(72i) MD Simulations with Petaflops Sustainable Performance for Trans-Scale Study On Transport and Reaction Processes

One of the fundamental challenges to chemical engineering is the vast scale difference between molecular structures that define the properties or functions of the chemical products and the reactors or equipments that actually produce these materials, it typically ranges from 10^-10m and 10^-15s to 10¹m and 10³s, and can be even wider. In particular, bottlenecks exist at the so-called meso-scales [1] where neither detailed dynamical study nor lumped statistical study seems to be sufficient or matured. Nano- and micro-fluidics, turbulent eddies and particle clusters in multiphase reaction systems are among the most well-known meso-scale structures.

The supercomputing capabilities available currently has provided an unique opportunity to breaking through some of these bottlenecks. In this presentation, we will demonstrate that, through consistent designing of the physical and mathematical models and the computer software and hardware [2], rigorous molecular dynamics (MD) can now reach truly Petaflops sustainable performance and micron scales in three dimensions or even millimeter scales in at least one dimension, which are reasonably “macro-scales” for statistical studies. For example, using up to 1728 GPUs of the Mole-8.5 system at IPE (http://www.top500.org/list/2010/06/100), a complete influenza virion, H1N1, constructed by 300 million atoms or radicals including the aqueous solution, is simulated at a speed of 0.77 ns per day [3]. Virtual experiments, such as the docking of the drugs can be conducted. In another example, using all CPUs and GPUs of the Tianhe-1A system (http://www.top500.org/system/10587), the world’s fastest computer currently, the silane pyrolysis and silicon deposition processes for the production of high-purity crystalline silicon were simulated with more than 100 billion atoms. The sustainable performance for the simulation of the bulk of silicon crystal has reached 1.87 Petaflops in single precision using all 7168 GPUs of Tianhe-1A [4].

Based on the strategy demonstrated in these examples, general purpose software and hardware platforms for high performance computing (HPC) of MD simulation can be established, which will proven to be a very powerful tool for exploring meso-scale structures, especially the fascinating nano- and micro-scale structures in chemical engineering.

References:

[1] Li J, Ge W, Kwauk M, 2009. Meso-scale phenomena from compromise -- a common challenge, not only for chemical engineering. arXiv:0912.5407.

[2] The EMMS group, 2011. Meso-scale oriented simulation towards virtual process engineering (VPE) ¯ the EMMS Paradigm. Chemical Engineering Science. In revision.

[3] Xu J, Wang X, He X et al., 2011. Application of Mole-8.5 HPC system ¯ Probing the whole influenza virion at the atomic level. Chinese Science Bulletin. In press.

[4] Ge W, Hou C, Xu J et al., 2011. Trans-scale Simulation of Silicon Deposition Process on Tianhe-1A. Submitted to Supercomputing 2011. Nov. 12-18, Seattle, USA.