Struct Subproblem 

pub struct Subproblem<'a> { /* private fields */ }

The trust-region subproblem at one outer iteration, in scaled coordinates.

Holds the column-scaled Jacobian J̃ and scaled gradient g̃; a solver returns a scaled step δ̃ with ‖δ̃‖ ≤ Δ.

Implementations

impl<'a> Subproblem<'a>

pub fn new(jacobian: MatRef<'a, f64>, gradient: MatRef<'a, f64>) -> Self

Build a subproblem view over the scaled Jacobian and scaled gradient.

pub fn n_params(&self) -> usize

Number of free parameters p.

pub fn gradient(&self) -> MatRef<'_, f64>

The scaled gradient g̃ (p×1).

pub fn jacobian(&self) -> MatRef<'_, f64>

The column-scaled Jacobian J̃ (m×p). Dense methods (dogleg) form the Gauss–Newton system from it; Krylov methods should prefer hvec.

pub fn jvec(&self, v: MatRef<'_, f64>) -> Mat<f64>

Matrix-free product J̃·v (m×1) — never forms J̃ᵀJ̃.

pub fn jtvec(&self, u: MatRef<'_, f64>) -> Mat<f64>

Matrix-free product J̃ᵀ·u (p×1).

pub fn hvec(&self, v: MatRef<'_, f64>) -> Mat<f64>

Hessian–vector product H·v = J̃ᵀ(J̃·v) (p×1) without forming H.

pub fn predicted_reduction(&self, s: MatRef<'_, f64>) -> f64

Predicted reduction of ½‖r‖² for a scaled step δ̃: −g̃ᵀδ̃ − ½‖J̃δ̃‖² (positive for a descent step).