The number of particles at sublinear distances from the tip in branching Brownian motion

Gabriel Flath · 2025

Read The number of particles at sublinear distances from the tip in branching Brownian motion by Gabriel Flath (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2504-20833.txt.

Introduction

Flath counts particles near the front tip of a branching, wandering swarm in twenty twenty-five. In branching Brownian motion, particles drift at random and split as they go. The lead particle's place after long time is well known from past work. Flath asks how many particles sit within a short stretch behind that tip.

He proves the count, scaled in a set way, settles toward a known random limit. For stretches up to a set size, the settling holds in chance but cannot be made stronger. He also shows how the family ties of two sampled particles look in the limit. The result gives a shared picture of the swarm's front.

Catalog record

CREATORGabriel Flath
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE76 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2504-20833

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

The number of particles at sublinear distances from the tip in branching Brownian motion Gabriel Flath Abstract Consider a branching Brownian motion (BBM). It is well known [ 8 , 15 ] that the rightmost particle is located near m t = 2 ​ t − 3 2 ​ 2 ​ log ⁡ t m_{t}=\sqrt{2}t-\frac{3}{2\sqrt{2}}\log t . Let 𝒩 ⁡ ( t , x ) \mathcal{N}(t,x) be the set of particles within distance x x from m t m_{t} , where x = o ⁡ ( t ) x=o(t) grows with t t . We prove that # 𝒩 ( t , x ) / π − 1 / 2 x e x ​ m t / t e − x 2 / ( 2 t ) \#\mathcal{N}(t,x)/\pi^{-1/2}xe^{xm_{t}/t}e^{-x^{2}/(2t)} converges in probability to Z ∞ Z_{\infty} , the limit of the so-called derivative martingale, and that, for x = O ⁡ ( t 1 / 3 ) x=O(t^{1/3}) , the convergence cannot be strengthened to an almost sure result. Moreover, we prove that the asymptotic overlap distribution of two particles sampled uniformly from 𝒩 ⁡ ( t , x ) \mathcal{N}(t,x) converges to that of the critical derivative martingale measure. This establishes a universal genealogical picture of the BBM front at sublinear distances from the tip. 1 Introduction and results 1.1 Introduction The (standard dyadic) Branching Brownian motion (BBM) is a classical stochastic process that describes the evolution of a population of particles undergoing both independent diffusion and branching. It is a staple of probability theory and statistical physics, with applications in e.g. population dynamics, reaction-diffusion equations, and random spin-glass models. Formally, BBM starts with a single particle at the origin, which diffuses as a standard Brownian motion. This particle lives for an exponentially distributed lifetime with mean one, and upon its death this particle splits into two offspring at the same position. Each daughter particle then starts an independent copy of the same process. The behaviour of the extremal particles of the BBM is by now a classical and well developed question which has seen some striking recent progress. Because the BBM is a toy model for log-correlated random fields, those developments have been instrumental in the study of the extremes of several models such as the Gaussian free field and the Random Energy Model universality class. Let \mathcal{N}_{t} be the collection of particles at time t . For v\in\mathcal{N}_{t} and s\in\left[0,t\right] , we say that u\in\mathcal{N}_{s} is its ancestor at time s if v is a descendant of u and denote it u\preceq v . For u\in\mathcal{N}_{t} and s\in\left[0,t\right] , we write X_{u}(s)\in\mathbb{R} for the position of particle u (or its ancestor) at time s . We define the branching Brownian motion process to be X(t)=(X_{u}(t),u\in\mathcal{N}_{t}) and its natural filtration \mathcal{F}_{t}=\sigma\{X(s),s\leq t\} . In a series of seminal works, Bramson [ 8 ] , showed that the position of the rightmost particle in a BBM is centred around m_{t}\coloneqq\sqrt{2}t-\frac{3}{2\sqrt{2}}\log t , and more precisely that \max_{u\in\mathcal{N}_{t}}X_{u}(t)-m_{t}\xrightarrow[t\rightarrow\infty]{d}W, where W is a random variable whose distribution function is given by the so-called critical traveling wave solution to the Fisher-KPP equation. Lalley and Sellke [ 15 ] then proved that there exists a constant C>0 such that \mathbb{P}(\max_{u\in\mathcal{N}_{t}}X_{u}(t)-m_{t}\leq x)\to\mathbb{E}\left[e^{-CZ_{\infty}e^{-\sqrt{2}x}}\right], (1) where Z_{\infty}>0 is the almost sure limit of the derivative martingale: Z_{t}=\sum_{u\in\mathcal{N}_{t}}(\sqrt{2}t-X_{u}(t))e^{\sqrt{2}(X_{u}(t)-\sqrt{2}t)}. Observe that this suggests that \max_{u\in\mathcal{N}_{t}}X_{u}(t)-m_{t}-\log(CZ_{\infty})/\sqrt{2}\xrightarrow[t\rightarrow\infty]{d}G, (2) where G follows a Gumbel distribution, as later confirmed; see, e.g., [ 1 ] . To study the genealogy of the process, we will frequently appeal to the branching property. For any time r\geq 0 and any particle u\in\mathcal{N}_{r} , the descendants of u form a branching Brownian motion shifted in time and space. We denote by Z_{\infty}^{(u)} the almost sure limit of the derivative martingale associated with this shifted subtree. Conditional on \mathcal{F}_{r} , the random variables (Z_{\infty}^{(u)})_{u\in\mathcal{N}_{r}} are independent and have the same distribution as the global limit Z_{\infty} . The aim of our work is to extend the understanding of the front up to sublinear distance of m_{t} . In particular, we investigate the number of particles within a sublinear distance of m_{t} . Understanding this quantity provides deeper insights into the fine structure of the front and other existing results. 1.2 Main results For t\geq 0 , define the set of particles to the right of m_{t}-x , \mathcal{N}(t,x)\coloneqq\{u\in\mathcal{N}_{t}:X_{u}(t)\geq m_{t}-x\}, (3) and N(t,x)\coloneqq\#\mathcal{N}(t,x) its cardinality. Theorem 1.1 . Let x_{t} be such that, as t\to\infty , x_{t}=o(t) and x_{t}\rightarrow\infty , then, \frac{N(t,x_{t})}{\pi^{-\frac{1}{2}}x_{t}e^{\frac{m_{t}}{t}x_{t}-\frac{x^{2}}{2t}}}\xrightarrow[t\rightarrow\infty]{\mathbb{P}}Z_{\infty}. (4) For convenience, we define f(t,x)\coloneqq\pi^{-1/2}xe^{\frac{m_{t}}{t}x-\frac{x^{2}}{2t}}. Remark 1.2 . The statement of Theorem 1.1 is equivalent to, for any \delta>0 , \lim_{x\to\infty}\lim_{c\to 0}\sup_{t:x\leq ct}\mathbb{P}\left(\left|\frac{N(t,x)}{f(t,x)}-Z_{\infty}\right|>\delta\right)=0. (5) The following theorem establishes that almost sure convergence fails when x_{t} grows slowly with t . Theorem 1.3 . Let x_{t} be such that 0\leq x_{t}\leq t^{1/3} . Then, \limsup_{t\to\infty}\frac{N(t,x_{t})}{x_{t}e^{\sqrt{2}x_{t}}}=\infty\qquad\text{a.s.} (6) The proof of Theorem 1.1 relies on bounding the number of pairs of particles that do not split early. Together with Theorem 1.1 , this bound yields an explicit limit overlap distribution for two particles sampled uniformly from \mathcal{N}(t,x) . Define the splitting time between two particles (u,v)\in\mathcal{N}_{t}^{2} as Q_{t}(u,v)\coloneqq\sup\left\{s\leq t:\forall\gamma\leq s,\ X_{u}(\gamma)=X_{v}(\gamma)\right\}. Theorem 1.4 . Let x_{t} be such that x_{t}\to\infty , and x_{t}=o(t) as t\to\infty . For t\geq 0 , let \mu_{t}\coloneqq\frac{1}{N(t,x_{t})^{2}}\sum_{(u,v)\in\mathcal{N}(t,x_{t})^{2}}\delta_{Q_{t}(u,v)}\quad(\text{setting }\mu_{t}\coloneqq\delta_{0}\text{ if }N(t,x_{t})=0) (7) the empirical overlap distribution for two particles sampled uniformly from \mathcal{N}(t,x_{t}) . Then (\mu_{t})_{t\geq 0} converges weakly in probability to a limit \mu_{\infty} , that is, \mu_{t}\xrightarrow[t\to\infty]{\mathbb{P}}\mu_{\infty}\quad\text{in the weak topology}. (8) Where \mu_{\infty} is characterized by \mu_{\infty}([r,\infty))=\frac{1}{Z_{\infty}^{2}}\sum_{u\in\mathcal{N}_{r}}\left(e^{\sqrt{2}X_{u}(r)-2r}\,Z_{\infty}^{(u)}\right)^{2},\qquad\forall r\geq 0,\quad\text{almost surely.} (9) Theorem 1.4 directly implies the following. Corollary 1.5 . Let U and V be chosen uniformly at random from \mathcal{N}(t,x_{t}) , and define Q_{t}=Q_{t}(U,V) if \mathcal{N}(t,x_{t})\neq\emptyset , and Q_{t}=0 otherwise. Then (Q_{t},Z_{t})_{t\geq 0} converges in distribution to (Q_{\infty},Z_{\infty}) , where, conditional on \mathcal{F}_{\infty}=\sigma\left(\cup_{t\geq 0}\mathcal{F}_{t}\right) , Q_{\infty} has distribution \mu_{\infty} . 1.3 Discussion and related work First, observe that a first-moment calculation for N(t,x) does not yield the correct order of magnitude. Indeed, a straightforward application of the many-to-One Lemma gives that \mathbb{E}[N(t,x)]\asymp te^{\frac{m_{t}}{t}x-\frac{x^{2}}{2t}}, (10) whereas the convergence in Theorem 1.1 suggests an order that is lower by a factor of x/t . This discrepancy arises from the correlation structure of branching Brownian motion. Specifically, Arguin, Bovier, and Kistler [ 3 ] proved that extremal particles are unlikely to cross a certain barrier to the right. It is natural to expect that likewise, the particles contributing to N(t,x) are similarly localised. The inflation of \mathbb{E}[N(t,x)] can thus be attributed to the contribution of unlikely events where particles do manage to cross this barrier. Theorem 1.4 , Proposition 4.1 and Proposition 2.1 provide the following description of the front of the branching Brownian motion. With high probability, for x=o(t) , most of the particles to the right of m_{t}-x are in fact located within m_{t}-x+O(1) at time t and remain below the curve s\frac{m_{t}}{t}-(s\wedge(t-s))^{\alpha} for any \alpha<1/2 , throughout the time interval [O(1),t-O(1)] . More precisely, the trajectory of a particle in \mathcal{N}(t,x) , shifted by m_{t}s/t , resembles a Brownian bridge B_{s} from distance O(1) at time r=O(1) and reaching x at time t conditioned to stay positive over the interval [r,t] . • For x\ll\sqrt{t} , B_{s} behaves like a Brownian excursion

Read the complete text