Professional Documents
Culture Documents
2y = 2λy .
x 2 + y 2 = 1
x
2 4 6 8
√ √
smallest y-intercepts of any line in the family that meets the curve x + y = 3.
These extreme values of k occur either when one of the lines is tangent to the
constraint curve or at an endpoint of the curve. (See Figure 4.27.)
This example illustrates the importance of locating all the points where ex-
trema may occur by considering places where ∇ f or ∇g is undefined (or where
∇g = 0) as well as the solutions to the system of equations determined using
Lagrange multipliers. ◆
Sketch of a proof of Theorem 3.1 We present the key ideas of the proof, which
are geometric in nature. Try to visualize the situation for the case n = 3, where
the constraint equation g(x, y, z) = c defines a surface S in R3 . (See Figure 4.28.)
In general, if S is defined as {x | g(x) = c} with ∇g(x0 ) #= 0, then (at least locally
near x0 ) S is a hypersurface in Rn . The proof that this is the case involves the
implicit function theorem (Theorem 6.5 in §2.6), and this is why our proof here
∇g (x0) is just a sketch.
Thus, suppose that x0 is an extremum of f restricted to S. We consider a
further restriction of f —to a curve lying in S and passing through x0 . This will
x0 S enable us to use results from one-variable calculus. The notation and analytic
x
particulars are as follows: Let x: I ⊆ R → S ⊂ R3 be a C 1 path lying in S with
x′(t0)
x(t0 ) = x0 for some t0 ∈ I . Then the restriction of f to x is given by the function
F, where
Figure 4.28 The gradient
∇g(x0 ) is perpendicular to F(t) = f (x(t)).
S = {x | g(x) = c}, hence, to the
tangent vector at x0 to any curve Because x0 is an extremum of f on S, it must also be an extremum on x. Conse-
x(t) lying in S and passing through quently, we must have F ′ (t0 ) = 0, and the chain rule implies that
x0 . If f has an extremum at x0 ,
then the restriction of f to the d
0 = F ′ (t0 ) = f (x(t))$t=t0 = ∇ f (x(t0 )) · x′ (t0 ) = ∇ f (x0 ) · x′ (t0 ).
$
curve also has an extremum at x0 . dt
Thus, ∇ f (x0 ) is perpendicular to any curve in S passing through x0 ; that is,
∇ f (x0 ) is normal to S at x0 . We’ve seen previously in §2.6 that the gradient
∇g(x0 ) is also normal to S at x0 . Since the normal direction to the level set S is