Mean square exponential stability of numerical methods for stochastic differential delay equations

Guangqiang Lan,Qi Liu

2023-06-21

Abstract:Mean square exponential stability of $\theta$-EM and modified truncated Euler-Maruyama (MTEM) methods for stochastic differential delay equations (SDDEs) are investigated in this paper. We present new criterion of mean square exponential stability of the $\theta$-EM and MTEM methods for SDDEs, which are different from most existing results under Khasminskii-type conditions. Two examples are provided to support our conclusions.

Numerical Analysis,Probability

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the mean - square exponential stability of numerical methods for stochastic differential - delay equations (SDDEs) under specific conditions. Specifically, the paper mainly focuses on the mean - square exponential stability of the θ - Euler - Maruyama (θ - EM) method and the modified truncated Euler - Maruyama (MTEM) method when dealing with SDDEs. The author proposes a new criterion to evaluate the mean - square exponential stability of these methods under Khasminskii - type conditions and verifies their conclusions through two examples.

### Background and Problem Description

As mentioned in the paper, for nonlinear stochastic differential - delay equations (SDDEs), they have the following form:
\[ dx(t) = f(x(t), x(t-\tau(t))) dt + g(x(t), x(t-\tau(t))) dB(t), \quad t \geq 0 \]
where \( f: \mathbb{R}^d \times \mathbb{R}^d \to \mathbb{R}^d \), \( g: \mathbb{R}^d \times \mathbb{R}^d \to \mathbb{R}^{d \times m} \), \( \tau(t): \mathbb{R}^+ \to \mathbb{R}^+ \) are Borel measurable functions, and \( 0 < \tau(t) \leq \tau \).

### Existing Research

Most of the existing research relies on Khasminskii - type conditions to ensure the exponential stability of numerical methods:
\[ 2\langle x, f(x, y) \rangle + \| g(x, y) \|^2 \leq -C_1 |x|^2 + C_2 |y|^2 \]
where \( C_1 > C_2 > 0 \).

### New Conditions

This paper proposes a new condition:
\[ 2x_i f_i(x, y) + \sum_{l = 1}^m (g_{il}(x, y))^2 \leq \sum_{j = 1}^d a_{ij} x_j^2 + \sum_{j = 1}^d b_{ij} y_j^2, \quad i \in d \]
where \( a_{ij} \) and \( b_{ij} \) satisfy certain appropriate conditions. This condition is different from the traditional Khasminskii - type condition and can be applied to a wider range of situations.

### Main Contributions

1. **New Criterion**: A new criterion for mean - square exponential stability is proposed, which is applicable to the θ - EM method and the MTEM method.
2. **Theoretical Proof**: The mean - square exponential stability of the θ - EM method and the MTEM method under the new condition is proved.
3. **Numerical Verification**: The correctness of the theoretical results is verified through two specific examples.

### Conclusion

By introducing new conditions, the paper extends the stability analysis of numerical methods for SDDEs in the existing literature. This provides new tools and methods for dealing with more complex stochastic differential - delay equations.