Abstract
I present the results of an investigation into the spectral properties of stars and galaxies. Firstly
I investigate methods for automatic MKK classification of stellar spectra, providing both a comparative study of some of the standard methods of automatic classification and a demonstration
of a state-of-the-art machine learning technique — Support Vector M achines. Using this technique I obtain a classification accuracy of δ = 1.7.
One of the limitations in the classification of stellar spectra is the lack of good training
data at high resolution. With this and also the application of population synthesis in mind, I
present a high resolution (λ/Δλ = 250000) library of 6410 synthetic stellar spectra which I
have generated from the Kurucz model atmospheres. The library covers the wavelength range
3000 - 10 000 with 54 values of effective temperature in the range 5250 - 50 000 K, 11 values
of log surface gravity between 0.0 and 5.0 and 19 metallicities in the range - 5 .0 to 1.0. By
com paring the new synthetic spectra with the STELIB library of observed spectra (Le Borgne
et al., 2003) I demonstrate their suitability for the application of population synthesis.
I then extend this library by supplementing the Kurucz spectra with other synthetic spectra,
to form a library for population synthesis similar to that of Lejeune et al. (1998) but at higher
resolution (2 ). I also investigate two methods of empirical population synthesis however I find
that even with modern computational resources these methods are not suitable for the number
and size of current spectra.
Finally I measure the Lick indices for a sample of Sloan Digital Sky Survey spectra and use
these in conjunction with the 2dF groups catalogue to investigate the change in these parameters
with the local density of galaxies. I find no strong trends in any of the Lick indices with group
velocity dispersion.
