I have digitized old data from CV Car that Sara included in the database. CV Car is an EA eclipsing binary, and I thought it would be fun to see if the period obtained through VSTAR would be the same or even better than the one derived in the pre-computer era.
The period in VSX is the same as the value given in the 1939 paper by Lukas Plaut: 14.414897 d. The more recent ASAS-3 observations fit this period nicely: http://www.astrouw.edu.pl/cgi-asas/asas_variable/104544-5755.0,asas3,14.414897,-24240.2700,500,0,0
In VSTAR, I scanned for a period between 14 and 15 days, with a resolution of 0.00001 and got as "biggest hit" this period:14.41887d, which gives a worst result when making the actual phase plot of the old and ASAS-3 data. What am I doing wrong? Or is the "manual 1939-method" just better than the one in VSTAR?
I think the DC DFT method is poor for finding the right period for eclipsing variables. I have noticed that for many stars I have worked on. Instead I use Peranso and one of the methods Lafler-Kinman, Dworetsky or Renson for eclipsers. For miras or cepheids I have got the best results with the ANOVA method.
I used Peranso and some of its methods on your data for CV Car and got the following "best" periods:
Lomb-Scargle 14.41825
DFT 14.41783
DCDFT 14.41887
CLEANest 14.41887
FALC 14.41570
ANOVA 14.41704
Dworetsky 14.41490
Renson 14.41477
PDM 14.41866
Lafler-Kinman 14.41474
Phase Binned AoV 14.41493
The periods from Dworetsky, Renson, Lafler-Kinman and AoV all show a nice and smooth phase curve and agree with the published period, so the correct period must be in that region and not the period from DC DFT. The next step would be to combine the old observations with the ASAS-3 data and see how they connect to get an even better value.
Thomas Karlsson
For the ASAS-3 data i got the following best periods:
Dworetsky 14.41530
Renson 14.41527
Lafler-Kinman 14.41527
For the combined data it is hard to say which period that is the best fit. The gap is to huge which mean that several values fits equal well, and it is also possible that the star have had changes in its period. But some values that could be considered are:
14.414732, 14.414818, 14.414903, 14.414989, 14.415075, 14.415161, 14.415247, 14.415332 and 14.415418
Hi Bruno, Thomas
Apologies for the delay in replying.
Yes, as you say Thomas, a Fourier method such as DCDFT does not seem ideal for period search in eclipsing binary data.
It's comforting to see that Peranso and VStar agree on the DCDFT result.
Also interesting is that only about 3 of the methods came close to the periods found in Plaut; he gives 3 on pages 26 and 27. Plaut used a least squares approach of some kind but doesn't go into detail.
I wrote a VStar plug-in called "AoV" which should be equivalent to Peranso's phase binned AoV. This is available from the Plug-in Manager in VStar. See:
The second link provides description, use, references.
For the AID CV Car data, with a period range of 14 to 15, a resolution of 0.001 days and number of bins set to 20, I get a top hit of 14.415 , very close to the Phase Binned AoV 14.41493 Peranso value you found. 14.415 gives a reasonable phase plot.
Increasing the resolution and number of bins over that range gave worse results so I widened the range: 13 to 16 days, 0.0001 resolution, 10 or 20 bins gives 14.4148 or 14.4149, respectively.
I find that getting the right combination of parameters (range, resolution, bins) takes some experimentation but the stats in the top-hits table give some guidance, as does a phase plot of course.
I can't look at the ASAS data at the moment because http://www.astrouw.edu.pl/cgi-asas/asas_cgi_get_data?104544-5755.0,asas3 gives an internal server error, as is sometimes the case with ASAS.
I want to write more period search plug-ins but with work and family committments, barely have enough time to keep VStar versions progressing. Other than the ones you mention in Peranso, there are some interesting ones that I'd like to make use of: Fast Chi-Square, PDM. I've spoken to the developer of the latter algorithm and he's happy for me to adapt it for VStar. I think I need to enlist the help of some students...
Thomas, one of the Peranso methods you show is ANOVA. Can you say how this differs from phase binned AoV?
David