MICDE Seminar: Ann Almgren, Lawrence Berkeley National Lab

By |

AnnAlmgren

Bio:  Ann Almgren is a senior scientist in the Computational Research Division of Lawrence Berkeley National Laboratory and the Group Lead of the Center for Computational Sciences and Engineering. Her primary research interest is in computational algorithms for solving PDE’s for fluid dynamics in a variety of application areas. Her current projects include the development and implementation of new multiphysics algorithms in high-resolution adaptive mesh codes that are designed for the latest multicore architectures.  She is a SIAM Fellow and serves on the editorial boards of CAMCoS and SIREV.

Next-Generation AMR

Block-structured adaptive mesh refinement (AMR) is a powerful tool for improving the computational efficiency and reducing the memory footprint of structured-grid numerical simulations. AMR techniques have been used for over 25 years to solve increasingly complex problems.  I will give an overview of recent and planned advances in AMR algorithms and implementations at BerkeleyLab to address the challenges of next-generation multicore architectures and the complexity of multiscale, multiphysics problems.  This will include new ways of thinking about multilevel algorithms and new approaches to data layout and load balancing, in situ and in transit visualization and analytics, and run-time performance modeling and control.

 

 

 

 

MICDE Seminar: Andrea Lodi, Polytechnique Montréal

By |

AndreaLodi

Bio:  Andrea Lodi received a PhD in System Engineering from the University of Bologna in 2000 and he was a Herman Goldstine Fellow at the IBM Mathematical Sciences Department, NY from 2005–2006. He was a full professor of Operations Research at DEI, University of Bologna between 2007 and 2015. Since 2015 he has been the Canada Excellence Research Chair in “Data Science for Real-time Decision Making” at the Polytechnique Montréal. His main research interests are in Mixed-Integer Linear and Nonlinear Programming and Data Science and his work has received recognition including the IBM and Google faculty awards. He is author of more than 80 publications in the top journals of the field of Mathematical Optimization. He serves as Associate Editor for several prestigious journals in the area. He has been the network coordinator and principal investigator of two large EU projects/networks, and, since 2006, consultant of the IBM CPLEX research and development team. Finally, Andrea Lodi is the co-principal investigator (with Yoshua Bengio) of the project “Data Serving Canadians: Deep Learning and Optimization for the Knowledge Revolution”, recently funded by the Canadian Federal Government under the Apogée Programme.

On Wide Split Cuts for Mixed-Integer Programming

Cutting planes (or simply cuts) are a fundamental component of modern Mixed-Integer Linear Programming (MILP) solvers because they help in strengthening the linear programming relaxation, a proxy to make the branchand-bound tree small. A classical way of devising cuts is to exploit disjunctions, for example in the domain of an integer variable, where, of course, no fractional value leads to any feasible solution. Cutting planes of this type, called split cuts, classically exploit disjunctions whose ‘width’ is always equal to one, i.e., no fractional value is feasible between two consecutive integer values. We investigate cutting planes that arise when widening the associated disjunctions. This allows, e.g., to model non contiguous domains of (integer) variables (or, stated differently, ‘holes’ in the domains). The validity of the disjunctions in a MILP can come from either primal or dual information, and we present examples and computational results in both cases. We further explore an exact MILP approach based on these cutting planes, that in addition tackles non-contiguity directly via branching and as a side-effect reduces the model size. (Joint work with P. Bonami, F. Serrano, A. Tramontani, S. Wiese.)

This seminar is co-sponsored by the U-M Department of Industrial & Operations Engineering

MICDE Seminar: Anthony Wachs, University of British Columbia

By |

cropped-Anthony_Wachs_photo2Bio: Anthony Wachs is an assistant professor with a joint appointment in the departments of Mathematics and of Chemical and Biological Engineering at the University of British Columbia, Vancouver, Canada. He received his B. Sc. and M. Sc. from the University Louis Pasteur of Strasbourg and his PhD from the Institut National Polytechnique of Grenoble in 2000. Right after, he was hired in 2001 as a Fluid Mechanics research engineer at IFP Energies nouvelles (IFPEN, at that time Institut Français du Pétrole) in Paris.

In 2009, he spent a one-year sabbatical at the nuclear research center of Cadarache in the south of France, where he worked for IRSN (the french national safety administration for nuclear energy). In 2010, he got his HDR (French Habilitation to Supervise Research) and was later promoted Scientific Advisor at IFPEN in Multiphase Flows and Scientific Computing. He then moved to IFPEN-Lyon where he supervised a group of researchers (including PhD and post-doc students) on the numerical simulation of reactive particulate flows (www.peligriff.com).

His main research interests are non-Newtonian Flows, Multiphase Flows and High Performance Computing. He collaborates extensively with academic groups in Canada, Brazil, France and Germany.

Micro/meso numerical modeling of flows laden with particles of arbitrary shape

Particulate flows are ubiquitous in environmental, geophysical and engineering processes. The intricate dynamics of these two-phase flows is governed by momentum transfer between the continuous fluid phase and the dispersed particulate phase. When significant temperature differences exist between the fluid and particles and/or chemical reactions take place at the fluid/particle interfaces, the phases also exchange heat and/or mass, respectively. While some multi-phase processes may be successfully modelled at the continuum scale through closure approximations, an increasing number of applications require resolution across scales, e.g. dense suspensions, fluidized beds. Within a multi-scale micro/meso/macro-framework, we develop robust numerical models at the micro and meso scales, based on a Distributed Lagrange Multiplier/Fictitious Domain method and a two-way Euler/Lagrange method, respectively. Collisions between finite size particles are modeled with a Discrete Element Method. Many real-life processes and/or flows involve non-spherical particles. Although there is still a lot to learn about flows laden with spherical particles, there is also a strong incentive to develop new modeling tools to account for non-spherical, angular, convex or even non-convex particles. We discuss assorted issues related to the numerical modelling of flows laden with particles of arbitrary shape. Along the way, we also address high performance computing issues related to our massively parallel numerical tools and challenges to efficiently transfer knowledge from small scales to large scales. We illustrate the modelling capabilities of our tools on the two following problems relevant of applications from the chemical engineering and process industry: (i) a rotating drum filled with non-convex particles and (ii) fixed and fluidized beds of multilobic (and hence non-convex) particles.

This seminar is co-organized with the Applied Interdisciplinary Mathematics program

MICDE Seminar: Jonathan Freund, University of Illinois at Urbana-Champaign

By |

Bio: Jonathan Freund is the Donald Biggar Willett Professor of Mechanical Science & Engineering and Aerospace at the University of Illinois at Urbana-Champaign.   He is a Fellow of the American Physical Society, and a winner of the 2008 Frenkiel Prize from its Division of Fluid Dynamics where he currently serves as the division secretary/treasurer.  He is an associate editor of Physical Review Fluids and on the editorial board of Annual Review of Fluid Mechanics.  Computational science has been central to his research, which has included simulations of turbulent jet noise and its control, the dynamics of molecularly thin liquid films, nanostructure formation by ion-bombardment of semiconductor materials, and most recently the dynamics of red blood cells flowing in the narrow confines of the microcirculation.  He co-directs the DOE-funded Center for Exascale Simulation of Plasma-Coupled Combustion at the University of Illinois.

Adjoint-based optimization for understanding and reducing flow noise

Advanced simulation tools, particularly large-eddy simulation techniques, are becoming capable of making quality predictions of jet noise for realistic nozzle geometries and at engineering relevant flow conditions.  Increasing computer resources will be a key factor in improving these predictions still further.  Quality prediction, however, is only a necessary condition for the use of such simulations in design optimization.  Predictions do not of themselves lead to quieter designs.  They must be interpreted or harnessed in some way that leads to design improvements.  As yet, such simulations have not yielded any simplifying principals that offer general design guidance. The turbulence mechanisms leading to jet noise remain poorly described in their complexity.  In this light, we have implemented and demonstrated an aeroacoustic adjoint-based optimization technique that automatically calculates gradients that point the direction in which to adjust controls in order to improve designs.  This is done with only a single flow solutions and a solution of an adjoint system, which is solved at computational cost comparable to that for the flow. Optimization requires iterations, but having the gradient information provided via the adjoint accelerates convergence in a manner that is insensitive to the number of parameters to be optimized.  The talk will review the formulation of the adjoint of the compressible flow equations for optimizing noise-reducing controls and present examples of its use.  We will particularly focus on some mechanisms of flow noise that have been revealed via this approach.

This seminar is co-sponsored by U-M Aerospace Engineering

MICDE Seminar: Jeremy Lichstein, University of Florida

By |

JeremyLichsteinBio: Jeremy Lichstein is an assistant professor of Biology at the University of Florida. Professor Lichstein got his Ph. D. from Princeton University and was a postdoctoral research fellow at Princeton’s department of Ecology and Evolutionary Biology. He was the recipient of the University of Florida Excellence Award for Assistant Professor, and was named a Florida Climate Institute Fellow for 2016-2017. His research interests are forest dynamics, biodiversity, carbon cycle and climate change.

Biodiversity and the changing Earth System: computational challenges and new answers to old questions

Terrestrial ecosystems currently offset roughly 25% of global annual anthropogenic fossil fuel emissions. However, the fate of this carbon sink is highly uncertain, in large part because global models diverge in their predictions of ecosystem responses to climate change, drought, and other perturbations. Although there is little agreement on how terrestrial ecosystems will respond to global change on decadal and longer time-scales, there is wide consensus that current global models are overly simplistic in their representation of important ecological processes. I will discuss our current understanding of how tree functional diversity is maintained in forests, the consequences of including more realistic levels of functional diversity in global models, and the computational challenges that need to be overcome in order to introduce ecological realism into the Earth System Models that the scientific and policy communities rely on for climate projections. A key result that is emerging from empirical and theoretical studies is that shifts in species composition across time or space (beta diversity) have different (and sometimes opposite) effects on ecosystem stability as local (alpha) diversity.

This seminar is co-sponsored by the U-M department of Ecology and Evolutionary Biology

MICDE Seminar: Rob Gardner, University of Chicago

By |

RobGardnerBio: Robert Gardner is a Senior Scientist at the Computation Institute from the University of Chicago,  and aSenior Scientist in the Enrico Fermi Institute. He spent his early academic career doing experimental high-energy physics research at different universities in the Midwest. He has been a member of the ATLAS experiment using the Large Hadron Collider at the CERN Laboratory, Geneva, Switzerland since 1998. His experimental work led him to specialize in developing and improving distributed computing technologies necessary for discoveries at the frontier of particle physics. He was instrumental in developing early research computing grids in the U.S.: the International Virtual Data Grid Laboratory (iVDGL), and the first deployment of the Open Science Grid (OSG) (NSF, Department of Energy). He have also generated systems for metrics collection for distributed systems (Grid Telemetry, PI, NSF-ITR). Currently, he directs the ATLAS Midwest Tier2 Center, which is comprised of integrated computing facilities from the University of Chicago, Indiana University, and the University of Illinois.

Leadership cyberinfrastructure for science and the humanities

In the past two decades high energy physics transformed its computing model from one relying on a single high performance computing center at the host laboratory to one incorporating resources distributed across institutional boundaries and geographic regions. Given the complexity of detectors and scale of data, the international collaborations of the Large Hadron Collider at CERN demanded it. By removing barriers to resource sharing, the resulting data and computation platform democratized the physics process across collaborations. Accelerated modes of scientific discovery by thousands of physicists were forged using hundreds of data centers linked by very high bandwidth networks. Meanwhile the explosion of commercial, social and enterprise data has driven innovation in resource abstraction and the creation of new service platforms, offering fresh opportunities to accelerate science and intellectual inquiry at all scales and across all domains. In this talk I’ll discuss the strategic significance that cyberinfrastructure technology plays in this regard and describe models for creating ubiquitous “substrates” that remove obstacles to connecting campuses, facilities, instruments and researchers.

This seminar is co-sponsored by the U-M department of Physics

MICDE Seminar: Nathan Kutz, University of Washington

By |

Nathan.KutzBio: Nathan Kutz is the Robert Bolles and Yasuko Endo Professor in the department of Applied Mathematics, and an adjunct professor of Electrical Engineering and Physics at the University of Washington. He was awarded the B.S. in Physics and Mathematics from the University of Washington in 1990 and the PhD in Applied Mathematics from Northwestern University in 1994. Following postdoctoral fellowships at the Institute for Mathematics and its Applications (University of Minnesota, 1994-1995) and Princeton University (1995-1997), he joined the faculty of applied mathematics and served as Chair from 2007-2015.

Data-driven discovery of dynamical systems in the engineering, physical and biological sciences

We demonstrate that the integration of data-driven dynamical systems and machine learning strategies with adaptive control are capable of producing efficient and optimal self-tuning algorithms for many complex systems arising in the engineering, physical and biological sciences. We demonstrate that we can use emerging, large-scale time-series data from modern sensors to directly construct, in an adaptive manner, governing equations, even nonlinear dynamics, that best model the system measured using sparsity-promoting techniques. Recent innovations also allow for handling multi-scale physics phenomenon and control protocols in an adaptive and robust way. The overall architecture is equation-free in that the dynamics and control protocols are discovered directly from data acquired from sensors. The theory developed is demonstrated on a number of example problems. Ultimately, the method can be used to construct adaptive controllers which are capable of obtaining and maintaining optimal states while the machine learning and sparse sensing techniques characterize the system itself for rapid state identification and improved optimization.

This seminar is co-sponsored by the U-M Department of Mathematics.

MICDE Fall 2016 Seminar Series speakers announced

By | Educational, Events, General Interest, News

The Michigan Institute for Computational Discovery and Engineering (MICDE) is proud to announce its fall lineup of seminar speakers. In cooperation with academic departments across campus, the seminar series brings nationally recognized speakers to campus.

This fall’s speakers are:

Sept. 13: Nathan Kutz, Professor of Applied Mathematics, University of Washington

Sept. 22: Rob Gardner, Senior Scientist at the Computation Institute, University of Chicago

Sept. 29: Jeremy Lichstein, Assistant Professor of Biology, University of Florida

Oct. 6: Jonathan Freund, Professor of Mechanical Science and Engineering and of Aerospace Engineering, University of Illinois, Urbana-Champaign

Oct. 14: Anthony Wachs, Assistant Professor of Mathematics and of Chemical and Biological Engineering, University of British Columbia

Oct. 26: Andrea Lodi, Professor of Mathematical and Industrial Engineering, Polytechnique Montreal

Nov. 11: David Higdon, Professor of the Biocomplexity Institute, Virginia Tech

Dec. 9: Ann Almgren, Staff Scientist at the Center for Computational Sciences and Engineering, Lawrence Berkeley National Laboratories

For more information, including links to bios and abstracts as available, please visit micde.umich.edu/seminar-series/.

Students in the Graduate Certificate in Computational Discovery and Engineering program are required to attend at least half of the seminars.

Advanced Research Computing at Michigan — An Overview

By |

Brock Palen, Associate Director of Advanced Research Computing – Technology Services, will provide an overview of the resources available to researchers engaged in computationally intensive science on the University of Michigan campus.

The talk is open to researchers from any department at U-M.

The session will address:

  • high performance computing services
  • data science services such as Hadoop and Spark
  • research storage
  • cloud services
  • networking services
  • grant consultation and collaboration
  • access to off-campus resources.

There will be time for questions and answers after the presentation.