Module batch_ls 

Gauss–Newton weighted batch least squares — the batch differential corrector.

This is the estimation core a batch orbit determination (or any parameter-fit) uses: given measurements z that depend on an unknown state x through a model h(x), it linearises h about the current estimate (a finite-difference Jacobian H = ∂h/∂x), forms and solves the weighted normal equations (HᵀWH)·Δx = HᵀW·(z − h(x)), and iterates to convergence.

It is deliberately generic over the model closure, so it is independent of any particular measurement set. For orbit determination the state would be the satellite epoch state (and a receiver clock bias) and h the range / range-rate / azimuth-elevation model from a ground station; here the engine and its convergence are delivered and tested, while the orbit-specific measurement model, the analytic state-transition matrix, and the published-case validation are follow-ons (see ROADMAP.md).

Structs

LsqResult

The outcome of a batch least-squares solve.

Functions

gauss_newton

Solve z ≈ h(x) by Gauss–Newton iteration with per-measurement weights weights (e.g. 1/σ²), starting from x0. Returns None on a dimension mismatch or a singular normal matrix (rank-deficient geometry).