Tag Archive for 'Octave'

Octave 3.4.2 on openSUSE 11.4

Finally, the science repository for openSUSE comes with a recent Octave version that supports UMFPACK. Add the corresponding repository

sudo zypper ar http://download.opensuse.org/repositories/science/openSUSE_Tumbleweed/ opensuse:science

and install the package

sudo zypper in octave

You verify that UMFPACK works by the following computation

A=sprandsym(1000,0.01)+eye(1000);b=rand(1000,1);norm(A*(A\b)-b)

which solves a linear system with a sparse symmetric positive definite system matrix A and prints the residual to the console.

Continue reading ‘Octave 3.4.2 on openSUSE 11.4′

Computational Magnetics

Eddy Currents

Next term the chair of numercial analysis, Bergische Universität Wuppertal, offers a lecture on Computational Magnetics. The aim is to develop an Octave software package to simulate magnetoquasistatic fields.

All the typical tasks of applied mathematics will be covered: modeling, software design, coding, testing, simulation und visualization.

Compile Octave on openSUSE 11.3 [Update 2]

Update: The science package for openSUSE contains a recent and working Octave package, see this blog post. It is not necessary to compile your own package.

There is no adequate octave package for openSUSE with SuiteSparse support, therefore one has to compile it manually. The following steps will install all dependencies for Octave 3.2.4, then compile octave, create a rpm package and install it on openSUSE 11.3.

Continue reading ‘Compile Octave on openSUSE 11.3 [Update 2]‘

Flux Distribution in a Transformer

I have written a simulation package FIDES to test multi rate time integration in coupled electromagnetic field and electric circruit equations. The package is part of the COMSON Demonstrator Platform, using OCTAVE as its interpreter.
Recently I wanted to visualize my results. I used ffmpeg to produce a movie from several plots of the flux distribution inside a transformer.

Continue reading ‘Flux Distribution in a Transformer’

Einbettungsverfahren

Abbildung des Konvergenzbereich des Arcustangens-BeispielsUm Nullstellen von nicht-linearen Problemen zu bestimmen, benötigt man Startwerte in der Nähe der Lösung. Wenn man diese nicht kennt, kann man versuchen das Newton-Verfahren zu globalisieren, zum Beispiel mit einem Einbettungsverfahren. Das eignet sich besonders für praxisnahe Probleme, die von einem ausgezeichneten Parameter abhängen. Dieser Steuerungsparameter kann zum Beispiel bei der Modellierung eines naturwissenschaftlichen Prozesses die Zeit repräsentieren. Details finden sich im Buch Newton Methods for Nonlinear Problems: Affine Invariance and Adaptive Algorithms von Peter Deuflhard, oder in einer Zusammenfassung, die ich als Seminararbeit angefertigt habe.