Radial Velocities

The APOGEE radial velocities (RVs) are determined with Doppler software (the SDSS-V version is here https://github.com/sdss/doppler)

Radial velocities (RV) are determined on a star-by-star basis using Doppler which fits a Cannon model to the data, determining Teff, logg, [Fe/H] and radial velocity in the process. It fits all of the visit spectra simultaneously, using a single set of stellar parameters for the star, but a separate radial velocity for each spectrum. The model spectrum is convolved with the correct LSF (line spread profile) for each visit spectrum.

These are the steps taken by rv.py:

1. Each visit spectrum is fit individually by Doppler.

2. Mean stellar parameters are calculated over the visits.

3. All of the visit spectra are fit simultaneously with a single set of stellar parameters and a separate RV for each spectrum.

4. Each visit spectrum is cross-correlated with the best-fit Cannon model spectrum and visit RV (these are the xcorr_ values). The peak pixel of the CCF is used to determine the xcorr RV (not interpolated).

5. Each visit CCF is Gaussian Decomposed and the number of components saved in N_COMPONENTS. If N_COMPONENTS=0, that means gausspy did not detect a peak or it crashed.

6. After extra quality cuts, the spectra and RVs are used to create the apStar combined spectrum.

Each time a star is observed, all of the existing visit spectra for that star are refit with Doppler. This means that there are multiple versions of RVs for each visit spectrum based on the latest visit MJD (Modified Julian Date) and sets the “STARVER” value.

The RV processing is parallelized over each object and log file is created in the stars/ directory in the HEALPix subdirectory for that star.

RV Columns and Flags

Visit spectra must pass through QA checks to be used in the final spectral combination:

1. Initial QA check: STARFLAG should not have any of the BAD flags set (BAD_PIXELS,VERY_BRIGHT_NEIGHBOR,BAD_RV_COMBINATION,RV_FAIL) and must have S/N>2. These are the visits that Doppler is run on and the ones that are counted in NGOODVISITS at the STAR level.

2. RV check: After Doppler has been run on all of the visits that passed the initial QA cut, then an additional selection is made for visits that do not have RV_REJECT set. These visits are used to determine the mean RV, RV scatter, and make the combined spectrum. These are the visits that have GOODVISIT=True set and are counted in NGOODRVS at the STAR level.

Relevant RV flags in STARFLAG:

Flag	Description
RV_REJECT	Rejected visit because cross-correlation RV differs significantly from least-squares RV: 50 km/s (rv_teff >= 6000) 20 km/s (rv_teff < 6000 and rv_logg > 3.8) 10 km/s (rv_teff < 6000 and rv_logg <= 3.8)
RV_SUSPECT	Suspect visit (but used) because cross-correlation RV differs slightly from least-squares RV (> 0 km/s)
MULTIPLE_SUSPECT	Suspect multiple components from Gaussian decomposition of cross-correlation (n_components > 1)
RV_FAIL	RV failure. No good visits or RVs for this star

DR17 also had:

Flag	Description
SUSPECT_ROTATION	CCFWHM > 2 * AUTOFWHM
SUSPECT_BROAD_LINES	Cross-correlation peak with template significantly broader than autocorrelation of template: WARN (AUTOFWHM > 300) (will be restarted in 1.6)

Before DR17 we also had (when we used synthetic and combined templates):

Flag	Description
SUSPECT_RV_COMBINATION	RVs from synthetic template differ significantly (~2 km/s) from those from combined template: WARN
BAD_RV_COMBINATION	RVs from synthetic template differ very significantly (~10 km/s) from those from combined template: BAD

Relevant Visit-level columns:

Column	Description
STARFLAG	Quality flags for this visit spectrum
STARVER	Version of the Doppler processing for this star. Generally, the MJD of the last visit used with Doppler.
GOODVISIT	Boolean flag indicating that the visit passed all the QA cuts and was used in determining the mean velocity and the combined spectrum
VREL	Doppler shift (km/s)
VRELERR	Uncertainty in VREL (km/s).
VRAD	Barycentric radial velocity (km/s)
BC	Barycentric correction for VREL (km/s)
CHISQ	Chi-squared of the best-fit Cannon model
RV_TEFF	Teff of the best-fit Cannon model
RV_LOGG	log(g) of the best-fit Cannon model
RV_FEH	Metallicity of the best-fit Cannon model
XCORR_VREL	Cross-correlation doppler shift (km/s)
XCORR_VRELERR	Uncertainty in XCORR_VREL (km/s)
XCORR_VRAD	Barycentric radial velocity from cross-correlation (km/s)
N_COMPONENTS	Number of components from the Gaussian decomposition of the cross-correlation function. If N_COMPONENTS=0, then this means that the Gaussian decomposition did not detect a peak or it failed for the visit
RV_COMPONENTS	Number of Gaussian components for each of the 3 detectors

Relevant Star-level columns:

Column	Description
STARFLAG	Quality flags OR-combined across visits
ANDFLAG	Quality flags AND-combined across visits
STARVER	Version of the Doppler processing for this star. Generally, the MJD of the last visit used with Doppler.
NVISITS	Total number of visits for this star (no QA cuts)
NGOODVISITS	Number of visits passing the initial QA cuts (no BAD STARFLAG flags and S/N>2)
NGOODRVS	Number of visits passing all QA cuts and used in determining the mean RV and combined spectrum
VRAD	Barycentric radial velocity (km/s)
VSCATTER	Standard deviation of visit-level VRAD values (km/s)
VERR	Uncertainty in VRAD. Standard deviation of the mean, VSCATTER/sqrt(NGOODRVS) (km/s).
VMEDERR	Median uncertainty of the visit-level RVs (km/s).
CHISQ	Chi-squared of the best-fit Cannon model to all visits
RV_TEFF	Teff of the best-fit Cannon model (K)
RV_TEFFERR	Uncertainty in RV_TEFF (K)
RV_LOGG	log(g) of the best-fit Cannon model
RV_LOGGERR	Uncertainty in RV_LOGG
RV_FEH	Metallicity of the best-fit Cannon model
RV_FEHERR	Uncertainty in RV_FEH
RV_CCPFWHM	FWHM of the cross-correlation peak of the spectrum and best-fit Cannon model (km/s)
RV_AUTOFWHM	FWHM of the cross-correlation peak of best-fit Cannon model with itself (km/s)
N_COMPONENTS	Number of components from the Gaussian decomposition of the cross-correlation function. This the maximum N_COMPONENTS of the visits for this star. If N_COMPONENTS=0, then the Gaussian decomposition failed for all visits for this star.
MEANFIB	Mean fiberID across the visits
SIGFIB	Standard deviation of the fiberID across the visits. Note that if SIGFIB=0, then RV uncertainty is significantly better than normal.