Astronomical Cardiology II: A Search For Heartbeat Stars Using APOGEE and TESS

Stellar binaries in short-period, highly eccentric systems with significant tidal deformations near pericenter, also known as heartbeat stars, are a laboratory for studying dynamical tides and oscillations in stars. We identify 50 heartbeat stars using TESS light curves of stars identified as binaries using SDSS APOGEE. We fit their phase-folded TESS light curves with an analytic model to measure their orbital periods, eccentricities, inclinations, and arguments of periastron. We measure the mass function of targets with enough APOGEE radial velocity observations to obtain a constraint on the secondary mass. We confirm our previous results that the non-giant heartbeat stars have started to evolve off the main sequence and that the fraction of (near) main sequence binaries that are heartbeat stars rises rapidly with effective temperature.

Improvements to AI Systems:

1. Automated Classification of Heartbeat Stars from Light Curves

• Train a deep learning model (e.g., CNN or transformer) on TESS light curves to automatically detect the characteristic periastron brightening and tidal distortion patterns, reducing the need for manual fitting.

• The improved system can scan millions of light curves in real time, flagging candidate heartbeat stars with high recall and precision, even in noisy or sparse data.

2. Analytic Model Fitting with Uncertainty Quantification

• Replace the current analytic fitting with a Bayesian neural network or normalizing flow that directly predicts orbital parameters (period, eccentricity, inclination, argument of periastron) and their posterior distributions from phase-folded light curves.

• The improved system can output probabilistic constraints for each parameter, enabling robust comparisons across stellar populations and automatic outlier detection.

3. Secondary Mass Inference from Sparse Radial Velocity Data

• Implement a Gaussian process or variational autoencoder to interpolate and predict radial velocity curves from limited APOGEE observations, then use a physics-informed neural network to derive the mass function and secondary mass.

• The improved system can estimate secondary masses for stars with as few as 3–5 RV points, expanding the sample size and enabling population studies of low-mass companions.

4. Evolutionary State Classification

• Use a gradient-boosted decision tree or random forest on photometric, spectroscopic, and derived orbital features to classify heartbeat stars as main-sequence, subgiant, or giant, and to predict the fraction of main-sequence binaries that are heartbeat stars as a function of effective temperature.

• The improved system can automatically map the evolutionary status of thousands of binaries, revealing trends in tidal interaction strength with stellar age and mass.

5. Population Synthesis and Tidal Dissipation Modeling

• Develop a generative model (e.g., a variational autoencoder or normalizing flow) trained on the observed parameter distributions to simulate synthetic heartbeat star populations, then use reinforcement learning to optimize tidal dissipation parameters (e.g., Q-factor) that match observations.

• The improved system can predict the expected number and properties of heartbeat stars in untested stellar mass and temperature ranges, guiding future survey designs.

6. Cross-Survey Data Fusion

• Build a multi-modal transformer that ingests TESS light curves, APOGEE spectra, Gaia astrometry, and photometric colors simultaneously to jointly infer orbital and stellar properties.

• The improved system can automatically identify and correct for systematic biases between surveys, producing a unified catalog of heartbeat stars with consistent parameters, and can discover new candidates missed by single-survey analyses.

Stellar binaries in short-period, highly eccentric systems with significant tidal deformations near pericenter, also known as heartbeat stars, are a laboratory for studying dynamical tides and oscillations in stars. We identify 50 heartbeat stars using TESS light curves of stars identified as binaries using SDSS APOGEE. We fit their phase-folded TESS light curves with an analytic model to measure their orbital periods, eccentricities, inclinations, and arguments of periastron. We measure the mass function of targets with enough APOGEE radial velocity observations to obtain a constraint on the secondary mass. We confirm our previous results that the non-giant heartbeat stars have started to evolve off the main sequence and that the fraction of (near) main sequence binaries that are heartbeat stars rises rapidly with effective temperature.