Abstract:In this article we design a novel quasi-regression Monte Carlo algorithm in order to approximate the solution of discrete time backward stochastic differential equations (BSDEs), and we analyze the convergence of the proposed method. The algorithm also approximates the solution to the related semi-linear parabolic partial differential equation (PDE) obtained through the well-known Feynman-Kac representation. For the sake of enriching the algorithm with high-order convergence a weighted approximation of the solution is computed and appropriate conditions on the parameters of the method are inferred. With the challenge of tackling problems in high dimensions we propose suitable projections of the solution and efficient parallelizations of the algorithm taking advantage of powerful many-core processors such as graphics processing units (GPUs).

What problem does this paper attempt to address?

The main problem that this paper attempts to solve is to design a new quasi - regression Monte Carlo algorithm to approximately solve discrete - time backward stochastic differential equations (BSDEs) and related semi - linear parabolic partial differential equations (PDEs). Specifically, the paper focuses on effectively solving these equations in high - dimensional cases and proposes an algorithm suitable for large - scale parallelization, especially for achieving efficient parallel computing on graphics processing units (GPUs). ### Problem Background 1. **Feynman - Kac Formula**: Since the 1940s, the Feynman - Kac formula has provided a probabilistic interpretation for linear partial differential equations. Therefore, the Monte Carlo method has become an alternative method for solving linear PDEs, especially when the dimension is high. Traditional deterministic numerical algorithms (such as the finite - difference, finite - element or finite - volume method) will encounter the "curse of dimensionality" problem, that is, as the dimension increases, the computational complexity required to reach a given accuracy increases exponentially. However, the convergence speed of the Monte Carlo method is independent of the dimension of the state space, so it has an advantage in high - dimensional problems. 2. **Backward Stochastic Differential Equations (BSDEs)**: In the 1990s, BSDE theory provided a probabilistic interpretation for nonlinear problems. BSDEs are a class of stochastic differential equations with terminal conditions, and their solutions are pairs of processes. Nonlinear BSDEs were first rigorously studied by Pardoux and Peng in 1990. The application of BSDEs in finance began with the research of El Karoui et al. in 1997, who applied them to the European option pricing problem. ### Core Problems of the Paper The paper aims to solve the following problems: - **Numerical Solution of High - Dimensional BSDEs and PDEs**: Traditional methods are difficult to effectively solve BSDEs and PDEs in high - dimensional cases, so new numerical schemes need to be designed. - **Efficient Parallelization**: In order to meet the computational challenges brought by high - dimensional problems, an algorithm suitable for large - scale parallelization is proposed, especially by using the powerful multi - core processor capabilities of GPUs. - **Improving the Convergence Order**: By introducing weighted approximations and appropriate parameter conditions, the convergence order of the algorithm is improved, thereby obtaining more accurate solutions. ### Solutions The paper proposes a new quasi - regression Monte Carlo algorithm, which has the following characteristics: 1. **Quasi - Regression Monte Carlo Scheme**: Utilize the orthogonal properties of basis functions to simplify the calculation of conditional expectations into simple Monte Carlo averages, avoiding the inversion of the linear system required for solving the least - squares problem. 2. **Smooth Basis Functions**: Use smooth and naturally orthogonal basis functions (such as Student - cosine basis) to ensure the global smoothness of the solution, which is especially suitable for solving smooth solutions. 3. **Large - Scale Parallelization**: By parallelizing between different samples and basis function coefficients, the parallel computing power of GPUs is fully utilized, and the computational efficiency is improved. ### Conclusion By designing a new quasi - regression Monte Carlo algorithm, this paper solves the numerical solution problems of high - dimensional BSDEs and PDEs and proposes a scheme suitable for large - scale parallelization, especially the efficient implementation on GPUs. This provides an effective tool for dealing with high - dimensional problems.

Quasi-Regression Monte-Carlo scheme for semi-linear PDEs and BSDEs with large scale parallelization on GPUs

Stratified regression Monte-Carlo scheme for semilinear PDEs and BSDEs with large scale parallelization on GPUs

Deep multi-step mixed algorithm for high dimensional non-linear PDEs and associated BSDEs

A test of backward stochastic differential equations solver for solving semilinear parabolic differential equations in 1D and 2D

A Deep BSDE approximation of nonlinear integro-PDEs with unbounded nonlocal operators

Numerical approximation of ergodic BSDEs using non linear Feynman-Kac formulas

Convergence of the Deep BSDE Method for Coupled FBSDEs

Quasi-Monte Carlo for Bayesian design of experiment problems governed by parametric PDEs

Stochastic methods for solving high-dimensional partial differential equations

The One Step Malliavin scheme: new discretization of BSDEs implemented with deep learning regressions

Local discontinuous Galerkin method for nonlinear BSPDEs of Neumann boundary conditions with deep backward dynamic programming time-marching

Numerical Approximations of Coupled Forward-Backward SPDEs

On Multilevel Picard Numerical Approximations for High-Dimensional Nonlinear Parabolic Partial Differential Equations and High-Dimensional Nonlinear Backward Stochastic Differential Equations

Deep Learning-Based Numerical Methods for High-Dimensional Parabolic Partial Differential Equations and Backward Stochastic Differential Equations

A Monte Carlo method for backward stochastic differential equations with Hermite martingales

Sparse grid approximation of stochastic parabolic PDEs: The Landau--Lifshitz--Gilbert equation

FBSDE based Neural Network Algorithms for High-Dimensional Quasilinear Parabolic PDEs

Gradient Convergence of Deep Learning-Based Numerical Methods for BSDEs

Machine Learning Approximation Algorithms for High-Dimensional Fully Nonlinear Partial Differential Equations and Second-order Backward Stochastic Differential Equations

On full history recursive multilevel Picard approximations and numerical approximations for high-dimensional nonlinear parabolic partial differential equations and high-dimensional nonlinear backward stochastic differential equations

Deep Operator BSDE: a Numerical Scheme to Approximate the Solution Operators