fix for stellar mass bias; more constrained broad+narrow fitting; and higher possible stellar vdisp

moustakas commented 7 months ago

This PR should address #159, #160, and #164.

Understanding the stellar mass bias required doing a deep dive into the choice of SPS models and priors used. I have large parts of a notebook written to show the results, but here are the main changes:

The choice of stellar library and isochrones clearly impacts the masses at the ~0.1 dex level. In particular, the C3K+MIST models used in FastSpecFit generally result in higher masses than the BC03+Padova models commonly used.
An even bigger effect is the choice of priors. As discussed by Leja et al. 2019, choosing a non-parametric star-formation history (SFH) can result in stellar masses which are easily 0.2-0.3 dex larger compared with parametric techniques (e.g., delayed $\tau$). However, I have been using far too many templates (i.e., free parameters) both in terms of ages and metallicities which are not well-constrained given the data.

With the preceding in mind, I've opted for a new template set (1.3.0; see here) which has just 40 models consisting of:

5 SPS ages (vs 8 previously) with mean ages 15 Myr, 65 Myr, 590 Myr, 6.4 Gyr, 12.7 Gyr;
solar metallicity only (vs three metallicity bins previously).

I tested these choices fairly extensively, including against doing the SPS inference "the slow way" using Prospector, and I'm pretty happy with the results. The FastSpecFit stellar masses will still be 0.1-0.25 dex higher than other existing catalogs (e.g., from Hu, Mar, and probably SDSS), but this is consistent with my choice of stellar library, ischrones, and priors and is also not inconsistent with published work by Leja et al. and others.

I also fixed a silly bug in how the light-weighted ages, dust attenuations, and SFRs were being computed.

Finally, to go along for the ride, I have an incorporated more constraints on the narrow Balmer+helium lines when fitting objects with broad lines (#160) and a velocity dispersion grid which extends all the way up to 500 km/s (#164).

github-actions[bot] commented 7 months ago