Abstract:In this paper, we investigate the strong convergence analysis of parareal algorithms for stochastic Maxwell equations with the damping term driven by additive noise. The proposed parareal algorithms proceed as two-level temporal parallelizable integrators with the stochastic exponential integrator as the coarse propagator and both the exact solution integrator and the stochastic exponential integrator as the fine propagator. It is proved that the convergence order of the proposed algorithms linearly depends on the iteration number. Numerical experiments are performed to illustrate the convergence of the parareal algorithms for different choices of the iteration number and the damping coefficient.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is about the strong convergence analysis of stochastic Maxwell equations with damping terms and driven by additive noise. Specifically, the author studied the performance of a class of numerical methods called the Parareal Algorithm when solving such equations. The Parareal Algorithm is a time - parallelized integrator, aiming to improve the efficiency and accuracy of solving partial differential equations (especially those containing stochastic factors).
The Parareal Algorithm proposed in the paper adopts a two - level time - parallelized integrator, where the Coarse G - Propagator uses a stochastic exponential integrator, and the Fine F - Propagator considers two options: the exact - solution integrator and the stochastic exponential integrator respectively. The author proved that the convergence order of the proposed algorithm depends linearly on the number of iterations \(k\), and verified this theoretical result through numerical experiments.
### Main contributions
1. **Strong convergence analysis**: The paper conducted a detailed analysis of the strong convergence of the Parareal Algorithm when solving stochastic Maxwell equations with damping terms and driven by additive noise, and proved that the convergence order is linearly related to the number of iterations \(k\).
2. **Numerical experiments**: Verified the correctness of the theoretical analysis through numerical experiments, showing the influence of different numbers of iterations \(k\) and damping coefficients \(\sigma\) on the convergence of the algorithm.
3. **Influence of noise**: Explored the influence of noise at different scales on the numerical solution, further verifying the stability and robustness of the algorithm.
### Mathematical model
The stochastic Maxwell equations studied in the paper can be expressed as:
\[
\begin{cases}
\epsilon\frac{\partial E(t,x)}{\partial t}=\nabla\times H(t,x)-\sigma E(t,x)-J_e(t,x,E,H)-J_r^e(t,x)\cdot\dot{W},&(t,x)\in(0,T]\times D,\\
\mu\frac{\partial H(t,x)}{\partial t}=-\nabla\times E(t,x)-\sigma H(t,x)-J_m(t,x,E,H)-J_r^m(t,x)\cdot\dot{W},&(t,x)\in(0,T]\times D,\\
E(0,x)=E_0(x),\quad H(0,x)=H_0(x),&x\in D,\\
n\times E = 0,&(t,x)\in(0,T]\times\partial D,
\end{cases}
\]
where \(D\subset\mathbb{R}^3\) is an open, bounded Lipschitz domain, and \(n\) is the unit outer normal vector of its boundary \(\partial D\). \(\epsilon\) and \(\mu\) are the permittivity and permeability respectively, and the damping terms \(\sigma E\) and \(\sigma H\) are used to simulate the attenuation of electromagnetic waves in the medium.
### Parareal Algorithm
The basic steps of the Parareal Algorithm are as follows:
1. **Initialization**: Use the Coarse G - Propagator to calculate the initial value \(u_n^{(0)}\).
2. **Time - parallel computation**: Use the Fine F - Propagator to independently calculate the intermediate value \(\bar{u}_n\) on each sub - interval.
3. **Prediction and correction**: Update the numerical solution \(u_n^{(k + 1)}\) through prediction and correction steps.
### Convergence analysis
The paper gives two main convergence results:
1. When the Fine F - Propagator is selected as the exact - solution integrator, the convergence estimate for a fixed number of iterations \(k\) is:
\[
\sup_{1\le