This section deals with the most frequently asked question about MIA and EWS. We will give some hints what you might check and what you can do, but we do not have the definitive answer. Reducing MIDI data is not trivial (even for enlightened pan-dimensional beings) and there are many things that can go wrong in the programs.
But first, you should celebrate. If the results of MIA and EWS are different, it means that at least one of the results is wrong, so you should be happy you noticed before you published it.
Second, do not expect too much precision from MIDI. Uncertainties of 10 to 15% are considered normal. Even larger differences occur with faint sources, where it's hard to obtain a reliable measurement of the photometric flux and the fringe amplitude.
If you still believe you have unusual differences, run MIA with the EWS mask. The mask EWS uses is in one of the files set with the environment variables prismhmask, prismsmask, grismhmask, and grismsmask. The mask is picked depending on whether the prism or grism and HIGH_SENSE- or SCI_PHOT-mode was used. You can specify the mask used by MIA when you call xmdv.
Next, make sure the universe is properly balanced and run EWS with the MIA mask. MIA creates its mask based on the data and stores it in the directory PHOTOMETRY. The filename is "Mask_" and the date and time of the photometric files.
If the results depend on the mask, but not on the program used, then one of the masks is not appropriate for the data. It may be that MIA has trouble finding the right spectra in the photometric files. It is also possible that the spectra are not exactly where they are supposed to be, so the predefined mask used by EWS does not pick up the flux. Your mission, should you choose to accept it, is to find out which mask is better for your data. This is the subject of ongoing discussions among the MIDI experts, as you can tell by the totally different strategies used by MIA and EWS.
Ok, you are still reading, so the results of MIA and EWS are different no matter which mask is used. Now you should check whether the difference is in the photometric (uncorrelated) flux or in the correlated flux. For this, it is useful to look at the plots produced by EWS in the final step, and by MIA if you call x->print_results.
The photometry should not be different, since MIA and EWS use the same program to measure it. It is possible, however, to override this in MIA with the NOEWSPHOT-switch, which may or may not be a good idea. The alternative code does not do a very good background subtraction, so its results for faint sources are almost certainly wrong. However, this should not matter for bright sources, and sometimes funny things happen to the background, which can cause EWS to go crazy.
When you compare the correlated fluxes, keep in mind that MIA by default bins three pixels together. Set the binning to one pixel (with x->set_lambda_bins) and try again. Of course, this lowers the signal-to-noise ratio.
Now you have to check if the programs (and their users!) worked correctly. The things to look out for are different for both programs:
In MIA, make sure the scans used to estimate the noise did not pick up signal from the fringes. This is not visible in the white light power spectra, but only in those of the individual spectral channels (in the plots you get if you click on the lower left plot in the GUI). The example in the section about the GUI shows how they should not look like, since there the signal in the noise scans is almost as high as the signal in the "good" scans.
Of course, you should also set a good threshold for the "good" scans. There is no simple recipe to do this, or else we would have implemented it in the code.
If both programs worked flawlessly and you still can't figure out why the results are different, then it is probably caused by a bug in the code. We recommend finding it, fixing it, and sending us a patch. You can be sure of the eternal gratitude of the MIA+EWS developers and all its users.