section.

HD:Documents:PS:TAACOPS:Art:

Berkeley Math Circle Feb. 20, 2000

Problem Solving for “Beginners”: Hints and

Solutions
Instructor: Paul Zeitz, University of San Francisco ([email protected])

Here are some hints, full solutions, and solution sketches to the problems you
got on Feb. 13, 2000. Some of these problems were quite difficult. Remember:
you don’t have to solve every problem, but you should cultivate a taste for
(and stamina for) investigating problems.
1 If there are n people in the room, after one minute, there will be either n + 1
or n − 2 people. The difference between these two possible outcomes is 3.
Continuing for longer times, we see that
At any fixed time t, all the possible values for the population of the
room differ from one another by multiples of 3.
In 31999 minutes, then, one possible population of the room is just 31999
people (assuming that one person entered each time). This is a multiple of
3, so all the possible populations for the room have to also be multiples of
3. Therefore 31000 + 2 will not be a valid population.
2 This is yet another application of the Handshake Lemma, which says that if
you add the number of handshakes each person in a group shakes, the sum
will be even (since you are double-counting). Let d0 , d1 , d2 , . . . , dn be the
number of doors in room i, where we include “outside” as room 0 (having as
its doors each of the doors in the house that open to the outside). Then by
the same reasoning as in 3.4.7, the sum d0 + d1 + d2 + . . . + dn must be even.
By hypothesis, d1 , d2 , . . . , dn are all even, which forces d0 to be even as well.

3 Experiment, and you will guess that only perfect squares stay closed. This
is due to the parity: the number of divisors of n is odd if and only if n is a
perfect square.

4 Consider the sum of the terms. We have

(a1 − 1) + (a2 − 2) + · · · + (an − n)
= (a1 + a2 + · · · + an ) − (1 + 2 + · · · + n)
= (1 + 2 + · · · + n) − (1 + 2 + · · · + n)
= 0,
so the sum is an invariant; it is equal to zero no matter what the arrangement.
A sum of an odd number of integers which equals zero (an even number) must
contain at least one even number!
5 Let us indicate the population at any time by an ordered triple (x, y, z).
Without loss of generality, consider an X-Z collision. The new population
 BMC Problem Solving for “Beginners”: Hints and Solutions Feb. 20, 2000 2

right, etc. The important thing about this sequence is that P7 ends up on
the left side, along with P1 . Therefore L cannot pass through the interior of
segment P1 P7 , a contradiction.

7 The answer is no. To see why, assume that there is a tiling, and we shall look
for a contradiction. Color the squares of 66 × 62 rectangle with 12 colors in
a cyclic “diagonal” pattern as follows (we are assuming that the height is 66
and the width is 62):
1 12 11 · · · 1 12
2 1 12 · · · 2 1 60 × 60 60 × 2
3 2 1 ··· 3 2
.. .. .. . . .. .. and then look at .
. . . . . .
5 4 3 ··· 5 4
6 5 4 ··· 6 5 6 × 60 6×2

8 It is easy to verify that if (x, y) is a legal point, then y − x will be a multiple

of 11. Since 1999 − 3 is not a multiple of 11, the answer is no.

9 Notice that uv + u + v + 1 = (u + 1)(v + 1), so that u, v are replaced with

(u+1)(v +1)−1. Therefore, if the sequence contains the values a1 , a2 , . . . , an ,
the quantity
(a1 + 1)(a2 + 1)(a3 + 1) · · · (an + 1)
is invariant! Hence the final number will equal 100! − 1, no matter what the
choices are.

10 For example, if n = 6 and the starting sequence was 362154, the cards evolve
as follows:

362154 → 263154 → 623154 → 451326 → 315426

→ 513426 → 243156 → 423156 → 132456.

It would be nice if the number of the card in the 1st place decreased mono-
tonically, but it didn’t (the sequence was 3, 2, 6, 4, 3, 5, 2, 4, 1). Nevertheless,
it is worth thinking about this sequence. We shall make use of a very simple,
but important, general principle:

If there are only finitely many states as something evolves, either

a state will repeat, or the evolution will eventually halt.

In our case, either the sequence of 1st-place numbers repeats (since there are
only finitely many), or eventually the 1st-place number will be 1 (and then
the evolution halts). We would like to prove the latter. How do we exclude
 BMC Problem Solving for “Beginners”: Hints and Solutions Feb. 20, 2000 3

the possibility of repeats? After all, in our example, there were plenty of
repeats!
Once again, the extreme principle saves the day. Since there are only finitely
many possibilities as a sequence evolves, there exists a largest 1st-place value
that ever occurs, which we will call L1 (in the example above, L1 = 6). So,
at some point in the evolution of the sequence, the 1st-place number is L1 ,
and thereafter, no 1st-place number is ever larger than L1 . What happens
immediately after L1 occurs in the 1st place? We reverse the first L1 cards,
so L1 appears in the L1 th place. We know that the 1st-place card can never
be larger than L1 , but can it ever again equal L1 ? The answer is no; as long
as the 1st place value is less than L1 , the reversals will never touch the card
in the L1 th place. We will never reverse more than the first L1 cards (by the
maximality of L1 ), so the only way to get the card numbered L1 to move at
all would be if we reversed exactly L1 places. But that would mean that the
1st-place and L1 th-place cards both had the value L1 , which is impossible.
That was the crux move. We now look at all the 1st-place values that occur
after L1 appeared in the 1st place. These must be strictly less than L1 .
Call the maximum of these values L2 . After L2 appears in 1st place, all
subsequent 1st-place values will be strictly less than L2 by exactly the same
argument as before.
Thus we can define a strictly decreasing sequence of maximum 1st-place
values. Eventually, this sequence must hit 1, and we are done!

11 Let us call any set {x1 , x2 , . . . , x23 } of integers “balanced” if it has the prop-
erty that no matter which of the xi is chosen for the referee, then one can
decompose the remaining 22 numbers into two sets of 11 which have equal
sums. Clearly if a set is balanced and we multiply or divide each element
by the same number, it will still be balanced. Likewise, if a set is balanced
and we add or subtract the same number to each element, it will still be
balanced.
Now, let us suppose we have a balanced set {x1 , x2 , . . . , x23 } of positive in-
tegers. Let S be the sum of the 23 elements. If we pick x1 as referee, then
we know that S − x1 must be even, since the remaining 22 elements can be
partitioned into two sets with equal integer sums. By the same reasoning,
S − x2 , S − x3 , . . . , S − x23 are all even. Therefore, if the set of integers is
balanced, then all the elements x1 , x2 , . . . , x23 are the same parity (i.e., all
are even, or all are odd).
Now consider our balanced set {x1 , x2 , . . . , x23 } of positive integers. We wish
to show that all elements are equal. Let a be the minimum value of the
elements. If we define bi = xi − a for i = 1, 2, . . . , 23, then the new set
{b1 , b2 , . . . , b23 } will also be a balanced set of nonnegative integers. Some of
 BMC Problem Solving for “Beginners”: Hints and Solutions Feb. 20, 2000 4

the elements will be zero, and perhaps some are not. We would like to prove
that they are all zero. Since some of the elements are zero, and zero is even,
then all of the elements must be even. Consequently we can form a new set
{c1 , c2 , . . . , c23 }, where ci = bi /2 for i = 1, 2, . . . , 23. But this set also has
some zero elements, hence all of its elements are even, hence we can divide
them all by 2 and get yet another balanced set of nonnegative integers. We
can do this forever! The only integer which one can divide by 2 endlessly and
still get an even integer as a result is zero. We conclude that the elements
of {b1 , b2 , . . . , b23 } are all zero, i.e., the elements of {x1 , x2 , . . . , x23 } are all
equal.
12 Consider the general problem of n marbles m0 , m1 , . . . , mn−1 with arbitrary
starting locations. Each marble has a “ghost path,” the path it would travel
if it did not bounce off its neighbors but instead passed through them. When-
ever the marbles bounce, the actual path of a marble coincides with another
marble’s ghost path. After one minute has passed, each ghost path has re-
turned to the original positions of each marble. Hence after one minute, the
actual locations of the marbles are a permutation of the original positions.
Moreover, this permutation must be a cyclic permutation, since the marbles
cannot pass through one one another.
We claim that the permutation takes m0 to md , where d is the “counter-
clockwise excess,” i.e. the difference modulo n between the number of coun-
terclockwise marbles and the number of clockwise marbles.
To see this, let vi (t) be the velocity function for marble mi , where the veloc-
ities of +1, −1 denote counterclockwise and clockwise motion, respectively.
Notice that for any time t,
n−1
X
vi (t) = d,
i=0
since the number of clockwise and the number of counterclockwise marbles
never changes (even when marbles collide). There will be finitely many
bounces, and in any time interval between bounces, each velocity function is
a constant. Let t1 , t2 , . . . , tk be time values inside each interval, and let each
each interval have length `i . For each marble mi , denote the net counter-
clockwise distance traveled from t = 0 to t = 1 by
si = vi (t1 )`1 + vi (t2 )`2 + · · · + vi (tk )`k .
Summing this over all marbles, we get
n−1
X
si = d(`1 + `2 + · · · + `k ) = d · 1 = d.
i=0

The only cyclic permutation associated with this sum of net distance traveled
is the one which takes m0 to md .
 BMC Problem Solving for “Beginners”: Hints and Solutions Feb. 20, 2000 5

13 A very sketchy hint: orient the large rectangle so that one corner is a lattice
point, and then consider the parity of the number of corner lattice points.

14 The idea is that triangular packing (so the centers of three circles are vertices
of an equilateral triangle) is a more efficient use of space, however, to do this,
you waste some space at the beginning of the rectangle. So you need a long
rectangle to catch up.

15 Let us work out the first few terms of the product. We get

(1 + x3 )(1 + 2x9 )(1 + 3x27 )(1 + 4x81 ) · · · =

= 1 + x3 + 2x9 + 2x12 + 3x27 + 3x30 + 6x36 + 6x39 + 4x81 + · · · .

What are the (positive) exponents ki ? All integers of the form 3u1 + 3u2 +
· · · + 3ur , where the integers uj satisfy 1 ≤ u1 < u2 < · · · < ur . In other
words, they will be numbers which, when written in base-3, only contain ones
and zeros and end with a zero. In order, the first few exponents are (written
in base-3) are

10, 100, 110, 1000, 1010, 1100, 1110, 10000, . . . .

Of course, these numbers are just the base-2 representations of the sequence
2, 4, 6, 8, . . .. In particular, to figure out k1996 , we just write 2 · 1996 = 3992
in base-2:

3992 = 2048 + 1024 + 512 + 256 + 128 + 16 + 8,

so the base-2 representation of 3994 is 111110011000, and k1996 is equal to

111110011000 (base-3).
In other words,

k1996 = 33 + 34 + 37 + 38 + 39 + 310 + 311 .

16 See Concrete Mathematics by Graham, Knuth and Patashnik for a very nice
discussion of this and related problems.

17 A bit of experimentation convinces us that if n = 3, the total private area

is also equal to the total area of one planet. Playing around with larger
n suggests the same result. We conjecture that the total private area is
always equal exactly to the area of one planet, no matter how the planets
are situated. It appears to be a nasty problem in solid geometry, but must
it be? The notions of “private” and “public” seem to be linked with a sort
of duality; perhaps the problem is really not geometric, but logical . We need
some “notation.” Let us assume that there is a universal coordinate system,
 BMC Problem Solving for “Beginners”: Hints and Solutions Feb. 20, 2000 6

such as longitude and latitude, so that we can refer to the “same” location
on any planet. For example, if the planets were little balls floating in a room,
the location “north pole” would mean the point on a planet which was closest
to the ceiling.
Given such a universal coordinate system, what can we say about a planet
P which has a private point at location x? Without loss of generality, let x
be at the “north pole.” Clearly, the centers of all the other planets must lie
on the south side of the P ’s “equatorial” plane. But that renders the north
poles of these planets public, for their north poles are visible from a point in
the southern hemisphere of P (or from the southern hemisphere of an planet
that lies between). In other words, we have shown pretty easily that

If location x is private on one planet, it is public on all the other

planets.

After this nice discovery, the penultimate step is clear: to prove that

Given any location x, it must be private on some planet.

We leave this as an exercise (problem?) for you!

18 Let R be the interior of the rectangle with vertices (0, 0), (b, 0), (b, a), (0, a).
The line y = ax/b intersects no lattice points in R (it passes through (0, 0)
and (b, a), but these points are not included in R, and there are no other
lattice points on the line, since a and b have no common divisors). Observe
that bai/bc is just the number of lattice points that lie below this line in
R for x = i. Thus the left-hand sum is just the number of lattice points
lying below the line in R . By similar reasoning, the right-hand sum is equal
to the number of lattice points lying above the line. The common value
must equal one-half of the total number of lattice points, which is of course
(a−1)(b−1). Observe that at least one of a and b must be odd, for otherwise
the two numbers would share a common divisor (namely, 2). Consequently
(a − 1)(b − 1) is even and can be divided by 2.

19 Assume that the values are not all equal. Let a > 0 be the smallest value on
the board. There must be a square containing a which is adjacent (WLOG)
on the east by a square containing the value b which is strictly greater than
a. But then a is equal to the average of 4 numbers, none less than a, one of
which is strictly greater than a. This is a contradiction.

20 This is very similar to the problem above.

21 The coin with smallest diameter cannot be tangent to more than 5 others.
 BMC Problem Solving for “Beginners”: Hints and Solutions Feb. 20, 2000 7

22 Certainly when n is even, it is not true: Just imagine a set of pairs of people
standing a few inches apart, with each pair quite far from every other pair.
Now, if n is odd, first eliminate all pairs as in the above case, where two
people end up shooting each other. Since n is odd, some dry people remain.
Now consider the person whose nearest neighbor is maximal (there may be
ties). This person will stay dry, since the only way that he could get shot
is if someone else is as close to him as he is to his nearest neighbor. But
that contradicts the fact that for each person, the distances to the others are
different.

23 g(n) = 2r , where r is the number of 1’s in the base-2 representation of n.

This can be proven with induction, as well as many other methods.

24 Let there be n people. Each person is seated a distance d from his or her
correct place, where 0 < d < n is measured counterclockwise. There are n
people, but n − 1 different values of d. Hence at least two people share the
same distance d.

25 The bug should travel along two line segments: first from (7, 11) to O =
(0, 0), and then from O to (−17, −3). This is a consequence of the following
principle: the bug must avoid quandrant II completely, even though a straight
line path from (7, 11) to (−17, −3) goes through quadrant II.
To see why this is true, let a and b be arbitrary positive numbers, and
consider a path starting at A = (0, a) and ending at B = (−b, 0). Certainly
√ within quadrant II is the line segment AB, and the length
the quickest route
of this path is a2 + b2 . Now consider the alternate route AO followed by
OB. This path lies outside quadrant II (since quadrant II does not include
the x- or y-axes) and has total length a + b. Compare these two lengths.
By the arithmetic-geometric mean inequality, we have a2 + b2 ≥ 2ab, which
implies that 2a2 + 2b2 ≥ a2 + 2ab + b2 = (a + b)2 . Hence
√ √
a + b ≤ 2 a2 + b 2 .

We conclude that as long as the speed in quadrant II is less than √12 , then any
path from A to B that passes through quadrant II will take more time than
the shortest non-quadrant-II path (along the y- and x-axes). Since 12 < √12 ,
our bug will save time by avoiding quadrant II.

26 This problem is rather tricky unless we start by considering the 2-dimensional

case. A bit of playing around convinces us that 5 is the magic number: If
5 lattices points are chosen in the plane (all distinct, of course), then one
of the line segments joining two of these points will have a lattice point in
the interior. The key ideas are parity and pigeonhole. There are only four
distinct parity types for lattice points: (odd, odd), (odd, even), (even, odd),
 BMC Problem Solving for “Beginners”: Hints and Solutions Feb. 20, 2000 8

and (even, even). Hence among any 5 distinct lattice points, two must be
of the same parity type, which means that the midpoint of the line segment
joining them is a lattice point! The argument adapts easily to 3-dimensions.

27 Let triangle ABC have the largest area among all triangles whose vertices
are taken from the given set of points. Let [ABC] denote the area of triangle
ABC. Then [ABC] ≤ 1. Let triangle LM N be the triangle whose medial
triangle is ABC. (In other words, A, B, C are the midpoints of the sides of
triangle LM N . See figure.)

medialt riangle.epsscaled500 Then [LM N ] = 4[ABC] ≤ 4. We claim that

the set of points must lie on the boundary or in the interior of LM N . Suppose
a point P lies outside LM N . Then we can connect P with two of the vertices
of ABC forming a triangle with larger area than ABC, contradicting the
maximality of [ABC].

28 We will show that no palindrome can exist by contradiction. Assume that

the concatenation of the numbers from 1 to n was the palindrome

P := 1234567891011 · · · 4321.

Consider the longest run of consecutive zeros in P ; note that this exists,
since n is surely greater than 10. There may be several runs of consecutive
zeros that are all equally long; pick the last (rightmost) one. Observe that
immediately to the left of this string is a single digit, and this digit plus
the zeros forms one of the numbers from 1 to n. For concreteness, suppose
that the longest string of zeros was 0000. Then the rightmost such string
obviously consists of the last digits of one of the numbers from 1 to n, not
the middle of one, and doesn’t straddle two (for example, if the number was,
say, 400005, then the number 400000 would have appeared to the left of
it, contradicting the fact that 0000 is longest string of zeros. Likewise, the
number that ends with 0000 had to start with a single digit, for if, say, the
number was 7310000 then there would have been the number 7000000 to the
left of it.
So, let us suppose that the rightmost string of 0000 is the last digits of the
number 70000. Then, writing the predecessor and successor numbers, these
four zeros are embedded in the string 699997000070001 Assume also, that
there is at least one other string of 0000 in P . Since P is a palindrome, the
first 0000 must be embedded in the string 100070000799996. But that makes
no sense, since the first time 0000 appears is as the last digits of the number
10000.
 BMC Problem Solving for “Beginners”: Hints and Solutions Feb. 20, 2000 9

So the only remaining possibility is that there is only one 0000 string, which
by necessity is at the exact center of P and is the last four digits of the
number 10000. Writing the predecessor and successor, and letting “|” mark
the exact midpoint of P , we must have the following string at the center:

· · · 9999100|0010001 · · ·

But this isn’t symmetrical (9 6= 0), achieving our contradiction.

29 Consider the shortest path joining 1 with n2 , where path means a walk along
adjacent squares. The worst case scenario is that the path has length n (if
1 and n2 are at opposite diagonal corners). In any event, the members of
the path will be at most n distinct numbers between 1 and n2 , inclusive. If
their successive differences were all less than or equal to n, that means there
are n − 1 successive differences which bridge the gap from 1 to n2 . Since
n2 − 1 = (n − 1)(n + 1), the largest difference must be at least n + 1.

30 A very brief hint: show that eventually, the sequence will form a chain where
each element will divide the next (when arranged in order). Moreover, the
least element and the greatest element of this chain are respectively the great-
est common divisor and least common multiple of all the original numbers.

31 Hint: look at diagonals.