In 1993, a new method was developed for finding the periodicity of stellar light curves
using cubic splines. The technique is similar in philosophy to the "minimum string
length" approach and is therefore insensitive to the actual light curve shape. This
work was described extensively in an article published in The Astrophysical Journal **436**,
pages 787-794 (1994 December 1).

The program source code for implementing the cubic spline spectral analysis can be found in two directories accessible with a web browser:

Both of these directories contain ".doc" text files which should be read first. The /Spline directory contains all of the source code required to create a library of spline routines. /Spline/Samples contains the source code for various main programs which use the spline library to find the periodicity of unequally sampled light curves. A brief description can be found by clicking "poster paper".

For more information, contact Carl Akerlof, akerlof@umich.edu

(*from The Astrophysical Journal)*

C. Akerlof, C. Alcock, R. Allsman, T. Axelrod, D.P. Bennett, K.H. Cook, F. Freeman, K. Griest, S. Marshall, H-S. Park, S. Perlmutter, B. Peterson, P. Quinn, J. Reimann, A. Rodgers, C.W. Stubbs, and W. Sutherland

In the absence of *a priori* information, nonparametric statistical techniques are
often useful in exploring the structure of data. A least-squares fitting program, based on
Cubic B-splines has been developed to analyze the periodicity of variable star light
curves. This technique takes advantage of the limited domain within which a particular
B-spline is nonzero to substantially reduce the number of calculations needed to generate
the regression matrix. By using simple approximations adapted to modern computer
workstations, the computational speed is competitive with most other common methods that
have been described in the literature. Since the number of arithmetic operations increases
as N2, where N is the number of data points, this method cannot compete with the FFT
modification of the Lomb-Scargle algorithm. However, for data sets with N < 10**4, it
should be quite useful. Examples are shown, taken from the MACHO experiment.