Instructor's Solutions Manual To Advanced Calculus - A Geometric View ( James J. Callahan )

Solutions:
Advanced Calculus
A Geometric View
James Callahan
DVI file created at 14:27, 26 April 2011

iii
Preface
This book consists of detailed solutions to the exercises in my text Advanced Calculus: A Geometric
View (Springer, New York, 2010, ISBN 978-1-4419-7331-3). My Web site,
http://maven.smith.edu/˜callahan/
contains additional material, including an errata file for the text,
ac/correct.pdf
This version of the solutions manual takes into account all corrections to the text found in the errata
file up to 17 April 2011.
Given the nature of the material, it is likely that the solutions themselves still have some errors. I
invite readers of this book to contact me at
callahan@math.smith.edu
to let me know about any errors or questions concerning the solutions found here. I will post a
solutions errata file on my Web site.
This solutions manual is available only directly from Springer, via its Web site
http://www.springer.com/instructors?SGWID=0-115-12-333200-0
J. Callahan
April 2011

iv

Contents
1 Starting Points 1
2 Geometry of Linear Maps 9
3 Approximations 23
4 The Derivative 32
5 Inverses 46
6 Implicit Functions 59
7 Critical Points 66
8 Double Integrals 74
9 Evaluating Double Integrals 80
10 Surface Integrals 91
11 Stokes’ Theorem 100

vi CONTENTS

Solutions: Chapter 1
Starting Points
1.1. The integrals are “improper,”, because there is an Therefore,

endpoint at ±∞. However, we know from the text that Z ∞ b
arctan x dx (arctan x)2 (π /2)2 π 2
Z b b = lim = = .
dx 0 1 + x2 b→∞ 2 2 8
2
= arctan x ; 0
a 1+x a
1.5. a. The push-forward substitution u = ln w yields
basic facts about improper integrals (cf. Chap. 9 of the Z Z
text) then allow us to write dw du
= .
w(ln w) p up
Z ∞ b
dx
= lim arctan x = π /2; Hence
0 1+x
2 b→∞
0 
Z 1 1 Z ln(ln w), p = 1,
dx dw
= lim arctan x = π /4 − (−π /2) = 3π /4. = 1
−∞ 1 + x
2 a→−∞
a w(ln w) p  , otherwise.
(1 − p)(lnw) p−1
1.2. We use the push-forward substitution u = 1 + x2, so 1.5. b. For the value of I to be finite, the expression just
that du = 2x dx and then given in part (a) must be finite as w → ∞. This rules out
Z Z 1 the first case (wherep = 1) and requires p − 1 > 0 in the
√ p
x dx 2 du 1 2 second case. Thus, I is finite precisely when p > 1, and
= = 2 ln u = ln u = ln 1 + x .
1 + x2 u then
b
1 −1
1.3. We use the pullback substitution x = R sin θ , so that I = = lim = .
b→∞ (1 − p)(lnw) p−1 (1 − p)(ln 2) p−1
dx = R cos θ d θ , R2 − x2 = R2 cos2 θ and θ = ±π /2 when 2
x = ±R. The result is

1.6. Assume first that the function x = ϕ (s) and its
Z Rp Z π /2
derivative are continuous on a ≤ s ≤ b. Then ϕ will have
R2 − x2 dx = R cos θ · R cos θ d θ an inverse on that interval if ϕ ′ (s) 6= 0 on a < s < b. To
−R −π /2
Z π /2 find the inverse we must solve the equation x = ϕ (s) for
1 + cos2θ
= R2 dθ the variable s.
−π /2 2
1.6. a. Here ϕ ′ = −1/s2, so ϕ is invertible on each of the
2 θ sin 2θ π /2 π R2 rays s > 0 and s < 0. The inverse is s = 1/x on each ray.
=R + = .
2 4 −π /2 2
1.6. b. Here ϕ ′ = 1 + 3s2 > 0 on the entire s-axis. The
inverse is the unique real root of s3 + s − x = 0 that is
1.4. Let u = arctan x (a push-forward substitution); then provided by Cardano’s formula:
du = dx/(1 + x2) so v v
u s u s
u 2 u
Z
arctan x dx
Z
u 2
(arctan x)2 t3 x x 1 t3 x x2 1
= u du = = . s = + + + − + .
1 + x2 2 2 2 4 27 2 4 27

2 SOLUTIONS: CHAPTER 1. STARTING POINTS
1.6. c. Here ϕ ′ = (1 − s2 )/(1 + s2 )2 , so ϕ has an inverse 1.6. e. First of all, ϕ is defined only when −1 < s < 1. In
on each of the three domains s ≤ −1, −1 ≤ s ≤ 1, and that case,
1 ≤ s. The graph of ϕ , shown below, makes it clear that ϕ ϕ′ =
1
>0
has an inverse on each of the three sets. (1 − s2)3/2
so ϕ is invertible for −1 < s < 1. To get the formula for
0.4
the inverse, note that
0.2
s2
-20 -10 10 20 x2 = or x2 = s2 + s2 x2 .
-0.2 1 − s2
√
Solving for s gives s = ±x/ 1 + x2. Because x = ϕ (s)
-0.4
has the same sign as s, we must choose the plus sign in

To find a formula for the inverse, we need to solve the the formula above, so the inverse of ϕ is
equation x = s/(1 + s2) for s. We have
x
2 2
s= √ .
x + xs = s or xs − s + x = 0. 1 + x2
The roots of this quadratic equation are 1.6. f. Here ϕ (s) = ms + b is invertible everywhere pro-
√ vided ϕ ′ = m 6= 0. In that case, the inverse is
1 ± 1 − 4x2
s= , |x| ≤ 1/2.
2x x−b
s= .
m
This formula defines all three inverse functions. first of
all, with the minus sign, the graph is:
1.6. g. Here ϕ ′ = sinh s; because ϕ ′ is nonzero on each of
1.0 the rays s < 0 and s > 0, ϕ will have an inverse on each
0.5
of these domains. To find the formulas, we just adapt the
procedure for sinh s, above. We have
-0.4 -0.2 0.2 0.4
-0.5 2x = es + e−s or (es )2 − 2xes + 1 = 0.
-1.0
This time the roots give us the two inverse functions
Clearly, this is the inverse for −1 ≤ s ≤ 1. Note that there ( √
is an indeterminate form when x → 0, but l’Hopital’s rule − ln x + x2 − 1 < 0,
s= √
shows that s → 0. With the plus sign, the formula defines ln x + x2 − 1 > 0.
the inverse on each of the other two domains:
−1/2 ≤ x < 0 ⇒ s ≤ −1; 0 < x ≤ 1/2 ⇒ 1 ≤ s. 1.6. h. Here ϕ ′ = 1 − 3s2 is nonzero on three separate do-
mains:
1.6. d. Here ϕ ′ = cosh s > 0 for all s, so ϕ is invertible −1 −1 1 1

s< √ , √ <s< √ , √ < s.
for all s. To get the formula for the inverse, we first write 3 3 3 3
2x = es − e−s or 2xes = e2s − 1. In each case the inverse is a root of the cubic equation
s3 − s + x = 0 as provided by Cardano’s formulas.
This is the quadratic equation (es )2 − 2xes − 1 = 0, and 1.6. i. Here ϕ ′ = sech2 s ≥ 1, so ϕ is invertible for all s.
has the roots To find the inverse, we have
√
2
2x ± 4x + 4 p
es = = x ± x2 + 1. es − e−s e2s − 1
2 x = = or e2s (x − 1) = −x − 1.
es + e−s e2s + 1
The minus sign is untenable, because es > 0, so we obtain Therefore,
the formula for the inverse as r

p 1 1+x 1+x
s = ln x + x2 + 1 . s = ln = ln .
2 1−x 1−x

3
√
1.6. j. Here ϕ ′ = −2/(1 + s)2 < 0 for all s 6= −1. Thus when ∆s ≈ 0. In this question, ϕ (s) = s and s0 = 100,
ϕ has an inverse on each ray s < −1, −1 < s. To get the so the approximation is given by
formula, we note
√ ∆s
1−x 100 + ∆s ≈ 10 + , ∆s ≈ 0.
x + xs = 1 − s or s = 20
1+x
Therefore we have
on each ray. √
102 ≈ 10.1, (∆s = +2),
1.7. a. Let θ = arcsin s, so s = sin θ . Therefore √
p 99.4 ≈ 9.97, (∆s = −0.6).
p
cos θ = 1 − sin θ = 1 − s = f (s),
2 2
sin θ s 1.9. c. Calculator values to nine decimal places are

tan θ = =√ = g(s). √ √
cos θ 1−s 2
102 = 10.099 504 938, 99.4 = 9.969 954 864.
1.7. b. Applying the chain rule to cos(arcsin s), we get Therefore the microscope estimates are too large by about
0.000 5 and 0.000 004 5, respectively.
1 −s
f ′ (s) = − sin(arcsin s) · √ =√ . 1.9. d. The microscope values √ are given by the tangent
1−s2 1 − s2
line to the graph of x = ϕ (s) = s at s = 100. The graph
√
Direct computation of the derivative of f (s) = 1 − s2 itself is concave down there (because ϕ ′′ (100) < 0), so it
gives the same result. lies below its tangent line at that point.
Applying the chain rule to tan(arcsin s), we get 1.10. a. Here ϕ ′ (s) = −1/s2 , so ϕ ′ (2) = −1/4 and the
1 microscope equation is ∆x ≈ −∆s/4. Therefore the ap-
g′ (s) = sec2 (arcsin s) · √ proximation is given by
1 − s2
1 1 1
= ·√ ≈ 0.5 − 0.25∆s, ∆s ≈ 0.
cos2 (arcsin s) 1 − s2 2 + ∆s
1 1 1 This gives us the estimates
= 2
·√ = .
1−s 1−s 2 (1 − s2 )3/2
√ 1
≈ 0.4925, (∆s = 0.03),
Direct computation of the derivative of g(s) = s/ 1 − s2 2.03
gives the same result. 1
≈ 0.505, (∆s = −0.02).
1.8. From the previous exercise we know that if we take 1.98
x = arcsin s, then
1.10. b. Calculator values are
p 3 ds
cos3 x = 1 − s2 = (1 − s2)3/2 , dx = √ . 1
= 0.492 610 837,
1
= 0.050 505 050.
1 − s2 2.03 1.98
Consequently, This time the microscope estimates are too small, by about
Z Z 3 −0.000 1 and −0.000 05, respectively.
s3 sin x
cos3 x dx = (1 − s2 ) ds = s − = sin x − ,
1.10. c. This time the tangent line (which gives the micro-
3 3
scope estimates) lies below the graph, because the graph
using s = sin x. is concave up: ϕ ′′ (2) > 0.
√
1.9. a. We have ϕ ′ (s) = 1/(2 s), so ϕ ′ (100) = 1/20 1.11. Let ϕ (s) = √s, and let s0 = 1. The approximation
and the microscope equation is we use (see the solution to Exercise 1.9) is
∆s ϕ (s0 + ∆s) ≈ ϕ (s0 ) + ϕ ′ (s0 )∆s, ∆s ≈ 0.
∆x ≈ ϕ ′ (100)∆s = .
20
Here ∆s = 2h, ϕ (s0 ) = 1, and ϕ ′ (s0 ) = 1/2, so our equa-
1.9. b. The microscope equation is used in the setting tion becomes
√
ϕ (s0 + ∆s) = ϕ (s0 ) + ∆x ≈ ϕ (s0 ) + ϕ ′ (s0 )∆s 1 + 2h ≈ 1 + h, h ≈ 0.

√ x2 = ϕ (s) = tan s, then ϕ (s) = sec s, so

1.12. a. When ′ 2 1.18. Again, work done is the dot product of force and
ϕ (π /4) = ( 2) = 2 and the microscope equation is
′ displacement:
∆x ≈ 2∆s. (a) W = −1 + 2 = 1, (b) W = 9, (c) W = 2.
1.12. b. Because tan(π /4) = 1, the approximating equa- 1.19. Write the unknown force as F = (P, Q). The given
tion (Ex. 1.9) takes the form information of work done on two particular displacements
provides two equations:
tan(π /4 + ∆s) ≈ 1 + 2∆s, ∆s ≈ 0;
F · (2, −1) = 2P − Q = 7,
with ∆s = h, this can be rewritten as
F · (4, 1) = 4P + Q = −3.
tan(h + π /4) ≈ 1 + 2h, h ≈ 0.
Solving these equations, we find P = 2/3, Q = −17/3, so
Because the graph of x = ϕ (s) = tan s is concave up at F = (2/3, −17/3). This allows us to determine
π /4 (i.e., ϕ ′′ (π /4) > 0), the graph lies above its tangent W = F · (1, 0) = 2/3, W = F · (0, 1) = −17/3.
line, so actual values of the tangent function are greater
than their approximations using the microscope equation. The work done will be zero if ∆x is perpendicular to F;
1.13. The local length multiplier at a point is the value of thus ∆x can be any nonzero multiple of (17/3, 2/3).
the derivative at that point, in this case the value of coss. 1.20. Consider first W = F · ∆x as function of its first ar-
Thus gument, F. We have
s 0 π /4 π /2 2π /3 π W (F1 + F2 ) = (F1 + F2 ) · ∆x
√
length multiplier at s 1 1/ 2 0 −1/2 −1 = F1 · ∆x + F2 · ∆x = W (F1 ) + W (F2 ),
W (rF) = (rF) · ∆x = rF · ∆x = rW (F).
1.14. At a point where the local length multiplier of the
function x = ϕ (s) is negative, the map ϕ : s → x reverses This shows W is a linear function of its first argument,
orientation. i.e., F. Symmetry of the dot product then shows that W is
likewise a linear function of its second argument, ∆x.
1.15. We have
s ′ 1.21. If θ is the angle between F and ∆u, then a formula
e − e−s es + e−s for the dot product allows us to express the work W in
(sinh s)′ = = = cosh s,
2 2 terms of θ as
s
−s ′ s − e−s p
e + e e
(cosh s)′ = = = sinh s. W = F · ∆u = kFkk∆uk cos θ = P2 + Q2 cos θ .
2 2
We have used the fact that k∆uk = 1. Because (P, Q) is
Furthermore, for all s we have fixed, W depends only on θ ; it assumes it maximum when
s −s
2 s −s
2 cos θ = +1 (i.e., when θ = 0) and its minimum when
e +e e −e
cosh2 s − sinh2 s = − cos θ = −1 (i.e., when θ = π ). Thus the unit displace-
2 2
ment that maximizes W is the one in the same direction as
e2s + 2 + e−2s e2s − 2 + e−2s the force, and the one that minimizes W is in the opposite
= −
4 4 direction.
2 − (−2)
= = 1. 1.22. Work as dot product gives us two linear equations
4
in the unknowns P and Q:
1.16. When x = sinh s, then 1 + x2 = 1 + sinh2 s = cosh2 s
aP + cQ = A, a c P A
and dx = cosh s ds, so = .
bP + dQ = B, b d Q B
Z Z Z
dx cosh s ds
√ = = ds = s = arcsinh x. The equations define the matrix equation shown above on
1 + x2 cosh s the right; to find P and Q, invert the matrix:
1.17. In each case, work done is the dot product of force dA − cB
P= ,
and displacement. Thus P
=
1 d −c A
; ∆
Q ∆ −b a B −bA + aB
(a) W = (2, −3)·(1, 2) = −4, (b) W = 8, (c) W = −2. Q= ,
∆

5
where ∆ = ad − bc. To find P and Q, these equations must 1.24. c. This problem requires us to determine the value
be solvable, that is, the determinant ∆ must be nonzero. of W no matter what path is chosen between the end-
points. Therefore we cannot simply chose our own path,
1.23. a. y
0
but must instead use an arbitrary parametrization
x = x(t), y = y(t), a ≤ t ≤ b.
~ we have
Because we know the endpoints of C,
x(a) = 5, y(a) = 2, x(b) = 7, y(b) = 11.
x
−1 1 Therefore,
Z b
W= y(t) · x′ (t) dt + x(t) · y′ (t) dt
a
t = −2 2 Z b b
′
= x(t) · y(t) dt = x(t) · y(t)
a a
The five parameter points are marked; the curve is the en- = x(b)y(b) − x(a)y(a) = 77 − 10 = 67.
tire unit circle minus the point (x, y) = (0, −1).
The key is to recognize that the original integrand is the
1.23. b. We can determine the limits by rewriting the ex- derivative of the product x(t) · y(t).
pressions for x and y; then 1.24. d. The path is now in (x, y, z)-space, where the work
2 done by the force F = (P, Q, R) is
2 (1/t ) − 1 Z
x= → 0, y= → −1,
(1/t) + t (1/t 2 ) + 1 W = P dx + Q dy + R dz.
~
C
as t → ±∞.
In the present case, this reduces to
1.23. c. We have Z 1 1

W= 0 · 2 dt + 0 · dt − mg · (−2t) dt = mgt 2 = mg.
4t 2 + 1 − 2t 2 + t 4 1 + 2t 2 + t 4 (1 + t 2)2
α= = = ≡ 1.
0 0
(1 + t 2)2 (1 + t 2)2 (1 + t 2)2
1.24. e. Here
By construction, α (x(t), y(t)) is the square of the distance Z A
from the point (x(t), y(t)) to the origin; α ≡ 1 means that W = − sin θ · (− sin θ ) d θ + cos θ · cos θ d θ + 1 · 3 d θ
0
all points on the curve lie on the unit circle. Z A
= (sin2 θ + cos2 θ + 3) d θ = 4A.
1.24. In every case, the work done by F = (P, Q) is the 0
path integral
Z Z 1.25. a. We need to parametrize C; ~ one possibility is
W= F · dx = P dx + Q dy. x(t) = (x(t), y(t)) = (3t − 2, 4t + 3), 0 ≤ t ≤ 1.
~
C ~
C
Then x + 2y = 11t + 4, x − y = −t − 5, and
1.24. a. Here x = τ 2 , dx = 2τ d τ , y = τ 3 , dy = 3τ 2 d τ ; Z Z 1
therefore, F · dx = (11t + 4, −t − 5) · (3, 4) dt
~
C 0
Z 2 2 Z 1 1
2τ 4 9τ 6 t2 13
W= τ 2 · 2τ d τ + 3τ 3 · 3τ 2 d τ = + = 102. = (29t − 8) dt = 29 − 8t = .
1 4 6 1 0 2 0 2
1.25. b. Here
~ as (x, y) = (2 cost, 2 sint), with
1.24. b. Parametrize C Z Z 1
0 ≤ t ≤ π . Then F · dx = (t 3 , 1 − t,t 2) · (2t, 1, −1) dt
~
C 0
Z π Z 1
W= −2 sint · (−2 sint) dt + 2 cost · 2 cost dt = (2t 4 − t 2 − t + 1) dt
0 0
Z π
2 1 1 17
= 4 dt = 4π . = − − +1 = .
0 5 3 2 30

1.25. c. Because x2 + y2 = R2 here, the integral becomes be done by another force acting against gravity to lift the
Z 8π object from the lower point to the higher.
−R sint R cost
, · (−R sint, R cost) dt 1.27. c. Because the points are at the same vertical height,
0 R2 R2
Z 8π c = γ , so no work is done by gravity. The gravitational
= (sin2 t + cos2 t) dt = 8π . force near the surface of the earth is conservative in the
0
sense that the work done by gravity in moving an object
over any closed path is zero; more generally, the path need
1.26. a. We show (cf. Exercise 1.23) that r(u) lies on the
merely begin and end at the same vertical height.
circle of radius 2 centered at the origin:
1.28. a. The explicit form of F(x, y, z) is
16u2 + 4u4 − 8u2 + 4 u4 + 2u2 + 1
kr(u)k2 = = 4 = 4.
(u2 + 1)2 u4 + 2u2 + 1 −µ x −µ y −µ z
√ , , .
(x2 + y2 + z2 )3/2 (x2 + y2 + z2 )3/2 (x2 + y2 + z2 )3/2
Next,√ we determine the endpoints r(2 ∓ 3). When u =
2 ∓ 3,
√ √ −µ x µ µ µ
8∓4 3 8∓4 3 1.28. b. We have kFk =
r3 = r3 kxk = r3 r = r2 .
x= √ = √ = 1,
4∓4 3+3+1 8∓4 3
√ √ √ 1.28. c. We can write Φ (x, y, z) = µ (x2 + y2 + z2 )−1/2 , so
8∓8 3+6−2 3∓2 3 2± 3 √
y= √ = √ · √ = ∓ 3.
8∓4 3 2∓ 3 2± 3 ∂Φ −µ x
√ √ = − 21 µ (x2 + y2 + z2 )−3/2 · 2x = 3 .
Thus r(u) begins at (1, − 3) and ends at (1, 3). The ∂x r
graphs of x(u) and y(u), (below right and left), show that By symmetry we have
x increases from 1 to 2 and then decreases
√ back
√ to 1 while
y increases monotonically from − 3 to + 3. ∂Φ −µ y ∂Φ −µ z
= 3 , = 3 ,
2.0 ∂y r ∂z r
1.5
1.5
1.0
so that grad Φ (x) = −µ x/r3 = F.
1.0 0.5
0.5 1.0 1.5 2.5 3.0 3.5 1.28. d. The work done is just the potential difference:
0.5 -0.5
µ µ 7µ
-1.0
0.5 1.0 1.5 2.0 2.5 3.0 3.5

-1.5
W = Φ (perihelion) − Φ (aphelion) = − = .
This confirms that r(u) traverses the circular arc once 3 10 10
√ √
counterclockwise from (1, − 3) to (1, + 3).
1.28. e. Because the gravitational force is conservative,
1.26. b. We have y′ (u) = 8u/(u2 + 1)2 , so the work done in one complete orbit (a closed path) is
Z Z 2+√3 zero. (The net energy expenditure is zero: energy is con-
32u2 √ 4π
x dy = √ 2 3
du = 3 + ; served.)
~
C 2− 3 (u + 1) 3 √
1.29. a. Here x′ = (6t, 8t), so kx′ k = 36t 2 + 64t 2 =
we have used a computer algebra system to evaluate the 10|t| = 10t because t ≥ 0. Thus
integral. This value of the integral agrees with those found
in the text using different parametrizations. Z 1 1

arc length = 10t dt = 5t 2 = 5.
1.27. a. We have ∂ Φ /∂ x = ∂ Φ /∂ y = 0 while ∂ Φ /∂ z = 0 0
−gm, showing that grad Φ = (0, 0, −mg) = F.
(The curve is the straight line from (0, 0) to (3, 4).)
1.27. b. Because the gravitational force F has the poten-
√
tial Φ (x, y, z), the work done is just the potential difference 1.29. b. Here x′ = (2t, 3t 2), so kx′ k = t 4 + 9t 2 and
(α ,β ,γ ) (α ,β ,γ ) 3
Z 3 p
W = Φ (x, y, z) = − gmz = gm(c − γ ). arc length = 2
t 4 + 9t dt =
2 2 3/2
(4 + 9t )
(a,b,c) (a,b,c) 1 3 · 18 1
This is negative if c − γ < 0, i.e., if c < γ . Negative work 853/2 − 133/2

= = 27.2885.
by gravity can be interpreted as positive work that has to 27

7
1.29. c. Here Therefore t = σ (s) = tan(s/2). To determine x(σ (s)), we

also need
x′ = (et cost − et sint, et sint + et cost)
1
= et (cost − sint, sint + cost), 1 + t 2 = sec2 (s/2), = cos2 (s/2),
1 + t2
kx′ k2 = e2t cos2 t − 2 cost sint + sin2 t which then give

+ sin2 t + 2 sint cost + cos2 t = 2e2t ,
1 − t2 2 sin2 (s/2)
= cos (s/2) 1 −
so Z b√ 1 + t2 cos2 (s/2)
√
arc length = 2 et dt = 2(eb − ea). = cos2 (s/2) − sin2 (s/2) = cos s,
a
2t sin(s/2)
√ =2 cos2 (s/2)
1.29. d. Here x′ = (− sint, cost, k) so kx′ k = 1 + k2 and 1+t 2 cos(s/2)
Z 2π p p = 2 sin(s/2) cos(s/2) = sin s.
arc length = 1 + k2 dt = 2π 1 + k2 .
0 Therefore,
1.29. e. We have y(s) = x(σ (s)) = (cos s, sin s), −π < s < π .

−4t 2 − 2t 2 The domain of s was determined here by noting that s =
x′ (t) = , 2 arctant = ±π when t = ±∞.
(1 + t ) (1 + t 2)2
2 2
16t 2 + 4 − 8t 2 + 4t 4 4 1.32. We use Definition 1.6 and Theorem 1.5 (text

kx′ (t)k2 = = ; page 18) to express total mass as the path integral of mass
(1 + t 2)4 (1 + t 2)2
density over the wire. If we use x(t) = (R cost, R sint) as
hence the parametrization of the curve, then kx′ (t)k = R, mass
Z 1 1 density is ρ (x(t)) = 1 + R2 cos2 t gm/cm, and
2
arc length = dt = 2 arctant = π. Z 2π
−1 1 + t
2
−1
total mass = (1 + R2 cos2 t) · R dt = 2π R + π R3 gm.
0
(This can also be seen by noting that the curve is a semi-
circle of radius 1; see Exercise 1.23.) 1.33. This exercise is similar to the example worked in
√
1.29. f. One possible parametrization is the text on pp. 18–19. In particular, ds = 2 dt, so

x(t) = (3 cost, 4 sint), 0 ≤ t ≤ 2π .
Z Z 4π √ √ t 2 4π √
z ds = t · 2 dt = 2 · = 8π 2 2,
p C 0 2 0
Then kx′ (t)k = 9 sin2 t + 16 cos2 t and √
Z Z 4π √ √ t 3 4π 64π 3 2
Z 2π p 2
z ds = 2
t · 2 dt = 2 · = .
arc length = 9 sin2 t + 16 cos2 t dt = 22.1035. C 0 3 0 3
0
1.34. Suppose C is a curve in the plane; then in an ex-
The value is determined by a computer algebra system.
pression of the form
1.30. a. We have kx′ (t)k = R; therefore, if we take t = 0 Z
and the initial point, then f (s) ds,
C
Z t
s(t) = R dt = Rt. the value f (s) is independent of the position of C. Thus
0 if C has arc length L, we can define
Z Z L
1.30. b. The inverse of s = s(t) is simply t = σ (s) = s/R. f (s) ds = f (s) ds,
The arc-length parametrization is C 0
where the integral on the right is just an ordinary one.
y(s) = R cos(s/R), R sin(s/R) .
Therefore, if C is a circle of radius 5, then L = 10π and
I Z 10π
1.31. Using kx′ (t)k from the solution to Exercise 1.29.e
cos s ds = cos s ds = 0,
and taking the initial value as t = 0 (so s(0) = 0), we have C 0
I Z 10π
Z t
s(t) =
2
dt = 2 arctant cos2 s ds = cos2 s ds = 5π .
C 0
0 1 + t2

1.35. The change from Cartesian to polar coordinates, For the general function, shown below, we use different
written as scales on the two axes, and mark the inflection points at
x = r cos θ , y = r sin θ , positions x = µ ± σ .
y
is a map (r, θ ) → (x, y) and is therefore a pullback substi- 1
tution.
y y = gµ,σ (x)
1.36. a. & b. In polar coordi-
nates,
µ − 3σ µ µ + 3σ x
√ 1.37. b. Let u = (x − µ )/σ ; then du = dx/σ and u = ±∞
1 ≤ r ≤ 10, D
D: when x = ±∞. Therefore,
0 ≤ θ ≤ π /2. Z ∞ Z ∞
x 2 √
gµ ,σ (x) dx = eu /2 σ du = σ · I = σ 2π .
1 √10 −∞ −∞
1.36. c. With x2 + y2 = r2 and dx dy = r dr d θ , we have
1.38. a. The narrowest and tallest graph below is the one
Z Z π /2 Z √10 with σ = 1/2; the widest and shortest has σ = 3. For
sin(x2 + y2 ) dx dy = sin(r2 ) r dr d θ clarity the vertical and horizontal scales are different.
D 0 1
√10 Z π /2 1.0
− cos(r2 )
= dθ 0.8
2 1 0 0.6
π cos 1 − cos10
= · = 1.08336. 0.4
2 2
0.2
1.37. a. To determine critical points and inflections, we -4 -2 0 2 4
need 1.38. b. The graphs are just horizontal translates of one

another; from left to right, they have µ = −2, 1, 10. The
−(x − µ ) −(x−µ )2 /2σ 2 −(x − µ )
g′µ ,σ (x) = e = gµ ,σ (x), vertical and horizontal scales are different.
σ2 σ2 0.5
−1 x−µ
g′′µ ,σ (x) = 2 gµ ,σ (x) − 2 g′µ ,σ (x) 0.4
σ σ
(x − µ )2
0.3
−1
= 2 gµ ,σ (x) + gµ ,σ (x)
σ σ4
0.2
(x − µ )2 − σ 2 0.1
= gµ ,σ (x)
σ4 -5 0 5 10
1.39. The procedure here is similar to the one used to

At a critical point, g′µ ,σ (x)
= 0; but because g µ ,σ > 0 and solve Exercise 1.37.b. If we use the suggested substitution
σ 2 > 0, g′µ ,σ (x) = 0 only when x − µ = 0, i.e., x = µ . z = (x − µ )/σ , then dz = dx/σ , z = 0 when x = µ , and
Moreover, because z = (b − µ )/σ when x = b; hence
Z
(µ − µ )2 − σ 2 −σ 2 1 b
e−(x−µ ) /2σ dx
2 2
g′′µ ,σ (µ ) = gµ ,σ (µ ) = 4 gµ ,σ (µ ) < 0, Prob(µ ≤ Xµ ,σ ≤ b) = √
σ 4 σ σ 2π µ
Z (b−µ )/σ
we see that x = µ is a maximum. 1 2
=√ e−z /2 dz
2π 0
= Prob(0 ≤ Z0,1 ≤ (b − µ )/σ ).
1
-1 1 3 5 7 9 11
1.40. When a < µ , we need to add Prob(a ≤ Xµ ,σ ≤ µ )
At an inflection, g′′µ ,σ (x) = 0; therefore, to the probabilities that must be determined. Now the nor-
mal density function Xµ ,σ is symmetric around x = µ , and
(x − µ )2 − σ 2 = 0 or x = µ ± σ . so are the pair of points x = a and x = 2µ − a (because
their average is µ ); therefore we can write
In the graph above (with µ = 5, σ = 2), the horizontal and Prob(a ≤ Xµ ,σ ≤ µ ) = Prob(µ ≤ Xµ ,σ ≤ 2µ − a).
vertical axes have the same scale.

Geometry of Linear Maps
2.1. a. By definition of M1 , the image of a line of slope 2.4. c. We must show the image of (2, 1) lies in the direc-
m = ∆v/∆u in the (u, v)-plane is a line in the (x, y)-plane tion of (2, 1). We have
whose slope is
6 2 2 14 2
= =7 .
∆y 3∆v/5 3∆v 3 2 3 1 7 1
= = = m 6= m when m 6= 0, ∞.
∆x 2∆u 10∆u 10
2.4. d. To show the image of (−1, 2) lies in the direction
2.2. According to the text, page 33, and the later discus- of (−1, 2), we have

sion, pages 35–36, the coordinate change matrix G that 6 2 −1 −2 −1
we need for M5 = G−1 M5 G has for its columns the eigen- = =2 .
2 3 2 4 2
vectors of M5 . From the work on page 34 of the text, we
see 2.4. e. y

1 −3 1 3
G= , G−1 = 41 .
1 1 −1 1
A straightforward calculation shows that G−1 M5 G = M5 .
2.3. By hypothesis, there are n × n matrices H and G for
which
C = H −1 BH and B = G−1 AG.
But then C = H −1 G−1 AGH = K −1 AK with K = GH (and
K −1 = H −1 G−1 ), showing that C is equivalent to A. x
2.4. a. The area multiplier is detM = 18 − 4 = 14.
2.4. b. y
2.4. f. The image of a single square of the new grid is a
large black rectangle in the figure above. In terms of the
grid of gray unit squares, the black rectangle’s dimensions
are 7 × 2. Because the image of a gray unit square is made
up of 14 such squares, the area multiplier is again 14.
2.5. a. We have tr M = −1, det M = −6, so the character-
x istic polynomial is λ 2 + λ − 6 = (λ + 3)(λ − 2). Thus the
eigenvalues are λ = −3, +2. For λ = −3, an eigenvector
is a nonzero solution of

4 2 x 0
(M − λ I)X = = .
2 1 y 0

−1
Thus for an eigenvector we can take .
2

10 SOLUTIONS: CHAPTER 2. GEOMETRY OF LINEAR MAPS
The eigenvector equation for λ = +2 is Hence the grid we choose is the same as the grid of gray
squares used in Exercise 2.4.
−1 2 x 0 y
(M − λ I)X = = ;
2 −4 y 0

2
For an eigenvector we can take .
1
2.5. b. Now tr M = 2and detM = −3, so the characteris-
tic polynomial is λ 2 − 2λ − 3 = (λ − 3)(λ + 1). For the
eigenvalue λ = +3 we have the eigenvector equation

−1 1 x 0
(M − λ I)X = = ; x
3 −3 y 0

1
1
For the eigenvalue λ = −1, the eigenvector equation is

3 1 x 0
(M − λ I)X = = ;
3 1 y 0
The image of the grid of gray squares is the grid of black
−1 rectangles. The gray square with its lower left corner at
3 the origin (shown slightly darker) has its boundary ori-
2.5. c. Here tr M = −1 and det M = −6, so the charac- ented counterclockwise. Its image is the black rectangle
teristic polynomial is λ 2 + λ − 6 = (λ + 3)(λ − 2). The with its boundary oriented clockwise. Each gray square is
eigenvalues are λ = −3, +2 (as they were for the matrix stretched by the factor +2 in one direction; in the other,
in part a). For λ = −3, the stretch, by a factor of 3, is combined with a flip. The
linear multipliers are therefore 2 and −3. The image of
√
3 6 x 0 each oriented gray unit square is a 2 × 3 rectangle of the
(M − λ I)X = √ = ; opposite orientation, so the area multiplier is −6.
6 2 y 0
For an eigenvector we can take 2.6. b. We first need to find the eigenvalues and eigenvec-
√ √ tors of M. We have tr M = 1, det M = −2, so the charac-
√
6 − 6 6 −2 teristic polynomial is λ 2 − λ − 2 = (λ − 2)(λ + 1). For
or = √ . λ = 2 the eigenvector equation is
−3 3 −3 6

For the eigenvalue λ = +2, the eigenvector equation is −1 1 x 0 x 1
(M − λ I)X = = , = .
√ 2 −2 y 0 y 1
−2 6 x 0
(M − λ I)X = √ = ;
6 −3 y 0 For λ = −1 the equation is
For an eigenvector we can take
2 1 x 0 x −1
√ √ (M − λ I)X = = , = .
√ 2 1 y 0 y 2
6 6 6 3
or = √ .
2 2 2 6 In the figure below (on the next page), the gray grid
consists of parallelograms that have their sides parallel to
2.6. The special grid for M is one built on the eigenvec- the vectors (1, 1) and (−1, 2). The black parallelograms
tors of M. are their images. Each one is made up of two gray paral-
2.6. a. The eigenvectors and eigenvalues of M were de- lelograms. The gray parallelogram with its lower left cor-
termined in the solution to Exercise 2.5a; they are ner at the origin has its sides oriented counterclockwise.
Its image has its sides oriented clockwise, so M reverses

2 −1 orientation. The area multiplier is therefore −2. In one
for λ = 2, for λ = −3. direction, a gray parallelogram is stretched double; in the
1 2

11
other, it is merely flipped. Thus the linear multipliers are 2.7. d. Using the various relations and the definition of
2 and −1. M8 we can write
y
x = y = 2u = 2u + 2v
and then
y = x− y = (4u − 2v)− 2u = 2u − 2v = 2(u+ v)− 2u = 2v.
This confirms that

2 2
M8 = .
0 2
x 2.8. a. We have tr Rθ = 2 cos θ and det Rθ = 1. Therefore
the characteristic equation is λ 2 − (2 cos θ )λ + 1 = 0 and
its roots are
√ q
2 cos θ ± 4 cos2 θ − 4
λ= = cos θ ± −(1 − cos2 θ )
p2
= cos θ ± − sin2 θ = cos θ ± isin θ .
Thus the eigenvalues of Rθ are cos θ ± isin θ = e±iθ .
2.7. a. The image of M8 − 2I consists of all vectors X of
the form 2.8. b. The addition formulas for the sine and cosine
functions establish this equality. We have
2 −2 u 2u − 2v x
X = (M8 − 2I)U = = = . cos θ − sin θ r cos α
2 −2 v 2u − 2v y
sin θ cos θ r sin α

Points in the image satisfy the condition y = x; that is, r(cos θ cos α − sin θ sin α )
=
the image is the line y = x in the (x, y)-plane and thus is r(sin θ cos α + cos θ sin α )
1-dimensional. The kernel of M8 − 2I is, by definition, the
r cos(θ + α )
set of eigenvectors of M8 with eigenvalue 2. According to = .
r sin(θ + α )
the rank–nullity theorem (text pp. 46–47, applied to the
map M8 − 2I), The computation says Rθ maps the line that makes the
angle α with the positive u-axis to the line that makes the
2 = dim(source) = dim(image) + dim(kernel) angle α + θ with that axis. If θ 6= nπ , these two lines
= 1 + dim(kernel). are different; in particular, Rθ can map no line to a real
multiple of itself. This means Rθ has no real eigenvalues.
Hence dim(kernel) = 1; that is, the set of eigenvectors has 2.8. c. For the eigenvalue λ = cos θ + i sin θ , the eigen-
dimension 1, not 2. vector equation is
2.7. b. Straightforward calculation shows M8 (u) = 2x
−i sin θ − sin θ x
and M8 (v) = 2x + 2y. Thus u is an eigenvector for M8 (Rθ − λ I)X =
sin θ −i sin θ y
with eigenvalue 2.
−i −1 x 0
2.7. c. Suppose the vector W has coordinates (u, v) based = sin θ = .
1 −i y 0
on the standard basis vectors e1 and e2 and has coordinates
(u, v) based on u and v. That is i
1
1 0 1 1 For the eigenvalue λ = cos θ − i sin θ , the eigenvector
u +v =W =u +v .
0 1 1 0 equation is

i sin θ − sin θ x i −1 x 0
From these equations we can read off the following: = sin θ = .
sin θ i sin θ y 1 i y 0
u = u + v and v = u.
−i
The same relation holds for the x- and y-variables. 1

Note that if θ = nπ , then sin θ = 0 so these eigenvector If x → 0− , then (x, y) is in the third quadrant (q > 0) and
equations reduce to 0 = 0.
π π
2.9. Every matrix equivalent to D = λ I has the form
θ (x, y) = arctan(q) − π → + − π = −
2 2
G−1 DG = λ G−1 IG = λ G−1 G = λ I = D, because q = y/x → +∞. This establishes the continuity of
θ for y < 0.
as claimed.
2.10. b. The spiral-ramp graph of z = θ (x, y) is shown in
2.10. a. We assume that the usual function arctan(q) has the text on page 430.
the graph
2.11. a. Because u1 and u2 are eigenvectors, they are
Π
2
nonzero. Therefore, if they were not linearly independent,
we could write u2 = ku1 , k 6= 0. But then
λ2 u2 = M(u2 ) = M(ku1 ) = kM(u1 ) = kλ1 u1 = λ1 u2 .
Because λ2 6= λ1 , we have u2 = 0, a contradiction.

Π
-
2
2.11. b. It is a basic result of linear algebra that any
In particular, note that arctan(q) has the same sign as q square matrix whose columns are linearly independent is
and arctan(q) → ±π /2 as q → ±∞. invertible. Suppose we write G as a a row of column vec-
For clarity, write arctan(y/x) as θ (x, y). With q = y/x, tors and G−1 as a column of row vectors:

we can consider the definition of θ to be −1 r
 G = c1 c2 and G = 1 ;
r2

 arctan(q), first and fourth quadrants,

arctan(q) + π , second quadrant,
By definition of an inverse, the following row–column
θ (x, y) =


 arctan(q) − π , third quadrant, products (i.e., dot products) hold:
 (
±π /2, positive, negative y-axis, resp.
1 if i = j,
ri c j = δi j =
Thus, to check continuity of θ on the positive y-axis, we 0 otherwise.
must therefore show
Hence
π
θ (x, y) → θ (0, y) = + as x → 0.
2 MG = M c1 c2 = Mc1 Mc2

If x → 0+ (i.e., x approaches 0 through positive values), = λ 1 c1 λ 2 c2 ,
then (x, y) is in the first quadrant (q > 0) and
r1 λ 1 r1 c1 λ 2 r1 c2
−1
G MG = λ 1 c1 λ 2 c2 =
π r2 λ 1 r2 c1 λ 2 r2 c2
θ (x, y) = arctan(z) → + because q = y/x → +∞.
2 λ1 0
= .
0 λ2
If, on the other hand, x → 0− , then (x, y) is in the second
quadrant (q < 0) and
2.12. a. According to the discussion after Theorem 2.2
π π (text p. 36), if
θ (x, y) = arctan(x) + π → − + π = +
2 2 a b
M=
c d
because q = y/x → −∞. This establishes the continuity of
θ for y > 0. has a repeated eigenvalue λ , then
For θ to be continuous on the negative y-axis, we must
have a+d −a2 + 2ad − d 2
π λ= and bc = .
θ (x, y) → θ (0, y) = − as x → 0. 2 4
2
If x → 0+ , then (x, y) is in the fourth quadrant (q < 0) and If we now follow the suggestion and write

π a−λ b
θ (x, y) = arctan(q) → − because q = y/x → −∞. (M − λ I)e = , (M − λ I)e = ,
2
1
c 2
d−λ

13
then we have 2.12. b. By definition, the eigenvectors of M for eigen-

2 value λ are the elements of the kernel of M − λ I. By
a b a−λ a − aλ + bc
M(M − λ I)e1 = = hypothesis, there is just a single [linearly independent]
c d c ac − cλ + cd eigenvector, so the kernel of M − λ I is 1-dimensional. Ac-
cording to the rank–nullity theorem (text pp. 46–47), the
But (using (a − d)/2 = a − (a + d)/2 = a − λ ),
image of M − λ I is also 1-dimensional.
We established in part (a) that the image of M − λ I con-
a2 ad −a2 + 2ad − d 2
a2 − aλ + bc = a2 − − + sists of eigenvectors of M. Given that u is an eigenvector,
2 2 4
every nonzero multiple, in particular λ u, is also an eigen-
a2 d 2 a+d a−d vector (recall that λ 6= 0). Thus, by part (a) again, there is
= − =
4 4 2 2 some vector v for which
= λ (a − λ );
(M − λ I)v = λ u, or Mv = λ u + λ v.
ca cd a+d
ac − cλ + cd = ac − − + cd = c
2 2 2 If we assume u and v are linearly dependent, then v = ku
= λ c. for some k, and
In other words, λ u = (M − λ I)v = k(M − λ I)u = k · 0.

This is a contradiction, so u and v are linearly indepen-
a−λ a−λ
M =λ . dent.
c c
2.12. c. Any square matrix with linearly independent
Next we have columns, such as G, is invertible. If we write

a b b ab + bd − bλ
M(M − λ I)e2 = = G= u v and G−1 =
r
,
c d d −λ cb + d 2 − d λ s
But (and this time using −(a − d)/2 = d − (a + d)/2 = by analogy with the previous solution, then G−1 G = I im-
d − λ ), plies the following row–column products

a+d a+d ru = 1, rv = 0,
ab + bd − bλ = b(a + d) − b =b
2 2 su = 0, sv = 1.
= λ b;
Therefore
−a2 + 2ad − d 2 da d 2
cb + d 2 − d λ = + d2 − −
4

2 2
MG = M u v = Mu Mv = λ u λ u + λ v ;
a2 d 2 a+d −a + d
=− + = r λ ru λ ru + λ rv
G−1 MG = λu λu+λv =
4 4 2 2 s λ su λ su + λ sv
= λ (d − λ ).
λ λ
= = Sλ .
0 λ
Thus we have the second eigenvector equation:

b b 2.13. a. Equating the real and the imaginary parts of
M =λ .
d −λ d −λ
Mu + iMv = M(u + iv) = (a − ib)(u + iv)
Linearity of M and M − λ I establishes that every nonzero = au + bv + i(−bu + av)
(M − λ I)w is an eigenvector of M: if w = c1 e1 + c2 e2 ,
then gives
Mu = au + bv, Mv = −bu + av
M(M − λ I)w = M(M − λ I)(c1e1 + c2 e2 )
2.13. b. As in the two previous solutions, we write
= c1 M(M − λ I)(e1 ) + c2 M(M − λ I)(e2 )

= c1 λ (M − λ I)(e1 ) + c2λ (M − λ I)(e2 ) G= u v ,

= λ (M − λ I)w. MG = Mu Mv = au + bv −bu + av .

To compute GCa,b , we write out the components of the Because λ1 6= λ2 , this equality will hold only if x · x2 = 0,
columns of G as that is, only if x1 and x2 are orthogonal.

u v a −b 2.14. c. If a symmetric matrix has equal eigenvalues, then
GCa,b = 1 1 the discriminant of the characteristic polynomial must be
u 2 v2 b a
zero; that is, (a − c)2 + 4b2 = 0. This holds only if a − c =
au1 + bv1 −bu1 + av1
= b = 0, i.e., a = c and b = 0.
au2 + bv2 −bu2 + av2
2.15. a. Let h be the altitude of the parallelogram v ∧ w
= au + bv −bu + av = MG.
from the tip of the vector v to the side w.
2.13. c. Suppose u = 0; then the equation Mu = au + bv v
implies that bv = 0. Now b 6= 0, so we must have v = 0 θ
h
and thus u + iv = 0. But u + iv 6= 0 because it is an eigen- v∧w
vector of M; this contradiction implies u 6= 0.
An entirely similar argument establishes that v 6= 0. w
Following the suggestion, we now assume ru+ sv = 0 and
deduce Then h = kvk sin θ (implying that h < 0 when θ < 0) and
area(v ∧ w) = hkwk = kvkkwk sin θ .
0 = M(ru + sv) = r(au + bv) + s(−bu + av)
= a(ru + sv) + b(−su + rv) = b(−su + rv). 2.15. b. We have
But then −su + rv = 0 because b 6= 0. Therefore we also area2 (v ∧ w) = kvk2 kwk2 sin2 θ = kvk2kwk2 (1 − cos2 θ )
have
= kvk2 kwk2 − (v · w)2.
2 2
0 = r(ru + sv) − s(−su + rv) = (r + s )u,
2.15. c. A straightforward calculation gives
but since u 6= 0, we must have r2 + s2 = 0, implying that 2 2 2 2 2 2 2 2
r = s = 0. It follows that u and v are linearly independent. kvk kwk − (v · w) = (v1 + v2 )(w1 + w2 ) − (v1 w1 + v2 w2 )
2 2 2 2 2 2 2 2 2 2 2 2
2.13. d. As in the previous exercise, the matrix G whose = v1 w1 + v1 w2 + v2 w1 − v2w2 − v1 w1 − 2v1w1 v2 w2 − v2w2
columns are the linearly independent vectors u and v must = v21 w22 − 2v1 w2 v2 w1 + v22 w21 = (v1 w2 − v2w1 )2 .
be invertible. Therefore the equation MG = GCa,b leads
to Thus area(v ∧ w) = ±(v1 w2 − v2 w1 ), as claimed.
−1 −1
G MG = G GCa,b = Ca,b . 2.15. d. When v ∧w is the positively oriented unit square,
that is, when
2.14. a. In the discussion following Theorem 2.2 in the
text (p. 36), the equation for the eigenvalues implies the (v1 , v2 ) = v = (1, 0) and (w1 , w2 ) = w = (0, 1),
eigenvalues will be real if the discriminant of the char- we find that v w − v w = (1 × 1) − (0 × 0) = +1. Be-
1 2 2 1
acteristic polynomial, tr2 (M) − 4 det(M), is nonnegative. cause we require area(v ∧ w) = +1, we conclude we must
For the given matrix M, we have choose the plus sign in the formula for the area that is
found in part (c).
tr2 (M) − 4 det(M) = (a + c)2 − 4ac + 4b2
= a2 − 2ac + c2 + 4b2 2.16. a. Ordinary matrix multiplication gives

= (a − c)2 + 4b2 ≥ 0, MV = Mv Mw = V where V = v w .
so the eigenvalues are always real. From linear algebra we know that the determinant of a
product is the product of the determinants; therefore
2.14. b. Suppose xi is an eigenvector for M with eigen-
value λi (i = 1, 2), and λ1 6= λ2 . Let x†i denote the trans- detV = det MV = detM × detV.
pose of xi (as in the text itself); then, using the symmetry
M † = M, we have 2.16. b. By definition, M(v ∧ w) is the image of the par-
allelogram v ∧ w under the linear map M; by part (a), this
λ1 (x1 · x2 ) = λ1 x†1 x2 = (Mx1 )† x2 = x†1 M † x2 is the parallelogram M(v) ∧ M(w). Thus
= x†1 Mx2 = λ2 x†1 x2 = λ2 (x1 · x2 ). areaM(v ∧ w) = areaM(v) ∧ M(w).

15
According to Exercise 2.15, the last area equals 2.20. a. To determine the orientation and volume of P,
we construct the matrix V whose columns are the vectors
detV = detMV = det M × detV. of P in the order given and then find the determinant of V :
Finally, detV = areav ∧ w, so we have  
1 1 1
areaM(v ∧ w) = detM × areav ∧ w. V = 0 1 1 , detV = +1.
0 0 1
2.17. The aim here is to assume only the three “bulleted”
Therefore P has positive orientation and volume +1.
properties of D and then deduce that D must be the deter-
minant function. 2.20. b. The parallelepiped M(P) is given by
2.17. a. Following the suggestion, we use the bilinearity            
1 1 1 0 1 1
of D to write
M 0 ∧M 1 ∧M 1 =  0  ∧  0  ∧  2  .
D(x + y, x + y) = D(x + y, x) + D(x + y, y) 0 0 1 −1 −1 −1
= D(x, x) + D(y, x) + D(x, y) + D(y, y)
2.20. c. Orientation and volume of M(P) are given by the
for arbitrary x and y. By property 2, the left hand side
determinant of the matrix whose columns are the vectors
and the first and last terms on the right-hand side are zero.
of M(P):
This leaves
0 = D(y, x) + D(x, y),
0 1 1
1 1
from which it follows that D(y, x) = −D(x, y).
vol M(P) = 0 0
2 = −1 × = −2.
−1 −1 −1 0 2
2.17. b. By property 1, we know D(e1 , e2 ) = +1; by
part (a), D(e2 , e1 ) = −D(e1 , e2 ) = −1. This shows M(P) is negatively oriented.
2.17. c. This time, bilinearity gives
2.20. d. The volume multiplier of M is its determinant,
D(v, y) = D(v1 e1 + v2 e2 , w1 e1 + w2 e2 ) namely −2. Because
= v1 w1 D(e1 , e1 ) + v2 w1 D(e2 , e1 )
−2 = vol M(P) = det M × volP = −2 × 1
+ v1w2 D(e1 , e2 ) + v2 w2 D(e2 , e2 )
= (v1 w1 × 0) + (v2w1 × −1) we see that the volume multiplier does indeed account for
+ (v1w2 × +1) + (v2w2 × 0) volM(P) as found in the previous part.
= v1 w2 − v2 w1 . 2.21. Because z = x × y 6= 0, we conclude that x, y, and
z are linearly independent and, in the given order, provide
2.18. By definition, a coordinate frame that has the same orientation as the
standard basis for R3 .
2 −1 −1 5 5 2
(5, 2, −1) × (3, 4, 2) = , , Thus we can use x, y, and z as a basis to define a ma-
4 2 2 3 3 4 trix for the linear map L and, although this matrix will be
= (8, −13, 14); different from the matrix defined using the standard basis,
their determinants will have the same sign. The compo-
1 1 1 1 1 1
(1, 1, 1) × (1, 1, −1) = , , nents of the ith column of the new matrix are the coor-
1 −1 −1 1 1 1
dinates of the image of the ith basis element in terms of
= (−2, 2, 0). the new basis. The defining equations for L give us these
components; we have
2.19. We can obtain the signed volumes as scalar triple  
products. We have the following: 0 1 0
L = 1 0 0  and det L = +1.
a. vol = (5, 2, −1) × (3, 4, 2) · (1, 0 − 1) 0 0 −1
= (8, −13, 14) · (1, 0, −1) = −6;
b. vol = (1, 1, 1) × (1, 1, −1) · (1, −1, 1) Thus L has positive determinant. Moreover,
= (−2, 2, 0) · (1, −1, 1) = −4. L(x ∧ y) = L(x) ∧ L(y) = y ∧ x.

2.22. We use the formula that expresses the volume of a For the each of the remaining two identities, shown on
parallelepiped as a scalar triple product. In particular, the left below, just apply the previous argument to the ex-
pression to its right:
vol(x ∧ y ∧ z) = (x × y) · z = (x × y) · z + (x × y) · (ax + by)
D(z, y, x) = −D(x, y, z), D(x + z, y, x + z);
By definition, x × y is orthogonal to both x and y, and D(y, x, z) = −D(x, y, z), D(x + y, x + y, z).
hence is orthogonal to any linear combination of x and y:
x × y · (ax + by) = 0. 2.24. b. By part (a) and property 1,

D(e1 , e3 , e2 ) = −D(e1 , e2 , e3 ) = −1,
This leaves us with
D(e3 , e2 , e1 ) = −D(e1 , e2 , e3 ) = −1,
vol(x ∧ y ∧ z) = (x × y) · z = vol(x ∧ y ∧ z). D(e2 , e1 , e3 ) = −D(e1 , e2 , e3 ) = −1.
In the figure below, the “base” is put on the top for better Each of the other two cubes can be transformed into the
visibility. standard cube with two transpositions:
y
ax + by x D(e2 , e3 , e1 ) = −D(e2 , e1 , e3 ) = D(e1 , e2 , e3 ) = +1,
D(e3 , e1 , e2 ) = −D(e1 , e3 , e2 ) = D(e1 , e2 , e3 ) = +1.
z x∧y∧z
2.24. c. The 27 terms are
x∧y∧z
z x1 y1 z1 D(e1 , e1 , e1 ) + x1 y1 z2 D(e1 , e1 , e2 )
+ x1 y1 z3 D(e1 , e1 , e3 ) + x1y2 z1 D(e1 , e2 , e1 )
2.23. Because the orientation of x ∧ y ∧ z is given by the
+ x1 y2 z2 D(e1 , e2 , e2 ) + x1y2 z3 D(e1 , e2 , e3 )
sign of the determinant of the 3 × 3 matrix whose rows are
x, y, and z, in that order, and because the sign of a 3 × 3 + x1 y3 z1 D(e1 , e3 , e1 ) + x1y3 z2 D(e1 , e3 , e2 )
determinant changes when any two rows are interchanged + x1 y3 z3 D(e1 , e3 , e3 ) + x2y1 z1 D(e2 , e1 , e1 )
(or “transposed”), we see immediately that the three par- + x2 y1 z2 D(e2 , e1 , e2 ) + x2y1 z3 D(e2 , e1 , e3 )
allelepipeds
+ x2 y2 z1 D(e2 , e2 , e1 ) + x2y2 z2 D(e2 , e2 , e2 )
y ∧ x ∧ z, x ∧ z ∧ y, and z ∧ y ∧ x + x2 y2 z3 D(e2 , e2 , e3 ) + x2y3 z1 D(e2 , e3 , e1 )
+ x2 y3 z2 D(e2 , e3 , e2 ) + x2y3 z3 D(e2 , e3 , e3 )
have orientation opposite that of x ∧ y ∧ z. By contrast,
the cyclic permutations can be accomplished by a pair of + x3 y1 z1 D(e3 , e1 , e1 ) + x3y1 z2 D(e3 , e1 , e2 )
transpositions: + x3 y1 z3 D(e3 , e1 , e3 ) + x3y2 z1 D(e3 , e2 , e1 )
+ x3 y2 z2 D(e3 , e2 , e2 ) + x3y2 z3 D(e3 , e2 , e3 )
x ∧ y ∧ z → y ∧ x ∧ z → y ∧ z ∧ x,
+ x3 y3 z1 D(e3 , e3 , e1 ) + x3y3 z2 D(e3 , e3 , e2 )
x ∧ y ∧ z → x ∧ z ∧ y → z ∧ x ∧ y.
+ x3 y3 z3 D(e3 , e3 , e3 )
Hence the two parallelepipeds on the right have the same
orientation as x ∧ y ∧ z. 2.24. d. Consider the nonzero terms xi y j zk D(ei , e j , ek );
by property 2, the three indices i, j, k must be different.
2.24. a. We first show D(x, z, y) = −D(x, y, z). The ap-
This makes three choices for i, two for j, and just one
proach is entirely similar to that taken in the solution to
for k, a total of six. The remaining 21 terms must be zero.
Exercise 2.17.a. The multilinearity of D (bulleted prop-
From the list above, and using the values of D(ei , e j , ek )
erty 3) implies
established in part (b), the nonzero terms are
D(x, y + z, y + z) = D(x, y + z, y) + D(x, y + z, z) x y z D(e , e , e ) = x y z × (+1) = x y z ,
1 2 3 1 2 3 1 2 3 1 2 3
= D(x, y, y) + D(x, z, y) x1 y3 z2 D(e1 , e3 , e2 ) = x1 y3 z2 × (−1) = −x1 y3 z2 ,
+ D(x, y, z) + D(x, z, z)
x2 y1 z3 D(e2 , e1 , e3 ) = x2 y1 z3 × (−1) = −x2 y1 z3 ,
By property 2, the left-hand side is zero, and so are the x2 y3 z1 D(e2 , e3 , e1 ) = x2 y3 z1 × (+1) = x2 y3 z1 ,
first and fourth terms on the right-hand side. This leaves x3 y1 z2 D(e3 , e1 , e2 ) = x3 y1 z2 × (+1) = x3 y1 z2 ,
0 = D(x, z, y) + D(x, y, z), or D(x, z, y) = −D(x, y, z). x3 y2 z1 D(e3 , e2 , e1 ) = x3 y2 z1 × (−1) = −x3 y2 z1 .

17
2.24. e. By definition, 2.27. Following the suggestion, we write

 
x1 y1 z1 x·y x·y
x1 y2 z3 + y1z2 x3 + z1 x2 y3 0 ≤ z·z = x− y · x− y
det x2 y2 z2  = kyk2 kyk2
−(x1 z2 y3 + y1 x2 z3 + z1 y2 x3 );
x3 y3 z3 x·y (x · y)2
= x·x−2 x · y + y·y
this equals the sum of the six terms listed in part (d). kyk2 kyk4
(x · y)2 (x · y)2
2.25. The multilinearity of D (property 3) gives = kxk2 − 2 + kyk2
kyk2 kyk4
D(x, y, z) = D(x, y, z + α x + β y) (x · y)2
= kxk2 − .
= D(x, y, z) + α D(x, y, x) + β D(x, y, y) kyk2
2 2 2
Because the second and third terms on the right have This implies (x · y) ≤ kxk kyk .
a repeated input, they are zero by property 2; hence 2.28. a. The given permutation is a transposition τ =
D(x, y, z) = D(x, y, z). τk,l ; that is, τ (k) = l, τ (l) = k, while τ (i) = i for every
2.26. a. We have i 6= k, l. We write this transposition of the vectors vi in the
form
1 0 2 −1 k l
(. . . , vl , . . . , vk , . . .);
P1 = ∧ , P2 = ∧ ,
2 −1 1 1
k
here vl indicates the vector vl in the kth position and . . .
1 1
P3 = ∧ . indicates the vectors vi that are left in place. With this
1 0
notation we can write the equality
2.26. b. We have D(vτ (1) , . . . , vτ (n) ) = −D(v1 , . . . , vn ),

1 0 2 −1 that is to be proven in the form
areaP1 = = −1, area P2 = =3
2 −1 1 1 k l k l
D(. . . , vl , . . . , vk , . . .) = −D(. . . , vk , . . . , vl , . . .).
1 1

area P3 = = −1;
1 0 The proof then mimics the earlier proofs when n = 2 or 3
p √ (Exercises 2.17 and 2.24); that is, we use the multilinear-
therefore, area P = (−1)2 + 32 + (−1)2 = 11. ity of D to write
2.26. c. We write k l
D(. . . , vk + vl , . . . , vk + vl , . . .)
v †
k l k l
V= v w ; then V † = , = D(. . . , vk , . . . , vk , . . .) + D(. . ., vl , . . . , vk , . . .)
w†
k l k l
+ D(. . . , vk , . . . , vl , . . .) + D(. . ., vl , . . . , vl , . . .).
where Let v† and w† are row vectors. Because V † has two
†
rows and V has two columns, V V will have two rows By property 2, the left-hand side and the first and fourth
and two columns. Ordinary row-by-column matrix multi- terms on the right-hand side are zero; this leaves
plication gives k l k l
† 0 = D(. . . , vl , . . . , vk , . . .) + D(. . ., vk , . . . , vl , . . .),
v v v† w v·v v·w
V †V = = . or
w† v w† w w·v w·w
k l k l
D(. . . , vl , . . . , vk , . . .) = −D(. . . , vk , . . . , vl , . . .),
Because v · v = kvk2 and similarly for w, we see
as we were to prove.
det V †V = kvk2kwk2 − (v · w)2.
2.28. b. Suppose the given permutation π can be written
We have already seen in Exercise 2.15.b that the expres- as a product of N transpositions; then sgn π = (−1) .
N
2
sion on the right is area (P). In the present case, Since each transposition of its inputs changes the sign
of D, and (eπ (1) , eπ (2) , . . . , eπ (n) ) can be transformed into
kvk2 = 6, kwk2 = 2, v · w = −1, (e1 , e2 , . . . , en ) by N transpositions, we find
q √
so area(P) = kvk2kwk2 − (v · w)2 = 11, D(eπ (1) , eπ (2) , . . . , eπ (n) ) = (−1)N D(e1 , e2 , . . . , en )
= sgn π × 1 = sgn π .
as required.

2.28. c. If the map π : {1, 2, . . . , n} → {1, 2, . . . , n} is not permutation, three products of pairs transpositions on dis-
a permutation, then it is not 1–1. Hence at least two of the tinct elements, eight and cyclic permutations of any three
values π (1), . . . , π (n) are equal. By property 2, elements:
{1, 2, 3, 4}, {2, 1, 4, 3}, {3, 4, 1, 2}, {4, 3, 2, 1},
D(eπ (1) , eπ (2) , . . . , eπ (n) ) = 0.
{3, 1, 2, 4}, {2, 3, 1, 4}, {4, 1, 3, 2}, {2, 4, 3, 1}
{4, 2, 1, 3}, {3, 2, 4, 1}, {1, 4, 2, 3}, {1, 3, 4, 2}.
2.28. d. To expand D(v1 , v2 , . . . , vn ) into nn distinct terms
using the expressions If we now transpose the first two elements of each of these
quadruples, we get distinct odd permutations:
v1 = v11 e1 + v21e2 + · · · + vn1 en , {2, 1, 3, 4}, {1, 2, 4, 3}, {4, 3, 1, 2}, {3, 4, 2, 1},
v2 = v12 e1 + v22e2 + · · · + vn2 en , {1, 3, 2, 4}, {3, 2, 1, 4}, {1, 4, 3, 2}, {4, 2, 3, 1}
.. {2, 4, 1, 3}, {2, 3, 4, 1}, {4, 1, 2, 3}, {3, 1, 4, 2}.
.
vn = v1n e1 + v2ne2 + · · · + vnn en , If the 4 × 4 matrix is V = (vi j ), then the terms of the de-
terminant are
and the multilinearity of D, we choose one term from the
expression for v1 and make it the first argument of D; then v11 v22 v33 v44 + v21 v12 v43 v34 + v31 v42 v13 v2,4 + · · ·
one term from the expression for v2 and make it the sec- − (v21 v12 v33 v44 + v11 v22 v43 v34 + v41 v32 v13 v24 + · · · ).
ond argument of D; and so on, choosing one term in the
expression for vn the nth argument of D. This yields one 2.30. Let Aπ = (sgn π )vπ (1),1 vπ (2),2 · · · vπ (n),n be a term
of the nn such choices. in detV . Because vi j = 0 for every i > j, Aπ 6= 0 only if
If we let π (i) be the first index of the term just chosen in π ( j) ≤ j for every j = 1, 2, . . . , n. But
the expression for vi , where i = 1, 2, . . . , n, then the term π (1) ≤ 1 =⇒ π (1) = 1,
just constructed is π (1) = 1, π (2) ≤ 2 =⇒ π (2) = 2,
π (1) = 1, π (2) = 2, π (3) ≤ 3 =⇒ π (3) = 3,
D(vπ (1),1 eπ (1) , vπ (2),2 eπ (2) , . . . , vπ (n),n eπ (n) )
etc.
= vπ (1),1 vπ (2),2 · · · vπ (n),n D(eπ (1) , eπ (2) , . . . , eπ (n) ).
Thus Aπ = 0 unless π is the identity permutation, so detV
The “choice” map π : {1, 2, . . ., n} → {1, 2, . . . , n} here is reduces to a single nonzero term, namely
any of the nn possibilities, not necessarily a permutation. v11 v22 · · · vnn ,
2.28. e. Parts (b) and (c) show that the product of the diagonal elements of V .
( 2.31. When we compute det M1 using the formula given
sgn π = ±1, π in Sn , after Exercise 2.28, the position of the 2 × 2 zero matrices
D(eπ (1) , eπ (2) , . . . , eπ (n) ) =
0, otherwise, in M1 implies that the only permutations π that can yield
a nonzero term have
where Sn is the group of permutations on n elements.
π (1) = 1 or 2, π (3) = 3 or 4,
2.28. f. By part (e), the expansion of D(v1 , v2 , . . . , vn ) π (2) = 1 or 2, π (4) = 3 or 4.
takes the following form:
An examination of the 24 permutations listed in the solu-
D(v1 , v2 , . . . , vn ) tion to Exercise 2.29 reveals that only two even permuta-
tions and two odd satisfy this condition; they give
= ∑ vπ (1),1 vπ (2),2 · · · vπ (n),n D(eπ (1) , eπ (2) , . . . , eπ (n) )
det M1 = v11 v22 v33 v44 − v21 v12 v33 v44
π in Sn
= ∑ (sgn π ) vπ (1),1 vπ (2),2 · · · vπ (n),n . − v11 v22 v43 v34 + v21 v12 v43 v34
π in Sn = a11 a22 b11 b22 − a21 a12 b11 b22
− a11 a22 b21 b12 + a21 a12 b21 b12
2.29. The key to the solution is to determine sgn π for = det A det B.
each π in S4 . This can be done systematically. One way is
to list first the twelve even permutations (which constitute The positions of A and B in M1 lead us to set vi j = ai j
the “alternating subgroup” A4 of S4 . They are the identity when i, j ≤ 2 and vi j = bi−2, j−2 when 3 ≤ i, j.

19
When we compute det M2 , the position of the 2 × 2 zero Because the same transpositions that produce π will also
matrices now implies that the only permutations π that can produce σ (when written in the opposite order), we see
yield a nonzero term have sgn σ = sgn π . Moreover, because π has unique inverse
in Sn , σ will run over all elements of Sn as π runs over all
π (1) = 3 or 4, π (3) = 1 or 2, elements of Sn , and we have
π (2) = 3 or 4, π (4) = 1 or 2.
detA† = ∑ (sgn π ) a1,π (1) a1,π (2) · · · an,π (n)
Once again only two even and two odd permutations in π in Sn
the list of 24 given in the solution to Exercise 2.29 satisfy = ∑ (sgn σ ) aσ (1),1 aσ (2),2 · · · aσ (n),n = det A.
the condition, and they give σ in Sn
detM2 = v31 v42 v13 v24 − v41 v32 v13 v24

2.34. Because our formula for det A permutes row in-
− v31 v42 v23 v14 + v41 v32 v23 v14 dices instead of column indices, it is easier to prove the
= b11 b22 a11 a22 − b21 b12 a11 a22 expansion by minors along a column index, J. That is, we
− b11 b22 a21 a12 + b21 b12 a21 a12 first show
= detA detB. detA = (−1)1+J a1,J det M1,J + (−1)2+J det M2,J
This time the positions of A and B in M2 lead us to set + · · · + (−1)n+J det Mn,J .
vi j = ai, j−2 when i = 1, 2, j = 3, 4 and vi j = bi−2, j when
i = 3, 4, j = 1, 2. In the formula for det A, move the factor aπ (J),J that occurs
When we compute M3 , vπ (i),i = 0 if π (i) = 3 or 4. But in each term to the beginning of the product, as follows:
every permutation π is onto (i.e., π (i) = 3 for some i), so
aπ (J),J · aπ (1),1 aπ (2),2 · · · a\
π (J),J · · · aπ (n),n .
every product vπ (1),1 vπ (2),2vπ (3),3 vπ (4),4 is zero, implying
det M3 = 0. The circumflex indicates that the factor is missing (in this
2.32. Suppose the elements of column J in A are all zero. case, because it has been moved elsewhere). Now break
Each term of down the sum defining detA into separate sums in which
π (J) takes the values 1, 2, . . . , I, . . . , n, in turn:
det A = ∑ (sgn π ) aπ (1),1 aπ (2),2 · · · aπ (n),n
π in Sn det A = a1,J ∑ (sgn π ) aπ (1),1 aπ (2),2 · · · ad 1,J · · · aπ (n),n
π (J)=1
contains the factor vπ (J),J = 0, so each term is zero and
hence det A = 0. + ···+
Suppose instead that the elements of row I are all zero. + aI,J ∑ (sgn π ) aπ (1),1 aπ (2),2 · · · ac
I,J · · · aπ (n),n
For each permutation π , choose jπ so that π ( jπ ) = I. This π (J)=I
can be one because π is onto. Then the term + ···+
(sgn π ) aπ (1),1 aπ (2),2 · · · aπ (n),n + an,J ∑ (sgn π ) aπ (1),1 aπ (2),2 · · · ad
n,J · · · aπ (n),n .
π (J)=n
contains the factor vI, jπ = 0, so detA = 0 once again.
Consider the Ith summation; no factor aπ ( j), j comes from
2.33. If A = (ai j ), then A† = (bi j ) where bi j = a ji . We either the Ith row or the Jth column of A; in other words,
then have all factors lie in the minor MI,J . In fact, this sum resembles
detMI,J , except that the permutations π in the formula are
det A† = ∑ (sgn π ) bπ (1),1 bπ (2),2 · · · bπ (n),n
in Sn instead of Sn−1 .
π in Sn
The remedy is to convert each π in Sn to an appropriate
= ∑ (sgn π ) a1,π (1) a1,π (2) · · · an,π (n) . πI in Sn−1 by
π in Sn
The last expression is not, formally at least, det A. How- πI ( j) = π ( j) for j 6= I;

ever, we can convert into the formal expression for det A and πI : {1, . . . , J,
b . . . , n} → {1, . . . , I,
b . . . , n}.
by introducing the inverse σ of the permutation π . The in-
verse satisfies the condition σ ( j) = i when j = π (i), and To establish the formula for expanding by minors along
thus the Jth column (stated at the beginning of this solution),
(σ ( j), j) = (i, π (i)). all that remains is to prove sgn π = (−1)I+J sgn πI . It will

be helpful to use the following explicit description of the Because v · v = kvk2 and w · w = kwk2, we have
action of a permutation σ on a set {1, 2, 3, . . . , n − 1, n}:
detV †V = kvk2kwk2 − (v · w)2.
1 2 3 ··· n−1 n
σ (1) σ (2) σ (3) · · · σ (n − 1) σ (n) 2.36. If v1 ∧v2 ∧· · · ∧vn is an n-parallelepiped in Rn , and
The following is the product of J − 1 successive transposi-
V = v1 v2 · · · vn ,
tion (that move J to the first position), followed by the the n×n
permutation πI coupled with the assignment J → I, then
followed by a further I − 1 transpositions (that move I out then, by Definition 2.5 (text page 46),
to the Jth position).
vol(v1 ∧ v2 ∧ · · · ∧ vn ) = detV.
 
1 2 ··· J −1 J ··· n
 1 2 ··· J J −1 ··· n  Therefore, because detAM = detA det M for square matri-
  ces A and M and detV † = detV , it follows that
 .. .. .. .. .. .. .. 
 . . . . . . . 
 
 J 1 ··· J −2 J −1 ··· n  vol2 (v1 ∧v2 ∧· · ·∧vn ) = (detV )2 = detV † detV = detV †V .
 
 I πI (1) ··· πI (J − 2) πI (J − 1) ··· πI (n)
π (1) I ··· πI (J − 2) πI (J − 1) ··· πI (n)
 I 
 . .. .. .. .. .. .. 
2.37. The calculation here is essentially the same as in
 .. . . . 
 . . . 
Exercises 2.35 and 2.26.c. We have
πI (1) πI (2) ··· I πI (J − 1) ··· πI (n)  †
πI (1) πI (2) ··· πI (J − 1) I ··· πI (n) v1
 †
The result is precisely the permutation π , demonstrating V = v1 v2 v3 , V =  † 
v2  .
n×3 3×n
that
v†3
sgn π = (−1)I−1+J−1 sgn πI = (−1)I+J sgn πI .
Therefore, with the usual row-by-column multiplications
By what has been said, this establishes the expansion by and their identification with dot products, we get
minors along a column.  † 
v1 v1 v†1 v2 v†1 v3  
To establish expansion by minors along a row of A, we   v1 · v1 v1 · v2 v1 · v3
can use the fact that a matrix and its transpose have the V †V =  † † †  
v2 v1 v2 v2 v2 v3  = v2 · v1 v2 · v2 v2 · v3 .

3×3
same determinant. Moreover, when Mi j is the minor ob- v 3 · v 1 v 3 · v 2 v 3 · v 3
tained by deleting the ith row and jth column of A, then v†3 v1 v†3 v2 v†3 v3
M †ji is the minor obtained by deleting the ith row and jth
column of A† . Thus if we expand detA† by minors along 2.38. When a1 , . . . , ak−1 are arbitrary real numbers, the
the Ith column, we get set L of vectors in Rn of the form
det A = detA† a1 v1 + · · · + ak−1vk−1
1+I † n+I †
= (−1) detM1,I + · · · + (−1) det Mn,I
is, by definition, the linear subspace spanned by the vec-
I+1
= (−1) detMI,1 + · · · + (−1)I+n det MI,n , tors v1 , . . . , vk−1 . For any vector w, the vectors
as required. w + a1v1 + · · · + ak−1vk−1
2.35. The argument here is essentially the same as the
one in the solution to Exercise 2.26.c. We take v and w define a hyperplane parallel to L. Because −z is in L
to be n × 1 column vectors, so v† and w† are 1 × n row whenever z is in L, the vectors
vectors and we set w − a1v1 − · · · − ak−1vk−1
†
† v
V = v w , V = † . also form a hyperplane parallel to L. Finally, notice that
n×2 2×n w
the parallel hyperplane
Ordinary row-by-column matrix multiplication gives
† vk − a1 v1 − · · · − ak−1vk−1
v v v† w v·v v·w
V †V = = . contains vk , because we can take a1 = · · · = ak−1 = 0.
2×2 w† v w† w w·v w·w

21
2.39. The discussion following Exercise 2.35 on pages 2.41. b. To solve for w and x, the matrix equation we use
67–68 in the text indicates that h is to be found as is
−3 1 w −5u − 3v
= .
h = v3 − a1 v1 − a2 v2 , 6 −2 x −3u − 2v
This time the matrix fails to be invertible, so we cannot
where a1 and a2 satisfy
solve for w and x in terms of u and v.
−1
a1 v · v v1 · v2 v1 · v3 2.41. c. To solve for u and x, the matrix equation we use
= 1 1 .
a2 v2 · v1 v2 · v2 v1 · v3 is
5 1 u −3v + 3w
= .
3 −2 x −2v − 6w
2.39. a. In this case, h = v3 − v1 = (−1, 1, 1, 0), because
The matrix is once again invertible, and we can write
a1 1 7 −3 2 1
= = . u 1 2 1 −3v + 3w
a2 5 −3 2 3 0 = ,
x 13 3 −5 −2v − 6w
or
2.39. b. This time h = v3 + 17 8 1
13 v1 − 13 v2 = 13 (1, 5, 1, −8), 8
because u = − 13 v,
1
x = 13 v + 3w.
a1 1 7 −1 −2 1 −17
= = .
a2 13 −1 2 3 13 8 2.42. a. The kernel of L consists of triples (u, v, w) that
satisfy the three equations
2.40. In the rank–nullity theorem for a linear map, the
u + v = 0, v + w = 0, u − w = 0.
dimension of the source is the number of columns in its
matrix, while its rank is the number of linearly indepen- Any v = −u and w = u with u arbitrary will satisfy these
dent rows. equations. Because there is one free variable (i.e., “u”),
dim ker L = 1. By fixing u = 1 we get (1, −1, 1) as a basis
2.40. a. Rank = 2 because the two rows are linearly inde-
element.
pendent. Nullity = 1 because the source is 3-dimensional.
2.42. b. Let M : R1 → R2 be defined by
2.40. b. Rank = 4 and nullity = 0 because the four rows (
are linearly independent. v = −u, −1
M: M= .
2.40. c. Rank = 2 because there are just two linearly in- w = u; 1
dependent rows. Nullity = 0 because the source is 2- Then the graph of M in R1 × R2 = R3 is the set of vectors
dimensional. (u, v, w) = (u, −u, u); this is obviously ker L as well. We
2.40. d. Rank = 3 and nullity = 0 because the three rows see p = 1, q = 2.
are linearly independent. 2.42. c. The equations are given in the solution to part (b).
2.41. a. We can write the equations as, for example, 2.42. d. The image f L is 2-dimensional, by the rank–
nullity theorem. The images of any two vectors not in the
5 3 u 3w − x
= , kernel, for example (1, 0, 0) and (0, 1, 0), provide a basis:
3 2 v −6w + 2x
(1, 0, 1), (1, 1, 0).
so the solution can be given using an inverse matrix, as
follows: 2.42. e. Because im L is a 2-dimensional subspace of R3 ,
−1 the map A must have a 2-dimensional graph in R3 ; thus
u 5 3 3w − x A : R2 → R1 ; j = 2, k = 1. Because
=
v 3 2 −6w + 2x
x − y = u + v − v − w = z,
2 −3 3w − x 24w − 8x
= = . we see that all points (x, y, z) in im L satisfy the equation
−3 5 −6w + 2x −39w + 13x
z = x − y. In other words, im L is the graph of
That is,
A : z = x − y; A = 1 −1 .
u = 24w − 8x,
2.42. f. The solution to part (e) gives the one equation in
v = −39w + 13x. two variables that defines A.

2.43. a. The second equation is just a multiple of the To show that the set G spans V , consider an arbitrary
first, so it provides no further restriction on the values of element v of V . Then there is some u in Rn for which
u and v beyond that given by the first equation. Solutions v = (u, L(u)). Because the vectors u1 , . . . , un are a basis
to the two equations can therefore be put into the form for Rn , they span Rn . Hence there are scalars a1 , . . . , an
(5 + 2v, v) where v is arbitrary. for which u = a1 u1 + · · · + anun . Therefore,
v
v = (u, L(u))
= (a1 u1 + · · · + anun , L(a1 u1 + · · · + anun ))
u
-5 5 = (a1 u1 + · · · + anun , a1 L(u1 ) + · · · + an L(un ))
-2 = (a1 u1 , a1 L(u1 )) + · · · + (an un , an L(un ))
= a1 (u1 , L(u1 )) + · · · + an (un , L(un ))
-4
= a1 v1 + · · · + an vn ,
-6
proving that G spans V .
2.43. b. The solution set is the graph of the function

u−5
v= .
2
The sketch above is the graph of this function (by con-
struction!).
2.43. c. The solution set here is (2v, v), with v arbitrary.
In the (u, v)-plane this set is the straight line through the
origin with slope 1/2; it is a translate of the solution set in
part (a).
2.44. a. For the given v1 and v2 we have
v1 +v2 = (u1 +u2 , L(u1 )+L(u2 )) = (u1 +u2 , L(u1 +u2 )),
using the linearity of L. Hence v1 + v2 is in V because it

has the form (w, L(w)) with w = u1 + u2 .
2.44. b. To show that cv is in V when v is in V and c is
an arbitrary scalar, we first write v = (u, L(u)) for some u
in Rn . Then
cv = (cu, cL(u)) = (cu, L(cu)),
again using the linearity of L. Hence cv is in V because it

has the form (w, L(w)) with w = cv.
2.44. c. We suppose the given v j satisfy the condition
c1 v1 +· · ·+cn vn = 0; we must show that c1 = · · · = cn = 0.
We have
0 = c1 v 1 + · · · + cn v n
= (c1 u1 + · · · + cn un , c1 L(u1 ) + · · · + cn L(vn ))
Hence c1 u1 + · · · + cn un = 0, but the vectors u1 , . . . , un

are linearly independent in Rn , so c1 = · · · = cn = 0, as
required. This shows that G is a linearly independent set
in Rn+p .

Approximations
Z 1
3.1. a. f =
1
xn dx = . 3.3. Lt f (x, y) = α x + β y + γ be the linear function, and
0 n+1 let the circular disk D have radius R and be centered at
Z
1 π 1 2 the point (x, y) = (p, q). To do the integration needed to
3.1. b. f = sin x dx = × 2 = .
π 0 π π find the average value of f over , it is helpful to use polar
coordinates based at the center of D:
3.1. c. The domain is the unit disk D so its area is π .
Therefore, x − p = r cos θ , y − q = r sin θ ,
1
ZZ Z
1 2π
Z 1 f = r(α cos θ + β sin θ ) + α p + β q + γ .
f= (x2 + y2 ) dx dy = dθ r2 · r dr
π D π 0 0 It is still the case that dx dy = r dr d θ ; thus
1 1 1 ZZ
= × 2π × = . πR f =2
(α x + β y + γ ) dx dy
π 4 2 D
Z R Z 2π
3.2. a. The coordinate ξ is halfway between a and b and = r2 ((α cos θ + β sin θ )
η is halfway between c and d; thus 0 0
+ r(α p + β q + γ ) dr d θ .
a+b c+d
(ξ , η ) = , . The integrals of cos θ and sin θ over a full period are zero,
2 2 so
Z R Z 2π
3.2. b. The value of f at the center of the rectangle R is π R2 f = r(α p + β q + γ ) dr d θ
0 0
ZR
a+b c+d α (a + b) + β (c + d) + 2γ
f , = . = 2π (α p + β q + γ ) r dr
2 2 2 0
R2
To determine the mean value f of f over R, we write = 2π (α p + β q + γ ) ×
2
Z b Z d
= π R2 (α p + β q + γ ).
(b − a)(d − c) f = (α x + β y + γ ) dy dx
a
d
c
Therefore f = α p + β q + γ , the value of f at the center
Z b
y2 of D.
= α xy + β + γ y dx
a 2 c of the integral is 1/(n + 1), and this equals
3.4. The value √
Z b
β (c + d) + 2γ cn when c = 1/ n n + 1.
= αx + (d − c) dx
a 2 3.5. The value of the integral is 2, and this equals π sin c
2 b when c = arcsin(2/π ).
x β (c + d) + 2γ
= α + x (d − c)
2 2 a 3.6. Using a standard factoring of bn+1 − an+1, we find
α (a + b) + β (c + d) + 2γ Z b
bn+1 − an+1
= (b − a)(d − c). n
c (b − a) = xn dx =
2 a n+1
Hence f = 12 (α (a + b) + β (c + d) + 2γ ), and this agrees (b + ab + · · · + an)(b − a)
n n−1
= .
with the value of f at the center of R. n+1
23

24 SOLUTIONS: CHAPTER 3. APPROXIMATIONS
Thus (n + 1)cn = bn + abn−1 + · + an, or But

s Z b Z b
n n−1 + · + an f (x)h(x) dx = − f (x)g(x) dx
n b + ab
c= . a a
n+1 Z b Z b
and h(x) dx = − g(x) dx,
To show that a < c < b, first note that a < b implies a a
so Z b Z b
(n + 1)cn < bn + b · bn−1 + · · · + bn = (n + 1)bn.
f (x)g(x) dx = f (c) g(x) dx,
Because b and c are positive, we conclude c < b. We also a a
have as we had to show.
(n + 1)cn > an + a · an−1 + · · · + an = (n + 1)an, 3.9. b. If f (x) = g(x) = x on [−1, 1], then
Z 1 Z 1
and hence a < c. f (c) g(x) dx = f (c) x dx = 0
−1 −1
3.7. From the solution to Exercise 3.1.c we see that the
value of the integral is π /2. Because this is to equal regardless of the value of c. But
π (α 2 + β 2 ), we can
√ choose (α , β ) to be any point on the Z 1 Z 1 1
circle of radius 1/ 2 centered at the origin. f (x)g(x) dx = x2 dx = ,
−1 −1 3
3.8. a. The graph of y = f (x) is the semicircle of radius r
centered at the origin in the (x, y)-plane. The value of the so Z b Z b
integral is the area under this graph, i.e., π r2 /2. Thus f (x)g(x) dx 6= f (c) g(x) dx
a a
π r2 p π r 2 for any c. There is no contradiction here because the sign
= 2r r2 − c2 , or = r 2 − c2 .
2 4 of g(x) does not remain constant on [−1, 1].
Hence 3.10. We cannot mimic the proof of the one-variable
q case (Theorem 3.2), because the fundamental theorem
c2 = r2 (1 − π 2/16), or c = ±r 1 − π 2/16. has no analogue for double integrals. We can, however,
mimic the proof of the generalized law of the mean (The-
3.8. b. The horizontal line is at the height orem 3.3).
Thus let m and M be the minimum and maximum val-
q
ues of F(x, y) on D; then
f (c) = r2 − r2 (1 − π 2/16) = rπ /4.
ZZ ZZ
y m × area(D) = m dx dy ≤ F(x, y) dx dy
y = f(x) ZZD D
y = f(c) ≤ M dx dy = M × area(D).
D
We can write this as

x ZZ
-r c r F(x, y) dx dy
D
m≤ ≤ M.
The hatched area above the horizontal line appears to be area(D)
about equal to the shaded area below the line, so the rect-
angle appears to contain the same area as the semicircle. Because F is continuous on D and D is connected, F takes
Because we can interpret integrals as areas, we expect on all values between its minimum and maximum; in par-
these two areas to be equal. ticular, there is a point (c, d) in D for which
ZZ
3.9. a. Let h(x) = −g(x); then h(x) ≥ 0 on [a, b] so the F(x, y) dx dy
generalized integral law of the mean in the form already F(c, d) = D
,
proven applies to h. Thus there is some c in [a, b] for area(D)
which Z Z
b b as was to be shown. If D is not connected, then F may not
f (x)h(x) dx = f (c) h(x) dx. take on all values between m and M. For example, let D
a a

25
consist of two disjoint closed circular disk, each of area 1, -1.0 -0.5 0.5 1.0
and let F = −1 on one and F = 1 on the other; then F is -14
-5. ´ 10
continuous on D. The value of the integral is 0, but there
is no point (c, d) in D for which F(c, d) = 0. -1. ´ 10-13
3.11. a. The terms of order 2 and 3 for Pn,100(∆x) are
already given in the text, and the computations have been -1.5 ´ 10-13
carried out for n = 2, 3. To compute the additional terms
we note -2. ´ 10-13
−3 · 5 −7/2 f (4) (100) −5 1 −1 The graph of R3,100 is too flat near the origin to be a
f (4) (x) = x , = 7 7 = ,
24 4! 2 10 256 × 106 parabola, but it does look like a multiple of (∆x)4 . The
3 · 5 · 7 −9/2 f (5) (100) 7 1 7 graph of R4,100 looks like a cubic, but it is similarly too
f (5) (x) = x , = 8 9 = .
25 5! 2 10 256 × 109 flat near the origin. It looks more like a multiple of (∆x)5 .
Thus the additional terms are The graph of R5,100 is even flatter than (∆x)4 ; it looks like
−(∆x)4 7(∆x)5 a (negative) multiple of (∆x)6 . To summarize,
and .
256 × 106 256 × 109 Rn,100(∆x) ≈ Cn (∆x)n+1 , n = 2, 3, 4, 5.
√
The values and the errors for 102 (i.e., ∆x = 2) are as It is in this sense that Rn,100 (∆x) = O(n + 1).
follows:
√ 3.11. c. The graphs just found suggest the value of Cn ,
Degree Sum Error = 102 − Sum namely, as the value of Rn,100 (∆x) when ∆x = 1. Thus, by
4 10.0995049375 8.62 × 10−10 inspection,
5 10.099504938375 −1.29 × 10−11 C3 = −4 × 10−9, C4 = 2.5 × 10−11, C5 = −2 × 10−13.
√
The corresponding results for 120 (i.e., ∆x = 20) are: (The text on page 84 gives C2 = 6.25 × 10−7.)
√ The graph of y = R3,100(∆x) is plotted below in gray;
Degree Sum Error = 120 − Sum
the graph of y = −4 × 10−9 (∆x)4 is shown dashed and
4 10.954375 7.62 × 10−5 shifted down by 0.1 × 10−9 to help make it visible.
5 10.9544625 −1.13 × 10−5
√ 5 n+1
-1.0 -0.5 0.5 1.0
When n = 4 the error√ for 102 is about 1/10 = 1/10
√ -9
-1. ´ 10
times the error for 120. When n = 5, the error
√ for 102
is about 1/106 = 1/10n+1 times the error for 120. -2. ´ 10-9
3.11. b. The graph of y = R2,100 (∆x) appears in the text
on page 84; the graphs for n = 3, 4, and 5 appear in that -3. ´ 10-9
order below. Note the vertical scales.
-4. ´ 10-9
-1.0 -0.5 0.5 1.0
The same scheme is used in the following figure, which
-1. ´ 10-9 show the graphs of
-2. ´ 10-9 y = R4,100 (∆x) and y = 2.7 × 10−11(∆x)5 .
-3. ´ 10-9
The revised value C4 = 2.7 × 10−11 gives a somewhat bet-
ter approximation; it is essentially the coefficient of the
-4. ´ 10-9 next power of ∆x in the Taylor expansion.
2. ´ 10-11 2. ´ 10-11
1. ´ 10-11 1. ´ 10-11
-1.0 -0.5 0.5 1.0 -1.0 -0.5 0.5 1.0

-1. ´ 10-11 -1. ´ 10-11
-2. ´ 10-11 -2. ´ 10-11

In the final figure y = −2 × 10−13(∆x)6 is plotted together 3.12. c. The graphs of y = Rn,1 (∆x) appear below, in the
with y = R5,100(∆x). order n = 1, 2, 3, 4. We see R1,1 looks parabolic, R2,1
looks cubic, R3,1 looks like a (negative) multiple of (∆x)4 ,
-1.0 -0.5 0.5 1.0 and R4,1 looks like a multiple of (∆x)5 . This demonstrates
-5. ´ 10-14 that Rn,1 (∆x) = O(n + 1), that is, that Rn,1 (∆x) looks like
a multiple of (∆x)n+1 when ∆x is small.
-1. ´ 10-13
-0.3 -0.2 -0.1 0.1 0.2 0.3

-1.5 ´ 10-13 -0.01
-0.02
-2. ´ 10-13
-0.03
3.12. a. We begin by computing derivatives of f (x) = -0.04
ln x and evaluating them at x = 1:
-0.05
f (x) = ln x, f (1) = 0,
f ′ (x) = x−1 , f ′ (1) = 1, 0.005
′′ −2 ′′
f (x) = −x , f (1) = −1,
′′′ −3
f (x) = 2x , f ′′′ (1) = 2, -0.3 -0.2 -0.1 0.1 0.2 0.3
f (4) (x) = −3!x−4, f (4) (1) = −3!.

-0.005
The polynomials Pn,1 (∆x) are therefore obtained by trun- -0.010

cating the following sum:
-0.3 -0.2 -0.1 0.1 0.2 0.3
(∆x)2 (∆x)3 (∆x)4
∆x − + − . -0.0005
2 3 4
-0.0010
3.12. b. We display the estimates in the following two ta- -0.0015
bles; the first is for ln 1.02.
-0.0020
Degree Sum Error = ln 1.02 − Sum
-0.0025
1 0.02 −1.97 × 10−4
2 0.0198 2.63 × 10−6 0.0004
3 0.019802667 −3.94 × 10−8 0.0002

4 0.01980262667 6.30 × 10−10
-0.3 -0.2 -0.1 0.1 0.2 0.3
The next is for ln 1.2.
-0.0002
Degree Sum Error = ln 1.2 − Sum
-0.0004
1 0.2 −1.77 × 10−2
2 0.18 2.32 × 10−3
3 0.182667 −3.45 × 10−4 3.13. As the text remarks (p. 81), the heart of the induc-
4 0.1822667 5.45 × 10−5 tion is an integration by parts of
The following table summarizes the relative size of the Z 1
two errors as a function of the degree n; we see the relative I= f (k+1) (a + t∆x)(1 − t)k dt
0
error is 1/10n+1.
To carry this out we take
Error in ln 1.02
Degree as a fraction of error in ln 1.2 u = f (k+1) (a + t∆x), dv = (1 − t)k dt;
1 1/102
2 1/103 then
3 1/104 −1
4 1/105 du = f (k+2) (a + t∆x) · ∆x, v= (1 − t)k+1.
k+1

27
Integration by parts gives In particular, ϕ (t)/t 2 is not bounded as t → 0, so it follows

1 that ϕ (t) 6= O(2).
−1
I= (1 − t)k+1 f (k+1) (a + t∆x) 3.16. We must show that ψ (u)/u p → 0 as u → 0 for any
k+1 0 p > 0. Let x = 1/|u|; then u p = ±1/x p and x → +∞ as
Z 1
∆x (k+1) u → 0. Therefore
+ f (a + t∆x)(1 − t)k+1 dt
k+1 0
Z 1 ψ (u) e−x xp
f (k+1) (a) ∆x lim p
= ± lim p
= ± lim x = 0
= + f (k+1) (a + t∆x)(1 − t)k+1 dt. u→0 u x→+∞ 1/x x→+∞ e
k+1 k+1 0
for any p > 0, as required. The graph of t = ψ (u) is sym-
Therefore metric around the y-axis.
(∆x)k+1 f (k+1) (a) t
Rk,a (∆x) = I= (∆x)k+1
k! (k + 1)!
Z 1
(∆x)k+2 0.2
+ f (k+1) (a + t∆x)(1 − t)k+1 dt
(k + 1)! 0 u
-1.0 -0.5 0.0 0.5 1.0
f (k+1) (a)
= (∆x)k+1 + Rk+1,a(∆x).
(k + 1)! Because 0 < e−x < 1 when ∞ > x > 0, we see
3.14. By hypothesis, the Taylor expansions for f and g 0 < ψ (u) < 1 when 0 < |u| < ∞.
at x = a are Because we can define ψ (0) = 0, we conclude that the im-
f (n) (a) f (n+1) (a + θ ∆x) age of ψ (u) on the t-axis is 0 ≤ t < 1, a half-open interval.
f (a + ∆x) = (∆x)n + (∆x)n+1 ,
n! (n + 1)! 3.17. Following the suggestion, we express ϕ (t)/t p → 0
g(n) (a) g(n+1)(a + θ ∆x) as t → 0 in the standard delta–epsilon form of analysis:
g(a + ∆x) = (∆x)n + (∆x)n+1 ; Given any ε > 0, there is a δ > 0 for which
n! (n + 1)!

ϕ (t)
therefore we get |t| < δ implies p < ε .

t
f (a + ∆x) f (n) (a) + f (n+1)(a + θ ∆x) ∆x/(n + 1) Hence
= (n) ,
g(a + ∆x) g (a) + g(n+1)(a + θ ∆x) ∆x/(n + 1) |t| < δ implies |ϕ (t)| < ε |t p |.
after cancelling factors (∆x)n /n! in the numerator and 3.18. a. On the domain u > 0 we can write ψ (u) = e−1/u
denominator. Because the factors f (n+1) (a + θ ∆x) and (i.e., the absolute value is unnecessary); then
g(n+1) (a + θ ∆x) remain bounded as ∆x → 0, we see
1
 ψ ′ (u) = e−1/u · 2 > 0
f (a + ∆x)  ∞ if g (n)
(a) = 0, u
→ f (a) so t = ψ (u) is invertible for all u > 0. One way to confirm
g(a + ∆x)  otherwise,
g(a) hat ϕ is the inverse is to derive it by solving t = e−1/u for
t. We have
as ∆x → 0.
1 1
3.15. Consider first ϕ (t)/t = t β , where β = α − 1. Be- e1/u = or = ln(1/t) = − lnt.
t u
cause 1 < α , we have 0 < β and thus
Thus the inverse is u = ϕ (t) = −1/ lnt on 0 < t < 1. We
ϕ (t) β extend ϕ to t = 0 as ϕ (0) = 0 by using 0 = ψ (0).
lim = lim t = 0.
t→0 t t→0
3.18. b. By Exercise 3.16, ψ (u) = o(p) for any p > 0.
This shows that ϕ (t) = o(1). Because q = 1/p is also positive, it is equally true that
2 γ
Now consider ϕ (t)/t = t , where γ = α − 2. Because ψ (u) = o(q) for all q > 0. By Exercise 3.17, we can ex-
α < 2, we have γ < 0 and thus press the condition ψ (u) = o(q) in the following way: For
a given ε < 1, there is a δ > 0 for which
ϕ (t)
lim = lim t γ = ∞. |u| < δ implies |ψ (u)| < |u|q .
t→0 t 2 t→0

Because we have u ≥ 0, ψ (u) ≥ 0, and q = 1/p, we can Therefore,

rewrite this as
S(∆x)
= aK + aK+1 ∆x + · · · + an (∆x)n−k → aK
0≤u<δ implies ψ (u) p < u. (∆x)K
Now choose some θ with 0 < θ < ψ (δ ), let 0 ≤ t < θ as ∆x → 0; hence
be arbitrary, and set u = ϕ (t); then we have
S(∆x) = f (a + ∆x) − Q(∆x) = O(K).
0 ≤ u = ϕ (t) < ϕ (θ ) < ϕ (ψ (δ )) = δ .
By contrast,
That is, 0 ≤ u < δ , so the condition ψ (u) = o(q) implies
S(∆x) aK
ψ (u) p < u or t p < ϕ (t) = + aK+1 + · · · + an(∆x)n−k−1 → ∞
(∆x)K+1 ∆x
for every 0 ≤ t < θ . as ∆x → 0; hence, by Lemma 3.1 (text p. 88),
3.19. Using mathematical induction we can prove the
S(∆x) = f (a + ∆x) − Q(∆x) 6= O(K + 1).
formula in the slightly refined version
(−1)k · 1 · 3 · · ·(2k − 1) 3.22. a. When f (x) = ex we have f (k) (x) = ex and

f (k+1) (x) = , k = 1, 2, . . . .
2k+1 xk+1/2 f (k) (0) = 1 for every k = 0, 1, . . .. Hence the coefficient
of xk in P(x) is 1/k! and we have
We must exclude k = 0 because of the factor (2k − 1). We
see from quick calculations, P(x) = 1 + x + 12 x2 + 16 x3 .
1 −1
f ′ (x) = and f ′′ (x) = ,
2 x1/2 3.22. b. On the interval −0.1 ≤ x ≤ 0.1, the graph of
22 x3/2
y = R(x), below, looks like a multiple of y = x4 and thus
that the second derivative does indeed fit the formula
indicates R(x) = O(4) but R(x) 6= O(5). Note the vertical
when k = 1. By induction we assume the formula is true
scale on this and every subsequent graph in this exercise.
for k and show it is true for k + 1. Thus
′ 4. ´ 10 -6
f (k+2) (x) = f (k+1) (x)

(−1)k · 1 · 3 · · ·(2k − 1) − 2k+1 ′
-6
3. ´ 10
= x 2
2k+1 2. ´ 10 -6
(−1)k · 1 · 3 · · ·(2k − 1)−(2k + 1) − 2k+3

= · x 2 1. ´ 10-6
2k+1 2
(−1)k+1 · 1 · 3 · · ·(2k − 1)(2k + 1)
= , -0.10 -0.05 0.05 0.10
2k+2 x(k+1)+1/2
which agrees with the formula for k + 2. 3.22. c. On the interval −0.1 ≤ x ≤ 0.1, the graph of
y = V1 (x), below, looks like a multiple of y = x3 and thus
3.20. Using big oh notation, Taylor’s theorem for degree
indicates V1 (x) = O(3) but V1 (x) 6= O(4).
n = 1 can be written (cf. the discussion following Def. 3.6
on p. 88 of the text)
0.00003
f (a + ∆x) = f (a) + f (a)∆x + O(2),

′ 0.00002
0.00001
or
∆y = f (a + ∆x) − f (a) = f ′ (a)∆x + O(2). -0.10 -0.05 0.05 0.10
-0.00001
This is the microscope equation with the nature of the ap- -0.00002
proximation clarified. -0.00003
3.21. The crucial difference between the proof of the the-

orem and the proof of the corollary is we can now assert 3.22. d. On the interval −0.1 ≤ x ≤ 0.1, the graph of
that y = V2 (x), below, looks like a multiple of y = x2 and thus
a0 = a1 = · · · = aK−1 = 0, aK 6= 0. indicates V2 (x) = O(2) but V2 (x) 6= O(3).

29
0.0015
3.23. c. Here the given function is a sum of one-variable
functions. For f (x) = x3 − 3x we have f (−1) = 2,
0.0010
f ′ (−1) = 0, f ′′ (−1) = −6 , so
x3 − 3x = 2 − 3(x + 1)2 + O(3)

0.0005
and x3 − 3x + y2 = 2 − 3(x + 1)2 + y2 + O(3).
-0.10 -0.05 0.05 0.10

3.23. d. The needed derivatives of f (x, y) = ln(x2 + y2 )
are
3.22. e. On the interval −0.1 ≤ x ≤ 0.1, the graph of
y = V3 (x), below, looks like a multiple of y = x and thus 2x 2y
fx = , fy = ,
indicates V3 (x) = O(1) but V3 (x) 6= O(2). x2 + y2 x2 + y2
0.010 2y2 − 2x2 −4xy 2x2 − 2y2
fxx = , f xy = , f yy = ;
(x2 + y2 )2 (x2 + y2 )2 (x2 + y2 )2
0.005
fx (1, 0) = 2, fy (1, 0) = 0,
fxx (1, 0) = −2, fxy (1, 0) = 0, fyy (1, 0) = 2.
-0.10 -0.05 0.05 0.10
-0.005
Because f (1, 0) = 0 we can now write
-0.010
ln(x2 + y2 ) = 2(x − 1) − (x − 1)2 + y2 + O(3).
3.22. f. The graph of y = V4 (x), below, underscores the 3.23. e. For f (x, y, z) = xyz, we have f (1, −2, 4) = −8
fact that V4 (0) 6= 0, although V4 (x) is bounded at and near and the derivatives are
x = 0. Thus V4 (x) = O(0) but V4 (x) 6= O(1). Notice that
V4 (x) ≈ 0.1 + kx4 for some k 6= 0. fx = yz, fy = xz, fz = xy,
0.15
fxy = z, fyz = x, fzx = y, fxx = fyy = fzz = 0;
0.10
fx (1, −2, 4) = −8, fy (1, −2, 4) = 4, fz (1, −2, 4) = −2,
0.05
fxy (1, −2, 4) = 4, fyz (1, −2, 4) = 1, fzx (1, −2, 4) = −2.
-1.0 -0.5 0.5 1.0

Therefore,
-0.05
xyz = −8 − 8(x − 1) + 4(y + 2) − 2(z − 4)
-0.10
+ 4(x − 1)(y + 2) + (y + 2)(z − 4) − 2(x − 1)(z − 4)
-0.15
+ O(3).
3.23. In the first two cases, the given function is a prod-
uct of one-variable functions; we construct its Taylor 3.23. f. This function follows the pattern of part (c); we
polynomial as the product of the Taylor polynomials of have
its factors. 1 − cos θ = 2 − 21 (θ − π )2,
x 1 2
3.23. a. We have e = 1 + x + 2 x + O(3) on the one hand v2 − (θ − π )2
and sin y = y + O(3) on the other; thus 1 − cos θ + 12 v2 = 2 + .
2

ex sin y = 1 + x + 12 x2 + O(3) y + O(3)
3.24. Because f (x, y) = (x2 + y2 )2 − (x2 + y2 ) is itself a
= y + xy + O(3).
polynomial of degree 4, it will be exactly equal to its Tay-
lor polynomial of degree 4 (at any chosen point). Simi-
3.23. b. We have cos x = 1 − 21 x2 + O(4); however, be- larly, the degree 2 Taylor polynomial for x2 will equal x2 ,
cause y ≈ π /2, we must write cos y = −(y − π /2) + O(3); and we can use this fact to build the degree 4 Taylor poly-
thus nomial for f (x, y) without computing a lot of derivatives.
To begin, it easy to see that
cos x cos y = 1 − 12 x2 + O(4) − (y − π /2) + O(3)
= −(y − π /2) + O(3).
1 2
1 1
4 + x− 2 + x− 2 = x2

is the degree 2 Taylor polynomial for x2 at x = 12 . For 3.27. We have the following Taylor expansions centered
clarity let us write x − 21 = X, y − 21 = Y ; then at ρ = ρ0 , θ = π /2, and ϕ = 0:
1
+ Y + Y 2 = y2 ρ = ρ0 + (ρ − ρ0),
4
cos θ = −(θ − π /2) + O(3),
1
is the degree 2 Taylor polynomial for y2 at y = 2 and sin θ = 1 − 21 (θ − π /2)2 + O(4),
Q(X,Y ) = 21 + X + Y + X 2 + Y 2 = x2 + y2 cos ϕ = 1 − 21 ϕ 2 + O(4),
sin ϕ = ϕ + O(3).
is the degree 2 Taylor polynomial for x2 + y2 at 21 , 21 .
Consequently Q2 − Q is the Multiplication of these expressions gives the degree 2
1 1
desired degree 4 Taylor poly- Taylor polynomial for s at (ρ , θ , ϕ ) = (ρ , π /2, 0) :
nomial for f (x, y) at 2 , 2 . A straightforward calculation 0
gives
x = −ρ0 (θ − π /2) − (ρ − ρ0 )(θ − π /2) + O(3),
Q2 − Q = − 41 + (X + Y )2 + 2(X 3 + X 2Y + XY 2 + Y 3 ) y = ρ0 + (ρ − ρ0 ) − 21 ρ0 (θ − π /2)2 − 12 ρ0 ϕ 2 + O(3),
+ (X 2 + Y 2 )2 . z = ρ0 ϕ + (ρ − ρ0 )ϕ + O(3).
3.25. We must find the derivatives of f (x, y) = ex cosy at 3.28. a. Suppose L : R → R : ∆u → ∆x is given by the
p q
(x, y) = (0, 0): q × p matrix (ai j ) as

    
fx = fxx = fxxx = fxxxx = ex cos y = 1, ∆x1 a11 · · · a p1 ∆u1
 ..   .. .. ..   ..  .
fy = fxy = fxxy = fxxxy = −ex sin y = 0,  . = . . .  . 
f =f =f = − f = −ex cos y = −1, ∆xq aq1 · · · aqp ∆u p
yy xyy xxyy
fyyy = fxyyy = − fy = ex sin y = 0, Because L(0) = 0, we may assume ∆u 6= 0 without loss
fyyyy = − fyy = ex cos y = 1. of generality. For each ∆u 6= 0, set
1
Thus we can write ∆v = ∆u where k = k∆uk > 0;
k
x2 − y2 x3 − 3xy2
ex cos y = 1 + x + + then k∆vk = 1 and thus |∆vi | ≤ 1 for all i = 1, . . . , p.
2! 3!
Choose A so that |ai j | ≤ A for all i, j. Then if ∆y = L(∆v),
x4 − 6x2y2 + y4 we have
+ + O(5);
4!
|∆y j | = |a j1 ∆v1 + · · · + a j p ∆v p | ≤ |a j1 | · 1 + |a j p| · 1 ≤ pA,
this agrees with the expression on page 97 of the text.
3.26. In words, the given equation asserts that and
the product of a function that vanishes at least k∆yk2 = (∆y1 )2 + · · · + (∆yq )2 ≤ qp2 A2 .
to order p and a function that vanishes at least
Therefore,
to order q vanishes at least to order p + q.
kL(∆u)k = kL(k∆v)k = kkL(∆v)k
To prove the assertion, let ϕ1 (t) = O(p) and ϕ2 (t) = O(q). √
Then there are constants δ1 , δ2 , C1 , and C2 for which = k · k∆yk ≤ k∆uk · pA q.
√
|ϕ1 (t)| ≤ C1 |t| p when |t| < δ1 , Thus we can take C = pA q and have
|ϕ2 (t)| ≤ C2 |t| p when |t| < δ2 . kL(∆u)k ≤ Ck∆uk for all ∆u.
Therefore, if we take δ = min(δ1 , δ2 ), C = C1C2 , then

3.28. b. The function kL(∆v)k is continuous on the unit
|t| < δ implies
sphere k∆vk = 1, a closed bounded set in R p . Therefore
|ϕ (t) · ϕ (t)| ≤ C |t| p ·C |t|q = C|t| p+q . it achieves its maximum M at some point on that set:
1 2 1 2
M = max kL(∆v)k.
It follows that ϕ1 (t) · ϕ2 (t) = O(p + q). k∆vk=1

31
By definition, kL(∆u)k ≤ 9L 9 k∆uk for all ∆u in R p .

In particular, this holds for all unit vectors ∆v in the form
kL(∆v)k ≤ 9L9. Hence
M = max kL(∆v)k ≤ 9L 9 .
On the other hand, maximality means that we have

kL(∆v)k ≤ M for all unit vectors ∆v. If ∆u is an arbitrary
nonzero vector, then we can write
∆u = k∆v
where k = k∆uk and ∆v is a unit vector. Therefore
kL(∆u)k = kkL(∆v)k ≤ kM = Mk∆uk,
so M provides a bound for all ∆u. But 9L9 is the smallest

such bound, so we must have 9L9 ≤ M. Together the two
inequalities show that
9L9 = M = max kL(∆v)k.

k∆vk=1
3.28. c. By assumption, L : R p → R p is now invertible,

so ∆x = L(∆u) spans R p when ∆u does. By part (a), we
can write
kL(∆u)k ≤ A2 k∆uk
for some constant A2 > 0 and for all ∆u. With ∆u =
L−1 (∆x), the previous inequality takes the form
k∆xk = kL(∆u)k ≤ A2 k∆uk = A2 kL−1 (∆x)k.
This holds for all ∆x; for all ∆x 6= 0, we can rewrite the
inequality as
1 kL−1 (∆x)k
≤ .
A2 k∆xk
By part (a), we can also write
kL−1 (∆x)k ≤ B2 k∆xk
for some B2 > 0 and for all ∆x. The previous argument
then shows, mutatis mutandis, that
1 kL(∆u)k
≤
B2 k∆uk
for all ∆u 6= 0. That is,
1 kL(∆u)k 1 kL−1 (∆x)k

≤ ≤ A2 , ≤ . ≤ B2 .
B2 k∆uk A2 k∆xk
3.28. d. The solution of the previous part shows that we

can take A1 = 1/B2 and B1 = 1/A2 .

The Derivative
4.1. a. A linear function is its own derivative at every 4.2. c. Shown below are contour plots of f (x, y) in two
point; thus d f(4,5) (∆x, ∆y) = 7∆x − 3∆y. windows centered at (2, −1). In the smaller window, on
4.1. b. We have fx = sin x = 0 when x = 0, fy = y = 1 the right, the contours do indeed appear to be the contours
when y = 1, so d f (∆x, ∆y) = ∆y. of a linear function.
(0,1)
0.0 -0.90
4.1. c. At the point (4, −3), fx = −y/(x2 + y2 ) = 3/25
and fy = x/(x2 + y2 ) = 4/25; thus d f(4,−3) (∆x, ∆y) = -0.5 -0.95
(−3∆x + 4∆y)/25.
-1.0 -1.00
4.1. d. Here fx = 2α x + 2β y + δ and fy = 2β x + 2γ y + ε ;
thus -1.5 -1.05
d f(a,b) (∆x, ∆y) = (2α a+2β b+ δ )∆x+(2β a+2γ b+ ε )∆y.

-2.0 -1.10
1.0 1.5 2.0 2.5 3.0 1.90 1.95 2.00 2.05 2.10
4.2. a. In the figure below, the window is 0.2 units wide;
the graph of z = f (x, y) does indeed appear flat. Shown below are the contours of f (x, y) (left) and its
derivative 3 + d f (2, −1) (right) to make it clear that f is
close to its derivative in a small window.
-0.90 -0.90
3.5 -0.95 -0.95

-0.90
3.0
-0.95 -1.00 -1.00
2.5
1.90 -1.00
1.95 -1.05 -1.05
2.00 -1.05
2.05
2.10
-1.10 -1.10 -1.10
1.90 1.95 2.00 2.05 2.10 1.90 1.95 2.00 2.05 2.10
4.2. b. The derivative is d f(2,−1) (∆x, ∆y) = 4∆x + 2∆y. 4.3. a. The derivatives of f (x, y) at (x, y) = (π /3, −π /2)
The figure below plots z = 3+d f(2,−1) (∆x, ∆y) in the same are
window; the two graphs appear indistinguishable.
1 1
fx = cos x sin y = · (−1) = − , fy = sin x cos y = 0.
2 2
√
In addition, f (π /3, −π /2) = − 3/2, so the equation of
the tangent plane is
3.5
√
-0.90
− 3 − ∆x
, where ∆x = x − π /3.
3.0
-0.95 z=
2.5 2
1.90 -1.00
1.95
-1.05
4.3. b. The figure below uses a viewpoint that helps make
2.00
2.05 it clear that the plane is tangent to the (curved) graph of
2.10
-1.10
f (x, y).
32

33
0
-1 To see that the statement is true, let f (t) = cost − 1 and
-3
-2
g(t) = t. Then f (0) = g(0) = 0, so l’Hôpital’s rule applies
0.0 and says that
-0.5
-1.0 cost − 1 f (t) f ′ (0)

lim 1
= lim = ′ = 0,
-1.5 t→0 t t→0 g(t) g (0)
because the fraction f ′ (0)/g′ (0) is not indeterminate.
0
1
2
3
4.4. b. The statement “sint = o(1)” would mean
4.3. c. The figure below uses a square in the (x, y)-plane sint
lim = 0.
that is 0.1 units on a side; the graph of z = f (x, y) is just t→0 t1
barely distinguishable from its tangent plane.
In fact, the value of the limit is 1 so the statement is false
-1.55 (i.e., sint 6= o(1)).
-1.60
-0.84
4.4. c. The statement “sint − t = o(2)” means
-0.86 sint − t
lim = 0.
-0.88
t→0 t2
1.00 Using l’Hôpital’s rule with f (x) = sint − t and g(t) = t 2 ,
1.05
we find this time f ′ (0)/g′ (0) = 0/0 is indeterminate so
we move on to the second derivatives:
4.3. d. Shown below in the large window are the contours
sint − t f (t) f ′′ (0)
of f (left) and its derivative (right). There is no resem- lim 2
= lim = ′′ = 0.
blance at this level.
t→0 t t→0 g(t) g (0)
0.0 0.0 The statement “sint − t = o(3)” would mean

-0.5 -0.5 sint − t
lim = 0,
-1.0 -1.0 t→0 t3
-1.5 -1.5
but this is false. In fact, using the Taylor expansion for
-2.0 -2.0 sint centered at a = 0, we have
-2.5 -2.5
sint − t − 61 t 3 + O(5) − 16 + O(2) 1
-3.0 -3.0 lim = lim = lim =− ,
t→0 t3 t→0 t3 t→0 1 6
0.0 0.5 1.0 1.5 2.0 2.5 3.0 0.0 0.5 1.0 1.5 2.0 2.5 3.0
not 0. The limit can also be found using l’Hôpital’s rule.
Show below are contours of the same two functions but The statement “sint − t = O(3)” is true, as can be seen
now in a square window that is just 0.1 units wide. At this from the degree 2 Taylor expansion of sin t centered at
level the resemblance has become clear. t = 0:
-1.52 -1.52
sint = t + O(3).
-1.54 -1.54
Subtraction gives sint − t = O(3).
-1.56 -1.56
4.5. To prove differentiability of f (x) at x = 0 using the
definition, we must show
-1.58 -1.58
f (∆x) = f (0) + f ′ (0)∆x + o(1).
-1.60 -1.60
-1.62 -1.62
We are required here to prove the more stringent condition
1.00 1.02 1.04 1.06 1.08 1.00 1.02 1.04 1.06 1.08
f (∆x) = f (0) + f ′ (0)∆x + O(2).
4.4. a. The statement “cost − 1 = o(1)” means that In the present case (where we aim to establish f ′ (0) = 0),
this reduces to showing
cost − 1
lim = 0. (∆x)2 sin(1/∆x) = O(2).
t→0 t1

34 SOLUTIONS: CHAPTER 4. THE DERIVATIVE
This means, by definition, that there should be positive That is, we write (x, y, z) in polar coordinates (see the dis-
numbers δ and C for which cussion of the “manta ray” function on page 108 of the
text). The illustration shows hat the graph is made up of
|(∆x)2 sin(1/∆x)| ≤ C|∆x|2 when |∆x| < δ . straight lines radiating from the origin. The ray above the
point (cos θ , sin θ ) in the (x, y)-plane has slope sin 2θ .
In fact, | sin(1/∆x)| ≤ 1 for all ∆x 6= 0 so we can take C = 1
and let δ be arbitrary. This establishes that f (x) is differ-
entiable at x = 0 and that f ′ (0) = 0.
For x 6= 0 direct calculation gives
f ′ (x) = 2x sin(1/x) − cos(1/x).
If f ′′ (0) existed, it would be
f ′ (x) − f ′ (0)
f ′′ (0) = lim
x→0 x
2x sin(1/x) − cos(1/x) − 0
= lim
x→0 x
cos(1/x)
= lim 2 sin(1/x) − lim
x→0 x→0 x
4.7. b. The figure shows that rays in the coordinate di-
The function in the first limit remains bounded as x → 0, rections have slope 0, showing quickly that fx (0, 0) =
but in the second the function is unbounded. Therefore fy (0, 0) = 0. More formally,
the limit cannot exist, so f ′′ (0) does not exist.
f (h, 0) − f (0, 0) 0−0
If xn = 1/(2π n), then xn → 0 as n → ∞. For any positive fx (0, 0) = lim = lim = 0;
integer n, h→0 h h→0 h
a similar calculation shows fy (0, 0) = 0.
sin(1/xn ) = sin(2π n) = 0, cos(1/xn ) = cos(2π n) = +1, 4.7. c. By definition, the directional derivative of f (x, y)
in the direction of the unit vector u = (u, v) at a = (0, 0) is
so
′ ′
+1 = lim f (xn ) 6= f lim xn = 0. d 2t 2 uv
n→∞ n→∞ f (tu,tv) = lim = lim ±2uv;
dt t→0 t|t| t→0
Thus f ′ (x) is not continuous at x = 0, providing another t=0
reason why f ′′ (0) does not exist. the sign is the sign of t. By assumption, u, v 6= 0 because
u is not a coordinate direction, so the limit does not exist.
4.6. The limiting value of the given expression along the The vertical section of the graph through the origin in the
ray ∆y = m∆x, ∆x > 0 is direction (cos θ , sin θ ) has the form z = sin 2θ |t|. When
sin 2θ 6= 0 (i.e., when θ does not point in a coordinate
m(∆x)3 m(∆x)3
lim = lim direction), this graph is nondifferentiable at the origin.
∆x→0 (∆x)2 + m2 (∆x)2 3/2 ∆x→0 (1 + m2 )3/2 (∆x)3
4.7. d. By the contrapositive to Theorem 4.3 (text p. 109),
m
= , f cannot be differentiable at the origin because some of its
(1 + m2)3/2 directional derivatives fail to exist there.
a constant. Along different rays this constant has different 4.8. a. Shown immediately below is the graph on the do-
values, so main −1 ≤ x ≤ 1, −1 ≤ y ≤ 1.
f (∆x, ∆y) − f (0, 0)
lim
(∆x,∆y)→(0,0) k(∆x, ∆yk
does not exist. 0.5
1.0
4.7. a. The suggestion about a polar-coordinates overlay 0.0
0.5
means we represent the graph of z = f (x, y) as a paramet- -0.5
0.0
ric plot in R3 using -1.0
-0.5
0.0 -0.5
(r, θ ) 7→ (r cos θ , r sin θ , 2r sin θ cos θ ). 0.5

-1.0
1.0

35
It appears that something unusual is happening near the The figure shows that the curve (x, x2 , x/2) lies along the
origin; we see the shape of the graph more clearly below ridge line of the escarpment of the graph of z = f (x, y) in
on the domain −0.1 ≤ x ≤ 0.1, −0.1 ≤ y ≤ 0.1. the first quadrant and along its valley line (“thalweg”) in
the second quadrant.
Every point in the (x, y)-plane lies on the line x = 0 or
else on one of the parabolas y = mx2 , −∞ < m < ∞. We
have f (0, y) = 0 and
0.05
0.10
0.00 mx5 m
0.05 f (x, mx) = = x.
-0.05 x 4 + m2 x 4 1 + m2
-0.10 0.00
-0.05
0.00 -0.05 Because the extreme values of m/(1 + m2) are ±1/2 when
0.05 m = ±1, we have
-0.10
0.10
| f (x, y)| ≤ 21 |x| ≤ 12 k(x, y)k for all (x, y);
The graph has an “escarpment” along the x-axis; at x = k
its slope in the y-direction is 1/k and it extends vertically this shows f (x, y) = O(1). But the fact that
the distance k/2 above and below the axis.
f (t,t 2 ) 1
4.8. b. To compute directional derivatives of f at the ori- lim = , not 0,
t→0 t 2
gin, first note that
is sufficient to show f (x, y) 6= o(1). Hence f (x, y) van-
t 4 u3 v t 2 u3 v
f (tu,tv) = 4 4 2 2 = 2 4 , ishes exactly to order 1 at the origin. Furthermore, if f
t u +t v t u + v2 were differentiable at the origin, its derivative would have
where (u, v) is a unit vector with v 6= 0. If v = 0 then to be the zero linear function (because, by part (b), all di-
f (tu,tv) = 0 and the directional derivative in each of the rectional derivatives of f are 0 there.) This would then
directions (±1, 0) is 0. The directional derivative at the imply f (x, y) − 0 = o(1), which we have just shown to be
origin in the direction of (u, v) with v 6= 0 is false.

d 2tu v 3 3 4.9. a. By Definition 3.16 (text p. 99), the derivative of
f (tu,tv) = 2 4 2 )2
= 0.
dt t=0 (t u + v t=0
a function is given by its matrix of partial derivatives. We
have
4.8. c. Away from the origin the partial derivatives are
∂x ∂x
3x2 y3 − x6 y x7 − x3 y2 = a, = b,
fx = 4 and f = . ∂ u ∂v
y
(x + y2 )2 (x4 + y2 )2 ∂y ∂y
= c, = d,
Note that fx (t,t 2 ) = 1/2 when t 6= 0, but fx (0, 0) = 0 so ∂u ∂v
fx in not continuous at (0, 0). Next, fy (x, 0) = 1/x when at all points u = (u, v), so
x 6= 0, but fy (0, 0) = 0, so fy is not continuous at (0, 0)
either. a b
df(u,v) = for all (u, v).
3 · x2 c d
x x
4.8. d. We have z = f (x, x2 ) = 4 = .
x + x4 2 The derivative is thus independent of (u, v); it is the “lin-
ear part” of f itself. That is,

x u u k a b u k
=f = dfu + = + .
y v v l c d v l
0.5
4.9. b. The partial derivatives of g are clear; they give
1.0

α β
0.0
0.5
dg(x,y) = for all (x, y).
-0.5 γ δ
-1.0 0.0
-0.5
With the substitution x = f(u) we have
0.0 -0.5

0.5 α β
dg(au+by+k,cx+dy+l) = for all (u, v).
γ δ
-1.0
1.0

4.9. c. Because f and g are (translates of) linear maps, 4.10. d. We know det dfr = r; we also have
their composite h = g ◦ f is obtained (mostly) by matrix x2 y2 1
multiplication. We have detdf−1
x = 2 2 3/2
+ 2 2 3/2
=p .
(x + y ) (x + y ) x + y2
2
p
r u u k κ With the substitution r = f−1 (x) we have x2 + y2 = r
=g f = dgf(u) dfu + +
s v v l λ and thus

α β a b u k κ 1 1
= + + det df−1 −1
γ δ c d v l λ x = detdff(r) = = .
r det dfr
α β a b u
= linear part of h 4.11. a. We find immediately that
γ δ c d v
2x 2y 2r cos θ 2r sin θ
α β k κ dgx = = = dgf(r) .
+
γ δ l
+
λ
translation 2x −2y 2r cos θ −2r sin θ
We can rewrite this as 4.11. b. With x2 − y2 = r2 cos2 θ − r2 sin2 θ = r2 cos 2θ

( we find (
r = (α a + β c)u + (α b + β d)v + α k + κ , u = r2 ,
h: h: .
s = (γ a + δ c)u + γ b + δ d)v + γ k + δ l + λ , v = r2 cos 2θ
but the matrix form is probably more enlightening. 4.11. c. The partial derivatives of u and v with respect to
r and θ give us
4.9. d. The matrix form of the map h makes it clear that
2r 0
the “linear part” of h, that is dhu , equals dgf(u) ◦ dfu . dhr = .
2r cos 2θ −2r2 sin 2θ
4.10. a. The partial derivatives we need are
For the product of the derivatives, we have

∂r x ∂r y 2r cos θ 2r sin θ cos θ −r sin θ
=p , =p , dgf(r) · dfr =
∂x x2 + y2 ∂y x2 + y2 2r cos θ −2r sin θ sin θ r cos θ

∂θ −y ∂θ x 2r cos θ + 2r sin θ −2r cos θ sin θ + 2r2 sin θ cos θ
2 2 2
= 2 , = ; =
∂x x + y2 ∂y x2 + y2 2r cos2 θ − 2r sin2 θ −2r2 cos θ sin θ − 2r2 sin θ cos θ

2r 0
they give = = dhr .
2r cos 2θ −2r2 sin 2θ
p p !
−1 x/ x2 + y2 y/ x2 + y2 4.12. We have
dfx = .
−y/(x2 + y2 ) x/(x2 + y2)
kf(∆u)k2 = (∆x)2 + (∆y)2
= (∆u)4 − 2(∆u)2(∆v)2 + (∆v)4 + 4(∆u)2(∆v)2
4.10. b. The text (p. 114) provides dfr as a 2 × 2 matrix
with determinant equal to r; its inverse is = (∆u)4 + 2(∆u)2(∆v)2 + (∆v)4
2

2 2
1 r cos θ r sin θ cos θ sin θ = (∆u) + (∆v) = k∆uk4 ;
−1
(dfr ) = = .
r − sin θ cos θ − sin θ /r cos θ /r thus kf(∆u)k = k∆uk2 .
4.13. a. To determine f(g+ (x)), we substitute the formu-
4.10. c. With x = r cos θ , y = r sin θ , we have las for u and v given by g into the formulas for x and y
given by f:
x r cos θ x r cos θ cos θ p p
p = = cos θ , 2 + y2
= 2
= ; x 2 + y2 + x x2 + y2 − x
2
x +y 2 r x r r 2
x = u −v =2
− = x,
s p2 p
2
the two terms with y can be converted in a similar way to 2 + y2 + x
give x x2 + y2 − x
y = 2uv = 2 ·
cos θ sin θ s
2 2
df−1
f(r) = − sin θ /r cos θ /r .
x2 + y2 − x2 p 2
= 2 = y = |y| = y.
This confirms that (dfr )−1 = df−1 .
f(r)
4

37
Note that y > 0 in the domain U 2 , so |y| = y there. what complicated. The derivatives with respect to x are
p !−1/2 p
4.13. b. We take as the polar-coordinate overlays the fol- ∂u 1 x2 + y2 + x x/ x2 + y2 + 1
lowing: = ·
∂x 2 2 2
p ! p
u = ρ cos ϕ , x = r cos θ , 1 x2 + y2 + x
−1/2
x + x2 + y2
v = ρ sin ϕ , y = r sin θ . = p ·
2 x2 + y2 2 2
p !1/2
The map g+ gives us 1 x2 + y2 + x u
= p = p ;
p 2
2 x +y 2 2 2 x2 + y2
x2 + y2 + x r + r cos θ p !−1/2 p
ρ cos ϕ = u =
2 2 2
= ∂v 1 x2 + y2 − x x/ x2 + y2 − 1
2 2 = ·
1 + cos θ ∂x 2 2 2
=r = r cos (θ /2);
2
p !−1/2 p
p 2 1 x2 + y2 − x x − x2 + y2
x2 + y2 − x r − r cos θ = p ·
ρ 2 sin2 ϕ = v2 = = 2 x2 + y2 2 2
2 2
1 − cos θ p !1/2
=r = r sin (θ /2).
2
−1 x2 + y2 − x −v
2 = p = p ;
2 x2 + y2 2 2 x2 + y2
Thus
√ To simplify the expressions, u and v were substituted for
ρ= r, ϕ = θ /2;
their values in terms of x and y. The derivatives with re-
in other words, g+ takes the square root of the distance spect to y are simpler:
from a point to the origin and cuts in half the angle that p !−1/2
the point makes with the horizontal. As the inverse of f, ∂u 1 x2 + y2 + x y y
= · p = p ;
g+ acts just as we would expect. ∂y 2 2 2
2 x +y 2 4u x2 + y2
p !−1/2
4.13. c. Because g+ cuts angles at the origin in half, the ∂v 1 x2 + y2 − x y y
image of U 2 is the open first quadrant: u > 0, v > 0. = · p = p .
∂y 2 2 2
2 x +y 2 4v x2 + y2
√
4.14. First consider x ≥ 0; then x2 = x and, as y → 0, Thus, keeping in mind that u and v represent functions of
sp x and y,
r
x2 + y2 + x x+x √  u y 
u= → = x, p p
2 2  2 x2 + y2
sp 4u x2 + y2 
r d(g+ )x = 
 −v y


x2 + y2 − x x−x p p
v= → = 0. 2 x2 + y2 2
4v x + y 2
2 2
√ p
By contrast, if x ≤ 0 then x2 = |x| = −x and, as y → 0, 4.15. b. Using x = u2 − v2 , y = 2uv, we have x2 + y2 =
sp u2 + v2 , y/2u = v, y/2v = u, and thus
r
x2 + y2 + x −x + x
u= → = 0, 1 u v
2 2 d(g+ )f(u) = .
sp 2(u2 + v2 ) −v u
r
x2 + y2 − x −x − x √ p
v= → = −x = |x|.
2 2 4.15. c. We have

The image of the positive x-axis is the positive u-axis; the 2u −2v −1 1 2u 2v
dfu = , (dfu ) = ,
image of the negative x-axis is the positive v-axis. 2v 2u 4(u2 + v2 ) −2v 2u
4.15. a. Calculation of the partial derivatives is some- showing that d(g+ )f(u) is the inverse of dfu .

√
4.16. a. Given that x0 = 1/2, y0 =
q 3/2, we see that The figure suggests that the map g+ rotates a small source
x20 + y20 = 1 and window centered at x0 about −30◦ while shrinking all di-
mensions to about 1/2 original size. As we have seen, this
s s is also the action of its derivative d(g+ )x0 , so g+ “looks
√
1 + 21 3 1 − 21 1 like” d(g+ )x0 near x0 .
u0 = = , v0 = = .
2 2 2 2 4.17. a. We can analyze the map g using the approach
−
we took with g+ in the solution to Exercise 4.13. The only
In polar coordinates (cf. the solution to Exercise 4.13.b),
difference is the sign of u; this affects only the computa-
r0 = 1 and θ0 = π /3 (or 60◦ ), so the polar coordinates of
tion for y:
the image are ρ0 = 1 and ϕ0 = π /6 (or 30◦ ).
p p
4.16. b. By the solution to Exercise 4.13, g+ maps the 60◦ 2 2 x2 + y2 + x x2 + y2 − x
x = u −v = − = x,
line through the origin to the 30◦ line through the origin. sp 2 2
As part (a) has just shown, x0 lies on the 60◦ line and its p
◦
x2 + y2 + x x2 + y2 − x
image on the 30 line. Because these lines pass through y = 2uv = −2 ·
2 2
the windows centered at x0 and g+ (x0 ), respectively, the s
60◦ line in the source window maps to the 30◦ line in the x2 + y2 − x2 p
target window. = −2 = − y2 = −|y| = y.
4
4.16. c. Using the calculation of the derivative of g+ in Note that |y| = −y because y < 0.
the solution to Exercise 4.15.a, we have
√ 4.17. b. The computations in the solution to Exer-
3/4 √1/4 cise 4.13.b carry over unchanged, yielding once again
d(g+ )x0 =
−1/4 3/4 √
ρ = r, ϕ = θ /2.
1 cos(−π /6) − sin(−π /6)
= = 21 R−π /6.
2 sin(−π /6) cos(−π /6) Thus g− takes the square root of the distance from a point
to the origin and cuts in half the angle that point makes
That is, λ = 1/2 and θ = −π /6. with the horizontal.
Shown below are square windows of width 0.2 units
centered at x0 (left) and g+ (x0 ) (right). The target win- 4.17. c. The formulas for g− indicate that u is negative
dow shows the image of the grid given in the source win- while v is positive, so the image is the open second quad-
dow. The marked region in the corner helps show how rant.
the image is oriented in relation to the source. The im- 4.18. a. Because y appears in the formulas for g− only in
age (without the orienting region) was produced with the the form y2 , the argument in the solution to Exercise 4.14
Mathematica command carries over here with the single change in the sign of u.
ParametricPlot[{Sqrt[(Sqrt[xˆ2 + yˆ2] + x)/2], The negative x-axis is rotated clockwise 90◦ to the positive
Sqrt[(Sqrt[xˆ2 + yˆ2] - x)/2]}, v-axis, and the positive x-axis is rotated 180◦ clockwise to
{x, 1/2 - .1, 1/2 + .1}, the negative u-axis.
{y, Sqrt[3]/2 - .1, Sqrt[3]/2 + .1}, 4.18. b. When a point (x, 0) is on thepnegative x-axis (i.e.,
PlotRange -> {{Sqrt[3]/2 - .1, Sqrt[3]/2 + .1}, x ≤ 0), it has the same image (0, |x|) on the positive
{1/2 - .1, 1/2 + .1}}, v-axis under each of the maps g± .
MeshShading -> {{None, None}, {None, None}}]
√when a point (x, 0) is on the positive x-axis,
By contrast,
0.60
image is ( x, 0) on the positive u-axis under g+ but is
its √
0.95 (− x, 0) on the negative u-axis under g− .
0.55
To see how g+ and g− work together, consider the map
0.90 f restricted to the closed upper half-plane (y ≥ 0). Be-
0.50
cause f doubles angles at the origin, f “fans out” the up-
0.85 per half of the (x, y)-plane to cover the entire (u, v)-plane.
This covers the positive u-axis twice, once by each half of
0.45
0.80 the x-axis. Therefore, if we delete the positive u-axis, the
maps g+ and g− together invert the map f. If we try to
0.40
0.40 0.45 0.50 0.55 0.60 0.80 0.85 0.90 0.95 include the positive u-axis, the inverse attempts to map it

39
to two different points: the inverse fails to be continuous (right). The target window shows the image under the
on the positive u-axis nonlinear map g− of the grid given in the source window,
including a marked corner to help show how the image is
4.19. a. Because the formulas for g− agree with those of
oriented in relation to the source.
g+ , except for the change in sign of the formula for u, the
two derivatives are identical when expressed, as we did in -2.15
1.15
the solution to Exercise 4.15.a, in terms of u and v; thus

 u y  -2.20
1.10
p p
 2 x2 + y2 4u x2 + y2 
d(g− )x =   -2.25
1.05
 −v y 
p p
2 x2 + y2 4v x2 + y2 -2.30
1.00
4.19. b. The solution to Exercise 4.15.c carries over with- -2.35

-0.10 -0.05 0.00 0.05 0.10 -1.15 -1.10 -1.05 -1.00
out change, to give
The figure suggests that the map g− rotates a small source
1 u v window centered at x0 about −135◦ while shrinking all
d(g− )f(u) = .
2(u2 + v2 ) −v u dimensions to about 1/3 original size. This is just what
the derivative does: g− “looks like” d(g− )x0 in a small
4.19. c. We saw in the solution to Exercise 4.15.c that window centered at x0 .

1 u v 4.21. a. The map f folds the (u, v)-plane along the u-axis
(dfu )−1 = .
2(u2 + v2 ) −v u and places the image of the fold on the x-axis, covering
the upper half-plane y > 0 twice. In the figure below the
4.20. a. The formulas for g− (x0 ) give formula for x has been changed to x = u + v/5 in order
sp to provide horizontal separation between the upper and
0 + 81/16+ 0 p √ lower halves of each vertical line u = k (with −2 ≤ u ≤ 2,
u=− = − 9/8 = −3 2/4, −1.9 ≤ v ≤ 2.1). The equation y = v2 accounts for the
2
sp folding and doubling.
0 + 81/16− 0 √
v= = 3 2/4. 4
2
Note that the polar angle of x0 is 270◦ ; the computations
3
and our knowledge of the action of g− confirm that the 2
polar angle of the image is 135◦.

1
4.20. b. Because the 270◦ line contains x0 it passes
through any window centered at x0 ; for a similar reason, 0
-2 -1 0 1 2
the 135◦ line passes through any window centered at its
image g− (x0 ).
1 0
4.21. b. The derivative is df(a,b) = .
4.20. c. From the solution to Exercise 4.19.a we have 0 2b
 √  4.21. c. The local area magnification factor is given by
−3 2/4 −9/4
 2 · 9/4 √  the determinant: det df(a,b) = 2b. The local area multiplier
 −3 2 · 9/4 
d(g− )x =  √  is zero on the horizontal axis because b = 0 there. The
 −3 2/4 −9/4  figure above shows that the image “bunches up” near the
√
2 · 9/4 3 2 · 9/4 horizontal axis (see part (e)). Specifically, the image of
√ √ !
a horizontal strip of vertical width 2ε along the u-axis is
1 − 2/2 2/2
= √ √ = 13 R−3π /4. a horizontal trip of vertical width ε 2 along the x-axis, so
3 − 2/2 − 2/2
the area magnification factor is ε /2, which becomes as
small as we wish by making the width of the original strip
Thus λ = 1/3 and θ = −3π /4 (or −135◦).
arbitrarily small.
To show that g− “looks like” d(g− )x0 in a small window
centered at x0 , we have constructed below square win- 4.21. d. Near (3, 2), the map f stretches vertical lengths
dows of width 0.2 units centered at x0 (left) and g− (x0 ) roughly by a factor of 4. The figure below shows this

clearly; it is a plot of the image of square window of width where

0.2 centered at (u, v) = (3, 2). Because df(3,2) stretches
vertical lengths exactly by the factor of 4, f “looks like” 3a2 − 3b2 6ab
P= and Q = .
df(3,2) near (3, 2). 3(a2 + b2 ) 3(a2 + b2)
4.4 We have P2 + Q2 = 1; so if we let θ (P, Q) = arctan(Q/P),

(see Exercise 2.10, text p. 59 for the definition of this
function defined for all (P, Q) 6= (0, 0)), then P = cos θ ,
Q = sin θ and

cos θ − sin θ
dqa = 3(a2 + b2 ) .
4.2 sin θ cos θ
That is, dq(a,b) is a similarity transformation λ Rθ with
uniform dilation by λ = 3(a2 + b2),

2
a − b2
4.0 rotation by θ = arctan .
2ab
We must exclude the origin (a, b) = (0, 0) from this de-

scription because the arctangent function, and hence the
rotation Rθ , is undefined there. A different reason for ex-
cluding the origin emerges in the solution to part (c): q is
3.8
not conformal at the origin because it triples angles there.
4.22. c. We first need “triple angle” formulas in order to
express cos 3ϕ and sin 3ϕ in terms of cos ϕ and sin ϕ . We
have
3.6
2.90 2.95 3.00 3.05 3.10
cos 3ϕ = cos(ϕ + 2ϕ ) = cos ϕ cos 2ϕ − sin ϕ sin 2ϕ
= cos3 ϕ − cos ϕ sin2 ϕ − 2 sin2 ϕ cos ϕ
4.21. e. The first figure below shows the image under f of
= cos3 ϕ − 3 cos ϕ sin2 ϕ ,
a square window of width 0.2; the second is the image of
the same window under the map modified to x = u + v/5 sin 3ϕ = sin(ϕ + 2ϕ ) = sin ϕ cos 2ϕ + cos ϕ sin 2ϕ
(in order to separate images of the upper and lower halves = cos2 ϕ sin ϕ − sin3 ϕ + 2 cos2 ϕ sin ϕ
of each vertical line). The images show the “bunching up”
= 3 cos2 ϕ sin ϕ − sin3 ϕ .
mentioned in part (c).
Thus, using
2.90 2.95 3.00 3.05 3.10
u = ρ cos ϕ , x = r cos θ ,
2.90 2.95 3.00 3.05 3.10 v = ρ sin ϕ , y = r sin θ ,
we find
4.22. a. Here is the derivative expressed in terms of the
coordinates of the point a = (a, b):
r cos θ = x = u3 − 3uv2
2
dqa =
3a − 3b2 −6ab
. = ρ 3 cos3 ϕ − 3ρ 3 cos ϕ sin2 ϕ = ρ 3 cos 3ϕ ,
6ab 3a2 − 3b2
r sin θ = y = 3u2 v − v3
4.22. b. We have det dqa = 9(a2 + b2 )2 , so we can write = 3ρ 3 cos2 ϕ sin ϕ − ρ 3 sin3 ϕ = ρ 3 sin 3ϕ ,

P −Q and finally r = ρ 3 , θ = 3ϕ . In other words, q cubes dis-
dqa = 3(a2 + b2 ) tances from the origin and triples angles there.
Q P

41
4.23. a. The derivative of s at the point a = (a, b) is With the polar overlays, we find

A −B A = 4a3 − 12ab2, r cos θ = x = u4 − 6u2v2 + v4
dsa = ,
B A B = 12a2 b − 4b3.
= ρ 4 cos4 ϕ − 6ρ 4 cos2 ϕ sin2 ϕ + ρ 4 sin4 ϕ
4.23. b. We have = ρ 4 cos 4ϕ ,
det dsa = A2 + B2 = 16(a2 + b2)3 . r sin θ = y = 4u3 v − 4uv3

= 4ρ 4 cos3 ϕ sin ϕ − 4ρ 4 cos ϕ sin3 ϕ
Therefore, we can write
= ρ 4 sin 4ϕ ,
P −Q
dsa = 4(a2 + b2 )3/2
Q P and finally r = ρ 4 , θ = 4ϕ . In other words, s raises dis-
tances from the origin to the fourth power and quadruples
where
angles there.
4a3 − 12ab2 12a2b − 4b3
P= and Q = . 4.24. The parametrization of the unit sphere used here is
4(a2 + b2 )3/2 4(a2 + b2)3/2

As in the previous exercise we have P2 + Q2 = 1, so we set x = cos θ cos ϕ ,

−π ≤ θ ≤ π ,
θ (P, Q) = arctan(Q/P) and then have P = cos θ , Q = sin θ f : y = sin θ cos ϕ ,

 − π /2 ≤ ϕ ≤ π /2.
to get z = sin ϕ ,

2 2 3/2 cos θ − sin θ
dsa = 4(a + b ) .
sin θ cos θ 4.24. a. The local area magnification factor for f at a point
(θ , ϕ ) is the square root of the sum of the squares of the
That is, ds(a,b) is a similarity transformation λ Rθ with 2 × 2 determinants of df(θ ,ϕ ) (cf. p. 123 of the text). The
2 × 2 determinants are
uniform dilation by λ = 4(a2 + b2)3/2 ,
2 − sin θ cos ϕ − cos θ sin ϕ
3a b − b3 = cos ϕ sin ϕ ,
rotation by θ = arctan 3 . cos θ cos ϕ − sin θ sin ϕ
a − 3ab2
cos θ cos ϕ − sin θ sin ϕ
We must exclude the origin (a, b) = (0, 0) from this de- = cos θ cos2 ϕ ,
0 cos ϕ
scription because the arctangent function, and hence the

0 cos ϕ
− sin θ cos ϕ − cos θ sin ϕ = sin θ cos ϕ ,
rotation Rθ , is undefined there. A different reason for ex- 2
cluding the origin emerges in the solution to part (c): s is
not conformal at the origin because it quadruples angles
and the sum of their squares is
there.
4.23. c. We now need “quadruple angle” formulas. We cos2 ϕ sin2 ϕ + cos4 ϕ = cos2 ϕ (sin2 ϕ + cos2 ϕ ) = cos2 ϕ .
could extend the work of the previous exercise, but here is p
a different approach that uses Euler’s formula Thus the local area magnification factor is cos2 ϕ =
| cos ϕ | = cos ϕ because cos ϕ ≥ 0 on the given domain.
cos α + isin α = eiα .
4.24. b. The parallel of latitude at latitude ϕ = ϕ0 can be
With α = 4ϕ we find given as the parametrized curve x1 (θ ) = f(θ , ϕ0 ), where
−π ≤ θ ≤ π . Its arc length is
cos 4ϕ + i sin 4ϕ = ei(4ϕ ) = (eiϕ )4
Z π
= (cos ϕ + i sin ϕ )4 L1 = kx′1 (θ )k d θ .
−π
= cos ϕ + 4icos ϕ sin ϕ + 6i cos ϕ sin ϕ
4 3 2 2 2
+ 4i3 cos ϕ sin3 ϕ + i4 sin4 ϕ . The derivative x′1 (θ ) is just the first column of the ma-
trix df(θ ,ϕ0 ) ; that is,
Collecting real and imaginary parts separately, we find
q
cos 4ϕ = cos ϕ − 6 cos ϕ sin ϕ + sin ϕ ,
4 2 2 4 kx′1 (θ )k = sin2 θ cos2 ϕ0 + cos2 θ cos2 ϕ0 = cos ϕ0 ,
sin 4ϕ = 4 cos3 ϕ sin ϕ − 4 cos ϕ sin3 ϕ . a value independent of θ . Therefore L1 = 2π cos ϕ0 .

We can parametrize the meridian of longitude θ = θ0 For proper comparison, all four boxes are shown from the
as x2 (ϕ ) = f(θ0 , ϕ ), −π /2 ≤ ϕ ≤ π /2. Its arc length is same viewpoint. The coordinate labels on the second pair
Z π /2 of boxes underscore the fact that the images inside are
L2 = kx′2 (ϕ )k d ϕ . indistinguishable: near the point (π /3, π /6), the map f
−π /2 “looks like” its derivative there.
The derivative x′2 (ϕ ) is the second column of df(θ0 ,ϕ ) : 4.26. For the parametrization f of the the crosscap, the
q local area magnification factor at (p, q) is the square root
kx′2 (ϕ )k = cos2 θ0 sin2 ϕ + sin2 θ0 sin2 ϕ + cos2 ϕ = 1. of the sum of the squares of the 2 × 2 determinants of
the derivative df(p,q) given on page 126 of the text. The
Therefore L2 = π , a value that is independent of θ0 . determinants are

4.25. a. At the point θ = π /3, ϕ = π /6, the derivative 1 0
= p, q
p 2
0 −2q
= 2q,
of f is   q p 0 −2q = −2q , 1 0
−3/4 −1/4
√ √ 
and the local area magnification factor is therefore
df(π /3,π /6) =  3/4 − 3/4 .
√ p
0 3/2 M = p2 + 4q2 + 4q4.
The image of df(π /3,π /6) is the plane in R3 spanned by the
two column vectors of the matrix df(π /3,π /6). The cross- Its value at each of the three points is
product of those √
√ two √ vectors gives a normal to the plane, a : p = 1, q = 0, M = 1 = 0 + 0 = 1;
namely (3, 3 3, 2 3). Hence the equation of the plane is √
√ √ b : p = −1, q = 1, M 1 + 4 + 4 = 3;
3x + 3 3y + 2 3z = 0. c : p = q = 0, M = 0.
Shown on the left below is the image of df(π /3,π /6).
4.27. The task here is to separate the linear and nonlinear
-1.0
0.0 -0.5 0.6
0.5
0.4 0.3
parts of ∆x = f(p + ∆u) − f(p):
0.5
1.0
0.6 ∆x = p + ∆u − p = ∆u,
0.5
∆y = (p + ∆u)(q + ∆v) − pq = q ∆u + p ∆v + ∆u · ∆v,
0.5
0.0
∆z = −(q + ∆v)2 − (−q2 ) = −2q ∆v − (∆v)2
0.4
-0.5
0.6
4.28. We can think of the torus tR,a as the surface gener-
0.7
-0.5
0.0 0.8 ated by a circle in a plane through the z-axis as the plane
0.5 0.9
is revolved around that axis. The generating circle has ra-
dius a; its center lies at the distance R from the z-axis and
4.25. b. Shown on the right above is the image under f
thus generates a circle of radius R that we can think of as
itself of a square window of width 0.4 centered at (θ , ϕ ) =
the “core” of the torus.
(π /3, π /6) in the parameter plane. The images below are
of the smaller window of width 0.04. The one on the left
is the image under df(π /3,π /6) plus a translation to map the
origin to f(π /3, π /6); the one on the right is the image
under f itself.
0.42 0.42
0.44 0.43 0.44 0.43
0.45 0.45
0.51 0.51
0.50 0.50 If R < a then the center of the generating circle is closer to
the z-axis than it is to the circle itself. In other words, the
0.49 0.49
generating circle crosses the z-axis, so the torus intersects
0.74 0.74
itself.
0.75 0.75
0.76 0.76

43
4.29. a. We have 4.30. b. The figures below repeat the solution of part(a) at
  (0, π /2), a point on the “top” of the torus. The boxes are
−(R + a cos ϕ ) sin θ −a cos θ sin ϕ seen from the same viewpoint, and the source window is
d(tR,a )(θ ,ϕ ) =  (R + a cos ϕ ) cos θ −a sin θ sin ϕ  . large enough to show that t3,1 is nonlinear but still “looks
0 a cos ϕ like” its derivative near (0, π /2).
4.29. b. To compute the local area magnification factor,

we need the 2 × 2 determinants of d(tR,a )(θ ,ϕ ) :
0.2 0.2

−(R + a cos ϕ ) sin θ −a cos θ sin ϕ

(R + a cos ϕ ) cos θ −a sin θ sin ϕ
0.0 0.0
1.02 1.02
= −a(R + a cos ϕ ) sin ϕ , 0.98

2.90 -0.2
0.98
2.90 -0.2
2.95 2.95
(R + a cos ϕ ) cos θ −a sin θ sin ϕ
3.00
3.05
3.00
3.05
3.10 3.10
0 a cos ϕ
= a(R + a cos ϕ ) cos θ cos ϕ , 4.31. a. We have
 

0 a cos ϕ −g(v) sin θ g′ (v) cos θ
−(R + a cos ϕ ) sin θ −a cos θ sin ϕ dr(θ ,v) =  g(v) cos θ g′ (v) sin θ  ,
= a(R + a cos ϕ ) sin θ cos ϕ . 0 1
The local area magnification factor M for tR,a is the square and the local are magnification factor is
root of the sum of the squares of these values. To com- q q
pute M, we first set B = a(R + a cos ϕ ); then M = (gg′ )2 + g2 cos2 θ + g2 sin2 θ = g(v) (g′ (v)2 + 1.
q
M = B2 sin2 ϕ + B2 cos2 θ cos2 ϕ + B2 sin2 θ cos2 ϕ
4.31. b. p We note that M depends only on v, not on θ . Be-
= B = a(R + a cos ϕ ). cause (g′ (v))2 + 1 ≥ 1, we have M ≥ g(v). Moreover,
g′ (v) = 0 where g has a minimum and thus M takes on its
We see that M depends only on ϕ , not on θ . The largest minimum value, namely, g(v), at such a point.
value of M is a(R+ a); it occurs when ϕ = 0. The smallest
value of M is a(R−a); it occurs when ϕ = ±π . The points 4.31. c. If we let g(v) = 2 + sinkv, then
(θ , 0) and (θ , ±π ) form the “outer” and “inner” equators p
of the torus. In the figure of the torus, above, the grid of M(v) = (2 + sinkv) k2 cos2 kv + 1
coordinate lines is most expanded near the outer equator
and most compressed near the inner equator; the figure and g(v) has its maximum when kv = π /2. Because
thus supports our deductions about largest and smallest
k coskv
values of M. M ′ (v) = √ k2 cos2kv − 2k2 sin kv + 1 ,
2 2
k cos kv + 1
4.30. a. The figures below show the image of a square
window of width 0.1 centered at (π /4, π /3). On the we see that M ′ (v) = 0 when kv = π /2; in other words,
left, the map is t3,1 itself, while on the right, the map is M has a critical point of some sort at v = π /2k. To de-
d(t3,1 )(π /4,π /3). termine its type, we can test the sign of M ′′ (π /2k). The
hand calculation is cumbersome, but a computer algebra
2.2 2.2
2.6
2.4
2.6
2.4 system provides
0.90 0.90
0.85 0.85 M ′′ (π /2k) = k2 (3k2 − 1).
2.2 2.2
2.4 2.4 √
2.6 2.6
Thus when 3k2 − 1 > 0, or k > 1/ 3, M ′′ (π /2k) > 0 so
The two boxes are seen from the same viewpoint, and the critical point is a minimum rather than a maximum.
their coordinate scales show how close the maps are. Nev- (Note: the maximum values of M occur where the other
ertheless, the source window is large enough to show the factor of M ′ vanishes, that is, where
nonlinearity of t3,1 while still suggesting that t3,1 “looks 1
like” its derivative near (π /4, π /3). k2 cos 2kv − 2k2 sin kv + 1 = 0, or cos 2kv = 2 sin kv − 2 .
k

The graphs of y = cos 2kv and y = 2 sin kv − 1/k2 intersect 4.35. The Jacobian of a map is the determinant of its
at two points symmetrically placed on either side of kv = derivative matrix. For the first, we use dq(u,v) from Ex-
π /2.) ercise 4.22. For the second, we use the fact the Jacobians
4.32. By definition, the derivative of f at a is the linear of inverse maps are reciprocals of each other. Thus,
map L for which ∂ (x, y)
= det dq(u,v) = 9(u2 + v2 )2
∂ (u, v)
f (a + ∆x) = f (a) + L(∆x) + o(1);
(from the solution to Exercise 4.22.b). Therefore
in other words, L must give
∂ (u, v) 1 1 1
f (a + ∆x) − f (a) − L(∆x) = = =
∂ (x, y) ∂ (x, y)/∂ (u, v) 9(u2 + v2)2 9(x2 + y2 )2/3
.
lim = 0.
∆x→0 ∆x
Because x and y are understood to be the independent vari-
If we let L(∆x) = f ′ (a) · ∆x, then ables in the Jacobian ∂ (u, v)/∂ (x, y), we needed to carry
out the last step to express the Jacobian in terms of x and y.
f (a + ∆x) − f (a) − f ′(a) · ∆x
lim That step follows from
∆x→0 ∆x

f (a + ∆x) − f (a) x2 + y2 = u6 − 6u4v2 + 9u2v4 + 9u4v2 − 6u2v4 + v6
= lim − f ′ (a) .
∆x→0 ∆x = (u2 + v2 )3 .
This limit is indeed 0, because
4.36. a. This is a straightforward application of the chain
′ f (a + ∆x) − f (a) rule:
f (a) = lim
∆x→0 ∆x dϕ ∂ Φ dx ∂ Φ dy
= + = (Φx , Φy ) · (x′ , y′ ).
by definition of f ′ (a). Thus we have shown that the dt ∂ x dt ∂ y dt
derivative of f at a is “multiplication by f ′ (a).”
4.36. b. Let F(x) = grad Φ (x) = (∂ Φ /∂ x, ∂ Φ /∂ y), and
4.33. By Definition 4.6 (text p, 129), the derivative dLa suppose that the oriented curve C ~ is parametrized as
is the linear map for which (x(t), y(t)), a ≤ t ≤ b. Then, by Theorem 1.1 (text p. 10)
L(a + ∆u) = L(a) + dLa (∆u) + o (1). and part (a), above,
Z Z b
But linearity of L gives F · dx = Φx (x(t), y(t))x′ (t) + Φy (x(t), y(t)y′ (t) dt
~
C a
L(a + ∆u) = L(a) + L(∆u); Z b b b

= ϕ ′ (t) dt = ϕ (t) = Φ (x(t), y(t))
hence, by the uniqueness of the derivative (Theorem 4.6), a a a
dLa = L (with o (1) = 0) for every a in R p . end of C~

4.34. This is essentially the same as Exercise 4.10. Re- = Φ (x(b), y(b)) − Φ (x(a), y(a)) = Φ (x) .
~
start of C
ferring to page 134 of the text, we have
4.37. a. We are given curves C ~ i in the (u, v)-plane that
−1 1 ρ cos ϕ ρ sin ϕ
df(ρ ,ϕ ) = are parametrized as u = ui (t), v = vi (t), ai ≤ t ≤ bi . The
ρ − sin ϕ cos ϕ ~ i ), that lie in the
map f gives new oriented curves, f(C
cos ϕ sin ϕ (x, y)-plane and are parametrized as
= .
− sin ϕ /ρ cos ϕ /ρ
x = x(ui (t), vi (t)), y = y(ui (t), vi (t)), ai ≤ t ≤ bi .
For f−1 we have
√ √ ~ i ) we have
For f(C
−1 u/ u2 + v2 v/ u2 + v2
d(f )(u,v) = , dx = xu u′i du + xv v′i dv, dy = yu u′i + yv v′i dv
−v/(u2 + v2 ) u/(u2 + v2 )
Z
and thus, if we set (u, v) = (ρ cos ϕ , ρ sin ϕ ) = f(ρ , ϕ ), and thus P dx + Q dy =
~i)
f(C

−1 ρ cos ϕ /ρ ρ sin ϕ /ρ Z bi
d(f )f(ρ ,ϕ ) =
−ρ sin ϕ /ρ 2 ρ cos ϕ /ρ 2 = P(x(ui (t), vi (t)), y(ui (t), vi (t)))(xu u′i + xv v′i )
ai
cos ϕ sin ϕ −1
= = df(ρ ,ϕ ) . + Q(x(ui (t), vi (t)), y(ui (t), vi (t)))(yu u′i + yv v′i ) dt.
− sin ϕ /ρ cos ϕ /ρ

45
If we use the abbreviations P∗ and Q∗ defined in the exer- and

cise and rearrange terms, we can rewrite the last expres-
−2uv u 2 − v2
sion as P ∗ x u + Q∗ y u = · 2u + · 2v
Z bi (u2 + v2 )2 (u2 + v2)2

= P∗ (ui (t), vi (t))xu + Q∗ (ui (t), vi (t))yu u′i dt −2u2v − 2v3 −2v
ai = = 2 ,
(u2 + v2)2 u + v2
+ P∗ (ui (t), vi (t))xv + Q∗ (ui (t), vi (t))yv v′i dt
Z −2uv u 2 − v2
P ∗ x v + Q∗ y v = · (−2v) + · 2u
= (P∗ xu + Q∗ yu ) du + (P∗xv + Q∗yv ) dv. (u2 + v2 )2 (u2 + v2 )2
~ci
2u
This establishes = 2 .
Z Z
u + v2
P dx+Q dy = (P∗ xu +Q∗ yu ) du+(P∗ xv +Q∗ yv ) dv Applying these equalities to Exercise 4.37, we find
~i)
f(C ~i
C
Z Z
for each i = 1, . . . , m. −y x −v u
dx+ 2 dy = 2 du + 2 dv.
x2 + y2 x + y2 u 2 + v2 u + v2
~ =C
4.37. b. For C ~ 1 + · · ·C
~ m , f(C)
~ = f(C
~ 1 ) + · · · + f(C
~ m ), ~
f(C) ~
C
we have
Z m Z By the previous exercise, the integral on the right is the
P dx + Q dy = ∑ P dx + Q dy change in θ = arctan(v/u) along the oriented curve C; ~
~ ~
f(C) i=1 f(Ci ) thus
m Z
=∑ (P∗ xu + Q∗ yu ) du + (P∗xv + Q∗ yv ) dv Z
−y x
end of C~

i=1 ~ci dx + 2 dy = 2 arctan(v/u) .
Z 2
f(C) x + y
~ 2 x +y 2
~
start of C
= (P∗ xu + Q∗yu ) du + (P∗xv + Q∗ yv ) dv.
~
C
4.38. Using the previous exercise with u = r, v = θ , and

−y x
P= , Q=
x2 + y2 x2 + y2
we have
−r sin θ − sin θ cos θ
P∗ = 2
= , Q∗ = ,
r r r
and
xr = cos θ , yr = sin θ ,
xθ = −r sin θ yθ = r cos θ .
Therefore,
− sin θ cos θ
P ∗ x r + Q∗ y r = cos θ + sin θ = 0,
r r
− sin θ cos θ
P ∗ x θ + Q∗ y θ = (−r sin θ ) + (r cos θ ) = 1,
r r
and thus
Z Z end of C~
−y x
2 2
dx + 2 dy = d θ = θ .
~ x +y
f(C) x + y2 ~
C ~
start of C
4.39. We solve this as well by using Exercise 4.37. We

have x2 + y2 = (u2 + v2 )2 ,
−2uv u 2 − v2
P∗ = , Q∗ = ,
(u2 + v2 )2 (u2 + v2)2

Inverses
5.1. We can adapt the method that was used on page 152 5.3. We add (1 + y2 )3/2 = cosh3 x to the equalities stated
of the text to get a formula for arccosh y. From y = in the solution to Exercise 5.1 to find
sinh x = (ex − e−x )/2 we get e2x − 2yex − 1 = 0; the Z Z Z
dy cosh x dx
quadratic formula then gives = = sech2 x dx
(1 + y2)3/2 cosh3 x
p p
ex = 2y ± 4y2 + 42 = y ± y2 + 1. sinh x y
= tanh x = =p .
p p cosh x 1 + y2
Now y2 + 1 > |y| and thus y − y2 + 1 < 0 for all y.
But ex > 0 for all x, so we can exclude the possibility of a
5.4. The identity cosh2 x − 1 = sinh2 x suggests that we
minus sign and conclude
use y = cosh x to determine the first integral. The result is
p Z Z Z
x = ln y + y2 + 1 = arcsinh y. dy sinh x dx
= = csch2 x dx
(y − 1)3/2
2 sinh3 x
To compute
p the integral,
p let y = sinh x; then dy = cosh x dx − cosh x −y
and 1 + y2 = 1 + sinh2 y = cosh x and so = − coth x = =p .
sinh x y2 − 1
Z Z
dy cosh x dx
p = The second integral has the same component, 1 − y2 , as
1 + y2 cosh x
Z the integral in Exercise 5.2.b, so we use y = tanh x again.
p
= dx = x = arcsinh y = ln y + y2 + 1 . At the point where we replace x p by y in the antiderivative,
we need cosh x = 1/ sech x = 1/ 1 − y2. The integral is
Z Z Z
5.2. a. We begin with dy sech2 x dx
= = cosh x dx
(1 − y2)3/2 sech3 x
sinh x ex − e−x e2x − 1 y
y = tanh x = = = 2x , = sinh x = tanh x · cosh x = p .
cosh x ex + e−x e +1 1 − y2
which we can write as 1 + y = (1 − y)e2x . It follows that

5.5. a. The graph of y = sech x (below left) is symmetric
about the y-axis; the graph of the inverse x = arcsech y
1+y 1 1+y
2x = ln or x = ln = arctanh y. (below right) is defined only on the interval 0 < y ≤ 1.
1−y 2 1−y
3
5.2. b. Let y = tanh x; then we have dy = sech2 x dx and

also 1 − y2 = 1 − tanh2 x = sech2 x. Therefore 2
Z Z 2
dy sech2 x dx
= 1
1 − y2 sech2 x 1
Z
1 1+y
= dx = x = arctanh y = ln .
2 1−y -4 -2 0 2 4 0 1
46

47
5.5. b. We begin with the equation 5.6. c. If y ≡ 0 then y′ ≡ 0 and y/(y − 1) ≡ 0 so the dif-
ferential equation is satisfied, implying f4 (x) ≡ 0 is yet
1 2
y = sech x = = or yex + ye−x = 2. another solution to the differential equation.
cosh x ex + e−x
5.6. d. By varying the value of c (or k), we get graphs
This gives ye2x − 2ex + y = 0, which we can solve for ex y = fc, j (x), j = 1, 2, 3, 4 that are horizontal translates of
using the quadratic equation to get one another. Together these graphs cover the (x, y)-plane
p p apart from the line y = 1.
2 ± 4 − 4y2 1 ± 1 − y2 y
ex = = .
2y y
It follows that
p !
1± 1 − y2
x = arcsech y = ln .
y
p
5.5. c. In the last formula, p 1 − y2 will not be real unless
|y| ≤ 1. In that case, −1 ≤ ± 1 − y2 ≤ 1 so x
p
0 ≤ 1 ± 1 − y2 ≤ 2.
p
But the fraction (1 ± 1 − y2)/y needs to be positive, and
that forces 0 < y. Thus, all together, 0 < y ≤ 1.
Because cosh(−x) = cosh(+x) ≥ 1 for all x, we have
0 < sech x ≤ 1 and sech(−x) = sech(+x). Hence the in- 5.7. The method of separation of variables transforms the
verse function x = arcsech y is double-valued and its do- differential equation into
main is 0 < y ≤ 1. The graph of x = arcsech y in part (a)
shows only the positive value. y2 + 1 1
dx = dy = y + dy.
5.5. d. We have − ln B = + ln A if and only if B = 1/A or y y
AB = 1. To show the two formulas for x = arcsech y are This gives functions
negatives of each other, we just need to check that
p p y2 y2
x= + ln y + c1, x= + ln(−y) + c2,
1 + 1 − y2 1 − 1 − y2 1 − (1 − y2) y2 2 2
· = = 2 = 1.
y y y2 y defined on the positive and negative y-axes, respectively.
Because dx/dy > 0 on the positive y-axis, the first func-
5.6. a. The graph of x = y − ln(−y) + 2 for −3.2 ≤ y < 0 tion has an inverse y = ϕ1 (x) > 0. One example is shown
is shown on the left below. The slope of the graph appears below; others are horizontal translates of this one. Like-
to approach −1 as y → −∞; in fact, x → −∞ as y → −∞. wise, dx/dy < 0 on the negative y-axis, so the second
x function has an inverse y = ϕ2 (x) < 0, a reflection of one
of the graphs y = ϕ1 (x) across the x-axis. These are solu-
tions to the original differential equation; they are defined
y for all x. Finally, y = ϕ3 (x) ≡ 0 is a solution. Given any
−1 + k x
point in the (x, y)-plane, precisely one of these functions
−1 + k −1 y = fk,3 (x) passes through that point.
y
−1 y
x
x = y − ln(y) + k
Incidentally, an ordinary plotting program cannot han-
5.6. b. The graph of y = fk,3 (x) is shown on the right dle an inverse function y = f −1 (x) that has no for-
above. It is defined for all x; its range is all y < 0. mula. Nevertheless, if the original function x = f (y)

48 SOLUTIONS: CHAPTER 5. INVERSES
does, then the graph of f −1 can be displayed as the para- For the Jacobian we have
metric plot of ( f (t),t). Show below is a Mathematica ∂ (x, y) −4 tan2 θ tan ϕ sec2 θ sec2 ϕ
ParametricPlot for f (y) = y2 /2 + lny − 1. =
∂ (θ , ϕ ) B4
−4 tan θ tan2 ϕ sec2 θ sec2 ϕ
3
−
2 B4
tan θ − tan ϕ
1
= −4 tan θ tan ϕ sec2 θ sec2 ϕ
B4
-4 -2 2 4
−4 tan θ tan ϕ sec2 θ sec2 ϕ
= .
B3
5.8. a. To begin, we have
In the other direction, we have
y y
tan θ = and tan ϕ = , ∂θ 1 −y −y
x−1 x+1 = · = ,
∂x 1 + y2/(x − 1)2 (x − 1)2 (x − 1)2 + y2
so
tan θ y/(x − 1) x + 1 ∂θ 1 1 x−1
A= = = . = · = ,
tan ϕ y/(x + 1) x − 1 ∂y 1 + y /(x − 1) x − 1 (x − 1)2 + y2
2 2
Hence (A − 1)x = A + 1 or ∂ϕ −y
= ,
∂x (x + 1)2 + y2
A + 1 (tan θ / tan ϕ ) + 1 tan θ + tan ϕ ∂ϕ x+1
x= = = .
A − 1 (tan θ / tan ϕ ) − 1 tan θ − tan ϕ ∂y
=
(x + 1)2 + y2
;
This gives x in terms of θ and ϕ . We also have ∂ (θ , ϕ ) −y(x + 1)

=
A+1 A−1 2A 2 tan θ ∂ (x, y) (x − 1)2 + y2 (x + 1)2 + y2
x+1 = + = = , −y(x − 1)
A − 1 A − 1 A − 1 tan θ − tan ϕ −
(x − 1)2 + y2
(x + 1)2 + y2
and thus
−2y
2 tan θ tan ϕ = .
y = (x + 1) tan ϕ = . (x − 1) + y2 (x + 1)2 + y2
2
tan θ − tan ϕ
We must now replace x and y in this expression by their
This gives y in terms of θ and ϕ . values in terms of θ and ϕ , and then to take the reciprocal
5.8. b. We obtain the partial derivatives by lengthy but of the result. We have
straightforward calculations (or a computer algebra sys- 2 2 tan ϕ 2 4 tan2 ϕ
tem). To simplify expressions, we set B = tan θ − tan ϕ . x − 1 = = , (x − 1) = .
A−1 B B2
∂x (tan θ − tan ϕ ) sec2 θ − (tan θ + tan ϕ ) sec2 θ Earlier computations give
=
∂θ B2 4 tan2 θ 4 tan2 θ tan2 ϕ
−2 tan ϕ sec2 θ (x + 1)2 = 2
, y2 = ,
= , B B2
B2 and thus
∂x (tan θ − tan ϕ ) sec2 ϕ − (tan θ + tan ϕ )(− sec2 ϕ )
= 4 tan2 ϕ 4 tan2 ϕ sec2 θ
∂ϕ B2 (x − 1)2 + y2 = 2
(1 + tan2 θ ) = ,
B B2
2 tan θ sec ϕ
2
4 tan2 θ 4 tan2 θ sec2 ϕ
= , (x + 1)2 + y2 = (1 + tan 2
ϕ ) = .
B2 B2 B2
∂y (tan θ − tan ϕ )2 sec θ tan ϕ − 2 tan θ tan ϕ sec θ
2 2
= In terms of θ and ϕ ,
∂θ B2
−2 tan2 ϕ sec2 θ ∂ (θ , ϕ ) −4 tan θ tan ϕ /B
= , =
B2 ∂ (x, y) 16 tan θ tan2 ϕ sec2 θ sec2 ϕ /B4
2
∂y (tan θ − tan ϕ )2 tan θ sec2 ϕ + 2 tan θ tan ϕ sec2 ϕ −B3 1

= = = ,
∂ϕ B 2 4 tan θ tan ϕ sec θ sec ϕ
2 2 ∂ (x, y)
2 tan2 θ sec2 ϕ ∂ (θ , ϕ )
= 2
.
B as we wished to show.

49
5.9. a. From r12 = (x − 1)2 + y2 and r22 = (x + 1)2 + y2 we 5.10. a. To begin, we note that
get
r22 − r12 = (x + 1)2 − (x − 1)2 = 4x, x2 + y2 = r2 = ρ 2 cos2 ϕ ,
x2 + y2 + z2 = ρ 2 cos2 ϕ + ρ 2 sin2 ϕ = ρ 2 ,
or x = (r22 − r12 )/4. thus x + 1 = (r22 − r12 + 4)/4 and
p
16r22 − (r22 − r12 )2 − 8r22 + 8r12 − 16 Next, with r = x2 + y2 = ρ cos ϕ ,
y2 = r22 − (x + 1)2 =
16 y ρ sin θ cos ϕ z ρ sin ϕ
−(r22 − r12 )2 + 8r22 + 8r12 − 16 = = tan θ , = = tan ϕ ;
= . x ρ cos θ cos ϕ r ρ cos ϕ
16
therefore, in summary we have
Thus we have p
s ρ = x2 + y2 + z2 ,
y
r2 − r12 −(r22 − r12 )2 + 8r22 + 8r12 − 16
x= 2 , y= . θ = arctan ,
4 16 x !
z
ϕ = arctan p .
5.9. b. The partial derivatives of x are x2 + y2
∂x −r1 ∂x r2
= and = . 5.10. b. The Jacobians here are 3 × 3 determinants. In
∂ r1 2 ∂ r2 2
one direction we have (by “expanding along the third
For the partial derivatives of y, it is more convenient to row”)
work with y2 :
cos θ cos ϕ −ρ sin θ cos ϕ −ρ cos θ sin ϕ
∂ (x, y, z)
∂y −2(r22 − r12 ) · (−2r1 ) + 16r1 r1 2 = sin θ cos ϕ ρ cos θ cos ϕ −ρ sin θ sin ϕ
2y = = (r2 − r12 + 4), ∂ (ρ , θ , ϕ ) sin ϕ 0 ρ cos ϕ
∂ r1 16 4
−ρ sin θ cos ϕ −ρ cos θ sin ϕ
∂y −2(r22 − r12 ) · 2r2 + 16r2 −r2 2 = sin ϕ
2y = = (r − r12 − 4). ρ cos θ cos ϕ −ρ sin θ sin ϕ
∂ r2 16 4 2
cos θ cos ϕ −ρ sin θ cos ϕ
+ ρ cos ϕ
Therefore sin θ cos ϕ ρ cos θ cos ϕ
∂ (x, y) r1 r2 2 r1 r2 2 = sin ϕ · ρ 2 sin ϕ cos ϕ + ρ cos ϕ · ρ cos2 ϕ
= (r2 − r12 − 4) − (r − r12 + 4)
∂ (r1 , r2 ) 8 · 2y 8 · 2y 2 = ρ 2 cos ϕ .
r1 r2 −8 −r1 r2
= · = .
2y 8 2y In the other direction
p it will help to simplify terms if we
use ρ and r = x2 + y2 where possible; then (expanding
For simplicity, y stands here in place of its expression in along the third column)
terms of r1 and r2 .
For the Jacobian of r1 and r2 with respect to x and y, we x/ρ y/ρ z/ρ
∂ (ρ , θ , ϕ )
have = −y/r2 x/r2 0
∂ (x, y, z) −xz/rρ 2 −yz/rρ 2 r/ρ 2
∂ r1 x−1 ∂ r1 y
= , = , z −y/r2 x/r2
∂x r1 ∂y r1 =
ρ −xz/rρ 2 −yz/rρ 2
∂ r2 x+1 ∂ r2 y
∂x
=
r2
,
∂y
= ;
r2 r x/ρ y/ρ
+ 2
ρ −y/r2 x/r2
here r1 and r2 stand in for their expressions in terms of x z z r 1 z2 + r2 1
and y. the Jacobian itself is = · 2+ 2· = = .
ρ rρ ρ ρ rρ 3 rρ
∂ (r1 , r2 ) y(x − 1) y(x + 1) −2y 1
= − = = , But ρ 2 cos ϕ = ρ · ρ cos ϕ = ρ · r, so the two Jacobians are
∂ (x, y) r1 r2 r1 r2 r1 r2 ∂ (x, y) reciprocals of each other.
∂ (r1 , r2 )
5.11. Because cylindrical coordinates in (x, y, z)-space
as was to be shown. just use polar coordinates in the (x, y)-plane together with

the z-coordinate, the 3-dimensional Jacobian for the cylin- Therefore

drical coordinate transformation should be the product of ∂ (ρ , θ , ϕ ) ∂ (r, z, θ ) 1 ∂ (ρ , θ , ϕ )
the 2-dimensional Jacobian for the polar coordinate trans- · = = .
∂ (r, z, θ ) ∂ (x, y, z) rρ ∂ (x, y, z)
formation with the 1-dimensional identity transformation
on the z-axis. In fact, √
5.13. We see 10 ≈ 3.162 277 660 168 to 12 decimal

cos θ −r sin θ 0 places. The first table shows that, when x0 = 3, the com-
∂ (x, y, z)
= sin θ r cos θ 0 = r(cos2 θ +sin2 θ ) = r, putation stabilizes after just four iterations.
∂ (r, θ , z) 0 0 1
iterate estimate
∂ (r, θ , z) 1 0 3
and thus = . 1 3.166 666 666 666 7
∂ (x, y, z) r
p 2 3.162 280 701 175 4
5.12. a. The key intermediate variable is r = x2 + y2 . 3 3.162 277 660 169 8
We have 4 3.162 277 660 168 4
p 5 3.162 277 660 168 4
ρ 2 = r2 + z2 , so f (r, z, θ ) = r2 + z2 .
By contrast, when x0 = 10, the computation does not sta-
We also have g(r, z, θ ) = θ and h(r, z, θ ) = arctan(z/r). bilize until after six iterations.
Put into words, iterate estimate
• the value of f is the square root of the sum of the 0 10
squares of its first two arguments; 1 5.5
2 3.659 090 909 090 9
• the value of g is its third argument; 3 3.196 005 081 874 6
4 3.162 455 622 803 9
• the value of h is the inverse tangent of its second ar- 5 3.162 277 665 175 7
gument divided by its first. 6 3.162 277 660 168 4
7 3.162 277 660 168 4
Thus
p √
r = f (x, y, z) = x2 + y2 , 5.14. a. To 8 decimal places, 3 10 ≈ 2.154 434 69, but
convergence here is significantly slower than for the reg-
z = g(x, y, z) = z, ular Babylonian algorithm. Around 30 iterations are
θ = h(x, y, z) = arctan(y/x), needed.
iterate estimate
which states correctly how cylindrical coordinates are re- 0 2
lated to Cartesian. 5 2.159 881 557
5.12. b. Each of the two Jacobians involves the deriva- 10 2.154 265 366
tives of f , g, and h with respect to their three arguments 15 2.154 439 982
in each case. For the first we have (using r + z = ρ ),
2 2 2 20 2.154 435 247
25 2.154 434 952

r/ρ z/ρ 0 30 2.154 434 690
∂ (ρ , θ , ϕ ) 35 3.154 434 690
= 0 0 1
∂ (r, z, θ ) −z/ρ 2 r/ρ 2 0
5.14. b. The weighted algorithm is fast; the estimates sta-
r/ρ z/ρ −1
= −1 · = . bilize after four iterations.
−z/ρ 2 r/ρ 2 ρ
iterate estimate
The dependent and independent variables in the second 0 2
Jacobian stand in the same relation to each others as do 1 2.166 666 667
those of the first Jacobian; hence we can write immedi- 2 2.154 503 362
ately 3 2.154 434 692
∂ (r, z, θ ) −1 4 2.154 434 690
= .
∂ (x, y, z) r 5 2.154 434 690

51
5.14. c. The successful algorithm in part (b) suggests that the (s,t)-plane does indeed carry the positive s-axis to the
we try g3 (x) = (3x + a/x3)/4.
√ The table below shows that image of the positive x-axis.
the fixed point of g3 gives 4 120 to 15 decimal places after t
just five iterations (and x0 = 3). y
x
iterate estimate
y
P
0 3 x α s
1 3.361 111 111 111 111
2 3.310 916 203 751 005
3 3.309 751 534 688 925
4 3.309 750 919 647 044
5 3.309 750 919 646 873
6 3.309 750 919 646 873 5.16. c. The figure below shows the same map P now act-
ing to pull back the (s,t)-coordinate frame to the (x, y)-
5.15. To find the roots of f (x) = x − 3x + 1 using the plane. In the (x, y)-plane, the image of the positive s-axis
3
Newton–Raphson method, we must find the fixed points lies at the angle −α from the positive x-axis.
of y t
f (x) 2x3 − 1
g(x) = x − ′ = .
f (x) 3x2 − 3
t
P
3
Because√ the roots of nearby polynomial x − 3x are at 0 x s
and ± 3 ≈ ±1.7, we use these as initial values x0 or the −α
iterations. The tables below give the estimates xn and in-
dicate how small f (xn ) is. The estimates stabilize to 15
decimal places after four of five iterations.
s
n xn f (xn ) 5.16. d. In general, when the (s,t) coordinate grid is
pulled back by Rθ to the (x, y)-coordinate plane, the new
0 0
grid lies at the angle −θ in relation to the original grid. In
1 0.333 333 333 333 333 3.7 × 10−2
particular, rotation by −θ turns the positive x-axis to the
2 0.347 222 222 222 222 2.0 × 10−4
pulled-back positive s-axis.
3 0.347 296 353 163 868 5.7 × 10−9
4 0.347 296 355 333 861 −5.3 × 10−17 5.17. a. The data is shown below using the Mathematica
PlotLogList command. By default t = 1 for the first
n xn f (xn )
point on the horizontal axis, so t − 1 marks decades since
0 1.7 1790 there; that is, 10(t − 1) = x, years since 1790.
1 1.556 613 756 613 757 1.0 × 10−1
2 1.532 743 354 256 828 2.6 × 10−3
2.0 × 10−6
200
3 1.532 089 372 729 457
100
4 1.532 088 886 238 225 1.1 × 10−12 50
5 1.532 088 886 237 956 −1.8 × 10−16 20
n xn f (xn ) 10
0 −1.7 5 10 15 20
1 −1.909 347 442 680 776 −2.3 × 10−1

2 −1.880 029 751 019 458 −4.9 × 10−3
5.17. b. The points lie roughly on a straight line. The one
3 −1.879 385 549 663 048 −2.3 × 10−6 shown on the plot was chosen, somewhat arbitrarily, to
4 −1.879 385 241 571 887 −5.4 × 10−13 contain the points (1, 6) and (23, 450). Thus, it has the
5 −1.879 385 241 571 817 −6.7 × 10−16 slope (log10 (450) − log10 (6))/220 = 0.008523 (in terms
√ of x), and Y -intercept Y = 6.
5.16. a. We have cos α = 1/2 and sin α = 3/2, we can
take α = π /3. 5.17. c. The slope–intercept information give us the
equation
5.16. b. When the images of the x- and y-axes appear on
the (s,t)-plane, we confirm that rotation by α = 60◦ in Y = 0.008523 x + log10 (6) or y = 6 × 100.008523x.

Note: The plot above suggests that two or three different 5.19. b. The instructions in the exercise have us write
straight lines will provide an even better fit for parts of the 

 ap2 + 2bpq − aq2
data. The graph below shows close-fitting straight lines ∆s = p + ,
for each of the three time periods 1790–1880, 1890–1970, 2(a2 + b2 )
ha :

 −bp2 + 2apq + bq2
and 1980–2010. ∆t = q + .
300
2(a2 + b2)
200
150
100
Thus when (p, q) = (0, 0), we see (∆s, ∆t) = (0, 0); that
70
50 is, ha (0) = 0. Next, when (p, q) = (−a, −b), we have
30
20
a3 + 2ab2 − ab2 a(a2 + b2) −a
∆s = −a +
15
10
2 2
= −a + = ,
2(a + b ) 2(a2 + b2) 2
5 10 15 20 −a2 b + 2a2b + b3 b(a2 + b2 ) −b
∆t = −b + 2 2
= −b + = ,
2(a + b ) 2(a2 + b2 ) 2
5.18. a. If y = ax p , then
confirming that ha (−a) = − 21 a. Finally, when we take
Y = log10 (y) = log10 (a) + p log10 (x) = pX + A,
(p, q) = (−2a, −2b), then
where A = log10 a. Thus the graph of y = ax p becomes
4a3 + 8ab2 − 4ab2 4a(a2 + b2 )
the graph of the straight line with slope p and Y -intercept ∆s = −2a + = −2a + = 0,
A under the map L. 2(a2 + b2 ) 2(a2 + b2)
2 2 3 2 2
5.18. b. The two linear graphs in a log–log coordinate ∆t = −b + −4a b + 8a b + 4b = −2b + 4b(a + b ) = 0.
plane are show below, the first with positive slope 2/3 and 2(a2 + b2 ) 2(a2 + b2)
the second with negative slope −2. The slopes appear the confirming that ha (−2a) = 0.
way they do because the two axes have different scales.
100.0
5.19. c. We can evaluate ha at the two points simultane-
50.0 ously. That is, take p = −a(1 ± ε ), q = −b(1 ± ε ); then
10.0
5.0
∆s = −a(1 ± ε )
a3 (1 ± ε )2 + 2ab2(1 ± ε )2 − ab2(1 ± ε )2
1.0
+
0.5
2(a2 + b2)
a(a2 + b2 )(1 ± ε )2
0.10 0.15 0.20 0.30 0.50 0.70 1.00 1.50 2.00
= −a(1 ± ε ) +
2(a2 + b2 )
The pullbacks in the (x, y)-plane are the power functions
a aε 2 −a + aε 2
y = 4x 2/3
and y = 1/x ; 2 = −a ∓ aε + ± aε + = ,
2 2 2
their graphs are the following. ∆t = −b(1 ± ε )
−a2 b(1 ± ε )2 + 2a2b(1 ± ε )2 + b3(1 ± ε )2
12 +
10
2(a2 + b2)2
b(a2 + b2 )(1 ± ε )2
= −b(1 ± ε ) +
8
6 2(a2 + b2 )
bε 2 −b + bε 2
4
b
2 = −b ∓ bε + ± bε + = .
2 2 2
1.0 1.5 2.0
Thus for any ε we have
Comment: Both figures are made by Mathematica, the
upper one with the LogLogPlotcommand. In fact, that ε2 − 1
ha − a(1 + ε ) = a = ha − a(1 − ε ) ;
command takes as input the nonlinear functions (i.e., the 2
pullbacks). the two points have the same image.
5.19. a. For every u we have 5.19. d. Because ha fails to be 1–1 on any neighborhood
2 2
of the point −a, the point −a = (−a, −b) must be ex-
f(−u) = (−u) − (−v) , 2(−u)(−v)
cluded from W . Therefore, if l is the length of a side of W ,
= u2 − v2 , 2uv = f(u). then l/2 < max |a|, |b| gives us an upper bound.

53
5.19. e. The derivative d(ha )p is given by the matrix Let a⊥ = (−b, a); then we can rewrite the angle as
 ⊥
ap + bq −bp + aq  a ·p
1+ 2 − θ = arctan
 a + b2 a2 + b2  a · (a + p)
 .
−bp + aq ap + bq
1 + According to this formula, θ = 0 when a⊥ · p = 0, that is,
a2 + b2 a2 + b2 when p is on the line through the origin that is perpendic-
This has the general form ular to the vector a⊥ . Moreover, because a · (a + p) > 0
in W , the sign of θ is determined solely by the sign of the
A −B
numerator. That is, θ > 0 if and only if p is on the same
B A
side of the line a⊥ · p = 0 as a⊥ itself.
of a dilation–rotation matrix if (A, B) 6= (0, 0). In that The (∆u, ∆v)-plane from the figure on page 164 of the
text appears below. The vector 12 a⊥ is shown together
case, the rotation√ angle is θ = arctan(B/A) and the di-
lation factor is A2 + B2 (cf. p. 118 of the text). Thus with the line a⊥ · ∆u = 0 perpendicular to it. (Recall
if we determine where A = B = 0 we will know when that p = (p, q) has replaced ∆u = (∆u, ∆v) in the solu-
d(ha )p fails to be a dilation–rotation matrix. We thus have tion to this exercise.) In “straightening out” the curvilin-
to solve the pair of linear equations ear grid in the (∆u, ∆v)-plane, the map ha must indeed
rotate points above the line a⊥ · ∆u = 0 counterclockwise
ap + bq = −(a2 + b2), (i.e., θ > 0) and points below, clockwise.
−bp + aq = 0,
∆t
for p and q. The solution p = −a, q = −b. That is,
d(ha )p is a dilation–rotation matrix everywhere except at
the point p = −a. By construction, this point lies out-
side W .
√
5.19. f. The dilation factor is A2 + B2 , where ∆s
(a2 + b2 + ap + bq)2 + (−bp + aq)2 p
A2 + B2 =
(a2 + b2)2
We simplify the numerator in the following (lengthy but
straightforward) computation:
h
(a2 + b2 + ap + bq)2 + (−bp + aq)2 _1 ⊥
W 2
a ∆v ∆t
= (a2 + b2)2 + 2(a2 + b2)(ap + bq)
+ a2 p2 + 2abpq + b2q2 + b2 p2 − 2abpq + a2q2
a⊥ ⋅ ∆u = 0
= (a2 + b2)2 + 2(a2 + b2)(ap + bq)
+ (a2 + b2 )(p2 + q2 ) ∆s
2 2 2 2 2 2
∆u
= (a + b ) a + b + 2ap + 2bp + p + q
p
= (a2 + b2) (a + p)2 + (b + q)2 .
Therefore the dilation factor for d(ha )p is
s
p (a + p)2 + (b + q)2
A2 + B2 = . 5.20. a. The derivative of s is
a2 + b2
cos x cosh y sin x sinh y
This is zero (and thus d(ha )p is the zero matrix) if and ds(x,y) = ,
− sin x sinh y cos x cosh y
only if p = −a.
and its Jacobian is
5.19. g. As mentioned in the solution to part (e), the rota-
tion angle is J(x, y) = cos2 x cosh2 y + sin2 x sinh2 y
= cos2 x(1 + sinh2 y) + sin2 x sinh2
B −bp + aq
θ = arctan = arctan 2 . = cos2 x + sinh2 y.
A a + b2 + ap + bq

Thus J > 0 unless cos x = 0 and sinh y = 0 simultaneously. each region marked of by two adjacent grid lines in each
On the given domain W this happens only when x = ±π /2 direction is roughly square-shaped, independent of the
and y = 0, i.e., only at the two points (±π /2, 0). spacing of the adjacent grid lines. These features suggest
that the map s defining the grid is conformal. The points
5.20. b. The map s is conformal at any point where its
(±1, 0) that are the common focal points of all the ellipses
derivative is given by a dilation–rotation matrix. In fact,
and hyperbolas appear on the horizontal axis halfway be-
the derivative matrix has the form
tween the center of the window and its vertical sides.
A −B A = cos x cosh y, 2 20
with ,
B A B = − sin x sinh y
1 10
and this is a dilation–rotation whenever A2 + B2 6= 0. But
A2 + B2 = J 6= 0 everywhere in W except the two points 0 0
(±π /2, 0).
5.20. c. If y = b 6= 0, then we can write u/ cosh b = sin x -1 -10
and v/ sinh b = cos x, so

u 2 v 2
-2
-2 -1 0 1 2
-20
-20 -10 0 10 20
+ = sin2 x + cos2 x = 1
cosh b sinh b
5.20. g. The same grid on the much larger square is
for all x. The locus is an ellipse in the (u, v)-plane whose
shown on the right above. The common focal points of the
semimajor and semiminor axes have lengths cosh b and
hyperbolas are now very close to the center of the window,
sinh b, respectively.
so the hyperbolas look—at this scale—like the radial lines
If b > 0 then sinh b > 0 and hence v = cos x sinh b ≥ 0
of a polar coordinate grid. Because the same focal points
(because cos x ≥ 0 for |x| ≤ π /2). The points (u, v) there-
serve for the ellipses, those ellipses look like circles. In a
fore lie on the upper half of the ellipse. If b < 0 then v < 0
polar coordinate grid, the spacing between circles would
and the points lie on the lower half. If b = 0 then v = 0
be constant, but the spacing of these ellipses is not. In fact,
and the image is just the line segment −1 ≤ u ≤ 1 on the
the spacing makes all regions between adjacent grid lines
u-axis.
have roughly the same (square) shape. This is a manifes-
5.20. d. If we set x = a and 0 < |a| < π /2, then we can tation of conformality on the larger scale.
write u/ sin a = cosh y and v/ cos a = sinh y, so 5.21. a. Because ev > 0 and e−v > 0 for all v, it is clear
u 2 v 2 that h(U) ⊆ Q. It remains to show h(U) ⊇ Q, i.e., that h
− = cosh2 y − sinh2 y = 1 is onto. Thus let (x, y) be any point in Q. We have
sin a cos a
for all y. This time the locus is the hyperbola whose y ev
xy = u2 and = −v = e2v ;
asymptotes are v = ±(cota)u; the two branches of the hy- x e
perbola lie to the right and the left of the asymptotes. √
therefore, if u = xy, v = 12 ln(y/x), then h(u, v) = (x, y)
If a is positive then sin a > 0 and u = sin a cosh y > 0.
so (x, y) is in h(U).
The points (u, v) thus lie on the right branch of the hy-
perbola. If a is negative then u = sin a cosh y < 0 and the 5.21. b. Work already done shows
points (u, v) lie on the left branch. If a = 0 then u = 0 ( √
and the image is the nonnegative v-axis. If a = π /2 then −1
h :
u = xy,
for every (x, y) in U.
v = 0 while u = + cosh y ≥ 1. The image is the ray u ≥ 1 v = 12 ln(y/x),
on the positive u-axis. For each y > 0, the points (π /2, y)
and (π /2, −y) have the same image. If a = −π /2 then 5.21. c. We have
y = 0 once again while u = − cosh y ≤ −1. The image is
the ray u ≤ −1 on the negative u-axis, and (−π /2, y) and ∂ (x, y) e−v −ue−v
= v = 2u;
(−π /2, −y) have the same image. ∂ (u, v) e uev
1p p
5.20. e. The preceding analysis shows that s is 1–1 every- ∂ (u, v) = 2 y/x 2 x/y = √1 + √1 = √1 .
1
∂ (x, y) −1/2x 1/2y 4 xy 4 xy 2 xy
where away from the the boundary lines x = ±π /2.
√
5.20. f. The grid lines in the curvilinear grid on the (u, v)- Because u = xy the Jacobians are reciprocals of each
plane (shown on the left below) meet at right angles, and other.

55
5.21. d. The curvilinear (u, v)-coordinate grid on Q is 5.22. c. The Jacobians are
shown on the left below; u = constant on the hyperbolas
∂ (p, q) es cosht es sinht
and v = constant on the radial lines. The curvilinear (x, y)- = s = e2s ,
∂ (s,t) e sinht es cosht
grid on U is show on the right; the x = constant grid lines
are the ones that are concave down. Because the curvilin- ∂ (s,t) p/(p2 − q2 ) −q/(p2 − q2)
= 1
= .
ear grid lines on Q are not mutually orthogonal, the map h ∂ (p, q) 2 2
−q/(p − q ) p/(p − q )2 2
p − q2
2
cannot be conformal. A further proof is that the derivative

does not have the form of a dilation–rotation. Because e2s = p2 − q2 , the Jacobians are reciprocals of
5 2 each other.
5.22. d. We compute p◦ h◦ u by making substitutions de-
4
1 fined by the sequence of maps p, h, and u.
3
y+x ev + e−v
0 p= =u = u cosh v = es cosht,
2 2 2
y−x ev − e−v
-1 q= =u = u sinh v = es cosht.
1 2 2
0 -2 Shown below is the image of the (s,t)-coordinate grid un-
0 1 2 3 4 5 0 1 2 3 4
der the successive maps u, h ◦ u, and p ◦ h ◦ u. The last
5.22. a. The image under m of the vertical line s = a is clearly agrees with the image of the (s,t)-coordinate grid
the parametrized curve under the map m.
5
p = ea cosht, q = ea sinht, 4
that lies on the hyperbola p2 − q2 = e2a whose asymptotes 4
2
are q = ±p. Because p > 0, the curve is the right half of 3
the hyperbola. As a runs from −∞ to ∞, these hyperbolas 0
fill up the quadrant 2
-2
0 < p,
Q1 : 1
−p < q < p. -4
The image under m of a horizontal line t = b is the radial 0 2 4 6 8 10

0
0 1 2 3 4 5
line 4
p = (cosh b)es > 0, q = (sinh b)es ,
with slope q/p = tanh b. Because −1 < tanh b < 1, the 2
radial lines also fill up Q1 . Here is the (s,t)-coordinate

0
grid in Q1 .
4 -2
2 -4
0 2 4 6 8 10
0
5.23. a. The arctangent function we use is the one in-

troduced on page 59 of the text. In particular, θ (u, v) =
-2
-4
arctan(v/u) takes values between 0 and π /2 in the first
quadrant and between π /2 and π in the second quadrant.
0 2 4 6 8 10
That is,

5.22. b. Because p2 − q2 = e2s (cosh2 t = sinh2 t) = e2s on y
the one hand, and q/p = tanht on the other, we have 0 < θ = arctan < π when y > 0.
x−1
(
−1 s = 21 ln(p2 − q2 ), The figure accompanying Exercise 5.8 (text p. 178) also
m :
t = arctanh(q/p). makes it clear that θ runs up from 0 to π as x runs

down from +∞ to −∞ (e.g., with y > 0 fixed). That is, 5.23. c. In the curvilinear (x, y)-coordinate grid in T , the
arctan y/(x − 1) maps the upper half-plane onto the inter- curves x = a form three separate “bundles” that radiate
val 0 < θ < π . out from the point (θ , ϕ ) = (π /2, π /2) to the three cor-
Next, note that x + 1 > x − 1 for all x; therefore, with ners of T . Curves in the bundle with a < −1 go to the
y > 0 we have y/(x + 1) < y/(x − 1) and thus upper right corner; those in the bundle with −1 < a < 1
go to the lower right corner. The horizontal line for which
y y a = −1 separates these two bundles. The curves in the
0 < ϕ = arctan < arctan = θ.
x+1 x−1 bundle with 1 < a go to the lower left corner. The vertical
line for which a = 1 separates this bundle from the one
Furthermore, if we fix y/(x − 1) = K (so θ is fixed) while whose curves go to the lower right corner.
letting y → ∞, then x → ∞ and The curves y = b cross the curves x = a; they look
y y x−1 x−1 somewhat like hyperbolas with common asymptotes the
= · =K· → K; lines ϕ = 0 and θ = π .
x+1 x−1 x+1 x+1
Therefore, 5.23. d. Both a and a−1 are orientation-reversing, be-
cause their Jacobians are negative; we established this in

y the solution to Exercise 5.8
ϕ = arctan → arctan K = θ .
x+1 Here is one way to describe the action of a−1 . Open
out the right angle of T to a straight (i.e. 180◦) angle.
In other words, for each fixed θ , the angle ϕ takes all Place the two now-opened-out sides of T along the x-axis
values between 0 and θ . We conclude that a maps U onto so the upper right corner of T is sent to x = −∞, the lower
the open triangular region T . left corner to x = +∞, and the straight angle to x = 0.
5.23. b. If x = 1, then θ = π /2. This the the gray vertical Points on the hypotenuse of T are sent to y = +∞. Do
line within T in the figure accompanying the exercise. If this so that the curves x = a in T become vertical and the
x = −1, then ϕ = π /2. This is the gray horizontal line curves y = b become horizontal. The reversal of orienta-
within T . tion is manifested by the way the two short sides of T are
When x < −1, then (x + 1, y) and (x − 1, y) both lie in mapped.
the second quadrant. If now y → 0, then these points ap- 5.23. e. The (θ , ϕ )-coordinate grid on the (x, y)-plane
proach the negative x-axis to the left of x = −1. By defi- just consists of radial lines from the two points (1, 0) and
nition of the arctangent function, (−1, 0).
y
y y
ϕ = arctan → π , θ = arctan → π.
x+1 x−1
In other words, a(x, y) approaches the upper right corner

of T , where (θ , ϕ ) = (π , π ). x
When −1 < x < 1, then (x + 1, y) now lies in the first
quadrant while (x − 1, y) continues to lie in the second.
Therefore, if y → 0, then

y y
ϕ = arctan → 0, θ = arctan → π.
x+1 x−1
5.24. a. It is clear From the geometric definition of r1
In other words, a(x, y) approaches the lower right corner and r2 in Exercise 5.9 that r1 + r2 > 2. To show the same
of T , where (θ , ϕ ) = (π , 0). computationally, we note first that, because y > 0, we have
Finally, when 1 < x, then both of the points (x + 1, y) q q
and (x − 1, y) lie in the first quadrant, so if y → 0, then r1 + r2 > (x − 1)2 + (x + 1)2 = |x − 1| + |x + 1|.

y y
ϕ = arctan → 0, θ = arctan → 0. By definition,
x+1 x−1
( (
In other words, a(x, y) approaches the lower left corner |x − 1| = x − 1 1 ≤ x,
|x + 1| =
x+1 −1 ≤ x,
of T , where (θ , ϕ ) = (0, 0). −x + 1 x ≤ 1; −x − 1 x ≤ −1.

57
Putting these together, we get Because S is defined by the inequalities |r2 − r1 | < 2 and
 r2 + r1 > 2, it follows that q(r1 , r2 ) > 0 on S, so the for-

2x 1 ≤ x, mula b−1 provides for y is defined on all of S. Shown
|x − 1| + |x + 1| = 2 −1 ≤ x ≤ 1, below is the curvilinear (x, y)-coordinate grid on S; the

 grid lines are not mutually perpendicular.
−2x x ≤ −1.
5
In all cases, |x − 1| + |x + 1| ≥ 2, so r1 + r2 > 2.
To show that |r2 − r1 | < 2, we begin by noting 4
|r22 − r12 | |(x − 1)2 − (x + 1)2| 3

|r2 − r1 | = <
r2 + r1 |x − 1| + |x + 1|

(|x − 1| + |x + 1|)|x − 1| − |x + 1| 2
=
|x − 1| + |x + 1|
1
= |x − 1| − |x + 1|.

0
We now break this down the way we did |x − 1| + |x + 1| 0 1 2 3 4 5
above; the result is

 5.24. c. The (r1 , r2 )-grid on U, shown below, consists of

−2 1 ≤ x, two families of concentric circles; one family is centered
|x − 1| − |x + 1| = −2x −1 ≤ x ≤ 1, at (1, 0), the other at (−1, 0). The grid is defined on the

 entire (x, y)-plane.
2 x ≤ −1.
y

Thus |r2 − r1 | < |x − 1| − |x + 1| ≤ 2.
5.24. b. The object is to show that b has a well defined in-
verse everywhere on the half-infinite strip S. The formula
x
for b−1 was found in Exercise 5.9; we must simply verify
that b−1 as found is defined everywhere in S. We have


 r2 − r12
x = 2 ,
b−1 : r4 .
 2 2 2 2 2
y = −(r2 − r1 ) + 8(r1 + r2 ) − 16 .

16 5.24. d. The image of the x-axis under the map b is the
boundary of S; the points (±1, 0) map to the corners of
The polynomial giving x is defined for all r1 and r2 ; the
that boundary. In detail, we have y = 0 and the map b
challenge is to show that the numerator of the radical ex-
reduces to
pression for y is defined everywhere on S. q q
A Mathematica plot of the zero-level contour of the r1 = (x − 1)2 = |x − 1|, r2 = (x + 1)2 = |x + 1|.
quartic polynomial
On the interval x ≤ −1, the map b defines
q(r1 , r2 ) = −(r22 − r12 )2 + 8(r12 + r22 ) − 16
r1 = −x + 1, r2 = −x − 1,
consists of the four lines v − u = ±2, v + u = ±2. It fol-
lows that q is a product of four linear factors; in fact, which is the portion of the straight line r2 = r1 − 2 for
which r1 ≥ 2. On the interval −1 ≤ x ≤ 1, the map b
−(r2 − r1 − 2)(r2 − r1 + 2)(r2 + r1 − 2)(r2 + r1 + 2) defines
= [4 − (r2 − r1 )2 ][(r2 + r1 )2 − 4] r1 = −x + 1, r2 = x + 1,
= −(r2 − r1 )2 (r2 + r1 )2 + 4(r2 + r1 )2 + 4(r2 − r1 )2 − 16 which is the portion of the line r2 = −r1 + 2 for which
0 ≤ r1 ≤ 2. On the last interval, 1 ≤ x, the map b defines
= −(r22 − r12 )2 + 4r22 + 8r2 r1 + 4r12
+ 4r22 − 8r2 r1 + 4r12 − 16 r1 = x − 1, r2 = x + 1,
= −(r22 − r12 )2 + 8r22 + 8r12 − 16 = q(r1 , r2 ). which is the portion of the line r2 = r1 + 2 with 0 ≤ r1 .

5.25. a. To determine the image V 4 = σ (U 4 ), it is help- and the second along its third row. The result is
ful to see how σ , when suitably restricted, defines polar
coordinates in R2 and spherical coordinates in R3 . −rs1 s2 c3 −rs1 c2 s3
det dσ (r,t) = rs1 c2 s3 c3
First set t2 = t3 = 0, and let R2 : (x1 , x2 , 0, 0). Then σ rc2 c3 −rs2 s3

gives polar coordinates on R2 : −rc1 s2 c3 −rc1 c2 s3
+ rc1 c2 s3 c3
( rc2 c3 −rs2 s3
x1 = r cost1 ,

2 −rs1 c2 c3 −rc1 s2 c3

σ (r,t1 , 0, 0) : + rs2 c3
x2 = r sint1 . rc1 c2 c3 −rs1 s2 c3

2

2 c1 c2 c3 −rs1 c2 c3

The image fails to cover points in R where t1 = ±π , that
2 + r c2 c3
s c c rc c c
1 2 3 1 2 3
is, points of the form (x1 , x2 ) = (−a2 , 0). This is a half-
line in the R2 -plane. = r c2 s3 c3 (s1 + c1 ) + r c2 c3 (s2 + c22 )
3 2 2 2 2 3 4 2
Now require only t3 = 0 and let R3 : (x1 , x2 , x3 , 0). Then = r3 c2 c23 (s23 + c23)
σ gives ordinary spherical coordinates on R3 :
= r3 cost2 cos2 t3 .


x1 = r cost1 cost2 ,
σ (r,t1 ,t2 , 0) : x2 = r sint1 cost2 , 5.25. c. Because 0 < r, |t2 | < π /2, and |t3 | < π /2 in U 4 ,

 we see that det dσ (r,t) 6= 0 at every point of U 4 , so dσ (r,t)
x3 = r sint2 .
is invertible at every point (r, t) in U 4 . By the inverse func-
tion theorem, σ is locally invertible at every point of U 4 .
The image fails to cover points in R3 where t1 = ±π
or t2 = ±π /2, that is, points of the form (x1 , x2 , x3 ) = 5.25. d. We must solve the equations for r, t1 , t2 , and t3 .
(−a2 , 0, b). This is a half-plane in R3 . A straightforward calculation gives x21 + x22 + x23 + x24 = r2 ,
When we allow both t2 and t3 to be nonzero and con- so we have q
sider the full map σ : U 4 → R4 , the image fails to cover r = x21 + x22 + x23 + x24 .
points in R4 where t1 = ±π or t2 = ±π /2 or t3 = ±π /2.
These points have the form We also see x2 /x1 = tant1 , so

(x1 , x2 , x3 , x4 ) = (−a2 , 0, b, c), a, b, c real, x2
t1 = arctan .
x1
and constitute a “half-hyperplane” H in R4 . Thus we find q
V 4 = R4 \ H. Next, because x21 + x22 = |r cost2 cost3 | = r cost2 cost3
(all factors inside the absolute value signs are positive be-
5.25. b. With the abbreviations si = sinti , ci = costi , we
cause of the restrictions on r, t2 , and t3 ), we have
can write the derivative of σ in the clear but condensed
form  
  x x
q 3 = tant2 , so t2 = arctan  q
3 .
c1 c2 c3 −rs1 c2 c3 −rc1 s2 c3 −rc1 c2 s3 2 2 2 2
s1 c2 c3 rc1 c2 c3 −rs1 s2 c3 −rs1 c2 s3  x 1 + x 2 x 1 + x 2
dσ (r,t) =  s2 c3
.
0 rc2 c3 −rs2 s3  q
s3 0 0 rc3 Finally, because x21 + x22 + x23 = r cost3 , we have
We can evaluate the determinant by beginning with an ex- x4

q = tant3 ,
pansion along the fourth row: x1 + x22 + x23
2

−rs1 c2 c3 −rc1 s2 c3 −rc1 c2 s3
and thus
det dσ (r,t) = −s3 rc1 c2 c3 −rs1 s2 c3 −rs1 c2 s3  
0 rc2 c3 −rs2 s3
x 4
c1 c2 c3 −rs1 c2 c3 −rc1 s2 c3 t3 = arctan  q .
2 2
x1 + x2 + x32
+ rc3 s1 c2 c3 rc1 c2 c3 −rs1 s2 c3 .
s2 c3 0 rc2 c3
These equations define σ −1 on V 4 .
Now expand first 3 × 3 determinant along its first column

Implicit Functions
3
6.1. a. Because
√ f √is the product of three linear factors,
namely ( 3y − x)( 3y + x)(x − 1), its zero-level contour 2
consists of the three line

√ √ 1
3y − x = 0, 3y + x = 0, x − 1 = 0.
0
The intersection of any two of these is a saddle point √ -1

of f . The saddle
√ points are therefore (0, 0), (1, 1/ 3),
and (1, −1/ 3). To find the relative maximum, we set -2
the partial derivatives equal to zero. If we write f (x, y) as

3xy2 − x3 − 3y2 + x2 , then we find
-3
-3 -2 -1 0 1 2 3
fx = 3y2 − 3x2 + 2x = 0, fy = 6xy − 6y = 6y(x − 1) = 0. 6.3. Given exy = e, it follows that xy = 1. Near (2, 1/2)
the solution is y = 1/x and thus dy/dx = −1/x2. The full
If we set y = 0 then fy = 0 and fx = −3x2 + 2x = 0, imply- locus exy = e is both branches of the hyperbola xy = 1; see
ing either x = 0 (a known saddle point) or x = 2/3. Thus above.
the relative maximum is at (2/3, 0).
6.4. We can solve y2 − 2y cosx − sin2 x = 0 for y using
6.1. b. The graph of z = f (x, y) rises in a bulge above the the quadratic formula:
plane z = 0 in the central triangle formed by the three in- p
tersecting lines of the locus f (x, y) = 0 of part (a). The 2 cos x ± 4 cos2 x + 4 sin2 x
y= = cos x ± 1.
relative maximum is at the top of the bulge, and the sad- 2
dles are at the three corners of the triangle. Thus dy/dx = − sin x. The zero locus consists of two
separate parallel curves (namely, vertical translates of
y = cos x).
3
2
0.0
-0.2
0.5 1
-0.4
-0.6
0.0 0
0.0
0.5 -0.5 -1
1.0
-2
6.2. Given that exy = 1, it follows that xy = 0. The point -3

(x, y) = (2, 0) satisfies this condition. We must have y = 0 -3 -2 -1 0 1 2 3
if xy = 0 while x is near 2. In other words, the implicitly 2

defined function is y = f (x) = 0; hence dy/dx = 0. The 6.5. The change in sign of sin x term means that
p
full locus exy = 1 is just the pair of coordinate axes; see 2 cos x ± 4 cos2 x − 4 sin2 x √
below. y= = cosx ± cos 2x.
2
59

60 SOLUTIONS: CHAPTER 6. IMPLICIT FUNCTIONS
These two functions are undefined when cos 2x < 0; this When x = 2 these formulas give y = 1, −20/8; thus for
happens when the function ϕ (x) we must use the plus sign:
√
π 3π −3x + 224 − 7x2
2π n + < 2x < 2π n + , n integer, ϕ (x) = .
2 2 8
or
π 3π Therefore
π n + < x < π n + , n integer.
4 4 3 7x 1
ϕ ′ (x) = − − √ , ϕ ′ (2) = − .
The derivatives are 8 8 224 − 7x2 2
dy sin 2x The derivatives ϕ ′ (x) and dx/dy are reciprocals at corre-
= − sin x ∓ √ .
dx cos 2x sponding points. In particular, at the point (x, y) = (2, 1),
They become infinite at points x where cos 2x = 0, that is, ϕ ′ (2) = −1/2 = dy/dx, while dx/dy = −2.
at the points x = π /4 + nπ /2. The locus

6
y2 − 2y cosx + sin2 x = 0 4
consists of a sequence of congruent ovals; each has width 2
π /2, and each is separated from the next by an empty ver- 0

tical band of width π /2. Points where the ovals are verti-
cal have x = π /4 + nπ /2, n integer.
-2
-4
9Π 7Π 5Π 3Π Π Π 3Π 5Π 7Π 9Π
- - - - -
4 4 4 4 4 4 4 4 4 4
-6
4
-6 -4 -2 0 2 4 6
2
0
6.7. The linearization of the locus f (x, y) = 0 at (a, b) is
L(∆x, ∆y) = fx (a, b)∆x + fy (a, b)∆y = 0, where we have
-2 ∆x = x − a and ∆y = y − b. If the linearization is not iden-
tically zero, its zero locus is the straight line tangent to the
-4
9Π 7Π 5Π 3Π Π Π 3Π 5Π 7Π 9Π
curve f = 0 at the point (a, b). The linearization is tangent
- - - - -
4 4 4 4 4 4 4 4 4 4 in parts (a), (c), and (g).
6.7. a. L = ∆x.
6.6. a. We apply the quadratic formula to the equation
6.7. b. L = 0.
f (x, y) − 14 = x2 + 3y · x + 4y2 − 14 = 0
6.7. c. L = −∆x.
to get
p p 6.7. d. e. & f. L = 0.
−3y ± 9y2 − 16y2 + 56 −3y ± 56 − 7y2 6.7. g. L = 3∆x + 3∆y.
x= = .
2 2
6.8. a. Contours of f in W are shown below. We have
The derivative is fx = y2 and fy = 2xy; both these expressions are nonzero
dx 3 7y when 1 ≤ x ≤ 2 and 1 ≤ y ≤ 2. Thus every point of W is
=− ∓ p
dy 2 2 56 − 7y2 a regular point of f .
2.0
When y = 1, x itself has two possible values, namely −5
and +2. The value of the derivative at each of these points 1.8
is −1 and −2, respectively.
6.6. b. The locus f (x, y) = 14 is shown below. 1.6
6.6. c. We solve f (x, y) − 14 = 0 for y applying the 1.4

quadratic formula once again:
√ √ 1.2
−3x ± 9x2 − 16x2 + 224 −3x ± 224 − 7x2
y= = .
8 8 1.0
1.0 1.2 1.4 1.6 1.8 2.0

61
6.8. b. According to the proof of Theorem 6.2, the map Shown on the left above is the window W with contours
h : W → R2 is ( that meet either its top or its bottom edges. The image
u = x, of W in the (u, v)-plane is the quadrilateral shown at the
h: 2
v = xy . same scale to its right. The image contours are horizontal
lines that meet one or the other of the sloping sides of the
A level curve of f satisfies the condition xy2 = c, where c image quadrilateral.
p constant. We can parametrize this curve as (x, y) = 6.8. d. The inverse h−1 : h(W ) → W : (u, v) → (x, y) is
is any
(t, c/t), 1 ≤ t ≤ 2. Its image under h is then
( given by (
p u = x = t, x = u,
h(t, c/t) : p −1
h : p
v = xy2 = t · ( c/t)2 = c, y = v/u.
the parametrization of a horizontal in the (u, v)-plane. 6.9. a. Here fx = 2x and fy = 2y, so every point other
6.8. c. Under the action of h, the vertical lines x = 1 and than the origin is a regular point of f . Therefore every
x = 2 map to the vertical lines u = 1 and u = 2. The hori- point of W is a regular point of f . The contours of f in W
zontal lines y = 1 and y = 2 map to the lines are circular arcs, as shown on the left below.
v = x · 1 = u, v = x · 4 = 4u.
The levels of f increase in value from f (1, 1) = 1 at the
lower left corner of W to f (2, 2) = 8 at the upper right.
The level of f at the lower right is f (2, 1) = 2, so the level
curve f (x, y) = c meets the bottom of W if 1 ≤ c ≤ 2. The
level of f at the upper left is f (1, 2) = 4, so f (x, y) = c
meets the top of W if 4 ≤ c ≤ 8. The spacing between
levels is ∆c = 0.25.
2.0
1.8
1.6 h
1.4 −→
1.2
1.0
1.0 1.2 1.4 1.6 1.8 2.0
2.0
1.8
1.6 h
1.4 −→
1.2
1.0
1.0 1.2 1.4 1.6 1.8 2.0
6.9. b. The map h : W → R2 that straightens level curves

is (
u = x,
h:
v = x2 + y2 .
We√can parametrize the level curve x2 + y2 = c as (x, y) =
(t, c − t 2), 1 ≤ t ≤ 2. Its image under h is then
(
p u = t,
2
h(t, c − t ) :
v = t 2 + c − t 2 = c,
the parametrization of a horizontal line in the (u, v)-plane.

6.9. c. Under the action of h, the vertical lines x = 1 and (x, y) = (t, k), 0 ≤ t ≤ 2; its image is (u, v) = (k,t 2 + k2 ),
x = 2 map to the vertical lines u = 1 and u = 2. The hori- a vertical line segment with k2 ≤ v ≤ k2 + 4.
zontal lines y = 1 and y = 2 map to the parabolas
6.11. a. By Corollary 6.4 (text p. 193), we have
v = x2 + 1 = u2 + 1, v = x2 + 4 = u2 + 4. − fu (u, v, ϕ (u, v))
ϕu (u, v) = ,
fw (u, v, ϕ (u, v))
The image is shown on the right in the previous figure,
above. Level curves of f have been mapped to horizontal and similarly for ϕv . Expressing, for the moment, the par-
lines. tial derivatives of f using w instead of ϕ (u, v), we have
6.10. a. The function f is the same as in Exercise 6.9. −2 1+w 1−v
The solution there establishes that the only critical point fu = · · ,
(1 + u)2 1 − w 1 + v
of f is at the origin. We have fy = 2y = 0 at every point −2 1+w 1−u
on the line y = 0, in particular, at the center (1, 0) of Z. fv = 2
· · ,
(1 + v) 1 − w 1 + u
Level curves of f in Z are shown below left.
2 1−u 1−v
fw = · · .
(1 − w)2 1 + u 1 + v
Therefore,
−2 1+w 1−v
− · ·
2
(1 + u) 1 − w 1 + v 1 − w2 1 − ϕ 2
1.0 ϕu = = = ,
2 1−u 1−v 1 − u2 1 − u2
0.5 · ·
0.0 h (1 − w)2 1 + u 1 + v
−→
1 − ϕ2
ϕv =
-0.5
.
-1.0 1 − v2
0.0 0.5 1.0 1.5 2.0
The equation for ϕv follows immediately from the equa-
tion for ϕu , given the symmetric way in which u and v
appear in f . Because ϕ (0, 0) = 0 by construction, and
ϕu (0, 0) = ϕv (0, 0) = 1 by the formulas just obtained, the
Taylor expansion of ϕ (u, v) at (u, v) = (0, 0) is
6.10. b. The given map h is defined by polynomial func-
tions so it is continuously differentiable everywhere. Be- ϕ (u, v) = 0 + 1 ·(u − 0)+ 1 ·(v− 0)+ O(2) = u + v+ O(2).
cause
6.11. b. We solve f (u, v, w) = 1 for w. To simplify the
0 1 0 1 computations, let
dh(1,0) = =
2x 2y (x,y)=(1,0) 2 0
(1 − u)(1 − v) 1+w
A= ; then ·A = 1
is invertible, the inverse function theorem (Theorem 5.2, (1 + u)(1 + v) 1−w
text p. 169) guarantees that h itself has a continuously dif-
and A + AW = 1 − w. Hence
ferentiable inverse on a neighborhood of h(1, 0) = (0, 1)
in the (u, v)-plane. (1 − u)(1 − v)
1−
6.10. c. The level curve f (x, y) = c > 0 in W under the 1 − A (1 + u)(1 + v)
w= =
action of the map h becomes a horizontal straight line in 1 + A (1 − u)(1 − v)
1+
h(W ); see√above right. We parametrize the level curve as (1 + u)(1 + v)
2
(x, y) = ( c − t ,t), with −1 ≤ t ≤ 1. Its image is then (1 + u)(1 + v) − (1 − u)(1 − v)
=
( (1 + u)(1 + v) + (1 − u)(1 − v)
p u = t, u+v
h( c − t 2,t) : = .
v = c − t 2 + t 2 = c, 1 + uv
the parametrization of a horizontal line in the (u, v)-plane. 6.11. c. The rational function ϕ (u, v) = u ⊕ v is defined if
A horizontal line in W becomes a vertical line under its denominator 1 + uv is nonzero. But |u| < 1 and |v| < 1
the action of h. To see this, parametrize the line in W as imply −1 < uv and thus 0 < 1 + uv, as required.

63
To show |u ⊕ v| < 1, we use |u| < 1, |v| < 1 to write and det dh(a,b,c) = fy (a, b, c) 6= 0. The rest of the argument
0 < 1 − u, 0 < 1 − v, and thus is as above.
But if fz (a, b, c) = fy (a, b, c) = 0, then fx (a, b, c) 6= 0
0 < (1 − u)(1 − v) = 1 − u − v + uv, or u + v < 1 + uv. because (a, b, c) is a regular point of f . In this case define
h by the formulas
Similarly we can write 0 < 1 + u, 0 < 1 + v, and thus 

u = y,
0 < (1 + u)(1 + v) = 1 + u + v + uv, −(1 + uv) < u + v. h : v = z,


Taken together these inequalities imply |u + v| < 1 + uv or w = f (x, y, z).
Then
u+v  
|u ⊕ v| = < 1.
1 + uv 0 1 0
dh(x, y, z) =  0 0 1 ,
fx (x, y, z) fy (x, y, z) fz (x, y, z)
6.11. d. Because u ⊕ v = ϕ (u, v) is a rational function, the
limit calculation is straightforward: and det dh(a,b,c) = fx (a, b, c) 6= 0. The rest of the argument
is as above.
u+v 1+v Corollary 6.4 asserts the existence of an implicit func-
lim u ⊕ v = lim = = 1.
u→1 u→1 1 + uv 1+v tion z = ϕ (x, y) with certain properties. To prove the
corollary we construct ϕ . By our proof of Theorem 6.3,
6.12. Theorem 6.3 asserts the existence of a coordinate above, we have a coordinate change h in a neighborhood
change h with certain properties. We begin the proof by of (a, b, c); h and its inverse can be given by the formulas
defining h. By hypothesis, (a, b, c) is a regular point of f .  
Suppose first that fz (a, b, c) 6= 0. We then define h by the 
u = x, 
x = u,
−1
formulas  h : v = y, h : y = v,
 
 

u = x, w = f (x, y, z), z = g(u, v, w),
h : v = y,

 for some continuously differentiable function g(u, v, w)
w = f (x, y, z).
uniquely determined by h and hence by f . Moreover,
By hypothesis, f has continuous first derivatives; hence h g is defined on some open neighborhood of h(a, b, c) =
does as well. Moreover, (a, b, k), where k = f (a, b, c).
  The inverse relation between h and h−1 allows us to
1 0 0 write, for all (x, y, z) sufficiently near (a, b, c),
dh(x, y, z) =  0 1 0 ,
fx (x, y, z) fy (x, y, z) fz (x, y, z) (x, y, z) = h−1 (h(x, y, z)) = h−1 (x, y, f (x, y, z))
= (x, y, g(x, y, f (x, y, z))).
and det dh(a,b,c) = fz (a, b, c) 6= 0, implying that dh(a,b,c)
is invertible. By the inverse function theorem, h has a In particular, the third components of these vectors are
continuously differentiable inverse on a neighborhood of equal:
of h(a, b, c) = (a, b, k) where k = f (a, b, c). Thus h is a z = g(x, y, f (x, y, z)).
valid coordinate change near (a, b, c). By construction, h Under the restriction w = f (x, y, z) = k, this equation re-
transforms the part of the level set f (x, y, z) = λ that is duces to
near (a, b, c) to the horizontal plane w = λ . z = g(x, y, k) = ϕ (x, y),
If fz (a, b, c) = 0 but fy (a, b, c) 6= 0, then define h by the
defining ϕ . Because g is defined on some open neighbor-
formulas  hood of (a, b, k), ϕ is defined on some open neighborhood

u = z, of (a, b) and
h : v = x, f (x, y, ϕ (x, y)) = k


w = f (x, y, z).
everywhere in that neighborhood. Also by construction,
Then we have ϕ (a, b) = c. Statements about the partial deriva-
  tives of ϕ follow directly from the chain rule applied to
0 0 1 f (x, y, ϕ (x, y)) = k. For example,
dh(x, y, z) =  1 0 0 ,
fx (x, y, z) fy (x, y, z) fz (x, y, z) 0 = fx + fz ϕx , so ϕx = − fx / fz .

6.13. Following the suggestion, let g(x, y) = 0 precisely 6.16. a. We have

on the set S in the (x, y)-plane, and define
1 0 0
df0 =
f (x, y, z) = [g(x, y)]2 + z2 . 0 3(y − z) −3(y − z) (x,y,z)=(0,0,0)
2 2

1 0 0
Because f is a sum of positive squares, f (x, y, z) = 0 if = .
0 0 0
and only if its square summands are both zero:
The map Df0 : R3 → R2 is therefore
g(x, y) = 0 and z = 0.
 
∆x
∆s 1 0 0   ∆x
This happens precisely at the points (x, y, 0) for which = ∆y = ;
∆t 0 0 0 0
(x, y) is in S . ∆z
6.14. The surfaces S f and S g intersect in the curve given
its image is the 1-dimensional line (∆x, 0) in R2 .
as the locus F(x, y) = x2 + y3 = 0 in the (x, y)-plane; see
the figure. 6.16. b. Let (x, y, 0) be an arbitrary point near (0, 0, 0);
then
1.0
f(x, y, z) = (x, (y − z)3 ) = (a, b)
√
0.5 if x = a and y− z = 3 √ b. In other words, the one-parameter
family of points (a, 3 b + z, z) maps to the single point
0.0
(a, b) in the target; z is the parameter.
6.16. c. We have just shown that the map f is onto its tar-
-0.5 get R2 , but the map df0 is not. The two maps have funda-
mentally different geometric character.
-1.0
-1.0 -0.5 0.0 0.5 1.0 6.17. a. A sketch of the two cylinders is shown below,
with the upper intersection curve marked.
The origin is on the locus but Fx = 2x, Fy = 3y2 , so the ori-
gin is a critical point of F. We cannot use implicit function
theorem at this point.
The locus defines the function y = ϕ (x) = −x2/3 , but
this fails to be differentiable when p x = 0. In the other
direction we have x = ψ (y) = ± −y3 ; this fails to be
defined for y > 0 and is double-valued for y < 0.
At any seed point (a, b) 6= (0, 0) for which a2 + b3 = 0,
the implicit function theorem guarantees that y = ϕ (x) and
x = ψ (y) are defined locally.
6.15. The surfaces z = 0 and (x + y)3 − z = 0 intersect
in the locus (x + y)3 = 0 in the (x, y)-plane. The locus
(x+y)3 = 0 is the same as x+y = 0, a straight line through To find the implicit functions, we must solve the equation
the origin. x2 + y2 = r2 for y and x2 + z2 = 1 for z. If we set
Two surfaces fail to be transverse at a point of intersec- p p
tion if they are tangent there—equivalently, if their nor- ϕ (x) = r2 − x2 , ψ = 1 − x2,
mals are collinear. The normal to the surface
then the curve associated with the seed (x, y, z) = (0, r, 1)
3
f (x, y, z) = (x + y) − z = 0 is the parametrized curve
is grad f = (3(x + y)2 , 3(x + y)2 , −1). The normal to the (x, ϕ (x), ψ (x)).
plane z = 0 at every point is (0, 0, 1). Therefore, along the For each of the other three seeds, the curve is
line of intersection x + y = 0, the normal to f = 0 reduces
to (0, 0, −1), and this is collinear with (0, 0, 1), the normal (0, r, −1) (0, −r, 1) (0, −r, −1)
to the plane. (x, ϕ (x), −ψ (x)) (x, −ϕ (x), ψ (x)) (x, −ϕ (x), −ψ (x))

65
√
The domains of ϕ (x) and ψ (x) are both −r ≤ x ≤ r. This 6.17. c. When r = 1 we can take ϕ (x) = ψ (x) = 1 − x2.
means that the minimum√ value of ϕ (x) is 0 but the min- The intersection consists of two plane curves; each is an
imum value of ψ is 1 − r2. The four curves are shown ellipse in its own plane. The planes are perpendicular and
in the following figure, with the seed for each. The first the ellipses form an “X” at each of the two points in space
curve is the thick black curve; the other three are faint. where they cross.
6.17. b. A sketch of the two cylinders in shown below.

The vertical cylinder now has the larger diameter, but the
intersection curves can be described with the aid of the
same four seed points.
√ √
We still have ϕ (x) = r2 − x2 and ψ (x) = 1 − x2, but
for both functions the domain is now −1 ≤ x ≤ 1. The
curve with the seed (0, r, 1) is once again parametrized as
(x, ϕ (x), ψ (x)),

√
but the minimum value of ϕ (x) is now r2 − 1 instead
of 0. The function ψ (x) is now the one that takes 0 for
its minimum value. The curves associated with the other
three seed points are likewise parametrized as in part (a).

Critical Points
7.1. a. We have h′B (u) = 1 + 2Bu so h′B (0) = 1, indepen- 7.4. The standard approach here would be to compute the
dently of B. By the inverse function theorem, hB is an partial derivatives of ∆u and ∆v with respect to ∆x and ∆y
invertible function near the origin, for any B; that is, hB is and then evaluate them for ∆x = ∆y = 0. However, if we
a coordinate change near the origin. construct instead the first-order Taylor polynomials of ∆u
and ∆v in terms of ∆x and ∆y, the derivatives we want will
7.1. b. If g(u) = f (hB (u)) = f (u + Bu2), then
appear as the coefficients of the linear terms. In fact, be-
g′ (u) = f ′ (u + Bu2)(1 + 2Bu), cause h expresses ∆u and ∆v as linear combinations of ∆x
and ∆y with variable coefficients, it is sufficient to extract
g′′ (u) = f ′′ (u + Bu2)(1 + 2Bu)2 + f ′ (u + Bu2) · 2B, just the constant part of each coefficient.
g′′ (0) = f ′′ (0) + 2B f ′(0). To begin, we note that
√ p p
By hypothesis, f ′ (0) 6= 0, so g′′ (0) = A when
A = 6p2 − 2 + O(1) = 6p2 − 2 + O(1),
A − f ′′ (0)
B= . where O(1) denotes, as usual, terms that vanish at least to
2 f ′ (0)
the same order as k(∆x, ∆y)k. Next,
7.2. We have B
B = O(1), so √ = O(1).
5 9 −2 5/2 A
a. c.
9 −2 5/2 −2 Thus
p
b.
−1 1/2
d.
1 2 ∆u = 6p2 − 2 ∆x + O(2),
1/2 −1 2 −5
implying that components of the first row of dh(0,0) are
p
7.3. We evaluate ( fm )x (x, 0) = 4x3 − 4x + m first when 6p2 − 2 and 0. For ∆v we have
x = ±1 − m/8. We have s
B 2
B2 p
3 m 3 3 2 −m 2 −C = 2−2p2 +O(1), − C = 2 − 2p2 +O(1),
4x = 4 ±1 − = 4 (±1) + 3(±1) + O(m ) A A
8 8
3m
= ±4 − + O(m2 ) as m → 0. and thus p
2
∆v = 2 − 2p2 ∆y + O(2).
Therefore
This establishes that
3 3m m !
4x − 4x + m = ±4 − ∓ 4 + + m + O(m2 ) = O(m2 ), p
2 2 6p2 − 2 p 0
dh(0,0) = .
as required. Next, when x = m/4 then 4x3 = O(m3 ) so 0 2 − 2p2
4m  
4x3 − 4x + m = − + m + O(m3) = O(m3 ), 0 5 1/2
4
7.5. a. M =  5 0 −1  .
also as required. 1/2 −1 0
66

67
7.5. b. Here Q = x2 − y2 + 2yz − z2, so 7.6. d. At (0, 0) the second-degree terms of the polyno-
  mial f give Q; that is, Q(x, y) = 3x2 − y2 . In the figure
1 0 0 below, the graphs of z = f (x, y) and z = Q(x, y) are virtu-
M = 0 −1 1  . ally indistinguishable on the small window.
0 1 −1
-0.10
-0.05
0.00
0.10 0.05
0.03
7.5. c. Here M is a diagonal matrix; the entry in the ith
0.02
position on the diagonal is i − 5: 0.01
  0.00
−4 0 · · · 0 -0.01
 0 −3 · · · 0  0.10 0.05
  0.00 -0.05 -0.10
 .. .. . .. .. .
 . . . 
0 0 ··· n−5 7.6. e. The critical point that Q has at (0, 0) Q is a sad-
dle. The figure shows that Q and f have the same type of
critical point there; the critical point of f is a saddle.
7.5. d. The coefficient of x2i in Q is 2i. The coefficient of
x j x j (i 6= j) in Q is (i + j) + ( j + i) = 2(i + j); half of this 7.7. a. We have fx = 3x2 − 3 and fy = 3y2 − 12; thus f
appears in the term in the ith row, jth column and half in has four critical points: (1, ±2) and (−1, ±2).
the term in the jth row, ith column. Thus 7.7. b. To get the second-order Taylor polynomials we
  need
2 3 ··· n + 1
 3 4 · · · n + 2
  fxx = 6x = ±6, fxy = 0, fyy = 6y = ±12.
M= . . .. ..  .
 .. .. . .  The four Taylor polynomials are
n + 1 n + 2 ··· 2n
type of
P TP point
7.5. e. The coefficient of x2i in Q is i − i = 0. The coef-
ficient of xi x j (i 6= j) in Q is (i − j) + ( j − i) = 0. Thus (1, 2) −18 + 3(x − 1)2 + 6(y − 2)2 minimum
Q ≡ 0; the corresponding symmetric matrix is the zero (1, −2) 14 + 3(x − 1)2 − 6(y + 2)2 saddle
matrix. (−1, 2) −14 − 3(x + 1)2 + 6(y − 2)2 saddle
7.6. a. We have fx = 6x − 3x2 and fy = −2y. Thus (−1, −2) 18 − 3(x + 1)2 − 6(y + 2)2 maximum
fx (0, 0) = fy (0, 0) = 0, so (0, 0) is a critical point of f .
Also fx (2, 0) = fy (2, 0) = 0 so (2, 0) is a critical point of f . In each case P is a critical point of TP , whose type 2is in-
dicated by the signs of the coefficients of (x ∓ 1) and
7.6. b. The first derivatives of f at (2, 0) are zero. The (y ∓ 2)2.
second derivatives are
7.7. c. The figures below show the graphs of f with each
TP in a small window centered at each critical point. The
fxx = 6 − 6x = −6, fxy = 0, fyy = −2.
(2,0)
graphs are virtually indistinguishable, implying that f has
Moreover f (2, 0) = 4, so P(x, y) = 4 − 3(x − 2)2 − y2 . the same kind of critical point as TP at P.
The figure below shows that the graphs of z = f (x, y)
and z = P(x, y) are virtually indistinguishable on the small
window.
-0.10
-0.05
0.00
0.10 0.05
4.00
3.99
3.98
3.97
3.96
2.10 2.05 2.00 1.95 1.90
7.6. c. Yes, P has a critical point at (2, 0); it is a maxi-

mum. The figure shows that P and f have the same type of
critical point at (2, 0); f has a (local) maximum at (2, 0).

68 SOLUTIONS: CHAPTER 7. CRITICAL POINTS
7.7. d. The list below is a consequence of the agreement 7.8. d. The list below is a consequence of the agreement
between f and TP near P. between f and TP near P.
type of type of
P critical point for f P critical point for f
(1, 2) minimum (0, 0) saddle
(1, −2) saddle (2/3, √0) minimum
(−1, 2) saddle (1, +1/√3) saddle
(−1, −2) maximum (1, −1/ 3) saddle
7.9. Critical points of f are solutions to the simultaneous

7.8. a. The four critical points of − f (and hence of f ) equations
were found in the solution to Exercise 6.1 (Solutions √
fx = 2ax + 2by + d = 0, a b x −1 d
page 59). The four points are (0, 0), (2/3, 0), (1, ±1/ 3). =
fy = 2bx + 2cy + e = 0; b c y 2 e
7.8. b. To get the second-order Taylor polynomials we
need If ac − b2 6= 0 then the matrix A is invertible and we have
the single critical point with coordinates
fxx = 6x − 2, fxy = −6y, fyy = −6x + 6. be − cd bd − ae
x= , y= .
2(ac − b2) 2(ac − b2)
We therefore find the four Taylor polynomials to be
These coordinates depend on all parameters except k. The
P TP type of critical point is determined by the eigenvalues
2 2 of A, hence upon its determinant and trace (cf. Chapter 2).
(0, 0) −x + 3y
In particular,
(2/3, 0) −4/27 + (x − 2/3)2 + y2 
√ √ √ 
(1, +1/ 3) 2(x − 1)2 − 2 3(x − 1)(y − 1/ 3) < 0 saddle(point,
√ √ √ ac − b 2
< 0 maximum,
(1, −1/ 3) 2(x − 1)2 + 2 3(x − 1)(y + 1/ 3) 
> 0 a + c
> 0 minimum.
In each case P is a critical point of TP . The second is a
minimum; the other three are saddles. The type of the If ac − b2 = 0, then the lines fx = 0 and fy = 0 are
first two is determined by the signs of the square terms; parallel. If they coincide, then fx = 0 defines a line of
the last two are saddles because TP splits into two linear critical points. The points are minima if a + c > 0 and
factors. maxima if a + c < 0. If the lines do not coincide, then Q
has no critical points.
7.8. c. The figures below show the graphs of f with each 7.10. We have Φ = sin θ = 0 if θ = nπ , where n is an
θ
TP in a small window centered at each critical point. The integer. Also, Φ = v = 0 if v = 0. Thus there are infinitely
v
graphs are virtually indistinguishable, implying that f has many critical points: (θ , v) = (nπ , 0).
the same kind of critical point as TP at P. To determine the type of each critical point, consider
the Hessian matrix

cos nπ 0 (−1)n 0
H(nπ , 0) = = .
0 1 0 1
It follows that (nπ , 0) is a minimum of Φ if n is even and
is a saddle of n is odd.
7.11. a. The surface is an upward-opening paraboloid;
vertical sections in the x-direction are gently curved, while
in the y-direction they are more sharply curved. Thus,
as measured from a particular point z = a on the positive
z-axis, the origin could be farther away than points on the
graph than lie above the y-axis while, at the same time,
closer than points above the x-axis. The work below es-
tablishes that this occurs when 1/2q2 < a < 1/2p2.

69
-1.0
1.0 0.5 0.0 -0.5 7.11. d. The signs of the diagonal elements of H(0,0) de-
termine the nature of the critical point that Da has at the
4 origin. If both signs are positive the point is a minimum,
if both are negative it is a maximum, and if they differ it
is a saddle. Thus for a minimum
2
1 − 2p2a > 0 and 1 − 2q2a > 0.

0
1.0 0.5
Because 1/2q2 < 1/2p2, it follows that Da has a minimum
0.0
at the origin when a < 1/2q2. For a maximum,
-0.5 -1.0
7.11. b. By definition, Da is the square of the length of 1 − 2p2a < 0 and 1 − 2q2a < 0,
the vector from (0, 0, a) to (x, y, f (x, y)):
or 1/2p2 < a. For a saddle, 1/2q2 < a < 1/2p2.
Da (x, y) = k(x, y, p2 y2 + q2y2 ) − (0, 0, a)k2
7.12. First consider the left-hand side of the equation.
= x2 + y2 + (p2 x2 + q2 y2 − a)2
We have
The partial derivatives are
∂ ∂ 2

∂ Da (∆u · ∇) f (a) = ∆u
2
+∆v f (u, v)
= 2x + 2(p2x2 + q2 y2 − a) · 2p2x, ∂u ∂v
∂x
(a,b)

2 ∂ ∂ 2 ∂
= 2x 1 + 2p2(p2 x2 + q2y2 − a) ,
2 2 2
= (∆u) + 2∆u ∆v + (∆v) f (u, v)
∂ Da ∂ u2 ∂ u∂ v ∂ v2 (a,b)
= 2y 1 + 2q2(p2 x2 + q2 y2 − a) .
∂y = (∆u)2 fuu (a, b) + 2∆u ∆v fuv(a, b) + (∆v)2 fvv (a, b).
Both derivatives vanish at the origin, so the origin is a The right-hand side is
critical point of Da for all values of a.

fuu (a, b) fuv (a, b) ∆u
7.11. c. We use the Hessian matrix to determine the type (∆u) Ha ∆u = ∆u ∆v
†
of the critical point of Da at the origin. To construct the fvu (a, b) fvv (a, b) ∆v
Hessian, we obtain the second derivatives. = (∆u)2 fuu (a, b) + 2∆u ∆v fuv(a, b) + (∆v)2 fvv (a, b);
∂ 2 Da the two sides agree.
(0, 0)
∂ x2
7.13. The solution of this exercise appears as the proof
= 2 1 + 2p2(p2 x2 + q2y2 − a + 2x · 4p4x of Theorem 7.12 (text p. 245).
(x,y)=(0,0)
= 2(1 − 2p2a), 7.14. a. For any two n × 1 vectors X1 and X2 , the matrix
∂ 2 Da product X2† X1 equals the ordinary dot product X2 · X1 . If
(0, 0) = 2x 4p2 q2 y = 0, MX1 = λ1 X1 , then
∂ x∂ y (x,y)=(0,0)
∂ 2 Da X2† MX1 = X2† λ X1 = λ1 X2† X1 = λ1 (X2 · X1 ).
(0, 0)
∂ y2

7.14. b. If, in addition, MX2 = λ2 X2 , then because M is
= 2 1 + 2p2(p2 x2 + q2y2 − a + 2y · 4q4y
(x,y)=(0,0) symmetric (M † = M), we also have
= 2(1 − 2q2a).
λ1 (X2 · X1 ) = X2† MX1 = X2† M † X1 = (MX2 )† X1
The Hessian is
= (λ2 X2 )† X1 = λ2 X2† X1 = λ2 (X2 · X1 ).
2(1 − 2p2a) 0
H(0,0) = ,
0 2(1 − 2q2a) Hence (λ1 − λ2 )(X2 · X1 ) = 0, but λ1 − λ2 6= 0, by hypoth-
esis. Therefore, X2 · X1 = 0; that is, X2 ⊥ X1 .
and det H(0,0) = 4(1 − 2p2 a)(1 − 2q2a). The critical point
of Da at the origin is degenerate when this determinant 7.15. a. According to the definition of pi in the proof of
equals zero, that is, when Lemma 7.3 (text p. 250), we need
1 1 ∂F ∂F
a= or a = . = ex sin y and = ex cos y.
2q2 2p2 ∂x ∂y

Then we have (using integral tables or a computer algebra This matrix is invertible; therefore, by the inverse function
system) theorem, k has a continuously differentiable inverse near
Z 1 k(0, 0) = (0, 0). That is, k is a local coordinate change
∂F
p1 (∆x, ∆y) = (t∆x,t∆y) dt near the origin.
0 ∂x
Z 1 7.16. d. The dilation–rotation map c−1 : (x, y) → (ξ , η )
= et∆x sin(t∆y) dt is given by the formulas
0
t=1
(
et∆x (∆x sin(t∆y) − ∆y cos(t∆y)) ξ = (y + x)/2,
= c−1 :
(∆x)2 + (∆y)2 t=0 η = (y − x)/2.
e∆x (∆x sin(∆y) − ∆y cos(∆y) + ∆y)
= ; Consequently, the map h = k◦ c−1 : (x, y) → (u, v) is given
(∆x)2 + (∆y)2 by
Z 1 ( p
∂F u = 3 − (y + x)· (y + x)/2,
p2 (∆x, ∆y) = (t∆x,t∆y) dt h: p
0 ∂y v = 3 + 3(y + x)· (y − x)/2.
Z 1
= et∆x cos(t∆y) dt Pullback by h of the level curves of −u2 + v2 to the (x, y)-
0
t=1 plane is shown on the left below. Shown on the right are
et∆x (∆x cos(t∆y) + ∆y sin(t∆y)) the level curves of the original folium f (x, y). The two
=
(∆x)2 + (∆y)2 t=0 figures are identical.
e∆x (∆x cos(∆y) + ∆y sin(∆y) − ∆x) 1.0 1.0
= .
(∆x)2 + (∆y)2
0.5 0.5
7.15. b. We have
0.0 0.0
p1 (x, y) x + p2 (x, y) y
ex (x2 sin y − xy cosy + xy + xy cosy + y2 sin y − xy) -0.5 -0.5
=
x2 + y2
= ex sin y, -1.0
-1.0 -0.5 0.0 0.5 1.0
-1.0
-1.0 -0.5 0.0 0.5 1.0
as required. The figure on the left was generated by the following

7.16. a. A straightforward calculation gives Mathematica command.
ϕ (ξ , η ) = f (ξ − η , ξ + η ) = 2ξ 3 + 6ξ η 2 − 3ξ 2 + 3η 2 . ContourPlot[
-(Sqrt[3 - (y + x)] (y + x)/2)ˆ2
7.16. b. We can write + (Sqrt[3 + 3 (y + x)] (y - x)/2)ˆ2,
{x, -1, 1}, {y, -1, 1},
ϕ (ξ , η ) = (2ξ − 3)ξ 2 + (6ξ + 3)η 2 ;
ContourShading -> False, Contours -> 16]
thus α (ξ ) = −3 + 2ξ and β (ξ ) = 3 + 6ξ . By contrast, the figure on the right was generated by this
7.16. c. The decomposition of ϕ suggests we define k as command:
( p
u = ξ · −α (ξ ), ContourPlot[
k: p xˆ3 + yˆ3 - 3 x y,
v = η · β (ξ ). {x, -1, 1},\\ {y, -1, 1},
Then u2 = −α (ξ ) · ξ 2 and v2 = β (ξ ) · η 2, so ContourShading -> False, Contours -> 16]
−u2 + v2 = α (ξ ) · ξ 2 + β (ξ ) · η 2 = ϕ (ξ , η ), The figure on the left above is comparable to the figure on

the left on page 239 of the text.
as required. To show that k is a local coordinate change
7.17. a. In the sketch below, the zero-level contour of
near (ξ , η ) = (0, 0), consider
p p ! f (x, y) is the one that has the “X” at the point (1, 1). The
−α (ξ ) + ξ / −α (ξ ) 0 “X” is the indicator of a saddle point. For future reference,

dk(0,0) = p p the negative levels c = −0.1, −0.2, −0.3, −0.4 are shown;
3η / β (ξ ) β (ξ ) (ξ ,η )=(0,0)
√ they appear to the northwest and southeast of the saddle
3 √0 point. Positive levels with double spacing (∆c = 0.2) ap-
= .
0 3 pear to the northeast and southwest.

71
∂ g1
= 5 + 16 ∆x + 163 ∆y + 4 (∆x) + 9∆x ∆y + 4 (∆y)
2.0 9 2 9 2
∂ (∆y) 2 3
5 (∆x) ∆y + 5 ∆x (∆y) + 2 (∆x) (∆y)
2 2 2 2
1.5 + 18 18 3
∂ g2
3 ∆x + 2∆y + 2 (∆x) + 2 ∆x ∆y
2
= 2 + 16 9 9
1.0 ∂ (∆y)
5 (∆x) ∆y + (∆x) ∆y.
+ 56 (∆x)3 + 18 2 3
0.5
We can now determine the hi j from these derivatives the
same way we determined the gi from the derivatives of ∆z.
0.0 We find
0.0 0.5 1.0 1.5 2.0
h11 (∆x, ∆y) = 2 + ∆x + 38 ∆y + 23 ∆x ∆y + 32 ∆x ∆y + 23 (∆y)2

7.17. b. A lengthy but straightforward calculation leads
to following (in which terms of the same degree appear
9
+ 10 ∆x (∆y)2 + 10
3
(∆y)3 + 15 ∆x (∆y)3 ,
on the same line):
h12 (∆x, ∆y) = 52 + 38 ∆x + 83 ∆y + 34 (∆x)2 + 3∆x ∆y + 43 (∆y)2
∆z = 2(∆x) + 5∆x ∆y + 2(∆y)
2 2
9
+ 10 (∆x)2 ∆y + 10
9
∆x (∆y)2 + 10
3
(∆x)2 (∆y)2
+ (∆x) + 8(∆x) ∆y + 8∆x(∆y) + (∆y)
3 2 2 3
+ 3(∆x)3 ∆y + 9(∆x)2(∆y)2 + 3∆x(∆y)3 h22 (∆x, ∆y) = 2 + 38 ∆x + ∆y + 23 (∆x)2 + 32 ∆x ∆y
+ 3(∆x)3 (∆y)2 + 3(∆x)2 (∆y)3

3
+ 10 (∆x)3 + 10
9
(∆x)2 ∆y + 15 (∆x)3 ∆y.
+ (∆x)3 (∆y)3 . 7.17. d. This is a lengthy but straightforward calculation.

7.17. c. First we need the partial derivatives of ∆z with 7.17. e. The general form that h takes to reduce ∆z to a
respect to ∆x and ∆y: sum of squares is shown on page 234 of the text (where
∂ (∆z) h11 , h12 and h22 play the roles of A, B, and C, respec-
= 4∆x + 5∆y + 3(∆x)2 + 16∆x ∆y + 8(∆y)3 tively):
∂ (∆x)
+ 9(∆x)2 ∆y + 18∆x (∆y)2 + 3(∆y)3  √ h12

 ∆u = h11 ∆x + √ ∆y,
+ 9(∆x)2 (∆y)2 + 6∆x(∆y)2 + 3(∆x)2(∆y)3 , 
 h11
h: s
∂ (∆z)  2
= 5∆x + 4∆y + 8(∆x)2 + 16∆x ∆y + 3(∆y)3 
∆v = ∆y h12 − h22.

∂ (∆y) h11
+ 3(∆x)3 + 18(∆x)2∆y + 9∆x (∆y)2
+ 6(∆x)3 ∆y + 9(∆x)2(∆y)2 + 3(∆x)3(∆y)2 . According to the inverse function theorem, h will be a
local coordinate change near the origin if its derivative
Because g1 (∆x, ∆y) and g2 (∆x, ∆y) are polynomials, their
dh(0,0) is invertible. Moreover, because h is the same kind
definition as integrals implies that they can be obtained
of map we dealt with in Exercise 7.4, above, we can use
quickly by multiplying a term of degree k of the integrand
the same argument to show that this derivative is invert-
(i.e., ∂ (∆z)/∂ (∆x) or ∂ (∆z)/∂ (∆y)) by 1/(k + 1). Thus
ible. That is, the components of dh(0,0) will be the con-
we have
stant terms in the coefficients of ∆x and ∆y in the formulas
g1 (∆x, ∆y) = 2∆x + 25 ∆y + (∆x)2 + 16 3 ∆x ∆y + 8
3 (∆y)2
for for h. We begin by noting that
p √
+ 4 (∆x) ∆y + 2 ∆x (∆y) + 4 (∆y)
9 2 9 2 3 3
h11 = 2 + O(1),
+ 59 (∆x)2 (∆y)2 + 65 ∆x (∆y)3 + 12 (∆x)2 (∆y)3 , h 5
√ 12 = √ + O(1),
g2 (∆x, ∆y) = 52 ∆x + 2∆y + 83 (∆x)2 + 16 ∆x ∆y + (∆y) 2 h 11 2 2
3 s r
+ 4 (∆x) + 2 (∆x) ∆y + 4 ∆x (∆y)
3 3 9 2 9 2
h212 25 3
− h22 = − 2 + O(1) = √ + O(1).
+ 5 (∆x) ∆y + 5 (∆x) (∆y) + 2 (∆x) (∆y) .
6 3 9 2 2 1 3 2 h 11 8 2 2
To determine hi j we first need the derivatives of gi . These Therefore √ √
are 2 5/2√2
dh(0,0) = ,
∂ g1 0 3/2 2
3 ∆y + 2 ∆x ∆y + 2 (∆y)
2
= 2 + 2∆x + 16 9 9
∂ (∆x) and because this matrix is invertible, h is a local coordi-
5 ∆x (∆y) + 5 (∆y) + ∆x (∆y) , nate change near (∆x, ∆y) = (0, 0).
2 3 3
+ 18 6

The following figure is a Mathematica plot that shows Therefore we have

together in a neighborhood of the origin in the (∆x, ∆y)
∂ (∆z)
the following: = 18∆x + 54(∆x)2 − 10(∆y)2 + 60(∆x)3
∂ (∆x)
• Contours of the function ∆u(∆x, ∆y); these are the − 36∆x (∆y)2 + 30(∆x)4 − 36(∆x)2(∆y)2
gray grid lines that run northwest–southeast. + 6(∆y)4 − 6(∆x)5 − 12(∆x)3(∆y)2 + 2∆x (∆y)4,
• Contours of the function ∆v(∆x, ∆y); these are the ∂ (∆z)
= −2∆y − 20∆x ∆y − 36(∆x)2∆y + 12(∆y)3
gray grid lines that run primarily west–east. ∂ (∆y)
− 24(∆x)3∆y + 24∆x (∆y)3 − 6(∆x)4 ∆y
• Contours of ∆u2 − ∆v2 in black; as in the figure in the
text, positive levels (to the northeast and southwest) + 4(∆x)2(∆y)3 − 6(∆y)5.
are twice as far apart as negative ones.
Next we compute gi using the quick method of the solu-
• A reference point at (0, 0.2) on the (∆u, ∆v)-grid. tion to Exercise 7.17.c.
Note that this figure matches the one in the text.
g1 = 9∆x + 18(∆x)2 − 10 2 3
3 (∆y) + 15(∆x) − 9∆x (∆y)
2
0.4 + 6(∆x)4 − 36 2 2 6
5 (∆x) (∆y) + 5 (∆y)
4
− (∆x)5 − 2(∆x)3(∆y)2 + 31 ∆x (∆y)4 ,

0.2
3 ∆x ∆y − 9(∆x) ∆y + 3(∆y)
2 3
g2 = −∆y − 20
5 (∆x) ∆y + 5 ∆x (∆y) − (∆x) ∆y

3 3 4
0.0 − 24 24
+ 23 (∆x)2 (∆y)3 − (∆y)5 .

-0.2
For the derivatives we have

-0.4
∂ g1
= 9 + 36∆x + 45(∆x)2 − 9(∆y)2 + 24(∆x)3
-0.4 -0.2 0.0 0.2 0.4 ∂ (∆x)
5 ∆x (∆y) − 5(∆x) − 6(∆x) (∆y) + 3 (∆y) ,
2 4 2 2 4
Here are some further details: Spacing for curves in the − 72 1
(∆u, ∆v)-grid is 0.1; spacing for the negative contours of ∂ g1

3 ∆y − 18∆x ∆y − 5 (∆x) ∆y + 5 (∆y)
2 3
= − 20 72 24
∆u2 − ∆v2 is 0.04. As a check, note that the reference ∂ (∆y)
point does indeed lie on the contour − 4(∆x)3∆y + 34 ∆x (∆y)3 ,
∆u2 − ∆v2 = −(0.2)2 = −0.04. ∂ g2
3 ∆x − 9(∆x) + 9(∆y) − 5 (∆x)
2 2 3
= −1 − 20 24
∂ (∆y)
5 ∆x (∆y) − (∆x) + 2(∆x) (∆y) − 5(∆y) .
2 4 2 2 4
7.18. a. Shown below are contours of g(x, y) near + 72
(x, y) = (1, 0). The X shape of the contour through (1, 0)
confirms that g has a saddle point there. We can now determine the hi j from these derivatives using
the quick method.
0.4
h11 = 9 + 18∆x + 15(∆x)2 − 3(∆y)2 + 6(∆x)3
5 ∆x (∆y) − (∆x) − 5 (∆x) (∆y) + 15 (∆y) ,

0.2 2 4 2 2 4
− 18 6 1
3 ∆y − 6∆x ∆y − 5 (∆x) ∆y + 5 (∆y)

0.0 2 3
h12 = − 10 18 6
-0.2 − 45 (∆x)3 ∆y + 15
4
∆x (∆y)3 ,
3 ∆x − 3(∆x) + 3(∆y) − 5 (∆x)

2 2 3
-0.4 h22 = −1 − 10 6
5 ∆x (∆y) − 5 (∆x) + 5 (∆x) (∆y) − (∆y) .

2 4 2 2 4
0.6 0.8 1.0 1.2 1.4 + 18 1 2
A lengthy but straightforward calculation shows that

A straightforward calculation shows that
∆z = 9(∆x)2 − (∆y)2 + 18(∆x)3 − 10∆x (∆y)2
h11 (∆x)2 + 2h12∆x ∆y + h22(∆y)2 = ∆z.
+ 15(∆x)4 − 18(∆x)2 (∆y)2 + 3(∆y)4
+ 6(∆x)5 − 12(∆x)3(∆y)2 + 6∆x (∆y)4 The map h : (∆x, ∆y) → (∆u, ∆v) has exactly the same gen-
eral form it did in the solution to the previous exercise.
− (∆x)6 − 3(∆x)4 (∆y)2 + (∆x)2 (∆y)4 − (∆y)6

73
Z1∂f Z 1
To show that it is likewise a local coordinate change, we g2 (x, y) = (tx,ty) dt = − cos(tx) sin(ty) dt
determine its derivative at (∆x, ∆y) = (0, 0) as the con- 0 ∂y 0
stant terms in the coefficients of ∆x and ∆y in the formu- y − y cosx cos y − x sin x sin y
las for h, exactly as in the previous solution. This time we = .
x2 − y2
have
p √ A computer algebra system provided the following inte-
h11 = 9 + O(1) = 3 + O(1), grals:
h12
√ = O(1), Z 1
h11 ∂ g1
s h 11 = (tx,ty) dt
0 ∂x
2
h12 √
− h22 = 1 + O(1) = 1 + O(1), (x2 + y2 )(cos x cos y − 1) + 2xy sinx sin y
h11 = ,
(x2 − y2 )2
so Z 1
3 0 ∂ g1
dh(0,0) = , h12 = (tx,ty) dt
0 1 0 ∂y
an invertible matrix. The following Mathematica plot 2xy(1 − cosx cosy) − (x2 + y2 ) sin x sin y
shows the (∆u, ∆v)-coordinate grid on the (∆x, ∆y)-plane = ,
(x2 − y2 )2
together with the pullback of contours of ∆u − ∆v under
2 2
Z 1
the map h. ∂ g2
h22 = (tx,ty) dt
0 ∂y
0.4 (x2 + y2 )(cos x cos y − 1) + 2xy sinx sin y
= .
(x2 − y2 )2
0.2
A straightforward calculation confirms that
0.0
cos x cos y = 1 + h11(x, y)x2 + 2h12(x, y)xy + h22(x, y)y2 .
-0.2
7.19. b. We find immediately that
1 ∂f 1 ∂f 1 ∂f
-0.4
−1
(0, 0) = (0, 0) = , (0, 0) = 0.
-0.4 -0.2 0.0 0.2 0.4 2 ∂ x2 2 ∂ y2 2 2 ∂ x∂ y
7.18. b. One way to see that the coordinate change h in However, hi j (0, 0) is indeterminate (= 0/0). But we have
this exercise is not just a rotation–dilation of the coordi-
2 y2
nate change in the previous exercise is to notice that the cos x cos y − 1 = 1 − x + O(4) 1 − + O(4) − 1
(∆u, ∆v)-grid here is not a rotation–dilation of the previ- 2 2
ous (∆u, ∆v)-grid. 2
−(x + y ) 2
Another way is to focus on the action of each coordi- = + O(4),
2
nate change near the origin, using the derivative. In the sin(x) sin y = xy + O(4).
present case, dh(0,0) is a pure dilation by 1 unit horizon-
tally and 3 vertically. The action of the other dh(0,0) is not Therefore we can write
the same; in particular, its eigenvectors are not orthogonal.
(The eigenvectors are (1, 0) and (5, −1). −(x2 + y2 )2 /2 + 2x2y2 + O(6)
h11 = h22 =
7.19. a. Because the critical point is at the origin, we can (x2 − y2 )2
use x and y in place of ∆x1 and ∆x2 for typographic sim- −x4 + 2x2y2 − y4 + O(6) −1
= = + O(2),
plicity. To begin, we calculate the gi (x, y). The integra- 2(x4 − 2x2 y2 + y4 ) 2
tions can be carried out using tables or a computer algebra −xy(x2 + y2 + O(4)) + (x2 + y2 )(xy + O(4))
system. h12 =
(x2 − y2)2
Z 1 Z 1
∂f = O(2),
g1 (x, y) = (tx,ty) dt = − sin(tx) cos(ty) dt
0 ∂x 0
x cos x cos y + y sin x sin y − x and thus conclude h11 (0, 0) = h22 (0, 0) = −1/2 while
= ;
x2 − y2 h12 (0, 0) = 0.

Double Integrals
8.1. a. The text provides a solution; the reader may con- dy = dx

struct an alternative. sum = 0
x = 1 + dx / 2
8.1. b. The computations can be carried out with just a FOR i = 1 TO k
single change to the program shown on page 271 of the y = 1 + dy / 2
text;: FOR j = 1 TO n
sum = sum + dx * dy /
sum = sum - a * dx * dy /
(4 * SQR(x ˆ 2 + y ˆ 2))
((1 + x ˆ 2 + y ˆ2)
y = y + dy
(x ˆ 2 + y ˆ 2 + a ˆ 2) ˆ (3 / 2)) NEXT j
The results are tabulated below. x = x + dx
NEXT i
a = 0.2 a = 0.1 a = 0.05 PRINT sum
k sum sum sum
64 −0.843 270 −0.542 976 −0.292 194 8.2. a. The program as written above estimates the value
128 −1.141 706 −1.021 734 −0.645 161 of the integral to be 0.204 806.
256 −1.178 579 −1.315 882 −1.125 753
512 −1.178 855 −1.351 878 −1.418 793 8.2. b. For a 20 × 40 grid, set k = 20; for a 200 × 400
1024 −1.178 855 −1.352 147 −1.454 575 grid, set k = 200. The program gives the following re-
2048 −1.352 150 −1.454 842 sults.
grid estimate error
These results provide an estimate of the size of the field at 20 × 40 0.205 209 1.4 × 10−5
each of the three distances: 200 × 400 0.205 213 < 10−6
field at a = 0.2: − 1.178 885; field at a = 0.1: − 1.3521; 8.3. Make the indicated changes to the following lines in
field at a = 0.05: − 1.454. the program in the previous solution.
n = k
8.1. c. The original plate has density ρ = 1/4G while new dx = .8 / k
plate has density x = .2 + dx / 2
y = .2 + dy / 2
1 1 1
ρ= · < sum = sum + dx * dy * 4 * (x ˆ 2 + y ˆ 2)
1 + x2 + y2 4G 4G
The program yields the indicated values when k = 4
The two plates have the same size, but the new one is less and k = 20. It appears that the grid size needs to be
dense so it is less massive. The less massive plate has the between 150 × 150 and 200 × 200 to make the value
weaker field, as expected. 2.11626: when k = 100 the value is 2.116 240; when
8.2. One way to modify the program to estimate the in- k = 150 the value is 2.116 255; when k = 200, the
tegral is as follows. value is 2.116 260.
k = 2 8.4. The two sets are not always equal, but the first is
n = 2 * k always a subset of the second; that is, ◦(Sc ) ⊆ (◦S)c for
dx = 1 / k every set S. To see the two may not be equal, let S be the
74

75
(open set of) negative real numbers: S = (−∞, 0) = ◦S. In the limit as m → ∞, we have 2m−k /22m = 1/2m+k → 0
Then (◦S)c = [0, ∞) includes 0 but ◦(Sc ) = (0, ∞) does not. and 1/22m → 0, so
To establish the inclusion, we use basic facts about inte-
riors and complements. Specifically, ◦S ⊆ S for any set S, 1
J(Q) = lim J m (Q) = 2k = lim J m (Q) = J(Q).
so m→∞ 2 m→∞
◦ c c ◦ c
( S) ⊇ S ⊃ (S ). This implies Q is Jordan measurable and J(Q) = 1/22k .
In the solution to Exercise 8.6, below, we use this result in 8.8. Let R denote the rectangle in the (x, y)-plane with
the form ◦S ∩ ◦(Sc ) = 0.
/ a ≤ x ≤ b, c ≤ y ≤ d. Set ak = ⌊2k a⌋ and bk = ⌈2k b⌉.
8.5. We are given that b is a boundary point of S. Sup- Then ak and bk are the integers for which
pose there is an open disk centered at b that contains no ak ak + 1 bk − 1 bk
point of S. That disk is then in Sc and b is in the interior k
≤a< k
and k
<b≤ k.
2 2 2 2
of Sc . This violates the definition of a boundary point of S.
Likewise, if there is an open disk centered at b that con- Similarly, set ck = ⌊2k c⌋ and dk = ⌈2k d⌉. It follows that
tains no point of Sc , then b is an interior point of S, again the squares of the grid J k that lie entirely inside R make
violating the definition of a boundary point. We conclude up a rectangle that is bk − ak squares wide and dk − ck
that, when b is a boundary point of a set S, then every open squares high. Their total area is
disk centered at b contains at least one point in S and one
point not in S. b k − a k d k − ck
J k (R) = · .
2k 2k
8.6. Suppose the continuous curve is given parametri-
cally as the map x : [0, 1] → S, with x(0) = p and x(1) = q. There are 2(bk − ak ) additional squares of J j that meet R
By hypothesis, no point x(t) is in ∂ S. Therefore if we de- along the top or the bottom of R; there are 2(dk − bk ) that
fine meet R along one of its two sides; and there are four that
meet at the four corners of R. Their total area is
A = {t : x(t) is in ◦S}, B = {t : x(t) is in ◦(Sc )}, 2(bk − ak ) + 2(dk − ck ) + 4
J k (R) = J k (R) + .
then A ∪ B = [0, 1]. Moreover, A ∩ B = 0/ from the solution 22k
to Exercise 8.4, above. The values in A are all bounded As k → ∞,
by 1; let T ≤ 1 be the least upper bound. We claim x(T )
must be in ∂ S. It cannot be an interior point of Sc because bk − ak d k − ck
k
→ b − a, → d − c.
the points x(t) with t < T are in ◦S and hence S itself but 2 2k
come arbitrarily close to x(T ), by continuity of x. Like- Furthermore, 4/22k → 0 and
wise, because A and B together exhaust [0, 1], any x(t)

with t > T must be in ◦(Sc ) and hence in Sc . But the con- 2(bk − ak ) + 2(dk − ck ) 1 b k − a k d k − ck
tinuity of x guarantees that such points come arbitrarily = k−1 +
22k 2 2k 2k
close to x(T ), so x(T ) cannot be an interior point S. The
→ 0 · ((b − a) + (d − c)) = 0.
only remaining possibility is that x(T ) is a boundary point
of S. Therefore
8.7. Along a side of the square Q there are 2m−k squares
J k (R) → J k (R) → (b − a)(d − c),
in the “finer” grid J m . Thus there are 2m−k × 2m−k such
small squares entirely contained in Q. Because the area of so R is Jordan measurable and J(R) = (b − a)(d − c).
a small square is 1/22m , their total area is
8.9. The first set of inequalities implies −a ≤ −αδ .
2m−2k 1 1 combining this with b ≤ β δ in the second set gives
J m (Q) = 2 × 2m = 2k .
2 2
b − a ≤ β δ − αδ .
Additional small squares intersect Q. There are 2m−k
along each of the four sides, and one more at each cor- But 2δ < b − a, so
ner. Their total area is
2δ < β δ − αδ or αδ + δ < β δ − δ ,
4 × 2m−k 4 × 1
J m (Q) = J m (Q) + + 2m . as required.
22m 2

76 SOLUTIONS: CHAPTER 8. DOUBLE INTEGRALS
8.10. We prove the contrapositives: J(S) 6= 0 ⇔ ◦S 6= 0. / 8.13. For the four measurable sets R, S, T , and U, we
Suppose first that ◦S 6= 0.
/ Then a disk of some radius r > 0 take them one at a time (four possibilities), then intersect-
lies√entirely within S. The disk contains a square of side ing two at a time (six possibilities), then three at a time
r/ 2, so J(S) ≥ r2 /2 > 0. (four possibilities), then four at a time; the formula is
Now suppose that J(S) = ε > 0. Since J k (S) increase
monotonically to J(S), we can choose a K for which J(R ∪ S ∪ T ∪U) = J(R) + J(S) + J(T) + J(U)
J K (S) > ε /2 6= 0. Thus there is at least one square of − J(R ∩ S) − J(R ∩ T) − J(R ∩U)
side 1/K entirely contained in S, and hence one open disk − J(S ∩ T ) − J(S ∩U) − J(T ∩U)
with radius 1/2K > 0. Thus the interior of S is nonempty.
+ J(R ∩ S ∩ T) + J(R ∩ S ∩U) + J(R ∩ T ∩U)
8.11. First consider S = S ∪ ∂ S (cf. Def. 8.5, text p. 277). + J(S ∩ T ∩U) − J(R ∩ S ∩ T ∩U).
Because S is Jordan measurable, so is ∂ S and J(∂ S) = 0
(Theorem 8.2, p. 282). By Theorem 8.12 (p. 285), S is To prove this, set A = R ∪ S ∪ T ; then
Jordan measurable and
J(R ∪ S ∪ T ∪U) = J(A ∪U) = J(A) + J(U) − J(A ∩U).
J(S) ≤ J(S) ≤ J(S) + J(∂ S) = J(S);
Eventually we replace J(A) by the known result for the
thus J(S) = J(S). union of three sets. For J(A ∩U) we need
Next consider ◦S; we have ◦S = S ∩ ∂ S by the remarks
following Def. 8.5. Therefore, ◦S is Jordan measurable by A ∩U = (R ∩U) ∪ (S ∩U) ∪ (T ∩U),
Theorem 8.12. If we write S = ◦S ∪ (S ∩ ∂ S), then
This is another three-set union; its various intersections
J(S) ≤ J(◦S) + J(S ∩ ∂ S) = J(◦S) + J(∂ S) = J(◦S). are
Now J(◦S) ≤ J(S) is immediate, so J(◦S) = J(S). (R ∪U) ∩ (S ∩U) = R ∩ S ∩U,
Finally, because T ⊆ S, we have J(T ) ≤ J(S) by Theo-
rem 8.3, p. 283. Inner Jordan content is similarly mono- (R ∪U) ∩ (T ∪U) = R ∩ T ∩U,
tonic; thus ◦S ⊆ T implies J(◦S) ≤ J(T ). Therefore (S ∪U) ∩ (T ∪U) = S ∩ T ∩U,
(R ∩U) ∩ (S ∩U) ∩ (T ∩U) = R ∩ S ∩ T ∩U.
J(T ) − J(T ) ≤ J(S) − J(◦S) = J(S) − J(S) = 0,
so T is Jordan measurable and we can write Hence
J(S) = J(◦S) ≤ J(T ) ≤ J(S) = J(S), J(A ∩U) = J(R ∪U) + J(S ∪U) + J(T ∪U)
− J(R ∩ S ∩U) − J(R ∩ T ∩U) − J(S ∩ T ∩U)
implying J(T ) = J(S).
+ J(R ∩ S ∩ T ∩U).
8.12. Theorem 8.12 implies that all finite unions and in-
tersections of measurable sets are measurable, so all the Substitute this into the expression for J(R ∪ S ∪ T ∪ U)
sets involved are measurable. Let A = R ∪ S; then Theo- above to get the identity.
rem 8.15 implies For the general case, part of the challenge is to have
a suitable notation. If S1 , S2 , . . . , S p are measurable sets,
J(R ∪ S ∪ T ) = J(A ∪ T ) = J(A) + J(T ) − J(A ∩ T) then all unions and intersections are measurable and
= J(R ∪ S) + J(T) − J(A ∩ T)
= J(R) + J(S) − J(R ∩ S) + J(T) − J(A ∩ T ). J(S1 ∪ · · · ∪ S p) = ∑ J(Si ) − ∑ J(Si1 ∩ Si2 )
i i1 <i2
Now A∩T = (R∪S)∩T = (R∩T )∪(S ∩T ); furthermore, + · · · + (−1)k−1 ∑ J(Si1 ∩ · · · ∩ Sik )
(R ∩ T ) ∩ (S ∩ T ) = R ∩ S ∩ T , so i1 <···<ik
J(A ∩ T ) = J(R ∩ T ) + J(S ∩ T ) − J((R ∩ T ) ∩ (S ∩ T ) + · · · + (−1) p−1J(S1 ∩ · · · ∩ S p ).

= J(R ∩ T ) + J(S ∩ T ) − J(R ∩ S ∩ T). We prove this formula by mathematical induction. We
Substituting this value of J(A ∩ T ) into the first equation know it is true for p = 2, 3, 4; we assume it true for all
and rearranging terms there gives p < P and then prove it for p = P. We have
J(R ∪ S ∪ T ) = J(R) + J(S) + J(T) J(S1 ∪ · · · ∪ SP−1 ∪ SP ) = J(S1 ∪ · · · ∪ SP−1) + J(SP)

− J(R ∩ S) − J(S ∩ T) − J(T ∩ R) + J(R ∩ S ∩ T). − J((S1 ∪ · · · ∪ SP−1) ∩ SP ).

77
We can write the last set as the union of P − 1 measurable boundary point of a set A, then every open disk D centered
sets: at b meets both A and Ac :
(S1 ∪ · · · ∪ SP−1) ∩ SP = A1 ∪ · · · ∪ AP−1 , D ∩ A 6= 0/ and D ∩ Ac 6= 0.

/
Let b be a boundary point of T \ S; then
where A j = S j ∩ SP , j = 1, . . . , P − 1.
Of the three terms on the right-hand side of the equation D ∩ (T \ S) = D ∩ T ∩ Sc 6= 0/ and
for J(S1 ∪ · · · ∪ SP ), the induction hypothesis provides the D ∩ (T \ S)c = D ∩ (T c ∪ S) = (D ∩ T c ) ∪ (D ∩ S) 6= 0/
expansion of the first, J(S1 ∪ · · · ∪ SP−1 ), as well as the
third, written as for every open disk centered at b. The first inequality im-
plies
−J((S1 ∪ · · · ∪ SP−1 ) ∩ SP ) = −J(A1 ∪ · · · ∪ AP−1). D ∩ T 6= 0/ and D ∩ Sc 6= 0. /
Expansion of this term involves all possible intersections If b is in ∂ T , we are done. Otherwise, if b is not in ∂ T ,
c
of the sets A j , which we relate back to intersections of the then it is an interior point of either T or T . That is, for ev-
b
ery open disk D centered at b whose radius is sufficiently
sets Si , in the following way. For any j1 < · · · < jk < P,
small,
A j1 ∩ A j2 ∩ · · · ∩ A jk = S j1 ∩ S j2 ∩ · · · ∩ S jk ∩ SP . b ∩ T = 0/ or D b ∩ T c = 0.
either D /
Therefore, translating from the A j to the Si , we find But Db ∩ T 6= 0, b ∩ T = 0/ is forced. The condition
/ so D c
−J(A1 ∪ · · · ∪ AP−1) b ∩ T c ) ∪ (D
(D b ∩ S) 6= 0/
= − ∑ J(Si ∩ SP ) + ∑ J(Si1 ∩ Si2 ∩ SP ) then implies Db ∩ S 6= 0. / We already know Db ∩ Sc 6= 0.
/
i<P i1 <i2 <P b is any sufficiently small open disk centered
Because D
− · · · − (−1)k−1 ∑ J(Si1 ∩ · · · ∩ Sik ∩ SP ) at b, we see b is in ∂ S.
i1 <···<ik <P
8.15. In R2 , a square in J k has 8 neighbors, but in R3 it
− · · · − (−1)P−2 J(S1 ∩ · · · ∩ SP−1 ∩ SP ). has 26. Therefore, to prove Lemma 8.5 in R3 , begin by
Adjusting the signs, we get choosing K so large that J K (S) < ε /27. If Q is a cube
counted in J K (S) that contains p, the point q is either in Q
−J(A1 ∪ · · · ∪ AP−1) or in one of the 26 cubes neighboring Q. Thus T is cov-
ered by the cubes Q and their immediate neighbors, whose
= − ∑ J(Si ∩ SP ) + ∑ J(Si1 ∩ Si2 ∩ SP )
i<P i1 <i2 <P
total Jordan content is less than 27 × ε /27 = ε . Hence
J(T ) ≤ J K (T ) < ε .
+ · · · + (−1)k ∑ J(Si1 ∩ · · · ∩ Sik ∩ SP) 8.16. The diameter of the square Q is the length w√2
i1 <···<ik <P
P−1 of its diagonal. The image of Q under any linear map is
+ · · · + (−1) J(S1 ∩ · · · ∩ SP−1 ∩ sP )
a parallelogram whose diameter is the larger of its two
The expansion of J(S1 ∪ · · · ∪ SP−1 ) contains all terms that diagonal lengths. Under the given map L the vertices of
exclude SP in the formula we are trying to prove, while L(Q) are the origin and the three points
−J(A1 ∪ · · · ∪ AP−1) contains all that the terms that do in-
a b a+b
clude SP (except for J(SP ), which has its own place in the P1 = w c , P2 = w d , and P3 = w c + d .
formula). This observation completes the induction.
The diagonal from the origin to P3 has length
8.14. a. For example, let T be the closed disk of radius 2 q
centered at the origin; ∂ T is the circle of radius 2. Let S δ1 = w (a + b)2 + (c + d)2
be the open disk of radius 1; ∂ S is the circle of radius 1. q
The set T \ S is a closed annulus and = w a2 + b2 + c2 + d 2 + 2(ab + cd).
∂ (T \ S) = ∂ T ∪ ∂ S. The other diagonal is given by the vector P1 − P2 ; its
length is
Hence ∂ S ⊂ ∂ (T \ S) but ∂ T ∩ ∂ S = 0;/ thus ∂ (T \ S) cer- q
tainly contains points of ∂ S that are not in ∂ T . δ2 = w (a − b)2 + (c − d)2
q
8.14. b. We must show that a boundary point of T \ S is
= w a2 + b2 + c2 + d 2 − 2(ab + cd).
a boundary point of either T or of S. In general, if b is a

78 SOLUTIONS: CHAPTER 8. DOUBLE INTEGRALS
The diameter δ (L(Q)) is the larger of δ1 and δ2 , and thus If S is a subset of R1 and f (x) is a function defined
the ratio δ (L(Q))/δ (Q) is the larger of the two numbers and bounded on S, then upper and lower Darboux sums
p for f over S and a grid G are defined formally as in R2 ,
a2 + b2 + c2 + d 2 ± 2(ab + cd) as are upper and lower Darboux integrals and Darboux
√ .
2 integrability.
Now let S = [a, b] be a closed interval, and suppose f
L(Q) is bounded and continuous everywhere on S except at the
points in a finite set Z. To demonstrate that f is integrable
w b on S, it is sufficient to show that, for any given ε > 0, there
d δ2 is a grid J k for which
δ1 w a
c
DJk ( f , S) − DJk ( f , S) < ε .
Let B be a global bound for f : | f (x)| ≤ B for all real x.

Because Z has length 0, it has tubular neighborhoods of
8.17. The following BASIC programs reproduce the con- arbitrarily small total positive length. Let T be a tubular
tents of the table given in the exercise. The first counts
neighborhood of Z for which ℓ(T ) < ε /4B. Because T
“Inner Squares” and computes J k ; the second, “Outer
is open, S \ T is closed and bounded, so f is uniformly
Squares” and J k . Just substitute for k the values 0, 1, . . . ,
9 in turn. continuous there. Thus, for the ε already given we can
choose a δ > 0 so that if p and q are in S \ T , then
k = 0
n = 2 ˆ k ε
|p − q| < δ =⇒ | f (p) − f (q)| < .
y = 0 2(b − a)
count = 0
FOR j = 1 TO n Now choose k so that 1/2k = kJ k k < δ (thus we can
x = 0 take k > log2 (1/δ )), and divide the segments of J k into
y = y + 1 / n
the two classes
FOR i = 1 TO n
x = x + i / n • Q1 , . . . , QN lie entirely within S \ T .
IF x ˆ 2 + y ˆ 2 <= 1
THEN count = count + 1 • R1 , . . . , RL meet T .
NEXT i
NEXT j Define mi , Mi , m bi as in the text; then by definition
b i , and M
PRINT k, n, 4 * count, 4 * count / n ˆ 2 of Darboux sums
N L
k = 0 DJk ( f , S)−DJk ( f , S) = ∑ (Mi −mi )ℓ(Qi )+ ∑ (M
bi − m
b i )ℓ(Ri ).
n = 2 ˆ k i=1 i=1
y = 0
count = 0 Because f is continuous on each of the closed bounded
FOR j = 0 TO n sets Qi , there are points pi and qi in Qi for which m1 =
y = j / n f (pi ), Mi = f (qi ). Furthermore, |pi − qi | < δ ; hence
FOR i = 0 TO n
x = i / n N N
IF x ˆ 2 + y ˆ 2 <= 1 ∑ (Mi − mi )ℓ(Qi ) = ∑ ( f (qi ) − f (pi)ℓ(Qi )
THEN count = count + 1 i=1 i=1
NEXT i ε N
NEXT j < ∑
2(b − a) i=1
ℓ(Qi )
PRINT k, n, 4 * count, 4 * count / n ˆ 2
ε ε
8.18. We define an integration grid G for R1 to be a col- ≤ (b − a) = .
2(b − a) 2
lection of nonoverlapping closed intervals that together
cover R1 . The Jordan content of such an interval I = [p, q] For the second summand, we have
is its length ℓ(I) = q − p. The grid J k consists of con-
L L
gruent intervals I with ℓ(I) = 1/2k starting at the origin.
k
Clearly kJ k k = 1/2 . ∑ (M b i )ℓ(Ri ) ≤ 2B ∑ ℓ(Ri ),
bi − m
i=1 i=1

79
because M bi and −m b i are both bounded by the global Therefore the mass of the piece S of the annulus from part
bound B. Because the Ri are precisely the segments of (a) has this fraction of the mass of the annulus:
J k that meet T , they give its outer length: ∆r ∆θ
M(S) = .
L 2 π ((1 + a) 2 − (∆r)2 /4)
∑ ℓ(Ri) = J k (T ). The area of the annulus itself is

i=1
A = π (a + ∆r/2)2 − π (a − ∆r/2)2 = 2aπ ∆r.
But J k (T ) ց J(T ) < ε /4B as k → ∞. Thus for some suf-
The area of the sector S is therefore
ficiently large K, J K (T ) < ε /4B as well. Therefore, if
∆θ
k > K as well as k > log2 (1/δ ), then A(S) = · 2aπ ∆r = a∆r ∆θ .
2π
L L
ε ε
∑ (Mbi − mb i)ℓ(Ri ) < 2B ∑ ℓ(Ri) < 2B · 4B = 2 8.21. c. The density ratio is
i=1 i=1 M(S) 1
= ;
A(S) 2π a((1 + a)2 − (∆r)2 /4)
and DJk ( f , S) − DJk ( f , S) < ε , as we sought to prove.
in the limit as ∆r → 0, this becomes
8.19. For an arbitrary grid G ,
1 1
ρ (a, b) = or ρ (r, θ ) = ,
DG ( f , S) = 0 and DG ( f , S) = 1. 2π a(1 + a)2 2π r(1 + r)2
if we use more familiar variables for radius a and angle b.
Because these values are independent of G , the upper and Note that ρ is independent of θ , as we would expect.
lower integrals are just
8.21. d. The element of area is dA = r dr d θ in polar co-
D( f , S) = 0 and D( f , S) = 1, ordinates, so the integral that gives back the mass is
Z Z 2π Z α
r dr
so f is not Darboux integrable on S. ρ (r, θ ) dA = d θ
0 2π r(1 + r)
2
0
r≤α
8.20. We need to consider how m and M > m lie on the α
−1
real line. = 2π · = −1 + 1
2π (1 + r) 0 1+α
• If 0 < m, then m∗ = m, M ∗ = M, and M ∗ − m∗ = α
M − m. = ,
1+α
• If M < 0, then m∗ = −M, M ∗ = −m, and M ∗ − m∗ = as required.
−m − (−M) = M − m. 8.21. e. Under the new assumptions, the mass of the an-
nulus is ∆r, and the mass of the sector S is
• If m ≤ 0 ≤ M, then m∗ = 0 and M ∗ is the larger of
−m and M. Hence M ∗ − m∗ is equal to the larger of ∆r ∆θ
M(S) = .
the positive numbers M and −m, but this is not larger 2π
than their sum M − m. The area of the sector is unchanged, so the density ratio is
M(S) 1
8.21. a. The whole disk of radius a ± ∆r/2 has mass = .
A(S) 2π a
a ± ∆r/2 This is constant, so ρ (r, θ ) = 1/2π r, and the integral that
M± = . gives back the mass is
1 + a ± ∆r/2
Z Z 2π Z α
r dr α
The annulus has mass ρ (r, θ ) dA = dθ = 2π · = α,
0 0 2π r 2π
r≤α
a + ∆r/2 a − ∆r/2
M+ − M− = − as required.
1 + a + ∆r/2 1 + a − ∆r/2
∆r 8.22. The only change we need make to the program
= . used in Exercise 8.3, above, is to the line computing the
(1 + a) − (∆r)2 /4
2
sum:
sum = sum + dx dy x
8.21. b. The disk has a sector cut off by an angle ∆θ * *
whose area is the fraction ∆θ /2π of the disk’s area. With both grids the result is 0.384.

Evaluating Double Integrals

√ √ √ p
9.1. Considering y = x2 + a2 tan θ as a function of θ , Next, write 2R2 + a2 as 2R 1 + a2/2R2 ; then
we have
p R2 R
dy = x2 + a2 sec2 θ d θ , √ = √ p
2
a 2R + a 2 a 2 1 + a2/2R2
x2 + y2 + a2 = (x2 + a2)(1 + tan2 θ ) R
= √ 1 + O((a/R)2)
= (x2 + a2) sec2 θ , a 2
dy 1 R
= 2 cos θ d θ , = √ + O(a/R) as a/r → 0.
(x2 + y2 + a2)3/2 x + a2 a 2
Z Z
dy 1 sin θ
= 2 cos θ d θ = 2 . The last step is a consequence of
(x2 + y2 + a2)3/2 x + a2 x + a2 2
kR a a
The definition of y in terms of θ implies that θ lies in a ·O = O .
a R2 R
right triangle opposite the
√ side with length y and adjacent
2 2
p x + a . The hypotenuse of the
to the side with length
9.3. The integrals in (a) and (b) have the same domain;
triangle is therefore x + y + a2, and
2 2
we expect them to have the same value (Theorem 9.1),
y
sin θ = p . and they do. The same is true of the integrals in (c) and
x + y2 + a 2
2
(d).
The integral we seek is therefore Z 5 y=5
2 3 2 xy4
y=R 9.3. a. (x + xy ) dy = x y + = 4x2 + 156x.
Z R
dy y 1 4 y=1
3
= p Z 3
(x2 + y2 + a2)3/2 (x + a ) x2 + y2 + a2
2 2 4x3
0
y=0 (4x2 + 156x) dx = + 78x2 = 738.
0 3 0
R
= √ . Z 3 x=3
(x2 + a2) x2 + R2 + a2 x3 x2 y3 9y3
9.3. b. (x2 + xy3 ) dx = + = 9 + .
0 3 2 x=0 2
9.2. First we have the general facts Z 5 5
9y3 9y4
√ 9+ dy = 9y + = 738.
1 + A = 1 + O(A) and
1
= 1 + O(B); 1 2 8 1
1+B Z 5 x=5
x3 x2 y3 124
putting these together we find 9.3. c. (x2 + xy3 ) dx = + = + 12y3.
1 3 2 x=1 3
1 1 Z 3 3
√ = = 1 + O(A) as A → 0. 124 124y
1+A 1 + O(A) 3
+ 12y dy = 4
+ 3y = 367.
0 3 3 0
With A = (a/2R)2 we thus have y=3
Z 3 4
xy 81x
1 9.3. d. (x2 + xy3 ) dy = x2 y + = 3x2 + .
p
2
= 1 + O (a/R)2 as a/R → 0, 0 4 y=0 4
1 + (a/2R) Z 5 5
81x 81x2
because (a/2R)2 → 0 if and only if a/R → 0. 3x2 + dx = x3 + = 367.
1 4 8 1
80

81
9.4. a. The inner integral is 9.6. a. The unit circle:

2y+1 p p
Z 2y+1 −1 ≤ x ≤ 1, − 1 − x2 ≤ y ≤ 1 − x2 .
x2 y 3y3
xy dx = = + 7y2 − 12y.
y−5 2 y−5 2
9.6. b. The circle of radius 3 centered at (5, −1):
The second, outer, integral is then
3 ≤ x ≤ 8,
Z 1 3 1 q q
3y 3y4 7y3 14
+ 7y2 − 12y dy = + − 6y2 = . −1 − 9 − (x − 5)2 ≤ y ≤ −1 + 9 − (x − 5)2.
−1 2 8 3 −1 3
9.6. c. The circle of radius R centered at (p, q):
9.4. b. The inner integral is
Z √x √x p − R ≤ x ≤ p + R,
3 2 y4 x2 y2 q q
(y + x y) dy = + ;
x/2 4 2 x/2 q − R2 − (x − p)2 ≤ y ≤ q + R2 − (x − p)2.
x2 x3 9x4
= + − . 9.6. d. Here y must satisfy the conditions x2 ≤ y ≤ 4; con-
4 2 64
sequently x2 ≤ 4 and thus
The outer integral is
4 −2 ≤ x ≤ 2, x2 ≤ y ≤ 4.
Z 4 2

x x3 9x4 x3 x4 9x5 = 128 .
+ − dx = + − 9.6.a. & b.
0 4 2 64 12 8 320 0 15
y
9.4. c. The inner integral is simple:
−1 1 2 8
1
Z 1 Z √x Z 1 3/2 3 x
√ 2x x 1 (5, −1)
dy dx = ( x − x2 ) dx = − = .
0 x2 0 3 3 3
0 3
9.4. d. The value of this integral is also 1/3 because the

substitutions y ←→ x convert it into the preceding integral. 9.6.c. 9.6.d.
9.5. 9.3.a. & b. 9.3.c. & d. y
y
y y 4
5
(p,q)
R
3
1 p−R p+R x −2 2 x
3 x 1 5 x 9.6. e. This region is also the domain of the integrals in

Exercises 9.4.b & c:
9.4.a. 9.4.b.
y y √
0 ≤ x ≤ 1, x2 ≤ y ≤ x.
2 (4, 2)
1 9.6. f. The triangle with vertices (0, 0), (5, 5), and (0, 5).
−1 1 x 4 x 0 ≤ x ≤ 5, x ≤ y ≤ 5.
9.4.c. & d; 9.6.e.
y 9.6. g. The graph of the absolute-value function is a con-
−1 (1, 1) venient way to bound the diamond-shaped region:
−1 ≤ x ≤ 1, |x| − 1 ≤ y ≤ 1 − |x|.
−5
√
1 x 9.6. h. 0 ≤ x ≤ p, x ≤ y ≤ 4 − x2 .

82 SOLUTIONS: CHAPTER 9. EVALUATING DOUBLE INTEGRALS
9.6.f. Another way is to split the domain into the two regions
y described in the second part of that solution, so
9.6.g.
Z 2 Z 10−x Z 8Z 8
(5, 5) y f (x, y) dy dx = √ f (x, y) dx dy
0 x3 0 3y
Z 10 Z 10−y
+ f (x, y) dx dy.
−1 1 x 8 0
9.9. The domain of integration is a triangle whose alter-

nate description is
x
9.7. a. The circle of radius R centered at (p, q): 0 ≤ x ≤ 1, 0 ≤ y ≤ x;
thus
q − R ≤ y ≤ q + R, Z 1Z 1 Z 1Z x Z 1 y=x
q q x2 x2 x2

e dx dy = e dy dx = y e dx
p − R2 − (y − q)2 ≤ x ≤ R2 − (y − q)2. 0 y 0 0 0 y=0
1
Z 1
x2 e e−1 x2
9.7. b. This is the same as the region in Exercise 9.6.e: = x e dx = = .
0 2 2
√ 0
0 ≤ y ≤ 1, y2 ≤ x ≤ y.
Without reversing the order of integration, we face the an-
tiderivative Z
9.7. c. The triangle with vertices (0, 0), (5, 5), and (0, 5): 2
ex dx
0 ≤ y ≤ 5, 0 ≤ x ≤ y.
which cannot be expressed in terms of the elementary
functions of calculus.
9.7. d. The triangle with vertices (0, 0), (5, 5), and (5, 0):
9.10. a. The domain is a semicircle whose “reverse” de-
0 ≤ y ≤ 5, y ≤ x ≤ 5. scription is
p
−2 ≤ y ≤ 2, 0 ≤ x ≤ 4 − y2.
9.7. e. This region is a triangle : (2, 2)
y
The integral is the area of this semicircle, so we know
0 ≤ y ≤ 2, y ≤ x ≤ 2. immediately that the value of the integral is 12 π · 22 = 2π .
Carrying out the requested steps , we find
9.7. f. If we define the function β as x √
Z 2 Z 4−x2 Z 2 Z √4−y2 Z 2p
(√ √ dy dx = dx dy = 4 − y2 dy
3 y, 0 ≤ y ≤ 8, 0 − 4−x2 −2 0 −2
β (y) = p
10 − y, 8 ≤ y, y 4 − y2 y 2

= + 2 arcsin = 2π .
then we can describe the region as 2 2
−2
0 ≤ y ≤ 10, 0 ≤ x ≤ β (y). 9.10. b. This is the integral in Exercise 9.4.c; when the
order of integration is reversed, the integral of Exercise
Alternatively, we can use two pairs of inequalities (and 9.4.d is produced. The solutions there showed these inte-
interpret the region as the union of the sets described by grals have the value 1/3.
each pair):
9.10. c. The domain is similar in shape to the domain of
0 ≤ y ≤ 8, 8 ≤ y ≤ 10, Exercise 9.4.b: straight on the “bottom” and curved on the
√
0 ≤ x ≤ 3 y; 0 ≤ x ≤ 10 − y. “top.” The reverse description is
0 ≤ y ≤ 1, y3 ≤ x ≤ y;
9.8. The domain of integration is the region described in
Exercise 9.7.f. One way to reverse the order of integration thus
Z 1Z √ Z 1Z y Z 1 y
is to use the function β (y) defined in the solution to that 3x
2 2

2
exercise, so y dy dx = y dx dy = xy dy
0 x 0 y 3 0 3 y
Z 2 Z 10−x Z 10 Z β (y) Z 1
1 1 1
f (x, y) dy dx = f (x, y) dx dy. = y3 − y5 dy = − = .
0 x3 0 0 0 4 6 12

83
9.10. d. This is the integral of Exercise 9.4.b; the reverse We can simplify this expression by translating from (p, q)
description of the domain is to (0, 0) by the substitutions u = x − p, v = y − q; then
2
0 ≤ y ≤ 2, y ≤ x ≤ 2y. √
Z r Z r2 −u2 Z r p
We therefore have I= √ (u+ p) dv du = 2(u+ p) r2 − u2 du.
Z 4 Z √x Z 2 Z 2y −r − r2 −u2 −r
3 2 3 2
(y + x y) dy, dx = (y + x y) dx dy We write the last integral as the sum of two: the first is
0 x/2 0 y2
Z 2 2y Z 2 essentially the same as the one in 9.11.e, the second is 2p
14y4 y7
= (xy3 + x3 y/3) dy = − y5 − dy times the area of the half disk of radius r. Thus,
0 y2 0 3 3 Z r Z
2 p r p
14y5 y6 y8 128 I= 2u r2 − u2 du + 2p r2 − u2 du
= − − = . −r −r
15 6 24 0 15 π r2
= 0 + 2p · = pπ r 2 .
9.11. a. This is the integral given, with the iteration car- 2
ried out in the two possible orders, in Exercises 9.3.a & b.
Its value is 738. 9.11. g. Write T as 0 ≤ x ≤ 4, x ≤ y ≤ 4 (cf. Ex. 9.6.f);
√
9.11. b. Write the region as 0 ≤ x ≤ 1, x ≤ y ≤ x; then then
2
the double integral becomes ZZ Z 4Z 4 Z 4 2 4

√ xy
Z 1Z x√
Z 2 2 x Z 1 2 xy dA = xy dy dx = dx
xy x − x5 T 0 x 0 2 x
xy dy dx = dx = 0 dx Z 4 4
0 x2 0 2 x2 2 x3 x4
2
= 8x − dx = 4x − = 32.
=
1
−
1
= .
1 0 2 8 0
2 · 3 2 · 6 12
9.11. c. The given domain of integration makes the itera- 9.12. The solid shape is a tetrahedron in the first octant.
tion clear: Its base B in the (x, y)-plane (z = 0) is the triangle bounded
ZZ Z 1Z 3 Z 1
by the two positive axes and the line x + y = 4. Thus
y2 dA = y2 dy dx = 9 dx = 18.
−1 0 −1
−1≤x≤1 B : 0 ≤ x ≤ 4, 0 ≤ y ≤ 4 − x.
0≤y≤3
9.11. d. The given domain makes the iteration clear: The “top” of the tetrahedron is the graph z = 4 − x − y, so
ZZ Z 2 Z 4−y Z 2 2 4−y its volume is the double integral
x
x dA = x dx dy = dy ZZ Z 4 Z 4−x
0 y 0 2 y

0≤y≤2 (4 − x − y) dA = (4 − x − y) dy dx
y≤x≤4−y B 0 0
2 Z 4 4−x
Z 2 y2
= (8 − 4y) dy = 8y − 2y2 = 8. = (4 − x)y − dx
0
0 2 0
0 4
Z 4
(4 − x)2 −(4 − x)3 32
9.11. e. We describe K as = dx = = 3.
p p 0 2 6 0
−2 ≤ x ≤ 2, − 4 − x2 ≤ y ≤ 4 − x2;
therefore (with the substitution u = 4 − x2) 9.13. The base of the solid (the domain of integration)
ZZ Z Z
√ Z is a disk of radius R, and its “top” lies in the horizontal
2 4−x2 2 p
x dA = √ x dy dx = 2x 4 − x2 dx plane at height z = H. Therefore the solid is a cylinder; its
K −2 −
2
4−x2 −2
volume, and the value of the integral, is π R2 H.
−2
2 3/2
p
= (4 − x ) = 0. 9.14. The graph z = R2 − x2 − y2 is a hemisphere of
3 −2 radius R whose base is a disk of radius R. The solid shape
9.11. f. From the solution to Exercise 9.6.c we have is a half-ball; its volume, and the value of the integral, is
3 πR .
ZZ Z Z √ 2 3
p+r q+ r2 −(x−p)2
I= x dA = √2 x dy dx. 9.15. We express the double integral as an iterated inte-
K p−r q− r −(x−p)2
gral, and then use the fundamental theorem of calculus

twice in carrying out the iterated integrations. We have hence, if z = ax + by + c, then

ZZ Z bZ d ZZ
∂2 f ∂ ∂f 1 cπ R 2
I= dA = (x, y) dy dx z= z dA = = c.
D ∂x∂y a c ∂y ∂x area π R2
Z b d Z b x2 +y2 ≤R2
∂f
∂f ∂f
= (x, y) dx = (x, d) − (x, c) dx
a ∂x c a ∂x ∂x The average value of z on the disk does not depend on
b either a or b.

= f (x, d) − f (x, c) 9.18. b. The base of the solid is the disk
a of radius R centered at the origin. The
= f (b, d) − f (a, d) − f (b, c) + f (a, c) ‘top” is a sloping plane: the graph of
= f (R) − f (S) − f (Q) + f (P), z = ax + by + c. The solid is a vertical
cylinder with a sloping top. Think of it
as required. as a candle; the candle wick in the cen-
ter has length c, the average height of the
9.16. If f (x, y) = xy, then ∂ 2 f /∂ x ∂ y = 1, so
candle. By definition (of z), the volume
ZZ ZZ is the area of the base times the average
∂2 f
dA = dA = area(D). height; this is cπ R2 .
D ∂x∂y D
9.18. c. The average value of z = ax + by on the disk cen-
This is true no matter what values a and c have. However, tered at the origin is the value of z at the origin, namely 0.
when a = c = 0, then From part (b), the value of the integral is also 0.
f (P) − f (Q) + f (R) − f (S) = 0 − 0 + b · d − 0 = area(D). 9.19. a. Here the average value is the integral of the
function over the interval divided by the length of the in-
9.17. The idea here is that we should set terval. The value of the integral is
Z R R
∂2 f mx2
(x, y) = sin x sin y. (mx + b) dx = + bx = 2bR,
∂x∂y −R 2 −R
One function for which this is true is f (x, y) = cosx cos y. and the average value of y on the interval is y
Therefore, invoking Exercise 9.15 we have Z R
1 b
Z πZ π y= (mx + b) dx = b.
sin x sin y dy dx length −R
0 0 −R R x
= f (0, 0) − f (π , 0) + f (π , π ) − f (0, π ) 9.19. b. No, y does not depend on the slope m. Visually,
= 1 − (−1) + 1 − (−1) = 4. the average value of a function is the height of rectangle
whose area is the same as the area under the graph of that
function. For a linear function on an interval, that height
9.18. a. According to Exercise 9.11.f,
occurs at the midpoint of the interval.
ZZ
ax dA = 0. 9.19. c. The evidence points to the fact that the average
value of a linear function is taken on at the center of a
x2 +y2 ≤R2
region that is symmetric with respect to the origin.
By symmetry, we must also have 9.20. a. The average value of z over a region is its inte-
ZZ gral over that region divided by the area of the region. In
by dA = 0. the present case, the integral is the volume of the half-ball
x2 +y2 ≤R2 of radius R (Exercise 9.14). Therefore
3 πR
2 3
Because the integral of a constant over any domain is that
z= = 23 R.
constant times the area of the domain, we have π R2
ZZ
c dA = c · π R2; 9.20. b. The graph is a hemisphere of radius R, shown
x2 +y2 ≤R2 below, with one-quarter cut away for better visibility.

85
9.20. c. The top of the cylinder is shown below with the By part (a), we therefore have
ZZ p
hemisphere (also with one-quarter cut away). It appears −π
that the volume above the plane z = z and inside the hemi- ln x2 + y2 dA = lim 1 − ε 2 + 2ε 2 ln ε .
ε →0 2
sphere is about equal to the volume below the plane and x2 +y2 ≤1
outside the hemisphere. In this sense the volumes inside
We know (text p. 335) that r ln r → 0 as r → 0; therefore
the hemisphere and the cylinder appear about equal.
the limit displayed above is −π /2.
9.23. The given integral is improper because the inte-
√ when x =√1. However, on the given
grand is not defined
interval we have x2 − x4 = x 1 − x2, so we write
Z 1 Z b
dx dx
√ = lim √ .
1/2 x − x4 b→1
2 1/2 x 1 − x2
With the substitution x = cos s we have

Z Z Z
9.21. a. The graph is a semicircle of radius R, so the in- dx − sin s ds −ds 1 1 − sin s
√ = = = 2 ln
tegral of y over the interval −R ≤ x ≤ R is the area of the x 1 − x2 cos s sin s cos s 1 + sin s
semi disk, namely π R2 /2. The length of the integration from standard integral tables. Now s = π /3 when x = 1/2
interval is 2R, so and s = arccosb when x = b. Furthermore,
p
π R2 /2 π R sin(arccos b) = 1 − b2,
y= = .
2R 4
so
√ Z b
9.21. b. The semicircle y = r2 − x2 and the horizontal dx 1 1 − sins arccos b
√ = 2 ln
line y = y are shown together below. 1/2 x2 − x4 1 + sins π /3
√ ! √ !
1 1 − 1 − b2 1 1 − 3/2
= 2 ln √ − 2 ln √
1 + 1 − b2 1 + 3/2
As b → 1, the first term on the right vanishes. The sec-
ond term is unchanged and equals (by “rationalizing the
denominator”)
√ ! √ !
9.21. c. By definition, y is the height of the rectangle 1 2+ 3 1 (2 + 3)2 √
+ 2 ln √ = + 2 ln = ln(2 + 3).
whose area is the same as the area under the graph of y. 2− 3 1
The figure above indicates this in the sense that the area
of above the line and below the circle appears to be equal This is the value of the original improper integral.
to the area below the line and above the circle. 9.24. a. From the definition of D± we have
p
9.22. a. Use polar coordinates, so that D2± = (a cosht ∓ a2 + b2 )2 + b2 sinh2 t
ZZ Z Z 1 p
p 2π
= a2 cosh2 t ∓ 2a cosht a2 + b2 + a2 + b| 2 + b{z
2
sinh2 }t
ln x2 + y2 dA = dθ ln(r) r dr
0 ε p
ε 2 ≤x2 +y2 ≤1 1 = (a2 + b2) cosh2 t ∓ 2a a2 + b2 cosht + a2
r2 p
= 2π · (2 ln r − 1) 2
4 = a2 + b2 cosht ∓ a ,
ε
π √
= − 1 − ε 2(2 ln ε − 1) and thus D± = a2 + b2 cosht ∓ a.
2
−π 9.24. b. If a = sin s, b = cos s, then for all s the hyperbolas
= 1 − ε 2 + 2ε 2 ln ε . x = a cosht, y = b sinht satisfy the equation
2
x2 y2 x2 y2
9.22. b. By definition, − = − =1
a2 b2 sin2 s cos2 s
ZZ p ZZ p √
ln x2 + y2 dA = lim ln x2 + y2 dA. and have the focal points p± = (± a2 + b2, 0) = (±1, 0).
ε →0
x2 +y2 ≤1 ε 2 ≤x2 +y2 ≤1
For all s these hyperbolas are confocal.

9.24. c. Let x = (x, y) = (a cost, b sint) parametrize the 9.26. With just the change of variable t = −s we have
ellipses. Set D± = kx − p± k; our aim is to show that Z 0 Z 0 Z 0
D− + D+ = 2a, a constant. We have f (t) dt = f (−s) · (−ds) = − f (−s) ds.
p −a a a
D2± = (a cost ∓ a2 − b2)2 + b2 sin2 t
p Because f is an odd function (that is, f (−s) = − f (s)), we
= a2 cos2 t ∓ a cost a2 − b2 + a2 − b2 + b2 sin2 t have
p Z 0 Z 0 Z 0 Z a
= (a2 − b2) cos2 t ∓ 2a a2 − b2 cost + a2
p 2 − f (−s) ds = f (s) ds = f (s) ds = − f (s) ds.
= a2 − b2 cost ∓ a . a a a 0
Therefore, because the “dummy” variable of integration

Now (recall that we assume a > 0)
can be changed at will, the equation
p √
| a2 − b2 cost| ≤ a2 = a. Z a Z 0 Z a
√ f (t) dt = f (t) dt + f (t) dt
−a −a 0
We must therefore write D± = a ∓ a2 − b2 cost from Z a Z a
which it follows that =− f (t) dt + f (t) dt = 0
p p 0 0
D− + D+ = a + a2 − b2 cost + a − a2 − b2 cost = 2a,
holds for any odd function.
as we intended. 9.27. a. If y = a, then s = 1 + x + a2 and t = x − a. If we
9.24. d. If a = cosh s, b = sinh s, then for all s the ellipses rewrite the equation for s as x = s − 1 − a2, then we have
x = a cost, y = b sint satisfy the equation
t = x − a = s − 1 − a2 − a or t = s + b,
x2 y2 x2 y2
+ = + =1 where b = −1 − a − a2 . This is the line of slope 1 and
a2 b2 cosh2 s sinh2 s t-intercept b in the (s,t)-plane. The graph of the rela-
√
and have the focal points p± = (± a2 − b2 , 0) = (±1, 0). tionship between a and b in the (a, b)-plane is a parabola
For all s these ellipses are confocal. whose vertex is at the point (a, b) = (−1/2, −3/4). On
the left side of the parabola, where a ≤ −1/2, the as-
9.25. The identity that is suggested follows from signment a → b is 1–1. Thus when a ≤ −1/2, the lines
t = s + b all have different t-intercepts and hence are dis-
cos 2s + 1 = cos2 s − sin2 s + cos2 s + sin2 s = 2 cos2 s,
tinct.
cosh 2t − 1 = cosh2 t + sinh2 t − (cosh2 t − sinh2 t)
9.27. b. The image of the line y = a under the map ϕ −1
2
= 2 sinh t. is the line that can be parametrized as
Note that the given integral is oriented; to integrate it, we ℓ(x) = ϕ −1 (x, a) = (x + 1 + a2, x − a).
use iterated integrals, and choose to integrate first with
respect to t. Thus, using the identity, the given integral If ℓ(x1 ) = ℓ(x2 ), then x1 − a = x2 − a so x1 = x2 . Thus ℓ is
becomes the iterated integral 1–1, so ϕ −1 is 1–1 on each line y = a.
Z bZ d 9.28. The spherical coordinate map is
1
(cos 2s + cosh 2t) dt ds 
2 a c
Z d
x = ρ cos θ cos ϕ ,

1 b
= 2t cos 2s + sinh 2t ds s : y = ρ sin θ cos ϕ ,
4 a 

Z b
c z = ρ sin ϕ .
1
= 2(d − c) cos 2s + sinh 2d − sinh 2c ds
4a The change of variables formula for triple integrals is de-
fined by analogy with the formula for double integrals:
1 b
= (d − c) sin 2s + s(sinh 2d − sinh2c)
4 a
ZZZ
(d − c)(sin 2b − sin 2a) + (b − a)(sinh2d − sinh 2c) f (x, y, z) dx dy dz
= . ZZ
s(D)
4
= f (ρ cos θ cos ϕ , ρ sin θ cos ϕ , ρ sin ϕ )|Js | d ρ d θ d ϕ .
This agrees with the value given on page 360 of the text. D

87
The Jacobian Js (ρ , θ , ϕ ) was obtained in the solution to On side 4, (x, y) = (0, 1 − t), 0 ≤ t ≤ 1, so
Exercise 5.10.b (Solutions page 49). It is Z Z 1
∂ (x, y, z) xy dx + y dy = (1 − t) (−dt) = − 12 .
Js (ρ , θ , ϕ ) = = ρ 2 cos ϕ ; side 4 0
∂ (ρ , θ , ϕ )
The double integral is
this is nonnegative everywhere so we can remove the ab-
ZZ Z 1Z 1 Z 1
solute value sign in the integral.
(Qx − Py ) dA = −x dy dx = x dx = − 21 .
9.29. The 4-dimensional analogue of the spherical coor- R 0 0 0
dinates map that we use is The integrals have the same value.

 x1 = r cost1 cost2 cost3 , 9.31. a. According to Green’s theorem,


 x2 = r sin t1 cost2 cost3 , I
σ:

 x3 = r sin 2 cost3 , f (x) dx + (ax2 + bxy + cy2) dy

 ZZ
x4 = r sin t3 . x2 +y2 =R2
= (2ax + by) dx dy = 0
We are given a domain D in (r,t1 ,t2 ,t3 )-space and a func- x2 +y2 ≤R2
tion f (x1 , x2 , x3 , x4 ) defined on σ (D). For notational sim-
plicity let by Exercise 9.18.c, for any f (x), a, b, and c.
9.31. b. Again by Green’s theorem and Exercise 9.18.c:
g(r,t1 ,t2 ,t3 ) = f (σ (r,t1 ,t2 ,t3 ))
I
The the change of variables formula in this setting is f (x) dx + (ax2 + bxy + cy2 + α x + β y + γ ) dy
ZZZZ ZZ
x2 +y2 =R2
f (x1 , x2 , x3 , x4 ) dx1 dx2 dx3 dx4 = (2ax + by + α ) dx dy = απ R2 .
σ (D)
ZZZZ x2 +y2 ≤R2
= g(r,t1 ,t2 ,t3 ) |Jσ (r,t1 ,t2 ,t3 )| dr dt1 dt2 dt3 .
D
9.32. a. By Green’s theorem,
The solution to Exercise 5.25.b (Solutions page 58) pro- I ZZ
vides the Jacobian of σ : 5y dx + 2x dy = (2 − 5) dx dy = −3 area(~T ),
~
C ~T
Jσ (r,t1 ,t2 ,t3 ) = r3 cost2 cos2 t3 ;
where ~T is the triangular region whose boundary, C, ~ is
this is nonnegative everywhere so we can remove the ab- oriented counterclockwise by its vertices. The oriented
solute value signs in the integral. area of T is one half the oriented area of the parallelogram
9.30. The path integral whose sides are
I
xy dx + y dy = − 21 (9, 2) − (1, 5) = (8, −3) and (8, 8) − (1, 5) = (7, 3).
∂R
Thus,
is calculated one side at a time, as follows. Side 1 (from

(0, 0) to (1, 0)) can be parametrized as (x, y) = (t, 0) with 3 8 −3 135
−3 area(~T ) = − =− .
0 ≤ t ≤ 1, so 2 7 3 2
Z
xy dx + y dy = 0.
side 1 9.32. b. Green’s theorem gives
I ZZ
On side 2 (from (1, 0) to (1, 1)), (x, y) = (1,t), 0 ≤ t ≤ 1,
(x2 − x3 ) dx + (x3 + y2) dy = 3x2 dx dy,
so Z Z ~
C ~D
1
xy dx + y dy = t dt = 12 .
side 2 0 where ~D is the unit disk with its positive orientation. The
On side 3 (from (1, 1) to (0, 1)), (x, y) = (1 − t, 1), double integral can be computed using polar coordinates,
0 ≤ t ≤ 1, so for example. Thus we have 3x2 = 3r2 cos2 θ and
Z Z 1 ZZ Z 2π Z 1
3 3π
xy dx + y dy = (1 − t) (−dt) = − 21 . 3x2 dx dy = cos2 θ d θ 3r3 dr = π · = .
side 3 0 ~
D 0 0 4 4

9.32. c. In this case Qx − Py = ey − ex and the double in- Theorem 9.21 (text p. 370) with g(x, y) ≡ 1, ~T = ~S = ~D
tegral can be calculated as the iterated integral gives the following:
Z 7Z 5 Z 7 5 ZZ ZZ

y x
(e − e ) dy dx = y x
(e − ye ) dx Jq ds dt = 3 dx dy = 3 area(~D) = 3π ,
−1 1 −1 ~D ~D
1
Z 7 ZZ ZZ
= 5 x
(e − e − 4e ) dx = 8(e − e) − 4(e − e ). 5 7 −1 Js ds dt = 4 dx dy = 4 area(~D) = 4π .
~D ~D
−1
9.33. Here Qx − Py = 1, so Green’s theorem gives

I ZZ 9.38. a. 2.0
x dy = 1 · dx dy = area(~R).
∂ ~R ~R
1.5
9.34. For the first path integral, Qx − Py = 0 − (−1) = 1;
for the second, Qx − Py = 12 − (− 12 ) = 1. Therefore, when
Green’s theorem is applied to either of these path inte-
grals, the result is the same as in the preceding solution. 1.0
9.35. y
0.5
2
0.0
−6 −4 −2 2 4 6 x -1.0 -0.5 0.0 0.5 1.0
−2
9.38. b. We have

∂ (u, v) 2x −2y
9.35. a. We can use the parametrization = = 4(x2 + y2 ).
∂ (x, y) 2y 2x
x(t) = a cost, y = b sint, 0 ≤ t ≤ 2π .
Then Furthermore, u2 + v2 = (x2 − y2 )2 + 4x2 y2 = (x2 + y2 )2 , so
x2 y2 a2 cos2 t b2 sin2 t
+ = + = 1, ∂ (x, y) 1 1 1
a2 b2 a2 b2 = = = √ .
as required. ∂ (u, v) ∂ (u, v)/∂ (x, y) 4(x + y ) 4 u + v2
2 2 2
9.35. b. By Exercise 9.33,

I Z 2π 9.38. c. By the change of variables formula,
area(D) = x dy = a cost · b cost dt ZZ ZZ ZZ
∂ ~D
Z 2π
0 ∂ (u, v)
area(A) = du dv = du dv = dx dy
= ab cos t dt = π ab.
2 A g(S) ∂ (x, y) S
0 Z 1Z 1 Z 1 1
y3
9.36. In this case Qx − Py = −Hxx − Hyy and this equals = 4(x2 + y2 ) dy dx = 4 x2 y + dx
0.2 0.2 0.2 3 0.2
~ = ∂ ~D, then
zero because H is harmonic. Therefore, if C Z 1
.992
I ZZ =4 0.8x2 + dx
∂H ∂H ∂ 2H ∂ 2H 0.2 3
dx − dy = − 2 − 2 dx dy
~ ∂y
C ∂x ~D ∂x ∂y 1
ZZ x3 .992
= 4 0.8 + x
= 0 dx dy = 0. 3 3 0.2
~D
.992 .992 3968
9.37. The solution to Exercise 4.22.c (Solutions page 40) = 4 0.8 + 0.8 = .
3 3 1875
shows that the map q triples angles at the origin; the solu-
tion to Exercise 4.23.c (Solutions page 41) shows that the The fraction equals 2.11626666. . . in decimal form. We
map s quadruples angles at the origin. It follows that q is can compare this value to estimates of the integral ob-
a three-fold and orientation-preserving cover of the unit tained in the solution to Exercise 8.3 (Solutions page 74);
disk, while s is four-fold. the closest is 2.11626.

89
9.38. d. The change of variables formula gives The second is

ZZ ZZ ZZ ZZ
du dv du dv du dv 4(x2 + y2 )
√ = √ 2 2
= 2 2 2
dx dy
A u 2 + v2 g(S) u2 + v2 A u +v S (x + y )
ZZ ZZ
1 ∂ (u, v) =
4 dx dy
= 4.37766.
= dx dy
S x + y ∂ (x, y)
2 2
2 2 S x +y
ZZ
1
= 2 + y2
4(x2 + y2 ) dx dy
S x 9.39. The image g(D) of the quarter-disk D is the semi-
ZZ
disk H : u2 + v2 ≤ 1, v ≥ 0. Obviously, areag(D) = π /2.
= 4 dx dy = 4 area(S).
S The value given by the change of variables formula is
ZZ ZZ
9.38. e. Use the change of variables formula to compute areag(D) = du dv = 4(x2 + y2 ) dx dy.
g(D) D
these integrals. The moment around the v-axis is
ZZ ZZ The shape of D suggests evaluating the last integral using
u du dv = 2 2 2
(x − y ) · 4(x + y ) dx dy 2 polar coordinates:
A S
Z 1Z 1 ZZ Z π /2 Z 1
π
=4 (x4 − y4 ) dy dx 4(x2 + y2 ) dx dy = dθ 4r2 · r dr = · 1.
0.2 0.2 D 0 0 2
Z 1 1
4 y5 The second integral can also be computed using polar co-
=4 x y − dx
0.2 5 0.2 ordinates, but this time directly in the the (u, v)-plane (i.e.,
Z 1
4 0.99968 without using g to change variables):
=4 0.8x − dx
0.2 5 ZZ p Z Z 1
π 1
0.99968 0.99968 u2 + v2 du dv = dθ r · r dr = π · .
= 4 0.8 − 0.8 = 0. g(D) 0 0 3
5 5
Here is an alternate computation that uses g to change
The moment around the u-axis is
variables:
ZZ ZZ
ZZ p ZZ
v du dv = 2xy · 4(x2 + y2) dx dy u2 + v2 du dv = (x2 + y2 ) · 4(x2 + y2 ) dx dy.
A S
Z 1Z 1 g(D) D
3 3
=8 (x y + xy ) dy dx
0.2 0.2 At this point it is again useful to switch to polar coordi-
Z 1 3 2 1 nates:
x y xy4
=8 − dx
0.2 2 4 0.2 ZZ
Z 1 (x2 + y2) · 4(x2 + y2 ) dx dy
0.96 x3 0.9984 x D
=8 + dx Z π /2 Z 1
0.2 2 4 π 4 π
1 = dθ 4r4 · r dr = · = .
x4 x2 0 0 2 6 3
= 8 0.96 + 0.9984
8 8 0.2
9.40. The given expression includes an improper integral
0.9984 0.96
= 8 0.96 + 0.9984 that we determine as follows:
8 8
29952 Z 1 Z 1 1
dx dx
= = 1.916928. = lim = lim ln x = lim (− ln a) = ∞.
15625 0 x a→0 a x a→0 a→0
a
9.38. f. Use a computer algebra system in conjunction 9.41. This improper integral has a finite value that is
with the change of variables formula to compute these in- most conveniently found by changing to polar coordi-
tegrals. The first is nates:
ZZ ZZ ZZ Z 2π Z 1
du dv 4(x2 + y2 ) dx dy r dr
= dx dy = 3.09012. = lim = lim 2π (1 − a) = 2π .
A v S 2xy B2 r a→0 0 a r a→0

9.42. Change to spherical coordinates (cf. the solution 9.45. a. Let T : R2 ≤ x2 + y2 + z2 ≤ (R + ∆R)2 be the thin
to Exercise 9.28, Solutions p. 86) to evaluate this im- shell. A change to spherical coordinates gives
proper integral. For the volume element dx dy dz we have ZZZ Z 2π Z π /2 Z R+∆R
ρ 2 cos ϕ d ρ d θ d ϕ ; thus dx dy dz = dθ cos ϕ d ϕ ρ 2 dρ
T 0 −π /2 R
ZZZ Z 2π Z π /2 Z 1 2
dx dy dz ρ dρ R+∆R
= lim dθ cos ϕ d ϕ ρ
3 4π
ρ −π /2 ρ = 4π · = (R + ∆R)3 − R3 .
3 R
B3 a→0 0 a
3
2 1
ρ
= lim 2π · 2 · = 2π .
a→0 2 a 9.45. b. We have

It is plausible to conjecture that the integral of 1/r over (R + ∆R)3 − R3 = R3 + 3R2∆R + O (∆R)2 − R3
the unit ball in Rn is finite if n ≥ 2.
= 3R2 ∆R + O (∆R)2 ;
9.43. a. We repeat the calculations in the solutions to Ex-
thus
ercises 9.40–9.43, changing the integrand to 1/r2 .
ZZZ
4π
• n = 1. Here 1/r2 = 1/x2 and dx dy dz = 3R2 ∆R + O (∆R)2
T 3

Z Z 1 Z 1 = 4π R2∆R + O (∆R)2 .
dx dx dx −2 1
= = lim 2 = lim = ∞.
B1 r2 −1 x2 a→0 a x2 a→0 x a
9.46. The domain of integration here is same thin shell T
that appeared in the preceding exercise. Change to spher-
• n = 2. Changing to polar coordinates, we get
ical coordinates to compute the integral:
Z Z 2π Z 1
dx dy r dr ZZZ Z 2π Z π /2 Z R+∆R 2
= lim dθ = 2π lim (− ln a) = ∞. dx dy dz ρ dρ
B2 r2 a→0 r2 a→0 = dθ cos ϕ d ϕ
0 a
T x2 + y2 + z2 0 −π /2 R ρ2
R+∆R
• n = 3. Changing to spherical coordinates, we get
= 4π · ρ = 4π ∆R.
ZZZ Z 2π Z π /2 Z 1 2 R
dx dy dz ρ dρ
= lim dθ cos ϕ d ϕ
B3 ρ2 a→0 0 −π /2 a ρ2
= lim 2π · 2 · (1 − a) = 4π .
a→0
It appears that the integral of 1/r2 over the unit ball in Rn

is finite for n ≥ 3.
9.43. b. We could say that 1/r2 is “more infinite” than

1/r at the origin in the sense that in every dimension in
which the integral of 1/r is infinite, so is that of 1/r2 , and
in addition the integral of 1/r2 is infinite in R2 while that
of 1/r is finite.
9.44. The integral is not improper. Under a change to

spherical coordinates it is
ZZZ
dx dy dz
1 + x2 + y2 + z2
x2 +y2 +z2 ≤1
Z 2π Z π /2 Z 1 2
ρ dρ
= dθ cos ϕ d ϕ
0 −π /2 0 1 + ρ2
Z 1
1 + ρ2 − 1
= 4π d ρ = 4π (1 − arctan1) = 4π − π 2.
0 1 + ρ2

Surface Integrals
10.1. a. For each square, ∆A = 1 and the unit normal is 10.2. a. The flux must now be computed as an integral,
the one of the basis vectors; thus but the computation may be simplified when the surface
through which the flow occurs is planar and thus has a
for (x, y)-plane: Φ = (2, −7, 1) · (0, 0, 1) = 1 kg,
constant normal. In the first case, n = (1, 0, 0) and
for (y, z)-plane: Φ = (2, −7, 1) · (1, 0, 0) = 2 kg,
V · n = (0, z, x) · (1, 0, 0) = 0 so Φ = 0.
for (z, x)-plane: Φ = (2, −7, 1) · (0, 1, 0) = −7 kg,
10.1. b. Here n = (0, 0, 1) and the matter flow is 10.2. b. Here V · n = (0, z, x) · (0, 0, −1) = −x so
ZZ Z 1 Z 1
Φ ∆t = V · n ∆A ∆t Φ= −x dx dy = −x dx dy
0 0
kg/m2 unit Z 1
= (2, −7, 1) · (0, 0, 1) × 12 m2 × 7 sec square 1 1
sec = − dy = − .
= 84 kg. 0 2 2
10.2. c. In this case the integral is readily calculated

10.1. c. The rectangle lies in the plane x = 5; it has area
using Definition 10.3 for the flux (text p. 395) and a
18 m2 and its unit normal is (1, 0, 0). Therefore
parametrization of the oriented triangle. Our parametriza-
Φ ∆t = V · n ∆A ∆t = 2 × 18 × 10 = 360 kg. tion is the linear inhomogeneous (i.e. affine) map f de-
fined on the (u, v)-plane by
10.1. d. The unit normal n for the region is determined by
z = 1 in which (0, 0) → (2, 2, 0), (1, 0) → (0, 2, 2), (0, 1) → (2, 0, 2).
the normal N = (1, 1, 1) to the√plane x√+ y +√
the region lies; thus n = (1/ 3, 1/ 3, 1/ 3). Because Thus,
the region has unit area, we find 
√ √ √ √ 
x = 2 − 2u,
Φ = (2, −7, 1) · (1/ 3, 1/ 3, 1/ 3) × 1 = −4/ 3 kg. ~T : 0 ≤ u ≤ 1,
f : y = 2 − 2v,

 0 ≤ v ≤ 1 − u.
10.1. e. We label each face of Q by the plane it lies in; z = 2u + 2v;
thus
In the integral of Def. 10.3 we take X = 0, Y = z = 2u+2v,
x = 1 : Φ = V · (1, 0, 0) = 2, Z = x = 2 − 2u, and
x = 0 : Φ = V · (−1, 0, 0) = −2, ∂ (z, x) ∂ (x, y)
= 4, = 4.
y = 1 : Φ = V · (0, 1, 0) = −7, ∂ (u, v) ∂ (u, v)
y = 0 : Φ = V · (0, −1, 0) = 7, The integral itself is
z = 1 : Φ = V · (0, 0, 1) = 1, ZZ
z = 0 : Φ = V · (0, 0, −1) = −1, Φf = 4(2u + 2v) + 4(2 − 2v) du dv
~T
Z 1 Z 1−u
The flows through pairs of opposite faces are negatives of = (8 − 8v) dv du
each other, because what flows into one (indicated by a 0 0
Z 1
negative value for Φ) flows out of the other. The net, or 8
= 8(1 − u) − 4(1 − u)2 du = .
algebraic, total flux Φ over the six faces is zero. 0 3
91

92 SOLUTIONS: CHAPTER 10. SURFACE INTEGRALS
10.3. We compute the flux using the integral of Def. 10.3. • y = 3: V · n = (−y, x, 0) · (0, 1, 0) = x.
The Jacobians in the integral are Z 2Z 5 Z 2
25
∂ (y, z) 2u −2v Φy=3 = x dx dz = dz = 25.
= = 4(u2 + v2 ), 0 0 0 2
∂ (u, v) 2v 2u
• z = 0: V · n = (−y, x, 0) · (0, 0 − 1) = 0. Φz=0 = 0.
∂ (z, x) 2v 2u
= = 2(v − u),
∂ (u, v) 1 1 • z = 2: V · n = (−y, x, 0) · (0, 0, 1) = 0. Φz=2 = 0.

∂ (x, y) 1 1 The total flux through the parallelepiped is Φ = 0.
= = −2(u + v),
∂ (u, v) 2u −2v
10.6. a. We show kg(u, v)k2 = 1 for every (u, v) in R2 .
and the integrand itself is Let 1 + u2 + v2 = D; then
(u + v) · 4(u2 + v2 ) + (u2 − v2 ) · 2(v − u) 4u2 4v2 1 − 2(u2 + v2 ) + (u2 + v2 )2

x2 = , y2 = , z2 = ,
+ 2uv · (−2(u + v)) D2 D2 D2
= 2u3 + 2u2v + 2uv2 + 2v3. and so

1 + 2(u2 + v2) + (u2 + v2 )2
Thus x2 + y2 + z2 =
Z 3 Z 1 D2
Φ= 3 2
(2u + 2u v + 2uv + 2v ) dv du 2 3 (1 + u + v )2
2 2
= = 1.
0

0
D2
Z 3
3 2 2u 1
= 2u + u + + du Hence g(R2 ) is contained in the unit sphere in R3 . Sup-
0 3 2
3 pose (u, v) were a point for which g(u, v) = (0, 0, −1). Be-
u4 u3 u2 u cause the first two coordinates here are 0, we must have
= + + + = 54.
2 3 3 2 0 u = v = 0. But then z = 1 − 0/(1 + 0) = +1, not −1. Thus
b
g(R2 ) excludes (0, 0, −1) so g(R2 ) ⊆ S.
10.4. On the sphere of radius R, the unit normal is 10.6. b. First suppose v = 0; then
x y z
n(x, y, z) = , , , 1 − u2 2
R R R z= and 1 + z = .
1 + u2 1 + u2
so V · n = (x2 + y2 + z2 )/R = R. Therefore,
ZZ ZZ When v = 0 we also have
V · n dA = R dA = R · area(S) = R · 4π R2 = 4π R3. 2u x
S S x= = (1 + z)u, so u = .
1 + u2 z+1
10.5. Here we calculate the flux as an integral over each Note u is well defined because z + 1 > 0 on S. b If we re-
of the six faces. Each face is parametrized by the two co- verse the roles of u and v, we find v = y/(z + 1). In other
ordinate variables that are not fixed on that face. Note that
words, the inverse of g on Sb is
the normals for opposite faces point in opposite directions.
x y
• x = 0: V · n = (−y, x, 0) · (−1, 0, 0) = y. u= , v= .
z+1 z+1
Z 2Z 3 Z 2
9
Φx=0 = y dy dz = dz = 9.
0 0 0 2 10.7. We assume that f : Ω → R3 and g : Ω∗ → R3 are
given by the formulas
• x = 5: V · n = (−y, x, 0) · (1, 0, 0) = −y.  
Z 2Z 3 Z 2 
x = x(u, v), x = ξ (s,t),

9
Φx=5 = −y dy dz = − dz = −9. f : y = y(u, v), g : y = η (s,t),
0 0 0 2 
 

z = z(u, v); z = ζ (s,t).
• y = 0: V · n = (−y, x, 0) · (0, −1, 0) = −x.
The key to the proof of Theorem 10.7 is the existence of a
Z 2Z 5 Z 2 coordinate change ϕ : Ω∗ → Ω for which ϕ (U ∗ ) = U and
25
Φy=0 = −x dx dz = − dz = −25. g(s,t) = f(ϕ (s,t)) for all (s,t) in Ω∗ .
0 0 0 2

93
We take as given for this proof all the Jacobians and Then
all the equations between them that appear in the proof of Z Z √ √
− cos ϕ d ϕ 1 du u 1 − sin ϕ
Theorem 10.6. If we abbreviate the local area multipliers √ = √ √ =√ = √ ,
2 3/2 1 − sin ϕ 2 2 u 2 2
for the parametrizations of f and g as
so
s √ √
Z b
∂ (y, z) 2 ∂ (z, x) 2 ∂ (x, y) 2 − cos ϕ d ϕ 1 − sinb − 2
Mf (u, v) = + + , lim √ = lim √ = −1,
∂ (u, v) ∂ (u, v) ∂ (u, v) b→π /2 −π /2 23/2 1 − sin ϕ b→π /2 2
s
∂ (η , ζ ) 2 ∂ (ζ , ξ ) 2 ∂ (ξ , η ) 2 as required.
Mg (s,t) = + + ,
∂ (s,t) ∂ (s,t) ∂ (s,t) 10.9. For the given parametrization of the torus,

then the chain rule implies ∂ (y, z) (R + a cosv) cos u −a sin u sin v
=
∂ (u, v) 0 a cosv
∂ (u, v)

Mg (s,t) = Mf (ϕ (s,t)) . = a(R + a cosv) cos u cos v,
∂ (s,t)
∂ (z, x) 0 a cos v
=
(The abbreviations make it easier to fit equations into the ∂ (u, v) −(R + a cosv) sin u −a cosu sin v
two-column format of this page.) Furthermore, = a(R + a cosv) sin u cos v,

∂ (x, y) −(R + a cosv) sin u −a cosu sin v
H(g(s,t)) = H(f(ϕ (s,t))) = H(f(u, v)). =
∂ (u, v) (R + a cosv) cos u −a sin u sin v
Therefore, we can express the integral over the domain U ∗ = a(R + a cosv) sin v.
in two ways, and then convert the second into an integral
over the domain U = ϕ (U ∗ ) by using the basic change of Therefore the local area multiplier is
s
variables formula (Theorem 9.11):
∂ (y, z) 2 ∂ (z, x) 2 ∂ (x, y) 2
ZZ + + = a(R+a cosv),
∂ (u, v) ∂ (u, v) ∂ (u, v)
H(g(s,t)) Mg (s,t) ds dt
U∗
ZZ and the surface area is
∂ (u, v)
= H(f(ϕ (s,t))) Mf (ϕ (s,t))
ds dt Z 2π Z 2π
U∗ ∂ (s,t) du (aR + a2 cos v) dv = 2π · 2π aR = 4π 2 aR.
ZZ 0 0
= H(f(u, v)) Mf (u, v) du dv (The average value of a2 cos v on 0 ≤ v ≤ 2π is zero, so
ϕ (U ∗ )
ZZ the integral of that term is zero.)
= H(f(u, v)) Mf (u, v) du dv. 10.10. a. dg = gx dx + gy dy = (3x2 − y2 ) dx − 6xy dy.
U
10.10. b. dg = y cos xy dx + x cosxy dy.
This proves that the surface integral
ZZ 10.10. c. dg = (cos y − y cosx) dx − (x sin y + sin x) dy.
H(x, y, z) dA x y
S 10.10. d. dg = 2 2
dx + 2 dy.
x +y x + y2
is defined independently of the parametrization of the sur- 10.10. e. dg = (y + z) dx + (z + x) dy + (x + y) dz.
face S.
10.10. f. dg = sin ϕ cos θ d ρ + ρ cos ϕ cos θ d ϕ
10.8. The integral is improper because the integrand be- − ρ sin ϕ sin θ d θ .
comes infinite as ϕ → π /2. To evaluate the integral, we −y x
first write the integrand as 10.10. g. dg = 2 dx + 2 dy.
x + y2 x + y2
(sin ϕ − 1) cos ϕ − cos ϕ d ϕ du 10.10. h. dg = u dx − v dy + x du − y dv.
d ϕ = 3/2 √ = 3/2 √ , u xu x xu
(2 − 2 sin ϕ )3/2 2 1 − sin ϕ 2 u 10.10. i. dg = dx − 2 dy + du − 2 dv.
yv y v yv yv
using the change of variable u = 1 − sin ϕ . Therefore, 10.10. j. We use xbi to indicate that the factor xi is missing;
Z π /2 Z b then
(sin ϕ − 1) cos ϕ − cos ϕ d ϕ n
d ϕ = lim √ . dg = ∑ x1 · · · xi−1 xbi xi+1 · · · xn dxi .
−π /2 (2 − 2 sin ϕ )3/2 b→π /2 −π /2 23/2 1 − sin ϕ i=1

10.11. The differential of a general 1-form is 10.11. k. From the preceding solution, we have
! !
∂ Pi cj · · · dxn = (−1)k−1 dx1 · · · dx j · · · dxn .
d x j dx1 · · · dx
d ∑ Pi dxi = ∑(dPi ) dxi = ∑ ∑ dx j dxi .
i i i j6=i ∂ x j Therefore d ω in this exercise is an alternating sum of
identical terms; consequently
10.11. a. d ω = dy dx − dx dy = −2 dx dy. (
0 n even,
10.11. b. d ω = (−2y dy) dx − (2x dx) dy dω =
= (2y − 2x) dx dy. dx1 · · · dxn n odd.

−dy dx 1 1
10.11. c. d ω = 2 dx + 2 dy = + dx dy. 10.12. a. d(dg) = d(3x2 − 3y2) dx − d(6xy) dy
y x x2 y2 = −6y dy dx − 6y dx dy
10.11. d. d ω = (dy − dz) dx + (dz − dx) dy = (6y − 6y) dx dy = 0.
+ (dx − dy) dz
10.12. b. d(dg) = (cos xy − xy sin xy) dy dx
= −2 dy dz − 2 dz dx − 2 dx dy.
+ (cos xy − xy sin xy) dx dy = 0.
10.11. e. The computation is clearer if we split the sum
10.12. c. d(dg) = (− sin y − cosx) dy dx
into two sums:
− (sin y + cosx) dx dy = 0.
n−1 n−1
dω = ∑ dx j−1 dx j − ∑ dx j+1 dx j 10.12. d. d(dg) = 2
−2xy
2 2
dy dx + 2
−2xy
dx dy = 0.
j=2 j=2 (x + y ) (x + y2 )2
n−1 n−1 10.12. e. d(dg) = (dy + dz) dx + (dz + dx) dy
= ∑ dx j−1 dx j + ∑ dx j dx j+1 + (dx + dy) dz
j=2 j=2
= −dx dy + dz dx − dy dz + dx dy
= dx1 dx2 + 2(dx2 dx3 + · · · + dxn−2 dxn−1 ) − dz dx + dy dz = 0.
+ dxn−1 dxn . 10.12. f. d(dg) = cos ϕ cos θ d ϕ d ρ − sin ϕ sin θ d θ d ρ
+ cos ϕ cos θ d ρ d ϕ − ρ cos ϕ sin θ d θ d ϕ
10.11. f. d ω = 2u du dv dw + 2v dv dw du + 2w dw du dv
− sin ϕ sin θ d ρ d θ − ρ cos ϕ sin θ d ϕ d θ
= 2(u + v + w) du dv dw.
= (− cos ϕ cos θ + cos ϕ cos θ ) d ρ d ϕ
10.11. g. d ω = (uex dx − vey dy + ex du − ey dv) dx dy + (ρ cos ϕ sin θ − ρ cos ϕ sin θ ) d ϕ d θ
+ (2x dx + 2y dy) du dv + (− sin ϕ sin θ + sin ϕ sin θ ) d θ d ρ
= ex dx dy du − ey dx dy dv + 2x dx du dv = 0.
+ 2y dy du dv.
y2 − x2 y2 − x2
10.11. h. d ω = cosh u cosh v dx dy du 10.12. g. d(dg) = 2 2 2
dy dx + 2 dx dy = 0.
(x + y ) (x + y2 )2
+ sinh u sinh v dx dy dv
+ cosx cos y dx du dv 10.12. h. d(dg) = du dx − dv dy + dx du − dy dv = 0.
− sin x sin y dy du dv. u 1 u
10.12. i. d(dg) = − 2 dy dx + du dx − 2 dv dx
n y v yv yv
10.11. i. d ω = ∑ d p j dq j . u x xu
− 2 dx dy − 2 du dy + 2 2 dv dy
j=1 y v y v y v
1 x x
10.11. j. As a first step we have + dx du − 2 dy du − 2 dv du
n
yv y v yv
u xu x
dω = ∑ (−1) j−1dx j dx1 · · · dx
cj · · · dxn ). − 2 dx dv + 2 2 dy dv − 2 du dv.
yv y v yv
j=1
These 12 terms cancel in pairs, giving d 2 g = 0.
We can move dx j to its natural numerical position in the
cj · · · dxn with a sequence of j − 1 trans- 10.12. j. Again using xbj to indicate that the factor x j is
product dx1 · · · dx
missing from a product, we have
positions with its neighbor to its immediate right. Each
transposition causes a change in sign; thus d(dg) = ∑ x1 · · · xbi · · · xbj · · · xn dx j dxi
j6=i
cj · · · dxn = (−1)k−1 dx1 · · · dx j · · · dxn .
dx j dx1 · · · dx
= ∑ x1 · · · xbj · · · xbi · · · xn dx j dxi
It follows that j<i
d ω = n dx1 · · · dx j · · · dxn . + ∑ x1 · · · xbi · · · xbj · · · xn dx j dxi .

i< j

95
In the second sum on the right, make the exchange i ↔ j. 10.15. a. d ω = dx dx − sin y dy dy = 0. We can recover
This changes i < j to j < i and dx j dxi to dxi dx j (with- f by “integrating” ω . Thus
out changing its sign). Continuing with the second sum,
x2
replace dxi dx j by −dx j dxi ; the result is f (x, y) = + sin y.
2
d(dg) = ∑ x1 · · · xbj · · · xbi · · · xn dx j dxi
j<i 10.15. b. Here d ω = h′ (x) dx dx + g′ (y) dy dy = 0. This
exercise generalizes the preceding one; we have
− ∑ x1 · · · xbi · · · xbj · · · xn dx j dxi = 0. Z Z
j<i x y
f (x, y) = h(s) ds + g(t) dt.
10.13. a. b. & c. Here d(d ω ) is a 3-form in two variables,

10.15. c. d ω = 2v dv du + 2u du dv = 0 and f (u, v) = 2uv.
so it is automatically zero.
10.15. d. We have
10.13. d. & e. The coefficients of d ω are all constants, so
d(d ω ) = 0. d ω = (z dy + y dz) dx + (z dx + x dz) dy + (y dx + x dy) dz
10.13. f. Here d(d ω ) is a 4-form in three variables, so it = −z dx dy + y dz dx + z dx dy − x dy dz
is automatically zero. − y dz dx + x dy dz = 0.
10.13. g. d(d ω ) = ex dx dx dy du − ey dy dx dy dv We can take f (x, y, z) = xyz.

+ 2 dx dx du dv + 2 dy dy du dv = 0 10.15. e. Here ω = dg where g is the function given in
because there is a repeated basic differential in each term. Exercise 10.10.e, so d ω = d 2 g = 0 is established in the
10.13. h. For the sake of brevity, in the following calcu- solution to Exercise 10.12.e. It follows that we can take
lation of d(d ω ) we exclude any term that has a repeated f (x, y, z) = g(x, y, z) = xy + yz + zx.
basic differential; thus d(d ω ) =
−1 1 x
= cosh u sinh v dv dx dy du + coshu sinh v du dx dy dv 10.15. f. d ω = 2
dy dx − 2 dx dy = 0 and f (x, y) = .
y y y
− cosx sin y dy dx du dv − cosx sin y dx dy du dv = 0 10.15. g. Here ω = dg from Exercise 10.10.i, and d ω =
d 2 g = 0 is confirmed in the solution to Exercise 10.12.i.
because dv dx dy du = −dx dy du dv = du dx dy dv and
Finally, we can take
dy dx du dv = −dx dy du dv.
xu
10.13. i. All of the coefficients of d ω are constants so f (x, y, u, v) = g(x, y, u, v) = .
yv
d(d ω ) = 0.
10.13. j. & k. The coefficients of d ω are constants, so 10.16. Note that if d ω = α , then d(ω + d f ) = α for any
d(d ω ) is automatically zero. Alternatively, d(d ω ) is au- 0-form f , so the range of possible choices for the integral
tomatically zero as an (n + 1)-form in n variables. ω here is much larger that in the preceding exercise.
10.16. a. Here d α = 0 because it is a 3-from in two vari-
10.14. a. du ∧ dv = dx ∧ (2y dy) = 2y dx ∧ dy.
ables. Two possible choices for ω are
10.14. b. Because terms that have a repeated differential 2 2
x y
vanish, we can write ω= − xy dy and ω = − xy dx.
2 2
du ∧ dv = (cos θ dr) ∧ (r cos θ d θ ) A different kind of choice is ω = −xy(dx + dy).
+ (−r sin θ d θ ) ∧ (sin θ dr)
10.16. b. Again d α = 0 because it is a 3-form in two vari-
= (r cos2 θ + r sin2 θ ) dr ∧ d θ = r dr ∧ d θ . ables. The preceding solution suggests that we introduce
Z x Z y
p(x, y) = ϕ (t, y) dt or q(x, y) = ϕ (x,t) dt;
10.14. c. du ∧ dv = 4x2 dx ∧ dy − 4y2 dy ∧ dx
= 4(x2 + y2 ) dx ∧ dy. then either
10.14. d. du ∧ dv = 9(x2 − y2 )2 dx ∧ dy − 36x2y2 dy ∧ dx
ω (x, y) = p(x, y) dy or ω (x, y) = −q(x, y) dx
= 9x4 − 18x2 y2 + 9y4 + 36x2y2 dx ∧ dy
= 9(x2 + y2)2 dx ∧ dy. provides an integral.

10.16. c. The coefficients of α are constants, so d α = 0 10.19. b. The coefficients of the basic differentials are
and we can take ω = x dy + y dz + z dx. just the corresponding components of the Jacobian ma-
10.17. Since d α = (dP) dx dy = (x2 + y2 ) dz dx dy = ω , trix:
we must have dx = sin ϕ cos θ d ρ + ρ cos ϕ cos θ d ϕ − ρ sin ϕ sin θ d θ ,
2 2
dP = (x + y ) dz + Px dx + Py dy. dy = sin ϕ sin θ d ρ + ρ cos ϕ sin θ d ϕ + ρ sin ϕ cos θ d θ ,
Thus, if we set P(x, y, z) = (x2 + y2 )z, then dP has the re- dz = cos ϕ d ρ − ρ sin ϕ d ϕ .
quired form. The corresponding condition on Q(x, y, z) is
10.19. c. We make this computation in stages. First we
dQ = (x2 + y2 ) dx + Qy dy + Qz dz, have
so we can take
dy ∧ dz = −ρ sin2 ϕ sin θ d ρ ∧ d ϕ + ρ cos2 ϕ sin θ d ϕ ∧ d ρ
x3
Q(x, y, z) =+ xy2 . + ρ sin ϕ cos ϕ cos θ d θ ∧ d ρ − ρ 2 sin2 ϕ cos θ d θ ∧ d ϕ
3
= −ρ sin θ d ρ ∧ d ϕ + ρ 2 sin2 ϕ cos θ d ϕ ∧ d θ
[Note: The existence of the integral is provided by the
Poincaré lemma; see Chapter 11 of the text. The proof + ρ sin ϕ cos ϕ cos θ d θ ∧ d ρ .
there gives an algorithm for constructing the integral,]
Finally we have
10.18. We first pull back the basic differentials:
dx ∧ dy ∧ dz = ρ 2 sin ϕ sin2 θ d θ ∧ d ρ ∧ d θ
ϕ ∗ dx = coss cosht ds + sin s sinht dt,
+ ρ 2 sin3 ϕ cos2 θ d ρ ∧ d ϕ ∧ d θ
ϕ ∗ dy = − sin s sinht ds + coss cosht dt.
+ ρ 2 sin ϕ cos2 ϕ cos2 θ d ϕ ∧ d θ ∧ d ρ
It follows that
= ρ 2 sin ϕ d ρ ∧ d ϕ ∧ d θ .
ϕ ω ∗
= 21 (− cos2 s sinht cosht ds − sin s cos s sinh2 t dt)
The factor ρ 2 sin ϕ that links the two volumes elements is
+ 12 (− sin2 s sinht cosht ds + sins cos s cosh2 t dt)
identical with the Jacobian.
= 21 (− sinht cosht ds + sin s cos s dt),
10.20. a. The questions here were answered in the solu-
ϕ ∗ α = cos2 s cosh2 t ds dt + sin2 s sinh2 t ds dt tions to Exercises 5.11 (Solutions p. 49). The result
= (cos2 s + cos2 s sinh2 t + sin2 s sinh2 t) ds dt ∂ (x, y, z)
=r
2 2
= (cos s + sinh t) ds dt, ∂ (r, θ , z)
dϕ ∗ ω = 21 (− cosh2 t − sinh2 t) dt ds is what we would expect.

+ 12 (cos2 s − sin2 s) ds dt 10.20. b. We have
= 21 (1 + sinh2 t + sinh2 t + cos2 s + cos2 s − 1) ds dt dx = cos θ dr − r sin theta d θ ,
= ϕ ∗α . dy = sin θ dr + r costheta d θ ,
dx ∧ dy = r cos2 θ dr ∧ d θ − r sin2 θ d θ ∧ dr
10.19. a. This Jacobian is slightly different from the one
produced in the solution to Exercise 5.10.b (Solutions = r dr ∧ d θ ,
p. 49): dx ∧ dy ∧ dz = r dr ∧ d θ ∧ dz.

sin ϕ cos θ ρ cos ϕ cos θ −ρ sin ϕ sin θ
∂ (x, y, z) The factor r that links the two volume elements is the Ja-
= sin ϕ sin θ ρ cos ϕ sin θ ρ sin ϕ cos θ cobian, as expected.
∂ (ρ , ϕ , θ ) cos ϕ −ρ sin ϕ 0
10.21. We already determined the pullback of dy dz in
ρ cos ϕ cos θ −ρ sin ϕ sin θ
= cos ϕ the first stage of he computation of (the pullback of)
ρ cos ϕ sin θ ρ sin ϕ cos θ dx dy dz. From that we get

sin ϕ cos θ −ρ sin ϕ sin θ
+ ρ sin ϕ σ ∗ (x dy dz) = −ρ 2 sin ϕ sin θ cos θ d ρ d ϕ
sin ϕ sin θ ρ sin ϕ cos θ
= cos ϕ · ρ 2 sin ϕ cos ϕ + ρ sin ϕ · ρ sin2 ϕ + ρ 3 sin3 ϕ cos2 θ d ϕ d θ
= ρ 2 sin ϕ . + ρ 2 sin2 ϕ cos ϕ cos2 θ d θ d ρ .

97
Next we have Therefore, taking ~D as the positively oriented region for

σ ∗ (dz dx) = ρ cos2 ϕ cos θ d ρ d ϕ which −1 ≤ u ≤ 3 and 0 ≤ v ≤ 2, we have
ZZ
+ ρ sin ϕ cos ϕ sin θ d θ d ρ
xy dx dy + xy dy dz + zx dz dx
~S
+ ρ sin2 ϕ cos θ d ρ d ϕ ZZ
+ ρ sin ϕ sin θ d ϕ d θ
2 2 = − 2(u2 − v2 ) + (uv − v2) − (uv + v2) du dv
~
D
Z 2 Z 3 3
= ρ cos θ d ρ d ϕ + ρ 2 sin2 ϕ sin θ d ϕ d θ 2 −2u3 −112
= dv −2u du = 2 · = .
+ ρ sin ϕ cos ϕ sin θ d θ d ρ , 0 −1 3
−1 3
σ (y dz dx) = ρ 2 sin ϕ sin θ cos θ d ρ d ϕ
∗
+ ρ 3 sin3 ϕ sin2 θ d ϕ d θ 10.24. a. The x- and y-coordinates of a point (x, y, z) in

~S satisfy the condition x2 + y2 = a2 . Thus the point lies
+ ρ 2 sin2 ϕ cos ϕ sin2 θ d θ d ρ , at the distance a from the z-axis, so ~S is a cylinder whose
σ (dx dy) = ρ sin ϕ cos ϕ sin θ cos θ d ρ d θ
∗
axis is the z-axis. The images of the two axes are shown on
+ ρ sin2 ϕ cos2 θ d ρ d θ the cylinder where they intersect at the coordinate origin.
(See the figure below.) The u-axis is the curved one that
+ ρ sin ϕ cos ϕ sin θ cos θ d ϕ d ρ
“wraps around” the cylinder; the v-axis runs straight up
+ ρ 2 sin ϕ cos ϕ cos2 θ d ϕ d θ a generator of the cylinder. Seen from the outside, the
− ρ sin2 ϕ sin2 θ d θ d ρ (u, v)-coordinates form a counterclockwise frame on the
cylinder.
− ρ 2 sin ϕ cos ϕ sin2 θ d θ d ϕ
= ρ 2 sin ϕ cos ϕ d ϕ d θ − ρ sin2 ϕ d θ d ρ ,
σ ∗ (z dx dy) = ρ 3 sin ϕ cos2 ϕ d ϕ d θ
− ρ 2 sin2 ϕ cos ϕ d θ d ρ .
The sum collapses to
σ ∗ α = σ ∗ (x dy dz + y dz dx + z dx dy) = ρ 3 sin ϕ d ϕ d θ .
10.22. We begin with the pullbacks of the basic differen-

tials and move on to wedge products.
f∗ (dx) = −α sin u cosh v du + α cos u sinh v dv, 10.24. b. We have
f (dy) = α cos u cosh v du + α sin u sinh v dv,
∗
dx = a cos u du, dy = −a sin u du, dz = dv,
f∗ (dz) = dv,
f∗ (dy dz) = α 2 cos u cosh v du dv, and the pullback of α is −a3 sin u du ∧ dv. Consequently
f∗ (dz dx) = α 2 sin u cosh v du dv, ZZ Z h Z 2π
α = −a 3
dv sin u du = 0,
f∗ (dx dy) = −α 2 sinh v cosh v du dv, ~S 0 0
f∗ (x dy dz) = α 3 cos2 u cosh2 v du dv, because the average value of sin u on [0, 2π ] in 0.
f∗ (y dz dx) = α 3 sin2 u cosh2 v du dv,
10.24. c. Because x and y depend only on u, the differen-
f∗ (−2z dx dy) = 2α 2 v sinh v cosh v du dv, tial dx ∧ dy depends only on du. Thus dx ∧ dy = 0 · du ∧ dv
f∗ (β ) = f∗ (x dy dz + y dz dx − 2z dx dy) and
= α 2 cosh v (α cosh v + 2v sinh v) du dv. ZZ
f (x, y, z) dx ∧ dy
~S
10.23. We have (the pullbacks) ZZ
dx = du + dv, dy = du − dv, z = dv; = f (a cos u, a sin u, v) · 0 · du ∧ dv = 0
0≤u≤2π
thus 0≤v≤h
dx dy = −2 u dv, dy dz = du dv, dz dx = −du dv. for any function f (x, y, z).

10.25. a. The basic pullbacks are Consequently,
m∗ (d p) = 2x dx + 2y dy, m∗ (dr) = y dx + x dy, I Z 4π

1 1
~ =
W (C) α= dt = 2,
m∗ (dq) = dx − dy, m∗ (ds) = dx + dy. 2π ~
C 2π 0
We then have as we expect.
m∗ (dq ∧ dr) = (x + y) dx ∧ dy, 10.27. Because the identity involves Jacobians, it is es-
∗ sentially a statement about determinants of linear maps.
m (dp ∧ ds) = (2x − 2y) dx ∧ dy,
In particular, if we suppose
∗
m (pq dq ∧ dr) = (x4 − y4 ) dx ∧ dy,
B A
m∗ (qr dp ∧ ds) = (2x3 y − 4x2y2 + 2xy3) dx ∧ dy, R2 −→ R3 −→ R2
B A
m∗ (β ) = (x4 + 2x3y − 4x2y2 + 2xy3 − y4 ) dx ∧ dy. (s,t) −→ (u, v, w) −→ (y, z)
are linear maps with matrix representations

10.25. b. We integrate over the pullback:
ZZ ZZ  
b 1 B1
β= m (β )
∗ a a1 a2 a3
~S ~
U
A = = , B = b B = b2 B2  ,
Z A A1 A2 A3
1Z 1 b 3 B3
= (x4 + 2x3y − 4x2y2 + 2xy3 − y4 ) dy dx
0 0 ab aB
Z 1 AB = Ab AB ,
4 3 4x2 x 1
= x +x − + − dx
0 3 2 5
then the identity connects the determinants of certain 2 × 2
1 1 4 1 1 1 submatrices of A and B with the determinant of AB . The
= + − + − = .
5 4 9 4 5 18 identity is

10.26. From the sketch, it is clear that the winding num- a1 a2 b1 B1 a2 a3 b2 B2 a3 a1 b3 B3
+ +
ber should equal 2. A1 A2 b2 B2 A2 A3 b3 B3 A3 A1 b1 B1

ab aB
2
= .
Ab AB
1
This identity is confirmed by a direct calculation that we
do not carry out here. The right-hand side consists of 18
terms, 6 of which cancel in pairs. The left-hand side con-
-2 -1 1
tains 12 terms that match the remaining ones on the right.
-1 10.28. When n = 3, the statement of the theorem has the
following form: For any differentiable map x = (x, y, z) =
f(u, v, w) = f(u) and k-form α (x) = α (x, y, z), we have
f∗ (d α ) = d(f∗ α ).
-2
For k ≥ 3 there is nothing to prove; we therefore con-

To compute the winding number as a path integral, we
sider, in turn, k = 0, 1, and 2.
first abbreviate esin(t/2) as B(t); then
For a 0-form α = g(x, y, z) = g(x), we have
dx = (−B sint + 12 B cos(t/2) cost) dt,
f∗ α = g(x(u, v, w), y(u, v, w), z(u, v, w)).
dy = (B cost + 21 B cos(t/2) sint) dt,
−y dx, = (B2 sin2 t − 12 B cos(t/2) sint cost) dt, Therefore (making use of the approach in the text in order
2 2 1 to shorten the argument here)
x dy = (B cos t + 2 B cos(t/2) sint cost) dt.
Because x2 + y2 = B2 , the integrand pulls back simply to d(f∗ α ) = (g∗1 xu + g∗2 yu + g∗3 zu ) du
+ (g∗1 xv + g∗2 yv + g∗3 zv ) dv
−y dx + x dy
α= = dt. + (g∗1 xw + g∗2 yw + g∗3 zw ) dw.
x2 + y2

99
On the other hand, d α = g1 dx + g2 dy + g3 dz and Finally, consider the 2-form α = P dx dy. The differen-
tial is simply d α = P3 dx dy dz, and
f∗ (d α ) = g∗1 f∗ (dx) + g∗2 f∗ (dy) + g∗3f∗ (dz)
∂ (x, y, z)
= g∗1 (xu du + xv dv + xw dw) f∗ (d α ) = P3∗ .
∂ (u, v, w)
+ g∗2(yu du + yv dv + yw dw)
+ g∗3(zu du + zv dv + zw dw) In the other direction,
= (g∗1 xu + g∗2 yu + g∗3 zu ) du
∂ (x, y) ∂ (x, y) ∂ (x, y)
+ (g∗1 xv + g∗2 yv + g∗3 zv ) dv f ∗ α = P∗ du dv + dv dw + dw du .
∂ (u, v) ∂ (v, w) ∂ (w, u)
+ (g∗1 xw + g∗2 yw + g∗3 zw ) dw
The three terms of d(f∗ α ) have the coefficients
= d(f∗ α ).
∂
Next we consider a 1-form α = P dx; the forms Q dy P∗ (xu yv − xv yu ) = (P1∗ xw + P2∗ yw + P3∗ zw )(xu yv − xv yu )
∂w
and R dz can be treated similarly. We have + P∗ (xuw yv + xuyvw − xvw yu − xv yuw ),
f∗ dx = xu du + xv dv + xw dw, ∂
P∗ (xv yw − xw yv ) = (P1∗ xu + P2∗ yu + P3∗ zu )(xv yw − xw yv )
∂u
f∗ α = P∗ xu du + P∗xv dv + P∗xw dw,
+ P∗ (xvu yw + xv ywu − xwu yv − xw yvu ),
so (omitting terms that eventually cancel with each other) ∂
P∗ (xw yu − xu yw ) = (P1∗ xv + P2∗ yv + P3∗ zv )(xw yu − xu yw )
∂v
d(f∗ α ) = (P∗ xu )v dv + (P∗xu )w dw du + P∗ (xwv yu + xw yuv − xuv yw − xu ywv ).

+ (P∗ xv )u du + (P∗xv )w dw dv
The six terms involving P1∗ cancel in pairs, as do the six
+ (P∗ xw )u du + (P∗xw )v dv dw involving P2∗ . The twelve terms involving P∗ likewise can-
= (P2∗ yv + P3∗zv ) xu dv du + (P2∗yw + P3∗ zw ) xu dw du cel in pairs. The remaining terms can be viewed as the
+ (P2∗ yu + P3∗ zu ) xv du dv + (P2∗yw + P3∗ zw ) xv dw dv expansion of a 3 × 3 determinant along is bottom row:
+ (P2∗ yu + P3∗ zu ) xw du dw + (P2∗yv + P3∗zv ) xw dv dw

∗ xv xw

∗ xw xu
xu xv
P3 zu + P3 zv ∗
+ P3 zw
∗ ∂ (x, y) ∗ ∂ (z, x) yv yw yw yu yu yv
= −P2 + P3 du dv
∂ (u, v) ∂ (u, v) xu xv xw
∂ (x, y, z)
∗ ∂ (x, y) ∗ ∂ (z, x) = P3∗ yu yv yw = P3∗ = f∗ (d α ).
+ −P2 + P3 dv dw zu zv zw ∂ (u, v, w)
∂ (v, w) ∂ (v, w)

∂ (x, y) ∂ (z, x)
+ −P2∗ + P3∗ dw du. Because Q dy dz and R dz dx can be treated the same
∂ (w, u) ∂ (w, u) way, and because pullback and exterior differentiation are
linear, we can conclude that f∗ (d α ) = d(f∗ α ) for any
In the other direction, we have
2-form α .
d α = (P2 dy + P3 dz) dx = −P2 dx dy + P3 dz dx.
Using the pullback of a basic 2-form in R3 , as given on

page 443 of he text, we have f∗ (d α ) =

∂ (x, y) ∂ (x, y) ∂ (x, y)
= − P2∗ du dv + dv dw + dw du
∂ (u, v) ∂ (v, w) ∂ (w, u)

∗ ∂ (z, x) ∂ (z, x) ∂ (z, x)
+ P3 du dv + dv dw + dw du ,
∂ (u, v) ∂ (v, w) ∂ (w, u)
and this agrees with d(f∗ α ). Because Q dy and R dz can

be treated the same way as P dx, and because pullback and
exterior differentiation are linear, we can conclude that
f∗ (d α ) = d(f∗ α ) for any 1-form α .

Stokes’ Theorem
11.1. a. div V = cosy + x cosy. ~ as

11.3. d. We can parametrize C
11.1. b. div V = 0 + 0 + 0 = 0. (p + r sint, q + r cost, 2), 0 ≤ t ≤ 2π ;
1 1 1 x+y+z this gives the circle a clockwise orientation when viewed
11.1. c. div V = + + = .
yz zx xy xyz from z > 2. The circulation is
Z 2π
11.1. d. V = (a, b, c) and div V = 0.
2 · (r cost dt) = 0.
11.1. e. V = ( fx , fy , fz ) and div V = fxx + fyy + fzz . 0
11.2. a. curlF = (x − x, y − y, z − z) = (0, 0, 0). 11.3. e. The clockwise orientation of C ~ when viewed
from y < 2 corresponds to the rotation of the positive
11.2. b. curlF = (1 − 1, 1 − 1, 1 − 1) = (0, 0, 0).
z-axis to the positive x-axis. The parametrization
11.2. c. F = (a, b, c) and curl F = (0, 0, 0).
(p + r sint, 2, q + r cost), 0 ≤ t ≤ 2π ,
11.2. d. Here
provides this orientation. The circulation is therefore
F = (2ax + 2by + 2 f z, 2bx + 2cy + 2dz, 2 f x + +2dy+ 2ez) Z 2π Z 2π
(q + r cost) · (r cost dt) = (qr cost + r2 cos2 t) dt
0 0
and curlF = (2d − 2d, 2 f − 2 f , 2b − 2b) = (0, 0, 0).
= π r2 .
11.2. e. curlF = (−2z, −2x, −2y).
11.2. f. curlF = (1, 1, 1). 11.3. f. Write C ~ as the oriented sum of the four line seg-
ments C~ 1 : (0, 0, 0) → (1, 1, 0), and so forth. We can use
11.2. g. Here F = (Hx , Hy , Hz ) and the parametrizations (in which 0 ≤ t ≤ 1)
curlF = (Hzy − Hyz , Hxz − Hzx , Hyx − Hxy ) = (0, 0, 0). C~ 1 : (x, y, z) = (t,t, 0),
~ 2 : (x, y, z) = (1 − t, 1 + t,t),
C
11.3. If C~ is parametrized as (x(t), y(t), z(t)), a ≤ t ≤ b
~ and hence the ~ 3 : (x, y, z) = (−t, 2 − t, 1),
C
and F = (z, 0, 0), then F · ds = z dx on C
circulation is ~ 4 : (x, y, z) = (−1 + t, 1 − t, 1 − t).
C
I Z b
Therefore
F · ds = z(t) x′ (t) dt. Z Z 1
~
C a
F · ds = 0 · dt = 0,
~1
C 0
11.3. a. The circulation is Z Z 1
Z 2π F · ds = t · (−dt) = − 12 ,
1 ~2
C 0
3 cost · (− sint dt) = 0 Z Z 1
0
F · ds = 1 · (−dt) = −1,
because the average value of cost sint on [0, 2π ] is zero. ~3
C 0
Z Z 1
11.3. b. The circulation is zero because x′ (t) = 0. F · ds = (1 − t) · dt = 21 ,
~4
C 0
11.3. c. we have z = 0 in the (x, y)-plane, so F = 0 and
the circulation is zero. ~1 +C
so the circulation around C ~2 +C
~3 +C
~4 = C
~ is −1.
100

101
11.4. Let n = (nx , ny , nz ); because curl F = (0, 1, 0) we 11.6. a. This surface integral is already calculated in the
can write the surface integral in either of the equivalent solution to Exercise 10.5 (Solutions page 92); the value
forms ZZ ZZ obtained there is 0.
ny dA or dy dz. 11.6. b. We see div V = 0; hence the divergence theorem
~R ~R
In every case below, the value of the surface integral of gives ZZ ZZZ
curlF over ~R agrees with the values of the path integral V · n dA = div V dV = 0.
∂ ~P ~P
giving the circulation of F around ∂ ~R.
This value agrees with the value of the surface integral
11.4. a. Here C ~ lies in the plane z = 1 x, which has the
3 computed directly.
normal vector N = − 13 , 0, 1 . Hence ny = 0 and the sur-
11.7. View the given surface integral as the integral of
face integral is 0.
the vector field V = 31 n; then
11.4. b. Here ~R has the oriented unit normal n = x y z 1 1 1
(−1, 0, 0); hence curl F·n = 0 and the surface integral is 0. div V = div , , = + + = 1.
3 3 3 3 3 3
11.4. c. Here n = (0, 0, 1), so curlF·n = 0 and the surface
integral is 0. If R is the 3-dimensional region enclosed by S, then the
divergence theorem says
11.4. d. Here n = (−1, 0, 0) as a result of the fact that ~R ZZ ZZ ZZZ
has clockwise orientation when viewed from z > 2; hence 1
3 x · n dA = V · n dA = 1 · dV = vol(R);
we still have curlF · n = 0 and the surface integral is 0. S S R
11.4. e. The properly oriented normal to the plane y = 2 this is, by definition, vol(S).
is n = (0, 1, 0), so the integral is 11.8. We have d α = (Px + Qy + Rz ) dx dy dz, and we take
ZZ
∂ ~B to be the union of the four faces Sx=0 , Sy=0 , Sz=0 , and
1 · dA = area(~R) = π r2 . S1 = Sx+y+z=1 , each with the outward orientation. (See
~R
below and the illustrations on page 455 of the text.)
11.4. f. A properly oriented normal to ~R is given by z
N = (1, 1, 0) × (−1, 1, 1) = (1, −1, 2);
√
(−1, 0, 0)
√ area(R) = kNk = 6. Therefore, because
furthermore,
ny = −1/ 6, the surface integral is
ZZ (0, −1, 0) Sx=0
−1 √
ny dA = √ · 6 = −1. Sy=0 y
~R 6 Sz=0
x y z
11.5. a. In this special case, n = , , , so x
R R R (0, 0, −1)
x2 + y2 + z2 R2 As with the proof of the divergence theorem for a ball

V·n = = = R. (Theorem 11.2, text pp. 453–454), we can treat the pieces
R R
Consequently, involving P, Q, and R separately. Thus we restrict our-
ZZ selves to showing
V · n dA = R · area(~S) = R · 4π R2 = 4π R3. ZZZ ZZ
~S Px dx dy dz = P dy dz.
~B ∂ ~B
11.5. b. Let ~B be the positively oriented ball of radius R
Consider the surface integral over ∂ ~B. On Sy=0 we have
centered at the origin. The divergence theorem says that
ZZ ZZZ dy = 0, and on Sz=0 we have dz = 0; therefore
ZZ ZZ ZZ
V · n dA = div V dV.
~S ~B P dy dz = P dy dz + P dy dz.
∂ ~B Sx=0 S1
In this case, div V = 1 + 1 + 1 = 3 so the divergence theo-
rem implies We can represent the oriented region S1 as an oriented sur-
ZZ ZZZ face patch (Def. 10.2, text p. 392) by the parametrization
V · n dA = 3 dV = 3 · vol(~B) = 3 · 34 π R3 = 4π R3. 
~S ~B 
x = 1 − u − v,
The two ways of obtaining the value of the surface integral f : y = u, ~T : 0 ≤ u ≤ 1,

 0 ≤ v ≤ 1 − u.
agree. z = v;

102 SOLUTIONS: CHAPTER 11. STOKES’ THEOREM
The parametrization pulls the surface integral back to a 11.9. & 10. These exercises are superfluous; they repeat
double integral and, ultimately, to an iterated integral, as Exercises 11.5 and 11.6.a.
follows.
ZZ ZZ 11.11. Let {b1 , b2 , . . . , bn } be a basis for Rn , and let
P dy dz = P(1 − u, v, u, v) du ∧ dv
S=1 ~T v = v1 b1 + · · · + vn bn , w = w1 b1 + · · · + wn bn .
Z 1 Z 1−u
= P(1 − u − v, u, v) dv du.
Assuming that A(bi , b j ) = B(bi , b j ) for every 1 ≤ i, j ≤ n,
0 0
the bilinearity of A and B allows us to write, for arbitrary
The region Sx=0 is also oriented; however, if we modify vectors v and w,
the map f by changing x = 1 − u − v to x = 0, the orienta- ! !
tions do not match. This problem is resolved if we work
instead with the oppositely oriented region −Sx=0 and its A(v, w) = A ∑ vi bi , ∑ w j b j = ∑ vi A bi , ∑ w j b j
i j i j
parametrization
 = ∑ vi w j A(bi , b j ) = ∑ vi w j B(bi , b j )

 x = 0, i, j i, j
0 ≤ u ≤ 1, ! !
bf : y = u, ~T :

 0 ≤ v ≤ 1 − u. = ∑ vi B bi , ∑ w j b j = B ∑ vi bi , ∑ w j b j
z = v; i j i j
Then = B(v, w).

ZZ ZZ
P dy dz = − P dy dz
Sx=0 −Sx=0 11.12. a. Because d Φ = Φx dx + Φy dy by definition, we
ZZ have
=− P(0, u, v) du ∧ dv −y x
~T Φx = 2 and Φy = 2 .
Z
1 Z 1−u
x + y2 x + y2
=− P(0, u, v) dv du. Therefore,
0 0
In summary, we express the surface integral over all of ∂ ~B ϕ ′ (t) = Φx (cost, sint) · (cost)′ + Φy (cost, sint) · (sint)′
as a single iterated integral: − sint cost
= · (− sint) + · (cost) = 1,
ZZ Z 1 Z 1−u 1 1
P dy dz = P(1−u−v, u, v)−P(0, u, v) dvdu.
0 0 for all t.
∂ ~B
11.12. b. From the general fact
Now consider the triple integral of Px over ~B. We can Z t
write the sloping face S1 as the graph of x = 1 − y − z over ϕ (t) − ϕ (0) = ϕ ′ (s) ds,
the triangle 0 ≤ y ≤ 1, 0 ≤ z ≤ 1 − y, and thus we can 0
define ~B by the inequalities
we have
0 ≤ y ≤ 1, Zt
0 ≤ z ≤ 1 − y, ϕ (t) = ϕ (0) + 1 · ds = Φ (1, 0) + t.
0
0 ≤ x ≤ 1 − y − z.
But then
Consequently
ZZZ Z 1 Z 1−y Z 1−y−z 2π + Φ (1, 0) = ϕ (2π ) = Φ (cos 2π , sin 2π ) = Φ (1, 0),
∂P
Px dx dy dz = dx dz dy
~B 0 0 0 ∂x
Z 1 Z 1−y x=1−y−z from which we get the contradiction 2π = 0.

= P(x, y, z) dz dy
11.13. a. From the solution to Exercise 5.25 (Solutions
0 0 x=0
Z 1 Z 1−y p. 58), we can see that ds(1,0) : R4 → R4 is the identity
= P(1 − y − z, y, z) − P(0, y, z) dz dy. map. (Note: in Exercise 5.25, s is written as σ .) If we let
0 0 ei , i = 0, 1, 2, 3 be the standard basis in (r,t1 ,t2 ,t3 )-space,
With the substitutions y ↔ u, z ↔ v, the triple integral and then the vector
the surface integral become equal. ds(1,0) (e0 )

103
is the outward-pointing normal to S3 at x0 = (1, 0, 0, 0), The integral is computed using the pullback:
while each of the vectors ZZZ ZZZ
β3 = s∗ (β3 )
ds(1,0) (ei ), i = 1, 2, 3 S3 s−1 (S3 )
Z π Z π /2 Z π /2
is tangent to S3
at x0 . Therefore s defines the positive = dt1 cost2 dt2 cos2 t3 dt3
−π −π /2 −π /2
orientation on S3 .
π
11.13. b. In Cartesian coordinates, the 3-form β3 is = 2π · 2 · = 2π 2 .
2
x1 dx2 dx3 dx4 − x2 dx1 dx3 dx4 + x3 dx1 dx2 dx4
β3 = 11.14. By Theorem 11.28 (text p. 508), the vector field F
(x21 + x22 + x23 + x24 )2 will have a scalar potential (at least locally) if curlF = 0.
x4 dx1 dx2 dx3 We have
− 2 .
(x1 + x22 + x23 + x24 )2
curlF = 0, 0, cosxy − xy sin xy − (cosxy − xy sin xy)
On S3 the denominator reduces to 1. To determine the = (0, 0, 0).
pullback s∗ (β3 ), we use the abbreviations si = sinti , ci =
costi introduced in the solution to Exercise 5.25. We can A bit of experimenting shows that ϕ = x2 + sin xy + 2z3 /3
then read off from the entries of dσ in that solution the is a scalar potential for F; that is, grad ϕ = F.
following differentials.
11.15. We know F can have a vector potential (at least
dx1 = −s1 c2 c3 dt1 −c1 s2 c3 dt2 − c1 c2 s3 dt3 , locally) only if div F = 0; but
dx2 = c1 c2 c3 dt1 −s1 s2 c3 dt2 − s1 c2 s3 dt3 , div F = 2 − y2 sin xy − x2 sin xy + 4z 6= 0,
dx3 = c2 c3 dt2 − s2 s3 dt3 ,
dx4 = c3 dt3 . so there is no vector potential.
11.16. a. Because curlV = 0, we know V has a scalar
The four terms in s∗ (β3 ) are computed as follows. potential. We can take ϕ = xy + yz + zx as the potential.
dx3 dx4 = c2 c23 dt2 dt3 , 11.16. b. Because div V = 0, we know V has a vector po-
dx2 dx3 dx4 = c1 c22 c33 dt1 dt2 dt3 , tential W = (P, Q, R). There are many possible values for
P, Q, and R. We can interpret curlW = V as imposing the
x1 dx2 dx3 dx4 = c21 c32 c43 dt1 dt2 dt3 , conditions
dx1 dx3 dx4 = −s1 c22 c33 dt1 dt2 dt3 ,
Ry = y + z, Pz = z + x, Qx = x + y,
−x2 dx1 dx3 dx4 = s21 c32 c43 dt1 dt2 dt3 ,
Qz = 0, Rx = 0, Py = 0.
dx2 dx4 = c1 c2 c23 dt1 dt3 − s1 s2 c23 dt2 dt3 ,
dx1 dx2 dx4 = s2 c2 c33 dt1 dt2 dt3 , One possibility for W is
2
x3 dx1 dx2 dx4 = s22 c2 c43 dt1 dt2 dt3 , z x2 y2
W = (P, Q, R) = + zx, + xy, + yz .
dx2 dx3 = c1 c22 c23 dt1 dt2 + s1 s3 c3 dt2 dt3 2 2 2
+ c1 s2 c2 s3 c3 dt3 dt1 ,
11.16. c. If fx = −P, then grad f = (−P, fy , fz ) and
dx1 dx2 dx3 = −c2 s3 c23 dt1 dt2 dt3 ,
W + grad f = (P, Q, R) + (−P, fy , fz ) = (0, Q + fy , R + fz ).
−x4 dx1 dx2 dx3 = c2 s23 c23 dt1 dt2 dt3 .
Therefore Following the suggestion, we want

∂f z2
s∗ (β3 ) = c21 c32 c43 + s21 c32 c43 + s22 c2 c43 + c2 s23 c23 dt1 dt2 dt3 = −P = − − zx,
∂x 2
= c22 c2 c43 + s22 c2 c43 + c2 s23 c23 dt1 dt2 dt3
so we can take f (x, y, z) = −xz2 /2 − zx2/2. The new vec-
= c2 c23 c23 + c2 s23 c23 dt1 dt2 dt3
tor potential for V is therefore
= c2 c23 dt1 dt2 dt3 , 2
x y2 − x2
as required. W + grad f = 0, + xy, + z(y − x) .
2 2

11.17. Let V = (yz, zx, xy); then div V = 0 (cf. the so- If d ω = 0, then the coefficients of each of the four basic
lution to Exercise 11.2.b, above). Therefore, by Theo- 3-forms above must be zero. That is,
rem 11.28 (text p. 508), there is a vector field P = (P, Q, R)
Au + Bx − Ey = 0 Cx + Du + Ev = 0,
for which V = curlP. In other words, there are three func-
tions P, Q, and R for which Bv + Cy − Fu = 0, Dy + Av + Fx = 0.
∂R ∂Q ∂P ∂R ∂Q ∂P 11.20. b. If α = P dx + Q dy + R du + S dv, then

− = yz, − = zx, − = xy.
∂y ∂z ∂z ∂x ∂x ∂y d α = (Py dy + Pu du + Pv dv) dx
To find these functions, let us suppose in addition that + (Qx dx + Qu du + Qv dv) dy
Qz = Rx = Py = 0, reducing the given partial differential + (Rx dx + Ry dy + Rv dv) du
equations to
+ (Sx dx + Sy dy + Su du) dv
∂R ∂P ∂Q = (Qx − Py ) dx dy + (Ry − Qu ) dy du
= yz, = zx, = xy.
∂y ∂z ∂x + (Su − Rv ) du dv + (Pv − Sx ) dv dx
One possible solution is then + (Rx − Pu ) dx du + (Sy − Qv ) dy dv.
If d α = ω , then the coefficients of each of the six basic
z2 x x2 y y2 z
P(x, y, z) = , Q(x, y, z) = , R(x, y, z) = . 2-forms in d α and ω must be equal:
2 2 2
Qx − Py = A, Ry − Qu = B, Su − Rv = C,
Many other solutions exist. Our argument involved diver-
gence and curl, but not gradient. Pv − Sx = D, Rx − Pu = E, Sy − Qv = F.
11.18. Let V = (−y + x, x + y, z); then, if there is a solu- 11.21. a. We have d ω = (Az + Bx + Cy ) dx dy dz; there-
tion to the given partial differential equations, it is a func- fore, if ω is closed (i.e., d ω = 0) then A, B, and C satisfy
tion f (x, y, z) for which grad f = V. In that case, Theorem the single partial differential equation
11.26 (text p. 508) then implies
∂ A ∂ B ∂C
+ + = 0.
curlV = curlgrad f = 0. ∂z ∂x ∂y
However, curl V = (0, 0, 2) 6= 0, contradicting our assump- 11.21. b. We have

tion: thus the partial differential equations have no solu- d α = (Qx − Py) dx dy + (Ry − Qz ) dy dz + (Pz − Rx ) dz dx;
tion.
therefore, if d α = ω , then the coefficients of d α must
11.19. Let F = (x, y, z); if there are solutions P, Q, R to equal the corresponding coefficients of ω . This means
the given partial differential equations, they constitute a that P, Q, and R, must satisfy the partial differential equa-
vector field V = (P, Q, R) for which curlV = F. In that tions
case, Theorem 11.26 then implies
∂Q ∂P ∂R ∂Q ∂P ∂R
− = A, − = B, − = C.
div curlV = div F = 0. ∂ x ∂ y ∂ y ∂ z ∂z ∂x
However, div F = 1 + 1 + 1 = 3 6= 0, contradicting our as- 11.22. a. We have

sumption: thus the partial differential equations have no d β = (Qx − Py ) dx dy + (Ry − Qz ) dy dz + (Pz − Rx ) dz dx;
solution.
therefore, if β is closed (d β = 0), then the coefficients of
11.20. a. The exterior derivative of the given 2-form is the basic 2-forms in d β must all be zero:
d ω = (Au du + Av dv) dx dy + (Bx dx + Bv dv) dy du ∂Q ∂P ∂R ∂Q ∂P ∂R
= , = , = .
+ (Cx dx + Cy dy) du dv + (Dy dy + Du du) dv dx ∂x ∂y ∂y ∂z ∂z ∂x
+ (Ey dy + Ev dv) dx du + (Fx dx + Fu du) dy dv 11.22. b. Because d f = fx dx + fy dy + fz dz, the func-
= (Au + Bx − Ey ) dx dy du tion f (x, y, z) must satisfy the following partial differential
+ (Bv + Cy − Fu ) dy du dv equations:
+ (Cx + Du + Ev ) du dv dx ∂f ∂f ∂f
= P, = Q, = R.
+ (Dy + Av + Fx ) dv dx dy. ∂x ∂y ∂z

105
11.23. The integrability conditions defined by d ω = 0 in Therefore, the equation d ω = 0 defines the n(n − 1)/2
this exercise are the same as those defined by d β = 0 in integrability conditions
the previous exercise: ∂ Pj ∂ Pi
∂Q ∂P ∂R ∂Q ∂P ∂R = , 1 ≤ i < j ≤ n.
= , = , = . ∂ x i ∂ xj
∂x ∂y ∂y ∂z ∂z ∂x
To obtain the conditions on Φ , note that z appears only in 11.24. b. We define Φ (x1 , . . . , xn ) as a sum of n integrals
the third term on the right in the definition of Φ . There- using aZ fixed point (a1 , . . . , an ):Z
x1 x2
fore (by the fundamental theorem of calculus), we find
Φ= P1 (t, a2 , . . . , an ) dt + P2 (x1 ,t, a3 , . . . , an ) dt
immediately that a1 a2
Z xn
Φz = R(x, y, z). Pn (x1 , x2 , . . . , xn−1 ,t) dt
+ ···+
an
Next, because y appears only in the second and third
terms, we have 11.24. c. Because xn appears in only the last integral
Z z defining Φ , the fundamental theorem of calculus implies
Φy = Q(x, y, c) + Ry (x, y,t) dt
Zc Φxn = Pn (x1 , x2 , . . . , xn−1 , xn ).
z
= Q(x, y, c) + Qz (x, y,t) dt We now establish ΦxI = PI (x1 , . . . , xn ) for an arbitrary
c
t=z I = 1, 2, . . . , n − 1. Because xI appears only in the last

= Q(x, y, c) + Q(x, y,t) n − (I + 1) integrals in the sum that defines Φ , we can
t=c write
= Q(x, y, c) + Q(x, y, z) − Q(x, y, c) = Q(x, y, z).
ΦxI = PI (x1 , . . . , xI , aI+1 , . . . , an )
We used one of the integrability conditions to replace Ry Z xI+1
∂ PI+1
by Qz in the integral (and then used the fundamental the- + (x1 , . . . , xI ,t, aI+2 , . . . , an ) dt
aI+1 ∂ xI
orem of calculus). Z xn
Finally, we have ∂ Pn
+ ···+ (x1 , . . . , xn−1 ,t) dt
Z y Z z a n ∂ xI
Φx = P(x, b, c) + Qx (x,t, c) dt + Rx (x, y,t) dt Now use the n − I integrability conditions
Zb y Z cz
= P(x, b, c) + Py (x,t, c) dt + Pz (x, y, c) dt ∂ PI+1 ∂ PI ∂ Pn ∂ PI
= , ... , =
b
y c
z ∂ xI ∂ xI+1 ∂ xI ∂ xn

= P(x, b, c) + P(x,t, c) + P(x, y,t) to make substitutions in the integrals. The result is
b c
ΦxI = PI (x1 , . . . , xI , aI+1 , . . . , an )
= P(x, b, c) + P(x, y, c) − P(x, b, c) Z xI+1
∂ PI
+ P(x, y, z) − P(x, y, c) = P(x, y, z). + (x1 , . . . , xI ,t, aI+2 , . . . , an ) dt
aI+1 ∂ xI+1
We used the remaining two integrability conditions to re- Z xn
∂ PI
place Qx by Py in one integral and Rx by Pz in the other. + ···+ (x1 , . . . , xn−1 ,t) dt
an ∂ xn
The preceding computations demonstrate that d Φ = ω .
= PI (x1 , . . . , xI , aI+1 , . . . , an )
11.24. a. If we write ω = ∑ Pi dxi , then xI+1

I
+ PI (x1 , . . . , xI ,t, aI+2 , . . . , an )
∂ Pi ∂ Pi ∂ Pi aI+1
dω = ∑ dx j dxi = ∑ dx j dxi + ∑ dx j dxi . xn
j6=i ∂ x j j<i ∂ x j j>i ∂ x j

+ · · · + PI (x1 , . . . , xn−1 ,t)
In the first sum on the right, exchange the dummy indices: an
j ↔ i. In the second sum, make the substitution dx j dxi = = PI (x1 , . . . , xI , aI+1 , aI+2 , . . . , an )
−dxi dx j . With these changes, + PI (x1 , . . . , xI , xI+1 , aI+2 , . . . , an )
∂ Pj ∂ Pi − PI (x1 , . . . , xI , aI+1 , aI+2 , . . . , an ) + · · ·
dω = ∑ dxi dx j − ∑ dxi dx j
i< j ∂ x i ∂
i< j x j + PI (x1 , . . . , xn ) − PI (x1 , . . . , xn−1 , an )

∂ Pj ∂ Pi = PI (x1 , . . . , xn ).
=∑ − dxi dx j .
i< j ∂ xi ∂xj Because ΦxI = PI , I = 1, 2, . . . , n, we have d Φ = ω .

11.25. Each basic k-form in n variables is a product of is true for n = k + 1 (the “base” case), assume it is true
k distinct factors dxi , wherei = 1, 2, . . . , n. There are nk for all n with k + 1 < n < N, and then prove it is true for
such products and hence nk terms in any k-form ω in n n = N.
variables. The base case is just the statement
For ω to be locally exact, we must have d ω = 0. Be-
k k+1
cause d ω is a (k + 1)-form in n variables, the equation = ,
k k+1
d ω = 0 requires each of the k+1 n
terms of d ω to be zero.
n

These k+1 equations are the integrability conditions. and this is true because both sides equal 1. For the induc-
To establish that ω is locally exact we must find a tion case, we have
(k − 1)-form α for which ω = d α . The coefficients of N N−1
m−1 m−1 N −1
the terms of α are k−1 ∑ = ∑
n
(unknown) functions, and the +
terms of d α involve their partial derivatives. The equa- m=k+1 k m=k+1 k k

tion d α = ω splits into nk equations for the individual N −1 N −1 N
= + = .
terms; these are the nk partial differential equations for k+1 k k+1
n
the k−1 unknown functions that define α .
We have used the induction hypothesis to evaluate the ex-
n−1 n−1 tended sum on the right, and then we used Exercise 11.26
11.26. We have + =
k−1 k−2 to add the the resulting two terms. We have thus proved
the induction case.
(n − 1)! (n − 1)!
= + 11.28. a. We know ∇ × F = (Cy − Bz , Az − Cx , Bx − Ay );
(k − 1)!(n − 1 − (k − 1))! (k − 2)!(n − 1 − (k − 2))!
(n − 1)! (n − 1)! thus, adding and subtracting Axx , Byy and Czz , we get
= +
(k − 1)!(n − k)! (k − 2)!(n − k + 1))! ∇ × (∇ × F)
(n − 1)!(n − k + 1) (n − 1)!(k − 1) = Bxy − Ayy − Azz + Cxz ,Cyz − Bzz − Bxx + Ayx ,
= +
(k − 1)!(n − k + 1)! (k − 1)!(n − k + 1))!
Azx − Cxx − Cyy + Bzy
(n − 1)!(n − k + 1 + k − 1) (n − 1)! × n
= = = (Ax + By + Cz )x , (Ax + By + Cz )y , (Ax + By + Cz )z
(k − 1)!(n − k + 1)! (k − 1)!(n − k + 1)!
− Axx + Ayy + Azz, Bxx + Byy + Bzz ,Cxx + Cyy + Czz
n! n
= = . = ∇(∇ · F) − (∇ · ∇)F.
(k − 1)!(n − (k − 1))! k−1
11.27. a. The totality of integrability conditions are in- 11.28. b. We identify U and V with ∇ and W with F.
dexed by the multi-index Because V represents a differential operator and U · W a
function, we should put V to the left of U · W in the given
L = (i1 , . . . , ik+1 ), 1 ≤ i1 < · · · < ik+1 ≤ n.
equation; thus
n
There are k+1 such multi-indices, and hence that many
integrability conditions. But the discussion on page 503 of U × (V × W) = V(U · W) − (U · V)W.
the text shows that the multi-indices actually used in the With the identifications, this equation becomes
induction step from n − 1 to n variables are of the form
L = I ∗ , n where ∇ × (∇ × F) = ∇(∇ · F) − (∇ · ∇)F,
I ∗ = (i1 , . . . , ik ), 1 ≤ i1 < · · · < ik ≤ n − 1. curl(curlF) = grad(div F) − div(grad F).

There are n−1 k such multi-indices, and they all have n 11.29. According to the divergence theorem,
as final
index. Now switch from n to m: then there are ZZ ZZZ
m−1
k integrability conditions used in the induction step n · f ∇g dA = ∇ · f ∇g dV.
from m − 1 variables to m variables. Moreover, the multi- ∂R R
indices used by different values of m all differ in at least If we set ∇g = (gx , gy , gz ), then
their last place.
∇ · f ∇g = ( f gx )x + ( f gy )y + f gz )z
11.27. b. We establish this identity by induction on n.
= fx gx + fy gy + fz gz + f (gxx + gyy + gzz)
That is, we show
n = ∇f · ∇g + f (∇· ∇g).
m−1 n
∑ k
=
k+1 This proves that the given equation is true.
m=k+1

Instructor's Solutions Manual To Advanced Calculus - A Geometric View ( James J. Callahan )

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Instructor's Solutions Manual To Advanced Calculus - A Geometric View ( James J. Callahan )

Uploaded by

Copyright:

Available Formats

Solutions:

DVI file created at 14:27, 26 April 2011

DVI file created at 14:27, 26 April 2011

DVI file created at 14:27, 26 April 2011

2 Geometry of Linear Maps 9

9 Evaluating Double Integrals 80

11 Stokes’ Theorem 100

DVI file created at 14:27, 26 April 2011

DVI file created at 14:27, 26 April 2011

1.1. The integrals are “improper,”, because there is an Therefore,

x = ±R. The result is

DVI file created at 14:27, 26 April 2011

has the same sign as s, we must choose the plus sign in

1.6. d. Here ϕ ′ = cosh s > 0 for all s, so ϕ is invertible −1 −1 1 1

DVI file created at 14:27, 26 April 2011

sin θ s 1.9. c. Calculator values to nine decimal places are

DVI file created at 14:27, 26 April 2011

√ x2 = ϕ (s) = tan s, then ϕ (s) = sec s, so

DVI file created at 14:27, 26 April 2011

DVI file created at 14:27, 26 April 2011

0.5 1.0 1.5 2.0 2.5 3.0 3.5

This is negative if c − γ < 0, i.e., if c < γ . Negative work 853/2 − 133/2

DVI file created at 14:27, 26 April 2011

1.29. c. Here Therefore t = σ (s) = tan(s/2). To determine x(σ (s)), we

16t 2 + 4 − 8t 2 + 4t 4 4 1.32. We use Definition 1.6 and Theorem 1.5 (text

DVI file created at 14:27, 26 April 2011

1.37. a. To determine critical points and inflections, we -4 -2 0 2 4

need 1.38. b. The graphs are just horizontal translates of one

1.39. The procedure here is similar to the one used to

DVI file created at 14:27, 26 April 2011

Geometry of Linear Maps

DVI file created at 14:27, 26 April 2011

DVI file created at 14:27, 26 April 2011

DVI file created at 14:27, 26 April 2011

λ2 u2 = M(u2 ) = M(ku1 ) = kM(u1 ) = kλ1 u1 = λ1 u2 .

Because λ2 6= λ1 , we have u2 = 0, a contradiction.

DVI file created at 14:27, 26 April 2011

then we have 2.12. b. By definition, the eigenvectors of M for eigen-

In other words, λ u = (M − λ I)v = k(M − λ I)u = k · 0.

DVI file created at 14:27, 26 April 2011

DVI file created at 14:27, 26 April 2011

DVI file created at 14:27, 26 April 2011

x × y · (ax + by) = 0. 2.24. b. By part (a) and property 1,

DVI file created at 14:27, 26 April 2011

2.24. e. By definition, 2.27. Following the suggestion, we write

DVI file created at 14:27, 26 April 2011

DVI file created at 14:27, 26 April 2011

detM2 = v31 v42 v13 v24 − v41 v32 v13 v24

The last expression is not, formally at least, det A. How- πI ( j) = π ( j) for j 6= I;

DVI file created at 14:27, 26 April 2011

DVI file created at 14:27, 26 April 2011

DVI file created at 14:27, 26 April 2011

2.43. b. The solution set is the graph of the function

using the linearity of L. Hence v1 + v2 is in V because it

cv = (cu, cL(u)) = (cu, L(cu)),

again using the linearity of L. Hence cv is in V because it

Hence c1 u1 + · · · + cn un = 0, but the vectors u1 , . . . , un

DVI file created at 14:27, 26 April 2011

DVI file created at 14:27, 26 April 2011

Thus (n + 1)cn = bn + abn−1 + · + an, or But

have as we had to show.

We can write this as

DVI file created at 14:27, 26 April 2011