
Gallery
Here are some results obtained with the
DUNE-FEM discretization module.
The first picture shows the solution to the Laplace equation on a locally
adapted grid (100.000 triangles) using first order polynomials. The second picture shows the
results for the same problem but using four cube elements and polynomial
degree 6. Both have about the same discretization error.
We solve the heat equation on a moving surface using a first order Lagrange
space. The color show the solution which is initially constant and then
increases in the regions where the surface is compressed.
The next is mean curvature flow of a surface. The colors represent the
distance of the surface from the origin.
Here we consider a stabilized Discontinuous Galerkin (DG) method for
hyperbolic and convection dominated problems. The presented scheme can
be used in several space dimensions and with a wide range of grid
element types for locally adaptive and load balanced
parallel computations. More details about the scheme can be found
here.
First we show a 2d computation of a forward facing step using YaspGrid
(left) and ALUGrid (right):
The following is the same code using ALUGrid with local adaptivity on 64
processors:
The last example shows the solution to the same problem but using a finite-volume
and a special filtered grid view included in DUNE-FEM - part of the grid is
removed by prescribing a filter excluding the element inside a half sphere:
This work was partially funded by the Landesstiftung Baden-Würtemberg..
The following movie shows a cold bubble set in a stratified atmosphere.
It falls down and slides along the ground, creating Kelvin-Helmholtz vortices.
A 5th order compact discontinuous Galerkin 2 (CDG2) method with a 3rd order
explicit Runge-Kutta scheme is used to solve the compressible Navier-Stokes
equations.
This is part of a DFG funded project.
More details about the implementation together with further atmospheric
test cases can be found
here.
The page also includes some comparison of the DUNE-FEM code with the
production code developed at the German weather service.
Two surfaces are matched by some minimization procedure. Using
preregistered images damaged parts can be reconstructed. The resulting
model is solved by gradient flow, resulting in a parabolic equation with a
non-local term. We used an adaptive DG method of second order.
In the following pictures
blue is the target image,
black is the image to be registered, and
red is the resulting registration
The first picture show the results without using preregistered images and
the second picture is with the statistical model included.
The following shows the registration of a femur bone. Again a damaged
specimen is being registered once without and once with statistical
information.
In the last example the method is used to transform a marking (coloring of
the mandible) from the reference image to a newly registered image.
In cooperation with the group of Thomas Vetter at the University of Basel.
A 3D version of the shallow water model is solved. The time dependent
domain is transformed to a fixed domain and the prismatic grid is used to
facilitate the integration over water columns. The colors represent the
magnitude of the horizontal velocity vector.
More details about the implementation and the settings used can be found
here.
This was part of a diploma thesis at the University of Freiburg.
The following movie shows a wave propagating through a realistic river bed.
A 2D shallow water model is solved, taking into account wetting and drying
and bottom friction.
Work funded by the BMBF project.
The shallow water equations is solved on the sphere including bottom
topography. The movie shows the first 15 hours of the propagation
of tsunami waves caused by an asteroid impact in the middle of the Atlantic Ocean.
The overall time maximum of the wave height above sea level is colored,
where
red depicts a wave height of at least 10m, and
blue depicts 0m.
The simulation was performed on an triangular surface grid with more than
3 million elements in a parallel run on 16 cores. The picture shows the
partitioning of the sphere.
More details about the implementation and the settings used can be found
here.
This was part of a diploma thesis at the University of Freiburg.