Struct Report 

pub struct Report {
    pub termination: Termination,
    pub n_iter: usize,
    pub n_residual_evals: usize,
    pub n_jacobian_evals: usize,
    pub cost: f64,
    pub gradient_norm: f64,
    pub cost_history: Vec<f64>,
    pub gradient_norm_history: Vec<f64>,
    pub params_history: Vec<Vec<f64>>,
}

Outcome of a solve. The optimised parameters live in the problem (read via TrustRegionProblem::params); this only carries diagnostics.

Not Copy: it carries the per-iteration convergence trajectory (cost_history / gradient_norm_history) as owned Vecs. These are observability only — they do not affect the optimisation and are recorded once per accepted point plus the terminal point.

Fields

termination: Termination

Why the loop stopped.

n_iter: usize

Accepted iterations.

n_residual_evals: usize

Residual evaluations.

n_jacobian_evals: usize

Jacobian evaluations.

cost: f64

Final ½‖r‖².

gradient_norm: f64

‖Jᵀr‖_∞ at the most recent point where the Jacobian was evaluated — the start of the final outer iteration.

This is the definition for every termination reason, and it is what makes the number comparable across them. It is deliberately not “‖Jᵀr‖_∞ at the returned parameters”: on an Ftol/Xtol/MaxEval stop the loop exits after accepting a step, and no Jacobian has been evaluated at that new point, so reporting a gradient there would mean either paying for an extra Jacobian evaluation or pairing the stale J with the new r — a product that is the gradient at neither point. The drivers therefore report the last genuinely-evaluated gradient instead. On a Gtol stop the two coincide, since that test fires before any step is taken.

cost_history: Vec<f64>

½‖r‖² at each accepted point (index 0 = initial), ending at the terminal cost. Empty only for an immediate pre-iteration numerical failure.

gradient_norm_history: Vec<f64>

‖Jᵀr‖_∞ recorded alongside each cost_history entry, under the same “at the most recent Jacobian evaluation” definition as gradient_norm — so the terminal entry repeats the previous iteration’s gradient when the loop stopped after an accepted step.

params_history: Vec<Vec<f64>>

The free-parameter vector θ at each accepted point, recorded alongside each cost_history entry (same length and ordering). This is the raw material for the convergence-to-truth metric dₖ = ‖(θₖ − θ_true)/s‖₂ on synthetic cases — observability only, it does not affect the solve. Empty for solvers that do not track it (only the faer LM driver records it today; the trust-region / dogleg / newton-cg drivers leave it empty).

Trait Implementations

impl Clone for Report

fn clone(&self) -> Report

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for Report

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.