Fast fitting of spectral lines with Gaussian and hyperfine structure models

Context. The fitting of spectral lines is a common step in the analysis of line observations and simulations. However, the observational noise, the presence of multiple velocity components, and potentially large data sets make it a non-trivial task. Aims. We present a new computer program Spectrum Iterative Fitter (SPIF) for the fitting of spectra with Gaussians or with hyperfine line profiles. The aim is to show the computational efficiency of the program and to use it to examine the general accuracy of approximating spectra with simple models. Methods. We describe the implementation of the program. To characterise its performance, we examined spectra with isolated Gaussian components or a hyperfine structure, also using synthetic observations from numerical simulations of interstellar clouds. We examined the search for the globally optimal fit and the accuracy to which single-velocity-component and multi-component fits recover true values for parameters such as line areas, velocity dispersion, and optical depth. Results. The program is shown to be fast, with fits of single Gaussian components reaching on graphics processing units speeds approaching one million spectra per second. This also makes it feasible to use Monte Carlo simulations or Markov chain Monte Carlo calculations for the error estimation. However, in the case of hyperfine structure lines, degeneracies affect the parameter estimation and can complicate the derivation of the error estimates. Conclusions. The use of many random initial values makes the fits more robust, both for locating the global χ 2 minimum and for the selection of the optimal number of velocity components.