Scientific Pictures
Roy Williams
Click on the thumbnail for full-size image
This is the Crab Nebula in Orion, constructed as an overlay of the
X-ray image from the Chandra observatory on an optical image from Palomar.
(Also available is a
2100x2100 jpg and
2100x2100 tif).
More information about the science of this image can be found here.
These pictures show a parallel 3D adaptive unstructured mesh.
The first two images show the pressure on the outside of a pipe of square
cross section, that has a slot in it. Fluid moves from the top right to the
bottom left; the main bow shock is reflected
multiple times as the fluid moves down the pipe.
The third image shows the surface of the unstructured tetrahedral mesh
that was used for the computation; the mesh has self-adapted to the pressure
gradient, meaning a finer mesh at the shocks. The fourth picture shows
the decomposition of the mesh among 512 parallel processors: each colour
represents a differentr processor. Adaptivity is evident here also, as
each processor has an equal number of tetrahedral mesh cells.
(with John Flower)
These pictures show 2D triangular meshes adapted through a rigourous
a posteriori error estimation. The equation being solved is the archetypal
reaction-diffusion equation; the bistable equation has two stable states
separated by a thin transition region -- many thin black contour lines that
merge into a single thick line. The mesh is adapted and load-balanced
as above.
(with Don Estep)
In converting a multiblock mesh into a tetrahedral mesh, each block must be
split into tetrahedra, inducing a splitting of the faces of the block into
triangles. Here we see the five distinct ways in which this can be done,
together with two more splittings of the faces that do not correspond
to a volume splitting.
This 2D area has been meshed and refined by the Dime software, then split
among 32 processors of a parallel machine. The two pictures show the
results of load-balancing by means of two algorithms: the first image by
simulated annealing, the second by orthogonal recursive bisection. The quality
of the load-balancing, measured by the number of triangle pairs that
are the same colour, is much better for the more expensive annealing
algorithm.
These images show the results for a load-balancing algorithm based on
eigenvectors of the adjacency matrix of the mesh graph. The mesh surrounds
a simple airfoil. The first picture shows the eigenvector itself for the
last stage of the load-balance process; the second image shows the result
of the load-balance itself.
These images show the results of a study of dynamic load balancing. A pair of
Gaussian peaks move randomly around the pentagon, and the adaptive mesh attempts
to keep up with it, then the mesh load balance responds in turn to the
adaptivity of the mesh.
Two-dimensional plane-strain computations, using an adaptive mesh.
Adaptive-mesh computation of compressible flow. The third pictures shows the
adapted mesh as well as the pressure field.
This highly relaxed mesh surrounds a 2D shape. The mesh has been adapted
by solving a Laplace equation in the space, with the surface charge being
proportional to the curvature of the surface.
Thes representations of three-dimensional unstructured meshes show the
different dimensionalities of the components: point, line, face, tet.
Unstructured mesh computations around a simple airfoil, with orthogonal
recursive bisection for the load balancing.
These tetrahedral meshes filling a sphere look nice.
This reminds me of a body in a coffin surrounded by muslin.
Brought to you by Roy Williams
Clickery.
