Aegean Associates, Inc.

Many engineering disciplines rely on supercomputers to simulate complicated physical phenomena — how cracks form in building materials, for instance, or fluids flow through irregular channels. Now, researchers in MIT’s Department of Mechanical Engineering have developed software that can perform such simulations on an ordinary smart phone. Although the current version of the software is for demonstration purposes, the work could lead to applications that let engineers perform complicated calculations in the field, and even to better control systems for vehicles or robotic systems.

The new software works in cases where the general form of a problem is known in advance, but not the particulars. For instance, says Phuong Huynh, a postdoc who worked on the project, a computer simulation of fluid flow around an obstacle in a pipe could depend on a single parameter: the radius of the obstacle. But for a given value of the parameter, calculating the fluid flow could take an hour on a supercomputer with 500 processing units. The researchers’ new software can provide a very good approximation of the same calculation in a matter of seconds.

“This is a very relevant situation,” says David Knezevic, another postdoc in the department who helped lead the project. “Often in engineering contexts, you know a priori that your problem is parameterized, but you don’t know until you get into the field what parameters you’re interested in.”

Each new problem the researchers’ software is called upon to solve requires its own mathematical model. The models, however, take up very little space in memory: A cell phone could hold thousands of them. The software, which is available for download, comes preloaded with models for nine problems, including heat propagation in objects of several different shapes, fluid flow around a spherical obstacle, and the effects of forces applied to a cracked pillar. As the researchers develop models for new classes of problems, they post them on a server, from which they can be downloaded.

Advance work

But while the models are small, creating them is a complicated process that does in fact require a supercomputer. “We’re not trying to replace a supercomputer,” Knezevic says. “This is a model of computation that works in conjunction with supercomputing. And the supercomputer is indispensable.”

Knezevic, his fellow postdoc Phuong Huynh, Ford Professor of Engineering Anthony T. Patera, and John Peterson of the Texas Advanced Computer Center describe their approach in a forthcoming issue of the journal Computers and Fluids. Once they have identified a parameterized problem, they use a supercomputer to solve it for somewhere between 10 and 50 different sets of values. Those values, however, are carefully chosen to map out a large space of possible solutions to the problem. The model downloaded to a smart phone finds an approximate solution for a new set of parameters by reference to the precomputed solutions.

The key to the system, Knezevic says, is the ability to quantify the degree of error in an approximation of a supercomputing calculation, a subject that Patera has been researching for almost a decade. As the researchers build a problem model, they select parameters that will successively minimize error, according to analytic techniques Patera helped developed. The calculation of error bounds is also a feature of the phone application itself. For each approximate solution of a parameterized problem, the app also displays the margin of error. The user can opt to trade speed of computation for a higher margin of error, but the app can generally get the error under 1 percent in less than a second.

Turning the tables

While the researchers’ software can calculate the behavior of a physical system on the basis of its parameters, it could prove even more useful by doing the opposite: calculating the parameters of a physical system on the basis of its behavior. Instead of, say, calculating fluid flow around an obstacle based on the obstacle’s size, the software could calculate the size of the obstacle based on measurements of the fluid flow at the end of a pipe. Ordinarily, that would require several different computations on a supercomputer, trying out several different sets of parameters. But if testing, say, 30 options on a supercomputer would take 30 hours, it might take 30 seconds on a phone. Indeed, the researchers have already developed a second application that calculates such “inverse problems.”

In the same way that a simulation of a physical system describes its behavior on the basis of parametric measurements, control systems, of the type that govern, say, automotive brake systems or autonomous robots, determine devices’ behavior on the basis of sensor measurements. Control-systems researchers spend a great deal of energy trying to come up with practical approximations of complex physics in order to make their systems responsive enough to work in real time. But Knezevic, Huynh and Patera’s approach could make those approximations both more accurate and easier to calculate.

Max Gunzberger, Frances Eppes Eminent Professor of Scientific Computing at Florida State University says that the MIT researchers’ work has a “cuteness aspect” that has already won it some attention. But “once you get over the cuteness factor,” he says, “if you talk about serious science or serious engineering, there’s a potential there.” Gunzberger points out that while the researchers’ demo concentrates on fluid mechanics, “there’s lots of other problems that their approach can be applied to. They built the structure that they themselves or others can start using to solve problems in different application areas.”

Deep in the ocean, sponges of the Agelas family, or bacteria living within the sponges, emit chemicals believed to help them defend their territory. Those chemicals, called agelastatins, have also shown the ability to kill cancer cells. For that reason, chemists have been trying to find ways to synthesize agelastatins in the laboratory since the chemicals were discovered in 1993.

Chemists at MIT, led by Associate Professor Mohammad Movassaghi, recently discovered the shortest and most productive way to synthesize all six of the known agelastatins. The team, which also includes graduate students Dustin Siegel and Sunkyu Han, described the new method in the Aug. 16 online edition of the journal Chemical Science.

“Movassaghi's very elegant synthesis demonstrates a nicely scalable, multi-gram preparation of all the known agelastatins,” says Tadeusz Molinski, the chemist who first isolated agelastatins C and D, the third and fourth agelastatins discovered, in 1998. Molinski, a professor of chemistry at the University of California at San Diego, says the new synthesis will allow researchers to produce enough of the compounds to test them as cancer drugs.

Agelastatins have been shown to inhibit cancer-cell proliferation by interfering with cell division. They also repress an enzyme known as glycogen synthase kinase-3, a potential target for treating Alzheimer’s disease and bipolar disorder.

“They have a very broad range of biological activity,” says Movassaghi. “The sponges are not interested in treating cancer or Alzheimer’s, but the agelastatins are potently active against them.”

Scientists speculate that sponges, or bacteria that live in symbiosis with them, release agelastatins into their watery environment to warn other sponge species not to colonize the area.

Copying nature

Agelas sponges, which have been found in the Coral Sea and Indian Ocean, are difficult to obtain, so researchers have had trouble generating enough agelastatin to do large-scale experiments in cancer cells. Since they were first discovered, chemists have reported about a dozen ways to produce one or more of the compounds, but none of the chemists have been able to produce all six. The MIT team can do so, and in relatively large quantities — a gram per reaction batch.

The reaction begins with a commonly available starting material, aspartic acid. The synthesis requires seven steps to produce agelastatin A, the first discovered and most potent of the compounds. Agelastatin A can then be converted to agelastatins B, C or E. The synthesis can also be altered slightly to produce D, which can then be converted to F.

In designing their synthesis, Movassaghi, Siegel and Han tried to mimic the way they believe the sponges naturally produce agelastatins.

Each agelastatin contains four rings, known as A, B, C and D, and most chemists have used syntheses in which the C ring forms before the B ring. The MIT team formed the B ring first, and the C ring last. The C ring is the only ring made solely of carbon atoms (all of the others contain at least one nitrogen atom), and it is where all four of the molecule’s stereocenters are found. (Stereocenters are atoms around which the molecule can take different three-dimensional orientations.)

Other chemists had theorized that the biological synthesis of agelastatins would use precursors with an electron-deficient carbon atom in the fourth carbon position and a carbon atom that wants to share its electrons in the eighth position. Movassaghi switched those features.

To show whether sponges do the same series of steps, more experiments are needed. Researchers could label the precursors with isotope tags, give them to the sponges and follow where the isotopic labels end up. Although some of the steps of Movassaghi’s synthesis require high temperatures or acidic conditions, those same reactions could occur under biological conditions if catalyzed by enzymes.

Movassaghi’s lab is now collaborating with researchers in academia and industry to test the biological activity of the compounds, with an emphasis on their anti-cancer activity. Using the new synthesis, the researchers should be able to easily produce variants not found in nature that might have even more powerful effects, says Movassaghi. The synthesis should also provide a good starting point for possible future large-scale production, should there be a need, he says.

Plans for Orion — the capsule that resembles the Apollo program’s spacecraft and was supposed to send humans to the moon by 2020 as part of NASA’s Constellation program — were changed in February when President Obama canceled Constellation, and then announced two months later that NASA would continue to develop Orion as an escape vehicle to be docked at the International Space Station for emergencies.

While it appears that Orion will eventually take flight, NASA continues to struggle with one crucial aspect of its design: minimizing the violent impact that astronauts would experience during landing. Although NASA initially designed Orion’s crew seats to be mounted onto a stiff structure supported by shock absorbers — essentially the same technology used to cushion Apollo’s water landings — this 1,100-pound structure would be too heavy to cushion astronauts if the vehicle landed on land. Whereas the Apollo capsule was designed to land in water, and Orion would likely do the same, NASA wants to make sure that Orion can land on land in case of an emergency.

A graduate student in MIT’s Department of Aeronautics and Astronautics has helped design a smaller alternative: a reusable, 700-pound air-bag system that could inflate during launch and landing, deflate for storage purposes, and partially inflate to provide seating while the vehicle is in space. Not only would the system be lighter than the one NASA originally proposed, but it would also be entirely mechanical, meaning not controlled by computers.

This is important because “the vast majority of accidents and failures in engineering systems” can be traced to computers misinterpreting situations, says Sydney Do, who helped design the air-bag system and spent several weeks in August testing a full-sized prototype designed to protect one astronaut. “Our goal was to see if it was possible to design a landing system that was purely mechanical.”

According to a paper presented at the American Institute of Aeronautics and Astronautics Space 2009 conference by Do and his thesis adviser, Olivier de Weck, an associate professor of aeronautics and astronautics and engineering systems, the air-bag system was inspired by the structure of seeds. Just as a fluid surrounds the embryo in seeds to provide protection as the seed is distributed, the Orion air-bag system would surround each astronaut in “a personal cushion of air,” according to current NASA astronaut Charlie Camarda, who seeks to develop more innovative space-engineering concepts that veer from the traditional. In 2008, Camarda helped organize a group of students from Pennsylvania State University and MIT, including Do, to explore how the physics of seeds could be applied to engineering principles. Do’s design for an Orion air-bag system, Camarda says, represents “a very novel” approach to mechanical design that could inspire more biological-based solutions in engineering.

Valve analysis

NASA’s Engineering and Safety Center agreed to fund the study by the Penn State and MIT students to explore the feasibility of an air-bag system that Orion astronauts could inflate before reentering Earth’s atmosphere. The students’ first step was to conduct tests to observe how the inflated bags behave when they are dropped from increasing one-foot increments while supporting an object that weighs about the same as an average male head — such drops simulate the impact velocity that an astronaut would feel upon landing.

These tests revealed how important timing is in terms of releasing gas from an air bag. Unlike car air bags, which inflate when hot gas is injected into them upon impact, the inflated Orion air bags already contain gas upon impact. If the air bags are either not big enough or don’t have enough air in them, the astronaut’s seat will directly impact the ground. Alternatively, if there is enough gas inside the bag, but it’s not released before the seat hits the ground, the impact will cause the seat to bounce upward, which could injure the astronaut. That’s because as an astronaut falls into the bag during the landing, the kinetic energy created from this motion is combined with the energy of the gas molecules moving inside the bags. This increases the pressure of the gas inside the bag, which could cause bouncing.

To prevent this bounce, enough gas needs to be vented between the point at which the floor of Orion impacts the ground and the point at which the seat and the astronaut impact the ground so that the kinetic energy caused by the falling seat and occupant have been removed. But even after some of this gas is vented, there still needs to be enough gas remaining in the bags to prevent direct impact between the seat and the ground. To get this balance right, the students decided to design valves that are triggered to open at a low pressure, which would allow gas to vent as soon as Orion’s floor comes to rest, but before the seat can impact the ground.


This clip shows several views of a “drop test” of an air-bag system being designed for a space capsule. During the test, a dummy attached to the air-bag system is raised and then dropped, simulating the velocity an astronaut would experience during landing.
Video courtesy of Sydney Do

Drop-test survival

When NASA decided to fund the research for another year last spring, Do took over the research for his master’s thesis and began testing a valve for the system. He then developed a computer model to analyze how certain variables, such as air-bag size, would affect the risk of astronaut injury upon impact. This helped him configure a prototype seat that would have four air bags — each about one foot long by two feet wide — containing two rectangular valves about six inches wide. Do then built the air bags from vectran, a high-strength material that was used to make the air bags for several rovers that landed on Mars.

Earlier this month, he tested the prototype through a series of drop tests conducted from as high as 10 feet involving a crash dummy that measured the acceleration of each drop. While Do still needs to analyze those results before presenting his final design to NASA later this fall, he says that the fact that the system survived dozens of drops suggests that certain variables he chose for the prototype, such as the material and manufacturing of the air bags, are adequate for an Orion landing. According to Camarda, future research could explore ways to ensure “a robust and fail-safe” system in the event that a valve malfunctions.

Do cautions that the air-bag system has one drawback: It’s likely only effective for vertical drops, meaning that the air bags could tip over if Orion descended at a sideways angle. But he says this might not be an issue if Orion is designed to land vertically. Although whatever NASA decides to do with Do’s research ultimately depends on the future of human spaceflight, he is hopeful that even if Orion never takes flight, his research could be used to guide designs of similar capsule-type spacecraft that commercial companies might be interested in building.

Across science and engineering, computers are often enlisted to find patterns in data. The data might be genetic information about a population, and the pattern could be which gene variants predispose people to asthma. Or the data might be frames of video, and the patterns could be objects that move or stand still from frame to frame, which data-compression or image-sharpening algorithms might want to locate.

In most cases, more data means more reliable inference of patterns. But how much data is enough? Vincent Tan, a graduate student in the Department of Electrical Engineering and Computer Science, and his colleagues in Professor Alan Willsky’s Stochastic Systems Group have taken the first steps toward answering that question.

Tan, Willsky and Animashree Anandkumar, a postdoc in Willsky’s group, envision data sets as what mathematicians call graphs. A graph is anything with nodes and edges: Nodes are generally depicted as circles and edges as lines connecting them. A typical diagram of a communications network, where the nodes represent electronic devices and the edges represent communications links, is a graph.

In the MIT researchers’ work, however, the nodes represent data and the edges correlations between them. For instance, one node might represent asthma, and the others could be a host of environmental, physiological and genetic factors. Some of the factors might be correlated with asthma, others not; other factors might be correlated with each other but not with asthma. Moreover, the edges can have different weights: The strength of the correlations can vary. From this perspective, a computer charged with pattern recognition is given a bunch of nodes and asked to infer the weights of the edges between them.

Stars and Chains

The MIT researchers are concentrating on graphs known as trees. Technically, a tree is a graph without any closed circuits: for any node, there’s no sequence of edges that will lead you back to it without backtracking. But intuitively, a tree is a graph that looks like a family tree, with a single node at the top connected to one or more nodes in the layer below, which are connected to nodes in the layer below them, and so on. The patterns formed by data of interest won’t always be trees, but when they aren’t, there’s generally a tree that closely approximates them. And trees are much easier to work with mathematically than graphs with random shapes.

In an article published this spring in IEEE Transactions on Signal Processing The researchers demonstrated that trees with a “star” pattern — in which one central node is connected to all the others — are the hardest to recognize; their shape can’t be inferred without lots of data. Suppose, for instance, that the central node represents asthma, and 100 other nodes represent all the factors that can contribute to it. If the computer system looks at 100 data samples, each one could imply a different predictor of asthma. It might require tens of thousands of samples before the system could reliably conclude which factors have stronger correlations than others.

Trees that form “chains,” on the other hand — where each node is linked to at most two others — are the easiest to recognize. Suppose that a computer system was analyzing concentrations of chemicals in biological cells. If the data reflected different stages of a single, complicated biochemical process, then the chemicals present at each stage might determine the chemicals present at the next. In that case, it would be fairly easy to conclude, with few data samples, the correlations between successive stages of the process.

The results for stars and chains led the researchers to hope that the difficulty of recognizing a tree pattern is purely a function of its diameter — the distance between the two farthest-apart nodes. (In a star, the diameter is two: No two nodes are more than two hops apart. In a chain, the diameter is equal to the number of nodes.) Unfortunately, that turns out not to be the case. For trees with shapes other than stars or chains, “Strength of the connectivity between the variables matters,” Tan says. “We cannot discount it.” Nonetheless, the tools that the researchers used to evaluate the best- and worst-case scenarios could shed further light on the intermediary cases.

“When you have large and high-dimensional data sets, it becomes very difficult to compute with more-general models,” says John Lafferty, a professor of computer science at Carnegie-Mellon University. “So having this tree-structured approximation is much more computationally efficient.” Lafferty points out, however, that the MIT researchers’ work assumes that the data being analyzed have a “normal distribution” — that a graph of them would have the familiar bell-curve shape. “You’d like the distribution to have any shape,” says Lafferty, an extension of the MIT researchers’ result that his own group his been working on. Lafferty also says that pattern-recognition techniques that “relax this tree assumption” — that can tolerate a few closed loops in the graph — would be more generally useful. But that’s something that Willsky’s group is already working to do, he says.

Human pluripotent stem cells, which can become any other kind of body cell, hold great potential to treat a wide range of ailments, including Parkinson’s disease, multiple sclerosis and spinal cord injuries. However, scientists who work with such cells have had trouble growing large enough quantities to perform experiments — in particular, to be used in human studies.

Furthermore, most materials now used to grow human stem cells include cells or proteins that come from mice embryos, which help stimulate stem-cell growth but would likely cause an immune reaction if injected into a human patient.

To overcome those issues, MIT chemical engineers, materials scientists and biologists have devised a synthetic surface that includes no foreign animal material and allows stem cells to stay alive and continue reproducing themselves for at least three months. It’s also the first synthetic material that allows single cells to form colonies of identical cells, which is necessary to identify cells with desired traits and has been difficult to achieve with existing materials.

The research team, led by Professors Robert Langer, Rudolf Jaenisch and Daniel G. Anderson, describes the new material in the Aug. 22 issue of Nature Materials. First authors of the paper are postdoctoral associates Ying Mei and Krishanu Saha.

Refining surfaces

Human stem cells can come from two sources — embryonic cells or body cells that have been reprogrammed to an immature state. That state, known as pluripotency, allows the cells to develop into any kind of specialized body cells.

It also allows the possibility of treating nearly any kind of disease that involves injuries to cells. Scientists could grow new neurons for patients with spinal cord injuries, for example, or new insulin-producing cells for people with type 1 diabetes.

To engineer such treatments, scientists would need to be able to grow stem cells in the lab for an extended period of time, manipulate their genes, and grow colonies of identical cells after they have been genetically modified. Current growth surfaces, consisting of a plastic dish coated with a layer of gelatin and then a layer of mouse cells or proteins, are notoriously inefficient, says Saha, who works in Jaenisch’s lab at the Whitehead Institute for Biomedical Research.

“For therapeutics, you need millions and millions of cells,” says Saha. “If we can make it easier for the cells to divide and grow, that will really help to get the number of cells you need to do all of the disease studies that people are excited about.”

Previous studies had suggested that several chemical and physical properties of surfaces — including roughness, stiffness and affinity for water — might play a role in stem-cell growth. The researchers created about 500 polymers (long chains of repeating molecules) that varied in those traits, grew stem cells on them and analyzed each polymer’s performance. After correlating surface characteristics with performance, they found that there was an optimal range of surface hydrophobicity (water-repelling behavior), but varying roughness and stiffness did not have much effect on cell growth.

They also adjusted the composition of the materials, including proteins embedded in the polymer. They found that the best polymers contained a high percentage of acrylates, a common ingredient in plastics, and were coated with a protein called vitronectin, which encourages cells to attach to surfaces.

Using their best-performing material, the researchers got stem cells (both embryonic and induced pluripotent) to continue growing and dividing for up to three months. They were also able to generate large quantities of cells — in the millions.

Creating new synthetic materials for stem-cell growth is a longstanding problem that many researchers have tried to solve, says Sheng Ding, an associate professor of chemistry at the Scripps Research Institute. “In the past, it was more of a trial-and-error process,” he says. “The beauty of this work is that they can design these in a very systematic way. This is really a platform that can be applied not just to human embryonic stem cells, but also other cells.”

The MIT researchers hope to refine their knowledge to help them build materials suited to other types of cells, says Anderson, from the MIT Department of Chemical Engineering, the Harvard-MIT Division of Health Sciences and Technology, and the David H. Koch Institute for Integrative Cancer Research. “We want to better understand the interactions between the cell, the surface and the proteins, and define more clearly what it takes to get the cells to grow,” he says.

Other MIT authors of the paper are Said Bogatyrev, Z. Ilke Kalcioglu, Maisam Mitalipova, Neena Pyzocha, Fredrick Rojas and Krystyn Van Vliet. Jing Yang, Andrew Hook, Martyn Davies and Morgan Alexander of the University of Nottingham (United Kingdom) and Seung-Woo Cho of Yonsei University (Korea) are also authors of the paper.

In July, Google released a trial version of new software, called the Google App Inventor, intended to let people with no previous programming experience design applications for phones that use Google’s Android operating system. The software has provoked much commentary in the technology press, and Google has been trumpeting it as a way to give people direct control of their own phones. But App Inventor is the latest outgrowth of a tradition of MIT research that dates back at least 40 years.

The App Inventor project was led by Hal Abelson, the Class of 1922 Professor of Computer Science and Engineering, who spent a sabbatical year at Google as a visiting professor. Instead of having to write traditional computer code, users of App Inventor can create programs by snapping together virtual, color-coded instruction “blocks.” For instance, to add a button to an application, the user would drag the button block into App Inventor’s workspace window and determine the button’s visual properties by selecting from pull-down menus. Then, to determine what the button will do, the user would snap a block that defines a function — like emitting a noise, or making a phone call, or changing the screen’s background color — into the button block.

The App Inventor blocks are based on the MIT master’s thesis of Ricarose Roque, for which she built a general version of a programming interface that Eric Klopfer, the director of MIT’s Teacher Education Program, had developed for a simulation program called StarLogo. StarLogo, in turn, began as the PhD thesis of Mitchel Resnick, who heads MIT’s Media Arts and Sciences Program, and whose graduate advisors were Abelson and Seymour Papert, a pioneer of educational computing.

Papert, who came to MIT in 1963, is most famous for inventing Logo, a simple computer language designed to introduce young children to the principles of programming. Initially, programs written in Logo would guide a robot with a pen attached to its undercarriage — a “turtle” — across a sheet of paper, executing a drawing. In the late 1960s, when Abelson was a graduate student at MIT, he helped Papert begin testing the system in Boston-area schools.

From Logo to Lego

“You really have to try hard to get into the mindset of that time, because a computer in those days was something that cost several million dollars,” Abelson says. “And the idea that you would take the most advanced computing research equipment around anywhere, and you would let fifth graders … start playing with it, it was just mind boggling. For the first 10 years of that, people just thought we were nuts.”

But that changed in the 1980s, Abelson says, when personal computers started invading elementary- and secondary-school classrooms. Many of today’s software-company executives first learned the principles of programming from a later version of Logo, in which a virtual turtle executed drawings on a computer screen.

Resnick’s PhD thesis was an extension of Logo called StarLogo, which enabled the interaction of hundreds or even thousands of turtles in complex simulated environments. “I was particularly interested in issues around emergent phenomena and decentralized systems,” Resnick says. “How do bird flocks work? How do the individual actions of individual birds lead to the behavior of a flock? Or how do ant colonies work? Or how do traffic jams form? Or how do market economies work, where there’s lots of individual buyers and sellers, but you get these larger-scale patterns that develop? I was interested in helping people understand how large-scale patterns arise from lots of local interactions. That’s always been a very difficult thing for people to understand. So the idea was to try to provide a way for people to play with those ideas by giving very simple rules to lots of individual objects.”

After graduating, Resnick became a professor in the MIT Media Lab, where he began developing a Logo-like system that would allow children to program robots built from Lego bricks with electromechanically activated moving parts. The project ultimately resulted in Lego’s Mindstorms kits. But along the way, an undergraduate named Andy Begel, who’s now at Microsoft Research, developed a graphical programming language — a precursor to the App Inventor’s programming blocks — as a way to let kids program their Lego robots more intuitively.

Crosstalk

Begel’s system influenced the software for Mindstorms, and Resnick and Klopfer have continued to expand on its central ideas, for two different projects. Klopfer’s is a 3-D adaptation of StarLogo called StarLogo TNG, which can be used to build video games, among other things. Resnick’s is Scratch, a successful educational system, launched in 2007, that allows children to design their own interactive stories and games for the Web. “We’ve been able to stay in touch about what’s going on, and we’re now having a chance to revisit what we were doing and making version two of both of our projects,” Klopfer says. “We made some different design decisions along the way, and I think that was really great as research projects, because we can really learn from each other’s successes and failures.”

For her master’s thesis, Roque decoupled Klopfer’s programming blocks from StarLogo, so that they could be reused in other software systems — and indeed, they were the basis for the App Inventor programming interface. And because Scratch is so popular — hundreds of thousands of registered users have used it create more than a million projects — members of Resnick’s group have been consulting with Google about how to build an online community around the system.

But beyond using actual code that was developed at MIT, Abelson says, the Google App Builder project was also motivated by the same philosophy that drove Papert’s “nutty” experiments in the 1960s. “It strikes me that most of the people who are teaching introductory programming have gotten out of step with the reality of how kids experience computing,” Abelson says. “It’s not about sitting at a desktop computer. It’s about these incredibly powerful computers that you now carry around with you that can do location-aware things, and they can find your friends, and they can make phone calls and do other stuff. App Inventor, as an educational program, is about giving young people who are trying to learn about computing power over the real computing that they’re using in their lives.”

Although unmanned, wheeled rovers have explored the surfaces of the moon and Mars for decades, these vehicles have limits — they can’t crawl inside craters, scale cliffs or travel long distances.

For more than two years, a team of students led by Professor of the Practice of Astronautics and former NASA astronaut Jeffrey Hoffman in MIT’s Department of Aeronautics and Astronautics has been collaborating with engineers from the Charles Stark Draper Laboratory to design and build a prototype for a new type of robotic explorer that would hop over, rather than traverse, a planetary surface. Hopping, they believe, would make it easier for an explorer to access tricky sites and travel greater distances, and thus collect more data during a mission.

Known as the Terrestrial Artificial Lunar and Reduced Gravity Simulator, or Talaris, the three-foot-wide vehicle is a prototype of a larger hopper that would be used in space. The team that built it wants to use Talaris on Earth to test guidance, navigation and control (GNC) software developed by Draper that would then be used to navigate the space-based hopper autonomously. Several graduate students involved with the project will present the latest details about the prototype later this month at the American Institute of Aeronautics and Astronautics Space 2010 conference in California.

The prototype is an outgrowth of MIT’s effort to win the Google Lunar X Prize, a $20 million competition to become the first team to send a privately funded spacecraft to the moon, travel 500 meters across its surface, and transmit video and images back to Earth. Both MIT and Draper are members of Next Giant Leap, one of about 20 teams registered in the competition.

Building Talaris to test the hopper waters

Talaris uses two propulsion systems. The main system consists of four downward-pointing electric ducted fans that provide lift to counter the vehicle’s weight and simulate the gravity environments of different planetary bodies. The second system uses compressed nitrogen gas to maneuver the vehicle as it operates in the simulated gravity conditions. With this setup, the researchers can repeatedly test different navigation algorithms on Earth to perfect the control software.

What distinguishes Talaris from other explorer prototypes is its ability to test how a hopper functions in different gravity scenarios before sending one into space, according to Seamus Tuohy, director of space systems at Draper, which is funding Talaris. “Other organizations had developed little lander prototypes, but the drawback was that they were essentially Earth landers,” Tuohy says. “MIT and Draper wanted to do better — to be able to say that we have the GNC expertise to do hopping in other environments, and that we have demonstrated it with Talaris.”

Although Talaris began with the goal of hopping across the moon, the Talaris team is pushing for hoppers to be used to explore any body in the solar system that has enough gravity to make hopping feasible, including asteroids. “There are limits to the terrain you can access on wheels, and with a hopper, you simply hop in, collect data and hop out,” Hoffman says, noting that hoppers can be used to explore deep craters on the moon that are thought to contain water, measure the magnetism of steep cliffs or set up a network of seismometers by placing sensors at multiple locations.

The ability of hoppers to travel long distances and visit multiple sites is also valuable. Whereas the rovers that have been used to explore Mars since the late 1990s traveled several kilometers over several years, hoppers could travel hundreds of kilometers per hop, depending on their size.

Hopper technology could even enable a human mission to Mars where astronauts orbiting the planet could use a high-bandwidth signal to tele-operate hoppers on the Martian surface, according to Phillip Cunio, an AeroAstro doctorate student who is working with Hoffman to lead the student group working on Talaris. “This would enable direct human oversight of exploration — with all the decision-making capability and flexibility that implies — without the risks of sending human bodies to the surface of Mars,” he says.

Filling the toolkit of planetary exploration

But hoppers do have one drawback: their engines require fuel. Electric rovers, in contrast, only need to recharge their batteries to run their wheels. Given that a hopper is limited to a certain number of hops, the Talaris researchers stress that hoppers should be thought of as an additional tool to complement rovers. Even so, Cunio says, engineers could design hoppers that could be useful even after their fuel runs out. For instance, they might act as solar array-powered rovers, or make fuel from local materials.

For now, the Talaris team is focused on finalizing the construction of Talaris. Although each component has been built and tested individually, the group has yet to test the entire system, which Cunio estimates will weigh about 110 pounds. The team hopes to complete a test hop — Talaris will hop about 20 meters by hovering, moving horizontally and descending — by the end of the calendar year. The researchers predict that if Next Giant Leap is able to secure funding, a large-scale planetary surface hopper explorer could take flight by the end of 2014, the deadline for contestants in the Google Lunar X Prize competition to complete a trip to the moon.

On Friday, Aug. 6, a mathematician at HP Labs named Vinay Deolalikar sent an e-mail to a host of other researchers with a 103-page attachment that purported to answer the most important outstanding question in computer science. That question is whether P = NP, and answering it will earn you $1 million from the Clay Mathematics Institute.

Last fall, MIT News published a fairly detailed explanation of what P = NP means. But roughly speaking, P is a set of relatively easy problems, NP is a set of incredibly hard problems, and if they’re equal, then a large number of computer science problems that seem to be incredibly hard are actually relatively easy. Problems in NP include the factoring of large numbers, the selection of an optimal route for a traveling salesman, and the so-called 3-SAT problem, in which you must find values that satisfy triplets of logical conditions. For instance, is it possible to contrive an attendance list for a party that satisfies the triplet (Alice OR Bob AND Carol) AND (David AND Ernie AND NOT Alice)? (Yes: Bob, Carol, David, and Ernie were there, but Alice wasn’t.) Interestingly, the related 2-SAT problem — which instead uses pairs of conditions, like (Alice OR Ernie) — is in the set of easy problems, P, as is the closely related XOR-SAT problem.

Most computer scientists believe that P doesn’t equal NP, and that’s what Deolalikar claimed to have proved. But by the following Monday, despite being on vacation in the Mediterranean and having time only to glance through the proof, MIT Associate Professor of Electrical Engineering and Computer Science Scott Aaronson had announced on his blog that he would mortgage his house and chip in another $200,000 if Deolalikar’s proof was correct. Last week, Aaronson took a few minutes to answer three questions about P and NP.

Q. Has the proof now been shown conclusively to be wrong?

A. I would say yeah. It was clear a couple days ago that there was a very serious gap in the statistical-physics part of the argument. It was not clear at all that the argument for showing why an NP-complete problem like 3-SAT was hard wouldn’t also show that problems like XOR-SAT are hard. Now, XOR-SAT is a variant of this satisfiability problem, which is known to be in P, which has an efficient solution. So if you’re proving that a problem is hard, but your proof could also be adapted to show that an easy problem is hard, then your proof must be fallacious: it proves too much. That’s the first check that people look for when someone announces a proof of P not equal to NP. Why doesn’t it also work for the easy problems?

The problem with saying that the thing has been conclusively refuted is that whenever anyone points to a problem like this, Vinay Deolalikar has been tending to respond, “Oh, yeah, sure, well, I’m going to address that in my next draft.” So it’s kind of a moving target. But I think it’s absolutely clear right now that at least the existing version does not solve the problem and furthermore wouldn’t solve the problem without some very, very major new ideas.

Q. Why were you so certain that there was a flaw in the proof?

A. P vs. NP is an absolutely enormous problem, and one way of seeing that is that there are already vastly, vastly easier questions that would be implied by P not equal to NP but that we already don’t know how to answer. So basically, if someone is claiming to prove P not equal to NP, then they’re sort of jumping 20 or 30 nontrivial steps beyond what we know today. So the first thing you look for is, What about steps one, two, and three? Can he explain even the easier questions, how he’s answering those? So I looked at the manuscript, and I didn’t see that.

The other check is the one that I already mentioned, which is, Why does the proof fail for variants of NP-complete problems which are known to be easy? What Deolalikar was doing is, he’s trying to argue that 3-SAT is hard by looking at its statistical properties. The problem is that 2-SAT and XOR-SAT, the problems that are easy, have very, very similar statistical properties, so it did not look like something that could distinguish the hard problems from the easy ones.

We have very strong reasons to believe that these problems cannot be solved without major — enormous — advances in human knowledge. So you look at the paper and you don’t see that it’s commensurate with the scale of the problem that it’s claiming to solve. This is not a problem that’s going to be solved by just combining or pushing around the ideas that we already have.

Q. Given that most people are pretty confident that P does not equal NP, what would the proof really do for us?

A. Yes, almost all of us believe already that P is not equal to NP. But this is one of those things where it’s not so much the destination as the journey. It’s the massive amount of new understanding of computation that’s going to be needed to prove such a statement. What are we trying to prove? That for solving all these natural optimization problems, or these search problems, or a proof of a theorem, finding the best schedule for airlines, breaking cryptographic codes — all these different things, that there’s no algorithm, no matter how clever, that’s going to solve them feasibly. So in order to prove such a thing, a prerequisite to it is to understand the space of all possible efficient algorithms. That is an unbelievably tall order. So the expectation is that on the way to proving such a thing, we’re going to learn an enormous amount about efficient algorithms, beyond what we already know, and very, very likely discover new algorithms that will likely have applications that we can’t even foresee right now.

Often in the history of theoretical computer science, the same ideas that you use to prove that something’s impossible can then be turned around to show that something else is possible, and vice versa. The simplest example of that is in cryptography, where you show some problem is hard to solve, and that gives you a code that is useful. But there are many other examples.

Central banks, such as the Federal Reserve in the United States, wield considerable economic clout by setting crucial short-term interest rates, which influence the cost of business loans, among other things. Lower rates tend to spur lending, spending and economic expansion, while higher rates limit lending, thus curbing growth and inflation.

Currently the economy — worldwide and in the United States — is showing sputtering growth. However, the Fed’s short-term interest rates are already near zero, making significant further cuts unrealistic. Given these circumstances, central banks, including the Fed, can try to spur growth through another financial tool, which is increasingly the subject of public discussion: quantitative easing.

At its core, quantitative easing is the attempt by a central bank to inject more money into the economy and to keep long-term interest rates low through the purchase of large amounts of assets, often held by financial institutions. In March 2009, for instance, the Bank of England, the U.K.’s central bank, engaged in quantitative easing by buying U.K. government bonds as well as debt issued by private companies. The means those firms now have more cash on their hands, which in theory makes business lending easier.

Quantitative easing may also lower the rates on five- or 10-year bonds. The Federal Reserve undertook a large quantitative easing measure by buying about $1 trillion in long-term Treasury bonds in March 2009. Taking bonds out of circulation might raise demand for the bonds left in the market, hence raising their prices and lowering their yields (bond prices and yields move in opposite directions) because the bond issuers do not have to promise higher yields in order to entice buyers. These lower rates should make further business investment more likely.

Alternately, economists note, long-term bond rates may stay low not because of the way quantitative easing affects the supply and demand of bonds, but because the central bank uses easing to send a clear policy message.

“It is not clear whether the size of a Fed intervention is large enough to affect these rates materially,” explains Ricardo Caballero, the Ford International Professor of Economics, Macroeconomics, and International Finance at MIT, and head of MIT’s Department of Economics. “If it does affect rates, it is probably not just the consequence of the direct, ‘quantitative’ intervention, but also of the signaling that the Fed will do whatever it takes to keep rates low until the economy is out of the woods.”

On the downside, quantitative easing should in theory create inflationary pressures (since central banks are in effect making new money to buy assets). But inflation has remained low, and indeed dropped, since the March 2009 quantitative easing efforts in the United States and Britain.

“I don’t think there is any major risk involved in quantitative easing,” notes Caballero. “It may be ineffective, perhaps, but not harmful.”

Clinical trials using patients’ own immune cells to target tumors have yielded promising results. However, this approach usually works only if the patients also receive large doses of drugs designed to help immune cells multiply rapidly, and those drugs have life-threatening side effects.

Now a team of MIT engineers has devised a way to deliver the necessary drugs by smuggling them on the backs of the cells sent in to fight the tumor. That way, the drugs reach only their intended targets, greatly reducing the risk to the patient.

The new approach could dramatically improve the success rate of immune-cell therapies, which hold promise for treating many types of cancer, says Darrell Irvine, senior author of a paper describing the technique in the Aug. 15 issue of Nature Medicine.

“What we’re looking for is the extra nudge that could take immune-cell therapy from working in a subset of people to working in nearly all patients, and to take us closer to cures of disease rather than slowing progression,” says Irvine, associate professor of biological engineering and materials science and engineering and a member of MIT’s David H. Koch Institute for Integrative Cancer Research.

The new method could also be used to deliver other types of cancer drugs or to promote blood-cell maturation in bone-marrow transplant recipients, according to the researchers.

T-cell therapy

To perform immune-cell therapy, doctors remove a type of immune cells called T cells from the patient, engineer them to target the tumor, and inject them back into the patient. Those T cells then hunt down and destroy tumor cells. Clinical trials are under way for ovarian and prostate cancers, as well as melanoma.

Immune-cell therapy is a very promising approach to treating cancer, says Glenn Dranoff, associate professor of medicine at Harvard Medical School. However, getting it to work has proved challenging. “The major limitation right now is getting enough of the T cells that are specific to the cancer cell,” says Dranoff, who was not involved in this study. “Another problem is getting T cells to function properly in the patient.” 

To overcome those obstacles, researchers have tried injecting patients with adjuvant drugs that stimulate T-cell growth and proliferation. One class of drugs that has been tested in clinical trials is interleukins — naturally occurring chemicals that help promote T-cell growth but have severe side effects, including heart and lung failure, when given in large doses. 

Irvine and his colleagues took a new approach: To avoid toxic side effects, they designed drug-carrying pouches made of fatty membranes that can be attached to sulfur-containing molecules normally found on the T-cell surface.

In the Nature Medicine study, the researchers injected T cells, each carrying about 100 pouches loaded with the interleukins IL-15 and IL-21, into mice with lung and bone marrow tumors. Once the cells reached the tumors, the pouches gradually degraded and released the drug over a weeklong period. The drug molecules attached themselves to receptors on the surface of the same cells that carried them, stimulating them to grow and divide.

Within 16 days, all of the tumors in the mice treated with T cells carrying the drugs disappeared. Those mice survived until the end of the 100-day experiment, while mice that received no treatment died within 25 days, and mice that received either T cells alone or T cells with injections of interleukins died within 75 days.

The study was funded by the National Institutes of Health, the National Science Foundation, the National Cancer Institute and a gift to the Koch Institute from Curtis ’63 and Kathy Marble.

‘A much simpler procedure’

Irvine’s approach to delivering the adjuvant drugs is both simple and innovative, says Dranoff. “The idea of modifying T cells in the lab to make them work better is something many people are exploring through more complicated approaches such as gene modification,” he says. “But here, the possibility of just attaching a carefully engineered nanoparticle to the surface of cells could be a much simpler procedure.”

While he is now focusing on immune-cell therapy, Irvine believes his cell pouches could be useful for other applications, including targeted delivery of chemotherapy agents. “There are lots of people studying nanoparticles for drug delivery, especially in cancer therapy, but the vast majority of nanoparticles injected intravenously go into the liver or the spleen. Less than 5 percent reach the tumor,” says Irvine, who is also a Howard Hughes Medical Institute Investigator.

With a new way to carry drugs specifically to tumors, scientists may be able to resurrect promising drugs that failed in clinical trials because they were cleared from the bloodstream before they could reach their intended targets, or had to be given in doses so high they had toxic side effects.

Irvine and his colleagues also demonstrated that they could attach their pouches to the surface of immature blood cells found in the bone marrow, which are commonly used to treat leukemia. Patients who receive bone-marrow transplants must have their own bone marrow destroyed with radiation or chemotherapy before the transplant, which leaves them vulnerable to infection for about six months while the new bone marrow produces blood cells.

Delivering drugs that accelerate blood-cell production along with the bone-marrow transplant could shorten the period of immunosuppression, making the process safer for patients, says Irvine. In the Nature Medicine paper, his team reports successfully enhancing blood-cell maturation in mice by delivering one such drug along with the cells.

Irvine is now starting to work on making sure the manufacturing process will yield nanoparticles safe to test in humans. Once that is done, he hopes the particles could be used in clinical trials in cancer patients, possibly within the next two or three years.

A team of MIT students has been working on testing a rapid-recharging system that could help to change public perceptions about electric vehicles and their practicality. They have already done extensive testing of the system with an individual battery cell and with a motorcycle they converted to all-electric operation, and in coming months they hope to be able to demonstrate the system on a full-sized sedan they converted.

The goal is to demonstrate that recharging can be accomplished routinely in under 30 minutes without severely reducing the operating lifetime of the batteries or causing other problems. In the year since the MIT Electric Vehicle Team started working on the project, new and established companies have begun to offer commercial rapid-recharging systems, and Japan has officially adopted a new standard for the connectors for such systems and has begun installing the systems in more than 100 locations. The Nissan Leaf, a pure electric five-passenger car to be introduced in the U.S. later this year, is already capable of rapid recharging in 30 minutes in places that have the necessary “Level III” charging system. (So far, there is just one such station in the U.S., in Portland, Oregon).

Next week, Lennon Rodgers, a doctoral student in mechanical engineering and a member of the MIT Electric Vehicle Team, will present a paper on the team’s rapid-charging tests at the 12th International Conference on Advanced Vehicle and Tire Technologies in Montreal. The paper was co-authored by fellow team members Radu Gogoana ’10, a master’s student in mechanical engineering, Paul Karplus (an undergraduate at Stanford) and Michael Nawrot ’11.

Rapid charging, also known as Level III, requires much higher voltages and current than what is supplied by conventional household circuits. The Japanese rapid-charging standard, called CHAdeMO, provides DC power at up to 500 volts with a current of 125 amps. Typical chargers operate on standard AC power, using either 110 volt household current (Level I), which generally can recharge an electric car’s batteries overnight, or special systems (similar to those needed for electric stoves or clothes dryers) that use 220 volts (Level II), which can cut the charging time in half. “Rapid charging” systems typically refer to those that can charge the batteries to at least 80 percent of capacity within 30 minutes.

Because of the large power requirements, this is not something you’d ever do in your home garage. Rather, this fast-recharge technology might be installed in central recharging stations comparable to today’s gas stations, where the cost of the necessary infrastructure could be warranted and where a fast turnaround is necessary. In many cases, rapid charging systems can provide a 50 percent charge — typically enough to travel 50 miles — in under 5 minutes, comparable to the time it takes to fill a gas tank.

While rapid charging — largely being promoted by companies with long commercial experience with recharging industrial fork lifts and similar vehicles — is beginning to attract attention, there has been relatively little testing on the effects of repeated rapid charging on battery life and performance. “Is it damaging over time? That’s the issue we wanted to study,” says Rodgers. That’s the kind of data the MIT team was collecting in an attempt to prove the potential for this technology.

Rodgers says that the chemistry used in lithium-ion batteries made by the MIT spinoff company A123 Systems is the best suited for rapid charging, and the company’s website declares that the batteries are capable of being fully recharged in 15 minutes. These batteries, based on research carried out at MIT in the lab of Yet-Ming Chiang, professor of materials science and engineering, have been selected for several planned new electric vehicles including cars from Fisker Automotive and buses and trucks from Daimler and Navistar.


A time-lapse video showing members of MIT’s Electric Vehicle Team working on the conversion of a car and a motorcycle from conventional gasoline power to all-electric operation. The conversions used batteries donated by MIT spinoff company A123 Systems.
Video: Melanie Gonick; still images/footage: MIT EVT

In the team’s tests, they ran one of these battery cells through 1,500 charge and discharge cycles, using an automated system. After 1,500 cycles, the battery had lost less than 10 percent of its initial power capacity, Rodgers says. The team used a fan to prevent overheating, which by stressing the chemical and mechanical components can lead to degradation.

To test a rapid-charging system under realistic conditions, the team converted a motorcycle to all-electric operation, and then performed a successful rapid-charging test, reaching more than 80 percent charge within 10 minutes.

The MIT EV Team has also completed the conversion of a 2010 Mercury Milan hybrid (donated by Ford) into a pure electric vehicle. The initial conversion was successfully completed last summer, and this summer they have been making major improvements — reinstalling the 8,000 lithium ion phosphate battery cells provided by A123, rewiring the system with a new control system, adding a powerful cooling system for the batteries, and making changes to make the car street-legal. They hope to use the car for testing of rapid charging technology, although they are still looking for funding to get the necessary equipment. Commercial rapid-charging systems can cost tens of thousands of dollars.

Kristen Helsel, vice president of EV Solutions for Aerovironment, a company that makes charging systems for electric vehicles, says that it’s unlikely anyone will start installing rapid-charging stations in the U.S. in substantial numbers until the country adopts an official standard, and “I don’t expect it in the near term. There are still multiple designs under consideration.” But a rapid charging capability is going to be crucial for widespread acceptance of electric vehicles, no matter what their driving range is on a full charge, because people will always want the possibility of being able to go farther, she says. In the meantime, research such as that being carried out by the MIT EV Team can play a useful role, she says. “Better batteries are coming, and because of that the ability to charge at any level is going to be a constantly evolving thing. We need to continually evolve the technology, and to better understand the effects of different things” on battery life and other factors, she says. “There’s all sorts of good work that needs to be done.”

For car designers, Rodgers says, there is a tradeoff they need to consider: They can include larger battery packs that provide a longer driving range, but are more difficult to recharge rapidly, or smaller packs that give a shorter range but cost less and can more easily be charged rapidly.

In addition to analyzing battery performance, the team analyzed the impact that rapid charging of electric vehicles might have on the electric grid. They concluded that spikes of usage that might present problems for the grid could be eliminated by using an intermediate battery system. Instead of directly charging the vehicle from the grid, a large battery pack — perhaps using batteries recycled from other cars — could be slowly charged using a “trickle charge” from the grid, thus using low-cost, off-peak power, and then rapidly transfer its charge to the vehicle’s batteries.

Neutrinos, elementary particles generated by nuclear reactions in the sun, suffer from an identity crisis as they cross the universe, morphing between three different “flavors.” Their antimatter counterparts (which are identical in mass but opposite in charge and spin) do the same thing.

A team of physicists including some from MIT has found surprising differences between the flavor-switching behavior of neutrinos and antineutrinos. If confirmed, the finding could help explain why matter, and not antimatter, dominates our universe.

“People are very excited about it because it suggests that there are differences between neutrinos and antineutrinos,” says Georgia Karagiorgi, an MIT graduate student and one of the leaders of the analysis of experimental data produced by the Booster Neutrino Experiment (MiniBooNE) at the Fermi National Accelerator Laboratory.

The new result, announced in June and submitted to the journal Physical Review Letters, appears to be one of the first observed violations of CP symmetry: the theory that matter and antimatter should behave in the same way. CP symmetry violation has been seen before in quarks, another type of elementary particle that makes up protons and neutrons, but never in neutrinos or electrons.

The finding could also force physicists to revise their Standard Model, which catalogs all of the known particles that make up matter. The model now posits only three flavors of neutrino, but a fourth (or fifth or sixth) may be necessary to explain the new results.

“If this should be proven to be correct, it would have major implications for particle physics,” says John Learned, professor of physics at the University of Hawaii, who is not part of the MiniBooNE team.

So far, the researchers have enough data to present their results with a confidence level of just below 99.7 percent (also called 3 sigma), which is not high enough to claim a new discovery. To reach that level, 5-sigma confidence (99.99994 percent) is required. “People are going to rightfully demand a really clean, 5-sigma result,” says Learned.

Unexpected oscillations

Since the 1960s, physicists have been gathering evidence that neutrinos can switch, or oscillate, between three different flavors — muon, electron and tau, each of which has a different mass. However, they have not yet been able to rule out the possibility that more types of neutrino might exist.

In an effort to help nail down the number of neutrinos, MiniBooNE physicists send beams of neutrinos or antineutrinos down a 500-meter tunnel, at the end of which sits a 250,000-gallon tank of mineral oil. When neutrinos or antineutrinos collide with a carbon atom in the mineral oil, the energy traces left behind allow physicists to identify what flavor of neutrino took part in the collision. Neutrinos, which have no charge, rarely interact with other matter, so such collisions are rare.

MiniBooNE was set up in 2002 to confirm or refute a controversial finding from an experiment at the Liquid Scintillator Neutrino Detector (LSND) at Los Alamos National Laboratory. In 1990, the LSND reported that a higher-than-expected number of antineutrinos appeared to be oscillating over relatively short distances, which suggests the existence of a fourth type of neutrino, known as a “sterile” neutrino.

In 2007, MiniBooNE researchers announced that their neutrino experiments did not produce oscillations similar to those seen at LSND. At the time, they assumed the same would hold true for antineutrinos. “In 2007, I would have told you that you can pretty much rule out LSND,” says MIT physics professor Janet Conrad, a member of the MiniBooNE collaboration and an author of the new paper.

MiniBooNE then switched to antineutrino mode and collected data for the next three years. The research team didn’t look at all of the data until earlier this year, when they were shocked to find more oscillations than would be expected from only three neutrino flavors — the same result as LSND.

Already, theoretical physicists are posting papers online with theories to account for the new results. However, “there’s no clear and immediate explanation,” says Karsten Heeger, a neutrino physicist at the University of Wisconsin. “To nail it down, we need more data from MiniBooNE, and then we need to experimentally test it in a different way.”

The MiniBooNE team plans to collect antineutrino data for another 18 months. Conrad also hopes to launch a new experiment that would use a cyclotron, a type of particle accelerator in which particles travel in a circle instead of a straight line, to help confirm or refute the MiniBooNE results.

The so-called “beads-on-a-string” phenomenon can be demonstrated with a very simple experiment. Stretch a glob of saliva between your thumb and forefinger, and you should see a string of beads form, before the strand eventually breaks.

Researchers have now discovered precisely why this occurs — a finding that could be used to improve industrial processes and for administering drugs in “personalized medicine.”

Saliva and other complex “viscoelastic” fluids, such as shaving gels and shampoo, contain long chains of molecules called polymers. In the case of saliva, the polymers are proteins known as mucopolysaccharides. In comparison, liquids such as water and other so-called “Newtonian” fluids do not form beads when stretched, because they lack dissolved polymers.

MIT’s Gareth McKinley, professor of mechanical engineering, and researchers at Purdue University and Rice University found that in addition to the elasticity of a fluid a key contributing factor in the beading mechanism is fluid inertia, or the tendency of a fluid to keep moving unless acted upon by an external force. Other major elements are a fluid's viscosity; the time it takes a stretched polymer molecule to "relax," or snap back to its original shape when stretching is stopped; and the "capillary time," or how long it would take for the surface of the fluid strand to vibrate if plucked.


Gareth McKinley describes how a material can be "stringy."
Video: Melanie Gonick; still images/additional footage: Gareth McKinley/Colin McKinley/NASA

The researchers discovered that the bead formation depends on a delicate balance of two ratios: the viscous force compared to inertial force and the relaxation time compared to the capillary time. The researchers reported their findings in the June 6 issue of Nature Physics.

“What makes this result a breakthrough is that bead formation is more complicated than a lot of people had assumed. The insight that inertia matters is a surprise. Earlier attempts to explain this phenomenon without inertia are either wrong or assume inappropriate physics,” says Yuriko Renardy, professor of mathematics at Virginia Tech University, who was not involved in the research.

Understanding how these beads arise could enable researchers to design systems that precisely control the formation of the beads, leading to improvements in various technologies, such as inkjet printing. The information also might be used to develop drug-dispensing systems for patients with certain disorders that require precise doses of medication depending on daily blood measurements.

Material from a Purdue University press release was used in this story.

People with type 1 diabetes must keep a careful eye on their blood glucose levels: Too much sugar can damage organs, while too little deprives the body of necessary fuel. Most patients must prick their fingers several times a day to draw blood for testing.

To minimize that pain and inconvenience, researchers at MIT’s Spectroscopy Laboratory are working on a noninvasive way to measure blood glucose levels using light.

First envisioned by Michael Feld, the late MIT professor of physics and former director of the Spectroscopy Laboratory, the technique uses Raman spectroscopy, a method that identifies chemical compounds based on the frequency of vibrations of the bonds holding the molecule together. The technique can reveal glucose levels by simply scanning a patient’s arm or finger with near-infrared light, eliminating the need to draw blood.

Spectroscopy Lab graduate students Ishan Barman and Chae-Ryon Kong are developing a small Raman spectroscopy machine, about the size of a laptop computer, that could be used in a doctor’s office or a patient’s home. Such a device could one day help some of the nearly 1 million people in the United States, and millions more around the world, who suffer from type 1 diabetes.

Researchers in the Spectroscopy Lab have been developing this technology for about 15 years. One of the major obstacles they have faced is that near-infrared light penetrates only about half a millimeter below the skin, so it measures the amount of glucose in the fluid that bathes skin cells (known as interstitial fluid), not the amount in the blood. To overcome this, the team came up with an algorithm that relates the two concentrations, allowing them to predict blood glucose levels from the glucose concentration in interstitial fluid.

However, this calibration becomes more difficult immediately after the patient eats or drinks something sugary, because blood glucose soars rapidly, while it takes five to 10 minutes to see a corresponding surge in the interstitial fluid glucose levels. Therefore, interstitial fluid measurements do not give an accurate picture of what’s happening in the bloodstream.

To address that lag time, Barman and Kong developed a new calibration method, called Dynamic Concentration Correction (DCC), which incorporates the rate at which glucose diffuses from the blood into the interstitial fluid. In a study of 10 healthy volunteers, the researchers used DCC-calibrated Raman spectroscopy to significantly boost the accuracy of blood glucose measurements — an average improvement of 15 percent, and up to 30 percent in some subjects.

The researchers described the new calibration method and results in the July 15 issue of the journal Analytical Chemistry. In addition to Feld, Barman and Kong, authors include Ramachandra Rao Dasari, associate director of the Spectroscopy Lab, and former postdoctoral associate Gajendra Pratap Singh.

Michael Morris, professor of chemistry at the University of Michigan, says the group appears to have solved a problem that has long stymied researchers. “Getting optical glucose measurements of any sort is something people have been trying to do since the 1980s,” says Morris, who was not involved in this study. “Usually people report that they can get good measurements one day, but not the next, or that it only works for a few people. They can’t develop a universal calibration system.”

Morris says the noninvasive nature of Raman spectroscopy could help boost quality of life for diabetes patients, but that to be practical, any device would need to become more affordable and very simple to use. The Spectroscopy Lab researchers believe that the smaller machine they are now developing should substantially drive down costs by miniaturizing and reducing the complexity of the instrument.

Barman and Kong plan to launch a clinical study to test the DCC algorithm in healthy volunteers this fall. Their work is funded by the National Institutes of Health and National Center for Research Resources.

In October, Barman will receive the Tomas A. Hirschfeld Award at the Federation of Analytical Chemistry and Spectroscopy Societies Conference, for his work on improving spectroscopy-based glucose measurements.

For decades, physicists have been trying to reconcile the two major theories that describe physical behavior. The first, Einstein’s theory of general relativity, uses gravity — forces of attraction — to explain the behavior of objects with large masses, such as falling trees or orbiting planets. However, at the atomic and subatomic level, particles with negligible masses are better described using another theory: quantum mechanics.

A “theory of everything” that marries general relativity and quantum mechanics would encompass all physical interactions, no matter the size of the object. One of the most popular candidates for a unified theory is string theory, first developed in the late 1960s and early 1970s.

String theory holds that electrons and quarks (the building blocks of larger particles) are one-dimensional oscillating strings, not the dimensionless objects they are traditionally thought to be.

Physicists are divided on whether string theory is a viable theory of everything, but many agree that it offers a new way to look at physical phenomena that have otherwise proven difficult to describe. In the past decade, physicists have used string theory to build a connection between quantum and gravitational mechanics, known as gauge/gravity duality.

MIT physicists, led by Hong Liu and John McGreevy, have now used that connection to describe a specific physical phenomenon — the behavior of a type of high-temperature superconductor, or a material that conducts electricity with no resistance. The research, published in the Aug. 5 online edition of Science, is one of the first to show that gauge/gravity duality can shed light on a material’s puzzling physical behavior.

So far, the team has described a few aspects of behavior of a type of superconducting materials called cuprates. However, the researchers hope their work could lead to more general theories to describe other materials, and eventually predict their behavior. “That’s the ultimate theoretical goal, and we haven’t really achieved that,” says Liu.

MIT graduate student Nabil Iqbal and recent PhD recipients Thomas Faulkner and David Vegh are also authors of the paper.

Strange metals

In 1986, physicists discovered that cuprates (ceramic compounds that contain copper) can superconduct at relatively high temperatures (up to 135 degrees Celsius above absolute zero).

At the atomic level, cuprates are classified as a “many-body system” — essentially a vast collection of electrons that interact with each other. Such systems are usually described using quantum mechanics. However, so far, physicists have found it difficult to describe cuprates, because their behavior is so different from other materials. Understanding that behavior could help physicists find new materials that superconduct at even higher temperatures. These new materials would have potentially limitless applications.

Unlike most materials, cuprates do not obey Fermi’s laws, a set of quantum-mechanics principles that govern microscopic behavior at very low temperatures (close to absolute zero, or -273 degrees Celsius). Instead, cuprates become superconductors. Just above the temperature at which they begin to superconduct, they enter a state called the “strange metal” state.

In this study, the researchers focused on two properties that distinguish those cuprate strange metals from Fermi liquids. In ordinary Fermi liquids, electrical resistivity and the rates of electron scattering (deflection from their original course caused by interactions with each other) are both proportional to the temperature squared. However, in cuprates (and other superconducting non-Fermi liquids), electron scattering and resistivity are proportional to the temperature. “There’s really no theory of how to explain that,” says Liu.

Drawing a parallel

Using gauge/gravity duality — the connection between quantum and gravitational mechanics — the MIT team identified a system that has the same unusual properties as strange metals, but could be explained by gravitational mechanics. In this case, the model they used was a gravitational system with a black hole. “It’s a mathematical abstraction which we hope may shed light on the physics of the real system,” says Liu. In their model, they can study behavior at high and low energy (determined by how the excitation energy of a single electron compares to the average energy of an electron in the system), and it turns out that at low energy, the black-hole model exhibits many of the same unusual traits seen in non-Fermi liquids such as cuprates.

For example, in both systems, when an electron at the lowest possible energy level is excited (by a photon or another particle), the resulting interaction between the electron and the hole left behind cannot be described as a quasiparticle (as it can in ordinary metals), because the electron excitation decays so quickly. (The electrons decay so quickly because their scattering rate is proportional to the temperature.) Furthermore, the electrical resistance of the black-hole system is directly proportional to temperature — just as it is in cuprates.

Gauge/gravity duality offers a “map” that correlates certain features of the black-hole model to corresponding features of strange metals. Therefore, once the physicists calculated the features of the model, using general relativity, those values could be translated to the corresponding values in the strange-metal system. For example, the value of an electromagnetic field in the gravitational system could correspond to the density of electrons in the quantum system.

Physicists have previously used gauge/gravity duality to describe some characteristics of quark gluon plasma, the “hot soup” of elementary particles that existed in the first millionths of a second after the Big Bang. However, this is the first time it has been used to give insight into a type of condensed matter (solids and liquids are condensed matter).

For that reason, the paper should have a significant impact in theoretical physics, says Joseph Polchinski, a theoretical physicist at the University of California at Santa Barbara. “Whenever people have systems they can’t understand in other ways, this might be a tool to try to understand it,” he says.

The MIT team believes the approach could shed light on a group of rare metal compounds known as heavy fermion metals, whose electrons behave as if their masses were 100 to 1,000 times greater than those in ordinary metals. They also display some of the same non-Fermi liquid behavior seen in the strange metal phase of cuprates.

By combining origami and electrical engineering, researchers at MIT and Harvard are working to develop the ultimate reconfigurable robot — one that can turn into absolutely anything. The researchers have developed algorithms that, given a three-dimensional shape, can determine how to reproduce it by folding a sheet of semi-rigid material with a distinctive pattern of flexible creases. To test out their theories, they built a prototype that can automatically assume the shape of either an origami boat or a paper airplane when it receives different electrical signals. The researchers reported their results in the July 13 issue of the Proceedings of the National Academy of Sciences.

As director of the Distributed Robotics Laboratory at the Computer Science and Artificial Intelligence Laboratory (CSAIL), Professor Daniela Rus researches systems of robots that can work together to tackle complicated tasks. One of the big research areas in distributed robotics is what’s called “programmable matter,” the idea that small, uniform robots could snap together like intelligent Legos to create larger, more versatile robots.

The U.S. Defense Department’s Defense Advanced Research Projects Agency (DARPA) has a Programmable Matter project that funds a good deal of research in the field and specifies “particles … which can reversibly assemble into complex 3D objects.” But that approach turns out to have drawbacks, Rus says. “Most people are looking at separate modules, and they’re really worried about how these separate modules aggregate themselves and find other modules to connect with to create the shape that they’re supposed to create,” Rus says. But, she adds, “actively gathering modules to build up a shape bottom-up, from scratch, is just really hard given the current state of the art in our hardware.”

A new wrinkle

So Rus has been investigating alternative approaches, which don’t require separate modules to locate and connect to each other before beginning to assemble more complex shapes. Fortunately, also at CSAIL is Erik Demaine, who joined the MIT faculty at age 20 in 2001, becoming the youngest professor in MIT history. One of Demaine’s research areas is the mathematics of origami, and he and Rus hatched the idea of a flat sheet of material with tiny robotic muscles, or actuators, which could fold itself into useful objects. In principle, flat sheets with flat actuators should be much easier to fabricate than three-dimensional robots with enough intelligence that they can locate and attach to each other.


Erik Demaine describes how the prototype robot can automatically fold itself into an airplane or an origami boat.
Video: Melanie Gonick; original footage: E. Hawkes/B. An/N.M. Benbernou/H. Tanaka/S. Kim/E.D. Demaine/D. Rus/R.J. Wood
About a year ago, Demaine and several colleagues — including his dad, who’s a visiting scientist at CSAIL, master’s student Aviv Ovadya, and Nadia Benbernou, a PhD student in applied mathematics who’s a coauthor on the new paper — proved that a large enough sheet creased in what’s called the “box pleat pattern” could be folded into a close approximation of any possible three-dimensional shape. The box pleat pattern divides the sheet into squares, each of which has a diagonal crease across it; but if two squares share an edge, their diagonal creases are mirror images. This paper marked the first time that the universality of a crease pattern had been shown, although Demaine and his collaborators have since proved that other crease patterns are universal as well.

Based on this result, Demaine, Rus, Harvard’s Robert Wood, and others developed algorithms that, given an arbitrary three-dimensional shape, could generate a sequence of folds that would produce it from a box-pleated sheet.

But as yet, no robotic system existed that could execute that sequence of folds automatically. In principle, a universal origami robot would have actuators on both sides of every crease, so that the sheet could fold in either direction at any point. But a system that complex is difficult to build, and before undertaking it, the researchers hoped to demonstrate the viability of their approach.

Theory into practice

So they designed yet another set of algorithms that, given sequences of folds for several different shapes, would determine the minimum number of actuators necessary to produce all of them. Then they set about building a robot that could actually assume multiple origami shapes. Their prototype, made from glass-fiber and hydrocarbon materials, with an elastic plastic at the creases, is divided into 16 squares about a centimeter across, each of which is further divided into two triangles. The actuators consist of a shape-memory alloy — a metal that changes shape when electricity is applied to it. Each triangle also has a magnet in it, so that it can attach to its neighbors once the right folds have been performed.

The sheet is too small — or, depending on your perspective, the triangles are too big — to do anything very useful yet. But, in principle, it’s possible to build either a similar sheet with much smaller moving parts, or a larger sheet with similar-sized moving parts. With a finer-grained sheet, “you could imagine downloading the new iPhone,” Demaine says. “In the same way that you download the latest CD from your favorite artist totally electronically, you could imagine downloading shapes electronically, and programming hardware the same way you program software.” Larger sheets could enable “a tent that can adapt its shape according to the wind so that it doesn’t blow over,” Demaine says, or “a solar cell that can adjust its shape to the sun and the cloud patterns and whatnot.”

“It’s a very nice merger of the abstract mathematical theory and the practical world,” says Joseph O’Rourke, the chair of the computer science department at Smith College, who was one of the referees for the PNAS paper. “It may be appropriate to say it’s some type of breakthrough.” O’Rourke points out, however, that Demaine’s universality proof relied on the assumption that the triangles in the box-pleated material were themselves somewhat flexible, which may not be the case with, for instance, tiny sheets of material carved out of silicon.

Demaine agrees but says that he and his group are approaching the problem from two directions. On one hand, they’re trying to mathematically model the flexibility of the triangles, so that the folding algorithms can take that into account. On the other, they’re taking some tentative first steps toward a theory of origami with rigid materials. “We are sorely lacking in rigid-origami theory, as Joe points out and I'd be quick to agree,” Demaine says.

So-called “cap and trade” legislation has often been portrayed as a regressive policy — one that would hit poor people the hardest. A new MIT study concluded that this is not the case.

The U.S. House of Representatives passed a cap-and-trade bill last year, and different versions of that bill had been working their way through the Senate until being yanked from consideration last month.

The study, co-authored by researchers at the MIT Joint Program on the Science and Policy of Global Change and at Tufts University, found that under all three versions of the bill submitted so far, the costs would fall hardest on wealthier households, and that lower-income households would see no change or a net benefit.

The basic concept of cap and trade is that greenhouse-gas emissions would be capped at some level (usually about the present level, or the level from a past year), and companies that produce those emissions, such as electric utilities, would receive permits for a given amount. If they choose to install lower-emissions plants, they would end up with extra permits, which could then be freely traded — that is, sold to companies that are unable to stay within their allotted limits.

The MIT study assumed that a cap-and-trade measure would take effect in 2012, and it estimated the legislation’s financial effects on U.S. households beginning in 2015 and continuing every five years through 2050. It found that incomes of the poorest Americans — households that earn less than $10,000 a year — would show a net increase of up to 1.5 percent in 2015, depending on the particular bill. Households earning less than $50,000 a year — about 45 percent of all households — would see some gains, or at worst no change. Those in the very highest income bracket would pay more, with total additional costs in 2015 amounting to less than 0.5 percent of their incomes. According to the study, these effects would become more pronounced over time.

The research, published this last month in the Berkeley Electronic Journal of Economic Analysis and Policy, was unique in its analysis of both the income and expense impacts of the legislation, and of regional differences on a scale that in some cases went down to the level of individual states.

Older computer models used to analyze the impacts of cap-and-trade legislation just looked at one or two typical households, explains John Reilly, co-director of the MIT Joint Program and one of the authors of the new study. “We decided to look at how carbon policies are going to affect different people,” Reilly says. “Conventional wisdom holds that by raising the cost of energy, policies to price carbon will have a negative effect on everyone,” he says. “Our research concludes that, by itself, pricing carbon tends to be progressive, rather than regressive.”

The database developed by the team, which also included researchers Sebastian Rausch and Sergey Paltsev of the MIT Joint Program and Gilbert Metcalf of Tufts, breaks the information down regionally, as well as by income level.

Overall, for all the different regions and income levels, the results were quite consistent: Those with the lowest incomes came out ahead, while those with higher incomes bore most of the additional costs. Reilly calls this finding “really unexpected,” and attributes it primarily to the fact that the study looked at both households’ incomes and expenses.

To understand why poorer households may fare better than richer ones, consider that those in the lowest income echelons tend to derive a larger portion of their incomes from government programs such as welfare or Social Security. These programs are all indexed to inflation, and because the cap-and-trade measures are expected to add to the cost-of-living index, those increases would be compensated by the adjustments.

The impacts of the additional costs would also be mitigated by mechanisms built into the bills. Among these provisions is one that would distribute dividends to households or regions likely to feel the greatest impacts of the carbon charges. “One way or the other,” Reilly says, “all of the proposals actually benefit low-income households, because the allowance allocation they receive is greater than their increases in energy costs and effects on income.”

Laurie Johnson, chief economist for the Natural Resources Defense Council, says that “it’s not at all surprising that the authors find the legislation [in its various specific forms] to be progressive, because some of the proceeds from the sale of pollution permits are redistributed back to households on a per-capita basis.” In fact, she says, if anything, the economic impacts on most people will be even less than this study suggests (or actually beneficial), because it doesn’t include the environmental benefits to be gained from the reductions in emissions. “Were these included,” she wrote in a blog post about the new findings, “the discussion of ‘costs’ of climate legislation would likely turn on its head, and instead be about benefits and savings.”

The researchers note that their analysis could help fine-tune cap-and-trade proposals: Policymakers, they say, could use the findings to revise legislation and mitigate the negative effects on particular regions or income levels. By using the computer model they developed, Rausch says, it’s now possible to take any specific proposal or modification of the existing bills and “run it through the model and see the effects, taking into account all the complex interactions.”

Although they have not yet analyzed the scaled-back legislation now being offered in place of cap and trade, Reilly says it contains provisions that would add to energy costs without generating revenue to offset these costs for lower-income households. Thus it might bring higher costs than the original legislation, while achieving much less reduction in emissions.

Many students learn about the Doppler effect in physics class, typically as part of a discussion of why the pitch of a siren is higher as an ambulance approaches and then lower as the ambulance passes by. The effect is useful in a variety of different scientific disciplines, including planetary science: Astronomers rely on the Doppler effect to detect planets outside of our solar system, or exoplanets. To date, 442 of the 473 known exoplanets have been detected using the Doppler effect, which also helps planetary scientists glean details about the newly found planets.

The Doppler effect, or Doppler shift, describes the changes in frequency of any kind of sound or light wave produced by a moving source with respect to an observer. Waves emitted by an object traveling toward an observer get compressed — prompting a higher frequency — as the source approaches the observer. In contrast, waves emitted by a source traveling away from an observer get stretched out.

In astronomy, that source can be a star that emits electromagnetic waves; from our vantage point, Doppler shifts occur as the star orbits around its own center of mass and moves toward or away from Earth. These wavelength shifts can be seen in the form of subtle changes in its spectrum, the rainbow of colors emitted in light. When a star moves toward us, its wavelengths get compressed, and its spectrum becomes slightly bluer. When the star moves away from us, its spectrum looks slightly redder.

To observe the so-called red shifts and blue shifts over time, planetary scientists use a high-resolution prism-like instrument known as a spectrograph that separates incoming light waves into different colors. In every star’s outer layer, there are atoms that absorb light at specific wavelengths, and this absorption appears as dark lines in the different colors of the star’s spectrum that are recorded from the light emanating from the star. Researchers use the shifts in these lines as convenient markers by which to measure the size of the Doppler shift.

If the star exists by itself — that is, if there is no exoplanet or companion star in its stellar system — then there will be no change in the pattern of its Doppler shifts over time. But if there is a planet or companion star in the system, the gravitational pull of this unseen body or star will perturb the host star’s movement at certain parts of its orbit, producing a noticeable change in the overall pattern and size of Doppler shifts over time. In other words, the pattern of a star’s Doppler shifts can change over time as a result of gravity affecting the star’s motion. “If this shift is large, then it must be caused by another star pulling it, but if this shift is small, then it is likely caused by a low-mass body like an exoplanet,” explains Joshua Winn, an assistant professor in MIT’s Department of Physics. As part of his work at MIT’s Kavli Institute for Astrophysics and Space Research, Winn studies the relationship between an exoplanet’s orbit and its parent star’s rotation for clues about how the planet may have formed.

How a planet’s Doppler shift changes over time can also shed light on the planet’s orbital period (the length of its “year”), the shape of its orbit and its minimum possible mass. Recently, Kavli postdoc Simon Albrecht used the Doppler effect to detect color shifts in the light absorbed by an exoplanet, which indicated strong winds in the planet’s atmosphere.

Doppler shifts are used in many fields besides astronomy. By sending radar beams into the atmosphere and studying the changes in the wavelengths of the beams that come back, meteorologists use the Doppler effect to detect water in the atmosphere. The Doppler phenomenon is also used in healthcare with echocardiograms that send ultrasound beams through a body to measure changes in blood flow to make sure that a heart valve is working properly or to diagnose vascular diseases. Police also rely on the Doppler effect when they use a radar gun to bounce radio beams off of your car; the change in frequency between the directed and reflected beams provides a measure of your car’s speed.

Like an ice cube on a warm day, most materials melt — that is, change from a solid to a liquid state — as they get warmer. But a few oddball materials do the reverse: They melt as they get cooler. Now a team of researchers at MIT has found that silicon, the most widely used material for computer chips and solar cells, can exhibit this strange property of “retrograde melting” when it contains high concentrations of certain metals dissolved in it.

The material, a compound of silicon, copper, nickel and iron, “melts” (actually turning from a solid to a slush-like mix of solid and liquid material) as it cools below 900 degrees Celsius, whereas silicon ordinarily melts at 1414 degrees C. The much lower temperatures make it possible to observe the behavior of the material during melting, based on specialized X-ray fluorescence microprobe technology using a synchrotron — a type of particle accelerator — as a source.

The material and its properties are described in a paper just published online in the journal Advanced Materials. Team leader Tonio Buonassisi, the SMA Assistant Professor of Mechanical Engineering and Manufacturing, is the senior author, and the lead authors are Steve Hudelson MS ’09, and postdoctoral fellow Bonna Newman PhD ’08.

The findings could be useful in lowering the cost of manufacturing some silicon-based devices, especially those in which tiny amounts of impurities can significantly reduce performance. In the material that Buonassisi and his researchers studied, impurities tend to migrate to the liquid portion, leaving regions of purer silicon behind. This could make it possible to produce some silicon-based devices, such as solar cells, using a less pure, and therefore less expensive, grade of silicon that would be purified during the manufacturing process.

“If you can create little liquid droplets inside a block of silicon, they serve like little vacuum cleaners to suck up impurities,” Buonassisi says. This research could also lead to new methods for making arrays of silicon nanowires — tiny tubes that are highly conductive to heat and electricity.

Buonassisi predicted in a 2007 paper that it should be possible to induce retrograde melting in silicon, but the conditions needed to produce such a state, and to study it at a microscopic level, are highly specialized and have only recently become available. To create the right conditions, Buonassisi and his team had to adapt a microscope “hot-stage” device that allowed the researchers to precisely control the rate of heating and cooling. And to actually observe what was happening as the material was heated and cooled, they drew upon high-power synchrotron-based X-ray sources at Lawrence Berkeley National Laboratory in California and at Argonne National Laboratory in Illinois (researchers from both national labs are co-authors of the paper).

The research was supported by the U.S. Department of Energy, the National Science Foundation, the Clare Booth Luce Foundation, Doug Spreng and the Chesonis Family Foundation, and some equipment was provided by McCrone Scientific.

The material for the tests consisted of a kind of sandwich made from two thin layers of silicon, with a filling of copper, nickel and iron between them. This was first heated enough to cause the metals to dissolve into the silicon, but below silicon’s melting point. The amount of metal was such that the silicon became supersaturated — that is, more of the metal was dissolved in the silicon than would normally be possible under stable conditions. For example, when a liquid is heated, it can dissolve more of another material, but then when cooled down it can become supersaturated, until the excess material precipitates out.

In this case, where the metals were dissolved into the solid silicon, “if you begin cooling it down, you hit a point where you induce precipitation, and it has no choice but to precipitate out in a liquid phase,” Buonassisi says. It is at that point that the material melts.

Matthias Heuer, a senior research scientist at Calisolar, a solar-energy startup company, says this work is “unique and new to our field,” and it “allows some very good insight into how transition metals and structural defects interact.” But he adds that there are a number of questions still to be answered in follow-up research: “Now that we know liquid inclusions can form, the question is, how efficient as sinks for impurities are they? How stable are they? Can they keep the impurities localized during other process steps — for example, during the final firing process of a solar cell?”

Among the most common tools in electrical engineering and computer science are rectangular grids of numbers known as matrices. The numbers in a matrix can represent data: The rows, for instance, could represent temperature, air pressure and humidity, and the columns could represent different locations where those three measurements were taken. But matrices can also represent mathematical equations. If the expressions t + 2p + 3h and 4t + 5p + 6h described two different mathematical operations involving temperature, pressure and humidity measurements, they could be represented as a matrix with two rows, [1 2 3] and [4 5 6]. Multiplying the two matrices together means performing both mathematical operations on every column of the data matrix and entering the results in a new matrix. In many time-sensitive engineering applications, multiplying matrices can give quick but good approximations of much more complicated calculations.

In a paper published in the July 13 issue of Proceedings of the National Academy of Science, MIT math professor Gilbert Strang describes a new way to split certain types of matrices into simpler matrices. The result could have implications for software that processes video or audio data, for compression software that squeezes down digital files so that they take up less space, or even for systems that control mechanical devices.

Strang’s analysis applies to so-called banded matrices. Most of the numbers in a banded matrix are zeroes; the only exceptions fall along diagonal bands, at or near the central diagonal of the matrix. This may sound like an esoteric property, but it often has practical implications. Some applications that process video or audio signals, for instance, use banded matrices in which each band represents a different time slice of the signal. By analyzing local properties of the signal, the application could, for instance, sharpen frames of video, or look for redundant information that can be removed to save memory or bandwidth.

Working backwards

Since most of the entries in a banded matrix — maybe 99 percent, Strang says — are zero, multiplying it by another matrix is a very efficient procedure: You can ignore all the zero entries. After a signal has been processed, however, it has to be converted back into its original form. That requires multiplying it by the “inverse” of the processing matrix: If multiplying matrix A by matrix B yields matrix C, multiplying C by the inverse of B yields A.

But the fact that a matrix is banded doesn’t mean that its inverse is. In fact, Strang says, the inverse of a banded matrix is almost always “full,” meaning that almost all of its entries are nonzero. In a signal-processing application, all the speed advantages offered by banded matrices would be lost if restoring the signal required multiplying it by a full matrix. So engineers are interested in banded matrices with banded inverses, but which matrices those are is by no means obvious.

In his PNAS paper, Strang describes a new technique for breaking a banded matrix up into simpler matrices — matrices with fewer bands. It’s easy to tell whether these simpler matrices have banded inverses, and if they do, their combination will, too. Strang’s technique thus allows engineers to determine whether some promising new signal-processing techniques will, in fact, be practical.

Faster than Fourier?

One of the most common digital-signal-processing techniques is the discrete Fourier transform (DFT), which breaks a signal into its component frequencies and can be represented as a matrix. Although the matrix for the Fourier transform is full, Strang says, “the great fact about the Fourier transform is that it happens to be possible, even though it’s full, to multiply fast and to invert it fast. That’s part of what makes Fourier wonderful.” Nonetheless, for some signal-processing applications, banded matrices could prove more efficient than the Fourier transform. If only parts of the signal are interesting, the bands provide a way to home in on them and ignore the rest. “Fourier transform looks at the whole signal at once,” Strang says. “And that’s not always great, because often the signal is boring for 99 percent of the time.”

Richard Brualdi, the emeritus UWF Beckwith Bascom Professor of Mathematics at the University of Wisconsin-Madison, points out that a mathematical conjecture that Strang presents in the paper has already been proven by three other groups of researchers. “It’s a very interesting theorem,” says Brualdi. “It’s already generated a couple of papers, and it’ll probably generate some more.” Brualdi points out that large data sets, such as those generated by gene sequencing, medical imaging, or weather monitoring, often yield matrices with regular structures. Bandedness is one type of structure, but there are others, and Brualdi expects other mathematicians to apply techniques like Strang’s to other types of structured matrices. “Whether or not those things will work, I really don’t know,” Brualdi says. “But Gil’s already said that he’s going to look at a different structure in a future paper.”