Telescope Reviews: Ensemble photometry techniques?

I've been thinking thru this question a bit today, and, on a high level, I think this is the direction we have to head brian, if we are to reduce that 5.5mmag number we've kind of bumped into.

Now, the process I've been thinking thru is just 'thoughts on the principles' as I've been driving around town doing various errands, but, I think it's valid.

Looking at what we do now for V-C1, in reality, there are 4 terms in there. V comes with n1 (uncertainty of n, or, more precisely, the amount of noise in V), and C1 comes with n2, another uncertainty value. The scatter we see in our plots is essentially n1+n2. So, now lets look at each individually, and, suddenly there is a way to think this thru conceptually.

Our V-C1 is actually V-C1+n1+n2. n1+n2 is the uncertainty we have, and, the thing we want to minimize at this point. Since n2 comes from a single star, and it's the target star, on a single frame, there is nothing we can do about n2, but, we can do something about n1. it's no different than stacking frames in time to reduce final image noise, here we want to reduce the noise component in the check star. So, we 'stack check stars'.

So, if we are using the same annulus and aperture for each of our check stars, it stands to reason, simple addition of c1+c2+c3 in reality is c1+c2+c2+3n. But, if we do it just a little differently, and instead, make an artificial aperture, and an artificial annulus. Into the artificial aperture we put the sum of 10 check stars, raw data, and into the artificial annulus we put the sum of the 10 sky backgrounds, raw data. We've essentially created an artificial 'check star' which has a much higher snr than any one individual check star. So, take these sums now, divide by the number of stars in them, and now carry the calcs thru with _that_ as the ref star value.

This should reduce the noise in the 'check star' by a factor of sqrt(numstars). So, back to our original data, 5.5mmag scatter, that's 2.75 from the target, and 2.75 from the check. If the check star is actually 10 stars all glued together into the artificial star, that 2.75 should become 2.75/sqrt(10) or about 0.86. So, we _should_ now see the final scatter become 2.75+0.86 = 3.6, a reduction of 1.9.

This ofc only holds true if the full 5.5 scatter we are seeing is due to noise equally amongst the check and target star.

It should be possible to see if this holds, by simply taking data from one of the existing transits, generate the output from aip4win in raw form, then in the spreadsheet to the V-c1 calcs, but also do the ensemble calcs, and generate the V-ensemble plot at the same time.

Then again, maybe I've missed an important detail here, a little bit punchy today, sat up for the LCROSS non-event last nite, and only got about 4 hours sleep after that.