Solutions to TSTST 2015

United States of America

57th IMO 2016, Hong Kong

§1 Solution to TSTST Problem 1

This problem was proposed by Mark Sellke.
Define
M (k) = max (ak + ak+1 + · · · + ak+`−1 ),
1≤`≤m

and let N (k) = max(M (k), 0). Note that k ∈ T ⇐⇒ M (k) = N (k). For fixed k0 ∈ T ,
take `0 ≥ 1 achieving the maximum value for M (k0 ), i.e. with

M (k0 ) = (ak0 + · · · + ak0 +`0 −1 ).

We may rewrite this as

ak0 = M (k0 ) − (ak0 +1 + · · · + ak0 +`0 −1 ).

From the definition of N we have now

ak0 = M (k0 ) − (ak0 +1 + · · · + ak0 +`0 −1 ) ≥ M (k0 ) − N (k0 + 1) = N (k0 ) − N (k0 + 1).

Now, divide the multi-set of values N (k) for 1 ≤ k ≤ n into the following three subsets:
the positive, zero, and negative values. The values N (k) for k ∈ T consist of all
positive values, some 0 values, andPno negative values. This means that among all sets
S ⊆ {1, 2, . . . n} with |S| = |T |, N (k) is maximized (perhaps not uniquely) when
k∈S
T = S. In particular, X X
N (k) ≥ N (k + 1).
k∈T k∈T
From this, we conclude the problem statement, as
X X
ak ≥ (N (k) − N (k + 1)) ≥ 0.
k∈T k∈T

§2 Solution to TSTST Problem 2

This problem was proposed by Ivan Borsenco.
Step 1. Let BC = a, AC = b, AB = c. Then the bisector theorem gives Ka B/Ka C =
La B/La C = c/b. Using the fact that Ka B + Ka C = a and |La B − La C| = a, we obtain
ca ca
Ka B = a , La B = a .
b+c |b − c|
Then
a |b − c| a b+c
Ma Ka = , Ma La = .
2 b+c 2 |b − c|

1
Solutions to TSTST 2015

Note also that Ka and La are on the same side of BC. Using the power of Ma with
respect to the circumcircle of AKa La we obtain
a2
Ma Xa · Ma A = Ma Ka · Ma La = = Ma B 2 = Ma C 2 .
4
The first step can be proved differently as follows. Because Xa lies on Apollonius circle
corresponding to vertex A, we have BX AB
CXa = AC . From Stewart’s formulae we have
a

AB 2 BXa2 AB 2 · Xa Ma + BMa2 · AXa − AXa · Xa Ma · AMa

= =
AC 2 CXa2 AC 2 · Xa Ma + CMa2 · AXa − AXa · Xa Ma · AMa
BMa2 · AXa − AXa · Xa Ma · AMa
= ,
CMa2 · AXa − AXa · Xa Ma · AMa

which is possible if and only if BMa2 = CMa2 = Xa Ma · AMa .

Step 2. By step 1 triangles Ma BA and Ma Xa B are similar, and triangles Ma CA and
Ma Xa C are similar. Consequently

∠BXa C = ∠BXa Ma + ∠Ma Xa C = ∠B + ∠C = 180◦ − ∠A.

Note also that Xa is on the same side of Ma as A. So Xa belongs to the circumcircle of

BHC, where H is the orthocenter.
Because Xa is on the median, ∠HXa Ma = 90◦ . This can be proved in two ways: If we
extend AMa to meet the circumcircle of BHC at A0 , then ABC and A0 CB are symmetric
with respect to Ma , which proves that CA0 is parallel to AB and hence perpendicular to
HC. By inscribed angles

∠HXa Ma = ∠HXa A0 = ∠HCA0 = 90◦ .

Or we can argue as follows. Denote by D, E, F the feet of the perpendiculars from

vertices A, B, and C. Consider inversion I with the pole A and power ρ = AH · HD =
AF · F B = AE · EC. Then the circle passing through B, C, H, Xa becomes the nine-point
circle of triangle ABC. It follows that I(Xa ) = Ma and AXa · AMa = AH · HD, yielding
that DHXa Ma is cyclic. Thus ∠HXa G = 90◦ .
We conclude that Xa , Xb , Xc lie on a circle of diameter GH, and hence their circum-
center is on the Euler line GH.

§3 Solution to TSTST Problem 3

This problem was proposed by Alex Zhai.
For a prime q, we say a residue mod q is q-abundant if it appears in M infinitely often.
S⊂M
Suppose for sake of contradiction that there is a prime q in P but not in M . Let Q
be the (finite) subset of primes whose residues are not q-abundant. Define X = p∈S p,
and let a be the remainder of X modulo q.
Let H ⊂ Z/q denote the subset of residues that can be obtained by multiplying finitely
many (not necessarily distinct) q-abundant residues. The definition Q of H ensures that
for any h ∈ H, we can find a finite subset
Q U h ⊂ M \ S such that p∈Uh p ≡ h (mod q).
Now, consider the number N = X · p∈Uh p + 1. Each prime factor of N must be in M .
However, N is relatively prime to X, so it can only have prime factors whose residues
are abundant. It follows that the residue of N modulo q is in H. Computing this, we
find that Y
N ≡X· p + 1 ≡ a · h + 1 (mod q),
p∈Uh

2
Solutions to TSTST 2015

so ah + 1 ∈ H.
This holds for each h ∈ H, so h 7→ ah + 1 maps H to itself. Moreover, this map is a
bijection, because

ah + 1 ≡ ah0 + 1 (mod q) =⇒ h ≡ h0 (mod q).

However, clearly 1 ∈ H and

ah + 1 ≡ 1 (mod q) =⇒ h ≡ 0 (mod q),

which is a contradiction.

§4 Solution to TSTST Problem 4

This problem was proposed by Alyazeed Basyoni.
We prove that the condition x4 + y 4 + z 4 + xyz = 4 implies
√ y+z
2−x≥ .
2
We first establish that 2 − x ≥ 0. Indeed, AM-GM gives that

3x4
 4 
x
5 = x4 + y 4 + (z 4 + 1) + xyz = + + y 4 + (z 4 + 1) + xyz
4 4
3x 4
≥ + x2 y 2 + 2z 2 + xyz.
4
We evidently have that x2 y 2 + 2z 2 + xyz ≥ 0 because the quadratic form a2 + ab + 2b2 is
positive definite, so x4 ≤ 203 =⇒ x ≤ 2. Now, the desired statement is implied by its
square, so it suffices to show that
 2
y+z
2−x≥
2

Assume for contradiction the reverse inequality holds. This rearranges to

4x + y 2 + 2yz + z 2 > 8.
y 4 +1
By AM-GM, we have x4 + 3 ≥ 4x and 2 ≥ y 2 which yields that

y4 + z4 y4 + z4
x4 + + 2yz + 4 > 8 =⇒ x4 + + 2yz > 4.
2 2
Subtracting the given condition x4 + y 4 + z 4 + xyz = 4 now gives

y4 + z4 y4 + z4
− + (2 − x)yz > 0 =⇒ (2 − x)yz > .
2 2
Since 2 − x and the right-hand side are positive, we have yz ≥ 0. So, we have
2
y4 + z4

y+z
<2−x< =⇒ 2y 4 + 2z 4 < yz(y + z)2 = y 3 z + 2y 2 z 2 + yz 3 .
2yz 2

This is clearly false by AM-GM, so we have a contradiction.

3
Solutions to TSTST 2015

§5 Solution to TSTST Problem 5

This problem was proposed by Iurie Boreico.
Let a be a positive integer. Define a relation on the integers by saying that two
numbers are a-equal iff their prime factorizations only differ by primes strictly less than
a. (Thus for primes p greater than or equal to a, two a-equal numbers are divisible by
the same power of p.) We will say that a set of integers {n1 , n2 , . . . , nk } is a-distinct
if no two of them are a-equal. Finally, we say that {n1 , n2 , . . . , nk } are a-compatible if
φ(n1 ), . . . , φ(nk ) are all a-equal.

Claim. If there exist k a-distinct, a-compatible integers x1 , . . . , xk , then there exist k

distinct integers n1 , . . . , nk such that φ(n1 ), . . . , φ(nk ) are all equal.

Proof. Suppose that x1 , . . . , xk are a-distinct and a-compatible. Let p be the first
prime smaller than a. Then x1 , . . . , xk are p-distinct, but may not be p-compatible,
because φ(x1 ), . . . , φ(xk ) may be divisible by different powers of p. However, if we
multiply x1 , . . . , xk by appropriate powers of p, they will become p-compatible. Thus,
continuing by downward induction on the primes less than a, we will arrive at integers
n1 , . . . , nk which are 2-distinct and 2-compatible, which is the same as being distinct
with φ(n1 ), . . . , φ(nk ) all equal.
Now set a = 2k (we can eventually take k = 2015). By Bertrand’s postulate, there
exist k primes p1 , . . . , pk between a and a2 . We will take p1 , . . . , pk to be the smallest
primes larger than a. Consider one of these primes pi . Then φ(pi ) = pi − 1, so that
a ≤ φ(pi ) < a2 , so that φ(pi ) is divisible by at most one other prime among the other
k − 1 primes, and all the other factors of φ(pi ) are less than a. Thus we can form a
directed graph with an arrow from pi to pj iff pj divides φ(pi ). Clearly this graph is a
union of connected components, each of which is a tree.
For each component Cr containing cr primes, we will find cr +1 a-distinct, a-compatible
integers. Now let p1 , . . . , pcr be the primes in Cr . If there is an arrow from pi to pjQ , define
pˆi = pj . If pi has no arrow coming out of it, define pˆi = 1. Now consider P0 = ci=1 r
pi .
pˆi
Let Pi = P0 · pi . By construction, P0 , P1 , . . . , Pcr are all a-distinct: P0 is divisible by
all the primes in Cr , while each of the other Pi is missting a different prime. Moreover,
P0 , P1 , . . . , Pcr are all a-compatible, because φ(pi · pˆi ) and φ(pˆi 2 ) are a-compatible.
Thus each component Cr gives rise to cr + 1 a-distinct, a-compatible integers. Let
us now call them Pr,0 , Pr,1 , . . . , Pr,cr . Note that the prime factors of these integers are
completely disjoint from component to component. Q ConsiderP all the products consisting
of one Pr,i for each value of r. There are (cr + 1) > cr = k of these products.
Moreover, because Pr,i is relatively prime to Pr0 ,i0 whenever r 6= r0 , these products are all
a-compatible and a-distinct. Thus we have found at least k a-distinct and a-compatible
integers, and we are done.

There are many ways of finding 2015 a-compatible, a-distinct integers. For example,

5, 7, 13, 17, 19, 37, 73, 97, 109, 163, 193, 257

are a set of twelve 4-distinct, 4-compatible integers. Taking all 212 combinations of
products gives 212 > 2015 4-distinct, 4-compatible integers. Similarly,

7, 11, 13, 17, 19, 31, 37, 41, 61, 73, 97, 101

are all 6-compatible and 6-distinct.

4
Solutions to TSTST 2015

§6 Solution to TSTST Problem 6

This problem was proposed by Linus Hamilton.
Here we present a solution with 14 registers and 22 moves. Initially X = n and all
other variables are zero.
0
X Y Go 0
SX SX SX SY0 SY SY0 Cl A B Die Die’
Init -1 1 1 1 1
Begin 1 -1 1 -1 1
Sleep 1 -1
StartX -1 1 -1 1
WorkX -1 -1 1 -1 1
WorkX’ -1 1 1 -1 -1 1
DoneX -1 1 -1 1
WrX -1 -1 -1 1 1 -1
StartY -1 1 -1 1
WorkY -1 -1 1 -1 1
WorkY’ 1 -1 1 -1 -1 1
DoneY 1 -1 -1 1
WrY -1 -1 1 -1
ClaimX -1 -1 1 -1 1
ClaimY -1 -1 1 -1 1
FakeX -1 -1 -1
FakeY -1 -1 -1
Win -1 -1
PunA -2
PunB -1 -1
Kill -1 -2 1
Kill’ -1 1 -2

Now, the “game” is played as follows. First, note that Alice must play Init, and then
Bob must play Sleep. From then on the mechanics are controlled by the turn counters A
and B. To be precise, we say that the game is

• In the main part if A + B = 1, and no one has played Init a second time.

• In the death part otherwise.

Observe that in the main state, on Alice’s turn we always have (A, B) = (1, 0) and on
Bob’s turn we always have (A, B) = (0, 1).

Claim. A player whose move leaves the game in the death part must lose.

Proof. • Suppose the offending player is in a situation where (A, B) = (0, 0). Then
he/she must play Init. At this point, the opposing player can respond by playing
Kill. Then the offending player must play Init again. The opposing player now
responds with Kill’. This iteration continues until InX reaches a negative number
and the offending player loses.

• Suppose Alice has (A, B) = (1, 0) but plays Init again anyways. Then Bob responds
with PunA to punish her; he then wins as in the first case.

• Suppose Bob has (A, B) = (0, 1) but plays Init again anyways. Alice responds with
PunB in the same way.
Situations with A + B ≥ 2 cannot occur during main part, so this is all cases.

5
Solutions to TSTST 2015

Now we return to analysis of the main part. Observe the main part starts with Alice
playing Init, Bob playing Sleep, and then Alice playing Begin (thus restoring the value of
n in X), then Bob playing Sleep.
Thereafter we say the game is in state S for S ∈ {SX0 , S , S 0 , S 0 , S , S 0 , Cl} if S = 1
X X Y Y Y
and all other variables are zero. By construction, this is always the case. From then on
the main part is divided into several phases:

• An X-phase: this begins with Alice at SX

0 , and ends when the game is in a state
0 . (She can never return to S 0 during an X-phase.)
other than SX and SX X

• A Y -phase: this begins with Alice at SY0 , and ends when the game is in a state
other than SY and SY0 . (She can never return to SY0 during a Y -phase.)

Claim. Consider an X-phase in which (X, Y ) = (x, 0), x > 1. Then Alice can complete
the phase without losing if and only if x is even; if so she begins a Y -phase with
(X, Y ) = (0, x/2).

Proof. As x > 1, Alice cannot play ClaimX since Bob will respond with FakeX and win.
Now by alternating between WorkX and WorkX’, Alice can repeatedly deduct 2 from X
and add 1 to Y , leading to (x − 2, y + 1), (x − 4, y + 2). (During this time, Bob can only
play Sleep.) Eventually, she must stop this process by playing DoneX, which begins a
Y -phase.
Now note that unless X = 0, Bob now has a winning move WrY (“wrong Y -phase”).
Conversely he may only play Sleep if X = 0.

We have an analogous claim for Y -phases. Thus if n is not a power of 2, we see that
Alice eventually loses.
Now suppose n = 2k is a power of 2, then Alice reaches (X, Y ) = (0, 2k−1 ), (2k−2 , 0), . . .
until either reaching (1, 0) or (0, 1). At this point she can play ClaimX or ClaimY, re-
spectively; the game is now in state Cl. Bob cannot play either FakeX or FakeY, so he
must play Sleep, and then Alice wins by playing Win. Thus Alice has a winning strategy
when n = 2k .