Weakly coupled mean-field game systems (Q739437): Difference between revisions

In this very interesting paper the authors study the existence of solutions to the first-order mean-field games (MFGs) model arising in optimal switching. We will remind the definition of the above mentioned MFG. It is a class of methods used to study large populations of rational, non-cooperative agents. Here the attention is focused on some application in the optimal switching. The model of the above mentioned games includes a weakly coupled system of Hamilton-Jacobi equations of the obstacle type, and in addition a system of transport equations. The periodic boundary conditions it seems being of interest here. The spatial domain is the \(N\)-dimensional flat torus \(\mathbb{T}^N\). This MFG model is determined by the following quantities: a value function \(u:\mathbb{T}^N\to \mathbb{R}^d\), a probability density \(\theta :\mathbb{T}^N\to (\mathbb{R}^{+})^d\), and a switching current \(\nu \). These together satisfy the system: \[ (i) \max\biggl(H^i(Du^i,x)+u^i-g(\theta^i), \max\limits_{j}(u^i-u^j-\psi^{ij})\biggr)=0, \] \[ (ii) -\text{div}(D_pH^i(Du^i,x)\theta^i)+\theta^i+ \sum\limits_{j\neq i}(\nu^{ij}-\nu^{ji})=1. \] If \(1\leq i,j\leq d\), then \(\nu^{ij}\) is a non-negative measure on \(\mathbb{T}^N\) supported in the set \(u^i-u^j-\psi^{ij}=0\). Here \(u^i (x)\) is the value function for an agent whose spatial location is \(x\), and whose mode is \(i\). The function \(\theta^i(x)\geq 0\) is the density of the agents on \(\mathbb{T}^N\times \{1, \ldots , d\}\). Note that the above stated system models a stationary population of agents. The authors explain that ``each agent moves in \(\mathbb{T}^N\), and can switch between different modes that are determined by the index \(i\). Their actions seek to minimize a certain cost. Agents can change their state by continuously modifying the spatial position \(x\), and by switching between different modes \(i\) to \(j\) at a cost \(\psi^{ij}\)''. There are some assumptions for the Hamiltonian \(H^i\), the nonlinearity \(g\), and the switching cost \(\psi^{ij}\): \(H^i\in C^{\infty }\) (\(H^i>0\)) \(\forall i\); \(g\in C^{\infty }\) and strictly increasing; \(\psi^{ij}\in C^{\infty }(\mathbb{T}^N)\), and \(\psi^{ij}(x)>0\); \(\int_{\mathbb{T}^N}\theta g(\theta)dx\geq -C_1\), the integral of \(\theta \) on \(\mathbb{T}^N\) is also limited; moreover, \(g(\theta )\leq (1/2)\theta g(\theta )+C\) (\(C>0\)). The Hamiltonian and its derivatives satisfy some estimates. Also there exists the Lagrangian \(L^i(x,v)=\sup \limits_{p\in \mathbb{R}^N}-p\cdot v-H^i(p,x)\), \(p\in\mathbb{R}^N\), \(x\in\mathbb{T}^N\), satisfying certain conditions. The main result is that there exists a solution \((u,\theta )\) of the system under consideration (i), (ii) with \(u\in (W^{2,2 }(\mathbb{T}^N))^d\cap (C^{\gamma }(\mathbb{T}^N))^d\), \(\gamma\in (0,1)\), \(\theta\in (W^{1,2 }(\mathbb{T}^N))^d\). To prove this the authors use an \(\varepsilon \)-system \[ (i_{\varepsilon}) H^i(Du^i,x)+u^i + \sum\limits_{j\neq i}\beta_{\varepsilon } (u^i-u^j-\psi^{ij})=g(\theta^i), \] \[ (ii_{\varepsilon}) \;\;-\text{div}(D_pH^i(Du^i,x)\theta^i)+\theta^i+ \sum\limits_{j\neq i}\beta_{\varepsilon }^{\prime } (u^{i}-u^{j}-\psi^{ij})\theta^i- \beta_{\varepsilon }^{\prime } (u^{j}-u^{i}-\psi^{ji})\theta^j=1, \] where the penalty function \(\beta_{\varepsilon } \in C^{\infty }\) is increasing, and \(\varepsilon >0\). Here assume that \(\beta_{\varepsilon }(s)\to \infty \) as \(\varepsilon\to 0\) (\(s>0\)); \(\beta_{\varepsilon }(s)=0\) for \(s\leq 0\). First the authors prove existence of solutions for the \(\varepsilon \)-system (\((i_{\varepsilon})\), \((ii_{\varepsilon})\)), which is \((u,\theta )\in (C^{\infty }(\mathbb{T}^N))^d\times (C^{\infty }(\mathbb{T}^N))^d\) (\(\theta^i\) positive). Then prove \(\varepsilon \)-independent bounds, and subsequently, consider the limit \(\varepsilon \to 0\).