Unofficial archive — problems, solutions & results © IMC, reproduced with permission.

IMC / 2025 / Problems / Day 2, P9

IMC 2025 · Day 2 · P9

very hard

Let nn be a positive integer. Consider the following random process which produces a sequence of nn distinct positive integers X1,X2,,XnX_1, X_2, \dots, X_n.

First, X1X_1 is chosen randomly with P(X1=i)=2i\mathsf{P}(X_1 = i) = 2^{-i} for every positive integer ii. For 1jn11 \le j \le n-1, having chosen X1,,XjX_1, \dots, X_j, arrange the remaining positive integers in increasing order as n1<n2<n_1 < n_2 < \cdots, and choose Xj+1X_{j+1} randomly with P(Xj+1=ni)=2i\mathsf{P}(X_{j+1} = n_i) = 2^{-i} for every positive integer ii.

Let Yn=max{X1,,Xn}Y_n = \max\{X_1, \dots, X_n\}. Show that E[Yn]=i=1n2i2i1\mathsf{E}[Y_n] = \sum_{i=1}^{n} \frac{2^i}{2^i - 1} where E[Yn]\mathsf{E}[Y_n] is the expected value of YnY_n.

(proposed by Jan Kuś and Jun Yan, University of Warwick)

Solution 1 of 3 (official)

For each j[n]={1,2,,n}j \in [n] = \{1, 2, \dots, n\}, let Yj=max{X1,,Xj}Y_j = \max\{X_1, \dots, X_j\}. We use induction on jj to show that E[Yj]=i=1j2i2i1for all j[n].\mathsf{E}[Y_j] = \sum_{i=1}^{j} \frac{2^i}{2^i - 1} \quad \text{for all } j \in [n]. The base case j=1j = 1 follows easily from definition.

For the inductive step, it suffices to show that E[Yj+1Yj]=2j+12j+11\mathsf{E}[Y_{j+1} - Y_j] = \frac{2^{j+1}}{2^{j+1} - 1} for every j[n1]j \in [n-1]. Note that Yj+1YjY_{j+1} \ne Y_j if and only if Xj+1>Yj=max{X1,,Xj}X_{j+1} > Y_j = \max\{X_1, \dots, X_j\}, in which case Yj+1=Xj+1Y_{j+1} = X_{j+1}. Thus, E[Yj+1Yj]=P[Xj+1>Yj]E[Xj+1YjXj+1>Yj].\mathsf{E}[Y_{j+1} - Y_j] = \mathsf{P}[X_{j+1} > Y_j] \cdot \mathsf{E}[X_{j+1} - Y_j \mid X_{j+1} > Y_j]. To compute P[Xj+1>Yj]\mathsf{P}[X_{j+1} > Y_j], note that for any fixed pairwise distinct positive integers a1,,aja_1, \dots, a_j and a>max{a1,,aj}a > \max\{a_1, \dots, a_j\}, P[(X1,,Xj+1)=(a,a1,,aj)]=P[(X1,,Xj+1)=(a1,a,,aj)]/2=P[(X1,,Xj+1)=(a1,a2,a,,aj)]/4==P[(X1,,Xj+1)=(a1,,aj,a)]/2j.\begin{align*} \mathsf{P}[(X_1, \dots, X_{j+1}) = (a, a_1, \dots, a_j)] &= \mathsf{P}[(X_1, \dots, X_{j+1}) = (a_1, a, \dots, a_j)] / 2 \\ &= \mathsf{P}[(X_1, \dots, X_{j+1}) = (a_1, a_2, a, \dots, a_j)] / 4 \\ &= \dots = \mathsf{P}[(X_1, \dots, X_{j+1}) = (a_1, \dots, a_j, a)] / 2^j. \end{align*} Therefore, summing over all possible a1,,aja_1, \dots, a_j and a>max{a1,,aj}a > \max\{a_1, \dots, a_j\}, we see that P[Xj+1>Yj]=2ji=0j2i=2j2j+11.\mathsf{P}[X_{j+1} > Y_j] = \frac{2^j}{\sum_{i=0}^{j} 2^i} = \frac{2^j}{2^{j+1} - 1}. To finish, it is easy to see that E[Xj+1YjXj+1>Yj]=t=1tP[Xj+1=Yj+tXj+1>Yj]=t=1t2t=2.\mathsf{E}[X_{j+1} - Y_j \mid X_{j+1} > Y_j] = \sum_{t=1}^{\infty} t \cdot \mathsf{P}[X_{j+1} = Y_j + t \mid X_{j+1} > Y_j] = \sum_{t=1}^{\infty} \frac{t}{2^t} = 2.

Solution 2 of 3 (official)

Since YnY_n takes values in Z>0\mathbb{Z}_{>0}, E[Yn]=k=1P[Ynk].\mathsf{E}[Y_n] = \sum_{k=1}^{\infty} \mathsf{P}[Y_n \ge k]. For each k[n]k \in [n], P[Ynk]=1\mathsf{P}[Y_n \ge k] = 1, while for each k>nk > n, P[Ynk]=1P[Yn<k]=1P[X1,,Xn<k]=1i=1n(112ki).\mathsf{P}[Y_n \ge k] = 1 - \mathsf{P}[Y_n < k] = 1 - \mathsf{P}[X_1, \dots, X_n < k] = 1 - \prod_{i=1}^{n} \left( 1 - \frac{1}{2^{k-i}} \right). Note that this formula also works for every k[n]k \in [n], as the i=ki = k term in the product is 0. Thus, it suffices to show that i=1n2i2i1=k=1(1i=1n(112ki)).\sum_{i=1}^{n} \frac{2^i}{2^i - 1} = \sum_{k=1}^{\infty} \left( 1 - \prod_{i=1}^{n} \left( 1 - \frac{1}{2^{k-i}} \right) \right). The case of n=1n = 1 is easy to verify. Using induction on nn, it suffices to show that for every n2n \ge 2, 1112n=k=1(i=1n1(112ki)i=1n(112ki))=k=112kni=1n1(112ki).\frac{1}{1 - \frac{1}{2^n}} = \sum_{k=1}^{\infty} \left( \prod_{i=1}^{n-1} \left( 1 - \frac{1}{2^{k-i}} \right) - \prod_{i=1}^{n} \left( 1 - \frac{1}{2^{k-i}} \right) \right) = \sum_{k=1}^{\infty} \frac{1}{2^{k-n}} \prod_{i=1}^{n-1} \left( 1 - \frac{1}{2^{k-i}} \right). Indeed, for every N>nN > n, after multiplying by 112n1 - \frac{1}{2^n}, the sum on the right telescopes as (112n)k=1N12kni=1n1(112ki)=k=1N(12kn12k)i=1n1(112ki)==k=1N(i=1n(112k+1i)i=1n(112ki))=i=1n(112N+1i).\begin{align*} \left( 1 - \frac{1}{2^n} \right) \sum_{k=1}^{N} \frac{1}{2^{k-n}} \prod_{i=1}^{n-1} \left( 1 - \frac{1}{2^{k-i}} \right) &= \sum_{k=1}^{N} \left( \frac{1}{2^{k-n}} - \frac{1}{2^k} \right) \prod_{i=1}^{n-1} \left( 1 - \frac{1}{2^{k-i}} \right) = \\ = \sum_{k=1}^{N} \left( \prod_{i=1}^{n} \left( 1 - \frac{1}{2^{k+1-i}} \right) - \prod_{i=1}^{n} \left( 1 - \frac{1}{2^{k-i}} \right) \right) &= \prod_{i=1}^{n} \left( 1 - \frac{1}{2^{N+1-i}} \right). \end{align*} Taking NN to infinity finishes the proof.

Solution 3 of 3 (official)

(sketch) It can be shown by induction or another method that for any sequence of positive integers a1<a2<<ana_1 < a_2 < \dots < a_n, P[{X1,,Xn}={a1,,an}]=2i=1naii=1n(2i1).\mathsf{P}[\{X_1, \dots, X_n\} = \{a_1, \dots, a_n\}] = 2^{-\sum_{i=1}^{n} a_i} \prod_{i=1}^{n} (2^i - 1). For any a1<a2<<ana_1 < a_2 < \dots < a_n, let d1=a1d_1 = a_1 and di+1=ai+1aid_{i+1} = a_{i+1} - a_i for i[n1]i \in [n-1], so P[{X1,,Xn}={d1,d1+d2,,d1++dn}]=2i=1n(n+1i)dii=1n(2i1).\mathsf{P}[\{X_1, \dots, X_n\} = \{d_1, d_1 + d_2, \dots, d_1 + \dots + d_n\}] = 2^{-\sum_{i=1}^{n} (n+1-i) d_i} \prod_{i=1}^{n} (2^i - 1). Note that (a1,,an)(d1,,dn)(a_1, \dots, a_n) \mapsto (d_1, \dots, d_n) is a bijection between strictly increasing sequences in Z>0\mathbb{Z}_{>0} of length nn and (Z>0)n(\mathbb{Z}_{>0})^n, so, using i1xi=x/(1x)\sum_{i \ge 1} x^i = x / (1 - x) and i1ixi=x/(1x)2\sum_{i \ge 1} i x^i = x / (1 - x)^2, we get E[Yn]=i=1n(2i1)d1,,dn1(i=1ndi)2j=1n(n+1j)dj=i=1n(2i1)i=1n(di1di2(n+1i)di)ji(dj12(n+1j)dj)==i=1n2i2i1.\begin{align*} \mathsf{E}[Y_n] &= \prod_{i=1}^{n} (2^i - 1) \sum_{d_1, \dots, d_n \ge 1} \left( \sum_{i=1}^{n} d_i \right) 2^{-\sum_{j=1}^{n} (n+1-j) d_j} \\ &= \prod_{i=1}^{n} (2^i - 1) \sum_{i=1}^{n} \left( \sum_{d_i \ge 1} d_i 2^{-(n+1-i) d_i} \right) \prod_{j \ne i} \left( \sum_{d_j \ge 1} 2^{-(n+1-j) d_j} \right) = \dots = \sum_{i=1}^{n} \frac{2^i}{2^i - 1}. \end{align*}

How the field did

contestants scored
425
average (of 10)
1.23
solved (≥ 80%)
5.9%
near-0 (≤ 10%)
73.9%
discrimination
0.54

Score distribution (field cohort)

Computed on contestants with a meaningful total (field cohort); discrimination is the corrected item–total correlation.

Similar problems

IMC 2022 · Day 2 · P8very hardavg 1.3/10 · solved 8% · near-0 80% · disc 0.46
IMC 2019 · Day 2 · P10killeravg 0.6/10 · solved 2% · near-0 89% · disc 0.36
IMC 2024 · Day 1 · P5killeravg 0.1/10 · solved 1% · near-0 97% · disc 0.25
IMC 2017 · Day 1 · P4very hardavg 1.4/10 · solved 11% · near-0 81% · disc 0.52