Tag

Conflux

Eric Parish, Aero Ph.D student, wins Von Neumann Fellowship from Sandia National Labs

By | Happenings, News, Research

Eric Parish

Eric Parish, who will graduate this spring with a Ph.D in Aerospace Engineering, is the 2018 recipient of the prestigious John von Neumann Postdoctoral Research Fellowship from Sandia National Laboratories (SNL). The highly competitive fellowship offers the opportunity to establish his own program at SNL to conduct innovative research in computational mathematics and scientific computing on advanced computing architectures.

Parish came to U-M from the University of Wyoming, and has developed innovative methodologies of computational math and physics with Prof. Karthik Duraisamy.

Parish said two of his accomplishments in his doctoral work have been developing data-driven solutions to computational physics problems using the NSF-funded ConFlux computing cluster, and bringing together ideas from statistical mechanics to develop efficient numerical solutions of complex partial differential equations.

“It was bridging a gap between communities,” he said of the latter effort.

“Eric came up with a particularly clever way of generalizing concepts from physics to develop a foundation to solve complex equations at a low cost in a mathematically rigorous fashion,” Duraisamy said. “He is one of the rare students who commands an exceptional grasp of applied mathematics, computing and physics, while being well-rounded in his organizational and communication skills. It has been a pleasure and a privilege to work with him.”

Parish said this research could eventually help usher the next generation of flight, for example, “hypersonic” aircraft that can travel at speeds of Mach 8-10. To help get there, his work moves the field toward a better understanding of the underlying physical phenomena via accurate numerical simulations.

At Sandia’s labs in Livermore, Calif., Parish said he plans to continue the work he started at U-M to develop “reduced order models”, which can process past simulation data to greatly reduce the computational cost of future simulations.

Parish said that conducting research at U-M, with the availability of high performance computing resources and a community of computational scientists to bounce ideas off of, helped push his research to a higher level.

“Within Aero, there are five or six very strong computational groups, which really helps me understand the fundamental aspects of what we’re doing, and what the addition of my small little delta means,” he said. “It’s very exciting to do computational research in that environment; it motivates me to come up with better code.”

In 2016, Parish received a $4,000 fellowship from the Michigan Institute for Computational Discovery and Engineering (MICDE). He used some of the funds to attend the International Workshop on Variational Multiscale Methods in Spain last year, where he met a few dozen people from around the world working on similar problems.

“It was fantastic to network and learn from them,” he said.

Parish grew up in Laramie, Wyo., before attending the University of Wyoming, where he played Division 1 golf. He said there was a small but active computational science community at U-W.

“For its size, there was a lot of good computational stuff there,” he said, adding that 10 years ago he would never have predicted the current direction of his career.

Golf played a significant role in his development as well, Parish said: “Being a successful student-athlete takes an extraordinary amount of work. The successes and failures I had … played an integral part in the development of my work ethic, time management skills, mental attitude, and overall growth as a person…I believe that the experience I gained as a student-athlete gave me a unique perspective and skill set that I was able to use to my advantage.”

As far as his future goes after Sandia, Parish said he plans to either continue in the national lab environment or to explore faculty positions so that he can teach and motivate students as his professors at Wyoming and Michigan did for him.

“I’m grateful for everyone’s help,” he said. “The doors that Michigan can open and the quality of people here are very apparent.”

A simulation of magnetohydrodynamic turbulence done on the ConFlux cluster with roughly 1 billion degree of freedom computation generating about 4TB of data.

ConFlux cluster expands

By | General Interest, Happenings, HPC, News

ARC-TS has installed 15 new compute nodes into the ConFlux cluster. These nodes have the same 20 cores CPU as the original set, but with 256 GB of RAM instead of 128 GB. Neither the original nodes nor the newly added ones contain any GPUs

As a result, jobs should spend less time in queue, and users can be more liberal in their memory requirements.

U-M, Toyota Research Institute partner in $2.4M battery project

By | General Interest, News, Research

With a $2.4 million investment from the Toyota Research Institute, University of Michigan researchers will develop computer simulation tools to predict automotive battery performance.

The project is part of a four-year, $35 million investment with research entities, universities and companies on research that uses artificial intelligence to help accelerate the design and discovery of advanced materials, TRI has announced.

Initially, the program will aim to help revolutionize materials science and identify new advanced battery materials and fuel cell catalysts that can power future zero-emissions and carbon-neutral vehicles.

“Toyota recognizes that artificial intelligence is a vital basic technology that can be leveraged across a range of industries, and we are proud to use it to expand the boundaries of materials science,” said Eric Krotkov, TRI chief science officer.

“Accelerating the pace of materials discovery will help lay the groundwork for the future of clean energy and bring us even closer to achieving Toyota’s vision of reducing global average new-vehicle CO2 emissions by 90 percent by 2050.”

The project, under the auspices of the Michigan Institute for Computational Discovery and Engineering at U-M, will combine mathematical models of the atomic nature and physics of materials with artificial intelligence.

“At the University of Michigan, we look forward to collaborating with TRI to advance computational materials science using machine learning principles,” said principal investigator Krishna Garikipati, professor of mechanical engineering and mathematics.

Also involved from U-M are Vikram Gavini, associate professor of mechanical engineering and materials science and engineering, and Karthik Duraisamy, assistant professor of aerospace engineering.

“The timing and goals of this program are well-aligned with the paradigm of data-enabled science that we have been promoting via the Michigan Institute for Computational Discovery and Engineering, and the Center for Data-Driven Computational Physics,” Duraisamy said.

The U-M project will use the ConFlux cluster, an innovative, new computing platform that enables computational simulations to interface with large datasets.

In addition to U-M, TRI’s newly funded research projects include collaborations with Stanford University, the Massachusetts Institute of Technology, University at Buffalo, University of Connecticut and the U.K.-based materials science company Ilika. TRI is also in ongoing discussions with additional research partners.

Research will merge advanced computational materials modeling, new sources of experimental data, machine learning and artificial intelligence in an effort to reduce the time scale for new materials development from a period that has historically been measured in decades.

Research programs will follow parallel paths, working to identify new materials for use in future energy systems as well as to develop tools and processes that can accelerate the design and development of new materials more broadly, according to TRI.

In support of these goals, TRI will partner on projects focused on areas including:

  • The development of new models and materials for batteries and fuel cells.
  • Broader programs to pursue novel uses of machine learning, artificial intelligence and materials informatics approaches for the design and development of new materials.
  • New automated materials discovery systems that integrate simulation, machine learning, artificial intelligence or robotics.

Accelerating materials science discovery represents one of four core focus areas for TRI, which was launched in 2015 with mandates to also enhance auto safety with automated technologies, increase access to mobility for those who otherwise cannot drive and help translate outdoor mobility technology into products for indoor mobility.