LSD - a nonlinear approach

摘要

Stellar spectra usually are very limited in the signal-to-noise ratio (SNR) that can beobtained. In order to increase their informative value, different techniques have been developed inthe past ten years which combine multiple spectral lines by cutting out individual line profiles andanalysing them by means of least-squares errors or PCA. They usually neglect that the bulk of linesare blended, resulting in artificial broadening of the retrieved common line pattern, while those thatcare to disentangle blended profiles assume linear line adding. Based on the well-known least-squares deconvolution (LSD) method, we developed a new tech-nique, which truly accounts for blended profiles and deconvolves them in a physically meaningfulway by taking into account the nonlinearity when abandoning the regime of optically thin lines. Theso-called interpolation formula by M. Minnaert is a unique tool to describe a line profile both in theoptically thin and the optically thick regime. It enables us to write a total line depth in terms of a(nonlinear) combination of contributing individual components. Applying different versions of LSD, among them our nonlinear (NL-) LSD, to simulated atomicand molecular intensity spectra shows the unrivaled functionality of our new method in terms ofinterpretability of the retrieved common line pattern. For the first time it is possible to recover anintrinsic line pattern from a molecular band.