Fig. 2: Visual comparison of models without and with perturber \({\mathcal{V}}\).

Top row, pixellated source surface-brightness maps. In the model without \({\mathcal{V}}\), the source model attempts to fit the gap in the bright lensed arc, resulting in a sharp discontinuity in the surface brightness along the lensing caustic (left column). The gravitational effect of \({\mathcal{V}}\) folds the caustic over onto the bright radio lobe, correcting this discontinuity and recovering a smooth, contiguous image (right column). The inset region around the bright, quadruply imaged source component has a side length of 4 mas. Also note the presence of a much fainter, doubly imaged source component ~50 mas to the north of the bright component. Middle row, lens-plane surface-brightness maps (reconvolved with the main lobe of the point spread function). Note that the gap in the arc is still present in the model without \({\mathcal{V}}\); in this case, the source has attempted to absorb the discontinuity. Bottom row, residuals, which have been normalized in the visibility plane and Fourier transformed into the image plane for ease of visualization. The inclusion of \({\mathcal{V}}\) corrects a >5σ peak in the residuals (corresponding to a 2% change in surface brightness) near the intersection between the critical curve and the lensed arc. The zoomed-in region in the middle and bottom rows is the same as the inset region in Fig. 1.