Research

My research combines observations, dynamical theory, and statistical methods to study the architectures and evolutionary histories of planetary systems. I use a data-driven dynamics approach, where dynamical simulations help interpret observational signatures and observations in turn constrain and refine theoretical models. In this framework, exoplanetary systems that we observe serve as natural laboratories for testing ideas about planet formation and evolution. An up-to-date list of refereed publications can be found on NASA/ADS.

Here are a brief summaries of my lead-author research publications:

The Astrometric Resoeccentric Degeneracy

Paper: on NASA/ADS // published in ApJL
Code: on GitHub
Yahalomi, Lu, et al. 2026

We identify and derive the astrometric resoeccentric degeneracy, in which a single eccentric planet can mimic the astrometric signal of two circular planets in a near 2:1 mean motion resonance. This effect is the astrometric analogue of the well-known radial velocity degeneracy and is particularly relevant for interpreting results from upcoming datasets such as Gaia DR4.

We show analytically that, to first order in eccentricity, the stellar astrometric motion induced by a single eccentric planet decomposes into a fundamental frequency and its first harmonic. A coplanar pair of circular planets in a 2:1 period ratio produces an identical harmonic structure, rendering the two scenarios observationally indistinguishable under certain conditions. The figure above shows two cool gas giants in a 2:1 configuration and the equivalent eccentric cool gas giant in simulated Gaia data with varying mutual inclinations.

We further map this degeneracy onto physical system parameters, demonstrating how combinations of planet masses and orbital periods can reproduce an apparent eccentric signal. This work highlights an important interpretability challenge for future astrometric surveys and provides a framework for diagnosing and breaking this degeneracy in long-period planetary systems.

TTV Orbital Landscape… the Circus Tent Diagram

Paper: on NASA/ADS // published in ApJ
Code: on GitHub
Yahalomi & Kipping 2026

We investigate the highly degenerate problem of interpreting transit timing variations (TTVs) in single-transiting-planet systems, where the perturbing companion is unseen. Using extensive N-body simulations, we map the multi-modal solution space of possible perturber orbital periods and introduce the “TTV circus tent” diagram as a visualization of this structure.

We show that the dominant TTV signal is governed by a combination of near mean-motion resonance super-periods, synodic periods, and their aliases, which generate predictable clusters of viable solutions in orbital period space. Building on this, we develop a Bayesian framework that partitions parameter space into physically motivated prior regions, enabling efficient and complete exploration of these modes.

Finally, we demonstrate how this framework can be used to distinguish between perturbations caused by non-transiting planets and those induced by exomoons. This work provides a practical roadmap for modeling single-planet TTV systems and significantly improves the interpretability of these otherwise highly degenerate signals.

The TTV Exoplanet Edge… Planets don’t Induce TTVs with Dominant Periods Faster than Half their own Orbital Period

Paper: on NASA/ADS // published in ApJL
Code: on GitHub
Yahalomi, Kipping, Agol, & Nesvorny 2025

We identify a fundamental feature of planet-induced transit timing variations: an “exoplanet edge”, a lower limit on the dominant TTV period. We show that planet–planet interactions do not produce observable TTVs with dominant periods shorter than half the orbital period of the perturbing planet.

Using numerical simulations and analytic arguments, we demonstrate that the dominant TTV signal from distant perturbers appears at either the perturber’s orbital period or half that period. These features arise from aliasing of synodic and near-resonant super-period signals, as well as tidal effects within the system.

We apply this framework to Kepler systems and identify a population of planets exhibiting TTVs below this limit, implying the presence of additional unseen mass (e.g., extra planets or exomoons). This provides a new diagnostic tool for identifying hidden companions and constraining planetary system architectures from TTV observations.

The democratic detrender: Ensemble-based Removal of the Nuisance Signal in Light Curves

Paper: on NASA/ADS // published in ApJS
Code: on GitHub
Code: on Read the Docs
Yahalomi, Kipping, et al. 2026

I am the lead developer on an open source Python code package, called the democratic_detrender, that implements democratic (also previously called method marginalized) detrending. To summarize, democratic detrending uses four different detrending algorithms, and then takes the median solution between the detrending methods for each data point (and adds in quadrature to the errors the median absolute deviation between the detrending solutions). The code has been extensively tested (>1000 stars) on Kepler and TESS light curves

In the package, we currently use four distinct detrending algorithms, which have each been shown in the literature to be efficient and accurate models for stellar noise: CoFiAM, polyAM, local, and GP. For a more detailed description of each detrending algorithm, see below and for a similar application see \citet{Kipping2022}:

• CoFiAM or Cosine Filtering with Autocorrelation Minimisation builds on cosine filtering approach used to study CoRoT data (Mazeh & Faigler 2010). In CoFiAM, we train 30 models with N cosines in our fit, where N ranges from 1 to 30, and at each epoch pick the cosine filter that leads to the least correlated light curve via the Durbin- Watson statistic (Durbin & Watson 1950; Kipping et al. 2013).
  
• polyAM or Polynomial detrending with Autocorrelation Min- imisation follows a similar process to CoFiAM, except we train 30 models with polynomial models as the basis function. Polynomial filtering is a common method for stellar activity detrending (Fab- rycky et al. 2012; Gautier et al. 2012; Giles et al. 2018). In polyAM, the 30 different bases models are 1st- to 30th-order polynomials. For each epoch, as above, the least correlated light curve via the Durbin- Watson statistic (Durbin & Watson 1950; Kipping et al. 2013) is chosen.
  
• local method again uses 1st- to 30th order polynomials, but the local detrended light curve is selected via the lowest Bayesian Information Criterion (Schwarz 1978) computed on the data within six transit durations of the time of mid-transit. This a fairly typical detrending method for the analysis of short-period transiters (Sandford & Kipping 2017).
• GP or Gaussian process regression,as the name suggests, uses a Gaussian process to detrend the stellar activity. We used a quasiperiodic Gaussian process, as it has been shown that it is possible to model stellar activity of a rotating star using a quasiperiodic kernel (Angus et al. 2017). Specifically, we used a SHOTerm kernel from celerite2 via the exoplanet package, which is a stochastically-driven, damped harmonic oscillator (exoplanet, Foreman-Mackey et al. 2021), (celerite2, Foreman-Mackey et al. 2017; Foreman- Mackey 2018).
  

“Killing” the Moon Hypothesis in the Kepler-1513b System

Paper: on NASA/ADS // published in MNRAS
Code: on GitHub
Research Summary: Twitter Thread
Yahalomi, Kipping, et al. 2024

We performed model selection (planet-planet vs. planet-moon vs. stellar activity) on the TTVs observed in Kepler-1513b using follow-up ground-based and space (TESS) observations. We found that the complete TTV signal, including two additional transit observations, a ~decade since the last Kepler observation, was inconsistent with both the planet-moon and the stellar activity hypothesis and was consistent with an external perturbing non-transiting planet near the 5:1 mean motion reference (MMR).

Using nested sampling with MultiNest, we modeled the TTVs with N-body simulations for the planet-planet model with SWIFT and a photodynamical planet-moon model with LUNA. We find that the planet-moon model is inconsistent with one the two new transit time observations at the ~3-σ level. We also show that the observed stellar activity (via stellar rotational period and slope of the LC around transits) is inconsistent with causing the observed TTVs. The main figure from this paper, showing this results, can be seen below. For more details, see the full paper and the code.

Solar System Analog Hunting

Paper: on NASA/ADS // published in AJ
Code: on GitHub
Research Summary: Twitter Thread
Yahalomi, Angus, Spergel, & Foreman-Macket 2024

Earth-mass exoplanets on year-long orbits and cool gas giants (CGG) on decade-long orbits lie at the edge of current detection limits. The Terra Hunting Experiment (THE) will take nightly radial velocity (RV) observations on HARPS3 of at least 40 bright nearby G and K dwarfs for 10 years, with a target 1σ measurement error of ~0.3 m/s, in search of exoplanets that are Earth-like in mass and temperature.

However, RV observations can only provide minimum mass estimates, due to the mass-inclination degeneracy. Astrometric observations of these same stars, with sufficient precision, could break this degeneracy. Gaia will soon release ~100-200 astrometric observations of the THE stars with a 10 year baseline and ~34.2 μas 1σ along-scan measurement error. The Nancy Grace Roman Space Telescope will be capable of precision astrometry using its wide field imager (target ~5-20 μas 1σ measurement error for bright stars) and could extend the astrometric observational baseline to ~25 years. We simulate and model an observing program that combines data from these three telescopes.

We found that (1) THE RVs and Gaia astrometry can detect Earth-like and CGG-like exoplanets around bright Sun-like stars at 10 parsecs and that (2) adding Roman astrometry improves the detection precision for CGG masses and periods by a factor up to ~10 and ~4, respectively. Such a survey could provide insight into the prevalence of Solar System analogs, exoplanet architectures reminiscent of the mass and orbital separation hierarchy of our Solar System, for the nearest Sun-like stars.

Determining the Mass and Radius for the White Dwarf in KOI-3278

Paper: on NASA/ADS // published in ApJ
Code: on GitHub
Yahalomi, Shvartzvald, Agol, Shporer, Latham, et al. 2019

I also led a follow-up study of KOI-3278, the first discovered self-lensing binary (Kruse and Agol, 2014). In the paper, we presented independent Einsteinian and Newtonian gravitational models for the system and showed that the respective estimates for the white dwarf mass in KOI-3278 agreed within 5.2%. We also presented a joint Einsteinian and Newtonian model that allowed us to remove white dwarf evolution models and assumptions on the white dwarf mass-radius relation. Doing so provided a model independent test of the mass-radius relation and thus white dwarf structure models. ​​

We are hopeful that further studies of the white dwarf in KOI-3278, as well as four other discovered white dwarf self-lensing systems, in conjunction with UV follow-up observations will provide a test of white dwarf structure models. I am currently working with Columbia Undergraduate student, Yassine Abaakil, on a project reproducing this work for the other four known Kepler self-lensing binaries.

There was a press release for this paper, which can be found here.

Gravitational Microlensing of the Strongly Lensed Type IA Supernova iPTF16geu

Paper: on NASA/ADS // published in MIT Journal of Undergraduate Research
Yahalomi, Schechter, & Wambsganss 2017

As an undergraduate at MIT, I led an investigation into the micro-lensing of the first discovered, strongly lensed type Ia supernova, iPTF16geu. This system cause commotion when it was discovered, as its standard candle nature plus time delay values, could theoretically lead to a constraint of the Hubble parameter. However, the macro-models published by More et al. 2017 showed significant flux ratio anomolies with the observations of the four images. I constructed a Monte-Carlo simulation, in Python, which showed that the likelihood for micro-lensing to cause these flux ratio anomalies between observations and macro-models was ~3/1000. I then showed that even if the macro-models fit perfectly, the microlensing probability density functions create a probability distribution on the intrinsic brightness with a full width half maximum of 0.73 magnitudes. As such, the error for the standard candle brightness is quite large. This reduces the utility of the standard candle nature of type Ia supernovae. This project led to a first author paper, and cover article, in the MIT Undergraduate Research Journal Fall 2017, which has 30 academic citations (as of September 2023). 

I then built on this work, leading three other micro-lensing projects that took advantage of the code that I had generated, culminating in my senior thesis, entitled “Statistical Analyses of Gravitational Microlensing Probability Densities.”