Math 627 - Introduction to Parallel Computing
Spring 2004 - Matthias K. Gobbert
Class Presentations of the Final Projects
This page can be reached via my homepage at
http://www.math.umbc.edu/~gobbert.
Program
Monday, May 17, 2004, 03:30 p.m., MP 401
-
03:35-03:55
Application of a Kinetic Transport and Reaction Model for
Chemical Vapor Deposition at the Feature Scale
Mark Breitenbach
The physical study of microelectronics manufacturing processes
such as chemical vapor deposition (CVD) involve modeling reactive gases.
Operating under certain assumptions, the flow of reactive gas can be
modeled by a system of linear Boltzmann transport equations (BTE). The
linear BTE is approximated in velocity space by a system of hyperbolic
conservation equations in space and time and is solved using the
discontinuous Galerkin finite element method (DGM). The DGM is
implemented in parallel and applied in three-dimensional physical
simulations designed to study how the flow of one reactive gas is affected
by pressure in the CVD process. Near optimal speedup for DGM is achieved
for large problem sizes.
-
04:00-04:20
A Parallel Multigrid Solver to the Poisson Equation
Jonathan Desi
The numerical solution of partial differential equations typically
involves the solution of large, sparse, highly structured system of linear
equations. Using a finite difference discretization on the Poisson
equation, we obtain a prototypical system of this type. The performance
in solving this system can be vastly improved by using a multigrid
algorithm in conjunction with the conjugate gradient method of iteratively
solving a system of linear equations. Additionally, we implement this
algorithm in parallel to further improve the performance.
-
04:25-04:45
Operator Upscaling for the Acoustic Wave Equation
Tetyana Vdovina
We present theoretical description and parallel algorithm of an operator
based upscaling technique for modeling wave propagation in heterogeneous
media, which contain many numerical scales. We consider constant density,
variable sound velocity two-dimensional acoustic wave equation. Operator
based upscaling captures the effect of the fine scales on the coarse
cells, but does not require solving the original full fine scale problem.
The method consists of two stages. We first solve for the subgrid
information internal to each coarse block. Then we define an upscaled
problem on the coarse grid, which includes the fine scale information from
the first stage. To demonstrate the accuracy of the method we compare the
upscaled solution with the solution produced by a second-order in space
and time finite-difference scheme. Our results are proved to be in good
agreement with finite-difference solution. We study the parallel
performance of the algorithm, and demonstrate that the numerical speedup
is very close to the theoretical estimates.
-
04:50-05:10
Parallel Spots Detection of 2-D Gel Images
Feng Li
Two-dimensional polyacrylamide gel eletrophoresis is the core technique
for separating proteins.
One important step in 2-D gel image analysis is to do spot detection.
A parallel processing method for spot detection of 2-D gel images is
discussed and explained in detail. Some experimented
results show the efficiency of this method.
Copyright © 2004 by Matthias K. Gobbert. All Rights Reserved.
This page version 1.1, May 2004.