meeting 2025 10 06 gw - JacobPilawa/TriaxSchwarzschild_wiki_6 GitHub Wiki

Context

  • Working a bit more on validating the DESI catalog in partciular against some other sources.

Siudek+24 Information

  • We're ultimately concerned with understanding how much we can trust our stellar masses from the combined Zou+24 catalogs. To that end, we have the Siudek+24 Value-Added Catalog, and in the last meeting, we had some questions about how they arrived at this particular sample. I've gathered that infomration, as well as some exploratory plots, for the Siudek+24 data:

  • Background/Sample Selection: the sample that one can download from Siudek+24 contains 1,337,250 sources.

    • Dec 2020: DESI begins 5 month survey validation with three phases: SV1, SV2, SV3

      • SV1 = validating target selection of the five main primary target classes (Milky Way Suryve, Bright Galaxy Survey, luminous red galaxies, emission line galaxies, and quasars).
      • SV2 = operation developments;
      • SV3 = "One-Percent Survey", optimized the observing procedures with very high fiber assignments and completness over 200 deg^2, (1% of the final DESI footprint)
      • Collectively, SV1/2/3 became known as "Fuji" and are the basis for the Siudek VAC.

    • DESI Spectra are processed with a fully automatic redshift estimation pipeline (Redrock). This produces a catalog with:

      • Z, ZERR, ZWARN, SPECTYPE to every source.
      • Note they also selected "GALAXY" or "QSO" sources from Redrock, and only sources which did not have fiber issues, but DO have a reliable redshift.
      • After these cuts, they are left with 1,345,137 objects.

    • Photometry is then acquired (and only including targets with all the photometry), brings this down to 1 286 124 unique objects.

      • It's not completely clear to me how to get this exact number of sources just yet, but I'll investigate this a bit further.

  • Regardless, it really seems like in large part the SV1/SV2/SV3 targets are quite uniformly sampled/it doesn't appear there were any location/parameter cuts that would bias the sample in one way or the other. With that said, here are some diagnostics from the Siudek+24 catalog alone (full catalog.dat without any additional cuts):

Cuts? Siudek+24 Diagnostics
None images/251006/diagnostics_all.png
Chi2<15 images/251006/diagnostics_cut.png
  • I also wanted to get a sense of what subsets of the data look like, errors and all, so here's logM vs. z with errors included for 1/250th of the full data:
logM vs z
images/251006/logM_vs_z_with_errors.png

Matching and Comparing with Siudek+24

  • We first wanted to compare the Zou+24 DESI LS9 results with the Siudek+24 VAC results:
    • Note that Siudek already has a small comparison in their paper between their results and Zou+24 (Table 4 and their Appendix C3).
    • Note that both the Siudek and Zou catalogs rely on CIGALE fits with the same SSP models and IMF, but they different in their prescriptions: Zou et al (2024) leave stellar metallicity as a free parameter for example, whereas Siudek fixes it to the solar value. Another example that's listed is in their treatment of AGN: Siudek's catalog accounts fo rthe AGN templates, whereas Zou et al (2024) do not.
    • I've extended their comparison to go beyond the subset of galaxies presented in Siudek+24 (seems to be limited to a subset of AGN sources):

Matching Diagnostics

  • The first task is to match the full Zou+24 DESI catalog with the targets from Fuji. There really should be a better way to do this (at least it seems to me), but it really seems like our best approach is matching based on RA and DEC. I've done a first pass at that matching, and here are some of those diagnostics.

  • The process:

    • The Zou+24 catalog comes as both a full file and split by RA. I've taken the approach to:
      1. Select the VAC in bins of 10 degrees of RA (with 0.01 degree buffer on either side)
      2. With that subset of the VAC, compute separations between the VAC coordinates and the corresponding RA bin from the Zou+24 data.
      3. Find the closest match (2d separation in arcsec) between the VAC susbet and the DESI subset, and store that data.

  • The results from the matching can be seen here:

    • Note that I can get a pretty similar matched fraction to the one that Siudek claims; though they don't specify how they did they matching so it's a bit hard to compare. Note that my matching criteria is currently 0.1 arcsec.
Matched Diagnostics
images/251006/match_sep_diagnostics.png

Redshift Comparisons

  • Both the Zou+22 and Siudek+24 catalogs quote a photo-z from their pipelines which we can compare to the specz measurements (if they exist). In this sense, we might get a sense of which pipeline is able to better reproduce the specz measurements.
  • From the cross-matched catalog, about 70,163 galaxies have a specz measurement from a different catalog. I have plotted the photo-z from Siudek and from Zou against these specz values here:
    • In general, it seems like the Redrock pipeline which runs on the DESI EDR data does an astonishing job of reproducing the specz value's that Zou compiled. A large majority of the points fall almost perfectly on the 1-to-1 line.
    • I've also included the dispersion, as defined by Zou+22, in each panel (technically its +/- 3*dispersion in these panels).
Redshift Comparisons
images/251006/siudek_zou_redshift_comparisons.png

Mass Comparisons

  • We can now start to dig a bit into the agreement of the stellar mass aross catalogs/with various criteria. I've also computed the dispersion of the residuals (in a similar style to the z calculation above) (the number in the title is the standard deviation of the residuals, since it's not exactly in the same form as the redshift calculation from above).
    • Overall, in particular for the subset of galaxies with specz measurements, the Zou and Siudek catalogs are in very nice agreement.
Mass Comparison
images/251006/logM_comparison.png
Trends in Mass Difference?
  • I was curious what might be driving the difference in stellar mass estimates between the Siudek and Zou catalogs. For example, are these particularly bad fits (in terms of chi2? in terms of stellar mass errors? in terms of other quantities?)
    • I've also included some local median and local dispersion trend lines for visualization. Seems like the trend with photo_z_err is the strongest, at least of the things I've checked. This makes sense and is in line with what Zou+22 and Zou+19 say.
Some diagnostics
logM vs. chi2 images/251006/mass_diff_v_chi2.png
logM vs. logM_err images/251006/mass_diff_v_e_logM.png
logM vs. photoz_err images/251006/mass_diff_v_photo_z_err.png
"Cleanest" Sample
  • One thing I was curious about -- if I take some of the most strict quality cuts across all the data, what does the resulting dispersion look like in mass (if the dispersions above are at the ~0.25 dex level?).
  • To do this, I've taken the cross-matched value-added catalog (~1.3 million sources) and did the following cuts:
    • matched sources <0.1 arcsec
    • 0.0001 < z < 1 for both Siudek's z and DESI_photo_z (Zou)
    • DESI_photo_zerr < 0.1*(1+DESI_photo_z) [following Zou+22]
    • valid masses (logM>0) from Siudek
    • chi2 from Siudek < 5
  • The result is ~517k sources, down from the 1.3 million. Still a lot! The redshift and mass comparisons are thus:
Redshift Stellar Masses
[images/251006/CLEAN_z.png]]](/JacobPilawa/TriaxSchwarzschild_wiki_6/wiki/[[images/251006/CLEAN_logM.png)
  • One thing I want to emphasize is the log scaling on the mass comparison. I'm currently showing all data in that plot, but you can get a better sense of the agreement if I only plot cells which have >N galaxies in that cell:
Min Count Plot
N=1 images/251006/CLEAN_logM_1.png
N=5 images/251006/CLEAN_logM_5.png
N=10 images/251006/CLEAN_logM_10.png
N=15 images/251006/CLEAN_logM_15.png
N=25 images/251006/CLEAN_logM_25.png
  • I've also plotted the agreement for a few different redshift bins:
Z Bin Plot
0–0.1 images/251006/CLEAN_logM_z0.0_0.1.png
0.1–0.2 images/251006/CLEAN_logM_z0.1_0.2.png
0.2–0.3 images/251006/CLEAN_logM_z0.2_0.3.png
0.3–0.4 images/251006/CLEAN_logM_z0.3_0.4.png
0.4–0.5 images/251006/CLEAN_logM_z0.4_0.5.png
0.5–0.6 images/251006/CLEAN_logM_z0.5_0.6.png
0.6–0.7 images/251006/CLEAN_logM_z0.6_0.7.png
0.7–0.8 images/251006/CLEAN_logM_z0.7_0.8.png
0.8–0.9 images/251006/CLEAN_logM_z0.8_0.9.png
0.9–1.0 images/251006/CLEAN_logM_z0.9_1.0.png

Matching with WISE2MBH Catalog

  • We had briefly discussed this toward the end of the last meeting, but the ultimate goal is to use the stellar masses and redshifts, so it'd be helpful to compare our stellar masses with some other catalogs, including something like WISE or 2MASS (or other very large sky surveys).

  • I've started with what should be one of the easier cases since the stellar masses are already published in WISE2MBH: a scaling-based algorithm for probing supermassive black hole masses through WISE catalogues.

    • At a very high level, they determined stellar masses via the process outlined in Culver+14:
    • W1 Luminosity is correlated with stellar masses
    • W1-W2 color is correlated with mass-to-light ratio
    • Combined W1 luminosity with W1-W2 colors allow for stellar mass estimation
    • Quoted accuracies are ~25% for galaxies >1e9 Msun, and this process seems to be more or less the standard for converting from WISE photometry to stellar masses
    • They then extend this to infer T types and thus bulge masses, leading to their final Mbh sample

  • I've taken the DESI DR9 (Zou+22) sample and tried my best to cross match the masses (though the cross matching takes a very long time, so only have a bit of progress so far!). Here's what these results look like:

    • Once again doing the matching simply based on coordinates with a 0.3 arcsec matching radius.
    • Note I've done NO additional cuts on the data for quality, so this is likely a "worst case" agreement.
    • It seems like there's a standard deviation of ~0.25 dex in the recovered mass. This is quite consistent with what we're seeing between Siudek and Zou (see above).
Zou vs. WISE2MBH Full Sample Zou vs. WISE2MBH (mincnt=10)
[images/251006/zou_vs_wise2mbh_full.png]]](/JacobPilawa/TriaxSchwarzschild_wiki_6/wiki/[[images/251006/zou_vs_wise2mbh_min10.png)

Comparison with Leja

  • One other small detaill - we had wanted to see the Leja+20 (z=0) GSMF on top of my current results, which is here. I've also computed this for two different redshift bins just to get a sense of how things shift around.
z=[0,1] z=[0,0.1]
[images/251006/vmax_method_combined_desi_z_0_1.png]]](/JacobPilawa/TriaxSchwarzschild_wiki_6/wiki/[[images/251006/vmax_method_combined_desi_z_0_0.1.png)