Scalability of Abinit on BlueGene/L for Identifying the Band Structure for Nanotechnology Materials

Abstract

The tradeoff between molecular complexity and accuracy of results continues to be a tug of war in computational material science. One of the challenges in computational nano materials science is the ability to parallelize the tools used, so as to facilitate the solution of more computationally extensive and complex problems with better accuracy. Though using more computationally powerful computers would enable solving larger problems, however real scalability is inherent to the underlying code used to implement the algorithms, and sometimes the algorithms themselves. Ab initio calculations are essential in identifying the properties of new nano materials. We investigate the performance scalability of Abinit1,2 in Band Structure calculations on a massively parallel architecture. We introduce a typical tool flow used for the computations involved. We study the scalability of the underlying tools while performing band structure calculations for C60 and EDOT molecules. We gained initial understanding of the factors affecting scalability for the molecules under consideration and we give some recommendations on how to address the possible issues that affect them.