1: \begin{proof}

2: The proof of this result is quite easy, but we prefer to first give an intuitive explanation for why the result holds true. Note that the time it takes to go from $n$ blocks to $n-1$ blocks is just an exponential random variable with rate $n(n-1)/2$. When $n$ is large, this is approximately $n^2/2$, so we can expect the number of blocks to approximately solve the differential equation:

3: \begin{equation}\index{Differential equation}\label{ode}

4: \begin{cases}

5: u'(t) &= - \ds \frac{u(t)^2}2\\

6: u(0)&=+\infty.

7: \end{cases}

8: \end{equation}

9: (\ref{ode}) has a well-defined solution $u(t) = 2/t$, which is finite for all $t>0$ but infinite for $t=0$. This explains why $N_t$ is finite almost surely for all $t>0$. in fact, one guesses from the ODE approximation:

10: \begin{equation}\index{Small-time behaviour}\label{K:smalltimes0}

11: N_t \sim \frac2t, \ \ t \to 0

12: \end{equation}

13: almost surely. This statement is correct indeed, but unfortunately it is tedious to make the ODE approximation rigorous. Instead, to show Theorem \ref{T:Kcdi}, we use the following simple argument. It is enough to show that, for every $\eps>0$, there exists $M>0$ such that $\P(N_t>M) \le \eps.$ For this, it suffices to look at the restrictions $\Pi^n$ of $\Pi$ to $[n]$, and show that

14: \begin{equation}\label{cdi0}

15: \limsup_{n\to \infty}\P( N^n_t >M) \le \eps.

16: \end{equation}

17: Here we used the notation $N^n_t $ for the number of blocks of $\Pi^n_t$. For every $n \ge 1$, let $E_n$ be an exponential random variable with rate $n(n-1)/2$. Then note that, by Markov's inequality:

18: \begin{align*}

19: \P( N_t^n >M) &= \P\left( \sum_{k=M}^n E_k >t\right) \\

20: & \le  \frac1t \E\left( \sum_{k=M}^n E_k \right) \\

21: & \le \frac1t \sum_{k=M}^\infty \frac2{k(k-1)}.

22: \end{align*}

23: The right-hand side of the above inequality is independent of $n$, and can be made as small as desired provided $M$ is chosen large enough. Thus (\ref{cdi0}) follows.

24: \end{proof}

25: