Solutions to Exercises

"For emergency use only."

That is how Buck begins his list of solutions. Ross comments, in the same
spirit, that students will simply "cheat themselves" if they turn to this section
before working hard to develop their own solutions. We trust that, at this level,
our students are mature enough to agree.
The answers here are intended, like the text, to teach. They are therefore sprin-
kled with comments, alternative approaches, and questions. Except for leaving
such questions to be pondered by the reader, these solutions are complete.

Chapter 1
Section 1.1
1. (a) Gl(X,y)-Gl(a,b)-[3a 2 _3b21[x-a]=x3_y3_a3+b3_
y-b
3a 2(x-a)+3b 2(y-b) = (x 2+ax+a 2 -3a 2)(x-a)+(v2+by+
b2 - 3b2)(y - b). Recalling that Ix-al and .J!-bl
,.j(x-a)2+(y-b)2 ,.j(x-a)2+(y-b)2
are bounded by 1, we have

as (x, y) ~ (a, b).

248 Solutions to Exercises: Chapter 1

(b) G2(X, y) - G2(a, b) - [b a] [~ =:] = xy - ab - b(x - a) -

a(y - b) = (x - a)(y - b). On division, this becomes (x - a)[(y -
b)IJ(x - a)2 + (y - b)2]. The bracketed factor has absolute value
::s: 1, and (x - a) -+- 0 as (x, y) -+- (a, b).

, [2(a+b)2(a+b)].
2. F (a, b) = 2(a _ b) 2(b _ a) . Frrst, by the mean value theorem, (x +
y)2_(a+b)2 = 2t(x+y-a-b) for some t betweenx+y anda+b. From
(x+y)2-(a+b)2 = 2t(x-a)+2t(y-b) ~ [2(a+b) 2(a+b)] [~=: J.
we guess Fl' = [2(a + b) 2(a + b)]. Confirmation comes from

Fl(X, y) - Fl(a, b) - [2(a +b) 2(a +b)] [ Xy _ b -a]

= (2t - 2a - 2b)(x - a + y - b),

because

l(2t - 2a - 2b)(x - a + y - b)1 41 bl 0

< t - a--+-
J
~-r==::::;<:=====:;;:---'-'-
(x - a)2 + (y - b)2 -

as (x, y) -+- (a, b). By similar algebra, F2' = [2(a - b) - 2(a - b)]. Then
apply Theorem 1.

a sin b] .
3. H'(a, b) = [ COS
2e2a 3e3b • Use the techmques of Answer 2:

sinx - cos y - (sina - cos b) = cos t(x - a) + sins(y - b)

-a]
~ [cosa sinb] [ Xy _ b '

because I cos t - cos a I + I sin s - sin bI -+- 0; and analogous constructions

for the exponentials.

4. Assume b =F O. Apply the mean value theorem to

../i: Ilxll-llbll = J(x? + ... +xn2) - J(b? + ... + bn2)

= 0.5t- 1/ 2[(Xl - bl)(Xl + bI) + ... + (x n - bn)(xn + bn)]

for a value t between x? + ... + xn2 and b? + ... + bn2 The value is no
problem if we confine x to N (b, ~). Then

suggests that I' (x) = Jb[bl+···bal

.. +b
. We verify as in Answers 2 and 3.
J2 a2
Solutions to Exercises: Section 1.2 249

5. (aF+~G)(x)- (aF+~G)(b)-(aF' +~G')(b)(x-b) = a(F(x)-F(b)-

F'(b) (x - b}) + ~(G(x) - G(b) - G'(b) (x - b}) = o(x - b) + o(x - b).

6. H' = A, because H(x) - H(c) = A(x} + b - (A(c) + b) = A(x - c}, which

certainly says that A(x - c} approximates H(x) - H(c) to within o(x - c).

Section 1.2
1. (a) (x, y) ~ (0,0) ::::} x ~ 0, y ~ O::::} (xy)1/3 ~ O.
(b) 1 == 0 along the x-axis. Hence *(0,0) == limHo [(t,O)~ [(0,0) = 0;
similarly with y.
(c) I(t, t) - 1(0, 0) - [0 O][t t]t = (tt)1/3 = t 2/ 3, which is much larger
than lI(t, t)1I = J2ltl; t 2: 3 ~ 00 as t ~ 0+.

2. (a) Away from the origin, the partials are given by the quotient rule. At
(0, 0), both partials are 0, because h == 0 along the axes.
(b) h(x, y) - h(O, 0) - [0 O][x y]f = -::th+
x y
. Along the line y = x, this
quantity is not small. Thus, h is not even continuous at (0, 0).

3. Away from the y-axis, ~~ = 2x sin (~) - cos (~) and ~~ = 0 are
continuous. By Theorem 2, G is differentiable there. For a point (0, b) on
the y-axis,

aG = l'
-
ax
1m
t_O
G(t,b)-G(O,b)
t
= l' t . (1)t = 0,
1m
t_O
sm -

answering (b). But limx_o ~~ (x, b) = limx_o - cos U) does not exist;
that answers (c). On the other hand,

IG(x,y)-G(O,b)-[O O][x y-WI = Ix2sin(~)I::s Ix1 2,

which is small compared to lI(x, y - b) II as (x, y) ~ (0, b). Therefore, G
is also differentiable on the y-axis.

4. (a) Differentiable everywhere; its partials are continuous.

f'(x, y) = [exp (: + y)
cosxcosy -smxsmy
ex~(: +. y)] .

(b) Differentiable everywhere. The formula

ag = 3x2y3 sin (~) - x i cos (~)

ax xy xy
250 Solutions to Exercises: Chapter 1

shows that if
is continuous away from the axes and --+ 0 as you get
close to any point on the axes. At (0, b),

ag = lim t 3 b3 sin{1/tb) = O.
ax t-O t

At (a, 0), if
= 0 because g == 0 along the x-axis. Hence if is always
continuous; analogously for y.

g' = [3x2 y 3 sin (:y) - xi cos (x~)

3x 3y2 sin (x~ ) - x 2y cos (x~) ]
or [0 0].
(c) Differentiable everywhere, even though the partials are discontinuous
along the axes; justifications and derivative matrix mimic Exercise 3.
(d) Differentiable off the axes, plus at the origin. Off the axes, ~~ =
~x-I/3y2/3 is continuous' same with iJG At (0 0) iJG = iJG = 0
3 'iJy' "iJx iJy ,
and

[G(X, y) - G(O, 0) - [0 0] [~]]

Jx2 + y2
=( x )2/3 ( Y )2/3 (x 2 + i)I/6 --+ 0
J x 2 + y2 J x 2 + y2
as (x, y) --+ (0,0). Anywhere else, G has an undefined partial: At
(a, 0),

aG I' G(a, t) - G(a, 0) I' 2/3 -1/3

- = 1m = 1m a t = 00;
ay t_O t t_O

by Theorem 1, G is undifferentiable. G' = [~x-I/3y2/3 ~x2/3y-I/3]

off the axes, [0 0] at the origin.
(e) Differentiable off the axes only, with H' = [tx-2/3 y l/3 txl/3y-2/3].
Away from the axes, both partials are continuous. At (a, 0),

aH I' H(a, t) - H(a, 0) I' 1/3 -2/3

- = 1m = 1m a t = 00;
ay t_O t t_O

by Theorem 1, H is undifferentiable. At (0, 0), ~~ = ~~ = 0, but

/ H(x, x) - H~: - [OO][x x F / = x 2/ 3/.J2lxl--+ 00,

making H undifferentiable.
Solutions to Exercises: Section 1.3 251

5. Since f: R -+ Rn, f' has to be n x 1. Because

f(t) - f(b) = (t - b)v = [VI ... vn]t (t - b),

we have f' = (v).

Section 1.3
1. (a)

~ = ~ a(u, v) = [2u 2v] [YCO~XY XCO~Xy]

a(x, y) a(u, v) a(x, y) -y SlDXY -x SlDXY
= [2uycosxy -2vysinxy 2uxcosxy -2vxsinxy] = [0 0].
(b) g = (sinxy)2 + (cosxy)2 = 1. Hence a(~~y) = o.
2. Decreasing. Distance r = (x 2 + y2)1/2 has
dr ar dx ar dy x y
- = -- + -- = -(-10sint) + -(Scost).
dt ax dt ay dt r r

When x = 5, cost = 0.5, so sint = ±J{. Quadrant I forces y = 4J3,

and r ~~ = 5( -5J3) + 4J3(4) < o.
3.
af af a(x, y, z) [a f af af ]
a(p, e, <p) = a(x, y, z) a(p, e, <p) = ax ay az
COS esin <p p sin e sin <p p cos ecos <P]
x [ sin e sin <p p cos e sin <p p sin e cos <p
cos <p 0 -p sin <p

= [cos e sin <p (:~) + sin esin <p ( :;) + cos <p (~~) etc.] .

4. ML. By Exercise 1.1:6, the derivative of a first-degree function is the asso-

ciated matrix.
5. Rule: If f and g are differentiable at band g(t) =fi 0, then h(x) == "ffit is
differentiable at b, and g

h'(b) = g(b)f'(b) - f(b)g'(b)

g(b)2

Proof: H(u, v) == ~ has ~~ = ~ and ~~ = -~ wherever v =fi O. Set u ==

f(x), v == g(x). Then h(x) = H(f(x), g(x)) is a differentiable composite
at b, with
ah
ax =
aH au
a; ax + a; ax =
aH av (1),
g(b) f (b)
(f(b)),
+ - g(b)2
g (b).
252 Solutions to Exercises: Chapter 1

6. u e V = L U j vj. By the product rule,

a(uev) ~(aVj aUj) ~ aVj ~ aUj
ax = ~ Uj a; + vj ax = ~ Uj a; + ~ vj ax

= [UI ... urn] [~] + [VI'" Vrn] [~]

~ fu!m.
ax ax
av au
= row(u)- + row(v)-;- .
ax ux

7. (a) Assume that r = x 2 + y2 > O. .J

=> If H is differentiable at r, then G(x, y) = H(r) is a differentiable
composite.
<= Assume that G is differentiable at (x, y). We have
H(r + s) - H(r)
s
G(x + 7-, y +~) - G(x, y) - G'(x, y)[7- ~r
=
s
+ G'(x, y)[7- ~y
s
As s ~ 0, the first fraction on the right vanishes, and the second is
fixed at G' (x, y) [~ f y.
Hence H is differentiable at r.
(b) => If G is differentiable at (0, 0), then
H(s) - H(O) = G(s, 0) - G(O, 0) ~ aG (0,0)
s s ax
as s ~ O. Hence H is differentiable at O. Also
H(s) - H(O) = G(-s, O) - G(O, 0) ~ _ aG (0, 0).
s s ax
Of necessity, H' (0) = ~~ (0, 0) = 0; same with ~~.
<= If H is differentiable at 0, then
G(x, y) - G(O, 0) - [OOUx y]f H(r) - H(O) ,
II(x, y)1I = r ~ H (0)
as (x, y) ~ O. Hence if H'(O) = 0, then G is differentiable at 0
(and G' (0, 0) = 0).
(c) The equality is trivial at the origin. Away, wherever G is differen-
tiable, G (x, y) = H (r) gives
aG dH ar H'(r)x
-=--=--
ax dr ax r

and Ty'
aG = !D!:1l.
r . CIearIy YTx'
aG = aG
xTy"
Solutions to Exercises: Section 1.4 253

8. L(x) = [Lk=l alkXk Lk=l amkXkr, so

IIL(x)112 = (ta lk Xk)2 + ... + (tamkXk)2
k=l k=l
~ (ta lk2 ) (tXk2) + ... + (ta mk2 ) (tXk2)
k=l k=l k=l k=l
(Cauchy) ~ (t~am,2) (~X,2);
take square roots.

Section 1.4
1. (a)
(Pf a
- - = -(2xyexp(x2y)) = 2xexp(x2y) +2xyx 2 exp(x 2y),
axay ay
af
2 a
- - = -(x 2 exp(x2y)) = 2xexp(x2y) + x 22xyexp(x2y).
ayax ax

(b) f" = exp(x2y) [[2 Y + 4x2y2 2x + 2x 3y ] [2x + 2x 3y x4]]

2. g' = [2; -;yJ. g" = [f~ ~~ [[1 ~f]l

3. fill is straightforward and long; g" is clearly 0, but requires sixteen O's.
4. By the formula for f" found in Example 1,
f"(x, y) - f"(a, b)
= [[2y - 2b 2x - 2a - 2y + 2b] [2x - 2a - 2y + 2b - 2x + 2a]]

= [[[02][~=~][2 -2][~=~]] [[2 -2][~=~J [-20{~=~JJJ

= [[[02] [2 -21] [[2 -2] [-201]] (x-a,y-b))

5. No. You would have iJ~~ = exp (x 2), iJ~iJ~ = sin2 y, continuous but not
symmetric.

6. There would exist Xl and Y1 E N(b, 1) with g(X1) = f(Y1), X2 and Y2 E

N (b, !)
with g(X2) = f(Y2), .... Then Xi -+ band Yi -+ b, so by continu-
ity
254 Solutions to Exercises: Chapter 2

Chapter 2
Section 2.1
1. (a) Directional = V'feu = (2x, -2y) e (cosO, sinO) = lO cos 0 -6 sinO.
(b) The directional is 0 in the directions with 10 cos 0 - 6 sin 0 = 0, or
tan 0 = l~. The derivative is found implicitly: 2x - 2yy' = 0 leads to
i;
y' = ~ = the tangent is in the zero-growth direction.

2. (a) V'g(x, y) = (2x, 8y), = (2a, 8b) at the point given.

(b) Implicitly,2x + 8yy' = 0, y' = - Iy • Hence the tangent at (a, b) lies
in the direction of vector (4b, -a). In that direction,
(4b, -a)
8ug=V'geu=(2a,8b)ell(4b,_a)11 =0.

3. (a) Excluding the origin, ~~= 0.5 (x 2 + ir l / 2 2x = ~. Similarly,

~~ = ~,and lIV'h11 2 = (~}2 + (n 2 = 1.
(b) Since h is distance from the origin, it increases fastest if you walk out-
ward along the line from you to the origin, in which case it increases
1 m for every meter you walk.

4. The unit vector in the direction of negative x j is

-ej == (0, ... ,0, -1, 0, ... ,0).

By Theorem 2(a), the derivative in that direction is

5. First, V'G(O) = (0, 1), because G == 0 along the x-axis, == y along the
y-axis. Therefore,

G(x, y) - G(O) - V'G(O)[x YY = G(x, y) - y.

Along the line y = x in quadrant I,

G(x,y) - y = y/2- y = _T 3/ 2
J(x 2 + y2) -/2Y
does not approach O.

6. Where the partials exist, the one-variable product rule gives

Solutions to Exercises: Section 2.2 255

Hence

(a~;:), ... ,a~;:») = f (::1"'" ::n) +g (:~, ... , :~).

7. (a) By Exercise 6, V(/f) = fV f + fV f = 2fV f·
(b) Write u = f2. Then ::. = ~fu if = 2f if for each j. Consequently,
J J J

au, ... ,au-) (a f

(-
aXI aXn
= aXI- , ... , 2afafXn- ) =2fVf·
2f

(c) The fastest increase for f2 is in the direction of V f if f is positive,

opposite direction if f is negative. Thus,

V([xy]2)(2xi, 2x2y) = 2xy(y, z) = 2fV f;

at (5,3), V(/2) = 30V f; at (-2,2), V(/2) = -SV f. It makes

sense: When f < 0, increasing f makes If I decrease. Note also that
if f = 0, then f2 is stationary; its rate of change is zero in every
direction.

S. The directional of f along v is ad = v f • v = VI it + ... + if. The

Vn
derivative of this quantity is

Then by Theorem 1, the directional of ad along u is [!,,(b)(v)](u).

Section 2.2
1. (a) f(b)- f(a) = 5-5 = 0, V f = (~, f), b-a = (0, S). Then 0 = ~
at y = 0, point (3, 0).
(b) f(b) - /(a) = e3 - e l , V f = (~+Y, ~+Y), b - a = (1,1). The
equation e 3 - e l = 2~+Y is not sufficient. But the segment is given
by y = x-I. The simultaneous solution of x + y = In e3 "2e 1 and
x - y = 1 is x = 0.5 + In[ (e 3 ;e 1) J. y = -0.5 + In ¥.
256 Solutions to Exercises: Chapter 2

2. F(x, y) == I~I has zero derivative for x > 0 and for x < 0, but has two
different values.

3. (a) /(x, y) == ../(x - 1)2 + y2 + J(x + 1)2 + y2 is undifferentiable at

just (±1, 0).
(b) /(1,0) - /(-1,0) = 0,

x-I x+l
V /= ( + ,
J(x - 1)2 + y2 J(x + 1)2 + y2
Y + Y )
J(x - 1)2 + y2 J(x + 1)2 + y2 .

The latter is 0 at the origin, so c == (0, 0) is one such point.

4. By Theorem 1.1:1, F' is a matrix with rows F{, ... ,F~. If F' = 0, then
each Fj = O. By Theorem 3, each Fj is constant.

5. Assume hI! = O. By Exercise 4, h' = constant = (matrix) M. Write H(x) ==

h(x) - M(x). Then H'(x) = h'(x) - (M(x))' (linearity, Exercise 1.1:5) =
h'(x) - M (Exercise 1.1:6) = O. Therefore H(x) = constant = (vector) b,
and h(x) = M(x) + b.

6. (a) From IIg(x)-g(y) II ::: Mllx-yll', we conclude that IIg(x)-g(y) II < e

as long as IIx - yll < 8 == (~ )1/,.
(b) If IIg(x) - g(b) II ::: Mllx - bll1+e, then

IIg(x) - g(b) - O(x - b) II ::: Mllx - blnx - bll = o(x - b).

Hence g is differentiable, with g' = 0; necessarily g is constant.

(c) Assume IIg'lI ::: M. We apply the mean value theorem separately to
the components of g. We get

IIg(x) - g(y) 112 = [gl (x) - gl (y)]2 + ... + [gn(x) - gn(y)f

= [gi (Cl)(X - y)]2 + ... + [g~(Cn)(x _ y)]2
::: IIgl (cI)1I 2 I1x - yll2 + ... + IIgn(cn)1I 2 I1x _ yll2

(operator inequality) ::: nM211x _ Y112.

(d) By properties of continuous functions, there is a neighborhood ofb in
which g'(x) is bounded. Apply part (c) there.

7. (a) any constant.

(b) any discontinuous function.
Solutions to Exercises: Section 2.2 257

(c) I(t) == Jjtf. If sand t are on the same side of zero, say 0::: t < s,
then I(s) - I(t) = .;s - ,Jt ::: ~, because

(.;s - ,Jt) 2 = s - 2,Jii + t ::: s - 2$t + t = s - t,

and similarly if s < t ::: O. If they are on opposite sides, t < 0 < s,
then

I/(s) - l(t)1 = IJS - HI ::: Jis - -tl ::: ~,

because Is + t I ::: Is I + It I = s - t. This proves that I is Lipschitz of
degree !. It is not Lipschitz of degree! + 8, because

I(s) - 1(0) -E
sl/2+E = s -+ 00 as s -+ O.

(d) g(x) == Ixl is Lipschitz, because the slope of every secant is between
-1 and 1: !g(I;=~?)! :::
1. But g' (0) is undefined.

(e) h (x) == l':'x' A Lipschitz function on a bounded domain has to be

bounded (Justify!), and h is not. But for any b E (-1, 1), h' is (con-
tinuous and therefore) bounded on [=.l,f!, 1 !b ],
so h is Lipschitz on
this latter interval.
(f) Part (c) suggests that t E is Lipschitz of degree 8 and no higher; you
should verify that principle. What we need is an arbitrarily small
power of t. This points to t 1/ t, but t 1/ t does not work. At infinity,
Int is smaller than any power of t, so we try I(t) == In,!/2!'
Since
I -+ 0 as t -+ 0, we set 1(0) = O. By L'Hospital's rule,
I(t) - 1(0) t- E _8t- E - 1
= ~ ~-oo
tE Int -ln2 t- 1
as t -+ 0+, so I is not Lipschitz of any degree.

8. Set I(x, y) == 0 if y < 0, == y2 if 0 ::: y ::: 2 and x > 0, == _y2 if

o ::: y ::: 2 and x < O. The domain has the points off the nonnegative y-
if
axis with y ::: 2. is always 0; if
is 0 on or below the x-axis, ±2y above;
both partials are continuous, and II I' II ::: 4. But 1(8, 1) - 1(-8, 1) =
1 - (-1) = 2, even though the points are close together.

(b) Set g(x) == x 3/ 2 sin 0) on (-1,1), g(O) == O. Then g' is defined on

the interval and continuous except at 0, but unbounded. Specifically,
g'(x) has arbitrarily large values near x = O. Thus, given M, there
will exist intervals [s, t] throughout which g'(x) > M, forcing

Ig(t) - g(s)1 = g'(t*)It - sl > M[t - s].

Hence g is not Lipschitz in any neighborhood of x = O.
258 Solutions to Exercises: Chapter 2

9. Assume that N(b, r) and N(e, s) have d is common. Then lib - ell ~
lib - dll + lid - ell < r +s, as the hint says. Ifx = (1- t)b + te is on the
segment, then

IIx - bll + IIx - ell = tile - bll + (1 - t)IIb - ell = lib - ell < r + s.
Of necessity, either IIx - bll < r, making x E N(b, r), or IIx - ell < s,
making x E N(e, s); either way, x is in their union.

10. (a) Trivial; they are continuous functions of z.

(b) Write F(x, y) == f~ f(x, y, z) dz. By the mean value theorem,

F(x+t,y)-F(x,y)
t
-
11 -a
0
af (
x
x,y,z
)d
z

10r ax(x+t *,y,z)- af

l
= [af ax(x,y,z) ] dz.

Since if is uniformly continuous on the cube,

t ~ 0 => t* ~ 0 => af (x + t* , y, z) ~ aaf (x, y, z)
ax x
uniformly. Hence F(x+t.Y~-F(x.y) ~ f~ if(x, y, z) dz. This proves
that the integral is ~~.
(c) ~~ is continuous, because

rl~ rl~
10 ax(x+s,y+t,z)dz- 10 ax(x,y,z)dz

~ max ( aaxf (x + s, y + t, z) - aaxf (x, y, z) ) ~ 0

as (s, t) ~ (0,0) by the uniform continuity of if, and analogously

for ~; ; therefore, F is differentiable.

Section 2.3
1. You should be able to figure out the critical points and their nature by in-
spection. Then match your conclusions against the prescribed method.

(a) f'(x, y) = [2x 2y], always defined, = 0 at the origin. Outward di-
rectional at (x, y) is

[2x 2y][x yf = 2( 2 2)1/2> o.

II(x, y)11 x +y

Origin is a strict minimum.

Solutions to Exercises: Section 2.3 259

(b) Vg = (2xy2,2x2y) = 0 everywhere on the axes. Near (a, 0), out-

ward directional is given by
(X a y) (4x 2y2 - 2axy2) 2a 2y2
(2xi,2x 2y). - '= ~ -- > 0
([x - a]2 + y2)1/2 ([x _ a]2 + y2)1/2 Iyl-
when x ~ a. Hence the x-axis is all loose minima; symmetrically, so
is the y-axis.
(c) Vh = (~, f), never zero, undefined at origin. Outwards are
( :', ~) • (x, y) = 1 > 0;
r r r
strict minimum.
(d) V F = (4x 3y 3, 3x 4 y2) =0 on the axes. Near (a, 0), outwards have
numerator
4x 3y\x - a) + 3x 4 y 3 ~ 3a 4 y 3.
They are positive or zero above the x-axis, ~ 0 below. Therefore,
along any line through (a, 0), there are higher or equal values above
the x-axis, lower or equal below. This means that the point is never a
strict maximum or strict minimum along a line; it is not a maximum,
minimum, or saddle.
Near (0, b), b #- 0, outwards have numerator
4x 3l x + 3x 4 i(y - b) ~ 4x 4 b 3.

This makes them positive to the right and left of the positive y-axis,
but zero on the axis; hence the positive y-axis is all loose minima. By
similar reasoning, the negative y-axis is all loose maxima.
(e) VG = (2x, -2y) = 0 at the origin. Outwards are equal to

(2x, -2y). (x, y) = 2 (x 2 - y2) .

r r
They are positive in the cone Iyl < lxi, negative in the conjugate cone
Iyl > Ixl. Hence the origin is a minimum along the line y = 0, a
maximum along x = 0; origin is a saddle.
(f) VB = (3x 2y3,3x 3y2) =0 on the axes. Near the origin, outwards
are equal to

(3x 2 y3 , 3x 3 y 2) • -
(x,y)
r
6x 3y 3
- = --.
r
They are positive in quadrants I and III, negative in II and IV; origin
is a saddle. Near (a, 0), a #- 0, outwards have numerator 3x 2y 3(x -
a) + 3x 3y2y ~ 3a 3y 3. They are zero along the x-axis, have one sign
above the axis, opposite sign below. Hence (a, 0), and likewise (0, a),
is not a strict extreme along any line; no minimum, no maximum, no
saddle.
260 Solutions to Exercises: Chapter 2

(g) By Exercise 1.2:4d, K'(x) is undefined on the axes, except that

K'(O,O) = O. Near (a, 0), outwards are equal to

point is loose minimum, and likewise (0, a).

2. Start with one variable. What is needed is a function that oscillates be-
tween, say, j = 2x2 and g = 2x4 near x = O. That suggests h(x) ==
(x 2 + x 4) - (x 2 - x 4) (Sin ~) for 0 < Ixl ::: 1, with h(O) == o. Clearly,
h(x) ::: (x 2 + x 4) - (x 2 - x 4) = 2x4 > 0, except when x = 0, so 0 is a
strict minimum. The outward directional toward the right is

h'(x) = 2x + 4x 3 - (2x - 4x 3) sin ~ + (1- x 2) cos~,

which has alternating signs at x = ~, ~, .... (Verify!) We need only set
G(x, y) == h(Jx 2 + y2).

Section 2.4

1. (a) For j, Hessian = [~ ~l for g, [~ ~2l

(b) For j, ~1 = 2, ~2 = 4. For g, ~1 = 2, ~2 = -4.
2. Here h'(x) = [y x], so (0,0) is the only critical point.
h" = [[0 1] [1 0]],

so
h"(0,0)(V)2=[[0 1][~~] [1 0][~~]][~~]=2V2Vl
is positive for v = (1, 1) and negative for v = (-1, 1). Theorem 2 says that
(0, 0) is a saddle.

3. (a)

l
F' = [cos x sin y sin x cos y] ,

F" = [[-sin x sin y cos x cos y ][cos x cos y - sin x sin y]

Two kinds of critical points: where x and y are both odd multiples of
~; and where x and y are both even multiples of ~ = multiples of 1f •
Solutions to Exercises: Section 2.3 261

(b) At the first kind, sin x and sin y are ± 1, so their product is as big or
small as possible; we expect extremes. At the others, sin x = sin y =
0; the changing signs should produce saddles.
For the first kind, let k be odd. At the places

( krr krr) (krr [k ± 4]rr)

2 ' 2 ' 2' 2 , ... ,

F" = [[-1 0] [0 - 1]], so F" (0, 0)(v}2 = -v? - vi

< O. There-
fore, these places are maxima. Between those are the places

[k ± 2]rr ) (krr [k ± 6]rr )

( krr
2' 2 '2' 2 , ....

At those,

F"(O, 0)(v}2 = [[10] [01]] (VI, V2)}2 = V? + vi> 0;

they are minima.
At the second kind of critical point, F" = [[0 ± 1] [±1 0]] (indepen-
dent signs), so F" (0, 0)(v}2 = ±2VI V2. For either sign, the product is
positive for some v, negative for some; all saddles.

4. (a) f(x, y) == x4 + y4; clearly strict minimum.

(b) F(x, y) == x 2y2; has minimal value everywhere on the two axes.
(c) g(x, y) == x 3y 3 has minimum along the line y = x, maximum along
y= -x.
(d) G (x, y) == x 3 + y3 . Along every line through the origin, the values on
one side are the negatives of those on the other. Hence no line offers
a strict maximum or minimum.

5. (a) h(x, y) == x 2 + y4. Clearly, 0 is a strict minimum.

h" (0,0) = [[2 0] [00]],

so h"(0,0)(v}2 = 2v I2 :::: 0 always, with h"(O, 0)(1, 0)}2 = 2,

h"(O, 0)(0, 1)}2 = O.
(b) H (x, y) == x 2 ; loose minimum, everything else like (a).
(c) k(x, y) == x 2 - y4; minimum along the x-axis, maximum along the
y-axis, all else like (a).
(d) K(x, y) == x 2 + y3. Along the y-axis, K has positive values above
the origin, negative below. Hence (0, 0) is not an extreme. Along y =
mx, K(x, y) = x 2(1 + m 3x) always has a strict minimum at (0,0).
Therefore no line gives a strict maximum, and (0, 0) is not a saddle.
262 Solutions to Exercises: Chapter 2

6. (a) Let b E N(a, 8). Write x(t) == [1 - t]a + tb and g(t) == f(x(t»,
III < Ilb~all' In view of the hint, if suffices to show that g has k + 1
derivatives in its domain, with

dj g (') .
dtj = f 1 (x(t»)(b - a)l.

The proof is a straightforward induction (compare the proof of Theo-

rem 2.4:1):

j<j+1) (x(t)) (b - a)j+! == [f(j+l) (X(/») (b - a)] (b - a)j

= [f(j+l)(X(/»)x'(t)] (b - a)j
d[j(j) (x(t))] .
= -':'::'-d~/"";";"~(b - a)l (chain rule)

d[j(j)(x(t»(b - a)j]
= (operator algebra).
dt

(b) At (0, 0), F = xeY = 0, F' = [eY xeY] = [10],

F" = [[0 eY] [e Y xeYJ] = [[0 1] [1 0]],

Fill = [[[00] [0 eY]] [[0 eY] [e Y xeY]]]

= [[[00] [01]] [[01] [10]]}

The polynomial is 0 + F'«x, y») + F"(X i1 »)2 + Fill «x31»)3 = x+

XYit YX + Xy2 +Y1?+YX] = x(1 + y + ~).
(c) It is easy to establish the following pattern in part (b): for c == (s, I),
f(k+!)(c) = [[ ... [0 0] ... [0 et ]] f(k) (C)] .

Hence
f(k+!) (c) «x, y»)k+!
= [[ ... [0 0] ... [0 et ]] «x, y»)k j<k) (c) «x, y»)k] «x, y»)

= [etyk j<k)(C)«X,y»)k] [~]

= xetyk + yf(k) (c) «x, y»)k.
Recursively, it follows that

f(k+!)( )«X,y»)k+! = (k+l)e t xl+set yk+!

c (k+ I)! (k+ I)!
Solutions to Exercises: Section 2.5 263

Its absolute value is at most etlxIIYlkk~~iW, which for fixed (x, y)

tends to 0 as k --+- 00.
(d) In the expression after "it follows" in part (c), set s =t = O. We
obtain

Ix 2xy 3xy2 ( y2 )
F(x y)
,
=0+-+-+-+
1! 2! 3!
.. · =x I+y+-+ .. ·
2!'

which is x times the series for eY •

(e) Assume that I is twice differentiable near b. With I'(b) = 0, Tay-
lor's theorem gives I(x)- I(b) = ["(C)ir b)2 forx in a neighborhood
N, c corresponding to x. Assume further that f" is positive definite at
b. We have seen that there is a positive e such that f" (b )(v) 2 ~ 2e for
every unit vector v. Finally, assume that I" is continuous at b. There
is a neighborhood P ofb in which 1If"(y) - f"(x)1I < e. Then for
x EN n P,

1"(c)(v)2 = !,'(b)(v)2 - [f"(b) - 1"(c)](v)2

~ 2e - II I" (b) - 1"(c)lIlIvIl 2 > e > O.

It follows that I(x) > I(b).

Section 2.5
1. (a) Anywhere but (1, 2). a(x-t+4Y) = 4 - 2y = 0 along the line y = 2.
Hence, away from (1,2), Theorem I guarantees a solution. Near that
point, the equation is not functional: x - y2 + 4y = 5 {} Y = 2 ±
JX=l, which guarantees two solutions for x > 1.
(b) !!1. - -aF/ax _ _1_
dx - aF/ay - 4-2y'
c) It is always possible: x = y2 - 4y + 5. Either directly or by Theorem
2,
dx = -8F/8y = 2y _ 4.
dy 8F/8x

2. a~;y = y~y = 0 on the x-axis. Hence you can solve for x on the rest of
the graph, meaning the y-axis minus the origin. Obviously the solution is
x = O. Similarly, you can solve for y along the x-axis minus the origin. At
the origin, you cannot solve for either variable; x = 0 gives multiple y's,
and vice versa.

(b) all the points in (a).

3. (a) Clearly, you can always solve: y = (e- X _ x3)1/3.

264 Solutions to Exercises: Chapter 2

(b) That function is undifferentiable at the point where e-x - x 3 = O.

(c) Theorem 1:

-at/ax
allay
=
provided y =f:. O. Explicitly,

y'
1
= _(e- X _x 3 )- 2/ 3 (_e- X _ 3x 2 ) -3x-2eX
= _1 - ---1
3 3y2 eX

provided y =f:. O. They match.

4. (a) Write F == x2 + y2 - z2. You can solve F = 0 for x wherever

~~ = 2x =f:. O.
(b) .2L _ -aF/iJ(y,z) _ -[2L-2z1 _ [_r _ &0]
a(y,z) - aF/ax 2X- - X x'
(c) No. Near any point where x = 0, the points (z, 0, z) and (-z, 0, z)
both solve the equation; the relation is not functional with respect to
x.
5. Let F(x, y) == g(x) - y. Clearly, ~~ = g'(x), ~; = -1. We always have

F(x, y) - F(s, t) - [g'(s) - 1] [~=:]

= g(x) - y - g(s) + t - g'(s)(x - s) +y - t
= [g(X) - g(s) _ g'(S)] (x _ s),
x -s
or zero if x = s. If g is differentiable at x = s, then this expression is
0(11 (x, y) - (s, t)11). Hence F(x, y) is a differentiable function oftwo vari-
ables at (s, t).
Now assume that g is differentiable near x = b. Then F is differentiable
near (b, g(b»). If g' is continuous at b, then F' is continuous at (b, g(b»).
Finally, ~; =f:. 0 there. By Theorem 1, F = 0 can be solved for x: g(x) -
y = 0 <=> F(x, y) = 0 <=> x = hey) in some box b - 8 ::: x ::: b + 8,
g(b) - B ::: y ::: g(b) + B; and

h'(y) = -aF/a(y) = _1_.

aF/ax g'(x)

Section 2.6
1. (a) Vector description:

t = (x', y', z') = (-2 sinO, 2cosO, i') = (--13,1, e1l'/3) ;

Solutions to Exercises: Section 2.6 265

then the line is (1,,,(3, e1l"/3) + (((-,,(3, 1, e1l"/3))). Parametric equa-

. -(x-I) y-./3 z-e,,/3
tIOns: ./3 = I = --;;rrr'
(b) t = (:~, ~) = (1,2x) = (1,4); the line is (2,4) + (((1,4»)); the
equations are x 12 = 9, or y - 4 = 4(x - 2).

2. (a) Write F(x, y, z) == x 2 + y2 - z. Then VF = (2x, 2y, -1) is never

0, so the surface is smooth. The tangent plane is given by (Example
5)

0= (x - b). VF(b) = (x - a, y - b, z - c). (2a, 2b, -1)

= 2ax + 2by - z - c,

because c = a 2 + b2 .
(b) G(x, y, z) == x 2 + y2 has VG = (2x, 2y, 0). The gradient is never
zero on the surface, because IIVGII 2 = 4x 2 + 4y2 = 16; the graph is
smooth. The plane is given by

0= (x - a, y - b, z - c). (2a, 2b, 0) = 2a(x - a) + 2b(y - b),

or ax +by = 4.
(c) H(x, y, z) == x 2 +y2+Z2 has IIVHII 2 = II (2x, 2y, 2z)1I 2 = 4r2 # 0,
smooth. The tangent plane equation is

0= (x - a, y - b, z - c) • (2a, 2b, 2c)

= 2ax + 2by + 2cz - 2(a 2 + b2 + c 2),
or ax + by + cz = r2.
(d) The given parametrization is undifferentiable where x = y, but it is
equivalenttoz 3 = x-Yo The latter is smooth, because V(x-y- z3) =
(1, -1, -3z2 ). The tangent plane has

0= (x - a) - (y - b) - 3c2(z - c) = x - y - 3c2z - a + b + 3c 3
= x - y - 3(a - b)2/3 z + 2(a - b).

The plane is vertical wherever a = b.

3. The given vector has y = 2x, so we try the image ofthe line y = 2x ~ xy-
plane. Thus, x = x, y = 2x, z = x 2 + y2 = 5x 2.1t crosses the origin, with
tangent vector
dX dy dZ)
( dx' dx' dx = (1,2,0).
266 Solutions to Exercises: Chapter 2

4. Solve for y: y = x + z2 - 1. By Example 4, (1, ~, 0)= (1, 1,0) and

(0, ~, 1) = (0, 2c, 1) make a basis. Reconcile the geometry: z2 = 1 -
(x - y) is a parabolic cylinder ruled by lines parallel to its spine, which is
the line x = y, z = 1; therefore, the direction of i + j is always tangent to
the cylinder.

5. Write G(x, y) == F(x) - y. By hypothesis, VG = (F'(x), -1) is continu-

ous, and it is never O. By Theorem 3, VG(a, F(a» is normal to the graph
of G = O. Hence the tangent is given by

0= (x - a, y - F(a». (F'(a), -1),

or y - F(a) = F'(a)(x - a).

6. (a) By the definition of curve, g is differentiable. Therefore,

g(s) - g(t) g'(s*)(s - t)

= ,
IIg(s) - g(t)11 IIg'(s*) II Is - tl
assuming that the division is legal. If g is smooth at g(t), then by def-
inition g' is continuous and nonzero at t, so g' stays nonzero nearby,
making the division legal. Set u == II~~~:~II. Then
g'(s*)(s - t)
-=-.:........:.....:....-----':.... --+ U
Ilg'(s) II Is - tl
as s --+ t+, --+ -u as s --+ t-, and those are the conclusions sought.
(b) Let g(s) = (x(s), x(s)2/3) and (0,0) = g(t). By the one-to-one as-
sumption, x(t+) and x(t-) are of opposite signs, say x(t+) > 0 >
x(t-). Then for s --+ t+, we have g(s) - g(t) = x(s)2/3(x(s)1/3, 1),
whose limiting direction is (0, 1). For s --+ t-, the forward secant is
g(t) - g(s) = -x (s)2/3(x(s)1/3 , 1). Since the scalar x(s)2/3 remains
positive, this secant lines up along -(0, 1).
(c) Argue as in (b) with g(s) = (x(s), Ix(s)l). The secant on the t+ side
is (x(s), Ix(s)1) = x(s)(I, 1); it is constantly along the line of (1,1).
On the t- side, Ix(s)1 = -x(s), and the secant is (x(s), Ix(s)1) =
x(s)(1, -1), different line.

7. The dot products fit the following pattern:

Solutions to Exercises: Section 3.1 267

Chapter 3
Section 3.1
1. The equation x = J 1 + x2 implies x2 = 1 + x2, impossible.
2. (a) One sample sequence:

= -0.1736 ... ;
cos 100
cos(Ans.) = -0.9999954073 ... ;
cos(Ans.) = -0.9998476966 ... ;
cos(Ans.) = -0.9998477415 ... ;
cos(Ans.) = -0.9998477415 ... ;

fixed.
(b) cosO° = 1 > 0, while cos 1° = 0.9 ... < 1, so cos x o = x occurs
between 0 and 1. There is a different fixed point because cos XO is not
cosx: cos x o = cos(fto).
(c) From most starts, the iterations get out of [-1, 1]. For example,
cos- l 0.1 ~ 1.47; similarly, cos- l 0.9 = 0.45 ... , cos- l (Ans.) ~
1.1. Once they exit, the result is ERROR. From a start near 0.739 RA-
DIAN, the iterations slowly move apart. The reason is the mean value
theorem:

I cos- l (Ans.2) - cos-I(Ans'I)1I = 1 > 1.

IAns.2 - Ans·tI ~

(d) The RADIAN answer converges slowly to zero. The absolute value
has to drop, because I sin x I < Ix I for x =f:. O. But the drop is slow,
because
I sin(Ans.) - 01 = Icost I ~ 1
.:..-.......:--.:..-..----.:.
IAns. -01
as Ans. descends to O.

3. (a) For this function, G(x) - 7 =3- ~l = 3(Xx-7). Therefore,

Xi+l - 7 G(Xi) - 7 3(Xi - 7)/Xi 3
Xi -7
= Xi -7
=
Xi -7
=-
Xi

This says that Xi+l is on the same side of 7 as Xi is, and at a distance
from 7 reduced by a factor ~, no more than the fixed < 1. :0
(b) A negative xo gives Xl = G(xo) = 10 - ;~ > 10. If 0 < xo < 2.1,
then Xl = 10 - ;~ < 0, X2 = 10 - ;: > 10. Either way, we are back
in (a).
268 Solutions to Exercises: Chapter 3

(c) G is one-to-one, with G- 1(y) =

IJ~y' If you set xo at any of 0,
G- 1(0) = 2.1, 2.66, ... , then you eventually iterate G
G- 1(2.1) ~
down to O. These numbers increase toward 3, because
3 - G- 1(y) 3(3 - y)/(10 - y) 3 3
3-y = 3-y
=--<-.
10- y 7
(d) If 0 < x < 3, then 3 - G(x) = -7 + ~1 = 7(3;x) > 2(3 - x). In
words, the distance from 3 more than doubles with each iteration. If
x is on list (c), then the iterates reach zero; if not, then they pass zero,
then immediately go over 10.
4. (a) Yes: Ilf(x) - f(y) II :::: IIx - yll implies that x ~ y =} f(x) ~ f(y).
(b) No: f (x) == ~ is a contraction, because
If(x) - f(y)1 = Illxil ; Ilylll :::: IIX; YII,

but f is not differentiable at x = O.

(c) With elementary tools, we can prove at least this: Ilfil :::: "foiK. Let
u be a unit vector. From
l' [f(X + tu) - f(x) - f (x)(tu) f(x) (tu) ]
d
I
) (
(x U -
) _
1m
1_0 t
+ t
,
we deduce that Ilf (x) (u) II = K. Setting u == e j, we get K ~ f (x) (e j)
= IIcolumn j off (x) II, so that
IIf'(x) 112 = IIcolumn 1 of /,(x) 112+ . + II column m of f'(x) 112 :::: mK2.
5. If x and y are both fixed points, then
IIx - yll = IIf(x) - f(y) II :::: Kllx - yll =} (1- K)IIx - yll :::: O.
Since K < 1, necessarily IIx - yll = o.
6. (a) fl(X) = 1 +0, h(x) = 1 + It
Idt = 1 +x, J3(x) = 1 + It (1 +
d x2 x k- 1
t) t=l+x+ , ... ,!k(x)=1+x+···+(k_l)!'
2
(b) The limit is clearly eX. That function solves the initial value problem.
(c) The first five iterates are
gl = 1 + (sinx), g2 = 1 + x + (1 - cosx),
x2
g3 = 1 +x +"2 + (x - sinx),

g4 = 1+ x + x; + (2~;3) + (x22 - 1 + cosx) ,

2 3
g5 = 1 + x + x + x + x4 + (x3 _ x + sin x) .
2 3! 4! 3!
Solutions to Exercises: Section 3.2 269

In these, the polynomial tends to eX. The part in parentheses looks like
either
± (sinx _[x _x3! + x5! 3 5
_
".
])

or ±(cosx - [1 - ~~ + ~; - ... D, both of which tend to O.

Section 3.2
1. The calculation is tedious but straightforward. The result is det(h') = 2.

COS 0 sin e/> - p sin 0 sin e/> p cos 0 cos e/>]

2. (a) :~J:~ = [ sinO sine/> pcosOsine/> psinOcose/> . See Answer 1.3:3.
cose/> 0 -p sine/>
(b) It is straightforward to calculate the cofactors and determinant, so we
will give a different view. Observe that

a(
x, y, z
) [1 0 0] [COSO sine/> -sinO cosocose/>]
0 ~ sin e/> 0 = sin 0 sin e/> cos 0 sin 0 cos e/> .
a(p, 0, e/» 0 0 1 cose/> 0 - sine/>
p

The last matrix is orthogonal: Its rows are orthonormal, as are its
columns. For such a matrix, the inverse is the transpose. Our equation
says "Jacobian times diagonal = orthogonal." It follows that J- 1 =
diagonal times transpose =transpose with rows 2 and 3 multiplied by
p s:n 1/>' ~, respectively. Thus,

COS 0 sin e/> sin 0 sin e/> cos e/> ]

-1 -sinO cosO 0
J = [ psml/> psml/> '
cos 0 cos t{> sinO cost{> - sint{>
p p p

defined as long as p sin e/> =1= 0; you have to stay off the z-axis.
(c) For example,

aF at a(p, 0, e/» at at at [ COS-sinO

0 sin e/>]
ax- ~ a(p, 0,4» ax ~ [ap ao a4> 1 a:~.
at at sinO at cosOcose/>
= -COSI7SlD'I"-
ap
ll' A..
- - - + - p .
ao p sine/> ae/>

3. (a) g(x, y, z) == (x 3, y3, z3) is one-to-one, even though a(x~~.z) = 0 at

(0,0,0).
270 Solutions to Exercises: Chapter 3

(b) Under the hypothesis, I I exists. Let x, == b + tel, y, == f(x,). Since

fis continuous at b, y, ~ f(b) as t ~ O. We have

I I (y,) - I I (f(b» = (lIf(X,) - f(b) II )-1

IIY, - f(b) II IIx, - bll

(Why are all those differences nonzero?) Since the fraction in paren-
theses approaches t(b)(el) = 0, the other fraction tends to 00. There-
fore, there cannot be an operator L with 1I(y,) _II (f(b» Rj L(y,-
f(b»); I I cannot be differentiable at f(b).

Section 3.3
1. The counterexample, as always, uses cubics: x - y - u 3 + v 3 = 0 =
x + y - u 3 - v 3 iff u = x l/3 and v = y1 / 3, even though :~t.~~ = 0 at
(x,y,u,v)=O.

abc
2. (a) e f g :F 0; d, h, I are immaterial.
i j k
(b) Set d = h = I = 0 and all the coefficients == 1.
3. (a) Write f == x 2 - y2 - U, g == x 2 + y2 - v. Then :~f.~~ = 8xy.
Away from the axes, Theorem 1 applies and guarantees a solution.
We always have u + v = 2x 2 • Hence near any point on the y-axis,
x = ±(~ + !)1/2 gives distinct solutions in any neighborhood, and
the relation is not functional in x; analogously for y near the x-axis.
(b) The points v = constant c > 0 constitute a circle x 2 + y2 = c. The
=
points u constant b > 0 make up a hyperbola x 2 - y2 b. Sketch=
such a circle and hyperbola. Along the latter, distance from the origin
increases as you move away from the vertex. Therefore, if the two
curves meet in a quadrant, then they are transversal. If instead they
meet at the x-axis, then they are tangent; that is the bad situation. The
same argument applies if b < 0, in which case the tangencies occur
along the y-axis. If b = 0, then the hyperbola becomes its asymptotes,
and the intersections are automatically transversal along y = ±X.
For v = 0, we no longer have a curve. Finally, v < 0 is not allowed.

4. The plane-cone intersection is a parabola given by (x + 4)2 = x 2 + y2, or

y2 = 8x + 16, along with z = x + 4. Where y = 8, we have x = 6. Hence
the tangent vector is

dy dx
( 1, dx' dZ) 8 1) = (1,0.5,1).
= ( 1, 2y'
Solutions to Exercises: Section 3.4 271

The equation is x - 6 = 2(y - 8) =z - 10. Alternatively,

0= (x - 6, y - 8, z - 10). V (x2 + y2 - Z2)

= 12(x - 6) + 16(y - 8) - 20(z - 10)

plus 0 = (x - 6, y - 8, z - 10) • V (x + 4 - z) = x - 6 - z + 10, a system
that reduces to z = arbitrary, y = 3 + ~, x = z - 4.

5 • If ( ill ill) (ill ill) are m

ox! ' ••• , OXn , ••• , OX!'···' OXn
.d den,
epen t then o(F! ..... Fk)
o(X! ..... Xn)
has k independent rows, the maximum possible. Hence it has k indepen-
dent columns, say

, ... ,

The implicit function theorem guarantees that you can solve the system for

Xl = gl (Xk+l, ... , xn), ... , Xk = gk(Xk+l, ... , xn)·

This means that the system is equivalent to

an (n - k)-surface.
6. By the argument in Theorem 2(a), we can solve

Xl = gl(tl, ... , tn-k),'" ,Xn = gn(tl, ... ,tn-k)

for n variables, meaning tl, ... ,tn-k and (say) Xl, •.• ,Xk. Then
Xl = gl(t) = gl(h(xk+l, ... ,xn»), ... ,
Xk = gk(t) = gk (h(Xk+l, ... ,xn»)

is the intersection of the k hypersurfaces

0= /j(x) == Xj - gj (h(Xk+1 , ... ,xn »), j = 1, ... ,k.

Section 3.4
1. V f = (y, x) = +
J... V(x 2y - 5) = (J..., 2J...) becomes y = J..., X = 5 - 2y =
2J..., leading to y = i, X = ~, f = 2i. At that point, the line X + 2y = 5
has slope -!, the hyperbola xy = 2i has slope fxf = 5~5, so they are
tangent.
2. To minimize f == distance2 = (x-a)2+(y-b)2, set V f = (2x -2a, 2y-
2b) = J...(c, d). The equation implies X - a = A{, Y - b = Af, and f =
A2(C~+d2). (Notice that (x -a, y -b) is in the line of (c, d), perpendicular to
the original line.) Also, J...c 2+J...d2 = 2(x-a)c+2(y-b)d = 2e-2ac-2bd,
so J... = 2(e-ac-bd). Hence the minimal distance is f = le-ac-bdl.
c2+d2 J c2+d2
272 Solutions to Exercises: Chapter 4

3. V(x +2y+3z) = AV(x 2 + y2+Z2 -1) gives 1 = 2AX, 2 = 2AY, 3 = 2AZ.

(These say that the radius vector (x, y, z) is perpendicular to the plane.)
Divide to get? = 2, i
= 3; then x 2 + (2x)2 + (3x)2 = 1 yields two
candidates, ±%3). The + gives max, - gives min.

4. From Answer 3.3:4, the intersection satisfies y2 = 8x + 16 and z = x + 4.

The lowest point there is (-2,0,2). If you are at (0, 0, 2) or lower on the z-
axis, then it pays to stay in the xz-plane, so the answer should be (-2,0,2).
We examine:

(a) V (x 2 + y2 + (z - 2)2) = J..V (x 2 + y2 - z2)+JLV(x+4-z) yields

2x = A2x + JL, 2y = A2y, 2z - 4 = -A2z - JL. If y =f:. 0, then we
get in succession A = 1, JL = 0, z = 1, x = -3, y undefined. Hence
necessarily y = 0, which specifies x = -2, z = 2.
(b) Change the equations to 2x = A.2x + JL, 2y = J..2y, 2z - 2b =
°
-A2z - JL. As before, y =f:. implies J.. = 1, JL = 0, z = ~,x = ~ -4,
y = ±v'4b - 16. Thus, if b ::: 4, we are again led to y = 0. In words,
if (0, 0, b) is on or below the plane of the parabola, then there is no
reason to leave the xz-plane. If instead b > 4, then you travel to a
place other than the vertex.

5. The intersection circle z = ,J5, x 2 + y2 = 4 has vertical tangent planes.

At the place (;fa, ;to, ,J5), the normal to the tangent plane is the circle's
radius vector (;fa, ~, 0). This vector is also normal to the plane x +3 y =
-20
JiO'
6. By Lagrange's method, / == (Xl - bl)2 + ... + (xn - bn)2 is least when

(2Xl - 2bl, ... ,2xn - 2bn) = Al V /1 (x) + ... + Ak V!k(x).

The solution x = c therefore satisfies

b - c is in the orthogonal complement of the tangent plane.

Chapter 4
Section 4.1
1. The "unit cube" {x: °: :
x j ::: I} and its translate to the upper right {x:
1 ::: Xj ::: 2} have precisely (1,1, ... ,1) in common, a point on each
boundary.
Solutions to Exercises: Section 4.1 273

2. (a) x E B n C <=> aj ::; Xj ::; bj and Cj ::; Xj ::; dj for each j <=> Xj :::
max{aj, Cj} andxj ::; min{bj,dj} for each j.
(b) In view of (a), B n C has an interior point <=> there exists x with
max{aj,cj} < Xj < min{bj,dj} for every j <=>thereisxwithboth
aj < Xj < bj and Cj < Xj < dj for every j <=> there is x E int(B) n
int(C).
(c) Assume that B and C do not overlap and x E B n C. The segment
from x to ~ is interior to B, except possibly for x. If x were in
int(C), then some neighborhood N(x, 8) would be interior to C, so
points near but unequal to x on the segment would be interior to B
and C. Hence x E int(C) is impossible. We conclude that x E bd(C),
and likewise with B.
(d) A neighborhood and a line through its center do not overlap, because
the line has no interior points, but their intersection consists of points
interior to the neighborhood.
(e) Sand T overlap <=> there is x E int(S) n int(T) (definition) <=> there
are N(x, $) S;;; S and N(x, t) S;;; T <=> there is N(x, r) contained in
both S and T <=> there is x E int(S n T).

3. In the picture,

VI = (Xl - xO)()'2 - YO) = (Xl - XO)(YI - YO) + (Xl - xO)()'2 - YI),

V2 = (X2 - XI)(Y2 - YI),
V3 = (X3 - XI)(YI - YO) = (X2 - XI)(YI - YO) + (X3 - X2)(YI - YO),
V4 = (X3 - X2)(Y3 - YI) = (X3 - X2)(Y2 - YI) + (X3 - X2)(Y3 - )'2),
Vs = (X3 - X2)(Y4 - Y3),
V6 = (X2 - XO)(Y4 - Y2) = (Xl - XO)(Y3 - Y2) + (Xl - XO)(Y4 - Y3)
+ (X2 - XI)(Y3 - Y2) + (X2 - XI)(Y4 - Y3).
The twelve right-hand elements sum to (X3 - XO)(Y4 - YO) = V ([a, bD.
4. Suppose Tk S;;; Sj. Since necessarily Tk S;;; C, we have Tk S;;; Sj n C.
Hence the union of the Tk contained in Sj is a subset of Sj n C. To prove
Sj n C S;;; union, let x E Sj n C. Since Sj and C overlap, Exercise 2b tells
us that Sj n C is a box. Hence x = lim Xi for some sequence (Xi) from
int(Sj n C) = int(Sj) n int(C) (Exercise 2e). Those Xi cannot come from
Tk contained in other Sm. They must come from the Tk contained in Sj; that
is, x is a closure point of the union. Since the union is closed, x E union,
and Sj n C S;;; union.

S. (a) Because Band C overlap, there must exist a neighborhood N S;;;

B n C. This N must have points interior to some Sj and some Tk.
Hence P'R. = {Sj n Tk-: Sj and Tk overlap} is nonempty. Any such
intersection is a box, by Exercise 2b.
274 Solutions to Exercises: Chapter 4

(b) If Sj n Tk and Sz n Tm overlap, then Sj overlaps S/, forcing j = I, and

similarly k = m.
(c) Suppose x E B n C. Then x E B = SI U ... U SJ, so that x E some
Sj. Let (Xi) -+ x from int(Sj)' An infinity of Xi must come from a
n
single Tk. Since these points are common to Sj and and are interior
to Sj, Exercise 2c implies that Sj and Tk overlap. Also, they are in Tk
and converge to x. Since Tk is closed, x E Tk. We have shown that x
is in some Sj n Tk E pn. Hence B n C 5; union of the S/ n Tm. The
opposite inclusion is trivial.
The product pn is the simplest refinement, in that if Q refines both
P and n, then Q refines pn: If each Q/ is contained in some Sj and
some n, then Qz 5; Sj n Tk.

Section 4.2
1. On any subinterval SJ of any partition p, sup f = inf f = 1. Therefore,
u(f, P) = l(f, P) = V(SI) + ... + V(SJ) = V(B); f is integrable, with
integral V(B).

2. (a) In the subinterval from (Xj-l, Yk-d to (Xj, Yk), supg = supx = Xj.
Hence
J K
u(f, P) = LLXj(Xj -Xj-l)(Yk - Yk-l).
j=1 k=1
Clearly, the k-summation is
K
Xj(Xj - Xj-l) L(Yk - Yk-l) = 4xj(xj - Xj-l),
k=1

and u(f, P) = Ll::U9 4xj(xj - xj-d.

(b) The answer in (a) is a Riemann sum for the one-variable integral
fg 4x dx = 18. If IIPII approaches 0, then each Xj -Xj-l approaches
0, so u(f, P) -+ 18.
(c) The box bounded by x = 0, x = 3, Y = 0, Y = 4, Z = 0, Z = 3, is cut
into congruent triangular prisms by the plane z = g(x, y) = x. The
box has volume 3(4)3 = 36. Each prism therefore has volume 18.

3. The function is 1 on the squares from n, !) to (1,1), (t, t) to (!, !),

(1,1) to (t, t), .... For the lower integral, consider the cross-partition
P3 defined by the first three. Those three are shaded in the figure below,
which also shows the (3 + 1)2 subintervals of P3. Clearly, minh = 1 on
°
the shaded subintervals, = on the others, so I(h, P3) = ir
+ :b +
We can do the same with the first k squares, producing a partition Pk with
ir.
I(h, Pk) = ir ir
+ ... + (21)2' Therefore, L(h, B) ~ +:b + + ... = ir t·
Solutions to Exercises: Section 4.2 275

:
I I
I I
I I

-.-
I I
I I
I I
I I

I
I

~L
I I
___ ~~~~~~~~
I
I I I
x
1 1 1
8 4 2
Exercise 3.

For the upper integral, let 8 be small. Define Q4 to be the cross-partition

determined by the boxes from x = y = (1;:e) tox = y = 1,x = y = (14 e )
tox=y= (He) 2 ,x=y=
(1-e)
8 tox=y=
(1+e) d
4 ,an x=y=
0 to
x = y = (1t e). (These are overlapping boxes that contain the ones where
h is nonzero.) On the subintervals of ~ not contained in those four, h is
identically O. Hence

the < sign needed because the boxes overlap. Extending the construction
to boxes surrounding the first k squares, we build Qk with

u(h, Qk) <

1+8)2 + ... + 1 (1+28)2
1 ( -2- ~ + 1 (1+8)2
~

< (1+28) 2(1

-+ 1- + 1)
22 ... +22k -
22k
< (1+28) 2(1-+-1)
3 2k .

It follows that U(h, B) ::::: (1+;e)2. Since this holds for arbitrarily small 8,
we have U(h, B) ::::: ~. Hence h is integrable, and Is h = ~.

4. (a) If I and g are integrable, then their integrals are limits of Riemann
sums. Then the Riemann sums for al + fig satisfy

This proves both assertions.

(b) If I ~ g, then u(f, P) ~ u(g, P) for every partition, forcing Is I =
U(f, B) ~ U(g, B) = Is I·
276 Solutions to Exercises: Chapter 4

(c) => Assume that I is integrable on B, and let 8 > 0. By Theorem

2(a), there is a fineness 8 below which upper sums are within 8 of
lower sums. Let Ql, ... , QJ be partitions of SI, ... , SJ finer than 8.
Clearly, Ql U ... U QJ is a partition of B, finer than 8. Also,

and similarly for lower sums. Hence for each j,

u(j, Qj) -J(j, Qj) !:: [u(/, Ql) -l(j, Ql}]

+ ... + [u(j, QJ) -l(j, QJ}]
= LU(j, Qj) - Ll(l, Qj) < 8.
We conclude that I is integrable on each Sj.
<= Assume that I is integrable on SI, ... , SJ. Given 8 > 0, there ex-
ist norms 81, ... ,8J below which partitions Ql, ... , QJ of SI, ... , SJ
7
have upper sums within of the corresponding lower sums. Let 'R be
a partition of B finer than min{81 , ... ,8J}. The partitions 'RQl, ... ,
'RQJ are refinements of Ql, ... , QJ. Let'R+ == 'RQl U ... U'RQJ.
Then

u(j, 'R+) -l(j, 'R+) = LU(j, 'RQj) - Ll(l, 'RQj)

= L[u(j, 'RQj) -l(j, 'RQj)]
!:: L[u(j, Qj) -l(j, Qj)]
8
< J-
- J
= 8.

By Theorem 2(a), I is integrable.

For the relation between the integrals, assume that I is integrable.
Establish individual partitions Ql, ... , QJ and Q == Ql U ... U QJ.
These have u(j, Q) = u(j, Ql) + ... + u(j, QJ). Letting the norms
of Ql, ... ,QJ tend to 0, we ha~e u(j, Q) .... fS I 1+··· + fs / /-
That forces U (I, B) !:: sum of mtegrals. Analogously, we estabbsh
L(j, B) ~ sum of integrals, and equality follows.

(d) By extending the boundaries of B, we can create a partition {B, SI, ... ,
SJ} of D. By part (c), fD 1= fB 1+ fS I 1+··· + fS I I· The last J
terms are nonnegative, by part (b). Hence fD I ~ fB I·

(e) LetM == sup/,m == inf/. ThenmV(B) = mfB 1 (by Exercise 1)

= fB m (part (a» !:: fB I (part (b» !:: fB M = MV(B).
Solutions to Exercises: Section 4.2 277

5. By linearity. f(ax + f3y + I5z) = a f x + f3 f y + 15 f z. From Exercise 2.

it is clear that fB x for a box B has upper sums
J K M
u(x. 'P) =L L I>j(Xj - Xj-l)(Yk - Yk-l)(Zm - Zm-l)
j=l k=l m=l

Since the last integral is ¥- ~ = (bl -al ~bl +a l ). we see that fB x =

V(B)~. By symmetry.

f (ax + f3y + I5z) = V(B) (aal + abl + f3 a2; f3b2 + l5a3 + 15b3)
(a+ b)
= V(B)(a. f3. 8). 2 .

°
6. On every subinterval Sj. inf F = and sup F = 1. Hence every upper sum
is L 1V(Sj) = V(B). every lower sum is LOV(Sj) = 0. and U(F. B) >
L(F. B).

7. (a) On any subinterval Sj.l/(x)I-I/(y)1 ~ I/(x) - l(y)1 ~ SUPj 1-

infj I for any x. y. Therefore. SUPj 1/1- infj III ~ SUPj I - infj I.
Summing over the subintervals and separating sups and infs. we ob-
tain
u(l/l. 'P) -1(1/1. 'P) = u(f. 'P) -l(f, 'P).
If I is integrable. then there are 'P with u(f, 'P) -l(f, 'P) < s. Hence
the same goes for III, and III is integrable.
F
(b) Let be Dirichlet's function on the unit square inR2, and G == F-!.
Then G is unintegrable, because every upper sum is every lower!,
sum is -!. !
But IGI = everywhere.
8. Assume that I is continuous on B. Then I has a minimum I(a) and a
maximum I(b). By Exercises 1 and 4, I(a) = fB ~ fB tf~) ~ tt:»
fB ~ = I(b). (Remember that V(B) is required to be positive.) Thus.
fB tt~» is between the extremes of I. By the intermediate value theorem,
there exists C E B such that I(c) = fB t~~).
9. By part (d), l(f, 'P) ~ u(f, R) for any 'P and n. This says that l(f, 'P) is
a lower bound for {u(f, R)}. so I (f, 'P) cannot exceed the infimum of the
set: l(f, 'P) ~ inf{u(f, R)} = U(f, B). This last says that U(f, B) is an
=
upper bound for {l(f, 'P)}. Therefore, L(f, B) sup{I(/, 'P)} ~ U(f, B).
278 Solutions to Exercises: Chapter 4

Section 4.3
1. It is easy to prove, by induction on the dimension n, that if X is on one wall
of Band y on a different wall, then the segment xy is interior to B, except
for x and y. If B n C has both x and y, then it contains xy. That says it has
points interior to both Band C; Band C overlap.

2. If I = 0 in int(B), then on every subinterval of a partition p, inf I = O.

Therefore, l(j, P) = 0, and L(j, B) = O. Assuming that III :::: M on the
walls of B, let Q be a cross-partition created by a big box C of volume
V(B)- iI. Thenu(j, Q):::: OV(C)+MI: V (other subintervals) =e. We
conclude that U(j, B) = 0, I is integrable, fn I = O.
3. (a) Let B be any box containing A == {Xl, ... ,xm }. Clearly, int(A) is
empty, so v*(A) = O. Let P partition B into kn congruent subinter-
vals. A subinterval touches cl(A) iff it contains some Xj. A given Xj
can lie in only 2n subintervals (induction on n). Hence V(A, P) ::::
2nrJB) , true for arbitrary k. We conclude that V*(A) = O. The set is
Archimedean, with V(A) = O.
(b) It has empty interior, so v* = O. Also, you can put it in the box

whose volume is 2e(b2 - a2) ... (b n - an). Let e ~ 0, and conclude

that V* = 0 = v*.
(c) We may assume th.at a, b >, O. Let k be a positive integer. Draw the
grid of lines x = T' Y = ~,O :::: j :::: k. They create a partition P
of the box 0 :::: x :::: a, 0 :::: Y :::: b, covering the set. Each subinterval
has volume ~. The lowest row of subintervals has k that meet the
closure, no interior ones. The next row up has k meeting the closure,
k - 3 interior ones; you lose 2 on the right, because the next-to-Iast
one has its upper-right corner on the boundary. In the next row, there
are k - 1 on the closure, k - 4 interior. Continue the pattern to obtain
ab ab
v(A, P) = (1 + 2 + ... + k - 3) k 2 = (k - 3)(k - 2) 2k2 '
ab ab
V(A, P) = (1 +2+ .. · +k +k) k 2 = k(k + 3)2k2 '
Passing to the limit, we conclude that V*(A) :::: at : : v*(A). There-
fore, A is Archimedean, with volume ¥.
4. Suppose b E int(A). By definition, some neighborhood N(b, 8) is a subset
of A. Then the box B from (bl- 2Jn' ... ,bn - 2Jn) to (bl + 2Jn' ... ,bn+
2Jn) is a subset of N(b, 8), therefore of A. Any partition P with B as a
subinterval has v(A, P) ~ v(B), making v*(A) ~ v(B) > O. By contrapo-
sition, v*(A) = 0 :::} int(A) has no members.
Solutions to Exercises: Section 4.4 279

5. (a) Unit cube n Qn. It is not Archimedean; its volume is undefined.

(b) If A is Archimedean, then V*(A) = v*(A). If A has empty interior,
then v*(A) = O. If both, then V(A) is defined, equals O.

6. Assume that A has zero volume and If I ~ M. Let B be any box superset of
A. There exists a partition of B with V*(A, P) < 2~' On the subintervals
ofP exteriorto A, fXA is identically zero. Henceu(fXA, P) ~ M L (vol-
umes of closure subintervals) = MV*(A, P) ~ ~, while I(fXA, P) ~ -M
L (volumes of closure subintervals) = -MV*(A, P) = 28 • Therefore
U (f XA, B) ~ 0 ~ L (f XA, B), and the conclusion follows.

7. (a) XA,XB,andXAnB all have values oand l,andXAnB(X) = 1 {} X E A

and x E B {} XA(X) = 1 and XB(X) = 1 {} XA(X)XB(X) = 1.
(b) AU B is the disjoint union of A - B, An B, B - A, and the two sides
match in each of those sets. The rest of the space is A * n B* , in which
both sides are O.
(c) If A £ N(O, M), then IIxll ~ M => XA(X) = 0, so limx_ oo XA(X) =
O. Conversely, if limx_oo XA (x) = 0, then there is N (0, R) outside
of which XA ~ !, forcing XA = O. Hence IIxll ~ R => x fj A; that is,
x E A => IIxll < R, and A is bounded.
8. (a) Ignore the value xs(b). If every neighborhood of b has points from
both Sand S* , then the limit does not exist. If instead some neighbor-
hood of b has points from only one of the two, then the limit exists.
Hence b has to be surrounded by S (interior to S or isolated point of
S*) or by S* (vice versa).
(b) In view of (a), b has to belong to the set that surrounds it. Hence b has
to be interior to S or interior to S* = exterior to S; in fewer words,
b fj bd(S).

Section 4.4
1. Unit square n Q2 has the unit square as boundary.

2. By Theorem 2, there is a set of boxes Bj I, ... , B jk(j) covering D j with

V(Bjl) + ... + V(Bjk(j» < j. Then U]=I U!~? Bjk is a union of boxes
covering DI U··· U DJ with volume-sum < 'f = e. By Theorem 2, DI U
... U D J is meager.

3. If D is meager, then you can cover it with BI, ... , B J having volume-sum
< e. If C £ D, then the same boxes cover C.

4. By induction: Let (XI, •.. ,xn) E S(O, a), so thatx l2 + . +xn2 = a2 . First,
X 12 ~ a 2 , equivalent to 1:;.1
~ 1. Hence there exists a unique tl E [0, 1l']
280 Solutions to Exercises: Chapter 4

with ~ = cos tl· Next, x? + xl ~ a 2 , or IJa:~xI21 = IaJ~tl I ~ 1. Hence

there exists one t2 E [O,1l'] with a~t
I
= cost2· Next, Ja2-x/-xl
X3 =
? t must equal some cos t3, and so on. Eventually, x l 2
asm Ism 2
+ ... +
x n2 =
a forces IXn I = a sin tl ... sin tn-I, and to cover the two possibilities Xn
2 >
oand Xn < 0, we extend the domain of tn-I to [0, 21l']. That parametrization
covers the sphere, one-to-one.
Obviously, each :~ is continuous. By induction, we may also show that
J
any two are orthogonal and have lengths a, a sin tl, ... ,a sin tl ... sin t n-2.
(Look ahead at Example 5.6:3.) Hence they are independent, making the
parametrization smooth, as long as tl, ... ,tn-2 are different from 0 and 1l'.

5. (a) Assume that OJ has some points of OJ. By construction, OJ possesses

no points from bd( OJ ).It cannot possess points from ext( OJ), because
then (being connected) it would also have to have points from bd( OJ).
Hence OJ S; int( OJ) = OJ. It cannot be OJ = OJ, because even the
boundary of OJ has no points from bd(Oj).
(b) Suppose OJ(1), OJ (2) , ••• is the subsequence of OJ that intersect OJ.
Then OJ intersects OJ(1) and j(1) ~ i. By (a), OJ S; OJ(I). This
means that OJ (2) , OJ(3), ... all intersect OJ(1). Hence OJ (2) , OJ(3), ...
are all subsets of OJ(1), and OJ(1) = OJ(1) U OJ(2) U .... Define
k(i) == j(1), and the conclusion follows.
(c) We cannot predict whether k(i) and k(m) match, or which is larger.
However, the following is certain: If k(i) =f:. k(m), then Ok(j) and
Ok(m) cannot have anything in common; if they did, then the one with
the larger index K == max{k(i), k(m)} would be a proper subset of
the other, violating the choice of OK as the lowest-numbered set in-
tersecting some OJ. Hence Ok(i) and

o- Ok(j) = U Ok(m)
k(m)#(j)

are disjoint open sets. That makes Ok(i) a maximal connected subset
ofO.

6. (a) Ifx E bd(A n B), then there is a sequence (Xj) -+ x from An Band
(y j) -+ x from outside An B . Either a subsequence (y j (i») comes from
outside A, so x E bd(A), or a subsequence (Yk(j») comes from outside
B, and x E bd(B). Hence x E bd(A n B) ~ x E bd(A) U bd(B).
(b) Assume that A and B are Archimedean. Then bd(A) and bd(B) are
meager. By Exercise 2, their union is meager; by (a), bd(A n B) S;
bd(A) U bd(B); by Exercise 3, bd(A n B) is meager; by Theorem I,
A n B is Archimedean.
Solutions to Exercises: Section 4.4 281

7. A and B overlap <=> int(A n B) is nonempty (Exercise 4.1:2e) <=> v*(A n

B) > 0 (Exercise 4.3:4 and definition of v*). Since A and B are Archime-
dean, so is AnB (Exercise 6). Hence v*(AnB) > oamounts to V(AnB) >
O.
8. (a) The non-Archimedean set 0 in Example 2 is the union of Archimed-
ean neighborhoods.
(b) Unit square - 0 is not Archimedean, because its boundary is bd( 0),
which is not meager.
(c) Again, bd(O) is not meager. However, it has no interior, because it
lacks all of Q2: every x E Q2 is in the interior of O.

9. (a) The given set is a bounded subset of a hyperplane. Such a set is nec-
essarily meager; compare Answer 4.3:3b. Therefore, every bounded
function is integrable on it.
(b) It matches Dirichlet's function on the box from (-0.5, -0.5, -0.5)
to (0.5,0.5,0.5), is therefore not integrable on that box. By Theorem
5, g cannot be integrable on S.
(c) This is the restriction of the modified Dirichlet function G, defined
on the box from (-1, -1, -1) to (1, 1, 1), to s. By Example 3 and
Theorem 5, G is integrable on S.

10. The finite set is meager (Exercise 4.3:3a). Therefore, Theorem 4 guarantees
that f is integrable.

11. Assume that f is continuous everywhere and A is Archimedean. Then A is

bounded, so cl(A) is closed and bounded. Consequently, f is bounded on
cl(A), is therefore bounded and continuous on A, is integrable on A.

12. Assume that every bounded function is integrable on S. Then g(x) == 1

is integrable, so by Theorem 4.3:3(c), S is Archimedean. If S were not
meager, then cl(S) would contain some neighborhood, and therefore some
box; on that box, Dirichlet's function F would beunintegrable; by Theorem
5, F would be unintegrable on S; contradiction.

13. The Cantor set is meager, because by its construction, it is contained in the
union of2k intervals oflength 3-k each, total length (~)k, for any k.

14. (a) Assume that S is meager. By Theorem 2, there is a class of boxes

CI, ... ,CJ covering S with volume-sum less than ~. If finite se-
quences are allowed, define (Bi) == (CI, ... ,CJ). Ifnot, set
282 Solutions to Exercises: Chapter 4

where ¥ is a box concentric with C J and t as wide. Then

(b) Unit square n Q2. Call it S, and enumerate it as

{(rl' st), (r2, S2), ... }.

Let Bi be the square centered at (ri' Si) with width 2- i B1/ n. Clearly,
S ~ Bl U B2 U ... , and

V(Bl) + V(B2) + ... ~ ~]TiBl/nr ~ + ~ + ... = i B.

That proves that S has zero measure. But S is not Archimedean, so its
volume is undefined.
(c) The Cantor set is uncountable and meager (Exercise 13 and refer-
ences).
(d) Assume that S has positive involume. Then int(S) is nonempty, so S
contains a neighborhood, and therefore a box B. Suppose now (Bi)
is a sequence of boxes covering S. Then (Bi) also covers B. By the
Heine-Borel theorem, some finite subcollection Bl, ... , BJ still cov-
ers B. By a familiar cross-partition argument, V(Bl)+' .. + V(BJ) :::
V(B). Hence V(Bl)+ V(B2) + ... ::: V(B). We conclude that S can-
not have zero measure. By contraposition, if S has zero measure, then
it has zero involume.
(e) By (d), A has zero measure => v*(A) = O. If A is Archimedean, then
V(A) = v*(A) = O.
(f) The words "closed and bounded" should give us a hint. Assume that S
is closed, bounded, and of zero measure. The last means that there is
a sequence (Bi) of boxes covering S with volume-sum < B. Since
S is compact, the same job can be done by a finite subcollection
Bl,'" ,BJ. Then S ~ Bl U .. ·UBJ with V(Bl)+"'+ V(BJ) < B.
By Theorem 2, S is meager.

15. (a) Assume x E bd(int(S». Then there are sequences (Xi) --+ X from
int(S) and (yj) --+ X from outside int(S). Each Xi E S. If an infinity
of Yi are in bd(S), then X is a closure point of bd(S); that forces x E
bd(S), because boundaries are closed sets. If instead finitely many Yi
are in bd( S), then a subsequence YJ, Yl+ 1, . .. comes from ext( S) ~
S*; that makes x E bd(S). We have shown that x E bd(int(S» => x E
bd(S).
Solutions to Exercises: Section 5.1 283

(b) Set S == {(X, y): x 2 + y2 ~ 1, or (x, y) E Q2 and 1 < x 2 + y2 ~ 2}.

Int(S) is the open unit disk, whose boundary is the unit circle. The
boundary of S is the annulus 1 ~ x 2 + y2 ~ 2.
(c) b E cl(S) <=> every N(b, c5) has points from S <=> (every N(b, c5)
intersects both S and S"') or (some N (b, c5) is all S) <=> (b E bd(S» or
(b E int(S».

Chapter 5
Section 5.1
1. Assume that Tk is defined by au ~ U1 ~ bk1, ... ,akm ~ Um ~ bkm, Sj
by Cj1 ~ VI ~ dj1, .•. ,Cj(n-m) ~ Vn- m ~ dj(n-m).

(a) The details are straightforward, so we provide an outline: Tk x Sj is

a box, given by the conjunction of all n inequalities; two such boxes
do not overlap, because if they did, then two Tk or else two Sj would
overlap; and their union is B. That would prove (a).
(b) The volume of Tk x Sj is [(bk1-ak1)'" (bkm -akm)][(dj1-Cj1)'"
(dj(n-m) - Cj(n-m»] = V(Tk)V(Sj).
(c) [diag(Tk x Sj)]2 = [(bk1 - au)2 + ... + (bkm - akm)2] + [(dj1 -
Cj1)2 + ... + (dj(n-m) - Cj(n_m»2] = diag(Tk)2 + diag(Sj)2.

2. (a) Fix a natural number k. For each fraction fe, 0 ~ j ~ k, paint the
horizontal band fe - -b -b.
~ y ~ fe + On those bands, f(x, y) ~ 1,
and the total area is [(k+1Hk+2)] 2 contributing at most 6 to the
2 F' ~
upper sum. For the unpainted bands, f (x, y) < t,
and the total area
t.
is less than 1, contributing less than Consequently, there are upper
sums = O(t), upper integral = 0 = lower integral.
(b) For fixed y = t
fe in lowest terms, f (x, y) == for rational x, == 0
t;
otherwise, is Dirichlet's function times it is not integrable.

3. (a) For a fixed y, g(x, y) is constantly 0 or constantly 1.

(b) If y is rational, then ~(y) = Jo1 1dx = 1; otherwise, ~(y) =
Jd Odx = O. Thus, ~ is Dirichlet's function.
4. (a) For a fixed rational y, h(x, y) = -1 if 0 ~ x ~ !, = 1 if! < x ~ 1.
This is a function with one discontinuity, is integrable. For a fixed
irrational y, h(x, y) = 1 if 0 ~ x ~ !, = -1 thereafter; likewise
integrable.
284 Solutions to Exercises: Chapter 5

(b) In (a), for every y, Jd h(x, y) dx = 0; <P is constant.

(c) In every subinterval of any partition, there are places where h =
sup h = 1 and places where h = inf h = -1. Hence the upper in-
tegral is 1, the lower integral -1.
5. Assume that f is continuous in B. First, f is integrable on B. Second,

=l
(~·····bn)
<PI (Xl) f(XI, ... ,Xn)d(X2, ... ,xn)
(a2 ... · .an)

is defined, because f(XI, ... ,xn) is a continuous function of (X2, ... ,xn )
for fixed Xl. Third, <PI is continuous, and therefore integrable, because

and If(x) - f(y)1 is uniformly small. Hence Fubini's theorem applies, to

l
give

{ f = bl
[( fXI(X2, ... ,xn)] dx.
l[a.b] aile
That proves the reduction formula, beginning a recursive proof.

6. Set C == Q n [0,1], A == {(x, x): X E C}, g(x) = G(x) == x. We know that

C is not Archimedean. In R2, A is a subset of a line, is therefore meager.
7. By Fubini,

f e"+Y+' = LLLI e"eYe' dz dydx = (LI <ldt)' = (e - e-I)'.

8. (a) Since the area is 1, the average is

1 L+ (x y) = 10 1 10 1(x +y)dydx = 10 1 (x +~) dx = 1.

This makes sense: Draw the line x + y = 1; for every patch of area to
its lower left where x + y = 1 - e, there is a corresponding one to the
upper right with x + y = 1 + e.
(b) Now the area is !, so the average is
{ {I {I-x
2 1A(x+ y )=2 10 10 (x+y)dydx

= 2 10{I ( x[l - x] +
[l - X]2)
2 dx =
2
3'
Solutions to Exercises: Section 5.2 285

This may seem surprising, because it is closer to the maximal x + y =

1 than to the minimal x + y = O. But again it makes sense, because
there is more area near x + y = 1 than near the origin.
9. The region is given by 0 =:: z =:: 3 - x 2 - 3y2 for x 2 + y2 =:: 1. Clearly, the
(x, y) integral is better switched to polar coordinates. Then

10. V = f1dxdydzdwdv = f~afx2+y2+Z2+w292_v21d(x,y,z,w)dv.

The inside integral is the volume V# of a ball of radius J a 2 - v 2 in R4.
# lT2(a2 v2)2
By Example 2, we see that V = 2- . Hence

Section 5.2
1. (a) Let f == 2, g == 1, A == line segment.
(b) Set h(x, y) == 1 if y = 1, == 0 otherwise, on A == unit square. Then
h ? 0,
r = Jint(A)
JA
hr h = 0, V(A) = 1.

(c) Same function h on A- == {(x, y): 0 =:: x =:: 1, either 0 =:: y =:: !
or y = I}. Alternative: A+ == unit square U segment from (-1,0) to
(0,0), with H(x, y) == x - 1xJ.

2. Assume that f ? 0 is continuous on A. Theorem l(c) says that fA f ? O.

Suppose f(b) > 0 for a place b E int(A). There must exist N(b, 8) in
which f(x) > f~b) and N(b, e) 5; A. Let N == N(b, 8) n N(b, e). Then

rf = rf + r f> r f(b) + r 0 = f(b)V(N) > O.

JA IN JA-N IN 2 JA- N 2
Thus, the integral is positive, or else there are no interior b E A with f (b) >
O. (The equivalence part comes from linearity.)

3. V(A U B) = fAUB 1 = fA 1 + fB 1 (Theorem 2(b» = V(A) + V(B).

286 Solutions to Exercises: Chapter 5

4. Assume S S;; T, both Archimedean. By Theorem 4.4:5(a), 1 is integrable

on T iff it is integrable on both Sand T - S. Hence if 1 2: 0 is integrable on
T, then 1 is integrable on Sand T - S, and fT 1 = f T- S 1 + fs 12: fs I·

5. By Theorem 1, k ~ 1 ~ K => V(A) = k fA 1 = fA k ~ fA 1 ~ fA K =

KV(A).

6. By Cauchy's inequality,

If 1 is integrable, then so are III and 12 , so these Riemann sums tend to

The stated inequality follows.

Section 5.3
1. Call the region A. Then (x, y) = <I>(u, v) == (au, bv) transforms the uv-
unit disk onto A. (Clearly one-to-one, with absdet <1>' = labl > 0, since we
take a and b to be positive.) Hence fA 1 = f<l>(D) 1 = fD(1 absdet <1>') =
ab area(D) = nab.

2. The region 0 ~ x ~ 3, 0 ~ y ~ 4{ is given by 0 ~ () ~ tan- 1 (~),

o ~ r ~ c1s(l. Hence

/o tan- (4/3) /0 / cos lr dr d()

1 3 (I
area =
o
tan- (4/3) (9)
0

= /o
1
2
- sec () d()
o 2

= (~) (~) =6,

1 /otan- (4/3) /0 3/ cosO
1
x - average = - r cos () r dr d()
6 0 0

= (tan- (4/3) cos ()(3/cos()3 d() = (~) (~) = 2

1

10 18 2 3 '
Solutions to Exercises: Section 5.3 287

y - average = -
l1
tan - 1 (4/3) 3/ COS ()1r sinOr dr dO
6 0 0
31 tan- 1 (4/3)
=- cos- 3 0 sinO dO
2 0
3
= 4[cos- 2 0]0tan-
1 4/3
= (3) (16)
4
4
9" = 3'

3. The cone is given by 0 ::: <jJ ::: l'

0 ::: 0 ::: 2rr, 0 ::: p ::: if>' The co!
transformation is singular along the z-axis, <jJ = 0, but we proceed as in
Example 1, as though the substitution hypothesis were met.

(a) Volume is

f27r
10 10 10
r/4 f 4/ cos if>
I p 2 sin<jJdpd<jJdO

= 2rr 1
o
7r/4 43
3cos <jJ
3 sin <jJ d<jJ
64rr
= -3- .

(b) Distance from the origin is p, so average distance is

(c) Symmetry suggests that x-average and y-average are O. The spherical
integral agrees, because x = p(sin<jJ) cosO and y = p(sin<jJ)sinO
make f;7r cos 0 and f;7r sin 0 factors. The z-average is

3 127r 17r/414/COSif>
- (pcos<jJ)p2sin<jJdpd<jJdO
64rr 0 0 0
4 r/
= 610 cos- 3 <jJsinqJd<jJ = 3.

4. The Jacobian a(r,~~t,u) is r(r sins)(r sins sint) (= product of the lengths of
the columns; consult Answer 4.4:4). We want the region r ::: a. Its volume
is

r r
10 10 10 10
f27r r 1(r 3 sin2 s sin t dr du dt ds

2rra 4 [s sin2s]7r rr 2a4

= --[-cost]~ - - - - = - - .
4 2 4 0 2
288 Solutions to Exercises: Chapter 5

5. From </J' (x) #- 0, we conclude (intermediate value theorem) that </J' is of one
sign. Suppose first </J' > O. Then </J is a strictly increasing function, mapping
[a, b] one-to-one onto [c, d]. The hypothesis of Theorem 5 is satisfied, so
we conclude that

{ f(u)du ={ f(</J(x))i</J'(x)ldx = {b f(</J(x»</J'(x)dx.

J[c,d] J[a,b] Ja

Suppose instead that </J' < O. Then </J maps [a, b] decreasingly onto [d, c].
By Theorem 5,

{ f(u)du = ( f(</J(x»I</J'(x)ldx.
~~] J~~l
The definition of
we have again
f: f(u) du is - J[d,c] f(u) du, and WI = -</J'. Hence

i d
f(u) du = lb f(</J(x»</J' (x) dx;

the change of sign required by the absolute-value sign is accomplished by

the order on R.
6. (a) Theorem 3.2:2 says that cI> maps some open set one-to-one onto some
neighborhood N(cI>(a), 8). Theorem 3.2:3 says that cI>-1 is differen-
tiable. Any continuous function with a continuous inverse, from one
open set to another, will map open sets to open sets: P open => image
of P = cI>(P) = (cI>-1 )-1 (P) = inverse image of open set, and the
last is necessarily open [Guzman, Section 4.4].
(b) Assume J(x) #- 0 and P ~ 0 is open. Let b E P. By Theorem
3.2:2, cI> maps some open set Q invertibly onto some neighborhood
N(cI>(b), 8), some N(b, 8) ~ Q n P. By (a), cI>(N(b, 8» is an open
subset of cI>(P). Thus, for each b E P, there exists an open set
cI>(N(b, 8» with cI>(b) E cI>(N(b, 8» ~ cI>(P); this proves that cI>(P)
is open.
(c) Under these hypotheses, cI> maps the open set int(A) into an open
subset of cI>(A), and 0 - cl(A) = 0 n ext(A) into an open subset of
cI>(O) - cI>(A). Ifx E bd(A), then neighborhoods of cI>(x) are images
of open sets surrounding x, open sets that must intersect A and A *;
hence cI>(x) E bd(cI>(A». This means that int(cI>(A» can get images
only from int(A), so cI>(int(A» has to be all of int(cI>(A», similarly
for the boundary and the remainder.
(d) Set A == {(r, 8): 1.1 ~ r ~ 1.9, m~ 8 ~ ~}. In the mapping
(r, 8) 1-+ (r, 28), the point at (1.5, 2ft;) is on the boundary of A, but
its image
Solutions to Exercises: Section 5.4 289

(1.5, 5~:;) = (1.5, 1~:;)

is interior to the image of A.
7. Assume that J (x) is continuous near a. Then

V(<I>(A») = r
J4>(A)
1= r
JA
J(x) = J(x*)V(A)

for some x* in A. Hence V~~i1» = J (x*) ~ J (a) as long as A S;;; N (a, cS)
with cS ~ O.

Section 5.4
1. (a) Suppose I/(x)1 =:: Kllxll-P for x outside the radius-R ball. Then for
any A out there,

IJr II
A
=:: frill =:: f
k=RA:SIlXII!Sk+l k=R
Mk-P[(k+l)n-kn]V(B(O, 1)).

The series converges, because (k + 1)n - k n = O(kn-l). Hence

{fA I} is bounded, and I is integrable on B(O, R)*.
(b)

12 (I+Jx2+y2)-P = fo21r fo oo (1+r)-P r drdfJ

(
u2-p u1-p ]00 21r
= 21r 2 - p - 1- P 1 = (p - 2)(p - 1) .

(c) Since (1 + IIxll)-P ~ (1 + IIxll)-n, it suffices to do the latter.

r
Jllxll!SR
(1 + IIxll)-n = Lr
R

k=l Jk-l!SlIxll!Sk
(1 + IIxll)-n
R
~ Lk-n[kn - (k - 1)n]V(B(O, 1»).
k=l

Since (k + 1)n - k n = nt n- 1 (by the mean value theorem) ~ n(~r-l

beginning with k = 3, the series looks like Lk~l 2n~ik' which di-
verges. Hence the integrals are unbounded.
2. (a) Suppose A S;;; unit ball is closed and does not include the origin. Then

Ii II =:: t.
00
il(k+l)!s IIxll!S 11k

[1
III
1]
=:: {;M(k+ l)q k n - (k+ l)n V(B(O,I»).
290 Solutions to Exercises: Chapter 5

Since (k + l)q ~ (2k)q and k~ - (k~l)n = _nt- n- 1(-1) ~ nk- n- 1,

the terms in the series are dominated by multiples of k q - n - 1• The
series converges, and the integrals of f are bounded.

(b) f D(x 2 + y2 )-q/2 = JOr2rr JOr1 r -q r d r dO = 2rr [r-q+2]1 2rr

-q+2 0 = 2-q'
(c) Let D be the disk, V its volume. For any K,

[
JD f ~ tr (1)k
K -q [ 1 +1]
k n - (k l)n V.

In the series, the terms exceed k qn(2k)-n-1 ~ constantlk; the series

diverges.

3. (a) No in both. Between the lines y = x and y = 2x, (X2:~2)2 ~ 2~:4 is

too big at 0 and at 00. Thus,

I 218tan- 1 r cos Or sin 0

4 rdrdO =K 18 dr
-
rr/4 8 r 8 r

diverges with either 8 -+ 00 or 0 -+ O.

(b) Yes in D, not D*. Near the origin, s~;r ~ 1, no problem. Between
the hyperbolas y = (2k+;/6)rr and y = (2k+;/6)rr, there is infinite area
and si~;r > (2k11)rr.

(c) Yes and yes. It is bounded in the circle, and is O(r-4) at infinity.
(d) No in D, yes in D*. We have x4 + y4 ~ (x 2 + y2)2. That makes
(x4 + y4f1 ~ II (x, y)II- 4, much too big near (0, 0). At infinity,

4.

1[0.rrjU ... U[2krr.(2k+1)rrj

sin-
-
x
x > -1 !orr sinxdx
rr 0
+ ... +
(2k
1
+ l)rr
!orr
0
sinxdx

= (1 + ! + ... + 1+ )
3 (2k 1)
~
rr'

which tends to infinity as k -+ 00.

Solutions to Exercises: Section 5.4 291

5. The arguments in the proof of Theorem 4.4:3 and Answer 4.4:6a, showing
that the boundaries of S U T, S - T, and S n T are all subsets of bd(S) U
bd(T), apply to all sets. Also, bd(S*) = bd(S). Suppose Sand T are LA.
Then bd(S) and bd(T) are locally meager, so that

[bd(S) U bd(T)] n N = [bd(S) n N] U [bd(T) n N]

is meager for every neighborhood N. It follows that union, intersection,
difference, and complement all have locally meager boundaries.

6. ::::} Assume that A is Archimedean. Then N is a neighborhood::::} AnN is

Archimedean (Theorem 4.4:3); this says that A is LA. Boundedness is part
of the definition.
<= Assume that S is bounded and LA. Bounded means that S 5; some
B(O, R). Because S is LA, S = S n N(O, 2R) is Archimedean.
7. (c) First, S U T and S n T are LA (Exercise 5). By extension of part (b),
I is integrable on S U T <=> I is integrable on S - T, S nT, and
T - S <=> I is integrable on Sand T; and

If Sand T do not overlap, then int(S n T) is empty, and (a) says that
fsnT I = O.
(d) Assume I ~ 0 on Sand T 5; S. Then for every A 5; T, fA I ~ 0
(Theorem 5.2:1). Therefore the infand sup of {fA I} are nonnegative,
making

i I == sup {i I} + {i I} ~ o.
inf

Similarly, fS-T I ~ O. Then by (c), Is I = IT I + IS-T I ~ IT I·

8. On an Archimedean A 5; S, IfAgl .::: fA Igl (Theorem 5.2:1(d» .::: fA I
(Theorem 5.2:1(c» .::: fs I (Theorem 5.4:2(d), because I ~ Igl ~ 0).
Hence Itg(g, S) is a bounded set.

9. No. By Exercise 2, (x 2 + y2fl/2 is integrable on the unit disk, but its

square is not.

10. (a) If I ~ 0, then 1+ = I and 1- = 0, so 1= 1+ - 1- = 1+ + 1- =

III· If instead I < 0, then 1+ = 0, 1- = -I, and I = 1+ - 1-,
III = - 1= 1+ + 1-·
292 Solutions to Exercises: Chapter 5

(b) :::} Assume that I is locally integrable on S, meaning that I is in-

tegrable on every closed A £ S - some U. Let A be one such
set. Necessarily III is integrable on A (Theorem 5.2:1(d)). Hence
1+ = J1Jt1.
and 1- = 1¥
are integrable on A (Theorem 5.2: l(a)).
This proves that 1+ and 1- are locally integrable.
{=: Assume that 1+ and 1- are locally integrable, meaning that 1+
is integrable on every closed A £ S - some U, 1- is integrable
on every closed A £ S - some V. Suppose now that A is a closed
subset of S - (U U V). Then 1+ and 1- are integrable on A, so that
1= 1+ - 1- is integrable on A (Theorem 5.2:1(a)).
(c) I is integrable on S <=} I is integrable on T+, TO (automatic), and
T- <=} 1+ is integrable on T+ and - I is integrable on T- <=} 1+ is
integrable on S (because 1+ is automatically integrable on S - T+,
where 1+ = 0) and 1- is integrable on S (similarly). If true, then

{ I = { 1+ { 1+ { I = { 1+0+ { I = { 1+ - { 1-·
Js JT+ JTO JT- JT+ JT- Js Js
(d) Dirichlet's function has T+ = Qn, not LA.

Section 5.5

1. J~ (~)2 + ... +(~fdt=J(bl-al)2+ ... +(bn-an)2.

2. (a) get) == (21t - ~ 1,0) has two linear pieces of length = 1. For an
example whose endpoints are (0,0) and (1,0), set h(t) == (2t, 0) if
o ~ t ~ == i, G- i i,
2t, 0) if ~ t ~ == (2[t - ~], O)iq ~ t ~ 1;
its three pieces have lengths i, i, i·
(b) get) == (tlsin(f,)I,O) if 0 < t ~ 1, g(O) == (0,0). Adapt Example
l(a) for the argument that len (g) = 00.
(c) There must exist a and b with g(a) = (0,0), g(b) = (1,0). Then the
partition
T = {O, min{a, b}, max{a, b}, I}
gives len(g, T) ::: IIg(a) - g(b) II = 1, forcing len (g) ::: 1.
3. get) == (t, h(t)), a ~ t ~ b, is a smooth parametrization. By Theorem 3(d),

len(g) = lba (dx)2

dt + (d)2
; dt = lb
a J1 + h'(t)2 dt.

4. (a) x = r cos () = cor, y = si~e leads to

{M (dX)2 y (M ~
J2rr d() + (dd() )2 d() = J2rr V1 + (j2 d() > M - 2rr -+ 00;
Solutions to Exercises: Section 5.5 293

the length is not bounded, and the spiral is unrectifiable.

(b) x = e-e cosO, y = e-e sinO leads to

as M ~ 00. The spiral is rectifiable.

5. (a) It is simply the fundamental theorem. Write the distance as

L(t) == len (g) = fot ! g~ (u)2 + ... + g~(u)2 duo

Then

(b) By (a), L'(t)2 = g'(t) • g'(t). Derivatives of those quantities are

2L' (t)L" (t) = 2g' (t) • g" (t) (dot-product rule). If speed, meaning
L', is constant, then 0 = g'(t). g"(t), as desired.

6. (a) Let g = (gl, ... ,gn)' Then for each k and any partition {to < ... <
tJ },

J J
L Igk(tj) - gk(tj-I)I ::: L IIg(tj) - g(tj_I)1I
j=1 j=1
N J
::: L L Igk(tj) - gk(tj-I)I.
k=lj=1

This shows that the middle sum is bounded iff each of the individual
sums on the left is bounded.
(b) Clearly,

J J
L Ih(tj) - h(tj_I)1 ::: K L Itj - tj-lI = K(b - a).
j=1 j=1

If h is differentiable and h' is bounded, then h is Lipschitz, owing to

the mean value theorem.
(c) H'(x) = 3x l / 2 sin(1/x)/2-x- I / 2 cos(1/x) has values near ±oo near
x = O. Let (Uj) be the sequence of extremes (critical points) of H,
in the intervals (i+\)1r ::: Uj ::: j~ (at whose ends H = 0). Given
T == {O = to < ... < tJ = I}, assume Uk+1 ::: tl < Uk. and
294 Solutions to Exercises: Chapter 5

let U = T U {UI, ... ,Uk}. Then Var(H, U), in reverse order, can be
broken up into

[H(1) - H(UI)] + [H(U2) - H(ut}] + ... + IH(Uk) - H(tI)1

+ IH(t,ll ,; + 2 (2 ~ IH(Uill) + I.

Since IH(uj)1 :::: u~/2 :::: L~ ]3/2, the series converges, and H E BV.
(d) First, some motivation: part (c) shows that x 3 / 2 sin(~) has bounded
i
variation, and clearly what is important there about is that it exceeds
1. We have also seen that x sin(~) has unbounded variation. The latter
does not fit this question, because it is not differentiable at x = O. It
appears that we need something intermediate between x I and all the
powers x He. Such a function is l:x' (Compare Answer 2.2:7f.) Ac-
cordingly, set G(x) == x~~;~~». It is differentiable on [0,1], because
G(x)-G(O) = sin(1/x) -+ 0 as x -+ 0 Let
x In(x/2) •

T== { 0 < -1- < 1 <,,·<-1-< }

1.
21 rr (21 - 0.5)rr 0.5rr

Then
4 4
Var( G, T) = + -:-:-:-~-::--:~--:'~
(41 - 1)rr[ln(41 - l)rr] (41 - 3)rr[ln(41 - 3)rr]

+". + rr In[rr] +
2 (Sin 1 2)
In2 - rr In[rr]

is unbounded, because [rr Inrrr I +[3rr In 3rrrI+ .. diverges. Hence

G ¢ BV.
(e) If I is increasing, then

for every partition. If I is decreasing, then similarly VarU, T) =

I(a) - I(b).
(f) sgn(x) == ~ if x =F 0, == 0 at 0, has variation = 2 on [-1, 1].
(g) See part (d).

7. (a) Yes. Set I(t) == t 2 sin( t), 1(0) == O. At 0, I has zero derivative;
elsewhere, I' (t) = 2t sin( t) - t)
cos ( is bounded. By Exercise 6b,
IE BV. By Exercise 6a, g(t) = (t, I(t)) is rectifiable.
(b) Yes; g(t) == t 3/ 2 sinn) is of bounded variation, by Exercise 6c.
Solutions to Exercises: Section 5.5 295

8. (a) The argument is identical to that in Theorem 1, except that h- 1(g)

is necessarily strictly decreasing. Therefore, c = h- 1(g(tJ)) < ... <
h- 1(g(tl)) < h- 1(g(to)) = disthepartition,anduj == h- 1(g(tJ_j)),
j =0, ... ,J.
(b) By the argument in Theorem 1, h- I (g) is an increasing map of [a, b)
to [c, d). Hence if a < t < b, then there is a unique t* E (c, d) with
h(t*) = g(t). Thus, g on [a, t] and h on [c, t*] are one-to-one, with
same start, same end. By Theorem 1, len(g[a, t]) = len(h[c, t*]). Let
t -+ b. Then t* = h-I(g(t)) -+ d. By Theorem 2(c), len(g[a, t]) -+
len(g), len(h[c, t*]) -+ len(h); hence len (g) = len(h).

9. (a)

x = len~C) fc x ds
1
= 2rr(a 2 +b)
2 1/2
121T
0
acos (a
t 2 + b2 )
1/2
dt = 0,
which makes sense, given the symmetry. Similarly, Y = 0.

which also makes sense, because half the helix is below z = brr, half
above.
(b) The mass center should be close to (a, 0, 1), where the helix is dens-
est.
296 Solutions to Exercises: Chapter 5

10. (a) With x = gl COSg2, Y = gl sing2,

= II (gi cos g2 -
= (gi)2 + (glg~)2 .
glg~ sin g2, gi sin g2 - glg~ cos g2) r
(b) For this fonn, 0 is the parameter, and gl (0) == g(O), g2(0) == O. Hence
2
(:;) = gi (0)2 + gl (0)2g~(0)2 = g'(0)2 + g(0)2.
11. The needed arguments are in:
(a) Answer 4.2:4a.
(b) Answer 4.2:4c, with Theorem 3(b).
(c) Answer 4.2:4b.
(d) Answer 4.2:4e.
(e) Theorem 5.2:1(d).
(t) Theorem 5.2:2(d).
(g) Theorem 5.2:3(b).
12. (a) The segment is given by x = 2t, Y = t, 0 :::: t :::: 1. Then
{ x 2 + l ds = {I 5t 2(22 + 12)1/2 dt = 5..j5,
1c 10 3
{ x 2 + l dx = (I 5t 22t dt = 10,
1c 10 3

L+ x2 ldy = 10 1 5t 21dt = 5.
(b) Same segment, opposite sense: x = 2 - 2t, Y = 1 - t, 0 :::: t :::: 1.
The integrand and ds stay the same, dx = -2dt, dy = -dt. Hence
the x-integral and y-integral change sign.
(c) x = 2 cos 0, y = 2 sin 0, 0 :::: 0 :::: rr. The integrand x 2 + y2 = 4 is
constant. Therefore,

L +l x2 ds = 4(length) = 8rr,
L+ x2 l dx = 4(x(rr) - x(O)) = -16,

L+ x2 y2 dy = 4(y(rr) - y(O)) = O.
Solutions to Exercises: Section 5.6 297

(d) Same equations as (c),

stant. Hence
°: : () : : 21l', and again the integrand is con-

1 x 2 + ids = 4(length) = 161l',

the others both zero.

13. (a) is an immediate consequence of operator boundedness and -III <

I ::: III·
(b) See the x-integrals in Exercises 12b, c.
(c) Set I(x, y) :: y, and let C be the broken line from (-1, -1) to (0, 0)
to (-2, 1). Then

1 1
C
I dx = 0
-1
x dx + 10-2 .::: dx =
0 2
=-,
2
3

and
Ief dx
x(end)-x(start) = 1:3 .
IS
I
not a va ue 0
f I .

Section 5.6
I. We are looking at z = x 2 + y2, over the disk D given by x 2 + y2 ::: 2. The
area is

a = Iv 1+ (:~r + (!;r = Iv JI+4x 2 +4y 2

r
= 10 21f 10../2 J1 + 4r2 r dr d()

~ 2" [ (1+ ~2)3/2 ~ 1~"

2. a(x;i!'z) = [cos () sin () 2r]t, a(x<ir) = [-rsin () r cos () 0]1. They are ev-
idently orthogonal, so area (~, ~) is the product (1 + 4r2) 1/2 r of their
lengths. Then

10r 10r../2 J 1 + 4r2 r dr d(),

21f
area(f) =

same as (1).

3. The right triangle from (0, 0, 0) to (0,0, z) to (x, y, z) has bottom angle

CIt::cos
-1 z = cos -1 -Z
Jx 2 + y2 +z2 a
Hence sec CIt = ~.
298 Solutions to Exercises: Chapter 5

(a) 1+(~~)2+(~~)2 = 1+(7)2+(=f)2 = z2+~~+y2 = ~.Forz > 0,

the square root is ~.
(b) V F = (2x, 2y, 2z), IIV FII2 = 4 (x 2 + y2 + z2) = 4a 2, V F.k = 2z.
Hence I~~~~~I = ~.
(c) Call the stated angle fJ. It is determined by

Reading ~; and ~~ from Example 1, we have

8 8 II i a cos () cos fjJ -a sin () sin fjJ

~ x ~ = j a sin () cos fjJ a cos () sin fjJ
11
8fjJ 8() k -a sin fjJ °
and (g; x ~~) • k = a 2 sin fjJ cos fjJ (the cofactor of k). Assuming
sin fjJ '# 0, meaning off the z-axis, the determining equation gives
cos fJ = cos fjJ. Hence sec a matches sec fJ.

4. (a) This is the area of the part of the plane x = z within the cylinder
x 2 + y2 ::::: a 2. The plane makes a 45° angle with the xy-plane, so
sec a = "fl. Hence

f sec a dx dy = J2(area of circle) = rra 2 J2.

(b) IIVGII = 11(1,0, -1)11 ="fl, IVG. kl = 1, so

f IIVGII d d -
IVG • kl x y - rra
2J2
.

5. Our current definition is area (g) = fA det[ ~ • ~ ]dt1 ... dtn . By Theo-
rem 5.3:1,

8g . -8g ] = absdet [ -8g ... -8g ] = absolute JacobIan

det [ - . of g.
8tj 8tk 8t1 8tn

Therefore, by Theorem 5.3:5: (a) area (g) = fA 1Jacobian 1 = area(g(A));

Solutions to Exercises: Section 5.6 299

(b)

r
Jg(A)
fda == r
JA
f(g(t)) det [~ • aa g ]
atj tk

= r f(g(t))IJacobianl = r
JA Jg(A)
f(u)du.

The one-to-one condition is always needed. Applying to Example

3.2:1, Answer 3.2:1 gives Jacobian = 2. Hence

3rr/2/2 9rr
area(h) == 1a0 1
2r drdO = -.
2
The range of h is the ring I ::: r ::: 2, whose area is 3rr. Thus, com-
pared to the area of the range, area (h) is 1.5 times too high, because
the parametrization h is "one-and-a-half-to-one."

6. The end on the left is a disk of radius f(a), area = rrf(a)2; similarly,
rrf(b)2 on the right. The surface may be described by f(x) = distance from
J
x-axis = y 2 + z2,leading to x = x, y = f(x) cosf3, z = f(x) sinf3, for
a ::: x ::: b, 0 ::: f3 ::: 2rr. Then

a x~
~ a = [ f'(x)cosf3
1 ] x [ -f(x)sinf3
0 ] = [f'(X)f(X)
-f(x)cosf3
]
,
ax af3 f'(x)sinf3 f(x)cosf3 -f(x)sinf3

so

f area ( ::' :;)

= 1a 2rr
lb JU'(x)f(x)]2 + f(x)2 cos2 f3 + f(x)2 sin2 f3 dx df3

= lb 2rrf(x)J 1 + f'(x)2 dx.

This integral matches f: 2rrf(x) ds along the curve; see Answer 5.5:3.

7. (a) Imagine painting the outside of the sphere with a layer of paint dr
thick. Then the volume dV added to the ball is area x thickness =
4rr r2 dr. Thus, ~~ = area.

(b) In R4, V = ~ (Example 5.1:2) and ~~ = 2rr2r 3 = area (Example

3 here).
300 Solutions to Exercises: Chapter 6

(c) In R5, the volume of the ball is 8~Is (Answer 5.1:10). This sug-
gests an area of 81T;r4. You should check that the usual parametriza-
tion (consult Answer 4.4:4) leads to

(1 = 10r 10r 10r 10{21T r 4 sin3 s sin2 t sin u dv du dt ds

1T
4 [ cos3 s ] t sin 2t 1T 1T 8rr 2r 4
= 2rrr -coss + -3- [-2 - --]
4 0
[-cosu]
0
= --.
3
o
8. In defining the cross product, we wrote V2 x ... X Vn = CI el + ... + Cnen ,
where the Cj are cofactors of the first column in a formal determinant.
Therefore, VI = Vll el + ... + Vlnen gives VI • (V2 x ... x vn) = Vll CI +
... + VlnCn = det[vi ... vnl. The last has, according to Theorem 5.3:2,
absolute value matching the volume of the parallelepiped of VI, ... , Vn.

Chapter 6
Section 6.1
1. Theorem 2 makes clear that the field does no work if the object moves
along a circle centered at the origin, because then the field, pointing along
the radius, is perpendicular to the path. Further, the symmetry of the field
implies that the work from one such circle to another is the same along
any radial path. Accordingly, you ,et the same work over the path from p
radially to the circle of q (at point mp), then counterclockwise Of. degrees
to q, whether you go directly (0 ~ Of. < 360°) or circuitously (Of. > 360°).

2. Go from p vertically to the x-axis, along the x-axis to the vertical line of
q, vertically to q. The work is zero, because G is perpendicular to the two
verticals and G == 0 on the axis, no matter how you actually travel along
the three parts.

3. It looks like a whirlpool, circulating counterclockwise with speed increas-

ing directly with distance from the origin.

(a) Use the definition: HI = -y, H2 = x; x = a + t(e - a), y =

b+t(d -b) =* dx = (e-a)dt,dy = (d -b)dt;

f Hldx+H2 dy= fol-(b+td-tb)(e-a)

+ (a + te - ta)(d - b) dt = ad - be.

(b) Use Theorem 2: On the straight part, H = Oi + xj, T = i, work =

0; on the circle, H = -yi + xj, T = unit perpendicular to (x, y) =
Solutions to Exercises: Section 6.1 301

f H. T ds = f (y2 +X 2)
4 = 4(length) = 8rr.

(c) As in (b), I H. T ds = 4(circumference) = 32rr.

4. By Theorem 1,

f F. ds = lab F(g(t)) • g' (t) dt = lab F(g(t)) • Tllg' (t) II dt,

and by Theorem 5.5:3(d), the last is Ie F • T ds.

5. (a) The field is tangential, so the flux should be zero. In detail: The normal
is along the radius, N = II~~:;~II = (x/). Hence H • N = (-YXa+XY) =
0, and flux = O.

(b) Think of H as air flow. For the part of the region (shaded in the fig-
ure below) near the origin, the air circulates slowly within the region,
producing no flux. Further outward, the air enters (negative outward
flux) from lower right, leaves (positive flux) at the same rate to upper
left. Furthest outward, the same thing happens at higher speed. The
net flux is zero.
302 Solutions to Exercises: Chapter 6

6. (a) The upward normal is along the radius, N = (x,!,z). Hence

{VeNda = {vke (x,y,z) da
is is a

- (!:)
a is
{ z ds -!:a ix2+y2~a2
{ z 1 + x2
z2
+ y2 dx dy
z2

= !: ( a dx dy = ll'a 2v.
a idisk
(b) We have a uniform upward flow. To escape through the hemisphere,
air must enter the upper half of the ball through the disk, to which the
flow is perpendicular.
7. The symmetry allows us to find the flux at the face z = a, then multiply by
6. For z = a, E = Kq (x2+y<x/,a;)3/2'
+a
N = k, 1 + an
2+ (~yZ)2 = 1. Hence

f E eNds =f
afa
-a -a (x2
Kqa
+ y2 + a2)
3 2 dx dy.
1

For this integral, we are best off if we let 0 ::: () ::: f, 0 ::: r < co~ 0 '
evaluate, and multiply by 8:
7r/41a1Coso Kqa
1 --"----::-;:- r dr d()
o 0 (r2 + a 2)3 /2

= Kqa r
io
l4
(~ _
a
1
Ja 2/cos 2 ()+a 2
) d()

= Kq(~- 107r14 J2~ss~n2()d())

1l' . -1 Sin(1l'/4») Kqll'
= Kq ( '4 - SID .fi = 12'
Multiply by 8 and 6 to get 41l' K q.
8. (a) Consult the long discussion below Theorem 5.6:2. Suppose the para-
metrization is Xn = f(Xl, ... ,Xn-l). Then

n=±(~,
aXl
... '~'-1).
aXn-l

Hence the upward unit normal is N = (-if, ... ,- a:L, 1) Illnll,

da = Ilnlldxl ... dXn-l, and
Is VeNds = i V(Xl, ... Xn-l,f(Xl, ... ,xn-t})
e (-aat , ... '-a at
Xl Xn-l
,1) dXl" ·dXn-l.
Solutions to Exercises: Section 6.1 303

(b) Suppose the parametrization is F(x) = 0, and :In

= V F • en > O.
Then F increases in the positive xn-direction. By (a), the flux to that
direction is

aXn axn)
• ( - -..... - - - , 1 dXl'" dXn-l
aXI aXn-1
r
= JA V •
(aFjaXI aFjaXn-l)
aFjaxn •···• aFjaxn.1 dXl··· d Xn-l

= 1v. a
A
VF
Fjaxn
dXl" ·dXn-l.

with :In :In

= 1 I· If :In < O. then F increases toward the negative
Xn -direction, and

flux = 1v. (-a , .... --.

A
axn aXn
aXn-1
)
-1 dXl" ·dXn-1

= 1v.
Xl
VF
dXl .. ·dXn-l.
A -aFjaxn

and - oXn
of = of 1•
1 oXn

9. (a) We know that L(t) = f~ 11g'(u)lIdu. L'(t) = IIg'(t)1I is just the fun-
damental theorem, and 1Ig'(t)1I > 0 because g is smooth.
(b) (c) First, (a) shows that L -1 exists. Write t = L -l(s). By the inverse
function theorem, L -1 is differentiable, with dL~~(S) = L'~t). Hence
G = g(L -1) is a differentiable composite, with

answering (c). Clearly, G'(s) # 0, so G is smooth. Also, G(O) =

g(L -1(0)) = g(a), G(len(C)) = g(b), and the range vector g(t) is
G(L(t)); that takes care of (b).
(d) Length of Gis fcien(c) IIG'(s)II ds. Make the substitution s = L(t).
Then

fo1en(c) IIG' (s) II ds = lb ~~:;

II II L' (t)dt

= lb IIg'(t)II dt = len(C).
304 Solutions to Exercises: Chapter 6

(e) By parts (c) and (a), !!J!- = ~rJ = II ~:~II; the last is T.
10. (a) The definition is K(s) = IIG"(s)lI. In terms of t, G'(s) = II~~:~II
(Exercise ge). We calculate the s-derivative of G'(s) by the chain and
quotient rules:

G"(s) == dG'(s} = dG'(s}/dt

ds ds/dt
= IIg'(t}lIg"(t} - g'(t}dllg'(t}lI/dt/lIg'(t) II.
IIg'(t}112

That establishes the desired denominator. From IIg'(t)1I = (I' .g'}I/2,

we get

dllg'(t}1I = !(g' • g'}-1/22(g • g') = (g'. g") .

dt 2 IIg'lI

The two-term numerator becomes 111'111" - ~, whose dot prod-

uct with itself works out to

This completes the proof.

(b) Recall that II g' x g" II is the area of their parallelogram, so

..I"
11& x g II = area(&..I '&..I' ) = det [g'
g".• g'
I' g"g'.• g"]
1"

= IIg(t)112 II g'(t)112 - [g(t). g'(t)]2.

Fit this into part (a).

(c) On the line, g"(t}= 0, so K = 0 by part (a).
(d) Yes. 0 = K(s} = IIG"(s}1I forces G'(s} = constant row = [Cl·· ·cn ].
Then mean value theorem gives

G(u} - G(O} = [C! ... cnHu - 0),

putting G(u} on the line through G(O} parallel to (Cl, ... ,cn).
(e) Write
(8 1/2
s(O) = 10 (a 2 sin2 u+a 2 cos 2 u) du=aO,

so that
G(s} = (x,y) = (acos[~] ,asin[~D.
Solutions to Exercises: Section 6.2 305

Then G"(s) = (-cos[~], - sin[~])la, and IIG"(s)1I = ~. If we

switch to cosine and sine of et , then we get simply s(t) = a(et - 1),

(x, y) = (a cos [ 1 + ~] ,asin [ 1 + ~]) .

Now G"(s) = (- cos[1 + ~], - sin[1 + ~])Ia, same norm. The cur-
vature is independent of parametrization; it is intrinsic to the curve.
(t) The principle sounds right, but the helix is a counterexample: s =
(a 2 + b2 ) 1/2 t (adapt Example 5.5:4(a», making

(x,y,z) = (acoss,asin s , !J~ ),

,.;a 2 + b2 ,.;a 2 + b2 ,.;a 2 + b2
and II G" (s ) II = a2~b2; constant curvature, but not a plane curve.
(g) This time, we may as well use part (a) on g(x) = (x, J(x». We find
that g = (1, J'), g' = (0, 1"), and

",(1 + /,2)(/,,2) - (f' /,,)2 11"1

K = (1 + /,2)3/2 = (1 + /,2)3/2 .
, -x , , _ a2
(h) Here y = = Part (g) gives K =
•
(a L x2)lf2' Y (a L x 2)3/2' 1 as
a'
expected.
(i) We have T = ~~~~g = ~. Within part (a), we obtained
dT _ dG' _ IIglIg' - g(g e g')/lIgli _ L _ g
ds - ds - 1Ig'1I 3 - 1Ig'1I2 a.
Therefore,

dT g (g" ) gxg"
T x ds = IIg'il x 1Ig'1I2 - ag = 1iiij3'
so that liT x ¥S I = II~;~/II, which by (b) is K.
We also characterized the norm of a cross product as the area of its paral-
lelogram: liT ¥S
x I = area (T, ¥S).
At the beginning of Section 5.6 we
matched that area with liT II 112' I sin (), where () is the angle between the
vectors. The component of ¥S
perpendicular to T is precisely I II sin (). ¥S
Section 6.2
1. On the broken line from (0,0) to (a, 0) to (a, b),

f GeTds= foaOieidx+ fob yiejdy=O,

and clearly G =F VO.

306 Solutions to Exercises: Chapter 6

(b) Assume a > 0 and b ~ 0, and write c 2 = a 2 + b 2. Go from (0, 0)

straight to (c, 0), then counterclockwise around the radius-c circle to
(a, b). Then

f rc {tan-
Heds= 10 (Oi+xj)eidx+ 10
1(b/a)

(-y,x)e
(-y x)
c' ds

= c(length of arc) = (a 2 + b2 ) tan-I (~) .

But

V ([x 2 + i]tan- I [~D

= (-y + 2xtan- 1 (~), x + 2ytan- 1 (~)) # H.
2. (a) GI=y,G2=O,andW=I#0=W.
(b) HI = -y, H2 = x, and ~ = -1 # 1 = ~.
3. Yes. Can you tell by inspection: 2xyi + x 2j = V (x 2y)? If not, integrate
from (0, 0) to (x, 0) to (x, y):

f Feds= fox x 2jeidt+ foY (2xti+x2j)ejdt=x2y.

4. and 5. Travel from a == (al, ... ,an) radially to the circle of radius Ilbl!,
then around to b. On the circle, the sense is immaterial, since the field is
orthogonal to the tangent. On the radial segment from a to 1I1~\i' we have
F = f(r)x, T = ~,andds = dr. Hence JF e Tds = ~II~~III f(r)rdr. This
proves (5), which implies (4).
6. IfF = fl(xI)el + ... + fn(xn)e n, then
fc F e ds = fc fl (xddxI + ... + fn(xn)dxn
is, by Theorem 5(b), the sum of PI integrals.

Section 6.3
1. By Green's theorem,

1c
-y dx = 1(- 80
-- - = area(A),
8(-y»)

i
A 8x 8y

fc x dy = 1 = area(A),

fc xdy - ydx =i 2 = 2 area(A).

Solutions to Exercises: Section 6.3 307

2. (a) At each point (x, y), -yi+xj is normal to the radial segment from the
origin. Therefore, ( - yi+ xj) eds is the component of ds perpendicular
to that radius. This component determines the rate at which the line
from 0 to (x, y) sweeps out area as (x, y) moves. In the figure for
this exercise, the shaded triangle has area roughly

1 ( -yi + x j ) 1
da+ = 2"1I(x,Y)1I dse II-yi+xjll = 2"(-ydx+xdy).

The integral of this quantity is the area of the region from the origin to
the remote part of A minus the area of the region from 0 to the near
part of A. Along this near part, (-yi + xj) e ds is negative, and the
integration subtracts the area of the part of the shaded triangle located
below the letter "C."

ds

~-----....,.x

Exercise 2a.

(b) Ifthecurvehasr = J(O), then it is parametrized by x = J(O) cosO,

y = J (0) sin 0, 0 ::: 0 ::: 21l'. The relation becomes

area(A) =! { xdy - ydx

21c
1 {21f
= 2" 10 J(O)cosO[f'(O) sinO + J(O) cosO]
- J(O) sinO[f'(O) cosO - J(O)sinO]dO
= (21f J(0)2 dO.
10 2

3. The counterclockwise sense in R2 around x 2 + y2 = 4 goes with the normal

in the positive z-direction; we use counterclockwise tangent and upward
normal. Along the edge, F = -yi + xj + Ok is tangent to the circle, so

L F e T ds = 1IFiliength = 2(21l'2) = 81l'.

308 Solutions to Exercises: Chapter 6

On the surface,
• il _y
~ ill
VxF= J - x =2k
k eZ In(1 Z + z)
and dO' = ( - ~~, - ~~, 1) dx dy, so

Is (V x F) e dO' = i 2dx dy = 2(area of disk) = 81T.

4. Recall that p == Ilxll has i!; = 7;. Then

~ il~ F(p)x
V xF= J ill F(p)y
k ilz F(p)z

= (F'(P)~Z
p
- F'(p)::"y, F'(p)::"x - F'(p):'z, F'(p):'y -
p p p p
F'(P)~X)
p
=0,

as expected.

5. Think in terms of R2; the general case is similar. In the first quadrant, the
field will exert a counterclockwise torque on the lower right edge of the pin-
wheel, and an equally strong, oppositely directed torque on the upper left.
The net torque will be zero. The same happens outside the first quadrant,
owing to the symmetry.

6. The i-component of V x UF) is

oU F3) oU F2) OF3 of OF2 of

-
--.:.:..~ = f - + - F 3 - f - - -F2
oy OZ oy oy oz oz
= f (OF3 _ OF2) + (Of F3 _ of F2) ,
oy oz oy OZ

which is the i-component of f (V x F) + (V f) x F. We may similarly check

the other components.

Section 6.4
1. (a) V e F = ~~ + ~ + ~~ fA V e F = 3V(A) = 31Ta 2b.
= 3, so
(b) The top ofthe cylinder is given by z = b, so that N = k,

1 top
FeNda=l zda=b(1Ta 2 ).
top
Solutions to Exercises: Section 6.4 309

On the bottom, N = -k, so F. N = -z = O. On the curved part, the

. therad'lUS from the Z-axlS,
normaI at (x, y, z) lS . (2(X,y,O) H
2)1/2' ence
x +y

1side
F.Nda= 1
side
(x,y,z).
(x,y,O)
(x2 + y2)
1/2da

= a area (side) = a 2rrab.

The total flux is 3rra 2b. The match is predicted by the divergence
theorem.

2. The defining inequalities require x 2 + y2 ::s 4; otherwise, the square root is

biggest.

(a) Again V. F = 3, so fA V. F = 3V(A). The volume is

[ 1= [ lS-X2-y2 dz dy dx
JA Jx2+y2~4 ../S-x L y2
[27r [2 2rr (19 53/2)
= Jo Jo (5-r 2 -J5-r 2)rdrdB= 3- .

That makes fA V. F = 2rr (19 - 53/ 2 ).

(b) Theparaboloidz = 5-x 2 -y 2 hasdO' = (-~~,-~~,I)dxdy,
giving

1 top
F.dO' = [
Jx2+y2~4
(x,y,z).(2x,2y,l)dxdy

[27r [2
= Jo Jo (r2 + 5) rdrdB = 28rr.

The hemisphere has dO' = (~~, ~~, -1) dx dy (downward normal),

so

[ F. dO' = Jx2+y2~4
Jbottom
[ (x, y, z). (-:., -~, -1) dx dy

1
z z
_x2_y2_Z2
= dxdy
x2+y2~4 z

__ [027r [0 2
dB = 10rr (1 - ../5) .
5
Jo Jo ~rdr
Total flux is 38rr - 1Orr../5.
We are dealing with a simple region; the match is as expected.
310 Solutions to Exercises: Chapter 6

3. () n1 h
a ne ave v. -
T7 F _ "n
L...-j=l
a(-kxj/r 3 ) _
aXj 76 L...-j=l r
-
-k "n
r
(3 _
-
Xj 3r 2Xj) _

k(3 j n) , r > O. If n = 2, then V • F =

r r
!J
is not integrable. If n = 3,
then V.F = 0 except at one point, so that fB V.F = O. Finally, if n :::
4, we may integrate based on the differential dV = area(r-sphere) dr
(see Answer 5.6:7) = r n - 1area(1-sphere) dr. Thus,

(b) On the sphere, dO' = ~ dO' , so

1 bd(B)
F.dO' = 1bd(B)
kx x
- - . -dO'
a3 a
k _
= -2'
a
area(sphere) = -ka n 3area (Sl)'

Here we had no reason to expect a match, because B is not a subset

of the domain of F. They do match, however, for n ::: 4. We leave it
to the reader to consider why.

4. The boundary of A is the 2-sphere S(O, 2), with normals pointing away
from the origin, together with the I-sphere S(O, 1), normals pointing to-
ward the origin. Write B for the unit ball, oriented in the standard way,
with normals pointing outward. We have

[ F • dO' + [ F • dO' = [ F • dO' ,

lbd(A) lbd(B) 1S(0,2)
because the flux over the inner border of A is the negative of the flux out of
B. Next,
[ F • dO' = [ V•F
1S(O,2) 1B(O,2)
by Gauss's theorem, and fB(0,2) V.F = fA V.F +fB V.F by set additivity.
Finally, fB V • F = fbd(B) F. dO' (Gauss). Consequently, fbd(A) F. dO' =
fA V • F.
5. Straightforward: By Gauss's theorem, fbd(A) (V x F) • dO' = fA V. (V x F),
and

as long as the mixed partials are symmetric.

Solutions to Exercises: Section 6.4 311

Tricky: Draw the "equator" z = H(x,y)th(X,y), for (x, y) in the boundary of

the plane region that defines A. Going eastward, this curve is the boundary
of the upper half U of the surface bd(A), so by Stokes' theorem,

r(V x F) •
1u
du = 1 equator east
F • ds.

Going westward, the same curve is the edge of the lower half L of bd(A),
so that
r
1L
(V x F) • du = r
1equator west
F • ds.

Hence
r
lbd(A)
(V x F) • du = r+r
least lwest
= O.

6. (a) Undefined; you may not apply V to a vector.

(b) Defined and equals O. We can calculate directly. Alternatively, in a
convex open set, use Theorems 6.2:2 and 6.3:6.
(c) Defined;

V. VI _ 8(8118x) 8 (8f18y) 8 (8118z)

( )- 8x + 8y + 8z
8 I 8 I
- +88z-I2 '
2 2 2
-
- -
8x 2+8y2

This is an important construction in mathematical physics, beginning

with one type of flow: heat. [See Kline, p. 672 among others.]
(d) Undefined, just as (a).
(e) Defined. It is inconvenient to work with, but we may show that its i
component is

Doing likewise with the j and k components, we have the symbols

V x (V x F) = V(V. F) - (V. V)F.
(t) Defined, = 0 by Exercise 5.
(g) Defined, no special form or significance.
(h) Undefined; you may not apply V x to a scalar.
(i) Undefined; you may not apply V. to a scalar.
312 Solutions to Exercises: Chapter 6

7. (a)

(b)

V. (fG) = OfGI + ofG2 + ofG3

ox oy oz
= foGI + foG2 + foG3 +GI of +G2 of +G3 of
ox oy oz ox oy oz
= f(V • G) + G • V f.
References

Alan F. Beardon, Limits: A New Approach to Real Analysis, Springer-Verlag, New

York,1997.

This interesting book is a development of one-variable analysis found-

ed entirely on the notion of limit. Its Chapter 6 gives strictly analytic
definitions for the exponential and trigonometric functions. The con-
text there is functions of a complex variable, where we have no ex-
perience. Still, the proof we cited-Theorem 6.3.1 on page 90-uses
from complex variables only the formulas for sine and cosine of a
sum. Those formulas are derivable from just rearrangement of real
Taylor series.

R.C. Buck, Advanced Calculus, 2nd edition, McGraw Hill, New York 1965.

Originally published in 1956, "Buck" (no further identification was

needed) is a classic, and the reader is urged to look there for a breadth
of coverage that we could not hope to match.

Alberto Guzman, Continuous Functions o/Vector Variables, Birkhiiuser Boston,

2002.

This is a favorite of ours. The material most needed from it is from

Chapters 4-5, on properties of continuous functions and the topology
of Euclidean space.
314 References

Morris Kline, Mathematical Thoughtfrom Ancient to Modern Times, Oxford Uni-

versity Press, New York, 1972.

Kline's book is a monumental achievement among histories of the

sciences and mathematics. Our focus on calculus restricts our interest
in history to only Europeans during (mostly) the nineteenth century.
However, Kline is a wonderful source of information about the de-
velopment of mathematical ideas worldwide and over thousands of
years.

David C. Lay, Linear Algebra and Its Applications, 2nd edition, Addison Wesley
Longman, Reading MA, 1997.

Lay is a good source for material we have assumed from elementary

linear algebra.

L. Mirsky, An Introduction to Linear Algebra, Clarendon Press, Oxford, 1995,

reprinted by Dover, Mineola, NY, 1990.
E.R. Peck, Electricity and Magnetism, McGraw Hill, New York, 1953.

This is a sentimental favorite; the author was taught from it by a

wonderful teacher named Fred Rose.

Kenneth A. Ross, Elementary Analysis: The Theory o/Calculus, Springer-Verlag,

New York, 1980.

This is the best introduction to advanced calculus we know. The sub-

title suggests that Ross's mission is an axiomatization of elementary
calculus. The book carries out that mission with an admirable combi-
nation of mathematical rigor and attention to pedagogy. It has all the
material we require with respect to functions of one variable, together
with (Section 13) much of the topology of Euclidean space.
Index

Akin, Ethan, 152 change of variable theorem, 155

arc length. See length of arc characteristic function, 118
Archimedean set closure, notation clO, 119
characterized by boundary vol-
ume, 124 constrained extremes, 96
defined, 120 contour integral. See line integral
locally Archimedean, 161 contraction (mapping), 73
Archimedes' principle, 245 contraction principle, 74
area
counterclockwise
of a parallelogram, 188
of a surface. See surface area defined,224
average rate of change, 38 critical point, 43
average value theorem, 116, 147 cross product
defined, 197
big-box technique, 121
direction and magnitude, 199
boundary, notation bdO, 118
bounded variation, 184 cross-partition, 106
box, 53 curl of a field, 232, 233
broken line, 39 relation to path-independence,
228,235
C 1, C j function, 172 curvature, 210
central field, 235
curve
centroid, 160
chain rule, 15, 16 defined,60
change of variable, 153 smooth curve, 60
316 Index

Darboux's lemma, 113, 174 Faraday, Michael, 207

del operator, 36, 232 Faraday's law, 207, 232
derivative field,201
componentwise, 5,18 conservative field, 215
defined,2 fineness of a partition, 105
directional derivative, 33 first derivative test, 43
higher derivatives, 29 first-degree approximation, 2
of an integral, 218 flux, 206
partial, defined, 11 Fubini's theorem, 135
second derivative, 23
uniqueness, 7 Gauss, Karl F., 243
determinant, as volume, 150 Gauss's law, 208, 243
differentiability, differentiable func- Gauss's theorem, 243, 245
tion gradient
defined, 2 and path-independence, 212
implied by partials, 12 defined,35
twice differentiable function, 23 direction and magnitude, 35
differential, 19 normal to hypersurface, 68
directional derivative relation to directional derivative,
as component of gradient, 35 35
defined,33 relation to hypersurfaces, 66
Dirichlet's function, 113 gradient operator. See del operator
modified Dirichlet function, 130, Gram-Schmidt process, 149
138 Green region, 222
discontinuities, 129 Green's theorem, 219
divergence of a field
defined, 236 Hessian matrix, 50
divergence theorem, 240 higher derivatives, 29
domain, as open set, 10 componentwise, 29
dot-product rule, 22 holes in a set, 224, 226
Douglas, Jesse, 114, 243 hyperarea
introduced, 193
Duhamel's principle, 189
relation to cross product, 199
hypersurface, 66
edge of parallelepiped, 149 as graph of a function, 91
element of arc length, 178 as level surface, 90
vector element, 205 as meager set, 126
element of surface area, 192
vector element of area, 207,232 implicit function theorem, 85
energy, kinetic and potential, 215 scalar form, 53
Euler, Leonhard implicit functions, 66
Euler's equation, 240 defined,53
exterior, notation extO, 118 incompressible fluid, 240
extreme values. See maximum indefinite operator, 49
Index 317

initial value problem, 76 length of arc, 170

solution by iteration, 77 additivity, continuity, 174
inscribed volume, involume. level curve, (hyper)surface, 66
See volume line integral
integrability as limit of Riemann sums, 176
along a curve, 176, 181 defined, 176
local integrability, 162 orientation, 181
of continuous functions, 128 properties, 180
on a box, defined, 112 with respect to a coordinate, 181
on a locally Archimedean set, line segment (notation), 37
163 Lipschitz function, 41, 76, 185
on a set, defined, 118 locally Lipschitz, 41
integral, Riemann integral, 112 loop, 212
additivity, boundedness, linear- lower sums, upper sums, 112, 176
ity, monotonicity, 115, 144,
146, 164, 180, 194 matrix of second partials.
line integral, 176 See Hessian matrix
lower, upper integrals, 112 maximum, minimum, extreme val-
on a locally Archimedean set, ues
163 defined,42
interior, notation intO, 118 relation to outward directional
intersection-partition, 109 gerivatives, 43
intrinsic properties, 61, 170,205,210 relation to second derivative, 48
inverse function theorem, 84, 152 Maxwell, James, 201
scalar form, 53 Maxwell's equations, 201, 233,
inverse functions 243
scalar, 59 meager set, 125
vector, 79ff integral on, 132
irrotational field, 222 locally meager, 161
iteration (method), 75 mean value theorem, 37
relation to average rate of change,
38
Jacobian (determinant), 81,152,159
minimum values. See maximum
Jacobian matrix, 12
mixed partial derivative. See partial
Jordan, Camille, 230
derivative
Jordan curve theorem, 230
modified Dirichlet function, 130

Kramer, David, 243 negative-definite, -semidefinite oper-

ator. See positive definite
Lagrange, Joseph Louis, 102, 201, neighborhood, 7
243 Newton's second law, 215, 240
Lagrange norm
method,102 of a linear map, 15
multipliers, 102 of a partition, 105
Lebesgue, Henri, 131, 137,230 Pythagorean norm, 2
318 Index

normal,67 row of rows, row(2) , 24

normal plane, 67 row of rows(i) , row(i+l), 29
relation to surface area, 197 R-sum. See Riemann sum
unit normal vector, 205
saddle, saddle point, 45, 48
o and 0 notation, 9, 18 second derivative, 23
open mapping, 160 second derivative test, 47
operator inequality, 16 sector, 7
optimization. See constrained extremes simple path, 227
orientation, 181 simple region, 240, 243
outward directional derivative, 43 simplest refinement of partitions, 109
overlap, overlapping sets, 105 simply connected set, 226
singularity, 62
parallelepiped, 149 space-filling arc, 172
volume, 150 speed,215
partial derivative spherical coordinates, 22, 160
defined, 11
Stokes' theorem, 232
implying differentiability, 12
subdeterminants (relation to definite-
mixed partial, 24
ness),51
symmetry of mixed partials, 27,
subintervals of a partition, 105
30
substitution rule, 155
partition of a box, 105
one variable, 160
path,203
surface
path integral, 202
as intersection of hypersurfaces,
path-independence, 212, 228, 235
92
piecewise smooth arc, 180
defined,62,189,192
polar coordinates, 18, 84, 158, 185
arc length, 185 in dimension 3, 90
polygonal path, 39 of dimension> 2,66, 193
positive definite, semidefinite opera- smooth surface, 62, 65, 192
tor,49 surface area
potential, 212 defined, 188, 193
product rule, 21 element of, 196
Pythagorean norm, 2 higher dimensional, 192
vector element of, 207, 232
quadratic form, 50 surface integral
defined, 192
rectifiability, 170 properties, 194
of a surface, 188
refinement, 107 tangent plane, hyperplane
region between graphs, 139 characterized, 65, 67
region under a graph, 143 defined,62
Riemann sum, 114, 136 dimension> 2, 66
right-hand rule, 208 relation to tangent to a curve,
rotational field, 222 63,92
Index 319

tangent, tangent line, 60 vector

to curve on surface, 63 as column matrix, 2
unit tangent vector, 204 vector field, 201
Taylor's theorem, 52 vector product. See cross product
total differential, 19 volume
transformation of coordinates. See change as integral, 121, 142
of variable circumscribed volume, circum-
transpose volume, 119, 124
column as transposed row, 11 inscribed volume, involume, 119,
transversal intersection, 94 124
of a set, defined, 120
upper sums, lower sums, 110, 176