Regensburg 2019 – wissenschaftliches Programm
MM 4.10: Vortrag
Montag, 1. April 2019, 12:45–13:00, H45
Machine learning enhanced atomistic simulation of ZrB2 at ultra-high temperatures — •Yanhui Zhang, Alessandro Lunghi, and Stefano Sanvito — School of Physics and CRANN, Trinity College Dublin, Dublin, Ireland
Machine-learned interatomic potentials (MLIP) are emerging as the tool of choice for molecular dynamics, since they exhibit robustness in large-scale atomistic simulations at a quasi-ab initio accuracy. Here we demonstrate the construction of such MLIP for the long-standing problem of extracting high-temperature properties of ultra-high temperature ceramics (UHTCs). Although some effort has been devoted in the past decades, the atomistic simulation at high temperatures is still sluggish. The development of a MLIP for UHTCs rises major challenges since it must simultaneously describe: 1) the constituent elements being much different from each other; 2) the nature of the bonding arising from a mixture of metallic, covalent and ionic interaction, 3) the complex response to heat and deformation loads, 4) the transferability across a wide range of temperatures and strains. All these attributes are extremely important for the accurate prediction of fundamental physical properties of UHTCs. We have developed a powerful MLIP using the SNAP model, which bears excellent temperature transferability. Thereafter, we have applied it to study the performances of UHTCs under heat and strain loads. Our simulations extend to temperature (up to 3000 K) far beyond what available in measurements. This work demonstrates that MLIPs are a very promising simulation tool in the field of ultra-high temperature materials.