…
26/July/2018 – Fakes news on STARLIGHT
Ge, Yan, Cappellari, Mao, Li and Lu (2018) have published a paper in MNRAS reporting crazy results of spectral fitting experiments with STARLIGHT, despite being explicitly told that all the biases they found in their tests with artificial data were due to a futile and irrelevant technicality. For a complete rebuttal of their results see CF2018. In that rebuttal article, a revised version of the code is used. This version is available below, but read on before thinking of downloading it.
We do not recommend you use this revised version! Why? Well, the new version is not better nor faster. In fact it can be slower and in some cases lead to worse results. Secondly, as shown in CF2018, it only differs from the original code in extreme situations where the results would already be suspicious anyway.
The only difference between this version and the original public distribution of STARLIGHT is a single line of code! The original version limits A_V to < 1 mag upon initialization of the Markov chains, while this revised version initializes the chains using A_V < AV_upp, where AV_upp is the highest extinction allowed for in the configuration file. As demonstrated in CF2018 this does not cause any problem in real life situations. The biases reported by Ge et al are triggered by large values of A_V, which may cause problems in some extreme cases. However, if you are dealing with a very dusty source then you should be much more worried about other things, like the over-simplistic treatment of dust attenuation in STARLIGHT (and most other codes too).
No one ever said STARLIGHT is the best nor the fastest or the most elegant full spectral fitting code in town. Informed users are in fact well aware of its several limitations and drawbacks. The A_V < 1 initialization thing which lead to the catastrophic results by Ge et al, however, does not even qualify as a bug, as its effects are only noticed under unrealistic conditions.
In short, we recommend you ignore this revision and wait for the upcoming distribution of the new version of STARLIGHT capable of simultaneously handling spectral and photometric constraints (eg, SDSS+GALEX).
If despite this you still think you need the revised code, here is a menu of executables:
