The sensitivity of TESS to transiting planets in TOIs with close-in stellar companions

The study quantifies how the presence of close-in stellar companions impacts the ability of TESS to detect transiting exoplanets. The authors examined 2052 TESS Objects of Interest (TOIs) that appear to be single-star systems based on speckle interferometric observations, as well as 188 TOIs in unresolved (< 1.2″) stellar binaries. For each planet, they took its transit signal-to-noise ratio (SNR), radius, and orbital period from the TOI catalog and, for planets in stellar binaries, corrected the radius for dilution by the companion. By applying a scaling relation to the measured transit SNR of each TOI, they determined the detectability of transits as a function of both planet radius and orbital period. The sensitivity grids quantify the bias against the detection of small planets in unresolved binaries by TESS and show that there is a particularly low sensitivity to planets transiting secondary stars in unresolved binaries, especially as the magnitude difference between the stars increases. These sensitivity grids are available for download to facilitate their use in other studies.

The sample was limited to TOIs with main-sequence host stars and high-resolution speckle interferometric observations from WIYN, Gemini, and SOAR. A TOI was classified as single if it had been vetted with these imagers and was not in the list of TOIs with stellar companions. Binaries lacking valid magnitude difference measurements at either 832 nm or 879 nm were excluded, as were binaries with separations > 1.2″. The final sample contained 2052 single hosts (310 confirmed planets, 223 known planets, 1519 planetary candidates) and 188 binary hosts (10 confirmed planets, 10 known planets, 168 planetary candidates). Nineteen TOIs with multiple companions were excluded.

For single-star TOIs, the transit SNR was taken directly from the TOI catalog. For binary TOIs, the planetary radius in the catalog is incorrect due to dilution, so a correction factor Xr was computed using the formula Xr = (Rp(true)/Rp(single)) = (Rt/R1) × sqrt(Ftot/Ft), where Rp(true) is the true planetary radius, Rp(single) is the incorrect radius assuming a single star, Rt is the true radius of the star being transited, R1 is the nominal radius under the single-star assumption, Ftot is the total flux of all stars contributing to the transit light curve, and Ft is the flux of the star being transited. Because the identity of the host star is usually unknown, two sensitivity grids were computed for each binary TOI—one assuming the planet transits the primary and one assuming it transits the secondary. The companion star's radius was estimated using the speckle magnitude difference and the Pecaut & Mamajek (2013) sequence.

The sensitivity of TESS to small planets improves dramatically at later spectral types because the transit depth is deeper for any given planet radius. For example, the sensitivity is 90% for a 5 Earth-radius planet in a 10-day orbit around an M-dwarf, but for an F-dwarf, the sensitivity for the same planet drops to about 50%. For single-star TOIs, the sensitivity grids were used to correct the observed period-radius distribution. In the uncorrected histograms, gas giants are conspicuous for all spectral types except M and have the highest relative frequency for spectral types A, F, and G. However, after dividing by the sensitivity grid, the relative frequency of gas giants is greatly reduced, while that of small planets is significantly enhanced.

For binary TOIs, the sensitivity depends strongly on the magnitude difference of the binary and which star hosts the planet. For binaries with magnitude difference < 1, the radius-correction factors are very similar for both primary-host and secondary-host TOIs, so the sensitivity grids are almost identical. But as the magnitude difference increases, the radius-correction factors diverge; this is because the dilution becomes minimal if the planet transits a much-brighter star but extreme if it orbits a much-fainter star. For magnitude difference ≥ 4, the sensitivity grid for primary-host TOIs is almost identical to that of a single-star TOI, with near-100% sensitivity to close-in gas giants. In contrast, for secondary-host TOIs with magnitude difference ≥ 4, only half of close-in gas giants would be detected and only 10% of Neptune-sized planets would be detected. Thus, even if both stars were equally likely to host transiting exoplanets, TESS would preferentially detect planets around the brighter star.

Geometric considerations further reduce the likelihood of detecting transiting exoplanets around the secondary. For exoplanets with circular orbits much smaller than their host star, the transit probability is approximately p = R*/a. The stellar radius has a much stronger dependence on the star's mass than does the semimajor axis, so the secondary's comparatively small radius reduces the transit probability more than the diminished semimajor axis boosts it. These geometric considerations are not taken into account in the sensitivity plots, so the prospects of detecting transiting exoplanets around the secondary in high-mass-ratio systems are even more dire than suggested by the sensitivity plots.

The authors note that their methodology differs from that of Ziegler et al. (2021) in several ways. Their approach allows for the possibility that the planet transits the secondary star and explores the poor prospects for detecting planets in such systems. Also, the binary systems simulated in Ziegler et al. (2021) allow for a wide range of angular separations, including those far in excess of the 1.2″ upper limit used here. The motivation is to explore the consequences of stellar companions that will be blended in typical seeing-limited observations.

The authors conclude that the multiplicity of TOI host stars has a significant effect on the detectability of exoplanets, underscoring the value of high-resolution imaging. If a host star is confirmed to be single, TESS is better able to detect small planets in comparison to binary TOIs, and this sensitivity improves with later spectral types. In binary host stars, the ability of TESS to detect transits depends strongly on the magnitude difference of the binary and which of the stars hosts the exoplanet. As the magnitude difference increases, TESS becomes increasingly insensitive to planets around the fainter star. The analysis suggests that there could be a population of transiting exoplanets around secondary stars in binary TOIs and that TESS will struggle to detect transits of these planets. Neptune-size planets (and smaller) in particular have a very low probability of detection if they orbit the secondary. This provides a justification for assuming that the primary star is the host in the absence of additional evidence, but it also means that the planetary content of binary systems remains poorly known in comparison to single-star TOIs. The sensitivity grids are provided as look-up tables in supplemental online material for use in future studies.

Improvements to AI Systems:

1. Dilution-Aware Exoplanet Detection Model

• Improve transit-detection algorithms (e.g., neural networks or Bayesian pipelines) to explicitly account for stellar companions by incorporating the correction factor X r = (R t/R 1) times sqrt F tot/F t.

• The improved AI can automatically adjust planet radius estimates and transit SNR for unresolved binaries, reducing false negatives for small planets around fainter secondary stars.

2. Sensitivity-Grid Integration for Survey Planning

• Train a reinforcement-learning or optimization agent to use the provided sensitivity grids (as lookup tables) to prioritize TESS targets or design follow-up observations.

• The improved system can predict detection completeness as a function of planet radius, orbital period, spectral type, and binary magnitude difference, then allocate telescope time to maximize yield of small planets.

3. Binary-Host Classification and Host-Star Identification

• Enhance multi-star system classifiers (e.g., random forests or deep CNNs) to predict the probability that a transiting planet orbits the primary vs. secondary, using magnitude difference and spectral type from speckle interferometry.

• The improved AI can output a posterior probability for each host-star scenario, enabling more accurate planet radius and occurrence-rate calculations.

4. Geometric Transit-Probability Correction

• Incorporate the geometric transit probability p = R*/a into AI-based occurrence-rate estimators for binary systems.

• The improved system can correct for the reduced likelihood of detecting planets around smaller secondary stars, yielding unbiased occurrence rates for planets in binaries.

5. Automated Correction of Catalog Biases

• Build a post-processing AI module that applies the sensitivity grids to any TOI catalog (or future missions like PLATO) to de-bias observed period–radius distributions.

• The improved system can output corrected histograms and occurrence rates, automatically separating single-star and binary-host populations.

6. Simulation-to-Real Transfer for Transit Surveys

• Use the paper’s methodology to generate synthetic training data for AI-based transit vetting, simulating both primary-host and secondary-host transits with varying magnitude differences.

• The improved AI can be trained to recognize diluted transit signals and avoid misclassifying them as noise or false positives.

7. Adaptive Observing Strategy for Follow-Up

• Develop a decision-support AI that, given a binary TOI’s magnitude difference, recommends whether to assume primary-host (justified for Δmag ≥ 4) or to request additional high-resolution imaging to resolve the host.

• The improved system can minimize wasted follow-up resources by flagging secondary-host scenarios with low detection probability (e.g., Neptune-sized planets around faint secondaries).

8. Occurrence-Rate Inference with Hierarchical Models

• Upgrade hierarchical Bayesian occurrence-rate models to include binary fraction, magnitude-difference distributions, and the provided sensitivity grids as priors.

• The improved AI can jointly infer planet occurrence rates for single vs. binary hosts, correcting for the severe incompleteness around secondary stars.

What the Improved AI System Can Do:

• Accurately estimate true planet radii and detectability in unresolved stellar binaries.

• Predict TESS (or similar mission) detection completeness for any target, including binaries.

• Produce unbiased exoplanet occurrence rates as a function of spectral type, period, and radius.

• Automatically flag binary systems where planets around the secondary are likely missed, guiding follow-up strategies.

• Generate synthetic training data for more robust transit vetting in multi-star systems.

• Provide real-time recommendations for target prioritization in ongoing and future transit surveys.