Skip to content



Visualization Tools to Guide Scientists
Through Their Simulation Results

Chandrajit Bajaj, Professor of Computer Sciences, Visualization Chair, Texas Institute for Computational and Applied Mathematics, Director, Center for Computational Visualization, TICAM,University of Texas

Computer simulations can help engineers drill for petroleum and natural gas more efficiently, show environmental scientists how pollution spreads in surface or underground water, and detail for biologists and chemists the workings of nucleic acids, proteins, and other biological molecules. But as simulations grow larger and more accurate, the resulting data sets grow larger and more intricate, and scientists themselves need a road map for analyzing these results and guiding them to the regions of interest. Chandrajit Bajaj of the University of Texas leads a project of the Interaction Environments thrust area that is building such a set of analysis and visualization tools to be used in a tight loop with large simulations.

"What we've developed is a set of core multiscale domain creation and visualization technologies," said Bajaj, professor of computer sciences and director of the Center for Computational Visualization (CCV) at the Texas Institute for Computational and Applied Mathematics (TICAM). "These affect both the preprocessing and post-processing of physics simulations for many disciplines."

The tools from Bajaj's team provide the foundation for meeting the needs of computational scientists who develop simulation software to address a new problem. The CCV tools accelerate the process of writing the companion software that both models the geometry of the problem domain and visualizes the results of their physics simulations.

Helical turbulence vortexFigure 1: Accelerated Isocontouring

CCV tools allow scientists to quickly create images such as this vorticity field from a collaboration with Greg Blaisdell of Purdue University. Through the helicoidal vortex (green), two orthogonal slices are colored with the intensity (red high, blue low) to show how the vortex evolves.



The CCV parallels in many ways the mission of the Interaction Environments thrust area. Bajaj and his colleagues work with scientists from various disciplines to integrate the visualization technologies and help advance their applications. Within the CCV, project Visual Eyes develops the core technologies--in particular, tools for adaptive, hierarchical mesh generation; accelerated isocontouring; hypervolume rendering; and scalar and vector topology analysis.

The mesh generation tools start with a set of 2-D images, such as slices from magnetic resonance imaging (MRI) scans, and extract 3-D geometry data from them. The object's geometry is expressed as a mesh, a lattice of tetrahedra or blocks, called finite elements.

"Such imaging data can be very dense, and the goal is to extract the geometry adaptively, providing high-resolution only where necessary," Bajaj said. "And by hierarchical, we mean that you can access higher-resolution meshes from coarser versions where needed. The challenge is to do it instantly for larger and larger data sets."

Simulation results must be plotted onto such underlying finite element meshes. The accelerated isocontouring tools analyze the physical features in the results--such as flow, temperature, stress, or electromagnetic fields--and plot isocontours on the mesh; points on the mesh with the same value are given the same colors and transparency. This is a basic visualization technique, and one that is used often by scientists for 2-D and 3-D simulation data. The CCV has developed methods with exponential speed-up over prior approaches (Figure 1).

Hypervolume rendering methods are used when simulations produce 3-D, or higher-dimensional, results that must be displayed on a 2-D screen. These methods project data sets onto standard color displays, taking into account the entire data set in the projection function. Coupling a fast projection operation with a graphical user interface to change the projection view allows users to rapidly explore the topology of higher-dimensional data and identify regions of interest. Exploring the multidimensional interaction energy landscape of large molecular configurations, for example, is a grand challenge and has spurred this visualization technique.

The scalar and vector topology analysis methods extract unique point, linear, and 2-D features from a 3-D data set that act as a road map for transitions in the data. These global feature sets provide a path for further local interactive exploration as well as acting as a guide for global tours of the data.

"For example, these methods find largest, smallest, and saddle transitions of physical functions such as molecular electrostatic energy, and places where air, gas or fluid flows may eddy or twist very rapidly," Bajaj said. "The road maps we construct are invariant to rotation, translation, and scale, so they are also useful for comparing the results of multiple simulation runs."

An additional effort is working on efficient data representations and human-computer interfaces to allow researchers to ask local and global questions about the data using the visualization. Such interrogative visualization involves building data structures to rapidly visualize a data set at multiple resolutions on large displays, simultaneously capturing both local and global features, and efficiently and accurately quantifying features in the data.

This numerical extraction of information--for example, accurate isosurface areas, distances between feature points, the number of critical transition points--can be invaluable to researchers while they visualize their simulation results. This data enhancement can be volumetric (areas, volumes), combinatorial (counts of features), or topological (showing connectivity) and can also be considered billboards on the visual road map of the simulation.

Shell structure of human headFigure 2: Adaptive, Hierarchical Mesh Generation

Chandrajit Bajaj and the CCV are collaborating with TICAM's Leszek Demkowicz to produce adaptive meshes from MRI data. This smooth reconstruction of a human head shell was built by smoothing a thickening an input triangulation model. Shell structure arises in the design of geometric objects such as aircraft, containers, or the human skull.



Besides the Visual Eyes effort, CCV projects are funded by other agencies. The Data-Intensive, Display-Intensive (DIDI) project, for example, is funded through the Department of Energy's ASCI effort. Project DIDI focuses on compression technologies that extract and display information from simulations on teraflops-scale computers. This effort studies the infrastructure needed to support interactive manipulation on large stereo displays of extremely large data streams. The challenges include progressive- and dynamic-resolution visualizations and scalability despite network and communications latency.

Another CCV project is also funded by the NSF to explore geometric modeling and the visualization of hierarchical A-splines--a new form of spline that was developed within CCV. NASA is sponsoring another project on large data set analysis.

Most recently, Bajaj is a co-principal investigator in an NSF Knowledge and Distributed Intelligence (KDI) award, led by NPACI researcher Mary Wheeler of TICAM, on multiscale simulations of multiphase fluid flows. Multiphase fluid flows of oil, gas, and water at and below the Earth's surface are the cause and cure for water and soil pollution, while petroleum and natural gas production depends on subsurface flows. The KDI project focuses on better understanding small-scale phenomena and linking simulations of these small-scale phenomena to larger-scale studies of such things as surface and groundwater environments. Bajaj and the CCV are involved in viewing the results interrogatively and collaboratively.


As part of the Interaction Environments thrust area, Bajaj is leading the "Visualization Tools" project, through which the tools from the Visual Eyes effort will be deployed in the NPACI infrastructure. Bajaj attended the September meeting at SDSC of the Molecular Science thrust area, for example, to identify possible matches between the CCV's work and the needs of molecular scientists. Past work by the CCV has looked at molecular surfaces, electrostatic and interaction potentials, and collaboration technologies.

The Visualization Tools project also incorporates Bajaj's work with collaborators at TICAM who are participating in the Engineering and Earth Systems Science thrust areas. In Engineering, Wheeler and colleagues at TICAM's Center for Subsurface Modeling are developing tools for simulating oil reservoirs and subsurface pollution. (See the April-June 1998 ENVISION.) This project is conducting realistic, high-resolution studies with a million or more mesh elements, and Visual Eyes tools help the researchers examine the results of these simulations. The accelerated isocontouring tool rapidly color-codes various functions such as permeability and hydraulic movement in the subsurface environments (Figure 1). The hypervolume rendering tools reveal such features as the flow of contaminants downstream over time.

Wheeler also leads a project on bay and estuary flow and transport within the Earth Systems Science thrust area. This project is linking two physical models--a water flow simulator and a chemical reaction simulator--to create a more realistic picture of the movement of pollutants in coastal waters. CCV tools for isocontouring and hypervolume rendering are being used to visualize the results.

Also in the Engineering thrust area, Bajaj is collaborating on a project led by Leszek Demkowicz on simulations of electromagnetic and acoustic fields. Electromagnetic waves, from cellular phones for example, travel through the head, bouncing off the skull and brain. To study how electronic signals interact with the irregular shape of the human head, Demkowicz and his team are creating simulators that use adaptively generated meshes to perform realistic simulations.

Since every head and brain are different, Demkowicz's team uses the CCV tools for adaptive, hierarchical mesh generation to create realistic brain models from MRI and CT scans (Figure 2). Demkowicz's simulations begin with a coarse mesh, solve the problem, estimate the error, then generate an adaptively refined mesh. The process is repeated until the error reaches an acceptable level.

"What we're providing are toolboxes of geometric modeling and visualization methods," Bajaj said. "NPACI gives us the opportunity to make these tools even more robust and put them into use in large metacomputing environments. The partnership brings us into contact not only with others in the Interaction Environments thrust area, but also with several scientific disciplines and potential NPACI users across the country." --DH END