COMMISSION 27 OF THE I. A. U. INFORMATION BULLETIN ON VARIABLE STARS Number 1347 Konkoly Observatory Budapest 1977 October 10 PRACTICAL FORMULAE TO SEARCH PERIODICITIES IN VARIABLE STARS The distribution of the dates in the series of observations of variable stars are often very irregular. The spacing between consecutive dates may range from hours to weeks or even months. Such unlikeness forbids the use of classical methods to search periodicities, or, at least, makes its use very hazardous. Nevertheless, the use of the fast Fourier transform is almost generalized among astronomers under the pretext that the results obtained are fairly good. In fact we obtained some good results in the study of some BY Dra stars observed at the I.T.A. Astronomical Observatory before 1974. However this has not been always true and practical questions arose when comparing some completely different results obtained for HD 319139 using the fast Fourier transform or the Lafler-Kinman method. The light curve of BY Dra stars are characterized by small amplitude and great noise, and in 1975 we adopted a new technique to deal with series where the dates are arbitrarily spaced. This technique is a natural generalization of the fast Fourier transform and gives more accurate estimates. In the figure, we show the periodogram obtained for 21 V-magnitudes of the star HD 319139 observed by C.A.O. Torres and I.C. Busko at Cerro Tololo in 1974 (solid line); we also compare it to the periodogram obtained with the classical fast Fourier transform (dashed line). Recently, in a review paper presented in the Colloquium on Close Binary Stars held in San Paulo we have shown the way in which the technique may be generalized to more complex cases like those were the data consist of N pieces of continuous records or where the data have different precisions. A typical example of problem involving unevenly spaced data of different precisions is the problem of the search of periodicities (and beats) in irregular, semi-regular or long-period variable stars. In this case the measurement I(w) of the power spectrum at the trial frequency omega is given by the expressions in the next page. [FIGURE 1] x_j=2(Pi)(omega)t_j (a_0)^-2 = Summa w_j (a_1)^-2 = Summa w_j (cos^2 x_j) - (a_0)^2(Summa w_j cos x_j)^2 (a_2)^-2 = Summa w_j (sin^2 x_j) - (a_0)^2 (Summa w_j sin^2 x_j)^2 - (a_1)^2 (Summa w_j cos x_j sin x_j)^2 - (a_0)^4 (a_1)^2 (Summa w_j cos x_j)^2 *Summa w_j sin x_j )^2 + 2(a_0)^2 (a_1)^2 (Summa w_j cos x_j) (Summa w_j sin x_j) (Summa w_j cos x_j sin x_j) c_1 = a_2 Summa w_j f(t_j) cos x_j c_2 = a_2 Summa w_j f(t_j) sin x_j - a_1 a_2 c_1 (Summa w_j cos x_j sin x_j - (a_0)^2 Summa w_j cos x_j Summa w_j sin x_j) I(omega) = (c_1)^2 + (c_2)^2 In these formulae t_j are the observation dates and w_j the weights of the observations. The measures f(t_j) are referred to their mean value, i.e., they are shifted in a way such that Summa w_j f(t ) = 0. All other symbols are intrinsic quantities. The light curves under study may be formed by several waves, not necessarily harmonic (as in Blazhko effect). In such cases, before searching for a secondary period, it is necessary to filter the time-series from the main wave. If omega_1 is the main frequency, the filtered time-series is given by g(t_j) = f(t_j) - d(t_j) where d(t_j) = d0 + d_1 cos x_j + d_2 sin x_j (x_j = 2 Pi omega t_j) and d_0 = - (a_0)^2 ( d_1 Summa w_j cos x_j + d_2 Summa w_j sin x_j) d_1 = a_1 c_1 + a_2 (a_1)^2 c_2 ( (a_0)^2 Summa w_j cos x_j Summa w_k sin x_k - Summa w_j cos x_j sin x_j ) d_2 = a_2 c_2 The secondary frequencies are searched in the periodogram of the filtered series g(t_j). The epoch of the minimum in the main wave is one of the two roots of tg x_0 = d_2/d_1. S. FERRAZ-MELLO Universidade de Sao Paulo Departamento de Astronomia (Brazil)