Elastic Brownian motion with random jumps from the boundary

Fausto Colantoni, Mirko D’Ovidio · 2025

Read Elastic Brownian motion with random jumps from the boundary by Fausto Colantoni, Mirko D’Ovidio (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-2511-01455.txt.

Introduction

Fausto Colantoni and Mirko D'Ovidio study a variant of Brownian motion, the random wandering used to model particles and prices alike. In their version, called elastic Brownian motion, the moving point is not stopped when it reaches the edge of its region. Instead, it restarts from a random position inside the region and keeps going.

The authors pin down the process in three ways: through its stochastic differential equation, its generator, and a direct description of its paths. They also work out the long-run probability distribution and a spectral representation. The final section turns to harmonic functions on the upper half-space to study the trace process left at the boundary.

Catalog record

CREATORFausto Colantoni, Mirko D’Ovidio
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE56 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2511-01455

Browse the full library · About the Tower of Babel

Full text

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

Elastic Brownian motion with random jumps from the boundary Fausto Colantoni, Mirko D’Ovidio Abstract . In this paper, we study elastic Brownian motion on a C 2 C^{2} domain. Instead of being killed at the boundary, the process restarts from a random position inside the domain. We characterize this process through its stochastic differential equation (SDE), its generator, and a description of the paths. We also derive the invariant probability measure and the spectral representation. At the end, we focus on the harmonic functions on the upper half-space to study the trace process. 2020 Mathematics Subject Classification Primary 60J65, 60J75; Secondary 60J60, 35P05, 35J25 1. Introduction In the field of probability theory, diffusions with boundary jumps are gaining increasing importance. These processes were first introduced by Feller [ 24 ] , who later studied the corresponding semigroups with these boundary conditions [ 25 ] . The model of Brownian motion killed at the boundary (with Dirichlet conditions) which, instead of being absorbed, is restarted from a fixed position inside the domain D , was successively investigated in both one and higher dimensions in [ 27 ] and [ 28 ] . The nonlocal Dirichlet condition, given by the formula (1) \displaystyle u(z)=\int_{D}u(x)\mu(z,dx),\quad\text{for }z\in\partial D, for a suitable domain D , was later generalized to include probability measures \mu that depend on the hitting point of the boundary. Diffusions with this boundary condition have been extensively studied, with a focus on their ergodic behavior and spectrum [ 9 , 10 ] , as well as a complete treatment of the semigroup in both bounded and unbounded domains [ 2 , 35 ] . A diffusion satisfying the boundary condition ( 1 ) can be represented by paths obtained through the concatenation of diffusions that, instead of being killed at the boundary, are restarted inside the domain at points distributed according to \mu . This type of concatenation of stochastic processes with jumps at the boundary is also well known in probability; see, for instance, [ 26 , Section 25] . In the case of Brownian motion, the associated process is not a Hunt process, since it cannot be realized with càdlàg paths. Similar boundary conditions have also been studied for stable processes, and in this case the associated stochastic process turns out to be a Hunt process [ 13 , 14 ] . The first probabilistic solution to the problem introduced by Feller in [ 24 ] was given by Itô and McKean [ 32 ] , where the measure appearing in the integral boundary condition is a Lévy measure. More recently, their approach has been revisited in the study of non-local boundary value problems for Brownian motion [ 15 , 19 ] . In these works, the integral condition is rewritten as a Marchaud-type operator, and techniques from fractional calculus are used to analyze the problem. Such models also have important physical applications. For instance, in [ 20 ] it is shown that, on the positive half-line, a Brownian motion with negative drift and a boundary condition combining reflection and jumps is the time-reversed process of a Brownian motion with Poissonian resetting to the origin [ 23 ] . This result highlights the versatility of such models, which naturally connect to search problems and stochastic thermodynamics. In this paper, we focus on elastic Brownian motion, where at the killing times the process has jumps inside the domain and restarts its paths instead of being killed. The corresponding boundary condition has been considered in [ 3 ] , where sufficient conditions for the semigroup to be Feller are established. In our setting, we study the simpler case where the boundary integral measure does not depend on the hitting point of the boundary, which nevertheless allows for a complete analysis from the SDE viewpoint. 2. Stochastic Differential Equation We consider the SDE in a bounded C^{2} domain D\subset\mathbb{R}^{n} : (2) \mathrm{d}X_{t}=\mathrm{d}B_{t}+n(X_{t})\,\mathrm{d}L_{t}+\int_{D}\bigl(x-X_{t-}\bigr)\,\nu\bigl(\mathrm{d}L_{t},\mathrm{d}x\bigr), where B is a standard Brownian motion on \mathbb{R}^{n} , L is the boundary local time of X on \partial D , n(\cdot) is the inward unit normal on \partial D , and \nu is a Poisson random measure, independent of B , on \mathbb{R}_{+}\times D with intensity \mathrm{d}\ell\otimes\mu(\mathrm{d}x) , where \mu is a probability measure. The last term induces jumps into D at boundary hits, distributed according to the measure \mu . Here, the filtration generated by (B,\nu) is the smallest filtration (\mathcal{F}_{t})_{t\geq 0} satisfying the usual conditions such that B_{s} and \nu\big((0,s]\times A\big) are \mathcal{F}_{t} -measurable for all s\leq t and all measurable A\subset\mathcal{B}(D) , the Borel \sigma- algebra of D . Let \{(\ell_{k},Z_{k})\}_{k\geq 1} be the countable family, respectively, of the random jump times and of the i.i.d. post-jump positions of \nu . Since we are dealing with a probability measure \mu , we have that \ell_{k}-\ell_{k-1}\sim\text{Exp}(1) and Z_{k}\sim\mu . We now introduce the jumping times for the process X \tau_{k}:=\inf\{t\geq 0:L_{t}\geq\ell_{k}\}\quad k\geq 1. Then, we rewrite the Lebesgue-Stieltjes integral as the absolutely convergent sum, see [ 31 , Formula (3.7) page 62] \int_{0}^{t}\!\int_{D}(x-X_{s-})\,\nu(\mathrm{d}L_{s},\mathrm{d}x)=\sum_{k:\,\ell_{k}\leq L_{t}}\bigl(Z_{k}-X_{\tau_{k}-}\bigr). Let \nu(d\ell,dy) be the Poisson random measure on [0,\infty)\times D with intensity d\ell\otimes\mu(dy) , and denote N_{t}:=\nu\big([0,t]\times D\big)=\sum_{k\geq 1}\mathbf{1}_{\{\ell_{k}\leq t\}}. Then the observable number of jumps up to t in our model is given by N_{L_{t}}:=\nu\big([0,L_{t}]\times D\big)=\sum_{k\geq 1}\mathbf{1}_{\{\ell_{k}\leq L_{t}\}}=\sum_{k\geq 1}\mathbf{1}_{\{\tau_{k}\leq t\}}, i.e. it is the process that governs the jumps from the boundary. The SDE in ( 2 ) can be rewritten as (3) X_{t}=X_{0}+B_{t}+\int_{0}^{t}n(X_{s})\,L_{s}+\sum_{k:\,\ell_{k}\leq L_{t}}\bigl(Z_{k}-X_{\tau_{k}-}\bigr);\quad X_{0}=x\in D, X_{t}\in\overline{D} for all t and \int_{0}^{\infty}\mathbf{1}_{\{X_{t}\in D\}}dL_{t}=0 a.s. 3. Existence and Uniqueness Theorem 1 . Let the assumptions above hold. Then the SDE ( 2 ) admits a pathwise unique strong solution (X_{t},L_{t})_{t\geq 0} , adapted to the filtration generated by (B,\nu) , with X càdlàg and L continuous and nondecreasing. Proof. The idea is to construct the process X piecewise, between successive jumps. Step 1 (before the first jump). Ignore the jumps and consider the reflected Brownian motion starting at X_{0} : \widehat{X}^{(0)}_{t}=X_{0}+B_{t}+\int_{0}^{t}n\big(\widehat{X}^{(0)}_{s}\big)\,d\widehat{L}^{(0)}_{s},\quad t<\tau_{1}. For C^{2} domains, strong existence and pathwise uniqueness for the reflected Brownian motion is classical (see for example [ 36 ] or [ 5 , Theorem 12.1] ). This defines (\widehat{X}_{t}^{(0)},\widehat{L}_{t}^{(0)}) up to t=\tau_{1} , where \tau_{1}=\inf\{t\geq 0:\widehat{L}^{(0)}_{t}\geq\ell_{1}\}. We then set X_{t}=\widehat{X}^{(0)}_{t} and L_{t}=\widehat{L}^{(0)}_{t} for t<\tau_{1} , and realize the jump: X_{\tau_{1}}:=Z_{1},\qquad L_{\tau_{1}}:=\widehat{L}^{(0)}_{\tau_{1}}, in order to extend L by continuity. Step 2 (between the first and second jump). Restart at t=\tau_{1} from X_{\tau_{1}}=Z_{1} . Using independence and the independent increments of B , note that s\mapsto B_{s+\tau_{1}}-B_{\tau_{1}} is a Brownian motion, define (\widehat{X}_{t}^{(1)},\widehat{L}_{t}^{(1)}) , for t\in[\tau_{1},\tau_{2}) , as a new reflected motion with initial condition Z_{1} : \widehat{X}^{(1)}_{t-\tau_{1}}=Z_{1}+\big(B_{t}-B_{\tau_{1}}\big)+\int_{\tau_{1}}^{t}n\big(\widehat{X}^{(1)}_{s-\tau_{1}}\big)\,d\widehat{L}^{(1)}_{s-\tau_{1}}, until \widehat{L}^{(0)}_{\tau_{1}}+\widehat{L}^{(1)}_{t-\tau_{1}}=L_{t} reaches \ell_{2} , i.e. \tau_{2}=\inf\{t\geq 0:\widehat{L}^{(0)}_{\tau_{1}}+\widehat{L}^{(1)}_{t-\tau_{1}}\geq\ell_{2}\} . Set X_{t}=\widehat{X}^{(1)}_{t-\tau_{1}} for t\in[\tau_{1},\tau_{2}) , then jump to Z_{2} at t=\tau_{2} , and continue recursively for all the jumping times. We observe that \tau_{k}=\inf\{t\geq 0:L_{t}\geq\ell_{k}\} are stopping times with respect to the filtration (B,\nu) since L is adapted and continuous and \ell_{k} are measurable. Then, B_{t+\tau_{k}}-B_{\tau_{k}} for t>0 is an independent Brownian motion as we need. Finite number of jumps in finite time. For T<\infty , the local time L is continuous a.s. on [0,T] , then it satisfies L_{T}<\infty a.s.. Conditioned on the path of L , since N_{L_{T}}\sim\mathrm{Poisson}(L_{T}) , only finitely

Read the complete text