New Perspectives on Ab Initio Calculation and Physical

We explore the use of ab initio, density-functional methods for the study of large-scale materials problems. Three examples are presented: (i) the interplay of surface and edge reconstructions in long silicon nanowires, where we examine the effects on electrical and mechanical properties; (ii) the calculation of solvation effects based on ab initio dielectric models, where we derive the dielectric treatment as a coarse-grained molecular description, apply our method to the hydrolysis reaction of methylene chloride, and examine simplifications to our ab initio method; and (iii) the study of screw dislocation cores in bcc molybdenum and tantalum, where we find core structures contrary to those commonly accepted and barriers to dislocation motion in better agreement with experiment. Methodologically, we present a new matrix-based, algebraic formalism for ab initio calculations which modularizes and isolates the roles played by the basis set, the energy functional, the algorithm used to achieve self-consistency, and the computational kernels. Development and implementation of new techniques amounts to derivation and transcription of algebraic expressions. Modularizing the computational kernels yields portable codes that are easily optimized and parallelized, and we present highly efficient kernels for scalar, shared, and distributed memory computers. We conclude with an analytical study of the spatial locality of the single-particle density matrix in solid-state systems. This locality reflects the localization of electronic states and is essential for real-space and O(N) methods. We derive new behavior for this spatial range contrary to previous proposals, and we verify our findings in model semiconductors, insulators, and metals. Thesis Supervisor: TomBs A. Arias Title: Professor Thesis Supervisor: John D. Joannopoulos Title: Professor