Skip to content

Constraint implementation

Low-level constraint violation helpers. For user-defined constrained problems, use the public problem-definition workflow instead.

Internal implementation API

These deep modules are not stable public imports. Documentation does not create a compatibility guarantee. See the stability policy.

Public problem definition · Constraints guide

ConstraintInfo dataclass

compute_constraint_info(G, eps=0.0)

Compute aggregate constraint violation and feasibility mask.

Parameters:

Name Type Description Default
G NDArray[float64] | None

Constraint values where <= 0 means satisfied.

required
eps float

Feasibility tolerance. Constraints <= eps are treated as satisfied.

``0.0``

Utility helpers for constraint handling.

compute_violation(G, *, n=None, eps=0.0)

Sum per solution of each constraint's excess over eps; assumes G shape (N, n_constraints), g<=0 satisfied.

When G is None (unconstrained), n must be provided so the output length is explicit.

eps is the feasibility tolerance of :func:is_feasible: the violation is sum(max(g - eps, 0)), so a solution has zero violation exactly when it is feasible with the same eps (default 0.0).

is_feasible(G, *, n=None, eps=0.0)

Boolean feasibility mask; assumes G shape (N, n_constraints).

When G is None (unconstrained), n must be provided so the output length is explicit.

eps is a feasibility tolerance: constraints with g(x) <= eps are treated as satisfied (default 0.0).