This web page is a repository for stuff relating to the O-type Binaries in Clusters project, involving Kathy Eastwood (NAU), Nidia Morrell (Carnegie), Phil Massey (Lowell), Doug Gies (GSU), and Laura Penny (College of Charleston), plus students Erin Darnell (Columbia) and Yelena Tsitkin (MIT).
The plan is to find masses for high-mass stars using eclipsing detached binaries. Step 1 is to find the eclipsing systems. Here are the clusters we're working on, along with finding charts and references: The Clusters
We will search for variability using data from several different telescopes. Here is the general philosophy:
The actual steps and links to the software:
We are making available a tar ball for the Y4KCam reductions and photometry. Enjoy.
The automatic photometry is in some ways the EASIEST part of all this. Nearly identical routines are used for the two instruments, and we describe them both here.
First, the various tasks have to be defined in your loginuser.cl Then the main calling program is autoswope.cl or y4kphot.cl Each program does the following:
The differences are: (1) the gain is extracted from the header for the Swope data, and explictly set to 1.5 in the case of the Y4KCam data, (2) datamax=saturation - 500 for the Swope and -1500 for the Y4KCam data, and (3) the "best" fwhm has to be handled somewhat differently in the two cases, as the seeing has been significantly worse on the Y4kCam data. In either case the frame is characterized, and values are put into the header so that the photometry can proceed. Also, the characterization of y4kcam data has been complicated by the fact that sometimes the frames are blank (surprise!) or the stars are donuts, so there has been some better error-handling put into the y4kcam stuff. The following is done as part of "character" (all flavors)
Here's how I did it...
When done with all filters, ran "color" and captured the output as "vubr.dat" Note that if you only have two filters (B and V, say) this still works.
fort.7: B-V, vi-v, d fort.8: U-B, ui-u, d fort.9: B-V, bi-b, d fort.10: V-R, ri-r, d MASSEY ET AL 1989 v-V=3.616-0.252(B-V) rms=0.05 u-U=5.958-0.249(U-B) rms=0.05 b-B=4.423-0.114(B-V) rms=0.05 Schmidt data to only the instrumental < 19. FINAL: v-V=3.747 -0.05(B-V) rms 0.05 mag vs 3.616 0.13 u-U=6.155 -0.21(U-B) rms 0.08 mag vs 5.958 0.20 b-B-4.513 -0.08(B-V) rms 0.07 mag vs 4.423 0.09 r-R=3.707 -0.04(V-R) rms 0.06 mag For N602c, did NOT restrict the magnitude range and still got rms of 0.03! When we did restrict it it did get cleaner, but no real difference in vals v-V=3.620-0.077(B-V) rms0.04 mag u-U=6.157-0.161(U-B) rms 0.05 mag b-B=4.156-0.074(B-V) rms 0.04 mag r-R=3.558-0.011(V-R) rms 0.06 mag For R136, the solutions are Truely Lousy, with large rms... Helped once I brought the sky in: v-V=3.808-0.075(B-V) rms=0.09 b-B=3.827-0.103(B-V) rms=0.07 Very tempted to adopt "old" photometry when we can.... N1910: deleted >19 and then: v-V=3.683-0.074(B-V) rms 0.09 mag u-U=6.063-0.101(U-B) rms 0.10 mag b-B=3.821-0.058(B-V) rms 0.08 mag and editing out > 18.... r-R=3.399-0.041(V-R) rms 0.10 mag N6611: EXCELLENT FITS... v-V=3.480+0.071(B-V) rms 0.04 mag u-U=5.747-0.236(U-B) rms 0.06mag b-B=3.584-0.073(B-V) rms 0.05 mag
How do we identify eclipsing binaries? Last summer undergraduate Erin Roye and I developed software to go through the photometry of each star and note variables. The criteria included
The problem is that none of these criteria work at detecting low-amplitude systems, such as R136-007, where the aplitudes are extremely low, such as in this example:
The "external" error (standard deviation of all the photometry) is only 0.03 mag. So, E/I=1.6, if I assume 0.02 mag for the internal error. The latter is reasonable, as the internal error is more than just the statistical uncertainty of the observation, since the zero-points are adjusted for each image.
An alternative way is to make use of the additional information that eclipses will be periodic, and so to look only for PERIODIC light variations. Mercedes Lopez-Morales argued this in her thesis analysis of a very large data set (looking for low-mass eclipsing systems, and the demonstrations she makes are quite convincing: see Lopez-Morales & Clemens 2004 PASP, 116, 22. She adopted the Schwarzenberg-Czerny (1989 MNRAS 241, 153) "analysis of variance" (AOV) code to produce periodgrams of star. Ones that show statistically significant spikes are probably winners. I've tried this on the first set of 50 R136 stars, and the results were quite interesting: the KNOWN eclipsing systems pop while most just show noise. One useful diaognostic is that we see harmonics of the correct period. Thus in a frequency periodogram if we see a statistically significant spike at frequence nu, and another spike at 2*nu, we might be suspicious that we have a winner.
Here are some examples:
The two largest spikes occur at frequencies of 0.146320 days^-1 (=6.8344 days) and 0.29282 days^-1 (=3.4151 days), where the centers have been determined by the centroid of the three highest points on a spike. So, we suspect that the period is 6.834 days. A Lafler-Kinman search between periods of 1-30 days on the same data yields a best period of 6.8282 days. We would never know this even WAS a light-variable if we were relying upon E/I or other criteria that did not take advantage of the periodic nature.
The two strongest spikes occur at 0.55369 days^-1 (1.80606 days), and 1.10730 days^-1 (0.9031 days), again differing by a factor of 2. A Lafler-Kinman search on the same data comes up with a period of 1.806 days.
But using only the first set of Swope data I thought I found a nice eclipse with a period of 31.6 days. What does the periodogram show? Not a thing!:
I don't think this is an eclipsing system.
Again, the two largest spikes happen to have a ratio of 2, with frequencies of 0.29507 days^-1 (3.389 days, same as what we found in our published paper) and 0.5903 days^-1 (1.6940 days). A Lafler-Kinman search fails to find any period!
Of course, now the trick is to do this for about 10,000 stars....
All automation programs feed off the "allstars" file
(C/P) * Dp/P = 0.005, or Dp/P^2 = 0.005 / C.
In frequence space, this is actually easier to consider. How well do we need to know nu? dnu = 1/P^2 dP, and so dnu = 0.005 / C. C is roughly 600 days, so...dnu = 8e-6. Call it 1e-5. But, that means we raov now geneates a STARNAME.res file containing 200,000 points, not 4000!
I packaged these all together in "period.cl", which also executes the igi files and generates IRAF "metacode" that can be examined with gkimosaic. Here's a sample output: