KktResidual<T>

Models & Types · AiDotNet.Solvers

How badly a candidate point violates each of the Karush-Kuhn-Tucker optimality conditions.

How It Works

All four residuals are zero exactly when the point, together with its multipliers, satisfies the KKT conditions. For a convex problem that is both necessary and sufficient for optimality, so these numbers constitute a certificate: a solver’s answer can be checked independently of the solver that produced it.

Constructors

Constructor Summary
KktResidual(T stationarity, T primalFeasibility, T dualFeasibility, T complementarySlackness, T worst) Creates a KKT residual report.

Properties

Property Summary
T ComplementarySlackness { get; } Gets the largest absolute value of λ_i · (a_iᵀx − b_i) over the inequality constraints.
T DualFeasibility { get; } Gets the worst negative inequality multiplier, as a non-negative magnitude.
T PrimalFeasibility { get; } Gets the worst violation of the constraints themselves — how far outside the feasible region the point sits.
T Stationarity { get; } Gets the largest absolute component of ∇f(x) + Aᵀ_eq·ν + Aᵀ_ub·λ.
T Worst { get; } Gets the largest of the four residuals.