Quantum-accurate supplies simulation wins Gordon Bell Prize
The distinguished award offered by the Affiliation for Computing Equipment goes to the workforce of UM mechanical engineering professor Vikram Gavini.
See full bio
Professor of Mechanical Engineering
Professor of Supplies Science and Engineering
Precisely calculating the interactions between electrons has been a significant barrier to reliably exploring and designing supplies by means of pc modeling, however a UM-led workforce has developed code that brings the quantum mechanical precision of enormous methods to the size of immediately’s supercomputers.
This achievement has now been acknowledged by the Affiliation for Computing Equipment with the Gordon Bell Award.
“As we’ve proven on this work, there’s now a scientific path to attaining large-scale simulations of supplies with quantum precision,” mentioned Vikram Gavini, a professor of mechanical engineering and supplies science who leads the workforce. “This in flip can speed up our understanding of fabric properties and help within the computational design of supplies. Utility areas the place this might have an effect embody designing higher alloys, designing higher catalysts, and drug discovery, to call a number of.”
The usage of quantum mechanical calculations has tremendously expanded the neighborhood’s skill to characterize the properties and conduct of supplies. Inside the class of quantum mechanics, quantum many-body (QMB) methodologies characterize one finish of the spectrum, attaining excessive accuracy. Nonetheless, the necessity to mannequin quantum wave capabilities of a number of, related electrons signifies that computational complexity limits their utility to methods starting from tens of electrons to some thousand.
As a substitute, density useful concept (DFT) is an easier and extra environment friendly approach to analyze complicated supplies by contemplating electron densities slightly than multi-electron wave capabilities. DFT is predicated on the common change correlation (XC) operate, which incorporates quantum mechanical interactions between electrons. Its actual form is unknown and due to this fact approximate. The accuracy of calculations offered by DFT is said to the accuracy of XC capabilities.
“Detailed calculations of methods of atoms and molecules have had to decide on between the accuracy that may be achieved and the scale of the system that may be achieved,” mentioned Karthik Doraisamy, professor of aerospace and mechanical engineering and director of the Michigan Institute for Computational Discovery and Engineering. “Professor Gavini and his collaborators have centered extensively on this drawback for greater than a decade and have lately developed an method to interrupt this barrier by bridging QMB strategies with DFT to deal with the decision limitations of DFT.”
Gavini’s workforce has developed a machine-learned XC operate with an accuracy equal to QMB strategies at floor state energies. The workforce achieved this by fixing the inverse DFT drawback – which hyperlinks DFT and QMB to find out the XC operate that reproduces the QMB electron densities – leading to an correct XC operate that offers very correct electron densities.
“This recognition is a testomony to the continued excellence of the Gavini Group. This gifted workforce of researchers has developed many inventions to make this potential, and this work has the potential to pave the best way for large-scale predictive modeling for supplies physics and past. Congratulations to the workforce on coming into Within the period of exascale computing at UM.”
As a part of the Gordon Bell Prize submission, Gavini’s workforce carried out a calculation of the dislocation of magnesium with random dissolved yttrium atoms – a system involving roughly 620,000 electrons. They achieved an unprecedented sustained efficiency of 660 petaflops (660 quadrillion floating level operations per second) on the world’s strongest supercomputer: the Frontier exascale system positioned at Oak Ridge Nationwide Laboratory. Notably, this represents a tenfold enchancment within the demonstrated sustained efficiency of any floor state DFT calculation.
“This has been a multi-year effort involving graduate college students, analysis scientists and collaborators,” Gavini mentioned. “We’re very happy that this work has been acknowledged with this 12 months’s Gordon Bell Award.”
The workforce members from the College of Michigan who participated on this work are Dr. Sambit Das (Assistant Analysis Scientist, Mechanical Engineering), Dr. Bikash Kanungo (Assistant Analysis Scientist, Mechanical Engineering), Vishal Subramanian (PhD Candidate, Supplies Science and Engineering), Professor Paul Zimmerman (Chemistry), Professor Vikram Gavini (Mechanical Engineering, Supplies Science and Engineering). Different workforce members are Gaurab Phangrahi and Professor Vani Motmari (PhD 2014, Gavini Group) from the Indian Institute of Science and Dr. David Rogers from Oak Ridge Nationwide Laboratory.
ACM press launch
Written by Connor Titsworth/UM Workplace of the Vice President for Analysis