Ryan Rubenzahl, Ph.D.

The Solar Calibrator (SoCal)

KPF SoCal is an autonomous solar feed on the roof of WMKO that, through an 80 m fiber run, provides KPF with disk-integrated sunlight. With SoCal, KPF acquires these ‘Sun-as-a-star’ spectra at signal to noise 500-1500 over 445-870 nm in 12 second exposures (61 sec cadence).

• Read the paper (Rubenzahl et al. 2023)

smolgp

smolgp is a Python/JAX library that enables fast and lightweight Gaussian process modeling by using their state space representation. Uniquely, smolgp can handle integrated measurements (even from multiple instruments that overlap in time!) - which must be accounted for if exposure times are comparable to the timescale of the process - in O(N), while traditional methods are O(N^3). smolgp also has the only true scalable implementation of the quasiperiodic kernel loved throughout astronomy. Derivative observations, multivariate observations, multicomponent GPs, predicting out of sample points, and drawing prior or posterior samples are all fast and convenient, and you can even parallelize on GPU to get even faster likelihoods and conditioned means and variances!

• View the docs
• Learn the math

ReloadedRM

An implementation of the Cegla et al. (2016) Reloaded Rossiter-McLaughlin method, written in a RadVel-like API to facilitate easy model building and MCMC with multiple parameter bases. A custom coordinate system allows the user to place starspots on the star and include their effects in the modeled Doppler shadow.

• View on GitHub

continuum

A continuum normalization routine, written in JAX, that is robust to outlier spikes (e.g. cosmic rays) and missing data. I made continuum for the use case of just needing a spectrum flattened to a continuum of 1 that can handle wide absorption lines and lines truncated by the edge of an order.

• View on GitHub

State Space Gaussian Processes

When working with the EPRV solar datasets, I ran into a problem. The exposure lengths were too similar to the p-mode timescales, which we traditionally model with a GP that assumes instantaneous timestamps. Combining data from multiple instruments added a second problem: exposure overlap. Integrating the covariance matrix gives the right answer, but in O(N³) time (Luhn, Rubenzahl et al. 2026), which is intractable for the solar data (~100,000 RVs). Instead, So Hattori and I found an alternative but equivalent formulation using the state space representation of GPs, which treats the process as a state evolving according to its governing stochastic differential equation (SDE <-> the kernel function in traditional GP language). Since the SDE defines the state and its derivatives, we added an integral state to model the effects of exposures. This way, we get the same answer, but in O(N) time.

• Read the paper (Rubenzahl & Hattori et al. 2026)

Sun-as-a-Star

The Sun is our best hope to understand how stellar variability impacts our observed spectra. It is the only star for which we can truly isolate only the (stochastic) stellar and instrument signals from the (known, deterministic) planet signals.

My current work at the CCA combines a year of solar spectra from KPF, NEID, HARPS-N, and EXPRES to trace solar variabiltiy and instrumental drift per pixel. For the Sun, that can be mapped to formation layers in the atmosphere. For drift, to location on the detector. Stay tuned for the paper coming very soon!

Obliquities

The stellar obliquities of close-in oddball systems like hot Jupiters and ultra low density super Neptunes, particularly when they have outer companions, give a key clue to how they could have formed. Using HIRES to observe the Rossiter-McLaughlin effect, I found the keystone super-puff WASP-107 b is on a polar orbit, a clear sign of a dynamic migration history. Similarly with KPF I found the ultra-hot Jupiter KELT-18 b is also on a polar orbit. Despite its high eccentricity (0.84), I found Kepler 1656 b to likely be in an aligned/low obliquity configuration. All three have outer companions that drive unique dynamics in each system.

Read more about each system

• WASP-107 b (Rubenzahl et al. 2021)
• Kepler-1656 b (Rubenzahl et al. 2024b)
• KELT-18 b (Rubenzahl et al. 2024c)

Planet Discovery

From photon to instrument to pipeline to analysis to vetting to planet. The fruit of all our labor. I’ve been fortunate enough to be a part of exoplanet discovery in two systems so far: an outer giant planet companion in the WASP-107 system, as well as confirming both the transiting candidates in the TOI-1347 system. TOI-1347 b is especially interesting in its own right, as an 11 M_⊕ and 1.8 R_⊕ rocky planet, it shows signs of a high mean-molecular weight atmosphere.

Read more about each system

• WASP-107 c (Piaulet, Benneke, Rubenzahl et al. 2021)
• TOI-1347 b & c (Rubenzahl et al. 2024a)

RV Surveys

I was a member of the TESS-Keck Survey, which obtained mass constraints for 126 planets or planet candidates in 86 TESS systems using the Keck/HIRES spectrograph. I personally (by hand) organized the observing schedule for the survey each of the 298 calendar nights we were on the telescope. I also contributed to the RV analysis (fitting & vetting) of a dozen or so TKS systems, which are cateloged in Polanski et al. 2024.

• Read the results of the TKS catalog here