4. Time delay estimation
Sliding the light curves against each other to find the best visual agreement lead us to an estimate of about half a year for the time delay A-B (component B leading) with a brightness difference of about 1.7 mag. For such a time delay, the observing periods of the one component coincide more or less with the seasonal gaps in the light curve of the other. This effect without doubt produces a bias for values around , and we had to find a quantitative method of estimation that is insensitive to such biases.
The dispersion method of Pelt et al. (1994; 1996) proved to be quite robust despite the windowing effects in the case of time delay determination for the double quasar 0957+561. To minimise the bias for our data, we used the dispersion (Pelt et al. 1996) which takes into account not just neighbouring data but uses the complete light curves with weight factors corresponding to the (shifted) time differences of two observations each. If the decorrelation length is chosen sufficiently large, the windowing effect becomes increasingly less significant. Fig. 3 shows the dispersion for different values of over an interval of a priori possible time delays.
The global minima of all dispersion curves are consistently located between 0.6 and 0.8 years, which indicates a value of the time delay in this range. Using a decorrelation length of 100 days which reduces the windowing but does not smooth the data on longer time scales, the minimum is at with magnitude difference of 1.70 for the weighted data. Note that the existence of slightly weaker local minima still permit a somewhat smaller time delay of 0.3-0.5 yrs. On the other hand, a of one year or more is not consistent with the data.
This estimate of a time delay is to be seen as a very preliminary result. A detailed analysis of the light curves has to wait until better sampled data are available.
© European Southern Observatory (ESO) 1998
Online publication: October 22, 1998