Open changhoonhahn opened 4 years ago
I examined the impact of inst. SFR = 0 galaxies on the observables in nb/sfr_res_issues.ipynb.
Galaxies with inst. SFR = 0 has a distinct feature in the color-mag and UV color-mag diagrams: The DEM currently does not attenuate galaxies with input SFR = 0 so these features are independent of the DEM parameters.
Even if we resample inst SFR between [0, SFR_res], the SFR = 0 galaxies are clustered in the observables:
I'll try running the ABC again with resampled SFRs, but I suspect DEM alone will not be able to "fix" this. More importantly, I don't think we want DEM to deal with this since getting rid of this feature will dominate the DEM parameter constraints. So @TStarkenburg I think we need to generate SEDs where the SFH is resampled to smooth out the features. What do you think?
I think this has the same conclusion as #39: the simulations predict too many quiescent galaxies (this can also be seen in other papers). Many of these SFR = 0 galaxies have no current SFR (instantaneous SFR doesn't have the resolution issues noted above), no recent SFR, and no gas. We will add flexibility to the DEM for the SFR = 0 galaxies. A more flexible DEM will probably not 'fix' this completely, but will give us some insight into whether these SFR = 0 galaxies should have dust, and we don't want to fundamentally change what the simulation predict.
Instead of sampling A_V separately for the SFR=0 galaxies, I figured we could directly sample the observables for the SFR=0 galaxies based on the SDSS quiescent population. That way we effectively remove SFR=0 galaxies from consideration[^1] in our inference.
I looked at the observables for quiescent galaxies (based on SSFR cut) in this notebook. Seems the simplest thing to do would be to sample (G-R) and (FUV-NUV) color for SFR=0 galaxies from the overall p(G-R) and p(FUV-NUV) distributions of SDSS quiescent galaxies:
[^1] this assumes that SFR~0 and Mgas~0 SDSS galaxies are not a special subpopulation of quiescent galaxies.
pesky resolution effect strikes again.
SFR resolution matters for galaxies with SFR=0 now that we have a log SFR dependence 😠At the moment, I uniformly assign a SFR between [0, SFR resolution] for SFR=0 galaxies. We may need to think more carefully about this(?).
@TStarkenburg what's the SFR resolution for TNG?