Inversion of line profiles using principal component analysis: some practical guidelines

Observatoire de Paris, LESIA, FRE 2461 du CNRS, 92195 Meudon Cedex, France

Abstract

Modern spectro-polarimeters produce huge amounts of data, and the computing time needed to achieve data inversion is then becoming an important bottleneck for the modelling activity. Recently, an alternative to the traditional non-linear least squares technique has been proposed; it relies on an old technique in multivariate analysis, principal component analysis (PCA), and proves to be much faster than traditional techniques (a gain of two orders of magnitude in computation time is not unusual). We will briefly recall this technique and discuss some means of making it still more robust.