Struct SolverSettings 

pub struct SolverSettings {

    pub max_iterations: usize,
    pub absolute_tolerance: f64,
    pub relative_tolerance: f64,
    pub rho: f64,
    pub sigma: f64,
    pub check_termination_every: usize,
    pub adaptive_rho: bool,
    pub adaptive_rho_interval: usize,
    pub adaptive_rho_tolerance: f64,
    pub adaptive_rho_multiplier: f64,
    pub over_relaxation: f64,
    pub minimum_rho: f64,
    pub maximum_rho: f64,
    pub scaling_iterations: usize,
    pub infeasibility_tolerance: f64,
    pub polish: bool,
    pub polish_regularization: f64,
    pub polish_refinement_iterations: usize,
}

ADMM settings.

Fields

max_iterations: usize

Maximum number of ADMM iterations.

absolute_tolerance: f64

Absolute stopping tolerance.

relative_tolerance: f64

Relative stopping tolerance.

rho: f64

Initial augmented-Lagrangian penalty.

sigma: f64

Positive proximal regularization in the x update.

check_termination_every: usize

Evaluate KKT residuals every this many iterations.

adaptive_rho: bool

Whether residual balancing may update rho.

adaptive_rho_interval: usize

Consider a penalty update every this many iterations.

adaptive_rho_tolerance: f64

Update when one ADMM residual exceeds the other by this ratio.

adaptive_rho_multiplier: f64

Multiplicative increase or decrease applied to rho.

over_relaxation: f64

Over-relaxation coefficient alpha blended into every consensus update: the slack and multiplier steps see alpha * Ax + (1 - alpha) * z_prev instead of Ax.

Must lie in the open interval (0, 2); 1.0 disables relaxation and recovers plain ADMM. Values around 1.6 typically cut iteration counts substantially on feasible convex QPs (see Boyd et al. §3.4.3 and the OSQP default).

minimum_rho: f64

Smallest penalty selected by adaptation.

maximum_rho: f64

Largest penalty selected by adaptation.

scaling_iterations: usize

Number of Ruiz equilibration passes over the problem data before iterating; 0 disables automatic scaling.

Scaling only changes the space ADMM iterates in. Termination checks and every reported residual are always evaluated on the original data.

infeasibility_tolerance: f64

Tolerance for declaring the problem primal or dual infeasible from ADMM iterate differences; 0 disables infeasibility detection.

Candidate certificates are normalized to unit infinity norm and evaluated on the original data, so the tolerance is relative: a direction qualifies when its defining residuals are below the tolerance and its gap is below the negated tolerance (see crate::check_primal_certificate / crate::check_dual_certificate). Problems infeasible by a margin smaller than this tolerance stop at SolveStatus::MaxIterations with diagnostics instead.

polish: bool

Whether to refine Solved iterates with one direct active-set solve (polishing).

The active set is guessed from the final multipliers and the resulting KKT system is solved through the same SMW reduction the iterations use, at the cost of one extra reduced factorization. The polished iterate is adopted only when its worst KKT residual — re-audited with crate::check_kkt on the original data — improves on the ADMM iterate’s, so enabling polish never degrades a solution. Solution::polished records the outcome.

polish_regularization: f64

Regularization added to the polishing KKT system so it stays factorable even when the active-set guess is degenerate.

The regularization error is removed afterwards by SolverSettings::polish_refinement_iterations rounds of iterative refinement against the unregularized system.

polish_refinement_iterations: usize

Iterative-refinement rounds applied to the polishing solve.

Trait Implementations

impl Clone for SolverSettings

fn clone(&self) -> SolverSettings

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for SolverSettings

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

Formats the value using the given formatter.

impl Default for SolverSettings

fn default() -> Self

Returns the “default value” for a type.

impl PartialEq for SolverSettings

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

Equality operator ==.

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

Inequality operator !=.

impl StructuralPartialEq for SolverSettings