-------------------------------- TEST_4 : Analysis of the results -------------------------------- F. Mignard 05/12/05 -------------------------------- These tests are specifically designed for the analysis of variable stars from the Gaia osbervations. The run is based on a simulation of the observing sequence and the variable stars simulated are typical RR-Lyr or Cepheids (simulation_rr_lyr) or eclipsing binaries (simulation_eb) The aim of these tests is not to prove the full efficiency of the method for the variable stars, but to demonstrate how to use FAMOUS for this purpose and to show that for the easy cases it performs well. Periods are searched between 3h and 1000 days (this is a very important parameter for the CPU). There are three outputs for each simulation : - a summary report giving the star id, S/N, true period, estimated period and the normalised difference. - the general report giving the detail of the run for each star (settings + run report) - the folded light curves when flfold = true in the setting. The driver is famous_driver_photom and read two input files : - the setting file as usual - the data file called rendez_vous_rr_lyr.txt or rendez_vous_eb.txt with the observations organised star by star with a header giving the number of observations, the simulated periods etc. - the version used in these tests has been optimised for the search of periodic signal in a time series. Comments on the results: period_rr_lyr.txt - with a S/N = 5 in the simulation (the definition is not universal, but here with S/N = 5, this means that the amplitude/sigma ~ 15 (amplitude peak to peak and not the half amplitude). These are then the easy cases, but the goal is just to show how to run the analyser. All the periods are recovered and the reduced errors (d(period)/estimated rms) do not show anomalous values. plots : two light curves for the 1st and 5nd star. running time : on the average with ~ 200 observations ( Astro + MBP ) the solution on a standard laptop comes with a runtime of 1.5s per star. It is directly proportional to the number of observations and to the range of frequencies where the search is performed : limiting the lowest periods to 3h instead of 2h decreases the runtime in the same proportion. period_eb.txt - with a S/N = 5 in the simulation ( but here with S/N = 5, this means that the amplitude/sigma ~ 30 . Again these are the easy cases, and in practice FAMOUS is not very good to search the period of Eclipsing binaries since the power is very diluted in the different harmonics. In addition when the secondary dip is more that about 0.1 the main one, the first harmonic (half the period) has more power than the fundamental. This is properly handled here in the 9th star which shows this effect. From a more extensive ru on similar simulation I have found that about 80% of the periods are found. Using the ratio between the frequency of the two lines (it must be very close to 2), one has an hint as to the period found is reliable or dubious. This indicator seems very efficient and its statistical properties should be ascertained with a more extensive Monte Carlo simulation. plots : Light curves for the 5th and the 9th run time : 0.2 s/star.