Speaker
Description
Spectroscopic data reduction is confounded by unknowns such as the sky spectrum and the instrumental transfer function. It is important to marginalize over these (and not merely subtract a point estimate) in order to obtain unbiased results with correct error bars. When the priors on the nuisance components can be expressed as a Gaussian in the high-dimension pixel space (i.e., an Npix x Npix covariance matrix) the marginalization integral is analytic. Andrew Saydjari and I have pioneered an approach called Marginalized Analytic Data-space Gaussian Inference for Component Separation (MADGICS) and obtained promising results from the APOGEE data. In particular, MADGICS allows easy separation of multiple objects within the fiber, leading to a more complete catalog of binary stars.
