Glossary term
KKT Conditions
Necessary optimality conditions for constrained optimization problems under suitable regularity assumptions.
Definition
theoremThe Karush-Kuhn-Tucker conditions are first-order optimality conditions for constrained optimization problems with equality and inequality constraints.
KKT conditions generalize the method of Lagrange multipliers to problems with inequality constraints. Under constraint qualification assumptions, a local optimum must satisfy stationarity, primal feasibility, dual feasibility, and complementary slackness. For convex optimization problems with suitable regularity, these conditions are also sufficient for global optimality. They are central to nonlinear programming, quadratic programming, optimal control, machine learning, structural optimization, and engineering design.
KKT conditions describe what must be true at an optimum of many constrained optimization problems. For a problem with objective f(x), inequality constraints g_i(x) \leq 0, and equality constraints h_j(x) = 0, the Lagrangian is:
where \lambda_i and \mu_j are Lagrange multipliers. The KKT conditions state that, at a qualified optimum, four groups of conditions must hold.
The four conditions
Stationarity requires the gradient of the Lagrangian with respect to the decision variables to be zero:
Primal feasibility requires the original constraints to be satisfied:
Dual feasibility requires inequality multipliers to be nonnegative:
Complementary slackness requires each inequality multiplier to be active only when its constraint is active:
If an inequality constraint is inactive, its multiplier is zero. If a multiplier is positive, the corresponding constraint is binding.
Engineering interpretation
KKT multipliers often have practical meaning. In structural optimization, a multiplier can indicate how much the objective would improve if a stress or displacement limit were relaxed. In operations research, it can represent a shadow price. In control and machine learning, KKT conditions explain active constraints, margins, and constrained parameter updates.
For convex problems, KKT conditions are especially powerful because they can certify global optimality under suitable regularity. For nonconvex problems, they are necessary conditions for local optima but do not guarantee that the point is the best solution.
Common mistakes
A common mistake is applying KKT conditions without checking constraint qualifications. If constraints are irregular, redundant, nonsmooth, or degenerate, the conditions may not hold in the usual form. Another mistake is interpreting a KKT point as automatically optimal in a nonconvex problem. It may be a local minimum, local maximum, saddle point, or numerically poor stationary point. Good optimization reports state the formulation, constraint activity, multipliers, tolerances, solver method, and whether convexity or other sufficiency arguments apply.