Abstract:This work introduces a new general approach for the numerical analysis of stable equilibria to second order mean field games systems in cases where the uniqueness of solutions may fail. For the sake of simplicity, we focus on a simple stationary case. We propose an abstract framework to study these solutions by reformulating the mean field game system as an abstract equation in a Banach space. In this context, stable equilibria turn out to be regular solutions to this equation, meaning that the linearized system is well-posed. We provide three applications of this property: we study the sensitivity analysis of stable solutions, establish error estimates for their finite element approximations, and prove the local converge of Newton's method in infinite dimensions.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is how to conduct numerical analysis in the second - order mean - field game system when the uniqueness of solutions may not hold. Specifically, the author focuses on a steady - state case with a pure quadratic Hamiltonian. The main contribution of the paper lies in introducing an abstract framework to study these solutions by reformulating the mean - field game system as an abstract equation in Banach spaces. In this context, the stable equilibrium points are proven to be regular solutions of this equation, that is, the linearized system is well - posed. Based on this property, the author provides three applications:
1. **Stability analysis**: Study the sensitivity analysis of stable solutions to perturbations of the coupling function \(f\) and the distribution \(m_0\).
2. **Error estimate of the finite - element method**: Establish the error estimate of the finite - element approximation of stable solutions.
3. **Local convergence of Newton's method**: Prove the local convergence of stable solutions when applying Newton's method in infinite - dimensional spaces.
### Abstract of the paper
This paper presents a new general method for the numerical analysis of stable equilibrium points in the second - order mean - field game system, especially when the uniqueness of solutions may not hold. We focus on a steady - state case with a pure quadratic Hamiltonian. We propose an abstract framework to study these solutions by reformulating the mean - field game system as an abstract equation in Banach spaces. In this case, the stable equilibrium points are proven to be regular solutions of this equation, that is, the linearized system is well - posed. We provide three applications of this property: study the sensitivity analysis of stable solutions to perturbations of the coupling function and the distribution, establish the error estimate of its finite - element approximation, and prove the local convergence when applying Newton's method in infinite - dimensional spaces.
### Main contributions
1. **Stability analysis**:
- Using the implicit function theorem, the sensitivity analysis of stable solutions to perturbations of the coupling function \(f\) and the distribution \(m_0\) is studied.
2. **Error estimate of the finite - element method**:
- Using the Brezzi - Rappaz - Raviart theory, the existence and error estimate of the finite - element approximation of stable solutions are obtained. Compared with previous works, our results are applicable to any stable solution without requiring the coupling term \(f\) to be monotonic.
3. **Local convergence of Newton's method**:
- The iterative convergence rate of stable solutions when applying Newton's method in various function spaces is provided. Different from previous methods, our results are directly based on the convergence theory of the classical Newton iteration in function spaces, allowing the handling of non - monotonic couplings and obtaining the convergence rate under a stronger norm.
### Mathematical background
- **Mean - field game system**:
\[
\begin{cases}
-\Delta u+\frac{1}{2}|Du|^{2}+\lambda u = f(m)&\text{in }\mathbb{T}^{d},\\
-\Delta m-\text{div}(m Du)+\lambda m=\lambda m_0&\text{in }\mathbb{T}^{d},
\end{cases}
\]
where \(\lambda> 0\) is a given constant, and \(m_0:\mathbb{T}^{d}\to\mathbb{R}\) and \(f:\mathbb{R}\to\mathbb{R}\) are given functions.
- **Definition of stable solutions**:
- Let \(f\in C_b^1(\mathbb{R})\), and \((u, m)\) be a classical solution of the above system. If \((v,\rho)=(0,0)\) is the unique classical solution of the following system, then \((u, m)\) is said to be stable:
\[
\begin{cases}
-\Delta v+Du\cdot Dv+\lambda v = f'(m)\rho&\text{in }\mathbb{T}^{d},\\
-\Delta \rho-\text{div}(\rho Du)+\lambda \rho=\text{d}