Problem Set 9 Solution Set

Anthony Varilly
Math 112, Spring 2002

1. End of Chapter 6, Exercise 5. Calculate the Jacobians of the following functions:

(a) f (x, y) = sin(x2 + y 3 ). We have
f
f
= cos(x2 + y 3 ) 2x and
= cos(x2 + y 3 ) 3y 2
x
y
Thus by Theorem 6.2.2, the Jacobian matrix is
2x cos(x2 + y 3 ) 3y 2 cos(x2 + y 3 )


(f) f (x, y) = xy+z . We have
f
f
f
= (y + z) xy+z1 and
= (log x) xy+z and
= (log x) xy+z
x
y
z
Thus by Theorem 6.2.2, the Jacobian matrix is
(y + z) xy+z1 (log x) xy+z (log x) xy+z


2. End of Chapter 6, Exercise 7. Find the critical points of the following functions and determine
whether they are local minima, maxima, or saddle points.
(a) f (x, y) = x3 + 6x2 + 3y 2 12xy + 9x.
Solution. The critical points are those points (x, y) for which
f
f
= 3x2 + 12x 12y + 9 = 0 and
= 6y 12x = 0.
x
y
Therefore y = 2x, which implies 3(x2 4x + 3) = 0, and so x = 1 or x = 3. Therefore
the critical points of f are (3, 6) and (1, 2). The matrix of the Hessian at (x, y) is


6x + 12 12
.
12
6
1

At (3, 6) the matrix of the Hessian evaluates to



30 12
12
6
and 1 = 30, 2 = 36. Hence, f has a local minimum at (3, 6).
At (1, 2) the matrix of the Hessian evaluates to


18 12
12
6
and 1 = 18, 2 = 36. Hence, f has a saddle point at (1, 2).
(c) f (x, y) = cos 2x sin y + z 2 .
Solution. The critical points are those points (x, y) for which
f
f
f
= 2 sin 2x sin y = 0,
= cos 2x cos y = 0, and
= 2z = 0
x
y
z
Clearly, it is always true that z = 0. From f /x = 0, either sin 2x = 0, which implies
x = n/2, or sin y = 0, which implies y = m. From f /y = 0, when x = n/2 we have
cos y = 0, which implies y = (2k + 1) /2, and when y = m we
which
 have cos 2x = 0, 
n (2k + 1)
,
, 0 and
implies x = (2j + 1) /4. Therefore the critical points of f are
2
2


(2j + 1)
, m, 0 , where k, j, m, n Z. The matrix of the Hessian at (x, y) is
4

4 cos 2x sin y 2 sin 2x cos y 0

2 sin 2x cos y cos 2x sin y 0 .
0
0
2


n (2k + 1)
At
,
, 0 , the matrix of the Hessian is
2
2

4 cos n sin ((2k + 1) /2)

0
0

0
cos n sin ((2k + 1) /2) 0 .
0
0
2

4 if n + k even
It follows that 1 =
, 2 = 4, and 3 = 8. Hence f has local
4 if n + k odd


n (2k + 1)
minima at
,
, 0 if n + k odd. If n + k even, then 3 > 0 implies f cannot
2
2


n (2k + 1)
have a minimum value at
,
, 0 , and 1 < 0 implies f cannot have
2
2 



n (2k + 1)
n (2k + 1)
a maximum value at
,
, 0 . But since
,
, 0 are critical
2
2
2
2
points, they are saddle points.
2

At


(2j + 1)
, m, 0 , the matrix of the Hessian is
4

0
2 sin ((2j + 1) /2) cos m 0
2 sin ((2j + 1) /2) cos m
0
0 .
0
0
2

It
 follows that 1 = 0, 2 = 4, and 3 = 8. Hence f has saddle points at
(2j + 1)
, m, 0 since 2 < 0.
4
3. End of Chapter 6, Exercise 8. Show that if f : A R2 R has a critical point x0 A and
we let
 2
2
2f
2f
f
=

x1 x1 x2 x2
x1 x2
be evaluated at x0 , then
(a) > 0 and 2 f /x1 x1 > 0 imply f has a local minimum at x0 .
Solution. From Theorem 6.9.4, f has a local minimum at x0 if x0 is a critical point of
f such that Hx0 (f ) is positive definite. Thus we need only show that Hx0 (f ) is positive
definite if > 0 and 2 f /x1 x1 > 0. To be sure, we prove the following lemma:


a b
Lemma: Let A =
be a symmetric 2 2 matrix, where a, b, c R. Then A is
b c
positive definite if a > 0 and ac b2 > 0. A is negative definite if a < 0 and ac b2 > 0.
Proof. Let v = (x, y)> be an arbitrary non-zero vector. Then

 

a b
x
>
x y
v Av =
= ax2 + 2bxy + cy 2
b c
y
b2 y 2 b 2 y 2
= ax2 + 2bxy +

+ cy 2
a
a




by 2
b2
= a x+
+ c
y2
a
a
If A is to be positive definite, then v > Av > 0 for all v. In particular, for y = 0, we must
have ax2 > 0, which implies a > 0. Also, when x = (b/a)y, (c b2 /a)y 2 > 0 implies
ac b2 > 0. If A is to be negative definite, then v > Av < 0 for all v. In particular,
for y = 0, we must have ax2 < 0, which implies a < 0. Also, when x = (b/a)y,
(c b2 /a)y 2 < 0 implies a(c b2 /a) > 0, which yields ac b2 > 0.
Since the matrix of the Hessian clearly satisfies the conditions of this lemma, it follows
that Hx0 (f ) is positive definite if = ac b2 > 0 and 2 f /x1 x1 = a > 0.
(b) > 0 and 2 f /x1 x1 < 0 imply f has a local maximum at x0 .
Solution. Since the matrix of the Hessian clearly satisfies the conditions of the lemma, it
follows that Hx0 (f ) is negative definite if = ac b2 > 0 and 2 f /x1 x1 = a < 0.
3

(c) < 0 implies x0 is a saddle point of f .

Solution. If x0 is not a local max or min, then it must be a saddle point. Similarly, if it
is not the case that > 0 and 2 f /x1 x1 > 0 or > 0 and 2 f /x1 x1 < 0, then it
must be true that < 0. Therefore, < 0 implies x0 is a saddle point of f .
4. End of Chapter 6, Exercise 12. A function f : Rn R is called homogenous of degree m if
f (tx) = tm f (x) for all x Rn and t R. If f is differentiable, show that for x Rn ,
Df (x) x = mf (x) , that is,

n
X

xi

i=1

f
= mf (x) .
xi

Show that maps multilinear in k variables give rise to homogenous functions of degree k. Give
other examples.
Solution. By definition of the directional derivative,
Df (x) x = lim

h0

f (x + hx) f (x)
f ((1 + h) x) f (x)
= lim
h0
h
h

Using the fact that f is homogeneous of degree m, we get



(1 + h)m f (x) f (x)
(1 + h)m 1
Df (x) x = lim
= lim
f (x)
h0
h0
h
h
!






m 2
m m
1m + m
1 h + 2 h + + m h 1
= lim
f (x)
h0
h

 
 

m
m m1
= lim m +
h + +
h
f (x) = mf (x).
h0
2
m
as desired.
k-linear maps are characterized by the property
L(x1 , . . . , xi1 , u + w, xi+1 , . . . , xn )
= L(x1 , . . . , xi1 , u, xi+1 , . . . , xn ) + (x1 , . . . , xi1 , w, xi+1 , . . . , xn )
If we define g(x) = L(x, . . . , x), then it follows that
| {z }
k times

g(tx) = L(tx, . . . , tx) = tk L(x, . . . , x) = tk g(x).

Therefore, maps multilinear in k variables give rise to homogenous functions of degree k.
An example of a non-linear homogenous
function is f (x, y) = x2 + y 2 . This is homogeneous
of degree 2 since f (kx, ky) = k 2 x2 + y 2 = k 2 f (x, y).
5. End of Chapter 6, Exercise 13. Use the chain rule to find derivatives of the following, where
f (x, y, z) = x2 + yz, g(x, y) = y 3 + xy, and h(x) = sin x:
4

(a) F (x, y, z) = f (h(x), g(x, y), z).

Solution. Here f (h, g, z) = h2 + gz. The chain rule gives
F
x
F
y
F
z

f h f g
f z
+
+
= 2 sin x cos x + zy + 0
h x g x z x
f h f g f z
=
+
+
= 0 + z(3y 2 + x) + 0
h y
g y
z y
f h f g f z
=
+
+
= 0 + 0 + (y 3 + xy)
h z
g z
z z


Therefore DF (x, y, z) = sin 2x + yz xz + 3y 2 z y 3 + xy .
Alternatively, we can use Jacobean matrices: DF (x, y, z) = DF (f (x)) Df (x). In this
case

cos x
0
0


3y 2 + x 0
DF (x, y, z) = 2h z g y
0
0
1
=

and we get the same answer as before.

6. End of Chapter 6, Exercise 15. Let f : R R be differentiable. Assume there is no x R
such that f and f 0 both vanish at x. Show that S = {x | 0 x 1, f (x) = 0} is finite.
Solution. Assume that S is infinite. Then by the Bolzano-Weierstrass theorem, S has an
accumulation point c [0, 1]. So there are xn in S with xn c. Since f is differentiable, it
is continuous, which implies that limn f (xn ) = f (c) = 0. Then using the definition of the
derivative,
f (xn ) f (c)
00
f 0 (c) = lim
= lim
=0
n
n
xn c
xn c
This last equality follows because xn S. But this contradicts the assumption that there is
no x R such that f and f 0 both vanish at x. Thus S must be finite.
7. End of Chapter 6, Exercise 18. Prove that the equation x3 + bx + c = 0 where b > 0 has
exactly one solution x R.
Solution. Let f (x) = x3 +bx+c, where b > 0. From the Intermediate Value Theorem, we know
that f must have at least one root since it is an odd polynomial. Since f 0 (x) = 3x2 + b > 0,
f is an increasing function for all x R. We now claim that f (x) = 0 only once. For if there
are x1 , x2 such that f (x1 ) = f (x2 ) = 0, then by Rolles Theorem, there exists (x1 , x2 )
such that f 0 () = 0, which contradicts the fact that f 0 (x) > 0 for all x R.

8. End of Chapter 6, Exercise 20. Let L : Rn Rn be a linear map. Define kLk = inf {M | kLxk M kxk for
Show that k k is a norm on the space of linear maps of Rn to Rm .

Solution. Let L(Rn , Rm ) denote the space of linear maps from Rn to Rm . We first show that
kL1 k 0 for L1 L(Rn , Rm ). We have
kL1 k = inf {M | kL1 xk M kxk} = inf {M | kyk M kxk}
if we let L1 x = y Rm . From properties of the vector norm, kyk 0 and kxk 0, so it
follows that M 0 and thus inf M 0.
Secondly, we show that kL1 k = 0 if and only if L1 = 0. We have
kL1 k

0 inf {M | kL1 xk M kxk} = 0

kL1 xk 0 for all x Rn

kL1 xk = 0 for all x Rn L1 = 0.

Thirdly, we show that for R, we have kL1 k = ||kL1 k. We have

kL1 k = inf {M | kL1 xk M kxk} = inf {M | kyk M kxk}


M
= inf M | kyk
kxk = inf {||N | kyk N kxk}
||
= || inf {N | kyk N kxk} = ||kL1 k.

Finally for L1 , L2 L(Rn , Rm ), we show that kL1 + L2 k kL1 k + kL2 k. Let kL1 k =
inf {M1 | kL1 xk M1 kxk}, kL2 k = inf {M2 | kL2 xk M2 kxk}, and kL1 +L2 k = inf {M | k(L1 + L2 )xk M
Note that
k(L1 + L2 )xk = kL1 x + L2 xk
kL1 xk + kL2 xk M1 kxk + M2 kxk
For all M1 and M2 for which the above inequalities kL1 xk M1 kxk and kL2 xk M2 kxk
hold, they must be valid for inf M1 and inf M2 . So we get
k(L1 + L2 )xk inf {M1 | kL1 xk M1 kxk} + inf {M2 | kL2 xk M2 kxk}
= kL1 k + kL2 k
It is now convenient to redefine kLk = sup {M | kLxk M kxk for all x Rn }. Then kL1 +
L2 k = sup {M | k(L1 + L2 )xk M kxk}. So we now have
M kxk k(L1 + L2 )xk kL1 k + kL2 k
For all M for which the above inequality holds, it must hold for sup M . Thus we have
sup {M | k(L1 + L2 )xk M kxk} kL1 k + kL2 k
which implies
kL1 + L2 k kL1 k + kL2 k
as desired.
9. End of Chapter 6, Exercise 29. Let fn (x) = xenx , x [0, ), n = 0, 1, 2, . . . .
6

(a) Show that f (x) =

n=0 fn (x)

exists. Compute f explicitly.

P
P
nx = x
x n
x
Solution. f (x) =
n=0 xe
n=0 (e ) . For x > 0, |e | < 1, so we have
x
x
a convergent geometric series which sums to 1/P
(1 e ) = e / (ex 1). Therefore
x
x
f (x) = xe / (e 1) for x > 0. At x = 0, f (0) =
n=0 fn (0) = 0.

(b) Is f continuous?
Solution. No, f is not continuous because
ex
f (x) f (0)
= lim x
=1
x0
x0 e 1
x0
lim

using lH
opitals Rule. Therefore f (x) 9 f (0) as x 0 and so f is not continuous.
(c) Find a suitable set on which the convergence is uniform.
Solution. Consider the interval [a, b], where 0 < a < b. Clearly x bPand ex

ea
fn (x) = xenx bean . If we let Mn = bean , then
n=0 Mn =
P. Therefore
n
a
a
b n=0 (e ) is a convergent geometric series since a > 0 implies e
< 1. By the
Weierstrass M test, we have uniform convergence on the interval [a, b]. But since a
and b are arbitrary positive numbers, we have shown uniform convergence on [a, ),
a > 0.
(d) May we differentiate term by term?
Solution. The fn are clearly differentiable for x (0, ), with fn0 (x) = (1 nx) enx .
FromP
this, we see that the fn0 are continuous for x (0, ). We have already showed
that
fn converges uniformly (hence pointwise) on (0, ).
P 0
We will now show that
fn converges
on
Consider
the interval
P uniformly
P(0, ).
P
0 (x) =
nx
nx . For
[a, b], where 0 < a < b. We
have
f
e
nxe
n=0 n
n=0
n=0
P

x [a, b], enx < 1, and so n=0 enx is a convergent geometric series.PIn Example
nx ,
5.1.9, this was shown to imply uniform convergence
series
n=0 nxe
Pon [a, b]. For
Pthe

an
an
take Mn = nbe
, similar to part (c). Then n=0 Mn = b n=0 ne
is seen to be
convergent
in [a, b] by an easy application of the ratio test.PSo by the Weierstrass M
P

nx converges uniformly on [a, b]. Thus, since

0
test,
nxe
n=0
n=0 fn (x) is the difference
between two uniformly convergent series, it is also uniformly convergent on [a, b]. But
since a and b are arbitrary positive numbers, we have shown uniform convergence on
(0, ).
Therefore, we conclude that term-by-term differentation is valid.
10. End of Chapter 6, Exercise 35. Let f : (a, b) R be twice differentiable. Suppose f vanishes
at three distinct points. Prove that there is a c (a, b) such that f 00 (c) = 0.
Solution. Suppose f (x1 ) = f (x2 ) = f (x3 ) = 0. WLOG, let x1 < x2 < x3 . Then by Rolles
Theorem, there exists c1 (x1 , x2 ) and c2 (x2 , x3 ) such that f 0 (c1 ) = f 0 (c2 ) = 0. Using
Rolles Theorem again, there exists c (c1 , c2 ) such that f 00 (c) = 0.