AI stitches 116 VLBI observations into a continuous, ultra-high-resolution view of the 3C 345 jet
AI-based neural-field reconstruction (kine) combines 116 MOJAVE VLBA epochs of blazar 3C 345 into a time-continuous, high‑resolution map, achieving ~113 microarcsecond effective resolution and ~500,000:1 dynamic range. It shows apparent jet speeds of 10–13 times light (bulk flow ~9–12c) due to geometry, not actual superluminal motion; interpolation between sparse epochs remains reliable for up to ~6 months. The knots may arise from overdense or emissive regions rather than strong shocks, and the method provides a new way to analyze long-running VLBI data that could be applied to other programs, without creating extra observations.