Date of Award
Doctor of Philosophy (PhD)
The field of exoplanets is currently poised to benefit hugely from improved radial velocity (RV) precision. Extreme precision radial-velocity (EPRV) measurements, capable of detecting planetary signals on the order of 10-30 cm/s, will deliver integral planetary parameters, be sensitive to a missing category of lower-mass planets, grant a deeper understanding of multi-planet architectures, and support both current and future space missions such as TESS and JWST. The ability of EPRV to deliver mass estimates is essential for comprehensively characterizing planets, understanding formation histories, and interpreting atmospheric spectra. Until recently, RV precision had stalled at around 1 m/s, i.e. signals with a semi-amplitude of less than 1 m/s could not be faithfully detected. We demonstrate with HARPS, UVES, and CHIRON observations of alpha Cen the need for better data, not just more data. Even with over a decade of observations at around 1 /mas precision, large areas of mass/period parameter space remained unprobed. Higher-fidelity data is needed to significantly push down detection limits. EXPRES, the EXtreme PREcision Spectrograph, was one of the first next-generation spectrographs to go on sky. Installed at the 4.3-m Lowell Discovery Telescope in 2017 and commissioned through 2019, EXPRES is a fiber-fed, ultra-stabilized, echelle spectrograph with a high median resolving power of R~137,000 and an instrument calibration stability of 4-7 cm/s, a factor of 10 better than previous instruments. The stringent requirements of EPRV measurements along with the stability of EXPRES and similar instruments changes how we must extract, calibrate, and model the resultant spectral data. This dissertation discusses the work that must be done in this new regime in terms of data pipelines and modeling stellar signals and showcases some initial progress. We present EXPRES' data pipeline, a new data-driven method for wavelength calibration, and the current state of the field for disentangling stellar signals. The EXPRES extraction pipeline implements a flat-relative, optimal extraction model and excalibur for wavelength calibration. Excalibur is a hierarchical, non-parametric method for wavelength calibration developed as part of this thesis work. Calibration line-positions are de-noised by using all calibration images to construct a model of the accessible calibration space of the instrument. This denoising returns wavelengths a factor of five more precise than previous polynomial-based methods. With EXPRES data, excalibur reduced the overall RMS of RV data sets for all targets tested by 0.2-0.5 m/s. This consistent reduction in overall RMS implies that excalibur is addressing an instrumental, red-noise component that would otherwise permeate all exposures. With instrumental noise lowered and extraction error reduced, intrinsic stellar variability and the resulting apparent RVs now dominate the error budget for EPRV measurements. The EXPRES Stellar Signals Project (ESSP) released high-fidelity, spectroscopic data from EXPRES and photometric observations from the automatic photoelectric telescopes (APT) for four different stars. This allowed for a self-consistent comparison of the 19 different methods submitted, which represent the current state of the field in disentangling stellar signals. The analysis of results is ongoing work. Currently, the best performing method give a final RV RMS of 1.2 m/s. Submitted methods nearly always do better than classic methods of decorrelating RVs from stellar signals. We found that methods returning the lowest RV RMS often used the full spectra and/or flexible statistical models such as Gaussian processes or principal component analysis. However, there was a concerning lack of agreement between methods. If we hope to improve on current advancements and develop methods achieving sub-meter-per-second RMS, we must introduce more interpretability to methods to understand what is and is not working. A densely sampled, high-resolution data sensitive to all categories of stellar variation is needed to understand all types of stellar signals. This dissertation work centers on the question of achieving EPRV capabilities for detecting planets incurring reflex velocity signals on the order of 10-30 cm/s. We consider what needs to be done, describe current development towards this goal, and discuss the future work that remains before sub-meter-per-second precision can become a regular reality. We emphasize the power of data-driven pipelines to account for variations in data for EPRV applications and beyond. Empirically backed conclusions for mitigating photospheric velocities are summarized from the results of the ESSP along with next steps and additional data requirements. Progress is being made, but there remains much work to be done.
Zhao, Lily Ling, "The Path to Extreme Precision Radial Velocity With EXPRES" (2021). Yale Graduate School of Arts and Sciences Dissertations. 446.