Modelling

Modelling is the only feasible way of assessing the reliability of the reconstructions. (Modelling here refers to the model used to generate a simulated data set, not the implicit modelling performed by the deconvolution algorithms.)

1. Select plausible model from from initial images.
2. Model-fit for parameters to the original data if possible.
3. Check that simulated observations can recover the significant features of the model reliably. In this case we were concerned that the natural asymmetry in the PSF was driving the position angle of the hot spots.
4. Run plausible models less the features in question though the imaging cycle - can the features be produced spuriously?
5. Vary the parameters of the observation to significantly alter the PSF. Pathological instabilities in the beam can drive all varieties of deconvolution to a similar error.
6. Use simulated observations to determine the effective resolution of the deconvolution sky model. MEM has a bias towards smoothness, and will only make a peak as sharp as needed to agree with the data within the noise. Consequently it has a (space variant) internal resolution in the sky model. The original model was smoothed to maximize agreement with the MEM sky model. In this case, internal resolutions about about 0."3 to 0."4 were typical. Consequently when the sky model is smoothed with a restoring beam of 0."4 the final image will have a resolution of 0."5 to 0."55.