Home › Activities › Seminars › Other Seminars and Lectures

Other Seminars and Lectures

Construction of multiscale materials models
November 3, 2011 - 9:00am - 12:30pm
SAMSI, Room 150
Khalil Elkhodary, Northwestern University

The meeting will take place from 9:00 to 12:30 in room 150 and will consist of two parts separated by a coffee break. First, the main speaker will introduce the topic through a technical presentation aimed at mathematicians and statisticians. In a second part, discussions will focus on the identification of UQ issues relevant to the problem at hand. No registration is necessary and everybody is welcome. There is a coffee break in the middle of the morning and a light lunch at the end.

Abstract:

The phenomenologies governing the macroscale performance of materials stem primarily from the mesoscale; that is, the length scale wherein the dominant microstructural features can be discerned as distinct components. At the mesoscale, meaningful and measurable material metrics are extracted to guide the process of materials design and synthesis. It follows that mesoscale mechanics plays a critical bridging role in the multiscale paradigm: It connects sub-mesoscale physics and chemistry to supra-mesoscale materials performance.

Clearly, an explicit account of all length and time scales governing mesoscale mechanics in any computation is both unnecessary and prohibitive. Thus, continuum mechanics theories that originally derive from the theory of simple bodies have been proposed to reduce the order of complexity of multiscale models and simulations. Multiscale mechanics theories have generalized the notions of a simple mesoscale body by positing the existence of multilevel structures underlying each point of the body. The multilevel structures are in turn posited to correspond to nested length scales, often orders of magnitude apart, all being small enough to be considered as fully contained in a point at the mesoscale of the simple body (thus appearing as a homogeneous continuum). These theories fail to recognize that in most modern engineering materials, which are inherently multi-component materials, the complexity of structure is often horizontal, i.e. at the same level of resolution. In fact, at the mesoscale, there may exist such complexity as is too costly to resolve explicitly in a direct numerical simulation (DNS) model, so that further homogenization into a simpler body continues to be needed for reasonable computational domains and feasible simulation times.

The consequence of all existing multilevel/multiscale approaches is that a single constitutive law of known form must be postulated a priori for a point at the mesoscale, which is inadequate for modern materials, as there exist multiple interacting components at that scale. Moreover, discovering the mesoscale response of these modern multi-component materials by large-scale simulations is precisely the desired outcome of computational modeling. If computational design of materials is desired, theories that propose a priori single constitutive laws for mesoscale material points inherently undermine such an endeavor.

We thus propose a new approach for multiscale microstructured materials, called the Archetype Blending Continuum (ABC) theory. Its purpose is to offer a generalized continuum framework that is valid across material systems, precisely to facilitate multiscale constitutive modeling for the design and analysis of complex microstructures, particularly for modern multi-component materials. Each mesoscale component is termed an archetype, and is thought of as a building block of the microstructure, thus recognizing the horizontal complexity of microstructures, and not requiring the assumption of scale separation. As archetypes appear at the mesoscale, they each admit their own sub-structures, which we term nanomorphisms; thus their constitutive behavior may be obtained from (1) constitutive laws or libraries established in literature, leveraging the expertise of materials scientists and constitutive modelers focused on detailed mechanisms (2) experimentation and imaging of archetypes and their sub-mesoscale features, or (3) by firing up multiscale (i.e. nested length scale) simulations from atomistic principles, to construct reduced-order mesoscale models for an archetype. Also, interactions between archetypes are defined via separate constitutive laws for the mechanisms evolving across imbedded interfaces. Archetype blending algorithms will then construct dynamically equivalent homogeneous mesoscale continua, and generate the desired macroscopic constitutive law by large-scale simulation. This approach permits feasible simulation of richly microstructured materials, while allowing the decoupling of mesoscale components from their interactions for a detailed analysis of their evolving properties and the extraction of multi-component and multiscale metrics for materials design. Specifically, this new approach sets up a modular framework for constitutive modeling of complex microstructures to facilitate the prediction of new material responses when components are separately modified in the design process.

Multiscale modeling of neutronics
November 17, 2011 - 10:00am - 12:30pm
SAMSI, Room 150
Hany Abdel-Khalik, NCSU and INL

The meeting will take place from 10:00 to 12:30 in room 150 and will consist of two parts separated by coffee break. First, the main speaker will introduce the topic through a technical presentation aimed at mathematicians and statisticians. In a second part, discussions will focus on the identification of UQ issues relevant to the problem at hand. No registration is necessary and everybody is welcome. There is a coffee break in the middle of the morning and a light lunch at the end.

Abstract:

Central to reactor calculations is the determination of the ensemble average of the neutron distribution as a function of space, energy, and direction of travel. Although, the associated physical phenomenon, i.e., neutron transport, is very well-understood, the numerical solution of the mathematical model, the Boltzmann Equation in a 7 variable phase space, is computationally intractable even with existing supercomputing platforms. To overcome this difficulty, reactor physicists have devised homogenization strategies to render solutions in practical times. The homogenization techniques employed have primarily been based on intuitive understanding of the physics process, and therefore has been closely tailored to reactor calculations. The level of accuracy attained is very good considering the complexity of the problem and sometimes the crude nature of the assumption, it is however still difficult to prove many important properties such as the existence of the solution, the uniqueness, convergence, rate of convergence, etc. This talk will overview the methods used in reactor calculations employing a generic mathematical framework to help stimulate discussions with the audience about the needed advances to improve the quality of reactor calculations.

Uncertainty Quantification & Propagation in Multiphase Flow Computational Fluid Dynamics Applications
December 15, 2011 - 9:00am - 12:30pm
SAMSI, Room 150
Aytekin Gel (NETL) and Charles Tong (LLNL)

The meeting will take place from 9:00 to 12:30 in room 150 and will consist of two parts separated by coffee break. First, the main speaker will introduce the topic through a technical presentation aimed at mathematicians and statisticians. In a second part, discussions will focus on the identification of UQ issues relevant to the problem at hand. No registration is necessary and everybody is welcome. There is a coffee break in the middle of the morning and a light lunch at the end.

Abstract: Recent years have seen a dramatic increase in the use of scientific computer simulations to design a diverse set of complex engineering systems ranging from transportation vehicles to clean energy technology development. These advances have essentially reshaped the system design process; in particular, the use of physical models or prototypes has reduced, resulting in significant savings in cost and time in the design cycle. This situation has improved time-to-market lifecycle. In spite of their increasingly widespread use and success, the current computer simulation approaches, in particular for multiphase flows, do not provide objective or statistically-meaningful confidence intervals for the predicted results. For example, the computational fluid dynamics (CFD) model of a gasifier predicts the performance of a gasifier for given set of input parameters (e.g., geometry specifications, gas/solid flow conditions, species mass fractions, reaction rates). In reality, these parameters are generally uncertain, and the associated variability (e.g., due to the source of fuel, degradation of the reactor with time, fluctuations in the flow controls) in them can have substantial impact on the design and performance of the gasifier. To increase the reliability of computer simulations, the performance predictions must include a measure of the uncertainty resulting from uncertainties in the input parameters, constitutive models, and numerical methods.

To this end, a new initiative has been launched at NETL to investigate the application of uncertainty quantification (UQ) and propagation methods to multiphase CFD with particular application to gasifier simulations [1,2]. The work presented here employs a non-intrusive parametric uncertainty quantification and propagation. The advantage of this approach is that the simulation code is treated as a black box and the code need not be modified. For this purpose, PSUADE [3], an open-source UQ toolbox developed at LLNL has been interfaced with the open source CFD code, MFIX [4]. For a simplified gasifier problem, a set of simulations have been performed to build a response surface to act as a surrogate model for a global sensitivity study to determine the reaction rate that has the most effect on the species composition of the product gas. Preliminary results of simulations conducted with the coupled MFIX-PSUADE codes will be presented.

1. Tong, C. and Gel, A., “Applying Uncertainty Quantification to Multiphase Flow CFDs”, Proceedings of the 2011 NETL Multiphase Flow Science Workshop, August 16-18, 2011, Pittsburgh, PA (http://www.netl.doe.gov/publications/proceedings/11/mfs/index.html)

2. Gel, A., Tong, C., Li, T., Shahnam, M., Guenther, C., Syamlal, M., and Garg, R.
“Uncertainty Quantification in Reactive Multiphase CFD”, The 6th Sino-US Joint Conference of Chemical Engineering, SINOPEC Conference Centre, Beijing, China, November 7-10, 2011.

3. PSUADE URL: https://computation.llnl.gov/casc/uncertainty_quantification/#psuade

4. MFIX (Multiphase Flow with Interphase eXchanges) URL: http://www.mfix.org

Scientific Uncertainty and Climate Change Risk Management: A Choice Between Managed or Unmanaged Change
January 11, 2012 - 11:00am - 12:00pm
via WebEx
Jay Gulledge, Pew Center on Global Climate Change

Dr. Jay Gulledge is the 2011 recipient of the Charles S. Falkenberg Award for his work communicating climate change science, risks, and meaningful response options to decision-makers and the public. The award is presented jointly by the American Geophysical Union (AGU) and the Earth Science Information Partnership (ESIP) to honor “a scientist under 45 years of age who has contributed to the quality of life, economic opportunities and stewardship of the planet through the use of Earth science information and to the public awareness of the importance of understanding our planet.”

Dr. Gulledge joined the Pew Center in 2005 and directs its science and impacts program, has worked to build public awareness of climate change science, the risks of climate change for natural and social systems, and approaches to managing those risks. In this role, he has communicated both an understanding of climate science and the need for urgent action to a diverse audience of non-scientists including policy-makers, the business community, and the media. Dr. Gulledge’s recent work uses a risk management framework to help explain that uncertainty over climate science is not a reason for inaction, rather it is a reason to act now to minimize both the risk that comes with climate change and the cost of mitigating it.

“He has the unique ability to translate scientific uncertainty into useful information for decision-makers and the public,” said the Center for Climate and Energy Solutions (C2ES) President Eileen Claussen, President of the Pew Center on Global Climate Change. “Jay often says, ‘Uncertainty is information.’ For the public, that notion is nothing short of revolutionary.”

In December Dr. Gulledge will be honored for his achievements at the 2011 AGU Fall Meeting in San Francisco. Established in 2002, the Falkenberg Award honors a scientist under age 45 who has contributed to the quality of life, economic opportunities, and stewardship of the planet through the use of Earth science information, and to the public awareness of the importance of understanding our planet.

Dr. Gulledge directs C2ES’s efforts to assess and communicate the latest scholarly information about the science and environmental and social impacts of climate change. In Pew Center reports, on the Climate Compass blog , and in numerous media interviews, Dr. Gulledge connects the dots between climate change and extreme weather , explains scientific developments in accessible terms, and delivers straight answers that increase public understanding of climate change.

Dr. Gulledge has also broken new ground in his work on the relationship between climate change and national security. As a non-resident Senior Fellow at the Center for a New American Security, he has co-authored influential reports, including The Age of Consequences: The Foreign Policy and National Security Implications of Global Climate Change and Lost in Translation: Closing the Gap Between Climate Science and National Security Policy < http://www.cnas.org/node/4391>. More recently he co-authored a ground-breaking report on re-framing international climate policy called Degrees of Risk: Defining a Risk Management Framework for Climate Security < http://www.c2es.org/publications/degrees-risk-defining-risk-management-f....

Dr. Gulledge is a Certified Senior Ecologist with two decades of experience teaching and conducting research in the biological and environmental sciences. He earned a Ph.D. from the University of Alaska Fairbanks and was a Life Sciences Research Foundation Postdoctoral Fellow at Harvard University. He has held faculty posts at Tulane University and the University of Louisville.

“The ability to effectively communicate Earth science to a wide range of audiences is rare, and Jay ranks among the very few who possess that skill,” said Claussen. “His dedication to transparency and accuracy and his unflagging defense of the scientific process in the face of political shenanigans have earned him the respect of his peers.”

Singular value decomposition for high-dimensional data
January 11, 2012 - 1:00pm - 2:00pm
SAMSI, Room 150
Dan Yang, University of Pennsylvania

Abstract: Singular value decomposition is a widely used tool for dimension reduction in multivariate analysis. However, when used for statistical estimation in high-dimensional low rank matrix models, singular vectors of the noise-corrupted matrix are inconsistent for their counterparts of the true mean matrix. In this talk, we suppose the true singular vectors have sparse representations in a certain basis. We propose an iterative thresholding algorithm that can estimate the subspaces spanned by leading left and right singular vectors and also the true mean matrix optimally under Gaussian assumption. We further turn the algorithm into a practical methodology that is fast, data-driven and robust to heavy-tailed noises. Simulations and a real data example further show its competitive performance.

This is joint work with Andreas Buja and Zongming Ma.

Stochastic Multiscale Analysis and Design
February 23, 2012 - 9:00am - 12:00pm
SAMSI, Room 150
Wei Chen (Northwestern University)

This series is part of the SAMSI UQ/Engineering program and its purpose is to explore issues of uncertainty and error that arise in complex multsicale and multiphysics models used in materials and energy research. The event will consist of two parts, separated by coffee break and followed by a light lunch at the end. First, the main speaker will introduce the topic through a technical presentation aimed at mathematicians and statisticians. In a second part, discussions will focus on the identification of UQ issues relevant to the problem at hand.

Abstract:

Stochastic multiscale analysis and design is an emerging research paradigm aiming for developing methodologies that support the predictive modeling and design of complex multiscale engineered systems. Under this paradigm, future engineered systems can achieve exceptional system performance through concurrent optimization of materials and structures across multiple length scales, accounting for the multiscale nature of physical behaviors. The design of complex materials system poses a set of challenges associated with the various sources of uncertainties beyond microstructure heterogeneity, the complexity and computational cost of multiscale analysis, the heterogeneity of information from both simulations and physical experiments, and the high-dimensionality of design space across multiple scales. This talk will present our research in developing an image-based, data-driven, hierarchical multiscale analysis and design approach in handling the aforementioned complexities. Methods for stochastic characterization and 3D reconstruction of microstructures, stochastic constitutive theory for upscaling, and hierarchical decomposition and stochastic reassembly to predict heterogeneous materials properties will be introduced. Research opportunities associated with a new stochastic Archetype-Blending Continuum (ABC) multiresolution theory will be discussed.

More information about Dr. Chen