SLT olympiad 2020

May 2020

Problem 1
Solve the differential equation

x2 y 00 + 5xy 0 + 3y = 0, y 0 (1) = 3, y(x) = O x−2 , x → +∞.

x > 0,

Answer: − x13 .
Solution: Let x = et , then y(x) → y(et ) and
dy
y0 = dt
dx
= ẏe−t ⇒ xy 0 = ẏ,
dt

dy 0
00 ÿe−t − ẏe−t
y = dt
dx
= = (ÿ − ẏ)e−2t ⇒ x2 y 00 = ÿ − ẏ.
dt
et
Hence our equation reduces to 2-nd order ODE with constant coefficients:

ÿ + 4ẏ + 3y = 0.

Characteristic equation:
λ2 + 4λ + 3 = 0 ⇒ λ = −1, λ = −3.
Then a general solution has the form
C1 C2
y(t) = C1 e−3t + C2 e−t ⇒ y(x) = 3
+ .
x x
C1
Since y(x) = O x−2 , C2 = 0, and y = y 0 (1) = 3 ⇒ y(x) = − x13 .


x3 ,

Problem 2
Let ξ, η be independent Poisson random variables with intensity 2. Calculate

Var [E [ξ|ξ + η]]

λk −λ
Distribution of Poisson random variable ξ with intensity λ is given by P(ξ = k) = k! e ,k = 0, 1, 2, . . .
Answer: 1.
Solution: Since ξ and η are i.i.d., it holds that
d
(ξ, ξ + η) = (η, ξ + η).

1
Hence, for any A ∈ σ(ξ + η), it holds
EξI(A) = EηI(A),
and this yields E[ξ|ξ + η] = E[η|ξ + η] almost surely. On the other hand,

E[ξ|ξ + η] + E[η|ξ + η] = E[ξ + η|ξ + η] = ξ + η.

ξ+η
Hence, 2 is a version of E [ξ|ξ + η], which implies that
 
ξ+η
Var E [ξ|ξ + η] = Var = 1.
2

Problem 3
Let u and v be unit vectors in Rd such that uT v = 0.5. Find det(I + uuT + vv T ).
Answer: 15/4.
Solution. Denote uuT + vv T by A and consider a polynomial P (λ) = det(I + λA). Since the matrix
A is symmetric, there exists an eigenvalue decomposition A = S T ΣS, where S is an orthogonal matrix,
i.e. S T S = Id , and Σ = diag(σ1 , σ2 , 0, . . . , 0). Note that only two eigenvalues of A are non-zero, because
rank(A) = 2. Then

P (λ) = det(S T (I + λΣ)S) = det(S T ) · det(I + λΣ) · det(S) = det(I + λΣ) = (1 + λσ1 )(1 + λσ2 ).

Here we used the fact that S is an orthogonal matrix and this yields det(S) = 1. Next, note that

λ2
Tr(Σ)2 − Tr(Σ2 ) .


(1 + λσ1 )(1 + λσ2 ) = 1 + λTr(Σ) +
2
Since the trace is invariant with respect to orthogonal transforms, we have Tr(A) = Tr(Σ). Thus, we obtain

λ2
Tr(A)2 − Tr(A2 ) .


P (λ) = 1 + λTr(A) +
2
It remains to compute Tr(A) and Tr(A2 ). Due to the equality Tr(AB) = Tr(BA), it holds

Tr(A) = Tr(uuT + vv T ) = uT u + v T v = 2,

Tr(A2 ) = Tr((uT u)uuT + (uT v)(uv T + vuT ) + (v T v)vv T ) = (uT u)2 + (v T v)2 + 2(uT v)2 = 2 + 2(uT v)2 .
Finally,
4 − 2 − 2(uT v)2 15
det(I + uuT + vv T ) = P (1) = 1 + 2 + = 4 − (uT v)2 = .
2 4

Problem 4
Find the limit lim P(ξn ≤ n) where ξn is a Poisson random variable with intensity n, n ∈ N.
n→∞
1
Answer: 2.
n
P nk −n d
Solution: Note that k! e = P(ξn ≤ n), where ξn ∼ P ois(n). Note that ξn = ξn,1 + . . . + ξn,n , with
k=0
ξn,1 , . . . , ξn,n - i.i.d. P ois(1) random variables. Hence,
n n
! n n
! !
X X 1 X X
P(ξn ≤ n) = P ξn,k − E ξn,k ≤ 0 = P √ ξn,k − E ξn,k ≤0 .
n
k=1 k=1 k=1 k=1

2
Due to the central limit theorem,
n n
!
1 X X d
√ ξn,k − E ξn,k → N (0, 1),
n
k=1 k=1
  n n
 
√1 ≤ 0 → 12 .
P P
which implies that P n
ξn,k − E ξn,k
k=1 k=1

Problem 5
Let {a1 , . . . , aN } be an array of naturalQ numbers and K be an integer number. Find number of subsets
m
{ai1 , . . . , aim } ⊆ {a1 , . . . , aN }, such that j=1 aij < K in O(N K) elementary operations.
Solution: Let P (k, j), 1 ≤ k ≤ K, 1 ≤ j ≤ N , be the number of subsets of {a1 , . . . , aj }, such that their
product is less then k. Note that we either use aj in our product or not, leading to the following recurrence

P (k, j) = P (k, j − 1) + (P (bk/aj c, j − 1) + 1) I(aj < K).

Hence, pseudo-polynomial tabular solution is possible:

def f(a,N,K):
P = np.zeros(N+1,K+1)
for i in range(1,K+1):
for j in range(1,N+1):
P[i,j] = P[i,j-1]
if (a[j] < K):
P[i,j] += (P[i/a[j],j-1] + 1)
return P[N,K].

Problem 6
Let X1 , . . . , X2n+1 be independent variables with uniform distribution on [0, 1]. Let X(1) ≤ . . . ≤ X(2n+1)
be corresponding order statistics. Find a density of the random variable X(n+1) − X(n) .
Answer: (2n + 1)(1 − x)2n , x ∈ [0, 1].
Solution. Let us find a density p(u, v), u, v ∈ [0, 1], of joint distribution of X(n+1) and X(n) . It is clear that
p(u, v) = 0 in the case u < v. Consider the case u ≥ v:


P X(n+1) ∈ [u, u + ∆u), X(n) ∈ [v, v + ∆v)
p(u, v) = lim .
∆u,∆v→0 ∆u∆v
For the numerator of the last expression, we have


P X(n+1) ∈ [u, u + ∆u), X(n) ∈ [v, v + ∆v)


= P ∃ i1 , . . . , i2n+1 : Xi1 , . . . , Xin−1 ∈ [0, v), Xn ∈ [v, v + ∆v), Xin+1 ∈ [u, u + ∆u), Xin+2 , . . . , Xi2n+1 ∈ [u + ∆u, 1]
X

= P Xi1 , . . . , Xin−1 ∈ [0, v), Xn ∈ [v, v + ∆v), Xin+1 ∈ [u, u + ∆u), Xin+2 , . . . , Xi2n+1 ∈ [u + ∆u, 1]
i1 ,...,i2n+1

(2n + 1)!
v n−1 (1 − u − ∆u)n ∆u∆v.
(n − 1)!1!1!n!

Then
(2n + 1)! n−1
p(u, v) = v (1 − u)n , 0 ≤ v ≤ u ≤ 1.
(n − 1)!n!

3
Thus, we have (
(2n+1)! n−1
(n−1)!n! v (1 − u)n , 0 ≤ v ≤ u ≤ 1;
p(u, v) =
0, otherwise.
Find a density q(t), t ∈ [0, 1], of the random variable ξ = X(n+1) − X(n) . By the convolution formula, we
have
Z1−t Z1−t
(2n + 1)!
q(t) = p(v + t, v)dv = v n−1 (1 − v − t)n dv
(n − 1)!n!
0 0
Z1−t n−1  n
(2n + 1)! v v dv
= (1 − t)2n 1−
(n − 1)!n! 1−t 1−t 1−t
0
(2n + 1)! Γ(n)Γ(n + 1)
= (1 − t)2n · = (2n + 1)(1 − t)2n .
(n − 1)!n! Γ(2n + 1)

Thus, ξ = X(n+1) − X(n) has beta distribution Beta(1, 2n + 1).