Vorträge und Workshops der Arbeitsgruppe Griewank


Shortcuts


Current      History



History


Roundout

Speaker
N.Gauger
Institution
HU
Event
Lectures serie: Methods and Applications of Multiobjective Optimization
Date
13.7.
Time
5:30-7:00 pm
Location
Humboldt Universität zu Berlin, Institut für Mathematik
Rudower Chaussee 25, 12489 Berlin
Room 1.115
More Information
Lectures Homepage

[ Back | Home | Institute ]

Vector optimization problems and parameteric optimization

Speaker
P.Mbunga
Institution
HU
Event
Lectures serie: Methods and Applications of Multiobjective Optimization
Date
6.7.
Time
5:30-7:00 pm
Location
Humboldt Universität zu Berlin, Institut für Mathematik
Rudower Chaussee 25, 12489 Berlin
Room 1.115
More Information
Lectures Homepage

[ Back | Home | Institute ]

Calculation of a feasible point of a non-convex set with application in multiobjective optimization

Speaker
J.Guddat
Institution
HU
Event
Lectures serie: Methods and Applications of Multiobjective Optimization
Date
29.6.
Time
5:30-7:00 pm
Location
Humboldt Universität zu Berlin, Institut für Mathematik
Rudower Chaussee 25, 12489 Berlin
Room 1.115
More Information
Lectures Homepage
Abstract
The calculation of a feasible point of non-convex set M described by non- linear equations and inequalities with an arbitrarily chosen starting point is an open problem until now.

We propose an algorithm (CAFENOCS) using pathfollowing methods and jumps in the set of non-degenerate critical points that are not stationary. We assume that the considered embeddings are Jongen-Jonker-Twilt regular.

There are a lot of interesting applications. We restrict ourself in this lecture to the applications in multiobjective optimization. Computational results are presented.


The Pareto front as range boundary and its generic singularities

Speaker
A.Griewank
Institution
HU
Event
Lectures serie: Methods and Applications of Multiobjective Optimization
Date
16.6.
Time
5:30-7:00 pm
Location
Humboldt Universität zu Berlin, Institut für Mathematik
Rudower Chaussee 25, 12489 Berlin
Room 1.115
More Information
Lectures Homepage

[ Back | Home | Institute ]

One Shot Approaches in Aerodynamic Shape Optimization

Speaker
N.Gauger
Institution
HU
Event
Lectures serie: Methods and Applications of Multiobjective Optimization
Date
8.6.
Time
5:30-7:00 pm
Location
Humboldt Universität zu Berlin, Institut für Mathematik
Rudower Chaussee 25, 12489 Berlin
Room 1.115
More Information
Lectures Homepage

[ Back | Home | Institute ]

Workshop:

Optimization Software at Matheon MF1

Institution
DFG Research Center Matheon - Mathematics for key technologies
Date
June 1, 2005
Location
Zuse Institute Berlin
Takustrasse 7
Berlin - Dahlem
D-14195
Germany
Room:
Lecture Hall in the ground floor
More information
See here

[ Back | Home | Institute ]

Some applications of multicriteria strategies in mechanical engineering

Speaker
D.Bestle
Institution
TU Cottbus
Event
Lectures serie: Methods and Applications of Multiobjective Optimization
Date
25.5.
Time
5:30-7:00 pm
Location
Humboldt Universität zu Berlin, Institut für Mathematik
Rudower Chaussee 25, 12489 Berlin
Room 1.115
More Information
Lectures Homepage

[ Back | Home | Institute ]

Efficient Optimization of Aerodynamic Coefficients I/II

Speaker
N.Gauger
Institution
HU
Event
Lectures serie: Methods and Applications of Multiobjective Optimization
Date
11.5., 18.5.
Time
5:30-7:00 pm
Location
Humboldt Universität zu Berlin, Institut für Mathematik
Rudower Chaussee 25, 12489 Berlin
Room 1.115
More Information
Lectures Homepage
Abstract
There are many ingredients required to establish an efficient and flexible numerical optimization capability. These include suitable techniques for geometry parametrization, meshing and mesh movement methods, efficient and accurate flow solvers as well as a flexible tool-set containing both deterministic and stochastic based optimizers.
Over the last years, numerical shape optimization is one of the major issues of the Institute of Aerodynamics and Flow Technology at DLR. One key activity is the derivation and implementation of a continuous adjoint approach for the DLR flow simulation software MEGAFLOW based on the solution of the Euler and Navier-Stokes equations. Its potential for efficient aerodynamic shape design in compressible flow will be demonstrated.
The two lectures will contain the following three sections: First, the adjoint method will be explained by means of simple linear equations. Second, as an exercise the continuous adjoint equations will be derived for the convection-diffusion equations. Third, the continuous adjoint Euler equations will be derived and their applications will be presented for 2D airfoil designs, wing-body optimizations as well as wing designs in MDO context.
Slides
Download (PDF)

[ Back | Home | Institute ]

Workshop:

Optimization in Aerodynamics

Institution
Department of Mathematics, Humboldt University Berlin
Date
May 09, 2005
Location
Humboldt Universität zu Berlin, Institut für Mathematik
Rudower Chaussee 25, 12489 Berlin
Room 1.410
Abstract
The workshop aims to bring together research scientists from universities, research institutes and industry that work in the field of aerodynamic shape optimization. The requirements for practical shape optimization will be worked out, new solution approaches and optimization tools from research institutes will be presented in conjunction with innovative ideas from the forefront of mathematical research at universities. Particular attention will be paid to the fact that aerodynamic shape optimisation is a part of a larger, multi-disciplinary design process.
More information
See here

[ Back | Home | Institute ]

A review of the normal-boundary intersection (NBI) method

Speaker
T.Steihaug
Institution
Uni Bergen & HU
Event
Lectures serie: Methods and Applications of Multiobjective Optimization
Date
4.5.
Time
5:00-6:30 pm
Location
Humboldt Universität zu Berlin, Institut für Mathematik
Rudower Chaussee 25, 12489 Berlin
Room 1.115
More Information
Lectures Homepage
Abstract
In two groundbreaking papers Indraneel Das and John. E. Dennis (SIAM J. Optimization Vol. 8(1998) 631-657 and Structural Optimization 14(1997)63-69 presented the normal- boundary intersection method for multiobjective optimization. Today, these two papers have close to 100 citations in the Science Citation Index. and the NBI method has in few years become de facto standard for multi objective optimization. In this talk we will review the (close to) trivial observation that lead to the method and discuss some of its strengths and weaknesses seen from a computational view. We will discuss the need for better warm starts in the optimization routines and we review some of the more recent improvements in terms of scaling and introducing a filter.
Slides
Download (PDF)

[ Back | Home | Institute ]

Mathematical Characterization of the Pareto Front (and why simple scalarization is not good enough)

Speaker
A.Griewank
Institution
HU
Event
Lectures serie: Methods and Applications of Multiobjective Optimization
Date
27.4.
Time
5:30-7:00 pm
Location
Humboldt Universität zu Berlin, Institut für Mathematik
Rudower Chaussee 25, 12489 Berlin
Room 1.115
More Information
Lectures Homepage

[ Back | Home | Institute ]

Need and use of multiobjective optimization in large scale engineering applications

Speaker
N.Kroll
Institution
DLR
Event
Lectures serie: Methods and Applications of Multiobjective Optimization
Date
20.4.
Time
5:30-7:00 pm
Location
Humboldt Universität zu Berlin, Institut für Mathematik
Rudower Chaussee 25, 12489 Berlin
Room 1.115
More Information
Lectures Homepage
Abstract
Aerospace industry is increasingly relying on advanced numerical flow simulation tools in the early aircraft design phase. Today s flow solvers based on the solution of the compressible Euler and Navier-Stokes equations are able to predict aerodynamic behaviour of aircraft components under different flow conditions quite well. Due to the high computational expense required for flow simulations around realistic 3D configurations, in industry computational fluid dynamics tools are rather used for analysis and assessment of given geometries than for shape design and optimization. However, within the next few years numerical shape optimization will play a strategic role for future aircraft design. It offers the possibility of designing or improving aircraft components with respect to a pre-specified figure of merit subject to geometrical and physical constraints. Very often the design problems lead to multipoint and multi-objective optimizations. In order to address these large scale optimization problems efficient and robust optimization strategies are required. In terms of multi-criteria optimization the relative importance of the various objectives are often not known until the best solution candidates are determined and the tradeoffs between the objectives are fully understood. The concept of Pareto optimality can be used to characterize the objectives. The goal of the numerical optimization is then to provide a set of Pareto optimal solutions (Pareto front) that represent a trade-off of information amongst the objectives.

This lecture gives an overview of the requirements, status and needs of single-objective and multi-objective optimization problems in detailed aerospace design. At the end some promising techniques to calculate Pareto optimal points are discussed.

Slides
Download (PDF)

[ Back | Home | Institute ]

Exakte Mathematische Beschreibung und Robuste Optimierung von Zahnräder

Speaker
Prof. Dr. A. Griewank
Institution
Humboldt Universität zu Berlin, DFG Research Center Matheon
Date
Dienstag, 15.02.2005
Time
16.30 Uhr
Location
Humboldt Universität zu Berlin, Institut für Mathematik
Erwin Schrödinger-Zentrum
Konferenzraum 0.119

[ Back | Home | Institute ]

Mikrostrukturen, Adaptivität und Moore

Speaker
Prof. Dr. C. Carstensen
Institution
Humboldt Universität zu Berlin, DFG Research Center Matheon
Date
Dienstag, 15.02.2005
Time
17.15 Uhr
Location
Humboldt Universität zu Berlin, Institut für Mathematik
Erwin Schrödinger-Zentrum
Konferenzraum 0.119

[ Back | Home | Institute ]

Automatic Differentiation of Navier-Stokes Solvers

Speaker
Dr. Thomas Kaminski
Institution
FastOpt GbR Hamburg
Date
26.01.2005
Time
16:15
Location
Technische Universität Berlin, Institut für Mathematik
Straße des 17. Juni 136, 10623 Berlin
Room: MA 313
Abstract
Automatic Differentiation (AD) is a technique to evaluate derivatives of functions that are defined by numerical programmes. In contrast to traditional derivative approximation by divided differences, AD employs the chain rule to provide accurate derivatives. Basic concepts of AD such as forward and reverse modes are explained and illustrated using simple examples. The AD tools TAF and TAC++ are introduced, and a number of large-scale applications for sensitivity studies, state/parameter estimation, and uncertainty propagation are presented. The selected examples focus on codes for the simulation of the global oceanic and atmospheric circulation, aerodynamic flows as well as the terrestrial biosphere. Integrating an AD tool in such a modelling system allows for quick updates of the derivative code after modifications of the underlying model.

[ Back | Home | Institute ]

Semidefinite programming an its application in multi load free material optimisation

Speaker
K. Kulshreshtha
Institution
Institut für Mathematik
Humboldt-Universität
Date
20.01.2005
Time
13:00 Uhr
Location
Humboldt Universität zu Berlin, Institut für Mathematik
Rudower Chaussee 25, 12489 Berlin
Room 2.009
Abstract
The talk shall give an overview of the semidefinite programming problem and its dual with the duality and optimality conditions of the problems. A simultaneous primal-dual interior point method for the solution shall also be discussed. A recent application of this method to solve free material optimisation problem with multiple load and contact conditions shall be briefly discussed.

[ Back | Home | Institute ]

An Algorithm for Large-Scale Nonlinear Optimization with Applications in Circuit Tuning

Speaker
Ph.D. Andreas Wächter
Institution
IBM
Date
17.12.2004
Time
13:00
Location
Humboldt Universität zu Berlin, Institut für Mathematik
Rudower Chaussee 25, 12489 Berlin
Room 1.115
Information
Website
Abstract
An interior point algorithm for large-scale nonlinear, nonconvex optimization will be presented. The method is based on a primal-dual barrier approach, and uses a line search filter method to ensure global convergence. Details and recent developments of the optimization algorithm, as well as numerical results for its implementation IPOPT will be presented.
As a practical application of the method we will consider circuit tuning which is an essential step in the design of digital circuits. The central task is to find the optimal widths of transistors in order to minimize signal delay or area requirement. The problem can be formulated as a large-scale nonlinearly constrained nonlinear optimization problem, where function evaluations are obtained by simulation of gates (small subcircuits). This approach has been implemented in the IBM-internal circuit tuning tool EinsTuner, which is currently used for the design of every custom digital circuit designed by IBM.

[ Back | Home | Institute ]

ADTAGEO -- Algorithmic Differentiation Through Automatic Graph Elimination Ordering

Andreas Griewank, Jan Riehme

Speaker
Jan Riehme
Institution
HU Berlin
Date
16.12.2004
Time
13:00
Location
Humboldt Universität zu Berlin, Institut für Mathematik
Rudower Chaussee 25, 12489 Berlin
Room 2.009
Abstract
We report first results on a prototype implementation of the Live - DAG - concept : Instead of storing some representation of the complete Computational Graph and applying some elimination sequence afterwards, we dynamically maintain a DAG representing the active variables that are alive at any one time. Whenever an active variable is deallocated or its value overwritten the corresponding vertex in the live DAG is eliminated by the standard rules. Consequently, the total memory requirement is similar to that of the sparse forward mode. However, because local variables are destructed in the opposite order of their construction, single assignment stretches of a code are in effect differentiated in the reverse mode. More generally, the user determines the elimination order intentionally or unintentionally by the order in which he declares variables. There is no explicit initiation of sweeps, and derivatives between active variables can be determined and used at any time.

The new concept is very simple and fits optimally into the strategy of overloaded operators for classes, it's seems to be a very natural example of Object Oriented Programming.

The current status of the project is to be a proof of concept. Hence we use classes from Standard Template Library to store the DAG. Ongoing activities will target performance issues by replacing the STL containers with specialized classes.


[ Back | Home | Institute ]

Matheon - Workshop:

Algebraic Complexity Theory meets Algorithmic Differentiation
Institution
Department of Mathematics, Humboldt University Berlin
Date
December 09 - 10, 2004
Location
Humboldt Universität zu Berlin, Institut für Mathematik
Rudower Chaussee 25, 12489 Berlin
Room 1.410
Pictures

[ Back | Home | Institute ]


[ Back | Home | Institute ]

Das Adjungiertenverfahren in der aerodynamischen Formoptimierung

Speaker
Nicolas Gauger
Institution
DLR Braunschweig
Date
7.12.2004
Time
15:00
Location
Humboldt Universität zu Berlin, Institut für Mathematik
Rudower Chaussee 25, 12489 Berlin
Room 1.1012
Information
Website
Slides
PDF [2.4Mb]
Abstract:
Die so genannten Adjungiertenverfahren zur aerodynamischen Formoptimierung ermöglichen eine effiziente und genaue Bestimmung der Ableitungen der Zielfunktion der Optimierungsaufgabe nach den Entwurfsvariablen. Im Gegensatz zu den herkömmlichen Finite-Differenzen-Verfahren ist das Adjungiertenverfahren unabhängig von der Anzahl der Entwurfsvariablen hinsichtlich des numerischen Aufwandes. Es wird beim Adjungiertenverfahren nur eine Strömungs- und eine adjungierte Strömungsberechnung benötigt, was eine drastische Reduzierung an numerisch aufwendigen Strömungsberechnungen für eine hohe Anzahl an Entwurfsvariablen mit sich bringt.

Die Umsetzung des Adjungiertenverfahrens in den DLR-Strömungslösser FLOWer wird erklärt und dessen Anwendung an Beispielen der Profiloptimierung bis hin zur Optimierung eines kompletten Überschallverkehrsflugzeuges gezeigt. Anhand der aerodynamischen Optimierung eines 3D-Flügels unter Berücksichtigung der statischen Deformation wird die Erweiterung des Adjungiertenverfahrens für die Aero-Strukturoptimierung erläutert. Schließlich wird ein Ausblick gegeben, wie es auf diesem Forschungsgebiet weiter gehen kann und welche Fragestellungen dabei von Interesse sind.


[ Back | Home | Institute ]

Taylorarithmetic (Final project report)

Speaker
Ljerka Jukic
Institution
Croatia
Date
1.12.2004
Time
13:00
Location
Humboldt Universität zu Berlin, Institut für Mathematik
Rudower Chaussee 25, 12489 Berlin
Room 1.114

Jan Riehme
Last modified: Thu Apr 21 12:44:18 CEST 2005