Improving the Efficiency of FP-LAPW Calculations

TL;DR

This paper analyzes and optimizes the computational efficiency of the FP-LAPW method for electronic structure calculations, comparing different iterative schemes and hardware considerations to reduce runtime.

Contribution

It identifies bottlenecks in FP-LAPW calculations and proposes optimized formulations and iterative schemes to enhance computational efficiency.

Findings

Significant speed-up of FP-LAPW calculations achieved

Comparison of iterative schemes for Hamiltonian diagonalization

Hardware architecture impacts computational performance

Abstract

The full-potential linearized augmented-plane wave (FP-LAPW) method is well known to enable most accurate calculations of the electronic structure and magnetic properties of crystals and surfaces. The implementation of atomic forces has greatly increased it's applicability, but it is still generally believed that FP-LAPW calculations require substantial higher computational effort compared to the pseudopotential plane wave (PPW) based methods. In the present paper we analyse the FP-LAPW method from a computational point of view. Starting from an existing implementation (WIEN95 code), we identified the time consuming parts and show how some of them can be formulated more efficiently. In this context also the hardware architecture plays a crucial role. The remaining computational effort is mainly determined by the setup and diagonalization of the Hamiltonian matrix. For the latter, two different iterative schemes are compared. The speed-up gained by these optimizations is compared to the runtime of the ``original'' version of the code, and the PPW approach. We expect that the strategies described here, can also be used to speed up other computer codes, where similar tasks must be performed.