Abstract:Partial differential equations are discretized using the mortar finite element method, where the mortar space contains piecewise quadratic and -cubic functions. We prove the wellposedness of the resulting saddle point system under a theoretical framework by Braess et al. [SIAM J. Numer. Analy. (1999)]. For solving this saddle point system, we introduce two domain decomposition (DD) algorithms. Under the assumption that every local problem approximates the global problem well in the interior of its own partition, the required communication complexity is proportional to the size of the interface. With a globally conforming fine mesh, a DD algorithm by Bank et al. [SIAM Rev. (2003), SIAM J. Sci. Comput. (2004)] is used, in which both the interface solution and the interface residual of each local problem are sent to the neighboring processors. To accommodate globally nonconforming fine meshes, we propose a new algorithm where only the interface solution is communicated. Consequently, the communication complexity is reduced by 1/2. The convergence property of these DD algorithms are studied on one-dimensional two-grid meshes. If the interior residual of each local problem is also communicated, the DD algorithms are shown to converge in one iteration. Otherwise, these algorithms are shown to converge in two iterations. Numerical results are presented, on solving the Poisson equation and a convection diffusion equation with a highly adaptive mesh underlying the finite element discretization. The scalability and parallel efficiency of our DD algorithms are studied. As the number of processors increases, linear and logarithmic speed-up of the solving time is observed for the convection-diffusion equation and the Poisson equation respectively. Furthermore, several parallel algebraic multilevel algorithms are introduced. Three two-level and one three-level algorithms are proposed and implemented, together with an overlapping Schwartz method, which uses a three-level algorithm followed by a two-level one at each iteration. The scalability results indicate that these algorithms are effective in reducing the communication complexity during the solving stage. In summary, our algorithms improve the coordination among processors during the solving stage and may provide a powerful tool for scientific computing.

Annealing-based approach to solving partial differential equations

Engineering Analysis Error Estimation when Removing Finite-Sized Features in Nonlinear Elliptic Problems.

Optimization of Approximate Solution of Integral Equations of Several Variables

Self-heating solving strategy for process dynamic optimization problems

Partial differential equation solver based on optimization methods

Variational quantum algorithm for generalized eigenvalue problems and its application to the finite element method

Initial-boundary value problems for discrete evolution equations: discrete linear Schrodinger and integrable discrete nonlinear Schrodinger equations

Scalable parallel multilevel algorithms for solving partial differential equations

Approximating partial differential equations without boundary conditions

An approximate Itô-SDE based simulated annealing algorithm for multivariate design optimization problems

Discovering Physics-Informed Neural Networks Model for Solving Partial Differential Equations through Evolutionary Computation

An Adaptive Inverse Iteration for Maxwell Eigenvalue Problem Based on Edge Elements.

An efficient preconditioner for evolutionary partial differential equations with $θ$-method in time discretization

Improved structural methods for nonlinear differential-algebraic equations via combinatorial relaxation

An efficient ODE-solving method based on heuristic and statistical computations: αII-(2 + 3)P method

Operational Solution of Non-Integer Ordinary and Evolution-Type Partial Differential Equations

An Overview on Machine Learning Methods for Partial Differential Equations: from Physics Informed Neural Networks to Deep Operator Learning

Adaptive Two-Layer ReLU Neural Network: II. Ritz Approximation to Elliptic PDEs

Deep-learning based discovery of partial differential equations in integral form from sparse and noisy data

Variational formulation based on duality to solve partial differential equations: Use of B-splines and machine learning approximants

Advances in Solution of Classical Generalized Eigenvalue Problem