Enum JointNewtonTerminalReason 

pub enum JointNewtonTerminalReason {
    CycleBudget,
    FullyRejectedExactFixedPoint {
        consecutive_cycles: usize,
        joint_trust_radius: f64,
        rejection_counts: [usize; 4],
    },
    FullyRejectedAtTrustRegionFloor {
        consecutive_cycles: usize,
        joint_trust_radius: f64,
        rejection_counts: [usize; 4],
    },
    SlowGeometricRate {
        rate_per_cycle: f64,
        window_cycles: usize,
        projected_cycles_to_tolerance: usize,
        residual: f64,
        residual_tol: f64,
        ray: Option<RayRestoration>,
    },
    StalledOnDescendingRay {
        residual: f64,
        residual_tol: f64,
        cycles: usize,
        ray: RayRestoration,
    },
}

Variants

CycleBudget

FullyRejectedExactFixedPoint

Fields

consecutive_cycles: usize

joint_trust_radius: f64

rejection_counts: [usize; 4]

FullyRejectedAtTrustRegionFloor

Fields

consecutive_cycles: usize

joint_trust_radius: f64

rejection_counts: [usize; 4]

SlowGeometricRate

The residual was still contracting — every step accepted, the model trusted — but at a geometric rate too slow to reach tolerance within the projection cap. The solve was descending, not stuck: on the #2695 1569 pair this is the scale coefficient walking the σ→0 ray, an objective with no finite minimizer along that direction at that ρ, and the outer needs to read it as under-penalization rather than as a failed seed.

Fields

rate_per_cycle: f64

window_cycles: usize

projected_cycles_to_tolerance: usize

residual: f64

residual_tol: f64

ray: Option<RayRestoration>

The block whose penalty is too weak to close the ray, and the strength at which it would: None when no block’s penalty opposes the accepted step (an unpenalized ray, which no ρ can close).

StalledOnDescendingRay

The solve left on a residual-stall or divergence guard while its last accepted step was still descending a direction no block’s penalty closes (gam#2695).

Read exactly like Self::SlowGeometricRate’s ray: the seed is under-penalized, not failed, and the outer restores it rather than discarding it. It is a SEPARATE reason because these exits certify nothing about a rate — the residual was flat or growing, not contracting slowly — and reporting them as a slow rate would claim a contraction that was not measured. On the #2695 1569 pair seeds 0-3 leave here (early-exit non-converged (divergence/stall guard)) with every step accepted at a good model ratio, the objective still falling and ‖β‖∞ still growing, which is the ray this names.

Fields

residual: f64

residual_tol: f64

cycles: usize

ray: RayRestoration

Trait Implementations

impl Clone for JointNewtonTerminalReason

fn clone(&self) -> JointNewtonTerminalReason

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Copy for JointNewtonTerminalReason

impl Debug for JointNewtonTerminalReason

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

Formats the value using the given formatter.

impl Display for JointNewtonTerminalReason

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

Formats the value using the given formatter.

impl PartialEq for JointNewtonTerminalReason

fn eq(&self, other: &JointNewtonTerminalReason) -> bool

Equality operator ==.

fn ne(&self, other: &Rhs) -> bool

Inequality operator !=.

impl StructuralPartialEq for JointNewtonTerminalReason