Compound Poisson Approximation for Stochastic Volterra Equations with Singular Kernels

Basic Information

• Paper ID: 2510.27323
• Title: Compound Poisson Approximation for Stochastic Volterra Equations with Singular Kernels
• Authors: Xicheng Zhang (Beijing Institute of Technology), Yuanlong Zhao (Wuhan University)
• Classification: math.PR (Probability Theory)
• Submission Date: October 31, 2025
• Paper Link: https://arxiv.org/abs/2510.27323

Abstract

This paper establishes strong convergence results for stochastic differential equations (SDEs) and Volterra-type SDEs under compound Poisson process approximation, and derives explicit convergence rates. The key advantage of compound Poisson approximation over classical Euler-Maruyama methods is that it does not require continuity of drift coefficients in the time variable and can even handle singularities. Numerical experiments verify the stability of the method.

Research Background and Motivation

1. Core Problem

This paper studies the numerical approximation of stochastic differential equations and stochastic Volterra equations, with particular focus on cases where coefficients exhibit temporal singularities.

2. Problem Significance

• Theoretical Importance: Stochastic Volterra equations hold an important position in stochastic analysis, particularly in describing stochastic systems with memory effects
• Application Value: Important models such as stochastic equations driven by fractional Brownian motion and stochastic Navier-Stokes equations can be reduced to this class of problems
• Numerical Challenge: When kernel functions exhibit singularities, traditional numerical methods face convergence and stability issues

3. Limitations of Existing Methods

• Euler-Maruyama Method: Requires Hölder continuity of coefficients in the time variable, unable to handle discontinuous or singular cases
• Improved EM Methods: While relaxing certain conditions (e.g., one-sided Lipschitz conditions), still require certain temporal regularity
• Methods for Singular Kernels: Existing literature primarily addresses non-singular kernels; theory for handling singular kernels remains incomplete

4. Research Motivation

• Develop numerical methods capable of handling temporal singularities
• Establish explicit convergence rate estimates
• Provide reliable numerical tools for important stochastic processes such as fractional Brownian motion

Core Contributions

1. Established Strong Convergence Theory for Compound Poisson Approximation: First systematic proof of strong convergence for SDEs and stochastic Volterra equations under compound Poisson approximation
2. Derived Explicit Convergence Rates:
   • For standard SDEs: convergence rate is $\varepsilon^{\gamma \wedge \frac{\beta}{2}}$
   • For stochastic Volterra equations: convergence rate is $\varepsilon^{\gamma/(2(2+\gamma))}$
3. Breakthrough in Handling Singular Coefficients: The method does not require temporal continuity of drift coefficients and can handle temporal singularities
4. Provided Complete Theoretical Framework: Including moment estimates, Gronwall inequality applications, and technical lemmas forming a complete proof system
5. Presented Concrete Application Examples: Detailed verification of assumption conditions for fractional Brownian motion (Hurst parameter $H \in (0,1/2) \cup (1/2,1)$)
6. Numerical Verification: Two numerical experiments verify the effectiveness and stability of the method

Methodology Details

Problem Formulation

Problem 1: Standard SDE Consider the stochastic differential equation on $\mathbb{R}^d$: $X_t = X_0 + \int_0^t \sigma(s, X_s) dW_s + \int_0^t b(s, X_s) ds$

where $\sigma: \mathbb{R}_+ \times \mathbb{R}^d \to \mathbb{R}^d \otimes \mathbb{R}^m$, $b: \mathbb{R}_+ \times \mathbb{R}^d \to \mathbb{R}^d$, and $W_t$ is an $m$-dimensional standard Brownian motion.

Problem 2: Stochastic Volterra Equation$Y_t = Y_0 + \int_0^t \sigma(t, s, Y_s) dW_s + \int_0^t b(t, s, Y_s) ds$

Key characteristic: coefficients depend on both current time $t$ and integration variable $s$, reflecting memory effects.

Model Architecture

1. Compound Poisson Process Construction

Basic Poisson Process: Let $(T_k)_{k \in \mathbb{N}}$ be i.i.d. exponential random variables with parameter 1: $P(T_k \geq t) = e^{-t}, \quad t \geq 0$

Define jump times: $S_0 = 0$, $S_n = S_{n-1} + T_n$

Standard Poisson process: $N_t = \max\{n: S_n \leq t\}$

Scaled Poisson Process: For given $\varepsilon > 0$, define: $N_t^\varepsilon := \varepsilon N_{t/\varepsilon}, \quad \tilde{N}_t^\varepsilon := N_t^\varepsilon - t$

where $N_t^\varepsilon$ is a Poisson process with jump size $\varepsilon$ and intensity $1/\varepsilon$.

Compound Poisson Process: $W_{N_t^\varepsilon} = \sum_{k=1}^{N_{t/\varepsilon}} (W_{k\varepsilon} - W_{(k-1)\varepsilon})$

This is a compound Poisson process with intensity measure $\nu_\varepsilon \sim N(0, \varepsilon I_d)$.

2. Numerical Scheme

Compound Poisson Approximation for SDEs: $X_t^\varepsilon = X_0 + \int_0^t \sigma(s, X_{s-}^\varepsilon) dW_{N_s^\varepsilon} + \int_0^t b(s, X_{s-}^\varepsilon) dN_s^\varepsilon$

Fully discrete form: $X_t^\varepsilon = X_0 + \sum_{k=1}^{N_{t/\varepsilon}} [\sigma(S_k^\varepsilon, X_{S_{k-1}^\varepsilon}^\varepsilon)(W_{k\varepsilon} - W_{(k-1)\varepsilon}) + \varepsilon b(S_k^\varepsilon, X_{S_{k-1}^\varepsilon}^\varepsilon)]$

Compound Poisson Approximation for Stochastic Volterra Equations: $Y_t^\varepsilon = Y_0 + \int_0^t \sigma(t, s, Y_{s-}^\varepsilon) dW_{N_s^\varepsilon} + \int_0^t b(t, s, Y_{s-}^\varepsilon) dN_s^\varepsilon$

Fully discrete form: $Y_t^\varepsilon = Y_0 + \sum_{k=1}^{N_{t/\varepsilon}} [\sigma(t, S_k^\varepsilon, Y_{S_{k-1}^\varepsilon}^\varepsilon)(W_{k\varepsilon} - W_{(k-1)\varepsilon}) + \varepsilon b(t, S_k^\varepsilon, Y_{S_{k-1}^\varepsilon}^\varepsilon)]$

Technical Innovations

1. Stochastic Time Discretization

Unlike deterministic time grids, this approach uses jump times of the Poisson process as discretization points, implementing a stochastic discretization strategy with key advantages:

• Naturally introduces randomness, independent of Brownian motion increments
• Does not require temporal continuity of coefficients
• Can handle temporal singularities

2. Error Decomposition Strategy

For SDEs, the error is decomposed into four terms: $X_t - X_t^\varepsilon = I_1(t) + I_2(t) + I_3(t) + I_4(t)$

where:

• $I_1(t)$: Difference between true path and its values at random time points
• $I_2(t)$: Difference between Brownian motion and compound Poisson process (most critical term)
• $I_3(t)$: Difference in coefficients across different paths
• $I_4(t)$: Discretization error of drift term

3. Key Technical Lemmas

Lemma 2.1 (Moment Estimates for Poisson Jump Times): For $\alpha \in (0,1]$, $\beta \geq 0$: $\mathbb{E}|r^\alpha - (S_k^\varepsilon)^\alpha|^\beta \lesssim (k\varepsilon)^{\alpha\beta} k^{-\beta/2}, \quad r \in [(k-1)\varepsilon, k\varepsilon]$

This estimate is more refined than direct application of $|a^\alpha - b^\alpha| \leq |a-b|^\alpha$.

Lemma 2.3 (Poisson Process Deviation Estimate): For any $p > 0$: $\mathbb{E}|N_t^\varepsilon - t|^p \leq C\varepsilon^{p/2}(t^{p/2} \vee t)$

4. Refined Treatment of $I_2(t)$ Term

This is the most difficult term, requiring:

• Representation via compensated Poisson random measures
• Doob's maximal inequality
• Hölder continuity assumption in time variable $(H_\sigma^t)$
• Refined summation estimates

For Volterra equations, handling $I_2$ is more complex, requiring introduction of intermediate discretization parameter $\delta$ and decomposition into four sub-terms.

Experimental Setup

Datasets (Numerical Experiment Scenarios)

Experiment 1: Linear SDE with Singular Temporal Drift

$dX_t = \mu(t)X_t dt + \sigma(t)X_t dW_t$

where: