Questions and Answers on Finite Difference Methods
- What is a two-point boundary value problem?
- What is a finite difference formula?
- What are three commonly used FD formulas for approximating
u'(x)? How accurate of those methods?
- What's the best 'h' that gives the best possible accuracy for
FD formulas for approximating u'(x) in terms of the machine
precision?
- What is a convergence order? How do we find them numerically?
- When we solve a 1D Sturm-Liouville problem using a FD method,
what kind of conditions should we have for the coefficients to
guarantee the solution exist and unique?
- How do we get finite difference formulas for u''(x)? How many
different possibilities? What's the best one?
- What are two most important parts in deriving the conservative
FD scheme for S-L BV problems?
- What is a finite difference stencil and a master grid point?
- What is a compact FD scheme?
- Why should we use a conservative FD scheme if available?
- What are local truncation errors and global errors?
- What conditions that can guarantee the convergence of a FD
method.
- What is the stability of a FD method?
- What is the ghost point method? Suppose we have q
u(a) + r u'(a) = s at x=a? How do we use the ghost point
method? * What is the order of the local truncation error
at x=a?
- What is the matrix-structure of the linear system terms of
equations from the FDM applied to S-L problem? Is it an
M-matrix?
- Given u''(x) = f(x), u'(a)=u'(b), does the problem exist? Or
under what kind of condition, the solution exists? If the
solution does exist, is it unique?
- Given u''(x) = f(x), u(a)=u(b), does the problem exist? Or
under what kind of condition, the solution exists? If the
solution does exist, is it unique? It is called a periodic
boundary condition.
- What is a S-L eigenvalue problem? Can you use a FDM to solve
it? How?
- What are the continuous and discrete maximum principle?
- Can you prove the convergence of the finite difference method
using the maximum principle?
- What are the matrix structure of the finite difference
equations for 2D Sturm_Liouville elliptic PDEs using natural
ordering and red-black ordering, respectively?
- How's 9-point compact 4th order finite difference scheme
derived? When we wish to use 4-th order schemes instead of 2nd
order?
Exercises:
- Modify the 1D two point BVP codes to use a tri-diagonal
solver.
- Extra Credit: Derive and implement a FDM for a general 1D
linear two-point BVP, using numerical integration if necessary.
Hint: Find an integration factor to convert the 1D problem to a
standard Sturm-Liouville problem.
- Extra Credit: Use the Green function to show that the 3-point
FD scheme for S-L two-point BVP is second order convergent in
the infinite norm. Hint: Refer to Dr. R. LeVeques's FD
book.
Suggested Projects:
- Develop and test a finite difference method in
spherical coordinates for Poisson equations.
- Explore the possibility of a finite difference
scheme for a 2D diffusion and advection elliptic PDEs.
- Develop and test a finite difference method in
Matlab for Helmholtz equations in Matlab using FFT (fast Fourier
Transforms). It should allow three linear boundary conditions.
- Develop and test a finite difference method in
Matlab for Helmholtz equations in in the entire 2D plane with
some numerical boundary conditions.
- Develop and test a multigrid solver for 2D
Sturm-Liouville elliptic PDEs.
- Develop MOL for 2D heat equations and compare
with the ADI method.
- Implement the projection method for Navier-Stokes
equations on a rectangular domain. Test your code using u(x,y,t)
= sin(pi*x)*sin(pi*y)*h(t).
- Simulate
the lid-driven flow using Helmholtz/Poisson solves (the
projection method)
- Use an extrapolation technique to get fourth order
finite difference schemes for 1D, 2D Poisson and/or elliptic
PDEs of BVP