Research Question
- How do binary black holes form across isolated and dynamical channels?
- What signatures in LIGO/Virgo observations discriminate between formation pathways?
Methodology
- Physics-informed deep learning with population synthesis priors (COMPAS, COSMIC, POSYDON)
- Simulation-based inference to map simulator outputs → observed GW parameters
- Domain adaptation to bridge simulated vs. detector data distributions
- Cross-modal Transformer attention to interrogate common-envelope efficiency effects
Falsification Logic
- Compare posterior predictive distributions against held-out GW events
- Stress-test with altered metallicity, kick prescriptions, and CE efficiencies
- Quantify epistemic vs. aleatoric uncertainty; reject models that overfit noise or collapse uncertainty
Current Status
- Prototype SBI pipeline trained on ensemble simulators
- Early domain-adaptation results aligning simulator and LIGO/Virgo feature spaces
- Building interpretable attention maps for parameter importance
Future Directions
- Expand to additional formation channels and spin evolution models
- Integrate hierarchical population inference for rate estimation
- Publish reproducible pipelines with open configs and figure scripts
