(September 6) Professor Noel Walkington:
Construction and Analysis of Unstructured Mesh
Generation Algorithms -
Given a collection of points, edges, and faces, in a bounded region
$\Omega \subset \Re^d$, the meshing problem is to construct the
coarsest possible triangulation of $\Omega$ from tetrahedra having
bounded aspect ratio which ``conforms'' to the input. These
requirements can lead to very complicated algorithms; so much so that
it can be difficult to verify correctness. I will give an overview of
the ideas and issues that arise when constructing algorithms to solve
the meshing problem, and will indicate how the mesh generation problem
touches on many areas of mathematics and computer science such as
approximation/interpolation theory, computational geometry, sphere
packing, graph theory, algorithm design. (September 13) Mr. Jae-Hong Pyo:
A Finite Element Gauge-Uzawa Method for the Evolution Navier-Stokes Equations -
The Navier-Stokes equations of incompressible fluids are still a computational
challenge. The numerical difficulty arises from the incompressibility
constraint, which requires a compatibility condition (discrete inf-sup)
between the finite element spaces for velocity and pressure. Several
projection methods have been introduced for time discretization to circumvent
the incompressibility constraint, but suffer from boundary layers. They are
either numerical or due to non-physical boundary conditions on pressure.
We introduce a first order Gauge-Uzawa method for time discretization coupled
with a stable finite element method for space discretization. The method is
unconditionally stable, consists of d+1 Poisson solves per time step, and
does not exhibit pronounced boundary layer effects. We prove error estimates
for both velocity and pressure under realistic regularity conditions via a
variational approach, and illustrate the performance with several numerical
netscapw&experiments. This work is joint with R. H. Nochetto.
(September 27) Dr. Axel Voigt:
From Macro- to Microscale in Modeling Silicon Crystal Growth -
In order to produce silicon wafers which are suitable for microelectronic
devices a high quality of the silicon material must be ensured. Silicon
crystals are grown from the melt and their quality is highly process
dependent. We present a mathematical model which relates macroscopic growth
conditions to lattice defects in the growing crystal. The heat and mass
transfer model is solved using an adaptive finite element method at different
time and length scales. The simulation results are verified with experimental
measurements. (October 2) Professor Wolfgang Dahmen:
Adaptive Multiscale Methods - A somewhat Different Approach
to Discretization -
This talk is concerned with the design and analysis of adaptive
schemes for a wide class of variational problems that are
well-posed in the sense that the induced operator is a
topological isomorphism from the ``energy space'' onto its dual.
This covers classical elliptic boundary value problems, corresponding
boundary integral formulations, transmission problems as well as
indefinite problems like mixed formulations and saddle point
problems. The main point is that, whenever the the energy space
has a wavelet characterization in terms of norm equivalences,
this combined with the well-posedness allows one to transform
the original problem into an equivalent one that is, however,
well-posed in the Euclidean metric. The idea is to apply
(conceptually) an iterative scheme to the latter (infinite dimensional)
problem. At each stage of the iteration the application of infinite
dimensional
operators to the current iterate is evaluated adaptively within
dynamically updated precision tolerances. We highlight some
concepts from nonlinear approximation and harmonic analysis
that can be used to show that the schemes exhibit in some sense
optimal work/accuracy rates. Some indications concerning
the treatment of time dependent and nonlinear problems are
given. The theoretical
results are illustrated by numerical examples for definite
and indefinite problems. The material presented in this talk is based
on joint work with A. Cohen, R. DeVore, S. Dahlke and K. Urban. (October 4) Professor Zhimin
Zhang:
Superconvergence of the Zienkiewicz-Zhu Patch Recovery for
Rectangular Elements -
The Zienkiewicz-Zhu patch recovery is a post-processing
technique for the finite element method. It recovers
gradient quantities in an element from element patches
surrounding the nodes of the element. It has been shown
numerically that ZZ provides superconvergent recovery
on regular meshes and provides recovery with much improved
accuracy on general meshes. In this talk, we present a
theoretical investigation on this remarkable recovery
technique and prove its superconvergence under rectangular
grids.
>
(October 11) Professor Howard Elman:
Performance and Analysis of Preconditioners for the Incompressible
Navier-Stokes Equations -
Discretization and linearization of the incompressible Navier-Stokes
equations leads to linear algebraic systems in which the coefficient
matrix has the form of a saddle point problem. In this talk, we describe
the development of efficient iterative solution algorithms for this class
of problems. In particular, we show how initial approaches used for the
Stokes equations, such as the Uzawa algorithm, can be viewed as solvers
for the Schur complement system, and we discuss the advantages of
explicitly working with the coupled form of saddle point problems and
using the Schur complement as a tool for developing preconditioners. We
describe some new algorithms derived using this point of view, and we
demonstrate their effectiveness for solving both the steady-state and
evolutionary Navier-Stokes equations.
(October 25) Professor Du Qiang:
Centroidal Voronoi Tessellations (CVTs) and their applications to
Numerical PDEs -
A centroidal Voronoi tessellation (CVT) is a Voronoi
tessellation of a given set such that the associated generating
points are centroids (centers of mass) of the corresponding
Voronoi regions. We discuss the applications of CVTs
to various scientific and engineering problems,
in particular, applications to numerical PDEs such as
unstructured grid generation/optimization and meshless computing.
We also present methods for computing these tessellations.
(November 1) Professor Zhiqiang Cai:
The Rate of Corrections and its Application in Scientific Computing -
One of the major issues in numerical analysis/scientific computing
is the accuracy of the underlying numerical methods. It is a common
knowledge that the accuracy of all current numerical methods for
differentiation, integration, ordinary and partial differential
equations etc are limited by the highest derivative of the underlying
approximated function. Since solutions of many partial differential
equations are not smooth, low order methods with adaptive mesh
refinements seem to be the method of choice for such problems without
the prior knowledge of singular behavior. But they are expensive.
Obviously, it is desirable to develop higher order accurate methods
for non-smooth problems. This seems to be a paradox. This talk presents an approach for computing higher order accurate
approximations for differentiation, integration, ODEs, and PDEs
when the underlying approximated functions are not smooth. The key
idea of this work is the introduction of the rate of corrections that
is of universality, that quantifies the accuracy of the numerical method
used, and that is computationally feasible. The rate of corrections is then
used to increase the accuracy of the approximation. Also, this approach
can be applied to problems without continuum background, such as the
sequence of period-doubling bifurcations in chaos.
(November 8) Dr. Maxim Olshanskii:
Finite element method for the Navier-Stokes equations:
A stabilization issue and iterative solvers
-
A Galerkin finite element method is considered to approximate
the steady incompressible Navier-Stokes equations together with iterative
methods to solve a resulting system of algebraic equations. This system
couples velocity and pressure unknowns, thus requiring a special technique
for handling. Doing this, we will involve the equations in
velocity-Bernoulli pressure variables. Also a consistent stabilization
method is considered from a new view point. The method suppress
instabilities occurring for low viscosity due to the pressure gradient in
momentum equations and has also impact on solvers performance. Theory and
numerical results in the talk address both the accuracy of discrete
solutions and the efficience of solvers. (November 15) Professor Roger Temam:
Robust Control of Turbulent Flows
-
Robust (or $H^\infty$) control is aimed at controlling systems
which are subjected to unpredictable small disturbances. The subject
has emerged from space study for the stabilization of light structures
in space. It has developed, especially in the context of linear
equations and linear operator theory.
In this lecture we will give some remarks concerning robust
control for nonlinear equations, especially for the control of
turbulent flows governed by the Navier Stokes equations.
(November 16) Professor Roger Temam:
Mathematical Problems in Meteorology and Oceanography
-
In this lecture we will present some recent and some less recent
mathematical results concerning the equations of the atmosphere, the ocean
and the coupled atmosphere ocean (the so-called Primitive Equations first
considered by Richardson). (November 29) Professor Ron DeVore:
Adaptive Methods for Solving PDEs -
Adaptive methods are often used to numerically resolve PDEs when
the solution to the PDE is known to exhibit singularities. Yet, it is
rare that there is even a convergence theory for an adaptive numerical
method much less an analysis of the decay of error in terms of the number
of computations. In this talk, we shall develop an analysis which will
allow the a priori determination of whether an adaptive numerical
method can perform better than the more standard (and numerically less
intensive) linear methods such as standard finite element methods.
Since adaptive methods are a form of nonlinear approximation, it is not
surprising that this theory has as one of its pillars the fundamental
theorems which characterize approximation rates (in terms of smoothness
conditions on the target function) of nonlinear methods. The other
pillar for this theory is regularity of the solution to the PDE. But the
new twist is that the regularity is not measured in the usual Sobolev
scale but rather in a scale of Besov spaces commensurate with nonlinear
methods. We shall give examples of how to apply this theory to hyperbolic
and elliptic problems. (December 6) Dr. Alan Sussman:
Storing and Processing Multi-dimensional Scientific Datasets -
Large datasets are playing an increasingly important role in many areas
of
scientific research. Such datasets can be obtained from various
sources,
including sensors on scientific instruments and simulations of physical
phenomena. The datasets often consist of a very large number of
records,
and have an underlying multi-dimensional attribute space. Because of
such
characteristics, traditional relational database techniques are not
adequate to
efficiently support ad hoc queries into the data. We have therefore
been
developing algorithms and designing systems to efficiently store and
process these datasets in both tightly coupled parallel computer systems
and more loosely coupled distributed computing environments. In this talk, I will discuss the design of two systems, the Active Data
Repository (ADR) and DataCutter, for managing large datasets in parallel
and distributed environments, respectively. Each of these systems
provides
both a programming model and a runtime framework for implementing high
performance data servers. These data servers provide efficient ad hoc
query
capabilities into very large (currently up to multiple terabytes)
multi-dimensional datasets. ADR is an object-oriented framework that
can
be customized to provide optimized storage and processing of disk-based
datasets on a parallel machine or network of workstations. DataCutter
is a
component-based programming model and runtime system for building data
intensive applications that can execute efficiently in a Grid
distributed
computing environment. I will present optimization techniques that
enable
both systems to achieve high performance in a wide range of application
areas. I will also present performance results on real applications on
various computing platforms to support that claim.
|