Enum ProblemClass 

pub enum ProblemClass {
    Lp,
    ConvexQp,
    ConvexQcqp,
    NonconvexQp,
    Nlp,
}

The mathematical class of a loaded problem, from most to least specialized. See the module docs and dev-notes/lp-qp-routing.md.

Variants

Lp

Linear objective, linear constraints.

ConvexQp

Convex quadratic objective, linear constraints (Hessian PSD).

ConvexQcqp

Convex quadratic objective and/or convex quadratic constraints. SOCP-representable; routes to the conic (SOCP) interior-point solver.

NonconvexQp

Quadratic objective with an indefinite (sense-adjusted) Hessian and linear constraints. auto falls through to the NLP solver for a local minimum; solver_selection=qp-active-set solves it directly with the pounce-qp active-set engine, which takes an indefinite H (gh #786) and returns a local minimum.

What “local minimum” means on that path is narrower than it reads, and was narrower still before gh #848. The §4.5 inertia control shifts H -> H + delta*I so the local model is convex; it does not move the iterate, and at a saddle g = 0 makes the shifted step zero. So the engine used to stop there and certify Optimal on the first-order evidence alone — vanishing projected gradient, sign-admissible working-set multipliers — which a saddle, and the constrained maximum of min x0*x1 on x0 + x1 = 2, satisfy exactly. The engine now exhibits a direction d with A_W d = 0 and d'Hd < 0 and follows it before certifying, so an Optimal here is second-order. It is still only local: nothing on this path rules out a better minimum elsewhere.

The linear-constraints half of that is load-bearing, not descriptive: both consumers reach the model through crate::qp_extract::extract_qp_with_map, which keeps only the degree-≤1 part of every row. A quadratic row here would be silently dropped, so a model carrying one classifies Self::Nlp instead — see ClassReason::NonconvexQcqp.

Nlp

General nonlinear (transcendental terms, higher-degree polynomials, or anything the classifier cannot prove quadratic).

Implementations

impl ProblemClass

pub fn name(self) -> &'static str

Human-readable name for diagnostics and the forced-solver-mismatch error message.

Trait Implementations

impl Clone for ProblemClass

fn clone(&self) -> ProblemClass

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Copy for ProblemClass

impl Debug for ProblemClass

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

Formats the value using the given formatter.

impl Eq for ProblemClass

impl PartialEq for ProblemClass

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

Equality operator ==.

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

Inequality operator !=.

impl StructuralPartialEq for ProblemClass