Struct SqpAlgorithm 

pub struct SqpAlgorithm { /* private fields */ }

SQP-side algorithm driver.

Implementations

impl SqpAlgorithm

pub fn new(qp_solver: ParametricActiveSetSolver, opts: SqpOptions) -> Self

pub fn with_qp_options(self, qp_opts: QpOptions) -> Self

Override the per-call QP-solver options. Defaults are the pounce_qp::QpOptions::default() (which include the use_schur_updates = false and anti_cycling = Expand from Phase 5a.2). Callers can pin tighter tolerances or flip use_schur_updates = true for warm-started workloads.

pub fn options(&self) -> &SqpOptions

pub fn iterates(&self) -> Option<&SqpIterates>

pub fn optimize<N: SqpProblemSpec>( &mut self, nlp: &mut N, ) -> Result<SqpResult, SqpError>

Run the SQP loop to convergence (or max_iter). Cold-starts the iterate from nlp.x_init() and an empty working set.

pub fn optimize_with_warm_start<N: SqpProblemSpec>( &mut self, nlp: &mut N, warm: Option<SqpIterates>, ) -> Result<SqpResult, SqpError>

Warm-start variant. warm = Some(prev) seeds the iterate from prev.{x, lambda_g, lambda_x, working} instead of the NLP’s cold defaults. Dimensions are validated against the problem; any mismatch is fatal. The QP solver consumes warm.working (when present) via solve_with_working_set.

warm = None is equivalent to Self::optimize.

Implements the §6 design-note warm-start contract: the tuple (x, λ_g, λ_x, 𝒲). The Hessian carry-forward (damped-BFGS / L-BFGS state) is not part of the warm-start payload — each optimize call rebuilds its own Hessian approximation from scratch.