PSF-ing and Chi

[hawcheng]

I have a question about PSF fitting and the chi statistic that is employed in Daophot and Allstar. Perhaps those of you who don't use Daophot or Allstar but do PSF fitting might also have some insight regarding this.

Basically, when I plot chi vs. raw magnitude for my stars, I find that instead of scattering about 1, it scatters about something less, typically 0.5 to 0.8.

The chi statistic that I am talking about is the ratio of the pixel-to-pixel residuals of the PSF fitting to the Poisson error. I have a hard time seeing how anything can beat Poisson statistics. Hence my concern at the chi.

And while I'm actually pretty confident in my photometry, I want to use chi to help me filter out bad measurements, and I'm not confident enough in such use if I don't understand why I'm getting the chi values I do.

Any insights would be appreciated.

[azangari]

The example that I've seen in the Davis '94 document suggested that chi could be lower because the star you were measuring was involved in creating the PSF and therefore, it wasn't statistically independent. I don't know if that helps you much.

I've got some chi trouble myself actually. I'm doing a variable PSF and I'm selecting lots of stars, many of them of moderate brightness. When I plot chi vs mag from an als file, I have chi values centered around 1.1-1.2 (yay?) for the dimmer stars, but the brightest stars (starting at an instrumental magnitude of about 13.5) have chi values that ramp upward in a way such that no bright star has a chi as good as the dimmer ones. I think I'm looking at some linearity issues. Unfortunately, there isn't a good way to get peak PSF values out of the .als files (correct me if I'm wrong, PLEASE), so I could peg, say anything over 35000 as officially garbage or what not.

I'm not sure what to do, model-PSF wise-- shoot for bright stars, with closer neighbors (and ignore the fact that neighbor subtraction WILL have to be smaller than the PSF), allow saturated stars or just keep with the neighbor-free moderate ones, or try to find some sort of correction.

Unfortunately, the target I'm interested in fitting, is among the brighter objects in the frame and the bright-not-saturated images have the best seeing.

Any advice?

[azangari]

@OP

Also, looking through the archives, it seems that you need to have your readnoise and gain parameters (in datapars) properly set as well. You can determine them from flats and biases using findgain.