One instability of `CLEAN' is well known: its images of extended sources are sometimes modulated at spatial frequencies corresponding to un-sampled parts of the u,v plane (e.g., Cornwell 1983). Convolution with a larger `CLEAN' beam than usual can mask this problem, especially if the un-sampled regions are in the outer parts of the u,v plane. Reducing the loop gain to very low values generally has little effect. Various modifications to CLEAN have been invented to try to combat this problem (see e.g. Cornwell (1983)), but overall the experience is that the best solution is to use another deconvolution algorithm, such as MEM.
The occurrence of the stripes is a natural consequence of the incorrect information about radio sources embodied in the `CLEAN' algorithm. Astronomers have not found much evidence for real stripes in radio sources, so they are skeptical about stripes in `CLEAN' images. Unfortunately the only a priori information built into `CLEAN', via the use of `CLEAN' boxes, is that astronomers prefer to see mainly blank images; there is no bias against stripes. These and other considerations motivated the development of deconvolution algorithms which incorporate extra constraints on astrophysically plausible brightness distributions or which are claimed to produce, in some way, optimal solutions to the deconvolution equation. MEM is an example of the latter.
1996 November 4