Known Issues

Below we list some issues known to crop up here and there in the everest light curves. This list is not comprehensive, so if you think you’ve found an issue with the de-trending, please let us know by opening an issue on the github page.

Missing eclipses

Some very deep transits and eclipses may be missing from the plots in the data validation summaries. This is almost always a plotting issue, and not an issue with the pipeline. We recommend plotting the light curve interactively using the user interface and zooming out – the eclipses are likely just below the plot limits!

Stars in crowded apertures

Everest 2.0 light curves are far more robust to crowding than the previous ones, but stars in crowded fields remain an issue for the algorithm. PLD implicitly assumes that the target star is the only source in the aperture and may overfit astrophysical signals if bright contaminants are present. It is important to inspect the apertures and high resolution images provided in the data validation summaries to ensure that crowding is not affecting the light curves. In general, crowding is only a problem when the contaminant source is of a similar magnitude (or brighter) than the target and is inside the optimal aperture. Future versions of the pipeline will improve aperture selection and adapt PLD to work for these stars.

RR Lyrae

Everest 2.0 again improves the performance of PLD on extremely variable stars relative to Everest 1.0. Nonetheless, about 40% of long cadence RR Lyrae seem to be overfitted by the algorithm. We recommend inspecting the raw and de-trended light curves in the data validation summaries to check whether PLD has dampened the astrophysical signal. Check out this file for a list of the EPIC numbers of 100 RR Lyrae stars from the first few K2 campaigns, obtained from the K2VARCAT catalog. We report (qualitative) scores for each of the EVEREST1, EVEREST2, K2SFF, and K2SC pipelines based on whether they overfitted the variability. Inspecting the light curves of several of these can give you a good sense of what an “overfitted” RR Lyrae star looks like. Remember that to bring up the EVEREST, K2SFF, and K2SC light curve of any star, just type

everest 210868876
everest 210868876 -f
everest 210868876 -s

in the terminal, respectively. This particular example suffered overfitting in both EVEREST and K2SFF, but not in K2SC.

Finally, I wrote a simple iPython Notebook showing how to go about manually de-trending an RR Lyrae light curve for which the EVEREST pipeline failed. Take a look at the code for ideas on how to handle these issues for other similar targets. Note that this file can also be viewed on github.

Aperture losses

Extremely saturated stars whose flux bleeds out of the target pixel file postage stamp are not properly de-trended, since a large portion of the astrophysical signal is missing from the light curve. We recommend inspection of the target aperture to ensure all of the stellar flux is enclosed by the chosen aperture.

Faint short cadence targets

Extremely faint stars observed in short cadence mode are not properly de-trended by PLD. This is because they are dominated by photon noise (instead of instrumental noise) and the algorithm has trouble mending the light curve segments at the breakpoints, resulting in discontinuous jumps for short cadence targets fainter than about 16th magnitude.