Applied Mathematics and Numerical Analysis Seminar

         Thursdays, 11:15 am-12:15 pm in Vincent Hall 570



                                   Spring 2005



January 20: Catalin Turc   (University of Minnesota)   Speaker's Homepage

Title: High-order solutions of three-dimensional rough surface scattering problems at high -frequencies

Abstract:
We present a new numerical scheme for three-dimensional acoustic and electromagnetic rough-surface scattering simulations that can deliver highly accurate results from low to high frequencies at a cost that is independent of the wavelength of the incoming radiation. The method is based on high-order asymptotic expansions of the oscillatory integrals that enter potential theoretic formulations of the scattering problems in the high-frequency regime, in a manner that bypasses the need to resolve the fields on the scale of the wavelength of radiation. Indeed, we seek the solutions of the integral equations (e.g. the normal velocity in acoustics and the current in electromagnetics) in the form of slow modulations of highly oscillatory exponentials of known phases, and we choose series expansions in inverse powers of the wavenumber to represent the unknown slowly varying envelopes. As we show, this framework can be made to yield an efficiently computable recursion for the terms of the series to any arbitrary order. The resulting algorithms generally provide a very significant improvement over classical (e.g. Kirchhoff's) approximations in both accuracy and applicability and they can, in fact, efficiently produce results with full double-precision accuracy for configurations of practical interest.




January 27: Peter Sternberg   (Indiana University)   Speaker's Homepage

Title: Elliptic Variational Problems on Constricted Networks with Applications to Ginzburg-Landau Theory

Abstract:
I will analyze variational problems set on a network of thin constricted tubes. In the asymptotic regime where the tubes collapse to a graph, one can identify a one dimensional variational problem with interesting natural boundary conditions at the nodes which in particular encourage jumps in the (limit of) minimizers across the various branches of the graph. An application to tunneling across weak links in a superconductor will be discussed.




February 3: Jon Wilkening   (Courant Institute)   Speaker's Homepage

Title: Stress-driven grain boundary diffusion: modeling, analysis and numerical methods

Abstract:
Microchips often fail when the metallic interconnects between transistors and diodes on the chip degrade due to extremely high current densities. The physics of this process is quite interesting; it is a non-local moving interface problem involving elastic deformation and diffusion. Stress singularities can develop which make boundary conditions difficult to understand and numerical simulation difficult to implement reliably.
After describing the model, I will outline our recent proof of well-posedness, which uses techniques from semigroup theory and requires an analysis of a type of Dirichlet to Neumann map involving the equations of elasticity. I will also briefly describe my recent work on computing stable asymptotics for singularities of Agmon-Douglis-Nirenberg elliptic systems near corners and interface junctions, and show how to adjoin these singular functions to the finite element basis to accurately and efficiently resolve stress singularities without mesh refinement. If time permits, I will also show that my least squares finite element formulation for elasticity transitions gracefully to the Stokes equations in the incompressible limit, and show how to incorporate the convection term to obtain an efficient Navier-Stokes solver for low to moderate Reynolds numbers.




February 10: Jacob Rubinstein   (Indiana University)   Speaker's Homepage

Title: Differential equations in optical design

Abstract:
Classically, differential equations play an indirect role in optical design. Indeed, optics is fundamentally based on Maxwell's equation. Even geometrical optics, which governs the bulk of optical design, is based on a differential (eikonal) equation. Yet, optical designers devised methods that are based to a large extent on ray tracing. Therefore, although PDEs are somewhere at the background, they are not used explicitly. In this talk I shall present examples in optical design where PDEs play a direct role.




February 17: Kumar Tamma   (University of Minnesota)   Speaker's Homepage

Title: The Time Dimension: A New Unified Theory Underlying Linear First Order Systems and a Prelude to Algorithms By Design

Abstract:
A new unified theory underlying the theoretical design of linear computational algorithms in the context of time dependent linear first order systems is presented. Providing new perspectives, and unlike various formulations existing in the literature, it involves the following considerations: i)it provides new avenues for designing new computational algorithms to foster the notion of algorithms by design and recovering existing algorithms in the literature, ii) describes a theory for the evolution of time operators via a unified mathematical framework, and iii) places into context and explains/contrasts future new developments including existing algorithm designs and the various relationships among the different classes of algorithms in the literature such as linear multistep methods, sub-stepping methods, Runge-Kutta type methods, higher order time accurate methods, etc. Subsequently, it provides design criteria and guidelines for contrasting and evaluating time dependent computational algorithms.

The notion of algorithms by design concept capitalizes upon :i) the recently developed unfied theory mentioned above, and ii) newly established algorithmic measures. The linear computaional algorithms in the context of first order systems are classified as distictly pertaining to Type 1, Type 2 and Type 3 classifications of time discretized operators. A generalized stability and accuracy limitation barrier theorem underlies the generic designs of computational algorithms with arbitrary order of accuracy(the Dahlquist theorem is a paricular case limited to second order accuracy), and establishes guidelines which cannot be circumvented.




February 24: Hongjie Dong   (University of Minnesota)   Speaker's Homepage

Title: On the Rate of Convergence of Finite-difference Approximations for Bellman Equations with Constant Coefficients

Abstract:
We consider elliptic Bellman equations with coefficients independent of the variable x. Error bounds for certain types of finite-difference schemes are obtained. These estimates are sharper than those of earlier results.




March 3: Yassine Boubendir   (University of Minnesota)   Speaker's Homepage

Title: The Non-Overlapping Domain Decomposition Method For Scattering Problems

Abstract:
In this talk we shall present some developments on the non-overlapping domain decomposition method in the context of acoustic scattering problems. After a brief introduction of the method, we will discuss the central issue of convergence of the iterative procedure that underlies the approach. More precisely, we shall show that the wave nature of solutions imposes subtle requirements on the choice of inter-domain (Robin) boundary conditions to optimize the convergence properties. If time permits, we shall also describe how to resolve the issue of "cross points" within domain decomposition methods, as well as how to use these schemes in BEM/FEM formulations of the scattering problem.




March 10: Xiantao Li   (IMA)   Speaker's Homepage

Title: A multiscale model for the dynamics of solids

Abstract:
We present a multiscale method for coupling atomistic and continuum models of solids. Both models are formulated in the form of physical conservation laws, and the coupling is achieved through balancing the fluxes. Error estimate is provided, which is of great help in choosing the size of the macro grid and microscale system. Finally we shall show some applications including phase transformation, and dynamic fracture mechanics.




March 17: NO SEMINAR   (Spring Break)



March 24: Richard Kollar   (IMA)   Speaker's Homepage

Title: Linear Stability of Stratified Fluids and the Associated Nonlinear Eigenvalue Problem

Abstract:
It's a problem that has long puzzled fluid dynamicists: How long does it take the waves in a container of fluid to settle? To date there is no complete mathematical analysis; the air/liquid/wall contact line and surface tension complicate things. But for "supercritical" fluids at high pressure (important in several industrial processes, such as decaffeination of coffee) modeled by the incompressible Navier-Stokes equations, the sharp distinction between liquid and vapor disappears. Viscosity can be expected to damp internal waves with a characteristic exponential relaxation time associated with the slowest decaying mode of the system. This work proves that, surprisingly, there is no slowest decaying mode in such stratified fluids. (This is a joint work with R. L. Pego and K. F. Gurski.)




March 31: Fatih Ecevit   (University of Minnesota)   Speaker's Homepage

Title: An efficient integral equation method for electromagnetic and acoustic scattering simulations: convergence of multiple scattering iterations

Abstract:
One of the main difficulties in high-frequency electromagnetic and acoustic scattering simulations is that any numerical scheme based on the full-wave model entails the resolution of wavelength. It is due to this challange that simulations involving even very simple geometries are beyond the reach of classical numerical schemes.
In this talk, we shall present an analysis of a recently proposed integral equation method that bypasses the need for the resolution of wavelength, and thereby delivers solutions in frequency-independent computational times. Within single scattering configurations, the method is based on the use of an appropriate ansatz for the unknown surface densities and on suitable extensions of the method of stationary phase. The extension to multiple-scattering configurations, in turn, is attained through consideration of an iterative (Neumann) series that successively accounts for multiple reflections. We show that the convergence properties of this series in the high-frequency regime depends solely on geometrical characteristics. Moreover,for periodic orbits, we determine the convergence rate with explicit error bounds. Finally, we show that this insight suggests the use of alternative summation mechanisms that can greatly accelerate the convergence of the series.




April 7: Jackie Shen   (University of Minnesota)   Speaker's Homepage

Title: Variational PDE Models in Imaging and Vision

Abstract:
Traditional signal and image processings have mainly been built upon spetral (Fourier) analysis. In the past two decades, variational and PDE methods have gradually gained their momentum in the modeling and analysis of imaging and vision, mainly due to their flexibilities in modeling and computation. In this talk, I will explain this novel approach based on some of my own recent works. I would like to thank Professors Tony Chan and Stan Osher for introducing to me this approach during my postdoctor training at UCLA several years ago.




April 14: NO SEMINAR   (IMA Workshop on Atomic Motion to Macroscopic Models: The Problem of Disparate Temporal and Spatial Scales in Matter, 4/11/05 to 4/15/05)



April 21: Michael P. Hennessey   (University of St. Thomas)   Speaker's Homepage

Title: Optimal Routing of a Sailboat in Steady Winds

Abstract:
A family of novel, insightful, but yet tractable problems regarding the optimal routing of a sailboat in the plane, such as a racing yacht, in steady (i.e. time-invariant) winds is considered. Typically, the optimization criterion is to minimize the route travel time from a start point to a finish point. The solution to the constant wind case is well-known and provides a solid basis for how the problem generalizes and becomes even more realistic. In particular, the mathematical properties of the boat polar are studied in detail and the significance of the luff and downwind dip zones noted. Limited corroborating experimental work was conducted on an instrumented C & C 38 yacht with regard to: (1) boat polar characterization and (2) optimal routing in constant winds. For the case of non-constant, but smooth wind fields, the minimum-time problem takes on the form of a Zermelo problem, resulting in a 2-point boundary-value problem generated from application of the calculus of variations and requires a single iteration of the initial boat heading angle. However, a major subtlety concerns the continuity level of the optimal route and how it is affected by the mathematical properties of the boat polar. Specifically, if the entire optimal route is sufficiently continuous (e.g. C2), the calculus of variations applies directly. If not, it may be applied in piece-wise fashion -- this provides motivation for introducing the notion of a discontinuity map associated with the set of finish points in the plane and their minimal discontinuity based optimal routes. Various formulations of increasing spatial dimensionality are identified and checked with Mathematica, as characterized by the type of wind field and the location of the finish point. Major cases investigated include: (1) 1-dimensional (1D) wind direction field with constant wind speed, (2) 2D wind direction field with constant wind speed, (3) constant direction wind field with speed variation, and (4) 2D wind speed and direction field. Other optimization criteria and associated solutions are also studied -- several velocity made good (VMG) techniques in particular, one of which is equivalent to the minimum-time criterion. Numerical simulation studies using MATLAB are provided for an interesting scenario involving 1D spatially oscillating winds that illustrate how the minimum-time route is generated and also how its continuity is affected by the boat polar's properties. Other optimization criteria and corresponding solutions are compared and contrasted as well (i.e. two VMG approaches, minimum-distance). In summary, this talk summarizes attempts to make a contribution in the area of sailing mechanics and optimal routing problems, with a special emphasis on, and appreciation of, continuity issues.




April 28: Joel Tropp   (University of Michigan)   Speaker's Homepage

Title: Signal recovery from partial information

Abstract:
This lecture will demonstrate theoretically and empirically that a greedy algorithm called Orthogonal Matching Pursuit (OMP) can reliably recover a signal with $m$ nonzero entries in dimension $d$ given $O(m \ln d)$ random linear measurements of that signal. This is a massive improvement over previous results for OMP, which require $O(m^2)$ measurements. The new results for OMP are comparable with recent results for another algorithm called Basis Pursuit (BP). The OMP algorithm is much faster and much easier to implement, which makes it an attractive alternative to BP for signal recovery problems. This work is in collaboration with Anna Gilbert.




May 5: NO SEMINAR   (IMA Workshop on Experiments in Physical Biology, Part I, 5/2/05 to 5/6/05)