The Art of Problem Solving Vol 2 Solutions

PHILLIPS ACADEMY
1 1
III 11
3 1 867 00 712
111;
-.####♦
igp i f?t *
*
# PHILLIPS * ACADEMY
♦
♦
♦
♦
¥
#
♦
# OUVER*WENDELL*HOLMES
##♦# # #
t LIBRARY
♦
*& §
**### ####%
1
Gift of
Anonymous Donor
the
ART
of
PROBLEM
SOLVING
Volume 2:
and BEYOND
Solutions
Richard Rusczyk
Sandor Lehoczky
Copyright © 1994,1996, 2003, 2004, 2006 Sandor Lehoczky and Richard Rusczyk.
All Rights Reserved.
Reproduction of any part of this book without expressed permission of the authors is strictly forbid¬
den.
For use outside the classroom of the problems contained herein, permission must be acquired from
the cited sources.
ISBN #: 978-0-9773045-9-2
Published by: AoPS Incorporated

P.O. Box 2185
Alpine, CA 91903-2185
(619) 659-1612
books@artofproblemsolving.com
Visit the Art of Problem Solving website at http://www.artofproblemsolving. com
Printed in the United States of America.

Seventh Edition; printed in 2010.
Editor: David Patrick
Cover image designed by Vanessa Rusczyk using KaleidoTile software.
Cover Image: "Niagara Falls" by Vanessa Rusczyk.
This book was produced using the UTpX document processing system.
Chapter 1
Logarithms
Solutions to Exercises
1-1 Let logrt b = x, logn c = y, and logc b - z. For Property 3, we must prove x - y = logfl b/c.
Writing our first two equations in exponential notation, we have b/c = ax~y, so that logfl b/c - x - y
as required. For Property 5, we wish to show that x/y - z. Once again writing the three equations in
exponential notation, we have ax = b, ay = c and cz = b. Raising the second equation to the z power,
we have ayz = cz - b = ax. Hence, yz - x, or x/y - z as desired. Finally, for the final property, let
loga„ bn = zv, so anw - bn. Raising this to the 1/n power, we have aw - b, so that w - x (since ax = b
also) and Property 6 is proven.
1-2 Let a = d in Property 4; then we have
(logfl b)(logc a) = (logfl fl)(logc b) = logc b,
proving the chain rule for logarithms.

1-3 First, we write all expressions in the simplest base possible. Since log3 4 = log3 22 = 2 log3 2,
we can write log3 2 = x/2. For the first problem, we note that 10 = 5-2, and we can write
x 1 x 1
63 63 63 2 log5 3 2 y
For the second problem, note that 6/5 = 1.2 and 3(2) = 6, so
log3 12 = log3(6/5) = loS3 3 + lo§3 2 “ lo§35 = 1 + \ ~

^ y
1-4 Writing Property 5 backwards, we have log23 = (log10 3)/(log10 2). My calculator can do
everything on the right side of this, so my calculator can figure out log2 3.
1-5 Let logx y = z, so xz = y. Hence,
xlo8x 3/ = xz = y.
< 3 >
4 > CHAPTER 1. LOGARITHMS
Solutions to Problems
1. Applying the chain rule repeatedly, we have (log2 3)(log3 4) = log2 4, (log2 4)(log4 5) = log2 5, and
so on, until finally we are just left with log2 8 = 3.
2. Writing the given information in the simplest base possible, we find log 6 = a/2 and log 5 = b/3
(since log 36 = log62 = 2 log 6 = a, for example). Thus,
log(l/12) = log(l/6) + log(l/2) = - log6- log2 = - (log 10 - log5) = -- + - - 1.
Make sure you see why log 1/2 = - log 2.

3. If the graphs intersect, then 2 log x = log 2x, or log x2 = log 2x, from which we find x2 - 2x and
x - 2 or x = 0. The expressions in the given equation are clearly not defined for x = 0, so we discard
that solution and we are left with only 1 point at which the graphs intersect.
4. We've got messy expressions in our exponents. One sure way to get them out of the exponents
is to take logarithms. Taking the logarithm base 10, we find
logxlos* = log —
(logx)(logx) = logx3-log 100

(logx)2 = 31ogx-2.
If we let y = logx, we have a quadratic in y, for which the solutions are y - 2 and y — 1. Solving for
x (x = 10^), we find the solutions x = 100 and x = 10.
5. From Property 5, p = logfl(logi, a). Hence, aP = \ogb a.
6. From the given information, we note that 210 > 103, so that log210 > loglO3, or 10 log 2 > 3.
We also find that 104 > 213, and taking the logarithm of this base 10, we find 4 > 13 log 2. Hence, the
best we can do is 3/10 < log 2 < 4/13.
7. Since 1/ logfl b = log^ a, the sum is logx 3 + logz 4 + logx 5 = logx 60.
8. Taking the logarithm base 10 of a number will tell us how many digits the number has. For
example, a number y with 25 digits, when written in scientific notation, is of the form x • 1024, where
x is some number from 1 to 10. The logarithm of this number is 24 + logx. Since x is between 1 and
10, 0 < logx < 1, so that [logyj = 24. Similarly, the integer part of the logarithm of any positive
integer is 1 less than the number of digits it has (remember, y above has 25 digits). (Can you make
this proof rigorous? Try using n instead of 25.) Thus, we take the logarithm of 544 in base 10, yielding
logS44 = 44 log 5 = 44(log 10 - log 2) = 44(1 - 0.3010) = 30.756,
so 5U has 30 + 1 = 31 digits.
9. First, we use the simplest bases possible. Since 8P = 3, we have 23P = 3 and log-, 3 = 3P. Since
there are no factors of 3 in our desired expression, but 3's occur in both our known expressions, we
want to introduce 3's into log10 5 somehow. Looking over our properties, we see that division is a
good way to do this. Let's try it:
lo 5 - l0g3 5 - Q Q
°8l° log3 10 loS3 5 + loS3 2 ~ Q+ !/(3P) ’
the ART of PROBLEM SOLVING: Volume 2 < 5
Simplifying this, we find log10 5 = (3PQ)/(1 + 3PQ).

10. Set all the given logarithms equal to t. Writing these as exponential equations, we have
9l = p, 12* = q, and 16f = p + q. Taking the quotient of the first two, we have 12* / 9* = (4/3)f = q/p.
Now we note that 16* / 9* = [(4/3)f]2 = (q/p)2 - q2/p2. Going back to the original equations, we find
16* - p + q = 12* + 9*. Dividing this by 9*, we have 16* /9* - 12* /9* + 1, or
since 16/9 = (4/3)2. Letting q/p = x, this equation becomes x2 - x + 1. Solving for x (and recalling
that we want the positive value since p and q are both positive), we find x = (1 + V5)/2.
11. Letting the two logarithmic expressions equal x and writing them in exponential notation,
we have (4n)x - 40 y/3 and (3n)x = 45. Dividing these gives (4/3)* = 8 V3/9, from which we find
x = 3/2. Thus, (3n)3/2 = 45, (3n)3 = 452, and n3 = 452/27 = 75.
12. Attack this just like the previous similar problem. Writing the given logarithms, all set equal to
x, in exponential form, we have 9X = a, 15x = b, and 25* = a + 2b. As before, b/a = (15/9)* = (5/3)* and
b2/a2 = [(5/3)2]* = 25*/9*. Combining the exponential equations, we find 25* = a + 2b = 9* + 2(15*).
Dividing by 9* and letting y = (5/3)* = b/a, we have y2 = 1 + 2y, which has solutions y - 1 ± V2.
Finally, we find b/a = 1 + V2 since a and b are both positive.
13. From the given information, we have a - log60 3 and b — log60 5. Using these and many of
our logarithmic properties, we have
12[(l-fl-fc)/2(l-fc)] 22[(l-(log60 3+log60 5))/2(l-log60 5)]

12(log60 60—ioggg 15)/(21og60 60-2 log60 5)
12(loS604)/(log60 602 /52)
12(log6o4)/(log60 144)
12(1°gl444) = 12(1°§122) zz 2.
Make sure you see how we have applied our properties at each step.
Chapter 2
Not Just For Right Triangles
2-1 30° is in quadrant I. Since 700° = 700° - 360° = 340°, it is in quadrant IV. 5n/3 is in quadrant
IV as well. Finally, -37i/5 = -3n/5 + 2tl = 7n/5, which is between n and 3n/2. Hence, -3n/5 is in
quadrant III.
2-2 If (1,0) is on the x axis then it is either (1,0) or (-1,0). In either case, sin 0 = 0. In the former
case, cos 6 = 1 and in the latter cos 6 = -1. Similarly, if (1,6) is on the y axis, x = 0 and therefore
cos 0 = 0, and sin 0 is either 1 or -1. If the point is on the x axis, 0 is an integral multiple of 7i. If it is
on the y axis, it is 7i/2 more than a multiple of 7i, or 2mz + n/2.
2-3 Evaluate each of these as described in the text. Pay close attention to your signs! You should
find that
sin 300° = - V3/2

cos 225° = -A/2/2
esc 150° = 1/sin 150° = l/(l/2) = 2
cot57i/3 = (cos 5tt/3)/(sin5n/3) = (l/2)/(- V3/2) = - V3/3
tan tl = (sin7r)/(cos7i) = 0/(—1) = 0
sec57i/6 = l/(cos5rr/6) = l/(- V3/2) = -2 V3/3.
2-4 Have fun!

2-5 Any angle in the first or fourth quadrant has a positive cosine (since x is positive there) and
any angle in the second or third has a negative cosine. For similar reasons, sine is positive in the first
and second quadrants (corresponding to positive y) and negative in the other two.
2-6 Let 0 = ZAOC. Since B is on the negative x axis, cos 0 = -OB = - cos lAOB = - cos(ti - 0).
2-7 Are you convinced? If not, keep working on it. Plug in some points.
2-8 The parent function is g(x) = sin ax, so /(x) = g(x + b/a), and the phase shift is -b/a.
2-9 The answer is not tl because the period of tanx is not 2tl\ Look at the graph of tanx. It
repeats every n, so its period is n. Hence, the period of tan2x is n/2. (Graph it and see!)
< 6 >
2-10 While Sin 11 = tz/2, sin 11 is not single valued! For example, we can write sin 11 -
Sin 11 = 571/2 - 7i/2 = 2n, so the given statement is not always true.
2-11 Since escx = 1/ sinx = -1 when x = 3n/2 + 2nn,
arccsc -1 = 3ti/2 + 2 nn.
Since cos7i/4 = V2/2, Cos-1 V2/2 = rr/4. The last is a bit trickier. We seek an x such that
(sinx)/(cosx) = - V3/3, or sinx = (-V3cosx)/3. Substituting this into sin2x + cos2x = 1, we
find that (4cos2x)/3 = 1, so cosx = + V3/2. Since Arctanx is between —tt/2 and ti/2, cosx = V3/2
and sinx = -1/2 so x = -tt/6.
2-12 Any function for which /(x) = /(-x) is even and those for which /(x) = -/(x) are odd.
Thus, sinx, tanx, cotx, and esex are all odd. We can see this just by noting that sinx and tanx are
odd. The other two are just reciprocals of these. Finally, cos x and sec x are even. Query: How could
we deduce that tanx is odd just by knowing that sinx is odd and cosx is even?
2-13 Since sec(270° + x) = 1/ cos(270° + x), we examine

cosine. Shifting cosx (solid line) to the left by 270° gives
cos(270° + x) (dashed line), which we see is the same as sinx.
Thus, sec(270° + x) = l/(sinx) = esex.
Shifting cos x to the left by n results in cos(x + tl), which
we see is the reflection of cos x in the x axis, so cos(x + n) = - cosx.
Shifting tan x (dashed line) to the left by 450° (resulting in the

dotted line), we see that the resulting graph is the graph of cot x
(bold line) reflected in the x axis, so tan(450° + x) = - cot x.
Finally, for sin(37i - x), we shift sin(-x) to the right by 371,
and the resulting graph is the same as sin x. Thus, sin(37i-x) =
sinx.
2-14 Let LEAD - a, AD = 1, and ZCAD = {$. Thus,
A3 - l/(cosn) and AC - 1/cosjS, BD = tana, DC - tanjS,
and
[ABC] [ABD] - [ACD]

(1/2)(AB)(AC) sin ABAC (1/2)(AD)(BD) - (1/2)(AD)(DC)
sin a sin f3
(—) ( ——fi) sin(tf - ^)
\cosa) \cospJ cos a cos
Multiplying both sides by cos a cos jS yields
sin(n - jS) = sin a cos jS - sin jS cos a.
2-15 Use the addition formula on a and —/S:
sin[n + (-jS)j = sinncos(-/3) + sin(-jS) cos(n) = sinncosjS - sin cos a,
where we have used sin(-x) = - sinx and cos(-x) - cosx.

8 > CHAPTER 2. NOT JUST FOR RIGHT TRIANGLES
2-16 We use the same approach as above:
cos[n + (—j8)] = cos a cos(-jS) - sin a sin(-jS) = cos a cos jS + sin a sin jS.
2-17 Let's try this in terms of sines and cosines:
cos(a -18) cos a cos jS + sin a sin j8 _ cot a cot jS + 1

cot(n - jS)
sin(n - jS) sin n cos j8 - sin jS cos n cot jS - cot a
where we have divided top and bottom of the fraction by sin a sin jS.
2-18 We try to write these as differences or sums of angles with which we are familiar. Since
sin 15° = sin(45° - 30°), we have
V6 — V2
sin 15° = sin 45° cos 30° - cos 45° sin 30°
Similarly,
sec(57i/12) = = V6+ VI
cos(7t/6 + 7i/4) cos7i/6cos7i/4 - sin7i/4sin7i/6
Finally,
cos(—345°) = cos 15° = cos(45° — 30°) — cos 45° cos 30° + sin 45° sin 30° =
V6 + V2
2-19 Using the half angle formula.
1 - cos 30°
sin 15° = sin(30°/2) = 2
2-20 Since (V6 + y/2)2 = 4(2 + V3), we have y/2 + V3 = (V6 + V2)/2. Putting this in one
expression for sin 15° yields the other, so the two expressions are equivalent.
2-21 We write tanx/2 in terms of sine and cosine:
tan - - sin(*/2) _ Vl ~ COS J

2 cos(x/2) Vl + cosx
We can get rid of the square roots by multiplying top and bottom by either Vl - cos x or Vl + cosx.
The former yields
1 - cos x 1 - cos x
Vl - cos2* sinx
and the latter gives
x sinx
tan - = --.
2 1 + cos x
2-22 Yes, this is true in general. Go through the derivation in the example with a in place of 3
and b in place of 1 to see so.
2-23 Since sin(ax/b) repeats every 2bn/a and cos(cx/d) repeats every 2dn/c, both functions
complete an integral number of periods from x = 0 to x = z if z/(2bn/a) and z/(2dn/c) are integers.
Since az/2bn and cz/2dn are both integers, the numerator of z must be an integer multiple of both
2nb and 2nd. Thus the numerator of z is at least 2n times the least common multiple of b and d. Since
az and cz must both be integers, the denominator of z can be no greater than the greatest common
factor of a and c. Putting this together, we find the minimum z when the numerator is minimized
and the denominator is maximized, or
_ 2n • [b, d]
(a, c)
2-24 We use our double angle formulas and other identities:
'2sinx - 2 cos2 xsinx /2sinx(l - cos2x) ^—

cos2* = sinx.
(2 sin x cos x) / cos x 2sinx
14. Using sinx = cos(90° - x) on the first four terms, our sum is
cos2(80°) + • • • + cos2 (50°) + sin2 (50°) + • • • + sin2(80°) + sin2(90°) = 5,
where we have used sin2 x + cos2 x = 1 four times.

15. esc (Arcsin \ - Arccos = csc(30° - 60°) = csc(-30°) = 1/ sin(-30°) = -2.
16. Taking sines of the given equation, we have sin y - x. To find tan y we need to find cos y.
From sin2 y + cos2 y - 1, we find cos y = Vl - x2. Thus, tan y = x/ Vl - x2.
17. For each period of sine, the value c is attained twice, (look at a graph of sinx to convince
yourself of this. Since the period of sinnq is 2n/n, there are n periods of the graph between 0 and
2n. Thus, the value c is attained 2n times.
18. The sides are sides of a right triangle since they satisfy the Pythagorean Theorem. Thus,
sinA = a/c = 3/5, sin2B = 2sinBcosB = 2(4/5)(3/5) = 24/25, and sin3C - sin37i/2 - —1. Our
desired sum is then 14/25.
19. Take tangents of both sides, yielding
„ 7i x . 2x, x/2 + 2x/3 7x16 7x

1 = tan — = tan(Arctan - + Arctan —) = -—, ■
4 v 2 3 ' 1 - (x/2)(2x/3) 1 - x2/3 6 - 2x2 ’
Solving this quadratic, we find the roots x = (-7 ± V97)/4. The two principal inverse tangents in the
sum will have the same sign as x. Since their sum is positive, x must be positive, so x = (-7 + V97)/4.
20. The period of the first term in the sum is 1/2 and that of the second is 2/5. Let z be the
period of the given sum. Thus, z is the smallest positive real number such that z/(2/5) = 5z/2 and
2/(1/2) = 2z are both integers. Clearly the smallest such z is z = 2. Notice that the phase shifts and
amplitudes of the terms in the sum are irrelevant.
21. This is good practice in the use of the identities sin2 x + cos2 x = 1 and cos 2x = 2 cos2 x - 1:
1 - cos2x\ 2
sin4 x — (1 — cos2 x)2 =
1-2 cos 2x + cos2 2x 1-2 cos 2x 1 + cos 4x

- + -
4 4 8
= (3 - 4cos2x + cos4x)/8.
Note that we have used the cosine double angle formula 'backwards/ as cos2 x = (cos 2x + l)/2 and
cos22x = (cos4x + 1)/2.
22. Express the sum in terms of sines and cosines and find a common denominator:
cos 10 sin 5 cos 10 cos 5 + sin 10 sin 5

cot 10 + tan 5
sin 10 cos 5 sin 10 cos 5
The numerator is the expansion of cos(10 - 5)! Thus our sum is
cos 5
esc 10.
sin 10 cos 5
23. Writing the equation in terms of sin x we have sin x = 1 - 2 sin2 x. Solving this as a quadratic
in sinx, we find sinx = 1/2 or sinx = -1. Given that 0 < x < n/2, x = 71/6.
24. Seeing that the angles often have ratios of 2 or 1/2, we use the sine double angle formula,
forwards and backwards, and sinx = cos(90°-x). For the first, sin 100° = sin(180°-100°) = sin 80° =
cos 10°, so
sin 10° sin 10° sin 100° = sin 10° sin 10° cos 10° = (sin 10°)(sin 20°)(1 /2)
= sin 10° sin 20° sin 30°.
For the second,
sin 10° sin20° sin 80° = sin 10° sin20°cosl0° = (sin20° sin20°)(l/2) = sin20° sin 20° sin30°.
For the third,
sin 10° sin40° sin 50° = sin 10° sin40° cos 40° = (sin 10°)(sin 80°)(1 /2)
- (sin 10°)(cos 10°)(l/2) - (sin20°)(l/2)(l/2) = sin20° sin2 30°.
25. We repeatedly factor the equation and apply double angle formulas:
8 sin x cos5 x - 8 sin5 x cos x = 8 sin x cos x (cos4 x - sin4 x)

= 4 sin 2x (cos2 x - sin2 x)(cos2 x + sin2 x)
= 4 sin 2x (cos 2x) = 2 sin 4x.
Thus, 2sin4x = 1 and sin4x = 1/2. Since we want the smallest solution, we solve 4x = 30°, so
x = 7.5°.
26. Square the given equation to find sin2 x + 2 sin x cos x + cos2 x = 1 /25, so 1 + sin 2x = 1 /25 and
sin2x = -24/25. Using sin2 2x + cos2 2x = 1 and noting that 2x is in the fourth quadrant, we solve
for cos 2x and cos 2x = 7/25.
27. As before, squaring the given equation gives 1 + sin 2.x = 1/4, so sin2x = -3/4. Thus, x
is in quadrant II. (If it were in quadrant I, sin2x would be positive.) Note that (cosx - sinx)2 =
cos2x - 2sinxcosx + sin2x = 1 - sin2x = 7/4. Since x is in quadrant II, cosx - sinx = - V7/2.
(Why can't it be positive?) Adding this to the given equation gives us cosx = 1/4 - V7/4, so
sinx = 1/4 + V7/4. Thus, tanx = (1 + y/7)/( 1 - V7) = -(4 + V7)/3, and (p,q) = (4,7).
28.Since AB = BC, AB = BC, so LBDA = LBDC. Thus, BD is the angle bisector of iCDA.
Let LBDA = x, so LCD A = 2x. From right triangle DBA (since LDBA is inscribed in a semicircle),
sinx = AB/AD = 1/4. Thus, cos LCD A = cos2x = 1 — 2sin2x = 7/8. From right triangle CAD,
CD/AD = cos LCD A - cos2x, so CD = 7/2.
29. The key here is to recognize that 1 - sin2x = (sinx - cosx)2. (Why?) Thus we have
sin2 x COS2X (sin x - cos x)2

+
3 7 10
Since we seek tan x, we divide both sides by cos2 x, yielding
tan2x 1 (tanx-1)2
3 + 7 = 10
Rearranging this and factoring the resulting quadratic in tan x, we have 49 tan2 x + 42 tan x + 9 =
(7tanx + 3)2 = 0, so tanx = -3/7.
30. Since
tan 20° 4- tan 25c
1 = tan(20° + 25°) =
1 — tan 20° tan 25°'
we have 1-tan 20° tan 25° = tan 20° +tan 25°. Thus, tan 20° + tan 25° + tan 20° tan 25° = 1. Expanding
the given product, we have
(1 + tanA)(l + tanB) = 1 + tan A + tanB + tan A tan B = 1 + 1=2.
31. Since sin 0/2 = V(1 ~ cos 0)/2, we find cos 6 as
x- 1
cos 0 = 1-2 sin2(0/2) = 1
x x
,
Since tan2 0 + 1 = sec2 0, we find tan 0 = Vsec2 0-1 = Vx2 - 1 where we take the positive square
root since 0 is acute.
D C 32. Draw altitude DE to AB. Thus,
tan lEDB + tan LED A

tanx = tan (LEDB + LED A) =
1 - tan LEDB tan LED A
g EBCD is a rectangle, so EB = 6 and ED = 4. Hence, we find tan LEDB = 3/2 and
A E
tan LADE = (y - 6)/4. Substituting this in the above yields
y/4
tanx =
1-3(y - 6)/8
Solving this equation for y gives us our answer.

26 sin x
V —-•
2 cos x + 3 sin x
The moral of this problem is that making rectangles is often very useful.
33. Using the cosine double angle formula backwards we can write
cos 72° = 2 cos2 36° - 1 and cos 36° = 1-2 sin2 18°.
Since sin 18° = cos72°, we can write the latter equation as cos 36° = 1 - 2 cos2 72°. Adding this to
the first equation above yields
cos 36° + cos 72° = 2(cos2 36° - cos2 72°) = 2(cos 36° - cos 72°)(cos 36° + cos 72°).
Dividing both sides by 2(cos36° + cos 72°) gives us cos 36° - cos 72° = 1/2. This may seem a bit
contrived, but there is a method involved. We try to express each term in the initial expression in
terms of the other.
34. We find the area of ABCDE, where AB = 1, in two different ways. In the A
first diagram, LABC = 3n/5 and LACB = n/5 (since A ABC is isosceles). Thus,
/.ACF - 3n/5 - tl/5 = 2n/5, AF - CF tan LACF = (1/2) tan(27i/5), and
[ABCDE] = 2 [ABC] + [ACD]
= 2(1/2)(1)(1) sin37i/5 + (CD)(AF)/2
A
= sin37i/5 + (l)(tan27i/5)/4
= sin37i/5 + (tan27i/5)/4.
Since sin37i/5 = sin2n/5, we have 4[ABCDE] = 4sin27r/5 + tan27i/5. In the

second diagram, ZCQF = 27t/10 = n/5, so OF = CF cot(7i/5) = (cot7i/5)/2, and
[ABCDE] = 10[OCF] = 10(OF)(FC)/2
= 5(l/2)(l/2)(cot~)= jCOt|.
Combining this with the first expression for [ABCDE], we have our desired expression.
35. Shown is triangle ABC with LABC = LACB - 2ABAC = 72°. We draw angle A
bisectors CF, BE, and AD. Since LABI - LCBI = LBCI = lACI - 36°, we have
A AEB ~ A AFC ~ ABIC

and A ABC ~ A CFB ~ aBEC ~ A BFI ~ ACIE. If we let AB = y and CB — x, we have
FB/FC = BC/AB, so FB = x2/y (since FC = BC = x). From FI/FB = BC/AB, we find
(since IC = IB = FB and FI = FC - IC)
C
* - *2/y = x
x2/y tf
Thus, yx - x2 = x3/y, or y2 - yx - x2 = 0. Applying the quadratic formula, we find y = (x ± V5x^)/2 =

x(l + V5)/2, where we take the positive square root since y and x are both positive. From right
triangle ABD, we have
sin LDAB - sin 18° - BD/AB = x/2y = (V5 - l)/4.
Similarly, cos LlCD = cos 36° = CD/IC = (x/2)/(x2/y) = y/2x = (1 + V5)/4.

Chapter 3
More Triangles!
3-1 First, when B is on the circle centered at A with radius AC, we have
b = a and cos C = CB/DC = a/2b
a
a2 + b2 - lab cos C = a2 + b2 - lab = b2 = c2
2b
For B outside the circle, we will address the case where BC extended in¬
tersects the circle on the opposite side of AC with respect to B as shown.
Since ZECD and ZACB are supplementary, cos ZECD = - cos ZACB and
EC - lb cos ZECD = -lb cos LACB. Since BF - c-b (AF = AC = b), from the
power of point B we have
(c - b)(c + b) = a(a - lb cos ZACB),
which expanded and rearranged is the law of cosines.

3-2 Since 2R = AC/ sin B = 12, R = 12/2 = 6.
3-3 When angle A is right, BC is the diameter of the circumcircle. Since

sin 90° = 1, BC/ sin .A = 2R is true. For the case of obtuse lA, consider the diagram
where CD is the diameter of the circumcircle of aABC. Since sin D = BC/CD = a/2R
and ZD = BC/2 = 180° - BDC/2 = 180° - LA,
™ = sinD = sin(180° - A) = sin A,
which completes our proof.
3-4 If one of the angles is obtuse, the proof is clear from the shown C
diagram, where AB < DB < BC. The first inequality can be seen from the
Pythagorean Theorem. If both angles are acute, we have BC/AC = sin A/ sinB B
from the law of sines. If ZA > ZB, then BC/AC = sin A/ sin B > 1, so BC > AC.
Similarly, if BC > AC, we can deduce sin A > sinB, from which ZA > ZB follows.
< 13 >
14 > CHAPTER 3. MORE TRIANGLES!
3-5 Let AB = x and AC = y. Since lADB = 180° - /.ADC, cos /ADB = - cos /ADC, and
x2 - 52 _ y2- 45
-48 “ -36 '
Rearranging this we find 3x2 + 4y1 - 336. From the Angle Bisector Theorem we have x/y = 4/3, and
our equation becomes 21x2/4 = 336, so x = 8.
3-6 Continuing from the previous example, we write a = 2R sin A, from which the desired result
follows.
3-7 From Heron's formula, we have [ABC] = \/21(8)(7)(6) = 84. The shortest altitude is the one
drawn to the longest side, so |(15)(x) = 84, and our shortest altitude has length 56/5.
3-8 Since /ABC and /BCD are supplementary, cos /ABC = - cos /BCD, and D_C
AC2 - AB2 - BC2 _ BD2 - BC2 - CD2

-2(AB)(BC) ~ -2 (BQ(CD) ‘ A B
Since CD = AB, we can rearrange this as the desired AC2 + BD2 - AB2 + BC2 + CD2 + DA2.
3-9 Solving 4AD2 + BC2 = 2(AB2 + AC2) for AD, we find
-1
AD = - yj2(AB2 + AC2) - BC2.
3-10 Solving for d in Stewart's Theorem, we find
bmb + cnc-man ,, /m\ 9 (n\

d =-= bl — + c - - mn.
a \a a
writing a as m + n and noting that m/(m + n) = 1/(1 + n/m) = 1/(1 + b/c) - c/(b + c) (remember,
n/m - b/c from the Angle Bisector Theorem), we have
d2 = b21 - C \ c2 [f --|
u ' +-2
b - mn
b + cJ \b + c
bc(b + c) ( m \ ( n
b+c \m + n/ \m + n.
Using our aforementioned expressions for m/(m + n) and n/(m + n), we have
b \ n l /. a*
d- \ be- a2 = \ be [ 1 -
b + c) V b + c (b + c)2
3-11 Since AD is a median, m = n = a/2 and Stewart's Theorem provides
a3 2 ab2 ac2
— +ad^ = — + —.
4 2 2
Solving for d, we find
b2 + c2 a2
d=
36. Since 62 + 82 > 92, the triangle is acute.

37. From Heron's formula, the area of the triangle is V21(8)(7)(6) = 84. Alternately, we can
express the area as (14)(B)/2. Setting these equal, we find h = 12.
38. From the Angle Bisector Theorem, CB/AB = CD/AD = 4/3, so BC = 4 and AC = 7. Thus,
BD = V/(8K6)(l-^)=6.
39. Let the sides be 4x, 6x, and 8x. Since the smallest angle is opposite the smallest side, we apply
the law of cosines to find
16x2 = 36x2 + 64x2 - 2(6x)(8x) cos 6.
Solving for cos 6, we find cos 6-7/8.
40. Extend median AM past M to D as shown so that ABDC is a parallel¬

ogram. Since AB = CD, AC = BD, and AD = 2AM, we have
AD2 + BC2 = AC2+ CD2+ DB2+AB2, so

36 + BC2 = 2(64 + 16).
Finally, BC - 2 V31-
41. Shown is AABC with angle bisector AD. Applying the law of sines to A ADC A
and AADB and noting that sin /.ADC - sin /ADB and sin /CAD = sin /DAB, we
have
AC _ sin/ADC
CD sin /CAD
sin /ADB
sin /BAD
AB
BD'
42. Multiplying both sides by a + b + c, we find a3 + b3 = c2(a + b). Factoring the left side, we find
(a + b){a2 + b2 - ab) = c2(a + b), so dividing by a + b gives c2 = a2 + b2- ab. Since from the law of cosines
we have c2 - a2 + b2 - lab cos C, we know cosC = 1/2 by comparison to the above equation. Thus
sinC = V3/2 (why?) and tanC = V3.
43. Drawing the altitude from A to BC, we can use the Pythagorean Theorem to determine the
altitude has length V49 - 1 = 4 V3. Thus, the triangle has area (2)(4 V3)/2 = 4 V3. Since abc = 4KR,
we find R = AO = 49 V3/24.
A B £ 44. Extend AB to E so that BECD is a parallelogram. Thus, AE =

AB + BE = AB + CD and applying the law of cosines to A ACE gives
AE2 = (AB + CD)2 = AC2 + EC2 - 2(AC)(CE) cos LACE.
Since EC = BD and (AB + CD)2 - (AC2 + BD2) = 0, we have cos LACE = 0. Thus, LACE = 90°. Since
BD || EC, BD _L AC and ABCD is orthodiagonal.
45. Seeing sines, we consider the law of sines. Writing sin A = a/2R and likewise for the other
two angles, our fraction becomes
+ b3 + c3
7= - 8R3
sin3 A + sin3 B + sin3 C
Thus, R - a/7/2. Since the triangle is inscribed in a circle with diameter ^7, the maximum length a
side can have is ^7.
46. Writing sin(A/2) = V(1 - cos A)/2 and applying the law of cosines yields
a2-b2-c2
1+ 2 be la2 - (b - C)2
sin
4frc
1 (a - b + c)(a + b - c) (s - b)(s - c)
V 4 be V be
47. Rather than using the half angle formula for cosine and the law of cosines, we note that
. . A A . A _ 2[ABC] _ 2 Vs(s - a)(s - b)(s - c)

2 sm — cos — = sm A
2 2 be be
Using our expression for sin(A/2) from the previous problem, we find
A ls(s - a)
cos - =
be
48. Since cot(A/2) = cos(A/2)/ sin(A/2), we have from the previous two problems
2 A B C 2
r cot — cot — cot — = r = r2s
2 2 2 (s - a)(s - b)(s - c) (s - a)(s - b)(s - c)'
Noting that r = ^/(sâ)(s:^F)(T:^c)Js, we recognize the right side of the above as r2s(l/r) = rs =
[ABC], as desired.
49. By connecting both B and C to the center, O, of the circle we can use SSS congruency to show
AOXB = AOXC, so LBXD = LDXC. Using SAS we can then show ABXA = aCXA (lBXA - lCXA,
BX — CX, and AX — AX) so that LBAX = LCAX = 6° (since LBAC — 12°). From right triangle ABD,
AB = AD cos LBAD = cos 6°. Since LBXD = 18°, LBXA = 162° and lABX = 12°. Finally, we apply
the law of sines to AABX to find
AX _ AB
sin 12° sin 162°’
Since sin 162° = sin 18° and AB = cos 6°, AX = cos 6° sin 12°/ sin 18°.
50. Like many times before, we extend AD past D to M so that ABMC is a

parallelogram. From the law of cosines applied to A ACM, we have
AM2 = AC2 + CM2 - 2(AC)(CM) cos ZACM.
Since CM = AB, AM = 2AD, and cos ZACM = cos Z(180° - ZCAB) = - cos ZCAB, this equation
becomes
4AD2 = AC2 + AB2 + 2(AC)(AB) cos ZBAC.
51. For this problem, we apply the fact that if in ADEF we have ZD > ZF, thenFF > DF. We use this
on ZxAXC, then A BYC, then A CZB. We can then relate the cevians of A ABC to the sides. Since c <b < a,
we have ZC < ZB < ZA. Since ZAXC > ZB > ZC, AC > AX. Similarly, since ZBYC > ZA > ZC,
BC > BY, and since zCZB > ZCAB > ZB, BC > CZ. Hence, AX + BY + CZ < AC + BC + BC = 2a + b.
A 52. First we apply the law of sines to A ABC to determine BC, the common
side of A ABC and A OBC. We find
sinZl
Since OB - 1 and sinA - sin 60° = V3/2, we have BC = V3. Since ZA = 60°,
BC = 120°. Thus, ZBOC = 120°. We can now apply the law of sines to ABOC,
finding Raboc — BC/2 sin ZBOC = 1.
53. Let the desired length be x. We attack this problem by finding the area as [ABC] - {cib/2) sin C
and [ABC] = [ACD] + [BCD] = (ax/2) sin(C/2) + (bx/2)sm(C/2). Since sinC = 2sin(C/2)cos(C/2), we
have
(flfr/2)(2sin(C/2)cos(C/2)) = (ax/2) sin(C/2) + (bx/2)sin(C/2).
Solving this for x, we find

2ab cos j
x-
a+b
54. Let ZACD = ZDCB = 0 and CD - x. We solve this problem by finding [ABC] C
in two ways:
A B
[ABC] = [ACD] + [BCD] D
(ab/2) sin ZACB = (ax/2) sin 6 + (bx/2) sin 6.
Since sin ZACB = sin 20 = 2 sin 0 cos 0, the above becomes
ab sin 0 cos 0 = (ax/2) sin 0 + (bx/2) sin 0.
Dividing by sin 0 and by ab (so we can get 1/a and 1/b terms as in the desired expression), we find
2 11
- cos 0 = - +
x a b
We are given x = 6 and cos 0 = 1/3, so we find 1/a + 1/b — 1/9.

55. Let x = AD and y = AE. From the Angle Bisector Theorem applied to triangles ABE and
ADC, we find AB = 2y/3 and AC = lx. Since we have angle bisectors of triangles ABE and ADC, we
apply the angle bisector formula d2 + mn = be proved in the text to them to get equations in x and y:
x2 + 6 = 2y2/3 and y1 + 18 - lx2.
Solving these, we find (x, y) = (3 V6,3 VlO). Ffence the sides of the triangle are AB - 2y/3 = 2 VlO,
AC = 2x - 6 V6, and BC = 11. The shortest of these is 2 VlO.
56. First, A AMD = A AME by ASA (lEAM = /.DAM, AM = AM, and /AMD = /AME). Similarly,
ACDM = aCEM. Hence, AD = AE = 6, DC' = EC = 12, and C'M = CM = BM = DM + 10.
Since [ABM] = [ACM] and [ADM] = [AEM], we have [ABM] - [ADM] = [ACM] - [ABM], or
[ABD] = [CBM] = [C'DM]. Hence,
[ABD] = sin ZBDA = sin ZC'DM = [C'DM],
so DM = 5 since ZBDA = /C'DM. Applying the law of cosines to A ABD and noting ZADB = /C'DM,
we have
AB2 100 + 36 - 2(10)(6) cos ZADM

136 -120 cos /C'DM
/ 225 - 144 ~ 25 \
136 -120 = 192.
V 2(12) (5) J
Thus, AB = 8 V3, and B = 8.
57. Seeing three angles which add to tl, we let these be the angles of triangle
ABC, where ZA — a, /B — /3, and ZC = y. Drawing the circumcircle of A ABC
centered at O, we have ZBOC = 2a, ZAOB = 2y, and ZAOC = 2j8. Letting
AO - OB = OC = 1, we have
C
111
[ABC] = - sin2n + - sin2jS + - sin2y.
Expressing the area as (ab/2) sin C and using the law of sines we have
tAvm-a^ ■ r- _ (2BsinA)(2BsinB) . ^ „ .
[ABC] - — sm C ----sm C = 2 sm a sm jS sin y.
Setting these two expressions for [ABC] equal we have the desired expression.
Chapter 4
Cyclic Quadrilaterals
4-1 Since their sides satisfy the Pythagorean Theorem, both A ABC and aACD
are right triangles. Hence, /.ABC + /ADC - 180°, and ABCD is a cyclic quadrilateral.
Since
tan a + tan j8
tan(a + jS)
1 - tana tan/T
we need only find tana and tan/1. The first is simply AB/BC since /ACB is in a right triangle. For
the second, note /ABD = /ACD, so tanjS = tan /ACD = AD/CD = 24/7. Using these values in the
above expression yields tan(a + jS) = -4/3.
4-2Rearranging the given equation as AO/DO = BO/CO and noting that/AOB =

/COD, we have A AOB ~ A DOC by SAS similarity. Hence, /CAB = /BDC and ABCD
is cyclic.
58. In an isosceles trapezoid the opposite interior angles are supplementary; thus, an isosceles
trapezoid is cyclic. This proves the 'if' part. To show the 'only if' part, we note that any inscribed
trapezoid has supplementary opposite interior angles and is thus isosceles.
59. Since /BAC = /BDC and /AEB = /CED, we have A ABE ~ aDCE. Hence,
B AB/BE = CD/CE and CE = 42/5.
60. Let /ABM = /ACB = x and /ADM = /ACD = y. Hence, /BAD = 180° - x-y
from A ABD and /BCD = x + y. Since /BAD + /BCD = 180°, ABCD is a cyclic
quadrilateral; thus, /ABM = /ACD = /ADM. From this, A ABD is isosceles and
< 19 >
20 > CHAPTER 4. CYCLIC QUADRILATERALS
61. The diameter of the circle has length 17. Since aABC is right, its sides must
satisfy the Pythagorean Theorem so we deduce that AB and BC are 8 and 15. Since / \\
CD = 12, AD = Vl45. Applying Ptolemy's Theorem, we find BD = (96 +15 Vl45)/17. y \ X/7/
62. In general, we apply the law of cosines to find the diagonals as shown in an pp—-q
example in the text. If we try this here, we find that the cosine of the angle between
the sides of length 25 and 60 is zero, so the diagonal of the quadrilateral opposite this angle is a
diameter of the circle. From the Pythagorean Theorem, or noticing the Pythagorean triple (5,12,13),
we find the diameter has length 65. The moral here is that when given the side lengths of a cyclic
quadrilateral, check to see if a diagonal is a diameter by seeing if the sides satisfy a2 + b2 = cz + d2 = x.
Make sure you see why this ensures that the quadrilateral has a pair of right interior angles.
63. If you don't remember the basics of transformational geometry, go back to Volume 1 and
review. Let the angle of rotation be a. By definition, /.AOA' = a. From the principles of rotation, AB
and its image A'B' will intersect in an angle of a, as will AC and A'C'. Hence, /AMA' = /AOA' =
/ANA'. From the first equality, M is on a circle with A, A', and O. From the second equality, N is on
a circle with A, A', and O. Hence, the five points are all on one circle.
64. Hold on to your hat. From parallel lines, vertical angles, and K
angle bisectors,
/DLA - /CLK (vertical angles)

= /KAB (since DC || AB)
= /KAD (AK bisects /BAD)
= /AKB (since AD || KB)
From these equalities triangles CLK, DLA, and ABK are isosceles, so LC = CK, AD = DL, and
DC = AB = BK. Since we also have OL = OC = OK, A LCO = A CKO by SSS congruency, so
ZLCO = /CKO. From SAS congruence, we have aDCO = ABKO. Therefore, /CBO = ZCDO and
DBCO is a cyclic quadrilateral.
65. Let Q be the center of square ABCD. Since /APB + /AQB - 90° +90° = 180°,
APBQ is a cyclic quadrilateral. Thus,
/APQ = /ABQ = /BAQ = /QPB,
so Q lies on the angle bisector of /APB (since /APQ = /QPB). As discussed in

Volume 1, any line through the center of a square bisects the area of the square
(try proving this yourself).
66. Let EFGH be the quadrilateral formed by connecting the midpoints of the sides of ABCD,
where E and F are the midpoints of AB and BC, respectively. Since A EBP ~ a ABC, EF || AC.
Similarly, we can show GH || AC || EF and EH || BD || FG. Thus, EFGH is a parallelogram. If ABCD
is orthodiagonal, AC _L BD implies EF _L FG and EFGH is a rectangle and hence cyclic. For the 'only
if' part, if EFGH is a cyclic parallelogram, it must be a rectangle, as its opposite angles are equal and
supplementary. Since EFGH is a rectangle, ABCD is orthodiagonal because the diagonals of ABCD
are parallel to the sides of EFGH and are therefore perpendicular.
67. Since /A + ZC = 180°, ABCD is cyclic and BD is a diameter since

/.BAD = 90°. If we continue DE to meet the circle at M, we find that EFBM is a
^ rectangle (since ZDMB = 90°, as it is inscribed in a semicircle). From the power
of point E, we have (EM)(ED) = (EC)(EA), so EM = BF = 21/5.
D
68. Since AO • CO = BO ■ DO, quadrilateral ABCD is cyclic. Since A ABO ~
ADCO and aBCO ~ A ADO, CD = AB(OD/AO) = 9/2 and BC = AD/2. From
Ptolemy's Theorem,
(6)(9/2) + (AD)(AD/2) = (10)(11),
so AD = Vl66.
69. Letting AB = a, BC = b, CD = c, and DA - d, we split the quadrilateral into two triangles with
diagonal AC. Thus,
1
[ABCD] = - (ab sin B + cd sin D).
Squaring this we have
[ABCD]2 = - [a2b2 sin2 B + labcd sin B sin D + c2d2 sin2 D) .
Writing the first and last terms in terms of cos B and cos D, we use the law of cosines and lots of
algebra like that in our proof of Heron's formula to show the desired expression. If the quadrilateral
is cyclic, B + D = 180° and the second term in Brahmagupta's formula vanishes, leaving [ABCD]2 =
(s - a)(s - b)(s - c)(s - d) for cyclic quadrilaterals.
Chapter 5
Conics and Polar Coordinates
5-1 First we'll discuss the effects of varying a on the general equation y — k = (1/4a)(x-h)2. First,
positive a's correspond to upward opening parabolas and negative a's cause the parabola to open
downward. As we increase a, 1 /4a decreases, so larger changes must be made to x to affect y. Hence,
increasing a causes the parabola to open wider. Similarly for the form x - h = (l/4a)(y - k)2, negative
a corresponds to leftward opening parabolas and positive a to rightward opening parabolas.
5-2 Since the axis divides the parabola in half, it must pass through the vertex. For upward or
downward opening parabolas, the axis is vertical and hence has the equation x = h. For parabolas
opening to the right or left, the equation is then y = k.
5-3 We'll just address the parabola y-k = (1/4 a)(x - h)z. The focus of this is (h, k + a), so the line
through the focus parallel to the directrix is y = k + a. To find the endpoints of the latus rectum, we
let y = k + a in our parabola equation, yielding a = (l/4a)(x - h)2, or x = h ± 2a. Hence, the endpoints
of the latus rectum are (h + 2a,k + a) and (h -2a,k + a), so the length of the segment is 4a. We can use
this to plot parabolas by plotting the vertex and focus, drawing the latus rectum, then drawing the
curve through the endpoints of the latus rectum and the vertex.
5-4 The line is called the directrix because it determines which direction the parabola points.
Pick a point as the focus, then use a few different lines (of all different slopes) as the directrix and
you'll see this connection.
5-5 Completing the square, we have
y = (l/2)(x2 + 6x) + 4
y + (l/2)(9) = (1/2)(x + 3)2 + 4
y = (l/2)(x + 3)2 - 1/2.
Hence, the parabola opens upward. The vertex is (-3, -1/2), the focus is (-3, -1/2 + 1/2) = (-3,0)
(since l/4a = 1/2, we find a = 1/2). The directrix is y = -1/2 - 1/2 = -1 and the length of the latus
rectum is 4(1/2) = 2.
5-6 Since the directrix is vertical and the focus is to the right of the directrix, the parabola points
to the right. From the focus we have k - 4 and h + a = 5/2. From the directrix we find h- a = 3/2.
< 22 >
Solving these we get h = 2 and a - 1/2. Hence, our parabola is x - 2 = (l/2)(y - 4)2.
5-7 If a = b, we have (x — h)2 + (y — k)2 = a2 and our ellipse is a circle, as the equation describes
the set of points which are a away from (h,k). Thus, a circle is just an ellipse whose axes have the
same length.
5-8 As c grows for a given a, the foci get further and further from the center and the ellipse
appears more and more elongated. Hence, c/a measures how eccentric, or odd, the ellipse appears.
In the case of a circle, c = 0 and the eccentricity of a circle, which has no elongation, is 0.
5-9 We segregate the x terms and y terms and complete the square:
3(x2 - 2x) + 4(y2 + 2y) - -3

3(x2 - 2x + 1) + 4(y2 + 2y + 1) = -3+ 3(1)+ 4(1)
3(x - l)2 + 4(y + l)2 = 4.
Dividing by 4, we have
(* -1)2 (y + i)2 = ,
4/3 1
Since 4/3 > 1, the major axis is parallel to the x axis. Thus, the center is (1, -1), a = y/4/3 = 2 V3/3,
and b - 1. Thus, c = x[\j3 = V3/3. The axes then have length 2a = 4 V3/3 and 2b = 2. Finally, the
foci have coordinates (1 ± V3/3, -1).
5-10 The only solution to the given equation is (2,-1). Thus our 'ellipse' is actually only a
point. If the number on the right is negative, there are no solutions since the sum of perfect squares
is always nonnegative.
5-11 We'll consider the case where the major axis is parallel to the x axis:
(* ~ h)2 (y - k)2
= 1.
a2 b2
Since the latus recti are segments on vertical lines through the foci, they have the equation x -h±c.
We'll consider the line x = h + c. Substituting this in the above equation to find the intersection
points, we have
c2 , (y - k)2 = 1.
a2 b2
Writing c2 = a2 - b2, our equation becomes
(y ~ k)2 = ^
bz a2
Solving for y, we find y - k± (b2/a). Hence, the distance between the endpoints of this latus rectum
is 2b2/a. This also holds for the ellipse whose major axis is parallel to the y axis.
5-12 The center is the midpoint of the segment connecting the foci, so the center is (-1,1). Since
the center is 4 away from each focus and the minor axis has length 4, we have c = 4 and b = 2, so
a2 = c2 + b2 - 20. Since the foci are on a horizontal line, the ellipse has major axis parallel to the x
axis. Thus, our equation is
(x + l)2 (y -1)2
= i.
20 4
24 > CHAPTER 5. CONICS AND POLAR COORDINATES
5-13 Letting y - k = ±\{x-h) in the hyperbola equation
(x - h)2 (y - k)2 _ 1
a2 b2
we find 0 = 1, which clearly can never be true. Hence, the lines y - k = ±^(x - h) never intersect
the hyperbola. The lines are asymptotes because we can choose points arbitrarily close to this line
which are on the hyperbola. (Why?)
5-14 By the same analysis as in the text, the center is (h, k), the vertices are a above and below the
center (at (h, k ± a)), the foci are c = Vfl2 + b2 above and below the center (at (h, k ± c)), the transverse
axis is again la, the conjugate axis is 2b, and finally, the asymptotes are y -k= ±^(x -h).
5-15 As we did with the ellipse, we'll consider the case where the transverse axis is parallel to
the x axis:
(x - h)2 (y ~k)2_
a2 b2
As before, we consider the latus rectum on the line x — h + c. Substituting this in the above equation
to find the intersection points, we have
_ (y - k)2
a2 b2
Writing c2 = a2 + b2, our equation becomes
(y - k)2
b2 a2'
Solving for y, we find y = k± (b2/a). Hence, the distance between the endpoints of this latus rectum
is lb2/a.
5-16 First, no point on a coordinate axis can lie on xy - c for nonzero c since if either x or y is
zero then we cannot solve for the other. This alone does not mean the axes are asymptotes. We must
show that the hyperbola comes arbitrarily close to the asymptotes. This is easily done by noting that
we can choose arbitrarily small x and find y as c/x so that (x, y) is on the hyperbola. Hence we can
find points on the hyperbola arbitrarily close to the y axis and similarly for the x axis.
5-17 The midpoint of the segment connecting the vertices is the center of the hyperbola, so the
center is (-2, -1). Since the vertices are on a vertical line, the hyperbola opens up and down. Since
the vertices are 2 yfl from the center, a = 2 V2. Since the conjugate axis has length 4, b = 2. Thus, our
hyperbola is
(y + l)2 _ (* + 2)2 _
8 4
5-18 For (6,-6V3) we have r = V36 + 108 = 12, and 6 = tan_1(-V3). Since the point is
in the fourth quadrant, we have 6 = 300°. Hence, the point is (12,300°). For the polar point
(-2,405°) (remember, negative radii are perfectly ok), we have x = rcosO = (-2) cos 405° = - \fl
and y = (-2) sin 405° = - V2. Thus, our point is (- V2, - V2).
5-19 From x — r cos 6 and y = r sin 8, we have
6(rcos0)(rsin0) = 8
3r2(2cos 0sin@) = 8
3r2 sin 26 = 8
as the polar equation.
5-20 For the first equation, we multiply by cos 6 and find r cos 6 = 4, so x = 4 is the rectangular
form. For the second, we multiply both sides by r to get r2 on one side and r sin 6 = y on the other,
resulting in x2 + y2 = 3y.
5-21 Horizontal lines are of the form y = a, which are r sin 6 = a in polar form and vertical lines
are x = b, or r cos 0 = bin. polar form. As discussed in Volume 1, lines through the origin are 8 - c, a
constant.
5-22 Multiplying both sides by r to force r2, r cos 8, and r sin 8 terms, we have
r2 = ar sin 8 + br cos 8
x2 + y2 = ay + bx.
This rectangular form describes a circle.

5-23 In our equations for (x,y) and (x',y') let 8 - a + /?, so that x = r cos 8, y = rsin0, and
P = 8 - a. Thus, we can write
x' = rcos(8 - a) = r cos 8 cos a + rsin0sinn = xcosa + ysina

y' - r sin(0 - a) = rsin 8 cos a - rcos 0sinn = ycosa - xsinn.
5-24 Yes. Any circle, no matter how it is oriented, can be described as the set of points equidistant
from a given point. As we saw in Volume 1, the distance formula yields the form of any circle and
the form cannot include an xy term. Thus, any circle can be described without an xy term.
70. Since the asymptotes of a hyperbola intersect at the center, we deduce that the center of the
hyperbola is (4,1). Since the given vertex is to the right of the center, the hyperbola opens to the
right and left. Since this vertex is 2 to the right of the center, we have a = 2. Returning to the our
asymptote, since the slopes of these lines are ±|, our value of b is 3. Hence, our hyperbola is
(x - 4)2 (y - l)2
4 9
71. If we instead return to our form of the parabola in the text, y - k = (l/4a)(x - h)2, we
immediately see h = 4 and k = 2 from the vertex. Hence, we have y - 2 = (l/4a)(x - 4)2. Substituting
the given point (2,0) in this equation for x and y (since the point is on the curve), we find -2 = 1 /a,
so a = -1/2. Thus, our parabola is
*=4FuZ)(X~4)2 +2= -X2+4X-6-
Our answer to the problem then is (-l/2)(4)(-6) = 12.

72. Since the distance between the centers is V242 + 72 = 25, the circles
clearly do not overlap. To fully contain the two circles, the third circle
must have a diameter equal to the longest segment which can be drawn
with endpoints on the two circles. This segment is clearly the segment which passes through the two
centers, as segment AB in the diagram. The length of AB is the distance between the two centers plus
the two radii of the circles. (Make sure you see this.) Our desired radius is then (25 + 3 + 4)/2 = 16.
73. First, since Vfl2 - b2 = V5 and the center of the ellipse is the origin, the points (± V5,0) are
the foci of the ellipse. Hence, the two legs of the bug's path which together go from one focus to
some point on the ellipse to the other focus have a total length equal to the major axis, or 2(3) = 6.
(This is from application of the constant sum of distances property of an ellipse.) The other leg of
the bug's journey is just the distance between the foci, or 2 V5. Hence, the bug's total journey has
length 6 + 2 V5-
74. A line can intersect a hyperbola in 0,1, or 2 points. Since we are given that the lines intersect
the hyperbola, we can rule out 0. We may think that 1 is ruled out as well by the non-tangency
condition, but that's not true. Consider a line parallel to one of the asymptotes. It will only meet
the hyperbola in one point. If it met the hyperbola in two points, it must then also intersect the
asymptote to which it is parallel, which is clearly a contradiction. Make sure you see this. Since
each line intersects the hyperbola in 1 or 2 points, and it is possible for the two lines to intersect the
hyperbola at the same point the possible numbers of intersections among the two lines are 1, 2, 3,
and 4. (To get 1 intersection, consider two lines such that each is parallel to an asymptote, and they
both pass through the same vertex of the hyperbola.)
75. Let the x coordinate of A be k, so that A = (k, -k2/2). Hence, B = {-k, -k2/T) (why?) and the
sides of the triangle have length 2k. Using the distance formula we have OA = AB, so
Squaring and rearranging, we find fc4 - 12k2 = k2(k2 - 12) = 0, so k = 2 V5. Hence, the sides of the
triangle have length 2k - 4 V3.
76. Let the center of the base be the origin of a coordinate system. From the given information, the
points (12,0), (-12,0), and (8,18) are all on the parabola. Since the center of the base is directly below
the vertex, the x coordinate of the vertex is 0. Hence, we can describe our parabola as y = cx2 + k for
some constants c and k. Using two of our given points on the parabola, we have
0 = c(122) + k
18 = c(82) + k.
Solving for c and k, we find c = -9/40 and k = 162/5. Thus, the vertex of the parabola is (0,32.4) and
the desired height is 32.4.
77. Any intersection of the given line and the ellipse will correspond to a simultaneous solution
of the two equations describing the curves. Letting y = mx + 1 in the ellipse equation, we have
x2 + 4(mx +1)2 = 1, so (4m2 + l)x2 + 8mx + 3 = 0. If the line and the ellipse are to meet only once, this
equation must have only one solution. Hence, we view the equation as a quadratic in x and set the
discriminant equal to 0, so
(8m)2 - 4(4m2 + 1)(3) = 0,
or m2 = 3/4.
78. First we multiply by cos 6 to get rid of the sec 6, yielding
r cos 6 = 2 + cos2 6,
or x = 2 + cos2 6. Multiplying both sides of this by r2 (to turn the cos2 6 into x2), we have
xr2 = 2r2 + r2 cos2 6

x(x2 + y2) = 2(x2 + y2) + x2.
Expanding this and rearranging, we have x3 + xy2 - 3x2 - 2y2 = 0.

79. Since F1F2 is a diameter of the circle, the triangle PF1F2 is a right triangle. Let x and y be the
lengths of PFi and PF2. Hence, [PF1F2] = xy/2 = 26 and PFi + PF2 = x + y = 15 from the constant sum
property of an ellipse (remember, P is on the ellipse and F\ and F2 are foci). The distance between
the foci is the hypotenuse of our right triangle PF1F2 and hence is sjx2 + y2. As discussed at length
in Volume 1, we can find x2 + y2 as x2 + y2 = (x + y)2 - 2xy = 225 - 104 = 121, so the desired distance
is Vl2T = 11.
80. Seeing distances and equations involved, we think to try analytic ge¬
ometry. We let A = (0,1), B = (0,0), C = (1,0), and D = (1,1). Our condition
u2 + v2 = w2 then becomes
AP2 + BP2 CP2

x2 + (y - l)2 + x2 + y2 (x - l)2 + y2
x2 + 2x + (y - l)2 1
(X + l)2 + (y - l)2 2.
This describes a circle of radius V2 centered at (-1,1), or point X in the diagram. Any point on this
circle can serve as P in the problem. The one furthest from D is found by drawing the line from D
through the center of the circle meeting the circle again at Y as shown. Our desired distance is then
DX + XY = 2 + V2.
81. Let the foci of the ellipse be X and Y and one point of tangency be P, where the circle has
center X. One special property of an ellipse (and one of the reasons for the name foci for these special
points) is that if we throw a ball from one focus and hit the ellipse, it will rebound directly to the
other focus! Think about why this occurs. Hence, if we through a ball from X to P, it will rebound
straight to Y. Since P is on the circle with X as its center, a ball thrown from X to P must go right
back through X as well. The only way these can both be true is if X, Y, and P are all collinear, i.e. P
is an endpoint of the major axis and the circle 'fits' into the end of the ellipse. The point of tangency
then is an endpoint of the major axis. If we let C be the center of the ellipse, we then have CP = 5
and CX = V52 - 42 = 3, so the radius of the resulting circle is XP = 2.
82. The given equations describe a hyperbola and a circle with the same center. Hence, we can
take considerable advantage of the symmetry of the resulting figure. Let the points of intersection
in order be A, B, C, and D, where A and B are one branch of the hyperbola and C and D are on the
other. By symmetry, we must have AB = CD, AD = BC, and AC = BD. Hence, our quadrilateral
ABCD is a parallelogram with equal diagonals, which we know better as a rectangle. (By inspection
of the equations, we can also find the solutions as (3,4), (4,3), (-3, -4), (-4, -3), but our intuitive
approach is more interesting.)
83. Since the circle is tangent to both sides of the parabola, we deduce that its center is on the y
axis. Thus, the circle can be described by x2 + (y - k)2 = r2, where the center is (0, k) and the radius
is r. The points of intersection of this and the parabola y = x2 are found by making this substitution
in our circle equation, or
x2 + (x2 - k)2 = r2.
Writing this as a quadratic in x2 and applying the quadratic formula, we have the equation x4 + (1 -
2k)x2 + k2 — r2 = 0, so
j2 _ 2k - 1 ± yj(2k - l)2 - 4(k2 - r2)
X — •
2
Since the points of tangency will clearly have the same y value and opposite x values, there can be
only 1 distinct value for x2. Hence, the discriminant above must be 0, leaving x2 = k - 1/2. Since
y = x2, we have y - k - 1/2 as the y coordinate of the intersection point. Since the y coordinate of
the center of the circle is k, the center is k - (k - 1/2) = 1/2 above the intersection point.
Chapter 6
Polynomials
6-1 To prove that these polynomials are unique, we show that if there is any pair of polynomials
(q'ix), r'ixfj which satisfy the conditions in the text for q{x) and r(x), then these must be equal to q
and r. Thus, we have
f{x) = q(x)g(x) + r(x) = q'(x)g(x) + r'(x).
Rearranging this gives
(q(x) - q'(xj)g(x) = r'(x) - r(x).
Now, if q(x) + q'(x), then the degree of the left-hand side of this equality is at least degy, while the
degree of the right-hand side is at most deg r. Since deg r < deg g, the degree of the left-hand side
is always greater than that of the right-hand side, so the polynomials cannot be identical. Thus, we
cannot have q{x) t q'(x), so q(x) = q'(x). This gives us
(q(x) - q'(x))g(x) = (0)f(x) = 0 = r'{x) - r(x),
so r{x) = r'(x) and the proof of uniqueness is complete.

6-2 Performing the synthetic division, we have
-3 32-1 1-7
-3 0 -6 21 -66
1 0 2 -7 22 -73
Hence, we find
x5 + 3*4 + 2x3 - x2 + x-7 73
= x4 + 2xz -7x + 22-
x+3 x+3
6-3 Let f(x) be the given polynomial. Since we clearly have /(1) = 0, we know x = 1 is a root.
Performing synthetic division, we find that fix) = (x - 1 )(x3 + 2x2 + 4x + 21). Since the signs of the
coefficients of the second factor are all positive, it has no positive roots. Thus we try x = -1, which
fails, then x = -3 (skipping x - -2 because it's disallowed by the Rational Root Theorem). This
works and we find fix) -{x- l)(x + 3)(x2 - x + 7). Hence, two of the roots are 1 and -3. The other
< 29 >
30 > CHAPTER 6. POLYNOMIALS
two are the solutions of the quadratic x2 - x + 7 = 0. We can't factor this, but from the quadratic
formula, we find the roots x = (1 ± 3i V3)/2.
6-4 The other roots are simply 3 + i and 4 - V2 as discussed in the text.
6-5 The first step is showing that our assertion is true for n = 1. This is quite simple, as we can
only have k = 1. The assertion states that the sum of the roots of a\X + uq is —ao/ai. Since this is, in
fact, the only root, it is indeed the sum of the roots.
Now we move to the inductive step. We assume that the assertion is true for n = m. Con¬
sider f(x) = am+\xm+l + • • • + flo- We can factor this as
f(x) = am+i(x- rm+1)g(x)
where g(x) is a degree m monic polynomial. As per our assertion, which we assume is true for all
polynomials of degree m, we can write g(x) in terms of its roots:
g(x) = xm - (n + • • • + rm)xm~l + (nr2 + • • • + rm-irm)xm~2 + • • • + (-l)”Vir2 • • • rm.
Thus, we can write f(x) as
f(x) = am+i(x - rm+1)(xm - (n + • • • + rm)xm_1 + • ■ • + (-l)"Vir2r3 • • • rm).
It is very important that you multiply this out for yourself and see that the product is indeed the
expression we have below. Carry out the product by considering the coefficient of xk in the product.
It is the result of multiplying the coefficient of xk in g(x) by rn+\ and adding this to x times the
coefficient of xfc_1 of g(x). Don't be intimidated by this product, it is very similar to multiplying
out (x + l)(x3 + x2 + x + 1). We find
f(x) = am+1 (xm+1 - (n + • • • + rm+i)xm + • • • + (-l)m+1rir2r3 • • • rm+1).
We see that f(x) satisfies our assertion, so we have completed our induction. If you don't see that f(x)
satisfies the assertion, look at the coefficient of xm. It is the negative of the sum of the roots of /(x).
Similarly, the coefficient of xm_1 is the 2nd symmetric sum of the roots and so on.
6-6 Factoring the polynomial will work, but it will take a while. Let's try to use the given
information. Let the roots be a, b, and c, where (a + b)/2 = c since one root is the average of the other
two. From the given polynomial, a + b + c = 27, or 2c + c - 27, so c = 9. Then the other two roots are
a and 18 - a, and we can use the polynomial's constant term to write fl(18 - a)(9) = 720, so a - 8 or
10. Either way, the three roots of the polynomial are 8, 9, and 10.
6-7 The desired polynomial has roots which are reciprocals of the polynomial whose roots are
a + 3, b + 3, and c + 3. This latter polynomial is
f(x - 3) = 3(x - 3)3 - 14(x - 3)2 + (x - 3) + 62 = 3x3 - 41x2 + 166x - 148.
(We could also use our synthetic division technique to determine this polynomial.) We find the
polynomial whose roots are the reciprocals of those of /(x - 3) by reversing the coefficients of
f(x - 3), finding
-148x3 + 166x2 - 41x + 3.
The sum of the roots of this polynomial is
1 1 1 _ 166 _ 83
a 3 b -\- 3 c+3 —148 74
6-8 If n < k, then several of the a, in ansk + an-\sk^\ + • • • + kan-k are 0 since i < 0. Eliminating
these terms we are left with ansk + a„_isfc_i + • • • + a0sk_n = 0.
6-9 Use Newton sums again:
2si + 3 = 0, si =
5
2s2 + 3si + 2(1) — 0, s2 = —/
39
2s3 + 3s2 + Si + 3(-4) — 0 s3
~8 '
84. Let f(x) - x13 + 1. The remainder is /(1) = 2.

85. First we use the Rational Root Theorem to decide what numbers could possibly be roots.
After trying a few (usually start with the integers) using synthetic division, we finally find that y = 3
is a solution and
f(y) = 2y4 - 9y3 + 14y2 + 6y - 63 = (y - 3)(2y3 - 3/ + 5y + 21).

Now we go on trying to find the roots of 2y3 - 3y2 + by + 21. Eventually you should find that y = -3/2
is a solution (you might have found that first) and that
m = (y - 3)(y + 3/2X2/ - 6y + 14),

so we can use the quadratic formula on the final quadratic to get the last two roots. Hence the roots
are 3, -3/2, and (3 ± i Vl9)/2.
86. Let f(x) = x3 + 3m2x2 + mx + 4. The binomial x + 2 is a factor of this polynomial if and only if
/(-2) = 0, so we have
-8 + 12m2 - 2m + 4 = 0.
Solving this equation for m, we find that x + 2 is a factor of f{x) if and only if m = -1/2 or m = 2/3.
87. The nth roots of 1 are solutions to the polynomial xn - 1, or xn - 1 = 0. The product of the
roots of this polynomial is (-1)”(-1) = (-l)n+1.
88. Let the doubled roots be r and s. From the coefficient of x3, we have 2r + 2s = 16, or r + s - 8.
From the coefficient of x2 we find r2 + s2 + 4rs = 94. Since p = -(2r2s + 2rs2) = -2rs(r + s) and q = r2s2,
if we find rs and r + s, we find p and q. We already have r + s, and we can find rs by squaring r + s = 8
to get r2 -l- 2rs + s2 - 64. Subtracting this from r2 + s2 + 4rs = 94, we find 2rs = 30, so rs = 15. Hence,
p = -30(8) - -240 and q = 152 = 225, so p + q = -15.
89. From our given information, we have
f(-7) = -a77 -b73 -c7 -5 = 7,
so ci77 + b73 +c7 = -12. Thus, /(7) = a77 + b73 + c7 - 5 = -12 - 5 - -17.
90. Let the roots be r, s, and t. From the coefficient of x2, we have r + s + t = —[(—12)/4] = 3.
Since r + s = 0, we find t = 3. Hence, c = rs + st + tr = rs + 3(r + s) = rs and d - -rst = -3rs. Thus,
d/c - -3rs/rs - -3.
91. There are a lot of really long ways to do this, but here we'll show you a clever method
involving the understanding of polynomials. Let f(x) be the given polynomial. Note that /(1) =
1 + A + B + C + D, so if we can find /(1), we can find the answer. The two roots 3 ± V2 have sum 6
and product 7, so they're the roots of x2 — 6x + 7. Similarly, —3 ± i V2 are the roots of x2 + 6x + 11,
so f{x) = (x2 - 6x + 7)(x2 + 6x + 11) and /(1) = 2(18) = 36. Since as we noted above we have
A + B + C + D- /(1) - 1, our answer is 35.
92. Since P(x) has only terms of odd degree, it is an odd function. (Remember odd functions
from Volume 1?) Thus, P(-x) = -P(x). Since P(3) = 6, we have P(-3) = -6. Now we finish this as
we did a similar example in the text:
P(x) = (x2 - 9)g(x) + r(x).
Since deg r(x) < deg(x2 - 9), r(x) = ax + b, and putting 3 and -3 in the above equation for P(x), we
find 6 = 0 + 3a + b and -6 = 0 - 3a + b, so b = 0 and a - 2. The desired remainder is 2x.
93. Note that for any even number, all the terms in the polynomial are even except the last term.
Hence, since p(0) = ao is odd, p(2N) is odd for all N. Similarly, all p(2N + 1) have the same parity
(meaning they're all odd or all even), so since p( 1) is odd, p(2N + 1) is odd for all N. Since for all
integers n, p(n) is odd, there are no integer roots of p(x) (since 0 is even).
94. Since we have a quartic (degree four polynomial) divided by a cubic, the quotient must be a
simple linear term like x + a. Multiplying this and the given cubic we have
(x + a)(x3 + 3x2 + 9x + 3) = x4 + (3 + a)x3 + (3a + 9)x2 + (9a + 3)x + 3a.
Since this equals the given quartic, we have
x4 + 4x3 + 6 px2 + 4 qx + r = x4 + (3 + a)x3 + (3a + 9)x2 + (9a + 3)x + 3a.
Matching coefficients of x3, we have a = 1. Matching coefficients of x2, we have 6p = 12, or p = 2.

From v we get q = 12/4 = 3, and from the constants, r - 3. Hence, the answer is (2 + 3)3 = 15.
95. The desired expression is the sum of the squares of the reciprocals of the roots. We find the
polynomial whose roots are the reciprocals of the roots of the original polynomial by reversing the
order of the coefficients of the original polynomial, or
-7x3 + 5x2 - 6x + 1 = 0.
Our desired sum is the sum of the squares of the roots of this new polynomial. We use Newton's
sums to get sums of powers of roots, so fl3si+fl2 = 0 gives si = 5/7. Thenextoneisfl3S2+fl2Si+2fl! = 0,
so -7s2 + 25/7 - 12 = 0, and s2 = -59/49.
96. We'll approach this in much the same way we proved that complex roots of polynomials with
real coefficients always come in conjugate pairs. The given x is a solution to the polynomial, so we
have
f(x) - C4x4 + ZC3X3 + C2X2 + iC\X + Co = 0.
Taking the conjugate of both sides, we have (remembering that i = —i)
0 = C4X4 + ZC3X3 + C2X2 + ic\x + Co

C4X4 + IC3X6 + C2XZ + IC\X + Co
= C4(x)4 - iCo{x)6 + C2(x)2 - iC\X + Co
Make sure you recognize this last line as /(-x); if you don't, try putting —x in the original polynomial.
Hence, if x is a root, so is —x, so —(a — hi) = —a + bi is a solution.
97. Let f{x) = x8. Since /(-1/2) = (-1/2)8, the remainder upon dividing x + 1/2 into /(x) is
T\ - (-1/2)8 = (1/2)8, so we can write
f(x) = x8 = (x + l/2)qi + (1/2)8.
Thus,
(x + 1/2)^ - x8 - (1/2)8
= [x4 - (1/2)4][x4 + (1/2)4]
= - (l/2)2][x2 + (1/2)2][x4 + (1/2)4]
= [x - l/2][x + 1/2][x2 + (l/2)2][x4 + (1/2)4].
Dividing by x + 1 /2, we get
<JiM = [* - 1/2][V + (l/2)2][x4 + (1/2)4].
Thus, our remainder upon dividing î(x) by x + 1/2 is £71 (—1/2) = -1/16.
98. This one's pretty tricky. We can rearrange the given equation as (x+l)(x+4)(x+2)(x+3) + l = 0,
which has the advantage that the product of the first two terms has the same x coefficient as the
product of the last two terms:
(x2 + 5x + 4)(x2 + 5x + 6) + 1 = 0
If we let t be the x2 + 5x common to our quadratics, we have (f + 4)(f + 6) + l = f2 + 10f + 25 = (f + 5)2 = 0.

Thus, t = -5 and solving x2 + 5x = -5 gives the two roots (—5 ± V5)/2, both of which are double
roots of the polynomial.
99. We can get the sum of all the by just letting x = 1:
(1 + 1 + l)n — do + U\ + #2 ’ ’ ' + Z?2n-
Now we need to get rid of the odd numbered terms. This suggests putting in x = -i, since then the
odd numbered terms will be subtracted:
(1 — 1 + l)n — Uq — U\ + U2 — #3 + • • • + Cl2n•
Adding this to the first equality, we find
3" + 1 - 2(flo + #2 ^4 4- • • • + Cl2n)-

Thus, the desired sum is (3n + l)/2. Remember the 'sticking numbers in' method for attacking
polynomial problems. It is often very useful, especially when we are looking for the sum of
coefficients of a polynomial, which is just the polynomial evaluated at x = 1.
100. As we've done many times before, we write
x203 - 1 = (x4 - l)g(x) + r(x),
where deg r(x) < 4, so we can write r(x) = ax3 + bx2 + cx + d. We put in the roots of x4 -1 for x, making
(x4 - l)g(x) = 0. These roots are 1, -1, i, and —i (since x4 - 1 = (x2 - l)(x2 + 1)); trying each of these in
turn, we have
0 — a+b+c+d
—2 —a + b — c + d
-i - 1 = -ai -b + ci + d
i-1 - ai -b - ci + d.
If we add the first two equations, we get b + d = -1. Equating real parts of the last equation, we have
d-b - -1. Adding these two gives d - -1, so b = 0. Similarly, subtracting the second equation from
the first gives a + c = 1 and equating imaginary parts of the last equation gives a- c = 1. From these
we find a = 1 and c = 0. Hence the remainder is x3 — 1.
101. Seeing the expression x2 - 3x - 2 in the first equation, we write x2 - 4x - 2 = 0 as x2 - 3x - 2 = x.
If x2 - 3x - 2 = x, we have
(x2 - 3x - 2) - 3(x2 - 3x - 2) - 2 - x = x2-3x-2-x

= x - x = 0,
so all solutions to x2 - 3x - 2 = x are solutions to the given polynomial. Multiplying out the given
polynomial, we find it is x4 - 6x3 + 2x2 + 20x + 8 = 0. Since we know x2 - 4x - 2 is one factor, we can
write
x4 - 6x3 + 2x2 + 20x + 8 = (x2 - 4x - 2)(x2 + ax + b)

= x4 + (a - 4)x3 + (b -2- 4 a)x2 - (4b + 2 a)x - lb.
Matching coefficients of the polynomials, we find a = -2 and b = -4, so the solutions of the initial
polynomial equation are the roots of x2 - 4x - 2 and x2 - 2x - 4, or 2 ± V6 and 1 ± V5.
102. Since P is a polynomial, it is either constant or it takes on infinitely many values. For the
first case, we have c = ck, so either k = 1 and c is arbitrary or k > 1 and c = -1 (if k is odd), c - 0
or c = 1. For non-constant P(x), let z = P(x). We then have P(z) = zk for infinitely many values of z.
Hence the polynomial P(z) - zk has infinitely many roots but a finite degree, so the polynomial must
be zero everywhere, or P(z) - zk = 0, so that P(x) = xk is the only family of non-constant polynomials
which solve the given equation.
103. From the coefficient of x3, a + b + c + d = 0, so a + b + c = -d, and (a + b + c)/d2 = -1/d.
Proceeding similarly, the roots of the desired polynomial are -1/a, -1/b, -1/c, and -1/d. The
polynomial with roots 1/a, 1/b, 1/c, and 1/d is found by reversing the coefficients of the original
polynomial, to get -3x4 - mx3 + 1 = 0. The equation which has roots which are c times those of
f(x) = -3x4 - mx3 + 1 is f(x/c) (remember this from the chapter?). Hence, one polynomial with the
given roots is f(-x) = -3x4 + mx3 +1; we want the leading term to be 3, so we multiply by -1, giving
3x4 - mx3 - 1.
104. Since g(x)h(x) = f(x), we have, for all a\, 02, ..., an, g(ai)h(ai) - /(«,-) = -1. Since g(cii) and
h(ai) are integers (why?), the two must be 1 and -1. Hence, g(dj) + h{ai) - 0 for each i, so that q(x) -
g(x)+h(x) = 0 has n distinct roots. Since g(x)h(x) - f(x), wehavedegg(x)/z(x) = deg f(x) = n. Because
g(x) and h(x) are nonconstant, neither g nor h has degree larger than n- 1. Hence, q(x) = g(x) + h(x)
has no terms of xn or higher degree. Therefore, either q{x) is the zero polynomial or deg^(x) < n. If
q{x) = 0 for all x, then g(x) = -h(x), and the leading term of g(x)h(x) has a negative coefficient, which
cannot match the leading term of /(x). For nonzero polynomials q(x), the fact that deg q{x) < n means
that q(x) has less than n factors and thus less than n roots. This is a contradiction to the above proof
that g(x) + h(x) has the n distinct a, as roots. Hence, we cannot factor /(x) as stated in the problem.
Chapter 7
Functions
7-1 We let the inverse function be g(x), so that f(g(x)) - \Jg{x) = x. Cubing both sides, we have
g(x) = x3.
7-2 We should have f° o fl (x) = fl (x) if the composition exponents are to add properly. But
f° o fl(x) = f°(fl(x)); setting this equal to fl{x) yields f°{fl(x)) = f1 (x). The only way this can be
satisfied is if /°(x) = x. This is not the same as [/(x)]° = 1.
7-3For the first, assume there are two numbers x and y such that /(x) = /(y), or x3 = y3, or
(x - y)(x2 + xy + y1) = 0. The only way this can be solved is if x = y or (using the quadratic formula)
x = -y(l ± V3 f)/2. Since the latter is ruled out because not both x and y can be real, we must have
x = y. Thus /(x) = x3 is 1 : 1.
The second is not 1 : 1 since, for example, g(2) = g(-2) = 2.
The third is not 1 : 1 since, for example, h( 1) = h( 1.5) = 1.
The fourth is 1 : 1 since if j(x) = j(y)r we have x/2 = y/2, or x = y.
7-4 With /(x) = sin x and g(x) = cos x, we have /(x + y) = sin(x + y) = sin x cos y + sin y cos x =
f(x)g(y) + f(y)g(x)/ so ê first identity holds. We also have [/(x)]2 + [g(x)]2 = sin2 x + cos2 x = 1, so
the second identity holds as well.
7-5 The first identity is |xy| = |x||y|. We look at three cases. If both x and y are positive,
then xy is as well, so we have |xy| - xy - \x\\y\. If exactly one of x and y is negative, then xy is
negative, so we have |xy| = -xy = |x||y|. If both x and y are negative, then xy is positive, so we have
|xy| - xy - (-x)(-y) = |x||y|. In all three cases the identity holds.
We can easily disprove the second identity by taking x = 1 and y = -1. This yields f(x + y) =
\x + y\ - |1 ~ 1| = |0| = 0 and /(x) + /(y) = |x| + |y| = |1| + | - 1| = 1 + 1 = 2. The two are clearly not
equal.
The third identity is false because x = -1 yields /(/(x)) = |1| = 1 ± -1.
7-6
i. If n and x are integers, then nx and x are integers, so [nxj = nx and n|_xj = nx—the identity
holds.
< 36 >
ii. Let x = 1.5 and n - 2. Then \ nxJ = |_3J = 3 and n|_xj = 2[1.5J = 2(1) = 2. The two are not
equal, so the identity does not hold.
iii. This case allows the same values of n and x as the previous case, so the identity cannot hold
in this case either.
7-7 f(x) = x satisfies many, many identities; see how many you can come up with. A few are
f(x + y) = f{x) + /(y), f{nx) = nf(x) for any n, f(xy) = /(x)/(y), and f(f(x)) = x.
7-8 For /(x) = cx, we have f(xy) = cxy and xf(y) - xcy. Since cxy = xcy, the identity holds for
any c.
7-9 Substituting y = 0 yields f(x) + f(x) = 2f(x) = lx2. Thus /(x) = x2 is the only possible
solution. Trying this solution in the original equation, we have f(x + y) + /(x - y) = 2x2 + 2y2, not
2x2 - 2y2. Thus our only candidate fails the test, and there are no solutions.
7-10
i. Let's start iterating the function. We have
f(x) = x+ -
X
x4 + 3x2 + 1
/(/M) = /M + 7S)=* + ; + i7i = - X3 + X
This, it is fairly clear, is not going anywhere. (Not quite a proof.)

ii. We have
x
fix) =
x-1
/(/(*)) = = X.
-1
x-1 l
Thus, /(x) = x/(x - 1) is cyclic and has order 2.

iii. We have
fix) =
1 + X
1 1
X +
/(/W) 1 + fix) 1 + jlj x+2
/(*) +1 _ ris + l _ 2 + X
/mm - f(x)+2 Oj + 2 3 + 2x
and it is clear that this function is again not cyclic.

iv. We have
f{x) = 1-x
/(/(x)) = l-(l-x) = x
and thus the function is cyclic and has order 2.

7-11 For h(x) = x2, we get /(x) = (x - 1/2)2 + x. With this function /, we have
/((l + <i)/2) = (a/2)2 + {a/2 +1)

38 > CHAPTER 7. FUNCTIONS
and
/((1 - a)/2) +a = {-a/2)1 + (1 - a/2) +a = (a/2)2 + (a/2 + 1).
Thus, we have /((1 + a)/2) = /((1 - a)/2) + a, so f(x) satisfies the given functional equation.
105. Substituting y = 2x, we have /(y) = 2/(2 + y/2), so 2/(y) = 4/(2 + y/2) = 8/(4 + y). Thus
2f(x) = 8/(4 + x). '
106. We just write f(g(x)) = /(2x) = 4/(2x - 1) and #(/(*)) = g(4/(x - 1)) = 8/(x - 1). Setting the
two equal and solving the resulting linear equation, we find x = 1/3 as the only solution.
107. Let a = 17/2 and x = 2, so that /(17) = 17/(2)/2 = 17(5)/2 = 85/2.
108. Substituting in y = 0 yields 2/(x) = 2, so the only possible candidate is f(x) = 1. Substituting
this into the given functional equation yields 2 - y + 2, which is not true for all y. Thus the given
functional equation has no solutions.
109. Rearranging, we have xf(x) = yf(y). If this is to hold for all x and y, we must have xf(x) = c
for some constant c. Thus f(x) = c/x, for constants c, are the only possible solutions. Substituting
this into the equation, we find that f(x) - c/x is a solution for any c.
110. To find y = f~l(-2), we write -2 = f(y) = l/(y + 2), to get y = -5/2. We then have
?(/“'(-2)) = g(-5/2) = 3.
111. We write y = f(x) = x/(l - x), so that x - y/(l + y). Examining the answers, we see that
y/( 1 + y) can be rewritten as —/(—y).
112. Let's try them. For the first, the composition of f(x) = ax+ b and g(x) - cx + d is a(cx + d) + b =
acx + (ad + b), which is clearly another function in the set. For the second, the composition of
f(x) = ax2 + ■ ■ ■ and g(x) = bx2 + • • • is a(bx2 + • • • )2 + • ■ • = ab2xA + ■ • •, which is clearly not in the
set. For the third, f(g(x)) is clearly a polynomial if / and g are both polynomials. For the fourth,
f(g( 12)) = /(0), which is not necessarily 0, since 0 is not necessarily a root of /; thus 12 is not
necessarily a root of / o g. Thus two of the given sets, the first and third, have the desired property.
113. Let y = ax; then f(y) = f(ax) = logfl x = loga(y/a) = logfl y - 1, or f(x) = logfl x - 1.
114. We use the cyclic function \/x, substituting x —> 1/x in the given equation to get the pair of
equations
21/m -7/(1) = 12*
21/Q)-7/M = 12/*.
Adding three times the first equation to the second to eliminate /(1/x), we have 56f(x) = 36x + 12/x,
or f(x) = (9x + 3/x)/14. Substituting this into the functional equation, we find that it is a solution; it
is the only solution.
115. To find /(1/2), we need to find x such that g(x) - 1 - x2 = 1/2. Solving, we find x = 1/ yj2.
Thus/(l/2) = /(g( 1/ V2» = (1 - (1/ V2)2)/(l/ V2)2 = (1 - l/2)/(l/2) = 1.
116. From /(*) = f(2a)x, we have /(2a) = /(2a)2". Dividing by /(2a), we find 1 = /(2a)2"-1, or
2a - 1 = 0, or a = 1/2.
117. Substituting a = b = 0 yields /(0)2 = /(O), so that /(0) equals 0 or 1. Since f(x) > 0 for all x,
only /(0) = 1 is allowed, so I is true.
Substituting b = -a, we have f{a)f{-a) - f(a - a) = /(0) = 1, so that /(a) = l//(-a), and II is also
true.
Substituting b - 2a, we have f(a)f(2a) = /(3a), or /(a)3 = /(3a). We thus have /(a) = f {3a), and
III is true.
One solution to the given equation is f(x) = 1 for all x. (Do you see why?) In this case, even
though 1 > 0, we have /(1) = /(0) rather than /(1) > /(0). Thus IV is not always true.
In sum, I, II, III are the true statements.
118. We write sec2 0 = x/(x - 1), yielding x = sec2 0/(sec2 0 - 1) = sec2 0/ tan2 0 = 1/ sin2 0. Then
/(sec2 0) = f(x/(x - 1)) = l/x = 1/(1/ sin2 0) = sin2 0.
119. For a number x to be in the domain of g o / = g(f(x)), x must first be in the domain of /, so
that x > -2. Moreover, /(x) must be in the domain of g, so that /(x) < 5, or x2 + x - 1 < 5. We thus
have x2 + x - 6 < 0, which yields (x + 3)(x - 2) < 0. The solution to this inequality is -3 < x < 2.
Finding all joint solutions of x > -2 and -3 < x < 2, we have -2 < x < 2 as the domain of g o /.
120. Setting t - x yields /(2x) - /(0) = 4x2. Substituting x = y/2, we then have f{y) - /(0) = y2.
Letting /(0) be some arbitrary constant a, we have /(x) = x2 + a. Testing this function, we find that it
satisfies the given functional equation for any a.
121. Substituting in x = 5 yields /(-4) - 4/(5) = 5; substituting x = -4 yields /(5) + 5/(-4) = 5.
Subtracting five times the first equation from the second equation yields 21/(5) = -20, so that
/(5) = —20/21. (Note that this solution is possible because of the cyclic nature of the function 1 - x.
Can you use this to find /(x) for all x?)
122. Substituting x = a into the equation /,(x) = 2afb(x - a), we have /,(a) = 2fl/,(0) = 2ab. Thus
fb{x) = 2xb, so fc{2x) = 22*c = (c/b2)[fb(x)]2.
123. We have
3 fix).
124. We have /(4x) = 4x/(4x - 1). To express this as a function of /(x), we solve /(x) = x/(x - 1)
for i,sor = /(*)/(/« - 1)- Then /(to) = 4 [/(*)/(/(*) - l)]/[4/(x)/(/(r) - 1) - 1] = 4/W/(3/W +1).
125. We substitute r = 2 to get
/(2) + 2/(-l) = 2.
40 > CHAPTER 7. FUNCTIONS
Seeing /(-1) above, we substitute x = -1 to get
/(-l) + 2/(l/2) = -1,
and then take x = 1/2 to get

/(1/2) + 2/(2) = 1/2.
We can solve these three equations by subtracting twice the second equation from the first to get
/(2) - 4/(1/2) = 4, then adding four times the third equation to get 9/(2) = 6, or /(2) = 2/3. (Again,
this solution is possible because of a cyclic function. What is the cyclic function? Can you use it to
solve for /(x)?)
126. Setting t = 0 we have /(x) = /(x) + /(0) + 2 ^/(x) ^/(0). This becomes 2 ^/(x) \Jf{0) - -/(0),
so squaring yields 4/(x) = /(0) as the only solution. Letting /(0) = a, an arbitrary constant, we
have /(x) = a/4. Is this a solution for any a/4? Substituting in the functional equation yields
a/4 = a/4 + a/4 + 2(a/4) = a, so that a = 0. Thus the only solution is /(x) = 0.
127. Substituting x = u - 1 and y = 1 yields f(u) - /(1) = (w - 1)/«, or f(u) = (u - l)/u + /(1).
These functions, for any constant /(l) = c, are the only candidates. Trying f(u) = (u - l)/u + c in the
functional equation, we have /(x + y) - /(y) = (x + y - l)/(x + y) + c - (y - l)/y - c = x/y(x + y). Thus
the function given is a solution for any c.
128. We have /(1985) = -1984 - 2/(1984), /(1984) = 1983 - 2/(1983), /(1983) = -1982 -
2/(1982),..., /(2) = 1 - 2/(1), /(1) = /(1986) = 1985 - 2/(1985). Adding these up, we have
/(1) + /(2) + • • • + /(1985) = 1- 2 + 3- 4h-+ 1985 - 2(/(l) + /(2) + • • • + /(1985)).
Rearranging, we have
3(/(l) + /(2) 1-+ /(1985)) = 1-2 + 3-4 +••• + 1985

= 1 + (3 - 2) + (5 - 4) + (7 - 6) + • • • + (1985 - 1984)
= 1 + 1984/2 = 993,
so the desired sum is 993/3 = 331.

129. Rearranging, we have /(I - x) - (1 - x) = /(x) - x. Letting /(y) - y = g(y), we have
g(l - x) = g(x). We can write this as g( 1/2 - (x - 1/2)) = g(l/2 + (x - 1/2)); this suggests writing
h(y) - g( 1/2 + y), so h(y) = /z(-y) and /z is an even function. We claim that for any even function h,
the corresponding function / is a solution. Starting from h(x), we construct g(x) = h(x - 1/2), and
correspondingly /(x) = g(x) + x = h(x - 1/2) + x. Substituting /(x) into the functional equation, we
have /(I - x) = h( 1/2 - x) + (1 - x) = h(x - 1/2) + (1 - x) = /(x) + 1 - 2x as long as h is even. Thus for
any even h, f(x) = h(x - 1/2) + x is a solution to the equation.
130. The given functional equation is
/(z) + / (coz + a) = g(z). (1)
Recalling that co2 + co + 1 = 0 for any third root of unity not equal to 1, we think to go for a cyclic
function-type simplification. Thus we substitute coz + a for z in the given equation to get
f(coz + a) + f(co2z + coa + a) = g(coz + a). (2)

We then substitute coz + a for z again in (2) to get
f(co2z + coa + a) + f(co3z + co2a + coa + a) = g(co2z + coa + a),
which because co2 + co + 1 = 0 and co3 = 1 becomes
/(oj2z + coa + a) + f(z) = g(co2z + am + a). (3)
Equations (1), (2), and (3) are linear in terms of f(z), f(coz + a), and f(co2z + aja + a). In fact, adding
the three equations gives
/(z) + f(a)Z + a) + f{co2z + coa + a) = [g(z) + g(coz + a) + g(co2z + coa + a)\/2.
Subtracting (2) from this gives
/(z) = [g(z) - g(coz + a) + g(co2z + coa + a)]/2.
Since this equation specifies /(z) specifically in terms of g(z), this is the one and only one /(z) which
solves the given functional equation.
Chapter 8
Taking it to the Limit
8-2 As n —» oo, the denominator of —r goes to oo for any k > 0. Since the numerator stays at 1,
n
1
the fraction tends to 0. Thus lim —r = 0 for any positive k.
n—>oo pi
8-3 The upshot of our discussion in the text is that we need only consider the leading terms of
each polynomial. We thus have
2x4 2 1
i. lim —j = lim 7 = -.
X—>oo 4x^ x—>oo 4 2
2x3 2
ii. lim —-j = lim — = 0.
X—>oo 4j4 x—>oo 4x
2.r5 2x
iii. lim —j = lim — = oo.
X—»oo 4x4 X—>oo 4
8-4 The functional definition is that for some N, there exists no x such that |/(x)| > |N|. The
sequence definition is that for some N there exists no n such that \an\ > \N\.
8-5 Yes. Either the sequence/function's values exceed N for any finite N, or they don't. In the
latter case the function is bounded, in the former case unbounded.
8-6 The "official" definition is rather tricky, but the essence of it is that no matter how small a
neighborhood (L - e, L + e) we take around the limit, for all sufficiently large x the function f(x) will
lie in the neighborhood.
8-7 If the degree of the top is less than or equal to the degree of the bottom, the rational function
is convergent. If the degrees of the top and bottom are the same, the limit will be nonzero.
8-8 Using the calculator, we find sin(l/1000)/(l/1000) = .9999998. It seems that the limit is
tending to 1.
8-9 For x < 0, \x\ = -x, so that
X X
lim — = lim — = lim (-1) = —1;
x—>0 \X\ x—>0~ X x—>0~
< 42 >
for x > 0, |x| = x, so that

X X
lim — = lim - = lim (1) = 1.
x->0+ |x| x->0+ X x-»0+
Since the limits from the left and from the right are different, the overall limit does not exist.
, sin(l/x)
8-10 It's more or less the same. In that case, though, we would have been looking at lim ^ ^ -;
there is no difference.
8-11 We have (1 +1/10)10 = 2.594 and (1 + l/lOO)100 = 2.705. It seems that the amount of interest
is tending to a limit.
131. We need to fill the hole at x = 2 by letting k be the limit as x tends to 2. We have
V2x + 5 - Vx + 7 (V2x + 5 - Vx + 7)( V2x + 5 + Vx + 7)

lim--- lim
x—>2 X — 2 x—>2 (x — 2)( V2x + 5 + Vx + 7)
x-2
lim
x—>2 (x - 2)( V2x + 5 + Vx + 7)
lim ■ -- ..- - = -.
V2x + 5 + Vx + 7 6
132.
i. The denominator goes to oo as the numerator oscillates between -1 and 1, so the limit is 0.
ii. The top of the fraction can be factored as (x + 2)(x2 - 2x + 4). Dividing x + 2 out of the top
and the bottom the limit becomes lim (x2 - 2x + 4) = 4 + 4 + 4 = 12.
X—»-2
iii. The bottom factors into (Vx - 4)( Vx + 4). Dividing the common factor of Vx - 4 from the
top and bottom leaves lim l/( Vx + 4) = 1/8.
x—>16
133. Dividing the polynomials, we find that
x2 - x - 2 4
/(x) = x-3+
x+2 x + 2’
As x —> +oo7 the fraction vanishes, and the graph tends to the line equation y = x - 3.
134. We have
(V4x2 + 5x - V4x2 + x)( V4x2 + 5x + V4x2 + x)

lim (V4x2 + 5x - V4x2 + x) = lim
X—>00
X—>oo
V4x2 + 5x + V4x2 + x
4x
lim - --
x->0° V4x2 + 5x + V4x2 + x
lim 4
X—>oo V4 + 5/x + V4 + 1/x
4
1.
4
44 > CHAPTER 8. TAKING IT TO THE LIMIT
135. We have
lim sin2 x/x lim sin x/x lim sin x = (1)(0) = 0.
x->0 Lx—>0 Lx->0
136. We have
lim 6 cot 6 lim 0/ sin 6 lim cos 6 7T-. nin lim cos 6 1.
e->o Leô L0->O lim sm 0/6 e->o
e-> o
137. We have
,. 1 - COS X sin2* .. sm2 x.. 1

lim- = lim = lim-lim —-r - (0)(1) = o.
x—>0 X x—>o x{l + cos x) x—>0 X x^0 (1 + COS X)
as desired.
138. Dividing the polynomials, we have
X13 X
xz - 1 = X+ ~2
~ 2—T
xl-7-
-l
As x —> ±oo the function tends to the slant asymptote y = x; as x —» ±1 the function diverges, so we
have the vertical asymptotes x = 1 and x = -1.
139. Dividing the top and bottom by x, we have lim 6/ yl9 + 17/x; thus the limit is 6/3 = 2.
J x—»oo
140. Dividing the top and bottom by x we have
V3 + 17/x
lim
X—>oo 1
Chapter 9
Complex Numbers
9-1 The complex number \ + \i corresponds to the point (|, |) in the plane and -\ + V2 i
corresponds to (-|, V2).
9-2 We can quickly analyze the graphs of each case by noting Re(z) = x and Im(z) = y, so that
Re(z) = 1 corresponds to the vertical line x - 1, Re(z) + Im(z) = 1 to the line x + y = 1, and Im(z) < 1
to y < 1 (or the half-plane below y = 1).
9-3 First, z corresponds to (3,2). Since z corresponds to (3, -2), it is the reflection of z in the x
axis. (This is always true of a complex number and its conjugate.) We find z2 is represented by the
point (5,12) and z - 1 by (2,2), so that z - 1 is a translation 1 unit to the left of z.
9-4 We graph these polar representations of complex numbers just as we would polar coordi¬
nates on the real Cartesian plane. The first point is in the first quadrant, while the second is on the
negative x axis (at (-3,0) to be exact).
9-5 The curve r - 1 represents all points which are 1 away from the origin; thus, the curve is
a circle with radius 1 centered at the origin. As discussed in Volume 1 and in our section on polar
coordinates, 9 = n/3 describes a line through the origin.
9-6 Applying the properties discussed in the chapter, we have
|(12 + 5z')(7-240| = 112 + 5z| |7 - 24z|

= Vl44 + 25 V49 + 576
= 325.
9-7 Let w = 5 and z = -5. Then \zv + z\ = \w\ + |z| gives 0 = 10, which is clearly false.
9-8 If z = cw and w has the polar form (r, 0), then z = (cr, 6) (why?), so that the equality condition
in polar form is that z and w have the same angle 6 in polar form.
9-9 Since \-w\ - \w\, we can replace w by -w in the inequality |z + w\ < |z| + \w\ to yield
\z-w\< |z| + \-w\ = |z| + \w\, as desired.
< 45 >
46 > CHAPTER 9. COMPLEX NUMBERS
9-10 Since i has the polar representation (1, n/2), multiplying the complex number z by i leaves
the magnitude unchanged but adds rr/2 to the angle of z. This corresponds to a 90° counterclockwise
rotation about the origin.
9-11 Are you convinced?
9-12 Since the polar form of i is (1, n/2), then in = (1, n/2)n = (ln, nn/2) = (1, nn/2).
9-13 Let the reciprocal be w - (r,<p). Since zw = 1 - (1,0), we have zw = (1 • r, 6 + (p) = (1,0).
Hence, r - 1 and <p = -6 and the reciprocal of z is (1, -6). Thus, 1/z is the reflection of z in the x axis.
Does this make sense? What is the relationship between z and 1/z for points not on the unit circle?
(It isn't a simple reflection!)
9-14 Since w - ~y + is on the unit circle, the powers of w are all on the unit circle (since
the magnitude of w is 1, all powers of w have magnitude 1). Since w = (1,45°), there are 8 distinct
powers of zu, because w9 = (1,405°) = (1,45°) brings us back to w and we start all over. (The same
holds for the negative powers, except these powers proceed clockwise around the circle rather than
counterclockwise.) For z = \ + i\, the magnitude is less than 1, so the positive powers get closer
and closer to 0, while the negative powers get larger and larger. Again, the positive powers go
counterclockwise (from 45° to 90° to 135°, etc.) and the negative clockwise.
9-15 First we write -4 V2 + 4 V2 i in polar form, as z = (8,135°). To raise z to the 3/4 power, we
first cube it then take the fourth root. Hence we have
z3'4 [(8,135°)3]1/4
(512,405°)1/4
(512,45°)1/4
360°k
(5121/4, ^ +
where k takes the values 0, 1, 2, 3. Simplifying \Z512 as 4 y/l, we have the following as our four
values (in polar form) of (-4 V2 + 4 V2 i)3^4:
4/r 45° 4 r- 405°, 4 r- 765° 1125°

(4V2,—), (4V5,—), (4V2,—-),
4 '
9-16 For sin2 x + cos2 x, we evaluate the two squares as
' eix _ e-ix \ 2 e2ix _ 2(fi + e~2ix _e2ix _ e~2ix + 2

sm x =
2i -4
and
e,x + e *x \ 2 c2'x + 2e® + e 2,x g2ix e 2ix + 2
COS X =
Adding these expressions for cos2 x and sin2 x clearly gives sin2 x+cos2 x = 1. For sin 2x = 2 sin x cos x
we multiply our expressions for sin x and cos x:
f4x — £ ix gix _|_ £ ix (3'iX_c 2^x I g2/x_^—2ix

sm x cos x =
2i 4i 2i
This final expression equals \ sin 2x, proving the assertion. (If you don't see that the last expression
above equals \ sin 2x, try putting 2x in our formula for sinx.)
9-17 Using our formulas for the hyperbolic sine and cosine, we have
2 2 e2x + 2 + e~2x e2x - 2 + e~2x e2x + 2 + e~2x - e2x + 2- e~2x

cosh x - sinh x =-=-= 1.
4 4 4
Try to show that sinh 2x-2 sinh x cosh x and cosh 2x = cosh2 x + sinh2 x.
9-18
We can equate the imaginary parts of the expression for cos 30 + i sin 30 in the previous
example, yielding
sin 30 = 3 cos2 0 sin 0 - sin3 0

= 3(1 - sin2 0) sin 0 - sin3 0
= -4 sin3 0 + 3 sin 0.
9-19 Since emd - cos nO + i sin n6, the sum on the right is clearly the imaginary part of the sum
i
on the left. This enables us to write the unwieldy sum YksinnO in terms of the geometric series
n=1
j
Y kem6, which is much easier to evaluate.
n-1
9-20The cube roots of unity are e° = 1, e2ln^ = -1/2 + i V3/2, and c4m/3 = -1/2 - i V3/2. The
, -,
fourth roots of unity are e° = 1 eln^2 = i, eln = 1 and e3m//2 = —i.
9-21 If -1 is an nth root of 1, then the point -1 must be on the regular polygon formed by the
nth roots. Since there are the same number of vertices of this polygon above the x axis as below,
and there are two vertices on the axis (if there's a vertex at (-1,0)), there must be an even number of
vertices of the polygon. Hence -1 is an nth root of unity if and only if n is even.
9-22
Since 5(2n)/17 > tt/2 > 4(2n)/17 and 9(2n)/17 > n > 8(2n)/17, the roots e2nkl117 are in the
second quadrant for k = 5,6, 7, and, 8, so there are 4 roots in the second quadrant.
141. Applying our properties from the chapter we have
7 - 24z’ _ [7 - 24z| _ V49 + 576 _

4 + 3z “ |4 + 3z| ” V16 + 9
142. We could multiply out the products or use polar form, but here we'll use clever algebraic
notation because we notice that 2 - 2i is twice the conjugate of i + 1:
(1 + i)4( 2 - 2i)3 = (i + 1)4(23)(1 - i)3

= 8(1 + i)[(l + 0(1 - i)]3
= 8(1 + i)(2)3 = 64 + 64i.
Try using polar form to complete this problem.

143. As mentioned in the text, the product of the nth roots of 1 are the solutions of the polynomial
equation xn - 1 = 0. From our discussion in the chapter on polynomials, the product of the roots of
this equation is (-1)M(-1) = (—1)”+1.
144. Let w = 2 + z. We evaluate f(zv), f2(w), f3(w), and look for a pattern. We find /(2 + z) =
(3 + z')/(l + 0 - 2 - i and /2(2 + z) = /(2 - 0 = 2 + z. Thus, we see that f2n(2 + i) = 2 + i and
f2n+1(2 + 0 = 2- z, so the answer to our problem is 2 - i.
145. Since the cube of the number z = (r, 0) is (r3,30), z3 will be real if and only if 30 = nn. Since
2 and K are positive, the point representing 2 + Ki is in the first quadrant. Hence, 0 < 6 < 90° for
2 + Ki. Thus, we must have 30 = 180°, so 0 = 60°. Since there is only one K such that (2 + Ki) has the
polar representation (r, 60°), only 1 K that satisfies the problem.
146. The slope of the line through two points is the ratio of the difference in y coordinates to the
difference in x coordinates. Since the difference in y coordinates of the points w and z in the complex
plane is the difference of the imaginary parts of the numbers and the difference in the x coordinates
is the difference in real parts of w and z, the slope between the points which represent w and z is
Im(z -w)l Re(z - w).
147. Let F = z = (r, 0). Hence, we have 1/z = (1/r, -0). Since F is outside the unit circle, r > 1,
so that 1/r < 1. Thus, the reciprocal of z is inside the circle. Since the angle of 1/z is -0, it is on the
opposite side of the x axis from z. The only point which satisfies these restrictions is point C.
148. To find one sixth root of -64, we write -64 as 64em. Hence, one sixth root of -64 is
641/6^71/6 _ 2em/6. The other roots have the same magnitude, 2, but different angles (which differ
from 7i / 6 by multiples otln/6 = n/3). Since we only want the roots with positive x, we only want the
roots with angle between -ti/2 and tt/2, exclusive. These roots are 2enl6 and 2e~n/6. Their product
then is 2 • 2en^6~n^6 = 4e° = 4.
149. Letting z = x + yi and writing the given equation in terms of x and y we have + y2 = y+l.
Squaring both sides, we have x2 + y2 - y2 + 2y + 1. Rearranging, we find x2 - 2y - 1 = 0, which is a
parabola.
150. From DeMoivre's Theorem, we have
(cos 0 + z sin 0)5 = cos 50 + z sin 50.
To find cos 50, we equate the real part on the right to that on the left, so we find
cos 50
cos5 0-10 cos3 0(1 - cos2 0) + 5 cos 0(1 - cos2 0)2

16 cos5 0-20 cos3 0 + 5 cos 0.
151. Let w — cos 12° + i sin 12° and z = cos 48° + i sin 48°. Point A represents
u> and B represents z. Point w + z is found by copying OB starting from A.
(Why? Compare this to adding 4 + 2i and 3 + 4z.) This gives us point C. Since
|z| = M = 1, we have OA = OB = BC = CA and OACB is a parallelogram. Since
/.BOA = 48° - 12° = 36°, /COA = 36°/2 = 18°. Thus, CO makes an angle of
18° + 12° = 30° with the real axis and hence has polar representation (r, 30°) for
some r. Thus, (w + z)6 = (r6,6 • 30°) = (r6,180°) and (w + z)6 = -r6. Hence, (w + z)6 has no imaginary
part and the answer to our problem is 0.
152. We can solve the given equation for x by first multiplying by x then using the quadratic
formula. Our equation becomes x2 - 2x cos 0 + 1=0. Applying the quadratic formula we have
2 cos 0 ± V4 cos2 0-4

x = cos 0 ± Vcos2 0-1 = cos 0 ± V - sin2 0 = cos 0 ± i sin 0.
Hence, x = cos 0 ± zsin 0 = e±ld. Thus we find xn = [e±l6]n = e±in0 and l/xn - l/e±ine = e ±inQ. Thus,
xn + —
xn
= e±in6 + e~±m0 = 2 cos nd.
Make sure you see why cos nd - (emd + e~tn6)/2.

153. The given sum looks very formidable; however, note that cos n6 = Re(emd). Thus, we can
write our sum, which we'll call S, as
S = Re
The last sum is a geometric series with first term 1 and ratio el6 /2. Hence, S becomes
S = Re (-—) =Re(-T-1-7-V
\l-e,d/2 J yl - \ cos 0 - | sin0y
To determine the real part of S, we must rationalize the denominator by multiplying top and bottom
by 1 - \ cos 0+5 sin 0, yielding
( 1 - l cos 0 + 4 sin 0 \ 1 - l cos 0

\ (1 - 5 cos 0)2 + | sin2 0 J (1 - 5 cos 0)2 + \ sin2 0
Since cos 0 = 1/5, we have sin2 0 = 1 - 1/25 = 24/25 and
r_ 1 1/10
- 6 _
" (9/10)2 + (l/4)(24/25) " 7'
154. Since z is a root of the polynomial, we have
zn + an-\Zn~l + • • • + a\Z + ao = 0.
To get a zn+l term, we multiply the given equation by (z - 1), giving
(z - l)(zM + an-izn~l + • • • + a\Z + flo) = 0

z"+1 + {an-\ - l)z" + (a„-2 -an-i)zn~l + ••• + (a0 - ai)z - a0 - 0.
We solve this equation for zn+1 since we wish to prove that zM+1 = 1. This yields
zn+1 - _ an-i)zn + (an-1 - dn-2)zn~x + ■•• + (d\- do)z + do-
We can now apply the Triangle Inequality to this relation, noting that each of the coefficients on the
right side are positive:
\zn+1\ < |(1 - an-i)zn\ + |(a„_i - an-i)zn~x\ + • • • + |(fli - a0)z\ + |a0|

\z\n+1 < (i - an-i)\zn\ + (fl„_i - an-2)\zn~x! + ••• + («!- a0)\z\ + a0
\z\n+1 < (1 -an-i)\z\n + (an-i -an-2)\z\n~l + ••• + (fli - a0)\z\ + a0.
Since we are given |z| > 1, we have |z|* > \z\i for all i > j. (Make sure you see why.) Hence, each of
the powers on the left side are less than or equal to \z\n, so we can write
\z\n+l < (1 - fl„-i)|z|n + (a„_i - an-2)\z\n + ••• + («!- a0)\z\n + a0\z\n = \z\n.
Make sure you understand why the sum equals \z\n. Thus, we have \z\n+l < \z\n and |z| > 1, which
can only be true if |z| = 1. Now let's return to our expression of the Triangle Inequality,
\z\n+1 < (1 - dn-i)\zn\ + {dn-i - dn-2)\zn~l\ + ••• + («!- flo)|z| + do.
For |z| = 1, this inequality becomes an equality. The Triangle Inequality can only be an equality if
each of the terms (1 - dn-\)zn, (a„_i - dn-2)zn~1,..., do have the same angle when written in polar form.
Since the last term, do, is real, all the terms must be real. From the equation
zn+i - _ an-\)zn + (dn-i - dn-2)zn_1 + ••• + («!- do)z + do,
we see that since each term on the right is real and positive (why positive?), z”+1 is real and positive.
Thus, |z|”+1 = 1 implies z”+1 = 1.
Chapter 10
Vectors and Matrices
10-1For Property 2, we know that v ■ zv = |M|||w|| cos 6, where 6 is the angle between v and w.
Assuming ||u|| and ||w|| are not zero, the only way this dot product can be zero is if cos 0 = 0; this
happens only when 0 - n/2 or 3n/2, and in both of these cases v and w are perpendicular.
Property 3 is immediate: (cv) • w = ||cu||||w|| cos 0 - c||u||||w|| cos 6 = cv-w.
10-2 The tail is at (0,0), while the head is at (2,3); thus the Pythagorean Theorem gives the
length as V22 + 32 = Vl3. (Make sure you can write down a formula for the length of an arbitrary
vector (x y).)
10-3 The dot product of the two is (l)(-6) + (17)(1) + (-3)(5) + (2)(2) = -6 + 17 - 15 + 4 = 0, so
the two are perpendicular.
10-4To find the top entry in the product vector, we go along the top row of the given matrix
and down the given vector, to get (2)(1) + (-4)(-l) = 6. To get the bottom entry, we go along the
bottom row of the given matrix and down the given vector, to get (-6)(1) + (8)(-l) = -14. thus the
product vector is
10-5
0 x' x
i. Let the general vector be ( * ); then transforming by [ _°.) yields (X
X V° 0 -1 X
The matrix reflects the vector across the x axis.
'ax + by'
ii. The general matrix X ^ j takes the general vector to For this to equal the
X cx + dyy
vector we started with, or we must have b = c = 0, a = d = 1. Thus the identity matrix is
1 0
0 1
< 51 >
52 > CHAPTER 10. VECTORS AND MATRICES
iii. A 3D vector looks like ( y ), and is transformed by a matrix as:
ax + by + cz
= ( dx + ey + fz
gx + hy + iz
You should easily be able to write down the 3D identity matrix, as well as matrices which reflect in
the xy, yz, and xz planes.
W\
10-6 We let v - (Vl ) and w = Then
V2. \w2.
'a b\ fvi + w\ 'av i + aw\ + bv 2 + bu>2

A(v + w) =
Kc d) yu2 + w>2 cv 1 + cw\ + dv 2 + dw2
'av i + bv 2
CV\+dV2j \CW\+dW2j
which is the first requirement of linearity, and
/a b\ f rvi\ aw1+brv2\fav1+bv2\
A(rv) =
c d \rv2, crv 1 + drv 2 J \ci>i + dv 2 J
which is the second.

10-7 First we do the multiplication inside the brackets. For the top entry we get (3)(l) + (-4)(l) =
-1, and for the bottom entry we get (5)(1) + (-6)(1) = -1. Thus the first multiplication yields
and our original product simplifies to
To execute this multiplication we do the same thing. The top entry is (-1)(-1) + (2)(-l) = —1, while
the second is (3)(-l) + (-4)(-l) = 1. Thus the overall product is
10-8 We stated in the text that
fa b\ (e f\ _ fae + bg af + bh\
yc dj \g hj ~ \ ce + dg cf + dh)‘
Considering only the first columns of the second multiplied matrix and the product matrix, our
multiplication looks like
fa b\ fe\ _ fae + bg\
Vc d) \g) -{pe + dg)'
which is a normal matrix multiplication of a vector. Similarly, considering only the second columns
yields
(a b\(f\ (af + bh\

\c d) \ h J =
\cf + dhj'
which is again simple matrix-vector multiplication. If you ever forget how to do matrix multiplica¬
tion, just do it column-by-column like this and you're home free.
0 1'
10-9 Comparing the form of the product matrix. |, to the general rotation matrix.
-1 0
/cos 6 - sin
, we see that this product matrix corresponds to a rotation by 270°. This is exactly
l sin 0 cos 6
what we would expect, since performing a 90° rotation then a 180° rotation should yield a 90° + 180° =
270° rotation!
'i iN
10-10 We evaluate each entry of the product ^ ^ ^ j ^ 3J separately. To get the upper
left entry, we go across the upper row of the first matrix and down the left column of the second, to
get (2)(1) + (—3)(2) = -4. (Make sure you see how the terms in this sum correspond to the entries
of the matrices.) To get the upper right entry, we go across the upper row of the first matrix and
down the left column of the second, to get (2)(1) + (-3)(3) = -7. To get the lower left entry of the
product, we go across the lower row of the first matrix and down the left column of the second, to
get (-4)(1) + (5)(2) = 6. To get the lower right entry of the product, we go across the lower row of
the first matrix and down the right column of the second, to get (-4)(1) + (5)(3) = 11. Thus the final
product is ^ 64 .
Matrix multiplication takes some getting used to. Practice until you can do it without too much
thought, as it is an essential tool.
10-11Geomtrically, the equivalent is a reflection through the origin. In matrices, we multiply
'-1 0N
the matrix for reflection through the y axis. (, by the matrix for reflection through the x axis.
,0 K
1 0 -1 o\ /1 0\ _ /-1 0N
, to get This corresponds exactly to the reflection through
0 -1 0 1 j lo -1 j - v 0 -1
the origin which we got geometrically. (Why?)
T 1 '1 r 1 2 r '1 r 2 J
10-12 We have , while '1
10-13 Using the form of the rotation matrix, we get A =

(1/2 -V3/2\
Since A6 corresponds
v V3/2 1/2 J
to a rotation by 60° X 6 = 360°, which is the same as a rotation by 0°, A6 must be the identity matrix
e 5-
10-14 In exactly the same way as we found the 2x2 identity, the 3x3 identity is
10-15
i. Rotation by an angle 0 about the x axis leaves the x-coordinates of all points the same, while
transforming the y- and z-coodinates according to
fy\ (cos 6 - sin d\ /y\

yzy :==^> ysint? cost? )\z)‘
The 3D matrix accomplishing the rotation is thus
/1 0
0 6 AI.
I 0 cos 9 - sin
yO sin 6 COS0 J
(Verify for yourself that this matrix leaves the v-coordinate alone and performs the desired transfor¬
mation to the y- and z-coordinates.)
ii. To squash any 3D vector to 0, we need the matrix
/0 0 0\
0 0 0 .
\0 0 0 /
iii. Reflection in the xy plane leaves the x- and y-coordinates of a point alone but changes the
z-coordinate to -z. Thus the matrix we want is
10-16 The 3x3 case you can do for yourself. For the 2 X 4 and 4x3 case, let's multiply
9 10 11
A -2 -3 4n 12 13 14
5 -6 -7 8, 15 16 17
18 19 20
The product matrix is
((1)(9) + (—2)(12) + (~3)(15) + (4)(18)

\(5)(9) + (-6)(12) + (-7)(15) + (8)(18)
(1)(10) + (-2)(13) + (-3)(16) + (4)(19)

(5)(10) + (-6)(13) + (-7)(16) + (8)(19)
(1)(11) + (-2)(14) + (-3)(17) + (4)(20)\

(5)(H) + (-6)(14) + (-7)(17) + (8)(20) J'
where we have staggered the entries of the 2 x 3 matrix to make them fit on the page. Make sure
you understand how these terms come about by going across rows and down columns. From a 2 X 4
and a 4 x 3 matrix, we get a product which is 2 X 3.
x
For the third case, let's multiply [a b cj by ( y \. We get (ax + by + cz}, a 1 x 1 matrix which
corresponds the the dot product of the two matrices if they are taken as vectors!
10-17 The two do indeed agree.
10-18 We can write = j - h which sums up the entire matrix in a single equation.
f cos 6 - sin d\
155. Recalling that the 2D rotation matrix is
^sin 6 cos 6 J , we simply substitute in 0 = 45° to get
/V2/2 -V2/2\
VV2/2 V2/2 J'
156. Since a rotation by x and then by y is the same as a rotation by x + y, we have
f cos x - sinx\ (cos y - sin y\ _ f cos(x + y) - sin(x + y)

ysinx cosx J ysiny cosy J ysin(x + y) cos(x + y)
Multiplying the matrices, we have
( cos x cos y - sin x sin y - cos x sin y - sin x cos y\ _ /cos(x + y) -sin(x + y)\
y+cosx sin y + sin x cosy cos x cos y - sin x sin y J ysin(x + y) cos(x + y) )’
and by comparing the matrices on the two sides of this equation we have the desired identities.
157. The product is
(2)(-2) + (1)(4) + (9)(3)

(-6)(-2) + (0)(4) + (-3)(3)
(l)(-2) + (3)(4) + (2)(3)
(2)(11) + (1)(4) + (9)(-2)
(-6)(11) + (0)(4) + (-3)(-2)
(1)(11) + (3)(4) + (2)(-2)
(2)( 1) + (1)(~3) + (9)(1) \

(_6)(_1) + (0)(-3) + (-3)(1)
(1)(-1) + (3)(-3) + (2)(1) )
f77 8 4 \
= 3 -60 3 .
\16 19 -8/
158. Since PRQS is a parallelogram with P and R diagonally opposite, PQ = SR. If S = (si,s2), we
thus have (4 -3) = (9 - Si 1 - s2), which yields Si = 5 and s2 = 4. Thus the coordinates of S are
(5,4).
159. We have
(3 4 5) • (-1 4 3) 28 14 Vl3
cos 6 -
||(3 4 5)||||(-1 4 3)|| 10Vl3 65
160. To multiply a 2 x 3 and a 4 x 2 matrix, we have to put the 4x2 first and the 2x3 second.
The dimension of the product is a 4 X 3 matrix, so has 12 elements.
161. Let B = Then AB = BA becomes
/15* + 2z 15y + 2w\ _ / 15x + 6y 2x + 7y\

\ 6x + 7z 6y + 7w J ylSz + 6w 2z + 7wJ'
which results in the four equations 15x + 2z = 15x + 6y, 15y + 2w = 2x + 7y, 6x + 7z - 15z + 6w, and
6y + 7w = 2z + 7w. The first and last equations yield z = 3y; plugging this into the second and third
equations gives 2x - 2w - 8y = 8z/3 for each. From this last equation, z must be divisible by 3 if x
and w are both integers. Trying z = 3 as the smallest such z, we get y - 1 and x-zv = 4. The smallest
positive integers x and w satisfying this equation are x = 5 and y = 1. The sum of the entries of the
matrix isx + y + z + w = l + l+ 5 + 3 = 10.
162. We multiply the vector by the matrix to get
(1 4 l\/3\
-2 0 0 1
V 3 2 -3j\2j
163. The given sum can be thought of as the dot product of the vectors
(2 3 6) and (sin x cosy sin x sin y cos x).
The maximum value of this dot product is the product of the lengths of the two vectors, \4 + 9 + 36 =
V49 = 7 and
a/ sin2 x cos2 y + sin2 x sin2 y +:0s2 x = \/ sin2 x + cos2 x = 1,
or 7. This maximum is attained when the two vectors are parallel. The minimum, attained when
the two vectors point in opposite directions, is the negative of the product of the lengths, or -7.
Chapter 11
Cross Products and Determinants
11-1 The area of the triangle so spanned is \ ||z/||||ze|| sin 6. The area of the parallelogram spanned
by the vectors is twice this, or ||^||||ie|| sin 6.
11-2 Let v and zv both have length 1, so that the area of the parallelogram spanned is just sin 6,
where 6 is the angle between v and zv and ranges between 0 and 2n. The cross product points straight
up and has length sin 6. For 0 = 0, sin 6 = 0, so the head of the cross product is at the origin. As 6
increases, the head rises straight up until, for 6 = 7i/2, it is at height 1. As 6 continues to increase,
the head swings back down to 0, until for 6 = tl it is at the origin again. As 6 increases to 37i/2, the
head drops to a depth of -1 below the xy plane, then it swings back up to 0 as 6 gets up to 2n.
11-3 Point the index finger of your right hand directly away from you and your middle finger
toward the left. Your thumb, the cross product, then points up. Now, without changing the position
of your fingers, point your middle finger away from you and your index finger to the left. (This will
take some twisting of your arm.) If you're doing it the way we intended, your thumb points straight
down. This illustrates the fact that v x zv = -zv x v.
11-4 All we have to do is show that the dot products of the defined vector with v and zv are 0.
Since v = (x\ y\ z\) and zv = (x2 yi these dot products are
xi(yiZ2 - yizi) + yi(zi*2 - Z2*i) + zx{xxy2 - x2yx)
and
x2{yiz2 - y2zx) + y2{z.\x2 - z2xi) + z2[xxy2 - x2yx),
both of which simplify to 0.
11-5 Letting the two 2D vectors be (xx y\ 0) and (x2 yi 0), with z\=z2 = 0, the formula
in the text gives the cross product as
(0 0 xxy2-x2yx),
which is indeed straight up or straight down.

11-6 The cross product of v and zv in this notation is (0 0 vxzv2 - zvxv2). The area of the
parallelogram spanned is the length of the cross product vector, or \vxzv2 - zvxv2\.
< 57 >
58 > CHAPTER 11. CROSS PRODUCTS/DETERMINANTS
11-7 When we multiply everything out, the term we're interested is actually the two terms
v\w2Ai x A] + w\v2A] x At.

—^ ^
To put them together, we have to use Aj x Ai = -Ai x Aj, which is the source of the - sign.
11-8 This can be seen in many ways. The simplest is that the area of the parallelogram spanned
by v and itself is zero.
11-9
Multiplying the vectors for each vertex of the rectangle by the given matrix, we find that
the image under transformation has vertices (13,7.5), (12,7), (19,11.5), and (18,11). This is not a
rectangle, though it is a parallelogram defined by the vectors (l .5) and (7 4.5). Evaluating
the cross product of these vectors, we get 1 as the area of the parallelogram, so since the original
rectangle has area 2, the parallelogram's area has changed by a factor of 0.5. We didn't actually
have to compute this cross product, though—the area is multiplied by the absolute value of the
determinant, or |-.5| = .5.
11-10The determinant of the matrix is 11, so the area of the image of the circle is 11 times the
original area, or ll(47i) = 4471.
11-11The individual determinants are 13, 2, and -8, so the determinant of the product is
(13)(2)(-8) - -208.
11-12 Let's use our shorthand on a generic determinant:
011 012 013 011 012

021 022 023 021 022
031 032 033 031 032
—> #11#22033 + 012023031 + a13a21^32 ~ 031022013 ~ 032023011 ~ 033021012
This last expression is exactly the one we used in the text, so our shorthand gives the correct answer.
11-13
Since one element from each row and column appears in each three-term product in the
expression for the 3 X 3 determinant, each three-term product is multiplied by c. Thus the entire
expression is multiplied by c.
11-14
Since every term in the expression for the determinant is multiplied bye, each three-term
product is multiplied by c3. Thus the entire expression is multiplied by c3, and the determinant of B
IS C
11-15 Each 2x2 determinant is preceded by a term #z;-; the sign of the term is given by (-1 )*+/.
11-16 For example, in the 4x4 case it looks like
The checkerboard pattern is the same for any matrix.
11-17 Expanding across the first column of f Un Ul2 , we have a.\\ (022) 012 (021)
\a2i a22/
011022 - 012021- This is the correct form of the determinant.
11-18 Let's do the determinant

3 -2 11
0-52.
7 7-4
Expanding by minors down the first column, we have
-5 2 -2 11 -2 11
-0 +7
7 -4 7 -4 -5 2 ‘
(Make sure you see how we got this expression.) Evaluating the individual determinants, the overall
determinant becomes 3(6) + 7(51) = 375. You can confirm for yourself that the shortcut method gives
the same result.
11-19 Expanding by minors across the first row, the determinant becomes
fl22 0 0
0 fl33 0
All
0 0 /
Expanding the remaining determinant by minors, we have
033 0
0 044
flllfl22
0 0
and so on until the overall determinant is flufl22 • • • ann.

11-20 Expanding by minors across the first row, the determinant becomes
fl22 0 • • • 0
032 033 • ‘ • 0
flu
0n2 0«3 lnn.
Expanding the remaining determinant by minors, we have
0 0
044 0
011022
, 0«3 0«4 *nn,
and so on until the overall determinant is 011022 •••0n«-

11-21 Expanding by minors along the row or column which is all zeros, the determinant is
O(something) + O(something) H-= 0.
11-22 Subtracting the first row from all the others yields
17 23 23 23
0-600
0 -6-6 O'
0 —6 —6 -6
Expanding by minors down the first column then gives
-6 0 0
17 -6 -6 0
-6 -6 -6
Since the new determinant is triangular, the result is the product of the diagonal elements, or
(-6)(-6)(-6) = -216. Hence the original determinant is 17(-216) - —3672.
11-23Subtracting one identical row or column from the other yields a matrix with one row or
column which is all zeros. We have seen already that the determinant of such a matrix is 0.
11-24 Let the elements in one row of an n x n matrix be multiplied by c . If the elements of that
row were an, ai2>- • • / ai3 and have corresponding minors An, A[2,..., Aq. The determinant of the
matrix before multiplying the row by c is
(—iy+lflziA'i + (—1)z+20z2A'2 +-+ (-1 y+nainAin/
and after multiplying by c the determinant is
canAn + (-l)z+2cflj2Af2 H-f (-1 )l+ncainAin.
Clearly this is c times the original determinant.

For the case of multiplying all the elements by c we do it one row at a time. If the original
determinant was D, then after multiplying one row c the determinant is cD, after multiplying a
second row by c it's c2D, by repeating our original argument, and so on. After multiplying all n rows
by c the determinant is cnD.
11-25 Expanding the given form by minors, we find that the determinant is x0(yiz2 - y2Zi) +
yo(zxx2 — z2X\) + zo(xiy2 — x2yx). This is the dot product of the vectors
(*o Vo zq) and (yiz2 - y2zi z\x2-z2x\ xiy2-x2yi).
But the second vector is the cross product of (x\ y\ Z\) and (x2 y2 z2), so the given determinant
is (*0 I/O z0)-((*i yi zi) x (x2 y2 z2)), as desired.
11-26 If we substitute (xi yx zx) or (x2 y2 z2) for (x0 y0 z0) in the determinant of the
previous exercise, we get a determinant with two identical rows. Since we earlier proved that such
a determinant is 0, the given cross product is perpendicular to both (ay yx zx) and (x2 y2 z2).
11-27 A matrix with determinant 0 takes any finite volume to 0. To invert such a matrix,
another matrix would have to take the volume 0 image back to the finite original volume, which is
impossible. Thus a determinant-0 matrix cannot have an inverse.
'a
b'
11-28 We start with the matrix . First we evaluate the determinant, ad - be. Second we
dc
'd c
replace each entry with its minor, to get ^J. Third we append the checkerboard of signs, to
d -c^
get Fourth, and finally, we transpose the matrix and divide by the determinant, to get
x-b a /
r fd '-b\ „
\ -c a j' exac% f°rm we 8ot in text-
11-29 The determinant is 1. Replacing each term by its minor, we get
Appending the checkerboard of signs leaves this matrix unchanged, since all the nonzero terms get
a +; dividing by the determinant also leaves it unchanged, since the determinant is 1. Thus none
of the steps changes the matrix, so the result is again the identity matrix. This is what we would
expect, since the identity times itself is the identity.
11-30 We'll use the matrix
4 3 2 \
-3 2 -1 .
2 5 2J
The determinant is 10. Replacing each term by its minor yields
/ 9 -4 -19\
-44 14 ,
\-7 2 17 /
and appending the checkerboard of signs makes this
/ 9 4 -19\
4 4 -14 .
\-7 -2 17 J
Dividing through by the determinant and transposing, the inverse is thus
If we multiply this by the original matrix we get the identity matrix, as desired. (Try it.)
164. The entries of A are 2,3,5, and 7. The determinant is ad - be, where a through d are the entries in
any order. To maximize this determinant, we let a = 5, d = 7, b = 2, and c = 3; then the determinant
is (7)(5) - (2)(3) = 29.
165. We do some row operations to simplify the matrix, adding the third row to the first to get
6 9 10
-3-2 2 3
2 5 4 -2'
3-3-2 0
then adding 3/2 times the third row to the second to get
6 9 10
0 11/2 8 0
2 5 4 -2
3-3-2 0
Expanding by minors down the last column, our determinant is then
6 9 1
-(-2) 0 11/2 8 /
3 -3 -2
which by adding -2 times the third row to the first becomes
0 15 5
2 0 11/2 8
3 -3 -2
Expanding by minors down the first column now gives
15 5
6 = 6[(15)(8) - 5(11/2)] = 6(185/2) = 555.
11/2 8
166. There will be no inverse exactly when the determinant is zero. We use our shorthand method
to find the determinant, writing
1 4 c 1 4
2-17 2 -1
3 -2 11 3 -2
The down diagonals yield (1)(-1)(11) = -11, (4)(7)(3) = 84, and (c)(2)(-2) = -4c. The up diagonals
yield -(3)(-l)(c) = 3c, -(-2)(7)(1) = 14, and -(11)(2)(4) = -88. Adding these six terms together
yields -c - 1 as the determinant, so the determinant is zero when c = —1.
167. We recall that the determinant of a product is the product of the determinants. The
determinant of A is (2)(4) - (1)(3) = 5, of B is (2)(4) - (3)(5) = -7, and of C is (2)(1) - (1)(6) = -4. The
product is (5)(-7)(-4) = 140.
168. We subtract the first row from the second and third rows to get
a 111
1-a a - 1 0 0
1-a 0 a-1 0'
111a
then we subtract a times the first row from the last row to get
a 111
1—a a-1 0 0
1-a 0 a-l 0
1 - a2 1 -a 1-a 0
Expanding by minors down the fourth column then yields
1-a a-1 0
- 1-a 0 a-1.
1 - a2 1 -a 1 -a
Factoring a-1 from each entry, we get (a -1)3 in front, since the matrix is 3 x 3. Thus our determinant
becomes
-1 1 0
-<« - i)3 -1 0 1
-1 - a -1 -1
Using the shorthand methods on the determinant which remains, we get 0 + (-1 - a)+ 0- 0-1-1 =
-a - 3. Thus the original determinant is -(a - 1 )3(-fl - 3) = (a - l)3(a + 3).
169. Since the matrix is 2 x 2, factoring a 4 out of every term brings a 42 = 16 out front. Thus
4W 4X W X
= 16 = 16(4) = 64.
4Y 4Z Y Z
170. Subtracting the first row from all the rest yields
3 11111
0 2 0 0 0 0
0 2 2 0 0 0
0 2 2 2 0 0'
0 2 2 2 2 0
0 2 2 2 2 2
then expanding by minors down the first column gives
2 0 0 0 0
2 2 0 0 0
3 2 2 2 0 0
2 2 2 2 0
2 2 2 2 2
Since the remaining determinant is triangular, its determinant is equal to the product of the diagonal
elements, or 25 = 32. Hence the original determinant is 3(32) = 96.
171. To find , we first divide by the determinant, 2, then replace every element by its minor.
to get \ 4 y' tîen transPose the result/ to get A 1 = - To find A 2, we square this
1/1 -2\ / 1 -2/ _ 1 / 3 ~10\ _ / 3/4 -5/2\

t° get 4 ^_i J
4 y-1 4 ) 4 \-5 18 ) V-5/4 9/2 )'
64 > CHAPTER 11. CROSS PRODUCTS AND DETERMINANTS
172. We will create a determinant equation which is satisfied only when (x, y) is on the circle:
x2 +1/2 x y 1
10 -3 11
20 2 4 1
29 5-21
Why does this have the desired property? Substituting in (x, y) = (-3,1), (2,4), and (5, -2), you can
see that in each case the determinant is zero because it has a repeated row. Since expanding the
determinant will give the equation of a circle (do you see why?) which is satisfied by the given three
points, this is the unique circle passing through the three points.
Similar techniques to this one can be used to write the equations of lines, planes, and conic
sections. Though this method is not too generally useful, you should play with the determinants
until you can put the desired equations in determinant form.
Chapter 12
Analytic Geometry
12-1 From the discussion in the chapter, we have
. a_ 2/3 -1/3 _ 3
an 1 + (2/3)(l/3) 11'
The answer is not 3/11, however, since the problem asks for the tangent of the obtuse angle. Hence,
we require the negative value, —3/11 (which we could have found from using 1/3 as m2 rather than
2/3).
12-2 First we recognize the equation as the description of a hyperbola. With a bit of algebra,
we write the equation as
(y - 2)2 _ (x - 4)2 =
9 4
Now recall that sec2 6 - tan2 6 = 1. Comparing this to the form of our hyperbola, we want
(y - 2)2 2 (x - 4)2
--= sec 6 and tan2 6,
9
so y = 2 + 3 sec 9 and x = 4 + 2 tan 6 is our parametric representation.

12-3 Since we know that sin2 6 + cos2 6 = 1, we write the first equation as x = (1 - cos2 6) / cos 6.
Since y = cos 6, we have
x = 1 -v2
y
which is the equation of the curve in rectangular coordinates.
12-4 As discussed in the text, the dot product WZ • ft is the product of the length of the projection
of WZ onto ft and the length of ft itself. Since we are only interested in the length of the projection,
we must divide WZ • ft by ||«||.
12-5 As in the example in the text, we consider our points in the plane as points in the space
with z = 0. Let a = (x2 - *i yi~ y\ 0) and b = (*3 - x\ yz~y\ 0). The area of the triangle is
< 65 >
66 > CHAPTER 12. ANALYTIC GEOMETRY
||a x 6||/2. We find that
a x b = (0 0 X\y2 + x2y3 + x3yi - xxy3 - x2yi - x3y2).
The area of the triangle is half the magnitude of this vector, or
Area = |(*iy2 + x2y3 + x3y\ - x\y3 - x2y\ - x3y2)/2|.
Evaluating the determinant expression in the problem, we get the same expression as above, so the
determinant in the problem does indeed give the area.
12-6 The previous exercise virtually proves the assertion for a triangle. For B
induction, suppose that the process works for n points. If we tack on an n + 1th
vertex, we just add another triangle as shown in the diagram. Let A through
E be the original n points and X be our added point. The area of ABCDEX is
[ABODE] + [AEX], Since ABODE has n vertices, our process works for finding
[ABODE], so we write
X\ Vi
X2 1/2
x2y\ x3 1/3 xi y2
Xn-lVn-2 Xn Vn Xn-2yn-l
Xnyn—1 Xl yi Xn-iyn
x\ yn xnyi
and let Kr be the sum of the right column and X; be the sum of the left. Then we have [ABODE] =
I(Kr - K\)/2|. For the triangle AEX, we note that point A is (x\, y{), E is (xn, yn), and X is {xn+i, yn+1),
so we find [AFX] by writing
X\ 3/i
Xn Vn
xnyi Xn+1 }/n+1 x\yn
Xn+l]/n Xl Vi Xnyn+l
X\yn+1 Xn+iyi
and letting Lr be the sum on the right and L/ be the sum on the left. Then [AEX] = \(Lr - L/)/2|. If we
write all the vertices of ABCDEX as the problem suggests, we have
Xl yi
x2 3/2
X2yi X3 3/3 xi yi
Xn}jn-1 Xn+1 3/n+l Xn-iyn

Xn+iyn Xl 3/i Xnyn+l
x\yn+\ xn+iyi
and we let Mr be the sum of the terms on the right and M/ be the sum of those on the left. By direct
comparison we find that the terms occurring in Kr - K, + Lr - L, are exactly those which occur in
Mr-Ml. Hence, \{Mr -M/)/2| represents the sum of the area of ABODE and our new triangle and thus
our induction is complete. Note that our induction is not completely 100% rigorous, but it should
give you a pretty good idea how it goes.
B C 12-7 Instead of dealing directly with right triangle ABC, we complete the rectangle
ABCD and view A ABC as half this rectangle. Notice that A ACD is just the rotation of
A ABC about the center of ABCD. Clearly, A ABC and A ACD have the same number of
interior points JA and the same number of boundary points. We'll divide the boundary
points of the triangles into those inside the rectangle (the points on AC) and those on
the perimeter of the rectangle. The latter is simple; there are a + b + 1 points on the
D perimeter of each triangle which are also on the perimeter of the rectangle. For the
former, we let BA be the number of lattice points on AC (besides A and C). Since the area of A ABC is
ab 12, to prove Pick's Theorem we must show that
Ba + a + b + 1
ab / 2 — 1A + -l
To prove this, we apply Pick's Theorem to ABCD. Since all the interior points of ABCD are either
interior points of one triangle or the other (but not both) or on segment AC, we have
. ,_r . 2a + 2b
ab — (2JA + Ba) h---1
Dividing this by 2 we have
ab BA+a + b 1
2 ~h+ 2-2'
which is equivalent to our desired expression. Hence we have proven that Pick's Theorem indeed
holds for right triangles whose sides are parallel to the coordinate axes.
12-8 Number the vertices of polygon P from 1 to n. We divide the polygon into n — 2 triangles
by drawing segments from vertex 1 to the vertices 3 through n- 1. Now let Ip and Bp be the number
of interior points and boundary points of the polygon and IA be the number of points inside any of
the triangles. If we apply Pick's Theorem to the triangles and add the results we will get the area
A of the polygon. Each interior point of the polygon is either an interior point of one triangle or a
boundary point of two triangles. Hence in summing the Pick's Theorem results from the triangles,
each interior point of the polygon is included exactly once. The boundary points of P are boundary
points of the triangles. Let the number of the boundary points of P which are not vertices of P be
B*; hence Bp = B* + n. Each of the members of B* is included once with weight 1/2 in the triangle
Pick's Theorem results because each one is on exactly one triangle. The vertices are another matter.
Vertex 1 appears in all n - 2 triangles, vertices 2 and n - 1 are only on 1, and the other n - 3 are on
2 triangles each. Summing these we have (n - 2) + 1 + 1 + 2(n - 3) = 3n - 6, each with weight 1/2.
Since we are adding n-2 Pick's Theorem equations to get the area A of the polygon, we have n-2
'-l’s, so our result is
B, 3/1-6
A - Ip + — + —^-(^-2)
Bp - n 3n - 6 2n - 4
- Ip + —— + .--—
Br
= Ip + ir-b
which proves Pick's Theorem for our polygon.

12-9 Order the points A, B, and C from least to greatest y coordinate and from least to greatest x
coordinate. Hence there are 4 points which have the largest or smallest of one of these coordinates.
Since there are only 3 distinct points, one point must be on an extreme in both lists. This point is our
corner as point B in the proof. The rectangle is then easily constructed by drawing the appropriate
lines from B. (Make sure you see this.)
12-10
We can extend our argument to non-convex polygons by noting that we can chop any
non-convex region into convex pieces. If we apply Pick's Theorem to each piece then sum the results
we will prove Pick's Theorem for the non-convex region.
12-11 The vector from the first point to the second is (2 - 1 3 - (-1) 1 - 3) = (l 4 -2) = v,
so to get a point on the line we can add any multiple of v to (1,-1,3). Thus, our parametric
representation is
x - 1 +t
V = -1+41
z = 3 - It.
12-12 First we find two vectors in the given plane. Letting the points in the order of the problem
be P, Q, and R, we have the vectors PQ = (-1 -5 0) and PR = (-2 -4 l). Thus, the vector
PQ x PR = (-5 1 -6) is normal to the plane. Hence, for any point S = (x,y,z) in the plane the
vector PS is normal to (-5 1 -6), so
(x -1 y-2 z - l) • (-5 1 -6) = 0.
Thus, our plane is 5x - y + 6z - 9 = 0.

12-13 First, just like the vector (A B^j is normal to the line Ax + By + C = 0, the vector
ft = (A B
C) is normal to the plane Ax + By + Cz + D = 0. From here we proceed just as we
did in finding the distance from a point to a line. Let P = {x\,y\,Z\) be a point in the plane
and point Q be the given (xo, yo,Zo). Our desired distance then is the length of the projection of
PQ = (xo — X\ yo ~ yi zo - zi) onto the normal vector through P. As before, we use the dot
product to find this length as
n = \PQ-n\ _ \A(x0 - xi) + B(y0 - yi) + C(z0 - zQ|

Ml YA2 + B2 + C2
|Axo + Byo + Czo — {Ax\ + By\ + Czi)|
y/A2 + B2 + C2
|Axo + Byo + Czo + D|
y/A2 + B2 + C2
where we have used the fact that (xi, yi,zi) is on the plane so that Axi + Byi + Cz\ = -D. If you don't
quite follow this, compare it to the example in the book where we find the distance from a point to
a line.
12-14 Since axb is normal to plane OAB, it is in the same direction as ft. Hence, we can write
a x b = \\a x b\\ft since the cross product has magnitude ||a x b\\ and direction ft. Thus, we can write
c ■ (ax b) = c • (||fl x b\\n) = II ax b\\(c■ ft).

12-15 Regardless of the order of the vectors in the product, the box product still has magnitude
equal to the volume of the parallelepiped spanned by the three vectors. This parallelepiped is the
same for any ordering of the vectors, so the volume remains the same as does the magnitude of the
box product.
12-16 Let the vectors be u, v, and w. Since
V X w = X2 1/2 Z2 /
*3 1/3 23
to find it ■ (vxw), we multiply the first term of it by the coefficient of i in the cross product, the second
term of it by the coefficient of j, and the third term of it by the coefficient of k, then add the three
results. This procedure is exactly the same as substituting the components of it for the i, j, and k in
the above cross product. (Make sure you see why!) Our box product then is
*1 1/1 Zi
it ■ (vxw) X2 1/2 Z2
*3 1/3 Z3
For a tetrahedron, we choose the first point to be the origin; hence, the vectors which form the tetra¬
hedron are (x2 — X\ 1/2 - i/i Z2-Z1)/ (x3-X\ 1/3 - 1/1 23-21), and (*4 - X\ 1/4 - 1/1 Z4-Z1).
The volume is the absolute value of 1 /6 of the box product of these, or
■y X2~ X\ 3/2 - 3/1 z2 — Zi

V = - X3-Xi 3/3 — 3/1 z3 - Zl
6
X4 - Xi 3/4 — 3/1 z4 — Zl
12-17 The z coordinate is easy; it's just z = -4. As for x and y, we convert the (3,120°) from
polar to rectangular coordinates as x - r cos 0 = -3/2 and y = r sin 0 = 3 y/3/2. Hence our point is
(—3/2,3 V3/2, —4).
12-18 The equation for a cylinder in rectangular coordinates is x2 + y2 = k2 for some constant
k. Letting x = rcos 6 and y = r sin 0, we have x2 + y2 = r2(cos2 0 + sin2 0) = r2 = k2 as our equation
in cylindrical coordinates. If we take the square root, we find that the equation for a cylinder in
cylindrical coordinates is just r = k. Does this make sense? A cylinder is the set of points k away
from the z axis, so the radius is constant while z and 0 are unrestricted.
12-19 Using our values for x, y, and z, we have
x2 + y2 + z2 = p2 cos2 0 sin2 <fi + p2 sin2 0 sin2 (p + p2 cos2 (p

= p2 sin2 <p(cos2 0 + sin2 0) + p2 cos2 (p
= p2 sin2 <p + p2 cos2 <p
as desired.
12-20 Every point on the sphere is a constant distance k from the origin, no matter what <p and
6 are. Since this distance from a point to the origin in polar coordinates is simply p, our equation for
a sphere centered at the origin is p = k.
12-21 We've already addressed the curve p = p\ (it's a sphere), so we move on to 6 = 0\. For
this surface, p and <p can take any value and the resulting graph is a plane perpendicular to the xy
plane which makes an angle of 6 with the positive x axis. Similarly, if we let <p be constant and p
and 6 vary, we trace out a cone, since varying 6 from 0 to 2n while keeping <p and p constant makes
circles. Varying p varies the radius, ultimately forming a cone.
12-22 The equation x2 + y2 = r2 in the (x, y) plane describes a circle; hence, for x2 -v y2 - z2 at
each value of z we have a circle. Thus, every cross-section of the graph is a circle; however, the graph
is not a cylinder, because the radii of these circles are not all the same. In fact, at z = Z\, the radius is
Zi and the distance from the center of the circle to the origin (which is clearly on the graph) is also Z\.
From this we see that the graph is a cone, since the ratio of the radius of any of the cross-sectional
circles to the distance (i.e. the height of the cone) from the center of the circle to vertex (the origin) is
constant. If you don't quite buy this, try graphing the equation or writing the equation in cylindrical
coordinates.
12-23 Since AB - B - A and likewise for BC, we can write
AB + BC = B-A + C-B = C-A=AC.
12-24 Again we write AB = B - A and likewise for AC and we find
AB - AC = B-A-(C-A) = B- C = CB.
12-25 Let O be the origin. If we extend the line past X to the point D such C D
that OA is the side of a parallelogram with sides parallel to OA and OB as shown,
we have AXBO ~ AXAD. Hence, AD/BO = AX/BX - 2 and OX/XD = 1/2.
Thus, we have OC = 2OB. Hence, C = 2B and D = 2B + CD -2B + A since A
and CD have the same direction and magnitude. Since OX/XD - 1/2, we find
OX/OD = 1/3 and X = D/3 = (2B + A)/3. Can you extend this argument to show that if point X is
on AB such that AX = c(BX), then X = (cB + A)/(c +1). Is this still true if c is a fraction? An irrational
number?
12-26 Let D and £ be the feet of the angle bisectors drawn from A and B, respectively. Since I is
on both AD and BE, we can write
I-A = fci (D-A)

I-B = k2(E-B).
Using the example in the text to find D and E, we have
-> fbB + cC
I h +A
V h+c
-> aA + cC
I k2 + B.
u+c
Equating coefficients of C and B, we find
kxc kic kx b
and 1 - =
b+c a+c ~ b+c
Solving for k2 we find k2 = (a + c)/(a + b + c), so using the second equation for / above, we find
aA + bB + cC
I =
u+b+c
12-27 Just as in the prior example, we apply the law of cosines to find
(A - C) • (A - C) = \\A\\2 + ||C||2 - 2||X||||<?|| cos MOC.
Comparing this to the given equation, we have cos ZAOC = 1/2, so that ZAOC = 60°.
12-28 The lines defined as AG in the previous example are the altitudes of regular tetrahedron
ABCD; hence, all that we have proved for those segments is also true for the altitudes of a regular
tetrahedron.
12-29 In the prior example, we have X - A - (3/4)(G-A), so that AX = 3GX. Segment XG is the
radius of the inscribed sphere since XG is perpendicular to BCD and the perpendicular segments from
X to the centers of each of the other 3 faces have the same length. Similarly, AX = BX - CX = DX and
the sphere with center X and radius AX is circumscribed about the tetrahedron. Since AX/AG = 3/4
and XG/AG = 1/4, the circumradius and inradius are 3/4 and 1/4, respectively, of the altitude length.
12-30 Let AG be the altitude from A to G, the centroid of A BCD. We find AG by considering
right triangle AGB. We know hypotenuse AB has length 6. The altitudes of equilateral triangle
BCD (and hence the medians) have length 3 V3. Since BG is 2/3 the median from B, it has length
(2/3) (3 V3) = 2 V3. Thus, from A AGB, we have AG - 2V6. From the prior exercise, the inradius is
(1/4) (2 Vb) = V*>/2 and the circumradius is 3V6/2. Since the area of A BCD is 9 V3, the volume of
ABCD is (AG)[BCD\/3 = 18 V2.
173. This is exactly like an example in the text. Our vectors spanning the tetrahedron are AB, AC,
and AD, or (3 3 -2), (l 0 -l), and (2 4 -2). Finding our volume as in the example, we
have
3 3-2
1 4 _ 2
v= 1 0 -1
6 6 ~ 3'
2 4-2
174. Solving the first equation for 2p, we have 2V = x - 1. We can write the second equation if
terms of 2p as y = 1 + 1/2^. Hence, we have y - 1 + l/(x - 1) = x/(x - 1).
175. As you should have determined in an exercise in the chapter, the graph of x2 + i/2 = z2
represents a cone. Bounding this by z < 6, we find that the cone has height 6 and radius 6, so the
volume is 6(62)n/3 = 727T.
176. Projecting P onto the x axis, l/ axis, and z axis, we find points X, Y, and Z, respectively. Hence,
we have cos0i = ZO/OP, cos @2 — XO/OP, and cos 03 = YO/OP. Since OP2 = XO2 + YO2 + ZO2,
we have
cos2 01 + COS2 02 + cos2 03 — 1.
From this we solve for cos 03 as Vl - R2 - S2.

177. The line is in the direction from the first point to the second, or (-3 -3 -6). Hence the
line goes through the point (1,2,3) and in the direction (-3 -3 -6), so the line is
(1,2,3) + f(-3, -3, -6) = 0,
just as in the text. Thus, our parametric equations describing the line are
x = 1-3t
y = 2-31
z = 3-6t.
178. Set up a coordinate system with B = (0,0,0), A = (v, 0,0), and C = (0, w, 0). Let D = (a, b, c).
Since /.BAD = n/2, point D is in the plane perpendicular to line BA. This plane is described by x — v,
so we have a = v. Since ZBCD = n/2, we have b - w. (Make sure you see both of these.) Applying
the Pythagorean Theorem to A CD A, we find
v2 + w2 = CA2 = CD2 + DA2 = v2 + c2 + w2 + c2.
so c = 0, and point D is in the xy plane along with the other three points.
179. First we observe that since BP and AP are in opposite directions and AP = k(BP), we have
kBP + AP = 0 and likewise kRD + RC - 0. We know that BP + PR + RD = BD and AP + PR + RC = AC.
Hence, we have
—* _ (k + 1 )PR + kBP + AP + kRD + RC
^ k+1
k(BP + PR + RD) + AP + PR + RC
k+1
kBD + AC
k+1 '
Similarly we find
-kAC + BD
k+1
We can find the area of PQRS as ||PP x QS||/2, since the area of a quadrilateral is half the product of
its diagonals times the sine of the angle between them. Make sure you see this. Hence, we have
kBD + AC -kAC + BD
[PQRS] = -
k+1 k+1
k2 1
BDxAC + ACxBD
2 (k + l)2 2 (k +1)2
since AC x AC = BD x BD = 0. Because BD x AC = -AC x BD, we then have
Thus, we have [PQRS]/[ABCD] = (k2 + 1 )/(k + l)2. Setting this equal to 0.52 from the information in
the problem, we have the quadratic (after a bit of algebra) 6k2 - 13k + 6 = 0. Factoring, we have the
values k = 2/3 and k = 3/2.
180. We have a sphere and a point on the sphere, so we can try analytic geometry. Let the corner
of the room be the origin and the radius of the sphere be r. The floor then is the xy plane and the
walls are the yz plane and the xz plane. Since the center of the sphere is r away from the floor and
the walls, the center is (r, r, r). Hence, the equation describing the sphere is
(x - r)2 + (y - r)2 + (z - r)2 - r2.
Fortunately, we are told that the point (5,5,10) is on the sphere. Putting this point in our equation,
we have
(5 - r)2 + (5 - r)2 + (10 - r)2 = 3r2 - 40r + 150 = r2.
Thus, r2 - 20r + 75 = (r - 5)(r -15) = 0 and the radius of the sphere is 5 or 15. These two radii account
for the two spheres in the problem. Hence the diameters are 10 and 30 and our desired sum is 40.
181. Since A ABC is an isosceles right triangle, we can describe the vertices A, B, and C as {k, 0),
(-k, 0), and (0, k), respectively. (Why?) Since point P is on AB, it is on the y axis and we can describe
it by (p, 0). Hence, we have 2CP2 = 2( \Jp2 + k2)2 - 2p2 + 2k2. Similarly, we find
AP2 + BP2 = (y/(k - p)2)2 + (\J(-k - p)2)2 = k2 - 2kp + p2 + k2 + 2kp + p2.
Thus, we have AP2 + BP2 = 2p2 + 2k2 = 2CP2.

182. Let a = x + y. Thus, we have \a + z| + \a - z| < 8. Since \b\ + \c\>b + c for all b and c, we have
8 > \a+z\+\a-z\ > 2a. Thus, we have a = x+y < 4. Similarly, we have 8 > \a+z\+\a-z\ > a+z-(a-z) = 2z.
Thus we have z < 4 as well. Our region then becomes 0<x + y<4, 0<z<4. Thus, we have a right
prism whose base is a right triangle (bounded by the line x + y - 4 and the x, y axes) and whose
height is 4. The vertices of our triangle are (0,0), (4,0), and (0,4). Hence the area of the triangle is
4(4)/2 = 8 and our volume is (base area)(height) = (8)(4) = 32.
183. Since we have a line perpendicular to a plane, the fact that a • b = 0 if and only if a _L b is
useful. Let the three edges of the rectangular box be described by x = AD, y - AB, and z = AA\.
Since C\ is across the box from A, we have AC\ = AD + DC + CCi = x + y + z (since DC = AB and
AA\ - CC\). Furthermore, we find A\D = AD - AA\ = x - z and A\B = AB - AA\ = y - z. Since the
edges of a rectangular box are perpendicular, we have x-y-y-z-z-x - 0. Similarly, AC\ being
perpendicular to AiBD gives us AC\ ± A\D and ACi ± A\B. From the first we have
0 = (x + y + z) ■ (x-z)
= x-x + y-x-y-z-z-z
= ltfl-114
so that ||x|| = ||z||. Similarly from AC\ _L A\Bwe find ||y|| = ||z||. Thus, the lengths of the three different
edges of the box are all equal, so the box is a cube.
184. Recalling our work in the chapter with tetrahedrons, we decide to try vectors! Let the origin
be the centroid of ABCD. Hence, we have A + B + C + D = 0 and we wish to show E + F + G + H = 0.
Subtracting these, we find that we can also prove the problem by showing
AE + BF + CG + DH = 0.
Since BD x CD is normal to BCD, it is parallel to AE (since AE is on the altitude from A to BCD).

Hence, we write
BD x CD ~(D-B)x(D-C) = BxC + CxD + DxB.
Since ||BD x CD\\ = 2[BCD], ha[BCD]/3 = V, and ||A£|| = k/ha (given), we have
. _ k _ k[BCD] _ k\\BD x CPU

11 11 ha 3V 6V
Since AE and BD x CD are in the same direction, we have
AE = ~ (BxC + CxD + DxB).

6V v y
We find similar expressions for BF, CG, and DH. Putting them all together, we find
— (AE + BF + CG + DH) =
(BxC + CxD + DxB) + (A xD + DxC + CxA) +
(A xB + BxD + DxA) + (AxC + CxB + BxA) = 0
as a consequence of repeated application of x X y = -y x x. Hence AE + BE + CG + DH = 0, so

E + F + G + H- 0, and the centroid of EFGH is also O.
Chapter 13
Equations and Expressions
13-1 What did you learn?

13-2 Let y - 1; from the repeated equation -3y - z - -1 we find z = -2 and using these we find
,
(x, y, z) - (7,1, -2). Similarly, we can pick y = 0 to get the solution (-1,0 1), and so on.
13-3 Eliminating x from the first two equations would leave
x + y + 3z = 2
-3y - z - -1
-3y -z = 0.
Since no solution satisfies the last two equations simultaneously, there are no solutions to this system.
13-4 Eliminating the x's is easier than eliminating the z's.
13-5 First we set up our matrix:
/4 2 1 3\
2-316.
\1 -3 2 6J
Multiplying the third row by 4 and subtracting the first row leaves
/4 2 1 3\
2 -3 1 6 .
\0 -14 7 21/
Multiplying the second row by 2 and subtracting the first row from it leaves
/4 2 1 3\
0 -8 1 9 .
\0 -14 7 21/
< 75 >
76 > CHAPTER 13. EQUATIONS AND EXPRESSIONS
Finally, we multiply the last row by 4 and subtract 7 times the second row from the result (to make
the second element in the last row 0) to get
/4 2 1 3\
0-8 1 9 .
\0 0 21 21 /
Writing this in equation form we have
4x + 2 y + z = 3
-8 y + z = 9
21z = 21
From these we quickly find (x, y, z) = (1, —1,1).

13-6 Just like our other examples in the text, we add the equations since the terms on the left
are patterned, giving
3w + 3x + 3y + 3z = 102.
Hence, w + x + y + z = 34. Since w + x + y = 20, we have 34 = w + x + y + z = 20 + z, so z = 14. Similarly,
we use w + x + z = 22 to get y - 12 and the other two equations to get (w, x, y, z) = (-2,10,12,14).
13-7 Buoyed by our previous success with addition, let's try adding them:
xy + yz + zx = 12 V6 + 54 V2 + 48 V3.
Not too helpful. How else can we combine the three equations to get a nice symmetric equation?
Try multiplication:
(xy)(yz)(zx) = (12 V6)(54 V2)(48 V3) = 12 • 54 • 48 • 6.
Hence, {xyz)2 = 12 • 54 • 48 • 6, or xyz = Vl2 • 54 • 48 • 6 = 432. Since xy = 12 V6, we can divide this
equation into the one for xyz, or
xyz _ 432
xy 12 V6'
so z = 6 V6. We can do the same to get x and y and we find (x, y, z) = (4 V2,3 V3,6 Vf>). Our bag of
tricks is apparently not limited to addition!
13-8 For the factorization of an - bn, consider the sum
an~l + an~2b h-f abn~2 + bn~l.
This is an n term geometric series with first term an 1 and common ratio b/a. From our discussion of
geometric series in Volume 1, we have
n-2 ,„-i _a”-b"

an~l + an~2b + • • • + abn~A + bn~l =
1-z a-b
Multiplying by a-b gives the desired factorization. The factorization of a2n+l + b2n+1 is similarly
proved by viewing
a2n - a2n 1b -t-- ab2n 1 + b2n
as a 2n + 1 term geometric series with first term a2n and common ratio -b/a.
13-9 Write the product as
(a + b c d) (a + b + c + d) (a b c -{- d).
The resulting terms are of the form a3, a2b, or abc. All the cubed terms can occur in only one
way. The terms with one squared and one linear variable, such as a2b, can each occur in 3 ways
since the linear term (b for a2b) can be taken from each of the three terms and the squared term
from the other two for a total of three ways to get a2b. If you don't see this, write the product as
(«i + b\ + c\ + di)(ci2 + b2 + C2 + ^2X^3 + + C3 + c/3) and note that the three a2b terms are a^b?,, aia^, and
02^3^1- Now we move to the terms like abc. The a can be selected in 3 ways and the b can be chosen
in 2 ways from the remaining two terms, and the c can be picked in only one. Hence, abc occurs six
times (try finding them all in our above subscripted product). Hence, our desired expansion is
a3 + b3 + c3 + d3
+3 (a2b + b2a + a2c + c2a + a2d + d2a + b2c + c2b + b2d + d2b + c2d + d2c)
+6 (abc + abd + acd + bed).
13-10 The common denominator is w2x2y2z2 and our desired fraction is
w2yz + wx2z + zvxy2 + xyz2

w2x2y2z2
13-11 Writing the desired expression with a common denominator, we have
1 1 1 z + x + 1/
— + — + — = ---
xy yz zx xyz
13-12 We can permute the three solutions among the labels x, y, and z in 3! = 6 ways. The
solutions are (5,6, -4); (6,5, -4); (5, -4,6); (6, -4,5); (-4,5,6); and (-4,6,5).
13-13 Writing the expression with the common denominator (a - b)(a - c)(b - c) we have
(b + c)(b - c) - (c + a)(a - c) + (a + b)(a -b) _ b2 - c2 + c2 - a2 + a2 - b2 _

(a - b)(a - c)(b - c) {a - b)(a - c)(b - c)
as desired. Remember that sometimes the algebra is simple enough to use instead of our clever
polynomial approach.
185. Factoring a3 -b3 as a difference of cubes, we find (a - b)(a2 + ab + b2) = 19x3. Since a -b - x, we
find a2 +ab + b2 = 19x2. Letting b - a-x, we can substitute for b and find a2 + a(a -x) + (a-x)2 = 19x2,
or a2 - ax - 6x2 = 0. Factoring this gives (a - 3x)(a + 2x) = 0, so a - 3x or a = -2x.
186. Seeing the pattern in the variables, we add the equations, yielding
(x + y + y + z + z + x)(x + y + z) = 66 + 77 + 99.
Simplifying the equation, we have 2(x + y + z)2 = 242, so x + y + z = ±11. Trying the positive result,
we write x + y + z = 11 and we can write the equations as
ll(x + y) = 66
ll(y + z) = 99
ll(z + x) = 77.
From the first equation we find x + y - 6. Since we know x + y + z = 11, we have 6 + z = 11, or z = 5.
Similarly, we use the second equation to get x and the last to get y. We find that (x, y,z) = (2,4,5).
Trying the negative solution for (x+y+z), we find (in the same way as above) the solution (-2, -4, —5).
187. Seeing the expressions a2 + b2 + c2 and ab + be + ca, we are reminded of our expression
(a + b + c)2 = (ia2 + b2 + c2) + 2 (ab + bc + ca).
Since all perfect squares are nonnegative, we can write
(a2 + b2 + c2) + 2(ab + bc + ca) > 0,
or ab + be + ca> -(a2 + b2 + c2)/2. Since a2 + b2 + c2 = 1, the minimum value of ab + be + ca is -1/2.

188. Since l22 isn't a perfect cube, we can't use those factorizations. We don't have a difference
of squares, either, so we 'force' the presence of a perfect square by introducing a 2(2n) term:
222 + 1 = 222 + 2(2n) + 1 - 2(2n).
Since 222 + 2(2n) + 1 = (211 + l)2, we have
(211 + l)2 - 212 = (211 + 1 - 26)(2n + 1 + 26),
where we have factored as a difference of squares. We can easily evaluate these factors as 1985 and
2113. Testing these, we find 2113 is prime but 1985 factors into (5)(397), so 222 + 1 = (2113)(5)(397).
189. Letting x, y, z, and w be the numbers, the given information is
x+y+z
w+ 29
~3~~
w+y+z
X + 23
^3~
W +X+z
y+ 21
w+x+y
z+ 17.
~3
Seeing the pattern on the left, we add the equations to find 2(w + x + y + z) = 90, or w + x + y + z = 45.
Using this and the first equation, we have x + y + z = 45 - w, so w + (x + y + z)/3 = w + (45 - w)/3 =
2w/3 + 15 = 29, and w = 21. Continuing in this manner for each of the other equations, we find that
(w, x, y, z) = (21,12,9,3).
190. Writing the second term with a common denominator (xyz) and doing the same for the
fourth, our product is
/ 1 \ fzy + xz + yx\ f 1 \/z + x + y\

\x + y + zj\ xyz J\xy + yz + zxJ\ xyz )'
Cancelling the common terms (x + y + z) and {xy + yz + zx), the product is merely 1 lx1y1z1.
191. First we can evaluate the last term \/3 - 2 V2 as \f (-1 + V2)2 = -1 + V2. We can evaluate
the numerator of the first expression in exactly the same way we evaluated similar expressions
in the chapter. We write z = \J yfb + 2 + \J y[5-2. Squaring both sides of this, we have z2 =
V5 + 2 + 2 V5 - 4 + V5-2 = 2V5 + 2. Hence, z = \/2 V5 + 2 (since we want the positive value of z).
Thus, our first term in the desired expression is
\/'V/5 + 2 + y/V5-2 _ a/2V5 + 2 _ V2y/V5 + 1 _ ^
\/ V5 + 1 \j \jb + 1 y/ V5 + 1
The answer is V2 - (-1 + V2) - 1.

192. Seeing equations near to those in the relationship between roots and coefficients of a
polynomial, we look for a way to manipulate the given equations into this form. The first equation
is for x + y -z, so we let w = -z and the equations become
x + y + iu = 0
-wx - xy - yw - 27
-xyw = 54.
Rewriting the last two equations as wx + xy + yw- -27 and xyw = -54, we let x, y and ze be the roots
of a polynomial, which is then t3 - 27f + 54 = 0. Factoring, we find f3 - 271 + 54 = (f — 3)2(f + 6) = 0, so
the solutions for t are 3,3, and -6. Letting each of the variables (x, y, and w) in turn take the value -6
and the other two variables equal 3, we find the solutions (remember z = -w) (x, y, z) = (-6,3, -3),
(3,-6, -3), and (3,3,6).
193. The system has no solution if
n 1 0
0 n 1 = 0,
1 0 n
or n3 + 1 = 0. Hence, if n = -1 we cannot find a unique solution. To show that this is a 'no solution'
situation rather than infinitely many solutions, let n = -1 and add the three equations to get 0 = 3.
194. Number the vertices from 1 to 20; let V, be the number at vertex i. Let S be the common sum
shared by all of the faces. For each face, we can write a sum of 5 V/'s which equals S. We have 12
such equations, and each Vj occurs in three of them (since each vertex is a vertex of three faces). If
we add the 12 equations, we get
20
3 J2 vi = 12S'
i=1
where the summation is a result of each Vj occurring in 3 of the twelve equations. Since
20
Vi = 1 + 2 + • • • + 20 = 210,
i=1
we have 12S = 630. Thus, S = 52.5; however S must be an integer since it is a sum of the five integers
at each vertex. Hence, we have a contradiction, so we cannot number the vertices as described.
195. As discussed in the chapter, to show that two cubic polynomials are identical, we need only
show that they are the same at four points. Let f(x) = a\(x +1)3 + ai(x + 2)3 + as(x + 3)3 + a^(x + 4)3 and
g(x) = (2x + l)3. Seeing the coefficients in the linear equations given, we see that /(0) = 1 from the
first linear equation. Similarly we see /(1) - 27, /(2) = 125 and /(3) = 343. Since g(0) = 1, g(l) = 27,
g(2) = 125, and g(3) = 343, g(x) and f(x) are cubic polynomials which agree at four points and thus
are the same polynomial. Matching the coefficient of x3 of f(x) and g(x) gives a\ + «2 + - 8
and f(-5) = g(-5) gives us 64«i + 217az + 803 + 04 = 729.
196. Let x, y, and z be the lengths of the edges of the solid. Hence, xy = 135, yz = 30, and
xz = 50. Since we seek xyz, we take the product of the three equations to find (xyz)2 = 135 • 30 • 50,
so xyz = 450.
197. Putting all expressions on one side of the equal sign and writing them with the common
denominator (a -b)(b- c)(c - a), we find that we must show that
-bc(b + c)(b - c) - ca(c + a)(c - a) - ab(a + b)(a -b) - (a + b + c)(a - b)(b - c)(c - a)
(a - b)(b - c)(c - a)
equals zero. Thus, we let
/(c) = -bc(b + c)(b - c) - ca(c + a)(c -a)- ab(a + b)(a -b) - (a + b + c)(a - b)(b - c)(c - a).
Since /(c) is a cubic equation in c, we can show that /(c) = 0 (and hence the identity holds) by
showing that /(c) has four distinct roots. Looking at each of the terms, we decide to try the roots 0,
a, b, and -a - b:
m = 0 - 0 - ab(a + b)(a -b) - (a + b)(a - b)(b)(-a) = 0;
m = -ba(b + a)(b -a)- 0 - ab(a + b)(a -b)- 0 = 0;

m - 0 - ba(b + a)(b - a) - ab(a + b){a -b) - 0 = 0;
f(-a-b) = -b(-a - b)(-a)(2b + a) - (-a - b)(a)(-b)(-2a - b) - ab(a + b)(a - b) - 0
[(a + b)(ab)][-(a + 2b) + (2a + b)] - ab(a + b)(a - b) = 0.
Thus, we have found four roots to a cubic. These four are all distinct unless -a - b = 0. We can treat
that case separately (and swiftly), just using direct algebra on the original identity. Thus, /(c) = 0
for all c and we have proven the identity.
198. Recognizing -y2 - z2 + 2yz as -(y - z)2, we can write the given expression as
x2 - (y - z)2 + x + y - z.
Now we factor as the difference of squares to find
(x-y + z)(x + y-z) + x + y-z = (x-y + z + l)(x + y-z)

as our desired factorization.

199. We can compare the fractions better if the numerators or denominators are the same. We
can simplify the numerators easily. For example,
a+b-c a+b
-=-1.
c c
If we do this then add one to all the resulting terms in the equation we find
a+b_a+c_b+c
c b a
We can now make the numerators the same by adding 1 to all parts of the equation, yielding
a+b+c _ a+b+c _ a+b+c

c b a
Hence, either the denominators are all the same (a = b - c), from which the desired expression equals
(2a)(2a)(2a) / a3 = 8, or the numerators are all 0 (a + b + c = 0). For this case, (a + b)(b + c)(c + a)/abc =
(-c)(-a)(-b) / abc = -1.
200. Generally, equations are easiest to attack if all the variables are on one side. Hence, we
divide by the (a + b) or the ab terms. Trying the former isn't promising (try and see), but the latter
yields
1 = 2(l/b + l/a)
1 = 3(l/c + l/b)
1 = 4(1 /a + 1/c).
We see now that dividing each equation by the coefficient of the reciprocals gives a nice patterned
system. We then add the equations to get 2(1 /a + l/b + 1/c) = 1/2 + 1/3 + 1/4, so 1/a + l/b + l/c = 13/24.
We combine this with the first equation above to get 1 = 2(13/24 - 1/c), so c = 24. Similarly, a = 24/5
and b = 24/7, and 5a + 7b + 9c = 264.
201. Seeing the cubes, we cube both sides, yielding
(x + 9) - 3y/(x + 9)2(x -9) + 3y/(x + 9)(x - 9)2 - (x - 9) = 27
-3 \J(x + 9)(x -9)(y/x + 9 -y/x- 9) = 9
-3ilx2 - 81(3) = 9.
Dividing by 9 and cubing to find x2, we get x2 = 80.

202. Recognizing expressions as squares of binomials, we write
a2 + b2 - c2 + 2ab _ (a + b)2 - c2
a2 + c2 -b2 + 2ac (a + c)2 - b2
Factoring these as the difference of squares, we have
(a + b + c)(a + b - c) _ a + b - c
(a + c + b)(a + c - b) a-b + c'
203. Seeing the cube roots, we think of cubing. Let
x= V5 + 2V13+ ^5 - 2 Vl3
and cube this to get
x3 = 10 + 3 ^5 + 2 Vl3^5 - 2 Vl3 (^5 + 2 Vl3 + \/5-2Vl3)
x3 = 10 + 3^25-52 (x),
so we have x3 + 9x -10 = 0. Factoring, we get (x - l)(x2 + x + 10) = 0. Since the quadratic has no real
roots, the only solution is x = 1.
204. We see lots of nice symmetric expressions. Recall that
(x + y + z)3 = x3 + y3 + z3 + 3(xy2 + x2y + yz2 + y2z + zx2 + z2x) + 6xyz,
so (x + y + z)3 = 4 + 3(12) + 24 = 64 and x + y + z = 4. Since
(x + y + z)(xy + yz + zx) = xy2 + x2y + yz2 + y2z + zx2 + z2x + 3xyz,
we find 4(xy + yz + zx) = 24, so xy + yz + zx = 6.

205. We write g(x12) = g(x)h(x) + r(x). Since degr(x) < degy(x), degr(x) < 5. Like other similar
problems, we find the roots of g(x) and use these in the initial equation above. Since (x-l)g(x) = x6 -1,
the roots of y(x) are the five sixth roots of unity besides 1. Hence each solution of g(x) satisfies
af - 1, so g(a}2) = g(l) = 6. Since g(ctj) = 0 for each of the five on, the initial equation evaluated at
these 5 points gives r(a*) = 6 for each on. Hence the polynomial r(x) - 6 has 5 distinct roots. Since the
degree of r(x) - 6 is at most 4, r(x) - 6 must therefore be everywhere 0. Hence, r(x) = 6 at all points x
and the remainder is 6.
206. Since (ifx)2 = i/x, we can combine the first and third equations as
16 + i/y + yfz = (4 + i/y + x[z)2,
from which we find (after a bit of algebra)
4 ^Jy + 4 yjz + ^Jyz = 0.
Since y and z are positive, this equation can never be satisfied, and there are no solutions to this
system.
207. Since the equation is a quadratic in a (or in b), we can apply the quadratic formula to find
-4 ± Vl6 + 16b2 + 16b -1 ± Vl + b + b2

a = ----=-.
8 2
Since b2 < b2 + b + 1 < (b + l)2, the quantity b2 + b + 1 cannot be the perfect square of an integer for
any positive integer b. Hence if b is an integer, a cannot be.
208. From the given equation, we know that x + 2 divides x2 + 3. We wish to find a linear
factor, or better yet, a constant term, which x + 2 divides. Since x + 2 divides x2 + 3, it also divides
x2+3-x(x+2) = 2x-3. Similarly we can knock offthex by noting that x+2 divides 2x-3-2(x+2) = -7.
Hence, x+ 2 equals either 1 or 7 (since we want positive integer solutions). The first gives a negative
solution for x, but for x + 2 = 7 we have x = 5, from which y - 4. Our only solution is {x, y) = (5,4).
209. Let A = be - a2, B = ca - b2, and C = ab -c2. We can link the given expression in the equality
to the desired expression by noting that
1 1 1 a b c a i b _ c b+cc+aa+b
A + B + C A + B + C ^V + B2 + C2 + ~BC + ~AC + ~AB
a b c t (b + c)A + (c + a)B + (a + b)C
A2 + B2 + C2+ ABC
Since direct algebra reveals (b + c)A + (c + a)B + (a + b)C = 0, if (1/A + 1/B + 1/C) = 0, we must have
(a/A2 + b/B2 + c/C2) = 0.
Chapter 14
Inequalities
14-1 The given expression is the perfect square of (2x - 3y). Since (2x - 3y)2 > 0, we have the
desired inequality.
14-2 Since the less than side is a perfect square, we can write (xy + l)2 > 0. Make sure you see
why this is nofuseful: the (xy + l)2 in the problem is on the less than side, not the greater than. Let's
try multiplying out the two sides instead:
x2y2 + x2 + y2 + 1 > x2y2 + 2xy + 1.
Rearranging this we have x2 - 2xy + y2 > 0. Since this expression is (x - y)2, we have the desired
inequality. Make sure you see that all of our steps are reversible.
14-3 To determine when equality holds, we examine the equality conditions on our three uses
of AM-GM. From (x2 + y2)/2 > xy, we have the condition x = y. From the other two we find y = z
and z - x. Hence, in order for equality to hold, we must have x - y - z.
14-4 By the AM-GM Inequality, we have
The right hand side is 1, so we have f + \ > 2. From the equality condition for AM-GM, this sum
equals 2 when a/b = b/a, or a = b (since they are both positive).
14-5 Since the square of a real number is nonnegative, any sum of squares is nonnegative as
well. Hence, we have
(a\x + b\)2 + (a2x + b2)2 + (a3x + b3)2 + • • • + (anx + bn)2 > 0.
We can write the left side as a quadratic expression by grouping x2 and x terms. This gives us
^ *2+(^2 it ^x+ibtf ^ °-
< 84 >
Since this quadratic is always greater than or equal to zero, it must have 1 or 0 real roots. If a
quadratic has two roots, it must at some point cross the x axis and therefore be negative. Since the
quadratic has no more than 1 real root, its discriminant must be 0 or negative; thus, we have
(2E«a)2-4(e4) (e*?)so.
A simple rearrangement of this yields Cauchy's inequality. The equality condition follows from
noting that the original quadratic can equal 0 if and only if all n squares equal 0 for the same x, so
diX' + bi = 0 for all i for some specific x'. Thus, bj/cii must be constant is the equality condition.
14-6
This is an example where assuming x - y - z will send us down the wrong path. Since
we're maximizing, we want xyz on the smaller side. Using AM-GM on the three quantities in the
sum, we find
2x + y + Z , rz-
-j--> y 2xyz.
Cubing both sides we find xyz < 32. The equality, or maximum value, is obtained when 2x = y - z
(the equality condition of AM-GM), or 2x + y + z = y + y + y = 12. Hence, we have equality when
y = z = 4 and x - 2. The maximum value is then 32.
14-7 No. We can only use 'without loss of generality' when the objects of interest are indistinct.
When we have x + y = 4, x and y are virtually indistinct, as exchanging them results in exactly the
same equation. Unfortunately, when 2x + y - 4, reversing x and y changes the problem to x + 2y = 4,
which is different from the original. We cannot proceed 'without loss of generality;' we must find
another way.
14-8
This statement follows immediately from the equality condition of AM-GM applied to the
prior example, from which a = b = c and A ABC is equilateral.
14-9 Using summations to denote the sums we have
14-10 For n = 1, there is no root and the resulting expression is the Arithmetic Mean. For m = 2
we have the root mean square, so if m - 2 and n = 1 the Power Mean Inequality is RMS > AM.
210. Since (x+yj1 = x2 + y2 + 2xy = 1 + 2xy, our problem is maximizing xy. From the Trivial Inequality,
(x2 + y2)/2 > \Jx2y2 = xy. Thus, xy < 1/2. Since the maximum value of xy is 1/2, the maximum of
(x + y)2 is 1 + 2(1/2) = 2. This is attained when x = y = ± V2/2.
211. Let x and y equal some very big number, say n, and letz = l/n2. Thus, xyz = (n)(n)(l/n2) - 1
as required, but the three sums in question are 2n, n + l/n2 and n + 1 In2. All three of these are at
least as large as n. Since we can make n as large as we want, min{x + y, x + z, y + z} has no maximum.
86 > CHAPTER 14. INEQUALITIES
212. As an example in the text, we showed that the given expression is greater than or equal to
one by using Cauchy's Inequality From the equality condition for Cauchy's Inequality, we have
(cos3 a)/ cos jS _ (sin3 a)/ sin jS

cos a cos jS sin a sin jS
from which we find tan2 a = tan2 (i. Since a and are both first quadrant angles, this implies a = jS.
213. From AM-GM, we have
A + 2B + 3C + 4D
> {/0)(2B)(3C)(4D),
4
so {/24ABCD < 2. Finally, ABCD < 16/24, so that the maximum value of the product is 2/3.
214. Without loss of generality, let x <y<z. Then, the largest of the three sums in question is
y+z. We can relate the given product to a sum with AM-GM, so (x + y + z) > 3 ^jxyz = 3. Since x is the
smallest of the three numbers, x < 1 (if x > 1, then xyz > 1, a contradiction). Hence, y + z>3-x>2.
Thus, the minimum value of y + z is 2 and this is attained when x = y = z = 1.
215. Without loss of generality, let
x < y < z. Hence, the smallest of the desired products is xy
and our problem is to maximize this product. Since x + y = 3 - z, we have
x+y _ 3-z
~ 2 2
" ~ ~
from AM-GM. Since z is the largest of the three numbers, it must at least 1 (otherwise the sum of the
three would be less than 3). Using this in the above expression, yfxy < 1, so that the maximum of xy
is 1 as desired. This can be achieved when x = y = z = 1.
216. Rearranging the given expression we have
a+b> (a2 + b2)/2 + yfab.
Squaring both sides of this gives
a2 + 2ab + b2 > (a2 + b2)/2 + ab + 2 \Jab(a2 + b2)/2,
which rearranges to (a2 + 2ab + b2)/2 > ^4ab(a2 + b2)/2, or
> yiab^TW).
Squaring out the left, we have
^2> + (2afc)>Vw + n
which is clearly true by AM-GM.
217. We see a product of n terms and an nth power. It looks like AM-GM. Rewrite the greater
than side as
f (x + ri) + (x + r2) + • ■ ■ + (x + rn)\n
n
Now, applying AM-GM to the terms (x + r{), (x + r2),..(x + rn), we have the desired result.
218. As written, the solution is very elusive. Let's try one of our little helping hints: take the
reciprocal of each side, yielding (after rationalizing the denominator and remembering to reverse
inequality sign) \fn + Vn - 1 >100. Now the answer is much clearer. Since V2500 = 50, for n = 2500
and all smaller n, we fail to satisfy the inequality but for n = 2501 the inequality is satisfied.
219. Seeing the absolute value signs, we consider the Triangle Inequality, or |x + y\ < |x| + \y\. Note
that since \y\ - \-y\, we can write |x| + \y\ > |x - y\ as well. In the problem, we have 10 terms of the
form |x - F(\. From the Triangle Inequality, we can write |x - F;| + \x- Fj\ > |(x Ff) - (x Fj)\ = |Fy - F,;|.
- -
Note that we have gotten rid of the x. Continuing this, we can use the Triangle Inequality to combine
all 10 terms:
10
f(x) - ^ \x ~ F/l > I ± (x - Fi) ± (x - F2) ± • • • ± (x - Fio)|,
i=1
where the ± signs show that we can use either plus or minus signs at each of these points and the
inequality will still be valid. (This is a result of the fact that |(x - Fi)\ = |-(x - F,-)|.) Since we want
to find a constant such that /(x) > c, we must choose 5 of the ± signs to be negative and 5 to be
positive to cancel out all of the x's. Since we can choose any five to be positive and still have a
valid inequality, we must choose the five so that the resulting constant is as great as possible. (For
example, if z > 5 and z > 6, the minimum value of z is 6 and the inequality z > 5 is irrelevant.) Since
our resulting sum is 5 positive Fibonacci numbers and 5 negatives, we choose the greatest 5 to be
positive, so that
/(x) > (55 + 34 + 21 + 13 + 8) - (5 + 3 + 2 + 1 + 1) = 119.
Challenge: Can you find all x that make /(x) = 119?

220. Lemma: If /(x) and g(x) are such that /(x) = anxn + a„_\xn~l + ■ ■ ■ + ao and g(x) = bnxn +
bn-\xn~l + ■■■ + bo and ai > bi for all z, then /(x) > g(x) for all positive x. Proof: /(x) - g(x) =
(1an - bn)xn + {an-1 - bn-\)xn~l -\-+ (ao - bo). Since 0; > bi for all i, (a* - bf) > 0, so the coefficients of
/(x) - y(x) are all positive. Thus, for all positive x, /(x) - g(x) > 0, and /(x) > ^(x).
Applying this lemma to the given problem, we see that if we can show that the coefficient of xk
on the left is greater than that of xk on the right for all k, then we have proven the given inequality.
On the left, the coefficient of xk is the sum of the products of the roots taken n - k at a time. (Recall
the discussion of the relationship of the roots of a polynomial to its coefficients.) From the Binomial
Theorem, the coefficient of xk on the right hand side is (^) (Vrir2 * ■ • rn)k. Now, apply AM-GM to the
terms in the sum which forms the coefficient of xk on the left side. Since there are (^) terms on the
left, and each r, appears in (nkZ\) of the terms (since after choosing a specific r,-, we must choose k - 1
of the remaining n - 1 rf s for the other k - 1 members of each term), we have
r\r2 • • • rk + • • • + r„jt+iN-/c+2 Ml)

- > ((rir2r3-*-r„)^)
0
from the AM-GM Inequality. Multiplying this by (jj) on both sides and noting that (j_j)/(j) = k/n,
we have , ,
n k/n
r\r2 • • • rk + • • • + rn-k+1rn-k+2 • • • rn > (rir2r3---rn)
k,
which is what we wanted to prove. Make sure you follow the algebra and the logic; it's a bit tricky.
Make sure you also understand why the above expressions are the coefficients of the polynomials
described in the original question.

221. Let the roots be r, s, and t. From the given form of the polynomial, we know r + s + t = 12
and rst = 64. Thus, the geometric mean of the three numbers equals the arithmetic mean. From
the AM-GM Inequality (which we can use since these three numbers are positive), we know that if
equality holds we must have r - s - t. Flence, the roots are all equal to 4. Finally, a = rs + st + tr = 48.
222. If we get rid of the radicals by squaring each side and then raising each side to the nth power,
we have nn < (n!)2. We can write this out as
n • n ■ n • • • n • n < [1 • 2 • • • (n - 1) • n][l • 2 • • • (n - 1) • n\.
We can prove this if we can group the two bracketed products into n pairs so that one number in
each pair comes from each set of brackets and each pair has a product greater than n. Make sure you
see why this will complete the problem; there are n pairs on the right and n n's on the left. If each
pair on the right has a product greater than or equal to n, then the right is greater than or equal to
the left.
Now, how do we pair the terms? Since 1 • 1 is not greater than n, we don't want to pair k with k
for each k. Let's try another way: pair 1 with n, 2 with (n - 1) and so on, so that we have
[(1)(«)][(2)(„ - 1)] • • ■ [(B)(1)] = (n!)2.
Now we must show that each of these products of the form k(n - k + 1) is greater than or equal to n.
Since k(n -k + 1) = k(n — k) + k = (k — 1 )(n — k) + n and (k - 1 )(n - k) > 0 for all k from 1 to n, we have
k(n — k + 1) > n. Make sure you see this. Thus, (n\)2 > nn and the problem is complete.
223. Seeing the product of sums, we think of Cauchy's Inequality, from which we have
(a + b + c + d)(ha + hb +hc + h^) > (\Jaha + \jbhi7 + \fchc + y/dhj)2.
Since V - aha/3 (and similarly for the other faces), the right side above is (V3V + V3V + V3V + V3V)2 =
481/, completing the proof.
224. Expanding the left hand side yields
x4 + y4 + z4 + lx2y2 + 2y2z2 + 2z2x2 < n(x4 + y4 + z4).
Since the x2y2 terms are on the less then side, we apply AM-GM using these as the geometric means,
or x2y2 = yfx4^4 = {x4 + y4)/2. Hence,
lx2y2 + 2y2z2 + 2zV < 2[(x4 + y4)/2] + 2[{y4 + z4)/2] + 2[(z4 + x4)/2],
and combining this with the above expression, we find
(x2 + y2 + z2)2 < 3(x4 + y4 + z4),
son-3. (See if you can find a faster solution using Cauchy's Inequality.)
225. Let the distance travelled be z, the time of Car A be s, and that of Car B be t. Car B is easy;
ut/2 + vt/2 = z, so z/f = (u + v)/2 and the average rate is the arithmetic mean of the two rates of the
car. Let the time that Car A travels at rate u be r. Hence, ur = v(s -r) = z/2 and ur + v(s -r)=z. From
the first equation, r = vs/(u + v) and substituting this in the second gives z/s = 2uv/{u + v). Perhaps
you recognize this as the average rate of the car being the harmonic mean of the two rates it travels.
We thus must show that
-<-
2 uv u+v
u+v 2
Multiplying both sides by 2(u + v) and subtracting 4uv, we have (u - v)2 * > 0, which is clearly true
and we have solved the problem.
226. We wish to show that ab > 4rR. From the Triangle Inequality, we have a + b > c. (We think
to use this because the problem involves a strict inequality). Since the problem also involves R and
r, we use the law of sines and [ABC] = rs. The expression ab occurs in [ABC] = (ab/2) sin C, so using
this we have
ab = 2[ABC]/sinC = 2rs/(c/2R) = 2rR(a + b + c)/c > 2rR(c + c)/c = 4rR,
as desired.
227. From Heron's formula we have [ABC] = 15 V7/4. We can also write the area in terms of the
sides and the given /, m, and n as (al + bm + cn)/2 = 15 V7/4. Seeing this sum of products and the
given sum of squares, we write out Cauchy's Inequality:
(a2 + b2 + c2)(l2 + m2 + n2) > (al + bm + cn)2.
Using all of the given information on the sides, the area, and the sum l2 + m2 + n2, we find that the
left and the right sides of the above expression are equal. Thus, the equality condition for Cauchy's
Inequality must hold, or a/l - b/m = c/n. Writing m and nin terms of I and using l2+m2+n2 = 225/44,
we find l = 30 V7/77.
228. We are asked to show that for two positive numbers, RMS - AM > GM - HM. Adding AM
to both sides and subtracting GM, after a bit of algebra on the less than side we have
(a ~ fr)2
2 (a + b)'
Squaring both sides and simplifying the greater than side, we have
(a + b)2 (a - b)A
(2ab)(a2 + b2) >
2 4 (a + b)2'
Now we put the radical expression on the right and the other two on the left. On the left, we find a
common denominator, multiply both sides by 2(a + b)2, and we find
2 (a + b)4 -(a- fr)4

\J (2 ab)(a2 + b2) or
4 (a + b)2
2 (a + bf -(a- bf
yj (2 ab)(a2 + b2)(4)(a + bf.
2
The expression on the right is suggestive of AM-GM; we can get an AM on the left by factoring as
the difference of squares:
(a + b)4 + [(a + bf -(a- bf]

yj(2 ab)(a2 + b2)(4)(a + bf, or
2
(a + bf + [4ab][2(a2 + b2)]
sj(2ab)(a2 + b2)(4)(a + bf,
2
where we note (a + fr)4 - (a - fr)4 = [(a + b)2 + (a - b)2)[(a + b)2 -(a- b)2]. By AM-GM this final inequality
is true, so since all steps are reversible, we have proven the inequality.
229. Seeing reciprocals, we try the AM-HM Inequality on a, b, and c, and we find
a+b+c ^ 3
3 -1a + 1b + 1'
c
Hence, ^ + ^ + ^>3/2. Now, seeing squares, we try RMS-AM, and we have
(a + 1/b)2 + (b + 1/c)2 + (c + 1/a)2 a + 1/b + b + 1/c + c + 1/a

-->---> 5/2.
Squaring and multiplying by 3 yields the desired inequality.

230. Let group i have pi people. Hence, p\ + p2 + • • • + pm = n. Similarly, let s\ be the side length
of the kth cake. Since no person has more than 25 cm2 of cake, s2 < 25pk, or < 5 yjp^. The total
amount of ribbon needed for the cakes is R - 4si + 4s2 H-1- 4sm, which from above we know is less
than or equal to 20 yjp{ + 20 ^[pi H-1-20 y/pj/. If we can now show that this second sum is less than
or equal to 20 yjmn, we will have
R < 20 yjpl + 20 yfpi H-+20 yfpj/ < 20 yfmn,
so R < 20 yfmn as desired. Sums of products; looks like a job for Cauchy's Inequality, from which
we have
[(^/fh)2 + (y/p2)2 + • • • + (y/p^)2][l2 + l2 H-+ l2] > (^/Pl + ^/p2 + • • ■ + yfp//i)2

(pi + P2 + • • • + Pm){m) > ( y/pi + yjpi + • • • + yjp^/)2
nm > (yfpl + y[p2 -\-+ yjp///)2
Taking the square root of this and multiplying by 20 gives the desired inequality.
231. First, we write the area of a general triangle in two ways as follows:
[ABC] = rs = s(s - a)(s - b)(s - c).
From the AM-GM Inequality, i/(s - a)(s - b)(s - c) < (s - a + s - b + s - c)/3 = s/3. Hence,
rS = S]j{s-a)(-s-b'l<~s~b= Vi(^(s-fl)(s-i,)(s-c))3/2< ViV^+27.
and we have r < s/3 V3 for any triangle. Applying this to each face of the tetrahedron, we find
sum of semiperimeters
rA + rB + rc + rD<--.
3 V3
Since each side belongs to two triangles, the sum of the semiperimeters of the triangles is the sum of
the sides of the tetrahedron, or 3. Thus, rA + rB + rc + rD < V3/3 and equality holds when all sides are
equal. We see this equality condition from the original AM-GM inequality, where s-a = s-b = s-c
implies the triangle is equilateral. Four equilateral faces means the tetrahedron is regular.
232. Let O be the circumcenter of A ABC. Since LAOB = 2 LACB and similarly A
for the other two angles of A ABC, we can write
P = [BOC] + [.AOC] + [AOB] = (r2/2)(sin2A + sin2B + sin2C).
By equating angles inscribed in the same arc, we find that the interior angles of
A'B'C' are (/A + zB)/2, (ZB + zC)/2, and (LA + zC)/2. Thus,
Q = (r2/2)[sin(A + B) + sin(A + C) + sin(B + C)].
Applying AM-GM to the sines and then using some of our trigonometric goodies (watch the steps
closely and make sure you understand the identity used in each one), we have
16Q3 = 16(r6/8)[sin(A + B) + sin(B + C) + sin(A + C)]3

> 27r6 [2 sin(B + C) sin(A + C) sin(A + B)]
= 27r6[cos(A - B) - cos(A + B + 2C)] sin(A + B)
= 27r6[cos(A - B) + cos C] sin(A + B)
= 27r6[cos(A - B) sin(A + B) + cos C sin(A + B)]
= (27/2)r6[sin2A + sin2B + sin(A + B + C) + sin (A + B - C)]
= (27/2)r6(sin2A + sin2B + sin2C)
= 27r4(r2/2)(sin2A + sin2B + sin2C)
= 2 7r4P.
Make sure you understand each of the steps! There are many subtle trigonometric identities, such
as sin(A + B - C) - sin(180° - 2C) = sin2C, used here; make sure each is clear to you.
233. Let Sk = x\ + H-nj. From Cauchy's Inequality, we find
> (x\ + x\ H-1- Xkn)2,
or Sk+iSk-i > S2. Starting from k = 20 and listing through k = 92, we have
S21S19 ^ ^20
S22S2O ^ S21
S93S91 >
Taking the product of all of these and cancelling out terms which appear on both sides, we have the
desired S19S93 > S2oS92-
234. Labelling the sides and angles a, b, c, A, B, C as usual, A

we first can find IRPQ = 90°, IPQR = /.BCR = 30°, and iQRP =
60°. The circumcenter of right triangle AMC is the midpoint of the
hypotenuse, so CP = b/2. From the law of sines applied to A CRB,
we find CR - a/ V3 (since IBCR - iCBR = (180° - iCRB)/2 =
(180° - iCMB)/2 = 30°). Applying the law of cosines to APCR, we
find
PR2 = PC2 + CR2 - 2{PC)(CR) cos(C + 30°)
= -7- + -1-(cos C cos 30° sin C sin 30°)

4 3 2V3
Now applying the law of cosines to A ABC to get cos C and noting that 2[ABC]/ab = sinC, we have
(after a bit of algebra)
1 (a2 2 4[ABC]\
PR2
4 U +C + V3 )-
Since PQ = (PR) tan 60° = PR V3 and APQR is right, we have
[ABC]
[PQR] = « = ^
2
Thus, [PQ-R] - [ABC] = (V3/8)(a2/3 + c2) - [ABC]/2. Applying AM-GM to a2/3 and c2, we find
[ABC]
[PQR] - [ABC] >
2
Simplifying, we have [PQR] - [ABC] > ac/4 - [ABC]/2. Since [ABC] = (ac/2) sin B, we have [PQR] -
[ABC] > (ac/4)(1 - sinB), so [PQR] - [ABC] > 0 (since 1 - sinB > 0), as desired.
Chapter 15
Com bina torics
15-1 Choosing k objects from a group of n without choosing some particular element A amounts
to choosing the k objects from the n — 1 non-A ones. Since this is choosing k objects from a group of
'n - 1\
n — 1, it can be done in ways.
15-2 Let's try n = 9, k = 4 for a tough one. Evaluating, Q) = (9 ■ 8 - 7 • 6)/(4 -3-2-1) = 126. Also,
(|) = (8 - 7 - 6)/(3 • 2 • 1) = 56 and (®) = (8 • 7 • 6 • 5)/(4 • 3 • 2 • 1) = 70. Since 56 + 70 = 126, the identity
works in this case.
15-3 When k = 0, we have (q) = ("q1) + Since (”_11) = 0, we have 1 = 1, so Pascal's
identity is still satisfied.
15-4 The counting argument is easy. To choose k things from a group of n is the same as to
choose the n - k elements you're not taking. The algebraic argument isn't much worse: we have
(fy = n\/k\(n - k)\, while = n\/(n - k)\(n - (n - k))\ = n\/(n - k)\k\. The two are clearly the same.
15-5 For n = 8 and k = 4 we have
= 70 + 35 + 15 + 5 + 1 = 126 =
as desired. If k = n - 1 the identity says that
This is obvious because the left hand side is n, while the right hand side is n l's.
15-6 For n = 8 and k = 4 we have
as desired. If k = 1 we have
< 93 >
94 > CHAPTER 15. COMBINATORICS
which is obvious because the right hand side is n and the left is n l's.
15-7 The first row is just 1; the second row is 11. We extend the table further not by evaluating
(j*) 's, but by simple addition:
1
1 1
1 2 1
1 3 3 1
1 4 6 4 1
1 5 10 10 5 1
1 6 15 20 15 6 1
1 7 21 35 35 21 7 1
1 8 28 56 70 56 28 8 1
15-8 There are several equally good explanations. One is Pascal's identity, which forces (^) =
/ clearly (^) and are symmetrically placed entries in the triangle.
15-9 The sums are 1,2,4, and 8; clearly we have the powers of 2 here. This happens because of
the identity (jj) + (") +-f (”) = 2n, since summing across a row of Pascal's triangle is just such a
sum.
15-10 We do the induction on n. The base case of the induction is clear: certainly we can walk
to ({) = 1 and (J) = 1 in 1 way each. For the inductive step, assume we can walk to in
steps for any k. To get to (”) for some j, we must pass through (*11) or ("y1), because we only walk
downward and to the left or right. Thus the number of ways we can walk to ("y1) is the sum of
the numbers of ways we can walk to and (ny1)- But these are and (n~l) by the inductive
assumption! Thus we can walk to (") in ("y1) + (”“*) = ('•) ways. (We used Pascal's identity in the
last equation.)
15-11 This is virtually the same argument as in the text. x

Here we associate the o's with the possible last left branchings, x
and we are done by the same argument—since there must be one
and only one last left branching and it must occur at one of the
o's, the sum of the numbers of ways to get to the o's is equal to
the number of ways to get to the •. x x x o
15-12 The fact that the sum of the bottoms of the combina- x x x • x x
tions is always r suggests the use of a committee selection model. In fact, each term can be seen as
the way to select an r-member committee, taking r - k from a group of n candidates and k from a
group of m candidates. Since we can take anywhere from 0 to r members from either group, this
is just the number of ways to choose an r-member committee from a group of ft + m candidates, or
(n+w) yê jiave hence proven that
as asked. (Try this with some small numbers, say n = 4, m = 5, and r = 3.)
Letting m = n = r, Vandermonde's identity is
which by („:t) = (J) is
15-13 Since (x + y)2 = x2 + 2xy + y2, (x + y)4 = (x2 + 2xy + y2)(x2 + 2xy + y2) = x4 + 2x3y + x2y2 +
2x3y + 4x2y2 + 2xy3 + x2y2 + 2xy3 + y4 - x4 + 4x3y + 6x2y2 + 4xy3 + y4, which agrees with the Binomial
Theorem, since (4) = (4) = 1, (4) = (^) = 4, and g) = 6.
4
4-k
15-14 In form, the expansion is ^
k=0
15-15 My logic is correct. The text gives the term as (£) xky4 k, and I gave the term as (4^) xky4 k,
but the two are the same since (£) = (4M.
15-16 For n = 1 we have (x + y)1 = (J)x + Q)y = x + y, which is the correct expansion. For
n = 2 we have (x + y)2 - (q)*2 + (2)xy + (^y2 = x2 + 2xy + y2, which is correct. For n - 3 we have
(x + y)3 = (o)*3 + (3)x2y + (3)xy2 + (^y3 = x3 + 3x2y + 3xy2 + y3, which is—again!—correct.
15-17 We can use L” J = to write the earlier expansion as
(x + y)n =
This does not correspond exactly (but is still equivalent) to expanded version in the text, as a
term-by-term expansion will show.
15-18 For the first one, we have
x4 + l*3(2 y) + (^x2(2y)2 + f^x(2y)3 + (2 y)4
x4 + 8xJy + 24x2y2 + 32xyJ + 16y4.
For the second one, the term in question is (4)x423 = (35)(8)x4 = 280x4, so the coefficient is 280.
For the third one, the generic term will be (fy(x2)k(l/x)6~k = (6k)x3k~6, so the x will vanish when
k - 2. The coefficient will be (2) = 15.
To find the sum of the coefficients, we just let a = b = 1. (Do you see why?) Thus the sum of the
coefficients in this case is (1 + l)10 = 210 = 1024.
15-19 Letting x = 1 and y = -1 in the Binomial Theorem yields
which can be rearranged into
If n is odd, the identity becomes
Applying (£) = to the left hand side makes it into the right hand side, so we're done right
away.
15-20 Let x- 2 and y - 1 in the Binomial Theorem to get
<2+1 r=t(j)^n-k=t(iy-
The last sum is the one we're trying to evaluate, so it equals (2 + l)n = 3n.
15-21 The coefficient of x4yz3w2 is the same as the number of ways in which the "word"
xxxxyzzzvovo can be arranged. This is 10!/4!1!3!2!.
15-22 The coefficient of abc2 is the same as the number of ways to rearrange abcc, which is
4!/l!l!2! = 4 • 3 • 2/2 = 12.
15-23 Let the exponents of x, y, and z be j, k, and n- j -k respectively. We want each of j and k
to range from 0 to n, while still having n - j - k be nonnegative. Thus k can only range up to n - j,
rather than all the way up to n. We thus write
n n-j
n\
EE jl k\ (n - j - k)\
;=0 k=0
'yk2
where we use the coefficient according to the exponents chosen. Can you see that every possible
combination of exponents is attained by this sum?
235. This is exactly the form of the identity
which we proved in the text. Thus the sum is 100 • 299.
236. We rewrite the given sum as then use, say, block walking. The sum
describes all paths with n-k steps left in the first n steps down, and k steps left in the next m steps
down. Thus the sum describes all possible ways to take a total of n steps left in n + m steps down,
so it equals (n+nm).
237. Evaluate the inner sum first. We know that
so that
10 i 10
(k) = -1) = (2 + 4 + 8h-+ 210) - (1 + 1 + • • • + 1) = 2046 - 10 = 2036.
238. According to the Binomial Theorem, the general term of the expansion looks like
16
l(xy)k(-2y-3)16-k.
. k ,
Simplifying this, we get

16s
(-2 )16-Wfc-48
We see that there is no power of y when k = 12; the coefficient is then 24 ) - 16 (146).
n fk\ m~l (kN

239. This is really just sum manipulation. We write the given sum as y j - y ( We
k=0 Jt=0
'Q + lN
proved in the text (using Pascal's identity over and over) that y for any Q; applying
”1) r+1
k=0
this to the two parts of the given sum yields as desired.
240. Let n - 45 (surprisingly, problems often look simpler with the numbers taken out), and
write the given sum asy(«-/c-l)wj, which can be split into

k-0
±(n-M;)-±(:
k=0 k=0
n
The second sum is equal to 2". The first is, by the symmetry identity, the same as y (n - k)
k=0
(n\ i
letting j - n - k, this becomes ) . I, which we showed in the text is equal to n2n . Thus the
/=o W/
sum we want is n2n_1 -2n = n2n~1 - 2 • 2n_1 - (n- 2)2n_1. Substituting n = 45, our sum is 43 • 244. (If
you find all the hard to follow, try writing the sums out. But if you do this, try to see how your
manipulations correspond to manipulations of the X/s—in the long run you'll find XI to be a much
more efficient notation.)
241. In this problem, we prove an identity from the text in a new way. Can you see which
identity?
For the first part, once we include the first block in our group of n, we are left to pick n - 1 other
blocks. We have to pick these blocks from the remaining n + k. By elementary counting, we can pick
n - 1 blocks out of n + k in ways.
The second part is similar. Once we pick the rath block, we need to pick n — 1 more blocks. Since
we can't choose any of the first ra, we have n + k + l-m blocks to choose from, so there are (”+^t.11 m)
ways to make the choice.
For the third part, we use the first two parts. Consider the ways in which we can choose n blocks
from a set of n + k + 1. In particular, break down the possible choices in terms of the position of the
first block in the group. If the first block in our chosen group is block 1, part ii with ra = 1 tells us
that there are choices. If the first block is block 2, part ii with m = 2 tells us that there are 1)
choices. Continuing in this way, the total number of ways to choose a group of n from the total of
n + k + 1 blocks is
fn + k\ n + k- 1\ n + k- 2\
+ + +... +
U-v n~ 1 ) n- 1 )
But of course, we can also write the number of ways to pick n objects from the group of n + k + 1 as
(n+«+1) • Thus we have proven that
n + k\ n + k-l\ fn + k - 2\ fn -1
+ ... +
n-y
+
n~ 1 )
+
V n-1 J \n - 1 (■+')
Applying the identity (£) = (n”k) to all the terms in this identity yields the desired identity; make
sure you see how. [Note the method of proof: you can often easily prove combinatorial identities by
counting the same thing in two ways.]
242. Each term is a coefficient times some numbers of a's, b's, c's, and d's. To find each coefficient
we simply leta = fr = c = d = l.In doing so, our expansion reduces to the sum of coefficients, which
is then (1 + 1 + 1 + l)10 = 410.
243. This time let's expand the sums to see what's going on. The left hand side of the identity is
(n\ fn + l\ fn + 2\ fn + m-T
UM 2 M 3 r-H » ,
which by a standard identity proved in the text is equal to
fn + mf
-1.
m
Similarly, the right hand side is
m\ (m +1\ fm + 2\ fm + n- l\ fm + n
-l.
ir 2 + 3 +-+ « = » .
„ n + m\ . . i m + n\
But I ^ I always equals I I, so we re done!
244. The trick is to realize that 1992 = (20 - l)92, which by the binomial theorem is
+ • • • +
Finding the last three digits of this sum is equivalent to considering the sum (mod 1000). In (mod
1000), all but the last three terms vanish, since other terms contain 20 to some power 3 or higher.
and are thus divisible by 1000. The third to last term is (922)400 = 92 • 91 • 400/2 = 1674400, which
is congruent to 400 (mod 1000). The second to last term is -(92)20 = 92 • 20 = -1840, which is
congruent to -840. The last term is 1. The last three digits will thus be 400 - 840 + 1 = -439 = 561
(mod 100), and their sum will be 12.
245. The given sum is equivalent to ^(-l)fc+1 ( U j. (Write the sums out with • • • if you can't see
k=1
why.) However, we proved in an exercise that ^(-1)^ f H j =0. We can get the given sum to look
k=0
more like this by pulling out a minus sign, so the given sum is
= -(-1) = l
as desired. (At the risk of being repetitive, we again urge you to write out the sums if this isn't 100%
clear.)
246. We first evaluate the sum for n even. We then wish to evaluate
s - (?)+ 3(3) + ••• + («-1 )(n"i)
We can use the identity (£) = to write this as
S = (n - 1) (") + (n - 3) (3) + • • • + 1 (").
We can then add the two representations of S to get
2S = «[(?) +(3)+ - + %)]•

In an exercise of the text, we showed that
G)+ ©+••• +(A) “2-1.

so we have 2S = n2”_1, or S = nln~2. Note that since we have earlier shown also that
(»)+2(»)+... + „(») =„2->,

(1) + 3(3) +5(5) +••• + («- l)(„”i) = nln~2 also proves that for even n,
2©+4(3+ - + »©= "2^-

If n is odd, the sum is
S=G)+3® + - + »Q-
Do you see why this is different from the even case? Using (j) = (nn_k) straightaway accomplishes
nothing. (Confirm this.) We pull out Pascal's identity instead, to write
S = (V) + (V) + 3(V)+3(V) + •■•+(»- 1)G:D + (»- D(a +«(”:!)-

Since n - 1 is even, we have already proven that
("I1) + 3 (V) + ••• + ("-!) £1) = (n - !)2"-3
and
2(V) + 4(V) +••• + (»- l)t!) = (» - 1)2"-3-
But S is the sum of these two sums and the leftover terms
Co1) + (V) + (V) + • • • + G:l) =2”'2-

Hence for odd n, S = (ft - l)2n~3 + (ft - 1)2M~3 + 2n~2 - n2n~2, just as in the even case.
247. This one requires some serious cleverness, namely in writing the given sum as Y%=1 (n-ik) (k-1) •
Once we do this, we can find the sum in many ways. One is committee selection. We have ft cats
and n dogs and we choose n-k cats and k- 1 dogs to get a term (M”fc) (fc”a). Adding these up over
all k, the sum is the total number of ways to select n - 1 animals from the total of 2n (ft cats plus ft
dogs). Thus the given sum is equal to (,)2n|).
248. The given quantity is
(2n + k)l(2n - k)\
ft!2 (ft + k)\ (ft - k)\'
We can write this expression as
((2ft + k)(2n + k - 1) • • • (ft + k + 1)) ((2n - k)(2n - k - 1) • • • (ft - k + 1))
^) ((2ft + k)(2n - kj) ((2ft + k- 1)(2n - k - 1)) • • • ((ft +k + 1)(« - k + 1)).
Applying the AM-GM inequality to each term in parentheses, we find that each term is maximized
for k - 0. Thus the maximum value is
249. This is perfectly suited for a committee selection-type model. Suppose we have n delegates
at a conference. Each term (£) (”~^) is the number of ways for Yalli to first choose k "best delegate"
winners from the n total delegates, then choose m-k "honorable mentions" from the n-k non-best
delegates. Observe that no matter what k is, Yalli always chooses m people for awards of some type.
Thus the sum over k is the number of ways to have some number of best delegates and some number
of honorable mentions, such that the total number of awards given is m.
Now we count them up another way. Let Yalli first choose the m people who will get any award
at all, then choose some subset of those award winners to be best delegates. She can choose the m
award winners from the n delegates in (") ways, then choose the subset of best delegates in 2m ways
(since each of the winners have 2 possible awards). Hence the given sum is equal to 2'" .
One note: our use of "best delegates" and "honorable mentions" is not entirely whimsical. It
shows that a model close to your heart will often make things clearer.
250. To get rid of the distracting numbers, let's let n-22,m = 15, and r = 10, so that our sum is
which by Vandermonde's identity (proven in an exercise) is (n+rm). Substituting numbers from the
present problem, we can roll the sum up into .
Chapter 16
Sequences and Series
16-1 The first is an arithmetic series with first term 2, last term In, and n terms, so the sum is
| (2n + 2) = nz + n. The second is an arithmetic series with first term 1, last term In - 1, and n terms,
so the sum is |[(2n - 1) + 1] = n2.
16-2 We have
,(i + l)(i)(i-l)(i-2) , ,(i + l)(i)(*-l) , (i + l)(0

6-24-+6-6-+
(f + l)(Q[(f - l)(z - 2) + 4(f - 1) + 2]
4
(z + l)(z)(z2 + i)
as desired. Note that this proves the unlikely-seeming result that
l3 + 23 + 33 • • • + n3 = (1 + 2 + 3 + • • • + n)2.
16-3 After 8 and 13 comes 8 + 13 = 21, then 13-1-21 = 34, then 21 + 34 = 55, then 89, 144, 233,
377, 610,987,...
16-4 By the quadratic formula, the two roots of the equation are —^—. If we took the - from
1 - V5
the ±, we would have —-—, which is negative and thus can't be the limiting ratio between two
Fibonacci numbers. Thus we take the + root.
< 102 >

16-5 For the first one, we just write
Fn Fn-2 + Fn-i
F n-2 + Fn-3 + Fn-2
Fn-l + Fn-3 + Fn-3 + F„_4
Fn-2 + Fn-3 + F„_4 + F„_5 + Fn-4
- Fn-2 + Fn_3 H-+ Fi + Fo + Fi

= F„_2+F„_3H-+F1+F0 + I.
For the second, we proceed by induction. As the base case, for n = 0 we have Fq = 0 = FoF\. For
the inductive step, assume it works for n - 1, so that Fq + F2 + ■ ■ ■ + F2n_x - F„_iF„. We then have
Fq + F^ +-f F\ - F„_!F„ + F2n — F„(F„_ 1 + F„) = FnFn+i, so the identity holds for n.
The third is like the first: F2n+i = F2n +F2n-i = F2n+F2n-2+F2n-3 = F2n+F2n-2 +F2n-4+F2n-5 + ■■■ -
Fin + F2n-2 + • • • + F2 + Fq.
16-6 Probably the easiest way to see it is to consider very large n. Then one of the terms arn~l
and bsn~l will get very large compared to the other. The only way to compensate is to multiply both
sides by 0.
16-7 This is simple enough. For r, we have
6 + 2V5 1 + V5 6 + 2V5-2-2V5-4
r2 - r - 1
4 2 ~ 4
as desired. The calculation for s is almost exactly the same.
16-8 For n = 0 the sum is a + b = 2. For n - 1 the sum is ar + bs = r + s = 1. For n-2 the sum is
r2 + s2 = [(6 + 2 V5) + (6-2 V5)]/4 = 12/4 = 3, and for n = 3 the sum is [(16 + 8 V5) + (16-8 V5)]/8 = 4.
The first four terms of the sequence—2,1, 3, 4—do indeed satisfy the Fibonacci relation.
16-9 Binet's formula gives
0,
1 + V5 1 - V5
Fi = 1,
V5 2 2
1 '6 + 2 V5 6-2V5
1,
V5 4 4
1 16 + 8 V5 16-8V5
and F3 = 2,
V5 8 8
as desired.
16-10 We write Gn = arn + bsn, as we did for the Fibonacci sequence. In order that this sequence
satisfy the relation Gn+i = Gn + 2G„_i, we must have arn+l + bsn+1 = arn + bsn + 2arn~l + 2bsn~l, which
is rearranged into arn~l(r2 - r - 2) = -bsn~l(s2 - s - 2). Assuming r + s, this can only be satisfied if
104 > CHAPTER 16. SEQUENCES AND SERIES
both r and s satisfy x2 — x — 2 = 0, which by the quadratic formula has solutions r = 2 and s = -1.
We then use Go = 0 and G\ = 1 to write a + b = 0 and ar + bs = 2a - b = 1; solving the system gives
a = 1/3 and & = -1/3. The solution is thus
Gn = l2" “ §(~1)n'
16-11 The solution would be similar, except we would need to write Xn as the sum of three
geometric series: Xn = arn + bsn + ctn. In order for Xn to satisfy the recursion relation, we would need
r, s, and t to satisfy the cubic equation
x3 - ax2 -.fix - y - 0.
If we could solve this equation to find r, s, and t, we would need three initial values, say Aq, A\ and
A2, to find a, b, and c.
As the number of terms in the recursion grows, the degree of the polynomial which must be
solved increases as well; even solving a cubic, to get the solution to a three-term recursion, is not
easy.
n 2 1
16-12 A partial fraction decomposition on --—-— reveals that it equals ---— --—,
tr ir (n + l)(n + 2) M (n + 2) (n +1)
so
v a TL 1
^ (n + 1 )(n + 2) = E n-1
(n + 2) (n + 1)
2 _ 1 2 _ 1 2 1
3 2+4 3+5 4
1111
2 + 3 + 4 + 5+ ‘“
111
which diverges since the series l + q + y + T + ’“ diverges.
16-13 Having written our sum as the double sum given, we evaluate
1/3"
E
m-n
3™ 1 -(1/3)'
using the formula for the sum of an infinite geometric series. Putting this back in the other summa¬
tion, our overall sum is
2 " 3,1 = 2 3 1/3 1
3 “ 2 ' 3" 3 2 1 — (1/3) 2
16-14
(_43) = (-3)(-4)(-5)(-6)/(4)(3)(2)(l) = 15;
(T) = (-7/3)(-10/3)(-13/3)/(3)(2)(l) = -455/81.
16-15 We have = n(n - 1 )(n - 2) • ■ ■ (n - k + l)/k\. If k > n, then n - /c + I < 0, so 0 will appear
somewhere in the product n(n - 1 )(n - 2) • • • (n - k + 1). Thus the product is 0, so (j) = 0 if k > n.
16-16 (-1) = (—1)(—2)(—3) • • • (-n)/(n)(n -!)■■• (2)(1) = (-1)".

16-17 We write
1/(2.12)2 = (2 + .12)"2 = 2~2 + ^~2^2-3(.12)+ j 2“4(.12)2 + • • •
= .25 + (-2)(.125)(.12) + (3)(.0625)(.12)2 • • •

W .25 - .03 + .0027 = .2227.
16-18 To first order, 1/101 = (100 + l)-1 « (100 - 1)/1002 = 99/10000 - .0099. Using a calculator
we get .009901—the first-order approximation is very good!
16-19 A second order approximation would take the first three terms into account, so would be
(A + e)-1 * A-1 - A~2e + A"V = ^ ~ ^ + ^.

A3
A “zeroth" order approximation would take only the first term; that is, (A + e)-1 « 1/A. Not too
interesting, but a decent approximation if e is small enough compared to A.
16-20 To first order,
Va2 + e = (A2 + e)1/z « (Az)1/z + (A2)~ll2e = A +
To evaluate V!7, we let A = 4 and e = 1 in the formula above to get 4 + 1/8 = 4.125 as the
approximation. Given that the actual value is 4.123..., this approximation isn't bad at all.
16-21 The reciprocals of the first sequence are \,l, |, 2,..., which is arithmetic. After getting
a common denominator, the reciprocals of the second are y, y, |, |, which is again arithmetic. The
reciprocals of the third are 2, 4, 8, 16, ..., which is not an arithmetic sequence. Thus the first and
second are harmonic sequences.
251. The sum of the first 3n positive integers is (3n)(3n + l)/2; the sum of the first n is (n)(n + l)/2.
We thus have (3n)(3n + 1) = n(n + 1) + 300, or 8n2 + 2n - 300 = 0. Dividing by the common factor of
2, we get 4n2 + n - 150 = 0, which factors as (4n + 25)(w - 6) = 0. Discarding the negative solution,
we find n - 6. The sum of the first 4n positive integers is hence (24)(25)/2 = 300.
252. The reciprocals of the given sequence are y ..., forming an arithmetic sequence with
first term \ and common difference The eighth term in the arithmetic sequence is|+7(^) = | = |;
the eighth term in the harmonic sequence is the reciprocal of this, or 3/4.
„ , . . , , (10)(11)(21) (10)(11) 1A
253. Splitting it up, the sum becomes + 2_^k + 2_^l = ---1---10 =
k=l k=l k=l 6 2
385 + 55 + 10 = 450.
254. This problem shows that products can telescope as well as sums. We have
In the right-hand expression, the numerator of each fraction is equal to the denominator of the
next one; thus the two can cancel. We can cancel all the numerators except the last one and all the
denominators except the first one, leaving 8/2 = 4.
255. We write a\ - = 1, so that a3 = (a\ - l)/«i- Furthermore, from al - aâi = -1, we get
«2 = + 1)Md = 10- Thus #3 = (100 - l)/3 = 33.
256. For any k, we can write
1 2 k- 1 1 + 2 +-h (*- 1) k(k - l)/2 fc-1

—1-(-•••-(-=---- = —- ■-, w
k k k k k 2
How does this apply? In our sum, we group the fractions like
1\ fl 2\ / 1 2 29
2 J + V3 + 3) + ' " + V30 + 30 + " ’ + 30
and each quantity in parentheses is a sum of the type (*). Thus the sum becomes
1 2 3 29 1 + 2 + • • • + 29 (29)(30)/2 435
— -f- — -f- — + • • • + - — - — - — -.
2 2 2 2 2 2 2
1/2
257. Each fraction is of the form —-—which can be written using partial fractions as
n
1/2
--. Thus the sum is
n+2
1/1_1\ 1/1_1\ 1 / 1 _ 1 \ _ 1 n _ J_\ _ 9_

2 \2 _ 4/ + 2 \4 _ 6/ + 2 \6 _ 8/ + ' ' + 2 \18 _ 20/ “ 2 V2_20y “ 40'
258. Making the exponents all have base 2, we get
21/322/923/2724/81 . . . _ 21/3+2/9+-
Thus the problem comes down to evaluating the sum S = (1/3) + (2/9) + (3/27) + • • • To do this, we
note that
S/3 = (1/9) + (2/27) + (3/81) + • • • = S - [(1/3) + (1/9) + (1/27) + • • • ]
The infinite geometric series (1/3) + (1/9) + (1/27) + • • • is easily summed as 1/2 (see Volume 1 if you
don't know how to do this), so we have S/3 = S - 1/2, or S = 3/4. Hence the product we were asked
to evaluate is 2s = 23/4 = ^8.
259. We just evaluate the terms in succession: a3 = 2^2 - 3a3 = -8, = 2a3 - 3«2 = -13,
#5 = 2a4 — 3#3 — —2.
260. Rationalizing the denominator of each term, we have
111 1
V2+ Vl V3+ V2 V4+ V3 V25+V24
V2-VI V3-V2 VI-V3 V25-V24
1 + 1 + i +"'+ 1 '
which immediately telescopes into V25 - Vl = 5 - 1 = 4.

261. Since the Cesaro sum of (a\, a2r..., a99) is 1000, the sum Si + S2 + ■ • • + S99 is 99(1000) = 99000.
Thus the Cesaro sum of (1, a\, a2/..., a99) is
1 + (1 + Si) + (1 + S2) + • • • + (1 + S99) 100 + 99000

100 ■ loo ■991
262. This is easily proven by induction. As the base cases, V\ = 2 = 2Fo and v2 - 4 = 2F2.
For the inductive step, assume that the relation v= 2F2ic_2 holds for all A: < n. For n we use
Fn - F„_ 1 = F„_2 repeatedly to write
vn - 3u„_i - vn-2 = 6F2H-4 - 2F2n-(, - 4F2n-4 + 7F2n-5 - 2F2n-3 + 2F2n-4 = 2F2n-2/
as desired. Our induction is complete.

263. We are asked to evaluate
„ 1 4 9 16 25
S “ 2 + 4 + 8 + 16 + 32 +
We write
1 4 9 16 25
S/2'4 + 8 + 16 + 32 + 64 +
and subtract this from S to find
S-S/2 = S/2 = H + f + l + !> +
To sum this new series for S/2, we must play the game again, writing
... 1 3 5 7 9
S/4~4 + 8 + l6 + 32 + M +
and subtracting:
1 2 2 2 2
S/2-S/4 = S/4_- + 5 + § + - + -
Since the sum of the geometric series 1/4 + 1/8 + 1/16 H-is 1/2, we have S/4 - 1/2 + 2(1/2) - 3/2,
so S = 4(3/2) = 6.
1 1
264. A partial fraction decomposition changes our sum which tele-
4n - 3 4n + 1
scopes straightforwardly:
+
265. We can expand the double sum as
OO OO ^
r l 1
+ + +
£ 0 +1)3 (;' + 2)3 (; + 3)3
;=i
ill 1 1 1
- 1 23 + 33 + 43 + + 1 33+43+53 +
1 1 1
+
43 + l, + + + 53 53
12 3 4
— - - —1— - 4- - •••
23 33 43 53
Ajt-i
“ t-3
k=1 K
A M _ 1
=
OO * OO
=
fc=l K k=1 K
= p-q-
266. The product is (|) (|) (|) • • • (^), which telescopes into 2/m.
267. We have
1 + 2z + 3z2 4-h (n + 1 )in = [1 + 2z - 3 - 4z] + [5 + 6z - 7 - 8z] +-1- (n + 1),
where there are n/4 of the expressions in brackets and an isolated n + 1. Each bracketed expression
adds to -2z - 2, so the sum is (rz)(-2z - 2)/4 + (zi + 1) = (« + 2 - ni)/2.
268. Consider instead the sequence g(n) = (-1 )nf(n). The first thing we note is that g(n) satisfies
the Fibonacci relation. Thus if we can find a and b such that g(0) = apQ + &F2 and g(l) = aF\ + bF3,
we will have g(2) = g(l) + g(0) = a(F0 + Fi) + b(F2 + F3) = aF2 + bF4, g(3) = g{2) + g(l) = a(Fi +
F2) + KF3 + F4) = + bFs, and so on: thus, g(n) = aFn + bFn+2 for all n. Of course, this is
dreaming until we can find such a and b. We need g(0) = 3 = Ofl + lb and g(l) = 1 = a + 2b, so
we immediately find b - 3 and a - -5; that is g(n) = -5Fn + 3Fn+2 for all n. For n = 100, we have
/(100) = g(100) = -5Fioo + 3Fio2 = -2Fioo + 3Fioi = F100 + 3F99. This could also be written F100 + 2F98,
F99 + 3F98, or in many other ways.
Do you see how this problem shows that any sequence An satisfying the Fibonacci relation can
be written as aFn+2 + bFn for some integers a and b? An interesting result.
269. Use 0 to represent odd numbers and e for evens, and consider only the first four entries in
each row. The first row is 0, the second is 000, the third is oeoe (first four entries only), the fourth is
ooeo. Each time we hit ooeo, the next row will be oeee (figure out why). Each time we hit oeee, the next
row is oooe; after oooe comes oeoe. This is what we started with in the third row, so the rows will cycle
through this pattern; oeoe —> ooeo —» oeee —> oooe —> oeoe. Thus every row after the third will have at
least one even entry.
270. While you could certainly do this one algebraically, here's a nicer solution. Draw the grid
1
1 2
1 2 3
1 2 3 ••• n
Now the sum across the kth row of the grid is k(k + l)/2, so the sum of all the numbers in the grid is
m 2g) + • • • +
n(n + 1)
(*)
2 2 2
On the other hand, the sum down the kth. column of the grid is (n - k + 1 )k, so the sum of all the
numbers in the grid is
l(n) + 2(n - 1) + 3(h - 2) + • • • + h(1). (**)
Comparing (*) and (**), we are immediately done.

a2 £2
271. Note that anbn = ^ ^ = an-\bn-\. This means that the product anbn is constant for all n, so
is equal to the value at n = 0: aobo-6. Substituting an-\ - 6/bn-\ into the recursion for bn, we have
bn — We thus have
fc8 = ^/6 = (fc3/6)3/6 = ... = ^J_.
Evaluating the power of 6 in the denominator as a geometric series, 37 + 36 H-1-3° = (1 — 3S)/(1 — 3) =

6560/2 = 3280, this last fraction becomes 36561/63280 = V,1S112?,lm.
272. The fourth term will have k = 3, so will be
li/3-3('_2x)3 = (s) ( 3) ( 3) -8r>) = -Â3.

3 J (3)(2)(1) ' 81
The coefficient is -40/81.

273. We have A1(un) = un+\ - un = (n + l)3 + (n + 1) - n3 - n - 3n2 + 3h + 2, A2(m„) = A1 (A1(un)) =
A1(3h2+3h+2) = 3(h + 1)2+3(h+1)+2-3h2-3h-2 = 6h+6, A3(h„) = A1(6n+6) = 6(h + 1)+6-6h-6 = 6,
and Af(u„) - A1 (6) = 6 - 6 = 0. The answer is 4.
274. We multiply the given equation by (a + b) to get
{a + b)Rn = ^(an + bn)(a + b) = i(an+1 + bn+l) + ^ (ban + abn) = Rn+1 + abRn_i.
Since a + b = 6 and ab = 1, this becomes the recursion 6Rn = Rn+1 + Rn-i, or Rn+1 = 6Rn - Rn-\. The
initial conditions for the recursion are easily found: Rq = 1 and R\ = (a + b)/2 = 3.
Finding the units digit of a term is the same as finding the term (mod 10). Thus, let's calculate
terms (mod 10): starting with Ro and going up, we have 1, 3, 6(3) - 1 = 17 = 7, 6(7) - 3 = 39 = 9,
47 = 7, 33 = 3, 1, 3, 7, 9, 7, 3, 1, 3, 7, 9, and so on. The recursion repeats in groups of 6! Thus
jRi2345 = 12345 (mod 6) = ^3 = 9 (mod 10), so 9 is the units digit we seek.
275. We have
0ioo - 099 + 2(99)

= fl98 + 2(98) + 2(99)
= #i+2(1+ 2+ 3 +••• + 99)

= 2 + 2[(99)(100)/2]
= 9902.
276. This is actually a trick question: if the polynomial is equal to un for all ft, then the sum of its
coefficients can be found by substituting 1 for n in the polynomial. Hence the sum of the coefficients
is equal to u\, or 5. We don't need the recursion information at all.
On the other hand, we can find the actual polynomial fairly easily. We write
un - un-1 - 3 + 4(n - 2)
un-1 - un—2 = 3 + 4(n - 3)
U„-2 - un-3 = 3 + 4(n - 4)
U.2 — U\ = 3 + 4(0).
Adding all these equations together, we get
un — U\ — 3(n - 1) + 4(1 + 2 + 3 +-1- (n - 2)) = 3(ft - 1) + 2(n - 2)(n - 1) = 2n2 — 3n + 1,
so that u„ = 2n2 - 3ft + 1 + Mi = 2ft2 - 3ft + 6.

277. We use a formula from complex numbers, writing
elx - e~lx
smx -
2i
We then have
Esm(fti)
OO . , V
_
oo
^ gmx — g-mx y
3n ^ ~2i 3"
n=0 n=0
OO / ix \ n ,-ix
- 2 i ^13
n=0 v
Since both sums are now infinite geometric series, this is just
1
2i . (l - (eix/3)) (l - (e~ix/3))_ ‘
Since sinx = 1/3, cosx = 2V2/3, so = cos x + z sinx = (2 V2 + z)/3 and = (2 V2 - z)/3.
Substituting in and rationalizing the denominators, we have
I2j (\_ l_1_1

(9 - 2 V2 )-i (9-2 V2) + i J
[(9 - 2 V2) + i] - [(9 - 2 V2) - i\'
90 - 36 V2
10 - 4 ^
10 + 4 V2
68
5 + 2V2
34
Whew.
278. First of all we will prove by induction that an - a\ + n - 1 for every n > 1. As the base
cases, we have 02 - + 1/ = 02 + 1 = 01 + 3; since 01 4-1 = 02 < «3 < = 01 + 3 and all 0/ are
integers, 03 = 01 + 2. For the inductive step, suppose that = a\ + (k - 1). Since 02jt = % + k, we
have fl2)t - + (2/c - 1). But 01 + (fc - 1) = 0jt < %+i < ak+2 < • < dik-i < a2k - ai + (2/c - 1), so
fljt+i = ak + 1 = fli + k. This completes the induction, so an = a\ + (n - 1) for all n > 1.
Thus all we need is to find a\. We will show that a\ = 1 by contradiction. Assume that 01 > 1.
Then the numbers (a\ + 1)! + 2, (a\ + 1)! + 3,..., (01 + 1)! + a\ + 1 are composite. (Why?) Let p be the
smallest prime number such that p > (a\ + 1)! + 01 +1 and let n = p - 01 + 1. Then p = 0i+n-l = 0M/
so that n is prime. On the other hand, (01 + 1)! + 2 < p - 01 + 1 < p - 1, so that p - a\ + 1 is composite,
a contradiction. Since 01 > 1 yields a contradiction, we must have a\ = 1. Since an = 01 + rc - 1, we
have an = n for every n. In particular, 01993 = 1993.
279. Let such a number be N - *1*2 •••Xfc, where each digit x,- is an element of {1,3,4} and
Xi + X2 H-\-Xk-n. It is fairly clear that an - an-\ + 0„_3 + an-4. (Do you see why?) We thus have
0-2 n — a2n-\ + #2n-3 + fl2«-4 = <?2n-2 + «2w-4 + (^2n-5 + a2n-f) + «2n-4
= fl2n-2 + «2n-4 + («2n-2 ~ &2n-(f) + «2n-4 = ^2(n-V) + 2«2(n-2) _ «2(n-3)-
We define a new number = 02«- Then = 2fr„_i + 2b„_2 - bn_3. Clearly b\ - 1 = l2, &2 = 4 = 22,
= 9 = 32, and so on. Now we must prove that the are all squares.
Define a third sequence c„ by ci = 1, c2 = 2, and c„+2 = cn+\ + c„ for all n. We can show by
induction that b„ = c2 for all n. As base case, you can confirm for yourself that b\ - c\ and b2 = c2.
For the inductive step, assume that bk = c2 holds for all k < n. Then
bn = 2bn-i + 2bn-2 - bn-3 = 2c2_2 + 2c2_2 - c2_3
= (Cn-1 + C„-2)2 + (Cn-1 - C„-2)2 ~ ^-3 = + c«-3 “ c«-3 = cn-
This completes the induction, so that bn = 02« = c2 for every n.

280. To break down the number of pairings, consider 2(n + 1) objects a\, a\, ci2, 02/- • •, b-
To each of the un pairings of a\,..., an one can add (b, b). From each partition of the 01,..., an one can
form n new ones by replacing {air aj) with (aj, b), {aif b). Among the (n +1 )un partitions thus obtained,
some repetitions exist. If the original pairing contains two identical pairs (0f,0;), (0/,0/), i 4- j, then
splitting each of them gives the same pairing of 01, 01, a2f. ..,b. There are n(n - l)/2 ways to select
i and and for any such choice the remaining 2(n - 2) objects can be paired in «n_2 ways. We must
subtract this total number, (n(n - l)/2)un-2, of repetitions, leaving un+1 = (n + 1 )un - (n(n - l)/2)un-2,
as desired.
281. Letting n = k and n = k - 1 for some k> 2 in the second given equation, we find
«! + ••• + - (^ - l)2%-i
fll +-b fljt = /c2%.
Subtracting the first from the second, we have, after a bit of algebra, (k2 - l)fl^ = (k - l)2flfc-i- Since
fc>2, we have
_k-l (k- l)(k - 2) (fc-1)! , 1

Uk~k + Iak~l~ (k + l)(k) Uk~2 (/c + 1) • • • 3 1 (k + l)k'
Hence we get an = l/n(n + 1).

Chapter 17
Counting in the Twilight Zone
17-1 Are you convinced that our basic techniques will fail?
17-2 Let there be x setters, y spaniels, and z wolfhounds. The sequence representing this has
x d's before the first_, y d's before the next_, and z d's after the last_. There is exactly one such
sequence. Similarly, if we start from a sequence with x d's then a then y d's, then a then z d's,
we can make exactly one choice for the breeder. If there are no setters, the sequence starts with a_;
if no spaniels, the two_'s are adjacent; and if no wolfhounds, the sequence ends with a_. This is
perfectly fine.
17-3 As before, we write a sequence with n d's and r - 1_'s to represent each choice. (We
use r - 1_'s because we wish to partition the d's into r groups.) Again, counting the number of
possible sequences is equivalent to counting the number of choices. Since we make our sequences
by choosing r — 1 places to hold_'s out of n + r — 1 possible places in the sequence, our general
formula is ("Z;1).
17-4 If we let X\ be the number of the first breed, x^ be the number of the second breed, etc., the
number of solutions to Xi + *2 ^-h xr = n is the number of ways we can choose n dogs from among
r breeds. We've already solved this problem, so we know that the number of solutions to the given
equation in nonnegative integers is (n^1).
17-5 Suppose you're just in class A. Then you're only counted once in +#(A). If you're in A and
B, you're added in both +#(A) and +#(B) and subtracted in -#(A D B). If you're in all three classes,
you're included in everything, so you're added four times and subtracted three.
17-6 As an extesion of our three class discussion, you should find if we add a class D, the
number of students is
#(A) + #(B) + #(C) + #(D)

-#(A n B) - #(A n C) - #(A n D) - #(B n C) - #(B n D) - #(C n D)
+#(A nBnC) + #(AnBnD) + #(AnCnD) + #(BnCnD)
-#(A nBnCnD).
You should be able to construct the expression for adding class E now.
< 113 >

114 > CHAPTER 17. COUNTING IN THE TWILIGHT ZONE
17-7 From the Principle of Inclusion-Exclusion, to find the total number of objects which are in
at least one category (or #(A\ U A2 U A3 U • • ■ U An)), we add the number of objects in each category
then subtract the number of objects which are in both Aj and Aj for all (i,j), and so on. Since #(A)
is the same for all i, we need only determine it for one category then multiply by the number of
categories. Similarly since #(Ai D Aj) is the same for all pairs of categories (i, /), we evaluate it for one
pair of categories, then multiply by the total number of categories (2). Continuing in this manner,
we get the desired expression since there are (3) groups of 3 categories, (4) groups of 4 categories,
and so on.
17-8 Don't go on unless you're sure that you understand this fact.
17-9 The generating for e\ consists of only even powers; that of £2 of only odd powers; and that
of C3 only prime powers. Thus our generating function is
(1 + x2 + x4 + x6 + • • • )(x + x3 + x5 + x7 + • • • )(x2 + x3 + x5 + x7 + ■••)•
17-10 The generating function for each die is (x + x2 + x3 + x4 + x5 + x6); thus, the generating
function for rolling all the dice is (x + x2 + x3 + x4 + x5 + x6)10. We proceed as in the text to find the
coefficient of x25. We first factor out an x: x10(l + x + x2 + x3 + x4 + x5)10. We thus want the coefficient
of x15 in
10
/1 - X6
(1 + X + x2 + x3 + x4 + X5)10 = (1 - x6)10(l - x)-10.
V 1 -X
Since the exponents of x in (1 - x6)10 are multiples of 6, we have only three terms contributing to
our total,namely (q0)(x6)0] [-(“/g0)x15], [-(\°)x6\ [-(“J0)x9], and [(a2°)(x6)2] [-("3°)x3]. Hence, our
desired number of rolls which produce 25 is
17-11 One 2 is 2, two 2's is 4, etc. Continuing in this way, we find our generating function for
the number of 2's is 1 + x2 + x4 -1- x6 H-.
17-12 For the x term, we only have one term: x-1-1-1 = x. Forx2, wehavexM-l-l+l-x2-!-!. = 2x2;
for x3, we get x3-l-l-l+x-x2-l-l + l- l-x3-l = 3x3; and for x4 we find
x4 • 1 • 1 • 1 + x2 • x2 • 1 • 1 +1 • x4 • 1 • 1 + x • 1 • x3 • 1 + 1 • 1 • 1 • x4 = 5x4.
The only partition of 1 is 1; there are two partitions of 2 (1 + 1 and 2); there are 3 partitions of 3
(1 + 1 + 1, 1 + 2, and 3); and there are 5 partitions of 4 (1 + 1 + 1 + 1, 1 + 1 + 2, 2 + 2, 1 + 3, and
4). Notice how each expanded term of the generating function corresponds to one of the partitions.
(For example, x • 1 • x3 • 1 corresponds to the 1+3 partition of 4.)
17-13 Certainly we can. Our backwards counting approach is very flexible. Try inventing a
simple three dimensional problem (like walking around on a cube).
17-14 Let Mike's initial vertex be 'home' and the other three 'away'. Let be the number of
ways he can get back to home after k moves starting from home and 4 be the number of ways to get
home in k moves starting from an away vertex. Since from the home vertex, our next move takes us
to any of 3 away vertices, we have = 3ffc-i, or 3 times the number of ways he can get back home
in k — 1 moves. From an away vertex, Mike can go straight home or to 2 other away vertices, so we
find ffc = sjt-i + 2tk-\- Clearly si = 0 and t\ - 1, so we can repeatedly use our recursions to solve the
problem and find = 183.
17-15 For the odds we have the correspondence
1 2 3 4 5 6 7 8 9 10 11
1 3 5 7 9 11 13 15 17 19 21
and for the evens we have
1 2 3 4 5 6 7 8 9 10 11
2 4 6 8 10 12 14 16 18 20 22
These correspondences show that the set of positive even integers and that of positive odd integers
are both the same size as the set of positive integers.
17-16 For the geometric proof, draw A ABC with AC = b and BC = a. Let any line parallel to AB
intersect AC at X and BC at Y. These parallel lines give us our correspondence between points on
AC and BC. Analytically, take 0 < x < a; we correspond this x to a number from 0 to b by bx/a.
282. We started off with an easy one. We can split the five balls into three groups as in the problem
in \ (2) (2) ({) =15 ways because we can pick 2 for the first group, two of the remaining 3 for the
second group, and the last ball is for the last group. We then divide the product by 2 because the
two groups of 2 are indistinguishable. Now we can put these three groups into numbered boxes in
3! = 6 ways, for a total of 6(15) = 90 ways to put the balls in the boxes.
283. There are 100/5 = 20 multiples of 5 and L100/7J = 14 multiples of 7, for a total of 34; however,
we have counted the multiples of 35 twice. Since we want to exclude these altogether, we subtract
them twice for a total of 34 - 2 • 2 = 30.
284. After choosing the first square, we eliminate the other squares in that row and column.
The remaining squares form a 4 by 4 grid. After choosing from among these squares, we have a
3 by 3 grid from which to choose the final square. Since we have 52 = 25 ways to pick the first
square, 42 = 16 ways to pick the second, and 32 = 9 ways to select the last, we have a total of
(25)(16)(9)/6 = 600 ways to pick the squares, where we have divided by 6 because the 3 squares can
be selected in 3! = 6 orders.
285. We count the number of intersections by counting the maximum number of possible in¬
tersections of 100 lines, and then subtracting the number of intersections Tost' to the parallel and
concurrent lines. The maximum number of intersections occurs when each pair of lines intersects
at a point through which no other lines pass. The number of such intersections is the number of
ways to select the lines, or (12°) = 4950. For the 25 parallel lines, there are 0 intersections where
there could have been (225) = 300. For the concurrent lines, there is only one intersection where there
could have been (225) = 300. Hence, we have Tost' 600 - 1 = 599 intersections and are left with a
possible 4950 - 599 = 4351.
286. We translate the given problem to one involving solutions in nonnegative integers by making
the substitution y,- = x* - 1. Since the x, are positive integers, the y\ are nonnegative integers. Our
given equation becomes
3/1 + 3/2 + • • • + 2/8 = 11-
From the discussion in the chapter there are ) = 31824 solutions to this equation in nonnegative
integers.
287. We call the 'minimum element' of a set the element with the smallest subscript. First
consider the case where a\ is the minimum element. To form a desired set we can include any of the
remaining elements. Since each element can either be in or out, there are 211 = 2048 such sets. If 02
is the minimum element, only even-subscripted elements can be included. There are 5 of these, so
there are 25 = 32 desired sets with 02 as the minimum element. Similarly, there are 23 = 8 desired
sets with 03 as the minimum element, 22 = 4 with a4 as minimum element, 21 = 2 with 05, and 21 = 2
with «6. Each of the other 6 elements can form their own subset which satisfies the problem, so our
total is
2048 + 32 + 8 + 4 + 2 + 2 + 6 = 2102.
288. The four countries can be ordered around the table in (4 - 1)! = 6 ways. The Americans and
Russians can be ordered (in their little group) in 3! = 6 ways. The Germans can be ordered in 4! = 24
ways and the French in 2. The number of possible seating is thus (6)(6)(6)(24)(2) = 10368.
289. We solve this problem by counting the total number of possible seatings, then subtracting
the ones in which Ginger sits next to Gilligan, his ape, or both. The number of possible seatings is
simply 8!. To count the number of seatings in which Gilligan and Ginger are adjacent we consider
the two as a unit, for a total of 7! seatings. Since Gilligan and Ginger can sit together in 2 orders, the
total number of seatings in which they are adjacent is 2(7!). Similarly, Ginger can sit next to the ape
in 2(7!) ways. Once again, we've overcounted by counting the cases of Ginger being between the
two undesirables twice. Ginger can be seated between the two in the row in 2(6!) ways (consider the
three as a unit and note that there are 2 ways to make the unit), so the number of admissible seatings
is
8! - 2(7!) - 2(7!) + 2(6!) = 30(6!) = 21600.
290. The question is pretty tough as stands, so let's try to draw a one-to-one correspondence
to a situation which is easily counted. Let x be a taken seat and 0 be an empty seat. Thus,
oxoooxoooxooxoxooooo is a possible seating sequence. There aren't (25°) permissible sequences because
there must be one 0 between each pair of x's. However, if we remove an 0, or a seat, from between
each pair of neighboring x's (forming oxooxooxoxxooooo from the sequence above), we get a sequence
with no restrictions. Similarly, we can go from a sequence of 11 o's and 5 x's with no restrictions to
a sequence as in the problem by adding an 0 between each pair of neighboring x's. In this way we
draw a one-to-one correspondence between the seating in the problem and the seating of 5 people
in 20 - 4 = 16 seats without restriction. (Make sure you see that this is a one-to-one correspondence;
try to prove it.) Thus, there are (g6) ways to select the seats of the participants, and 5! = 120 to put
the people in the seats, for a total of (120)(4368) = 524160 seatings.
291. Guided by our counting on graphs discussion in the text, we work 1 2 3 4 5

backwards. We can get to vertex 30 from 29 in only 1 way. Similarly, there's 671 616 511 365 190
10 9 8 7 6
only one way to go from 25, 26, 27, or 28. From point 24 we can go to 25 or
55 105 146 175 190
to 27. Each of these points have one path to 30, so there are 2 paths from 24. 11 12 13 14 15
Similarly, if we start from 23, we have 2 paths via point 24 and 1 path via 28 55 50 41 29 15
(directly to 28). In this manner, we continue backwards finding the number 20 19 18 17 16
5 9 12 14 15
of paths from each point on the grid. These numbers are shown on the grid 21 22 23 24 25
beneath the numbered points. We see that the number of paths from point 1 5 4 3 2 i
is 671. 30 29 28 27 26
i i i I 1
292. We can just count these in the conventional casework way, but there
is a significantly slicker way. Let A be the original set of points with integral c nponents from 1 to 4,
and B the set of points with integral components from 0 to 5. Clearly any line through four points of
A also goes through two points of B. Now consider the points in B but not in A. Each of these points
are on exactly one of the lines through four points in A. For example, the corners are on diagonals of
the cube A, the points on the edges are on diagonals of square cross-sections, and the points on the
interior of the faces are on the lines parallel to the coordinate axes through four points of A. Hence,
we can draw a one-to-one correspondence between pairs of points in B - A (meaning the points in
B but not in A) and lines through four members of A. There are 63 points in B and 43 in A, so there
are (63 - 43)/2 = 76 pairs of points in B - A. Extending this as suggested, we have [(n + 2)3 - n3]/2
lines for i, j,k < n.
293. We can count the total number of ways to seat six people in a row by counting the number
of ways none are seated properly, the number of ways one person is seated properly, two people
are seated properly, etc. This leads us to the Principle of Inclusion-Exclusion. If we let #(A) be
the number of ways the people can be seated with person A in the proper seat, the Principle of
Inclusion-Exclusion becomes
6!= (jW)- (*WnB)+ CjWnBnC)- \#(A n B n C n D n E n F) + X,
where X is the number of seatings in which no one is in the proper seat. Here we note that #(A) is the
same for each individual. Similarly, #(A D B) is the same for each pair. This accounts for (^)#(A n B)
in the Principle of Inclusion-Exclusion since there are Q) pairs. To evaluate #(A), we note that 1
person is fixed and the others can be seated in 5! = 120 ways. Similarly #(A D B) = 4! = 24 since the
4 non-fixed people can be seated in 4! ways. Hence, we have
X = 6! -
so X = 265.
294.
i. Here we are only interested in the number of 2's, 4's, 6's, etc., so our generating function is
(1 + x2 + x4 + • • • )(1 + x4 + x8 + x12 + • • • )(1 + X6 + x12 + x18 + • • •) • • • .
ii. We only want the odds, but only one of each; thus, each factor of the generating function
terminates after two terms. Our generating function therefore is
(1 + *)(1 + x3)(l + x5)(l + x7)(l + x9) ■ ■ ■ .

iii. The problem is essentially finding the partitions of a number containing only l's, 5's, 10's,
20's, 50's, and 100's, so the generating function is
(1 + X + x2 + • • • )(1 + x5 + J10 + • • • )(1 + x10 + x20 + • • • )(1 + x20 + x40 + ■ ■ ■ )(1 + x50 + • • • )(1 + x100 + •••).
iv. This one's a bit tricky. Whatever x, y, and z are, 2x is a multiple of 2, 3y is a multiple of 3,
and 7z is a multiple of 7. Hence, each solution of 2x + 3y + 7z = n represents a partition of n with x
2's, y 3's, and z 7's. Since z < 4, the generating function factor for the 7's terminates after the fourth
term. Thus, our full generating function is
(1 + x2 + x4 + x6 + • • • )(1 + x3 + x6 + x9 + • • • )(1 + x7 + x14 + X21).
295. Each term can be characterized by the number of a's, b's, c's, and d's it has. Let xz be
the number of z's the term has. Since each term is the product of 10 a's, b's, c's, and d's, we have
xa + x& + xc + Xrf = 10. Each term in the expansion corresponds to a different solution (xfl,x&,xc,Xd).
Hence, as discussed in the text, there are (g3) solutions to this equation and therefore 286 terms.
296. First note that if there are n elements in a set, then there are 2" subsets of that set (since in
forming a subset, we can either include or exclude each element). Hence, from n(A) + n(B) + n(C) =
n(A U B U C), we have 2^ + 2^ + 2^ = 2ÛSuCl, or 2101 + 2^ = 2ÂuBuCK Since |C| > 0, the only possible
solution of this equation is |C| = 101 and \A U B U C| = 102. We can relate our determined expressions
to the desired \A D B D C| via the Principle of Inclusion-Exclusion:
\A u b u C| = \a\ + \B\ + \c\-\An B\ - \A n c\ - \B n c\ +\A n B n C|.

Solving for \A D B n C|, we find
\A nbnq = -199 + \A n B\ + \A n q + \B n q.
We now apply the Principle of Inclusion-Exclusion to the \A D B\ terms to get
\AnB nc\ = -199 + 2(\a\ + \b\ + |q) - \a u b\ - |A u q - \b u q

= 403 - \A U B\ - \A u q - \B u q.
Each of the |A U B\ terms can be no larger than \A U B U C| = 102, so the smallest possible value of
\A n B n q is 403 - 3(102) = 97. Can you construct sets which satisfy this problem and our discovered
minimum?
297. We'll derive the general result then apply it to the specific case in the problem. If we consider
the boys indistinguishable and also the girls, our problem becomes finding the expected number of
bg's or gb's in a sequence of m b's and n g's. Label the chairs from 1 to m + n. Consider the pair of
chairs (i, i +1). The probability that there is a boy and a girl in these chairs is the same as the expected
number of boy-girl pairs in the two chairs (since there can be only one boy-girl pair in the chairs at
a time). To count the number of sequences with bg or gb in chairs (i, i + 1) we place the boy and girl
in the chairs (2 ways, bg or gb) and then put the remaining m - 1 boys in the other m + n- 2 chairs
^ (^m-i2) ways- Since there are (w^) ways to place the m b's in the m + n places and (m + n - 1)
adjacent pairs of chairs, our desired expected number is
(m + n- l)(2)(m^~2) _ 2(m + n - l)(m + n - 2)\m\n\ _ 2 mn

(m+n) {m + n)\(m - l)\(n -1)! m+n
In the given case, we have 2(7)(13)/(7 + 13) = 91/10.

298. First we count the number of sets of n w-cubes, no two of which are on the same level. Start
from the top of the cube and work downward. There are n2 choices from the top level. Eliminating all
the cubes in the same level as the first chosen level, we are left with (n -1)2 choices from the next level
down. (Make sure you see this.) Continuing in this manner, we have n2(n - 1 )2(n - 2)2 • • • l2 = (w!)2
sets of n w-cubes, no two of which are on the same level. Let A be the set of these sets.
Now we find the maximum number of these sets which have two or more w-cubes of the same
color. If we can show that the number of such sets is less than the number of possible sets, we can
conclude that there must be some sets with no two w-cubes of the same color. Suppose we pick two
w-cubes of the same color. There are w3/4 ways to pick the color and (2) = 6 ways to pick the pair of
cubes from among the cubes of that color. To count the ways to pick the remaining cubes, we have,
by the same argument as above, [(n - 2)!]2 ways to pick the remaining n - 2 cubes to complete some
member of the set A. Thus, our total number of members of A which have more than one w-cube of
some color is no more than 6(«3/4)[(w - 2)!]2. (Make sure you see that this is the maximum possible
number of such sets, not the exact number; we have overcounted because we have included several
sets with at least two blocks of the same color much more than one time.) Hence, the maximum
portion of the members of A which have more than one w-cube of the same color is
6(w3/4)[(n - 2)!]2 _ 3n
(w!)2 = 2(w - l)2'
For all n > 4, this fraction is less than 1. (Why?) Hence, there must be some portion of the members
of A which have no two w-cubes of the same color.
299. To show that 2” < /(n), we note that if we only take steps north and west, all of the paths thus
formed are self-avoiding. Since at each juncture we have 2 choices (north or west), we have at least
2n self-avoiding paths. If we count our paths by only excluding those in which we don't directly
backtrack (i.e. go west then on the next step go east), we have 4 choices on the first step, then 3 on
all subsequent steps, for a total of 4 • 3"_1 paths. Unfortunately, this may not accurately count the
self-avoiding paths because it is possible in this set-up to form loops rather than just backtracking to
cross our own path. Hence, the number 4 • 3n~l is an upper bound and we have 2” < f(n) < 4 • 3n_1.
300. For n = 1, the sum is 2. Let n > 2. The first digit is 1, and there are = (2n~2) ways to
arrange the l's among the remaining positions. Now consider any position except the first. If we
put a 1 there, there are then (2"_J22) = (2"“2) numbers with a 1 in that position, so that when we add
all the numbers there are (2nw-1) 22"-1's and (2"~2) of for the other k's. Our sum then is
+ 22m~2) + 2” ~ 22n~1 =
2w - 2N 2n ~
(22”-1 -1) +
n
Chapter 18
Again and Again
18-1 The continued power is

2l2 _ r,2z4 _ n216
2Z‘ = 2Z~ =2
which is absolutely enormous. The continued fraction is
1 „. 1- n 12 70
2+
-2+2+~5_2+29 "29'
2+*k
The continued root can be evaluated only with a calculator, where it equals
+ V2 + V^2 + V3.414 \/2+ \/2 +V3.848 « ^2 + V3.962 « V3.990 1.998.
(Can you guess where this continued root is heading?)

18-2 The first yields the equation
x=2+
x
which gives the quadratic x2 - 2x - 1 = 0. Solving for the positive root, we get x = 1 + V2.
The second yields the equation
x = 2+—i-p
3+ 1
which simplifies (with some elbow grease) to the quadratic 3x2 - 6x - 2 = 0. Solving for the positive
root, we get x = 1 + Vl5/3.
18-3 If it had been 2, the expression
1
something
< 120 >

would have had to be greater than 1, even though the "something" is greater than 1 (because it is a
positive integer plus something else positive). It it had been 4, then 1/(something) would have had
to be negative, an impossibility.
18-4 Removing the fractions one by one, the continued fraction does equal the proper fraction.
18-5 Since 29 • 5 = 145 = 147 — 2, we have
147 2 1
- — 5 "I- — — 5 ■ -
29 29 29/2'
Since 2 • 14 = 28, this yields
1
5+
14 + \
as the final continued fraction.
Since 7 • 4 = 28 = 29 - 1, the next fraction gives
29 , 1
7 ~ + 7
immediately.
Since 12-5 = 60 = 70-10, the last one yields
70 _ c 10 _ c 5 _ 1 c 1
12 12 6 6/5 1 + 1
18-6 We have bt = l/(b,-_i - «,•_i) = l/(b,-_i - |&i-iJ) = l/{b,-i}.

18-7 We can take any continued fraction which terminates and roll up all the fractions until
we get a single, rational, number. Thus this finite continued fraction cannot represent an irrational
number.
18-8 We take n and subtract a\ = 3. We take the reciprocal of what remains, to get 7.0625. We
subtract *4 = 7 and take the reciprocal of what remains to get 15.996595. We subtract a3 = 15 and
take the reciprocal of what remains to get 1.003417. We subtract *4 = 1 and take the reciprocal of
what remains to get 292.646, from which we take *4 = 292. We could go on, but eventually round-off
errors would catch up with us, so we'll be satisfied with
71 3+
7+
15+-
' 292-t-
18-9 The method is the same as we used in the previous exercise. Enter the number you're
finding the continued fraction for and subtract off its integer part, writing down the integer part as
a\. Take the reciprocal of what remains, and subtract off and write down the integer part *4. Take the
reciprocal of what remains and write down the integer part *4. And so on. And so on. And... you
get the picture.
18-10 The first few convergents are 3,3 + \ = y, 3 + 1/(7 + 1/15) = 3 + ^ How good are
these as approximations? This third one already yields 3.1415, an excellent approximation to n.
18-11 We prove it by induction. For the base case, we have P\ = a\, Qi = 1, P2 = *4*4 + 1, and
0.2 = 02/ so that
P1Q2 ~ P2Q1 = 0102 ~ (0102 + 1)(1) = -1 = (-1)1-
122 > CHAPTER 18. AGAIN AND AGAIN
For the inductive step, we assume it holds for k — 1. For k we have
PkQk+l ~ Pk+lQk ~ Pk(ak+lQk + Qfc-l) ” (ak+\Pk + Pk-l)Qk

= PkQk_! - Pt-aCt = -(-l)t_1 = (-1)*/
and our induction is complete.
301. Since
= 2,
we have
e*2 =2.
Taking natural logs of both sides, we have x2 = In 2, or x = Vln2.

302. We have
1 11
cn = n + --i— l = n +-i— = n +
In + 2 n+... n + n + 2 n+... n + cn
2 „2 J2 ™
Multiplying through by n + cn, we have c„ + ncn = nl + ncn + 1, or c„ = ft2 + 1. Thus _
E4 = X>2 +1) = E«2 + E i = + u = 517.

k=1 fc=l fc=l Jt=l
303. Since L V/c2 + 1J - k, we have
Vfc2 + 1 = k + 4
A
for some A. Solving for A, we have
(V/c2 + 1 + it)
A = = Vfc2 + i + fc,
V/c2 + 1 - (V/c2 + l-/c)(V/c2 + l + /t)
so that the continued fraction expansion is
1
V/c2 + 1 — k H—— — k i-= k h- ,
a jc+ VFTT k + k + l;^-.i
= k+ - k+
k + k+-i 2k +
k+k+y/W+1 ^'v ' 2fc+^
The period is 1.
304. The golden ratio (p satisfies the quadratic equation
(p2-(p- 1,
so that (p = 1 + 1/cp, or
1
<P 1 +
0
1 +
1+1
<P
1 +
1 +
i+ i+-
(You could also use the method of the text to find the continued fraction methodically.) The first five
convergents are 1,1 + 1/1 = 2,1 + 1/(1 + 1/1) = 1 + 1/2 = 3/2,1 + 1/(1 + 1/(1 + 1/1)) = 1 + l/(3/2) =
1 + 2/3 = 5/3, and 1 + 1/(1 + 1/(1 + 1/(1 + 1/1))) = 8/5. (Do you see a pattern here? Can you prove
that the nth convergent is F„/F„_\, where F*- is the A:th Fibonacci number?)
305. We square the equation A = \/6 + 2 V5 - \J6- 2 y/5, to get
A2 = (V6 + 2V5)2 - 2 ^6 + 2 V5 V6-2V5 + (V6-2V5)2 = 6 + 2 V5 - 2 Vl6 + 6 - 2 V5 = 4,
so that A = 2. We further have B = A - or B2 - AB + 1 = 0, or B = (A + VA2 - 4)/2 = 1. Thus

A + S = 3.
Chapter 19
Probability
19-1 If two events are mutually exclusive, they cannot both occur. Thus, P(A n B) = 0 and
P(A U B) = P(A) + P(B).
19-2 Applying the Principle of Inclusion-Exclusion, we must first add the probabilities of each
of the events occurring, then subtract the probabilities of each pair of events occurring, and finally
add the probabilities of all three events happening. Thus, we have
P(A U B U C) = P(A) + P(B) + P(C) - P(A n B) - P(A DC)- P(B n C) + P(A nBnC).
19-3 The square, or possible region, has area 42 = 16. To satisfy the problem, the
point must be outside the smaller square shown. (Make sure you see why.) The area
of the desired region is then 16 - 22 = 12. Hence the probability is 12/16 = 3/4.
19-4 The segment CG represents those cases where the woman arrives at 12:00
and still meets her husband, since he arrives at 11:40 at the earliest for those points.
Similarly, CH represents the cases where the man arrives at 11:50 and still is able to pick up his wife
because she arrives between 11:40 and 12:00.
19-5 Graphing the two numbers against each other gives a 10 by 10 square
of possible choices. We find the desired points as before. Let the vertical
number be 0 and note that the horizontal can be between 0 and 5. Gradually
increasing the vertical number yields the shaded region as the area of all points
such that the positive difference between the horizontal and vertical numbers
is less than 5. The area of the possible region is 100 and that of the desired
region is 100 - 25/2 - 25/2 = 75; hence, our desired probability is 75/100 = 3/4.
19-6 If we were only to consider integers, our problem would become a discrete one rather than
continuous as we could count the individual cases of the choices of numbers. Our ratio of areas
method would fail.
19-7 Convinced yet?
19-8 Still don't buy it?
19-9 Let event A be having the disease and event B be the test being positive. We are given
P(A) = 0.01, P(B\A) = 0.9, P(B|A') = 0.2,
< 124 >

where the last two are the probabilities of the test being positive given that I have the disease and
that I don't, respectively. We are asked to find P(A\B). Thus we must determine P(A n B), the
probability that I have the disease and the test is positive, and P(B), the probability that the test is
positive. We have P(A D B) = P(A)P(B\A) = (0.01)(0.9) = 0.009 (the probability I have the disease
times the probability the test is positive when I am diseased). To find P(B), we consider the two
mutually exclusive cases of my being diseased and testing positive and my not being afflicted and
testing positive: P(B) = (0.01)(0.9) + (0.99)(0.2) = 0.207. Hence, the probability that I am afflicted
is (0.009)/(0.207) = 1/23. Kind of surprising, isn't it? We can improve the accuracy of testing by
repeating the procedure on those individuals who test positive the first time.
306. This is simply a counting problem. There are 52! ways to order the cards. To count the ways the
two given cards can be adjacent, we consider the two cards as a unit, so we can order the 51 units
(our 2 cards together plus the other 50 cards) in 51! ways. Since the two given cards can be ordered
in two ways, our final probability is [2(51!)]/52! = 1/26.
307. Applying P(A\B) = P(A Pi B)/P(B), we have
= (l/2)(4/6) = 16
(l/2)(4/6) + (l/2)(3/8) 25 ‘
308. The winning team must win 3 of the first 5 games and then the last game (not 4 of the 6
games, since the winning team must win the last game). We can choose the winning team in 2 ways
and the 3 games they win in ( ) = 10 ways. The
3 6 games each have 2 possible outcomes, for a total
of 2 = 64 possible overall outcomes of the six games. Hence, our probability is 2(10)/64 = 5/16.
6
309. Although this is a conditional probability problem, we can attack this best with our simple
counting techniques. There are 3(3)(3) = 27 ways to make three rolls over 3 and there are 3 ways (all
4's, all 5's, all 's) to have a desired outcome, for a probability of 3/27 = 1/9.
6
310. The only way we can have a desired outcome is if the center of the coin is directly in the
center of one of the chessboard squares. Thus, the area of desired region is 0, since there is no two
dimensional region of desirable points. Hence, our probability is 0.
311. The described numbers can only be those with four 9's and a 7 and those with three 9's and
two 's. There are ( ) = 5 of the former type and Q) = 10 of the latter. Of all of these, only 3 are
8 3
divisible by 11. (Using the divisibility rule for 11 makes these easy to find.) Thus, our probability is
3/15 = 1/5.
312. There are two distinct combinations which total six: 1-2-3 and 2-2-2. The latter can only
occur in one way while the former can happen in ways. (There are ways to order the members
6 6
of the set (1,2,3}.) Since all 7 of these cases are equally likely to occur, our probability is 1/7.
313. We can choose the 6 cards in (562) ways. The two fives can be chosen in (4) ways and the
remaining 4 non-five cards in (448) ways. Hence, our probability is
126 > CHAPTER 19. PROBABILITY
314. We find the probability that no two girls are adjacent and subtract from 1. There are 9! ways
to seat the children in a circle. To count the number of ways to seat the children so that no two girls
are adjacent, we consider one girl fixed and count the ways the rest of the students can be seated
about her. Because no two girls can be next to each other, we know which 4 other seats contain girls
and which boys. Thus, the number of ways we can seat the children in these is (4!)(5!). (Once we fix
the location of one girl, seating the others is like seating them in a row.) Hence, our probability is
4! 5! _ 125
1 ~ ~9T ~~ 126'
315. Since all three people give the same answer, they are either all liars or all telling the truth.
If they are all telling the truth, it is raining. Our desired probability then is the probability that all
three are telling the truth divided by the probability that all three give the same answer, or
(4/5)3 _ 64
(1/5)3 + (4/5)3 65'
Compare this to our discussion of conditional probability with event B being all three people giving
the same answer and event A being all three people being truth-tellers.
316. There are (554) ways to select the cards. To find the number of ways to get 5 of a kind, we
consider 2 cases: one joker plus 4 of a kind, and two jokers plus 3 of a kind. The former can occur
in 13(2) ways, with 13 ways to choose the duplicated card and 2 ways to pick the joker. The latter
can occur in 13(4) ways with 13 ways to pick the three of a kind and 4 ways to pick the three suits
included among these. Hence,
13(2) + 13(4) _ 13(6!)(49!)

541/49! 5! “ 54!
317. There are 500 possible choices. Since there are [500/7J = 71 multiples of 7, [500/11J = 45
multiples of 11, and L500/77J = 6 multiples of 77 less than 500, there are 71 + 45 - 6 = 110 numbers
between 0 and 500 which are multiples of 7 or 11 (by the Principle of Inclusion-Exclusion). The
desired probability is then 110/500 = 11/50.
318. Shown are five of the holes in the mesh; one hole plus the four
surrounding it. Since the spark has a 1 mm radius, its center must be at least 1
mm from the wires which form the mesh. Thus, the center of the spark must
be inside the shown shaded square which has side 3 mm (since the center
must be at least 1 mm from each wire). Thus, the area of the desired region is
9 mm2. The possible region is a bit tricky. We can't just consider the interior of
the square mesh because that would exclude the possibility of the center of the
spark actually hitting the wire mesh. Thus, we extend the possible region to the center of the wires
which form the mesh as shown in the diagram. The possible region then is 0.5 mm further outside
the hole in the mesh since the wires have diameter 1 mm. Hence, the probability is 9/36 = 1/4. This
approach is valid because we can cover the entire screen with these 36 mm2 'possible' regions, each
containing 9 mm2 'desired' regions.
319. Without loss of generality, let AB = 2. Hence, AM = 1. Clearly either AP or PB is greater
than or equal to AM. Let AP be this largest side. Thus, we require PB + AM > AP in order for the
three lengths to form a triangle. Letting BP = x, where x < 1 (since BP < AP), we have x + 1 > 2 - x,
so x > 1/2 must be true for the three lengths to be the sides of a triangle. Our possible region has
length 1 < x < ) and our desired region has length
(0 1 (
1/2 < x < ) so our probability is 1/2.
1/2 1
0 0.5 1.0 1.5 2.0 2.5

320. We plot x and y on number lines, one ascending and one descending as
y •—__—_—. shown. The bold portions of the lines show where the numbers will be rounded
up. The described sum will only equal three when both x and y are rounded
up. As the number lines make evident, both x and y are rounded up for a total length of 1 out of the
possible 2.5 for a probability of 1/2.5 = 2/5.
321. Since all lattice squares are identical, we look at one lattice square. In the
diagram, circular arcs of length 1 have been drawn from each vertex; the points
within each quarter-circle are within 1 unit of the vertex which is the center. The
possible area is the area in the lattice square, or 1. The desired area is that area in the
lattice square which is inside exactly two of the quarter-circles drawn. The Xed regions are regions
inside either 4 (the center region) or 3 (the rest of the Xes) quarter-circles. The other regions are the
ones we desire. Consider the region covered by the quarter-circles centered at A and B. The leftover
region in the square but not in these quarter-circles is one of the four desired regions. We can find
the area covered by the two quarter-circles by adding the area of the two and subtracting the area
of their intersection. Doing this as discussed in Volume 1, we find the area contained in the quarter
circles is n/6 + V3/4. Therefore, our desired area, and hence the probability (since the possible area
is 1), is
4[1 - (ti/6 + V3/4)] = 4 - ~ - V3.
322. Let the three arcs AB, BC, and AC have measures x, y, and z, where x > y > z. Since the three
together form a circle, x + y + z = 360°. Since x is the largest arc, it must be at least 120°. The three
points lie on a semicircle if this largest arc is greater than 180°. Since x is equally likely to be any
value from 120° to 360° (we can see this by supposing we select the points by first choosing x, then y
and then labelling the points), our possible range has length 240° and our desired region has length
180° (from 180° to 360°). Thus, our desired probability is 3/4.
323. Since we have continuous variables, we turn to a graphical solution. (LI)

y
Let the three numbers be (x, y, z), where x, y > z. First we find our possible
range. Suppose z - a for some a such that 0 < a < 1. In the figure at right, the
shaded region corresponds to all (x, y) such that both x and y are greater than (a, a)
or equal to z. We see that for any value of a we get a square for the desired
area, where the square is the entire unit square when z — 0 and gets smaller as
we increase z. Thus, we find that the possible region is a pyramid which has (0,0)
height 1 (z ranges from 0 to 1) and whose base is a square with side length 1
(corresponding to z = 0).
128 > CHAPTER 19. PROBABILITY
The desired range is a bit trickier. First, if z > 2/3, then we are in the _ (1/1)
desired region, since the largest number is at most 1, the difference between
the largest and smallest cannot be larger than 1 - 2/3 = 1/3. This corresponds
to the entire possible region for z > 2/3 and is thus a pyramid with height 1/3
and square base with side length 1/3. For z < 2/3 consider the diagram, letting
z = a. The shaded square shows all points for which neither x nor y is more
than 1/3 greater than a. Any point outside this square has either x or y more (0,0)
than 1/3 greater than z. Thus, for all z < 2/3, the desired region is a square
of area (l/3)(l/3) = 1/9. Since all these cross-sections have the same area, the volume of the solid
formed by putting these together is (l/9)(2/3) = 2/27 since the solid has height 2/3. Our probability
is a ratio of volumes. The volume of the possible region is (l)(l2)/3 = 1/3 and that of the desired
region is 2/27 + (l/3)(l/3)2/3 = 7/81, so our probability is (7/81)/(l/3) = 7/27.
Chapter 20
Find It and Make It
20-1 First let point C be a point such that its distances from lines / and
m (CB and CA, respectively) are equal. Since CB and CA are distances to
the respective lines, they are perpendicular to the lines. Since AC = BC,
LCBO = ZCAO = 90°, and CO = CO, we have aOCA = aOCB by HL
congruence. Hence, /COB = ZCOA, so any point in the locus is on one of
the angle bisectors of the angle formed at O. Conversely, if C is on one of the
angle bisectors at O, then from SA congruence applied to right triangles CBO and CAO, CB = CA,
so C is on the locus. Hence, our locus is the angle bisectors.
20-2 The locus is a pair of planes, one above and one below the given plane.
20-3 Continue CO to X. Since CX is a diameter, ZCDX = 90°. Since DC

is perpendicular to both AB and DX, we have AB || DX. Since ZDCP = ZPCX,
we have DP = PX and point P is the midpoint of DX. Since DX || AB, we have
BD - AX and point P is the midpoint of AB as well. Hence, the locus is the
endpoints of the diameter perpendicular to AB. (The other endpoint occurs
when C is on the other side of AB.)
20-4 Did you really do the construction?
20-5 The incenter is the intersection of bisectors of the angles of the triangle. Thus, to get the
incenter, we draw the angle bisectors of two of the angles and their intersection is the incenter.
Similarly, we draw the perpendicular bisectors of two of the sides and find their intersection to get
the circumcenter. To get the orthocenter, we draw the line from A perpendicular to BC and the line
from B perpendicular to AC. Their intersection is the orthocenter. To get the median from A to BC,
we locate the midpoint M of BC in the same way we find the perpendicular bisector. We then draw
AM and similarly draw another median. Their intersection is the centroid. To draw the circumcircle,
we draw a circle with center O (the circumcenter we found) and radius OA. The incircle is a bit
tougher. We have the center, but the radius is the perpendicular distance from the incenter to the
sides. Hence, we draw the perpendicular to a side and find where it meets the side (call this D). We
then draw a circle with center I (the incenter) and radius ID.
< 129 >

130 > CHAPTER 20. FIND IT AND MAKE IT
20-6 First we make the square with side AB. We draw lines per¬ A D
pendicular to AB through A and B. We wish to find the point D on line l
l such that AD - AB. We do this by drawing an arc with center A and X \jY /
zf/
radius AB. The intersection of this with l is D. Drawing the line through A t/
D perpendicular to /, we find C as shown. Finding the square with
diagonal AB is a bit less complicated. Since the diagonals of a square c m
B
are perpendicular bisectors of each other, we draw t, the perpendicular n
bisector of AB. The vertices of a square are equidistant from the intersection of the diagonals, so we
draw the circle with center X and radius XA. The intersections of this circle with line t give the other
two vertices of the square.
20-7 Since we can make a square, we can make a segment of length V2

by making a square with side length 1 and drawing the diagonal. This is a
result of the Pythagorean Theorem. If we instead draw a rectangle with side
lengths 1 and 2, the diagonal has length V5. We can continue in this way
to note that the diagonal of a rectangle with side lengths 1 and 4 has length VT7. Thus, we start
with AB of length 1 and draw a line through B perpendicular to AB. We then draw a circle with
center B and radius AB to get C on the perpendicular line so that BC = AB = 1. Drawing a circle
with center C and radius AB locates D such that BD = 2. Similarly we can make E and F (or just a
circle with center D and radius BD locates F). Connecting A to F gives us our desired segment since
AF = VAB2 + BF2 = Vl + 16 = VT7.
20-8 We use the construction in the previous example to divide AB into p + q equal segments.
We can then easily select the appropriate P from among the dividing points on AB. (Make sure you
see how!)
20-9 Let the length of the desired segment be x. Hence we have x = Aab, or
(x)(x) = (a)(b), which reminds us of the Power of a Point Theorem. If we have two
chords which intersect each other such that one is in pieces of lengths a and b and
the other is in two equal pieces of length x, then x2 = ab. The most common and
useful instance of one chord bisecting another is when one chord is a diameter
and the second chord is perpendicular to this diameter. Hence, we draw diameter
AB by drawing AE = a then EB - b on the same line. We find the midpoint M of this line and
draw the circle with radius AM (and hence diameter AB). To get our second chord, we draw a line
through £ perpendicular to AB and let C and D be the points where the line meets the circle. Since
AB is a diameter and AB _L CD, we have CE = DE. From the Power of a Point Theorem, we have
CE - x2 - (AE)(BE) = ab; hence, CE is a segment with the desired length.
324. Two triangles with the same base have the same area if their altitudes are the same. Hence, the
locus is the set of points equidistant from the line containing the given base. The locus of points
in a plane equidistant from a given line is a pair of parallel lines, or if the locus is in space, it is a
cylinder.
325. We know that the center of a circle passing through three vertices of a triangle is the
intersection of the perpendicular bisectors of the sides. Hence we just pick three points A, B, and C
on the circle. The center then is the intersection of the perpendicular bisectors of AB and BC.
326. Start with segment BC. Since point A, the third vertex of the equilateral
triangle, is the length BC away from C, it is on the circle with center C and
radius BC. Similarly, A is on the circle with center B and radius BC. Hence
we draw these two circles and the point A is the intersection of the two (either
intersection will do).
327. Call the point X. Any circle with radius a passing through X has a center which is a away
from X; therefore, the locus of all possible points is the set of points a away from X, or the circle with
center X and radius a.
328. First we find the points equidistant from the two lines. Since the lines are
parallel, the set of points equidistant from them is a third line, /, parallel to the first
two and exactly between them. Let the radius of the circle be r. Thus, the third line
from above is exactly r away from the other two. For a point then to be equidistant
from the two lines and the circle, it must also be r away from the circle. The points
which are r away from a circle with radius r are the center of the circle and the entire circle with
radius 2r (make sure you see this). Line l passes through the center of the circle and also meets the
circle with radius 2r in two points. Hence, there are 3 points in the plane which satisfy the problem.
329. We can make a 30° angle by bisecting a 60° angle. Since we can make an equilateral triangle,
we can make a 60° angle. We then bisect one of the interior angles of the triangle to form a 30° angle.
330. We can fill half the tank exactly by tilting the tank as shown
and filling it so that the water level connects the corners as shown in
the first diagram. Make sure you see why the shaded triangle is half
the area of the tank. Setting the tank level, we mark the height of half
the tank with a line as shown in the second rectangle. We then tilt the
tank again, filling until the water level connects the end of the chalk
mark to the corner of the tank as shown in the third diagram. The
darker shaded area is clearly half the upper rectangle, or 25% of the
whole tank. Hence, we have filled 50% + 25% = 75% of the tank with water.
331. If we can find one side of the triangle, we can easily construct the rest
of the triangle. If an equilateral triangle is inscribed in a circle, each side cuts off
a 120° arc. We can find AC, one side of the triangle, by finding a 120° arc of the
circle. We thus construct adjacent equilateral triangles BDC and ABD as shown.
The resulting ZABC has measure 120°. Hence, AC is one side of the triangle.
The third vertex can be found in many ways; one way is to draw the circle with
center A (or C) and radius AC (the third vertex is the intersection of this circle and circle B), another
is to extend BD to meet the circle again. Make sure you see how both of these produces E, the third
vertex of A ACE.
332. Point G is the intersection of medians BE and CM of the original C

triangle. Choose any point X on the line / which C moves on and find Y,
the centroid of AABX as the intersection of medians AF and XM. Since
CM/GM = XM/YM = 3, ACMX ~ aGMY and thus GY || CX. Hence, any
centroid formed as in the problem is on the line through G parallel to line
/ and the desired locus is this line.
132 > CHAPTER 20. FIND IT AND MAKE IT
333. Draw a line through P to circle 0 at point Z. The midpoint X^

of PZ is in the locus, so point Y is in the locus. We then connect Y f \^X^\
to the midpoint of OP (point X in the diagram) because this forms /-—'—" ^
similar triangles OZP and XYP (since PX/PO = PY/PZ -1/2 and \X^-——wVlIIx
ZXPY = zOPZ). Hence we have XY/OZ = 1/2 and XY = r/2. Z
Similarly, we can show that any point in the locus is r/2 away from
X. Thus, the locus is some portion of the circle with radius r/2 centered at X. (We might have
guessed this by drawing several points on the locus.) We can show that every point on the circle is
in the locus by drawing a line through P and an arbitrary point Y on the circle. Extending the line to
Z, since PX/PO = XY/OZ we have A OZP ~ A XYP by SAS similarity. Hence, PY/PZ = 1/2 and Y is
in the locus.
334. First note that AA'B'A is a trapezoid and that OO' is perpendicular to AB. Since OO' is the
median of AA'B'A, we have
OO' = (AA' + BB')/2 = (AG + GB)/2 = AB/2.
Hence, OO' has constant length and is always perpendicular to AB. Thus, there is only one point
which can be O', regardless of the position of G. The locus is the point X on the circle on the same
side of AB as A’ such that XO _L AB.
335. In order for a circle with center C to be tangent to two parallel lines,
its center must be midway between the two lines. To find the lines through A
and B which are parallel and have C midway between them, we note that the
line through C parallel to our two lines must pass through the midpoint of AB.
Hence, we construct the midpoint M of AB, then draw MC. The lines parallel
to MC through A and B are the tangents. We construct the circle then by drawing the line through
C perpendicular to the lines through A and B. Let X and Y be the intersections thus formed. Our
circle is then the circle with center C and radius CX (since the radius is perpendicular to the tangent,
CX is a radius of the desired circle).
336. In triangle ABC, ZABC > iBAC if and only if AC > BC. Hence, the points B such that
BC < AC for a given AC compose the interior of the circle (except for the segment AC), and therefore
the locus, centered at C with radius AC. (Make sure that you see that any point in the interior of the
circle with center C and radius AC is closer to C than point A is.)
337. Given the centroid G and a vertex A, what else can we construct on the
triangle? Let I be the midpoint of BC. Since AG/GI = 2/1, we can construct point I
by finding the midpoint, M of AG, then drawing a circle with center G and radius
GM. This intersects the ray from A through G at point I. Point I is then the midpoint
of BC, but how do we find B and C? We know that the diameter through I is
perpendicular to side BC since a diameter passing through the midpoint of a chord
is perpendicular to the chord. Hence, we draw the diameter through I, then the line
through I perpendicular to the diameter. The intersections of this line and the circle are points B and
C.
338. We'll use analytic geometry to attack this problem. Without loss of generality, let the circles
be
x2 + y2 r\ and
(x - a)2 + y1 = ri.
Drawing a line through a point P and the center 0 of a circle of radius r, we find that the power of
point P with respect to O is (OP - r)(OP + r). Applying this to the point (x, y) and the above circles,
we have
( \!xl + V2 - ri) ( \/x2 + y2 + ri) = ( \]{x- a)2 + y1 - r2) ( \/(x- a)2 + y2 + r2)
x2 + y2 - r2 = x2 - lax + a2 + y2 - r2.
Hence, all (x, y) which have the same power with respect to the two circles satisfy the equation
x = (r2- r^j/la. Thus, the locus is a line perpendicular to the line connecting the centers of the circles.
339. We can find three circles which pass through a single point by D
drawing the circles with centers A, B, and C which pass through the
circumcenter P of aABC. Since AP = BP = CP, these circles all pass
through P. We can then construct a triangle such that each of the drawn
circles are tangent to two sides of the triangle by constructing the common
tangents of the circles as shown. We draw these by constructing the
perpendiculars to the sides of the triangle through the vertices (the six
segments outside A ABC). The intersections of these with the circles are
the tangent points (such as M and N). Drawing the line through M
and N we form the common tangent to circles A and C, because MN is
perpendicular to AM and CN so MN is tangent to the circles. The intersections of these three tangents
form A DEF. The sides of aDEF are parallel to those of A ABC. Since these triangles are similar and
have the same orientation, corresponding lines in the triangles will be parallel. Hence, we locate X
by drawing PE and PF. Since XB || PE and XC || PF (since corresponding lines are parallel), we draw
lines through B and C parallel to EP and PP. The intersection of these lines is X. This is a tough
construction; read through it a few times to make sure you understand it. Then get a compass and
ruler and perform it.
340. Let points Y and Z be on segments AB and BC as shown.

Since AZYX = ABAO, we have
ZYXZ + LYBZ = ZBOC + ZOBC + lOBA = 180°.
Thus, quadrilateral BYXZ is cyclic. Since ZXYZ = ZXBZ, we find
zOBZ = lOBC = LXYZ = ZXBZ.
From zOBZ = ZXBZ, we deduce that O, B, and X are collinear. Thus

we deduce the locus is a set of n line segments emanating from O equal in length and equally spaced
about O. (Each line segment corresponds to a vertex of the polygon. For example, in the diagram
point X is on the extension of BO past O.)
Chapter 21
Collinearity and Concurrency
21-1 Clearly AB + BC = AC if B is on AC. If B is not on AC, then let X be the foot of the altitude
from B to AC. By the Pythagorean Theorem,
AB + BC = VAX2 + XB2 + VXC2 + BX2 > AX + XC = AC.
Hence B is on AC if and only if AB + BC = AC.

21-2 Since AC is a straight line through B, LDBC = 180° - lABD. Hence,
LEBD = ZEBC + ICBD = 180° + (LEBC - LABD).
Thus, ZEBD = 180° (and hence E, B, and D are concurrent) if and only if LDBA = LEBC.
21-3 In general, vectors a and b are in the same direction if one is a scalar times the other; that
is if there exists some number n such that a = nb.
21-4 If we choose X to be the origin, then x = 0, so that we need only prove that y and z are in
the same direction. Yes, this usually simplifies our problem.
21-5 Each of the ratios, using the notion of directed segments, is negative for our initial diagram,
so the product must be negative.
21-6 Create a number line through B and C such that B = 0 and C = 1. Let X' = x be on segment
BC. Hence, we have BX'/CX' = (x - 0)/(l - x) = c. There is only one solution for x, x = c/(c + 1).
Hence, there's only one point X' on segment BC that satisfies BX'/CX' = c. If X' is on line BC but
not on the segment, our ratio BX'/CX' = ~(x - 0)/(l -x) - c, and we find x = c/(c - 1). (How is this
related to directed segments?) Hence, there are two points on line BC which satisfy the problem,
one on segment BC and one outside the endpoints B and C.
21-7 In the previous exercise you proved that there is one possible X' on segment BC and one
other outside segment BC on line BC. For the former, BX'/CX' is negative and for the latter the ratio
is positive. (Make sure you see this!) Now there is only one point which satisfies our restriction.
21-8 We can't apply the Angle Bisector Theorem as suggested because the heart of our problem
is proving that Al is the bisector of LCAZ.
< 134 >

21-9
In saying that AAXC is right, we are assuming that point A is on HX (since HX is the
perpendicular to BC through X). This is what we are trying to prove, so we can't assume it is true!
Be very careful in using collinearity techniques on concurrency problems; it is very easy to make
erroneous assumptions like this. Check your steps closely.
21-10
Let G be the intersection of medians AX and BY. We wish to show that the segment CZ,
where Z is the midpoint of AB, passes through G. We'll do this by using our vector method of the
previous section. From our discussion on vectors, we have
7* -> —>
j_ A+B A+B+C
and
2 ~T~
Note that we can show that G has the above vector representation without assuming all three medians
go through G; just two medians are enough to find the vector form of G. Hence,
g_g=d±B-2C ^ g_g=A + B-2C
so that CG = (2/3)CZ. From this we conclude that C, G, and Z are collinear and AX, BY, and CZ are
concurrent. Note that we can also conclude that the centroid divides each median in a 2 : 1 ratio.
21-11 Suppose X, Y, and Z are on BC, AC, and AB such that
AY CX BZ _
cy'bx'az~ '
Let Z' be the point on AB such that AX, BY, and CZ' are concurrent. From Ceva's Theorem, we have
AY CX BZ' _
CY BX AZ/ ~ '
so that BZ/AZ = BZ'/AZ', from which we deduce that Z and Z' are the same point using the notion
of directed segments. This completes our proof of the converse of Ceva's Theorem.
21-12 Written in the alternative form, Menelaus's Theorem is easily distinguished from Ceva's
Theorem.
21-13
Let X, Y, and Z be the points of interest (midpoints for medians and feet of angle bisectors
for angle bisectors) of BC, AC, and AB, respectively. Medians are easy; BX = CX, AY = CY, and
AZ = BZ since the points are midpoints. The expression for Ceva's Theorem clearly completes this
case. For angle bisectors, we look for a way to evaluate the ratios. The Angle Bisector Theorem gives
us our answer: BX/CX - BA/CA, AY/CY - AB/CB, and BZ/AZ = BC/AC. Multiplying these gives
us the desired result.
341. The sum of all of our Greek angles is 180° since together they form a triangle. Using the
given equalities, we have 2(a + )3 + 6) = 180°, so a + + 5 = 90°. Hence, from AADB, lADB =
180° - (a + /? + 5) = 90°. In the same manner, we can show that BE and CF are also altitudes. Finally,
from these angle equalities, we can use the sine form of Ceva's Theorem to prove the altitudes are
concurrent. Yet another method to prove the concurrency of altitudes...
136 > CHAPTER 21. COLLINEARITY AND CONCURRENCY
342. Applying Menelaus's Theorem to AECB, since A, P, and D are collinear and are on sides EB,
EC, and BC, respectively, of the triangle, we have
PE CD AB _
cp'db'ae~ '
The first ratio is 1/r, the second is 3/1, and the last is (AE + EB)/AE = 1 + 2/3 = 5/3. Hence, we find
r- 5.
A 343. What a messy diagram... and we didn't even include the angle
bisectors! Since we are dealing primarily with angles, we'll use the sine
form of Ceva's Theorem which we proved in the chapter. Since AD, BE,
and CF are concurrent, we have
sin AD AC sin IBCF sin LABE _

sin /DAB sin lACF sin ZEBC
Suppose X on BC is the foot of the angle bisector from A. Since AD’ is the
reflection of AD in AX, AD'AX - ZDAX. Hence
AD' AB = AXAB - lXAD' = AXAC - /XAD = AD AC.
Similarly, we can show five other angle congruences like this one. Using these, we have
sin ZD'AB sin /ACE' sin ICBE' _ sin ZDAC sin /.BCE sin /ABE
sin AD'AC sin ABCF' sin /E'BA sin /DAB sin /ACE sin lEBC
Hence, if AD, BE, and CF are concurrent, then so are AD', BE', and CF'.
344. Since HC = HZ and CZ ± AB, AB is the perpendicular bisector of CZ. Hence, AZ = AC = AX
and BZ = BC = BY. Thus, AZ and BZ are medians of triangles CZX and CZY, respectively, which
are half the sides to which they are drawn. Hence, triangles CZX and CZY are right, so that
AXZY = ZCZX + /.CZY = 90° + 90° = 180° and X, Z, and Y are collinear.
d 345. For the first part, we determine BD in terms of the sides of the
triangle. Labelling AW, CW, and BX as shown and letting DG = y,
we have, since tangents from a point to a circle are equal, AW = AG =
AX = x, CW = CZ = w, DG = DZ = DY = y, and BY = BX = z. Hence,
BD = z + y, BC - a = w + 2y + z, AC = b - x + w, and AB = c = x + z.
B
From these and a little algebra, we find AB + BC-AC = 2z + 2y = 2BD.
Hence, BD - (a + c - b)/2, so that D must be the point where the incircle of AABC is tangent to BC.
Thus, the perpendicular to BC through D passes through the center of the incircle, as do the angle
bisectors l and m. Hence, the three lines in question are concurrent at the incenter of A ABC. The
second part of the problem follows easily from observations made above. The other points defined
like D are the other two points where the incircle is tangent to A ABC. Hence, BD - BI - s - b,
CD = CH = s - c, and AH - AI - s - a, and Ceva's Theorem completes the problem.
346. Seeing cosines involved, we think to try the sine form of Ceva's A
Theorem. Thus, we need to find all our Greek angles in the diagram in
terms of the angles of the triangle. Since the line through A is an angle
bisector, a = <p = A/2. The line through C is just an altitude, so 6 = 90° - B
and e = 90° - A. Finally, the line through B is a bit tougher. Let O be
the circumcenter and draw the circumcircle. We find /AOB = 2ZC. Since
AAOB is isosceles, /3 = lOBA = (180° - 2zC)/2 = 90° - C. Similarly,
5 = 90° - A. Now we apply the sine form of Ceva, finding that these lines are concurrent if and only
if
sin a sine sin 5 _ sin A/2 sin(90°-A) sin(90°-A)_^
sine/) sin@ sinjS sin A/2 sin(90° - B) sin(90° - C)
Applying the identity sin(90° - x) = cos x attains the desired result.
347. As we saw in an earlier chapter, the orthocenter and the centroid have fairly simple vector
representations upon taking the circumcenter as the origin. This highly suggests that vectors may
be the best way to go for this problem. Using our vectors, we have
—> A + B + C —> -* -> ->

OG =--- and OH = A + B + C.
Clearly, OH = 3OG, so O, G, and H are collinear since OH and OG are in the same direction. Since
OH has three times the length of OG, we conclude 2(OG) = GH, completing the second part of the
problem.
348. As we showed in a previous problem in the chapter on locus and construction, the locus of
all points with equal powers with respect to two circles is a straight line. If the two circles intersect,
then this line must contain the common chord of the circles, since the endpoints of the chord have
power 0 with respect to each circle. Let the circles in the diagram in the problem be Ci, C2, and C3,
labelled clockwise from the upper left. Let I be the intersection of AD and BE. Since I is on AD, it
has the same power with respect to C\ and C2. Since it is on BE, it has the same power with respect
to C2 and C3. Thus, I has the same power with respect to both Ci and C3 and hence is on CF, so the
three segments are collinear at I.
349. Triangles, lines, and concurrency—looks like a job for Ceva! A

Let X', Y', and Z' be the points where AX, BY, and CZ intersect the
sides of the triangles as shown. Seeing angles, we think of the angle
form of Ceva's Theorem. From the law sines applied to AABX and
aACX, we find
sin a sin(B + J3) sin 6 _ sin(C + cp)

~BX = AX an ~CX ~ AX ‘
Applying the law of sines to ABCX gives (sinj3)/CX = (sin<p)/BX

and combining this and the two above equalities yields
sin a BX sin(B + fi) _ sin cp sin(B +18)

sin 6 CX sin(C + cp) sin sin(C + cp)
We find similar expressions for the ratios of sines at the other vertices and find that the product
of the three yields 1, so by the converse of the sine form of Ceva's Theorem, AX, BY, and CZ are
138 > CHAPTER 21. COLLINEARITY AND CONCURRENCY
concurrent. If you aren't comfortable with the sine form of Ceva and would rather use lengths,
note that BX'/X'C = (c sin a) / (b sin 6) (from the law of sines applied to AABX and AAXC and
sin /AXB - sin /.AXC) and proceed as above to find the three ratios appearing in Ceva's Theorem.
Once again, their product is 1 and the lines are concurrent.
350. From the given equal angles, ABCD, ABEF, and CDEF are all cyclic
quadrilaterals. Let the three circumcircles be C\, Cj, and C3. If any two
of these are the same circle, then all three must be the same circle. Make
sure you see this; suppose C\ and C2 are the same. If so, then CDEF is also
inscribed in this circle since its vertices are also vertices of at least one of the
other 2 quadrilaterals. Hence, either these three circles are distinct, or they
are all the same. From a previous problem, if the three circles are different,
then the common chords AB, CD, and EF must be concurrent. Clearly this is
impossible for a convex hexagon, so the circles must all be the same. Finally, BC, DE, and AF are
chords subtending arcs of equal measure (since /CAB = /ECD - /FEA) and thus must have equal
length.
Chapter 22
Geometry Tidbits
22-1 Draw planes X and Y such that they are perpendicular. In plane X draw segments AB
and CD such that AB is perpendicular to Y and CD is parallel to Y. Thus, when we project plane X
into plane Y, the image of AB is a point and that of CD is still a segment. Since the ratio of images
A'B'/C'D' is 0 and the ratio of the original segments is nonzero, we have our desired example.
22-2 No. Again let the planes be X and Y. If the planes are parallel, a projection from X to Y
leaves the figure unchanged (make sure you see this). Otherwise they intersect in some line, call
it /. Now break up the original region into rectangles, no matter how small, with one pair of sides
parallel to l and the other pair perpendicular to l. Upon projection from plane X to plane Y, the sides
parallel to l remain unchanged (again, make sure you see why), while those perpendicular (and
hence the area) are all scaled by the same factor. Hence, no matter how figures are oriented, upon a
projection all of them are scaled by the same factor.
22-3 No. The image of a line segment under an orthogonal projection cannot be longer than
the original segment. This is a simple consequence of the Pythagorean Theorem: suppose A'B'C is
the image of ABC. If we move plane A'B'C' (straight up or straight down) so that A coincides with
A', right triangle ABB' has hypotenuse AB and legs BB' and AB' (or A'B'). Hence, A'B' < AB.
22-4 Let line / be in the plane projected by a central projection into another
plane such that m is the image of l. Clearly E, the intersection of the lines, is
its own image. Let /.OED = 90°, lAEC = 60°, lOBA = 45°, and lOAE = 90°.
Letting OE = 12, we find AE = 6 V3, OA = AB = 6, so BE = 6 V3 - 6. We
also find CE = 12 V3 and DE = 12(2 - V3), so DC = 24 V3 - 24. Hence,
AB/BE t CD IDE.
22-5 You should find that OA' gets larger and larger as OA gets smaller and smaller and that
OA' shrinks as OA grows.
< 139 >

140 > CHAPTER 22. GEOMETRY TIDBITS
22-6 We seek the point X on OP such that (OX)(OP) - r2, where r is the
radius of the circle. Writing this as OX/r = r/OP, we think to hunt for similar
triangles. By constructing the circle with diameter OP we locate points Q and
R, the intersections of this circle with circle O. Since OP is a diameter, AOQP
is right. Drawing QR which intersects OP at S, we have QS _L OP (why?),
so A OQP ~ aOSQ. Hence we have OQ/OS = OP/OQ, or (OS)(OP) = r2 as
desired. Thus, point S is the inversion of P with respect to O. What if P is
inside circle O?
22-7 Any two circles are similar and have the same orientation, so any pair of circles are
homothetic to each other. We can find the center by drawing lines through corresponding parts and
finding where they intersect. We can easily prove that two such lines go through Q, the point of
tangency by noting that the common tangent and the lines through the centers of the circles both go
through the point of tangency.
22-8 By SAS similarity we find AAEF ~ AACB. Hence, PE || BC and similarly A

we find that the sides of A ABC are parallel to those of ADEF. Hence, the triangles
ABC and DEF are both similar and homothetic. To find the center of homothecy,
consider the midpoint M of FE. By SAS similarity (AF/AB - FM/BD and ZAPM =
LABD) we have AAFM ~ AABD so that point M is on AD. Similarly we can show
that the medians of A DEF are on the same lines as the medians of A ABC. The center of homothecy
is the intersection of these lines, the centroid of the triangles.
22-9 Draw one ray from P, r\, pointing directly 'up' and let the area of the region be K. We wish
to create another ray, r2, from P such that the area of T between r\ and ri is K/12. We do so by letting
r2 be initially coincident with r\, in which case the area between the two is 0. Rotate r2 clockwise and
let x be the area inside T between r\ and rj (measuring clockwise from r\). As T2 rotates from r\ all
the way around the circle back to r\, x varies continuously from 0 to K. Hence, for some r2 we must
have x - K/12. We fix T2 at this position and hunt for r$ by rotating clockwise from 7*2 and so on.
22-10 Are you convinced that the theorem isn't very easily proven?
22-11 We can't use this approach on an infinite number of points because we have chosen the
three points with the minimum a. If we have an infinite number of points, there is not necessarily a
minimum value of a\ A simple counterexample is the set of lattice points on a Cartesian plane (i.e.
the points (z, j) where i and j are integers.)
351. Let G be the centroid of ABC. Let AD = 3x. Then, GD = x, AG - 2x, AD' = 3x, and GD' = 5x.
Hence, D', E' and F' are dilations of D, E, and P with ratio -5, meaning that GD' = 5(GD) and D' is
on the other side of G from D. Thus, D'E'F' is homothetic to ABC (since A D'E'F' ~ a DEF ~ a ABC
and lines through corresponding parts are concurrent) with center G and ratio D'G/AG = 5/2. Thus,
[D'E'F'],/[ABC] = (5/2)2 and [D'E'F'] = 75.
352. Two parallel lines never intersect; rather, they only meet at the 'point at infinity.' Since
their images are circles through the center which do not intersect anywhere but the center (because
the lines only intersect at the point at infinity, whose image is the center of inversion), the images
are circles through the center of inversion which are tangent at the center. This provides a way of
proving that lines are parallel (or that circles are tangent).
353. A circle with radius one has circumference 2n. Consider the four arcs in which the vertices
of the quadrilateral divides the circle. Since their sum is 2n, at least one has length less than or equal
to 7i/2. The central angle of this arc is then less than or equal to 7i/2 and hence the chord is less than
or equal to Vl + 1 = V2.
354. Assume that F is two-dimensional. If its image upon projection onto a is a straight line, F
must be perpendicular to a. Similarly F must be perpendicular to jS. Hence, a must be parallel to jS,
a contradiction. We conclude that F cannot be two-dimensional.
355. Let the right angle be at A. From SAS similarity, aAEF ~ A ABC. Since AB - AE, aAEF =
A ABC. Let the altitude from A to EF be AX and to BC be AY. Since AY = DY by reflection
and AX - AY from the congruent triangles, we have XD = 3(AY). Hence, triangles DEF and
ABC have congruent bases (EF and BC) and altitudes DX and AY such that DX/AY = 3. Finally,
[DEF] - 3[ABC] = 3.
A 356. It is very important to note that proving that the area of an ortho¬
diagonal quadrilateral is half the product of its diagonals does not solve this
problem. We are asked to prove the converse of this fact. We do this by
drawing the altitudes from A and C to diagonal BD and showing that their
D
feet coincide, so the line from C perpendicular to BD passes through A and
the diagonals are perpendicular. Let E and F be the aforementioned altitude
feet as shown. We are given [ABCD] = (AC)(BD)/2. From the diagram,
[ABCD] = [(AF)(BD) + (EC)(BD)]/2. Setting the two expressions for [ABCD]
equal to each other, we conclude that AF + EC = AC. Thus,
AC = AG + CG = VAF2 + FG2 + VeC^TeG2.
Since AC = VAF2 + VCE2, we deduce that FG = EG = 0, so F, G, and E are the same point and the
quadrilateral ABCD is indeed orthodiagonal. Make sure you understand the general method we
employed here in showing that points E and F are the same.
357. First we show that a line tangent to a circle at the center of inversion is parallel to the image,
line /, of the circle. Let the tangent be line m. Consider the images of / and m. The image of l is the
original circle and the image of m is m itself. Since the images of / and m intersect only at the center
of inversion, l and m only intersect at the point at infinity. Hence, lines / and m are parallel. Since
the tangents of the two circles in the original problem are perpendicular, the images of the circles are
lines parallel to these two and are thus perpendicular as well.
358. Let the points where the circles intersect be A, B, C, D, E,

and F, connected as shown in the diagram. Since the measure of
an arc is twice an inscribed angle, we can write
ZACB = LECB + LECA = (EB +AE )/2

LABC . = LABE + AFBC = (AF + FC )/2
/.BAC = aBAD + LGAD = (BD + CD )/2.
Summing these, we find that the sum of the desired arcs is 2(aACB+
LABC + ABAC) = 2(180°) = 360°, so the boundary's measure is
invariant.
359. Let H be the midpoint of altitudes AD and BE. Thus, ABDE is a A

parallelogram, so AE || BD, which is a contradiction to ABC being a triangle.
Hence, the orthocenter cannot be the midpoint of two altitudes. Remember
that if any two lines bisect each other, they are the diagonals of a parallelogram;
it will come in handy!
360. As we showed in the chapter, given any orientation, there

exists a line that bisects the area of a given region. Let the regions
be a and jS. It is not hard to see from the proof in the book that
the line of a given orientation that bisects the area of a is unique.
As we slide a line up from below a to above a, the area below it
increases continuously from 0 to [a\ and thus at only one point can
it bisect the area of [a]. Start with an orientation, say 6, and the
line with orientation 6 that bisects the area of a. Choose 6 such
G +180° that the line passes through jS. (We can do this by choosing a point
in j6 and finding the line through this point which bisects the area
of a.) Let x be the area of /3 to the right of the line bisecting the area
of a, where we initially arbitrarily choose which side is right. In the diagram, where the top is our
initial orientation 0, the right side is indicated by a star. As we rotate from 6 to 6 + 180° (finding
the line which bisects the area of a for each orientation) we come to the second diagram, where our
bisecting line is the same as the first, but the star, which rotates along with the line, is on the other
side. Now the area of jS to the right of the bisecting line is [/l] - x. Hence, as we continuously change
orientation from 6 to 0 +180°, the area in jS to the right of the line goes continuously from x to [jS] - x.
At some point in between, this area must be [j3]/2. This line is the desired line which bisects both
figures.
361. Since AC is common to both triangles, it is tangent to both incircles. Let these points of
tangency be E and F, where AE is tangent to the incircle of A ABC. We will show that these two points
are the same, so the two circles are tangent. We start with
AE = (AB +AC- BC)/2 and

AF = (AC + AD - CD)/l.
Subtracting these gives AE - AF - [(AB + CD) - (BC + DA)\/2. This last quantity is 0 because ABCD
is circumscribed about a circle. Hence AE = AF and E and F are the same point since both are on
segment AC.
362. Draw a line through 2 of the points A and B. Let x be the number of the 2n - 2 points (the
original 2n minus the two on the line) on the right side of the line (where we choose the right side
arbitrarily). As many times before, we spin the line 180° (so that the line always goes through A)
and the result is 2n - 2 - x of the points on the right side. Since in spinning we go from x to 2n - 2 - x
by l's, we must hit n - 1 somewhere. If we do this for all 2n points, we get at least n different lines
(2 points are 'used' for each line).
363. Extend side AD of a triangle which fits the description of the prob¬
lem to point B on the larger circle. Since aABC and A ADC share an altitude,
[ADC]/[ABC\ = AD/AB. Since the two circles are homothetic with center A,
AD/AB = 1/3. Finally, we see that by maximizing [ABC], we maximize [ADC].
The maximum area of a circle inscribed in a circle of radius 3 is equilateral and
has side length 3 V3, so its area is 27 V3/4. Hence, the maximum value of [ADC]
is 9 V3/4.
364. The argument here is the same as for the Pancake Theorem. Choose a line with orientation
a which bisects the area of S. Let x be the length of the boundary to the left of the line, rotate the
orientation from a to a + 180° (finding the area bisecting line for each orientation). The length of the
boundary to the left of the line must at some point be half the length of the boundary of S and we
have found our desired line.
365. Draw CE parallel to BD. Since ABCD is an isosceles trape¬

zoid and BECD is a parallelogram, AC = BD = EC, so altitude CE
is a median since A ACE is isosceles. Since CE = (AB + CD)/ 2 =
(AB + BE)/2 - AE/2, A ACE is a right triangle (the median from C
is half the opposite side). Since AC _L CE, AC _L BD and ABCD is
orthodiagonal.
366. Let the smooth convex figure be S. Given any orientation 6, there are two tangents to S
with that orientation, one on either side of S. Draw these two tangents. Draw also the two tangents
with orientation 6 + 90°. The intersection of these lines with the first two lines forms a rectangle.
Designate the sides of the rectangle with orientation 0 as the length and the other side as the width.
Let x be the ratio of the length to the width. As we rotate the initial orientation from 6 to 6 + 90°, this
ratio will go from x to 1/x. (Why?) At some point the ratio must be 1 and the rectangle is a square.
367. Instead of projecting a single plane as we have done several times, D

in this problem we project three planes (three sides of the drilled cube) onto
one plane. Hold the cube so that you are looking directly through the hole.
What you are looking at is what results when you project the cube down
upon a plane perpendicular to the axis of the drill. The figure is a portion
of this projection. The circle centered at A (which is the image of vertex A')
is the hole formed by the drill and thus has radius 1. Points B, C, and D are
the images of the cube vertices connected to vertex A' by edges of the cube.
Let the original cube vertices be B', C, and D'. The orientation of segment B'D' is perpendicular
to the orientation of the drill and hence is parallel to the image plane. Any segment which is
orthogonally projected onto a parallel plane has an image of equal length. Thus, BD — B'D' = 3 y/2.
Hence, BCD is equilateral and [BCD] = 9 V3/2. Now the ratio of [BCD] to the area of the shown
circle is equal to the ratio of [B'C'A'] + [B'D'A'] + [D'C'A'] to the area of the portion of the faces on
these three faces which are zapped by the drill. Thus,
area lost to drill _n

[B'C'A] + [B'D'A] + [D'C'A] ~ 9xf3/2'
From this we find that the area lost on these three faces is n V3 (since B'C'A and the other two are
isosceles right triangles with side length 3). Hence, on each face, n V3/3 is lost to the drill, leaving
9— 71V3/3.
368. The original diagram is on the left. Seeing lots

of circles with special intersections, we try an inversion
with respect to the third circle. The image of the other two
circles are lines since the circles pass through the center of
inversion. These lines are perpendicular since the circles
are orthogonal (proven in an earlier problem) and they
meet on the circle of inversion (at B, which is its own
image) because the original circles intersect at B. Points C
and D are their own images. Since iCBD is right and inscribed, it is inscribed in a semicircle. Hence,
CD is a diameter of the image circle. Since C, D, and the circle of inversion are their own images, CD
is the diameter of the original circle as desired.
A G B 369. First we show that it is impossible to cover the square with three triangles.
Since there are four vertices of the square, two of them must be covered with
one triangle. This can only be done when a side of the triangle is a side of the
square as shown. Consider points E, F, and G on the sides of the square such that
AG - 0.5, ED = 0.01 and FC = 0.02. A simple application of the Pythagorean
Theorem reveals that no two of these three points are within 1 unit of each other
and hence we need three different triangles to cover all three. Our minimum
number is 4 then, and the second diagram shows that this is indeed possible.
370. First note that if circle a is obtained by inverting circle ft with respect to
circle y, then the centers of a, ft, and y are collinear. (Why?) We can show that the
circumcircle of A ABC is the image of the circumcircle of AXYZ upon inversion
with respect to incircle I by showing that points A, B, C are the inverses of X, Y,
Z. (Remember, we only need three points to determine a circle; this is why we
need only show that three points on one circle are the images of three points on the other!)
C Let segment Al intersect EF at M. Hence, lEAM = ZFAM (AI is an

angle bisector), AM = AM, and EA - FA (equal tangents), so AEAM =
AFAM. From this, EM = EM and M is the same point as X. Since
LAF1 = ZFXJ = 90°, we have AAFI ~ AFXI so that (IX)(IA) = (IF)2.
Since IF is the radius of inversion and I the center, this proves that A
is the image of X. Similarly, we show that B and C are the images of Y
and Z and our proof is complete.
371. To construct the first outer medial triangle of A ABC,

we construct the line through A parallel to BC, the line through
B parallel to AC, and the line through C parallel to AB. The
intersections of these lines form our triangle. For the second
part, since we want to contain all the points inside the first
outer medial triangle, we look for a triangle with some maximal
quantity, such as maximum area or maximum perimeter. Since
we are dealing with the region inside the triangle, we'll try area. Because there are a finite number
of points, there is some group of three points among the n such that these three form the triangle of
largest area possible among all those formed by connecting 3 of the n points. Let this triangle be ABC
and its first outer medial triangle be DEF. Let the region on the opposite side of line EF from BC be
region I, and let regions II and III be defined similarly as shown. Clearly all points outside DEF lie
in at least one of these regions. Consider point P in region I. The distance from P to BC (the distance
from a point to a line is the length of the altitude from the point to the line) is greater than that from
A to BC, so [PBC] > [ABC]. Since [ABC] is the triangle of maximal area among those formed by the
original n points, P cannot be among those points. Similarly, none of the remaining n - 3 points can
be outside DEF and we have found our desired triangle.
For the third part, we want no points inside so we minimize the D

area. Let A ABC be the triangle of minimum area among those with
all vertices among the n points and let A DEF be its second outer
medial triangle. Let point P be one of the other n points. First,
it cannot be inside A ABC, or it would be closer to BC than point
A; hence, we would have [PBC] < [ABC], a contradiction. Next, it
cannot be between line / and segment EF. By SSS congruence we
have A ABC = aMCB, so that EF is the same distance from BC as point A is. Thus, any point between
/ and EF is closer to BC than point A. For a point P in this region we would then have [PBC] < [ABC],
a contradiction as before. Similarly we can eliminate the other regions of A DEF and A ABC is proven
to be our desired triangle.
372. Consider any point P inside A ABC. Draw two lines through P parallel to the directions of
the sides of the rectangle and let the four points of intersection of these lines with the sides of the
triangle be D, E, F, and G. By the Pigeonhole Principle (which creeps into a great many problems),
at least two of these points must be in the same rectangle (of the three that cover the sides). Let
the points be D and E and the rectangle which contains them both be Rj. If these are on the same
line through P, then P, also contains P because every point on DE will also be in P*. If D and E are
on perpendicular lines through P, the rectangle formed with opposite vertices at D and E and sides
parallel to the sides of P; is contained entirely within P/. Since P is also a vertex of this rectangle with
vertices at D and E, P is in P; as well. Since point P is arbitrarily chosen within AABC, all points in
AABC are contained in at least one of the rectangles.
Chapter 23
Number Theory
23-1 The quotient (b/ai)/(b/(a,b)) = (a,b)/ai. Hence, if a\\(a,b), this quotient is an integer, so
b/(a, b)\b/ai.
23-2 Since (2,6) = 2, we have to divide the modulus, 6, by 2 when we divide everything else by
2. This jives with the previous example, where we had to divide 6 by 2 to get 10 = 1 (mod 3).
23-3 In the first one, the modulus is 20 and we are dividing by 6. Since (20,6) = 2, we have to
divide the modulus by 2. We are thus left with a = b (mod 10).
In the second one, we can divide out 23. Since (23,5) = 1, the modulus remains the same and we
have 1 = 6 (mod 5).
In the third one, we are dividing by 6 and the modulus is 9. Since (6,9) = 3, we divide the
modulus by 3 to get 2 = 5 (mod 3).
23-4
i. First we subtract 45 from both sides to get 1235x = 9045 (mod 24). We then mod both
sides out by 24 (divide by 24 and keep the remainder) to get some smaller, manageable numbers:
llx = 21 (mod 24). All we need to do is divide both sides by 11. However, 11 doesn't divide 21, so
we add 24 to the right side until it is divisible by 11: 21 = 45 = 69 = 93 = 117 = 141 = 165 - 11 • 15.
We thus have llx = 11 • 15 (mod 24), so that x = 15 (mod 24) is a general solution.
ii. As above, we subtract 45 from both sides to get 1235x = 9045 (mod 11). We then mod both
sides by 11 to get 3x = 3 (mod 11). Dividing both sides by 3, we have x = 1 (mod 11) as the general
solution.
iii. We subtract 45 from both sides to get 1235x = 9042 (mod 11). We mod both sides out by 11
to get 3x = 11 (mod 11). This becomes 3x = 0 (mod 11), so that x = 0 (mod 11) is the general solution.
iv. We have 1232x = 9045 (mod 24), or 8x = 21 (mod 24). To divide a 2 from both sides, we try
adding 24 to the left: 21 = 45 = 69 = • • •. But the right side will never be divisible by 2 when we add
24's to it, since 24 is divisible by 2. Thus the equation has no solutions.
23-5 We solve the first congruence: 3x = 4 = 11 = 18 (mod 7), or x = 6 (mod 7). This has the
general solution x = 6 + 7j. Unfortunately, the second congruence has no solution, since adding any
< 146 >

number congruent to 5 is of the form 8k+5, and thus cannot be divisible by 4, as 4x is. Thus the system
has no solutions. Just for fun, let's solve the first and third congruences alone. For the third, we
have 5x = 6 = 15 (mod 9), so that x = 3 (mod 9). This has the general solution x = 3 + 9k; combining
this with the solution of the first congruence, we have 3 + 9k = 6 + 7j. This leads us to the additional
linear congruence 3 = 6 + 7/ (mod 9), or 7/ = -3 = 6 = 15 = 24 = 33 = 42 (mod 9), or / = 6 (mod 9).
Thus j = 6 + 91 for any /; substituting this into x - 6 + 7/, we have x = 6 + 7(6 + 91) = 48 + 63/ for any /.
23-6 In mod 7, the squares are 0, 1, 4, 9 = 2 (mod 7), 16 = 2, 25 = 4, and 36 = 1: the quadratic
residues are 0,1, 4, and 2. In mod 8, the squares are 0,1, 4, 9 = 1,16 = 0, 25 = 1, 36 = 4, 49 = 1: the
quadratics residues are 0,1, 4. In mod 9, they are 0, 1,4,9 s 0, 16 = 7, 25 = 7, 36 = 0, 49 = 4, and
64 = 1: the squares are 0,1, 4, 7.
23-7 The second m positive integers will have the same squares (mod 2m + 1) as the first m: the
square of 2m + 1 - k is (2m + l)2 - 2k(2m + 1) + k2 = kz (mod 2m + 1). Thus if the numbers 1 through
m all have different squares, we have m + l = (n + l)/2 quadratic residues: these m plus 0.
23-8 The first 16 squares are 1,4, 9,16, 25, 36, 49, 64, 81,100,121,144,169,196,225, 256. In mod
11, the first 10 are 1,4,9, 5, 3,3,5, 9,4,1. Adding in 0, we have 6 = (11 +1)/2 quadratic residues—the
maximum. In mod 17, the squares are 1, 4, 9, 16, 8, 2, 15, 13, 13, 15, 2, 8, 16, 9, 4, 1, so we have
9 = (17 + l)/2 quadratic residues including 0. Again the maximum.
23-9 In mod 3, the residues are 0,1, and 22 = 1. Thus 0 and 1 are the residues. (The important
lesson here is that no square is ever congruent to 2 (mod 3).) We found in an earlier example that
the quadratic residues (mod 8) are 0, 1, and 4—these are the only possible squares (mod 8). This
is another exceedingly useful tool: to prove some expression cannot be a square, just prove it is
congruent to, say, 3 (mod 8).
23-10 For n = 16 = 24 the product is (1+ 2 + 4 + 8 + 16) = 31. For n = 20 = 22 • 5 it is
(1 + 2 + 4)(1 + 5) = 7 • 6 = 42. For n = 28 = 22 • 7 it is (1 + 2 + 4)(1 + 7) = 7 • 8 = 56.
23-11 The key is the distributive law. For example, consider the above product for s(28). We have
(1 +2+4)(l+7), which on expansion by the distributive law is (1)(1) + (1)(7) + (2)(1) + (2)(7) + (4)(1)+4(7).
We get all possible combinations of prime powers, which is to say all possible divisors!
23-12 Since 1 + x + x2 -\-hi” = (xn+1 - l)/(x - 1), the product becomes
pT1 -1
Vi ~ 1 Vk~ 1
23-13 The "d\n" underneath the £ means that the sum is taken over all d which divide n. If we
add up 1 for all such d, we get the number of d which divide n, which is the number of divisors. If
we add up d for all such d, we get the sum of the d which divide n, which is the sum of the divisors.
23-14 s(24) = s(23 • 3) = (1 + 2 + 4 + 8)(1 + 3) = 15 • 4 = 60 > 2 • 24; 24 is abundant. s(26) = s(2 • 13) =
(1 + 2)(1 +13) = 3 • 14 = 42 < 2 • 26; 26 is deficient. s(28) = s(22 • 7) = (1 + 2 + 4)(1 + 7) = 7 ■ 8 = 56 = 2 • 28;
28 is perfect.
23-15 Since 2k+l - 1 is assumed to be prime, 2k(2k+1 - 1) is a complete factorization. Thus the
sum of the divisors is
(1 + 2 + 4 + • • • + 2*)(l + (2k+1 - 1)) = (2k+1 - l)(2k+l),
which is twice our original number. Thus any number of the form 2k(2k+1 - 1) is perfect.
148 > CHAPTER 23. NUMBER THEORY
23-16 For k — 1 we have 21(22 — 1) = 2 • 3 = 6. For k — 2 we get 22(23 - 1) = 2 • 7 = 28. To find

the next perfect number, we need to find k such that 2k+1 - 1 is prime. 24 - 1 = 15 is not prime, but
25 - 1 = 31 is. Thus the next perfect number is 24(25 - 1) = 16 • 31 = 496.
23-17 We take powers: l1 = 1, so 1 has period 1; 22 = 4, 23 = 8 = 1, so 2 has period 3;
3—>9 = 2—>6 = -l—> -3 —> -9 = -2 —> -6 = 1 (where —> represents going from one power to the
next), so 3 has period 6; 4 —» 16 = 2 —» 8 = 1, so 4 has period 3. Similarly 5 has period 6, and 6 has
period 2.
23-18 We must show that a6 = 1 (mod 7) for a = 1, 2,..., 7. The sixth powers are 1, 64 = 1,
36 = (33)2 = (-1)2 = 1,46 = (-3)6 = 36 = 1, 56 = {-if = 1, and 66 = (-1)6 = 1.
23-19 Fermat's Theorem is the key to this universal type of problem. We know that 622 =
1 (mod 23), so that
6iooo = 51000 (mod 22) _ 6io = (65)2 = 77762 s 22 = 4(mod23).
23-20 The period of a number (mod 23) must divide 23 - 1 = 22, so must be 1, 2,11, or 22. The
period of a number (mod 17) must divide 16, so must be 1, 2, 4, 8, or 16. The period of a number
(mod 7) must divide 6, so must be 1, 2, 3, or 6.
23-21 Since g has period p-1, the numbers gip-O/fa^ have periods 0, ^2/- • • respectively.
Why? Clearly these numbers taken to the respective powers all yield = 1. Can these numbers
be taken to smaller powers yield 1? Suppose {g^P-V/hy = ^ (modp) for c < d^. Then ydp-i)/4 = 1 for
c(p - 1)A4 < p - 1, violating the assertion that g has period p - 1. Thus the given numbers have the
desired periods.
23-22 20 has the prime factors 2 and 5, so the numbers relatively prime to it are those divisible
by neither 2 nor 5, namely 1, 3, 7, 9, 11,13, 17,19. The numbers relatively prime to 15 are divisible
by neither 3 nor 5, so are 1, 2, 4, 7, 8,11,13, 14. The numbers relatively prime to 12 are divisible by
niether 2 nor 3, so are 1, 5, 7, and 11 only.
23-23 As we found in the previous exercise, there are 4 numbers less than 12 and relatively
prime to 12, so that 0(12) = 4. Since 11 is prime, every number less than it is relatively prime to it.
There are 10 positive numbers less than 11, so 0(11) = 10.
23-24 For p prime, every number less than p is relatively prime to it. There are p — 1 positive
numbers less than p, so (p(p) - p — 1. For p prime, Euler's generalization yields a^P) = \ (modp), or
aP~l = 1 (modp). This is exactly Fermat's Theorem.
23-25 Every number less than pk is relatively prime to pk, except the multiples of p. Between 1
and p there are p - 1 numbers not divisible by p; between p + 1 and 2p there are p - 1, and so on.
There are pk~l such sets of p integers, each containing p - 1 integers relatively prime to p, for a total
of pk~l{p - 1) numbers less than pk and relatively prime to it. Hence <p(pk) = pk~l{p - 1). (Try this for
p - 3 and k = 2.)
23-26 We factor 6876 = 22 • 1719 = 22 • 32 • 191. Thus
<£(6876) = 6876 (l -\) (l - \) (l - Tj) = 6876 • \ • | • § = 2280.
23-27 With a = b = 2, we have 3! = 6 = 2 (mod 4). But for any higher a = b, the factorial {a2 - 1)!
contains a at least twice, so the factorial is congruent to 0. (Try it for a = 3 or a = 4.)
373. 24 divides a number if 8 and 3 divide it. We can factor p2 — 1 into (p — 1 )(p +1). Since p is a prime
greater than 3, p is odd. Thus either p = 1 (mod 4) or p = 3 (mod 4). In each case one of p + 1 and p — 1
is congruent to 0 (mod 4) and one is congruent to 2. The one which is congruent to 0 is divisible by
4, and the other is even; thus 8 divides p2 — 1. To show 3 divides p2 — 1 we use a similar argument.
Since p is a prime greater than 3, p is not divisible by 3, so either p = 1 (mod 3) or p = 2 (mod 3). In
either case one of p - 1 and p + 1 is congruent to 0 (mod 3), and is thus divisible by 3. Thus the
product (p - l)(p + 1) is divisible by 3.
374. Since n — 4 is divisible by 5, n = 4 (mod 5). Thus n2 = 16 = 1 (mod 5) and n4 = 1 (mod 5).
Hence n2 -1 = 0, n2 - 4 = -3 = 2, n2 -16 = -15 = 0, rz + 4 = 8 = 3, and n4 -1 = 0. All those congruent
to 0 are divisible by 5, so the first, third, and fifth are divisible.
375. For some integers q, we can write 1059 = q\d + r, 1417 = q2d + r, and 2312 = q3d + r. We thus
have 1417 - 1059 = 358 = (q2 - q\)d and 2312 - 1417 = 895 = (q3 - q2)d. This means that d divides
both 358 and 895. Factored, 358 = 2 • 179 and 895 = 5 • 179. Since d divides both, it must equal 179.
Dividing 179 into 1059, we find r - 164 as the remainder. Thus d - r = 179 - 164 - 15.
376. If 7 divides x2 + 15x +1, we have x2 + 15x + l = 0(mod7), orx2+x + l = 0(mod7). Substituting
in 0,1,2,..., 6 for x to find solutions, we discover that x = 2 and x = 4 are the solutions. The solutions
less than or equal to 100 are thus
2, 4, 9, 11, 16, 18, ..., 93, 95, 100,
the sum of which can be divided into the two arithmetic series 2 + 9 + 16 + • • • + 100 = y (102) =
(15)(51) = 765 and 4 + 11 + 18 H-f 95 = y (99) = (7)(99) = 693. The answer is the sum of these two,
or 1458.
377. Since n - 2 gives 32 - 2 = 30, this is the largest number which must divide n5 -n for any n.
To see that 30 does divide n5 - n for all n, we realize that 30 = 2 • 3 • 5, so we have only to show that
2, 3, and 5 divide.
So consider the equation successively in mods 2,3, and 5. In mod 2, n = 0orl,son5-n = 05-0 = 0
or l5 -1 =0. Since n5 - n = 0 (mod 2) no matter what n is, 2 always divides. For n - 3, n is congruent
to either 0, 1, or 2; testing each of these gives the same result, so 3 always divides. For 5, Fermat's
Theorem guarantees that n5 = n (mod 5) for any n; do you see why? Alternatively, we can test 0,1,
2, 3, and 4 (or, easier, -2, -1, 0,1, and 2) to see that each yields a result congruent to 0 (mod 5). Since
2, 3, and 5 divide n5 - n for all n, 30 does as well. Since this is also the largest possible candidate, 30
is the answer.
378. The units digits of powers of 7 go 7, 9, 3, 1, 7, 9, 3, 1, ..., repeating in sets of 4. Thus
the units digits of 7(/7^ is equal to the units digit of 7t(7?(mod 4)1. To evaluate 77(mod4), we write
77 = 37 = (-1)7 = -1 = 3 (mod 4). Hence the units digit of 7^ is equal to the units digit of 73, or 3.
379. Consider the elements of S (mod 7); then our criterion is that S can't contain any two
elements whose sum is congruent to 0 (mod 7). We cannot take both an element congruent to 1 and
an element congruent to 6; we must choose one or the other. The same holds true for the pairs (2,5)
and (3,4). Since there are 8 elements congruent to 1 (1, 8, 15, ..., 50) and only 7 for the others, we
take all the elements congruent to 1, 2, and 3, for 22 elements. What about elements congruent to 0?
We can take one, but taking two would yield a sum congruent to 0. Thus to the other 22 we can add
one element congruent to 0, for 23 total elements.
380. f(x) is congruent to 0 when divided by 6 if x2 + 3x + 2 = 0 (mod 6). Testing 0,1,..., 5, we get
the remainders 2, 0, 0, 2, 0, 0. We thus have four solutions between 0 and 5, four between 6 and 11,
four between 12 and 17, four between 18 and 23, then one more, at 25. The total is 17 elements.
381. There aren't any. Since all the numbers 2, 3,..., n divide n\, we can see that 2 divides n\ + 2,
3 divides n\ + 3 and so on. Each number greater than n\ + 1 and less than n\ + n has a divisor other
than 1 or itself, so there are 0 prime numbers in this range.
382. Finding the last three digits of 9105 is equivalent to evaluating 9105 (mod 1000). By the
Binomial Theorem, we have
9105
(10 -1)105 =
Since 10* = 0 (mod 1000) for k > 3, only the last three terms of the expansion matter (mod 1000). These
last three terms are - P®5) 100 = -(105)(104)(100)/2, which is divisible by 1000 and thus congruent to
0, (f)l0 = 1050 = 50, and - (1q5) = -1. Thus the last two digits are found by looking at 50 - 1 = 49,
so are 049.
383. The product telescopes: the 3 in 3 cancels with the 3 in the 4 in | cancels with the 4 in |,
and so on. What's left is yj(n + 1 ){n + 2)/2. Since n + 1 and n + 2 are consecutive integers, they are
relatively prime, so the only way (n + 1 ){n + 2)/2 can be a perfect square is if either (n + l)/2 and
n + 2 are perfect squares or n + 1 and (n + 2)/2 are perfect squares. In either case we seek two perfect
squares x and y such that lx and y are consecutive integers. Writing down the first few squares, 1, 4,
9, we see that 4 and 9 are just such squares. We thus let (n +1)/2 = 4 and n + 2 = 9, so n = 7. (Another
pair of squares, 25 and 49, yields n - 48 as the next largest solution. The equation x2 = 2y2 ± 1 is the
Pell equation, so we could find infinitely many solutions n.)
384. Since Adam's schedule repeats every 4 days and Ben's every 10, the overall schedule repeats
every 20 days. In this 20-day period, Ben rests on days 8, 9, 10, 18,19, and 20, while Adam rests on
days 4, 8,12, 16, 20. In each 20-day period they share 2 rest days. In the first 1000 days, or 50 even
20-day periods, they thus share 100 rest days.
385. Looking at the powers of 7 (mod 100) to get the last two digits, we have 71 = 07, 72 = 49,
73 = 343 = 43, 74 = (43)(7) = 301 - 01, and 75 = (01)(7) = 07. Thus the last two digits repeat in cycles
of 4, so 7"" = 7"" <mod 4> = 73 = 43 (mod 100).
386. Two consecutive integers n and n -1-1 have sum 2n + 1; setting this equal to 1000, we get
n = 999/2, which is not an integer. Three consecutive integers n, n + 1, and n + 2 have sum 3n + 3;
3n + 3 - 1000 yields n = 997/3, which is again not an integer. For n = 4 we have 4n + 6 = 1000, or
n = 994/4 = 497/2. For n - 5 we get 5n + 10 = 1000, or n- 990/5 = 198—at last an integral solution.
Hence n - 5 is the smallest solution.
387. We can see that 17 divides 9x + 5y by noting that 9x + 5y = 17x + 17y - 4(2* + 3y). Each term
on the right is divisible by 17, so 9x + 5y is.
388. The squares (mod 10) can easily be verified to be 0,1, 4, 5, 6, 9, and 0, so that d must be one
of these digits. Moreover, in (mod 4) any number is congruent to its last two digits, or 10 + d. But a
square must be congruent to 0 or 1 (mod 4), so that 10 + d must be 12, 13, 16, or 17. Since d cannot
be 2, 3, or 7, d must be 6. Thus our proposed square ends with the digits 19916. Since any number
is congruent to its last four digits (mod 16), our number is congruent to 9916 = 12 (mod 16). But all
even squares are congruent to 0 or 4 (mod 16), so this cannot be a square. [We came across using
(mod 16) only after finding that (mod 8) didn't give any new information; always feel free to play
around. To see that all even squares are congruent to 0 or 4 (mod 16), note that all even numbers are
of either the form 4k or the form 4k + 2. The square in the first case is 16k2 = 0 (mod 16), while the
square in the second case is 16k2 + 16k + 4 = 4 (mod 16). All in all, a pretty slick solution.]
.
389 The simplest way to do it is to look at the fifth power of each digit (mod 10). We have 05 = 0,
l5 = 1, 25 = 32 = 2, 35 = 243 = 3, 45 = 1024 = 4, 55 = 5, 65 = (-4)5 = -4 = 6, 75 = (-3)5 = -3 = 7,
85 = (-2)5 = -2 = 8, and 95 = (-1)5 = -1 = 9. No matter what a number q is, it will be congruent to
one of these digits (mod 10), so q5 = q (mod 10), and 10 divides q5 - q.
.
390 We use the result of the previous problem—that q5 = q (mod 10) for all q. As long as q is not
divisible by 5, we can divide a q from both sides of this to get q4 = 1 (mod 5). (Note that we have to
divide the modulus by 2, since q might be even.) Thus, if neither x nor y is divisible by 5, we have
x4 = 1 (mod 5) and 4 y4 = 4(1) = 4 (mod 5) and x4 + 4 y4 =l+4=5=0 (mod 5).
.
391 If x is the multiple of 5, then x4 + 4y4 = 0 + 4 = 4 (mod 5), so it is not divisible by 5. If y is the
multiple of 5, then x4 + 4y4 = 1 + 0 = 1, so it is not divisible by 5.
392. Every odd prime can be written as either 6n + 1 or 6n + 5 (not 6n + 3 because it's divisible by
3). For the former, 6n + 1 + 2 = 3(2n + 1), so p + 2 is not composite in this case. Hence, p = 6n + 5 and
p + 1 = 6n + 6 is divisible by 6 as desired.
393. Let n = lOx + y where x and y are integers and o < y < 9. Hence, n2 = 100x2 + 20xy + y2, so
the tens digit of n2 is odd if and only if the tens digit of y2 is. This only occurs if y = 4 or y - 6. In
both cases, the units digit of n2 is 6.
.
394 Since a\ = 1 is not prime, U2 = 1001 is not prime, and an is clearly divisible by 3 if n is, so we
need only consider n > 3 such that n is not divisible by 3.
We have an = 1 + 103 + 106 + • • • + 103^-1^ = = 0° ~h(^+io +1)> since 10” - 1 is divisible

by 9, we need only show that 102” + 10” + 1 is divisible by 111 to prove our assertion. We first
note that 103 = 1000 = 1 (mod 111), so 103m = 1 (mod 111), 103m+1 = 103m • 10 = 10 (mod 111), and
103w+2 = 100 (mod 111) for any positive integer m.
Since n is not divisible by 3, we consider the two cases n - 3k + 1 and n - 3k + 2. In the first
cast we have 10” = 103k+l = 10 (mod 111) and 102” = 103^+2 = 100 (mod 111). Thus in this case
102” + 10” + 1 = 100 + 10 + 1 = 0 (mod 111). In the second case we similarly get 10” = 100 and
102” = 10 (mod 111), so 102” +10” + 1=0 (mod 111) in this case as well. Since 9 always divides 10” -1
and 111 always divides 102" + 10” + 1, the fraction an = (10” - 1)(102” + 10” + l)/999 is an integer for
n> 1. Furthermore, since 102” + 10” + 1 > 10” - 1 > 999 for n > 3, the above factorization shows that
all an with n > 3 can be expressed as the product of two integers larger than 1; hence, an is not prime.
395. We need to prove both the "if" and the "only if" parts. For the "only if," we start with
p\(a2-a-3). Letb = 3a-l; thenb2-b+25 = (3a-l)2-(3a-l)+25 = 9az-9a+27 = 9(a2-a+3) = 0(modp),
where the last equivalence follows from the assumption that a2 - a+ 3 = 0 (mod p).
To prove the "if," we start with p\(b2 - b + 25). We need two cases. If p ^ 3, then there exists
some r such that 3r = 1 (modp). We let a = r(b + 1); then a2 - a + 3 = r2(b + l)2 — r(b + 1) + 3 =
r2(b + l)2 - (3r)r(b + 1) + 3(3r)2 = r2[(b + l)2 -3(b + 1) + 27] = r2(b2 -b + 25) = 0(modp).
If p = 3, we just let b = 2 and a = 1; then a2 - a + 3 = 3, which 3 divides, and b2 - b + 25 = 27,
which 3 divides.
We have thus proven that if p\(a2 -a+ 3) for some a, then p\(b2 -b + 25) for some b.
396. We set yk = xkxk+ixk+2xk+2l/ with the modification that the k can wrap around as in yy—i =
Xj-\XjX\X2. Each y\ is equal to either 1 or -1 and y\ + y2 H-f- Jh =0. Hence n is even: n = 2m. Also,
yil/2 * * * y« - (-1 )m; but each Xj appears four times in this product, so the product is +1. Thus m is
also even: m - 2p, so n = 4p is divisible by 4.
397. First d(n). Let m = p^1 ■ ■ ■ pekk and n - q{' • • • eft be the factorizations of the relatively prime
integers m and n. Since m and n are relatively prime, all the q's are different from all the p's, so the
factorization of the product mn is pe^ • • • pekft\ "'ft- We thus have
d{m) = (e1 + l)(e2 + l)---(ek + l)

d{n) = (/1 + l)(/2 + 1) - - • (/fc + 1)
d(mn) = (ei + l)(e2 + l)---(ek + l)(/i + l)(f2 + 1 )---(fk + 1),
so clearly d{mn) = d(m)d(n) for relatively prime m and n.

For s(n) we do nearly the same thing. With m, n, and mn as above, we have
s(m) = (l+pi+p? + - ■• + Pi)-'(X+Pk + vl + '
s(n) = (1 + e}i + q\ + ■ ■" + ft\)■ ■ ■ (i+ dk + qk + •• • + 9t)

s(mn) = (1 + pi + p\ + • • • + p^1) ■ ■ ■ (i + pk + p\ + • - + P?)
X (1 + q\ + q\ -+-f q{') • • • (1 + qk + (ft 4-1- ft)
so s(mn) = s(m)s(n) for relatively prime m and n.

.
398 Rather than use long division on the polynomials, we note that if the polynomial f(x) divides
g(x), then the integer f(p) divides g(p) for any fixed p. Thus, substituting p = 0 tells us that ra|30;
substituting p = 1 yields (m + 3)|66. Just these first two yield, upon testing divisors of 30, either m = 3
or m - 30. p - -1 gives (m - 3)|0, which yields no information, but p = 2 gives (p + 12)|5280, which
fails for p = 30. Thus 3 is the only remaining candidate.
399. If a < b and b > 2, 2b - 1 is greater than 2a + 1, so cannot divide it. If a = b, 2b - 1 clearly
cannot divide 2a + 1 because their difference is only 2. For the case where a > b, we write a - bq + r
for integers q > 1 and 0 < r < b. We then write
2a + 1 _ 2^+r + 1 _ or2^ - 1 2r + 1
2b 1 ~ 2b 1 “ 2 2b 1
- - - + 2b 1
- '
Now (2fc - 1)1(2^ - 1) for any q and b (do you see why?), so the first term is an integer. The second
term, on the other hand, is never an integer, because 2r + 1 < 2b - 1. Thus the sum is not an integer,
so 2^ — 1 cannot divide 2a + 1.
400. It suffices to show that at least one of 2d -1,5d-1, and 13d -1 is not a square. Assume for the
sake of contradiction that all three are squares; then we have 2d - x2 + 1, 5d = y2 +1, and 13d = z2 +1
for some x, y, and z. Clearly x is odd; then x2 = 1 (mod 8) and 2d = x2 + 1 = 2 (mod 8). Thus d is odd.
This forces y and z to be even: y = 2u, z = 2v. Since z2 - y2 = 8d, we have (v - u)(v + u) = 2d. But
v — u and v + u have the same parity, so they must both be even for their product to be even. Since
both are even, their product is moreover divisible by 4. But that product is also equal to 2d, for d an
odd integer. Thus the product cannot be divisible by 4, so we have a contradiction.
.
401 Let p be any prime; we show that p divides the numerator at least as much as it divides the
denominator. The largest a for which pa divides the denominator, n\, is [n/p\ + [n/p2\ + • • •. (This is
fairly clear based on our discussion of factorials in Volume 1; read back over that if you need to.) If p
divides b, then p can be cancelled from the quotient. If p does not divide b, then it must divide one of
the factors a,a + b,... ,a + (p-1 )b. (Why?) Altogether there are at least [n/p] factors in the numerator
divisible by p (one for each block of p consecutive factors); similarly there are at least [n/p2] factors
divisible by p2, and so on for higher powers of p. Altogether the product a(a + b)(a + 2b) ■ • • (a + (n - l)b)
has p as a factor at least a times. Since any prime divides the numerator at least as many times at it
divides the denominator, the fraction is an integer.
402. We write n as the sum of consecutive integers as
(2k + /)(/ + 1)
n - k + (k + 1) + • • • + (k + /)
2
If l is odd, 2k + l is odd. If Z is even, Z + 1 is odd. Hence, any number which is the sum of consecutive
integers has an odd factor and thus cannot be a power of 2. If n is not a power of 2 we can write n
as 2x(2y + 1). If y < 2X, then we can write n as
n = (2x-y) + (2x-y+ !) + ■■■+ (2X+ y-l) + (2x + y).
If y > 2X, we can write

n = (y-2* + l) + (y-2* + 2) + --- + (2* + y).
Chapter 24
Diophantine Equations
24-1 The solutions are given by (x, y) - (4k, 3k) for any k.
24-2 The graph is shown at right, with the solutions highlighted in

black. The solutions are obtained by starting at (0,0) and moving right 3,
down 2, left 3, up 2, and so on.
24-3 Starting with (0,0), we have (3,2), (6,4), (9,6),-With b = 6,
a = 4, the greatest common divisor is g = 2, yielding the solutions
(3k,2k) = (-3,-2), (0,0), (3,2),....
24-4 Dividing out a common factor of 2, the given equation becomes
2x - -3y. Thus the graph is the same as the graph we drew above. Rather
than getting the solutions by moving right 6 and down 4, we get them by
going right 3 and down 2 as before. Thus the common factor of 2 does not change the solutions; we
just divide it out before we start.
24-5 We have 5(2) + 3(-l) = 7, so (2, -1) is a solution.
24-6 Substituting in, we have a(r) + b((c - ar)/b) = ar + c - ar - c, so (r, (c - ar)/b) is a solution.
(You might wonder why we need to specify that r is a solution of ar = c (mod b). This is necessary to
make (c - ar)/b an integer. Do you see how?)
24-7
i. Dividing out the common factor of 2, the equation becomes 3x + 2y = 2. To find a specific
solution, we solve 3x = 2 = 0(mod2). One solution is x = 0, yielding (x,y) = (0,1) as a solution of
the equation. To get more solutions, we add a multiple -2k to x and a multiple 3k to y. This yields
(-2k, 3k + 1) as the general solution of the equation. Putting in various k, we can get some solutions
if we have a mind to: (-4,7), (-2,4), (2, -2).
ii. In this case, the left side is always divisible by 2 but the right side never is. Thus the equation
has no solutions.
24-8 No. The only positive y which doesn't make 3x+7y greater than 12 is y = 1, but substituting
y = 1 yields 3x = 5. There is no x which satisfies this equation.
< 154 >

24 9 We fill in the same grid as in the text, with x along the top and y down the side:
0 1 2 3
0 0 m 2m 3m
1 n n+m n + 2m n + 3m
2 2n 2n + m 2 n + 2m 2 n + 3m
3 3n 3n + m 3 n + 2m 3 n + 3m
All c = 0 (mod m) with c > 0 can be obtained from the first row; c = n (mod m) with c > n can be
obtained from the second row; and so on until all c = (m - l)n (mod m) with c > (m - l)n can be
obtained from the last row.
We can thus get all c > n(m - 1); the largest c we can't get is the largest c = n(m - 1) which is still
less than n(m -1). (Compare to the argument in the text if following all these letters bogs you down.)
The largest c congruent to n(m - 1) (mod m) which is less than n(m - 1) is n(m -1) - m = mn -m-n,
as desired.
Since this argument is quite abstract, we strongly recommend that you go through it for, say,
3x + 7y = c. What is the largest c we can't get? Get a feel for why we can't get this c, and why this is
the largest such c.
24-10 We have x = yjz2 - y2 = yJ(r2 + s2)2 - (r2 - s2)2 = yj(2r2)(2s2) = 2rs.

24-11 If r and s have a common factor d, then d divides 2rs, r2 - sz, and r2 + s2. But it was
specified that the triangle was primitive, meaning that x, y, and z can have no common divisors, so
d cannot exist. Thus r and s must be relatively prime.
24-12 For (3,4,5), we have 2rs = 4, so that rs = 2, implying that r - 2 and s = 1.
For (6,8,10), we similarly have rs = 6/2 = 3 or 8/2 = 4. Since r + s, the possibilities are r - 4,
s = 1 and r = 3, s = 1. The second pair produces the correct triangle.
For (5,12,13), we have rs = 12/2 = 6, so that either r - 6, s = 1 or r - 3, s = 2. The latter gives the
correct triangle.
24-13 Clearly r < 10, since otherwise z = r2 + s2 = 100 + s2 is greater than 100. Moreover, one of
r and s must be even, or else all three sides of the triangle will be even. We further need r and s to be
relatively prime, and r > s. We find the pairs (2,1), (4,1), (6,1), (8,1), (3,2), (5,2), (7,2), (9,2), (4,3),
(8,3), (5,4), (7,4), (9,4), (6,5), (8,5), (7,6), (8,7), and (9,8). Not all these are OK, though, since some
of them make z > 100. In particular, (8,7) and (9,8) make r2 + s2 > 100. We are left with 16 pairs (r, s),
and thus 16 primitive Pythagorean triples.
24-14 Not both r and s can be odd, since otherwise 2rs, r2 - s2, and r2 + s2 are all even, whereas we
know that the three are relatively prime. Both r and s cannot be even for the same reason. Moreover,
we cannot have r even and s odd, since then we have y\ = r2 - s2 = -1 = 3 (mod 4), which is not
possible. The only remaining possibility is that s is even and r odd.
24-15 Since r and s are relatively prime and t - s/2, r and t must clearly be relatively prime.
24-16 We already established that r and t are relatively prime, so their square roots, rx and t\,
must certainly be relatively prime as well. Moreover, we know that r is odd, so that its square root
n is odd. (Of course this exercise is simple, but it makes sure that you follow what's going on.)
24-17 We know that y\ = r2 - s2; substituting in several definitions, we get y\ - (rj)2 - (21)2 =
(rf)2 - (2f2)2. This can be rearranged to get the desired equation.
156 > CHAPTER 24. DIOPHANTINE EQUATIONS
24-18 If some common factor divided both 2t2 and t/q, then it would also divide V\ and t, and
hence r and s. But r and s are relatively prime, so such a common factor cannot exist. Thus 2t2 and
y\ are relatively prime.
24-19 Because zq = r2 + s2 = r\ + s2 > r\ > r\.
24-20 You might have found the solutions (x,y) = (1,1), (3,2), (7,5), for example. (Make sure
you see why these are solutions.)
24-21 In this case we can factor x2 — E2y2 = 1 as (x - Ey)(x + Ey) — 1. Since x — Ey and x + Ey
are integers, the only way this can be is if x — Ey = x + Ey = ±1. Thus y = 0, x = ±1 are the only two
solutions, both of which can be considered trivial.
24-22 Using the continued fraction
1
V2 = 1 +
2+
with period 1, we find the convergents |, j=, and the corresponding Pell solutions (1,1), (3,2),
(7,5), and (17,12). The first corresponds to the - sign, the second to +, the third to -, and the fourth
to +.
We can do the others similarly; the only tough part is finding the continued fraction expansion.
Using the usual method for continued fractions, we find
V3 = 1 + (period 2),
1+
2+-
1+ 2+-
V5 = 2 + (period 1),
4+ 4+ 4+~
and
V6 = 2 + (period 2).
2+ 4+-
2+ 4+..
Since V5's continued fraction has period 1, every convergent will be a solution. We find (2,1),
corresponding to -, (9,4), corresponding to +, and (38,17), corresponding to -.
For V3 and V6, we want every second convergent, because the continued fractions are period 2.
(What do we get for the other convergents?) For a/3 we get the solutions (2,1), corresponding to +,
and (7,4), corresponding to +. For V6 we get (5,2), corresponding to +, and (49,20), corresponding
to +.
24-23 We proceed by induction. For n - 1, we get a - xq, b = yo, which is a solution by
assumption. For any solution (m, n), multiplying (m + Vu n) by (xq + Vd yo) yields
(mxo + Dnyo) + (nxo + myo) Vd,
for the pair (mxo + Dnyo, nxo + myo). Testing this pair as a solution, we have
(mxo + Dny0)2 ~ D(nx0 + myo)2 = m2(xl - Dy0) - Dn2(x% - Dy§) = (±1 )(m2 - Dn2) = ±1,
as desired.
This allows us to make the inductive step. Assume that (xo + VD yo)k yields a solution; we have
shown that multiplying this by xo + VD t/q will yield another solution. Thus if (xo + y[Dyo)k yields a
solution, (xo + VD yo)k+l also yields a solution. By induction (xo + VD yo)n yields a solution for any n.
24-24 The squares x2 and y2 are both congruent to 0 or 1 (mod 4), so their sum is congruent to
0,1, or 2. Since 100000003 is congruent to 3 (mod 4), it cannot be the sum of the two squares.
403. To solve this Pell equation, we need the continued fraction expansion of V8. We can get this
without computation by recalling that
1
V2 = 1 + l '
2+ 2+-
so that
V8 = 2V2 = 2 + -2 + = 2 +
2+ 1+ 1 +
2+ — 4+- 4+-
2+ 1+
2+— 4+-
(Of course, we could find this continued fraction in the usual way, but this method is a little easier.)
Since the period of the continued fraction is 2, we get the first solution to our equation from the
second convergent: 2 + \ = y. Testing (x, y) = (3,1), we find that it is indeed a solution.
We can find more solutions without the continued fraction by taking powers of 3 + V8. The
square is 17 + 6 V8, yielding (x, y) = (17,6) as the next solution. (Test it; does it work?)
404. We seek the number of solutions to 4x + 3y = 1776 such that x and y are nonnegative integers.
Letting y = 0, we find the solution (444,0); succeeding solutions can be found by decreasing x by
3 and increasing y by 4. The next few solutions are thus (441,4), (438,8), and so on. This process
continues until we reach (0,592). How many solutions does this entail? The arithmetic sequence
444, 441,..., 0 has 444/3 + 1 = 149 terms.
405. We can rearrange the given equation into n — m(n - 1), or m - n/(n — 1). For m to be an
integer, we must have n = 0 or n = 2 (do you see why?). Thus we have the 2 solutions (m, n) = (2,2)
and (0,0).
406. Clearly the right side of the equation is congruent to 1 (mod 3). The left side factors into
m(m + 5)(m + 1). One of m, m + 1, and m + 5 must be divisible by 3 (why?), so the left side of the
equation is always congruent to 0 (mod 3). Hence the equation has no solutions.
407. We factor 1984 as 26 • 31 to see that V1984 = 8 V3l. Since x and y are integers and
yfc + yfy = 8 V3l, we must have yfx = a V3T and yfy = b V3L Furthermore, we must have a + b = 8
and b > a (since the problem stipulates that y > x). The only pairs {a, b) are (1,7), (2,6), and (3,5), so
there are 3 solutions.
408. Assume for the sake of contradiction that 3q can be expressed as a sum of squares, and let
q = al + b\ and 3q = (% + <%. Further suppose that q is the smallest positive integer with the desired
property. All squares are congruent to either 0 or 1 (mod 3), so = 3q = 0 (mod 3) implies
that cl = dl = 0, which in turn implies that c0 and d0 are divisible by 3. We thus write c0 = 3ci and
do = 3d\. Then 3q = 9(c2 + d\), so q = 3(c2 + d2) is divisible by 3. Writing q = 3q\, we have qi = c2 + d2
and 3q\ — q — Aq + &q. In we have thus constructed a smaller positive integer with the desired
property which is a contradiction to our assumption that q is the smallest. Hence there is no such q.
409. First of all, we use the fact that a square is always congruent to 0 or 1 (mod 4). In order, the
given pairs are congruent to 3, 1, 1, 3, and 1, so 07 and 63 are immediately out. Furthermore, any
square which ends in 5 must end in 25: if the nonsquared number was congruent to 10x+5 (mod 100),
then the square is congruent to 100x2 + lOOx + 25 = 25 (mod 100). Thus 85 is out. The remaining
two, 29 and 41, can be the last two digits of a square: 272 = 729 and 212 = 441. Thus 29 and 41 are
answers.
410. Any Pythagorean triangle is a multiple of a primitive triangle, with sides 2rs, r2-s2, and r2+s2.
If the sides of our triangle are 2 Os, C(r2—s2), and C(r2+s2), then the area is C2rs(r2—s2) = C2rs(r+s)(r—s)
and the perimeter is 2Crs + 2Cr2 = 2Cr(r + s). Setting the area equal to the perimeter, we have
C2rs(r + s)(r - s) - 2Cr(r + s), or, cancelling Cr(r + s) from both sides, Cs(r - s) = 2. The possibilities
are C = 2, s = 1, r - s - 1; C = 1, s = 2, r - s = 1; and C = 1, s = 1, r - s = 2. The first case
gives (C, r,s) = (2,2,1), so that the triangle we obtain has sides 2(2)(2)(1) = 8, 2(22 - l2) = 6, and
2(22 + l2) = 10. The second case gives (C,r,s) = (1,2,3), yielding the sides 2(2)(3) = 12, 32 - 22 = 5,
and 32 + 22 = 13. The third case gives (C,r,s) = (1,1,3), yielding the sides 2(1)(3) = 6, 32 - l2 = 8,
and 32 + l2 = 10—this is the same triangle as we got in the first case. Thus there are only 2 solutions:
(5,12,13) and (6,8,10).
411. Since 3m, 3n, and 1 are all odd, their sum, 3m + 3n + 1, is odd as well. Thus we write
3m + 3n + 1 - k2, for k some odd integer. This yields 3m + 3n = (k2 - 1) - (k - 1 )(k + 1). The right
hand side of this equation is always divisible by 8 (since k is odd, it is between a multiple of 4 and a
multiple of 2). However, powers of 3 are always congruent to either 1 or 3 (mod 8), so the left side
of the equation is congruent to 1 + 1 = 2, 1 + 3 = 4, or 3 + 3 = 6 (mod 8). Hence the left side of the
equation can never be divisible by 8, while the right hand side always is, so 3m + 3” + 1 can never be
a perfect square.
412. We could square the numbers, add them, and take the square root, but that wouldn't be
much fun. Instead we divide out the common factors until the legs are relatively prime. Dividing
out 2, the legs are 204 and 10403, and these are relatively prime. We thus set 2rs = 204, so that
rs = 102 = 2-3-17. The only way to get r2 - s2 large enough is to take r = 102, s = 1, whence
r2 - s2 = 10404 - 1 = 10403. The hypotenuse of the primitive triangle is r2 + s2 = 10405, and the
hypotenuse of the original triangle is twice this, or 20810.
413. For x > 5,x\ has units digit 0. Since 1! = 1, 2! = 2, 3! = 6, and 4! = 24, 1! + 2! + 3! + • • • + x\
thus has units digit 1 + 2 + 6 +24+ 0 + 0l-= 33 —> 3 for x > 4. We wish for this to equal a perfect
square; but a perfect square cannot have units digit 3. Thus the only possible solutions are for x < 3.
In fact, we find the only two solutions, (1,1) and (3,3), by testing these values of x.
414. Only one of the two legs of a primitive Pythagorean triangle is even, the one of the form 2rs.
Not both of r and s can be odd, because r2 - s2 would also be even if r and s were both odd. Since
one of r or s is even, 2rs is divisible by 4, so cannot equal 90. Thus no primitive Pythagorean triangle
can exist with one leg 90. (Can you find a non-primitive Pythagorean triangle with one leg 90?)
415. If the sides are 2rs, r2-s2, and r2+s2, then the area is rs(r2-s2) = rs(r+s)(r-s) and the perimeter
is 2rs + 2r2 - 2r(r + s). Setting the area equal to twice the perimeter, we have rs(r + s)(r - s) = 4r(r + s),
or, cancelling r(r + s) from both sides, s(r - s) = 4. Since s and r - s are integers, the only possibilities
are s = 1, r -s = 4; s = 2, r-s = 2; and s = 4, r-s = 1. The first case is not primitive, since both r and
s are odd then. In the second case, we have s = 2, r = 4, so again the triangle is not primitive. In the
third case, s = 4 and r = 5, so the sides of the triangle are 2(4)(5) = 40, 52 - 42 = 9, and 52 + 42 = 41.
The only triangle with the desired property is (9,40,41).
416. Every Pythagorean triangle is some multiple of a primitive triangle, so we only need to
prove it for primitive triangles. We consider r and s (mod 5). If either is congruent to 0, then we
are done because 2rs is congruent to 0. If the two are congruent to one another, then we are done
because r2 — s2 is congruent to 0. If r = -s (mod 5), then we are done because r2 — s2 is congruent to
0. What's left? Only the pairs (r,s) = (1,2), (3,4), (1,3), and (2,4). Testing these shows that r2 + s2 is
congruent to 0 (mod 5) for each.
417. In (mod 5), the possible fourth powers are 04 = 0, l4 = 1, 24 = 16 = 1, 34 = (-2)4 = 1, and
44 = (-1)4 = 1—that is, only 0 and 1. So consider the five numbers a through e. If e4 is congruent to
0 (mod 5), then a4 through d4 must all be congruent to 0 (mod 5). If e4 is congruent to 1, then one
of a4 through d4 must be congruent to 1 and the other three to 0. In either case, at least three of the
numbers are divisible by 5.
418. Since na+nb = nc, a and k must be less thane. We thus divide through by a to get l+nb~a = nc~a.
Looking at this equation (mod n), we must have k - a — 0, since otherwise nb~a and nc~a would both
be divisible by n and we would have 0 = 1 (mod n). With b - a = 0, we get 2 = nc~a, which forces
n = 2, c = a + 1. The expression an + bn - cn becomes a2 + a2 - (a + l)2, or a2 - 2a - 1. The largest this
can be with a < 11 is for a = 10, which yields 100 - 20 - 1 = 79.
419. Since 14x2 and 71990 are divisible by 7,15y2 must be also. Hence we let y = 7yi and we have
14x2 +15 • 49y2 = 71990. Dividing by 7 gives us lx2 + 105y2 = 71989, so x is divisible by 7 as well. We let
x = 7xi and we find, after dividing by 7, that 14x2 + 15y2 = 71988. We can continue like this over and
over until we finally come to 14x2 + 15t/2 = 7° = 1. This equation clearly has no integer solutions, so
the original equation has no integer solutions.
420. Multiplying the equation by 5 and completing the square to get perfect squares, we obtain
(5m - 3n)2 + 26n2 - 9925. Taking the equation (mod 13) to eliminate the 26n2, we have (5m - 3n)2 =
6 (mod 13). But the squares mod 13 are 0,1, 4, 9, 3,12, and 10; since 6 is not a square, there can be no
such m and n.
421. The even leg of such a primitive Pythagorean triple is 2rs — 28, or rs = 14. We thus have
r - 7 and s = 2orr=14 and s = 1. In the first case, the hypotenuse is 72 + 22 = 53; in the second
case, the hypotenuse is 142 + l2 = 197. The sum of these two possible values is 250.
422. Suppose there are such a and b; then 25a2 + 30ab + 35b2 = 5 • 1993, or (5a + 3b)2 + 26b2 = 5 • 1993.
Clearly 5a + 3b is odd, since 26b2 is even and 5 • 1993 is odd. We let 5a + 3b = 2k + 1, so that
5 • 1993 - 26b2 = (2k + l)2 = 4k2 + 4k + 1 = 8s + 1 for some s, yielding 4982 - 13k2 = 4s. Since 4982
and 4s are both even, then 13k2 must be even as well. Since k is thus even, k2 is divisible by 4. But
we now have a contradiction, because 13k2 and 4s are both divisible by 4 but 4982 is not. Thus there
are no such a and k.
423. Any three consecutive numbers are congruent to 0,1, and 2 (mod 3). Their squares are thus
congruent to 0,1, and 1 (mod 3), so their sum is congruent to 2 (mod 3). But all squares are congruent
to 0 or 1 (mod 3), so this sum cannot be a square.
424. Taking the equation (mod D), we have x2 = -1 (mod D). This only has a solution if -1 is a
quadratic residue.
425. Let (xq, yo, zq) be the solution of x3 + 3y3 = 9z3 with the smallest z. Since 3yl and 9z^ are both
divisible by 3, xjj must also be divisible by 3. Thus we let xo = 3xi, so that 27x3 + 3yg = 9zq. Since 27x3
and 9zg are both divisible by 9,3yg must be also, so that yo is divisible by 3. Letting yo = 3yi, we have
27x3 + 81y3 = 9zq. Since 27x3 and 81y^ are divisible by 27, 9zq must be also, so that zo is divisible by
3. Letting zo = 3zi, we have 27x3 + 81y3 = 243z3. Dividing out a factor of 27, we have x3 + 3y3 = 9z3.
But this is a solution of the original equation with z\ < Zo, contradicting our assertion that (xo, yo/ô)
had the smallest z value. Thus we have a contradiction, so the equation has no solutions.
426. If all three of x, y, and z are odd, then all seven terms of x3 + y3 + z3 + x2y + y2z + z2x + xyz
are odd, so the sum itself is odd. Since 0 is even, the equation cannot be satisfied in this case.
If exactly two of x, y and z are odd, then three terms of x3 + y3 + z3 + x2y + y2z + z2x + xyz are
odd, so the sum itself is odd. (For example, if x and y are odd then x3, y3, and x2y are odd.) Again,
the equation cannot be solved in this case.
If exactly one of x, y, and z is odd, then one term of x3 + y3 + z3 + x2y + y2z + z2x + xyz is odd, so
again the sum is odd and the equation has no solution.
Thus the only way the equation can have a solution is if x, y, and z are all even. Letting x = 2xi,
y = 2yi, and z = 2zi, the equation becomes
8x3 + 8y3 + 8z3 + 8x2yi + 8y2zi + 8z2Xi + 8xiyiZi = 0.
Dividing out by 8, we find that (x\, y\,z{) is a solution to the original equation. But this means that
Xi, yi, and z\ are all even, so that x\ - 2x2, yi = 2y2, and Z\ - 2z2- But then (X2, y2/Zz) is a solution, so
all these are even, and so on. Since we can find a smaller solution for any given solution, there can
be no solution at all.
Chapter 25
Graph Theory
25-2 There are as many edges as there are pairs of vertices, which from combinatorics is (2) =
n(n -1)/2.
25-3 The vertices comprising the largest clique, of size 3, have been colored
black in the graph at right; the vertices of the largest independent set, size 5, are
squares. (The black square is in both.) Can you prove that there is no bigger
independent set in this graph?
25-4 We have tried at left; however, the last edge cannot be drawn, since it
would intersect one of the other edges we drew.
25-5 Some four of the vertices can always be formed into a square. The fifth
vertex may be either inside the square or outside it. If the fifth vertex is inside
the square, then the two edges connecting opposite vertices of the square must
both lie outside the square. This cannot be. Similarly, if the fifth vertex is outside
the square, the two edges connecting opposite vertices of the square must both lie inside the square,
another impossibility.
25-6 It has 4 faces, where we make sure to count the unbounded face.
25-7 If we add up the degrees of all the faces, we count each edge twice, so that D = 2E.
Substituting this into D > 3F, E > 3F/2 is immediate.
25-8 Some ways to add an edge look different, because the edge goes into or through the
unbounded face. But these are the same basic ways.
25-9 Just do it.
< 161 >

162 > CHAPTER 25. GRAPH THEORY
25-10 If we add up the degrees of all the vertices, we get a total of dV. In doing so, we have
counted each edge twice, so that 2E = dV, or E = dV/2.
25-11 Since there are / edges around each face, we can add up the numbers of edges around all
the faces to get fF. Since this counts each edge twice, we have fF = 2£, or F = 2E/f. Substituting in
E = dV/2 from the previous exercise, we thus have F = dV//, as desired.
25-12 From the inequality (d - 2)(/ - 2) < 4, we can write (d - 2)(/ - 2) = 1, (d - 2)(/ - 2) = 2, or
(d - 2)(/ — 2) = 3. In the first case, we must have d - 2 = / - 2 = 1, so that d = f = 3. In the second
case we can have d- 2 — 1, / — 2 = 2, so that d — 3 and / = 4, or d — 2 = 2, / — 2 = 1, so that d — 4 and
/ = 3. In the last case we can have d - 2 = 1, / - 2 = 3, so that d = 3 and / = 5, or d - 2 = 1, / - 2 = 3,
so that d = 5 and / = 3. Thus the five solution pairs are (3,3), (3,4), (4,3), (3,5), and (5,3).
25-13 Looking at the graph, this is fairly clear.
25-14 The pair (3,3) corresponds to the tetrahedron, (3,4) to the cube, (3,5) to the dodecahedron,
and (5,3) to the icosahedron, as you can see by counting edges around the faces and edges emerging
from the vertices of the corresponding solids. (For pictures of the five Platonic solids, see Volume 1.)
25-15
From left to right below, we have the path BCADE, the trail ABCADE, which is not a
path because it goes through A twice, and the walk CABCADE, which is not a trail because it passes
through edge CA twice.
C C c
25-16
There are n edges, since we take n steps by walking around the cycle, passing through
every edge once.
25-17 There are n - 1 edges, since each of the n vertices, except the very top one, has one edge
above it in the "levels" structure described in the text.
25-18 Try to think about the "why" before you read the explanation in the text.
25-19Since Q has leftover edges, we can move out from it on an unused edge. Since we have
passed through every vertex an even number of times, and every vertex started with even degree,
each vertex has an even number of unused edges. Thus the vertex we go to from Q must have an
unused edge besides the one we came in on, and similarly for each edge in the trail—if the path
ended on a vertex, that vertex would have to have odd degree. (This is exactly the same as our
original argument that the path has to end up at its starting point.) Thus the path can always keep
going on unused edges until it gets back to Q.
25-20 A graph with exactly two odd degree vertices will always have an Euler trail, and an
Euler trail of such a graph will always start and end on the odd-degree vertices. To see that the
graph has an Euler trail, imagine an edge between the two odd-degree vertices. Since adding this
edge increases the degree of each by 1, their degrees are then even. Since the entire graph is now
even-degree, we can use our previous method to draw an Euler trail which starts and ends at the
same place. Cutting the added edge back out, we have an Euler trail which starts on one odd-degree
vertex and ends on the other.
To see that any Euler trail of the graph must start and end on odd-degree vertices, recall that the
unused degree of a vertex goes down by 2 each time we pass through the vertex. Thus if the walk
did not start or end on one of the odd-degree vertices, the unused degree of that odd-degree edge
would go 2n + 1 —> 2n — 1 —» 2n — 3 —» • • • —> 1, and would never hit 0. Thus some edge connected
to that vertex would never be used in the walk.
25-21 Each vertex corresponds to the country it sits in.
25-22 Any two vertices which are connected together must have different colors. Thus if a
graph has any edges at all, it has y > 2! The only graphs with y = 1 are null graphs.
25-23
25-24 The normal way is easy. If you try to draw with no edges crossing, though, you will
find it impossible.
25-25 K.3'3 has E = 9 and V - 6. Thus 3V - 6 = 10, so E < 3V - 6 is indeed satisfied. However,
this does NOT mean X33 is planar. It just passes one test of being planar.
25-26 Ksj has E = st and V = s + t. To satisfy E < 2V - 4, we must thus have st < 2s + 2t - 4.
Rearranging and factoring, this yields (s - 2)(f - 2) < 0. This inequality is a necessary condition for
KS/t to be planar, but it cannot be satisied for s,t > 3.
25-27 Put the set of t vertices on a line, and put one of the set
of 2 vertices on each side of the line. It is clear that all the desired
connections can be drawn without crossing, as shown at right.
25-28 Since every face has at least g edges around it, we can count
the edges around all the faces to get a total of at least Fg. Since this sum counts every edge twice,
we have Fg < 2E, or F < 2E/g. Substituting this into Euler's formula, we have V - E + 2E/g > 2, or
E < (V - 2)/(l — 2/g). (Verify that we get familiar inequalities for g - 3 and g = 4.)
427. When we add up the degrees of all the vertices, each edge gets counted twice—once for each of
its endpoints. Thus the total sum is equal to twice the number of edges.
428. If we add up the degrees of all the vertices, we get a sum Vd, where d is the common degree.
This sum counts each edge twice, so Vd = 2E = 40. Thus our graph could have 2 vertices of degree
20, 4 vertices of degree 10, and so on. However, the degree of a vertex must be less than the total
number of vertices, so the only real possibilities are V = 40, d = 1; V = 20, d = 2; V = 10, d = 4; and
V = 8, d = 5.
164 > CHAPTER 25. GRAPH THEORY
Are all these graphs possible? The first is: just divide the 40 vertices into 20 pairs
and connect the pairs—then each vertex has degree 1 and there are a total of 20 edges.
The second combination is similarly attainable, with a cycle of length 20. How about
the third? We can construct a graph with 10 degree-four vertices as at upper right,
arranging the points in a circle and connecting every point to its two nearest neighbors
and its two next-nearest neighbors. We can construct a graph with 8 degree-five vertices
in the same way, except additionally connecting each point to the one directly opposite,
as at lower right. Hence all our candidates are possible graphs, and the answers are
40, 20,10, and 8.
429. A graph model is the easiest way to figure this out. Since each player is on exactly two
teams, we can let each player represent an edge on a graph where the teams are the vertices. Since
the problem tells us that there are seven teams (vertices) and each pair of teams shares a player
(edge), the graph in question is X7, which has 7 • 6/2 = 21 edges (players).
430. To get the most edges out of the fewest vertices, we let our graph be Kn for some n. It then
has n(n - l)/2 vertices. Since 10(9)/2 = 45 and 11(10)/2 = 55, our smallest graph has 11 vertices.
(How close does our graph come to being Xu?)
431. Here we cannot take Kn, because Kn is not planar for n > 5. Recalling the restriction that
E < 3V - 6 for planar graphs, we find V > 56/3, so the smallest possible V is 19. This is only
a candidate, though; we need to show that a planar graph with 19 vertices and 50 edges exists.
Experimentation shows that it does, so we're done. (Can you draw such a graph?)
432.
433. Let's assume, for the sake of contradiction, that all V vertices of some planar graph have
degree > 6. The sum of the degrees of all the vertices is equal to 2E, as we have seen repeatedly, so
2E > 6 V, or E > 3V. But we know that in any planar graph, E < 3V - 6, so we get 3V < 3V - 6, a
contradiction.
434. This looks like a job for Euler's formula. Since the sum of the degrees of the vertices is
4V, we have E = 2V. Then Euler's formula gives V-E + F- V- 2V + 10 = 2, so V = 8. Find a
generalization to the case of V vertices, each with degree k, and E faces.
435. Without loss of generality, suppose segment BC is red. We consider three cases.
Case I: At least three of the edges BAi are red. Let three red edges be BAr, BAS, and BAt. At least
one of the sides of the triangle ArAsAt, say AsAt, is a diagonal of the base, so is colored. If AsAt is
red, then triangle BAsAt is all red. Moreover, if CAS or CAt is red, then triangle BCAS or BCAt is all
red. Thus all of AsAt, CAS and CAt must be blue to avoid an all-red triangle; but this makes CAsAt
an all-blue triangle. Thus we must always have a unicolored triangle in this case.
Case II: Exactly two of the edges BAi are red. Let the two edges be As and Atr and consider two
subcases. If AsAt is a diagonal of the base, then we can reason exactly as in Case I above. Otherwise,
we can without loss of generality suppose that s = 1 and t = 2, which means that BAi is blue for
i t 1,2. Consider the three base vertices A3, A5, and A7. Since BA3, BA5/ and BA7 are all blue, the
diagonals A3A5, A5A7, and A7A3 must all be red to avoid an all-blue triangle with B. But these edges
all being red forces the all-red triangle A3A5A7. Thus we must always have a unicolored triangle in
this case.
Case III: Exactly one edge BAj is red. Let the red edge be BA\, so that BAj is blue for all i + 1.
We can repeat the argument of Case II on A3, A5, and Ay to show that there is always a unicolored
triangle in this case.
Since there is a unicolored triangle in every case, there must be a unicolored triangle no matter
what.
436. The Pigeonhole Principle tells us that the maximum answer is 499. If we played for more
than 499 days, each player would have 500 or more partners from the 499 other players, so would
have to play with some other player twice.
The tough part is to show that a 499 day tournament is possible. To do this, we
arrange 499 of the players around a circle with one player at the center. We can choose
any player on the circumference to play with the central player, then pair up the other
players using lines perpendicular to the line connecting this first pair. (The procedure
is shown as right using 10 players instead of 500.) By rotating the resulting figure,
we can get 499 different pairings so that no player plays with the same other player twice. Since we
have shown that we can get 499, and that nothing greater than 499 is possible, 499 is the answer.
Chapter 26
Parting Shots
437. Rewriting and factoring, we have
(lnx)3 - Inx4 = (lnx)J - 41nx = (lnx)(lnx - 2)(lnx + 2) = 0.
Hence lnx = 0, lnx = 2, or lnx = -2 and the solutions are 1, e2r and 1/e2.
438. A function has a well defined inverse if there is one and only one element x for which f(x) - y
for all y. This means that our function must assign a different element of A to each element of A.
Let A = [a\r... ,013}. If all the /(a*) are to be distinct, then there are 13 choices for f{a\), leaving 12
choices for f(a2), then 11 for f(a3), and so on, for a total of 13! different functions with a well-defined
inverse. Since in defining a function from A to A we have 13 choices for each /(fl;), there are 1313
functions from A to A and our desired fraction is 13!/1313 = 12I/1312.
439. Let there be s steers and c cows. Thus, 25s + 26c = 1000. Since 26c = 1000 - 25s = 25(40 - s)
and s and c are integers, c must be divisible by 25. Since c > 1 and c < 50 (since (26)(50) > 1000),
c = 25.
440. Since In b - In a = 1, In b/a = 1, so b/a - e. Similarly, we can show d = ec - e2b = e3a, so a, b, c,
d is a geometric sequence with ratio e.
441. Since 3x2 +9x + 17 = (3x2 + 9x + 7) + 10, our given expression is equal to
1 | 10
3x2 + 9 x + 7'
To maximize this, we must minimize the denominator of the second term. Completing the square
of the denominator yields 3x2 + 9x+ 7 = 3(x + 3/2)2 + 1/4. The minimum value of this is clearly 1 /4
(when x = -3/2). Thus, the maximum value of our given expression is 1 + 10/(1 /4) = 41.
B 442. The figure formed upon rotation is a pair of cones joined at the base. If
we draw altitude BD of the triangle, we see that in the solid formed by rotation,
BD is the radius of the cones. Segment DC is the altitude of the right hand
cone and AD is the altitude of the left cone. Thus, our problem is now finding
BD, AD, and DC. We find the altitude by determining [ABC] in two ways, as
(AC)(BD)/2 and by using Heron's formula. Heron yields [ABC] = 84 and from
< 166 >

this we find BD — 12. Since AB — 13 and BC — 15, AD — 5 and DC = 9. Hence, our desired volume is
(122)(5)h , (122)(9)tz
+ = 672tt.
443. Shown is the case where n = 4. We see that the 2n radii divide the
circle into 2n regions. As shown in our diagram, we can always draw a secant
that passes through n + 1 of these regions, cutting each of them into two regions
and thus forming n + 1 new regions in addition to the original 2n. Our answer
therefore is 3« + 1.
444. Writing
n+1
5 + 3V5 fl + V5\"+1 5-3V5 /l- V5\
f(n +1) =
10 V 2 / + 10 V 2 /
5 + 3 V5 (l + y/5\ 5 - 3 V5 (l - V5X
/(n-1) =
10 10
we have
n-l
as the desired expression. Evaluating the inner term in each of the above products, we find that
f(n + 1) - f(n - 1) = f(n).
445. If we write the point (x, y) in polar coordinates, s and c are sin 6 and cos 6, respectively. Since
sin2 6 - cos2 6 = cos 20, s2 - c2 can range from -1 to 1, inclusive, or -1 < s2 - c2 < 1.
446. Writing the number with just twos and threes as bases it is 2636 + 212 + 312. This problem
gets a bit tricky; recall from Volume 1 that x2 + xy + y2 = (x + y)2 - xy. Let's try this here, with x = 26
and y = 36:
212 + 2636 + 312 = 212 + 2(2636) + 312 - 2636
= (26 + 36)2 - 2636
= (26 + 36)2 - (2333)2
= (26 + 2333 + 36) (26 - 2333 + 36)
Thus, our numbers are 1009 and 577.

447. Squaring the second quantity and multiplying by the first, we have
J.2X /0„\2
r=, =64.
T V x
Thus, the possible values of z are the cube roots of 64, which, by DeMoivre's Theorem, are 4,
-2 + 2i V3, and -2 - 2i V3.
168 > CHAPTER 26. PARTING SHOTS
448. Note that z(l + x + xy) = z + xz + xyz = 1+z + xz. Similarly, we note that we can get a common
denominator of the three fractions by multiplying the first term top and bottom by z and the second
term top and bottom by xz, yielding:
z xz 1
- + - + - — 1.
Z + XZ + 1 XZ + l+Z 1+Z + ZJ
449. First we write each of P, S, and S' in closed form:
P = a(ar)(ar2) ■ ■ ■ (arn = anr{-n
a - arn 1 -r
S = a + ar + ar2 + ■•■ + arn 1 =-= a--, and
1-r 1 -r
111 1 /1\ 1 -rn _ /_1_\ 1 ~rn
S' = - +-1-~ -\-\-—r -
a ar ar1 arn 1 \ajl-r~1 \arn~l) 1 -r'
where in evaluating S' we have noted that it is a geometric series with first term 1/a and ratio r~l.
Thus the quotient S/S' (which gets rid of the (1 - rn)/( 1 - r) term) is a2rn_1. Comparing this to our
expression for P, we see that
(S/S')"/2 = {a2rn-l)n/2 = anrn{n~1)l2 = P.
450. Since cos(l80° +x) = cos 180° cosx-sinl80° sinx = - cos x, we see that cos 45° = i2 cos 225° =
z4cos315° = ••• = z40 cos 3645°. Similarly, we can show that the second, fourth, etc. terms are the
same. Flence, our sum becomes 21 cos 45° + 20z cos 135° = 21 V2/2 — 10* V2.
451. The expression yjx2 + y2 is the distance of the point (x, y) from the origin. The point on the
line 5x + 12y = 60 which minimizes this quantity is the closest point on the line to the origin, and the
desired minimum is the distance of this line from the origin, or |5(0) +12(0) - 60|/ V52 + 122 = 60/13.
452. The center of the circle is on the side shared by the two bottom squares
(by the symmetry of the figure). Let the distance OC be x. By the Pythagorean
Theorem,
Vx2 + l2 = OB = OA= yJ(2-x)2 + (l/2)2,
where the final expression comes from considering the vertical and horizontal
distances between O and A. Equating these expressions for x, we find x = 13/16.
Using the above expression to find OB, our desired radius is 5 Vl7/16.
453. When omitting one number between 1 and n inclusive, the highest average that could occur
happens when 1 is the number erased, leaving n-1 numbers with an average of (n - l)(n + 2)/2(n -1) =
(n + 2)/2 (since the n-1 numbers form an arithmetic sequence). Similarly, the lowest average occurs
when n is erased, leaving n-1 integers with an average of n(n - l)/2 (n - 1) = n/2. Since 18.8 must
fall between these values, we have n/2 < 18.8 < (n + 2)/2, or
n < 37.6 < n + 2.
For this to be true, n must be 36 or 37. Since the average of the n-1 remaining numbers is 18.8,
n cannot be 37, since division by 37 - 1 = 36 cannot yield a decimal of .8. Hence, there were 36
numbers originally. The sum of these was 36(37)/2 = 666. Let the number omitted be x. From the
given information, (666 - x)/35 = 18.8 and we find x = 8.
454. Fiom the second equation, we have c(a + b) = 23. Since 23 is prime, either c — 23 and
(a + b) — 1 or c — 1 and (a + b) — 23. The former case is impossible since a and b are positive integers,
so c = 1 and a + b = 23. Using this in the first equation, we have ab + b = 44. Substituting a = 23 - b
for a, we find b2 - 24 b + 44 = 0. Thus, b = 2orb = 22, which give the solutions (21,2,1) and (1,22,1).
There are 2 solutions.
455. Letting N be the initial population, we have
N = x2, N + 100 = y2 + 1, N + 200 = z2.
Subti acting the first two equations, we find (y — x)(y + x) = 99. Since y > x and both are positive
integers, we have the following possibilities:
y~x= 1 and y + x = 99 y-x = 3 and y + x = 33
y - x = 9 and y + x = 11.
From these, our possibilities for (x,y) are (49,50), (15,18), and (1,10). For the last two of these,
x2 + 200 is not a perfect square, so these cases must be eliminated. The first case checks out, as
492 + 200 = 512. Hence, N = 492 = 2401.
456. Since AB = AB', /.ABB' = /AB'B and by AA similarity, A BAB' ~ A _ C

AACB. Since /C'AB' = /CAB, we have /CAC = /BAB' = /ACB. Thus,
by SAS congruence we have A CAC' = A ACB = A AC'B'. From these /
congruences, AC = BC and AB = CC', so ABCC' is a parallelogram (part ^ B C
ii). Since AC || BC, /CB'C' = /B'C'A = /ACB, so AB'DC is isosceles (part i). Since AC = AC
and aADC' ~ A B'DC, DB'/CB' = AD/AC' = AD/AC (part iii). Since /C'B'C = /ACB' = /CAC,
AC'CB' is a cyclic quadrilateral (part iv). Finally, since A BAB' ~ A ACB, AB/BB' = BC/AB so that
AB2 = (BB')(BC). By the converse of the Power of a Point Theorem, this means that AB is tangent to
the circumcircle of AB’CC (part v).
457. Let the chords be AB = 8 and CD = 7 and their midpoints be F and

E, respectively. The perpendicular bisector of any chord passes through the
center of the circle, so the center of the circle is intersection of the perpendicular
bisectors of the chords. Let the chords intersect at X (not labelled). Hence
OE = AB/2 - XB = 2 and CE = CD/2 = 7/2. From the Pythagorean Theorem,
OC = VO£2 + CE2 = V65/2. Hence, the diameter is V65-
458. First we must show that in any group of 4 points, no three of which are collinear and no three
of which form a right triangle, there is some group of three points which forms an obtuse triangle.
This is a simple exercise in contradiction. If the four points form a convex quadrilateral and all its
angles are acute, then the sum of the angles would be less than 360°, a contradiction. If the points
form a concave quadrilateral (i.e. three points form a triangle containing the fourth), the three angles
formed at the interior point when connected to the vertices of the triangle must add to 360°. If all
three of these are acute, the sum will be less than 270°.
Now we can move on. Each obtuse triangle which can be formed among the n points is contained
in n - 3 groups of 4 of the n points (since we can choose any of n - 3 other points to include with the
three triangle vertices). Thus, if there are T obtuse triangles, then there are at most T(n - 3) groups
of four points which contain obtuse triangles. If less than 1/4 of the triangles formed among the n
points are obtuse, then T < (1/4) (3). Hence,
n(n - 1 )(n - 2)(n - 3)

T(n - 3) <
1(2)(3)(4)
Thus, the maximum possible number of groups of four points which contain an obtuse triangle is
less than the total number of groups of 4 points. This means there is some group of 4 points among
the n for which we cannot form an obtuse triangle by connecting any 3 of the 4 vertices. This is a
contradiction to our initial proved assertion, so our assumption that less than 1 /4 of the triangles
formed by choosing 3 of the n points are obtuse must be false. Our proof is complete.
459. Starting from (0,0), every time the line passes through either a vertical line, x = n for some
integer n, or a horizontal line, y = m for some integer m, it passes into a new square. Since it crosses
118 vertical lines (x = 1 through x = 118) and 152 horizontal ones, we have 118 + 152 = 270. In
this count, we have not included the unit square with (0,0) as its lower left vertex, so this gives us
271. Unfortunately, we are still not right because every time the line passes through a lattice point,
it enters a new square, but we have counted it as passing through two new squares, one for the
horizontal and one for the vertical line. Hence, by the Principle of Inclusion-Exclusion, we subtract
the number of lattice points (besides the initial and final points) through which the line passes. The
given line is described by 119y = 153x, or 7y = 9x. Hence, we have one lattice point on the line
for each x that is a multiple of 7. Since there are 16 positive multiples of 7 less than 119, the line
passes through 16 lattice points besides the endpoints. Finally, our total number of lattice squares is
271 - 16 = 255.
460. Since at x = 0 we have 0/(-l) = -3P(0), x = 0 is a root of P(x). At x = 3, we find that 3P(2) = 0,
so x = 2 is a root of P(x) as well. Using x = 2 in the given equation gives 2P(1) = -P(2) = 0, so x = 1
is a root. Putting this together, we find P(x) = x(x-T)(x-2), which does satisfy the restrictions of the
problem.
461. If a fraction is reducible, so is its reciprocal. Using long division on (5n + 6)/(n - 13) as in
our chapter on polynomials, we find
5n + 6 _ 71
-— 5 -I-.
n — 13 n - 13
Hence (5n + 6)/(n - 13) is reducible if and only if 71/(n - 13) is. (Make sure you see why.) Since 71 is
prime, the smallest n for which 71/(n - 13) can be reduced is n - 84.
462. Since (r3 + 1/x3)2 = x6 + 1/x6 + 2, we can factor the numerator as a difference of squares:
/(*)
From AM-GM, we have x + ^ > 2, so that the minimum value of f(x) is 6.

463. The sum of the smallest three sides of a quadrilateral must be greater than the fourth side,
so that Fa+Fb + Fc > Fd. If c < d - 1, then Fb + Fc < Fd_lf so Fa + Fb + Fc < Fa + Fd_x < Fd. Hence
c = d - 1. Similarly, we can show that b = d- 2. Hence, d-b must be 2.
464. As suggested in the chapter on trigonometry, square both equations, yielding
sin2 A + sin2B + 2 sin A sin B = 1/9

cos2 A + cos2 B + 2 cos A cos 6 = 16/9.
Adding these and noting that cos2 x + sin2 x = 1 gives
2 + 2(sinAsinB + cos A cos B) = 2 + 2cos(A - B) = 17/9,
so that cos(A - B) = -1/18.
F B 465. First, [ABC] = [ABCD]/2 = 126. Since A ABC and A EBC share an
altitude, [FBC]/[ABC] = EB/AB, so [EBC] — 126/2 = 63. Similarly, we find
^—A/ [FBC] = (2/3)(126) = 84. Since AB || CD, we have AEBG ~ A CFG. Since
D
EB/AB = 1/2 and FC/CD = 2/3, we find (EB/AB)(CD/FC) = EB/FC = 3/4.
From our similar triangles, GC/EG = 4/3, so GC/EC = 4/7. Thus, [GBC] = (4/7)[£BC] = 36. Now we
can find [EGFCB]:
[EGFCB] = [EGB] + [BFC] = [EBC] - [BGC] + [BFC] = 63 - 36 + 84 = 111.
Finally, [AEGFD] = [ABCD] - [EGFCB] = 141.

466. Lots of equal angles means lots of similar triangles. Namely, we determine that A ADM ~
aACD and aBCM ~ aACB. Hence, AD/AC = AM/AD, BC/CM - CA/BC, and
AD2 + BC2 = (AM)(AC) + (MC)(AC) = (AM + MC)(AC) = AC2,
completing our proof, since AD, BC, and AC satisfy the Pythagorean Theorem.
467. We can prove a number is not a complete square by proving that it is always between a pair
of consecutive perfect squares. The given quadratic suggests placing the number between (n - 10)2
and (n - 9)2. Since (n - 9)2 - (n2 - 19n + 89) = n - 8, (n - 9)2 > n2 - 19n + 89 for all n > 8. Similarly,
(n - 10)2 < n2 - 19« + 89 if and only if n > 11. Thus, for all n > 11, (n - 9)2 > n2 - 19n + 89 > (n -10)2,
showing that the expression cannot be a perfect square since it is always between two consecutive
squares.
468. Through the two parts we are proving that for any square set we can make a new set by
adding a certain element or by replacing one of the elements with a certain new element. The proofs
that follow are largely algebraic and a good exercise in manipulation. Let x2 be the sum of the
products of all pairs in the set and s be the sum of the elements of the set so that b - s + 2x. For the
new set [a\, 02, • • •, 0«, b), we have
0102 + 0103 + • • • + 0M-I0n + b(a 1 + 02 + • • • + 0«) = x2 + b(s)

= x2 + (s + 2x)(s) = (x + s)2.
so that the new set is a square set. For the second part, suppose flf is the element replaced by b. Now
in our sum of products of pairs of a's, we must remove all the terms involving flf. We can write this
as
U\Cl2 + + • • • + Cln-\Cln — Q-i{s — Cli),
where the last term 'erases' the a* terms from the sum. Hence our sum of the products of pairs in the
new set is
ci\ci2 H-h an-\an - ai(s - cij) + b(a\ H-(-«„)- baj x2 - ai(s - af) + b(s - «;)
x2 + (b- flf)(s - at)
x2 + (s + 2x- ai)(s - cij)
x2 + 2x(s - at) + (s - cii)2
(X + S - fli)2.
Thus the new set is a square set as well.

469. Since (e?1)7 = e7ai = cos 7a + i sin 7a, we have
i-i ,
(cos a + i sin a) - cos 7a + i sin 7a.
To find cos 7a in terms of cos a, we equate the real parts of the equation above:
cos 7a = cos7 a + f 2)z”2 cos5 a s’zn2 a H-f (^ j i6 cosflsm a.
We can convert the right hand side to a polynomial in cos a by noting sin2 a = 1 - cos2 a. Letting
x - cos a we have the polynomial desired in the problem. Looking at the polynomial above, it is
clear that we will only have odd powers of cos a, so that the coefficient of cos2 a, or x2 in terms of the
problem, is 0.
470. Clearly there are (g3) sets of three numbers we can choose. Let's look at the chosen numbers
mod 4. The sum of the three integers is divisible by four if the resulting set (after evaluating each
number mod 4) is congruent to {0,0,0}, {0,2,2}, {0,1,3}, {1,1,2}, or {2,3,3}. These can be done in
respectively (3), 3 • Q), 3(4)(3), (2) • 3, and 3 • Q) ways. Thus, our probability is
1 + 9 + 36 + 18 + 9 73
13-12-11 286'
3-2-1
471. Expanding the given equation, we have
x2 + 2xy V2 + 2y2 + z2 + 2zf V2 + 2f2 = 5 + 4 V2.
Just like when dealing with imaginary numbers, the rational part of the left equals that on the right
and the irrational part on the left equals that on the right. Hence, we can change the sign of the
irrational parts and equality still holds (just like taking the conjugate of a complex number). Thus,
we can write
x2 -2xyyJl + 2y2 + z2 - 2zf V2 + 2t2 = 5 - 4 V2,
from which we have

(x-y V2)2 + (z - f V2)2 = 5 - 4 V2.
Since (52) < (4 V2)2, the number on the right is negative and thus cannot be the sum of squares of
real numbers. Hence, the initial equation has no rational solutions.
472. We are given
r 3 7 3 7
Fl = ^ + Rf + Rf + Rf+"'and
r 7 3 7 3
2 Ri + R? + R\ + R* + "' '
Finding the sum and difference of these, we find
F1+F2 = 1°(j1 + ^ + Jj+...) and
F2- F1 = 4 (__— + —_L

Ui R2i Rl Ri
Doing the same with base R2/ we find
F\ + F2 = 7
J_ and
Ri R? R?
ft-*=3GHrn-k+-)
We can set the two expressions for F\ + F2 and F2 — F\ equal and evaluate the series as geometric
series, yielding
10 7
Fi + F2 - and
Ri- 1 R2 — 1
R\ + 1 R2 + 1
Hence, we have two linear equations in Ri and R2. Solving this system for (Ri,R2), we find
(R\,R2) = (11,8) and our desired sum is 19.
473. Draw the incircle and label the equal tangents as shown. Now we can
find the area of the triangle in two ways, with Heron's formula and as rs. Hence,
rs = 4(14 + x) = V(14 + x)(6)(8)(x) (since the tangent lengths x, 8, and 6 are s - a, s-b, 6
and s - c). Dividing each side by 4 Vl4 + x gives Vl4 + x = V3x- Hence, x = 7 and
the sides of the triangle have lengths 13,14, and 15. The shortest is obviously 13.
474. From Heron's formula, the area of A ABC is 4 V5. We find XY by finding the area of
AAXC in two ways. First, from the Angle Bisector Theorem, XC/BX = AC/AB - 6/3 = 2, so
[AXC]/[ABC] = XC/BC = 2/3. Hence, [AXC] = 8 V5/3. Since [AXC] = (AC)(XY)/2, we solve for XY
and find XY = 8 V5/9.
475. Any number whose digits are all the same is the product of a single digit and a number
consisting only of ones, i.e. we can write the number as k( 11 • • • 11). Either k - 7 and 11 • • • 11 is a
multiple of 7 or 11 ••• 11 is a multiple of 49. From direct checking we can find that the lowest number
consisting only of ones which is also a multiple of 7 is 111111. Unfortunately this is not also a
multiple of 49, so we must multiply by 7 to get a multiple of 49. Our desired number is 777777.
476. For any integer n which is not a multiple of 3, make the list n, 3n, 9n, etc. Clearly we make
our maximal subset by including n and every other member of each list (since no member of one
list can be three times a member of a different list). Thus, our maximal subset contains all numbers
which are not multiples of 3, all multiples of 9 which are not multiples of 27, and the number 81, for
a total of (100 - 33) + (11 - 3) + 1 = 67 + 8 + 1 = 76.
477. First we attack n — 1. We factor the resulting polynomial in x:
y2 = x(x1991 - x1989 + x2 - 1) = x(x - l)(x1990 + x1989 + x + 1)

= x(x - l)(x + l)(x1989 + 1) = x(x - l)(x + l)2(x1988 - x1987 + x1986 -+ 1).
Since (x + l)2 is a perfect square, the remaining product, x(x - l)(x1988 - • • • + 1) is a perfect square.
Now we compare the three terms of this product. The last two terms differ by 1 from a multiple
of x and therefore cannot share any factors with x. Similarly, the final two factors do not share any
factors. Thus, the three factors are relatively prime numbers whose product is a perfect square and
as such each factor must be a perfect square. Since x and x - 1 can both be perfect squares only if
x = 1, the only solution to the equation is (x, y) = (1,0). Since we want solutions in positive integers,
we discard this and move on to n - 2.
Again we factor, getting x(x1991 — x1989 + 3x2 -1) = y2. And again, our factors are relatively prime,
so we must have x1991 - x1989 + 3x2 - 1 = z2 for some integer z. Attacking this mod 3 to eliminate the
3x2 term, we find
x1991 - x1989 = z2 + 1 (mod 3).
Since x1991 - x1989 = x1989(x2 - 1), we see that x1991 - x1989 is congruent to 0 mod 3 (try x = 0,1,2 (mod
3)). However, z2 +1 can never be conruent to 0 mod 3 since no square is congruent to 2 mod 3. Thus,
the equation has no solutions in positive integers for n = 2.
478. Let the greater part have length 1 so that the lesser part has length R. Thus, we have
R _ 1
I ~ R+l'
so that R2 + R = 1. Returning to the desired expression, we want to find R2 + R_1. From above we
find R~l = R + 1, so that 0 = R2 + R - 1 = R2 + R-1 - 2. Thus, we have R2 + R_1 = 2. Using this
repeatedly, we find
rR(r2+r f+r-i]
479. Let point M be some point on the extension of AB past B. We will

prove that A\B\ and B\Ci are the bisectors of lBB\C and LBB\A. Since
BBi bisects lABC, we have
lMBA\ — /A\BB\ — iBiBA = 60°.
Since a point on the angle bisector of an angle is equidistant from the sides of the angle, we find that
Ai is equidistant from AM and BB1 (since BA\ bisects iMBBi) and also equidistant from AB and AC
(from angle bisector AA\). Hence, A\ is equidistant from BB\ and AC, from which we deduce that
B\Ai bisects zBBiC. Similarly, we can show that B\Ci bisects LBB\A. Finally,
ZCBiB + zABiB
1A1B1C1 = ZA1B1B + ZBB1C1 = = 90°.
2 ~
480. We can draw a simple one-to-one correspondence between the numbers in base three which
have no 2's and numbers written in base two. In other words, we can interpret each base three
number with no 2's as a base two number. Clearly for each base two number there is exactly one
corresponding base three number with no 2's. Hence our task is to find the largest base three
number less than or equal to 1992 which has no 2's, then interpret that as a base two number.
Since 1992io = 22011223, the largest number in base three less than 1992io which contains no 2's is
IIIIIII3. Evaluating this in base two, we have 11111112 = 127. For all base two numbers less than
this there is exactly one corresponding base three number with no 2's. Thus, as we discussed above,
there must be 127 base three numbers from 1 to 1992 with no 2's.
481. Seeing products of cosines, we think of all of our trigonometric identities involving products.
We also note that many of the angles in the product are twice those of other angles in the product.
Hence, we come to the identity sin 2x = 2 sin x cos x, but with no sines, this doesn't look helpful;
however, we can always put one in. Let the product be P. Hence, we have (notice that we'll use
cos(180° — x) = -cosx and sin(180° - x) = sinx among our manipulations to use acute angles as
much as possible)
. 71 tz n 2n 3n 4tl 5n 6n
P sm- sin — cos — cos — cos — cos — cos — cos —
7 7 7 7 7 7 7
1 2n 2n 3n 4n 5n 6tl
- sin — cos — cos — cos — cos — cos —
2 7 7 7 7 7 7
1 471 3tl 3tl 2n tl
- v sin — cos — cos — cos — cos —
4 7 7 7 7 7
1 . 3ti 371 3tl 2n n
—- sm — cos — cos — cos — cos —
4 7 7 7 7 7
1 . 671 371 2tt 71
- - sm — cos — cos — cos —
8 7 7 7 7
1 7i 37i 2n tl 1 . 2tl 3n 2n
--sm — cos — cos — cos — = sm — cos — cos —
877 77 16 7 7 7
371 1 . 371 3n
cos — = - — sm — cos —
7 32 7 7
.71 1 . 671 1 . 71 1
Hence, Psm 7 = --sm y = --sm7, or P = --.
482. Let x(i, j) be the number appearing on the card in row i and column j after both rear¬
rangements. Suppose that the x(i, j) do not increase from left to right. Then for some i, j, we have
x(i, j) > x(i, j + 1). Since column j is ordered from top to bottom, there are exactly i - 1 entries in the
column less than x(i, j). Hence, there are at most i -1 entries in column j which are less than x(i, j +1).
Now consider the first i entries in column j + 1. At most i — 1 of the entries in column j are not
greater than all of these i elements. Thus, one of the original rows before the column rearrangement
must not have been in order, a contradiction. (The smallest i entries in column j + 1 cannot be paired
with i elements in column j so that each element in column j + 1 is greater than the corresponding
element in column j.)
483. We use a Principle of Inclusion-Exclusion approach to this problem. We can get AXYZ by
starting with A ABC and cutting out A CBF, A AEB, and A ACD. In doing this we cut out triangles
CDY, FZB and AEX twice, so we must add these pieces back in. Since CD/CB = 1/3, we have
[ADC] = [ABC]/3 and similarly for the initial triangles removed. Since YD/AD = 1/7, we have
[CYD] = [ACD]/7 = [ABC]/21 and similarly for the other two triangles added back in. Hence we
have
[XYZ] = [ABC] - 3([ABC]/3) + 3([ABC]/21) = [ABC]/7.
484. Since CF bisects BD at E, we can form a parallelogram by selecting the point

G on CF such that EF = EG, since then GF and BD bisect each other. Remember that
introducing parallel lines or parallelograms is often very helpful! Since DG || AT,
we have aCGD ~ ACFA. Since CD/CA = 1/2, we have AF - 2(DG) - 2(BF) = 10
and AB = AF + FB = 15.
485. We can use the Pythagorean Theorem to determine AB = 5 and DB = 13.
The desired ratio is just sin ZDBE. Since ZDBE + IDBC = 180°, we have sin ZDBE = sin ZDBC. Angle
DBC is composed of two angles in easily determined right triangles. Hence, we write
sin ZDBC = sin(/DBA + /.ABC)

= sin /DBA cos /ABC + sin /ABC cos /DBA
= (.AD/DB)(CB/AB) + (AC/AB)(AB/DB) = 48/65 + 15/65 = 63/65.
486. From the relationships between the roots of a polynomial and its coefficients, we have
a2 + b2 + c2 = P and a2b2c2 = R. Seeing cosines in the desired sum, we apply the law of cosines to
A ABC and our desired sum becomes
b2 + c2 - a2 a2 + c2 - b2 a2 + b2 - c2 a2 + b2 + c2 P
2abc 2abc 2abc 2abc 2 Vk
487. As in a prior problem, the equality (5 + 3 y/2)m = (3 + 5 y[2)n holds if and only the equality
(5-3 V2)m = (3 - 5 V2)" holds. Since 0<5-3V2<l and |3 — 5 V2| > 1, the magnitude of (5 - 3 V2)m
is always less than one and the magnitude of (3 - 5 V2)” is always greater than 1. Hence, the two
can never be equal and the initial equation therefore has no solutions in positive integers.
488. From the given limit and recursion definition, we can write
bun-\ + CUn—2
3 = lim = lim ( b + c———
n—»oo un-1
Un-1 »->0° V
From our limit, we have lim (un-\/un-i) - lim(u„/w„_i) = 3, so lim (m„_2/mm_i) = 1/3. Hence, b and
n—xx) n—>oo n—xx> 7
c must satisfy 3 = b + c/3. Since b and c are nonnegative integers, our possible solutions then are
(0,9), (1,6), (2,3), and (3,0). The last three are valid solutions, but the first yields the sequence 1,1,
9, 9, 81, etc. The limit Jim (un/un-1) is not defined for this sequence.
489. Seeing only even numbers available for our desired representation, we think of doubling
or halving. Since n/2 is an integer, it can be written in normal base three using 0's, l's, and 2's.
Doubling this representation for n/2 gives us a base three representation for n consisting of 0's, 2's,
and 4's. For example, 5 = 1 • 31 + 2 • 3°, so 2 • 5 = 10 = 2 • 31 + 4 • 3°.
expressions are anx2n, a2nx2n, and a^+1xn\ respectively. Equating these, we find 2n = n2 and an = a2.
Since n and an are nonzero integers, we have n - 2 and an = 1. Thus, f(x) = x2 + bx + c for some b
and c. Putting this in the given equality /(x2) = [/(*)] gives
x4 + bx2 + c = x4 + 2 bx3 + (b2 + 2c)x2 + Ibex + c2.
Equating coefficients of x3 gives b = 0 and the coefficients of x2 then give c - 0. Hence, only one
function satisfies the problem, f(x) = x2.
491. We've already seen that we can write any even number in a base three representation with
only 0's, 2's, and 4's, so how can we use this to get a representation with only -l's, l's, and 3's? We
can get the latter by subtracting 1 from each digit of a representation with just 0's, 2's, and 4's. For
example,
4 • 33 + 2 • 32 + 0 • 31 + 2 • 3° - 11113 = 3 • 33 + 1 • 32 + (-1) • 31 + 1 • 3°.
Now how do we form the first number above with the 4's, 2's and 0's? Since we want to get our
original n by subtracting 11 • • • II3 from this number, we must add 11 • • • II3 to n to get the number
we write with 4's, 2's and 0's. We must be careful, however, because the sum 11 • • • II3 + n must be
even in order to write it in base 3 with 0's, 2's, and 4's, and we must have at least as many l's in
11 • • • II3 as there are digits in the normal base three representation of n; otherwise, when subtracting
it from the even number, we won't be able to change all the 0's, 2's, and 4's to odd numbers. Hence,
we can write n as suggested by following this method: write n in base 3; then choose a number of
the form 11 • • • II3 which is of the same parity (odd or even) as n; add these two and express the sum
in base three with 4's, 2's, and 0's as in the prior example; finally subtract the 11 • • • II3 from this to
get the representation for n. Let's try it for n = 18. Hence, n in base three is 2OO3. Thus, we add IIII3
(since III3 is odd, we can't use it). We find n + IIII3 = 18 + 40 = 58. Hence, we have
58 = 2(1 • 33 + 0 • 32 + 0 • 31 + 2 • 3°) = 2 • 33 + 0 • 32 + 0 • 31 + 4 • 3°.
Subtracting HH3 from this, we have
18 = 1 • 33 + (-1) • 32 + (-1) • 31 + 3 • 3°
as the desired representation for 18.

492. We find the area of the pentagon in two ways. First, connecting O to each of the vertices
of the pentagon forms five triangles congruent to aODC. Hence, we have [ABODE] = 5[ODC] =
5(OP)(DC)/2 = 5x/2, where x is the side length of the pentagon. Second, we have
[ABODE] = [AOD] + [ABC] + [ADE] = (AP)(DC)/2 + (AQ)(BC)/2 + (AR)(DE)/2

= (x/2)(AO + AQ + AR + 1),
where we have used AP = AO + OP = AO + 1. Setting this expression equal to 5x/2, we find our
desired sum is 4.
493. We could divide, but that would take a long time. Instead, let /(x) = x2 - x + a and
g(x) = x8 + 5x6 + 13x4 + 20x2 + 36. Since f(x) divides g(x) evenly then for any integer n, f(n) divides
g(n). Hence, /(0) = a divides y(0) = 36. Since /(1) - a,a divides g( 1) = 75. Comparing the common
factors of 36 and 75, we find that a is 1 or 3 (since a is known to be positive). Trying one or two
more integers can determine which of these is the answer. For example, since /(-2) - a + 6 and
g(-2) = 900, we exclude a = 1 because 900 is not divisible by 1 + 6 = 7. Hence, the answer is 3.
494. Proceed by induction. The case where n - 3 is obvious; the closest pair shoot each other and
the third person remains dry. Now suppose any group of 2m - 1 are such that at least one person is
left dry. We shall prove the same is true for 2m +1 people. Consider the two people among the 2m +1
who are closest to each other. These two shoot each other. If someone else shoots one of these two,
there are then 2m - 2 shots left and 2m - 1 dry people, so someone will be dry at the end. Otherwise,
we are left with 2m - 1 dry people and 2m - 1 shots among them. By our induction hypothesis,
someone will be left dry then. Hence, someone will be left dry if there is an odd number of people.
495. This one's a bit tricky! If we expand (V7 + V5)6, we find that the odd powered terms contain
V35- Since we are looking for an integer, we would do well to eliminate these odd powered terms.
In ( V7 + V5)6 these terms are positive, but if we expand (V7 - V5)6, these terms are negative. Hence,
if we add the two, we find
(V7+V5)6 + (V7-V5)6 = 2(^(V7)6+Q(V7)4(V5)2+Qj(V7)2(V5)4 + (V5)6j
= 13536.
,
Since (V7 - V5) < 1 we have
(V7 + V5)6 = 13536 - (V7 - V5)6 > 13535.
Since 13535 < (V7 + V5)6 < 13536, the answer is 13535.
496. Seeing the circumcircles involved, we think of the expression R - abc/AK for the circumra-
dius. Hence, the sum of the desired areas is
((.AX)2(BX)2(ABf (AX)2(CX)2(AC)2 \
n(ÂABX + AXc) - n
V 16[ABX]2 + 16[ACX]2 ) '
Since aABC and A ABX share an altitude from A, [ABX] = (BX/BC)[ABC]. Finding a similar expres¬
sion for [.ACX], our sum becomes
(BC2)(AX2)tl
(AB2 + AC2).
16[ABC]2
The only quantity which varies in this expression is AX2. Hence, our desired area is at a minimum
when AX is minimized, or when AX is an altitude. (Why?) From Heron's formula, the area of a ABC
is 6 V6, so (AX)(BC) = 2(6 V6) and AX = 12 V6/7. From the Pythagorean Theorem, we then find
BX = 19/7.
497. The value of y/x at a point in the plane is equal to the slope of the line
through the origin and the point (x, y). Hence, we seek the point on the circle
for which this slope is the greatest. Clearly this point will be such that the line
through the origin and (x, y) is tangent to the circle as shown. Since XY = V6
and OY = 3 V2, we have OX = Vl8 - 6 = 2V3 and tan ZYOX = V2/2. The
value of y/x at point X is the slope of the line through O and X, which is in
turn equal to the tangent of the angle formed by OX and the positive x axis.
Since this angle is 45° + ZYOX, we have
1 + tan ZYOX
= tan(45° + ZYOX) =
1 - (l)(tan ZYOX)
= 3+ 2 VI
498. First we establish that E is on AC by showing that if DE passes through

C, then [AED] > [BED], so we have to move DE towards A to get equal areas.
If DE goes through C, we have BD = CD < AD (by looking at the angles of
A ACD and A BCD). Hence we have the desired [ADC] > [BCD] since the two
triangles share an altitude. Now we draw altitudes CX and EY and assume that CX = 1. Thus,
AX = 1 and BX = V3 (since ZCAX = 45° and ZCBX = 30°), so [ABC] = (1 + V3)/2. Letting
EY = x, we find AY = x and YD = x V3/3, so [AED] = x2(l + V3/3)/2. Since 2[AED] = [ABC],
we have x2(l + V3/3) = (1 + V3)/2. Solving for x, we find x = -^3/4. Our desired ratio then is
AD/AB = x(l + V3/3)/(l + V3) = l/<fu.
A
499. Lemma: Quadrilateral DYAZ is circumscriptible if and only if
BD-CD = AB- AC.
Proof: If DYAZ is circumscriptible, then let the shown points be the points of
tangency. Then since SD = DP, BP = BR, and CS = CQ, we have
BD - CD = (BD + DP) - (CD + DS) = BP-CS = BR- CQ.
Since AR = AQ, we have BR - CQ = (BR + RA) - (CQ + QA) = AB - AC, as desired. We can prove
the converse by assuming that the line from C tangent to the incircle of AABY intersects BY at some
point E different from D. As above we have AB - AC = BE - EC. Let BE = BD + aDE, where a - ±1.
Combining AB - AC = BE - EC with the obvious EC = CD + (EC - CD), we find
AB - AC = BD - CD + (aDE - EC + CD).
This last expression equals zero only when A DEC is degenerate, i.e. when point D and E are the
same. If points D and E are the same, DYAZ is circumscriptible. Hence, AB - AC = BD - CD if and
only if DYAZ is circumscriptible.
Applying this to our problem, we find that we must prove that if AB - AC = BD - CD and
BC - AB = CD - AD, then we have AC - BC = AD - BD. This is clearly true (just add the two given
equations to get the desired one), so our lemma proves the problem completely.
500. In the figure, H is the orthocenter, I is the incenter, and D A

is the midpoint of BC. If we continue BH to meet AC at M and CH
to meet AB at N, we will find from cyclic quadrilateral AMHN that
LCHB = LMHN = 180° - lA. Similarly, ABIC = 180° - (ZB + zC)/2 =
180° - (180° - zA)/2 = 90° + lA/2. Letting IH = ID = 1, we note
that BD = 2 tan ZBHD = tan LBID from right triangles BHD and BID.
Letting x = LA/4, this equation becomes
2 tan(90° - 2x) = tan(45° + x),
where LBHD = LBHC/2 and LBID = LBIC/2. Applying various trigonometric identities (such as
tan(90° - 2x) = sin(90° - 2x)/ cos(90° - 2x) = cos2x/ sin2x = 1/ tan2x), we find
2 1 - tan2 x 1 + tan x _ tan 45° + tan x

tan2x ~ tanx 1-tanx 1 - tan45° tanx'
Writing tan x = (sin x)/(cos x), we work through much algebra (be careful!) to find 3 cos x sin x = 1 or
• „ 2
sm2x = -.
Since sin2x = sin(A/2) = y/(l - cos A)/2 = 2/3, we find cos A = 1/9.
501. First note that from the AM-GM Inequality we have
" + 17/,,> V17

2
for any positive n. Hence, if x > 0, then 2y = (x + 17/x) > 2 Vl7. From this we get z > Vl7, then
w > x/V7, and x > Vl7. We can rewrite the first equality in the problem as
Since (x-Vl7)/2x = l/2-Vl7/2x, (x-Vl7)/2x < 1/2 and we have y- Vl7 < (x-y]Y7)/2. Rearranging
this we have x - y > y - Vl7. Since y > xlY7, x - y > 0, so x > y. Similarly we can show y > z,
z > w, and w > x, so that x>y>z>w>x. Thus, x = y = z~w = Vl7 is the only possible
positive solution. We check it and find that it works. Since {-w, -x, -y, -z) is a solution if and only
if (tv, x, y,z) is, our only solutions are ( Vl7, VL7, Vl7, Vl7) and (- V17, - Vl7, - Vl7, - Vl7).
502. Look at x in base 2. The recursion in the problem is then equivalent to sliding the decimal
place over to the right and chopping off any integer part. For example, let xo = O.OIOIII2. Hence,
x\ = 2xo = O.IOIII2, X2 = 2xi - 1 = O.OIII2 (since 2xi > 1), X3 = 2x2 = O.III2, etc. As we can see, the
number of decimal places of Xf decreases as i increases unless xo repeats indefinitely. Since xo = X5,
xo must go on indefinitely. Since the 0th and 5th term differ by sliding the decimal 5 places to the
right, Xo must consist of a block of 5 digits repeating indefinitely in order for Xo and X5 to be the
same. Since each of these digits can be 0 or 1, there are 25 = 32 possible blocks. However, since the
block xo = 0.11111 is equal to 1 and xo < 1, we exclude this from our count of 32. Thus, there are 31
Xo such that xq = X5.
503. For the first part. Let k be one of the elements and S*- the sum of the remaining elements.
The sum of all the elements is k + S^. Since SjJk = d, where d is an integer, (Sjt + k)/k = d + 1 and k
divides the sum of all the elements.
Since 1 is an element of the set in the second part, the greatest common divisor of all the numbers
in the set is clearly 1. The sum of all the members of the set is 1 +1 + 2 + • • • + 2" = 1 + 2n+1 -1 = 2n+l.
Hence, for any member of the set 2k we have (2n+1 - 2k)/2k = 2n+l~k - 1, so any element of the set
divides the sum of the remaining elements and the set is magical.
Let n be the perfect number whose proper divisors form our set. The sum of the numbers in the
set is then ft. Let k be the number removed from the set. The sum of the remaining elements is n — k.
Since ft is a multiple of k, n — k is also a multiple of k and the set is thus magical.
For the fourth part, let the members of the set be a, b, and c with a < b < c. Since a + b must be a
multiple of c for the set to magical and a + b <2c, we must have either a + b - c or a + b - 2c. The latter
set is easy, since in this case if a < b < c and a + b - 2c, then a = b = c. Since the greatest common
factor of the three numbers must be 1, the only set of this type is {1,1,1}. For the case a + b = c, a
and b cannot both be even, or then all three will have the common factor 2. Thus, let a be odd and
b even. We know that a\b + a + b and b\a + b + a, or a\2b and b\2a. Since b is even, we let b = 2k, so
a\4k and 2k\2a. Since a is odd we find a\k and k\a, so k = a and our set is {a, 2a, 3a}. Since the greatest
common factor of these is 1, our set is {1,2,3}. Finally, if a and b are odd we go through exactly the
same steps as above to show a\b and b\a so a - b and we thus find the solution {1,1,2}.
Let n > m. Since the set is magical, we have n\m + 4, so 4 + m = kn for some k. Since n > m, we
have kn > km, so 4 + m > km. Solving for k, we have k < 1 + 4/m. Since m > 3, k can only be 1 or
2. For k = 1, our set is {1,3, m, m + 4}. Thus, m|8 + m or m|8. Since m > 3, m = 4 or m - 8. Only the
latter leads to a magical set, namely {1,3,8,12}. For k = 2, from k < 1 + 4/m, we must have m = 3 or
4. Only the value m = 4 leads to a magical set, namely {1,3,4,4}.
Finally, for the last part, we can easily see that adding the sum S of the elements in the original
set we form a new magical set. The proof that this new set is magical is straightforward. Let k be
an element of the original set. Since k\S, k will also divide 2S. Hence, we need only prove that S
divides the sum of the elements of the new set. Since S clearly divides 2S, we have shown that all
the elements of the new set divides the sum of the elements of the set. Thus, the new set is magical.
504. Remove the barriers. Instead of standing still at the barrier, let the particle move beyond it
according to the toss of the coin. Hence, if the particle ever goes to (ft, n + k) from point (n - l,n +k)
without barriers, then it would have gone to (n, n) in 2n + k moves with barriers since it would stay
stuck on the north barrier for the k moves it takes to get up to the (n + A:)th row. (Why don't we
consider the case where the non-barriered particle gets to (n, n+k) from (n, n + k- 1)? Because then
with barriers, it would have taken less than 2n + k moves to get to (ft, n).) Thus, we remove the
barriers and find the probability that the particle goes through (n, n + k) from {n-\,n + k). It can get
to (n - 1, n + k) by taking n- 1 east steps out of 2n + k - 1. Hence, the probability that the particle
gets to (n - 1, n + k) is
CtVKlT"
There is a 1/2 chance that the particle will then take the right step from this point to (ft, n + k), so our
probability is
P
iCv.r')(!)“"
We're not quite done; if the unbarriered particle gets to (ft + k, n) via point (n + k,n - 1), then it would
have gotten to (n, n) inln+k steps as well. These paths to (n, n) are just the mirror images of the paths
to (n, n + k), so the probability of this course is the same as for the one to (n, n + k) via (n - 1, n + k).
Hence, our final probability of getting to (n, n) in 2 n+k moves is 2 P, or
505. Let the numbers be a\, a2, ..., flz. Since the range of tanx for -n/2 < x < nil is all real
numbers, then for each a, there is an x* such thata; = tanx*. If we divide the interval {-n/2, n/2) into
six equal subintervals, there are two of the X; in one of the subintervals by the Pigeonhole Principle.
Let these two be xk and Xj with xk the larger of the two. Hence we have 0 < x* - Xj < n/6. Since tan x
increases over the interval (0, n/6), this inequality becomes tan 0 < tan(xjt - Xj) < tan n/6, or
tanxfc-tanx,- 1
0 < --- <
1 + tan xk tan Xj y3
Hence we have
0<AZA< *
1 + akUj V3
as desired.
506. Since ZAED + ZAFD = 180°, AEDF is a cyclic quadrilateral. Since these two angles are right
angles, AD is a diameter of this circle. Hence, the circumradius of AAEF has length AD/2. The
circumradius of A ABC is abc/^[ABC]. Since (AD)(a/2) = [ABC], the product of these circumradii is
/AD\ f abc \ _ f[ABC]\ / abc \ _ be

V 2 ) U[ABC]) ~ V a ) U[ABC]) ~ J'
Thus, we have [^4BC] = be/4. Since we want LA, we write [ABC] = (be/2) sin A and we find
sin A - 1/2, so LA = 30° since A ABC is acute.
507. The coin could end up heads 0, 2, 4, ..., or 50 times to satisfy the problem. Hence, our
probability is
This sum is the the sum of the even powered terms of (1/3 + 2/3)50. To isolate the even powered
terms, we note that the odd powered terms of (1/3 - 2/3)50 are negative. Hence, adding (1 /3 + 2/3)50
and (1/3 - 2/3)50 eliminates all the odd terms and leaves us with double the sum of the even terms.
Our probability then is
_ (1/3 + 2/3)50 + (1/3 - 2/3)50 _ 1 + (1/3)50

2 2
508. This is a tough one, so pay close attention. First we draw the tangents MX and MY. Since
LACM = ZMYA = LMXA = 90°, points A, Y, C, M, and X are all on a circle as shown below. Since
AY = AX, we have AY = AX, so LAYX = ZAXY = ZACY and AAYC ~ aAKY, so AK = AY2/AC.
Since AB = AY, we find AK = AB2/AC = (4/5)AB. Hence, point K is fixed. Since MY = MX, the
orthocenter H of AYMX is on line AM and YH = XH. Since YH and AX are both perpendicular to
XM, they are parallel. Similarly, AY || HX and AYHX is a parallelogram. Since AY = AX, AYHX is a
rhombus. Let point L be the intersection of the diagonals of AYHX. Since lALK = 90° (the diagonals
of a rhombus are perpendicular), as M moves, point L moves around the circle Y with diameter AK.
Since AH = 2(AL) (why?), point H is on the circle homothetic to T with ratio 2 : 1 and center of
homothecy A. Thus, the locus of H is the circle through A with center K (except point A). To show
that all these points are in the locus, we note that for any point H on the circle, we draw AH through
to CM in the diagram to determine the point / such that the tangents to circle A from / form the
triangle with orthocenter H as the problem requires.
509. First we recall that the interior angles of a regular n-gon have measure 180(n - 2)/n. Hence,
let's consider the case where we fit 3 polygons about a point. Let the polygons have a, b, and c sides,
where a > b > c > 3. Since the sum of the angles about a point is 360°, we have
180(fl - 2) + 180(fr - 2) + 180(c - 2) = ^

a b c
Simplifying this equation we find

1111
-^ 7—I — ^•
a b c 2
Note that if c > 7, then the sum on the left will always be less than or equal to 3/7. Hence, c - 3,
4, 5, or 6. These lead to 1/a + l/b = 1/6, 1/4, 3/10, or 1/3. Now we need a systematic way to solve
equations of the form
11 p
~ + T ~ ~/
a b q
where p and q are relatively prime and a and b are integers. Write the left side of this as (a + b)/ab.
Let a = (m + q)/p and b = (n + q)/p. Hence,
a+b m + n + 2q l
ab = V q2 + (m + n)q + mn
Thus, we see that the big fraction must equal 1 /q. Thus, q(m + n + 2q) = q2+ (m + n)q + mn, so mn - q2.
Hence, we have a way to generate our solutions; list all ways to write q as a product mn of integers,
keep only those for which p divides q + m and q + n (since a and b must be integers), and the solutions
are (m + q)/p and b = (n + q)/p. Coming back to our above values for 1/a + l/b, we thus generate the
solutions (42,7,3); (24,8,3); (18,9,3); (15,10,3); (12,12,3); (20,5,4); (12,6,4); (8,8,4); (10,5,5); and
(6,6,6).
For four polygons, we have a>b>c>d> 3 and working through the algebra as above we get
1/a + l/b + l/c + l/d = 1. Hence, if d > 5, the sum is less than 1. We thus consider d = 3 and d- 4.
For d - 4 we have only (4,4,4,4). For d = 3, we get 1/a + 1/b + 1/c = 2/3. Again, we deduce c > 5 is
impossible. We then let c = 3 or 4. Thus we get 1/a + 1/b = 1/3 and 5/12, which we then solve with
our process to get (12,4,3,3); (6,6,3,3); and (6,4,4,3).
The same processes work for the cases of five or six polygons (or we can find these by inspection),
and we get (6,3,3,3,3); (4,4,3,3,3); and (3,3,3,3,3,3) as the solutions. Clearly there can't be more
than six polygons since the smallest angle possible among the polygons is 60°. Hence there are a
total of 17 solutions.
This is a very difficult problem to do without the use of a computer. The authors of this problem
(the authors of this text together with Sam Vandervelde) used a computer to determine there are
indeed 17 solutions before developing a methodical method to find the solutions without a computer.
Therefore, don't feel too bad if you had to resort to peeking at the solutions to solve this one!
the Art of Problem Solving
Volume 2: and Beyond
the Art of Problem Solving Volumes l and 2 have been used by many
of the top students and programs in the United States since 1993. The
texts offer a challenging exploration of problem solving mathematics
and preparation for programs such as MATHCOUNTS® and the
American Mathematics Competitions.
Richard Rusczyk is the founder of www.artofproblemsolving.com. He was a national MATHCOUNTS participant
in 1985, a three-time participant in the Math Olympiad Summer Program, and a USA Math Olympiad winner
in 1989. He is the author of Art of Problem Solving’s Introduction to Geometry and Introduction to Algebra
textbooks. Sandor Lehoczky participated in the Math Olympiad Summer Program in 1989, and in 1990 earned
the sole perfect AIME score and led the national first place team on the AHSME (now AMC 12). Lehoczky and
Rusczyk are co-founders of the Mandelbrot Competition.

The Art of Problem Solving Vol 2 Solutions

Uploaded by

Copyright:

Available Formats

The Art of Problem Solving Vol 2 Solutions

Uploaded by

Document Information

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

The Art of Problem Solving Vol 2 Solutions

Uploaded by

Copyright:

Available Formats

What is the book about based on the information provided?

What is the book about based on the information provided?

What are some of the accomplishments of the authors Richard Rusczyk and Sandor Lehoczky?

What are some of the accomplishments of the authors Richard Rusczyk and Sandor Lehoczky?

PHILLIPS ACADEMY

Published by: AoPS Incorporated

Visit the Art of Problem Solving website at http://www.artofproblemsolving. com

Printed in the United States of America.

(logfl b)(logc a) = (logfl fl)(logc b) = logc b,

proving the chain rule for logarithms.

log3 12 = log3(6/5) = loS3 3 + lo§3 2 “ lo§35 = 1 + \ ~

log(l/12) = log(l/6) + log(l/2) = - log6- log2 = - (log 10 - log5) = -- + - - 1.

Make sure you see why log 1/2 = - log 2.

(logx)(logx) = logx3-log 100

logS44 = 44 log 5 = 44(log 10 - log 2) = 44(1 - 0.3010) = 30.756,

Simplifying this, we find log10 5 = (3PQ)/(1 + 3PQ).

12[(l-fl-fc)/2(l-fc)] 22[(l-(log60 3+log60 5))/2(l-log60 5)]

12(loS604)/(log60 602 /52)

Not Just For Right Triangles

sin 300° = - V3/2

2-4 Have fun!

arccsc -1 = 3ti/2 + 2 nn.

2-13 Since sec(270° + x) = 1/ cos(270° + x), we examine

Shifting tan x (dashed line) to the left by 450° (resulting in the

[ABC] [ABD] - [ACD]

Multiplying both sides by cos a cos jS yields

sin(n - jS) = sin a cos jS - sin jS cos a.

2-15 Use the addition formula on a and —/S:

sin[n + (-jS)j = sinncos(-/3) + sin(-jS) cos(n) = sinncosjS - sin cos a,

where we have used sin(-x) = - sinx and cos(-x) - cosx.

2-16 We use the same approach as above:

2-17 Let's try this in terms of sines and cosines:

cos(a -18) cos a cos jS + sin a sin j8 _ cot a cot jS + 1

2-19 Using the half angle formula.

tan - - sin(*/2) _ Vl ~ COS J

2-24 We use our double angle formulas and other identities:

'2sinx - 2 cos2 xsinx /2sinx(l - cos2x) ^—

cos2(80°) + • • • + cos2 (50°) + sin2 (50°) + • • • + sin2(80°) + sin2(90°) = 5,

where we have used sin2 x + cos2 x = 1 four times.

„ 7i x . 2x, x/2 + 2x/3 7x16 7x

1-2 cos 2x + cos2 2x 1-2 cos 2x 1 + cos 4x

cos 10 sin 5 cos 10 cos 5 + sin 10 sin 5

The numerator is the expansion of cos(10 - 5)! Thus our sum is

For the second,

For the third,

8 sin x cos5 x - 8 sin5 x cos x = 8 sin x cos x (cos4 x - sin4 x)

sin2 x COS2X (sin x - cos x)2

(1 + tanA)(l + tanB) = 1 + tan A + tanB + tan A tan B = 1 + 1=2.

31. Since sin 0/2 = V(1 ~ cos 0)/2, we find cos 6 as

tan lEDB + tan LED A

Solving this equation for y gives us our answer.

Since sin37i/5 = sin2n/5, we have 4[ABCDE] = 4sin27r/5 + tan27i/5. In the

[ABCDE] = 10[OCF] = 10(OF)(FC)/2

A AEB ~ A AFC ~ ABIC

Thus, yx - x2 = x3/y, or y2 - yx - x2 = 0. Applying the quadratic formula, we find y = (x ± V5x^)/2 =

sin LDAB - sin 18° - BD/AB = x/2y = (V5 - l)/4.

Similarly, cos LlCD = cos 36° = CD/IC = (x/2)/(x2/y) = y/2x = (1 + V5)/4.

which expanded and rearranged is the law of cosines.

3-3 When angle A is right, BC is the diameter of the circumcircle. Since

™ = sinD = sin(180° - A) = sin A,