B. Leimkuhler,A. Sharma,M.V. Tretyakov
Abstract:A simple-to-implement weak-sense numerical method to approximate reflected stochastic differential equations (RSDEs) is proposed and analysed. It is proved that the method has the first order of weak convergence. Together with the Monte Carlo technique, it can be used to numerically solve linear parabolic and elliptic PDEs with Robin boundary condition. One of the key results of this paper is the use of the proposed method for computing ergodic limits, i.e. expectations with respect to the invariant law of RSDEs, both inside a domain in $\mathbb{R}^{d}$ and on its boundary. This allows to efficiently sample from distributions with compact support. Both time-averaging and ensemble-averaging estimators are considered and analysed. A number of extensions are considered including a second-order weak approximation, the case of arbitrary oblique direction of reflection, and a new adaptive weak scheme to solve a Poisson PDE with Neumann boundary condition. The presented theoretical results are supported by several numerical experiments.
What problem does this paper attempt to address?
The main problems that this paper attempts to solve are how to numerically solve stochastic differential equations with reflecting boundaries (RSDEs) and their associated linear parabolic partial differential equations (PDEs), especially equations with Robin boundary conditions. Specifically, the article proposes a simple and easy - to - implement numerical method in the weak sense to approximate the reflecting stochastic differential equations and analyzes the convergence of this method.
### Main Problems and Objectives
1. **How to numerically solve stochastic differential equations with reflecting boundaries (RSDEs) and their associated linear parabolic PDEs?**
- A simple numerical method in the weak sense is proposed to approximate the reflecting stochastic differential equations (RSDEs), and it is proved that this method has first - order weak convergence.
- Through Monte Carlo techniques, it can be used to numerically solve linear parabolic and elliptic PDEs with Robin boundary conditions.
2. **How to calculate the ergodic limits in the domain \( G \) and on the boundary \( \partial G \)?**
- Time - average and ensemble - average estimators are introduced to calculate the expected values under the invariant measure of the reflecting diffusion process \( X(t) \).
- A method for calculating the expected values of integrals with respect to local time is proposed.
3. **How to sample from distributions with compact support?**
- The reflecting Brownian dynamics is used to efficiently sample from target distributions with compact support, which has wide applications in machine learning and molecular dynamics.
4. **How to numerically solve linear elliptic PDEs with Robin boundary conditions?**
- The probabilistic representation method is used to transform the elliptic Robin problem into a functional integral of \( X(t) \) on \([0, \infty)\).
- The weak method is applied and the elliptic problem is analyzed. In particular, the Poisson equation with Neumann boundary conditions is dealt with, and a new adaptive time - step algorithm is proposed.
### Specific Application Scenarios
- **Models in Physics, Biology, Engineering and Finance**: Many models from these fields can be described by stochastic differential equations with reflecting boundaries, such as inventory management, quality control, heavy - traffic behavior in queuing systems, chemical reaction kinetics, etc.
- **Feynman - Kac Formula**: It is used to give the probabilistic representation of solutions to parabolic and elliptic PDEs with Neumann/Robin boundary conditions.
- **High - Dimensional Distribution Sampling**: The reflecting Brownian dynamics is used to sample from high - dimensional compact - support distributions, which is applicable in fields such as machine learning and molecular dynamics.
### Method Characteristics
- **Simple and Easy to Implement**: The proposed weak approximation method does not require orthogonal transformation of the diffusion matrix or change of the local coordinate system, so it is easy to implement.
- **First - Order Weak Convergence**: It is proved that this method has first - order weak convergence and can maintain this convergence in time - average and ensemble - average estimations.
- **Extended Application**: The method can be extended to any obliquely inward reflecting direction, and a new procedure is introduced to obtain a second - order method.
In conclusion, this paper aims to provide a simple and effective numerical method to approximate reflecting stochastic differential equations and apply it to various practical problems, especially in solving PDEs with Robin boundary conditions and sampling from distributions with compact support.