MARVELS Caveats

We strongly caution the users of the RV data for many of the same reasons given for DR11. Although processed in a completely different way from DR11, the results generally show the same level of RV systematics over long periods of time. Data taken during the same observation period (typically 7-10 days) can generally be considered reliable. Inversely, RV changes that are only apparent when looking at data taken over many months is suspect. Short period cycles are therefore much more reliable, when observed. All DR11 caveats concerning CAS tables also apply to DR12.

However, with a more complete understanding of the systematics it became possible to predict the true error bars as a function of not only target magnitude but also beam location. The predicted 'offset error' in the starrv_platename.fits files includes not only photon limits and ideal instrument performance but also factors in the systematics, which vary depending on the location of the spectra on the CCD.

We strongly caution users of the radial velocity (RV) data that there is a significant systematic, temporally correlated noise ("red noise") component in the RV curves. The temporally correlated noise is of the same order or greater than the uncorrelated noise ("white noise"), and grows at a faster rate as the flux gets lower. The correlation is particularly strong on timescales of one to a few months, but there is not much correlation on timescales of a week. The ratio between the red noise and the white noise is different for every star, because the noise characteristics of the star depend not only on the observed flux, but also on the depth and sharpness of the spectral lines. Thus, the error bars on the data are not strictly usable for the purposes of algorithms designed to find sinusoidal signals in data that are not affected by systematics.

The error bars on the RV data derived using the interferometer fringes in DR11 (these data are the "DFDI RV" data) are indicative of the theoretical best white noise only (accounting for the photon counts and for the visibility of the interferometer fringes); red noise must be estimated by the user.

The error bars on the RV data derived using a cross-correlation function in DR11 (these data are the "CCF RV" data) are not very accurate (they are based on an unweighted fit for the width of the CCF peak), and should not be used for weighting purposes when modelling the RV curve. It is better to estimate the error empirically over a sample of RV curves than to rely on the CCF RV error bars.

Star names in CAS and Binary Results files

The STARNAME column in the Catalog Archive Server (CAS) tables and the binary fits (see Data Model) files are a unique identifier for a star (although the same star may appear on multiple plates). This identifier is normally constructed out of a catalog name for the star (Hipparcos, HST Guide Star Catalog, Tycho 2, or 2MASS). In the v001.88 results files, stars that have Hipparcos-based STARNAMEs are not the actual Hipparcos names due to a programming error in the target selection code. In the CAS, the HIP_Name column contains the correct Hipparcos name. The binary files have not been corrected, so we recommend that you use one of the other catalog names in your queries of other information sources.

Stellar effective temperatures in CAS and Binary Results files

Since preselection data were only available for years 1 + 2, SSPP temperatures could not be assigned to stars monitored in years 3 + 4. So, temperatures (Teff) provided for years 3 + 4 are from the Casagrande InfraRed Flux Method using V-K or J-K, depending on which is available, or, if both are available, then the average of the two results. Temperatures provided for years 1 + 2 are from SSPP.