Skip to content

Nondegenerate Constraint Qualification

Definition

Write the constraint map as h=(h1,…,hm)\boldsymbol{h}=(h_1,\ldots,h_m). For equality constraints h1(x)=c1,…,hm(x)=cmh_1(\boldsymbol{x})=c_1,\ldots,h_m(\boldsymbol{x})=c_m, the nondegenerate constraint qualification is satisfied at x∗\boldsymbol{x}^\ast if

rank⁡(Dh(x∗))=m.\operatorname{rank}(\boldsymbol{D}\boldsymbol{h}(\boldsymbol{x}^\ast))=m.

With inequality constraints, the same rank condition is applied to the gradients of the binding inequality constraints together with the equality constraints.


Interpretation

NDCQ says the active constraints are locally independent. If it fails, some constraint may be redundant or the feasible set may have a singular shape.

Two panels of two constraint curves. Left: the curves cross and their gradients are independent, so the qualification holds. Right: the curves touch and their gradients are parallel, so it fails.

On the left, the constraints cross transversally and rank⁡Dh(x∗)=2=m\operatorname{rank}\boldsymbol{D}\boldsymbol{h}(\boldsymbol{x}^\ast)=2=m, so the qualification holds. On the right, they touch with parallel gradients and rank⁡Dh(x∗)=1<m\operatorname{rank}\boldsymbol{D}\boldsymbol{h}(\boldsymbol{x}^\ast)=1<m, so it fails.

The Jacobian collects the constraint gradients as rows. Its rank equals the number of active constraints exactly when these rows are independent, excluding a nonzero linear relation between their first-order effects.


Key Results


Examples

  • For h(x,y)=x+yh(x,y)=x+y, NDCQ holds at every point because the gradient is nonzero.
  • For h1(x,y)=x+yh_1(x,y)=x+y and h2(x,y)=2x+2yh_2(x,y)=2x+2y, NDCQ fails because the gradients are dependent.

Prerequisites

Sources

Aliases: NDCQ, LICQ Keywords: constraint qualification, active constraint rank condition, linear independence constraint qualification

Last updated: 26 October 2026 00:09