Struct Tolerance 

pub struct Tolerance {
    pub residual: f64,
    pub step: f64,
    pub max_iterations: usize,
    pub max_backtracks: usize,
    pub max_sweeps: usize,
}

How hard the solver tries.

Two criteria, because a residual alone is not scale-free. A residual is in the instrument’s own money, so the pillar solve reads it against the rung’s money_scale — the notional a CDS upfront rung carries in money_per_quote, one for the unit-notional par rungs the templates build — and residual is a tolerance per unit of that scale. Even so scaled it can be asked below what the pricing arithmetic can express, so the step criterion is the backstop: it is on the unknown itself, a log discount factor (or hazard, or index level) of order one, and is meaningful whatever the notional.

A pillar is solved when either holds: the residual is small against the rung’s money scale, or the Newton step has stopped moving the parameter.

Fields

residual: f64

Residual below which a pillar is solved, per unit of the rung’s money_scale. The default, 1e-12, leaves a template ladder’s par quote reading back to about the quote’s own last digit, where 1e-8 left par rates off by 1e-10 — visible in any P&L explain that reprices the rungs.

step: f64

Newton step below which the parameter is taken as converged, in log-discount-factor space.

max_iterations: usize

Iterations allowed per pillar before reporting no convergence.

max_backtracks: usize

How many times a single Newton step may be halved when the full step would make the residual worse.

The pillar solve has no bracket to fall back into, so this is its only defence against a step that lands where the objective is exponentially wrong — the failure a curve with a bending short end produces.

max_sweeps: usize

How many full-curve sweeps the fit may take before giving up.

Separate from max_iterations, which bounds a single pillar’s Newton solve, because under a global scheme these are different budgets: the sweeps are the fit, and a fit needing more passes says nothing about any one pillar needing more steps. Conflating them meant a caller who wanted to bound the fit starved the Newton solves first.

Trait Implementations

impl Clone for Tolerance

fn clone(&self) -> Tolerance

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Copy for Tolerance

impl Debug for Tolerance

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

Formats the value using the given formatter.

impl Default for Tolerance

fn default() -> Self

Returns the “default value” for a type.