A Parareal exponential integrator finite element method for linear parabolic equations
Jianguo Huang, Yuejin Xu · 2024
Read A Parareal exponential integrator finite element method for linear parabolic equations by Jianguo Huang, Yuejin Xu (2024) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2412-01138.txt.
Introduction
Jianguo Huang and Yuejin Xu present a faster solver for a class of heat-like equations, in a 2024 numerical methods paper. The equations, called linear parabolic equations, model spread driven by gradients over time. Their solver works on box-shaped domains. Space is handled with a grid method that uses simple block basis shapes. Time is handled with a Runge-Kutta scheme wrapped in a parallel-in-time frame, so many time steps can be worked at once.
The team proves error bounds for the full scheme under mild smoothness rules. A fast solver path comes from a shared matrix split tied to a fast Fourier method. Tests in several dimensions match the theory and show strong speed. The authors offer the method for large spread problems where serial time stepping is the bottleneck.
Catalog record
| CREATOR | Jianguo Huang, Yuejin Xu |
|---|---|
| YEAR | 2024 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 58 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2412-01138
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
A Parareal exponential integrator finite element method for linear parabolic equations Jianguo Huang, Yuejin Xu Abstract In this paper, for solving a class of linear parabolic equations in rectangular domains, we have proposed an efficient Parareal exponential integrator finite element method. The proposed method first uses the finite element approximation with continuous multilinear rectangular basis function for spatial discretization, and then takes the Runge-Kutta approach accompanied with Parareal framework for time integration of the resulting semi-discrete system to produce parallel-in-time numerical solution. Under certain regularity assumptions, fully-discrete error estimates in L 2 L^{2} -norm are derived for the proposed schemes with random interpolation nodes. Moreover, a fast solver can be provided based on tensor product spectral decomposition and fast Fourier transform (FFT), since the mass and coefficient matrices of the proposed method can be simultaneously diagonalized with an orthogonal matrix. A series of numerical experiments in various dimensions are also presented to validate the theoretical results and demonstrate the excellent performance of the proposed method. 1 Introduction In this paper, we are intended to study a fast solution of the following linear parabolic equation: \left\{\begin{split}&u_{t}=\mathcal{D}\Delta u+f(t),\quad\bm{x}\in\Omega,\quad t_{0}\leq t\leq t_{0}+T,\\ &u(t_{0}):=u(t_{0},\bm{x})=u_{0},\quad\bm{x}\in\Omega,\end{split}\right. (1.1) where \Omega is an open rectangular domain in \mathbb{R}^{d} ( d\geq 1 ), T>0 is the duration time, \mathcal{D}>0 is the diffusion coefficient, and f(t) is the reaction term for the above equation, which has been widely used in various scientific and engineering models. In this paper, we mainly discuss the problem above, which is associated with Dirichlet boundary conditions. The theoretical conclusion can also be extended to equations with periodic boundary conditions after minor modification. In the past few decades, many researchers are devoted to developing fast solver for parabolic equation like ( 1.1 ). Various methods have been applied to solving the numerical result for spatial and temporal discretization, respectively. Some common methods such as finite element method, finite difference method, finite volume method and spectral method are used for spatial discretization, while temporal discretization is mainly implemented by implicit-explicit method [ 1 ] , fully-implicit scheme [ 9 ] , split step method [ 4 , 30 ] , integrating factor method (IF) [ 25 ] , integrating factor Runge-Kutta method (IFRK) [ 21 , 26 ] , sliders method [ 5 ] , exponential time differencing method (ETD) [ 16 , 3 ] , invariant energy quadratization method (IEQ) [ 38 ] , scalar auxiliary variable method (SAV) [ 31 ] and so on. ETD method has received much attention in the past two decades since it approximates the nonlinear part of function f using polynomial interpolations to obtain higher accuracy for highly oscillation problems [ 39 ] . Due to the great efficiency and stability of ETD method, the analysis of convergence [ 15 , 14 , 6 , 24 , 28 ] and modification [ 24 , 17 , 35 , 23 , 39 , 18 , 19 ] has been widely proposed. For temporal discretization, Parareal algorithm is a crucial method firstly proposed by Lions, Maday, Y. and Turinici in [ 27 ] , which solves equations in the parallel-in-time pattern to improve the calculating efficiency. Motivated by the two-level method, the whole time interval is divided into coarse and fine subintervals, attaching with coarse propagator \mathcal{G} and fine propagator \mathcal{F} respectively. The great advantage of Parareal method is that the numerical results obtained on coarse grids can reach to the accuracy of applying fine propagator squentially on fine grids with significant speedup. The convergence property of Parareal is discussed in [ 12 , 2 ] for scalar ODEs, showing that the algorithm is superlinear convergent on bounded time intervals, while that is linear convergent on unbounded intervals based on the convergence factor and the property of Banach spaces respectively. As for systems with complex eigenvalues, the convergence of Parareal and Parareal-Euler method are proposed in [ 32 , 29 ] by deriving the corresponding stability function or convergence factor. What’s more, the convergence property is discussed for some specific combinations of coarse propagator \mathcal{G} and fine propagator \mathcal{F} [ 37 , 34 , 36 ] , which are mainly derived by constructing the convergence factors and plotting the stability regions. In recent years, with the development of Parareal method, many scholars have proposed some modified Parareal algorithm mainly in two aspects, decreasing the iteration times to speedup the convergence rate and improving the parallel efficiency for each iteration. Many researchers have tried to decrease the iteration times from different aspects, for instance, implicit time-integration schemes and Newton-like correction are used to accelerate the simulations in [ 8 ] . Some modified Parareal methods are proposed in [ 13 ] in the framework of two-level scheme, which regard the operators on coarse grid and fine grid as predictor and corrector respectively and can ensure to converge within a fixed number of iterations. Moreover, based on the relationship among Parareal, space-time multigrid and multiple shooting methods, some variants of Parareal are derived in [ 12 ] to accelerate the convergence. As for improving the efficiency of parallelism, multilevel extension can alleviate some of the load-balance issues of the two-level method [ 13 ] . A nonintrusive, optimal-scaling time-parallel method based on multigrid reduction (MGR) is also designed to obtain greater parallelism [ 7 ] . To reduce the communication or synchronization among the processors, Paraexp method is proposed in [ 11 ] to decouple the original problem to inhomogeneous and homogeneous subproblems, by untilizing an overlapping time-domain decomposition. Diagonalization is a crucial technique to reduce the computational cost in each iteration, which has been applied to solve the advection-diffusion equation with periodic boundary conditions in [ 10 ] , to develop the parallel coarse grid correction (CGC), and so on. As far as we know, for most numerical methods of PDEs, the spatial variables are often discretized by finite difference methods [ 23 , 39 , 19 , 18 ] and the corresponding convergence analysis highly depends on discrete maximum principle [ 22 ] and references therein, which does not always hold for classical finite element or pseudo-spectral discretization. Moreover, most of Parareal convergence analysis is based on deriving the convergent factor, thus, the stability regions obtained are always approximate but not accurate when the convergent factor is complicated. In order to overcome the above obstacles, in this paper, for linear parabolic equation ( 1.1 ) in rectangular domains, we propose an efficient Parareal exponential integrator finite element method (abbreviated as PEIFE-linear) to solve large-scale problems in parallel-in-time pattern. Similar to [ 20 ] , we first derive a semi-discrete scheme with continuous linear elements on a uniform rectangular partition and then obtain a fully-discrete scheme with explicit exponential Runge-Kutta scheme. Later, the whole time interval is divided into several subintervals and is applied with Parareal method in the temporal direction. Global error estimates measured in L^{2} -norm can be derived for PEIFE-linear method with random RK stages for the linear problem with Dirichlet boundary condition. It’s worthy to note that the error estimate of PEIFE-linear method can be derived explicitly, by following the similar arguments of [ 16 ] and [ 12 , 2 ] . Furthermore, motivated by [ 23 ]