The ANNUAL 2011 DEIXIS uate F ellow s hip S c ien e G ra d uate of E nergy Computational Department

FROM THE GALACTIC TO THE ATOMIC Practicum Takes Paul Sutter Deep into Inner Space

page 5

Ying Hu’s Road Trips Lead to Scattering Discoveries

page 9

The ANNUAL Table of Contents DEIXIS 9 22 DEIXIS, The DOE CSGF Annual is published by the Krell Institute. Krell administers the Department of Energy Computational Science Graduate Fellowship (DOE CSGF) program for the DOE under contract DE-FG02-97ER25308.

For additional information about the DOE CSGF program, the Krell Institute or topics covered in this publication, please contact: 5 24 26

Editor, DEIXIS Krell Institute 1609 Golden Aspen Drive, Suite 101 Ames, IA 50010 (515) 956-3696 www.krellinst.org/csgf A TOAST TO 20 YEARS 4 22 28 Copyright 2011 Krell Institute. Practicum Profiles Alumni Profiles Winning Essays All rights reserved. In 1991, the fates of two A Season of Discovery From Alumni to Leaders Encouraging Communication technological tools — the Internet and computational science — became 5 paul Sutter 22 Brian Moore 28 Kenley Pelzer, Winner DEIXIS (ΔΕΙΞΙΣ — pronounced da¯ksis) transliterated linked in the High-Performance Fellow Spans Scales from A Career at the Can Peeling an Onion from classical Greek into the Roman alphabet, means a display, mode or process of proof; the process of Computing and Communications the Universal to the Atomic Nuclear Intersection Cure Cancer? showing, proving or demonstrating. DEIXIS can Act. The law backed creation of the “information superhighway,” but also refer to the workings of an individual’s keen intellect, or to the means by which such individuals, also directed the Department of Energy (DOE) to support computational 9 ying Hu 24 Dan Martin 30 hayes Stripling, e.g. DOE CSGF fellows, are identified. science education. Illuminating a Path “Cool Problems” Draw Honorable Mention The Internet’s phenomenal expansion is renowned, but the parallel to Cancer Cells Alumnus to Laboratory On the Quantification DEIXIS is an annual publication of the Department of Energy Computational Science Graduate Fellowship escalation in high-performance computing for science has been equally of “Maybe”: A Niche program that highlights the work of fellows and alumni. exceptional. Since 1991, the DOE Computational Science Graduate 13 anne Warlaumont 26 mala Radhakrishnan for Computation Fellowship has led the field’s maturation into the “third leg” of scientific Modeling a Neural Network Alumna Busts Resistance — The DOE CSGF is funded by the Office of Science and the National Nuclear Security Administration’s Office discovery. With more than 250 alumni, it’s helped create a community in the Classroom and Lab of Defense Programs. of computational science leaders. 17 Scott Clark This issue bears witness to the fellowship’s success. Our cover story Practicum Charts Career 32 features Paul Sutter, who used his practicum to step outside his scientific Course Beyond the Howes Scholars comfort zone. We also announce this year’s Frederick A. Howes Scholar Velvet Rope Living Up to a Legacy in Computational Science: Alejandro Rodriguez, who has matched his work in computational physics with his devotion to outreach. And we feature three alumni, including Brian Moore, who found a calling in Editor Publication nuclear energy. 35 Shelly Olsan Coordinator Alumni and fellows are welcome to participate in the program’s Directory Lisa Frerichs essay contest, which provides them the chance to convey their research Senior Science Writer Design to a broader, non-technical audience. The winner, fellow Kenley 35 fellows Thomas R. O’Donnell julsdesign, inc. Pelzer, connects onions to quantum mechanics. Here’s to 20 years — and many more to come. 38 alumni Copy Editor Ron Winther 28

P2 DEIXIS 11 DOE CSGF ANNUAL DEIXIS 11 DOE CSGF ANNUAL P3 practicum profiles A SEASON OF DISCOVERY For fellows, the summer practicum nurtures professional and personal growth

The Department of Fellow spans scales from Energy Computational the universAl to the atomic Science Graduate

Fellowship supports the PAUL SUTTER nation’s brightest science University of Illinois at Urbana-Champaign and engineering students, Oak Ridge National Laboratory allowing them to concentrate on learning When it came time to choose a practicum for summer 2009, Paul Sutter and research. The work went from the enormous to the minuscule. Sutter, a Department of Energy Computational Science Graduate Fellowship recipient, of more than 250 DOE usually simulated galaxy clusters, the largest structures in the universe. At Oak Ridge National CSGF alumni has helped Laboratory (ORNL), he calculated electronic structure in atoms. “The energy scales, the spatial scales, the time scales are the complete opposite end of the the United States remain spectrum from what I usually do,” says Sutter, who earned his doctoral degree in astrophysics competitive in a from the University of Illinois at Urbana-Champaign (UIUC) this spring. “And the architecture of the computer code was completely different.” global economy. WIt’s just what he wanted. Practicums let fellows try research in areas largely outside their studies, but few go as far afield as Sutter did. “I realized that there aren’t a lot of opportunities in ~~~~~ your research career to go off and do something completely different,” he says, and he took advantage of it. After meeting several ORNL researchers Sutter opted to work with Robert Harrison, leader of the Computational Chemical Sciences Group. Sutter focused on MADNESS (Multiresolution Adaptive Numerical Environment for Scientific Simulation), a code that solves the electronic structures of atoms, molecules and Left to right: Summer plans often take more thought for recipients of the Department of nanoscale systems. Harrison led creation of the program, which is used in computational Scott Clark, Paul Sutter, Anne Energy Computational Science Graduate Fellowship than for other graduate students. chemistry, superconductor modeling and other areas. Warlaumont and Ying Hu in Summer usually is when fellows head to DOE national laboratories to complete their required Calculating electronic structure is so difficult scientists must limit their models’ accuracy Washington, D.C., at the DOE practicums. Often, they’ve met the scientist they’ll work with at a conference or through a mutual and the size of the systems they compute. Using MADNESS, researchers can compute the CSGF Annual Conference. acquaintance. But some fellows — like Ying Hu, who is profiled here — also visit several labs properties of larger systems with higher accuracy than traditional quantum chemistry before choosing one. codes allow. It divides the physical domain being modeled into parts for calculation on Practicums push fellows to explore subjects or test skills with little or no connection to parallel processing computers, while repeatedly subdividing the most interesting pieces for their doctoral research. It frequently leads them to refine a computer science skill — as fellow more precision. Paul Sutter did — or, like fellow Anne Warlaumont, to develop a greater appreciation for As the program subdivides the domain it generates an “octree” structure of branching hands-on lab work. nodes. The number of nodes depends on the level of detail researchers want, the problem size Some fellows even end the summer determined to take on entirely new research subjects. and the computer they use. Typical MADNESS problems involve octrees with 10,000 to 100 Fellow Scott Clark, for instance, headed to his practicum planning to focus his dissertation on million nodes. computational fluid dynamics. By summer’s end, he’d found a new love: metagenomics. That’s the power of the practicum: reinvigorating, redirecting and remarkable. P4 DEIXIS 11 DOE CSGF ANNUAL DEIXIS 11 DOE CSGF ANNUAL P5 practicum profiles

Sutter and Hartman-Baker implemented a “melding” algorithm that splits up nodes so each processor has roughly the same amount of communication and work. ~~~~~

Sutter and Hartman-Baker implemented The second test used a real-world balancing algorithm, calculation time held Sutter’s research focused on magnetic a “melding” algorithm that splits up nodes problem: solving the Poisson equation to steady at around 60 seconds — a 60 fields within clusters of hundreds or so each processor has roughly the same calculate the electrostatic potential of a percent cut. “Overall, we found a pretty thousands of galaxies. Astrophysicists amount of communication and work. But large molecule-like structure. “This problem solid performance boost when doing this think supermassive black holes found in a “we didn’t really have a way of figuring can get very, very big very, very quickly,” improved load balancing,” Sutter says. galaxy at the clusters’ center may generate out how much time was required on each Sutter says. “There is a nasty list of things Working with MADNESS gave the fields. node” to weight them for redistribution, that have to be done.” Sutter insights into extending FLASH, The fields are weak, Ricker says — Hartman-Baker says. Again, Sutter tested the algorithm a multiphysics, multiscale code he used perhaps a fraction of Earth’s, but spread Sutter addressed that problem with a on large and small problems and averaged in his dissertation research. In fact, over a large enough area that they force “profiling” algorithm. In effect, MADNESS results over five runs. Profiling and he returned to Oak Ridge last fall accelerated particles like electrons to runs with the random load-balancing melding produced a modest gain on the to work with Hartman-Baker on move in a spiral-like pattern, generating algorithm and gathers data on how long large problem, cutting calculation time by improving FLASH. radiation in the radio spectrum. Two nodes take to compute. Then it uses that almost 10 percent when compared to the Sutter also got the chance to work clusters may look similar in the optical data to rebalance the load on the fly. standard approach. Calculation time with Jaguar, even getting the machine or X-ray spectra, Sutter says, but not in Overloaded processors send work to changed little for both even as the number almost all to himself as MADNESS ran the radio spectrum. “Since radio is others while underutilized processors of processors increased. during testing after an upgrade. He and nonthermal emission we actually get gather data to work on. Results were more significant for the his UIUC doctoral advisor, Paul Ricker, signatures of the cluster’s history, such as “Melding with profiling tries to smaller problem. Calculation time for the also got an allocation on Jaguar. “It’s just past collisions” with another cluster. balance the two — to only move work to standard approach held steady at a bit a dream to use,” Sutter adds. another processor if it makes sense for more than 150 seconds. With the load- both computation and communication,” Sutter says. This visualization shows a molecular orbital of the benzene dimer computed using Sutter ran two tests comparing MADNESS. It shows the adaptive “grids” used to focus computation on the MADNESS with and without the load- most important areas. balancing algorithm. The first compared Image courtesy of Robert Harrison, Oak Ridge National Laboratory, with support from the Scientific basic mathematical operations — multiply Discovery through Advanced Computing program. two functions, copy one function into another and “compress,” in which MADNESS devolves a function into The structure makes scaling MADNESS Scientist Rebecca Hartman-Baker says. its constituents. a challenge; running it on larger computers The sometimes-uneven distribution of Using ORNL’s Jaguar, a Cray XT5 doesn’t lead to proportionately faster and work and the constant communication rated one of the world’s fastest computers, more efficient performance. Some nodes slow things down. Sutter averaged the cost of compress, require lots of work, while others require What was needed, Sutter says, was multiply and copy operations across five Above: MADNESS has applications in solid-state physics, little, but there’s no way of knowing which “some way of balancing them out. Instead runs on two problems — one with about including calculating the Coloumb potential for electrostatic is which, so MADNESS randomly assigns of, say, 10 nodes on all processors, maybe 10,000 nodes and one with about a million. interaction, as shown here for lithium fluoride. them to processors. some processors have just one node, but The smaller problem produced little “It results in a lot of communication, computationally it’s a very expensive node. difference on compress and multiply Left: MADNESS can apply the Hartree-Fock and Density because if you want to find out what’s Other processors might then have 100 operations but improved scaling for Functional Theory methods to calculate the ground state going on with your neighbor on the tree, nodes, but they would be very, very cheap. copy. The results were essentially the energies of atoms, molecules and nanoscale systems. This you know your neighbor’s not going to So in the end the amount of work per same for the larger problem — except visualization shows a MADNESS calculation of the spin density be on your processor. It’s going to be processor stays the same.” that computing times for copy operations of a solvated electron. somewhere else,” ORNL Computational improved dramatically as the number of Images courtesy of Robert Harrison, Oak Ridge National Laboratory, with support from the Scientific Discovery through Advanced Computing program. processors increased. “That was good, because copies happen a lot,” Sutter says.

P6 DEIXIS 11 DOE CSGF ANNUAL DEIXIS 11 DOE CSGF ANNUAL P7 practicum profiles

Illuminating a path to cancer cells

This simulation shows the distribution of gas in a section of the universe 5 billion light-years wide. The enlarged portion zooms in on a pair of merging galaxy clusters. The simulation uses YING HU adaptive mesh refinement to focus calculations on areas of Rice University interest: the central red turbulent region of the clusters, where Lawrence Berkeley National Laboratory peak resolution is a computational cell about the width of the Milky Way galaxy.

Ying Hu swears his decision to pursue a practicum in California had nothing to do with the snowstorm that trapped him in his hotel during a Astronomers and cosmologists are few high-mass, high-power halos. That Sutter finished his doctorate while “What we’re proposing January 2009 visit to Argonne National Laboratory near Chicago. building radio telescope arrays to capture may result from assumptions made in the living the last two years in Newark, Ohio, Hu, a Department of Energy Computational Science Graduate Fellowship (DOE CSGF) is that by just using signals the clusters emit. The simulations simulation, Ricker says. five hours from Urbana-Champaign. recipient, also visited Lawrence Livermore (LLNL) and Lawrence Berkeley (LBNL) give them ideas of the number and strength Ricker and Sutter later ran a version Ricker approved the arrangement so Sutter’s one added step you national laboratories in the San Francisco area as he researched practicum locations. of objects they’ll see. And when researchers simulating a volume of space 1.5 gigaparsecs new wife, Mandi, could stay at her job. “All the groups I visited are fabulous,” says Hu, a doctoral candidate in biomedical get radio emissions data, simulations (1.5 billion parsecs) on a side. The simulation “With just about any other student, I can improve the engineering at Rice University in sunny Houston, Texas. He learned how different labs will help interpret them into a history identified the most massive clusters and the would have said this is really a bad idea,” sensitivity and probably operate and discussed shared scientific interests with researchers, including LBNL’s of the cluster. researchers used adaptive mesh refinement he says, but Sutter is “disciplined enough Jim Schuck and Jeff Neaton. Schuck studies nanoscale optical imaging spectroscopy, Those simulations must cover enormous to focus calculations, “zooming in” on those that he’s been able to get an awful lot done.” detect something you Yinvestigating how structures hundreds of times smaller than human hairs interact with swaths of space, but with enough detail to and a large number of less massive halos. That long-distance relationship is light. Neaton works on the theory side, understanding nanoscale phenomena and guiding capture the injection of magnetic fields At its finest resolution, one cell in the nothing compared with Sutter’s postdoctoral couldn’t detect before,” experimental researchers like Schuck. emanating from black holes. “It just becomes computational data mesh was about the research post. He’s studying the cosmic Hu says. Their work meshed well with Hu’s doctoral project: Developing computer codes to a very, very large problem,” Sutter adds. size of the Milky Way galaxy. microwave background and helping design investigate optical properties of nanostructures and optimize their designs — specifically In one project he, Ricker and graduate Sutter and Ricker plan to compare the next generation of cosmic probes with ~~~~~ to target cancer cells. And LBNL had the Molecular Foundry, a nearly new facility where student Hsiang-Yi Yang used FLASH to simulation results with actual radio telescope Ben Wandelt, a former UIUC faculty nanoscale theory, fabrication, testing and simulation research share space in a dramatic create simulated radio astronomy maps. observations. Sutter adds, “Hopefully we can member now at the Paris Institute of building overlooking the bay area. They computationally evolved galaxy match what they see and that will give us Astrophysics. Sutter will spend a few weeks “It played a major role in my decision,” Hu says. “It’s a very integrated facility. You can clusters and their radio halos over a span some confidence that we’re moving in the each year in Paris, but most of the time fabricate a sample in the clean room and then take it to the first floor and do measurements. of 12 billion years in a “box” of space 366 right direction.” he’ll be in Ohio, working with Ohio At the same time people on the third floor are doing calculations” to understand the results. megaparsecs on a side. (One megaparsec, He’s fascinated by the idea that every State University Astronomy Professor Hu did all three while researching surface-enhanced Raman (RAH-mon) scattering or million parsecs, equals about 3 million plot, simulation or visualization he does is David Weinberg. (SERS), in which scientists decipher the molecular and crystal structures of even miniscule light years.) The simulation had to an attempt to portray an unimaginably What’s a few thousand miles when amounts of materials by analyzing the way they react to light. include the influence of invisible dark enormous reality. “It’s a pretty crazy idea. you’re discussing objects billions of light “You excite the molecule at one wavelength and you collect the whole spectrum and matter that makes up the bulk of the When I actually sit down and think about, years across? look for the Raman peaks,” Hu says. Like fingerprints, “The locations of the peaks are universe and then calculate the movement outside of the simulations I’m doing, what unique to each molecule.” The effect is weak, but in the 1970s researchers found that and behavior of hot gas. Yet, it’s still only this means in the real universe — it boggles placing samples on roughened metal surfaces significantly boosts scattering — thus the precise enough to estimate just the my mind sometimes.” SERS name. magnetic fields, rather than radio Scientists have postulated two explanations for the SERS effect, one arising from emissions directly. chemical bonds between the material and the metal surface, and one from electromagnetic Observations show only a small interactions. Hu explored each. percentage of lower-mass clusters has Sutter is fascinated by the idea that every plot, simulation Both projects focused on a substrate of nanocones made of silicon-germanium detectable halos, Ricker says, while about or visualization he does is an attempt to portray an and coated with a thin layer of gold. It’s like a field packed with a jumble of irregularly sized a third of higher-mass clusters do. Yet, gold-plated traffic cones, each a thousand times thinner than a hair. the simulated radio maps predicted more unimaginably enormous reality. low-mass, low-power radio halos and too ~~~~~

P8 DEIXIS 11 DOE CSGF ANNUAL DEIXIS 11 DOE CSGF ANNUAL P9 practicum profiles

This scanning electron microscope image shows an overhead Rewarding requirement view of the nanocone substrate for SERS, seeded with gold nanoparticles (red arrows) to test how seeding enhances Raman scattering. Scale bar (lower left) is 100 nanometers.

All DOE CSGF recipients must complete a practicum at a During his practicum, Hu studied experimentally, says Hu, who prepared Electromagnetic enhancement of SERS While at LBNL Hu also adapted national laboratory by the end of how the way that a benzene thiol (BT) and scanned the samples. “It’s a very is believed to arise from surface-plasmon MEEP, an electromagnetic physics code, their second year — a requirement molecule binds to the substrate — like potent chemical and it has a very strong, resonance — the nanostructures’ response to run on parallel processing machines at Ying Hu found a bit intimidating. how a Tinkertoy stick-and-ball model skunk-like odor.” to electromagnetic radiation. In essence, the Molecular Foundry. The task required “In the beginning you feel like, might fit in a socket on the gold traffic the gold nanocones act like tiny antennae. installing and coordinating multiple ‘OK, it just adds another three months cones — affects SERS. The researchers Enhanced understanding Their electrons oscillate in response to software packages. “That was really a that I need for graduate school. It computed relative Raman scattering of enhancement visible light because the structures are learning experience for me because my delays my graduation date,’” says enhancement for each of three binding They found the way light perturbs the similar in size to the light’s wavelength — background is not computer science,” Hu was familiar with calculating Hu, who earned his doctorate at configurations and compared the results charge transfer complex between BT and 400 to 800 nanometers. Hu says. nanoplasmonic effects. His doctoral Rice University earlier this year. He changed his mind after with actual SERS measurements. the gold atoms affects SERS. To continue In a typical SERS setup, a laser hits Hu arrived at LBNL, Schuck says, just research focused on nanoshells: minuscule doing research at Lawrence Berkeley The researchers chose BT because it’s the analogy, how much scattering is the molecules to be tested — located on as he and Neaton “realized there was a hollow or layered gold balls. In Drezek’s National Laboratory in summer 2009. relatively simple to compute. “But it by no enhanced relies on how light affects the the metal substrate — from directly whole new range of things we could do lab, researchers conjugate the particles The experience, Hu says, expanded means is an easy molecule to work with” connection between the Tinkertoy and gold above, Hu says. “Most of the time, if you that needed to go parallel.” Since Hu with antibodies that target cancer cells. his horizons. He learned how another on the traffic cones. Some configurations think of light as electromagnetic waves, installed MEEP, “we use it a ton.” Once injected into the body, the lab in an area similar to his doctoral yielded a stronger response than others. In the polarization can only be perpendicular nanoshells attach to tumors. research operates, what computing some cases, computer calculations agreed to the direction it’s propagating,” he adds. Tiny ties By adjusting their size and geometry, and experimental resources are with the experimental results. A laser coming directly from above will Hu used MEEP to simulate the researchers can “tune” nanoshells to available in the field, and what other Hu presented the research in a paper be polarized in the transverse plane — nanoplasmonic response of bow-tie absorb or scatter specific light wavelengths. researchers are thinking. coauthored with Schuck, Neaton, and horizontal to the surface. It can’t be antennae — tiny gold triangles pointing to The Rice team engineers nanoshells to Hu’s summer in Berkeley — and LBNL and University of California, axially polarized — perpendicular to each other. In Hu’s calculations, there were resonate with near-infrared light, a spectrum a second practicum at Lawrence Berkeley researchers Alexey Zayak, Hyuck the substrate. four triangles, arranged like a Maltese minimally absorbed by surrounding tissue. Livermore National Laboratory in fall 2010 — also allowed him to begin Choo and Stefano Cabrini. “It’s a relatively But the researchers’ simulation cross, on a dielectric substrate. He studied When a near-infrared laser shines on the building a network of colleagues and new theory to explain chemical showed axial polarization would generate the consequences of moving a triangle particles, they illuminate, defining the possible collaborators. enhancement,” Hu says. “I think it’s stronger electromagnetic enhancement in off center. tumor’s edges. “It’s really a start for me to think going to have a relatively big impact.” the nanocones. “So that creates problems,” “By shifting one element we can actually The surface temperature of the about my career path, to develop Schuck says that while different Hu says. The researchers tackled them concentrate light at different locations in the nanoshells also rises quickly as they absorb my own perspective on what I think mechanisms for enhancement have been by spreading gold nanoparticles on the structure at different wavelengths,” Hu says, light. By increasing the laser’s power it may should be done or can be done (in proposed over the past 30 years, no one substrate — like dropping gold balls on and different configurations can sort light of be possible to “cook” and kill cancer cells his field) in a more realistic way,” knew which were important until Hu and the field of traffic cones. Tests found different wavelengths. with minimal harm to healthy tissues. Hu adds. “It’s something that’s hard his fellow researchers came along. With nestling the nanoparticles between the The color-sorting antennae could be Just a few years ago most nanoshell to get from staying in the lab in the interpretation in the paper, “For the cones boosted SERS activity by more than used in optical computers, which transmit designs were simple enough that standard graduate school.” first time, we have a general quantitative a factor of 10, even with transversely information with photons instead of tiny computers usually were enough to basis for understanding chemical polarized light, Hu says. Experiments bore wires printed on silicon chips, Schuck says. calculate their properties, Drezek says. enhancement for all SERS measurements, out the results, which are reported in a But a more near-term application may be Now the nanoparticles labs produce are so which I think is pretty amazing.” paper accepted for publication in the in extremely small and fast multicolor light complicated it takes parallel processing on Hu tackled SERS’s electromagnetic journal ACS Nano. detection for cameras and sensors. a cluster to understand them. foundation after returning to Rice and “What we’re proposing is that by just LBNL researchers were still analyzing “We weren’t doing any of that in our The graph shows Raman cross sections of benzene thiol with the isolated molecule discussing ideas with Seunghyun Lee, a using one added step you can improve the Hu’s data and producing papers based on lab before Ying started. We just did our tracked in black and the molecule chemisorbed on a Au(111) surface in orange. chemistry graduate student. Other sensitivity and probably detect something it more than a year after he finished his simulations on desktop computers,” Chemical enhancement leads to the much stronger intensity from the chemisorbed collaborators include Berkeley Lab’s you couldn’t detect before,” Hu says. practicum, Schuck says. “We can’t help but she says. “He was able to take all the molecule. The image below the graphic simulates the binding geometry of benzene Choo; Jaeseok Jeon and Tae Seok of UC Fine-tuning nanoparticle size and send Ying e-mails and ask him questions, computational coursework, move what he thiol on a Au(111) surface at 300 K. This shows that at room temperature the binding Berkeley; Hu’s advisor, Rebekah Drezek; concentration could enhance scattering because we know he knows the answers to was doing into the clusters and open up all geometry of benzene thiol on Au(111) is not well defined, both with respect to and Lee’s advisor, Jason Hafner. even more, he adds. a lot of these things. We’d love nothing sorts of different opportunities for us.” molecular orientation and sulfur-gold bonding coordination. Sulfur typically tends to more than to work with Ying” again. pull one gold atom from the surface.

P10 DEIXIS 11 DOE CSGF ANNUAL DEIXIS 11 DOE CSGF ANNUAL P11 practicum profiles

Modeling a This illustration shows axial and transverse (in-plane) neural network polarizations (in relation to the SERS substrate) of laser light for Raman excitation. Research by Ying Hu and colleagues suggests that seeding the surface with gold nanoparticles, as shown, boosts SERS activity by more ANNE WARLAUMONT than a factor of 10 with transversely polarized light. University of Memphis Argonne National Laboratory

Warlaumont says the Usually, Anne Warlaumont deals only with the rough “We weren’t doing any of that In a paper published in the Nov. 24, on a piece of tissue paper. The wet paper computer counterpart to a brain — building “neural” networks that classify sounds from project gave her a in our lab before Ying started. 2008 issue of the journal Optics Express, sank, leaving the needle pointing north on babies and emulate the way they learn to speak. Hu, Drezek and Rice biochemistry student the surface. “It was the first compass I’d different perspective. But Warlaumont, a Department of Energy Computational Science Graduate Fellowship We just did our simulations on Ryan Fleming simulated a nanoshell ever seen and I have to say it is the coolest recipient, saw the real thing during her summer 2009 practicum at Argonne National comprised of a gold outer layer, a silicon compass I’ve ever seen,” Hu says. “In my main research Laboratory (ANL) near Chicago. desktop computers,” Drezek inner layer and a gold core. The calculations Hu’s dissertation includes a chapter I build neural network Warlaumont shadowed two students in University of Chicago (UC) researcher Wim van says. “He was able to take all found each layer’s plasmonic resonance on his LBNL work and he’s still collaborating Drongelen’s lab and witnessed electrophysiology experiments on mouse and human brain mode interacts with the others. “That with scientists he met there. “It’s just another models. Those models tissue. As her practicum ended she also watched brain surgeons operate on a girl who the computational coursework, interaction is what renders these particles proof of the value of the practicum,” he suffered epileptic seizures since infancy. are even more abstract move what he was doing into the tunable, like how we can tune it to the near says. “It’s really the start of my own U“I definitely never thought I would see something like that. It was a special bonus,” says infrared spectrum,” Hu says. independent research.” than the ones we worked Warlaumont, a doctoral student in the School of Audiology and Speech-Language Pathology clusters and open up all sorts of Hu graduated this summer, just nine In fact, Hu did another practicum at at the University of Memphis (UM). years after moving from China to Houston, LLNL in fall 2010, with electromagnetics with (at ANL). I was Warlaumont’s peek inside the skull was only appropriate. In her practicum, she designed different opportunities for us.” where his parents are researchers at the and photonics researchers Daniel White a program that compared detailed and abstract computational models of an epileptic brain. happy about the ~~~~~ Baylor College of Medicine. His mother’s and Tiziana Bond. “I talked to them While working with Mark Hereld, an ANL experimental systems engineer in the Mathematics father, a physics professor, inspired Hu’s during my practicum-hunting trip,” he opportunity to work with and Computer Science Division; ANL Associate Laboratory Director for Computing, interest in science. He recalls his adds. “I just decided to do a second one Environment and Life Sciences, Rick Stevens; and van Drongelen, Hyong Lee and Marc grandfather floating a magnetized needle because I liked the first so much.” a detailed model and see Benayoun at UC, she also compared the models with activity recorded from slices of mouse what I was missing.” brain tissue. Warlaumont says the project gave her a different perspective. “In my main research I ~~~~~ build neural network models. Those models are even more abstract than the ones we worked with (at ANL). I was happy about the opportunity to work with a detailed model and see These computer simulations of what I was missing.” gold-coated nanoscale cones show The models Warlaumont develops in her UM research “learn” to identify infant how spacing affects electromagnetic utterances — the vowels, squeals, growls, grunts, babbling, crying and laughing babies do enhancement for surface-enhanced as they test their speech abilities and learn to talk. Raman scattering. “There are theories of how infants develop the ability to vocalize and theories of how infants learn sound and why they produce some sounds before others,” she adds. “There is A single cone (a) or a doublet (b) a small group of us interested in translating some of the theories into a way to more rigorously does not exhibit strong enhancement test those computationally.” The end results could include better speech analysis tools or under transversely polarized light, even a model that perceives sounds — both from others and from itself — and learns to while a narrow gap between two “speak” much as infants do. nanocones (c) does.

P12 DEIXIS 11 DOE CSGF ANNUAL DEIXIS 11 DOE CSGF ANNUAL P13 practicum profiles

Each colored point in this figure represents a single neural network model run’s behavior, plotted in an abstract 2-dimensional “behavior space.” Red and orange points are from the two versions of the detailed model; green and blue are from the two versions of the of the reasons we had to struggle to try are always better, because there are Eugene Buder, Robert Kozma and Rick abstract model. Each run within a model version has a different and eliminate artifacts and find real advantages to detailed models.” Dale, is a neural network that recognizes combination of excitation- and inhibition-related parameters. Black differences,” Warlaumont says. As to what this work means for protophones — early categories of points and the star represent samples from a recording of real mouse The researchers filtered simulation understanding and treating epilepsy, infant vocalizations — with potential brain tissue. results to make them comparable to mouse Warlaumont says computational models ramifications for speech analysis. brain traces, then extracted eight primary of brain disorders are imperfect, but still Currently infant speech research metrics from each time series. “We looked informative and can help guide research. depends on assistants who spend days at things like, within the network, are all “Assuming you accept that listening to recordings and manually the neurons spiking? If there is a lot of computational modeling is valuable coding each sound for its type and other heavy synchrony in the firing of neurons, for those purposes, it’s a logical next step properties. It’s time-consuming, expensive “I have several different research On the other hand, simpler models aren’t In the abstract it would end up leaving traces of big peaks to ask how we are going to objectively work, and the amount of data to be sifted threads, all related to infant vocalization as readily tweaked to match reality. The more abstract model, developed and valleys in the network signal,” a sign of compare these models,” she adds. “It’s not is growing. Coders also must make research and understanding the computational The researchers had their work cut in 2003 by Eugene Izhikevich, then of the a possible seizure, Warlaumont says. “We a big problem now because there are only subjective judgments about how to or technological components involved,” out for them. “The problem is pretty Neurosciences Institute in La Jolla, Calif., also looked at the spectral character of a few, but if you see a future for this there classify a protophone, leading to Warlaumont says. “I see these as part of a difficult because it’s terra incognita,” treats each neuron as a single compartment those network voltage signals, the amount will be more models. I think figuring out inevitable inconsistencies. very long-term research program.” Hereld says. Scientists typically have some and randomly varies parameters to model of power in different frequency bands.” how to compare and evaluate those Warlaumont and Hereld hope her intuition about the relevant processes different types. Neurons are networked They ran a principal components is important.” practicum project will help us better behind a phenomenon, but epilepsy’s more randomly and a simpler mathematical analysis to reduce the dimensionality of understand brain activity by improving complex, variable nature resists prediction. method models ion channels. the behavioral features space. That let the Grasping Speech neural models, which range from highly The researchers ran two models, one The researchers also ran two researchers compare the range of behaviors Development complex and computationally demanding detailed and one abstract, then compared versions of this simpler model: one with a given model produced. The details of speech recognition and to abstract and easy to run. average cell membrane potential — a instantaneous transmissions between Ultimately, the abstract models learning Warlaumont hopes to emulate in “Our project was comparing temporal measure of the voltage difference between neurons and one with a six-millisecond seemed to produce a range of behaviors her UM dissertation research are too signatures of neural network data produced the interior and exterior of a cell. Neurons delay. They compared simulation results as broad and nearly as similar to mouse complex to capture with the models and by a couple of very different types of use electrical membrane potentials to with data recorded from slices of mouse data as the detailed ones. “You are not computing technologies available today. computational models. We wanted to transmit signals between different parts frontal lobe tissue that was excited to necessarily disadvantaged, from that Still, she hopes her work will help researchers compare them with each other and with a of the cell and to initiate communication produce normal and seizure behavior. perspective, if you use the less detailed better understand speech development. real system of brain cells,” Warlaumont says. across cells. The detailed simulations ran on version,” Warlaumont says. Yet, “I wouldn’t One of Warlaumont’s models, built For example, the researchers wanted Van Drongelen designed the detailed Jazz, ANL’s recently replaced 350-node want to make it sound like simple models in collaboration with UM colleagues to know whether the advantages of a model, which simulated 656 neurons of computing cluster. Each persistent sodium highly realistic model outweigh the six different kinds. Each cell is modeled version took about 200 seconds to run. In demand it places on computer resources. as a set of compartments corresponding to contrast, each run of the abstract model’s “This project is helping us to understand its parts and includes chemical channels instantaneous transmission version took how low can we go — how simple and that regulate spontaneous firing and about 10 seconds on a standard laptop. Top right: The points in this plot are the same as in the page 14 plot of therefore computationally fast we can transmission of nerve impulses For each model and for the mouse the detailed model, abstract model and mouse tissue in 2-D abstract make a model that will still deliver between cells. data the researchers averaged neuron “behavior space.” In this figure, convex hulls are drawn around each appropriate results,” Hereld says. The researchers ran two versions: activity across all the cells. That was model version’s behaviors, indicating the range of behavior and the amount Warlaumont adds, “Another factor is One with persistent sodium ion channels tricky: Both models generated data in of overlap across the models. The abstract models produce a wider range of behaviors than the detailed models and the real mouse brain tissue. your ability to understand what’s going on and one without. Persistent channels physical units — microvolt waveforms with a model. The more detailed a model could be important to understanding ­­— but they had different temporal is, the more like a real system it is, but it network behavior, Hereld says. resolutions and some unimportant Right: Each colored line in this plot represents one neural network may be so complex it’s hard to understand.” differences. Those disparities are “one model run or one sample from a mouse brain tissue recording. Different principal components (i.e., dimensions of an abstract, 22-dimensional “behavior space”) are on the x-axis, and models’ coordinates with respect to those principal component dimensions are on the y-axis.

P14 DEIXIS 11 DOE CSGF ANNUAL DEIXIS 11 DOE CSGF ANNUAL P15 practicum profiles

“Our project was comparing temporal signatures of neural network data produced by a couple of very different types of computational models. We wanted to compare them with each other and with a real system Practicum charts career course beyond the velvet rope of brain cells,” Warlaumont says.

~~~~~ SCOTT CLARK Cornell University The neural networks Warlaumont updating that node’s weights and its more than half the time, the infant’s age Los Alamos National Laboratory and colleagues are developing and testing neighbor’s weights. It’s similar to how real 42.8 percent of the time, and its identity automatically classify infant vocalizations, brain cells develop connections based on 32.4 percent of the time. “It is understandably helping researchers better understand how the animal’s particular previous experiences, very hard to classify age and identity on humans perceive them and creating data Warlaumont says. the basis of a single second of vocalization. After a summer working One of Scott Clark’s weaknesses, if it can be called that’s more standardized. Each vocalization triggers states of We might be able to average many that, is that there are just too many captivating challenges. with Hengartner and Their model, reported in an April 2010 SOM node activations, which are sent to vocalizations in a recording and get better “I’ve always liked to explore new and interesting things,” says Clark, a Department of paper in the Journal of the Acoustical Society a second layer, a neural network called a performance,” Warlaumont says. Joel Berendzen in LANL’s Energy Computational Science Graduate Fellowship (DOE CSGF) recipient. “One of my of America, first converts utterances into perceptron. The perceptron measures The model is a step forward, says problems is I don’t really settle on one.” spectrograms — frequency, duration and the relevance of the learned SOM D. Kimbrough Oller, Warlaumont’s Applied Modern Physics That willingness to follow what fascinates him set Clark on a new course after intensity represented as 225 shaded pixels on features to various categories, classifies doctoral advisor. “We’re testing its basic Group, Clark made an completing his DOE CSGF practicum in summer 2009. Characteristically, he told coordinators a square. Those are sent to a type of neural the protophone and determines which capabilities and developing the scripts at the national laboratories he just wanted to work in an interesting group investigating network, called a self-organizing map SOM nodes best distinguish one utterance and tools we need to go on to much more unusual leap: from the interesting problems. Aric Hagberg, coordinator at Los Alamos National Laboratory (SOM), of 16 nodes mathematically from another. exact things.” (LANL) and himself a DOE CSGF alumnus, connected Clark with Nick Hengartner, a arranged in a four-by-four grid with After training, the perceptron can Warlaumont will have lots to contribute engineering-oriented Oresearcher with the Information Sciences and Technology Metagenomics team. randomly weighted connections classify each protophone input by type to the effort, Oller says. She “is going to have world of fluid flow to the Metagenomics approaches biotechnology in a new way. Rather than sequencing the between each. and infant age and identity, based on SOM a very significant academic career in helping DNA of individual microorganisms, it decodes genomes from entire communities of The SOM is “trained” by matching layer node activations. In tests, the model to establish not only new foundations in biological realm of organisms, like in the human gut or a tidal pool. Then it compares chunks of sequence data random spectrograms with the nodes performed significantly better than the theory of vocal development, but the to learn important information about the organisms’ genes. metagenomics. whose weights are most similar to it, then chance. It guessed the correct protophone application of the tools that she’s developing.” When he started his doctoral studies at Cornell University in 2008, Clark had planned to do research in computational fluid dynamics (CFD). He loves the subject, which he ~~~~~ began studying as an undergraduate at Oregon State University, but it’s a fundamental field that’s been scrutinized since computers were invented. He wanted to get into an area where new things are happening and changing and “I might be able to make some big contributions right off the start.” Goodbye, CFD. Hello, DNA. After a summer working with Hengartner and Joel This schematic shows the neural network Berendzen in LANL’s Applied Modern Physics Group, Clark made an unusual leap: from used to classify infant vocalizations. It the engineering-oriented world of fluid flow to the biological realm of metagenomics. combines a self-organizing map (SOM) He returned to Cornell with a new research focus. “I was hooked and I haven’t looked and a single-layer perceptron. Pixels back,” Clark adds. of an utterance are presented first to “It’s a field where you can very quickly describe what’s going on, which I really like the SOM. Activations of SOM nodes because my friends and family can understand what’s happening,” Clark says, “but then are then sent to the perceptron output getting the actual solution is much more difficult.” nodes for classification according to protophone, age and infant identity. Metagenomics is possible thanks to rapid “shotgun” genome sequencing technology, The weights from the input layer to which breaks an organism’s DNA into chunks, then copies and “reads” them to record the the SOM layer are trained first, then order of the base pairs. With today’s fast, relatively inexpensive sequencing scientists can weights to the SOM are frozen and read genomes multiple times for greater accuracy. The DOE supports metagenomics the perceptron’s weights are trained. research as it seeks genes that can help microorganisms convert plant materials into biofuels more efficiently and cheaply.

P16 DEIXIS 11 DOE CSGF ANNUAL DEIXIS 11 DOE CSGF ANNUAL P17 practicum profiles

Welcome to the (sequence) club

Local Alignments for soi: Escherichia coli str. K12 substr. DH10B vs Bacillus subtilis subsp. subtilis str. 168

Velvetrope uses two filters to find offsets — areas where The Velvetrope name is meant two genetic sequences are Some programs approach sequence Velvetrope uses bitwise comparison to evoke the feeling of an exclusive more similar than expected. alignment by seeking matches of a to find similar offsets. A sequence of club, Scott Clark says. The algorithm This visualization shows four specified length, called “k-mers,” where k interest, with letters representing the DNA compares DNA or amino acid vectors that passed a global is a specified number of base pairs or base pairs or amino acids, is lined up with sequences to find similar sections, filter test with regions marked amino acids within a read. For example, a sequence from another genome. To find gradually restricting criteria to find and shaded as having possible the most relevant areas. sequence alignments. Blue the DNA read ATCGGC has the 4-mers matching offsets, the first sequence ATCG, TCGG and CGGC within it. The remains stationary while the second is “The idea was it was like a means no match. Cyan marks nightclub and you would get deeper program uses k-mer matches as “seeds” to shifted to the right or left, one letter at a possible random match and deeper into it with more and index the genome. But “if you have a a time. because it doesn’t occur in more credentials,” Clark says. “The the same region. Yellow marks sequence that’s not superconserved “You’re comparing No. 1 to No. 1. barrier at a nightclub is always a regions of possible alignment (completely identical) or where every third Then you shift the second sequence over velvet rope.” Only sequences that and red marks positions in the amino acid might be slightly different — one. Now you’re comparing No. 1 to No. 2,” match certain specifications get to region that have matches. The statistically that’s a significant conservation” Clark says. “By shifting one sequence and the “club’s” most exclusive part. location of the region is marked worthy of study, Clark says, but some comparing it to a stationary sequence, Clark and Los Alamos National along with the total density of programs won’t spot it. you’re effectively comparing all the Laboratory researchers Nick matches within that region. different offsets, which allows for some Hengartner and Joel Berendzen behind the velvetrope really quick computational techniques.” kicked around several names, like But metagenomics also produces an It’s rare for large sequence chunks to The approach Clark helped develop, With metagenomics producing “bouncer,” before Berendzen came enormous and complex biological puzzle, be identical from genome to genome, but named Velvetrope (see sidebar), skips datasets in the terabytes — millions of up with “Velvetrope.” They weren’t concerned that another genomic as scientists reassemble and analyze how similar reads align helps scientists finding k-mers in favor of seeking areas megabytes — speed was important to tool is called Velvet because it billions of DNA chunks representing understand the sequences’ importance. “If where sequences share many similarities. Clark, Hengartner and Berendzen. They involves genome assembly, trillions of base pairs. Only massive a particular region was conserved across “It’s a complex search for mostly conserved took a computer science approach that not searching. computational power can do the job. many different genomes, possibly in areas or statistically significant conservation, emphasized simplicity and parallel Ironically, since starting different places, it almost surely has some using filtering effectively and doing it in a operation. Velvetrope’s modular approach sequence assembly research last getting aliened genetic benefit,” Clark says. “You also can very modular way so it can be readily makes it especially amenable to computation summer at Lawrence Berkeley At LANL Clark worked on algorithms find relative comparisons like mutation parallelized,” Clark says. on general-purpose graphics processing National Laboratory, “I’m using to find alignments: chunks of DNA in and silent mutation rates within these Velvetrope uses two filters. The global units (GPGPUs), the chips made for video Velvet all the time,” Clark says. which a sequence of base pairs or the conserved sequences across all genomes. filter finds offsets — areas where two games that now are a major part of the amino acids they code for are highly That would allow us to get some sequences share a higher-than-expected newest supercomputers. similar in the genomes of different information about the evolutionary similarity. The second, local filter looks organisms. Scientists say these sequences distance” between two organisms. at those offsets more closely to find areas are “conserved” because they weren’t Sequence alignment algorithms are with higher-than-expected similarities excised through natural selection. Once available for parallel computers, but they within them. “Each filter is a statistical researchers have multiple DNA chunks, demand a lot of power, Clark says. Other step to see whether there’s a significant or reads, with signatures of conserved algorithms have difficulty dealing with amount of matches,” Clark adds. “You By shifting one sequence and comparing it to a stationary sequence, sequence, Hengartner says, they can see transpositions — like gene A appearing slowly refine that search until you get just you’re effectively comparing all the different offsets, which allows for how sequences align in multiple genomes before gene B in one sequence but B the area you’re interested in.” “just to make sure they’re all pretty much appearing before A in another. They try some really quick computational techniques. from the same place.” to force the two sequences together, “and invariably either give you that the As align ~~~~~ and nothing else aligns, or the Bs align but nothing else does.”

P18 DEIXIS 11 DOE CSGF ANNUAL DEIXIS 11 DOE CSGF ANNUAL P19 practicum profiles

Hengartner says he and his Los Alamos difficult, Clark says. “When you compare developing builds formal likelihood biologists and … sometimes it would be colleagues pushed Clark to adapt the code two sequences you found in the wild, models of an assembly given the reads sort of hard to have each of us understand for GPGPUs, stretching the fellow’s there’s no answer key that says this should already available. That lets researchers what the other one meant.” Clark “was Above: This shows two sets of alignments of Below: Comparing a sequence abilities and enabling the program to run align perfectly or this conservation is find the probability of a given assembly very valuable from that point of view ­— two Lectin protein sequence domains, with of interest against a large set of on desktop computers. “We’re developing enough.” Yet, Velvetrope did find the same and use it to validate the assembly or suggest and then he’s also just wicked in terms red corresponding to the Lectin1 domain test sequences lets Velvetrope tools that a biologist or practitioner can or similar areas of alignment as BLAST and changes to improve the score. of computation.” and Green the Lectin2 domain. Orange find areas within that sequence use on his desktop,” he adds. “This is HMMer, two gene-alignment software Peter Frazier, Clark’s doctoral advisor Clark is one of the few people who is the Velvetrope alignment obtained by remapping the areas of high shared identity that are homologous to multiple something that is very dear to my heart tools. The researchers are making Velvetrope and an assistant professor in Cornell’s could tie a career in computers to cows. test sequences. This histogram across the previous set to the sequence of — developing something that is usable by available for free download as an open School of Operations Research and The Oregon native started programming combines information about shared interest. Blue is the alignment from BAliBASE, the community.” source project. Information Engineering, is collaborating at age 8, “much to the chagrin of my identity (solid blue columns) and a traditional sequence alignment program. “club membership” (light, always Hengartner praises Clark for achieving with Clark, Wang and LBNL’s Rob Egan parents, who were both journalism It resolves the two-domain comparison by larger green columns) — areas the goal with a minimum of direction. “It’s some assembly required on a paper describing the scoring algorithm. majors.” He played with PCs from garage manually specifying which domain to align of high local homology — across not like there was a tremendous amount of Clark’s fascination with metagenomics Next they want to optimize it to improve the sales, and in high school launched a Web (Lectin1 in the first alignment and Lectin2 many test sequences. This helps hand-holding. I was highly impressed” with led him to a second practicum in summer quality of an assembly, Frazier says, with an page design business. At Oregon State he in the second). This causes the aligning find areas of the sequence of Clark’s initiative. 2010 at Lawrence Berkeley National ultimate goal of applying the algorithm to started in mathematics and computer algorithm to append whatever domain interest that are shared among a Comparing Velvetrope’s performance Laboratory (LBNL), where he worked with metagenomic assembly. science but switched to math and physics. is not specified to the beginning or end large percentage of the whole set. with other gene alignment software is Zhong Wang, a researcher in the Joint Clark and Frazier also are working While still a freshman, he talked his way into of the alignment. Velvetrope picks out Genome Institute, on a project connecting with researchers from Cornell’s College of a graduate-level computational physics homologous areas regardless of position in multiple DOE labs. Wang’s lab emphasizes Veterinary Medicine and Department of course with legendary professor Rubin the sequence without specification. Lectin1 has low homologous identity in the latter InClub and Score matches across all alignments. sequence assembly — using computers to Computer Science on genomic approaches Landau, who later became his mentor. “From Global filter of sigma > 7.0 then local filter of CDF. part of its domain; therefore Velvetrope Total genes compared: 20. Reference Gene: Escherichia coli str. K12 substr. DH10B put DNA reads in the correct order. One to find bacteriophages — viruses that that day on I was in love with computational doesn’t pick it up while the non-homology of the algorithms Clark worked on uses infect bacteria — able to treat Escherichia science,” says Clark, who earned bachelor’s Residues: components of BAliBASE align it. It’s still 401-500 sequence reads early in the process, rather coli-related ailments in cows. The researchers degrees in mathematics, physics and picked up by the Velvetrope “club” filter. Cons. Ident. than toward the end as is traditionally done, hope to find useful features in the phages’ computational physics. Start(m) and End(c) to accelerate assembly. The approach also genomes, then use machine learning, In the long run, Clark sees himself Residues: 501-600 helps make the calculations easy to run on optimization and experimental design to starting a biotech or computational

Cons. Ident. parallel computers. mix the best viruses in a “cocktail” for science business. His lab experience, Start(m) and End(c) At LBNL Clark also worked on a bovine infections. however, also has him considering the Residues: 601-700 machine-learning algorithm that scores a With his unusual combination of DOE labs.

Cons. Ident. sequence assembly’s accuracy based on the expertise in computation, mathematics Start(m) and End(c) data it generates. Current assembly scoring and biology, “Scott’s been playing more Residues: metrics are loose and somewhat arbitrary, of an advisory role,” to bridge the gap

Frequency of matches across all other genes. 701-800

Cons. Ident. Clark says. The approach he and Wang are between the two research communities, Start(m) and End(c) Score = blue and InClub = green and further in shades of red Frazier says. “We’d be talking with Sequence compared against, red text indicates at least ratio=0.5 InSetInClub Sequence from 401 to 800 of 1343.

P20 DEIXIS 11 DOE CSGF ANNUAL DEIXIS 11 DOE CSGF ANNUAL P21 alumni profiles

Alumni Profiles From Alumni to Leaders

DOE CSGF graduates move to the front in their fields

A Career at the Since 1991, the Department of Energy Computational Science Graduate Fellowship Brian Moore Nuclear Intersection has seeded government laboratories, industry GE Hitachi Nuclear Energy and academia with scientists trained to apply shigh-performance computing to challenging research problems. Alumni have made outstanding Brian Moore‘s father had some tough questions for Nuclear Energy, an alliance between the iconic American and contributions, including improved fusion plasma him when a Soviet nuclear reactor exploded in April 1986. Japanese companies to develop and market nuclear technology, models, insights into evolution of drug-resistant Moore was just beginning his nuclear engineering studies services and fuels. diseases and more efficient and economical nuclear when the Chernobyl accident occurred, spreading radioactivity “I still feel like I’m sitting on this intersection between all power plant designs. More importantly, they’ve over Western Europe. The accident hobbled the entire nuclear these cool things,” Moore says from his office in Wilmington, N.C. helped develop computer simulation and benergy industry, even though — as Moore told his father — a “I get to play in all of it. If we’re doing computational fluid computational science into a research tool that poor design and lax safety standards were to blame. dynamics simulation of part of the reactor, I get to see that. If equals theory and experiment. As these profiles “I did have some heartburn from, ‘Well, why are you still we’re doing the heterogeneous full transport calculations for part show, alumni are the DOE CSGF’s best promoters. going into this stagnant industry?’” Moore says, but he liked of the fuel bundle, I get to play in that. If I’m getting the whole the field’s combination of nuclear physics, reactor engineering, system aspect or even what happens when someone wants to applied mathematics and computational science. “It’s this sort do a new application, I get to play in that, too.” of in-between world where they all interact.” Moore and his team create predictive reactor core models, Predicted linear heat generation rate (kW/ft) distribution Moore, a Department of Energy Computational Science combining the physics of fission, fluid dynamics and other across a boiling water reactor radial plane at an elevation of Graduate Fellowship recipient from 1992 to 1995 — part of just processes with decades of reactor operation data. The resulting 21 inches from the base of active fuel early in the operating the second class in the program’s 20-year history — has inhabited calculation is fairly straightforward, Moore says. “The only cycle. The predictive capability is verified via non-destructive that world since earning his doctoral degree from North Carolina problem is we can’t solve it explicitly; it’s too complicated and fission product signature measurements. Examined for every pin at each point in the cycle and compared to the fuel design State University in 1996. He joined the nuclear energy arm of the it’s too big, so we end up making any number of different kinds Moore also helped analyze new reactor core monitoring limit, the bundle and/or core design can be optimized to General Electric Co. and now manages the Methods and Software of approximations.” instruments and core and fuel properties in the Economic maximize performance while meeting all safety limits. Development Center of Excellence. He’s also part of GE Hitachi To help compensate, Moore and his colleagues may couple Simplified Boiling Water Reactor, which, in addition to the detailed models into broad simulations. For instance, they may tie Image Copyright GE Hitachi Nuclear Energy. PRISM Advanced Recycling Center, is a leading-edge design two-dimensional “slices” of fuel assemblies into a three-dimensional for GE Hitachi. setting to calculate a reactor’s full power distribution. The data is And since 2000, Moore has been part of Global Nuclear Fuel, “in a system where we can both use it to prognosticate what the to know how you’re going to use it” and whether the application a venture linking GE, Hitachi and Toshiba. “I have the opportunity This cutaway drawing of the GE next two years might be or, in what we’re required to do in licensing, is conservative enough to account for biases and uncertainties. to look for areas of synergy where we can help each other out … Hitachi Economic Simplified Boiling what the next 5 seconds” might hold. The last nuclear plant built in the United States went on line and try to avoid duplication of effort so we can maximize the Water Reactor pressure vessel shows “After we’ve created these nice big physics tools [we] walk in 1996 — the same year Moore earned his doctorate. But GE still brilliant people we have.” the core and other features. Brian them through regulatory licensing approval,” Moore says. The U.S. builds in other countries and, just as importantly, helps modify Moore’s elevation to management has given him another area Moore and his team at the Methods Nuclear Regulatory Commission (NRC) — and similar bodies in existing plants for greater efficiency and capacity. Moore and his to play in: spreading the word about how computational science and Software Development Center countries where GE and GE Hitachi build and service reactors colleagues calculate the consequences of putting more fuel rods can boost GE Hitachi’s overall business. “Taking on the leadership of Excellence helped analyze new — not only review reactor designs, but also mathematical tools in an assembly and worked on MELLLA (Maximum Extended Load positions has been very good for me to stretch beyond my comfort reactor core monitoring instruments used to estimate their capacities and limits. Regulators aren’t Line Limit Analysis) Plus, GE Hitachi’s technology to boost output zone, and that’s what I really like.” and core and fuel properties for the design. “just concerned with, ‘Is it doing the right thing?’ They also want on boiling water reactors.

Image Copyright GE Hitachi Nuclear Energy.

P22 DEIXIS 11 DOE CSGF ANNUAL DEIXIS 11 DOE CSGF ANNUAL P23 alumni profiles

Dan Martin Lawrence Berkeley National Laboratory

“Cool Problems” Draw Alumnus to Laboratory

It may help to think of Dan Martin as a carpenter That’s just what one of Martin’s latest projects needs. He and Martin spends a lot of time extending AMR and other algorithmic whose main tool is a trusty multipurpose saw. his colleagues are part of ISICLES, the Ice Sheet Initiative for techniques to new and bigger problems. Application scientists Martin, a Department of Energy Computational Science CLimate ExtremeS, a DOE program to develop better models of can’t “just take a problem and drop it into AMR. Everything has its Graduate Fellowship (DOE CSGF) recipient from 1993 to 1996, land-based ice sheets. “Ice sheets have an enormous range of own refinements and its own algorithmic complexity. Doing it works with scientists in a variety of fields to build computer dynamic scales. You need incredibly fine resolution to get some correctly for any given problem often requires work and research.” imodels. He’s helped fashion algorithmic “lumber” into simulations features right,” like the grounding line, the point at which an ice The Applied Numerical Algorithms Group also focuses on portraying devices and phenomena ranging from fusion plasmas’ sheet slides off land and onto the ocean. “But there also are large improving AMR to higher orders of accuracy and on finding rapid reactions to ice sheets’ glacial movements. regions where nothing’s going on. You really can’t computationally better ways to solve the partial differential equations at the “It’s fun to work on really cool problems,” says the researcher afford to resolve everything at the fine scale.” heart of computer models. in the Applied Numerical Algorithms Group (ANAG) at Lawrence The LBNL project, called Berkeley-ISICLES or BISICLES, applies Martin also is on the development team for Chombo, an Berkeley National Laboratory (LBNL). “That’s one of the advantages AMR to the Glimmer Community Ice Sheet Model, often a component ANAG-developed software package for adaptively refined of being a computational scientist. I’ve been all over the place. I’ve of larger climate simulations. The team also uses other algorithmic rectangular grids, and frequently answers questions from gotten to learn a lot about a lot of really cool and interesting areas.” adjustments, including approximating the vertical structure or researchers who use it. “There’s an impressive number of really The versatile tool Martin and his colleagues often use is distribution of the ice-sheet velocity and coupling it to equations for capable people who are using our software to attack problems adaptive mesh refinement (AMR), a computational technique that the sheet’s horizontal movement and thickness evolution. That “lets us we don’t have time or resources to look at, but are still good focuses a computer’s power where it’s needed most. AMR get away with a two-dimensional vertically integrated solver that’s also candidates for AMR.” automatically casts fine, high-resolution grids or meshes of data a big win” in saving computational power. One of AMR’s creators, ANAG leader Phillip Colella, was his points in areas of interest in a simulated domain — like the point doctoral advisor at the University of California, Berkeley, but where two fluids interact — and coarser meshes elsewhere. Martin says his 1994 practicum under combustion modeler and Computers calculate changes in physical properties at each data AMR pioneer John Bell was what set him on course for a point to provide a complete picture. “You can think of AMR as laboratory career. “I really enjoyed working with the people,” This is a demonstration of an algorithm that uses giving you essentially fine-grid accuracy for essentially coarse-grid Martin adds. A lab career “seemed like a way to be able to do lots computational meshes localized in both space and time (computational) cost,” Martin says. of fun, exciting research without having to go through the tenure to compute incompressible viscous flows. It shows two process.” He’s also found a national laboratory is “one of the vortex rings angled toward each other at initial time places where teamwork … just works so much better.” and after 60, 90 and 120 time steps. Black lines depict streamlines. Green boxes are a level 1 mesh, with a As LBNL’s DOE CSGF practicum coordinator, Martin pairs resolution four times finer than the background mesh. students with researchers in complementary disciplines. “I meet Blue boxes are a level 2 mesh, with a resolution four This visualization of a simulation of Antarctic ice sheet movement people all across the lab and they think I’m great because I bring times finer than level 1. For clarity, grid boxes are shown shows the magnitude of the ice sheet’s velocity in meters per them all these fantastic people,” he adds. “It’s definitely improved only in the rear half of the computational domain. year, with blue-white the lowest velocity magnitude and my knowledge of what’s going on at the lab in a way that’s blue-green higher magnitude. The visualization also shows been fun.” the adaptive computational grid used to calculate the velocity magnitude. Small, black boxes have a resolution of 2.5 kilometers and purple boxes are at 5 kilometers. The base resolution for the simulation is 10 kilometers.

P24 DEIXIS 11 DOE CSGF ANNUAL DEIXIS 11 DOE CSGF ANNUAL P25 alumni profiles

Mala Radhakrishnan Wellesley College

Alumna Busts Resistance — in the Classroom and Lab

Mala Radhakrishnan is out to override Just as importantly, Radhakrishnan is sharing her expertise She’s already compiled a strong record of achievement. In resistance. As an assistant professor of chemistry at Wellesley and setting a standard for Wellesley students. She found she was a paper published last year in the journal Biotechnology Progress1, College, much of her research aims to outmaneuver mutations at home in the classroom while participating in the Teach for Radhakrishnan and Bruce Tidor, her doctoral advisor at the in viruses like HIV that make some drugs nearly useless. America program after earning her undergraduate degree. Massachusetts Institute of Technology (MIT), computationally Radhakrishnan, a Department of Energy Computational “During grad school I actually missed teaching. That’s another varied the binding behavior of cytokines, proteins that attach to mScience Graduate Fellowship (DOE CSGF) recipient from 2004 to reason I knew academia would probably be a better fit for me,” cells to regulate them. Erythropoietin (Epo), for instance, binds to 2007, builds computational models for drug design. Sometimes Radhakrishnan says. “The teaching, the mentorship is the best surface receptors to regulate red blood cell maturation. those models design and analyze “promiscuous” molecules — part,” whether it’s a chemistry class or writing computer code. Radhakrishnan and Tidor modeled how changing the time ones capable of targeting multiple viral variants. Radhakrishnan “It’s really fun for me to see a student who has never written a cytokines bind to their receptors could affect their longer-term also uses optimization methods to help find effective molecular piece of code before get excited when something works.” potency. The longer a cytokine remains bound, the more likely combinations to block many variants with minimal side effects. That’s an especially vital message at Wellesley, Radhakrishnan it is to be internalized and degraded, making it unavailable to bind And in current research she’s searching for common factors in says. Teaching at a women’s college is important because “there at another receptor. Their model indicated that a weakly binding Mala Radhakrishnan and Bruce Tidor built a computational HIV drug resistance. still aren’t enough female computational scientists out there” cytokine works better when less of it is available or when cells model of cytokine behavior illustrated here. Cytokines are — something she and her students are reminded of every time quickly internalize and degrade it, for example. In those cases small proteins that bind to specific cell surface receptors. they attend a conference or take a job. the cytokine releases from the receptor and is recycled for binding Cytokines may bind to receptors to activate a signal in the At Wellesley, Radhakrishnan carries out most of her projects at another site. Strongly binding cytokines are better in other cell, then release to bind at another site. Others are brought into the cell where they may remain activated before they are on a computing cluster with 80 processing cores, but she’s situations — when they activate slowly or when the cell has recycled back to the cell surface or degraded in the lysosome. contemplating bigger computers soon, when she takes a year fewer receptors. off from teaching to focus solely on research. “A lot of researchers, when they design drugs, they want things to bind as tightly as possible,” Radhakrishnan says. “This shows an example where that can be a detriment. You lose the Radhakrishnan adds. It could lead to simulations that better capture long-term potency.” the fine detail of biochemical interactions. It took a large-scale cellular model to test that idea. In other At the other end of the scale, Radhakrishnan is working with Results of simulations to determine instantaneous and integrated cytokine activity research Radhakrishnan used molecular-scale models to design Stanford University medical researcher Robert Shafer to analyze a at 15,000 minutes as a function of surface/endosomal dissociation rate constant two mutant Epo receptors. The receptors selectively bound to one database of HIV drug-resistance data. “I’m looking for patterns and (koff) and activation rate constant with a liganded receptor (ksigf). Figures (A) and (B) of two different parts of the Epo molecule, letting Radhakrishnan trying to understand what major drug-resistant genotypes and are for endosomal koff and ksigf for parameters similar to those for the cytokine and her MIT colleagues explore how selective binding affects phenotypes are out there,” she says. “Is there a set of representative Granulocyte-colony stimulating factor (GCSF). Figures (C) and (D) show activity and cell response. ways a person could be resistant to HIV drugs?” By identifying key integrated activity as a function of endosomal koff and ksigf for the same parameters, Radhakrishnan takes an equally small-scale approach in a drug-resistance factors, researchers could design “promiscuous” with surface koff held fixed to reflect mutant GCSF with poorer binding only in the collaboration with fellow DOE CSGF alumnus Jaydeep Bardhan. new drugs or drug cocktails with multiple infection targets. endosome. The alteration further maximizes instantaneous and integrated activity. Bardhan, an assistant professor at Rush University Medical Center Radhakrishnan will focus on new challenges during her year Figures (E) and (F) show plots for instantaneous and integrated activity for koff and in Chicago, has developed a way to efficiently calculate of research. Meanwhile, she’s also tending to her infant son, Samay, ksigf under parameters similar to those for the cytokine Erythropoietin (Epo). The plots show that tighter binders were consistently preferred in the Epo system. In electrostatics — electrical charges and fields — that influence and recently published a collection of “chemistry poetry” entitled

all cases the black line indicates the experimentally measured koff at cell surface pH. biomolecules. “I’m working with him to try to apply it to biological Atomic Romances, Molecular Dances. systems and see in what cases it works well and when it doesn’t,”

1 Radhakrishnan ML, Tidor B. Cellular-level models as tools for cytokine design. Biotechnol Prog. 2010;26:919-937. P26 DEIXIS 11 DOE CSGF ANNUAL DEIXIS 11 DOE CSGF ANNUAL P27 essay contest winners Winning Essays Encouraging communication through AN annual writing contest Pictured here is dihydrofolate reductase, a protein whose function is crucial for cell proliferation. Drugs that bind to the active site of this protein and inhibit its activity Winner Can peeling an are used to stop the rapid So how can anyone possibly study a proliferation of cancer cells. onion cure cancer? protein with thousands of these slippery, The yellow shaded region mysterious little particles? surrounds the active site. by Kenley Pelzer Since electrons are too small to see through a microscope, chemists predict their behavior with sophisticated computer Awarding “With no offense to the quantum However, it is absolutely crucial to simulations. Not only do these simulations they are usually referred to as “quantum theorist in the room,” the professor said, consider quantum theory in some cases, have the power to study electrons, they mechanics” calculations. ComMunication casting me a quick look and smiling gently, because its messy, complicated rules can also have economic advantages: Once the When a lower level of accuracy is electron distribution might affect the “you should never, ever deal with quantum help us understand diseases — and how to software has been developed, it costs acceptable for a particular layer, the active site’s behavior. So scientists effects unless you have to.” As an eager cure them. virtually nothing to use it to study large program may use a less advanced (and compromise and use a less precise method: The DOE CSGF young graduate student in theoretical The human body contains a multitude numbers of potential drugs. Given the less time-consuming) quantum either a quantum mechanics method that launched an annual quantum chemistry, I was enthralled by of protein molecules that play a central massive costs involved in pharmaceutical mechanics method. In cases where is a little less accurate (and hence faster), essay contest in 2005 the wonderful and bizarre laws Einstein role in whether we stay healthy or get sick. research, this advantage has important even more approximate information or a molecular mechanics method. to give current and and other great scientists had discovered. Each protein contains thousands of electrons implications in the search for is sufficient, the computer performs The farther we are from the active former fellows an Yet I knew the professor had a point: If you — tiny, negatively charged particles that effective medicines. a “molecular mechanics” calculation. site, the less we need detailed information opportunity to write Wunnecessarily involve these strange are attracted to positive charges. These The bad news is that — thanks to the Because molecular mechanics methods about electrons and their quantum about their work with a principles in your approach to a scientific attractions influence whether a protein influence of quantum properties — these are less precise and incorporate less mechanical behavior. So we use a less broader, non-technical problem, it will be harder — or perhaps “sticks” to another molecule. simulations may take weeks to generate quantum mechanical theory, they’re fast precise simulation technique with each audience in mind. The impossible — to solve it. This biological stickiness is especially information about just a few electrons. — really fast. So by applying molecular successive layer. By using faster techniques competition encourages In the world of quantum mechanics, important in treating cancer. In some Since a protein has thousands, accurately mechanics and working with information to treat the outer layers of the “onion,” it’s better communication of objects behave in very peculiar ways. An types of cancer cells, there are protein predicting the distribution of electrons that’s a little less accurate, the simulations possible to predict how a large protein computational science object can pass through walls. It can be in molecules with a region called an “active could take decades. Not very helpful gain a lot of speed. will interact with a particular drug. And and engineering and two places at the same time. No matter site.” When certain molecules attach to the when cancer patients are hoping for a To understand how the ONIOM fortunately for patients waiting for the its value to society to how hard you try, you can never be sure active site, they trigger a cascade of events new drug now. approach can help us answer important next new medicine, the answers can be non-expert audiences. where it is or where it is going. Fortunately that leads to disease. But if a drug can This is where the idea of peeling an questions, consider the example of a drug obtained in days or weeks — not decades. In addition to for scientists, these quantum effects are target a particular protein and stick to its onion comes to the rescue. With the aptly that needs to bind to a particular protein. By peeling apart a protein as though recognition and a cash only important for the smallest particles active site, it can block other molecules named ONIOM method (“Our own At the protein’s active site — the onion’s it were an onion, scientists follow my prize, the winners and can be ignored in many situations. from binding and prevent this dangerous N-layered Integrated molecular Orbital center — the computer uses an advanced professor’s advice to never, ever deal with receive the opportunity spiral. If a drug can’t stick to the protein molecular Mechanics”) the protein quantum mechanics method to calculate quantum effects unless you have to. The to work with a — if it falls off and circulates in the molecule is divided into a chosen number the behavior of electrons and predict beauty of this approach is that for many professional science bloodstream — it may just float around (N) of layers. The program treats each whether a drug will “stick.” physiological proteins, neither quantum writer to critique and doing nothing or, worse yet, cause toxic with a different simulation technique, Then the computer must simulate the mechanics nor molecular mechanics alone copy-edit their essays. This image shows a molecular side effects. carefully selected based on what next layer of the onion — or rather, the could effectively answer our questions. The latest winning essays orbital calculated using To predict which drugs will attach information is needed. When highly next layer of the protein — which wraps Fortunately, collaboration between are published here. a quantum mechanics firmly to an active site, scientists must accurate information on the electronic around the active site but isn’t part of the scientists who specialize in each method For more information program. By calculating understand how protein molecules and structure is needed, the program uses a active site. Since this layer doesn’t actually has led to ingenious hybrid programs that on the essay contest, visit the shapes and orientations their electrons behave. But because “molecular orbital” method that rigorously stick to the drug of interest, we don’t need can contribute great insights to drug of the molecular orbitals, www.krellinst.org/csgf. electrons are so small, they obey all of the incorporates quantum theory. The details to worry quite as much about its electrons. design. And then with the press of a key, a which may contain electrons, complicated laws of quantum mechanics. of quantum theory are so important for On the other hand, we can’t totally ignore computer can guide us in the urgent quest scientists obtain detailed information on the distribution these molecular orbital calculations that this layer, because changes in its shape or to develop life-saving medicines. of the electrons in space.

P28 DEIXIS 11 DOE CSGF ANNUAL DEIXIS 11 DOE CSGF ANNUAL P29 essay contest winners

On the Quantification of “Maybe”: Honorable Mention A Niche for Computation

by Hayes Stripling

The error bars represent predictions (with uncertainty) Consider perhaps the most science. Using computational power to of the time required for laser energy to ablate a pressing question facing the global simulate a system or experiment is beryllium disk. We use existing experimental data community in this and coming decades: relatively inexpensive and has increasing to “tune” our uncertain parameters and reduce “Is human activity adversely affecting the potential to provide accurate, thorough the magnitude of the predictive uncertainty. The analysis shows that between experimental results, environment and global climate?” A brief results. Most importantly, computation the error bars are comparable to the experimental skimming of newspaper headlines, also has surfaced as the most valuable variability. Predictions at extrapolated (larger) disk scholarly journals and political debates tool for exploring uncertainty in thicknesses, however, are less informed by the reveals that credible responses to this “yes/no” questions. experimental results and have larger uncertainty. “yes/no” question all have a flavor of The specific branch of computation C“probably,” “maybe” or “probably not.” In that addresses this issue is uncertainty fact, most scientifically based stances quantification (UQ). UQ — or the concede we cannot provide a definitive quantification of maybe — aims to Each column of the figure illustrates “yes/no” answer without many more years’ comprehend how uncertainty in nature 215, or 32,768 experiments — certainly a consumer than results from only a handful something we could never determine a calibrated or posterior distribution worth of data, investigation and discovery. affects a particular quantity of interest number too expensive and time-consuming of experiments. using physical experiments alone. of a single uncertain input to a So the correct answer today is just (QOI). It then uses this understanding to to consider. Cutting-edge simulations running on Of course, it’s impossible for massive climate model. The bar “maybe,” with some evidence and make predictions or informed decisions Computation has proven extremely some of the largest computer architectures computation to ever make experimentation charts (histograms) represent the proponents on either side. about the quantity. Achieving this goal, valuable in such cases, in which in the world use this kind of approach to a moot practice. We must compare single-variable distribution and In response to scientific and political however, is not so simple, for the QOIs experimentation is economically or address society’s toughest challenges. But experimental measurements with the scatter plots illustrate how the pressure for a more definitive answer, we seek are complex results of politically infeasible. What if we can afford in some cases, the problems are becoming computational results to ensure our calibration of one variable correlates to the other two variables. For researchers have focused on this multidisciplinary systems that are only five crash tests? We may approach the too large and complex. For example, models are valid. Further, example, these results suggest reformulation of the question: “To what confounded by uncertain inputs and problem computationally instead. The models designed to describe the long-term experimentation’s maturity as a technique larger values of cldfrc_rhminl and extent does human activity affect the less-than-perfectly-understood physics. computer model would solve equations global climate predict thousands of and the ability to witness its physical smaller values of cldwat_icritc will environment and global climate, and how To illustrate the utility of computation that govern (or simulate) the physics that quantities of interest from thousands of results with your own eyes makes it more improve the accuracy and reduce do we manage the risks of this activity as in uncertain systems, let’s consider a simple take place during a crash. We could design uncertain inputs. To help handle data of credible to humans than computational uncertainty in future model runs. the scientific investigation continues?” example: determining the forces a passenger the model to accept any number of this magnitude we employ sensitivity number crunching. For example, let’s This is not posed as a “yes/no” question; experiences in a car crash. Imagine we can adjustable inputs that contribute to analysis, a sub-discipline of uncertainty return to our car crash case: Would you be instead, it calls for a quantification of identify the five most important factors the final QOI (in this case, the forces a quantification. By sampling results from more inclined to believe a vehicle’s safety “maybe” — an estimation of uncertainty (such as speed or seat-belt use) contributing passenger experiences). Inputs could span previous computer experiments we can report based on five physical experiments in our response. Answering the question to our QOI but that we can’t know exactly a range instead of adhering to a binary determine which outputs are most using real crash-test dummies or 1,000 also requires a multidisciplinary effort — what the settings or statuses of these inputs choice, like fast or slow speed. It would sensitive to which inputs. For example, if simulated smash-ups in which you can’t involving experts in science, engineering will be (that is, our inputs are uncertain). be common, given sufficient computer we find that the computed prediction of see, hear or feel the impact? Would you be and policy development — that will evolve If we restrict each input to one of two resources, to run hundreds or thousands global surface temperature isn’t sensitive more inclined to believe the computer if it as we gain understanding. settings (for example, speed is either fast of “computer experiments” corresponding to lunar cycles but is highly sensitive to exactly predicted the results of the five These kinds of questions have led to or slow) and wish to test each possible to hundreds or thousands of input cloud coverage, we can adapt our sampling physical crashes we could afford? We can an exploding demand for knowledge. In combination of inputs, we would require combinations. The final result is a strategies to explore cloud coverage more never abandon experimentation — but we reaction, researchers are expanding the the design, construction, crashing and distribution of our QOI, allowing us to thoroughly and hold off on varying the can leverage validated computer models to scientific method, which is founded on the analysis of 25 (32) test cars. This may be make an informed statement of the form, lunar effects. This will give us a more develop and explore scientific concepts long-lived pillars of theory and experiment, an acceptable number of experiments, “We are 99 percent certain that the precise understanding of global surface that will guide our society’s policies in the to explore new domains and strategies to depending on economic, administrative passenger G-force will be below the injury temperature behavior and use fewer face of the great challenges ahead. support decisions and conclusions. These and/or political constraints. But five inputs threshold in 99.9 percent of vehicle evaluations of the (potentially time- include computation, an invaluable tool probably are too few. A more realistic collisions.” Such a statement is much consuming) computer model. It’s also many now accept as the third pillar of approach may consider 15 inputs — that’s more useful to a policy-making board or

P30 DEIXIS 11 DOE CSGF ANNUAL DEIXIS 11 DOE CSGF ANNUAL P31 howes scholars scholars

The Frederick A. Howes Scholar in Computational Science award was established in 2001 to honor the late Frederick Anthony Howes, who was a champion for computational science education.

Howes award About Fred Howes Honors Leadership As Well As Scholarship

A Cuban émigré who has helped School of Engineering and Applied “To be honored for my work In the 10 years since it was first scientists better understand a strange Sciences and the MIT Department conferred, the Frederick A. Howes Scholar and, most importantly, for my physical force is the 2011 Frederick A. of Mathematics. in Computational Science award has Howes Scholar in Computational Science. “To be honored for my work and, most devotion to science is not only become emblematic of research excellence and outstanding The award recognizes outstanding importantly, for my devotion to science is leadership. It’s a fitting tribute to Howes, who was known recent doctoral graduates of the not only a tremendous honor but also an a tremendous honor but for his scholarship, intelligence and humor. Department of Energy Computational invaluable source of encouragement,” Howes earned his bachelor’s and doctoral degrees also an invaluable source Science Graduate Fellowship (DOE Rodriguez says. in mathematics at the University of Southern California, an obituary on the Society of Industrial and Applied CSGF) program. Winners are chosen 2010 Howes Scholar Julianne Chung of encouragement.” Mathematics website reports. He held teaching posts not just for their technical achievements, also felt invigorated. She was chosen, in at the universities of Wisconsin and Minnesota before but also for outstanding leadership, part, for her work supporting budding ~~~~~ joining the faculty of the University of California, Davis, integrity and character — qualities that scholars through groups such as the in 1979. Ten years later Howes served a two-year rotation brought Howes wide admiration. Association for Women in Mathematics with the National Science Foundation’s Division of AAs manager of DOE’s Applied and the Society for Industrial and Applied Mathematical Sciences. He joined DOE in 1991. Mathematical Sciences Program, Howes Mathematics. She’s continued that completed a National Science Foundation In 2000, colleagues formed an informal committee was a staunch defender of the fellowship involvement, but “with the award I feel Mathematical Sciences Postdoctoral to honor Howes. They chose the DOE CSGF as the and its goals. He died unexpectedly on more empowered, in that it’s not just my Research Fellowship at the University of vehicle and gathered donations, including a generous Dec. 4, 1999, at age 51. experience I’m sharing with students. It’s Maryland at College Park. This fall she contribution from Howes’ family, to endow an award in 2011 Winner Alejandro Rodriguez, a fellow from also Dr. Howes’ vision for computational, starts a tenure-track position in the his name. Alejandro 2006 to 2010, is the 15th Howes Scholar. interdisciplinary research, as well as the Mathematics Department at the Rodriguez Rodriguez graduated from the Massachusetts greater computational science community, University of Texas at Arlington, where Alejandro Rodriguez has been selected as the Institute of Technology (MIT) in June that I represent.” she’ll continue her work on inverse 2011 Howes Scholar in 2010. He’s now a postdoctoral researcher Since earning her doctoral degree problems and image processing. Computational Science. with a joint appointment in the Harvard from Emory University in 2009, Chung The Howes award occasionally Dr. Rodriguez was a fellow attracted attention as Chung interviewed from 2006 to 2010. He for positions. A faculty member at one PAST HOWES SCHOLARS received his Ph.D. from the major research university told her he’d Massachusetts Institute of known Howes personally. “In some sense, Technology and currently it was an immediate acknowledgement of 2010 Julianne Chung holds joint postdoctoral the value of my research, as well as the 2009 David Potere positions at the Harvard importance of my work,” she adds. “It 2008 Mala Radhakrishnan School of Engineering definitely has an effect on the connection 2007 Jaydeep Bardhan and Kristen Grauman and Applied Sciences and I have with the community.” 2006 Matthew Wolinsky and Kevin Chu at the MIT Department 2005 Ryan Elliott and Judith Hill of Mathematics. DOE CSGF recipients who have completed or plan to complete requirements 2004 Collin Wick for their doctoral degree in a calendar year, 2003 Oliver Fringer and Jon Wilkening both with fellowship support or after 2001 Mayya Tokman and Jeffrey Hittinger receiving support for the maximum number of years, are eligible for that year’s Howes award. They’re nominated by Above left: 2010 winner Julianne Chung receives her award from Dr. David Brown, department chairs, advisors and fellowship chair of the Howes selection committee and longtime friend of the fellowship. coordinators at their universities. A review Above right: David Potere, the 2009 Howes Scholar, presents his research at the DOE CSGF Annual Conference in Washington. P32 DEIXIS 11 DOE CSGF ANNUAL DEIXIS 11 DOE CSGF ANNUAL P33 howes scholars

fellows directory class of 2011

Alejandro Rodriguez, the 2011 Eric Chi Anubhav Jain Matthew Reuter Howes Scholar in Computational Rice University Massachusetts Institute of Technology Northwestern University Science, explains his research and Bioinformatics/Statistics Materials Science and Engineering Theoretical Chemistry award-winning poster at the 2009 Advisor: David Scott Advisor: Gerbrand Ceder Advisor: Mark Ratner annual conference. Practicums: Lawrence Livermore National Laboratory Practicum: Lawrence Berkeley National Laboratory Practicum: Oak Ridge National Laboratory and two at Sandia National Laboratories – California Contact: [email protected] Contact: [email protected] Contact: [email protected]

Milo Lin Sarah Richardson committee chooses one or two scholars each year to MIT summer MITES program for under-represented Gregory Crosswhite California Institute of Technology Johns Hopkins University School University of Washington Physics of Medicine receive an honorarium and engraved crystal memento high school students. He also has represented both the Physics Human Genetics and Molecular Biology at the DOE CSGF Annual Conference in Washington, MIT physics department and DOE CSGF at national Advisor: Ahmed Zewail D.C., where the recipients also deliver lectures conferences and is profiled in the Physics Society Advisor: Dave Bacon Practicum: Brookhaven National Laboratory Advisor: Joel Bader Practicum: Lawrence Livermore National Laboratory Contact: [email protected] Practicum: Lawrence Berkeley National Laboratory describing their research. “People in Physics” video series. Contact: [email protected] Contact: [email protected] David Potere, the 2009 Howes Scholar, said the “I love to share my work and passion for science talk was one of the honor’s most significant elements. with others,” Rodriguez says. “The less the audience Paul Loriaux At Princeton University he focused on applying knows or cares about science, the more I enjoy Hal Finkel University of California, San Diego Danilo Scepanovic Yale University Computational Biology Harvard University/Massachusetts computer power to analyze the huge quantity of data convincing them of its importance and beauty.” The Physics Institute of Technology remote sensing satellites generate, using it for things award recognizes the significance of mentorship and Advisor: Alexander Hoffmann Signal Processing/Cardiovascular Modeling Advisor: Richard Easther Practicum: Argonne National Laboratory like predicting epidemics. His Howes lecture focused scholarship, he adds, and he plans to use it “as fuel Practicum: Princeton Plasma Physics Laboratory Contact: [email protected] Advisor: Richard Cohen on what the satellites see. to continue to inspire others.” Contact: [email protected] Practicum: Sandia National Laboratories – New Mexico “I was hoping to just leave the room with a sense Rodriguez’s research has led the way in Contact: [email protected]

of how beautiful the remote sensing satellite imagery understanding the Casimir forces. These strange James Martin is and how much things are changing,” Potere says. attractions arise from quantum-scale fluctuations Steven Hamilton University of Texas Emory University Computational and Applied Mathematics Paul Sutter “We’re really at a tipping point now.” After graduation in the electromagnetic field and are affecting Computational Mathematics University of Illinois at Urbana-Champaign he joined the Boston Consulting Group, where he’s microelectromechanical systems now in development. Advisor: Omar Ghattas Cosmology Advisor: Michele Benzi Practicum: Sandia National Laboratories – California helped found a “geoanalytics team.” It plans to apply Rodriguez’s work has led to methods capable of Practicum: Los Alamos National Laboratory Contact: [email protected] Advisor: Paul Ricker geospatial technologies to problems such as store calculating these forces between arbitrarily shaped Contact: [email protected] Practicums: Two at Oak Ridge National Laboratory location, demographics and distribution optimization. objects. Previous techniques could only predict them Contact: [email protected] Potere says the Howes award surprised him between simply shaped objects. Rodriguez is first Matthew Norman because “I hadn’t really thought of myself that way.” author on numerous papers describing the research, Ying Hu North Carolina State University Rice University Atmospheric Sciences Cameron Talischi Yet, like Chung, he found “it was a nice piece of including ones published in the Proceedings of the Biomedical Engineering University of Illinois at Urbana-Champaign confirmation that there was a leadership impact National Academy of Sciences and Physical Review Letters. Advisor: Fredrick Semazzi Computational Mechanics Advisor: Rebekah Drezek Practicum: Oak Ridge National Laboratory and a leadership role” to his research. Rodriguez was 13 when he and his family fled Practicums: Lawrence Berkeley National Laboratory Contact: [email protected] Advisor: Glaucio Paulino For Rodriguez, the honor has special meaning Cuba after his stepfather, a physics professor, was fired and Lawrence Livermore National Laboratory Practicum: Sandia National Laboratories – New Mexico Contact: [email protected] Contact: [email protected] “after spending the past five years in the company of for refusing to identify the writers of a letter opposing extremely talented and passionate CSGF fellows.” the communist government. They settled in the Miami Geoffrey Oxberry Massachusetts Institute of Technology Rodriguez, the selection committee wrote in its area and became U.S. citizens. Joshua Hykes Chemical Kinetics/Transport Phenomena John Ziegler citation, “embodies the qualities that Fred Howes The award, Rodriguez says, recognizes the role his North Carolina State University California Institute of Technology promoted in all young scientists.” He is “not just an entire family, including those still in his native land, Nuclear Engineering Advisor: William Green Aeronautics Practicum: Sandia National Laboratories – California exemplary computational physicist,” but also “a had in his success. Naturally, they’re overjoyed: “My Advisor: Yousry Azmy Contact: [email protected] Advisor: Dale Pullin dedicated spokesman for physics and a mentor for father in Cuba almost went as far as publishing the Practicum: Oak Ridge National Laboratory Practicum: Oak Ridge National Laboratory Contact: [email protected] Contact: [email protected] younger scientists.” Rodriguez taught physics in the news in the national Cuban newspaper.” Alex Perkins University of California, Davis Theoretical Ecology

Advisor: Alan Hastings Practicum: Oak Ridge National Laboratory P34 DEIXIS 11 DOE CSGF ANNUAL Contact: [email protected] DEIXIS 11 DOE CSGF ANNUAL P35 fellows directory

th Year RD Year nd Year Fellows Fellows Fellows

Carl Boettiger Douglas Mason Edward Baskerville Noah Reddell Mary Benage Christopher Ivey Edgar Solomonik Heather Mayes University of California, Davis Harvard University University of Michigan University of Washington Georgia Institute of Technology Stanford University University of California, Berkeley Northwestern University Biology – Ecology and Evolution Physics Ecology and Scientific Computing Computational Plasma Modeling for Geophysics Computational Fluid Dynamics Computer Science Chemical Engineering Advisor: Alan Hastings Advisor: Eric Heller Advisor: Mercedes Pascual Fusion Energy Advisor: Josef Dufek Advisor: Parviz Moin Advisor: James Demmel Advisor: Linda Broadbelt Practicum: Lawrence Berkeley Practicum: Lawrence Berkeley Practicum: Lawrence Berkeley Advisor: Uri Shumlak Practicum: Lawrence Livermore Contact: [email protected] Practicum: Argonne National Laboratory Contact: [email protected] National Laboratory National Laboratory National Laboratory Practicum: Princeton Plasma National Laboratory Contact: [email protected] Contact: [email protected] Contact: [email protected] Contact: [email protected] Physics Laboratory Contact: [email protected] Irene Kaplow Jarrod McClean Contact: [email protected] Stanford University Zachary Ulissi Harvard University Scott Clark Britton Olson Sanjeeb Bose Aleah Caulin Computational Biology Massachusetts Institute of Technology Chemical Physics Cornell University Stanford University Stanford University Troy Ruths University of Pennsylvania Advisor: Daphne Koller Chemical Engineering Advisor: Alan Aspuru-Guzik Applied Mathematics Computational Fluid Dynamics Computational Fluid Dynamics Rice University Genomics and Computational Biology Practicum: Lawrence Berkeley Advisor: Michael Strano Contact: [email protected] Advisor: Peter Frazier Advisor: Sanjiva Lele Advisor: Parviz Moin Bioinformatics Advisor: Carlo Maley National Laboratory Contact: [email protected] Practicums: Los Alamos National Laboratory Practicums: Lawrence Berkeley National Practicum: Lawrence Livermore Advisor: Luay Nakhleh Contact: [email protected] Contact: [email protected] Robert Parrish and Lawrence Berkeley National Laboratory Laboratory and Lawrence Livermore National Laboratory Practicum: Lawrence Berkeley Georgia Institute of Technology Contact: [email protected] National Laboratory Contact: [email protected] National Laboratory Seth Davidovits Miles Lopes Theoretical Chemistry Contact: [email protected] Contact: [email protected] Princeton University University of California, Berkeley Advisor: David Sherrill Curtis Hamman Kurt Brorsen Plasma Physics Machine Learning Contact: [email protected] Stanford University Cyrus Omar Iowa State University Samuel Skillman Advisor: Nathaniel Fisch Advisor: Martin Wainwright Flow Physics and Computational Engineering Carnegie Mellon University Physical Chemistry University of Colorado at Boulder Contact: [email protected] Contact: [email protected] st Year Aurora Pribram-Jones Advisor: Parviz Moin Programming Language Design/Neurobiology Advisor: Mark Gordon Astrophysics Fellows University of California, Irvine Practicum: Sandia National Laboratories – Advisor: Jonathan Aldrich Practicum: Argonne National Laboratory Advisor: Jack Burns Leslie Dewan Peter Maginot Theoretical Chemistry California Practicum: Los Alamos National Laboratory Contact: [email protected] Practicum: Oak Ridge National Laboratory Massachusetts Institute of Technology Texas A&M University Advisor: Kieron Burke Contact: [email protected] Contact: [email protected] Contact: [email protected] Nuclear Waste Materials Nuclear Engineering Contact: [email protected] Jeffrey Donatelli Advisor: Linn Hobbs Advisor: Jim Morel Jason Bender Armen Kherlopian Claire Ralph University of California, Berkeley Hayes Stripling Practicum: Lawrence Berkeley Contact: [email protected] University of Minnesota Alexander Rattner Cornell University Cornell University Applied Mathematics Texas A&M University National Laboratory Hypersonic Computational Fluid Dynamics Georgia Institute of Technology Computational and Systems Biology Theoretical Chemistry Advisor: James Sethian Nuclear Engineering/Uncertainty Quantification Contact: [email protected] Devin Matthews Advisor: Graham Candler Mechanical Engineering Advisor: David Christini Advisor: Garnet Chan Practicum: Oak Ridge National Laboratory Advisor: Marvin Adams University of Texas Contact: [email protected] Advisor: Srinivas Garimella Practicum: Princeton Plasma Practicums: Two at Sandia National Contact: [email protected] Practicums: Lawrence Livermore Carmeline Dsilva Chemistry Contact: [email protected] Physics Laboratory Laboratories – New Mexico National Laboratory and Argonne Princeton University Advisor: John Stanton Rogelio Cardona-Rivera Contact: [email protected] Contact: [email protected] Virgil Griffith National Laboratory Chemical Engineering Practicum: Argonne National Laboratory North Carolina State University Phoebe Robinson California Institute of Technology Contact: [email protected] Advisor: Yannis Kevrekidis Contact: [email protected] Artificial Intelligence Harvard University Kathleen King Brenda Rubenstein Theoretical Neuroscience Contact: [email protected] Advisor: R. Michael Young Earth Science Cornell University Columbia University Advisor: Christof Koch Travis Trahan Scot Miller Contact: [email protected] Advisor: Brendan Meade Operations Research Theoretical Chemistry Practicum: Lawrence Berkeley University of Michigan Christopher Eldred Harvard University Contact: [email protected] Advisor: John Muckstadt Advisor: David Reichman National Laboratory Nuclear Engineering University of Utah Atmospheric Sciences Daniel Dandurand Practicum: Argonne National Laboratory Practicums: Los Alamos National Laboratory Contact: [email protected] Advisor: Edward Larsen Climate Modeling Advisor: Steven Wofsy University of California, Berkeley Michael Rosario Contact: [email protected] and Lawrence Livermore National Laboratory Practicums: Argonne National Advisor: Thomas Reichler Practicum: Lawrence Berkeley Astrophysics/Cosmology University of Massachusetts, Amherst Contact: [email protected] Tobin Isaac Laboratory and Lawrence Livermore Practicum: Lawrence Berkeley National Laboratory Advisor: Eliot Quataert Organismic and Evolutionary Biology Eric Liu University of Texas National Laboratory National Laboratory Contact: [email protected] Contact: [email protected] Advisor: Sheila Patek Massachusetts Institute of Technology Anne Warlaumont Computational and Applied Mathematics Contact: [email protected] Contact: [email protected] Contact: [email protected] Computational Fluid Mechanics University of Memphis Advisor: Omar Ghattas Kenley Pelzer Omar Hafez Advisor: David Darmofal Computational Developmental Practicum: Los Alamos National Laboratory Sean Vitousek Thomas Fai University of Chicago University of California, Davis Hansi Singh Practicum: Lawrence Berkeley Psycholinguistics Contact: [email protected] Stanford University New York University Theoretical Physical Chemistry Structural Mechanics University of Washington National Laboratory Advisor: David Kimbrough Oller Environmental Fluid Mechanics and Hydrology Applied Mathematics Advisor: David Mazziotti Advisor: Yannis Dafalias Geophysics Contact: [email protected] Practicum: Argonne National Laboratory Mark Maienschein-Cline Advisor: Oliver Fringer Advisor: Charles Peskin Contact: [email protected] Contact: [email protected] Advisor: Cecilia Bitz Contact: [email protected] University of Chicago Practicum: Lawrence Berkeley Practicum: Lawrence Berkeley Contact: [email protected] Brian Lockwood Physical Chemistry National Laboratory National Laboratory Amanda Peters Maxwell Hutchinson University of Wyoming Advisor: Aaron Dinner Contact: [email protected] Contact: [email protected] Harvard University University of Chicago Chris Smillie Fluid Dynamics Practicum: Los Alamos National Laboratory Applied Physics Physics Massachusetts Institute of Technology Advisor: Dimitri Mavriplis Contact: [email protected] Norman Yao Charles Frogner Advisor: Efthimios Kaxiras Contact: [email protected] Biology, Computer Science and Bioengineering Practicum: Argonne National Laboratory Harvard University Massachusetts Institute of Technology Practicum: Lawrence Livermore Advisor: Eric Alm Contact: [email protected] Condensed Matter Physics Computational Biology National Laboratory Curtis Lee Contact: [email protected] Advisor: Mikhail Lukin Advisor: Tomaso Poggio Contact: [email protected] Duke University Practicum: Los Alamos National Laboratory Practicum: Lawrence Berkeley Computational Mechanics Joshua Vermaas Contact: [email protected] National Laboratory Christopher Quinn Advisor: John Dolbow University of Illinois at Urbana-Champaign Contact: [email protected] University of Illinois at Urbana-Champaign Contact: [email protected] Biophysics Communications Advisor: Emad Tajkhorshid Evan Gawlik Advisor: Todd Coleman Sarah Loos Contact: [email protected] Stanford University Contact: [email protected] Carnegie Mellon University Applied Mathematics Verification of Hybrid Systems Matthew Zahr Advisor: Margot Gerritsen Aaron Sisto Advisor: Andre Platzer Stanford University Practicum: Argonne National Laboratory Purdue University Contact: [email protected] Computational and Mathematical Engineering Contact: [email protected] Computational Materials Science Advisor: Charbel Farhat Advisor: Xiulin Ruan Contact: [email protected] Contact: [email protected]

P36 DEIXIS 11 DOE CSGF ANNUAL DEIXIS 11 DOE CSGF ANNUAL P37 Alumni DIRECTORY

Matthew Adams Allison Baker Arnab Bhattacharyya Kent Carlson James Comer Tal Danino Amanda Duncan Matthew Farthing University of Washington University of Colorado Massachusetts Institute of Technology Florida State University North Carolina State University University of California, San Diego University of Illinois at Urbana-Champaign University of North Carolina Computational Electromagnetics Iterative Methods for Linear Systems, Parallel Computer Science Solidification of Cast Metals Computational Fluid Mechanics, Flow in Dynamics of Systems Biology Electrical Engineering Flow and Transport Phenomena in Fellowship Years: 2007-2008 Computing, Software for Scientific Computing Fellowship Years: 2006-2010 Fellowship Years: 1991-1995 Absorbent Structures Fellowship Years: 2006-2010 Fellowship Years: 1991-1995 Porous Media Fellowship Years: 1999-2003 Current Status: Graduate Student Current Status: Assistant Research Fellowship Years: 1991-1995 Current Status: Postdoctoral Researcher, Current Status: Senior Staff Engineer, Fellowship Years: 1997-2001 Joshua Adelman Current Status: Lawrence Livermore Engineer/Adjunct Assistant Professor, Current Status: Technology Leader, Boston University/Massachusetts Intel Corporation Current Status: Research Hydraulic University of California, Berkeley National Laboratory Mary Biddy University of Iowa Modeling and Simulation, Family Care, Institute of Technology Engineer, Coastal and Hydraulics Biophysics University of Wisconsin Procter & Gamble Mary Dunlop Laboratory, U.S. Army Corps of Fellowship Years: 2005-2009 Devin Balkcom Engineering Nathan Carstens William Daughton California Institute of Technology Engineers Engineer Research and Current Status: Postdoctoral Fellow, Carnegie Mellon University Fellowship Years: 2002-2006 Massachusetts Institute of Technology Gavin Conant Massachusetts Institute of Technology Bioengineering, Synthetic Biology, Biofuels, Development Center University of Pittsburgh Robotics Current Status: British Petroleum Computational Simulation of BWR Fuel University of New Mexico Plasma Physics Dynamical Systems Fellowship Years: 2000-2004 Fellowship Years: 2001-2004 Molecular Evolution/Bioinformatics Fellowship Years: 1992-1996 Fellowship Years: 2002-2006 Michael Feldmann Zlatan Aksamija Current Status: Faculty, Dartmouth College Edwin Blosch Current Status: AREVA Fellowship Years: 2000-2004 Current Status: Staff Scientist, Current Status: Assistant Professor, California Institute of Technology University of Illinois at Urbana-Champaign University of Florida Current Status: Assistant Professor, Los Alamos National Laboratory University of Vermont Computational Finance Nanostructured Semiconductor Thermoelectrics Michael Barad Aerospace Engineering Edward Chao Animal Sciences and Informatics, Fellowship Years: 1999-2002 Fellowship Years: 2005-2009 University of California, Davis Fellowship Years: 1991-1994 Princeton University University of Missouri, Columbia Gregory Davidson Lewis Dursi Current Status: Executive Vice President, Current Status: Computing Innovation Computational Fluid Dynamics Current Status: Technical Lead, Computed Tomography/Radiation Therapy University of Michigan University of Chicago Quantitative Research, Walleye Trading Postdoctoral Fellowship, NTG Group, Fellowship Years: 2002-2006 CFD-FASTRAN Fellowship Years: 1992-1995 William Conley Nuclear Engineering and Computational Astrophysics, Large-Scale Software LLC University of Wisconsin-Madison Current Status: NASA Ames Research Center Current Status: Scientist, TomoTherapy Purdue University Computational Science Simulation, Hydrodynamics, Combustion, Nawaf Bou-Rabee Nonlinear Mechanics of Fellowship Years: 2002-2006 Magnetohydrodynamics Krzysztof Fidkowski Bree Aldridge Jaydeep Bardhan California Institute of Technology Jarrod Chapman Nano-Mechanical Systems Current Status: Oak Ridge Fellowship Years: 1999-2003 Massachusetts Institute of Technology Massachusetts Institute of Technology Massachusetts Institute of Technology Monte Carlo Methods, Numerical Solution of University of California, Berkeley Fellowship Years: 2003-2008 National Laboratory Current Status: Senior Research Computational Fluid Dynamics Systems Biology/Computational Biology Numerical Methods for Molecular Analysis Stochastic Differential Equations, Whole Genome Shotgun Assembly, Current Status: Engineer, Crane Associate, CITA Fellowship Years: 2003-2007 Fellowship Years: 2002-2006 and Design Molecular Dynamics Computational Genomics Naval Warfare Center Jimena Davis Current Status: Assistant Professor, Current Status: Postdoctoral Research Fellowship Years: 2002-2006 Fellowship Years: 2002-2006 Fellowship Years: 1999-2003 North Carolina State University Ryan Elliott Aerospace Engineering, University Fellow, Harvard School of Public Health Current Status: Assistant Professor, Current Status: Courant Instructor, Current Status: Postdoctoral Fellow, Natalie Cookson Uncertainty Quantification, Physiologically University of Michigan of Michigan Molecular Biophysics and Physiology, New York University Computational Genomics Program, University of California, San Diego Based Pharmacokinetic Modeling, Shape Memory Alloys and Active Materials Erik Allen Rush University DOE Joint Genome Institute Systems Biodynamics and Risk Assessment Fellowship Years: 2000-2004 Piotr Fidkowski Massachusetts Institute of Technology Jenelle Bray Computational Biology Fellowship Years: 2004-2008 Current Status: Assistant Professor, Massachusetts Institute of Technology Molecular Simulation Edward Barragy California Institute of Technology Eric Charlton Fellowship Years: 2005-2009 Current Status: Postdoctoral Fellow, Department of Aerospace Engineering Structural/Computational Engineering Fellowship Years: 2004-2008 University of Texas Computational Structural Biology University of Michigan Current Status: Postdoctoral Researcher, U.S. Environmental Protection Agency and Mechanics, University of Minnesota Fellowship Years: 2009-2011 Engineering Mechanics Fellowship Years: 2006-2009 Aerodynamics and Computational University of California, San Diego Current Status: Software Engineer, Marcelo Alvarez Fellowship Years: 1991-1993 Fluid Dynamics Jack Deslippe Thomas Epperly Google Inc. University of Texas Current Status: Intel Corporation J. Dean Brederson Fellowship Years: 1992-1996 Ethan Coon University of California, Berkeley University of Wisconsin Astrophysics University of Utah Current Status: Lockheed Martin Columbia University Computational Condensed Matter Theory Component Technology for Stephen Fink Fellowship Years: 2001-2005 William Barry Synergistic Data Display Aeronautics Company Computational Geophysics Fellowship Years: 2006-2010 High-Performance Computing University of California, San Diego Current Status: Postdoctoral Researcher, Carnegie Mellon University Fellowship Year: 1996 Fellowship Years: 2005-2009 Current Status: Graduate Student Fellowship Years: 1991-1995 Computer Science Kavli Institute for Particle Astrophysics Computational Mechanics, Michael Chiu Current Status: Postdoctoral Researcher, Current Status: Lawrence Livermore Fellowship Years: 1994-1998 and Cosmology, Stanford University Engineering Education Paul Bunch Massachusetts Institute of Technology Los Alamos National Laboratory Mark DiBattista National Laboratory Current Status: IBM Fellowship Years: 1994-1998 Purdue University Mechanical Engineering Columbia University Asohan Amarasingham Current Status: Associate Professor, Chemical Engineering Fellowship Years: 1992-1996 John Costello Computational Fluid Dynamics Susanne Essig Robert Fischer Brown University Trine University Fellowship Years: 1994-1997 Current Status: Engineering Manager, University of Arizona Fellowship Years: 1992-1994 Massachusetts Institute of Technology Harvard University Theoretical Neuroscience Current Status: Merck & Co. Inc. Teradyne Applied Mathematics Aeronautics and Astronautics, Security, Privacy, Mobile Agents, Fellowship Years: 1998-2002 Paul Bauman Fellowship Years: 1998-2002 John Dolbow Computational Turbulence Software Engineering Current Status: Postdoctoral Fellow, University of Texas Jeffrey Butera Kevin Chu Current Status: Software Engineer, Microsoft Northwestern University Fellowship Years: 1997-2002 Fellowship Years: 1994-1998 Center for Molecular and Behavioral Multiscale Modeling, Error Estimation, North Carolina State University Massachusetts Institute of Technology Computational Methods for Evolving Current Status: Quant Neuroscience, Rutgers University Automatic Adaptivity, Validation, Mathematics Computer Science and Engineering, Nathan Crane Discontinuities and Interfaces Annette Evangelisti Uncertainty Quantification Fellowship Years: 1993-1997 Applied Math, Artificial Intelligence, University of Illinois at Urbana-Champaign Fellowship Years: 1997-1999 University of New Mexico Jasmine Foo Kristopher Andersen Fellowship Years: 2003-2007 Current Status: Administrative Computing, High-Performance Computing Computational Mechanics Current Status: Yoh Family Professor, Computational Molecular Biology Brown University University of California, Davis Current Status: Research Associate, Hampshire College Fellowship Years: 2002-2005 Fellowship Years: 1999-2002 Duke University Fellowship Years: 2001-2005 Applied Mathematics Computational Materials Science University of Texas at Austin Current Status: Research Scientist/ Current Status: Sandia National Current Status: Postdoctoral Researcher, Fellowship Years: 2004-2008 Fellowship Years: 2001-2005 Michael Bybee Consultant, Serendipity Research; CEO, Laboratories – New Mexico Laura Dominick University of New Mexico Current Status: Postdoctoral Fellow, Current Status: HPTi Martin Bazant University of Illinois at Urbana-Champaign Velexi Corporation Florida Atlantic University Harvard University Harvard University Hydrodynamic Simulation of Colloidal Stephen Cronen-Townsend Computational Electromagnetics/ John Evans Matthew Anderson Applied Mathematics, Fluid Mechanics, Suspensions Julianne Chung Cornell University Electromagnetic Performance of Materials University of Texas Ashlee Ford University of Texas Electrochemical Systems Fellowship Years: 2004-2008 Emory University Computational Materials Physics Fellowship Years: 1993-1997 Computational and Applied Mathematics University of Illinois at Urbana-Champaign Numerical Relativity, Magnetohydrodynamics, Fellowship Years: 1992-1996 Current Status: Senior Engineer, Computational Science, Applied Mathematics, Fellowship Years: 1991-1995 Current Status: Large Military Engines Fellowship Years: 2006-2010 Modeling of Drug Delivery, Numerical Methods Relativistic High-Performance Computing Current Status: Associate Professor, Gamma Technologies Inc. Biomedical Imaging Current Status: Drupal Developer, Division, Pratt & Whitney Current Status: Graduate Student for Partial Differential Equations Fellowship Years: 2000-2004 Department of Mechanical Engineering, Fellowship Years: 2006-2009 Cronen-Townsend Consulting Fellowship Years: 2006-2010 Current Status: Postdoctoral Researcher, Stanford University Brandoch Calef Current Status: NSF Mathematical Michael Driscoll Matt Fago Current Status: Graduate Student Louisiana State University University of California, Berkeley Sciences Postdoctoral Research Fellow, Robert Cruise Boston University California Institute of Technology Mark Berrill Imaging Research University of Maryland Indiana University Bioinformatics and Systems Biology Computational Structural Mechanics Gregory Ford Jordan Atlas Colorado State University Fellowship Years: 1996-2000 Computational Physics Fellowship Years: 2002-2006 Fellowship Years: 2000-2003 University of Illinois at Urbana-Champaign Cornell University Computational Engineering and Current Status: Scientist, Boeing Kristine Cochran Fellowship Years: 1997-2001 Current Status: Dataspora Inc. Current Status: Research Scientist Chemical Engineering Computational Biology Energy Sciences University of Illinois at Urbana-Champaign Current Status: Department of Defense Fellowship Year: 1993 Fellowship Years: 2005-2009 Fellowship Years: 2006-2010 Patrick Canupp Computational Mechanics, Material Modeling, Jeffrey Drocco Michael Falk Current Status: Software Development Current Status: Postdoctoral Researcher, Stanford University Fracture Mechanics Aron Cummings Princeton University University of California, Santa Barbara Robin Friedman Engineer, Microsoft Eugene P. Wigner Fellowship, Oak Ridge Aeronautical and Astronautical Engineering Fellowship Years: 2002-2006 Arizona State University Biophysics and Computation Stress-Driven Materials Processes Including Massachusetts Institute of Technology National Laboratory Fellowship Years: 1991-1995 Current Status: Engineering Researcher, Nanoscale Electronics Fellowship Years: 2004-2008 Fracture, Deformation and Semiconductor Computational and Systems Biology Teresa Bailey Current Status: Chief Aerodynamicist, Oak Ridge National Laboratory Fellowship Years: 2003-2007 Current Status: Postdoctoral Researcher, Crystal Growth Fellowship Years: 2007-2010 Texas A&M University Kathleen Beutel Joe Gibbs Racing Current Status: Postdoctoral Researcher, Los Alamos National Laboratory Fellowship Years: 1995-1998 Current Status: Postdoctoral Fellow, Deterministic Transport Theory University of Minnesota Joshua Coe Sandia National Laboratories – California Current Status: Associate Professor, Institut Pasteur Fellowship Years: 2002-2006 Computational Chemistry Christopher Carey University of Illinois at Urbana-Champaign Brian Dumont Materials Science and Engineering, Current Status: Lawrence Livermore Fellowship Years: 2009-2010 University of Wisconsin Chemical Physics, Electronically Excited Joseph Czyzyk University of Michigan Johns Hopkins University Oliver Fringer National Laboratory Plasma Physics States, Monte Carlo Methodology Northwestern University Aerospace Engineering Stanford University Bonnie Beyer Fellowship Years: 2005-2009 Fellowship Years: 2001-2002 Industrial Engineering and Management Fellowship Year: 1994 Parallel Coastal Ocean Modeling University of Illinois at Urbana-Champaign Current Status: Scientific Staff, MIT Current Status: Los Alamos Fellowship Years: 1991-1994 Current Status: Airflow Sciences Corporation Fellowship Years: 1997-2001 Avionics for Business and Regional Aircraft Lincoln Laboratory National Laboratory Current Status: Business Intelligence Current Status: Assistant Professor, Fellowship Years: 1991-1995 Analyst, Central Michigan University Stanford University Current Status: Technical Project Manager, Research Corporation Rockwell Collins

P38 DEIXIS 11 DOE CSGF ANNUAL DEIXIS 11 DOE CSGF ANNUAL P39 Kenneth Gage Kristen Grauman Jeff Hammond Jeffrey Hittinger Yan Karklin Justin Koo Alex Lindblad Lisa Mesaros University of Pittsburgh Massachusetts Institute of Technology University of Chicago University of Michigan Carnegie Mellon University University of Michigan University of Washington University of Michigan Molecular Imaging, Computational Fluid Computer Vision, Machine Learning Supercomputing, Computational Chemistry, Computational Plasma Physics Computational Neuroscience Electric Propulsion Modeling and Simulation Computational Solid Mechanics Aerospace Engineering and Dynamics Design of Artificial Organs Fellowship Years: 2001-2005 Programming Models, Verification Fellowship Years: 1996-2000 Fellowship Years: 2002-2006 Fellowship Years: 2000-2004 Fellowship Years: 2002-2006 Scientific Computing Fellowship Years: 1998-2002 Current Status: Clare Boothe Luce and Validation Current Status: Staff Member, Center for Current Status: Postdoctoral Researcher, Current Status: Engineer, Air Force Current Status: Senior Member of Fellowship Years: 1991-1995 Current Status: Radiology Resident Assistant Professor, Department of Fellowship Years: 2005-2009 Applied Scientific Computing, Lawrence Center for Neural Science/Howard Hughes Research Laboratory, Edwards Air Technical Staff, Sandia National Current Status: Business Manager, (Research Track), Johns Hopkins Computer Science, University of Texas Current Status: Director’s Postdoctoral Livermore National Laboratory Medical Institute, New York University Force Base Laboratories – California Fluent Inc. Medical Institutions at Austin Fellow, Argonne Leadership Computing Facility Gordon Hogenson Richard Katz Michael Kowalok Tasha Lopez Richard Mills Nouvelle Gebhart Corey Graves University of Washington Columbia University University of Wisconsin University of California, Los Angeles College of William and Mary University of New Mexico North Carolina State University Jeff Haney Physical Chemistry Geodynamics, Coupled Fluid-Solid Dynamics Monte Carlo Methods for Radiation Therapy Chemical Engineering Scientific Computing Chemistry Pervasive Computing/Image Processing Texas A&M University Fellowship Years: 1993-1996 Fellowship Years: 2001-2005 Treatment Planning Fellowship Years: 2000-2001 Fellowship Years: 2001-2004 Fellowship Years: 2001-2003 Fellowship Years: 1996-2000 Physical Oceanography Current Status: Technical Writer, Microsoft Current Status: Academic Fellow, Fellowship Years: 2000-2004 Current Status: Sales Specialist, Current Status: Computational Scientist, Current Status: Deceased Current Status: Business Owner, Scholars’ Fellowship Years: 1993-1996 Department of Earth Science, University Current Status: Medical Physicist, IBM Cognos Business Analytics Oak Ridge National Laboratory Advocate; Assistant Professor, North Current Status: IT Manager, Dynacon Inc. Daniel Horner of Oxford Waukesha Memorial Hospital Sommer Gentry Carolina A&T State University University of California, Berkeley Christie Lundy Julian Mintseris Massachusetts Institute of Technology Heath Hanshaw Breakup Processes, Quantum Benjamin Keen Yury Krongauz Missouri University of Science and Technology Boston University Optimization Michael Greminger University of Michigan Molecular Dynamics University of Michigan Northwestern University Physics Computational Biology Fellowship Years: 2001-2005 University of Minnesota High Energy Density Physics Fellowship Years: 2000-2004 Conservation Laws in Complex Geometries Theoretical and Applied Mechanics Fellowship Years: 1991-1994 Fellowship Years: 2001-2005 Current Status: Assistant Professor, Mechanical Engineering Fellowship Years: 2001-2005 Current Status: Research Analyst, Fellowship Years: 2000-2004 Fellowship Years: 1993-1996 Current Status: Missouri State Government Current Status: Postdoctoral Fellow, Mathematics, U.S. Naval Academy Fellowship Years: 2002-2005 Current Status: Sandia National Advanced Technology and Systems Current Status: IDA Center for Current Status: BlackRock Harvard Medical School Current Status: Seagate Technology Laboratories – New Mexico Analysis Division, Center for Computing Sciences William Marganski Charles Gerlach Naval Analysis Eric Lee Boston University Erik Monsen Northwestern University Noel Gres Rellen Hardtke Peter Kekenes-Huskey Rutgers University Computational Biology, Imaging, Modeling Stanford University Finite Elements, High Strain Rate University of Illinois at Urbana-Champaign University of Wisconsin William Humphrey California Institute of Technology Mechanical Engineering Fellowship Years: 1998-2002 Entrepreneurship, Organization Development Solid Mechanics Electrical Engineering Particle Astrophysics (Neutrinos from University of Illinois at Urbana-Champaign Computational Chemistry and Biology Fellowship Years: 1999-2003 Current Status: Research Scientist, and Change Fellowship Years: 1995-1999 Fellowship Years: 1999-2001 Gamma-Ray Bursts), Gender and Science Physics Fellowship Years: 2004-2007 Current Status: Engineer, Northrop Systems Biology Department, Harvard Fellowship Years: 1991-1993 Current Status: Southwest Fellowship Years: 1998-2002 Fellowship Years: 1992-1994 Current Status: Postdoctoral Scholar, Grumman Corporation Medical School Current Status: Senior Research Fellow, Research Institute Boyce Griffith Current Status: Associate Professor, Current Status: NumeriX LLC McCammon Group, University of Max Planck Institute of Economics New York University University of Wisconsin-River Falls California, San Diego Miler Lee David Markowitz (Jena, Germany) Timothy Germann Mathematical Modeling and Computer Jason Hunt University of Pennsylvania Princeton University Harvard University Simulation in Cardiac and Kristi Harris University of Michigan Jeremy Kepner Computational Biology, Developmental Genetics Computational Neurobiology Brian Moore Physical Chemistry Cardiovascular Physiology University of Maryland, Baltimore County Aerospace Engineering and Princeton University Fellowship Years: 2005-2009 Fellowship Years: 2005-2009 North Carolina State University Fellowship Years: 1992-1995 Fellowship Years: 2000-2004 Theoretical Solid State Physics Scientific Computing High-Performance Embedded Computing Current Status: Postdoctoral Associate, Computational Simulation of Nuclear and Current Status: Staff Member, Los Alamos Current Status: Assistant Professor of Fellowship Years: 2006-2010 Fellowship Years: 1999-2003 Fellowship Years: 1993-1996 Yale University Daniel Martin Thermal-Hydraulic Processes in Boiling National Laboratory Medicine, Leon H. Charney Division of Current Status: Analyst, Department Current Status: General Dynamics – Current Status: Senior Technical Staff, University of California, Berkeley Water Nuclear Reactors Cardiology, New York University School of Defense Advanced Information Systems MIT Lincoln Laboratory Seung Lee Adaptive Mesh Refinement Algorithm Fellowship Years: 1992-1995 Christopher Gesh of Medicine Massachusetts Institute of Technology and Software Development Current Status: Leader, Methods Texas A&M University Owen Hehmeyer E. McKay Hyde David Ketcheson Computational Molecular Biology Fellowship Years: 1993-1996 and Software Development Center Computational Transport Theory, Nuclear Eric Grimme Princeton University California Institute of Technology University of Washington Fellowship Years: 2001-2005 Current Status: Lawrence Berkeley of Excellence, GE Hitachi Nuclear Energy Reactor Analysis, Nuclear Non-Proliferation University of Illinois at Urbana-Champaign Chemical Engineering Efficient, High-Order Integral Equation Applied Mathematics: Numerical Analysis Current Status: Management Consultant, National Laboratory Fellowship Years: 1993-1997 Electrical Engineering Fellowship Years: 2002-2006 Methods in Computational and Scientific Computing Boston Consulting Group (Seoul Office) Nathaniel Morgan Current Status: Pacific Northwest Fellowship Years: 1994-1997 Current Status: ExxonMobil Upstream Electromagnetics and Acoustics Fellowship Years: 2006-2009 Marcus Martin Georgia Institute of Technology National Laboratory Current Status: Intel Corporation Research Corporation Fellowship Years: 1999-2002 Current Status: Assistant Professor, Jack Lemmon University of Minnesota Computational Fluid Dynamics Current Status: Assistant Professor, King Abdullah University of Science Georgia Institute of Technology Monte Carlo Molecular Simulation Fellowship Years: 2002-2005 Matthew Giamporcaro John Guidi Eric Held Computational and Applied Mathematics, and Technology Mechanical Engineering (Algorithm Development Focus) Current Status: Los Alamos Boston University University of Maryland University of Wisconsin Rice University Fellowship Years: 1991-1994 Fellowship Years: 1997-1999 National Laboratory Adaptive Algorithms, Artificial Neural Networks Computer Science Plasma/Fusion Theory Sven Khatri Current Status: Medtronic Inc. Current Status: Director, Useful Bias Inc. Fellowship Years: 1998-2000 Fellowship Years: 1994-1997 Fellowship Years: 1995-1999 Eugene Ingerman California Institute of Technology James Morrow Current Status: Engineering Consultant, Current Status: High School Math Teacher Current Status: Associate Professor, University of California, Berkeley Electrical Engineering Mary Ann Leung Randall McDermott Carnegie Mellon University GCI Inc. Physics Department, Utah Applied Mathematics/Numerical Methods Fellowship Years: 1993-1996 University of Washington University of Utah Sensor-Based Control of Robotic Systems Brian Gunney State University Fellowship Years: 1997-2001 Current Status: Honeywell Contractor Computational Physical Chemistry Numerical Methods for Large-Eddy Simulation Fellowship Years: 1992-1995 Ahna Girshick University of Michigan Current Status: Senior Scientist, Fellowship Years: 2001-2005 of Turbulent Reacting Flows Current Status: Principal Member of University of California, Berkeley Computational Fluid Dynamics, Multi-Physics Asegun Henry General Electric Jeffrey Kilpatrick Current Status: Program Manager, Fellowship Years: 2001-2005 Technical Staff, Sandia National Computational Models of Vision Simulations, Adaptive Mesh Refinement, Massachusetts Institute of Technology Rice University Krell Institute Current Status: Staff Scientist, Laboratories – New Mexico and Perception Parallel Computing Renewable Energy, Atomistic Level Heat Ahmed Ismail Computer Science National Institute of Standards and Fellowship Years: 2001-2005 Fellowship Years: 1993-1996 Transfer, First Principles Electronic Massachusetts Institute of Technology Fellowship Years: 2008-2010 Brian Levine Technology (NIST) Sarah Moussa Current Status: Postdoctoral Researcher, Current Status: Lawrence Livermore Structure Calculations Molecular Simulations and Current Status: Software Development Cornell University University of California, Berkeley University of California, Berkeley National Laboratory Fellowship Years: 2005-2009 Multiscale Modeling Engineer, Microsoft Transport Systems Matthew McGrath Machine Learning and Genomics Current Status: Fellow, Advanced Fellowship Years: 2000-2004 Fellowship Years: 2006-2010 University of Minnesota Fellowship Years: 2003-2005 Kevin Glass Aric Hagberg Research Projects Agency – Energy, Current Status: Junior Professor, Benjamin Kirk Current Status: Graduate Student Computational Aerosol Physics Current Status: Senior Software University of Oregon University of Arizona U.S. Department of Energy Mechanical Engineering, RWTH University of Texas Fellowship Years: 2004-2007 Engineer, Google Inc. Computational Ecology Applied Mathematics Aachen University Aerospace Engineering Jeremy Lewi Current Status: University Researcher, Fellowship Years: 1996-2000 Fellowship Years: 1992-1994 Judith Hill Fellowship Years: 2001-2004 Georgia Institute of Technology University of Helsinki (Finland) Michael Mysinger Current Status: Scientist, Molecular Current Status: Staff Member, Los Alamos Carnegie Mellon University Amber Jackson Current Status: NASA Johnson Neuroengineering Stanford University Science Computing Facility, National Laboratory Computational Fluid Dynamics, University of North Carolina Space Center Fellowship Years: 2005-2009 Richard McLaughlin Molecular Docking Solvation Models and Environmental Molecular Sciences Partial Differential Equation- Applied Mathematics Current Status: Engineering Princeton University G Protein-Coupled Receptor Docking Laboratory, Pacific Northwest Glenn Hammond Constrained Optimization Fellowship Years: 2004-2008 Bonnie Kirkpatrick Scientist, Intellisis Fluid Dynamics Fellowship Years: 1996-2000 National Laboratory University of Illinois at Urbana-Champaign Fellowship Years: 1999-2003 Current Status: Graduate Student University of California, Berkeley Fellowship Years: 1991-1994 Current Status: University of California, Multiphase Flow and Multicomponent Current Status: Computational Computer Science Benjamin Lewis Current Status: Professor of Mathematics, San Francisco Larisa Goldmints Biogeochemical Transport, Mathematics, Oak Ridge Nickolas Jovanovic Fellowship Years: 2004-2008 Massachusetts Institute of Technology University of North Carolina at Carnegie Mellon University Parallel Computation National Laboratory Yale University Current Status: Graduate Student Computational Biology Chapel Hill Heather Netzloff Structural and Computational Mechanics Fellowship Years: 1999-2003 Preconditioned Iterative Solution Techniques Fellowship Years: 2002-2006 Iowa State University Fellowship Years: 1997-2001 Current Status: Scientist III, Pacific Charles Hindman in Boundary Element Analysis Kevin Kohlstedt Matthew McNenly Quantum/Theoretical/Computational Chemistry Current Status: General Electric; Northwest National Laboratory University of Colorado Fellowship Years: 1992-1994 Northwestern University Lars Liden University of Michigan Fellowship Years: 2000-2004 Rensselaer Polytechnic Institute Aerospace Engineering Current Status: Founding Associate Coulomb Interactions in Soft Materials Boston University Rarefied Gas Dynamics Current Status: Community College Math Fellowship Years: 1999-2003 Professor of Systems Engineering, Fellowship Years: 2005-2009 Educational Tools for Special Needs Children Fellowship Years: 2001-2005 Instructor and Tutor William Gooding Current Status: Air Force Research University of Arkansas at Little Rock Current Status: Research Fellow, Chemical Fellowship Years: 1994-1998 Current Status: Staff, Lawrence Livermore Purdue University Laboratory, Space Vehicles Directorate Engineering, University of Michigan Current Status: Chief Technical Officer, National Laboratory Chemical Engineering TeachTown LLC; Software Technology Fellowship Years: 1991-1994 Manager, University of Washington

P40 DEIXIS 11 DOE CSGF ANNUAL DEIXIS 11 DOE CSGF ANNUAL P41 Elijah Newren Joel Parriott Alejandro Quezada Robin Rosenfeld Amoolya Singh Shilpa Talwar Joshua Waterfall Michael Wolf University of Utah University of Michigan University of California, Berkeley Scripps Research Institute University of California, Berkeley Stanford University Cornell University University of Illinois at Urbana-Champaign Computational Biofluid Dynamics Elliptical Galaxies, Computational Fluid Geophysics Computational Biophysics Dynamics and Evolution of Stress Array Signal Processing Molecular Biology Computer Science (Parallel and Combinatorial Fellowship Years: 2001-2005 Dynamics, Parallel Computing Fellowship Year: 1997 Fellowship Years: 1996-1997 Response Networks Fellowship Years: 1992-1994 Fellowship Years: 2002-2006 Scientific Computing) Current Status: Staff Member, Sandia Fellowship Years: 1992-1996 Current Status: ActiveSight Fellowship Years: 2002-2006 Current Status: Senior Research Scientist, Current Status: Postdoctoral Researcher, Fellowship Years: 2003-2007 National Laboratories – New Mexico Current Status: Program Examiner, Office Catherine Quist Current Status: Scientist, Amyris Intel Corporation Cornell University Current Status: Postdoctoral Researcher, of Management and Budget, Executive Cornell University Mark Rudner Biotechnologies Sandia National Laboratories – New Pauline Ng Office of the President Bioinformatics Massachusetts Institute of Technology Brian Taylor Philip Weeber Mexico University of Washington Fellowship Years: 2000-2004 Theoretical Condensed Matter Physics Melinda Sirman University of Illinois at Urbana-Champaign University of North Carolina Computational Biology Ian Parrish Current Status: Postdoctoral Fellow, Fellowship Years: 2003-2007 University of Texas Detonation, Shock Waves, Reacting Flow Interest Rate Derivative Consulting Matthew Wolinsky Fellowship Years: 2000-2002 Princeton University University of Michigan Cancer Center Current Status: Postdoctoral Fellow Engineering Mechanics Fellowship Years: 2003-2007 Fellowship Years: 1994-1996 Duke University Current Status: Group Leader, Genome Computational Astrophysics Fellowship Years: 1994-1996 Current Status: National Research Current Status: Chatham Financial Computational Geoscience Institute of Singapore Fellowship Years: 2004-2007 Mala Radhakrishnan Ariella Sasson Current Status: At Home Council Postdoctoral Researcher, Naval Fellowship Years: 2001-2005 Current Status: Einstein/Chandra Massachusetts Institute of Technology Rutgers University Research Laboratory Adam Weller Current Status: Research Scientist, Shell Diem-Phuong Nguyen Postdoctoral Fellow, University of Computational Drug and Biomolecular Design Computational Biology and Benjamin Smith Princeton University International Exploration and Production University of Utah California, Berkeley and Analysis Molecular Biophysics Harvard University Mayya Tokman Chemical Engineering Computational Fluid Dynamics (Combustion Fellowship Years: 2004-2007 Fellowship Years: 2006-2010 Cloud and Mobile Computing California Institute of Technology Fellowship Years: 2001-2002 Allan Wollaber and Reaction) Tod Pascal Current Status: Assistant Professor Current Status: Bioinformatics Specialist, Fellowship Years: 2006-2010 Numerical Methods, Scientific Computing University of Michigan Fellowship Years: 1999-2003 California Institute of Technology of Chemistry, Wellesley College Children’s Hospital of Philadelphia Current Status: Software Engineer Fellowship Years: 1996-2000 Gregory Whiffen Nuclear Engineering Current Status: Staff, University of Utah Physical Chemistry Current Status: Assistant Professor, Cornell University Deep Space Trajectory and Fellowship Years: 2004-2008 Fellowship Years: 2003-2007 Emma Rainey David Schmidt Steven Smith University of California, Merced Mission Design, Low-Thrust Mission Current Status: Los Alamos Debra Nielsen California Institute of Technology University of Illinois at Urbana-Champaign North Carolina State University Design, Nonlinear Optimal Control National Laboratory Colorado State University Virginia Pasour Planetary Sciences Communications Chemical Engineering William Triffo Fellowship Years: 1991-1995 Civil Engineering North Carolina State University Fellowship Years: 2003-2006 Fellowship Years: 2002-2006 Fellowship Years: 1992-1994 Rice University Current Status: Senior Engineer, Outer Brandon Wood Fellowship Years: 1992-1996 Physical/Biological Modeling, Modeling Current Status: Arete Associates Current Status: Epic Systems Current Status: Invista Biophysical Imaging, 3-D Electron Microscopy Planets Mission Design Group, NASA Jet Massachusetts Institute of Technology of Epidemiological Dynamics Fellowship Years: 2003-2007 Propulsion Laboratory Computational Materials Science Oaz Nir Fellowship Years: 1998-1999 Nathan Rau Samuel Schofield Benjamin Sonday Current Status: Graduate Student Fellowship Years: 2003-2007 Massachusetts Institute of Technology Current Status: Program Manager, University of Illinois at Urbana-Champaign University of Arizona Princeton University Collin Wick Current Status: Postdoctoral Fellow, Computational Biology Biomathematics, Army Research Office Civil Engineering Computational Fluid Dynamics, Hydrodynamic Dimensionality Reduction/Computational Mario Trujillo University of Minnesota Lawrence Livermore National Laboratory Fellowship Years: 2006-2009 Fellowship Years: 2000-2001 Stability, Interface Methods Nonlinear Dynamics University of Illinois at Urbana-Champaign Computational Chemistry Christina Payne Current Status: Civil Engineer, Hanson Fellowship Years: 2001-2005 Fellowship Years: 2006-2010 Two-Phase Flow, Computational Fluid Fellowship Years: 2000-2003 Lee Worden Joyce Noah-Vanhoucke Vanderbilt University Professional Services Current Status: Scientist, T-5, Los Alamos Current Status: Goldman Sachs Mechanics, Atomization Phenomena Current Status: Assistant Professor, Princeton University Stanford University Molecular Dynamics Simulations National Laboratory Fellowship Years: 1997-2000 Louisiana Tech University Applied Mathematics Fellowship Years: 2001-2003 Fellowship Years: 2003-2007 Clifton Richardson Eric Sorin Current Status: Assistant Professor, Fellowship Years: 1998-2002 Current Status: Scientist, Archimedes Current Status: Postdoctoral Researcher, Cornell University Christopher Schroeder Stanford University Mechanical Engineering, University of James Wiggs Current Status: S. V. Ciriacy-Wantrup National Renewable Energy Laboratory Physics University of California, San Diego Simulational Studies of Biomolecular Wisconsin-Madison University of Washington Postdoctoral Fellow, Environmental Peter Norgaard Fellowship Years: 1991-1995 Theoretical Particle Physics, Lattice Assembly and Conformational Dynamics Physical Chemistry Studies, Policy and Management, Princeton University Chris Penland Gauge Theory Fellowship Years: 2002-2004 Obioma Uche Fellowship Years: 1991-1994 University of California, Berkeley Computational Plasma Dynamics Duke University Christopher Rinderspacher Fellowship Years: 2005-2009 Current Status: Assistant Professor, Princeton University Current Status: Novum Millennium Fellowship Years: 2005-2009 Computational and Statistical Modeling of University of Georgia Current Status: Postdoctoral Researcher Computational and Physical Chemistry, Molecular Simulation, Statistical Mechanics Michael Wu Current Status: Ph.D. (ABD) Candidate, Pharmacokinetic/Pharmacodynamic Inverse Design, Quantum Chemistry California State University, Long Beach Fellowship Years: 2002-2006 Stefan Wild University of California, Berkeley Princeton University Systems for Biopharma Fellowship Years: 2001-2005 Robert Sedgewick Current Status: Research Associate, Cornell University Social Analytics, Graph and Social Network Fellowship Years: 1993-1997 Current Status: Army Research Laboratory University of California, Santa Barbara Scott Stanley University of Virginia, Charlottesville Operations Research Analysis, Predictive Modeling, High Dim Catherine Norman Current Status: Expert Modeler, Computational Biology University of California, San Diego Fellowship Years: 2005-2008 Data Visualization Northwestern University Pharmacometrics – Modeling and John Rittner Fellowship Years: 2000-2003 Large Scale Data Analysis, Search Engine Anton Van der Ven Current Status: Assistant Computational Fellowship Years: 2002-2006 Computational Fluid Dynamics Simulation, Novartis Institutes for Northwestern University Current Status: Research Associate, Technology, Fluid Mechanics, Massachusetts Institute of Technology Mathematician, Mathematics and Current Status: Principal Scientist of Fellowship Years: 2000-2004 Biomedical Research Grain Boundary Segregation Carnegie Mellon University Turbulence Modeling First Principles Modeling of Thermodynamic Computer Science Division, Argonne Analytics, Lithium Technologies Current Status: Research Analyst, Center Fellowship Years: 1991-1995 Fellowship Year: 1994 and Kinetic Properties of Solids National Laboratory for Naval Analyses Carolyn Phillips Current Status: Chicago Board Michael Sekora Current Status: Vice President of Fellowship Years: 1996-2000 Pete Wyckoff University of Michigan Options Exchange Princeton University Engineering, Buyful Current Status: Assistant Professor, Jon Wilkening Massachusetts Institute of Technology Gregory Novak Applied Physics Numerical Analysis, Godunov Methods, Department of Materials Science, University of California, Berkeley Parallel Architectures and Distributed Networks University of California, Santa Cruz Fellowship Years: 2006-2010 Courtney Roby Multiscale Algorithms, Asymptotic Samuel Stechmann University of Michigan Numerical Analysis, Computational Fellowship Years: 1992-1995 Theoretical Astrophysics Current Status: Graduate Student University of Colorado Preserving Methods New York University Physics, Partial Differential Equations, Current Status: Research Scientist, Fellowship Years: 2002-2006 History of Science in the Ancient World Fellowship Years: 2006-2010 Applied Math, Atmospheric Science Michael Veilleux Scientific Computing Ohio Supercomputer Center Current Status: Postdoctoral Fellow, James Phillips Fellowship Years: 2002-2003 Current Status: Laurion Capital Fellowship Years: 2003-2007 Cornell University Fellowship Years: 1997-2001 Princeton University University of Illinois at Urbana-Champaign Current Status: Assistant Professor, Current Status: Assistant Professor, Computational Fracture Mechanics Current Status: Assistant Professor, Charles Zeeb Parallel Molecular Dynamics Simulation of Cornell University Marc Serre University of Wisconsin-Madison Fellowship Years: 2004-2008 University of California, Berkeley Colorado State University Christopher Oehmen Large Biomolecular Systems University of North Carolina Current Status: Technical Staff Member, Mechanical Engineering University of Memphis/University of Tennessee Fellowship Years: 1995-1999 Alejandro Rodriguez Environmental Stochastic Modeling James Strzelec Sandia National Laboratories – California Glenn Williams Fellowship Years: 1993-1997 Health Science Center Current Status: Senior Research Massachusetts Institute of Technology and Mapping Stanford University University of North Carolina Current Status: Deceased High-Performance Computing in Programmer, University of Illinois Nanophotonics, Casimir Effect Fellowship Years: 1996-1999 Computational Mathematics Rajesh Venkataramani Applied and Computational Mathematics, Computational Biology Fellowship Years: 2006-2010 Current Status: Assistant Professor, Fellowship Years: 1992-1994 Massachusetts Institute of Technology Computational Biology, Etay Ziv Fellowship Years: 1999-2003 Todd Postma Current Status: Postdoctoral Fellow, University of North Carolina Chemical Engineering Environmental Modeling Columbia University Current Status: Senior Research University of California, Berkeley Harvard University Rajeev Surati Fellowship Years: 1995-1999 Fellowship Years: 1993-1996 Computational Biology Scientist, Computational Biology and Nuclear Engineering, Computational Neutronics Jason Sese Massachusetts Institute of Technology Current Status: Goldman Sachs Current Status: Assistant Professor, Fellowship Years: 2004-2008 Bioinformatics Group, Pacific Northwest Fellowship Years: 1994-1998 David Rogers Stanford University Electrical Engineering and Computer Science Department of Mathematics and National Laboratory Current Status: Director of University of Cincinnati Hydrogen Storage on Carbon Nanotubes Fellowship Years: 1995-1997 Stephen Vinay III Statistics, Old Dominion University Scott Zoldi Engineering, Totality Computational Physical Chemistry Fellowship Years: 2003-2005 Current Status: Scalable Carnegie Mellon University Duke University Steven Parker Fellowship Years: 2006-2009 Current Status: Chemical Engineer, Display Technologies Application of Smoothed Particle Eric Williford Analytical Modeling University of Utah David Potere Current Status: Postdoctoral Research Environmental Consulting Company Hydrodynamics to Problems in Florida State University Fellowship Years: 1996-1998 Computational Science Princeton University Fellow, Sandia National Laboratories – Laura Swiler Fluid Mechanics Meteorology Current Status: Vice President of Analytic Fellowship Years: 1994-1997 Demography/Remote Sensing New Mexico Elsie Simpson Pierce Carnegie Mellon University Fellowship Years: 1998-2000 Fellowship Years: 1993-1996 Science, Fair Isaac Corporation Current Status: Research Assistant Fellowship Years: 2004-2008 University of Illinois at Urbana-Champaign Reliability Analysis, Prognostics, Current Status: Manager, T&H Analysis Current Status: Weather Predict Inc. Professor, Computer Science, University Current Status: Consultant, Boston David Ropp Nuclear Engineering Network Vulnerability Analysis, Methods Development, Bettis Atomic John ZuHone of Utah Consulting Group University of Arizona Fellowship Years: 1991-1993 Combinatorial Optimization Power Laboratory University of Chicago Adaptive Radar Array Processing Current Status: Computer Scientist, Fellowship Years: 1992-1994 Astrophysics Rick Propp Fellowship Years: 1992-1995 Lawrence Livermore National Laboratory Current Status: Principal Member Fellowship Years: 2004-2008 University of California, Berkeley Current Status: Senior Scientist, SAIC of Technical Staff, Sandia National Current Status: Postdoctoral Researcher, Computational Methods for Flow Through Laboratories – New Mexico NASA Goddard Space Flight Center Porous Media Fellowship Years: 1993-1996 Current Status: Senior Software Engineer, WorkDay

P42 DEIXIS 11 DOE CSGF ANNUAL DEIXIS 11 DOE CSGF ANNUAL P43 The Krell Institute 1609 Golden Aspen Drive, Suite 101 Ames, IA 50010 (515) 956-3696 Funded by the Office of Science and the www.krellinst.org/csgf National Nuclear Security Administration’s Office of Defense