- Workshop Advances in Computational Science
- 21.06.2011 - 01.07.2011
- Institute of Computational Science - Lugano
- ICS Past Events
Event VideosItem not found.
Advances in Computational Science - Models, Methods and Applications
The Institute of Computational Science (ICS) of the Faculty of Informatics at USI is pleased to announce the Workshop on Advances in Computational Science - Models, Methods and Applications that will take place in two sessions on Monday, June 21st and Thursday July 1st at USI Lugano.
During the workshop, we will discuss a wide variety of high performance methods for advanced simulations in engineering, life sciences and other application areas. The focus of this workshop will be put on mathematical and methodological questions, as well as on the capability of the methods to be applicable on modern high performance computer systems.
Thursday, July 1st, A21 - Red Building, USI, Via G. Buffi 13, Lugano
|08:00 - 08:50||Maddalena Venturoli - Time-scale challenges in molecular dynamics simulations|
|08:50 - 09:40||Simone Melchionna - Multi-Scale Response of Soft Matter to Flow Conditions: Insights from Computer Simulation|
|09:40 - 10:30||Frank Noè - Scientific computing in molecular biology|
|10:30 - 10.45||Break|
|10:45 - 11:35||Luca Maragliano - Novel methods for potential of mean force and reaction paths calculations in Molecular Dynamics simulations of bio-molecules|
|11:35 - 12:15||Igor Pivkin - Multiscale modeling of biological flows|
|12:15 - 13:05||Masha Sosonkina - Multilevel methods for solving sparse linear systems and their integration into high-performance simulations|
MADDALENA VENTUROLI - Time-scale challenges in molecular dynamics simulations
Abstract: In recent years, molecular dynamics (MD) simulations have become a tool of choice in chemistry, material sciences and molecular biology. MD allows one to simulate the motion of the individuals atoms of complex systems involved in complicated processes and thereby obtain a detailed information about the dynamics of these systems. MD can be used to analyze from first principles chemical processes such as allosteric transitions of biomolecules, nucleation events during phase transformation, fracture and crack propagation in solids, defect dynamics in crystals, etc. in a way which is potentially more precise and less expensive than by doing actual experiments and without having to rely on effective macroscopic models. To achieve its full potential, however, MD must be integrated as a part in multiscale algorithms. This is the only possibility to be able to tackle problems involving systems of realistic size made of an enormously large number of atoms, and to simulate their dynamics on the very long time-scales of actual interest. In my talk, I will discuss challenges related to this second issue and show how to access time-scales which are much too long for direct simulations but on which rare but important reactive events arise. In particular, I will discuss techniques to investigate mechanism, rate and energetics of rare events and illustrate their recent application to the understanding at the molecular level the unfolding mechanism of the protein ubiquitin under forces comparable to those used in atomic force microscopy experiments.
Biography: Maddalena Venturoli is a postdoctoral research fellow at the Courant Intitute of Mathematical Sciences at New York University. She holds a Ph.D. in Physics from the University of Amsterdam, The Netherlands and a Laurea in Physics from the University of Roma "La Sapienza'", Italy. She has been a postdoctoral research fellow at CECAM (Centre Européen de Calcul Atomique et Moléculaire) where she was also project manager for a Marie Curie Early Stage Training program. From 2002 to 2006 she has been a postdoctoral research fellow in the EPSRC RealityGrid project, the U.K. government's initiative on e-Science during which she worked in collaboration between the Center for Computational Science at UCL in London and Schlumberger Cambridge Research. Her research interests include the development and application of advanced simulation techniques to the study of activated processes in phase transitions and conformational changes of biological macromolecules. She has also contributed to the development of a mesoscopic model for biological membranes with embedded proteins, and to the application of large scale lattice-Boltzmann simulations to the study of flow in porous media.
SIMONE MELCHIONNA - Multi-Scale Response of Soft Matter to Flow Conditions: Insights from Computer Simulation
Abstract: A large class of soft matter systems displays multi-scale behavior and emerging complexity. The presence of multiple concurring phenomena spans several temporal and spatial scales and produces non-universal and multi-functional behavior. This is probably the main reason why soft matter is so important in sustaining the long-range organization of life, with patterns that extend the size of the elementary components by orders of magnitude.
In this talk, I will describe two prototypical examples of such behavior, namely the translocation of biopolymers in narrow pores and non-Newtonian flows in the cardiovascular circulatory system. In my recent work, I have devised and employed a concurrent, scale-adaptive computational approach to account for this behavior, overcoming the limitations of conventional numerical strategies. The two showcases illustrate how the interplay between scales has profound implications on our understanding of physical processes involving soft matter in flow conditions. Results on dense suspensions of red blood cells is probably the most striking example of how local structural and dynamical heterogeneities reflect on the macroscopic rheological properties. Finally, I will discuss prospects to tackle cellular dynamics and suspended macromolecules embedded in fluid dynamic environments.
Biography: Dr. Simone Melchionna is a Senior Research Scientist at the National Institute of Condensed Matter, Italy, and Collaborateur Scientifique at the Institute of Materials at EPFL. He is an expert on computational models of complex and biological systems, confined fluids, proteins and DNA. He obtained his master degree in physics and a PhD in chemistry in 1996 at the University of Rome. After some initial studies on organic molecules in Rome, Dr. Melchionna spent three years at the University of Cambridge, working on simple fluids and water under confinement. Subsequently he worked on problems in soft matter, membranes, models of ionic transport and proteins under extreme conditions. More recently, he spent two years at Harvard University working on DNA translocation and multi-scale cardiovascular blood flows.
FRANK NOÈ - Scientific computing in molecular biology
Abstract: Computational molecular biology involves particle simulation methods that pose extreme challenges to high performance computing. In particular, molecular dynamics simulation is limited by the sampling problem which still prevents us from straightforwardly simulating macromolecular processes of biological interest. This calls for efficient and optimal computation methods which allow the desired quantities to be calculated with available computing equipment. Here I will give a brief overview of our contributions to such methods and give examples from protein folding and cellular dynamics:
- optimal estimation of stationary properties from nonstationary sampling methods such as metadynamics and umbrella sampling.
- calculating kinetic properties of molecular systems based on Markov models
- from molecules to cells: a perspective of a full cell model of rod vision phototransduction
- 1996-2002 Study of electrical engineering and computer science in Germany (Stuttgart) and Ireland (Cork)
- 2002-2006 PhD in computer science at IWR Heidelberg, Germany
- 2006-2007 Postdoctoral fellow, BIOMS program, Heidelberg, Germany
- 2007-now Independent junior group leader, MATHEON, Berlin, Germany
- 2009-2010 Substitute professor (W2) at IWR Heidelberg for the chair "scientific computing in the biosciences"
LUCA MARAGLIANO - Novel methods for potential of mean force and reaction paths calculations in Molecular Dynamics simulations of bio-molecules
Abstract: In this talk I will present new methods to reconstruct multi-dimensional potential of mean force (PMF) landscapes and/or reaction paths on them using Molecular Dynamics simulations. Standard ways to perform such calculations are based on extensive sampling, and for this reason are limited to cases in which the number of collective variables (or reaction coordinates) is small. We have developed an alternative approach where we compute locally, in the unknown PMF space, some suitable quantities and then use these data to explore and reconstruct the PMF, or to determine reaction paths.
Such procedure allows us to considerably increase the number of collective variables used.
I will describe the methodology in detail and illustrate its application on examples as the conformational transition of the Voltage Sensor Domain in a potassium channel and the diffusion of a CO molecule in myoglobin.
Biography: I graduated in Physics in 2000 at the University of Rome La Sapienza, Italy, under the supervision of Prof. Giovanni Ciccotti, with a thesis entitled "Stability of Thermophilic proteins by Molecular Dynamics". I obtained the Ph.D. in Physics in 2004 from the University of Modena, Italy, with a thesis on "Methods and applications of Molecular Dynamics simulations to bio-molecules", working with Profs. Mauro Ferrario and Giovanni Ciccotti.
From 2005 to the end of 2007 I was post-doc at the Courant Institute of Mathematical Sciences, New York University, were I have worked with Prof. Eric Vanden-Eijnden. Since January 2008 I am post-doc at the University of Chicago, Department of Biochemistry and Molecular Biology, working in the group of Prof. Benoit Roux.
IGOR PIVKIN - Multiscale modeling of biological flows
Abstract: The increasing power of supercomputers -- from today's petaflop to exaflop by the year 2018 -- can play a catalytic role in advancing biology at all scales. However, new approaches are required for simulations of the multiscale phenomena encountered in biological systems. In this talk, we will present multiscale modeling approaches for continuum (PDE-based) and atomistic (dissipative particle dynamics) modeling with specific applications to the vascular system and microcirculation. Examples of simulations of arterial flows and blood cells will be presented along with systematic validation based on microfluidic experiments in order to address specific pathologies, e.g. malaria.
Biography: Igor V. Pivkin received his B.Sc. and M.Sc. degrees in Mathematics from Novosibirsk State University, Russia, M.Sc. degree in Computer Science and Ph.D. in Applied Mathematics from Brown University, USA. He is currently a Postdoctoral Associate in the Department of Materials Science and Engineering at Massachusetts Institute of Technology, USA. His research interests lie in the area of multiscale/multiphysics modeling, corresponding numerical methods and parallel large-scale simulations of biological and physical systems.
Specific areas include biophysics, cellular and molecular biomechanics, stochastic multiscale modeling, and coarse-grained molecular simulations.
MASHA SOSONKINA - Multilevel methods for solving sparse linear systems and their integration into high-performance simulations
Abstract: Sparse matrix computations are ubiquitous in high-performance scientific applications. An efficient large-scale linear system solver may drastically improve the overall performance of such applications. It is difficult, however, to navigate through the available solver packages and to tune their performance to the problem and architecture at hand. This is mainly due to the plethora of interfaces and adaptation tools, each of which may require application modifications. For a high-performance scientific application, it may be desirable to select dynamically a solution method based on the progress made in the course of computation. This appears possible if sparse matrix computations are designed as components, which provide various levels of encapsulation and enhance solver reuse.
In this talk, we overview several efforts to abstract the minimal common set of interfaces among solver packages leading to their integration with applications. The benefits of sparse matrix kernel compositions are considered in the parallel SPARSKIT library, which becomes easily adaptable and extensible with novel techniques.
We present variants of Algebraic Recursive Multilevel (ARMS) and Schur Complement methods used in parallel preconditioning of large-scale sparse linear systems that generally exhibit poor iterative convergence. Then, we employ these techniques in the numerical experiments with the systems arising in multi-particle simulations and compare their parallel performance and iterative convergence.
Biography: Masha Sosonkina has received her B.S. and M.S. degrees in Applied Mathematics from Kiev National University in Ukraine. She has graduated from Virginia Tech in 1997 with a Ph.D. degree in Computer Science and Applications. Upon graduation, she was appointed as Assistant professor at the University of Minnesota Duluth, where she became an Associate professor in 2002. Since 2003, Dr. Sosonkina is a scientist at the U.S. Department of Energy Ames Laboratory and an adjunct Associate professor at Iowa State University. She has also been a visiting research scientist at the Minnesota Supercomputing Institute in 1998 and at CERFACS and INRIA French research centers in 2000 and 2005, respectively. Dr. Sosonkina's research interests include high-performance computing and applications, petascale and many-core systems, parallel numerical algorithms, performance analysis, and adaptive algorithms.
Monday, June 21st, Auditorium USI, Via G. Buffi 13, Lugano
|08:15 - 09:05||Alexandre Caboussat - A Numerical Framework for the Simulation of Two-Phases Flow and Some Applications|
|09:05 - 09:55||Patrick Ilg - Hybrid and Multiscale Simulation Approaches to Complex Fluids|
|09:55 - 10:45||Gregor Gassner - A Space and Time Adaptive Discontinuous Galerkin Framework for the High Performance Computation of Multiscale Problems|
|11:00 - 11:50||Michael Bronstein - Computational metric geometry|
|11:50 - 12:40||Carsten Burstedde - Scalable adaptive mesh refinement for simulation and inverse problems in solid earth geophysics|
ALEXANDRE CABOUSSAT - A Numerical Framework for the Simulation of Two-Phases Flow and Some Applications
Abstract: The numerical simulation of two-phases flow has many applications in science and engineering, for instance in mold casting, glacier modeling, aluminum production, or bubbles simulation.
We present a numerical framework for the simulation of three-dimensional two-phases flow with free surfaces, involving an incompressible liquid and a compressible gas. It allows to solve the incompressible Navier-Stokes equations only in the liquid domain, while taking into account the influence of the surrounding gas.
We use an implicit time splitting scheme to decouple the physical phenomena.
It couples a method of characteristics for the solution of advection problems and an implicit scheme for the solution of a time dependent Stokes problem. Algorithms for the detection of gas bubbles and computation of surface tension effects are added.
A two-grids method that couples a structured grid of small cells and a finite element mesh of tetrahedrons is used. An interface tracking technique involving local adaptive mesh refinement around the interfaces is advocated to obtain a more accurate approximation of the interfaces and reduce numerical diffusion. Finite element techniques are used for the discretization of the Stokes problems.
Numerical experiments in a large variety of physical problems illustrate the efficiency and robustness of the method.
We present several applications ranging from mold casting in complex geometries, bubbles in aluminum electrolysis or glacier simulations. We conclude by presenting future directions of research, going from coupled multiphysics problems (such as solidification) to particles flow.
Biography: Alexandre Caboussat has been an Assistant Professor at the University of Houston since 2005. He completed his Ph.D. in numerical analysis and scientific computing at the Ecole Polytechnique Federale de Lausanne (EPFL) in 2003. His Ph.D. thesis on the 'Analysis and numerical simulation of free surface flows' has been awarded the EPFL doctorate award in 2003.
He then obtained an individual fellowship from the Swiss National Science Foundation in 2004 to go to the University of Houston and work on the modeling and numerical simulation of the thermodynamics and dynamics of aerosol particles. He continued working on free surfaces flow, first in collaboration with the Los Alamos National Laboratory, and, since 2009, with Ycoor systems (a spin-off from the EPFL working on scientific software) and Rio-Tinto Alcan.
More recently, he has been working on non-smooth problems arising in continuum mechanics and on numerical methods for fully or implicitly nonlinear elliptic equations (such as the Monge-Ampere equation). This project is currently funded by the US National Science Foundation.
PATRICK ILG - Hybrid and Multiscale Simulation Approaches to Complex Fluids
Abstract: Complex fluids and soft matter constitute an important class of materials, covering among others polymer solutions and melts, colloids, liquid crystals, as well as supercooled liquids and amorphous systems. The presence of many length- and time scales in soft matter systems is responsible for their very interesting macroscopic properties, but also poses serious challenges on modeling and simulation approaches.
Here, I want to show how methods of nonequilibrium statistical mechanics and computational physics can be used to design new simulation methods that overcome the limitations of traditional brute force approaches to such systems.
One example is a hybrid simulation scheme, that combines a macroscopic finite element method with a mesoscopic stochastic simulation of liquid crystalline dynamics. Since the hybrid simulations do not rely on closure approximations, the simulation results are more reliable compared to traditional approaches. Including thermal fluctuations is an additional benefit of the hybrid scheme which is especially advantageous in microfluidics. Another example is a systematic multiscale approach that we have developed very recently in order to bridge the enormous gap in length- and time scales between microscopic and macroscopic quantities. We implemented this method in an efficient and thermodynamically consistent manner. The method has been tested successfully for flowing polymer melts and can now be applied to a wide variety of other problems.
Biography: Patrick Ilg has been a senior researcher at ETH Zurich, Polymer Physics group since 2006. Before that he was at University of Lyon with a fellowship from the Alexander von Humboldt Foundation. From 2003 to 2005 he was assistant researcher at Technical University of Berlin.
GREGOR GASSNER - A Space and Time Adaptive Discontinuous Galerkin Framework for the High Performance Computation of Multiscale Problems
Abstract: Discontinuous Galerkin methods have gained considerable attention in recent years due to their low dispersion and dissipation errors, their capability of handling complex domains and their proven potential for massive parallel computations. Furthermore, for problems with multiple spatial and temporal scales the Discontinuous Galerkin framework offers high flexibility in a natural way by allowing to adjust the resolution adaptively. In spite of significant advances over the last decade, Discontinuous Galerkin methods still suffer from being computationally too expensive when compared to more traditional methods such as Finite Difference and Finite Volume methods. In this talk, we will discuss several recent progresses to enhance the computational efficiency, such as a spatial integration based on polymorphic nodal elements and a Runge-Kutta based predictor-corrector time integration with time accurate local time-stepping. We will demonstrate the potential of this framework with applications to selected fluid dynamics problems.
Biography: Gregor Gassner started to study Astrophysics at University of Innsbruck (Austria) in 1998. After one year, he changed to Aerospace Engineering at the University of Stuttgart (Germany). In 2002 he got a special permit from the University to study Aerospace Engineering and Mathematics in parallel and was awarded with the diplomas in 2004 and 2005. He obtained his PhD in Aerospace Engineering from the University of Stuttgart in 2009, where he worked in the group of Prof. C.-D. Munz on the topic of Discontinuous Galerkin methods for compressible Navier-Stokes equations. In November 2009 he was promoted to Akademischer Rat auf Zeit (Lecturer) at the Institute of Aerodynamics and Gasdynamics. His main research interests are the development of mathematical models and their application to large scale engineering problems using high performance computing.
MICHAEL BRONSTEIN - Computational metric geometry
Abstract: Metric geometry is a branch of theoretical mathematics formalizing and studying the notions of "distance" and "similarity". In the past decades, this field has experienced revolutionary developments (recognized in 2009 by the Abel prize), remaining, however, practically unnoticed in applied sciences. Since many problems in computational sciences and engineering often have some underlying notion of similarity, metric geometry is a natural common denominator allowing to address such problems.
In this talk, I will showcase a few problems that can be addressed from the metric perspective. The first problem is the analysis of deformable shapes, arising in a broad variety of applications on all levels from nano to macro. Modeling shapes as metric spaces allows elegantly overcoming some fundamental difficulties in shape representation and analysis on the one hand, and provides for practical computational methods on the other. In this framework, shape similarity is related to the Gromov-Hausdorff distance and multidimensional scaling (MDS) problems, and can be efficiently computed using multi-resolution and multi-grid optimization algorithms.
As additional examples of applications, time permitting, I will show problems from the field of "Internet vision", a recent field dealing with the analysis of large-scale web repositories of visual data such as community photo albums, video blogs, and social networks. In this context, I will discuss metric geometry approaches to similarity learning, comparison of multimodal data, and information retrieval.
Biography: Michael Bronstein received his Ph.D. (with distinction) in Computer Science from the Technion (Israel Institute of Technology) in 2007. He spent the post-doctoral period in California at Stanford university and as a technology executive in the industry. His main research interests are theoretical and computational methods in metric geometry and their application to problems in computer vision, pattern recognition, computer graphics, medical imaging, and machine learning.
Highlights of his research were featured in CNN, SIAM News, Wired, and in the Abel lecture given in Oslo in honor of the 2009 Abel Prize laureate Mikhail Gromov. To date, Michael has authored a book, over 60 publications in leading journals and conferences, and over a dozen of patents. He was the chair of the Workshop on Non-rigid shapes and deformable image alignment (NORDIA) in 2008-2010, the organizer of a symposium on shape analysis in the SIAM Conference on Imaging Sciences in 2010, and has served on review and program committees of many international conferences and workshops. In addition to academic activities, Michael is involved as an advisor in several companies licensing the technology he has developed.
CARSTEN BURSTEDDE - Scalable adaptive mesh refinement for simulation and inverse problems in solid earth geophysics
Abstract: Many problems in solid earth geophysics are characterized by multiscale dynamic phenomena, which complicates the numerical solution of the governing partial differential equations (PDEs). One approach to overcoming the tyranny of scales is adaptive mesh refinement (AMR), which locally and dynamically adapts the mesh to resolve spatio-temporal scales and features of interest.
While AMR promises to help overcome the challenges inherent in modeling multiscale problems, the benefits are difficult to achieve in practice, particularly on highly parallel supercomputers. Due to complex mesh topology and communication patterns, and frequent data exchange and redistribution, scaling dynamic AMR to tens of thousands of processors has long been considered a challenge. Additional numerical difficulties are encountered when extending parallel AMR to high-order-accurate, complex-geometry-respecting discretizations that are favored for many classes of solid earth geophysical problems.
We will discuss our approach to develop highly scalable dynamic adaptive mesh discretizations, and the implications for a continuous Galerkin solver for nonlinear Stokes systems and a discontinuous Galerkin method for high-order seismic wave propagation. We outline the formulation of associated inverse problems and include scientific breakthrough results obtained from high-resolution simulations of convection in the earth's mantle enabled by our new-generation parallel AMR techniques.
Biography: Carsten Burstedde studied Physics at the University of Cologne, Germany, and obtained his doctorate in Applied Mathematics at the University of Bonn, Germany. For the past four years he has been working as a postdoc and more recently as Research Associate at The University of Texas at Austin, USA, developing scalable parallel adaptive numerical methods for solid earth Geophysics problems.
- Institute of Computational Science - Website
- Via Giuseppe Buffi 13
The Institute of Computational Science provides a unique research environment, where strong competences in modeling, simulation and information science come together in an open and application oriented atmosphere.
EventManagement powered by scriptplaza.com