Expanding The Search For A Range Of New Materials

Finding new functional materials for batteries and catalysts and lots of other uses is a major goal of researchers around the world. And the design and discovery of new materials often requires computer simulations running on the world’s fastest supercomputers using specialized software that can determine properties at the quantum level.

Artificial intelligence and machine learning techniques are providing new avenues to speed up the search for promising new materials. This process will be further accelerated by the US Department of Energy’s powerful exascale supercomputers, including the Aurora system at Argonne National Laboratory and on other exascale-class systems that the DOE builds and supports.

To prepare for the exascale era of computing, a team of researchers has been enhancing the QMCPACK code to predict the properties of larger and more complex materials than were possible on previous-generation supercomputers. Supported by the Argonne Leadership Computing Facility’s Early Science Program, the team is improving the speed and accuracy of QMCPACK’s predictive capabilities to expedite the search for materials and molecules for applications ranging from batteries to medicine. Argonne computational scientist Anouar Benali is the principal investigator of the ESP project to prepare QMCPACK for Aurora. Benali’s project has been carried out in close collaboration with a larger QMCPACK project supported by DOE’s Exascale Computing Project. Led by Paul Kent at Oak Ridge National Laboratory, the ECP effort focused on readying QMCPACK software for the nation’s exascale systems, including Oak Ridge’s Frontier supercomputer.

Accelerating the Discovery Of New Materials

One of Benali’s targets is to speed up the search for promising new battery materials that are safer, more cost-effective, and longer-lasting to help meet the escalating needs of sustainable energy storage for electric vehicles, grid-level energy storage, and other applications.

As opposed to previous studies that focus on optimizing known materials, the team is employing an inverse design approach that uses QMCPACK in conjunction with ML to identify new materials with specific properties of interest.

“With inverse design, we are trying to find a material that corresponds to the desired property rather than evaluating the property based on a material we have,” Benali explained. “This requires research at the level of quantum mechanics.”

Predicting the properties of a material requires solving the many-body problem of interacting particles in a quantum system. The QMCPACK code uses the Quantum Monte Carlo (QMC) method which explicitly solves the many-body Schrödinger equation in a stochastic manner, and therefore captures strong correlation and weak interactions (van der Waals interactions), enabling more accurate treatment of quantum effects needed to simulate complex materials and molecules.

“By solving the Schrödinger equation using statistical methods, large and complex systems can be studied to unprecedented accuracy – including systems where other electronic structure methods have difficulty,” Benali explained.

Getting Ready for The Aurora Supercomputer

Work is in progress to get the Aurora exascale supercomputer ready for production. The system is powered by the latest Intel compute engines, specifically the Xeon Max Series CPUs, known by the codenamed “Sapphire Rapids,” and the Max Series GPUs, codenamed “Ponte Vecchio.” (The Next Platform drilled down into the Aurora system architecture here.)

The open source QMCPACK code used by the team is maintained by Argonne, Oak Ridge National Laboratory, Lawrence Livermore National Laboratory, Sandia National Laboratories, and several universities. The team has worked with ECP, ALCF, and the Intel Parallel Computing Center to modify the code to run efficiently on the Aurora supercomputer. Preparing software for an exascale system requires a new approach to software design and code refactoring.

“Our team has released multiple code updates with the goal of being able to port QMCPACK software to run on any exascale supercomputer,” said Ye Luo, an Argonne computational scientist and one of the lead developers of QMCPACK. “We completely refactored the QMCPACK code using new parallelization schemes to improve scalability and state-of-art algorithms to make the code faster and more efficient, which will allow us to study significantly larger size of systems at increased accuracy and shorter time to solution.”

“The initial performance results from our early runs on Aurora’s GPUs have been extremely promising,” Luo added.

With the arrival of the exascale era, systems like Aurora will enable the simulation of more realistic and complex materials in situ, replicating what experimentalists do. Benali and team are looking forward to the research advances that will be possible with the combination of QMCPACK, exascale computing power, and AI.

“With the boost we’re getting from exascale machines and our software, we’re now at a point where we can work together with AI and machine learning specialists to reverse engineer material design instead of trying everything at the simulation level,” Benali said. “If we know which properties we need for a particular application, we can use AI to scan for promising materials and tell us which ones to investigate further. This approach has the potential to revolutionize computer-aided materials discovery.”

Linda Barney is the founder and owner of Barney and Associates, a technical/marketing writing, training, and web design firm in Beaverton, Oregon.