Visualization of Magneto-rotational instability and turbulent angular momentum transport (Reloaded)
Table of Contents
Introduction
This project, led by by Fausto Cattaneo, University of Chicago, used a
previous allotment of 2 million processor-hours to study the forces that help
newly born stars and black holes increase in size. In space, gases and other
matter often form swirling disks around attracting central objects such as
newly formed stars. The presence of magnetic fields can cause the disks to
become unstable and develop turbulence, thereby causing the disk material to
fall onto the central object. This run at NERSC was used to set up initial
conditions for a larger scale simulations. The Visualization Group assisted
this project in generating High-quality visualizations of
data produced in these runs. Based on these initial results, the project
continues to carry out large-scale simulations to test theories on how
turbulence can develop in such disks.
The visualization group continues to work with the researchers to visualize
the results of their simulations. We use VisIt, which was initially developed at
LLNL and now serves as one of main deployment vehicles for state-of-the-art
production-quality visualization algorithms by the Visualization and Anautics Center for Enabling
Technologies. Our recent work focused on defining an appropriate file
format suitable for exchanging the results of these larger-scale simulations
and using VisIt for parallel voume rendering.
Data Exchange
To facilitate data exchange with Incite 4 researchers, we developed an
HDF5-based file format for their data and a corresponding VisIt database
plugin. Simulations are performed on a mesh specified in cylindrical
coordinates, which the plugin converts into a Cartesian representation at load
time. Furthermore, the plugin automatically splits the data into as many blocks
along the z-axis as there are processors used by VisIt, supporting VisIt's
parallel operation on these data sets.
First Light
As a first step we generated volume rendered images of the azimuthal
magnetic field (bt) and the magnetic energy (em), i.e., the square root of sum
of squares of the radial, azimuthal and axial components, using various
transfer functions. Figures 1-6 show the initial results.
Figure 1: First Attempt for bt and em.
Figure 2: Reverse background and revised transfer function for bt and em.
Figure 3: Cranking up the number of samples to 3500 Samples per ray for bt and em using the same transfer function as in Figure 2.
Figure 4: 5000 Samples per ray for bt and em, still the same transfer functions as in Figures 2 and 3.
Figure 5: 2000 and 3500 Samples per ray for fh with a slightliy different transfer function.
Figure 6: 2000 and 3500 Samples per ray for fm with a sliughtly different transfer function.
3D Visualization
After seeing these preliminary results, we were asked to generate
visualizations for another variable (vtd). Again, we experimented with
different transfer functions (Figures 7 - 9) until we settled on a final
transfer function and rendered a movie for several time steps in the
simulations. Since the volume rendered image contains large opaque regions, we
decided to "cut out" a quarter of the cylinder to make the center region of the
simulation visible.
Figure 7: Two different transfer functions for vtd. Clipped quarter of volume. 2000 samples per ray.
Figure 8: Two more transfer functions for vtd. Clipped quarter of volume. 2000 samples per ray.
Figure 9: (Left) Yet another transfer function for vtd. Clipped quarter of volume (using index select). 2000 samples per ray. (Right) Redone version of Figure 8 (right) with fixed transfer function min and max to help with movie generation.
Figure 10: Redone version of Figure 8 (left) with fixed transfer function min and max to help with movie generation.
Source Code
- First attempt at a VisIt Plugin to read data. Files should have .i4h5 as extension since .h5 is already taken by another database plugin. Use xml2makefile to create makefile for your visit setup.
- VisIt Plugin used to read and automatically split data into as many blocks as processors are used to compute the image. Files should have .i4h5 as extension since .h5 is already taken by another database plugin. The README file contains installation instructions.
References
- MHD computations by F. Cattaneo, P. Fischer and A. Obabko on NERSC resources supported by DOE under Contract No. DE-AC03-76SF00098
- Slides with more details of the project can be found here and here.