(367e) Development of a Hierarchical Computational Architecture for Automated Nanomaterial Design
AIChE Annual Meeting
2008
2008 Annual Meeting
Computing and Systems Technology Division
Numerical Methods for Molecular and Mesoscopic Systems
Tuesday, November 18, 2008 - 4:35pm to 4:55pm
Molecular modeling and simulation has been recognized as an effective tool for improving more insightful and comprehensive understanding of nanomaterial structures, properties, which is critical for nanomaterial development. However, computational efficiency has been always a major challenge in molecular simulation, due to the existence of a huge design space which requires a thorough investigation to achieve design optimality. Tremendous efforts have been made on the development of parallel computing focused techniques.1-3 The developed techniques are very helpful for reducing the computational time of individual in silico experiments. However, nanomaterial development is a much more sophisticated task, involving the design and management of massive runs of simulations and numerous types of subtasks in multiple computational stages. Thus, an effective and efficient nanomaterial development must rely on the appropriate design of a computational architecture as well.
In this work, we will introduce a generic task-distributed hierarchical computational architecture for nanomaterial design, which can be readily constructed on a usual cluster system. In this architecture, the upper layer, designed using the master node of a cluster machine, is for managing the entire material analysis and design process. The lower layer consists of two computational zones, one for material sample development, and the other for material property testing. In each design iteration, nanomaterial formulation candidates are transmitted down to the sample development zone in the lower layer to generate samples by each compute node. The samples generated from this zone will be transmitted back to the master node in the upper layer, and the master node will then manage the sample transmission to the sample testing zone in the lower layer. The test results will be transmitted up to the master node. The material formulation-coating property correlation will be analyzed and the most preferable formulation will be identified for that design iteration. Based on the preset design objective, the Optimal Material Design Algorithm resided in the master node will determine if the next design iteration will be needed or not. If yes, then that algorithm will determine how to modify the formulation and subsequently a new iteration starts; otherwise, the identified material design will be considered the design solution. The developed computational architecture, together with the algorithms for hierarchical computations, can realize design automation, with significant computational efficiency. The introduced computational architecture has been successfully implemented in a Beowulf Cluster system (10 Xeon dual processor nodes @ 2.66 GHz). The efficacy of the architecture will be fully demonstrated through a case study, which will identify an optimal super-scratch-resistant nanopaint formulation.
References:
1. Xiao SP, Ni J, Wang SW. The bridging domain multiscale method and its high performance computing implementation. Journal of Computational and Theoretical Nanoscience. in press, 2008.
2. Ogata S, Lidorikis E, Shimojo F, Nakano A, Vashishta P, Kalia RK. Hybrid finite-element/molecular dynamics/electronic density functional approach to materials simulations on parallel computers. Computer Physics Communications. 2001; 138: 143-154.
3. Liu Z, Chen L, Paghavan P, Du Q, Sofo JO, Langer SA, Wolverton C. An integrated framework for multi-scale materials simulation and design. Journal of Computer Aided Materials Design. 2004; 11: 183-199.