A major drawback to the use of `CLEAN' is the way in which its answers depend upon the various control parameters: the location of `CLEAN' boxes, the loop gain and the number of `CLEAN' subtractions. By changing these one can, even for a relatively well-sampled u,v plane, produce noticeably different final images. In the absence of an error analysis of `CLEAN', one can do nothing about this except practise vigilance and avoid interpreting any aspects of an image that are unstable to the choice of control parameters.
Part of our purpose in this tutorial is to make you aware of effects that should keep you from being over-confident in the final images produced by `CLEAN'. In almost any astronomical application, Monte Carlo tests of `CLEAN', and comparisons of its results with those of other deconvolution methods, are illuminating. They remain the only practical way to estimate the effects of data errors and of different `CLEAN'ing strategies on the final image.
Eventually, you will gain experience of applying `CLEAN' to a wide range of different images. This experience will let you guide `CLEAN' to plausible results more quickly. The `CLEAN' images that you then produce may not be intrinsically more reliable, but you will have calibrated your use of them for astrophysics much better!
1996 November 4