Thu, 07/05/2018 - 03:53
I'm trying to look at an eclipsing binary that also has G dor pulsations. There are too many frequencies to pre-whiten and look at residuals.
If I model the out of eclipse data (short spans, excluding the eclipse), is there a way to show the values of the model for the excluded data points?
Duane
Sure. Here you go:
http://adsabs.harvard.edu/full/2000ASPC..210..423G
Hi Duane
I haven't read the paper Stephen pointed you to, but when you say:
that sounds a lot like: is there a way to show the residuals resulting from applying the model to the out of eclipse data.
Is that what you mean or something else?
David
Hello,
Stephen, thanks for the link. I haven't read that one, but my question is a more practical one.
I'll briefly give a little more background to my project.
I've carefully analyzed 15 quarters of kepler data using period04 . From this I found over 200 frequencies with S/N over 5 in the range 0.15d-1 to 13d-1. There is a sequence of evenly space frequencies that appeared to be orbital harmonics of a companion. A closer look at the lightcurve showed eclipses at the period (half period) of about 10 days predicted from the frequency spacing.
The variability range of the star is about 0.44 mmag and the eclipse drops are on the order of 6 mmag. It does't look as though there are alternating deep and shallow minima.
With my limited computing capability I looked at the residuals prewhiten by about 75 frequencies and there was still too much variability to try to characterize the eclipses.
Each eclipse is about 6 data pionts or roughly 3 hours. (I'm using the long cadence data). I constructed smaller data sets of 20 points on either side of the eclipse low point. Using VStar I exclude the eclipse and model the out of eclipse points using a high degree polynomial (usually 13 gives a nice fit). I then exculde all the out of eclipse points, leaving 6, and model the eclipse with a degree 5 polynomial (n-1). This gives a minimum point. I then evaluate the out of eclipse polynomial at the minimum point and take the difference between the minimum and the evaluated point to get an estimate of the drop. Using several quarters of consecutive eclipses I get about a 1mmag spread in drop values but no clear alternation.
Now, my question. I wanted to be able to look at the out of eclipse model to see if there was anything happening during the eclipse. I've suspected that there was a critical point in the area of the eclipse. VStar doesn't show any model points during the eclipse. For a couple of cases I've evaluated the polynomial at the excluded points to make a new data set and added the eclipse. I've also used the residuals and added back the eclipse by hand calculating the eclipse residuals. Since I would like to do this for an extended set of eclipses it is tedious to do.
I'm also attempting to look at frequency, amplitude, and phase modulation which may give, among other things, more orbital information.
Thanks for the responses. Brad, you can chime in any time.
Duane
Duane
It would help if VStar showed a continuous line as well as model data points; been on my list to add for quite awhile.
Page 70 of the user manual may help. You could look at the model in Excel or R since VStar provides the model function in each. I should add Python too. VeLa models ("function" tab in Models dialog) may be useful for this in future.
David
David,
I've been using the Excel functions for my difference values and to plot a few cases. Since I want to look at a sequence of 40 or so eclipses, I wanted to be lazy and not plot each one using that. With a large sequence I want to see if I can detect if there are alternating minima or that they are all statisically equal.
Does my technique seem valid?
Thanks,
Duane
Hi Duane
Seems reasonable.
If you had a plugin that allowed you to plot all 40 eclipses with a continuous line superimposed to show an arbitrary model in full, not just model points corresponding to observation times, would that help? What other information would you need (extrema, fit metrics)?
David
Hi David,
Showing the model with a continuous curve is what I was originally wanting.
Since the star is kind of wild, I have to model a small region around each eclipse with a different polynomial. I use the fit metrics and such already available on each. I'll attach a word doc showing some close ups of the lightcurve so you can see what I'm up against. The lightcurve is from the kepler fits files from quarter 4. The first is about a 25 day stretch which contains 3 eclipses. Even in the 25 day span the eclipses are hard to see. The next three are 3 consecutive eclipses in a closer view. (not all to the same scale)
As the lightcurve is so complicated, I think that the best I can do is to construct a file of residuals for each eclipse using Excel. I'll then paste them together and use the phase plot. First with the period given from the frequency spacing, then with double that.The combined filewill contain very large gaps but I don't think that will be a problem.
Thanks again,
Duane