Numerical Solution Method for Stochastic Partial Differential Equations Based on the Feynman-Kac Formula and Random Forest Regression

Zichang Liu

doi:10.54097/gd2z9b66

Authors

Zichang Liu

DOI:

https://doi.org/10.54097/gd2z9b66

Keywords:

Feynman-Kac Formula, Numerical Solution of Stochastic Differential Equations, Random Forest, Adaptive Step-Size Algorithm

Abstract

The Feynman-Kac formula establishes a crucial connection between stochastic processes and partial differential equations (PDEs). By expressing the solution of a PDE as the expected value of a stochastic variable, this formula provides an innovative approach to numerically solving PDEs and can be extended to compute the expected solutions of stochastic partial differential equations (SPDEs). However, in practical applications, the numerical stability of this formula is often affected by the range of stochastic variable values. When the stochastic variable takes excessively large or small values, computational errors may increase sharply, potentially rendering the formula ineffective. To address this issue, this paper proposes an adaptive algorithm that dynamically adjusts computational parameters, enhancing the applicability of the Feynman-Kac formula to a broader class of problems. Specifically, the algorithm adaptively optimizes the sampling strategy based on the range of stochastic variable values, ensuring computational stability and convergence. Furthermore, this paper extends the adaptive algorithm to the solution of SPDEs and integrates it with the random forest method to achieve data-driven fitting and approximation of the expected solution.

Downloads

Download data is not yet available.

References

[1] Hawkins K P , Pakniyat A , Tsiotras P .Value function estimators for Feynman–Kac forward–backward SDEs in stochastic optimal control[J].Automatica, 2023, 158(000):-1.DOI:10.1016/j.automatica.2023.111281.

[2] Choi B S , Choi M Y .On Some Results of the Nonuniqueness of Solutions Obtained by the Feynman–Kac Formula[J].Mathematics (2227-7390), 2024, 12(1).DOI:10.3390/math12010129.

[3] LAN Chengming, Xu Zhenqian, Ma Junming, et al. Comparison of Markov Chain Monte Carlo Sampling Algorithms in Subset Simulation [J]. Journal of Civil Engineering, 2022, 55(10):14. DOI: 10.15951/j.tmgcxb.22030200.

[4] Ma Guozhen, Wang Yunjia, Wang Zhumei, Du Wentong. Electric Vehicle Disorderly Charging Load Calculation Model Based on Monte Carlo Random Simulation Algorithm [J]. Journal of Computational Technology and Automation, 2024.

[5] Yang Xu, Tang Mengtong. Existence and Uniqueness of Strong Solutions for Stochastic Partial Differential Equations with Gradients [J]. Journal of Applied Probability and Statistics, 2022, 38(6):931-940.

[6] He Yugong. Convergence and Variance-Finiteness Conditions of Monte Carlo Methods for Solving Elliptic Partial Differential Equations [J]. Journal of Tsinghua University (Natural Science Edition), 1964(02):109-128. DOI: CNKI: SUN: QHXB.0.1964-02-006.

[7] Li Yan. Study on Implicit Runge-Kutta Algorithm for Richards Equation [D]. Hunan University, 2022.

[8] Sang Lin, Zhang Cheng. Adaptive Interpolation Wavelet Method for Numerical Solutions of Partial Differential Equations [J]. Journal of Biomathematics, 2001, 016(002):145-150.

[9] Chen Jingfei. Machine Learning Methods for Solving Partial Differential Equations: From Network Structures to Optimization Algorithms [D]. Soochow University, 2023.

[10] Zhu Quanhui. Using Machine Learning Methods to Solve Fractional Order Partial Differential Equations and Discontinuous Problems [D]. Harbin Institute of Technology, 2020.

[11] Peng Pai, Feng Xinlong. Solving Steady-State Partial Differential Equations Using Deep Learning Methods Based on Two-Level Grids [J]. Journal of Xinjiang University: Natural Science Edition (Chinese and English), 2022, 39(4):9.

[12] Fang Kuangnan, Wu Jianbin, Zhu Jianping, et al. A Review of Random Forest Methods [J]. Statistics and Information Forum, 2011, 26(3):7. DOI: 10.3969/j.issn.1007-3116.2011.03.006.