High Performance Algorithms for Electronic Materials

点击翻译按钮获取中文摘要

9873664Chelikowsky and SaadThis is an interdisciplinary research grant funded jointly by the Divisions of Materials Research, of Advanced Computational Infrastructure and Research, and of Mathematical Sciences. The research involves algorithm development and implementation on massively parallel computers, and calculations of the properties of real materials using electronic structure theory. Materials of interest include amorphous materials, atomic clusters, liquids, glasses, extended defects, reconstructed point defects, non-ideal interfaces, etc. The recent development of high performance and novel architecture computers is creating new opportunities for the application of sophisticated electronic structure techniques to study these materials. Another two or more orders of magnitude improvement in speed, which is likely in the near future, will enable us to step beyond the current computational limits in materials science and reach an exciting era of understanding and discovery in real materials. These gains are not achieved by progress in hardware alone, but may depend more on innovations in algorithms. Developing new algorithms in multidisciplinary research requires an understanding of the different disciplines involved by the collaborating members, in this case a close collaboration between computer scientists and physical scientists.We will use real space algorithms (based on ab initio pseudopotentials) to predict the dielectric properties of quantum dots and nanostructured matter. We will apply a variety of methodologies to this problem including quasiparticle energies determined from density functional theory. We will consider quantum dots up to several thousand atoms in size. This will allow us to examine the transition from "atomic" to "crystalline" in terms of optical and dielectric response functions. Also, we will examine the structural properties of quantum dots, and extrinsic properties such as photoluminescence. We intend to continue our ongoing activities in other areas such as defects in solids, the structure and electronic properties of semiconductor liquids, and crystal growth.We will continue to examine new and efficient algorithms for computing eigenvectors and eigenvalues, with an emphasis on methods for computing a large number of eigenvectors. These algorithms include, but are not limited to, spectrum slicing via polynomial iterations and preconditioning techniques. In addition, we will examine new algorithms which avoid the computation of eigenvalues and eigenvectors altogether. One of these methods will be based on utilizing the Lanczos procedure.%%% This is an interdisciplinary research grant funded jointly by the Divisions of Materials Research, of Advanced Computational Infrastructure and Research, and of Mathematical Sciences. The research involves algorithm development and implementation on massively parallel computers, and calculations of the properties of real materials using electronic structure theory. Materials of interest include amorphous materials, atomic clusters, liquids, glasses, extended defects, reconstructed point defects, non-ideal interfaces, etc. The recent development of high performance and novel architecture computers is creating new opportunities for the application of sophisticated electronic structure techniques to study these materials. Another two or more orders of magnitude improvement in speed, which is likely in the near future, will enable us to step beyond the current computational limits in materials science and reach an exciting era of understanding and discovery in real materials. These gains are not achieved by progress in hardware alone, but may depend more on innovations in algorithms. Developing new algorithms in multidisciplinary research requires an understanding of the different disciplines involved by the collaborating members, in this case a close collaboration between computer scientists and physical scientists.***