The Carbon Giant Population Revealed by Gaia DR3

Using the all-sky catalog of G < 16.5 mag Gaia DR3 carbon stars from Roulston et al. (2025), we investigate a uniform, high-confidence sample of carbon giants in the Milky Way, its globular clusters, and nearby Local Group galaxies. We identify 6937 carbon giants in the Large Magellanic Cloud, 2148 in the Small Magellanic Cloud, and 219 associated with the Sagittarius dwarf spheroidal, nearly quadrupling the previously known sample in the latter. We identify several candidate carbon symbiotic stars through their X-ray counterparts. The mean absolute C-AGB magnitude, MJ0 = −6.30 in the LMC and −6.15 in the SMC, agrees closely with the photometrically derived values of Ripoche et al. (2020), reinforcing the utility of carbon AGB stars as distance indicators. From 4276 Galactic carbon giants with MG 5◦, we measure a local disk space density of n0 = 11±0.5×10−8 pc−3 and a scale height of Hz = 257±4 pc. Lower-luminosity carbon giants exhibit significantly larger Gaia DR3 RUWE values than C-AGB stars, providing strong evidence that most are extrinsic post-mass-transfer systems, likely evolved dwarf carbon stars. C-AGB stars are more tightly confined to the Galactic disk and are rare inside R ≲ 7 kpc, consistent with Galactic age and metallicity gradients, whereas the lower-luminosity carbon giants have a larger scale height and require an additional halo component.

Improvements to AI systems:

1. Astrophysical Distance Calibration Model – Train a regression model on the 9,085 LMC/SMC carbon AGB stars with known mean absolute magnitudes (MJ0 = −6.30/−6.15) to predict distances to carbon giants in unresolved or dusty regions, improving 3D mapping of the Milky Way’s outer disk and halo.

2. Classification Pipeline for Extrinsic vs. Intrinsic Carbon Stars – Use the RUWE (astrometric noise) discrepancy between lower-luminosity carbon giants (high RUWE) and C-AGB stars (low RUWE) to build a binary classifier that flags post-mass-transfer systems (e.g., dwarf carbon stars) in large photometric surveys, enabling automated identification of binary evolution products.

3. Galactic Structural Prior for Density Estimation – Incorporate the measured disk scale height (Hz = 257 pc) and local space density (n0 = 11×10−8 pc−3) as Bayesian priors in a hierarchical model that infers stellar population distributions from sparse Gaia data, improving predictions of carbon star counts in unobserved sky regions.

4. Metallicity-Gradient-Aware Generator – Use the finding that C-AGB stars are rare inside R < 7 kpc (due to age/metallicity gradients) to condition a generative model (e.g., GAN or diffusion model) that synthesizes synthetic carbon star catalogs with realistic radial distributions, useful for testing survey completeness or training detection algorithms.

5. Symbiotic Star Candidate Finder – Leverage the cross-match between carbon stars and X-ray counterparts to train a multi-wavelength classifier that predicts symbiotic star candidacy from Gaia photometry + X-ray flux alone, enabling rapid triage in future all-sky X-ray surveys (e.g., eROSITA).

6. Distance Indicator for Local Group Galaxies – Apply the calibrated C-AGB absolute magnitude as a standard candle in a neural network that estimates distances to dwarf galaxies from resolved stellar photometry, reducing reliance on Cepheids or RR Lyrae in metal-poor, dusty environments.

What the improved AI system can do:

• Automatically map the Milky Way’s carbon star distribution in 3D with 5–10× better distance precision than current photometric methods.

• Instantly classify millions of unresolved stellar sources into intrinsic C-AGB vs. extrinsic post-mass-transfer systems, enabling population synthesis of binary evolution.

• Predict carbon star counts and spatial density in unobserved Galactic regions, optimizing target selection for future spectroscopic surveys.

• Generate realistic synthetic carbon star catalogs for training detection algorithms or testing survey biases.

• Flag new symbiotic star candidates in real-time from multi-wavelength alerts, accelerating follow-up observations.

• Provide accurate distances to nearby dwarf galaxies using only optical photometry, independent of variable-star methods.

Classical carbon stars (atmospheric C/O >1) are asymptotic giant branch (AGB) stars that become carbon-rich through third dredge-up during the thermally pulsing AGB phase. However, carbon stars also occur among red giants and main-sequence stars, where the carbon is thought to originate from binary mass transfer from a former AGB companion. Using the all-sky catalog of G<16.5 mag Gaia DR3 carbon stars from Roulston et al. (2025), we investigate a uniform, high-confidence sample of carbon giants in the Milky Way, its globular clusters, and nearby Local Group galaxies. We identify 6937 carbon giants in the Large Magellanic Cloud, 2148 in the Small Magellanic Cloud, and 219 associated with the Sagittarius dwarf spheroidal, nearly quadrupling the previously known sample in the latter. We identify several candidate carbon symbiotic stars through their X-ray counterparts. The mean absolute C-AGB magnitude, M 0 J=-6.30 in the LMC and-6.15 in the SMC, agrees closely with the photometrically derived values of Ripoche et al. (2020), reinforcing the utility of carbon AGB stars as distance indicators. From 4276 Galactic carbon giants with M G<0 and b>5, we measure a local disk space density of n 0=11 plus or minus0.5 times10-8, pc-3 and a scale height of H z=257 plus or minus4 pc. Lower-luminosity carbon giants exhibit significantly larger Gaia DR3 RUWE values than C-AGB stars, providing strong evidence that most are extrinsic post-mass-transfer systems, likely evolved dwarf carbon stars. C-AGB stars are more tightly confined to the Galactic disk and are rare inside R 7 kpc, consistent with Galactic age and metallicity gradients, whereas the lower-luminosity carbon giants have a larger scale height and require an additional halo component.