Applied Mathematics and Numerical Analysis Seminar

         Thursdays, 11:30 am-12:15 pm in Vincent Hall 570



                                   Fall 2002



Sept 12: Speaker: Fernando Reitich   (U. Minn.)

Title: Shape deformations and analytic continuation in free boundary problems

Abstract: In this talk I will present a general approach to the analysis of steady states and nonlinear stability for a class of free-boundary problems. The approach is based on the explicit consideration of series solutions in a parameter measuring variations of domains (e.g. steady state or initial configurations) from separable geometries (e.g. planes, spheres, etc). The method relies on the derivation of formulas for the recursive evaluation of (Taylor) approximations of arbitrary order in the variation parameters, and on the iterative estimation of the growth, in appropriately defined spaces, of the resulting functional coefficients. We shall present a variety of instances where this procedure leads to general results on analyticity of solutions, on existence of complex steady states and on nonlinear stability of equilibria. Particular attention will be given to examples of "curvature driven" free-boundary problems, including the classical Stefan problems, models of capillary fluid drops, water waves, tumor growth, etc. We shall further show that the relevance of our studies goes beyond the theoretical, as it uncovered the mechanism behind the observed performance of a class of numerical algorithms, based on shape-perturbation theory, that have been used in these and other contexts. Indeed, our research shows that the standard implementations of these schemes can suffer from severe ill-conditioning as a result of pronounced cancellations in the underlying recursions. Moreover, our work further suggests alternative implementations with greatly improved properties of numerical stability and convergence, enhanced by methods of analytic continuation. As we shall demonstrate these modifications can have a dramatic effect on the accuracy and applicability of perturbative numerical approaches to (boundary value and) free-boundary problems.



Sept 19: Speaker: Guido Kanschat   (U. Heidelberg)

Title: Multilevel-Preconditioning for Discontinuous Galerkin Methods

Abstract: Variants of a multi-level preconditioner for the interior penalty method are presented. They are analyzed in the framework established by Bramble/Pasciak for symmetric problems and shown to be optimal. Then, the methods are extended to other discontinuous discretizations of elliptic problems. Finally, they are applied to problems of computational fluid dynamics.



Sept 26: Speaker: Dominik Schoetzau   (U. Basel)

Title: Mixed hp-DGFEM for incompressible flows

Abstract: We consider the discretization of incompressible fluid flow problems by mixed discontinuous Galerkin methods. The derivation of these methods is explained for velocity-pressure elements of the type ${\cal Q}_k-{\cal Q}_k$ and ${\cal Q}_k-{\cal Q}_{k-1}$. It is shown that these elements satisfy suitable inf-sup conditions on geometric meshes in polygonal and polyhedral domains. The meshes may be refined anisotropically and non quasi-uniformly towards edges and corners, respectively. The discrete inf-sup constants are proven to be independent of the aspect ratio of the anisotropic elements and only weakly depending on the approximation orders. Using these results, we derive a-priori error estimates for $hp$-approximations on geometric meshes. A series of numerical tests is presented for Stokes and Oseen flow. In particular, the results demonstrate that the methods perform well for a wide range of the Reynolds numbers.



Oct 3: Speaker: Bernardo Cockburn   (U. Minn.)

Title: A new characterization of hybridized mixed methods

Abstract: Mixed finite element methods are a very powerful method for numerically solving second-order elliptic problems. Since their approximate solution is difficult to compute, it is necessary to recast them in a suitable way; this can be achieved by what is called a hybridization procedure. The price to pay, however, is that we must introduce new unknowns, called Lagrange multipliers, which satisfy a matrix equation difficult to compute. In this talk, we give a new characterization of the approximate solution given by hybridized mixed methods for second-order, self-adjoint elliptic problems. We then apply this characterization to obtain an explicit formula for the entries of the matrix equation for the so-called Lagrange multipliers. We also obtain necessary and sufficient conditions under which the multipliers of two well-known methods (the Raviart-Thomas and the Brezzi-Douglas-Marini) of similar order are identical.



Oct 10: Speaker: E. McKay Hyde   (U. Minn.)

Title: A new characterization of hybridized mixed methods

Abstract: Mixed finite element methods are a very powerful method for numerically solving second-order elliptic problems. Since their approximate solution is difficult to compute, it is necessary to recast them in a suitable way; this can be achieved by what is called a hybridization procedure. The price to pay, however, is that we must introduce new unknowns, called Lagrange multipliers, which satisfy a matrix equation difficult to compute. In this talk, we give a new characterization of the approximate solution given by hybridized mixed methods for second-order, self-adjoint elliptic problems. We then apply this characterization to obtain an explicit formula for the entries of the matrix equation for the so-called Lagrange multipliers. We also obtain necessary and sufficient conditions under which the multipliers of two well-known methods (the Raviart-Thomas and the Brezzi-Douglas-Marini) of similar order are identical.



Oct 17: Speaker: Qi Wang   (F.S.U.)

Title: Modeling flows of nematic liquid crystalline polymers using kinetic theory

Abstract: I will present a general Doi-type kinetic theory for nematic liquid crystalline polymers (LCPs) accounting for the molecular aspect ratio, excluded volume interaction, long-range intermolecular interaction, and chirality of the molecules. I will derive an approximate intermolecular potential for LCP molecules of the spheroidal shape and the elastic as well as the visocus stress expression corresponding to the shaped molecule being transported in viscous solvent. I will then show the theory obeys the second law of thermodynamics and "reduces" to the well-known Ericksen-Leslie theory for nematic liquid crystals in the weak flow, weak elasticity, and slow time limit. Applications of the theory and its moment approximations in simple flows will be discussed in the end.



Oct 24: Speaker: Alexander Alekseenko   (U. Minn.)

Title: Hyperbolic formulations in linearized gravity

Abstract: The problem of running stable evolution for the Einstein's equation keeps challenging scientists for many years. With the development of Laser Interferometer Gravitational Observatories whose main purpose is the detection of the gravity waves, the need in reliable computations had grown tremendously. At the same time, difficulties of numerical relativity are so intense and various that one can hardly expect a real success without the deepest analysis of the subject. What are the challenges that make the discretization of the relativity equations difficult? Possibly, many questions will drop out once we know a natural way of writing the Einstein's equation. Indeed, in terms of curvature and stress-energy tensors the Einstein's equation is short, but when translated into a coordinate system, it turns into a system of 10 nonlinear PDE's with a huge amount of terms, and from 10 unknown components of the space-time metric only 6 can be determined. Due to this one and due to other reasons which will be addressed in the talk various reformulation were proposed to write the Einstein's equation in a manner suitable for numeric computations. The most promising approach is to write the Einstein equation in the form of a symmetric hyperbolic system. In this talk we will discuss examples of well-known symmetric hyperbolic formulations and introduce a new one which is the result of our work with Professor Arnold during the last year. Finally, we will briefly mention the future challenges in particular, the admissible boundary conditions and approaches to discretization.



Oct 31: Speaker: John Lowengrub   (U. Minn.)

Title: Numerical simulation of drop coalescence with surfactant in 3D

Abstract: In the processing of emulsions and polymer blends, the drop size distributions are determined by two coexisting processes: drop breakup and coalescence. Here we study the effects of surfactants, e.g. block copolymers, on coalescence. We use a newly developed 3D adaptive finite-element algorithm. The method is based on unstructured adaptive triangulated and tetrahedral meshes that discretize the interfaces and the bulk respectively, and on an efficient parallelization of the numerical solvers.

A nonlinear Langmuir equation of state for the surfactant is used and Van der Waals forces, which are responsible for coalescence, are included in the numerical method. Surfactants are transported by convection-diffusion on the drop/matrix interface and between the interface and the bulk phases. Our accurate and robust numerical method features parallel computation and adaptive reconstruction of the finite element meshes describing the bulk phases and the interface.

Our results reveal a nontrivial dependence of the critical capillary number Ca_c, below which coalescence occurs, on the surface coverage of surfactant. Marangoni stresses inhibit coalescence and thus decrease Ca_c with respect to the clean-drop case. However, at large surfactant coverages close to the maximum packing of surfactant molecules, surfactant redistribution is prohibited (the surfactant is nearly incompressible) and thus the effect of Marangoni stresses is weakened, leading to an increase of Ca_c. In some cases, Ca_c at high coverages is even higher than in the clean-drop case: surfactant near-incompressibility renders the interface more rigid which results in less drop deformation and thus coalescence can occur at higher capillary number.

Finally, our results also reveal a nontrivial dependence of Ca_c on surfactant solubility in the bulk. At moderate surfactant concentration, diffusion in the bulk decreases surfactant redistribution on the interface and thus weakens Marangoni stresses resulting in higher Ca_c than in the insoluble case. However, when the surfactant bulk concentration is large, high adsorption fluxes maintain a higher surface concentration in equilibrium than for the insoluble case, thus resulting in larger drop deformation and in lower Ca_c.

This work is joint with H. Zhou, V. Cristini and C.W. Macosko.




Nov 7: Speaker: Olga Brezhneva   (IMA)

Title: Methods for finding singular solutions to nonlinear problems

Abstract: JOINT COMPUTATIONAL OPTIMIZATION SEMINAR LOCATED THIS DAY IN LINDH 409. We consider nonlinear problems for which the Jacobian matrix is singular at the solution. A new approach for finding singular solutions to nonlinear equations will be presented. The main idea of this approach is to replace the original nonlinear equation with a new one having the same solution as its nonsingular (regular) solution. To construct the new equation, the information about the first derivative of the original nonlinear problem is used. The results in this field find wide applications in bifurcation theory and optimization. An application of this approach to solving the Chandrasekhar $H$--equation and reformulations of the nonlinear complementarity problems will be illustrated.



Nov 14: Speaker: Michael Efroimsky   (IMA)

Title: Evolution of planetary orbits. Hidden symmetry and its relevance to orbit integration.

Abstract: We revisit the system of equations, which describes orbit evolution in the celestial mechanics, and point out a previously neglected aspect of these equations. This system was first written down by Lagrange, and later rewritten, in a canonical form, by Delaunay. A careful re-examination of the derivation of this system shows that, both in the Lagrange and Delaunay formulations, the orbit resides on a certain 9-dimensional submanifold of some 12-dimensional space We show that there exists a certain amount of freedom in choosing the submanifold. This choice is mathematically equivalent to gauge fixing. The freedom of choice (=freedom of gauge fixing) reveals a symmetry hiding behind the equations of celestial mechanics. This symmetry is analogous to the gauge invariance in electrodynamics (and reveals a fibre-bundle-type structure). Just as a convenient choice of gauge simplifies calculations in electrodynamics, the freedom of choice of the submanifold may, potentially, be used to create simpler schemes of orbit integration. On the other hand, the presence of this feature may be a previously unrecognised source of numerical error.



Nov 21: Speaker: Yu Chen   (NYU)

Title: Merging and splitting for scattering calculations

Abstract: We will show how merging subscatterers is useful for rapid solution of forward scattering problem, and present an approach to solution of the inverse problem based on splitting a scatterer into subscatterers. This splitting process partitions the ill-posed, nonlinear, inverse problem into a problem that is nonlinear but well-posed, and a problem that is ill-posed but linear.



Special Seminar Nov 26 (3pm, VINH 570): Speaker: Jay Gopalakrishnan   (U. Florida)

Title: Some iterative techniques for time harmonic Maxwell equations

Abstract: Time harmonic Maxwell equations in a lossless cavity lead to a second order partial differential equation for electric field involving a differential operator that is neither elliptic nor definite. A Galerkin method using Nedelec spaces can be employed to get approximate solutions numerically. In this talk, we will discuss the problem of efficient solution of the indefinite linear system arising from this method. Not all standard techniques work well for this application due to complications arising from the non-ellipticity and indefiniteness of the underlying differential operator. The suitability of some Schwarz methods and multigrid methods for the Maxwell application will be shown. Motivated by certain analytical techniques developed in this analysis of iterative techniques, some recent efforts have been made in simplifying the convergence analysis of finite element methods for time harmonic Maxwell equations. These will be briefly discussed.



Nov 28: Speaker:   ()

Title: No seminar. Thanksgiving Holiday.

Abstract:



Dec 5: Speaker: Bagisa Mukherjee   (Penn State, Scranton )

Title: Some Mathematical Issues in Liquid Crystal Flows

Abstract: We will discuss Ericksen's model of liquid crystals with its variable degree of orientation. The model consists of governing equations for the velocity field, the pressure, the director and the order parameter. The consitutive functions for the Leslie coefficients, derived from the molecular theory of Doi, play a crucial role in the modeling. One of the goals of the analysis is to examine the role of the order parameter in describing defects as well as in obtaining new regimes which cannot be predicted by the previous Leslie-Ericksen model.



Dec 12: Speaker: William Cooper   (U. Minn., Dept. Mech. Eng.)

Title: Revenue Management and Markov Decision Processes

Abstract: JOINT COMPUTATIONAL OPTIMIZATION SEMINAR (LINDH409). The talk will begin with a description of the basic revenue management problem, in which a firm must dynamically manage the availability of a variety of products that can be sold for different prices to different customer classes and that are constructed out of a common pool of scarce resources. One important example of such a problem is the fare class management problem faced by commercial airlines, where the resources are seats on flight legs, and the products are possible fare-class/itinerary ticket combinations. The remainder of the talk will describe progress in analyzing these issues using Markov decision processes.