1

st
United States of America Junior Mathematical Olympiad 2010
1. Solution from Andy Niedermier: Every integer in [n] can be
uniquely written in the form x
2
q, where q either 1 or square free,
that is, a product of distinct primes. Let q denote the set {1
2
q, 2
2

q, 3
2
q, . . .} [n].
Note that for f to satisfy the square-free property, it must permute
q for every q = 1, 2, 3, . . . . To see this, notice that given an arbitrary
square-free q, in order for qf(q) to be a square, f(q) needs to contribute
one of every prime factor in q, after which it can take only even powers
of primes. Thus, f(q) is equal to the product of q and some perfect
square.
The number of f that permute the q is equal to

qn
q is square-free
__
n
q
_
!
For 2010 = 2 3 5 67 to divide P(n), we simply need 67! to appear in
this product, which will rst happen in 1 so long as
_
n/q 67 for
some n and q. The smallest such n is 67
2
= 4489.
This problem was proposed by Andy Niedermier.
2. Solution from Razvan Gelca: There is a unique sequence 2, 4, 6, . . . , 2n
2 satisfying the conditions of the problem.
Note that (b) implies x
i
< 2n for all i. We will examine the possible
values of x
1
.
If x
1
= 1, then (c) implies that all numbers less than 2n should be
terms of the sequence, which is impossible since the sequence has only
n 1 terms.
If x
1
= 2, then by (c) the numbers 2, 4, 6, . . . , 2n 2 are terms of the
sequence, and because the sequence has exactly n 1 terms we get
x
i
= 2i, i = 1, 2, . . . , n 1. This sequence satises conditions (a) and
(b) as well, so it is a solution to the problem.
For x
1
3, we will show that there is no sequences satisfying the
conditions of the problem. Assume on the contrary that for some n
there is such a sequence with x
1
3. If n = 2, the only possibility is
x
1
= 3, which violates (b). If n = 3, then by (a) we have the possibilities
(x
1
, x
2
) = (3, 4), or (3, 5), or (4, 5), all three of which violate (b). Now
we assume that n > 3. By (c), the numbers
x
1
, 2x
1
, . . . ,
_
2n
x
i
_
x (1)
are terms of the sequence, and no other multiples of x
1
are. Because
x
1
3, the above accounts for at most
2n
3
terms of the sequence. For
n > 3, we have
2n
3
< n 1, and so there must be another term besides
the terms in (1). Let x
j
be the smallest term of the sequence that does
not appear in (1). Then the rst j terms of the sequence are
x
1
, x
2
= 2x
1
, . . . , x
j1
= (j 1)x
1
, x
j
, (2)
and we have x
j
< jx
1
. Condition (b) implies that the last j terms of
the sequence must be
x
nj
= 2n x
j
, x
nj+1
= 2n (j 1)x
1
, . . . ,
x
n2
= 2n 2x
1
, x
n1
= 2n x
1
.
But then x
1
+ x
nj
< x
1
+ x
n1
= 2n, hence by condition (c) there
exists k such that x
1
+x
nj
= x
k
. On the one hand, we have
x
k
= x
1
+x
nj
= x
1
+ 2n x
j
= 2n (x
j
x
1
)
> 2n (jx
1
x
1
) = 2n (j 1)x
1
= x
nj+1
.
One the other hand, we have
x
k
= x
1
+x
nj
< x
1
+x
nj+1
= x
nj+2
.
This means that x
k
is between two consecutive terms x
nj+1
and x
nj+2
,
which is impossible by (a). (In the case j = 2, x
k
> x
nj+1
= x
n1
,
which is also impossible.) We conclude that there is no such sequence
with x
1
3.
Remark. This problem comes from the study of Weierstrass gaps in
the theory of Riemann surfaces.
Alternate Solution from Richard Stong: Assume that x
1
, x
2
, . . . , x
n1
is a sequence satisfying the conditions of the problem. By condition
(a), the following terms
x
1
, 2x
1
, x
1
+x
2
, x
1
+x
3
, x
1
+x
4
, . . . , x
1
+x
n2
form an increasing sequence. By condition (c), this new sequence is
a subsequence of the original sequence. Because both sequences have
exactly n1 terms, these two sequences are identical; that is, 2x
1
= x
2
and x
1
+ x
j
= x
j+1
for 2 j n 2. It follows that x
j
= jx
1
for
1 j n 1. By condition (b), we conclude that (x
1
, x
2
, . . . , x
n1
) =
(2, 4, . . . , 2n 2).
Remark. The core of the second solution is a result due to Freiman:
Let A be a set of positive integers. Then the set A+A = {a
1
+
a
2
| a
1
, a
2
A} has at least 2|A| 1 elements and equality
holds if and only if A is a set of an arithmetic progression.
Freimans theorem and its generalization below are very helpful in
proofs of many contest problems, such as, USAMO 2009 problem 2,
IMO 2000 problem 1, and IMO 2009 problem 5.
Let A and B be nite nonempty subsets of integers. Then
the set A+B = {a+b | a A, b B} has at least |A|+|B|1
elements. Equality holds if and only if either A and B are
arithmetic progressions with equal dierence or |A| or |B| is
equal to 1.
This problem was suggested by Razvan Gelca.
3. Solution by Titu Andreescu: Let T be the foot of the perpendicular
from Y to line AB. We note the P, Q, T are the feet of the perpen-
diculars from Y to the sides of triangle ABX. Because Y lies on the
circumcircle of triangle ABX, points P, Q, T are collinear, by Simsons
theorem. Likewise, points S, R, T are collinear.
% $
;
<
=
3
4
5
6
7
2
We need to show that XOZ = 2PTS or
PTS =
XOZ
2
=

XZ
2
=

XY
2
+

Y Z
2
= XAY +ZBY = PAY +SBY.
Because PTS = PTY +STY , it suces to prove that
PTY = PAY and STY = SBY ;
that is, to show that quadrilaterals APY T and BSY T are cyclic, which
is evident, because APY = ATY = 90

and BTY = BSY =

90

.
Alternate Solution from Lenny Ng and Richard Stong: Since
Y Q, Y R are perpendicular to BX, AZ respectively, RY Q is equal to
the acute angle between lines BX and AZ, which is
1
2
(

AX +

BZ) =
1
2
(180

XZ) since X, Z lie on the circle with diameter AB. Also,
AXB = AZB = 90

and so PXQY and SZRY are rectangles,

whence PQY = 90

Y XB = 90

Y B/2 and Y RS = 90

AZY = 90

AY /2. Finally, the angle between PQ and RS is

PQY +Y RS RY Q = (90

Y B/2) + (90

AY /2) (90

XZ /2)
=

XZ /2
= (XOZ)/2,
as desired.
This problem was proposed by Titu Andreescu.
4. Solution from Zuming Feng:
Let A = (a, a
2
), B = (b, b
2
), and C = (c, c
2
), with a < b < c. We have

AB = [b a, b
2
a
2
] and

AC = [c a, c
2
a
2
]. Hence the area of
triangle ABC is equal to
[ABC] = (2
n
m)
2
=
|(b a)(c
2
a
2
) (c a)(b
2
a
2
)|
2
=
(b a)(c a)(c b)
2
.
Setting b a = x and c b = y (where both x and y are positive
integers), the above equation becomes
(2
n
m)
2
=
xy(x +y)
2
. (3)
If n = 0, then (m, x, y) = (1, 1, 1) is clearly a solution to (3). If n 1,
it is easy to check that,
(m, x, y) =
_
(2
4n2
1, 2
2n+1
, (2
2n1
1)
2
)
_
satises (3).
Alternate Solution from Jacek Fabrykowski:
The beginning is the same up to (2
n
m)
2
=
xy(x+y)
2
. If n = 0, we take
m = x = y = 1. If n = 1, we take m = 3, x = 1, y = 8. Assume
that n 2. Let a, b, c be a primitive Pythagorean triple with b even.
Let b = 2
r
d where d is odd and r 2. Let x = 2
2k
, y = 2
2k
b and
z = 2
2k
c where k 0. We let m = adc and r = 2 if n = 3k + 2, r = 3
if n = 3k + 3 and r = 4 if n = 3k + 4.
Assuming that x = a 2
s
, y = b 2
2
, other triples are possible:
(a) If n = 3k, then let m = 1 and x = y = 2
2k
.
(b) If n = 3k + 1, then take m = 3, x = 2
2k
, y = 2
2k+3
.
(c) If n = 3k + 2, then take m = 63, x = 49 2
2k
, and y = 2
2k+5
.
This problem was suggested by Zuming Feng.
5. Solution from Gregory Galperin:
Let us create the following 1006 permutations X
1
, ..., X
1006
, the rst
1006 positions of which are all possible cyclic rotations of the sequence
1, 2, 3, 4, ..., 1005, 1006, and the remaining 1004 positions are lled ar-
bitrarily with the remaining numbers 1006, 1007, ..., 2009, 2010:
X
1
= 1, 2, 3, 4, ... , 1005, 1006, , , ... , ;
X
2
= 2, 3, 4, ... , 1005, 1006, 1, , , ... , ;
X
3
= 3, 4, ... , 1005, 1006, 1, 2, , , ... , ;

X
1006
= 1006, 1, 2, 3, 4, ... , 1005, , , ... , .
We claim that at least one of these 1006 sequences has the same integer
at the same position as the initial (unknown) permutation X.
Suppose not. Then the set of the rst (leftmost) integers in the permu-
tation X contains no integers from 1 to 1006. Hence it consists of the
1004 integers in the range from 1007 to 2010 only. By the pigeon-hole
principle, some two of the integers from the permutation X must be
equal, which is a contradiction: there are not two identical integers in
the permutation X.
Consequently, the permutation X has at last one common element with
some sequence X
i
, i = 1, . . . 1006 and we are done.
This problem was proposed by Gregory Galperin.
6. Solution from Zuming Feng: The answer is no, it is not possible
for segments AB, BC, BI, ID, CI, IE to all have integer lengths.
Assume on the contrary that these segments do have integer side lengths.
We set = ABD = DBC and = ACE = ECB. Note that
I is the incenter of triangle ABC, and so BAI = CAI = 45

.
Applying the Law of Sines to triangle ABI yields
AB
BI
=
sin(45

+)
sin 45

= sin + cos ,
by the addition formula (for the sine function). In particular, we con-
clude that s = sin + cos is rational. It is clear that + = 45

.
By the subtraction formulas, we have
s = sin(45

) + cos(45

) =

2 cos ,
from which it follows that cos is not rational. On the other hand,
from right triangle ACE, we have cos = AC/EC, which is rational by
assumption. Because cos cannot not be both rational and irrational,
our assumption was wrong and not all the segments AB, BC, BI, ID,
CI, IE can have integer lengths.
Alternate Solution from Jacek Fabrykowski: Using notations as
introduced in the problem, let BD = m, AD = x, DC = y, AB = c,
BC = a and AC = b. The angle bisector theorem implies
x
b x
=
c
a
and the Pythagorean Theorem yields m
2
= x
2
+ c
2
. Both equations
imply that
2ac =
(bc)
2
m
2
c
2
a
2
c
2
and since a
2
= b
2
+c
2
is rational, a is rational too (observe that to reach
this conclusion, we only need to assume that b, c, and m are integers).
Therefore, x =
bc
a+c
is also rational, and so is y. Let now (similarly to the
notations above from the solution by Zuming Feng) ABD = and
ACE = where + = /4. It is obvious that cos and cos are
both rational and the above shows that also sin = x/m is rational.
On the other hand, cos = cos(/4) = (

2/2)(sin +sin ), which

is a contradiction. The solution shows that a stronger statement holds
true: There is no right triangle with both legs and bisectors of acute
angles all having integer lengths.
Alternate Solution from Zuming Feng: Prove an even stronger
result: there is no such right triangle with AB, AC, IB, IC having ra-
tional side lengths. Assume on the contrary, that AB, AC, IB, IC have
rational side lengths. Then BC
2
= AB
2
+AC
2
is rational. On the other
hand, in triangle BIC, BIC = 135

. Applying the law of cosines to

triangle BIC yields
BC
2
= BI
2
+CI
2

2BI CI
which is irrational. Because BC
2
cannot be both rational and irra-
tional, we conclude that our assumption was wrong and that not all of
the segments AB, AC, IB, IC can have rational lengths.
This problem was proposed by Zuming Feng.
Copyright c Mathematical Association of America