Math 621 - Numerical Analysis II
Numerical Methods for Partial Differential Equations
Fall 2007 - Matthias K. Gobbert
Class Presentations of the Final Projects
Friday, December 14, 2007, 02:30 p.m., MP 401
-
02:30-02:45
The Matrix-Free Preconditioned Conjugate Gradient Method:
Demonstrated on a 2D Poisson Equation Example
Amanda Gassman
The existing Preconditioned Conjugate Gradient method in Matlab does not
efficiently compute the numerical solution and therefore an optimized
method was developed. The algorithm was optimized in several stages and
tested repeatedly at each stage on a two dimensional Poisson problem to
ensure that it produces the same numerical solution as the original Matlab
function. Some of the memory saving techniques include: creation of a
function which reuses some vectors, development of a matrix-free method of
computing matrix-vector multiplications, and derivation of a matrix-free
method that implements preconditioners on the system matrix.
All techniques were combined to create a
matrix-free Preconditioned Conjugate Gradient method,
which requires less time and less memory than the original.
Ultimately, convergence results confirm the accuracy of the new method and
demonstrate its superiority.
-
02:45-03:00
Parallelism of the Load Vector Calculation in a
3D Upscaling Elastic Wave Propagation Code
Sean Griffith
Solving the elastic wave equation often requires dealing with data on
multiple scales. However, computational limits often require the use of a
grid that is coarse relative to finer features of the domain. The upscaling
technique adequately deals with this issue by breaking a large domain into
subdomains, and solving the problem on each sub-domain separately. Each
sub-domain solution is then used to solve the problem on the coarse scale.
Using fine grid information to solve the coarse problem imparts some of the
fine scale accuracy to the coarse solution. Additionally, this technique
has the advantage that subdomain problems can be solved in parallel. In
this presentation, it is seen that this parallelizability extends to the
coarse grid solution as well, in particular to the calculation of the load
vector needed to solve the coarse problem via finite elements.
-
03:00-03:15
A Convergence Study of Finite Element Solutions Using True Solutions in
COMSOL Multiphysics
Shiming Yang
Finite element solutions' L2-norm errors
based on true solutions are studied for Poisson equations. Smooth and
non-smooth test problems are solved with Lagrange elements of all
polynomial degrees available in COMSOL Multiphysics, with respect to
two types of domain shapes. The results agree with the convegence
theorem, but need more explanation for certain cases. Results in this
work leads up to the development and testing of a procedure based
on a reference solution in cases where a true solution is not available.
-
03:15-03:30
Numerical Studies of the Asymptotic Behavior of a
Reaction-Diffusion System with a Fast Reaction
Yushu Yang
A system of reaction-diffusion equations with a fast reaction is
studied with an initial condition with three interfaces between regions
of dominance by the two main reactants. The efficiency
and accuracy of the numerical solver is studied for two cases of
faster reaction rate values. We observe that the asymptotic limit
of the problem is already approached for the values considered.
The talk will also report ODE order k and
Newton iteration steps for studying behavior of the numerical solution.
-
03:30-03:45
Numerical Solution of the Second Order Wave Equation
Noemi Zakarias
The second-order two-dimensional wave equation is solved analytically
using traditional techniques in Partial Differential Equations and
numerically using both the finite difference method (implemented in
MATLAB) and COMSOL Multiphysics solver package based on the finite element
method. Stability theory for the finite difference method and numerical
experiments are presented. Accuracy and stability are confirmed for the
leapfrog finite difference scheme (centered second differences in time and
space) and the convergence order of the finite element solution is
estimated in the L2-norm using the COMSOL Script feature.
The results show agreement between the true and numerical solutions.
Copyright © 2001-2007 by Matthias K. Gobbert. All Rights Reserved.
This page version 1.0, December 2007.