Abstract:We consider a general interacting particle system with interactions on a random graph, and study the large population limit of this system. When the sequence of underlying graphs converges to a graphon, we show convergence of the interacting particle system to a so called graphon stochastic differential equation. This is a system of uncountable many SDEs of McKean-Vlasov type driven by a continuum of Brownian motions. We make sense of this equation in a way that retains joint measurability and essentially pairwise independence of the driving Brownian motions of the system by using the framework of Fubini extension. The convergence results are general enough to cover nonlinear interactions, as well as various examples of sparse graphs. A crucial idea is to work with unbounded graphons and use the $L^p$ theory of sparse graph convergence.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is the asymptotic behavior of interacting particle systems on sparse random graphs. In particular, when graph theory converges to a graphon, how to describe the behavior of this system. Specifically, the paper focuses on the limit behavior of the interacting particle system when the underlying graph sequence $G_N$ converges to a graphon $g$. Here, graphon is a generalized graph representation method that can handle the case of uncountable vertices, and the graphon $g(x,y)$ represents the weight of the edge.
### Main research questions
1. **Well - posedness of graphon SDE**:
- The paper first explores the well - posedness problem of graphon SDE (Stochastic Differential Equation). The form of graphon SDE is as follows:
\[
dX_x^t = b\left(t, X_x^t, \int_I g(x, y) L(X_y^t) \lambda(dy)\right) dt+\sigma\left(t, X_x^t, \int_I g(x, y) L(X_y^t) \lambda(dy)\right) dB_x^t, \quad X_x^0\sim\gamma_x
\]
- Here, $g$ is a square - integrable graphon, $b$ and $\sigma$ are coefficient functions, $B_x$ is Brownian motion, and $\gamma_x$ is the initial distribution.
2. **Convergence of particle systems**:
- The paper studies the problem of the convergence of the particle system $Y_i^N$ on sparse graphs to the graphon SDE $X_x$. Specifically, the form of the particle system is as follows:
\[
Y_i^N = Y_i^N_0+\int_0^t b(s, Y_i^N_s, m_i^N Y_s) ds+\int_0^t \sigma(s, Y_i^N_s, m_i^N Y_s) dW_i^s, \quad Y_i^N_0\sim\gamma_i
\]
- Here, $m_i^N Y_s=\frac{1}{N}\sum_{j = 1}^N\zeta_{ij}^N\delta_{Y_j^N_s}$ is a positive measure, representing the interaction between particles.
3. **Propagation of chaos**:
- The paper also studies the propagation of chaos phenomenon of the particle system, that is, the independence of the particle system in the large - population limit. Specifically, the paper proves that under certain conditions, the empirical measure of the particle system $Y_i^N$ will converge to the distribution of the solution of the graphon SDE $X_x$.
### Key contributions
1. **Well - posedness of graphon SDE**:
- The paper proves the existence of the unique solution of the graphon SDE under the square - integrable graphon $g$. This is achieved by adapting the existence and uniqueness proof method of the classical McKean - Vlasov SDE and dealing with the subtle measurability problems related to Fubini extension.
2. **Convergence of particle systems**:
- The paper proves the $L^1$ convergence of the particle system $Y_i^N$ on sparse graphs to the graphon SDE $X_x$. Specifically, when the graphon $g$ and the graphon sequence $g_N$ are both bounded and the diffusion term $\sigma$ does not depend on the measure parameter, we have:
\[
\frac{1}{N}\sum_{i = 1}^N E\left[\sup_{t\in[0, T]}\left|X_i^N_t - X_x^t\right|\right]+\frac{1}{N}\sum_{i = 1}^N E\left[d_{BL}\left(m_i^N X_t, \int_I g(x_i, y) L(X_y^t)\lambda\right)\right]