(1) Problem with flux error bars. The flux error bars for actively deblended sources were found to be grossly underestimated, as shown in Figure 4c of Laurent's page . Examination of previous results with synthetic data showed that this problem had existed before, and had not been properly addressed. The problem appeared to be present only for the secondary components, and became worse for the brighter sources. Since the latter are PSF error dominated, this suggested that the problem was connected with the noise model used in active deblending. This turned out to be the case; the contribution of the secondary PSF error had accidentally been omitted. This error was rectified, and the active deblending algorithm in prophot was re-tested with synthetic data. The results are shown in:

http://spider.ipac.caltech.edu/staff/kam/2mass/deblending/act_syn.html#Apr10

and show substantial improvement over the previous ones.

The color-color plots were regenerated after rerunning prophot on the same data as before (971116n), and the results are shown in Figures 7a, 7b, and 7c of Laurent's page . It is apparent that the performance of the deblending algorithm has improved substantially -- not only has the halo of outliers disappeared, but also, the flux errors are now well represented by the error bars.

Also on that page (after Fig 7) are some animated gifs of the color-color plots as a function of delta(mag) and separation.


(2) Flux bias from not deblending. This is plotted as a function of flux ratio and separation in:

http://spider.ipac.caltech.edu/staff/kam/2mass/deblending/repeat.html#Sept29

The key plots to look at are:

Figures 1a,2a,3a: Flux bias as a function of pri-sec magnitude difference

Figures 1b,2b,3b: Flux bias as a function of component separation.


(3) Position repeatability. See:

http://spider.ipac.caltech.edu/staff/kam/2mass/deblending/repeat.html#Apr26


(4) Selection of magnitude thresholds.

The selection of appropriate values for the magnitude cutoffs for active deblending is facilitated by an analysis of chi squared values from version 2 processing, carried out by J. Rho, and presented in:

http://spider.ipac.caltech.edu/staff/rho/2mass/psfchi/psfchi.html

Magnitude cutoffs for active deblending can be assessed from plots of the fraction of sources whose chi squared values are below a set of specified thresholds, as a function of magnitude. Particularly useful are the plots which show a comparision between the results for low and high latitude sources (b=0 and b=89 deg), since these plots show most clearly the effects of confusion at low latitudes. They are presented for a chi squared cutoff of 2.0 at J-band, H-band, and K-band. An appropriate value for the magnitude cutoff at each band corresponds to the magnitude at which the source fraction rises back up towards 1.0, since that represents the point at which the signal to noise ratio is too low for the chi squared to be a discriminant of single versus blended sources. Based on these considerations, appropriate values for the magnitude thresholds would be 14.5, 14.0, and 13.0, at J, H, and K, respectively. The band-to-band differences of these magnitudes are remarkably close to those of the nominal point source sensitivities (15.8, 15.1, and 14.3), which lends credence to our interpretation of the above plots. A small adjustment of the H-band cutoff would make the three magnitude cutoffs consistent with a single value of signal to noise. The appropriate values of the cutoffs are then: 14.5, 13.8, and 13.0 at J, H, and K, respectively.

After having said all that, it is probably not necessary to impose magnitude cutoffs at all, since the chi squared cutoff is implicitly magnitude-limiting. This can be seen from the above plots, which demonstrate that at large magnitudes, there is a decrease in the number of sources for which the chi squared exceeds 2.0 (note that the latter is close to the threshold of 1.8 used in the decision whether to make a deblend attempt). Quantitatively, if we imposed no magnitude cutoffs at all, the above plots, in conjunction with the total source counts as a function of magnitude, suggest that the number of attempted deblends would increase by less than 10%.

Not only does it not appear necessary to impose magnitude cutoffs, there is an actual disadvantage in imposing them, since they would adversely affect source count statistics. Specifically, a magnitude cutoff would produce a discontinuity in the source counts as a function of magnitude. My recommendation, therefore is not to impose magnitude limits in active deblending, but rather to rely on the chi squared threshold completely.