Sie sind hier

## Absolventen-Seminar • Numerische Mathematik

Absolventen-Seminar
Verantwortliche Dozenten:
Prof. Dr. Christian MehlProf. Dr. Volker Mehrmann
Koordination:
Benjamin Unger
Termine:
Do 10:00-12:00 in MA 376
Inhalt:
Vorträge von Diplomanden, Doktoranden, Postdocs und manchmal auch Gästen zu aktuellen Forschungsthemen
Sommersemester 2015
Datum
Zeit
Raum
Vortragende(r)
Titel
Do 16.04.
10:15
Uhr
MA 376
Vorbesprechung
Matthew Salewski
Solutions of turbulent flows [abstract]
Do 23.04.
10:15
Uhr
MA 376
Andreas Steinbrecher
QUALIDAES - A Software Package for the Numerical Integration of Quasi-Linear DAEs [abstract]
Peter Kunkel
Optimal control for DAEs, formal adjoints and formal optimality conditions [abstract]
Mi 29.04.
9:15 Uhr
MA 376
Matthias Voigt
On Linear-Quadratic Optimal Control and Robustness of Differential-Algebraic Systems [abstract]
Do 30.04.
10:15
Uhr
MA 376
Balmohan V. Limaye
Condition Numbers of Bases [abstract]
Philipp Schulze, Benjamin Unger
A step towards the realization problem for retarded descriptor systems [abstract]
Do 07.05.
10:15
Uhr
MA 376
- kein Seminar -
Conference in Honor of Volker Mehrmann on the Occasion of his 60th Birthday
Do 14.05.
10:15
Uhr
MA 376
- kein Seminar -
Christi Himmelfahrt
Do
21.05.
10:15
Uhr
MA 376
Robert Altmann
Regularization and Simulation of Constrained PDEs [abstract]
Mark Embree
Interpolatory Matrix Approximations [abstract]
Do
28.05.
10:15
Uhr
MA 376
Olga Markova
Lengths of matrix sets with quasi-commuting elements [abstract]
Do 04.06.
10:15
Uhr
MA 376
Felix Held
Optimal control of mechanical multi-body systems [abstract]
Volker Mehrmann
Optimal control of delay differential-algebraic equations [abstract]
Do 11.06.
10:15
Uhr
MA 376
Fredy Sosa
On  first order asymptotic expansions for multiplicative perturbation of eigenvalues [abstract]
Lia Strenge
Modeling and simulation of a droop controlled swarm type low voltage DC microgrid in a DAE framework [abstract]
Do 18.06.
10:15
Uhr
MA 376
Melina Freitag
Data assimilation as a very large inverse problem - an introduction [abstract]
Philipp Schulze
Data-Driven Model Reduction of Linear Port-Hamiltonian Systems [abstract]
Do 25.06.
10:15
Uhr
MA 376
Jeroen Stolwijk
Numerical Solution and Error Analysis for the Euler Equations in Stationary Form [abstract]
Jon Paul
Analysis and solution of dynamic flash equations [abstract]
Di 30.06
11:15 Uhr
MA 415
Thomas Berger
Regularization of linear descriptor systems [abstract]
Zoran Tomljanović
Optimization of semi-active damping and external damping in mechanical systems with external force [abstract]
Do 02.07.
10:15
Uhr
MA 376
Leonhard Batzke
Low-rank perturbations of structured matrix pencils [abstract]
Benjamin Unger
Delay DAEs and their regularization [abstract]
Do 09.07.
10:15
Uhr
MA 376
Christoph Zimmer
On Linear Operator-differential-algebraic Equations with Delay [abstract]
Daniel Bankmann
OVDBDF - A Software Package for the Numerical Integration of Differential Algebraic Equations [abstract]
Michael Götte
Backend design of a Modelica compiler [abstract]
Do 16.07.
10:15
Uhr
MA 376
Ute Kandler
A certain Gram-Schmidt variant - Compensated Gram-Schmidt [abstract]
A DAE Modeling Approach toward the Crosstalk Phenomenon [abstract]

## Abstracts zu den Vorträgen:

### Matthew Salewski (TU Berlin)

Donnerstag, 16. April 2015

Solutions of turbulent flows

In the dynamical systems approach to hydrodynamics, turbulence is seen as the chaotic path that the state vector traces out in the space of allowed solutions to the equations of motion, e.g. Navier-Stokes equations. Rather than a random walk in this space, the approach contends that there are a set of equilibria and periodic orbits that organize the state-space trajectory via their unstable manifolds. From this follows a belief (or hope) that this set is approximately finite, with only a few significant elemental solutions needed to accurately capture the dynamics of turbulent flows. In my talk I will (re-)introduce these topics and place them in the context of turbulent rotating shear flows, where there is growing evidence of such solutions dictating the global statistics of the fluid dynamics even at the highest measured intensities for turbulent flows.

### Andreas Steinbrecher (TU Berlin)

Donnerstag, 23. April 2015

QUALIDAES - A Software Package for the Numerical Integration of Quasi-Linear DAEs

The behavior of dynamical systems often are modeled with differential-algebraic equations (DAEs) in quasi-linear form

(1)   E(x,t)x'=f(x,t)

as model equations with the vector of unknowns x and the time t. Unfortunately, in general the direct numerical integration of DAEs is not feasible due to so-called hidden constraints

(2)   0=h(x,t).

They are contained in the DAE but not explicitly stated as equations. The occurrence of hidden constraints leads to difficulties like instabilities or order reduction in the numerical integration. Hence, before a robust numerical integration is possible it is necessary to regularize or remodel the model equations.

In this talk we will present the software package QUALIDAES for the numerical integration of quasi-linear DAEs of the form (1) which covers the model equations of many dynamical processes. A key point for the robust and efficient numerical integration in QUALIDAES is the precise consideration of the hidden constraints (2). Therefore, the approach implemented in QUALIDAES is based on the interaction of a regularization of the DAE (1) with an efficient numerical treatment of the regularization in form of an overdetermined fomulation.

### Peter Kunkel (Universität Leipzig)

Donnerstag, 23. Oktober 2014

Optimal control for DAEs, formal adjoints and formal optimality conditions

Deriving necessary conditions for the solution of linear-quadratic optimal control problems for differential-algebraic equations (DAEs) with arbitrary index, one should replace the given DAE by an associated index-reduced DAE. Possessing the same (smooth) solutions as the original DAE, the reduced DAE allows for a suitable solution operator for the application of abstract results from optimization theory. The resulting necessary conditions are then in terms of the reduced DAE. It was, however, observed that formally replacing the reduced DAE by the orginal DAE may still make sense to some extent. It shall therefore be discussed how this formal approach is related to the necessary conditions and to what extent this relation can be utilized.

### Matthias Voigt (TU Berlin)

Mittwoch, 29. April 2015

On Linear-Quadratic Optimal Control and Robustness of Differential-Algebraic Systems

In this talk, I will practise my PhD defense presentation. The main focus will be on linear-quadratic optimal control and its relation to descriptor Kalman-Yakubovich-Popov (KYP) inequalities, Lur'e equations and even matrix pencils. Further results on robustness questions will be touched.

### Balmohan V. Limaye (Indian Institute of Technology Bombay)

Donnerstag, 30. April 2015

Condition Numbers of Bases

Let X=[x_1,...,x_m] be an ordered basis of a finite dimensional subspace of L^p (1<= p <=\infty). The condition number \kappa_p(X) of X is known to measure the sensitivity of computing the coefficients in the expression of a function in L^p as a unique linear combination of the basis elements.

We consider an ordered basis X of an arbitrary  finite dimensional normed space M, and introduce a number cond(X) which controls `near linear dependence' of the basis elements as well as overflow and underflow during computations involving the basis. We also describe optimal scaling strategies for cond(X).

If X is an ordered basis of an inner product space, then cond(X) can be calculated explicitly in terms of the diagonal entries of the Gram matrix corresponding to X and the diagonal entries of its inverse.

The number cond(X) can be characterized in terms of the norms of the basis elements and the norms of the elements of the ordered dual basis of M'. In fact, cond(X) measures the sensitivity of the problem of finding the unique ordered basis of M' which is dual to the basis X.

A  comparison of  cond(X) and \kappa_p(X) shows how intimately they are related, although they measure reliability of computations from  different perspectives.

### Philipp Schulze, Benjamin Unger (TU Berlin)

Donnerstag, 30. April 2015

A step towards the realization problem for retarded descriptor systems

We introduce a method to construct low dimensional descriptor systems with retarded argument of the form

(1a)    Ex'(t) = A_1 x(t) + A_2 x(t-\tau) + Bu(t),
(1b)      y(t) = Cx(t),

directly from measurements at low computational cost. The key tool is a generalization of the Loewner and shifted Loewner matrix to allow interpolation of the transfer function in the delayed setting. We show that our approach extends the well-known Loewner framework in the sense that they coincide for vanishing delay $\tau=0$. A family of reduced-order models for (1) based on Moment Matching is introduced and we show its strong connection to the generalized Loewner approach.

### Robert Altmann (TU Berlin)

Donnerstag, 21. Mai 2015

Regularization and Simulation of Constrained PDEs

In this talk, I will practise my PhD defense presentation. The main focus will be on operator DAEs. We discuss the formulation of such systems, consistency conditions, a regularization, and the resulting advantages for the simulation of those systems.

### Mark Embree (Virginia Tech, VA)

Donnerstag, 21. Mai 2015

Interpolatory Matrix Approximations

Interpolatory matrix factorizations provide alternatives to the singular value decomposition for obtaining low-rank approximations; this class includes the CUR factorization, where the C and R matrices are formed from a subset of columns and rows of the target matrix.  While interpolatory approximations lack the SVD's optimality, their ingredients are easier to interpret than singular vectors: since they come directly from the matrix itself, they inherit the data's key properties (e.g., nonnegative/integer values, sparsity, etc.). We shall provide an overview of these approximate factorizations, describe how they can be analyzed using interpolatory projectors, and introduce a new method for their construction based on the Discrete Empirical Interpolation Method (DEIM).  This talk describes joint work with Dan Sorensen (Rice).

### Olga Markova (Moscow State University)

Donnerstag, 28. Mai 2015

Lengths of matrix sets with quasi-commuting elements

Given a finite set  S of square  matrices over a field, we consider the linear span A of all products in S and define the length of S to be the least non-negative integer k such that the products in these generators of lengths not exceeding k span A.

In this talk we  discuss  the known bounds for the lengths of matrix sets which elements pairwise quasi-commute (i.e. commute up to a factor depending on the matrices).   We also consider the realizability problem for the lengths of such sets consisting of two matrices.

### Volker Mehrmann (TU Berlin)

Donnerstag, 04. Juni 2015

Optimal control of delay differential-algebraic equations

We discuss the solution of optimal control problems with linear delay differential-algebraic equation (DDAE) constraints. Compared to the already complicated case of standard differential-algebraic equation (DAE) constraints, several further difficulties arise.

These include the decreased regularity in the solution and the occurrence of higher derivatives of the input functions. We present the necessary optimality conditions under the assumption that the method of steps leads to a reasonable solution and discuss the algebraic properties of the optimality system.

Joint work with Peter Kunkel

### Felix Held (TU Berlin)

Donnerstag, 04. Juni 2015

Optimal control of mechanical multi-body systems

In this talk, I will motivate the topic of my master thesis and outline the employed methods and goals. The mechanical model is presented and the derivation of its equations is discussed. Furthermore some remarks on their structure will be given. I also will focus on possibilities to solve these equations numerically.

Donnerstag, 11. Juni 2015

On first order asymptotic expansions for multiplicative perturbation of eigenvalues

Let A be a complex matrix with arbitrary Jordan structure, and \lambda an eigenvalue of A whose largest Jordan block has size n. Based on the use of the Newton diagram, it has been shown that for a small multiplicative perturbation \hat{A}=(I+\epsilon C)A(I+\epsilon B) of the matrix A, the splitting of \lambda under this perturbation is, generically, of order \epsilon^{1/n} if \lambda\neq 0.  Explicit formulas for the leading coefficients are obtained, involving the perturbation matrices B and C and the eigenvectors of A. In the special case of \lambda=0, similar results has been found for leading coefficients in the splitting of \lambda in this case the splitting of \lambda, is generically, of order \epsilon^{\frac{1}{n+1}}.

Joint work with Julio Moro Carreno.

Abstract as PDF.

### Lia Strenge (TU Berlin)

Donnerstag, 11. Juni 2015

Modeling and simulation of a droop controlled swarm type low voltage DC microgrid in a DAE framework

In this talk, I present the preliminary results of my Master's thesis. We derive a model of a droop controlled swarm type low voltage (LV) direct current (DC) microgrid and discuss simulation methods and results for the derived model. This type of microgrid is a technical solution for the bottom-up electrification of the off-grid population in the Global South (referring to less economically developed countries).

Mathematically, the closed-loop model is a stiff strangeness-free differential-algebraic equation (DAE) system. It can be of arbitrarily high (finite) dimension due to the generic topology of the grid being modeled by an undirected graph. In addition, the modeling allows for linear and nonlinear representation and includes hybrid (switching) characteristics. Regarding the simulation, the reaction of the voltages and power flows to changes in consumption (i.e. loads) for an example topology of the grid are of major interest to understand the behaviour of swarm type LVDC microgrids and their control. We discuss first results on numerical and technical level obtained with the DAE solvers DASSL (in Dymola/Modelica) and QUALIDAES.

### Melina Freitag (University of Bath)

Freitag, 19. Juni 2015

Data assimilation as a very large inverse problem - an introduction

In this talk we aim to provide a theoretical framework for data assimilation, a specific type of an inverse problem arising for example in numerical weather prediction, hydrology and geology. We consider the general mathematical theory for inverse problems and regularisation, before introducing Tikhonov regularisation as one of the most popular methods for solving inverse problems. We show that data assimilation techniques such as 3DVar and 4DVar as well as the Kalman filter and Bayes' data assimilation are, in the linear case, a form of cycled Tikhonov regularisation. We give an introduction to key data assimilation methods as currently used in practice and explain computational challenges.

### Philipp Schulze (TU Berlin)

Donnerstag, 18. Juni 2015

Data-Driven Model Reduction of Linear Port-Hamiltonian Systems

The main idea of data-driven model reduction is to create a small-dimensional model from data of numerical or real-world experiments without needing a state-space representation in the beginning. However, in general the obtained realization does not exhibit any structure or related properties of the original system.

In this talk we extend the Loewner framework for data-driven model reduction to preserve the structure of linear port-Hamiltonian systems. This provides the realization with both physical meaning and mathematical properties as stability and passivity. An example illustrates how to obtain a port-Hamiltonian realization only from input-output data.

Joint work with Arjan van der Schaft

### Jeroen Stolwijk (TU Berlin)

Donnerstag, 25. Juni 2015

Numerical Solution and Error Analysis for the Euler Equations in Stationary Form

Natural gas plays a crucial role in the energy supply of Europe and the world. After oil, it is the second most used energy supplier in Germany. The high and probably increasing demand for natural gas calls for a robust mathematical modeling, simulation and optimisation of the gas transport through the existing pipeline network.

Natural gas transportation is commonly modelled by the one-dimensional Euler equations. Although this is a widely accepted model in academia, we will see in this presentation that even the great mathematician Leonhard Euler himself encountered difficulties applying this model in a practical engineering problem.

Several simplifications of the Euler equations lead us to a system with three stationary partial differential equations together with two algebraic constraints. We will numerically solve this system using a first order forward difference scheme, Newton's method and adequate boundary and parameter values. Finally, we will perform an error analysis for the impulse equation by determining the magnitude of the discretisation, rounding and data errors as well as their amplification in the computation of the pressure.

Joint work with V. Mehrmann.
Supported by the German Research Foundation DFG.

### Jon Paul Janet (TU Berlin)

Donnerstag, 25. Juni 2015

Analysis and solution of dynamic flash equations

In this talk, I will present preliminary results from my master’s thesis as part of the COSSE program.  The topic concerns the simulation and control of two-phase flash separators, which are important and ubiquitous components of many industrial processes.  The topic will be motivated, and two dynamic models proposed in literature will be presented: a full model and one based on simplifying physical assumptions. In general, the system is described by a hybrid physical/empirical system of DAEs with a high degree of nonlinearity.

The index of the models will be analysed in a behaviour setting and the impact of the dependencies of the empirical terms is explored, resulting in deriving explicit conditions for both systems to be index 1. It is claimed in the literature that the simplified model has an increased d-index and this can be clearly understood from the behaviour setting.

Finally, some numerical results for a test problem are presented and analysed.  Both systems and two solvers are compared.

### Thomas Berger (Universität Hamburg)

Montag, 29. Juni 2015

Regularization of linear descriptor systems

For linear time-invariant descriptor systems we consider different regularization approaches. First, the question whether there exists a feedback which renders the closed-loop system regular is considered. This property can be equivalently characterized by simple algebraic and geometric conditions in terms of the involved matrices and the augmented Wong sequences. We also consider the slightly more general problem of existence of a feedback such that an autonomous closed-loop system is obtained. For systems which are not regularizable by feedback, an additional behavioral equivalence transformation and a  reorganization of input and state variables leads to a regular system, the index of which is at most one. This procedure is known (see Campbell et al., Regularization of linear and nonlinear descriptor systems, 2012) and we present a new approach which allows for a detailed characterization of the resulting regular system. In particular, this system is fully determined by the augmented Wong sequences. The aforementioned result can be further improved and we show that a feedback is actually not necessary. To this end, we provide an algorithmic procedure for the construction of the regularization and discuss computational aspects.

### Zoran Tomljanović (University J.J. Strossmayer in Osijek, Croatia)

Montag, 06. Juli 2015

Optimization of semi-active damping and external damping in mechanical system with external force

We present an efficient approach for determination on an optimal semi-active damping and we also consider damping optimization in systems excited by external force.

First we study the problem of determining an optimal semi-active damping of vibrating systems. For this damping optimization we use a minimization criterion based on the impulse response energy of the system. In this case the optimization approach yields a large number of Lyapunov equations which have to be solved, thus we propose an optimization approach that works with
reduced systems which accelerate optimization process. Reduced systems are generated using the parametric dominant pole algorithm. The optimization process is additionally accelerated with a modal approach while the initial parameters for the parametric dominant pole algorithm are chosen during optimization procedure using residual bounds. Our approach calculates a
satisfactory approximation of the impulse response energy while providing a significant acceleration of the optimization process.

In the second part we consider optimization of external damping in mechanical system excited by external force. We introduce two criteria based on the minimization of the energy functions, that allow a damping optimization in mechanical systems with external force. This optimization problem is a very demanding due to the numerous linear systems that have to be solved.
For that purpose we have derived the new formulas which allow us to calculate energy functions very efficiently.

Numerical results illustrate the effectiveness of the proposed approaches.

Joint work with Peter Benner, Patrick Kürschner, Krešimir Veselić and Ninoslav Truhar.

Abstract as pdf.

### Leonhard Batzke (TU Berlin)

Donnerstag, 02. Juli 2015

Low-rank perturbations of structured matrix pencils

In this talk, I will present several results from my thesis on generic low-rank perturbations of structured regular matrix pencils. The focus will be on structure-preserving rank-1 perturbations of T-alternating matrix pencils. I will also briefly touch on other types of structured matrix pencils and perturbations of higher rank.

### Benjamin Unger (TU Berlin)

Donnerstag, 02. Juli 2015

Delay DAEs and their regularization

We recall the concept of well-posedness in terms of delay differential- algebraic equations (DDAE) and outline the need for a regularization procedure. Based on a review of the existing procedure for linear time-invariant DDAEs, introduced in [Ha, Mehrmann, 2014], we introduce a new regularization methodology that is a generalization of the strangeness index concept for DAEs [Kunkel, Mehrmann, 2006] and allows for an efficient determination of the strangeness and shift index.

### Christoph Zimmer (TU Berlin)

Donnerstag, 16. Juli 2015

On Linear Operator-differential-algebraic Equations with Delay

Constrained linear partial differential equations (lin. PDAEs) have an important role in modeling practical systems such as the incompressible linearized Navier-Stokes equation. On the other hand, time-delays occur naturally in closed-loop controlled dynamical systems, since measurements, signal transmissions, and calculations of the control require a certain time. The combination of lin. PDAEs and time-delays leads to a new mathematical object, which in includes a various number of challenges. In this talk, we investigate this kind of object in the abstract setting of linear operator-differential-algebraic equations. In particular we consider the unsteady Stokes equation with and without delay and show existence results.

### Daniel Bankmann (TU Berlin)

Donnerstag, 16. Juli 2015

OVDBDF - A Software Package for the Numerical Integration of Differential Algebraic Equations

Differential algebraic equations (DAEs) arise in many applications as multi-body systems or networks (e.g. electrical circuits) when modeling their dynamicalbehavior and can be obtained in particular via automatic modeling. We consider DAEs in the most general form

(1) F(t,x,x') = 0,

where t is the independent variable, x the state vector and x' its derivative and the partial derivative of F with respect to x' is possibly singular.

In this context the so-called hidden constraints of the system play an important role in terms of numerical robustness. These are constraints that are not explicitly given in the equations (1). Failing to provide these constraints explicitly might lead to numerical drifts. There exist various (numerical) techniques to determine these constraints, e.g. the procedure mentioned in [1] for quasi-linear DAEs. Thus, we can additionally impose

(2) G(t,x) = 0,

where G contains all the explicit and hidden constraints of the original system F, without changing the solution set.

In this talk we present the solver OVDBDF for the numerical integration of such an overdetermined system fulfilling (1) and (2) that is based on the DASSL code for square nonlinear DAEs using backward difference formulas (BDF). A similar approach has been taken by QUALIDAES for quasi-linear DAEs using Radau IIA methods. Besides a sophisticated stepsize and order control a key point for the robust and efficient numerical integration in OVDBDF is the precise consideration of the hidden constraints (2).

[1] Steinbrecher,A. Analysis of Quasi-Linear Differential-Algebraic Equations. Institut für Mathematik, Technische Universität Berlin, Berlin, Germany, number 11-2006. 2006.

### Michael Götte (TU Berlin)

Donnerstag, 09. Juli 2015

Backend design of a Modelica compiler

AMSUN is an interdisciplinary project combining modeling, compiling and simulation on the basis of Modelica. In my short talk I will present my work as a student assistant designing an C-interface for solvers like QUALIDAES and ovdBDF. The focus is dealing with automation. For that we are testing different ideas for index reduction based on the method of Pryce and automatic differentiation. This new approaches are exploiting the fact that the two solvers can deal with overdetermined systems.

### Ute Kandler (TU Berlin)

Donnerstag, 16. Juli 2015

A certain Gram-Schmidt variant - Compensated Gram-Schmidt

In various applications like tensor calculus or mixed precision arithmetic vector operations like matrix-vector multiplication, summing, scaling and inner products can not be evaluated exactly. We investigate the behavior of the QR decomposition using different variants of the Gram-Schmidt orthogonalization scheme. In particular, we introduce a variant, called the Compensated Gram-Schmidt orthogonalization, that  uses a slightly different projector to alleviate the damage, the perturbations inflict on the orthogonality of Q. We compare this new variant with the well known classical and modified Gram Schmidt methods.

### Helia Niroomand Rad (TU Berlin)

Donnerstag, 16. Juli 2015

A DAE Modeling Approach toward the Crosstalk Phenomenon

In this talk, we mainly propose a modeling approach in a general framework in order to describe the crosstalk phenomenon in electro-magnetic systems, and in particular, within electrical circuits. Crosstalk in an electrical circuit, loosely speaking, refers to the undesirable disturbance coupling in between the electrical elements and its effect on the entire circuit.

We model the phenomenon by bilateral coupling of two sets of differential equations where the main set is the circuit equations formulated in the framework of modified nodal analysis, and the second set consists of the non-stationary Maxwell equations. By spacial discretization of the second set, e.g., the non-stationary Maxwell equations, we obtain a large set of differential-algebraic equations (DAEs) modeling the crosstalk phenomenon.

In the last part, we shortly focus on analysis of the DAE system corresponding to the non-stationary Maxwell equations.