Abstract:In several recent works on infinite-dimensional systems of ODEs \cite{cao_derivation_2021,cao_explicit_2021,cao_iterative_2024,cao_sticky_2024}, which arise from the mean-field limit of agent-based models in economics and social sciences and model the evolution of probability distributions (on the set of non-negative integers), it is often shown that the Gini index serves as a natural Lyapunov functional along the solution to a given system. Furthermore, the Gini index converges to that of the equilibrium distribution. However, it is not immediately clear whether this convergence at the level of the Gini index implies convergence in the sense of probability distributions or even stronger notions of convergence. In this paper, we prove several results in this direction, highlighting the interplay between the Gini index and other popular metrics, such as the Wasserstein distance and the usual $\ell^p$ distance, which are used to quantify the closeness of probability distributions.
What problem does this paper attempt to address?
The main problem that this paper attempts to solve is: when the Gini index converges to its equilibrium value, does it mean that the underlying probability distribution also converges in some sense? Specifically, the author explores the behavior of the Gini index as a Lyapunov functional in an infinite - dimensional ordinary differential equation system, and attempts to prove that this convergence can be generalized to stronger forms of convergence, such as the Wasserstein distance or the $\ell^p$ distance.
### Background and Motivation of the Paper
1. **Infinite - Dimensional ODE System**:
- The infinite - dimensional ordinary differential equation system obtained from the mean - field limit of agent - based stochastic models has wide applications in economics and social sciences.
- These systems are usually represented as $p' = Q[p]$, where $p(t)$ is a probability distribution on the set of non - negative integers, and the total probability mass and the mean remain constant.
2. **Gini Index as a Lyapunov Functional**:
- The Gini index $G[p]$ is defined as:
\[
G[p]=\frac{1}{2\mu}\sum_{i\in\mathbb{N}}\sum_{j\in\mathbb{N}}|i - j|p_ip_j
\]
- In many cases, the Gini index is decreasing along the solutions of the system, that is:
\[
\frac{d}{dt}G[p]\leq0
\]
3. **The Core of the Problem**:
- Although the Gini index converges to its equilibrium value $G[p^*]$, it is not clear whether this means that the probability distribution $p(t)$ converges to its unique equilibrium distribution $p^*$.
- The goal of the paper is to answer the following question: if the convergence of the Gini index can be proved:
\[
G[p(t)] - G[p^*]\to0\quad\text{as}\quad t\to\infty
\]
then in what sense will $p(t)$ converge to $p^*$?
### Main Results
- **Theorem 1**: Suppose $p\in V_\mu$ and $\mu\in\mathbb{R}^+$, then:
\[
W_1(p, p^*)\leq
\begin{cases}
2\mu(G[p] - G[p^*]),&\text{if}\ \mu\in\mathbb{N}^+\\
\frac{2\mu}{\min\{\mu-\lfloor\mu\rfloor,\lfloor\mu\rfloor + 1-\mu\}}(G[p] - G[p^*]),&\text{if}\ \mu\in\mathbb{R}^+\setminus\mathbb{N}^+
\end{cases}
\]
where $W_1(p, p^*)$ represents the Wasserstein distance (first - order) between $p$ and $p^*$.
- **Theorem 2**: Suppose $p\in V_\mu$ and $\mu\in\mathbb{R}^+$, then:
\[
\|p - \delta_0\|_{\ell^1}\leq2\sqrt{\mu}\sqrt{1 - G[p]}
\]
In particular, when $G[p]\to1$, $\|p - \delta_0\|_{\ell^1}\to0$, that is, $p$ converges to the Dirac delta distribution $\delta_0$ concentrated at 0.
### Conclusion
Through these results, the paper fills the gap in the Gini index.