization signals to test general relativity predictions (gravitational lensing, frame-dragging, etc.);
4. Multi-Wavelength Joint Analysis: Integrate optical, radio, and other data to provide a comprehensive portrait of the black hole-accretion disk system.

Current Challenges: Scarcity of X-ray polarization observation data, simplified physical models (gap between synthetic data and real observations), high computational resource requirements, and systematic errors (instrument calibration, background estimation).

Future Directions: Develop real-time analysis pipelines, multi-source joint analysis, improve physical models (Monte Carlo simulations), explainable AI (understand the physical mechanisms behind predictions), and automated end-to-end pipelines to lower technical barriers.