Further processing

Sources that could not be fitted by the program (about one percent of the sources, mostly because the starting conditions for the algorithm were not good enough, or the number of iterations for fitting exceeded the maximum number of iterations) are automatically fitted by another program which is better at fitting sources than SAD, especially at lower flux levels, but is also slower. This program must also have peak positions as input in order to work properly. These positions are taken from the output from SAD which contains the peak positions of the bad fitted peaks.

After fitting these sources a residual map is made with all the fitted sources subtracted. The source finding program doesn't search down to in one step, because then the chance is big that you will miss sources with small fluxdensities. The finding routine is used iteratively in 3 steps. First sources are searched with a peak flux higher than 100 mJy. Then the residual map is used to find the rms over the entire map. This is not the true rms, because there are still sources in that map with a peak flux between 100 mJy and the calculated level. Tests show that the calculated rms is about 10 percent overestimated. That is why we use a third iteration with a search level of of the residual map from the second iteration. This rms value is close to the true rms value (about 0.5 mJy difference at level). In order to obtain a better estimate of the r.m.s. when there are strong sources in the map or large negative gaps due to bad source subtraction an adopted version of the AIPS task IMEAN was written, which makes a histogram of the pixelvalues in the map and discards large values at the tails of the distribution. The values of these binned pixelvalues plus information about binsize and peak position are outputted, so one can check later how the r.m.s. estimate was calculated.

The last residual map is saved on tape in case we want to search deeper in the map to lets say . These deep searches are not used for our primary search, because this would lead to a worse reliability of the catalogue , but it could contain interesting sources. Another argument for making the residual maps is for inspection reasons. It is possible that the extraction program will fit a source wrong and not report about it. This type of error will leave a residual component in the map if the fit is underestimated or a negative hole in the residual map if the fit is overestimated. If a positive component is left in the map it isn't bad, because this component will be found in the next iteration. The negative holes are not found by SAD, because this program only searches for positive flux. A special routine had to written to find these negative places in the residual map (inverting the residual map and then find positive sources just like above). If they are found the source entries in the output of the previous iteration are automatically deleted and a new fit is made with another fitting program (see above). The final result is a (hopefully) good source list and a perfect residual map.

The final residual maps and the inverted residual maps are then checked by eye to see if everything went satisfactory. Usually life isn't perfect and a few sources show unwanted fitting components. At this moment it looks like the very strong sources (above 1 Jy), irregular sources and the very elliptical ones are not always fitted well. These sources are cut out of the map and are fitted by hand. Usually there are about ten of these sources per frame. The map with these cut-away sources is then used for the second run of the source finding batch. This gives in almost all cases a satisfactory result. The final source lists are then stored in another subdirectory where further processing can take place. This processing involves storing the source-lists into a database and making adjustments to the dataformat. More details can be found in the next chapter.