To achieve that level of accuracy, the team trained a novel machine-learning interatomic model by using an extensive cache of quantum mechanical data on various states of carbon, including BC8.
“We basically fingerprinted every atomic environment around each atom in a billion-atom system that could result during the system’s evolution at extreme pressures and temperatures,” Oleynik said.
Only an exascale supercomputing system like Frontier could deliver the computational power required to achieve quantum accuracy while efficiently simulating a billion atoms by using the machine-learning model. Oleynik and his team turned to the Oak Ridge Leadership Computing Facility, a DOE Office of Science user facility and home to Frontier at ORNL.
The team applied for and received an allocation of time on Frontier through DOE’s Innovative and Novel Computational Impact on Theory and Experiment program.
“Without Frontier, this would have been impossible,” Oleynik said. “For this study, we needed to simulate more than a billion atoms while performing up to a million time steps in molecular dynamics simulations. We had access to other supercomputers, but none of them even had enough computational power to handle that many atoms.”
Traditional supercomputing architectures based on CPUs bogged down when trying to run the Large-scale Atomic/Molecular Massively Parallel Simulator software module, or LAMMPS, code used by Oleynik and his team on such large simulations. Frontier and its hybrid architecture, built on a mix of CPUs and GPUs, offered a 50-times speedup over the fastest CPU-based competitors.
The research team used LAMMPS on Frontier to generate a range of potential scenarios and simulate conditions that might lead to BC8’s formation. Those conditions included pressures from 0 to 20 megabars — 20 million times the atmospheric pressure on Earth at sea level — and temperatures from 0 to 10,000 K, nearly twice as hot as the surface of the sun.
The computational power of Frontier allowed researchers to run the LAMMPS code in a single day — about 24 hours — by using 8,000 of Frontier’s more than 9,400 nodes.
The study captured time-stamped, atomically resolved snapshots of billion-atom dynamics for each half of a femtosecond — about a quadrillionth of a second — of each scenario.
“Frontier allowed us to achieve near-quantum accuracy in a tour-de-force effort,” Oleynik said. “Using this unprecedented computational power of Frontier, we discovered previous experiments had been focused on the wrong place.”
Frontier allowed the research team to predict the diamond’s transformation into BC8 and observe the atomic mechanism of that transformation at scale. The simulations showed the diamond first melts; then BC8 forms from the hot, dense carbon liquid.
“It’s a new discovery in that sense because in most cases materials transform from one crystalline phase into another by concerted rearrangement of an atomic structure,” Oleynik said. “But the carbon bonds that make up a diamond are so strong, we have to melt the diamond in order to transform it into a new BC8 crystalline phase. So that adds another layer to this process with even more extreme pressures and temperatures — 12 million times the pressure of Earth’s atmosphere and 5,000 K, which is close to the temperature of the sun’s surface. How do we connect all these material states at such extreme conditions and still reach the desired result?”
Oleynik and his team found that although a strong compression wave, known as a shock wave, might generate high pressures and temperatures, those conditions tend to be less than ideal for BC8’s synthesis.
“We found you can’t do this with a single shock,” he said. “So we used the simulations to design a sequence of shocks that bring the diamond precisely to the temperatures and pressures required to synthesize BC8.”
The team has begun to test their findings by attempting to synthesize BC8 at Lawrence Livermore National Laboratory’s National Ignition Facility.
“Thanks to Frontier, we have a good chance of success,” Oleynik said. “It’s still an extreme challenge with no guarantees, but we have great confidence in these results.”
Co-authors of the study included Kien Nguyen Cong, Jonathan Willman, Joseph Gonzalez and Ashley Williams at USF; Anatoly Belonoshko of the Swedish Royal Institute of Technology; Stan Moore, Aidan Thompson and Mitchell Wood at Sandia National Laboratories; and Marius Millot, Jon Eggert and Luis A. Zepeda-Ruiz at LLNL.
Support for this research came from the DOE Office of Science’s Advanced Scientific Computing Research program.
UT-Battelle manages ORNL for DOE’s Office of Science, the single largest supporter of basic research in the physical sciences in the United States. The Office of Science is working to address some of the most pressing challenges of our time. For more information, please visit energy.gov/science. — Matt Lakin
This Oak Ridge National Laboratory news article "Frontier simulations could help build a better diamond" was originally found on https://www.ornl.gov/news