Abstract
A simple practical step-forward procedure is employed for the rapid diagonalization of large matrices occurring in ab initio pseudopotential (PP) and linearized augmented plane wave (LAPW) calculations in the process leading to the charge density self-consistency. Much smaller energy cut-offs can be adopted to reduce the dimensionality of the eigenvalue equation by transforming the Hamiltonian and overlap (for LAPW) matrices into a diagonally-dominant form with the eigenvectors obtained in previous iterations. For the FLAPW method, the possible variation of basis functions in the muffin-tin and vacuum regions is treated by adopting both the fixed basis approach and the direct step-forward approach. Test calculations carried out for an Fe monolayer and the MgO (001) surface indicate that this procedure works very well for both metallic and ionic systems.
| Original language | English (US) |
|---|---|
| Pages (from-to) | 58-62 |
| Number of pages | 5 |
| Journal | Computer Physics Communications |
| Volume | 76 |
| Issue number | 1 |
| DOIs | |
| State | Published - Jun 1993 |
Funding
We thank Prof. D.S. Wang for helpful discussions. Work supported by the National Science Foundation (Grant No. DMR 91-17818) and by a grant of computer time at the Pittsburgh Supercomputing Center through its Division of Advanced Scientific Computing.
ASJC Scopus subject areas
- Hardware and Architecture
- General Physics and Astronomy