Yukiko's room

It is difficult to estimate the most suitable domain on the linear regression for totally warped, but partially linearizable data. The solution to this problem influences, especially, the subject for the characterization of the power spectral densities (PSD) of fluctuations in, for example, the physical and natural penomena. This characteristic can be found as the slope of the regression line which is fitted for the log-log PSD. However, the general forms of the spectral components are not linear as a whole, and we have to make a decision the domain on the regression by ourselves before fitting. In this paper, we investigate a way for an automatic estimation of such a domain, and give a numerical experiment and applications to some observed data, namely, the set of time series of the fluctuations in the intervals between the human blinks.