Galaxy Spectra
The goals of this project are to derive physical parameters (e.g. age, metallicity, star formation history….) from a galaxy spectrum. We can do this by brute force maximum likelihood comparison with spectral population synthesis models, but we can use data compression methods to do the same job to the same accuracy much faster, and, in fact, much more reliably. This opens up the possibility of routine analysis of large surveys such as Sloan and 2dF.
Find b1 which captures most information about age.
Find b2 (orthogonal to b1) which captures most information about stellar mass… etc. [Solve generalised eigenvalue problem; Gram-Schmidt orthogonalisation in 1000-dimensional (curved!) space.]
Compress data by forming a few linear combinations y=b.x
(Heavens, Jimenez, Lahav 2000)