(September 12) Professor Walter Gander:
Generating Numerical Algorithms Using Computer Algebra -
We show how numerical algorithms can be derived in a simple way using
computer
algebra. Examples are numerical differentiation, quadrature and multi-step
methods for ODE. It is also shown how the discretization error of a
method
can be computed automatically. This approach not only makes formularies
obsolete (in fact some errors were found in Abramowitz/Stegun) but is also
useful in teaching since principles and fundamentals are emphasized and we
can
leave the sometimes tedious derivation of the specific algorithm to the
machine. (September 19) Dr. Milan Mihajlovic:
Efficient Preconditioning of the Navier-Stokes Equations with the Free
Surface Condition -
A numerical model is presented for the simulation of viscoelastic flows (Oldroyd-B)
with complex free surfaces in three space dimensions. The mathematical formulation of
the model is similar to that of the volume of fluid (VOF) method, but the numerical
procedures are different. The time discretization is based on a splitting scheme. The
first step consists in solving advection equations whilst the second step corresponds
to a simplified Oldroyd-B problem in a fixed domain. Numerical results will be
presented for the buckling of a jet and for the stretching of a filament. Extension to the stochastic Hookean dumbbells model is then discussed and existence
on a fixed time interval is proved provided the data are small enough. This result is
based on the maximal regularity property of the Stokes operator and on the analycity
behavior of the corresponding semi-group. A finite element discretization in space is
then proposed. Existence of the numerical solution is proved for small data, as well
as a priori error estimates, using an implicit function theorem framework.
(September 26) Dr. Andrea Bonito:
Three-Dimensional Viscoelastic Flows with Complex Free Surfaces:
Numerical Simulations and Mathematical Analysis of Simplified
Problems
-
A numerical model is presented for the simulation of viscoelastic flows (Oldroyd-B)
with complex free surfaces in three space dimensions. The mathematical formulation of
the model is similar to that of the volume of fluid (VOF) method, but the numerical
procedures are different. The time discretization is based on a splitting scheme. The
first step consists in solving advection equations whilst the second step corresponds
to a simplified Oldroyd-B problem in a fixed domain. Numerical results will be
presented for the buckling of a jet and for the stretching of a filament. Extension to the stochastic Hookean dumbbells model is then discussed and existence
on a fixed time interval is proved provided the data are small enough. This result is
based on the maximal regularity property of the Stokes operator and on the analycity
behavior of the corresponding semi-group. A finite element discretization in space is
then proposed. Existence of the numerical solution is proved for small data, as well
as a priori error estimates, using an implicit function theorem framework.
(October 10) Professor Ivo Babuska:
Are the Computational Results Reliable? -
The notion of the reliability is complex. It will be understood as the reliability of the prediction of the phenomena based on the computation. This prediction is then the basis for the decision.
Few major engineering accidents representing different reasons for the non reliability will be presented.
a.Modeling. The incorrect mathematical problem was solved.
b.Discretization. The error in the numerical solution (of the correct problem).
c.Computer science. The round off, underflow.
d.Human error.
Category a. belongs to the field of Validation, Categories b.c.d. to the Verification.
Notions of the Mathematical problem, Verification, Validation, Quantification of the uncertainty and the regulatory assessment will be presented.
The Validation pyramid of the Airbus A380 and the failure of the wing test (February 2006) will be discussed.
A specific example of the frame where the regulatory assessment is to compute the probability that the displacement in a point is not larger than 3mm and the estimate of the confidence in the computed result. For disposition are the experimental data for the calibration, validation and the accreditation (which create the validation pyramid).
Three groups of a different number of available data are given. The purpose was to show the influence of the amount of experimental data on the reliability of the results. This was one of the problems of the Sandia workshop. (May 2006).
Mathematics of the Validation is very broad and new. Validation has many aspects, physics, experimentation, sensitivity, philosophy of science.
Finally some very basic literature will be mentioned.
(October 17) Dr. Manuel Cascon:
Convergence and Optimality of an AFEM for General Second Order Linear Elliptic PDE -
Despite the overwhelming computational evidence that AFEM lead to optimal meshes,
the mathematical theory started very recently with Binev et al
[1]and Stevenson [3]. Paper [1] discusses geometric properties of bisection which turn out to be crucial for optimality; The AFEM in[1] includes a coarsening step which is not necessary in practise for linear
elliptic PDE. Stevenson presents in [3] an algorithm with optimal complexity without coarsening, but it does not work with general linear elliptic operator and is unpractical. In this paper, we modify the Morin-Nochetto-Siebert algorithm
[2] with a new procedure for oscillation reduction relative to error estimator. We prove that the resulting AFEM is a contraction for the total error between two consecutive adaptive loops. This strict error reduction property then leads to the
optimality of AFEM, that is to a total error decay in terms of number of degrees of freedom as predicted by the best approximation. Bibliography [1] P. Dinev, W. Dahmen, and R. DeVore
Adaptive finite element methods with convergence rate,
Numer. Math., 97(2), pp.~219-268 (2004). [2] P. Morin, R.H. Nochetto, and K.G.
Siebert, Data Oscillation and Convergence of adaptive FEM.
SIAM J. Numer. Anal. 38, pp.~466--488 (2000). [3] R. Stevenson,
Optimality of a standard adaptive finite element method
(to appear).
(October 24) Professor Leszek Demkowitz:
hp-FINITE ELEMENTS FOR MAXWELL EQUATIONS: Discrete Compactness and hp-Convergence for Maxwell Eigenvalues
-
We will present the Finite Element discretization of time-harmonic
Maxwell equations. We shall start with a variational formulation,
and discuss the so-called stabilized formulation that reveals
that we are dealing with a mixed problem. The satisfaction
of two Brezzi's inf-sup conditions leads then to the discrete
exact sequence property, and the analysis of the corresponding
Maxwell eigenvalue problem. We shall introduce then the Kikuchi's
idea of discrete compactness property, and discuss how it implies the
convergence of the Maxwell eigenvalues. Finally, we will
present a proof of the discrete compactness for a two-dimensional
hp-method based on Nedelec's rectangle of the first kind. The result is a joint work with D. Boffi, M. Dauge and M. Costabel.
(October 31) Professor Luc Mieussens:
Numerical simulations of rarefied gases in curved channels: thermal creep, circulating flow, and pumping effect -
Thermal creep is a flow of a slightly rarefied gas caused by the
temperature gradient along a wall. Although it is one of the classical
phenomena in rarefied gas dynamics, its importance was revived in
recent years in connection with micro gas flows. The aim of this work is to study the possibility to use the thermal
creep to make a pump without moving part for a gas under rarefied
conditions or in micro-scales. We first use an original 2D numerical solving of a Boltzmann kinetic model equation. We also derive an asymptotic model (by the diffusion approximation theory) in case of thin channels. The results of these two methods are compared. In collaboration with K. Aoki, S. Takata and H. Yoshida (Kyoto, Japan)
and P. Degond (Toulouse, France)
(November 07) Dr. Alexei Lozinski:
Modeling and simulations of gas bubbles in a fluid
-
In this talk we shall discuss a model allowing to describe
motion and coalescence of gas bubbles in a liquid under the action of
gravitation and surface tension. The shape of the bubbles is described by
a pre-defined family of mappings specified by a small number of
parameters, and the effects of the gas motions inside the bubbles are
neglected. The motion of a bubble is obtained then in a Lagrangian form
using the principle of virtual works. The set of equations is numerically
solved with the help of the fictitious domain technique with Lagrange
multipliers on the bubbles boundaries. Numerical results in 2D and 3D
will be presented.
(November 14) Ms. Olena Shevchenko:
Some Aspects of the Theory of Long-Step Interior Point Methods Beyond
Symmetric Cone Programming -
Interior-Point methods play an important role in modern optimization. These
methods are popular because of their efficiency and polynomial-time
complexity. The algorithms that allow long steps are the most efficient
among various interior-point methods. They were first developed for linear
programming and then generalized to semidefinite and symmetric cone
programming. Later, Guler showed that it is possible to extend long-step
methods to larger classes of problems like hyperbolic or homogeneous
programming problems. The interior-point methods are based on generating iterates that follow some
path in the interior of the feasible region that consists of the minimizers
of a family of self-concordant barrier functions. These are auxiliary
functions that keep the iterations inside the feasible region. They satisfy
certain (restrictive) relations between the first, the second, and the third
directional derivatives. It is crucial to have efficiently computable
self-concordant barriers preferably with the best possible complexity value.
In this talk, we introduce a new recursive formula for an optimal dual
barrier function for homogeneous cones and discuss some of its properties.
If
time permits, we will talk about the work done jointly with Osman Guler on
hyperbolic programming problems, hyperbolic polynomials, the Lax conjecture,
and its applications.
(November 21) Professor Igor Griva:
Case Studies in Shape and Trajectory Optimization: Catenary Problem -
This talk presents a case study in modern large-scale constrained
optimization to illustrate how recent advances in algorithms and modeling
languages have made it easy to solve difficult problems using optimization
software. We consider the shape of a hanging chain, which, in equilibrium,
minimizes the potential energy of the chain. We emphasize the importance
of the modeling aspect, present several models of the problem and
demonstrate differences in iteration numbers and solution time.
(November 30) Professor George Papanicolaou:
Optimal Illumination in Array Imaging -
I will discuss the mathematical problem of optimally illuminating an object for imaging by an array. In a certain regime of parameters and for a special class of objects this can be done by using spheroidal wave functions. In general situations one must use algorithms that optimally image the object by detecting its edges. Such algorithms are very different from the ones that maximize the energy of the signals received by the array so as to enhance detectability. I will analyze and compare the two types of algorithms. (Joint work with L. Borcea and C. Tsogka.) (December 01) Professor George Papanicolaou:
Imaging in Random Media
-
I will present an overview of some recently developed methods for imaging
with array and distributed sensors when the environment between the objects
to be imaged and the sensors is complex and only partially known to the
imager. This brings in modeling and analysis in random media, and the need
for statistical algorithms that increase the computational complexity of
imaging, which is done by backpropagating local correlations rather than
traces (interferometry). I will illustrate the theory with applications from
non-destructive testing and from other areas.
(December 05) Dr. Christian Kreuser:
Convergence of Adaptive Finite Element Methods for $p$-Laplace
Equation
-
We consider the homogeneous Dirichlet Problem for the $p$-Laplacian,
$p\in(1,\infty)$. We propose an adaptive algorithm with continuous piecewise
affine finite elements and prove an error reduction rate of approximate
solutions to the exact one. We improve the a posteriori estimations for
quasi-norms and generalize the error reduction property of the linear case
to an energy reduction property in the nonlinear case. For adaptive
refinement we use a marking strategy incorporating only error estimators.
Thus we obtain a strict reduction of energy differences plus oscillation.
Since energy differences are proportional to the error measured in
quasi-norms we get linear convergence.
(December 12) Dr. Jing Zou:
The Super Fast Spectral Method for Multiscale PDEs -
We develop a new sublinear super fast spectral method. It takes time
$O(log N)$, as compared with the O(N log N) cost of the traditional
spectral method, where N is the number of gird points. We apply this
new method to solve the following 1-dimensional multiscale problem:
u_t - (a(x, x/eps) u_x)_x =0, u(0,x)=f(x),
where the coefficient a is bounded, periodic, continuous and uniformly
positive, a and f are each well represented by sparse Fourier
representations. Both theoretical analysis and numerical results show
the encouraging advantage of the new method in speed over the traditional
spectral methods, when the number of grid points N is very large.
Finally, we also apply this method to solve elliptic and hyperbolic
multiscale PDEs.
|