TY - BOOK

T1 - Optimization and Parallelization of Complex DFT Codes

AU - Meisenbichler, Christian

N1 - no embargo

PY - 2012

Y1 - 2012

N2 - Density-functional theory (DFT) is one of the biggest fields in todays condensed-matter and materials science. It constitutes a precise and practical way to predict and interpret many material properties. DFT is a computational method to calculate the groundstate electron-density of systems like molecules and crystals. DFT is concerned with solving the Kohn-Sham equation. The DFT toolbox contains many codes, which differ mostly by the choice of the basis set on which the Kohn-Sham wave functions are represented. One of those is the LAPW ( linearized augmented plane-wave) basis set. This thesis presents a variety of implementations for the exciting code from the fields of numerical mathematics, user interfaces and general scientific software development. We examine possible optimizations, also within respect to enable the exciting code to run on parallel computer architectures. The presented optimizations involve the k-point parallelization, finding a better solver for the generalized eigenvalue problem, and convergence improving algorithms for the self consistent field loop. We introduce new input and output methods using XML which serves as the base for a new concept for the graphical user interface and results database exciting@web. Finally we discuss the necessary changes in the development process to allow for a better collaboration on the exciting code.

AB - Density-functional theory (DFT) is one of the biggest fields in todays condensed-matter and materials science. It constitutes a precise and practical way to predict and interpret many material properties. DFT is a computational method to calculate the groundstate electron-density of systems like molecules and crystals. DFT is concerned with solving the Kohn-Sham equation. The DFT toolbox contains many codes, which differ mostly by the choice of the basis set on which the Kohn-Sham wave functions are represented. One of those is the LAPW ( linearized augmented plane-wave) basis set. This thesis presents a variety of implementations for the exciting code from the fields of numerical mathematics, user interfaces and general scientific software development. We examine possible optimizations, also within respect to enable the exciting code to run on parallel computer architectures. The presented optimizations involve the k-point parallelization, finding a better solver for the generalized eigenvalue problem, and convergence improving algorithms for the self consistent field loop. We introduce new input and output methods using XML which serves as the base for a new concept for the graphical user interface and results database exciting@web. Finally we discuss the necessary changes in the development process to allow for a better collaboration on the exciting code.

KW - DFT

KW - Dichte-Functional Theorie

KW - LAPW

KW - linearized augmented plane wave

KW - parallelisierung

KW - XML

KW - Benutzerschnittstelle

KW - Wissenschaftliche Software-Entwicklung

KW - iterative eigenvector Löser

KW - Broyden mixing

KW - XML Datenbank

KW - XSLT

KW - refactoring

KW - Modularität

KW - ARPACK

KW - SMP

KW - exciting@web

KW - Scaling

KW - XForms

KW - ASE

KW - Software Tests

KW - DFT

KW - density-functional theory

KW - LAPW

KW - linearized augmented plane wave

KW - parallelization

KW - XML

KW - user interface

KW - scientific software development

KW - iterative eigenvector solver

KW - Broyden mixing

KW - XML Database

KW - XSLT

KW - refactoring

KW - modularity

KW - ARPACK

KW - SMP

KW - exciting@web

KW - Scaling

KW - XForms

KW - ASE

KW - Testing

M3 - Doctoral Thesis

ER -