Report for Simulation #6

The bottom line is that we get 99% completeness, and efficiency of 54%. These bottom line numbers seem to be a constant of the simulation, since we have to adjust the parameters to exclude basically all of the stars, and the majority of the QSOs are quite separated from the locus. We missed three QSOs, all of which are in the 2.5 < z < 3.5 range. In this range, the completeness drops to 86%. Of the selected candidates, 54% are QSOs, 15% are NELGs, 4% are white dwarfs, and 27% are galactic stars. We targeted less than one percent of the galactic stars. We could eliminate some of these galacitic stars by eliminating bright (15th magnitude) A stars. This would require extending the bright locus, and has not yet been done.

The new algorithm handles stars which are not detected in all bands. It also uses the individual errors in the measurements. Using the individual errors did not help much for this simulation, since the errors were photon noise only. The addition of errors is expected to be of greatest benefit when errors are enlarged by blending with neighbors. I have not yet implemented code to calculate the locus given that the input data has errors; fixing this will make it a little easier to generate the loci, but will probably not change the results much for this simulation.

I have successfully run the fifth simulation, using the sixth simulation as a training set. The sixth simulation used u exposures which were 10 times longer than usual. The reduced errors made it possible to fit the locus for all but the faintest bin (19 < i < 20), which still had errors large enough to make a slightly weird fit at the very red end. If necessary, the algorithm could have been made to work on this. Since the locus does not move much as a function of magnitude in the simulation, this did not help much. There is evidence that this might happen, though, in the real data.

Even though I could now fit the red end of the locus, I had to edit the locus files to increase the width of the locus at the red end. This is because there are *so many* stars out there, and *no* QSOs, that we bring in an enormous number of stars if we use the same number of sigma limits out there. I did not have to edit the blue end.

New in this version of the target selection code, the blue end is described by a half ellipsoid which fits onto the end of the curved elliptical cylinder which describes the locus. Before, the elliptical cylinder ended in a plane.

You will see in the u-g vs g-r plot that there is one QSO about half way down the locus which looks like it is well enough separated from the locus that it should have been targetted. However, the error in the u measurement is 0.228 magnitudes. The three sigma error cut moves that QSO easily within the locus, so it is not flagged. There is another QSO near there which is closer to the locus, but brighter. With the smaller errors, this QSO can be distinguished from the stars.

This is the figure of merit:

                                        # targets/#objects brighter than 20. # objects

Targets selected (by type):     1       59/6558 6558
                                2       47/4700 4700
                                3       0/0     0
                                4       0/0     0
                                5       0/0     0
                                6       0/0     0
                                7       0/0     0
                                8       57/58   58
                                9       14/16   16
                                10      210/213 213
                                        -----------------
                                        387/11545       11545

QSO completeness (z):           -10 - 1 85/85   85
                                1 - 1.5 49/49   49
                                1.5 - 2 32/32   32
                                2 - 2.5 24/24   24
                                2.5 - 3 14/16   16
                                3 - 3.5 5/6     6
                                3.5 - 4 1/1     1
                                4 - 6   0/0     0
                                6 - 10  0/0     0
                                        -----------------
                                        210/213 213

QSO completeness (m):           -10 - 15        0/0     0
                                15 - 16 1/1     1
                                16 - 17 1/1     1
                                17 - 18 20/20   20
                                18 - 19 91/92   92
                                19 - 20 97/99   99
                                20 - 30 0/0     0
                                        -----------------
                                        210/213 213