Seminar Scientific Computing

Seminar registration: If one has interests to participate at the seminar in the coming summer term, contact Martin Krosche. We arrange a short appointment to speak about the topic and the supervisor. The registration ends in the first week of the summer semester (or if no free topics are available). Usually the seminar takes place Thursdays between 16:45-18:30 o'clock in the seminar room RZ012 of the computing center.

Every participant of our seminar has to write an elaboration about his seminar topic (a summery in his/her own words). This written document must be delivered to the supervisor two weeks before the talk is scheduled (at the latest!). But we assume that every participant informs his supervisor continuously about the current state of his work. Is this not the case and the participant do not contact the supervisor continuously, the topic is opened again for other students or is removed from the schedule. The presentation foils must be delivered one week before the talk is scheduled (at the latest!). The foils must be checked by the supervisor. The duration of the talk should be 30 minutes. After the seminar talk there will be a small discussion up to 15 min. It is oligatory to be present at all other seminar talks!

The time one needs for a seminar is nearly equivalent with the time one needs for a 2 hours lecture with excercises and homeworks (4 credits). One credit is equivalent to 30 hours of work, i.e. for a seminar (4 credits) one have to calculate 120 hours of work (3 weeks (Mo-Fr) with 8 hours per day).

Each participant gets a research paper. We expect that the participants go through the material independently, e.g. doing a small literature review. As a result of this phase an elaboration which summarises the paper has to be written. This document is the source for the foils for the talk. Additionally all participants have to be present in all seminar talks.

The elaboration has to be written in latex, so the foils. You can find latex frameworks for the document and the foils in the download section on this page.

• Holonomic Constraints for Molecular Dynamics
  Abstract: In classic models of the molecular mechanics for Molecular Dynamics Simulations the high-frequency parts of the degree of freedoms for the bond- and angle potentials are often to be regarded as fixed. The lengths of the chemical bonds and values of the angles in molecules are considered as fixed. By the use of holonomic constraints in molecular simulation systems the bondlengths and angles are permanently restraint on an default value. For an implementation of such geometrical constraints within the molecules different Methods exits. One of them is the so called symplectical RATTLE-Algorithm, which fullfills in every time step the constraints by means of Lagrange multipliers. In the RATTLE a set of nonlinear equations are generated and solved iteratively for the Lagrange multipliers. The first task is to give a formulation of the symplectical RATTLE-algorithm in the framework of general hamiltonian functions and compare the RATTLE with the related SHAKE-algorithm. For complex constraints one can use other iterative solver for the Lagrange multipliers for instance the Newton-Method. Outline the advantage of the RATTLE-iteration in comparison to the Newton-iteration.
• Splitting methods
  Abstract: In many instationary problems the spatial differential operator can be split into a sum of operators, each having it's own distinct properties. An example is a reaction diffusion system where the operator can be split into the reaction and the diffusion part. Each of them can be integrated separately taking into account the special properties of the operator (e.g. stiffness, conservativity). The method shall be explained in detail together with it's implication on the order and stability of the resulting scheme and some sample applications showing the method in practice.
• Adaptive Finite Element Method
  Keywords: error estimation/indication, adaptive discretisations, h-, p-refinement
• Uncertainty Quantification in Complex Systems
  Significant research has been expanded over the past several decades to develop model-based predictions into sharp estimators of the actual behavior of natural and physical phenomena. The vision of computational experiments paralleling and predicting the outcomes of physical tests is already a driving force and an accepted model for the future of scientific computing. A key component in realizing this vision is the accurate and meaningful quantification of errors in model-based predictions. The estimation of errors associated with the discretization of the partial differential equations governing a particular problem is a very active research field. The interpretation let alone estimation of errors associated with natuarl variability and limited data is, on the other hand, an emerging field. This field of uncertainty quantification addresses issues that are paramount to the validation of model-based predictions and their use as surrogates to physical tests.
• Computational Optimization of Systems Governed by Partial Differential Equations (PDE)
  Optimization problems governed by partial differential equation (PDE) constraints arise in many important applications. Progress in computational and applied mathematics combined with the availability of rapidly increasing computer power steadily enlarges the range of applications that can be simulated numerically and for which optimization tasks, such as optimal design, parameter identification, and control are being considered. For most of these optimization problems, simple approaches combining off-the-shelf PDE solvers and optimization algorithms often lack robustness or can be very inefficient. Successful solution approaches have to overcome challenges arising from, e.g., the increasing complexity of applications and their mathematical models, the influence of the underlying infinite dimensional problem structure on optimization algorithms, and the interaction of PDE discretization and optimization. Also the solution of pde-constrained optimization requires techniques from a number of mathematical disciplines including functional analysis, optimal control theory, numerical optimization, numerical PDEs and numerical analysis.
• Modeling and Analysis of Non-Markovian Stochastic Processes using Stochastic Petri Nets
  Stochastic processes without the Markov property (memoryless property) are called Non-Markovian processes. This seminar deals with modeling and analysis techniques for this kind of stochastic processes.

• Direct Simulation Monte Carlo
  Organiser: Dr. Christian Oldiges
• Quadrature Formulas for High Dimensional Integration
  Organiser: Martin Krosche

Latex foils framework for the presentation: presentation.tar.gz

Latex framework for the written elaboration: seminarausarbeitung.tar.gz

• Der goldene Weg zum perfekten Seminarvortrag

Martin Krosche: martin.krosche(at)tu-bs.de