Combinatorics A Solutions

1. Select two distinct diagonals at random from a regular octagon. What is the probability that
the two diagonals intersect at a point strictly within the octagon? Express your answer as
a + b, where the probability is ab and a and b are relatively prime positive integers.
Proposed by: Rishi Dange
Answer: 26
7
The answer is 19 . We use complementary counting.
Split into three cases in which the diagonals don’t intersect strictly within the octagon.
Case 1: The diagonals intersect on the octagon. For each vertex, there are 5 possible diagonals
from which we must choose 2 in this case. This gives us 8 52 = 80 total ways.

Case 2: The diagonals never intersect. Take a diagonal separated by 1 vertex. There are 2
other

diagonals parallel to this one. There are 4 distinct possible rotations, giving a total of
4 32 = 12 possibilities. The only other way for the diagonals to not intersect is to take a
diagonal separated by 2 vertices, in which case there is only 1 parallel choice and 4 rotations.
Thus, we have a total of 12 + 4 = 16 ways for this case.
Case 3: The diagonals intersect outside of the octagon. There are two subcases here. First, we
can have two non-parallel diagonals that are each separated by 1 vertex. By rotation, there
are 8 ways for this. Second, we can have one diagonal separated by 1 vertex and one separated
by 2. For each possible diagonal separated by 1 vertex, there are 2 possibilities for the other
diagonal. Thus, we have 8 × 2 = 16 possibilities here. That gives us a total of 8 + 16 = 24
ways for this case.
The total number of ways to select diagonals that do not intersect

strictly within the octagon
20
is 80 + 16 + 24 = 120. There are 8×5
2 = 20 diagonals, giving 2 = 190 total choices. Thus,
120 7
by complementary counting, we arrive at a final answer of 1 − 190 = 19 .
2. Eighteen people are standing in a (socially-distanced) line to enter a grocery store. Five people
are wearing a black mask, 6 are wearing a gray mask, and 7 are wearing a white mask. Suppose
that these 18 people got on line in a random order. The expected number of pairs of adjacent
people wearing different-colored masks can be given by ab , where gcd(a, b) = 1. Compute a + b.
Proposed by: Nancy Xu
Answer: 116
For each pair of adjacent people, the probability that they are wearing different-colored masks
5
is 18 · 13 6 12 7 11 107
17 + 18 · 17 + 18 · 17 = 153 . There are 17 pairs of adjacent people, so the expected value
107 107
is 153 · 17 = 9 , and our answer is 107 + 9 = 116.
3. Nelson is having his friend drop his unique bouncy ball from a 12 foot building, and Nelson
will only catch the ball at the peak of its trajectory between bounces. On any given bounce,
there is an 80% chance that the next peak occurs at 13 the height of the previous peak and a
20% chance that the next peak occurs at 3 times the height of the previous peak (where the
first peak is at 12 feet). If Nelson can only reach 4 feet into the air and will catch the ball
as soon as possible, the probability that Nelson catches the ball after exactly 13 bounces is
2a × 3b × 5c × 7d × 11e for integers a, b, c, d, and e. Find |a| + |b| + |c| + |d| + |e|.
Proposed by: Rishi Dange
Answer: 31

1
 Call an 80% bounce A and a 20% bounce B. It’s easy to see that for 13 bounces to occur before
ending, there must be 7 As and 6 Bs. Now we simply need to find the number of orderings of
the As and Bs that are possible.
The first bounce must be B and the last two must both be A, otherwise Nelson will catch the
ball before 13 bounces. For the remaining 10 slots, there must be 5 As and 5 Bs such that
within those 10, there is never an instance where the number of As exceeds the number of
Bs by 2 or more (not including the bounce before those 10). That way, Nelson won’t catch
the ball before the 13 bounce total. This is a simple Catalan counting problem that gives 132
orderings, or C(6).
Thus, the final probability is ( 45 )7 × ( 15 )6 × 132, or 216 × 31 × 5−13 × 70 × 111 .
4. There are n lilypads in a row labeled 1, 2, ..., n from left to right. Fareniss the Frog picks a
lilypad at random to start on, and every second she jumps to an adjacent lilypad; if there are
two such lilypads, she is twice as likely to jump to the right as to the left. After some finite
number of seconds, there exists two lilypads A and B such that Fareniss is more than 1000
times as likely to be on A as she is to be on B. What is the minimal number of lilypads n
such that this situation must occur?
Proposed by: Austen Mazenko
Answer: 12
This situation is modeled by a Markov chain; calculating the equilibrium distribution for each
state gives the probabilities as (1/3)n−2 , (1/3)n−3 , 2 ∗ (1/3)n−3 , ...2n−3 ∗ (1/3)n−3 , (2/3)n−2 .
The maximum is (2/3)n−2 and the minimum is (1/3)n−3 , and their ratio is 2n−2 × 3. This
must exceed 1000, so n must be at least 12.
5. A Princeton slot machine has 100 pictures, each equally likely to occur. One is a picture
of a tiger. Alice and Bob independently use the slot machine, and each repeatedly makes
independent plays. Alice keeps playing until she sees a tiger, at which point she stops. Similarly,
Bob keeps playing until he sees a tiger. Given that Bob played twice longer than Alice, let the
expected number of plays for Alice be ab with a, b relatively prime positive integers. Find the
remainder when a + b is divided by 1000.
Proposed by: Sunay Joshi
Answer: 701
For simplicity, let M = 100. Let the random variables NA and NB denote the number of plays
for Alice and Bob, respectively. We want to find the conditional expectation E = E(NA |NB =
2NA ). By the Tower Law, we have
∞
X ∞
X
E = E(NA |NB = 2NA ) = E(NA |NA = k, NB = 2NA )P (NA = k|NB = 2NA ) = kP (NA = k|NB = 2NA ).
k=1 k=1

By definition,
P (NA = k, NB = 2k) P (NA = k, NB = 2k)
P (NA = k|NB = 2NA ) = =P .
P (NB = 2NA ) ℓ≥1 P (NA = ℓ, NB = 2ℓ)

k−1 1 M −1 2k−1
As P (NA = k) = MM−1 1


M and P (NB = 2k) = M M, we find P (NA = k, NB =

3k
2k) = MM−1 1
(M −1)2 . Thus our denominator becomes

X X  M − 1 3ℓ 1
P (NA = ℓ, NB = 2ℓ) =
M (M − 1)2
ℓ≥1 ℓ≥1

2
 M −1 3


1 M (M − 1)3 1
= 2
· 3 = 3 .
(M − 1) 1 − M −1 M − (M − 1)3 (M − 1)2


M
Thus our conditional probability is given as

M −1 3k 1


M (M −1)2
P (NA = k|NB = 2NA ) = (M −1)3 1
M 3 −(M −1)3 (M −1)2

3k
M 3 − (M − 1)3

M −1
= .
(M − 1)3 M
Therefore our expected value reduces to
3k ∞
M 3 − (M − 1)3 M 3 − (M − 1)3 X k

X M −1
E= k· = kr ,
(M − 1)3 M (M − 1)3
k≥1 k=1

3
where r = MM−1 . By differentiating the geometric series 1−r
1
= 1+r+r2 +. . . and multiplying
P∞ k r
by r, we see that k=1 kr = (1−r)2 , hence the desired expectation is

M 3 − (M − 1)3 r M3
E= = .
(M − 1)3 (1 − r)2 M 3 − (M − 1)3
a 1000000
For M = 100, the reduced fraction is b = 29701 . Thus a + b ≡ 701 (mod 1000), our answer.
6. Alice, Bob, and Carol are playing a game. Each turn, one of them says one of the 3 players’
names, chosen from {Alice, Bob, Carol} uniformly at random. Alice goes first, Bob goes
second, Carol goes third, and they repeat in that order. Let E be the expected number of
names that are have been said when, for the first time, all 3 names have been said twice. If
E= m n for relatively prime positive integers m and n, find m + n. (Include the last name to
be said twice in your count.)
Proposed by: Sunay Joshi
Answer: 383
Let Em,n denote the expected number of additional names that must be said until all 3 names
have been said twice, starting with m names said once, n names said at least twice, and
3 − m − n names said 0 times.
We know that E0,3 = 0, and we wish to find E = E0,0 .
We derive a recurrence relation for Em,n . If a name is said for the first time on the next turn,
then it contributes 3−m−n
3 (Em+1,n + 1) to the expected value. Similarly, if one of the m names
already said once is said again, we get a term of m 3 (Em−1,n+1 + 1). Finally, if one of the n
names already said twice is said again, we get n3 (Em,n + 1). Adding these up, we find the
recurrence
3−m−n m n
Em,n = 1 + Em+1,n + Em−1,n+1 + Em,n ,
3 3 3
which simplifies to
3−n 3−m−n m
Em,n = Em+1,n + Em−1,n+1 + 1.
3 3 3
347
Solving the resulting system of 9 equations, we find that E = E0,0 = 36 , and our answer is
m + n = 347 + 36 = 383.

3
7. Cassidy has string of n bits, where n is a positive integer, which initially are all 0s or 1s. Every
second, Cassidy may choose to do one of two things:
1. Change the first bit (so the first bit changes from a 0 to a 1, or vice versa)
2. Change the first bit after the first 1.
Let M be the minimum number of such moves it takes to get from 1 . . . 1 to 0 . . . 0 (both of
length 12), and N the number of starting sequences with 12 bits that Cassidy can turn into
all 0s. Find M + N.
Proposed by: Frank Lu
Answer: 6826
To show that one such sequence exists, we employ induction. Our base cases are n = 1 and
n = 2, which are clear. Now, suppose that it is possible to do this for all n ≤ k. For the string
with k + 1 bits, we can perform the following procedure using our inductive hypothesis. First,
turn the first k − 1 bits into all zeroes. Then, convert the k + 1st bit into a zero using the
second step. From here, turn the first k − 1 bits back into ones; this is possible from the fact
that both of these steps are involutions, so they their own inverses (so reversing the moves
used to go from all 1s to all 0s will perform this). The result is k 1s followed by a 0, which
again can be turned into all 0s by the inductive hypothesis. Thus, we have N = 4096.
To compute M, notice that in order to turn our sequence into all zeroes, we need to reach the
state 00 . . . 011, in order to turn the last bit into a 0. Let an be the minimal number of moves
it takes to go from 11 . . . 1 to 00 . . . 0, and bn the minimal number of moves it takes to go from
00 . . . 01 to 00 . . . 0. Then, we see that an = an−2 + 1 + bn−1 , using the above logic. We just
need to find bn now.
For bn , in order to reach all zeroes, the same logic tells us that we need to first go from 00 . . . 01
to 00 . . . 011, which takes bn−1 moves. Applying the logic above yields us that bn = 2bn−1 + 1.
Substituting in the first recurrence yields us with the linear recurrence an+1 − 2an − an−1 +
2an−2 = 0. Using the theory of linear recurrences, we know that this takes the form an =
c1 r1n + c2 r2n + c3 r3n , where the ri are roots of the equation r3 − 2r2 − r + 2 = 0. But these roots
are 2, 1, −1. Therefore, we have that an = c1 2n + c2 + c3 (−1)n .
We just need to find the value of these ci . It’s not hard to manually compute that a1 = 1, a2 =
2, a3 = 5. Substituting these values into the above recurrence and solving the resulting system
n+1 n+1
yields an = 2 3 − 12 + (−1)6 . Substituting n = 12 yields a12 = 8192 1 1 8190
3 − 2 − 6 = 3 = 2730.
So our answer is 4096 + 2730 = 6826.
8. Physicists at Princeton are trying to analyze atom entanglement using the following experi-
ment. Originally there is one atom in the space and it starts splitting according to the following
procedure. If after n minutes there are atoms a1 , ..., aN , in the following minute every atom ai
(1) (2) (3) (4) (j) (j)
splits into four new atoms, ai , ai , ai , ai . Atoms ai and ak are entangled if and only
(j) (j+1)
if the atoms ai and ak were entangled after n minutes. Moreover, atoms ai and ak are
entangled for all 1 ≤ i, k ≤ N and j = 1, 2, 3. Therefore, after one minute there is 4 atoms,
after two minutes there are 16 atoms and so on.
Physicists are now interested in the number of unordered quadruplets of atoms {b1 , b2 , b3 , b4 }
among which there is an odd number of entanglements. What is the number of such quadru-
plets after 3 minutes?
Remark. Note that atom entanglement is not transitive. In other words, if atoms ai , aj
are entangled and if aj , ak are entangled, this does not necessarily mean that ai and ak are
entangled.
Proposed by: Aleksa Milojevic

4
Answer: 354476
Let Gn be the graph arising after n minutes and let f (n) be the number of quadruples
b1 , b2 , b3 , b4 with an odd number of edges in their induced graph. We establish a recursive
relation on f (n).
(1) (2)
The idea is to split naturally Gn into 4 parts: let A = {ai |1 ≤ i ≤ 4n−1 }, B = {ai |1 ≤ i ≤
(3) (4)
4n−1 }, C = {ai |1 ≤ i ≤ 4n−1 }, and D = {ai |1 ≤ i ≤ 4n−1 }.
Let b1 , b2 , b3 , b4 be an arbitrary quadruple. We analyze this quadruple based on the sets
A, B, C, D in which b1 , b2 , b3 , b4 lie and split into cases. For example, if all bi ∈ A, we denote
this case as (A, A, A, A).
Case 1: (A, A, A, A). In this case, the answer is f (n − 1), inductively. Similarly, for all other
cases (B, B, B, B), (C, C, C, C), (D, D, D, D) we have the same answer.
Case 2: (A, B, B, B). As the induced graphs on A, B, C, D are isomorphic, fix an isomorphism
φ taking B to C. Then, the idea is that exactly half of all quadruplets on the sets (A, B, B, B) or
the sets (A, C, C, C) have an odd number of edges. Why is this? If a quadruple (b1 , b2 , b3 , b4 ) ∈
(A, B, B, B) has an odd number of edges, then (b1 , φ(b2 ), φ(b3 ), φ(b4 ) has an even number of
edges and vice versa. (This is because the existence of exactly three edges changed between
the two quadruples, namely the edges between b1 and the rest).
One can apply the same argument to all other cases of the form (A, D, D, D), (B, C, C, C),
etc, to get that exactly one half of all quadruples have an odd number of edges.
Case 3: (A, A, B, C). The induced graph on A has exactly half of the edges of the complete
graph on the same number of vertices. Therefore, by going through all pairs (a1 , a2 ) for the
A-vertices, we get again that exactly half of all possible quadruples have an odd number of
edges.
Case 4: (A, A, B, B) similar to the previous one, by the same argument.
Case 5: (A, B, C, D) - all quadruples of this type work, so we have 44(n−1) contribution.
In total, we have 4f (n) for the case 1, density 1/2 for cases 2 through 4 and 44(n−1) for case 5.
Therefore the final recurrence is
1 h 4n
   n−1 i
4
f (n) = 4f (n − 1) + 44(n−1) + − 44(n−1) − 4 ,
2 4 4

with the initial terms f (0) = 0, f (1) = 1.