Hutter's Work On Communication Costs Leads to DOE Computational Science Graduate Fellowship
12/4/2018 1:39:18 PM
And Hutter (BS CompE ’17) put what he learned to immediate use. He led a project that produced a new communication-avoiding algorithm for obtaining an orthogonal factorization of a matrix. He also created a way to employ that algorithm, showing for the first time how additional memory can be used to speed execution time over state-of-the-art libraries for what is a near-universal problem, Solomonik said.
Now a second-year computer science PhD student, Hutter’s work has earned him a Computational Science Graduate Fellowship from the U.S. Department of Energy. The CSGF is the most prestigious and competitive computational science fellowship available from the federal agency, according to Solomonik, who is Hutter’s advisor.
Hutter was one of the two University of Illinois students to win one this fall, along with biochemistry PhD student Anda Trifan. Fewer than 6 percent of applicants are chosen each year.
The fellowships are administered by the Krell Institute in Ames, Iowa. They support PhD students focused on using high-performance computers to solve science and engineering problems that are of national importance. In Hutter’s case, that means using high-performance computing to improve the scalability of existing methods in quantum chemistry.
“The Department of Energy recognizes the importance of applying high performance computing to the field of quantum chemistry,” he said. “Improving existing algorithms to reduce communication will lead to computational methods that achieve higher accuracy at less cost.”
Hutter, who is from Geneva, Ill., plans to use his fellowship to develop practical algorithms to minimize communication costs of QR and eigenvalue factorizations conducted over a network. He also plans to investigate tensor factorizations for representing high-accuracy wave functions necessary in many electronic structure calculations.
“My research seeks to improve the scalability of electronic structure calculations in both theory and practice through the development of communication-avoiding tensor factorizations," Hutter said.
Solomonik said the fellowship should provide Hutter with a demanding and closely guided experience.
The DOE CSG Fellowship “exposes students to a wide network of alumni from a diverse set of disciplines in sciences and computation, who are present at every major national research lab and at most top universities,” Solomonik said.
Hutter, who called being chosen for the fellowship “an honor and a privilege,” echoed Solomonik on the uniqueness of the opportunity ahead of him.
“This fellowship provides the opportunity to explore electronic structure calculations from a computer science background that would not have been possible otherwise,” including spending part of a summer at Oak Ridge National Laboratory in Tennessee.