[ad_1]
A Sandia Nationwide Laboratories supercomputer simulation mannequin known as SNAP that quickly predicts the conduct of billions of interacting atoms has captured the melting of diamond when compressed by excessive pressures and temperatures. At a number of million atmospheres, the inflexible carbon lattice of the toughest recognized substance on Earth is proven in SNAP (Spectral Neighbor Evaluation Potential) simulations to crack, soften into amorphous carbon after which recrystallize. The work may support understanding of the inner construction of carbon-based exoplanets and have necessary implications for nuclear fusion efforts that make use of capsules fabricated from polycrystalline diamond.
This multi-billion atom simulation of shockwave propagation into initially uncompressed diamond (blue) makes use of a high-accuracy SNAP mannequin from Sandia Nationwide Laboratories to foretell that the ultimate state (orange) is shaped by recrystallization of amorphous cracks (crimson) that take form within the mild blue, inexperienced and yellow compressed materials.
Designing novel supplies and implications for large planets
“We are able to now research the response of many supplies below the identical excessive pressures,” stated Sandia scientist Aidan Thompson, who originated SNAP. “Purposes embody planetary science questions — for instance, what sort of influence stress would have led to the formation of our moon. It additionally opens the door to design and manufacture of novel supplies at excessive situations.”
The impact of utmost pressures and temperatures on supplies is also necessary for devising inside fashions of big planets. Highly effective DOE amenities like Sandia’s Z machine and Lawrence Livermore Nationwide Laboratory’s Nationwide Ignition Facility can recreate near-identical situations of those worlds in earthly experiments that supply close-up examinations of radically compressed supplies. However even these uniquely highly effective machines can’t pinpoint key microscopic mechanisms of change below these excessive situations, as a result of limitations in diagnostics on the stage of atoms.
“Solely pc simulations can try this,” stated Thompson.
Gordon Bell paper finalist is about ‘a micron-sized hunk of compressed diamond’
A technical paper describing the simulation was chosen as a finalist for the Gordon Bell prize, sponsored yearly by the Affiliation of Computing Equipment. The diamond-specific modeling, which took solely a day on the Summit supercomputer (the quickest within the U.S.) at Oak Ridge Nationwide Laboratory, was led by Prof. Ivan Oleynik on the College of South Florida. Along with Sandia and USF, the collaborative staff additionally included software program builders on the Division of Power’s Nationwide Power Analysis Scientific Computing Heart and the NVIDIA Company.
The staff’s simulations relied on SNAP, one of many main machine-learning descriptions of interatomic interactions, to mannequin and resolve an important downside, stated Thompson.
“We created gigantic simulations of a micron-sized hunk of compressed diamond,” stated Thompson. “To do that, we monitor the movement of billions of atoms by repeatedly calculating the atomic forces over very many, exceedingly tiny, intervals of time.”
Machine-learning bridged with quantum mechanical calculations
SNAP used machine-learning and different information science methods to coach a surrogate mannequin that faithfully reproduced the proper atomic forces. These have been calculated utilizing high-accuracy quantum mechanical calculations, that are solely doable for programs containing just a few hundred atoms. The surrogate mannequin was then scaled as much as predict forces and accelerations for programs containing billions of atoms. All native atomic buildings that emerged within the large-scale simulations have been well-represented within the small-scale coaching information, a essential situation for accuracy.
One other vital a part of the ultimate outcome was efficiency optimization of the software program to run effectively on GPU-based supercomputers like Summit, stated Thompson. “Since 2018, simply by bettering the software program, we now have been capable of make the SNAP code over 30 occasions quicker, shortening the time for these sorts of simulations by 97%. On the similar time, every technology of {hardware} is extra highly effective than the final. Because of this, calculations which may have till not too long ago taken a complete yr can now be run in a day on Summit.”
Run time shortened by 97 p.c
“Since supercomputer time is dear and extremely aggressive,” stated Thompson, “every shortening of SNAP’s run time saves cash and will increase the usefulness of the mannequin.”
Sandia researchers Stan Moore and Mitchell Wooden made necessary contributions to the SNAP mannequin and the dramatic efficiency enhancements.
The optimized software program for working SNAP on supercomputers is accessible within the open supply distribution of Sandia’s LAMMPS (https://www.lammps.org/) molecular dynamics code. The Sandia FitSNAP (https://github.com/FitSNAP/FitSNAP) software program for constructing new SNAP fashions can also be publicly obtainable.
The primary model of SNAP was created in 2012 with assist from Sandia’s Laboratory Directed Analysis and Improvement program. Software program enchancment has been supported repeatedly since 2017 by the DOE Exascale Computing Undertaking, a collaborative effort of the U.S. Division of Power Workplace of Science and the Nationwide Nuclear Safety Administration.
[ad_2]
