by Gordy Slack
the Oak Ridge National laboratory by Dr. Chen and her colleagues at the
Sandia National Laboratories.
Scientists across the board, from chemists to astrophysicists, today have the tools to explore and model phenomenon at a mind-blowing level of fidelity and resolution. But the data sets they generate not only blow minds, they also blow the gaskets of the computers and the traditional schemas that scientists try to examine them on.
In Professor Kwan-Liu Ma’s UC Davis VIDI (Visualization and Interface Design Innovation) lab, the objective is to take data sets that can be on the peta or tera scale and turn them into explorable, workable, and visualizable units.
“In the past, most data was viewed as a two-dimensional cross-section or at best as so-called isosurfaces,” says Ma. “But by employing our visualization techniques we are able to let researchers see the full extent of their data at the highest possible resolution and in both three-dimensional space and the temporal domain. So scientists can begin to visualize things they just couldn’t see in the past.”
Computer scientists have been helping other types of researchers visualize their work since the early days of computer graphics. What is new, and the subject of Ma’s Department of Energy-supported research, is working with vast oceans of data. In some cases, the data is so detailed that visualization techniques like Ma’s are needed before hypotheses can be validated. In other cases, the visualizations allow researchers to connect dots they might not even suspect existed.
As part of a five-year SciDAC grant, Ma’s lab has worked with Dr. Jaqueline Chen’s research team at Sandia National Labs in Livermore, CA, on depicting the properties of turbulent combustion with detailed chemistry. At first, Ma simply helped them to visualize their data in three dimensions over time. But then Dr. Chen said that they wanted to see multiple variables, and how these variables related to each other over time, in the same visualization. So Ma and his colleagues had to find ways to superimpose several elements in one image so that they were distinct, but so that their relationships, as they changed over time, were clear.
“In the past, they did most of the data analysis as a post process. But now they are able to interact with their data, to explore, because we’ve developed techniques that have allowed them to go into these different domains of the data. That has become very powerful.”
“We even developed a user interface so they could move between different spaces. They can shift from the temporal space to the spatial domain, and then look at the interaction between different variables at different levels.”
For example, there were delicate properties of turbulent combustion that were hidden in the data due to the multi-scale nature of turbulence flow, Ma says. His visualizations were able to draw some of those features out, allowing, for instance, researchers to view small turbulence eddies that are very close to a known feature surface of interest. Until now, those features have usually been eclipsed by other, more salient phenomena. Ma’s visualization software allows the Sandia scientists to zoom in on the feature surfaces and move closer and further away from the surfaces, examining these different features at different scales, including very detailed and subtle micro-eddies near the surface.
Powerful as computer visualizations are, there is always some uncertainty introduced when looking at data indirectly, says Ma. Depicting that level of uncertainty, making it a part of the visualization itself, is another subject of his team’s research.
Ma wants users of his tools to be able to “visualize the process of visualization itself,” he says. “If we can convey what has been done to the data to generate the image they’re working with, — not just the information loss but also the important mapping done to the data to get the image– then the user can have much greater control.”
Ma and his colleagues at VIDI work with scientists from a wide range of disciplines: climate science, chemistry, astrophysics and practical sciences like groundwater engineering and combustion studies. For Ma and his colleagues and students, the first phase of any project is a total-immersion crash course in the language and basic principles of the science itself. Only then can the task of translating data generated by the science into manipulatable, computer-generated images begin.
On the one hand, Ma helps some groups to exploit the largest available computers to process and visualize their data. He has created algorithms that work on massively parallel computers and cluster computers, for groups such as the Stanford Linear Accelerator Center, Argonne National Laboratory, and Sandia National Laboratories that have large computing power needs. On the other hand, he also goes the other direction, taking huge data sets and making them visualizable on smaller computers. He recently worked with the Harvard-Smithsonian Center for Astrophysics, taking vast amounts of observational data viewable only on computers with hundreds of gigabytes of processing power, and scaling it down–without losing important information, of course—so that it can be examined and explored on a single-gigabyte desktop.
Ma is clearly driven by a desire to empower scientists to better do their most important work, like finding ways to model climate change and to develop technologies that will mitigate global warming, but he also seems driven by the pure aesthetics of his field.
As John Keats wrote, “Truth is beauty,” and, when they reveal the subtle secrets of their subjects, Ma’s true visualizations can be stunningly beautiful as well.