Comparison of modified black-body fits for the estimation of dust optical depths in interstellar clouds

TL;DR

This study compares single and multi-component modified black-body models for estimating dust optical depths in interstellar clouds, highlighting trade-offs between accuracy and computational cost and suggesting improvements for large-scale mapping.

Contribution

It systematically evaluates the performance of single versus multi-component MBB models in fitting dust SEDs, including their accuracy and computational feasibility.

Findings

Multi-component models reduce systematic errors compared to single MBB.

Single MBB remains valid with empirical bias corrections.

Fitting multi-component models is feasible for large pixel maps.

Abstract

When dust far-infrared spectral energy distributions (SEDs) are fitted with a single modified black body (MBB), the optical depths tend to be underestimated. This is caused by temperature variations, and fits with several temperature components could lead to smaller errors. We want to quantify the performance of the standard model of a single MBB in comparison with some multi-component models. We are interested in both the accuracy and computational cost. We examine some cloud models relevant for interstellar medium studies. Synthetic spectra are fitted with a single MBB, a sum of several MBBs, and a sum of fixed spectral templates, but keeping the dust opacity spectral index fixed. When observations are used at their native resolution, the beam convolution becomes part of the fitting procedure. This increases the computational cost, but the analysis of large maps is still feasible with direct optimisation or even with Markov chain Monte Carlo methods. Compared to the single MBB fits, multi-component models can show significantly smaller systematic errors, at the cost of more statistical noise. The $\chi^2$ values of the fits are not a good indicator of the accuracy of the $\tau$ estimates, due to the potentially dominant role of the model errors. The single-MBB model also remains a valid alternative if combined with empirical corrections to reduce its bias. It is technically feasible to fit multi-component models to maps of millions of pixels. However, the SED model and the priors need to be selected carefully, and the model errors can only be estimated by comparing alternative models.