CRD Summer Students, Advisor Use Virtual Visits to Conduct Real Research
September 18, 2020
By Jon Bashor
At a time when they should have been gathering at Berkeley Lab with dozens of other students from around the country to present posters about their summer experiences and learn about their peers’ efforts, Vanessa Meraz, Adrian De la Rocha, and Armando Garcia logged in from El Paso, Texas to explain their research via Zoom during an August 4 virtual poster session.
Angel Boada Velazco did likewise from San Diego. And Serges Love Teutu Talla stayed home in Baltimore, consulting with his advisor as he compiled a final summary of the knowledge and information he had gleaned over the past 10 weeks.
Welcome to the Computational Research Division (CRD) summer student program in the time of the COVID-19 pandemic. Although Computing Sciences canceled on-site student research projects, many of the students accepted for the summer were able to gain valuable research experience and knowledge.
“Typically, coming to Berkeley and experiencing the lab up close and personal is an important part of the summer student experience, but this year was anything but typical,” CRD Director David Brown said. “But the commitments by the students, their advisors, and our researchers made this summer a productive one and some teams will continue to collaborate into the school year.”
The students and their faculty advisors were part of a joint effort between Computing Sciences and the Sustainable Horizons Institute through the Sustainable Research Pathways (SRP) program. This program aims to recruit students and faculty from a variety of institutions, including minority-serving institutions and women’s colleges supporting students from under-represented or under-privileged backgrounds, for meaningful summer research opportunities.
Looking to Reduce the Computational Cost of the Prediction of Materials Properties
Meraz, De la Rocha, and Garcia, all physics students at The University of Texas at El Paso (UTEP), worked with their advisor, Jorge Munoz, to research machine learning methods to reduce the computational cost of density functional theory-based molecular dynamics simulations to enable the prediction of thermodynamic quantities from quantum mechanics. Their lab mentor was Bert De Jong, head of the Computational Chemistry, Materials, and Climate Group, as well as Alvarez Fellows Yu-Hang Tang and Tess Smidt.
“It has been a productive summer and now we are looking to develop a longer-term collaboration,” de Jong said. “Some of their work has already been submitted as posters at a major conference next year. They also got to do a lot of work at NERSC and use some of Berkeley Lab's unique machine learning capabilities through the Alvarez postdocs Yu-Hang Tang and Tess Smidt.”
Understanding the properties of materials under different temperatures or pressures is highly desirable for industrial applications and can open avenues for novel physics, but the number of possible configurations of the system grows rapidly when temperature causes the atoms to vibrate about their equilibrium positions. It is currently possible to simulate certain systems from quantum mechanics to understand their thermodynamics, but this normally requires a lot of supercomputing power.
The mathematical methods developed at Berkeley Lab that are being applied by the UTEP team reduce the amount of computation necessary by leveraging machine learning. The goal of the collaboration is to reduce computation enough that these calculations can be performed in smaller computer clusters so that they are available to more researchers, and this is important because being able to predict these thermodynamic effects computationally can accelerate the discovery of new materials.
“Every material has different crystalline structures and goes through phase transitions at different temperatures and pressures,” said Garcia, who is in the second year of his master’s program. “With these calculations, we hope to more accurately predict phase transitions and how that affects the properties of the material. There will still be testing in the lab, but it helps if you can make the calculations before testing.”
The team investigated simulations of Body-Centered Cubic Zirconium (bcc Zr) which is not thermodynamically stable at room temperature. Traditional density-functional theory methods simulate Zr atoms at 0 degrees Kelvin when the atoms are not vibrating. In order to simulate thermodynamics, up to 25,000 time steps with different atomic configurations are necessary to produce reliable results. To test their ideas, the team members used different methods to build a dataset using a bcc Zr supercell, first with 16 atoms to fine-tune their calculations and then with 128 atoms for their final results.
“Getting access to the Cori supercomputer for our calculations was exciting,” Maraz said. “It has been the fastest computer I've had the opportunity to use. It was impressive how, on average, the calculations took a shorter amount of time than what I'd been used to; even if they came out to be computationally costly.”
They also investigated the behavior of bcc-based FeTi, an intermetallic alloy of iron and titanium which shows interesting behavior for the atomic vibrations at high pressures. For example, the energies of some vibrations decrease when the system is squeezed, which is the opposite of what most materials do.
“I am extremely thankful for the opportunity to work with such outstanding scientists and propel my career plans forward,” De la Rocha said. “Not only did I gain a great amount of scientific knowledge, but it also served to give me more insight into the process of science itself and I am very satisfied with the results we have produced so far.”
Long-Distance Learning on Python
The Computing Sciences’ summer research program proved enlightening for Serges Love from Morgan State University, a historically Black university in Baltimore. His faculty advisor in mathematics, Dr. Isabelle Kemajou-Brown, also added to her computing knowledge along the way.
Just entering the first year of his Ph.D. program in mathematics, Love did not have much programming experience before working with Cy Chan in CRD’s Computer Architecture Group and Bin Wang, a former CRD post-doc now with Berkeley Lab’s Energy Analysis and Environmental Impacts Division.
“A week before I started, Dr. Chan and Dr. Wang thought that Python programming would be suitable for what I wanted to do and sent me links to some videos to learn about Python,” Love said. “It was challenging, but I enjoyed it.”
Love’s project was to see how vehicle flows could be improved across road network links depending on the time, congestion on the link, and other parameters.;
“The problem was both mathematical and computational, and I learned new things in both areas,” Love said. “As we got into the applications, I could see how beautiful it was. We came out with a model that gave us the best estimation of traffic flow under the set parameters.”
His advisor helped him with modeling and optimization while Chan and Wang helped with coding and data. Kemajou-Brown said the two of them spent up to eight hours a day working on the problem to make sure everything lined up.
Love admits he wasn’t sure how working remotely would work out, given that his lab mentors scheduled just two one-hour meetings per week, and he wondered if that would be enough.
“It amazed me how present they were, available to answer my questions all day during the week and on weekends,” Love said. “They also sent links so I could find the answers myself. It really boosted my confidence and helped me push through to the finish.”
Kemajou-Brown noticed that too.
“It’s really important that Serges got experience in working on a project from start to finish,” she said. “I was amazed at how fast he was learning the material. At one point, I thought ‘Whoa, he’s really rolling!’.”
Among the topics covered were the Python libraries that are available to solve different kinds of problems, what sorts of challenges arise when using those tools, how to measure and identify trends in performance, how to decompose large, difficult problems into smaller, more tractable ones, and what the trade-offs are, such as accuracy versus execution time, Chan explained.
“Serges did a great job learning some Python in the short time we had and we've discussed how to solve statistical problems with a computational mindset,” Chan said.
And some of it rubbed off on Kemajou-Brown, too.
“I myself learned a lot during this project,” she said. “I’ve used computers to solve problems, but often needed help to understand the code. But now, not only do I understand better, I can solve some of these problems myself.”
Back at Berkeley Lab (Virtually) for Second Summer
Although he was disappointed that, like his fellow students in Computing Sciences’ 2020 summer research program, he would not be working at Berkeley Lab in person, Angel Boada did have his 2019 in-person summer experience to help him adapt to the virtual environment.
Now in the last year of his Ph.D. program at San Diego State University, Boada spent this summer working with David Trebotich and Dan Graves of the Applied Numerical Algorithms Group (ANAG) integrating SuperLU, a sparse matrix solver developed by LBNL’s Sherry Li, into the Chombo-Crunch library developed over the years by ANAG staff, including Trebotich. Chombo-Crunch is a high-performance computational fluid dynamics (CFD) and reactive transport code used to model problems ranging from subsurface flows to aircraft turbulence.
The project is aimed at ensuring the Chombo-Crunch will perform well on high-performance systems using GPUs.
“It was a great experience being able to work with them,” Boada said. “Working remotely was challenging and they helped guide me through the library and to understand more of the details and how the code works internally so I could see how Chombo would communicate with SuperLU.”
Boada said he had several meetings a week with his lab mentors, each of which he said was a learning experience. He also said the work was aided by the strong documentation for SuperLU.
Although Boada participated in the August 4 and 11 virtual poster sessions, he missed the energy and chance to meet his peers and lab staff as he did at the 2019 session. He spent that summer also working with Trebotich and others.
“I did miss the experience of being onsite at the lab this year, and the interactions I had last year, as well as spending time in Berkeley,” he said, “but it was a great experience and has impacted my own research. I learned so much, even advice on how to present my results.”
About Computing Sciences at Berkeley Lab
The Computing Sciences Area at Lawrence Berkeley National Laboratory provides the computing and networking resources and expertise critical to advancing Department of Energy Office of Science research missions: developing new energy sources, improving energy efficiency, developing new materials, and increasing our understanding of ourselves, our world, and our universe.
Founded in 1931 on the belief that the biggest scientific challenges are best addressed by teams, Lawrence Berkeley National Laboratory and its scientists have been recognized with 13 Nobel Prizes. Today, Berkeley Lab researchers develop sustainable energy and environmental solutions, create useful new materials, advance the frontiers of computing, and probe the mysteries of life, matter, and the universe. Scientists from around the world rely on the Lab’s facilities for their own discovery science. Berkeley Lab is a multiprogram national laboratory, managed by the University of California for the U.S. Department of Energy’s Office of Science.
DOE’s Office of Science is the single largest supporter of basic research in the physical sciences in the United States, and is working to address some of the most pressing challenges of our time. For more information, please visit energy.gov/science.