Hyperbolic branching Brownian motion: the empirical limit measure

David Geldbach · 2025

Read Hyperbolic branching Brownian motion: the empirical limit measure by David Geldbach (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-2509-06730.txt.

Introduction

Branching particles spreading through curved hyperbolic space are analyzed by David Geldbach in a solo 2025 probability paper. Brownian motion in hyperbolic space wanders off to the boundary and never returns, so the cloud of descendants accumulates there. The normalized empirical measure, a record of where mass ends up, settles into a random limit measure on the boundary.

Geldbach computes the Hausdorff dimension of that limit measure in terms of the branching rate, and shows it has a density with respect to boundary volume once branching is fast enough. This behavior differs sharply from the set of accumulation points studied earlier, whose threshold is much lower. The results answer questions posed by Woess and by Candellero and Hutchcroft. The author believes the methods also extend to branching random walks on hyperbolic groups, broadening the reach of the argument.

Catalog record

CREATORDavid Geldbach
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE54 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2509-06730

Browse the full library · About the Tower of Babel

Full text

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

Hyperbolic branching Brownian motion: the empirical limit measure David Geldbach Abstract We study branching Brownian motion in hyperbolic space. As hyperbolic Brownian motion is transient, the normalised empirical measure of branching Brownian motion converges to a random measure μ ∞ \mu_{\infty} on the boundary. We show that the Hausdorff dimension of supp ​ μ ∞ \mathrm{supp}\ \mu_{\infty} is ( 2 ​ β ) ∧ 1 (2\beta)\wedge 1 where β \beta is the branching rate, and that μ ∞ \mu_{\infty} admits a Lebesgue density for β > 1 / 2 \beta>1/2 . This is very different to the behaviour of the set of accumulation points on the boundary where β c = 1 / 8 \beta_{c}=1/8 which has been shown by Lalley and Selke [ 10 ] . This answers several questions posed by Woess [ 12 ] and similar questions posed by Candellero and Hutchcroft [ 3 ] . We believe that our methods also apply to branching random walks on non-elementary hyperbolic groups. 2010 Mathematics Subject Classification: 60J80, 60D05, 60J65. Keywords: Branching Brownian motion, Hausdorff dimension, hyperbolic space. Figure 1: Three simulations of hyperbolic BBM and the limit of its empirical distribution \mu_{\infty} on the boundary. The branching rates are \beta\in\{0.12,0.4,1\} from left to right. The Hausdorff dimensions of the support of \mu_{\infty} are \{0.24,0.8,1\} whereas the Hausdorff dimensions of the set of accumulation points on the boundary are \{0.4,1,1\} . Observe that in the middle picture where \beta=0.4 there are a lot of paths accumulating on the boundary that do not contribute significantly to \mu_{\infty} . 1 Introduction Given a transient stochastic process, one can often define a natural extension of the state space so that the process converges almost surely to a point on the boundary of this space. We can think of the distribution of this limit as an exit distribution of the process. Understanding these exit distributions is a well studied topic in itself, but it becomes even richer when combined with branching processes. The idea here is that we have multiple, possibly infinitely many, correlated particles that all escape towards the boundary. This induces different random subsets of the boundary, notably the set \Lambda of accumulation points on the boundary, and the set \mathcal{S}\mathrel{\mathop{\mathchar 58\relax}}=\mathrm{supp}\ \mu_{\infty} which is the support of the limit of the empirical measure \mu_{\infty} . Loosely speaking, \Lambda is determined by all particles including rare exceptional particles, while \mathcal{S} is determined only by the bulk of the particles. We always have that \mathcal{S}\subseteq\Lambda , it is natural to ask if \mathcal{S} can be a proper subset of \Lambda and, if yes, how we can quantify this difference. This is the object of this paper in the case of branching Brownian motion in hyperbolic space. Branching Brownian motion (BBM) on \mathbb{R} is an interacting particle system. Particles move as independent Brownian motions and split into two at a given rate \beta . Here the exceptionally fast particles, that is the maximal displacement at time t , is a major object of interest [ 1 , 2 ] . On the other hand, the number of particles near the origin always grows exponentially for any \beta>0 . If the underlying space is hyperbolic, the behaviour of BBM is markedly different: there is \beta_{c}=1/8 such that for any \beta<\beta_{c} the process eventually vacates any compact set almost surely. On the other hand, for any \beta>\beta_{c} , the number of particles near the origin again grows exponentially. The same is true for a discrete version of this model, a branching random walk on a homogeneous tree. We give a precise definition of branching Brownian motion in hyperbolic space in the next section but refer to Woess [ 12 ] and the references therein for background on hyperbolic BBM. The limit set \Lambda of hyperbolic BBM has first been studied by Lalley and Selke [ 10 ] who show that \Lambda is a fractal like random set and compute its Hausdorff dimension. (See for example [ 5 ] for some background on fractals and Hausdorff dimension.) Others have studied similar sets of accumulation points of branching random walks on the boundary on discrete hyperbolic spaces, see for example [ 4 , 7 , 11 ] . Much less is known about \mu_{\infty} and its support. Even the existence of \mu_{\infty} has only been shown recently [ 3 , 8 ] . In fact, we think that this paper is the first to show quantitative properties of \mu_{\infty} . 1.1 The model Hyperbolic space is usually modelled with the Poincaré disk model \mathbb{D} or the upper half plane model \mathbb{H} . We use them interchangeably. They are Riemannian manifolds with metrics given by \frac{2\sqrt{dx^{2}+dy^{2}}}{1-x^{2}-y^{2}}\text{ for }(x,y)\in\mathbb{D}=\{(x,y)\in\mathbb{R}^{2}\mathrel{\mathop{\mathchar 58\relax}}x^{2}+y^{2}<1\}, for the disk model and for the upper half plane model by \frac{\sqrt{dx^{2}+dy^{2}}}{y^{2}}\text{ for }(x,y)\in\mathbb{H}=\{(x,y)\in\mathbb{R}\times\mathbb{R}_{+}\}. The two models are isometric, an isometry f\mathrel{\mathop{\mathchar 58\relax}}\mathbb{D}\to\mathbb{H} is given by f(z)=i\frac{1+z}{1-z} where we identify \mathbb{D} and \mathbb{H} with subsets of \mathbb{C} by z=x+iy . Note that origin 0\in\mathbb{D} corresponds to i\in\mathbb{H} . Both \mathbb{D} and \mathbb{H} are endowed with natural boundaries \partial\mathbb{D} and \partial\mathbb{H} given by \partial\mathbb{D}=\{z\in\mathbb{C}\mathrel{\mathop{\mathchar 58\relax}}|z|=1\} and \partial\mathbb{H}=\{z\in\mathbb{C}\mathrel{\mathop{\mathchar 58\relax}}\Im(z)=0\} . The hyperbolic Laplacian is given by \mathcal{L}_{\mathbb{D}}=\frac{(1-|z|^{2})^{2}}{4}\left(\partial_{x}^{2}+\partial_{y}^{2}\right),\qquad\text{respectively}\qquad\mathcal{L}_{\mathbb{H}}=y^{2}\left(\partial_{x}^{2}+\partial_{y}^{2}\right). From this we define hyperbolic Brownian motion to be the stochastic process with generator \frac{1}{2}\mathcal{L}_{\mathbb{D}} (respectively \frac{1}{2}\mathcal{L}_{\mathbb{H}} ). In the upper half plane model we could also do this by solving the pair of stochastic differential equations dX_{t}=Y_{t}dW_{t},\quad dY_{t}=Y_{t}dB_{t}, where (W_{t})_{t} and (B_{t})_{t} are independent Brownian motions. This process is then canonically started from (X_{0},Y_{0})=(0,1) . Observe that (Y_{t})_{t} is a geometric Brownian motion, hence we can solve the SDE explicitly in the second coordinate, Y_{t}=\exp\left(-\frac{t}{2}+B_{t}\right). This also tells us that X_{t} is Gaussian with mean 0 and random variance \int_{0}^{t}\exp\left(-s+2B_{s}\right)ds . From this we can see that hyperbolic Brownian motion converges to a random point (X_{\infty},0) on the boundary \partial\mathbb{H} where X_{\infty} is Gaussian with (random) variance \int_{0}^{\infty}\exp\left(-s+2B_{s}\right)ds . Having defined hyperbolic Brownian motion, we define hyperbolic branching Brownian motion (BBM) to be the following particle process on \mathbb{D} : At time 0 , we start with one particle at the origin. Particles move as independent hyperbolic Brownian motions. At rate \beta , each particle independently branches into two, both offspring particles branch and move independently. This results in a cloud of particles, we denote it by ((X_{u}(t),Y_{u}(t)),u\in\mathcal{N}(t)) where \mathcal{N}(t) is the set of particles alive at time t . By abuse of notation, we also denote the (isometric) process on \mathbb{H} by ((X_{u}(t),Y_{u}(t)),u\in\mathcal{N}(t)) . Here the process is started from a single particle at (0,1) . We can relate certain expectations for hyperbolic BBM to expectations of hyperbolic Brownian motion by the many–to–one formula, \mathbb{E}_{(x,y)}\left[\sum_{u\in\mathcal{N}(t)}f((X_{u}(s),Y_{u}(s))_{0\leq s\leq t})\right]=e^{\beta t}\mathbb{E}_{(x,y)}\left[f((X_{s},Y_{s})_{0\leq s\leq t})\right], (1.1) for any (x,y)\in\mathbb{D} and measurable non–negative f . This follows from linearity due to the independence of movement and branching. 1.2 Results We are interested in the long term behaviour of the cloud ((X_{u}(t),Y_{u}(t)),u\in\mathcal{N}(t)) , especially related to the boundary. We define the normalised empirical measure at time t to be \mu_{t}=\frac{1}{|\mathcal{N}(t)|}\sum_{u\in\mathcal{N}(t)}\delta_{(X_{u}(t),Y_{u}(t))}. One can show that there is a measure \mu_{\infty} , supported on the boundary, such that \mu_{t} converges weakly to \mu_{\infty} with probability one, see [ 12 ] . This is a simple argument: let h\mathrel{\mathop{\mathchar 58\relax}}\mathbb{D}\to\mathbb{R} be

Read the complete text