[vsnet-computing 13] Re: period analysis

[Taichi Kato <tkato>]

Re: period analysis

> To me, it seems like several software packages are giving 
> quite different results. For the very same dataset, the DOS 
> executable (pdm.exe) I downloaded from the VSNET website 
> (and I thank  Taichi for pointing my attention there) gives 
> 0.05669d. So, I'm a bit confused on this issue. In the past 
> I had several cases where I did a reasonable period estimate 
> "by eyes". This time this operation looks more difficult, as 
> the maxima, during the same run, do not have a constant 
> separation (and this could explain why different software 
> gives different results?).

   I'm not sure about various software packages, but there have been known
"bugs" in this subject.  Of course, there may be additional sorts of
"unclarified" bugs in many softwares...

   Some softwares handle data with single-precision values, which lead
to the loss of siginificant digits when handling JDs.  Some softwares
don't subtract the average before period analysis (you may just do this
manually).  Some software have "buffer overrun" bugs when handling
a large number of data, and many, many...

   More subtle treatments involve the subtraction of trends of decline,
which becomes important in analyzing dwarf nova-type light curves.
This process is also prone to errors, since JDs are usually large numbers,
which will cause a loss of significant digits if one uses the simplest
implementation of linear fitting.

   Searching for the true minimum is another important issue.  Instead
of finding a minimum in mathematical sense, you had better fit some
function, or fold, the resultant theta diagram, as one does for determining
a minimum of eclipsing binary.  In doing this, the analysis in the frequency
domain is more convenient.  Your PDM results on your web look a bit noisy,
which may have caused this kind of scatter in minimum finding.

Regards,
Taichi Kato

VSNET Home Page

---

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