Finite-difference Errors - Part 2: Amplitude Error Analysis of Linear Eqs; the von Neuman method (continued_
Under construction (this web page, and most other web pages for this course).
Instructor: Roland Stull
Learning Goals: By the end of this module, you will be able to ...
- use stability analyses to recommend which finite-different schemes are best.
- discuss the advantages and disadvantages of the finite difference schemes used in the WRF model.
- compare the finite difference schemes used in WRF-ARW vs. WRF-NMM.
- identify the appropriate stability criteria for the diffusion eq.
- compare explicit vs. implicit time-differencing schemes.
Readings BEFORE class (same as previous readings):
- Stull p 759-761, and Warner p72-118.
- Press et al "Numerical Recipes 3rd Ed", Chapter 20 PDEs, start on p1032. List of ...
Homework AFTER class: Homework 5
- Do a von Neumann linear stability analysis using the eigenmode method
as we used in class (Press et al, Numerical Recipes), to answer the
following questions regarding the 3-point-centered-in-time 5-point-centered-in-space approximation to the linear advection eq.:
[T(j,n+1) - T(j,n-1)] / (2 ∆t) = - [Uo / (12∆x)] [T(j-2,n) - 8T(j-1,n) + 8T(j+1,n) - T(j+2,n)]
- What criterion is required to ensure numerical stability. (Hint, the answer of CR ≤ 0.729 is in Warner p69 top line.) But you need to show all the steps to get this answer.
- What is the amount of damping (amplitude reduction) as a function of wavenumber k (or as a function of m, for m∆x wavelengths)?
- For the linear advection equation
that is numerically approximated by a 3rd-order Runge-Kutta
(RK3)-in-time with 2nd-order-centered-in-space scheme discussed in
class:
- Check the algebra of my in-class derivation of the single-equation version of RK3.
- Do a von Neuman linear
stability analysis of this single RK3 equation using the eigenmode
method of Press et al (Numerical Recipes), and find the equation for
the magnitude of the amplitude |A(k)| , or for |A(k)|2 if easier.
- For the worst case wavelength, what is the critical value of CR (Courant number) is needed to maintain stability, and what wavelength is the worst-case wavelength?
- How does the Courant
stability criterion for RK3 with 2nd-order-Spatial Differencing (that
you just derived) compare with the higher-order stability criteria given in Table
3.1 (p28 of the WRF-ARW4.3 tech note; see link to it in the Resources web page)?
- Plot |A(k)| vs. CR (Courant number in the range of 0 to 3) for various wavelengths: L = 2∆x, 2.5∆x, 3∆x, 4∆x, 5∆x, 10∆x, 20∆x.
Topics (same as last week)
A. List of Error Types.
B. Truncation Error.
- Centered difference as an example
C. Amplitude Errors for Linear eqs.
- von Neumann stability analysis method
- The linear advection eq.
- Stability analysis of different explicit finite-diff approximations (from the list below) for the linear advection eq.
- Diffusion eq. -- stability analysis of different finite-diff approximations.
a) forward in time;
b) centered in time. approximations.
- Implicit finite diff schemes-- stability analyses:
a) Advection example using Crank-Nicholson method.
b) Diffusion example.
c) Diffusion using Crank-Nicholson
List of some of the Explicit Finite Difference Approximations
- Euler-forward in time, centered in space.
- Forward in time, backward in space.
- Forward in time from spatial average (Lax method), centered in space.
- Centered in time, 3-point centered in space (Leapfrog method)
- Centered in time, 5-point centered in space
- Two-step Lax-Wendroff
- Lax single-step in 2-D
- 3rd-order Runge-Kutta in time, 2nd-order centered in space (WRF-ARW)
- Adams-Bashforth (2nd order in time)
- Off-centered Adams-Bashforth (WRF-NMM)