4 septembre 2007

Ilse Ipsen
Department of Mathematics
North Carolina State University
Raleigh, NC, USA


Titre/Title: 

The Mathematics Behind Google's PageRank


Résumé/Abstract:

How does Google decide in which order to display web pages as the
result of a search?  A major ingredient in this decision is PageRank,
which is a score associated with every web page.  Increasing the
PageRank of a company's web page has become an important factor in
many marketing strategies.

PageRank only depends on the links among web pages, but not on their
content. Mathematically, PageRank can be viewed as the stationary
distribution of a stochastic matrix whose dimension is now in the
hundreds of billions. Hence the computation of PageRank is often
referred to as the world's largest matrix computation.

We discuss the mathematical problem behind PageRank, its numerical
computation, as well as efficient criteria that can guarantee correct
ranking of the PageRank scores. A round off error analysis
demonstrates the validity of the ranking criteria for matrices with
dimension up to 10^(14), and numerical experiments illustrate that
they can effectively rank the top PageRank scores.


Lieu/Location: McGill

11 septembre 2007

Abderrazak Ramadane
Département de Mathématiques et de statistique
Université de Montréal

Titre/Title: 

Numerical simulation of clogging and back flushing operation in a filter element in liquid filtration


Résumé/Abstract:

The objective of the study is to develop a numerical approach that could provide some
technical knowledge regarding the influence of the filter mesh, the optimisation of operating
conditions (flow rate, element design) and as well as back flushing operation for hydraulic
and filtration performances of automatic filters. The problem requires a multi-scale approach
from the particle /mesh unit cell size up to the scale of a complete filter, which implies the
consideration of several intermediate scales (section of a disk-type filter element, and disk-
type filter element). In this project, we concentrate on a section of a disk-type filter element.
First, we have to model the flow within the domain under consideration. Then, we have to develop
models of particle transport, clogging and  back flushing operation.

Lieu/Location: McGill

18 septembre 2007

Louis F. Rossi
Department of Mathematical Sciences
University of Delaware

Titre/Title: 

Re-projection: Mathematical and computational challenges for high order
vortex methods


Résumé/Abstract:

Vortex methods are numerical schemes for approximating solutions to
the Navier-Stokes equations using a linear combination of moving basis
functions to approximate the vorticity field of a fluid. Typically,
the basis function velocity is determined through a Biot-Savart
integral applied at the basis function centroid. Since vortex methods
are naturally adaptive, they are advantageous in flows dominated by
localized regions of vorticity such as jets, wakes and boundary
layers. While they have been successful in numerous engineering
application, the complexity of understanding grid-free methods make
their analysis a uniquely mathematical endeavor. One recent outcome
of rigorous analysis is an new naturally adaptive high order method
with basis functions that deform as they move according to flow
properties. This new class of methods is very unusual because the
basis functions do not move with the physical flow velocity at the
basis function centroid as is usually specified in vortex methods.
One of the leading edge research problems associated with high
accuracy methods of this type is how to re-project extremely deformed
elements onto a configuration of regular elements to prevent
catastrophic growth of interpolation errors. Recent progress in this
area brings together ideas from radial basis function interpolation,
pre-conditioners, image processing, and partial differential equations.


Lieu/Location: McGill

25 septembre 2007

Beatrice Riviere
Department of Mathematics
University of Pittsburgh

Titre/Title: 

On coupled flow and two-phase flow problems

Résumé/Abstract:


In this talk, we present high order numerical methods for solving multiphysics
problems. First, we investigate the  coupling of surface flow with subsurface flow. 
Surface flow is characterized either by Stokes or Navier-Stokes
equations whereas subsurface flow is characterized by Darcy equations.
Special interface conditions are considered between the subregions.
Existence of a weak solution is obtained. For the spatial discretization,
continuous finite element methods and discontinuous Galerkin, are used.
Optimal error estimates for both velocity and pressure are obtained.
Second, we consider numerical discretizations of the incompressible
two-phase flow problem.  In particular, adaptive simulations of a
sequential approach are shown. We also investigate a fully coupled
approach: error estimates are derived and numerical convergence in both
the mesh size and the polynomial degree are shown.  One advantage of the
coupled approach is that no slope limiters are needed.

Lieu/Location: McGill


2 octobre 2007

Jason Cooper
Department of Chemistry
Queen's University, Kingston, Ontario

Titre/Title: 

A pruned basis method for exact wavepacket propagation in scattering  
problems with many bound degrees of freedom

Résumé/Abstract:


The accurate simulation of molecular dynamics including the often  
central effects of quantum mechanics can be achieved through the  
numerical solution of the time-dependent Schrodinger equation.  
However, exponential growth of computational complexity for  
simulation of the full quantum system limits naive methods to systems  
with four or five degrees of freedom.  Larger systems are generally  
simulated using the simpler but approximate laws of classical  
mechanics or using combined quantum-classical approaches.

Wavepacket methods are a promising avenue for the extension of exact  
quantum methods to larger systems where the wavefunction stays well-
localized.  Most existing methods employ a dynamic, nonorthogonal set  
of basis functions to give a linear system of moderate size, but  
suffer a considerable computational penalty associated with the  
nonorthogonality of the basis.  Furthermore, the efficient use of  
these methods is generally limited to potential energies of a special  
form, rendering them inappropriate for the simulation of realistic  
molecular systems.

We have developed a new method in which a static "simultaneous  
diagonalization" basis, which is both well-localized and orthogonal,  
is pruned at each time step, leaving only a relatively small set of  
basis functions which follows the motion of the wavepacket.  We will  
describe the method and its characterization, with a focus on recent  
simulations of model problems in which a single atom is scattered  
from a molecule with many interacting internal degrees of freedom.

Lieu/Location: McGill


9 octobre 2007

Samuel Isaacson
Department of Mathematics
University of Utah

Titre/Title: 

Connections Between Several Stochastic Reaction Diffusion Methods for Modeling Biochemical Systems

Résumé/Abstract:

Recently three different, but fundamentally related stochastic reaction diffusion models have been used in modeling biochemical systems at the single-cell scale. Two of these methods create realizations of the stochastic process described by the spatially continuous Smoluchowski equation. One, Green's Function Reaction Dynamics, provides exact realizations of this process, while the other, based on a Brownian Dynamics approach, uses a time discretization. Both methods are spatially continuous. An alternative approach used by several authors is based on the reaction-diffusion master equation (RDME); an extension of the well-known chemical master equation. In the RDME approach space is discretized, while time is kept continuous. We will give an overview of each approach, with emphasis on the behavior of the RDME as the lattice spacing is varied. In particular, we will show that while molecules never react in the continuum limit that the lattice spacing approaches zero, for intermediate lattice spacings the RDME can be thought of as an asymptotic approximation to the Smoluchowski equation. Numerical results demonstrating the accuracy of the RDME in approximating the Smoluchowski equation for biologically relevant parameter regimes will be given.


Lieu/Location: McGill

16 octobre 2007

Tim Phillips
School of Mathematics
Cardiff University, UK

Titre/Title: 

The spectral element method with applications to non-Newtonian flows

Résumé/Abstract:

The spectral element method (SEM) is a high-order method for solving partial differential equations. This talk will begin by outlining some of the key components and advantages of the SEM before describing how it can be utilised to discretize the equations governing fluid flow. Important considerations include the compatibility of the discrete spaces, the treatment of the continuity equation and the efficient solution of the resulting linear systems of equations. The application of the SEM to some benchmark flow problems in non-Newtonian fluid mechanics will be described and the salient features of the schemes will be highlighted. The benchmark problems considered are the flonal centre manifold. Realization theorems in delay-differential equations ask the question of surjectivity of the centre manifold reduction mapping from classes of delay equations to possible classes of ordinary differential equations on the centre manifold. In particular, for a given class of ordinary differential equations on the centre manifold, one wants to find the class of delay equations with the least number of delays for which the centre manifold mapping is surjective. In this talk, I will discuss several realization theorems for linear and nonlinear systems of delay-differential equations and discuss applications to a few mathematical models.

Lieu/Location: McGill

 6 novembre 2007

Jayme De Luca
Department of Physics
UFSCAR, Brazil


Titre/Title: 

Variational structure and delay equation of the electromagnetic two-body problem


Résumé/Abstract:

We will discuss the state-dependent delay equation of the electromagnetic two-body problem. This is a mixed-type delay equation involving two singular denominators, one in the future light-cone and one in the past light-cone. These denominators spam some nontrivial qualitative dynamics with a stiff timescale and are the main hindrance for a numerical integration as an algebraic-differential neutral-delay equation. The problem has a variational structure that differs from the usual Hamiltonian principle of Galilean physics because it needs future and past stories. We shall discuss a natural use of this variational principle to solve the equations as a boundary-value problem and discuss a variational integrator. Last, the variational structure also includes the same denominators, but in a form easier to regularize than the mixed-type equations of motion.


Lieu/Location: McGill

13 novembre 2007

Frédéric Lesage
Département de génie électrique
École Polytechnique, Montréal

Titre/Title: 

Application de l'acquisition parcimonieuse à l'imagerie photo-acoustique

Résumé/Abstract:


Les dernières années ont vu émerger de nouvelles techniques permettant la compression quasi-optimale de signaux moyennant certaines contraintes. Plus récemment, cette compression s'est révélée être utile
dans le cadre non seulement du transfert des données mais de leur acquisition même. Dans ce séminaire nous montrerons que la minimisation d'un problème d'optimisation de type L1 permet d'obtenir avec une grande probabilité une représentation parcimonieuse d'images photo-acoustiques. L'application concrète de cet outil permet de développer des techniques d'acquisition d'images biomédicales à un rythme accru en prenant moins de mesures. Des exemples de mesures concrètes seront présentées


Lieu/Location: McGill

27 novembre 2007

Nathalie Lanson
Applied Mathematics Department
University of Waterloo

Titre

Méthodes sans grille renormalisées: analyse de convergence et applications

Title: 

Renormalized mesh-free schemes: convergence analysis and applications

Résumé

Les méthodes sans grille, appelées également méthodes particulaires, ont été développées pour
lapproximation des lois de conservation de la dynamique des fluides et des solides et sont maintenant
utilisées dans de nombreux domaines. Le principal avantage de ces méthodes réside dans leur capacité
à sappliquer avec succès à des simulations complexes impliquant de larges déformations, comme par
exemple des problèmes dimpacts. Les méthodes avec maillage tels que les éléments finis ne sont pas
adaptées à des problèmes tels que les impacts rapides à cause de la structure même du maillage. La
principale idée des méthodes sans grille est de traiter la partie convective des équations séparément
en la résolvant le long des courbes caractéristiques associées au champs de transport à laide de
particules numériques. Cet exposé introduit les méthodes sans grille renormalisées pour les lois de
conservation. Les caractéristiques analytiques et numériques de ces méthodes sont présentées ainsi
que leur application à des multimatériaux solides, avec modèle dendommagement et à des problèmes
de dynamique des fluides.

Abstract:

Meshfree methods, also referred to as particle methods, have been recently developed for the
approximation of conservation laws in many fields of hydrodynamics and solid dynamics and are now
used in a wide range of applications. The main advantage of these methods lies in their ability to
handle complex situations involving highly distorted systems, such as crash or impact problems. In
fact, mesh-based methods such as Finite Element Methods are not well suited for problems such as
high velocity impacts, the main drawback being the structural aspect of the mesh. Meshfree methods
offer a potential solution to these difficulties. The main idea of these methods is to treat separately
the convective part of the equations by solving it along the characteristic curves associated with the
transport field. In this talk I will present the so-called renormalized meshfree schemes for conservation
laws, discussing both their analytical and numerical aspects. I will then illustrate the methods with
an application to multimaterial solid systems, including damage models, as well as to fluid dynamics
problems.


Lieu/Location: McGill

4 décembre 2007

Lennaert van Veen
Department of Mathematics and Statistics
Concordia University

Titre/Title: 

Connecting orbits: the missing link in the theory of bursting shear flows

Résumé/Abstract: 

Shear flows, such as flow in a pipe or along a wall, display the phenomenon of bursting. Turbulent flow profiles develop and decay spontaneously at sufficiently high flow rates, even though linear theory predicts that the flow remains smooth. These turbulent bursts increase the drag, with important consequences in many practical applications. Recently, a new theory of bursting was formulated. This theory is based entirely on concepts of nonlinear dynamics, most importantly on heteroclinic connections. Although the qualitative predictions of this theory agree well with numerical and laboratory experiments, no direct computation of such connecting orbits has been performed for realistic models of shear flow. In this talk I will present some preliminary computations on low-order models, as well as ideas for full-scale computations which might verify the new theory of bursting.


Lieu/Location: McGill

19 décembre 2007

Giuseppe Geymonat:
Pôle Mathématiques, Informatique, Physique, Structures et Systèmes
Laboratoire Mécanique et Génie Civil (LMGC)
Département de Mécanique
CC055, Place Eugène Bataillon
34 095 Montpellier Cedex 5, France

Titre/Title: 

Saint Venant, Beltrami, Volterra and the elasticity complex in a Sobolev space framework for Lipschitz domains.

Résumé/Abstract: 

The characterization of smooth symmetric second order tensor fields that are strain fields was obtained by Saint Venant (1864, necessary conditions)), Beltrami (1886, sufficiency for simply connected domains) and V. Volterra (1906, multiply connected domains). Their results can be interpreted in geometric language in terms of the elasticity complex.

We study the extension of these results to Sobolev spaces and general Lipschitz domains.


Lieu/Location: UdeM

19 décembre 2007

Françoise Krasucki
Institut de Mathématiques et de Modélisation de Montpellier
Université Montpellier 2
Case Courrier 051
Place Eugène Bataillon
34095 MONTPELLIER Cedex
France

Titre/Title: 

Highly contrasted elastic multi-materials: multi-scale variational modelings and adapted domain decomposition algorithms

Résumé/Abstract: 


Multi-materials are obtained by joining together materials with different properties. In this talk two elastic materials are joined together by a third elastic material occupying an " ε-small" layer and having elastic properties of order ε^p with respect to the two other materials. Two significant situations are p=1 (soft adhesive) and p=-1 (strengthening layer). For both situations one can characterize the limit for ε→0 where the layer becomes an interface supporting suitable transmission conditions. In the linear case a domain decomposition algorithm is then natural and one can study the convergence of a GMRES iterative scheme.

Lieu/Location: UdeM