Abstract:We study a fully-coupled system of conditional slow-fast McKean-Vlasov Stochastic Differential Equations that exhibit full dependence on both the slow and fast components, as well as on the conditional law of the slow component. Our aim is to derive convergence rates to its homogenized limit, without making periodicity assumptions. To prove our results, we utilize a perturbation method for equations posed in the Wasserstein space.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is the behavior of the slow variable \(X^\varepsilon_t\) when the scale parameter \(\varepsilon\) of the fast variable tends to zero in the conditional slow - fast McKean - Vlasov stochastic differential equations (SDEs) system. Specifically, the authors aim to study the limit behavior of the slow variable \(X^\varepsilon_t\) and determine the rate at which it converges to the averaged process. This process is analyzed in the presence of the common noise \(B^0_t\) without the need to assume periodicity. To prove their results, the authors use the perturbation method in the Wasserstein space.
### Specific problem description
- **Research object**: A fully coupled conditional slow - fast McKean - Vlasov SDEs system, which depends on the slow and fast variables as well as the conditional distribution of the slow variable.
- **Objective**: Derive the rate at which the slow variable \(X^\varepsilon_t\) converges to its homogenized limit.
- **Method**: Utilize the perturbation method in the Wasserstein space without assuming periodicity.
### Mathematical model
Consider the following slow - fast McKean - Vlasov SDEs system:
\[
dX^\varepsilon_t=\left(\frac{\lambda}{\varepsilon} f(X^\varepsilon_t, Y^\varepsilon_t, L(X^\varepsilon_t|\mathcal{F}^0_t)) + c(X^\varepsilon_t, Y^\varepsilon_t, L(X^\varepsilon_t|\mathcal{F}^0_t))\right)dt+\sigma(X^\varepsilon_t, Y^\varepsilon_t, L(X^\varepsilon_t|\mathcal{F}^0_t))dW_t+\sigma_0^1(X^\varepsilon_t, L(X^\varepsilon_t|\mathcal{F}^0_t))dB^0_t
\]
\[
dY^\varepsilon_t = \frac{1}{\varepsilon}\left(\frac{1}{\varepsilon}f(X^\varepsilon_t, Y^\varepsilon_t, L(X^\varepsilon_t|\mathcal{F}^0_t))+g(X^\varepsilon_t, Y^\varepsilon_t, L(X^\varepsilon_t|\mathcal{F}^0_t))\right)dt+\frac{1}{\varepsilon}\left(\tau_1(X^\varepsilon_t, Y^\varepsilon_t, L(X^\varepsilon_t|\mathcal{F}^0_t))dW_t+\tau_2(X^\varepsilon_t, Y^\varepsilon_t, L(X^\varepsilon_t|\mathcal{F}^0_t))dB_t+\sigma_0^2(X^\varepsilon_t, L(X^\varepsilon_t|\mathcal{F}^0_t))dB^0_t\right)
\]
where:
- \(L(X^\varepsilon_t|\mathcal{F}^0_t)\) is the conditional distribution of \(X^\varepsilon_t\).
- \(b, c, f, g:\mathbb{R}^d\times\mathbb{R}^d\times\mathcal{P}_2(\)