Scientific Computing Seminar

Date and Place: Thursdays and hybrid (live in 32-349/online via Zoom). For detailed dates see below!

Content

In the Scientific Computing Seminar we host talks of guests and members of the SciComp team as well as students of mathematics, computer science and engineering. Everybody interested in the topics is welcome.

List of Talks

  • Thu
    28
    Apr
    2016

    11:30SC Seminar Room 32-349

    Ole Burghardt, SciComp/Robert Bosch GmbH

    Title:
    Wärmeübergangsmodelle und –sensitivitäten an Fluid-Festkörper-Grenzflächen

    Abstract:

    Die Wärmeverteilung in einem Festkörper wird durch reine Wärmeleitung – das heißt, nach dem Fourierschen Gesetz durch die Temperaturgradienten – bestimmt, im Fluid hingegen auch durch die Konvektion der Wärme und den Einfluss der Turbulenz auf die Temperaturleitfähigkeit. Im Vortrag werden vier verschiedene Modelle vorgestellt, wie der Wärmeübergang an Grenzflächen abgeschätzt werden kann und welche Konsequenzen die Vernachlässigung bestimmter physikalischer Effekte für Sensitivitätsberechnungen hat. Wir beginnen mit einer Abschätzung, die allein das Fluid betrachtet und mit den zugehörigen Impulswerten auskommt, und gehen vor bis zum sogenannten Conjugate-Heat-Transfer, bei dem der gegenseitige Einfluss der PDEs über die Randbedingungen an der Grenzfläche berücksichtigt wird.

  • Fri
    29
    Apr
    2016

    11:30SC Seminar Room 32-349

    Joshua Billert, RHRK/SciComp

    Title:
    Performance beim HTTP – Beschleunigte Ladezeiten bei Webseiten durch technisch angepasste Templates

    Abstract:

    Diese Arbeit befasst sich mit verschiedenen Aspekten, welche die Ladezeit im Browser beeinflussen und liefert Lösungsansätze, um Webseiten performanter zu gestalten.

  • Tue
    10
    May
    2016

    10:00SC Seminar Room 32-349

    PD Dr. Gabor Janiga, Universität Magdeburg, Institut für Strömungstechnik und Thermodynamik (ISUT)

    Title:
    CFD-based flow optimization for engineering and medical applications

    Abstract:

    The present talk discusses various optimization problems based on Computational Fluid Dynamics (CFD) simulations. The application of stochastic optimization methods, such as evolutionary algorithms, requires a large number of individual evaluations to explore the often high-dimensional design variable space. CFD simulations for engineering and medical applications are often unsteady and three-dimensional leading to large computational times. Specific examples, including geometrical optimization of wind turbines as well as flow-diverter stent optimization for the treatment of intracranial aneurysms considering the complex patient-specific geometry, involving enormous computational efforts will be presented during the talk.

  • Thu
    02
    Jun
    2016

    11:30SC Seminar Room 32-349

    Max Sagebaum, SciComp

    Title:
    Current and future developments of the AD tool CoDiPack

    Abstract:

    One year after the first release of CoDiPack, the advancements of the tool in this time period are shown. The features are explained and use cases will be presented. The second half of the talk will give an overview of features that are currently under development and are planed for the next year.

  • Mon
    27
    Jun
    2016

    10:30SC Seminar Room 32-349

    Prof. Andreas Griewank, School of Mathematics and Computer Science, Yachay Tech

    Title:
    Optimality and convexity conditions for piecewise smooth objective functions

    Abstract:

    Any piecewise smooth function that is specified by an evaluation procedure involving smooth elemental functions and piecewise linear functions like min and max can be represented in abs normal form. This is in particular true for most popular nonsmooth test functions. By an extension of algorithmic, or automatic differentiation, one can then compute certain first and second order derivative vectors and matrices that represent a local piecewise linearization and provide additional curvature information. On the basis of these quantities we characterize local optimality by first and second order necessary and sufficient conditions, which generalize the corresponding KKT and SSC theory for smooth problems. The key assumption is the Linear Independence Kink Qualification (LIKQ), a generalization of LICQ familiar from NLOP. It implies that the objective has locally a so-called VU decomposition and renders everything tractable in terms of matrix factorizations and other simple linear algebra operations.

  • Thu
    07
    Jul
    2016

    11:30SC Seminar Room 32-349

    Valentin Fütterling, Fraunhofer ITWM

    Title:
    Photo-realistic image synthesis with Path Tracing – An optimization problem?

    Abstract:

    Path tracing is a versatile, physically-based algorithm to generate photo-realistic images from virtual objects and environments.
    Applications are wide-spread in movie production, product design and advertising. A promising new area is emerging with virtual and augmented reality. The basis of Path Tracing is Monte Carlo Integration which requires a non-uniform distribution of samples over the integration domain in order to minimize noise and maximize efficiency. This talk will give an introduction to the Path Tracing algorithm, and pose the question whether the method can profit from formulating it as an optimization problem.

  • Thu
    21
    Jul
    2016

    11:30SC Seminar Room 32-349

    Thomas Economon, Aerospace Design Laboratory, Stanford University

    Title:
    Simulation-based Analysis and Design for Efficient Aerospace Systems

    Abstract:

    The aerospace industry regularly depends upon computational methods for analyzing and designing advanced aircraft, launch vehicles, or jet engines, for example. However, these calculations can carry high computational cost at the required fidelity, and the expense is compounded by embedding the calculations within design loops. To combat this, we seek advances in design methodologies, including the development of high-fidelity, adjoint-based techniques that can provide tremendous efficiency improvements for gradient-based optimization using computational fluid dynamics (CFD). Moreover, by making effective use of the extreme computational resources becoming available, problem complexity can be increased, and predictions can become more accurate, reliable, and robust while maintaining turnaround times that fit within the industrial design cycle. Consequently, the combination of these theoretical and computational advances will directly enable the design of next-generation aerospace vehicles with reduced fuel burn, emissions, and noise.

    As demonstrations of these components, this talk contains two separate but related topics. The first portion presents the development and application of a new unsteady continuous adjoint formulation for optimal shape design of aerodynamic surfaces in motion, such as rotating or pitching applications. By leveraging shape calculus, the resulting surface formulation efficiently provides the sensitivity information necessary for performing gradient-based aerodynamic shape optimization. Several open issues and future directions for adjoint-based methods will be highlighted. The second part of the presentation covers recent work in the area of high performance computing. As supercomputers push toward exascale, it is increasingly difficult to achieve high levels of performance and portability on emerging hardware architectures. The scalability needed by many crucial applications will only be realized through research and investment in both algorithmic improvements and new computer science techniques. Accordingly, investigations into scalable algorithms and the harnessing of multiple levels of parallelism in a modern CFD code will be presented.