THE UNIVERSITY OF TORONTO

UNDERGRADUATE MATHEMATICS COMPETITION

March 12, 2006

Time: 3 21 hours
No aids or calculators permitted.
It is not necessary to do all the problems. Complete solutions to fewer problems are preferred to partial
solutions to many.

1. (a) Suppose that a 6 × 6 square grid of unit squares (chessboard) is tiled by 1 × 2 rectangles (dominoes).
Prove that it can be decomposed into two rectangles, tiled by disjoint subsets of the dominoes.
(b) Is the same thing true for an 8 × 8 array?

2. Let u be a unit vector in R3 and define the operator P by P (x) = u × x for x ∈ R3 (where × denotes
the cross product).
(a) Describe the operator I + P 2 .
(b) Describe the action of the operator I + (sin θ)P + (1 − cos θ)P 2 .

3. Let p(x) be a polynomials of positive degree n with n distinct real roots a1 < a2 < · · · < an . Let b be a
real number for which 2b < a1 + a2 . Prove that

2n−1 |p(b)| ≥ |p0 (a1 )(b − a1 )| .

4. Two parabolas have parallel axes and intersect in two points. Prove that their common chord bisects
the segments whose endpoints are the points of contact of their common tangent.

5. Suppose that you have a 3 × 3 grid of squares. A line is a set of three squares in the same row, the same
column or the same diagonal; thus, there are eight lines.
Two players A and B play a game. They take alternate turns, A putting a 0 in any unoccupied square
of the grid and B putting a 1. The first player is A, and the game cannot go on for more than nine
moves. (The play is similar to noughts-and-crosses, or tictactoe.) A move is legitimate if it does not
result in two lines of squares being filled in with different sums. The winner is the last player to make
a legitimate move.
(For example, if there are three 0s down the diagonal, then B can place a 1 in any vacant square provided
it completes no other line, for then the sum would differ from the diagonal sum. If there are two zeros
at the top of the main diagonal and two ones at the left of the bottom line, then the lower right square
cannot be filled by either player, as it would result in two lines with different sums.)

(a) What is the maximum number of legitimate moves possible in a game?

(b) What is the minimum number of legitimate moves possible in a game that would not leave a
legitimate move available for the next player?
(c) Which player has a winning strategy? Explain.

6. Suppose that k is a positive integer and that

f (t) = a1 eλ1 t + a2 eλ2 t + · · · + ak eλk t

where a1 , · · · , ak , λ1 , · · · , λk are real numbers with λ1 < λ2 < · · · < λk . Prove that f (t) = 0 has finitely
many real solutions. What is the maximum number of solutions possible, as a function of k?

1
 7. Let A be a real 3 × 3 invertible matrix for which the sums of the rows, columns and two diagonals are
all equal. Prove that the rows, columns and diagonal sums of A−1 are all equal.

8. Let f (x) be a real function defined and twice differentiable on an open interval containing [−1, 1].
Suppose that 0 < α ≤ γ and that |f (x)| ≤ α and |f 00 (x)| ≤ γ for −1 ≤ x ≤ 1. Prove that
√
|f 0 (x)| ≤ 2 αγ

for −1 ≤ x ≤ 1. (Part marks are possible for the weaker inequality |f 0 (x)| ≤ α + γ.)

9. A high school student asked to solve the surd equation

√ √
3x − 2 − 2x − 3 = 1

gave the following answer: Squaring both sides leads to

3x − 2 − 2x − 3 = 1

so x = 6. The answer is, in fact, correct.

Show that there are infinitely many real quadruples (a, b, c, d) for which this method leads to a correct
solution of the surd equation √ √
ax − b − cx − d = 1 .

10. Let P be a planar polygon that is not convex. The vertices can be classified as either convex or concave
according as to whether the angle at the vertex is less than or greater than 180◦ respectively. There
must be at least two convex vertices. Select two consecutive convex vertices (i.e., two interior angles
less than 180◦ for which all interior angles in between exceed 180◦ ) and join them by a segment. Reflect
the edges between these two convex angles in the segment to form along with the other edges of P a
polygon P1 . If P1 is not convex, repeat the process, reflecting some of the edges of P1 in a segment
joining two consecutive convex vertices, to form a polygon P2 . Repeat the process. Prove that, after a
finite number of steps, we arrive at a polygon Pn that is convex.

2
 Solutions.

1. (a) Suppose that a 6 × 6 square grid of unit squares (chessboard) is tiled by 1 × 2 rectangles (dominoes).
Prove that it can be decomposed into two rectangles, tiled by disjoint subsets of the dominoes.
(b) Is the same thing true for an 8 × 8 array?

Solution. (a) There are 18 dominoes and 10 interior lines in the grid. For the decomposition not to
occur, each of the lines must be straddled by at least one domino. We argue that, in fact, at least two
dominos must straddle each line. Since no domino can straddle more than one line, this would require 20
dominos and so yield a contradiction.
Each interior line has six segments. For a line next to the side of the square grid, an adjacent domino
between it and the side must either cross one segment or be adjacent to two segments. Since the number of
segments is even, evenly many dominos must cross a segment. For the next line in, an adjacent domino must
be adjacent to two segments, be adjacent to one segment and cross the previous line, or cross one segment.
Since the number of dominoes straddling the previous line is even, there must be evenly many that cross the
segment. In this way, we can work our way from one line to the next.

(b) Number the squares in the grid by pairs ij of digits where the square is in the ith row and jth
column. Here is a tiling with dominos in which each interior line is straddled and no decomposition into
subrectangles is possible:

(11 − 12), (13 − 14), (15 − 16), (17 − 27), (18 − 28), (21 − 31), (22 − 32), (23 − 33)

(24 − 34), (25 − 26), (35 − 45), (36 − 46), (37 − 38), (41 − 42), (43 − 44), (47 − 57)
(48 − 58), (51 − 61), (52 − 53), (54 − 55), (56 − 66), (62 − 72), (63 − 73), (64 − 74)
(65 − 75), (67 − 68), (71 − 81), (76 − 77), (78 − 88), (82 − 83), (84 − 85), (86 − 87) .

2. Let u be a unit vector in R3 and define the operator P by P (x) = u × x for x ∈ R3 (where × denotes
the cross product).
(a) Describe the operator I + P 2 .
(b) Describe the action of the operator I + (sin θ)P + (1 − cos θ)P 2 .

Solution 1. Recall that a × (b × c) = (a · c)b − (a · b)c for any 3-vectors a, b, c.

(a) We have that

P 2 (x) = u × (u × x) = (u · x)u − (u · u)x = (u · x)u − x

from which we see that (I + P 2 )(x) = (u · x)u. Thus, I + P 2 is an orthogonal projection onto the one-
dimensional spaces spanned by {u}.
(b) Let R = I + (sin θ)P + (1 − cos θ)P 2 . Observe that P (u) = o. Then R(u) = u, so that the
one-dimensional space spanned by {u} is invariant under R. Let v be a unit vector orthogonal to u and
w = u × v. Then {u, v, w} is an orthonormal basis for R3 . We have that

R(v) = v + (sin θ)w + (1 − cos θ)(−v) = (cos θ)v + (sin θ)w

and
R(w) = w + (sin θ)(u × w) + (1 − cos θ)(−w) = (− sin θ)v + (cos θ)w .
It follows that the plane through o orthogonal to u is rotated about the axis u through an angle θ; by
linearity, the same holds for all of R3 .

3
 Solution 2. Wolog, we may assume that u = (1, 0, 0). Let x = (x, y, z). The u × x = (0, −z, y) and
u × (u × x) = (0, −y, −z). Hence (I + P 2 )x = (x, 0, 0) and

(I + (sin θ)P + (1 − cos θ)P 2 )x = (x, 0, 0) + (0, (cos θ)y − (sin θ)z, (sin θ)y + (cos θ)z) .

Thus, I + P 2 is an orthogonal projection onto u and I + (sin θ)P + (1 − cos θ)P 2 is a rotation about the axis
u through the angle θ.

Solution 3. Let v be the image of u after a 90◦ rotation and let w = u×v. Then u, v, w is an orthonormal
basis of R3 and u × w = −v. Hence, if x = xu + yv + xw, then we find that (I + P 2 )x = xu = (x · u)u
and (I + (sin θ)P + (1 − cos θ)P 2 )(x) = xu + [(cos θ)y − (sin θ)z]v + [(cos θ)z + (sin θ)y]w, and we reach the
same conclusion as before.

3. Let p(x) be a polynomials of positive degree n with n distinct real roots a1 < a2 < · · · < an . Let b be a
real number for which 2b < a1 + a2 . Prove that

2n−1 |p(b)| ≥ |p0 (a1 )(b − a1 )| .

Solution. Wolog, let p(x) have leading coefficient 1. Observe that, for i > 1,

ai − a1 = ai − b + b − a1 < ai − b + a2 − b < 2(ai − b)

(x − ai ), from which p0 (a1 ) =

Q Q
and that p(x) = i≥2 (a1 − ai ). Then
Y
|p(b)| = |b − a1 | |b − ai |
i≥2
Y1
≥ |b − a1 | (ai − a1 ) = |b − a1 ||p0 (ai )|2−(n−1) .
2
i≥2

4. Two parabolas have parallel axes and intersect in two points. Prove that their common chord bisects
the segments whose endpoints are the points of contact of their common tangent.

Solution 1. Wolog, we may assume that the parabolas have the equations y = ax2 and y = b(x − 1)2 + c.
The common chord has equation

0 = a[b(x − 1)2 + c − y] − b[ax2 − y] = 0 ,

or
(a − b)y + 2abx − a(b + c) = 0 . (1)

Consider a point (u, au2 ) on the first parabola. The tangent at this point has equation y = 2aux − au2 .
The abscissa of the intersection point of this tangent with the parabola of equation y = b(x − 1)2 + c is given
by the question
bx2 − 2(b + au)x + (au2 + b + c) = 0 .
This has coincident roots if and only if

(b + au)2 = b(au2 + b + c) ⇐⇒ a(a − b)u2 + 2abu − bc = 0 . (2)

In this situation, the coincident roots are x = 1 + (au)/b and the point of contact of the common tangent
with the second parabola is
au a2 u2
 
1+ , +c .
b b

4
The midpoint of the segment joining the two contact points is

b + au + bu abu2 + a2 u2 + bc
 
, .
2b 2b

Plugging this into the left side of (1) and using (2) yields that

[1/(2b)][(a − b)a(a + b)u2 + (a − b)bc + 2ab2 + 2ab(a + b)u − 2ab(b + c)]

= [(a + b)/(2b)][a(a − b)u2 + 2abu − bc] = 0 .

Thus, the coordinates of the midpoint of the segment satisfy (1) and the result follows.

Solution 2. [A. Feizmohammadi] Let the two parabolas have equations y = ax(x − u) and y = bx(x − v).
Since the two parabolas must open the same way for the situation to occur, wolog, we may suppose that
a, b > 0. The parabolas intersect at the points (0, 0) and ((au − bv)/(a − b), (ab(au − bv)(u − v)/(a − b)2 ),
and the common chord has equation (a − b)y − ab(u − v)x = 0.
Let y = mx+k be the equation of the common tangent. Then both of the equations ax2 −(au+m)x−k =
0 and bx2 − (bv + m)x − k = 0 have double roots. Therefore (au + m)2 + 4ak = (bv + m)2 + 4bk = 0, from
which (by eliminating k),
ab(au2 − bv 2 ) + 2ab(u − v)m + (b − a)m2 = 0 .
The common tangent of equation y = mx + k touches the first parabola at

au + m m2 − a2 u2
 
,
2a 4a

and the second parabola at

bv + m m2 − b2 v 2
 
, .
2b 4b
The midpoint of the segment joining these two points is

ab(u + v) + (a + b)m (a + b)m2 − ab(au2 + bv 2 )

 
, .
4ab 8ab

Using these coordinates as the values of x and y, we find that

8ab[(a − b)y − ab(u − v)x] = (a − b)[(a + b)m2 − ab(au2 + bv 2 )] − 2a2 b2 (u − v)(u + v)

+ 2ab(a + b)(u − v)m
= (a − b2 )m2 − 2ab(a + b)(u − v)m
2

− [(a − b)a2 bu2 + (a − b)ab2 v 2 + 2a2 b2 u2 − 2a2 b2 v 2 ]

= (a2 − b2 )m2 − 2ab(a + b)(u − v)m − [a3 bu2 + a2 b2 u2 − a2 b2 v 2 − ab3 v 2 ]
= (a + b)[(a − b)m2 − 2ab(u − v)m] − [ab(a(a + b)u2 − ab(b(a + b)v 2 ]
= (a + b)[(a − b)m2 − 2ab(u − v)m − ab(au2 − bv 2 )] = 0 .

5. Suppose that you have a 3 × 3 grid of squares. A line is a set of three squares in the same row, the same
column or the same diagonal; thus, there are eight lines.
Two players A and B play a game. They take alternate turns, A putting a 0 in any unoccupied square
of the grid and B putting a 1. The first player is A, and the game cannot go on for more than nine
moves. (The play is similar to noughts-and-crosses, or tictactoe.) A move is legitimate if it does not

5
 result in two lines of squares being filled in with different sums. The winner is the last player to make
a legitimate move.
(For example, if there are three 0s down the diagonal, then B can place a 1 in any square provided it
completes no other line, for then the sum would differ from the diagonal sum. If there are two zeros at
the top of the main diagonal and two ones at the left of the bottom line, then the lower right square
cannot be filled by either player, as it would result in two lines with different sums.)

(a) What is the maximum number of legitimate moves possible in a game?

(b) What is the minimum number of legitimate moves possible in a game that would not leave a
legitimate move available for the next player?
(c) Which player has a winning strategy? Explain.

Solution. (a) A game cannot continue to nine moves. Otherwise, the line sum must be three times the
value on the centre square of the grid (why?) and so must be 0 or 3. But some line must contain both zeros
and ones, yielding a contradiction. [An alternative argument is that, if the array is filled, not all the rows
can have the same numbers of 0s and 1s, and therefore cannot have the same sums.] However, an 8-move
grid is possible, in which one player selects the corner squares and the other the squares in the middle of the
edges.
(b) Consider any game after four moves have occurred and it is A’s turn to play a zero. Suppose, first
of all, that no lines have been filled with numbers. The only way an inaccessible square can occur is if it
is the intersection of two lines each having the other two squares filled in. This can happen in at most one
way. So A would have at least four possible squares to fill in. On the other hand, if three of the first four
moves complete a line, the fourth number can bar at most three squares for A in the three lines determined
by the fourth number and one of the other three. Thus, A would have at least two possible positions to fill.
Thus, a game must go to at least five moves.
A five-move game can be obtained when A has placed three 0’s down the left column and B has 1 in the
centre square and another square of the middle column. Each remaining position is closed to B as it would
complete a line whose sum is not 0.
(c) A has a winning strategy. Let A begin by putting 0 in the centre square. After three moves, A can
guarantee either a row, column or diagonal with two 0’s and one 1. Suppose that we have 0, 0, 1 down a
diagonal. Wherever B puts his 1, A can complete the line through the centre containing it. For each of B’s
responses, A can make a move that stymies B. Suppose that we have 1, 0, 0 down a column through the
centre parallel to a side of the grid. If B puts 1 in a row with a 0, then A can complete that row. Whatever
B does next, A has a move that will stymie him. If B puts 1 in a row with another 1, A puts a 0 at the
other end of the diagonal. Whatever B does next, A has a move that will stymie him.

6. Suppose that k is a positive integer and that

f (t) = a1 eλ1 t + a2 eλ2 t + · · · + ak eλk t

where a1 , · · · , ak , λ1 , · · · , λk are real numbers with λ1 < λ2 < · · · < λk . Prove that f (t) = 0 has finitely
many real solutions. What is the maximum number of solutions possible as a function of k?

Solution. We prove by induction that the equation can have at most k − 1 solutions. It holds for k = 1.
Suppose that it holds up to k, and that a1 eλ1 t + · · · + ak+1 eλk+1 t vanishes for m distinct values of t. Then

g(t) ≡ a1 e(λ1 −λk+1 )t + a2 e(λ2 −λk+1 )t + · · · + ak e(λk −λk+1 )t + ak+1

has the same roots, so that, by Rolle’s theorem,

g 0 (t) ≡ a1 (λ1 − λk+1 )e(λ1 −λk+1 ) t + a2 (λ2 − λk+1 )e(λ2 −λk+1 )t + · · · + ak (λk − λk+1 )e(λk −λk+1 )t

6
has m − 1 distinct roots. Hence m − 1 ≤ k − 1 and m ≤ k.
For k ≥ 2, let pk (x) = (x − 1)(x − 2)(x − 3) · · · (x − k − 1) ≡ a1 + a2 x + · · · + ak xk−1 . Then fk (t) =
e pk (et ) = a1 et + · · · + ak ekt has k − 1 roots, namely t = log i (1 ≤ i ≤ k − 1). Hence k − 1 roots is possible.
t

7. Let A be a real 3 × 3 invertible matrix for which the sums of the rows, columns and two diagonals are
all equal. Prove that the rows, columns and diagonal sums of A−1 are all equal.

Solution 1. Let r be the common sum and let e = (1, 1, 1)|bf t . Then Ae = re. Since A is invertible,
r 6= 0 and A−1 e = r−1 e. The other eigenvalues (counting repetitions) of A are u and −u, since the trace of
A is equal to r. Since A is invertible, u 6= 0.
The eigenvalues of A−1 are r−1 , u−1 and −u−1 , and the trace of A−1 is equal to r−1 + u−1 − u−1 = r−1 .
Since also A−1 e = r−1 e, the row and main diagonal sums are r−1 .
Since (A−1 )t = (At )−1 , we can apply the same reasoning to At to find that the row sums of At and,
hence, the column sums of A−1 are equal to r−1 .
 
0 0 1
Finally, let B =  0 1 0 . Then B −1 = A−1 P and we can apply the result to B, which has the
1 0 0
same row, column and diagonal sums as A. The main diagonals of B and B −1 are respectively equal to the
cross diagonals of A and A−1 , so we find that the cross diagonal of A−1 is equal to r−1 . The desired result
follows.

Solution 2. Since the common sum of the rows, columns and diagonals of a magic square is three times
the central entry, the general form of the matrix A is
 
a 3c − a − b b
c + b − a c c + a − b .
2c − b a+b−c 2c − a
The adjugate, adj A of this matrix is
 2 
3c − a2 + b2 − ac − 2bc b2 − a2 − 6c2 + 5ac + bc 3c2 + b2 − a2 + 2ac − 5bc
 5ac − 5bc + b2 − a2 b2 − a2 + 2ac − 2bc b2 − a2 − ac + bc  .
2 2 2 2 2 2 2 2 2
b − a − 3c + 2ac + bc 6c + b − a − ac − 5bc b − a − 3c + 5ac − 2bc
The sum of each row, column and diagonal of adj A is 3(b − a)(b + a − 2c). Since (det A)A−1 = adj A, the
desired result follows.

Comment. A direct solution for the rows and columns can be obtained from the relationships between
the elements of A and A−1 . Let A = (aij ) and A−1 = (bij ). Then, for 1 ≤ i, k ≤ 3,
3
X
aij bjk = δik ,
i=1

where δij is the Kronecker delta (1 if its subscripts agree, 0 otherwise). Let the common sums for A be s.
Then
X3 X 3 X 3 3 X
X 3 3
X
s bjk = ( aij )bjk = ( aij bjk ) = δik = 1 ,
j=1 j=1 i=1 i=1 j=1 i=1
P3
whence s 6= 0 and j=1 bjk = 1/s. Hence, each column sum of A−1 is 1/s. A similar argument holds for
the row sums.

8. Let f (x) be a real function defined and twice differentiable on an open interval containing [−1, 1].
Suppose that 0 < α ≤ γ and that |f (x)| ≤ α and |f 00 (x)| ≤ γ for −1 ≤ x ≤ 1. Prove that
√
|f 0 (x)| ≤ 2 αγ

7
 for −1 ≤ x ≤ 1. (Part marks are possible for the weaker inequality |f 0 (x)| ≤ α + γ.)

Solution. Let |f 0 (x)| attain its maximum value on [−1, 1] at x = u. Then, for −1 ≤ a ≤ u ≤ b ≤ 1, we
have, by Taylor’s theorem, for some v, w in [a, b],

1
f (a) − f (u) = (a − u)f 0 (u) + (a − u)2 f 00 (v) ,
2
1
f (b) − f (u) = (b − u)f 0 (u) + (b − u)2 f 00 (w) .
2
Therefore
1
(b − a)f 0 (u) = f (b) − f (a) + [(a − u)2 f 00 (v) − (b − u)2 f 00 (w)]
2
so that
1
(b − a)|f 0 (u)| ≤ |f (a)| + |f (b)| + [(b − u)2 |f 00 (w)| + (a − u)2 |f 00 (v)|]
2
γ γ
≤ 2α + [(b − u) + (a − u)2 ] ≤ 2α + (b − a)2 ,
2
2 2
since
(b − a)2 − (b − u)2 − (a − u)2 = 2(b − u)(u − a) ≥ 0 .
p
Select a and b so that b − a = 2 α/γ. Then
r 
√
r
0 γ α
|f (u)| ≤ (α) + γ = 2 αγ .
α γ

Comment. If a = −1 and b = 1, then we get in the above that 2|f 0 (u)| ≤ 2α + 2γ.

9. A high school student asked to solve the surd equation

√ √
3x − 2 − 2x − 3 = 1

gave the following answer: Squaring both sides leads to

3x − 2 − 2x − 3 = 1

so x = 6. The answer is, in fact, correct.

Show that there are infinitely many real quadruples (a, b, c, d) for which this method leads to a correct
solution of the surd equation √ √
ax − b − cx − d = 1 .

Solution 1. Solving the general equation properly leads to

√ √ √
ax − b − cx − d = 1 =⇒ ax − b = 1 + cx − d − 2 cx − d
√
=⇒ (a − c)x = (b + 1 − d) − 2 cx − d .
To make the manipulation simpler, specialize to a = c + 1 and d = b + 1. Then the equation becomes

x2 = 4(cx − d) =⇒ 0 = x2 − 4cx + 4d .

Using the student’s “method” to solve the same equation gives ax − b − cx − d = 1 which yields
x = (1 + b + d)/(a − c) = 2d. So, for the “method” to work, we need

0 = 4d2 − 8cd + 4d = 4d(d − 2c + 1)

8
which can be achieved by making 2c = d + 1. So we can take

(a, b, c, d) = (t + 1, 2(t − 1), t, 2t − 1)

for some real t. The original problem corresponds to t = 2.

The equation p p
(t + 1)x − 2(t − 1) − tx − (2t − 1) = 1
is satisfied by x = 2 and x = 4t − 2. The first solution works for all values of t, while the second is valid if
and only if t ≥ 21 . The equation (t + 1)x − 2(t − 1) − tx − (2t − 1) = 1 is equivalent to x = 4t − 2.

Solution 2. [G. Goldstein] Analysis. We want to solve simultaneously the equations

√ √
ax − b − cx − d = 1 (1)

and
ax − b − cx − d = 1 . (2)
From (1), we find that √
ax − b = 1 + (cx − d) + 2 cx − d . (3)
√
From (2) and (3), we obtain that d = cx − d, so that x = (d2 + d)/c. From (2), we have that x =
(1 + b + d)/(a − c).
Select a, c, d so that d > 0 and ac(a − c) 6= 0, and choose b to satisfy

d2 + d 1+b+d
= .
c a−c
Let
d2 + d 1+b+d (d2 + d) + (1 + b + d) (d + 1)2 + b
x= = = = .
c a−c c + (a − c) a
Then √ √ p √
ax − b − cx − d = (d + 1)2 − d2 = (d + 1) − d = 1
and
ax − b − cx − d = (d + 1)2 + b − b − d2 − d − d = 1 .

Comments. In Solution 2, if we take c = d = 1, we get the family of parameters (a, b, c, d) = (a, 2a −

4, 1, 1). R. Barrington Leigh found the set of parameters given in Solution 1. A. Feizmohammadi provided
the parameters (a, b, c, d) = (2c, 0, c, 1).

10. Let P be a planar polygon that is not convex. The vertices can be classified as either convex or concave
according as to whether the angle at the vertex is less than or greater than 180◦ respectively. There
must be at least two convex vertices. Select two consecutive convex vertices (i.e., two interior angles
less than 180◦ for which all interior angles in between exceed 180◦ ) and join them by a segment. Reflect
the edges between these two convex angles in the segment to form along with the other edges of P a
polygon P1 . If P1 is not convex, repeat the process, reflecting some of the edges of P1 in a segment
joining two consecutive convex vertices, to form a polygon P2 . Repeat the process. Prove that, after a
finite number of steps, we arrive at a polygon Pn that is convex.

Solution. Note that for the problem to work, we need to assume that, at each stage, we create a polygon
that does not cross itself. Suppose that P is a m−gon. Then each Pi has at most m sides, as some angles
of 180◦ may be created at the vertices. Also, the perimeter of Pi is the same as that of P for each index
i. Also P ⊆ P1 ⊆ P2 ⊆ · · · ⊆ Pi ⊆ · · ·. Therefore there is a disc that contains all of the polygons Pi (the
perimeter is at least as long as any line segment contained within the polygon).

9
 Let a be a vertex of P , and let ai be the corresponding vertex obtained from a in Pi . each ai being
equal to its predecessor or obtained from it by a reflection that takes it outside of Pi−1 . In the latter case,
ai must lie outside of Pj when j < i. Because the sequence {ai } is bounded, it must have an accumulation
point a0 .
Let  > 0 be small enough that an −neighbourhood about a excludes all other vertices of P . Consider
an −neighbourhood of a0 . It contains a point al . If al+1 6∈ U , then al+1 is the reflection of al in a line
joining two other vertices of Pl and lying on the opposite side of the centre of U to al , and so U ⊆ Pl+1 .
This entails that ak 6∈ U for k ≥ l + 1, contradicting that a0 is an accumulation point. Hence ak ∈ U for
k ≥ l. It follows that limi→∞ ai = a0 . We show, in fact, that ai = a0 for sufficiently large i.
Let P 0 be the limiting polygon with vertices a0 . Suppose, if possible that a0 is a concave vertex. We
can find a neighbourhood Vx0 around each vertex x0 of P 0 such that, selecting any point from each of these
neighbourhoods Vx0 will create a vertex with a convex vertex in Va0 . Thus, there is an index r such that the
polygon Pr has a vertex inside each of these neighbourhoods. If the convex vertex ar is reflected about a
line joining two other vertices in Pr to ar+1 , then ar will be an interior point of Pr+1 and each subsequent
ai will lie outdie of the interior of Pr+1 , and so outside of Ua0 , yielding a contradiction.
Hence P 0 has acute angles at all of its vertices. But then for sufficiently large n, so will Pn and so
ai = an for i ≥ n. The result follows.

10