Struct ConstrainedSubspaceKernel 

pub struct ConstrainedSubspaceKernel<'a> { /* private fields */ }

Per-evaluation handle that combines a penalty-projected PenaltySubspaceTrace with an active inequality-constraint block, producing the constraint-aware pseudo-inverse K_T = K_S − K_S Aᵀ (A K_S Aᵀ)⁻¹ A K_S. See PenaltySubspaceTrace::with_active_constraints for the math.

Caches the small k_active × k_active Schur inverse so subsequent per-coordinate apply calls only do O(p · k_active) work each.

Implementations

impl<'a> ConstrainedSubspaceKernel<'a>

pub fn apply_pseudo_inverse(&self, a: &Array1<f64>) -> Array1<f64>

Apply K_T = K_S − K_S Aᵀ (A K_S Aᵀ)⁻¹ A K_S to a. The result lies in range(S₊) ∩ ker(A_act) — the smooth manifold the inner solver actually moves on at a constrained-stationary point.

pub fn has_active_constraints(&self) -> bool

Whether any active constraints contribute (when false this kernel is identical to the bare PenaltySubspaceTrace::apply_pseudo_inverse).