Krylov-Veretennikov desomposition for measure-valued processes induced by SDEs with interaction on Riemannian manifolds
2025
Read Krylov-Veretennikov desomposition for measure-valued processes induced by SDEs with interaction on Riemannian manifolds (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-2508-18995.txt.
Introduction
Andrey Dorogovtsev and Alexander Weiss wrote a 2025 mathematics paper on stochastic differential equations that play out on curved spaces called manifolds. In their setup, the drift and pull on a moving particle can depend both on its place and on how the whole group of particles is spread out. They work on compact, connected spaces of a set dimension.
The pair prove that strong solutions exist and are unique under broad rules on the inputs. They then study the process that tracks the full spread of particles over time and give an explicit breakdown of its random motion. Further results cover smooth response when the starting spread changes. The paper builds theory for linked random motion on curved spaces.
Catalog record
| YEAR | 2025 |
|---|---|
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 62 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2508-18995
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Krylov-Veretennikov desomposition for measure-valued processes induced by SDEs with interaction on Riemannian manifolds Abstract We introduce a framework for stochastic differential equations (SDEs) with interaction on compact, connected, d d -dimensional manifolds. For SDEs whose drift and diffusion coefficients may depend on both the state variable and the empirical distribution, we establish existence and uniqueness of strong solutions under general regularity assumptions. We study the associated measure-valued process on the Wasserstein space over the manifold, deriving an explicit Itô–Wiener decomposition. We prove Malliavin differentiability of the solution and, using directional derivatives in the Wasserstein space, establish smooth dependence of the solution on the measure component for a class of coefficients. Andrey Dorogovtsev 1 1 1 NAS Ukraine, [email protected] , Alexander Weiß 2 2 2 University of Leipzig, [email protected] 1 Introduction In this paper, we introduce the notion of stochastic differential equations with interactions on compact, connected, d -dimensional Riemannian manifolds M , as introduced by Dorogovtsev [ DK97 ] . These equations are of the following type in the case M=\mathbb{R}^{d} \displaystyle\begin{cases}\mathrm{d}x(u,t)&=a(x(u,t),\mu_{t})\mathrm{d}t+\int_{\mathbb{R}^{d}}b(x(u,t),\mu_{t},p)W(\mathrm{d}p,\mathrm{d}t)\\ x(u,0)&=u\in\mathbb{R}^{d}\\ \mu_{t}&=\mu\circ x^{-1}(\cdot,t).\end{cases} Here, \mu is a probability measure on \mathbb{R}^{d} , and W is a Wiener sheet on \llll^{2}(\mathbb{R}^{d}) .We can interpret points in \mathbb{R}^{d} as particles distributed in space according to \mu , and the flow x as the temporal evolution of these particles. This evolution depends not only on individual particle dynamics but also on the evolution of their spatial distribution, reflecting particle interaction. The fact that equations with interaction are stochastic flows is advantageous for generalizing them to manifolds and simplifies their treatment. To our knowledge, the first and only consideration of SDEs with interactions on manifolds was conducted by Ding, Fang, and Li [ DFL23 ] , where the authors considered special interaction kernels in the drift but no interaction term in the noise. We are interested in ergodicity results for the measure-valued process (\mu_{t})_{t\geq 0} . While the Krylov–Bogoliubov theorem ensures the existence of an invariant measure on the space \mathcal{P}_{2}(M) , ergodicity results cannot be established using classical techniques. This is due to the fact that the Markov process (\mu_{t})_{t\geq 0} takes values in an infinite-dimensional and non-linear space, while the driving noise is finite-dimensional. Hence, "smoothing" properties of the induced semigroup P_{t}f(\mu)=\mathbb{E}(f(\mu_{t})) cannot be expected, due to the highly degenerate noise. Therefore, we study the Krylov–Veretennikov decomposition of functionals (F(\mu_{t}))_{t\geq 0} , where F:\mathcal{P}_{2}(M)\to\mathbb{R} . This provides a formula for the kernels of the Itô–Wiener decomposition, offering insight into the structure of functionals of (\mu_{t})_{t\geq 0} .The explicit formula enables manual calculations of long-term behavior by analising individual terms of the decomposition separately. To our knowledge this is the first contribution towards a Krylov-Veretennikov decomposition for the measure valued process induced by SDEs with interaction. The Krylov-Veretennikov decomposition has been established in [ VK76 ] by Krylov and Veretennikov. In their work they considered solutions to ordinary one-dimensional SDEs and described the Itô-Wiener kernels only through the semigroup induced by the SDE and a differential operator, they showed with the help of the decomposition existence of strong solutions, the same ansatz has been applied to SDEs with more general coefficients [ KZ00 , Kry25 ] . After the work of Krylov and Veretennikov, there have been more general considerations, in [ Dor12 ] Dorogovtsev showed a Krylov-Veretennikov type decomposition for general stochastic semigroups on Hilbert spaces moreover he showed the Krylov-Veretennikov decomposition for the Arratia flow. In [ Ria15 ] the Krylov-Veretennikov decomposition for a stopped Brownian motion has been proven and in [ Gli15 ] the Krylov-Veretennikov for functionals of the Harris flow. To our knowledge this is the first paper treating the Krylov-Veretennikov decomposition result for the measure valued process induced by SDEs with interactions. The structure of the paper is as follows. In Chapter 2, we present a definition and a uniqueness and existence result for SDEs with interaction on general Riemannian manifolds. In chapter 3 we present differential operators on the Wasserstein space and discuss smoothness of the solution to SDEs with interaction with respect to these differential operators. In chapter 4 we present and prove our main result which in simplified form looks as follows Theorem 1.1 . Let f be a "smooth" and bounded functional then \displaystyle f(\mu_{t})=T_{t}f(\mu)+\sum^{n}_{i=1}\sum^{\infty}_{k=1}\underset{\Delta^{k}([0,t])}{\int\dots\int}T_{\tau_{1}}A_{i}T_{\tau_{2}-\tau_{1}}\dots T_{\tau_{k}-\tau_{k-1}}A_{i}T_{t-\tau_{k}}f(\mu)\mathrm{d}B^{i}(\tau_{1})\dots\mathrm{d}B^{i}(\tau_{k}) where A_{i} is a differential operator on the Wasserstein space more precisely a directional derivative into the direction of the noise vector fields and T_{t}f(\mu)=\mathbb{E}(f(\mu_{t})) . The form of our Krylov-Veretennikov expansion is very close to the original in [ VK76 ] the authors obtained for T_{t}f(u)=\mathbb{E}(f(y(u,t))) and y(u,t) solves a one dimensional SDE \displaystyle\begin{cases}dy(u,t)&=\alpha(y(u,t))\mathrm{d}t+\beta(y(u,t))\mathrm{d}B_{t}\\ x(u,0)&=u\in\mathbb{R}\end{cases} where \alpha,\beta:\mathbb{R}\to\mathbb{R} . Their result was \displaystyle f(x(u,t))=E(f(x(u,t)))+\sum^{\infty}_{k=0}\underset{\Delta([0,t]^{k})}{\int\dots\int}T_{t-\tau_{k}}\beta(u)\frac{\partial}{\partial u}T_{\tau_{k}-\tau_{k-1}}\dots\beta(u)\frac{\partial}{\partial u}T_{\tau_{1}}f(u)\mathrm{d}B(\tau_{1})\dots\mathrm{d}B_{\tau_{k}}. One can observe that the decomposition here only depends on the semigroup and directional derivative operator b\frac{\partial}{\partial u} where b is the noise coefficient of the SDE describing x(u,t) similar to our main result. 2 Existence and Uniqueness of SDEs with interaction on compact Riemannian manifolds Let M be a compact smooth connected d -dimensional Riemannian manifold without boundary or M=\mathbb{R}^{d} . (V_{0},\dots,V_{n}) be mappings such that V_{i}:\mathfrak{M}\times\mathcal{P}(M)\to TM for all i=0,\dots,n , such that V_{i}(\cdot,\mu) is a smooth vector field for all \mu\in\mathcal{P}(M) and V_{i}(x,\cdot):\mathcal{P}(M)\to T_{x}M and its derivative are Lipschitz map with respect to the Wasserstein- 2 distance on \mathcal{P}(M) for all x\in M , moreover the partial derivatives of V_{i} suffice the same properties. We consider equations of the following form \displaystyle\begin{cases}dx_{\mu}(u,t)&=F(x_{\mu}(u,t),\circ\mathrm{d}t)\\ x_{\mu}(u,0)&=u\in M\\ F(u,t)&=\int_{0}^{t}V_{0}(u,\mu_{t})\mathrm{d}t+\sum^{n}_{i=1}\int_{0}^{t}V_{i}(u,\mu_{t})\mathrm{d}B^{i}_{t}\\ \mu_{t}&=\mu\circ x_{\mu}(\cdot,t)^{-1}\end{cases} (1) A M -valued process x solves the system ( 1 ) if for all f\in\ccc^{\infty}(M) \displaystyle f(x_{\mu}(u,t)) \displaystyle=f(u)+\int_{0}^{t}F(x(u,t),\circ\mathrm{d}t)f \displaystyle=f(u)+\int\tilde{V}_{0}(\cdot,\mu_{t})f(x_{\mu}(u,s))\mathrm{d}s+\sum^{n}_{i=1}\int_{0}^{t}V_{i}(\cdot,\mu_{s})f(x_{\mu}(u,s))\mathrm{d}B_{s} where \displaystyle\tilde{V}_{0}(\cdot,\mu)f=V_{0}(\cdot,\mu)f+\frac{1}{2}\sum_{i=1}^{n}V^{2}_{i}(\cdot,\mu)f in local coordinates (x^{1},\dots,x^{d}) . Lemma 2.1 . Let M be a connected d -dimensional smooth Riemannian manifold without boundary and let \iota:M\to\mathbb{R}^{N} be the Nash embedding then we have for all x,y\in M and C>0 \displaystyle\left|\iota(x)-\iota(y)\right|\leq d_{M}(x,y)\leq C\left|\iota(x)-\iota(y)\right| (2) Proof. It suffices, by isometry to show \displaystyle\left|\iota(x)-\iota(y)\right|\leq d_{\iota(M)}(\iota(x),\iota(y))\leq C\left|\iota(x)-\iota(y)\right| since the first inequality is trivial we shall only show the second. Let x\in M Choose a submanifold chart (U_{x},\varphi) where U_{x} is an open \mathbb{R}^{N} neighbourhood around \iota(x) such that \varphi(U) is a ball and \varphi_{i}^{d+1}(\iota(y))=\dots=\varphi_{i}^{N}(\iota(y))=0 for y\in U_{x}\cap\iota(M) , moreover by shrinking U_{x} if necessary we can assume w.l.o.g that \varphi\in\ccc^{\infty}_{b}(U_{x}) . Now let \iota(y),\iota(z)\in U_{x}\cap M .Let \gamma(t)=\varphi^{-1}_{i}(\varphi(\iota(z))+t(\varphi(\iota(y))-\varphi(\iota(z)))) which can be chosen since \varphi(U) is a ball. Now \gamma(0)=\iota(x) and \gamma(1)=\iota(y) we get \gamma^{\prime}\equiv\varphi(\iota(x))-\varphi(\iota(y)) in local coordinates. We denote by U^{\varepsilon}_{x} a \mathbb{R}^{N} -open neighbourhood of \iota(x) such that U^{\varepsilon}_{x}\cap M=B_{\iota(M)}(\iota(x),\varepsilon) for all \varepsilon>0 . Choose \delta>0 such that \overline{B_{\mathbb{R}^{N}}(\iota(x),\delta)}\subset U_{x} . Now let \varepsilon>0 be small enough such that B_{\iota(M)}(\iota(x),\varepsilon)\subset B_{\mathbb{R}^{N}}(\iota(x),\delta)\cap\iota(M) we can choose now U^{\frac{\varepsilon}{2}}_{x}\subset U^{\varepsilon}_{x} such that U^{\varepsilon}_{x}\subset B_{\mathbb{R}^{N}}(\iota(x),\delta) then we have for all y,z\in B_{M}(x,\varepsilon) \displaystyle d_{\iota(M)}(\iota(y),\iota(z))\leq C\left|\varphi(\iota(z))-\varphi(\iota(y))\right|. where C>0 depends solely depends on the Riemannian metric and the chart \varphi . Moreover \displaystyle d_{\iota(M)}(\iota(z),\iota(y))\leq\left|\varphi(\iota(y))-\varphi(\iota(z))\right|\leq\sup_{x\in\overline{B_{\mathbb{R}^{N}}(\iota(x),\delta)}}\left|\nabla\varphi(x)\right|\left|\iota(y)-\iota(z)\right|. Now cover M with such B_{M}(x_{i},\frac{\varepsilon_{i}}{2})_{i=1,\dots,m} as constructed above, then B_{M}(x_{i},\varepsilon_{i}) is also a cover of M on which we have the same coordinate chart (\varphi_{i})_{i=1,\dots,n} as constructed. Then we can choose by construction a universal constant C>0 such that, whenever y,z\in B_{M}(x_{i},\varepsilon_{i}) for some i=1,\dots,n holds, we have \displaystyle d_{\iota(M)}(\iota(y),\iota(z))\leq C\left|\iota(y)-\iota(z)\right|_{\mathbb{R}^{N}}. (3) Set \varepsilon=\min_{i=1,\dots,n}\varepsilon_{i} , hence for all y,z\in M such that d_{M}(y,z)<\frac{\varepsilon}{2} , we get ( 3 ). To observe this, note that