[vsnet-chat 229] Re: PDM troubles

[Taichi Kato <tkato>]

Dear Rudolf Novak and all,

   My opinion is that the PDM statistic (theta) is not suited for estimating
the _error_ of the period, since theta statistic is based on F-statistic,
which tests the null hypothesis of non-significant difference between phase
bins -- this statistic uses no phase information of individual bins.
(However, the phase information is implicitly reflected to the shape of
theta(frequency) ).

   I rather use residuals to traditional least squares fit to observed times
of certain phases (maxima, minima, mid-rising/fading points, maximum
correlation etc.) to estimate the error.  The timing error of superhump
maxima is usually an order of 0.001 to 0.002 day.  The baseline of 100
cycles (typically less than a week) would then yield a typical error of
0.000001 to 0.000002 day, assuming no variation in period and superhump
profile.  A more sophisticated way is to "add" random noise to the observed
or template light curve and then to apply PDM to find how the best period
drifts.  Choosing the random noise as to reflect the observational errors
(photon noise etc.), the resultant standard deviation of the period would
approximate (internal) 1-sigma error.

Regards,
Taichi Kato

VSNET Home Page

Return to Daisaku Nogami

---

vsnet-adm@kusastro.kyoto-u.ac.jp