Struct ArrowProximalCorrectionOptions 

pub struct ArrowProximalCorrectionOptions {
    pub initial_ridge: f64,
    pub ridge_growth: f64,
    pub max_attempts: usize,
    pub armijo_c1: f64,
    pub gradient_tolerance: f64,
    pub convergence_objective_rel_tol: f64,
}

Globalization guard for non-convex arrow-Schur inner steps.

The raw Schur solve is exactly Newton. For non-convex analytic penalties, full Newton can cycle. This controller adds a proximal LM shift mu I to both blocks and accepts only Armijo-decreasing trial points.

Fields

initial_ridge: f64

ridge_growth: f64

max_attempts: usize

armijo_c1: f64

gradient_tolerance: f64

convergence_objective_rel_tol: f64

Relative objective resolution below which the proximal correction declares convergence instead of failing.

Near a stationary point the largest decrease the damped Newton model can still achieve shrinks to the floating-point resolution of the objective itself: at proximal ridge μ → μ_max the accepted step length is O(‖g‖ / μ), so the realised change in the objective falls below rel_tol · (|f| + 1). At that scale the Armijo sufficient-decrease test compares two values that differ only by rounding noise, and no further productive decrease is achievable. Rather than raise AdaptiveCorrectionFailed, the loop then returns the incumbent state (a zero step) as converged. This does NOT mask genuine non-convergence: it triggers only when every attempted step either fails to decrease the objective by more than this resolution OR increases it by no more than this resolution (pure rounding). A step that genuinely reduces the objective is always taken first.

Trait Implementations

impl Clone for ArrowProximalCorrectionOptions

fn clone(&self) -> ArrowProximalCorrectionOptions

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for ArrowProximalCorrectionOptions

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

Formats the value using the given formatter.

impl Default for ArrowProximalCorrectionOptions

fn default() -> Self

Returns the “default value” for a type.