Stability of backward propagation of chaos

Antonis Papapantoleon, Alexandros Saplaouras, Stefanos Theodorakopoulos · 2025

Read Stability of backward propagation of chaos by Antonis Papapantoleon, Alexandros Saplaouras, Stefanos Theodorakopoulos (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-2506-03562.txt.

Introduction

Antonis Papapantoleon, Alexandros Saplaouras, and Stefanos Theodorakopoulos set out in 2025 to define what stability means for a key limit in financial mathematics. The limit is called backward propagation of chaos. It describes how a large system of backward stochastic equations settles into many free-standing copies of one limit equation.

In their sense, a stable limit still holds when the noise that drives the system is swapped for a close stand-in. One such stand-in is a discrete-time martingale that nears a continuous one. The team sets out the terms under which the limit stays intact. This matters because real models are always run with stand-ins for the ideal noise, not the ideal noise itself.

Catalog record

CREATORAntonis Papapantoleon, Alexandros Saplaouras, Stefanos Theodorakopoulos
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE134 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2506-03562

Browse the full library · About the Tower of Babel

Full text

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

Stability of backward propagation of chaos Thanks: AS gratefully acknowledges the financial support from the Hellenic Foundation for Research and Innovation Grant No. 235 (2nd Call for H.F.R.I. Research Projects to support Post-Doctoral Researchers). ST gratefully acknowledges the financial support from the Hellenic Foundation for Research and Innovation Grant No. 05724 (3rd Call for H.F.R.I. Scholarships for PhD candidates). Antonis Papapantoleon, Alexandros Saplaouras, Stefanos Theodorakopoulos Abstract . The purpose of the present paper is to introduce and establish a notion of stability for the backward propagation of chaos, with respect to (initial) data sets. Consider, for example, a sequence of discrete-time martingales converging to a continuous-time limit, and a system of mean-field BSDEs that satisfies the backward propagation of chaos, i.e. converges to a sequence of i.i.d. McKean–Vlasov BSDEs. Then, we say that the backward propagation of chaos is stable if the system of mean-field BSDEs driven by the discrete-time martingales converges to the sequence of McKean–Vlasov BSDEs driven by the continuous-time limit. We consider the convergence scheme of the backward propagation of chaos as the image of the corresponding data set under which this scheme is established. Then, using an appropriate notion of convergence for data sets, we are able to show a variety of continuity properties for this functional point of view. 2020 Mathematics Subject Classification 60F17, 60H20, 60G42, 60G44, 60G51 1. Introduction The backward propagation of chaos states that, under appropriate conditions, the solution of a mean-field system of backward stochastic differential equations (BSDEs) with N players (or particles) converges to the solutions of N independent and identically distributed McKean–Vlasov BSDEs, as the number of players N goes to infinity. Every such phenomenon is, of course, associated with a set of data \mathscr{D} that provides the basis for its mathematical description. Consider a well-posed equation and assume that the sequence of data \{\mathscr{D}^{k}\}_{k\in\mathbb{N}} converges to the data \mathscr{D}^{\infty} ; then, if the corresponding solutions also converge, we say that the equation is stable. Naturally, the framework to answer this kind of questions should determine a lot of the technical details; for example, in what sense do the data converge, how do we measure the distance of the solutions, and so on. In the theory of BSDEs, frameworks for the stability of BSDEs are provided by Hu and Peng 1997 in the special case of a constant filtration, by Briand et al. 2001 , Briand et al. 2002 for Brownian drivers, and more recently by Papapantoleon et al. 2023 where a very general framework is established. The aim of the present paper is to introduce and establish a stability result for the backward propagation of chaos. In order to motivate the reader with a simple example, let us consider the framework for Donsker’s theorem. More specifically, we associate with each player an independent sequence of symmetric random walks converging to a Brownian motion. When these symmetric random walks are sufficiently close to their corresponding Brownian motion, and when the number of players is sufficiently large, the stability of backward propagation of chaos ensures that the solution of the mean-field system of BSDEs driven by the symmetric random walks approximates the solution of the McKean–Vlasov BSDEs driven by the Brownian motions. In this work, we will combine the frameworks of Papapantoleon et al. 2024 and Papapantoleon et al. 2023 in order to study the stability of the backward propagation of chaos, in a general setting that includes BSDEs driven by either discrete-time or continuous-time martingales with independent increments. Let us describe what we mean by that. Set \overline{\mathbb{N}}:=\mathbb{N}\cup\{\infty\} and assume a sequence of standard data \{\mathscr{D}^{k}\}_{k\in\overline{\mathbb{N}}} , where \mathscr{D}^{k}:=\Big(\{\overline{X}^{k,i}\}_{i\in\mathbb{N}},T^{k},\left\{\{\xi^{k,i,N}\}_{i\in\mathscr{N}}\right\}_{N\in\mathbb{N}},\{\xi^{k,i}\}_{i\in\mathbb{N}},\Theta^{k},\Gamma,f^{k}\Big) as defined in [ 20 , Definition 4.2] , i.e. , the set of Assumptions (B1) - (B7) (presented in Section 3 ) are set in force. Let (k,N)\in\overline{\mathbb{N}}\times\mathbb{N} and i=1,\dots,N, then the following system of mean-field BSDEs under the standard data \mathscr{D}^{k} , for t\in[0,T^{k}] , \displaystyle\begin{multlined}Y^{k,i,N}_{t}=\xi^{k,i,N}+\int^{T^{k}}_{t}f^{k}\left(s,Y^{k,i,N}_{s},Z^{k,i,N}_{s}c^{k}_{s},\Gamma^{(\mathbb{F}^{k,(1,\dots,N)},\overline{X}^{k,i},\Theta^{k})}(U^{k,i,N})_{s},L^{N}(\textbf{Y}^{k,N}_{s})\right)\,\mathrm{d}C^{k}_{s}\\ -\int^{T^{k}}_{t}Z^{k,i,N}_{s}\,\mathrm{d}X^{k,i,\circ}_{s}-\int^{T^{k}}_{t}\int_{\mathbb{R}^{n}}U^{k,i,N}_{s}(x)\,\widetilde{\mu}^{(\mathbb{F}^{k,(1,\dots,N)},X^{k,i,\natural})}(\mathrm{d}s,\mathrm{d}x)-\int^{T^{k}}_{t}\,\mathrm{d}M^{k,i,N}_{s},\end{multlined} is well-defined and has the unique solution \textbf{S}^{k,N}:=\left(\textbf{Y}^{k,N},\textbf{Z}^{k,N},\textbf{U}^{k,N},\textbf{M}^{k,N}\right) , where in particular \textbf{Y}^{k,N}:=\left(Y^{k,1,N},\dots,Y^{k,N,N}\right) ; we use analogous notation for \textbf{Z}^{k,N} , \textbf{U}^{k,N} and \textbf{M}^{k,N} . Here, N denotes the total number of players in the mean-field system and i denotes one of these players, while L^{N} denotes the empirical measure associated to \textbf{Y}^{k,N}_{t} , i.e. L^{N}(\textbf{Y}^{k,N}_{t}):=\frac{1}{N}\sum_{i=1}^{N}Y^{k,i,N}_{t}. The notation will be clarified in the next section, however let us quickly state the following: \xi^{k,i,N} is the terminal condition of the BSDE for the i -th player in the N -th game for the k -th data set, \overline{X}^{k,i}=(X^{k,i,\circ},X^{k,i,\natural}) is a tuple of square-integrable martingales (where one could think of them as the continuous and purely discontinuous part of a martingale) with quadratic covariation C^{\overline{X}^{k,i}} , M^{k,i,N} is the orthogonal martingale with respect to \overline{X}^{k,i} , Z^{k,i,N} is the integrand of the Itō integral with respect to X^{k,i,\circ} , U^{k,i,N} is the integrand of the stochastic integral with respect to the random measure associated with X^{k,i,\natural} , and \Gamma is a suitable function of U^{k,i,N} chosen from a large class. In particular, the \Gamma function can be chosen as the Radon–Nikodym derivative of the stochastic integral of U^{k,i,N} with respect to the jump measure relative to the measure induced by C^{\overline{X}^{k,i}} , for a range of free parameters \Theta ; see Section 2.7 . In addition, for (k,i)\in\overline{\mathbb{N}}\times\mathbb{N} , the following McKean–Vlasov BSDE under the standard data \mathscr{D}^{k} , for t\in[0,T^{k}] , \displaystyle\begin{multlined}Y^{k,i}_{t}=\xi^{k,i}+\int^{T^{k}}_{t}f^{k}\left(s,Y^{k,i}_{s},Z^{k,i}_{s}c^{k}_{s},\Gamma^{(\mathbb{F}^{k,i},\overline{X}^{k,i},\Theta^{k})}(U^{k,i})_{s},\mathcal{L}(Y^{k,i}_{s})\right)\,\mathrm{d}C^{k}_{s}\\ -\int^{T^{k}}_{t}Z^{k,i}_{s}\,\mathrm{d}X^{k,i,\circ}-\int^{T^{k}}_{t}\int_{\mathbb{R}^{n}}U^{k,i}_{s}\,\widetilde{\mu}^{(\mathbb{F}^{k,i},X^{k,i,\natural})}(\mathrm{d}s,\mathrm{d}x)-\int^{T^{k}}_{t}\,\mathrm{d}M^{k,i}_{s},\end{multlined} is well-defined and has the unique solution (Y^{k,i},Z^{k,i},U^{k,i},M^{k,i}) , where \mathcal{L}(Y^{k,i}_{t}) denotes the law of Y^{k,i}_{t} . We have shown in Papapantoleon et al. 2024 that the system of mean-field BSDEs satisfies a backward propagation of chaos property, i.e. we know that for every (k,i)\in\overline{\mathbb{N}}\times\mathbb{N} we have (1.5) \displaystyle\big\|\big(Y^{k,i,N}-Y^{k,i},Z^{k,i,N}-Z^{k,i},U^{k,i,N}-U^{k,i},M^{k,i,N}-M^{k,i}\big)\big\|\xrightarrow[N\rightarrow\infty]{}0, under some suitable norm. Our goal now is to show that the property of backward propagation of chaos is also stable, namely, we want to show that for every i\in\mathbb{N} holds \left(Y^{k,i,N},Z^{k,i,N}\cdot X^{k,i,\circ}+U^{k,i,N}\star\widetilde{\mu}^{X^{k,i,\natural}},M^{k,i,N}\right)\\ \xrightarrow[(k,N)\rightarrow(\infty,\infty)]{}\left(Y^{\infty,i},Z^{\infty,i}\cdot X^{\infty,i,\circ}+U^{\infty,i}\star\widetilde{\mu}^{X^{\infty,i,\natural}},0\right), under some appropriate metric. To the best of our knowledge, this will be the first result of this kind. In order to accommodate the reader, we provide in Table 1.1 a roadmap of our approach. Using this table, we will describe the implications of the setting inherited from Papapantoleon et al. 2024 , i.e. , the set of Assumptions (B1) – (B7) presented in Section 3 , as well as the forthcoming results we intend to prove. The validity, for every k\in\overline{\mathbb{N}} , of the backward propagation of chaos is denoted by a solid, horizontal right arrow in Table 1.1 . The set of Assumptions (S1) – (S10) , which is presented in Section 3 and amounts to the convergence of the sequence of standard data, is denoted by the solid, vertical down arrow in the first column of the table. Now, we claim that the framework we have already set allows for the validity, on the one hand, of the stability of McKean–Vlasov BSDEs and, on the other hand, of the uniform (over k\in\overline{\mathbb{N}} ) backward propagation of chaos. The former is denoted by the dashed, vertical down arrow in the last column of the table. The latter was chosen not to be represented in Table 1.1 in order to keep the table as simple as possible. The conjunction of the two aforementioned results

Read the complete text