Mean-field backward stochastic Volterra integral equations: well-posedness and related particle systems

Tao Hao, Ying Hu, Jiaqiang Wen · 2025

Read Mean-field backward stochastic Volterra integral equations: well-posedness and related particle systems by Tao Hao, Ying Hu, Jiaqiang Wen (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-2511-07173.txt.

Introduction

Tao Hao, Ying Hu, and Jiaqiang Wen establish the theory of mean-field backward stochastic Volterra integral equations. In these equations, the solution at each time depends on its own future path and on the distribution of the whole population. Such systems arise in large-population control and finance. The authors prove existence and uniqueness of solutions in two regimes for the generator's growth in the control variable: linear growth and quadratic growth.

They then connect the system to particle systems, finite populations whose collective behavior should approximate the mean-field limit. This propagation of chaos is proved under both growth conditions, with convergence rates derived in each case. The quadratic case requires the generator to be independent of the law of the control variable. The rates quantify how large a particle system must be before it faithfully represents the mean-field equation.

Catalog record

CREATORTao Hao, Ying Hu, Jiaqiang Wen
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE137 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2511-07173

Browse the full library · About the Tower of Babel

Full text

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

Mean-field backward stochastic Volterra integral equations: well-posedness and related particle systems Tao Hao, Ying Hu, Jiaqiang Wen Abstract Abstract: This paper studies the mean-field backward stochastic Volterra integral equations (mean-field BSVIEs) and associated particle systems. We establish the existence and uniqueness of solutions to mean-field BSVIEs when the generator g is of linear growth or quadratic growth with respect to Z , respectively. Moreover, the propagation of chaos is analyzed for the corresponding particle systems under two conditions. When g is of linear growth in Z , the convergence rate is proven to be of order {\cal Q}(N) . When g is of quadratic growth in Z and is independent of the law of Z , we not only establish the convergence of the particle systems but also derive a convergence rate of order {\cal O}(N^{-\frac{1}{2\lambda}}) , where \lambda>1 . Key words: Backward stochastic Volterra integral equation; mean-field; particle system; quadratic growth; convergence rate AMS subject classifications. 60H10; 60H30 1 Introduction Let (\Omega,\mathscr{F},\mathbb{P}) be a complete probability space, where a d -dimensional standard Brownian motion W is defined, and the completed augmented natural filtration is denoted by \mathbb{F}=\{\mathscr{F}_{t}\}_{0\leqslant t\leqslant T} . In 1990, Pardoux and Peng [ 22 ] first established the existence and uniqueness of solutions to nonlinear backward stochastic differential equations (BSDEs, for short) under Lipschitz assumptions. Since then, this theory has found extensive applications in stochastic optimal control, partial differential equations, and financial mathematics. One important application of BSDEs in economics is their use in modeling stochastic differential utility (SDU, for short), as illustrated below: Y(t)=\mathbb{E}\Big[\xi+\int_{t}^{T}h(s,Y(s))ds\Big|\mathscr{F}_{t}\Big], where \xi is a square-integrable random variable, h:[0,T]\times\mathbb{R}\rightarrow\mathbb{R} is an appropriate function, and \mathbb{E}[\cdot|\mathscr{F}_{t}] represents the conditional expectation under the probability \mathbb{P} . This SDU can be interpreted as the solution to a particular BSDE with terminal value \xi and generator h (see Duffie and Epstein [ 5 ] ). Moreover, El Karoui, Peng, and Quenez [ 6 ] extended their work by introducing the concept of a general SDU. Nonetheless, the standard expected utility function (EUF, for short) is typically expressed as Y(t)=\mathbb{E}\Big[\xi e^{\gamma(T-t)}+\int_{t}^{T}u(c(s))e^{-\gamma(s-t)}ds\Big|\mathscr{F}_{t}\Big], where u(\cdot) is a utility function, c(\cdot) is a consumption process, and \gamma is the discount rate. The standard EUF, which includes a consumption process, a discount rate, and a utility function, emphasizes that both the terminal value \overline{\psi}(t)\equiv\xi e^{\gamma(T-t)} and the generator \overline{g}(t,s)\equiv u(c(s))e^{-\gamma(s-t)} are time-dependent and have memory effects, i.e., a feature not naturally captured by classical BSDEs. To better describe this phenomenon, many researchers have turned it to be backward stochastic Volterra integral equations (BSVIEs, for short) of the following form: Y(t)=\psi(t)+\int_{t}^{T}g(t,s,Y(s),Z(t,s))ds-\int_{t}^{T}Z(t,s)dW(s),\quad t\in[0,T]. For this class of equations, Lin [ 21 ] established the existence and uniqueness of adapted solutions under Lipschitz conditions, Yong [ 32 ] introduced a class of general BSVIEs and proved their well-posedness. Readers may refer to [ 11 , 12 , 24 , 25 , 26 , 27 , 28 , 29 , 30 ] for new developments of BSVIEs and their applications. Note that although the Lipschitz condition facilitates mathematical derivations, it often restricts the range of interesting applications, such as risk-sensitive problems. To address this limitation, Peng [ 23 ] proposed relaxing the Lipschitz assumption on the generator g when studying the existence and/or uniqueness of adapted solutions to BSDEs. This suggestion has inspired numerous research efforts in the field. For instance, Lepeltier and San Martin [ 19 ] studied the case where the generator g is continuous in (Y,Z) and has linear growth. Kobylanski [ 18 ] , Briand and Hu [ 3 , 4 ] , as well as Bahlali, Eddahbi, and Ouknine [ 1 ] , worked on one-dimensional BSDEs with quadratic growth (QBSDEs). Hu and Tang [ 15 ] , Xing and Zitkovic [ 31 ] , and Fan, Hu, and Tang [ 8 ] focused on multi-dimensional QBSDEs. For BSVIEs, similar efforts have been made to relax the Lipschitz assumption on the generator g . For instance, Shi and Wang [ 24 ] investigated general BSVIEs under non-Lipschitz conditions. Moreover, Wang, Sun, and Yong [ 28 ] extended Lin’s work [ 21 ] from linear growth to quadratic growth and established the well-posedness of both local and global solutions for BSVIEs with quadratic growth. However, to the best of our knowledge, no fundamental results have yet been established for general mean-field BSVIEs (i.e., the generator g depends on the distribution of the solution (Y,Z)) , let alone for related particle systems. Note that the work of Shi, Wang, and Yong [ 25 ] dealt with classical mean-field BSVIEs, in which the generator g depends on the expectation of the solution (Y,Z) ; meanwhile, Jiang, Li, and Wei [ 16 ] discussed a special case within the general setting. In this paper, motivated by the wide applications of particle systems, we study general mean-field BSVIEs. Specifically, we focus on analyzing the convergence rate of particle systems associated with the following BSVIEs: For i=1,\cdots,N , \displaystyle Y^{N,i}(t) \displaystyle=\psi^{i}(t)+\int_{t}^{T}g(t,s,Y^{N,i}(s),Z^{N,i,i}(t,s),\mu^{N}(s),\nu^{N}(t,s))ds (1.1) \displaystyle-\int_{t}^{T}\sum\limits_{j=1}^{N}Z^{N,i,j}(t,s)dW^{j}(s),\quad t\in[0,T], where \{W^{j};1\leqslant j\leqslant N\} are N independent d -dimensional standard Brownian motions on the complete probability space (\Omega,\mathscr{F},\mathbb{P}) , and \mu^{N}(s)\triangleq\frac{1}{N}\sum\limits_{i=1}^{N}\delta_{Y^{N,i}(s)}\quad\hbox{and}\quad\nu^{N}(t,s)\triangleq\frac{1}{N}\sum\limits_{i=1}^{N}\delta_{Z^{N,i,i}(t,s)}. (1.2) The equation above is a multi-dimensional BSVIE. As the system size becomes very large, i.e., as N\rightarrow\infty , based on the theory of convergence under the Wasserstein distance for the empirical measure (as discussed in Fournier and Guillin [ 10 ] ), the particle system described in Eq. ( 1.1 ) will converge in structure to the following mean-field BSVIE: Y(t)=\psi(t)+\int_{t}^{T}g(t,s,Y(s),Z(t,s),\mathbb{P}_{Y(s)},\mathbb{P}_{Z(t,s)})ds-\int_{t}^{T}Z(t,s)dW(s),\quad t\in[0,T]. (1.3) The process \psi(\cdot) and the mapping g are referred to as the free term and the generator, respectively. We denote equation ( 1.3 ) as a general mean-field BSVIE. The term “general” here indicates that the generator g depends on the distributions of Y(s) and that of Z(t,s) , rather than merely their expectations. This convergence implies that various analyses related to the particle system in equation ( 1.1 ), such as error estimates of solutions and asymptotic Nash equilibrium, can be effectively addressed using the framework of equation ( 1.3 ). The main goal of this paper is to study the convergence and its rate for the particle system described by ( 1.1 )-( 1.2 ). Before studying our main results, we first establish the well-posedness of BSVIE ( 1.3 ). To obtain more general results, we consider both linear growth and quadratic growth cases. For the linear growth case, we initially prove the existence and uniqueness of the solution to BSVIE ( 1.3 ) by parameterizing the time variable t and applying the contraction mapping principle (see Theorem 3.2 ). Subsequently, we derive a convergence rate of order {\cal Q}(N) by combining the t -parameterization method with convergence theory in the Wasserstein distance for empirical measures (see Theorem 4.2 ). Note that when g depends on the law of Z(\cdot,\cdot) , we need to work within the space L^{p}_{\mathbb{F}}(\Delta[0,T];\mathbb{R}^{d}) with 1<p<2 , because we only have that \int_{0}^{T}\int_{t}^{T}\mathbb{E}|\overline{Z}^{i}(t,s)|^{2}dsdt<\infty (see ( 4.11 )). However, when g is independent of the law of Z(\cdot,\cdot) , we can work in the space L^{p}_{\mathbb{F}}(\Delta[0,T];\mathbb{R}^{d}) for p>1 (see 4.4 ). Furthermore, even for general mean-field BSDEs with Lipschitz conditions, our result concerning the convergence rate is new. To the best

Read the complete text