Numerous bow shocks in the outer Helix Nebula

Near the end of their lives, low- and intermediate-mass stars expel metal-enriched material in winds and outflows, ultimately producing planetary nebulae (PNe). The ejected material is expected to fragment and mix into the interstellar medium (ISM), but this final assimilation step has been difficult to observe directly. Here we report evidence for this process in the form of twenty-two bow shocks in the eastern outskirts of the Helix Nebula, detected in Hα emission with the partially completed MOTHRA telescope. Unlike the large-scale wind–ISM bow shocks commonly observed around evolved stars, the shocks are compact and associated with individual clumps of gas. Going outward from the central star, the radius of curvature Rc decreases by a factor of ∼ 102 over the radial range r = 0.4–1.4 pc. This is accompanied by a morphological transition from thin, well-defined bows to fuzzy, patchy structures. We interpret these changes as progressive stripping and fragmentation of AGB-shell remnants as they interact with the ISM. The slope of the observed Rc–r relation implies a loss of fragment coherence on a timescale of ≈ 104 yr, providing a rare direct constraint on the time-scale for disruption and entrainment of fragmented stellar ejecta into the ISM.

Improvements to AI Systems:

1. Astrophysical Shock-Morphology Classifier

• Improvement: Train a vision transformer on multi-band Hα/optical images to automatically detect and classify bow-shock morphologies (thin/well-defined vs. fuzzy/patchy) across radial distances.

• Capability: The AI can autonomously map the transition zone of stellar ejecta in any PN or AGB-shell image, flagging candidate fragmentation events without manual inspection.

2. Predictive Model for Fragment Coherence Timescale

• Improvement: Build a regression model (e.g., Gaussian process or neural network) that ingests radial distance, shock curvature (Rc), and ISM density estimates to predict the disruption timescale (104 yr) and entrainment rate.

• Capability: The AI can forecast how long clumps of stellar ejecta survive before mixing into the ISM, enabling predictions for other PNe or evolved stars with incomplete data.

3. Simulation-to-Observation Inference Engine

• Improvement: Use a normalizing flow or diffusion model to invert hydrodynamic simulations of AGB-shell fragmentation, conditioned on observed Rc–r slopes.

• Capability: The AI can infer hidden physical parameters (e.g., initial clump mass, ISM pressure, magnetic field strength) directly from observed shock geometries, providing a fast alternative to full radiative-transfer modeling.

4. Automated Radial-Profile Analyzer for Curvature Evolution

• Improvement: Implement a curve-fitting algorithm (e.g., robust spline or Bayesian piecewise model) to extract Rc as a function of radius from noisy Hα maps, with uncertainty quantification.

• Capability: The AI can produce high-confidence Rc–r relations for any PN, detecting subtle breaks or power-law slopes that indicate different stripping regimes (e.g., ram-pressure vs. turbulent fragmentation).

5. Multi-Object Bow-Shock Detector for Large Surveys

• Improvement: Fine-tune an object-detection model (e.g., YOLO or DETR) on synthetic and real bow-shock images to find compact, clump-associated shocks in wide-field surveys (e.g., LSST, Euclid).

• Capability: The AI can scan entire sky surveys to discover dozens of similar ejecta–ISM interaction sites, enabling statistical studies of metal-enrichment timescales across the Galaxy.

6. Physics-Informed Neural Network for ISM Entrainment Dynamics

• Improvement: Embed the observed Rc–r slope and timescale constraint as a loss term in a PINN that solves the equations of motion for fragmented clumps in a moving ISM.

• Capability: The AI can simulate the full life cycle of AGB ejecta—from clump formation to full mixing—and output time-resolved maps of metal distribution, directly comparable to future observations.

7. Cross-Scale Anomaly Detector

• Improvement: Train an autoencoder on large-scale wind–ISM bow shocks (from evolved stars) and fine-tune it to flag compact, clump-scale shocks as anomalies.

• Capability: The AI can automatically distinguish between large-scale stellar wind interactions and small-scale ejecta fragmentation in mixed datasets, improving source classification in archival images.