March_2022_AMP_Digital

A D V A N C E D M A T E R I A L S & P R O C E S S E S | M A R C H 2 0 2 2 6 MACHINE LEARNING | AI SIMULATION CAPTURES DIAMOND MELTING A supercomputer simulation mod- el at Sandia National Laboratories, Al- buquerque, N.M., called SNAP (spectral neighbor analysis potential) can rap- idly predict the behavior of billions of interacting atoms. The model recent- ly captured the melting of diamond when compressed by extreme pres- sures and temperatures. At several mil- lion atmospheres, the diamond’s rig- id carbon lattice is shown in the simu- lations to crack, melt into amorphous carbon, and recrystallize. Scientists say the work could lead to a better under- standing of carbon-based exoplanets and have important implications for nu- clear fusion efforts that use polycrystal- line diamond. “We can now study the response of many materials under the same ex- treme pressures,” says SNAP creator Aidan Thompson. “Applications include planetary science, and it also opens the door to the design and manufacture of novel materials at extreme conditions.” SNAP used machine learning and other data science techniques to train a surrogate model that accurately re- produced the correct atomic forces. These were computed using high-accu- racy quantum mechanical calculations, which are only possible for systems containing a few hundred atoms. The surrogate model was then scaled up to predict forces and accelerations for systems containing billions of atoms. sandia.gov. USING AI TO OPTIMIZE BATTERIES Researchers working on flow bat- teries must find target molecules that offer the ability to both store a lot of en- ergy and remain stable for long periods of time. To discover the right molecules, scientists at the DOE’s Argonne Na- tional Laboratory, Lem- ont, Ill., are using artifi- cial intelligence (AI) to search through a vast chemical space of over a million molecules. Chemist Rajeev Assary and his colleagues mod- eled anolyte redoxmers, the electrically active molecules in a flow battery. For each redoxmer, the team identified three properties they wanted to optimize. The first two, re- duction potential and solvation free energy, relate to how much energy the molecule can store. The third, fluores- cence, serves as a self-reporting mark- er that indicates overall battery health. Because it is extremely time con- suming to calculate the properties of in- terest for all potential candidates, the team employed a machine learning and AI technique called active learning, in which a model can train itself to identi- fy increasingly plausible targets. The re- searchers began with a fairly small sam- ple—a dataset of 1400 redoxmer can- didates whose properties they already knew from quantum mechanical sim- ulations. By using this dataset as prac- tice, they were able to see that the algo- rithm correctly identified the molecules with the best properties. Once they had explored the small dataset, the team expanded their search to more than a million different candidates. Through the model’s iterative performance im- provement, better and better mole- cules began to be identified for further study. According to Assary, the optimi- zation algorithm could likely be applied to other types of batteries and other fields . anl.gov. Multibillion atom simulation of shockwave propagation into uncompressed diamond (blue) predicts that the final state (orange) is formed by recrystallization of amorphous cracks (red) that take shape in the light blue, green, and yellow compressed material. Researchers are using machine learning and AI to optimize new, potential chemicals for use in redox flow batteries that can modernize the electric grid.