Research
In one sentence: find and understand rare, hot compact stars among tens of millions of survey spectra.
Hot Subdwarf Identification
Problem
Hot subdwarfs are rare among survey spectra (a few per ten thousand); manual vetting cannot cover tens of millions of LAMOST spectra and billions of Gaia photometric and astrometric measurements. How do we find them reliably at scale?
Methods
- Gaia parallax/color cross-matching with LAMOST spectra
- Machine-learning spectral classification (attention CNN)
- MK-like spectral subclassing
Ongoing work
Extending identified samples with LAMOST DR12 and deep-learning models; packaging the pipeline as an online service.
Selected papers
- Deep learning-driven atmospheric parameter prediction for hot subdwarf stars
- Hot subdwarf stars identified in LAMOST DR8 with single-lined and composite spectra
- Hot Subdwarf Stars Identified in Gaia DR2 with Spectra of LAMOST DR6 and DR7. I.
- New hot subdwarf stars identified in Gaia DR2 with LAMOST DR5 spectra -- II
Binaries & Formation Channels
Problem
Most hot subdwarfs form through binary interactions, but the relative contribution of each channel — and how extreme systems (e.g. highly eccentric wide binaries) survive — remain open.
Methods
- Single-lined / composite binary identification and orbital solutions
- Reflection-effect systems and radial-velocity variability surveys
- Comparison against binary-evolution models
Ongoing work
Constraining orbital-parameter distributions of hot-subdwarf binaries with multi-survey radial velocities.
Selected papers
AI for Large Spectral Surveys
Problem
Classical spectral fitting needs manual initial guesses and per-target computation, which cannot scale to surveys. Can models learn to "read" spectra and return atmospheric parameters in milliseconds?
Methods
- Channel/spatial-attention CNN regression
- NLTE synthetic-grid training samples
- Large-scale catalog construction
Ongoing work
Productizing the published parameter-prediction model (this site's service) and studying uncertainty quantification.
Stellar Evolution Theory
Problem
Blue hook stars in globular clusters and the bimodality of horizontal-branch morphology resist a single stellar-evolution picture; how do binary fractions and winds shape the observed populations?
Methods
- Stellar-evolution models with tidally enhanced winds
- Synthetic horizontal-branch population synthesis
- Quantitative bimodality parameters
Ongoing work
Connecting formation-channel models to the observed parameter distributions of hot-subdwarf binaries.