THE UNIVERSITY OF TORONTO

UNDERGRADUATE MATHEMATICS COMPETITION

Sunday, March 16, 2003

Time: 3 12 hours
No aids or calculators permitted.

1. Evaluate
∞  
X
−1 2
tan .
n=1
n2

[tan−1 denotes the (composition) inverse function for tan.]

2. Let a, b, c be positive real numbers for which a + b + c = abc. Prove that

1 1 1 3
√ +√ +√ ≤ .
1 + a2 1 + b2 1 + c2 2

3. Solve the differential equation

y 00 = yy 0 .

4. Show that n divides the integer nearest to

(n + 1)!
.
e

5. For x > 0, y > 0, let g(x, y) denote the minimum of the three quantities, x, y + 1/x and 1/y. Determine
the maximum value of g(x, y) and where this maximum is assumed.

6. A set of n lightbulbs, each with an on-off switch, numbered 1, 2, · · · , n are arranged in a line. All are
initially off. Switch 1 can be operated at any time to turn its bulb on of off. Switch 2 can turn bulb 2
on or off if and only if bulb 1 is off; otherwise, it does not function. For k ≥ 3, switch k can turn bulb k
on or off if and only if bulb k − 1 is off and bulbs 1, 2, · · · , k − 2 are all on; otherwise it does not function.
(a) Prove that there is an algorithm that will turn all of the bulbs on.
(b) If xn is the length of the shortest algorithm that will turn on all n bulbs when they are initially off,
determine the largest prime divisor of 3xn + 1 when n is odd.

7. Suppose that the polynomial f (x) of degree n ≥ 1 has all real roots and that λ > 0. Prove that the set
{x ∈ R : |f (x)| ≤ λ|f 0 (x)|} is a finite union of closed intervals whose total length is equal to 2nλ.

8. Three matrices A, B and A + B have rank 1. Prove that either all the rows of A and B are multiples
of one and the same vector, or that all of the columns of A and B are multiples of one and the same
vector.

9. Prove that the integral

∞
sin2 x
Z
dx
0 π 2 − x2
exists and evaluate it.

10. Let G be a finite group of order n. Show that n is odd if and only if each element of G is a square.

END

1
 Solutions

1. Evaluate
∞  
X
−1 2
tan .
n=1
n2

[tan−1 denotes the (composition) inverse function for tan.]

Solution 1. Let an = tan−1 n for n ≥ 0. Then

(n + 1) − (n − 1) 2
tan(an+1 − an−1 ) = 2
= 2
1 + (n − 1) n

for n ≥ 1. Then
m
X 2
tan−1 = tan−1 (m + 1) + tan−1 m − tan−1 1 − tan−1 0 .
n=1
n2

Letting m → ∞ yields the answer π/2 + π/2 − π/4 − 0 = 3π/4.

Solution 2. Let bn = tan−1 (1/n) for n ≥ 0. Then

2
tan(bn−1 − bn+1 ) =
n2
for n ≥ 2, whence
m m
X 2 X 1 1 1
tan−1 2
= tan−1
2 + (bn−1 − bn+1 ) = tan−1 2 + tan−1 1 + tan−1 − tan−1 − tan−1
n=1
n n=2
2 m m+1
1 1
= (tan−1 2 + cot−1 2) + tan−1 1 − tan−1 − tan−1
m m+1
π π 1 1
= + − tan−1 − tan−1
2 4 m m+1
for m ≥ 3, from which the result follows by letting m tend to infinity.
Pn
Solution 3. [S. Huang] Let sn = k=1 tan−1 (2/k 2 ) and tn = tan sn . Then {tn } = {2, ∞, −9/2, −14/5, −20/9, · · ·}
where the numerators of the fractions are {−2, −5, −9, −14, −20, · · ·} and the denominators are {−1, 0, 2, 5, 8, · · ·}.
We conjecture that
−n(n + 3)
tn =
(n − 2)(n + 1)
for n ≥ 1. This is true for 1 ≤ n ≤ 5. Suppose that it holds to n = k − 1 ≥ 5, so that tk−1 =
−(k − 1)(k + 2)/(k − 3)k. Then

tk−1 + (2/k 2 )
tk =
1 − 2tk−1 k −2
−k 2 (k − 1)(k + 2) + 2(k − 3)k
=
k 3 (k − 3) + 2(k − 1)(k + 2)
−k(k + 3)(k 2 − 2k + 2) −k(k + 3)
= 2
= .
(k − 2)(k + 1)(k − 2k + 2) (k − 2)(k + 1)

The desired expression for tn holds by induction and so limn→∞ tn = −1. For n ≥ 3, tn < 0 and
tan−1 (2/n2 ) < π/2, so we must have π/2 < sn < π and sn = π − tan−1 tn . Therefore

lim sn = tan−1 (π + lim tn ) = π − (π/4) = (3π)/4 .

n→∞ n→∞

2
 2. Let a, b, c be positive real numbers for which a + b + c = abc. Prove that

1 1 1 3
√ +√ +√ ≤ .
1+a 2 1+b2 1+c2 2

Solution 1. Let a = tan α, b = tan β, c = tan β, where α, β, γ ∈ (0, π/2). Then

tan α + tan β + tan γ − tan α tan β tan γ a + b + c − abc

tan(α + β + γ) = = =0,
1 − tan α tan β − tan β tan γ − tan γ tan α 1 − ab − bc − ca

whence α + β + γ = π. Then, the left side of the inequality is equal to

cos α + cos β + cos γ = cos α + cos β − cos(α + β)

     
α+β α−β α+β
= 2 cos cos − 2 cos2 +1
2 2 2
   
α+β α+β
≤ 2 cos − 2 cos2 +1
2 2
   
γ γ
= 2 sin − 2 sin2 +1
2 2
3 1 3
= − (2 sin(γ/2) − 1)2 ≤ ,
2 2 2
with equality if and only if α = β = γ = π/3.

Solution 2. Define α, β and γ and note that α + β + γ = π as in Solution 1. Since cos x is a concave
function on [0, π/2], we have that
 
cos α + cos β + cos γ α+β+γ π 1
≤ cos = cos = ,
3 3 3 2

from which the result follows.

3. Solve the differential equation

y 00 = yy 0 .

Solution. The equation can be rewritten

dy 0 0
y = yy 0
dy

from which either y 0 = 0 and y is a constant, or dy 0 = ydy, whence

dy 1
= y 0 = (y 2 + k)
dx 2
for some constant k. (One can also get to the same place by noting that the equation can be rewritten as
2y 00 = (y 2 )0 and integrating.)
In case k = 0, we have that 2dy/y 2 = dx, whence −2/y = x + c and y = −2(x + c)−1 for some constant
c.
In the case that k = a2 > 0, we have that

2dy
= dx
y 2 + a2

3
whence (2/a) tan−1 (y/a) = x + c and
a(x + c)
y = a tan
2
for some constant c.
In the case that k = −b2 < 0, we have that
 
2dy 1 dy dy
= − = dx
y 2 − b2 b y−b y+b
whence
1 y−b
ln =x+c
b y+b
for some constant c. Solving this for y yields that

1 + eb(x+c) 1 − eb(x+c)
   
y=b or y=b .
1 − eb(x+c) 1 + eb(x+c)

4. Show that the positive integer n divides the integer nearest to

(n + 1)!
.
e

Solution. By Taylor’s Theorem, we have that

1 1 1 1 ec
e−1 = 1 − + − + · · · + (−1)n+1 + (−1)n+2
1! 2! 3! (n + 1)! (n + 2)!
where −1 < c < 0. Hence
(n + 1)! (n + 1)! ec
(n + 1)!e−1 = (n + 1)! − (n + 1)! + − + · · · + (−1)n [(n + 1) − 1] + (−1)n−2 .
2! 3! n+2
The last term does not exceed 1/(n + 2), which is less than 1/2. The second last term is equal to ±n, and
each previous term has a factor n in the numerator that is not cancelled out by the denominator. Since the
sum of all but the last term is an integer divisible by n and is the nearest integer to (n + 1)!/e, the result
holds.

5. For x > 0, y > 0, let g(x, y) denote the minimum of the three quantities, x, y + 1/x and 1/y. Determine
the maximum value of g(x, y) and where this maximum is assumed.

Solution 1. Consider the attached graph showing the √ curves

√ of equations x = y + 1/x, x = 1/y and
y + 1/x = 1/y, all three of which contain the point P ∼ ( 2, 1/ 2), and the regions in which√g(x, y) is each
one of the three given functions. When g(x, y) = x, the maximum value of g(x,√ y) is equal to 2, assumed at
the point P . When g(x, y) = 1/y, the maximum value of g(x, y) is equal
√ to 2, also assumed at P . Suppose
(x, y) are such that g(x, y) = y + 1/x. Note that the curve y = 2 − (1/x) passes through the point P ,
where it intersects each of the curves y + (1/x) = (1/y) and y + (1/x) = x. It intersects neither of these
curves
√ at any other point, and so√lies vertically above the region where g(x, y) = y + (1/x). In this region,
√
y ≤ 2 − (1/x) and so g(x, y) ≤ √ 2, √with equality only at the point P . Hence the required maximum is 2,
and it is assumed at (x, y) = ( 2, 1/ 2).

Solution 2. [R. Appel] If x ≤ 1 and y ≤ 1, then g(x, y) ≤ x ≤ 1. If y ≥ 1, then g(x, y) ≤ 1/y ≤ 1. It

remains to examine the case x > 1 and y < 1, so that y + (1/x) < 2. Suppose that min (x, 1/y) = a and
max (x, 1/y) = b. Then min (1/x, y) = 1/a and max (1/x, y) = 1/b, so that
1 1 1 a+b
y+ = + = .
x a b ab
4
Hence g(x, y) = min (a, (a + b)/(ab)). Either a2 ≤ 2 or a2 ≥ 2. But in the latter case,

a+b 2b √
≤√ = 2.
ab 2b
√ √ √
In either case, g(x, y) ≤ 2. This maximum value is attained when (x, y) = ( 2, 1/ 2).

Solution 3. [D. Varodayan] By the continuity of the functions, each of the regions {(x, y) : 0 < x <
y + (1/x), xy < 1}, {(x, y) : 0 < x, y + (1/x) < x, y + (1/x) < (1/y)}, and {(x, y) : 0 < (1/y) < x, (1/y) <
y + (1/x)} is an open subset of the plane; using partial derivatives, we see that none of the three functions
being minimized have any critical values there. It follows that any extreme values of g(x, y) must occur on
one of the curves defined by the equations
x = y + (1/x) (1)
x = 1/y (2)
y + (1/x) = (1/y) (3)
On the curve (1), x > 1 and  
x
g(x, y) = min x, 2
x −1
 √
x, if x ≤ √2;
= x
x2 −1 , if x ≥ 2.
On the curve (2),
g(x, y) = min (x, 2/x)
 √
x, if x ≤ √2;
=
2/x, if x ≥ 2.
On the curve (3), 0 < y < 1 and
 
y 1
g(x, y) = min ,
1 − y2 y
( y
2, if 0 < y < √12 ;
= 1−y
1/y, if √12 ≤ y ≤ 1.
√ √ √
On each of these curves, g(x, y) reaches its maximum value of 2 when (x, y) = ( 2.1/ 2).

Solution 4. [J. Sparling] Let z = 1/y. For fixed z, let

vz (x) = min {x, z, (1/x) + (1/z)}

and
w(z) = max {vz (x) : x > 0} .
Suppose that z ≤ 1. Then (1/x) + (1/z) ≥ z, so vz (x) = min {x, z} and

x, for x ≤ z;
vz (x) =
z, for x ≥ z;
√
so that w(z) = z when z ≤ 1. Suppose that 1 < z ≤ 2, so that z ≤ z/(z 2 − 1). Then

 x, for x ≤ z;
vz (x) = z, for z ≤ x < z/(z 2 − 1);
(1/x) + (1/z), for z/(z 2 − 1) ≤ x;


5
 √ √
so that w(z) = z when 1 < z ≤ 2. Finally, suppose that 2 > z. Note that x ≤ (1/x) √ + (1/z) ⇔
zx2 −x−z ≤ 0. Then the minimum of x and (1/x)+(1/z) is x when zx2 −x−z ≤ 0, or x ≤ (1+ 1 + 4z 2 )/2z.
Since √ √ √
√ 1 + 1 + 4z 2 (2 2z − 1) − 1 + 4z 2
 
2− =
2z 2z
2
√
4z − 4 2z
= √ √
2z[(2 2z − 1) + 1 + 4z 2 ]
√
2(z − 2)
= √ √ ≥0,
(2 2z − 1) + 1 + 4z 2
this minimum is always less than z, so that
( √
1+ 1+4z 2
x, for x ≤
vz (x) = √2z
1 1 1+ 1+4z 2
x + z , for x ≥ 2z ,
√ √ √ √
so that w(z) = (1 + 1 + 4z 2 )/(2z) ≤ 2 when 2 ≤ z. Hence,
√ √the minimum value of w(z) = 2 and this
is the maximum value of g(x, y), assumed when (x, y) = ( 2, 1/ 2).

Solution 5. For x > 0, let  

1 1
hx (y) = min x, y + , .
x y
√
Suppose that x ≤ 2. Then x − (1/x) ≤ (1/x) and
1 1

 y + x , if 0 < y ≤ x − x ;
hx (y) = x, if x − x ≤ y ≤ x1 ;
1
 1, if x1 ≤ y;
y
√
so that the minimum value of hx (y) is x, and this occurs when x − (1/x) ≤ y ≤ (1/x). Suppose that x ≥ 2.
Then y + (1/x) ≤ (1/y) ⇔ xy 2 + y − x ≤ 0 and
√ √ √
√ 1 + 1 + 4x2 ( 8x − 1) − 1 + 4x2
 
2− =
2x 2x
2
√
4x − 4 2x
= √ √
2x[( 8x − 1) + 1 + 4x2 )
√
2(x − 2)
= √ √ ≥0,
( 8x − 1) + 1 + 4x2

so that √
1+ 1 + 4x2 √
≤ 2≤x.
2x
( √
2
y + 1 , when 0 < y ≤ −1+ 2x1+4x ;
hx (y) = 1 x √
−1+ 1+4x2
y , when 2x ≤ y;
√ √
so that the minimum value of hx (y) is (1 + 1 + 4x2 )/(2x), and this occurs when y = (−1 + 1 + 4x2 )/(2x).
Thus, we have to maximize the function u(x) where
 √
x, √ if 0 < x ≤ 2;
u(x) = 1+ 1+4x2
√
2x , if 2 ≤ x.
√ √
By what we have shown, this maximum is 2 and is attained when x = 2. The result follows.

6
 6. A set of n lightbulbs, each with an on-off switch, numbered 1, 2, · · · , n are arranged in a line. All are
initially off. Switch 1 can be operated at any time to turn its bulb on of off. Switch 2 can turn bulb 2
on or off if and only if bulb 1 is off; otherwise, it does not function. For k ≥ 3, switch k can turn bulb k
on or off if and only if bulb k − 1 is off and bulbs 1, 2, · · · , k − 2 are all on; otherwise it does not function.
(a) Prove that there is an algorithm that will turn all of the bulbs on.
(b) If xn is the length of the shortest algorithm that will turn on all n bulbs when they are initially off,
determine the largest prime divisor of 3xn + 1 when n is odd.

Solution. (a) Clearly x1 = 1 and x2 = 2. Let n ≥ 3. The only way that bulb n can be turned on is for
bulb n − 1 to be off and for bulbs 1, 2, · · · , n − 2 to be turned on. Once bulb n is turned on, then we need
get bulb n − 1 turned on. The only way to do this is to turn off bulb n − 2; but for switch n − 2 to work,
we need to have bulb n − 3 turned off. So before we can think about dealing with bulb n − 1, we need to
get the first n − 2 bulbs turned off. Then we will be in the same situation as the outset with n − 1 rather
than n bulbs. Thus the process has the following steps: (1) Turn on bulbs 1, · · · , n − 2; (2) Turn on bulb n;
(3) Turn off bulbs n − 2, · · · , 1; (3) Turn on bulbs 1, 2, · · · , n − 1. So if, for each positive integer k, yk is the
length of the shortest algorithm to turn them off after all are lit, then

xn = xn−2 + 1 + yn−2 + xn−1 .

Since we can clearly find an algorithm for turning on or off bulb 1, an induction argument will provide an
algorithm for turning any number of bulbs on or off.
We show that xn = yn for n = 1, 2, · · ·. Suppose that we have an algorithm that turns all the bulbs on.
We prove by induction that at each step we can legitimately reverse the whole sequence to get all the bulbs
off again. Clearly, the first step is to turn either bulb 1 or bulb 2 on; since the switch is functioning, we can
turn the bulb off again. Suppose that we can reverse the first k − 1 steps and are at the kth step. Then the
switch that operates the bulb at that step is functioning and can restore us to the situation at the end of
the (k − 1)th step. By the induction hypothesis, we can go back to having all the bulbs off. Hence, given
the bulbs all on, we can reverse the steps of the algorithm to get the bulbs off again. A similar argument
allows us to reverse the algorithm that turns the bulbs off. Thus, for each turning-on algorithm there is a
turning-off algorithm of equal length, and vice versa. Thus xn = yn .
We have that xn = xn−1 + 2xn−2 + 1 for n ≥ 3. By, induction, we show that, for m = 1, 2, · · ·,

x2m = 2x2m−1 and x2m+1 = 2x2m + 1 = 4x2m−1 + 1 .

This is true for m = 1. Suppose it is true for m ≥ 1. Then

x2(m+1) = x2m+1 + 2x2m + 1 = 2(x2m + 1) + 4x2m−1

= 2(x2m + 2x2m−1 + 1) = 2x2m+1 ,

and
x2(m+1)+1 = x2(m+1) + 2x2m+1 + 1 = 2x2m+1 + 4x2m + 3
= 2(x2m+1 + 2x2m + 1) + 1 = 2x2(m+1) + 1 .
Hence, for m ≥ 1,

3x2m+1 + 1 = 4(3x2m−1 + 1) = · · · = 4m (3x1 + 1) = 4m+1 = 22(m+1) .

Thus, the largest prime divisor is 2.

7. Suppose that the polynomial f (x) of degree n ≥ 1 has all real roots and that λ > 0. Prove that the set
{x ∈ R : |f (x)| ≤ λ|f 0 (x)|} is a finite union of closed intervals whose total length is equal to 2nλ.

7
 Qk
Solution. Wolog, we may assume that the leading coefficient is 1. Let f (x) = i=1 (x − ri )mi , where
Pk
n = i=1 mi . Then
k
f 0 (x) X mi
= .
f (x) i=1
x − ri
Pk
Note that the derivative of this function, − i=1 mi (x − ri )−2 < 0, so that it decreases on each interval
upon which it is defined. By considering the graph of f 0 (x)/f (x), we see that f 0 (x)/f (x) ≥ 1/λ on finitely
many intervals of the form (ri , si ], where ri < si and the ri and sj interlace, and f 0 (x)/f (x) ≤ −1/λ on
finitely many intervals of the form [ti , ri ), where ti < ri and the ti and rj interlace. For each i, we have
ti < ri < si < ti+1 .
The equation f 0 (x)/f (x) = 1/λ can be rewritten as

k
X
0 =(x − r1 )(x − r2 ) · · · (x − rk ) − λ mi (x − r1 ) · · · (x\
− ri ) · · · (x − rk )
i=1
k
X k
X 
= xk − ri − λ mi xk−1 + · · · .
i=1 i=1

The sum of the roots of this polynomial is

s1 + s2 + · · · + sk = r1 + · · · + rk − λn ,
Pm
so that i=1 (si − ri ) = λn. This is the sum of the lengths of the intervals (ri , si ] on which f 0 (x)/f (x) ≥ 1/λ.
Similarly, we can show that f 0 (x)/f (x) ≤ −1/λ on a finite collection of intervals of total length λn. The set
on which the inequality of the problem holds is equal to the union of all of these half-open intervals and the
set {r1 , r2 , · · · , rk }. The result follows.

8. Three matrices A, B and A + B have rank 1. Prove that either all the rows of A and B are multiples
of one and the same vector, or that all of the columns of A and B are multiples of one and the same
vector.

Solution 1. Let A = (aij ) and B = (bij ). Since the image each of the matrix A and the matrix B as an
operator on Cn is the span of its column vectors, there exist nonzero vectors (ai ), (bi ), (uj ), (vj ) for which

aij = ai uj and bij = bi vj .

Hence aij + bij = ai uj + bi vj . Similarly, since A + B = (aij + bij ) has rank 1, there are vectors (ci ) and (wj )
for which
aij + bij = ci wj .
Thus, for all i and j,
ai uj + bi vj = ci wj .

If the rows of A and B are multiples of the same row, then there exists a constant k for which vj = kuj
for each j, and so ci wj = (ai + kbi )uj , and so the ith row of A + B is also a multiple of (uj ).
Suppose if that the vectors (uj ) and (vj ) are linearly independent. Wolog, we may assume that (u1 , u2 )
and (v1 , v2 ) are linearly independent. Then u1 v2 − u2 v1 6= 0, and for each i and j, the system

u1 ai + v1 bi = w1 ci

u2 ai + v2 bi = w2 ci

8
has the solution ai = pci , bi = qci , where

w 1 v2 − w 2 v1
p=
u1 v2 − u2 v1

and
u1 w2 − u2 w1
q=
u1 v2 − u2 v1
are independent of i. The result follows.

Solution 2. Suppose that A = uat and B = vbt where a = (aj ), b = (bj ), u = (ui ) and v = (vi ) are
column vectors. If the set {u, v} is linearly dependent, then the columns of A + B, A and B are multiplies
of each other and the result follows. Otherwise, let {u, v} be linearly independent. The ith columns of A
and B are respectively ai u and bi v. Since A + B is of rank 1, there exists a column vector w and scalars ci
for which
ai u + bi v = ci w
for each i. Hence, for each i, j,
ci (aj u + bj v) − cj (ai u + bi v) = 0
=⇒ (ci aj − cj ai )u + (ci bj − cj bi )v = 0
=⇒ ci aj − cj ai = ci bj − cj bi = 0
=⇒ ci aj = cj ai and ci bj = cj bi
=⇒ ci (aj bi − ai bj ) = 0 .
Suppose, wolog, that c1 6= 0. Then either a1 or b1 is nonzero; suppose the former. Then bi = (ai /a1 )b1
and ai = (ai /a1 )ai for each i, so that the vectors b and a are parallel, and the rows of A + B are sums of
multiples of these parallel vectors. The result follows.

9. Prove that the integral

∞
sin2 x
Z
dx
0 π 2 − x2
exists and evaluate it.

Solution. Since limx→π (sin2 x)/(π 2 − x2 ) exists and equals 0 (by l’Hôpital’s Rule), there is a removable
singularity in the integrand at x = π. The integral on the infinite interval converges by comparison with the
integral of 1/x2 .
First, note that, for k ≥ 0,

(k+1)π (k+1)π (k+1)π

sin2 x sin2 x sin2 x
Z Z Z 
1
2 2
dx = − dx − dx
kπ π −x 2π kπ x−π kπ x+π
kπ 2 (k+2)π
sin2 x
Z Z
1 sin x 1
=− dx + dx .
2π (k−1)π x 2π (k+1)π x

Hence
nπ π (n+1)π
sin2 x sin2 x sin2 x
Z Z Z
1 1
dx = − dx + dx .
0 π 2 − x2 2π −π x 2π (n−1)π x
The first integral on the right vanishes, because the integrand is odd, and so

nπ (n+1)π (n+1)π
sin2 x sin2 x
Z Z Z
1 1 1 1
2 2
dx = dx ≤ dx ≤ ,
0 π −x 2π (n−1)π x 2π (n−1)π x n−1

9
 R∞ sin2 x
with the result that 0 π 2 −x2 dx = 0.

10. Let G be a finite group of order n. Show that n is odd if and only if each element of G is a square.

Solution. Suppose that n is odd. Then, by Lagrange’s theorem, each element of G is of odd order, so
that, if a ∈ G, then a2k+1 = e (the identity) for some nonnegative integer k. Hence a = (ak+1 )2 is a square.
On the other hand, suppose that n is even. Pair off each element of G with its inverse. Some elements get
paired off with a distinct element, and others get paired off with themselves. Since n is even and an even
number of elements get paired off with a distinct element, there must be an even number of elements that get
paired off with themselves. Since the identity gets paired off with itself, there must be some other element v
that also is its own inverse, i.e. v 2 = e. Consider the mapping x → x2 defined from G to itself. Since e and
v have the same image and since G is finite, this mapping cannot be onto. Hence, there must be an element
of G which is not the square of another element. The result follows.

10