The Kuhn-Tucker conditions have been used to derive many significant results in economics, particularly in decision problems that occur in static situations, for instance, to show the existence of an equilibrium for a competitive economy (Negishi [1] ), to carry out the first-order approach to principal-agent problems (Rogerson [2] ), and to examine the need for land reform (Grossman [3] ). Also, the Kuhn-Tucker conditions and/or the method of Lagrange multipliers are usually contained in standard microeconomics textbooks, for instance, Mas-Colell, Whinston and Green [4] , where the Kuhn-Tucker conditions for the problem with both inequality and equality constraints are discussed.

The Kuhn-Tucker conditions for the optimization problem with inequality and equality constraints have a comprehensive form that incorporates the method of Lagrange multipliers (introduced by Lagrange in 1788) in a natural way; therefore, the simple derivation of the Kuhn-Tucker conditions would shed light on the problem’s true nature.

In this paper, the Kuhn-Tucker conditions under the Mangasarian-Fromovitz constraint qualification are derived directly by applying a corollary of Farkas’s lemma without resorting to the Fritz John conditions and the boundedness of Lagrange multipliers is also shown.

The problem to be addressed is as follows:

where, , and are continuously Fréchet differentiable functions, and, (). If there are no inequality (equality) constraints, we think that ().

Here, we should pay attention to the fact that the problem (P) naturally includes the optimization problem with equality constraints considered by Lagrange in the 18th century.

We postulate the following Mangasarian-Fromovitz constraint qualification (MF) (Mangasarian and Fromovitz [5] ) in association with (P). We define.

(MF) For, , are linearly independent, and there exists an s.t. and.

Remarks

Ÿ The linearly independent constraint qualification, which is usually assumed in practice, implies (MF) (see Nocedal and Wright [6] ).

Ÿ (MF) is equal to the Cottle constraint qualification without the presence of equality constraints, and if the problem (P) is a concave program without equality constraints, the Slater constraint qualification implies the Cottle constraint qualification (Bazaraa and Shetty [7] ).

Finally, we recall the following result to the linear system including equalities for the sake of convenience.

Lemma 1. ([7] , Corollary 2 to Theorem 2.3.5) For, , and, either

(a), , ,

or

(b), and,

but never both.

We now establish the main result, which differs from [7] Theorem 5.3.1 in that our result includes complementarity conditions and the boundedness of Lagrange multipliers under (MF).

Theorem 1. Suppose that is a local solution for (P), and that the constraint qualification (MF) holds at. Then, it holds that for and,

and, are bounded.

Proof. At a local solution, if we choose in the feasible region such that,

which shows that satisfies, and,.

Then, for a local solution, it follows that

does not hold, since, if so, as for, which con-

tradicts the local optimality of at.

Note that (MF) guarantees the existence of such from the implicit function theorem ([8] , Appendix D-3 with and) if; otherwise (2) does not hold for. By applying Lemma 1 to (2) and (MF), even if the active constraints are empty, we obtain

or, equivalently,

for,.

The rest part of the proof is as follows. From (3) we obtain

So, if,

and if, vanishes. In any case, (3) reduced to

Since, are linearly independent by (MF), is determined to a single bounded vector.

Example 1.

Consider the problem

with an optimal solution. At, the linearly independent constraint qualification does not hold, whereas (MF) holds for.

Indeed, (MF) is valid for problems with a number of inequality constraints and admits feasible directions around in the orthogonal complementary space of,.

In this paper, the Kuhn-Tucker conditions under the Mangasarian-Fromovitz constraint qualification were derived directly by applying a corollary of Farkas’s lemma without resorting to the Fritz John conditions, or without introducing the Bouligand tangent cone, and the boundedness of Lagrange multipliers was also shown.

Considerable effort has been devoted to the generalization of Farkas’s lemma. However, what seems to be lacking is a discrete version of Farkas’s lemma under a mild condition; such a version would be theoretically meaningful and would be help solve the discrete optimization problems that emerge in the economics studying indivisible goods.