The degeneracy between dust colour temperature and spectral index. Comparison of methods for estimating the beta(T) relation

TL;DR

This paper compares various statistical methods for estimating the intrinsic beta(T) relation in dust emission, highlighting biases and advantages of Bayesian and hierarchical models over traditional chi^2 fitting.

Contribution

It systematically evaluates multiple methods, including Bayesian and hierarchical models, for separating intrinsic dust properties from noise-induced correlations in sub-millimetre observations.

Findings

Bayesian and hierarchical models show lower bias than chi^2 fitting.

All methods exhibit some bias, especially at high noise levels.

Hierarchical models bias estimates towards high S/N data, affecting the beta(T) relation.

Abstract

Sub-millimetre dust emission provides information on the physics of interstellar clouds and dust. Noise can produce anticorrelation between the colour temperature T_C and the spectral index beta. This must be separated from the intrinsic beta(T) relation of dust. We compare methods for the analysis of the beta(T) relation. We examine sub-millimetre observations simulated as simple modified black body emission or using 3D radiative transfer modelling. In addition to chi^2 fitting, we examine the results of the SIMEX method, basic Bayesian model, hierarchical models, and one method that explicitly assumes a functional form for beta(T). All methods exhibit some bias. Bayesian method shows significantly lower bias than direct chi^2 fits. The same is true for hierarchical models that also result in a smaller scatter in the temperature and spectral index values. However, significant bias was observed in cases with high noise levels. Beta and T estimates of the hierarchical model are biased towards the relation determined by the data with the highest S/N ratio. This can alter the recovered beta(T) function. With the method where we explicitly assume a functional form for the beta(T) relation, the bias is similar to the Bayesian method. In the case of an actual Herschel field, all methods agree, showing some degree of anticorrelation between T and beta. The Bayesian method and the hierarchical models can both reduce the noise-induced parameter correlations. However, all methods can exhibit non-negligible bias. This is particularly true for hierarchical models and observations of varying signal-to-noise ratios and must be taken into account when interpreting the results.