(Prof. Inese Ivans' research group)
Tuesday, April 10, 2012
12:30pm (JFB Library)
Title: Squirrels in the Spectra: Dealing with unmodeled effects in spectral data for chemical composition determination of stars
In principle the spectrum of a star contains nearly all the information that can be learned about it. In order to extract the most information from a stellar spectrum it is necessary to model radiative transfer through the stellar atmosphere. I will discuss the limitations of modeled stellar spectra and talk about known effects which limit our ability to model the spectrum well. Stellar chemical abundance analysis consists of comparing observed absorption features in the spectrum with predictions based on the number density of the associated element in the atmosphere. Each transition between atomic or molecular energy levels has the potential to give rise to a feature in the observed spectrum. Of the millions of known atomic transitions only a relatively select few have had their transition probabilities measured in the laboratory with high precision. Theoretical calculations of transitions can give answers which are wrong by orders of magnitude. This creates a situation in which the majority of the features in a stellar spectrum either have properties which are poorly known (and therefore can only be poorly modeled) or in some cases completely unknown. Computational constraints and poorly known atomic physics data create a situation in which the error distribution of observed and modeled spectra is non-gaussian. The typical approach to avoiding this problem is to attempt to use only features whose transition probabilities are well known. However because of the density of weak features throughout a stellar spectrum this is only partially effective since there may be contamination from many poorly understood features which are too weak to resolve as they are on the order the noise in the data. I will discuss detecting and dealing with non-gaussian error distributions and analyze the actual error distribution found when comparing modeled spectra to observed spectra.