On the zero-noise limit for SDE’s singular at the initial time

Jules Pitcho · 2025

Read On the zero-noise limit for SDE’s singular at the initial time by Jules Pitcho (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-2503-22905.txt.

Introduction

When a noisy system calms down, which of its possible paths does the randomness choose? Jules Pitcho studies that zero-noise limit in a solo 2025 paper for equations whose drift is singular right at the starting moment.

The drift here swirls without divergence and belongs to a rough class tied to a classic condition from fluid theory. Pitcho proves the vanishing noise selects one definite probability spread over the drift's own flow lines. A famous tangled example fits the class. For almost every starting point in that example, the limit spreads across several distinct flow lines at once. Uniqueness of the limit, not of the path, is the point.

Catalog record

CREATORJules Pitcho
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE47 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2503-22905

Browse the full library · About the Tower of Babel

Full text

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

On the zero-noise limit for SDE’s singular at the initial time Jules Pitcho Abstract . We investigate the zero-noise limit for SDE’s driven by Brownian motion with a divergence-free drift singular at the initial time and prove that a unique probability measure concentrated on the integral curves of the drift is selected. More precisely, we prove uniqueness of the zero-noise limit for divergence-free drifts in L l ​ o ​ c 1 ​ ( ( 0 , T ] , B ​ V ​ ( 𝕋 d , ℝ d ) ) ∩ L q ​ ( ( 0 , T ) , L p ​ ( 𝕋 d , ℝ d ) ) L^{1}_{loc}((0,T];BV(\mathbb{T}^{d};\mathbb{R}^{d}))\cap L^{q}((0,T);L^{p}(\mathbb{T}^{d};\mathbb{R}^{d})) where p p and q q satisfy a Prodi-Serrin condition. The vector field constructed by Depauw [ 12 ] lies in this class and we show that for almost every intial datum, the zero-noise limit selects a probability measure concentrated on several distinct integral curves of this vector field. 1. Introduction This work studies selection by a zero-noise limit of probability measures concentrated on the integral curves of a rough vector field. Let us begin by defining integral curves. Definition 1.1 . Consider a Borel vector field b:[0,T]\times\mathbb{T}^{d}\to\mathbb{R}^{d} . We shall say that a continuous curve \gamma_{x}:[0,T]\to\mathbb{T}^{d} is an integral curve of b starting from x\in\mathbb{T}^{d} , if for every t\in[0,T] we have (1.1) \gamma_{x}(t)-x=\int_{0}^{t}b(s,\gamma_{x}(s))ds\qquad\text{and}\qquad\int_{0}^{T}|b(s,\gamma_{x}(s))|ds<+\infty. By the Cauchy-Lipschitz theory, existence and uniqueness of integral curves of b holds when the time integral of the spatial Lipschitz constant of b is finite. However, uniqueness of integral curves can fail almost everywhere when the vector field is rough. Such vector fields are constructed in [ 19 , 12 , 2 , 17 ] . An interesting question is then to select a probability measure concentrated on the set of integral curves starting from x . Let us review some known selection results. The work of Ambrosio [ 3 ] , building upon the work of DiPerna and Lions [ 13 ] , proves that for divergence-free vector fields b\in L^{1}((0,T);BV(\mathbb{T}^{d};\mathbb{R}^{d})) , there exists an essentially unique 1 1 1 essential uniqueness is understood in the following sense: for any other Borel family \{\tilde{\gamma}_{x}\} satisfying ( 1.2 ), it holds \gamma_{x}=\tilde{\gamma}_{x} for Lebesgue almost every x\in\mathbb{T}^{d} . Borel family of integral curves of b \{\gamma_{x}\} 2 2 2 by convention, when the indexing set of a family is not specified, we shall take it to be \mathbb{T}^{d} . such that for every t\in[0,T] , we have (1.2) \int_{\mathbb{T}^{d}}\phi(x)dx=\int_{\mathbb{T}^{d}}\phi(\gamma_{x}(t))dx\qquad\forall\phi\in C(\mathbb{T}^{d}). In the one-dimensional setting, Bafico and Baldi [ 4 ] show that for a class of autonomous vector fields, smooth except at a single point, the zero-noise limit selects a probability measure concentrated on two integral curves of b with suitable weights. Delarue and Flandoli [ 11 ] also give another proof of the same result. In a series two works Bressan, Mazzola and Nguyen [ 6 , 7 ] study the one-dimensional problem for autonomous vector fields satisfying a suitable condition, and characterise the probability measures on integral curves compatible with a semigroup structure deterministic or Markovian. Furthermore, these authors show that the transition kernels of these semigroups are limits of the transition kernels of diffusion processes with smooth coefficients. We here investigate the zero-noise limit for a subclass of divergence-free vector fields b\in L^{1}_{loc}((0,T];BV(\mathbb{T}^{d};\mathbb{R}^{d})) including the vector field constructed by Depauw [ 12 ] , a two-dimensional vector field for which integral curves are almost everywhere non-unique. Importantly, and in contrast to [ 4 , 11 ] , upon taking the zero-noise limit, we will retain randomness in the initial data, which will be essential to the uniqueness of the zero-noise limit. Previous work motivates this investigation. Building upon my work [ 20 ] , in collaboration with Mescolini and Sorella [ 18 ] , we show that for vector fields in b\in L^{1}_{loc}((0,T];BV(\mathbb{T}^{d};\mathbb{R}^{d}))\cap L^{2}((0,T)\times\mathbb{T}^{d};\mathbb{R}^{d}) , the vanishing diffusivity scheme and regularisation by convolution of the vector field select a unique solution of the continuity equation. In [ 21 ] , I also show that for bounded, divergence-free vector fields in b\in L^{1}_{loc}((0,T];BV(\mathbb{T}^{d};\mathbb{R}^{d})) , there exists a unique probabilistic flow of b under which the Lebesgue measure is invariant. For the vector field b_{DP} constructed by Depauw [ 12 ] , which is in this class, for almost every initial condition, this flow is concentrated on several distinct integral curves of b_{DP} . Moreover, this flow is the weak limit of the classical flows of approximations by convolution of b . To introduce our result precisely, let us first fix notations and conventions. 1.1. Notations and conventions \mathbb{T}^{d}\cong\mathbb{R}^{d}/\mathbb{Z}^{d} is the d -dimensional flat torus. \Gamma_{T} is the space of continuous paths C([0,T];\mathbb{T}^{d}) . e_{t}:\Gamma_{T}\ni\gamma\longmapsto\gamma(t)\in\mathbb{T}^{d} is the evaluation map at time t\in[0,T] . \mathcal{P}(\Gamma_{T}) is the space of Radon probability measures on \Gamma_{T} . {\mathscr{L}}^{d} is the d -dimensional Lebesgue measure. \delta_{\gamma} is the Dirac mass on \gamma\in\Gamma_{T} . {\mathbb{E}}(Y) is the expactation of the random variable Y . P_{Y} is the law of the random variable Y . The Borel \sigma -algebra of a topological space X is denoted \mathcal{B}(X) . A measure on a metric space is always assumed to be a Radon measure. 1.2. Review of SDE theory Consider the stochastic differential equation (SDE) X^{\nu}_{t}(x)=x+\int_{0}^{t}b(s,X^{\nu}_{s}(x))ds+\nu W_{t},\qquad t\in[0,T] on \mathbb{T}^{d} , where b:[0,T]\times\mathbb{T}^{d}\to\mathbb{R}^{d} is a Borel vector field, \nu>0 , and (W_{t})_{t\in[0,T]} is a \mathbb{T}^{d} -valued Brownian motion started from zero on the canonical filtered probability space (\Omega,\mathcal{F},(\mathcal{F}_{t})_{t\in[0,T]},P) . Even when integral curves of b starting from x are non-unique, strong existence and strong uniqueness of solutions of ( SDE ) holds. This phenomenon is called regularisation by noise in the litterature. We say that continuous adapted process (X^{\nu}_{t}(x))_{t\in[0,T]} on the filtered probability space (\Omega,\mathcal{F},(\mathcal{F}_{t})_{t\in[0,T]},P) is a strong solution, if P -almost surely, we have ( SDE ). We say that strong uniqueness holds for ( SDE ), if for any two strong solutions are indistinguishable. Veretennikov [ 24 ] proves strong existence and strong uniqueness for ( SDE ) when b is bounded. Krylov and Röckner [ 16 ] prove strong existence and strong uniqueness for ( SDE ) when b belongs to L^{q}_{p}(T) defined below in ( 1.4 ) where p and q satisfy ( 1.5 ). Fedrizzi and Flandoli [ 14 ] then give another proof of this result and further show that the random continuous sample paths \{(X^{\nu}_{t}(x))_{t\in[0,T]}\;:\;x\in\mathbb{T}^{d}\} have a continuous modification in x . In order to include randomness in the initial data, let us form the filtered probability space (1.3) \mathscr{P}:=(\Omega\times\mathbb{T}^{d},\mathcal{F}\otimes\mathcal{B}(\mathbb{T}^{d}),(\mathcal{F}_{t}\otimes\mathcal{B}(\mathbb{T}^{d}))_{t\in[0,T]},P\otimes{\mathscr{L}}^{d}). We also define the function space (1.4) L^{q}_{p}(T):=L^{q}((0,T);L^{p}(\mathbb{T}^{d};\mathbb{R}^{d})), which forms a Banach space under the norm \|b\|_{L^{q}_{p}}:=\Big(\int_{0}^{T}\big(\int_{\mathbb{T}^{d}}|b(t,x)|^{p}dx\big)^{q/p}dt\Big)^{1/q}. We assume that the exponents p and q satisfy the following Prodi-Serrin condition (1.5) \frac{d}{p}+\frac{2}{q}<1,\qquad p,q\in(1,+\infty]. We then view X^{\nu}=(X_{t}^{\nu})_{t\in[0,T]} as a random continuous process on \mathscr{P} , with randomness coming both from the Brownian motion and from the initial data. For future use, we record the following proposition extracted from [ 14 ] . Proposition 1.2 . Consider a Borel vector field b:[0,T]\times\mathbb{T}^{d}\to\mathbb{R}^{d} , and p,q\in(1,+\infty] satisfying ( 1.5 ). Assume that b\in L^{q}_{p}(T) . Then there exists a continuous stochastic process X^{\nu} on \mathscr{P} defined in ( 1.3 )

Read the complete text