Forums / The Science / Data Analysis / Period analysis of eclipse so far!
Period analysis of eclipse so far!
Hi,
Just out of interest I have done a period analysis of the last three months of observations, both the visual observations and the Johnson V CCD observations, using Peranso [ http://www.peranso.com/ ].
I should add that I am new to period analysis and am still learning the methodology and terminology.
Using the ANOVA method over the period range 1h to 50 hours has thrown up the expected artefacts at approx 24h, with their associated harmonics of 48h, 12h, 6h and 3h. (The results are shown below.)
There are two results which are less easy to explain.
Firstly, each set of data shows peaks around periods of approximately 8 hours. My thinking is that this is due to the difference in longitude of the 'European' observers and the US observers, which are the largest concentrations of observers contributing to the data sets. Is this conceivable?
Secondly, there are periodicities for which I can think of no explanation. The visual data shows a peak at a periodicity of ~4.8 h, while the CCD data shows peaks at ~3.4h and 24.8h. Might these be just noise (hence the different values from the two data sets), or artefacts of some other systematic errors?
Regards,
David Conner.
=================
Peranso analysis of all CitizenSky visual observations of epsilon Aurigae for three months to Nov 4 2009 (ie from July 17)
Anova 1h to 50 h, resolution 10000
Freq time/h theta
0.04156 24.061598 42.00
0.04185 23.892579 37.15
0.08321 12.017786 35.83
0.12506 7.996418 25.21
0.08370 11.947431 24.48
0.06253 15.991812 24.07
0.02098 47.664442 19.80
0.12555 7.965208 18.57
0.06273 15.941844 13.40
0.12457 8.027873 11.18
0.16641 6.009182 9.99
0.06224 16.067354 8.84
0.33311 3.002011 7.80
0.33409 2.993205 7.62
0.33350 2.998483 7.47
0.20875 4.790465 7.12
0.16700 5.988024 6.99
0.02078 48.113934 6.90
0.04097 24.406912 6.48
0.41759 2.394716 6.48
Peranso analysis of all CitizenSky CCD Johnson V observations of epsilon Aurigae for three months to Nov 5 2009 (ie from July 18)
Anova 1h to 50 h, resolution 10000
Freq time/h theta
0.12492 8.005203 84.19
0.06245 16.013344 74.22
0.08340 11.990108 72.00
0.12541 7.973952 71.61
0.06274 15.938372 70.89
0.12570 7.955318 70.30
0.04179 23.930601 66.12
0.04140 24.157021 64.71
0.04159 24.043278 62.81
0.16761 5.966216 60.17
0.16673 5.997751 60.07
0.17045 5.866823 59.16
0.25446 3.929852 57.68
0.04247 23.544416 56.50
0.25143 3.977296 53.73
0.08516 11.741970 53.59
0.08370 11.948026 53.56
0.29559 3.383093 53.17
0.04032 24.802356 52.62
0.33603 2.975951 52.59
Hi David, I have done extensive work with Peranso and it is an excellent program. While there is much to be learned from the eclipse ingress, I'm afraid period analysis will not be of much value. To work you need something that is varying periodically during the time of the set of observations. During ingress you are viewing only a very small portion of one cycle (one cycle is 27 years). While it is possible to glean information from the ingress, getting a period is extremely difficult if not impossible. What you can determine are first and second contact points, the length of ingress and the depth of the eclipse (once past second contact). This information is very important and tells much about the star system. There is a quasi-periodic out-of-eclipse variation that has defied period analysis. I and several others have spent many hours trying to make sense out of it. Taking all the data to date will give a period, add a new season and the period changes. Do season by season analysis and the periods vary greatly. If you would like to experiment with the data let me know via email and I will send you a text file of the V photometric data from 2003 to 2009 and you can experiment with it using Peranso. More information, including suggestions and methods for analyzing the eclipse, on this can be found on the Campaign web site at http://www.hposoft.com/Campaign09.html Good luck! Jeff Counting Photons Hopkins Phoenix Observatory Phoenix, Arizona USA phxjeff@hposoft.com
In my opinion, period analysis of the ingress data is a valid endeavor. We are collecting much more precise data during this eclipse than any other eclipse. Periods that defied earlier analysis due to lack of coverage, sensitivity, etc. may be detected with a better data set - if they exist. The periods you found are indeed real - they are just not caused by the star. However, *if* epsilon Aurigae did have some astrophysical induced periods at short time scales then your technique may work. As long as the observations are not evenly distributed then you can detect short periods using long time spans of data that have gaps in it. Our data meets that criteria thanks to having lots of observers at different longitudes, observing at different times of nights, at different intervals, etc. This works because the uneven data catches different phases of the periods. As we collect more data the signal will rise further out of the noise as more gaps are filled in and confirmed. We know that there are lots of aliases of the daily cycle in the data. Using the TS program* you can remove these aliases and then look for other signals. I do not know if this is possible with Peranso, but I wouldn't be surprised if it is as Peranso is an excellent and powerful tool. It will likely be an option of Phase 2 of the VStar program (being programmed by one of the Citizen Sky teams). Phase 1 of VStar should be released in weeks and Phase 2 should be released before the data analysis workshop.At the AAVSO annual meeting today, Gary Billings presented a paper where he is using telephoto lenses on ST-7 cameras and getting precision photometry of epsilon Aurigae at ultra high temporal resolution for many hours each night he observes. His preliminary analysis doesn't reveal any periodicity "from minutes to hours". But he has some more calibration work to do to be completely sure. And he plans to continue to collect tons of data. He will upload his data to the web site when he is done with calibration so that anyone can use it. You may want to check back in later and rerun your analysis on his data when it is available. That reminds me, all CS data is free for download at any time using this form.*TS is a powerful time series analysis tool, but it is only available in QBASIC (runs on DOS and XP only) and Fortran (runs on any machine with a fortran compiler, such as OS X and most UNIX distros). Many/Most/All of its tools will likely be added to VStar. The Phase 2 specs of Vstar are currently being finalized.
Hi,Thanks for all the responses.It shows that this sort of analysis is as much an art as a science, if you'll excuse the phrase - you really need to be heads up about what the results actually mean!I appreciate that there are probably no physical causes to the periods that Peranso has highlighted, but given that there are so many aliases, how can we be sure that we aren't - in principle - discounting any real periodicity? Is a dataset with no gaps the only solution?More work to do with Peranso, and also VStar!As an aside, the wide range of estimates coming in is interesting. It's also interesting to note that my own estimates tend to be below the mean, even when deliberately trying to not mentally rush the eclipse along. There's so much going on here! Learning a lot already,Many thanks,David Conner.
Ingress is a fairly smooth time in the light curve and indeed approaches a straight line. Doing period analysis of a straight line certainly seems strange. What could possibly be seen?Aside from the 27.1 year period of the star system, there are out-of-eclipse variations that have been previously mentioned. These warrant more study, but are best studied out-of-eclipse and certainly NOT during ingress or egress.While the ingress may have some bumps, those are most likely due to the out-of-eclipse phenomena and it would be a very poor time to study or try to analyze them. The best time is out-of-eclipse. While it is certainly possible some of the ingress bumps are due to holes or gaps in the secondary, it seems unlikely and again period analysis would not help.Some observers have reported seeing short term flare activity of the system. I have seen strange things too on the order of seconds. These flares would most likely be missed visually or with CCD/DSLR photometry. There are efforts underway to do several all night observations using high resolution single channel photometry to look for variation in the range of seconds. Because of the extremely short integration times (in seconds) the fact that the system is in ingress (or any other phase) should have minor effect on that project. Because the flare activity appears to be non-periodic, period analysis will be of little or no value.Epsilon Aurigae offers many avenues for investigation, but analysis of the system is very different from most variable stars. What observers should be looking into is determining first contact time (not at all easy - try using linear regression to determine a slope), ingress time and second contact time along with eclipse depth. These parameters tell an important story about the star system and can be investigated by anyone from the published data.For example what does the ingress time and depth tell us. What if the ingress time was shorter or longer? I encourage observers and those interested, to think of some models. Use current thoughts on distance, orbital velocities, masses and sizes along with current observational data. Compare this with the last eclipse data.I suggest using the period analysis for studying the out-of-eclipse variations, but not to get caught up in trying to see things with it during the eclipse. Again, most variable star analysis techniques do not apply to epsilon Aurigae, but there are other techniques that can help present a picture of the star system and its parameters. These will be discussed in more detail on the Campaign web site and future Campaign Newsletters.JeffCounting PhotonsHopkins Phoenix ObservatoryPhoenix, Arizona USAphxjeff@hposoft.com
>Ingress is a fairly smooth time in the light curve and indeed approaches a straight line. Doing period >analysis of a straight line certainly seems >strange. What could possibly be seen? The thing is you don't know if it is smooth until you do the analysis. Visual inspection of the current light curve would never notice periods of less than a few days. Also, as stated before, we have a much better data set than in the past so we have the ability to detect periods that eluded past analysis. As for what could be seen, just look at the literature for exoplanet transits. There are plenty of physical mechanisms behind variability in ingress (rings, dust clouds, atmospheres, etc.) If it is a dust cloud, for example, then perhaps there could be gaps in the cloud allowing light to peak through. If accretion is going on, then there could be turbulence. If there is magnetism, then collimation. And since we know even less about this bizarre system there could be causes we haven't yet seen. I'm not suggesting or predicting any of these phenomenon. I'm just giving some examples of what could cause variation in ingress light curves. In fact, my prediction is the same as yours. I bet the slope remains smooth. But we should not assume that hypothesis - we should test it. And, ideally, many people should do the testing independently of each other. David's analysis has told us something we didn't know before. It told us that our data doesn't show any unexplained short term periodicity. Before he did the analysis, we didn't know that! You could assume it based on past eclipses, but that would not be scientific. Past eclipse data wasn't good enough to rule it out and, as you have astutely stated, there is lots of eclipse-to-eclipse variation in the light curve. As financial analysts like to say "past performance does not guarantee future performance".
Hi,One motivation for my doing a period analysis during ingress was the wide scatter in the reported visual magnitudes; a range of +- 0.2 mags being fairly typical. The Johnson V observations were done as a control, to see if there were any similarities or differences between the two sets of data.Since we have now eliminated any causes due to the system itself (periodic and otherwise), this scatter must have another cause. As a visual variable star observer of some years I have experienced some of the problems that can affect the accuracy of estimates. Indeed, my own tendency to 'rush' eclipses is rearing its ugly head in this project (as it has on some of my observations of other eclipsing binaries over the years)!There is a goldmine of information in this dataset. Perhaps an analysis of the observations submitted by the different participants of this project might show some systematic differences between observers.Regards,David Conner.
Perhaps an analysis of the observations submitted by the different participants of this project might show some systematic differences between observers.Wow, you really should come to the data analysis workshop and talk to Grant Foster. You two think alike. :) He had the same comment on Friday night while we were looking at your results. TS has a "bias observer" function that allows you to apply observer corrections. We made a graph that shows how this works using the epsilon Aurigae data and I intend to write a post about it soon.
Hi, David! This is an awesome first start, especially since we have no period analysis training or tutorials online yet. I'm currently typing this with Grant Foster looking over my shoulder. Grant is our statistics-guru-dude here at the AAVSO and wrote many of the period finding routines and software that variable star researchers often use. We think your suggestion that these are daily aliases is the correct one. The 8 hour, 4.8 hour and 3.4 hour aliases are all daily aliases (3, 5 and 7 cycles per day, respectively). Supporting that is that the magnitude of the periods are around 1.5 magnitudes, which is about the average difference in accuracy between observers. So the signals you are finding is really in the data. But they are artifacts of the observing process and not related to the star itself. A warning: You may want to detrend the data to remove the overall slope of the decline. This will remove the strongest signal and weaken its aliases. This can be done by subtracting a linear fit from the data and working with the residuals. Keep on playing with the data and looking for more stuff! For those new to the discussion, rest assured, we plan lots of tutorials on this. We will also hold a session on this at the 2nd Citizen Sky Workshop (which, hopefully, Grant will present!).