Consider a continuous time doubling branching random walk on the nonnegative integers. It starts with a particle at $0$. The initial particle dies after producing two offspring at $1$, each after an independent exponential$(1)$-distributed amount of time. In the same fashion, a particle born at $k$ will die upon producing two offspring at $k+1$ after, each after an independent exponential$(1)$-distributed amount of time. Suppose the last of the $2^n$ particles born at $n$ occurs at time $T$. This is known as the "Last-Birth Problem" and is fairly well studied. We are curious how many particles reach $n$ just before $T$.

To make this concrete let $X_n$ be the number of particles that reach $n$ at times in $[T-1,T)$. Is there a function $f(n)\to \infty$ such that $P(X_n > f(n)) \to 1$?

Perhaps this is false and only $O(1)$ particles arrive in the final second. However, there may be a huge wave of particles just in front of the last one.


Consider the binary tree of ancestry (thus, the vertices at depth $n$ are the particles that reach location $n$). Write on each edge $e$ between level $n-1$ to level $n$ the time it takes to generate the particle at level $n$, and call this variable $X_e$. If I understand correctly your model, these times are iid exponential (1). For any vertex $v$, let $S_v$ be the partial sum of $X_e$ for $e$ on the geodesic connecting the root to $v$. If I understand correctly, $T=\max_{v: |v|=n} S_v$. The question you are asking is then a question on the extremal process of this branching random walk (how many particles in an interval of size $1$ to the left of $T$). By a result of Madaule, https://arxiv.org/pdf/1107.2543.pdf, the extremal process converges to a decorated Poisson point process (with intensity $e^{-cx}$, some explicit $c$) viewed from its tip, and therefore $X_n$ is a tight sequence. So $f(n)=O(1)$.

  • $\begingroup$ Thanks Ofer. Maduale's and the preceding work on the min in branching processes are really interesting. $\endgroup$ – Matthew Junge Jan 26 '18 at 18:36

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.