For this reason I created an "IRAC Test Pattern" optimized for this task. This consists of a grid of point sources all of which are exactly 1 mJy in brightness and which are located on a euclidean grid exactly 72 pixels apart. It eliminates many of the problems encountered in analyzing more realistic simulated data like that generated for IRAC areal galaxy surveys. Specifically, the points are all bright enough that they can be easily recognized by any automated extraction software and extracted without worrying about algorithmic effects on the completeness and reliability. They are all separated enough that confusion is irrelevant, and large extraction apertures can be used if desired. Finally, since they are all the same flux, it is easy to quickly characterize the extracted population in terms of means and uncertainties.

The sims were generated using an observing strategy similar to that of SWIRE. Namely, a cycling dither pattern, 30 second frames, and two dithered frames per map pointing. A 3x3 map was actually used, since the SWIRE AORs are much larger than needed for this. Because there are dithers and the spacing of the stars does not beat against the map spacing, the stars are observed at many different locations on the array. The sims were generated in several different "flavors": raw (all instrumental signatures), bcd (just noise), and a test set that just contains the dark and flat signatures.

The data was reduced by subtracting the actual dark and then dividing by the flat-field derived previously via the mask and coadd approach. The chosen flat-field was the 2.5 hour field. The resulting reduced data was then analyzed with the imaginitively named SExtractor. Sextractor indicates uncertainties in the extraction (based on noise statistics) of 0.2%.

IRAC test pattern with dark and flat signatures present. Shown are channels 1-4.

Data reduction. From left to right: raw, dark-subtracted, and flat-fielded data.

Locations on the array sampled in this test by artifical stars.

The resulting 204 extracted fluxes have a 1-sigma standard deviation from their mean value of 0.85%. The maximum deviations are +/-2.2%. The histogram of the percent errors are shown in the figure below. Overall, this result could have been predicted entirely from the difference between the derived flat and the flat-field used by the ISDS. This is shown in red below. The IRAC psf, particularly in band 1, is not much more than a few pixels in area. As a result, the errors in point source fluxes are likely to be very similar to the pixel-scale errors in the flat itself - there is not very much averaging over a wide area. Also, the distribution is a little "fatter" than would have been expected purely from the gaussian pixel-wise error distribution. This is because the errors are also convolved with an additional error term, namely the errors on larger spatial scales.

Thus, we can consider the errors derived in previous experiments indicative of what we can expect to get in terms of extracted point source accuracy. The above flat yields a result that is within 1%, 1-sigma.

Histogram of % difference relative to mean flux for 204 stars (black). Also shown in red is the distribution of % differences for each pixel in the flat relative to the known flat used in the simulation.