You are on page 1of 1

26:711:564 OPTIMIZATION MODELS IN FINANCE

Professor Andrzej Ruszczynski

Optimality Conditions in Nonlinear Optimization


Let us consider the nonlinear optimization problem

min f (x)
gi (x) 0, i = 1, . . . , m, (1)
hi (x) = 0, i = 1, . . . , p.

We assume that the functions f : Rn R, gi : Rn R, i = 1, . . . , m, and hi : Rn R,


i = 1, . . . , p, are continuously differentiable. We call such a nonlinear optimization problem
smooth. The feasible set of this problem is denoted X.
Let I 0 (x) be the set of active inequality constraints at x:

I 0 (x) = {i : gi (x) = 0}.

We say that problem (1) satisfies at x the MangasarianFromovitz condition if

the gradients hi (x), i = 1, . . . , p, are linearly independent; and

there exists a direction d such that

gi (x), di < 0 i I 0 (x),


hi (x), di = 0, i = 1, . . . , p.

Problem (1) is said to satisfy Slaters condition, if the functions gi , i = 1, . . . , m are convex, the
functions hi , i = 1, . . . , p are affine, and there exists a feasible point xS such that gi (xS ) < 0,
i = 1, . . . , m.
In what follows we shall say that problem (1) satisfies the constraint qualification condition,
if either the MangasarianFromovitz or Slaters condition holds true.

Theorem
Let x be a local minimum of problem (1), where the functions f , gi and hi are continuously
differentiable. Assume that at x the constraint qualification condition is satisfied. Then there
exist multipliers i 0, i = 1, . . . , m, and i R, i = 1, . . . , p, such that
m
X p
X
f (x) + i gi (x) + i hi (x) = 0, (2)
i=1 i=1

and
i gi (x) = 0, i = 1, . . . , m. (3)
Conversely, if the functions f and gi are convex, and the functions hi are affine, then every point
x X satisfying these conditions for some 0 and Rp is an optimal solution of (1).

You might also like