Fastspecfit V2.0 does not detect many Broad-Ha candidates that were originally detected in V1.0

desihub / fastspecfit

Fast spectral synthesis and emission-line fitting of DESI spectra.

https://fastspecfit.readthedocs.org

BSD 3-Clause "New" or "Revised" License

13 stars 2 forks source link

Fastspecfit V2.0 does not detect many Broad-Ha candidates that were originally detected in V1.0 #118

Closed Ragadeepika-Pucha closed 1 year ago

Ragadeepika-Pucha commented 1 year ago

I have compared broad-line selection between V1.0 vs V2.0 and found that many of the obvious broad-line sources, that were detected from V1.0 are not detected in V2.0. It would be useful to keep the RCHI2 difference over the Balmer line region only (like what was done in V1.0) or perhaps add it in addition to the full spectrum DELTA_LINERCHI2. Here are some obvious BL candidates that are selected in V1.0, but missed in V2.0:

These sources are selected with Broad Ha AoN >= 5 in V1.0 and having Broad Ha Flux = 0 in V2.0. The 15 sources are randomly selected in different stellar mass (from CIGALE) bins. The SPECPROD, TARGETID, SURVEY, PROGRAM, HEALPIX, and Z columns for these 15 example candidates are as follows:

Thank you!

Ragadeepika-Pucha commented 1 year ago

I made some figures from V1 and V2 -- comparing the Ha linewidth to the delta_rchi2 to access the parameter space that would be worth re-fitting the broad components.

Selection Criterion:

Narrow Ha AoN >= 3 in both V1 and V2
Broad Ha flux != 0 and Broad Ha AoN >= 3 from V1
Broad Ha flux == 0 in V2

From the right side plot, we propose to select the right-side cloud of galaxies to refit using broad-lines. Here is the proposed cut in "red":

I further selected some broad-line galaxies with Broad Ha AoN >= 5 from V1.0 and not detected in V2.0 and ran my emission-line fitting code. The galaxies that are selected as BL candidates from my code are shown as red points in this plot, which justifies the proposed cut.

Thank you.

moustakas commented 1 year ago

@Ragadeepika-Pucha @stephjuneau here's a comparison of the Halpha broad line-width in the Fuji/v1 VAC vs the imminent Fuji/v3 VAC (produced using tag 2.4.1) for the set of objects selected with:

I = (v1['HALPHA_BROAD_AMP'] * np.sqrt(v1['HALPHA_BROAD_AMP_IVAR']) > 5) * (v2['HALPHA_BROAD_AMP'] == 0.)

There are N=21,127 objects in this comparison sample; however, only 647 have non-zero broad Halpha in the v3 catalog the reason being that most of the remaining objects have broad-line widths <250 km/s, which is the cutoff in tag 2.4.1 for the broad-line model to be dropped. In v1, these objects survived because of the computational bug / inconsistency I describe in https://github.com/desihub/fastspecfit/pull/151.

Hopefully this is acceptable to you, but if not please let me know soon. Otherwise, my plan is to move forward with v3.

moustakas commented 1 year ago

@stephjuneau has signed off on the changes with the following off-list comments, which I received permission to post here (lightly edited):

Hey John, looking into this now. So your new cutoff corresponds to FWHM~590 km/s and one could 
convert this to a minimum BH mass given a range of Ha-broad luminosities. [snip] 

From the Reines+2013 paper, it seems their lowest FWHM for BL-AGN was about 630 km/s so above 
your cutoff. But they didn’t find any candidates with less than 10**5 Msun black holes. To some extent, 
this choice will be setting the limit in terms of which minimum mass can we achieve when searching for 
IMBHs with DESI. [snip]

It gets really challenging in any case at these lower widths because it starts to overlaps with the 
width of outflow components. So a general recipe will always struggle to cover all bases. But allowing 
broad components down to 590km/s FWHM is much more helpful than applying a stringent cut at 
FWHM>1000km/s or so!

Let's continue any discussion of these issues in new ticket(s).