When you select the `Auto' option in the menu, `arc' tries to use the fit you have obtained already as a starting point and tries to find lines in the arc that match lines in the tables. The algorithm used is very simple, and is based on the principle that the automatic fit should not add lines that will make the fit significantly worse than do the lines you have already got-and are presumably happy with. There are only two parameters (at present) involved and, really, only one is important.
`arc' takes each pixel in the spectrum in turn. If that pixel is more than `chfact' times the current sigma value from any line already found, it uses that pixel as the starting point for a line search. This is exactly as if you had selected that pixel with the cursor during the interactive part of the process. If anything resembling a line can be found, it calculates its wavelength and looks in the line tables for a line close to that wavelength.
A line is accepted if the discrepancy between its calculated and tabulated wavelength is less than `sigfact' times the current RMS value. It is this that means that the criterion for accepting new lines is based on how their wavelength discrepancies compare with those for the lines that have already been accepted. `sigfact' is the more important parameter of the two. The default value of 3.0 means that the automatic search can make the overall RMS of the fit somewhat worse, but it will give the program a fair go at finding some lines. Setting `sigfact' to 1 or less, which you may do with the `Modify' menu option, ensures that the automatic search will not make the fit worse, but it will probably not find many lines either.
Just how best to use the automatic line finder is a matter of experience and, probably, opinion. At the moment, it is a relatively new feature and so the experience is lacking, even if the opinion is not. It does, however, seem fair to say that the better the original fit, the more likely the automatic fit is to make correct rather than misleading identifications. The `Xauto' menu option, which causes all automatic line identifications to be deleted, at least means that you can experiment a little without doing irreparable damage to your fit.
One approach is to let the automatic fit loose, and then tidy up after it, deleting those lines that it found but that you don't like the look of. You can do this from the line list produced by the `Fit' menu option, using the `Edit' option to delete the lines affecting the fit the most. Lines found automatically are flagged in the list by a plus sign (and are shown in `arlines.lis' with a (A) symbol).
An alternative, and quite a simple operation, is to return to the interactive selection (the `Reselect' menu option) and examine the lines found. The lines found by the Auto option are displayed with their wavelengths in parentheses, so it is fairly straightforward to run through the spectrum with the cursor, hitting `d' at every bracketed line that looks like nothing more than a noise bump.
FIGARO A general data reduction system