Abstract:This paper studies large deviations of a ``fully coupled" finite state mean-field interacting particle system in a fast varying environment. The empirical measure of the particles evolves in the slow time scale and the random environment evolves in the fast time scale. Our main result is the path-space large deviation principle for the joint law of the empirical measure process of the particles and the occupation measure process of the fast environment. This extends previous results known for two time scale diffusions to two time scale mean-field models with jumps. Our proof is based on the method of stochastic exponentials. We characterise the rate function by studying a certain variational problem associated with an exponential martingale.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is the large deviation principle (LDP) in the path space of mean - field interacting particle systems in a rapidly changing environment. Specifically, the research object is a "fully coupled" finite - state mean - field interacting particle system, where the empirical measure of particles evolves on a slow time scale, while the random environment evolves on a fast time scale. The main objective is to establish the joint path - space large deviation principle for the particle empirical measure process and the fast - environment occupation measure process.
### Background and Motivation of the Paper
1. **Multi - time - scale Models**: Many real - world network systems can be modeled as two - time - scale mean - field models. For example, retry queuing systems and wireless local area networks with local interactions. There may be multiple seemingly "stable operating points", that is, metastable points, in these networks. Understanding these metastable phenomena is crucial for evaluating the performance of the system.
2. **Metastable Phenomena**: Researchers are usually interested in the following metastable phenomena:
- The average time that the network stays near an operating point.
- The average time required to transition from one stable operating point to another.
- The average time for the system to reach a state sufficiently close to the stationary state.
### Main Results
The main result of the paper is the establishment of the joint path - space large deviation principle for the particle empirical measure process and the fast - environment occupation measure process. Specifically, assuming that the initial conditions satisfy the large deviation principle, then the sequence \(\{(µ_N(t), θ_N(t)), 0 \leq t \leq T\}\) satisfies the large deviation principle on \(D([0, T], M_1(X)) \times D^\uparrow([0, T], M(Y))\) with the rate function:
\[ I(µ, θ) := I_0(µ(0)) + J(µ, θ), \]
where \(J(µ, θ)\) is defined as:
\[ J(µ, θ) := \int_{[0, T]} \left\{ \sup_{α \in \mathbb{R}^{|X|}} \left( \langle α, (\dot{µ}_t - \bar{Λ}^*_{µ_t, m_t} µ_t) \rangle - \int_{X \times E_X} τ(Dα(x, ∆)) \bar{λ}_{x, x'}(µ_t, m_t) µ_t(dx) \right) + \sup_{g \in B(Y)} \int_Y \left( -L_{µ_t} g(y) - \int_{E_Y} τ(Dg(y, ∆)) γ_{y, y'}(µ_t) \right) m_t(dy) \right\} dt, \]
when the mapping \([0, T] \ni t \mapsto µ_t \in M_1(X)\) is absolutely continuous, and \(\theta\) can be represented as \(\theta(dtdy) = m_t(dy)dt\) as a measure on \([0, T] \times Y\), where \(m_t \in M_1(Y)\) holds for almost all \(t \in [0, T]\); otherwise \(J(µ, θ) = +\infty\).
### Proof Methods
1. **Exponential Compactness**: First, prove that the sequence \(\{(µ_N, θ_N)\}\) is exponentially compact on \(D([0, T], M_1(X)) \times D^\uparrow([0, T], M(Y))\).
2. **Sub - sequence Rate Functions**: By defining the exponential martingale related to the Markov process \((µ_N, Y_N)\), derive the equations that the sub - sequence rate functions must satisfy.
3. **Variational Problems**: Define the candidate rate function \(I^*\) as a variational problem and prove that the form of \(I^*\) is consistent with \(J(µ, θ)\).
4. **Approximation Arguments**: Extend the solution of the above variational problem to the entire space \(D([0, T], M_1(X)) \times D^\uparrow([0, T], M(Y))\) through approximation methods.