Abstract:Coupled multi-physics problems are encountered in countless applications and pose significant numerical challenges. Although monolithic approaches offer possibly the best solution strategy, they often require ad-hoc preconditioners and numerical implementations. Sequential (also known as splitted, partitioned or segregated) approaches are iterative methods for solving coupled problems where each equation is solved independently and the coupling is achieved through iterations. These methods offer the possibility to flexibly add or remove equations from a model and to rely on existing black-box solvers for every specific equation. Furthermore, when problems are non-linear, inner iterations need to be performed even in monolithic solvers, therefore making a sequential iterative approach a viable alternative. The cost of running inner iterations to achieve the coupling, however, could easily becomes prohibitive, or, in some cases the iterations might not converge. In this work we present a general formulation of splitting schemes for continuous operators, with arbitrary implicit/explicit splitting, like in standard iterative methods for linear systems. By introducing a generic relaxation operator we find the conditions for the convergence of the iterative schemes. We show how the relaxation operator can be thought as a preconditioner and constructed based on an approximate Schur-complement. We propose a Schur-based Partial Jacobi relaxation operator to stabilise the coupling and show its effectiveness. Although we mainly focus on scalar-scalar linear problems, most results are easily extended to non-linear and higher-dimensional problems. Numerical tests (1D and 2D) for two PDE systems, namely the Dual-Porosity model and a Quad-Laplacian operator, are carried out to confirm the theoretical results.

Convergence Analysis of the Deep Splitting Scheme: the Case of Partial Integro-Differential Equations and the associated FBSDEs with Jumps

Full error analysis of the random deep splitting method for nonlinear parabolic PDEs and PIDEs

Deep learning numerical methods for high-dimensional fully nonlinear PIDEs and coupled FBSDEs with jumps

FBSJNN: A Theoretically Interpretable and Efficiently Deep Learning method for Solving Partial Integro-Differential Equations

Convergence of the Deep BSDE Method for Coupled FBSDEs

Efficient Solution of Backward Jump-Diffusion PIDEs with Splitting and Matrix Exponentials

Convergence of the Deep BSDE Method for FBSDEs with Non-Lipschitz Coefficients

High Order Splitting Methods for SDEs Satisfying a Commutativity Condition

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

Splitting Schemes for Coupled Differential Equations: Block Schur-Based Approaches and Partial Jacobi Approximation

Generalized convergence of the deep BSDE method: a step towards fully-coupled FBSDEs and applications in stochastic control

Convergence of particles and tree based scheme for singular FBSDEs

Deep learning-based numerical methods for high-dimensional parabolic partial differential equations and backward stochastic differential equations

Gradient Convergence of Deep Learning-Based Numerical Methods for BSDEs

Three algorithms for solving high-dimensional fully-coupled FBSDEs through deep learning

Solving High-Dimensional Partial Integral Differential Equations: The Finite Expression Method

An IMEX-RK scheme for capturing similarity solutions in the multidimensional Burgers' equation

A discretization scheme for path-dependent FBSDEs

Convergence Analysis of a Quasi-Monte Carlo-based Deep Learning Algorithm for Solving Partial Differential Equations

Sharp error estimate of variable time-step IMEX BDF2 scheme for parabolic integro-differential equations with initial singularity arising in finance