California Institute of Technology on the Cover Saturn’S Rings Are Made of Particles Ranging from Tiny Pieces of Dust to Huge Boul- Ders

In This Issue:

Seeking WIMPY Particles • Exploring Saturn’s Rings • Picking Five-Star Movies

Volume LXXIII, Number 2, spring 2010

California Institute of Technology on the cover Saturn’s rings are made of particles ranging from tiny pieces of dust to huge boul- ders. On May 3, 2005, the Cassini spacecraft sent radio signals back to Earth through the rings to measure the particles’ size distribution. In this simulated image, which has a resolution of 10 kilometers, purple represents regions where all the particles are bigger than five centimeters wide. Green areas contain particles smaller than five centimeters, and blue ones have particles smaller than one centimeter. The white zones are so dense that the signal didn’t get through. For more on what Cassini is showing us about the rings, go to page 22.

40 years ago in E&S

“The Future Isn’t What It Used to Be,” Arthur C. Clarke told Caltech on April 12, 1970. The dean of science- fiction writers had been invited to campus to speculate on what the world would be like in—of course!—2001. He cautioned that his conjectures were far from predictions, and indeed many of his ideas seem outlandish today. Yet the themes he touched on are still very much relevant. He recognized the inefficiency of raising animals for food, and suggested that we domesticate animals such as antelopes, tapirs, or hippos that could live off “marginal land that’s of no use for anything else.” We could also herd whales instead of cattle, an idea he wrote about in his 1957 novel, The Deep Range. And we could engineer microorganisms to transform “inedible materials—such as sawdust and wastes of various kinds—into food which we or our animals can eat.” He even suggested, those being the days before the first oil crisis, that much of our protein could be derived from petroleum products.

Perhaps fortunately for our taste buds, none of these have come to pass. But Clarke did predict the communications revolution and an interconnected world thriving on a vast “information grid.” Although he thought this revolution would come by way of satellites instead of fiber-optic cables and Internet cafés, the globalized world he envisioned has essentially come true: “The home will have a kind of communications console with a television May 1970’s cover shot of screen, television camera, computer keyboard, microphone, and probably hard-copy readout.” He also foresaw the troubles now facing print journalism, saying that “the newspaper as we know it will be extinct.” red blood cells was taken with a scanning electron Since everyone will be connected remotely, “many people will be able to do most of their work without leav- microscope by Richard ing home—unless their wives insist.” (He apparently didn’t anticipate the rise of the working woman.) Everyone Baker, a professor at USC around the globe would live and work together in step, he said, and time zones will be abolished—or else sleep and a Caltech research itself, via some chemical or electronic innovation. The future would give us more free time, and education would associate. He, Professor be the greatest industry, followed by entertainment. Clarke was hopeful that the world would turn the corner by of Engineering Science the 21st century, that “2001 could mark the great divide between barbarism and civilization.” Maybe it didn’t quite J. Harold Wayland (MS turn out that way, but we can always be optimistic and raise a tall glass of whale milk to the future. ’35, PhD ’37), and JPL’s John Devaney were study- Also in the May 1970 issue, JPL’s Ray Newburn (BS ’54, MS ’55) discussed the prospect of a “Grand Tour” of ing the cells’ mechanical the outer planets, using a then-untried method of gravitational slingshots to take advantage of a 1976 planetary structure. alignment that had last occurred during the Jefferson administration and wouldn’t happen again until 2148. Ac- companying the article were two grainy, telescopic photos of Jupiter and Saturn, and neither Newburn nor anyone else could’ve anticipated the stunning images that the grand tourists—the Voyager spacecraft—would return.

Beyond Saturn, the planets were just fuzzy blobs. Newburn noted that we weren’t even sure of Neptune’s size, much less its density. Seeing the gas giants up close was a once-in-several-lifetimes opportunity. “We have a chance to do that with reasonable economy this decade; otherwise we must develop new vehicles with greater performance and spacecraft with very long lives, or wait until the middle of the 22nd century.” Fortunately, we didn’t wait. —MW

Engineering & Science is printed on recycled paper containing a minimum of 10 percent postconsumer waste. The paper’s virgin pulp comes from trees grown in the U.S. in accordance with the Forest Stewardship Council’s sustainability standards.

2 Random Walk

14 A Mine for Dark Matter by Marcus y. woo Researchers have built an experiment 230 stories underground in search of dark matter, the mysterious stuff that makes up a quarter of the universe.

22 Cassini’s Ringside Seat by linda doran As Cassini explores Saturn, its moons have grabbed most of the headlines. But we’re finding out a lot about its rings, as well.

32 Recommend a Movie, Win a Million Bucks by Joseph sill Can a computer learn to choose movies you’re sure to like? Artificial intelligence took on this real-world problem in the Netflix contest. A Caltech alum recounts his adventures in the quest for the million-dollar grand prize. 41 Letters

43 Books A Dream of Undying Fame: How Freud Betrayed His Mentor and Invented Psychoanalysis by Louis Breger

Engineering & Science (ISSN 0013-7812) is published quarterly at the California Institute of Technology, 1200 East California Boulevard, Pasadena, CA 91125. Annual subscription $20.00 domestic, $30.00 foreign air mail; single copies $5.00. Send subscriptions to Caltech 1-71, Pasadena, CA 91125. Third-class postage paid at Pasadena, CA. All rights reserved. Reproduction of material contained herein forbidden without authorization. © 2010, California Institute of Technology. Visit us online at http://EandS.caltech.edu Published by Caltech and the Alumni Association. Telephone: 626-395-3630. for videos and web links.

PICTURE CREDITS Jasmine R. Bryant STAFF Copy Editors Cover, 23 — NASA/JPL; 2, 14, 16, 17, 22, 34, 36 — Bill Youngblood; 3 — Mike President of the Editor Sara Arnold, Allison Rogers; 5 — Doug Smith; 7 — Caltech Archives, 8 — Vicky Chiu; 12 — Gaby Maimon Alumni Association Douglas L. Smith Benter, Barbara DiPalma, and Michael Dickinson; 15, 18 — CDMS Collaboration; 17 — NASA/CXC/CfA/M. Michael Farquhar Peter D. Hero Writer Markevitch et al., NASA/STScI, Magellan/U.Arizona/D.Clowe et al., NASA/STScI, ESO Vice President for Marcus Y. Woo Business Manager WFI, Magellan/U.Arizona/D.Clowe et al.; 18, 19 — Fermilab; 22-23, 24, 26–31 — NASA/ Development and Alumni Debbie Bradbury JPL/SSI; 26, 27 — NASA/JPL/University of Colorado; 30 — NASA/JPL/USGS; 32-33, Contributing Writers Relations 37 — Doug Cummings; 33 — Joe Sill Allison Benter, Shelley Kristen Brown Erwin, Lori Oliwenstein, Associate Vice President Kathy Svitil for Communications Graphic Artist, Layout Doug Cummings Design Random Walk

Schlinger Lab Dedicated

Caltech’s third new building in as ing “had a department made up of many issues of E&S opened its doors two professors and a secretary”— on a blustery March 9. Almost a Katharine Stewart, whom he married decade in the planning, the Warren the year he got his master’s degree. and Katharine Schlinger Laboratory (Schlinger contributed to the faculty’s for Chemistry and Chemical Engi- growth by staying on as an instructor neering is Caltech’s first new facility until 1953.) specifically and exclusively designed Like the recently opened Cahill and From left: Will Webster (BS ’49); around the research needs of chem- Annenberg buildings, the Schlinger Katharine Schlinger; Warren ists and chemical engineers since the Lab is on track to earn a gold certifi- Schlinger (BS ’44, MS ’46, construction of the adjacent Noyes cation under the LEED (Leadership PhD ’49); Chemistry and Chemi- Laboratory of Chemical Physics in in Energy and Environmental Design) cal Engineering Division Chair 1967. At the dedication, Warren Green Building Rating System. Be- Jacqueline Barton, the Hanisch Schlinger noted that there had been sides using locally derived or recycled Memorial Professor and profes- some growth in the years since he building materials, the Schlinger Lab sor of chemistry; and President arrived on campus as a freshman uses 28 percent less energy and 30 Chameau cut the ribbon. in 1941, when chemical engineer- percent less water than typical build-

2 ENGINEERING & SCIENCE Spring 2010 Random Walk

ings designed for chemical research. noisy, sewing-machine-sized beasts Meeting the stringent energy savings that like to leak their oil all over the required for a gold rating was particu- lab floor. Instead, each of Schlinger’s larly challenging, as the fume hoods rotovaps gets its suction from a pump gulp electricity 24/7. the size of a large paperback book, Fume hoods, for the chemically efficiently and quietly powered by the declined, are the enclosed cabinets in campus’s compressed-air system. which experiments are done. The front Energy-efficient double-glazed side of the hood is a shatterproof win- floor-to-ceiling windows flood the dowpane that can be raised for ac- labs with natural light, a feat that cess or lowered until it is almost, but was made possible by relocating all not quite, shut. Powerful fans up on the plumbing and ducting—normally the roof suck a steady draft of room carried from floor to floor by a “wet” air in under the sash and through the wall—to within a set of elliptical pillars hood in order to keep noxious vapors out in the central hallway. By contrast, away from the lab’s occupants. The the entire west facade of Noyes Lab higher the sash is raised, the more air is windowless, hiding the giant utility whooshes through the hood and the core that serves the labs. Schlinger’s harder the fans have to work. pillars, inlaid with green glass tiles, Above: Even while you’re standing at a Above each of the Schlinger’s complement the rich maple accent fume hood, the outdoors beckons. hoods is an electric eye that con- panels and similarly hued flooring Below: The Grand Promenade. stantly scans its vicinity. If nobody is to give an effect reminiscent of the To see more pictures, check out the around, the sash automatically lowers grand corridors of the Queen Mary. slideshow at http://images.caltech.edu/ to the fully closed position, minimizing Other eco-friendly features include slideshows/Schlinger-architecture/. the volume of air being pulled in. (A a “bio-swale” on the north side that second eye on the sash’s underside keeps a lookout for protruding glass- ware or other objects, stopping the descent if the beam is broken.) This high-tech hood design was pioneered in Europe but is new to the States. Another European innovation new here can be found in Schlinger’s rotary evaporators, which are vacuum- assisted stills. Banks of rotovaps, as they are affectionately known, are essential to any synthetic-chemistry lab—whenever you dissolve something to make it react, you eventually have to get rid of the solvent in order to retrieve your product. A rotovap needs a strong vacuum to get the solvent out as fast as possible, which means either a centralized system with heavy-duty piping, or lots and lots of individual vacuum pumps—

spring 2010 ENGINEERING & SCIENCE 3 collects the runoff from the adjoining planters, sidewalks, and parking stalls, Of Flies and Men filtering it naturally before returning it to the groundwater. There’s also a dedicated room in the basement for The poet and mystic William Blake pheromones, and finding the olfac- collecting and sorting recyclables. saw a world in a grain of sand. A fly’s tory sensory neurons that express Weather permitting, you can even brain is scarcely larger, yet Caltech these receptors.” According to Wang, get in touch with nature without ever scientists see in it a window for the paper’s first author, the only leaving the building. One entire wall of exploring the biological roots of our insect meeting these requirements the first-floor classroom/conference own behavior and emotions. The was Drosophila melanogaster. “The room folds up into the ceiling like a brain of Drosophila melanogaster, genetic and molecular architecture of set of glass garage doors, opening the common fruit fly, contains barely Drosophila’s olfactory system is well onto a courtyard. 20,000 neurons—yet two recent pa- understood,” he explains. “One can As Division Chair Jacqueline Barton pers from the lab of David Anderson, easily test whether a specific recep- remarked at the dedication, “When Benzer Professor of Biology and an tor or neuron is involved in a given you bring chemists and chemical investigator with the Howard Hughes behavior.” engineers together in one laboratory, Medical Institute, offer glimpses into Wang discovered that 11-cis- the results will be far greater than its genetic hardwiring that may throw Vaccenyl Acetate (cVA), a pheromone the sum of the parts.” Schlinger’s light on what makes us tick. present in the male fly’s cuticle, or reconfigurable lab spaces will house For example, both inconsiderate exoskeleton, promotes aggression in the research groups of three chemists boors and unthinking flies will elbow pairs of male flies. An aggressive fly and three chemical engineers, work- their way to the front of the press at will “lunge,” rearing up on its hind legs ing in fields ranging from drug design a crowded lunch counter, causing and snapping its forelegs down on its to pollution control. The Center for the less assertive to go elsewhere. opponent. When Wang and Anderson Catalysis and Chemical Synthesis Now grad student Liming Wang and added synthetic cVA to the “arena” will also move in, and there’s enough Anderson, writing in the January 10 in which combatant flies were tested, room remaining for two new hires. edition of Nature, have identified an the frequency of lunges dramatically The building’s architects, Bohlin aggression-promoting pheromone increased. Building on earlier work Cywinski Jackson, are known for sus- that appears to help drive competi- elsewhere that had identified cVA’s tainable design and have done labs tors away from a crowded piece of, receptors, Wang next showed that and other academic buildings across say, overripe banana, and pinpointed silencing the cVA-sensitive neurons in the nation. Rudolph and Sletten was the neurons in the fly’s antennae that the antennae mellowed the flies out. the general contractor. detect it. To find out whether natural cVA Besides the Schlingers, support Pheromones are chemicals used from other flies had the same effect, for the building and its research was by particular species to communicate Wang and Anderson then trapped provided by the Gordon and Betty with their own kind, but proving that between 20 and 100 “donor” male Moore Foundation, Will and Helen a pheromone released by the insects flies—so called because they donate Webster, Victor and Elizabeth Atkins, themselves—rather than being pro- their pheromones into the surround- the John Stauffer Charitable Trust, vided in a synthetic form by inquisitive ing environment—in a tiny cage Barbara Dickinson, the Ralph M. scientists—normally controls aggres- surrounded by a fine mesh screen. Parsons Foundation, John Willard sive behavior “required the ability The screen allowed the pheromones Jones (BS ’41), Patricia Beckman, to experimentally interfere with the to escape, but not the flies. A pair of and Gregory P. Stone (BS ’74, MS insects’ capacity to sense the phero- “tester” males would be placed on ’74). — DS mone,” Anderson notes. “And that, in top of the cage, where they could turn, meant identifying the receptor sense the pheromone but not interact molecules that detect aggression with the donors. “Remarkably,” says

4 ENGINEERING & SCIENCE Spring 2010 On March 9, Mylar dirigibles battled for the skies—or at least for the airspace within Brown Gymnasium—in “Re- venge of the Hindenberg,” this year’s installment of the ME 72 (Engineering Design Laboratory) contest. For full coverage, see http://weblab. caltech.edu/features/16.

Anderson, “the presence of the caged donor flies strongly increased aggression between the tester flies, and this aggression-promoting effect increased with the number of donors.” And again, the testers’ testiness was assuaged by inactivating their cVA- sensing neurons. Which brings us back to the lunch counter—or more aptly, the free food at happy hour. Male flies are attracted to food not only to eat, but also to mate with feeding females. And, of course, the more guys there are, the harder it gets to score. Since feisty flies tend to chase away their competitors, an aggression-promoting pheromone might keep the number of males down to an equitable level. Wang tested this hypothesis by allowing a small number of flies to compete for a limited food supply, after genetically manipulating their Because pheromones evolved as order (ADHD). Most of Drosophila’s cVA-receptor neurons to make them “private” communications channels genes are also found in humans— more excitable. The flies quickly dis- within a given species, it’s unlikely including those for the neurons that persed. “They fought one another until the fly pheromone would work on produce dopamine and serotonin, a dominant fly became ‘king of the hill’ us. However, that doesn’t necessar- both of which have been implicated in and drove the others away,” Anderson ily mean that humans lack aggres- psychiatric disorders. explains. sion pheromones, Anderson notes. A team led by then-postdoc Tim According to Wang and Anderson, They’ve been discovered in mice, Lebestky found that a rapid succes- this suggests that when the popula- which are evolutionarily closer to us sion of brief, brisk puffs of air caused tion of male flies reaches a certain than flies, so it’s possible we might flies to run around their test chamber density, the concentration of cVA have our own as well. But whether in what Anderson calls a “frantic man- rises to a level that promotes ag- such pheromones can keep lines ner” for several minutes after the last gression, forcing some of the flies off short at the buffet, Anderson remarks, puff. “Even after the flies had calmed the food. Their departure decreases “only time will tell.” down,” he adds, “they remained hy- the ambient concentration of the Anderson’s lab has also seen signs persensitive to a single air puff.” These pheromone, decreasing aggression. of a primitive emotion-like behavior, “hyperactive” flies were picked out “The population becomes stabilized specifically a state of agitation, that from the crowd via an automated ma- at an optimal density until more flies might illuminate the relationship be- chine-vision-based system developed become attracted to the food, and the tween the neurotransmitter dopamine in the lab of Anderson’s colleague cycle repeats itself,” says Wang. and attention deficit hyperactivity dis- Pietro Perona, the Puckett Profes-

spring 2010 ENGINEERING & SCIENCE 5 sor of Electrical Engineering. These cause they are hyperactive and easily the powerful genetics of this model flies proved to have a mutation called distracted. But this work shows that organism to understanding how these DopR that inactivated a dopamine hyperactivity and learning disabilities chemicals influence behavior through receptor known as D1—a result that are unconnected—in flies, at least. their actions on specific brain circuits,” was published in the November 25, “We could separately ‘rescue’ the says Anderson. “While the specific 2009, issue of Neuron. hyperactivity and learning deficits in a details of where and how this occurs This discovery dovetails with what completely independent manner,” says are likely to be different in flies and is known about ADHD, which is Anderson, “by genetically restoring in humans, the basic principles are characterized by impulsivity, hyperac- the dopamine receptor to different likely to be evolutionarily conserved, tivity, and a short attention span, and regions of the fly’s brain.” If it turns out and may aid in our understanding of is often treated with drugs such as that ADHD works in a similar way, An- what goes wrong in disorders such as Ritalin that increase dopamine levels derson believes that it may be better ADHD.” in the brain. The way the mutant flies to develop drugs to treat the two is- The research described in both responded to the air puffs is, more- sues separately. The broad-spectrum papers was supported by grants from over, “reminiscent of how individuals pharmaceuticals now used to attack the National Science Foundation and with ADHD display hypersensitivity both at once tend to have undesirable the Howard Hughes Medical Institute. to environmental stimuli and are more side effects. —LO easily aroused by such influences,” Besides Lebestky, Anderson, and says Anderson. Furthermore, ADHD Perona, the other people involved in often goes hand in hand with learning the work are Caltech biology research disabilities, and Anderson’s collabora- technician Jung-Sook Chang, then- tors at Penn State have shown that postdocs Heiko Dankert and Lihi flies with theDopR mutation can’t Zelnik; Young-Cho Kim and Kyung-An learn to associate a particular odor Han from Penn State; and Fred Wolf with an electric shock. They don’t from UC San Francisco. avoid the odor afterward, while flies That flies exhibit emotion-like without the mutation quickly catch on. behaviors controlled by some of the It’s often assumed that ADHD kids same brain chemicals as in humans have difficulty learning precisely be- “opens up the possibility of applying

Another Alum Goes to Washington

The brain drain from Pasadena to Edmonds, an engineer by train- Policy Act of 2005. “I am pleased to Foggy Bottom continues. Karina ing, has held key technology-transfer have Karina join our team,” says En- Edmonds, MS ’93, PhD ’98, direc- positions at Caltech and JPL for over ergy Secretary Steven Chu, who was tor of JPL technology transfer, joined a decade. As director of JPL technol- also Caltech’s commencement speak- the Department of Energy as its ogy transfer, she was responsible for er last year. “Having Karina oversee a first technology transfer coordinator technology licensing, managing the coordinated, strategic effort on behalf on April 12. She will work with the JPL patent portfolio, and assisting of the department will help increase DOE’s national laboratories to help Caltech/JPL start-up companies. the rate of successful technology accelerate the process of moving This is the first time the DOE has transfers, creating clean-energy jobs discoveries from the laboratory to the appointed a full-time person to fill the and providing more solutions to our private sector. role, which was created by the Energy energy challenges.” —AB

6 ENGINEERING & SCIENCE Spring 2010 The six figures on Calder’s arches represent (from left) Nature, Art, Energy, Science, Imagina- tion, and Law.

Caltech TURNs 100—AGAIN

Although founded in 1891, Caltech The leap from vo-tech to high-tech benefactor secured a site for a new, can once again celebrate its 100th was the work of noted solar as- expanded campus—some 22 acres of anniversary. On June 8, 1910, the first tronomer George Ellery Hale, who orange groves dotted with stately oak building on the present-day campus had come west in 1903 to be the trees in the southeast part of the city. was dedicated before local digni- founding director of the Mount Wilson Throop’s original campus— taries and a large public audience. Observatory. Hale was soon deeply in acquired after a start-up year in the Dubbed Pasadena Hall (and renamed the flow of Pasadena’s civic, cultural, old Wooster Block, still a presence Throop Hall in 1920), it was hailed as and educational schemes, becom- on the corner of Fair Oaks Avenue a monument to civic pride. The first ing a tireless booster of Southern and Green Street in the heart of Old students to be educated in it arrived California in general and Pasadena in Pasadena—crammed all six schools the following September—30 in total, particular. He soon became bent on into three buildings at Lincoln Avenue all male, and all enrolled in a college- establishing a local technical school and Fair Oaks, a site that is today un- level engineering curriculum. Tuition to train engineers (construed at that der the 210 freeway. Hale envisioned was $150 annually. time to mean men only) to meet the the new campus two miles east as an Throop Polytechnic Institute, as it needs of a booming region—in partic- opportunity for an idealized building was then known, had just split apart ular, to bring water and electricity over scheme in harmony with a new civic at the seams. The old Throop had the mountains to a sun-drenched but center, a campus whose laboratories evolved into an agglomeration of utility-starved Los Angeles basin. By would be fitted out with the latest and six schools, teaching at levels from 1907 he had begun a campaign for best equipment. Such an institute elementary to collegiate, with a heavy the creation of a “high-grade institute would redound to the glory of Pasa- emphasis in the upper division on of technology” in Pasadena and was dena and would surely inspire the such practical skills as stenography, elected to the Throop board of trust- generosity of Pasadena’s well-to-do typing, and operating machine tools. ees. In that same year, an anonymous residents. Hale was right. The mission-style structure by architects Myron Hunt and Elmer Grey was paid for entirely by local subscriptions, to the tune of approximately $165,000. The arcaded entrance was adorned by a set of reliefs created by Pasadena’s Alexander Stirling Calder, whose son would invent the mobile. Touted at the time as the most significant artwork

Throop Hall (with Dabney Hall to the left) in April 1965. The president, the provost, the treasurer, and the deans had offices on the first floor. Various business offices—payroll, personnel, accounting, central files, and so on—occupied the second floor. Ed Hutchings, editor of E&S, lived in the basement with the news bureau, the alumni office, and most of development.

spring 2010 ENGINEERING & SCIENCE 7 Throop Hall was demolished after the San Fernando earthquake, and the Calder Arches now adorn the facade of the Arnold and Mabel Beckman Laboratory of Chemical Synthesis. The Throop site is now a vest-pocket park in the middle of campus—a perfect place for a photo op. Here Lemelson winner Heather Agnew (right) and finalist Yvonne Chen enjoy their accolades.

in the city, the elaborate, allegorical figures were, in Calder’s words, “to give plastic utterance to the aims and scope of the school.” The new building’s 62 rooms housed what the 1910 catalog boast- Lemelson Winners ANNOUNCED ed as being “the only college devoted primarily to Technology west of the Mississippi River.” Meanwhile, the Deep in the Amazon, a woman is package or the antibodies degrade, other five schools were closed down keeling over with stomach pains and Agnew says. But the developing or divested. The elementary school vomiting. Does she just have the flu, world, which needs such simple moved to a new location a block west or is she one of two billion people diagnostic tools the most, isn’t always of the new campus and became the worldwide who has been afflicted air-conditioned. Furthermore, antibod- Polytechnic School. Throop Acad- with Hepatitis B, a potentially deadly ies are expensive to produce. Today, emy remained at the old campus liver disease? Today’s diagnostic tools many tests look for just one or two and eventually merged with a new are too delicate for health workers to proteins, Agnew says. But diseases public high school. And, after almost use in the steamy environment of a re- like cancer are complex, so an ac- becoming UC Pasadena in 1911 and mote jungle. But in the future, a drop curate diagnosis might require the completing the Gates Laboratory of of blood from a prick of the finger and measurement of more than a dozen Chemistry in 1917, Throop College a cheap, simple device that works proteins, each by its own antibody. of Technology rebranded itself as the in nearly all conditions may change Agnew and her colleagues are California Institute of Technology in that. Heather Agnew (PhD ’10) and building cheap, durable antibody 1920—so we have another 10 years Jim Heath, the Gilloon Professor and replacements called protein-capture to wait for that party. professor of chemistry, are working to agents out of synthetic peptides, An online exhibit about the 1910 make such devices a reality. For her which are relatively short chains of campus may be found at the Caltech role in this effort, Agnew has won the amino acids—the building blocks of Archives website: http://archives. $30,000 2010 Lemelson-MIT Caltech proteins. Peptides are cheap to make, caltech.edu/ . —SE Student Prize. and can be designed to have all sorts A diagnostic test, or assay, can of nice properties, including heat measure the amount of a protein spe- resistance and biological or chemical cific to some disease by allowing it to stability. But since they’re small mol- bind to another molecule, called an ecules, they don’t stick to their target antibody, that is tailor-made to recog- proteins as well as antibodies do. nize it. Assays can be packaged into Reasoning that two peptides of easy-to-use kits for diagnosis outside middling stickiness might do the the lab, and they’re not restricted trick if they worked together, Agnew to blood. For example, the home and her coworkers tested millions of pregnancy test assays a hormone them. And here the project got help called human chorionic gonadotropin from the target protein itself—when in urine. appropriately primed versions of The problem with such as- the peptides recognized the protein says, though, is that the antibodies and bound to it, it held them in just themselves are proteins, sensitive the right orientations that they could to heat, humidity, and other factors. “click” together to create a new mol- For instance, HIV assays have to be ecule that is 10 to 100 times better performed within hours of opening the at binding to the target protein than

8 ENGINEERING & SCIENCE Spring 2010 either peptide alone. Repeating the molecule of RNA—which is similar to it more versatile and easier to control. process to add a third peptide further DNA—that acts like a switch, turning The Lemelson-MIT Caltech Prize enhances the binding. cytokine production on when exposed is funded by the Lemelson-MIT As for durability, Heath’s benchmark to theophylline, a caffeine-like mol- Program, founded in 1994 by Jerome is what he calls the Pasadena Test: ecule (see E&S 2005, No. 4). When H. Lemelson to inspire young innova- will it work even after a year spent the theophylline infusion stops, so tors. Chen’s prize as a finalist was baking in the trunk of his car? Agnew does cytokine production, and the T donated by Michael Hunkapiller (PhD says her protein-capture agents have cell dies. This is just a demo, as large ’74). Lemelson-MIT student prizes are withstood airplane travel and years of doses of theophylline can cause an also at MIT, Rensselaer Polytechnic sitting on a shelf in her office. irregular heartbeat and even death. Institute, and the University of Illinois A second award of $10,000 went Fortunately, the RNA switch can eas- at Urbana Champaign. The Caltech to Yvonne Chen (MS ’07), a grad stu- ily be designed so that it responds prize was first awarded last year. dent working with Christina Smolke, a to a harmless molecule, such as a —MW former assistant professor of chemical vitamin. Chen is now working to make engineering at Caltech who’s now at Stanford. Chen developed a way to help T cells fight cancer. T cells are a part of the body’s protective army, Voyager nears the local fluff and other researchers have been able to engineer them to attack cancerous tumors. “We can keep putting them Our solar system is plunging through show that the Fluff is much more in the blood supply until they home a vast cloud of wispy gas called the strongly magnetized than anyone in on the tumor,” Chen explains. “The local interstellar cloud, also known as had previously suspected—between problem is that they die really quickly.” the “Local Fluff.” About 30 light-years four and five microgauss,” Opher Because T cells are a part of the wide, the Fluff is made of 6,000°C told Science@NASA. “This magnetic body’s natural immune response, they hydrogen and helium. The Fluff is field can provide the extra pressure die by default if they aren’t instructed about twice as dense as the interstel- required to resist destruction.” to attack. “Our challenge then is to lar meduim surrounding it, and what Previous estimates of the Fluff’s figure out how to engineer this T-cell holds it together has been a mys- field had been in the 1.8 to 2.5 population to be sustainable so they tery—until now, thanks to a discovery microgauss range. By comparison, can finish killing the tumor cells.” by JPL’s twin Voyager spacecraft. Earth’s magnetic field is about half T cells are kept alive by molecules By rights, the Fluff shouldn’t exist. a gauss, or roughly a million times called cytokines. But you can’t just A group of nearby stars exploded stronger. inject cytokines into someone to about 10 million years ago, and the Inside the Fluff—and encompass- keep the T cells going—you’d need resulting blast of million-degree gas ing us—is a 10-billion-kilometer- a lot, and too much would put the is now blowing past us. The Fluff is wide bubble called the heliosphere, patient into shock. One solution is neither hot enough nor dense enough which helps shield us from constant to engineer the T cell to produce its to withstand the onslaught, says bombardment by high-energy cosmic own cytokines. But you also have to Merav Opher, a former JPL postdoc rays from the depths of space. The regulate cytokine production carefully, now on the faculty at George Mason heliosphere is kept inflated by the because an excess will cause the T University. But in the December 24, solar wind, a stream of charged cells to reproduce nonstop, resulting 2009, issue of Nature, Opher and her particles emitted by the sun, so its in leukemia. colleagues reported that the latest size is determined by the balance With Smolke, Michael Jensen from data from Voyager 2 reveal a mag- of forces between the solar wind City of Hope medical center, and netic field strong enough to enable pushing out and the local interstellar other researchers, Chen made a the Fluff to push back. “Voyager data cloud pressing back. In 2004 and

spring 2010 ENGINEERING & SCIENCE 9 2007, respectively, Voyagers 1 and 2 crossed into the heliosphere’s outer Money on Your Mind layers, a region called the heliosheath. (See E&S 2008, No. 3.) Once there, they could measure the size of the It’s perhaps not surprising that aver- These two people, as well as other heliosphere, allowing scientists to sion to losing money is hardwired into volunteers, were each given $50 in calculate how much pressure the Fluff our brains, but a sense of fairness cash and then offered a series of is exerting on it. This pressure, in turn, seems to be as well. These are just bets on the outcome of a computer- partly depends on the strength of the two results from recent work at Cal generated coin toss. Each potential Fluff’s magnetic field. tech’s Brain Imaging Center, where wager had the same odds, 50/50, but This discovery raises the possibil- a multidisciplinary team of biologists a different ratio of payout to loss. For ity that other clouds in our galactic and social scientists are using func- example, you might get the chance to neighborhood are also strongly mag- tional Magnetic Resonance Imaging win $20 or lose $5 (a risk most peo- netized, and when the solar system (fMRI) to map behavior onto brain ple will accept), or you might stand collides with them, they will push structure with millimetric precision. to lose $20 for the same $20 return back even harder. If the heliosphere An fMRI scanner tracks blood (a bet most people will decline). In is further compressed, more cosmic flow in the brain as a proxy for brain general, people shied away from the rays might reach Earth. “There could activity. The test subject lies in the prospect of large losses, so even the be interesting times ahead,” Opher scanner and is then asked a series proposition of winning $20 versus says. But there’s no need to get out of questions or told to perform some losing $15 got few takers, “even the tinfoil hats quite yet—we won’t run other sort of mental activity, such though the net expected outcome is into the next cloud for hundreds of as memorizing a list of names, while positive,” Adolphs says. thousands of years. —MW the experimenters literally watch him Neither of the amygdala-damaged or her think. Many experiments use patients were fazed by the prospect pairs of volunteers, each in their own of losing money, taking risky gam- scanner, trying to outwit each other in bles much more often than control various strategy games where cash is subjects. “We think this shows that TUNE IN TO “TODAY” on the line. the amygdala is critical for trigger- For more on what’s happening with It turns out that the fear of losing ing a sense of caution,” explains Caltech and the Caltech community, check out the articles on Caltech Today, money lives in the amygdalae, two Camerer. This function, he says, may which offers online coverage of the almond-shaped clusters of tis- be similar to the amygdala’s role in Institute and its activities. There you’ll sue located in the medial temporal fear and anxiety. “Loss aversion has find press releases covering the latest lobes. (The amygdala registers rapid been observed in many economics Institute research and feature articles emotional reactions and appears to studies, from monkeys trading tokens highlighting faculty, student, alumni, play a role in depression, anxiety, and for food to people on high-stakes and campus activities. Click on the autism.) Benedetto de Martino, a visit- game shows,” he adds, “but this is the links with icons for the latest stories or ing researcher from University College first clear evidence of a special brain scroll down to the Feature and News London; Colin Camerer, the Kirby structure that is responsible for fear of archives at the foot of the page and Professor of Behavioral Econom- such losses.” click on those links to browse through ics; and Ralph Adolphs (PhD ’92), A paper on this research appeared older articles. the Bren Professor of Psychology in the February 23 issue of the Pro- Caltech Today can be accessed through and Neuroscience and professor of ceedings of the National Academy of its link on the Caltech home page or biology, found the seat of this fear by Sciences. The work was supported directly at http://today.caltech.edu/. studying two patients whose amygda- by the Gordon and Betty Moore You can also subscribe via email or RSS lae had been destroyed by a very rare Foundation, the Human Frontier Sci- feed. genetic disease. ence Program, the Wellcome Trust,

10 ENGINEERING & SCIENCE Spring 2010 the National Institutes of Health, the more money they could potentially Camerer, too, found the results Simons Foundation, and a Global get—from zero dollars up to another thought provoking. “We economists Center of Excellence grant from the $50—in a payout scheme selected at have a widespread view that most Japanese government. random at the end of the run. people are basically self-interested, Another study, by Professor of As it turned out, the way the and won’t try to help other people,” Psychology John O’Doherty, Camerer, volunteers—or, to be more precise, he says. “But if that were true, you then-postdoc Elizabeth Tricomi, and the reward centers in their brains— wouldn’t see these sort of reactions Associate Professor of Economics reacted depended strongly upon to other people getting money.” Still, Antonio Rangel (BS ’93), looked whether the volunteer was the “poor” he says, the rich may have been at at the brain’s reward centers. It’s or the “rich” member of the pair. least partly motivated by self-inter- long been known that we don’t like “People who started out poor had a est—or a reduction of their own dis- inequality, especially when it comes strong reaction to getting money, and comfort. “We think that, for the people to money. Tell two people working the essentially no reaction to money go- who start out rich, seeing another same job that their salaries are differ- ing to another person,” Camerer says. person get money reduces their guilt ent, and there’s going to be trouble, “By itself, that wasn’t too surprising.” over having more than the others.” notes O’Doherty. “It’s not just the What was surprising was the other O’Doherty says that the next step is application of a social rule or conven- side of the coin—“people who started to attempt to figure out how these re- tion; there’s really something about out rich had a stronger reaction to actions translate into changes in be- the basic processing of rewards in other people getting money than to havior. “For example, the person who the brain that reflects these consid- themselves getting money. In other finds out they’re being paid less than erations.” words, their brains liked it better when someone else for doing the same job The experimenters watched how their poorer partner got the money.” might end up working less hard. It will the ventromedial prefrontal cortex “We now know that these areas be interesting to try to understand the (VMPFC) and the ventral striatum— are not just self-interested,” adds brain mechanisms that underlie such two well-known reward centers in O’Doherty. “They don’t exclusively changes.” the brain—reacted to the prospect respond to the rewards that one gets These findings were published in of being offered various amounts of as an individual.” Instead, contrary to the February 25 issue of Nature. The money. But there was a twist—the 40 the prevailing wisdom about human project was funded by grants from volunteers were paired off before- nature, the brain evaluates the overall the National Science Foundation, the hand, and one person in each pair equity of the situation. “It shows that Human Frontier Science Program, the was given an extra $50 before the ex- the basic reward structures in the hu- Gordon and Betty Moore Founda- periments even began. Then, in each man brain are sensitive to even subtle tion, and the Caltech Brain Imaging trial, the pair would be told how much differences in social context.” Center. —KS/LO

“People who started out rich had a stronger reaction . . . . Their brains liked it better when their poorer partner got the money.”

spring 2010 ENGINEERING & SCIENCE 11 Getting Inside a Fly’s head

What goes on in the tiny brain of a place—“but free to flap its wings.” By what causes the change in sensitivity. fruit fly? We’re beginning to find out, slicing off a patch of the hard cuticle Is the system turned off when the fly now that Michael Dickinson, the Za- covering the brain, “we were able to is on the ground? What neurochemi- rem Professor of Bioengineering, and target our electrodes onto geneti- cals are involved? We can use all postdocs Gaby Maimon and Andrew cally marked neurons,” he says. As the genetic tricks that are available in Straw have succeeded in recording the electrodes took data, high-speed fruit flies to get a better idea of what the activity of individual brain cells digital cameras simultaneously re- is going on.” Adds Maimon, “Sensory as the fly flies. This is no mean feat, corded the flies’ behavior. neurons in many species—including considering that each fly is only about The study focused on a set of birds, rodents, and primates—change 2.5 millimeters long. visual-system neurons that “basically their response strength depending on “Researchers have recorded the help the fly detect when its body the behavioral state of the animal, but neural-cell activity of fruit flies before, posture changes” in order to maintain why these changes take place is not but only in animals that had been stable flight, Dickinson says. When entirely clear.” stuck or glued down,” Dickinson the wings started flapping, these cells The researchers also plan to spy on explains. “Gaby was able to develop immediately ramped up their activity. olfactory and motor cells to see if they a preparation where the animal is “The neurons’ responses to visual display similar behavior. “The ques- tethered”—its head clamped into motion roughly double when the flies tion is, ‘Is the entire brain completely begin to fly, which suggests that the different in flight?’” Dickinson says. system is more sensitive during flight,” “We suspect that this phenomenon is Dickinson says. “The increase is very not unique to the visual cells we have abrupt. It’s not at all a subtle change, studied. Most cells care whether the and so we suspect that there is a animal is flying or not.” neurochemical quickly released during A paper describing the research flight that sets the animal’s brain in was published in the March issue of this different state.” Nature Neuroscience; the work was Previous studies in locusts—which funded by the National Science Foun- are far bigger and thus far easier to dation and a Caltech Della Martin study—had suggested the existence Fellowship. —KS of this effect. However, the genetics of locusts are not nearly as well understood as those of Drosophila. Now, says Dickinson, it should be possible to “figure out specifically

This fruit fly has a dye-filled glass electrode (pink) inserted into its brain. The fly’s head is clamped to the underside of a reservoir filled with a sterile saline solution (colored blue here) that bathes the electrode and the brain. At rest, the fly clings to the reservoir; a gentle puff of air starts it flapping its wings in tethered flight.

12 ENGINEERING & SCIENCE Spring 2010 Old Magazines Never Die . . .

The Intel Science Talent Search, DeBenedictis built on research have been a fifth-grader. In a presen- formerly the Westinghouse Science by JPL’s Martin Lo (BS ’75 and tation she gave at JPL on April 15, Talent Search, is to high-school sci- a Science Talent Search winner she cited E&S as her inspiration. ence fairs what the World Series is to himself), in collaboration with control Contacted by email, she elabo- sandlot baseball. The grand prize is and dynamical systems professor rated, “I think what happened (as with $100,000, and recent winners of this Jerrold Marsden’s research group, most interesting science articles) was nationwide competition have done on what Lo calls the “Interplanetary that I saw something I liked and asked such things as creating a 50-gene Superhighway”—a set of low-energy my dad to explain it to me. That’s why model for predicting the probability routes connecting every massive when I thought of it a few years later of a specific colon cancer recurring, body in the solar system through the he remembered it too and was able to building a Littrow-type spectrograph, intersections of rotating Poincaré find it again. and designing a nanosensor for manifolds. In fact, Lo, a colleague of “You would probably be surprised neurotoxins. Erika’s father, Sandia National Lab’s how much difference the articles you This year’s top honor went to Erika Erik DeBenedicits (BS ’78, PhD ’83), write make—E&S is one of my favorite DeBenedictis of Albuquerque, New helped her get started on a precursor magazines to flip through and look at Mexico, for “a software navigation project in 2007–8. the cool stuff.” system that would allow spacecraft to If the Interplanetary Superhighway DeBenedictis will be matriculat- exploit low-energy orbits . . . for more sounds vaguely familiar, it’s because ing at Caltech in the fall, and hopes efficient transit routes through the an article on it appeared in E&S in to work at JPL when she graduates. solar system.” 2002, when DeBenedicits would —DS

Making Book on the LHC

As you no doubt know by now, the just in time for the regularly scheduled release received by this office, a Large Hadron Collider, or LHC, is winter shutdown. publicly traded Irish online betting back up and running again at a stable, Caltech physics faculty, staff, and firm named Paddy Power is laying record-setting collision energy of sev- students pulled an all-nighter to watch odds on what the LHC will discover en trillion electron volts. The LHC was the restart on a live video feed from first. “The mysterious and previously switched on with great fanfare in Sep- Geneva, where the European Organi- undetectable form of matter known tember 2008 (see “Beam On!,” E&S zation for Nuclear Research (CERN) as Dark Matter is the red-hot 11/10 2008, No. 3) and shut down again and the LHC are located. And it was favourite, followed by Black Holes at nine days later due to a faulty electri- a long night—after two false starts, 8/1 and Dark Energy at 12/1. God cal connection that led to a massive the countercirculating proton beams remains the 100/1 outsider.” coolant leak and ultimately damaged were finally brought into collision —DS 53 of the more than 1,600 supercon- just after lunch in Switzerland, which ducting magnets. It took over a year unfortunately translated into 3:58 a.m. to repair everything, and the LHC was our time. restarted again in November 2009, Meanwhile, according to a press

spring 2010 ENGINEERING & SCIENCE 13 A Mine for Dark Matter

For two weeks in mid-December 2009, the the results would be published in Nature; an Researchers have built an physics world was buzzing with anticipa- announcement to be made in such a widely experiment 230 stories tion and speculation. A team of researchers read and prestigious journal must mean big was rumored to have made an astounding news. This was soon debunked, but it was underground in search of an discovery—they’d detected dark matter, too late. The rumor had already drawn the elusive particle that may be the unknown stuff that makes up nearly a attention of physics blogs, including Cosmic dark matter, the mysterious quarter of the universe. The world’s leading Variance (in a post by Caltech senior re- experiment to find dark matter, the Cryo- search associate Sean Carroll), and those of stuff that makes up nearly a genic Dark Matter Search (CDMS), had just New Scientist and symmetry magazines. quarter of the universe. finished analyzing its final data set. Word To further fan the flames, a pair of talks had somehow reached the blogosphere that announcing the results were scheduled

Assistant Professor of Physics Sunil Golwala stands in front of a three-dimensional map of dark matter. Measurements of how light bends around massive galaxy clusters, an effect called gravitational lensing, allow astronomers to better estimate not only how much dark matter is out there, but also how it has governed cosmic evolution. This map, which has time as its third dimension, goes halfway back to the Big Bang (over Golwala’s shoulder). As your eye moves to the left, the dark matter goes from smooth to clumpy, dragging the visible matter with it to form the galactic structures of today. This map, the first of its kind (see E&S 2007, No. 1), was published in a 2007 paper by then-postdoc Richard Massey and others, including JPL scientist Jason Rhodes and Steele Family Professor of Astronomy Richard Ellis. The data came from the Hubble Space Telescope’s Cosmic Evolution Survey, led by Moseley Professor of Astronomy Nick Scoville, which spent nearly 1,000 hours looking at a patch of sky about the size of nine full moons.

14 ENGINEERING & SCIENCE spring 2010 Each CDMS detector is a 230- gram germanium crystal. Six detectors are stacked to form one of the five towers that make up the whole apparatus.

By Marcus Y. Woo

to be given simultaneously at Fermilab in sits about 230 stories deep in the Soudan What’s the Matter? Illinois and the Stanford Linear Accelerator Underground Laboratory, a research facility For decades, dark matter remained an Center (SLAC) in California. These talks built by the University of Minnesota, Fermi- abstraction, living within the confines of con- were set weeks before, says Assistant Pro- lab, and the Minnesota Department of Natu- jecture and theory. Caltech’s Fritz Zwicky— fessor of Physics Sunil Golwala, a member ral Resources. Why the DNR? Because the the eccentric, cantankerous iconoclast who of the CDMS team. But, coupled with the lab sits in the bowels of an old iron mine was a professor of physics from 1941 to rumors, the scheduled talks only added to nestled among the lakes and forests at the 1968—coined the term nearly 80 years the rampant speculation. “Then it got crazy northeast tip of Minnesota. The CDMS team ago. In 1933, he found that galaxies in a for a couple of weeks,” Golwala recalls. The numbers nearly 80 people from 16 institu- group called the Coma Cluster were zipping CDMS team had decided early on not to tions around the world, including Caltech. around a common center of gravity much discuss the results before the presentations, Although CDMS is far from alone in trying faster than they should’ve been—at those to ensure that data wouldn’t be released to catch WIMPs, it’s been the standard- speeds they should have been flying apart. before a thorough vetting. The secrecy, bearer for the past few years. No experiment The only way the galaxies could stay clus- however, just got people more suspicious. has yet detected anything, but each silent tered was if there was more mass to them “People came up to me and tried to read result narrows down what WIMPs might than met the eye—some new type of matter my facial expression,” says Jeff Filippini, a look like—any theory that predicts some- that only interacted with stars, dust, and postdoc and CDMS team member. thing the experiments don’t see has to be gas through mutual gravitation. This stuff No one knows for sure what dark matter refined or ruled out. CDMS has provided couldn’t be seen, and was therefore “dark.” is made of, but so far the best guess is the tightest constraints yet, and these latest Other astronomers, however, didn’t take his that it consists of a type of particle called a results, taken over a period of more than a pronouncement seriously. (In fact, much of weakly interacting massive particle (WIMP). year, have doubled the collaboration’s data. Zwicky’s research was ahead of its time, and If physicists and astronomers are right about If physicists were close to finding WIMP now many consider him to be an overlooked this, then WIMPs should be all around collisions, then CDMS would have been the genius.) us, zooming about at hundreds of kilome- first experiment to do so—which explains Not until the 1970s, when Vera Rubin ters per second. But because they hardly why people became so anxious upon hear- of the Carnegie Institution of Washington interact with regular matter, you can’t see or ing the rumors. The hype underscores just measured how fast spiral galaxies spin, did feel them. There could be billions of them how momentous a dark-matter discovery the notion of dark matter gain greater ac- streaming through your body right now. would be. “It’s a really exciting topic,” says ceptance. The principle behind her discov- Once in a while, though, a WIMP could Golwala, whose two graduate students, ery is the same as Zwicky’s—she looked crash into an atomic nucleus like a cue Zeeshan Ahmed (MS ’08) and David Moore, at dozens of spiral galaxies and found that ball hitting an eight ball, and that’s the idea did a lot of the number crunching for the the outer stars were circling so fast that the behind most dark-matter searches, including new data. “Suppose you have conclusive galaxies should’ve been ripping apart. Since CDMS. evidence that you just discovered the dark then, even more accurate measurements of The detector consists of 30 hockey- matter in the universe,” he says. “I mean, galaxies and galaxy clusters have revealed puck-sized crystals of germanium waiting that’s just amazing.” a universe filled with mass we can’t see, for a WIMP to come along. To block cosmic holding galaxies together like cosmic glue. rays that might confuse the signal, CDMS Nearly all galaxies appear to be embedded

spring 2010 ENGINEERING & SCIENCE 15 In addition to CDMS, Golwala’s research group works on a variety of other topics in observational cosmology, exploring the origin of the universe and the nature of dark matter. Golwala uses Bolocam—a camera built with Andrew Lange, JPL scientists Jamie Bock and Hien Nguyen, and Jason Glenn of the University of Colorado—at the Caltech Submillimeter Observatory to study tiny fluctuations in the cosmic microwave background (CMB) caused by galaxy clusters. Golwala is also developing a new camera called MUSIC with Professor of Physics Jonas Zmuidzinas (BS ’81), Nguyen, Glenn, JPL postdoc Jack Sayers (MS ’04, PhD ’08), and JPL scientists Peter Day (PhD ’93) and Rick LeDuc. Golwala also plays a role in two more experiments that analyze the CMB. Spider, on which postdoc Jeff Filippini is a team member, is a balloon experiment that will fly from Antarctica and observe at millimeter wavelengths. BICEP2, on which postdoc Walt Ogburn is a team member, is a sister experiment situated at the South Pole.

in huge dark-matter halos several times the pass through each other like ghosts. When Deuterium, the next heaviest hydrogen iso- size of galaxies themselves. According to astronomers measured how light bends tope, has a proton and a neutron, and tritium the latest estimates, about 85 percent of around the Bullet Cluster, an effect called has a proton and two neutrons. The total all the matter in the universe is dark. (And gravitational lensing, they found that most density of protons and neutrons dictated the it turns out that most of the cosmos isn’t of the mass is not in the hot ball of colliding isotopes produced and in what proportions. even matter—three-fourths is dark energy, gas, but in the places where the dark matter This primordial process is the only way to an entirely different beast altogether and an would be. For the first time, astronomers make deuterium, so by measuring the abun- even bigger mystery.) had isolated dark matter from visible matter. dance of deuterium now, we can gauge the Physicists and astronomers have come Some hypotheses say that dark matter is total density of protons and neutrons then. up with no end of ideas to account for the composed of familiar-but-dim objects, like It turns out that there aren’t enough protons universe’s invisible mass. Some astrono- black holes or brown dwarfs—Jupiter-sized and neutrons—collectively called baryons— mers have even proposed that our theory balls of gas too small to form stars. These to account for all the mass in the universe. of gravity is incomplete, that it behaves things have been dubbed massive compact The cosmic microwave background differently on cosmic scales. But in 2006, halo objects (MACHOs—since they’re (CMB), the sky-filling afterglow of the Big observations of the Bullet Cluster, published obviously not WIMPs), but there doesn’t Bang, also suggests that dark matter must by a team of astronomers that included lead seem to be enough of them out there. Two be different—or nonbaryonic, in physics author Douglas Clowe (BS ’93), Anthony lines of evidence from the Big Bang have nomenclature. Observations of the CMB— Gonzales (BS ’95), and Dennis Zaritsky now convinced most astronomers that dark including those from BOOMERanG, the (BS ’86), made a convincing case that dark matter isn’t normal stuff, made from protons Antarctic balloon telescope experiment led matter is indeed real. (Clowe, Gonzales, and and neutrons, but something completely by the late Goldberger Professor of Physics Stephen Murray [PhD ’71] were also among different. Andrew Lange—show a stipple that betrays the authors of a lead-up study on the Bullet Right after the Big Bang, the cosmos the universe’s baryonic density. (See E&S Cluster’s dark matter in 2004.) was a soup of hot plasma. As the universe 2000, No. 3). “You’re looking back in time Most of the visible matter in galaxy clus- expanded and cooled from 1032 degrees to to a plasma a few hundred thousand years ters consists of X-ray-emitting hot gas, and a balmy 109 degrees, protons and neutrons after the Big Bang,” Filippini explains. “You the Bullet is actually two galaxy clusters in formed. In this fiery cosmic cauldron, nuclear can think of it as a snapshot of the way this the midst of a high-speed crash. When two fusion took place as protons and neutrons plasma is frothing and sloshing from place clusters smack into each other, the gas slammed into one another, forging the to place.” The normal stuff—baryons—is collides, and the clusters slow each first elements—hydrogen, helium, being pushed around by the energy in this other down. Dark matter, on the and lithium—and some of their cosmic soup, adds Walt Ogburn (BS ’99), a other hand, hardly interacts isotopes, which have vary- postdoc who was also on the CDMS team. with anything, so the ing numbers of neutrons. “If you have some kind of nonbaryonic dark two giant blobs of For example, the matter that’s not interacting with the pro- dark matter nucleus of ordi- tons, then it does its own thing and interacts would nary hydrogen gravitationally with the soup.” So how much is a bare the plasma sloshes—its amplitude, which is proton. represented by the prominence of the spots

16 ENGINEERING & SCIENCE spring 2010 A composite image of the Bullet Cluster, which is two galaxy clusters that began colliding about 3.5 billion years ago. The red represents each cluster’s X-ray-emitting gas, which was slowed down by the collision and remains near the crash site. The blue represents the dark matter. Since dark matter hardly interacts with anything, it sailed right on through the other cluster unimpeded, and it’s still going. in the CMB—tells astronomers what kinds of ingredients are in the soup . . . and a lot seems to be that strange, nonbaryonic dark matter. Theorists have concocted a smorgasbord of potential dark-matter candidates, exotic particles with names like Q-balls, cryptons, and WIMPzillas. But so far, the particle that is considered to be the most credible candi- cooked up several theories. One of these “There’s no reason this coincidence had to date is the WIMP. “In my mind, it’s the best is supersymmetry, in which every particle happen, and therefore we should take it se- option,” says Golwala. A WIMP is about a has a new partner—for instance, a quark’s riously.” Whereas many other potential dark- hundred times more massive than a proton “superpartner” is a squark. Another group matter particles are pure invention, WIMPs and interacts with other particles through of theories posits that there are more spatial are born naturally out of particle physics. “It gravity and the weak force (which are two dimensions than the three we normally doesn’t rely on one specific theory of how of the four fundamental forces of nature— experience—up, down; left, right; forward, things work—that’s the most compelling the others being the electromagnetic force, backward. The extra dimensions are curled thing about it.” which is actually the same as the weak up into sizes too small for us to see. The How exactly a WIMP behaves—how force at sufficiently higher energies, and the theories of supersymmetry and extra dimen- frequently it would smash into the nuclei in strong force, which holds nuclei together). sions each suggest that there could be par- CDMS’s detectors, for instance—depends What distinguishes WIMPs as good candi- ticles with the general properties of WIMPs. on the specific variation on supersymmetry dates is that they’re really not special at all, If there are WIMPs, they would’ve been or extra-dimension theory, and there are but a type of particle predicted by many of produced in the early universe, back when it many. “For every theorist who’s thought the latest theories in particle physics. was so dense that despite their weak inter- about it, there’s going to be a slightly dif- The Standard Model, which explains how actions, they would’ve been dashing around, ferent theory,” Filippini says. Still, most of all the fundamental particles interact with smashing into each other. When they did, these ideas require, or are consistent with, one another, is one of the most successful they would have annihilated one another, WIMPs. “A particle physicist and a cosmol- theories in physics. But it’s incomplete— bursting into other particles and energy. But ogist would both think this kind of particle for example, it doesn’t explain gravity. It as the universe expanded and became less should exist,” he remarks. “That’s such a cool also turns out that if you were to solve the dense, it became harder for WIMPs to find coincidence that it makes people actually relevant equations, you would find that every and annihilate each other, and they soon want to look for these things.” particle—except the photon or gluon, anoth- were no longer able to, leaving a bunch of er massless particle—should have a mass of them hanging around with nowhere to go. 1019 gigaelectron volts, the so-called Planck Calculations predict that the quantity of Such a WIMP mass. Instead, a proton has a mass of about leftover WIMPs is roughly the same as the The Soudan iron mine opened in 1884 100 gigaelectron volts—and the Standard estimated amount of dark matter. “It’s just and would become Minnesota’s deepest, Model can’t account for the enormous gap. a complete coincidence—and that is what as well as oldest, mine, helping the state To fix the problem, physicists have got people really excited,” Golwala says. become the nation’s leader in producing iron

Far left: David Moore, Golwala, Jeff Filippini, and Zeeshan Ahmed in front of the Cahill Center for Astronomy and Astrophysics.

spring 2010 ENGINEERING & SCIENCE 17 ore. It closed in 1962, and is now a state cosmic-ray strike anywhere in the instrument crystal. On the detector’s surface are thin park as well as a physics lab, home also would trigger a cascade of other particles aluminum films, which act as phonon- to a neutrino-detection experiment called such as neutrons, which could mimic a collecting antennas. When aluminum is as MINOS. Most flights take the researchers to WIMP’s signal if they hit the detector. cold as these detectors are, its electrons— Minneapolis, where they then drive for four Developed by groups at Stanford and UC which freely roam in the metal—arrange hours to Soudan—a journey made perilous Berkeley in the 1990s, led by Blas Cabrera themselves in so-called Cooper pairs. These by Minnesota winters. “California drivers and Bernard Sadoulet, respectively, the pairs form when a free electron tugs at the don’t do well in the snow,” says graduate detectors are composed of 230-gram ger- positively charged lattice of the aluminum student Moore, who, along with his fellow manium crystals. The experiment’s detection nuclei, creating a tiny ripple in the crystal grad Ahmed, has spent plenty of time in method requires a semiconductor, and of structure. The displacement caused by that the mine. “Sometimes with the wind chill, them all, germanium has the heaviest nuclei, ripple then pulls at another free electron, it’s minus 40 degrees—it’s kind of crazy,” maximizing the possibility of a collision. The coupling it with the first. The partnership adds Ahmed, who played a leading role in crystals are arranged into five stacks of six is weak, so the phonons from the crashing the latest analysis. Researchers have spun detectors each, and then the stacks are WIMP easily split the pairs, transferring their off roads and driven into ditches; they’ve hit encased in copper, chosen for its low ra- energy as heat to the newly single electrons. deer and, reportedly, grazed a bald eagle. dioactivity. The stacks are kept at a frigid 40 The electrons then find their way toward If they make it to Soudan in one piece, millikelvins—that’s 0.04 degrees above ab- tungsten thermometers. they stay for a week or two at a time in solute zero, almost a hundred times colder The tungsten is a transition-edge a gray, four-bedroom house five minutes than outer space—and surrounded by layers superconductor, meaning that any bump from the mine. Every morning at 7:30, they of polyethylene, for more neutron protec- in temperature would nudge it out of its squeeze into the mine shaft’s rusty cage, tion, and lead, to shield against gamma rays. superconducting state. The electrons—now where an operator lowers them into the lab. Because the lead will inevitably have some They don’t return to the surface until 5:30 radioactive isotopes, spouting the occa- in the evening. The lab is within a four-story sional particle, the detectors are surrounded high cavern, and with walls, bright lights, a by ancient lead taken from an old sunken steady temperature of 70 degrees, and even ship. Two centuries old, this lead has already a ping-pong table, it isn’t too much different decayed away most of its radioactivity. Sur- from a lab above ground. rounding the apparatus is something called Because they’re waiting for a single a plastic scintillator anticoincidence detec- particle among billions, at a rate of just a tor, a shield built by UC Santa Barbara that few potential hits per year with the current lights up when muons pass through, notify- detectors, the team must minimize as much ing researchers of events to ignore. Finally, background noise as possible. With almost the whole thing sits in a clean room. 700 meters of dirt and rock above, CDMS When a WIMP smacks into the disk- is protected from cosmic rays, which are not shaped detector, it rattles a germanium actually rays but primarily particles—protons, nucleus, sending phonons—quantized helium nuclei, electrons, and muons. A packets of vibrational energy—across the

Top right: A top view of the five stacks that make up the CDMS experiment. Copper provides a first layer of protection against background particles. Bottom right: This model of the apparatus provides a multisensory demonstration of CDMS in action. It lights up and chimes when it detects particles.

18 ENGINEERING & SCIENCE spring 2010 Far left: The Soudan mineshaft in winter. Middle: The park’s mine tour includes a cart ride. Left: Inside the CDMS lab, where Rupak Mahapatra (left), now on the faculty at Texas A&M University, and Xinjie Qiu (right), now a Stanford postdoc at Fermilab, take a break from dark-matter hunting with a game of ping pong. Below: Dan Bauer, from Fermilab and project manager for CDMS, removes one of the towers to make room for the first of the next generation of towers, called SuperCDMS. With him are Jim Beaty of the University of Minnesota (center) and Steve Leman of MIT. Below right: Walt Ogburn is in the “cryopad,” the area that houses the liquid-helium and nitrogen dewars and pumps. The computer controls the circulation of the liquids that keep the detectors frigid.

warm from the phonons—push the tung- field in the crystal corrals the free charges er any fit the criteria, revealing whether any sten toward being a normal conductor. By onto the detector’s rear surface (oppo- of the elusive particles have been caught. measuring the sudden changes in electrical site the tungsten and aluminum), where current, the researchers can determine the electrodes measure the ionization energy. energy of the WIMP-generated phonons— By calculating the ratio of ionization to total Finding Dark Matter and therefore the energy of the collision. collision energy, the researchers can identify WIMP fever was running high on De- To be sure, one of the biggest challenges which events were electron collisions and cember 17, when physicists packed into for the CDMS team is to minimize and discard them. auditoriums in California and Illinois to hear understand background particles. Despite The team employs several more tricks to what, if anything, CDMS had discovered. the mine’s depth, the shielding, and the ef- distinguish WIMPs from background par- The Economist had written on that day, “If fort to keep materials clean and pure, stray ticles. They only look at the signals from the the rumors are true, a solution to one of particles still slip through. Electrons from most probable energy range for WIMP col- the great problems of physics may now be the beta decay of radioactive elements that lisions. They discount anything that happens within reach.” JoAnne Hewett, a particle found their way onto the detector surfaces, at the edges of the detector, where signals physicist at SLAC, even liveblogged the and high-energy photons called gamma rays can get fuzzy. They only consider single- event on Cosmic Variance. “The excitement from impurities in the apparatus, both knock collision events—since a WIMP interacts in the air is palpable,” she wrote. “Not much electrons loose in the germanium. To distin- so weakly, it would only hit one atom in one work is being done—everyone is pretty guish these events from the WIMP-nucleus detector. If the researchers see some par- much talking in the hallways, trying to pass collisions investigators are looking for, the ticle plowing into multiple nuclei down the the time until 2:00.” detector collects and counts the free charge whole stack, they know it can’t be a WIMP. Finally, the results were announced— created by the loose electrons. An electric The team also throws away any collisions two events had been found! The team had that occur at the crystal’s surface, as those actually unblinded the data set a month tend to be caused by stray particles from before, while Golwala was at the Caltech the apparatus itself—no matter how pure Submillimeter Observatory on Mauna Kea your materials, some radioactive isotopes in Hawaii, where he was doing astronomical are bound to sneak in. observations with Bolocam for another proj- To prevent bias on their part—or maybe ect. “I was in my dorm room recovering from wishful thinking—the team uses the above altitude sickness that morning,” he recalls. “I requirements to decide on what consti- was floating in and out of the teleconference tutes a signal before they look at the data. as the unblinding was happening. I was lying Calibrating with particles generated from on the bed and heard something about two radioactive sources, the team determines events. I thought, ‘Uh oh. What did we do a set of criteria that allows a reasonable wrong?’” number of WIMPs, but rejects virtually all After checking the data and instruments, background particles. Finally, they “unblind” the team concluded that nothing had gone their data by looking at it and seeing wheth- wrong. These two collisions were real. But,

spring 2010 ENGINEERING & SCIENCE 19 The barn was born during the Manhattan Project, when American physicists needed a shorter way of saying 10-24 square centimeters, which is about the size of a uranium nucleus. In the July 1972 issue of Physics Today, Marshall Holloway and Charles Baker re- count how they coined the term in 1942, after mulling over other candidates like “Oppenheimer” or “Bethe,” in honor of the physicists who made the atomic bomb possible. But “Oppenheimer” was too long and “Bethe” was too similar to the Greek letter. In no small part because of where they both were from—Purdue University in Indiana, surrounded by farmland—they decided on the “barn,” since a uranium nucleus was as “big as a barn.” For security reasons, physicists also wanted to avoid discussing overtly technical terms on the phone. The U.S. government ended up classifying the term anyway, and didn’t declassify it until 1948. “Zepto” is the prefix for 10-21, and so a zeptobarn is 10-45 square centimeters.

before booking flights to Stockholm, they to surpass CDMS’s results with new data built about 2.5 kilometers underground in calculated that there was a 23 percent soon. In the next couple of years, half a South Dakota. The experiment won’t happen chance these signals were caused by dozen more projects will begin—and they’ll until 2017 at the earliest, but these are the background—likely electron recoils that had be many times more sensitive than CDMS. kinds of projects that will increase WIMP snuck past their set of criteria. As Golwala One of those is the next generation of sensitivities by a hundred times over the points out, “No one claims discovery with CDMS, called SuperCDMS. next decade, making physicists optimistic that high of a chance.” The team couldn’t say The bigger a detector’s crystal, the better about the future. “It’s very possible that in they had discovered dark matter, but they the chance a WIMP will hit it. Boasting five the next five years we might be talking about couldn’t rule it out, either. “supertowers” with a total of 15 kilograms of WIMP astronomy, rather than just trying to So CDMS hasn’t revolutionized our germanium, SuperCDMS will improve that detect something,” says Filippini. understanding of the universe—yet. As for chance about fourfold. The team will install Now that the latest data run is complete, all the hype? “In a couple of months, no these new detectors in Soudan as early as the CDMS team is turning toward making better detectors. For example, Moore is working on a new sensor design with improved resolution. Based on technol- “It’s very possible that in the next five years we might ogy developed by Jonas Zmuidzinas (BS be talking about WIMP astronomy, rather than just ’81), professor of physics, these so-called trying to detect something.” microwave kinetic inductance detectors would allow researchers to better pinpoint a WIMP’s crash site. Ahmed, whose disserta- tion will include a lot of the latest results, is one will remember this,” Golwala says. Still, this year. “The experiment could legitimately now designing a device to measure minus- their results—published in the March 26 detect something in the next few years,” cule amounts of radioactivity. As detectors issue of Science—are noteworthy, placing Filippini remarks. become more sensitive, materials need to the most stringent constraints yet again on A couple more years down the line, the be even cleaner, with as little radioactivity what WIMPs could be. “It’s an exciting time team will build even larger detectors—70 as possible. Current instruments, however, in the bigger sense, because we’ve been units for a total of 105 kilograms. With a can’t detect such low levels. producing results from this experiment for twentyfold boost in sensitivity, the team will WIMP-search experiments are approach- about five years,” he says. “We’ve been the need to move SuperCDMS to SNOLAB, ing what physicists call the zeptobarn scale premier experiment in this field.” a mine near Sudbury, Ontario, just north of (see box). The zeptobarn, which is 10-45 These data sets are marking the end of Lake Huron. At around two kilometers in square centimeters, is a unit of cross sec- the current chapter in dark-matter searches. depth, it’s the world’s deepest underground tion, the measure of how frequently a par- But new experiments are already under way, laboratory. ticle interacts. The larger the cross section, including a rival project called XENON100, If that weren’t enough, researchers hope the more likely those particles will collide. which uses liquid xenon in lieu of germa- to bury 1.5 tons of germanium in DUSEL, The latest CDMS results rule out WIMPs nium crystals. Situated in Italy’s Gran Sasso the Deep Underground Science and En- with cross sections bigger than about 10-44 National Laboratory, the experiment is likely gineering Laboratory, which is slated to be square centimeters, or 10 zeptobarns. If

20 ENGINEERING & SCIENCE spring 2010 WIMPs have cross sections bigger than Eureka! . . . or Not this limit, then CDMS should’ve already There have been plenty of tantalizing “discoveries” of dark matter. One experi- detected them. ment in Italy, called DAMA, has claimed to see dark matter not once, but many It turns out that the most compelling times. DAMA, like CDMS, tries to measure WIMP collisions. Instead of germa- theories of supersymmetry or extra dimen- nium, DAMA has sodium iodide crystals that glow when struck by particles, sions predict WIMPs to have cross sections and instead of picking out individual WIMPs, it tries to measure the periodic of around a zeptobarn. Because the next ups and downs of collisions as Earth sails through the dark matter in our solar generation of WIMP detectors will be sensi- system. This annual modulation, as it’s called, could be a signature for WIMPs. tive to these kinds of particles, they could be The group said they detected this signature first in 1997 and that they con- discovered in the coming decade, Golwala firmed the signal multiple times, in 2000, 2004, and 2008. says. But below a zeptobarn, things get The problem, though, is that no one else has been able to confirm the weird. If you don’t see anything by a zepto- results. You have to be sure the signal isn’t coming from another, non-WIMP barn, “you’re excluding the most reasonable source, such as an exploding star or just some artifact of your technique. models,” he says. “But no one said nature “Over the last decade, DAMA has managed to slowly knock down all the had to be reasonable.” At cross sections of objections I have had to their result, in terms of possible non-dark-matter hundredths or thousandths of a zeptobarn, explanations,” Golwala says. “But that doesn’t mean that all non-dark-matter the likelihood of dark matter being WIMPs options have been exhausted. And they have never been particularly forth- drops, and with limited funding, it might not coming with their data, in spite of the extraordinary claim to have detected be worthwhile to continue the search. dark matter. Again, there is no way to conclusively prove it, but I tend to think Regardless of when scientists discover it’s not dark matter.” dark-matter particles, it likely won’t happen Another way to detect WIMPs is to see them annihilate one another in with a single, big announcement in one- space. In the moments following the Big Bang, pairs of WIMPs—if WIMPs inch headlines—despite what rumors might actually exist—would have been smashing into each other, igniting bursts of lead you to believe. The search is a slow, energy and showers of particles. Even though they wouldn’t be crashing as systematic sweep of the possible identities frequently now as they did then, they would still manage to find one another of dark matter, and any finding will have to once in a while. In 2008, a satellite called PAMELA (Payload for Antimatter withstand close scrutiny and be confirmed Matter Exploration and Light-nuclei Astrophysics) and a balloon over Ant- by multiple experiments. arctica, ATIC (Advanced Thin Ionization Calorimeter), found complementary Direct-detection experiments like CDMS data that could be explained by WIMP annihilations. ATIC detected an excess are just one of three roads to determining of high-energy electrons, while PAMELA found extra high-energy positrons, whether dark matter is made of WIMPs. or antielectrons. Last year, Fermi, NASA’s gamma-ray space telescope, also Physicists are trying to detect WIMPs indi- detected a subtle boost in electrons. rectly with satellites like Fermi, which aims But these extra particles could also have come from other sources, such as to measure the gamma rays that are shot a type of dense rotating star called a pulsar. Golwala remains skeptical. “It’s out from WIMP-WIMP annihilations. They not possible to say this conclusively, but there have been too many of these also hope to make WIMPs from scratch at signals that appear and then can be explained astrophysically to put much the Large Hadron Collider in Geneva, which stock in any of them.” All the different data don’t fit together in a coherent will be able to recreate the conditions of the picture, adds Moore. “There’s no real smoking gun for any of those detections universe moments after the Big Bang. These yet.” experiments represent a confluence of esoteric theory and tangible hardware that may soon solve one of the great mysteries Sunil Golwala received his BA in phys- sistant Professor of Physics in 2003, and of nature. “It’s amazing that we can even ask ics from the University of Chicago in Associate Professor of Physics in 2010. the questions, what is the universe made 1993. He received his MA in 1995 and His CDMS work is supported by the of, and why it’s made of that stuff,” Golwala PhD in 2000 from UC Berkeley, where he Department of Energy (DOE). The current says. “And it’s even more amazing that we worked on the previous generation of the CDMS experiment was built by the DOE can attempt to answer them.” CDMS experiment with Bernard Sadoulet. and the National Science Foundation. He came to Caltech in 2000 as a Millikan Postdoctoral Scholar, became an As-

spring 2010 ENGINEERING & SCIENCE 21 Cassini’s Ringside Seat

Saturn’s rings and moons are a model for how planetary systems may be forming around nearby stars. The Cassini spacecraft has been taking a close look at the rings of late, and it turns out that gravity and granules with momentum can do some amazing things.

Of all the planets in our solar system, only one’s got a whole lot of bling. Yet, for most of human history, we didn’t know about Saturn’s luminous rings, even though the gigantic, gaseous planet is readily visible in the night sky. When Galileo became the first person to peer at Saturn through a telescope in 1610, he sketched a compan- ion moon on each side. A couple of years later, he became utterly perplexed when the moons vanished. When the mysterious objects returned, he saw them as elliptical arms resembling handles. In 1655, Dutch astronomer Christiaan Huygens, using a better telescope, figured out that the arms were actually a flat ring and that, like the edge of a sheet of paper held horizontally at eye level, it disappeared from view when Saturn’s tilt presented it to us edge-on. (The rings truly are paper-thin—a mere 10 to 20 meters thick on average, and yet so broad that they would fit neatly between Earth and

22 ENGINEERING & SCIENCE spring 2010 Opposite: Linda Spilker, Cassini’s project scientist, with one of the Deep Space Network’s 34-meter dishes at Goldstone, California. Right: The Cassini orbiter. Below: The anatomy of Saturn’s rings. The planet itself is out of view to the left.

By Linda Doran

its moon.) Saturn’s rotational axis is tilted by 4,800 kilometers wide, demonstrated that surface of smog-shrouded Titan. Discovered about 27 degrees, and the rings, which gir- Saturn’s rings were not a single, solid by Christiaan Huygens in 1655 and bigger dle Saturn’s equator, are tilted by the same object. Quite perceptively, Cassini theorized than the planet Mercury, Titan is the second- amount. Seasons change slowly on Saturn, that they were swarms of tiny moonlets too largest moon in the solar system and has an with its nearly 30-year orbit; but once every small to be seen from afar. Earth-like surface with highlands, lowlands, 14 to 15 years, when the sun straddles Sat- Three centuries later, spacefaring vehicles stream channels, and liquid methane–filled urn’s equator at the solar equinox, the rings carried telescopic cameras to Saturn itself. lakes. Huygens sent pictures and data back vanish from view—they’re pointed directly at Our first emissary, NASA’s Pioneer 11, re- to Cassini during the entire journey down, the sun and more or less at us. vealed in 1979 that the Cassini Division was and then continued to transmit information Seen through the 60-inch telescope at not actually empty, but was merely a zone from the surface for some 70 minutes, until Caltech’s Palomar Observatory, Saturn’s with a very low density of ring material. JPL’s the mother ship flew out of range. Mean- rings beckon like diamond-paved, circular twin Voyager spacecraft, flying past Saturn while, Cassini had settled into its four-year highways. (Every now and then, the Friends in 1980 and 1981, showed that the rings primary mission—a grand tour of a planet of Palomar Observatory get a tour that, are astonishingly complex. Astronomers and that many scientists regard as a miniature weather permitting, includes an opportunity stargazers alike marveled at images of gaps solar system in its own right. A team of to turn that telescope on the heavens.) They swept clean by tiny moons, mysterious dark navigators guides Cassini on the multiple nearly fill the telescope’s field of view, and spokes whirling within the B ring as if it were loops of its itinerary, firing rocket thrusters seem as close as the moon does in a pair a wagon wheel, and the “braided” multiple to fine-tune its orbit and using the gravita- of good binoculars. In reality, they’re more strands of the F ring. The Voyagers showed tional pull of Titan like a slingshot to adjust than a billion kilometers away. If you were to that Saturn’s rings were really thousands of the spacecraft’s path. Cassini’s travels have count those kilometers at one per second, discrete ringlets, and confirmed that they taken it on equatorial orbits in the plane of you’d still be counting three decades from are very nearly pure—99 percent—water ice. Saturn’s rings and moons, as well as on now—an entire Saturnian year later. Put an- (We’ll come back to that other 1 percent steeply inclined orbits above and below the other way, Saturn is about 10 times farther later.) The Voyagers also determined that plane in order to make observations from as away from the sun than Earth. Understand- the ring particles range from flecks the size many angles as possible. And Cassini will ing something so remote is a challenge not of dust, about a millionth of a meter across, be going strong for years to come—in Janu- only for parents with curious children—at to chunks of ice as big as two-story houses. ary 2010, NASA authorized a seven-year Palomar and elsewhere—but for scientists Most of them, however, run from pebbles to mission extension to follow Saturn through a as well. To this day, we’re not even sure snowballs. complete change of seasons. how old the rings are or exactly what they’re On June 30, 2004, a schoolbus-sized made of. spacecraft named Cassini, a joint U.S.-Eu- Our first clue came in 1675, when ropean mission built and operated by the Jet Plunging into the Dark Giovanni Domenico Cassini, the founding Propulsion Laboratory, which Caltech man- Which brings us back to the solar equi- director of the Paris Observatory under ages for NASA, entered orbit around Saturn. nox. In August 2009, for the first time in Louis XIV, observed not one bright ring but Six months later, on December 25, Cassini history, astronomers got a close-up look at two, separated by a dark gap. The discovery released the European Space Agency’s Saturn’s rings shrouded in darkness, just of what’s now called the Cassini Division, Huygens probe, which parachuted to the as they were when Galileo couldn’t see

spring 2010 ENGINEERING & SCIENCE 23 The gravitational pull of a tiny moon named Daphnis, only eight kilometers in diameter, sweeps stray material out of the 42-kilometer- wide Keeler Gap near the outer edge of the A ring. Cassini discovered that Daphnis’s gravity also kicks up bright, white rooster tails in the ring material.

them. One of those astronomers was JPL’s Linda Spilker, Cassini’s project scientist. “At equinox, the sun is essentially edge-on to Saturn’s rings,” she says. “This is the time when the ring temperature is at its very coldest—down to 43 kelvins (–230 degrees C). In a sense, it’s like experimentally turning off the sun. The rings are heated only by Later, she worked on the Voyager mission, tions as a result of data collected during the Saturn’s thermal radiation and by sunlight and she likes to tell her daughters that their equinox. In addition to her management role reflected from Saturn.” When Cassini arrived births were literally based on the alignment coordinating Cassini’s many investigations, at Saturn, the rings averaged 90 to 110 K of the planets. The girls were born between Spilker is coinvestigator on an instrument (–183 degrees C to –163 degrees C). Dur- Voyager 2’s flybys of Saturn in 1981 and called the Composite Infrared Spectrometer, ing equinox, a 60-degree temperature drop Uranus in 1986—in other words, during the or CIRS. Humans sense infrared light as happened overnight, as it were. “cruise phase” when there wasn’t so much heat. Like a snake’s tongue, CIRS seeks out Spilker belongs to that generation of going on. sources of heat, experiencing the world at space explorers who spent July 20, 1969, in Spilker’s fascination with destinations wavelengths longer than those the human wide-eyed wonder in front of their television beyond Earth is a constant in her life. “If you eye can see. As it distinguishes variations in infrared light, CIRS can map the temperature and composition of Saturn’s rings because different materials emit and absorb Spilker belongs to that generation of space explorers heat in characteristic ways. who spent July 20, 1969, in wide-eyed wonder in front “I think that the outsides of Saturn’s ring particles would certainly be very fluffy and of their television sets watching Neil Armstrong set foot porous,” Spilker says. “That’s based on the on the moon. fact that they heat up and cool down very quickly, within about half an hour. They have what we call a very low thermal inertia—in fact, it’s four orders of magnitude lower than sets watching Neil Armstrong set foot on could scoop up a particle out of Saturn’s that of a solid block of ice.” Rapid heating the moon. While an undergrad majoring in rings,” she asks, “and hold it in your hand, and cooling implies a lot of surface area in physics at Cal State Fullerton, she studied what would it look like? Would you see this order to exchange so much thermal energy other space rocks at Caltech—helping her fluffy snowball? Would it have an icy core? with the surroundings so fast. That’s why physics professor, Dorothy “Dotty” Woolum, If you could take your knapsack and collect crushed ice in a glass of soda melts faster who had been a postdoc at Caltech in the particles as you collect shells on a beach, than cubed ice. ’70s, and Don Burnett, Caltech professor would you find different kinds of particles in For about four days, the rings were of nuclear geophysics, analyze the distri- each of the rings? I like being an explorer.” cloaked in shadow. Indeed, sunlight on bution of bismuth and lead in meteorites. She has a partial answer to those ques- Saturn’s rings during the equinox is so dim

24 ENGINEERING & SCIENCE spring 2010 that the rings are nearly invisible in the raw camera images, even those taken from overhead. The only reason we can see them is that scientists led by Carolyn Porco (MS ’79, PhD ’83) digitally enhanced the Planetary scientist Carolyn Porco did her for viewing at JPL’s Cassini website and shots from Cassini’s wide-angle camera. PhD thesis on Voyager’s observations at the website of SSI’s Cassini Imag- (Porco, too, worked on the Voyager mission, of Saturn’s rings and spokes. She now ing Central Laboratory for Operations and is now also working on NASA’s New leads Cassini’s imaging team, which is (CICLOPS), where the images are pro- Horizons mission en route to Pluto.) To bring headquartered at the Space Science cessed. The images paint a portrait of a the rings to life, she and her colleagues Institute (SSI) in Boulder, Colorado. In planetary system that, in Porco’s words, at the Cassini Imaging Central Laboratory 2009, she received the most prestigious “is so alien in comparison to Earth that for Operations (CICLOPS) at the Space award in scientific photography, the Len- we might as well have visited a planet Science Institute in Boulder, Colorado, nart Nilsson Award, “for combining the around another star in another quadrant increased the brightness of the dark half finest techniques of planetary exploration of the galaxy. When this mission is over, of the rings by a factor of three relative to and scientific research with aesthetic we will leave behind a stunning visual the half illuminated by Saturn-shine, and finesse and educational talent.” legacy, and a body of work that will bumped up the brightness of the entire ring “From the very beginning of the guide future explorers, both robotic and system by a factor of 20. Cassini mission,” says Porco, “I took it as human, in their excursions around the The photos revealed walls of icy rubble as my personal goal that we on the imaging Saturn system.” tall as mountain peaks rising out of the ring team would be the planetary equivalent A much sought-after speaker and Carl plane. “It’s like standing outside right before of nature photographers. We would try Sagan protégée who served as a the sun sets,” notes Spilker. “Your shadow to capture, whenever possible, images consultant on the movie Contact, she gets very long. Anything that’s a little bit and movies whose primary purpose went also consulted with Industrial Light and bigger, or sticks up, casts shadows on the beyond science and conveyed the sheer Magic’s visual-effects supervisor Roger part of the ring that’s behind it. We see magnificence and otherworldly beauty to Guyett on the recent Star Trek reboot. what look like towering mountains, in some be found around Saturn.” This picture, taken at ILM, shows her with cases as high as four kilometers, created by The vast library of aesthetically striking some familiar faces from yet another the particles that Saturn’s moon Daphnis is images taken during equinox as well blockbuster franchise. (Photo courtesy of pulling out of the ring plane.” This happens as throughout the mission is available Carolyn Porco and ILM.) because Daphnis’s orbit is slightly tilted with respect to the rings. When Daphnis crosses the ring plane, it drags some of the particles region in one of Saturn’s rings. Orbital reso- relative to their fellows. These density waves out of the plane. Like the rooster tail behind nance brings two bodies back to the same are analogous to the arms of a spiral galaxy a speedboat at full throttle, this disturbance region of space over and over again—for such as our own Milky Way, says Porco. quickly subsides once Daphnis passes by. example, if the ring particles travel around “They’re much more tightly wound than the If you were a pilot flying above the great Saturn exactly twice for every orbit of the spiral structures that you see in galaxies, plains that are Saturn’s rings, it would be a moon, they have a 2-to-1 resonance. The but they are in fact the same creatures. The good idea to keep plenty of altitude over the repeated gravitational tug at that location physics behind them is the same.” landscape below, lest you crash into an icy causes the ring particles to crowd together Spiral density waves are compressional, curtain as tall as the Rockies. into a coherent spiral structure one or two existing only in the two dimensions of the As you kept an eye out for ice peaks, you kilometers wide. The resulting density wave ring plane. But if the orbit of the resonant might also notice regularly spaced basins sweeps through the ring material. As the moon is slightly askew, yet another set of and ranges. Close analysis of the Voyager crest overtakes the ring particles in their or- waves forms—this time, in three dimen- pictures had revealed spiral density waves in bits, they get pushed together by the wave; sions. The orbit of Saturn’s moon Mimas the rings. These waves occur wherever one once the wave passes, the trough pulls the also takes it above and below the ring plane, of Saturn’s moons is in resonance with a particles back into their original positions pulling ring particles out of the plane along

spring 2010 ENGINEERING & SCIENCE 25 with it. Orbital resonance then keeps these Blinking in the Sunlight particles’ orbits tilted relative to the plane, If, on a bright sunny day (you certainly resulting in a towering spiral wave called a wouldn’t want to be using visual flight rules bending wave that wraps all the way around during equinox), you were to buzz down be- the ring. The density waves and the bending tween the A and B rings through the Cassini waves propagate in opposite directions, Division, you would find that the rings’ unlit says Spilker, adding a further level of com- side is not completely dark. Instead of be- plexity to the rings’ structure. ing continuous, each of Saturn’s ringlets And now Cassini’s equinox pictures show is like a dashed line of clouds. “As we fly yet another kind of three-dimensional wave, underneath,” says University of Colorado at heretofore undiscovered. These waves only Boulder planetary scientist Larry Esposito, rise about 100 meters above the ring plane, “we see flashes of sunlight come through yet during the equinoctial photo session the gaps. We originally thought we would they too remained illuminated after the rings see a uniform cloud of particles. Instead we were plunged into darkness. “From high find that the particles are clumped together above, the rippled surface of Saturn’s D ring with empty spaces in between.” Esposito looks like a corrugated roof,” says Spilker. is principal investigator for the Ultraviolet “The ripple extends for more than 17,000 Imaging Spectrograph, or UVIS, which sees kilometers across the ring system. When the light at wavelengths shorter than we can Voyagers flew by, it wasn’t there. Some- perceive—ultraviolet rays are what burn your thing caused this wonderful rippling to get skin if you forget the SPF 30. Esposito’s started, and then it expanded outward all team measured ultraviolet emissions from a the way into the C ring. Based on computer distant star, Alpha Arae, which waxed and models of the ripple’s expansion, we can waned as the rings passed in front of it. run the clock backward, and we think that it Cassini confirmed the presence of the began some time in the early to mid-1980s.” taffylike clumps of material by sending radio Scientists remain baffled about its origin, signals through the rings to Earth. These but one possible scenario has a meteoroid signals—with a strength of less than a slamming into the rings—the pebble in the billionth of a watt by the time they arrived pond, as it were. Like the waves the Voyag- here!—were picked up by the ultrasensitive ers discovered, this ripple is also a tightly receivers of JPL’s Deep Space Network, wound spiral. In fact, it has become more which has stations in California, Spain, and tightly wound over time. In Hubble Space Australia in order to ensure that one set of Telescope images taken before Cassini’s ears is above the horizon at all times. Be- discovery and since reexamined, the ripples sides communicating with NASA’s robotic were farther apart. explorers (and spacecraft of other nations Cassini also got a close-up look at the as well), the 70- and 34-meter dishes are mysterious “spokes” in Saturn’s rings first also used for radio-science experiments. In observed by the Voyagers. These spokes this case, the ring particles absorbed and may be the result of meteoroids hitting the scattered the radio waves in a manner that ring and generating charged particles, says revealed their size and distribution. Porco. The particles levitate, in the same The clumps, called self-gravity wakes, are manner that a comb rubbed vigorously on a special case of the spiral density waves a wool sweater will cause your hair to rise mentioned earlier, which Cassini has now if held near your head, until eventually they shown to exist all over Saturn’s rings. “We lose their charges and settle back into the first saw self-gravity wakes in the A ring, ring plane. The low-angled sunlight at equi- which is less dense, so they are farther apart nox caught the high-flying spokes, setting and easier to see. They just leaped out at us,” them off against the shadowed rings below. says Spilker. “They are harder to tease out in

26 ENGINEERING & SCIENCE spring 2010 the B ring, because it is so optically thick.” All the particles in all the rings are constantly colliding, but the dense B ring affords the most chances for the particles to adhere to one another afterward, held together by gravity and their own stickiness. The resulting flat sheets of material grow by accretion until they are 30 to 50 meters wide, at which point Saturn’s gravity pulls them apart. Under different circumstances, if they were farther from the planet, the wakes might have been the seeds of moons. And what of those ring particles? What are they made of? Though we know that water ice is 99 percent of the ring material, the other 1 percent remains a puzzle. Cassini has returned spectacular images showing that the rings when bathed in sunlight appear golden, golden-brown, or even slightly pinkish. If the rings were made only of water ice, they would be frosty white or bluish-white, like glacial ice on Earth. Something else is creating the subtle hues seen by Cassini, which has cameras with much greater color sensitivity than those on earlier spacecraft, notes Spilker. “Originally, after Voyager 2 flew past Saturn, we thought the fact that the rings were so bright and clean must have meant that they were young. Now we’re starting to think that maybe they’re a mix of young and old. There may be some processes that periodically break the ring particles open and recycle the contents. You can imagine breaking open a snowball and releasing fresh material into the system, making the rings appear cleaner and brighter. We’d really like to get an idea of the rate at which the rings are accumulating additional, non- icy material and then figure out what kind of processing is going on within the rings.” Spilker thinks that soon we’ll be able to tease out the subtle signals associated with ring contaminants from the very large signal that corresponds to water ice. One of the instruments working overtime on identifying the chemical makeup of the rings is Cassini’s Visual and Infrared Mapping Spec- trometer, or VIMS, which maps colors at

Clockwise, from top left: 1. An artist’s rendition of the ring particles. They continually clump and disperse again, forming thin, curved transitory aggregates with nearly empty space in between. 2. This false-color image from UVIS shows the density and orientation of the clumps. The brightest regions are the densest—in fact, the middle of the B ring was too dense for Alpha Arae’s light to penetrate. The clumps are tilted into oblique wakes in the blue regions, and are oriented along concentric circles in the yellow ones. 3. A natural-color view of the sunlit rings, as seen from below. The translucent C ring runs through the center of the frame, while the denser B ring arcs across the top. 4. As Saturn neared equinox in November 2008, the “spokes” (seen here as dark smudges) first seen by the Voyagers one Saturn year earlier returned to prominence. 5. In September 2009, a month after equinox, the spokes stand out much more clearly. Janus, a small moon 179 kilometers in diameter, can be seen at upper left.

spring 2010 ENGINEERING & SCIENCE 27 Below: A natural-color shot of icy Dione, snapped when Cassini was very nearly in the ring plane. The rings form a razor- thin horizontal stripe across the bottom of the picture, while the set of narrow, curving shadows cast on Saturn by the C ring is visible behind Dione. A portion of the B ring’s shadow adorns the upper right corner.

different wavelengths than does CIRS. Like Moons, Moons, and More Moons CIRS, the VIMS spectrometer breaks light And then there are the moons. Compared into its component wavelengths. Because to Earth and its monogamous relationship, every molecule reflects a specific set of Saturn is downright promiscuous, be- wavelengths and absorbs the rest, scien- ing accompanied by a retinue of—at last tists can identify molecules by their spectral count—62 orbiting companions. Saturn “fingerprints.” Shortly after Cassini arrived at is surrounded by real moons, wan- Saturn, VIMS data hinted at the presence of nabe moons, pieces of moons, some kind of iron-bearing compound—per- and fleeting moons. Cassini has haps not too surprising, as many scientists discovered seven moons—not think the rings are the remnants of icy com- to mention hints of numerous ets and iron-bearing meteoroids torn apart suspected but unconfirmed by Saturn’s gravity. moons that we infer from The composition of the rings has a direct their effects on the structure bearing on their age. Water ice is easily of the rings—and during eroded, and ring particles are pulverized by roughly the same interval, micrometeoroids and ground up by colli- other observers have found sions with each other. Saturn itself plays a 24 more. These moons are all role in breaking down the ice grains—UVIS tiny—less than 18 kilometers has detected an immense cloud of oxygen wide—and irregularly shaped. atoms liberated from ring ice by bombard- They travel in eccentric, steeply in- ment from Saturn’s own internal radiation. clined orbits far from Saturn, and are The cloud surrounds Saturn and extends for thought to be minor planets captured millions of kilometers beyond Saturn itself, by Saturn’s gravitational pull . . . or perhaps says Esposito. The rings also have their own debris from the breakup of such bodies after atmosphere of oxygen gas, produced when they were captured. Some of the irregular ultraviolet light from the sun interacts with moons are retrograde, orbiting in the direc- the water ice. Other kinds of debris, such as tion opposite to Saturn’s rotation. Porco’s tiny pieces of rock or minerals from micro- team even discovered objects in Saturn’s F meteorites, are harder to erode and have ring that could only be tracked for a couple greater longevity. “It could very well be that of orbits before disappearing altogether, different parts of the ring system have differ- suggesting that creation and destruction of ent ages,” says Porco. “The massive middle moonlike bodies around Saturn continues to B ring might be a lot older than the A ring. this day. One might be billions of years old and the Saturn’s moons and Saturn’s rings are in- other only a few tens of millions or hundreds extricably linked in a cosmic dance choreo- of millions of years. We don’t know.” graphed by the forces of gravity. “If a moon

Compared to Earth and its monogamous relationship, Saturn is downright promiscuous, being accompanied by a retinue of—at last count—62 orbiting companions. Saturn is surrounded by real moons, wannabe moons, pieces of moons, and fleeting moons.

28 ENGINEERING & SCIENCE spring 2010 The propellers in context. The leftmost photo shows the F, A, and B rings. Zooming in on the box in the middle of the A ring (center), we see a large density wave caused by Janus and Epimetheus at the bottom of the frame and two smaller density waves in the middle. Zooming in on the feature- less regions between the waves, we find four propellers (circled), each about five kilometers from tip to tip and caused by an unseen moonlet about 100 meters in diameter. The leading blade is about 300 meters closer to Saturn than the trailing one; the resolution is 52 meters per pixel. is large enough,” explains Spilker, “it can F ring in exquisite detail. We can now see clear a gap in Saturn’s rings by exerting a that it clumps in places where aggregates gravitational pull on the ring particles within of ring material or hidden moonlets lie, and a certain distance of itself. Good examples kinks where the gravitational pull of Pro- of that are two of Saturn’s ‘ring moons,’ Pan metheus tugs on material in the ring. In fact, and Daphnis. Pan orbits in the center of the Prometheus literally collides with the F ring Encke gap, and it keeps that gap open and at the point in its orbit most distant from pretty much free of ring particles, except for Saturn. As it moves back toward the planet a few dusty ringlets in the gap. Daphnis, a and away from the ring, the moon pulls ma- much smaller moon, keeps the Keeler gap terial out of the F ring, leaving dark channels open. It’s a much narrower gap because the behind. moon is smaller.” In a way, Galileo was not that far off In a way, Saturn’s rings are a labora- when he mistook Saturn’s rings for com- tory for determining how small an orbiting panion moons. Sometimes Saturn’s rings object can be and still be considered a are a source of material for the moons, and moon. “Cassini found still smaller objects sometimes Saturn’s moons are a source of that would like to open a gap but don’t have material for the rings. Cassini has confirmed enough gravity to succeed,” says Spilker. In- that at least three of Saturn’s moons are stead, like speedboats, they create “wakes” gaining girth from the rings. Pan, Daphnis, that travel with them. But rather than being and Atlas started out as “football-shaped V-shaped, these wakes resemble two- bodies,” says Porco. “But as they sweep bladed airplane propellers. One blade sticks their paths clean, some material accretes on out ahead of the moonlet, pulled forward by their surfaces, forming waistline bulges that gravitational attraction to faster-moving ring make them look a bit like flying saucers.” material nearer to Saturn. The other blade Several of Saturn’s moons produce rings stretches out behind, tugging on slower- or partial rings. The most celebrated of moving material farther from the planet. The these is Enceladus, which spews geysers propellers can be as long as five kilometers of water ice and vapor laced with organic from tip to tip. A propeller’s span depends material. (See E&S No. 1, 2006.) The icy on the size of its moonlet, but even the big- particles are pulled into orbit around Saturn gest moonlets are still too small to be seen. to form the diffuse E ring. “The geysers of Cassini has so far discovered more than Enceladus,” says Porco, “are perhaps our 150 such moonlets. most stupendous and significant discovery, Out at the edge of the main ring system because they very likely erupt from pockets lies the narrow, quirky F ring, held in place of organic-rich liquid water. Not only does by the shepherd moons Prometheus and this suggest a potentially habitable environ- Pandora. Cassini has photographed the ment below the south pole of the moon, but

Top: Potato-shaped Pandora (left), Prometheus (right), and the F ring. Middle: A hidden moonlet’s gravity disturbs the F ring, swirling it like a wisp of smoke. Bottom: Cassini snapped this shot half an hour after Prometheus (in the bottom left corner) burst up through the F ring—in fact, its shadow on the ring shows that it’s not quite all the way out yet. It’s beginning to pull a streamer of material after itself, which will eventually leave a dark channel in the ring. On its previous pass about 15 hours earlier, it tore open the channel that stretches downward across the center of the image; an even earlier channel that has started to fill back in extends into the upper right corner and out of frame.

spring 2010 ENGINEERING & SCIENCE 29 the way in which Enceladus is generating vinyl LP (remember those?), it turns out that Michigan, filled with paint thinner.” The lakes heat—and it is generating a lot of heat—is a some of the spiral density waves in Saturn’s are found in the polar regions, but there are fascinating problem in the study of planetary rings carry a record of the duo’s past orbital more in the north than the south—including moons.” Several other rings and ring arcs swaps, as he discovered after the 2006 ex- all the really big ones that might properly be discovered by Cassini are believed to be change. There is undoubtedly other informa- called seas. Planetary scientists are curious made of dust ejected from the surfaces of tion encoded in the rings as well, he says, to see if the coming of spring to the north various moons by meteoroid impacts. These adding that we are just starting to figure out will cause the lakes to evaporate and their include the Janus/Epimetheus ring, the how to extract it. contents to turn into rain in the southern Methone ring arc, the Anthe ring arc, and What’s next for Cassini? The mission hemisphere, creating fresh lakes there. the Pallene ring. extension to 2017 will allow us to follow a But the changing seasons alone may Janus and Epimetheus, by the way, are complete change of seasons on Saturn, as not be sufficient to explain the asymmetric unique in the solar system, as far as we the south pole goes into darkness and sum- distribution of Titan’s lakes. A team led by know. The two moons essentially share the mer arrives in the north. Spilker points out Associate Professor of Planetary Science same orbit, with Janus until recently being that Cassini arrived at Saturn just a couple Oded Aharonson and including grad stu- some 50 kilometers “inside” Epimetheus. of years after the northern winter solstice. dent Alexander Hayes (MS ’08), Jonathan The inside moon travels slightly faster Scientists noticed at the time that the atmo- Lunine (MS ’83, PhD ’85) of the Lunar and because it is slightly closer to Saturn, and sphere in the north looked much bluer than Planetary Lab at the University of Arizona, since Janus is 181 kilometers across and in the south, so they will be watching to see Ralph Lorenz of the Applied Physics Lab at Epimetheus is 116 kilometers wide, you’d if a similar bluish-colored haze starts to form Johns Hopkins, Michael Allison of the NASA think they’d plow into each other. Not so— down under as the southern hemisphere Goddard Institute for Space Studies, and when the inner moon overtakes the outer receives less sunlight. JPL Director Charles Elachi (MS ’69, PhD one once every four years, gravity steps in. Cassini will also keep an eye on Saturn’s ’71) has proposed that Titan may have a The outer moon pulls on the inner moon, largest moon, Titan. The spacecraft’s radar much longer cycle of climate change. Just giving it extra momentum and flinging it mapper has peered through Titan’s smoggy as our ice ages are widely believed to be into a higher orbit in which it, paradoxi- skies to identify about 400 lakes, some quite driven by regular, predictable variations in cally, moves more slowly. At the same time, large, of a liquid methane-ethane mixture. the tilt of Earth’s axis and the eccentric- the inner moon tugs on the outer moon, Porco describes them as being like “Lake ity and precession of its orbit around the siphoning off some of its momentum and dropping it into a lower, faster orbit. The moons trade orbits as they retreat back the way they came. The chase begins again, but now with a new pursuer. This cosmic game of tag was most recently played out this past January, putting Janus on the outside track until 2014, says Cornell’s Matthew Tiscareno (BS ’98), an associate on the Cassini Imaging Team. Like the grooves on a

Above: The fountains of Enceladus. More than 30 jets of all sizes can be seen here, over 20 of which were previously unidentified. Cassini took this shot just before barreling through the spray on November 21, 2009, in order to sample its composition. Right: The oily seas of Titan, tinted blue and black in this radar mosaic of the north polar region. The large one at upper-right whose coastline we can completely see is bigger than Lake Superior.

30 ENGINEERING & SCIENCE spring 2010 “The geysers of Enceladus,” says Porco, “are perhaps our most stupendous and significant discovery.”

sun—collectively called the Milankovitch spread out into a flattened disk. We stand to cycles—similar changes in Saturn’s (and learn a great deal about the early stages of therefore Titan’s) tilt and orbit may be at solar systems throughout the galaxy, and the work on Titan. If the lakes don’t fly south for cosmos, by studying Saturn’s rings.” the winter, it could bolster this theory. When the time comes to end the mission, NASA may decide to send Cassini into the depths of Saturn, much as Galileo (the Linda Doran is a freelance science writer, spacecraft, not the scientist) was inciner- member of the National Association of ated in Jupiter’s atmosphere. Such a fiery Science Writers, and former Jet Propul- demise would preclude any chance of sion Laboratory employee. Prior to that, Cassini one day crashing into—and perhaps she was a corporate communications contaminating—one of Saturn’s moons. If specialist for Sandia National Laboratories a later mission does discover microbes on in New Mexico and a reporter for the San Enceladus, for example, we’d like ironclad Gabriel Valley Tribune in California. She assurance that they’re not stowaways holds undergraduate degrees in geology, from Earth. First, however, Cassini would anthropology, and German. complete a final set of experiments—orbiting This article was edited by Douglas L. near Saturn’s cloud tops, perhaps sampling Smith. the dusty D ring, measuring the mass of the B ring for the first time, and acquiring precise measurements of Saturn’s gravitational and magnetic fields. In that way, Cassini will end its mission as it began, providing views of Saturn in greater detail than ever before. By studying Saturn, says Spilker, we can put ourselves into context. Titan’s primordial chemistry—a nitrogen atmosphere, no free oxygen, and small amounts of organic materials—may show us what Earth was like a very long time ago. And Saturn’s bling may give us hints about how and where to look for planets around other stars. Notes Porco, “The processes occurring at Saturn today are similar to those that occurred in the very early days of the solar system, when the material now contained in the planets was

Top: A frame from the first-ever visible-light movie of Saturn’s auroras. (The aurora has been colored orange to make it easier to see.) The movie was shot over 81 hours spanning October 5–8, 2009. Bottom: Saturn’s atmosphere is not as showy as Jupiter’s, with its baroque, swirling bands of clouds in vivid colors and long-lived, eye-catching storms like the Great Red Spot. But in its own pastel way, Saturn’s atmosphere is just as complex and intriguing. This true-color image of the north pole was taken in November 2008, as the northern winter was drawing to a close. We can see hundreds of bright storm systems, and a blue that has since faded to other colors with spring’s return.

spring 2010 ENGINEERING & SCIENCE 31 Recommend a Movie, Win a Million Bucks

Can a computer learn to choose movies you’re sure to like? Artificial intelligence took on this real-world problem in the Netflix contest. A Caltech alum recounts his adventures in the quest for the million-dollar grand prize.

but designing machines that “learn” in an even remotely similar manner will provide fodder for PhD theses for decades to come. The Netflix challenge, to build a better method for predicting a customer’s future movie preferences based on past movie ratings, ultimately led to many discoveries, including profound advances in statistical modeling—as well as the fact that fans of the TV series Friends tend to dislike Stanley Kubrick movies. A visitor to the Netflix website looking for something to rent faces a sea of more than 100,000 choices. Cinematch, the website’s Reproduced by permission of Netflix, Inc., Copyright © 2010 Netflix, Inc. All rights reserved. recommendation system, helps you sort Many a Caltech PhD goes into research, but teams competing, and with one day to go through them. Cinematch bases its sug- for the better part of a year I did so alone we were in first place. gestions on a staggeringly huge database and unemployed—and, yes, often in my Modeled after the 2003 X Prize chal- of movie ratings collected whenever one of pajamas. Yet, without ever leaving my apart- lenge to build a privately funded spaceship, Netflix’s 10-million-plus subscribers clicks ment, I ended up on an 11-country team the Netflix Prize was intended to harness on one of the five “star” buttons below the chasing the million-dollar Netflix Prize in the creativity of computer scientists and thumbnail picture of each movie’s cover art. what Caltech professor Yaser Abu-Mostafa statisticians worldwide. Humans (most of us The average customer has rated some 200 [PhD ’83] has aptly described as the Super anyway), instinctively learn from experience movies, so Cinematch has plenty of prefer- Bowl of machine learning. There were 5,169 to recognize patterns and make predictions, ence data to work with, but still, as of 2005,

32 ENGINEERING & SCIENCE spring 2010 Contrary to popular belief, self- employed computer gurus do occasionally spend time in the sun. Sill enjoys a balmy spring day on Chicago’s lakefront.

By Joseph Sill

Netflix was not entirely happy with it. Some came close to living up to—or down to—its of its suggestions seemed pretty random. name, dead last went to Avia Vampire Hunt- The company had been tweaking the er at 1.29 stars. (A zero-star rating was not system for years, and finally Netflix founder permitted.) But winning the contest would and CEO Reed Hastings tried to fix things require much more complex mathematics. himself. He neglected his wife and family The competition itself involved another over the 2005 Christmas vacation while 2.8 million data points where only the cus- manipulating spreadsheets and trying to tomer IDs, titles, and dates were revealed. course. I visited him in early 2008 to help make sense of mountains of customer data. This data had been randomly split into two interview prospective grad students, and he He soon concluded that he was in over his subsets, the quiz set and the test set, and filled me in. We’d been friends for more than head, and Netflix eventually decided to look contestants were not told which data points 15 years, so his interest excited mine; but outside the company for help. On October belonged to what set. Teams were allowed still, my job left me no free time. 2, 2006, Netflix formally offered one mil- to upload their predictions for the entire My own career in machine learning lion dollars to whoever submitted the best set to the Netflix Prize server once a day. had begun in Caltech’s computation and algorithm, as long as it beat Cinematch’s The server then compared the predictions neural systems program, which occupies performance by at least 10 percent. to the true ratings and calculated the root the intersection of computer science and Entrants were allowed to download a mean squared error, or RMSE, for each engineering with neurobiology. In it, neurobi- data set of more than 100,000,000 ratings set. Loosely speaking, RMSE measures the ologists use computers to help understand spanning seven years; 480,189 customers, average error. If an algorithm predicted three the brain, and engineers look to neurobiology represented anonymously by ID numbers; stars when the true rating was four, for ex- for inspiration in designing machines. With and 17,770 movies and TV series. Each data ample, the RMSE would be one. The teams’ a background in applied math, I was on the point consisted of four pieces of information: RMSEs on the quiz set were posted in rank theoretical periphery, and only survived the the customer ID, the movie or TV show’s order on a public web page known as the two demanding lab courses through the title, the rating, and the date the rating was leaderboard. The RMSEs on the test set, patience and generosity of my unlucky lab made. This was the “training set”—the data which would determine who—if anyone— partners. One final required us to determine you let your computer puzzle over to tease would win the million dollars, were known the inner workings of an analog VLSI chip (a out the patterns. only to Netflix. particularly esoteric device) by running vari- Some simple calculations showed that Things started out swiftly, with an im- ous diagnostic tests. I passed only by divine the highest-rated movies were the Lord of provement of more than 8.5 percent over intervention—the fabrication plant acciden- the Rings trilogy, with the final installment, Cinematch in the first 18 months. I had a tally burned the chips, and the exam was The Return of the King, in first place with an highly demanding job as a quantitative ana- cancelled. I’m much more comfortable work- average rating of 4.72 stars. Interestingly, lyst for a hedge fund at the time, so I was ing with data, which in the mid-’90s meant a every single one of the bottom 10 movies not following the action very closely. But my few thousand data points if we were lucky. was a horror flick. In one case, this may old mentor, Professor of Electrical Engineer- The sheer enormity of the Netflix dataset have been intentional, since the sixth-lowest ing and Computer Science Abu-Mostafa, was alluring, and in the summer of 2008 movie (1.40 stars) was titled The Worst was—he was using the Netflix Prize as the a career decision provided the time to Horror Movie Ever Made. Although this film basis for a project-based machine-learning dig into it. The full fury of that fall’s market

spring 2010 ENGINEERING & SCIENCE 33 Left: A CS 156b discussion section. Abu-Mostafa (seated, at left) and grad student Panna Felson (BS ’09) look on as senior Constantine (Costis) Sideris holds forth. Grad student and teaching assistant Chess Stetson is sitting on the right.

Right: At the end of 2007, BellKor ruled the Netflix leaderboard.

meltdown had not yet hit, and my employer THE Netflix prize as a teaching Tool was doing fine, but I was ready to move on The most important feature of the Netflix be the most promising technique known, from finance. I needed some time off, and competition was not the size of the prize, blending the results of such duplicative I needed a project to keep my skills sharp one million dollars, (although that didn’t efforts would not give a lot of improve- while I mulled things over. The Netflix Prize hurt!) but rather the size of the data: 100 ment. was perfect. million points. Machine learning research- Therefore, I announced that a team’s ers are used to much smaller data sets, grade would not depend on their algo- and Netflix provided several orders of rithm’s individual performance, but on the Getting Up to Speed magnitude more data than what we incremental improvement in the class- The contest had now been going on for al- usually have to be content with. It was a wide solution when the algorithm was most two years, so I had a lot of catching up machine learning bonanza. incorporated into the blend. This gave the to do. But I had ample time, now that I had Machine learning lives or dies by the students an incentive to explore the less- quit my job, and there was a road map to the data that the algorithm learns from. The traveled roads that might offer a better top of the leaderboard. To keep things mov- whole idea of this technology is to infer chance of shining in the blended solu- ing, Netflix was offering an annual $50,000 the rules governing some underlying tion. There was great educational value “Progress Prize” to the team with the lowest process—in this case, how people decide in pushing people to venture “outside the RMSE, as long as it represented at least a 1 whether they like or dislike a movie— b ox .” percent improvement on the previous year’s from a sample of data generated by that How did we do? Well, I registered two best score. To claim the prize, however, process. The data are our only window team names with Netflix. I told the stu- you had to publish a paper describing your into the process, and if the data are not dents that if we did really, really well, we techniques at a reproducible level of detail. enough, nothing can be done but make would submit under “Caltech.” The other The first Progress Prize had gone to BellKor, guesses. That doesn’t work very well. name was that of a rival school back east, a team from AT&T Research made up of I saw the Netflix data as an opportunity and we would use it if we did really, really Yehuda Koren (now with Yahoo! Research), to get my CS 156 Learning Systems stu- badly. It turned out that we fared neither Bob Bell, and Chris Volinsky. All I had to do dents to try out the algorithms that they too badly nor too well. Our effort gave was follow their instructions, and in theory I had learned in class. With 100 million data about a 6 percent improvement over the should get similar results. points, the students could experiment original Cinematch system, so we did not As I read BellKor’s paper, I quickly real- with all kinds of ideas to their heart’s officially submit it. ized that ironic quotation marks belonged content. The second time the course was of- around the phrase “all I had to do.” Their The first time I gave the Netflix problem fered, the Netflix competition had already solution was a blend of over 100 mathemati- as a class project, the competition was ended. There was no blending this time cal models, some fairly simple and others still going on. For the project, each team around, and the grading process was complex and subtle. Fortunately, the paper had to come up with its own algorithm more of a judgment call. suggested that a smaller set of models— and maximize its performance. Then all With the benefit of hindsight, the class’ perhaps as few as a dozen—might suffice. the algorithms would be blended to give performance was much better. One team Even so, nailing down the details would be a solution for the class as a whole. Like was getting a weekly improvement that it no easy task. Joe, our hope was to rapidly climb the took the teams in the actual competition leaderboard with each submission. months to achieve. In fact, at one point a Perhaps my biggest challenge as an in- guy in the class wanted to set an appoint- structor was to work out a way of ensur- ment to meet with me. He had the top ing that the various teams tried different individual score of 5.3 percent, so I techniques. As you will see, blending radi- jokingly told him that he needed to get to cally different solutions is key to getting 6 percent before I would see him. Within a good performance, so if everybody tried few days he had gotten to 6.1 percent. the same approach because it seemed to —YA-M

of 0.9570, which was nearly as good as the 0.9513 of the original Cinematch software. I got perverse enjoyment out of imagining a customer being told that since he hated Rambo III, he might like Annie Hall. Nearest-neighbor systems improved greatly over the contest’s first year, as teams delved into the math behind the models. These new twists represented significant progress in the design of recommendation systems, but a bigger breakthrough came in the form of another technique called matrix Several other leading teams had also no exception. With 17,770 movies, as op- factorization. published papers, even though they were posed to half a million customers, comput- Matrix factorization catapulted into promi- not compelled to do so, and some of their ing the correlations between all possible nence when “Simon Funk” suddenly ap- methods also looked promising. Three pairings in the training set was much more peared out of nowhere, vaulting into fourth general approaches seemed to be the most manageable, and the resulting table fit easily place on the leaderboard. Funk, whose real successful: nearest neighbors, matrix factor- into a computer’s RAM. These precomputed name is Brandyn Webb, had graduated ization, and restricted Boltzmann machines. correlations led to odd insights—it’s how I from UC San Diego at 18 with a computer Nearest-neighbor models are among learned that fans of Friends tend to dislike science degree, and has since had a spec- the oldest, tried-and-true approaches in Stanley Kubrick. (If you correlate Netflix’s tacular career designing algorithms. He was machine learning, so it wasn’t surprising to catalog with the ratings given to a DVD of a quite open about his method, describing it see them pop up here. There are two basic season of Friends and rank the results, Dr. in detail on his website, and matrix factoriza- variations—the user-based approach and Strangelove and 2001: A Space Odys- tion eventually became the leading tech- the product-based approach. sey end up at the bottom.) Typically, only nique within the Netflix Prize community. Suppose the task is to predict how many positive correlations are used—for instance, Matrix factorization represents each cus- stars customer 317459 would give Titanic. Pulp Fiction is highly correlated with other tomer and each movie as a vector, that is, a The user-based approach would look for Quentin Tarantino movies like Reservoir set of numbers called factors. The custom- like-minded people and see what they Dogs and Kill Bill, as well as movies with a er’s predicted rating of that movie is the dot thought. This similarity is measured by cal- similar dark and twisted sensibility like Fight product of the two vectors—a simple mathe- culating correlation coefficients, which track Club and American Beauty. Thus a product- matical operation involving multiplying each the tendency of two variables to rise and based approach allows Netflix to say, “since customer factor by each movie factor and fall in tandem. However, this approach has you enjoyed Fight Club, we thought you summing up the result. Essentially, the movie some serious drawbacks. The correlations might like Pulp Fiction.” vector encodes an assortment of traits, with between every possible pairing of custom- Surprisingly, I found that negative prod- ers need to be determined, which for Netflix uct-based correlations were nearly as useful amounts to 125 billion calculations—a very as positive ones. Although the connections DOT PRODUCTS heavy computational burden. Furthermore, were not as obvious, there was often a Let’s say that movie vectors have four factors: e-commerce companies have found that certain logic to them. For instance, Pulp Fic- sex (S), violence (V), humor (H), and music (M). people tend to trust recommendations tion and Jennifer Lopez romantic comedies Maggie, a customer who has just watched more if the reasoning behind them can be like The Wedding Planner and Maid in Brigadoon and is looking for more of bonnie explained. “Selling” a user-based nearest- Manhattan were strongly anticorrelated, and Scotland, has preferences S = 2, V = 1, H = 3, neighbor recommendation is not so easy: fans of the TV series Home Improvement M = 5. If Braveheart scores S = 2, V = 5, “There’s this woman in Idaho who usually generally disliked quirky movies such as The H = 1, M = 1 on this scale, then the dot product agrees with you. She liked Titanic, so we Royal Tenenbaums, Eternal Sunshine of the predicting Maggie’s likely reaction to Mel think you will, too—trust us on this one.” Spotless Mind, and Being John Malkovich. Gibson’s gorefest is (2 × 2) + (1 × 5) + (3 × 1) For these reasons, product-based I ended up devising a negative-nearest- + (5 × 1) = 17. Since the maximum possible nearest-neighbor techniques are now much neighbors algorithm—a “furthest opposites” score in this example is 100, she should prob- more common, and the Netflix contest was technique, if you will—that scored an RMSE ably give Braveheart a miss.

Reproduced by permission of Netflix, Inc., Copyright © 2010 Netflix, Inc. All rights reserved. spring 2010 ENGINEERING & SCIENCE 35 each factor’s numerical value indicating how “the temperature at which the brain begins the week, movies were more likely to get a much of that trait the movie has. The cus- to die” and intended as a Republican bad rating on Mondays and more likely to get tomer vector represents the viewer’s prefer- response) lowest. As the number of factors a good rating on weekends. On its own, this ences for those traits. It’s tempting to define grew—beyond 100, in some cases!—they model performed horribly, but if none of the the factors ahead of time—how much money got much harder to interpret. The trends other models took the day of the week into the film grossed, how many Oscar winners being discerned got more subtle, but no account, this tiny but statistically significant are in the cast, how much nudity or violence less real. signal could lead to a noticeable boost in the it has—but in fact they are not predeter- The third class of approaches, the ensemble’s accuracy. mined in any way. They are generated by the restricted Boltzmann machine, is hard to model itself, inside the “black box,” and only describe concisely. Basically, it’s a type of the model knows exactly what they mean. neural network, which is a mathematical Blended Models, Blended Teams By the fall of 2008, having familiarized myself with all the major techniques and the art of blending large numbers of models, I The default number of teams displayed on the web set out to climb the leaderboard. I started with what I thought was a modest goal—the page was 40, so if I cracked the Top 40 I could at least default number of teams displayed on the point myself out to people easily. web page was 40, so if I could crack the Top 40 I could at least point myself out to people easily. Although my girlfriend was being very supportive, I wanted a tangible achievement to show for my solitary efforts All we know is that as the model learns, it construct loosely inspired by the intercon- hunched over the desk in the living room of continually adjusts the factors’ values until nections of the neurons in the brain. As a my Chicago apartment, pounding away on they reliably give the right output, or rating, computation and neural systems alumnus, my laptop. But with more than 5,000 teams for any set of inputs. I was glad to see neural networks well competing, landing in the Top 40 meant But if a model had just a handful of fac- represented. being in the 99th percentile. I should have tors, sometimes their meanings leapt out. BellKor’s $50,000 winning blend incor- realized it wouldn’t be so easy. One such factor tagged teen movies: the porated numerous twists and tweaks of all For one thing, there were subtleties American Pie series and Dude, Where’s My three major techniques as well as several unmentioned in the papers that made the Car? came out on top, and old-school clas- less-prominent methods. This blending difference between a working model and sics like Citizen Kane and The Bridge on approach became standard practice in the a failure. Deducing these tricks on my own the River Kwai were at the bottom. Another Netflix Prize community. The process of sometimes took weeks at a time. Another loved romantic comedies such as Sleepless blending predictions was itself a meta-prob- challenge was weighing how much time to in Seattle and hated the Star Trek franchise. lem in machine learning, in that the blending spend on known methods versus inventing A third chose left-leaning films, placing algorithm had to be trained how to combine original approaches. I probably spent too Michael Moore movies such as Fahrenheit the outputs of the component models. much time trying for a home run—a stupen- 9/11 highest Teams working with blends soon discov- dous, brand-new technique. and movies ered that it didn’t make sense to fixate on Still, I scored one triumph. When blend- such as improving any particular technique. A better ing their models, most of my competitors Celsius strategy was to come up with new models relied on linear regression, a time-honored 41.11 that captured subtle effects that had eluded technique whose many virtues include a (sub- your other models. My favorite example was straightforward, analytically exact procedure titled a model based solely on what day of the for obtaining the best fit to the data. My week the rating had been made. Even after intuition suggested that the blend should be controlling for the types of movies that adaptive, depending on such side informa- get rated on tion as the number of ratings associated different with each customer or movie. By making days of every model’s coefficients a linear function

Abu-Mostafa uses telepathy to pick movies for you. Machine learning systems can’t do that yet.

36 ENGINEERING & SCIENCE spring 2010 Time was running out in the spring of these side variables, I got a significant of 2009, but nobody knew it. Then boost in accuracy while still retaining linear suddenly, on June 26, a newly regression’s key merits. formed team reached the goal. Despite this breakthrough, come January The competition promptly shifted 2009 I was still hovering just below where I into overdrive, as the contest wanted to be, stuck in the mid-40s. I would included a one-month “last call” often peruse the top 10 with a mixture of for all comers to take (or defend) respect and jealousy. One day, a new team the lead. The team with the best score suddenly appeared there—the Grand Prize at the end of that month would be Team, a name that struck me as presumptu- declared the winner. ous and cocky. When I read about them, though, I was intrigued. GPT’s founders—a team of Hungarian researchers called Grav- of BellKor and two Austrian ity, and a team of Princeton undergrads graduate students calling named Dinosaur Planet—had merged in themselves Big Chaos. 2007. That team, dubbed When Gravity and The amalgam, BellKor in Dinosaurs Unite, had very nearly won the Big Chaos, had won the first Progress Prize, but was overtaken by second Progress Prize BellKor in the final hours. in October 2008, with a GPT issued a standing invitation to all score of 0.8616. They had comers to join them, but in order to be since climbed to 0.8598, admitted the applicant had to demonstrably and there they had ground to improve GPT’s score—not an easy task. a halt. It seemed like that figure As of GPT’s founding, their RMSE stood at was chiseled in stone—so close 0.8655. The million-dollar goal was 0.8563, to the magic number, and yet so very so only 0.0092—or 92 basis points, as they far away. were called—remained to go. My stake as one of a dozen members of Shaving off just one basis point was ex- GPT gave me newfound motivation, as did cruciatingly hard, and GPT’s offer reflected the contrast between our leap and our com- dragged on, GPT narrowed the gap, but this. The original members would only claim petitors’ glacial progress. We were still un- the front-runners continued to inch ahead. one-third of the prize, or $333,333, and the derdogs, but underdogs with momentum on I’d check the standings every day, and often remainder would be split among the new our side. Over the next few months, I found several times a day, wincing on the rare days members in proportion to their basis-point some other side variables I could exploit, when a new score was posted. Their prog- contribution. Thus one basis point, the and we crept up the leaderboard. A new ress was so slow that I felt we had plenty of smallest measurable improvement, was recruit from Israel, Dan Nabutovsky, boosted time. I was wrong. worth almost $7,000. This may seem overly our score by another six basis points. generous, but it reflected how difficult pro Our collaborative style varied. Some GPT gress had become. Weeks or even months members (or prospective members) simply Endgame would go by with nothing happening at the sent Gábor their predictions. He’d add them Late afternoon on Friday, June 26, I top of the leaderboard, and a boost of just a to the mix and see whether the overall blend checked the leaderboard, and what I saw few basis points was cause for celebration. improved—something he could do without felt like a punch in the gut. A new team— The old 80/20 rule—that 80 percent of the even knowing how the new models worked. BellKor’s Pragmatic Chaos—had broken the payoff comes in the first 20 percent of the Gábor and I, however, were collaborating at million-dollar barrier, with a score of 0.8558. work—was in full force. a much deeper level, exchanging code and They hadn’t won yet, though. The contest My advancement as a lone wolf sty- debating which approaches were most likely rules provided for a one-month “last call” mied, I crossed my fingers and sent my to give us a boost. during which everyone had a final shot at adaptive-blending code to GPT captain Meanwhile, another underdog was also the prize, and the team with the best score Gábor Takács. Reading the email I got in making a run. Calling themselves Pragmatic at the end of this period would be declared response was the most satisfying moment Theory, Martin Chabbert and Martin Piotte the winner. So we weren’t toast, but things I had yet experienced—I’d boosted their of Quebec had been steadily rising up the didn’t look good. We were 36 basis points score by more than 10 basis points! GPT’s leaderboard. Neither of them had any formal behind, a gap that felt as wide as the Grand next official submission to Netflix a few days training in machine learning, and both were Canyon. We’d been running an ultrama- later jumped from 0.8626 to 0.8613 on the holding down full-time jobs. Nonetheless, by rathon—months-long for me; years-long strength of my contribution, leapfrogging March 2009 they had surpassed BellKor in for some of my teammates—and only the a few other entrants in the process. As of Big Chaos and taken the lead with a score day before, we’d been on the heels of the February 2009, I was suddenly a significant of 0.8597. After a bit of jockeying, the lead- leaders. Now, just short of the finish line, we shareholder on a leading team. erboard settled into a new equilibrium, with looked up and found they were so far ahead However, first place belonged to a merger the two teams tied at 0.8596. As the weeks that we had to squint to see them.

spring 2010 ENGINEERING & SCIENCE 37 We decloaked on Saturday afternoon. With less than 24 hours to go, we were on top.

With so little time left, GPT became much the prize, and I began to worry whether we’d the final minutes ticked away, Peng Zhou an- more collaborative, with most members be able to reconstruct what we had done. nounced from Shanghai that he thought he exchanging detailed information about their With so little time left, though, we decided could achieve 0.8553. We sent in his blend, algorithms via email and chat sessions. Ev- to deal with that later. retook the lead, and the deadline passed. eryone else was scrambling to catch up as The competition was scheduled to end We had finished first, but had we won? well. A few leaders were on a quixotic quest on Sunday, July 26, 2009, at 1:42 p.m. Shortly after the final buzzer sounded, to win on their own, but many could see that Chicago time. That Thursday, we obtained Netflix announced on the Netflix Prize there was strength in numbers. During the a thrilling result: 0.8554. We could take discussion board that two teams had quali- first two weeks of the final month, another the lead! Ah, but should we do so publicly? fied, and that their submissions were being coalition formed. Vandelay Industries, the As we phrased it in our email discussions, evaluated. We still didn’t know who’d won, name of a fake company from an episode should we decloak? Our members’ agendas but at least we knew we’d achieved a 10 of Seinfeld, and Opera Solutions, a real varied—some of us were all about win- percent (or better!) improvement on the test consulting firm, united and began aggres- ning, while others were more interested in set. We were relieved that most of our ac- sively recruiting anyone in the top 100. Soon the publicity. If we decloaked too soon, we curacy on the quiz set had carried over. their score was nearly the equal of ours, but might spur BellKor’s Pragmatic Chaos to In previous years, the Progress Prize even so, we were both still far behind. It was work extra hard or go on a recruiting binge. winners had been notified via email shortly clear to all what we had to do. We negoti- On the other hand, our window of oppor- after the deadline but well before any public ated a merger and a 50/50 split of the prize tunity might not last. If our next submission announcement. We had a gentleman’s money, halving the value of my basis points. was merely the runner-up, we’d get less agreement with our rivals that if either team We were now a group of 30 people, and media attention than if we had suddenly received such an email, we’d notify the since the technique of blending models is seized first place. Ultimately, we decloaked other. An agonizing 90 minutes passed, and known in the trade as ensemble learning, we on Saturday afternoon. With less than 24 finally, around 3:00 p.m., we got the email named ourselves The Ensemble. hours to go, we were on top. we dreaded. BellKor’s Pragmatic Chaos had We worked hard but in stealth, keeping However, we were Number One with an won. our new name off the leaderboard. BellKor’s asterisk. Remember that the leaderboard I went for a long jog to clear my head. I Pragmatic Chaos might have tried to recruit posted the quiz set’s results, while the true was a little depressed, but I wasn’t feel- the uncommitted if they felt threatened, so winner would be the highest performer on ing so bad. It was safe to say that more of we hoped to lull them into complacency the test set—a ranking known only to the their original work was in our solution than by keeping our existence a secret. We had Netflix engineers running the contest. Since vice versa, since they had been required to developed accurate ways to estimate our both sets were drawn from the same data, publish two papers in order to win the Prog- score in-house, and before long we deter- the two scores should be close, but pre- ress Prizes. I was also in awe of Pragmatic mined that we could surpass the 0.8563 cisely how close? Our learning and tweaking Theory, who had pulled off a stunning victory million-dollar barrier. Meanwhile, the opposi- process had been influenced in subtle ways in a field where they were newcomers work- tion also inched ahead, to 0.8555. Little did by feedback from the quiz set—an example ing in their spare time. I couldn’t begrudge they know that their lead was shrinking. of what’s known in the education biz as them the victory. Our efforts intensified as the contest “teaching to the test.” What would happen Netflix’s official stance for the next two entered its final week. With team mem- now that the test itself had changed? months was that the contest had not yet bers in Europe, India, China, Australia, and Our first-place debut brought a rush of been decided, as the winning software was the United States, we were working the adrenaline, but the game was far from over. still being validated. This led to a fair amount problem 24/7. We now had thousands of We planned to make one last submission a of confusion. We still held first place on the models, and there were countless ways to few minutes before the contest ended. We leaderboard, and it was entirely reasonable blend them. We could even blend a collec- worked frantically through Saturday night for a casual observer to assume that we had tion of blends. In fact, our best solution was and into Sunday morning. With 20 minutes won. Netflix asked us to say only that we many-layered—a blend of blends of blends to go, BellKor’s Pragmatic Chaos matched were happy to have qualified, which led to of blends. If we won, we’d have to docu- our quiz score of 0.8554. A cacophony of some awkward situations. ment our methods before being awarded emails and chat messages ensued, and as Netflix finally announced the winner at a

38 ENGINEERING & SCIENCE spring 2010 press conference in New York on Septem- Chabbert and Piotte had used the dates on that the overall prediction improved. Big ber 21. It was there that we learned that we which the ratings had been made in a very Chaos had found a way to train the indi- had, in fact, tied, with both teams scoring an creative way. A movie could be rated in one vidual models to optimize their contributions RMSE of 0.8567 on the test set. However, of two contexts: either within a few days of to the blend, rather than optimizing their own BellKor’s Pragmatic Chaos had sent in their watching it (Netflix asks customers to rate accuracy. submission 20 minutes before we did. The the movies they’ve just rented), or when The power of blending, which the contest pain of losing by such a minuscule margin browsing the website and rating previously demonstrated over and over again, was the was somewhat allayed by being quoted in seen movies. If a customer rated dozens or Netflix Prize’s take-away lesson. You could the New York Times, and I bought several hundreds of movies on a single day, it was even reverse-engineer a blend of models, copies as souvenirs. a safe bet that most of those films had been using the results to design an “integrated” The press conference was my first op- seen a long time ago. However, certain mov- single model that incorporated the effects portunity to meet both my teammates and ies got better ratings immediately after view- captured by the various simpler models. my competitors. I asked the Pragmatic ing, while others fared better in retrospect. Indeed, Pragmatic Theory created a single Theory duo whether they were ready to quit I had done a little work along these lines model that scored 0.8713, equal to the their jobs and become machine-learning re- myself, and had squeezed out an additional RMSE of the 100-model blend that had won searchers. I was only half joking, but Martin 1-basis-point share in GPT as a result, but BellKor the first Progress Prize. Piotte just shrugged and said he might take evidently the effect was more significant Although not all of the techniques up another hobby instead. After less than than I realized. Pragmatic Theory modeled involved in the winning solution have been 18 months in the field, and having published it in detail and got a significant accuracy incorporated into the working version of perhaps the year’s most important paper, I boost as a consequence. Cinematch, many of them have, and Netflix guess he felt it was time to move on. Another interesting advance came from has seen increased customer loyalty since At the awards ceremony that followed, Big Chaos’s Austrians. Most contestants implementing these advances. Hastings, the some members of BellKor’s Pragmatic optimized each individual model’s accuracy Netflix CEO, said in theNew York Times Chaos gave a talk outlining the advances on its own, and then blended it into the that the contest had been “a big winner” for made in their final push. It turned out that collection and crossed their fingers, hoping the company. As for myself, I didn’t quite finish as a winner in the formal sense, but I’m still thrilled with how the experience turned out. The contest has led to actual paid consulting work, and a group of my teammates and I are writing a paper on some of our techniques. Many, many other papers have and will come out of the contest, to the great benefit of the broader machine-learning research community. Only one team won the million dollars, but the Netflix competition ended up producing prizes of many kinds.

Joseph Sill is an analytics consultant. He earned his Caltech PhD in 1998, and a BS in applied math from Yale in 1993, where he was elected to Phi Beta Kappa. Before competing for the Netflix Prize, he had worked for Citadel Investment Group and NASA’s Ames Research Center. This article was edited by Douglas L. Smith.

It ain’t over till it’s over, as Yogi Berra used to say. Well, it’s finally over, and 20 minutes made all the difference.

A TRUE ALTERNATIVE FOR YOUR FAMILY

We’re not your run-of-the-mill, try-to-do- everything Credit Union. We choose to deliver only those products and services that we can do BEST. And year after year, this selective philosophy works! The result is an uncommonly successful and cost-efficient operation that is able to return a far greater percentage of its earnings to its members than the industry average. We consistently offer among the highest savings rates in the nation...the lowest loan rates available...and state-of-the-art online access. All with the unsurpassed safety and warm, personal service that has made Caltech Employees Federal Credit Union the trusted choice of our members — for both saving and borrowing — for over half a century. As a proud member of the extended Caltech family, we never forget that we work for YOU! Expect more So, when we offer a financial product, you can be sure it will be one of the best. That’s our from the policy, and the heart of our vision: to always be YOUR FIRST CHOICE FOR FINANCIAL SERVICES. Credit Union

exclusively

serving the

extended 800.592.3328 • www.cefcu.org A Strong, Secure and Exclusive Financial Resource Caltech & JPL Foothill Office: 528 Foothill Blvd. • La Cañada Flintridge JPL Office: 4800 Oak Grove Dr., Bldg. 218 • Pasadena community. Campus Office: 515 S. Wilson Ave. • Pasadena

Must qualify for CEFCU membership to join. Minimum $5 deposit and 25¢ membership fee due upon opening any CEFCU account. All loans subject to credit approval. CEFCU is an Equal Opportunity Lender. 40 ENGINEERING & SCIENCE spring 2010 Engineering & Science welcomes letters. Send correspondence to Douglas L. Smith, editor, E&S magazine, Caltech mail code 1-71, Pasadena, CA 91125, or email [email protected]. We reserve the right to edit any letters selected for publication for length, content, and clarity.

Letters

SUNLIGHT IN YOUR TANK The syngas can then be used as Consider the following closed First a general comment: the Fall feed stock for a Fisher-Tropsch unit to system. One atom of carbon-13 is in 2009 issue of E&S seems meatier produce usable fuels. its first excited state and stationary than usual. Thank you. Weigert quite rightly makes the in the rest frame of the observer, and While the science from Professor point that solar power towers and another atom of carbon-13 is in its Sossina Haile’s group is interesting, FT processes, as they’re known ground state and moving with respect isn’t the pursuit of chemical fuels from today, are not cheap and cannot com- to the observer along a vector directly sunlight for use in internal combustion pete with oil and gas production. toward the first atom, and at a speed (or even gas-fired turbine) engines He forgets that science and engi- equal to the recoil speed that the first a bit of a futile exercise if the energy neering never stands still, as proven atom will have as it makes a radiative conversion in the engine is only 25%? all the time in E&S magazine. transition to the ground state. Assume 25% × 25% = 6% overall (sunlight I can’t comment on the future cost further that the photon emitted by the to fuel × fuel to mechanical). Pho- of solar tower power, but I can on FT first atom is directed at the second tovoltaic and electric motors seem processes. atom and is absorbed, putting the to do better at 13% × 80% = 10% The Oxford Catalysts Group and its second atom in its first excited state. (sunlight to electrical × electrical to subsidiary, Velocys Inc., has invented Causality requires that the ab- mechanical). Even if one adds stor- a microchannel reactor and a hyper- sorption of the photon occurs at a age, for example 70% for pumped active FT catalyst. The two of them later time than the emission of the hydroelectric, the final result is 7% together outperform the best in class photon. This is certainly true in the versus 6%. fixed-bed reactor by a factor of 15. observer’s rest frame, and accord- Vertically migrating zooplankton This opens the possibility of ing to Einstein, it will be true in all would seem to contribute to vertical economic applications of small-scale frames. Please correct me if I’m transport of energy in at least two gas-to-liquid or waste-to-liquid. One wrong about that. The final state is other ways—the current needed to demonstration unit is being installed not the same as the initial state, but maintain position in the water column in Güssing, Austria, to test a feed- it is essentially the same as a time- if they are not neutrally buoyant, stock from a wood-chip gasifier, and reversal of the initial state, so it will and the transfer of heat when they in Brazil to test its potential to avoid have the same entropy. I do not think descend from warmer surface waters gas flaring in offshore production. that quantum mechanical uncertain- coupled with the “heat deficit” when In other words, science never ties—line width of the ground and they return to the surface after spend- stands still, and every avenue to stop excited states, the fact that the atoms’ ing time at depth. or to recycle CO2 should be investi- positions and velocities have uncer- Phelps Freeborn [BS ’65] gated with an open mind. tainties, and that the photon energy • • • Pierre Jungels [PhD ’73] is also somewhat uncertain, since it In his letter to E&S in the Winter has a finite lifetime—will change the 2010 issue, Frank Weigert (PhD ’68) THE ARROW OF TIME conclusion. (Again, please correct me comments on the brilliant research Sean Carroll, in his interesting if I’m wrong). reported in the previous issue that article, “The Arrow of Time,” says that Not only must the final state occur gives some hope of finding a solution “effects always follow causes. at a later time than the initial state, for CO2 recycling. It turns out that all (emphasis in the there is the intermediate state— He noted that the process re- original) of these phenomena can consisting of two atoms, both in the quires high temperature from a solar be traced back to the second law ground state, and a photon—which power tower to strip the catalyst of [of thermodynamics].” I am a big fan must occur after the initial state and its oxygen, then using the catalyst to of entropy—I’ve even written poems before the final state, and which will strip oxygen out of a mixture of steam about it—however, there are examples have the same entropy. and CO2 to produce mixture of H2 of effects following causes that are Bill Tivol [BS ’62] and CO known as syngas. unrelated to changes in entropy. • • •

spring 2010 ENGINEERING & SCIENCE 41 The article in the Winter 2010 issue to the earth’s crust, which is less than recall Lange’s interest and involve- of E&S entitled “The Arrow of Time” 30 miles thick). ment with the student body, through had some interesting information, but Of course, it is possible that our mentoring, student government and neglected to mention two very impor- planet could cycle through another the House system. I’m sure that he tant applications of the second law period of several hundred million would have wanted this mentioned, of thermodynamics in understanding years of high temperatures and even in this tightly abbreviated listing today’s crisis in energy and mineral enough carbon dioxide to produce of his contributions to the Caltech resources. a biomass large enough to lay down community. One online collection of Any machine produces work (me- another rich layer of fossil fuels. Or folks remembering Lange through tes- chanical energy) by transferring heat the planet could undergo extreme vol- timonials and photos can be found on (thermal energy) from a high-temper- canic activity to produce new mineral Facebook; search for “Andrew Lange.” ature reservoir, usually some sort of deposits by bringing precious ele- Kohl S. Gill [BS ’98] energy-consuming heat source, to ments to the surface in very localized a low-temperature reservoir, or heat magmatic intrusions. Sadly, this would Professor Lange’s obit, short by sink; the second law of thermodynam- take a very long time, during which necessity, was adapted from the ics gives the maximum efficiency of the planet would not be very habitable press release. A campus memorial

this process as (Thi – Tlo)/Thi, where Thi for homo sapiens. service for him will be held on Friday, is the temperature of the heat source Peter Gottleib [BS ’56] May 7. Look for complete coverage of

and Tlo is the temperature of the heat • • • it in the next issue of E&S. —ed. sink. In order to achieve an efficiency I just read “The Arrow of Time” in of one (the holy grail of all perpetual the latest E&S. It sounds like the au- ANNENBERG CENTER DEDICATED motion machines), the engine would thor thinks that it is possible to know I was pleased to read in the article have to exhaust into a reservoir having the exact state of the universe, but about the dedication of the An- a temperature of absolute zero. the Heisenberg uncertainty principle nenberg Center that Caltech has a Naive optimists viewing the ever- states that this is not possible. In professor of computer science and increasing cost of mineral resources, addition, Conway and Kochen’s free applied mathematics. particularly fossil fuels, are fond of will theorem denies any possibility of In about 1976, I applied to study saying that “we can always find more “hidden variables”—assuming that the application of then-nascent prin- by digging deeper.” Unfortunately, humans have free will; this cannot be ciples of computer science to the de- the second law of thermodynamics proven, but without it, all human en- velopment of mathematical software. tells us that our exploitation of the deavor is pointless. If there are indeed Professor Francis Buffington, who limited number of highly concentrated no hidden variables for predicting the I had known as an undergraduate mineral resources has resulted in outcomes of quantum measurements, student lab technician in Engineering a vast increase in entropy as these then this is a much stronger argument 91, had the duty to inform me that the resources are chemically transformed than entropy for the arrow of time. computer science department (then and/or spread around the world. This John Lindal called information science) thought is a process that can’t be reversed, [BS ’94, MS ’95, PhD ’01] my interests and qualifications fit well just like Sean Carroll’s arrow of with the applied mathematics depart- time. It is obvious that the key to the ANDREW LANGE REMEMBERED ment, while the applied mathematics survival of our civilization in the near I was saddened to hear of Profes- department thought I fit well with the term, i.e. the next several hundred sor Andrew Lange’s passing recently. computer science department. years, is to switch to more renewable I was also disappointed to see that Van Snyder, La Cresenta, CA energy resources and to recycle pre- his obituary in the Winter 2010 E&S cious materials such as gold. Digging omitted his work with students of deeper is not the answer; the high Caltech, and undergraduates in concentrations of minerals are limited particular. More than 10 years on, I

42 ENGINEERING & SCIENCE spring 2010 A Dream of Undying Fame: How Freud Betrayed His Mentor and Invented Psychoanalysis by Louis Breger Basic Books, 2009 160 pages, $22.95

books

This small biography is a gem, a bril- sisters who took his mother from him confirmatory evidence, we can never liantly polished examination of a vital as a child. . . .” know the accuracy of assertions such aspect of Sigmund Freud’s life and Confirmatory bias occurs through- as Freud’s claim that a young girl had work: his creation of the theory of out Freud’s writings, and Breger been “sexually molested each night by psychoanalysis and, in the service of brings it compassionately to life. He her governess.” this accomplishment, his remarkable makes it clear that Freud’s mind, In fact, one might interpret the perfidy to his teacher, Josef Breuer, brilliant as it could be in generating following to be a reflection of Freud’s the physician who had mentored ideas, was simply not open to the awareness of a fictional element to his Freud and, notably, had introduced possibility of disconfirmation of these case studies: him to the “talking cure.” Louis Breger, same ideas. Although he initially “It still strikes me myself as strange a psychoanalytic scholar, clinician, framed his formulations as hypoth- that the case histories I write should and Caltech professor of psychology eses, with the passage of time—but read like short stories and that, as one emeritus, has published two previ- with no supporting evidence—he might say, they lack the serious stamp ous biographies, Freud: Darkness came to express these formulations of science.” in the Midst of Vision, which was a as fact. He had to be right, and he Freud’s thinking was highly deep exploration of Freud’s life and required unwavering loyalty from his metaphorical—a fact that enlivened work, and Dostoevsky: The Author as colleagues. This was a central tragedy his writings but that also limited their Psychoanalyst. limiting his greatness and his human- basis in science. More, this dissoci- As he did in his earlier Freud biog- ity. ated his ideas from reality. He was, raphy, Breger provides a nuanced, Freud and Breuer claimed that apparently, sometimes aware of this. balanced, and ultimately respectful cathartic recall of traumatic memories Breger quotes a letter to Fliess, in examination of the history of Freud’s “immediately and permanently” re- which Freud confesses, “I no longer initial forays into psychotherapeutic lieved hysterical symptoms. However, believe in my neurotica,” adding that treatment under the guidance of as Breger notes, “it was much more “there are no indications of reality in Breuer and the nascent psychoana- difficult to achieve cures in practice the unconscious.” A contemporary lytic ideas they developed in collabo- than this statement implies.” Contem- critic of Freudian theory would be ration. Breger’s criticism of Freud’s porary evidence suggests that trau- hard-pressed to express it more ideas is tempered by compassion matic symptoms are less likely to be concisely. for Freud’s evident long-standing relieved by such clinical approaches Although Breger observes that it is personal torment. Breger describes than by methods intended to produce this “fictional” quality to Freud’s narra- Freud’s lifelong obsession with competing, benign memories. If Freud tive style that enriches his writing and becoming famous and, to that end, and Breuer were correct, one would renders it so compelling, the question his betrayal of Breuer, with whom he expect, for instance, that the vivid, of its factual accuracy remains open. broke professionally and socially and cathartic flashbacks of soldiers would One glimpses the apparent ease with whom he eventually failed to credit for be self-limiting, if not altogether cura- which a thought Freud may have had their critically important early collabo- tive, which they are not. about his own psychology becomes ration. Since Freud is otherwise so vulner- a “fact” about everyone’s. “A single Breger’s description of Freud’s ear- able to confirmatory bias and other idea of general value dawned on me. ly history increases our understanding logical errors in the service of promot- I have found, in my own case too, the of Freud’s development of some later, ing his assertion that, for example, phenomena of being in love with my unsubstantiated theoretical views. sexual factors underlie all forms of mother and jealous of my father, and For instance, Breger characterizes neurosis, one wonders how reliable I now consider it a universal event in the conflict between Freud and his Freud’s claims might be of any of the early childhood (italics mine).” fiancée’s mother and older brother: details unearthed from patients’ re- Thus, Freud admits that he takes “These fights were unconscious ports. Breger does not raise this spe- an experience of childhood feelings of remnants of his reactions to all those cific concern, but, in the absence of his own (taken by him as real, though

spring 2010 ENGINEERING & SCIENCE 43 IF A MATHEMATICAL EQUATION COULD we cannot know why) and then, as if became Freud’s bid for ‘eternal fame’; lay in our understanding of personality by magic, he considers the experi- it would make him a great scientist.” and psychopathology. His overwhelm- ence to be a universal one. One can Breger’s depiction of Freud’s ing need for fame—and his corollary make the case that Freud’s theories— personality, especially his desperate need, to be always correct—were, E X P R E S S his assertions about others’ psychol- quest for fame, shows us a pitiable perhaps, satisfied. But his distortion ogy—were essentially projections of man. For all of Freud’s substantial of psychoanalysis from testable theory his own sense of self. The fact that talent at observation and, more, for to a “cause” cast it into the category GRATITUDE, there is a kernel of truth to some of his distinct and articulate expression, of dubious beliefs. “How different them reflects that his own psychology his personal torment was a profoundly things would have been if, instead of WE’D USE IT. was not so different than that of most limiting quality. With the benefit of a cult-like ‘cause,’ psychoanalysis had humans. contemporary psychological knowl- really been the science that it claimed Breger’s clinical acumen is edge, one can readily see that many to be. . . .” If only, Breger suggests, We’re TIAA-CREF, and we’ve been a trusted partner of Caltech since beautifully rendered in his analysis of of Freud’s ideas were based primarily Freud could have continued to col- 1920. Now, we’re honored to be their single administrative services Oedipal theory. “Freud’s substitution upon his rich imagination. His talent laborate with Breuer. provider for retirement planning as we continue to offer personalized, of his universal Oedipal theory for one for fantasy was surely as capacious This book is a remarkably success- objective advice with a long-term commitment to their fi nancial goals.* based on real traumas was a mixture as that of his hysterical patients. This ful depiction of a central aspect of For over 90 years, we’ve helped millions of people in the academic, of truth and speculation. It revealed commonality likely contributed to his Freud’s life. Breger has written it as a medical, cultural, and research fi elds live well in retirement. his wish for his mother’s love and her extensive range of ideas, but not to scholar, yet it reads like a mystery, the loss to a rival, though he made the their critical examination. solution to which is both compelling Join us, and make a change for the good. Call to need for mother-infant attachment Breger suggests that Freud’s Stud- and tragic. —JB discuss your portfolio at 888 556-2451 or visit us at ‘sexual’ and substituted his father for ies in Hysteria “began a revolution in tiaa-cref.org/hereforyou. For a personalized counseling the many babies who took his place. our understanding of human person- Joseph Barber is a clinical profes- At the same time, it made him into a ality and psychological disturbance.” sor of rehabilitation medicine at the session, schedule an appointment at our local offi ce. warrior, a young Oedipus, in combat He adds that Freud “pushed the University of Washington School of with a king. It also did away with real field in a number of fruitful direc- Medicine in Seattle. 2 North Lake Avenue, Suite 150, Pasadena traumas, sexual or any other kind, and tions.” Freud’s emphasis on both the 626 432-6363 gave primary emphasis to instincts existence and the ubiquity of uncon- and fantasies. In this new theory, it scious motivation and on the potential was not what actually happened that meaningfulness of dreams continue was the source of fear, depression, to be powerful cultural influences. and symptoms—‘the worries that However, when Freud chose to not robbed me of my youth’—but rather test his theoretical formulations with the young child’s drive for pleasure, empirical research, he also led the Oedipal fantasies, and sexual wishes field astray. Moreover, one might ar- that conflicted with moral standards. gue, he contributed significantly to the In addition, the theory itself—immedi- delay in the growth of the science of ately promoted to ‘universal’ status— psychology, and especially to the de-

*Our advisors receive no commissions. TIAA-CREF compensates the advisors through a salary-plus-incentive program based on client service excellence and ﬁ nancial results. Advisors will only recommend products that help achieve our clients’ goals. TIAA-CREF products may be subject to market and other risk factors. See the applicable product literature, or visit tiaa-cref.org for details. Before transferring assets, consider the differences in features, costs, surrender charges, services, company strength and any tax consequences. C46514 © 2010 Teachers Insurance and Annuity Association- College Retirement Equities Fund (TIAA-CREF), 730 Third Avenue, New York, NY 10017. TIAA-CREF Individual & Institutional Services, LLC, and Teachers Personal Investors Services, Inc., members FINRA, distribute securities products. Annuity contracts and certiﬁ cates are issued by Teachers Insurance and Annuity Association (TIAA) and College Retirement Equities Fund (CREF), New York, NY. Consider the investment objectives, risks, charges and expenses carefully before investing. Visit tiaa-cref.org for a prospectus that 44 ENGINEERING & SCIENCE spring 2010 contains this and other information. Read carefully before investing.

44122-10_TT010PT1632_Engineering&Scinece_PG.indd 1 4/16/10 10:30 AM IF A MATHEMATICAL EQUATION COULD E X P R E S S GRATITUDE, WE’D USE IT.

We’re TIAA-CREF, and we’ve been a trusted partner of Caltech since 1920. Now, we’re honored to be their single administrative services provider for retirement planning as we continue to offer personalized, objective advice with a long-term commitment to their ﬁ nancial goals.* For over 90 years, we’ve helped millions of people in the academic, medical, cultural, and research ﬁ elds live well in retirement.

Join us, and make a change for the good. Call to discuss your portfolio at 888 556-2451 or visit us at tiaa-cref.org/hereforyou. For a personalized counseling session, schedule an appointment at our local ofﬁ ce.

2 North Lake Avenue, Suite 150, Pasadena 626 432-6363

*Our advisors receive no commissions. TIAA-CREF compensates the advisors through a salary-plus-incentive program based on client service excellence and ﬁ nancial results. Advisors will only recommend products that help achieve our clients’ goals. TIAA-CREF products may be subject to market and other risk factors. See the applicable product literature, or visit tiaa-cref.org for details. Before transferring assets, consider the differences in features, costs, surrender charges, services, company strength and any tax consequences. C46514 © 2010 Teachers Insurance and Annuity Association- College Retirement Equities Fund (TIAA-CREF), 730 Third Avenue, New York, NY 10017. TIAA-CREF Individual & Institutional Services, LLC, and Teachers Personal Investors Services, Inc., members FINRA, distribute securities products. Annuity contracts and certiﬁ cates are issued by Teachers Insurance and Annuity Association (TIAA) and College Retirement Equities Fund (CREF), New York, NY. Consider the investment objectives, risks, charges and expenses carefully before investing. Visit tiaa-cref.org for a prospectus that contains this and other information. Read carefully before investing.

44122-10_TT010PT1632_Engineering&Scinece_PG.indd 1 4/16/10 10:30 AM California Institute of Technology Pasadena, California 91125