Optimization

The heterodyne optimization pipeline follows a two-stage workflow:

1. NLSQ warm-start – Fast non-linear least squares via scipy.optimize.least_squares with JAX JIT-compiled residuals and Jacobians. Provides point estimates and covariance for posterior initialization.

2. CMC (Bayesian MCMC) – Full posterior sampling via NumPyro NUTS, initialized from the NLSQ warm-start. Produces credible intervals, convergence diagnostics, and ArviZ-compatible inference data.

• NLSQ Optimization
  • Core Fitting
  • Configuration
  • Results
  • Adapter
  • JIT Strategies
  • Fourier Reparameterization
  • CMA-ES Wrapper
  • Multi-Start Optimizer
  • Input Validation & Fit Quality
• CMC (Bayesian)
  • Core Fitting
  • Configuration
  • Results
  • NumPyro Model
  • Priors
  • Sampler
  • Diagnostics
  • Reparameterization
  • Scaling
  • Data Preparation
  • CMC I/O
  • CMC Plotting
  • Backends