Algebra: Mathematics Olympiad Coachs Seminar, Zhuhai, China
Algebra: Mathematics Olympiad Coachs Seminar, Zhuhai, China
Algebra: Mathematics Olympiad Coachs Seminar, Zhuhai, China
03/22/2004
Algebra
1. There exists a polynomial P of degree 5 with the following property: if z is a complex number such
that z 5 + 2004z = 1, then P (z 2 ) = 0. Find all such polynomials P .
2. Let N denote the set of positive integers. Find all functions f : N → N such that
f (m + n)f (m − n) = f (m2 )
for all m, n ∈ N.
Solution: Function f (n) = 1, for all n ∈ N, is the only function satisfying the conditions of
the problem.
Note that
f (1)f (2n − 1) = f (n2 ) and f (3)f (2n − 1) = f ((n + 1)2 )
for n ≥ 3. Thus
f (3) f ((n + 1)2 )
= .
f (1) f (n2 )
f (3)
Setting f (1) = k yields f (n2 ) = k n−3 f (9) for n ≥ 3. Similarly, for all h ≥ 1,
f (h + 2) f ((m + 1)2 )
=
f (h) f (m2 )
for sufficiently large m and is thus also k. Hence f (2h) = k h−1 f (2) and f (2h + 1) = k h f (1).
But
f (25) f (25) f (11)
= · ··· · = k8
f (9) f (23) f (9)
and
f (25) f (25) f (16)
= · = k2 ,
f (9) f (16) f (9)
so k = 1 and f (16) = f (9). This implies that f (2h + 1) = f (1) = f (2) = f (2j) for all j, h, so f is
constant. From the original functional equation it is then clear that f (n) = 1 for all n ∈ N.
holds for all real numbers x and y. Prove that no very convex function exists.
from which,
µ ¶ µ ¶ µ ¶ µ ¶
i+2 i+1 i+1 i 4
f −f ≥f −f + ,
n n n n n
that is, ∆i+1 ≥ ∆i + 4/n. Combining this for n consecutive values of i gives ∆i+n ≥ ∆i + 4. Summing
this inequality for i = 0 to i = n − 1 and cancelling terms yields
This cannot hold for all n ≥ 1. Hence there are no very convex functions.
for all nonnegative integers n. This will yield a contradiction, because for fixed x and y the right side
gets arbitrarily large, while the left side remains fixed.
We are given the base case n = 0. Now if the inequality holds for a given n, then for a, b real,
f (a) + f (a + 2b)
≥ f (a + b) + 2n+1 |b|,
2
f (a + b) + f (a + 3b) ≥ 2(f (a + 2b) + 2n+1 |b|),
and
f (a + 2b) + f (a + 4b)
≥ f (a + 3b) + 2n+1 |b|.
2
Adding these three inequalities and canceling terms yields
f (a) + f (a + 4b)
≥ f (a + 2b) + 2n+3 |b|.
2
Thus, for each n, at least one of f (1/2n ) − f (0) and f (−1/2n ) − f (0) is greater than 0. It follows
that at least one of f (1) − f (0) and f (−1) − f (0) is greater than 2n for all n ≥ 1, which is impossible.
Hence there is no very convex functions.
P P
Solution: Let bi = 2 − ai , and let S = bi and T = b2i . Then the given conditions are
that
(2 − a1 ) + · · · + (2 − an ) ≥ n
and
(4 − 4b1 + b21 ) + · · · + (4 − 4bn + b2n ) ≥ n2 ,
which is to say S ≤ n and T ≥ n2 − 4n + 4S.
From these inequalities, we obtain
T ≥ n2 − 4n + 4S ≥ (n − 4)S + 4S = nS.
P
On the other hand, if bi > 0 for i = 1, . . . , n, then certainly bi < bi = S ≤ n, and so
Thus we cannot have bi > 0 for i = 1, . . . , n, so bi ≤ 0 for some i, and ai ≥ 2 for that i, proving the
claim.
First Solution: (by Alison Miller) For any k, we have the following inequality:
1 1 1 1
ak+1 + ≥ a2k + + = a2k + ≥ ak ,
20 5 20 4
because (ak − 12 )2 ≥ 0. Summing up this inequalities over k = n + 1, . . . , n + 4, we have
1 1
an+5 + ≥ an+1 ≥ a2n + ,
5 5
and similarly an ≥ a2n−5 , so
an+5 ≥ a2n ≥ a4n−5 ,
implying the desired result.
which is equivalent to
2 sin 2 · sin 1 + 2(2 sin 4 · sin 1) + · · · + 89(2 sin 178 · sin 1) = 90 cos 1.
so (∗) is true.
Hence, we obtain
where x = sin a, y = sin b, and z = sin c (hence x, y, z > 0). Since the last inequality is symmetric
with respect to x, y, z, we may assume that x ≥ y ≥ z > 0. It suffices to prove that
which is evident as
x(y 2 − x2 )(z 2 − x2 ) ≥ 0
and
z(z 2 − x2 )(z 2 − y 2 ) ≥ z(y 2 − x2 )(z 2 − y 2 ) ≥ y(z 2 − y 2 )(y 2 − x2 ).
Note: The key step of the proof is an instance of Schur’s Inequality with r = 21 .
Note that this inequality is symmetric with respect to α, β, γ, so we can assume without loss of
generality that 0◦ < α < β < γ < 90◦ . Then regrouping the terms on the left-hand side gives
which is positive because the function y = sin x is increasing for 0◦ < x < 90◦ .
6 Zuming Feng ([email protected]), Phillips Exeter Academy, Exeter 03833, USA
Second Solution: We keep the notation of the first solution. By the Addition formulas,
we have
Note that sin x is increasing, cos x and cos(2x) are decreasing for 0 < x < 90◦ . Since 0 < α, γ, α +γ <
90◦ , each of the two products in the last addition is less than or equal to 0. Hence
and
sin 3γ + sin 3β − cos(α − γ) − cos(α − β) ≤ 0.
First Solution: (Based on work by Matthew Tang and Anders Kaseorg) By multiplying a, b,
and c by a suitable factor, we reduce the problem to the case when a + b + c = 3. The desired
inequality reads
(a + 3)2 (b + 3)2 (c + 3)2
+ + ≤ 8.
2a2 + (3 − a)2 2b2 + (3 − b)2 2c2 + (3 − c)2
Set
(x + 3)2
f (x) =
2x2 + (3 − x)2
Mathematics Olympiad Coachs Seminar, Zhuhai, China 7
It is not difficult to see that the desired result follows from summing up the above inequality and its
analogous forms.
Thus, we have
(2a + b + c)2 3(2a2 + (b + c)2 ) − 2(a − b − c)2
=
2a2 + (b + c)2 2a2 + (b + c)2
2(a − b − c)2
= 3− 2 .
2a + (b + c)2
8 Zuming Feng ([email protected]), Phillips Exeter Academy, Exeter 03833, USA
Because (b + c)2 ≤ 2(b2 + c2 ), we have 2a2 + (b + c)2 ≤ 2(a2 + b2 + c2 ) and its analogous forms. It
suffices to show that
(a − b − c)2 (b − a − c)2 (c − a − b)2 1
+ + ≥ ,
2(a2 + b2 + c2 ) 2(a2 + b2 + c2 ) 2(a2 + b2 + c2 ) 2
or,
(a − b − c)2 + (b − a − c)2 + (c − a − b)2 ≥ a2 + b2 + c2 .
Multiplying this out, the left-hand side of the last inequality becomes 3(a2 + b2 + c2 ) − 2(ab + bc + ca).
Therefore the last inequality is equivalent to 2[a2 + b2 + c2 − (ab + bc + ca)] ≥ 0, which is evident
because
2[a2 + b2 + c2 − (ab + bc + ca)] = (a − b)2 + (b − c)2 + (c − a)2 .
Equalities hold if and only if (b + c)2 = 2(b2 + c2 ) and (c + a)2 = 2(c2 + a2 ), that is, a = b = c.
which reduces to
(2a − x)(10a2 − 3ax − x2 ) = (2a − x)2 (5a + x) ≥ 0,
which is evident. We proved that
4a2 − 12a(b + c) + 5(b + c)2 4(2a − b − c)
2 2
≥− ,
3[2a + (b + c) ] 3(a + b + c)
hence (1) follows. Equality holds if and only if 2a = b + c, 2b = c + a, 2c = a + b, i.e., when a = b = c.
where σ runs over all permutations of 1, . . . , n (for a total of n! terms). For example, if n = 3, and
we write x, y, z for x1 , x2 , x3 ,
X
x3 = 2x3 + 2y 3 + 2z 3
sym
X
x2 y = x2 y + y 2 z + z 2 x + x2 z + y 2 x + z 2 y
sym
X
xyz = 6xyz.
sym
We combine the terms in the desired inequality over a common denominator and use symmetric sum
notation to simplify the algebra. The numerator of the difference between the two sides is
X
2 4a6 + 4a5 b + a4 b2 + 5a4 bc + 5a3 b3 − 26a3 b2 c + 7a2 b2 c2 , (3)
sym
and it suffices to show the the expression in (3) is always greater or equal to 0. By the Weighted
AM-GM Inequality, we have 4a6 + b6 + c6 ≥ 6a4 bc, 3a5 b + 3a5 c + b5 a + c5 a ≥ 8a4 bc, and their
analogous forms. Adding those inequalities yields
X X X X
6a6 ≥ 6a4 bc and 8a5 b ≥ 8a4 bc.
sym sym sym sym
Consequently, we obtain
X X
4a6 + 4a5 b + 5a4 bc ≥ 13a4 bc. (4)
sym sym
Again by the AM-GM Inequality, we have a4 b2 + b4 c2 + c4 a2 ≥ 4a2 b2 c2 , a3 b3 + b3 c3 + c3 a3 ≥ 3a2 b2 c2 ,
and their analogous forms. Thus,
X X
a4 b2 + 5a3 b3 ≥ 6a2 b2 c2 ,
sym sym
or X X
a4 b2 + 5a3 b3 + 7a2 b2 c2 ≥ 13a2 b2 c2 . (5)
sym sym
Recalling Schur’s Inequality, we have
or X
a3 − 2a2 b + abc ≥ 0.
sym
Thus X X
13a4 bc − 26a3 b2 c + 13a2 b2 c2 ≥ 13abc a3 − 2a2 b + abc ≥ 0. (6)
sym sym
Adding (4), (5), (6) yields (3).
10 Zuming Feng ([email protected]), Phillips Exeter Academy, Exeter 03833, USA
Note: While the last two methods seem inefficient for this problem, they hold the keys to proving
the following inequality:
a+b+c √ √ √ √ √ √ √
− abc ≤ max{( a − b)2 , ( b − c)2 , ( c − a)2 }.
3
Second Solution: (by Ian Le) We again prove the stronger inequality (1), which can be rewritten
X
[a − 2(ab)1/2 + (abc)1/3 ] ≥ 0,
sym
where the sum is taken over all six permutations of a, b, c. This inequality follows from adding the
two inequalities X
[a − 2a2/3 b1/3 + (abc)1/3 ] ≥ 0
sym
and X
[a2/3 b1/3 + a1/3 b2/3 − 2a1/2 b1/2 ] ≥ 0.
sym
Mathematics Olympiad Coachs Seminar, Zhuhai, China 11
The first of these is the Schur’s inequality with x = a1/3 , y = b1/3 , z = c1/3 , while the second
follows from the AM-GM inequality.
Third Solution: Without loss of generality, assume that b is between a and c. The desired
inequality reads √ √
3
a + b + c − 3 abc ≤ 3(c + a − 2 ac).
As a function of b, the right side minus the left side is concave (its second derivative is −(2/3)(ac)1/3 b−5/3 ),
so its minimum value in the range [a, c] occurs at one of the endpoints. Thus, without loss of gener-
ality, we may assume a = b. Moreover, we may rescale the variables to get a = b = 1. Now the claim
reads
2c + 3c1/3 + 1
≥ c1/2 .
6
This is an instance of weighted AM-GM inequality.
The right inequality can be proved, by using the method of the third solution above. We leave the
details as an exercise for the reader.
√ √
The left inequality falls apart when we replace m by c, the average of ( ai − aj )2 for 1 ≤ i < j ≤ n.
Since
P √ √ 2
m c 1≤i<j≤n ( ai − aj )
≤ = ¡ ¢
2 2 2 n2
P √ √ 2
1≤i<j≤n ( ai − aj )
=
n(n − 1)
P √
(n − 1)(a1 + a2 + · · · + an ) − 2 1≤i<j≤n ai aj
= ,
n(n − 1)
the inequality now reads √
X √ n(n − 1) n a1 a2 · · · an
ai aj ≥ .
2
1≤i<j≤n
in (2) to obtain, in a way, a sharper lower bound. A similar proof works. We leave it to the reader
as an exercise.
Even more generally, one can ask for a comparison between the difference between the arithmetic and
geometric means of a set of n nonnegative real numbers, and the maximum (or average) difference
between the arithmetic and geometric means over all k-element subsets. The authors do not know
what the correct inequalities should look like or how they may be proved.
12 Zuming Feng ([email protected]), Phillips Exeter Academy, Exeter 03833, USA
n µ ¶−1
X n+1
n n + 1 X 2i
= .
i 2n+1 i
i=0 i=1
Solution: Let
n µ ¶ n
1 X n −1 X k!(n − k)!
Sn = = .
n+1 k (n + 1)!
k=0 k=0
P
We must show that Sn = ( n+1 k
k=1 2 /k)/2
n+1 . To do so, it suffices to check that S = 1, which is
1
clear, and that 2n+2 Sn+1 − 2n+1 Sn = 2n+2 /(n + 2). Now
n+1 µ ¶ n+1 µ ¶
1 X n + 1 −1 X n + 1 −1
2Sn+1 = +
n+2 i j
i=0 j=0
n
õ ¶ µ ¶ !
2 1 X n + 1 −1 n + 1 −1
= + +
n+2 n+2 i i+1
i=0
n
2 1 X i!(n + 1 − i)! + (i + 1)!(n − i)!
= +
n+2 n+2 (n + 1)!
i=0
n
2 1 X i!(n − i)!(n + 1 − i + i + 1)
= +
n+2 n+2 (n + 1)!
i=0
2
= Sn + ,
n+2
as claimed.
12. Express
n
X
(−1)k (n−k)!(n+k)!
k=0
in closed form.
f (k) = (n+1−k)!(n+k)!
f (k) + f (k + 1)
= (n+1−k)!(n+k)! + (n−k)!(n+k+1)!
= (n + 1 − k + n + k + 1)(n−k)!(n+k)!
= 2(n + 1)(n−k)!(n+k)!.
Mathematics Olympiad Coachs Seminar, Zhuhai, China 13
Therefore,
n
X
(−1)k (n−k)!(n+k)!
k=0
X n
1
= (−1)k [f (k) + f (k + 1)]
2(n + 1)
k=0
f (0) + (−1)n f (n + 1)
=
2(n + 1)
(n+1)!n! + (−1)n 0!(2n+1)!
=
2(n + 1)
(n!)2 (−1)n (2n+1)!
= + .
2 2(n + 1)
Similarly
1 1 a
≤ = ,
b3 + c3 + abc bc(b + c) + abc abc(a + b + c)
and
1 1 b
≤ = .
c3 3
+ a + abc ca(c + a) + abc abc(a + b + c)
Thus
1 1 1 a+b+c 1
+ 3 + 3 ≤ = .
a3 3 3 3
+ b + abc b + c + abc c + a + abc abc(a + b + c) abc
a + b + c ≥ abc.
Solution: (by David Shin) Perform the substitutions x = 1/a, y = 1/b, and z = 1/c. It suf-
fices to prove that at least two of the inequalities
2x + 3y + 6z ≤ 6, 2y + 3z + 6x ≤ 6, 2z + 3x + 6y ≤ 6
14 Zuming Feng ([email protected]), Phillips Exeter Academy, Exeter 03833, USA
a contradiction. Thus, our assumptions is false and at least two of the desired inequalities must be
true. To obtain equalities, that is two of the numbers
2 3 6 2 3 6 2 3 6
+ + , + + , + +
a b c b c a c a b
are equal to 6, we must have 8x − 4y = 3y − 9z = 0, or a : b : c = (1/x) : (1/y) : (1/z) = 2 : 1 : 3.
Therefore the equalities hold if and only if (a, b, c) = (1, 3, 2), (3, 2, 1), (2, 1, 3).
16. Let a0 , a1 , · · · , an be numbers from the interval (0, π/2) such that
π π π
tan(a0 − ) + tan(a1 − ) + · · · + tan(an − ) ≥ n − 1.
4 4 4
Prove that
tan a0 tan a1 · · · tan an ≥ nn+1 .
Solution: Let bk = tan(ak − π/4), k = 0, 1, . . . , n. It follows from the hypothesis that for each k,
−1 < bk < 1, and X
1 + bk ≥ (1 − bl ). (1)
0≤l6=k≤n
n
à n
!1/n
Y Y
(1 + bk ) ≥ nn+1 (1 − bl )n ,
k=0 l=0
Mathematics Olympiad Coachs Seminar, Zhuhai, China 15
Because ³³
1 + bk 1 + tan(ak − π4 ) π´ π´
= = tan ak − + = tan ak ,
1 − bk 1 − tan(ak − π4 ) 4 4
the conclusion follows.
17. Let R be the set of real numbers. Determine all functions f : R → R such that
we have f (−x) = −f (x) for x 6= 0. Hence f (x) is odd. From now on, we assume x, y ≥ 0.
Setting y = 0 in the given condition yields f (x2 ) = xf (x). Hence f (x2 − y 2 ) = f (x2 ) − f (y 2 ), or,
f (x2 ) = f (x2 − y 2 ) + f (y 2 ). Since for x ≥ 0 there is a unique t ≥ 0 such that t2 = x, it follows that
By (2) and by setting x = 2t + 1 and y = 1 in (1), the left-hand side of (3) becomes
On the other hand, by setting x = t + 1 and y = 1 in (1), the right-hand side of (3) reads
or,
(t + 1)f (t + 1) − tf (t) = f (t) + (t + 1)f (1). (5)
Putting (3), (4), and (5) together leads to 2f (t) + f (1) = f (t) + (t + 1)f (1), or,
f (t) = tf (1)
for t ≥ 0. Recall that f (x) is odd; we conclude that f (−t) = −f (t) = −tf (1) for t ≥ 0. Hence
f (x) = kx for all x, where k = f (1) is a constant. It is not difficult to see that all such functions
indeed satisfy the conditions of the problem.
16 Zuming Feng ([email protected]), Phillips Exeter Academy, Exeter 03833, USA
a2 + b2 + c2 + abc = 4.
Prove that
0 ≤ ab + bc + ca − abc ≤ 2.
First Solution: (By Richard Stong) From the condition, at least one of a, b, and c does not
exceed 1, say a ≤ 1. Then
(1 − b)(1 − c) ≥ 0. (1)
bc ≤ 2 − a. (2)
ab + bc + ac − abc ≤ ab + 2 − a + ac(1 − b)
= 2 − a(1 + bc − b − c)
= 2 − a(1 − b)(1 − c) ≤ 2,
as desired.
The last equality holds if and only if b = c and a(1−b)(1−c) √ = 0.√Hence,
√ equality
√ for the√upper
√ bound
holds if and only if (a, b, c) is one of the triples (1, 1, 1), (0, 2, 2), ( 2, 0, 2), and ( 2, 2, 0).
Second Solution: (by Oaz Nir) We prove only the upper bound here. Either two of a, b, c
are less than or equal to 1, or two are greater than or equal to 1. Assume b and c have this property.
Then
b + c − bc = 1 − (1 − b)(1 − c) ≤ 1. (3)
Viewing the given equality as a quadratic equation in a and solving for a yields
p
−bc ± (b2 − 4)(c2 − 4)
a= .
2
Mathematics Olympiad Coachs Seminar, Zhuhai, China 17
Note that
or
ab + ac + bc − abc ≤ 2,
as desired.
Third Solution: We prove only the upper bound here. Define functions f , g as
for all nonnegative numbers x, y, z. Observe that if z ≤ 1, then both f and g are unbounded,
increasing functions of x and y.
Assume that f (a, b, c) = 4 and, without loss of generality, that a ≥ b ≥ c ≥ 0. Then c ≤ 1.
¡ ¢2
Let a0 = (a + b)/2. Because a + b = a0 + a0 and ab ≤ a−b 2 + ab = a0 2 , we have
Now increase a0 to e ≥ 0 such that f (e, e, c) = 4. Note that g(e, e, c) ≥ g(a0 , a0 , c). It suffices to prove
that g(e, e, c) ≤ 2.
Since f (e, e, c) = 2e2 + c2 + e2 c = 4, e2 = (4 − c2 )/(2 + c) = 2 − c. We obtain that
as desired.
19. Let a, b, and c be positive real numbers, not all equal. Find all solutions to the system of equations
x2 − yz = a,
y 2 − zx = b,
z 2 − xy = c,
Solution: Squaring each equation and subtracting the product of the other two yields
a2 − bc = x(x3 + y 3 + z 3 − 3xyz),
b2 − ca = y(x3 + y 3 + z 3 − 3xyz),
c2 − ab = z(x3 + y 3 + z 3 − 3xyz).
The same computation that produced the system above shows tha the expression on the left is
a(a3 + b3 + c3 − 3abc), and the latter is positive by the AM-GM inequality. Hence
p
k = ± a3 + b3 + c3 − 3abc
and the solutions to the system (one for each choice of k) are
a2 − bc b2 − ca c2 − ab
x= , y= , and z = .
k k k
20. Find all functions f : N → N satisfying
for all n ∈ N.
Solution: We first look for a function of the form f (n) = n + a. The relation from the state-
ment yields a = 667, and hence f (n) = n + 667 is a solution. Let us show that this is the only
solution.
Fix some positive integer n and define a0 = n, and ak = f (ak−1 0 for k ≥ 1. The sequence {ak }k≥0
satisfies the recursive relation
x3 − 3x2 + 6x − 4 = 0.
21. Let a1 , . . . , an and b1 , . . . , bn be two sequences of distinct numbers such that ai + bj 6= 0 for all i, j.
Show that if ½
Xn
cj,k 1 if i = k
=
ai + bj 0 otherwise,
j=1
Pn
Solution: Let rj = k=1 cj,k . Then
n
X n Pn
X n X
X n n X
X n
rj k=1 cj,k cj,k cj,k
= = = = 1, (1)
ai + bj ai + bj ai + bj ai + bj
j=1 j=1 j=1 k=1 k=1 j=1
Then R(x) = P (x)/Q(x) where Q(x) = (x + b1 )(x + b2 ) · · · (x + bn ) and P (x) has degree at most
n − 1. By (1), R(a1 ) = R(a2 ) = · · · = R(an ) = 1, so if we write
S(x)
R(x) = 1 − ,
Q(x)
then S(x) is a monic polynomial of degree n and S(a1 ) = S(a2 ) = · · · = S(an ) = 0. Hence
S(x) = (x − a1 )(x − a2 ) · · · (x − an ).
Consider the coefficient of xn−1 in P (x) = Q(x) − S(x). Form (2), this coefficient is r1 + r2 + · · · + rn .
On the other hand,
22. Suppose the positive integers have been expressed as a disjoint union of arithmetic progressions
{ai + ndi }∞
n=0 , i = 1, 2, . . . , k. Show that
k
X ai k+1
= .
di 2
i=1
as each power of x appears with coefficient 1 on both sides of the equations. Multiplying both sides
of (1) by (1 − x) yields
X k
xai −1
= 1.
1 + x + · · · + xdi −1
i=1
Setting x = 1 in the last equation gives
k
X 1
= 1.
di
i=1
We now rewriting (1) as
k µ ai −1
X ¶ k
X 1
x −1 1 di
+ = ,
1 − xdi 1 − xdi 1−x
i=1 i=1
that is, Ã !
k
X 1 k
X
1 di 1 + x + · · · + xai −2
− = .
1 − xdi 1−x 1 + x + · · · + xdi −1
i=1 i=1
Hence
k
X k
X
di − 1 − x − · · · − xdi −1 1 + x + · · · + xai −2
= ,
di (1 − xdi ) 1 + x + · · · + xdi −1
i=1 i=1
or
k
X k
X
(1 − x)[(di − 1) + (di − 2)x + · · · + xdi −1 ] 1 + x + · · · + xai −2
= ,
di (1 − x)(1 + x + · · · + xdi −1 ) 1 + x + · · · + xdi −1
i=1 i=1
Setting x = 1 in the last equation yields
k
X k
X
(di − 1) + (di − 2) + · · · + 1 ai − 1
= ,
i=1
d2i i=1
di
or
k
X k
X k
X
di (di − 1) ai 1
= − .
i=1
2d2i i=1
di di
i=1
Hence
k
X k µ
X ¶
ai 1 1 k+1
= + = ,
di 2 2di 2
i=1 i=1
as desired.
23. Let b
P S be the set of perfect powers, that is, the numbers of the form a for some a, b > 1. Show that
n∈S 1/(n − 1) = 1.
But this is just the same as my first sum. Indeed, for any denominator d, the number of times 1/d
occurs in the first sum is the number of representations d = ab (a, b ≥ 2), while the number of times
1/d occurs in the second sum is the number of representations d = nm (n ∈ S, m ≥ 1). If we write
d = pq with p minimal, then both these multiplicities are equal to f (q) − 1, where f (q) denote the
number of divisors of q. So the sums really are the same.
24. Prove that the product of any k consecutive Fibonacci numbers is divisible by the product of the
first k Fibonacci numbers.
Solution: Let [k]! = F1 F2 · · · Fk for k ≥ 1 and [0]! = 1, and let Ra,b = [a + n]!/([a]![b]!)
√ for
a, b ≥ 0. √We need to show that Ra,b is an integer for all a, b ≥ 0. Putting α = (1 + 5)/2 and
β = (1 − 5)/2, we have
αn − β n
Fn = = αn−1 + αn−2 β + · · · + β n−1 (1)
α−β
for all n ≥ 1. From this one can derive the recursice relation
for a, b ≥ 1, since dividing both sides of (2) by [a + b − 1]!/([a]![b]!) converts it into the formula
Fa+b = αa Fb + β b Fa
which in turn follows easily from (1). The recursive relation (2), combined with the initial conditions
Ra,0 = R0,b = 1 for all a, b ≥ 0, guarantees that Ra,b is always expressible as a polynomial in α and
β with integer coefficients.
Now since β = 1 − α, Ra,b can be expressed as a polynomial in α with integer coefficients. Conse-
quently, iteratively applying αn = αn−1 + αn−2 for n ≥ 2, we can write Ra,b in the form r + sα, where
r and s are integers. Note that Ra,b is rational and α is irrational. Hence s = 0 and Ra,b = r is an
integer.
25. Prove that any monic polynomial (a polynomial with leading coefficient 1) of degree n with real
coefficients is the average of two monic polynomials of degree n with n real roots.
First Solution: (Tiankai Liu) Let us begin with the following lemma.
Lemma. For any set of n ordered pairs of real numbers {(xi , yi )}ni=1 , with xi 6= xj for all i 6= j,
there exists a unique monic real polynomial P (x) of degree n such that P (xi ) = yi for all integers
i ∈ [1, n].
We present two proofs of the lemma.
• First proof By the Lagrange Interpolation Formula, there exists a unique real polynomial
Q(x) of degree less than n such that Q(xi ) = yi − xni for all integers i = 1, 2, . . . , n. Then we
can, and must, have P (x) = Q(x) + xn .
• Second proof By the Lagrange Interpolation Formula, there exists a unique real polynomial
Q(x) of degree less than n suchQthat Q(xi ) = yi for all integers i = 1, 2, . . . , n. Then we can,
and must, have P (x) = Q(x) + ni=1 (x − xi ).
22 Zuming Feng ([email protected]), Phillips Exeter Academy, Exeter 03833, USA
Let the given monic real polynomial be F (x). Choose any strictly decreasing sequence of n real
numbers x1 , . . . , xn . For each odd i, choose a yi such that yi < min{0, 2F (xi )}. For each even i,
choose a yi such that yi > max{0, 2F (xi )}.
Let P (x) be the monic real polynomial of degree n such that P (xi ) = yi for all integers i = 1, 2, . . . , n;
the existence of such a P (x) is guaranteed by the lemma. Let Q(x) = 2F (x) − P (x).
Note that, for each integer i = 1, 2, . . . , n − 1, P (x) has a root in (xi+1 , xi ) because P (xi+1 ) and
P (xi ) have opposite signs and polynomial functions are continuous. Similarly, for each integer i =
1, 2, . . . , n − 1, Q(X) has a root in (xi+1 , xi ) because 2F (xi+1 ) − P (xi+1 ) and 2F (xi ) − P (xi ) have
opposite signs.
Therefore, we are guaranteed that P (x) and Q(x) each have n roots with at least n − 1 of them are
real. Because imaginary roots of real polynomials come in conjugate pairs, all real polynomials have
an even number of imaginary roots. Hence the other root of P (x) must also be real, and the same
holds for Q(x). Thus, P (x) and Q(x) each have n real roots.
Note: One could also prove that P (x) and Q(x) both have n real roots by the following argu-
ment. Because P (x) has a positive leading coefficient, limx→∞ P (x) = ∞. Because P (x1 ) < 0, then,
P (x) has another root in (x1 , +∞). Similarly, Q(x) also has a leading coefficient of +1. If n is odd,
then limx→−∞ Q(x) = −∞, so because Q(xn ) > 0, Q(x) has another root in (−∞, xn ). If n is even,
then limx→−∞ Q(x) = ∞, so because Q(xn ) < 0, Q(x) has another root in (−∞, xn ).
Second Solution: Let F (x) be the monic real polynomial of degree n. If n = 1, then F (x) = x + a
for some real number a. Then F (x) is the average of x + 2a and x, each of which has 1 real root.
Now we assume that n > 1. Define the polynomial
G(x) = (x − 2)(x − 4) · · · (x − 2(n − 1)).
The degree of G(x) is n − 1. Consider the polynomials
P (x) = xn − kG(x)
and
Q(x) = 2F (x) − P (x) = 2F (x) − xn + kG(x).
We will show that for large enough k these two polynomials have n real roots. Since they are monic
and their average is F (x), this will solve the problem.
Consider the values of polynomial G(x) at n points x = 1, 3, 5, . . . , 2n − 1. These values alternate
in sign and have magnitude at least 1 (because at most two of the factors have magnitude 1 and
the others have magnitude at least 2). On the other hand, there is a constant c > 0 such that for
0 ≤ x ≤ 2n − 1, we have |xn | < c and |2F (x) − xn | < c. Take k > c. Then we see that P (x) and Q(x)
evaluated at n points x = 1, 3, 5, . . . , 2n − 1 alternate in sign. Thus, polynomials P (x) and Q(x) each
have at least n − 1 real roots — one in each interval (1, 3), . . ., (2n − 3, 2n − 1). However, since they
are polynomials of degree n, they must then each have n real roots (as in the previous solution), as
desired.
26. Prove that for any integer n, there exists a unique polynomial Q with coefficients in {0, 1, . . . , 9} such
that Q(−2) = Q(−5) = n.
Solution: First suppose there exists a polynomial Q with coefficients in {0, 1, . . . , 9} such that
Q(−2) = Q(−5) = n. We shall prove that this polynomial is unique. By the Factor Theorem, we can
Mathematics Olympiad Coachs Seminar, Zhuhai, China 23
write Q(x) = P (x)R(x)+n where P (x) = (x+2)(x+5) = x2 +7x+10 and R(x) = r0 +r1 x+r2 x2 +· · ·
is a polynomial. Then r0 , r1 , r2 , . . . are integers such that
(with the understanding that r−1 = 0). For each k, (∗) uniquely determines rk once rj is known for
all j < k. Uniqueness of R, and therefore of Q, follows.
Existence will follow from the fact that for the unique sequence {rk } satisfying (∗), there exists some
N such that rk = 0 for all k ≥ N . First note that {rk } is bounded, since |r0 |, |r1 | ≤ B and B ≥ 9
imply |rk | ≤ B for all k. This follows by induction, using 10|rk | ≤ 7|rk−1 | + |rk−2 | + 9 ≤ 10B. More
specifically, if ri ≤ M for i = k − 1, k − 2, then
7rk−1 rk−2 4M
rk ≥ − − ≥− ,
10 10 5
while if ri ≥ L for i = k − 1, k − 2, then
7rk−1 rk−2 9 4L 9
rk ≤ − − + ≤− + .
10 10 10 10 10
Since the sequence {rk } is bounded, we can define
we may deduce the result from the induction hypothesis unless we are in the following situation:
For i = 1, . . . , n, let ri = max{ai /bi , bi /ai }. Without loss of generality, we may assume 1 < r1 <
· · · < rn , and that a1 < b1 . Now notice that f (x) = L(a1 , x, a2 , b2 , . . . , an , bn ) is a linear function of
x in the interval [a1 , r2 a1 ]. Explicitly,
f (r2 a1 ) = L(a1 , r2 a1 , a2 , b2 , . . . )
½
L(a1 + a2 , r2 a1 + b2 , a3 , b3 , . . . ) if r2 = b2 /a2 ,
=
L(a2 − r2 a1 , b2 − a1 , a3 , b3 , . . . ) if r2 = a2 /b2 .
Thus we deduce the desired inequality from the induction hypothesis in all cases.
Note: More precisely, it can be shown that for ai , bi > 0, equality holds if and only if, for each
r > 1, the set Sr of indices i in {1, . . . , n} such that ai /bi ∈ {r, 1/r} has the property that
X X
ai = bi .
i∈Sr i∈Sr
Namely, assume this is the case for n − 1 pairs. Given n pairs, if conditions (a)-(c) are not all met, we
may deduce the result from the induction hypothesis by the same reductions as that at the beginning
of the proof. If (a)-(c) are met, then for equality to hold, 0 = f (b1 ) ≥ min{f (a1 ), f (r2 a1 )} ≥ 0.
Since f (x) is linear on the interval [a1 , r2 a1 ], f (x) is identically zero on the interval. Since f (a1 ) = 0,
L(a2 , b2 , . . . ) = 0. Applying the induction hypothesis, with all of ai and bi nonzero ((a) is met) and
ri > 1 ((b) is met), we have
(i) for each i ≥ 2, there exists an r > 1 such that either ai = rbi or bi = rai .
Mathematics Olympiad Coachs Seminar, Zhuhai, China 25
n
X
(ii) (aj − bj ) = 0.
j=2
Therefore, if aj > bj , aj = rbj and cjP (1 − r) = (−bj )(1 − r) = aj − bj ; if aj < bj , raj = bj and
cj (1 − r) = aj (1 − r) = aj − bj . Hence nj=2 cj = 0. But then 0 = f (r2 a1 ) yields
n
X
0 = (r2 a1 − a1 )(a1 + 2 cj ) + L(a2 , b2 , . . . , an , bn ) = (r2 − 1)a21
j=2
and a1 = 0, a contradiction.