Neural-field algorithm delivers high-resolution video of a variable blazar jet
A new imaging method called kine uses neural representations to turn multi-epoch VLBI data into continuous, full-polarization videos, demonstrated on the blazar 3C 345.
Researchers introduced kine, a forward-modeling algorithm that represents very long baseline interferometry (VLBI) data as a neural field, jointly fitting all epochs to produce a smooth, full-polarimetric video of a changing source. Using 116 observations of the blazar 3C 345 taken by the VLBA between 1995 and 2022, the method delivered an effective resolution of about 113 µas—roughly four times finer than the nominal beam—and a dynamic range near 5 × 10^5, far surpassing traditional CLEAN imaging.
By applying optical-flow techniques to the high-resolution video, the team extracted a two-dimensional velocity field, showing that the apparent speeds of bright components (10-13 c) match the bulk plasma flow (9-12 c), challenging shock-driven interpretations. Polarimetric analysis revealed a persistent electric-vector alignment consistent with a toroidal magnetic field, and the lack of localized fractional-polarization enhancements further argued against strong shocks. The authors suggest that kine can be extended to other VLBI programs, including horizon-scale imaging of black holes, and may find applications beyond astronomy, such as in medical imaging.
Why it matters
The method dramatically improves how astronomers visualize and analyze rapidly changing cosmic jets, potentially reshaping models of black-hole outflows.
In this story
