Massively Parallel Computational Studies of Material Response at High Strain Rate Deformation
Abstract
Large scale molecular dynamics (MD) simulations are now commonly utilized to study materials at extreme conditions: high pressure and/or temperatures and ultra-high strain rates of deformation. A variety of emerging architectures such as general purpose graphics processing units (GP-GPU) and many-integrated core (MIC) architecture as well as new execution models have changed the traditional approach of high-performance computing. ExMatEx, the DoE initiative for enabling exa-scale (1018 flops) performance in scientific applications, has developed several proxy applications to facilitate co-design of novel algorithms and hardware by software developers and microchip vendors. We have used CoMD, a proxy application for classical MD, to investigate load balancing of shock wave simulation problems in various platforms as well as how best to improve performance of embedded-atom method (EAM) force evaluation kernels in GPUs. We also have implemented quasi-isentropic (QI) compression and expansion model in Los Alamos MD code SPaSM. QI uniaxial compression is achieved by incorporating a strain rate function in the position and velocity equations of motion. In this new formalism the change in internal energy is exactly equal to the work done during the compression or expansion. Large-Scale molecular dynamics (MD) simulations comprising 4 to 34 million atoms were performed to systematically study the poorly understood interplay between initial dislocation density and strain rate on deformation twinning in bcc metals. Using tantalum as a test case, for which a large body of experimental data exist. The atomic interactions were modeled employing an embedded-atom method (EAM) potential of Ta, we examined both compressive and tensile deformation at strain rates in the range of 108 – 1011 1/s . At these high-strain rates, twin nucleation thresholds can clearly be measured. Under both expansion and compression, deformation twinning increases with strain rate for strain-rates > 109 1/s.
Subject Area
Computational physics|Macroecology
Recommended Citation
Abeywardhana, Jayalath Abeywardhana Mudiyanselage Madawa, "Massively Parallel Computational Studies of Material Response at High Strain Rate Deformation" (2018). ETD Collection for University of Texas, El Paso. AAI10931981.
https://scholarworks.utep.edu/dissertations/AAI10931981