Higher Order A-Stable Schemes for the Wave Equation Using a Successive Convolution Approach

Abstract

In several recent works, we developed a new second order, A-stable approach to wave propagation problems based on the method of lines transpose (MOL T ) formulation combined with alternating direction implicit (ADI) schemes. Because our method is based on an integral solution of the ADI splitting of the MOL T formulation, we are able to easily embed non-Cartesian boundaries and include point sources with exact spatial resolution. Further, we developed an efficient O(N ) convolution algorithm for rapid evaluation of the solution, which makes our method competitive with explicit finite difference (e.g., finite difference time domain) solvers, in terms of both accuracy and time to solution, even for Courant numbers slightly larger than 1. We have demonstrated the utility of this method by applying it to a range of problems with complex geometry, including cavities with cusps. In this work, we present several important modifications to our recently developed wave solver. We obtain a family of wave solvers which are unconditionally stable, accurate of order 2P , and require O(P d N ) operations per time step, where N is the number of spatial points and d the number of spatial dimensions. We obtain these schemes by including higher derivatives of the solution, rather than increasing the number of time levels. The novel aspect of our approach is that the higher derivatives are constructed using successive applications of the convolution operator. We develop these schemes in one spatial dimension, and then extend the results to higher dimensions, by reformulating the ADI scheme to include recursive convolution. Thus, we retain a fast, unconditionally stable scheme, which does not suffer from the large dispersion errors characteristic to the ADI method. We demonstrate the utility of the method by applying it to a host of wave propagation problems. This method holds great promise for developing higher order, parallelizable algorithms for solving hyperbolic PDEs and can also be extended to parabolic PDEs. Introduction.In recent works [2, 3], the method of lines transpose (MOL T ) has been utilized to solve the wave equation, resulting in a second order accurate, A-stable numerical scheme. The solution is constructed using a boundary-corrected integral equation, which is derived in a semidiscrete setting. Thus, the solution at time t n+1 is found by convolving the solution at previous time levels against the semidiscrete Green's function. Upon spatial discretization, traditional convolution requires O(N 2 ) operations for a total of N spatial points per time step. However, we have developed a fast convolution algorithm for the one-dimensional (1d) problem, which reduces the computational cost to O(N ) operations [2]. This efficiency was additionally extended to the wave equation in higher dimensions by applying alternate direction implicit (ADI) splitting to the semidiscrete elliptic differential operator.

Method of Lines Transpose: Energy Gradient Flows Using Direct Operator Inversion for Phase-Field Models

“…It remains to discretize the convolution integral (18), and obtain a fully discrete algorithm. In our previous works [4,7,5,6], we accomplish this with fast convolution; for convenience, we will summarize the fast algorithm here. First, the particular solution (19) is split into…”

A WENO-based Method of Lines Transpose approach for Vlasov simulations

“…The PP-limiter can keep the solutions non-negative without loss of accuracy. While for the discontinuous case (Figure 4.1), we consider the 2π-periodic boundary condition and the following Dirichlet boundary condition u(−π, t) = 1, t ∈ [ 3 4 π, 5 4 π], 0, t ∈ (0, 3 4 π) ∪ ( 5 4 π, 2π).…”

“…In [8], the computational cost of the scheme is reduced from O(N 2 ) to O(N ), where N is the number of discrete mesh points, by taking advantage of an important property of the analytical solution of the wave equations. More recently, a successive convolution technique [5] is developed to enhance the temporal accuracy to arbitrary high order. A notable advantage of such a scheme is that, even though it is implicit in time, we do not need to explicitly solve a linear system, while the BVP is inverted analytically in an integral formulation.…”

An Asymptotic Preserving Maxwell Solver Resulting in the Darwin Limit of Electrodynamics

“…The resulting method is A-stable, easy to implement and the computation complexity can be reduced to O(N ) by utilizing the analytical properties of the one-dimensional Green's function, see [8]. Arbitrary temporal accuracy is attained by successive convolutions in [6]. However, there are still several challenges for the extension of this method to Maxwell's equations.…”