Lecture 22.ppt

Lecture 22 MA471 Fall 2003 Advection Equation Recall the 2D advection equation: We will use a Runge-Kutta time integrator and spectral representation in space. Periodic Data Let’s assume we are given N values of a function f at N data points on the unit interval. For instance N=8 on a unit interval: Discrete Fourier Transform We can seek a trigonometric interpolation of a function f as a linear combination of N (even) trigonometric functions: Such that: Transform The interpolation formula defines a linear system for the unknown fhat coefficients: Code for the DFT Code for Inverse DFT Fast Fourier Transform See handout Spectral Derivative We can differentiate the interpolant by: Detail We note that the derivative of the k=(N/2) mode is technically: However, as we show on the next slide – this mode has turning points at all the data points. Real Component of N/2 Mode Modified Derivative Formula So we can create an alternative symmetric derivative formula: Spectral Differentiation Scheme Use an fft to compute: Differentiate in spectral space. i.e. compute: cont 3) Then: 4) Summary: a) fft transform data to compute coefficients b) scale Fourier coefficients c) inverse fft (ifft) scaled coefficients Final Twist Matlab stores the coefficients from the fast Fourier transform in a slightly odd order: The derivative matrix will now be a matrix with diagonal entries: Spectral Differentiation Code Two-Dimensional Fourier Transform We can now construct a Fourier expansion in two variables for a function of two variables.. The 2D inverse discrete transform and discrete transform are: Advection Equation Recall the 2D advection equation: We will use a Runge-Kutta time integrator and spectral representation in space. Runge-Kutta Time Integrator We will now discuss a particularly simple Runge-Kutta time integrator introduced by Jameson-Schmidt-Turkel The idea is each time step is divided into s substeps, which taken together approximate the update t