In Progress, 2010
During 2010, about 1.4 million Americans will be diagnosed with cancer and about 65-percent of them will receive radiation therapy. Applied Computational Technologies (ACT), a software development company in Windber, Pennsylvania, works to make radiation therapy better for patients and easier for professionals to make good decisions. They have developed a sophisticated software tool, called ProACTive, to dramatically improve the radiation oncologist’s ability to implement the best possible course of radiation treatment.
Current radiation-treatment software is a compromise between speed and accuracy. ProACTive eliminates this trade-off, and in 2010 testing it has calculated accurate treatment doses more than 170 times faster than other software in this field. In July at the American Association of Physicists in Medicine conference in Philadelphia, ACT presented results from rigorous benchmark tests that, says McClatchey, were superior to the top commercial systems and scientific codes.
“The key benefits of improved planning accuracy,” says McClatchey, “are local tumor control and reduced complications in normal tissue. This is especially true when the cancer is close to organs. Improved accuracy can enable an increase in the dose prescription and physicians may be more comfortable that normal tissue complications can be kept within acceptable limits.”
ACT’s goal is to position itself to transfer ProACTive to a radiation treatment system provider. Currently in final stages of prototype testing, ProACTive is expected to dramatically increase the precision of radiation dose calculation, which means more effective cancer therapy and additional lives saved.
While it’s often true that giving makes you feel good, the “warm glow” of contributing to a worthy cause goes only so far in attracting private donations to charitable and cultural organizations. Far more effective for organizations that depend on wealthy private donors are exclusive, high-profile events, such as dinner parties, that convey social status and offer visibility with other affluent people. That’s the main finding from computational modeling using PSC resources by Holger Sieg, J. M. Cohen Term Chair in Economics at the University of Pennsylvania’s Department of Economics.
Sieg and University of Pittsburgh Ph.D. student Jipeng Zhang compiled an extensive dataset from publicly available donor lists for 10 large Pittsburgh nonprofits: Pittsburgh Ballet Theater, Children’s Museum, City Theater, Pittsburgh Opera, Phipps Conservatory, Pittsburgh Public Theater, Pittsburgh Symphony, Western Pennsylvania Conservancy, Pittsburgh Zoo and the PPG Aquarium. To analyze the data, they used a method new to philanthropy research, classifying individual tiers of giving as “prices” associated with different bundles of benefits.
Their modeling approach — “repeated discrete choice with multiple choice occasions” — is flexible, says Sieg, and has many other potential applications where consumers demand multiple units of different products. Estimating these models is computationally intensive. Initially, the researchers ran their model many times using 300 BigBen processors. For subsequent revisions, they relied on Pople, PSC’s SGI Altix shared-memory system. “This would have taken forever with a high-end desktop system,” says Sieg, “but the computational burden is easily feasible with the current generation of supercomputers available at the PSC.”
They found that charities relying heavily on private benefits to attract donors would see lower giving if these benefits were eliminated. In addition, the nearly eight-percent of individuals who supported multiple charities would decline significantly if there were no high-value private benefits.
To create a visualization of a hurricane brewing its swirl of energy so that it can be projected in 3D stereo in the spectacular Imiloa Astronomy Center at the University of Hawaii in Hilo, Hawaii — that’s the task that PSC visualization specialist Greg Foss undertook in 2009. The image shown here, from that visualization, is from a simulation of Hurricane Ike, the third major hurricane of the 2008 Atlantic hurricane season, and the costliest hurricane ever to make landfall in the United States.
Foss created this and other animations for a November 2009 National Science Foundation workshop at the Imiloa Planetarium titled “Data Visualization: Taking the Presentation of Methods and Results to the Next Level.” The workshop highlighted an NSF EPSCoR (Experimental Program to Stimulate Competitive Research) award to the University of Hawaii system to develop advanced data manipulation and visualization in Hawaii. The world-class 120-seat Imiloa Planetarium, with a 52-foot diameter dome, was the first in the world to have 3D stereo capability.
Simulations on PSC’s BigBen and Pople systems, using the Advanced Regional Prediction System software from the Center for Analysis and Prediction of Storms at the University of Oklahoma, Norman, produced the data. The Hurricane Ike simulation ran on 64 processors of PSC’s shared-memory SGI Altix system Pople. Foss used the visualization software VisIt to create the animation, which represents a domain 1600 kilometers on each side and 16.8 kilometers vertically. The arrows represent wind direction and velocity (increasing in magnitude with color from yellow to red). The bright coloring of the clouds corresponds to radar reflectivity, which shows the amount of water vapor in the atmosphere (increasing from blue to green, yellow and red).
PSC’s director of Scientific Applications and User Support, Sergiu Sanielevici, notes that Hutchinson is an “early explorer” of PSC’s new SGI UV system, Blacklight (p. 4). “Max has excellent knowledge of computing systems,” says Sanielevici, “and how to program them to productively generate new scientific knowledge.”
To work on quantum physics software — even if it involves an expenses paid sojourn in Vienna — isn’t the usual undergrad spring break. For Carnegie Mellon physics student Max Hutchinson, however, it’s a natural extension of his involvement, since high school, with PSC and TeraGrid educational outreach. From a supercomputing award at the Pittsburgh Science Fair to awards at the 2008 and 2009 TeraGrid conference student competitions, to working as an intern in PSC’s Strategic Applications Group, Hutchinson has made the leap from student competitions to real world scientist. This year he began working in the research group of Carnegie Mellon physicist Michael Widom, which led him to Vienna on spring break to work on widely used quantum physics software called VASP (Vienna Ab-initio Simulation Package).
“His PSC experience taught him a great deal about machine architecture and the link with efficient algorithms,” says Widom of Hutchinson, “and through him this experience is improving my own ability to solve interesting problems.” Some of Hutchinson’s work involved GPU (graphics processing unit) systems. The TeraGrid includes several GPU systems, which have processors specialized for graphics rendering. Because of this expertise, Hutchinson was invited to the University of Vienna over spring break, where he worked on porting the VASP software to test on a GPU system. He will apply this method to study the complex crystal structure of the element boron, the stable form of which (called “beta” boron, represented in the graphic shown here) Widom and collaborators confirmed using VASP.
In addition to the study of boron, Widom’s work with VASP has included new understanding of the liquid-liquid transition in supercooled silicon, reported in Physical Review Letters (February 2009). He has relied on PSC’s Pople system for many of his VASP projects, since its shared-memory architecture allows these quantum calculations to run more efficiently than they would otherwise. In this work, Widom has often collaborated closely with PSC scientist Yang Wang.