VAC Overview 

Description 

The following pages describe the content and construction of the FastSpecFit value-added catalogs (VACs). For a complete list of available VACs organized by data release, please see the VAC index.

Note

A DESI data release has been publicly released whereas a data assembly is only available internally to DESI collaboration members; not all data assemblies become data releases.

Data Organization 

Within the data release directory of each VAC, there are two key subdirectories, healpix and catalogs, which we now describe in more detail.

We run FastSpecFit on the healpix-coadded DESI spectra, and organize the files identically to how the spectra, redshift catalogs, and other data products are organized in the DESI data releases and assemblies (as documented here). In other words, relative to the VAC data release directory VACDIR, the individual per-healpix files can be found at the following location:

$VACDIR/healpix/SURVEY/PROGRAM/HPIXGROUP/HEALPIX/{fastspec,fastphot}-SURVEY-PROGRAM-HEALPIX.fits{.gz}

where SURVEY, PROGRAM, HPIXGROUP, and HEALPIX are fully documented here.

Note

The fastspec healpix catalogs are gzipped because they contain the best-fitting model spectra in addition to the fitting results (see the fastspec data model). The fastphot healpix catalogs are not gzipped because they do not include a MODELS extension (see the fastphot data model).

Merged Catalogs 

Most users will be interested in the merged FastSpecFit catalogs, which combine all the individual healpix catalogs for a given SURVEY and PROGRAM into a single file. Relative to the top-level data release directory VACDIR, the merged catalog filenames have the following syntax:

$VACDIR/catalogs/{fastspec,fastphot}-SURVEY-PROGRAM.fits

Note

In order to keep the size of the files reasonable, the merged fastspec catalogs do not contain the MODELS FITS extension (see the fastspec data model page for a description of this FITS extension). The fastphot merged catalogs also omit the MODELS extension, which is not produced by fastphot.

Note

For large survey-program combinations (e.g., main-bright and main-dark), the merged catalogs are further subdivided by nside=1 healpixel to keep individual file sizes manageable. In that case the filenames follow the pattern:

{fastspec,fastphot}-SURVEY-PROGRAM-nside1-hpNN.fits

where NN runs from 00 to 11, corresponding to the 12 nside=1 healpixels covering the full sky. Individual VAC pages document which survey-program combinations are split in this way.

Sample Selection 

The sample selection criteria were chosen to be very inclusive so that modeling results would be available for as many objects as possible.

In brief, we fit all extragalactic (redshift greater than \(10^{-3}\)), non-sky targets with no cuts on targeting bits, redshift or fiber-assignment warning bits other than a simple cut which throws out spectra with no data.

Specifically, let redrockfile be the full pathname to a given Redrock catalog. The following bit of Python code illustrates which targets we fit:

import fitsio
from fastspecfit.util import ZWarningMask

zb = fitsio.read(redrockfile, 'REDSHIFTS')
fm = fitsio.read(redrockfile, 'FIBERMAP')

I = (zb['Z'] > 1e-3) & (fm['OBJTYPE'] == 'TGT') & (zb['ZWARN'] & ZWarningMask.NODATA == 0)

Here, the ZWarningMask.NODATA bit indicates a spectrum which contains no data (all inverse variance pixel values in the extracted spectrum are zero). Navigate to the DESI ZWARN bit definitions documentation for more details.

Updated QSO Redshifts 

For a small but important fraction of quasar (QSO) targets, the redshift determined by Redrock is incorrect. To mitigate this issue, the DESI team has developed an approach to rectify the redshift nominally measured by Redrock using the machine-learning algorithm QuasarNet.

We update the Redrock redshift Z for affected QSO targets using the QuasarNet catalog and MgII catalog afterburners distributed alongside each Redrock output file. The original Redrock redshift and ZWARNING bitmask are preserved in the Z_RR and ZWARN_RR output columns (see the fastspec data model); note that only Z is modified, not ZWARN.

The update is applied separately to two classes of targets:

Primary QSO targets — objects assigned to a QSO targeting bit (or the equivalent SV0_QSO / MINI_SV_QSO commissioning bits for CMX data) — have their redshift corrected when two conditions are met: the QuasarNet afterburner independently classifies the spectrum as a QSO (IS_QSO_QN_NEW_RR = True), and the maximum QuasarNet line confidence across the six lines Lyα, C IV, C III], Mg II, Hβ, and Hα exceeds a threshold. For DR2 (Loa) this threshold is 0.99; for DR1 (Iron) it was 0.95.

Variable WISE QSO secondary targets (WISE_VAR_QSO secondary targeting bit, present in main-survey and some SV programs but not CMX data) are updated under a broader criterion: IS_QSO_QN_NEW_RR must be set and at least one of the following must hold — the Redrock spectral type is QSO, the MgII afterburner identifies the object as a QSO (IS_QSO_MGII), or the maximum QuasarNet line confidence exceeds the threshold above.

Issues 

For questions (or problems) regarding any of the VACs catalogs or their construction, please open a ticket and/or contact John Moustakas (Siena University).