Applications of Real-Time Multiresolution Analysis for Molecular Dynamics Simulations of Infrequent Events
- PDF / 282,675 Bytes
- 7 Pages / 612 x 792 pts (letter) Page_size
- 83 Downloads / 181 Views
Applications of Real-Time Multiresolution Analysis for Molecular Dynamics Simulations of Infrequent Events
David A. Richie,1 Jeongnim Kim and John W. Wilkins Department of Physics, The Ohio State University Columbus OH 43210, U.S.A. 1 The Ohio State University, 74 West 18th Avenue Columbus OH 43210, U.S.A. ABSTRACT
The simulation of defect dynamics (e.g., transient enhanced di usion of boron in the presence of silicon interstitials) is a technologically relevant challenge for computational materials science. The dynamics of defect structures in bulk unfolds as a sequence of thermally induced structural transitions. Identifying and characterizing reaction paths, as well as extracting dynamical quantities (e.g., di usion constants) is important for modeling the macroscopic properties of real materials. Applying real-time multiresolution analysis (RTMRA) to various dynamical quantities using simple Haar wavelets, we have developed a computationally cheap data compression scheme to handle the massive data sets generated in molecular dynamics (MD) simulations; data storage has been reduced hundredfold with no loss of relevant information. More importantly, the same RTMRA techniques are developed into a sophisticated event detection scheme capable of solving three major challenges to multiscale MD simulations, speci cally, (1) identifying meta-stable structures against the background of thermal vibrations, (2) detecting infrequent events, e.g., structural transitions, in the presence of thermal noise, and (3) accurately identifying transition times to further enhance recently emerging MD acceleration techniques. INTRODUCTION
The structural stability and dynamics of material defects are fundamental to materials science. In many cases, the origin of the complexities in real materials is not that they are defected in the trivial sense of being imperfect or impure, but rather that their material properties depend critically on their nonideality. As an example, the enhanced di usion of dopants in the presence of extended f311g defects in silicon is a limiting factor in the fabrication of shallow junction devices [1]. The growth of such extended defects involves the di usion, capture and dissociation of silicon point defects [2-4]. The dynamics will proceed by thermally induced infrequent events. Molecular dynamics (MD) simulations o er the opportunity to study these dynamics, identifying rate-limiting processes and relevant energy and time scales. Performing MD simulations of such a system presents numerous challenges. The long time scales involved in many processes put their simulation beyond today's computing power. However, recently emerging acceleration techniques [5-7] o er an opportunity to bridge the gap between these time-scales and achievable simulation times. The massive data sets generated by MD simulations of large systems over long time-scales also presents a problem. Using conventional approaches to recording the con guration of the system, AA5.1.1
gigabytes of data can be generated for the most modest calcu
Data Loading...
