Multi-level Monte Carlo methods with the Truncated Euler-Maruyama Scheme for Stochastic Differential Equations

Qian Guo,Wei Liu,Xuerong Mao,Weijun Zhan

DOI: https://doi.org/10.48550/arXiv.1611.07833

2017-02-21

Abstract:In this paper, the truncated Euler-Maruyama (EM) method is employed together with the Multi-level Monte Carlo (MLMC) method to approximate the expectations of functions of solutions to stochastic differential equations (SDEs). The convergence rate and the computational cost of the approximations using the truncated EM method with the MLMC method are proved when the coefficients of SDEs fulfill the local Lipschitz and Khasminskii-type conditions. Numerical examples are given to demonstrate the theoretical results.

Numerical Analysis

What problem does this paper attempt to address?

The main problem that this paper attempts to solve is to combine the multi - level Monte Carlo (MLMC) method with the truncated Euler - Maruyama method to approximate the expected values of certain functions of the solutions of stochastic differential equations (SDEs). Specifically, when the coefficients of SDEs satisfy the local Lipschitz condition and the Khasminskii - type condition, the convergence rate and computational cost of this method combination are studied.

### Background of the Paper

Stochastic differential equations are widely used in many fields to capture the uncertain phenomena of systems. However, explicit solutions are difficult to find, so numerical approximation has become an important means of applying SDEs. For non - globally Lipschitz continuous drift and diffusion terms, the strong convergence of explicit numerical methods is a research hotspot. The multi - level Monte Carlo method has become an effective tool by improving the convergence rate and reducing the computational cost of estimating the expected value.

### Research Objectives

1. **Combining MLMC and the Truncated Euler - Maruyama Method**: Study the convergence rate and computational cost of this combined method when approximating the expected values of certain nonlinear functions of the solutions of SDEs.
2. **Theoretical Analysis**: Prove the convergence rate and computational complexity of this method when the coefficients of SDEs satisfy the local Lipschitz condition and the Khasminskii - type condition.
3. **Numerical Experiments**: Provide numerical examples to verify the theoretical results.

### Main Contributions

- **Method Innovation**: Combine the MLMC method with the truncated Euler - Maruyama method and propose a new numerical method.
- **Theoretical Proof**: Prove the convergence rate and computational complexity of this method, especially for non - globally Lipschitz continuous SDEs.
- **Numerical Verification**: Verify the correctness of the theoretical analysis through specific numerical experiments.

### Specific Contents

1. **Mathematical Preliminaries**:
- Define the basic form and related symbols of SDEs.
- Introduce the basic concepts of the truncated Euler - Maruyama method and the multi - level Monte Carlo method.
- Give the assumed conditions of the coefficients of SDEs, such as the local Lipschitz condition and the Khasminskii - type condition.

2. **Main Results**:
- **Generalized Theorem**: Generalize Giles' theorem of the multi - level Monte Carlo method and relax the requirements for parameters.
- **Specific Theorem**: Prove the convergence rate and computational complexity of the multi - level Monte Carlo method combined with the truncated Euler - Maruyama method.

3. **Numerical Simulation**:
- Through an example of a nonlinear scalar SDE, show the divergence of the classical Euler method and the multi - level Monte Carlo method.
- Use the truncated Euler - Maruyama method combined with the multi - level Monte Carlo method to show its convergence trend and the improvement of computational efficiency.

### Conclusion

This paper successfully combines the multi - level Monte Carlo method and the truncated Euler - Maruyama method, providing an effective method for solving the problem of estimating the expected value of SDEs with non - globally Lipschitz continuous coefficients. Both theoretical analysis and numerical experiments verify the effectiveness and efficiency of this method.