Sibeller/gaia-dr3-long-period-variables

Gaia DR3 Long Period Variables

Credit: NASA/ESA/Hubble

Part of a dataset collection on Hugging Face.

Dataset description

The Gaia DR3 Long Period Variables (LPV) catalog contains ~1.7 million variable giant star candidates identified by the ESA Gaia mission's variability processing pipeline. Each source includes a dominant pulsation frequency, G-band amplitude, RP spectral variability proxy, and a carbon-star classification flag.

Long Period Variables are evolved giant stars on the Asymptotic Giant Branch (AGB) that pulsate with periods ranging from ~10 to more than 1000 days. They encompass Mira variables (large-amplitude, near-sinusoidal light curves driven by fundamental-mode pulsation), semi-regular variables (SRb/SRa, multi-periodic or irregular), and OSARG (OGLE Small Amplitude Red Giants, overtone pulsators). LPVs are important for multiple reasons: they follow tight period-luminosity relations in near-infrared bands (analogous to Cepheids) and serve as distance indicators to nearby galaxies; they are among the most prolific producers of dust and chemically enriched material in the interstellar medium; and their pulsation properties constrain AGB stellar evolution models.

Carbon stars (is_cstar=True) have carbon-to-oxygen ratios greater than one in their atmospheres due to dredge-up episodes bringing carbon from the helium-burning shell to the surface. Their RP spectra show distinctive CN and C2 molecular band signatures that differ markedly from the TiO-dominated spectra of oxygen-rich M-type AGB stars. With 1.7 million candidates, this Gaia DR3 catalog is the largest LPV compilation ever assembled, exceeding previous large-scale surveys such as OGLE, 2MASS, and WISE by an order of magnitude in sky coverage and sample size.

This dataset is suitable for tabular classification, tabular regression tasks.

Schema

Column	Type	Description	Sample	Null %
source_id	int64	Gaia DR3 unique source identifier; use for cross-matching with gaia_source to obtain sky coordinates and photometry	154863636444160	0.0%
frequency	float64	Dominant pulsation frequency in cycles per day; the primary periodicity of the LPV light curve	0.0045675368518594	77.2%
frequency_error	float64	Formal uncertainty on the pulsation frequency in cycles per day	0.00021300006	77.2%
amplitude	float64	Peak-to-peak variability amplitude in the Gaia G-band in magnitudes; large amplitudes (>1 mag) indicate Mira-type pulsators	0.99943	77.2%
median_delta_wl_rp	float64	Median wavelength shift of the RP (red photometer) spectrum relative to the template spectrum in nm; a proxy for spectral variability driven by TiO/CN molecular band changes over the pulsation cycle	5.953	1.6%
is_cstar	boolean	Boolean classification flag: True if the source is classified as a carbon star (C/O > 1, C-type AGB), False for oxygen-rich LPV (M- or S-type AGB); based on RP spectral shape	False	1.6%
period_days	float64	Dominant pulsation period in days, derived as 1/frequency; ranges from ~10 days (OSARG-type) to >1000 days (long-period Miras)	218.93638353304354	77.2%

Quick stats

• 1,720,588 LPV candidates total
• Carbon stars (is_cstar=True): 546,468 (31.8%)
• Oxygen-rich LPVs (is_cstar=False): 1,174,120 (68.2%)
• Median pulsation period: 252.3 days
• Median G-band amplitude: 0.130 mag

Usage

from datasets import load_dataset

ds = load_dataset("juliensimon/gaia-dr3-long-period-variables", split="train")
df = ds.to_pandas()

from datasets import load_dataset
import matplotlib.pyplot as plt

ds = load_dataset("juliensimon/gaia-dr3-long-period-variables", split="train")
df = ds.to_pandas()

# Carbon stars vs oxygen-rich LPVs
n_cstar = df["is_cstar"].sum()
print(f"Carbon stars: {n_cstar:,} ({100*n_cstar/len(df):.1f}%)")
print(f"Oxygen-rich LPVs: {(~df['is_cstar']).sum():,}")

# Period distribution (log scale)
df["period_days"].clip(upper=2000).hist(bins=200, log=True)
plt.xlabel("Period (days)")
plt.ylabel("Count (log scale)")
plt.title("Gaia DR3 LPV Period Distribution")
plt.show()

# Amplitude vs period scatter (random subsample)
sample = df.sample(50_000)
plt.scatter(sample["period_days"], sample["amplitude"],
            c=sample["is_cstar"].astype(int),
            cmap="coolwarm", s=1, alpha=0.3)
plt.xscale("log")
plt.xlabel("Period (days)")
plt.ylabel("G-band amplitude (mag)")
plt.title("Amplitude vs Period — Gaia DR3 LPVs")
plt.colorbar(label="Carbon star (1=True)")
plt.show()

# Cross-match with gaia_source for sky coordinates
# (source_id is the Gaia DR3 64-bit identifier)
print("Use source_id to join with gaia_source for RA/Dec, parallax, etc.")

Data source

https://gea.esac.esa.int/archive/

Related datasets

• juliensimon/gaia-dr3-cepheids

• juliensimon/gaia-dr3-rrlyrae

• juliensimon/gcvs-variable-stars

If you find this dataset useful, please consider giving it a like on Hugging Face. It helps others discover it.

About the author

Created by Julien Simon — AI Operating Partner at Fortino Capital. Part of the Space Datasets collection.

Citation

@dataset{gaia_dr3_long_period_variables,
  title = {Gaia DR3 Long Period Variables},
  author = {juliensimon},
  year = {2026},
  url = {https://huggingface.co/datasets/juliensimon/gaia-dr3-long-period-variables},
  publisher = {Hugging Face}
}

License

CC-BY-NC-3.0-IGO