sysnetwebmailadmin

Upcoming Seminars

Seminars are held Tuesdays and Thursdays in POB 6.304 from 3:30-5:00 pm, unless otherwise noted. Speakers include scientists, researchers, visiting scholars, potential faculty, and ICES/UT Faculty or staff. Everyone is welcome to attend. Refreshments are served at 3:15 pm.

 

ICES Seminar
Thursday, Feb 11, 2016 from 3:30PM to 5PM
POB 6.304

A generalized multiscale model reduction technique for heterogeneous problems
by Yalchin Efendiev

Professor of Mathematics, Texas A&M University

In this talk, I will discuss multiscale model reduction techniques for problems in heterogeneous media. I will describe a framework for constructing local (space-time) reduced order models for problems with multiple scales and high contrast. I will focus on a recently proposed method, Generalized Multiscale Finite Element Method, that systematically constructs local multiscale finite element basis functions on a coarse grid, which is much larger than the underlying resolved fine grid. The multiscale basis functions take into account the fine-scale information of the resolved solution space via careful choices of local snapshot spaces and local spectral decompositions. I will discuss the issues related to the construction of multiscale basis functions, main ingredients of the method, and a number of applications. These methods are intended for multiscale problems without scale separation and high contrast.

Hosted by Todd Arbogast





 

ICES Seminars-Computational Medicine Spring Seminar Series
Tuesday, Feb 16, 2016 from 3:30PM to 5PM
POB 6.304

Modeling protein-ligand interactions
by Pengyu Ren

UT Austin

Molecular recognition between biomolecules such as protein-ligand is essential for biological processes and accurate prediction protein-ligand interaction is needed for accelerating the discovery of new drugs and biological probes. Many factors, including shape complementary, electrostatic interaction and entropic effect, are responsible for the specificity and selectivity in molecular recognition. Physics-based modeling is useful for a wide range of applications from high-throughput lead screening to understanding physical principles of biomolecular structure and function. To achieve its potential for quantitative prediction and chemical accuracy, we have been working on improving the understanding and modeling of the fundamental interatomic forces, as well as developing efficient simulation algorithms. I will present a few examples, our current developments and future prospects in these areas throughout this talk.

Bio:
Dr. Ren is Associate Professor of Biomedical Engineering and William J. Murray Jr. Fellow #4 in Engineering at UT Austin. He received his Ph.D. in Chemical Engineering from the University of Cincinnati followed by a postdoctoral training in the Department of Biochemistry and Molecular Biophysics at Washington University Medical School. He was a recipient of Hewlett-Packard Outstanding Junior Faculty Award from the American Chemical Society and was elected to the AIMBE College of Fellows in 2015. His research focuses on understanding the physical driving forces underlying biomolecules and developing physics-based computational methods to accelerate the discovery of new medicines and materials.

Hosted by Michael Sacks and George Biros





 

ICES Seminar
Thursday, Feb 18, 2016 from 3:30PM to 5PM
POB 6.304

Multigrid at Scale?
by Mark Ainsworth

Professor, Division of Applied Mathematics,Brown University

Multigrid and multilevel iterative algorithms are often the method of choice for the solution of large-scale systems of linear equations arising from discretisation of partial differential equations using finite element or finite difference methods. The method involves a number of components including smoothing relaxation, coarse grid solve, prolongation and restriction operators between grids in the multilevel hierarchy, and the convergence behavior of the method has been extensively analysed in the context of standard computer architectures. However, comparatively little is known about the resilience or fault-tolerance of the algorithm on next generation hardware architectures which are expected to suffer from frequent data corruption and hardware failures. We will address this issue and present some of the results of our recent work showing that the issue is anything but clear.

This is joint work with Christian Glusa (Brown University).

Bio
Mark Ainsworth obtained his PhD in Mathematics at Durham University in the United Kingdom. Prior to moving to Brown, he held the 1825 Chair in Mathematics at Strathclyde University and was Director of
NAIS , a joint centre between the Universities of Edinburgh, Heriot-Watt and Strathclyde, and Edinburgh Parallel Computing Centre to develop UK capacity in high performance computing and numerical analysis.
Learn More About Mark Ainsworth

Hosted by Leszek Demkowicz





 

ICES Seminar
Tuesday, Feb 23, 2016 from 3:30PM to 5PM
POB 6.304

Data assimilation by the Ensemble Kalman filter and other particle filters — why are 50 ensemble members enough?
by Matthias Morzfeld

Professor, Department of Mathematics, University of Arizona.

Suppose you have a mathematical model for the weather and that you want to use it to make a forecast. If the model calls for rain but you wake up to sunshine, then you should recalibrate your model to this observation before you make a prediction for the next day. This is an example of Bayesian inference, also called “data assimilation” in geophysics. The ensemble Kalman filter (EnKF) is used routinely and daily in numerical weather prediction (NWP) for solving such a data assimilation problem. Particle filters (PF) are sequential Monte Carlo methods and are in principle applicable to this problem as well. However, PF are never used in operational NWP because a linear analysis shows that the computational requirements of PF scale exponentially with the dimension of the problem. In this talk I will show that a similar analysis applies to EnKF, i.e., “in theory” EnKF’s computational requirements also scale poorly with the dimension. Thus, there is a dramatic contradiction between theory and practice. I will explain that this contradiction arises from different assessments of errors (local vs. global). I will also offer an explanation of why EnKF can work reliably with a very small ensemble (typically 50), and what these findings imply for the applicability of PF to high-dimensional estimation problems.

Hosted by Tan Bui-Thanh





 

ICES Seminar
Tuesday, Feb 23, 2016 from 1:30PM to 3PM
POB 2.302 (AVAYA)

Numerical Study of a Monolithic Fluid-Structure Formulation
by Olivier Pironneau

Sorbonne Universites, UPMC Paris VI, Laboratoire Jacques-Louis Lions

The conservation laws of continuum mechanics are naturally written in an Eulerian frame where the difference between a fluid and a solid is only in the expression of the stress tensors, usually with Newton's hypothesis for the fluids and with Helmholtz potentials of energy for hyperelastic solids.

There are currently two favored approaches to Fluid Structured Interactions both working with the equations for the solid in the initial domain; one uses an ALE formulation for the fluid and the other matches the fluid-structure interfaces using Lagrange multipliers and the fictitious domain method.

By contrast the proposed formulation works in the frame of physically deformed solid and propose a discretization where the structures have large displacements computed in the deformed domain together with the fluid in the same, in a monolithic formulation where the velocity of both are computed all at once by a semi-implicit in time plus finite element method.

Besides the simplicity of the formulation the advantage is a single algorithm for a variety of problems including multi-fluid, free boundary and FSI. On the flipped side it needs a robust mesh generator.

Stability will be studied showing were are the difficulties and why we were able to show convergence of an earlier monolithic algorithm for a fluid within a shell restricted to small displacements.

Numerical examples will be given.

Hosted by Irene Gamba





 

ICES Seminar
Tuesday, Feb 23, 2016 from 1:30PM to 3PM
POB 2.302 (AVAYA)

Event Update Notification: NOTE: Different TIME - Change in date and location

Numerical Study of a Monolithic Fluid-Structure Formulation - Note Different TIME.
by Olivier Pironneau

Professor, Sorbonne Universite, Laboratoire Jacques-Louis Lions, UPMC (Paris VI)

The conservation laws of continuum mechanics are naturally written in an Eulerian frame where the difference between a fluid and a solid is only in the expression of the stress tensors, usually with Newton's hypothesis for the fluids and with Helmholtz potentials of energy for hyperelastic solids.

There are currently two favored approaches to Fluid Structured Interactions both working with the equations for the solid in the initial domain; one uses an ALE formulation for the fluid and the other matches the fluid-structure interfaces using Lagrange multipliers and the fictitious domain method.

By contrast the proposed formulation works in the frame of physically deformed solid and propose a discretization where the structures have large displacements computed in the deformed domain together with the fluid in the same, in a monolithic formulation where the velocity of both are computed all at once by a semi-implicit in time plus finite element method.

Besides the simplicity of the formulation the advantage is a single algorithm for a variety of problems including multi-fluid, free boundary and FSI. On the flipped side it needs a robust mesh generator.

Stability will be studied showing were are the difficulties and why we were able to show convergence of an earlier monolithic algorithm for a fluid within a shell restricted to small displacements.

Numerical examples will be given.

Hosted by Irene Gamba





 

ICES Seminar
Thursday, Mar 3, 2016 from 1:30PM to 3PM
POB 6.304

Event Update Notification: NOTE: Special Time for Seminar

COFFE, a Conservative Field Finite Element solver for PDEs
by Ryan S. Glasby

Professor, University of Tennessee at Chattanooga

HPCMP CREATETM-AV Conservative Field Finite Element (COFFE) is envisioned as a numerical solver for partial differential equations designed to robustly solve a broad range of problems. To realize this goal, COFFE employs a modular software design that separates discretization, physics, parallelization, and linear algebra into distinct components. These components are developed with modern software engineering principles and are rigorously unit tested. Within COFFE, the Streamline Upwind/Petrov-Galerkin finite-element method is employed to discretize the compressible Reynolds-Averaged Navier-Stokes equations. This equation set is highly nonlinear, and an efficient and robust nonlinear solver is utilized to drive complex aerospace flow problems to a steady-state solution. The mathematics and philosophy of the methodology that makes up COFFE are presented as well as solutions of a variety of two and three dimensional test cases.

Hosted by Mary Wheeler





 

ICES Seminar
Thursday, Mar 10, 2016 from 3:30PM to 5PM
POB 6.304

On the Solution of Elliptic Partial Differential Equations on Regions with Corners
by Vladimir Rokhlin

Professor, Yale

Solution of elliptic partial differential equations on regions with non-smooth boundaries (edges, corners, etc.) is a notoriously refractory problem, especially when high accuracy is desired. In this talk, I observe that when the problems are formulated as boundary integral equations of classical potential theory, the solutions (of the integral equations) in the vicinity of corners can be represented by a series of elementary functions. In addition to being analytically perspicuous, the resulting expressions lend themselves to the construction of discretization schemes displaying convergence of arbitrarily high order. For most practical purposes, the loss of accuracy associated with the presence of corners (edges, etc.) disappears entirely. The results are illustrated by a number of numerical examples.

Hosted by George Biros and Greg Rodin





 

ICES Seminar
Thursday, Apr 21, 2016 from 3:30PM to 5PM
POB 6.304

When does the finite element method (not) converge?
by Vaclav Kucera

Department of Numerical Mathematics, Faculty of Mathematics and Physics, Charles University in Prague

The piecewise linear finite element method (FEM) exhibits O(h) convergence on triangulations satisfying the maximum angle condition. This is a sufficient condition, however it can be easily shown that it is not necessary. In fact the FEM meshes can contain an arbitrary number of arbitrarily 'bad' elements while still exhibiting O(h) convergence in the H^1 (semi)norm. In this talk we generalize known sufficient conditions for convergence of various orders by modifying the Lagrange interpolation procedure on degenerating elements. Moreover, we derive the first nontrivial necessary condition for various types of FEM convergence. As a byproduct, this allows us to construct simple counterexamples, systems of triangulations on which the FEM cannot have O(h) convergence or cannot converge at all. Up to now, only one such counterexample was known in the literature. Although a necessary and sufficient condition remains unknown, the gap between the derived conditions is small in special cases, giving hope for future work.

Hosted by Ivo Babuska





 

ICES Seminar - ICES Student Forum
Friday, May 6, 2016 from 11AM to 12PM
POB 6.304

A Large-Scale Ensemble Transform for Bayesian Inverse Problems Governed by PDEs
by Aaron Myers

ICES, The University of Texas at Austin

We present an ensemble-based method for prior samples to posterior ones. This method avoids Markov chain simulation and hence discards expensive work when a sample is rejected. The idea is to cast the problem of finding posterior samples into a large-scale linear programming problem for which efficient and scalable solver can be developed. Large-scale numerical results will be presented to demonstrate the capability of the method.

Hosted by Teresa Portone and Travis Sanders