| Kohn-Sham density functional theory (DFT) is an important computational tool in materials science for studying molecules, metals, insulators, and more. However, using DFT to study even modest size systems and timescales is extremely computationally demanding. To enable calculations on larger systems and longer timescales, much research has been done on reduced scaling methods based on localized orbitals. One such reduced scaling method called the Recursive Subspace Bisection (RSB) method was recently proposed to reduce the cost of hybrid density functional theory calculations. However, its applicability was limited to the case of computing hybrid functionals and to the study of relatively homogeneous systems. In this work, we expand and refine the RSB method so that it may be applied to a wider class of applications. First, we develop scheduling algorithms to allow for strong scaling parallel performance when applied to inhomogeneous systems. Second, we perform a detailed analysis of the error introduced by the RSB method, providing important reference data for future calculations on new systems. Finally, we expand the use of the RSB method beyond accelerating hybrid functionals to compressing wavefunction data. Throughout this work, we will also use the RSB method to provide new insight into the localization properties for a variety of systems. Together, the improvements and insights in this work will enable a new class of once intractable large-scale DFT calculations. |
