Using Radial Velocities

This page contains information about using the DR17 APOGEE combined and individual visit radial velocities (RVs) and provides several examples to illustrate their uses. An understanding of how the quality of the supplied measurements are flagged using bitmasks is important to properly and effectively utilize these datasets. For a general description of how the radial velocities are determined, which is different in DR17 than in previous data releases, see the radial velocities page.

Visit-level Radial Velocities

For each visit to an object, APOGEE reports the measured radial velocity as VREL and the associated uncertainty is stored in VRELERR. The barycentric correction is given by BC, and applied to the relative RV to yield the visit barycenter-corrected velocity, VHELIO.

Note: The header keyword VHELIO is retained for historical reasons, but the pipeline is calculating a true barycentric velocity, not a heliocentric velocity.

Because the barycentric correction can be calculated very precisely (at the m/s level, such that the associated errors are very small relative to VRELERR), the total uncertainty for VHELIO is VRELERR. If you are interested in utilizing the visit RVs to search for variations in RV over time, you should look for variations in VHELIO as a function of the Julian Date, JD. The Julian Date given in the summary allVisit file is the JD-MID from the apVisit file, which refers to the middle of an exposure sequence (a "visit") determined from exposure-time weighted mean of the mid-exposure times and the exposure times are nearly always the same. For RV-variations on timescales shorter than a typical "visit" (normally, ~80 min including overheads), this may be insufficient time resolution. Note that the MJD columns gives an integer value of the Modified Julian Date on which that star was observed.

At the visit-level, the STARFLAG bitmask provides detailed quality information about the individual visit spectra. Results from a visit that has the LOW_SNR flag, which is set automatically for all spectra with signal-to-noise (SNR) less than 5, should be used with caution. Ideally, using a lower limit on the SNR of ~10 will produce the best results. Additionally, visits with the RV_REJECT, RV_FAIL, BAD_PIXELS, or VERY_BRIGHT_NEIGHBOR flags should not be used. These five flags (STARFLAG bits 0, 3, 4, 19, 22) are the most obvious indicators that the visit radial velocity may not be trustworthy, but for the highest quality visit velocities, you should also exclude visits with PERSIST_HIGH, PERSIST_JUMP_POS, and PERSIST_JUMP_NEG. Other STARFLAGS or additional cuts based on VRELERR (e.g., all visit velocities must have uncertainties less than 2 km/s) may be utilized to further remove RVs determined from low-quality visit spectra. In DR17, the MULTIPLE_SUSPECT flag indicates that multiple components are apparent in the Gaussian decomposition of the cross-correlation; stars with this flag set (in either the allVisit or allStar summary files) is likely be SB2 candidates, and additional inspection of their visit spectra may be necessary; note that MULTIPLE_SUSPECT has been set conservatively, and there may be spectroscopic multiples that do not have this bit set.

Please see Nidever et al. 2015 for a detailed discussion of the pipeline. Troup et al. 2016, Price-Whelan et al. 2018, and Price-Whelan et al. 2020 provide more detailed examples of using the visit-level APOGEE RVs for scientific analyses.

Average Radial Velocities

APOGEE also reports the SNR-weighted average velocity associated with each combined spectrum. This is provided in VHELIO_AVG and the scatter around this average is stored in VSCATTER. The number of visit spectra included in the combined spectra is given by NVISITS. Please note that the number of visits given for the combined spectra (e.g., in allStar file) does not always match the total number of visits obtained (e.g., the number of unique entries in the allVisit file). This occurs because an object that was observed in two different fields will be passed through the visit-combination pipeline as two distinct objects and, thus, will be presented as separate entries in the summary catalogs containing combined data; the object will have the same APOGEE_ID, but with different FIELD tags (the sum of FIELD and APOGEE_ID define a unique object for which visit spectra are combined). The number of visits recorded for a combined spectrum corresponds to the number of visits contributing to that combination. The sum of the NVISITS for each entry of that object will be equal to the total number of visits across all fields (e.g., the number of unique observations present at the visit level).

The SNR-weighted uncertainty is stored in VERR. These uncertainties are known to be underestimated; in some cases (e.g., a combined spectrum with high SNR obtained over many visits) VSCATTER may represent a better estimate of the overall measurement precision. If VSCATTER is very large (i.e., much larger than VERR) and measured from multiple (more than 3) high SNR visit spectra, this may be an indicator that the star is in a stellar binary. For stars with only a single visit, VSCATTER is set to zero.

Both Visit Combination and Radial Velocities pages provide more information about the visit combination and average RV derivation process.

Example: Eclipsing Binary KIC 2161623

The following analysis gives a brief example of how the parameters outlined above might be applied.

KIC 2161623 (APOGEE_ID of 2M19265867+3735276) is a known eclipsing binary discovered by Kepler (see Conroy et al. 2014 for details) that has a period of 2.2835 days (Kirk et al. 2016). This star has been observed by the APOGEE survey with a total of 20 visits (each with SNR ≤ 10, though 5 visits are rejected with the RV_REJECT bit in STARFLAG set), resulting in a combined spectrum with high signal-to-noise (SNR ~ 160).

In this example, we assume a user wants to (1) confirm that the star is in a binary system, and (2) derive its orbital period using only the data provided by APOGEE.

Detecting Binarity
The binarity of any star can be determined by comparing VSCATTER to the measurement error. For KIC 2161623, the VSCATTER is ~100 km/s, which is much greater than the median visit RV error (< 1 km/s), so it can be inferred that the star is a binary. In general, if VSCATTER > 1 km/s (and VSCATTER > 10×VERR_MED), the star is likely a binary; however, if a star has only a few visits or many very low-SNR visits, VSCATTER may not be a reliable indicator of binarity. For this reason, a large value for VSCATTER is not a definitive marker for binarity; individual systems should be studied more closely for this determination.
Orbital Period
Now that the binarity of the star is confirmed using the VSCATTER derived from the APOGEE visits, the visit radial velocities can be used to derive the period of the orbit. Note, this method works best for stars with many visits (at least eight visits) that span a temporal baseline longer than the orbital period.

From the allVisit file, the STARFLAGS detailed above should be used to filter out low-SNR or poor-quality visits. For KIC 2161623, 5 out of the 20 visits have any important quality flags set; these 5 visits have the RV_REJECT flag set, meaning that these visits should be left out of this orbital analysis. The user should consider whether other flags (e.g., RV_SUSPECT, MULTIPLE_SUSPECT, etc.) will be ignored in all or some cases. Additionally, the user should check that the uncertainties are suitable for this analysis. In this case, the VRELERR given for each visit is < 1.5 km/s; again, the measurement errors are much less than the scatter, so the remaining 15 visits without the RV_REJECT flag can be included. Using the visit radial velocities (VHELIO) and the associated errors, a Lomb-Scargle periodogram (Press & Rybicki 1989) can be used to detect periodic signals in the observations. The period of maximum power in the periodogram for this set of observations falls at 2.28345 days, so the APOGEE RVs confirm the period derived from the Kepler transits of the system.

Users interested in fitting other orbital parameters (e.g., semi-amplitude, eccentricity, etc.), might use one of the various available RV variation fitters. RV variation fitters include, but are not limited to: RadVel (Fulton et al. 2018), Systemic Console (Meschiari et al. 2009), and The Joker (Price-Whelan et al. 2017).

With 15 visits, all of these algorithms should converge on a single orbital solution for KIC 2161623. The user should find a low-eccentricity orbit (e < 0.1), with a period of 2.2835 days and a semi-amplitude of 166 km/s. The derived systemic velocity for the system is v0 = -45.3 km/s, whereas the SNR weighted average velocity is VHELIO_AVG = -52.3 km/s with VERR = 0.5 km/s. The systemic velocity is not within the radial velocity uncertainty but is well within the scatter of the data. The phase-folded orbit is shown below.

Both Price-Whelan et al. 2020 and your paper here provide more detailed examples of using the visit-level APOGEE radial velocities for scientific analyses.

A Lomb-Scargle periodogram for the APOGEE RV data of Eclipsing Binary, KIC 2161623. Maximum power occurs at 2.28 days, which is taken as the period of the system.  <em>Figure courtesy of H. Lewis.</em>
A Lomb-Scargle periodogram for the APOGEE RV data of Eclipsing Binary, KIC 2161623. Maximum power occurs at 2.28 days, which is taken as the period of the system. Figure courtesy of H. Lewis.
 The phased RV curve for Eclipsing Binary, KIC 2161623, using the period determined from the RVs alone. <em>Figure courtesy of H. Lewis.</em>
The phased RV curve for Eclipsing Binary, KIC 2161623, using the period determined from the RVs alone. Figure courtesy of H. Lewis.