Skip to content Skip to main navigation Report an accessibility issue

A Modern Forum

April 4, 2021

Some three decades ago, Alan Tennant was working toward a bachelor’s degree in physics in Scotland and Cristian Batista was doing the same in Argentina. They had never met, but their paths would merge years later through a shared interest in the structure and dynamics of quantum materials. It’s a common story in science, this bringing together people from the world over, but the Shull Wollan Center has emerged as a rich and unique chapter, connecting scientists from different specialties, generations, and institutions and placing within their reach some of the world’s most sophisticated facilities. As such, it has become an incubator for driving the creative thinking and innovation needed to thrive both in a quantum economy and in the education of next-generation scientists.

Filtering out the Noise

Sometimes it takes a little while to smooth out the edges in any system. Neutron scattering traces its roots to the science of the 1940s and has been tremendously successful at helping scientists understand the structure and behavior of materials at the atomic scale. Yet the results haven’t been without limitations.

The first challenge for researchers trying to understand a quantum material is finding a model describing that material’s various constituents. At the Spallation Neutron Source, for example, experiments provide vast amounts of data, but it can be difficult to look at all those findings and put them back to together—like trying to reassemble a piece of furniture with no instructions.

“You have the scattering data, and you need to somehow have the instructions that you can invert (so) you can create the model of what’s actually happening,” Tennant explained. “That’s the inverse scattering problem, which has been there since the beginning of scattering.”

Batista added there are other complications in creating these models: there simply may not be enough data for the desired parameters, or there might be noise or errors clouding the results. Magnetic materials—like those that power laptops and smart phones—are complex, with intricate dynamics at the quantum scale, which makes creating a model from their scattering data an even more formidable task.

In 2018 (?), a multi-disciplinary group of researchers began meeting at the Shull Wollan Center to consider how modern tools—artificial intelligence and machine-learning—could help sort out this problem. The team comprised computer scientists as well as computational, theoretical, and experimental physicists from Oak Ridge National Laboratory, Los Alamos National Laboratory, and the University of Tennessee, Knoxville. They met every week for two years, drawing on their collective expertise and experience. They saw tremendous potential in training an artificial neural network (an encoder) to recognize patterns and help point scientists in the right direction as they reassembled the seemingly inexhaustible possibilities for calculations.

“You create models that depend on lots of parameters,” Tennant said. “There’s far too many possibilities compared to what you could ever look at. What the neural net’s extremely good at is it kind of sees the overall pattern and it comes out with almost a hypothesis of what will happen in between the points that you’ve trained it on. A machine can learn to recognize so many more things than we can, because it takes us years to go through all the cases.”

Batista added that with huge amounts of data an encoder is much more efficient in extracting the essential features necessary for creating a structural model for a material, and it can help scientists distinguish between competing model candidates and again, help focus their calculations.

The center’s efforts, published in Nature Communications, culminated in an autoencoder that categorized different magnetic behaviors and eliminated background noise. The methodology not only cleared a fundamental bottleneck in condensed matter physics by offering optimized modeling but can also be applied to other types of materials and scattering problems.

Standing Alongside the Oracle

While there are plenty of successful collaborations in science, the Shull Wollan Center under Tennant’s and Batista’s leadership has a clear mission that takes into account a distinctive mix of expertise, facilities, and what Tennant refers to as “wherewithal.” The partnership between ORNL and UT is squarely positioned to take advantage of the SNS and leverage the combined resources: both in terms of personnel and instrumentation.

“You’ve got some really experienced people with a very high international profile who are from different realms of science but are connected by some common scientific things that we all know about,” Tennant said. “We’ve got a lot of young people around as well, and a lot of people coming through. And we’re in an interdisciplinary environment.”

He explained that other places have those assets, “but not in neutrons. There are other neutron centers, but not with the combination we have of researchers, the university, the national lab, (and) new instrumentation. We’re in position to forge a path in this whole realm of science, which is one that connects with so much other stuff—from structural biology to engineering materials to the future of quantum information.”

For Batista, who previously worked at LANL and is a faculty member with UT Physics, the center combines the multi-disciplinary resources of a national laboratory with the academic freedom of the university.

“What the center is giving us is a kind of flexibility that allows us to bring the best of both worlds,” he said.

That flexibility fuels incubation. While most organizations and managers are focused on delivering metrics, Tennant said “modern science is a combination of new ideas and the ability to carry them out.”

The team working on the machine-learning methodology included physicists and computer scientists, theorists and experimentalists. Joining the efforts were Kipton Barros of LANL (an expert in artificial intelligence), as well as ORNL researchers Ying Wai Li and Markus Eisenbach—both of whom, like Barros, are experts in computational physics. The group also crossed generations. SNS Postdoc Anjana Samarakoon, who played a key role in the work, came to the research as a graduate student with the University of Virginia.

“We did it as an interdisciplinary team,” Tennant said. “Where people are in their careers doesn’t matter for us. We’re not a hierarchy as such. You’ve got to have this bigger picture framework and huge knowledge base to pull this through from an idea to something that’s actually done. That’s incubating. You’ve got a big reservoir of skills and knowledge and experience to do that.”

With Samarakoon eventually moving to Argonne National Laboratory and Li currently working at Los Alamos, the research expands the center’s network and fosters the ebb and flow of talented scientists and expertise.

Building a framework of multi-disciplinary expertise and tools also accelerates progress. Batista explained that sometimes it turns out a problem has already been solved in a different scientific community, and it “just takes someone who speaks both languages” to transfer the solution.

“Innovation is where you’ve got ideas, skills, and perspectives in clusters” Tennant said. “(It’s) where you connect those clusters together. People who understand enough about the other cluster because they’ve got the wherewithal to absorb it are the ones that make all these connections happen. Once that connection is made, the flow happens. That’s the key thing.”

Or to borrow from the ancients: “It’s a bit like the Oracle at Delphi,” he said, “being in one place where you’re seeing a lot of what happens. You’re kind of at the nerve center.”

Machine learning in neutron scattering pipelines open new directions in autonomous scientific discovery. Image Credit: Alan Tennant

New Science in a Time-Honored Tradition

The Oracle at Delphi isn’t the only classical metaphor for the Shull Wollan Center’s success. Tennant and Batista are big believers in communication, something Tennant compares to the power of dialogue mastered by Aristotle and Plato.

“There’s an incredible power in dialogue itself,” he said. “Just reading papers: that’s a monologue. Dialogue is much more powerful because you can really get to the heart of the questions. And to have a dialogue, you need a forum, and you have to have the right people. When you’ve got real-quality dialogue happening in a forum, it’s transformative, and particularly at a time of disruptive change. You have people from different perspectives, but then they start having a common vision. It’s a creative process that requires interaction between people, and I think that’s critical. That’s where you get initiative.”

Batista drew on the more modern example of large tech companies building campuses with lots of open space for the spontaneous exchange of ideas.

“In some sense this Shull Wollan Center is a common space that we have within UT and Oak Ridge where people from different fields can meet,” he said.

For example, his group can develop theory ideas without having a material, but someone running an experiment at the SNS often comes in to offer their results.

“Every week, someone would come with new data,” he said. “Not everything would be interesting in the sense that you would learn something new, but from time to time you find exciting discoveries.”

This model—experimentalists and theorists working and talking together, often from different branches of study—is actually a throwback to an earlier time.

“Science was interdisciplinary for many centuries,” Batista said. “Now we’ve reached this level of specialization that we need to do something to recover that.”

Right Place, Right Time, Right People

While the center looks to return science to its interdisciplinary roots, Tennant and Batista also have an eye on the future. Inspired by the center’s successful endeavors in machine learning, Batista proposed a cluster hire at UT, which ultimately brought Adrian Del Maestro to the university in 2020. He holds a joint appointment in physics and computer science, connecting two traditional UT strengths. He’s also helping coordinate searches for additional hires for fall 2021, including an experimental neutron scatterer and a theoretician in topological materials.

“Clusters will probably be the rule in the future, following the funding,” Batista explained. “Having a model that is working at the small scale, the Shull Wollan Center, for me was very important, because then you can build.”

Building ties to education with a multi-disciplinary component is key for the center as it continues to drive new science. Tennant explained that the growing quantum economy will require engineers and physicists to understand the principles underpinning quantum information. The first cluster hire has already proven successful in that vein: Del Maestro taught a spring 2020 course called Special Topics in Physics: Introduction to Machine Learning, which focused on neural networks. Like other Shull Wollan Center affiliates, his group makes code and data available as open source, adding another component of expertise and education that can be shared quickly as the science unfolds.

Batista explained that having interdisciplinary teams already in place gives UT an advantage in going after external funding.

“We are already working together when a call comes,” he said. “If you’re already working with a team, it’s much easier to put together a project. The ideas for the proposals we’re writing right now were incubated at the Shull Wollan Center.”

Though the center takes advantage of some phenomenal scientific instruments, Tennant puts the center’s purpose in more relatable terms.

“Facilities are industrial-level science,” he said. “So you need this other, human, approach. Science is amazing (in) the way that it creates new knowledge and changes the world . . . but it requires leadership, vision, and environment.”

That, he explained, is the opportunity the Shull Wollan Center offers.

“It’s the right place, the right time, the right people,” he said. “So change happens.”