Tool predicts how fast code will run on a chip

MIT researchers have invented a machine-learning tool that predicts how fast computer chips will execute code from various applications.  

To get code to run as fast as possible, developers and compilers — programs that translate programming language into machine-readable code — typically use performance models that run the code through a simulation of given chip architectures. 

Compilers use that information to automatically optimize code, and developers use it to tackle performance bottlenecks on the microprocessors that will run it. But performance models for machine code are handwritten by a relatively small group of experts and are not properly validated. As a consequence, the simulated performance measurements often deviate from real-life results. 

In series of conference papers, the researchers describe a novel machine-learning pipeline that automates this process, making it easier, faster, and more accurate. In a paper presented at the International Conference on Machine Learning in June, the researchers presented Ithemal, a neural-network model that trains on labeled data in the form of “basic blocks” — fundamental snippets of computing instructions — to automatically predict how long it takes a given chip to execute previously unseen basic blocks. Results suggest Ithemal performs far more accurately than traditional hand-tuned models. 

Then, at the November IEEE International Symposium on Workload Characterization, the researchers presented a benchmark suite of basic blocks from a variety of domains, including machine learning, compilers, cryptography, and graphics that can be used to validate performance models. They pooled more than 300,000 of the profiled blocks into an open-source dataset called BHive. During their evaluations, Ithemal predicted how fast Intel chips would run code even better than a performance model built by Intel itself. 

Ultimately, developers and compilers can use the tool to generate code that runs faster and more efficiently on an ever-growing number of diverse and “black box” chip designs. “Modern computer processors are opaque, horrendously complicated, and difficult to understand. It is also incredibly challenging to write computer code that executes as fast as possible for these processors,” says co-author Michael Carbin, an assistant professor in the Department of Electrical Engineering and Computer Science (EECS) and a researcher in the Computer Science and Artificial Intelligence Laboratory (CSAIL). “This tool is a big step forward toward fully modeling the performance of these chips for improved efficiency.”

Most recently, in a paper presented at the NeurIPS conference in December, the team proposed a new technique to automatically generate compiler optimizations.  Specifically, they automatically generate an algorithm, called Vemal, that converts certain code into vectors, which can be used for parallel computing. Vemal outperforms hand-crafted vectorization algorithms used in the LLVM compiler — a popular compiler used in the industry.

Learning from data

Designing performance models by hand can be “a black art,” Carbin says. Intel provides extensive documentation of more than 3,000 pages describing its chips’ architectures. But there currently exists only a small group of experts who will build performance models that simulate the execution of code on those architectures. 

“Intel’s documents are neither error-free nor complete, and Intel will omit certain things, because it’s proprietary,” Mendis says. “However, when you use data, you don’t need to know the documentation. If there’s something hidden you can learn it directly from the data.”

To do so, the researchers clocked the average number of cycles a given microprocessor takes to compute basic block instructions — basically, the sequence of boot-up, execute, and shut down — without human intervention. Automating the process enables rapid profiling of hundreds of thousands or millions of blocks. 

Domain-specific architectures

In training, the Ithemal model analyzes millions of automatically profiled basic blocks to learn exactly how different chip architectures will execute computation. Importantly, Ithemal takes raw text as input and does not require manually adding features to the input data. In testing, Ithemal can be fed previously unseen basic blocks and a given chip, and will generate a single number indicating how fast the chip will execute that code. 

The researchers found Ithemal cut error rates in accuracy — meaning the difference between the predicted speed versus real-world speed — by 50 percent over traditional hand-crafted models. Further, in their next paper, they showed that Ithemal’s error rate was 10 percent, while the Intel performance-prediction model’s error rate was 20 percent on a variety of basic blocks across multiple different domains.

The tool now makes it easier to quickly learn performance speeds for any new chip architectures, Mendis says. For instance, domain-specific architectures, such as Google’s new Tensor Processing Unit used specifically for neural networks, are now being built but aren’t widely understood. “If you want to train a model on some new architecture, you just collect more data from that architecture, run it through our profiler, use that information to train Ithemal, and now you have a model that predicts performance,” Mendis says.

Next, the researchers are studying methods to make models interpretable. Much of machine learning is a black box, so it’s not really clear why a particular model made its predictions. “Our model is saying it takes a processor, say, 10 cycles to execute a basic block. Now, we’re trying to figure out why,” Carbin says. “That’s a fine level of granularity that would be amazing for these types of tools.”

They also hope to use Ithemal to enhance the performance of Vemal even further and achieve better performance automatically.

de MIT News https://ift.tt/2sTmsn1

Professor Emeritus Ali Argon, pioneer in the mechanics of materials, dies at 89

Ali S. Argon SM '53, ScD '56, the Quentin Berg Emeritus Professor of Mechanical Engineering, passed away on Dec. 21, 2019, at the age of 89. A world-leading expert in the mechanics of materials, Argon’s pioneering research furthered the field’s understanding of inelastic deformation and fracture of materials including metals and alloys, ceramics, glasses, polymers, and composites.

Argon was born in 1930 in Istanbul, Turkey, to a Turkish father and a German mother. After completing high school in Turkey, Argon moved to the United States, where he obtained a bachelor of science degree from Purdue University in 1952. He then enrolled in graduate school at MIT, where he studied materials science under Professor Egon Orowan. Argon received his master’s degree in mechanical engineering from MIT in 1953 and his doctoral degree in 1956. His doctoral thesis examined the strength and anelasticity of glass.

After receiving his doctorate, Argon spent two years working on Van de Graaf particle accelerators for both research and medical applications at the High Voltage Engineering Corporation in Burlington, Massachusetts. He then returned to Turkey in 1958 to serve in the Turkish Army Ordnance Corps.

In 1960, after two years of military service, Argon returned to MIT, having accepted a faculty position in mechanical engineering. By 1968, he was named a full professor. In 2001, Argon was named the Quentin Berg Professor of Mechanical Engineering at MIT.

Throughout his career, Argon combined novel experiments with theoretical and computational modeling to deepen the understanding of inelastic deformation and fracture of engineering materials. His research shed light on the connections between microstructure and macroscopic deformation and failure properties of engineering solids.

Having published 335 research works, Argon is one of the most-cited researchers in the field of mechanics of materials. Along with co-author Frank A. McClintock, he wrote the seminal text “Mechanical Behavior of Materials” (Addison-Wesley, 1966). As one of the first books to provide an overview of the mechanical behavior of metals as well as ceramics, rubbers, and polymers, many consider the work as the beginning of the mechanics and materials field.  

In addition to his impactful research contributions, Argon was a dedicated educator throughout his career at MIT. He mentored over 30 doctoral students, many of whom have gone on to become leading experts in the field. In the mid-1990s, Argon helped reshape the graduate program in mechanical engineering at MIT by leading an ad hoc committee. Under his leadership, the committee put forth recommendations for graduate programs designed for students interested in pursuing careers in industry.

Argon received numerous awards and honors in recognition for his research contributions. In 1989, he was elected to the National Academy of Engineering for “major contributions to the understanding of deformation and fracture of engineering materials through the application of mechanics to microstructure." He was also made a fellow of the American Physical Society. Among his many awards are the ASME Nadai Medal, ETH’s Staudinder Durrer Medal, and the Heyn Medal of the German Materials Society. In 2005 he received an honorary doctoral degree from his alma mater, Purdue University.

Argon is survived by his wife, Xenia (nee Lacher), and his son, Kermit. He was predeceased by his daughter, Alice, in 2015.

de MIT News https://ift.tt/2QNFhjg

Top MIT research stories of 2019

With a new year just begun, we take a moment to look back at the most popular articles of 2019 reflecting innovations, breakthroughs, and new insights from the MIT community. The following 10 research-related stories published in the previous 12 months received top views on MIT News. A selection of additional top news that you might have missed follows.

10. We’ve seen a black hole. An international team of astronomers, including scientists from MIT’s Haystack Observatory, announced the first direct images of a black hole in April. They accomplished this remarkable feat by coordinating the power of eight major radio observatories on four continents, to work together as a virtual, Earth-sized telescope.

9. The kilo is dead. Long live the kilo! On World Metrology Day, MIT Professor Wolfgang Ketterle delivered a talk on scientists’ new definition of the kilogram and the techniques for its measurement. As of May 20, a kilo is now defined by fixing the numerical value of a fundamental constant of nature known as the Planck constant.

8. A new record for blackest black. MIT engineers led by Professor Brian Wardle cooked up a material that is 10 times blacker than anything previously reported. The material is made from carbon nanotubes grown on chlorine-etched aluminum foil and captures at least 99.995 percent of incoming light. The material was featured as part of an exhibit at the New York Stock Exchange that was conceived by Diemut Strebe, MIT Center for Art, Science, and Technology artist-in-residence, in collaboration with Wardle and his lab.

7. Further evidence that Einstein was right. Physicists from MIT and elsewhere studied the ringing of an infant black hole, and found that the pattern of this ringing accurately predicts the black hole’s mass and spin — more evidence that Albert Einstein’s general theory of relativity is correct.

6. Understanding infections and autism. MIT and Harvard Medical School researchers uncovered a cellular mechanism that may explain why some children with autism experience a temporary reduction in behavioral symptoms when they have a fever.

5. A step toward pain-free diabetes treatments. An MIT-led research team developed a drug capsule that could be used to deliver oral doses of insulin, potentially replacing the injections that people with type 1 diabetes have to give themselves every day.

4. Da Vinci’s design holds up. Some 500 years after his death, MIT engineers and architects tested a design by Leonardo da Vinci for what would have been the world’s longest bridge span of its time. Their proof of the bridge’s feasibility sheds light on what ambitious construction projects might have been possible using only the materials and methods of the early Renaissance.

3. A novel kind of airplane wing. MIT and NASA engineers built and tested a radically new kind of airplane wing, assembled from hundreds of tiny identical pieces. The wing can change shape to control the plane’s flight, and, according to the researchers, could provide a significant boost in aircraft production, flight, and maintenance efficiency.

2. Simple programming for everyone. MIT researchers created a programming system with artificial intelligence that can easily be used by novices and experts alike. Users can create models and algorithms with the system, “Gen,” without having to deal with equations or handwrite high-performance code; experts can also use it to write sophisticated models and inference algorithms that were previously infeasible.

1. A new way to remove carbon dioxide from air. MIT researchers developed a system that can remove carbon dioxide from a stream of air at virtually any concentration level. The new method is significantly less energy-intensive and expensive than existing processes, and could provide a significant tool in the battle against climate change.

In case you missed it…

Additional top research stories of 2019 included a study finding better sleep habits lead to better college grades; a meta-study on the efficacy of educational technology; findings that science blooms after star researchers die; a system for converting the molecular structures of proteins into musical passages; and the answer to life, the universe, and everything. 

de MIT News https://ift.tt/2QkTDIX

School of Science recognizes members with 2020 Infinite Kilometer Awards

The MIT School of Science has announced the winners of the 2020 Infinite Kilometer Awards, which are presented annually to researchers within the school who are exceptional contributors to their communities.

These winners are nominated by their peers and mentors for their hard work, which can include mentoring and advising, supporting educational programs, providing service to groups such as the MIT Postdoctoral Association, or some other form of contribution to their home departments, labs, and research centers, the school, and the Institute.

The 2020 Infinite Kilometer Award winners in the School of Science are:

• Edgar Costa, a research scientist in the Department of Mathematics, nominated by Professor Bjorn Poonen and Principal Research Scientist Andrew Sutherland;
• Casey Rodriguez, an instructor in the Department of Mathematics, nominated by Professor Gigliola Staffilani;
• Rachel Ryskin, a postdoc in the Department of Brain and Cognitive Sciences, nominated by Professor Edward Gibson; and
• Grayson Sipe, a postdoc in the Picower Institute for Learning and Memory, nominated by Professor Mriganka Sur.

A monetary award is granted to recipients, and a celebratory reception will be held later this spring in their honor, attended by those who nominated them, family, and friends, in addition to the soon-to-be-announced recipients of the 2020 Infinite Mile Award.

de MIT News https://ift.tt/2Ql5uXp

How strong is your knot?

In sailing, rock climbing, construction, and any activity requiring the securing of ropes, certain knots are known to be stronger than others. Any seasoned sailor knows, for instance, that one type of knot will secure a sheet to a headsail, while another is better for hitching a boat to a piling. 

But what exactly makes one knot more stable than another has not been well-understood, until now. 

MIT mathematicians and engineers have developed a mathematical model that predicts how stable a knot is, based on several key properties, including the number of crossings involved and the direction in which the rope segments twist as the knot is pulled tight. 

“These subtle differences between knots critically determine whether a knot is strong or not,” says Jörn Dunkel, associate professor of mathematics at MIT. “With this model, you should be able to look at two knots that are almost identical, and be able to say which is the better one.”

“Empirical knowledge refined over centuries has crystallized out what the best knots are,” adds Mathias Kolle, the Rockwell International Career Development Associate Professor at MIT. “And now the model shows why.”

Dunkel, Kolle, and PhD students Vishal Patil and Joseph Sandt have published their results today in the journal Science. 

Pressure’s color

In 2018, Kolle’s group engineered stretchable fibers that change color in response to strain or pressure. The researchers showed that when they pulled on a fiber, its hue changed from one color of the rainbow to another, particularly in areas that experienced the greatest stress or pressure. 

An example of overhand knots.

Kolle, an associate professor of mechanical engineering, was invited by MIT’s math department to give a talk on the fibers. Dunkel was in the audience and began to cook up an idea: What if the pressure-sensing fibers could be used to study the stability in knots? 

Mathematicians have long been intrigued by knots, so much so that physical knots have inspired an entire subfield of topology known as knot theory — the study of theoretical knots whose ends, unlike actual knots, are joined to form a continuous pattern. In knot theory, mathematicians seek to describe a knot in mathematical terms, along with all the ways that it can be twisted or deformed while still retaining its topology, or general geometry. 

“In mathematical knot theory, you throw everything out that’s related to mechanics,” Dunkel says. “You don’t care about whether you have a stiff versus soft fiber — it’s the same knot from a mathematician’s point of view. But we wanted to see if we could add something to the mathematical modeling of knots that accounts for their mechanical properties, to be able to say why one knot is stronger than another.” 

Spaghetti physics

Dunkel and Kolle teamed up to identify what determines a knot’s stability. The team first used Kolle’s fibers to tie a variety of knots, including the trefoil and figure-eight knots — configurations that were familiar to Kolle, who is an avid sailor, and to rock-climbing members of Dunkel’s group. They photographed each fiber, noting where and when the fiber changed color, along with the force that was applied to the fiber as it was pulled tight.

The researchers used the data from these experiments to calibrate a model that Dunkel’s group previously implemented to describe another type of fiber: spaghetti. In that model, Patil and Dunkel described the behavior of spaghetti and other flexible, rope-like structures by treating each strand as a chain of small, discrete, spring-connected beads. The way each spring bends and deforms can be calculated based on the force that is applied to each individual spring. 

Kolle’s student Joseph Sandt had previously drawn up a color map based on experiments with the fibers, which correlates a fiber’s color with a given pressure applied to that fiber. Patil and Dunkel incorporated this color map into their spaghetti model, then used the model to simulate the same knots that the researchers had tied physically using the fibers. When they compared the knots in the experiments with those in the simulations, they found the pattern of colors in both were virtually the same — a sign that the model was accurately simulating the distribution of stress in knots. 

With confidence in their model, Patil then simulated more complicated knots, taking note of which knots experienced more pressure and were therefore stronger than other knots. Once they categorized knots based on their relative strength, Patil and Dunkel looked for an explanation for why certain knots were stronger than others. To do this, they drew up simple diagrams for the well-known granny, reef, thief, and grief knots, along with more complicated ones, such as the carrick, zeppelin, and Alpine butterfly.

An example of a reef knot.

Each knot diagram depicts the pattern of the two strands in a knot before it is pulled tight. The researchers included the direction of each segment of a strand as it is pulled, along with where strands cross. They also noted the direction each segment of a strand rotates as a knot is tightened. 

In comparing the diagrams of knots of various strengths, the researchers were able to identify general “counting rules,” or characteristics that determine a knot’s stability. Basically, a knot is stronger if it has more strand crossings, as well as more “twist fluctuations” — changes in the direction of rotation from one strand segment to another. 

For instance, if a fiber segment is rotated to the left at one crossing and rotated to the right at a neighboring crossing as a knot is pulled tight, this creates a twist fluctuation and thus opposing friction, which adds stability to a knot. If, however, the segment is rotated in the same direction at two neighboring crossing, there is no twist fluctuation, and the strand is more likely to rotate and slip, producing a weaker knot. 

They also found that a knot can be made stronger if it has more “circulations,” which they define as a region in a knot where two parallel strands loop against each other in opposite directions, like a circular flow. 

By taking into account these simple counting rules, the team was able to explain why a reef knot, for instance, is stronger than a granny knot. While the two are almost identical, the reef knot has a higher number of twist fluctuations, making it a more stable configuration. Likewise, the zeppelin knot, because of its slightly higher circulations and twist fluctuations, is stronger, though possibly harder to untie, than the Alpine butterfly — a knot that is commonly used in climbing. 

“If you take a family of similar knots from which empirical knowledge singles one out as “the best,” now we can say why it might deserve this distinction,” says Kolle, who envisions the new model can be used to configure knots of various strengths to suit particular applications. “We can play knots against each other for uses in suturing, sailing, climbing, and construction. It’s wonderful.”

This research was supported, in par,t by the Alfred P. Sloan Foundation, the James S. McDonnell Foundation, the Gillian Reny Stepping Strong Center for Trauma Innovation at Brigham and Women’s Hospital, and the National Science Foundation

de MIT News https://ift.tt/39B1lWS

How long will a volcanic island live?

When a hot plume of rock rises through the Earth’s mantle to puncture the overlying crust, it can create not only a volcanic ocean island, but also a swell in the ocean floor hundreds to thousands of kilometers long. Over time the island is carried away by the underlying tectonic plate, and the plume pops out another island in its place. Over millions of years, this geological hotspot can produce a chain of trailing islands, on which life may flourish temporarily before the islands sink, one by one, back into the sea. 

The Earth is pocked with dozens of hotspots, including those that produced the island chains of Hawaii and the Galapagos. While the process by which volcanic islands form is similar from chain to chain, the time that any island spends above sea level can vary widely, from a few million years in the case of the Galapagos to over 20 million for the Canary Islands. An island’s age can determine the life and landscapes that evolve there. And yet the mechanisms that set an island’s lifespan are largely unknown.

Now scientists at MIT have an idea about the processes that determine a volcanic island’s age. In a paper published today in Science Advances, they report an analysis of 14 major volcanic island chains around the world. They found that an island’s age is related to two main geological factors: the speed of the underlying plate and the size of the swell generated by the hotspot plume.

For instance, if an island lies on a fast-moving plate, it is likely to have a short lifespan, unless, as is the case with Hawaii, it was also created by a very large plume. The plume that gave rise to the Hawaiian islands is among the largest on Earth, and while the Pacific plate on which Hawaii sits is relatively speedy compared with other oceanic plates, it takes considerable time for the plate to slide over the plume’s expansive swell. 

The researchers found that this interplay between tectonic speed and plume size explains why the Hawaiian islands persist above sea level for million years longer than the oldest Galapagos Islands, which also sit on plates that travel at a similar speed but over a much smaller plume. By comparison, the Canary Islands, among the oldest island chains in the world, sit on the slow-moving Atlantic plate and over a relatively large plume. 

“These island chains are dynamic, insular laboratories that biologists have long focused on,” says former MIT graduate student Kimberly Huppert, the study’s lead author. “But besides studies on individual chains, there’s not a lot of work that related them to processes of the solid Earth, kilometers below the surface.”

“You can imagine all these organisms living on a sort of treadmill made of islands, like stepping stones, and they’re evolving, diverging, migrating to new islands, and the old islands are drowning,” adds Taylor Perron, associate head of MIT’s Department of Earth, Atmospheric and Planetary Sciences. “What Kim has shown is, there’s a geophysical mechanism that controls how fast this treadmill is moving and how long the island chains go before they drop off the end.”

Huppert and Perron co-authored the study with Leigh Royden, professor of earth, atmospheric and planetary sciences at MIT. 

Sinking a blowtorch

The new study is a part of Huppert’s MIT thesis work, in which she looked mainly at the evolution of landscapes on volcanic island chains, the Hawaiian islands in particular. In studying the processes that contribute to island erosion, she dug up a controversy in the literature regarding the processes that cause the seafloor to swell around hotspot islands. 

“The idea was, if you heat some of the bottom of the plate, you can make it go up really fast by just thermal uplift,  basically like a blowtorch under the plate,” Royden says. 

If this idea is correct, then by the same token, cooling of the heated plate should cause the seafloor to subside and islands to eventually sink back into the ocean. But in studying the ages of drowned islands in hotspot chains around the world, Huppert found that islands drown at a faster rate than any natural cooling mechanism could explain.

“So most of this uplift and sinking couldn’t have been from heating and cooling,” Royden says. “It had to be something else.”

Huppert’s observation inspired the group to compare major volcanic island chains in hopes of identifying the mechanisms of island uplift and sinking — which are likely the same processes that set an island’s lifespan, or time above sea level. 

Evolution, on a treadmill

In their analysis, the researchers looked at 14 volcanic island chains around the world, including the Hawaiian, Galapagos, and Canary islands. For each island chain, they noted the direction in which the underlying tectonic plate was moving and measured the plate’s average speed relative to the hotspot. They then measured, in the direction of each island chain, the distance between the beginning and the end of the swell, or uplift in the crust, created by the underlying plume. For every island chain, they divided the swell distance by plate velocity to arrive at a number representing the average time a volcanic island should spend atop the plume’s swell — which should determine how long an island remains above sea level before sinking into the ocean.

When the researchers compared their calculations with the actual ages of each island in each of the 14 chains, including islands that had long since sunk below sea level, they found a strong correlation between the time spent atop the swell and the typical amount of time that islands remain above sea level. A volcanic island’s lifespan, they concluded, depends on a combination of the underlying plate’s speed and the size of the plume, or swell that it creates. 

Huppert says that the processes that set an island’s age can help scientists better understand biodiversity and how life looks different from one island chain to another. 

“If an island spends a long time above sea level, that provides a long time for speciation to play out,” Huppert says. “But if you have an island chain where you have islands that drown at a faster rate, then it will affect the ability of fauna to radiate to neighboring islands, and how these islands are populated.”

The researchers posit that, in some sense, we have the interplay of tectonic speed and plume size to thank for our modern understanding of evolution. 

“You’re looking at a process in the solid Earth which is contributing to the fact that the Galapagos is a very fast moving treadmill, with islands moving off very quickly, with not a long time to erode, and this was the system that led to people discovering evolution,” Royden notes. “So in a sense this process really set the stage for humans to figure out what evolution was about, by doing it in this microcosm. If there hadn’t been this process, and the Galapagos hadn’t been on that short residence time, who knows how long it would have taken for people to figure it out.”

This research was supported, in part, by NASA.

de MIT News https://ift.tt/2tq86dx

Scientists pin down timing of lunar dynamo’s demise

A conventional compass would be of little use on the moon, which today lacks a global magnetic field. 

But the moon did produce a magnetic field billions of years ago, and it was likely even stronger than the Earth’s field today. Scientists believe that this lunar field, like Earth’s, was generated by a powerful dynamo — the churning of the moon’s core. At some point, this dynamo, and the magnetic field it generated, petered out.

Now scientists from MIT and elsewhere have pinned down the timing of the lunar dynamo’s end, to around 1 billion years ago. The findings appear today in the journal Science Advances. 

The new timing rules out some theories for what drove the lunar dynamo in its later stages and favors one particular mechanism: core crystallization. As the moon’s inner iron core crystallized, the liquid core’s electrically charged fluid was buoyantly stirred, producing the dynamo. 

“The magnetic field is this nebulous thing that pervades space, like an invisible force field,” says Benjamin Weiss, professor of earth, atmospheric, and planetary sciences at MIT. “We’ve shown that the dynamo that produced the moon’s magnetic field died somewhere between 1.5 and 1 billion years ago, and seems to have been powered in an Earth-like way.”

Weiss’ co-authors on the paper are co-lead authors Saied Mighani and Huapei Wang, as well as Caue Borlina and Claire Nichols of MIT, along with David Shuster of the University of California at Berkeley. 

Dueling dynamo theories

Over the past few years, Weiss’ group and others have discovered signs of a strong magnetic field, of around 100 microteslas, in lunar rocks as old as 4 billion years. For comparison, Earth’s magnetic field today is around 50 microteslas. 

In 2017, Weiss’s group studied a sample collected from NASA’s Apollo project, and found traces of a much weaker magnetic field, below 10 microteslas, in a moon rock they determined to be about 2.5 billion years old. Their thinking at the time was that perhaps two mechanisms for the lunar dynamo were at play: The first could have generated a much stronger, earlier magnetic field around 4 billion years ago, before being replaced by a second, more long-lived mechanism that sustained a much weaker field, through to at least 2.5 billion years ago. 

“There are several ideas for what mechanisms powered the lunar dynamo, and the question is, how do you figure out which one did it?” Weiss says. “It turns out all these power sources have different lifetimes. So if you could figure out when the dynamo turned off, then you could distinguish between the mechanisms that have been proposed for the lunar dynamo. That was the purpose of this new paper.”

Most of the magnetic studies lunar samples from the Apollo missions have been from ancient rocks, dating to about 3 billion to 4 billion years old. These are rocks that originally spewed out as lava onto a very young lunar surface, and as they cooled, their microscopic grains aligned in the direction of the moon’s magnetic field. Much of the moon’s surface is covered in such rocks, which have remained unchanged since, preserving a record of the ancient magnetic field.

However, lunar rocks whose magnetic histories began less than 3 billion years ago have been much harder to find because most lunar volcanism had ceased by this time. 

“The past 3 billion years of lunar history has been a mystery because there’s almost no rock record of it,” Weiss says.

“Little compasses”

Nevertheless, he and his colleagues identified two samples of lunar rock, collected by astronauts during the Apollo missions, that appear to have suffered a massive impact about 1 billion years ago and as a result were melted and welded back together in such a way that their ancient magnetic record was all but erased. 

The team took the samples back to the lab and first analyzed the orientation of each rock’s electrons, which Weiss describes as “little compasses” that either align in the direction of an existing magnetic field or appear in random orientations in the absence of one. For both samples, the team observed the latter: random configurations of electrons, suggesting that the rocks formed in an extremely weak to essentially zero magnetic field, of no more than 0.1 microteslas. 

The team then determined the age of both samples using a radiometric dating technique that Weiss and Shuster were able to adapt for this study.

The team put the samples through a battery of tests to see whether they were indeed good magnetic recorders. In other words, once they were reheated by some massive impact, could they have still been sensitive enough to record even a weak magnetic field on the moon, if it existed?

To answer this, the researchers placed both samples in an oven and blasted them with high temperatures to effectively erase their magnetic record, then exposed the rocks to an artificially generated magnetic field in the laboratory as they cooled. 

The results confirmed that the two samples were indeed reliable magnetic recorders and that the field strength they initially measured, of 0.1 microteslas, accurately represented the maximum possible value of the moon’s extremely weak magnetic field 1 billion years ago. Weiss says a field of 0.1 microteslas is so low that it’s likely the lunar dynamo ended by this time. 

The new findings line up with the predicted lifetime of core crystallization, a proposed mechanism for the lunar dynamo that could have generated a weak and long-lived magnetic field in the later part of the moon’s history. Weiss says that prior to core crystallization, a mechanism known as precession may have powered a much stronger though shorter-lived dynamo. Precession is a phenomenon by which the solid outer shell of a body such as the moon, in close proximity to a much larger body such as the Earth, wobbles in response to the Earth’s gravity. This wobbling stirs up the fluid in the core, the way swishing a cup of coffee stirs up the liquid inside.   

Around 4 billion years ago, the infant moon was likely much closer to the Earth than it is today, and much more susceptible to the planet’s gravitational effects. As the moon moved slowly away from the Earth, the effect of precession decreased, weakening the dynamo and the magnetic field in turn. Weiss says it’s likely that around 2.5 billion years ago, core crystallization became the dominant mechanism by which the lunar dynamo continued, producing a weaker magnetic field that continued to dissipate as the moon’s core eventually fully crystallized. 

The group is looking next to measure the direction of the moon’s ancient magnetic field in hopes of gleaning more information about the moon’s evolution.

This research was supported, in part, by NASA.

de MIT News https://ift.tt/2ZNh5Sj