For a function of two or more variables, the surface at a saddle-point resembles a saddle that curves up in one or more directions, and curves down in one or more other directions. ....For example, two hills separated by a high pass will show up a saddle point, at the top of the pass,....
A saddle point is a very important idea for economics, though it is somewhat mathematical. For econ learners, it is necessary to keep it in mind.
An econ example is about the consumer's utility maximization problem(UMP):
Suppose he or she had two kinds of goods: apples and oranges. The price of an apple is p, and the price of an orange is q. The number of apples is A, and the number of oranges is O. He or she has a utility function like U(A,O). It expresses his preference and assumes that the more oranges or apples he gets the happier he becomes. His income is M. And then his or her utility maximization problem is:
max U(A,O)
such that pA+qO≦M
When an economist usually solves the above problem, she uses the Lagrangian function:
L(A,O,l)=U(A,O) - l(pA+qO-M)
where l is the Lagrange multiplier, or the marginal utility of income.(If you see it as the marginal utility of income, differenciate L with respect to M and you get l. As M increases L also increases. The multiplier l is the rate of change of L to a small change in M. )
And she differentiates L with respect to A,O and l to get the first-order necessary conditions for the maximization problem.
f.o.c.
dL/dA=dU/dA - lp=0....(1)
dL/dO=dU/dO - lq=0....(2)
dL/dl=- (pA+qO-M)≧0,
l(pA+qO-M)=0 and l≧0....(3)
The above conditions are called "the Kuhn-Tucker necessary conditions". (Note that dL/dx means the partial derivative.) The Kuhn-Tucker conditions (or the Kuhn-Tucker Theorem) say that there exists l such that dL/dA=dU/dA - lp=0, dL/dO=dU/dO - lq=0 and dL/dl=- (pA+qO-M)=0.
And the condition (3) says that l>0 and -(pA+qO-M)>0 cannot hold at the same time. (Of course, l=0 and (pA+qO-M)=0 can hold simultaneously.)
If l=0, then -(pA+qO-M)≧0. In this case, this problem is unconstrained. And so you can solve the problem just by equating the derivatives of U(A,O) with respect to A and O to zero.
If l>0, then -(pA+qO-M)=0. In this case, this problem is constrained to income. And so you can solve the problem by solving the simultaneous equations of A,O and l, (1), (2) and (3).
Well, I'll reach the conclusion soon. The optimal solutions (A*,O*,l*) are at a saddle point: These give the maxima of the utility in A and O, and the minimum in l. That is, for a function of two variables A and O, the surface at a saddle-point curves up in two directions of A and O, and curves down in one other direction of l.
The meaning is simple. If you have a little more income left, you would be better to spend it on goods in order to make your utility as large as possible. If you maximize your utility, no income is left. That's the optimum!
Appendix:
If you have a maximization problem with nonnegative constraints, you'll have the following first-order conditions:
max U(A,O)
such that pA+qO≦M, A≧0 and O≧0
f.o.c.
dL/dA=dU/dA - lp≦0, A(dL/dA)=0 and A≧0....(1)'
dL/dO=dU/dO - lq≦0, O(dL/dO)=0 and O≧0....(2)'
dL/dl=- (pA+qO-M)≧0, l(pA+qO-M)=0 and l≧0....(3)'


