Concepedia

ON THE COMPUTATIONAL STABILITY OF NUMERICAL SOLUTIONS OF TIME-DEPENDENT NON-LINEAR GEOPHYSICAL FLUID DYNAMICS PROBLEMS

Douglas K. Lilly

Monthly Weather Review · 1965 · 353 citations · 12 references

Concepts

TL;DR

The satisfactory numerical solution of fluid‑dynamics equations for atmospheric and oceanic problems requires high computational stability and accurate conservation of statistical moments. The study aims to describe methods that satisfy these stability and conservation requirements for low‑Mach fluid‑dynamics systems. The authors analyze and test Arakawa’s spatial differencing schemes and various time‑stepping methods—such as midpoint, Adams‑Bashforth, Matsuno, and Lax‑Wendroff—using spectral analysis and finite‑spectral equations to assess conservation, stability, and accuracy in low‑Mach, non‑divergent fluid dynamics. Linear and quadratic conserving schemes eliminate the nonlinear instability identified by Phillips, though aliasing errors remain; the midpoint rule is unstable, whereas Matsuno and second‑order Adams‑Bashforth methods are stable and accurate, while Matsuno, Lax‑Wendroff are less accurate.

Abstract

The satisfactory numerical solution of the equations of fluid dynamics applicable to atmospheric and oceanic problems characteristically requires a high degree of computational stability and accurate conservation of certain statistical moments. Methods for satisfying these requirements are described for various systems of equations typical of low. Mach number fluid dynamics systems, and are investigated in detail as applied to the two-dimensional, inertial-plane equation for conservation of vorticity in a frictionless non-divergent fluid. The conservation and stability properties of the spatial differencing methods devised by A. Arakawa are investigated by means of spectral analysis of the stream function into finite Fourier modes. Any of two classes of linear and quadratie conserving schemes are shown to eliminate the non-linear instability discussed by Phillips, although the “aliasing” error remains. Stability related to the time derivative term is investigated through analytic and numerical solutions of a limited-component system of finite spectral equations, equivalent to one of the quadratic conserving difference schemes, and a number of first and second order representations of the time derivative term are tested separately. The commonly used midpoint rule (“leapfrog”) method is shown to be unstable in some cases. Of the stable methods, that devised by Matsuno, and the second order Adams-Bashforth method exhibit satisfactory accuracy, while those due to Matsuno, and Lax and Wendroff are much less accurate. A systematic derivation of the Arakawa difference schemes is contained in an appendix, which shows their unique satisfaction of certain prescribed accuracy and conservation properties.

References

12

Deterministic Nonperiodic Flow

Edward N. Lorenz · Journal of the Atmospheric Sciences · 1963

+13

19K citations

Systems of conservation laws

Peter D. Lax, Burton Wendroff · Communications on Pure and Applied Mathematics · 1960

+3

2.6K citations

Weather Prediction by Numerical Process

Norman A. Phillips, Lewis Fry Richardson · Mathematics of Computation · 1966

+15

1.2K citations

On the solution of nonlinear hyperbolic differential equations by finite differences

Richard Courant, Eugene Isaacson, Mina Rees · Communications on Pure and Applied Mathematics · 1952

+4

1K citations