The Convergence of Spectral and Finite Difference Methods for Initial-Boundary Value Problems

  • Authors:
  • Natasha Flyer;Paul N. Swarztrauber

  • Affiliations:
  • -;-

  • Venue:
  • SIAM Journal on Scientific Computing
  • Year:
  • 2001

Quantified Score

Hi-index 0.02

Visualization

Abstract

The general theory of compatibility conditions for the differentiability of solutions to initial-boundary value problems is well known. This paper introduces the application of that theory to numerical solutions of partial differential equations and its ramifications on the performance of high-order methods. Explicit application of boundary conditions (BCs) that are independent of the initial condition (IC) results in the compatibility conditions not being satisfied. Since this is the case in most science and engineering applications, it is shown that not only does the error in a spectral method, as measured in the maximum norm, converge algebraically, but the accuracy of finite differences is also reduced. For the heat equation with a parabolic IC and Dirichlet BCs, we prove that the Fourier method converges quadratically in the neighborhood of t=0 and the boundaries and quartically for large t when the first-order compatibility conditions are violated. For the same problem, the Chebyshev method initially yields quartic convergence and exponential convergence for t0. In contrast, the wave equation subject to the same conditions results in inferior convergence rates with all spectral methods yielding quadratic convergence for all t. These results naturally direct attention to finite difference methods that are also algebraically convergent. In the case of the wave equation, we prove that a second-order finite difference method is reduced to 4/3-order convergence and numerically show that a fourth-order finite difference scheme is apparently reduced to 3/2-order. Finally, for the wave equation subject to general ICs and zero BCs, we give a conjecture on the error for a second-order finite difference scheme, showing that an O(N-2log N) convergence is possible.