Effective filtering for slow-fast systems via Wong-Zakai approximation (Q6134096)

In the present paper, the authors consider the filtering problem for the following slow-fast system \[ \left\{ \begin{array}{lr} dX^{\varepsilon} =\frac{1}{\varepsilon} AX^{\varepsilon}dt+ \frac{1}{\varepsilon} f(X^{\varepsilon}, Y^{\varepsilon})dt+ +\frac{1}{\sqrt{\varepsilon}}q_1dw_1,&\quad \text{in}\, H_1, \\ dY^{\varepsilon} =BY^{\varepsilon}dt+ g(X^{\varepsilon}, Y^{\varepsilon})dt+ q_2dw_2,&\quad \text{in}\, H_2, \end{array}\right. \tag{1} \] where \(H_1\) and \(H_2\) are infinite separable Hilbert spaces, \(A, B\) are linear operators, \(f, g\) are nonlinear functions, parameter \(\varepsilon\) is sufficiently small, \(w_1\) and \(w_2\) are mutually independent standard real-valued Brownian motions. Recently \textit{H. Qiao} [J. Math. Anal. Appl. 487, No. 2, Article ID 123979, 22 p. (2020; Zbl 1440.37057)] and \textit{H. Qiao} et al. [Nonlinearity 31, No. 10, 4649--4666 (2018; Zbl 1394.60037)] published results on filtering problems, and proved that the filtering on the corresponding slow manifold can approximate that of the original system (1). The fact that Brownian motion is not differentiable stimulated people to develop a smoother approximation for the filtering of the considered stochastic problems. One of the well-known methods is Wong-Zakai approximation. The key of the technique is to replace white noise with colored noise, the systems driven by which have many advantages in numerical simulations. In this paper the authors use the Wong-Zakai approximation to investigate the filtering problem for (1), i.e., they consider the corresponding Wong-Zakai system as follows: \[ \left\{ \begin{array}{lr} dU^{\mu,\varepsilon} =\frac{1}{\varepsilon} A U^{\mu,\varepsilon} dt+ \frac{1}{\varepsilon} f(U^{\mu,\varepsilon}, V^{\mu,\varepsilon} )dt+ +\frac{1}{\sqrt{\varepsilon}}q_1z_1^{\mu}dt,&\quad \text{in}\, H_1, \\ d V^{\mu,\varepsilon}=B V^{\mu,\varepsilon} dt+ g(U^{\mu,\varepsilon},V^{\mu,\varepsilon} )dt+ q_2 z_2^{\mu}dt,&\quad \text{in}\, H_2, \end{array}\right.\tag{2} \] where \(z_k^{\mu}\) satisfy \[ dz_k^{\mu}=-\frac{1}{\mu} z_k^{\mu}dt +\frac{1}{\mu}dw_k, k=1,2, \] and \(\mu\) is a small parameter. The authors use the filtering of the smooth reduced system of (2) on the slow manifold to approximate that of (1) via random invariant manifold theory and Wong-Zakai approximation. They establish the existence of the random invariant manifold for the Wong-Zakai system (2), show the solution of the original system can be well approximated by that of a smooth reduced system on the obtained slow manifold, and state that the filtering of the reduced system can well approximate the one of the original system.