Non-parametric stellar LOSVD analysis

Ill-posed inverse problems are common in astronomy, and their solutions are unstable with respect to noise in the data. Solutions of such problems are typically found using two classes of methods: parametrization and fitting the data against some predefined function or a solution with a non-parametrical function using regularization. Here we are focusing on the latter non-parametric approach applied for the recovery of complex stellar line-of-sight velocity distribution (LOSVD) from the observed galaxy spectra. Development of such an approach is crucial for galaxies hosting multiple kinematically misaligned stellar components, such as 2 stellar counter-rotating disks, thin and thick disks, kinematically decoupled cores, and others. Stellar LOSVD recovery from the observed galaxy spectra is equivalent to a deconvolution and can be solved as a linear inverse problem. To overcome its ill-posed nature we apply smoothing regularization. Searching for an optimal degree of smoothing regularization is a challenging part of this approach. Here we present a non-parametric fitting technique, discuss its potential caveats, perform numerous tests based on synthetic mock spectra, and show real-world application to MaNGA spectral data cubes and some long-slit spectra of stellar counter-rotating galaxies. GitHub repository: https://github.com/gasymovdf/sla
Comment: 4 pages, 3 figure; to appear in the proceedings of the XXXI Astronomical Data Analysis Software and Systems (ADASS) conference (published by ASP); python pip package https://pypi.org/project/sla/