SOME HINTS AND ANSWERS (#2)

TO 18.S34 SUPPLEMENTARY PROBLEMS

(FALL 2007)

55. (b) See F. Ardila, The number of halving circles, Amer. Math. Monthly
111 (2004), 586–592.

56. (a) Let p be a prime dividing k. Since n5 − 1 = (n − 1)(n4 + n3 + n2 +

n + 1), we have that p divides n5 − 1, i.e., n5 � 1 (mod p). Also

(n − 1)(n3 + 2n2 + 3n + 4) − (n4 + n3 + n2 + n + 1) = −5.

Thus either p = 5 or p doesn’t divide n−1 [why?]. In other words,

p = 5 or n ≥� 1 (mod p).

Assume that p = ≥ 5. Let t be the least positive integer for which

nt � 1 (mod p). A standard property of congruences states that

if nm � 1 (mod p), then t divides m. Since n5 � 1 (mod p) but
n ≥� 1 (mod p), we have t = 5. Note that n is not divisible by
p since n5 � 1 (mod p). Hence by Fermat’s theorem we have
np−1 � 1 (mod p). Thus p − 1 is divisible by 5, as desired. Since
every prime factor of k is congruent to 0 or 1 (mod 5), the same
is true for k, completing the proof.
(b) This is immediate from (a). Essentially the same argument shows
that there are infinitely many primes of the form qn + 1 for any
fixed prime q. With a little more work one can show there are
infinitely many primes of the form qn + 1 for any q > 1. It is in
fact true that there are infinitely many primes of the form qn + r
for any fixed relatively prime integers q and r, but this is much
harder to prove.

60. (c) The surprising answer is that it is impossible to reach a point with
x-coordinate equal to 5. See R. Honsberger, Mathematical Gems II,
Mathematical Association of America, 1976, Chapter 3.

68. If n is even, then x4 + 4n is even and greater than 2. If n is odd, say

n = 2m + 1, then

n4 + 4n = n4 + 4(2m )4 = (n2 + 2 · 4m − 2m+1 )(n2 + 2 · 4m + 2m+1 ).

1
70. h(n) = 1+ n2 + n4 . For an elegant proof see R. Honsberger, Mathemat­
� � � �

ical Morsels, Mathematical Association of America, 1978, Problem 3.

71. The motion of the fly is not precisely defined by the conditions of the
problem. The fly could be anywhere between the man and the point
x = 0, for if we put the fly in any such position and let time run
backwards, then both the man and the fly end up at x = 0 after one
hour.

72. See J. Borwein and K.-K. S. Choi, On the representations of xy+yz+xz,

Experiment. Math. 9 (2000), 153–158;

http://projecteuclid.org/Dienst/UI/1.0/Summarize/euclid.em/1046889597.

73. Yes. Let w be the Fibonacci word that is the unique fixed point of the
transformation 0 ≡ 01 and 1 ≡ 0. Equivalently, we have w = x1 x2 · · ·
where x1 = 0, x2 = 10, and xi+1 = xi−1 xi for i ∼ 2. Thus

w = 010010100100101001010 · · · .

Then w is not eventually periodic, and every prefix of w of length at

least 6 ends in a square of length at most 5.

79. In general, if a point p is in the interior of an equilateral triangle of size

d and is at distance a, b, c from the vertices, then

3(a4 + b4 + c4 + d4 ) = (a2 + b2 + c2 + d2 )2 .

Note the curious symmetry between a, b, c, d. The symmetry between

a, b, c is obvious, but why also d? (A simple, noncomputational

� reason

→
can be given.) For the case a = 3, b = 4, c = 5, we have d = 25 + 3.
For some references and generalizations, see R. J. Gregorac, A general

3-4-5 puzzle, European J. Combinatorics 17 (1996), 533–541.

80. Four, as shown in the figure below.

 This famous dissection is due to Henry E. Dudeney. A good general ref­

erence to problem of this nature is H. Lindgren, Recreational Problems
in Geometric Dissections and How to Solve Them, revised and enlarged
by G. Frederickson, Dover, New York, 1972. See page 25 for the “rea­
son” behind the figure above. Moreover, on page 9 is a dissection of
a square into five pieces that can be reassembled into two congruent
equilateral triangles. A further reference is Chapter 4 of M. Gardner,
The Unexpected Hanging and Other Mathematical Diversions, Simon
and Schuster, New York, 1969.

81. Hint: Consider the largest power of 2 dividing any of 1, 2, . . . , n.

83. Two proofs may be found in Solution to Problem 11114, Amer. Math.
Monthly 113 (2006), pp. 760–761. The first proof is elementary, while
the second is an elegant argument based on 2-adic analysis.

��
��
� �
�� ���
� �� � �� �� ��
�
� � �� � �
� � � � � �
� �
� �� �� �
92.
p(1 − q)
93. Answer: f (p, q) =
p(1 − q) + q(1 − p)
95. Let u = (1−xy)−1 and v = (1−yx)−1 . Note that (1−yx)y = y(1−xy),

3
 and therefore yu = vy. Thus

(1 + x)v(1 + y) = v + xv + yu + xyu
= v + xv + yu + u − (1 − xy)u
= u + v + xv + yu − 1.

This last expression is symmetric with respect to the permutation (writ­

ten in disjoint cycle form) (x, y)(u, v), so an affirmative answer follows.
This argument is due to S. Fomin.
A vast generalization is due to D. Krob in Topics in invariant theory
(M.-P. Malliavin, ed.), Lecture Notes in Math., vol. 1478, Springer-
Verlag, Berlin/Heidelberg/New York, 1991, pp. 215–243. (A short dis­
cussion also appears in §8 of C. Reutenauer, in Formal Power Series
and Algebraic Combinatorics (New Brunswick, NJ, 1994), DIMACS
Series Discrete Math. Theoret. Comput. Sci. 24, American Mathemat­
ical Society, Providence, RI, 1996, pp. 159–169.) Namely, any identity
in noncommutative variables which holds when formally expanded into
power series continues to hold in any ring for which the identity is
defined.
97. The nth term is the number 16 written in base 17 − n, so the missing
term is 121.
98. These are the positive integers whose spelling begins with the letter t.
100. When we rotate ourselves 180� to compare ourselves with our mirror
image, there is a preferred axis of rotation, namely, the up-down axis.
Rotation about this axis reverses left and right. We can imagine instead
rotating about a line that passes through our two hips, say. In this case,
up and down would be reversed, but not left and right. It is the earth’s
gravity that leads to our preferred axis of rotation.
103. See R. Honsberger, Mathematical Morsels, problem 12, and Ian Stew­
art, Pursuing polygonal privacy, Scientific American, February 2001,
pp. 88–89.
104. In general, if the ladders are of length a and b (with a ∼ b) and the
height of their intersection is c, then their distance x apart is given by

x4 − 2cx3 + (a2 − b2 )x2 − 2c(a2 − b2 )x + c2 (a2 − b2 ) = 0,

4
 Here a = 119, b = 70, c = 30, x = 56. See M. Gardner, Mathematical
Circus, Knopf, New York, 1979, page 62.
105. (a) Let x = (x1 , x2 , . . .). Since 2n and 3n are relatively prime, there
are integers an and bn for which xn = an 2n + bn 3n . Hence f (x) =
f (y)+f (z), where y = (2a1 , 4a2 , 8a3 , . . .) and z = (3b1 , 9b2 , 27b3 , . . .).
Now for any k ∼ 1 we have
f (y) = f (2a1 , 4a2 , . . . , 2k−1 ak−1 , 0, 0, . . .)
+f (0, 0, . . . , 0, 2k ak , 2k+1 ak+1 , . . .)
= 0 + 2k f (0, 0, . . . , 0, ak , 2ak+1 , 4ak+2 , . . .).
Hence f (y) is divisible by 2k for all k ∼ 1, so f (y) = 0. Similarly
f (z) is divisible by 3k for all k ∼ 1, so f (z) = 0. Hence f (x) = 0.
(b) Let ai = f (ei ). Define integers 0 < n1 < n2 < · · · such that for all
k ∼ 1,
k
� 1
|ai |2ni < 2nk+1 .
i=1
2
(Clearly this is possible — once n1 , . . . , nk have been chosen, sim­
ply choose nk+1 sufficiently large.) Consider x = (2n1 , 2n2 , . . .).
Then
f (x) = f (a1 e1 + · · · + ak ek + 2nk+1 (ek+1 + 2nk+2 −nk+1 ek+2 + · · ·))
�k
= ai 2ni + 2nk+1 bk ,
i=1

where bk = f (ek+1 + 2nk+2 −nk+1 ek+2 + · · ·). Thus by the triangle

inequality,
� �
�� k �
|2nk+1 bk | < � ai 2ni � + |f (x)|
� �
� �
i=1
1 nk+1
< 2 + |f (x)|.
2
Thus for sufficiently large k we have bk = 0 [why?]. Since
bj − 2nj+2 −nj+1 bj+1 = f (ej+1 ) [why?],
we have f (ek ) = 0 for k sufficiently large.

5
106. This identity has been verified to over 20,000 decimal digit accuracy.
See J. M. Borwein and D. H. Bailey, Mathematics by Experiment: Plau­
sible Reasoning in the 21st Century, A K Peters, Natick, MA, 2004
(pages 90–91).

109. The smallest n such that the solution x to

3
(3 ·
10n + x) = 10x + 3
2
is an integer is n = 15, yielding

3 · 10n + x = 3529411764705882.

110. (b) � = e1/e

111. For some references and recent work related to this problem (where one
has stamps of value a1 , . . . , an ), see P. Erdős and R. L. Graham, Old
and New Problems in Combinatorial Number Theory, pp. 85–86, and
mathworld.wolfram.com/CoinProblem.html. A interesting recent pa­
per is A. Barvinok and K. Woods, Short rational generating functions
for lattice point problems, www.math.lsa.umich.edu/�barvinok/sem.ps.
→
112. x = 3 − 2 2 = 0.17157287 · · ·

113. The best way to understand this problem is via the theory of ordinal
numbers (which is taught in beginning courses in logic or set theory).
Let � denote the first infinite ordinal (the ordinal number of the se­
quence 1, 2, 3, . . .). In the total bn -ary expansion of an , replace each bn
with �. This defines a certain ordinal number �n . From the definition
of ai it follows immediately that �n+1 is a smaller ordinal number than
�n unless no bn ’s appear in the total bn -ary expansion of an . But a
strictly decreasing sequence of ordinal numbers must be finite, so some
an must have only 1’s in its complete bn -ary expansion (equivalently,
an < bn ), and the proof follows. For some further examples of proce­
dures that unexpectedly terminate, see Chapter 2 of M. Gardner, The
Last Recreations. For further information see

http://en2.wikipedia.org/wiki/Goodstein’s theorem

6
114. An arbitarily large overhang can be achieved. The first appearance
of this result seems to be Problem 3009, American Math. Monthly 30
(1923), 76. For further information and references see

http://mathworld.wolfram.com/BookStackingProblem.html.

115. (knowledge of linear algebra assumed) Let n = #V (G). Let A be the

adjacency matrix of G mod 2, i.e., with entries in the finite field F2 .
Let I denote the n × n identity matrix, and y the n × 1 row vector of
all 1’s. We need to show that y is in the row space of A + I [why?].
Now in general we know from linear algebra that if M is any m × n
matrix and z is any n × 1 row vector, then z is in the row space of M
if and only if zv = 0 whenever Mv = 0 (where v is an n × 1 column
vector). For the special case that z = y, we obtain the following: y is
in the row space of M if and only if there do not exist an odd number
of rows of M whose sum is the 0 vector. Assume then that there are
an odd number of rows of A + I whose sum is 0, say the rows indexed
by vertices v1 , . . . , vk . Let H be the subgraph �k of G induced by the
vertices v1 , . . . , vk . Let A + I = (bij ). Since i=1 bij = 0 for 1 ≈ j ≈ n
and bii = 1, it follows that every vertex of H has odd degree. This is
impossible since H has an odd number of vertices, a contradiction that
completes the proof.
A special case of this problem was open for several years until K. Sutner,
a graduate student at the time, found a proof by linear algebra more
complicated than the above prood. The above proof is due to Yair
Caro, Ars Combinatoria 42 (1996), 175–180.
Note. This problem is equivalent to Problem 10 from the Linear
Algebra and Determinants problem set.
→
116. The farthest distance is 130/4 � 2.8504 · · ·. This problem is due to
Yoshiyuki Kotani. It belongs to the genre known as “Spider and Fly
problems,” as does Problem 91 from Supplementary Problem Set #8.
See Dick Hess, Kotani’s ant problem, in Puzzlers’ Tribute (D. Wolfe
and T. Rogers, eds.), A K Peters, Natick, MA, 2002, pp. 407–411.

118. (e) Suppose to the contrary that every integer can be uniquely written
in exactly one of the forms mi x + ai . By adjusting the value of ai

7
 modulo mi we can assume that 0 ≈ ai < mi . This implies [why?]
the generating function identity
1 z a1 z a2 z ak
= + + · · · + .
1−z 1 − z m1 1 − z m2 1 − z mk
Multiply both sides by (1 − z m1 ) · · · (1 − z mk ), and let � = e2�i/mk ,
a primitive mk th root of unity. The left-hand side then vanishes at
z = �; but since mk is larger than all the other mi ’s the right-hand
side doesn’t vanish at z = �, a contradiction.

119. (a) Answer: 1/(1 − x). One can view this result as the “analytic
form” of the uniqueness of the binary expansion of a nonnegative
integer.
(b) This remarkable result goes back to M. A. Stern in 1858. For fur­
ther information and references, see www.math.uiuc.edu/�reznick/stern.pdf.
A forthcoming paper by Bruce Reznick entitled “A Stern intro­
duction to combinatorial number theory” should be the definitive
reference.
(c) This astonishing result was proved by David Newman in 2002.

121. If m and n are integers and f is a polynomial with integer coeffi­

cients, then it follows from elementary properties of congruences that
f ( m + t ) � f (m) (mod t), where m and t are integers. Let t = f (m)
to get
f (m + f (m)) � f (m) (mod f (m)),
so

f (m + f (m)) � 0 (mod f (m)).

Since f is nonconstant we can find infinitely many values of m for which

|f (m)| > 1 and the numbers m + f (m) are all distinct. Thus f (n) is

composite for the infinitely many distinct values n = m + f (m) for

which | f (m)| > 1.

123. Person A must have a “rule” for deciding what numbers y and y/2
to write down. Such a rule is essentially a probability distribution f
on the positive real numbers. Thus the probability P (a, b) that the
�b
two numbers y and y/2 satisfy a ≈ y ≈ b is P (a, b) = a f (y)dy. (If
some of the probability distribution is discrete, so that the probability

8
 P (y) of writing down y and y/2 is positive, then the integral must be
interpreted as a sum over the discrete part and an integral over the
continuous part.) From the statement of the problem it follows that
f (x) = f (2x) for all x, i.e., it is assumed that it is equally likely that
x and x/2 are written down as x and 2x. However, it is not hard to
show that there does not exist a probability distribution f with this
property.

124. See C. Freiling and D. Rinne, Tiling a square with similar rectangles,
Math. Res. Lett. 1 (1994), 547–558, and M. Laczkovich and G. Szekeres,
Tilings of the square with similar rectangles, Discrete Comput. Geom.
13 (1995), 569–572.

125. See F. Ardila, Fibonacci Quart. 42 (2004), 202–204; math.CO/0409418.

126. Yes. More generally, if m and n are relatively prime positive integers
such that f m and f n are infinitely differentiable, then so is f . This is
a result of Henri Joris in 1982. For a simple proof, see R. Myer, Amer.
Math. Monthly 112 (2005), 829–831.

129. Note that 21 (p + q) is not prime.

130. Answer:
1 � 6
log x + 12x5 + 45x4 + 44x3 − 33x2 + 43
6
4 3 2
→ �
+(x + 10x + 30x + 22x − 11) x + 4x − 6x + 4x + 1 .
4 3 2

See solution to Advanced Problem 5812, American Math. Monthly 79

(1972), 1144–1146.

132. See P. Erdős and R. L. Graham, On packing squares with equal squares,
J. Combinatorial Theory (A) 19 (1975), 119–123. A more recent ref­
erence is E. Friedman, Packing unit squares in squares: a survey and
new results, Elec. J. Combinatorics DS7; available at

http://www.combinatorics.org/Surveys.

133. Hint. Consider the triangle of least altitude formed by any three of
the points that don’t lie on a line.

9
134. This result is due to David Gale and Richard Karp, as a special case
of a more general result appearing in J. Comput. System Sci. 6 (1972),
103–115. One way to prove it is as follows. It suffices to show that
if N is a matrix whose rows are in increasing order, then sorting the
columns into increasing order keeps the rows increasing. Clearly [why?]
we can assume that N has only two columns c1 and c2 . If the entries
of c2 , written in increasing order, are y1 ≈ y2 ≈ · · · ≈ ym , then yi is at
least as large as the i elements in c1 in the same rows as y1 , y2 , . . . , yi .
Hence yi is at least as large as the ith smallest element xi of c1 . But
after we sort the two columns of N the ith row is [xi , yi ], so this row
is increasing. Another reference is pages 184–185 of M. Gardner, The
Last Recreations. For a generalization see B. Tenner, A Non-messing­
up phenomenon for posets, math.CO/0404396.

139. Surprisingly, the proof of this result requires the classification of finite
simple groups! See W. Feit, Some consequences of the classification of
finite simple groups, in The Santa Cruz Conference on Finite Groups,
Proc. Sympos. Pure Math. 37, American Mathematical Society, Prov­
idence, RI, 1980, pp. 175–181.

Solution to #22 of “Problems on Congruence and

Divisibility”

19. No. Note that [why?]

1
an = (a2 + (n − 2)an−1 ), n ∼ 3.
n − 1 n−1
Working modulo 43, it is easy to compute a0 , a1 , . . . , a43 , all mod 43. (Note
that we never have to divide by 43 in the process of computing these num­
bers.) We can then check that a243 + 42 · a43 � 24 (mod 43). Hence a44 is not
an integer (its denominator is divisible by 43). An interesting point about
this problem is that it’s computationally infeasible to compute a44 exactly.
There’s also the interesting question of why p = 43 is the smallest prime for
which the above argument works. For further discussion, see

www-groups.dcs.st-and.ac.uk/�john/Zagier/Solution5.3.html.

10
 Comments on #18 of “Problems on Linear Algebra
and Determinants”

Let us call an integer n satisfying the conditions of the problem per­

missible. Then the permissible integers are precisely 1,2,4, or 8. Suppose
that A is an n-dimensional division algebra over R, not necessarily asso­
ciative or commutative. Thus A is an n-dimensional vector space over R
with a binary multiplication compatible with the vector space structure (i.e.,
�(xy) = (�x)y = x(�y) for � ≤ R, x, y ≤ A) satisfying (1) A has a multi­
plicative identity 1, and (2) every 0 =≥ x ≤ A has a multiplicative inverse y,
i.e., xy = yx = 1. We can regard any element x ≤ A as defining a linear
transformation Tx : A ≡ A via Tx (y) = xy. The set {Tx : x ≤ A} is then an
n-dimensional space of linear transformations on A for which every nonzero
element is invertible. Regarding these linear transformations as matrices
shows that n is permissible.
It is well-known that the above division algebras A exist for n = 1, 2, 4, 8,
namely, R (n = 1), C (n = 2), the quaternions (n = 4), and the octonions or
Cayley numbers (n = 8). Hence these values of n are permissible.
It is a very deep result that there do not exist division algebras A (over
R) of dimensions other than 1, 2, 4, 8. Moreover, it can be shown that
if n is permissible, then there exists an n-dimensional real divison alge­
bra. Hence 1, 2, 4, 8 are the only permissible values of n. See e.g. math-
world.wolfram.com/DivisionAlgebra.html for some references. These results
are proved by connecting them with a third problem, namely, for which n do
there exist n − 1 linear independent nonvanishing vector fields on an (n − 1)­
sphere? (Such spheres are called parallelizable.)
Given any n, let f (n) denote the maximum dimension of a space V of
n × n real matrices such that all nonzero elements of V are invertible. (Thus
f (n) = n if and only if n = 1, 2, 4, 8.) Write n = (2a + 1)2c+4d , where
0 ≈ c ≈ 3. Then Adams proved in Ann. Math. (2) 75 (1962), 603–632 (in
the context of vector fields), that f (n) = 2c + 8d. In particular, f (n) = 1 if
n is odd. A simple proof follows from the fact that if A, B are two nonzero
square matrices of the same odd order, then det(A + xB) is a polynomial in
x of odd degree and therefore has a real zero.

11