J/A+A/711/A217 Hot subdwarf binaries with deepL from Gaia and SDSS (Li+, 2026)
Identifying hot subdwarf binaries with multimodal deep learning.
Based on Gaia and SDSS data.
Li Q., Bu Y., Zhang J., Zhang Y., Wang K., Yang B., Yi Z., Kong X., Liu M.
<Astron. Astrophys. 711, A217 (2026)>
=2026A&A...711A.217L 2026A&A...711A.217L (SIMBAD/NED BibCode)
ADC_Keywords: Binaries, spectroscopic ; Stars, double and multiple ;
Stars, subdwarf ; Positional data ; Photometry ; Spectroscopy ;
Optical ; Effective temperatures
Keywords: methods: data analysis - techniques: photometric -
techniques: spectroscopic - binaries: general - subdwarfs
Abstract:
Hot subdwarf stars are important tracers of stellar structure and
evolution, while their binary systems provide key constraints on their
formation channels. Nevertheless, the number of confirmed hot subdwarf
binaries is still limited.
We aim to identify hot subdwarf binaries efficiently and reliably in
large candidate samples by combining heterogeneous observational
information.
We developed a two-stage classification framework. A Bayesian neural
network was first used to perform preliminary classification and
uncertainty quantification from Gaia photometric and astrometric data.
We then constructed a multimodal deep learning model, HsdB-FusionNet,
which combines Gaia photometric and astrometric features with SDSS
spectra through a cross-attention mechanism. We applied this framework
to a catalog of 61 585 hot subdwarf candidates. The resulting
candidates were further examined through spectral energy distribution
fitting, and atmospheric parameters were derived for high-quality
systems.
We identified 1369 hot subdwarf binary candidates from the parent
sample. Among them, 1057 objects were strongly supported as binaries
by spectral energy distribution fitting. Atmospheric parameters were
obtained for 579 high-quality binary systems. The interpretability
analysis further indicates that the model decisions are consistent
with known physical priors.
This work provides a substantially enlarged sample of hot subdwarf
binary candidates and demonstrates the potential of multimodal deep
learning for binary identification. The released catalog offers a
useful basis for future studies on the formation and evolution of hot
subdwarf binaries.
Description:
We present a catalogue of hot subdwarf binary candidates identified
with a multimodal deep-learning framework. The model combines Gaia
photometric and astrometric information with SDSS spectra and provides
binary probabilities for 1369 candidates. The catalogue also includes
uncertainty estimates from the Bayesian neural network classifier and
spectral energy distribution fitting quantities used to validate the
candidates when available.
This catalogue contains 1369 hot subdwarf binary candidates identified
with the HsdB-FusionNet framework. The table gives the ICRS position,
the final binary probability, the predictive uncertainty from the
Gaia-based Bayesian neural network (BNN), the predicted class, and the
results of spectral energy distribution (SED) fitting when available.
The SED fitting was performed with the VOSA tool. For objects with
available SED-fitting results, the table provides the chi-square
values for the single-star and binary-star fits, their difference and
ratio, and the atmospheric parameters derived from the binary fit when
a reliable parameter estimate was obtained.
File Summary:
--------------------------------------------------------------------------------
FileName Lrecl Records Explanations
--------------------------------------------------------------------------------
ReadMe 80 . This file
catalog.dat 190 1369 Astrometrics, binary probabilities, atmospheric
parameters from Chi2 SED fits VOSA tool of hot
subdwarf binary candidates identified with the
HsdB-FusionNet framework
--------------------------------------------------------------------------------
Byte-by-byte Description of file: catalog.dat
--------------------------------------------------------------------------------
Bytes Format Units Label Explanations
--------------------------------------------------------------------------------
1- 13 F13.9 deg RAdeg Right ascension (ICRS) at Epoch=J2000
(RA_ICRS)
15- 27 F13.9 deg DEdeg Declination (ICRS) at Epoch=J2000 (DE_ICRS)
29- 38 F10.8 --- Pbinary Binary probability from the final
HsdB-FusionNet model (binary_probability)
40- 53 E14.9 --- e_Pbinary Predictive uncertainty from the Gaia-based
BNN classifier (bnn_uncertainty)
55 I1 --- Binary Predicted class label; 1=binary
(predicted_class)
57- 62 A6 --- Class Text label of predicted class
(predicted_label)
64- 71 A8 --- Data Data used for final classification
(data_source)
73- 76 I4 --- ID ? Internal SED-fitting identifier (object_id)
78- 92 F15.9 --- Chi2s ? Chi-square of the single-star fit
(Chi2_single)
94-108 F15.9 --- Chi2b ? Chi-square of the binary-star fit
(Chi2_binary)
110-131 F22.16 --- Chi2d ? Chi2_single minus Chi2_binary (Delta_Chi2)
133-151 F19.15 --- Chi2r ? Chi2_single divided by Chi2_binary
(Chi2_ratio)
153-157 I5 K Teff1 ? Effective temperature of primary
(Teff1_binary) (1)
159-163 I5 K Teff2 ? Effective temperature of companion
(Teff2_binary) (1)
165-167 F3.1 [cm/s2] logg1 ? Surface gravity of primary
(logg1_binary) (1)
169-171 F3.1 [cm/s2] logg2 ? Surface gravity of companion
(logg2_binary) (1)
173-190 A18 --- SEDq SED-fitting validation flag (sed_quality) (2)
--------------------------------------------------------------------------------
Note (1): The atmospheric parameters Teff1, Teff2, logg1 and logg2 are provided
only for objects with reliable binary-fit parameter estimates.
For other objects these fields are left blank. Also, blank fields indicate
unavailable or not applicable values. The value of data_source is all_data
when the final classification used the available Gaia photometric and
astrometric information together with SDSS spectral information.
Note (2): The column sed_quality meanings are as follows:
confirmed = The candidate is supported by the binary SED-fitting
comparison, 1012 sources in our sample
failed = No reliable SED-fitting comparison or parameter estimate
is available, 275 sources in our sample
excess-confirmed = The object shows infrared excess and has an acceptable
binary SED-fitting result, 44 sources in our sample
excess-Fitted-Poor = The object shows infrared excess, but the binary SED fit
has poor fitting quality, 38 sources in our sample
--------------------------------------------------------------------------------
Acknowledgements:
Qin Li, qinlee0606(at)163.com
License: CC-BY-4.0
(End) Luc Trabelsi [CDS] 15-Jun-2026